diff --git a/sys/compat/linuxkpi/common/src/linux_compat.c b/sys/compat/linuxkpi/common/src/linux_compat.c index 9962154cffd9..ebc1dbabd567 100644 --- a/sys/compat/linuxkpi/common/src/linux_compat.c +++ b/sys/compat/linuxkpi/common/src/linux_compat.c @@ -1,2871 +1,2867 @@ /*- * Copyright (c) 2010 Isilon Systems, Inc. * Copyright (c) 2010 iX Systems, Inc. * Copyright (c) 2010 Panasas, Inc. * Copyright (c) 2013-2021 Mellanox Technologies, Ltd. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include #include "opt_global.h" #include "opt_stack.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(__i386__) || defined(__amd64__) #include #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(__i386__) || defined(__amd64__) #include #include #endif #include #ifdef XENHVM #undef xen_pv_domain #undef xen_initial_domain /* xen/xen-os.h redefines __must_check */ #undef __must_check #include #endif SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "LinuxKPI parameters"); int linuxkpi_debug; SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN, &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable."); int linuxkpi_warn_dump_stack = 0; SYSCTL_INT(_compat_linuxkpi, OID_AUTO, warn_dump_stack, CTLFLAG_RWTUN, &linuxkpi_warn_dump_stack, 0, "Set to enable stack traces from WARN_ON(). Clear to disable."); static struct timeval lkpi_net_lastlog; static int lkpi_net_curpps; static int lkpi_net_maxpps = 99; SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN, &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second."); MALLOC_DEFINE(M_KMALLOC, "lkpikmalloc", "Linux kmalloc compat"); #include /* Undo Linux compat changes. */ #undef RB_ROOT #undef file #undef cdev #define RB_ROOT(head) (head)->rbh_root static void linux_destroy_dev(struct linux_cdev *); static void linux_cdev_deref(struct linux_cdev *ldev); static struct vm_area_struct *linux_cdev_handle_find(void *handle); cpumask_t cpu_online_mask; static cpumask_t **static_single_cpu_mask; static cpumask_t *static_single_cpu_mask_lcs; struct kobject linux_class_root; struct device linux_root_device; struct class linux_class_misc; struct list_head pci_drivers; struct list_head pci_devices; spinlock_t pci_lock; struct uts_namespace init_uts_ns; unsigned long linux_timer_hz_mask; wait_queue_head_t linux_bit_waitq; wait_queue_head_t linux_var_waitq; int panic_cmp(struct rb_node *one, struct rb_node *two) { panic("no cmp"); } RB_GENERATE(linux_root, rb_node, __entry, panic_cmp); #define START(node) ((node)->start) #define LAST(node) ((node)->last) INTERVAL_TREE_DEFINE(struct interval_tree_node, rb, unsigned long,, START, LAST,, lkpi_interval_tree) static void linux_device_release(struct device *dev) { pr_debug("linux_device_release: %s\n", dev_name(dev)); kfree(dev); } static ssize_t linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf) { struct class_attribute *dattr; ssize_t error; dattr = container_of(attr, struct class_attribute, attr); error = -EIO; if (dattr->show) error = dattr->show(container_of(kobj, struct class, kobj), dattr, buf); return (error); } static ssize_t linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf, size_t count) { struct class_attribute *dattr; ssize_t error; dattr = container_of(attr, struct class_attribute, attr); error = -EIO; if (dattr->store) error = dattr->store(container_of(kobj, struct class, kobj), dattr, buf, count); return (error); } static void linux_class_release(struct kobject *kobj) { struct class *class; class = container_of(kobj, struct class, kobj); if (class->class_release) class->class_release(class); } static const struct sysfs_ops linux_class_sysfs = { .show = linux_class_show, .store = linux_class_store, }; const struct kobj_type linux_class_ktype = { .release = linux_class_release, .sysfs_ops = &linux_class_sysfs }; static void linux_dev_release(struct kobject *kobj) { struct device *dev; dev = container_of(kobj, struct device, kobj); /* This is the precedence defined by linux. */ if (dev->release) dev->release(dev); else if (dev->class && dev->class->dev_release) dev->class->dev_release(dev); } static ssize_t linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf) { struct device_attribute *dattr; ssize_t error; dattr = container_of(attr, struct device_attribute, attr); error = -EIO; if (dattr->show) error = dattr->show(container_of(kobj, struct device, kobj), dattr, buf); return (error); } static ssize_t linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf, size_t count) { struct device_attribute *dattr; ssize_t error; dattr = container_of(attr, struct device_attribute, attr); error = -EIO; if (dattr->store) error = dattr->store(container_of(kobj, struct device, kobj), dattr, buf, count); return (error); } static const struct sysfs_ops linux_dev_sysfs = { .show = linux_dev_show, .store = linux_dev_store, }; const struct kobj_type linux_dev_ktype = { .release = linux_dev_release, .sysfs_ops = &linux_dev_sysfs }; struct device * device_create(struct class *class, struct device *parent, dev_t devt, void *drvdata, const char *fmt, ...) { struct device *dev; va_list args; dev = kzalloc(sizeof(*dev), M_WAITOK); dev->parent = parent; dev->class = class; dev->devt = devt; dev->driver_data = drvdata; dev->release = linux_device_release; va_start(args, fmt); kobject_set_name_vargs(&dev->kobj, fmt, args); va_end(args); device_register(dev); return (dev); } struct device * device_create_groups_vargs(struct class *class, struct device *parent, dev_t devt, void *drvdata, const struct attribute_group **groups, const char *fmt, va_list args) { struct device *dev = NULL; int retval = -ENODEV; if (class == NULL || IS_ERR(class)) goto error; dev = kzalloc(sizeof(*dev), GFP_KERNEL); if (!dev) { retval = -ENOMEM; goto error; } dev->devt = devt; dev->class = class; dev->parent = parent; dev->groups = groups; dev->release = device_create_release; /* device_initialize() needs the class and parent to be set */ device_initialize(dev); dev_set_drvdata(dev, drvdata); retval = kobject_set_name_vargs(&dev->kobj, fmt, args); if (retval) goto error; retval = device_add(dev); if (retval) goto error; return dev; error: put_device(dev); return ERR_PTR(retval); } struct class * lkpi_class_create(const char *name) { struct class *class; int error; class = kzalloc(sizeof(*class), M_WAITOK); class->name = name; class->class_release = linux_class_kfree; error = class_register(class); if (error) { kfree(class); return (NULL); } return (class); } static void linux_kq_lock(void *arg) { spinlock_t *s = arg; spin_lock(s); } static void linux_kq_unlock(void *arg) { spinlock_t *s = arg; spin_unlock(s); } static void linux_kq_assert_lock(void *arg, int what) { #ifdef INVARIANTS spinlock_t *s = arg; if (what == LA_LOCKED) mtx_assert(s, MA_OWNED); else mtx_assert(s, MA_NOTOWNED); #endif } static void linux_file_kqfilter_poll(struct linux_file *, int); struct linux_file * linux_file_alloc(void) { struct linux_file *filp; filp = kzalloc(sizeof(*filp), GFP_KERNEL); /* set initial refcount */ filp->f_count = 1; /* setup fields needed by kqueue support */ spin_lock_init(&filp->f_kqlock); knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock, linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock); return (filp); } void linux_file_free(struct linux_file *filp) { if (filp->_file == NULL) { if (filp->f_op != NULL && filp->f_op->release != NULL) filp->f_op->release(filp->f_vnode, filp); if (filp->f_shmem != NULL) vm_object_deallocate(filp->f_shmem); kfree_rcu(filp, rcu); } else { /* * The close method of the character device or file * will free the linux_file structure: */ _fdrop(filp->_file, curthread); } } struct linux_cdev * cdev_alloc(void) { struct linux_cdev *cdev; cdev = kzalloc(sizeof(struct linux_cdev), M_WAITOK); kobject_init(&cdev->kobj, &linux_cdev_ktype); cdev->refs = 1; return (cdev); } static int linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot, vm_page_t *mres) { struct vm_area_struct *vmap; vmap = linux_cdev_handle_find(vm_obj->handle); MPASS(vmap != NULL); MPASS(vmap->vm_private_data == vm_obj->handle); if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) { vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset; vm_page_t page; if (((*mres)->flags & PG_FICTITIOUS) != 0) { /* * If the passed in result page is a fake * page, update it with the new physical * address. */ page = *mres; vm_page_updatefake(page, paddr, vm_obj->memattr); } else { /* * Replace the passed in "mres" page with our * own fake page and free up the all of the * original pages. */ VM_OBJECT_WUNLOCK(vm_obj); page = vm_page_getfake(paddr, vm_obj->memattr); VM_OBJECT_WLOCK(vm_obj); vm_page_replace(page, vm_obj, (*mres)->pindex, *mres); *mres = page; } vm_page_valid(page); return (VM_PAGER_OK); } return (VM_PAGER_FAIL); } static int linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type, vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last) { struct vm_area_struct *vmap; int err; /* get VM area structure */ vmap = linux_cdev_handle_find(vm_obj->handle); MPASS(vmap != NULL); MPASS(vmap->vm_private_data == vm_obj->handle); VM_OBJECT_WUNLOCK(vm_obj); linux_set_current(curthread); down_write(&vmap->vm_mm->mmap_sem); if (unlikely(vmap->vm_ops == NULL)) { err = VM_FAULT_SIGBUS; } else { struct vm_fault vmf; /* fill out VM fault structure */ vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx); vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0; vmf.pgoff = 0; vmf.page = NULL; vmf.vma = vmap; vmap->vm_pfn_count = 0; vmap->vm_pfn_pcount = &vmap->vm_pfn_count; vmap->vm_obj = vm_obj; err = vmap->vm_ops->fault(&vmf); while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) { kern_yield(PRI_USER); err = vmap->vm_ops->fault(&vmf); } } /* translate return code */ switch (err) { case VM_FAULT_OOM: err = VM_PAGER_AGAIN; break; case VM_FAULT_SIGBUS: err = VM_PAGER_BAD; break; case VM_FAULT_NOPAGE: /* * By contract the fault handler will return having * busied all the pages itself. If pidx is already * found in the object, it will simply xbusy the first * page and return with vm_pfn_count set to 1. */ *first = vmap->vm_pfn_first; *last = *first + vmap->vm_pfn_count - 1; err = VM_PAGER_OK; break; default: err = VM_PAGER_ERROR; break; } up_write(&vmap->vm_mm->mmap_sem); VM_OBJECT_WLOCK(vm_obj); return (err); } static struct rwlock linux_vma_lock; static TAILQ_HEAD(, vm_area_struct) linux_vma_head = TAILQ_HEAD_INITIALIZER(linux_vma_head); static void linux_cdev_handle_free(struct vm_area_struct *vmap) { /* Drop reference on vm_file */ if (vmap->vm_file != NULL) fput(vmap->vm_file); /* Drop reference on mm_struct */ mmput(vmap->vm_mm); kfree(vmap); } static void linux_cdev_handle_remove(struct vm_area_struct *vmap) { rw_wlock(&linux_vma_lock); TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry); rw_wunlock(&linux_vma_lock); } static struct vm_area_struct * linux_cdev_handle_find(void *handle) { struct vm_area_struct *vmap; rw_rlock(&linux_vma_lock); TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) { if (vmap->vm_private_data == handle) break; } rw_runlock(&linux_vma_lock); return (vmap); } static int linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred, u_short *color) { MPASS(linux_cdev_handle_find(handle) != NULL); *color = 0; return (0); } static void linux_cdev_pager_dtor(void *handle) { const struct vm_operations_struct *vm_ops; struct vm_area_struct *vmap; vmap = linux_cdev_handle_find(handle); MPASS(vmap != NULL); /* * Remove handle before calling close operation to prevent * other threads from reusing the handle pointer. */ linux_cdev_handle_remove(vmap); down_write(&vmap->vm_mm->mmap_sem); vm_ops = vmap->vm_ops; if (likely(vm_ops != NULL)) vm_ops->close(vmap); up_write(&vmap->vm_mm->mmap_sem); linux_cdev_handle_free(vmap); } static struct cdev_pager_ops linux_cdev_pager_ops[2] = { { /* OBJT_MGTDEVICE */ .cdev_pg_populate = linux_cdev_pager_populate, .cdev_pg_ctor = linux_cdev_pager_ctor, .cdev_pg_dtor = linux_cdev_pager_dtor }, { /* OBJT_DEVICE */ .cdev_pg_fault = linux_cdev_pager_fault, .cdev_pg_ctor = linux_cdev_pager_ctor, .cdev_pg_dtor = linux_cdev_pager_dtor }, }; int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, unsigned long size) { vm_object_t obj; vm_page_t m; obj = vma->vm_obj; if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0) return (-ENOTSUP); VM_OBJECT_RLOCK(obj); for (m = vm_page_find_least(obj, OFF_TO_IDX(address)); m != NULL && m->pindex < OFF_TO_IDX(address + size); m = TAILQ_NEXT(m, listq)) pmap_remove_all(m); VM_OBJECT_RUNLOCK(obj); return (0); } void vma_set_file(struct vm_area_struct *vma, struct linux_file *file) { struct linux_file *tmp; /* Changing an anonymous vma with this is illegal */ get_file(file); tmp = vma->vm_file; vma->vm_file = file; fput(tmp); } static struct file_operations dummy_ldev_ops = { /* XXXKIB */ }; static struct linux_cdev dummy_ldev = { .ops = &dummy_ldev_ops, }; #define LDEV_SI_DTR 0x0001 #define LDEV_SI_REF 0x0002 static void linux_get_fop(struct linux_file *filp, const struct file_operations **fop, struct linux_cdev **dev) { struct linux_cdev *ldev; u_int siref; ldev = filp->f_cdev; *fop = filp->f_op; if (ldev != NULL) { if (ldev->kobj.ktype == &linux_cdev_static_ktype) { refcount_acquire(&ldev->refs); } else { for (siref = ldev->siref;;) { if ((siref & LDEV_SI_DTR) != 0) { ldev = &dummy_ldev; *fop = ldev->ops; siref = ldev->siref; MPASS((ldev->siref & LDEV_SI_DTR) == 0); } else if (atomic_fcmpset_int(&ldev->siref, &siref, siref + LDEV_SI_REF)) { break; } } } } *dev = ldev; } static void linux_drop_fop(struct linux_cdev *ldev) { if (ldev == NULL) return; if (ldev->kobj.ktype == &linux_cdev_static_ktype) { linux_cdev_deref(ldev); } else { MPASS(ldev->kobj.ktype == &linux_cdev_ktype); MPASS((ldev->siref & ~LDEV_SI_DTR) != 0); atomic_subtract_int(&ldev->siref, LDEV_SI_REF); } } #define OPW(fp,td,code) ({ \ struct file *__fpop; \ __typeof(code) __retval; \ \ __fpop = (td)->td_fpop; \ (td)->td_fpop = (fp); \ __retval = (code); \ (td)->td_fpop = __fpop; \ __retval; \ }) static int linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td, struct file *file) { struct linux_cdev *ldev; struct linux_file *filp; const struct file_operations *fop; int error; ldev = dev->si_drv1; filp = linux_file_alloc(); filp->f_dentry = &filp->f_dentry_store; filp->f_op = ldev->ops; filp->f_mode = file->f_flag; filp->f_flags = file->f_flag; filp->f_vnode = file->f_vnode; filp->_file = file; refcount_acquire(&ldev->refs); filp->f_cdev = ldev; linux_set_current(td); linux_get_fop(filp, &fop, &ldev); if (fop->open != NULL) { error = -fop->open(file->f_vnode, filp); if (error != 0) { linux_drop_fop(ldev); linux_cdev_deref(filp->f_cdev); kfree(filp); return (error); } } /* hold on to the vnode - used for fstat() */ vhold(filp->f_vnode); /* release the file from devfs */ finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops); linux_drop_fop(ldev); return (ENXIO); } #define LINUX_IOCTL_MIN_PTR 0x10000UL #define LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX) static inline int linux_remap_address(void **uaddr, size_t len) { uintptr_t uaddr_val = (uintptr_t)(*uaddr); if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR && uaddr_val < LINUX_IOCTL_MAX_PTR)) { struct task_struct *pts = current; if (pts == NULL) { *uaddr = NULL; return (1); } /* compute data offset */ uaddr_val -= LINUX_IOCTL_MIN_PTR; /* check that length is within bounds */ if ((len > IOCPARM_MAX) || (uaddr_val + len) > pts->bsd_ioctl_len) { *uaddr = NULL; return (1); } /* re-add kernel buffer address */ uaddr_val += (uintptr_t)pts->bsd_ioctl_data; /* update address location */ *uaddr = (void *)uaddr_val; return (1); } return (0); } int linux_copyin(const void *uaddr, void *kaddr, size_t len) { if (linux_remap_address(__DECONST(void **, &uaddr), len)) { if (uaddr == NULL) return (-EFAULT); memcpy(kaddr, uaddr, len); return (0); } return (-copyin(uaddr, kaddr, len)); } int linux_copyout(const void *kaddr, void *uaddr, size_t len) { if (linux_remap_address(&uaddr, len)) { if (uaddr == NULL) return (-EFAULT); memcpy(uaddr, kaddr, len); return (0); } return (-copyout(kaddr, uaddr, len)); } size_t linux_clear_user(void *_uaddr, size_t _len) { uint8_t *uaddr = _uaddr; size_t len = _len; /* make sure uaddr is aligned before going into the fast loop */ while (((uintptr_t)uaddr & 7) != 0 && len > 7) { if (subyte(uaddr, 0)) return (_len); uaddr++; len--; } /* zero 8 bytes at a time */ while (len > 7) { #ifdef __LP64__ if (suword64(uaddr, 0)) return (_len); #else if (suword32(uaddr, 0)) return (_len); if (suword32(uaddr + 4, 0)) return (_len); #endif uaddr += 8; len -= 8; } /* zero fill end, if any */ while (len > 0) { if (subyte(uaddr, 0)) return (_len); uaddr++; len--; } return (0); } int linux_access_ok(const void *uaddr, size_t len) { uintptr_t saddr; uintptr_t eaddr; /* get start and end address */ saddr = (uintptr_t)uaddr; eaddr = (uintptr_t)uaddr + len; /* verify addresses are valid for userspace */ return ((saddr == eaddr) || (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS)); } /* * This function should return either EINTR or ERESTART depending on * the signal type sent to this thread: */ static int linux_get_error(struct task_struct *task, int error) { /* check for signal type interrupt code */ if (error == EINTR || error == ERESTARTSYS || error == ERESTART) { error = -linux_schedule_get_interrupt_value(task); if (error == 0) error = EINTR; } return (error); } static int linux_file_ioctl_sub(struct file *fp, struct linux_file *filp, const struct file_operations *fop, u_long cmd, caddr_t data, struct thread *td) { struct task_struct *task = current; unsigned size; int error; size = IOCPARM_LEN(cmd); /* refer to logic in sys_ioctl() */ if (size > 0) { /* * Setup hint for linux_copyin() and linux_copyout(). * * Background: Linux code expects a user-space address * while FreeBSD supplies a kernel-space address. */ task->bsd_ioctl_data = data; task->bsd_ioctl_len = size; data = (void *)LINUX_IOCTL_MIN_PTR; } else { /* fetch user-space pointer */ data = *(void **)data; } #ifdef COMPAT_FREEBSD32 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { /* try the compat IOCTL handler first */ if (fop->compat_ioctl != NULL) { error = -OPW(fp, td, fop->compat_ioctl(filp, cmd, (u_long)data)); } else { error = ENOTTY; } /* fallback to the regular IOCTL handler, if any */ if (error == ENOTTY && fop->unlocked_ioctl != NULL) { error = -OPW(fp, td, fop->unlocked_ioctl(filp, cmd, (u_long)data)); } } else #endif { if (fop->unlocked_ioctl != NULL) { error = -OPW(fp, td, fop->unlocked_ioctl(filp, cmd, (u_long)data)); } else { error = ENOTTY; } } if (size > 0) { task->bsd_ioctl_data = NULL; task->bsd_ioctl_len = 0; } if (error == EWOULDBLOCK) { /* update kqfilter status, if any */ linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); } else { error = linux_get_error(task, error); } return (error); } #define LINUX_POLL_TABLE_NORMAL ((poll_table *)1) /* * This function atomically updates the poll wakeup state and returns * the previous state at the time of update. */ static uint8_t linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate) { int c, old; c = v->counter; while ((old = atomic_cmpxchg(v, c, pstate[c])) != c) c = old; return (c); } static int linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key) { static const uint8_t state[LINUX_FWQ_STATE_MAX] = { [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY, [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */ }; struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq); switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { case LINUX_FWQ_STATE_QUEUED: linux_poll_wakeup(filp); return (1); default: return (0); } } void linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p) { static const uint8_t state[LINUX_FWQ_STATE_MAX] = { [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY, [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */ [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED, }; /* check if we are called inside the select system call */ if (p == LINUX_POLL_TABLE_NORMAL) selrecord(curthread, &filp->f_selinfo); switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { case LINUX_FWQ_STATE_INIT: /* NOTE: file handles can only belong to one wait-queue */ filp->f_wait_queue.wqh = wqh; filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback; add_wait_queue(wqh, &filp->f_wait_queue.wq); atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED); break; default: break; } } static void linux_poll_wait_dequeue(struct linux_file *filp) { static const uint8_t state[LINUX_FWQ_STATE_MAX] = { [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT, [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT, [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT, }; seldrain(&filp->f_selinfo); switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { case LINUX_FWQ_STATE_NOT_READY: case LINUX_FWQ_STATE_QUEUED: case LINUX_FWQ_STATE_READY: remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq); break; default: break; } } void linux_poll_wakeup(struct linux_file *filp) { /* this function should be NULL-safe */ if (filp == NULL) return; selwakeup(&filp->f_selinfo); spin_lock(&filp->f_kqlock); filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ | LINUX_KQ_FLAG_NEED_WRITE; /* make sure the "knote" gets woken up */ KNOTE_LOCKED(&filp->f_selinfo.si_note, 1); spin_unlock(&filp->f_kqlock); } static void linux_file_kqfilter_detach(struct knote *kn) { struct linux_file *filp = kn->kn_hook; spin_lock(&filp->f_kqlock); knlist_remove(&filp->f_selinfo.si_note, kn, 1); spin_unlock(&filp->f_kqlock); } static int linux_file_kqfilter_read_event(struct knote *kn, long hint) { struct linux_file *filp = kn->kn_hook; mtx_assert(&filp->f_kqlock, MA_OWNED); return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0); } static int linux_file_kqfilter_write_event(struct knote *kn, long hint) { struct linux_file *filp = kn->kn_hook; mtx_assert(&filp->f_kqlock, MA_OWNED); return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0); } static struct filterops linux_dev_kqfiltops_read = { .f_isfd = 1, .f_detach = linux_file_kqfilter_detach, .f_event = linux_file_kqfilter_read_event, }; static struct filterops linux_dev_kqfiltops_write = { .f_isfd = 1, .f_detach = linux_file_kqfilter_detach, .f_event = linux_file_kqfilter_write_event, }; static void linux_file_kqfilter_poll(struct linux_file *filp, int kqflags) { struct thread *td; const struct file_operations *fop; struct linux_cdev *ldev; int temp; if ((filp->f_kqflags & kqflags) == 0) return; td = curthread; linux_get_fop(filp, &fop, &ldev); /* get the latest polling state */ temp = OPW(filp->_file, td, fop->poll(filp, NULL)); linux_drop_fop(ldev); spin_lock(&filp->f_kqlock); /* clear kqflags */ filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ | LINUX_KQ_FLAG_NEED_WRITE); /* update kqflags */ if ((temp & (POLLIN | POLLOUT)) != 0) { if ((temp & POLLIN) != 0) filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ; if ((temp & POLLOUT) != 0) filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE; /* make sure the "knote" gets woken up */ KNOTE_LOCKED(&filp->f_selinfo.si_note, 0); } spin_unlock(&filp->f_kqlock); } static int linux_file_kqfilter(struct file *file, struct knote *kn) { struct linux_file *filp; struct thread *td; int error; td = curthread; filp = (struct linux_file *)file->f_data; filp->f_flags = file->f_flag; if (filp->f_op->poll == NULL) return (EINVAL); spin_lock(&filp->f_kqlock); switch (kn->kn_filter) { case EVFILT_READ: filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ; kn->kn_fop = &linux_dev_kqfiltops_read; kn->kn_hook = filp; knlist_add(&filp->f_selinfo.si_note, kn, 1); error = 0; break; case EVFILT_WRITE: filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE; kn->kn_fop = &linux_dev_kqfiltops_write; kn->kn_hook = filp; knlist_add(&filp->f_selinfo.si_note, kn, 1); error = 0; break; default: error = EINVAL; break; } spin_unlock(&filp->f_kqlock); if (error == 0) { linux_set_current(td); /* update kqfilter status, if any */ linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); } return (error); } static int linux_file_mmap_single(struct file *fp, const struct file_operations *fop, vm_ooffset_t *offset, vm_size_t size, struct vm_object **object, int nprot, bool is_shared, struct thread *td) { struct task_struct *task; struct vm_area_struct *vmap; struct mm_struct *mm; struct linux_file *filp; vm_memattr_t attr; int error; filp = (struct linux_file *)fp->f_data; filp->f_flags = fp->f_flag; if (fop->mmap == NULL) return (EOPNOTSUPP); linux_set_current(td); /* * The same VM object might be shared by multiple processes * and the mm_struct is usually freed when a process exits. * * The atomic reference below makes sure the mm_struct is * available as long as the vmap is in the linux_vma_head. */ task = current; mm = task->mm; if (atomic_inc_not_zero(&mm->mm_users) == 0) return (EINVAL); vmap = kzalloc(sizeof(*vmap), GFP_KERNEL); vmap->vm_start = 0; vmap->vm_end = size; vmap->vm_pgoff = *offset / PAGE_SIZE; vmap->vm_pfn = 0; vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL); if (is_shared) vmap->vm_flags |= VM_SHARED; vmap->vm_ops = NULL; vmap->vm_file = get_file(filp); vmap->vm_mm = mm; if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) { error = linux_get_error(task, EINTR); } else { error = -OPW(fp, td, fop->mmap(filp, vmap)); error = linux_get_error(task, error); up_write(&vmap->vm_mm->mmap_sem); } if (error != 0) { linux_cdev_handle_free(vmap); return (error); } attr = pgprot2cachemode(vmap->vm_page_prot); if (vmap->vm_ops != NULL) { struct vm_area_struct *ptr; void *vm_private_data; bool vm_no_fault; if (vmap->vm_ops->open == NULL || vmap->vm_ops->close == NULL || vmap->vm_private_data == NULL) { /* free allocated VM area struct */ linux_cdev_handle_free(vmap); return (EINVAL); } vm_private_data = vmap->vm_private_data; rw_wlock(&linux_vma_lock); TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) { if (ptr->vm_private_data == vm_private_data) break; } /* check if there is an existing VM area struct */ if (ptr != NULL) { /* check if the VM area structure is invalid */ if (ptr->vm_ops == NULL || ptr->vm_ops->open == NULL || ptr->vm_ops->close == NULL) { error = ESTALE; vm_no_fault = 1; } else { error = EEXIST; vm_no_fault = (ptr->vm_ops->fault == NULL); } } else { /* insert VM area structure into list */ TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry); error = 0; vm_no_fault = (vmap->vm_ops->fault == NULL); } rw_wunlock(&linux_vma_lock); if (error != 0) { /* free allocated VM area struct */ linux_cdev_handle_free(vmap); /* check for stale VM area struct */ if (error != EEXIST) return (error); } /* check if there is no fault handler */ if (vm_no_fault) { *object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE, &linux_cdev_pager_ops[1], size, nprot, *offset, td->td_ucred); } else { *object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE, &linux_cdev_pager_ops[0], size, nprot, *offset, td->td_ucred); } /* check if allocating the VM object failed */ if (*object == NULL) { if (error == 0) { /* remove VM area struct from list */ linux_cdev_handle_remove(vmap); /* free allocated VM area struct */ linux_cdev_handle_free(vmap); } return (EINVAL); } } else { struct sglist *sg; sg = sglist_alloc(1, M_WAITOK); sglist_append_phys(sg, (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len); *object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len, nprot, 0, td->td_ucred); linux_cdev_handle_free(vmap); if (*object == NULL) { sglist_free(sg); return (EINVAL); } } if (attr != VM_MEMATTR_DEFAULT) { VM_OBJECT_WLOCK(*object); vm_object_set_memattr(*object, attr); VM_OBJECT_WUNLOCK(*object); } *offset = 0; return (0); } struct cdevsw linuxcdevsw = { .d_version = D_VERSION, .d_fdopen = linux_dev_fdopen, .d_name = "lkpidev", }; static int linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; ssize_t bytes; int error; error = 0; filp = (struct linux_file *)file->f_data; filp->f_flags = file->f_flag; /* XXX no support for I/O vectors currently */ if (uio->uio_iovcnt != 1) return (EOPNOTSUPP); if (uio->uio_resid > DEVFS_IOSIZE_MAX) return (EINVAL); linux_set_current(td); linux_get_fop(filp, &fop, &ldev); if (fop->read != NULL) { bytes = OPW(file, td, fop->read(filp, uio->uio_iov->iov_base, uio->uio_iov->iov_len, &uio->uio_offset)); if (bytes >= 0) { uio->uio_iov->iov_base = ((uint8_t *)uio->uio_iov->iov_base) + bytes; uio->uio_iov->iov_len -= bytes; uio->uio_resid -= bytes; } else { error = linux_get_error(current, -bytes); } } else error = ENXIO; /* update kqfilter status, if any */ linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ); linux_drop_fop(ldev); return (error); } static int linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; ssize_t bytes; int error; filp = (struct linux_file *)file->f_data; filp->f_flags = file->f_flag; /* XXX no support for I/O vectors currently */ if (uio->uio_iovcnt != 1) return (EOPNOTSUPP); if (uio->uio_resid > DEVFS_IOSIZE_MAX) return (EINVAL); linux_set_current(td); linux_get_fop(filp, &fop, &ldev); if (fop->write != NULL) { bytes = OPW(file, td, fop->write(filp, uio->uio_iov->iov_base, uio->uio_iov->iov_len, &uio->uio_offset)); if (bytes >= 0) { uio->uio_iov->iov_base = ((uint8_t *)uio->uio_iov->iov_base) + bytes; uio->uio_iov->iov_len -= bytes; uio->uio_resid -= bytes; error = 0; } else { error = linux_get_error(current, -bytes); } } else error = ENXIO; /* update kqfilter status, if any */ linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE); linux_drop_fop(ldev); return (error); } static int linux_file_poll(struct file *file, int events, struct ucred *active_cred, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; int revents; filp = (struct linux_file *)file->f_data; filp->f_flags = file->f_flag; linux_set_current(td); linux_get_fop(filp, &fop, &ldev); if (fop->poll != NULL) { revents = OPW(file, td, fop->poll(filp, LINUX_POLL_TABLE_NORMAL)) & events; } else { revents = 0; } linux_drop_fop(ldev); return (revents); } static int linux_file_close(struct file *file, struct thread *td) { struct linux_file *filp; int (*release)(struct inode *, struct linux_file *); const struct file_operations *fop; struct linux_cdev *ldev; int error; filp = (struct linux_file *)file->f_data; KASSERT(file_count(filp) == 0, ("File refcount(%d) is not zero", file_count(filp))); if (td == NULL) td = curthread; error = 0; filp->f_flags = file->f_flag; linux_set_current(td); linux_poll_wait_dequeue(filp); linux_get_fop(filp, &fop, &ldev); /* * Always use the real release function, if any, to avoid * leaking device resources: */ release = filp->f_op->release; if (release != NULL) error = -OPW(file, td, release(filp->f_vnode, filp)); funsetown(&filp->f_sigio); if (filp->f_vnode != NULL) vdrop(filp->f_vnode); linux_drop_fop(ldev); ldev = filp->f_cdev; if (ldev != NULL) linux_cdev_deref(ldev); linux_synchronize_rcu(RCU_TYPE_REGULAR); kfree(filp); return (error); } static int linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; struct fiodgname_arg *fgn; const char *p; int error, i; error = 0; filp = (struct linux_file *)fp->f_data; filp->f_flags = fp->f_flag; linux_get_fop(filp, &fop, &ldev); linux_set_current(td); switch (cmd) { case FIONBIO: break; case FIOASYNC: if (fop->fasync == NULL) break; error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC)); break; case FIOSETOWN: error = fsetown(*(int *)data, &filp->f_sigio); if (error == 0) { if (fop->fasync == NULL) break; error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC)); } break; case FIOGETOWN: *(int *)data = fgetown(&filp->f_sigio); break; case FIODGNAME: #ifdef COMPAT_FREEBSD32 case FIODGNAME_32: #endif if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) { error = ENXIO; break; } fgn = data; p = devtoname(filp->f_cdev->cdev); i = strlen(p) + 1; if (i > fgn->len) { error = EINVAL; break; } error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i); break; default: error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td); break; } linux_drop_fop(ldev); return (error); } static int linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot, vm_prot_t maxprot, int flags, struct file *fp, vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp) { /* * Character devices do not provide private mappings * of any kind: */ if ((maxprot & VM_PROT_WRITE) == 0 && (prot & VM_PROT_WRITE) != 0) return (EACCES); if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0) return (EINVAL); return (linux_file_mmap_single(fp, fop, foff, objsize, objp, (int)prot, (flags & MAP_SHARED) ? true : false, td)); } static int linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { struct linux_file *filp; const struct file_operations *fop; struct linux_cdev *ldev; struct mount *mp; struct vnode *vp; vm_object_t object; vm_prot_t maxprot; int error; filp = (struct linux_file *)fp->f_data; vp = filp->f_vnode; if (vp == NULL) return (EOPNOTSUPP); /* * Ensure that file and memory protections are * compatible. */ mp = vp->v_mount; if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { maxprot = VM_PROT_NONE; if ((prot & VM_PROT_EXECUTE) != 0) return (EACCES); } else maxprot = VM_PROT_EXECUTE; if ((fp->f_flag & FREAD) != 0) maxprot |= VM_PROT_READ; else if ((prot & VM_PROT_READ) != 0) return (EACCES); /* * If we are sharing potential changes via MAP_SHARED and we * are trying to get write permission although we opened it * without asking for it, bail out. * * Note that most character devices always share mappings. * * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE * requests rather than doing it here. */ if ((flags & MAP_SHARED) != 0) { if ((fp->f_flag & FWRITE) != 0) maxprot |= VM_PROT_WRITE; else if ((prot & VM_PROT_WRITE) != 0) return (EACCES); } maxprot &= cap_maxprot; linux_get_fop(filp, &fop, &ldev); error = linux_file_mmap_sub(td, size, prot, maxprot, flags, fp, &foff, fop, &object); if (error != 0) goto out; error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, FALSE, td); if (error != 0) vm_object_deallocate(object); out: linux_drop_fop(ldev); return (error); } static int linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) { struct linux_file *filp; struct vnode *vp; int error; filp = (struct linux_file *)fp->f_data; if (filp->f_vnode == NULL) return (EOPNOTSUPP); vp = filp->f_vnode; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_STAT(vp, sb, curthread->td_ucred, NOCRED); VOP_UNLOCK(vp); return (error); } static int linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct linux_file *filp; struct vnode *vp; int error; filp = fp->f_data; vp = filp->f_vnode; if (vp == NULL) { error = 0; kif->kf_type = KF_TYPE_DEV; } else { vref(vp); FILEDESC_SUNLOCK(fdp); error = vn_fill_kinfo_vnode(vp, kif); vrele(vp); kif->kf_type = KF_TYPE_VNODE; FILEDESC_SLOCK(fdp); } return (error); } unsigned int linux_iminor(struct inode *inode) { struct linux_cdev *ldev; if (inode == NULL || inode->v_rdev == NULL || inode->v_rdev->si_devsw != &linuxcdevsw) return (-1U); ldev = inode->v_rdev->si_drv1; if (ldev == NULL) return (-1U); return (minor(ldev->dev)); } static int linux_file_kcmp(struct file *fp1, struct file *fp2, struct thread *td) { struct linux_file *filp1, *filp2; if (fp2->f_type != DTYPE_DEV) return (3); filp1 = fp1->f_data; filp2 = fp2->f_data; return (kcmp_cmp((uintptr_t)filp1->f_cdev, (uintptr_t)filp2->f_cdev)); } struct fileops linuxfileops = { .fo_read = linux_file_read, .fo_write = linux_file_write, .fo_truncate = invfo_truncate, .fo_kqfilter = linux_file_kqfilter, .fo_stat = linux_file_stat, .fo_fill_kinfo = linux_file_fill_kinfo, .fo_poll = linux_file_poll, .fo_close = linux_file_close, .fo_ioctl = linux_file_ioctl, .fo_mmap = linux_file_mmap, .fo_chmod = invfo_chmod, .fo_chown = invfo_chown, .fo_sendfile = invfo_sendfile, .fo_cmp = linux_file_kcmp, .fo_flags = DFLAG_PASSABLE, }; /* * Hash of vmmap addresses. This is infrequently accessed and does not * need to be particularly large. This is done because we must store the * caller's idea of the map size to properly unmap. */ struct vmmap { LIST_ENTRY(vmmap) vm_next; void *vm_addr; unsigned long vm_size; }; struct vmmaphd { struct vmmap *lh_first; }; #define VMMAP_HASH_SIZE 64 #define VMMAP_HASH_MASK (VMMAP_HASH_SIZE - 1) #define VM_HASH(addr) ((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE]; static struct mtx vmmaplock; static void vmmap_add(void *addr, unsigned long size) { struct vmmap *vmmap; vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL); mtx_lock(&vmmaplock); vmmap->vm_size = size; vmmap->vm_addr = addr; LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next); mtx_unlock(&vmmaplock); } static struct vmmap * vmmap_remove(void *addr) { struct vmmap *vmmap; mtx_lock(&vmmaplock); LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next) if (vmmap->vm_addr == addr) break; if (vmmap) LIST_REMOVE(vmmap, vm_next); mtx_unlock(&vmmaplock); return (vmmap); } #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv) void * _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr) { void *addr; addr = pmap_mapdev_attr(phys_addr, size, attr); if (addr == NULL) return (NULL); vmmap_add(addr, size); return (addr); } #endif void iounmap(void *addr) { struct vmmap *vmmap; vmmap = vmmap_remove(addr); if (vmmap == NULL) return; #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv) pmap_unmapdev(addr, vmmap->vm_size); #endif kfree(vmmap); } void * vmap(struct page **pages, unsigned int count, unsigned long flags, int prot) { vm_offset_t off; size_t size; size = count * PAGE_SIZE; off = kva_alloc(size); if (off == 0) return (NULL); vmmap_add((void *)off, size); pmap_qenter(off, pages, count); return ((void *)off); } void vunmap(void *addr) { struct vmmap *vmmap; vmmap = vmmap_remove(addr); if (vmmap == NULL) return; pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE); kva_free((vm_offset_t)addr, vmmap->vm_size); kfree(vmmap); } static char * devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap) { unsigned int len; char *p; va_list aq; va_copy(aq, ap); len = vsnprintf(NULL, 0, fmt, aq); va_end(aq); if (dev != NULL) p = devm_kmalloc(dev, len + 1, gfp); else p = kmalloc(len + 1, gfp); if (p != NULL) vsnprintf(p, len + 1, fmt, ap); return (p); } char * kvasprintf(gfp_t gfp, const char *fmt, va_list ap) { return (devm_kvasprintf(NULL, gfp, fmt, ap)); } char * lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...) { va_list ap; char *p; va_start(ap, fmt); p = devm_kvasprintf(dev, gfp, fmt, ap); va_end(ap); return (p); } char * kasprintf(gfp_t gfp, const char *fmt, ...) { va_list ap; char *p; va_start(ap, fmt); p = kvasprintf(gfp, fmt, ap); va_end(ap); return (p); } static void linux_timer_callback_wrapper(void *context) { struct timer_list *timer; timer = context; /* the timer is about to be shutdown permanently */ if (timer->function == NULL) return; if (linux_set_current_flags(curthread, M_NOWAIT)) { /* try again later */ callout_reset(&timer->callout, 1, &linux_timer_callback_wrapper, timer); return; } timer->function(timer->data); } int mod_timer(struct timer_list *timer, int expires) { int ret; timer->expires = expires; ret = callout_reset(&timer->callout, linux_timer_jiffies_until(expires), &linux_timer_callback_wrapper, timer); MPASS(ret == 0 || ret == 1); return (ret == 1); } void add_timer(struct timer_list *timer) { callout_reset(&timer->callout, linux_timer_jiffies_until(timer->expires), &linux_timer_callback_wrapper, timer); } void add_timer_on(struct timer_list *timer, int cpu) { callout_reset_on(&timer->callout, linux_timer_jiffies_until(timer->expires), &linux_timer_callback_wrapper, timer, cpu); } int del_timer(struct timer_list *timer) { if (callout_stop(&(timer)->callout) == -1) return (0); return (1); } int del_timer_sync(struct timer_list *timer) { if (callout_drain(&(timer)->callout) == -1) return (0); return (1); } int timer_delete_sync(struct timer_list *timer) { return (del_timer_sync(timer)); } int timer_shutdown_sync(struct timer_list *timer) { timer->function = NULL; return (del_timer_sync(timer)); } /* greatest common divisor, Euclid equation */ static uint64_t lkpi_gcd_64(uint64_t a, uint64_t b) { uint64_t an; uint64_t bn; while (b != 0) { an = b; bn = a % b; a = an; b = bn; } return (a); } uint64_t lkpi_nsec2hz_rem; uint64_t lkpi_nsec2hz_div = 1000000000ULL; uint64_t lkpi_nsec2hz_max; uint64_t lkpi_usec2hz_rem; uint64_t lkpi_usec2hz_div = 1000000ULL; uint64_t lkpi_usec2hz_max; uint64_t lkpi_msec2hz_rem; uint64_t lkpi_msec2hz_div = 1000ULL; uint64_t lkpi_msec2hz_max; static void linux_timer_init(void *arg) { uint64_t gcd; /* * Compute an internal HZ value which can divide 2**32 to * avoid timer rounding problems when the tick value wraps * around 2**32: */ linux_timer_hz_mask = 1; while (linux_timer_hz_mask < (unsigned long)hz) linux_timer_hz_mask *= 2; linux_timer_hz_mask--; /* compute some internal constants */ lkpi_nsec2hz_rem = hz; lkpi_usec2hz_rem = hz; lkpi_msec2hz_rem = hz; gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div); lkpi_nsec2hz_rem /= gcd; lkpi_nsec2hz_div /= gcd; lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem; gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div); lkpi_usec2hz_rem /= gcd; lkpi_usec2hz_div /= gcd; lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem; gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div); lkpi_msec2hz_rem /= gcd; lkpi_msec2hz_div /= gcd; lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem; } SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL); void linux_complete_common(struct completion *c, int all) { - int wakeup_swapper; - sleepq_lock(c); if (all) { c->done = UINT_MAX; - wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0); + sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0); } else { if (c->done != UINT_MAX) c->done++; - wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0); + sleepq_signal(c, SLEEPQ_SLEEP, 0, 0); } sleepq_release(c); - if (wakeup_swapper) - kick_proc0(); } /* * Indefinite wait for done != 0 with or without signals. */ int linux_wait_for_common(struct completion *c, int flags) { struct task_struct *task; int error; if (SCHEDULER_STOPPED()) return (0); task = current; if (flags != 0) flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; else flags = SLEEPQ_SLEEP; error = 0; for (;;) { sleepq_lock(c); if (c->done) break; sleepq_add(c, NULL, "completion", flags, 0); if (flags & SLEEPQ_INTERRUPTIBLE) { DROP_GIANT(); error = -sleepq_wait_sig(c, 0); PICKUP_GIANT(); if (error != 0) { linux_schedule_save_interrupt_value(task, error); error = -ERESTARTSYS; goto intr; } } else { DROP_GIANT(); sleepq_wait(c, 0); PICKUP_GIANT(); } } if (c->done != UINT_MAX) c->done--; sleepq_release(c); intr: return (error); } /* * Time limited wait for done != 0 with or without signals. */ int linux_wait_for_timeout_common(struct completion *c, int timeout, int flags) { struct task_struct *task; int end = jiffies + timeout; int error; if (SCHEDULER_STOPPED()) return (0); task = current; if (flags != 0) flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; else flags = SLEEPQ_SLEEP; for (;;) { sleepq_lock(c); if (c->done) break; sleepq_add(c, NULL, "completion", flags, 0); sleepq_set_timeout(c, linux_timer_jiffies_until(end)); DROP_GIANT(); if (flags & SLEEPQ_INTERRUPTIBLE) error = -sleepq_timedwait_sig(c, 0); else error = -sleepq_timedwait(c, 0); PICKUP_GIANT(); if (error != 0) { /* check for timeout */ if (error == -EWOULDBLOCK) { error = 0; /* timeout */ } else { /* signal happened */ linux_schedule_save_interrupt_value(task, error); error = -ERESTARTSYS; } goto done; } } if (c->done != UINT_MAX) c->done--; sleepq_release(c); /* return how many jiffies are left */ error = linux_timer_jiffies_until(end); done: return (error); } int linux_try_wait_for_completion(struct completion *c) { int isdone; sleepq_lock(c); isdone = (c->done != 0); if (c->done != 0 && c->done != UINT_MAX) c->done--; sleepq_release(c); return (isdone); } int linux_completion_done(struct completion *c) { int isdone; sleepq_lock(c); isdone = (c->done != 0); sleepq_release(c); return (isdone); } static void linux_cdev_deref(struct linux_cdev *ldev) { if (refcount_release(&ldev->refs) && ldev->kobj.ktype == &linux_cdev_ktype) kfree(ldev); } static void linux_cdev_release(struct kobject *kobj) { struct linux_cdev *cdev; struct kobject *parent; cdev = container_of(kobj, struct linux_cdev, kobj); parent = kobj->parent; linux_destroy_dev(cdev); linux_cdev_deref(cdev); kobject_put(parent); } static void linux_cdev_static_release(struct kobject *kobj) { struct cdev *cdev; struct linux_cdev *ldev; ldev = container_of(kobj, struct linux_cdev, kobj); cdev = ldev->cdev; if (cdev != NULL) { destroy_dev(cdev); ldev->cdev = NULL; } kobject_put(kobj->parent); } int linux_cdev_device_add(struct linux_cdev *ldev, struct device *dev) { int ret; if (dev->devt != 0) { /* Set parent kernel object. */ ldev->kobj.parent = &dev->kobj; /* * Unlike Linux we require the kobject of the * character device structure to have a valid name * before calling this function: */ if (ldev->kobj.name == NULL) return (-EINVAL); ret = cdev_add(ldev, dev->devt, 1); if (ret) return (ret); } ret = device_add(dev); if (ret != 0 && dev->devt != 0) cdev_del(ldev); return (ret); } void linux_cdev_device_del(struct linux_cdev *ldev, struct device *dev) { device_del(dev); if (dev->devt != 0) cdev_del(ldev); } static void linux_destroy_dev(struct linux_cdev *ldev) { if (ldev->cdev == NULL) return; MPASS((ldev->siref & LDEV_SI_DTR) == 0); MPASS(ldev->kobj.ktype == &linux_cdev_ktype); atomic_set_int(&ldev->siref, LDEV_SI_DTR); while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0) pause("ldevdtr", hz / 4); destroy_dev(ldev->cdev); ldev->cdev = NULL; } const struct kobj_type linux_cdev_ktype = { .release = linux_cdev_release, }; const struct kobj_type linux_cdev_static_ktype = { .release = linux_cdev_static_release, }; static void linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate) { struct notifier_block *nb; struct netdev_notifier_info ni; nb = arg; ni.ifp = ifp; ni.dev = (struct net_device *)ifp; if (linkstate == LINK_STATE_UP) nb->notifier_call(nb, NETDEV_UP, &ni); else nb->notifier_call(nb, NETDEV_DOWN, &ni); } static void linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; struct netdev_notifier_info ni; nb = arg; ni.ifp = ifp; ni.dev = (struct net_device *)ifp; nb->notifier_call(nb, NETDEV_REGISTER, &ni); } static void linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; struct netdev_notifier_info ni; nb = arg; ni.ifp = ifp; ni.dev = (struct net_device *)ifp; nb->notifier_call(nb, NETDEV_UNREGISTER, &ni); } static void linux_handle_iflladdr_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; struct netdev_notifier_info ni; nb = arg; ni.ifp = ifp; ni.dev = (struct net_device *)ifp; nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni); } static void linux_handle_ifaddr_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; struct netdev_notifier_info ni; nb = arg; ni.ifp = ifp; ni.dev = (struct net_device *)ifp; nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni); } int register_netdevice_notifier(struct notifier_block *nb) { nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER( ifnet_link_event, linux_handle_ifnet_link_event, nb, 0); nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER( ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0); nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER( ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0); nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER( iflladdr_event, linux_handle_iflladdr_event, nb, 0); return (0); } int register_inetaddr_notifier(struct notifier_block *nb) { nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER( ifaddr_event, linux_handle_ifaddr_event, nb, 0); return (0); } int unregister_netdevice_notifier(struct notifier_block *nb) { EVENTHANDLER_DEREGISTER(ifnet_link_event, nb->tags[NETDEV_UP]); EVENTHANDLER_DEREGISTER(ifnet_arrival_event, nb->tags[NETDEV_REGISTER]); EVENTHANDLER_DEREGISTER(ifnet_departure_event, nb->tags[NETDEV_UNREGISTER]); EVENTHANDLER_DEREGISTER(iflladdr_event, nb->tags[NETDEV_CHANGEADDR]); return (0); } int unregister_inetaddr_notifier(struct notifier_block *nb) { EVENTHANDLER_DEREGISTER(ifaddr_event, nb->tags[NETDEV_CHANGEIFADDR]); return (0); } struct list_sort_thunk { int (*cmp)(void *, struct list_head *, struct list_head *); void *priv; }; static inline int linux_le_cmp(const void *d1, const void *d2, void *priv) { struct list_head *le1, *le2; struct list_sort_thunk *thunk; thunk = priv; le1 = *(__DECONST(struct list_head **, d1)); le2 = *(__DECONST(struct list_head **, d2)); return ((thunk->cmp)(thunk->priv, le1, le2)); } void list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv, struct list_head *a, struct list_head *b)) { struct list_sort_thunk thunk; struct list_head **ar, *le; size_t count, i; count = 0; list_for_each(le, head) count++; ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK); i = 0; list_for_each(le, head) ar[i++] = le; thunk.cmp = cmp; thunk.priv = priv; qsort_r(ar, count, sizeof(struct list_head *), linux_le_cmp, &thunk); INIT_LIST_HEAD(head); for (i = 0; i < count; i++) list_add_tail(ar[i], head); free(ar, M_KMALLOC); } #if defined(__i386__) || defined(__amd64__) int linux_wbinvd_on_all_cpus(void) { pmap_invalidate_cache(); return (0); } #endif int linux_on_each_cpu(void callback(void *), void *data) { smp_rendezvous(smp_no_rendezvous_barrier, callback, smp_no_rendezvous_barrier, data); return (0); } int linux_in_atomic(void) { return ((curthread->td_pflags & TDP_NOFAULTING) != 0); } struct linux_cdev * linux_find_cdev(const char *name, unsigned major, unsigned minor) { dev_t dev = MKDEV(major, minor); struct cdev *cdev; dev_lock(); LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) { struct linux_cdev *ldev = cdev->si_drv1; if (ldev->dev == dev && strcmp(kobject_name(&ldev->kobj), name) == 0) { break; } } dev_unlock(); return (cdev != NULL ? cdev->si_drv1 : NULL); } int __register_chrdev(unsigned int major, unsigned int baseminor, unsigned int count, const char *name, const struct file_operations *fops) { struct linux_cdev *cdev; int ret = 0; int i; for (i = baseminor; i < baseminor + count; i++) { cdev = cdev_alloc(); cdev->ops = fops; kobject_set_name(&cdev->kobj, name); ret = cdev_add(cdev, makedev(major, i), 1); if (ret != 0) break; } return (ret); } int __register_chrdev_p(unsigned int major, unsigned int baseminor, unsigned int count, const char *name, const struct file_operations *fops, uid_t uid, gid_t gid, int mode) { struct linux_cdev *cdev; int ret = 0; int i; for (i = baseminor; i < baseminor + count; i++) { cdev = cdev_alloc(); cdev->ops = fops; kobject_set_name(&cdev->kobj, name); ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode); if (ret != 0) break; } return (ret); } void __unregister_chrdev(unsigned int major, unsigned int baseminor, unsigned int count, const char *name) { struct linux_cdev *cdevp; int i; for (i = baseminor; i < baseminor + count; i++) { cdevp = linux_find_cdev(name, major, i); if (cdevp != NULL) cdev_del(cdevp); } } void linux_dump_stack(void) { #ifdef STACK struct stack st; stack_save(&st); stack_print(&st); #endif } int linuxkpi_net_ratelimit(void) { return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps, lkpi_net_maxpps)); } struct io_mapping * io_mapping_create_wc(resource_size_t base, unsigned long size) { struct io_mapping *mapping; mapping = kmalloc(sizeof(*mapping), GFP_KERNEL); if (mapping == NULL) return (NULL); return (io_mapping_init_wc(mapping, base, size)); } /* We likely want a linuxkpi_device.c at some point. */ bool device_can_wakeup(struct device *dev) { if (dev == NULL) return (false); /* * XXX-BZ iwlwifi queries it as part of enabling WoWLAN. * Normally this would be based on a bool in dev->power.XXX. * Check such as PCI PCIM_PCAP_*PME. We have no way to enable this yet. * We may get away by directly calling into bsddev for as long as * we can assume PCI only avoiding changing struct device breaking KBI. */ pr_debug("%s:%d: not enabled; see comment.\n", __func__, __LINE__); return (false); } static void devm_device_group_remove(struct device *dev, void *p) { const struct attribute_group **dr = p; const struct attribute_group *group = *dr; sysfs_remove_group(&dev->kobj, group); } int lkpi_devm_device_add_group(struct device *dev, const struct attribute_group *group) { const struct attribute_group **dr; int ret; dr = devres_alloc(devm_device_group_remove, sizeof(*dr), GFP_KERNEL); if (dr == NULL) return (-ENOMEM); ret = sysfs_create_group(&dev->kobj, group); if (ret == 0) { *dr = group; devres_add(dev, dr); } else devres_free(dr); return (ret); } #if defined(__i386__) || defined(__amd64__) bool linux_cpu_has_clflush; struct cpuinfo_x86 boot_cpu_data; struct cpuinfo_x86 *__cpu_data; #endif cpumask_t * lkpi_get_static_single_cpu_mask(int cpuid) { KASSERT((cpuid >= 0 && cpuid <= mp_maxid), ("%s: invalid cpuid %d\n", __func__, cpuid)); KASSERT(!CPU_ABSENT(cpuid), ("%s: cpu with cpuid %d is absent\n", __func__, cpuid)); return (static_single_cpu_mask[cpuid]); } bool lkpi_xen_initial_domain(void) { #ifdef XENHVM return (xen_initial_domain()); #else return (false); #endif } bool lkpi_xen_pv_domain(void) { #ifdef XENHVM return (xen_pv_domain()); #else return (false); #endif } static void linux_compat_init(void *arg) { struct sysctl_oid *rootoid; int i; #if defined(__i386__) || defined(__amd64__) static const uint32_t x86_vendors[X86_VENDOR_NUM] = { [X86_VENDOR_INTEL] = CPU_VENDOR_INTEL, [X86_VENDOR_CYRIX] = CPU_VENDOR_CYRIX, [X86_VENDOR_AMD] = CPU_VENDOR_AMD, [X86_VENDOR_UMC] = CPU_VENDOR_UMC, [X86_VENDOR_CENTAUR] = CPU_VENDOR_CENTAUR, [X86_VENDOR_TRANSMETA] = CPU_VENDOR_TRANSMETA, [X86_VENDOR_NSC] = CPU_VENDOR_NSC, [X86_VENDOR_HYGON] = CPU_VENDOR_HYGON, }; uint8_t x86_vendor = X86_VENDOR_UNKNOWN; for (i = 0; i < X86_VENDOR_NUM; i++) { if (cpu_vendor_id != 0 && cpu_vendor_id == x86_vendors[i]) { x86_vendor = i; break; } } linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH); boot_cpu_data.x86_clflush_size = cpu_clflush_line_size; boot_cpu_data.x86_max_cores = mp_ncpus; boot_cpu_data.x86 = CPUID_TO_FAMILY(cpu_id); boot_cpu_data.x86_model = CPUID_TO_MODEL(cpu_id); boot_cpu_data.x86_vendor = x86_vendor; __cpu_data = mallocarray(mp_maxid + 1, sizeof(*__cpu_data), M_KMALLOC, M_WAITOK | M_ZERO); CPU_FOREACH(i) { __cpu_data[i].x86_clflush_size = cpu_clflush_line_size; __cpu_data[i].x86_max_cores = mp_ncpus; __cpu_data[i].x86 = CPUID_TO_FAMILY(cpu_id); __cpu_data[i].x86_model = CPUID_TO_MODEL(cpu_id); __cpu_data[i].x86_vendor = x86_vendor; } #endif rw_init(&linux_vma_lock, "lkpi-vma-lock"); rootoid = SYSCTL_ADD_ROOT_NODE(NULL, OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys"); kobject_init(&linux_class_root, &linux_class_ktype); kobject_set_name(&linux_class_root, "class"); linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid), OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class"); kobject_init(&linux_root_device.kobj, &linux_dev_ktype); kobject_set_name(&linux_root_device.kobj, "device"); linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device"); linux_root_device.bsddev = root_bus; linux_class_misc.name = "misc"; class_register(&linux_class_misc); INIT_LIST_HEAD(&pci_drivers); INIT_LIST_HEAD(&pci_devices); spin_lock_init(&pci_lock); mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF); for (i = 0; i < VMMAP_HASH_SIZE; i++) LIST_INIT(&vmmaphead[i]); init_waitqueue_head(&linux_bit_waitq); init_waitqueue_head(&linux_var_waitq); CPU_COPY(&all_cpus, &cpu_online_mask); /* * Generate a single-CPU cpumask_t for each CPU (possibly) in the system. * CPUs are indexed from 0..(mp_maxid). The entry for cpuid 0 will only * have itself in the cpumask, cupid 1 only itself on entry 1, and so on. * This is used by cpumask_of() (and possibly others in the future) for, * e.g., drivers to pass hints to irq_set_affinity_hint(). */ static_single_cpu_mask = mallocarray(mp_maxid + 1, sizeof(static_single_cpu_mask), M_KMALLOC, M_WAITOK | M_ZERO); /* * When the number of CPUs reach a threshold, we start to save memory * given the sets are static by overlapping those having their single * bit set at same position in a bitset word. Asymptotically, this * regular scheme is in O(n²) whereas the overlapping one is in O(n) * only with n being the maximum number of CPUs, so the gain will become * huge quite quickly. The threshold for 64-bit architectures is 128 * CPUs. */ if (mp_ncpus < (2 * _BITSET_BITS)) { cpumask_t *sscm_ptr; /* * This represents 'mp_ncpus * __bitset_words(CPU_SETSIZE) * * (_BITSET_BITS / 8)' bytes (for comparison with the * overlapping scheme). */ static_single_cpu_mask_lcs = mallocarray(mp_ncpus, sizeof(*static_single_cpu_mask_lcs), M_KMALLOC, M_WAITOK | M_ZERO); sscm_ptr = static_single_cpu_mask_lcs; CPU_FOREACH(i) { static_single_cpu_mask[i] = sscm_ptr++; CPU_SET(i, static_single_cpu_mask[i]); } } else { /* Pointer to a bitset word. */ __typeof(((cpuset_t *)NULL)->__bits[0]) *bwp; /* * Allocate memory for (static) spans of 'cpumask_t' ('cpuset_t' * really) with a single bit set that can be reused for all * single CPU masks by making them start at different offsets. * We need '__bitset_words(CPU_SETSIZE) - 1' bitset words before * the word having its single bit set, and the same amount * after. */ static_single_cpu_mask_lcs = mallocarray(_BITSET_BITS, (2 * __bitset_words(CPU_SETSIZE) - 1) * (_BITSET_BITS / 8), M_KMALLOC, M_WAITOK | M_ZERO); /* * We rely below on cpuset_t and the bitset generic * implementation assigning words in the '__bits' array in the * same order of bits (i.e., little-endian ordering, not to be * confused with machine endianness, which concerns bits in * words and other integers). This is an imperfect test, but it * will detect a change to big-endian ordering. */ _Static_assert( __bitset_word(_BITSET_BITS + 1, _BITSET_BITS) == 1, "Assumes a bitset implementation that is little-endian " "on its words"); /* Initialize the single bit of each static span. */ bwp = (__typeof(bwp))static_single_cpu_mask_lcs + (__bitset_words(CPU_SETSIZE) - 1); for (i = 0; i < _BITSET_BITS; i++) { CPU_SET(i, (cpuset_t *)bwp); bwp += (2 * __bitset_words(CPU_SETSIZE) - 1); } /* * Finally set all CPU masks to the proper word in their * relevant span. */ CPU_FOREACH(i) { bwp = (__typeof(bwp))static_single_cpu_mask_lcs; /* Find the non-zero word of the relevant span. */ bwp += (2 * __bitset_words(CPU_SETSIZE) - 1) * (i % _BITSET_BITS) + __bitset_words(CPU_SETSIZE) - 1; /* Shift to find the CPU mask start. */ bwp -= (i / _BITSET_BITS); static_single_cpu_mask[i] = (cpuset_t *)bwp; } } strlcpy(init_uts_ns.name.release, osrelease, sizeof(init_uts_ns.name.release)); } SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL); static void linux_compat_uninit(void *arg) { linux_kobject_kfree_name(&linux_class_root); linux_kobject_kfree_name(&linux_root_device.kobj); linux_kobject_kfree_name(&linux_class_misc.kobj); free(static_single_cpu_mask_lcs, M_KMALLOC); free(static_single_cpu_mask, M_KMALLOC); #if defined(__i386__) || defined(__amd64__) free(__cpu_data, M_KMALLOC); #endif mtx_destroy(&vmmaplock); spin_lock_destroy(&pci_lock); rw_destroy(&linux_vma_lock); } SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL); /* * NOTE: Linux frequently uses "unsigned long" for pointer to integer * conversion and vice versa, where in FreeBSD "uintptr_t" would be * used. Assert these types have the same size, else some parts of the * LinuxKPI may not work like expected: */ CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t)); diff --git a/sys/compat/linuxkpi/common/src/linux_schedule.c b/sys/compat/linuxkpi/common/src/linux_schedule.c index 3349a4aa8d72..fa20a11f5ec7 100644 --- a/sys/compat/linuxkpi/common/src/linux_schedule.c +++ b/sys/compat/linuxkpi/common/src/linux_schedule.c @@ -1,423 +1,417 @@ /*- * Copyright (c) 2017 Mark Johnston * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conds * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conds, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conds and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include #include #include #include #include #include #include #include #include #include #include #include static int linux_add_to_sleepqueue(void *wchan, struct task_struct *task, const char *wmesg, int timeout, int state) { int flags, ret; MPASS((state & ~(TASK_PARKED | TASK_NORMAL)) == 0); flags = SLEEPQ_SLEEP | ((state & TASK_INTERRUPTIBLE) != 0 ? SLEEPQ_INTERRUPTIBLE : 0); sleepq_add(wchan, NULL, wmesg, flags, 0); if (timeout != 0) sleepq_set_timeout(wchan, timeout); DROP_GIANT(); if ((state & TASK_INTERRUPTIBLE) != 0) { if (timeout == 0) ret = -sleepq_wait_sig(wchan, 0); else ret = -sleepq_timedwait_sig(wchan, 0); } else { if (timeout == 0) { sleepq_wait(wchan, 0); ret = 0; } else ret = -sleepq_timedwait(wchan, 0); } PICKUP_GIANT(); /* filter return value */ if (ret != 0 && ret != -EWOULDBLOCK) { linux_schedule_save_interrupt_value(task, ret); ret = -ERESTARTSYS; } return (ret); } unsigned int linux_msleep_interruptible(unsigned int ms) { int ret; /* guard against invalid values */ if (ms == 0) ms = 1; ret = -pause_sbt("lnxsleep", mstosbt(ms), 0, C_HARDCLOCK | C_CATCH); switch (ret) { case -EWOULDBLOCK: return (0); default: linux_schedule_save_interrupt_value(current, ret); return (ms); } } static int wake_up_task(struct task_struct *task, unsigned int state) { - int ret, wakeup_swapper; + int ret; - ret = wakeup_swapper = 0; + ret = 0; sleepq_lock(task); if ((atomic_read(&task->state) & state) != 0) { set_task_state(task, TASK_WAKING); - wakeup_swapper = sleepq_signal(task, SLEEPQ_SLEEP, 0, 0); + sleepq_signal(task, SLEEPQ_SLEEP, 0, 0); ret = 1; } sleepq_release(task); - if (wakeup_swapper) - kick_proc0(); return (ret); } bool linux_signal_pending(struct task_struct *task) { struct thread *td; sigset_t pending; td = task->task_thread; PROC_LOCK(td->td_proc); pending = td->td_siglist; SIGSETOR(pending, td->td_proc->p_siglist); SIGSETNAND(pending, td->td_sigmask); PROC_UNLOCK(td->td_proc); return (!SIGISEMPTY(pending)); } bool linux_fatal_signal_pending(struct task_struct *task) { struct thread *td; bool ret; td = task->task_thread; PROC_LOCK(td->td_proc); ret = SIGISMEMBER(td->td_siglist, SIGKILL) || SIGISMEMBER(td->td_proc->p_siglist, SIGKILL); PROC_UNLOCK(td->td_proc); return (ret); } bool linux_signal_pending_state(long state, struct task_struct *task) { MPASS((state & ~TASK_NORMAL) == 0); if ((state & TASK_INTERRUPTIBLE) == 0) return (false); return (linux_signal_pending(task)); } void linux_send_sig(int signo, struct task_struct *task) { struct thread *td; td = task->task_thread; PROC_LOCK(td->td_proc); tdsignal(td, signo); PROC_UNLOCK(td->td_proc); } int autoremove_wake_function(wait_queue_t *wq, unsigned int state, int flags, void *key __unused) { struct task_struct *task; int ret; task = wq->private; if ((ret = wake_up_task(task, state)) != 0) list_del_init(&wq->task_list); return (ret); } int default_wake_function(wait_queue_t *wq, unsigned int state, int flags, void *key __unused) { return (wake_up_task(wq->private, state)); } void linux_init_wait_entry(wait_queue_t *wq, int flags) { memset(wq, 0, sizeof(*wq)); wq->flags = flags; wq->private = current; wq->func = autoremove_wake_function; INIT_LIST_HEAD(&wq->task_list); } void linux_wake_up(wait_queue_head_t *wqh, unsigned int state, int nr, bool locked) { wait_queue_t *pos, *next; if (!locked) spin_lock(&wqh->lock); list_for_each_entry_safe(pos, next, &wqh->task_list, task_list) { if (pos->func == NULL) { if (wake_up_task(pos->private, state) != 0 && --nr == 0) break; } else { if (pos->func(pos, state, 0, NULL) != 0 && --nr == 0) break; } } if (!locked) spin_unlock(&wqh->lock); } void linux_prepare_to_wait(wait_queue_head_t *wqh, wait_queue_t *wq, int state) { spin_lock(&wqh->lock); if (list_empty(&wq->task_list)) __add_wait_queue(wqh, wq); set_task_state(current, state); spin_unlock(&wqh->lock); } void linux_finish_wait(wait_queue_head_t *wqh, wait_queue_t *wq) { spin_lock(&wqh->lock); set_task_state(current, TASK_RUNNING); if (!list_empty(&wq->task_list)) { __remove_wait_queue(wqh, wq); INIT_LIST_HEAD(&wq->task_list); } spin_unlock(&wqh->lock); } bool linux_waitqueue_active(wait_queue_head_t *wqh) { bool ret; spin_lock(&wqh->lock); ret = !list_empty(&wqh->task_list); spin_unlock(&wqh->lock); return (ret); } int linux_wait_event_common(wait_queue_head_t *wqh, wait_queue_t *wq, int timeout, unsigned int state, spinlock_t *lock) { struct task_struct *task; int ret; if (lock != NULL) spin_unlock_irq(lock); /* range check timeout */ if (timeout < 1) timeout = 1; else if (timeout == MAX_SCHEDULE_TIMEOUT) timeout = 0; task = current; sleepq_lock(task); if (atomic_read(&task->state) != TASK_WAKING) { ret = linux_add_to_sleepqueue(task, task, "wevent", timeout, state); } else { sleepq_release(task); ret = 0; } if (lock != NULL) spin_lock_irq(lock); return (ret); } int linux_schedule_timeout(int timeout) { struct task_struct *task; int ret; int state; int remainder; task = current; /* range check timeout */ if (timeout < 1) timeout = 1; else if (timeout == MAX_SCHEDULE_TIMEOUT) timeout = 0; remainder = ticks + timeout; sleepq_lock(task); state = atomic_read(&task->state); if (state != TASK_WAKING) { ret = linux_add_to_sleepqueue(task, task, "sched", timeout, state); } else { sleepq_release(task); ret = 0; } set_task_state(task, TASK_RUNNING); if (timeout == 0) return (MAX_SCHEDULE_TIMEOUT); /* range check return value */ remainder -= ticks; /* range check return value */ if (ret == -ERESTARTSYS && remainder < 1) remainder = 1; else if (remainder < 0) remainder = 0; else if (remainder > timeout) remainder = timeout; return (remainder); } static void wake_up_sleepers(void *wchan) { - int wakeup_swapper; - sleepq_lock(wchan); - wakeup_swapper = sleepq_signal(wchan, SLEEPQ_SLEEP, 0, 0); + sleepq_signal(wchan, SLEEPQ_SLEEP, 0, 0); sleepq_release(wchan); - if (wakeup_swapper) - kick_proc0(); } #define bit_to_wchan(word, bit) ((void *)(((uintptr_t)(word) << 6) | (bit))) void linux_wake_up_bit(void *word, int bit) { wake_up_sleepers(bit_to_wchan(word, bit)); } int linux_wait_on_bit_timeout(unsigned long *word, int bit, unsigned int state, int timeout) { struct task_struct *task; void *wchan; int ret; /* range check timeout */ if (timeout < 1) timeout = 1; else if (timeout == MAX_SCHEDULE_TIMEOUT) timeout = 0; task = current; wchan = bit_to_wchan(word, bit); for (;;) { sleepq_lock(wchan); if ((*word & (1 << bit)) == 0) { sleepq_release(wchan); ret = 0; break; } set_task_state(task, state); ret = linux_add_to_sleepqueue(wchan, task, "wbit", timeout, state); if (ret != 0) break; } set_task_state(task, TASK_RUNNING); return (ret); } void linux_wake_up_atomic_t(atomic_t *a) { wake_up_sleepers(a); } int linux_wait_on_atomic_t(atomic_t *a, unsigned int state) { struct task_struct *task; void *wchan; int ret; task = current; wchan = a; for (;;) { sleepq_lock(wchan); if (atomic_read(a) == 0) { sleepq_release(wchan); ret = 0; break; } set_task_state(task, state); ret = linux_add_to_sleepqueue(wchan, task, "watomic", 0, state); if (ret != 0) break; } set_task_state(task, TASK_RUNNING); return (ret); } bool linux_wake_up_state(struct task_struct *task, unsigned int state) { return (wake_up_task(task, state) != 0); } diff --git a/sys/dev/qat/qat_common/adf_aer.c b/sys/dev/qat/qat_common/adf_aer.c index 7fdeba873420..6be8ab04bbd7 100644 --- a/sys/dev/qat/qat_common/adf_aer.c +++ b/sys/dev/qat/qat_common/adf_aer.c @@ -1,339 +1,335 @@ /* SPDX-License-Identifier: BSD-3-Clause */ /* Copyright(c) 2007-2022 Intel Corporation */ #include "qat_freebsd.h" #include "adf_cfg.h" #include "adf_common_drv.h" #include "adf_accel_devices.h" #include "icp_qat_uclo.h" #include "icp_qat_fw.h" #include "icp_qat_fw_init_admin.h" #include "adf_cfg_strings.h" #include "adf_transport_access_macros.h" #include "adf_transport_internal.h" #include #include #include #include #define ADF_PPAERUCM_MASK (BIT(14) | BIT(20) | BIT(22)) static struct workqueue_struct *fatal_error_wq; struct adf_fatal_error_data { struct adf_accel_dev *accel_dev; struct work_struct work; }; static struct workqueue_struct *device_reset_wq; void linux_complete_common(struct completion *c, int all) { - int wakeup_swapper; - sleepq_lock(c); c->done++; if (all) - wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0); + sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0); else - wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0); + sleepq_signal(c, SLEEPQ_SLEEP, 0, 0); sleepq_release(c); - if (wakeup_swapper) - kick_proc0(); } /* reset dev data */ struct adf_reset_dev_data { int mode; struct adf_accel_dev *accel_dev; struct completion compl; struct work_struct reset_work; }; int adf_aer_store_ppaerucm_reg(device_t dev, struct adf_hw_device_data *hw_data) { unsigned int aer_offset, reg_val = 0; if (!hw_data) return -EINVAL; if (pci_find_extcap(dev, PCIZ_AER, &aer_offset) == 0) { reg_val = pci_read_config(dev, aer_offset + PCIR_AER_UC_MASK, 4); hw_data->aerucm_mask = reg_val; } else { device_printf(dev, "Unable to find AER capability of the device\n"); return -ENODEV; } return 0; } void adf_reset_sbr(struct adf_accel_dev *accel_dev) { device_t pdev = accel_to_pci_dev(accel_dev); device_t parent = device_get_parent(device_get_parent(pdev)); uint16_t bridge_ctl = 0; if (accel_dev->is_vf) return; if (!parent) parent = pdev; if (!pcie_wait_for_pending_transactions(pdev, 0)) device_printf(GET_DEV(accel_dev), "Transaction still in progress. Proceeding\n"); device_printf(GET_DEV(accel_dev), "Secondary bus reset\n"); pci_save_state(pdev); bridge_ctl = pci_read_config(parent, PCIR_BRIDGECTL_1, 2); bridge_ctl |= PCIB_BCR_SECBUS_RESET; pci_write_config(parent, PCIR_BRIDGECTL_1, bridge_ctl, 2); pause_ms("adfrst", 100); bridge_ctl &= ~PCIB_BCR_SECBUS_RESET; pci_write_config(parent, PCIR_BRIDGECTL_1, bridge_ctl, 2); pause_ms("adfrst", 100); pci_restore_state(pdev); } void adf_reset_flr(struct adf_accel_dev *accel_dev) { device_t pdev = accel_to_pci_dev(accel_dev); pci_save_state(pdev); if (pcie_flr(pdev, max(pcie_get_max_completion_timeout(pdev) / 1000, 10), true)) { pci_restore_state(pdev); return; } pci_restore_state(pdev); device_printf(GET_DEV(accel_dev), "FLR qat_dev%d failed trying secondary bus reset\n", accel_dev->accel_id); adf_reset_sbr(accel_dev); } void adf_dev_pre_reset(struct adf_accel_dev *accel_dev) { struct adf_hw_device_data *hw_device = accel_dev->hw_device; device_t pdev = accel_to_pci_dev(accel_dev); u32 aer_offset, reg_val = 0; if (pci_find_extcap(pdev, PCIZ_AER, &aer_offset) == 0) { reg_val = pci_read_config(pdev, aer_offset + PCIR_AER_UC_MASK, 4); reg_val |= ADF_PPAERUCM_MASK; pci_write_config(pdev, aer_offset + PCIR_AER_UC_MASK, reg_val, 4); } else { device_printf(pdev, "Unable to find AER capability of the device\n"); } if (hw_device->disable_arb) { device_printf(GET_DEV(accel_dev), "Disable arbiter.\n"); hw_device->disable_arb(accel_dev); } } void adf_dev_post_reset(struct adf_accel_dev *accel_dev) { struct adf_hw_device_data *hw_device = accel_dev->hw_device; device_t pdev = accel_to_pci_dev(accel_dev); u32 aer_offset; if (pci_find_extcap(pdev, PCIZ_AER, &aer_offset) == 0) { pci_write_config(pdev, aer_offset + PCIR_AER_UC_MASK, hw_device->aerucm_mask, 4); } else { device_printf(pdev, "Unable to find AER capability of the device\n"); } } void adf_dev_restore(struct adf_accel_dev *accel_dev) { struct adf_hw_device_data *hw_device = accel_dev->hw_device; device_t pdev = accel_to_pci_dev(accel_dev); if (hw_device->pre_reset) { dev_dbg(GET_DEV(accel_dev), "Performing pre reset save\n"); hw_device->pre_reset(accel_dev); } if (hw_device->reset_device) { device_printf(GET_DEV(accel_dev), "Resetting device qat_dev%d\n", accel_dev->accel_id); hw_device->reset_device(accel_dev); pci_restore_state(pdev); pci_save_state(pdev); } if (hw_device->post_reset) { dev_dbg(GET_DEV(accel_dev), "Performing post reset restore\n"); hw_device->post_reset(accel_dev); } } static void adf_device_reset_worker(struct work_struct *work) { struct adf_reset_dev_data *reset_data = container_of(work, struct adf_reset_dev_data, reset_work); struct adf_accel_dev *accel_dev = reset_data->accel_dev; if (adf_dev_restarting_notify(accel_dev)) { device_printf(GET_DEV(accel_dev), "Unable to send RESTARTING notification.\n"); return; } if (adf_dev_stop(accel_dev)) { device_printf(GET_DEV(accel_dev), "Stopping device failed.\n"); return; } adf_dev_shutdown(accel_dev); if (adf_dev_init(accel_dev) || adf_dev_start(accel_dev)) { /* The device hanged and we can't restart it */ /* so stop here */ device_printf(GET_DEV(accel_dev), "Restart device failed\n"); if (reset_data->mode == ADF_DEV_RESET_ASYNC) kfree(reset_data); WARN(1, "QAT: device restart failed. Device is unusable\n"); return; } adf_dev_restarted_notify(accel_dev); clear_bit(ADF_STATUS_RESTARTING, &accel_dev->status); /* The dev is back alive. Notify the caller if in sync mode */ if (reset_data->mode == ADF_DEV_RESET_SYNC) complete(&reset_data->compl); else kfree(reset_data); } int adf_dev_aer_schedule_reset(struct adf_accel_dev *accel_dev, enum adf_dev_reset_mode mode) { struct adf_reset_dev_data *reset_data; if (!adf_dev_started(accel_dev) || test_bit(ADF_STATUS_RESTARTING, &accel_dev->status)) return 0; set_bit(ADF_STATUS_RESTARTING, &accel_dev->status); reset_data = kzalloc(sizeof(*reset_data), GFP_ATOMIC); if (!reset_data) return -ENOMEM; reset_data->accel_dev = accel_dev; init_completion(&reset_data->compl); reset_data->mode = mode; INIT_WORK(&reset_data->reset_work, adf_device_reset_worker); queue_work(device_reset_wq, &reset_data->reset_work); /* If in sync mode wait for the result */ if (mode == ADF_DEV_RESET_SYNC) { int ret = 0; /* Maximum device reset time is 10 seconds */ unsigned long wait_jiffies = msecs_to_jiffies(10000); unsigned long timeout = wait_for_completion_timeout(&reset_data->compl, wait_jiffies); if (!timeout) { device_printf(GET_DEV(accel_dev), "Reset device timeout expired\n"); ret = -EFAULT; } kfree(reset_data); return ret; } return 0; } int adf_dev_autoreset(struct adf_accel_dev *accel_dev) { if (accel_dev->autoreset_on_error) return adf_dev_reset(accel_dev, ADF_DEV_RESET_ASYNC); return 0; } static void adf_notify_fatal_error_work(struct work_struct *work) { struct adf_fatal_error_data *wq_data = container_of(work, struct adf_fatal_error_data, work); struct adf_accel_dev *accel_dev = wq_data->accel_dev; adf_error_notifier((uintptr_t)accel_dev); if (!accel_dev->is_vf) { adf_dev_autoreset(accel_dev); } kfree(wq_data); } int adf_notify_fatal_error(struct adf_accel_dev *accel_dev) { struct adf_fatal_error_data *wq_data; wq_data = kzalloc(sizeof(*wq_data), GFP_ATOMIC); if (!wq_data) { device_printf(GET_DEV(accel_dev), "Failed to allocate memory\n"); return ENOMEM; } wq_data->accel_dev = accel_dev; INIT_WORK(&wq_data->work, adf_notify_fatal_error_work); queue_work(fatal_error_wq, &wq_data->work); return 0; } int __init adf_init_fatal_error_wq(void) { fatal_error_wq = create_workqueue("qat_fatal_error_wq"); return !fatal_error_wq ? EFAULT : 0; } void adf_exit_fatal_error_wq(void) { if (fatal_error_wq) destroy_workqueue(fatal_error_wq); fatal_error_wq = NULL; } int adf_init_aer(void) { device_reset_wq = create_workqueue("qat_device_reset_wq"); return !device_reset_wq ? -EFAULT : 0; } void adf_exit_aer(void) { if (device_reset_wq) destroy_workqueue(device_reset_wq); device_reset_wq = NULL; } diff --git a/sys/kern/kern_condvar.c b/sys/kern/kern_condvar.c index a55863595ce2..517b83d90243 100644 --- a/sys/kern/kern_condvar.c +++ b/sys/kern/kern_condvar.c @@ -1,484 +1,475 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2000 Jake Burkholder . * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #include #endif /* * A bound below which cv_waiters is valid. Once cv_waiters reaches this bound, * cv_signal must manually check the wait queue for threads. */ #define CV_WAITERS_BOUND INT_MAX #define CV_WAITERS_INC(cvp) do { \ if ((cvp)->cv_waiters < CV_WAITERS_BOUND) \ (cvp)->cv_waiters++; \ } while (0) /* * Common sanity checks for cv_wait* functions. */ #define CV_ASSERT(cvp, lock, td) do { \ KASSERT((td) != NULL, ("%s: td NULL", __func__)); \ KASSERT(TD_IS_RUNNING(td), ("%s: not TDS_RUNNING", __func__)); \ KASSERT((cvp) != NULL, ("%s: cvp NULL", __func__)); \ KASSERT((lock) != NULL, ("%s: lock NULL", __func__)); \ } while (0) /* * Initialize a condition variable. Must be called before use. */ void cv_init(struct cv *cvp, const char *desc) { cvp->cv_description = desc; cvp->cv_waiters = 0; } /* * Destroy a condition variable. The condition variable must be re-initialized * in order to be re-used. */ void cv_destroy(struct cv *cvp) { #ifdef INVARIANTS struct sleepqueue *sq; sleepq_lock(cvp); sq = sleepq_lookup(cvp); sleepq_release(cvp); KASSERT(sq == NULL, ("%s: associated sleep queue non-empty", __func__)); #endif } /* * Wait on a condition variable. The current thread is placed on the condition * variable's wait queue and suspended. A cv_signal or cv_broadcast on the same * condition variable will resume the thread. The mutex is released before * sleeping and will be held on return. It is recommended that the mutex be * held when cv_signal or cv_broadcast are called. */ void _cv_wait(struct cv *cvp, struct lock_object *lock) { WITNESS_SAVE_DECL(lock_witness); #ifdef KTRACE char wmesg[WMESGLEN + 1]; #endif struct lock_class *class; struct thread *td __ktrace_used; uintptr_t lock_state; td = curthread; CV_ASSERT(cvp, lock, td); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, "Waiting on \"%s\"", cvp->cv_description); if (SCHEDULER_STOPPED()) return; #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) { strlcpy(wmesg, cv_wmesg(cvp), sizeof(wmesg)); ktrcsw(1, 0, wmesg); } else { wmesg[0] = '\0'; } #endif class = LOCK_CLASS(lock); lock_state = 0; sleepq_lock(cvp); CV_WAITERS_INC(cvp); if (lock == &Giant.lock_object) mtx_assert(&Giant, MA_OWNED); DROP_GIANT(); sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR, 0); if (lock != &Giant.lock_object) { if (class->lc_flags & LC_SLEEPABLE) sleepq_release(cvp); WITNESS_SAVE(lock, lock_witness); lock_state = class->lc_unlock(lock); if (class->lc_flags & LC_SLEEPABLE) sleepq_lock(cvp); } sleepq_wait(cvp, 0); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 0, wmesg); #endif PICKUP_GIANT(); if (lock != &Giant.lock_object) { class->lc_lock(lock, lock_state); WITNESS_RESTORE(lock, lock_witness); } } /* * Wait on a condition variable. This function differs from cv_wait by * not acquiring the mutex after condition variable was signaled. */ void _cv_wait_unlock(struct cv *cvp, struct lock_object *lock) { #ifdef KTRACE char wmesg[WMESGLEN + 1]; #endif struct lock_class *class; struct thread *td __ktrace_used; td = curthread; CV_ASSERT(cvp, lock, td); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, "Waiting on \"%s\"", cvp->cv_description); KASSERT(lock != &Giant.lock_object, ("cv_wait_unlock cannot be used with Giant")); class = LOCK_CLASS(lock); if (SCHEDULER_STOPPED()) { class->lc_unlock(lock); return; } #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) { strlcpy(wmesg, cv_wmesg(cvp), sizeof(wmesg)); ktrcsw(1, 0, wmesg); } else { wmesg[0] = '\0'; } #endif sleepq_lock(cvp); CV_WAITERS_INC(cvp); DROP_GIANT(); sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR, 0); if (class->lc_flags & LC_SLEEPABLE) sleepq_release(cvp); class->lc_unlock(lock); if (class->lc_flags & LC_SLEEPABLE) sleepq_lock(cvp); sleepq_wait(cvp, 0); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 0, wmesg); #endif PICKUP_GIANT(); } /* * Wait on a condition variable, allowing interruption by signals. Return 0 if * the thread was resumed with cv_signal or cv_broadcast, EINTR or ERESTART if * a signal was caught. If ERESTART is returned the system call should be * restarted if possible. */ int _cv_wait_sig(struct cv *cvp, struct lock_object *lock) { WITNESS_SAVE_DECL(lock_witness); #ifdef KTRACE char wmesg[WMESGLEN + 1]; #endif struct lock_class *class; struct thread *td __ktrace_used; uintptr_t lock_state; int rval; td = curthread; CV_ASSERT(cvp, lock, td); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, "Waiting on \"%s\"", cvp->cv_description); if (SCHEDULER_STOPPED()) return (0); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) { strlcpy(wmesg, cv_wmesg(cvp), sizeof(wmesg)); ktrcsw(1, 0, wmesg); } else { wmesg[0] = '\0'; } #endif class = LOCK_CLASS(lock); lock_state = 0; sleepq_lock(cvp); CV_WAITERS_INC(cvp); if (lock == &Giant.lock_object) mtx_assert(&Giant, MA_OWNED); DROP_GIANT(); sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR | SLEEPQ_INTERRUPTIBLE, 0); if (lock != &Giant.lock_object) { if (class->lc_flags & LC_SLEEPABLE) sleepq_release(cvp); WITNESS_SAVE(lock, lock_witness); lock_state = class->lc_unlock(lock); if (class->lc_flags & LC_SLEEPABLE) sleepq_lock(cvp); } rval = sleepq_wait_sig(cvp, 0); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 0, wmesg); #endif PICKUP_GIANT(); if (lock != &Giant.lock_object) { class->lc_lock(lock, lock_state); WITNESS_RESTORE(lock, lock_witness); } return (rval); } /* * Wait on a condition variable for (at most) the value specified in sbt * argument. Returns 0 if the process was resumed by cv_signal or cv_broadcast, * EWOULDBLOCK if the timeout expires. */ int _cv_timedwait_sbt(struct cv *cvp, struct lock_object *lock, sbintime_t sbt, sbintime_t pr, int flags) { WITNESS_SAVE_DECL(lock_witness); #ifdef KTRACE char wmesg[WMESGLEN + 1]; #endif struct lock_class *class; struct thread *td __ktrace_used; int lock_state, rval; td = curthread; CV_ASSERT(cvp, lock, td); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, "Waiting on \"%s\"", cvp->cv_description); if (SCHEDULER_STOPPED()) return (0); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) { strlcpy(wmesg, cv_wmesg(cvp), sizeof(wmesg)); ktrcsw(1, 0, wmesg); } else { wmesg[0] = '\0'; } #endif class = LOCK_CLASS(lock); lock_state = 0; sleepq_lock(cvp); CV_WAITERS_INC(cvp); if (lock == &Giant.lock_object) mtx_assert(&Giant, MA_OWNED); DROP_GIANT(); sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR, 0); sleepq_set_timeout_sbt(cvp, sbt, pr, flags); if (lock != &Giant.lock_object) { if (class->lc_flags & LC_SLEEPABLE) sleepq_release(cvp); WITNESS_SAVE(lock, lock_witness); lock_state = class->lc_unlock(lock); if (class->lc_flags & LC_SLEEPABLE) sleepq_lock(cvp); } rval = sleepq_timedwait(cvp, 0); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 0, wmesg); #endif PICKUP_GIANT(); if (lock != &Giant.lock_object) { class->lc_lock(lock, lock_state); WITNESS_RESTORE(lock, lock_witness); } return (rval); } /* * Wait on a condition variable for (at most) the value specified in sbt * argument, allowing interruption by signals. * Returns 0 if the thread was resumed by cv_signal or cv_broadcast, * EWOULDBLOCK if the timeout expires, and EINTR or ERESTART if a signal * was caught. */ int _cv_timedwait_sig_sbt(struct cv *cvp, struct lock_object *lock, sbintime_t sbt, sbintime_t pr, int flags) { WITNESS_SAVE_DECL(lock_witness); #ifdef KTRACE char wmesg[WMESGLEN + 1]; #endif struct lock_class *class; struct thread *td __ktrace_used; int lock_state, rval; td = curthread; CV_ASSERT(cvp, lock, td); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, "Waiting on \"%s\"", cvp->cv_description); if (SCHEDULER_STOPPED()) return (0); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) { strlcpy(wmesg, cv_wmesg(cvp), sizeof(wmesg)); ktrcsw(1, 0, wmesg); } else { wmesg[0] = '\0'; } #endif class = LOCK_CLASS(lock); lock_state = 0; sleepq_lock(cvp); CV_WAITERS_INC(cvp); if (lock == &Giant.lock_object) mtx_assert(&Giant, MA_OWNED); DROP_GIANT(); sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR | SLEEPQ_INTERRUPTIBLE, 0); sleepq_set_timeout_sbt(cvp, sbt, pr, flags); if (lock != &Giant.lock_object) { if (class->lc_flags & LC_SLEEPABLE) sleepq_release(cvp); WITNESS_SAVE(lock, lock_witness); lock_state = class->lc_unlock(lock); if (class->lc_flags & LC_SLEEPABLE) sleepq_lock(cvp); } rval = sleepq_timedwait_sig(cvp, 0); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 0, wmesg); #endif PICKUP_GIANT(); if (lock != &Giant.lock_object) { class->lc_lock(lock, lock_state); WITNESS_RESTORE(lock, lock_witness); } return (rval); } /* - * Signal a condition variable, wakes up one waiting thread. Will also wakeup - * the swapper if the process is not in memory, so that it can bring the - * sleeping process in. Note that this may also result in additional threads - * being made runnable. Should be called with the same mutex as was passed to - * cv_wait held. + * Signal a condition variable, wakes up one waiting thread. Note that this may + * also result in additional threads being made runnable. Should be called with + * the same mutex as was passed to cv_wait held. */ void cv_signal(struct cv *cvp) { - if (cvp->cv_waiters == 0) return; sleepq_lock(cvp); if (cvp->cv_waiters == 0) { sleepq_release(cvp); return; } if (cvp->cv_waiters == CV_WAITERS_BOUND && sleepq_lookup(cvp) == NULL) { cvp->cv_waiters = 0; sleepq_release(cvp); } else { if (cvp->cv_waiters < CV_WAITERS_BOUND) cvp->cv_waiters--; - if (sleepq_signal(cvp, SLEEPQ_CONDVAR | SLEEPQ_DROP, 0, 0)) - kick_proc0(); + sleepq_signal(cvp, SLEEPQ_CONDVAR | SLEEPQ_DROP, 0, 0); } } /* * Broadcast a signal to a condition variable. Wakes up all waiting threads. * Should be called with the same mutex as was passed to cv_wait held. */ void cv_broadcastpri(struct cv *cvp, int pri) { - int wakeup_swapper; - if (cvp->cv_waiters == 0) return; /* * XXX sleepq_broadcast pri argument changed from -1 meaning * no pri to 0 meaning no pri. */ - wakeup_swapper = 0; if (pri == -1) pri = 0; sleepq_lock(cvp); if (cvp->cv_waiters > 0) { cvp->cv_waiters = 0; - wakeup_swapper = sleepq_broadcast(cvp, SLEEPQ_CONDVAR, pri, 0); + sleepq_broadcast(cvp, SLEEPQ_CONDVAR, pri, 0); } sleepq_release(cvp); - if (wakeup_swapper) - kick_proc0(); } diff --git a/sys/kern/kern_lock.c b/sys/kern/kern_lock.c index 4700ee0f8f98..4771496f950a 100644 --- a/sys/kern/kern_lock.c +++ b/sys/kern/kern_lock.c @@ -1,1859 +1,1840 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2008 Attilio Rao * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice(s), this list of conditions and the following disclaimer as * the first lines of this file unmodified other than the possible * addition of one or more copyright notices. * 2. Redistributions in binary form must reproduce the above copyright * notice(s), this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH * DAMAGE. */ #include "opt_ddb.h" #include "opt_hwpmc_hooks.h" #include #include #include #include #include #include #include #include #include #include #include #ifdef DEBUG_LOCKS #include #endif #include #include #include #ifdef DDB #include #endif #ifdef HWPMC_HOOKS #include PMC_SOFT_DECLARE( , , lock, failed); #endif /* * Hack. There should be prio_t or similar so that this is not necessary. */ _Static_assert((PRILASTFLAG * 2) - 1 <= USHRT_MAX, "prio flags wont fit in u_short pri in struct lock"); CTASSERT(LK_UNLOCKED == (LK_UNLOCKED & ~(LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS))); #define SQ_EXCLUSIVE_QUEUE 0 #define SQ_SHARED_QUEUE 1 #ifndef INVARIANTS #define _lockmgr_assert(lk, what, file, line) #endif #define TD_SLOCKS_INC(td) ((td)->td_lk_slocks++) #define TD_SLOCKS_DEC(td) ((td)->td_lk_slocks--) #ifndef DEBUG_LOCKS #define STACK_PRINT(lk) #define STACK_SAVE(lk) #define STACK_ZERO(lk) #else #define STACK_PRINT(lk) stack_print_ddb(&(lk)->lk_stack) #define STACK_SAVE(lk) stack_save(&(lk)->lk_stack) #define STACK_ZERO(lk) stack_zero(&(lk)->lk_stack) #endif #define LOCK_LOG2(lk, string, arg1, arg2) \ if (LOCK_LOG_TEST(&(lk)->lock_object, 0)) \ CTR2(KTR_LOCK, (string), (arg1), (arg2)) #define LOCK_LOG3(lk, string, arg1, arg2, arg3) \ if (LOCK_LOG_TEST(&(lk)->lock_object, 0)) \ CTR3(KTR_LOCK, (string), (arg1), (arg2), (arg3)) #define GIANT_DECLARE \ int _i = 0; \ WITNESS_SAVE_DECL(Giant) #define GIANT_RESTORE() do { \ if (__predict_false(_i > 0)) { \ while (_i--) \ mtx_lock(&Giant); \ WITNESS_RESTORE(&Giant.lock_object, Giant); \ } \ } while (0) #define GIANT_SAVE() do { \ if (__predict_false(mtx_owned(&Giant))) { \ WITNESS_SAVE(&Giant.lock_object, Giant); \ while (mtx_owned(&Giant)) { \ _i++; \ mtx_unlock(&Giant); \ } \ } \ } while (0) static __always_inline bool LK_CAN_SHARE(uintptr_t x, int flags, bool fp) { if ((x & (LK_SHARE | LK_EXCLUSIVE_WAITERS | LK_EXCLUSIVE_SPINNERS)) == LK_SHARE) return (true); if (fp || (!(x & LK_SHARE))) return (false); if ((curthread->td_lk_slocks != 0 && !(flags & LK_NODDLKTREAT)) || (curthread->td_pflags & TDP_DEADLKTREAT)) return (true); return (false); } #define LK_TRYOP(x) \ ((x) & LK_NOWAIT) #define LK_CAN_WITNESS(x) \ (((x) & LK_NOWITNESS) == 0 && !LK_TRYOP(x)) #define LK_TRYWIT(x) \ (LK_TRYOP(x) ? LOP_TRYLOCK : 0) #define lockmgr_xlocked_v(v) \ (((v) & ~(LK_FLAGMASK & ~LK_SHARE)) == (uintptr_t)curthread) #define lockmgr_xlocked(lk) lockmgr_xlocked_v(lockmgr_read_value(lk)) static void assert_lockmgr(const struct lock_object *lock, int how); #ifdef DDB static void db_show_lockmgr(const struct lock_object *lock); #endif static void lock_lockmgr(struct lock_object *lock, uintptr_t how); #ifdef KDTRACE_HOOKS static int owner_lockmgr(const struct lock_object *lock, struct thread **owner); #endif static uintptr_t unlock_lockmgr(struct lock_object *lock); struct lock_class lock_class_lockmgr = { .lc_name = "lockmgr", .lc_flags = LC_RECURSABLE | LC_SLEEPABLE | LC_SLEEPLOCK | LC_UPGRADABLE, .lc_assert = assert_lockmgr, #ifdef DDB .lc_ddb_show = db_show_lockmgr, #endif .lc_lock = lock_lockmgr, .lc_unlock = unlock_lockmgr, #ifdef KDTRACE_HOOKS .lc_owner = owner_lockmgr, #endif }; static __read_mostly bool lk_adaptive = true; static SYSCTL_NODE(_debug, OID_AUTO, lockmgr, CTLFLAG_RD, NULL, "lockmgr debugging"); SYSCTL_BOOL(_debug_lockmgr, OID_AUTO, adaptive_spinning, CTLFLAG_RW, &lk_adaptive, 0, ""); #define lockmgr_delay locks_delay struct lockmgr_wait { const char *iwmesg; int ipri; int itimo; }; static __always_inline bool lockmgr_slock_try(struct lock *lk, uintptr_t *xp, int flags, bool fp); static __always_inline bool lockmgr_sunlock_try(struct lock *lk, uintptr_t *xp); static void -lockmgr_exit(u_int flags, struct lock_object *ilk, int wakeup_swapper) +lockmgr_exit(u_int flags, struct lock_object *ilk) { struct lock_class *class; if (flags & LK_INTERLOCK) { class = LOCK_CLASS(ilk); class->lc_unlock(ilk); } - - if (__predict_false(wakeup_swapper)) - kick_proc0(); } static void lockmgr_note_shared_acquire(struct lock *lk, int contested, uint64_t waittime, const char *file, int line, int flags) { LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(lockmgr__acquire, lk, contested, waittime, file, line, LOCKSTAT_READER); LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, file, line); WITNESS_LOCK(&lk->lock_object, LK_TRYWIT(flags), file, line); TD_LOCKS_INC(curthread); TD_SLOCKS_INC(curthread); STACK_SAVE(lk); } static void lockmgr_note_shared_release(struct lock *lk, const char *file, int line) { WITNESS_UNLOCK(&lk->lock_object, 0, file, line); LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, file, line); TD_LOCKS_DEC(curthread); TD_SLOCKS_DEC(curthread); } static void lockmgr_note_exclusive_acquire(struct lock *lk, int contested, uint64_t waittime, const char *file, int line, int flags) { LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(lockmgr__acquire, lk, contested, waittime, file, line, LOCKSTAT_WRITER); LOCK_LOG_LOCK("XLOCK", &lk->lock_object, 0, lk->lk_recurse, file, line); WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE | LK_TRYWIT(flags), file, line); TD_LOCKS_INC(curthread); STACK_SAVE(lk); } static void lockmgr_note_exclusive_release(struct lock *lk, const char *file, int line) { if (!lockmgr_disowned(lk)) { WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE, file, line); TD_LOCKS_DEC(curthread); } LOCK_LOG_LOCK("XUNLOCK", &lk->lock_object, 0, lk->lk_recurse, file, line); } static __inline struct thread * lockmgr_xholder(const struct lock *lk) { uintptr_t x; x = lockmgr_read_value(lk); return ((x & LK_SHARE) ? NULL : (struct thread *)LK_HOLDER(x)); } /* * It assumes sleepq_lock held and returns with this one unheld. * It also assumes the generic interlock is sane and previously checked. * If LK_INTERLOCK is specified the interlock is not reacquired after the * sleep. */ static __inline int sleeplk(struct lock *lk, u_int flags, struct lock_object *ilk, const char *wmesg, int pri, int timo, int queue) { GIANT_DECLARE; struct lock_class *class; int catch, error; class = (flags & LK_INTERLOCK) ? LOCK_CLASS(ilk) : NULL; catch = pri & PCATCH; pri &= PRIMASK; error = 0; LOCK_LOG3(lk, "%s: %p blocking on the %s sleepqueue", __func__, lk, (queue == SQ_EXCLUSIVE_QUEUE) ? "exclusive" : "shared"); if (flags & LK_INTERLOCK) class->lc_unlock(ilk); if (queue == SQ_EXCLUSIVE_QUEUE && (flags & LK_SLEEPFAIL) != 0) { if (lk->lk_exslpfail < USHRT_MAX) lk->lk_exslpfail++; } GIANT_SAVE(); sleepq_add(&lk->lock_object, NULL, wmesg, SLEEPQ_LK | (catch ? SLEEPQ_INTERRUPTIBLE : 0), queue); if ((flags & LK_TIMELOCK) && timo) sleepq_set_timeout(&lk->lock_object, timo); /* * Decisional switch for real sleeping. */ if ((flags & LK_TIMELOCK) && timo && catch) error = sleepq_timedwait_sig(&lk->lock_object, pri); else if ((flags & LK_TIMELOCK) && timo) error = sleepq_timedwait(&lk->lock_object, pri); else if (catch) error = sleepq_wait_sig(&lk->lock_object, pri); else sleepq_wait(&lk->lock_object, pri); GIANT_RESTORE(); if ((flags & LK_SLEEPFAIL) && error == 0) error = ENOLCK; return (error); } -static __inline int +static __inline void wakeupshlk(struct lock *lk, const char *file, int line) { uintptr_t v, x, orig_x; u_int realexslp; - int queue, wakeup_swapper; + int queue; - wakeup_swapper = 0; for (;;) { x = lockmgr_read_value(lk); if (lockmgr_sunlock_try(lk, &x)) break; /* * We should have a sharer with waiters, so enter the hard * path in order to handle wakeups correctly. */ sleepq_lock(&lk->lock_object); orig_x = lockmgr_read_value(lk); retry_sleepq: x = orig_x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS); v = LK_UNLOCKED; /* * If the lock has exclusive waiters, give them preference in * order to avoid deadlock with shared runners up. * If interruptible sleeps left the exclusive queue empty * avoid a starvation for the threads sleeping on the shared * queue by giving them precedence and cleaning up the * exclusive waiters bit anyway. * Please note that lk_exslpfail count may be lying about * the real number of waiters with the LK_SLEEPFAIL flag on * because they may be used in conjunction with interruptible * sleeps so lk_exslpfail might be considered an 'upper limit' * bound, including the edge cases. */ realexslp = sleepq_sleepcnt(&lk->lock_object, SQ_EXCLUSIVE_QUEUE); if ((x & LK_EXCLUSIVE_WAITERS) != 0 && realexslp != 0) { if (lk->lk_exslpfail != USHRT_MAX && lk->lk_exslpfail < realexslp) { lk->lk_exslpfail = 0; queue = SQ_EXCLUSIVE_QUEUE; v |= (x & LK_SHARED_WAITERS); } else { lk->lk_exslpfail = 0; LOCK_LOG2(lk, "%s: %p has only LK_SLEEPFAIL sleepers", __func__, lk); LOCK_LOG2(lk, "%s: %p waking up threads on the exclusive queue", __func__, lk); - wakeup_swapper = - sleepq_broadcast(&lk->lock_object, - SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE); + sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, 0, + SQ_EXCLUSIVE_QUEUE); queue = SQ_SHARED_QUEUE; } } else { /* * Exclusive waiters sleeping with LK_SLEEPFAIL on * and using interruptible sleeps/timeout may have * left spourious lk_exslpfail counts on, so clean * it up anyway. */ lk->lk_exslpfail = 0; queue = SQ_SHARED_QUEUE; } if (lockmgr_sunlock_try(lk, &orig_x)) { sleepq_release(&lk->lock_object); break; } x |= LK_SHARERS_LOCK(1); if (!atomic_fcmpset_rel_ptr(&lk->lk_lock, &x, v)) { orig_x = x; goto retry_sleepq; } LOCK_LOG3(lk, "%s: %p waking up threads on the %s queue", __func__, lk, queue == SQ_SHARED_QUEUE ? "shared" : "exclusive"); - wakeup_swapper |= sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, - 0, queue); + sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, 0, queue); sleepq_release(&lk->lock_object); break; } LOCKSTAT_PROFILE_RELEASE_RWLOCK(lockmgr__release, lk, LOCKSTAT_READER); - return (wakeup_swapper); } static void assert_lockmgr(const struct lock_object *lock, int what) { panic("lockmgr locks do not support assertions"); } static void lock_lockmgr(struct lock_object *lock, uintptr_t how) { panic("lockmgr locks do not support sleep interlocking"); } static uintptr_t unlock_lockmgr(struct lock_object *lock) { panic("lockmgr locks do not support sleep interlocking"); } #ifdef KDTRACE_HOOKS static int owner_lockmgr(const struct lock_object *lock, struct thread **owner) { panic("lockmgr locks do not support owner inquiring"); } #endif void lockinit(struct lock *lk, int pri, const char *wmesg, int timo, int flags) { int iflags; MPASS((flags & ~LK_INIT_MASK) == 0); ASSERT_ATOMIC_LOAD_PTR(lk->lk_lock, ("%s: lockmgr not aligned for %s: %p", __func__, wmesg, &lk->lk_lock)); iflags = LO_SLEEPABLE | LO_UPGRADABLE; if (flags & LK_CANRECURSE) iflags |= LO_RECURSABLE; if ((flags & LK_NODUP) == 0) iflags |= LO_DUPOK; if (flags & LK_NOPROFILE) iflags |= LO_NOPROFILE; if ((flags & LK_NOWITNESS) == 0) iflags |= LO_WITNESS; if (flags & LK_QUIET) iflags |= LO_QUIET; if (flags & LK_IS_VNODE) iflags |= LO_IS_VNODE; if (flags & LK_NEW) iflags |= LO_NEW; iflags |= flags & LK_NOSHARE; lock_init(&lk->lock_object, &lock_class_lockmgr, wmesg, NULL, iflags); lk->lk_lock = LK_UNLOCKED; lk->lk_recurse = 0; lk->lk_exslpfail = 0; lk->lk_timo = timo; lk->lk_pri = pri; STACK_ZERO(lk); } /* * XXX: Gross hacks to manipulate external lock flags after * initialization. Used for certain vnode and buf locks. */ void lockallowshare(struct lock *lk) { lockmgr_assert(lk, KA_XLOCKED); lk->lock_object.lo_flags &= ~LK_NOSHARE; } void lockdisableshare(struct lock *lk) { lockmgr_assert(lk, KA_XLOCKED); lk->lock_object.lo_flags |= LK_NOSHARE; } void lockallowrecurse(struct lock *lk) { lockmgr_assert(lk, KA_XLOCKED); lk->lock_object.lo_flags |= LO_RECURSABLE; } void lockdisablerecurse(struct lock *lk) { lockmgr_assert(lk, KA_XLOCKED); lk->lock_object.lo_flags &= ~LO_RECURSABLE; } void lockdestroy(struct lock *lk) { KASSERT(lk->lk_lock == LK_UNLOCKED, ("lockmgr still held")); KASSERT(lk->lk_recurse == 0, ("lockmgr still recursed")); KASSERT(lk->lk_exslpfail == 0, ("lockmgr still exclusive waiters")); lock_destroy(&lk->lock_object); } static __always_inline bool lockmgr_slock_try(struct lock *lk, uintptr_t *xp, int flags, bool fp) { /* * If no other thread has an exclusive lock, or * no exclusive waiter is present, bump the count of * sharers. Since we have to preserve the state of * waiters, if we fail to acquire the shared lock * loop back and retry. */ while (LK_CAN_SHARE(*xp, flags, fp)) { if (atomic_fcmpset_acq_ptr(&lk->lk_lock, xp, *xp + LK_ONE_SHARER)) { return (true); } } return (false); } static __always_inline bool lockmgr_sunlock_try(struct lock *lk, uintptr_t *xp) { for (;;) { if (LK_SHARERS(*xp) > 1 || !(*xp & LK_ALL_WAITERS)) { if (atomic_fcmpset_rel_ptr(&lk->lk_lock, xp, *xp - LK_ONE_SHARER)) return (true); continue; } break; } return (false); } static bool lockmgr_slock_adaptive(struct lock_delay_arg *lda, struct lock *lk, uintptr_t *xp, int flags) { struct thread *owner; uintptr_t x; x = *xp; MPASS(x != LK_UNLOCKED); owner = (struct thread *)LK_HOLDER(x); for (;;) { MPASS(owner != curthread); if (owner == (struct thread *)LK_KERNPROC) return (false); if ((x & LK_SHARE) && LK_SHARERS(x) > 0) return (false); if (owner == NULL) return (false); if (!TD_IS_RUNNING(owner)) return (false); if ((x & LK_ALL_WAITERS) != 0) return (false); lock_delay(lda); x = lockmgr_read_value(lk); if (LK_CAN_SHARE(x, flags, false)) { *xp = x; return (true); } owner = (struct thread *)LK_HOLDER(x); } } static __noinline int lockmgr_slock_hard(struct lock *lk, u_int flags, struct lock_object *ilk, const char *file, int line, struct lockmgr_wait *lwa) { uintptr_t tid, x; int error = 0; const char *iwmesg; int ipri, itimo; #ifdef KDTRACE_HOOKS uint64_t sleep_time = 0; #endif #ifdef LOCK_PROFILING uint64_t waittime = 0; int contested = 0; #endif struct lock_delay_arg lda; if (SCHEDULER_STOPPED()) goto out; tid = (uintptr_t)curthread; if (LK_CAN_WITNESS(flags)) WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line, flags & LK_INTERLOCK ? ilk : NULL); x = lockmgr_read_value(lk); lock_delay_arg_init(&lda, &lockmgr_delay); if (!lk_adaptive) flags &= ~LK_ADAPTIVE; /* * The lock may already be locked exclusive by curthread, * avoid deadlock. */ if (LK_HOLDER(x) == tid) { LOCK_LOG2(lk, "%s: %p already held in exclusive mode", __func__, lk); error = EDEADLK; goto out; } for (;;) { if (lockmgr_slock_try(lk, &x, flags, false)) break; lock_profile_obtain_lock_failed(&lk->lock_object, false, &contested, &waittime); if ((flags & (LK_ADAPTIVE | LK_INTERLOCK)) == LK_ADAPTIVE) { if (lockmgr_slock_adaptive(&lda, lk, &x, flags)) continue; } #ifdef HWPMC_HOOKS PMC_SOFT_CALL( , , lock, failed); #endif /* * If the lock is expected to not sleep just give up * and return. */ if (LK_TRYOP(flags)) { LOCK_LOG2(lk, "%s: %p fails the try operation", __func__, lk); error = EBUSY; break; } /* * Acquire the sleepqueue chain lock because we * probabilly will need to manipulate waiters flags. */ sleepq_lock(&lk->lock_object); x = lockmgr_read_value(lk); retry_sleepq: /* * if the lock can be acquired in shared mode, try * again. */ if (LK_CAN_SHARE(x, flags, false)) { sleepq_release(&lk->lock_object); continue; } /* * Try to set the LK_SHARED_WAITERS flag. If we fail, * loop back and retry. */ if ((x & LK_SHARED_WAITERS) == 0) { if (!atomic_fcmpset_acq_ptr(&lk->lk_lock, &x, x | LK_SHARED_WAITERS)) { goto retry_sleepq; } LOCK_LOG2(lk, "%s: %p set shared waiters flag", __func__, lk); } if (lwa == NULL) { iwmesg = lk->lock_object.lo_name; ipri = lk->lk_pri; itimo = lk->lk_timo; } else { iwmesg = lwa->iwmesg; ipri = lwa->ipri; itimo = lwa->itimo; } /* * As far as we have been unable to acquire the * shared lock and the shared waiters flag is set, * we will sleep. */ #ifdef KDTRACE_HOOKS sleep_time -= lockstat_nsecs(&lk->lock_object); #endif error = sleeplk(lk, flags, ilk, iwmesg, ipri, itimo, SQ_SHARED_QUEUE); #ifdef KDTRACE_HOOKS sleep_time += lockstat_nsecs(&lk->lock_object); #endif flags &= ~LK_INTERLOCK; if (error) { LOCK_LOG3(lk, "%s: interrupted sleep for %p with %d", __func__, lk, error); break; } LOCK_LOG2(lk, "%s: %p resuming from the sleep queue", __func__, lk); x = lockmgr_read_value(lk); } if (error == 0) { #ifdef KDTRACE_HOOKS if (sleep_time != 0) LOCKSTAT_RECORD4(lockmgr__block, lk, sleep_time, LOCKSTAT_READER, (x & LK_SHARE) == 0, (x & LK_SHARE) == 0 ? 0 : LK_SHARERS(x)); #endif #ifdef LOCK_PROFILING lockmgr_note_shared_acquire(lk, contested, waittime, file, line, flags); #else lockmgr_note_shared_acquire(lk, 0, 0, file, line, flags); #endif } out: - lockmgr_exit(flags, ilk, 0); + lockmgr_exit(flags, ilk); return (error); } static bool lockmgr_xlock_adaptive(struct lock_delay_arg *lda, struct lock *lk, uintptr_t *xp) { struct thread *owner; uintptr_t x; x = *xp; MPASS(x != LK_UNLOCKED); owner = (struct thread *)LK_HOLDER(x); for (;;) { MPASS(owner != curthread); if (owner == NULL) return (false); if ((x & LK_SHARE) && LK_SHARERS(x) > 0) return (false); if (owner == (struct thread *)LK_KERNPROC) return (false); if (!TD_IS_RUNNING(owner)) return (false); if ((x & LK_ALL_WAITERS) != 0) return (false); lock_delay(lda); x = lockmgr_read_value(lk); if (x == LK_UNLOCKED) { *xp = x; return (true); } owner = (struct thread *)LK_HOLDER(x); } } static __noinline int lockmgr_xlock_hard(struct lock *lk, u_int flags, struct lock_object *ilk, const char *file, int line, struct lockmgr_wait *lwa) { struct lock_class *class; uintptr_t tid, x, v; int error = 0; const char *iwmesg; int ipri, itimo; #ifdef KDTRACE_HOOKS uint64_t sleep_time = 0; #endif #ifdef LOCK_PROFILING uint64_t waittime = 0; int contested = 0; #endif struct lock_delay_arg lda; if (SCHEDULER_STOPPED()) goto out; tid = (uintptr_t)curthread; if (LK_CAN_WITNESS(flags)) WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file, line, flags & LK_INTERLOCK ? ilk : NULL); /* * If curthread already holds the lock and this one is * allowed to recurse, simply recurse on it. */ if (lockmgr_xlocked(lk)) { if ((flags & LK_CANRECURSE) == 0 && (lk->lock_object.lo_flags & LO_RECURSABLE) == 0) { /* * If the lock is expected to not panic just * give up and return. */ if (LK_TRYOP(flags)) { LOCK_LOG2(lk, "%s: %p fails the try operation", __func__, lk); error = EBUSY; goto out; } if (flags & LK_INTERLOCK) { class = LOCK_CLASS(ilk); class->lc_unlock(ilk); } STACK_PRINT(lk); panic("%s: recursing on non recursive lockmgr %p " "@ %s:%d\n", __func__, lk, file, line); } atomic_set_ptr(&lk->lk_lock, LK_WRITER_RECURSED); lk->lk_recurse++; LOCK_LOG2(lk, "%s: %p recursing", __func__, lk); LOCK_LOG_LOCK("XLOCK", &lk->lock_object, 0, lk->lk_recurse, file, line); WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE | LK_TRYWIT(flags), file, line); TD_LOCKS_INC(curthread); goto out; } x = LK_UNLOCKED; lock_delay_arg_init(&lda, &lockmgr_delay); if (!lk_adaptive) flags &= ~LK_ADAPTIVE; for (;;) { if (x == LK_UNLOCKED) { if (atomic_fcmpset_acq_ptr(&lk->lk_lock, &x, tid)) break; continue; } lock_profile_obtain_lock_failed(&lk->lock_object, false, &contested, &waittime); if ((flags & (LK_ADAPTIVE | LK_INTERLOCK)) == LK_ADAPTIVE) { if (lockmgr_xlock_adaptive(&lda, lk, &x)) continue; } #ifdef HWPMC_HOOKS PMC_SOFT_CALL( , , lock, failed); #endif /* * If the lock is expected to not sleep just give up * and return. */ if (LK_TRYOP(flags)) { LOCK_LOG2(lk, "%s: %p fails the try operation", __func__, lk); error = EBUSY; break; } /* * Acquire the sleepqueue chain lock because we * probabilly will need to manipulate waiters flags. */ sleepq_lock(&lk->lock_object); x = lockmgr_read_value(lk); retry_sleepq: /* * if the lock has been released while we spun on * the sleepqueue chain lock just try again. */ if (x == LK_UNLOCKED) { sleepq_release(&lk->lock_object); continue; } /* * The lock can be in the state where there is a * pending queue of waiters, but still no owner. * This happens when the lock is contested and an * owner is going to claim the lock. * If curthread is the one successfully acquiring it * claim lock ownership and return, preserving waiters * flags. */ v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS); if ((x & ~v) == LK_UNLOCKED) { v &= ~LK_EXCLUSIVE_SPINNERS; if (atomic_fcmpset_acq_ptr(&lk->lk_lock, &x, tid | v)) { sleepq_release(&lk->lock_object); LOCK_LOG2(lk, "%s: %p claimed by a new writer", __func__, lk); break; } goto retry_sleepq; } /* * Try to set the LK_EXCLUSIVE_WAITERS flag. If we * fail, loop back and retry. */ if ((x & LK_EXCLUSIVE_WAITERS) == 0) { if (!atomic_fcmpset_ptr(&lk->lk_lock, &x, x | LK_EXCLUSIVE_WAITERS)) { goto retry_sleepq; } LOCK_LOG2(lk, "%s: %p set excl waiters flag", __func__, lk); } if (lwa == NULL) { iwmesg = lk->lock_object.lo_name; ipri = lk->lk_pri; itimo = lk->lk_timo; } else { iwmesg = lwa->iwmesg; ipri = lwa->ipri; itimo = lwa->itimo; } /* * As far as we have been unable to acquire the * exclusive lock and the exclusive waiters flag * is set, we will sleep. */ #ifdef KDTRACE_HOOKS sleep_time -= lockstat_nsecs(&lk->lock_object); #endif error = sleeplk(lk, flags, ilk, iwmesg, ipri, itimo, SQ_EXCLUSIVE_QUEUE); #ifdef KDTRACE_HOOKS sleep_time += lockstat_nsecs(&lk->lock_object); #endif flags &= ~LK_INTERLOCK; if (error) { LOCK_LOG3(lk, "%s: interrupted sleep for %p with %d", __func__, lk, error); break; } LOCK_LOG2(lk, "%s: %p resuming from the sleep queue", __func__, lk); x = lockmgr_read_value(lk); } if (error == 0) { #ifdef KDTRACE_HOOKS if (sleep_time != 0) LOCKSTAT_RECORD4(lockmgr__block, lk, sleep_time, LOCKSTAT_WRITER, (x & LK_SHARE) == 0, (x & LK_SHARE) == 0 ? 0 : LK_SHARERS(x)); #endif #ifdef LOCK_PROFILING lockmgr_note_exclusive_acquire(lk, contested, waittime, file, line, flags); #else lockmgr_note_exclusive_acquire(lk, 0, 0, file, line, flags); #endif } out: - lockmgr_exit(flags, ilk, 0); + lockmgr_exit(flags, ilk); return (error); } static __noinline int lockmgr_upgrade(struct lock *lk, u_int flags, struct lock_object *ilk, const char *file, int line, struct lockmgr_wait *lwa) { uintptr_t tid, v, setv; int error = 0; int op; if (SCHEDULER_STOPPED()) goto out; tid = (uintptr_t)curthread; _lockmgr_assert(lk, KA_SLOCKED, file, line); op = flags & LK_TYPE_MASK; v = lockmgr_read_value(lk); for (;;) { if (LK_SHARERS(v) > 1) { if (op == LK_TRYUPGRADE) { LOCK_LOG2(lk, "%s: %p failed the nowait upgrade", __func__, lk); error = EBUSY; goto out; } if (atomic_fcmpset_rel_ptr(&lk->lk_lock, &v, v - LK_ONE_SHARER)) { lockmgr_note_shared_release(lk, file, line); goto out_xlock; } continue; } MPASS((v & ~LK_ALL_WAITERS) == LK_SHARERS_LOCK(1)); setv = tid; setv |= (v & LK_ALL_WAITERS); /* * Try to switch from one shared lock to an exclusive one. * We need to preserve waiters flags during the operation. */ if (atomic_fcmpset_ptr(&lk->lk_lock, &v, setv)) { LOCK_LOG_LOCK("XUPGRADE", &lk->lock_object, 0, 0, file, line); WITNESS_UPGRADE(&lk->lock_object, LOP_EXCLUSIVE | LK_TRYWIT(flags), file, line); LOCKSTAT_RECORD0(lockmgr__upgrade, lk); TD_SLOCKS_DEC(curthread); goto out; } } out_xlock: error = lockmgr_xlock_hard(lk, flags, ilk, file, line, lwa); flags &= ~LK_INTERLOCK; out: - lockmgr_exit(flags, ilk, 0); + lockmgr_exit(flags, ilk); return (error); } int lockmgr_lock_flags(struct lock *lk, u_int flags, struct lock_object *ilk, const char *file, int line) { struct lock_class *class; uintptr_t x, tid; u_int op; bool locked; if (SCHEDULER_STOPPED()) return (0); op = flags & LK_TYPE_MASK; locked = false; switch (op) { case LK_SHARED: if (LK_CAN_WITNESS(flags)) WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line, flags & LK_INTERLOCK ? ilk : NULL); if (__predict_false(lk->lock_object.lo_flags & LK_NOSHARE)) break; x = lockmgr_read_value(lk); if (lockmgr_slock_try(lk, &x, flags, true)) { lockmgr_note_shared_acquire(lk, 0, 0, file, line, flags); locked = true; } else { return (lockmgr_slock_hard(lk, flags, ilk, file, line, NULL)); } break; case LK_EXCLUSIVE: if (LK_CAN_WITNESS(flags)) WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file, line, flags & LK_INTERLOCK ? ilk : NULL); tid = (uintptr_t)curthread; if (lockmgr_read_value(lk) == LK_UNLOCKED && atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid)) { lockmgr_note_exclusive_acquire(lk, 0, 0, file, line, flags); locked = true; } else { return (lockmgr_xlock_hard(lk, flags, ilk, file, line, NULL)); } break; case LK_UPGRADE: case LK_TRYUPGRADE: return (lockmgr_upgrade(lk, flags, ilk, file, line, NULL)); default: break; } if (__predict_true(locked)) { if (__predict_false(flags & LK_INTERLOCK)) { class = LOCK_CLASS(ilk); class->lc_unlock(ilk); } return (0); } else { return (__lockmgr_args(lk, flags, ilk, LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, file, line)); } } static __noinline int lockmgr_sunlock_hard(struct lock *lk, uintptr_t x, u_int flags, struct lock_object *ilk, const char *file, int line) - { - int wakeup_swapper = 0; - - if (SCHEDULER_STOPPED()) - goto out; - - wakeup_swapper = wakeupshlk(lk, file, line); - -out: - lockmgr_exit(flags, ilk, wakeup_swapper); + if (!SCHEDULER_STOPPED()) + wakeupshlk(lk, file, line); + lockmgr_exit(flags, ilk); return (0); } static __noinline int lockmgr_xunlock_hard(struct lock *lk, uintptr_t x, u_int flags, struct lock_object *ilk, const char *file, int line) { uintptr_t tid, v; - int wakeup_swapper = 0; u_int realexslp; int queue; if (SCHEDULER_STOPPED()) goto out; tid = (uintptr_t)curthread; /* * As first option, treact the lock as if it has not * any waiter. * Fix-up the tid var if the lock has been disowned. */ if (lockmgr_disowned_v(x)) tid = LK_KERNPROC; /* * The lock is held in exclusive mode. * If the lock is recursed also, then unrecurse it. */ if (lockmgr_recursed_v(x)) { LOCK_LOG2(lk, "%s: %p unrecursing", __func__, lk); lk->lk_recurse--; if (lk->lk_recurse == 0) atomic_clear_ptr(&lk->lk_lock, LK_WRITER_RECURSED); goto out; } if (tid != LK_KERNPROC) LOCKSTAT_PROFILE_RELEASE_RWLOCK(lockmgr__release, lk, LOCKSTAT_WRITER); if (x == tid && atomic_cmpset_rel_ptr(&lk->lk_lock, tid, LK_UNLOCKED)) goto out; sleepq_lock(&lk->lock_object); x = lockmgr_read_value(lk); v = LK_UNLOCKED; /* * If the lock has exclusive waiters, give them * preference in order to avoid deadlock with * shared runners up. * If interruptible sleeps left the exclusive queue * empty avoid a starvation for the threads sleeping * on the shared queue by giving them precedence * and cleaning up the exclusive waiters bit anyway. * Please note that lk_exslpfail count may be lying * about the real number of waiters with the * LK_SLEEPFAIL flag on because they may be used in * conjunction with interruptible sleeps so * lk_exslpfail might be considered an 'upper limit' * bound, including the edge cases. */ MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0); realexslp = sleepq_sleepcnt(&lk->lock_object, SQ_EXCLUSIVE_QUEUE); if ((x & LK_EXCLUSIVE_WAITERS) != 0 && realexslp != 0) { if (lk->lk_exslpfail != USHRT_MAX && lk->lk_exslpfail < realexslp) { lk->lk_exslpfail = 0; queue = SQ_EXCLUSIVE_QUEUE; v |= (x & LK_SHARED_WAITERS); } else { lk->lk_exslpfail = 0; LOCK_LOG2(lk, "%s: %p has only LK_SLEEPFAIL sleepers", __func__, lk); LOCK_LOG2(lk, "%s: %p waking up threads on the exclusive queue", __func__, lk); - wakeup_swapper = sleepq_broadcast(&lk->lock_object, - SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE); + sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, 0, + SQ_EXCLUSIVE_QUEUE); queue = SQ_SHARED_QUEUE; } } else { /* * Exclusive waiters sleeping with LK_SLEEPFAIL * on and using interruptible sleeps/timeout * may have left spourious lk_exslpfail counts * on, so clean it up anyway. */ lk->lk_exslpfail = 0; queue = SQ_SHARED_QUEUE; } LOCK_LOG3(lk, "%s: %p waking up threads on the %s queue", __func__, lk, queue == SQ_SHARED_QUEUE ? "shared" : "exclusive"); atomic_store_rel_ptr(&lk->lk_lock, v); - wakeup_swapper |= sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, 0, queue); + sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, 0, queue); sleepq_release(&lk->lock_object); out: - lockmgr_exit(flags, ilk, wakeup_swapper); + lockmgr_exit(flags, ilk); return (0); } /* * Lightweight entry points for common operations. * * Functionality is similar to sx locks, in that none of the additional lockmgr * features are supported. To be clear, these are NOT supported: * 1. shared locking disablement * 2. returning with an error after sleep * 3. unlocking the interlock * * If in doubt, use lockmgr_lock_flags. */ int lockmgr_slock(struct lock *lk, u_int flags, const char *file, int line) { uintptr_t x; MPASS((flags & LK_TYPE_MASK) == LK_SHARED); MPASS((flags & LK_INTERLOCK) == 0); MPASS((lk->lock_object.lo_flags & LK_NOSHARE) == 0); if (LK_CAN_WITNESS(flags)) WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line, NULL); x = lockmgr_read_value(lk); if (__predict_true(lockmgr_slock_try(lk, &x, flags, true))) { lockmgr_note_shared_acquire(lk, 0, 0, file, line, flags); return (0); } return (lockmgr_slock_hard(lk, flags | LK_ADAPTIVE, NULL, file, line, NULL)); } int lockmgr_xlock(struct lock *lk, u_int flags, const char *file, int line) { uintptr_t tid; MPASS((flags & LK_TYPE_MASK) == LK_EXCLUSIVE); MPASS((flags & LK_INTERLOCK) == 0); if (LK_CAN_WITNESS(flags)) WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file, line, NULL); tid = (uintptr_t)curthread; if (atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid)) { lockmgr_note_exclusive_acquire(lk, 0, 0, file, line, flags); return (0); } return (lockmgr_xlock_hard(lk, flags | LK_ADAPTIVE, NULL, file, line, NULL)); } int lockmgr_unlock(struct lock *lk) { uintptr_t x, tid; const char *file; int line; file = __FILE__; line = __LINE__; _lockmgr_assert(lk, KA_LOCKED, file, line); x = lockmgr_read_value(lk); if (__predict_true(x & LK_SHARE) != 0) { lockmgr_note_shared_release(lk, file, line); if (lockmgr_sunlock_try(lk, &x)) { LOCKSTAT_PROFILE_RELEASE_RWLOCK(lockmgr__release, lk, LOCKSTAT_READER); } else { return (lockmgr_sunlock_hard(lk, x, LK_RELEASE, NULL, file, line)); } } else { tid = (uintptr_t)curthread; lockmgr_note_exclusive_release(lk, file, line); if (x == tid && atomic_cmpset_rel_ptr(&lk->lk_lock, tid, LK_UNLOCKED)) { LOCKSTAT_PROFILE_RELEASE_RWLOCK(lockmgr__release, lk,LOCKSTAT_WRITER); } else { return (lockmgr_xunlock_hard(lk, x, LK_RELEASE, NULL, file, line)); } } return (0); } int __lockmgr_args(struct lock *lk, u_int flags, struct lock_object *ilk, const char *wmesg, int pri, int timo, const char *file, int line) { GIANT_DECLARE; struct lockmgr_wait lwa; struct lock_class *class; const char *iwmesg; uintptr_t tid, v, x; u_int op, realexslp; - int error, ipri, itimo, queue, wakeup_swapper; + int error, ipri, itimo, queue; #ifdef LOCK_PROFILING uint64_t waittime = 0; int contested = 0; #endif if (SCHEDULER_STOPPED()) return (0); error = 0; tid = (uintptr_t)curthread; op = (flags & LK_TYPE_MASK); iwmesg = (wmesg == LK_WMESG_DEFAULT) ? lk->lock_object.lo_name : wmesg; ipri = (pri == LK_PRIO_DEFAULT) ? lk->lk_pri : pri; itimo = (timo == LK_TIMO_DEFAULT) ? lk->lk_timo : timo; lwa.iwmesg = iwmesg; lwa.ipri = ipri; lwa.itimo = itimo; MPASS((flags & ~LK_TOTAL_MASK) == 0); KASSERT((op & (op - 1)) == 0, ("%s: Invalid requested operation @ %s:%d", __func__, file, line)); KASSERT((flags & (LK_NOWAIT | LK_SLEEPFAIL)) == 0 || (op != LK_DOWNGRADE && op != LK_RELEASE), ("%s: Invalid flags in regard of the operation desired @ %s:%d", __func__, file, line)); KASSERT((flags & LK_INTERLOCK) == 0 || ilk != NULL, ("%s: LK_INTERLOCK passed without valid interlock @ %s:%d", __func__, file, line)); KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread), ("%s: idle thread %p on lockmgr %s @ %s:%d", __func__, curthread, lk->lock_object.lo_name, file, line)); class = (flags & LK_INTERLOCK) ? LOCK_CLASS(ilk) : NULL; if (lk->lock_object.lo_flags & LK_NOSHARE) { switch (op) { case LK_SHARED: op = LK_EXCLUSIVE; break; case LK_UPGRADE: case LK_TRYUPGRADE: case LK_DOWNGRADE: _lockmgr_assert(lk, KA_XLOCKED | KA_NOTRECURSED, file, line); if (flags & LK_INTERLOCK) class->lc_unlock(ilk); return (0); } } - wakeup_swapper = 0; switch (op) { case LK_SHARED: return (lockmgr_slock_hard(lk, flags, ilk, file, line, &lwa)); break; case LK_UPGRADE: case LK_TRYUPGRADE: return (lockmgr_upgrade(lk, flags, ilk, file, line, &lwa)); break; case LK_EXCLUSIVE: return (lockmgr_xlock_hard(lk, flags, ilk, file, line, &lwa)); break; case LK_DOWNGRADE: _lockmgr_assert(lk, KA_XLOCKED, file, line); WITNESS_DOWNGRADE(&lk->lock_object, 0, file, line); /* * Panic if the lock is recursed. */ if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) { if (flags & LK_INTERLOCK) class->lc_unlock(ilk); panic("%s: downgrade a recursed lockmgr %s @ %s:%d\n", __func__, iwmesg, file, line); } TD_SLOCKS_INC(curthread); /* * In order to preserve waiters flags, just spin. */ for (;;) { x = lockmgr_read_value(lk); MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0); x &= LK_ALL_WAITERS; if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x, LK_SHARERS_LOCK(1) | x)) break; cpu_spinwait(); } LOCK_LOG_LOCK("XDOWNGRADE", &lk->lock_object, 0, 0, file, line); LOCKSTAT_RECORD0(lockmgr__downgrade, lk); break; case LK_RELEASE: _lockmgr_assert(lk, KA_LOCKED, file, line); x = lockmgr_read_value(lk); if (__predict_true(x & LK_SHARE) != 0) { lockmgr_note_shared_release(lk, file, line); return (lockmgr_sunlock_hard(lk, x, flags, ilk, file, line)); } else { lockmgr_note_exclusive_release(lk, file, line); return (lockmgr_xunlock_hard(lk, x, flags, ilk, file, line)); } break; case LK_DRAIN: if (LK_CAN_WITNESS(flags)) WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file, line, flags & LK_INTERLOCK ? ilk : NULL); /* * Trying to drain a lock we already own will result in a * deadlock. */ if (lockmgr_xlocked(lk)) { if (flags & LK_INTERLOCK) class->lc_unlock(ilk); panic("%s: draining %s with the lock held @ %s:%d\n", __func__, iwmesg, file, line); } for (;;) { if (lk->lk_lock == LK_UNLOCKED && atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid)) break; #ifdef HWPMC_HOOKS PMC_SOFT_CALL( , , lock, failed); #endif lock_profile_obtain_lock_failed(&lk->lock_object, false, &contested, &waittime); /* * If the lock is expected to not sleep just give up * and return. */ if (LK_TRYOP(flags)) { LOCK_LOG2(lk, "%s: %p fails the try operation", __func__, lk); error = EBUSY; break; } /* * Acquire the sleepqueue chain lock because we * probabilly will need to manipulate waiters flags. */ sleepq_lock(&lk->lock_object); x = lockmgr_read_value(lk); /* * if the lock has been released while we spun on * the sleepqueue chain lock just try again. */ if (x == LK_UNLOCKED) { sleepq_release(&lk->lock_object); continue; } v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS); if ((x & ~v) == LK_UNLOCKED) { v = (x & ~LK_EXCLUSIVE_SPINNERS); /* * If interruptible sleeps left the exclusive * queue empty avoid a starvation for the * threads sleeping on the shared queue by * giving them precedence and cleaning up the * exclusive waiters bit anyway. * Please note that lk_exslpfail count may be * lying about the real number of waiters with * the LK_SLEEPFAIL flag on because they may * be used in conjunction with interruptible * sleeps so lk_exslpfail might be considered * an 'upper limit' bound, including the edge * cases. */ if (v & LK_EXCLUSIVE_WAITERS) { queue = SQ_EXCLUSIVE_QUEUE; v &= ~LK_EXCLUSIVE_WAITERS; } else { /* * Exclusive waiters sleeping with * LK_SLEEPFAIL on and using * interruptible sleeps/timeout may * have left spourious lk_exslpfail * counts on, so clean it up anyway. */ MPASS(v & LK_SHARED_WAITERS); lk->lk_exslpfail = 0; queue = SQ_SHARED_QUEUE; v &= ~LK_SHARED_WAITERS; } if (queue == SQ_EXCLUSIVE_QUEUE) { realexslp = sleepq_sleepcnt(&lk->lock_object, SQ_EXCLUSIVE_QUEUE); if (lk->lk_exslpfail >= realexslp) { lk->lk_exslpfail = 0; queue = SQ_SHARED_QUEUE; v &= ~LK_SHARED_WAITERS; if (realexslp != 0) { LOCK_LOG2(lk, "%s: %p has only LK_SLEEPFAIL sleepers", __func__, lk); LOCK_LOG2(lk, "%s: %p waking up threads on the exclusive queue", __func__, lk); - wakeup_swapper = - sleepq_broadcast( + sleepq_broadcast( &lk->lock_object, SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE); } } else lk->lk_exslpfail = 0; } if (!atomic_cmpset_ptr(&lk->lk_lock, x, v)) { sleepq_release(&lk->lock_object); continue; } LOCK_LOG3(lk, "%s: %p waking up all threads on the %s queue", __func__, lk, queue == SQ_SHARED_QUEUE ? "shared" : "exclusive"); - wakeup_swapper |= sleepq_broadcast( - &lk->lock_object, SLEEPQ_LK, 0, queue); + sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, 0, + queue); /* * If shared waiters have been woken up we need * to wait for one of them to acquire the lock * before to set the exclusive waiters in * order to avoid a deadlock. */ if (queue == SQ_SHARED_QUEUE) { for (v = lk->lk_lock; (v & LK_SHARE) && !LK_SHARERS(v); v = lk->lk_lock) cpu_spinwait(); } } /* * Try to set the LK_EXCLUSIVE_WAITERS flag. If we * fail, loop back and retry. */ if ((x & LK_EXCLUSIVE_WAITERS) == 0) { if (!atomic_cmpset_ptr(&lk->lk_lock, x, x | LK_EXCLUSIVE_WAITERS)) { sleepq_release(&lk->lock_object); continue; } LOCK_LOG2(lk, "%s: %p set drain waiters flag", __func__, lk); } /* * As far as we have been unable to acquire the * exclusive lock and the exclusive waiters flag * is set, we will sleep. */ if (flags & LK_INTERLOCK) { class->lc_unlock(ilk); flags &= ~LK_INTERLOCK; } GIANT_SAVE(); sleepq_add(&lk->lock_object, NULL, iwmesg, SLEEPQ_LK, SQ_EXCLUSIVE_QUEUE); sleepq_wait(&lk->lock_object, ipri & PRIMASK); GIANT_RESTORE(); LOCK_LOG2(lk, "%s: %p resuming from the sleep queue", __func__, lk); } if (error == 0) { lock_profile_obtain_lock_success(&lk->lock_object, false, contested, waittime, file, line); LOCK_LOG_LOCK("DRAIN", &lk->lock_object, 0, lk->lk_recurse, file, line); WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE | LK_TRYWIT(flags), file, line); TD_LOCKS_INC(curthread); STACK_SAVE(lk); } break; default: if (flags & LK_INTERLOCK) class->lc_unlock(ilk); panic("%s: unknown lockmgr request 0x%x\n", __func__, op); } if (flags & LK_INTERLOCK) class->lc_unlock(ilk); - if (wakeup_swapper) - kick_proc0(); return (error); } void _lockmgr_disown(struct lock *lk, const char *file, int line) { uintptr_t tid, x; if (SCHEDULER_STOPPED()) return; tid = (uintptr_t)curthread; _lockmgr_assert(lk, KA_XLOCKED, file, line); /* * Panic if the lock is recursed. */ if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) panic("%s: disown a recursed lockmgr @ %s:%d\n", __func__, file, line); /* * If the owner is already LK_KERNPROC just skip the whole operation. */ if (LK_HOLDER(lk->lk_lock) != tid) return; lock_profile_release_lock(&lk->lock_object, false); LOCKSTAT_RECORD1(lockmgr__disown, lk, LOCKSTAT_WRITER); LOCK_LOG_LOCK("XDISOWN", &lk->lock_object, 0, 0, file, line); WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE, file, line); TD_LOCKS_DEC(curthread); STACK_SAVE(lk); /* * In order to preserve waiters flags, just spin. */ for (;;) { x = lockmgr_read_value(lk); MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0); x &= LK_ALL_WAITERS; if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x, LK_KERNPROC | x)) return; cpu_spinwait(); } } void lockmgr_printinfo(const struct lock *lk) { struct thread *td; uintptr_t x; if (lk->lk_lock == LK_UNLOCKED) printf("lock type %s: UNLOCKED\n", lk->lock_object.lo_name); else if (lk->lk_lock & LK_SHARE) printf("lock type %s: SHARED (count %ju)\n", lk->lock_object.lo_name, (uintmax_t)LK_SHARERS(lk->lk_lock)); else { td = lockmgr_xholder(lk); if (td == (struct thread *)LK_KERNPROC) printf("lock type %s: EXCL by KERNPROC\n", lk->lock_object.lo_name); else printf("lock type %s: EXCL by thread %p " "(pid %d, %s, tid %d)\n", lk->lock_object.lo_name, td, td->td_proc->p_pid, td->td_proc->p_comm, td->td_tid); } x = lk->lk_lock; if (x & LK_EXCLUSIVE_WAITERS) printf(" with exclusive waiters pending\n"); if (x & LK_SHARED_WAITERS) printf(" with shared waiters pending\n"); if (x & LK_EXCLUSIVE_SPINNERS) printf(" with exclusive spinners pending\n"); STACK_PRINT(lk); } int lockstatus(const struct lock *lk) { uintptr_t v, x; int ret; ret = LK_SHARED; x = lockmgr_read_value(lk); v = LK_HOLDER(x); if ((x & LK_SHARE) == 0) { if (v == (uintptr_t)curthread || v == LK_KERNPROC) ret = LK_EXCLUSIVE; else ret = LK_EXCLOTHER; } else if (x == LK_UNLOCKED) ret = 0; return (ret); } #ifdef INVARIANT_SUPPORT FEATURE(invariant_support, "Support for modules compiled with INVARIANTS option"); #ifndef INVARIANTS #undef _lockmgr_assert #endif void _lockmgr_assert(const struct lock *lk, int what, const char *file, int line) { int slocked = 0; if (SCHEDULER_STOPPED()) return; switch (what) { case KA_SLOCKED: case KA_SLOCKED | KA_NOTRECURSED: case KA_SLOCKED | KA_RECURSED: slocked = 1; case KA_LOCKED: case KA_LOCKED | KA_NOTRECURSED: case KA_LOCKED | KA_RECURSED: #ifdef WITNESS /* * We cannot trust WITNESS if the lock is held in exclusive * mode and a call to lockmgr_disown() happened. * Workaround this skipping the check if the lock is held in * exclusive mode even for the KA_LOCKED case. */ if (slocked || (lk->lk_lock & LK_SHARE)) { witness_assert(&lk->lock_object, what, file, line); break; } #endif if (lk->lk_lock == LK_UNLOCKED || ((lk->lk_lock & LK_SHARE) == 0 && (slocked || (!lockmgr_xlocked(lk) && !lockmgr_disowned(lk))))) panic("Lock %s not %slocked @ %s:%d\n", lk->lock_object.lo_name, slocked ? "share" : "", file, line); if ((lk->lk_lock & LK_SHARE) == 0) { if (lockmgr_recursed(lk)) { if (what & KA_NOTRECURSED) panic("Lock %s recursed @ %s:%d\n", lk->lock_object.lo_name, file, line); } else if (what & KA_RECURSED) panic("Lock %s not recursed @ %s:%d\n", lk->lock_object.lo_name, file, line); } break; case KA_XLOCKED: case KA_XLOCKED | KA_NOTRECURSED: case KA_XLOCKED | KA_RECURSED: if (!lockmgr_xlocked(lk) && !lockmgr_disowned(lk)) panic("Lock %s not exclusively locked @ %s:%d\n", lk->lock_object.lo_name, file, line); if (lockmgr_recursed(lk)) { if (what & KA_NOTRECURSED) panic("Lock %s recursed @ %s:%d\n", lk->lock_object.lo_name, file, line); } else if (what & KA_RECURSED) panic("Lock %s not recursed @ %s:%d\n", lk->lock_object.lo_name, file, line); break; case KA_UNLOCKED: if (lockmgr_xlocked(lk) || lockmgr_disowned(lk)) panic("Lock %s exclusively locked @ %s:%d\n", lk->lock_object.lo_name, file, line); break; default: panic("Unknown lockmgr assertion: %d @ %s:%d\n", what, file, line); } } #endif #ifdef DDB int lockmgr_chain(struct thread *td, struct thread **ownerp) { const struct lock *lk; lk = td->td_wchan; if (LOCK_CLASS(&lk->lock_object) != &lock_class_lockmgr) return (0); db_printf("blocked on lockmgr %s", lk->lock_object.lo_name); if (lk->lk_lock & LK_SHARE) db_printf("SHARED (count %ju)\n", (uintmax_t)LK_SHARERS(lk->lk_lock)); else db_printf("EXCL\n"); *ownerp = lockmgr_xholder(lk); return (1); } static void db_show_lockmgr(const struct lock_object *lock) { struct thread *td; const struct lock *lk; lk = (const struct lock *)lock; db_printf(" state: "); if (lk->lk_lock == LK_UNLOCKED) db_printf("UNLOCKED\n"); else if (lk->lk_lock & LK_SHARE) db_printf("SLOCK: %ju\n", (uintmax_t)LK_SHARERS(lk->lk_lock)); else { td = lockmgr_xholder(lk); if (td == (struct thread *)LK_KERNPROC) db_printf("XLOCK: LK_KERNPROC\n"); else db_printf("XLOCK: %p (tid %d, pid %d, \"%s\")\n", td, td->td_tid, td->td_proc->p_pid, td->td_proc->p_comm); if (lockmgr_recursed(lk)) db_printf(" recursed: %d\n", lk->lk_recurse); } db_printf(" waiters: "); switch (lk->lk_lock & LK_ALL_WAITERS) { case LK_SHARED_WAITERS: db_printf("shared\n"); break; case LK_EXCLUSIVE_WAITERS: db_printf("exclusive\n"); break; case LK_ALL_WAITERS: db_printf("shared and exclusive\n"); break; default: db_printf("none\n"); } db_printf(" spinners: "); if (lk->lk_lock & LK_EXCLUSIVE_SPINNERS) db_printf("exclusive\n"); else db_printf("none\n"); } #endif diff --git a/sys/kern/kern_sx.c b/sys/kern/kern_sx.c index 18e6ba232c4a..4cc8ac8ce106 100644 --- a/sys/kern/kern_sx.c +++ b/sys/kern/kern_sx.c @@ -1,1575 +1,1563 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2007 Attilio Rao * Copyright (c) 2001 Jason Evans * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice(s), this list of conditions and the following disclaimer as * the first lines of this file unmodified other than the possible * addition of one or more copyright notices. * 2. Redistributions in binary form must reproduce the above copyright * notice(s), this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH * DAMAGE. */ /* * Shared/exclusive locks. This implementation attempts to ensure * deterministic lock granting behavior, so that slocks and xlocks are * interleaved. * * Priority propagation will not generally raise the priority of lock holders, * so should not be relied upon in combination with sx locks. */ #include "opt_ddb.h" #include "opt_hwpmc_hooks.h" #include "opt_no_adaptive_sx.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(SMP) && !defined(NO_ADAPTIVE_SX) #include #endif #ifdef DDB #include #endif #if defined(SMP) && !defined(NO_ADAPTIVE_SX) #define ADAPTIVE_SX #endif #ifdef HWPMC_HOOKS #include PMC_SOFT_DECLARE( , , lock, failed); #endif /* Handy macros for sleep queues. */ #define SQ_EXCLUSIVE_QUEUE 0 #define SQ_SHARED_QUEUE 1 /* * Variations on DROP_GIANT()/PICKUP_GIANT() for use in this file. We * drop Giant anytime we have to sleep or if we adaptively spin. */ #define GIANT_DECLARE \ int _giantcnt = 0; \ WITNESS_SAVE_DECL(Giant) \ #define GIANT_SAVE(work) do { \ if (__predict_false(mtx_owned(&Giant))) { \ work++; \ WITNESS_SAVE(&Giant.lock_object, Giant); \ while (mtx_owned(&Giant)) { \ _giantcnt++; \ mtx_unlock(&Giant); \ } \ } \ } while (0) #define GIANT_RESTORE() do { \ if (_giantcnt > 0) { \ mtx_assert(&Giant, MA_NOTOWNED); \ while (_giantcnt--) \ mtx_lock(&Giant); \ WITNESS_RESTORE(&Giant.lock_object, Giant); \ } \ } while (0) /* * Returns true if an exclusive lock is recursed. It assumes * curthread currently has an exclusive lock. */ #define sx_recursed(sx) ((sx)->sx_recurse != 0) static void assert_sx(const struct lock_object *lock, int what); #ifdef DDB static void db_show_sx(const struct lock_object *lock); #endif static void lock_sx(struct lock_object *lock, uintptr_t how); #ifdef KDTRACE_HOOKS static int owner_sx(const struct lock_object *lock, struct thread **owner); #endif static uintptr_t unlock_sx(struct lock_object *lock); struct lock_class lock_class_sx = { .lc_name = "sx", .lc_flags = LC_SLEEPLOCK | LC_SLEEPABLE | LC_RECURSABLE | LC_UPGRADABLE, .lc_assert = assert_sx, #ifdef DDB .lc_ddb_show = db_show_sx, #endif .lc_lock = lock_sx, .lc_unlock = unlock_sx, #ifdef KDTRACE_HOOKS .lc_owner = owner_sx, #endif }; #ifndef INVARIANTS #define _sx_assert(sx, what, file, line) #endif #ifdef ADAPTIVE_SX #ifdef SX_CUSTOM_BACKOFF static u_short __read_frequently asx_retries; static u_short __read_frequently asx_loops; static SYSCTL_NODE(_debug, OID_AUTO, sx, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "sxlock debugging"); SYSCTL_U16(_debug_sx, OID_AUTO, retries, CTLFLAG_RW, &asx_retries, 0, ""); SYSCTL_U16(_debug_sx, OID_AUTO, loops, CTLFLAG_RW, &asx_loops, 0, ""); static struct lock_delay_config __read_frequently sx_delay; SYSCTL_U16(_debug_sx, OID_AUTO, delay_base, CTLFLAG_RW, &sx_delay.base, 0, ""); SYSCTL_U16(_debug_sx, OID_AUTO, delay_max, CTLFLAG_RW, &sx_delay.max, 0, ""); static void sx_lock_delay_init(void *arg __unused) { lock_delay_default_init(&sx_delay); asx_retries = 10; asx_loops = max(10000, sx_delay.max); } LOCK_DELAY_SYSINIT(sx_lock_delay_init); #else #define sx_delay locks_delay #define asx_retries locks_delay_retries #define asx_loops locks_delay_loops #endif #endif void assert_sx(const struct lock_object *lock, int what) { sx_assert((const struct sx *)lock, what); } void lock_sx(struct lock_object *lock, uintptr_t how) { struct sx *sx; sx = (struct sx *)lock; if (how) sx_slock(sx); else sx_xlock(sx); } uintptr_t unlock_sx(struct lock_object *lock) { struct sx *sx; sx = (struct sx *)lock; sx_assert(sx, SA_LOCKED | SA_NOTRECURSED); if (sx_xlocked(sx)) { sx_xunlock(sx); return (0); } else { sx_sunlock(sx); return (1); } } #ifdef KDTRACE_HOOKS int owner_sx(const struct lock_object *lock, struct thread **owner) { const struct sx *sx; uintptr_t x; sx = (const struct sx *)lock; x = sx->sx_lock; *owner = NULL; return ((x & SX_LOCK_SHARED) != 0 ? (SX_SHARERS(x) != 0) : ((*owner = (struct thread *)SX_OWNER(x)) != NULL)); } #endif void sx_sysinit(void *arg) { struct sx_args *sargs = arg; sx_init_flags(sargs->sa_sx, sargs->sa_desc, sargs->sa_flags); } void sx_init_flags(struct sx *sx, const char *description, int opts) { int flags; MPASS((opts & ~(SX_QUIET | SX_RECURSE | SX_NOWITNESS | SX_DUPOK | SX_NOPROFILE | SX_NEW)) == 0); ASSERT_ATOMIC_LOAD_PTR(sx->sx_lock, ("%s: sx_lock not aligned for %s: %p", __func__, description, &sx->sx_lock)); flags = LO_SLEEPABLE | LO_UPGRADABLE; if (opts & SX_DUPOK) flags |= LO_DUPOK; if (opts & SX_NOPROFILE) flags |= LO_NOPROFILE; if (!(opts & SX_NOWITNESS)) flags |= LO_WITNESS; if (opts & SX_RECURSE) flags |= LO_RECURSABLE; if (opts & SX_QUIET) flags |= LO_QUIET; if (opts & SX_NEW) flags |= LO_NEW; lock_init(&sx->lock_object, &lock_class_sx, description, NULL, flags); sx->sx_lock = SX_LOCK_UNLOCKED; sx->sx_recurse = 0; } void sx_destroy(struct sx *sx) { KASSERT(sx->sx_lock == SX_LOCK_UNLOCKED, ("sx lock still held")); KASSERT(sx->sx_recurse == 0, ("sx lock still recursed")); sx->sx_lock = SX_LOCK_DESTROYED; lock_destroy(&sx->lock_object); } int sx_try_slock_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF) { uintptr_t x; if (SCHEDULER_STOPPED()) return (1); KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread), ("sx_try_slock() by idle thread %p on sx %s @ %s:%d", curthread, sx->lock_object.lo_name, file, line)); x = sx->sx_lock; for (;;) { KASSERT(x != SX_LOCK_DESTROYED, ("sx_try_slock() of destroyed sx @ %s:%d", file, line)); if (!(x & SX_LOCK_SHARED)) break; if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, x + SX_ONE_SHARER)) { LOCK_LOG_TRY("SLOCK", &sx->lock_object, 0, 1, file, line); WITNESS_LOCK(&sx->lock_object, LOP_TRYLOCK, file, line); LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx, 0, 0, file, line, LOCKSTAT_READER); TD_LOCKS_INC(curthread); curthread->td_sx_slocks++; return (1); } } LOCK_LOG_TRY("SLOCK", &sx->lock_object, 0, 0, file, line); return (0); } int sx_try_slock_(struct sx *sx, const char *file, int line) { return (sx_try_slock_int(sx LOCK_FILE_LINE_ARG)); } int _sx_xlock(struct sx *sx, int opts, const char *file, int line) { uintptr_t tid, x; int error = 0; KASSERT(kdb_active != 0 || SCHEDULER_STOPPED() || !TD_IS_IDLETHREAD(curthread), ("sx_xlock() by idle thread %p on sx %s @ %s:%d", curthread, sx->lock_object.lo_name, file, line)); KASSERT(sx->sx_lock != SX_LOCK_DESTROYED, ("sx_xlock() of destroyed sx @ %s:%d", file, line)); WITNESS_CHECKORDER(&sx->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file, line, NULL); tid = (uintptr_t)curthread; x = SX_LOCK_UNLOCKED; if (!atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid)) error = _sx_xlock_hard(sx, x, opts LOCK_FILE_LINE_ARG); else LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx, 0, 0, file, line, LOCKSTAT_WRITER); if (!error) { LOCK_LOG_LOCK("XLOCK", &sx->lock_object, 0, sx->sx_recurse, file, line); WITNESS_LOCK(&sx->lock_object, LOP_EXCLUSIVE, file, line); TD_LOCKS_INC(curthread); } return (error); } int sx_try_xlock_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF) { struct thread *td; uintptr_t tid, x; int rval; bool recursed; td = curthread; tid = (uintptr_t)td; if (SCHEDULER_STOPPED()) return (1); KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(td), ("sx_try_xlock() by idle thread %p on sx %s @ %s:%d", curthread, sx->lock_object.lo_name, file, line)); KASSERT(sx->sx_lock != SX_LOCK_DESTROYED, ("sx_try_xlock() of destroyed sx @ %s:%d", file, line)); rval = 1; recursed = false; x = SX_LOCK_UNLOCKED; for (;;) { if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid)) break; if (x == SX_LOCK_UNLOCKED) continue; if (x == tid && (sx->lock_object.lo_flags & LO_RECURSABLE)) { sx->sx_recurse++; atomic_set_ptr(&sx->sx_lock, SX_LOCK_RECURSED); break; } rval = 0; break; } LOCK_LOG_TRY("XLOCK", &sx->lock_object, 0, rval, file, line); if (rval) { WITNESS_LOCK(&sx->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK, file, line); if (!recursed) LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx, 0, 0, file, line, LOCKSTAT_WRITER); TD_LOCKS_INC(curthread); } return (rval); } int sx_try_xlock_(struct sx *sx, const char *file, int line) { return (sx_try_xlock_int(sx LOCK_FILE_LINE_ARG)); } void _sx_xunlock(struct sx *sx, const char *file, int line) { KASSERT(sx->sx_lock != SX_LOCK_DESTROYED, ("sx_xunlock() of destroyed sx @ %s:%d", file, line)); _sx_assert(sx, SA_XLOCKED, file, line); WITNESS_UNLOCK(&sx->lock_object, LOP_EXCLUSIVE, file, line); LOCK_LOG_LOCK("XUNLOCK", &sx->lock_object, 0, sx->sx_recurse, file, line); #if LOCK_DEBUG > 0 _sx_xunlock_hard(sx, (uintptr_t)curthread, file, line); #else __sx_xunlock(sx, curthread, file, line); #endif TD_LOCKS_DEC(curthread); } /* * Try to do a non-blocking upgrade from a shared lock to an exclusive lock. * This will only succeed if this thread holds a single shared lock. * Return 1 if if the upgrade succeed, 0 otherwise. */ int sx_try_upgrade_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF) { uintptr_t x; uintptr_t waiters; int success; if (SCHEDULER_STOPPED()) return (1); KASSERT(sx->sx_lock != SX_LOCK_DESTROYED, ("sx_try_upgrade() of destroyed sx @ %s:%d", file, line)); _sx_assert(sx, SA_SLOCKED, file, line); /* * Try to switch from one shared lock to an exclusive lock. We need * to maintain the SX_LOCK_EXCLUSIVE_WAITERS flag if set so that * we will wake up the exclusive waiters when we drop the lock. */ success = 0; x = SX_READ_VALUE(sx); for (;;) { if (SX_SHARERS(x) > 1) break; waiters = (x & SX_LOCK_WAITERS); if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, (uintptr_t)curthread | waiters)) { success = 1; break; } } LOCK_LOG_TRY("XUPGRADE", &sx->lock_object, 0, success, file, line); if (success) { curthread->td_sx_slocks--; WITNESS_UPGRADE(&sx->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK, file, line); LOCKSTAT_RECORD0(sx__upgrade, sx); } return (success); } int sx_try_upgrade_(struct sx *sx, const char *file, int line) { return (sx_try_upgrade_int(sx LOCK_FILE_LINE_ARG)); } /* * Downgrade an unrecursed exclusive lock into a single shared lock. */ void sx_downgrade_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF) { uintptr_t x; - int wakeup_swapper; if (SCHEDULER_STOPPED()) return; KASSERT(sx->sx_lock != SX_LOCK_DESTROYED, ("sx_downgrade() of destroyed sx @ %s:%d", file, line)); _sx_assert(sx, SA_XLOCKED | SA_NOTRECURSED, file, line); #ifndef INVARIANTS if (sx_recursed(sx)) panic("downgrade of a recursed lock"); #endif WITNESS_DOWNGRADE(&sx->lock_object, 0, file, line); /* * Try to switch from an exclusive lock with no shared waiters * to one sharer with no shared waiters. If there are * exclusive waiters, we don't need to lock the sleep queue so * long as we preserve the flag. We do one quick try and if * that fails we grab the sleepq lock to keep the flags from * changing and do it the slow way. * * We have to lock the sleep queue if there are shared waiters * so we can wake them up. */ x = sx->sx_lock; if (!(x & SX_LOCK_SHARED_WAITERS) && atomic_cmpset_rel_ptr(&sx->sx_lock, x, SX_SHARERS_LOCK(1) | (x & SX_LOCK_EXCLUSIVE_WAITERS))) goto out; /* * Lock the sleep queue so we can read the waiters bits * without any races and wakeup any shared waiters. */ sleepq_lock(&sx->lock_object); /* * Preserve SX_LOCK_EXCLUSIVE_WAITERS while downgraded to a single * shared lock. If there are any shared waiters, wake them up. */ - wakeup_swapper = 0; x = sx->sx_lock; atomic_store_rel_ptr(&sx->sx_lock, SX_SHARERS_LOCK(1) | (x & SX_LOCK_EXCLUSIVE_WAITERS)); if (x & SX_LOCK_SHARED_WAITERS) - wakeup_swapper = sleepq_broadcast(&sx->lock_object, SLEEPQ_SX, - 0, SQ_SHARED_QUEUE); + sleepq_broadcast(&sx->lock_object, SLEEPQ_SX, 0, + SQ_SHARED_QUEUE); sleepq_release(&sx->lock_object); - if (wakeup_swapper) - kick_proc0(); - out: curthread->td_sx_slocks++; LOCK_LOG_LOCK("XDOWNGRADE", &sx->lock_object, 0, 0, file, line); LOCKSTAT_RECORD0(sx__downgrade, sx); } void sx_downgrade_(struct sx *sx, const char *file, int line) { sx_downgrade_int(sx LOCK_FILE_LINE_ARG); } #ifdef ADAPTIVE_SX static inline void sx_drop_critical(uintptr_t x, bool *in_critical, int *extra_work) { if (x & SX_LOCK_WRITE_SPINNER) return; if (*in_critical) { critical_exit(); *in_critical = false; (*extra_work)--; } } #else #define sx_drop_critical(x, in_critical, extra_work) do { } while (0) #endif /* * This function represents the so-called 'hard case' for sx_xlock * operation. All 'easy case' failures are redirected to this. Note * that ideally this would be a static function, but it needs to be * accessible from at least sx.h. */ int _sx_xlock_hard(struct sx *sx, uintptr_t x, int opts LOCK_FILE_LINE_ARG_DEF) { GIANT_DECLARE; uintptr_t tid, setx; #ifdef ADAPTIVE_SX struct thread *owner; u_int i, n, spintries = 0; enum { READERS, WRITER } sleep_reason = READERS; bool in_critical = false; #endif #ifdef LOCK_PROFILING uint64_t waittime = 0; int contested = 0; #endif int error = 0; #if defined(ADAPTIVE_SX) || defined(KDTRACE_HOOKS) struct lock_delay_arg lda; #endif #ifdef KDTRACE_HOOKS u_int sleep_cnt = 0; int64_t sleep_time = 0; int64_t all_time = 0; #endif #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING) uintptr_t state = 0; int doing_lockprof = 0; #endif int extra_work = 0; tid = (uintptr_t)curthread; #ifdef KDTRACE_HOOKS if (LOCKSTAT_PROFILE_ENABLED(sx__acquire)) { while (x == SX_LOCK_UNLOCKED) { if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid)) goto out_lockstat; } extra_work = 1; doing_lockprof = 1; all_time -= lockstat_nsecs(&sx->lock_object); state = x; } #endif #ifdef LOCK_PROFILING extra_work = 1; doing_lockprof = 1; state = x; #endif if (SCHEDULER_STOPPED()) return (0); if (__predict_false(x == SX_LOCK_UNLOCKED)) x = SX_READ_VALUE(sx); /* If we already hold an exclusive lock, then recurse. */ if (__predict_false(lv_sx_owner(x) == (struct thread *)tid)) { KASSERT((sx->lock_object.lo_flags & LO_RECURSABLE) != 0, ("_sx_xlock_hard: recursed on non-recursive sx %s @ %s:%d\n", sx->lock_object.lo_name, file, line)); sx->sx_recurse++; atomic_set_ptr(&sx->sx_lock, SX_LOCK_RECURSED); if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p recursing", __func__, sx); return (0); } if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR5(KTR_LOCK, "%s: %s contested (lock=%p) at %s:%d", __func__, sx->lock_object.lo_name, (void *)sx->sx_lock, file, line); #if defined(ADAPTIVE_SX) lock_delay_arg_init(&lda, &sx_delay); #elif defined(KDTRACE_HOOKS) lock_delay_arg_init_noadapt(&lda); #endif #ifdef HWPMC_HOOKS PMC_SOFT_CALL( , , lock, failed); #endif lock_profile_obtain_lock_failed(&sx->lock_object, false, &contested, &waittime); #ifndef INVARIANTS GIANT_SAVE(extra_work); #endif THREAD_CONTENDS_ON_LOCK(&sx->lock_object); for (;;) { if (x == SX_LOCK_UNLOCKED) { if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid)) break; continue; } #ifdef INVARIANTS GIANT_SAVE(extra_work); #endif #ifdef KDTRACE_HOOKS lda.spin_cnt++; #endif #ifdef ADAPTIVE_SX if (x == (SX_LOCK_SHARED | SX_LOCK_WRITE_SPINNER)) { if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid)) break; continue; } /* * If the lock is write locked and the owner is * running on another CPU, spin until the owner stops * running or the state of the lock changes. */ if ((x & SX_LOCK_SHARED) == 0) { sx_drop_critical(x, &in_critical, &extra_work); sleep_reason = WRITER; owner = lv_sx_owner(x); if (!TD_IS_RUNNING(owner)) goto sleepq; if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR3(KTR_LOCK, "%s: spinning on %p held by %p", __func__, sx, owner); KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread), "spinning", "lockname:\"%s\"", sx->lock_object.lo_name); do { lock_delay(&lda); x = SX_READ_VALUE(sx); owner = lv_sx_owner(x); } while (owner != NULL && TD_IS_RUNNING(owner)); KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread), "running"); continue; } else if (SX_SHARERS(x) > 0) { sleep_reason = READERS; if (spintries == asx_retries) goto sleepq; if (!(x & SX_LOCK_WRITE_SPINNER)) { if (!in_critical) { critical_enter(); in_critical = true; extra_work++; } if (!atomic_fcmpset_ptr(&sx->sx_lock, &x, x | SX_LOCK_WRITE_SPINNER)) { critical_exit(); in_critical = false; extra_work--; continue; } } spintries++; KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread), "spinning", "lockname:\"%s\"", sx->lock_object.lo_name); n = SX_SHARERS(x); for (i = 0; i < asx_loops; i += n) { lock_delay_spin(n); x = SX_READ_VALUE(sx); if (!(x & SX_LOCK_WRITE_SPINNER)) break; if (!(x & SX_LOCK_SHARED)) break; n = SX_SHARERS(x); if (n == 0) break; } #ifdef KDTRACE_HOOKS lda.spin_cnt += i; #endif KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread), "running"); if (i < asx_loops) continue; } sleepq: #endif sleepq_lock(&sx->lock_object); x = SX_READ_VALUE(sx); retry_sleepq: /* * If the lock was released while spinning on the * sleep queue chain lock, try again. */ if (x == SX_LOCK_UNLOCKED) { sleepq_release(&sx->lock_object); sx_drop_critical(x, &in_critical, &extra_work); continue; } #ifdef ADAPTIVE_SX /* * The current lock owner might have started executing * on another CPU (or the lock could have changed * owners) while we were waiting on the sleep queue * chain lock. If so, drop the sleep queue lock and try * again. */ if (!(x & SX_LOCK_SHARED)) { owner = (struct thread *)SX_OWNER(x); if (TD_IS_RUNNING(owner)) { sleepq_release(&sx->lock_object); sx_drop_critical(x, &in_critical, &extra_work); continue; } } else if (SX_SHARERS(x) > 0 && sleep_reason == WRITER) { sleepq_release(&sx->lock_object); sx_drop_critical(x, &in_critical, &extra_work); continue; } #endif /* * If an exclusive lock was released with both shared * and exclusive waiters and a shared waiter hasn't * woken up and acquired the lock yet, sx_lock will be * set to SX_LOCK_UNLOCKED | SX_LOCK_EXCLUSIVE_WAITERS. * If we see that value, try to acquire it once. Note * that we have to preserve SX_LOCK_EXCLUSIVE_WAITERS * as there are other exclusive waiters still. If we * fail, restart the loop. */ setx = x & (SX_LOCK_WAITERS | SX_LOCK_WRITE_SPINNER); if ((x & ~setx) == SX_LOCK_SHARED) { setx &= ~SX_LOCK_WRITE_SPINNER; if (!atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid | setx)) goto retry_sleepq; sleepq_release(&sx->lock_object); CTR2(KTR_LOCK, "%s: %p claimed by new writer", __func__, sx); break; } #ifdef ADAPTIVE_SX /* * It is possible we set the SX_LOCK_WRITE_SPINNER bit. * It is an invariant that when the bit is set, there is * a writer ready to grab the lock. Thus clear the bit since * we are going to sleep. */ if (in_critical) { if ((x & SX_LOCK_WRITE_SPINNER) || !((x & SX_LOCK_EXCLUSIVE_WAITERS))) { setx = x & ~SX_LOCK_WRITE_SPINNER; setx |= SX_LOCK_EXCLUSIVE_WAITERS; if (!atomic_fcmpset_ptr(&sx->sx_lock, &x, setx)) { goto retry_sleepq; } } critical_exit(); in_critical = false; } else { #endif /* * Try to set the SX_LOCK_EXCLUSIVE_WAITERS. If we fail, * than loop back and retry. */ if (!(x & SX_LOCK_EXCLUSIVE_WAITERS)) { if (!atomic_fcmpset_ptr(&sx->sx_lock, &x, x | SX_LOCK_EXCLUSIVE_WAITERS)) { goto retry_sleepq; } if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p set excl waiters flag", __func__, sx); } #ifdef ADAPTIVE_SX } #endif /* * Since we have been unable to acquire the exclusive * lock and the exclusive waiters flag is set, we have * to sleep. */ if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p blocking on sleep queue", __func__, sx); #ifdef KDTRACE_HOOKS sleep_time -= lockstat_nsecs(&sx->lock_object); #endif sleepq_add(&sx->lock_object, NULL, sx->lock_object.lo_name, SLEEPQ_SX | ((opts & SX_INTERRUPTIBLE) ? SLEEPQ_INTERRUPTIBLE : 0), SQ_EXCLUSIVE_QUEUE); /* * Hack: this can land in thread_suspend_check which will * conditionally take a mutex, tripping over an assert if a * lock we are waiting for is set. */ THREAD_CONTENTION_DONE(&sx->lock_object); if (!(opts & SX_INTERRUPTIBLE)) sleepq_wait(&sx->lock_object, 0); else error = sleepq_wait_sig(&sx->lock_object, 0); THREAD_CONTENDS_ON_LOCK(&sx->lock_object); #ifdef KDTRACE_HOOKS sleep_time += lockstat_nsecs(&sx->lock_object); sleep_cnt++; #endif if (error) { if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: interruptible sleep by %p suspended by signal", __func__, sx); break; } if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p resuming from sleep queue", __func__, sx); x = SX_READ_VALUE(sx); } THREAD_CONTENTION_DONE(&sx->lock_object); if (__predict_true(!extra_work)) return (error); #ifdef ADAPTIVE_SX if (in_critical) critical_exit(); #endif GIANT_RESTORE(); #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING) if (__predict_true(!doing_lockprof)) return (error); #endif #ifdef KDTRACE_HOOKS all_time += lockstat_nsecs(&sx->lock_object); if (sleep_time) LOCKSTAT_RECORD4(sx__block, sx, sleep_time, LOCKSTAT_WRITER, (state & SX_LOCK_SHARED) == 0, (state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state)); if (lda.spin_cnt > sleep_cnt) LOCKSTAT_RECORD4(sx__spin, sx, all_time - sleep_time, LOCKSTAT_WRITER, (state & SX_LOCK_SHARED) == 0, (state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state)); out_lockstat: #endif if (!error) LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx, contested, waittime, file, line, LOCKSTAT_WRITER); return (error); } /* * This function represents the so-called 'hard case' for sx_xunlock * operation. All 'easy case' failures are redirected to this. Note * that ideally this would be a static function, but it needs to be * accessible from at least sx.h. */ void _sx_xunlock_hard(struct sx *sx, uintptr_t x LOCK_FILE_LINE_ARG_DEF) { uintptr_t tid, setx; - int queue, wakeup_swapper; + int queue; if (SCHEDULER_STOPPED()) return; tid = (uintptr_t)curthread; if (__predict_false(x == tid)) x = SX_READ_VALUE(sx); MPASS(!(x & SX_LOCK_SHARED)); if (__predict_false(x & SX_LOCK_RECURSED)) { /* The lock is recursed, unrecurse one level. */ if ((--sx->sx_recurse) == 0) atomic_clear_ptr(&sx->sx_lock, SX_LOCK_RECURSED); if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p unrecursing", __func__, sx); return; } LOCKSTAT_PROFILE_RELEASE_RWLOCK(sx__release, sx, LOCKSTAT_WRITER); if (x == tid && atomic_cmpset_rel_ptr(&sx->sx_lock, tid, SX_LOCK_UNLOCKED)) return; if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p contested", __func__, sx); sleepq_lock(&sx->lock_object); x = SX_READ_VALUE(sx); MPASS(x & (SX_LOCK_SHARED_WAITERS | SX_LOCK_EXCLUSIVE_WAITERS)); /* * The wake up algorithm here is quite simple and probably not * ideal. It gives precedence to shared waiters if they are * present. For this condition, we have to preserve the * state of the exclusive waiters flag. * If interruptible sleeps left the shared queue empty avoid a * starvation for the threads sleeping on the exclusive queue by giving * them precedence and cleaning up the shared waiters bit anyway. */ setx = SX_LOCK_UNLOCKED; queue = SQ_SHARED_QUEUE; if ((x & SX_LOCK_EXCLUSIVE_WAITERS) != 0 && sleepq_sleepcnt(&sx->lock_object, SQ_EXCLUSIVE_QUEUE) != 0) { queue = SQ_EXCLUSIVE_QUEUE; setx |= (x & SX_LOCK_SHARED_WAITERS); } atomic_store_rel_ptr(&sx->sx_lock, setx); /* Wake up all the waiters for the specific queue. */ if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR3(KTR_LOCK, "%s: %p waking up all threads on %s queue", __func__, sx, queue == SQ_SHARED_QUEUE ? "shared" : "exclusive"); - wakeup_swapper = sleepq_broadcast(&sx->lock_object, SLEEPQ_SX, 0, - queue); + sleepq_broadcast(&sx->lock_object, SLEEPQ_SX, 0, queue); sleepq_release(&sx->lock_object); - if (wakeup_swapper) - kick_proc0(); } static __always_inline bool __sx_can_read(struct thread *td, uintptr_t x, bool fp) { if ((x & (SX_LOCK_SHARED | SX_LOCK_EXCLUSIVE_WAITERS | SX_LOCK_WRITE_SPINNER)) == SX_LOCK_SHARED) return (true); if (!fp && td->td_sx_slocks && (x & SX_LOCK_SHARED)) return (true); return (false); } static __always_inline bool __sx_slock_try(struct sx *sx, struct thread *td, uintptr_t *xp, bool fp LOCK_FILE_LINE_ARG_DEF) { /* * If no other thread has an exclusive lock then try to bump up * the count of sharers. Since we have to preserve the state * of SX_LOCK_EXCLUSIVE_WAITERS, if we fail to acquire the * shared lock loop back and retry. */ while (__sx_can_read(td, *xp, fp)) { if (atomic_fcmpset_acq_ptr(&sx->sx_lock, xp, *xp + SX_ONE_SHARER)) { if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR4(KTR_LOCK, "%s: %p succeed %p -> %p", __func__, sx, (void *)*xp, (void *)(*xp + SX_ONE_SHARER)); td->td_sx_slocks++; return (true); } } return (false); } static int __noinline _sx_slock_hard(struct sx *sx, int opts, uintptr_t x LOCK_FILE_LINE_ARG_DEF) { GIANT_DECLARE; struct thread *td; #ifdef ADAPTIVE_SX struct thread *owner; u_int i, n, spintries = 0; #endif #ifdef LOCK_PROFILING uint64_t waittime = 0; int contested = 0; #endif int error = 0; #if defined(ADAPTIVE_SX) || defined(KDTRACE_HOOKS) struct lock_delay_arg lda; #endif #ifdef KDTRACE_HOOKS u_int sleep_cnt = 0; int64_t sleep_time = 0; int64_t all_time = 0; #endif #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING) uintptr_t state = 0; #endif int extra_work __sdt_used = 0; td = curthread; #ifdef KDTRACE_HOOKS if (LOCKSTAT_PROFILE_ENABLED(sx__acquire)) { if (__sx_slock_try(sx, td, &x, false LOCK_FILE_LINE_ARG)) goto out_lockstat; extra_work = 1; all_time -= lockstat_nsecs(&sx->lock_object); state = x; } #endif #ifdef LOCK_PROFILING extra_work = 1; state = x; #endif if (SCHEDULER_STOPPED()) return (0); #if defined(ADAPTIVE_SX) lock_delay_arg_init(&lda, &sx_delay); #elif defined(KDTRACE_HOOKS) lock_delay_arg_init_noadapt(&lda); #endif #ifdef HWPMC_HOOKS PMC_SOFT_CALL( , , lock, failed); #endif lock_profile_obtain_lock_failed(&sx->lock_object, false, &contested, &waittime); #ifndef INVARIANTS GIANT_SAVE(extra_work); #endif THREAD_CONTENDS_ON_LOCK(&sx->lock_object); /* * As with rwlocks, we don't make any attempt to try to block * shared locks once there is an exclusive waiter. */ for (;;) { if (__sx_slock_try(sx, td, &x, false LOCK_FILE_LINE_ARG)) break; #ifdef INVARIANTS GIANT_SAVE(extra_work); #endif #ifdef KDTRACE_HOOKS lda.spin_cnt++; #endif #ifdef ADAPTIVE_SX /* * If the owner is running on another CPU, spin until * the owner stops running or the state of the lock * changes. */ if ((x & SX_LOCK_SHARED) == 0) { owner = lv_sx_owner(x); if (TD_IS_RUNNING(owner)) { if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR3(KTR_LOCK, "%s: spinning on %p held by %p", __func__, sx, owner); KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread), "spinning", "lockname:\"%s\"", sx->lock_object.lo_name); do { lock_delay(&lda); x = SX_READ_VALUE(sx); owner = lv_sx_owner(x); } while (owner != NULL && TD_IS_RUNNING(owner)); KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread), "running"); continue; } } else { if ((x & SX_LOCK_WRITE_SPINNER) && SX_SHARERS(x) == 0) { MPASS(!__sx_can_read(td, x, false)); lock_delay_spin(2); x = SX_READ_VALUE(sx); continue; } if (spintries < asx_retries) { KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread), "spinning", "lockname:\"%s\"", sx->lock_object.lo_name); n = SX_SHARERS(x); for (i = 0; i < asx_loops; i += n) { lock_delay_spin(n); x = SX_READ_VALUE(sx); if (!(x & SX_LOCK_SHARED)) break; n = SX_SHARERS(x); if (n == 0) break; if (__sx_can_read(td, x, false)) break; } #ifdef KDTRACE_HOOKS lda.spin_cnt += i; #endif KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread), "running"); if (i < asx_loops) continue; } } #endif /* * Some other thread already has an exclusive lock, so * start the process of blocking. */ sleepq_lock(&sx->lock_object); x = SX_READ_VALUE(sx); retry_sleepq: if (((x & SX_LOCK_WRITE_SPINNER) && SX_SHARERS(x) == 0) || __sx_can_read(td, x, false)) { sleepq_release(&sx->lock_object); continue; } #ifdef ADAPTIVE_SX /* * If the owner is running on another CPU, spin until * the owner stops running or the state of the lock * changes. */ if (!(x & SX_LOCK_SHARED)) { owner = (struct thread *)SX_OWNER(x); if (TD_IS_RUNNING(owner)) { sleepq_release(&sx->lock_object); x = SX_READ_VALUE(sx); continue; } } #endif /* * Try to set the SX_LOCK_SHARED_WAITERS flag. If we * fail to set it drop the sleep queue lock and loop * back. */ if (!(x & SX_LOCK_SHARED_WAITERS)) { if (!atomic_fcmpset_ptr(&sx->sx_lock, &x, x | SX_LOCK_SHARED_WAITERS)) goto retry_sleepq; if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p set shared waiters flag", __func__, sx); } /* * Since we have been unable to acquire the shared lock, * we have to sleep. */ if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p blocking on sleep queue", __func__, sx); #ifdef KDTRACE_HOOKS sleep_time -= lockstat_nsecs(&sx->lock_object); #endif sleepq_add(&sx->lock_object, NULL, sx->lock_object.lo_name, SLEEPQ_SX | ((opts & SX_INTERRUPTIBLE) ? SLEEPQ_INTERRUPTIBLE : 0), SQ_SHARED_QUEUE); /* * Hack: this can land in thread_suspend_check which will * conditionally take a mutex, tripping over an assert if a * lock we are waiting for is set. */ THREAD_CONTENTION_DONE(&sx->lock_object); if (!(opts & SX_INTERRUPTIBLE)) sleepq_wait(&sx->lock_object, 0); else error = sleepq_wait_sig(&sx->lock_object, 0); THREAD_CONTENDS_ON_LOCK(&sx->lock_object); #ifdef KDTRACE_HOOKS sleep_time += lockstat_nsecs(&sx->lock_object); sleep_cnt++; #endif if (error) { if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: interruptible sleep by %p suspended by signal", __func__, sx); break; } if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p resuming from sleep queue", __func__, sx); x = SX_READ_VALUE(sx); } THREAD_CONTENTION_DONE(&sx->lock_object); #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING) if (__predict_true(!extra_work)) return (error); #endif #ifdef KDTRACE_HOOKS all_time += lockstat_nsecs(&sx->lock_object); if (sleep_time) LOCKSTAT_RECORD4(sx__block, sx, sleep_time, LOCKSTAT_READER, (state & SX_LOCK_SHARED) == 0, (state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state)); if (lda.spin_cnt > sleep_cnt) LOCKSTAT_RECORD4(sx__spin, sx, all_time - sleep_time, LOCKSTAT_READER, (state & SX_LOCK_SHARED) == 0, (state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state)); out_lockstat: #endif if (error == 0) { LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx, contested, waittime, file, line, LOCKSTAT_READER); } GIANT_RESTORE(); return (error); } int _sx_slock_int(struct sx *sx, int opts LOCK_FILE_LINE_ARG_DEF) { struct thread *td; uintptr_t x; int error; KASSERT(kdb_active != 0 || SCHEDULER_STOPPED() || !TD_IS_IDLETHREAD(curthread), ("sx_slock() by idle thread %p on sx %s @ %s:%d", curthread, sx->lock_object.lo_name, file, line)); KASSERT(sx->sx_lock != SX_LOCK_DESTROYED, ("sx_slock() of destroyed sx @ %s:%d", file, line)); WITNESS_CHECKORDER(&sx->lock_object, LOP_NEWORDER, file, line, NULL); error = 0; td = curthread; x = SX_READ_VALUE(sx); if (__predict_false(LOCKSTAT_PROFILE_ENABLED(sx__acquire) || !__sx_slock_try(sx, td, &x, true LOCK_FILE_LINE_ARG))) error = _sx_slock_hard(sx, opts, x LOCK_FILE_LINE_ARG); else lock_profile_obtain_lock_success(&sx->lock_object, false, 0, 0, file, line); if (error == 0) { LOCK_LOG_LOCK("SLOCK", &sx->lock_object, 0, 0, file, line); WITNESS_LOCK(&sx->lock_object, 0, file, line); TD_LOCKS_INC(curthread); } return (error); } int _sx_slock(struct sx *sx, int opts, const char *file, int line) { return (_sx_slock_int(sx, opts LOCK_FILE_LINE_ARG)); } static __always_inline bool _sx_sunlock_try(struct sx *sx, struct thread *td, uintptr_t *xp) { for (;;) { if (SX_SHARERS(*xp) > 1 || !(*xp & SX_LOCK_WAITERS)) { if (atomic_fcmpset_rel_ptr(&sx->sx_lock, xp, *xp - SX_ONE_SHARER)) { if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR4(KTR_LOCK, "%s: %p succeeded %p -> %p", __func__, sx, (void *)*xp, (void *)(*xp - SX_ONE_SHARER)); td->td_sx_slocks--; return (true); } continue; } break; } return (false); } static void __noinline _sx_sunlock_hard(struct sx *sx, struct thread *td, uintptr_t x LOCK_FILE_LINE_ARG_DEF) { - int wakeup_swapper = 0; uintptr_t setx, queue; if (SCHEDULER_STOPPED()) return; if (_sx_sunlock_try(sx, td, &x)) goto out_lockstat; sleepq_lock(&sx->lock_object); x = SX_READ_VALUE(sx); for (;;) { if (_sx_sunlock_try(sx, td, &x)) break; /* * Wake up semantic here is quite simple: * Just wake up all the exclusive waiters. * Note that the state of the lock could have changed, * so if it fails loop back and retry. */ setx = SX_LOCK_UNLOCKED; queue = SQ_SHARED_QUEUE; if (x & SX_LOCK_EXCLUSIVE_WAITERS) { setx |= (x & SX_LOCK_SHARED_WAITERS); queue = SQ_EXCLUSIVE_QUEUE; } setx |= (x & SX_LOCK_WRITE_SPINNER); if (!atomic_fcmpset_rel_ptr(&sx->sx_lock, &x, setx)) continue; if (LOCK_LOG_TEST(&sx->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p waking up all thread on" "exclusive queue", __func__, sx); - wakeup_swapper = sleepq_broadcast(&sx->lock_object, SLEEPQ_SX, - 0, queue); + sleepq_broadcast(&sx->lock_object, SLEEPQ_SX, 0, queue); td->td_sx_slocks--; break; } sleepq_release(&sx->lock_object); - if (wakeup_swapper) - kick_proc0(); out_lockstat: LOCKSTAT_PROFILE_RELEASE_RWLOCK(sx__release, sx, LOCKSTAT_READER); } void _sx_sunlock_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF) { struct thread *td; uintptr_t x; KASSERT(sx->sx_lock != SX_LOCK_DESTROYED, ("sx_sunlock() of destroyed sx @ %s:%d", file, line)); _sx_assert(sx, SA_SLOCKED, file, line); WITNESS_UNLOCK(&sx->lock_object, 0, file, line); LOCK_LOG_LOCK("SUNLOCK", &sx->lock_object, 0, 0, file, line); td = curthread; x = SX_READ_VALUE(sx); if (__predict_false(LOCKSTAT_PROFILE_ENABLED(sx__release) || !_sx_sunlock_try(sx, td, &x))) _sx_sunlock_hard(sx, td, x LOCK_FILE_LINE_ARG); else lock_profile_release_lock(&sx->lock_object, false); TD_LOCKS_DEC(curthread); } void _sx_sunlock(struct sx *sx, const char *file, int line) { _sx_sunlock_int(sx LOCK_FILE_LINE_ARG); } #ifdef INVARIANT_SUPPORT #ifndef INVARIANTS #undef _sx_assert #endif /* * In the non-WITNESS case, sx_assert() can only detect that at least * *some* thread owns an slock, but it cannot guarantee that *this* * thread owns an slock. */ void _sx_assert(const struct sx *sx, int what, const char *file, int line) { #ifndef WITNESS int slocked = 0; #endif if (SCHEDULER_STOPPED()) return; switch (what) { case SA_SLOCKED: case SA_SLOCKED | SA_NOTRECURSED: case SA_SLOCKED | SA_RECURSED: #ifndef WITNESS slocked = 1; /* FALLTHROUGH */ #endif case SA_LOCKED: case SA_LOCKED | SA_NOTRECURSED: case SA_LOCKED | SA_RECURSED: #ifdef WITNESS witness_assert(&sx->lock_object, what, file, line); #else /* * If some other thread has an exclusive lock or we * have one and are asserting a shared lock, fail. * Also, if no one has a lock at all, fail. */ if (sx->sx_lock == SX_LOCK_UNLOCKED || (!(sx->sx_lock & SX_LOCK_SHARED) && (slocked || sx_xholder(sx) != curthread))) panic("Lock %s not %slocked @ %s:%d\n", sx->lock_object.lo_name, slocked ? "share " : "", file, line); if (!(sx->sx_lock & SX_LOCK_SHARED)) { if (sx_recursed(sx)) { if (what & SA_NOTRECURSED) panic("Lock %s recursed @ %s:%d\n", sx->lock_object.lo_name, file, line); } else if (what & SA_RECURSED) panic("Lock %s not recursed @ %s:%d\n", sx->lock_object.lo_name, file, line); } #endif break; case SA_XLOCKED: case SA_XLOCKED | SA_NOTRECURSED: case SA_XLOCKED | SA_RECURSED: if (sx_xholder(sx) != curthread) panic("Lock %s not exclusively locked @ %s:%d\n", sx->lock_object.lo_name, file, line); if (sx_recursed(sx)) { if (what & SA_NOTRECURSED) panic("Lock %s recursed @ %s:%d\n", sx->lock_object.lo_name, file, line); } else if (what & SA_RECURSED) panic("Lock %s not recursed @ %s:%d\n", sx->lock_object.lo_name, file, line); break; case SA_UNLOCKED: #ifdef WITNESS witness_assert(&sx->lock_object, what, file, line); #else /* * If we hold an exclusve lock fail. We can't * reliably check to see if we hold a shared lock or * not. */ if (sx_xholder(sx) == curthread) panic("Lock %s exclusively locked @ %s:%d\n", sx->lock_object.lo_name, file, line); #endif break; default: panic("Unknown sx lock assertion: %d @ %s:%d", what, file, line); } } #endif /* INVARIANT_SUPPORT */ #ifdef DDB static void db_show_sx(const struct lock_object *lock) { struct thread *td; const struct sx *sx; sx = (const struct sx *)lock; db_printf(" state: "); if (sx->sx_lock == SX_LOCK_UNLOCKED) db_printf("UNLOCKED\n"); else if (sx->sx_lock == SX_LOCK_DESTROYED) { db_printf("DESTROYED\n"); return; } else if (sx->sx_lock & SX_LOCK_SHARED) db_printf("SLOCK: %ju\n", (uintmax_t)SX_SHARERS(sx->sx_lock)); else { td = sx_xholder(sx); db_printf("XLOCK: %p (tid %d, pid %d, \"%s\")\n", td, td->td_tid, td->td_proc->p_pid, td->td_name); if (sx_recursed(sx)) db_printf(" recursed: %d\n", sx->sx_recurse); } db_printf(" waiters: "); switch(sx->sx_lock & (SX_LOCK_SHARED_WAITERS | SX_LOCK_EXCLUSIVE_WAITERS)) { case SX_LOCK_SHARED_WAITERS: db_printf("shared\n"); break; case SX_LOCK_EXCLUSIVE_WAITERS: db_printf("exclusive\n"); break; case SX_LOCK_SHARED_WAITERS | SX_LOCK_EXCLUSIVE_WAITERS: db_printf("exclusive and shared\n"); break; default: db_printf("none\n"); } } /* * Check to see if a thread that is blocked on a sleep queue is actually * blocked on an sx lock. If so, output some details and return true. * If the lock has an exclusive owner, return that in *ownerp. */ int sx_chain(struct thread *td, struct thread **ownerp) { const struct sx *sx; /* * Check to see if this thread is blocked on an sx lock. * First, we check the lock class. If that is ok, then we * compare the lock name against the wait message. */ sx = td->td_wchan; if (LOCK_CLASS(&sx->lock_object) != &lock_class_sx || sx->lock_object.lo_name != td->td_wmesg) return (0); /* We think we have an sx lock, so output some details. */ db_printf("blocked on sx \"%s\" ", td->td_wmesg); *ownerp = sx_xholder(sx); if (sx->sx_lock & SX_LOCK_SHARED) db_printf("SLOCK (count %ju)\n", (uintmax_t)SX_SHARERS(sx->sx_lock)); else db_printf("XLOCK\n"); return (1); } #endif diff --git a/sys/kern/kern_synch.c b/sys/kern/kern_synch.c index 9bb6079c0c20..8e956324ee23 100644 --- a/sys/kern/kern_synch.c +++ b/sys/kern/kern_synch.c @@ -1,696 +1,680 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1990, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include "opt_ktrace.h" #include "opt_sched.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #ifdef EPOCH_TRACE #include #endif #include static void synch_setup(void *dummy); SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, NULL); int hogticks; static const char pause_wchan[MAXCPU]; static struct callout loadav_callout; struct loadavg averunnable = { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ /* * Constants for averages over 1, 5, and 15 minutes * when sampling at 5 second intervals. */ static uint64_t cexp[3] = { 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 0.9944598480048967 * FSCALE, /* exp(-1/180) */ }; /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, "Fixed-point scale factor used for calculating load average values"); static void loadav(void *arg); SDT_PROVIDER_DECLARE(sched); SDT_PROBE_DEFINE(sched, , , preempt); static void sleepinit(void *unused) { hogticks = (hz / 10) * 2; /* Default only. */ init_sleepqueues(); } /* * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure * it is available. */ SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, NULL); /* * General sleep call. Suspends the current thread until a wakeup is * performed on the specified identifier. The thread will then be made * runnable with the specified priority. Sleeps at most sbt units of time * (0 means no timeout). If pri includes the PCATCH flag, let signals * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a * signal becomes pending, ERESTART is returned if the current system * call should be restarted if possible, and EINTR is returned if the system * call should be interrupted by the signal (return EINTR). * * The lock argument is unlocked before the caller is suspended, and * re-locked before _sleep() returns. If priority includes the PDROP * flag the lock is not re-locked before returning. */ int _sleep(const void *ident, struct lock_object *lock, int priority, const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) { struct thread *td __ktrace_used; struct lock_class *class; uintptr_t lock_state; int catch, pri, rval, sleepq_flags; WITNESS_SAVE_DECL(lock_witness); TSENTER(); td = curthread; #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(1, 0, wmesg); #endif WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, "Sleeping on \"%s\"", wmesg); KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL || (priority & PNOLOCK) != 0, ("sleeping without a lock")); KASSERT(ident != NULL, ("_sleep: NULL ident")); KASSERT(TD_IS_RUNNING(td), ("_sleep: curthread not running")); if (priority & PDROP) KASSERT(lock != NULL && lock != &Giant.lock_object, ("PDROP requires a non-Giant lock")); if (lock != NULL) class = LOCK_CLASS(lock); else class = NULL; if (SCHEDULER_STOPPED()) { if (lock != NULL && priority & PDROP) class->lc_unlock(lock); return (0); } catch = priority & PCATCH; pri = priority & PRIMASK; KASSERT(!TD_ON_SLEEPQ(td), ("recursive sleep")); if ((uintptr_t)ident >= (uintptr_t)&pause_wchan[0] && (uintptr_t)ident <= (uintptr_t)&pause_wchan[MAXCPU - 1]) sleepq_flags = SLEEPQ_PAUSE; else sleepq_flags = SLEEPQ_SLEEP; if (catch) sleepq_flags |= SLEEPQ_INTERRUPTIBLE; sleepq_lock(ident); CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", td->td_tid, td->td_proc->p_pid, td->td_name, wmesg, ident); if (lock == &Giant.lock_object) mtx_assert(&Giant, MA_OWNED); DROP_GIANT(); if (lock != NULL && lock != &Giant.lock_object && !(class->lc_flags & LC_SLEEPABLE)) { KASSERT(!(class->lc_flags & LC_SPINLOCK), ("spin locks can only use msleep_spin")); WITNESS_SAVE(lock, lock_witness); lock_state = class->lc_unlock(lock); } else /* GCC needs to follow the Yellow Brick Road */ lock_state = -1; /* * We put ourselves on the sleep queue and start our timeout * before calling thread_suspend_check, as we could stop there, * and a wakeup or a SIGCONT (or both) could occur while we were * stopped without resuming us. Thus, we must be ready for sleep * when cursig() is called. If the wakeup happens while we're * stopped, then td will no longer be on a sleep queue upon * return from cursig(). */ sleepq_add(ident, lock, wmesg, sleepq_flags, 0); if (sbt != 0) sleepq_set_timeout_sbt(ident, sbt, pr, flags); if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { sleepq_release(ident); WITNESS_SAVE(lock, lock_witness); lock_state = class->lc_unlock(lock); sleepq_lock(ident); } if (sbt != 0 && catch) rval = sleepq_timedwait_sig(ident, pri); else if (sbt != 0) rval = sleepq_timedwait(ident, pri); else if (catch) rval = sleepq_wait_sig(ident, pri); else { sleepq_wait(ident, pri); rval = 0; } #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 0, wmesg); #endif PICKUP_GIANT(); if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { class->lc_lock(lock, lock_state); WITNESS_RESTORE(lock, lock_witness); } TSEXIT(); return (rval); } int msleep_spin_sbt(const void *ident, struct mtx *mtx, const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) { struct thread *td __ktrace_used; int rval; WITNESS_SAVE_DECL(mtx); td = curthread; KASSERT(mtx != NULL, ("sleeping without a mutex")); KASSERT(ident != NULL, ("msleep_spin_sbt: NULL ident")); KASSERT(TD_IS_RUNNING(td), ("msleep_spin_sbt: curthread not running")); if (SCHEDULER_STOPPED()) return (0); sleepq_lock(ident); CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", td->td_tid, td->td_proc->p_pid, td->td_name, wmesg, ident); DROP_GIANT(); mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); WITNESS_SAVE(&mtx->lock_object, mtx); mtx_unlock_spin(mtx); /* * We put ourselves on the sleep queue and start our timeout. */ sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); if (sbt != 0) sleepq_set_timeout_sbt(ident, sbt, pr, flags); /* * Can't call ktrace with any spin locks held so it can lock the * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold * any spin lock. Thus, we have to drop the sleepq spin lock while * we handle those requests. This is safe since we have placed our * thread on the sleep queue already. */ #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) { sleepq_release(ident); ktrcsw(1, 0, wmesg); sleepq_lock(ident); } #endif #ifdef WITNESS sleepq_release(ident); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", wmesg); sleepq_lock(ident); #endif if (sbt != 0) rval = sleepq_timedwait(ident, 0); else { sleepq_wait(ident, 0); rval = 0; } #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 0, wmesg); #endif PICKUP_GIANT(); mtx_lock_spin(mtx); WITNESS_RESTORE(&mtx->lock_object, mtx); return (rval); } /* * pause_sbt() delays the calling thread by the given signed binary * time. During cold bootup, pause_sbt() uses the DELAY() function * instead of the _sleep() function to do the waiting. The "sbt" * argument must be greater than or equal to zero. A "sbt" value of * zero is equivalent to a "sbt" value of one tick. */ int pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) { KASSERT(sbt >= 0, ("pause_sbt: timeout must be >= 0")); /* silently convert invalid timeouts */ if (sbt == 0) sbt = tick_sbt; if ((cold && curthread == &thread0) || kdb_active || SCHEDULER_STOPPED()) { /* * We delay one second at a time to avoid overflowing the * system specific DELAY() function(s): */ while (sbt >= SBT_1S) { DELAY(1000000); sbt -= SBT_1S; } /* Do the delay remainder, if any */ sbt = howmany(sbt, SBT_1US); if (sbt > 0) DELAY(sbt); return (EWOULDBLOCK); } return (_sleep(&pause_wchan[curcpu], NULL, (flags & C_CATCH) ? PCATCH : 0, wmesg, sbt, pr, flags)); } /* * Make all threads sleeping on the specified identifier runnable. */ void wakeup(const void *ident) { - int wakeup_swapper; - sleepq_lock(ident); - wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); + sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); sleepq_release(ident); - if (wakeup_swapper) { - KASSERT(ident != &proc0, - ("wakeup and wakeup_swapper and proc0")); - kick_proc0(); - } } /* * Make a thread sleeping on the specified identifier runnable. * May wake more than one thread if a target thread is currently * swapped out. */ void wakeup_one(const void *ident) { - int wakeup_swapper; - sleepq_lock(ident); - wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP | SLEEPQ_DROP, 0, 0); - if (wakeup_swapper) - kick_proc0(); + sleepq_signal(ident, SLEEPQ_SLEEP | SLEEPQ_DROP, 0, 0); } void wakeup_any(const void *ident) { - int wakeup_swapper; - sleepq_lock(ident); - wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP | SLEEPQ_UNFAIR | - SLEEPQ_DROP, 0, 0); - if (wakeup_swapper) - kick_proc0(); + sleepq_signal(ident, SLEEPQ_SLEEP | SLEEPQ_UNFAIR | SLEEPQ_DROP, 0, 0); } /* * Signal sleeping waiters after the counter has reached zero. */ void _blockcount_wakeup(blockcount_t *bc, u_int old) { KASSERT(_BLOCKCOUNT_WAITERS(old), ("%s: no waiters on %p", __func__, bc)); if (atomic_cmpset_int(&bc->__count, _BLOCKCOUNT_WAITERS_FLAG, 0)) wakeup(bc); } /* * Wait for a wakeup or a signal. This does not guarantee that the count is * still zero on return. Callers wanting a precise answer should use * blockcount_wait() with an interlock. * * If there is no work to wait for, return 0. If the sleep was interrupted by a * signal, return EINTR or ERESTART, and return EAGAIN otherwise. */ int _blockcount_sleep(blockcount_t *bc, struct lock_object *lock, const char *wmesg, int prio) { void *wchan; uintptr_t lock_state; u_int old; int ret; bool catch, drop; KASSERT(lock != &Giant.lock_object, ("%s: cannot use Giant as the interlock", __func__)); catch = (prio & PCATCH) != 0; drop = (prio & PDROP) != 0; prio &= PRIMASK; /* * Synchronize with the fence in blockcount_release(). If we end up * waiting, the sleepqueue lock acquisition will provide the required * side effects. * * If there is no work to wait for, but waiters are present, try to put * ourselves to sleep to avoid jumping ahead. */ if (atomic_load_acq_int(&bc->__count) == 0) { if (lock != NULL && drop) LOCK_CLASS(lock)->lc_unlock(lock); return (0); } lock_state = 0; wchan = bc; sleepq_lock(wchan); DROP_GIANT(); if (lock != NULL) lock_state = LOCK_CLASS(lock)->lc_unlock(lock); old = blockcount_read(bc); ret = 0; do { if (_BLOCKCOUNT_COUNT(old) == 0) { sleepq_release(wchan); goto out; } if (_BLOCKCOUNT_WAITERS(old)) break; } while (!atomic_fcmpset_int(&bc->__count, &old, old | _BLOCKCOUNT_WAITERS_FLAG)); sleepq_add(wchan, NULL, wmesg, catch ? SLEEPQ_INTERRUPTIBLE : 0, 0); if (catch) ret = sleepq_wait_sig(wchan, prio); else sleepq_wait(wchan, prio); if (ret == 0) ret = EAGAIN; out: PICKUP_GIANT(); if (lock != NULL && !drop) LOCK_CLASS(lock)->lc_lock(lock, lock_state); return (ret); } static void kdb_switch(void) { thread_unlock(curthread); kdb_backtrace(); kdb_reenter(); panic("%s: did not reenter debugger", __func__); } /* * mi_switch(9): The machine-independent parts of context switching. * * The thread lock is required on entry and is no longer held on return. */ void mi_switch(int flags) { uint64_t runtime, new_switchtime; struct thread *td; td = curthread; /* XXX */ THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); #ifdef INVARIANTS if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) mtx_assert(&Giant, MA_NOTOWNED); #endif /* thread_lock() performs spinlock_enter(). */ KASSERT(td->td_critnest == 1 || KERNEL_PANICKED(), ("mi_switch: switch in a critical section")); KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, ("mi_switch: switch must be voluntary or involuntary")); KASSERT((flags & SW_TYPE_MASK) != 0, ("mi_switch: a switch reason (type) must be specified")); KASSERT((flags & SW_TYPE_MASK) < SWT_COUNT, ("mi_switch: invalid switch reason %d", (flags & SW_TYPE_MASK))); /* * Don't perform context switches from the debugger. */ if (kdb_active) kdb_switch(); if (SCHEDULER_STOPPED()) return; if (flags & SW_VOL) { td->td_ru.ru_nvcsw++; td->td_swvoltick = ticks; } else { td->td_ru.ru_nivcsw++; td->td_swinvoltick = ticks; } #ifdef SCHED_STATS SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); #endif /* * Compute the amount of time during which the current * thread was running, and add that to its total so far. */ new_switchtime = cpu_ticks(); runtime = new_switchtime - PCPU_GET(switchtime); td->td_runtime += runtime; td->td_incruntime += runtime; PCPU_SET(switchtime, new_switchtime); td->td_generation++; /* bump preempt-detect counter */ VM_CNT_INC(v_swtch); PCPU_SET(switchticks, ticks); CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name); #ifdef KDTRACE_HOOKS if (SDT_PROBES_ENABLED() && ((flags & SW_PREEMPT) != 0 || ((flags & SW_INVOL) != 0 && (flags & SW_TYPE_MASK) == SWT_NEEDRESCHED))) SDT_PROBE0(sched, , , preempt); #endif sched_switch(td, flags); CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name); /* * If the last thread was exiting, finish cleaning it up. */ if ((td = PCPU_GET(deadthread))) { PCPU_SET(deadthread, NULL); thread_stash(td); } spinlock_exit(); } /* * Change thread state to be runnable, placing it on the run queue. * * Requires the thread lock on entry, drops on exit. */ void setrunnable(struct thread *td, int srqflags) { THREAD_LOCK_ASSERT(td, MA_OWNED); KASSERT(td->td_proc->p_state != PRS_ZOMBIE, ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); switch (TD_GET_STATE(td)) { case TDS_RUNNING: case TDS_RUNQ: case TDS_INHIBITED: if ((srqflags & (SRQ_HOLD | SRQ_HOLDTD)) == 0) thread_unlock(td); break; case TDS_CAN_RUN: KASSERT((td->td_flags & TDF_INMEM) != 0, ("setrunnable: td %p not in mem, flags 0x%X inhibit 0x%X", td, td->td_flags, td->td_inhibitors)); /* unlocks thread lock according to flags */ sched_wakeup(td, srqflags); break; default: panic("setrunnable: state 0x%x", TD_GET_STATE(td)); } } /* * Compute a tenex style load average of a quantity on * 1, 5 and 15 minute intervals. */ static void loadav(void *arg) { int i; uint64_t nrun; struct loadavg *avg; nrun = (uint64_t)sched_load(); avg = &averunnable; for (i = 0; i < 3; i++) avg->ldavg[i] = (cexp[i] * (uint64_t)avg->ldavg[i] + nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; /* * Schedule the next update to occur after 5 seconds, but add a * random variation to avoid synchronisation with processes that * run at regular intervals. */ callout_reset_sbt(&loadav_callout, SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US, loadav, NULL, C_DIRECT_EXEC | C_PREL(32)); } static void ast_scheduler(struct thread *td, int tda __unused) { #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(1, 1, __func__); #endif thread_lock(td); sched_prio(td, td->td_user_pri); mi_switch(SW_INVOL | SWT_NEEDRESCHED); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 1, __func__); #endif } static void synch_setup(void *dummy __unused) { callout_init(&loadav_callout, 1); ast_register(TDA_SCHED, ASTR_ASTF_REQUIRED, 0, ast_scheduler); /* Kick off timeout driven events by calling first time. */ loadav(NULL); } bool should_yield(void) { return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks); } void maybe_yield(void) { if (should_yield()) kern_yield(PRI_USER); } void kern_yield(int prio) { struct thread *td; td = curthread; DROP_GIANT(); thread_lock(td); if (prio == PRI_USER) prio = td->td_user_pri; if (prio >= 0) sched_prio(td, prio); mi_switch(SW_VOL | SWT_RELINQUISH); PICKUP_GIANT(); } /* * General purpose yield system call. */ int sys_yield(struct thread *td, struct yield_args *uap) { thread_lock(td); if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) sched_prio(td, PRI_MAX_TIMESHARE); mi_switch(SW_VOL | SWT_RELINQUISH); td->td_retval[0] = 0; return (0); } int sys_sched_getcpu(struct thread *td, struct sched_getcpu_args *uap) { td->td_retval[0] = td->td_oncpu; return (0); } diff --git a/sys/kern/subr_sleepqueue.c b/sys/kern/subr_sleepqueue.c index 0b792d752e94..0af666653352 100644 --- a/sys/kern/subr_sleepqueue.c +++ b/sys/kern/subr_sleepqueue.c @@ -1,1511 +1,1507 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2004 John Baldwin * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * Implementation of sleep queues used to hold queue of threads blocked on * a wait channel. Sleep queues are different from turnstiles in that wait * channels are not owned by anyone, so there is no priority propagation. * Sleep queues can also provide a timeout and can also be interrupted by * signals. That said, there are several similarities between the turnstile * and sleep queue implementations. (Note: turnstiles were implemented * first.) For example, both use a hash table of the same size where each * bucket is referred to as a "chain" that contains both a spin lock and * a linked list of queues. An individual queue is located by using a hash * to pick a chain, locking the chain, and then walking the chain searching * for the queue. This means that a wait channel object does not need to * embed its queue head just as locks do not embed their turnstile queue * head. Threads also carry around a sleep queue that they lend to the * wait channel when blocking. Just as in turnstiles, the queue includes * a free list of the sleep queues of other threads blocked on the same * wait channel in the case of multiple waiters. * * Some additional functionality provided by sleep queues include the * ability to set a timeout. The timeout is managed using a per-thread * callout that resumes a thread if it is asleep. A thread may also * catch signals while it is asleep (aka an interruptible sleep). The * signal code uses sleepq_abort() to interrupt a sleeping thread. Finally, * sleep queues also provide some extra assertions. One is not allowed to * mix the sleep/wakeup and cv APIs for a given wait channel. Also, one * must consistently use the same lock to synchronize with a wait channel, * though this check is currently only a warning for sleep/wakeup due to * pre-existing abuse of that API. The same lock must also be held when * awakening threads, though that is currently only enforced for condition * variables. */ #include #include "opt_sleepqueue_profiling.h" #include "opt_ddb.h" #include "opt_sched.h" #include "opt_stack.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef EPOCH_TRACE #include #endif #include #include #ifdef DDB #include #endif /* * Constants for the hash table of sleep queue chains. * SC_TABLESIZE must be a power of two for SC_MASK to work properly. */ #ifndef SC_TABLESIZE #define SC_TABLESIZE 256 #endif CTASSERT(powerof2(SC_TABLESIZE)); #define SC_MASK (SC_TABLESIZE - 1) #define SC_SHIFT 8 #define SC_HASH(wc) ((((uintptr_t)(wc) >> SC_SHIFT) ^ (uintptr_t)(wc)) & \ SC_MASK) #define SC_LOOKUP(wc) &sleepq_chains[SC_HASH(wc)] #define NR_SLEEPQS 2 /* * There are two different lists of sleep queues. Both lists are connected * via the sq_hash entries. The first list is the sleep queue chain list * that a sleep queue is on when it is attached to a wait channel. The * second list is the free list hung off of a sleep queue that is attached * to a wait channel. * * Each sleep queue also contains the wait channel it is attached to, the * list of threads blocked on that wait channel, flags specific to the * wait channel, and the lock used to synchronize with a wait channel. * The flags are used to catch mismatches between the various consumers * of the sleep queue API (e.g. sleep/wakeup and condition variables). * The lock pointer is only used when invariants are enabled for various * debugging checks. * * Locking key: * c - sleep queue chain lock */ struct sleepqueue { struct threadqueue sq_blocked[NR_SLEEPQS]; /* (c) Blocked threads. */ u_int sq_blockedcnt[NR_SLEEPQS]; /* (c) N. of blocked threads. */ LIST_ENTRY(sleepqueue) sq_hash; /* (c) Chain and free list. */ LIST_HEAD(, sleepqueue) sq_free; /* (c) Free queues. */ const void *sq_wchan; /* (c) Wait channel. */ int sq_type; /* (c) Queue type. */ #ifdef INVARIANTS struct lock_object *sq_lock; /* (c) Associated lock. */ #endif }; struct sleepqueue_chain { LIST_HEAD(, sleepqueue) sc_queues; /* List of sleep queues. */ struct mtx sc_lock; /* Spin lock for this chain. */ #ifdef SLEEPQUEUE_PROFILING u_int sc_depth; /* Length of sc_queues. */ u_int sc_max_depth; /* Max length of sc_queues. */ #endif } __aligned(CACHE_LINE_SIZE); #ifdef SLEEPQUEUE_PROFILING static SYSCTL_NODE(_debug, OID_AUTO, sleepq, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "sleepq profiling"); static SYSCTL_NODE(_debug_sleepq, OID_AUTO, chains, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "sleepq chain stats"); static u_int sleepq_max_depth; SYSCTL_UINT(_debug_sleepq, OID_AUTO, max_depth, CTLFLAG_RD, &sleepq_max_depth, 0, "maxmimum depth achieved of a single chain"); static void sleepq_profile(const char *wmesg); static int prof_enabled; #endif static struct sleepqueue_chain sleepq_chains[SC_TABLESIZE]; static uma_zone_t sleepq_zone; /* * Prototypes for non-exported routines. */ static int sleepq_catch_signals(const void *wchan, int pri); static inline int sleepq_check_signals(void); static inline int sleepq_check_timeout(void); #ifdef INVARIANTS static void sleepq_dtor(void *mem, int size, void *arg); #endif static int sleepq_init(void *mem, int size, int flags); static void sleepq_resume_thread(struct sleepqueue *sq, struct thread *td, int pri, int srqflags); static void sleepq_remove_thread(struct sleepqueue *sq, struct thread *td); static void sleepq_switch(const void *wchan, int pri); static void sleepq_timeout(void *arg); SDT_PROBE_DECLARE(sched, , , sleep); SDT_PROBE_DECLARE(sched, , , wakeup); /* * Initialize SLEEPQUEUE_PROFILING specific sysctl nodes. * Note that it must happen after sleepinit() has been fully executed, so * it must happen after SI_SUB_KMEM SYSINIT() subsystem setup. */ #ifdef SLEEPQUEUE_PROFILING static void init_sleepqueue_profiling(void) { char chain_name[10]; struct sysctl_oid *chain_oid; u_int i; for (i = 0; i < SC_TABLESIZE; i++) { snprintf(chain_name, sizeof(chain_name), "%u", i); chain_oid = SYSCTL_ADD_NODE(NULL, SYSCTL_STATIC_CHILDREN(_debug_sleepq_chains), OID_AUTO, chain_name, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "sleepq chain stats"); SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO, "depth", CTLFLAG_RD, &sleepq_chains[i].sc_depth, 0, NULL); SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO, "max_depth", CTLFLAG_RD, &sleepq_chains[i].sc_max_depth, 0, NULL); } } SYSINIT(sleepqueue_profiling, SI_SUB_LOCK, SI_ORDER_ANY, init_sleepqueue_profiling, NULL); #endif /* * Early initialization of sleep queues that is called from the sleepinit() * SYSINIT. */ void init_sleepqueues(void) { int i; for (i = 0; i < SC_TABLESIZE; i++) { LIST_INIT(&sleepq_chains[i].sc_queues); mtx_init(&sleepq_chains[i].sc_lock, "sleepq chain", NULL, MTX_SPIN); } sleepq_zone = uma_zcreate("SLEEPQUEUE", sizeof(struct sleepqueue), #ifdef INVARIANTS NULL, sleepq_dtor, sleepq_init, NULL, UMA_ALIGN_CACHE, 0); #else NULL, NULL, sleepq_init, NULL, UMA_ALIGN_CACHE, 0); #endif thread0.td_sleepqueue = sleepq_alloc(); } /* * Get a sleep queue for a new thread. */ struct sleepqueue * sleepq_alloc(void) { return (uma_zalloc(sleepq_zone, M_WAITOK)); } /* * Free a sleep queue when a thread is destroyed. */ void sleepq_free(struct sleepqueue *sq) { uma_zfree(sleepq_zone, sq); } /* * Lock the sleep queue chain associated with the specified wait channel. */ void sleepq_lock(const void *wchan) { struct sleepqueue_chain *sc; sc = SC_LOOKUP(wchan); mtx_lock_spin(&sc->sc_lock); } /* * Look up the sleep queue associated with a given wait channel in the hash * table locking the associated sleep queue chain. If no queue is found in * the table, NULL is returned. */ struct sleepqueue * sleepq_lookup(const void *wchan) { struct sleepqueue_chain *sc; struct sleepqueue *sq; KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__)); sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); LIST_FOREACH(sq, &sc->sc_queues, sq_hash) if (sq->sq_wchan == wchan) return (sq); return (NULL); } /* * Unlock the sleep queue chain associated with a given wait channel. */ void sleepq_release(const void *wchan) { struct sleepqueue_chain *sc; sc = SC_LOOKUP(wchan); mtx_unlock_spin(&sc->sc_lock); } /* * Places the current thread on the sleep queue for the specified wait * channel. If INVARIANTS is enabled, then it associates the passed in * lock with the sleepq to make sure it is held when that sleep queue is * woken up. */ void sleepq_add(const void *wchan, struct lock_object *lock, const char *wmesg, int flags, int queue) { struct sleepqueue_chain *sc; struct sleepqueue *sq; struct thread *td; td = curthread; sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); MPASS(td->td_sleepqueue != NULL); MPASS(wchan != NULL); MPASS((queue >= 0) && (queue < NR_SLEEPQS)); /* If this thread is not allowed to sleep, die a horrible death. */ if (__predict_false(!THREAD_CAN_SLEEP())) { #ifdef EPOCH_TRACE epoch_trace_list(curthread); #endif KASSERT(0, ("%s: td %p to sleep on wchan %p with sleeping prohibited", __func__, td, wchan)); } /* Look up the sleep queue associated with the wait channel 'wchan'. */ sq = sleepq_lookup(wchan); /* * If the wait channel does not already have a sleep queue, use * this thread's sleep queue. Otherwise, insert the current thread * into the sleep queue already in use by this wait channel. */ if (sq == NULL) { #ifdef INVARIANTS int i; sq = td->td_sleepqueue; for (i = 0; i < NR_SLEEPQS; i++) { KASSERT(TAILQ_EMPTY(&sq->sq_blocked[i]), ("thread's sleep queue %d is not empty", i)); KASSERT(sq->sq_blockedcnt[i] == 0, ("thread's sleep queue %d count mismatches", i)); } KASSERT(LIST_EMPTY(&sq->sq_free), ("thread's sleep queue has a non-empty free list")); KASSERT(sq->sq_wchan == NULL, ("stale sq_wchan pointer")); sq->sq_lock = lock; #endif #ifdef SLEEPQUEUE_PROFILING sc->sc_depth++; if (sc->sc_depth > sc->sc_max_depth) { sc->sc_max_depth = sc->sc_depth; if (sc->sc_max_depth > sleepq_max_depth) sleepq_max_depth = sc->sc_max_depth; } #endif sq = td->td_sleepqueue; LIST_INSERT_HEAD(&sc->sc_queues, sq, sq_hash); sq->sq_wchan = wchan; sq->sq_type = flags & SLEEPQ_TYPE; } else { MPASS(wchan == sq->sq_wchan); MPASS(lock == sq->sq_lock); MPASS((flags & SLEEPQ_TYPE) == sq->sq_type); LIST_INSERT_HEAD(&sq->sq_free, td->td_sleepqueue, sq_hash); } thread_lock(td); TAILQ_INSERT_TAIL(&sq->sq_blocked[queue], td, td_slpq); sq->sq_blockedcnt[queue]++; td->td_sleepqueue = NULL; td->td_sqqueue = queue; td->td_wchan = wchan; td->td_wmesg = wmesg; if (flags & SLEEPQ_INTERRUPTIBLE) { td->td_intrval = 0; td->td_flags |= TDF_SINTR; } td->td_flags &= ~TDF_TIMEOUT; thread_unlock(td); } /* * Sets a timeout that will remove the current thread from the * specified sleep queue at the specified time if the thread has not * already been awakened. Flags are from C_* (callout) namespace. */ void sleepq_set_timeout_sbt(const void *wchan, sbintime_t sbt, sbintime_t pr, int flags) { struct sleepqueue_chain *sc __unused; struct thread *td; sbintime_t pr1; td = curthread; sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); MPASS(TD_ON_SLEEPQ(td)); MPASS(td->td_sleepqueue == NULL); MPASS(wchan != NULL); if (cold && td == &thread0) panic("timed sleep before timers are working"); KASSERT(td->td_sleeptimo == 0, ("td %d %p td_sleeptimo %jx", td->td_tid, td, (uintmax_t)td->td_sleeptimo)); thread_lock(td); callout_when(sbt, pr, flags, &td->td_sleeptimo, &pr1); thread_unlock(td); callout_reset_sbt_on(&td->td_slpcallout, td->td_sleeptimo, pr1, sleepq_timeout, td, PCPU_GET(cpuid), flags | C_PRECALC | C_DIRECT_EXEC); } /* * Return the number of actual sleepers for the specified queue. */ u_int sleepq_sleepcnt(const void *wchan, int queue) { struct sleepqueue *sq; KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__)); MPASS((queue >= 0) && (queue < NR_SLEEPQS)); sq = sleepq_lookup(wchan); if (sq == NULL) return (0); return (sq->sq_blockedcnt[queue]); } static int sleepq_check_ast_sc_locked(struct thread *td, struct sleepqueue_chain *sc) { struct proc *p; int ret; mtx_assert(&sc->sc_lock, MA_OWNED); if ((td->td_pflags & TDP_WAKEUP) != 0) { td->td_pflags &= ~TDP_WAKEUP; thread_lock(td); return (EINTR); } /* * See if there are any pending signals or suspension requests for this * thread. If not, we can switch immediately. */ thread_lock(td); if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND)) return (0); thread_unlock(td); mtx_unlock_spin(&sc->sc_lock); p = td->td_proc; CTR3(KTR_PROC, "sleepq catching signals: thread %p (pid %ld, %s)", (void *)td, (long)p->p_pid, td->td_name); PROC_LOCK(p); /* * Check for suspension first. Checking for signals and then * suspending could result in a missed signal, since a signal * can be delivered while this thread is suspended. */ ret = sig_ast_checksusp(td); if (ret != 0) { PROC_UNLOCK(p); mtx_lock_spin(&sc->sc_lock); thread_lock(td); return (ret); } ret = sig_ast_needsigchk(td); /* * Lock the per-process spinlock prior to dropping the * PROC_LOCK to avoid a signal delivery race. * PROC_LOCK, PROC_SLOCK, and thread_lock() are * currently held in tdsendsignal() and thread_single(). */ PROC_SLOCK(p); mtx_lock_spin(&sc->sc_lock); PROC_UNLOCK(p); thread_lock(td); PROC_SUNLOCK(p); return (ret); } /* * Marks the pending sleep of the current thread as interruptible and * makes an initial check for pending signals before putting a thread * to sleep. Enters and exits with the thread lock held. Thread lock * may have transitioned from the sleepq lock to a run lock. */ static int sleepq_catch_signals(const void *wchan, int pri) { struct thread *td; struct sleepqueue_chain *sc; struct sleepqueue *sq; int ret; sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); MPASS(wchan != NULL); td = curthread; ret = sleepq_check_ast_sc_locked(td, sc); THREAD_LOCK_ASSERT(td, MA_OWNED); mtx_assert(&sc->sc_lock, MA_OWNED); if (ret == 0) { /* * No pending signals and no suspension requests found. * Switch the thread off the cpu. */ sleepq_switch(wchan, pri); } else { /* * There were pending signals and this thread is still * on the sleep queue, remove it from the sleep queue. */ if (TD_ON_SLEEPQ(td)) { sq = sleepq_lookup(wchan); sleepq_remove_thread(sq, td); } MPASS(td->td_lock != &sc->sc_lock); mtx_unlock_spin(&sc->sc_lock); thread_unlock(td); } return (ret); } /* * Switches to another thread if we are still asleep on a sleep queue. * Returns with thread lock. */ static void sleepq_switch(const void *wchan, int pri) { struct sleepqueue_chain *sc; struct sleepqueue *sq; struct thread *td; bool rtc_changed; td = curthread; sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); THREAD_LOCK_ASSERT(td, MA_OWNED); /* * If we have a sleep queue, then we've already been woken up, so * just return. */ if (td->td_sleepqueue != NULL) { mtx_unlock_spin(&sc->sc_lock); thread_unlock(td); return; } /* * If TDF_TIMEOUT is set, then our sleep has been timed out * already but we are still on the sleep queue, so dequeue the * thread and return. * * Do the same if the real-time clock has been adjusted since this * thread calculated its timeout based on that clock. This handles * the following race: * - The Ts thread needs to sleep until an absolute real-clock time. * It copies the global rtc_generation into curthread->td_rtcgen, * reads the RTC, and calculates a sleep duration based on that time. * See umtxq_sleep() for an example. * - The Tc thread adjusts the RTC, bumps rtc_generation, and wakes * threads that are sleeping until an absolute real-clock time. * See tc_setclock() and the POSIX specification of clock_settime(). * - Ts reaches the code below. It holds the sleepqueue chain lock, * so Tc has finished waking, so this thread must test td_rtcgen. * (The declaration of td_rtcgen refers to this comment.) */ rtc_changed = td->td_rtcgen != 0 && td->td_rtcgen != rtc_generation; if ((td->td_flags & TDF_TIMEOUT) || rtc_changed) { if (rtc_changed) { td->td_rtcgen = 0; } MPASS(TD_ON_SLEEPQ(td)); sq = sleepq_lookup(wchan); sleepq_remove_thread(sq, td); mtx_unlock_spin(&sc->sc_lock); thread_unlock(td); return; } #ifdef SLEEPQUEUE_PROFILING if (prof_enabled) sleepq_profile(td->td_wmesg); #endif MPASS(td->td_sleepqueue == NULL); sched_sleep(td, pri); thread_lock_set(td, &sc->sc_lock); SDT_PROBE0(sched, , , sleep); TD_SET_SLEEPING(td); mi_switch(SW_VOL | SWT_SLEEPQ); KASSERT(TD_IS_RUNNING(td), ("running but not TDS_RUNNING")); CTR3(KTR_PROC, "sleepq resume: thread %p (pid %ld, %s)", (void *)td, (long)td->td_proc->p_pid, (void *)td->td_name); } /* * Check to see if we timed out. */ static inline int sleepq_check_timeout(void) { struct thread *td; int res; res = 0; td = curthread; if (td->td_sleeptimo != 0) { if (td->td_sleeptimo <= sbinuptime()) res = EWOULDBLOCK; td->td_sleeptimo = 0; } return (res); } /* * Check to see if we were awoken by a signal. */ static inline int sleepq_check_signals(void) { struct thread *td; td = curthread; KASSERT((td->td_flags & TDF_SINTR) == 0, ("thread %p still in interruptible sleep?", td)); return (td->td_intrval); } /* * Block the current thread until it is awakened from its sleep queue. */ void sleepq_wait(const void *wchan, int pri) { struct thread *td; td = curthread; MPASS(!(td->td_flags & TDF_SINTR)); thread_lock(td); sleepq_switch(wchan, pri); } /* * Block the current thread until it is awakened from its sleep queue * or it is interrupted by a signal. */ int sleepq_wait_sig(const void *wchan, int pri) { int rcatch; rcatch = sleepq_catch_signals(wchan, pri); if (rcatch) return (rcatch); return (sleepq_check_signals()); } /* * Block the current thread until it is awakened from its sleep queue * or it times out while waiting. */ int sleepq_timedwait(const void *wchan, int pri) { struct thread *td; td = curthread; MPASS(!(td->td_flags & TDF_SINTR)); thread_lock(td); sleepq_switch(wchan, pri); return (sleepq_check_timeout()); } /* * Block the current thread until it is awakened from its sleep queue, * it is interrupted by a signal, or it times out waiting to be awakened. */ int sleepq_timedwait_sig(const void *wchan, int pri) { int rcatch, rvalt, rvals; rcatch = sleepq_catch_signals(wchan, pri); /* We must always call check_timeout() to clear sleeptimo. */ rvalt = sleepq_check_timeout(); rvals = sleepq_check_signals(); if (rcatch) return (rcatch); if (rvals) return (rvals); return (rvalt); } /* * Returns the type of sleepqueue given a waitchannel. */ int sleepq_type(const void *wchan) { struct sleepqueue *sq; int type; MPASS(wchan != NULL); sq = sleepq_lookup(wchan); if (sq == NULL) return (-1); type = sq->sq_type; return (type); } /* * Removes a thread from a sleep queue and makes it runnable. * * Requires the sc chain locked on entry. If SRQ_HOLD is specified it will * be locked on return. Returns without the thread lock held. */ static void sleepq_resume_thread(struct sleepqueue *sq, struct thread *td, int pri, int srqflags) { struct sleepqueue_chain *sc; bool drop; MPASS(td != NULL); MPASS(sq->sq_wchan != NULL); MPASS(td->td_wchan == sq->sq_wchan); sc = SC_LOOKUP(sq->sq_wchan); mtx_assert(&sc->sc_lock, MA_OWNED); /* * Avoid recursing on the chain lock. If the locks don't match we * need to acquire the thread lock which setrunnable will drop for * us. In this case we need to drop the chain lock afterwards. * * There is no race that will make td_lock equal to sc_lock because * we hold sc_lock. */ drop = false; if (!TD_IS_SLEEPING(td)) { thread_lock(td); drop = true; } else thread_lock_block_wait(td); /* Remove thread from the sleepq. */ sleepq_remove_thread(sq, td); /* If we're done with the sleepqueue release it. */ if ((srqflags & SRQ_HOLD) == 0 && drop) mtx_unlock_spin(&sc->sc_lock); /* Adjust priority if requested. */ MPASS(pri == 0 || (pri >= PRI_MIN && pri <= PRI_MAX)); if (pri != 0 && td->td_priority > pri && PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) sched_prio(td, pri); /* * Note that thread td might not be sleeping if it is running * sleepq_catch_signals() on another CPU or is blocked on its * proc lock to check signals. There's no need to mark the * thread runnable in that case. */ if (TD_IS_SLEEPING(td)) { MPASS(!drop); TD_CLR_SLEEPING(td); setrunnable(td, srqflags); } else { MPASS(drop); thread_unlock(td); } } static void sleepq_remove_thread(struct sleepqueue *sq, struct thread *td) { struct sleepqueue_chain *sc __unused; MPASS(td != NULL); MPASS(sq->sq_wchan != NULL); MPASS(td->td_wchan == sq->sq_wchan); MPASS(td->td_sqqueue < NR_SLEEPQS && td->td_sqqueue >= 0); THREAD_LOCK_ASSERT(td, MA_OWNED); sc = SC_LOOKUP(sq->sq_wchan); mtx_assert(&sc->sc_lock, MA_OWNED); SDT_PROBE2(sched, , , wakeup, td, td->td_proc); /* Remove the thread from the queue. */ sq->sq_blockedcnt[td->td_sqqueue]--; TAILQ_REMOVE(&sq->sq_blocked[td->td_sqqueue], td, td_slpq); /* * Get a sleep queue for this thread. If this is the last waiter, * use the queue itself and take it out of the chain, otherwise, * remove a queue from the free list. */ if (LIST_EMPTY(&sq->sq_free)) { td->td_sleepqueue = sq; #ifdef INVARIANTS sq->sq_wchan = NULL; #endif #ifdef SLEEPQUEUE_PROFILING sc->sc_depth--; #endif } else td->td_sleepqueue = LIST_FIRST(&sq->sq_free); LIST_REMOVE(td->td_sleepqueue, sq_hash); if ((td->td_flags & TDF_TIMEOUT) == 0 && td->td_sleeptimo != 0 && td->td_lock == &sc->sc_lock) { /* * We ignore the situation where timeout subsystem was * unable to stop our callout. The struct thread is * type-stable, the callout will use the correct * memory when running. The checks of the * td_sleeptimo value in this function and in * sleepq_timeout() ensure that the thread does not * get spurious wakeups, even if the callout was reset * or thread reused. * * We also cannot safely stop the callout if a scheduler * lock is held since softclock_thread() forces a lock * order of callout lock -> scheduler lock. The thread * lock will be a scheduler lock only if the thread is * preparing to go to sleep, so this is hopefully a rare * scenario. */ callout_stop(&td->td_slpcallout); } td->td_wmesg = NULL; td->td_wchan = NULL; td->td_flags &= ~(TDF_SINTR | TDF_TIMEOUT); CTR3(KTR_PROC, "sleepq_wakeup: thread %p (pid %ld, %s)", (void *)td, (long)td->td_proc->p_pid, td->td_name); } void sleepq_remove_nested(struct thread *td) { struct sleepqueue_chain *sc; struct sleepqueue *sq; const void *wchan; MPASS(TD_ON_SLEEPQ(td)); wchan = td->td_wchan; sc = SC_LOOKUP(wchan); mtx_lock_spin(&sc->sc_lock); sq = sleepq_lookup(wchan); MPASS(sq != NULL); thread_lock(td); sleepq_remove_thread(sq, td); mtx_unlock_spin(&sc->sc_lock); /* Returns with the thread lock owned. */ } #ifdef INVARIANTS /* * UMA zone item deallocator. */ static void sleepq_dtor(void *mem, int size, void *arg) { struct sleepqueue *sq; int i; sq = mem; for (i = 0; i < NR_SLEEPQS; i++) { MPASS(TAILQ_EMPTY(&sq->sq_blocked[i])); MPASS(sq->sq_blockedcnt[i] == 0); } } #endif /* * UMA zone item initializer. */ static int sleepq_init(void *mem, int size, int flags) { struct sleepqueue *sq; int i; bzero(mem, size); sq = mem; for (i = 0; i < NR_SLEEPQS; i++) { TAILQ_INIT(&sq->sq_blocked[i]); sq->sq_blockedcnt[i] = 0; } LIST_INIT(&sq->sq_free); return (0); } /* * Find thread sleeping on a wait channel and resume it. */ -int +void sleepq_signal(const void *wchan, int flags, int pri, int queue) { struct sleepqueue_chain *sc; struct sleepqueue *sq; struct threadqueue *head; struct thread *td, *besttd; CTR2(KTR_PROC, "sleepq_signal(%p, %d)", wchan, flags); KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__)); MPASS((queue >= 0) && (queue < NR_SLEEPQS)); sq = sleepq_lookup(wchan); if (sq == NULL) { if (flags & SLEEPQ_DROP) sleepq_release(wchan); - return (0); + return; } KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE), ("%s: mismatch between sleep/wakeup and cv_*", __func__)); head = &sq->sq_blocked[queue]; if (flags & SLEEPQ_UNFAIR) { /* * Find the most recently sleeping thread, but try to * skip threads still in process of context switch to * avoid spinning on the thread lock. */ sc = SC_LOOKUP(wchan); besttd = TAILQ_LAST_FAST(head, thread, td_slpq); while (besttd->td_lock != &sc->sc_lock) { td = TAILQ_PREV_FAST(besttd, head, thread, td_slpq); if (td == NULL) break; besttd = td; } } else { /* * Find the highest priority thread on the queue. If there * is a tie, use the thread that first appears in the queue * as it has been sleeping the longest since threads are * always added to the tail of sleep queues. */ besttd = td = TAILQ_FIRST(head); while ((td = TAILQ_NEXT(td, td_slpq)) != NULL) { if (td->td_priority < besttd->td_priority) besttd = td; } } MPASS(besttd != NULL); sleepq_resume_thread(sq, besttd, pri, (flags & SLEEPQ_DROP) ? 0 : SRQ_HOLD); - return (0); } static bool match_any(struct thread *td __unused) { return (true); } /* * Resume all threads sleeping on a specified wait channel. */ -int +void sleepq_broadcast(const void *wchan, int flags, int pri, int queue) { struct sleepqueue *sq; CTR2(KTR_PROC, "sleepq_broadcast(%p, %d)", wchan, flags); KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__)); MPASS((queue >= 0) && (queue < NR_SLEEPQS)); sq = sleepq_lookup(wchan); - if (sq == NULL) - return (0); - KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE), - ("%s: mismatch between sleep/wakeup and cv_*", __func__)); + if (sq != NULL) { + KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE), + ("%s: mismatch between sleep/wakeup and cv_*", __func__)); - return (sleepq_remove_matching(sq, queue, match_any, pri)); + sleepq_remove_matching(sq, queue, match_any, pri); + } } /* * Resume threads on the sleep queue that match the given predicate. */ -int +void sleepq_remove_matching(struct sleepqueue *sq, int queue, bool (*matches)(struct thread *), int pri) { struct thread *td, *tdn; /* * The last thread will be given ownership of sq and may * re-enqueue itself before sleepq_resume_thread() returns, * so we must cache the "next" queue item at the beginning * of the final iteration. */ TAILQ_FOREACH_SAFE(td, &sq->sq_blocked[queue], td_slpq, tdn) { if (matches(td)) sleepq_resume_thread(sq, td, pri, SRQ_HOLD); } - - return (0); } /* * Time sleeping threads out. When the timeout expires, the thread is * removed from the sleep queue and made runnable if it is still asleep. */ static void sleepq_timeout(void *arg) { struct sleepqueue_chain *sc __unused; struct sleepqueue *sq; struct thread *td; const void *wchan; td = arg; CTR3(KTR_PROC, "sleepq_timeout: thread %p (pid %ld, %s)", (void *)td, (long)td->td_proc->p_pid, (void *)td->td_name); thread_lock(td); if (td->td_sleeptimo == 0 || td->td_sleeptimo > td->td_slpcallout.c_time) { /* * The thread does not want a timeout (yet). */ } else if (TD_IS_SLEEPING(td) && TD_ON_SLEEPQ(td)) { /* * See if the thread is asleep and get the wait * channel if it is. */ wchan = td->td_wchan; sc = SC_LOOKUP(wchan); THREAD_LOCKPTR_ASSERT(td, &sc->sc_lock); sq = sleepq_lookup(wchan); MPASS(sq != NULL); td->td_flags |= TDF_TIMEOUT; sleepq_resume_thread(sq, td, 0, 0); return; } else if (TD_ON_SLEEPQ(td)) { /* * If the thread is on the SLEEPQ but isn't sleeping * yet, it can either be on another CPU in between * sleepq_add() and one of the sleepq_*wait*() * routines or it can be in sleepq_catch_signals(). */ td->td_flags |= TDF_TIMEOUT; } thread_unlock(td); } /* * Resumes a specific thread from the sleep queue associated with a specific * wait channel if it is on that queue. */ void sleepq_remove(struct thread *td, const void *wchan) { struct sleepqueue_chain *sc; struct sleepqueue *sq; /* * Look up the sleep queue for this wait channel, then re-check * that the thread is asleep on that channel, if it is not, then * bail. */ MPASS(wchan != NULL); sc = SC_LOOKUP(wchan); mtx_lock_spin(&sc->sc_lock); /* * We can not lock the thread here as it may be sleeping on a * different sleepq. However, holding the sleepq lock for this * wchan can guarantee that we do not miss a wakeup for this * channel. The asserts below will catch any false positives. */ if (!TD_ON_SLEEPQ(td) || td->td_wchan != wchan) { mtx_unlock_spin(&sc->sc_lock); return; } /* Thread is asleep on sleep queue sq, so wake it up. */ sq = sleepq_lookup(wchan); MPASS(sq != NULL); MPASS(td->td_wchan == wchan); sleepq_resume_thread(sq, td, 0, 0); } /* * Abort a thread as if an interrupt had occurred. Only abort * interruptible waits (unfortunately it isn't safe to abort others). * * Requires thread lock on entry, releases on return. */ -int +void sleepq_abort(struct thread *td, int intrval) { struct sleepqueue *sq; const void *wchan; THREAD_LOCK_ASSERT(td, MA_OWNED); MPASS(TD_ON_SLEEPQ(td)); MPASS(td->td_flags & TDF_SINTR); MPASS((intrval == 0 && (td->td_flags & TDF_SIGWAIT) != 0) || intrval == EINTR || intrval == ERESTART); /* * If the TDF_TIMEOUT flag is set, just leave. A * timeout is scheduled anyhow. */ if (td->td_flags & TDF_TIMEOUT) { thread_unlock(td); - return (0); + return; } CTR3(KTR_PROC, "sleepq_abort: thread %p (pid %ld, %s)", (void *)td, (long)td->td_proc->p_pid, (void *)td->td_name); td->td_intrval = intrval; /* * If the thread has not slept yet it will find the signal in * sleepq_catch_signals() and call sleepq_resume_thread. Otherwise * we have to do it here. */ if (!TD_IS_SLEEPING(td)) { thread_unlock(td); - return (0); + return; } wchan = td->td_wchan; MPASS(wchan != NULL); sq = sleepq_lookup(wchan); MPASS(sq != NULL); /* Thread is asleep on sleep queue sq, so wake it up. */ sleepq_resume_thread(sq, td, 0, 0); - return (0); } void sleepq_chains_remove_matching(bool (*matches)(struct thread *)) { struct sleepqueue_chain *sc; struct sleepqueue *sq, *sq1; int i; for (sc = &sleepq_chains[0]; sc < sleepq_chains + SC_TABLESIZE; ++sc) { if (LIST_EMPTY(&sc->sc_queues)) { continue; } mtx_lock_spin(&sc->sc_lock); LIST_FOREACH_SAFE(sq, &sc->sc_queues, sq_hash, sq1) { for (i = 0; i < NR_SLEEPQS; ++i) sleepq_remove_matching(sq, i, matches, 0); } mtx_unlock_spin(&sc->sc_lock); } } /* * Prints the stacks of all threads presently sleeping on wchan/queue to * the sbuf sb. Sets count_stacks_printed to the number of stacks actually * printed. Typically, this will equal the number of threads sleeping on the * queue, but may be less if sb overflowed before all stacks were printed. */ #ifdef STACK int sleepq_sbuf_print_stacks(struct sbuf *sb, const void *wchan, int queue, int *count_stacks_printed) { struct thread *td, *td_next; struct sleepqueue *sq; struct stack **st; struct sbuf **td_infos; int i, stack_idx, error, stacks_to_allocate; bool finished; error = 0; finished = false; KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__)); MPASS((queue >= 0) && (queue < NR_SLEEPQS)); stacks_to_allocate = 10; for (i = 0; i < 3 && !finished ; i++) { /* We cannot malloc while holding the queue's spinlock, so * we do our mallocs now, and hope it is enough. If it * isn't, we will free these, drop the lock, malloc more, * and try again, up to a point. After that point we will * give up and report ENOMEM. We also cannot write to sb * during this time since the client may have set the * SBUF_AUTOEXTEND flag on their sbuf, which could cause a * malloc as we print to it. So we defer actually printing * to sb until after we drop the spinlock. */ /* Where we will store the stacks. */ st = malloc(sizeof(struct stack *) * stacks_to_allocate, M_TEMP, M_WAITOK); for (stack_idx = 0; stack_idx < stacks_to_allocate; stack_idx++) st[stack_idx] = stack_create(M_WAITOK); /* Where we will store the td name, tid, etc. */ td_infos = malloc(sizeof(struct sbuf *) * stacks_to_allocate, M_TEMP, M_WAITOK); for (stack_idx = 0; stack_idx < stacks_to_allocate; stack_idx++) td_infos[stack_idx] = sbuf_new(NULL, NULL, MAXCOMLEN + sizeof(struct thread *) * 2 + 40, SBUF_FIXEDLEN); sleepq_lock(wchan); sq = sleepq_lookup(wchan); if (sq == NULL) { /* This sleepq does not exist; exit and return ENOENT. */ error = ENOENT; finished = true; sleepq_release(wchan); goto loop_end; } stack_idx = 0; /* Save thread info */ TAILQ_FOREACH_SAFE(td, &sq->sq_blocked[queue], td_slpq, td_next) { if (stack_idx >= stacks_to_allocate) goto loop_end; /* Note the td_lock is equal to the sleepq_lock here. */ (void)stack_save_td(st[stack_idx], td); sbuf_printf(td_infos[stack_idx], "%d: %s %p", td->td_tid, td->td_name, td); ++stack_idx; } finished = true; sleepq_release(wchan); /* Print the stacks */ for (i = 0; i < stack_idx; i++) { sbuf_finish(td_infos[i]); sbuf_printf(sb, "--- thread %s: ---\n", sbuf_data(td_infos[i])); stack_sbuf_print(sb, st[i]); sbuf_putc(sb, '\n'); error = sbuf_error(sb); if (error == 0) *count_stacks_printed = stack_idx; } loop_end: if (!finished) sleepq_release(wchan); for (stack_idx = 0; stack_idx < stacks_to_allocate; stack_idx++) stack_destroy(st[stack_idx]); for (stack_idx = 0; stack_idx < stacks_to_allocate; stack_idx++) sbuf_delete(td_infos[stack_idx]); free(st, M_TEMP); free(td_infos, M_TEMP); stacks_to_allocate *= 10; } if (!finished && error == 0) error = ENOMEM; return (error); } #endif #ifdef SLEEPQUEUE_PROFILING #define SLEEPQ_PROF_LOCATIONS 1024 #define SLEEPQ_SBUFSIZE 512 struct sleepq_prof { LIST_ENTRY(sleepq_prof) sp_link; const char *sp_wmesg; long sp_count; }; LIST_HEAD(sqphead, sleepq_prof); struct sqphead sleepq_prof_free; struct sqphead sleepq_hash[SC_TABLESIZE]; static struct sleepq_prof sleepq_profent[SLEEPQ_PROF_LOCATIONS]; static struct mtx sleepq_prof_lock; MTX_SYSINIT(sleepq_prof_lock, &sleepq_prof_lock, "sleepq_prof", MTX_SPIN); static void sleepq_profile(const char *wmesg) { struct sleepq_prof *sp; mtx_lock_spin(&sleepq_prof_lock); if (prof_enabled == 0) goto unlock; LIST_FOREACH(sp, &sleepq_hash[SC_HASH(wmesg)], sp_link) if (sp->sp_wmesg == wmesg) goto done; sp = LIST_FIRST(&sleepq_prof_free); if (sp == NULL) goto unlock; sp->sp_wmesg = wmesg; LIST_REMOVE(sp, sp_link); LIST_INSERT_HEAD(&sleepq_hash[SC_HASH(wmesg)], sp, sp_link); done: sp->sp_count++; unlock: mtx_unlock_spin(&sleepq_prof_lock); return; } static void sleepq_prof_reset(void) { struct sleepq_prof *sp; int enabled; int i; mtx_lock_spin(&sleepq_prof_lock); enabled = prof_enabled; prof_enabled = 0; for (i = 0; i < SC_TABLESIZE; i++) LIST_INIT(&sleepq_hash[i]); LIST_INIT(&sleepq_prof_free); for (i = 0; i < SLEEPQ_PROF_LOCATIONS; i++) { sp = &sleepq_profent[i]; sp->sp_wmesg = NULL; sp->sp_count = 0; LIST_INSERT_HEAD(&sleepq_prof_free, sp, sp_link); } prof_enabled = enabled; mtx_unlock_spin(&sleepq_prof_lock); } static int enable_sleepq_prof(SYSCTL_HANDLER_ARGS) { int error, v; v = prof_enabled; error = sysctl_handle_int(oidp, &v, v, req); if (error) return (error); if (req->newptr == NULL) return (error); if (v == prof_enabled) return (0); if (v == 1) sleepq_prof_reset(); mtx_lock_spin(&sleepq_prof_lock); prof_enabled = !!v; mtx_unlock_spin(&sleepq_prof_lock); return (0); } static int reset_sleepq_prof_stats(SYSCTL_HANDLER_ARGS) { int error, v; v = 0; error = sysctl_handle_int(oidp, &v, 0, req); if (error) return (error); if (req->newptr == NULL) return (error); if (v == 0) return (0); sleepq_prof_reset(); return (0); } static int dump_sleepq_prof_stats(SYSCTL_HANDLER_ARGS) { struct sleepq_prof *sp; struct sbuf *sb; int enabled; int error; int i; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sb = sbuf_new_for_sysctl(NULL, NULL, SLEEPQ_SBUFSIZE, req); sbuf_cat(sb, "\nwmesg\tcount\n"); enabled = prof_enabled; mtx_lock_spin(&sleepq_prof_lock); prof_enabled = 0; mtx_unlock_spin(&sleepq_prof_lock); for (i = 0; i < SC_TABLESIZE; i++) { LIST_FOREACH(sp, &sleepq_hash[i], sp_link) { sbuf_printf(sb, "%s\t%ld\n", sp->sp_wmesg, sp->sp_count); } } mtx_lock_spin(&sleepq_prof_lock); prof_enabled = enabled; mtx_unlock_spin(&sleepq_prof_lock); error = sbuf_finish(sb); sbuf_delete(sb); return (error); } SYSCTL_PROC(_debug_sleepq, OID_AUTO, stats, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0, dump_sleepq_prof_stats, "A", "Sleepqueue profiling statistics"); SYSCTL_PROC(_debug_sleepq, OID_AUTO, reset, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, reset_sleepq_prof_stats, "I", "Reset sleepqueue profiling statistics"); SYSCTL_PROC(_debug_sleepq, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, enable_sleepq_prof, "I", "Enable sleepqueue profiling"); #endif #ifdef DDB DB_SHOW_COMMAND(sleepq, db_show_sleepqueue) { struct sleepqueue_chain *sc; struct sleepqueue *sq; #ifdef INVARIANTS struct lock_object *lock; #endif struct thread *td; void *wchan; int i; if (!have_addr) return; /* * First, see if there is an active sleep queue for the wait channel * indicated by the address. */ wchan = (void *)addr; sc = SC_LOOKUP(wchan); LIST_FOREACH(sq, &sc->sc_queues, sq_hash) if (sq->sq_wchan == wchan) goto found; /* * Second, see if there is an active sleep queue at the address * indicated. */ for (i = 0; i < SC_TABLESIZE; i++) LIST_FOREACH(sq, &sleepq_chains[i].sc_queues, sq_hash) { if (sq == (struct sleepqueue *)addr) goto found; } db_printf("Unable to locate a sleep queue via %p\n", (void *)addr); return; found: db_printf("Wait channel: %p\n", sq->sq_wchan); db_printf("Queue type: %d\n", sq->sq_type); #ifdef INVARIANTS if (sq->sq_lock) { lock = sq->sq_lock; db_printf("Associated Interlock: %p - (%s) %s\n", lock, LOCK_CLASS(lock)->lc_name, lock->lo_name); } #endif db_printf("Blocked threads:\n"); for (i = 0; i < NR_SLEEPQS; i++) { db_printf("\nQueue[%d]:\n", i); if (TAILQ_EMPTY(&sq->sq_blocked[i])) db_printf("\tempty\n"); else TAILQ_FOREACH(td, &sq->sq_blocked[i], td_slpq) { db_printf("\t%p (tid %d, pid %d, \"%s\")\n", td, td->td_tid, td->td_proc->p_pid, td->td_name); } db_printf("(expected: %u)\n", sq->sq_blockedcnt[i]); } } /* Alias 'show sleepqueue' to 'show sleepq'. */ DB_SHOW_ALIAS(sleepqueue, db_show_sleepqueue); #endif diff --git a/sys/sys/proc.h b/sys/sys/proc.h index 5e3f0e2e117d..57c910d8fce0 100644 --- a/sys/sys/proc.h +++ b/sys/sys/proc.h @@ -1,1341 +1,1340 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #ifndef _SYS_PROC_H_ #define _SYS_PROC_H_ #include /* For struct callout. */ #include /* For struct klist. */ #ifdef _KERNEL #include #endif #include #ifndef _KERNEL #include #endif #include #include #include #include #include #include #include /* XXX. */ #include #include #include #include #include #ifndef _KERNEL #include /* For structs itimerval, timeval. */ #else #include #include #endif #include #include #include #include #include /* Machine-dependent proc substruct. */ #ifdef _KERNEL #include #endif /* * One structure allocated per session. * * List of locks * (m) locked by s_mtx mtx * (e) locked by proctree_lock sx * (c) const until freeing */ struct session { u_int s_count; /* Ref cnt; pgrps in session - atomic. */ struct proc *s_leader; /* (m + e) Session leader. */ struct vnode *s_ttyvp; /* (m) Vnode of controlling tty. */ struct cdev_priv *s_ttydp; /* (m) Device of controlling tty. */ struct tty *s_ttyp; /* (e) Controlling tty. */ pid_t s_sid; /* (c) Session ID. */ /* (m) Setlogin() name: */ char s_login[roundup(MAXLOGNAME, sizeof(long))]; struct mtx s_mtx; /* Mutex to protect members. */ }; /* * One structure allocated per process group. * * List of locks * (m) locked by pg_mtx mtx * (e) locked by proctree_lock sx * (c) const until freeing */ struct pgrp { LIST_ENTRY(pgrp) pg_hash; /* (e) Hash chain. */ LIST_HEAD(, proc) pg_members; /* (m + e) Pointer to pgrp members. */ struct session *pg_session; /* (c) Pointer to session. */ struct sigiolst pg_sigiolst; /* (m) List of sigio sources. */ pid_t pg_id; /* (c) Process group id. */ struct mtx pg_mtx; /* Mutex to protect members */ int pg_flags; /* (m) PGRP_ flags */ struct sx pg_killsx; /* Mutual exclusion between group member * fork() and killpg() */ }; #define PGRP_ORPHANED 0x00000001 /* Group is orphaned */ /* * pargs, used to hold a copy of the command line, if it had a sane length. */ struct pargs { u_int ar_ref; /* Reference count. */ u_int ar_length; /* Length. */ u_char ar_args[1]; /* Arguments. */ }; /*- * Description of a process. * * This structure contains the information needed to manage a thread of * control, known in UN*X as a process; it has references to substructures * containing descriptions of things that the process uses, but may share * with related processes. The process structure and the substructures * are always addressable except for those marked "(CPU)" below, * which might be addressable only on a processor on which the process * is running. * * Below is a key of locks used to protect each member of struct proc. The * lock is indicated by a reference to a specific character in parens in the * associated comment. * * - not yet protected * a - only touched by curproc or parent during fork/wait * b - created at fork, never changes * (exception aiods switch vmspaces, but they are also * marked 'P_SYSTEM' so hopefully it will be left alone) * c - locked by proc mtx * d - locked by allproc_lock lock * e - locked by proctree_lock lock * f - session mtx * g - process group mtx * h - callout_lock mtx * i - by curproc or the master session mtx * j - locked by proc slock * k - only accessed by curthread * k*- only accessed by curthread and from an interrupt * kx- only accessed by curthread and by debugger * l - the attaching proc or attaching proc parent * n - not locked, lazy * o - ktrace lock * q - td_contested lock * r - p_peers lock * s - see sleepq_switch(), sleeping_on_old_rtc(), and sleep(9) * t - thread lock * u - process stat lock * w - process timer lock * x - created at fork, only changes during single threading in exec * y - created at first aio, doesn't change until exit or exec at which * point we are single-threaded and only curthread changes it * * If the locking key specifies two identifiers (for example, p_pptr) then * either lock is sufficient for read access, but both locks must be held * for write access. */ struct cpuset; struct filecaps; struct filemon; struct kaioinfo; struct kaudit_record; struct kcov_info; struct kdtrace_proc; struct kdtrace_thread; struct kmsan_td; struct kq_timer_cb_data; struct mqueue_notifier; struct p_sched; struct proc; struct procdesc; struct racct; struct sbuf; struct sleepqueue; struct socket; struct td_sched; struct thread; struct trapframe; struct turnstile; struct vm_map; struct vm_map_entry; struct epoch_tracker; struct syscall_args { u_int code; u_int original_code; struct sysent *callp; register_t args[8]; }; /* * XXX: Does this belong in resource.h or resourcevar.h instead? * Resource usage extension. The times in rusage structs in the kernel are * never up to date. The actual times are kept as runtimes and tick counts * (with control info in the "previous" times), and are converted when * userland asks for rusage info. Backwards compatibility prevents putting * this directly in the user-visible rusage struct. * * Locking for p_rux: (cu) means (u) for p_rux and (c) for p_crux. * Locking for td_rux: (t) for all fields. */ struct rusage_ext { uint64_t rux_runtime; /* (cu) Real time. */ uint64_t rux_uticks; /* (cu) Statclock hits in user mode. */ uint64_t rux_sticks; /* (cu) Statclock hits in sys mode. */ uint64_t rux_iticks; /* (cu) Statclock hits in intr mode. */ uint64_t rux_uu; /* (c) Previous user time in usec. */ uint64_t rux_su; /* (c) Previous sys time in usec. */ uint64_t rux_tu; /* (c) Previous total time in usec. */ }; /* * Kernel runnable context (thread). * This is what is put to sleep and reactivated. * Thread context. Processes may have multiple threads. */ struct thread { struct mtx *volatile td_lock; /* replaces sched lock */ struct proc *td_proc; /* (*) Associated process. */ TAILQ_ENTRY(thread) td_plist; /* (*) All threads in this proc. */ TAILQ_ENTRY(thread) td_runq; /* (t) Run queue. */ union { TAILQ_ENTRY(thread) td_slpq; /* (t) Sleep queue. */ struct thread *td_zombie; /* Zombie list linkage */ }; TAILQ_ENTRY(thread) td_lockq; /* (t) Lock queue. */ LIST_ENTRY(thread) td_hash; /* (d) Hash chain. */ struct cpuset *td_cpuset; /* (t) CPU affinity mask. */ struct domainset_ref td_domain; /* (a) NUMA policy */ struct seltd *td_sel; /* Select queue/channel. */ struct sleepqueue *td_sleepqueue; /* (k) Associated sleep queue. */ struct turnstile *td_turnstile; /* (k) Associated turnstile. */ struct rl_q_entry *td_rlqe; /* (k) Associated range lock entry. */ struct umtx_q *td_umtxq; /* (c?) Link for when we're blocked. */ lwpid_t td_tid; /* (b) Thread ID. */ sigqueue_t td_sigqueue; /* (c) Sigs arrived, not delivered. */ #define td_siglist td_sigqueue.sq_signals u_char td_lend_user_pri; /* (t) Lend user pri. */ u_char td_allocdomain; /* (b) NUMA domain backing this struct thread. */ u_char td_base_ithread_pri; /* (t) Base ithread pri */ struct kmsan_td *td_kmsan; /* (k) KMSAN state */ /* Cleared during fork1(), thread_create(), or kthread_add(). */ #define td_startzero td_flags int td_flags; /* (t) TDF_* flags. */ int td_ast; /* (t) TDA_* indicators */ int td_inhibitors; /* (t) Why can not run. */ int td_pflags; /* (k) Private thread (TDP_*) flags. */ int td_pflags2; /* (k) Private thread (TDP2_*) flags. */ int td_dupfd; /* (k) Ret value from fdopen. XXX */ int td_sqqueue; /* (t) Sleepqueue queue blocked on. */ const void *td_wchan; /* (t) Sleep address. */ const char *td_wmesg; /* (t) Reason for sleep. */ volatile u_char td_owepreempt; /* (k*) Preempt on last critical_exit */ u_char td_tsqueue; /* (t) Turnstile queue blocked on. */ u_char _td_pad0[2]; /* Available. */ int td_locks; /* (k) Debug: count of non-spin locks */ int td_rw_rlocks; /* (k) Count of rwlock read locks. */ int td_sx_slocks; /* (k) Count of sx shared locks. */ int td_lk_slocks; /* (k) Count of lockmgr shared locks. */ struct lock_object *td_wantedlock; /* (k) Lock we are contending on */ struct turnstile *td_blocked; /* (t) Lock thread is blocked on. */ const char *td_lockname; /* (t) Name of lock blocked on. */ LIST_HEAD(, turnstile) td_contested; /* (q) Contested locks. */ struct lock_list_entry *td_sleeplocks; /* (k) Held sleep locks. */ int td_intr_nesting_level; /* (k) Interrupt recursion. */ int td_pinned; /* (k) Temporary cpu pin count. */ struct ucred *td_realucred; /* (k) Reference to credentials. */ struct ucred *td_ucred; /* (k) Used credentials, temporarily switchable. */ struct plimit *td_limit; /* (k) Resource limits. */ int td_slptick; /* (t) Time at sleep. */ int td_blktick; /* (t) Time spent blocked. */ int td_swvoltick; /* (t) Time at last SW_VOL switch. */ int td_swinvoltick; /* (t) Time at last SW_INVOL switch. */ u_int td_cow; /* (*) Number of copy-on-write faults */ struct rusage td_ru; /* (t) rusage information. */ struct rusage_ext td_rux; /* (t) Internal rusage information. */ uint64_t td_incruntime; /* (t) Cpu ticks to transfer to proc. */ uint64_t td_runtime; /* (t) How many cpu ticks we've run. */ u_int td_pticks; /* (t) Statclock hits for profiling */ u_int td_sticks; /* (t) Statclock hits in system mode. */ u_int td_iticks; /* (t) Statclock hits in intr mode. */ u_int td_uticks; /* (t) Statclock hits in user mode. */ int td_intrval; /* (t) Return value for sleepq. */ sigset_t td_oldsigmask; /* (k) Saved mask from pre sigpause. */ volatile u_int td_generation; /* (k) For detection of preemption */ stack_t td_sigstk; /* (k) Stack ptr and on-stack flag. */ int td_xsig; /* (c) Signal for ptrace */ u_long td_profil_addr; /* (k) Temporary addr until AST. */ u_int td_profil_ticks; /* (k) Temporary ticks until AST. */ char td_name[MAXCOMLEN + 1]; /* (*) Thread name. */ struct file *td_fpop; /* (k) file referencing cdev under op */ int td_dbgflags; /* (c) Userland debugger flags */ siginfo_t td_si; /* (c) For debugger or core file */ int td_ng_outbound; /* (k) Thread entered ng from above. */ struct osd td_osd; /* (k) Object specific data. */ struct vm_map_entry *td_map_def_user; /* (k) Deferred entries. */ pid_t td_dbg_forked; /* (c) Child pid for debugger. */ u_int td_no_sleeping; /* (k) Sleeping disabled count. */ struct vnode *td_vp_reserved;/* (k) Preallocated vnode. */ void *td_su; /* (k) FFS SU private */ sbintime_t td_sleeptimo; /* (t) Sleep timeout. */ int td_rtcgen; /* (s) rtc_generation of abs. sleep */ int td_errno; /* (k) Error from last syscall. */ size_t td_vslock_sz; /* (k) amount of vslock-ed space */ struct kcov_info *td_kcov_info; /* (*) Kernel code coverage data */ long td_ucredref; /* (k) references on td_realucred */ #define td_endzero td_sigmask /* Copied during fork1(), thread_create(), or kthread_add(). */ #define td_startcopy td_endzero sigset_t td_sigmask; /* (c) Current signal mask. */ u_char td_rqindex; /* (t) Run queue index. */ u_char td_base_pri; /* (t) Thread base kernel priority. */ u_char td_priority; /* (t) Thread active priority. */ u_char td_pri_class; /* (t) Scheduling class. */ u_char td_user_pri; /* (t) User pri from estcpu and nice. */ u_char td_base_user_pri; /* (t) Base user pri */ uintptr_t td_rb_list; /* (k) Robust list head. */ uintptr_t td_rbp_list; /* (k) Robust priv list head. */ uintptr_t td_rb_inact; /* (k) Current in-action mutex loc. */ struct syscall_args td_sa; /* (kx) Syscall parameters. Copied on fork for child tracing. */ void *td_sigblock_ptr; /* (k) uptr for fast sigblock. */ uint32_t td_sigblock_val; /* (k) fast sigblock value read at td_sigblock_ptr on kern entry */ #define td_endcopy td_pcb /* * Fields that must be manually set in fork1(), thread_create(), kthread_add(), * or already have been set in the allocator, constructor, etc. */ struct pcb *td_pcb; /* (k) Kernel VA of pcb and kstack. */ enum td_states { TDS_INACTIVE = 0x0, TDS_INHIBITED, TDS_CAN_RUN, TDS_RUNQ, TDS_RUNNING } td_state; /* (t) thread state */ /* Note: td_state must be accessed using TD_{GET,SET}_STATE(). */ union { syscallarg_t tdu_retval[2]; off_t tdu_off; } td_uretoff; /* (k) Syscall aux returns. */ #define td_retval td_uretoff.tdu_retval u_int td_cowgen; /* (k) Generation of COW pointers. */ /* LP64 hole */ struct callout td_slpcallout; /* (h) Callout for sleep. */ struct trapframe *td_frame; /* (k) */ vm_offset_t td_kstack; /* (a) Kernel VA of kstack. */ u_short td_kstack_pages; /* (a) Size of the kstack. */ u_short td_kstack_domain; /* (a) Domain backing kstack KVA. */ volatile u_int td_critnest; /* (k*) Critical section nest level. */ struct mdthread td_md; /* (k) Any machine-dependent fields. */ struct kaudit_record *td_ar; /* (k) Active audit record, if any. */ struct lpohead td_lprof[2]; /* (a) lock profiling objects. */ struct kdtrace_thread *td_dtrace; /* (*) DTrace-specific data. */ struct vnet *td_vnet; /* (k) Effective vnet. */ const char *td_vnet_lpush; /* (k) Debugging vnet push / pop. */ struct trapframe *td_intr_frame;/* (k) Frame of the current irq */ struct proc *td_rfppwait_p; /* (k) The vforked child */ struct vm_page **td_ma; /* (k) uio pages held */ int td_ma_cnt; /* (k) size of *td_ma */ /* LP64 hole */ void *td_emuldata; /* Emulator state data */ int td_lastcpu; /* (t) Last cpu we were on. */ int td_oncpu; /* (t) Which cpu we are on. */ void *td_lkpi_task; /* LinuxKPI task struct pointer */ int td_pmcpend; void *td_remotereq; /* (c) dbg remote request. */ off_t td_ktr_io_lim; /* (k) limit for ktrace file size */ #ifdef EPOCH_TRACE SLIST_HEAD(, epoch_tracker) td_epochs; #endif }; struct thread0_storage { struct thread t0st_thread; uint64_t t0st_sched[10]; }; struct mtx *thread_lock_block(struct thread *); void thread_lock_block_wait(struct thread *); void thread_lock_set(struct thread *, struct mtx *); void thread_lock_unblock(struct thread *, struct mtx *); #define THREAD_LOCK_ASSERT(td, type) \ mtx_assert((td)->td_lock, (type)) #define THREAD_LOCK_BLOCKED_ASSERT(td, type) \ do { \ struct mtx *__m = (td)->td_lock; \ if (__m != &blocked_lock) \ mtx_assert(__m, (type)); \ } while (0) #ifdef INVARIANTS #define THREAD_LOCKPTR_ASSERT(td, lock) \ do { \ struct mtx *__m; \ __m = (td)->td_lock; \ KASSERT(__m == (lock), \ ("Thread %p lock %p does not match %p", td, __m, (lock))); \ } while (0) #define THREAD_LOCKPTR_BLOCKED_ASSERT(td, lock) \ do { \ struct mtx *__m; \ __m = (td)->td_lock; \ KASSERT(__m == (lock) || __m == &blocked_lock, \ ("Thread %p lock %p does not match %p", td, __m, (lock))); \ } while (0) #define TD_LOCKS_INC(td) ((td)->td_locks++) #define TD_LOCKS_DEC(td) do { \ KASSERT(SCHEDULER_STOPPED() || (td)->td_locks > 0, \ ("Thread %p owns no locks", (td))); \ (td)->td_locks--; \ } while (0) #else #define THREAD_LOCKPTR_ASSERT(td, lock) #define THREAD_LOCKPTR_BLOCKED_ASSERT(td, lock) #define TD_LOCKS_INC(td) #define TD_LOCKS_DEC(td) #endif /* * Flags kept in td_flags: * To change these you MUST have the scheduler lock. */ #define TDF_BORROWING 0x00000001 /* Thread is borrowing pri from another. */ #define TDF_INPANIC 0x00000002 /* Caused a panic, let it drive crashdump. */ #define TDF_INMEM 0x00000004 /* Thread's stack is in memory. */ #define TDF_SINTR 0x00000008 /* Sleep is interruptible. */ #define TDF_TIMEOUT 0x00000010 /* Timing out during sleep. */ #define TDF_IDLETD 0x00000020 /* This is a per-CPU idle thread. */ #define TDF_UNUSED11 0x00000040 /* Available */ #define TDF_SIGWAIT 0x00000080 /* Ignore ignored signals */ #define TDF_KTH_SUSP 0x00000100 /* kthread is suspended */ #define TDF_ALLPROCSUSP 0x00000200 /* suspended by SINGLE_ALLPROC */ #define TDF_BOUNDARY 0x00000400 /* Thread suspended at user boundary */ #define TDF_UNUSED1 0x00000800 /* Available */ #define TDF_UNUSED2 0x00001000 /* Available */ #define TDF_SBDRY 0x00002000 /* Stop only on usermode boundary. */ #define TDF_UPIBLOCKED 0x00004000 /* Thread blocked on user PI mutex. */ #define TDF_UNUSED3 0x00008000 /* Available */ #define TDF_UNUSED4 0x00010000 /* Available */ #define TDF_UNUSED5 0x00020000 /* Available */ #define TDF_NOLOAD 0x00040000 /* Ignore during load avg calculations. */ #define TDF_SERESTART 0x00080000 /* ERESTART on stop attempts. */ #define TDF_THRWAKEUP 0x00100000 /* Libthr thread must not suspend itself. */ #define TDF_SEINTR 0x00200000 /* EINTR on stop attempts. */ #define TDF_UNUSED12 0x00400000 /* Available */ #define TDF_UNUSED6 0x00800000 /* Available */ #define TDF_SCHED0 0x01000000 /* Reserved for scheduler private use */ #define TDF_SCHED1 0x02000000 /* Reserved for scheduler private use */ #define TDF_SCHED2 0x04000000 /* Reserved for scheduler private use */ #define TDF_SCHED3 0x08000000 /* Reserved for scheduler private use */ #define TDF_UNUSED7 0x10000000 /* Available */ #define TDF_UNUSED8 0x20000000 /* Available */ #define TDF_UNUSED9 0x40000000 /* Available */ #define TDF_UNUSED10 0x80000000 /* Available */ enum { TDA_AST = 0, /* Special: call all non-flagged AST handlers */ TDA_OWEUPC, TDA_HWPMC, TDA_VFORK, TDA_ALRM, TDA_PROF, TDA_MAC, TDA_SCHED, TDA_UFS, TDA_GEOM, TDA_KQUEUE, TDA_RACCT, TDA_MOD1, /* For third party use, before signals are */ TAD_MOD2, /* processed .. */ TDA_SIG, TDA_KTRACE, TDA_SUSPEND, TDA_SIGSUSPEND, TDA_MOD3, /* .. and after */ TAD_MOD4, TDA_MAX, }; #define TDAI(tda) (1U << (tda)) #define td_ast_pending(td, tda) ((td->td_ast & TDAI(tda)) != 0) /* Userland debug flags */ #define TDB_SUSPEND 0x00000001 /* Thread is suspended by debugger */ #define TDB_XSIG 0x00000002 /* Thread is exchanging signal under trace */ #define TDB_USERWR 0x00000004 /* Debugger modified memory or registers */ #define TDB_SCE 0x00000008 /* Thread performs syscall enter */ #define TDB_SCX 0x00000010 /* Thread performs syscall exit */ #define TDB_EXEC 0x00000020 /* TDB_SCX from exec(2) family */ #define TDB_FORK 0x00000040 /* TDB_SCX from fork(2) that created new process */ #define TDB_STOPATFORK 0x00000080 /* Stop at the return from fork (child only) */ #define TDB_CHILD 0x00000100 /* New child indicator for ptrace() */ #define TDB_BORN 0x00000200 /* New LWP indicator for ptrace() */ #define TDB_EXIT 0x00000400 /* Exiting LWP indicator for ptrace() */ #define TDB_VFORK 0x00000800 /* vfork indicator for ptrace() */ #define TDB_FSTP 0x00001000 /* The thread is PT_ATTACH leader */ #define TDB_STEP 0x00002000 /* (x86) PSL_T set for PT_STEP */ #define TDB_SSWITCH 0x00004000 /* Suspended in ptracestop */ #define TDB_BOUNDARY 0x00008000 /* ptracestop() at boundary */ #define TDB_COREDUMPREQ 0x00010000 /* Coredump request */ #define TDB_SCREMOTEREQ 0x00020000 /* Remote syscall request */ /* * "Private" flags kept in td_pflags: * These are only written by curthread and thus need no locking. */ #define TDP_OLDMASK 0x00000001 /* Need to restore mask after suspend. */ #define TDP_INKTR 0x00000002 /* Thread is currently in KTR code. */ #define TDP_INKTRACE 0x00000004 /* Thread is currently in KTRACE code. */ #define TDP_BUFNEED 0x00000008 /* Do not recurse into the buf flush */ #define TDP_COWINPROGRESS 0x00000010 /* Snapshot copy-on-write in progress. */ #define TDP_ALTSTACK 0x00000020 /* Have alternate signal stack. */ #define TDP_DEADLKTREAT 0x00000040 /* Lock acquisition - deadlock treatment. */ #define TDP_NOFAULTING 0x00000080 /* Do not handle page faults. */ #define TDP_SIGFASTBLOCK 0x00000100 /* Fast sigblock active */ #define TDP_OWEUPC 0x00000200 /* Call addupc() at next AST. */ #define TDP_ITHREAD 0x00000400 /* Thread is an interrupt thread. */ #define TDP_SYNCIO 0x00000800 /* Local override, disable async i/o. */ #define TDP_SCHED1 0x00001000 /* Reserved for scheduler private use */ #define TDP_SCHED2 0x00002000 /* Reserved for scheduler private use */ #define TDP_SCHED3 0x00004000 /* Reserved for scheduler private use */ #define TDP_SCHED4 0x00008000 /* Reserved for scheduler private use */ #define TDP_GEOM 0x00010000 /* Settle GEOM before finishing syscall */ #define TDP_SOFTDEP 0x00020000 /* Stuck processing softdep worklist */ #define TDP_NORUNNINGBUF 0x00040000 /* Ignore runningbufspace check */ #define TDP_WAKEUP 0x00080000 /* Don't sleep in umtx cond_wait */ #define TDP_INBDFLUSH 0x00100000 /* Already in BO_BDFLUSH, do not recurse */ #define TDP_KTHREAD 0x00200000 /* This is an official kernel thread */ #define TDP_CALLCHAIN 0x00400000 /* Capture thread's callchain */ #define TDP_IGNSUSP 0x00800000 /* Permission to ignore the MNTK_SUSPEND* */ #define TDP_AUDITREC 0x01000000 /* Audit record pending on thread */ #define TDP_RFPPWAIT 0x02000000 /* Handle RFPPWAIT on syscall exit */ #define TDP_RESETSPUR 0x04000000 /* Reset spurious page fault history. */ #define TDP_NERRNO 0x08000000 /* Last errno is already in td_errno */ #define TDP_UIOHELD 0x10000000 /* Current uio has pages held in td_ma */ #define TDP_INTCPCALLOUT 0x20000000 /* used by netinet/tcp_timer.c */ #define TDP_EXECVMSPC 0x40000000 /* Execve destroyed old vmspace */ #define TDP_SIGFASTPENDING 0x80000000 /* Pending signal due to sigfastblock */ #define TDP2_SBPAGES 0x00000001 /* Owns sbusy on some pages */ #define TDP2_COMPAT32RB 0x00000002 /* compat32 ABI for robust lists */ #define TDP2_ACCT 0x00000004 /* Doing accounting */ /* * Reasons that the current thread can not be run yet. * More than one may apply. */ #define TDI_SUSPENDED 0x0001 /* On suspension queue. */ #define TDI_SLEEPING 0x0002 /* Actually asleep! (tricky). */ #define TDI_LOCK 0x0008 /* Stopped on a lock. */ #define TDI_IWAIT 0x0010 /* Awaiting interrupt. */ #define TD_IS_SLEEPING(td) ((td)->td_inhibitors & TDI_SLEEPING) #define TD_ON_SLEEPQ(td) ((td)->td_wchan != NULL) #define TD_IS_SUSPENDED(td) ((td)->td_inhibitors & TDI_SUSPENDED) #define TD_ON_LOCK(td) ((td)->td_inhibitors & TDI_LOCK) #define TD_AWAITING_INTR(td) ((td)->td_inhibitors & TDI_IWAIT) #ifdef _KERNEL #define TD_GET_STATE(td) atomic_load_int(&(td)->td_state) #else #define TD_GET_STATE(td) ((td)->td_state) #endif #define TD_IS_RUNNING(td) (TD_GET_STATE(td) == TDS_RUNNING) #define TD_ON_RUNQ(td) (TD_GET_STATE(td) == TDS_RUNQ) #define TD_CAN_RUN(td) (TD_GET_STATE(td) == TDS_CAN_RUN) #define TD_IS_INHIBITED(td) (TD_GET_STATE(td) == TDS_INHIBITED) #define TD_ON_UPILOCK(td) ((td)->td_flags & TDF_UPIBLOCKED) #define TD_IS_IDLETHREAD(td) ((td)->td_flags & TDF_IDLETD) #define TD_CAN_ABORT(td) (TD_ON_SLEEPQ((td)) && \ ((td)->td_flags & TDF_SINTR) != 0) #define KTDSTATE(td) \ (((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep" : \ ((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" : \ ((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" : \ ((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding") #define TD_SET_INHIB(td, inhib) do { \ TD_SET_STATE(td, TDS_INHIBITED); \ (td)->td_inhibitors |= (inhib); \ } while (0) #define TD_CLR_INHIB(td, inhib) do { \ if (((td)->td_inhibitors & (inhib)) && \ (((td)->td_inhibitors &= ~(inhib)) == 0)) \ TD_SET_STATE(td, TDS_CAN_RUN); \ } while (0) #define TD_SET_SLEEPING(td) TD_SET_INHIB((td), TDI_SLEEPING) #define TD_SET_LOCK(td) TD_SET_INHIB((td), TDI_LOCK) #define TD_SET_SUSPENDED(td) TD_SET_INHIB((td), TDI_SUSPENDED) #define TD_SET_IWAIT(td) TD_SET_INHIB((td), TDI_IWAIT) #define TD_SET_EXITING(td) TD_SET_INHIB((td), TDI_EXITING) #define TD_CLR_SLEEPING(td) TD_CLR_INHIB((td), TDI_SLEEPING) #define TD_CLR_LOCK(td) TD_CLR_INHIB((td), TDI_LOCK) #define TD_CLR_SUSPENDED(td) TD_CLR_INHIB((td), TDI_SUSPENDED) #define TD_CLR_IWAIT(td) TD_CLR_INHIB((td), TDI_IWAIT) #ifdef _KERNEL #define TD_SET_STATE(td, state) atomic_store_int(&(td)->td_state, state) #else #define TD_SET_STATE(td, state) (td)->td_state = state #endif #define TD_SET_RUNNING(td) TD_SET_STATE(td, TDS_RUNNING) #define TD_SET_RUNQ(td) TD_SET_STATE(td, TDS_RUNQ) #define TD_SET_CAN_RUN(td) TD_SET_STATE(td, TDS_CAN_RUN) #define TD_SBDRY_INTR(td) \ (((td)->td_flags & (TDF_SEINTR | TDF_SERESTART)) != 0) #define TD_SBDRY_ERRNO(td) \ (((td)->td_flags & TDF_SEINTR) != 0 ? EINTR : ERESTART) /* * Process structure. */ struct proc { LIST_ENTRY(proc) p_list; /* (d) List of all processes. */ TAILQ_HEAD(, thread) p_threads; /* (c) all threads. */ struct mtx p_slock; /* process spin lock */ struct ucred *p_ucred; /* (c) Process owner's identity. */ struct filedesc *p_fd; /* (b) Open files. */ struct filedesc_to_leader *p_fdtol; /* (b) Tracking node */ struct pwddesc *p_pd; /* (b) Cwd, chroot, jail, umask */ struct pstats *p_stats; /* (b) Accounting/statistics (CPU). */ struct plimit *p_limit; /* (c) Resource limits. */ struct callout p_limco; /* (c) Limit callout handle */ struct sigacts *p_sigacts; /* (x) Signal actions, state (CPU). */ int p_flag; /* (c) P_* flags. */ int p_flag2; /* (c) P2_* flags. */ enum p_states { PRS_NEW = 0, /* In creation */ PRS_NORMAL, /* threads can be run. */ PRS_ZOMBIE } p_state; /* (j/c) Process status. */ pid_t p_pid; /* (b) Process identifier. */ LIST_ENTRY(proc) p_hash; /* (d) Hash chain. */ LIST_ENTRY(proc) p_pglist; /* (g + e) List of processes in pgrp. */ struct proc *p_pptr; /* (c + e) Pointer to parent process. */ LIST_ENTRY(proc) p_sibling; /* (e) List of sibling processes. */ LIST_HEAD(, proc) p_children; /* (e) Pointer to list of children. */ struct proc *p_reaper; /* (e) My reaper. */ LIST_HEAD(, proc) p_reaplist; /* (e) List of my descendants (if I am reaper). */ LIST_ENTRY(proc) p_reapsibling; /* (e) List of siblings - descendants of the same reaper. */ struct mtx p_mtx; /* (n) Lock for this struct. */ struct mtx p_statmtx; /* Lock for the stats */ struct mtx p_itimmtx; /* Lock for the virt/prof timers */ struct mtx p_profmtx; /* Lock for the profiling */ struct ksiginfo *p_ksi; /* Locked by parent proc lock */ sigqueue_t p_sigqueue; /* (c) Sigs not delivered to a td. */ #define p_siglist p_sigqueue.sq_signals pid_t p_oppid; /* (c + e) Real parent pid. */ /* The following fields are all zeroed upon creation in fork. */ #define p_startzero p_vmspace struct vmspace *p_vmspace; /* (b) Address space. */ u_int p_swtick; /* (c) Tick when swapped in or out. */ u_int p_cowgen; /* (c) Generation of COW pointers. */ struct itimerval p_realtimer; /* (c) Alarm timer. */ struct rusage p_ru; /* (a) Exit information. */ struct rusage_ext p_rux; /* (cu) Internal resource usage. */ struct rusage_ext p_crux; /* (c) Internal child resource usage. */ int p_profthreads; /* (c) Num threads in addupc_task. */ volatile int p_exitthreads; /* (j) Number of threads exiting */ int p_traceflag; /* (o) Kernel trace points. */ struct ktr_io_params *p_ktrioparms; /* (c + o) Params for ktrace. */ struct vnode *p_textvp; /* (b) Vnode of executable. */ struct vnode *p_textdvp; /* (b) Dir containing textvp. */ char *p_binname; /* (b) Binary hardlink name. */ u_int p_lock; /* (c) Prevent exit. */ struct sigiolst p_sigiolst; /* (c) List of sigio sources. */ int p_sigparent; /* (c) Signal to parent on exit. */ int p_sig; /* (n) For core dump/debugger XXX. */ u_int p_ptevents; /* (c + e) ptrace() event mask. */ struct kaioinfo *p_aioinfo; /* (y) ASYNC I/O info. */ struct thread *p_singlethread;/* (c + j) If single threading this is it */ int p_suspcount; /* (j) Num threads in suspended mode. */ struct thread *p_xthread; /* (c) Trap thread */ int p_boundary_count;/* (j) Num threads at user boundary */ int p_pendingcnt; /* how many signals are pending */ struct itimers *p_itimers; /* (c) POSIX interval timers. */ struct procdesc *p_procdesc; /* (e) Process descriptor, if any. */ u_int p_treeflag; /* (e) P_TREE flags */ int p_pendingexits; /* (c) Count of pending thread exits. */ struct filemon *p_filemon; /* (c) filemon-specific data. */ int p_pdeathsig; /* (c) Signal from parent on exit. */ /* End area that is zeroed on creation. */ #define p_endzero p_magic /* The following fields are all copied upon creation in fork. */ #define p_startcopy p_endzero u_int p_magic; /* (b) Magic number. */ int p_osrel; /* (x) osreldate for the binary (from ELF note, if any) */ uint32_t p_fctl0; /* (x) ABI feature control, ELF note */ char p_comm[MAXCOMLEN + 1]; /* (x) Process name. */ struct sysentvec *p_sysent; /* (b) Syscall dispatch info. */ struct pargs *p_args; /* (c) Process arguments. */ rlim_t p_cpulimit; /* (c) Current CPU limit in seconds. */ signed char p_nice; /* (c) Process "nice" value. */ int p_fibnum; /* in this routing domain XXX MRT */ pid_t p_reapsubtree; /* (e) Pid of the direct child of the reaper which spawned our subtree. */ uint64_t p_elf_flags; /* (x) ELF flags */ void *p_elf_brandinfo; /* (x) Elf_Brandinfo, NULL for non ELF binaries. */ sbintime_t p_umtx_min_timeout; /* End area that is copied on creation. */ #define p_endcopy p_xexit u_int p_xexit; /* (c) Exit code. */ u_int p_xsig; /* (c) Stop/kill sig. */ struct pgrp *p_pgrp; /* (c + e) Pointer to process group. */ struct knlist *p_klist; /* (c) Knotes attached to this proc. */ int p_numthreads; /* (c) Number of threads. */ struct mdproc p_md; /* Any machine-dependent fields. */ struct callout p_itcallout; /* (h + c) Interval timer callout. */ u_short p_acflag; /* (c) Accounting flags. */ struct proc *p_peers; /* (r) */ struct proc *p_leader; /* (b) */ void *p_emuldata; /* (c) Emulator state data. */ struct label *p_label; /* (*) Proc (not subject) MAC label. */ STAILQ_HEAD(, ktr_request) p_ktr; /* (o) KTR event queue. */ LIST_HEAD(, mqueue_notifier) p_mqnotifier; /* (c) mqueue notifiers.*/ struct kdtrace_proc *p_dtrace; /* (*) DTrace-specific data. */ struct cv p_pwait; /* (*) wait cv for exit/exec. */ uint64_t p_prev_runtime; /* (c) Resource usage accounting. */ struct racct *p_racct; /* (b) Resource accounting. */ int p_throttled; /* (c) Flag for racct pcpu throttling */ /* * An orphan is the child that has been re-parented to the * debugger as a result of attaching to it. Need to keep * track of them for parent to be able to collect the exit * status of what used to be children. */ LIST_ENTRY(proc) p_orphan; /* (e) List of orphan processes. */ LIST_HEAD(, proc) p_orphans; /* (e) Pointer to list of orphans. */ TAILQ_HEAD(, kq_timer_cb_data) p_kqtim_stop; /* (c) */ LIST_ENTRY(proc) p_jaillist; /* (d) Jail process linkage. */ }; #define p_session p_pgrp->pg_session #define p_pgid p_pgrp->pg_id #define NOCPU (-1) /* For when we aren't on a CPU. */ #define NOCPU_OLD (255) #define MAXCPU_OLD (254) #define PROC_SLOCK(p) mtx_lock_spin(&(p)->p_slock) #define PROC_SUNLOCK(p) mtx_unlock_spin(&(p)->p_slock) #define PROC_SLOCK_ASSERT(p, type) mtx_assert(&(p)->p_slock, (type)) #define PROC_STATLOCK(p) mtx_lock_spin(&(p)->p_statmtx) #define PROC_STATUNLOCK(p) mtx_unlock_spin(&(p)->p_statmtx) #define PROC_STATLOCK_ASSERT(p, type) mtx_assert(&(p)->p_statmtx, (type)) #define PROC_ITIMLOCK(p) mtx_lock_spin(&(p)->p_itimmtx) #define PROC_ITIMUNLOCK(p) mtx_unlock_spin(&(p)->p_itimmtx) #define PROC_ITIMLOCK_ASSERT(p, type) mtx_assert(&(p)->p_itimmtx, (type)) #define PROC_PROFLOCK(p) mtx_lock_spin(&(p)->p_profmtx) #define PROC_PROFUNLOCK(p) mtx_unlock_spin(&(p)->p_profmtx) #define PROC_PROFLOCK_ASSERT(p, type) mtx_assert(&(p)->p_profmtx, (type)) /* These flags are kept in p_flag. */ #define P_ADVLOCK 0x00000001 /* Process may hold a POSIX advisory lock. */ #define P_CONTROLT 0x00000002 /* Has a controlling terminal. */ #define P_KPROC 0x00000004 /* Kernel process. */ #define P_UNUSED3 0x00000008 /* --available-- */ #define P_PPWAIT 0x00000010 /* Parent is waiting for child to exec/exit. */ #define P_PROFIL 0x00000020 /* Has started profiling. */ #define P_STOPPROF 0x00000040 /* Has thread requesting to stop profiling. */ #define P_HADTHREADS 0x00000080 /* Has had threads (no cleanup shortcuts) */ #define P_SUGID 0x00000100 /* Had set id privileges since last exec. */ #define P_SYSTEM 0x00000200 /* System proc: no sigs or stats. */ #define P_SINGLE_EXIT 0x00000400 /* Threads suspending should exit, not wait. */ #define P_TRACED 0x00000800 /* Debugged process being traced. */ #define P_WAITED 0x00001000 /* Someone is waiting for us. */ #define P_WEXIT 0x00002000 /* Working on exiting. */ #define P_EXEC 0x00004000 /* Process called exec. */ #define P_WKILLED 0x00008000 /* Killed, go to kernel/user boundary ASAP. */ #define P_CONTINUED 0x00010000 /* Proc has continued from a stopped state. */ #define P_STOPPED_SIG 0x00020000 /* Stopped due to SIGSTOP/SIGTSTP. */ #define P_STOPPED_TRACE 0x00040000 /* Stopped because of tracing. */ #define P_STOPPED_SINGLE 0x00080000 /* Only 1 thread can continue (not to user). */ #define P_PROTECTED 0x00100000 /* Do not kill on memory overcommit. */ #define P_SIGEVENT 0x00200000 /* Process pending signals changed. */ #define P_SINGLE_BOUNDARY 0x00400000 /* Threads should suspend at user boundary. */ #define P_HWPMC 0x00800000 /* Process is using HWPMCs */ #define P_JAILED 0x01000000 /* Process is in jail. */ #define P_TOTAL_STOP 0x02000000 /* Stopped in stop_all_proc. */ #define P_INEXEC 0x04000000 /* Process is in execve(). */ #define P_STATCHILD 0x08000000 /* Child process stopped or exited. */ #define P_INMEM 0x10000000 /* Loaded into memory, always set. */ #define P_UNUSED1 0x20000000 /* --available-- */ #define P_UNUSED2 0x40000000 /* --available-- */ #define P_PPTRACE 0x80000000 /* PT_TRACEME by vforked child. */ #define P_STOPPED (P_STOPPED_SIG|P_STOPPED_SINGLE|P_STOPPED_TRACE) #define P_SHOULDSTOP(p) ((p)->p_flag & P_STOPPED) #define P_KILLED(p) ((p)->p_flag & P_WKILLED) /* These flags are kept in p_flag2. */ #define P2_INHERIT_PROTECTED 0x00000001 /* New children get P_PROTECTED. */ #define P2_NOTRACE 0x00000002 /* No ptrace(2) attach or coredumps. */ #define P2_NOTRACE_EXEC 0x00000004 /* Keep P2_NOPTRACE on exec(2). */ #define P2_AST_SU 0x00000008 /* Handles SU ast for kthreads. */ #define P2_PTRACE_FSTP 0x00000010 /* SIGSTOP from PT_ATTACH not yet handled. */ #define P2_TRAPCAP 0x00000020 /* SIGTRAP on ENOTCAPABLE */ #define P2_ASLR_ENABLE 0x00000040 /* Force enable ASLR. */ #define P2_ASLR_DISABLE 0x00000080 /* Force disable ASLR. */ #define P2_ASLR_IGNSTART 0x00000100 /* Enable ASLR to consume sbrk area. */ #define P2_PROTMAX_ENABLE 0x00000200 /* Force enable implied PROT_MAX. */ #define P2_PROTMAX_DISABLE 0x00000400 /* Force disable implied PROT_MAX. */ #define P2_STKGAP_DISABLE 0x00000800 /* Disable stack gap for MAP_STACK */ #define P2_STKGAP_DISABLE_EXEC 0x00001000 /* Stack gap disabled after exec */ #define P2_ITSTOPPED 0x00002000 #define P2_PTRACEREQ 0x00004000 /* Active ptrace req */ #define P2_NO_NEW_PRIVS 0x00008000 /* Ignore setuid */ #define P2_WXORX_DISABLE 0x00010000 /* WX mappings enabled */ #define P2_WXORX_ENABLE_EXEC 0x00020000 /* WXORX enabled after exec */ #define P2_WEXIT 0x00040000 /* exit just started, no external thread_single() is permitted */ #define P2_REAPKILLED 0x00080000 #define P2_MEMBAR_PRIVE 0x00100000 /* membar private expedited registered */ #define P2_MEMBAR_PRIVE_SYNCORE 0x00200000 /* membar private expedited sync core registered */ #define P2_MEMBAR_GLOBE 0x00400000 /* membar global expedited registered */ /* Flags protected by proctree_lock, kept in p_treeflags. */ #define P_TREE_ORPHANED 0x00000001 /* Reparented, on orphan list */ #define P_TREE_FIRST_ORPHAN 0x00000002 /* First element of orphan list */ #define P_TREE_REAPER 0x00000004 /* Reaper of subtree */ #define P_TREE_GRPEXITED 0x00000008 /* exit1() done with job ctl */ /* * These were process status values (p_stat), now they are only used in * legacy conversion code. */ #define SIDL 1 /* Process being created by fork. */ #define SRUN 2 /* Currently runnable. */ #define SSLEEP 3 /* Sleeping on an address. */ #define SSTOP 4 /* Process debugging or suspension. */ #define SZOMB 5 /* Awaiting collection by parent. */ #define SWAIT 6 /* Waiting for interrupt. */ #define SLOCK 7 /* Blocked on a lock. */ #define P_MAGIC 0xbeefface #ifdef _KERNEL /* Types and flags for mi_switch(9). */ #define SW_TYPE_MASK 0xff /* First 8 bits are switch type */ #define SWT_OWEPREEMPT 1 /* Switching due to owepreempt. */ #define SWT_TURNSTILE 2 /* Turnstile contention. */ #define SWT_SLEEPQ 3 /* Sleepq wait. */ #define SWT_RELINQUISH 4 /* yield call. */ #define SWT_NEEDRESCHED 5 /* NEEDRESCHED was set. */ #define SWT_IDLE 6 /* Switching from the idle thread. */ #define SWT_IWAIT 7 /* Waiting for interrupts. */ #define SWT_SUSPEND 8 /* Thread suspended. */ #define SWT_REMOTEPREEMPT 9 /* Remote processor preempted. */ #define SWT_REMOTEWAKEIDLE 10 /* Remote processor preempted idle. */ #define SWT_BIND 11 /* Thread bound to a new CPU. */ #define SWT_COUNT 12 /* Number of switch types. */ /* Flags */ #define SW_VOL 0x0100 /* Voluntary switch. */ #define SW_INVOL 0x0200 /* Involuntary switch. */ #define SW_PREEMPT 0x0400 /* The invol switch is a preemption */ /* How values for thread_single(). */ #define SINGLE_NO_EXIT 0 #define SINGLE_EXIT 1 #define SINGLE_BOUNDARY 2 #define SINGLE_ALLPROC 3 #define FOREACH_PROC_IN_SYSTEM(p) \ LIST_FOREACH((p), &allproc, p_list) #define FOREACH_THREAD_IN_PROC(p, td) \ TAILQ_FOREACH((td), &(p)->p_threads, td_plist) #define FIRST_THREAD_IN_PROC(p) TAILQ_FIRST(&(p)->p_threads) /* * We use process IDs <= pid_max <= PID_MAX; PID_MAX + 1 must also fit * in a pid_t, as it is used to represent "no process group". */ #define PID_MAX 99999 #define NO_PID (PID_MAX + 1) #define THREAD0_TID NO_PID extern pid_t pid_max; #define SESS_LEADER(p) ((p)->p_session->s_leader == (p)) /* Lock and unlock a process. */ #define PROC_LOCK(p) mtx_lock(&(p)->p_mtx) #define PROC_TRYLOCK(p) mtx_trylock(&(p)->p_mtx) #define PROC_UNLOCK(p) mtx_unlock(&(p)->p_mtx) #define PROC_LOCKED(p) mtx_owned(&(p)->p_mtx) #define PROC_WAIT_UNLOCKED(p) mtx_wait_unlocked(&(p)->p_mtx) #define PROC_LOCK_ASSERT(p, type) mtx_assert(&(p)->p_mtx, (type)) /* Lock and unlock a process group. */ #define PGRP_LOCK(pg) mtx_lock(&(pg)->pg_mtx) #define PGRP_UNLOCK(pg) mtx_unlock(&(pg)->pg_mtx) #define PGRP_LOCKED(pg) mtx_owned(&(pg)->pg_mtx) #define PGRP_LOCK_ASSERT(pg, type) mtx_assert(&(pg)->pg_mtx, (type)) #define PGRP_LOCK_PGSIGNAL(pg) do { \ if ((pg) != NULL) \ PGRP_LOCK(pg); \ } while (0) #define PGRP_UNLOCK_PGSIGNAL(pg) do { \ if ((pg) != NULL) \ PGRP_UNLOCK(pg); \ } while (0) /* Lock and unlock a session. */ #define SESS_LOCK(s) mtx_lock(&(s)->s_mtx) #define SESS_UNLOCK(s) mtx_unlock(&(s)->s_mtx) #define SESS_LOCKED(s) mtx_owned(&(s)->s_mtx) #define SESS_LOCK_ASSERT(s, type) mtx_assert(&(s)->s_mtx, (type)) /* * A non-zero p_lock prevents the process from exiting; it will sleep in exit1() * until the count reaches zero. * * PHOLD() asserts that the process (except the current process) is * not exiting and increments p_lock. * _PHOLD() is same as PHOLD(), it takes the process locked. */ #define PHOLD(p) do { \ PROC_LOCK(p); \ _PHOLD(p); \ PROC_UNLOCK(p); \ } while (0) #define _PHOLD(p) do { \ PROC_LOCK_ASSERT((p), MA_OWNED); \ KASSERT(!((p)->p_flag & P_WEXIT) || (p) == curproc, \ ("PHOLD of exiting process %p", p)); \ (p)->p_lock++; \ } while (0) #define PROC_ASSERT_HELD(p) do { \ KASSERT((p)->p_lock > 0, ("process %p not held", p)); \ } while (0) #define PRELE(p) do { \ PROC_LOCK((p)); \ _PRELE((p)); \ PROC_UNLOCK((p)); \ } while (0) #define _PRELE(p) do { \ PROC_LOCK_ASSERT((p), MA_OWNED); \ PROC_ASSERT_HELD(p); \ (--(p)->p_lock); \ if (((p)->p_flag & P_WEXIT) && (p)->p_lock == 0) \ wakeup(&(p)->p_lock); \ } while (0) #define PROC_ASSERT_NOT_HELD(p) do { \ KASSERT((p)->p_lock == 0, ("process %p held", p)); \ } while (0) #define PROC_UPDATE_COW(p) do { \ struct proc *_p = (p); \ PROC_LOCK_ASSERT((_p), MA_OWNED); \ atomic_store_int(&_p->p_cowgen, _p->p_cowgen + 1); \ } while (0) #define PROC_COW_CHANGECOUNT(td, p) ({ \ struct thread *_td = (td); \ struct proc *_p = (p); \ MPASS(_td == curthread); \ PROC_LOCK_ASSERT(_p, MA_OWNED); \ _p->p_cowgen - _td->td_cowgen; \ }) /* Control whether or not it is safe for curthread to sleep. */ #define THREAD_NO_SLEEPING() do { \ curthread->td_no_sleeping++; \ MPASS(curthread->td_no_sleeping > 0); \ } while (0) #define THREAD_SLEEPING_OK() do { \ MPASS(curthread->td_no_sleeping > 0); \ curthread->td_no_sleeping--; \ } while (0) #define THREAD_CAN_SLEEP() ((curthread)->td_no_sleeping == 0) #define THREAD_CONTENDS_ON_LOCK(lo) do { \ MPASS(curthread->td_wantedlock == NULL); \ curthread->td_wantedlock = lo; \ } while (0) #define THREAD_CONTENTION_DONE(lo) do { \ MPASS(curthread->td_wantedlock == lo); \ curthread->td_wantedlock = NULL; \ } while (0) #define PIDHASH(pid) (&pidhashtbl[(pid) & pidhash]) #define PIDHASHLOCK(pid) (&pidhashtbl_lock[((pid) & pidhashlock)]) extern LIST_HEAD(pidhashhead, proc) *pidhashtbl; extern struct sx *pidhashtbl_lock; extern u_long pidhash; extern u_long pidhashlock; #define PGRPHASH(pgid) (&pgrphashtbl[(pgid) & pgrphash]) extern LIST_HEAD(pgrphashhead, pgrp) *pgrphashtbl; extern u_long pgrphash; extern struct sx allproc_lock; extern int allproc_gen; extern struct sx proctree_lock; extern struct mtx ppeers_lock; extern struct mtx procid_lock; extern struct proc proc0; /* Process slot for swapper. */ extern struct thread0_storage thread0_st; /* Primary thread in proc0. */ #define thread0 (thread0_st.t0st_thread) extern struct vmspace vmspace0; /* VM space for proc0. */ extern int hogticks; /* Limit on kernel cpu hogs. */ extern int lastpid; extern int nprocs, maxproc; /* Current and max number of procs. */ extern int maxprocperuid; /* Max procs per uid. */ extern u_long ps_arg_cache_limit; LIST_HEAD(proclist, proc); TAILQ_HEAD(procqueue, proc); TAILQ_HEAD(threadqueue, thread); extern struct proclist allproc; /* List of all processes. */ extern struct proc *initproc, *pageproc; /* Process slots for init, pager. */ extern struct uma_zone *proc_zone; extern struct uma_zone *pgrp_zone; struct proc *pfind(pid_t); /* Find process by id. */ struct proc *pfind_any(pid_t); /* Find (zombie) process by id. */ struct proc *pfind_any_locked(pid_t pid); /* Find process by id, locked. */ struct pgrp *pgfind(pid_t); /* Find process group by id. */ void pidhash_slockall(void); /* Shared lock all pid hash lists. */ void pidhash_sunlockall(void); /* Shared unlock all pid hash lists. */ struct fork_req { int fr_flags; int fr_pages; int *fr_pidp; struct proc **fr_procp; int *fr_pd_fd; int fr_pd_flags; struct filecaps *fr_pd_fcaps; int fr_flags2; #define FR2_DROPSIG_CAUGHT 0x00000001 /* Drop caught non-DFL signals */ #define FR2_SHARE_PATHS 0x00000002 /* Invert sense of RFFDG for paths */ #define FR2_KPROC 0x00000004 /* Create a kernel process */ }; /* * pget() flags. */ #define PGET_HOLD 0x00001 /* Hold the process. */ #define PGET_CANSEE 0x00002 /* Check against p_cansee(). */ #define PGET_CANDEBUG 0x00004 /* Check against p_candebug(). */ #define PGET_ISCURRENT 0x00008 /* Check that the found process is current. */ #define PGET_NOTWEXIT 0x00010 /* Check that the process is not in P_WEXIT. */ #define PGET_NOTINEXEC 0x00020 /* Check that the process is not in P_INEXEC. */ #define PGET_NOTID 0x00040 /* Do not assume tid if pid > PID_MAX. */ #define PGET_WANTREAD (PGET_HOLD | PGET_CANDEBUG | PGET_NOTWEXIT) int pget(pid_t pid, int flags, struct proc **pp); /* ast_register() flags */ #define ASTR_ASTF_REQUIRED 0x0001 /* td_ast TDAI(TDA_X) flag set is required for call */ #define ASTR_TDP 0x0002 /* td_pflags flag set is required */ #define ASTR_KCLEAR 0x0004 /* call me on ast_kclear() */ #define ASTR_UNCOND 0x0008 /* call me always */ void ast(struct trapframe *framep); void ast_kclear(struct thread *td); void ast_register(int ast, int ast_flags, int tdp, void (*f)(struct thread *td, int asts)); void ast_deregister(int tda); void ast_sched_locked(struct thread *td, int tda); void ast_sched_mask(struct thread *td, int ast); void ast_sched(struct thread *td, int tda); void ast_unsched_locked(struct thread *td, int tda); struct thread *choosethread(void); int cr_bsd_visible(struct ucred *u1, struct ucred *u2); int cr_cansee(struct ucred *u1, struct ucred *u2); int cr_canseesocket(struct ucred *cred, struct socket *so); int cr_cansignal(struct ucred *cred, struct proc *proc, int signum); int enterpgrp(struct proc *p, pid_t pgid, struct pgrp *pgrp, struct session *sess); int enterthispgrp(struct proc *p, struct pgrp *pgrp); int fork1(struct thread *, struct fork_req *); void fork_exit(void (*)(void *, struct trapframe *), void *, struct trapframe *); void fork_return(struct thread *, struct trapframe *); int inferior(struct proc *p); void itimer_proc_continue(struct proc *p); void kqtimer_proc_continue(struct proc *p); void kern_proc_vmmap_resident(struct vm_map *map, struct vm_map_entry *entry, int *resident_count, bool *super); void kern_yield(int); -void kick_proc0(void); void killjobc(void); int leavepgrp(struct proc *p); int maybe_preempt(struct thread *td); void maybe_yield(void); void mi_switch(int flags); int p_candebug(struct thread *td, struct proc *p); int p_cansee(struct thread *td, struct proc *p); int p_cansched(struct thread *td, struct proc *p); int p_cansignal(struct thread *td, struct proc *p, int signum); int p_canwait(struct thread *td, struct proc *p); struct pargs *pargs_alloc(int len); void pargs_drop(struct pargs *pa); void pargs_hold(struct pargs *pa); void proc_add_orphan(struct proc *child, struct proc *parent); int proc_get_binpath(struct proc *p, char *binname, char **fullpath, char **freepath); int proc_getargv(struct thread *td, struct proc *p, struct sbuf *sb); int proc_getauxv(struct thread *td, struct proc *p, struct sbuf *sb); int proc_getenvv(struct thread *td, struct proc *p, struct sbuf *sb); void procinit(void); int proc_iterate(int (*cb)(struct proc *, void *), void *cbarg); void proc_linkup0(struct proc *p, struct thread *td); void proc_linkup(struct proc *p, struct thread *td); struct proc *proc_realparent(struct proc *child); void proc_reap(struct thread *td, struct proc *p, int *status, int options); void proc_reparent(struct proc *child, struct proc *newparent, bool set_oppid); void proc_set_p2_wexit(struct proc *p); void proc_set_traced(struct proc *p, bool stop); void proc_wkilled(struct proc *p); struct pstats *pstats_alloc(void); void pstats_fork(struct pstats *src, struct pstats *dst); void pstats_free(struct pstats *ps); void proc_clear_orphan(struct proc *p); void reaper_abandon_children(struct proc *p, bool exiting); int securelevel_ge(struct ucred *cr, int level); int securelevel_gt(struct ucred *cr, int level); void sess_hold(struct session *); void sess_release(struct session *); void setrunnable(struct thread *, int); void setsugid(struct proc *p); bool should_yield(void); int sigonstack(size_t sp); void stopevent(struct proc *, u_int, u_int); struct thread *tdfind(lwpid_t, pid_t); void threadinit(void); void tidhash_add(struct thread *); void tidhash_remove(struct thread *); void cpu_idle(int); int cpu_idle_wakeup(int); extern void (*cpu_idle_hook)(sbintime_t); /* Hook to machdep CPU idler. */ void cpu_switch(struct thread *, struct thread *, struct mtx *); void cpu_sync_core(void); void cpu_throw(struct thread *, struct thread *) __dead2; bool curproc_sigkilled(void); void userret(struct thread *, struct trapframe *); void cpu_exit(struct thread *); void exit1(struct thread *, int, int) __dead2; void cpu_copy_thread(struct thread *td, struct thread *td0); bool cpu_exec_vmspace_reuse(struct proc *p, struct vm_map *map); int cpu_fetch_syscall_args(struct thread *td); void cpu_fork(struct thread *, struct proc *, struct thread *, int); void cpu_fork_kthread_handler(struct thread *, void (*)(void *), void *); int cpu_procctl(struct thread *td, int idtype, id_t id, int com, void *data); void cpu_set_syscall_retval(struct thread *, int); int cpu_set_upcall(struct thread *, void (*)(void *), void *, stack_t *); int cpu_set_user_tls(struct thread *, void *tls_base); void cpu_thread_alloc(struct thread *); void cpu_thread_clean(struct thread *); void cpu_thread_exit(struct thread *); void cpu_thread_free(struct thread *); struct thread *thread_alloc(int pages); int thread_check_susp(struct thread *td, bool sleep); void thread_cow_get_proc(struct thread *newtd, struct proc *p); void thread_cow_get(struct thread *newtd, struct thread *td); void thread_cow_free(struct thread *td); void thread_cow_update(struct thread *td); void thread_cow_synced(struct thread *td); int thread_create(struct thread *td, struct rtprio *rtp, int (*initialize_thread)(struct thread *, void *), void *thunk); void thread_exit(void) __dead2; void thread_free(struct thread *td); void thread_link(struct thread *td, struct proc *p); void thread_reap_barrier(void); int thread_recycle(struct thread *, int pages); int thread_single(struct proc *p, int how); void thread_single_end(struct proc *p, int how); void thread_stash(struct thread *td); void thread_stopped(struct proc *p); void childproc_stopped(struct proc *child, int reason); void childproc_continued(struct proc *child); void childproc_exited(struct proc *child); void thread_run_flash(struct thread *td); int thread_suspend_check(int how); bool thread_suspend_check_needed(void); void thread_suspend_switch(struct thread *, struct proc *p); void thread_suspend_one(struct thread *td); void thread_unlink(struct thread *td); void thread_unsuspend(struct proc *p); void thread_wait(struct proc *p); bool stop_all_proc_block(void); void stop_all_proc_unblock(void); void stop_all_proc(void); void resume_all_proc(void); static __inline int curthread_pflags_set(int flags) { struct thread *td; int save; td = curthread; save = ~flags | (td->td_pflags & flags); td->td_pflags |= flags; return (save); } static __inline void curthread_pflags_restore(int save) { curthread->td_pflags &= save; } static __inline int curthread_pflags2_set(int flags) { struct thread *td; int save; td = curthread; save = ~flags | (td->td_pflags2 & flags); td->td_pflags2 |= flags; return (save); } static __inline void curthread_pflags2_restore(int save) { curthread->td_pflags2 &= save; } static __inline __pure2 struct td_sched * td_get_sched(struct thread *td) { return ((struct td_sched *)&td[1]); } #define PROC_ID_PID 0 #define PROC_ID_GROUP 1 #define PROC_ID_SESSION 2 #define PROC_ID_REAP 3 void proc_id_set(int type, pid_t id); void proc_id_set_cond(int type, pid_t id); void proc_id_clear(int type, pid_t id); EVENTHANDLER_LIST_DECLARE(process_ctor); EVENTHANDLER_LIST_DECLARE(process_dtor); EVENTHANDLER_LIST_DECLARE(process_init); EVENTHANDLER_LIST_DECLARE(process_fini); EVENTHANDLER_LIST_DECLARE(process_exit); EVENTHANDLER_LIST_DECLARE(process_fork); EVENTHANDLER_LIST_DECLARE(process_exec); EVENTHANDLER_LIST_DECLARE(thread_ctor); EVENTHANDLER_LIST_DECLARE(thread_dtor); EVENTHANDLER_LIST_DECLARE(thread_init); #endif /* _KERNEL */ #endif /* !_SYS_PROC_H_ */ diff --git a/sys/sys/sleepqueue.h b/sys/sys/sleepqueue.h index 9e456141393b..f28a7f75a50c 100644 --- a/sys/sys/sleepqueue.h +++ b/sys/sys/sleepqueue.h @@ -1,122 +1,122 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2004 John Baldwin * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #ifndef _SYS_SLEEPQUEUE_H_ #define _SYS_SLEEPQUEUE_H_ /* * Sleep queue interface. Sleep/wakeup, condition variables, and sx * locks use a sleep queue for the queue of threads blocked on a sleep * channel. * * A thread calls sleepq_lock() to lock the sleep queue chain associated * with a given wait channel. A thread can then call call sleepq_add() to * add themself onto a sleep queue and call one of the sleepq_wait() * functions to actually go to sleep. If a thread needs to abort a sleep * operation it should call sleepq_release() to unlock the associated sleep * queue chain lock. If the thread also needs to remove itself from a queue * it just enqueued itself on, it can use sleepq_remove() instead. * * If the thread only wishes to sleep for a limited amount of time, it can * call sleepq_set_timeout() after sleepq_add() to setup a timeout. It * should then use one of the sleepq_timedwait() functions to block. * * A thread is normally resumed from a sleep queue by either the * sleepq_signal() or sleepq_broadcast() functions. Sleepq_signal() wakes * the thread with the highest priority that is sleeping on the specified * wait channel. Sleepq_broadcast() wakes all threads that are sleeping * on the specified wait channel. A thread sleeping in an interruptible * sleep can be interrupted by calling sleepq_abort(). A thread can also * be removed from a specified sleep queue using the sleepq_remove() * function. Note that the sleep queue chain must first be locked via * sleepq_lock() before calling sleepq_abort(), sleepq_broadcast(), or * sleepq_signal(). These routines each return a boolean that will be true * if at least one swapped-out thread was resumed. In that case, the caller * is responsible for waking up the swapper by calling kick_proc0() after * releasing the sleep queue chain lock. * * Each thread allocates a sleep queue at thread creation via sleepq_alloc() * and releases it at thread destruction via sleepq_free(). Note that * a sleep queue is not tied to a specific thread and that the sleep queue * released at thread destruction may not be the same sleep queue that the * thread allocated when it was created. * * XXX: Some other parts of the kernel such as ithread sleeping may end up * using this interface as well (death to TDI_IWAIT!) */ struct lock_object; struct sleepqueue; struct thread; #ifdef _KERNEL #define SLEEPQ_TYPE 0x0ff /* Mask of sleep queue types. */ #define SLEEPQ_SLEEP 0x00 /* Used by sleep/wakeup. */ #define SLEEPQ_CONDVAR 0x01 /* Used for a cv. */ #define SLEEPQ_PAUSE 0x02 /* Used by pause. */ #define SLEEPQ_SX 0x03 /* Used by an sx lock. */ #define SLEEPQ_LK 0x04 /* Used by a lockmgr. */ #define SLEEPQ_INTERRUPTIBLE 0x100 /* Sleep is interruptible. */ #define SLEEPQ_UNFAIR 0x200 /* Unfair wakeup order. */ #define SLEEPQ_DROP 0x400 /* Return without lock held. */ void init_sleepqueues(void); -int sleepq_abort(struct thread *td, int intrval); +void sleepq_abort(struct thread *td, int intrval); void sleepq_add(const void *wchan, struct lock_object *lock, const char *wmesg, int flags, int queue); struct sleepqueue *sleepq_alloc(void); -int sleepq_broadcast(const void *wchan, int flags, int pri, int queue); +void sleepq_broadcast(const void *wchan, int flags, int pri, int queue); void sleepq_chains_remove_matching(bool (*matches)(struct thread *)); void sleepq_free(struct sleepqueue *sq); void sleepq_lock(const void *wchan); struct sleepqueue *sleepq_lookup(const void *wchan); void sleepq_release(const void *wchan); void sleepq_remove(struct thread *td, const void *wchan); -int sleepq_remove_matching(struct sleepqueue *sq, int queue, +void sleepq_remove_matching(struct sleepqueue *sq, int queue, bool (*matches)(struct thread *), int pri); void sleepq_remove_nested(struct thread *td); -int sleepq_signal(const void *wchan, int flags, int pri, int queue); +void sleepq_signal(const void *wchan, int flags, int pri, int queue); void sleepq_set_timeout_sbt(const void *wchan, sbintime_t sbt, sbintime_t pr, int flags); #define sleepq_set_timeout(wchan, timo) \ sleepq_set_timeout_sbt((wchan), tick_sbt * (timo), 0, C_HARDCLOCK) u_int sleepq_sleepcnt(const void *wchan, int queue); int sleepq_timedwait(const void *wchan, int pri); int sleepq_timedwait_sig(const void *wchan, int pri); int sleepq_type(const void *wchan); void sleepq_wait(const void *wchan, int pri); int sleepq_wait_sig(const void *wchan, int pri); #ifdef STACK struct sbuf; int sleepq_sbuf_print_stacks(struct sbuf *sb, const void *wchan, int queue, int *count_stacks_printed); #endif #endif /* _KERNEL */ #endif /* !_SYS_SLEEPQUEUE_H_ */ diff --git a/sys/vm/vm_glue.c b/sys/vm/vm_glue.c index 2627ee75dbff..5128c46a1d9f 100644 --- a/sys/vm/vm_glue.c +++ b/sys/vm/vm_glue.c @@ -1,846 +1,838 @@ /*- * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) * * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ #include "opt_vm.h" #include "opt_kstack_pages.h" #include "opt_kstack_max_pages.h" #include "opt_kstack_usage_prof.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if VM_NRESERVLEVEL > 1 #define KVA_KSTACK_QUANTUM_SHIFT (VM_LEVEL_1_ORDER + VM_LEVEL_0_ORDER + \ PAGE_SHIFT) #elif VM_NRESERVLEVEL > 0 #define KVA_KSTACK_QUANTUM_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT) #else #define KVA_KSTACK_QUANTUM_SHIFT (8 + PAGE_SHIFT) #endif #define KVA_KSTACK_QUANTUM (1ul << KVA_KSTACK_QUANTUM_SHIFT) /* * MPSAFE * * WARNING! This code calls vm_map_check_protection() which only checks * the associated vm_map_entry range. It does not determine whether the * contents of the memory is actually readable or writable. In most cases * just checking the vm_map_entry is sufficient within the kernel's address * space. */ bool kernacc(void *addr, int len, int rw) { boolean_t rv; vm_offset_t saddr, eaddr; vm_prot_t prot; KASSERT((rw & ~VM_PROT_ALL) == 0, ("illegal ``rw'' argument to kernacc (%x)\n", rw)); if ((vm_offset_t)addr + len > vm_map_max(kernel_map) || (vm_offset_t)addr + len < (vm_offset_t)addr) return (false); prot = rw; saddr = trunc_page((vm_offset_t)addr); eaddr = round_page((vm_offset_t)addr + len); vm_map_lock_read(kernel_map); rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); vm_map_unlock_read(kernel_map); return (rv == TRUE); } /* * MPSAFE * * WARNING! This code calls vm_map_check_protection() which only checks * the associated vm_map_entry range. It does not determine whether the * contents of the memory is actually readable or writable. vmapbuf(), * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be * used in conjunction with this call. */ bool useracc(void *addr, int len, int rw) { boolean_t rv; vm_prot_t prot; vm_map_t map; KASSERT((rw & ~VM_PROT_ALL) == 0, ("illegal ``rw'' argument to useracc (%x)\n", rw)); prot = rw; map = &curproc->p_vmspace->vm_map; if ((vm_offset_t)addr + len > vm_map_max(map) || (vm_offset_t)addr + len < (vm_offset_t)addr) { return (false); } vm_map_lock_read(map); rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); vm_map_unlock_read(map); return (rv == TRUE); } int vslock(void *addr, size_t len) { vm_offset_t end, last, start; vm_size_t npages; int error; last = (vm_offset_t)addr + len; start = trunc_page((vm_offset_t)addr); end = round_page(last); if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) return (EINVAL); npages = atop(end - start); if (npages > vm_page_max_user_wired) return (ENOMEM); error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); if (error == KERN_SUCCESS) { curthread->td_vslock_sz += len; return (0); } /* * Return EFAULT on error to match copy{in,out}() behaviour * rather than returning ENOMEM like mlock() would. */ return (EFAULT); } void vsunlock(void *addr, size_t len) { /* Rely on the parameter sanity checks performed by vslock(). */ MPASS(curthread->td_vslock_sz >= len); curthread->td_vslock_sz -= len; (void)vm_map_unwire(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); } /* * Pin the page contained within the given object at the given offset. If the * page is not resident, allocate and load it using the given object's pager. * Return the pinned page if successful; otherwise, return NULL. */ static vm_page_t vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset) { vm_page_t m; vm_pindex_t pindex; pindex = OFF_TO_IDX(offset); (void)vm_page_grab_valid_unlocked(&m, object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); return (m); } /* * Return a CPU private mapping to the page at the given offset within the * given object. The page is pinned before it is mapped. */ struct sf_buf * vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset) { vm_page_t m; m = vm_imgact_hold_page(object, offset); if (m == NULL) return (NULL); sched_pin(); return (sf_buf_alloc(m, SFB_CPUPRIVATE)); } /* * Destroy the given CPU private mapping and unpin the page that it mapped. */ void vm_imgact_unmap_page(struct sf_buf *sf) { vm_page_t m; m = sf_buf_page(sf); sf_buf_free(sf); sched_unpin(); vm_page_unwire(m, PQ_ACTIVE); } void vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz) { pmap_sync_icache(map->pmap, va, sz); } static vm_object_t kstack_object; static vm_object_t kstack_alt_object; static uma_zone_t kstack_cache; static int kstack_cache_size; static vmem_t *vmd_kstack_arena[MAXMEMDOM]; static vm_pindex_t vm_kstack_pindex(vm_offset_t ks, int npages); static vm_object_t vm_thread_kstack_size_to_obj(int npages); static int vm_thread_stack_back(vm_offset_t kaddr, vm_page_t ma[], int npages, int req_class, int domain); static int sysctl_kstack_cache_size(SYSCTL_HANDLER_ARGS) { int error, oldsize; oldsize = kstack_cache_size; error = sysctl_handle_int(oidp, arg1, arg2, req); if (error == 0 && req->newptr && oldsize != kstack_cache_size) uma_zone_set_maxcache(kstack_cache, kstack_cache_size); return (error); } SYSCTL_PROC(_vm, OID_AUTO, kstack_cache_size, CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &kstack_cache_size, 0, sysctl_kstack_cache_size, "IU", "Maximum number of cached kernel stacks"); /* * Allocate a virtual address range from a domain kstack arena, following * the specified NUMA policy. */ static vm_offset_t vm_thread_alloc_kstack_kva(vm_size_t size, int domain) { #ifndef __ILP32__ int rv; vmem_t *arena; vm_offset_t addr = 0; size = round_page(size); /* Allocate from the kernel arena for non-standard kstack sizes. */ if (size != ptoa(kstack_pages + KSTACK_GUARD_PAGES)) { arena = vm_dom[domain].vmd_kernel_arena; } else { arena = vmd_kstack_arena[domain]; } rv = vmem_alloc(arena, size, M_BESTFIT | M_NOWAIT, &addr); if (rv == ENOMEM) return (0); KASSERT(atop(addr - VM_MIN_KERNEL_ADDRESS) % (kstack_pages + KSTACK_GUARD_PAGES) == 0, ("%s: allocated kstack KVA not aligned to multiple of kstack size", __func__)); return (addr); #else return (kva_alloc(size)); #endif } /* * Release a region of kernel virtual memory * allocated from the kstack arena. */ static __noinline void vm_thread_free_kstack_kva(vm_offset_t addr, vm_size_t size, int domain) { vmem_t *arena; size = round_page(size); #ifdef __ILP32__ arena = kernel_arena; #else arena = vmd_kstack_arena[domain]; if (size != ptoa(kstack_pages + KSTACK_GUARD_PAGES)) { arena = vm_dom[domain].vmd_kernel_arena; } #endif vmem_free(arena, addr, size); } static vmem_size_t vm_thread_kstack_import_quantum(void) { #ifndef __ILP32__ /* * The kstack_quantum is larger than KVA_QUANTUM to account * for holes induced by guard pages. */ return (KVA_KSTACK_QUANTUM * (kstack_pages + KSTACK_GUARD_PAGES)); #else return (KVA_KSTACK_QUANTUM); #endif } /* * Import KVA from a parent arena into the kstack arena. Imports must be * a multiple of kernel stack pages + guard pages in size. * * Kstack VA allocations need to be aligned so that the linear KVA pindex * is divisible by the total number of kstack VA pages. This is necessary to * make vm_kstack_pindex work properly. * * We import a multiple of KVA_KSTACK_QUANTUM-sized region from the parent * arena. The actual size used by the kstack arena is one kstack smaller to * allow for the necessary alignment adjustments to be made. */ static int vm_thread_kstack_arena_import(void *arena, vmem_size_t size, int flags, vmem_addr_t *addrp) { int error, rem; size_t kpages = kstack_pages + KSTACK_GUARD_PAGES; KASSERT(atop(size) % kpages == 0, ("%s: Size %jd is not a multiple of kstack pages (%d)", __func__, (intmax_t)size, (int)kpages)); error = vmem_xalloc(arena, vm_thread_kstack_import_quantum(), KVA_KSTACK_QUANTUM, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX, flags, addrp); if (error) { return (error); } rem = atop(*addrp - VM_MIN_KERNEL_ADDRESS) % kpages; if (rem != 0) { /* Bump addr to next aligned address */ *addrp = *addrp + (kpages - rem) * PAGE_SIZE; } return (0); } /* * Release KVA from a parent arena into the kstack arena. Released imports must * be a multiple of kernel stack pages + guard pages in size. */ static void vm_thread_kstack_arena_release(void *arena, vmem_addr_t addr, vmem_size_t size) { int rem; size_t kpages __diagused = kstack_pages + KSTACK_GUARD_PAGES; KASSERT(size % kpages == 0, ("%s: Size %jd is not a multiple of kstack pages (%d)", __func__, (intmax_t)size, (int)kpages)); KASSERT((addr - VM_MIN_KERNEL_ADDRESS) % kpages == 0, ("%s: Address %p is not properly aligned (%p)", __func__, (void *)addr, (void *)VM_MIN_KERNEL_ADDRESS)); /* * If the address is not KVA_KSTACK_QUANTUM-aligned we have to decrement * it to account for the shift in kva_import_kstack. */ rem = addr % KVA_KSTACK_QUANTUM; if (rem) { KASSERT(rem <= ptoa(kpages), ("%s: rem > kpages (%d), (%d)", __func__, rem, (int)kpages)); addr -= rem; } vmem_xfree(arena, addr, vm_thread_kstack_import_quantum()); } /* * Create the kernel stack for a new thread. */ static vm_offset_t vm_thread_stack_create(struct domainset *ds, int pages) { vm_page_t ma[KSTACK_MAX_PAGES]; struct vm_domainset_iter di; int req = VM_ALLOC_NORMAL; vm_object_t obj; vm_offset_t ks; int domain, i; obj = vm_thread_kstack_size_to_obj(pages); if (vm_ndomains > 1) obj->domain.dr_policy = ds; vm_domainset_iter_page_init(&di, obj, 0, &domain, &req); do { /* * Get a kernel virtual address for this thread's kstack. */ ks = vm_thread_alloc_kstack_kva(ptoa(pages + KSTACK_GUARD_PAGES), domain); if (ks == 0) continue; ks += ptoa(KSTACK_GUARD_PAGES); /* * Allocate physical pages to back the stack. */ if (vm_thread_stack_back(ks, ma, pages, req, domain) != 0) { vm_thread_free_kstack_kva(ks - ptoa(KSTACK_GUARD_PAGES), ptoa(pages + KSTACK_GUARD_PAGES), domain); continue; } if (KSTACK_GUARD_PAGES != 0) { pmap_qremove(ks - ptoa(KSTACK_GUARD_PAGES), KSTACK_GUARD_PAGES); } for (i = 0; i < pages; i++) vm_page_valid(ma[i]); pmap_qenter(ks, ma, pages); return (ks); } while (vm_domainset_iter_page(&di, obj, &domain) == 0); return (0); } static __noinline void vm_thread_stack_dispose(vm_offset_t ks, int pages) { vm_page_t m; vm_pindex_t pindex; int i, domain; vm_object_t obj = vm_thread_kstack_size_to_obj(pages); pindex = vm_kstack_pindex(ks, pages); domain = vm_phys_domain(vtophys(ks)); pmap_qremove(ks, pages); VM_OBJECT_WLOCK(obj); for (i = 0; i < pages; i++) { m = vm_page_lookup(obj, pindex + i); if (m == NULL) panic("%s: kstack already missing?", __func__); KASSERT(vm_page_domain(m) == domain, ("%s: page %p domain mismatch, expected %d got %d", __func__, m, domain, vm_page_domain(m))); vm_page_xbusy_claim(m); vm_page_unwire_noq(m); vm_page_free(m); } VM_OBJECT_WUNLOCK(obj); kasan_mark((void *)ks, ptoa(pages), ptoa(pages), 0); vm_thread_free_kstack_kva(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), ptoa(pages + KSTACK_GUARD_PAGES), domain); } /* * Allocate the kernel stack for a new thread. */ int vm_thread_new(struct thread *td, int pages) { vm_offset_t ks; u_short ks_domain; /* Bounds check */ if (pages <= 1) pages = kstack_pages; else if (pages > KSTACK_MAX_PAGES) pages = KSTACK_MAX_PAGES; ks = 0; if (pages == kstack_pages && kstack_cache != NULL) ks = (vm_offset_t)uma_zalloc(kstack_cache, M_NOWAIT); /* * Ensure that kstack objects can draw pages from any memory * domain. Otherwise a local memory shortage can block a process * swap-in. */ if (ks == 0) ks = vm_thread_stack_create(DOMAINSET_PREF(PCPU_GET(domain)), pages); if (ks == 0) return (0); ks_domain = vm_phys_domain(vtophys(ks)); KASSERT(ks_domain >= 0 && ks_domain < vm_ndomains, ("%s: invalid domain for kstack %p", __func__, (void *)ks)); td->td_kstack = ks; td->td_kstack_pages = pages; td->td_kstack_domain = ks_domain; return (1); } /* * Dispose of a thread's kernel stack. */ void vm_thread_dispose(struct thread *td) { vm_offset_t ks; int pages; pages = td->td_kstack_pages; ks = td->td_kstack; td->td_kstack = 0; td->td_kstack_pages = 0; td->td_kstack_domain = MAXMEMDOM; if (pages == kstack_pages) { kasan_mark((void *)ks, 0, ptoa(pages), KASAN_KSTACK_FREED); uma_zfree(kstack_cache, (void *)ks); } else { vm_thread_stack_dispose(ks, pages); } } /* * Calculate kstack pindex. * * Uses a non-identity mapping if guard pages are * active to avoid pindex holes in the kstack object. */ static vm_pindex_t vm_kstack_pindex(vm_offset_t ks, int kpages) { vm_pindex_t pindex = atop(ks - VM_MIN_KERNEL_ADDRESS); #ifdef __ILP32__ return (pindex); #else /* * Return the linear pindex if guard pages aren't active or if we are * allocating a non-standard kstack size. */ if (KSTACK_GUARD_PAGES == 0 || kpages != kstack_pages) { return (pindex); } KASSERT(pindex % (kpages + KSTACK_GUARD_PAGES) >= KSTACK_GUARD_PAGES, ("%s: Attempting to calculate kstack guard page pindex", __func__)); return (pindex - (pindex / (kpages + KSTACK_GUARD_PAGES) + 1) * KSTACK_GUARD_PAGES); #endif } /* * Allocate physical pages, following the specified NUMA policy, to back a * kernel stack. */ static int vm_thread_stack_back(vm_offset_t ks, vm_page_t ma[], int npages, int req_class, int domain) { vm_object_t obj = vm_thread_kstack_size_to_obj(npages); vm_pindex_t pindex; vm_page_t m; int n; pindex = vm_kstack_pindex(ks, npages); VM_OBJECT_WLOCK(obj); for (n = 0; n < npages;) { m = vm_page_grab(obj, pindex + n, VM_ALLOC_NOCREAT | VM_ALLOC_WIRED); if (m == NULL) { m = vm_page_alloc_domain(obj, pindex + n, domain, req_class | VM_ALLOC_WIRED); } if (m == NULL) break; ma[n++] = m; } if (n < npages) goto cleanup; VM_OBJECT_WUNLOCK(obj); return (0); cleanup: for (int i = 0; i < n; i++) { m = ma[i]; (void)vm_page_unwire_noq(m); vm_page_free(m); } VM_OBJECT_WUNLOCK(obj); return (ENOMEM); } static vm_object_t vm_thread_kstack_size_to_obj(int npages) { return (npages == kstack_pages ? kstack_object : kstack_alt_object); } static int kstack_import(void *arg, void **store, int cnt, int domain, int flags) { struct domainset *ds; int i; if (domain == UMA_ANYDOMAIN) ds = DOMAINSET_RR(); else ds = DOMAINSET_PREF(domain); for (i = 0; i < cnt; i++) { store[i] = (void *)vm_thread_stack_create(ds, kstack_pages); if (store[i] == NULL) break; } return (i); } static void kstack_release(void *arg, void **store, int cnt) { vm_offset_t ks; int i; for (i = 0; i < cnt; i++) { ks = (vm_offset_t)store[i]; vm_thread_stack_dispose(ks, kstack_pages); } } static void kstack_cache_init(void *null) { vm_size_t kstack_quantum; int domain; kstack_object = vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS)); kstack_cache = uma_zcache_create("kstack_cache", kstack_pages * PAGE_SIZE, NULL, NULL, NULL, NULL, kstack_import, kstack_release, NULL, UMA_ZONE_FIRSTTOUCH); kstack_cache_size = imax(128, mp_ncpus * 4); uma_zone_set_maxcache(kstack_cache, kstack_cache_size); kstack_alt_object = vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS)); kstack_quantum = vm_thread_kstack_import_quantum(); /* * Reduce size used by the kstack arena to allow for * alignment adjustments in vm_thread_kstack_arena_import. */ kstack_quantum -= (kstack_pages + KSTACK_GUARD_PAGES) * PAGE_SIZE; /* * Create the kstack_arena for each domain and set kernel_arena as * parent. */ for (domain = 0; domain < vm_ndomains; domain++) { vmd_kstack_arena[domain] = vmem_create("kstack arena", 0, 0, PAGE_SIZE, 0, M_WAITOK); KASSERT(vmd_kstack_arena[domain] != NULL, ("%s: failed to create domain %d kstack_arena", __func__, domain)); vmem_set_import(vmd_kstack_arena[domain], vm_thread_kstack_arena_import, vm_thread_kstack_arena_release, vm_dom[domain].vmd_kernel_arena, kstack_quantum); } } SYSINIT(vm_kstacks, SI_SUB_KMEM, SI_ORDER_ANY, kstack_cache_init, NULL); #ifdef KSTACK_USAGE_PROF /* * Track maximum stack used by a thread in kernel. */ static int max_kstack_used; SYSCTL_INT(_debug, OID_AUTO, max_kstack_used, CTLFLAG_RD, &max_kstack_used, 0, "Maximum stack depth used by a thread in kernel"); void intr_prof_stack_use(struct thread *td, struct trapframe *frame) { vm_offset_t stack_top; vm_offset_t current; int used, prev_used; /* * Testing for interrupted kernel mode isn't strictly * needed. It optimizes the execution, since interrupts from * usermode will have only the trap frame on the stack. */ if (TRAPF_USERMODE(frame)) return; stack_top = td->td_kstack + td->td_kstack_pages * PAGE_SIZE; current = (vm_offset_t)(uintptr_t)&stack_top; /* * Try to detect if interrupt is using kernel thread stack. * Hardware could use a dedicated stack for interrupt handling. */ if (stack_top <= current || current < td->td_kstack) return; used = stack_top - current; for (;;) { prev_used = max_kstack_used; if (prev_used >= used) break; if (atomic_cmpset_int(&max_kstack_used, prev_used, used)) break; } } #endif /* KSTACK_USAGE_PROF */ /* * Implement fork's actions on an address space. * Here we arrange for the address space to be copied or referenced, * allocate a user struct (pcb and kernel stack), then call the * machine-dependent layer to fill those in and make the new process * ready to run. The new process is set up so that it returns directly * to user mode to avoid stack copying and relocation problems. */ int vm_forkproc(struct thread *td, struct proc *p2, struct thread *td2, struct vmspace *vm2, int flags) { struct proc *p1 = td->td_proc; struct domainset *dset; int error; if ((flags & RFPROC) == 0) { /* * Divorce the memory, if it is shared, essentially * this changes shared memory amongst threads, into * COW locally. */ if ((flags & RFMEM) == 0) { error = vmspace_unshare(p1); if (error) return (error); } cpu_fork(td, p2, td2, flags); return (0); } if (flags & RFMEM) { p2->p_vmspace = p1->p_vmspace; refcount_acquire(&p1->p_vmspace->vm_refcnt); } dset = td2->td_domain.dr_policy; while (vm_page_count_severe_set(&dset->ds_mask)) { vm_wait_doms(&dset->ds_mask, 0); } if ((flags & RFMEM) == 0) { p2->p_vmspace = vm2; if (p1->p_vmspace->vm_shm) shmfork(p1, p2); } /* * cpu_fork will copy and update the pcb, set up the kernel stack, * and make the child ready to run. */ cpu_fork(td, p2, td2, flags); return (0); } /* * Called after process has been wait(2)'ed upon and is being reaped. * The idea is to reclaim resources that we could not reclaim while * the process was still executing. */ void vm_waitproc(struct proc *p) { vmspace_exitfree(p); /* and clean-out the vmspace */ } - -/* - * This used to kick the thread which faults in threads. - */ -void -kick_proc0(void) -{ -}