Index: stable/11/sys/compat/linuxkpi/common/include/linux/module.h =================================================================== --- stable/11/sys/compat/linuxkpi/common/include/linux/module.h (revision 331995) +++ stable/11/sys/compat/linuxkpi/common/include/linux/module.h (revision 331996) @@ -1,108 +1,106 @@ /*- * Copyright (c) 2010 Isilon Systems, Inc. * Copyright (c) 2010 iX Systems, Inc. * Copyright (c) 2010 Panasas, Inc. * Copyright (c) 2013, 2014 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. * * $FreeBSD$ */ #ifndef _LINUX_MODULE_H_ #define _LINUX_MODULE_H_ #include #include #include #include #include #include #include #include #include #include #define MODULE_AUTHOR(name) #define MODULE_DESCRIPTION(name) #define MODULE_LICENSE(name) #define MODULE_INFO(tag, info) #define MODULE_FIRMWARE(firmware) #define THIS_MODULE ((struct module *)0) #define EXPORT_SYMBOL(name) #define EXPORT_SYMBOL_GPL(name) #define __MODULE_STRING(x) __stringify(x) /* OFED pre-module initialization */ #define SI_SUB_OFED_PREINIT (SI_SUB_ROOT_CONF - 2) /* OFED default module initialization */ #define SI_SUB_OFED_MODINIT (SI_SUB_ROOT_CONF - 1) #include static inline void _module_run(void *arg) { void (*fn)(void); #ifdef OFED_DEBUG_INIT char name[1024]; caddr_t pc; long offset; pc = (caddr_t)arg; if (linker_search_symbol_name(pc, name, sizeof(name), &offset) != 0) printf("Running ??? (%p)\n", pc); else printf("Running %s (%p)\n", name, pc); #endif fn = arg; - DROP_GIANT(); fn(); - PICKUP_GIANT(); } #define module_init(fn) \ SYSINIT(fn, SI_SUB_OFED_MODINIT, SI_ORDER_FIRST, _module_run, (fn)) #define module_exit(fn) \ SYSUNINIT(fn, SI_SUB_OFED_MODINIT, SI_ORDER_SECOND, _module_run, (fn)) /* * The following two macros are a workaround for not having a module * load and unload order resolver: */ #define module_init_order(fn, order) \ SYSINIT(fn, SI_SUB_OFED_MODINIT, (order), _module_run, (fn)) #define module_exit_order(fn, order) \ SYSUNINIT(fn, SI_SUB_OFED_MODINIT, (order), _module_run, (fn)) #define module_get(module) #define module_put(module) #define try_module_get(module) 1 #define postcore_initcall(fn) module_init(fn) #endif /* _LINUX_MODULE_H_ */ Index: stable/11/sys/compat/linuxkpi/common/src/linux_compat.c =================================================================== --- stable/11/sys/compat/linuxkpi/common/src/linux_compat.c (revision 331995) +++ stable/11/sys/compat/linuxkpi/common/src/linux_compat.c (revision 331996) @@ -1,2266 +1,2267 @@ /*- * Copyright (c) 2010 Isilon Systems, Inc. * Copyright (c) 2010 iX Systems, Inc. * Copyright (c) 2010 Panasas, Inc. * Copyright (c) 2013-2018 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 __FBSDID("$FreeBSD$"); #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 #endif #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 #endif SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW, 0, "LinuxKPI parameters"); MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat"); #include /* Undo Linux compat changes. */ #undef RB_ROOT #undef file #undef cdev #define RB_ROOT(head) (head)->rbh_root static struct vm_area_struct *linux_cdev_handle_find(void *handle); 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; unsigned long linux_timer_hz_mask; int panic_cmp(struct rb_node *one, struct rb_node *two) { panic("no cmp"); } RB_GENERATE(linux_root, rb_node, __entry, panic_cmp); int kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args) { va_list tmp_va; int len; char *old; char *name; char dummy; old = kobj->name; if (old && fmt == NULL) return (0); /* compute length of string */ va_copy(tmp_va, args); len = vsnprintf(&dummy, 0, fmt, tmp_va); va_end(tmp_va); /* account for zero termination */ len++; /* check for error */ if (len < 1) return (-EINVAL); /* allocate memory for string */ name = kzalloc(len, GFP_KERNEL); if (name == NULL) return (-ENOMEM); vsnprintf(name, len, fmt, args); kobj->name = name; /* free old string */ kfree(old); /* filter new string */ for (; *name != '\0'; name++) if (*name == '/') *name = '!'; return (0); } int kobject_set_name(struct kobject *kobj, const char *fmt, ...) { va_list args; int error; va_start(args, fmt); error = kobject_set_name_vargs(kobj, fmt, args); va_end(args); return (error); } static int kobject_add_complete(struct kobject *kobj, struct kobject *parent) { const struct kobj_type *t; int error; kobj->parent = parent; error = sysfs_create_dir(kobj); if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) { struct attribute **attr; t = kobj->ktype; for (attr = t->default_attrs; *attr != NULL; attr++) { error = sysfs_create_file(kobj, *attr); if (error) break; } if (error) sysfs_remove_dir(kobj); } return (error); } int kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...) { va_list args; int error; va_start(args, fmt); error = kobject_set_name_vargs(kobj, fmt, args); va_end(args); if (error) return (error); return kobject_add_complete(kobj, parent); } void linux_kobject_release(struct kref *kref) { struct kobject *kobj; char *name; kobj = container_of(kref, struct kobject, kref); sysfs_remove_dir(kobj); name = kobj->name; if (kobj->ktype && kobj->ktype->release) kobj->ktype->release(kobj); kfree(name); } static void linux_kobject_kfree(struct kobject *kobj) { kfree(kobj); } static void linux_kobject_kfree_name(struct kobject *kobj) { if (kobj) { kfree(kobj->name); } } const struct kobj_type linux_kfree_type = { .release = linux_kobject_kfree }; 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); } int kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype, struct kobject *parent, const char *fmt, ...) { va_list args; int error; kobject_init(kobj, ktype); kobj->ktype = ktype; kobj->parent = parent; kobj->name = NULL; va_start(args, fmt); error = kobject_set_name_vargs(kobj, fmt, args); va_end(args); if (error) return (error); return kobject_add_complete(kobj, parent); } 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_lock_owned(void *arg) { #ifdef INVARIANTS spinlock_t *s = arg; mtx_assert(&s->m, MA_OWNED); #endif } static void linux_kq_lock_unowned(void *arg) { #ifdef INVARIANTS spinlock_t *s = arg; mtx_assert(&s->m, 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_lock_owned, linux_kq_lock_unowned); return (filp); } void linux_file_free(struct linux_file *filp) { if (filp->_file == NULL) { if (filp->f_shmem != NULL) vm_object_deallocate(filp->f_shmem); kfree(filp); } else { /* * The close method of the character device or file * will free the linux_file structure: */ _fdrop(filp->_file, curthread); } } 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_checked(page, vm_obj, (*mres)->pindex, *mres); vm_page_lock(*mres); vm_page_free(*mres); vm_page_unlock(*mres); *mres = page; } page->valid = VM_PAGE_BITS_ALL; 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; linux_set_current(curthread); /* 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); 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)pidx << PAGE_SHIFT); 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(vmap, &vmf); while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) { kern_yield(PRI_USER); err = vmap->vm_ops->fault(vmap, &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 }, }; #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; 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; linux_set_current(td); if (filp->f_op->open) { error = -filp->f_op->open(file->f_vnode, filp); if (error) { 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); 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(int rw, 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, 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; } #if defined(__amd64__) if (td->td_proc->p_elf_machine == EM_386) { /* try the compat IOCTL handler first */ if (filp->f_op->compat_ioctl != NULL) error = -OPW(fp, td, filp->f_op->compat_ioctl(filp, cmd, (u_long)data)); else error = ENOTTY; /* fallback to the regular IOCTL handler, if any */ if (error == ENOTTY && filp->f_op->unlocked_ioctl != NULL) error = -OPW(fp, td, filp->f_op->unlocked_ioctl(filp, cmd, (u_long)data)); } else #endif if (filp->f_op->unlocked_ioctl != NULL) error = -OPW(fp, td, filp->f_op->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.m, 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.m, 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) { int temp; if (filp->f_kqflags & kqflags) { struct thread *td = curthread; /* get the latest polling state */ temp = OPW(filp->_file, td, filp->f_op->poll(filp, NULL)); 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)) { if (temp & POLLIN) filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ; if (temp & POLLOUT) 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, vm_ooffset_t *offset, vm_size_t size, struct vm_object **object, int nprot, 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 (filp->f_op->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); 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, filp->f_op->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; 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); if (filp->f_op->read) { bytes = OPW(file, td, filp->f_op->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); 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; 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); if (filp->f_op->write) { bytes = OPW(file, td, filp->f_op->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; } 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); return (error); } static int linux_file_poll(struct file *file, int events, struct ucred *active_cred, struct thread *td) { struct linux_file *filp; int revents; filp = (struct linux_file *)file->f_data; filp->f_flags = file->f_flag; linux_set_current(td); if (filp->f_op->poll != NULL) revents = OPW(file, td, filp->f_op->poll(filp, LINUX_POLL_TABLE_NORMAL)) & events; else revents = 0; return (revents); } static int linux_file_close(struct file *file, struct thread *td) { struct linux_file *filp; int error; filp = (struct linux_file *)file->f_data; KASSERT(file_count(filp) == 0, ("File refcount(%d) is not zero", file_count(filp))); filp->f_flags = file->f_flag; linux_set_current(td); linux_poll_wait_dequeue(filp); error = -OPW(file, td, filp->f_op->release(filp->f_vnode, filp)); funsetown(&filp->f_sigio); if (filp->f_vnode != NULL) vdrop(filp->f_vnode); 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; int error; filp = (struct linux_file *)fp->f_data; filp->f_flags = fp->f_flag; error = 0; linux_set_current(td); switch (cmd) { case FIONBIO: break; case FIOASYNC: if (filp->f_op->fasync == NULL) break; error = -OPW(fp, td, filp->f_op->fasync(0, filp, fp->f_flag & FASYNC)); break; case FIOSETOWN: error = fsetown(*(int *)data, &filp->f_sigio); if (error == 0) { if (filp->f_op->fasync == NULL) break; error = -OPW(fp, td, filp->f_op->fasync(0, filp, fp->f_flag & FASYNC)); } break; case FIOGETOWN: *(int *)data = fgetown(&filp->f_sigio); break; default: error = linux_file_ioctl_sub(fp, filp, cmd, data, td); break; } return (error); } static int linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, struct file *fp, vm_ooffset_t *foff, vm_object_t *objp) { /* * Character devices do not provide private mappings * of any kind: */ if ((*maxprotp & VM_PROT_WRITE) == 0 && (prot & VM_PROT_WRITE) != 0) return (EACCES); if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0) return (EINVAL); return (linux_file_mmap_single(fp, foff, objsize, objp, (int)prot, 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; 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; error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp, &foff, &object); if (error != 0) return (error); error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, FALSE, td); if (error != 0) vm_object_deallocate(object); return (error); } static int linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { 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 = vn_stat(vp, sb, td->td_ucred, NOCRED, td); VOP_UNLOCK(vp, 0); return (error); } static int linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { return (0); } 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)); } 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_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__) 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__) pmap_unmapdev((vm_offset_t)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); } char * kvasprintf(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); p = kmalloc(len + 1, gfp); if (p != NULL) vsnprintf(p, len + 1, fmt, 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; linux_set_current(curthread); timer = context; timer->function(timer->data); } void mod_timer(struct timer_list *timer, int expires) { timer->expires = expires; callout_reset(&timer->callout, linux_timer_jiffies_until(expires), &linux_timer_callback_wrapper, timer); } 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); } static void linux_timer_init(void *arg) { /* * 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--; } 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); } else { if (c->done != UINT_MAX) c->done++; wakeup_swapper = 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); - DROP_GIANT(); - 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 + } else { + DROP_GIANT(); sleepq_wait(c, 0); + PICKUP_GIANT(); + } } if (c->done != UINT_MAX) c->done--; sleepq_release(c); intr: - PICKUP_GIANT(); - 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); - DROP_GIANT(); - 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: - PICKUP_GIANT(); - 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_release(struct kobject *kobj) { struct linux_cdev *cdev; struct kobject *parent; cdev = container_of(kobj, struct linux_cdev, kobj); parent = kobj->parent; if (cdev->cdev) destroy_dev(cdev->cdev); kfree(cdev); kobject_put(parent); } static void linux_cdev_static_release(struct kobject *kobj) { struct linux_cdev *cdev; struct kobject *parent; cdev = container_of(kobj, struct linux_cdev, kobj); parent = kobj->parent; if (cdev->cdev) destroy_dev(cdev->cdev); kobject_put(parent); } 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; nb = arg; if (linkstate == LINK_STATE_UP) nb->notifier_call(nb, NETDEV_UP, ifp); else nb->notifier_call(nb, NETDEV_DOWN, ifp); } static void linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; nb = arg; nb->notifier_call(nb, NETDEV_REGISTER, ifp); } static void linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; nb = arg; nb->notifier_call(nb, NETDEV_UNREGISTER, ifp); } static void linux_handle_iflladdr_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; nb = arg; nb->notifier_call(nb, NETDEV_CHANGEADDR, ifp); } static void linux_handle_ifaddr_event(void *arg, struct ifnet *ifp) { struct notifier_block *nb; nb = arg; nb->notifier_call(nb, NETDEV_CHANGEIFADDR, ifp); } 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(void *priv, const void *d1, const void *d2) { 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 *), &thunk, linux_le_cmp); INIT_LIST_HEAD(head); for (i = 0; i < count; i++) list_add_tail(ar[i], head); free(ar, M_KMALLOC); } void linux_irq_handler(void *ent) { struct irq_ent *irqe; linux_set_current(curthread); irqe = ent; irqe->handler(irqe->irq, irqe->arg); } #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_init(cdev, 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_init(cdev, 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); } } #if defined(__i386__) || defined(__amd64__) bool linux_cpu_has_clflush; #endif static void linux_compat_init(void *arg) { struct sysctl_oid *rootoid; int i; #if defined(__i386__) || defined(__amd64__) linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH); #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, 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]); } 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); 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)); Index: stable/11/sys/compat/linuxkpi/common/src/linux_pci.c =================================================================== --- stable/11/sys/compat/linuxkpi/common/src/linux_pci.c (revision 331995) +++ stable/11/sys/compat/linuxkpi/common/src/linux_pci.c (revision 331996) @@ -1,328 +1,324 @@ /*- * Copyright (c) 2015-2016 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 __FBSDID("$FreeBSD$"); #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 static device_probe_t linux_pci_probe; static device_attach_t linux_pci_attach; static device_detach_t linux_pci_detach; static device_suspend_t linux_pci_suspend; static device_resume_t linux_pci_resume; static device_shutdown_t linux_pci_shutdown; static device_method_t pci_methods[] = { DEVMETHOD(device_probe, linux_pci_probe), DEVMETHOD(device_attach, linux_pci_attach), DEVMETHOD(device_detach, linux_pci_detach), DEVMETHOD(device_suspend, linux_pci_suspend), DEVMETHOD(device_resume, linux_pci_resume), DEVMETHOD(device_shutdown, linux_pci_shutdown), DEVMETHOD_END }; static struct pci_driver * linux_pci_find(device_t dev, const struct pci_device_id **idp) { const struct pci_device_id *id; struct pci_driver *pdrv; uint16_t vendor; uint16_t device; vendor = pci_get_vendor(dev); device = pci_get_device(dev); spin_lock(&pci_lock); list_for_each_entry(pdrv, &pci_drivers, links) { for (id = pdrv->id_table; id->vendor != 0; id++) { if (vendor == id->vendor && device == id->device) { *idp = id; spin_unlock(&pci_lock); return (pdrv); } } } spin_unlock(&pci_lock); return (NULL); } static int linux_pci_probe(device_t dev) { const struct pci_device_id *id; struct pci_driver *pdrv; if ((pdrv = linux_pci_find(dev, &id)) == NULL) return (ENXIO); if (device_get_driver(dev) != &pdrv->bsddriver) return (ENXIO); device_set_desc(dev, pdrv->name); return (0); } static int linux_pci_attach(device_t dev) { struct resource_list_entry *rle; struct pci_bus *pbus; struct pci_dev *pdev; struct pci_devinfo *dinfo; struct pci_driver *pdrv; const struct pci_device_id *id; device_t parent; devclass_t devclass; int error; linux_set_current(curthread); pdrv = linux_pci_find(dev, &id); pdev = device_get_softc(dev); parent = device_get_parent(dev); devclass = device_get_devclass(parent); if (pdrv->isdrm) { dinfo = device_get_ivars(parent); device_set_ivars(dev, dinfo); } else { dinfo = device_get_ivars(dev); } pdev->dev.parent = &linux_root_device; pdev->dev.bsddev = dev; INIT_LIST_HEAD(&pdev->dev.irqents); pdev->devfn = PCI_DEVFN(pci_get_slot(dev), pci_get_function(dev)); pdev->device = id->device; pdev->vendor = id->vendor; pdev->subsystem_vendor = dinfo->cfg.subvendor; pdev->subsystem_device = dinfo->cfg.subdevice; pdev->class = pci_get_class(dev); pdev->revision = pci_get_revid(dev); pdev->dev.dma_mask = &pdev->dma_mask; pdev->pdrv = pdrv; kobject_init(&pdev->dev.kobj, &linux_dev_ktype); kobject_set_name(&pdev->dev.kobj, device_get_nameunit(dev)); kobject_add(&pdev->dev.kobj, &linux_root_device.kobj, kobject_name(&pdev->dev.kobj)); rle = linux_pci_get_rle(pdev, SYS_RES_IRQ, 0); if (rle != NULL) pdev->dev.irq = rle->start; else pdev->dev.irq = LINUX_IRQ_INVALID; pdev->irq = pdev->dev.irq; if (pdev->bus == NULL) { pbus = malloc(sizeof(*pbus), M_DEVBUF, M_WAITOK | M_ZERO); pbus->self = pdev; pbus->number = pci_get_bus(dev); pdev->bus = pbus; } - DROP_GIANT(); spin_lock(&pci_lock); list_add(&pdev->links, &pci_devices); spin_unlock(&pci_lock); + error = pdrv->probe(pdev, id); - PICKUP_GIANT(); if (error) { spin_lock(&pci_lock); list_del(&pdev->links); spin_unlock(&pci_lock); put_device(&pdev->dev); error = -error; } return (error); } static int linux_pci_detach(device_t dev) { struct pci_dev *pdev; linux_set_current(curthread); pdev = device_get_softc(dev); - DROP_GIANT(); + pdev->pdrv->remove(pdev); - PICKUP_GIANT(); + spin_lock(&pci_lock); list_del(&pdev->links); spin_unlock(&pci_lock); put_device(&pdev->dev); return (0); } static int linux_pci_suspend(device_t dev) { const struct dev_pm_ops *pmops; struct pm_message pm = { }; struct pci_dev *pdev; int error; error = 0; linux_set_current(curthread); pdev = device_get_softc(dev); pmops = pdev->pdrv->driver.pm; if (pdev->pdrv->suspend != NULL) error = -pdev->pdrv->suspend(pdev, pm); else if (pmops != NULL && pmops->suspend != NULL) { error = -pmops->suspend(&pdev->dev); if (error == 0 && pmops->suspend_late != NULL) error = -pmops->suspend_late(&pdev->dev); } return (error); } static int linux_pci_resume(device_t dev) { const struct dev_pm_ops *pmops; struct pci_dev *pdev; int error; error = 0; linux_set_current(curthread); pdev = device_get_softc(dev); pmops = pdev->pdrv->driver.pm; if (pdev->pdrv->resume != NULL) error = -pdev->pdrv->resume(pdev); else if (pmops != NULL && pmops->resume != NULL) { if (pmops->resume_early != NULL) error = -pmops->resume_early(&pdev->dev); if (error == 0 && pmops->resume != NULL) error = -pmops->resume(&pdev->dev); } return (error); } static int linux_pci_shutdown(device_t dev) { struct pci_dev *pdev; linux_set_current(curthread); pdev = device_get_softc(dev); - if (pdev->pdrv->shutdown != NULL) { - DROP_GIANT(); + if (pdev->pdrv->shutdown != NULL) pdev->pdrv->shutdown(pdev); - PICKUP_GIANT(); - } return (0); } static int _linux_pci_register_driver(struct pci_driver *pdrv, devclass_t dc) { int error; linux_set_current(curthread); spin_lock(&pci_lock); list_add(&pdrv->links, &pci_drivers); spin_unlock(&pci_lock); pdrv->bsddriver.name = pdrv->name; pdrv->bsddriver.methods = pci_methods; pdrv->bsddriver.size = sizeof(struct pci_dev); mtx_lock(&Giant); error = devclass_add_driver(dc, &pdrv->bsddriver, BUS_PASS_DEFAULT, &pdrv->bsdclass); mtx_unlock(&Giant); return (-error); } int linux_pci_register_driver(struct pci_driver *pdrv) { devclass_t dc; dc = devclass_find("pci"); if (dc == NULL) return (-ENXIO); pdrv->isdrm = false; return (_linux_pci_register_driver(pdrv, dc)); } int linux_pci_register_drm_driver(struct pci_driver *pdrv) { devclass_t dc; dc = devclass_create("vgapci"); if (dc == NULL) return (-ENXIO); pdrv->isdrm = true; pdrv->name = "drmn"; return (_linux_pci_register_driver(pdrv, dc)); } void linux_pci_unregister_driver(struct pci_driver *pdrv) { devclass_t bus; bus = devclass_find("pci"); spin_lock(&pci_lock); list_del(&pdrv->links); spin_unlock(&pci_lock); mtx_lock(&Giant); if (bus != NULL) devclass_delete_driver(bus, &pdrv->bsddriver); mtx_unlock(&Giant); } Index: stable/11/sys/compat/linuxkpi/common/src/linux_rcu.c =================================================================== --- stable/11/sys/compat/linuxkpi/common/src/linux_rcu.c (revision 331995) +++ stable/11/sys/compat/linuxkpi/common/src/linux_rcu.c (revision 331996) @@ -1,398 +1,398 @@ /*- * Copyright (c) 2016 Matthew Macy (mmacy@mattmacy.io) * 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 __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * By defining CONFIG_NO_RCU_SKIP LinuxKPI RCU locks and asserts will * not be skipped during panic(). */ #ifdef CONFIG_NO_RCU_SKIP #define RCU_SKIP(void) 0 #else #define RCU_SKIP(void) unlikely(SCHEDULER_STOPPED() || kdb_active) #endif struct callback_head { STAILQ_ENTRY(callback_head) entry; rcu_callback_t func; }; struct linux_epoch_head { STAILQ_HEAD(, callback_head) cb_head; struct mtx lock; struct task task; } __aligned(CACHE_LINE_SIZE); struct linux_epoch_record { ck_epoch_record_t epoch_record; TAILQ_HEAD(, task_struct) ts_head; int cpuid; } __aligned(CACHE_LINE_SIZE); /* * Verify that "struct rcu_head" is big enough to hold "struct * callback_head". This has been done to avoid having to add special * compile flags for including ck_epoch.h to all clients of the * LinuxKPI. */ CTASSERT(sizeof(struct rcu_head) == sizeof(struct callback_head)); /* * Verify that "epoch_record" is at beginning of "struct * linux_epoch_record": */ CTASSERT(offsetof(struct linux_epoch_record, epoch_record) == 0); static ck_epoch_t linux_epoch; static struct linux_epoch_head linux_epoch_head; static DPCPU_DEFINE(struct linux_epoch_record, linux_epoch_record); static void linux_rcu_cleaner_func(void *, int); static void linux_rcu_runtime_init(void *arg __unused) { struct linux_epoch_head *head; int i; ck_epoch_init(&linux_epoch); head = &linux_epoch_head; mtx_init(&head->lock, "LRCU-HEAD", NULL, MTX_DEF); TASK_INIT(&head->task, 0, linux_rcu_cleaner_func, NULL); STAILQ_INIT(&head->cb_head); CPU_FOREACH(i) { struct linux_epoch_record *record; record = &DPCPU_ID_GET(i, linux_epoch_record); record->cpuid = i; ck_epoch_register(&linux_epoch, &record->epoch_record, NULL); TAILQ_INIT(&record->ts_head); } } SYSINIT(linux_rcu_runtime, SI_SUB_CPU, SI_ORDER_ANY, linux_rcu_runtime_init, NULL); static void linux_rcu_runtime_uninit(void *arg __unused) { struct linux_epoch_head *head; head = &linux_epoch_head; /* destroy head lock */ mtx_destroy(&head->lock); } SYSUNINIT(linux_rcu_runtime, SI_SUB_LOCK, SI_ORDER_SECOND, linux_rcu_runtime_uninit, NULL); static void linux_rcu_cleaner_func(void *context __unused, int pending __unused) { struct linux_epoch_head *head; struct callback_head *rcu; STAILQ_HEAD(, callback_head) tmp_head; linux_set_current(curthread); head = &linux_epoch_head; /* move current callbacks into own queue */ mtx_lock(&head->lock); STAILQ_INIT(&tmp_head); STAILQ_CONCAT(&tmp_head, &head->cb_head); mtx_unlock(&head->lock); /* synchronize */ linux_synchronize_rcu(); /* dispatch all callbacks, if any */ while ((rcu = STAILQ_FIRST(&tmp_head)) != NULL) { uintptr_t offset; STAILQ_REMOVE_HEAD(&tmp_head, entry); offset = (uintptr_t)rcu->func; if (offset < LINUX_KFREE_RCU_OFFSET_MAX) kfree((char *)rcu - offset); else rcu->func((struct rcu_head *)rcu); } } void linux_rcu_read_lock(void) { struct linux_epoch_record *record; struct task_struct *ts; if (RCU_SKIP()) return; /* * Pin thread to current CPU so that the unlock code gets the * same per-CPU epoch record: */ sched_pin(); record = &DPCPU_GET(linux_epoch_record); ts = current; /* * Use a critical section to prevent recursion inside * ck_epoch_begin(). Else this function supports recursion. */ critical_enter(); ck_epoch_begin(&record->epoch_record, NULL); ts->rcu_recurse++; if (ts->rcu_recurse == 1) TAILQ_INSERT_TAIL(&record->ts_head, ts, rcu_entry); critical_exit(); } void linux_rcu_read_unlock(void) { struct linux_epoch_record *record; struct task_struct *ts; if (RCU_SKIP()) return; record = &DPCPU_GET(linux_epoch_record); ts = current; /* * Use a critical section to prevent recursion inside * ck_epoch_end(). Else this function supports recursion. */ critical_enter(); ck_epoch_end(&record->epoch_record, NULL); ts->rcu_recurse--; if (ts->rcu_recurse == 0) TAILQ_REMOVE(&record->ts_head, ts, rcu_entry); critical_exit(); sched_unpin(); } static void linux_synchronize_rcu_cb(ck_epoch_t *epoch __unused, ck_epoch_record_t *epoch_record, void *arg __unused) { struct linux_epoch_record *record = container_of(epoch_record, struct linux_epoch_record, epoch_record); struct thread *td = curthread; struct task_struct *ts; /* check if blocked on the current CPU */ if (record->cpuid == PCPU_GET(cpuid)) { bool is_sleeping = 0; u_char prio = 0; /* * Find the lowest priority or sleeping thread which * is blocking synchronization on this CPU core. All * the threads in the queue are CPU-pinned and cannot * go anywhere while the current thread is locked. */ TAILQ_FOREACH(ts, &record->ts_head, rcu_entry) { if (ts->task_thread->td_priority > prio) prio = ts->task_thread->td_priority; is_sleeping |= (ts->task_thread->td_inhibitors != 0); } if (is_sleeping) { thread_unlock(td); pause("W", 1); thread_lock(td); } else { /* set new thread priority */ sched_prio(td, prio); /* task switch */ mi_switch(SW_VOL | SWT_RELINQUISH, NULL); /* * Release the thread lock while yielding to * allow other threads to acquire the lock * pointed to by TDQ_LOCKPTR(td). Else a * deadlock like situation might happen. */ thread_unlock(td); thread_lock(td); } } else { /* * To avoid spinning move execution to the other CPU * which is blocking synchronization. Set highest * thread priority so that code gets run. The thread * priority will be restored later. */ sched_prio(td, 0); sched_bind(td, record->cpuid); } } void linux_synchronize_rcu(void) { struct thread *td; int was_bound; int old_cpu; int old_pinned; u_char old_prio; if (RCU_SKIP()) return; WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "linux_synchronize_rcu() can sleep"); td = curthread; - DROP_GIANT(); - /* * Synchronizing RCU might change the CPU core this function * is running on. Save current values: */ thread_lock(td); + + DROP_GIANT(); old_cpu = PCPU_GET(cpuid); old_pinned = td->td_pinned; old_prio = td->td_priority; was_bound = sched_is_bound(td); sched_unbind(td); td->td_pinned = 0; sched_bind(td, old_cpu); ck_epoch_synchronize_wait(&linux_epoch, &linux_synchronize_rcu_cb, NULL); /* restore CPU binding, if any */ if (was_bound != 0) { sched_bind(td, old_cpu); } else { /* get thread back to initial CPU, if any */ if (old_pinned != 0) sched_bind(td, old_cpu); sched_unbind(td); } /* restore pinned after bind */ td->td_pinned = old_pinned; /* restore thread priority */ sched_prio(td, old_prio); thread_unlock(td); PICKUP_GIANT(); } void linux_rcu_barrier(void) { struct linux_epoch_head *head; linux_synchronize_rcu(); head = &linux_epoch_head; /* wait for callbacks to complete */ taskqueue_drain(taskqueue_fast, &head->task); } void linux_call_rcu(struct rcu_head *context, rcu_callback_t func) { struct callback_head *rcu = (struct callback_head *)context; struct linux_epoch_head *head = &linux_epoch_head; mtx_lock(&head->lock); rcu->func = func; STAILQ_INSERT_TAIL(&head->cb_head, rcu, entry); taskqueue_enqueue(taskqueue_fast, &head->task); mtx_unlock(&head->lock); } int init_srcu_struct(struct srcu_struct *srcu) { return (0); } void cleanup_srcu_struct(struct srcu_struct *srcu) { } int srcu_read_lock(struct srcu_struct *srcu) { linux_rcu_read_lock(); return (0); } void srcu_read_unlock(struct srcu_struct *srcu, int key __unused) { linux_rcu_read_unlock(); } void synchronize_srcu(struct srcu_struct *srcu) { linux_synchronize_rcu(); } void srcu_barrier(struct srcu_struct *srcu) { linux_rcu_barrier(); } Index: stable/11/sys/compat/linuxkpi/common/src/linux_schedule.c =================================================================== --- stable/11/sys/compat/linuxkpi/common/src/linux_schedule.c (revision 331995) +++ stable/11/sys/compat/linuxkpi/common/src/linux_schedule.c (revision 331996) @@ -1,430 +1,418 @@ /*- * 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 __FBSDID("$FreeBSD$"); #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_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; ret = wakeup_swapper = 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); 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_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); - DROP_GIANT(); - /* range check timeout */ if (timeout < 1) timeout = 1; else if (timeout == MAX_SCHEDULE_TIMEOUT) timeout = 0; task = current; /* * Our wait queue entry is on the stack - make sure it doesn't * get swapped out while we sleep. */ PHOLD(task->task_thread->td_proc); 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; } PRELE(task->task_thread->td_proc); - PICKUP_GIANT(); - 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; - DROP_GIANT(); - 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); - PICKUP_GIANT(); - 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_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; - DROP_GIANT(); - /* 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); - PICKUP_GIANT(); - 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; - DROP_GIANT(); - 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); - - PICKUP_GIANT(); return (ret); } bool linux_wake_up_state(struct task_struct *task, unsigned int state) { return (wake_up_task(task, state) != 0); } Index: stable/11 =================================================================== --- stable/11 (revision 331995) +++ stable/11 (revision 331996) Property changes on: stable/11 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r331828