diff --git a/sys/sparc64/sparc64/machdep.c b/sys/sparc64/sparc64/machdep.c index 382fbbf3799f..fa173cbb68f0 100644 --- a/sys/sparc64/sparc64/machdep.c +++ b/sys/sparc64/sparc64/machdep.c @@ -1,939 +1,940 @@ /*- * Copyright (c) 2001 Jake Burkholder. * Copyright (c) 1992 Terrence R. Lambert. * Copyright (c) 1982, 1987, 1990 The Regents of the University of California. * All rights reserved. * * This code is derived from software contributed to Berkeley by * William Jolitz. * * 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. * 4. 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. * * from: @(#)machdep.c 7.4 (Berkeley) 6/3/91 * from: FreeBSD: src/sys/i386/i386/machdep.c,v 1.477 2001/08/27 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_ddb.h" #include "opt_kstack_pages.h" #include "opt_msgbuf.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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include typedef int ofw_vec_t(void *); #ifdef DDB extern vm_offset_t ksym_start, ksym_end; #endif struct tlb_entry *kernel_tlbs; int kernel_tlb_slots; int cold = 1; long Maxmem; long realmem; char pcpu0[PCPU_PAGES * PAGE_SIZE]; struct trapframe frame0; vm_offset_t kstack0; vm_paddr_t kstack0_phys; struct kva_md_info kmi; u_long ofw_vec; u_long ofw_tba; char sparc64_model[32]; static int cpu_use_vis = 1; cpu_block_copy_t *cpu_block_copy; cpu_block_zero_t *cpu_block_zero; void sparc64_init(caddr_t mdp, u_long o1, u_long o2, u_long o3, ofw_vec_t *vec); void sparc64_shutdown_final(void *dummy, int howto); static void cpu_startup(void *); SYSINIT(cpu, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL); CTASSERT((1 << INT_SHIFT) == sizeof(int)); CTASSERT((1 << PTR_SHIFT) == sizeof(char *)); CTASSERT(sizeof(struct reg) == 256); CTASSERT(sizeof(struct fpreg) == 272); CTASSERT(sizeof(struct __mcontext) == 512); CTASSERT((sizeof(struct pcb) & (64 - 1)) == 0); CTASSERT((offsetof(struct pcb, pcb_kfp) & (64 - 1)) == 0); CTASSERT((offsetof(struct pcb, pcb_ufp) & (64 - 1)) == 0); CTASSERT(sizeof(struct pcb) <= ((KSTACK_PAGES * PAGE_SIZE) / 8)); CTASSERT(sizeof(struct pcpu) <= ((PCPU_PAGES * PAGE_SIZE) / 2)); static void cpu_startup(void *arg) { vm_paddr_t physsz; int i; physsz = 0; for (i = 0; i < sparc64_nmemreg; i++) physsz += sparc64_memreg[i].mr_size; printf("real memory = %lu (%lu MB)\n", physsz, physsz / (1024 * 1024)); realmem = (long)physsz / PAGE_SIZE; vm_ksubmap_init(&kmi); bufinit(); vm_pager_bufferinit(); EVENTHANDLER_REGISTER(shutdown_final, sparc64_shutdown_final, NULL, SHUTDOWN_PRI_LAST); printf("avail memory = %lu (%lu MB)\n", cnt.v_free_count * PAGE_SIZE, cnt.v_free_count / ((1024 * 1024) / PAGE_SIZE)); if (bootverbose) printf("machine: %s\n", sparc64_model); cpu_identify(rdpr(ver), PCPU_GET(clock), curcpu); } void cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size) { struct intr_request *ir; int i; pcpu->pc_irtail = &pcpu->pc_irhead; for (i = 0; i < IR_FREE; i++) { ir = &pcpu->pc_irpool[i]; ir->ir_next = pcpu->pc_irfree; pcpu->pc_irfree = ir; } } void spinlock_enter(void) { struct thread *td; register_t pil; td = curthread; if (td->td_md.md_spinlock_count == 0) { pil = rdpr(pil); wrpr(pil, 0, PIL_TICK); td->td_md.md_saved_pil = pil; } td->td_md.md_spinlock_count++; critical_enter(); } void spinlock_exit(void) { struct thread *td; td = curthread; critical_exit(); td->td_md.md_spinlock_count--; if (td->td_md.md_spinlock_count == 0) wrpr(pil, td->td_md.md_saved_pil, 0); } void sparc64_init(caddr_t mdp, u_long o1, u_long o2, u_long o3, ofw_vec_t *vec) { char type[8]; char *env; struct pcpu *pc; vm_offset_t end; caddr_t kmdp; phandle_t child; phandle_t root; uint32_t portid; end = 0; kmdp = NULL; /* * Find out what kind of CPU we have first, for anything that changes * behaviour. */ cpu_impl = VER_IMPL(rdpr(ver)); /* * Do CPU-specific Initialization. */ if (cpu_impl >= CPU_IMPL_ULTRASPARCIII) cheetah_init(); /* * Clear (S)TICK timer (including NPT). */ tick_clear(); /* * UltraSparc II[e,i] based systems come up with the tick interrupt * enabled and a handler that resets the tick counter, causing DELAY() * to not work properly when used early in boot. * UltraSPARC III based systems come up with the system tick interrupt * enabled, causing an interrupt storm on startup since they are not * handled. */ tick_stop(); /* * Initialize Open Firmware (needed for console). */ OF_init(vec); /* * Parse metadata if present and fetch parameters. Must be before the * console is inited so cninit gets the right value of boothowto. */ if (mdp != NULL) { preload_metadata = mdp; kmdp = preload_search_by_type("elf kernel"); if (kmdp != NULL) { boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int); kern_envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *); end = MD_FETCH(kmdp, MODINFOMD_KERNEND, vm_offset_t); kernel_tlb_slots = MD_FETCH(kmdp, MODINFOMD_DTLB_SLOTS, int); kernel_tlbs = (void *)preload_search_info(kmdp, MODINFO_METADATA | MODINFOMD_DTLB); } } init_param1(); /* * Prime our per-CPU data page for use. Note, we are using it for * our stack, so don't pass the real size (PAGE_SIZE) to pcpu_init * or it'll zero it out from under us. */ pc = (struct pcpu *)(pcpu0 + (PCPU_PAGES * PAGE_SIZE)) - 1; pcpu_init(pc, 0, sizeof(struct pcpu)); pc->pc_addr = (vm_offset_t)pcpu0; pc->pc_mid = UPA_CR_GET_MID(ldxa(0, ASI_UPA_CONFIG_REG)); pc->pc_tlb_ctx = TLB_CTX_USER_MIN; pc->pc_tlb_ctx_min = TLB_CTX_USER_MIN; pc->pc_tlb_ctx_max = TLB_CTX_USER_MAX; /* * Determine the OFW node and frequency of the BSP (and ensure the * BSP is in the device tree in the first place). */ pc->pc_node = 0; root = OF_peer(0); for (child = OF_child(root); child != 0; child = OF_peer(child)) { if (OF_getprop(child, "device_type", type, sizeof(type)) <= 0) continue; if (strcmp(type, "cpu") != 0) continue; if (OF_getprop(child, cpu_impl < CPU_IMPL_ULTRASPARCIII ? "upa-portid" : "portid", &portid, sizeof(portid)) <= 0) continue; if (portid == pc->pc_mid) { pc->pc_node = child; break; } } if (pc->pc_node == 0) OF_exit(); if (OF_getprop(child, "clock-frequency", &pc->pc_clock, sizeof(pc->pc_clock)) <= 0) OF_exit(); /* * Provide a DELAY() that works before PCPU_REG is set. We can't * set PCPU_REG without also taking over the trap table or the * firmware will overwrite it. Unfortunately, it's way to early * to also take over the trap table at this point. */ clock_boot = pc->pc_clock; delay_func = delay_boot; /* * Initialize the console before printing anything. * NB: the low-level console drivers require a working DELAY() at * this point. */ cninit(); /* * Panic if there is no metadata. Most likely the kernel was booted * directly, instead of through loader(8). */ if (mdp == NULL || kmdp == NULL) { printf("sparc64_init: no loader metadata.\n" "This probably means you are not using loader(8).\n"); panic("sparc64_init"); } /* * Sanity check the kernel end, which is important. */ if (end == 0) { printf("sparc64_init: warning, kernel end not specified.\n" "Attempting to continue anyway.\n"); end = (vm_offset_t)_end; } cache_init(pc); + cache_enable(); uma_set_align(pc->pc_cache.dc_linesize - 1); cpu_block_copy = bcopy; cpu_block_zero = bzero; getenv_int("machdep.use_vis", &cpu_use_vis); if (cpu_use_vis) { switch (cpu_impl) { case CPU_IMPL_SPARC64: case CPU_IMPL_ULTRASPARCI: case CPU_IMPL_ULTRASPARCII: case CPU_IMPL_ULTRASPARCIIi: case CPU_IMPL_ULTRASPARCIIe: cpu_block_copy = spitfire_block_copy; cpu_block_zero = spitfire_block_zero; break; } } #ifdef SMP mp_init(); #endif /* * Initialize virtual memory and calculate physmem. */ pmap_bootstrap(end); /* * Initialize tunables. */ init_param2(physmem); env = getenv("kernelname"); if (env != NULL) { strlcpy(kernelname, env, sizeof(kernelname)); freeenv(env); } /* * Initialize the interrupt tables. */ intr_init1(); /* * Initialize proc0, set kstack0, frame0, curthread and curpcb. */ proc_linkup0(&proc0, &thread0); proc0.p_md.md_sigtramp = NULL; proc0.p_md.md_utrap = NULL; thread0.td_kstack = kstack0; thread0.td_pcb = (struct pcb *) (thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; frame0.tf_tstate = TSTATE_IE | TSTATE_PEF | TSTATE_PRIV; thread0.td_frame = &frame0; pc->pc_curthread = &thread0; pc->pc_curpcb = thread0.td_pcb; /* * Initialize global registers. */ cpu_setregs(pc); /* * It's now safe to use the real DELAY(). */ delay_func = delay_tick; /* * Initialize the message buffer (after setting trap table). */ msgbufinit(msgbufp, MSGBUF_SIZE); mutex_init(); intr_init2(); /* * Finish pmap initialization now that we're ready for mutexes. */ PMAP_LOCK_INIT(kernel_pmap); OF_getprop(root, "name", sparc64_model, sizeof(sparc64_model) - 1); kdb_init(); #ifdef KDB if (boothowto & RB_KDB) kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger"); #endif } void set_openfirm_callback(ofw_vec_t *vec) { ofw_tba = rdpr(tba); ofw_vec = (u_long)vec; } void sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct trapframe *tf; struct sigframe *sfp; struct sigacts *psp; struct sigframe sf; struct thread *td; struct frame *fp; struct proc *p; u_long sp; int oonstack; int sig; oonstack = 0; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sig = ksi->ksi_signo; psp = p->p_sigacts; mtx_assert(&psp->ps_mtx, MA_OWNED); tf = td->td_frame; sp = tf->tf_sp + SPOFF; oonstack = sigonstack(sp); CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm, catcher, sig); /* Make sure we have a signal trampoline to return to. */ if (p->p_md.md_sigtramp == NULL) { /* * No signal trampoline... kill the process. */ CTR0(KTR_SIG, "sendsig: no sigtramp"); printf("sendsig: %s is too old, rebuild it\n", p->p_comm); sigexit(td, sig); /* NOTREACHED */ } /* Save user context. */ bzero(&sf, sizeof(sf)); get_mcontext(td, &sf.sf_uc.uc_mcontext, 0); sf.sf_uc.uc_sigmask = *mask; sf.sf_uc.uc_stack = td->td_sigstk; sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE; /* Allocate and validate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { sfp = (struct sigframe *)(td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct sigframe)); } else sfp = (struct sigframe *)sp - 1; mtx_unlock(&psp->ps_mtx); PROC_UNLOCK(p); fp = (struct frame *)sfp - 1; /* Translate the signal if appropriate. */ if (p->p_sysent->sv_sigtbl && sig <= p->p_sysent->sv_sigsize) sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)]; /* Build the argument list for the signal handler. */ tf->tf_out[0] = sig; tf->tf_out[2] = (register_t)&sfp->sf_uc; tf->tf_out[4] = (register_t)catcher; if (SIGISMEMBER(psp->ps_siginfo, sig)) { /* Signal handler installed with SA_SIGINFO. */ tf->tf_out[1] = (register_t)&sfp->sf_si; /* Fill in POSIX parts. */ sf.sf_si = ksi->ksi_info; sf.sf_si.si_signo = sig; /* maybe a translated signal */ } else { /* Old FreeBSD-style arguments. */ tf->tf_out[1] = ksi->ksi_code; tf->tf_out[3] = (register_t)ksi->ksi_addr; } /* Copy the sigframe out to the user's stack. */ if (rwindow_save(td) != 0 || copyout(&sf, sfp, sizeof(*sfp)) != 0 || suword(&fp->fr_in[6], tf->tf_out[6]) != 0) { /* * Something is wrong with the stack pointer. * ...Kill the process. */ CTR2(KTR_SIG, "sendsig: sigexit td=%p sfp=%p", td, sfp); PROC_LOCK(p); sigexit(td, SIGILL); /* NOTREACHED */ } tf->tf_tpc = (u_long)p->p_md.md_sigtramp; tf->tf_tnpc = tf->tf_tpc + 4; tf->tf_sp = (u_long)fp - SPOFF; CTR3(KTR_SIG, "sendsig: return td=%p pc=%#lx sp=%#lx", td, tf->tf_tpc, tf->tf_sp); PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } #ifndef _SYS_SYSPROTO_H_ struct sigreturn_args { ucontext_t *ucp; }; #endif /* * MPSAFE */ int sigreturn(struct thread *td, struct sigreturn_args *uap) { struct proc *p; mcontext_t *mc; ucontext_t uc; int error; p = td->td_proc; if (rwindow_save(td)) { PROC_LOCK(p); sigexit(td, SIGILL); } CTR2(KTR_SIG, "sigreturn: td=%p ucp=%p", td, uap->sigcntxp); if (copyin(uap->sigcntxp, &uc, sizeof(uc)) != 0) { CTR1(KTR_SIG, "sigreturn: efault td=%p", td); return (EFAULT); } mc = &uc.uc_mcontext; error = set_mcontext(td, mc); if (error != 0) return (error); PROC_LOCK(p); td->td_sigmask = uc.uc_sigmask; SIG_CANTMASK(td->td_sigmask); signotify(td); PROC_UNLOCK(p); CTR4(KTR_SIG, "sigreturn: return td=%p pc=%#lx sp=%#lx tstate=%#lx", td, mc->mc_tpc, mc->mc_sp, mc->mc_tstate); return (EJUSTRETURN); } #ifdef COMPAT_FREEBSD4 int freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap) { return sigreturn(td, (struct sigreturn_args *)uap); } #endif /* * Construct a PCB from a trapframe. This is called from kdb_trap() where * we want to start a backtrace from the function that caused us to enter * the debugger. We have the context in the trapframe, but base the trace * on the PCB. The PCB doesn't have to be perfect, as long as it contains * enough for a backtrace. */ void makectx(struct trapframe *tf, struct pcb *pcb) { pcb->pcb_pc = tf->tf_tpc; pcb->pcb_sp = tf->tf_sp; } int get_mcontext(struct thread *td, mcontext_t *mc, int flags) { struct trapframe *tf; struct pcb *pcb; tf = td->td_frame; pcb = td->td_pcb; bcopy(tf, mc, sizeof(*tf)); if (flags & GET_MC_CLEAR_RET) { mc->mc_out[0] = 0; mc->mc_out[1] = 0; } mc->mc_flags = _MC_VERSION; critical_enter(); if ((tf->tf_fprs & FPRS_FEF) != 0) { savefpctx(pcb->pcb_ufp); tf->tf_fprs &= ~FPRS_FEF; pcb->pcb_flags |= PCB_FEF; } if ((pcb->pcb_flags & PCB_FEF) != 0) { bcopy(pcb->pcb_ufp, mc->mc_fp, sizeof(mc->mc_fp)); mc->mc_fprs |= FPRS_FEF; } critical_exit(); return (0); } int set_mcontext(struct thread *td, const mcontext_t *mc) { struct trapframe *tf; struct pcb *pcb; uint64_t wstate; if (!TSTATE_SECURE(mc->mc_tstate) || (mc->mc_flags & ((1L << _MC_VERSION_BITS) - 1)) != _MC_VERSION) return (EINVAL); tf = td->td_frame; pcb = td->td_pcb; /* Make sure the windows are spilled first. */ flushw(); wstate = tf->tf_wstate; bcopy(mc, tf, sizeof(*tf)); tf->tf_wstate = wstate; if ((mc->mc_fprs & FPRS_FEF) != 0) { tf->tf_fprs = 0; bcopy(mc->mc_fp, pcb->pcb_ufp, sizeof(pcb->pcb_ufp)); pcb->pcb_flags |= PCB_FEF; } return (0); } /* * Exit the kernel and execute a firmware call that will not return, as * specified by the arguments. */ void cpu_shutdown(void *args) { #ifdef SMP cpu_mp_shutdown(); #endif openfirmware_exit(args); } /* Get current clock frequency for the given CPU ID. */ int cpu_est_clockrate(int cpu_id, uint64_t *rate) { struct pcpu *pc; pc = pcpu_find(cpu_id); if (pc == NULL || rate == NULL) return (EINVAL); *rate = pc->pc_clock; return (0); } /* * Duplicate OF_exit() with a different firmware call function that restores * the trap table, otherwise a RED state exception is triggered in at least * some firmware versions. */ void cpu_halt(void) { static struct { cell_t name; cell_t nargs; cell_t nreturns; } args = { (cell_t)"exit", 0, 0 }; cpu_shutdown(&args); } void sparc64_shutdown_final(void *dummy, int howto) { static struct { cell_t name; cell_t nargs; cell_t nreturns; } args = { (cell_t)"SUNW,power-off", 0, 0 }; /* Turn the power off? */ if ((howto & RB_POWEROFF) != 0) cpu_shutdown(&args); /* In case of halt, return to the firmware */ if ((howto & RB_HALT) != 0) cpu_halt(); } void cpu_idle(int busy) { /* Insert code to halt (until next interrupt) for the idle loop. */ } int cpu_idle_wakeup(int cpu) { return (0); } int ptrace_set_pc(struct thread *td, u_long addr) { td->td_frame->tf_tpc = addr; td->td_frame->tf_tnpc = addr + 4; return (0); } int ptrace_single_step(struct thread *td) { /* TODO; */ return (0); } int ptrace_clear_single_step(struct thread *td) { /* TODO; */ return (0); } void exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings) { struct trapframe *tf; struct pcb *pcb; struct proc *p; u_long sp; /* XXX no cpu_exec */ p = td->td_proc; p->p_md.md_sigtramp = NULL; if (p->p_md.md_utrap != NULL) { utrap_free(p->p_md.md_utrap); p->p_md.md_utrap = NULL; } pcb = td->td_pcb; tf = td->td_frame; sp = rounddown(stack, 16); bzero(pcb, sizeof(*pcb)); bzero(tf, sizeof(*tf)); tf->tf_out[0] = stack; tf->tf_out[3] = p->p_sysent->sv_psstrings; tf->tf_out[6] = sp - SPOFF - sizeof(struct frame); tf->tf_tnpc = entry + 4; tf->tf_tpc = entry; tf->tf_tstate = TSTATE_IE | TSTATE_PEF | TSTATE_MM_TSO; td->td_retval[0] = tf->tf_out[0]; td->td_retval[1] = tf->tf_out[1]; } int fill_regs(struct thread *td, struct reg *regs) { bcopy(td->td_frame, regs, sizeof(*regs)); return (0); } int set_regs(struct thread *td, struct reg *regs) { struct trapframe *tf; if (!TSTATE_SECURE(regs->r_tstate)) return (EINVAL); tf = td->td_frame; regs->r_wstate = tf->tf_wstate; bcopy(regs, tf, sizeof(*regs)); return (0); } int fill_dbregs(struct thread *td, struct dbreg *dbregs) { return (ENOSYS); } int set_dbregs(struct thread *td, struct dbreg *dbregs) { return (ENOSYS); } int fill_fpregs(struct thread *td, struct fpreg *fpregs) { struct trapframe *tf; struct pcb *pcb; pcb = td->td_pcb; tf = td->td_frame; bcopy(pcb->pcb_ufp, fpregs->fr_regs, sizeof(fpregs->fr_regs)); fpregs->fr_fsr = tf->tf_fsr; fpregs->fr_gsr = tf->tf_gsr; return (0); } int set_fpregs(struct thread *td, struct fpreg *fpregs) { struct trapframe *tf; struct pcb *pcb; pcb = td->td_pcb; tf = td->td_frame; tf->tf_fprs &= ~FPRS_FEF; bcopy(fpregs->fr_regs, pcb->pcb_ufp, sizeof(pcb->pcb_ufp)); tf->tf_fsr = fpregs->fr_fsr; tf->tf_gsr = fpregs->fr_gsr; return (0); } struct md_utrap * utrap_alloc(void) { struct md_utrap *ut; ut = malloc(sizeof(struct md_utrap), M_SUBPROC, M_WAITOK | M_ZERO); ut->ut_refcnt = 1; return (ut); } void utrap_free(struct md_utrap *ut) { int refcnt; if (ut == NULL) return; mtx_pool_lock(mtxpool_sleep, ut); ut->ut_refcnt--; refcnt = ut->ut_refcnt; mtx_pool_unlock(mtxpool_sleep, ut); if (refcnt == 0) free(ut, M_SUBPROC); } struct md_utrap * utrap_hold(struct md_utrap *ut) { if (ut == NULL) return (NULL); mtx_pool_lock(mtxpool_sleep, ut); ut->ut_refcnt++; mtx_pool_unlock(mtxpool_sleep, ut); return (ut); } diff --git a/sys/sparc64/sparc64/mp_machdep.c b/sys/sparc64/sparc64/mp_machdep.c index 58921fca208b..0fcd6e60e5d5 100644 --- a/sys/sparc64/sparc64/mp_machdep.c +++ b/sys/sparc64/sparc64/mp_machdep.c @@ -1,617 +1,618 @@ /*- * Copyright (c) 1997 Berkeley Software Design, Inc. 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. * 3. Berkeley Software Design Inc's name may not be used to endorse or * promote products derived from this software without specific prior * written permission. * * THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``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 BERKELEY SOFTWARE DESIGN INC 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. * * from BSDI: locore.s,v 1.36.2.15 1999/08/23 22:34:41 cp Exp */ /*- * Copyright (c) 2002 Jake Burkholder. * Copyright (c) 2007 Marius Strobl * 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 __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 #include #include #include #include static ih_func_t cpu_ipi_ast; static ih_func_t cpu_ipi_preempt; static ih_func_t cpu_ipi_stop; /* * Argument area used to pass data to non-boot processors as they start up. * This must be statically initialized with a known invalid CPU module ID, * since the other processors will use it before the boot CPU enters the * kernel. */ struct cpu_start_args cpu_start_args = { 0, -1, -1, 0, 0, 0 }; struct ipi_cache_args ipi_cache_args; struct ipi_tlb_args ipi_tlb_args; struct pcb stoppcbs[MAXCPU]; struct mtx ipi_mtx; cpu_ipi_selected_t *cpu_ipi_selected; static vm_offset_t mp_tramp; static u_int cpuid_to_mid[MAXCPU]; static int isjbus; static volatile u_int shutdown_cpus; static void cpu_mp_unleash(void *v); static void spitfire_ipi_send(u_int mid, u_long d0, u_long d1, u_long d2); static void sun4u_startcpu(phandle_t cpu, void *func, u_long arg); static void sun4u_stopself(void); static cpu_ipi_selected_t cheetah_ipi_selected; static cpu_ipi_selected_t spitfire_ipi_selected; SYSINIT(cpu_mp_unleash, SI_SUB_SMP, SI_ORDER_FIRST, cpu_mp_unleash, NULL); CTASSERT(MAXCPU <= IDR_CHEETAH_MAX_BN_PAIRS); CTASSERT(MAXCPU <= sizeof(u_int) * NBBY); CTASSERT(MAXCPU <= sizeof(int) * NBBY); void mp_init(void) { struct tte *tp; int i; mp_tramp = (vm_offset_t)OF_claim(NULL, PAGE_SIZE, PAGE_SIZE); if (mp_tramp == (vm_offset_t)-1) panic("%s", __func__); bcopy(mp_tramp_code, (void *)mp_tramp, mp_tramp_code_len); *(vm_offset_t *)(mp_tramp + mp_tramp_tlb_slots) = kernel_tlb_slots; *(vm_offset_t *)(mp_tramp + mp_tramp_func) = (vm_offset_t)mp_startup; tp = (struct tte *)(mp_tramp + mp_tramp_code_len); for (i = 0; i < kernel_tlb_slots; i++) { tp[i].tte_vpn = TV_VPN(kernel_tlbs[i].te_va, TS_4M); tp[i].tte_data = TD_V | TD_4M | TD_PA(kernel_tlbs[i].te_pa) | TD_L | TD_CP | TD_CV | TD_P | TD_W; } for (i = 0; i < PAGE_SIZE; i += sizeof(vm_offset_t)) flush(mp_tramp + i); /* * On UP systems cpu_ipi_selected() can be called while * cpu_mp_start() wasn't so initialize these here. */ if (cpu_impl == CPU_IMPL_ULTRASPARCIIIi || cpu_impl == CPU_IMPL_ULTRASPARCIIIip) isjbus = 1; if (cpu_impl >= CPU_IMPL_ULTRASPARCIII) cpu_ipi_selected = cheetah_ipi_selected; else cpu_ipi_selected = spitfire_ipi_selected; } /* * Probe for other CPUs. */ void cpu_mp_setmaxid(void) { char buf[128]; phandle_t child; u_int cpus; all_cpus = 1 << curcpu; mp_ncpus = 1; cpus = 0; for (child = OF_child(OF_peer(0)); child != 0; child = OF_peer(child)) if (OF_getprop(child, "device_type", buf, sizeof(buf)) > 0 && strcmp(buf, "cpu") == 0) cpus++; mp_maxid = cpus - 1; } int cpu_mp_probe(void) { return (mp_maxid > 0); } struct cpu_group * cpu_topo(void) { return (smp_topo_none()); } static void sun4u_startcpu(phandle_t cpu, void *func, u_long arg) { static struct { cell_t name; cell_t nargs; cell_t nreturns; cell_t cpu; cell_t func; cell_t arg; } args = { (cell_t)"SUNW,start-cpu", 3, }; args.cpu = cpu; args.func = (cell_t)func; args.arg = (cell_t)arg; openfirmware(&args); } /* * Stop the calling CPU. */ static void sun4u_stopself(void) { static struct { cell_t name; cell_t nargs; cell_t nreturns; } args = { (cell_t)"SUNW,stop-self", }; openfirmware_exit(&args); panic("%s: failed.", __func__); } /* * Fire up any non-boot processors. */ void cpu_mp_start(void) { char buf[128]; volatile struct cpu_start_args *csa; struct pcpu *pc; register_t s; vm_offset_t va; phandle_t child; u_int mid; u_int clock; u_int cpuid; mtx_init(&ipi_mtx, "ipi", NULL, MTX_SPIN); intr_setup(PIL_AST, cpu_ipi_ast, -1, NULL, NULL); intr_setup(PIL_RENDEZVOUS, (ih_func_t *)smp_rendezvous_action, -1, NULL, NULL); intr_setup(PIL_STOP, cpu_ipi_stop, -1, NULL, NULL); intr_setup(PIL_PREEMPT, cpu_ipi_preempt, -1, NULL, NULL); cpuid_to_mid[curcpu] = PCPU_GET(mid); csa = &cpu_start_args; for (child = OF_child(OF_peer(0)); child != 0 && mp_ncpus <= MAXCPU; child = OF_peer(child)) { if (OF_getprop(child, "device_type", buf, sizeof(buf)) <= 0 || strcmp(buf, "cpu") != 0) continue; if (OF_getprop(child, cpu_impl < CPU_IMPL_ULTRASPARCIII ? "upa-portid" : "portid", &mid, sizeof(mid)) <= 0) panic("%s: can't get module ID", __func__); if (mid == PCPU_GET(mid)) continue; if (OF_getprop(child, "clock-frequency", &clock, sizeof(clock)) <= 0) panic("%s: can't get clock", __func__); csa->csa_state = 0; sun4u_startcpu(child, (void *)mp_tramp, 0); s = intr_disable(); while (csa->csa_state != CPU_TICKSYNC) ; membar(StoreLoad); csa->csa_tick = rd(tick); if (cpu_impl >= CPU_IMPL_ULTRASPARCIII) { while (csa->csa_state != CPU_STICKSYNC) ; membar(StoreLoad); csa->csa_stick = rdstick(); } while (csa->csa_state != CPU_INIT) ; csa->csa_tick = csa->csa_stick = 0; intr_restore(s); cpuid = mp_ncpus++; cpuid_to_mid[cpuid] = mid; cpu_identify(csa->csa_ver, clock, cpuid); va = kmem_alloc(kernel_map, PCPU_PAGES * PAGE_SIZE); pc = (struct pcpu *)(va + (PCPU_PAGES * PAGE_SIZE)) - 1; pcpu_init(pc, cpuid, sizeof(*pc)); pc->pc_addr = va; pc->pc_clock = clock; pc->pc_mid = mid; pc->pc_node = child; cache_init(pc); all_cpus |= 1 << cpuid; intr_add_cpu(cpuid); } KASSERT(!isjbus || mp_ncpus <= IDR_JALAPENO_MAX_BN_PAIRS, ("%s: can only IPI a maximum of %d JBus-CPUs", __func__, IDR_JALAPENO_MAX_BN_PAIRS)); PCPU_SET(other_cpus, all_cpus & ~(1 << curcpu)); smp_active = 1; } void cpu_mp_announce(void) { } static void cpu_mp_unleash(void *v) { volatile struct cpu_start_args *csa; struct pcpu *pc; register_t s; vm_offset_t va; vm_paddr_t pa; u_int ctx_inc; u_int ctx_min; int i; ctx_min = TLB_CTX_USER_MIN; ctx_inc = (TLB_CTX_USER_MAX - 1) / mp_ncpus; csa = &cpu_start_args; csa->csa_count = mp_ncpus; SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { pc->pc_tlb_ctx = ctx_min; pc->pc_tlb_ctx_min = ctx_min; pc->pc_tlb_ctx_max = ctx_min + ctx_inc; ctx_min += ctx_inc; if (pc->pc_cpuid == curcpu) continue; KASSERT(pc->pc_idlethread != NULL, ("%s: idlethread", __func__)); pc->pc_curthread = pc->pc_idlethread; pc->pc_curpcb = pc->pc_curthread->td_pcb; for (i = 0; i < PCPU_PAGES; i++) { va = pc->pc_addr + i * PAGE_SIZE; pa = pmap_kextract(va); if (pa == 0) panic("%s: pmap_kextract", __func__); csa->csa_ttes[i].tte_vpn = TV_VPN(va, TS_8K); csa->csa_ttes[i].tte_data = TD_V | TD_8K | TD_PA(pa) | TD_L | TD_CP | TD_CV | TD_P | TD_W; } csa->csa_state = 0; csa->csa_pcpu = pc->pc_addr; csa->csa_mid = pc->pc_mid; s = intr_disable(); while (csa->csa_state != CPU_BOOTSTRAP) ; intr_restore(s); } membar(StoreLoad); csa->csa_count = 0; smp_started = 1; } void cpu_mp_bootstrap(struct pcpu *pc) { volatile struct cpu_start_args *csa; csa = &cpu_start_args; if (cpu_impl >= CPU_IMPL_ULTRASPARCIII) cheetah_init(); + cache_enable(); pmap_map_tsb(); /* * Flush all non-locked TLB entries possibly left over by the * firmware. */ tlb_flush_nonlocked(); cpu_setregs(pc); tick_start(); smp_cpus++; KASSERT(curthread != NULL, ("%s: curthread", __func__)); PCPU_SET(other_cpus, all_cpus & ~(1 << curcpu)); printf("SMP: AP CPU #%d Launched!\n", curcpu); csa->csa_count--; membar(StoreLoad); csa->csa_state = CPU_BOOTSTRAP; while (csa->csa_count != 0) ; /* Ok, now enter the scheduler. */ sched_throw(NULL); } void cpu_mp_shutdown(void) { int i; critical_enter(); shutdown_cpus = PCPU_GET(other_cpus); if (stopped_cpus != PCPU_GET(other_cpus)) /* XXX */ stop_cpus(stopped_cpus ^ PCPU_GET(other_cpus)); i = 0; while (shutdown_cpus != 0) { if (i++ > 100000) { printf("timeout shutting down CPUs.\n"); break; } } /* XXX: delay a bit to allow the CPUs to actually enter the PROM. */ DELAY(100000); critical_exit(); } static void cpu_ipi_ast(struct trapframe *tf) { } static void cpu_ipi_stop(struct trapframe *tf) { CTR2(KTR_SMP, "%s: stopped %d", __func__, curcpu); savectx(&stoppcbs[curcpu]); atomic_set_acq_int(&stopped_cpus, PCPU_GET(cpumask)); while ((started_cpus & PCPU_GET(cpumask)) == 0) { if ((shutdown_cpus & PCPU_GET(cpumask)) != 0) { atomic_clear_int(&shutdown_cpus, PCPU_GET(cpumask)); sun4u_stopself(); } } atomic_clear_rel_int(&started_cpus, PCPU_GET(cpumask)); atomic_clear_rel_int(&stopped_cpus, PCPU_GET(cpumask)); CTR2(KTR_SMP, "%s: restarted %d", __func__, curcpu); } static void cpu_ipi_preempt(struct trapframe *tf) { sched_preempt(curthread); } static void spitfire_ipi_selected(u_int cpus, u_long d0, u_long d1, u_long d2) { u_int cpu; KASSERT((cpus & (1 << curcpu)) == 0, ("%s: CPU can't IPI itself", __func__)); while (cpus) { cpu = ffs(cpus) - 1; cpus &= ~(1 << cpu); spitfire_ipi_send(cpuid_to_mid[cpu], d0, d1, d2); } } static void spitfire_ipi_send(u_int mid, u_long d0, u_long d1, u_long d2) { register_t s; u_long ids; int i; KASSERT((ldxa(0, ASI_INTR_DISPATCH_STATUS) & IDR_BUSY) == 0, ("%s: outstanding dispatch", __func__)); for (i = 0; i < IPI_RETRIES; i++) { s = intr_disable(); stxa(AA_SDB_INTR_D0, ASI_SDB_INTR_W, d0); stxa(AA_SDB_INTR_D1, ASI_SDB_INTR_W, d1); stxa(AA_SDB_INTR_D2, ASI_SDB_INTR_W, d2); membar(Sync); stxa(AA_INTR_SEND | (mid << IDC_ITID_SHIFT), ASI_SDB_INTR_W, 0); /* * Workaround for SpitFire erratum #54; do a dummy read * from a SDB internal register before the MEMBAR #Sync * for the write to ASI_SDB_INTR_W (requiring another * MEMBAR #Sync in order to make sure the write has * occurred before the load). */ membar(Sync); (void)ldxa(AA_SDB_CNTL_HIGH, ASI_SDB_CONTROL_R); membar(Sync); while (((ids = ldxa(0, ASI_INTR_DISPATCH_STATUS)) & IDR_BUSY) != 0) ; intr_restore(s); if ((ids & (IDR_BUSY | IDR_NACK)) == 0) return; /* * Leave interrupts enabled for a bit before retrying * in order to avoid deadlocks if the other CPU is also * trying to send an IPI. */ DELAY(2); } if ( #ifdef KDB kdb_active || #endif panicstr != NULL) printf("%s: couldn't send IPI to module 0x%u\n", __func__, mid); else panic("%s: couldn't send IPI", __func__); } static void cheetah_ipi_selected(u_int cpus, u_long d0, u_long d1, u_long d2) { register_t s; u_long ids; u_int bnp; u_int cpu; int i; KASSERT((cpus & (1 << curcpu)) == 0, ("%s: CPU can't IPI itself", __func__)); KASSERT((ldxa(0, ASI_INTR_DISPATCH_STATUS) & IDR_CHEETAH_ALL_BUSY) == 0, ("%s: outstanding dispatch", __func__)); if (cpus == 0) return; ids = 0; for (i = 0; i < IPI_RETRIES * mp_ncpus; i++) { s = intr_disable(); stxa(AA_SDB_INTR_D0, ASI_SDB_INTR_W, d0); stxa(AA_SDB_INTR_D1, ASI_SDB_INTR_W, d1); stxa(AA_SDB_INTR_D2, ASI_SDB_INTR_W, d2); membar(Sync); bnp = 0; for (cpu = 0; cpu < mp_ncpus; cpu++) { if ((cpus & (1 << cpu)) != 0) { stxa(AA_INTR_SEND | (cpuid_to_mid[cpu] << IDC_ITID_SHIFT) | (isjbus ? 0 : bnp << IDC_BN_SHIFT), ASI_SDB_INTR_W, 0); membar(Sync); bnp++; } } while (((ids = ldxa(0, ASI_INTR_DISPATCH_STATUS)) & IDR_CHEETAH_ALL_BUSY) != 0) ; intr_restore(s); if ((ids & (IDR_CHEETAH_ALL_BUSY | IDR_CHEETAH_ALL_NACK)) == 0) return; bnp = 0; for (cpu = 0; cpu < mp_ncpus; cpu++) { if ((cpus & (1 << cpu)) != 0) { if ((ids & (IDR_NACK << (isjbus ? (2 * cpuid_to_mid[cpu]) : (2 * bnp)))) == 0) cpus &= ~(1 << cpu); bnp++; } } /* * Leave interrupts enabled for a bit before retrying * in order to avoid deadlocks if the other CPUs are * also trying to send IPIs. */ DELAY(2 * bnp); } if ( #ifdef KDB kdb_active || #endif panicstr != NULL) printf("%s: couldn't send IPI (cpus=0x%u ids=0x%lu)\n", __func__, cpus, ids); else panic("%s: couldn't send IPI", __func__); } void ipi_selected(u_int cpus, u_int ipi) { cpu_ipi_selected(cpus, 0, (u_long)tl_ipi_level, ipi); } void ipi_all(u_int ipi) { panic("%s", __func__); } void ipi_all_but_self(u_int ipi) { cpu_ipi_selected(PCPU_GET(other_cpus), 0, (u_long)tl_ipi_level, ipi); }