2 * Copyright (c) 1992 Terrence R. Lambert.
3 * Copyright (C) 1994, David Greenman
4 * Copyright (c) 1982, 1987, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * from: @(#)machdep.c 7.4 (Berkeley) 6/3/91
39 * $FreeBSD: src/sys/i386/i386/machdep.c,v 1.385.2.30 2003/05/31 08:48:05 alc Exp $
43 #include "opt_atalk.h"
44 #include "opt_compat.h"
46 #include "opt_directio.h"
49 #include "opt_msgbuf.h"
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/sysproto.h>
55 #include <sys/signalvar.h>
56 #include <sys/kernel.h>
57 #include <sys/linker.h>
58 #include <sys/malloc.h>
61 #include <sys/reboot.h>
63 #include <sys/msgbuf.h>
64 #include <sys/sysent.h>
65 #include <sys/sysctl.h>
66 #include <sys/vmmeter.h>
68 #include <sys/upcall.h>
69 #include <sys/usched.h>
73 #include <vm/vm_param.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vm_extern.h>
82 #include <sys/thread2.h>
83 #include <sys/mplock2.h>
91 #include <machine/cpu.h>
92 #include <machine/clock.h>
93 #include <machine/specialreg.h>
94 #include <machine/md_var.h>
95 #include <machine/pcb_ext.h> /* pcb.h included via sys/user.h */
96 #include <machine/globaldata.h> /* CPU_prvspace */
97 #include <machine/smp.h>
99 #include <machine/perfmon.h>
101 #include <machine/cputypes.h>
103 #include <bus/isa/rtc.h>
104 #include <machine/vm86.h>
105 #include <sys/random.h>
106 #include <sys/ptrace.h>
107 #include <machine/sigframe.h>
108 #include <unistd.h> /* umtx_* functions */
109 #include <pthread.h> /* pthread_yield */
111 extern void dblfault_handler (void);
113 #ifndef CPU_DISABLE_SSE
114 static void set_fpregs_xmm (struct save87 *, struct savexmm *);
115 static void fill_fpregs_xmm (struct savexmm *, struct save87 *);
116 #endif /* CPU_DISABLE_SSE */
118 extern void ffs_rawread_setup(void);
119 #endif /* DIRECTIO */
122 int64_t tsc_offsets[MAXCPU];
124 int64_t tsc_offsets[1];
127 #if defined(SWTCH_OPTIM_STATS)
128 extern int swtch_optim_stats;
129 SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats,
130 CTLFLAG_RD, &swtch_optim_stats, 0, "");
131 SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count,
132 CTLFLAG_RD, &tlb_flush_count, 0, "");
136 sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
138 u_long pmem = ctob(physmem);
140 int error = sysctl_handle_long(oidp, &pmem, 0, req);
144 SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_ULONG|CTLFLAG_RD,
145 0, 0, sysctl_hw_physmem, "LU", "Total system memory in bytes (number of pages * page size)");
148 sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
150 int error = sysctl_handle_int(oidp, 0,
151 ctob((int)Maxmem - vmstats.v_wire_count), req);
155 SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD,
156 0, 0, sysctl_hw_usermem, "IU", "");
158 SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, 0, "");
163 sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
167 /* Unwind the buffer, so that it's linear (possibly starting with
168 * some initial nulls).
170 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr,
171 msgbufp->msg_size-msgbufp->msg_bufr,req);
172 if(error) return(error);
173 if(msgbufp->msg_bufr>0) {
174 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr,
175 msgbufp->msg_bufr,req);
180 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
181 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
183 static int msgbuf_clear;
186 sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
189 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
191 if (!error && req->newptr) {
192 /* Clear the buffer and reset write pointer */
193 bzero(msgbufp->msg_ptr,msgbufp->msg_size);
194 msgbufp->msg_bufr=msgbufp->msg_bufx=0;
200 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
201 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
202 "Clear kernel message buffer");
207 * Send an interrupt to process.
209 * Stack is set up to allow sigcode stored
210 * at top to call routine, followed by kcall
211 * to sigreturn routine below. After sigreturn
212 * resets the signal mask, the stack, and the
213 * frame pointer, it returns to the user
217 sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
219 struct lwp *lp = curthread->td_lwp;
220 struct proc *p = lp->lwp_proc;
221 struct trapframe *regs;
222 struct sigacts *psp = p->p_sigacts;
223 struct sigframe sf, *sfp;
226 regs = lp->lwp_md.md_regs;
227 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
229 /* save user context */
230 bzero(&sf, sizeof(struct sigframe));
231 sf.sf_uc.uc_sigmask = *mask;
232 sf.sf_uc.uc_stack = lp->lwp_sigstk;
233 sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
234 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_gs, sizeof(struct trapframe));
236 /* make the size of the saved context visible to userland */
237 sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext);
239 /* save mailbox pending state for syscall interlock semantics */
240 if (p->p_flag & P_MAILBOX)
241 sf.sf_uc.uc_mcontext.mc_xflags |= PGEX_MAILBOX;
244 /* Allocate and validate space for the signal handler context. */
245 if ((lp->lwp_flag & LWP_ALTSTACK) != 0 && !oonstack &&
246 SIGISMEMBER(psp->ps_sigonstack, sig)) {
247 sfp = (struct sigframe *)(lp->lwp_sigstk.ss_sp +
248 lp->lwp_sigstk.ss_size - sizeof(struct sigframe));
249 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
252 sfp = (struct sigframe *)regs->tf_esp - 1;
254 /* Translate the signal is appropriate */
255 if (p->p_sysent->sv_sigtbl) {
256 if (sig <= p->p_sysent->sv_sigsize)
257 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
260 /* Build the argument list for the signal handler. */
262 sf.sf_ucontext = (register_t)&sfp->sf_uc;
263 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
264 /* Signal handler installed with SA_SIGINFO. */
265 sf.sf_siginfo = (register_t)&sfp->sf_si;
266 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
268 /* fill siginfo structure */
269 sf.sf_si.si_signo = sig;
270 sf.sf_si.si_code = code;
271 sf.sf_si.si_addr = (void*)regs->tf_err;
274 /* Old FreeBSD-style arguments. */
275 sf.sf_siginfo = code;
276 sf.sf_addr = regs->tf_err;
277 sf.sf_ahu.sf_handler = catcher;
282 * If we're a vm86 process, we want to save the segment registers.
283 * We also change eflags to be our emulated eflags, not the actual
286 if (regs->tf_eflags & PSL_VM) {
287 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
288 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
290 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
291 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
292 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
293 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
295 if (vm86->vm86_has_vme == 0)
296 sf.sf_uc.uc_mcontext.mc_eflags =
297 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
298 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
301 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
302 * syscalls made by the signal handler. This just avoids
303 * wasting time for our lazy fixup of such faults. PSL_NT
304 * does nothing in vm86 mode, but vm86 programs can set it
305 * almost legitimately in probes for old cpu types.
307 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
312 * Save the FPU state and reinit the FP unit
314 npxpush(&sf.sf_uc.uc_mcontext);
317 * Copy the sigframe out to the user's stack.
319 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
321 * Something is wrong with the stack pointer.
322 * ...Kill the process.
327 regs->tf_esp = (int)sfp;
328 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
331 * i386 abi specifies that the direction flag must be cleared
334 regs->tf_eflags &= ~(PSL_T|PSL_D);
336 regs->tf_cs = _ucodesel;
337 regs->tf_ds = _udatasel;
338 regs->tf_es = _udatasel;
339 if (regs->tf_trapno == T_PROTFLT) {
340 regs->tf_fs = _udatasel;
341 regs->tf_gs = _udatasel;
343 regs->tf_ss = _udatasel;
347 * Sanitize the trapframe for a virtual kernel passing control to a custom
350 * Allow userland to set or maintain PSL_RF, the resume flag. This flag
351 * basically controls whether the return PC should skip the first instruction
352 * (as in an explicit system call) or re-execute it (as in an exception).
355 cpu_sanitize_frame(struct trapframe *frame)
357 frame->tf_cs = _ucodesel;
358 frame->tf_ds = _udatasel;
359 frame->tf_es = _udatasel;
361 frame->tf_fs = _udatasel;
362 frame->tf_gs = _udatasel;
364 frame->tf_ss = _udatasel;
365 frame->tf_eflags &= (PSL_RF | PSL_USERCHANGE);
366 frame->tf_eflags |= PSL_RESERVED_DEFAULT | PSL_I;
371 cpu_sanitize_tls(struct savetls *tls)
373 struct segment_descriptor *desc;
376 for (i = 0; i < NGTLS; ++i) {
378 if (desc->sd_dpl == 0 && desc->sd_type == 0)
380 if (desc->sd_def32 == 0)
382 if (desc->sd_type != SDT_MEMRWA)
384 if (desc->sd_dpl != SEL_UPL)
386 if (desc->sd_xx != 0 || desc->sd_p != 1)
393 * sigreturn(ucontext_t *sigcntxp)
395 * System call to cleanup state after a signal
396 * has been taken. Reset signal mask and
397 * stack state from context left by sendsig (above).
398 * Return to previous pc and psl as specified by
399 * context left by sendsig. Check carefully to
400 * make sure that the user has not modified the
401 * state to gain improper privileges.
405 #define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
406 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
409 sys_sigreturn(struct sigreturn_args *uap)
411 struct lwp *lp = curthread->td_lwp;
412 struct proc *p = lp->lwp_proc;
413 struct trapframe *regs;
419 error = copyin(uap->sigcntxp, &ucp, sizeof(ucp));
423 regs = lp->lwp_md.md_regs;
424 eflags = ucp.uc_mcontext.mc_eflags;
427 if (eflags & PSL_VM) {
428 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
429 struct vm86_kernel *vm86;
432 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
433 * set up the vm86 area, and we can't enter vm86 mode.
435 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
437 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
438 if (vm86->vm86_inited == 0)
441 /* go back to user mode if both flags are set */
442 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
443 trapsignal(lp->lwp_proc, SIGBUS, 0);
445 if (vm86->vm86_has_vme) {
446 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
447 (eflags & VME_USERCHANGE) | PSL_VM;
449 vm86->vm86_eflags = eflags; /* save VIF, VIP */
450 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
452 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
453 tf->tf_eflags = eflags;
454 tf->tf_vm86_ds = tf->tf_ds;
455 tf->tf_vm86_es = tf->tf_es;
456 tf->tf_vm86_fs = tf->tf_fs;
457 tf->tf_vm86_gs = tf->tf_gs;
458 tf->tf_ds = _udatasel;
459 tf->tf_es = _udatasel;
461 tf->tf_fs = _udatasel;
462 tf->tf_gs = _udatasel;
468 * Don't allow users to change privileged or reserved flags.
471 * XXX do allow users to change the privileged flag PSL_RF.
472 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
473 * should sometimes set it there too. tf_eflags is kept in
474 * the signal context during signal handling and there is no
475 * other place to remember it, so the PSL_RF bit may be
476 * corrupted by the signal handler without us knowing.
477 * Corruption of the PSL_RF bit at worst causes one more or
478 * one less debugger trap, so allowing it is fairly harmless.
480 if (!EFL_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
481 kprintf("sigreturn: eflags = 0x%x\n", eflags);
486 * Don't allow users to load a valid privileged %cs. Let the
487 * hardware check for invalid selectors, excess privilege in
488 * other selectors, invalid %eip's and invalid %esp's.
490 cs = ucp.uc_mcontext.mc_cs;
491 if (!CS_SECURE(cs)) {
492 kprintf("sigreturn: cs = 0x%x\n", cs);
493 trapsignal(lp, SIGBUS, T_PROTFLT);
496 bcopy(&ucp.uc_mcontext.mc_gs, regs, sizeof(struct trapframe));
500 * Restore the FPU state from the frame
503 npxpop(&ucp.uc_mcontext);
506 * Merge saved signal mailbox pending flag to maintain interlock
507 * semantics against system calls.
509 if (ucp.uc_mcontext.mc_xflags & PGEX_MAILBOX) {
510 lwkt_gettoken(&p->p_token);
511 p->p_flag |= P_MAILBOX;
512 lwkt_reltoken(&p->p_token);
515 if (ucp.uc_mcontext.mc_onstack & 1)
516 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
518 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
520 lp->lwp_sigmask = ucp.uc_sigmask;
521 SIG_CANTMASK(lp->lwp_sigmask);
527 * Stack frame on entry to function. %eax will contain the function vector,
528 * %ecx will contain the function data. flags, ecx, and eax will have
529 * already been pushed on the stack.
540 sendupcall(struct vmupcall *vu, int morepending)
542 struct lwp *lp = curthread->td_lwp;
543 struct trapframe *regs;
544 struct upcall upcall;
545 struct upc_frame upc_frame;
549 * If we are a virtual kernel running an emulated user process
550 * context, switch back to the virtual kernel context before
551 * trying to post the signal.
553 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
554 lp->lwp_md.md_regs->tf_trapno = 0;
555 vkernel_trap(lp, lp->lwp_md.md_regs);
559 * Get the upcall data structure
561 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
562 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
565 kprintf("bad upcall address\n");
570 * If the data structure is already marked pending or has a critical
571 * section count, mark the data structure as pending and return
572 * without doing an upcall. vu_pending is left set.
574 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
575 if (upcall.upc_pending < vu->vu_pending) {
576 upcall.upc_pending = vu->vu_pending;
577 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
578 sizeof(upcall.upc_pending));
584 * We can run this upcall now, clear vu_pending.
586 * Bump our critical section count and set or clear the
587 * user pending flag depending on whether more upcalls are
588 * pending. The user will be responsible for calling
589 * upc_dispatch(-1) to process remaining upcalls.
592 upcall.upc_pending = morepending;
594 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
595 sizeof(upcall.upc_pending));
596 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
600 * Construct a stack frame and issue the upcall
602 regs = lp->lwp_md.md_regs;
603 upc_frame.eax = regs->tf_eax;
604 upc_frame.ecx = regs->tf_ecx;
605 upc_frame.edx = regs->tf_edx;
606 upc_frame.flags = regs->tf_eflags;
607 upc_frame.oldip = regs->tf_eip;
608 if (copyout(&upc_frame, (void *)(regs->tf_esp - sizeof(upc_frame)),
609 sizeof(upc_frame)) != 0) {
610 kprintf("bad stack on upcall\n");
612 regs->tf_eax = (register_t)vu->vu_func;
613 regs->tf_ecx = (register_t)vu->vu_data;
614 regs->tf_edx = (register_t)lp->lwp_upcall;
615 regs->tf_eip = (register_t)vu->vu_ctx;
616 regs->tf_esp -= sizeof(upc_frame);
621 * fetchupcall occurs in the context of a system call, which means that
622 * we have to return EJUSTRETURN in order to prevent eax and edx from
623 * being overwritten by the syscall return value.
625 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
626 * and the function pointer in %eax.
629 fetchupcall (struct vmupcall *vu, int morepending, void *rsp)
631 struct upc_frame upc_frame;
632 struct lwp *lp = curthread->td_lwp;
633 struct trapframe *regs;
635 struct upcall upcall;
638 regs = lp->lwp_md.md_regs;
640 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
644 * This jumps us to the next ready context.
647 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
650 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
653 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
654 regs->tf_eax = (register_t)vu->vu_func;
655 regs->tf_ecx = (register_t)vu->vu_data;
656 regs->tf_edx = (register_t)lp->lwp_upcall;
657 regs->tf_eip = (register_t)vu->vu_ctx;
658 regs->tf_esp = (register_t)rsp;
661 * This returns us to the originally interrupted code.
663 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
664 regs->tf_eax = upc_frame.eax;
665 regs->tf_ecx = upc_frame.ecx;
666 regs->tf_edx = upc_frame.edx;
667 regs->tf_eflags = (regs->tf_eflags & ~PSL_USERCHANGE) |
668 (upc_frame.flags & PSL_USERCHANGE);
669 regs->tf_eip = upc_frame.oldip;
670 regs->tf_esp = (register_t)((char *)rsp + sizeof(upc_frame));
679 * cpu_idle() represents the idle LWKT. You cannot return from this function
680 * (unless you want to blow things up!). Instead we look for runnable threads
681 * and loop or halt as appropriate. Giant is not held on entry to the thread.
683 * The main loop is entered with a critical section held, we must release
684 * the critical section before doing anything else. lwkt_switch() will
685 * check for pending interrupts due to entering and exiting its own
688 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
689 * to wake a HLTed cpu up.
691 static int cpu_idle_hlt = 1;
692 static int cpu_idle_hltcnt;
693 static int cpu_idle_spincnt;
694 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
695 &cpu_idle_hlt, 0, "Idle loop HLT enable");
696 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
697 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
698 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
699 &cpu_idle_spincnt, 0, "Idle loop entry spins");
704 struct thread *td = curthread;
705 struct mdglobaldata *gd = mdcpu;
709 KKASSERT(td->td_critcount == 0);
713 * See if there are any LWKTs ready to go.
718 * The idle loop halts only if no threads are scheduleable
719 * and no signals have occured.
722 (td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
725 KKASSERT(MP_LOCK_HELD() == 0);
727 if ((td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
729 struct timeval tv1, tv2;
730 gettimeofday(&tv1, NULL);
732 reqflags = gd->mi.gd_reqflags &
733 ~RQF_IDLECHECK_WK_MASK;
734 umtx_sleep(&gd->mi.gd_reqflags, reqflags,
737 gettimeofday(&tv2, NULL);
738 if (tv2.tv_usec - tv1.tv_usec +
739 (tv2.tv_sec - tv1.tv_sec) * 1000000
741 kprintf("cpu %d idlelock %08x %08x\n",
752 __asm __volatile("pause");
762 * Called by the spinlock code with or without a critical section held
763 * when a spinlock is found to be seriously constested.
765 * We need to enter a critical section to prevent signals from recursing
769 cpu_spinlock_contested(void)
777 * Clear registers on exec
780 exec_setregs(u_long entry, u_long stack, u_long ps_strings)
782 struct thread *td = curthread;
783 struct lwp *lp = td->td_lwp;
784 struct trapframe *regs = lp->lwp_md.md_regs;
785 struct pcb *pcb = lp->lwp_thread->td_pcb;
787 /* was i386_user_cleanup() in NetBSD */
790 bzero((char *)regs, sizeof(struct trapframe));
791 regs->tf_eip = entry;
792 regs->tf_esp = stack;
793 regs->tf_eflags = PSL_USER | (regs->tf_eflags & PSL_T);
801 /* PS_STRINGS value for BSD/OS binaries. It is 0 for non-BSD/OS. */
802 regs->tf_ebx = ps_strings;
805 * Reset the hardware debug registers if they were in use.
806 * They won't have any meaning for the newly exec'd process.
808 if (pcb->pcb_flags & PCB_DBREGS) {
815 if (pcb == td->td_pcb) {
817 * Clear the debug registers on the running
818 * CPU, otherwise they will end up affecting
819 * the next process we switch to.
823 pcb->pcb_flags &= ~PCB_DBREGS;
827 * Initialize the math emulator (if any) for the current process.
828 * Actually, just clear the bit that says that the emulator has
829 * been initialized. Initialization is delayed until the process
830 * traps to the emulator (if it is done at all) mainly because
831 * emulators don't provide an entry point for initialization.
833 pcb->pcb_flags &= ~FP_SOFTFP;
836 * note: do not set CR0_TS here. npxinit() must do it after clearing
837 * gd_npxthread. Otherwise a preemptive interrupt thread may panic
842 load_cr0(rcr0() | CR0_MP);
846 /* Initialize the npx (if any) for the current process. */
847 npxinit(__INITIAL_NPXCW__);
852 * note: linux emulator needs edx to be 0x0 on entry, which is
853 * handled in execve simply by setting the 64 bit syscall
865 cr0 |= CR0_NE; /* Done by npxinit() */
866 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
867 cr0 |= CR0_WP | CR0_AM;
874 sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
877 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
879 if (!error && req->newptr)
884 SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
885 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
887 extern u_long bootdev; /* not a cdev_t - encoding is different */
888 SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
889 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
892 * Initialize 386 and configure to run kernel
896 * Initialize segments & interrupt table
899 extern struct user *proc0paddr;
904 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
905 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
906 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
907 IDTVEC(page), IDTVEC(mchk), IDTVEC(fpu), IDTVEC(align),
908 IDTVEC(xmm), IDTVEC(syscall),
911 IDTVEC(int0x80_syscall);
915 #ifdef DEBUG_INTERRUPTS
916 extern inthand_t *Xrsvdary[256];
920 ptrace_set_pc(struct lwp *lp, unsigned long addr)
922 lp->lwp_md.md_regs->tf_eip = addr;
927 ptrace_single_step(struct lwp *lp)
929 lp->lwp_md.md_regs->tf_eflags |= PSL_T;
934 fill_regs(struct lwp *lp, struct reg *regs)
936 struct trapframe *tp;
938 tp = lp->lwp_md.md_regs;
939 regs->r_gs = tp->tf_gs;
940 regs->r_fs = tp->tf_fs;
941 regs->r_es = tp->tf_es;
942 regs->r_ds = tp->tf_ds;
943 regs->r_edi = tp->tf_edi;
944 regs->r_esi = tp->tf_esi;
945 regs->r_ebp = tp->tf_ebp;
946 regs->r_ebx = tp->tf_ebx;
947 regs->r_edx = tp->tf_edx;
948 regs->r_ecx = tp->tf_ecx;
949 regs->r_eax = tp->tf_eax;
950 regs->r_eip = tp->tf_eip;
951 regs->r_cs = tp->tf_cs;
952 regs->r_eflags = tp->tf_eflags;
953 regs->r_esp = tp->tf_esp;
954 regs->r_ss = tp->tf_ss;
959 set_regs(struct lwp *lp, struct reg *regs)
961 struct trapframe *tp;
963 tp = lp->lwp_md.md_regs;
964 if (!EFL_SECURE(regs->r_eflags, tp->tf_eflags) ||
965 !CS_SECURE(regs->r_cs))
967 tp->tf_gs = regs->r_gs;
968 tp->tf_fs = regs->r_fs;
969 tp->tf_es = regs->r_es;
970 tp->tf_ds = regs->r_ds;
971 tp->tf_edi = regs->r_edi;
972 tp->tf_esi = regs->r_esi;
973 tp->tf_ebp = regs->r_ebp;
974 tp->tf_ebx = regs->r_ebx;
975 tp->tf_edx = regs->r_edx;
976 tp->tf_ecx = regs->r_ecx;
977 tp->tf_eax = regs->r_eax;
978 tp->tf_eip = regs->r_eip;
979 tp->tf_cs = regs->r_cs;
980 tp->tf_eflags = regs->r_eflags;
981 tp->tf_esp = regs->r_esp;
982 tp->tf_ss = regs->r_ss;
986 #ifndef CPU_DISABLE_SSE
988 fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
990 struct env87 *penv_87 = &sv_87->sv_env;
991 struct envxmm *penv_xmm = &sv_xmm->sv_env;
994 /* FPU control/status */
995 penv_87->en_cw = penv_xmm->en_cw;
996 penv_87->en_sw = penv_xmm->en_sw;
997 penv_87->en_tw = penv_xmm->en_tw;
998 penv_87->en_fip = penv_xmm->en_fip;
999 penv_87->en_fcs = penv_xmm->en_fcs;
1000 penv_87->en_opcode = penv_xmm->en_opcode;
1001 penv_87->en_foo = penv_xmm->en_foo;
1002 penv_87->en_fos = penv_xmm->en_fos;
1005 for (i = 0; i < 8; ++i)
1006 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
1010 set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
1012 struct env87 *penv_87 = &sv_87->sv_env;
1013 struct envxmm *penv_xmm = &sv_xmm->sv_env;
1016 /* FPU control/status */
1017 penv_xmm->en_cw = penv_87->en_cw;
1018 penv_xmm->en_sw = penv_87->en_sw;
1019 penv_xmm->en_tw = penv_87->en_tw;
1020 penv_xmm->en_fip = penv_87->en_fip;
1021 penv_xmm->en_fcs = penv_87->en_fcs;
1022 penv_xmm->en_opcode = penv_87->en_opcode;
1023 penv_xmm->en_foo = penv_87->en_foo;
1024 penv_xmm->en_fos = penv_87->en_fos;
1027 for (i = 0; i < 8; ++i)
1028 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
1030 #endif /* CPU_DISABLE_SSE */
1033 fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
1035 #ifndef CPU_DISABLE_SSE
1037 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
1038 (struct save87 *)fpregs);
1041 #endif /* CPU_DISABLE_SSE */
1042 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
1047 set_fpregs(struct lwp *lp, struct fpreg *fpregs)
1049 #ifndef CPU_DISABLE_SSE
1051 set_fpregs_xmm((struct save87 *)fpregs,
1052 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
1055 #endif /* CPU_DISABLE_SSE */
1056 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
1061 fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
1067 set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1074 * Return > 0 if a hardware breakpoint has been hit, and the
1075 * breakpoint was in user space. Return 0, otherwise.
1078 user_dbreg_trap(void)
1080 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1081 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1082 int nbp; /* number of breakpoints that triggered */
1083 caddr_t addr[4]; /* breakpoint addresses */
1087 if ((dr7 & 0x000000ff) == 0) {
1089 * all GE and LE bits in the dr7 register are zero,
1090 * thus the trap couldn't have been caused by the
1091 * hardware debug registers
1098 bp = dr6 & 0x0000000f;
1102 * None of the breakpoint bits are set meaning this
1103 * trap was not caused by any of the debug registers
1109 * at least one of the breakpoints were hit, check to see
1110 * which ones and if any of them are user space addresses
1114 addr[nbp++] = (caddr_t)rdr0();
1117 addr[nbp++] = (caddr_t)rdr1();
1120 addr[nbp++] = (caddr_t)rdr2();
1123 addr[nbp++] = (caddr_t)rdr3();
1126 for (i=0; i<nbp; i++) {
1128 (caddr_t)VM_MAX_USER_ADDRESS) {
1130 * addr[i] is in user space
1137 * None of the breakpoints are in user space.
1150 cpu_feature = regs[3];
1156 Debugger(const char *msg)
1158 kprintf("Debugger(\"%s\") called.\n", msg);