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 $
40 * $DragonFly: src/sys/platform/pc64/amd64/Attic/cpu_regs.c,v 1.6 2008/04/21 15:47:56 dillon Exp $
43 #include "use_ether.h"
46 #include "opt_atalk.h"
47 #include "opt_compat.h"
49 #include "opt_directio.h"
52 #include "opt_msgbuf.h"
55 #include <sys/param.h>
56 #include <sys/systm.h>
57 #include <sys/sysproto.h>
58 #include <sys/signalvar.h>
59 #include <sys/kernel.h>
60 #include <sys/linker.h>
61 #include <sys/malloc.h>
64 #include <sys/reboot.h>
66 #include <sys/msgbuf.h>
67 #include <sys/sysent.h>
68 #include <sys/sysctl.h>
69 #include <sys/vmmeter.h>
71 #include <sys/upcall.h>
72 #include <sys/usched.h>
76 #include <vm/vm_param.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_pager.h>
83 #include <vm/vm_extern.h>
85 #include <sys/thread2.h>
93 #include <machine/cpu.h>
94 #include <machine/clock.h>
95 #include <machine/specialreg.h>
96 #include <machine/md_var.h>
97 #include <machine/pcb_ext.h> /* pcb.h included via sys/user.h */
98 #include <machine/globaldata.h> /* CPU_prvspace */
99 #include <machine/smp.h>
101 #include <machine/perfmon.h>
103 #include <machine/cputypes.h>
105 #include <bus/isa/rtc.h>
106 /* #include <machine/vm86.h> */
107 #include <sys/random.h>
108 #include <sys/ptrace.h>
109 #include <machine/sigframe.h>
110 #include <unistd.h> /* umtx_* functions */
112 extern void dblfault_handler (void);
114 #ifndef CPU_DISABLE_SSE
115 static void set_fpregs_xmm (struct save87 *, struct savexmm *);
116 static void fill_fpregs_xmm (struct savexmm *, struct save87 *);
117 #endif /* CPU_DISABLE_SSE */
119 extern void ffs_rawread_setup(void);
120 #endif /* DIRECTIO */
123 int64_t tsc_offsets[MAXCPU];
125 int64_t tsc_offsets[1];
128 #if defined(SWTCH_OPTIM_STATS)
129 extern int swtch_optim_stats;
130 SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats,
131 CTLFLAG_RD, &swtch_optim_stats, 0, "");
132 SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count,
133 CTLFLAG_RD, &tlb_flush_count, 0, "");
137 sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
139 int error = sysctl_handle_int(oidp, 0, ctob((int)Maxmem), req);
143 SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_INT|CTLFLAG_RD,
144 0, 0, sysctl_hw_physmem, "IU", "");
147 sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
149 int error = sysctl_handle_int(oidp, 0,
150 ctob((int)Maxmem - vmstats.v_wire_count), req);
154 SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD,
155 0, 0, sysctl_hw_usermem, "IU", "");
157 SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, NULL, "");
162 sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
166 /* Unwind the buffer, so that it's linear (possibly starting with
167 * some initial nulls).
169 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr,
170 msgbufp->msg_size-msgbufp->msg_bufr,req);
171 if(error) return(error);
172 if(msgbufp->msg_bufr>0) {
173 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr,
174 msgbufp->msg_bufr,req);
179 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
180 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
182 static int msgbuf_clear;
185 sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
188 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
190 if (!error && req->newptr) {
191 /* Clear the buffer and reset write pointer */
192 bzero(msgbufp->msg_ptr,msgbufp->msg_size);
193 msgbufp->msg_bufr=msgbufp->msg_bufx=0;
199 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
200 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
201 "Clear kernel message buffer");
206 * Send an interrupt to process.
208 * Stack is set up to allow sigcode stored
209 * at top to call routine, followed by kcall
210 * to sigreturn routine below. After sigreturn
211 * resets the signal mask, the stack, and the
212 * frame pointer, it returns to the user
216 extern int _ucodesel, _udatasel;
218 sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
220 struct lwp *lp = curthread->td_lwp;
221 struct proc *p = lp->lwp_proc;
222 struct trapframe *regs;
223 struct sigacts *psp = p->p_sigacts;
224 struct sigframe sf, *sfp;
227 regs = lp->lwp_md.md_regs;
228 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
230 /* save user context */
231 bzero(&sf, sizeof(struct sigframe));
232 sf.sf_uc.uc_sigmask = *mask;
233 sf.sf_uc.uc_stack = lp->lwp_sigstk;
234 sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
235 /* bcopy(regs, &sf.sf_uc.uc_mcontext.mc_gs, sizeof(struct trapframe)); */
237 /* make the size of the saved context visible to userland */
238 sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext);
240 /* save mailbox pending state for syscall interlock semantics */
241 if (p->p_flag & P_MAILBOX)
242 sf.sf_uc.uc_mcontext.mc_flags |= PGEX_MAILBOX;
245 /* Allocate and validate space for the signal handler context. */
246 if ((lp->lwp_flag & LWP_ALTSTACK) != 0 && !oonstack &&
247 SIGISMEMBER(psp->ps_sigonstack, sig)) {
248 sfp = (struct sigframe *)(lp->lwp_sigstk.ss_sp +
249 lp->lwp_sigstk.ss_size - sizeof(struct sigframe));
250 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
253 sfp = (struct sigframe *)regs->tf_rsp - 1;
255 /* Translate the signal is appropriate */
256 if (p->p_sysent->sv_sigtbl) {
257 if (sig <= p->p_sysent->sv_sigsize)
258 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
261 /* Build the argument list for the signal handler. */
263 sf.sf_ucontext = (register_t)&sfp->sf_uc;
264 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
265 /* Signal handler installed with SA_SIGINFO. */
266 sf.sf_siginfo = (register_t)&sfp->sf_si;
267 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
269 /* fill siginfo structure */
270 sf.sf_si.si_signo = sig;
271 sf.sf_si.si_code = code;
272 sf.sf_si.si_addr = (void*)regs->tf_err;
275 /* Old FreeBSD-style arguments. */
276 sf.sf_siginfo = code;
277 sf.sf_addr = regs->tf_err;
278 sf.sf_ahu.sf_handler = catcher;
283 * If we're a vm86 process, we want to save the segment registers.
284 * We also change eflags to be our emulated eflags, not the actual
287 if (regs->tf_rflags & PSL_VM) {
288 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
289 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
291 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
292 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
293 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
294 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
296 if (vm86->vm86_has_vme == 0)
297 sf.sf_uc.uc_mcontext.mc_eflags =
298 (tf->tf_rflags & ~(PSL_VIF | PSL_VIP)) |
299 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
302 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
303 * syscalls made by the signal handler. This just avoids
304 * wasting time for our lazy fixup of such faults. PSL_NT
305 * does nothing in vm86 mode, but vm86 programs can set it
306 * almost legitimately in probes for old cpu types.
308 tf->tf_rflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
313 * Save the FPU state and reinit the FP unit
315 npxpush(&sf.sf_uc.uc_mcontext);
318 * Copy the sigframe out to the user's stack.
320 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
322 * Something is wrong with the stack pointer.
323 * ...Kill the process.
328 regs->tf_rsp = (int)sfp;
329 regs->tf_rip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
332 * amd64 abi specifies that the direction flag must be cleared
335 regs->tf_rflags &= ~(PSL_T|PSL_D);
337 regs->tf_cs = _ucodesel;
338 /* regs->tf_ds = _udatasel;
339 regs->tf_es = _udatasel; */
340 if (regs->tf_trapno == T_PROTFLT) {
341 /* regs->tf_fs = _udatasel;
342 regs->tf_gs = _udatasel; */
344 regs->tf_ss = _udatasel;
348 * Sanitize the trapframe for a virtual kernel passing control to a custom
351 * Allow userland to set or maintain PSL_RF, the resume flag. This flag
352 * basically controls whether the return PC should skip the first instruction
353 * (as in an explicit system call) or re-execute it (as in an exception).
356 cpu_sanitize_frame(struct trapframe *frame)
358 frame->tf_cs = _ucodesel;
360 frame->tf_ds = _udatasel;
361 frame->tf_es = _udatasel;
362 frame->tf_fs = _udatasel;
363 frame->tf_gs = _udatasel;
365 frame->tf_ss = _udatasel;
366 frame->tf_rflags &= (PSL_RF | PSL_USERCHANGE);
367 frame->tf_rflags |= PSL_RESERVED_DEFAULT | PSL_I;
372 cpu_sanitize_tls(struct savetls *tls)
374 struct segment_descriptor *desc;
377 for (i = 0; i < NGTLS; ++i) {
379 if (desc->sd_dpl == 0 && desc->sd_type == 0)
381 if (desc->sd_def32 == 0)
383 if (desc->sd_type != SDT_MEMRWA)
385 if (desc->sd_dpl != SEL_UPL)
387 if (desc->sd_xx != 0 || desc->sd_p != 1)
394 * sigreturn(ucontext_t *sigcntxp)
396 * System call to cleanup state after a signal
397 * has been taken. Reset signal mask and
398 * stack state from context left by sendsig (above).
399 * Return to previous pc and psl as specified by
400 * context left by sendsig. Check carefully to
401 * make sure that the user has not modified the
402 * state to gain improper privileges.
404 #define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
405 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
408 sys_sigreturn(struct sigreturn_args *uap)
410 struct lwp *lp = curthread->td_lwp;
411 struct proc *p = lp->lwp_proc;
412 struct trapframe *regs;
418 error = copyin(uap->sigcntxp, &ucp, sizeof(ucp));
422 regs = lp->lwp_md.md_regs;
423 rflags = ucp.uc_mcontext.mc_rflags;
426 if (eflags & PSL_VM) {
427 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
428 struct vm86_kernel *vm86;
431 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
432 * set up the vm86 area, and we can't enter vm86 mode.
434 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
436 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
437 if (vm86->vm86_inited == 0)
440 /* go back to user mode if both flags are set */
441 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
442 trapsignal(lp->lwp_proc, SIGBUS, 0);
444 if (vm86->vm86_has_vme) {
445 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
446 (eflags & VME_USERCHANGE) | PSL_VM;
448 vm86->vm86_eflags = eflags; /* save VIF, VIP */
449 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
451 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
452 tf->tf_eflags = eflags;
453 tf->tf_vm86_ds = tf->tf_ds;
454 tf->tf_vm86_es = tf->tf_es;
455 tf->tf_vm86_fs = tf->tf_fs;
456 tf->tf_vm86_gs = tf->tf_gs;
457 tf->tf_ds = _udatasel;
458 tf->tf_es = _udatasel;
460 tf->tf_fs = _udatasel;
461 tf->tf_gs = _udatasel;
467 * Don't allow users to change privileged or reserved flags.
470 * XXX do allow users to change the privileged flag PSL_RF.
471 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
472 * should sometimes set it there too. tf_eflags is kept in
473 * the signal context during signal handling and there is no
474 * other place to remember it, so the PSL_RF bit may be
475 * corrupted by the signal handler without us knowing.
476 * Corruption of the PSL_RF bit at worst causes one more or
477 * one less debugger trap, so allowing it is fairly harmless.
479 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
480 kprintf("sigreturn: eflags = 0x%x\n", rflags);
485 * Don't allow users to load a valid privileged %cs. Let the
486 * hardware check for invalid selectors, excess privilege in
487 * other selectors, invalid %eip's and invalid %esp's.
489 cs = ucp.uc_mcontext.mc_cs;
490 if (!CS_SECURE(cs)) {
491 kprintf("sigreturn: cs = 0x%x\n", cs);
492 trapsignal(lp, SIGBUS, T_PROTFLT);
495 /* bcopy(&ucp.uc_mcontext.mc_gs, regs, sizeof(struct trapframe)); */
499 * Restore the FPU state from the frame
501 npxpop(&ucp.uc_mcontext);
504 * Merge saved signal mailbox pending flag to maintain interlock
505 * semantics against system calls.
507 if (ucp.uc_mcontext.mc_flags & PGEX_MAILBOX)
508 p->p_flag |= P_MAILBOX;
510 if (ucp.uc_mcontext.mc_onstack & 1)
511 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
513 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
515 lp->lwp_sigmask = ucp.uc_sigmask;
516 SIG_CANTMASK(lp->lwp_sigmask);
521 * Stack frame on entry to function. %eax will contain the function vector,
522 * %ecx will contain the function data. flags, ecx, and eax will have
523 * already been pushed on the stack.
534 sendupcall(struct vmupcall *vu, int morepending)
536 struct lwp *lp = curthread->td_lwp;
537 struct trapframe *regs;
538 struct upcall upcall;
539 struct upc_frame upc_frame;
543 * If we are a virtual kernel running an emulated user process
544 * context, switch back to the virtual kernel context before
545 * trying to post the signal.
547 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
548 lp->lwp_md.md_regs->tf_trapno = 0;
549 vkernel_trap(lp, lp->lwp_md.md_regs);
553 * Get the upcall data structure
555 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
556 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
559 kprintf("bad upcall address\n");
564 * If the data structure is already marked pending or has a critical
565 * section count, mark the data structure as pending and return
566 * without doing an upcall. vu_pending is left set.
568 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
569 if (upcall.upc_pending < vu->vu_pending) {
570 upcall.upc_pending = vu->vu_pending;
571 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
572 sizeof(upcall.upc_pending));
578 * We can run this upcall now, clear vu_pending.
580 * Bump our critical section count and set or clear the
581 * user pending flag depending on whether more upcalls are
582 * pending. The user will be responsible for calling
583 * upc_dispatch(-1) to process remaining upcalls.
586 upcall.upc_pending = morepending;
587 crit_count += TDPRI_CRIT;
588 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
589 sizeof(upcall.upc_pending));
590 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
594 * Construct a stack frame and issue the upcall
596 regs = lp->lwp_md.md_regs;
597 upc_frame.eax = regs->tf_rax;
598 upc_frame.ecx = regs->tf_rcx;
599 upc_frame.edx = regs->tf_rdx;
600 upc_frame.flags = regs->tf_rflags;
601 upc_frame.oldip = regs->tf_rip;
602 if (copyout(&upc_frame, (void *)(regs->tf_rsp - sizeof(upc_frame)),
603 sizeof(upc_frame)) != 0) {
604 kprintf("bad stack on upcall\n");
606 regs->tf_rax = (register_t)vu->vu_func;
607 regs->tf_rcx = (register_t)vu->vu_data;
608 regs->tf_rdx = (register_t)lp->lwp_upcall;
609 regs->tf_rip = (register_t)vu->vu_ctx;
610 regs->tf_rsp -= sizeof(upc_frame);
615 * fetchupcall occurs in the context of a system call, which means that
616 * we have to return EJUSTRETURN in order to prevent eax and edx from
617 * being overwritten by the syscall return value.
619 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
620 * and the function pointer in %eax.
623 fetchupcall (struct vmupcall *vu, int morepending, void *rsp)
625 struct upc_frame upc_frame;
626 struct lwp *lp = curthread->td_lwp;
627 struct trapframe *regs;
629 struct upcall upcall;
632 regs = lp->lwp_md.md_regs;
634 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
638 * This jumps us to the next ready context.
641 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
644 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
645 crit_count += TDPRI_CRIT;
647 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
648 regs->tf_rax = (register_t)vu->vu_func;
649 regs->tf_rcx = (register_t)vu->vu_data;
650 regs->tf_rdx = (register_t)lp->lwp_upcall;
651 regs->tf_rip = (register_t)vu->vu_ctx;
652 regs->tf_rsp = (register_t)rsp;
655 * This returns us to the originally interrupted code.
657 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
658 regs->tf_rax = upc_frame.eax;
659 regs->tf_rcx = upc_frame.ecx;
660 regs->tf_rdx = upc_frame.edx;
661 regs->tf_rflags = (regs->tf_rflags & ~PSL_USERCHANGE) |
662 (upc_frame.flags & PSL_USERCHANGE);
663 regs->tf_rip = upc_frame.oldip;
664 regs->tf_rsp = (register_t)((char *)rsp + sizeof(upc_frame));
673 * cpu_idle() represents the idle LWKT. You cannot return from this function
674 * (unless you want to blow things up!). Instead we look for runnable threads
675 * and loop or halt as appropriate. Giant is not held on entry to the thread.
677 * The main loop is entered with a critical section held, we must release
678 * the critical section before doing anything else. lwkt_switch() will
679 * check for pending interrupts due to entering and exiting its own
682 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
683 * to wake a HLTed cpu up. However, there are cases where the idlethread
684 * will be entered with the possibility that no IPI will occur and in such
685 * cases lwkt_switch() sets TDF_IDLE_NOHLT.
687 static int cpu_idle_hlt = 1;
688 static int cpu_idle_hltcnt;
689 static int cpu_idle_spincnt;
690 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
691 &cpu_idle_hlt, 0, "Idle loop HLT enable");
692 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
693 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
694 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
695 &cpu_idle_spincnt, 0, "Idle loop entry spins");
700 struct thread *td = curthread;
701 struct mdglobaldata *gd = mdcpu;
704 KKASSERT(td->td_pri < TDPRI_CRIT);
707 * See if there are any LWKTs ready to go.
712 * The idle loop halts only if no threads are scheduleable
713 * and no signals have occured.
715 if (cpu_idle_hlt && !lwkt_runnable() &&
716 (td->td_flags & TDF_IDLE_NOHLT) == 0) {
718 if (!lwkt_runnable()) {
720 struct timeval tv1, tv2;
721 gettimeofday(&tv1, NULL);
723 /* umtx_sleep(&gd->mi.gd_runqmask, 0, 1000000); */
725 gettimeofday(&tv2, NULL);
726 if (tv2.tv_usec - tv1.tv_usec +
727 (tv2.tv_sec - tv1.tv_sec) * 1000000
729 kprintf("cpu %d idlelock %08x %08x\n",
738 __asm __volatile("pause");
743 td->td_flags &= ~TDF_IDLE_NOHLT;
746 /*__asm __volatile("sti; pause");*/
747 __asm __volatile("pause");
749 /*__asm __volatile("sti");*/
759 * Called by the LWKT switch core with a critical section held if the only
760 * schedulable thread needs the MP lock and we couldn't get it. On
761 * a real cpu we just spin in the scheduler. In the virtual kernel
762 * we sleep for a bit.
765 cpu_mplock_contested(void)
771 * Called by the spinlock code with or without a critical section held
772 * when a spinlock is found to be seriously constested.
775 cpu_spinlock_contested(void)
783 * Clear registers on exec
786 exec_setregs(u_long entry, u_long stack, u_long ps_strings)
788 struct thread *td = curthread;
789 struct lwp *lp = td->td_lwp;
790 struct trapframe *regs = lp->lwp_md.md_regs;
791 struct pcb *pcb = lp->lwp_thread->td_pcb;
793 /* was i386_user_cleanup() in NetBSD */
796 bzero((char *)regs, sizeof(struct trapframe));
797 regs->tf_rip = entry;
798 regs->tf_rsp = stack;
799 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
807 /* PS_STRINGS value for BSD/OS binaries. It is 0 for non-BSD/OS. */
808 regs->tf_rbx = ps_strings;
811 * Reset the hardware debug registers if they were in use.
812 * They won't have any meaning for the newly exec'd process.
814 if (pcb->pcb_flags & PCB_DBREGS) {
821 if (pcb == td->td_pcb) {
823 * Clear the debug registers on the running
824 * CPU, otherwise they will end up affecting
825 * the next process we switch to.
829 pcb->pcb_flags &= ~PCB_DBREGS;
833 * Initialize the math emulator (if any) for the current process.
834 * Actually, just clear the bit that says that the emulator has
835 * been initialized. Initialization is delayed until the process
836 * traps to the emulator (if it is done at all) mainly because
837 * emulators don't provide an entry point for initialization.
839 /* pcb->pcb_flags &= ~FP_SOFTFP; */
842 * note: do not set CR0_TS here. npxinit() must do it after clearing
843 * gd_npxthread. Otherwise a preemptive interrupt thread may panic
848 load_cr0(rcr0() | CR0_MP);
852 /* Initialize the npx (if any) for the current process. */
853 npxinit(__INITIAL_NPXCW__);
858 * note: linux emulator needs edx to be 0x0 on entry, which is
859 * handled in execve simply by setting the 64 bit syscall
871 cr0 |= CR0_NE; /* Done by npxinit() */
872 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
873 cr0 |= CR0_WP | CR0_AM;
880 sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
883 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
885 if (!error && req->newptr)
890 SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
891 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
893 extern u_long bootdev; /* not a cdev_t - encoding is different */
894 SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
895 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
898 * Initialize 386 and configure to run kernel
902 * Initialize segments & interrupt table
905 extern struct user *proc0paddr;
910 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
911 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
912 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
913 IDTVEC(page), IDTVEC(mchk), IDTVEC(fpu), IDTVEC(align),
914 IDTVEC(xmm), IDTVEC(syscall),
917 IDTVEC(int0x80_syscall);
921 #ifdef DEBUG_INTERRUPTS
922 extern inthand_t *Xrsvdary[256];
926 ptrace_set_pc(struct lwp *lp, unsigned long addr)
928 lp->lwp_md.md_regs->tf_rip = addr;
933 ptrace_single_step(struct lwp *lp)
935 lp->lwp_md.md_regs->tf_rflags |= PSL_T;
940 fill_regs(struct lwp *lp, struct reg *regs)
942 struct trapframe *tp;
944 tp = lp->lwp_md.md_regs;
945 /* regs->r_gs = tp->tf_gs;
946 regs->r_fs = tp->tf_fs;
947 regs->r_es = tp->tf_es;
948 regs->r_ds = tp->tf_ds; */
949 regs->r_rdi = tp->tf_rdi;
950 regs->r_rsi = tp->tf_rsi;
951 regs->r_rbp = tp->tf_rbp;
952 regs->r_rbx = tp->tf_rbx;
953 regs->r_rdx = tp->tf_rdx;
954 regs->r_rcx = tp->tf_rcx;
955 regs->r_rax = tp->tf_rax;
956 regs->r_rip = tp->tf_rip;
957 regs->r_cs = tp->tf_cs;
958 regs->r_rflags = tp->tf_rflags;
959 regs->r_rsp = tp->tf_rsp;
960 regs->r_ss = tp->tf_ss;
965 set_regs(struct lwp *lp, struct reg *regs)
967 struct trapframe *tp;
969 tp = lp->lwp_md.md_regs;
970 if (!EFL_SECURE(regs->r_rflags, tp->tf_rflags) ||
971 !CS_SECURE(regs->r_cs))
973 /* tp->tf_gs = regs->r_gs;
974 tp->tf_fs = regs->r_fs;
975 tp->tf_es = regs->r_es;
976 tp->tf_ds = regs->r_ds; */
977 tp->tf_rdi = regs->r_rdi;
978 tp->tf_rsi = regs->r_rsi;
979 tp->tf_rbp = regs->r_rbp;
980 tp->tf_rbx = regs->r_rbx;
981 tp->tf_rdx = regs->r_rdx;
982 tp->tf_rcx = regs->r_rcx;
983 tp->tf_rax = regs->r_rax;
984 tp->tf_rip = regs->r_rip;
985 tp->tf_cs = regs->r_cs;
986 tp->tf_rflags = regs->r_rflags;
987 tp->tf_rsp = regs->r_rsp;
988 tp->tf_ss = regs->r_ss;
992 #ifndef CPU_DISABLE_SSE
994 fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
996 struct env87 *penv_87 = &sv_87->sv_env;
997 struct envxmm *penv_xmm = &sv_xmm->sv_env;
1000 /* FPU control/status */
1001 penv_87->en_cw = penv_xmm->en_cw;
1002 penv_87->en_sw = penv_xmm->en_sw;
1003 penv_87->en_tw = penv_xmm->en_tw;
1004 penv_87->en_fip = penv_xmm->en_fip;
1005 penv_87->en_fcs = penv_xmm->en_fcs;
1006 penv_87->en_opcode = penv_xmm->en_opcode;
1007 penv_87->en_foo = penv_xmm->en_foo;
1008 penv_87->en_fos = penv_xmm->en_fos;
1011 for (i = 0; i < 8; ++i)
1012 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
1014 sv_87->sv_ex_sw = sv_xmm->sv_ex_sw;
1018 set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
1020 struct env87 *penv_87 = &sv_87->sv_env;
1021 struct envxmm *penv_xmm = &sv_xmm->sv_env;
1024 /* FPU control/status */
1025 penv_xmm->en_cw = penv_87->en_cw;
1026 penv_xmm->en_sw = penv_87->en_sw;
1027 penv_xmm->en_tw = penv_87->en_tw;
1028 penv_xmm->en_fip = penv_87->en_fip;
1029 penv_xmm->en_fcs = penv_87->en_fcs;
1030 penv_xmm->en_opcode = penv_87->en_opcode;
1031 penv_xmm->en_foo = penv_87->en_foo;
1032 penv_xmm->en_fos = penv_87->en_fos;
1035 for (i = 0; i < 8; ++i)
1036 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
1038 sv_xmm->sv_ex_sw = sv_87->sv_ex_sw;
1040 #endif /* CPU_DISABLE_SSE */
1043 fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
1045 #ifndef CPU_DISABLE_SSE
1047 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
1048 (struct save87 *)fpregs);
1051 #endif /* CPU_DISABLE_SSE */
1052 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
1057 set_fpregs(struct lwp *lp, struct fpreg *fpregs)
1059 #ifndef CPU_DISABLE_SSE
1061 set_fpregs_xmm((struct save87 *)fpregs,
1062 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
1065 #endif /* CPU_DISABLE_SSE */
1066 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
1071 fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
1074 dbregs->dr[0] = rdr0();
1075 dbregs->dr[1] = rdr1();
1076 dbregs->dr[2] = rdr2();
1077 dbregs->dr[3] = rdr3();
1078 dbregs->dr[4] = rdr4();
1079 dbregs->dr[5] = rdr5();
1080 dbregs->dr[6] = rdr6();
1081 dbregs->dr[7] = rdr7();
1085 pcb = lp->lwp_thread->td_pcb;
1086 dbregs->dr[0] = pcb->pcb_dr0;
1087 dbregs->dr[1] = pcb->pcb_dr1;
1088 dbregs->dr[2] = pcb->pcb_dr2;
1089 dbregs->dr[3] = pcb->pcb_dr3;
1092 dbregs->dr[6] = pcb->pcb_dr6;
1093 dbregs->dr[7] = pcb->pcb_dr7;
1099 set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1102 load_dr0(dbregs->dr[0]);
1103 load_dr1(dbregs->dr[1]);
1104 load_dr2(dbregs->dr[2]);
1105 load_dr3(dbregs->dr[3]);
1106 load_dr4(dbregs->dr[4]);
1107 load_dr5(dbregs->dr[5]);
1108 load_dr6(dbregs->dr[6]);
1109 load_dr7(dbregs->dr[7]);
1112 struct ucred *ucred;
1114 uint32_t mask1, mask2;
1117 * Don't let an illegal value for dr7 get set. Specifically,
1118 * check for undefined settings. Setting these bit patterns
1119 * result in undefined behaviour and can lead to an unexpected
1122 for (i = 0, mask1 = 0x3<<16, mask2 = 0x2<<16; i < 8;
1123 i++, mask1 <<= 2, mask2 <<= 2)
1124 if ((dbregs->dr[7] & mask1) == mask2)
1127 pcb = lp->lwp_thread->td_pcb;
1128 ucred = lp->lwp_proc->p_ucred;
1131 * Don't let a process set a breakpoint that is not within the
1132 * process's address space. If a process could do this, it
1133 * could halt the system by setting a breakpoint in the kernel
1134 * (if ddb was enabled). Thus, we need to check to make sure
1135 * that no breakpoints are being enabled for addresses outside
1136 * process's address space, unless, perhaps, we were called by
1139 * XXX - what about when the watched area of the user's
1140 * address space is written into from within the kernel
1141 * ... wouldn't that still cause a breakpoint to be generated
1142 * from within kernel mode?
1145 if (suser_cred(ucred, 0) != 0) {
1146 if (dbregs->dr[7] & 0x3) {
1147 /* dr0 is enabled */
1148 if (dbregs->dr[0] >= VM_MAX_USER_ADDRESS)
1152 if (dbregs->dr[7] & (0x3<<2)) {
1153 /* dr1 is enabled */
1154 if (dbregs->dr[1] >= VM_MAX_USER_ADDRESS)
1158 if (dbregs->dr[7] & (0x3<<4)) {
1159 /* dr2 is enabled */
1160 if (dbregs->dr[2] >= VM_MAX_USER_ADDRESS)
1164 if (dbregs->dr[7] & (0x3<<6)) {
1165 /* dr3 is enabled */
1166 if (dbregs->dr[3] >= VM_MAX_USER_ADDRESS)
1171 pcb->pcb_dr0 = dbregs->dr[0];
1172 pcb->pcb_dr1 = dbregs->dr[1];
1173 pcb->pcb_dr2 = dbregs->dr[2];
1174 pcb->pcb_dr3 = dbregs->dr[3];
1175 pcb->pcb_dr6 = dbregs->dr[6];
1176 pcb->pcb_dr7 = dbregs->dr[7];
1178 pcb->pcb_flags |= PCB_DBREGS;
1186 * Return > 0 if a hardware breakpoint has been hit, and the
1187 * breakpoint was in user space. Return 0, otherwise.
1190 user_dbreg_trap(void)
1192 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1193 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1194 int nbp; /* number of breakpoints that triggered */
1195 caddr_t addr[4]; /* breakpoint addresses */
1199 if ((dr7 & 0x000000ff) == 0) {
1201 * all GE and LE bits in the dr7 register are zero,
1202 * thus the trap couldn't have been caused by the
1203 * hardware debug registers
1210 bp = dr6 & 0x0000000f;
1214 * None of the breakpoint bits are set meaning this
1215 * trap was not caused by any of the debug registers
1221 * at least one of the breakpoints were hit, check to see
1222 * which ones and if any of them are user space addresses
1226 addr[nbp++] = (caddr_t)rdr0();
1229 addr[nbp++] = (caddr_t)rdr1();
1232 addr[nbp++] = (caddr_t)rdr2();
1235 addr[nbp++] = (caddr_t)rdr3();
1238 for (i=0; i<nbp; i++) {
1240 (caddr_t)VM_MAX_USER_ADDRESS) {
1242 * addr[i] is in user space
1249 * None of the breakpoints are in user space.
1259 Debugger(const char *msg)
1261 kprintf("Debugger(\"%s\") called.\n", msg);