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/vkernel/i386/cpu_regs.c,v 1.14 2007/02/19 01:14:40 corecode 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>
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 int error = sysctl_handle_int(oidp, 0, ctob((int)Maxmem), req);
142 SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_INT|CTLFLAG_RD,
143 0, 0, sysctl_hw_physmem, "IU", "");
146 sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
148 int error = sysctl_handle_int(oidp, 0,
149 ctob((int)Maxmem - vmstats.v_wire_count), req);
153 SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD,
154 0, 0, sysctl_hw_usermem, "IU", "");
156 SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, NULL, "");
161 sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
165 /* Unwind the buffer, so that it's linear (possibly starting with
166 * some initial nulls).
168 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr,
169 msgbufp->msg_size-msgbufp->msg_bufr,req);
170 if(error) return(error);
171 if(msgbufp->msg_bufr>0) {
172 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr,
173 msgbufp->msg_bufr,req);
178 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
179 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
181 static int msgbuf_clear;
184 sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
187 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
189 if (!error && req->newptr) {
190 /* Clear the buffer and reset write pointer */
191 bzero(msgbufp->msg_ptr,msgbufp->msg_size);
192 msgbufp->msg_bufr=msgbufp->msg_bufx=0;
198 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
199 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
200 "Clear kernel message buffer");
205 * Send an interrupt to process.
207 * Stack is set up to allow sigcode stored
208 * at top to call routine, followed by kcall
209 * to sigreturn routine below. After sigreturn
210 * resets the signal mask, the stack, and the
211 * frame pointer, it returns to the user
215 sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
217 struct lwp *lp = curthread->td_lwp;
218 struct proc *p = lp->lwp_proc;
219 struct trapframe *regs;
220 struct sigacts *psp = p->p_sigacts;
221 struct sigframe sf, *sfp;
224 regs = lp->lwp_md.md_regs;
225 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
227 /* save user context */
228 bzero(&sf, sizeof(struct sigframe));
229 sf.sf_uc.uc_sigmask = *mask;
230 sf.sf_uc.uc_stack = lp->lwp_sigstk;
231 sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
232 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_gs, sizeof(struct trapframe));
234 /* make the size of the saved context visible to userland */
235 sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext);
237 /* save mailbox pending state for syscall interlock semantics */
238 if (p->p_flag & P_MAILBOX)
239 sf.sf_uc.uc_mcontext.mc_xflags |= PGEX_MAILBOX;
242 /* Allocate and validate space for the signal handler context. */
243 if ((lp->lwp_flag & LWP_ALTSTACK) != 0 && !oonstack &&
244 SIGISMEMBER(psp->ps_sigonstack, sig)) {
245 sfp = (struct sigframe *)(lp->lwp_sigstk.ss_sp +
246 lp->lwp_sigstk.ss_size - sizeof(struct sigframe));
247 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
250 sfp = (struct sigframe *)regs->tf_esp - 1;
252 /* Translate the signal is appropriate */
253 if (p->p_sysent->sv_sigtbl) {
254 if (sig <= p->p_sysent->sv_sigsize)
255 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
258 /* Build the argument list for the signal handler. */
260 sf.sf_ucontext = (register_t)&sfp->sf_uc;
261 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
262 /* Signal handler installed with SA_SIGINFO. */
263 sf.sf_siginfo = (register_t)&sfp->sf_si;
264 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
266 /* fill siginfo structure */
267 sf.sf_si.si_signo = sig;
268 sf.sf_si.si_code = code;
269 sf.sf_si.si_addr = (void*)regs->tf_err;
272 /* Old FreeBSD-style arguments. */
273 sf.sf_siginfo = code;
274 sf.sf_addr = regs->tf_err;
275 sf.sf_ahu.sf_handler = catcher;
280 * If we're a vm86 process, we want to save the segment registers.
281 * We also change eflags to be our emulated eflags, not the actual
284 if (regs->tf_eflags & PSL_VM) {
285 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
286 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
288 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
289 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
290 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
291 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
293 if (vm86->vm86_has_vme == 0)
294 sf.sf_uc.uc_mcontext.mc_eflags =
295 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
296 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
299 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
300 * syscalls made by the signal handler. This just avoids
301 * wasting time for our lazy fixup of such faults. PSL_NT
302 * does nothing in vm86 mode, but vm86 programs can set it
303 * almost legitimately in probes for old cpu types.
305 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
310 * Copy the sigframe out to the user's stack.
312 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
314 * Something is wrong with the stack pointer.
315 * ...Kill the process.
320 regs->tf_esp = (int)sfp;
321 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
322 regs->tf_eflags &= ~PSL_T;
323 regs->tf_cs = _ucodesel;
324 regs->tf_ds = _udatasel;
325 regs->tf_es = _udatasel;
326 if (regs->tf_trapno == T_PROTFLT) {
327 regs->tf_fs = _udatasel;
328 regs->tf_gs = _udatasel;
330 regs->tf_ss = _udatasel;
334 * Sanitize the trapframe for a virtual kernel passing control to a custom
337 * Allow userland to set or maintain PSL_RF, the resume flag. This flag
338 * basically controls whether the return PC should skip the first instruction
339 * (as in an explicit system call) or re-execute it (as in an exception).
342 cpu_sanitize_frame(struct trapframe *frame)
344 frame->tf_cs = _ucodesel;
345 frame->tf_ds = _udatasel;
346 frame->tf_es = _udatasel;
348 frame->tf_fs = _udatasel;
349 frame->tf_gs = _udatasel;
351 frame->tf_ss = _udatasel;
352 frame->tf_eflags &= (PSL_RF | PSL_USERCHANGE);
353 frame->tf_eflags |= PSL_RESERVED_DEFAULT | PSL_I;
358 cpu_sanitize_tls(struct savetls *tls)
360 struct segment_descriptor *desc;
363 for (i = 0; i < NGTLS; ++i) {
365 if (desc->sd_dpl == 0 && desc->sd_type == 0)
367 if (desc->sd_def32 == 0)
369 if (desc->sd_type != SDT_MEMRWA)
371 if (desc->sd_dpl != SEL_UPL)
373 if (desc->sd_xx != 0 || desc->sd_p != 1)
380 * sigreturn(ucontext_t *sigcntxp)
382 * System call to cleanup state after a signal
383 * has been taken. Reset signal mask and
384 * stack state from context left by sendsig (above).
385 * Return to previous pc and psl as specified by
386 * context left by sendsig. Check carefully to
387 * make sure that the user has not modified the
388 * state to gain improper privileges.
390 #define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
391 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
394 sys_sigreturn(struct sigreturn_args *uap)
396 struct lwp *lp = curthread->td_lwp;
397 struct proc *p = lp->lwp_proc;
398 struct trapframe *regs;
404 error = copyin(uap->sigcntxp, &ucp, sizeof(ucp));
408 regs = lp->lwp_md.md_regs;
409 eflags = ucp.uc_mcontext.mc_eflags;
412 if (eflags & PSL_VM) {
413 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
414 struct vm86_kernel *vm86;
417 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
418 * set up the vm86 area, and we can't enter vm86 mode.
420 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
422 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
423 if (vm86->vm86_inited == 0)
426 /* go back to user mode if both flags are set */
427 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
428 trapsignal(lp->lwp_proc, SIGBUS, 0);
430 if (vm86->vm86_has_vme) {
431 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
432 (eflags & VME_USERCHANGE) | PSL_VM;
434 vm86->vm86_eflags = eflags; /* save VIF, VIP */
435 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
437 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
438 tf->tf_eflags = eflags;
439 tf->tf_vm86_ds = tf->tf_ds;
440 tf->tf_vm86_es = tf->tf_es;
441 tf->tf_vm86_fs = tf->tf_fs;
442 tf->tf_vm86_gs = tf->tf_gs;
443 tf->tf_ds = _udatasel;
444 tf->tf_es = _udatasel;
446 tf->tf_fs = _udatasel;
447 tf->tf_gs = _udatasel;
453 * Don't allow users to change privileged or reserved flags.
456 * XXX do allow users to change the privileged flag PSL_RF.
457 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
458 * should sometimes set it there too. tf_eflags is kept in
459 * the signal context during signal handling and there is no
460 * other place to remember it, so the PSL_RF bit may be
461 * corrupted by the signal handler without us knowing.
462 * Corruption of the PSL_RF bit at worst causes one more or
463 * one less debugger trap, so allowing it is fairly harmless.
465 if (!EFL_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
466 kprintf("sigreturn: eflags = 0x%x\n", eflags);
471 * Don't allow users to load a valid privileged %cs. Let the
472 * hardware check for invalid selectors, excess privilege in
473 * other selectors, invalid %eip's and invalid %esp's.
475 cs = ucp.uc_mcontext.mc_cs;
476 if (!CS_SECURE(cs)) {
477 kprintf("sigreturn: cs = 0x%x\n", cs);
478 trapsignal(lp, SIGBUS, T_PROTFLT);
481 bcopy(&ucp.uc_mcontext.mc_gs, regs, sizeof(struct trapframe));
485 * Merge saved signal mailbox pending flag to maintain interlock
486 * semantics against system calls.
488 if (ucp.uc_mcontext.mc_xflags & PGEX_MAILBOX)
489 p->p_flag |= P_MAILBOX;
491 if (ucp.uc_mcontext.mc_onstack & 1)
492 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
494 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
496 lp->lwp_sigmask = ucp.uc_sigmask;
497 SIG_CANTMASK(lp->lwp_sigmask);
502 * Stack frame on entry to function. %eax will contain the function vector,
503 * %ecx will contain the function data. flags, ecx, and eax will have
504 * already been pushed on the stack.
515 sendupcall(struct vmupcall *vu, int morepending)
517 struct lwp *lp = curthread->td_lwp;
518 struct proc *p = lp->lwp_proc;
519 struct trapframe *regs;
520 struct upcall upcall;
521 struct upc_frame upc_frame;
525 * If we are a virtual kernel running an emulated user process
526 * context, switch back to the virtual kernel context before
527 * trying to post the signal.
529 if (p->p_vkernel && p->p_vkernel->vk_current) {
530 lp->lwp_md.md_regs->tf_trapno = 0;
531 vkernel_trap(p, lp->lwp_md.md_regs);
535 * Get the upcall data structure
537 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
538 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
541 kprintf("bad upcall address\n");
546 * If the data structure is already marked pending or has a critical
547 * section count, mark the data structure as pending and return
548 * without doing an upcall. vu_pending is left set.
550 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
551 if (upcall.upc_pending < vu->vu_pending) {
552 upcall.upc_pending = vu->vu_pending;
553 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
554 sizeof(upcall.upc_pending));
560 * We can run this upcall now, clear vu_pending.
562 * Bump our critical section count and set or clear the
563 * user pending flag depending on whether more upcalls are
564 * pending. The user will be responsible for calling
565 * upc_dispatch(-1) to process remaining upcalls.
568 upcall.upc_pending = morepending;
569 crit_count += TDPRI_CRIT;
570 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
571 sizeof(upcall.upc_pending));
572 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
576 * Construct a stack frame and issue the upcall
578 regs = lp->lwp_md.md_regs;
579 upc_frame.eax = regs->tf_eax;
580 upc_frame.ecx = regs->tf_ecx;
581 upc_frame.edx = regs->tf_edx;
582 upc_frame.flags = regs->tf_eflags;
583 upc_frame.oldip = regs->tf_eip;
584 if (copyout(&upc_frame, (void *)(regs->tf_esp - sizeof(upc_frame)),
585 sizeof(upc_frame)) != 0) {
586 kprintf("bad stack on upcall\n");
588 regs->tf_eax = (register_t)vu->vu_func;
589 regs->tf_ecx = (register_t)vu->vu_data;
590 regs->tf_edx = (register_t)lp->lwp_upcall;
591 regs->tf_eip = (register_t)vu->vu_ctx;
592 regs->tf_esp -= sizeof(upc_frame);
597 * fetchupcall occurs in the context of a system call, which means that
598 * we have to return EJUSTRETURN in order to prevent eax and edx from
599 * being overwritten by the syscall return value.
601 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
602 * and the function pointer in %eax.
605 fetchupcall (struct vmupcall *vu, int morepending, void *rsp)
607 struct upc_frame upc_frame;
608 struct lwp *lp = curthread->td_lwp;
609 struct trapframe *regs;
611 struct upcall upcall;
614 regs = lp->lwp_md.md_regs;
616 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
620 * This jumps us to the next ready context.
623 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
626 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
627 crit_count += TDPRI_CRIT;
629 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
630 regs->tf_eax = (register_t)vu->vu_func;
631 regs->tf_ecx = (register_t)vu->vu_data;
632 regs->tf_edx = (register_t)lp->lwp_upcall;
633 regs->tf_eip = (register_t)vu->vu_ctx;
634 regs->tf_esp = (register_t)rsp;
637 * This returns us to the originally interrupted code.
639 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
640 regs->tf_eax = upc_frame.eax;
641 regs->tf_ecx = upc_frame.ecx;
642 regs->tf_edx = upc_frame.edx;
643 regs->tf_eflags = (regs->tf_eflags & ~PSL_USERCHANGE) |
644 (upc_frame.flags & PSL_USERCHANGE);
645 regs->tf_eip = upc_frame.oldip;
646 regs->tf_esp = (register_t)((char *)rsp + sizeof(upc_frame));
655 * cpu_idle() represents the idle LWKT. You cannot return from this function
656 * (unless you want to blow things up!). Instead we look for runnable threads
657 * and loop or halt as appropriate. Giant is not held on entry to the thread.
659 * The main loop is entered with a critical section held, we must release
660 * the critical section before doing anything else. lwkt_switch() will
661 * check for pending interrupts due to entering and exiting its own
664 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
665 * to wake a HLTed cpu up. However, there are cases where the idlethread
666 * will be entered with the possibility that no IPI will occur and in such
667 * cases lwkt_switch() sets TDF_IDLE_NOHLT.
669 static int cpu_idle_hlt = 1;
670 static int cpu_idle_hltcnt;
671 static int cpu_idle_spincnt;
672 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
673 &cpu_idle_hlt, 0, "Idle loop HLT enable");
674 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
675 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
676 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
677 &cpu_idle_spincnt, 0, "Idle loop entry spins");
682 struct thread *td = curthread;
685 KKASSERT(td->td_pri < TDPRI_CRIT);
688 * See if there are any LWKTs ready to go.
693 * If we are going to halt call splz unconditionally after
694 * CLIing to catch any interrupt races. Note that we are
695 * at SPL0 and interrupts are enabled.
697 * We must poll our mailbox signals prior to calling
698 * sigpause() in order to properly interlock with them.
700 if (cpu_idle_hlt && !lwkt_runnable() &&
701 (td->td_flags & TDF_IDLE_NOHLT) == 0) {
704 if (!lwkt_runnable()) {
709 __asm __volatile("pause");
714 td->td_flags &= ~TDF_IDLE_NOHLT;
718 /*__asm __volatile("sti; pause");*/
719 __asm __volatile("pause");
721 /*__asm __volatile("sti");*/
729 * Clear registers on exec
732 exec_setregs(u_long entry, u_long stack, u_long ps_strings)
734 struct thread *td = curthread;
735 struct lwp *lp = td->td_lwp;
736 struct trapframe *regs = lp->lwp_md.md_regs;
737 struct pcb *pcb = lp->lwp_thread->td_pcb;
739 /* was i386_user_cleanup() in NetBSD */
742 bzero((char *)regs, sizeof(struct trapframe));
743 regs->tf_eip = entry;
744 regs->tf_esp = stack;
745 regs->tf_eflags = PSL_USER | (regs->tf_eflags & PSL_T);
753 /* PS_STRINGS value for BSD/OS binaries. It is 0 for non-BSD/OS. */
754 regs->tf_ebx = ps_strings;
757 * Reset the hardware debug registers if they were in use.
758 * They won't have any meaning for the newly exec'd process.
760 if (pcb->pcb_flags & PCB_DBREGS) {
767 if (pcb == td->td_pcb) {
769 * Clear the debug registers on the running
770 * CPU, otherwise they will end up affecting
771 * the next process we switch to.
775 pcb->pcb_flags &= ~PCB_DBREGS;
779 * Initialize the math emulator (if any) for the current process.
780 * Actually, just clear the bit that says that the emulator has
781 * been initialized. Initialization is delayed until the process
782 * traps to the emulator (if it is done at all) mainly because
783 * emulators don't provide an entry point for initialization.
785 pcb->pcb_flags &= ~FP_SOFTFP;
788 * note: do not set CR0_TS here. npxinit() must do it after clearing
789 * gd_npxthread. Otherwise a preemptive interrupt thread may panic
794 load_cr0(rcr0() | CR0_MP);
798 /* Initialize the npx (if any) for the current process. */
799 npxinit(__INITIAL_NPXCW__);
804 * note: linux emulator needs edx to be 0x0 on entry, which is
805 * handled in execve simply by setting the 64 bit syscall
817 cr0 |= CR0_NE; /* Done by npxinit() */
818 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
820 if (cpu_class != CPUCLASS_386)
822 cr0 |= CR0_WP | CR0_AM;
829 sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
832 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
834 if (!error && req->newptr)
839 SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
840 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
842 extern u_long bootdev; /* not a cdev_t - encoding is different */
843 SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
844 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
847 * Initialize 386 and configure to run kernel
851 * Initialize segments & interrupt table
854 extern struct user *proc0paddr;
859 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
860 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
861 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
862 IDTVEC(page), IDTVEC(mchk), IDTVEC(fpu), IDTVEC(align),
863 IDTVEC(xmm), IDTVEC(syscall),
866 IDTVEC(int0x80_syscall);
870 #ifdef DEBUG_INTERRUPTS
871 extern inthand_t *Xrsvdary[256];
875 ptrace_set_pc(struct lwp *lp, unsigned long addr)
877 lp->lwp_md.md_regs->tf_eip = addr;
882 ptrace_single_step(struct lwp *lp)
884 lp->lwp_md.md_regs->tf_eflags |= PSL_T;
889 fill_regs(struct lwp *lp, struct reg *regs)
891 struct trapframe *tp;
893 tp = lp->lwp_md.md_regs;
894 regs->r_gs = tp->tf_gs;
895 regs->r_fs = tp->tf_fs;
896 regs->r_es = tp->tf_es;
897 regs->r_ds = tp->tf_ds;
898 regs->r_edi = tp->tf_edi;
899 regs->r_esi = tp->tf_esi;
900 regs->r_ebp = tp->tf_ebp;
901 regs->r_ebx = tp->tf_ebx;
902 regs->r_edx = tp->tf_edx;
903 regs->r_ecx = tp->tf_ecx;
904 regs->r_eax = tp->tf_eax;
905 regs->r_eip = tp->tf_eip;
906 regs->r_cs = tp->tf_cs;
907 regs->r_eflags = tp->tf_eflags;
908 regs->r_esp = tp->tf_esp;
909 regs->r_ss = tp->tf_ss;
914 set_regs(struct lwp *lp, struct reg *regs)
916 struct trapframe *tp;
918 tp = lp->lwp_md.md_regs;
919 if (!EFL_SECURE(regs->r_eflags, tp->tf_eflags) ||
920 !CS_SECURE(regs->r_cs))
922 tp->tf_gs = regs->r_gs;
923 tp->tf_fs = regs->r_fs;
924 tp->tf_es = regs->r_es;
925 tp->tf_ds = regs->r_ds;
926 tp->tf_edi = regs->r_edi;
927 tp->tf_esi = regs->r_esi;
928 tp->tf_ebp = regs->r_ebp;
929 tp->tf_ebx = regs->r_ebx;
930 tp->tf_edx = regs->r_edx;
931 tp->tf_ecx = regs->r_ecx;
932 tp->tf_eax = regs->r_eax;
933 tp->tf_eip = regs->r_eip;
934 tp->tf_cs = regs->r_cs;
935 tp->tf_eflags = regs->r_eflags;
936 tp->tf_esp = regs->r_esp;
937 tp->tf_ss = regs->r_ss;
941 #ifndef CPU_DISABLE_SSE
943 fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
945 struct env87 *penv_87 = &sv_87->sv_env;
946 struct envxmm *penv_xmm = &sv_xmm->sv_env;
949 /* FPU control/status */
950 penv_87->en_cw = penv_xmm->en_cw;
951 penv_87->en_sw = penv_xmm->en_sw;
952 penv_87->en_tw = penv_xmm->en_tw;
953 penv_87->en_fip = penv_xmm->en_fip;
954 penv_87->en_fcs = penv_xmm->en_fcs;
955 penv_87->en_opcode = penv_xmm->en_opcode;
956 penv_87->en_foo = penv_xmm->en_foo;
957 penv_87->en_fos = penv_xmm->en_fos;
960 for (i = 0; i < 8; ++i)
961 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
963 sv_87->sv_ex_sw = sv_xmm->sv_ex_sw;
967 set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
969 struct env87 *penv_87 = &sv_87->sv_env;
970 struct envxmm *penv_xmm = &sv_xmm->sv_env;
973 /* FPU control/status */
974 penv_xmm->en_cw = penv_87->en_cw;
975 penv_xmm->en_sw = penv_87->en_sw;
976 penv_xmm->en_tw = penv_87->en_tw;
977 penv_xmm->en_fip = penv_87->en_fip;
978 penv_xmm->en_fcs = penv_87->en_fcs;
979 penv_xmm->en_opcode = penv_87->en_opcode;
980 penv_xmm->en_foo = penv_87->en_foo;
981 penv_xmm->en_fos = penv_87->en_fos;
984 for (i = 0; i < 8; ++i)
985 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
987 sv_xmm->sv_ex_sw = sv_87->sv_ex_sw;
989 #endif /* CPU_DISABLE_SSE */
992 fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
994 #ifndef CPU_DISABLE_SSE
996 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
997 (struct save87 *)fpregs);
1000 #endif /* CPU_DISABLE_SSE */
1001 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
1006 set_fpregs(struct lwp *lp, struct fpreg *fpregs)
1008 #ifndef CPU_DISABLE_SSE
1010 set_fpregs_xmm((struct save87 *)fpregs,
1011 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
1014 #endif /* CPU_DISABLE_SSE */
1015 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
1020 fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
1023 dbregs->dr0 = rdr0();
1024 dbregs->dr1 = rdr1();
1025 dbregs->dr2 = rdr2();
1026 dbregs->dr3 = rdr3();
1027 dbregs->dr4 = rdr4();
1028 dbregs->dr5 = rdr5();
1029 dbregs->dr6 = rdr6();
1030 dbregs->dr7 = rdr7();
1034 pcb = lp->lwp_thread->td_pcb;
1035 dbregs->dr0 = pcb->pcb_dr0;
1036 dbregs->dr1 = pcb->pcb_dr1;
1037 dbregs->dr2 = pcb->pcb_dr2;
1038 dbregs->dr3 = pcb->pcb_dr3;
1041 dbregs->dr6 = pcb->pcb_dr6;
1042 dbregs->dr7 = pcb->pcb_dr7;
1048 set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1051 load_dr0(dbregs->dr0);
1052 load_dr1(dbregs->dr1);
1053 load_dr2(dbregs->dr2);
1054 load_dr3(dbregs->dr3);
1055 load_dr4(dbregs->dr4);
1056 load_dr5(dbregs->dr5);
1057 load_dr6(dbregs->dr6);
1058 load_dr7(dbregs->dr7);
1061 struct ucred *ucred;
1063 uint32_t mask1, mask2;
1066 * Don't let an illegal value for dr7 get set. Specifically,
1067 * check for undefined settings. Setting these bit patterns
1068 * result in undefined behaviour and can lead to an unexpected
1071 for (i = 0, mask1 = 0x3<<16, mask2 = 0x2<<16; i < 8;
1072 i++, mask1 <<= 2, mask2 <<= 2)
1073 if ((dbregs->dr7 & mask1) == mask2)
1076 pcb = lp->lwp_thread->td_pcb;
1077 ucred = lp->lwp_proc->p_ucred;
1080 * Don't let a process set a breakpoint that is not within the
1081 * process's address space. If a process could do this, it
1082 * could halt the system by setting a breakpoint in the kernel
1083 * (if ddb was enabled). Thus, we need to check to make sure
1084 * that no breakpoints are being enabled for addresses outside
1085 * process's address space, unless, perhaps, we were called by
1088 * XXX - what about when the watched area of the user's
1089 * address space is written into from within the kernel
1090 * ... wouldn't that still cause a breakpoint to be generated
1091 * from within kernel mode?
1094 if (suser_cred(ucred, 0) != 0) {
1095 if (dbregs->dr7 & 0x3) {
1096 /* dr0 is enabled */
1097 if (dbregs->dr0 >= VM_MAX_USER_ADDRESS)
1101 if (dbregs->dr7 & (0x3<<2)) {
1102 /* dr1 is enabled */
1103 if (dbregs->dr1 >= VM_MAX_USER_ADDRESS)
1107 if (dbregs->dr7 & (0x3<<4)) {
1108 /* dr2 is enabled */
1109 if (dbregs->dr2 >= VM_MAX_USER_ADDRESS)
1113 if (dbregs->dr7 & (0x3<<6)) {
1114 /* dr3 is enabled */
1115 if (dbregs->dr3 >= VM_MAX_USER_ADDRESS)
1120 pcb->pcb_dr0 = dbregs->dr0;
1121 pcb->pcb_dr1 = dbregs->dr1;
1122 pcb->pcb_dr2 = dbregs->dr2;
1123 pcb->pcb_dr3 = dbregs->dr3;
1124 pcb->pcb_dr6 = dbregs->dr6;
1125 pcb->pcb_dr7 = dbregs->dr7;
1127 pcb->pcb_flags |= PCB_DBREGS;
1135 * Return > 0 if a hardware breakpoint has been hit, and the
1136 * breakpoint was in user space. Return 0, otherwise.
1139 user_dbreg_trap(void)
1141 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1142 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1143 int nbp; /* number of breakpoints that triggered */
1144 caddr_t addr[4]; /* breakpoint addresses */
1148 if ((dr7 & 0x000000ff) == 0) {
1150 * all GE and LE bits in the dr7 register are zero,
1151 * thus the trap couldn't have been caused by the
1152 * hardware debug registers
1159 bp = dr6 & 0x0000000f;
1163 * None of the breakpoint bits are set meaning this
1164 * trap was not caused by any of the debug registers
1170 * at least one of the breakpoints were hit, check to see
1171 * which ones and if any of them are user space addresses
1175 addr[nbp++] = (caddr_t)rdr0();
1178 addr[nbp++] = (caddr_t)rdr1();
1181 addr[nbp++] = (caddr_t)rdr2();
1184 addr[nbp++] = (caddr_t)rdr3();
1187 for (i=0; i<nbp; i++) {
1189 (caddr_t)VM_MAX_USER_ADDRESS) {
1191 * addr[i] is in user space
1198 * None of the breakpoints are in user space.
1208 Debugger(const char *msg)
1210 kprintf("Debugger(\"%s\") called.\n", msg);
1214 #include <sys/disklabel.h>
1217 * Determine the size of the transfer, and make sure it is
1218 * within the boundaries of the partition. Adjust transfer
1219 * if needed, and signal errors or early completion.
1221 * On success a new bio layer is pushed with the translated
1222 * block number, and returned.
1225 bounds_check_with_label(cdev_t dev, struct bio *bio,
1226 struct disklabel *lp, int wlabel)
1229 struct buf *bp = bio->bio_buf;
1230 struct partition *p = lp->d_partitions + dkpart(dev);
1231 int labelsect = lp->d_partitions[0].p_offset;
1232 int maxsz = p->p_size,
1233 sz = (bp->b_bcount + DEV_BSIZE - 1) >> DEV_BSHIFT;
1234 daddr_t blkno = (daddr_t)(bio->bio_offset >> DEV_BSHIFT);
1236 /* overwriting disk label ? */
1237 /* XXX should also protect bootstrap in first 8K */
1238 if (blkno + p->p_offset <= LABELSECTOR + labelsect &&
1239 #if LABELSECTOR != 0
1240 blkno + p->p_offset + sz > LABELSECTOR + labelsect &&
1242 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1243 bp->b_error = EROFS;
1247 #if defined(DOSBBSECTOR) && defined(notyet)
1248 /* overwriting master boot record? */
1249 if (blkno + p->p_offset <= DOSBBSECTOR &&
1250 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1251 bp->b_error = EROFS;
1257 * Check for out of bounds, EOF, and EOF clipping.
1259 if (bio->bio_offset < 0)
1261 if (blkno + sz > maxsz) {
1263 * Past EOF or B_BNOCLIP flag was set, the request is bad.
1265 if (blkno > maxsz || (bp->b_flags & B_BNOCLIP))
1269 * If exactly on EOF just complete the I/O with no bytes
1270 * transfered. B_INVAL must be set to throw away the
1271 * contents of the buffer. Otherwise clip b_bcount.
1273 if (blkno == maxsz) {
1274 bp->b_resid = bp->b_bcount;
1275 bp->b_flags |= B_INVAL;
1278 bp->b_bcount = (maxsz - blkno) << DEV_BSHIFT;
1280 nbio = push_bio(bio);
1281 nbio->bio_offset = bio->bio_offset + ((off_t)p->p_offset << DEV_BSHIFT);
1285 * The caller is responsible for calling biodone() on the passed bio
1286 * when we return NULL.
1289 bp->b_error = EINVAL;
1291 bp->b_resid = bp->b_bcount;
1292 bp->b_flags |= B_ERROR | B_INVAL;