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 1.6 2007/01/09 06:04:57 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>
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", "");
159 sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
163 /* Unwind the buffer, so that it's linear (possibly starting with
164 * some initial nulls).
166 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr,
167 msgbufp->msg_size-msgbufp->msg_bufr,req);
168 if(error) return(error);
169 if(msgbufp->msg_bufr>0) {
170 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr,
171 msgbufp->msg_bufr,req);
176 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
177 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
179 static int msgbuf_clear;
182 sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
185 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
187 if (!error && req->newptr) {
188 /* Clear the buffer and reset write pointer */
189 bzero(msgbufp->msg_ptr,msgbufp->msg_size);
190 msgbufp->msg_bufr=msgbufp->msg_bufx=0;
196 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
197 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
198 "Clear kernel message buffer");
203 * Send an interrupt to process.
205 * Stack is set up to allow sigcode stored
206 * at top to call routine, followed by kcall
207 * to sigreturn routine below. After sigreturn
208 * resets the signal mask, the stack, and the
209 * frame pointer, it returns to the user
213 sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
215 struct lwp *lp = curthread->td_lwp;
216 struct proc *p = lp->lwp_proc;
217 struct trapframe *regs;
218 struct sigacts *psp = p->p_sigacts;
219 struct sigframe sf, *sfp;
222 regs = lp->lwp_md.md_regs;
223 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
225 /* save user context */
226 bzero(&sf, sizeof(struct sigframe));
227 sf.sf_uc.uc_sigmask = *mask;
228 sf.sf_uc.uc_stack = lp->lwp_sigstk;
229 sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
230 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_gs, sizeof(struct trapframe));
232 /* Allocate and validate space for the signal handler context. */
234 if ((p->p_flag & P_ALTSTACK) != 0 && !oonstack &&
235 SIGISMEMBER(psp->ps_sigonstack, sig)) {
236 sfp = (struct sigframe *)(lp->lwp_sigstk.ss_sp +
237 lp->lwp_sigstk.ss_size - sizeof(struct sigframe));
238 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
241 sfp = (struct sigframe *)regs->tf_esp - 1;
243 /* Translate the signal is appropriate */
244 if (p->p_sysent->sv_sigtbl) {
245 if (sig <= p->p_sysent->sv_sigsize)
246 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
249 /* Build the argument list for the signal handler. */
251 sf.sf_ucontext = (register_t)&sfp->sf_uc;
252 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
253 /* Signal handler installed with SA_SIGINFO. */
254 sf.sf_siginfo = (register_t)&sfp->sf_si;
255 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
257 /* fill siginfo structure */
258 sf.sf_si.si_signo = sig;
259 sf.sf_si.si_code = code;
260 sf.sf_si.si_addr = (void*)regs->tf_err;
263 /* Old FreeBSD-style arguments. */
264 sf.sf_siginfo = code;
265 sf.sf_addr = regs->tf_err;
266 sf.sf_ahu.sf_handler = catcher;
271 * If we're a vm86 process, we want to save the segment registers.
272 * We also change eflags to be our emulated eflags, not the actual
275 if (regs->tf_eflags & PSL_VM) {
276 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
277 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
279 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
280 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
281 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
282 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
284 if (vm86->vm86_has_vme == 0)
285 sf.sf_uc.uc_mcontext.mc_eflags =
286 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
287 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
290 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
291 * syscalls made by the signal handler. This just avoids
292 * wasting time for our lazy fixup of such faults. PSL_NT
293 * does nothing in vm86 mode, but vm86 programs can set it
294 * almost legitimately in probes for old cpu types.
296 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
301 * Copy the sigframe out to the user's stack.
303 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
305 * Something is wrong with the stack pointer.
306 * ...Kill the process.
311 regs->tf_esp = (int)sfp;
312 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
313 regs->tf_eflags &= ~PSL_T;
314 regs->tf_cs = _ucodesel;
315 regs->tf_ds = _udatasel;
316 regs->tf_es = _udatasel;
317 if (regs->tf_trapno == T_PROTFLT) {
318 regs->tf_fs = _udatasel;
319 regs->tf_gs = _udatasel;
321 regs->tf_ss = _udatasel;
325 * Sanitize the trapframe for a virtual kernel passing control to a custom
328 * Allow userland to set or maintain PSL_RF, the resume flag. This flag
329 * basically controls whether the return PC should skip the first instruction
330 * (as in an explicit system call) or re-execute it (as in an exception).
333 cpu_sanitize_frame(struct trapframe *frame)
335 frame->tf_cs = _ucodesel;
336 frame->tf_ds = _udatasel;
337 frame->tf_es = _udatasel;
339 frame->tf_fs = _udatasel;
340 frame->tf_gs = _udatasel;
342 frame->tf_ss = _udatasel;
343 frame->tf_eflags &= (PSL_RF | PSL_USERCHANGE);
344 frame->tf_eflags |= PSL_RESERVED_DEFAULT | PSL_I;
349 cpu_sanitize_tls(struct savetls *tls)
351 struct segment_descriptor *desc;
354 for (i = 0; i < NGTLS; ++i) {
356 if (desc->sd_dpl == 0 && desc->sd_type == 0)
358 if (desc->sd_def32 == 0)
360 if (desc->sd_type != SDT_MEMRWA)
362 if (desc->sd_dpl != SEL_UPL)
364 if (desc->sd_xx != 0 || desc->sd_p != 1)
371 * sigreturn(ucontext_t *sigcntxp)
373 * System call to cleanup state after a signal
374 * has been taken. Reset signal mask and
375 * stack state from context left by sendsig (above).
376 * Return to previous pc and psl as specified by
377 * context left by sendsig. Check carefully to
378 * make sure that the user has not modified the
379 * state to gain improper privileges.
381 #define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
382 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
385 sys_sigreturn(struct sigreturn_args *uap)
387 struct lwp *lp = curthread->td_lwp;
388 struct trapframe *regs;
394 error = copyin(uap->sigcntxp, &ucp, sizeof(ucp));
398 regs = lp->lwp_md.md_regs;
399 eflags = ucp.uc_mcontext.mc_eflags;
402 if (eflags & PSL_VM) {
403 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
404 struct vm86_kernel *vm86;
407 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
408 * set up the vm86 area, and we can't enter vm86 mode.
410 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
412 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
413 if (vm86->vm86_inited == 0)
416 /* go back to user mode if both flags are set */
417 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
418 trapsignal(lp->lwp_proc, SIGBUS, 0);
420 if (vm86->vm86_has_vme) {
421 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
422 (eflags & VME_USERCHANGE) | PSL_VM;
424 vm86->vm86_eflags = eflags; /* save VIF, VIP */
425 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
427 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
428 tf->tf_eflags = eflags;
429 tf->tf_vm86_ds = tf->tf_ds;
430 tf->tf_vm86_es = tf->tf_es;
431 tf->tf_vm86_fs = tf->tf_fs;
432 tf->tf_vm86_gs = tf->tf_gs;
433 tf->tf_ds = _udatasel;
434 tf->tf_es = _udatasel;
436 tf->tf_fs = _udatasel;
437 tf->tf_gs = _udatasel;
443 * Don't allow users to change privileged or reserved flags.
446 * XXX do allow users to change the privileged flag PSL_RF.
447 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
448 * should sometimes set it there too. tf_eflags is kept in
449 * the signal context during signal handling and there is no
450 * other place to remember it, so the PSL_RF bit may be
451 * corrupted by the signal handler without us knowing.
452 * Corruption of the PSL_RF bit at worst causes one more or
453 * one less debugger trap, so allowing it is fairly harmless.
455 if (!EFL_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
456 kprintf("sigreturn: eflags = 0x%x\n", eflags);
461 * Don't allow users to load a valid privileged %cs. Let the
462 * hardware check for invalid selectors, excess privilege in
463 * other selectors, invalid %eip's and invalid %esp's.
465 cs = ucp.uc_mcontext.mc_cs;
466 if (!CS_SECURE(cs)) {
467 kprintf("sigreturn: cs = 0x%x\n", cs);
468 trapsignal(lp->lwp_proc, SIGBUS, T_PROTFLT);
471 bcopy(&ucp.uc_mcontext.mc_gs, regs, sizeof(struct trapframe));
474 if (ucp.uc_mcontext.mc_onstack & 1)
475 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
477 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
479 lp->lwp_sigmask = ucp.uc_sigmask;
480 SIG_CANTMASK(lp->lwp_sigmask);
485 * Stack frame on entry to function. %eax will contain the function vector,
486 * %ecx will contain the function data. flags, ecx, and eax will have
487 * already been pushed on the stack.
498 sendupcall(struct vmupcall *vu, int morepending)
500 struct lwp *lp = curthread->td_lwp;
501 struct trapframe *regs;
502 struct upcall upcall;
503 struct upc_frame upc_frame;
507 * Get the upcall data structure
509 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
510 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
513 kprintf("bad upcall address\n");
518 * If the data structure is already marked pending or has a critical
519 * section count, mark the data structure as pending and return
520 * without doing an upcall. vu_pending is left set.
522 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
523 if (upcall.upc_pending < vu->vu_pending) {
524 upcall.upc_pending = vu->vu_pending;
525 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
526 sizeof(upcall.upc_pending));
532 * We can run this upcall now, clear vu_pending.
534 * Bump our critical section count and set or clear the
535 * user pending flag depending on whether more upcalls are
536 * pending. The user will be responsible for calling
537 * upc_dispatch(-1) to process remaining upcalls.
540 upcall.upc_pending = morepending;
541 crit_count += TDPRI_CRIT;
542 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
543 sizeof(upcall.upc_pending));
544 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
548 * Construct a stack frame and issue the upcall
550 regs = lp->lwp_md.md_regs;
551 upc_frame.eax = regs->tf_eax;
552 upc_frame.ecx = regs->tf_ecx;
553 upc_frame.edx = regs->tf_edx;
554 upc_frame.flags = regs->tf_eflags;
555 upc_frame.oldip = regs->tf_eip;
556 if (copyout(&upc_frame, (void *)(regs->tf_esp - sizeof(upc_frame)),
557 sizeof(upc_frame)) != 0) {
558 kprintf("bad stack on upcall\n");
560 regs->tf_eax = (register_t)vu->vu_func;
561 regs->tf_ecx = (register_t)vu->vu_data;
562 regs->tf_edx = (register_t)lp->lwp_upcall;
563 regs->tf_eip = (register_t)vu->vu_ctx;
564 regs->tf_esp -= sizeof(upc_frame);
569 * fetchupcall occurs in the context of a system call, which means that
570 * we have to return EJUSTRETURN in order to prevent eax and edx from
571 * being overwritten by the syscall return value.
573 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
574 * and the function pointer in %eax.
577 fetchupcall (struct vmupcall *vu, int morepending, void *rsp)
579 struct upc_frame upc_frame;
580 struct lwp *lp = curthread->td_lwp;
581 struct trapframe *regs;
583 struct upcall upcall;
586 regs = lp->lwp_md.md_regs;
588 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
592 * This jumps us to the next ready context.
595 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
598 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
599 crit_count += TDPRI_CRIT;
601 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
602 regs->tf_eax = (register_t)vu->vu_func;
603 regs->tf_ecx = (register_t)vu->vu_data;
604 regs->tf_edx = (register_t)lp->lwp_upcall;
605 regs->tf_eip = (register_t)vu->vu_ctx;
606 regs->tf_esp = (register_t)rsp;
609 * This returns us to the originally interrupted code.
611 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
612 regs->tf_eax = upc_frame.eax;
613 regs->tf_ecx = upc_frame.ecx;
614 regs->tf_edx = upc_frame.edx;
615 regs->tf_eflags = (regs->tf_eflags & ~PSL_USERCHANGE) |
616 (upc_frame.flags & PSL_USERCHANGE);
617 regs->tf_eip = upc_frame.oldip;
618 regs->tf_esp = (register_t)((char *)rsp + sizeof(upc_frame));
627 * cpu_idle() represents the idle LWKT. You cannot return from this function
628 * (unless you want to blow things up!). Instead we look for runnable threads
629 * and loop or halt as appropriate. Giant is not held on entry to the thread.
631 * The main loop is entered with a critical section held, we must release
632 * the critical section before doing anything else. lwkt_switch() will
633 * check for pending interrupts due to entering and exiting its own
636 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
637 * to wake a HLTed cpu up. However, there are cases where the idlethread
638 * will be entered with the possibility that no IPI will occur and in such
639 * cases lwkt_switch() sets TDF_IDLE_NOHLT.
641 static int cpu_idle_hlt = 1;
642 static int cpu_idle_hltcnt;
643 static int cpu_idle_spincnt;
644 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
645 &cpu_idle_hlt, 0, "Idle loop HLT enable");
646 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
647 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
648 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
649 &cpu_idle_spincnt, 0, "Idle loop entry spins");
654 struct thread *td = curthread;
657 KKASSERT(td->td_pri < TDPRI_CRIT);
660 * See if there are any LWKTs ready to go.
665 * If we are going to halt call splz unconditionally after
666 * CLIing to catch any interrupt races. Note that we are
667 * at SPL0 and interrupts are enabled.
669 if (cpu_idle_hlt && !lwkt_runnable() &&
670 (td->td_flags & TDF_IDLE_NOHLT) == 0) {
673 if (!lwkt_runnable()) {
680 __asm __volatile("pause");
685 td->td_flags &= ~TDF_IDLE_NOHLT;
688 /*__asm __volatile("sti; pause");*/
689 __asm __volatile("pause");
691 /*__asm __volatile("sti");*/
699 * Clear registers on exec
702 setregs(struct lwp *lp, u_long entry, u_long stack, u_long ps_strings)
704 struct trapframe *regs = lp->lwp_md.md_regs;
705 struct pcb *pcb = lp->lwp_thread->td_pcb;
707 /* was i386_user_cleanup() in NetBSD */
710 bzero((char *)regs, sizeof(struct trapframe));
711 regs->tf_eip = entry;
712 regs->tf_esp = stack;
713 regs->tf_eflags = PSL_USER | (regs->tf_eflags & PSL_T);
721 /* PS_STRINGS value for BSD/OS binaries. It is 0 for non-BSD/OS. */
722 regs->tf_ebx = ps_strings;
725 * Reset the hardware debug registers if they were in use.
726 * They won't have any meaning for the newly exec'd process.
728 if (pcb->pcb_flags & PCB_DBREGS) {
735 if (pcb == curthread->td_pcb) {
737 * Clear the debug registers on the running
738 * CPU, otherwise they will end up affecting
739 * the next process we switch to.
743 pcb->pcb_flags &= ~PCB_DBREGS;
747 * Initialize the math emulator (if any) for the current process.
748 * Actually, just clear the bit that says that the emulator has
749 * been initialized. Initialization is delayed until the process
750 * traps to the emulator (if it is done at all) mainly because
751 * emulators don't provide an entry point for initialization.
753 lp->lwp_thread->td_pcb->pcb_flags &= ~FP_SOFTFP;
756 * note: do not set CR0_TS here. npxinit() must do it after clearing
757 * gd_npxthread. Otherwise a preemptive interrupt thread may panic
762 load_cr0(rcr0() | CR0_MP);
766 /* Initialize the npx (if any) for the current process. */
767 npxinit(__INITIAL_NPXCW__);
772 * note: linux emulator needs edx to be 0x0 on entry, which is
773 * handled in execve simply by setting the 64 bit syscall
785 cr0 |= CR0_NE; /* Done by npxinit() */
786 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
788 if (cpu_class != CPUCLASS_386)
790 cr0 |= CR0_WP | CR0_AM;
797 sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
800 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
802 if (!error && req->newptr)
807 SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
808 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
810 extern u_long bootdev; /* not a cdev_t - encoding is different */
811 SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
812 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
815 * Initialize 386 and configure to run kernel
819 * Initialize segments & interrupt table
822 extern struct user *proc0paddr;
827 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
828 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
829 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
830 IDTVEC(page), IDTVEC(mchk), IDTVEC(fpu), IDTVEC(align),
831 IDTVEC(xmm), IDTVEC(syscall),
834 IDTVEC(int0x80_syscall);
838 #ifdef DEBUG_INTERRUPTS
839 extern inthand_t *Xrsvdary[256];
843 ptrace_set_pc(struct proc *p, unsigned long addr)
845 p->p_md.md_regs->tf_eip = addr;
850 ptrace_single_step(struct lwp *lp)
852 lp->lwp_md.md_regs->tf_eflags |= PSL_T;
857 fill_regs(struct lwp *lp, struct reg *regs)
859 struct trapframe *tp;
861 tp = lp->lwp_md.md_regs;
862 regs->r_gs = tp->tf_gs;
863 regs->r_fs = tp->tf_fs;
864 regs->r_es = tp->tf_es;
865 regs->r_ds = tp->tf_ds;
866 regs->r_edi = tp->tf_edi;
867 regs->r_esi = tp->tf_esi;
868 regs->r_ebp = tp->tf_ebp;
869 regs->r_ebx = tp->tf_ebx;
870 regs->r_edx = tp->tf_edx;
871 regs->r_ecx = tp->tf_ecx;
872 regs->r_eax = tp->tf_eax;
873 regs->r_eip = tp->tf_eip;
874 regs->r_cs = tp->tf_cs;
875 regs->r_eflags = tp->tf_eflags;
876 regs->r_esp = tp->tf_esp;
877 regs->r_ss = tp->tf_ss;
882 set_regs(struct lwp *lp, struct reg *regs)
884 struct trapframe *tp;
886 tp = lp->lwp_md.md_regs;
887 if (!EFL_SECURE(regs->r_eflags, tp->tf_eflags) ||
888 !CS_SECURE(regs->r_cs))
890 tp->tf_gs = regs->r_gs;
891 tp->tf_fs = regs->r_fs;
892 tp->tf_es = regs->r_es;
893 tp->tf_ds = regs->r_ds;
894 tp->tf_edi = regs->r_edi;
895 tp->tf_esi = regs->r_esi;
896 tp->tf_ebp = regs->r_ebp;
897 tp->tf_ebx = regs->r_ebx;
898 tp->tf_edx = regs->r_edx;
899 tp->tf_ecx = regs->r_ecx;
900 tp->tf_eax = regs->r_eax;
901 tp->tf_eip = regs->r_eip;
902 tp->tf_cs = regs->r_cs;
903 tp->tf_eflags = regs->r_eflags;
904 tp->tf_esp = regs->r_esp;
905 tp->tf_ss = regs->r_ss;
909 #ifndef CPU_DISABLE_SSE
911 fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
913 struct env87 *penv_87 = &sv_87->sv_env;
914 struct envxmm *penv_xmm = &sv_xmm->sv_env;
917 /* FPU control/status */
918 penv_87->en_cw = penv_xmm->en_cw;
919 penv_87->en_sw = penv_xmm->en_sw;
920 penv_87->en_tw = penv_xmm->en_tw;
921 penv_87->en_fip = penv_xmm->en_fip;
922 penv_87->en_fcs = penv_xmm->en_fcs;
923 penv_87->en_opcode = penv_xmm->en_opcode;
924 penv_87->en_foo = penv_xmm->en_foo;
925 penv_87->en_fos = penv_xmm->en_fos;
928 for (i = 0; i < 8; ++i)
929 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
931 sv_87->sv_ex_sw = sv_xmm->sv_ex_sw;
935 set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
937 struct env87 *penv_87 = &sv_87->sv_env;
938 struct envxmm *penv_xmm = &sv_xmm->sv_env;
941 /* FPU control/status */
942 penv_xmm->en_cw = penv_87->en_cw;
943 penv_xmm->en_sw = penv_87->en_sw;
944 penv_xmm->en_tw = penv_87->en_tw;
945 penv_xmm->en_fip = penv_87->en_fip;
946 penv_xmm->en_fcs = penv_87->en_fcs;
947 penv_xmm->en_opcode = penv_87->en_opcode;
948 penv_xmm->en_foo = penv_87->en_foo;
949 penv_xmm->en_fos = penv_87->en_fos;
952 for (i = 0; i < 8; ++i)
953 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
955 sv_xmm->sv_ex_sw = sv_87->sv_ex_sw;
957 #endif /* CPU_DISABLE_SSE */
960 fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
962 #ifndef CPU_DISABLE_SSE
964 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
965 (struct save87 *)fpregs);
968 #endif /* CPU_DISABLE_SSE */
969 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
974 set_fpregs(struct lwp *lp, struct fpreg *fpregs)
976 #ifndef CPU_DISABLE_SSE
978 set_fpregs_xmm((struct save87 *)fpregs,
979 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
982 #endif /* CPU_DISABLE_SSE */
983 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
988 fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
991 dbregs->dr0 = rdr0();
992 dbregs->dr1 = rdr1();
993 dbregs->dr2 = rdr2();
994 dbregs->dr3 = rdr3();
995 dbregs->dr4 = rdr4();
996 dbregs->dr5 = rdr5();
997 dbregs->dr6 = rdr6();
998 dbregs->dr7 = rdr7();
1002 pcb = lp->lwp_thread->td_pcb;
1003 dbregs->dr0 = pcb->pcb_dr0;
1004 dbregs->dr1 = pcb->pcb_dr1;
1005 dbregs->dr2 = pcb->pcb_dr2;
1006 dbregs->dr3 = pcb->pcb_dr3;
1009 dbregs->dr6 = pcb->pcb_dr6;
1010 dbregs->dr7 = pcb->pcb_dr7;
1016 set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1019 load_dr0(dbregs->dr0);
1020 load_dr1(dbregs->dr1);
1021 load_dr2(dbregs->dr2);
1022 load_dr3(dbregs->dr3);
1023 load_dr4(dbregs->dr4);
1024 load_dr5(dbregs->dr5);
1025 load_dr6(dbregs->dr6);
1026 load_dr7(dbregs->dr7);
1029 struct ucred *ucred;
1031 uint32_t mask1, mask2;
1034 * Don't let an illegal value for dr7 get set. Specifically,
1035 * check for undefined settings. Setting these bit patterns
1036 * result in undefined behaviour and can lead to an unexpected
1039 for (i = 0, mask1 = 0x3<<16, mask2 = 0x2<<16; i < 8;
1040 i++, mask1 <<= 2, mask2 <<= 2)
1041 if ((dbregs->dr7 & mask1) == mask2)
1044 pcb = lp->lwp_thread->td_pcb;
1045 ucred = lp->lwp_proc->p_ucred;
1048 * Don't let a process set a breakpoint that is not within the
1049 * process's address space. If a process could do this, it
1050 * could halt the system by setting a breakpoint in the kernel
1051 * (if ddb was enabled). Thus, we need to check to make sure
1052 * that no breakpoints are being enabled for addresses outside
1053 * process's address space, unless, perhaps, we were called by
1056 * XXX - what about when the watched area of the user's
1057 * address space is written into from within the kernel
1058 * ... wouldn't that still cause a breakpoint to be generated
1059 * from within kernel mode?
1062 if (suser_cred(ucred, 0) != 0) {
1063 if (dbregs->dr7 & 0x3) {
1064 /* dr0 is enabled */
1065 if (dbregs->dr0 >= VM_MAX_USER_ADDRESS)
1069 if (dbregs->dr7 & (0x3<<2)) {
1070 /* dr1 is enabled */
1071 if (dbregs->dr1 >= VM_MAX_USER_ADDRESS)
1075 if (dbregs->dr7 & (0x3<<4)) {
1076 /* dr2 is enabled */
1077 if (dbregs->dr2 >= VM_MAX_USER_ADDRESS)
1081 if (dbregs->dr7 & (0x3<<6)) {
1082 /* dr3 is enabled */
1083 if (dbregs->dr3 >= VM_MAX_USER_ADDRESS)
1088 pcb->pcb_dr0 = dbregs->dr0;
1089 pcb->pcb_dr1 = dbregs->dr1;
1090 pcb->pcb_dr2 = dbregs->dr2;
1091 pcb->pcb_dr3 = dbregs->dr3;
1092 pcb->pcb_dr6 = dbregs->dr6;
1093 pcb->pcb_dr7 = dbregs->dr7;
1095 pcb->pcb_flags |= PCB_DBREGS;
1103 * Return > 0 if a hardware breakpoint has been hit, and the
1104 * breakpoint was in user space. Return 0, otherwise.
1107 user_dbreg_trap(void)
1109 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1110 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1111 int nbp; /* number of breakpoints that triggered */
1112 caddr_t addr[4]; /* breakpoint addresses */
1116 if ((dr7 & 0x000000ff) == 0) {
1118 * all GE and LE bits in the dr7 register are zero,
1119 * thus the trap couldn't have been caused by the
1120 * hardware debug registers
1127 bp = dr6 & 0x0000000f;
1131 * None of the breakpoint bits are set meaning this
1132 * trap was not caused by any of the debug registers
1138 * at least one of the breakpoints were hit, check to see
1139 * which ones and if any of them are user space addresses
1143 addr[nbp++] = (caddr_t)rdr0();
1146 addr[nbp++] = (caddr_t)rdr1();
1149 addr[nbp++] = (caddr_t)rdr2();
1152 addr[nbp++] = (caddr_t)rdr3();
1155 for (i=0; i<nbp; i++) {
1157 (caddr_t)VM_MAX_USER_ADDRESS) {
1159 * addr[i] is in user space
1166 * None of the breakpoints are in user space.
1176 Debugger(const char *msg)
1178 kprintf("Debugger(\"%s\") called.\n", msg);
1182 #include <sys/disklabel.h>
1185 * Determine the size of the transfer, and make sure it is
1186 * within the boundaries of the partition. Adjust transfer
1187 * if needed, and signal errors or early completion.
1189 * On success a new bio layer is pushed with the translated
1190 * block number, and returned.
1193 bounds_check_with_label(cdev_t dev, struct bio *bio,
1194 struct disklabel *lp, int wlabel)
1197 struct buf *bp = bio->bio_buf;
1198 struct partition *p = lp->d_partitions + dkpart(dev);
1199 int labelsect = lp->d_partitions[0].p_offset;
1200 int maxsz = p->p_size,
1201 sz = (bp->b_bcount + DEV_BSIZE - 1) >> DEV_BSHIFT;
1202 daddr_t blkno = (daddr_t)(bio->bio_offset >> DEV_BSHIFT);
1204 /* overwriting disk label ? */
1205 /* XXX should also protect bootstrap in first 8K */
1206 if (blkno + p->p_offset <= LABELSECTOR + labelsect &&
1207 #if LABELSECTOR != 0
1208 blkno + p->p_offset + sz > LABELSECTOR + labelsect &&
1210 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1211 bp->b_error = EROFS;
1215 #if defined(DOSBBSECTOR) && defined(notyet)
1216 /* overwriting master boot record? */
1217 if (blkno + p->p_offset <= DOSBBSECTOR &&
1218 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1219 bp->b_error = EROFS;
1225 * Check for out of bounds, EOF, and EOF clipping.
1227 if (bio->bio_offset < 0)
1229 if (blkno + sz > maxsz) {
1231 * Past EOF or B_BNOCLIP flag was set, the request is bad.
1233 if (blkno > maxsz || (bp->b_flags & B_BNOCLIP))
1237 * If exactly on EOF just complete the I/O with no bytes
1238 * transfered. B_INVAL must be set to throw away the
1239 * contents of the buffer. Otherwise clip b_bcount.
1241 if (blkno == maxsz) {
1242 bp->b_resid = bp->b_bcount;
1243 bp->b_flags |= B_INVAL;
1246 bp->b_bcount = (maxsz - blkno) << DEV_BSHIFT;
1248 nbio = push_bio(bio);
1249 nbio->bio_offset = bio->bio_offset + ((off_t)p->p_offset << DEV_BSHIFT);
1253 * The caller is responsible for calling biodone() on the passed bio
1254 * when we return NULL.
1257 bp->b_error = EINVAL;
1259 bp->b_resid = bp->b_bcount;
1260 bp->b_flags |= B_ERROR | B_INVAL;