Do not allow umtx_sleep() to restart on a restartable signal. We want to
[dragonfly.git] / sys / platform / vkernel / i386 / cpu_regs.c
CommitLineData
6f7b98e0
MD
1/*-
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.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * William Jolitz.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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.
25 *
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
36 * SUCH DAMAGE.
37 *
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 $
24eb47e0 40 * $DragonFly: src/sys/platform/vkernel/i386/cpu_regs.c,v 1.18 2007/07/01 02:51:43 dillon Exp $
6f7b98e0
MD
41 */
42
43#include "use_ether.h"
44#include "use_npx.h"
45#include "use_isa.h"
46#include "opt_atalk.h"
47#include "opt_compat.h"
48#include "opt_ddb.h"
49#include "opt_directio.h"
50#include "opt_inet.h"
51#include "opt_ipx.h"
52#include "opt_msgbuf.h"
53#include "opt_swap.h"
54
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>
62#include <sys/proc.h>
63#include <sys/buf.h>
64#include <sys/reboot.h>
65#include <sys/mbuf.h>
66#include <sys/msgbuf.h>
67#include <sys/sysent.h>
68#include <sys/sysctl.h>
69#include <sys/vmmeter.h>
70#include <sys/bus.h>
71#include <sys/upcall.h>
72#include <sys/usched.h>
73#include <sys/reg.h>
74
75#include <vm/vm.h>
76#include <vm/vm_param.h>
77#include <sys/lock.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>
84
85#include <sys/thread2.h>
86
87#include <sys/user.h>
88#include <sys/exec.h>
89#include <sys/cons.h>
90
91#include <ddb/ddb.h>
92
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>
100#ifdef PERFMON
101#include <machine/perfmon.h>
102#endif
103#include <machine/cputypes.h>
104
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
111extern void dblfault_handler (void);
112
113#ifndef CPU_DISABLE_SSE
114static void set_fpregs_xmm (struct save87 *, struct savexmm *);
115static void fill_fpregs_xmm (struct savexmm *, struct save87 *);
116#endif /* CPU_DISABLE_SSE */
117#ifdef DIRECTIO
118extern void ffs_rawread_setup(void);
119#endif /* DIRECTIO */
120
121#ifdef SMP
122int64_t tsc_offsets[MAXCPU];
123#else
124int64_t tsc_offsets[1];
125#endif
126
127#if defined(SWTCH_OPTIM_STATS)
128extern int swtch_optim_stats;
129SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats,
130 CTLFLAG_RD, &swtch_optim_stats, 0, "");
131SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count,
132 CTLFLAG_RD, &tlb_flush_count, 0, "");
133#endif
134
6f7b98e0
MD
135static int
136sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
137{
71152ac6 138 int error = sysctl_handle_int(oidp, 0, ctob((int)Maxmem), req);
6f7b98e0
MD
139 return (error);
140}
141
142SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_INT|CTLFLAG_RD,
143 0, 0, sysctl_hw_physmem, "IU", "");
144
145static int
146sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
147{
148 int error = sysctl_handle_int(oidp, 0,
71152ac6 149 ctob((int)Maxmem - vmstats.v_wire_count), req);
6f7b98e0
MD
150 return (error);
151}
152
153SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD,
154 0, 0, sysctl_hw_usermem, "IU", "");
155
44afc5dc
SS
156SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, NULL, "");
157
6f7b98e0
MD
158#if 0
159
160static int
161sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
162{
163 int error;
164
165 /* Unwind the buffer, so that it's linear (possibly starting with
166 * some initial nulls).
167 */
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);
174 }
175 return(error);
176}
177
178SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
179 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
180
181static int msgbuf_clear;
182
183static int
184sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
185{
186 int error;
187 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
188 req);
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;
193 msgbuf_clear=0;
194 }
195 return (error);
196}
197
198SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
199 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
200 "Clear kernel message buffer");
201
202#endif
203
204/*
205 * Send an interrupt to process.
206 *
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
212 * specified pc, psl.
213 */
214void
215sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
216{
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;
222 int oonstack;
223
224 regs = lp->lwp_md.md_regs;
225 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
226
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;
4e7c41c5 232 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_gs, sizeof(struct trapframe));
6f7b98e0 233
ce3d8c4d
MD
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);
236
4b486183
MD
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;
240
241
6f7b98e0 242 /* Allocate and validate space for the signal handler context. */
08f2f1bb 243 if ((lp->lwp_flag & LWP_ALTSTACK) != 0 && !oonstack &&
6f7b98e0
MD
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;
248 }
249 else
250 sfp = (struct sigframe *)regs->tf_esp - 1;
251
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)];
256 }
257
258 /* Build the argument list for the signal handler. */
259 sf.sf_signum = sig;
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;
265
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;
270 }
271 else {
272 /* Old FreeBSD-style arguments. */
273 sf.sf_siginfo = code;
274 sf.sf_addr = regs->tf_err;
275 sf.sf_ahu.sf_handler = catcher;
276 }
277
278#if 0
279 /*
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
282 * eflags.
283 */
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;
287
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;
292
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));
297
298 /*
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.
304 */
305 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
306 }
307#endif
308
309 /*
310 * Copy the sigframe out to the user's stack.
311 */
312 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
313 /*
314 * Something is wrong with the stack pointer.
315 * ...Kill the process.
316 */
317 sigexit(p, SIGILL);
318 }
319
320 regs->tf_esp = (int)sfp;
321 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
322 regs->tf_eflags &= ~PSL_T;
4e7c41c5
MD
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;
329 }
330 regs->tf_ss = _udatasel;
6f7b98e0
MD
331}
332
333/*
334 * Sanitize the trapframe for a virtual kernel passing control to a custom
335 * VM context.
336 *
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).
340 */
341int
342cpu_sanitize_frame(struct trapframe *frame)
343{
4e7c41c5
MD
344 frame->tf_cs = _ucodesel;
345 frame->tf_ds = _udatasel;
346 frame->tf_es = _udatasel;
347#if 0
348 frame->tf_fs = _udatasel;
349 frame->tf_gs = _udatasel;
350#endif
351 frame->tf_ss = _udatasel;
352 frame->tf_eflags &= (PSL_RF | PSL_USERCHANGE);
6f7b98e0
MD
353 frame->tf_eflags |= PSL_RESERVED_DEFAULT | PSL_I;
354 return(0);
355}
356
4e7c41c5
MD
357int
358cpu_sanitize_tls(struct savetls *tls)
359{
360 struct segment_descriptor *desc;
361 int i;
362
363 for (i = 0; i < NGTLS; ++i) {
364 desc = &tls->tls[i];
365 if (desc->sd_dpl == 0 && desc->sd_type == 0)
366 continue;
367 if (desc->sd_def32 == 0)
368 return(ENXIO);
369 if (desc->sd_type != SDT_MEMRWA)
370 return(ENXIO);
371 if (desc->sd_dpl != SEL_UPL)
372 return(ENXIO);
373 if (desc->sd_xx != 0 || desc->sd_p != 1)
374 return(ENXIO);
375 }
376 return(0);
377}
378
6f7b98e0
MD
379/*
380 * sigreturn(ucontext_t *sigcntxp)
381 *
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.
389 */
390#define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
391#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
392
393int
394sys_sigreturn(struct sigreturn_args *uap)
395{
396 struct lwp *lp = curthread->td_lwp;
4b486183 397 struct proc *p = lp->lwp_proc;
6f7b98e0 398 struct trapframe *regs;
aaf8b91f
MD
399 ucontext_t ucp;
400 int cs;
401 int eflags;
402 int error;
6f7b98e0 403
aaf8b91f
MD
404 error = copyin(uap->sigcntxp, &ucp, sizeof(ucp));
405 if (error)
406 return (error);
6f7b98e0
MD
407
408 regs = lp->lwp_md.md_regs;
aaf8b91f 409 eflags = ucp.uc_mcontext.mc_eflags;
6f7b98e0
MD
410
411#if 0
412 if (eflags & PSL_VM) {
413 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
414 struct vm86_kernel *vm86;
415
416 /*
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.
419 */
420 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
421 return (EINVAL);
422 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
423 if (vm86->vm86_inited == 0)
424 return (EINVAL);
425
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);
429
430 if (vm86->vm86_has_vme) {
431 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
432 (eflags & VME_USERCHANGE) | PSL_VM;
433 } else {
434 vm86->vm86_eflags = eflags; /* save VIF, VIP */
435 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
436 }
aaf8b91f 437 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
6f7b98e0
MD
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;
4e7c41c5
MD
442 tf->tf_vm86_gs = tf->tf_gs;
443 tf->tf_ds = _udatasel;
444 tf->tf_es = _udatasel;
445#if 0
446 tf->tf_fs = _udatasel;
447 tf->tf_gs = _udatasel;
448#endif
6f7b98e0
MD
449 } else
450#endif
451 {
452 /*
453 * Don't allow users to change privileged or reserved flags.
454 */
455 /*
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.
464 */
465 if (!EFL_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
466 kprintf("sigreturn: eflags = 0x%x\n", eflags);
467 return(EINVAL);
468 }
469
470 /*
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.
474 */
aaf8b91f 475 cs = ucp.uc_mcontext.mc_cs;
6f7b98e0
MD
476 if (!CS_SECURE(cs)) {
477 kprintf("sigreturn: cs = 0x%x\n", cs);
08f2f1bb 478 trapsignal(lp, SIGBUS, T_PROTFLT);
6f7b98e0
MD
479 return(EINVAL);
480 }
aaf8b91f 481 bcopy(&ucp.uc_mcontext.mc_gs, regs, sizeof(struct trapframe));
6f7b98e0
MD
482 }
483
4b486183
MD
484 /*
485 * Merge saved signal mailbox pending flag to maintain interlock
486 * semantics against system calls.
487 */
488 if (ucp.uc_mcontext.mc_xflags & PGEX_MAILBOX)
489 p->p_flag |= P_MAILBOX;
490
aaf8b91f 491 if (ucp.uc_mcontext.mc_onstack & 1)
6f7b98e0
MD
492 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
493 else
494 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
495
aaf8b91f 496 lp->lwp_sigmask = ucp.uc_sigmask;
6f7b98e0
MD
497 SIG_CANTMASK(lp->lwp_sigmask);
498 return(EJUSTRETURN);
499}
500
501/*
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.
505 */
506struct upc_frame {
507 register_t eax;
508 register_t ecx;
509 register_t edx;
510 register_t flags;
511 register_t oldip;
512};
513
514void
515sendupcall(struct vmupcall *vu, int morepending)
516{
517 struct lwp *lp = curthread->td_lwp;
518 struct trapframe *regs;
519 struct upcall upcall;
520 struct upc_frame upc_frame;
521 int crit_count = 0;
522
1fdc022f
MD
523 /*
524 * If we are a virtual kernel running an emulated user process
525 * context, switch back to the virtual kernel context before
526 * trying to post the signal.
527 */
39005e16 528 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1fdc022f 529 lp->lwp_md.md_regs->tf_trapno = 0;
287ebb09 530 vkernel_trap(lp, lp->lwp_md.md_regs);
1fdc022f
MD
531 }
532
6f7b98e0
MD
533 /*
534 * Get the upcall data structure
535 */
536 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
537 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
538 ) {
539 vu->vu_pending = 0;
540 kprintf("bad upcall address\n");
541 return;
542 }
543
544 /*
545 * If the data structure is already marked pending or has a critical
546 * section count, mark the data structure as pending and return
547 * without doing an upcall. vu_pending is left set.
548 */
549 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
550 if (upcall.upc_pending < vu->vu_pending) {
551 upcall.upc_pending = vu->vu_pending;
552 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
553 sizeof(upcall.upc_pending));
554 }
555 return;
556 }
557
558 /*
559 * We can run this upcall now, clear vu_pending.
560 *
561 * Bump our critical section count and set or clear the
562 * user pending flag depending on whether more upcalls are
563 * pending. The user will be responsible for calling
564 * upc_dispatch(-1) to process remaining upcalls.
565 */
566 vu->vu_pending = 0;
567 upcall.upc_pending = morepending;
568 crit_count += TDPRI_CRIT;
569 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
570 sizeof(upcall.upc_pending));
571 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
572 sizeof(int));
573
574 /*
575 * Construct a stack frame and issue the upcall
576 */
577 regs = lp->lwp_md.md_regs;
578 upc_frame.eax = regs->tf_eax;
579 upc_frame.ecx = regs->tf_ecx;
580 upc_frame.edx = regs->tf_edx;
581 upc_frame.flags = regs->tf_eflags;
582 upc_frame.oldip = regs->tf_eip;
583 if (copyout(&upc_frame, (void *)(regs->tf_esp - sizeof(upc_frame)),
584 sizeof(upc_frame)) != 0) {
585 kprintf("bad stack on upcall\n");
586 } else {
587 regs->tf_eax = (register_t)vu->vu_func;
588 regs->tf_ecx = (register_t)vu->vu_data;
589 regs->tf_edx = (register_t)lp->lwp_upcall;
590 regs->tf_eip = (register_t)vu->vu_ctx;
591 regs->tf_esp -= sizeof(upc_frame);
592 }
593}
594
595/*
596 * fetchupcall occurs in the context of a system call, which means that
597 * we have to return EJUSTRETURN in order to prevent eax and edx from
598 * being overwritten by the syscall return value.
599 *
600 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
601 * and the function pointer in %eax.
602 */
603int
604fetchupcall (struct vmupcall *vu, int morepending, void *rsp)
605{
606 struct upc_frame upc_frame;
607 struct lwp *lp = curthread->td_lwp;
608 struct trapframe *regs;
609 int error;
610 struct upcall upcall;
611 int crit_count;
612
613 regs = lp->lwp_md.md_regs;
614
615 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
616 if (error == 0) {
617 if (vu) {
618 /*
619 * This jumps us to the next ready context.
620 */
621 vu->vu_pending = 0;
622 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
623 crit_count = 0;
624 if (error == 0)
625 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
626 crit_count += TDPRI_CRIT;
627 if (error == 0)
628 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
629 regs->tf_eax = (register_t)vu->vu_func;
630 regs->tf_ecx = (register_t)vu->vu_data;
631 regs->tf_edx = (register_t)lp->lwp_upcall;
632 regs->tf_eip = (register_t)vu->vu_ctx;
633 regs->tf_esp = (register_t)rsp;
634 } else {
635 /*
636 * This returns us to the originally interrupted code.
637 */
638 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
639 regs->tf_eax = upc_frame.eax;
640 regs->tf_ecx = upc_frame.ecx;
641 regs->tf_edx = upc_frame.edx;
642 regs->tf_eflags = (regs->tf_eflags & ~PSL_USERCHANGE) |
643 (upc_frame.flags & PSL_USERCHANGE);
644 regs->tf_eip = upc_frame.oldip;
645 regs->tf_esp = (register_t)((char *)rsp + sizeof(upc_frame));
646 }
647 }
648 if (error == 0)
649 error = EJUSTRETURN;
650 return(error);
651}
652
653/*
654 * cpu_idle() represents the idle LWKT. You cannot return from this function
655 * (unless you want to blow things up!). Instead we look for runnable threads
656 * and loop or halt as appropriate. Giant is not held on entry to the thread.
657 *
658 * The main loop is entered with a critical section held, we must release
659 * the critical section before doing anything else. lwkt_switch() will
660 * check for pending interrupts due to entering and exiting its own
661 * critical section.
662 *
663 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
664 * to wake a HLTed cpu up. However, there are cases where the idlethread
665 * will be entered with the possibility that no IPI will occur and in such
666 * cases lwkt_switch() sets TDF_IDLE_NOHLT.
667 */
668static int cpu_idle_hlt = 1;
669static int cpu_idle_hltcnt;
670static int cpu_idle_spincnt;
671SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
672 &cpu_idle_hlt, 0, "Idle loop HLT enable");
673SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
674 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
675SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
676 &cpu_idle_spincnt, 0, "Idle loop entry spins");
677
6f7b98e0
MD
678void
679cpu_idle(void)
680{
681 struct thread *td = curthread;
682
683 crit_exit();
684 KKASSERT(td->td_pri < TDPRI_CRIT);
685 for (;;) {
686 /*
687 * See if there are any LWKTs ready to go.
688 */
689 lwkt_switch();
690
691 /*
692 * If we are going to halt call splz unconditionally after
693 * CLIing to catch any interrupt races. Note that we are
694 * at SPL0 and interrupts are enabled.
7c1212ec
MD
695 *
696 * We must poll our mailbox signals prior to calling
697 * sigpause() in order to properly interlock with them.
6f7b98e0
MD
698 */
699 if (cpu_idle_hlt && !lwkt_runnable() &&
700 (td->td_flags & TDF_IDLE_NOHLT) == 0) {
6f7b98e0 701 splz();
4e7c41c5
MD
702 if (!lwkt_runnable()) {
703 sigpause(0);
4e7c41c5 704 }
6f7b98e0 705#ifdef SMP
4e7c41c5 706 else {
6f7b98e0 707 __asm __volatile("pause");
4e7c41c5 708 }
6f7b98e0
MD
709#endif
710 ++cpu_idle_hltcnt;
711 } else {
712 td->td_flags &= ~TDF_IDLE_NOHLT;
713 splz();
714#ifdef SMP
715 /*__asm __volatile("sti; pause");*/
716 __asm __volatile("pause");
717#else
718 /*__asm __volatile("sti");*/
719#endif
720 ++cpu_idle_spincnt;
721 }
722 }
723}
724
725/*
726 * Clear registers on exec
727 */
728void
08f2f1bb 729exec_setregs(u_long entry, u_long stack, u_long ps_strings)
6f7b98e0 730{
08f2f1bb
SS
731 struct thread *td = curthread;
732 struct lwp *lp = td->td_lwp;
6f7b98e0
MD
733 struct trapframe *regs = lp->lwp_md.md_regs;
734 struct pcb *pcb = lp->lwp_thread->td_pcb;
735
6f7b98e0
MD
736 /* was i386_user_cleanup() in NetBSD */
737 user_ldt_free(pcb);
738
739 bzero((char *)regs, sizeof(struct trapframe));
740 regs->tf_eip = entry;
741 regs->tf_esp = stack;
742 regs->tf_eflags = PSL_USER | (regs->tf_eflags & PSL_T);
743 regs->tf_ss = 0;
744 regs->tf_ds = 0;
745 regs->tf_es = 0;
746 regs->tf_fs = 0;
4e7c41c5 747 regs->tf_gs = 0;
6f7b98e0
MD
748 regs->tf_cs = 0;
749
750 /* PS_STRINGS value for BSD/OS binaries. It is 0 for non-BSD/OS. */
751 regs->tf_ebx = ps_strings;
752
753 /*
754 * Reset the hardware debug registers if they were in use.
755 * They won't have any meaning for the newly exec'd process.
756 */
757 if (pcb->pcb_flags & PCB_DBREGS) {
758 pcb->pcb_dr0 = 0;
759 pcb->pcb_dr1 = 0;
760 pcb->pcb_dr2 = 0;
761 pcb->pcb_dr3 = 0;
762 pcb->pcb_dr6 = 0;
763 pcb->pcb_dr7 = 0;
08f2f1bb 764 if (pcb == td->td_pcb) {
6f7b98e0
MD
765 /*
766 * Clear the debug registers on the running
767 * CPU, otherwise they will end up affecting
768 * the next process we switch to.
769 */
770 reset_dbregs();
771 }
772 pcb->pcb_flags &= ~PCB_DBREGS;
773 }
774
775 /*
776 * Initialize the math emulator (if any) for the current process.
777 * Actually, just clear the bit that says that the emulator has
778 * been initialized. Initialization is delayed until the process
779 * traps to the emulator (if it is done at all) mainly because
780 * emulators don't provide an entry point for initialization.
781 */
08f2f1bb 782 pcb->pcb_flags &= ~FP_SOFTFP;
6f7b98e0
MD
783
784 /*
785 * note: do not set CR0_TS here. npxinit() must do it after clearing
786 * gd_npxthread. Otherwise a preemptive interrupt thread may panic
787 * in npxdna().
788 */
789 crit_enter();
790#if 0
791 load_cr0(rcr0() | CR0_MP);
792#endif
793
794#if NNPX > 0
795 /* Initialize the npx (if any) for the current process. */
796 npxinit(__INITIAL_NPXCW__);
797#endif
798 crit_exit();
799
800 /*
801 * note: linux emulator needs edx to be 0x0 on entry, which is
802 * handled in execve simply by setting the 64 bit syscall
803 * return value to 0.
804 */
805}
806
807void
808cpu_setregs(void)
809{
810#if 0
811 unsigned int cr0;
812
813 cr0 = rcr0();
814 cr0 |= CR0_NE; /* Done by npxinit() */
815 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
816#ifdef I386_CPU
817 if (cpu_class != CPUCLASS_386)
818#endif
819 cr0 |= CR0_WP | CR0_AM;
820 load_cr0(cr0);
821 load_gs(_udatasel);
822#endif
823}
824
825static int
826sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
827{
828 int error;
829 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
830 req);
831 if (!error && req->newptr)
832 resettodr();
833 return (error);
834}
835
836SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
837 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
838
839extern u_long bootdev; /* not a cdev_t - encoding is different */
840SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
841 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
842
843/*
844 * Initialize 386 and configure to run kernel
845 */
846
847/*
848 * Initialize segments & interrupt table
849 */
850
851extern struct user *proc0paddr;
852
853#if 0
854
855extern inthand_t
856 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
857 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
858 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
859 IDTVEC(page), IDTVEC(mchk), IDTVEC(fpu), IDTVEC(align),
860 IDTVEC(xmm), IDTVEC(syscall),
861 IDTVEC(rsvd0);
862extern inthand_t
863 IDTVEC(int0x80_syscall);
864
865#endif
866
867#ifdef DEBUG_INTERRUPTS
868extern inthand_t *Xrsvdary[256];
869#endif
870
871int
08f2f1bb 872ptrace_set_pc(struct lwp *lp, unsigned long addr)
6f7b98e0 873{
08f2f1bb 874 lp->lwp_md.md_regs->tf_eip = addr;
6f7b98e0
MD
875 return (0);
876}
877
878int
879ptrace_single_step(struct lwp *lp)
880{
881 lp->lwp_md.md_regs->tf_eflags |= PSL_T;
882 return (0);
883}
884
885int
886fill_regs(struct lwp *lp, struct reg *regs)
887{
6f7b98e0
MD
888 struct trapframe *tp;
889
890 tp = lp->lwp_md.md_regs;
4e7c41c5 891 regs->r_gs = tp->tf_gs;
6f7b98e0
MD
892 regs->r_fs = tp->tf_fs;
893 regs->r_es = tp->tf_es;
894 regs->r_ds = tp->tf_ds;
895 regs->r_edi = tp->tf_edi;
896 regs->r_esi = tp->tf_esi;
897 regs->r_ebp = tp->tf_ebp;
898 regs->r_ebx = tp->tf_ebx;
899 regs->r_edx = tp->tf_edx;
900 regs->r_ecx = tp->tf_ecx;
901 regs->r_eax = tp->tf_eax;
902 regs->r_eip = tp->tf_eip;
903 regs->r_cs = tp->tf_cs;
904 regs->r_eflags = tp->tf_eflags;
905 regs->r_esp = tp->tf_esp;
906 regs->r_ss = tp->tf_ss;
6f7b98e0
MD
907 return (0);
908}
909
910int
911set_regs(struct lwp *lp, struct reg *regs)
912{
6f7b98e0
MD
913 struct trapframe *tp;
914
915 tp = lp->lwp_md.md_regs;
916 if (!EFL_SECURE(regs->r_eflags, tp->tf_eflags) ||
917 !CS_SECURE(regs->r_cs))
918 return (EINVAL);
4e7c41c5 919 tp->tf_gs = regs->r_gs;
6f7b98e0
MD
920 tp->tf_fs = regs->r_fs;
921 tp->tf_es = regs->r_es;
922 tp->tf_ds = regs->r_ds;
923 tp->tf_edi = regs->r_edi;
924 tp->tf_esi = regs->r_esi;
925 tp->tf_ebp = regs->r_ebp;
926 tp->tf_ebx = regs->r_ebx;
927 tp->tf_edx = regs->r_edx;
928 tp->tf_ecx = regs->r_ecx;
929 tp->tf_eax = regs->r_eax;
930 tp->tf_eip = regs->r_eip;
931 tp->tf_cs = regs->r_cs;
932 tp->tf_eflags = regs->r_eflags;
933 tp->tf_esp = regs->r_esp;
934 tp->tf_ss = regs->r_ss;
6f7b98e0
MD
935 return (0);
936}
937
938#ifndef CPU_DISABLE_SSE
939static void
940fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
941{
942 struct env87 *penv_87 = &sv_87->sv_env;
943 struct envxmm *penv_xmm = &sv_xmm->sv_env;
944 int i;
945
946 /* FPU control/status */
947 penv_87->en_cw = penv_xmm->en_cw;
948 penv_87->en_sw = penv_xmm->en_sw;
949 penv_87->en_tw = penv_xmm->en_tw;
950 penv_87->en_fip = penv_xmm->en_fip;
951 penv_87->en_fcs = penv_xmm->en_fcs;
952 penv_87->en_opcode = penv_xmm->en_opcode;
953 penv_87->en_foo = penv_xmm->en_foo;
954 penv_87->en_fos = penv_xmm->en_fos;
955
956 /* FPU registers */
957 for (i = 0; i < 8; ++i)
958 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
959
960 sv_87->sv_ex_sw = sv_xmm->sv_ex_sw;
961}
962
963static void
964set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
965{
966 struct env87 *penv_87 = &sv_87->sv_env;
967 struct envxmm *penv_xmm = &sv_xmm->sv_env;
968 int i;
969
970 /* FPU control/status */
971 penv_xmm->en_cw = penv_87->en_cw;
972 penv_xmm->en_sw = penv_87->en_sw;
973 penv_xmm->en_tw = penv_87->en_tw;
974 penv_xmm->en_fip = penv_87->en_fip;
975 penv_xmm->en_fcs = penv_87->en_fcs;
976 penv_xmm->en_opcode = penv_87->en_opcode;
977 penv_xmm->en_foo = penv_87->en_foo;
978 penv_xmm->en_fos = penv_87->en_fos;
979
980 /* FPU registers */
981 for (i = 0; i < 8; ++i)
982 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
983
984 sv_xmm->sv_ex_sw = sv_87->sv_ex_sw;
985}
986#endif /* CPU_DISABLE_SSE */
987
988int
989fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
990{
991#ifndef CPU_DISABLE_SSE
992 if (cpu_fxsr) {
993 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
994 (struct save87 *)fpregs);
995 return (0);
996 }
997#endif /* CPU_DISABLE_SSE */
998 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
999 return (0);
1000}
1001
1002int
1003set_fpregs(struct lwp *lp, struct fpreg *fpregs)
1004{
1005#ifndef CPU_DISABLE_SSE
1006 if (cpu_fxsr) {
1007 set_fpregs_xmm((struct save87 *)fpregs,
1008 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
1009 return (0);
1010 }
1011#endif /* CPU_DISABLE_SSE */
1012 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
1013 return (0);
1014}
1015
1016int
1017fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
1018{
1019 if (lp == NULL) {
1020 dbregs->dr0 = rdr0();
1021 dbregs->dr1 = rdr1();
1022 dbregs->dr2 = rdr2();
1023 dbregs->dr3 = rdr3();
1024 dbregs->dr4 = rdr4();
1025 dbregs->dr5 = rdr5();
1026 dbregs->dr6 = rdr6();
1027 dbregs->dr7 = rdr7();
1028 } else {
1029 struct pcb *pcb;
1030
1031 pcb = lp->lwp_thread->td_pcb;
1032 dbregs->dr0 = pcb->pcb_dr0;
1033 dbregs->dr1 = pcb->pcb_dr1;
1034 dbregs->dr2 = pcb->pcb_dr2;
1035 dbregs->dr3 = pcb->pcb_dr3;
1036 dbregs->dr4 = 0;
1037 dbregs->dr5 = 0;
1038 dbregs->dr6 = pcb->pcb_dr6;
1039 dbregs->dr7 = pcb->pcb_dr7;
1040 }
1041 return (0);
1042}
1043
1044int
1045set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1046{
1047 if (lp == NULL) {
1048 load_dr0(dbregs->dr0);
1049 load_dr1(dbregs->dr1);
1050 load_dr2(dbregs->dr2);
1051 load_dr3(dbregs->dr3);
1052 load_dr4(dbregs->dr4);
1053 load_dr5(dbregs->dr5);
1054 load_dr6(dbregs->dr6);
1055 load_dr7(dbregs->dr7);
1056 } else {
1057 struct pcb *pcb;
1058 struct ucred *ucred;
1059 int i;
1060 uint32_t mask1, mask2;
1061
1062 /*
1063 * Don't let an illegal value for dr7 get set. Specifically,
1064 * check for undefined settings. Setting these bit patterns
1065 * result in undefined behaviour and can lead to an unexpected
1066 * TRCTRAP.
1067 */
1068 for (i = 0, mask1 = 0x3<<16, mask2 = 0x2<<16; i < 8;
1069 i++, mask1 <<= 2, mask2 <<= 2)
1070 if ((dbregs->dr7 & mask1) == mask2)
1071 return (EINVAL);
1072
1073 pcb = lp->lwp_thread->td_pcb;
1074 ucred = lp->lwp_proc->p_ucred;
1075
1076 /*
1077 * Don't let a process set a breakpoint that is not within the
1078 * process's address space. If a process could do this, it
1079 * could halt the system by setting a breakpoint in the kernel
1080 * (if ddb was enabled). Thus, we need to check to make sure
1081 * that no breakpoints are being enabled for addresses outside
1082 * process's address space, unless, perhaps, we were called by
1083 * uid 0.
1084 *
1085 * XXX - what about when the watched area of the user's
1086 * address space is written into from within the kernel
1087 * ... wouldn't that still cause a breakpoint to be generated
1088 * from within kernel mode?
1089 */
1090
1091 if (suser_cred(ucred, 0) != 0) {
1092 if (dbregs->dr7 & 0x3) {
1093 /* dr0 is enabled */
1094 if (dbregs->dr0 >= VM_MAX_USER_ADDRESS)
1095 return (EINVAL);
1096 }
1097
1098 if (dbregs->dr7 & (0x3<<2)) {
1099 /* dr1 is enabled */
1100 if (dbregs->dr1 >= VM_MAX_USER_ADDRESS)
1101 return (EINVAL);
1102 }
1103
1104 if (dbregs->dr7 & (0x3<<4)) {
1105 /* dr2 is enabled */
1106 if (dbregs->dr2 >= VM_MAX_USER_ADDRESS)
1107 return (EINVAL);
1108 }
1109
1110 if (dbregs->dr7 & (0x3<<6)) {
1111 /* dr3 is enabled */
1112 if (dbregs->dr3 >= VM_MAX_USER_ADDRESS)
1113 return (EINVAL);
1114 }
1115 }
1116
1117 pcb->pcb_dr0 = dbregs->dr0;
1118 pcb->pcb_dr1 = dbregs->dr1;
1119 pcb->pcb_dr2 = dbregs->dr2;
1120 pcb->pcb_dr3 = dbregs->dr3;
1121 pcb->pcb_dr6 = dbregs->dr6;
1122 pcb->pcb_dr7 = dbregs->dr7;
1123
1124 pcb->pcb_flags |= PCB_DBREGS;
1125 }
1126
1127 return (0);
1128}
1129
1130#if 0
1131/*
1132 * Return > 0 if a hardware breakpoint has been hit, and the
1133 * breakpoint was in user space. Return 0, otherwise.
1134 */
1135int
1136user_dbreg_trap(void)
1137{
1138 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1139 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1140 int nbp; /* number of breakpoints that triggered */
1141 caddr_t addr[4]; /* breakpoint addresses */
1142 int i;
1143
1144 dr7 = rdr7();
1145 if ((dr7 & 0x000000ff) == 0) {
1146 /*
1147 * all GE and LE bits in the dr7 register are zero,
1148 * thus the trap couldn't have been caused by the
1149 * hardware debug registers
1150 */
1151 return 0;
1152 }
1153
1154 nbp = 0;
1155 dr6 = rdr6();
1156 bp = dr6 & 0x0000000f;
1157
1158 if (!bp) {
1159 /*
1160 * None of the breakpoint bits are set meaning this
1161 * trap was not caused by any of the debug registers
1162 */
1163 return 0;
1164 }
1165
1166 /*
1167 * at least one of the breakpoints were hit, check to see
1168 * which ones and if any of them are user space addresses
1169 */
1170
1171 if (bp & 0x01) {
1172 addr[nbp++] = (caddr_t)rdr0();
1173 }
1174 if (bp & 0x02) {
1175 addr[nbp++] = (caddr_t)rdr1();
1176 }
1177 if (bp & 0x04) {
1178 addr[nbp++] = (caddr_t)rdr2();
1179 }
1180 if (bp & 0x08) {
1181 addr[nbp++] = (caddr_t)rdr3();
1182 }
1183
1184 for (i=0; i<nbp; i++) {
1185 if (addr[i] <
1186 (caddr_t)VM_MAX_USER_ADDRESS) {
1187 /*
1188 * addr[i] is in user space
1189 */
1190 return nbp;
1191 }
1192 }
1193
1194 /*
1195 * None of the breakpoints are in user space.
1196 */
1197 return 0;
1198}
1199
1200#endif
1201
1202
1203#ifndef DDB
1204void
1205Debugger(const char *msg)
1206{
1207 kprintf("Debugger(\"%s\") called.\n", msg);
1208}
1209#endif /* no DDB */
1210