kernel - Major signal path adjustments to fix races, tsleep race fixes, +more
[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 $
6f7b98e0
MD
40 */
41
6f7b98e0 42#include "use_npx.h"
6f7b98e0
MD
43#include "opt_compat.h"
44#include "opt_ddb.h"
45#include "opt_directio.h"
46#include "opt_inet.h"
47#include "opt_ipx.h"
48#include "opt_msgbuf.h"
49#include "opt_swap.h"
50
51#include <sys/param.h>
52#include <sys/systm.h>
53#include <sys/sysproto.h>
54#include <sys/signalvar.h>
55#include <sys/kernel.h>
56#include <sys/linker.h>
57#include <sys/malloc.h>
58#include <sys/proc.h>
59#include <sys/buf.h>
60#include <sys/reboot.h>
61#include <sys/mbuf.h>
62#include <sys/msgbuf.h>
63#include <sys/sysent.h>
64#include <sys/sysctl.h>
65#include <sys/vmmeter.h>
66#include <sys/bus.h>
67#include <sys/upcall.h>
68#include <sys/usched.h>
69#include <sys/reg.h>
70
71#include <vm/vm.h>
72#include <vm/vm_param.h>
73#include <sys/lock.h>
74#include <vm/vm_kern.h>
75#include <vm/vm_object.h>
76#include <vm/vm_page.h>
77#include <vm/vm_map.h>
78#include <vm/vm_pager.h>
79#include <vm/vm_extern.h>
80
81#include <sys/thread2.h>
684a93c4 82#include <sys/mplock2.h>
6f7b98e0
MD
83
84#include <sys/user.h>
85#include <sys/exec.h>
86#include <sys/cons.h>
87
88#include <ddb/ddb.h>
89
90#include <machine/cpu.h>
91#include <machine/clock.h>
92#include <machine/specialreg.h>
93#include <machine/md_var.h>
94#include <machine/pcb_ext.h> /* pcb.h included via sys/user.h */
95#include <machine/globaldata.h> /* CPU_prvspace */
96#include <machine/smp.h>
97#ifdef PERFMON
98#include <machine/perfmon.h>
99#endif
100#include <machine/cputypes.h>
101
102#include <bus/isa/rtc.h>
103#include <machine/vm86.h>
104#include <sys/random.h>
105#include <sys/ptrace.h>
106#include <machine/sigframe.h>
b402c633 107#include <unistd.h> /* umtx_* functions */
b68e846f 108#include <pthread.h> /* pthread_yield */
6f7b98e0
MD
109
110extern void dblfault_handler (void);
111
112#ifndef CPU_DISABLE_SSE
113static void set_fpregs_xmm (struct save87 *, struct savexmm *);
114static void fill_fpregs_xmm (struct savexmm *, struct save87 *);
115#endif /* CPU_DISABLE_SSE */
116#ifdef DIRECTIO
117extern void ffs_rawread_setup(void);
118#endif /* DIRECTIO */
119
120#ifdef SMP
121int64_t tsc_offsets[MAXCPU];
122#else
123int64_t tsc_offsets[1];
124#endif
125
126#if defined(SWTCH_OPTIM_STATS)
127extern int swtch_optim_stats;
128SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats,
129 CTLFLAG_RD, &swtch_optim_stats, 0, "");
130SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count,
131 CTLFLAG_RD, &tlb_flush_count, 0, "");
132#endif
133
6f7b98e0
MD
134static int
135sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
136{
39d69dae
AH
137 u_long pmem = ctob(physmem);
138
139 int error = sysctl_handle_long(oidp, &pmem, 0, req);
6f7b98e0
MD
140 return (error);
141}
142
39d69dae 143SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_ULONG|CTLFLAG_RD,
9b9532a0 144 0, 0, sysctl_hw_physmem, "LU", "Total system memory in bytes (number of pages * page size)");
6f7b98e0
MD
145
146static int
147sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
148{
149 int error = sysctl_handle_int(oidp, 0,
71152ac6 150 ctob((int)Maxmem - vmstats.v_wire_count), req);
6f7b98e0
MD
151 return (error);
152}
153
154SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD,
155 0, 0, sysctl_hw_usermem, "IU", "");
156
201b3f37 157SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, 0, "");
44afc5dc 158
6f7b98e0
MD
159#if 0
160
161static int
162sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
163{
164 int error;
165
166 /* Unwind the buffer, so that it's linear (possibly starting with
167 * some initial nulls).
168 */
169 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr,
170 msgbufp->msg_size-msgbufp->msg_bufr,req);
171 if(error) return(error);
172 if(msgbufp->msg_bufr>0) {
173 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr,
174 msgbufp->msg_bufr,req);
175 }
176 return(error);
177}
178
179SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
180 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
181
182static int msgbuf_clear;
183
184static int
185sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
186{
187 int error;
188 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
189 req);
190 if (!error && req->newptr) {
191 /* Clear the buffer and reset write pointer */
192 bzero(msgbufp->msg_ptr,msgbufp->msg_size);
193 msgbufp->msg_bufr=msgbufp->msg_bufx=0;
194 msgbuf_clear=0;
195 }
196 return (error);
197}
198
199SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
200 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
201 "Clear kernel message buffer");
202
203#endif
204
205/*
206 * Send an interrupt to process.
207 *
208 * Stack is set up to allow sigcode stored
209 * at top to call routine, followed by kcall
210 * to sigreturn routine below. After sigreturn
211 * resets the signal mask, the stack, and the
212 * frame pointer, it returns to the user
213 * specified pc, psl.
214 */
215void
216sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
217{
218 struct lwp *lp = curthread->td_lwp;
219 struct proc *p = lp->lwp_proc;
220 struct trapframe *regs;
221 struct sigacts *psp = p->p_sigacts;
222 struct sigframe sf, *sfp;
223 int oonstack;
224
225 regs = lp->lwp_md.md_regs;
226 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
227
228 /* save user context */
229 bzero(&sf, sizeof(struct sigframe));
230 sf.sf_uc.uc_sigmask = *mask;
231 sf.sf_uc.uc_stack = lp->lwp_sigstk;
232 sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
4e7c41c5 233 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_gs, sizeof(struct trapframe));
6f7b98e0 234
ce3d8c4d
MD
235 /* make the size of the saved context visible to userland */
236 sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext);
237
6f7b98e0 238 /* Allocate and validate space for the signal handler context. */
4643740a 239 if ((lp->lwp_flags & LWP_ALTSTACK) != 0 && !oonstack &&
6f7b98e0
MD
240 SIGISMEMBER(psp->ps_sigonstack, sig)) {
241 sfp = (struct sigframe *)(lp->lwp_sigstk.ss_sp +
242 lp->lwp_sigstk.ss_size - sizeof(struct sigframe));
243 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
244 }
245 else
246 sfp = (struct sigframe *)regs->tf_esp - 1;
247
248 /* Translate the signal is appropriate */
249 if (p->p_sysent->sv_sigtbl) {
250 if (sig <= p->p_sysent->sv_sigsize)
251 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
252 }
253
254 /* Build the argument list for the signal handler. */
255 sf.sf_signum = sig;
256 sf.sf_ucontext = (register_t)&sfp->sf_uc;
257 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
258 /* Signal handler installed with SA_SIGINFO. */
259 sf.sf_siginfo = (register_t)&sfp->sf_si;
260 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
261
262 /* fill siginfo structure */
263 sf.sf_si.si_signo = sig;
264 sf.sf_si.si_code = code;
265 sf.sf_si.si_addr = (void*)regs->tf_err;
266 }
267 else {
268 /* Old FreeBSD-style arguments. */
269 sf.sf_siginfo = code;
270 sf.sf_addr = regs->tf_err;
271 sf.sf_ahu.sf_handler = catcher;
272 }
273
274#if 0
275 /*
276 * If we're a vm86 process, we want to save the segment registers.
277 * We also change eflags to be our emulated eflags, not the actual
278 * eflags.
279 */
280 if (regs->tf_eflags & PSL_VM) {
281 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
282 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
283
284 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
285 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
286 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
287 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
288
289 if (vm86->vm86_has_vme == 0)
290 sf.sf_uc.uc_mcontext.mc_eflags =
291 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
292 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
293
294 /*
295 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
296 * syscalls made by the signal handler. This just avoids
297 * wasting time for our lazy fixup of such faults. PSL_NT
298 * does nothing in vm86 mode, but vm86 programs can set it
299 * almost legitimately in probes for old cpu types.
300 */
301 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
302 }
303#endif
1b251f0a
MD
304
305 /*
306 * Save the FPU state and reinit the FP unit
307 */
308 npxpush(&sf.sf_uc.uc_mcontext);
6f7b98e0
MD
309
310 /*
311 * Copy the sigframe out to the user's stack.
312 */
313 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
314 /*
315 * Something is wrong with the stack pointer.
316 * ...Kill the process.
317 */
b276424c 318 sigexit(lp, SIGILL);
6f7b98e0
MD
319 }
320
321 regs->tf_esp = (int)sfp;
322 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
8688c24a
AE
323
324 /*
325 * i386 abi specifies that the direction flag must be cleared
326 * on function entry
327 */
328 regs->tf_eflags &= ~(PSL_T|PSL_D);
329
4e7c41c5
MD
330 regs->tf_cs = _ucodesel;
331 regs->tf_ds = _udatasel;
332 regs->tf_es = _udatasel;
333 if (regs->tf_trapno == T_PROTFLT) {
334 regs->tf_fs = _udatasel;
335 regs->tf_gs = _udatasel;
336 }
337 regs->tf_ss = _udatasel;
6f7b98e0
MD
338}
339
340/*
341 * Sanitize the trapframe for a virtual kernel passing control to a custom
342 * VM context.
343 *
344 * Allow userland to set or maintain PSL_RF, the resume flag. This flag
345 * basically controls whether the return PC should skip the first instruction
346 * (as in an explicit system call) or re-execute it (as in an exception).
347 */
348int
349cpu_sanitize_frame(struct trapframe *frame)
350{
4e7c41c5
MD
351 frame->tf_cs = _ucodesel;
352 frame->tf_ds = _udatasel;
353 frame->tf_es = _udatasel;
354#if 0
355 frame->tf_fs = _udatasel;
356 frame->tf_gs = _udatasel;
357#endif
358 frame->tf_ss = _udatasel;
359 frame->tf_eflags &= (PSL_RF | PSL_USERCHANGE);
6f7b98e0
MD
360 frame->tf_eflags |= PSL_RESERVED_DEFAULT | PSL_I;
361 return(0);
362}
363
4e7c41c5
MD
364int
365cpu_sanitize_tls(struct savetls *tls)
366{
367 struct segment_descriptor *desc;
368 int i;
369
370 for (i = 0; i < NGTLS; ++i) {
371 desc = &tls->tls[i];
372 if (desc->sd_dpl == 0 && desc->sd_type == 0)
373 continue;
374 if (desc->sd_def32 == 0)
375 return(ENXIO);
376 if (desc->sd_type != SDT_MEMRWA)
377 return(ENXIO);
378 if (desc->sd_dpl != SEL_UPL)
379 return(ENXIO);
380 if (desc->sd_xx != 0 || desc->sd_p != 1)
381 return(ENXIO);
382 }
383 return(0);
384}
385
6f7b98e0
MD
386/*
387 * sigreturn(ucontext_t *sigcntxp)
388 *
389 * System call to cleanup state after a signal
390 * has been taken. Reset signal mask and
391 * stack state from context left by sendsig (above).
392 * Return to previous pc and psl as specified by
393 * context left by sendsig. Check carefully to
394 * make sure that the user has not modified the
395 * state to gain improper privileges.
3919ced0
MD
396 *
397 * MPSAFE
6f7b98e0
MD
398 */
399#define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
400#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
401
402int
403sys_sigreturn(struct sigreturn_args *uap)
404{
405 struct lwp *lp = curthread->td_lwp;
406 struct trapframe *regs;
aaf8b91f
MD
407 ucontext_t ucp;
408 int cs;
409 int eflags;
410 int error;
6f7b98e0 411
aaf8b91f
MD
412 error = copyin(uap->sigcntxp, &ucp, sizeof(ucp));
413 if (error)
414 return (error);
6f7b98e0
MD
415
416 regs = lp->lwp_md.md_regs;
aaf8b91f 417 eflags = ucp.uc_mcontext.mc_eflags;
6f7b98e0
MD
418
419#if 0
420 if (eflags & PSL_VM) {
421 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
422 struct vm86_kernel *vm86;
423
424 /*
425 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
426 * set up the vm86 area, and we can't enter vm86 mode.
427 */
428 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
429 return (EINVAL);
430 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
431 if (vm86->vm86_inited == 0)
432 return (EINVAL);
433
434 /* go back to user mode if both flags are set */
435 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
436 trapsignal(lp->lwp_proc, SIGBUS, 0);
437
438 if (vm86->vm86_has_vme) {
439 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
440 (eflags & VME_USERCHANGE) | PSL_VM;
441 } else {
442 vm86->vm86_eflags = eflags; /* save VIF, VIP */
443 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
444 }
aaf8b91f 445 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
6f7b98e0
MD
446 tf->tf_eflags = eflags;
447 tf->tf_vm86_ds = tf->tf_ds;
448 tf->tf_vm86_es = tf->tf_es;
449 tf->tf_vm86_fs = tf->tf_fs;
4e7c41c5
MD
450 tf->tf_vm86_gs = tf->tf_gs;
451 tf->tf_ds = _udatasel;
452 tf->tf_es = _udatasel;
453#if 0
454 tf->tf_fs = _udatasel;
455 tf->tf_gs = _udatasel;
456#endif
6f7b98e0
MD
457 } else
458#endif
459 {
460 /*
461 * Don't allow users to change privileged or reserved flags.
462 */
463 /*
464 * XXX do allow users to change the privileged flag PSL_RF.
465 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
466 * should sometimes set it there too. tf_eflags is kept in
467 * the signal context during signal handling and there is no
468 * other place to remember it, so the PSL_RF bit may be
469 * corrupted by the signal handler without us knowing.
470 * Corruption of the PSL_RF bit at worst causes one more or
471 * one less debugger trap, so allowing it is fairly harmless.
472 */
473 if (!EFL_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
474 kprintf("sigreturn: eflags = 0x%x\n", eflags);
475 return(EINVAL);
476 }
477
478 /*
479 * Don't allow users to load a valid privileged %cs. Let the
480 * hardware check for invalid selectors, excess privilege in
481 * other selectors, invalid %eip's and invalid %esp's.
482 */
aaf8b91f 483 cs = ucp.uc_mcontext.mc_cs;
6f7b98e0
MD
484 if (!CS_SECURE(cs)) {
485 kprintf("sigreturn: cs = 0x%x\n", cs);
08f2f1bb 486 trapsignal(lp, SIGBUS, T_PROTFLT);
6f7b98e0
MD
487 return(EINVAL);
488 }
aaf8b91f 489 bcopy(&ucp.uc_mcontext.mc_gs, regs, sizeof(struct trapframe));
6f7b98e0
MD
490 }
491
4b486183 492 /*
1b251f0a
MD
493 * Restore the FPU state from the frame
494 */
3919ced0 495 crit_enter();
1b251f0a
MD
496 npxpop(&ucp.uc_mcontext);
497
aaf8b91f 498 if (ucp.uc_mcontext.mc_onstack & 1)
6f7b98e0
MD
499 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
500 else
501 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
502
aaf8b91f 503 lp->lwp_sigmask = ucp.uc_sigmask;
6f7b98e0 504 SIG_CANTMASK(lp->lwp_sigmask);
3919ced0 505 crit_exit();
6f7b98e0
MD
506 return(EJUSTRETURN);
507}
508
509/*
510 * Stack frame on entry to function. %eax will contain the function vector,
511 * %ecx will contain the function data. flags, ecx, and eax will have
512 * already been pushed on the stack.
513 */
514struct upc_frame {
515 register_t eax;
516 register_t ecx;
517 register_t edx;
518 register_t flags;
519 register_t oldip;
520};
521
522void
523sendupcall(struct vmupcall *vu, int morepending)
524{
525 struct lwp *lp = curthread->td_lwp;
526 struct trapframe *regs;
527 struct upcall upcall;
528 struct upc_frame upc_frame;
529 int crit_count = 0;
530
531 /*
1fdc022f
MD
532 * If we are a virtual kernel running an emulated user process
533 * context, switch back to the virtual kernel context before
534 * trying to post the signal.
535 */
39005e16 536 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1fdc022f 537 lp->lwp_md.md_regs->tf_trapno = 0;
287ebb09 538 vkernel_trap(lp, lp->lwp_md.md_regs);
1fdc022f
MD
539 }
540
541 /*
6f7b98e0
MD
542 * Get the upcall data structure
543 */
544 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
545 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
546 ) {
547 vu->vu_pending = 0;
548 kprintf("bad upcall address\n");
549 return;
550 }
551
552 /*
553 * If the data structure is already marked pending or has a critical
554 * section count, mark the data structure as pending and return
555 * without doing an upcall. vu_pending is left set.
556 */
557 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
558 if (upcall.upc_pending < vu->vu_pending) {
559 upcall.upc_pending = vu->vu_pending;
560 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
561 sizeof(upcall.upc_pending));
562 }
563 return;
564 }
565
566 /*
567 * We can run this upcall now, clear vu_pending.
568 *
569 * Bump our critical section count and set or clear the
570 * user pending flag depending on whether more upcalls are
571 * pending. The user will be responsible for calling
572 * upc_dispatch(-1) to process remaining upcalls.
573 */
574 vu->vu_pending = 0;
575 upcall.upc_pending = morepending;
f9235b6d 576 ++crit_count;
6f7b98e0
MD
577 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
578 sizeof(upcall.upc_pending));
579 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
580 sizeof(int));
581
582 /*
583 * Construct a stack frame and issue the upcall
584 */
585 regs = lp->lwp_md.md_regs;
586 upc_frame.eax = regs->tf_eax;
587 upc_frame.ecx = regs->tf_ecx;
588 upc_frame.edx = regs->tf_edx;
589 upc_frame.flags = regs->tf_eflags;
590 upc_frame.oldip = regs->tf_eip;
591 if (copyout(&upc_frame, (void *)(regs->tf_esp - sizeof(upc_frame)),
592 sizeof(upc_frame)) != 0) {
593 kprintf("bad stack on upcall\n");
594 } else {
595 regs->tf_eax = (register_t)vu->vu_func;
596 regs->tf_ecx = (register_t)vu->vu_data;
597 regs->tf_edx = (register_t)lp->lwp_upcall;
598 regs->tf_eip = (register_t)vu->vu_ctx;
599 regs->tf_esp -= sizeof(upc_frame);
600 }
601}
602
603/*
604 * fetchupcall occurs in the context of a system call, which means that
605 * we have to return EJUSTRETURN in order to prevent eax and edx from
606 * being overwritten by the syscall return value.
607 *
608 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
609 * and the function pointer in %eax.
610 */
611int
612fetchupcall (struct vmupcall *vu, int morepending, void *rsp)
613{
614 struct upc_frame upc_frame;
615 struct lwp *lp = curthread->td_lwp;
616 struct trapframe *regs;
617 int error;
618 struct upcall upcall;
619 int crit_count;
620
621 regs = lp->lwp_md.md_regs;
622
623 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
624 if (error == 0) {
625 if (vu) {
626 /*
627 * This jumps us to the next ready context.
628 */
629 vu->vu_pending = 0;
630 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
631 crit_count = 0;
632 if (error == 0)
633 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
f9235b6d 634 ++crit_count;
6f7b98e0
MD
635 if (error == 0)
636 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
637 regs->tf_eax = (register_t)vu->vu_func;
638 regs->tf_ecx = (register_t)vu->vu_data;
639 regs->tf_edx = (register_t)lp->lwp_upcall;
640 regs->tf_eip = (register_t)vu->vu_ctx;
641 regs->tf_esp = (register_t)rsp;
642 } else {
643 /*
644 * This returns us to the originally interrupted code.
645 */
646 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
647 regs->tf_eax = upc_frame.eax;
648 regs->tf_ecx = upc_frame.ecx;
649 regs->tf_edx = upc_frame.edx;
650 regs->tf_eflags = (regs->tf_eflags & ~PSL_USERCHANGE) |
651 (upc_frame.flags & PSL_USERCHANGE);
652 regs->tf_eip = upc_frame.oldip;
653 regs->tf_esp = (register_t)((char *)rsp + sizeof(upc_frame));
654 }
655 }
656 if (error == 0)
657 error = EJUSTRETURN;
658 return(error);
659}
660
661/*
662 * cpu_idle() represents the idle LWKT. You cannot return from this function
663 * (unless you want to blow things up!). Instead we look for runnable threads
664 * and loop or halt as appropriate. Giant is not held on entry to the thread.
665 *
666 * The main loop is entered with a critical section held, we must release
667 * the critical section before doing anything else. lwkt_switch() will
668 * check for pending interrupts due to entering and exiting its own
669 * critical section.
670 *
671 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
b12defdc 672 * to wake a HLTed cpu up.
6f7b98e0
MD
673 */
674static int cpu_idle_hlt = 1;
675static int cpu_idle_hltcnt;
676static int cpu_idle_spincnt;
677SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
678 &cpu_idle_hlt, 0, "Idle loop HLT enable");
679SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
680 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
681SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
682 &cpu_idle_spincnt, 0, "Idle loop entry spins");
683
6f7b98e0
MD
684void
685cpu_idle(void)
686{
687 struct thread *td = curthread;
b402c633 688 struct mdglobaldata *gd = mdcpu;
c5724852 689 int reqflags;
6f7b98e0
MD
690
691 crit_exit();
f9235b6d 692 KKASSERT(td->td_critcount == 0);
792a98ed 693 cpu_enable_intr();
6f7b98e0
MD
694 for (;;) {
695 /*
696 * See if there are any LWKTs ready to go.
697 */
698 lwkt_switch();
699
700 /*
b402c633
MD
701 * The idle loop halts only if no threads are scheduleable
702 * and no signals have occured.
6f7b98e0 703 */
cbdd23b1
MD
704 if (cpu_idle_hlt &&
705 (td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
6f7b98e0 706 splz();
c5724852 707#ifdef SMP
b5d16701 708 KKASSERT(MP_LOCK_HELD() == 0);
c5724852 709#endif
cbdd23b1 710 if ((td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
e77dc05b
MD
711#ifdef DEBUGIDLE
712 struct timeval tv1, tv2;
713 gettimeofday(&tv1, NULL);
714#endif
c5724852 715 reqflags = gd->mi.gd_reqflags &
cbdd23b1 716 ~RQF_IDLECHECK_WK_MASK;
c5724852
MD
717 umtx_sleep(&gd->mi.gd_reqflags, reqflags,
718 1000000);
e77dc05b
MD
719#ifdef DEBUGIDLE
720 gettimeofday(&tv2, NULL);
721 if (tv2.tv_usec - tv1.tv_usec +
722 (tv2.tv_sec - tv1.tv_sec) * 1000000
723 > 500000) {
724 kprintf("cpu %d idlelock %08x %08x\n",
725 gd->mi.gd_cpuid,
f9235b6d 726 gd->mi.gd_reqflags,
e77dc05b
MD
727 gd->gd_fpending);
728 }
729#endif
730 }
6f7b98e0
MD
731 ++cpu_idle_hltcnt;
732 } else {
6f7b98e0
MD
733 splz();
734#ifdef SMP
6f7b98e0 735 __asm __volatile("pause");
6f7b98e0
MD
736#endif
737 ++cpu_idle_spincnt;
738 }
739 }
740}
741
b402c633
MD
742#ifdef SMP
743
744/*
06615ccb
MD
745 * Called by the spinlock code with or without a critical section held
746 * when a spinlock is found to be seriously constested.
03bcb4ab
MD
747 *
748 * We need to enter a critical section to prevent signals from recursing
749 * into pthreads.
06615ccb
MD
750 */
751void
752cpu_spinlock_contested(void)
753{
c5724852 754 cpu_pause();
06615ccb
MD
755}
756
b402c633
MD
757#endif
758
6f7b98e0
MD
759/*
760 * Clear registers on exec
761 */
762void
08f2f1bb 763exec_setregs(u_long entry, u_long stack, u_long ps_strings)
6f7b98e0 764{
08f2f1bb
SS
765 struct thread *td = curthread;
766 struct lwp *lp = td->td_lwp;
6f7b98e0
MD
767 struct trapframe *regs = lp->lwp_md.md_regs;
768 struct pcb *pcb = lp->lwp_thread->td_pcb;
769
6f7b98e0
MD
770 /* was i386_user_cleanup() in NetBSD */
771 user_ldt_free(pcb);
772
773 bzero((char *)regs, sizeof(struct trapframe));
774 regs->tf_eip = entry;
775 regs->tf_esp = stack;
776 regs->tf_eflags = PSL_USER | (regs->tf_eflags & PSL_T);
777 regs->tf_ss = 0;
778 regs->tf_ds = 0;
779 regs->tf_es = 0;
780 regs->tf_fs = 0;
4e7c41c5 781 regs->tf_gs = 0;
6f7b98e0
MD
782 regs->tf_cs = 0;
783
784 /* PS_STRINGS value for BSD/OS binaries. It is 0 for non-BSD/OS. */
785 regs->tf_ebx = ps_strings;
786
787 /*
788 * Reset the hardware debug registers if they were in use.
789 * They won't have any meaning for the newly exec'd process.
790 */
791 if (pcb->pcb_flags & PCB_DBREGS) {
792 pcb->pcb_dr0 = 0;
793 pcb->pcb_dr1 = 0;
794 pcb->pcb_dr2 = 0;
795 pcb->pcb_dr3 = 0;
796 pcb->pcb_dr6 = 0;
797 pcb->pcb_dr7 = 0;
08f2f1bb 798 if (pcb == td->td_pcb) {
6f7b98e0
MD
799 /*
800 * Clear the debug registers on the running
801 * CPU, otherwise they will end up affecting
802 * the next process we switch to.
803 */
804 reset_dbregs();
805 }
806 pcb->pcb_flags &= ~PCB_DBREGS;
807 }
808
809 /*
810 * Initialize the math emulator (if any) for the current process.
811 * Actually, just clear the bit that says that the emulator has
812 * been initialized. Initialization is delayed until the process
813 * traps to the emulator (if it is done at all) mainly because
814 * emulators don't provide an entry point for initialization.
815 */
08f2f1bb 816 pcb->pcb_flags &= ~FP_SOFTFP;
6f7b98e0
MD
817
818 /*
819 * note: do not set CR0_TS here. npxinit() must do it after clearing
820 * gd_npxthread. Otherwise a preemptive interrupt thread may panic
821 * in npxdna().
822 */
823 crit_enter();
824#if 0
825 load_cr0(rcr0() | CR0_MP);
826#endif
827
828#if NNPX > 0
829 /* Initialize the npx (if any) for the current process. */
830 npxinit(__INITIAL_NPXCW__);
831#endif
832 crit_exit();
833
834 /*
835 * note: linux emulator needs edx to be 0x0 on entry, which is
836 * handled in execve simply by setting the 64 bit syscall
837 * return value to 0.
838 */
839}
840
841void
842cpu_setregs(void)
843{
844#if 0
845 unsigned int cr0;
846
847 cr0 = rcr0();
848 cr0 |= CR0_NE; /* Done by npxinit() */
849 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
4db955e1 850 cr0 |= CR0_WP | CR0_AM;
6f7b98e0
MD
851 load_cr0(cr0);
852 load_gs(_udatasel);
853#endif
854}
855
856static int
857sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
858{
859 int error;
860 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
861 req);
862 if (!error && req->newptr)
863 resettodr();
864 return (error);
865}
866
867SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
868 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
869
870extern u_long bootdev; /* not a cdev_t - encoding is different */
871SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
872 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
873
874/*
875 * Initialize 386 and configure to run kernel
876 */
877
878/*
879 * Initialize segments & interrupt table
880 */
881
882extern struct user *proc0paddr;
883
884#if 0
885
886extern inthand_t
887 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
888 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
889 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
890 IDTVEC(page), IDTVEC(mchk), IDTVEC(fpu), IDTVEC(align),
891 IDTVEC(xmm), IDTVEC(syscall),
892 IDTVEC(rsvd0);
893extern inthand_t
894 IDTVEC(int0x80_syscall);
895
896#endif
897
898#ifdef DEBUG_INTERRUPTS
899extern inthand_t *Xrsvdary[256];
900#endif
901
902int
08f2f1bb 903ptrace_set_pc(struct lwp *lp, unsigned long addr)
6f7b98e0 904{
08f2f1bb 905 lp->lwp_md.md_regs->tf_eip = addr;
6f7b98e0
MD
906 return (0);
907}
908
909int
910ptrace_single_step(struct lwp *lp)
911{
912 lp->lwp_md.md_regs->tf_eflags |= PSL_T;
913 return (0);
914}
915
916int
917fill_regs(struct lwp *lp, struct reg *regs)
918{
6f7b98e0
MD
919 struct trapframe *tp;
920
921 tp = lp->lwp_md.md_regs;
4e7c41c5 922 regs->r_gs = tp->tf_gs;
6f7b98e0
MD
923 regs->r_fs = tp->tf_fs;
924 regs->r_es = tp->tf_es;
925 regs->r_ds = tp->tf_ds;
926 regs->r_edi = tp->tf_edi;
927 regs->r_esi = tp->tf_esi;
928 regs->r_ebp = tp->tf_ebp;
929 regs->r_ebx = tp->tf_ebx;
930 regs->r_edx = tp->tf_edx;
931 regs->r_ecx = tp->tf_ecx;
932 regs->r_eax = tp->tf_eax;
933 regs->r_eip = tp->tf_eip;
934 regs->r_cs = tp->tf_cs;
935 regs->r_eflags = tp->tf_eflags;
936 regs->r_esp = tp->tf_esp;
937 regs->r_ss = tp->tf_ss;
6f7b98e0
MD
938 return (0);
939}
940
941int
942set_regs(struct lwp *lp, struct reg *regs)
943{
6f7b98e0
MD
944 struct trapframe *tp;
945
946 tp = lp->lwp_md.md_regs;
947 if (!EFL_SECURE(regs->r_eflags, tp->tf_eflags) ||
948 !CS_SECURE(regs->r_cs))
949 return (EINVAL);
4e7c41c5 950 tp->tf_gs = regs->r_gs;
6f7b98e0
MD
951 tp->tf_fs = regs->r_fs;
952 tp->tf_es = regs->r_es;
953 tp->tf_ds = regs->r_ds;
954 tp->tf_edi = regs->r_edi;
955 tp->tf_esi = regs->r_esi;
956 tp->tf_ebp = regs->r_ebp;
957 tp->tf_ebx = regs->r_ebx;
958 tp->tf_edx = regs->r_edx;
959 tp->tf_ecx = regs->r_ecx;
960 tp->tf_eax = regs->r_eax;
961 tp->tf_eip = regs->r_eip;
962 tp->tf_cs = regs->r_cs;
963 tp->tf_eflags = regs->r_eflags;
964 tp->tf_esp = regs->r_esp;
965 tp->tf_ss = regs->r_ss;
6f7b98e0
MD
966 return (0);
967}
968
969#ifndef CPU_DISABLE_SSE
970static void
971fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
972{
973 struct env87 *penv_87 = &sv_87->sv_env;
974 struct envxmm *penv_xmm = &sv_xmm->sv_env;
975 int i;
976
977 /* FPU control/status */
978 penv_87->en_cw = penv_xmm->en_cw;
979 penv_87->en_sw = penv_xmm->en_sw;
980 penv_87->en_tw = penv_xmm->en_tw;
981 penv_87->en_fip = penv_xmm->en_fip;
982 penv_87->en_fcs = penv_xmm->en_fcs;
983 penv_87->en_opcode = penv_xmm->en_opcode;
984 penv_87->en_foo = penv_xmm->en_foo;
985 penv_87->en_fos = penv_xmm->en_fos;
986
987 /* FPU registers */
988 for (i = 0; i < 8; ++i)
989 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
6f7b98e0
MD
990}
991
992static void
993set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
994{
995 struct env87 *penv_87 = &sv_87->sv_env;
996 struct envxmm *penv_xmm = &sv_xmm->sv_env;
997 int i;
998
999 /* FPU control/status */
1000 penv_xmm->en_cw = penv_87->en_cw;
1001 penv_xmm->en_sw = penv_87->en_sw;
1002 penv_xmm->en_tw = penv_87->en_tw;
1003 penv_xmm->en_fip = penv_87->en_fip;
1004 penv_xmm->en_fcs = penv_87->en_fcs;
1005 penv_xmm->en_opcode = penv_87->en_opcode;
1006 penv_xmm->en_foo = penv_87->en_foo;
1007 penv_xmm->en_fos = penv_87->en_fos;
1008
1009 /* FPU registers */
1010 for (i = 0; i < 8; ++i)
1011 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
6f7b98e0
MD
1012}
1013#endif /* CPU_DISABLE_SSE */
1014
1015int
1016fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
1017{
1018#ifndef CPU_DISABLE_SSE
1019 if (cpu_fxsr) {
1020 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
1021 (struct save87 *)fpregs);
1022 return (0);
1023 }
1024#endif /* CPU_DISABLE_SSE */
1025 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
1026 return (0);
1027}
1028
1029int
1030set_fpregs(struct lwp *lp, struct fpreg *fpregs)
1031{
1032#ifndef CPU_DISABLE_SSE
1033 if (cpu_fxsr) {
1034 set_fpregs_xmm((struct save87 *)fpregs,
1035 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
1036 return (0);
1037 }
1038#endif /* CPU_DISABLE_SSE */
1039 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
1040 return (0);
1041}
1042
1043int
1044fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
1045{
956a62ec 1046 return (ENOSYS);
6f7b98e0
MD
1047}
1048
1049int
1050set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1051{
956a62ec 1052 return (ENOSYS);
6f7b98e0
MD
1053}
1054
1055#if 0
1056/*
1057 * Return > 0 if a hardware breakpoint has been hit, and the
1058 * breakpoint was in user space. Return 0, otherwise.
1059 */
1060int
1061user_dbreg_trap(void)
1062{
1063 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1064 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1065 int nbp; /* number of breakpoints that triggered */
1066 caddr_t addr[4]; /* breakpoint addresses */
1067 int i;
1068
1069 dr7 = rdr7();
1070 if ((dr7 & 0x000000ff) == 0) {
1071 /*
1072 * all GE and LE bits in the dr7 register are zero,
1073 * thus the trap couldn't have been caused by the
1074 * hardware debug registers
1075 */
1076 return 0;
1077 }
1078
1079 nbp = 0;
1080 dr6 = rdr6();
1081 bp = dr6 & 0x0000000f;
1082
1083 if (!bp) {
1084 /*
1085 * None of the breakpoint bits are set meaning this
1086 * trap was not caused by any of the debug registers
1087 */
1088 return 0;
1089 }
1090
1091 /*
1092 * at least one of the breakpoints were hit, check to see
1093 * which ones and if any of them are user space addresses
1094 */
1095
1096 if (bp & 0x01) {
1097 addr[nbp++] = (caddr_t)rdr0();
1098 }
1099 if (bp & 0x02) {
1100 addr[nbp++] = (caddr_t)rdr1();
1101 }
1102 if (bp & 0x04) {
1103 addr[nbp++] = (caddr_t)rdr2();
1104 }
1105 if (bp & 0x08) {
1106 addr[nbp++] = (caddr_t)rdr3();
1107 }
1108
1109 for (i=0; i<nbp; i++) {
1110 if (addr[i] <
1111 (caddr_t)VM_MAX_USER_ADDRESS) {
1112 /*
1113 * addr[i] is in user space
1114 */
1115 return nbp;
1116 }
1117 }
1118
1119 /*
1120 * None of the breakpoints are in user space.
1121 */
1122 return 0;
1123}
1124
1125#endif
1126
514621ec
SS
1127void
1128identcpu(void)
1129{
1130 int regs[4];
1131
1132 do_cpuid(1, regs);
1133 cpu_feature = regs[3];
1134}
1135
6f7b98e0
MD
1136
1137#ifndef DDB
1138void
1139Debugger(const char *msg)
1140{
1141 kprintf("Debugger(\"%s\") called.\n", msg);
1142}
1143#endif /* no DDB */
1144