* Remove some zoneinfo files from 'make upgrade' that were re-added 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 $
60297eb4 40 * $DragonFly: src/sys/platform/vkernel/i386/cpu_regs.c,v 1.3 2007/01/07 00:44:30 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
156#if 0
157
158static int
159sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
160{
161 int error;
162
163 /* Unwind the buffer, so that it's linear (possibly starting with
164 * some initial nulls).
165 */
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);
172 }
173 return(error);
174}
175
176SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
177 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
178
179static int msgbuf_clear;
180
181static int
182sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
183{
184 int error;
185 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
186 req);
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;
191 msgbuf_clear=0;
192 }
193 return (error);
194}
195
196SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
197 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
198 "Clear kernel message buffer");
199
200#endif
201
202/*
203 * Send an interrupt to process.
204 *
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
210 * specified pc, psl.
211 */
212void
213sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
214{
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;
220 int oonstack;
221
222 regs = lp->lwp_md.md_regs;
223 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
224
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 sf.sf_uc.uc_mcontext.mc_gs = rgs();
231 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_fs, sizeof(struct trapframe));
232
233 /* Allocate and validate space for the signal handler context. */
234 /* XXX lwp flags */
235 if ((p->p_flag & P_ALTSTACK) != 0 && !oonstack &&
236 SIGISMEMBER(psp->ps_sigonstack, sig)) {
237 sfp = (struct sigframe *)(lp->lwp_sigstk.ss_sp +
238 lp->lwp_sigstk.ss_size - sizeof(struct sigframe));
239 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
240 }
241 else
242 sfp = (struct sigframe *)regs->tf_esp - 1;
243
244 /* Translate the signal is appropriate */
245 if (p->p_sysent->sv_sigtbl) {
246 if (sig <= p->p_sysent->sv_sigsize)
247 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
248 }
249
250 /* Build the argument list for the signal handler. */
251 sf.sf_signum = sig;
252 sf.sf_ucontext = (register_t)&sfp->sf_uc;
253 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
254 /* Signal handler installed with SA_SIGINFO. */
255 sf.sf_siginfo = (register_t)&sfp->sf_si;
256 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
257
258 /* fill siginfo structure */
259 sf.sf_si.si_signo = sig;
260 sf.sf_si.si_code = code;
261 sf.sf_si.si_addr = (void*)regs->tf_err;
262 }
263 else {
264 /* Old FreeBSD-style arguments. */
265 sf.sf_siginfo = code;
266 sf.sf_addr = regs->tf_err;
267 sf.sf_ahu.sf_handler = catcher;
268 }
269
270#if 0
271 /*
272 * If we're a vm86 process, we want to save the segment registers.
273 * We also change eflags to be our emulated eflags, not the actual
274 * eflags.
275 */
276 if (regs->tf_eflags & PSL_VM) {
277 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
278 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
279
280 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
281 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
282 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
283 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
284
285 if (vm86->vm86_has_vme == 0)
286 sf.sf_uc.uc_mcontext.mc_eflags =
287 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
288 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
289
290 /*
291 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
292 * syscalls made by the signal handler. This just avoids
293 * wasting time for our lazy fixup of such faults. PSL_NT
294 * does nothing in vm86 mode, but vm86 programs can set it
295 * almost legitimately in probes for old cpu types.
296 */
297 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
298 }
299#endif
300
301 /*
302 * Copy the sigframe out to the user's stack.
303 */
304 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
305 /*
306 * Something is wrong with the stack pointer.
307 * ...Kill the process.
308 */
309 sigexit(p, SIGILL);
310 }
311
312 regs->tf_esp = (int)sfp;
313 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
314 regs->tf_eflags &= ~PSL_T;
315 regs->tf_cs = 0;
316 regs->tf_ds = 0;
317 regs->tf_es = 0;
318 regs->tf_fs = 0;
319 regs->tf_ss = 0;
320}
321
322/*
323 * Sanitize the trapframe for a virtual kernel passing control to a custom
324 * VM context.
325 *
326 * Allow userland to set or maintain PSL_RF, the resume flag. This flag
327 * basically controls whether the return PC should skip the first instruction
328 * (as in an explicit system call) or re-execute it (as in an exception).
329 */
330int
331cpu_sanitize_frame(struct trapframe *frame)
332{
333 frame->tf_cs = 0;
334 frame->tf_ds = 0;
335 frame->tf_es = 0;
336 frame->tf_fs = 0;
337 frame->tf_ss = 0;
338 frame->tf_eflags &= (PSL_USER | PSL_RF);
339 frame->tf_eflags |= PSL_RESERVED_DEFAULT | PSL_I;
340 return(0);
341}
342
343/*
344 * sigreturn(ucontext_t *sigcntxp)
345 *
346 * System call to cleanup state after a signal
347 * has been taken. Reset signal mask and
348 * stack state from context left by sendsig (above).
349 * Return to previous pc and psl as specified by
350 * context left by sendsig. Check carefully to
351 * make sure that the user has not modified the
352 * state to gain improper privileges.
353 */
354#define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
355#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
356
357int
358sys_sigreturn(struct sigreturn_args *uap)
359{
360 struct lwp *lp = curthread->td_lwp;
361 struct trapframe *regs;
362 ucontext_t *ucp;
363 int cs, eflags;
364
365 ucp = uap->sigcntxp;
366
367 if (!useracc((caddr_t)ucp, sizeof(ucontext_t), VM_PROT_READ))
368 return (EFAULT);
369
370 regs = lp->lwp_md.md_regs;
371 eflags = ucp->uc_mcontext.mc_eflags;
372
373#if 0
374 if (eflags & PSL_VM) {
375 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
376 struct vm86_kernel *vm86;
377
378 /*
379 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
380 * set up the vm86 area, and we can't enter vm86 mode.
381 */
382 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
383 return (EINVAL);
384 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
385 if (vm86->vm86_inited == 0)
386 return (EINVAL);
387
388 /* go back to user mode if both flags are set */
389 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
390 trapsignal(lp->lwp_proc, SIGBUS, 0);
391
392 if (vm86->vm86_has_vme) {
393 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
394 (eflags & VME_USERCHANGE) | PSL_VM;
395 } else {
396 vm86->vm86_eflags = eflags; /* save VIF, VIP */
397 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
398 }
399 bcopy(&ucp->uc_mcontext.mc_fs, tf, sizeof(struct trapframe));
400 tf->tf_eflags = eflags;
401 tf->tf_vm86_ds = tf->tf_ds;
402 tf->tf_vm86_es = tf->tf_es;
403 tf->tf_vm86_fs = tf->tf_fs;
404 tf->tf_vm86_gs = ucp->uc_mcontext.mc_gs;
405 tf->tf_ds = 0;
406 tf->tf_es = 0;
407 tf->tf_fs = 0;
408 } else
409#endif
410 {
411 /*
412 * Don't allow users to change privileged or reserved flags.
413 */
414 /*
415 * XXX do allow users to change the privileged flag PSL_RF.
416 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
417 * should sometimes set it there too. tf_eflags is kept in
418 * the signal context during signal handling and there is no
419 * other place to remember it, so the PSL_RF bit may be
420 * corrupted by the signal handler without us knowing.
421 * Corruption of the PSL_RF bit at worst causes one more or
422 * one less debugger trap, so allowing it is fairly harmless.
423 */
424 if (!EFL_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
425 kprintf("sigreturn: eflags = 0x%x\n", eflags);
426 return(EINVAL);
427 }
428
429 /*
430 * Don't allow users to load a valid privileged %cs. Let the
431 * hardware check for invalid selectors, excess privilege in
432 * other selectors, invalid %eip's and invalid %esp's.
433 */
434 cs = ucp->uc_mcontext.mc_cs;
435 if (!CS_SECURE(cs)) {
436 kprintf("sigreturn: cs = 0x%x\n", cs);
437 trapsignal(lp->lwp_proc, SIGBUS, T_PROTFLT);
438 return(EINVAL);
439 }
440 bcopy(&ucp->uc_mcontext.mc_fs, regs, sizeof(struct trapframe));
441 }
442
443 if (ucp->uc_mcontext.mc_onstack & 1)
444 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
445 else
446 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
447
448 lp->lwp_sigmask = ucp->uc_sigmask;
449 SIG_CANTMASK(lp->lwp_sigmask);
450 return(EJUSTRETURN);
451}
452
453/*
454 * Stack frame on entry to function. %eax will contain the function vector,
455 * %ecx will contain the function data. flags, ecx, and eax will have
456 * already been pushed on the stack.
457 */
458struct upc_frame {
459 register_t eax;
460 register_t ecx;
461 register_t edx;
462 register_t flags;
463 register_t oldip;
464};
465
466void
467sendupcall(struct vmupcall *vu, int morepending)
468{
469 struct lwp *lp = curthread->td_lwp;
470 struct trapframe *regs;
471 struct upcall upcall;
472 struct upc_frame upc_frame;
473 int crit_count = 0;
474
475 /*
476 * Get the upcall data structure
477 */
478 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
479 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
480 ) {
481 vu->vu_pending = 0;
482 kprintf("bad upcall address\n");
483 return;
484 }
485
486 /*
487 * If the data structure is already marked pending or has a critical
488 * section count, mark the data structure as pending and return
489 * without doing an upcall. vu_pending is left set.
490 */
491 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
492 if (upcall.upc_pending < vu->vu_pending) {
493 upcall.upc_pending = vu->vu_pending;
494 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
495 sizeof(upcall.upc_pending));
496 }
497 return;
498 }
499
500 /*
501 * We can run this upcall now, clear vu_pending.
502 *
503 * Bump our critical section count and set or clear the
504 * user pending flag depending on whether more upcalls are
505 * pending. The user will be responsible for calling
506 * upc_dispatch(-1) to process remaining upcalls.
507 */
508 vu->vu_pending = 0;
509 upcall.upc_pending = morepending;
510 crit_count += TDPRI_CRIT;
511 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
512 sizeof(upcall.upc_pending));
513 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
514 sizeof(int));
515
516 /*
517 * Construct a stack frame and issue the upcall
518 */
519 regs = lp->lwp_md.md_regs;
520 upc_frame.eax = regs->tf_eax;
521 upc_frame.ecx = regs->tf_ecx;
522 upc_frame.edx = regs->tf_edx;
523 upc_frame.flags = regs->tf_eflags;
524 upc_frame.oldip = regs->tf_eip;
525 if (copyout(&upc_frame, (void *)(regs->tf_esp - sizeof(upc_frame)),
526 sizeof(upc_frame)) != 0) {
527 kprintf("bad stack on upcall\n");
528 } else {
529 regs->tf_eax = (register_t)vu->vu_func;
530 regs->tf_ecx = (register_t)vu->vu_data;
531 regs->tf_edx = (register_t)lp->lwp_upcall;
532 regs->tf_eip = (register_t)vu->vu_ctx;
533 regs->tf_esp -= sizeof(upc_frame);
534 }
535}
536
537/*
538 * fetchupcall occurs in the context of a system call, which means that
539 * we have to return EJUSTRETURN in order to prevent eax and edx from
540 * being overwritten by the syscall return value.
541 *
542 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
543 * and the function pointer in %eax.
544 */
545int
546fetchupcall (struct vmupcall *vu, int morepending, void *rsp)
547{
548 struct upc_frame upc_frame;
549 struct lwp *lp = curthread->td_lwp;
550 struct trapframe *regs;
551 int error;
552 struct upcall upcall;
553 int crit_count;
554
555 regs = lp->lwp_md.md_regs;
556
557 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
558 if (error == 0) {
559 if (vu) {
560 /*
561 * This jumps us to the next ready context.
562 */
563 vu->vu_pending = 0;
564 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
565 crit_count = 0;
566 if (error == 0)
567 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
568 crit_count += TDPRI_CRIT;
569 if (error == 0)
570 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
571 regs->tf_eax = (register_t)vu->vu_func;
572 regs->tf_ecx = (register_t)vu->vu_data;
573 regs->tf_edx = (register_t)lp->lwp_upcall;
574 regs->tf_eip = (register_t)vu->vu_ctx;
575 regs->tf_esp = (register_t)rsp;
576 } else {
577 /*
578 * This returns us to the originally interrupted code.
579 */
580 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
581 regs->tf_eax = upc_frame.eax;
582 regs->tf_ecx = upc_frame.ecx;
583 regs->tf_edx = upc_frame.edx;
584 regs->tf_eflags = (regs->tf_eflags & ~PSL_USERCHANGE) |
585 (upc_frame.flags & PSL_USERCHANGE);
586 regs->tf_eip = upc_frame.oldip;
587 regs->tf_esp = (register_t)((char *)rsp + sizeof(upc_frame));
588 }
589 }
590 if (error == 0)
591 error = EJUSTRETURN;
592 return(error);
593}
594
595/*
596 * cpu_idle() represents the idle LWKT. You cannot return from this function
597 * (unless you want to blow things up!). Instead we look for runnable threads
598 * and loop or halt as appropriate. Giant is not held on entry to the thread.
599 *
600 * The main loop is entered with a critical section held, we must release
601 * the critical section before doing anything else. lwkt_switch() will
602 * check for pending interrupts due to entering and exiting its own
603 * critical section.
604 *
605 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
606 * to wake a HLTed cpu up. However, there are cases where the idlethread
607 * will be entered with the possibility that no IPI will occur and in such
608 * cases lwkt_switch() sets TDF_IDLE_NOHLT.
609 */
610static int cpu_idle_hlt = 1;
611static int cpu_idle_hltcnt;
612static int cpu_idle_spincnt;
613SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
614 &cpu_idle_hlt, 0, "Idle loop HLT enable");
615SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
616 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
617SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
618 &cpu_idle_spincnt, 0, "Idle loop entry spins");
619
620static void
621cpu_idle_default_hook(void)
622{
623 /*
624 * We must guarentee that hlt is exactly the instruction
625 * following the sti.
626 */
60297eb4 627 kprintf("idle halt\n");
6f7b98e0
MD
628 __asm __volatile("hlt"); /* sti; hlt */
629}
630
631/* Other subsystems (e.g., ACPI) can hook this later. */
632void (*cpu_idle_hook)(void) = cpu_idle_default_hook;
633
634void
635cpu_idle(void)
636{
637 struct thread *td = curthread;
638
639 crit_exit();
640 KKASSERT(td->td_pri < TDPRI_CRIT);
641 for (;;) {
642 /*
643 * See if there are any LWKTs ready to go.
644 */
645 lwkt_switch();
646
647 /*
648 * If we are going to halt call splz unconditionally after
649 * CLIing to catch any interrupt races. Note that we are
650 * at SPL0 and interrupts are enabled.
651 */
652 if (cpu_idle_hlt && !lwkt_runnable() &&
653 (td->td_flags & TDF_IDLE_NOHLT) == 0) {
654 /* __asm __volatile("cli"); */
655 splz();
656 if (!lwkt_runnable())
657 cpu_idle_hook();
658#ifdef SMP
659 else
660 __asm __volatile("pause");
661#endif
662 ++cpu_idle_hltcnt;
663 } else {
664 td->td_flags &= ~TDF_IDLE_NOHLT;
665 splz();
666#ifdef SMP
667 /*__asm __volatile("sti; pause");*/
668 __asm __volatile("pause");
669#else
670 /*__asm __volatile("sti");*/
671#endif
672 ++cpu_idle_spincnt;
673 }
674 }
675}
676
677/*
678 * Clear registers on exec
679 */
680void
681setregs(struct lwp *lp, u_long entry, u_long stack, u_long ps_strings)
682{
683 struct trapframe *regs = lp->lwp_md.md_regs;
684 struct pcb *pcb = lp->lwp_thread->td_pcb;
685
686 /* Reset pc->pcb_gs and %gs before possibly invalidating it. */
687 pcb->pcb_gs = 0;
688#if 0
689 load_gs(_udatasel);
690#endif
691
692 /* was i386_user_cleanup() in NetBSD */
693 user_ldt_free(pcb);
694
695 bzero((char *)regs, sizeof(struct trapframe));
696 regs->tf_eip = entry;
697 regs->tf_esp = stack;
698 regs->tf_eflags = PSL_USER | (regs->tf_eflags & PSL_T);
699 regs->tf_ss = 0;
700 regs->tf_ds = 0;
701 regs->tf_es = 0;
702 regs->tf_fs = 0;
703 regs->tf_cs = 0;
704
705 /* PS_STRINGS value for BSD/OS binaries. It is 0 for non-BSD/OS. */
706 regs->tf_ebx = ps_strings;
707
708 /*
709 * Reset the hardware debug registers if they were in use.
710 * They won't have any meaning for the newly exec'd process.
711 */
712 if (pcb->pcb_flags & PCB_DBREGS) {
713 pcb->pcb_dr0 = 0;
714 pcb->pcb_dr1 = 0;
715 pcb->pcb_dr2 = 0;
716 pcb->pcb_dr3 = 0;
717 pcb->pcb_dr6 = 0;
718 pcb->pcb_dr7 = 0;
719 if (pcb == curthread->td_pcb) {
720 /*
721 * Clear the debug registers on the running
722 * CPU, otherwise they will end up affecting
723 * the next process we switch to.
724 */
725 reset_dbregs();
726 }
727 pcb->pcb_flags &= ~PCB_DBREGS;
728 }
729
730 /*
731 * Initialize the math emulator (if any) for the current process.
732 * Actually, just clear the bit that says that the emulator has
733 * been initialized. Initialization is delayed until the process
734 * traps to the emulator (if it is done at all) mainly because
735 * emulators don't provide an entry point for initialization.
736 */
737 lp->lwp_thread->td_pcb->pcb_flags &= ~FP_SOFTFP;
738
739 /*
740 * note: do not set CR0_TS here. npxinit() must do it after clearing
741 * gd_npxthread. Otherwise a preemptive interrupt thread may panic
742 * in npxdna().
743 */
744 crit_enter();
745#if 0
746 load_cr0(rcr0() | CR0_MP);
747#endif
748
749#if NNPX > 0
750 /* Initialize the npx (if any) for the current process. */
751 npxinit(__INITIAL_NPXCW__);
752#endif
753 crit_exit();
754
755 /*
756 * note: linux emulator needs edx to be 0x0 on entry, which is
757 * handled in execve simply by setting the 64 bit syscall
758 * return value to 0.
759 */
760}
761
762void
763cpu_setregs(void)
764{
765#if 0
766 unsigned int cr0;
767
768 cr0 = rcr0();
769 cr0 |= CR0_NE; /* Done by npxinit() */
770 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
771#ifdef I386_CPU
772 if (cpu_class != CPUCLASS_386)
773#endif
774 cr0 |= CR0_WP | CR0_AM;
775 load_cr0(cr0);
776 load_gs(_udatasel);
777#endif
778}
779
780static int
781sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
782{
783 int error;
784 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
785 req);
786 if (!error && req->newptr)
787 resettodr();
788 return (error);
789}
790
791SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
792 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
793
794extern u_long bootdev; /* not a cdev_t - encoding is different */
795SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
796 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
797
798/*
799 * Initialize 386 and configure to run kernel
800 */
801
802/*
803 * Initialize segments & interrupt table
804 */
805
806extern struct user *proc0paddr;
807
808#if 0
809
810extern inthand_t
811 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
812 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
813 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
814 IDTVEC(page), IDTVEC(mchk), IDTVEC(fpu), IDTVEC(align),
815 IDTVEC(xmm), IDTVEC(syscall),
816 IDTVEC(rsvd0);
817extern inthand_t
818 IDTVEC(int0x80_syscall);
819
820#endif
821
822#ifdef DEBUG_INTERRUPTS
823extern inthand_t *Xrsvdary[256];
824#endif
825
826int
827ptrace_set_pc(struct proc *p, unsigned long addr)
828{
829 p->p_md.md_regs->tf_eip = addr;
830 return (0);
831}
832
833int
834ptrace_single_step(struct lwp *lp)
835{
836 lp->lwp_md.md_regs->tf_eflags |= PSL_T;
837 return (0);
838}
839
840int
841fill_regs(struct lwp *lp, struct reg *regs)
842{
843 struct pcb *pcb;
844 struct trapframe *tp;
845
846 tp = lp->lwp_md.md_regs;
847 regs->r_fs = tp->tf_fs;
848 regs->r_es = tp->tf_es;
849 regs->r_ds = tp->tf_ds;
850 regs->r_edi = tp->tf_edi;
851 regs->r_esi = tp->tf_esi;
852 regs->r_ebp = tp->tf_ebp;
853 regs->r_ebx = tp->tf_ebx;
854 regs->r_edx = tp->tf_edx;
855 regs->r_ecx = tp->tf_ecx;
856 regs->r_eax = tp->tf_eax;
857 regs->r_eip = tp->tf_eip;
858 regs->r_cs = tp->tf_cs;
859 regs->r_eflags = tp->tf_eflags;
860 regs->r_esp = tp->tf_esp;
861 regs->r_ss = tp->tf_ss;
862 pcb = lp->lwp_thread->td_pcb;
863 regs->r_gs = pcb->pcb_gs;
864 return (0);
865}
866
867int
868set_regs(struct lwp *lp, struct reg *regs)
869{
870 struct pcb *pcb;
871 struct trapframe *tp;
872
873 tp = lp->lwp_md.md_regs;
874 if (!EFL_SECURE(regs->r_eflags, tp->tf_eflags) ||
875 !CS_SECURE(regs->r_cs))
876 return (EINVAL);
877 tp->tf_fs = regs->r_fs;
878 tp->tf_es = regs->r_es;
879 tp->tf_ds = regs->r_ds;
880 tp->tf_edi = regs->r_edi;
881 tp->tf_esi = regs->r_esi;
882 tp->tf_ebp = regs->r_ebp;
883 tp->tf_ebx = regs->r_ebx;
884 tp->tf_edx = regs->r_edx;
885 tp->tf_ecx = regs->r_ecx;
886 tp->tf_eax = regs->r_eax;
887 tp->tf_eip = regs->r_eip;
888 tp->tf_cs = regs->r_cs;
889 tp->tf_eflags = regs->r_eflags;
890 tp->tf_esp = regs->r_esp;
891 tp->tf_ss = regs->r_ss;
892 pcb = lp->lwp_thread->td_pcb;
893 pcb->pcb_gs = regs->r_gs;
894 return (0);
895}
896
897#ifndef CPU_DISABLE_SSE
898static void
899fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
900{
901 struct env87 *penv_87 = &sv_87->sv_env;
902 struct envxmm *penv_xmm = &sv_xmm->sv_env;
903 int i;
904
905 /* FPU control/status */
906 penv_87->en_cw = penv_xmm->en_cw;
907 penv_87->en_sw = penv_xmm->en_sw;
908 penv_87->en_tw = penv_xmm->en_tw;
909 penv_87->en_fip = penv_xmm->en_fip;
910 penv_87->en_fcs = penv_xmm->en_fcs;
911 penv_87->en_opcode = penv_xmm->en_opcode;
912 penv_87->en_foo = penv_xmm->en_foo;
913 penv_87->en_fos = penv_xmm->en_fos;
914
915 /* FPU registers */
916 for (i = 0; i < 8; ++i)
917 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
918
919 sv_87->sv_ex_sw = sv_xmm->sv_ex_sw;
920}
921
922static void
923set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
924{
925 struct env87 *penv_87 = &sv_87->sv_env;
926 struct envxmm *penv_xmm = &sv_xmm->sv_env;
927 int i;
928
929 /* FPU control/status */
930 penv_xmm->en_cw = penv_87->en_cw;
931 penv_xmm->en_sw = penv_87->en_sw;
932 penv_xmm->en_tw = penv_87->en_tw;
933 penv_xmm->en_fip = penv_87->en_fip;
934 penv_xmm->en_fcs = penv_87->en_fcs;
935 penv_xmm->en_opcode = penv_87->en_opcode;
936 penv_xmm->en_foo = penv_87->en_foo;
937 penv_xmm->en_fos = penv_87->en_fos;
938
939 /* FPU registers */
940 for (i = 0; i < 8; ++i)
941 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
942
943 sv_xmm->sv_ex_sw = sv_87->sv_ex_sw;
944}
945#endif /* CPU_DISABLE_SSE */
946
947int
948fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
949{
950#ifndef CPU_DISABLE_SSE
951 if (cpu_fxsr) {
952 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
953 (struct save87 *)fpregs);
954 return (0);
955 }
956#endif /* CPU_DISABLE_SSE */
957 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
958 return (0);
959}
960
961int
962set_fpregs(struct lwp *lp, struct fpreg *fpregs)
963{
964#ifndef CPU_DISABLE_SSE
965 if (cpu_fxsr) {
966 set_fpregs_xmm((struct save87 *)fpregs,
967 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
968 return (0);
969 }
970#endif /* CPU_DISABLE_SSE */
971 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
972 return (0);
973}
974
975int
976fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
977{
978 if (lp == NULL) {
979 dbregs->dr0 = rdr0();
980 dbregs->dr1 = rdr1();
981 dbregs->dr2 = rdr2();
982 dbregs->dr3 = rdr3();
983 dbregs->dr4 = rdr4();
984 dbregs->dr5 = rdr5();
985 dbregs->dr6 = rdr6();
986 dbregs->dr7 = rdr7();
987 } else {
988 struct pcb *pcb;
989
990 pcb = lp->lwp_thread->td_pcb;
991 dbregs->dr0 = pcb->pcb_dr0;
992 dbregs->dr1 = pcb->pcb_dr1;
993 dbregs->dr2 = pcb->pcb_dr2;
994 dbregs->dr3 = pcb->pcb_dr3;
995 dbregs->dr4 = 0;
996 dbregs->dr5 = 0;
997 dbregs->dr6 = pcb->pcb_dr6;
998 dbregs->dr7 = pcb->pcb_dr7;
999 }
1000 return (0);
1001}
1002
1003int
1004set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1005{
1006 if (lp == NULL) {
1007 load_dr0(dbregs->dr0);
1008 load_dr1(dbregs->dr1);
1009 load_dr2(dbregs->dr2);
1010 load_dr3(dbregs->dr3);
1011 load_dr4(dbregs->dr4);
1012 load_dr5(dbregs->dr5);
1013 load_dr6(dbregs->dr6);
1014 load_dr7(dbregs->dr7);
1015 } else {
1016 struct pcb *pcb;
1017 struct ucred *ucred;
1018 int i;
1019 uint32_t mask1, mask2;
1020
1021 /*
1022 * Don't let an illegal value for dr7 get set. Specifically,
1023 * check for undefined settings. Setting these bit patterns
1024 * result in undefined behaviour and can lead to an unexpected
1025 * TRCTRAP.
1026 */
1027 for (i = 0, mask1 = 0x3<<16, mask2 = 0x2<<16; i < 8;
1028 i++, mask1 <<= 2, mask2 <<= 2)
1029 if ((dbregs->dr7 & mask1) == mask2)
1030 return (EINVAL);
1031
1032 pcb = lp->lwp_thread->td_pcb;
1033 ucred = lp->lwp_proc->p_ucred;
1034
1035 /*
1036 * Don't let a process set a breakpoint that is not within the
1037 * process's address space. If a process could do this, it
1038 * could halt the system by setting a breakpoint in the kernel
1039 * (if ddb was enabled). Thus, we need to check to make sure
1040 * that no breakpoints are being enabled for addresses outside
1041 * process's address space, unless, perhaps, we were called by
1042 * uid 0.
1043 *
1044 * XXX - what about when the watched area of the user's
1045 * address space is written into from within the kernel
1046 * ... wouldn't that still cause a breakpoint to be generated
1047 * from within kernel mode?
1048 */
1049
1050 if (suser_cred(ucred, 0) != 0) {
1051 if (dbregs->dr7 & 0x3) {
1052 /* dr0 is enabled */
1053 if (dbregs->dr0 >= VM_MAX_USER_ADDRESS)
1054 return (EINVAL);
1055 }
1056
1057 if (dbregs->dr7 & (0x3<<2)) {
1058 /* dr1 is enabled */
1059 if (dbregs->dr1 >= VM_MAX_USER_ADDRESS)
1060 return (EINVAL);
1061 }
1062
1063 if (dbregs->dr7 & (0x3<<4)) {
1064 /* dr2 is enabled */
1065 if (dbregs->dr2 >= VM_MAX_USER_ADDRESS)
1066 return (EINVAL);
1067 }
1068
1069 if (dbregs->dr7 & (0x3<<6)) {
1070 /* dr3 is enabled */
1071 if (dbregs->dr3 >= VM_MAX_USER_ADDRESS)
1072 return (EINVAL);
1073 }
1074 }
1075
1076 pcb->pcb_dr0 = dbregs->dr0;
1077 pcb->pcb_dr1 = dbregs->dr1;
1078 pcb->pcb_dr2 = dbregs->dr2;
1079 pcb->pcb_dr3 = dbregs->dr3;
1080 pcb->pcb_dr6 = dbregs->dr6;
1081 pcb->pcb_dr7 = dbregs->dr7;
1082
1083 pcb->pcb_flags |= PCB_DBREGS;
1084 }
1085
1086 return (0);
1087}
1088
1089#if 0
1090/*
1091 * Return > 0 if a hardware breakpoint has been hit, and the
1092 * breakpoint was in user space. Return 0, otherwise.
1093 */
1094int
1095user_dbreg_trap(void)
1096{
1097 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1098 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1099 int nbp; /* number of breakpoints that triggered */
1100 caddr_t addr[4]; /* breakpoint addresses */
1101 int i;
1102
1103 dr7 = rdr7();
1104 if ((dr7 & 0x000000ff) == 0) {
1105 /*
1106 * all GE and LE bits in the dr7 register are zero,
1107 * thus the trap couldn't have been caused by the
1108 * hardware debug registers
1109 */
1110 return 0;
1111 }
1112
1113 nbp = 0;
1114 dr6 = rdr6();
1115 bp = dr6 & 0x0000000f;
1116
1117 if (!bp) {
1118 /*
1119 * None of the breakpoint bits are set meaning this
1120 * trap was not caused by any of the debug registers
1121 */
1122 return 0;
1123 }
1124
1125 /*
1126 * at least one of the breakpoints were hit, check to see
1127 * which ones and if any of them are user space addresses
1128 */
1129
1130 if (bp & 0x01) {
1131 addr[nbp++] = (caddr_t)rdr0();
1132 }
1133 if (bp & 0x02) {
1134 addr[nbp++] = (caddr_t)rdr1();
1135 }
1136 if (bp & 0x04) {
1137 addr[nbp++] = (caddr_t)rdr2();
1138 }
1139 if (bp & 0x08) {
1140 addr[nbp++] = (caddr_t)rdr3();
1141 }
1142
1143 for (i=0; i<nbp; i++) {
1144 if (addr[i] <
1145 (caddr_t)VM_MAX_USER_ADDRESS) {
1146 /*
1147 * addr[i] is in user space
1148 */
1149 return nbp;
1150 }
1151 }
1152
1153 /*
1154 * None of the breakpoints are in user space.
1155 */
1156 return 0;
1157}
1158
1159#endif
1160
1161
1162#ifndef DDB
1163void
1164Debugger(const char *msg)
1165{
1166 kprintf("Debugger(\"%s\") called.\n", msg);
1167}
1168#endif /* no DDB */
1169
1170#include <sys/disklabel.h>
1171
1172/*
1173 * Determine the size of the transfer, and make sure it is
1174 * within the boundaries of the partition. Adjust transfer
1175 * if needed, and signal errors or early completion.
1176 *
1177 * On success a new bio layer is pushed with the translated
1178 * block number, and returned.
1179 */
1180struct bio *
1181bounds_check_with_label(cdev_t dev, struct bio *bio,
1182 struct disklabel *lp, int wlabel)
1183{
1184 struct bio *nbio;
1185 struct buf *bp = bio->bio_buf;
1186 struct partition *p = lp->d_partitions + dkpart(dev);
1187 int labelsect = lp->d_partitions[0].p_offset;
1188 int maxsz = p->p_size,
1189 sz = (bp->b_bcount + DEV_BSIZE - 1) >> DEV_BSHIFT;
1190 daddr_t blkno = (daddr_t)(bio->bio_offset >> DEV_BSHIFT);
1191
1192 /* overwriting disk label ? */
1193 /* XXX should also protect bootstrap in first 8K */
1194 if (blkno + p->p_offset <= LABELSECTOR + labelsect &&
1195#if LABELSECTOR != 0
1196 blkno + p->p_offset + sz > LABELSECTOR + labelsect &&
1197#endif
1198 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1199 bp->b_error = EROFS;
1200 goto error;
1201 }
1202
1203#if defined(DOSBBSECTOR) && defined(notyet)
1204 /* overwriting master boot record? */
1205 if (blkno + p->p_offset <= DOSBBSECTOR &&
1206 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1207 bp->b_error = EROFS;
1208 goto error;
1209 }
1210#endif
1211
1212 /*
1213 * Check for out of bounds, EOF, and EOF clipping.
1214 */
1215 if (bio->bio_offset < 0)
1216 goto bad;
1217 if (blkno + sz > maxsz) {
1218 /*
1219 * Past EOF or B_BNOCLIP flag was set, the request is bad.
1220 */
1221 if (blkno > maxsz || (bp->b_flags & B_BNOCLIP))
1222 goto bad;
1223
1224 /*
1225 * If exactly on EOF just complete the I/O with no bytes
1226 * transfered. B_INVAL must be set to throw away the
1227 * contents of the buffer. Otherwise clip b_bcount.
1228 */
1229 if (blkno == maxsz) {
1230 bp->b_resid = bp->b_bcount;
1231 bp->b_flags |= B_INVAL;
1232 goto done;
1233 }
1234 bp->b_bcount = (maxsz - blkno) << DEV_BSHIFT;
1235 }
1236 nbio = push_bio(bio);
1237 nbio->bio_offset = bio->bio_offset + ((off_t)p->p_offset << DEV_BSHIFT);
1238 return (nbio);
1239
1240 /*
1241 * The caller is responsible for calling biodone() on the passed bio
1242 * when we return NULL.
1243 */
1244bad:
1245 bp->b_error = EINVAL;
1246error:
1247 bp->b_resid = bp->b_bcount;
1248 bp->b_flags |= B_ERROR | B_INVAL;
1249done:
1250 return (NULL);
1251}
1252