Continue fleshing out the VKERNEL.
[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 $
40 * $DragonFly: src/sys/platform/vkernel/i386/cpu_regs.c,v 1.1 2007/01/05 22:18:18 dillon Exp $
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
135int physmem = 0;
136
137static int
138sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
139{
140 int error = sysctl_handle_int(oidp, 0, ctob(physmem), req);
141 return (error);
142}
143
144SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_INT|CTLFLAG_RD,
145 0, 0, sysctl_hw_physmem, "IU", "");
146
147static int
148sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
149{
150 int error = sysctl_handle_int(oidp, 0,
151 ctob(physmem - vmstats.v_wire_count), req);
152 return (error);
153}
154
155SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD,
156 0, 0, sysctl_hw_usermem, "IU", "");
157
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;
232 sf.sf_uc.uc_mcontext.mc_gs = rgs();
233 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_fs, sizeof(struct trapframe));
234
235 /* Allocate and validate space for the signal handler context. */
236 /* XXX lwp flags */
237 if ((p->p_flag & P_ALTSTACK) != 0 && !oonstack &&
238 SIGISMEMBER(psp->ps_sigonstack, sig)) {
239 sfp = (struct sigframe *)(lp->lwp_sigstk.ss_sp +
240 lp->lwp_sigstk.ss_size - sizeof(struct sigframe));
241 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
242 }
243 else
244 sfp = (struct sigframe *)regs->tf_esp - 1;
245
246 /* Translate the signal is appropriate */
247 if (p->p_sysent->sv_sigtbl) {
248 if (sig <= p->p_sysent->sv_sigsize)
249 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
250 }
251
252 /* Build the argument list for the signal handler. */
253 sf.sf_signum = sig;
254 sf.sf_ucontext = (register_t)&sfp->sf_uc;
255 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
256 /* Signal handler installed with SA_SIGINFO. */
257 sf.sf_siginfo = (register_t)&sfp->sf_si;
258 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
259
260 /* fill siginfo structure */
261 sf.sf_si.si_signo = sig;
262 sf.sf_si.si_code = code;
263 sf.sf_si.si_addr = (void*)regs->tf_err;
264 }
265 else {
266 /* Old FreeBSD-style arguments. */
267 sf.sf_siginfo = code;
268 sf.sf_addr = regs->tf_err;
269 sf.sf_ahu.sf_handler = catcher;
270 }
271
272#if 0
273 /*
274 * If we're a vm86 process, we want to save the segment registers.
275 * We also change eflags to be our emulated eflags, not the actual
276 * eflags.
277 */
278 if (regs->tf_eflags & PSL_VM) {
279 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
280 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
281
282 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
283 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
284 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
285 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
286
287 if (vm86->vm86_has_vme == 0)
288 sf.sf_uc.uc_mcontext.mc_eflags =
289 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
290 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
291
292 /*
293 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
294 * syscalls made by the signal handler. This just avoids
295 * wasting time for our lazy fixup of such faults. PSL_NT
296 * does nothing in vm86 mode, but vm86 programs can set it
297 * almost legitimately in probes for old cpu types.
298 */
299 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
300 }
301#endif
302
303 /*
304 * Copy the sigframe out to the user's stack.
305 */
306 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
307 /*
308 * Something is wrong with the stack pointer.
309 * ...Kill the process.
310 */
311 sigexit(p, SIGILL);
312 }
313
314 regs->tf_esp = (int)sfp;
315 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
316 regs->tf_eflags &= ~PSL_T;
317 regs->tf_cs = 0;
318 regs->tf_ds = 0;
319 regs->tf_es = 0;
320 regs->tf_fs = 0;
321 regs->tf_ss = 0;
322}
323
324/*
325 * Sanitize the trapframe for a virtual kernel passing control to a custom
326 * VM context.
327 *
328 * Allow userland to set or maintain PSL_RF, the resume flag. This flag
329 * basically controls whether the return PC should skip the first instruction
330 * (as in an explicit system call) or re-execute it (as in an exception).
331 */
332int
333cpu_sanitize_frame(struct trapframe *frame)
334{
335 frame->tf_cs = 0;
336 frame->tf_ds = 0;
337 frame->tf_es = 0;
338 frame->tf_fs = 0;
339 frame->tf_ss = 0;
340 frame->tf_eflags &= (PSL_USER | PSL_RF);
341 frame->tf_eflags |= PSL_RESERVED_DEFAULT | PSL_I;
342 return(0);
343}
344
345/*
346 * sigreturn(ucontext_t *sigcntxp)
347 *
348 * System call to cleanup state after a signal
349 * has been taken. Reset signal mask and
350 * stack state from context left by sendsig (above).
351 * Return to previous pc and psl as specified by
352 * context left by sendsig. Check carefully to
353 * make sure that the user has not modified the
354 * state to gain improper privileges.
355 */
356#define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
357#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
358
359int
360sys_sigreturn(struct sigreturn_args *uap)
361{
362 struct lwp *lp = curthread->td_lwp;
363 struct trapframe *regs;
364 ucontext_t *ucp;
365 int cs, eflags;
366
367 ucp = uap->sigcntxp;
368
369 if (!useracc((caddr_t)ucp, sizeof(ucontext_t), VM_PROT_READ))
370 return (EFAULT);
371
372 regs = lp->lwp_md.md_regs;
373 eflags = ucp->uc_mcontext.mc_eflags;
374
375#if 0
376 if (eflags & PSL_VM) {
377 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
378 struct vm86_kernel *vm86;
379
380 /*
381 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
382 * set up the vm86 area, and we can't enter vm86 mode.
383 */
384 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
385 return (EINVAL);
386 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
387 if (vm86->vm86_inited == 0)
388 return (EINVAL);
389
390 /* go back to user mode if both flags are set */
391 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
392 trapsignal(lp->lwp_proc, SIGBUS, 0);
393
394 if (vm86->vm86_has_vme) {
395 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
396 (eflags & VME_USERCHANGE) | PSL_VM;
397 } else {
398 vm86->vm86_eflags = eflags; /* save VIF, VIP */
399 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
400 }
401 bcopy(&ucp->uc_mcontext.mc_fs, tf, sizeof(struct trapframe));
402 tf->tf_eflags = eflags;
403 tf->tf_vm86_ds = tf->tf_ds;
404 tf->tf_vm86_es = tf->tf_es;
405 tf->tf_vm86_fs = tf->tf_fs;
406 tf->tf_vm86_gs = ucp->uc_mcontext.mc_gs;
407 tf->tf_ds = 0;
408 tf->tf_es = 0;
409 tf->tf_fs = 0;
410 } else
411#endif
412 {
413 /*
414 * Don't allow users to change privileged or reserved flags.
415 */
416 /*
417 * XXX do allow users to change the privileged flag PSL_RF.
418 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
419 * should sometimes set it there too. tf_eflags is kept in
420 * the signal context during signal handling and there is no
421 * other place to remember it, so the PSL_RF bit may be
422 * corrupted by the signal handler without us knowing.
423 * Corruption of the PSL_RF bit at worst causes one more or
424 * one less debugger trap, so allowing it is fairly harmless.
425 */
426 if (!EFL_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
427 kprintf("sigreturn: eflags = 0x%x\n", eflags);
428 return(EINVAL);
429 }
430
431 /*
432 * Don't allow users to load a valid privileged %cs. Let the
433 * hardware check for invalid selectors, excess privilege in
434 * other selectors, invalid %eip's and invalid %esp's.
435 */
436 cs = ucp->uc_mcontext.mc_cs;
437 if (!CS_SECURE(cs)) {
438 kprintf("sigreturn: cs = 0x%x\n", cs);
439 trapsignal(lp->lwp_proc, SIGBUS, T_PROTFLT);
440 return(EINVAL);
441 }
442 bcopy(&ucp->uc_mcontext.mc_fs, regs, sizeof(struct trapframe));
443 }
444
445 if (ucp->uc_mcontext.mc_onstack & 1)
446 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
447 else
448 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
449
450 lp->lwp_sigmask = ucp->uc_sigmask;
451 SIG_CANTMASK(lp->lwp_sigmask);
452 return(EJUSTRETURN);
453}
454
455/*
456 * Stack frame on entry to function. %eax will contain the function vector,
457 * %ecx will contain the function data. flags, ecx, and eax will have
458 * already been pushed on the stack.
459 */
460struct upc_frame {
461 register_t eax;
462 register_t ecx;
463 register_t edx;
464 register_t flags;
465 register_t oldip;
466};
467
468void
469sendupcall(struct vmupcall *vu, int morepending)
470{
471 struct lwp *lp = curthread->td_lwp;
472 struct trapframe *regs;
473 struct upcall upcall;
474 struct upc_frame upc_frame;
475 int crit_count = 0;
476
477 /*
478 * Get the upcall data structure
479 */
480 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
481 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
482 ) {
483 vu->vu_pending = 0;
484 kprintf("bad upcall address\n");
485 return;
486 }
487
488 /*
489 * If the data structure is already marked pending or has a critical
490 * section count, mark the data structure as pending and return
491 * without doing an upcall. vu_pending is left set.
492 */
493 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
494 if (upcall.upc_pending < vu->vu_pending) {
495 upcall.upc_pending = vu->vu_pending;
496 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
497 sizeof(upcall.upc_pending));
498 }
499 return;
500 }
501
502 /*
503 * We can run this upcall now, clear vu_pending.
504 *
505 * Bump our critical section count and set or clear the
506 * user pending flag depending on whether more upcalls are
507 * pending. The user will be responsible for calling
508 * upc_dispatch(-1) to process remaining upcalls.
509 */
510 vu->vu_pending = 0;
511 upcall.upc_pending = morepending;
512 crit_count += TDPRI_CRIT;
513 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
514 sizeof(upcall.upc_pending));
515 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
516 sizeof(int));
517
518 /*
519 * Construct a stack frame and issue the upcall
520 */
521 regs = lp->lwp_md.md_regs;
522 upc_frame.eax = regs->tf_eax;
523 upc_frame.ecx = regs->tf_ecx;
524 upc_frame.edx = regs->tf_edx;
525 upc_frame.flags = regs->tf_eflags;
526 upc_frame.oldip = regs->tf_eip;
527 if (copyout(&upc_frame, (void *)(regs->tf_esp - sizeof(upc_frame)),
528 sizeof(upc_frame)) != 0) {
529 kprintf("bad stack on upcall\n");
530 } else {
531 regs->tf_eax = (register_t)vu->vu_func;
532 regs->tf_ecx = (register_t)vu->vu_data;
533 regs->tf_edx = (register_t)lp->lwp_upcall;
534 regs->tf_eip = (register_t)vu->vu_ctx;
535 regs->tf_esp -= sizeof(upc_frame);
536 }
537}
538
539/*
540 * fetchupcall occurs in the context of a system call, which means that
541 * we have to return EJUSTRETURN in order to prevent eax and edx from
542 * being overwritten by the syscall return value.
543 *
544 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
545 * and the function pointer in %eax.
546 */
547int
548fetchupcall (struct vmupcall *vu, int morepending, void *rsp)
549{
550 struct upc_frame upc_frame;
551 struct lwp *lp = curthread->td_lwp;
552 struct trapframe *regs;
553 int error;
554 struct upcall upcall;
555 int crit_count;
556
557 regs = lp->lwp_md.md_regs;
558
559 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
560 if (error == 0) {
561 if (vu) {
562 /*
563 * This jumps us to the next ready context.
564 */
565 vu->vu_pending = 0;
566 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
567 crit_count = 0;
568 if (error == 0)
569 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
570 crit_count += TDPRI_CRIT;
571 if (error == 0)
572 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
573 regs->tf_eax = (register_t)vu->vu_func;
574 regs->tf_ecx = (register_t)vu->vu_data;
575 regs->tf_edx = (register_t)lp->lwp_upcall;
576 regs->tf_eip = (register_t)vu->vu_ctx;
577 regs->tf_esp = (register_t)rsp;
578 } else {
579 /*
580 * This returns us to the originally interrupted code.
581 */
582 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
583 regs->tf_eax = upc_frame.eax;
584 regs->tf_ecx = upc_frame.ecx;
585 regs->tf_edx = upc_frame.edx;
586 regs->tf_eflags = (regs->tf_eflags & ~PSL_USERCHANGE) |
587 (upc_frame.flags & PSL_USERCHANGE);
588 regs->tf_eip = upc_frame.oldip;
589 regs->tf_esp = (register_t)((char *)rsp + sizeof(upc_frame));
590 }
591 }
592 if (error == 0)
593 error = EJUSTRETURN;
594 return(error);
595}
596
597/*
598 * cpu_idle() represents the idle LWKT. You cannot return from this function
599 * (unless you want to blow things up!). Instead we look for runnable threads
600 * and loop or halt as appropriate. Giant is not held on entry to the thread.
601 *
602 * The main loop is entered with a critical section held, we must release
603 * the critical section before doing anything else. lwkt_switch() will
604 * check for pending interrupts due to entering and exiting its own
605 * critical section.
606 *
607 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
608 * to wake a HLTed cpu up. However, there are cases where the idlethread
609 * will be entered with the possibility that no IPI will occur and in such
610 * cases lwkt_switch() sets TDF_IDLE_NOHLT.
611 */
612static int cpu_idle_hlt = 1;
613static int cpu_idle_hltcnt;
614static int cpu_idle_spincnt;
615SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
616 &cpu_idle_hlt, 0, "Idle loop HLT enable");
617SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
618 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
619SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
620 &cpu_idle_spincnt, 0, "Idle loop entry spins");
621
622static void
623cpu_idle_default_hook(void)
624{
625 /*
626 * We must guarentee that hlt is exactly the instruction
627 * following the sti.
628 */
629 __asm __volatile("hlt"); /* sti; hlt */
630}
631
632/* Other subsystems (e.g., ACPI) can hook this later. */
633void (*cpu_idle_hook)(void) = cpu_idle_default_hook;
634
635void
636cpu_idle(void)
637{
638 struct thread *td = curthread;
639
640 crit_exit();
641 KKASSERT(td->td_pri < TDPRI_CRIT);
642 for (;;) {
643 /*
644 * See if there are any LWKTs ready to go.
645 */
646 lwkt_switch();
647
648 /*
649 * If we are going to halt call splz unconditionally after
650 * CLIing to catch any interrupt races. Note that we are
651 * at SPL0 and interrupts are enabled.
652 */
653 if (cpu_idle_hlt && !lwkt_runnable() &&
654 (td->td_flags & TDF_IDLE_NOHLT) == 0) {
655 /* __asm __volatile("cli"); */
656 splz();
657 if (!lwkt_runnable())
658 cpu_idle_hook();
659#ifdef SMP
660 else
661 __asm __volatile("pause");
662#endif
663 ++cpu_idle_hltcnt;
664 } else {
665 td->td_flags &= ~TDF_IDLE_NOHLT;
666 splz();
667#ifdef SMP
668 /*__asm __volatile("sti; pause");*/
669 __asm __volatile("pause");
670#else
671 /*__asm __volatile("sti");*/
672#endif
673 ++cpu_idle_spincnt;
674 }
675 }
676}
677
678/*
679 * Clear registers on exec
680 */
681void
682setregs(struct lwp *lp, u_long entry, u_long stack, u_long ps_strings)
683{
684 struct trapframe *regs = lp->lwp_md.md_regs;
685 struct pcb *pcb = lp->lwp_thread->td_pcb;
686
687 /* Reset pc->pcb_gs and %gs before possibly invalidating it. */
688 pcb->pcb_gs = 0;
689#if 0
690 load_gs(_udatasel);
691#endif
692
693 /* was i386_user_cleanup() in NetBSD */
694 user_ldt_free(pcb);
695
696 bzero((char *)regs, sizeof(struct trapframe));
697 regs->tf_eip = entry;
698 regs->tf_esp = stack;
699 regs->tf_eflags = PSL_USER | (regs->tf_eflags & PSL_T);
700 regs->tf_ss = 0;
701 regs->tf_ds = 0;
702 regs->tf_es = 0;
703 regs->tf_fs = 0;
704 regs->tf_cs = 0;
705
706 /* PS_STRINGS value for BSD/OS binaries. It is 0 for non-BSD/OS. */
707 regs->tf_ebx = ps_strings;
708
709 /*
710 * Reset the hardware debug registers if they were in use.
711 * They won't have any meaning for the newly exec'd process.
712 */
713 if (pcb->pcb_flags & PCB_DBREGS) {
714 pcb->pcb_dr0 = 0;
715 pcb->pcb_dr1 = 0;
716 pcb->pcb_dr2 = 0;
717 pcb->pcb_dr3 = 0;
718 pcb->pcb_dr6 = 0;
719 pcb->pcb_dr7 = 0;
720 if (pcb == curthread->td_pcb) {
721 /*
722 * Clear the debug registers on the running
723 * CPU, otherwise they will end up affecting
724 * the next process we switch to.
725 */
726 reset_dbregs();
727 }
728 pcb->pcb_flags &= ~PCB_DBREGS;
729 }
730
731 /*
732 * Initialize the math emulator (if any) for the current process.
733 * Actually, just clear the bit that says that the emulator has
734 * been initialized. Initialization is delayed until the process
735 * traps to the emulator (if it is done at all) mainly because
736 * emulators don't provide an entry point for initialization.
737 */
738 lp->lwp_thread->td_pcb->pcb_flags &= ~FP_SOFTFP;
739
740 /*
741 * note: do not set CR0_TS here. npxinit() must do it after clearing
742 * gd_npxthread. Otherwise a preemptive interrupt thread may panic
743 * in npxdna().
744 */
745 crit_enter();
746#if 0
747 load_cr0(rcr0() | CR0_MP);
748#endif
749
750#if NNPX > 0
751 /* Initialize the npx (if any) for the current process. */
752 npxinit(__INITIAL_NPXCW__);
753#endif
754 crit_exit();
755
756 /*
757 * note: linux emulator needs edx to be 0x0 on entry, which is
758 * handled in execve simply by setting the 64 bit syscall
759 * return value to 0.
760 */
761}
762
763void
764cpu_setregs(void)
765{
766#if 0
767 unsigned int cr0;
768
769 cr0 = rcr0();
770 cr0 |= CR0_NE; /* Done by npxinit() */
771 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
772#ifdef I386_CPU
773 if (cpu_class != CPUCLASS_386)
774#endif
775 cr0 |= CR0_WP | CR0_AM;
776 load_cr0(cr0);
777 load_gs(_udatasel);
778#endif
779}
780
781static int
782sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
783{
784 int error;
785 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
786 req);
787 if (!error && req->newptr)
788 resettodr();
789 return (error);
790}
791
792SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
793 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
794
795extern u_long bootdev; /* not a cdev_t - encoding is different */
796SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
797 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
798
799/*
800 * Initialize 386 and configure to run kernel
801 */
802
803/*
804 * Initialize segments & interrupt table
805 */
806
807extern struct user *proc0paddr;
808
809#if 0
810
811extern inthand_t
812 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
813 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
814 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
815 IDTVEC(page), IDTVEC(mchk), IDTVEC(fpu), IDTVEC(align),
816 IDTVEC(xmm), IDTVEC(syscall),
817 IDTVEC(rsvd0);
818extern inthand_t
819 IDTVEC(int0x80_syscall);
820
821#endif
822
823#ifdef DEBUG_INTERRUPTS
824extern inthand_t *Xrsvdary[256];
825#endif
826
827int
828ptrace_set_pc(struct proc *p, unsigned long addr)
829{
830 p->p_md.md_regs->tf_eip = addr;
831 return (0);
832}
833
834int
835ptrace_single_step(struct lwp *lp)
836{
837 lp->lwp_md.md_regs->tf_eflags |= PSL_T;
838 return (0);
839}
840
841int
842fill_regs(struct lwp *lp, struct reg *regs)
843{
844 struct pcb *pcb;
845 struct trapframe *tp;
846
847 tp = lp->lwp_md.md_regs;
848 regs->r_fs = tp->tf_fs;
849 regs->r_es = tp->tf_es;
850 regs->r_ds = tp->tf_ds;
851 regs->r_edi = tp->tf_edi;
852 regs->r_esi = tp->tf_esi;
853 regs->r_ebp = tp->tf_ebp;
854 regs->r_ebx = tp->tf_ebx;
855 regs->r_edx = tp->tf_edx;
856 regs->r_ecx = tp->tf_ecx;
857 regs->r_eax = tp->tf_eax;
858 regs->r_eip = tp->tf_eip;
859 regs->r_cs = tp->tf_cs;
860 regs->r_eflags = tp->tf_eflags;
861 regs->r_esp = tp->tf_esp;
862 regs->r_ss = tp->tf_ss;
863 pcb = lp->lwp_thread->td_pcb;
864 regs->r_gs = pcb->pcb_gs;
865 return (0);
866}
867
868int
869set_regs(struct lwp *lp, struct reg *regs)
870{
871 struct pcb *pcb;
872 struct trapframe *tp;
873
874 tp = lp->lwp_md.md_regs;
875 if (!EFL_SECURE(regs->r_eflags, tp->tf_eflags) ||
876 !CS_SECURE(regs->r_cs))
877 return (EINVAL);
878 tp->tf_fs = regs->r_fs;
879 tp->tf_es = regs->r_es;
880 tp->tf_ds = regs->r_ds;
881 tp->tf_edi = regs->r_edi;
882 tp->tf_esi = regs->r_esi;
883 tp->tf_ebp = regs->r_ebp;
884 tp->tf_ebx = regs->r_ebx;
885 tp->tf_edx = regs->r_edx;
886 tp->tf_ecx = regs->r_ecx;
887 tp->tf_eax = regs->r_eax;
888 tp->tf_eip = regs->r_eip;
889 tp->tf_cs = regs->r_cs;
890 tp->tf_eflags = regs->r_eflags;
891 tp->tf_esp = regs->r_esp;
892 tp->tf_ss = regs->r_ss;
893 pcb = lp->lwp_thread->td_pcb;
894 pcb->pcb_gs = regs->r_gs;
895 return (0);
896}
897
898#ifndef CPU_DISABLE_SSE
899static void
900fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
901{
902 struct env87 *penv_87 = &sv_87->sv_env;
903 struct envxmm *penv_xmm = &sv_xmm->sv_env;
904 int i;
905
906 /* FPU control/status */
907 penv_87->en_cw = penv_xmm->en_cw;
908 penv_87->en_sw = penv_xmm->en_sw;
909 penv_87->en_tw = penv_xmm->en_tw;
910 penv_87->en_fip = penv_xmm->en_fip;
911 penv_87->en_fcs = penv_xmm->en_fcs;
912 penv_87->en_opcode = penv_xmm->en_opcode;
913 penv_87->en_foo = penv_xmm->en_foo;
914 penv_87->en_fos = penv_xmm->en_fos;
915
916 /* FPU registers */
917 for (i = 0; i < 8; ++i)
918 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
919
920 sv_87->sv_ex_sw = sv_xmm->sv_ex_sw;
921}
922
923static void
924set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
925{
926 struct env87 *penv_87 = &sv_87->sv_env;
927 struct envxmm *penv_xmm = &sv_xmm->sv_env;
928 int i;
929
930 /* FPU control/status */
931 penv_xmm->en_cw = penv_87->en_cw;
932 penv_xmm->en_sw = penv_87->en_sw;
933 penv_xmm->en_tw = penv_87->en_tw;
934 penv_xmm->en_fip = penv_87->en_fip;
935 penv_xmm->en_fcs = penv_87->en_fcs;
936 penv_xmm->en_opcode = penv_87->en_opcode;
937 penv_xmm->en_foo = penv_87->en_foo;
938 penv_xmm->en_fos = penv_87->en_fos;
939
940 /* FPU registers */
941 for (i = 0; i < 8; ++i)
942 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
943
944 sv_xmm->sv_ex_sw = sv_87->sv_ex_sw;
945}
946#endif /* CPU_DISABLE_SSE */
947
948int
949fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
950{
951#ifndef CPU_DISABLE_SSE
952 if (cpu_fxsr) {
953 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
954 (struct save87 *)fpregs);
955 return (0);
956 }
957#endif /* CPU_DISABLE_SSE */
958 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
959 return (0);
960}
961
962int
963set_fpregs(struct lwp *lp, struct fpreg *fpregs)
964{
965#ifndef CPU_DISABLE_SSE
966 if (cpu_fxsr) {
967 set_fpregs_xmm((struct save87 *)fpregs,
968 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
969 return (0);
970 }
971#endif /* CPU_DISABLE_SSE */
972 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
973 return (0);
974}
975
976int
977fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
978{
979 if (lp == NULL) {
980 dbregs->dr0 = rdr0();
981 dbregs->dr1 = rdr1();
982 dbregs->dr2 = rdr2();
983 dbregs->dr3 = rdr3();
984 dbregs->dr4 = rdr4();
985 dbregs->dr5 = rdr5();
986 dbregs->dr6 = rdr6();
987 dbregs->dr7 = rdr7();
988 } else {
989 struct pcb *pcb;
990
991 pcb = lp->lwp_thread->td_pcb;
992 dbregs->dr0 = pcb->pcb_dr0;
993 dbregs->dr1 = pcb->pcb_dr1;
994 dbregs->dr2 = pcb->pcb_dr2;
995 dbregs->dr3 = pcb->pcb_dr3;
996 dbregs->dr4 = 0;
997 dbregs->dr5 = 0;
998 dbregs->dr6 = pcb->pcb_dr6;
999 dbregs->dr7 = pcb->pcb_dr7;
1000 }
1001 return (0);
1002}
1003
1004int
1005set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1006{
1007 if (lp == NULL) {
1008 load_dr0(dbregs->dr0);
1009 load_dr1(dbregs->dr1);
1010 load_dr2(dbregs->dr2);
1011 load_dr3(dbregs->dr3);
1012 load_dr4(dbregs->dr4);
1013 load_dr5(dbregs->dr5);
1014 load_dr6(dbregs->dr6);
1015 load_dr7(dbregs->dr7);
1016 } else {
1017 struct pcb *pcb;
1018 struct ucred *ucred;
1019 int i;
1020 uint32_t mask1, mask2;
1021
1022 /*
1023 * Don't let an illegal value for dr7 get set. Specifically,
1024 * check for undefined settings. Setting these bit patterns
1025 * result in undefined behaviour and can lead to an unexpected
1026 * TRCTRAP.
1027 */
1028 for (i = 0, mask1 = 0x3<<16, mask2 = 0x2<<16; i < 8;
1029 i++, mask1 <<= 2, mask2 <<= 2)
1030 if ((dbregs->dr7 & mask1) == mask2)
1031 return (EINVAL);
1032
1033 pcb = lp->lwp_thread->td_pcb;
1034 ucred = lp->lwp_proc->p_ucred;
1035
1036 /*
1037 * Don't let a process set a breakpoint that is not within the
1038 * process's address space. If a process could do this, it
1039 * could halt the system by setting a breakpoint in the kernel
1040 * (if ddb was enabled). Thus, we need to check to make sure
1041 * that no breakpoints are being enabled for addresses outside
1042 * process's address space, unless, perhaps, we were called by
1043 * uid 0.
1044 *
1045 * XXX - what about when the watched area of the user's
1046 * address space is written into from within the kernel
1047 * ... wouldn't that still cause a breakpoint to be generated
1048 * from within kernel mode?
1049 */
1050
1051 if (suser_cred(ucred, 0) != 0) {
1052 if (dbregs->dr7 & 0x3) {
1053 /* dr0 is enabled */
1054 if (dbregs->dr0 >= VM_MAX_USER_ADDRESS)
1055 return (EINVAL);
1056 }
1057
1058 if (dbregs->dr7 & (0x3<<2)) {
1059 /* dr1 is enabled */
1060 if (dbregs->dr1 >= VM_MAX_USER_ADDRESS)
1061 return (EINVAL);
1062 }
1063
1064 if (dbregs->dr7 & (0x3<<4)) {
1065 /* dr2 is enabled */
1066 if (dbregs->dr2 >= VM_MAX_USER_ADDRESS)
1067 return (EINVAL);
1068 }
1069
1070 if (dbregs->dr7 & (0x3<<6)) {
1071 /* dr3 is enabled */
1072 if (dbregs->dr3 >= VM_MAX_USER_ADDRESS)
1073 return (EINVAL);
1074 }
1075 }
1076
1077 pcb->pcb_dr0 = dbregs->dr0;
1078 pcb->pcb_dr1 = dbregs->dr1;
1079 pcb->pcb_dr2 = dbregs->dr2;
1080 pcb->pcb_dr3 = dbregs->dr3;
1081 pcb->pcb_dr6 = dbregs->dr6;
1082 pcb->pcb_dr7 = dbregs->dr7;
1083
1084 pcb->pcb_flags |= PCB_DBREGS;
1085 }
1086
1087 return (0);
1088}
1089
1090#if 0
1091/*
1092 * Return > 0 if a hardware breakpoint has been hit, and the
1093 * breakpoint was in user space. Return 0, otherwise.
1094 */
1095int
1096user_dbreg_trap(void)
1097{
1098 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1099 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1100 int nbp; /* number of breakpoints that triggered */
1101 caddr_t addr[4]; /* breakpoint addresses */
1102 int i;
1103
1104 dr7 = rdr7();
1105 if ((dr7 & 0x000000ff) == 0) {
1106 /*
1107 * all GE and LE bits in the dr7 register are zero,
1108 * thus the trap couldn't have been caused by the
1109 * hardware debug registers
1110 */
1111 return 0;
1112 }
1113
1114 nbp = 0;
1115 dr6 = rdr6();
1116 bp = dr6 & 0x0000000f;
1117
1118 if (!bp) {
1119 /*
1120 * None of the breakpoint bits are set meaning this
1121 * trap was not caused by any of the debug registers
1122 */
1123 return 0;
1124 }
1125
1126 /*
1127 * at least one of the breakpoints were hit, check to see
1128 * which ones and if any of them are user space addresses
1129 */
1130
1131 if (bp & 0x01) {
1132 addr[nbp++] = (caddr_t)rdr0();
1133 }
1134 if (bp & 0x02) {
1135 addr[nbp++] = (caddr_t)rdr1();
1136 }
1137 if (bp & 0x04) {
1138 addr[nbp++] = (caddr_t)rdr2();
1139 }
1140 if (bp & 0x08) {
1141 addr[nbp++] = (caddr_t)rdr3();
1142 }
1143
1144 for (i=0; i<nbp; i++) {
1145 if (addr[i] <
1146 (caddr_t)VM_MAX_USER_ADDRESS) {
1147 /*
1148 * addr[i] is in user space
1149 */
1150 return nbp;
1151 }
1152 }
1153
1154 /*
1155 * None of the breakpoints are in user space.
1156 */
1157 return 0;
1158}
1159
1160#endif
1161
1162
1163#ifndef DDB
1164void
1165Debugger(const char *msg)
1166{
1167 kprintf("Debugger(\"%s\") called.\n", msg);
1168}
1169#endif /* no DDB */
1170
1171#include <sys/disklabel.h>
1172
1173/*
1174 * Determine the size of the transfer, and make sure it is
1175 * within the boundaries of the partition. Adjust transfer
1176 * if needed, and signal errors or early completion.
1177 *
1178 * On success a new bio layer is pushed with the translated
1179 * block number, and returned.
1180 */
1181struct bio *
1182bounds_check_with_label(cdev_t dev, struct bio *bio,
1183 struct disklabel *lp, int wlabel)
1184{
1185 struct bio *nbio;
1186 struct buf *bp = bio->bio_buf;
1187 struct partition *p = lp->d_partitions + dkpart(dev);
1188 int labelsect = lp->d_partitions[0].p_offset;
1189 int maxsz = p->p_size,
1190 sz = (bp->b_bcount + DEV_BSIZE - 1) >> DEV_BSHIFT;
1191 daddr_t blkno = (daddr_t)(bio->bio_offset >> DEV_BSHIFT);
1192
1193 /* overwriting disk label ? */
1194 /* XXX should also protect bootstrap in first 8K */
1195 if (blkno + p->p_offset <= LABELSECTOR + labelsect &&
1196#if LABELSECTOR != 0
1197 blkno + p->p_offset + sz > LABELSECTOR + labelsect &&
1198#endif
1199 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1200 bp->b_error = EROFS;
1201 goto error;
1202 }
1203
1204#if defined(DOSBBSECTOR) && defined(notyet)
1205 /* overwriting master boot record? */
1206 if (blkno + p->p_offset <= DOSBBSECTOR &&
1207 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1208 bp->b_error = EROFS;
1209 goto error;
1210 }
1211#endif
1212
1213 /*
1214 * Check for out of bounds, EOF, and EOF clipping.
1215 */
1216 if (bio->bio_offset < 0)
1217 goto bad;
1218 if (blkno + sz > maxsz) {
1219 /*
1220 * Past EOF or B_BNOCLIP flag was set, the request is bad.
1221 */
1222 if (blkno > maxsz || (bp->b_flags & B_BNOCLIP))
1223 goto bad;
1224
1225 /*
1226 * If exactly on EOF just complete the I/O with no bytes
1227 * transfered. B_INVAL must be set to throw away the
1228 * contents of the buffer. Otherwise clip b_bcount.
1229 */
1230 if (blkno == maxsz) {
1231 bp->b_resid = bp->b_bcount;
1232 bp->b_flags |= B_INVAL;
1233 goto done;
1234 }
1235 bp->b_bcount = (maxsz - blkno) << DEV_BSHIFT;
1236 }
1237 nbio = push_bio(bio);
1238 nbio->bio_offset = bio->bio_offset + ((off_t)p->p_offset << DEV_BSHIFT);
1239 return (nbio);
1240
1241 /*
1242 * The caller is responsible for calling biodone() on the passed bio
1243 * when we return NULL.
1244 */
1245bad:
1246 bp->b_error = EINVAL;
1247error:
1248 bp->b_resid = bp->b_bcount;
1249 bp->b_flags |= B_ERROR | B_INVAL;
1250done:
1251 return (NULL);
1252}
1253