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 */
41
42 #include "opt_ddb.h"
43 #include "opt_inet.h"
44 #include "opt_msgbuf.h"
45 #include "opt_swap.h"
46
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/sysmsg.h>
50 #include <sys/signalvar.h>
51 #include <sys/kernel.h>
52 #include <sys/linker.h>
53 #include <sys/malloc.h>
54 #include <sys/proc.h>
55 #include <sys/buf.h>
56 #include <sys/reboot.h>
57 #include <sys/mbuf.h>
58 #include <sys/msgbuf.h>
59 #include <sys/sysent.h>
60 #include <sys/sysctl.h>
61 #include <sys/vmmeter.h>
62 #include <sys/bus.h>
63 #include <sys/usched.h>
64 #include <sys/reg.h>
65
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <sys/lock.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_pager.h>
74 #include <vm/vm_extern.h>
75
76 #include <sys/thread2.h>
77
78 #include <sys/exec.h>
79 #include <sys/cons.h>
80
81 #include <ddb/ddb.h>
82
83 #include <machine/cpu.h>
84 #include <machine/clock.h>
85 #include <machine/specialreg.h>
86 #include <machine/md_var.h>
87 #include <machine/pcb.h>
88 #include <machine/pcb_ext.h>
89 #include <machine/globaldata.h> /* CPU_prvspace */
90 #include <machine/smp.h>
91 #include <machine/cputypes.h>
92
93 #include <bus/isa/rtc.h>
94 #include <sys/random.h>
95 #include <sys/ptrace.h>
96 #include <machine/sigframe.h>
97 #include <unistd.h> /* umtx_* functions */
98
99 extern void dblfault_handler (void);
100
101 static void set_fpregs_xmm (struct save87 *, struct savexmm *);
102 static void fill_fpregs_xmm (struct savexmm *, struct save87 *);
103
104 int64_t tsc_offsets[MAXCPU];
105
106 #if defined(SWTCH_OPTIM_STATS)
107 extern int swtch_optim_stats;
108 SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats,
109 CTLFLAG_RD, &swtch_optim_stats, 0, "");
110 SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count,
111 CTLFLAG_RD, &tlb_flush_count, 0, "");
112 #endif
113
114 static int
sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)115 sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
116 {
117 u_long pmem = ctob(physmem);
118 int error;
119
120 error = sysctl_handle_long(oidp, &pmem, 0, req);
121
122 return (error);
123 }
124
125 SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_ULONG|CTLFLAG_RD,
126 0, 0, sysctl_hw_physmem, "LU", "Total system memory in bytes (number of pages * page size)");
127
128 static int
sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)129 sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
130 {
131 u_long usermem = ctob(Maxmem - vmstats.v_wire_count);
132 int error;
133
134 error = sysctl_handle_long(oidp, &usermem, 0, req);
135
136 return (error);
137 }
138
139 SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_ULONG|CTLFLAG_RD,
140 0, 0, sysctl_hw_usermem, "LU", "");
141
142 SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, 0, "");
143
144 /*
145 * Send an interrupt to process.
146 *
147 * Stack is set up to allow sigcode stored
148 * at top to call routine, followed by kcall
149 * to sigreturn routine below. After sigreturn
150 * resets the signal mask, the stack, and the
151 * frame pointer, it returns to the user
152 * specified pc, psl.
153 */
154 void
sendsig(sig_t catcher,int sig,sigset_t * mask,u_long code)155 sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
156 {
157 struct lwp *lp = curthread->td_lwp;
158 struct proc *p = lp->lwp_proc;
159 struct trapframe *regs;
160 struct sigacts *psp = p->p_sigacts;
161 struct sigframe sf, *sfp;
162 int oonstack;
163 char *sp;
164
165 regs = lp->lwp_md.md_regs;
166 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
167
168 /* Save user context */
169 bzero(&sf, sizeof(struct sigframe));
170 sf.sf_uc.uc_sigmask = *mask;
171 sf.sf_uc.uc_stack = lp->lwp_sigstk;
172 sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
173 KKASSERT(__offsetof(struct trapframe, tf_rdi) == 0);
174 /* gcc8 craps out on -Warray-bounds w/ optimized bcopy */
175 _bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(struct trapframe));
176
177 /* Make the size of the saved context visible to userland */
178 sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext);
179
180 /* Allocate and validate space for the signal handler context. */
181 if ((lp->lwp_flags & LWP_ALTSTACK) != 0 && !oonstack &&
182 SIGISMEMBER(psp->ps_sigonstack, sig)) {
183 sp = (char *)lp->lwp_sigstk.ss_sp + lp->lwp_sigstk.ss_size -
184 sizeof(struct sigframe);
185 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
186 } else {
187 /* We take red zone into account */
188 sp = (char *)regs->tf_rsp - sizeof(struct sigframe) - 128;
189 }
190
191 /* Align to 16 bytes */
192 sfp = (struct sigframe *)((intptr_t)sp & ~0xFUL);
193
194 /* Translate the signal is appropriate */
195 if (p->p_sysent->sv_sigtbl) {
196 if (sig <= p->p_sysent->sv_sigsize)
197 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
198 }
199
200 /*
201 * Build the argument list for the signal handler.
202 *
203 * Arguments are in registers (%rdi, %rsi, %rdx, %rcx)
204 */
205 regs->tf_rdi = sig; /* argument 1 */
206 regs->tf_rdx = (register_t)&sfp->sf_uc; /* argument 3 */
207
208 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
209 /*
210 * Signal handler installed with SA_SIGINFO.
211 *
212 * action(signo, siginfo, ucontext)
213 */
214 regs->tf_rsi = (register_t)&sfp->sf_si; /* argument 2 */
215 regs->tf_rcx = (register_t)regs->tf_err; /* argument 4 */
216 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
217
218 /* fill siginfo structure */
219 sf.sf_si.si_signo = sig;
220 sf.sf_si.si_pid = psp->ps_frominfo[sig].pid;
221 sf.sf_si.si_uid = psp->ps_frominfo[sig].uid;
222 sf.sf_si.si_code = code;
223 sf.sf_si.si_addr = (void *)regs->tf_addr;
224 } else {
225 /*
226 * Old FreeBSD-style arguments.
227 *
228 * handler (signo, code, [uc], addr)
229 */
230 regs->tf_rsi = (register_t)code; /* argument 2 */
231 regs->tf_rcx = (register_t)regs->tf_addr; /* argument 4 */
232 sf.sf_ahu.sf_handler = catcher;
233 }
234
235 #if 0
236 /*
237 * If we're a vm86 process, we want to save the segment registers.
238 * We also change eflags to be our emulated eflags, not the actual
239 * eflags.
240 */
241 if (regs->tf_eflags & PSL_VM) {
242 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
243 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
244
245 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
246 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
247 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
248 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
249
250 if (vm86->vm86_has_vme == 0)
251 sf.sf_uc.uc_mcontext.mc_eflags =
252 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
253 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
254
255 /*
256 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
257 * syscalls made by the signal handler. This just avoids
258 * wasting time for our lazy fixup of such faults. PSL_NT
259 * does nothing in vm86 mode, but vm86 programs can set it
260 * almost legitimately in probes for old cpu types.
261 */
262 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
263 }
264 #endif
265
266 /*
267 * Save the FPU state and reinit the FP unit
268 */
269 npxpush(&sf.sf_uc.uc_mcontext);
270
271 /*
272 * Copy the sigframe out to the user's stack.
273 */
274 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
275 /*
276 * Something is wrong with the stack pointer.
277 * ...Kill the process.
278 */
279 sigexit(lp, SIGILL);
280 }
281
282 regs->tf_rsp = (register_t)sfp;
283 regs->tf_rip = trunc_page64(PS_STRINGS - *(p->p_sysent->sv_szsigcode));
284 regs->tf_rip -= SZSIGCODE_EXTRA_BYTES;
285
286 /*
287 * x86 abi specifies that the direction flag must be cleared
288 * on function entry
289 */
290 regs->tf_rflags &= ~(PSL_T|PSL_D);
291
292 /*
293 * 64 bit mode has a code and stack selector but
294 * no data or extra selector. %fs and %gs are not
295 * stored in-context.
296 */
297 regs->tf_cs = _ucodesel;
298 regs->tf_ss = _udatasel;
299 }
300
301 /*
302 * Sanitize the trapframe for a virtual kernel passing control to a custom
303 * VM context. Remove any items that would otherwise create a privilage
304 * issue.
305 *
306 * XXX at the moment we allow userland to set the resume flag. Is this a
307 * bad idea?
308 */
309 int
cpu_sanitize_frame(struct trapframe * frame)310 cpu_sanitize_frame(struct trapframe *frame)
311 {
312 frame->tf_cs = _ucodesel;
313 frame->tf_ss = _udatasel;
314 /* XXX VM (8086) mode not supported? */
315 frame->tf_rflags &= (PSL_RF | PSL_USERCHANGE | PSL_VM_UNSUPP);
316 frame->tf_rflags |= PSL_RESERVED_DEFAULT | PSL_I;
317
318 return(0);
319 }
320
321 /*
322 * Sanitize the tls so loading the descriptor does not blow up
323 * on us. For x86_64 we don't have to do anything.
324 */
325 int
cpu_sanitize_tls(struct savetls * tls)326 cpu_sanitize_tls(struct savetls *tls)
327 {
328 return(0);
329 }
330
331 /*
332 * sigreturn(ucontext_t *sigcntxp)
333 *
334 * System call to cleanup state after a signal
335 * has been taken. Reset signal mask and
336 * stack state from context left by sendsig (above).
337 * Return to previous pc and psl as specified by
338 * context left by sendsig. Check carefully to
339 * make sure that the user has not modified the
340 * state to gain improper privileges.
341 */
342 #define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
343 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
344
345 int
sys_sigreturn(struct sysmsg * sysmsg,const struct sigreturn_args * uap)346 sys_sigreturn(struct sysmsg *sysmsg, const struct sigreturn_args *uap)
347 {
348 struct lwp *lp = curthread->td_lwp;
349 struct trapframe *regs;
350 ucontext_t uc;
351 ucontext_t *ucp;
352 register_t rflags;
353 int cs;
354 int error;
355
356 /*
357 * We have to copy the information into kernel space so userland
358 * can't modify it while we are sniffing it.
359 */
360 regs = lp->lwp_md.md_regs;
361 error = copyin(uap->sigcntxp, &uc, sizeof(uc));
362 if (error)
363 return (error);
364 ucp = &uc;
365 rflags = ucp->uc_mcontext.mc_rflags;
366
367 /* VM (8086) mode not supported */
368 rflags &= ~PSL_VM_UNSUPP;
369
370 #if 0
371 if (eflags & PSL_VM) {
372 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
373 struct vm86_kernel *vm86;
374
375 /*
376 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
377 * set up the vm86 area, and we can't enter vm86 mode.
378 */
379 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
380 return (EINVAL);
381 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
382 if (vm86->vm86_inited == 0)
383 return (EINVAL);
384
385 /* go back to user mode if both flags are set */
386 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
387 trapsignal(lp->lwp_proc, SIGBUS, 0);
388
389 if (vm86->vm86_has_vme) {
390 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
391 (eflags & VME_USERCHANGE) | PSL_VM;
392 } else {
393 vm86->vm86_eflags = eflags; /* save VIF, VIP */
394 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
395 }
396 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
397 tf->tf_eflags = eflags;
398 tf->tf_vm86_ds = tf->tf_ds;
399 tf->tf_vm86_es = tf->tf_es;
400 tf->tf_vm86_fs = tf->tf_fs;
401 tf->tf_vm86_gs = tf->tf_gs;
402 tf->tf_ds = _udatasel;
403 tf->tf_es = _udatasel;
404 #if 0
405 tf->tf_fs = _udatasel;
406 tf->tf_gs = _udatasel;
407 #endif
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(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
425 kprintf("sigreturn: rflags = 0x%lx\n", (long)rflags);
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, SIGBUS, T_PROTFLT);
438 return(EINVAL);
439 }
440 bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(struct trapframe));
441 }
442
443 /*
444 * Restore the FPU state from the frame
445 */
446 npxpop(&ucp->uc_mcontext);
447
448 if (ucp->uc_mcontext.mc_onstack & 1)
449 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
450 else
451 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
452
453 lp->lwp_sigmask = ucp->uc_sigmask;
454 SIG_CANTMASK(lp->lwp_sigmask);
455 return(EJUSTRETURN);
456 }
457
458 /*
459 * cpu_idle() represents the idle LWKT. You cannot return from this function
460 * (unless you want to blow things up!). Instead we look for runnable threads
461 * and loop or halt as appropriate. Giant is not held on entry to the thread.
462 *
463 * The main loop is entered with a critical section held, we must release
464 * the critical section before doing anything else. lwkt_switch() will
465 * check for pending interrupts due to entering and exiting its own
466 * critical section.
467 *
468 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
469 * to wake a HLTed cpu up.
470 */
471 __read_mostly static int cpu_idle_hlt = 1;
472 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
473 &cpu_idle_hlt, 0, "Idle loop HLT enable");
474
475 void
cpu_idle(void)476 cpu_idle(void)
477 {
478 struct thread *td = curthread;
479 struct mdglobaldata *gd = mdcpu;
480 int reqflags;
481
482 crit_exit();
483 KKASSERT(td->td_critcount == 0);
484 cpu_enable_intr();
485
486 for (;;) {
487 /*
488 * See if there are any LWKTs ready to go.
489 */
490 lwkt_switch();
491
492 /*
493 * The idle loop halts only if no threads are scheduleable
494 * and no signals have occured.
495 */
496 if (cpu_idle_hlt &&
497 (td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
498 splz();
499 if ((td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
500 #ifdef DEBUGIDLE
501 struct timeval tv1, tv2;
502 gettimeofday(&tv1, NULL);
503 #endif
504 reqflags = gd->mi.gd_reqflags &
505 ~RQF_IDLECHECK_WK_MASK;
506 KKASSERT(gd->mi.gd_processing_ipiq == 0);
507 umtx_sleep(&gd->mi.gd_reqflags, reqflags,
508 1000000);
509 #ifdef DEBUGIDLE
510 gettimeofday(&tv2, NULL);
511 if (tv2.tv_usec - tv1.tv_usec +
512 (tv2.tv_sec - tv1.tv_sec) * 1000000
513 > 500000) {
514 kprintf("cpu %d idlelock %08x %08x\n",
515 gd->mi.gd_cpuid,
516 gd->mi.gd_reqflags,
517 gd->gd_fpending);
518 }
519 #endif
520 }
521 } else {
522 splz();
523 __asm __volatile("pause");
524 }
525 }
526 }
527
528 /*
529 * Called by the spinlock code with or without a critical section held
530 * when a spinlock is found to be seriously constested.
531 *
532 * We need to enter a critical section to prevent signals from recursing
533 * into pthreads.
534 */
535 void
cpu_spinlock_contested(void)536 cpu_spinlock_contested(void)
537 {
538 cpu_pause();
539 }
540
541 /*
542 * Clear registers on exec
543 */
544 void
exec_setregs(u_long entry,u_long stack,u_long ps_strings)545 exec_setregs(u_long entry, u_long stack, u_long ps_strings)
546 {
547 struct thread *td = curthread;
548 struct lwp *lp = td->td_lwp;
549 struct pcb *pcb = td->td_pcb;
550 struct trapframe *regs = lp->lwp_md.md_regs;
551
552 user_ldt_free(pcb);
553
554 bzero((char *)regs, sizeof(struct trapframe));
555 regs->tf_rip = entry;
556 regs->tf_rsp = ((stack - 8) & ~0xFul) + 8; /* align the stack */
557 regs->tf_rdi = stack; /* argv */
558 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
559 regs->tf_ss = _udatasel;
560 regs->tf_cs = _ucodesel;
561 regs->tf_rbx = ps_strings;
562
563 /*
564 * Reset the hardware debug registers if they were in use.
565 * They won't have any meaning for the newly exec'd process.
566 */
567 if (pcb->pcb_flags & PCB_DBREGS) {
568 pcb->pcb_dr0 = 0;
569 pcb->pcb_dr1 = 0;
570 pcb->pcb_dr2 = 0;
571 pcb->pcb_dr3 = 0;
572 pcb->pcb_dr6 = 0;
573 pcb->pcb_dr7 = 0; /* JG set bit 10? */
574 if (pcb == td->td_pcb) {
575 /*
576 * Clear the debug registers on the running
577 * CPU, otherwise they will end up affecting
578 * the next process we switch to.
579 */
580 reset_dbregs();
581 }
582 pcb->pcb_flags &= ~PCB_DBREGS;
583 }
584
585 /*
586 * Initialize the math emulator (if any) for the current process.
587 * Actually, just clear the bit that says that the emulator has
588 * been initialized. Initialization is delayed until the process
589 * traps to the emulator (if it is done at all) mainly because
590 * emulators don't provide an entry point for initialization.
591 */
592 pcb->pcb_flags &= ~FP_SOFTFP;
593
594 /*
595 * NOTE: do not set CR0_TS here. npxinit() must do it after clearing
596 * gd_npxthread. Otherwise a preemptive interrupt thread
597 * may panic in npxdna().
598 */
599 crit_enter();
600 #if 0
601 load_cr0(rcr0() | CR0_MP);
602 #endif
603
604 /*
605 * NOTE: The MSR values must be correct so we can return to
606 * userland. gd_user_fs/gs must be correct so the switch
607 * code knows what the current MSR values are.
608 */
609 pcb->pcb_fsbase = 0; /* Values loaded from PCB on switch */
610 pcb->pcb_gsbase = 0;
611 /* Initialize the npx (if any) for the current process. */
612 npxinit();
613 crit_exit();
614
615 /*
616 * note: linux emulator needs edx to be 0x0 on entry, which is
617 * handled in execve simply by setting the 64 bit syscall
618 * return value to 0.
619 */
620 }
621
622 void
cpu_setregs(void)623 cpu_setregs(void)
624 {
625 #if 0
626 unsigned int cr0;
627
628 cr0 = rcr0();
629 cr0 |= CR0_NE; /* Done by npxinit() */
630 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
631 cr0 |= CR0_WP | CR0_AM;
632 load_cr0(cr0);
633 load_gs(_udatasel);
634 #endif
635 }
636
637 static int
sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)638 sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
639 {
640 int error;
641 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
642 req);
643 if (!error && req->newptr)
644 resettodr();
645 return (error);
646 }
647
648 SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
649 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
650
651 /*
652 * Initialize x86 and configure to run kernel
653 */
654
655 /*
656 * Initialize segments & interrupt table
657 */
658
659 extern struct user *proc0paddr;
660
661 #if 0
662
663 extern inthand_t
664 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
665 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
666 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
667 IDTVEC(page), IDTVEC(mchk), IDTVEC(rsvd), IDTVEC(fpu), IDTVEC(align),
668 IDTVEC(xmm), IDTVEC(dblfault),
669 IDTVEC(fast_syscall), IDTVEC(fast_syscall32);
670 #endif
671
672 int
ptrace_set_pc(struct lwp * lp,unsigned long addr)673 ptrace_set_pc(struct lwp *lp, unsigned long addr)
674 {
675 lp->lwp_md.md_regs->tf_rip = addr;
676 return (0);
677 }
678
679 int
ptrace_single_step(struct lwp * lp)680 ptrace_single_step(struct lwp *lp)
681 {
682 lp->lwp_md.md_regs->tf_rflags |= PSL_T;
683 return (0);
684 }
685
686 int
fill_regs(struct lwp * lp,struct reg * regs)687 fill_regs(struct lwp *lp, struct reg *regs)
688 {
689 struct trapframe *tp;
690
691 if ((tp = lp->lwp_md.md_regs) == NULL)
692 return EINVAL;
693 bcopy(&tp->tf_rdi, ®s->r_rdi, sizeof(*regs));
694 return (0);
695 }
696
697 int
set_regs(struct lwp * lp,struct reg * regs)698 set_regs(struct lwp *lp, struct reg *regs)
699 {
700 struct trapframe *tp;
701
702 tp = lp->lwp_md.md_regs;
703 if (!EFL_SECURE(regs->r_rflags, tp->tf_rflags) ||
704 !CS_SECURE(regs->r_cs))
705 return (EINVAL);
706 bcopy(®s->r_rdi, &tp->tf_rdi, sizeof(*regs));
707 return (0);
708 }
709
710 static void
fill_fpregs_xmm(struct savexmm * sv_xmm,struct save87 * sv_87)711 fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
712 {
713 struct env87 *penv_87 = &sv_87->sv_env;
714 struct envxmm *penv_xmm = &sv_xmm->sv_env;
715 int i;
716
717 /* FPU control/status */
718 penv_87->en_cw = penv_xmm->en_cw;
719 penv_87->en_sw = penv_xmm->en_sw;
720 penv_87->en_tw = penv_xmm->en_tw;
721 penv_87->en_fip = penv_xmm->en_fip;
722 penv_87->en_fcs = penv_xmm->en_fcs;
723 penv_87->en_opcode = penv_xmm->en_opcode;
724 penv_87->en_foo = penv_xmm->en_foo;
725 penv_87->en_fos = penv_xmm->en_fos;
726
727 /* FPU registers */
728 for (i = 0; i < 8; ++i)
729 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
730 }
731
732 static void
set_fpregs_xmm(struct save87 * sv_87,struct savexmm * sv_xmm)733 set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
734 {
735 struct env87 *penv_87 = &sv_87->sv_env;
736 struct envxmm *penv_xmm = &sv_xmm->sv_env;
737 int i;
738
739 /* FPU control/status */
740 penv_xmm->en_cw = penv_87->en_cw;
741 penv_xmm->en_sw = penv_87->en_sw;
742 penv_xmm->en_tw = penv_87->en_tw;
743 penv_xmm->en_fip = penv_87->en_fip;
744 penv_xmm->en_fcs = penv_87->en_fcs;
745 penv_xmm->en_opcode = penv_87->en_opcode;
746 penv_xmm->en_foo = penv_87->en_foo;
747 penv_xmm->en_fos = penv_87->en_fos;
748
749 /* FPU registers */
750 for (i = 0; i < 8; ++i)
751 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
752 }
753
754 int
fill_fpregs(struct lwp * lp,struct fpreg * fpregs)755 fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
756 {
757 if (lp->lwp_thread == NULL || lp->lwp_thread->td_pcb == NULL)
758 return EINVAL;
759 if (cpu_fxsr) {
760 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
761 (struct save87 *)fpregs);
762 return (0);
763 }
764 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
765 return (0);
766 }
767
768 int
set_fpregs(struct lwp * lp,struct fpreg * fpregs)769 set_fpregs(struct lwp *lp, struct fpreg *fpregs)
770 {
771 if (cpu_fxsr) {
772 set_fpregs_xmm((struct save87 *)fpregs,
773 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
774 return (0);
775 }
776 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
777 return (0);
778 }
779
780 int
fill_dbregs(struct lwp * lp,struct dbreg * dbregs)781 fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
782 {
783 return (ENOSYS);
784 }
785
786 int
set_dbregs(struct lwp * lp,struct dbreg * dbregs)787 set_dbregs(struct lwp *lp, struct dbreg *dbregs)
788 {
789 return (ENOSYS);
790 }
791
792 #if 0
793 /*
794 * Return > 0 if a hardware breakpoint has been hit, and the
795 * breakpoint was in user space. Return 0, otherwise.
796 */
797 int
798 user_dbreg_trap(void)
799 {
800 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
801 u_int32_t bp; /* breakpoint bits extracted from dr6 */
802 int nbp; /* number of breakpoints that triggered */
803 caddr_t addr[4]; /* breakpoint addresses */
804 int i;
805
806 dr7 = rdr7();
807 if ((dr7 & 0x000000ff) == 0) {
808 /*
809 * all GE and LE bits in the dr7 register are zero,
810 * thus the trap couldn't have been caused by the
811 * hardware debug registers
812 */
813 return 0;
814 }
815
816 nbp = 0;
817 dr6 = rdr6();
818 bp = dr6 & 0x0000000f;
819
820 if (!bp) {
821 /*
822 * None of the breakpoint bits are set meaning this
823 * trap was not caused by any of the debug registers
824 */
825 return 0;
826 }
827
828 /*
829 * at least one of the breakpoints were hit, check to see
830 * which ones and if any of them are user space addresses
831 */
832
833 if (bp & 0x01) {
834 addr[nbp++] = (caddr_t)rdr0();
835 }
836 if (bp & 0x02) {
837 addr[nbp++] = (caddr_t)rdr1();
838 }
839 if (bp & 0x04) {
840 addr[nbp++] = (caddr_t)rdr2();
841 }
842 if (bp & 0x08) {
843 addr[nbp++] = (caddr_t)rdr3();
844 }
845
846 for (i=0; i<nbp; i++) {
847 if (addr[i] <
848 (caddr_t)VM_MAX_USER_ADDRESS) {
849 /*
850 * addr[i] is in user space
851 */
852 return nbp;
853 }
854 }
855
856 /*
857 * None of the breakpoints are in user space.
858 */
859 return 0;
860 }
861
862 #endif
863
864 void
identcpu(void)865 identcpu(void)
866 {
867 int regs[4];
868
869 do_cpuid(1, regs);
870 cpu_feature = regs[3];
871 }
872
873
874 #ifndef DDB
875 void
Debugger(const char * msg)876 Debugger(const char *msg)
877 {
878 kprintf("Debugger(\"%s\") called.\n", msg);
879 }
880 #endif /* no DDB */
881