kernel: Make SMP support default (and non-optional).
[dragonfly.git] / sys / platform / vkernel64 / x86_64 / cpu_regs.c
CommitLineData
da673940
JG
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 $
da673940
JG
40 */
41
da673940
JG
42#include "opt_compat.h"
43#include "opt_ddb.h"
44#include "opt_directio.h"
45#include "opt_inet.h"
46#include "opt_ipx.h"
47#include "opt_msgbuf.h"
48#include "opt_swap.h"
49
50#include <sys/param.h>
51#include <sys/systm.h>
52#include <sys/sysproto.h>
53#include <sys/signalvar.h>
54#include <sys/kernel.h>
55#include <sys/linker.h>
56#include <sys/malloc.h>
57#include <sys/proc.h>
58#include <sys/buf.h>
59#include <sys/reboot.h>
60#include <sys/mbuf.h>
61#include <sys/msgbuf.h>
62#include <sys/sysent.h>
63#include <sys/sysctl.h>
64#include <sys/vmmeter.h>
65#include <sys/bus.h>
66#include <sys/upcall.h>
67#include <sys/usched.h>
68#include <sys/reg.h>
69
70#include <vm/vm.h>
71#include <vm/vm_param.h>
72#include <sys/lock.h>
73#include <vm/vm_kern.h>
74#include <vm/vm_object.h>
75#include <vm/vm_page.h>
76#include <vm/vm_map.h>
77#include <vm/vm_pager.h>
78#include <vm/vm_extern.h>
79
80#include <sys/thread2.h>
0e6594a8 81#include <sys/mplock2.h>
da673940
JG
82
83#include <sys/user.h>
84#include <sys/exec.h>
85#include <sys/cons.h>
86
87#include <ddb/ddb.h>
88
89#include <machine/cpu.h>
90#include <machine/clock.h>
91#include <machine/specialreg.h>
92#include <machine/md_var.h>
93#include <machine/pcb_ext.h> /* pcb.h included via sys/user.h */
94#include <machine/globaldata.h> /* CPU_prvspace */
95#include <machine/smp.h>
96#ifdef PERFMON
97#include <machine/perfmon.h>
98#endif
99#include <machine/cputypes.h>
100
101#include <bus/isa/rtc.h>
102#include <sys/random.h>
103#include <sys/ptrace.h>
104#include <machine/sigframe.h>
105#include <unistd.h> /* umtx_* functions */
b68e846f 106#include <pthread.h> /* pthread_yield() */
da673940
JG
107
108extern void dblfault_handler (void);
109
110#ifndef CPU_DISABLE_SSE
111static void set_fpregs_xmm (struct save87 *, struct savexmm *);
112static void fill_fpregs_xmm (struct savexmm *, struct save87 *);
113#endif /* CPU_DISABLE_SSE */
114#ifdef DIRECTIO
115extern void ffs_rawread_setup(void);
116#endif /* DIRECTIO */
117
da673940 118int64_t tsc_offsets[MAXCPU];
da673940
JG
119
120#if defined(SWTCH_OPTIM_STATS)
121extern int swtch_optim_stats;
122SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats,
123 CTLFLAG_RD, &swtch_optim_stats, 0, "");
124SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count,
125 CTLFLAG_RD, &tlb_flush_count, 0, "");
126#endif
127
128static int
129sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
130{
39d69dae
AH
131 u_long pmem = ctob(physmem);
132
133 int error = sysctl_handle_long(oidp, &pmem, 0, req);
da673940
JG
134 return (error);
135}
136
39d69dae 137SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_ULONG|CTLFLAG_RD,
9b9532a0 138 0, 0, sysctl_hw_physmem, "LU", "Total system memory in bytes (number of pages * page size)");
da673940
JG
139
140static int
141sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
142{
143 /* JG */
144 int error = sysctl_handle_int(oidp, 0,
145 ctob((int)Maxmem - vmstats.v_wire_count), req);
146 return (error);
147}
148
149SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD,
150 0, 0, sysctl_hw_usermem, "IU", "");
151
152SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, 0, "");
153
154#if 0
155
156static int
157sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
158{
159 int error;
160
161 /* Unwind the buffer, so that it's linear (possibly starting with
162 * some initial nulls).
163 */
164 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr,
165 msgbufp->msg_size-msgbufp->msg_bufr,req);
166 if(error) return(error);
167 if(msgbufp->msg_bufr>0) {
168 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr,
169 msgbufp->msg_bufr,req);
170 }
171 return(error);
172}
173
174SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
175 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
176
177static int msgbuf_clear;
178
179static int
180sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
181{
182 int error;
183 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
184 req);
185 if (!error && req->newptr) {
186 /* Clear the buffer and reset write pointer */
187 bzero(msgbufp->msg_ptr,msgbufp->msg_size);
188 msgbufp->msg_bufr=msgbufp->msg_bufx=0;
189 msgbuf_clear=0;
190 }
191 return (error);
192}
193
194SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
195 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
196 "Clear kernel message buffer");
197
198#endif
199
200/*
201 * Send an interrupt to process.
202 *
203 * Stack is set up to allow sigcode stored
204 * at top to call routine, followed by kcall
205 * to sigreturn routine below. After sigreturn
206 * resets the signal mask, the stack, and the
207 * frame pointer, it returns to the user
208 * specified pc, psl.
209 */
210void
211sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
212{
213 struct lwp *lp = curthread->td_lwp;
214 struct proc *p = lp->lwp_proc;
215 struct trapframe *regs;
216 struct sigacts *psp = p->p_sigacts;
217 struct sigframe sf, *sfp;
218 int oonstack;
219 char *sp;
220
221 regs = lp->lwp_md.md_regs;
222 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
223
224 /* Save user context */
225 bzero(&sf, sizeof(struct sigframe));
226 sf.sf_uc.uc_sigmask = *mask;
227 sf.sf_uc.uc_stack = lp->lwp_sigstk;
228 sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
229 KKASSERT(__offsetof(struct trapframe, tf_rdi) == 0);
230 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(struct trapframe));
231
232 /* Make the size of the saved context visible to userland */
233 sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext);
234
da673940 235 /* Allocate and validate space for the signal handler context. */
4643740a 236 if ((lp->lwp_flags & LWP_ALTSTACK) != 0 && !oonstack &&
da673940
JG
237 SIGISMEMBER(psp->ps_sigonstack, sig)) {
238 sp = (char *)(lp->lwp_sigstk.ss_sp + lp->lwp_sigstk.ss_size -
239 sizeof(struct sigframe));
240 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
241 } else {
242 /* We take red zone into account */
243 sp = (char *)regs->tf_rsp - sizeof(struct sigframe) - 128;
244 }
245
246 /* Align to 16 bytes */
247 sfp = (struct sigframe *)((intptr_t)sp & ~0xFUL);
248
249 /* Translate the signal is appropriate */
250 if (p->p_sysent->sv_sigtbl) {
251 if (sig <= p->p_sysent->sv_sigsize)
252 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
253 }
254
255 /*
256 * Build the argument list for the signal handler.
257 *
258 * Arguments are in registers (%rdi, %rsi, %rdx, %rcx)
259 */
260 regs->tf_rdi = sig; /* argument 1 */
261 regs->tf_rdx = (register_t)&sfp->sf_uc; /* argument 3 */
262
263 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
264 /*
265 * Signal handler installed with SA_SIGINFO.
266 *
267 * action(signo, siginfo, ucontext)
268 */
269 regs->tf_rsi = (register_t)&sfp->sf_si; /* argument 2 */
270 regs->tf_rcx = (register_t)regs->tf_err; /* argument 4 */
271 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
272
273 /* fill siginfo structure */
274 sf.sf_si.si_signo = sig;
275 sf.sf_si.si_code = code;
c55fa5ee 276 sf.sf_si.si_addr = (void *)regs->tf_addr;
da673940
JG
277 } else {
278 /*
279 * Old FreeBSD-style arguments.
280 *
281 * handler (signo, code, [uc], addr)
282 */
283 regs->tf_rsi = (register_t)code; /* argument 2 */
1a482e3f 284 regs->tf_rcx = (register_t)regs->tf_addr; /* argument 4 */
da673940
JG
285 sf.sf_ahu.sf_handler = catcher;
286 }
287
288#if 0
289 /*
290 * If we're a vm86 process, we want to save the segment registers.
291 * We also change eflags to be our emulated eflags, not the actual
292 * eflags.
293 */
294 if (regs->tf_eflags & PSL_VM) {
295 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
296 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
297
298 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
299 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
300 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
301 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
302
303 if (vm86->vm86_has_vme == 0)
304 sf.sf_uc.uc_mcontext.mc_eflags =
305 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
306 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
307
308 /*
309 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
310 * syscalls made by the signal handler. This just avoids
311 * wasting time for our lazy fixup of such faults. PSL_NT
312 * does nothing in vm86 mode, but vm86 programs can set it
313 * almost legitimately in probes for old cpu types.
314 */
315 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
316 }
317#endif
318
319 /*
320 * Save the FPU state and reinit the FP unit
321 */
322 npxpush(&sf.sf_uc.uc_mcontext);
323
324 /*
325 * Copy the sigframe out to the user's stack.
326 */
327 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
328 /*
329 * Something is wrong with the stack pointer.
330 * ...Kill the process.
331 */
332 sigexit(lp, SIGILL);
333 }
334
335 regs->tf_rsp = (register_t)sfp;
336 regs->tf_rip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
337
338 /*
339 * i386 abi specifies that the direction flag must be cleared
340 * on function entry
341 */
342 regs->tf_rflags &= ~(PSL_T|PSL_D);
343
344 /*
345 * 64 bit mode has a code and stack selector but
346 * no data or extra selector. %fs and %gs are not
347 * stored in-context.
348 */
349 regs->tf_cs = _ucodesel;
350 regs->tf_ss = _udatasel;
351}
352
353/*
354 * Sanitize the trapframe for a virtual kernel passing control to a custom
355 * VM context. Remove any items that would otherwise create a privilage
356 * issue.
357 *
358 * XXX at the moment we allow userland to set the resume flag. Is this a
359 * bad idea?
360 */
361int
362cpu_sanitize_frame(struct trapframe *frame)
363{
364 frame->tf_cs = _ucodesel;
365 frame->tf_ss = _udatasel;
366 /* XXX VM (8086) mode not supported? */
367 frame->tf_rflags &= (PSL_RF | PSL_USERCHANGE | PSL_VM_UNSUPP);
368 frame->tf_rflags |= PSL_RESERVED_DEFAULT | PSL_I;
369
370 return(0);
371}
372
373/*
374 * Sanitize the tls so loading the descriptor does not blow up
a76ca9b9 375 * on us. For x86_64 we don't have to do anything.
da673940
JG
376 */
377int
378cpu_sanitize_tls(struct savetls *tls)
379{
380 return(0);
381}
382
383/*
384 * sigreturn(ucontext_t *sigcntxp)
385 *
386 * System call to cleanup state after a signal
387 * has been taken. Reset signal mask and
388 * stack state from context left by sendsig (above).
389 * Return to previous pc and psl as specified by
390 * context left by sendsig. Check carefully to
391 * make sure that the user has not modified the
392 * state to gain improper privileges.
393 */
394#define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
395#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
396
397int
398sys_sigreturn(struct sigreturn_args *uap)
399{
400 struct lwp *lp = curthread->td_lwp;
da673940
JG
401 struct trapframe *regs;
402 ucontext_t uc;
403 ucontext_t *ucp;
404 register_t rflags;
405 int cs;
406 int error;
407
408 /*
409 * We have to copy the information into kernel space so userland
410 * can't modify it while we are sniffing it.
411 */
412 regs = lp->lwp_md.md_regs;
413 error = copyin(uap->sigcntxp, &uc, sizeof(uc));
414 if (error)
415 return (error);
416 ucp = &uc;
417 rflags = ucp->uc_mcontext.mc_rflags;
418
419 /* VM (8086) mode not supported */
420 rflags &= ~PSL_VM_UNSUPP;
421
422#if 0
423 if (eflags & PSL_VM) {
424 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
425 struct vm86_kernel *vm86;
426
427 /*
428 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
429 * set up the vm86 area, and we can't enter vm86 mode.
430 */
431 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
432 return (EINVAL);
433 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
434 if (vm86->vm86_inited == 0)
435 return (EINVAL);
436
437 /* go back to user mode if both flags are set */
438 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
439 trapsignal(lp->lwp_proc, SIGBUS, 0);
440
441 if (vm86->vm86_has_vme) {
442 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
443 (eflags & VME_USERCHANGE) | PSL_VM;
444 } else {
445 vm86->vm86_eflags = eflags; /* save VIF, VIP */
446 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
447 }
448 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
449 tf->tf_eflags = eflags;
450 tf->tf_vm86_ds = tf->tf_ds;
451 tf->tf_vm86_es = tf->tf_es;
452 tf->tf_vm86_fs = tf->tf_fs;
453 tf->tf_vm86_gs = tf->tf_gs;
454 tf->tf_ds = _udatasel;
455 tf->tf_es = _udatasel;
456#if 0
457 tf->tf_fs = _udatasel;
458 tf->tf_gs = _udatasel;
459#endif
460 } else
461#endif
462 {
463 /*
464 * Don't allow users to change privileged or reserved flags.
465 */
466 /*
467 * XXX do allow users to change the privileged flag PSL_RF.
468 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
469 * should sometimes set it there too. tf_eflags is kept in
470 * the signal context during signal handling and there is no
471 * other place to remember it, so the PSL_RF bit may be
472 * corrupted by the signal handler without us knowing.
473 * Corruption of the PSL_RF bit at worst causes one more or
474 * one less debugger trap, so allowing it is fairly harmless.
475 */
476 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
477 kprintf("sigreturn: rflags = 0x%lx\n", (long)rflags);
478 return(EINVAL);
479 }
480
481 /*
482 * Don't allow users to load a valid privileged %cs. Let the
483 * hardware check for invalid selectors, excess privilege in
484 * other selectors, invalid %eip's and invalid %esp's.
485 */
486 cs = ucp->uc_mcontext.mc_cs;
487 if (!CS_SECURE(cs)) {
488 kprintf("sigreturn: cs = 0x%x\n", cs);
489 trapsignal(lp, SIGBUS, T_PROTFLT);
490 return(EINVAL);
491 }
492 bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(struct trapframe));
493 }
494
495 /*
496 * Restore the FPU state from the frame
497 */
498 npxpop(&ucp->uc_mcontext);
499
da673940
JG
500 if (ucp->uc_mcontext.mc_onstack & 1)
501 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
502 else
503 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
504
505 lp->lwp_sigmask = ucp->uc_sigmask;
506 SIG_CANTMASK(lp->lwp_sigmask);
507 return(EJUSTRETURN);
508}
509
510/*
511 * Stack frame on entry to function. %rax will contain the function vector,
512 * %rcx will contain the function data. flags, rcx, and rax will have
513 * already been pushed on the stack.
514 */
515struct upc_frame {
516 register_t rax;
517 register_t rcx;
518 register_t rdx;
519 register_t flags;
520 register_t oldip;
521};
522
523void
524sendupcall(struct vmupcall *vu, int morepending)
525{
526 struct lwp *lp = curthread->td_lwp;
527 struct trapframe *regs;
528 struct upcall upcall;
529 struct upc_frame upc_frame;
530 int crit_count = 0;
531
532 /*
533 * If we are a virtual kernel running an emulated user process
534 * context, switch back to the virtual kernel context before
535 * trying to post the signal.
536 */
537 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
538 lp->lwp_md.md_regs->tf_trapno = 0;
539 vkernel_trap(lp, lp->lwp_md.md_regs);
540 }
541
542 /*
543 * Get the upcall data structure
544 */
545 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
546 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
547 ) {
548 vu->vu_pending = 0;
549 kprintf("bad upcall address\n");
550 return;
551 }
552
553 /*
554 * If the data structure is already marked pending or has a critical
555 * section count, mark the data structure as pending and return
556 * without doing an upcall. vu_pending is left set.
557 */
558 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
559 if (upcall.upc_pending < vu->vu_pending) {
560 upcall.upc_pending = vu->vu_pending;
561 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
562 sizeof(upcall.upc_pending));
563 }
564 return;
565 }
566
567 /*
568 * We can run this upcall now, clear vu_pending.
569 *
570 * Bump our critical section count and set or clear the
571 * user pending flag depending on whether more upcalls are
572 * pending. The user will be responsible for calling
573 * upc_dispatch(-1) to process remaining upcalls.
574 */
575 vu->vu_pending = 0;
576 upcall.upc_pending = morepending;
f9235b6d 577 ++crit_count;
da673940
JG
578 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
579 sizeof(upcall.upc_pending));
580 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
581 sizeof(int));
582
583 /*
584 * Construct a stack frame and issue the upcall
585 */
586 regs = lp->lwp_md.md_regs;
587 upc_frame.rax = regs->tf_rax;
588 upc_frame.rcx = regs->tf_rcx;
589 upc_frame.rdx = regs->tf_rdx;
590 upc_frame.flags = regs->tf_rflags;
591 upc_frame.oldip = regs->tf_rip;
592 if (copyout(&upc_frame, (void *)(regs->tf_rsp - sizeof(upc_frame)),
593 sizeof(upc_frame)) != 0) {
594 kprintf("bad stack on upcall\n");
595 } else {
596 regs->tf_rax = (register_t)vu->vu_func;
597 regs->tf_rcx = (register_t)vu->vu_data;
598 regs->tf_rdx = (register_t)lp->lwp_upcall;
599 regs->tf_rip = (register_t)vu->vu_ctx;
600 regs->tf_rsp -= sizeof(upc_frame);
601 }
602}
603
604/*
605 * fetchupcall occurs in the context of a system call, which means that
606 * we have to return EJUSTRETURN in order to prevent eax and edx from
607 * being overwritten by the syscall return value.
608 *
609 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
610 * and the function pointer in %eax.
611 */
612int
613fetchupcall(struct vmupcall *vu, int morepending, void *rsp)
614{
615 struct upc_frame upc_frame;
616 struct lwp *lp = curthread->td_lwp;
617 struct trapframe *regs;
618 int error;
619 struct upcall upcall;
620 int crit_count;
621
622 regs = lp->lwp_md.md_regs;
623
624 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
625 if (error == 0) {
626 if (vu) {
627 /*
628 * This jumps us to the next ready context.
629 */
630 vu->vu_pending = 0;
631 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
632 crit_count = 0;
633 if (error == 0)
634 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
f9235b6d 635 ++crit_count;
da673940
JG
636 if (error == 0)
637 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
638 regs->tf_rax = (register_t)vu->vu_func;
639 regs->tf_rcx = (register_t)vu->vu_data;
640 regs->tf_rdx = (register_t)lp->lwp_upcall;
641 regs->tf_rip = (register_t)vu->vu_ctx;
642 regs->tf_rsp = (register_t)rsp;
643 } else {
644 /*
645 * This returns us to the originally interrupted code.
646 */
647 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
648 regs->tf_rax = upc_frame.rax;
649 regs->tf_rcx = upc_frame.rcx;
650 regs->tf_rdx = upc_frame.rdx;
651 regs->tf_rflags = (regs->tf_rflags & ~PSL_USERCHANGE) |
652 (upc_frame.flags & PSL_USERCHANGE);
653 regs->tf_rip = upc_frame.oldip;
654 regs->tf_rsp = (register_t)((char *)rsp + sizeof(upc_frame));
655 }
656 }
657 if (error == 0)
658 error = EJUSTRETURN;
659 return(error);
660}
661
662/*
663 * cpu_idle() represents the idle LWKT. You cannot return from this function
664 * (unless you want to blow things up!). Instead we look for runnable threads
665 * and loop or halt as appropriate. Giant is not held on entry to the thread.
666 *
667 * The main loop is entered with a critical section held, we must release
668 * the critical section before doing anything else. lwkt_switch() will
669 * check for pending interrupts due to entering and exiting its own
670 * critical section.
671 *
672 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
b12defdc 673 * to wake a HLTed cpu up.
da673940
JG
674 */
675static int cpu_idle_hlt = 1;
676static int cpu_idle_hltcnt;
677static int cpu_idle_spincnt;
678SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
679 &cpu_idle_hlt, 0, "Idle loop HLT enable");
680SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
681 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
682SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
683 &cpu_idle_spincnt, 0, "Idle loop entry spins");
684
685void
686cpu_idle(void)
687{
688 struct thread *td = curthread;
689 struct mdglobaldata *gd = mdcpu;
c5724852 690 int reqflags;
da673940
JG
691
692 crit_exit();
f9235b6d 693 KKASSERT(td->td_critcount == 0);
da673940 694 cpu_enable_intr();
da0b0e8b 695
da673940
JG
696 for (;;) {
697 /*
698 * See if there are any LWKTs ready to go.
699 */
700 lwkt_switch();
701
702 /*
703 * The idle loop halts only if no threads are scheduleable
b12defdc 704 * and no signals have occured.
da673940 705 */
cbdd23b1
MD
706 if (cpu_idle_hlt &&
707 (td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
da673940 708 splz();
cbdd23b1 709 if ((td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
da673940
JG
710#ifdef DEBUGIDLE
711 struct timeval tv1, tv2;
712 gettimeofday(&tv1, NULL);
713#endif
c5724852 714 reqflags = gd->mi.gd_reqflags &
cbdd23b1 715 ~RQF_IDLECHECK_WK_MASK;
da0b0e8b 716 KKASSERT(gd->mi.gd_processing_ipiq == 0);
c5724852
MD
717 umtx_sleep(&gd->mi.gd_reqflags, reqflags,
718 1000000);
da673940
JG
719#ifdef DEBUGIDLE
720 gettimeofday(&tv2, NULL);
721 if (tv2.tv_usec - tv1.tv_usec +
722 (tv2.tv_sec - tv1.tv_sec) * 1000000
723 > 500000) {
724 kprintf("cpu %d idlelock %08x %08x\n",
725 gd->mi.gd_cpuid,
f9235b6d 726 gd->mi.gd_reqflags,
da673940
JG
727 gd->gd_fpending);
728 }
729#endif
730 }
da673940
JG
731 ++cpu_idle_hltcnt;
732 } else {
da673940 733 splz();
da673940 734 __asm __volatile("pause");
da673940
JG
735 ++cpu_idle_spincnt;
736 }
737 }
738}
739
da673940
JG
740/*
741 * Called by the spinlock code with or without a critical section held
742 * when a spinlock is found to be seriously constested.
743 *
744 * We need to enter a critical section to prevent signals from recursing
745 * into pthreads.
746 */
747void
748cpu_spinlock_contested(void)
749{
b68e846f 750 cpu_pause();
da673940
JG
751}
752
da673940
JG
753/*
754 * Clear registers on exec
755 */
756void
757exec_setregs(u_long entry, u_long stack, u_long ps_strings)
758{
759 struct thread *td = curthread;
760 struct lwp *lp = td->td_lwp;
761 struct pcb *pcb = td->td_pcb;
762 struct trapframe *regs = lp->lwp_md.md_regs;
763
764 /* was i386_user_cleanup() in NetBSD */
765 user_ldt_free(pcb);
766
767 bzero((char *)regs, sizeof(struct trapframe));
768 regs->tf_rip = entry;
769 regs->tf_rsp = ((stack - 8) & ~0xFul) + 8; /* align the stack */
770 regs->tf_rdi = stack; /* argv */
771 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
772 regs->tf_ss = _udatasel;
773 regs->tf_cs = _ucodesel;
774 regs->tf_rbx = ps_strings;
775
776 /*
777 * Reset the hardware debug registers if they were in use.
778 * They won't have any meaning for the newly exec'd process.
779 */
780 if (pcb->pcb_flags & PCB_DBREGS) {
781 pcb->pcb_dr0 = 0;
782 pcb->pcb_dr1 = 0;
783 pcb->pcb_dr2 = 0;
784 pcb->pcb_dr3 = 0;
785 pcb->pcb_dr6 = 0;
786 pcb->pcb_dr7 = 0; /* JG set bit 10? */
787 if (pcb == td->td_pcb) {
788 /*
789 * Clear the debug registers on the running
790 * CPU, otherwise they will end up affecting
791 * the next process we switch to.
792 */
793 reset_dbregs();
794 }
795 pcb->pcb_flags &= ~PCB_DBREGS;
796 }
797
798 /*
799 * Initialize the math emulator (if any) for the current process.
800 * Actually, just clear the bit that says that the emulator has
801 * been initialized. Initialization is delayed until the process
802 * traps to the emulator (if it is done at all) mainly because
803 * emulators don't provide an entry point for initialization.
804 */
805 pcb->pcb_flags &= ~FP_SOFTFP;
806
807 /*
808 * NOTE: do not set CR0_TS here. npxinit() must do it after clearing
809 * gd_npxthread. Otherwise a preemptive interrupt thread
810 * may panic in npxdna().
811 */
812 crit_enter();
813#if 0
814 load_cr0(rcr0() | CR0_MP);
815#endif
816
817 /*
818 * NOTE: The MSR values must be correct so we can return to
819 * userland. gd_user_fs/gs must be correct so the switch
820 * code knows what the current MSR values are.
821 */
822 pcb->pcb_fsbase = 0; /* Values loaded from PCB on switch */
823 pcb->pcb_gsbase = 0;
824 /* Initialize the npx (if any) for the current process. */
a8f1df17 825 npxinit(__INITIAL_FPUCW__);
da673940
JG
826 crit_exit();
827
828 /*
829 * note: linux emulator needs edx to be 0x0 on entry, which is
830 * handled in execve simply by setting the 64 bit syscall
831 * return value to 0.
832 */
833}
834
835void
836cpu_setregs(void)
837{
838#if 0
839 unsigned int cr0;
840
841 cr0 = rcr0();
842 cr0 |= CR0_NE; /* Done by npxinit() */
843 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
844 cr0 |= CR0_WP | CR0_AM;
845 load_cr0(cr0);
846 load_gs(_udatasel);
847#endif
848}
849
850static int
851sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
852{
853 int error;
854 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
855 req);
856 if (!error && req->newptr)
857 resettodr();
858 return (error);
859}
860
861SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
862 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
863
864extern u_long bootdev; /* not a cdev_t - encoding is different */
865SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
866 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
867
868/*
869 * Initialize 386 and configure to run kernel
870 */
871
872/*
873 * Initialize segments & interrupt table
874 */
875
876extern struct user *proc0paddr;
877
878#if 0
879
880extern inthand_t
881 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
882 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
883 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
884 IDTVEC(page), IDTVEC(mchk), IDTVEC(rsvd), IDTVEC(fpu), IDTVEC(align),
885 IDTVEC(xmm), IDTVEC(dblfault),
886 IDTVEC(fast_syscall), IDTVEC(fast_syscall32);
887#endif
888
889#ifdef DEBUG_INTERRUPTS
890extern inthand_t *Xrsvdary[256];
891#endif
892
893int
894ptrace_set_pc(struct lwp *lp, unsigned long addr)
895{
896 lp->lwp_md.md_regs->tf_rip = addr;
897 return (0);
898}
899
900int
901ptrace_single_step(struct lwp *lp)
902{
903 lp->lwp_md.md_regs->tf_rflags |= PSL_T;
904 return (0);
905}
906
907int
908fill_regs(struct lwp *lp, struct reg *regs)
909{
da673940
JG
910 struct trapframe *tp;
911
d64d3805
MD
912 if ((tp = lp->lwp_md.md_regs) == NULL)
913 return EINVAL;
da673940 914 bcopy(&tp->tf_rdi, &regs->r_rdi, sizeof(*regs));
da673940
JG
915 return (0);
916}
917
918int
919set_regs(struct lwp *lp, struct reg *regs)
920{
da673940
JG
921 struct trapframe *tp;
922
923 tp = lp->lwp_md.md_regs;
924 if (!EFL_SECURE(regs->r_rflags, tp->tf_rflags) ||
925 !CS_SECURE(regs->r_cs))
926 return (EINVAL);
927 bcopy(&regs->r_rdi, &tp->tf_rdi, sizeof(*regs));
da673940
JG
928 return (0);
929}
930
931#ifndef CPU_DISABLE_SSE
932static void
933fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
934{
935 struct env87 *penv_87 = &sv_87->sv_env;
936 struct envxmm *penv_xmm = &sv_xmm->sv_env;
937 int i;
938
939 /* FPU control/status */
940 penv_87->en_cw = penv_xmm->en_cw;
941 penv_87->en_sw = penv_xmm->en_sw;
942 penv_87->en_tw = penv_xmm->en_tw;
943 penv_87->en_fip = penv_xmm->en_fip;
944 penv_87->en_fcs = penv_xmm->en_fcs;
945 penv_87->en_opcode = penv_xmm->en_opcode;
946 penv_87->en_foo = penv_xmm->en_foo;
947 penv_87->en_fos = penv_xmm->en_fos;
948
949 /* FPU registers */
950 for (i = 0; i < 8; ++i)
951 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
da673940
JG
952}
953
954static void
955set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
956{
957 struct env87 *penv_87 = &sv_87->sv_env;
958 struct envxmm *penv_xmm = &sv_xmm->sv_env;
959 int i;
960
961 /* FPU control/status */
962 penv_xmm->en_cw = penv_87->en_cw;
963 penv_xmm->en_sw = penv_87->en_sw;
964 penv_xmm->en_tw = penv_87->en_tw;
965 penv_xmm->en_fip = penv_87->en_fip;
966 penv_xmm->en_fcs = penv_87->en_fcs;
967 penv_xmm->en_opcode = penv_87->en_opcode;
968 penv_xmm->en_foo = penv_87->en_foo;
969 penv_xmm->en_fos = penv_87->en_fos;
970
971 /* FPU registers */
972 for (i = 0; i < 8; ++i)
973 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
da673940
JG
974}
975#endif /* CPU_DISABLE_SSE */
976
977int
978fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
979{
d64d3805
MD
980 if (lp->lwp_thread == NULL || lp->lwp_thread->td_pcb == NULL)
981 return EINVAL;
da673940
JG
982#ifndef CPU_DISABLE_SSE
983 if (cpu_fxsr) {
984 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
985 (struct save87 *)fpregs);
986 return (0);
987 }
988#endif /* CPU_DISABLE_SSE */
989 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
990 return (0);
991}
992
993int
994set_fpregs(struct lwp *lp, struct fpreg *fpregs)
995{
996#ifndef CPU_DISABLE_SSE
997 if (cpu_fxsr) {
998 set_fpregs_xmm((struct save87 *)fpregs,
999 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
1000 return (0);
1001 }
1002#endif /* CPU_DISABLE_SSE */
1003 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
1004 return (0);
1005}
1006
1007int
1008fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
1009{
1010 return (ENOSYS);
1011}
1012
1013int
1014set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1015{
1016 return (ENOSYS);
1017}
1018
1019#if 0
1020/*
1021 * Return > 0 if a hardware breakpoint has been hit, and the
1022 * breakpoint was in user space. Return 0, otherwise.
1023 */
1024int
1025user_dbreg_trap(void)
1026{
1027 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1028 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1029 int nbp; /* number of breakpoints that triggered */
1030 caddr_t addr[4]; /* breakpoint addresses */
1031 int i;
1032
1033 dr7 = rdr7();
1034 if ((dr7 & 0x000000ff) == 0) {
1035 /*
1036 * all GE and LE bits in the dr7 register are zero,
1037 * thus the trap couldn't have been caused by the
1038 * hardware debug registers
1039 */
1040 return 0;
1041 }
1042
1043 nbp = 0;
1044 dr6 = rdr6();
1045 bp = dr6 & 0x0000000f;
1046
1047 if (!bp) {
1048 /*
1049 * None of the breakpoint bits are set meaning this
1050 * trap was not caused by any of the debug registers
1051 */
1052 return 0;
1053 }
1054
1055 /*
1056 * at least one of the breakpoints were hit, check to see
1057 * which ones and if any of them are user space addresses
1058 */
1059
1060 if (bp & 0x01) {
1061 addr[nbp++] = (caddr_t)rdr0();
1062 }
1063 if (bp & 0x02) {
1064 addr[nbp++] = (caddr_t)rdr1();
1065 }
1066 if (bp & 0x04) {
1067 addr[nbp++] = (caddr_t)rdr2();
1068 }
1069 if (bp & 0x08) {
1070 addr[nbp++] = (caddr_t)rdr3();
1071 }
1072
1073 for (i=0; i<nbp; i++) {
1074 if (addr[i] <
1075 (caddr_t)VM_MAX_USER_ADDRESS) {
1076 /*
1077 * addr[i] is in user space
1078 */
1079 return nbp;
1080 }
1081 }
1082
1083 /*
1084 * None of the breakpoints are in user space.
1085 */
1086 return 0;
1087}
1088
1089#endif
1090
1091void
1092identcpu(void)
1093{
1094 int regs[4];
1095
1096 do_cpuid(1, regs);
1097 cpu_feature = regs[3];
1098}
1099
1100
1101#ifndef DDB
1102void
1103Debugger(const char *msg)
1104{
1105 kprintf("Debugger(\"%s\") called.\n", msg);
1106}
1107#endif /* no DDB */