Add a missing break in the 64 bit vkernel, too.
[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 $
40 * $DragonFly: src/sys/platform/vkernel/i386/cpu_regs.c,v 1.29 2008/06/06 13:19:25 swildner Exp $
41 */
42
43#include "use_ether.h"
44#include "use_isa.h"
45#include "opt_atalk.h"
46#include "opt_compat.h"
47#include "opt_ddb.h"
48#include "opt_directio.h"
49#include "opt_inet.h"
50#include "opt_ipx.h"
51#include "opt_msgbuf.h"
52#include "opt_swap.h"
53
54#include <sys/param.h>
55#include <sys/systm.h>
56#include <sys/sysproto.h>
57#include <sys/signalvar.h>
58#include <sys/kernel.h>
59#include <sys/linker.h>
60#include <sys/malloc.h>
61#include <sys/proc.h>
62#include <sys/buf.h>
63#include <sys/reboot.h>
64#include <sys/mbuf.h>
65#include <sys/msgbuf.h>
66#include <sys/sysent.h>
67#include <sys/sysctl.h>
68#include <sys/vmmeter.h>
69#include <sys/bus.h>
70#include <sys/upcall.h>
71#include <sys/usched.h>
72#include <sys/reg.h>
73
74#include <vm/vm.h>
75#include <vm/vm_param.h>
76#include <sys/lock.h>
77#include <vm/vm_kern.h>
78#include <vm/vm_object.h>
79#include <vm/vm_page.h>
80#include <vm/vm_map.h>
81#include <vm/vm_pager.h>
82#include <vm/vm_extern.h>
83
84#include <sys/thread2.h>
0e6594a8 85#include <sys/mplock2.h>
da673940
JG
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 <sys/random.h>
107#include <sys/ptrace.h>
108#include <machine/sigframe.h>
109#include <unistd.h> /* umtx_* functions */
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
135static int
136sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
137{
138 /* JG */
139 int error = sysctl_handle_int(oidp, 0, ctob((int)Maxmem), req);
140 return (error);
141}
142
143SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_INT|CTLFLAG_RD,
144 0, 0, sysctl_hw_physmem, "IU", "");
145
146static int
147sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
148{
149 /* JG */
150 int error = sysctl_handle_int(oidp, 0,
151 ctob((int)Maxmem - 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
158SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, 0, "");
159
160#if 0
161
162static int
163sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
164{
165 int error;
166
167 /* Unwind the buffer, so that it's linear (possibly starting with
168 * some initial nulls).
169 */
170 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr,
171 msgbufp->msg_size-msgbufp->msg_bufr,req);
172 if(error) return(error);
173 if(msgbufp->msg_bufr>0) {
174 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr,
175 msgbufp->msg_bufr,req);
176 }
177 return(error);
178}
179
180SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
181 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
182
183static int msgbuf_clear;
184
185static int
186sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
187{
188 int error;
189 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
190 req);
191 if (!error && req->newptr) {
192 /* Clear the buffer and reset write pointer */
193 bzero(msgbufp->msg_ptr,msgbufp->msg_size);
194 msgbufp->msg_bufr=msgbufp->msg_bufx=0;
195 msgbuf_clear=0;
196 }
197 return (error);
198}
199
200SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
201 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
202 "Clear kernel message buffer");
203
204#endif
205
206/*
207 * Send an interrupt to process.
208 *
209 * Stack is set up to allow sigcode stored
210 * at top to call routine, followed by kcall
211 * to sigreturn routine below. After sigreturn
212 * resets the signal mask, the stack, and the
213 * frame pointer, it returns to the user
214 * specified pc, psl.
215 */
216void
217sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
218{
219 struct lwp *lp = curthread->td_lwp;
220 struct proc *p = lp->lwp_proc;
221 struct trapframe *regs;
222 struct sigacts *psp = p->p_sigacts;
223 struct sigframe sf, *sfp;
224 int oonstack;
225 char *sp;
226
227 regs = lp->lwp_md.md_regs;
228 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
229
230 /* Save user context */
231 bzero(&sf, sizeof(struct sigframe));
232 sf.sf_uc.uc_sigmask = *mask;
233 sf.sf_uc.uc_stack = lp->lwp_sigstk;
234 sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
235 KKASSERT(__offsetof(struct trapframe, tf_rdi) == 0);
236 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(struct trapframe));
237
238 /* Make the size of the saved context visible to userland */
239 sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext);
240
241 /* Save mailbox pending state for syscall interlock semantics */
242 if (p->p_flag & P_MAILBOX)
243 sf.sf_uc.uc_mcontext.mc_xflags |= PGEX_MAILBOX;
244
245 /* Allocate and validate space for the signal handler context. */
246 if ((lp->lwp_flag & LWP_ALTSTACK) != 0 && !oonstack &&
247 SIGISMEMBER(psp->ps_sigonstack, sig)) {
248 sp = (char *)(lp->lwp_sigstk.ss_sp + lp->lwp_sigstk.ss_size -
249 sizeof(struct sigframe));
250 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
251 } else {
252 /* We take red zone into account */
253 sp = (char *)regs->tf_rsp - sizeof(struct sigframe) - 128;
254 }
255
256 /* Align to 16 bytes */
257 sfp = (struct sigframe *)((intptr_t)sp & ~0xFUL);
258
259 /* Translate the signal is appropriate */
260 if (p->p_sysent->sv_sigtbl) {
261 if (sig <= p->p_sysent->sv_sigsize)
262 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
263 }
264
265 /*
266 * Build the argument list for the signal handler.
267 *
268 * Arguments are in registers (%rdi, %rsi, %rdx, %rcx)
269 */
270 regs->tf_rdi = sig; /* argument 1 */
271 regs->tf_rdx = (register_t)&sfp->sf_uc; /* argument 3 */
272
273 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
274 /*
275 * Signal handler installed with SA_SIGINFO.
276 *
277 * action(signo, siginfo, ucontext)
278 */
279 regs->tf_rsi = (register_t)&sfp->sf_si; /* argument 2 */
280 regs->tf_rcx = (register_t)regs->tf_err; /* argument 4 */
281 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
282
283 /* fill siginfo structure */
284 sf.sf_si.si_signo = sig;
285 sf.sf_si.si_code = code;
286 sf.sf_si.si_addr = (void *)regs->tf_err;
287 } else {
288 /*
289 * Old FreeBSD-style arguments.
290 *
291 * handler (signo, code, [uc], addr)
292 */
293 regs->tf_rsi = (register_t)code; /* argument 2 */
294 regs->tf_rcx = (register_t)regs->tf_err; /* argument 4 */
295 sf.sf_ahu.sf_handler = catcher;
296 }
297
298#if 0
299 /*
300 * If we're a vm86 process, we want to save the segment registers.
301 * We also change eflags to be our emulated eflags, not the actual
302 * eflags.
303 */
304 if (regs->tf_eflags & PSL_VM) {
305 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
306 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
307
308 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
309 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
310 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
311 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
312
313 if (vm86->vm86_has_vme == 0)
314 sf.sf_uc.uc_mcontext.mc_eflags =
315 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
316 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
317
318 /*
319 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
320 * syscalls made by the signal handler. This just avoids
321 * wasting time for our lazy fixup of such faults. PSL_NT
322 * does nothing in vm86 mode, but vm86 programs can set it
323 * almost legitimately in probes for old cpu types.
324 */
325 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
326 }
327#endif
328
329 /*
330 * Save the FPU state and reinit the FP unit
331 */
332 npxpush(&sf.sf_uc.uc_mcontext);
333
334 /*
335 * Copy the sigframe out to the user's stack.
336 */
337 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
338 /*
339 * Something is wrong with the stack pointer.
340 * ...Kill the process.
341 */
342 sigexit(lp, SIGILL);
343 }
344
345 regs->tf_rsp = (register_t)sfp;
346 regs->tf_rip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
347
348 /*
349 * i386 abi specifies that the direction flag must be cleared
350 * on function entry
351 */
352 regs->tf_rflags &= ~(PSL_T|PSL_D);
353
354 /*
355 * 64 bit mode has a code and stack selector but
356 * no data or extra selector. %fs and %gs are not
357 * stored in-context.
358 */
359 regs->tf_cs = _ucodesel;
360 regs->tf_ss = _udatasel;
361}
362
363/*
364 * Sanitize the trapframe for a virtual kernel passing control to a custom
365 * VM context. Remove any items that would otherwise create a privilage
366 * issue.
367 *
368 * XXX at the moment we allow userland to set the resume flag. Is this a
369 * bad idea?
370 */
371int
372cpu_sanitize_frame(struct trapframe *frame)
373{
374 frame->tf_cs = _ucodesel;
375 frame->tf_ss = _udatasel;
376 /* XXX VM (8086) mode not supported? */
377 frame->tf_rflags &= (PSL_RF | PSL_USERCHANGE | PSL_VM_UNSUPP);
378 frame->tf_rflags |= PSL_RESERVED_DEFAULT | PSL_I;
379
380 return(0);
381}
382
383/*
384 * Sanitize the tls so loading the descriptor does not blow up
385 * on us. For AMD64 we don't have to do anything.
386 */
387int
388cpu_sanitize_tls(struct savetls *tls)
389{
390 return(0);
391}
392
393/*
394 * sigreturn(ucontext_t *sigcntxp)
395 *
396 * System call to cleanup state after a signal
397 * has been taken. Reset signal mask and
398 * stack state from context left by sendsig (above).
399 * Return to previous pc and psl as specified by
400 * context left by sendsig. Check carefully to
401 * make sure that the user has not modified the
402 * state to gain improper privileges.
403 */
404#define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
405#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
406
407int
408sys_sigreturn(struct sigreturn_args *uap)
409{
410 struct lwp *lp = curthread->td_lwp;
411 struct proc *p = lp->lwp_proc;
412 struct trapframe *regs;
413 ucontext_t uc;
414 ucontext_t *ucp;
415 register_t rflags;
416 int cs;
417 int error;
418
419 /*
420 * We have to copy the information into kernel space so userland
421 * can't modify it while we are sniffing it.
422 */
423 regs = lp->lwp_md.md_regs;
424 error = copyin(uap->sigcntxp, &uc, sizeof(uc));
425 if (error)
426 return (error);
427 ucp = &uc;
428 rflags = ucp->uc_mcontext.mc_rflags;
429
430 /* VM (8086) mode not supported */
431 rflags &= ~PSL_VM_UNSUPP;
432
433#if 0
434 if (eflags & PSL_VM) {
435 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
436 struct vm86_kernel *vm86;
437
438 /*
439 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
440 * set up the vm86 area, and we can't enter vm86 mode.
441 */
442 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
443 return (EINVAL);
444 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
445 if (vm86->vm86_inited == 0)
446 return (EINVAL);
447
448 /* go back to user mode if both flags are set */
449 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
450 trapsignal(lp->lwp_proc, SIGBUS, 0);
451
452 if (vm86->vm86_has_vme) {
453 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
454 (eflags & VME_USERCHANGE) | PSL_VM;
455 } else {
456 vm86->vm86_eflags = eflags; /* save VIF, VIP */
457 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
458 }
459 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
460 tf->tf_eflags = eflags;
461 tf->tf_vm86_ds = tf->tf_ds;
462 tf->tf_vm86_es = tf->tf_es;
463 tf->tf_vm86_fs = tf->tf_fs;
464 tf->tf_vm86_gs = tf->tf_gs;
465 tf->tf_ds = _udatasel;
466 tf->tf_es = _udatasel;
467#if 0
468 tf->tf_fs = _udatasel;
469 tf->tf_gs = _udatasel;
470#endif
471 } else
472#endif
473 {
474 /*
475 * Don't allow users to change privileged or reserved flags.
476 */
477 /*
478 * XXX do allow users to change the privileged flag PSL_RF.
479 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
480 * should sometimes set it there too. tf_eflags is kept in
481 * the signal context during signal handling and there is no
482 * other place to remember it, so the PSL_RF bit may be
483 * corrupted by the signal handler without us knowing.
484 * Corruption of the PSL_RF bit at worst causes one more or
485 * one less debugger trap, so allowing it is fairly harmless.
486 */
487 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
488 kprintf("sigreturn: rflags = 0x%lx\n", (long)rflags);
489 return(EINVAL);
490 }
491
492 /*
493 * Don't allow users to load a valid privileged %cs. Let the
494 * hardware check for invalid selectors, excess privilege in
495 * other selectors, invalid %eip's and invalid %esp's.
496 */
497 cs = ucp->uc_mcontext.mc_cs;
498 if (!CS_SECURE(cs)) {
499 kprintf("sigreturn: cs = 0x%x\n", cs);
500 trapsignal(lp, SIGBUS, T_PROTFLT);
501 return(EINVAL);
502 }
503 bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(struct trapframe));
504 }
505
506 /*
507 * Restore the FPU state from the frame
508 */
509 npxpop(&ucp->uc_mcontext);
510
511 /*
512 * Merge saved signal mailbox pending flag to maintain interlock
513 * semantics against system calls.
514 */
515 if (ucp->uc_mcontext.mc_xflags & PGEX_MAILBOX)
516 p->p_flag |= P_MAILBOX;
517
518 if (ucp->uc_mcontext.mc_onstack & 1)
519 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
520 else
521 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
522
523 lp->lwp_sigmask = ucp->uc_sigmask;
524 SIG_CANTMASK(lp->lwp_sigmask);
525 return(EJUSTRETURN);
526}
527
528/*
529 * Stack frame on entry to function. %rax will contain the function vector,
530 * %rcx will contain the function data. flags, rcx, and rax will have
531 * already been pushed on the stack.
532 */
533struct upc_frame {
534 register_t rax;
535 register_t rcx;
536 register_t rdx;
537 register_t flags;
538 register_t oldip;
539};
540
541void
542sendupcall(struct vmupcall *vu, int morepending)
543{
544 struct lwp *lp = curthread->td_lwp;
545 struct trapframe *regs;
546 struct upcall upcall;
547 struct upc_frame upc_frame;
548 int crit_count = 0;
549
550 /*
551 * If we are a virtual kernel running an emulated user process
552 * context, switch back to the virtual kernel context before
553 * trying to post the signal.
554 */
555 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
556 lp->lwp_md.md_regs->tf_trapno = 0;
557 vkernel_trap(lp, lp->lwp_md.md_regs);
558 }
559
560 /*
561 * Get the upcall data structure
562 */
563 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
564 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
565 ) {
566 vu->vu_pending = 0;
567 kprintf("bad upcall address\n");
568 return;
569 }
570
571 /*
572 * If the data structure is already marked pending or has a critical
573 * section count, mark the data structure as pending and return
574 * without doing an upcall. vu_pending is left set.
575 */
576 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
577 if (upcall.upc_pending < vu->vu_pending) {
578 upcall.upc_pending = vu->vu_pending;
579 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
580 sizeof(upcall.upc_pending));
581 }
582 return;
583 }
584
585 /*
586 * We can run this upcall now, clear vu_pending.
587 *
588 * Bump our critical section count and set or clear the
589 * user pending flag depending on whether more upcalls are
590 * pending. The user will be responsible for calling
591 * upc_dispatch(-1) to process remaining upcalls.
592 */
593 vu->vu_pending = 0;
594 upcall.upc_pending = morepending;
595 crit_count += TDPRI_CRIT;
596 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
597 sizeof(upcall.upc_pending));
598 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
599 sizeof(int));
600
601 /*
602 * Construct a stack frame and issue the upcall
603 */
604 regs = lp->lwp_md.md_regs;
605 upc_frame.rax = regs->tf_rax;
606 upc_frame.rcx = regs->tf_rcx;
607 upc_frame.rdx = regs->tf_rdx;
608 upc_frame.flags = regs->tf_rflags;
609 upc_frame.oldip = regs->tf_rip;
610 if (copyout(&upc_frame, (void *)(regs->tf_rsp - sizeof(upc_frame)),
611 sizeof(upc_frame)) != 0) {
612 kprintf("bad stack on upcall\n");
613 } else {
614 regs->tf_rax = (register_t)vu->vu_func;
615 regs->tf_rcx = (register_t)vu->vu_data;
616 regs->tf_rdx = (register_t)lp->lwp_upcall;
617 regs->tf_rip = (register_t)vu->vu_ctx;
618 regs->tf_rsp -= sizeof(upc_frame);
619 }
620}
621
622/*
623 * fetchupcall occurs in the context of a system call, which means that
624 * we have to return EJUSTRETURN in order to prevent eax and edx from
625 * being overwritten by the syscall return value.
626 *
627 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
628 * and the function pointer in %eax.
629 */
630int
631fetchupcall(struct vmupcall *vu, int morepending, void *rsp)
632{
633 struct upc_frame upc_frame;
634 struct lwp *lp = curthread->td_lwp;
635 struct trapframe *regs;
636 int error;
637 struct upcall upcall;
638 int crit_count;
639
640 regs = lp->lwp_md.md_regs;
641
642 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
643 if (error == 0) {
644 if (vu) {
645 /*
646 * This jumps us to the next ready context.
647 */
648 vu->vu_pending = 0;
649 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
650 crit_count = 0;
651 if (error == 0)
652 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
653 crit_count += TDPRI_CRIT;
654 if (error == 0)
655 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
656 regs->tf_rax = (register_t)vu->vu_func;
657 regs->tf_rcx = (register_t)vu->vu_data;
658 regs->tf_rdx = (register_t)lp->lwp_upcall;
659 regs->tf_rip = (register_t)vu->vu_ctx;
660 regs->tf_rsp = (register_t)rsp;
661 } else {
662 /*
663 * This returns us to the originally interrupted code.
664 */
665 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
666 regs->tf_rax = upc_frame.rax;
667 regs->tf_rcx = upc_frame.rcx;
668 regs->tf_rdx = upc_frame.rdx;
669 regs->tf_rflags = (regs->tf_rflags & ~PSL_USERCHANGE) |
670 (upc_frame.flags & PSL_USERCHANGE);
671 regs->tf_rip = upc_frame.oldip;
672 regs->tf_rsp = (register_t)((char *)rsp + sizeof(upc_frame));
673 }
674 }
675 if (error == 0)
676 error = EJUSTRETURN;
677 return(error);
678}
679
680/*
681 * cpu_idle() represents the idle LWKT. You cannot return from this function
682 * (unless you want to blow things up!). Instead we look for runnable threads
683 * and loop or halt as appropriate. Giant is not held on entry to the thread.
684 *
685 * The main loop is entered with a critical section held, we must release
686 * the critical section before doing anything else. lwkt_switch() will
687 * check for pending interrupts due to entering and exiting its own
688 * critical section.
689 *
690 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
691 * to wake a HLTed cpu up. However, there are cases where the idlethread
692 * will be entered with the possibility that no IPI will occur and in such
693 * cases lwkt_switch() sets TDF_IDLE_NOHLT.
694 */
695static int cpu_idle_hlt = 1;
696static int cpu_idle_hltcnt;
697static int cpu_idle_spincnt;
698SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
699 &cpu_idle_hlt, 0, "Idle loop HLT enable");
700SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
701 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
702SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
703 &cpu_idle_spincnt, 0, "Idle loop entry spins");
704
705void
706cpu_idle(void)
707{
708 struct thread *td = curthread;
709 struct mdglobaldata *gd = mdcpu;
710
711 crit_exit();
712 KKASSERT(td->td_pri < TDPRI_CRIT);
713 cpu_enable_intr();
714 for (;;) {
715 /*
716 * See if there are any LWKTs ready to go.
717 */
718 lwkt_switch();
719
720 /*
721 * The idle loop halts only if no threads are scheduleable
722 * and no signals have occured.
723 */
724 if (cpu_idle_hlt && !lwkt_runnable() &&
725 (td->td_flags & TDF_IDLE_NOHLT) == 0) {
726 splz();
727 if (!lwkt_runnable()) {
728#ifdef DEBUGIDLE
729 struct timeval tv1, tv2;
730 gettimeofday(&tv1, NULL);
731#endif
732 umtx_sleep(&gd->mi.gd_runqmask, 0, 1000000);
733#ifdef DEBUGIDLE
734 gettimeofday(&tv2, NULL);
735 if (tv2.tv_usec - tv1.tv_usec +
736 (tv2.tv_sec - tv1.tv_sec) * 1000000
737 > 500000) {
738 kprintf("cpu %d idlelock %08x %08x\n",
739 gd->mi.gd_cpuid,
740 gd->mi.gd_runqmask,
741 gd->gd_fpending);
742 }
743#endif
744 }
745#ifdef SMP
746 else {
747 __asm __volatile("pause");
748 }
749#endif
750 ++cpu_idle_hltcnt;
751 } else {
752 td->td_flags &= ~TDF_IDLE_NOHLT;
753 splz();
754#ifdef SMP
755 /*__asm __volatile("sti; pause");*/
756 __asm __volatile("pause");
757#else
758 /*__asm __volatile("sti");*/
759#endif
760 ++cpu_idle_spincnt;
761 }
762 }
763}
764
765#ifdef SMP
766
767/*
768 * Called by the LWKT switch core with a critical section held if the only
769 * schedulable thread needs the MP lock and we couldn't get it. On
770 * a real cpu we just spin in the scheduler. In the virtual kernel
771 * we sleep for a bit.
772 */
773void
774cpu_mplock_contested(void)
775{
776 usleep(1000);
777}
778
779/*
780 * Called by the spinlock code with or without a critical section held
781 * when a spinlock is found to be seriously constested.
782 *
783 * We need to enter a critical section to prevent signals from recursing
784 * into pthreads.
785 */
786void
787cpu_spinlock_contested(void)
788{
789 crit_enter();
790 usleep(1000);
791 crit_exit();
792}
793
794#endif
795
796/*
797 * Clear registers on exec
798 */
799void
800exec_setregs(u_long entry, u_long stack, u_long ps_strings)
801{
802 struct thread *td = curthread;
803 struct lwp *lp = td->td_lwp;
804 struct pcb *pcb = td->td_pcb;
805 struct trapframe *regs = lp->lwp_md.md_regs;
806
807 /* was i386_user_cleanup() in NetBSD */
808 user_ldt_free(pcb);
809
810 bzero((char *)regs, sizeof(struct trapframe));
811 regs->tf_rip = entry;
812 regs->tf_rsp = ((stack - 8) & ~0xFul) + 8; /* align the stack */
813 regs->tf_rdi = stack; /* argv */
814 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
815 regs->tf_ss = _udatasel;
816 regs->tf_cs = _ucodesel;
817 regs->tf_rbx = ps_strings;
818
819 /*
820 * Reset the hardware debug registers if they were in use.
821 * They won't have any meaning for the newly exec'd process.
822 */
823 if (pcb->pcb_flags & PCB_DBREGS) {
824 pcb->pcb_dr0 = 0;
825 pcb->pcb_dr1 = 0;
826 pcb->pcb_dr2 = 0;
827 pcb->pcb_dr3 = 0;
828 pcb->pcb_dr6 = 0;
829 pcb->pcb_dr7 = 0; /* JG set bit 10? */
830 if (pcb == td->td_pcb) {
831 /*
832 * Clear the debug registers on the running
833 * CPU, otherwise they will end up affecting
834 * the next process we switch to.
835 */
836 reset_dbregs();
837 }
838 pcb->pcb_flags &= ~PCB_DBREGS;
839 }
840
841 /*
842 * Initialize the math emulator (if any) for the current process.
843 * Actually, just clear the bit that says that the emulator has
844 * been initialized. Initialization is delayed until the process
845 * traps to the emulator (if it is done at all) mainly because
846 * emulators don't provide an entry point for initialization.
847 */
848 pcb->pcb_flags &= ~FP_SOFTFP;
849
850 /*
851 * NOTE: do not set CR0_TS here. npxinit() must do it after clearing
852 * gd_npxthread. Otherwise a preemptive interrupt thread
853 * may panic in npxdna().
854 */
855 crit_enter();
856#if 0
857 load_cr0(rcr0() | CR0_MP);
858#endif
859
860 /*
861 * NOTE: The MSR values must be correct so we can return to
862 * userland. gd_user_fs/gs must be correct so the switch
863 * code knows what the current MSR values are.
864 */
865 pcb->pcb_fsbase = 0; /* Values loaded from PCB on switch */
866 pcb->pcb_gsbase = 0;
867 /* Initialize the npx (if any) for the current process. */
868 npxinit(__INITIAL_NPXCW__);
869 crit_exit();
870
871 /*
872 * note: linux emulator needs edx to be 0x0 on entry, which is
873 * handled in execve simply by setting the 64 bit syscall
874 * return value to 0.
875 */
876}
877
878void
879cpu_setregs(void)
880{
881#if 0
882 unsigned int cr0;
883
884 cr0 = rcr0();
885 cr0 |= CR0_NE; /* Done by npxinit() */
886 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
887 cr0 |= CR0_WP | CR0_AM;
888 load_cr0(cr0);
889 load_gs(_udatasel);
890#endif
891}
892
893static int
894sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
895{
896 int error;
897 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
898 req);
899 if (!error && req->newptr)
900 resettodr();
901 return (error);
902}
903
904SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
905 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
906
907extern u_long bootdev; /* not a cdev_t - encoding is different */
908SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
909 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
910
911/*
912 * Initialize 386 and configure to run kernel
913 */
914
915/*
916 * Initialize segments & interrupt table
917 */
918
919extern struct user *proc0paddr;
920
921#if 0
922
923extern inthand_t
924 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
925 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
926 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
927 IDTVEC(page), IDTVEC(mchk), IDTVEC(rsvd), IDTVEC(fpu), IDTVEC(align),
928 IDTVEC(xmm), IDTVEC(dblfault),
929 IDTVEC(fast_syscall), IDTVEC(fast_syscall32);
930#endif
931
932#ifdef DEBUG_INTERRUPTS
933extern inthand_t *Xrsvdary[256];
934#endif
935
936int
937ptrace_set_pc(struct lwp *lp, unsigned long addr)
938{
939 lp->lwp_md.md_regs->tf_rip = addr;
940 return (0);
941}
942
943int
944ptrace_single_step(struct lwp *lp)
945{
946 lp->lwp_md.md_regs->tf_rflags |= PSL_T;
947 return (0);
948}
949
950int
951fill_regs(struct lwp *lp, struct reg *regs)
952{
953 struct pcb *pcb;
954 struct trapframe *tp;
955
956 tp = lp->lwp_md.md_regs;
957 bcopy(&tp->tf_rdi, &regs->r_rdi, sizeof(*regs));
958
959 pcb = lp->lwp_thread->td_pcb;
960 return (0);
961}
962
963int
964set_regs(struct lwp *lp, struct reg *regs)
965{
966 struct pcb *pcb;
967 struct trapframe *tp;
968
969 tp = lp->lwp_md.md_regs;
970 if (!EFL_SECURE(regs->r_rflags, tp->tf_rflags) ||
971 !CS_SECURE(regs->r_cs))
972 return (EINVAL);
973 bcopy(&regs->r_rdi, &tp->tf_rdi, sizeof(*regs));
974 pcb = lp->lwp_thread->td_pcb;
975 return (0);
976}
977
978#ifndef CPU_DISABLE_SSE
979static void
980fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
981{
982 struct env87 *penv_87 = &sv_87->sv_env;
983 struct envxmm *penv_xmm = &sv_xmm->sv_env;
984 int i;
985
986 /* FPU control/status */
987 penv_87->en_cw = penv_xmm->en_cw;
988 penv_87->en_sw = penv_xmm->en_sw;
989 penv_87->en_tw = penv_xmm->en_tw;
990 penv_87->en_fip = penv_xmm->en_fip;
991 penv_87->en_fcs = penv_xmm->en_fcs;
992 penv_87->en_opcode = penv_xmm->en_opcode;
993 penv_87->en_foo = penv_xmm->en_foo;
994 penv_87->en_fos = penv_xmm->en_fos;
995
996 /* FPU registers */
997 for (i = 0; i < 8; ++i)
998 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
999
1000 sv_87->sv_ex_sw = sv_xmm->sv_ex_sw;
1001}
1002
1003static void
1004set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
1005{
1006 struct env87 *penv_87 = &sv_87->sv_env;
1007 struct envxmm *penv_xmm = &sv_xmm->sv_env;
1008 int i;
1009
1010 /* FPU control/status */
1011 penv_xmm->en_cw = penv_87->en_cw;
1012 penv_xmm->en_sw = penv_87->en_sw;
1013 penv_xmm->en_tw = penv_87->en_tw;
1014 penv_xmm->en_fip = penv_87->en_fip;
1015 penv_xmm->en_fcs = penv_87->en_fcs;
1016 penv_xmm->en_opcode = penv_87->en_opcode;
1017 penv_xmm->en_foo = penv_87->en_foo;
1018 penv_xmm->en_fos = penv_87->en_fos;
1019
1020 /* FPU registers */
1021 for (i = 0; i < 8; ++i)
1022 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
1023
1024 sv_xmm->sv_ex_sw = sv_87->sv_ex_sw;
1025}
1026#endif /* CPU_DISABLE_SSE */
1027
1028int
1029fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
1030{
1031#ifndef CPU_DISABLE_SSE
1032 if (cpu_fxsr) {
1033 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
1034 (struct save87 *)fpregs);
1035 return (0);
1036 }
1037#endif /* CPU_DISABLE_SSE */
1038 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
1039 return (0);
1040}
1041
1042int
1043set_fpregs(struct lwp *lp, struct fpreg *fpregs)
1044{
1045#ifndef CPU_DISABLE_SSE
1046 if (cpu_fxsr) {
1047 set_fpregs_xmm((struct save87 *)fpregs,
1048 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
1049 return (0);
1050 }
1051#endif /* CPU_DISABLE_SSE */
1052 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
1053 return (0);
1054}
1055
1056int
1057fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
1058{
1059 return (ENOSYS);
1060}
1061
1062int
1063set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1064{
1065 return (ENOSYS);
1066}
1067
1068#if 0
1069/*
1070 * Return > 0 if a hardware breakpoint has been hit, and the
1071 * breakpoint was in user space. Return 0, otherwise.
1072 */
1073int
1074user_dbreg_trap(void)
1075{
1076 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1077 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1078 int nbp; /* number of breakpoints that triggered */
1079 caddr_t addr[4]; /* breakpoint addresses */
1080 int i;
1081
1082 dr7 = rdr7();
1083 if ((dr7 & 0x000000ff) == 0) {
1084 /*
1085 * all GE and LE bits in the dr7 register are zero,
1086 * thus the trap couldn't have been caused by the
1087 * hardware debug registers
1088 */
1089 return 0;
1090 }
1091
1092 nbp = 0;
1093 dr6 = rdr6();
1094 bp = dr6 & 0x0000000f;
1095
1096 if (!bp) {
1097 /*
1098 * None of the breakpoint bits are set meaning this
1099 * trap was not caused by any of the debug registers
1100 */
1101 return 0;
1102 }
1103
1104 /*
1105 * at least one of the breakpoints were hit, check to see
1106 * which ones and if any of them are user space addresses
1107 */
1108
1109 if (bp & 0x01) {
1110 addr[nbp++] = (caddr_t)rdr0();
1111 }
1112 if (bp & 0x02) {
1113 addr[nbp++] = (caddr_t)rdr1();
1114 }
1115 if (bp & 0x04) {
1116 addr[nbp++] = (caddr_t)rdr2();
1117 }
1118 if (bp & 0x08) {
1119 addr[nbp++] = (caddr_t)rdr3();
1120 }
1121
1122 for (i=0; i<nbp; i++) {
1123 if (addr[i] <
1124 (caddr_t)VM_MAX_USER_ADDRESS) {
1125 /*
1126 * addr[i] is in user space
1127 */
1128 return nbp;
1129 }
1130 }
1131
1132 /*
1133 * None of the breakpoints are in user space.
1134 */
1135 return 0;
1136}
1137
1138#endif
1139
1140void
1141identcpu(void)
1142{
1143 int regs[4];
1144
1145 do_cpuid(1, regs);
1146 cpu_feature = regs[3];
1147}
1148
1149
1150#ifndef DDB
1151void
1152Debugger(const char *msg)
1153{
1154 kprintf("Debugger(\"%s\") called.\n", msg);
1155}
1156#endif /* no DDB */