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/sysproto.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 #include <sys/mplock2.h>
78 
79 #include <sys/exec.h>
80 #include <sys/cons.h>
81 
82 #include <ddb/ddb.h>
83 
84 #include <machine/cpu.h>
85 #include <machine/clock.h>
86 #include <machine/specialreg.h>
87 #include <machine/md_var.h>
88 #include <machine/pcb.h>
89 #include <machine/pcb_ext.h>
90 #include <machine/globaldata.h>		/* CPU_prvspace */
91 #include <machine/smp.h>
92 #include <machine/cputypes.h>
93 
94 #include <bus/isa/rtc.h>
95 #include <sys/random.h>
96 #include <sys/ptrace.h>
97 #include <machine/sigframe.h>
98 #include <unistd.h>		/* umtx_* functions */
99 #include <pthread.h>		/* pthread_yield() */
100 
101 extern void dblfault_handler (void);
102 
103 static void set_fpregs_xmm (struct save87 *, struct savexmm *);
104 static void fill_fpregs_xmm (struct savexmm *, struct save87 *);
105 
106 int64_t tsc_offsets[MAXCPU];
107 
108 #if defined(SWTCH_OPTIM_STATS)
109 extern int swtch_optim_stats;
110 SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats,
111 	CTLFLAG_RD, &swtch_optim_stats, 0, "");
112 SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count,
113 	CTLFLAG_RD, &tlb_flush_count, 0, "");
114 #endif
115 
116 static int
117 sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
118 {
119 	u_long pmem = ctob(physmem);
120 	int error;
121 
122 	error = sysctl_handle_long(oidp, &pmem, 0, req);
123 
124 	return (error);
125 }
126 
127 SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_ULONG|CTLFLAG_RD,
128 	0, 0, sysctl_hw_physmem, "LU", "Total system memory in bytes (number of pages * page size)");
129 
130 static int
131 sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
132 {
133 	u_long usermem = ctob(Maxmem - vmstats.v_wire_count);
134 	int error;
135 
136 	error = sysctl_handle_long(oidp, &usermem, 0, req);
137 
138 	return (error);
139 }
140 
141 SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_ULONG|CTLFLAG_RD,
142 	0, 0, sysctl_hw_usermem, "LU", "");
143 
144 SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, 0, "");
145 
146 /*
147  * Send an interrupt to process.
148  *
149  * Stack is set up to allow sigcode stored
150  * at top to call routine, followed by kcall
151  * to sigreturn routine below.  After sigreturn
152  * resets the signal mask, the stack, and the
153  * frame pointer, it returns to the user
154  * specified pc, psl.
155  */
156 void
157 sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
158 {
159 	struct lwp *lp = curthread->td_lwp;
160 	struct proc *p = lp->lwp_proc;
161 	struct trapframe *regs;
162 	struct sigacts *psp = p->p_sigacts;
163 	struct sigframe sf, *sfp;
164 	int oonstack;
165 	char *sp;
166 
167 	regs = lp->lwp_md.md_regs;
168 	oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
169 
170 	/* Save user context */
171 	bzero(&sf, sizeof(struct sigframe));
172 	sf.sf_uc.uc_sigmask = *mask;
173 	sf.sf_uc.uc_stack = lp->lwp_sigstk;
174 	sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
175 	KKASSERT(__offsetof(struct trapframe, tf_rdi) == 0);
176 	/* gcc8 craps out on -Warray-bounds w/ optimized bcopy */
177 	_bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(struct trapframe));
178 
179 	/* Make the size of the saved context visible to userland */
180 	sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext);
181 
182 	/* Allocate and validate space for the signal handler context. */
183         if ((lp->lwp_flags & LWP_ALTSTACK) != 0 && !oonstack &&
184 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
185 		sp = (char *)lp->lwp_sigstk.ss_sp + lp->lwp_sigstk.ss_size -
186 		    sizeof(struct sigframe);
187 		lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
188 	} else {
189 		/* We take red zone into account */
190 		sp = (char *)regs->tf_rsp - sizeof(struct sigframe) - 128;
191 	}
192 
193 	/* Align to 16 bytes */
194 	sfp = (struct sigframe *)((intptr_t)sp & ~0xFUL);
195 
196 	/* Translate the signal is appropriate */
197 	if (p->p_sysent->sv_sigtbl) {
198 		if (sig <= p->p_sysent->sv_sigsize)
199 			sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
200 	}
201 
202 	/*
203 	 * Build the argument list for the signal handler.
204 	 *
205 	 * Arguments are in registers (%rdi, %rsi, %rdx, %rcx)
206 	 */
207 	regs->tf_rdi = sig;				/* argument 1 */
208 	regs->tf_rdx = (register_t)&sfp->sf_uc;		/* argument 3 */
209 
210 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
211 		/*
212 		 * Signal handler installed with SA_SIGINFO.
213 		 *
214 		 * action(signo, siginfo, ucontext)
215 		 */
216 		regs->tf_rsi = (register_t)&sfp->sf_si;	/* argument 2 */
217 		regs->tf_rcx = (register_t)regs->tf_err; /* argument 4 */
218 		sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
219 
220 		/* fill siginfo structure */
221 		sf.sf_si.si_signo = sig;
222 		sf.sf_si.si_pid = psp->ps_frominfo[sig].pid;
223 		sf.sf_si.si_uid = psp->ps_frominfo[sig].uid;
224 		sf.sf_si.si_code = code;
225 		sf.sf_si.si_addr = (void *)regs->tf_addr;
226 	} else {
227 		/*
228 		 * Old FreeBSD-style arguments.
229 		 *
230 		 * handler (signo, code, [uc], addr)
231 		 */
232 		regs->tf_rsi = (register_t)code;	/* argument 2 */
233 		regs->tf_rcx = (register_t)regs->tf_addr; /* argument 4 */
234 		sf.sf_ahu.sf_handler = catcher;
235 	}
236 
237 #if 0
238 	/*
239 	 * If we're a vm86 process, we want to save the segment registers.
240 	 * We also change eflags to be our emulated eflags, not the actual
241 	 * eflags.
242 	 */
243 	if (regs->tf_eflags & PSL_VM) {
244 		struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
245 		struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
246 
247 		sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
248 		sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
249 		sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
250 		sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
251 
252 		if (vm86->vm86_has_vme == 0)
253 			sf.sf_uc.uc_mcontext.mc_eflags =
254 			    (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
255 			    (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
256 
257 		/*
258 		 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
259 		 * syscalls made by the signal handler.  This just avoids
260 		 * wasting time for our lazy fixup of such faults.  PSL_NT
261 		 * does nothing in vm86 mode, but vm86 programs can set it
262 		 * almost legitimately in probes for old cpu types.
263 		 */
264 		tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
265 	}
266 #endif
267 
268 	/*
269 	 * Save the FPU state and reinit the FP unit
270 	 */
271 	npxpush(&sf.sf_uc.uc_mcontext);
272 
273 	/*
274 	 * Copy the sigframe out to the user's stack.
275 	 */
276 	if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
277 		/*
278 		 * Something is wrong with the stack pointer.
279 		 * ...Kill the process.
280 		 */
281 		sigexit(lp, SIGILL);
282 	}
283 
284 	regs->tf_rsp = (register_t)sfp;
285 	regs->tf_rip = trunc_page64(PS_STRINGS - *(p->p_sysent->sv_szsigcode));
286 	regs->tf_rip -= SZSIGCODE_EXTRA_BYTES;
287 
288 	/*
289 	 * x86 abi specifies that the direction flag must be cleared
290 	 * on function entry
291 	 */
292 	regs->tf_rflags &= ~(PSL_T|PSL_D);
293 
294 	/*
295 	 * 64 bit mode has a code and stack selector but
296 	 * no data or extra selector.  %fs and %gs are not
297 	 * stored in-context.
298 	 */
299 	regs->tf_cs = _ucodesel;
300 	regs->tf_ss = _udatasel;
301 }
302 
303 /*
304  * Sanitize the trapframe for a virtual kernel passing control to a custom
305  * VM context.  Remove any items that would otherwise create a privilage
306  * issue.
307  *
308  * XXX at the moment we allow userland to set the resume flag.  Is this a
309  * bad idea?
310  */
311 int
312 cpu_sanitize_frame(struct trapframe *frame)
313 {
314 	frame->tf_cs = _ucodesel;
315 	frame->tf_ss = _udatasel;
316 	/* XXX VM (8086) mode not supported? */
317 	frame->tf_rflags &= (PSL_RF | PSL_USERCHANGE | PSL_VM_UNSUPP);
318 	frame->tf_rflags |= PSL_RESERVED_DEFAULT | PSL_I;
319 
320 	return(0);
321 }
322 
323 /*
324  * Sanitize the tls so loading the descriptor does not blow up
325  * on us.  For x86_64 we don't have to do anything.
326  */
327 int
328 cpu_sanitize_tls(struct savetls *tls)
329 {
330 	return(0);
331 }
332 
333 /*
334  * sigreturn(ucontext_t *sigcntxp)
335  *
336  * System call to cleanup state after a signal
337  * has been taken.  Reset signal mask and
338  * stack state from context left by sendsig (above).
339  * Return to previous pc and psl as specified by
340  * context left by sendsig. Check carefully to
341  * make sure that the user has not modified the
342  * state to gain improper privileges.
343  */
344 #define	EFL_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
345 #define	CS_SECURE(cs)		(ISPL(cs) == SEL_UPL)
346 
347 int
348 sys_sigreturn(struct sigreturn_args *uap)
349 {
350 	struct lwp *lp = curthread->td_lwp;
351 	struct trapframe *regs;
352 	ucontext_t uc;
353 	ucontext_t *ucp;
354 	register_t rflags;
355 	int cs;
356 	int error;
357 
358 	/*
359 	 * We have to copy the information into kernel space so userland
360 	 * can't modify it while we are sniffing it.
361 	 */
362 	regs = lp->lwp_md.md_regs;
363 	error = copyin(uap->sigcntxp, &uc, sizeof(uc));
364 	if (error)
365 		return (error);
366 	ucp = &uc;
367 	rflags = ucp->uc_mcontext.mc_rflags;
368 
369 	/* VM (8086) mode not supported */
370 	rflags &= ~PSL_VM_UNSUPP;
371 
372 #if 0
373 	if (eflags & PSL_VM) {
374 		struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
375 		struct vm86_kernel *vm86;
376 
377 		/*
378 		 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
379 		 * set up the vm86 area, and we can't enter vm86 mode.
380 		 */
381 		if (lp->lwp_thread->td_pcb->pcb_ext == 0)
382 			return (EINVAL);
383 		vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
384 		if (vm86->vm86_inited == 0)
385 			return (EINVAL);
386 
387 		/* go back to user mode if both flags are set */
388 		if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
389 			trapsignal(lp->lwp_proc, SIGBUS, 0);
390 
391 		if (vm86->vm86_has_vme) {
392 			eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
393 			    (eflags & VME_USERCHANGE) | PSL_VM;
394 		} else {
395 			vm86->vm86_eflags = eflags;	/* save VIF, VIP */
396 			eflags = (tf->tf_eflags & ~VM_USERCHANGE) |					    (eflags & VM_USERCHANGE) | PSL_VM;
397 		}
398 		bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
399 		tf->tf_eflags = eflags;
400 		tf->tf_vm86_ds = tf->tf_ds;
401 		tf->tf_vm86_es = tf->tf_es;
402 		tf->tf_vm86_fs = tf->tf_fs;
403 		tf->tf_vm86_gs = tf->tf_gs;
404 		tf->tf_ds = _udatasel;
405 		tf->tf_es = _udatasel;
406 #if 0
407 		tf->tf_fs = _udatasel;
408 		tf->tf_gs = _udatasel;
409 #endif
410 	} else
411 #endif
412 	{
413 		/*
414 		 * Don't allow users to change privileged or reserved flags.
415 		 */
416 		/*
417 		 * XXX do allow users to change the privileged flag PSL_RF.
418 		 * The cpu sets PSL_RF in tf_eflags for faults.  Debuggers
419 		 * should sometimes set it there too.  tf_eflags is kept in
420 		 * the signal context during signal handling and there is no
421 		 * other place to remember it, so the PSL_RF bit may be
422 		 * corrupted by the signal handler without us knowing.
423 		 * Corruption of the PSL_RF bit at worst causes one more or
424 		 * one less debugger trap, so allowing it is fairly harmless.
425 		 */
426 		if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
427 			kprintf("sigreturn: rflags = 0x%lx\n", (long)rflags);
428 			return(EINVAL);
429 		}
430 
431 		/*
432 		 * Don't allow users to load a valid privileged %cs.  Let the
433 		 * hardware check for invalid selectors, excess privilege in
434 		 * other selectors, invalid %eip's and invalid %esp's.
435 		 */
436 		cs = ucp->uc_mcontext.mc_cs;
437 		if (!CS_SECURE(cs)) {
438 			kprintf("sigreturn: cs = 0x%x\n", cs);
439 			trapsignal(lp, SIGBUS, T_PROTFLT);
440 			return(EINVAL);
441 		}
442 		bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(struct trapframe));
443 	}
444 
445 	/*
446 	 * Restore the FPU state from the frame
447 	 */
448 	npxpop(&ucp->uc_mcontext);
449 
450 	if (ucp->uc_mcontext.mc_onstack & 1)
451 		lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
452 	else
453 		lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
454 
455 	lp->lwp_sigmask = ucp->uc_sigmask;
456 	SIG_CANTMASK(lp->lwp_sigmask);
457 	return(EJUSTRETURN);
458 }
459 
460 /*
461  * cpu_idle() represents the idle LWKT.  You cannot return from this function
462  * (unless you want to blow things up!).  Instead we look for runnable threads
463  * and loop or halt as appropriate.  Giant is not held on entry to the thread.
464  *
465  * The main loop is entered with a critical section held, we must release
466  * the critical section before doing anything else.  lwkt_switch() will
467  * check for pending interrupts due to entering and exiting its own
468  * critical section.
469  *
470  * Note on cpu_idle_hlt:  On an SMP system we rely on a scheduler IPI
471  * to wake a HLTed cpu up.
472  */
473 __read_mostly static int	cpu_idle_hlt = 1;
474 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
475     &cpu_idle_hlt, 0, "Idle loop HLT enable");
476 
477 void
478 cpu_idle(void)
479 {
480 	struct thread *td = curthread;
481 	struct mdglobaldata *gd = mdcpu;
482 	int reqflags;
483 
484 	crit_exit();
485 	KKASSERT(td->td_critcount == 0);
486 	cpu_enable_intr();
487 
488 	for (;;) {
489 		/*
490 		 * See if there are any LWKTs ready to go.
491 		 */
492 		lwkt_switch();
493 
494 		/*
495 		 * The idle loop halts only if no threads are scheduleable
496 		 * and no signals have occured.
497 		 */
498 		if (cpu_idle_hlt &&
499 		    (td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
500 			splz();
501 			if ((td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) {
502 #ifdef DEBUGIDLE
503 				struct timeval tv1, tv2;
504 				gettimeofday(&tv1, NULL);
505 #endif
506 				reqflags = gd->mi.gd_reqflags &
507 					   ~RQF_IDLECHECK_WK_MASK;
508 				KKASSERT(gd->mi.gd_processing_ipiq == 0);
509 				umtx_sleep(&gd->mi.gd_reqflags, reqflags,
510 					   1000000);
511 #ifdef DEBUGIDLE
512 				gettimeofday(&tv2, NULL);
513 				if (tv2.tv_usec - tv1.tv_usec +
514 				    (tv2.tv_sec - tv1.tv_sec) * 1000000
515 				    > 500000) {
516 					kprintf("cpu %d idlelock %08x %08x\n",
517 						gd->mi.gd_cpuid,
518 						gd->mi.gd_reqflags,
519 						gd->gd_fpending);
520 				}
521 #endif
522 			}
523 		} else {
524 			splz();
525 			__asm __volatile("pause");
526 		}
527 	}
528 }
529 
530 /*
531  * Called by the spinlock code with or without a critical section held
532  * when a spinlock is found to be seriously constested.
533  *
534  * We need to enter a critical section to prevent signals from recursing
535  * into pthreads.
536  */
537 void
538 cpu_spinlock_contested(void)
539 {
540 	cpu_pause();
541 }
542 
543 /*
544  * Clear registers on exec
545  */
546 void
547 exec_setregs(u_long entry, u_long stack, u_long ps_strings)
548 {
549 	struct thread *td = curthread;
550 	struct lwp *lp = td->td_lwp;
551 	struct pcb *pcb = td->td_pcb;
552 	struct trapframe *regs = lp->lwp_md.md_regs;
553 
554 	user_ldt_free(pcb);
555 
556 	bzero((char *)regs, sizeof(struct trapframe));
557 	regs->tf_rip = entry;
558 	regs->tf_rsp = ((stack - 8) & ~0xFul) + 8; /* align the stack */
559 	regs->tf_rdi = stack;		/* argv */
560 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
561 	regs->tf_ss = _udatasel;
562 	regs->tf_cs = _ucodesel;
563 	regs->tf_rbx = ps_strings;
564 
565 	/*
566 	 * Reset the hardware debug registers if they were in use.
567 	 * They won't have any meaning for the newly exec'd process.
568 	 */
569 	if (pcb->pcb_flags & PCB_DBREGS) {
570 		pcb->pcb_dr0 = 0;
571 		pcb->pcb_dr1 = 0;
572 		pcb->pcb_dr2 = 0;
573 		pcb->pcb_dr3 = 0;
574 		pcb->pcb_dr6 = 0;
575 		pcb->pcb_dr7 = 0; /* JG set bit 10? */
576 		if (pcb == td->td_pcb) {
577 			/*
578 			 * Clear the debug registers on the running
579 			 * CPU, otherwise they will end up affecting
580 			 * the next process we switch to.
581 			 */
582 			reset_dbregs();
583 		}
584 		pcb->pcb_flags &= ~PCB_DBREGS;
585 	}
586 
587 	/*
588 	 * Initialize the math emulator (if any) for the current process.
589 	 * Actually, just clear the bit that says that the emulator has
590 	 * been initialized.  Initialization is delayed until the process
591 	 * traps to the emulator (if it is done at all) mainly because
592 	 * emulators don't provide an entry point for initialization.
593 	 */
594 	pcb->pcb_flags &= ~FP_SOFTFP;
595 
596 	/*
597 	 * NOTE: do not set CR0_TS here.  npxinit() must do it after clearing
598 	 *	 gd_npxthread.  Otherwise a preemptive interrupt thread
599 	 *	 may panic in npxdna().
600 	 */
601 	crit_enter();
602 #if 0
603 	load_cr0(rcr0() | CR0_MP);
604 #endif
605 
606 	/*
607 	 * NOTE: The MSR values must be correct so we can return to
608 	 *	 userland.  gd_user_fs/gs must be correct so the switch
609 	 *	 code knows what the current MSR values are.
610 	 */
611 	pcb->pcb_fsbase = 0;	/* Values loaded from PCB on switch */
612 	pcb->pcb_gsbase = 0;
613 	/* Initialize the npx (if any) for the current process. */
614 	npxinit();
615 	crit_exit();
616 
617 	/*
618 	 * note: linux emulator needs edx to be 0x0 on entry, which is
619 	 * handled in execve simply by setting the 64 bit syscall
620 	 * return value to 0.
621 	 */
622 }
623 
624 void
625 cpu_setregs(void)
626 {
627 #if 0
628 	unsigned int cr0;
629 
630 	cr0 = rcr0();
631 	cr0 |= CR0_NE;			/* Done by npxinit() */
632 	cr0 |= CR0_MP | CR0_TS;		/* Done at every execve() too. */
633 	cr0 |= CR0_WP | CR0_AM;
634 	load_cr0(cr0);
635 	load_gs(_udatasel);
636 #endif
637 }
638 
639 static int
640 sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
641 {
642 	int error;
643 	error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
644 		req);
645 	if (!error && req->newptr)
646 		resettodr();
647 	return (error);
648 }
649 
650 SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
651 	&adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
652 
653 /*
654  * Initialize x86 and configure to run kernel
655  */
656 
657 /*
658  * Initialize segments & interrupt table
659  */
660 
661 extern  struct user *proc0paddr;
662 
663 #if 0
664 
665 extern inthand_t
666 	IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
667 	IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
668 	IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
669 	IDTVEC(page), IDTVEC(mchk), IDTVEC(rsvd), IDTVEC(fpu), IDTVEC(align),
670 	IDTVEC(xmm), IDTVEC(dblfault),
671 	IDTVEC(fast_syscall), IDTVEC(fast_syscall32);
672 #endif
673 
674 int
675 ptrace_set_pc(struct lwp *lp, unsigned long addr)
676 {
677 	lp->lwp_md.md_regs->tf_rip = addr;
678 	return (0);
679 }
680 
681 int
682 ptrace_single_step(struct lwp *lp)
683 {
684 	lp->lwp_md.md_regs->tf_rflags |= PSL_T;
685 	return (0);
686 }
687 
688 int
689 fill_regs(struct lwp *lp, struct reg *regs)
690 {
691 	struct trapframe *tp;
692 
693 	if ((tp = lp->lwp_md.md_regs) == NULL)
694 		return EINVAL;
695 	bcopy(&tp->tf_rdi, &regs->r_rdi, sizeof(*regs));
696 	return (0);
697 }
698 
699 int
700 set_regs(struct lwp *lp, struct reg *regs)
701 {
702 	struct trapframe *tp;
703 
704 	tp = lp->lwp_md.md_regs;
705 	if (!EFL_SECURE(regs->r_rflags, tp->tf_rflags) ||
706 	    !CS_SECURE(regs->r_cs))
707 		return (EINVAL);
708 	bcopy(&regs->r_rdi, &tp->tf_rdi, sizeof(*regs));
709 	return (0);
710 }
711 
712 static void
713 fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
714 {
715 	struct env87 *penv_87 = &sv_87->sv_env;
716 	struct envxmm *penv_xmm = &sv_xmm->sv_env;
717 	int i;
718 
719 	/* FPU control/status */
720 	penv_87->en_cw = penv_xmm->en_cw;
721 	penv_87->en_sw = penv_xmm->en_sw;
722 	penv_87->en_tw = penv_xmm->en_tw;
723 	penv_87->en_fip = penv_xmm->en_fip;
724 	penv_87->en_fcs = penv_xmm->en_fcs;
725 	penv_87->en_opcode = penv_xmm->en_opcode;
726 	penv_87->en_foo = penv_xmm->en_foo;
727 	penv_87->en_fos = penv_xmm->en_fos;
728 
729 	/* FPU registers */
730 	for (i = 0; i < 8; ++i)
731 		sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
732 }
733 
734 static void
735 set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
736 {
737 	struct env87 *penv_87 = &sv_87->sv_env;
738 	struct envxmm *penv_xmm = &sv_xmm->sv_env;
739 	int i;
740 
741 	/* FPU control/status */
742 	penv_xmm->en_cw = penv_87->en_cw;
743 	penv_xmm->en_sw = penv_87->en_sw;
744 	penv_xmm->en_tw = penv_87->en_tw;
745 	penv_xmm->en_fip = penv_87->en_fip;
746 	penv_xmm->en_fcs = penv_87->en_fcs;
747 	penv_xmm->en_opcode = penv_87->en_opcode;
748 	penv_xmm->en_foo = penv_87->en_foo;
749 	penv_xmm->en_fos = penv_87->en_fos;
750 
751 	/* FPU registers */
752 	for (i = 0; i < 8; ++i)
753 		sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
754 }
755 
756 int
757 fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
758 {
759 	if (lp->lwp_thread == NULL || lp->lwp_thread->td_pcb == NULL)
760 		return EINVAL;
761 	if (cpu_fxsr) {
762 		fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
763 				(struct save87 *)fpregs);
764 		return (0);
765 	}
766 	bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
767 	return (0);
768 }
769 
770 int
771 set_fpregs(struct lwp *lp, struct fpreg *fpregs)
772 {
773 	if (cpu_fxsr) {
774 		set_fpregs_xmm((struct save87 *)fpregs,
775 			       &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
776 		return (0);
777 	}
778 	bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
779 	return (0);
780 }
781 
782 int
783 fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
784 {
785 	return (ENOSYS);
786 }
787 
788 int
789 set_dbregs(struct lwp *lp, struct dbreg *dbregs)
790 {
791 	return (ENOSYS);
792 }
793 
794 #if 0
795 /*
796  * Return > 0 if a hardware breakpoint has been hit, and the
797  * breakpoint was in user space.  Return 0, otherwise.
798  */
799 int
800 user_dbreg_trap(void)
801 {
802         u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
803         u_int32_t bp;       /* breakpoint bits extracted from dr6 */
804         int nbp;            /* number of breakpoints that triggered */
805         caddr_t addr[4];    /* breakpoint addresses */
806         int i;
807 
808         dr7 = rdr7();
809         if ((dr7 & 0x000000ff) == 0) {
810                 /*
811                  * all GE and LE bits in the dr7 register are zero,
812                  * thus the trap couldn't have been caused by the
813                  * hardware debug registers
814                  */
815                 return 0;
816         }
817 
818         nbp = 0;
819         dr6 = rdr6();
820         bp = dr6 & 0x0000000f;
821 
822         if (!bp) {
823                 /*
824                  * None of the breakpoint bits are set meaning this
825                  * trap was not caused by any of the debug registers
826                  */
827                 return 0;
828         }
829 
830         /*
831          * at least one of the breakpoints were hit, check to see
832          * which ones and if any of them are user space addresses
833          */
834 
835         if (bp & 0x01) {
836                 addr[nbp++] = (caddr_t)rdr0();
837         }
838         if (bp & 0x02) {
839                 addr[nbp++] = (caddr_t)rdr1();
840         }
841         if (bp & 0x04) {
842                 addr[nbp++] = (caddr_t)rdr2();
843         }
844         if (bp & 0x08) {
845                 addr[nbp++] = (caddr_t)rdr3();
846         }
847 
848         for (i=0; i<nbp; i++) {
849                 if (addr[i] <
850                     (caddr_t)VM_MAX_USER_ADDRESS) {
851                         /*
852                          * addr[i] is in user space
853                          */
854                         return nbp;
855                 }
856         }
857 
858         /*
859          * None of the breakpoints are in user space.
860          */
861         return 0;
862 }
863 
864 #endif
865 
866 void
867 identcpu(void)
868 {
869 	int regs[4];
870 
871 	do_cpuid(1, regs);
872 	cpu_feature = regs[3];
873 }
874 
875 
876 #ifndef DDB
877 void
878 Debugger(const char *msg)
879 {
880 	kprintf("Debugger(\"%s\") called.\n", msg);
881 }
882 #endif /* no DDB */
883