xref: /dragonfly/sys/cpu/x86_64/include/cpufunc.h (revision 335b9e93)
1 /*-
2  * Copyright (c) 2003 Peter Wemm.
3  * Copyright (c) 1993 The Regents of the University of California.
4  * Copyright (c) 2008 The DragonFly Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  * $FreeBSD: src/sys/amd64/include/cpufunc.h,v 1.139 2004/01/28 23:53:04 peter Exp $
32  */
33 
34 /*
35  * Functions to provide access to special i386 instructions.
36  * This in included in sys/systm.h, and that file should be
37  * used in preference to this.
38  */
39 
40 #ifndef _CPU_CPUFUNC_H_
41 #define	_CPU_CPUFUNC_H_
42 
43 #include <sys/cdefs.h>
44 #include <sys/thread.h>
45 #include <machine/clock.h>
46 #include <machine/psl.h>
47 #include <machine/smp.h>
48 
49 struct thread;
50 struct region_descriptor;
51 struct pmap;
52 
53 __BEGIN_DECLS
54 #define readb(va)	(*(volatile u_int8_t *) (va))
55 #define readw(va)	(*(volatile u_int16_t *) (va))
56 #define readl(va)	(*(volatile u_int32_t *) (va))
57 #define readq(va)	(*(volatile u_int64_t *) (va))
58 
59 #define writeb(va, d)	(*(volatile u_int8_t *) (va) = (d))
60 #define writew(va, d)	(*(volatile u_int16_t *) (va) = (d))
61 #define writel(va, d)	(*(volatile u_int32_t *) (va) = (d))
62 #define writeq(va, d)	(*(volatile u_int64_t *) (va) = (d))
63 
64 #ifdef	__GNUC__
65 
66 #include <machine/lock.h>		/* XXX */
67 
68 struct trapframe;
69 
70 static __inline void
71 breakpoint(void)
72 {
73 	__asm __volatile("int $3");
74 }
75 
76 static __inline void
77 cpu_pause(void)
78 {
79 	__asm __volatile("pause":::"memory");
80 }
81 
82 static __inline u_int
83 bsfl(u_int mask)
84 {
85 	u_int	result;
86 
87 	__asm __volatile("bsfl %1,%0" : "=r" (result) : "rm" (mask));
88 	return (result);
89 }
90 
91 static __inline u_long
92 bsfq(u_long mask)
93 {
94 	u_long	result;
95 
96 	__asm __volatile("bsfq %1,%0" : "=r" (result) : "rm" (mask));
97 	return (result);
98 }
99 
100 static __inline u_long
101 bsflong(u_long mask)
102 {
103 	u_long	result;
104 
105 	__asm __volatile("bsfq %1,%0" : "=r" (result) : "rm" (mask));
106 	return (result);
107 }
108 
109 static __inline u_int
110 bsrl(u_int mask)
111 {
112 	u_int	result;
113 
114 	__asm __volatile("bsrl %1,%0" : "=r" (result) : "rm" (mask));
115 	return (result);
116 }
117 
118 static __inline u_long
119 bsrq(u_long mask)
120 {
121 	u_long	result;
122 
123 	__asm __volatile("bsrq %1,%0" : "=r" (result) : "rm" (mask));
124 	return (result);
125 }
126 
127 static __inline void
128 clflush(u_long addr)
129 {
130 	__asm __volatile("clflush %0" : : "m" (*(char *) addr));
131 }
132 
133 static __inline void
134 do_cpuid(u_int ax, u_int *p)
135 {
136 	__asm __volatile("cpuid"
137 			 : "=a" (p[0]), "=b" (p[1]), "=c" (p[2]), "=d" (p[3])
138 			 :  "0" (ax));
139 }
140 
141 static __inline void
142 cpuid_count(u_int ax, u_int cx, u_int *p)
143 {
144 	__asm __volatile("cpuid"
145 			 : "=a" (p[0]), "=b" (p[1]), "=c" (p[2]), "=d" (p[3])
146 			 :  "0" (ax), "c" (cx));
147 }
148 
149 #ifndef _CPU_DISABLE_INTR_DEFINED
150 
151 static __inline void
152 cpu_disable_intr(void)
153 {
154 	__asm __volatile("cli" : : : "memory");
155 }
156 
157 #endif
158 
159 #ifndef _CPU_ENABLE_INTR_DEFINED
160 
161 static __inline void
162 cpu_enable_intr(void)
163 {
164 	__asm __volatile("sti");
165 }
166 
167 #endif
168 
169 /*
170  * Cpu and compiler memory ordering fence.  mfence ensures strong read and
171  * write ordering.
172  *
173  * A serializing or fence instruction is required here.  A locked bus
174  * cycle on data for which we already own cache mastership is the most
175  * portable.
176  */
177 static __inline void
178 cpu_mfence(void)
179 {
180 	__asm __volatile("mfence" : : : "memory");
181 }
182 
183 /*
184  * cpu_lfence() ensures strong read ordering for reads issued prior
185  * to the instruction verses reads issued afterwords.
186  *
187  * A serializing or fence instruction is required here.  A locked bus
188  * cycle on data for which we already own cache mastership is the most
189  * portable.
190  */
191 static __inline void
192 cpu_lfence(void)
193 {
194 	__asm __volatile("lfence" : : : "memory");
195 }
196 
197 /*
198  * cpu_sfence() ensures strong write ordering for writes issued prior
199  * to the instruction verses writes issued afterwords.  Writes are
200  * ordered on intel cpus so we do not actually have to do anything.
201  */
202 static __inline void
203 cpu_sfence(void)
204 {
205 	/*
206 	 * NOTE:
207 	 * Don't use 'sfence' here, as it will create a lot of
208 	 * unnecessary stalls.
209 	 */
210 	__asm __volatile("" : : : "memory");
211 }
212 
213 /*
214  * cpu_ccfence() prevents the compiler from reordering instructions, in
215  * particular stores, relative to the current cpu.  Use cpu_sfence() if
216  * you need to guarentee ordering by both the compiler and by the cpu.
217  *
218  * This also prevents the compiler from caching memory loads into local
219  * variables across the routine.
220  */
221 static __inline void
222 cpu_ccfence(void)
223 {
224 	__asm __volatile("" : : : "memory");
225 }
226 
227 /*
228  * This is a horrible, horrible hack that might have to be put at the
229  * end of certain procedures (on a case by case basis), just before it
230  * returns to avoid what we believe to be an unreported AMD cpu bug.
231  * Found to occur on both a Phenom II X4 820 (two of them), as well
232  * as a 48-core built around an Opteron 6168 (Id = 0x100f91  Stepping = 1).
233  * The problem does not appear to occur w/Intel cpus.
234  *
235  * The bug is likely related to either a write combining issue or the
236  * Return Address Stack (RAS) hardware cache.
237  *
238  * In particular, we had to do this for GCC's fill_sons_in_loop() routine
239  * which due to its deep recursion and stack flow appears to be able to
240  * tickle the amd cpu bug (w/ gcc-4.4.7).  Adding a single 'nop' to the
241  * end of the routine just before it returns works around the bug.
242  *
243  * The bug appears to be extremely sensitive to %rip and %rsp values, to
244  * the point where even just inserting an instruction in an unrelated
245  * procedure (shifting the entire code base being run) effects the outcome.
246  * DragonFly is probably able to more readily reproduce the bug due to
247  * the stackgap randomization code.  We would expect OpenBSD (where we got
248  * the stackgap randomization code from) to also be able to reproduce the
249  * issue.  To date we have only reproduced the issue in DragonFly.
250  */
251 #define __AMDCPUBUG_DFLY01_AVAILABLE__
252 
253 static __inline void
254 cpu_amdcpubug_dfly01(void)
255 {
256 	__asm __volatile("nop" : : : "memory");
257 }
258 
259 #ifdef _KERNEL
260 
261 #define	HAVE_INLINE_FFS
262 
263 static __inline int
264 ffs(int mask)
265 {
266 #if 0
267 	/*
268 	 * Note that gcc-2's builtin ffs would be used if we didn't declare
269 	 * this inline or turn off the builtin.  The builtin is faster but
270 	 * broken in gcc-2.4.5 and slower but working in gcc-2.5 and later
271 	 * versions.
272 	 */
273 	return (mask == 0 ? mask : (int)bsfl((u_int)mask) + 1);
274 #else
275 	/* Actually, the above is way out of date.  The builtins use cmov etc */
276 	return (__builtin_ffs(mask));
277 #endif
278 }
279 
280 #define	HAVE_INLINE_FFSL
281 
282 static __inline int
283 ffsl(long mask)
284 {
285 	return (mask == 0 ? mask : (int)bsfq((u_long)mask) + 1);
286 }
287 
288 #define	HAVE_INLINE_FLS
289 
290 static __inline int
291 fls(int mask)
292 {
293 	return (mask == 0 ? mask : (int)bsrl((u_int)mask) + 1);
294 }
295 
296 #define	HAVE_INLINE_FLSL
297 
298 static __inline int
299 flsl(long mask)
300 {
301 	return (mask == 0 ? mask : (int)bsrq((u_long)mask) + 1);
302 }
303 
304 #define	HAVE_INLINE_FLSLL
305 
306 static __inline int
307 flsll(long long mask)
308 {
309 	return (flsl((long)mask));
310 }
311 
312 #endif /* _KERNEL */
313 
314 static __inline void
315 halt(void)
316 {
317 	__asm __volatile("hlt");
318 }
319 
320 /*
321  * The following complications are to get around gcc not having a
322  * constraint letter for the range 0..255.  We still put "d" in the
323  * constraint because "i" isn't a valid constraint when the port
324  * isn't constant.  This only matters for -O0 because otherwise
325  * the non-working version gets optimized away.
326  *
327  * Use an expression-statement instead of a conditional expression
328  * because gcc-2.6.0 would promote the operands of the conditional
329  * and produce poor code for "if ((inb(var) & const1) == const2)".
330  *
331  * The unnecessary test `(port) < 0x10000' is to generate a warning if
332  * the `port' has type u_short or smaller.  Such types are pessimal.
333  * This actually only works for signed types.  The range check is
334  * careful to avoid generating warnings.
335  */
336 #define	inb(port) __extension__ ({					\
337 	u_char	_data;							\
338 	if (__builtin_constant_p(port) && ((port) & 0xffff) < 0x100	\
339 	    && (port) < 0x10000)					\
340 		_data = inbc(port);					\
341 	else								\
342 		_data = inbv(port);					\
343 	_data; })
344 
345 #define	outb(port, data) (						\
346 	__builtin_constant_p(port) && ((port) & 0xffff) < 0x100		\
347 	&& (port) < 0x10000						\
348 	? outbc(port, data) : outbv(port, data))
349 
350 static __inline u_char
351 inbc(u_int port)
352 {
353 	u_char	data;
354 
355 	__asm __volatile("inb %1,%0" : "=a" (data) : "id" ((u_short)(port)));
356 	return (data);
357 }
358 
359 static __inline void
360 outbc(u_int port, u_char data)
361 {
362 	__asm __volatile("outb %0,%1" : : "a" (data), "id" ((u_short)(port)));
363 }
364 
365 static __inline u_char
366 inbv(u_int port)
367 {
368 	u_char	data;
369 	/*
370 	 * We use %%dx and not %1 here because i/o is done at %dx and not at
371 	 * %edx, while gcc generates inferior code (movw instead of movl)
372 	 * if we tell it to load (u_short) port.
373 	 */
374 	__asm __volatile("inb %%dx,%0" : "=a" (data) : "d" (port));
375 	return (data);
376 }
377 
378 static __inline u_int
379 inl(u_int port)
380 {
381 	u_int	data;
382 
383 	__asm __volatile("inl %%dx,%0" : "=a" (data) : "d" (port));
384 	return (data);
385 }
386 
387 static __inline void
388 insb(u_int port, void *addr, size_t cnt)
389 {
390 	__asm __volatile("cld; rep; insb"
391 			 : "+D" (addr), "+c" (cnt)
392 			 : "d" (port)
393 			 : "memory");
394 }
395 
396 static __inline void
397 insw(u_int port, void *addr, size_t cnt)
398 {
399 	__asm __volatile("cld; rep; insw"
400 			 : "+D" (addr), "+c" (cnt)
401 			 : "d" (port)
402 			 : "memory");
403 }
404 
405 static __inline void
406 insl(u_int port, void *addr, size_t cnt)
407 {
408 	__asm __volatile("cld; rep; insl"
409 			 : "+D" (addr), "+c" (cnt)
410 			 : "d" (port)
411 			 : "memory");
412 }
413 
414 static __inline void
415 invd(void)
416 {
417 	__asm __volatile("invd");
418 }
419 
420 #if defined(_KERNEL)
421 
422 #ifndef _CPU_INVLPG_DEFINED
423 
424 /*
425  * Invalidate a particular VA on this cpu only
426  *
427  * TLB flush for an individual page (even if it has PG_G).
428  * Only works on 486+ CPUs (i386 does not have PG_G).
429  */
430 static __inline void
431 cpu_invlpg(void *addr)
432 {
433 	__asm __volatile("invlpg %0" : : "m" (*(char *)addr) : "memory");
434 }
435 
436 #endif
437 
438 static __inline void
439 cpu_nop(void)
440 {
441 	__asm __volatile("rep; nop");
442 }
443 
444 #endif	/* _KERNEL */
445 
446 static __inline u_short
447 inw(u_int port)
448 {
449 	u_short	data;
450 
451 	__asm __volatile("inw %%dx,%0" : "=a" (data) : "d" (port));
452 	return (data);
453 }
454 
455 static __inline u_int
456 loadandclear(volatile u_int *addr)
457 {
458 	u_int   result;
459 
460 	__asm __volatile("xorl %0,%0; xchgl %1,%0"
461 			: "=&r" (result) : "m" (*addr));
462 	return (result);
463 }
464 
465 static __inline void
466 outbv(u_int port, u_char data)
467 {
468 	u_char	al;
469 	/*
470 	 * Use an unnecessary assignment to help gcc's register allocator.
471 	 * This make a large difference for gcc-1.40 and a tiny difference
472 	 * for gcc-2.6.0.  For gcc-1.40, al had to be ``asm("ax")'' for
473 	 * best results.  gcc-2.6.0 can't handle this.
474 	 */
475 	al = data;
476 	__asm __volatile("outb %0,%%dx" : : "a" (al), "d" (port));
477 }
478 
479 static __inline void
480 outl(u_int port, u_int data)
481 {
482 	/*
483 	 * outl() and outw() aren't used much so we haven't looked at
484 	 * possible micro-optimizations such as the unnecessary
485 	 * assignment for them.
486 	 */
487 	__asm __volatile("outl %0,%%dx" : : "a" (data), "d" (port));
488 }
489 
490 static __inline void
491 outsb(u_int port, const void *addr, size_t cnt)
492 {
493 	__asm __volatile("cld; rep; outsb"
494 			 : "+S" (addr), "+c" (cnt)
495 			 : "d" (port));
496 }
497 
498 static __inline void
499 outsw(u_int port, const void *addr, size_t cnt)
500 {
501 	__asm __volatile("cld; rep; outsw"
502 			 : "+S" (addr), "+c" (cnt)
503 			 : "d" (port));
504 }
505 
506 static __inline void
507 outsl(u_int port, const void *addr, size_t cnt)
508 {
509 	__asm __volatile("cld; rep; outsl"
510 			 : "+S" (addr), "+c" (cnt)
511 			 : "d" (port));
512 }
513 
514 static __inline void
515 outw(u_int port, u_short data)
516 {
517 	__asm __volatile("outw %0,%%dx" : : "a" (data), "d" (port));
518 }
519 
520 static __inline void
521 ia32_pause(void)
522 {
523 	__asm __volatile("pause");
524 }
525 
526 static __inline u_long
527 read_rflags(void)
528 {
529 	u_long	rf;
530 
531 	__asm __volatile("pushfq; popq %0" : "=r" (rf));
532 	return (rf);
533 }
534 
535 static __inline u_int64_t
536 rdmsr(u_int msr)
537 {
538 	u_int32_t low, high;
539 
540 	__asm __volatile("rdmsr" : "=a" (low), "=d" (high) : "c" (msr));
541 	return (low | ((u_int64_t)high << 32));
542 }
543 
544 static __inline u_int64_t
545 rdpmc(u_int pmc)
546 {
547 	u_int32_t low, high;
548 
549 	__asm __volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (pmc));
550 	return (low | ((u_int64_t)high << 32));
551 }
552 
553 #define _RDTSC_SUPPORTED_
554 
555 static __inline tsc_uclock_t
556 rdtsc(void)
557 {
558 	u_int32_t low, high;
559 
560 	__asm __volatile("rdtsc" : "=a" (low), "=d" (high));
561 	return (low | ((tsc_uclock_t)high << 32));
562 }
563 
564 #ifdef _KERNEL
565 #include <machine/cputypes.h>
566 #include <machine/md_var.h>
567 
568 static __inline tsc_uclock_t
569 rdtsc_ordered(void)
570 {
571 	if (cpu_vendor_id == CPU_VENDOR_INTEL)
572 		cpu_lfence();
573 	else
574 		cpu_mfence();
575 	return rdtsc();
576 }
577 #endif
578 
579 static __inline void
580 wbinvd(void)
581 {
582 	__asm __volatile("wbinvd");
583 }
584 
585 #if defined(_KERNEL)
586 void cpu_wbinvd_on_all_cpus_callback(void *arg);
587 
588 static __inline void
589 cpu_wbinvd_on_all_cpus(void)
590 {
591 	lwkt_cpusync_simple(smp_active_mask, cpu_wbinvd_on_all_cpus_callback, NULL);
592 }
593 #endif
594 
595 static __inline void
596 write_rflags(u_long rf)
597 {
598 	__asm __volatile("pushq %0;  popfq" : : "r" (rf));
599 }
600 
601 static __inline void
602 wrmsr(u_int msr, u_int64_t newval)
603 {
604 	u_int32_t low, high;
605 
606 	low = newval;
607 	high = newval >> 32;
608 	__asm __volatile("wrmsr" : : "a" (low), "d" (high), "c" (msr));
609 }
610 
611 static __inline void
612 xsetbv(u_int ecx, u_int eax, u_int edx)
613 {
614 	__asm __volatile(".byte 0x0f,0x01,0xd1"
615 	    :
616 	    : "a" (eax), "c" (ecx), "d" (edx));
617 }
618 
619 static __inline void
620 load_cr0(u_long data)
621 {
622 
623 	__asm __volatile("movq %0,%%cr0" : : "r" (data));
624 }
625 
626 static __inline u_long
627 rcr0(void)
628 {
629 	u_long	data;
630 
631 	__asm __volatile("movq %%cr0,%0" : "=r" (data));
632 	return (data);
633 }
634 
635 static __inline u_long
636 rcr2(void)
637 {
638 	u_long	data;
639 
640 	__asm __volatile("movq %%cr2,%0" : "=r" (data));
641 	return (data);
642 }
643 
644 static __inline void
645 load_cr3(u_long data)
646 {
647 
648 	__asm __volatile("movq %0,%%cr3" : : "r" (data) : "memory");
649 }
650 
651 static __inline u_long
652 rcr3(void)
653 {
654 	u_long	data;
655 
656 	__asm __volatile("movq %%cr3,%0" : "=r" (data));
657 	return (data);
658 }
659 
660 static __inline void
661 load_cr4(u_long data)
662 {
663 	__asm __volatile("movq %0,%%cr4" : : "r" (data));
664 }
665 
666 static __inline u_long
667 rcr4(void)
668 {
669 	u_long	data;
670 
671 	__asm __volatile("movq %%cr4,%0" : "=r" (data));
672 	return (data);
673 }
674 
675 #ifndef _CPU_INVLTLB_DEFINED
676 
677 /*
678  * Invalidate the TLB on this cpu only
679  */
680 static __inline void
681 cpu_invltlb(void)
682 {
683 	load_cr3(rcr3());
684 #if defined(SWTCH_OPTIM_STATS)
685 	++tlb_flush_count;
686 #endif
687 }
688 
689 #endif
690 
691 void smp_invltlb(void);
692 void smp_sniff(void);
693 void cpu_sniff(int);
694 void hard_sniff(struct trapframe *);
695 
696 static __inline u_short
697 rfs(void)
698 {
699 	u_short sel;
700 	__asm __volatile("movw %%fs,%0" : "=rm" (sel));
701 	return (sel);
702 }
703 
704 static __inline u_short
705 rgs(void)
706 {
707 	u_short sel;
708 	__asm __volatile("movw %%gs,%0" : "=rm" (sel));
709 	return (sel);
710 }
711 
712 static __inline void
713 load_ds(u_short sel)
714 {
715 	__asm __volatile("movw %0,%%ds" : : "rm" (sel));
716 }
717 
718 static __inline void
719 load_es(u_short sel)
720 {
721 	__asm __volatile("movw %0,%%es" : : "rm" (sel));
722 }
723 
724 #ifdef _KERNEL
725 /* This is defined in <machine/specialreg.h> but is too painful to get to */
726 #ifndef	MSR_FSBASE
727 #define	MSR_FSBASE	0xc0000100
728 #endif
729 static __inline void
730 load_fs(u_short sel)
731 {
732 	/* Preserve the fsbase value across the selector load */
733 	__asm __volatile("rdmsr; movw %0,%%fs; wrmsr"
734             : : "rm" (sel), "c" (MSR_FSBASE) : "eax", "edx");
735 }
736 
737 #ifndef	MSR_GSBASE
738 #define	MSR_GSBASE	0xc0000101
739 #endif
740 static __inline void
741 load_gs(u_short sel)
742 {
743 	/*
744 	 * Preserve the gsbase value across the selector load.
745 	 * Note that we have to disable interrupts because the gsbase
746 	 * being trashed happens to be the kernel gsbase at the time.
747 	 */
748 	__asm __volatile("pushfq; cli; rdmsr; movw %0,%%gs; wrmsr; popfq"
749             : : "rm" (sel), "c" (MSR_GSBASE) : "eax", "edx");
750 }
751 #else
752 /* Usable by userland */
753 static __inline void
754 load_fs(u_short sel)
755 {
756 	__asm __volatile("movw %0,%%fs" : : "rm" (sel));
757 }
758 
759 static __inline void
760 load_gs(u_short sel)
761 {
762 	__asm __volatile("movw %0,%%gs" : : "rm" (sel));
763 }
764 #endif
765 
766 /* void lidt(struct region_descriptor *addr); */
767 static __inline void
768 lidt(struct region_descriptor *addr)
769 {
770 	__asm __volatile("lidt (%0)" : : "r" (addr));
771 }
772 
773 /* void lldt(u_short sel); */
774 static __inline void
775 lldt(u_short sel)
776 {
777 	__asm __volatile("lldt %0" : : "r" (sel));
778 }
779 
780 /* void ltr(u_short sel); */
781 static __inline void
782 ltr(u_short sel)
783 {
784 	__asm __volatile("ltr %0" : : "r" (sel));
785 }
786 
787 static __inline u_int64_t
788 rdr0(void)
789 {
790 	u_int64_t data;
791 	__asm __volatile("movq %%dr0,%0" : "=r" (data));
792 	return (data);
793 }
794 
795 static __inline void
796 load_dr0(u_int64_t dr0)
797 {
798 	__asm __volatile("movq %0,%%dr0" : : "r" (dr0));
799 }
800 
801 static __inline u_int64_t
802 rdr1(void)
803 {
804 	u_int64_t data;
805 	__asm __volatile("movq %%dr1,%0" : "=r" (data));
806 	return (data);
807 }
808 
809 static __inline void
810 load_dr1(u_int64_t dr1)
811 {
812 	__asm __volatile("movq %0,%%dr1" : : "r" (dr1));
813 }
814 
815 static __inline u_int64_t
816 rdr2(void)
817 {
818 	u_int64_t data;
819 	__asm __volatile("movq %%dr2,%0" : "=r" (data));
820 	return (data);
821 }
822 
823 static __inline void
824 load_dr2(u_int64_t dr2)
825 {
826 	__asm __volatile("movq %0,%%dr2" : : "r" (dr2));
827 }
828 
829 static __inline u_int64_t
830 rdr3(void)
831 {
832 	u_int64_t data;
833 	__asm __volatile("movq %%dr3,%0" : "=r" (data));
834 	return (data);
835 }
836 
837 static __inline void
838 load_dr3(u_int64_t dr3)
839 {
840 	__asm __volatile("movq %0,%%dr3" : : "r" (dr3));
841 }
842 
843 static __inline u_int64_t
844 rdr4(void)
845 {
846 	u_int64_t data;
847 	__asm __volatile("movq %%dr4,%0" : "=r" (data));
848 	return (data);
849 }
850 
851 static __inline void
852 load_dr4(u_int64_t dr4)
853 {
854 	__asm __volatile("movq %0,%%dr4" : : "r" (dr4));
855 }
856 
857 static __inline u_int64_t
858 rdr5(void)
859 {
860 	u_int64_t data;
861 	__asm __volatile("movq %%dr5,%0" : "=r" (data));
862 	return (data);
863 }
864 
865 static __inline void
866 load_dr5(u_int64_t dr5)
867 {
868 	__asm __volatile("movq %0,%%dr5" : : "r" (dr5));
869 }
870 
871 static __inline u_int64_t
872 rdr6(void)
873 {
874 	u_int64_t data;
875 	__asm __volatile("movq %%dr6,%0" : "=r" (data));
876 	return (data);
877 }
878 
879 static __inline void
880 load_dr6(u_int64_t dr6)
881 {
882 	__asm __volatile("movq %0,%%dr6" : : "r" (dr6));
883 }
884 
885 static __inline u_int64_t
886 rdr7(void)
887 {
888 	u_int64_t data;
889 	__asm __volatile("movq %%dr7,%0" : "=r" (data));
890 	return (data);
891 }
892 
893 static __inline void
894 load_dr7(u_int64_t dr7)
895 {
896 	__asm __volatile("movq %0,%%dr7" : : "r" (dr7));
897 }
898 
899 static __inline register_t
900 intr_disable(void)
901 {
902 	register_t rflags;
903 
904 	rflags = read_rflags();
905 	cpu_disable_intr();
906 	return (rflags);
907 }
908 
909 static __inline void
910 intr_restore(register_t rflags)
911 {
912 	write_rflags(rflags);
913 }
914 
915 #else /* !__GNUC__ */
916 
917 int	breakpoint(void);
918 void	cpu_pause(void);
919 u_int	bsfl(u_int mask);
920 u_int	bsrl(u_int mask);
921 void	cpu_disable_intr(void);
922 void	cpu_enable_intr(void);
923 void	cpu_invlpg(u_long addr);
924 void	cpu_invlpg_range(u_long start, u_long end);
925 void	do_cpuid(u_int ax, u_int *p);
926 void	halt(void);
927 u_char	inb(u_int port);
928 u_int	inl(u_int port);
929 void	insb(u_int port, void *addr, size_t cnt);
930 void	insl(u_int port, void *addr, size_t cnt);
931 void	insw(u_int port, void *addr, size_t cnt);
932 void	invd(void);
933 void	invlpg_range(u_int start, u_int end);
934 void	cpu_invltlb(void);
935 u_short	inw(u_int port);
936 void	load_cr0(u_int cr0);
937 void	load_cr3(u_int cr3);
938 void	load_cr4(u_int cr4);
939 void	load_fs(u_int sel);
940 void	load_gs(u_int sel);
941 struct region_descriptor;
942 void	lidt(struct region_descriptor *addr);
943 void	lldt(u_short sel);
944 void	ltr(u_short sel);
945 void	outb(u_int port, u_char data);
946 void	outl(u_int port, u_int data);
947 void	outsb(u_int port, void *addr, size_t cnt);
948 void	outsl(u_int port, void *addr, size_t cnt);
949 void	outsw(u_int port, void *addr, size_t cnt);
950 void	outw(u_int port, u_short data);
951 void	ia32_pause(void);
952 u_int	rcr0(void);
953 u_int	rcr2(void);
954 u_int	rcr3(void);
955 u_int	rcr4(void);
956 u_short	rfs(void);
957 u_short	rgs(void);
958 u_int64_t rdmsr(u_int msr);
959 u_int64_t rdpmc(u_int pmc);
960 tsc_uclock_t rdtsc(void);
961 u_int	read_rflags(void);
962 void	wbinvd(void);
963 void	write_rflags(u_int rf);
964 void	wrmsr(u_int msr, u_int64_t newval);
965 u_int64_t	rdr0(void);
966 void	load_dr0(u_int64_t dr0);
967 u_int64_t	rdr1(void);
968 void	load_dr1(u_int64_t dr1);
969 u_int64_t	rdr2(void);
970 void	load_dr2(u_int64_t dr2);
971 u_int64_t	rdr3(void);
972 void	load_dr3(u_int64_t dr3);
973 u_int64_t	rdr4(void);
974 void	load_dr4(u_int64_t dr4);
975 u_int64_t	rdr5(void);
976 void	load_dr5(u_int64_t dr5);
977 u_int64_t	rdr6(void);
978 void	load_dr6(u_int64_t dr6);
979 u_int64_t	rdr7(void);
980 void	load_dr7(u_int64_t dr7);
981 register_t	intr_disable(void);
982 void	intr_restore(register_t rf);
983 
984 #endif	/* __GNUC__ */
985 
986 int	rdmsr_safe(u_int msr, uint64_t *val);
987 int wrmsr_safe(u_int msr, uint64_t newval);
988 void	reset_dbregs(void);
989 void	smap_open(void);
990 void	smap_close(void);
991 
992 __END_DECLS
993 
994 #endif /* !_CPU_CPUFUNC_H_ */
995