xref: /dragonfly/sys/cpu/x86_64/include/cpufunc.h (revision cecb9aae)
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. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  * $FreeBSD: src/sys/amd64/include/cpufunc.h,v 1.139 2004/01/28 23:53:04 peter Exp $
36  */
37 
38 /*
39  * Functions to provide access to special i386 instructions.
40  * This in included in sys/systm.h, and that file should be
41  * used in preference to this.
42  */
43 
44 #ifndef _CPU_CPUFUNC_H_
45 #define	_CPU_CPUFUNC_H_
46 
47 #include <sys/cdefs.h>
48 #include <sys/thread.h>
49 #include <machine/psl.h>
50 #include <machine/smp.h>
51 
52 struct thread;
53 struct region_descriptor;
54 
55 __BEGIN_DECLS
56 #define readb(va)	(*(volatile u_int8_t *) (va))
57 #define readw(va)	(*(volatile u_int16_t *) (va))
58 #define readl(va)	(*(volatile u_int32_t *) (va))
59 #define readq(va)	(*(volatile u_int64_t *) (va))
60 
61 #define writeb(va, d)	(*(volatile u_int8_t *) (va) = (d))
62 #define writew(va, d)	(*(volatile u_int16_t *) (va) = (d))
63 #define writel(va, d)	(*(volatile u_int32_t *) (va) = (d))
64 #define writeq(va, d)	(*(volatile u_int64_t *) (va) = (d))
65 
66 #ifdef	__GNUC__
67 
68 #include <machine/lock.h>		/* XXX */
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 #endif /* _KERNEL */
305 
306 static __inline void
307 halt(void)
308 {
309 	__asm __volatile("hlt");
310 }
311 
312 /*
313  * The following complications are to get around gcc not having a
314  * constraint letter for the range 0..255.  We still put "d" in the
315  * constraint because "i" isn't a valid constraint when the port
316  * isn't constant.  This only matters for -O0 because otherwise
317  * the non-working version gets optimized away.
318  *
319  * Use an expression-statement instead of a conditional expression
320  * because gcc-2.6.0 would promote the operands of the conditional
321  * and produce poor code for "if ((inb(var) & const1) == const2)".
322  *
323  * The unnecessary test `(port) < 0x10000' is to generate a warning if
324  * the `port' has type u_short or smaller.  Such types are pessimal.
325  * This actually only works for signed types.  The range check is
326  * careful to avoid generating warnings.
327  */
328 #define	inb(port) __extension__ ({					\
329 	u_char	_data;							\
330 	if (__builtin_constant_p(port) && ((port) & 0xffff) < 0x100	\
331 	    && (port) < 0x10000)					\
332 		_data = inbc(port);					\
333 	else								\
334 		_data = inbv(port);					\
335 	_data; })
336 
337 #define	outb(port, data) (						\
338 	__builtin_constant_p(port) && ((port) & 0xffff) < 0x100		\
339 	&& (port) < 0x10000						\
340 	? outbc(port, data) : outbv(port, data))
341 
342 static __inline u_char
343 inbc(u_int port)
344 {
345 	u_char	data;
346 
347 	__asm __volatile("inb %1,%0" : "=a" (data) : "id" ((u_short)(port)));
348 	return (data);
349 }
350 
351 static __inline void
352 outbc(u_int port, u_char data)
353 {
354 	__asm __volatile("outb %0,%1" : : "a" (data), "id" ((u_short)(port)));
355 }
356 
357 static __inline u_char
358 inbv(u_int port)
359 {
360 	u_char	data;
361 	/*
362 	 * We use %%dx and not %1 here because i/o is done at %dx and not at
363 	 * %edx, while gcc generates inferior code (movw instead of movl)
364 	 * if we tell it to load (u_short) port.
365 	 */
366 	__asm __volatile("inb %%dx,%0" : "=a" (data) : "d" (port));
367 	return (data);
368 }
369 
370 static __inline u_int
371 inl(u_int port)
372 {
373 	u_int	data;
374 
375 	__asm __volatile("inl %%dx,%0" : "=a" (data) : "d" (port));
376 	return (data);
377 }
378 
379 static __inline void
380 insb(u_int port, void *addr, size_t cnt)
381 {
382 	__asm __volatile("cld; rep; insb"
383 			 : "+D" (addr), "+c" (cnt)
384 			 : "d" (port)
385 			 : "memory");
386 }
387 
388 static __inline void
389 insw(u_int port, void *addr, size_t cnt)
390 {
391 	__asm __volatile("cld; rep; insw"
392 			 : "+D" (addr), "+c" (cnt)
393 			 : "d" (port)
394 			 : "memory");
395 }
396 
397 static __inline void
398 insl(u_int port, void *addr, size_t cnt)
399 {
400 	__asm __volatile("cld; rep; insl"
401 			 : "+D" (addr), "+c" (cnt)
402 			 : "d" (port)
403 			 : "memory");
404 }
405 
406 static __inline void
407 invd(void)
408 {
409 	__asm __volatile("invd");
410 }
411 
412 #if defined(_KERNEL)
413 
414 /*
415  * If we are not a true-SMP box then smp_invltlb() is a NOP.  Note that this
416  * will cause the invl*() functions to be equivalent to the cpu_invl*()
417  * functions.
418  */
419 void smp_invltlb(void);
420 void smp_invltlb_intr(void);
421 
422 #ifndef _CPU_INVLPG_DEFINED
423 
424 /*
425  * Invalidate a patricular VA on this cpu only
426  */
427 static __inline void
428 cpu_invlpg(void *addr)
429 {
430 	__asm __volatile("invlpg %0" : : "m" (*(char *)addr) : "memory");
431 }
432 
433 #endif
434 
435 static __inline void
436 cpu_nop(void)
437 {
438 	__asm __volatile("rep; nop");
439 }
440 
441 #endif	/* _KERNEL */
442 
443 static __inline u_short
444 inw(u_int port)
445 {
446 	u_short	data;
447 
448 	__asm __volatile("inw %%dx,%0" : "=a" (data) : "d" (port));
449 	return (data);
450 }
451 
452 static __inline u_int
453 loadandclear(volatile u_int *addr)
454 {
455 	u_int   result;
456 
457 	__asm __volatile("xorl %0,%0; xchgl %1,%0"
458 			: "=&r" (result) : "m" (*addr));
459 	return (result);
460 }
461 
462 static __inline void
463 outbv(u_int port, u_char data)
464 {
465 	u_char	al;
466 	/*
467 	 * Use an unnecessary assignment to help gcc's register allocator.
468 	 * This make a large difference for gcc-1.40 and a tiny difference
469 	 * for gcc-2.6.0.  For gcc-1.40, al had to be ``asm("ax")'' for
470 	 * best results.  gcc-2.6.0 can't handle this.
471 	 */
472 	al = data;
473 	__asm __volatile("outb %0,%%dx" : : "a" (al), "d" (port));
474 }
475 
476 static __inline void
477 outl(u_int port, u_int data)
478 {
479 	/*
480 	 * outl() and outw() aren't used much so we haven't looked at
481 	 * possible micro-optimizations such as the unnecessary
482 	 * assignment for them.
483 	 */
484 	__asm __volatile("outl %0,%%dx" : : "a" (data), "d" (port));
485 }
486 
487 static __inline void
488 outsb(u_int port, const void *addr, size_t cnt)
489 {
490 	__asm __volatile("cld; rep; outsb"
491 			 : "+S" (addr), "+c" (cnt)
492 			 : "d" (port));
493 }
494 
495 static __inline void
496 outsw(u_int port, const void *addr, size_t cnt)
497 {
498 	__asm __volatile("cld; rep; outsw"
499 			 : "+S" (addr), "+c" (cnt)
500 			 : "d" (port));
501 }
502 
503 static __inline void
504 outsl(u_int port, const void *addr, size_t cnt)
505 {
506 	__asm __volatile("cld; rep; outsl"
507 			 : "+S" (addr), "+c" (cnt)
508 			 : "d" (port));
509 }
510 
511 static __inline void
512 outw(u_int port, u_short data)
513 {
514 	__asm __volatile("outw %0,%%dx" : : "a" (data), "d" (port));
515 }
516 
517 static __inline void
518 ia32_pause(void)
519 {
520 	__asm __volatile("pause");
521 }
522 
523 static __inline u_long
524 read_rflags(void)
525 {
526 	u_long	rf;
527 
528 	__asm __volatile("pushfq; popq %0" : "=r" (rf));
529 	return (rf);
530 }
531 
532 static __inline u_int64_t
533 rdmsr(u_int msr)
534 {
535 	u_int32_t low, high;
536 
537 	__asm __volatile("rdmsr" : "=a" (low), "=d" (high) : "c" (msr));
538 	return (low | ((u_int64_t)high << 32));
539 }
540 
541 static __inline u_int64_t
542 rdpmc(u_int pmc)
543 {
544 	u_int32_t low, high;
545 
546 	__asm __volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (pmc));
547 	return (low | ((u_int64_t)high << 32));
548 }
549 
550 #define _RDTSC_SUPPORTED_
551 
552 static __inline u_int64_t
553 rdtsc(void)
554 {
555 	u_int32_t low, high;
556 
557 	__asm __volatile("rdtsc" : "=a" (low), "=d" (high));
558 	return (low | ((u_int64_t)high << 32));
559 }
560 
561 static __inline void
562 wbinvd(void)
563 {
564 	__asm __volatile("wbinvd");
565 }
566 
567 #if defined(_KERNEL)
568 void cpu_wbinvd_on_all_cpus_callback(void *arg);
569 
570 static __inline void
571 cpu_wbinvd_on_all_cpus(void)
572 {
573 	lwkt_cpusync_simple(smp_active_mask, cpu_wbinvd_on_all_cpus_callback, NULL);
574 }
575 #endif
576 
577 static __inline void
578 write_rflags(u_long rf)
579 {
580 	__asm __volatile("pushq %0;  popfq" : : "r" (rf));
581 }
582 
583 static __inline void
584 wrmsr(u_int msr, u_int64_t newval)
585 {
586 	u_int32_t low, high;
587 
588 	low = newval;
589 	high = newval >> 32;
590 	__asm __volatile("wrmsr" : : "a" (low), "d" (high), "c" (msr));
591 }
592 
593 static __inline void
594 xsetbv(u_int ecx, u_int eax, u_int edx)
595 {
596 	__asm __volatile(".byte 0x0f,0x01,0xd1"
597 	    :
598 	    : "a" (eax), "c" (ecx), "d" (edx));
599 }
600 
601 static __inline void
602 load_cr0(u_long data)
603 {
604 
605 	__asm __volatile("movq %0,%%cr0" : : "r" (data));
606 }
607 
608 static __inline u_long
609 rcr0(void)
610 {
611 	u_long	data;
612 
613 	__asm __volatile("movq %%cr0,%0" : "=r" (data));
614 	return (data);
615 }
616 
617 static __inline u_long
618 rcr2(void)
619 {
620 	u_long	data;
621 
622 	__asm __volatile("movq %%cr2,%0" : "=r" (data));
623 	return (data);
624 }
625 
626 static __inline void
627 load_cr3(u_long data)
628 {
629 
630 	__asm __volatile("movq %0,%%cr3" : : "r" (data) : "memory");
631 }
632 
633 static __inline u_long
634 rcr3(void)
635 {
636 	u_long	data;
637 
638 	__asm __volatile("movq %%cr3,%0" : "=r" (data));
639 	return (data);
640 }
641 
642 static __inline void
643 load_cr4(u_long data)
644 {
645 	__asm __volatile("movq %0,%%cr4" : : "r" (data));
646 }
647 
648 static __inline u_long
649 rcr4(void)
650 {
651 	u_long	data;
652 
653 	__asm __volatile("movq %%cr4,%0" : "=r" (data));
654 	return (data);
655 }
656 
657 #ifndef _CPU_INVLTLB_DEFINED
658 
659 /*
660  * Invalidate the TLB on this cpu only
661  */
662 static __inline void
663 cpu_invltlb(void)
664 {
665 	load_cr3(rcr3());
666 #if defined(SWTCH_OPTIM_STATS)
667 	++tlb_flush_count;
668 #endif
669 }
670 
671 #endif
672 
673 /*
674  * TLB flush for an individual page (even if it has PG_G).
675  * Only works on 486+ CPUs (i386 does not have PG_G).
676  */
677 static __inline void
678 invlpg(u_long addr)
679 {
680 
681 	__asm __volatile("invlpg %0" : : "m" (*(char *)addr) : "memory");
682 }
683 
684 static __inline u_short
685 rfs(void)
686 {
687 	u_short sel;
688 	__asm __volatile("movw %%fs,%0" : "=rm" (sel));
689 	return (sel);
690 }
691 
692 static __inline u_short
693 rgs(void)
694 {
695 	u_short sel;
696 	__asm __volatile("movw %%gs,%0" : "=rm" (sel));
697 	return (sel);
698 }
699 
700 static __inline void
701 load_ds(u_short sel)
702 {
703 	__asm __volatile("movw %0,%%ds" : : "rm" (sel));
704 }
705 
706 static __inline void
707 load_es(u_short sel)
708 {
709 	__asm __volatile("movw %0,%%es" : : "rm" (sel));
710 }
711 
712 #ifdef _KERNEL
713 /* This is defined in <machine/specialreg.h> but is too painful to get to */
714 #ifndef	MSR_FSBASE
715 #define	MSR_FSBASE	0xc0000100
716 #endif
717 static __inline void
718 load_fs(u_short sel)
719 {
720 	/* Preserve the fsbase value across the selector load */
721 	__asm __volatile("rdmsr; movw %0,%%fs; wrmsr"
722             : : "rm" (sel), "c" (MSR_FSBASE) : "eax", "edx");
723 }
724 
725 #ifndef	MSR_GSBASE
726 #define	MSR_GSBASE	0xc0000101
727 #endif
728 static __inline void
729 load_gs(u_short sel)
730 {
731 	/*
732 	 * Preserve the gsbase value across the selector load.
733 	 * Note that we have to disable interrupts because the gsbase
734 	 * being trashed happens to be the kernel gsbase at the time.
735 	 */
736 	__asm __volatile("pushfq; cli; rdmsr; movw %0,%%gs; wrmsr; popfq"
737             : : "rm" (sel), "c" (MSR_GSBASE) : "eax", "edx");
738 }
739 #else
740 /* Usable by userland */
741 static __inline void
742 load_fs(u_short sel)
743 {
744 	__asm __volatile("movw %0,%%fs" : : "rm" (sel));
745 }
746 
747 static __inline void
748 load_gs(u_short sel)
749 {
750 	__asm __volatile("movw %0,%%gs" : : "rm" (sel));
751 }
752 #endif
753 
754 /* void lidt(struct region_descriptor *addr); */
755 static __inline void
756 lidt(struct region_descriptor *addr)
757 {
758 	__asm __volatile("lidt (%0)" : : "r" (addr));
759 }
760 
761 /* void lldt(u_short sel); */
762 static __inline void
763 lldt(u_short sel)
764 {
765 	__asm __volatile("lldt %0" : : "r" (sel));
766 }
767 
768 /* void ltr(u_short sel); */
769 static __inline void
770 ltr(u_short sel)
771 {
772 	__asm __volatile("ltr %0" : : "r" (sel));
773 }
774 
775 static __inline u_int64_t
776 rdr0(void)
777 {
778 	u_int64_t data;
779 	__asm __volatile("movq %%dr0,%0" : "=r" (data));
780 	return (data);
781 }
782 
783 static __inline void
784 load_dr0(u_int64_t dr0)
785 {
786 	__asm __volatile("movq %0,%%dr0" : : "r" (dr0));
787 }
788 
789 static __inline u_int64_t
790 rdr1(void)
791 {
792 	u_int64_t data;
793 	__asm __volatile("movq %%dr1,%0" : "=r" (data));
794 	return (data);
795 }
796 
797 static __inline void
798 load_dr1(u_int64_t dr1)
799 {
800 	__asm __volatile("movq %0,%%dr1" : : "r" (dr1));
801 }
802 
803 static __inline u_int64_t
804 rdr2(void)
805 {
806 	u_int64_t data;
807 	__asm __volatile("movq %%dr2,%0" : "=r" (data));
808 	return (data);
809 }
810 
811 static __inline void
812 load_dr2(u_int64_t dr2)
813 {
814 	__asm __volatile("movq %0,%%dr2" : : "r" (dr2));
815 }
816 
817 static __inline u_int64_t
818 rdr3(void)
819 {
820 	u_int64_t data;
821 	__asm __volatile("movq %%dr3,%0" : "=r" (data));
822 	return (data);
823 }
824 
825 static __inline void
826 load_dr3(u_int64_t dr3)
827 {
828 	__asm __volatile("movq %0,%%dr3" : : "r" (dr3));
829 }
830 
831 static __inline u_int64_t
832 rdr4(void)
833 {
834 	u_int64_t data;
835 	__asm __volatile("movq %%dr4,%0" : "=r" (data));
836 	return (data);
837 }
838 
839 static __inline void
840 load_dr4(u_int64_t dr4)
841 {
842 	__asm __volatile("movq %0,%%dr4" : : "r" (dr4));
843 }
844 
845 static __inline u_int64_t
846 rdr5(void)
847 {
848 	u_int64_t data;
849 	__asm __volatile("movq %%dr5,%0" : "=r" (data));
850 	return (data);
851 }
852 
853 static __inline void
854 load_dr5(u_int64_t dr5)
855 {
856 	__asm __volatile("movq %0,%%dr5" : : "r" (dr5));
857 }
858 
859 static __inline u_int64_t
860 rdr6(void)
861 {
862 	u_int64_t data;
863 	__asm __volatile("movq %%dr6,%0" : "=r" (data));
864 	return (data);
865 }
866 
867 static __inline void
868 load_dr6(u_int64_t dr6)
869 {
870 	__asm __volatile("movq %0,%%dr6" : : "r" (dr6));
871 }
872 
873 static __inline u_int64_t
874 rdr7(void)
875 {
876 	u_int64_t data;
877 	__asm __volatile("movq %%dr7,%0" : "=r" (data));
878 	return (data);
879 }
880 
881 static __inline void
882 load_dr7(u_int64_t dr7)
883 {
884 	__asm __volatile("movq %0,%%dr7" : : "r" (dr7));
885 }
886 
887 static __inline register_t
888 intr_disable(void)
889 {
890 	register_t rflags;
891 
892 	rflags = read_rflags();
893 	cpu_disable_intr();
894 	return (rflags);
895 }
896 
897 static __inline void
898 intr_restore(register_t rflags)
899 {
900 	write_rflags(rflags);
901 }
902 
903 #else /* !__GNUC__ */
904 
905 int	breakpoint(void);
906 void	cpu_pause(void);
907 u_int	bsfl(u_int mask);
908 u_int	bsrl(u_int mask);
909 void	cpu_disable_intr(void);
910 void	cpu_enable_intr(void);
911 void	cpu_invlpg(u_long addr);
912 void	cpu_invlpg_range(u_long start, u_long end);
913 void	do_cpuid(u_int ax, u_int *p);
914 void	halt(void);
915 u_char	inb(u_int port);
916 u_int	inl(u_int port);
917 void	insb(u_int port, void *addr, size_t cnt);
918 void	insl(u_int port, void *addr, size_t cnt);
919 void	insw(u_int port, void *addr, size_t cnt);
920 void	invd(void);
921 void	invlpg(u_int addr);
922 void	invlpg_range(u_int start, u_int end);
923 void	cpu_invltlb(void);
924 u_short	inw(u_int port);
925 void	load_cr0(u_int cr0);
926 void	load_cr3(u_int cr3);
927 void	load_cr4(u_int cr4);
928 void	load_fs(u_int sel);
929 void	load_gs(u_int sel);
930 struct region_descriptor;
931 void	lidt(struct region_descriptor *addr);
932 void	lldt(u_short sel);
933 void	ltr(u_short sel);
934 void	outb(u_int port, u_char data);
935 void	outl(u_int port, u_int data);
936 void	outsb(u_int port, void *addr, size_t cnt);
937 void	outsl(u_int port, void *addr, size_t cnt);
938 void	outsw(u_int port, void *addr, size_t cnt);
939 void	outw(u_int port, u_short data);
940 void	ia32_pause(void);
941 u_int	rcr0(void);
942 u_int	rcr2(void);
943 u_int	rcr3(void);
944 u_int	rcr4(void);
945 u_short	rfs(void);
946 u_short	rgs(void);
947 u_int64_t rdmsr(u_int msr);
948 u_int64_t rdpmc(u_int pmc);
949 u_int64_t rdtsc(void);
950 u_int	read_rflags(void);
951 void	wbinvd(void);
952 void	write_rflags(u_int rf);
953 void	wrmsr(u_int msr, u_int64_t newval);
954 u_int64_t	rdr0(void);
955 void	load_dr0(u_int64_t dr0);
956 u_int64_t	rdr1(void);
957 void	load_dr1(u_int64_t dr1);
958 u_int64_t	rdr2(void);
959 void	load_dr2(u_int64_t dr2);
960 u_int64_t	rdr3(void);
961 void	load_dr3(u_int64_t dr3);
962 u_int64_t	rdr4(void);
963 void	load_dr4(u_int64_t dr4);
964 u_int64_t	rdr5(void);
965 void	load_dr5(u_int64_t dr5);
966 u_int64_t	rdr6(void);
967 void	load_dr6(u_int64_t dr6);
968 u_int64_t	rdr7(void);
969 void	load_dr7(u_int64_t dr7);
970 register_t	intr_disable(void);
971 void	intr_restore(register_t rf);
972 
973 #endif	/* __GNUC__ */
974 
975 int	rdmsr_safe(u_int msr, uint64_t *val);
976 void	reset_dbregs(void);
977 
978 __END_DECLS
979 
980 #endif /* !_CPU_CPUFUNC_H_ */
981