xref: /freebsd/sys/amd64/include/cpufunc.h (revision 63d45d7da0eac8efdeb765ac5caddfc2c5ca021e)
1 /*-
2  * Copyright (c) 2003 Peter Wemm.
3  * Copyright (c) 1993 The Regents of the University of California.
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 4. Neither the name of the University nor the names of its contributors
15  *    may be used to endorse or promote products derived from this software
16  *    without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  * $FreeBSD$
31  */
32 
33 /*
34  * Functions to provide access to special i386 instructions.
35  * This in included in sys/systm.h, and that file should be
36  * used in preference to this.
37  */
38 
39 #ifndef _MACHINE_CPUFUNC_H_
40 #define	_MACHINE_CPUFUNC_H_
41 
42 #ifndef _SYS_CDEFS_H_
43 #error this file needs sys/cdefs.h as a prerequisite
44 #endif
45 
46 struct region_descriptor;
47 
48 #define readb(va)	(*(volatile u_int8_t *) (va))
49 #define readw(va)	(*(volatile u_int16_t *) (va))
50 #define readl(va)	(*(volatile u_int32_t *) (va))
51 #define readq(va)	(*(volatile u_int64_t *) (va))
52 
53 #define writeb(va, d)	(*(volatile u_int8_t *) (va) = (d))
54 #define writew(va, d)	(*(volatile u_int16_t *) (va) = (d))
55 #define writel(va, d)	(*(volatile u_int32_t *) (va) = (d))
56 #define writeq(va, d)	(*(volatile u_int64_t *) (va) = (d))
57 
58 #if defined(__GNUCLIKE_ASM) && defined(__CC_SUPPORTS___INLINE)
59 
60 static __inline void
61 breakpoint(void)
62 {
63 	__asm __volatile("int $3");
64 }
65 
66 static __inline u_int
67 bsfl(u_int mask)
68 {
69 	u_int	result;
70 
71 	__asm __volatile("bsfl %1,%0" : "=r" (result) : "rm" (mask));
72 	return (result);
73 }
74 
75 static __inline u_long
76 bsfq(u_long mask)
77 {
78 	u_long	result;
79 
80 	__asm __volatile("bsfq %1,%0" : "=r" (result) : "rm" (mask));
81 	return (result);
82 }
83 
84 static __inline u_int
85 bsrl(u_int mask)
86 {
87 	u_int	result;
88 
89 	__asm __volatile("bsrl %1,%0" : "=r" (result) : "rm" (mask));
90 	return (result);
91 }
92 
93 static __inline u_long
94 bsrq(u_long mask)
95 {
96 	u_long	result;
97 
98 	__asm __volatile("bsrq %1,%0" : "=r" (result) : "rm" (mask));
99 	return (result);
100 }
101 
102 static __inline void
103 disable_intr(void)
104 {
105 	__asm __volatile("cli" : : : "memory");
106 }
107 
108 static __inline void
109 do_cpuid(u_int ax, u_int *p)
110 {
111 	__asm __volatile("cpuid"
112 			 : "=a" (p[0]), "=b" (p[1]), "=c" (p[2]), "=d" (p[3])
113 			 :  "0" (ax));
114 }
115 
116 static __inline void
117 enable_intr(void)
118 {
119 	__asm __volatile("sti");
120 }
121 
122 #ifdef _KERNEL
123 
124 #define	HAVE_INLINE_FFS
125 #define        ffs(x)  __builtin_ffs(x)
126 
127 #define	HAVE_INLINE_FFSL
128 
129 static __inline int
130 ffsl(long mask)
131 {
132 	return (mask == 0 ? mask : (int)bsfq((u_long)mask) + 1);
133 }
134 
135 #define	HAVE_INLINE_FLS
136 
137 static __inline int
138 fls(int mask)
139 {
140 	return (mask == 0 ? mask : (int)bsrl((u_int)mask) + 1);
141 }
142 
143 #define	HAVE_INLINE_FLSL
144 
145 static __inline int
146 flsl(long mask)
147 {
148 	return (mask == 0 ? mask : (int)bsrq((u_long)mask) + 1);
149 }
150 
151 #endif /* _KERNEL */
152 
153 static __inline void
154 halt(void)
155 {
156 	__asm __volatile("hlt");
157 }
158 
159 #if !defined(__GNUCLIKE_BUILTIN_CONSTANT_P) || __GNUCLIKE_ASM < 3
160 
161 #define	inb(port)		inbv(port)
162 #define	outb(port, data)	outbv(port, data)
163 
164 #else /* __GNUCLIKE_BUILTIN_CONSTANT_P  && __GNUCLIKE_ASM >= 3 */
165 
166 /*
167  * The following complications are to get around gcc not having a
168  * constraint letter for the range 0..255.  We still put "d" in the
169  * constraint because "i" isn't a valid constraint when the port
170  * isn't constant.  This only matters for -O0 because otherwise
171  * the non-working version gets optimized away.
172  *
173  * Use an expression-statement instead of a conditional expression
174  * because gcc-2.6.0 would promote the operands of the conditional
175  * and produce poor code for "if ((inb(var) & const1) == const2)".
176  *
177  * The unnecessary test `(port) < 0x10000' is to generate a warning if
178  * the `port' has type u_short or smaller.  Such types are pessimal.
179  * This actually only works for signed types.  The range check is
180  * careful to avoid generating warnings.
181  */
182 #define	inb(port) __extension__ ({					\
183 	u_char	_data;							\
184 	if (__builtin_constant_p(port) && ((port) & 0xffff) < 0x100	\
185 	    && (port) < 0x10000)					\
186 		_data = inbc(port);					\
187 	else								\
188 		_data = inbv(port);					\
189 	_data; })
190 
191 #define	outb(port, data) (						\
192 	__builtin_constant_p(port) && ((port) & 0xffff) < 0x100		\
193 	&& (port) < 0x10000						\
194 	? outbc(port, data) : outbv(port, data))
195 
196 static __inline u_char
197 inbc(u_int port)
198 {
199 	u_char	data;
200 
201 	__asm __volatile("inb %1,%0" : "=a" (data) : "id" ((u_short)(port)));
202 	return (data);
203 }
204 
205 static __inline void
206 outbc(u_int port, u_char data)
207 {
208 	__asm __volatile("outb %0,%1" : : "a" (data), "id" ((u_short)(port)));
209 }
210 
211 #endif /* __GNUCLIKE_BUILTIN_CONSTANT_P */
212 
213 static __inline u_char
214 inbv(u_int port)
215 {
216 	u_char	data;
217 	/*
218 	 * We use %%dx and not %1 here because i/o is done at %dx and not at
219 	 * %edx, while gcc generates inferior code (movw instead of movl)
220 	 * if we tell it to load (u_short) port.
221 	 */
222 	__asm __volatile("inb %%dx,%0" : "=a" (data) : "d" (port));
223 	return (data);
224 }
225 
226 static __inline u_int
227 inl(u_int port)
228 {
229 	u_int	data;
230 
231 	__asm __volatile("inl %%dx,%0" : "=a" (data) : "d" (port));
232 	return (data);
233 }
234 
235 static __inline void
236 insb(u_int port, void *addr, size_t cnt)
237 {
238 	__asm __volatile("cld; rep; insb"
239 			 : "+D" (addr), "+c" (cnt)
240 			 : "d" (port)
241 			 : "memory");
242 }
243 
244 static __inline void
245 insw(u_int port, void *addr, size_t cnt)
246 {
247 	__asm __volatile("cld; rep; insw"
248 			 : "+D" (addr), "+c" (cnt)
249 			 : "d" (port)
250 			 : "memory");
251 }
252 
253 static __inline void
254 insl(u_int port, void *addr, size_t cnt)
255 {
256 	__asm __volatile("cld; rep; insl"
257 			 : "+D" (addr), "+c" (cnt)
258 			 : "d" (port)
259 			 : "memory");
260 }
261 
262 static __inline void
263 invd(void)
264 {
265 	__asm __volatile("invd");
266 }
267 
268 static __inline u_short
269 inw(u_int port)
270 {
271 	u_short	data;
272 
273 	__asm __volatile("inw %%dx,%0" : "=a" (data) : "d" (port));
274 	return (data);
275 }
276 
277 static __inline void
278 outbv(u_int port, u_char data)
279 {
280 	u_char	al;
281 	/*
282 	 * Use an unnecessary assignment to help gcc's register allocator.
283 	 * This make a large difference for gcc-1.40 and a tiny difference
284 	 * for gcc-2.6.0.  For gcc-1.40, al had to be ``asm("ax")'' for
285 	 * best results.  gcc-2.6.0 can't handle this.
286 	 */
287 	al = data;
288 	__asm __volatile("outb %0,%%dx" : : "a" (al), "d" (port));
289 }
290 
291 static __inline void
292 outl(u_int port, u_int data)
293 {
294 	/*
295 	 * outl() and outw() aren't used much so we haven't looked at
296 	 * possible micro-optimizations such as the unnecessary
297 	 * assignment for them.
298 	 */
299 	__asm __volatile("outl %0,%%dx" : : "a" (data), "d" (port));
300 }
301 
302 static __inline void
303 outsb(u_int port, const void *addr, size_t cnt)
304 {
305 	__asm __volatile("cld; rep; outsb"
306 			 : "+S" (addr), "+c" (cnt)
307 			 : "d" (port));
308 }
309 
310 static __inline void
311 outsw(u_int port, const void *addr, size_t cnt)
312 {
313 	__asm __volatile("cld; rep; outsw"
314 			 : "+S" (addr), "+c" (cnt)
315 			 : "d" (port));
316 }
317 
318 static __inline void
319 outsl(u_int port, const void *addr, size_t cnt)
320 {
321 	__asm __volatile("cld; rep; outsl"
322 			 : "+S" (addr), "+c" (cnt)
323 			 : "d" (port));
324 }
325 
326 static __inline void
327 outw(u_int port, u_short data)
328 {
329 	__asm __volatile("outw %0,%%dx" : : "a" (data), "d" (port));
330 }
331 
332 static __inline void
333 ia32_pause(void)
334 {
335 	__asm __volatile("pause");
336 }
337 
338 static __inline u_long
339 read_rflags(void)
340 {
341 	u_long	rf;
342 
343 	__asm __volatile("pushfq; popq %0" : "=r" (rf));
344 	return (rf);
345 }
346 
347 static __inline u_int64_t
348 rdmsr(u_int msr)
349 {
350 	u_int32_t low, high;
351 
352 	__asm __volatile("rdmsr" : "=a" (low), "=d" (high) : "c" (msr));
353 	return (low | ((u_int64_t)high << 32));
354 }
355 
356 static __inline u_int64_t
357 rdpmc(u_int pmc)
358 {
359 	u_int32_t low, high;
360 
361 	__asm __volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (pmc));
362 	return (low | ((u_int64_t)high << 32));
363 }
364 
365 static __inline u_int64_t
366 rdtsc(void)
367 {
368 	u_int32_t low, high;
369 
370 	__asm __volatile("rdtsc" : "=a" (low), "=d" (high));
371 	return (low | ((u_int64_t)high << 32));
372 }
373 
374 static __inline void
375 wbinvd(void)
376 {
377 	__asm __volatile("wbinvd");
378 }
379 
380 static __inline void
381 write_rflags(u_long rf)
382 {
383 	__asm __volatile("pushq %0;  popfq" : : "r" (rf));
384 }
385 
386 static __inline void
387 wrmsr(u_int msr, u_int64_t newval)
388 {
389 	u_int32_t low, high;
390 
391 	low = newval;
392 	high = newval >> 32;
393 	__asm __volatile("wrmsr" : : "a" (low), "d" (high), "c" (msr));
394 }
395 
396 static __inline void
397 load_cr0(u_long data)
398 {
399 
400 	__asm __volatile("movq %0,%%cr0" : : "r" (data));
401 }
402 
403 static __inline u_long
404 rcr0(void)
405 {
406 	u_long	data;
407 
408 	__asm __volatile("movq %%cr0,%0" : "=r" (data));
409 	return (data);
410 }
411 
412 static __inline u_long
413 rcr2(void)
414 {
415 	u_long	data;
416 
417 	__asm __volatile("movq %%cr2,%0" : "=r" (data));
418 	return (data);
419 }
420 
421 static __inline void
422 load_cr3(u_long data)
423 {
424 
425 	__asm __volatile("movq %0,%%cr3" : : "r" (data) : "memory");
426 }
427 
428 static __inline u_long
429 rcr3(void)
430 {
431 	u_long	data;
432 
433 	__asm __volatile("movq %%cr3,%0" : "=r" (data));
434 	return (data);
435 }
436 
437 static __inline void
438 load_cr4(u_long data)
439 {
440 	__asm __volatile("movq %0,%%cr4" : : "r" (data));
441 }
442 
443 static __inline u_long
444 rcr4(void)
445 {
446 	u_long	data;
447 
448 	__asm __volatile("movq %%cr4,%0" : "=r" (data));
449 	return (data);
450 }
451 
452 /*
453  * Global TLB flush (except for thise for pages marked PG_G)
454  */
455 static __inline void
456 invltlb(void)
457 {
458 
459 	load_cr3(rcr3());
460 }
461 
462 /*
463  * TLB flush for an individual page (even if it has PG_G).
464  * Only works on 486+ CPUs (i386 does not have PG_G).
465  */
466 static __inline void
467 invlpg(u_long addr)
468 {
469 
470 	__asm __volatile("invlpg %0" : : "m" (*(char *)addr) : "memory");
471 }
472 
473 static __inline u_int
474 rfs(void)
475 {
476 	u_int sel;
477 	__asm __volatile("movl %%fs,%0" : "=rm" (sel));
478 	return (sel);
479 }
480 
481 static __inline u_int
482 rgs(void)
483 {
484 	u_int sel;
485 	__asm __volatile("movl %%gs,%0" : "=rm" (sel));
486 	return (sel);
487 }
488 
489 static __inline u_int
490 rss(void)
491 {
492 	u_int sel;
493 	__asm __volatile("movl %%ss,%0" : "=rm" (sel));
494 	return (sel);
495 }
496 
497 static __inline void
498 load_ds(u_int sel)
499 {
500 	__asm __volatile("movl %0,%%ds" : : "rm" (sel));
501 }
502 
503 static __inline void
504 load_es(u_int sel)
505 {
506 	__asm __volatile("movl %0,%%es" : : "rm" (sel));
507 }
508 
509 #ifdef _KERNEL
510 /* This is defined in <machine/specialreg.h> but is too painful to get to */
511 #ifndef	MSR_FSBASE
512 #define	MSR_FSBASE	0xc0000100
513 #endif
514 static __inline void
515 load_fs(u_int sel)
516 {
517 	register u_int32_t fsbase __asm("ecx");
518 
519 	/* Preserve the fsbase value across the selector load */
520 	fsbase = MSR_FSBASE;
521         __asm __volatile("rdmsr; movl %0,%%fs; wrmsr"
522             : : "rm" (sel), "c" (fsbase) : "eax", "edx");
523 }
524 
525 #ifndef	MSR_GSBASE
526 #define	MSR_GSBASE	0xc0000101
527 #endif
528 static __inline void
529 load_gs(u_int sel)
530 {
531 	register u_int32_t gsbase __asm("ecx");
532 
533 	/*
534 	 * Preserve the gsbase value across the selector load.
535 	 * Note that we have to disable interrupts because the gsbase
536 	 * being trashed happens to be the kernel gsbase at the time.
537 	 */
538 	gsbase = MSR_GSBASE;
539         __asm __volatile("pushfq; cli; rdmsr; movl %0,%%gs; wrmsr; popfq"
540             : : "rm" (sel), "c" (gsbase) : "eax", "edx");
541 }
542 #else
543 /* Usable by userland */
544 static __inline void
545 load_fs(u_int sel)
546 {
547 	__asm __volatile("movl %0,%%fs" : : "rm" (sel));
548 }
549 
550 static __inline void
551 load_gs(u_int sel)
552 {
553 	__asm __volatile("movl %0,%%gs" : : "rm" (sel));
554 }
555 #endif
556 
557 static __inline void
558 lidt(struct region_descriptor *addr)
559 {
560 	__asm __volatile("lidt (%0)" : : "r" (addr));
561 }
562 
563 static __inline void
564 lldt(u_short sel)
565 {
566 	__asm __volatile("lldt %0" : : "r" (sel));
567 }
568 
569 static __inline void
570 ltr(u_short sel)
571 {
572 	__asm __volatile("ltr %0" : : "r" (sel));
573 }
574 
575 static __inline u_int64_t
576 rdr0(void)
577 {
578 	u_int64_t data;
579 	__asm __volatile("movq %%dr0,%0" : "=r" (data));
580 	return (data);
581 }
582 
583 static __inline void
584 load_dr0(u_int64_t dr0)
585 {
586 	__asm __volatile("movq %0,%%dr0" : : "r" (dr0));
587 }
588 
589 static __inline u_int64_t
590 rdr1(void)
591 {
592 	u_int64_t data;
593 	__asm __volatile("movq %%dr1,%0" : "=r" (data));
594 	return (data);
595 }
596 
597 static __inline void
598 load_dr1(u_int64_t dr1)
599 {
600 	__asm __volatile("movq %0,%%dr1" : : "r" (dr1));
601 }
602 
603 static __inline u_int64_t
604 rdr2(void)
605 {
606 	u_int64_t data;
607 	__asm __volatile("movq %%dr2,%0" : "=r" (data));
608 	return (data);
609 }
610 
611 static __inline void
612 load_dr2(u_int64_t dr2)
613 {
614 	__asm __volatile("movq %0,%%dr2" : : "r" (dr2));
615 }
616 
617 static __inline u_int64_t
618 rdr3(void)
619 {
620 	u_int64_t data;
621 	__asm __volatile("movq %%dr3,%0" : "=r" (data));
622 	return (data);
623 }
624 
625 static __inline void
626 load_dr3(u_int64_t dr3)
627 {
628 	__asm __volatile("movq %0,%%dr3" : : "r" (dr3));
629 }
630 
631 static __inline u_int64_t
632 rdr4(void)
633 {
634 	u_int64_t data;
635 	__asm __volatile("movq %%dr4,%0" : "=r" (data));
636 	return (data);
637 }
638 
639 static __inline void
640 load_dr4(u_int64_t dr4)
641 {
642 	__asm __volatile("movq %0,%%dr4" : : "r" (dr4));
643 }
644 
645 static __inline u_int64_t
646 rdr5(void)
647 {
648 	u_int64_t data;
649 	__asm __volatile("movq %%dr5,%0" : "=r" (data));
650 	return (data);
651 }
652 
653 static __inline void
654 load_dr5(u_int64_t dr5)
655 {
656 	__asm __volatile("movq %0,%%dr5" : : "r" (dr5));
657 }
658 
659 static __inline u_int64_t
660 rdr6(void)
661 {
662 	u_int64_t data;
663 	__asm __volatile("movq %%dr6,%0" : "=r" (data));
664 	return (data);
665 }
666 
667 static __inline void
668 load_dr6(u_int64_t dr6)
669 {
670 	__asm __volatile("movq %0,%%dr6" : : "r" (dr6));
671 }
672 
673 static __inline u_int64_t
674 rdr7(void)
675 {
676 	u_int64_t data;
677 	__asm __volatile("movq %%dr7,%0" : "=r" (data));
678 	return (data);
679 }
680 
681 static __inline void
682 load_dr7(u_int64_t dr7)
683 {
684 	__asm __volatile("movq %0,%%dr7" : : "r" (dr7));
685 }
686 
687 static __inline register_t
688 intr_disable(void)
689 {
690 	register_t rflags;
691 
692 	rflags = read_rflags();
693 	disable_intr();
694 	return (rflags);
695 }
696 
697 static __inline void
698 intr_restore(register_t rflags)
699 {
700 	write_rflags(rflags);
701 }
702 
703 #else /* !(__GNUCLIKE_ASM && __CC_SUPPORTS___INLINE) */
704 
705 int	breakpoint(void);
706 u_int	bsfl(u_int mask);
707 u_int	bsrl(u_int mask);
708 void	disable_intr(void);
709 void	do_cpuid(u_int ax, u_int *p);
710 void	enable_intr(void);
711 void	halt(void);
712 void	ia32_pause(void);
713 u_char	inb(u_int port);
714 u_int	inl(u_int port);
715 void	insb(u_int port, void *addr, size_t cnt);
716 void	insl(u_int port, void *addr, size_t cnt);
717 void	insw(u_int port, void *addr, size_t cnt);
718 register_t	intr_disable(void);
719 void	intr_restore(register_t rf);
720 void	invd(void);
721 void	invlpg(u_int addr);
722 void	invltlb(void);
723 u_short	inw(u_int port);
724 void	lidt(struct region_descriptor *addr);
725 void	lldt(u_short sel);
726 void	load_cr0(u_long cr0);
727 void	load_cr3(u_long cr3);
728 void	load_cr4(u_long cr4);
729 void	load_dr0(u_int64_t dr0);
730 void	load_dr1(u_int64_t dr1);
731 void	load_dr2(u_int64_t dr2);
732 void	load_dr3(u_int64_t dr3);
733 void	load_dr4(u_int64_t dr4);
734 void	load_dr5(u_int64_t dr5);
735 void	load_dr6(u_int64_t dr6);
736 void	load_dr7(u_int64_t dr7);
737 void	load_fs(u_int sel);
738 void	load_gs(u_int sel);
739 void	ltr(u_short sel);
740 void	outb(u_int port, u_char data);
741 void	outl(u_int port, u_int data);
742 void	outsb(u_int port, const void *addr, size_t cnt);
743 void	outsl(u_int port, const void *addr, size_t cnt);
744 void	outsw(u_int port, const void *addr, size_t cnt);
745 void	outw(u_int port, u_short data);
746 u_long	rcr0(void);
747 u_long	rcr2(void);
748 u_long	rcr3(void);
749 u_long	rcr4(void);
750 u_int64_t rdmsr(u_int msr);
751 u_int64_t rdpmc(u_int pmc);
752 u_int64_t rdr0(void);
753 u_int64_t rdr1(void);
754 u_int64_t rdr2(void);
755 u_int64_t rdr3(void);
756 u_int64_t rdr4(void);
757 u_int64_t rdr5(void);
758 u_int64_t rdr6(void);
759 u_int64_t rdr7(void);
760 u_int64_t rdtsc(void);
761 u_int	read_rflags(void);
762 u_int	rfs(void);
763 u_int	rgs(void);
764 void	wbinvd(void);
765 void	write_rflags(u_int rf);
766 void	wrmsr(u_int msr, u_int64_t newval);
767 
768 #endif	/* __GNUCLIKE_ASM && __CC_SUPPORTS___INLINE */
769 
770 void	reset_dbregs(void);
771 
772 #endif /* !_MACHINE_CPUFUNC_H_ */
773