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
46struct region_descriptor;
47
48#define readb(va)	(*(volatile uint8_t *) (va))
49#define readw(va)	(*(volatile uint16_t *) (va))
50#define readl(va)	(*(volatile uint32_t *) (va))
51#define readq(va)	(*(volatile uint64_t *) (va))
52
53#define writeb(va, d)	(*(volatile uint8_t *) (va) = (d))
54#define writew(va, d)	(*(volatile uint16_t *) (va) = (d))
55#define writel(va, d)	(*(volatile uint32_t *) (va) = (d))
56#define writeq(va, d)	(*(volatile uint64_t *) (va) = (d))
57
58#if defined(__GNUCLIKE_ASM) && defined(__CC_SUPPORTS___INLINE)
59
60static __inline void
61breakpoint(void)
62{
63	__asm __volatile("int $3");
64}
65
66static __inline u_int
67bsfl(u_int mask)
68{
69	u_int	result;
70
71	__asm __volatile("bsfl %1,%0" : "=r" (result) : "rm" (mask));
72	return (result);
73}
74
75static __inline u_long
76bsfq(u_long mask)
77{
78	u_long	result;
79
80	__asm __volatile("bsfq %1,%0" : "=r" (result) : "rm" (mask));
81	return (result);
82}
83
84static __inline u_int
85bsrl(u_int mask)
86{
87	u_int	result;
88
89	__asm __volatile("bsrl %1,%0" : "=r" (result) : "rm" (mask));
90	return (result);
91}
92
93static __inline u_long
94bsrq(u_long mask)
95{
96	u_long	result;
97
98	__asm __volatile("bsrq %1,%0" : "=r" (result) : "rm" (mask));
99	return (result);
100}
101
102static __inline void
103clflush(u_long addr)
104{
105
106	__asm __volatile("clflush %0" : : "m" (*(char *)addr));
107}
108
109static __inline void
110clts(void)
111{
112
113	__asm __volatile("clts");
114}
115
116static __inline void
117disable_intr(void)
118{
119	__asm __volatile("cli" : : : "memory");
120}
121
122static __inline void
123do_cpuid(u_int ax, u_int *p)
124{
125	__asm __volatile("cpuid"
126			 : "=a" (p[0]), "=b" (p[1]), "=c" (p[2]), "=d" (p[3])
127			 :  "0" (ax));
128}
129
130static __inline void
131cpuid_count(u_int ax, u_int cx, u_int *p)
132{
133	__asm __volatile("cpuid"
134			 : "=a" (p[0]), "=b" (p[1]), "=c" (p[2]), "=d" (p[3])
135			 :  "0" (ax), "c" (cx));
136}
137
138static __inline void
139enable_intr(void)
140{
141	__asm __volatile("sti");
142}
143
144#ifdef _KERNEL
145
146#define	HAVE_INLINE_FFS
147#define        ffs(x)  __builtin_ffs(x)
148
149#define	HAVE_INLINE_FFSL
150
151static __inline int
152ffsl(long mask)
153{
154	return (mask == 0 ? mask : (int)bsfq((u_long)mask) + 1);
155}
156
157#define	HAVE_INLINE_FLS
158
159static __inline int
160fls(int mask)
161{
162	return (mask == 0 ? mask : (int)bsrl((u_int)mask) + 1);
163}
164
165#define	HAVE_INLINE_FLSL
166
167static __inline int
168flsl(long mask)
169{
170	return (mask == 0 ? mask : (int)bsrq((u_long)mask) + 1);
171}
172
173#endif /* _KERNEL */
174
175static __inline void
176halt(void)
177{
178	__asm __volatile("hlt");
179}
180
181static __inline u_char
182inb(u_int port)
183{
184	u_char	data;
185
186	__asm __volatile("inb %w1, %0" : "=a" (data) : "Nd" (port));
187	return (data);
188}
189
190static __inline u_int
191inl(u_int port)
192{
193	u_int	data;
194
195	__asm __volatile("inl %w1, %0" : "=a" (data) : "Nd" (port));
196	return (data);
197}
198
199static __inline void
200insb(u_int port, void *addr, size_t count)
201{
202	__asm __volatile("cld; rep; insb"
203			 : "+D" (addr), "+c" (count)
204			 : "d" (port)
205			 : "memory");
206}
207
208static __inline void
209insw(u_int port, void *addr, size_t count)
210{
211	__asm __volatile("cld; rep; insw"
212			 : "+D" (addr), "+c" (count)
213			 : "d" (port)
214			 : "memory");
215}
216
217static __inline void
218insl(u_int port, void *addr, size_t count)
219{
220	__asm __volatile("cld; rep; insl"
221			 : "+D" (addr), "+c" (count)
222			 : "d" (port)
223			 : "memory");
224}
225
226static __inline void
227invd(void)
228{
229	__asm __volatile("invd");
230}
231
232static __inline u_short
233inw(u_int port)
234{
235	u_short	data;
236
237	__asm __volatile("inw %w1, %0" : "=a" (data) : "Nd" (port));
238	return (data);
239}
240
241static __inline void
242outb(u_int port, u_char data)
243{
244	__asm __volatile("outb %0, %w1" : : "a" (data), "Nd" (port));
245}
246
247static __inline void
248outl(u_int port, u_int data)
249{
250	__asm __volatile("outl %0, %w1" : : "a" (data), "Nd" (port));
251}
252
253static __inline void
254outsb(u_int port, const void *addr, size_t count)
255{
256	__asm __volatile("cld; rep; outsb"
257			 : "+S" (addr), "+c" (count)
258			 : "d" (port));
259}
260
261static __inline void
262outsw(u_int port, const void *addr, size_t count)
263{
264	__asm __volatile("cld; rep; outsw"
265			 : "+S" (addr), "+c" (count)
266			 : "d" (port));
267}
268
269static __inline void
270outsl(u_int port, const void *addr, size_t count)
271{
272	__asm __volatile("cld; rep; outsl"
273			 : "+S" (addr), "+c" (count)
274			 : "d" (port));
275}
276
277static __inline void
278outw(u_int port, u_short data)
279{
280	__asm __volatile("outw %0, %w1" : : "a" (data), "Nd" (port));
281}
282
283static __inline u_long
284popcntq(u_long mask)
285{
286	u_long result;
287
288	__asm __volatile("popcntq %1,%0" : "=r" (result) : "rm" (mask));
289	return (result);
290}
291
292static __inline void
293lfence(void)
294{
295
296	__asm __volatile("lfence" : : : "memory");
297}
298
299static __inline void
300mfence(void)
301{
302
303	__asm __volatile("mfence" : : : "memory");
304}
305
306static __inline void
307ia32_pause(void)
308{
309	__asm __volatile("pause");
310}
311
312static __inline u_long
313read_rflags(void)
314{
315	u_long	rf;
316
317	__asm __volatile("pushfq; popq %0" : "=r" (rf));
318	return (rf);
319}
320
321static __inline uint64_t
322rdmsr(u_int msr)
323{
324	uint32_t low, high;
325
326	__asm __volatile("rdmsr" : "=a" (low), "=d" (high) : "c" (msr));
327	return (low | ((uint64_t)high << 32));
328}
329
330static __inline uint64_t
331rdpmc(u_int pmc)
332{
333	uint32_t low, high;
334
335	__asm __volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (pmc));
336	return (low | ((uint64_t)high << 32));
337}
338
339static __inline uint64_t
340rdtsc(void)
341{
342	uint32_t low, high;
343
344	__asm __volatile("rdtsc" : "=a" (low), "=d" (high));
345	return (low | ((uint64_t)high << 32));
346}
347
348static __inline uint32_t
349rdtsc32(void)
350{
351	uint32_t rv;
352
353	__asm __volatile("rdtsc" : "=a" (rv) : : "edx");
354	return (rv);
355}
356
357static __inline void
358wbinvd(void)
359{
360	__asm __volatile("wbinvd");
361}
362
363static __inline void
364write_rflags(u_long rf)
365{
366	__asm __volatile("pushq %0;  popfq" : : "r" (rf));
367}
368
369static __inline void
370wrmsr(u_int msr, uint64_t newval)
371{
372	uint32_t low, high;
373
374	low = newval;
375	high = newval >> 32;
376	__asm __volatile("wrmsr" : : "a" (low), "d" (high), "c" (msr));
377}
378
379static __inline void
380load_cr0(u_long data)
381{
382
383	__asm __volatile("movq %0,%%cr0" : : "r" (data));
384}
385
386static __inline u_long
387rcr0(void)
388{
389	u_long	data;
390
391	__asm __volatile("movq %%cr0,%0" : "=r" (data));
392	return (data);
393}
394
395static __inline u_long
396rcr2(void)
397{
398	u_long	data;
399
400	__asm __volatile("movq %%cr2,%0" : "=r" (data));
401	return (data);
402}
403
404static __inline void
405load_cr3(u_long data)
406{
407
408	__asm __volatile("movq %0,%%cr3" : : "r" (data) : "memory");
409}
410
411static __inline u_long
412rcr3(void)
413{
414	u_long	data;
415
416	__asm __volatile("movq %%cr3,%0" : "=r" (data));
417	return (data);
418}
419
420static __inline void
421load_cr4(u_long data)
422{
423	__asm __volatile("movq %0,%%cr4" : : "r" (data));
424}
425
426static __inline u_long
427rcr4(void)
428{
429	u_long	data;
430
431	__asm __volatile("movq %%cr4,%0" : "=r" (data));
432	return (data);
433}
434
435static __inline u_long
436rxcr(u_int reg)
437{
438	u_int low, high;
439
440	__asm __volatile("xgetbv" : "=a" (low), "=d" (high) : "c" (reg));
441	return (low | ((uint64_t)high << 32));
442}
443
444static __inline void
445load_xcr(u_int reg, u_long val)
446{
447	u_int low, high;
448
449	low = val;
450	high = val >> 32;
451	__asm __volatile("xsetbv" : : "c" (reg), "a" (low), "d" (high));
452}
453
454/*
455 * Global TLB flush (except for thise for pages marked PG_G)
456 */
457static __inline void
458invltlb(void)
459{
460
461	load_cr3(rcr3());
462}
463
464#ifndef CR4_PGE
465#define	CR4_PGE	0x00000080	/* Page global enable */
466#endif
467
468/*
469 * Perform the guaranteed invalidation of all TLB entries.  This
470 * includes the global entries, and entries in all PCIDs, not only the
471 * current context.  The function works both on non-PCID CPUs and CPUs
472 * with the PCID turned off or on.  See IA-32 SDM Vol. 3a 4.10.4.1
473 * Operations that Invalidate TLBs and Paging-Structure Caches.
474 */
475static __inline void
476invltlb_globpcid(void)
477{
478	uint64_t cr4;
479
480	cr4 = rcr4();
481	load_cr4(cr4 & ~CR4_PGE);
482	/*
483	 * Although preemption at this point could be detrimental to
484	 * performance, it would not lead to an error.  PG_G is simply
485	 * ignored if CR4.PGE is clear.  Moreover, in case this block
486	 * is re-entered, the load_cr4() either above or below will
487	 * modify CR4.PGE flushing the TLB.
488	 */
489	load_cr4(cr4 | CR4_PGE);
490}
491
492/*
493 * TLB flush for an individual page (even if it has PG_G).
494 * Only works on 486+ CPUs (i386 does not have PG_G).
495 */
496static __inline void
497invlpg(u_long addr)
498{
499
500	__asm __volatile("invlpg %0" : : "m" (*(char *)addr) : "memory");
501}
502
503#define	INVPCID_ADDR	0
504#define	INVPCID_CTX	1
505#define	INVPCID_CTXGLOB	2
506#define	INVPCID_ALLCTX	3
507
508struct invpcid_descr {
509	uint64_t	pcid:12 __packed;
510	uint64_t	pad:52 __packed;
511	uint64_t	addr;
512} __packed;
513
514static __inline void
515invpcid(struct invpcid_descr *d, int type)
516{
517
518	/* invpcid (%rdx),%rax */
519	__asm __volatile(".byte 0x66,0x0f,0x38,0x82,0x02"
520	    : : "d" (d), "a" ((u_long)type) : "memory");
521}
522
523static __inline u_short
524rfs(void)
525{
526	u_short sel;
527	__asm __volatile("movw %%fs,%0" : "=rm" (sel));
528	return (sel);
529}
530
531static __inline u_short
532rgs(void)
533{
534	u_short sel;
535	__asm __volatile("movw %%gs,%0" : "=rm" (sel));
536	return (sel);
537}
538
539static __inline u_short
540rss(void)
541{
542	u_short sel;
543	__asm __volatile("movw %%ss,%0" : "=rm" (sel));
544	return (sel);
545}
546
547static __inline void
548load_ds(u_short sel)
549{
550	__asm __volatile("movw %0,%%ds" : : "rm" (sel));
551}
552
553static __inline void
554load_es(u_short sel)
555{
556	__asm __volatile("movw %0,%%es" : : "rm" (sel));
557}
558
559static __inline void
560cpu_monitor(const void *addr, u_long extensions, u_int hints)
561{
562
563	__asm __volatile("monitor"
564	    : : "a" (addr), "c" (extensions), "d" (hints));
565}
566
567static __inline void
568cpu_mwait(u_long extensions, u_int hints)
569{
570
571	__asm __volatile("mwait" : : "a" (hints), "c" (extensions));
572}
573
574#ifdef _KERNEL
575/* This is defined in <machine/specialreg.h> but is too painful to get to */
576#ifndef	MSR_FSBASE
577#define	MSR_FSBASE	0xc0000100
578#endif
579static __inline void
580load_fs(u_short sel)
581{
582	/* Preserve the fsbase value across the selector load */
583	__asm __volatile("rdmsr; movw %0,%%fs; wrmsr"
584	    : : "rm" (sel), "c" (MSR_FSBASE) : "eax", "edx");
585}
586
587#ifndef	MSR_GSBASE
588#define	MSR_GSBASE	0xc0000101
589#endif
590static __inline void
591load_gs(u_short sel)
592{
593	/*
594	 * Preserve the gsbase value across the selector load.
595	 * Note that we have to disable interrupts because the gsbase
596	 * being trashed happens to be the kernel gsbase at the time.
597	 */
598	__asm __volatile("pushfq; cli; rdmsr; movw %0,%%gs; wrmsr; popfq"
599	    : : "rm" (sel), "c" (MSR_GSBASE) : "eax", "edx");
600}
601#else
602/* Usable by userland */
603static __inline void
604load_fs(u_short sel)
605{
606	__asm __volatile("movw %0,%%fs" : : "rm" (sel));
607}
608
609static __inline void
610load_gs(u_short sel)
611{
612	__asm __volatile("movw %0,%%gs" : : "rm" (sel));
613}
614#endif
615
616static __inline void
617lidt(struct region_descriptor *addr)
618{
619	__asm __volatile("lidt (%0)" : : "r" (addr));
620}
621
622static __inline void
623lldt(u_short sel)
624{
625	__asm __volatile("lldt %0" : : "r" (sel));
626}
627
628static __inline void
629ltr(u_short sel)
630{
631	__asm __volatile("ltr %0" : : "r" (sel));
632}
633
634static __inline uint64_t
635rdr0(void)
636{
637	uint64_t data;
638	__asm __volatile("movq %%dr0,%0" : "=r" (data));
639	return (data);
640}
641
642static __inline void
643load_dr0(uint64_t dr0)
644{
645	__asm __volatile("movq %0,%%dr0" : : "r" (dr0));
646}
647
648static __inline uint64_t
649rdr1(void)
650{
651	uint64_t data;
652	__asm __volatile("movq %%dr1,%0" : "=r" (data));
653	return (data);
654}
655
656static __inline void
657load_dr1(uint64_t dr1)
658{
659	__asm __volatile("movq %0,%%dr1" : : "r" (dr1));
660}
661
662static __inline uint64_t
663rdr2(void)
664{
665	uint64_t data;
666	__asm __volatile("movq %%dr2,%0" : "=r" (data));
667	return (data);
668}
669
670static __inline void
671load_dr2(uint64_t dr2)
672{
673	__asm __volatile("movq %0,%%dr2" : : "r" (dr2));
674}
675
676static __inline uint64_t
677rdr3(void)
678{
679	uint64_t data;
680	__asm __volatile("movq %%dr3,%0" : "=r" (data));
681	return (data);
682}
683
684static __inline void
685load_dr3(uint64_t dr3)
686{
687	__asm __volatile("movq %0,%%dr3" : : "r" (dr3));
688}
689
690static __inline uint64_t
691rdr4(void)
692{
693	uint64_t data;
694	__asm __volatile("movq %%dr4,%0" : "=r" (data));
695	return (data);
696}
697
698static __inline void
699load_dr4(uint64_t dr4)
700{
701	__asm __volatile("movq %0,%%dr4" : : "r" (dr4));
702}
703
704static __inline uint64_t
705rdr5(void)
706{
707	uint64_t data;
708	__asm __volatile("movq %%dr5,%0" : "=r" (data));
709	return (data);
710}
711
712static __inline void
713load_dr5(uint64_t dr5)
714{
715	__asm __volatile("movq %0,%%dr5" : : "r" (dr5));
716}
717
718static __inline uint64_t
719rdr6(void)
720{
721	uint64_t data;
722	__asm __volatile("movq %%dr6,%0" : "=r" (data));
723	return (data);
724}
725
726static __inline void
727load_dr6(uint64_t dr6)
728{
729	__asm __volatile("movq %0,%%dr6" : : "r" (dr6));
730}
731
732static __inline uint64_t
733rdr7(void)
734{
735	uint64_t data;
736	__asm __volatile("movq %%dr7,%0" : "=r" (data));
737	return (data);
738}
739
740static __inline void
741load_dr7(uint64_t dr7)
742{
743	__asm __volatile("movq %0,%%dr7" : : "r" (dr7));
744}
745
746static __inline register_t
747intr_disable(void)
748{
749	register_t rflags;
750
751	rflags = read_rflags();
752	disable_intr();
753	return (rflags);
754}
755
756static __inline void
757intr_restore(register_t rflags)
758{
759	write_rflags(rflags);
760}
761
762#else /* !(__GNUCLIKE_ASM && __CC_SUPPORTS___INLINE) */
763
764int	breakpoint(void);
765u_int	bsfl(u_int mask);
766u_int	bsrl(u_int mask);
767void	clflush(u_long addr);
768void	clts(void);
769void	cpuid_count(u_int ax, u_int cx, u_int *p);
770void	disable_intr(void);
771void	do_cpuid(u_int ax, u_int *p);
772void	enable_intr(void);
773void	halt(void);
774void	ia32_pause(void);
775u_char	inb(u_int port);
776u_int	inl(u_int port);
777void	insb(u_int port, void *addr, size_t count);
778void	insl(u_int port, void *addr, size_t count);
779void	insw(u_int port, void *addr, size_t count);
780register_t	intr_disable(void);
781void	intr_restore(register_t rf);
782void	invd(void);
783void	invlpg(u_int addr);
784void	invltlb(void);
785u_short	inw(u_int port);
786void	lidt(struct region_descriptor *addr);
787void	lldt(u_short sel);
788void	load_cr0(u_long cr0);
789void	load_cr3(u_long cr3);
790void	load_cr4(u_long cr4);
791void	load_dr0(uint64_t dr0);
792void	load_dr1(uint64_t dr1);
793void	load_dr2(uint64_t dr2);
794void	load_dr3(uint64_t dr3);
795void	load_dr4(uint64_t dr4);
796void	load_dr5(uint64_t dr5);
797void	load_dr6(uint64_t dr6);
798void	load_dr7(uint64_t dr7);
799void	load_fs(u_short sel);
800void	load_gs(u_short sel);
801void	ltr(u_short sel);
802void	outb(u_int port, u_char data);
803void	outl(u_int port, u_int data);
804void	outsb(u_int port, const void *addr, size_t count);
805void	outsl(u_int port, const void *addr, size_t count);
806void	outsw(u_int port, const void *addr, size_t count);
807void	outw(u_int port, u_short data);
808u_long	rcr0(void);
809u_long	rcr2(void);
810u_long	rcr3(void);
811u_long	rcr4(void);
812uint64_t rdmsr(u_int msr);
813uint64_t rdpmc(u_int pmc);
814uint64_t rdr0(void);
815uint64_t rdr1(void);
816uint64_t rdr2(void);
817uint64_t rdr3(void);
818uint64_t rdr4(void);
819uint64_t rdr5(void);
820uint64_t rdr6(void);
821uint64_t rdr7(void);
822uint64_t rdtsc(void);
823u_long	read_rflags(void);
824u_int	rfs(void);
825u_int	rgs(void);
826void	wbinvd(void);
827void	write_rflags(u_int rf);
828void	wrmsr(u_int msr, uint64_t newval);
829
830#endif	/* __GNUCLIKE_ASM && __CC_SUPPORTS___INLINE */
831
832void	reset_dbregs(void);
833
834#ifdef _KERNEL
835int	rdmsr_safe(u_int msr, uint64_t *val);
836int	wrmsr_safe(u_int msr, uint64_t newval);
837#endif
838
839#endif /* !_MACHINE_CPUFUNC_H_ */
840