2 * Copyright (c) 2003 Peter Wemm.
3 * Copyright (c) 1993 The Regents of the University of California.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
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
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.
39 #ifndef _MACHINE_CPUFUNC_H_
40 #define _MACHINE_CPUFUNC_H_
43 #error this file needs sys/cdefs.h as a prerequisite
46 struct region_descriptor;
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))
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))
58 #if defined(__GNUCLIKE_ASM) && defined(__CC_SUPPORTS___INLINE)
63 __asm __volatile("int $3");
71 __asm __volatile("bsfl %1,%0" : "=r" (result) : "rm" (mask));
75 static __inline u_long
80 __asm __volatile("bsfq %1,%0" : "=r" (result) : "rm" (mask));
89 __asm __volatile("bsrl %1,%0" : "=r" (result) : "rm" (mask));
93 static __inline u_long
98 __asm __volatile("bsrq %1,%0" : "=r" (result) : "rm" (mask));
106 __asm __volatile("clflush %0" : : "m" (*(char *)addr));
110 clflushopt(u_long addr)
113 __asm __volatile(".byte 0x66;clflush %0" : : "m" (*(char *)addr));
120 __asm __volatile("clts");
126 __asm __volatile("cli" : : : "memory");
130 do_cpuid(u_int ax, u_int *p)
132 __asm __volatile("cpuid"
133 : "=a" (p[0]), "=b" (p[1]), "=c" (p[2]), "=d" (p[3])
138 cpuid_count(u_int ax, u_int cx, u_int *p)
140 __asm __volatile("cpuid"
141 : "=a" (p[0]), "=b" (p[1]), "=c" (p[2]), "=d" (p[3])
142 : "0" (ax), "c" (cx));
148 __asm __volatile("sti");
153 #define HAVE_INLINE_FFS
154 #define ffs(x) __builtin_ffs(x)
156 #define HAVE_INLINE_FFSL
161 return (mask == 0 ? mask : (int)bsfq((u_long)mask) + 1);
164 #define HAVE_INLINE_FFSLL
167 ffsll(long long mask)
169 return (ffsl((long)mask));
172 #define HAVE_INLINE_FLS
177 return (mask == 0 ? mask : (int)bsrl((u_int)mask) + 1);
180 #define HAVE_INLINE_FLSL
185 return (mask == 0 ? mask : (int)bsrq((u_long)mask) + 1);
188 #define HAVE_INLINE_FLSLL
191 flsll(long long mask)
193 return (flsl((long)mask));
201 __asm __volatile("hlt");
204 static __inline u_char
209 __asm __volatile("inb %w1, %0" : "=a" (data) : "Nd" (port));
213 static __inline u_int
218 __asm __volatile("inl %w1, %0" : "=a" (data) : "Nd" (port));
223 insb(u_int port, void *addr, size_t count)
225 __asm __volatile("cld; rep; insb"
226 : "+D" (addr), "+c" (count)
232 insw(u_int port, void *addr, size_t count)
234 __asm __volatile("cld; rep; insw"
235 : "+D" (addr), "+c" (count)
241 insl(u_int port, void *addr, size_t count)
243 __asm __volatile("cld; rep; insl"
244 : "+D" (addr), "+c" (count)
252 __asm __volatile("invd");
255 static __inline u_short
260 __asm __volatile("inw %w1, %0" : "=a" (data) : "Nd" (port));
265 outb(u_int port, u_char data)
267 __asm __volatile("outb %0, %w1" : : "a" (data), "Nd" (port));
271 outl(u_int port, u_int data)
273 __asm __volatile("outl %0, %w1" : : "a" (data), "Nd" (port));
277 outsb(u_int port, const void *addr, size_t count)
279 __asm __volatile("cld; rep; outsb"
280 : "+S" (addr), "+c" (count)
285 outsw(u_int port, const void *addr, size_t count)
287 __asm __volatile("cld; rep; outsw"
288 : "+S" (addr), "+c" (count)
293 outsl(u_int port, const void *addr, size_t count)
295 __asm __volatile("cld; rep; outsl"
296 : "+S" (addr), "+c" (count)
301 outw(u_int port, u_short data)
303 __asm __volatile("outw %0, %w1" : : "a" (data), "Nd" (port));
306 static __inline u_long
311 __asm __volatile("popcntq %1,%0" : "=r" (result) : "rm" (mask));
319 __asm __volatile("lfence" : : : "memory");
326 __asm __volatile("mfence" : : : "memory");
333 __asm __volatile("sfence" : : : "memory");
339 __asm __volatile("pause");
342 static __inline u_long
347 __asm __volatile("pushfq; popq %0" : "=r" (rf));
351 static __inline uint64_t
356 __asm __volatile("rdmsr" : "=a" (low), "=d" (high) : "c" (msr));
357 return (low | ((uint64_t)high << 32));
360 static __inline uint64_t
365 __asm __volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (pmc));
366 return (low | ((uint64_t)high << 32));
369 static __inline uint64_t
374 __asm __volatile("rdtsc" : "=a" (low), "=d" (high));
375 return (low | ((uint64_t)high << 32));
378 static __inline uint32_t
383 __asm __volatile("rdtsc" : "=a" (rv) : : "edx");
390 __asm __volatile("wbinvd");
394 write_rflags(u_long rf)
396 __asm __volatile("pushq %0; popfq" : : "r" (rf));
400 wrmsr(u_int msr, uint64_t newval)
406 __asm __volatile("wrmsr" : : "a" (low), "d" (high), "c" (msr));
410 load_cr0(u_long data)
413 __asm __volatile("movq %0,%%cr0" : : "r" (data));
416 static __inline u_long
421 __asm __volatile("movq %%cr0,%0" : "=r" (data));
425 static __inline u_long
430 __asm __volatile("movq %%cr2,%0" : "=r" (data));
435 load_cr3(u_long data)
438 __asm __volatile("movq %0,%%cr3" : : "r" (data) : "memory");
441 static __inline u_long
446 __asm __volatile("movq %%cr3,%0" : "=r" (data));
451 load_cr4(u_long data)
453 __asm __volatile("movq %0,%%cr4" : : "r" (data));
456 static __inline u_long
461 __asm __volatile("movq %%cr4,%0" : "=r" (data));
465 static __inline u_long
470 __asm __volatile("xgetbv" : "=a" (low), "=d" (high) : "c" (reg));
471 return (low | ((uint64_t)high << 32));
475 load_xcr(u_int reg, u_long val)
481 __asm __volatile("xsetbv" : : "c" (reg), "a" (low), "d" (high));
485 * Global TLB flush (except for thise for pages marked PG_G)
495 #define CR4_PGE 0x00000080 /* Page global enable */
499 * Perform the guaranteed invalidation of all TLB entries. This
500 * includes the global entries, and entries in all PCIDs, not only the
501 * current context. The function works both on non-PCID CPUs and CPUs
502 * with the PCID turned off or on. See IA-32 SDM Vol. 3a 4.10.4.1
503 * Operations that Invalidate TLBs and Paging-Structure Caches.
506 invltlb_globpcid(void)
511 load_cr4(cr4 & ~CR4_PGE);
513 * Although preemption at this point could be detrimental to
514 * performance, it would not lead to an error. PG_G is simply
515 * ignored if CR4.PGE is clear. Moreover, in case this block
516 * is re-entered, the load_cr4() either above or below will
517 * modify CR4.PGE flushing the TLB.
519 load_cr4(cr4 | CR4_PGE);
523 * TLB flush for an individual page (even if it has PG_G).
524 * Only works on 486+ CPUs (i386 does not have PG_G).
530 __asm __volatile("invlpg %0" : : "m" (*(char *)addr) : "memory");
533 #define INVPCID_ADDR 0
534 #define INVPCID_CTX 1
535 #define INVPCID_CTXGLOB 2
536 #define INVPCID_ALLCTX 3
538 struct invpcid_descr {
539 uint64_t pcid:12 __packed;
540 uint64_t pad:52 __packed;
545 invpcid(struct invpcid_descr *d, int type)
548 /* invpcid (%rdx),%rax */
549 __asm __volatile(".byte 0x66,0x0f,0x38,0x82,0x02"
550 : : "d" (d), "a" ((u_long)type) : "memory");
553 static __inline u_short
557 __asm __volatile("movw %%fs,%0" : "=rm" (sel));
561 static __inline u_short
565 __asm __volatile("movw %%gs,%0" : "=rm" (sel));
569 static __inline u_short
573 __asm __volatile("movw %%ss,%0" : "=rm" (sel));
580 __asm __volatile("movw %0,%%ds" : : "rm" (sel));
586 __asm __volatile("movw %0,%%es" : : "rm" (sel));
590 cpu_monitor(const void *addr, u_long extensions, u_int hints)
593 __asm __volatile("monitor"
594 : : "a" (addr), "c" (extensions), "d" (hints));
598 cpu_mwait(u_long extensions, u_int hints)
601 __asm __volatile("mwait" : : "a" (hints), "c" (extensions));
605 /* This is defined in <machine/specialreg.h> but is too painful to get to */
607 #define MSR_FSBASE 0xc0000100
612 /* Preserve the fsbase value across the selector load */
613 __asm __volatile("rdmsr; movw %0,%%fs; wrmsr"
614 : : "rm" (sel), "c" (MSR_FSBASE) : "eax", "edx");
618 #define MSR_GSBASE 0xc0000101
624 * Preserve the gsbase value across the selector load.
625 * Note that we have to disable interrupts because the gsbase
626 * being trashed happens to be the kernel gsbase at the time.
628 __asm __volatile("pushfq; cli; rdmsr; movw %0,%%gs; wrmsr; popfq"
629 : : "rm" (sel), "c" (MSR_GSBASE) : "eax", "edx");
632 /* Usable by userland */
636 __asm __volatile("movw %0,%%fs" : : "rm" (sel));
642 __asm __volatile("movw %0,%%gs" : : "rm" (sel));
647 lidt(struct region_descriptor *addr)
649 __asm __volatile("lidt (%0)" : : "r" (addr));
655 __asm __volatile("lldt %0" : : "r" (sel));
661 __asm __volatile("ltr %0" : : "r" (sel));
664 static __inline uint64_t
668 __asm __volatile("movq %%dr0,%0" : "=r" (data));
673 load_dr0(uint64_t dr0)
675 __asm __volatile("movq %0,%%dr0" : : "r" (dr0));
678 static __inline uint64_t
682 __asm __volatile("movq %%dr1,%0" : "=r" (data));
687 load_dr1(uint64_t dr1)
689 __asm __volatile("movq %0,%%dr1" : : "r" (dr1));
692 static __inline uint64_t
696 __asm __volatile("movq %%dr2,%0" : "=r" (data));
701 load_dr2(uint64_t dr2)
703 __asm __volatile("movq %0,%%dr2" : : "r" (dr2));
706 static __inline uint64_t
710 __asm __volatile("movq %%dr3,%0" : "=r" (data));
715 load_dr3(uint64_t dr3)
717 __asm __volatile("movq %0,%%dr3" : : "r" (dr3));
720 static __inline uint64_t
724 __asm __volatile("movq %%dr4,%0" : "=r" (data));
729 load_dr4(uint64_t dr4)
731 __asm __volatile("movq %0,%%dr4" : : "r" (dr4));
734 static __inline uint64_t
738 __asm __volatile("movq %%dr5,%0" : "=r" (data));
743 load_dr5(uint64_t dr5)
745 __asm __volatile("movq %0,%%dr5" : : "r" (dr5));
748 static __inline uint64_t
752 __asm __volatile("movq %%dr6,%0" : "=r" (data));
757 load_dr6(uint64_t dr6)
759 __asm __volatile("movq %0,%%dr6" : : "r" (dr6));
762 static __inline uint64_t
766 __asm __volatile("movq %%dr7,%0" : "=r" (data));
771 load_dr7(uint64_t dr7)
773 __asm __volatile("movq %0,%%dr7" : : "r" (dr7));
776 static __inline register_t
781 rflags = read_rflags();
787 intr_restore(register_t rflags)
789 write_rflags(rflags);
792 #else /* !(__GNUCLIKE_ASM && __CC_SUPPORTS___INLINE) */
794 int breakpoint(void);
795 u_int bsfl(u_int mask);
796 u_int bsrl(u_int mask);
797 void clflush(u_long addr);
799 void cpuid_count(u_int ax, u_int cx, u_int *p);
800 void disable_intr(void);
801 void do_cpuid(u_int ax, u_int *p);
802 void enable_intr(void);
804 void ia32_pause(void);
805 u_char inb(u_int port);
806 u_int inl(u_int port);
807 void insb(u_int port, void *addr, size_t count);
808 void insl(u_int port, void *addr, size_t count);
809 void insw(u_int port, void *addr, size_t count);
810 register_t intr_disable(void);
811 void intr_restore(register_t rf);
813 void invlpg(u_int addr);
815 u_short inw(u_int port);
816 void lidt(struct region_descriptor *addr);
817 void lldt(u_short sel);
818 void load_cr0(u_long cr0);
819 void load_cr3(u_long cr3);
820 void load_cr4(u_long cr4);
821 void load_dr0(uint64_t dr0);
822 void load_dr1(uint64_t dr1);
823 void load_dr2(uint64_t dr2);
824 void load_dr3(uint64_t dr3);
825 void load_dr4(uint64_t dr4);
826 void load_dr5(uint64_t dr5);
827 void load_dr6(uint64_t dr6);
828 void load_dr7(uint64_t dr7);
829 void load_fs(u_short sel);
830 void load_gs(u_short sel);
831 void ltr(u_short sel);
832 void outb(u_int port, u_char data);
833 void outl(u_int port, u_int data);
834 void outsb(u_int port, const void *addr, size_t count);
835 void outsl(u_int port, const void *addr, size_t count);
836 void outsw(u_int port, const void *addr, size_t count);
837 void outw(u_int port, u_short data);
842 uint64_t rdmsr(u_int msr);
843 uint64_t rdpmc(u_int pmc);
852 uint64_t rdtsc(void);
853 u_long read_rflags(void);
857 void write_rflags(u_int rf);
858 void wrmsr(u_int msr, uint64_t newval);
860 #endif /* __GNUCLIKE_ASM && __CC_SUPPORTS___INLINE */
862 void reset_dbregs(void);
865 int rdmsr_safe(u_int msr, uint64_t *val);
866 int wrmsr_safe(u_int msr, uint64_t newval);
869 #endif /* !_MACHINE_CPUFUNC_H_ */