cpufunc.h revision 35976
1193267Sjkim/*- 2193267Sjkim * Copyright (c) 1993 The Regents of the University of California. 3193267Sjkim * All rights reserved. 4193267Sjkim * 5193267Sjkim * Redistribution and use in source and binary forms, with or without 6193267Sjkim * modification, are permitted provided that the following conditions 7217355Sjkim * are met: 8298714Sjkim * 1. Redistributions of source code must retain the above copyright 9193267Sjkim * notice, this list of conditions and the following disclaimer. 10193267Sjkim * 2. Redistributions in binary form must reproduce the above copyright 11217355Sjkim * notice, this list of conditions and the following disclaimer in the 12217355Sjkim * documentation and/or other materials provided with the distribution. 13217355Sjkim * 3. All advertising materials mentioning features or use of this software 14217355Sjkim * must display the following acknowledgement: 15217355Sjkim * This product includes software developed by the University of 16217355Sjkim * California, Berkeley and its contributors. 17217355Sjkim * 4. Neither the name of the University nor the names of its contributors 18217355Sjkim * may be used to endorse or promote products derived from this software 19217355Sjkim * without specific prior written permission. 20217355Sjkim * 21217355Sjkim * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22217355Sjkim * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23217355Sjkim * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24217355Sjkim * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25193267Sjkim * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26217355Sjkim * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27217355Sjkim * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28217355Sjkim * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29193267Sjkim * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30217355Sjkim * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31217355Sjkim * SUCH DAMAGE. 32217355Sjkim * 33217355Sjkim * $Id: cpufunc.h,v 1.77 1998/05/11 02:13:47 dyson Exp $ 34217355Sjkim */ 35217355Sjkim 36217355Sjkim/* 37217355Sjkim * Functions to provide access to special i386 instructions. 38217355Sjkim */ 39217355Sjkim 40217355Sjkim#ifndef _MACHINE_CPUFUNC_H_ 41217355Sjkim#define _MACHINE_CPUFUNC_H_ 42217355Sjkim 43193267Sjkim#include <sys/cdefs.h> 44193267Sjkim#include <sys/types.h> 45213800Sjkim 46213800Sjkim#include <machine/lock.h> 47193267Sjkim 48228110Sjkim#if defined(SWTCH_OPTIM_STATS) 49228110Sjkimextern int tlb_flush_count; 50228110Sjkim#endif 51228110Sjkim 52228110Sjkim#ifdef __GNUC__ 53223480Sjkim 54193267Sjkimstatic __inline void 55193267Sjkimbreakpoint(void) 56193267Sjkim{ 57193267Sjkim __asm __volatile("int $3"); 58193267Sjkim} 59193267Sjkim 60193267Sjkimstatic __inline void 61209734Sjkimdisable_intr(void) 62193267Sjkim{ 63193267Sjkim __asm __volatile("cli" : : : "memory"); 64193267Sjkim MPINTR_LOCK(); 65193267Sjkim} 66193267Sjkim 67272444Sjkimstatic __inline void 68193267Sjkimenable_intr(void) 69193267Sjkim{ 70193267Sjkim MPINTR_UNLOCK(); 71193267Sjkim __asm __volatile("sti"); 72193267Sjkim} 73193267Sjkim 74193267Sjkim#define HAVE_INLINE_FFS 75193267Sjkim 76193267Sjkimstatic __inline int 77193267Sjkimffs(int mask) 78193267Sjkim{ 79245582Sjkim int result; 80245582Sjkim /* 81272444Sjkim * bsfl turns out to be not all that slow on 486's. It can beaten 82272444Sjkim * using a binary search to reduce to 4 bits and then a table lookup, 83245582Sjkim * but only if the code is inlined and in the cache, and the code 84272444Sjkim * is quite large so inlining it probably busts the cache. 85193267Sjkim * 86272444Sjkim * Note that gcc-2's builtin ffs would be used if we didn't declare 87272444Sjkim * this inline or turn off the builtin. The builtin is faster but 88272444Sjkim * broken in gcc-2.4.5 and slower but working in gcc-2.5 and 2.6. 89272444Sjkim */ 90272444Sjkim __asm __volatile("testl %0,%0; je 1f; bsfl %0,%0; incl %0; 1:" 91272444Sjkim : "=r" (result) : "0" (mask)); 92272444Sjkim return (result); 93272444Sjkim} 94272444Sjkim 95272444Sjkim#define HAVE_INLINE_FLS 96272444Sjkim 97272444Sjkimstatic __inline int 98272444Sjkimfls(int mask) 99272444Sjkim{ 100272444Sjkim int result; 101272444Sjkim __asm __volatile("testl %0,%0; je 1f; bsrl %0,%0; incl %0; 1:" 102272444Sjkim : "=r" (result) : "0" (mask)); 103272444Sjkim return (result); 104272444Sjkim} 105272444Sjkim 106272444Sjkim#if __GNUC__ < 2 107193267Sjkim 108193267Sjkim#define inb(port) inbv(port) 109272444Sjkim#define outb(port, data) outbv(port, data) 110272444Sjkim 111272444Sjkim#else /* __GNUC >= 2 */ 112272444Sjkim 113272444Sjkim/* 114272444Sjkim * The following complications are to get around gcc not having a 115272444Sjkim * constraint letter for the range 0..255. We still put "d" in the 116272444Sjkim * constraint because "i" isn't a valid constraint when the port 117272444Sjkim * isn't constant. This only matters for -O0 because otherwise 118272444Sjkim * the non-working version gets optimized away. 119272444Sjkim * 120272444Sjkim * Use an expression-statement instead of a conditional expression 121272444Sjkim * because gcc-2.6.0 would promote the operands of the conditional 122272444Sjkim * and produce poor code for "if ((inb(var) & const1) == const2)". 123272444Sjkim * 124272444Sjkim * The unnecessary test `(port) < 0x10000' is to generate a warning if 125272444Sjkim * the `port' has type u_short or smaller. Such types are pessimal. 126272444Sjkim * This actually only works for signed types. The range check is 127272444Sjkim * careful to avoid generating warnings. 128272444Sjkim */ 129272444Sjkim#define inb(port) __extension__ ({ \ 130272444Sjkim u_char _data; \ 131272444Sjkim if (__builtin_constant_p(port) && ((port) & 0xffff) < 0x100 \ 132272444Sjkim && (port) < 0x10000) \ 133272444Sjkim _data = inbc(port); \ 134272444Sjkim else \ 135272444Sjkim _data = inbv(port); \ 136272444Sjkim _data; }) 137272444Sjkim 138272444Sjkim#define outb(port, data) ( \ 139272444Sjkim __builtin_constant_p(port) && ((port) & 0xffff) < 0x100 \ 140272444Sjkim && (port) < 0x10000 \ 141272444Sjkim ? outbc(port, data) : outbv(port, data)) 142272444Sjkim 143272444Sjkimstatic __inline u_char 144272444Sjkiminbc(u_int port) 145272444Sjkim{ 146272444Sjkim u_char data; 147272444Sjkim 148272444Sjkim __asm __volatile("inb %1,%0" : "=a" (data) : "id" ((u_short)(port))); 149272444Sjkim return (data); 150272444Sjkim} 151272444Sjkim 152272444Sjkimstatic __inline void 153272444Sjkimoutbc(u_int port, u_char data) 154272444Sjkim{ 155272444Sjkim __asm __volatile("outb %0,%1" : : "a" (data), "id" ((u_short)(port))); 156272444Sjkim} 157272444Sjkim 158272444Sjkim#endif /* __GNUC <= 2 */ 159272444Sjkim 160272444Sjkimstatic __inline u_char 161272444Sjkiminbv(u_int port) 162272444Sjkim{ 163272444Sjkim u_char data; 164272444Sjkim /* 165272444Sjkim * We use %%dx and not %1 here because i/o is done at %dx and not at 166272444Sjkim * %edx, while gcc generates inferior code (movw instead of movl) 167272444Sjkim * if we tell it to load (u_short) port. 168272444Sjkim */ 169272444Sjkim __asm __volatile("inb %%dx,%0" : "=a" (data) : "d" (port)); 170272444Sjkim return (data); 171272444Sjkim} 172272444Sjkim 173272444Sjkimstatic __inline u_long 174272444Sjkiminl(u_int port) 175272444Sjkim{ 176272444Sjkim u_long data; 177272444Sjkim 178272444Sjkim __asm __volatile("inl %%dx,%0" : "=a" (data) : "d" (port)); 179272444Sjkim return (data); 180272444Sjkim} 181272444Sjkim 182272444Sjkimstatic __inline void 183272444Sjkiminsb(u_int port, void *addr, size_t cnt) 184272444Sjkim{ 185272444Sjkim __asm __volatile("cld; rep; insb" 186272444Sjkim : : "d" (port), "D" (addr), "c" (cnt) 187272444Sjkim : "di", "cx", "memory"); 188272444Sjkim} 189272444Sjkim 190272444Sjkimstatic __inline void 191272444Sjkiminsw(u_int port, void *addr, size_t cnt) 192272444Sjkim{ 193272444Sjkim __asm __volatile("cld; rep; insw" 194272444Sjkim : : "d" (port), "D" (addr), "c" (cnt) 195272444Sjkim : "di", "cx", "memory"); 196272444Sjkim} 197272444Sjkim 198272444Sjkimstatic __inline void 199272444Sjkiminsl(u_int port, void *addr, size_t cnt) 200272444Sjkim{ 201272444Sjkim __asm __volatile("cld; rep; insl" 202272444Sjkim : : "d" (port), "D" (addr), "c" (cnt) 203272444Sjkim : "di", "cx", "memory"); 204272444Sjkim} 205272444Sjkim 206193267Sjkimstatic __inline void 207193267Sjkiminvd(void) 208193267Sjkim{ 209193267Sjkim __asm __volatile("invd"); 210193267Sjkim} 211193267Sjkim 212272444Sjkim#ifdef KERNEL 213193267Sjkim#ifdef SMP 214193267Sjkim 215193267Sjkim/* 216193267Sjkim * When using APIC IPI's, the inlining cost is prohibitive since the call 217213800Sjkim * executes into the IPI transmission system. 218213800Sjkim */ 219193267Sjkimvoid invlpg __P((u_int addr)); 220272444Sjkimvoid invltlb __P((void)); 221193267Sjkim 222272444Sjkimstatic __inline void 223193267Sjkimcpu_invlpg(void *addr) 224272444Sjkim{ 225272444Sjkim __asm __volatile("invlpg %0"::"m"(*(char *)addr):"memory"); 226272444Sjkim} 227272444Sjkim 228272444Sjkim#else /* !SMP */ 229272444Sjkim 230272444Sjkimstatic __inline void 231272444Sjkiminvlpg(u_int addr) 232272444Sjkim{ 233193267Sjkim __asm __volatile("invlpg %0"::"m"(*(char *)addr):"memory"); 234193267Sjkim} 235193267Sjkim 236193267Sjkim 237213800Sjkimstatic __inline void 238213800Sjkiminvltlb(void) 239193267Sjkim{ 240193267Sjkim u_long temp; 241272444Sjkim /* 242193267Sjkim * This should be implemented as load_cr3(rcr3()) when load_cr3() 243193267Sjkim * is inlined. 244193267Sjkim */ 245193267Sjkim __asm __volatile("movl %%cr3, %0; movl %0, %%cr3" : "=r" (temp) 246272444Sjkim : : "memory"); 247193267Sjkim#if defined(SWTCH_OPTIM_STATS) 248193267Sjkim ++tlb_flush_count; 249193267Sjkim#endif 250193267Sjkim} 251193267Sjkim 252193267Sjkim#endif /* SMP */ 253193267Sjkim#endif /* KERNEL */ 254193267Sjkim 255213800Sjkimstatic __inline u_short 256193267Sjkiminw(u_int port) 257193267Sjkim{ 258193267Sjkim u_short data; 259193267Sjkim 260193267Sjkim __asm __volatile("inw %%dx,%0" : "=a" (data) : "d" (port)); 261193267Sjkim return (data); 262193267Sjkim} 263193267Sjkim 264272444Sjkimstatic __inline u_int 265193267Sjkimloadandclear(u_int *addr) 266272444Sjkim{ 267193267Sjkim u_int result; 268193267Sjkim 269193267Sjkim __asm __volatile("xorl %0,%0; xchgl %1,%0" 270193267Sjkim : "=&r" (result) : "m" (*addr)); 271193267Sjkim return (result); 272193267Sjkim} 273213800Sjkim 274213800Sjkimstatic __inline void 275193267Sjkimoutbv(u_int port, u_char data) 276213800Sjkim{ 277193267Sjkim u_char al; 278193267Sjkim /* 279193267Sjkim * Use an unnecessary assignment to help gcc's register allocator. 280193267Sjkim * This make a large difference for gcc-1.40 and a tiny difference 281193267Sjkim * for gcc-2.6.0. For gcc-1.40, al had to be ``asm("ax")'' for 282193267Sjkim * best results. gcc-2.6.0 can't handle this. 283193267Sjkim */ 284193267Sjkim al = data; 285193267Sjkim __asm __volatile("outb %0,%%dx" : : "a" (al), "d" (port)); 286193267Sjkim} 287193267Sjkim 288193267Sjkimstatic __inline void 289193267Sjkimoutl(u_int port, u_long data) 290193267Sjkim{ 291193267Sjkim /* 292193267Sjkim * outl() and outw() aren't used much so we haven't looked at 293193267Sjkim * possible micro-optimizations such as the unnecessary 294193267Sjkim * assignment for them. 295193267Sjkim */ 296193267Sjkim __asm __volatile("outl %0,%%dx" : : "a" (data), "d" (port)); 297193267Sjkim} 298193267Sjkim 299193267Sjkimstatic __inline void 300193267Sjkimoutsb(u_int port, const void *addr, size_t cnt) 301193267Sjkim{ 302193267Sjkim __asm __volatile("cld; rep; outsb" 303213800Sjkim : : "d" (port), "S" (addr), "c" (cnt) 304213800Sjkim : "si", "cx"); 305213800Sjkim} 306193267Sjkim 307213800Sjkimstatic __inline void 308193267Sjkimoutsw(u_int port, const void *addr, size_t cnt) 309193267Sjkim{ 310193267Sjkim __asm __volatile("cld; rep; outsw" 311193267Sjkim : : "d" (port), "S" (addr), "c" (cnt) 312193267Sjkim : "si", "cx"); 313193267Sjkim} 314193267Sjkim 315193267Sjkimstatic __inline void 316193267Sjkimoutsl(u_int port, const void *addr, size_t cnt) 317193267Sjkim{ 318193267Sjkim __asm __volatile("cld; rep; outsl" 319193267Sjkim : : "d" (port), "S" (addr), "c" (cnt) 320193267Sjkim : "si", "cx"); 321193267Sjkim} 322193267Sjkim 323193267Sjkimstatic __inline void 324193267Sjkimoutw(u_int port, u_short data) 325193267Sjkim{ 326193267Sjkim __asm __volatile("outw %0,%%dx" : : "a" (data), "d" (port)); 327193267Sjkim} 328193267Sjkim 329193267Sjkimstatic __inline u_long 330193267Sjkimrcr2(void) 331193267Sjkim{ 332193267Sjkim u_long data; 333193267Sjkim 334193267Sjkim __asm __volatile("movl %%cr2,%0" : "=r" (data)); 335193267Sjkim return (data); 336193267Sjkim} 337193267Sjkim 338193267Sjkimstatic __inline u_long 339193267Sjkimread_eflags(void) 340231844Sjkim{ 341231844Sjkim u_long ef; 342231844Sjkim 343231844Sjkim __asm __volatile("pushfl; popl %0" : "=r" (ef)); 344231844Sjkim return (ef); 345231844Sjkim} 346231844Sjkim 347231844Sjkimstatic __inline quad_t 348231844Sjkimrdmsr(u_int msr) 349231844Sjkim{ 350231844Sjkim quad_t rv; 351231844Sjkim 352231844Sjkim __asm __volatile(".byte 0x0f, 0x32" : "=A" (rv) : "c" (msr)); 353231844Sjkim return (rv); 354231844Sjkim} 355231844Sjkim 356231844Sjkimstatic __inline quad_t 357231844Sjkimrdpmc(u_int pmc) 358231844Sjkim{ 359231844Sjkim quad_t rv; 360231844Sjkim 361231844Sjkim __asm __volatile(".byte 0x0f, 0x33" : "=A" (rv) : "c" (pmc)); 362231844Sjkim return (rv); 363231844Sjkim} 364231844Sjkim 365231844Sjkimstatic __inline quad_t 366231844Sjkimrdtsc(void) 367193267Sjkim{ 368193267Sjkim quad_t rv; 369193267Sjkim 370193267Sjkim __asm __volatile(".byte 0x0f, 0x31" : "=A" (rv)); 371193267Sjkim return (rv); 372193267Sjkim} 373193267Sjkim 374193267Sjkimstatic __inline void 375193267Sjkimsetbits(volatile unsigned *addr, u_int bits) 376193267Sjkim{ 377193267Sjkim __asm __volatile( 378193267Sjkim#ifdef SMP 379193267Sjkim "lock; " 380193267Sjkim#endif 381193267Sjkim "orl %1,%0" : "=m" (*addr) : "ir" (bits)); 382193267Sjkim} 383193267Sjkim 384193267Sjkimstatic __inline void 385193267Sjkimwbinvd(void) 386193267Sjkim{ 387193267Sjkim __asm __volatile("wbinvd"); 388193267Sjkim} 389193267Sjkim 390213800Sjkimstatic __inline void 391193267Sjkimwrite_eflags(u_long ef) 392193267Sjkim{ 393193267Sjkim __asm __volatile("pushl %0; popfl" : : "r" (ef)); 394245582Sjkim} 395245582Sjkim 396193267Sjkimstatic __inline void 397193267Sjkimwrmsr(u_int msr, quad_t newval) 398193267Sjkim{ 399193267Sjkim __asm __volatile(".byte 0x0f, 0x30" : : "A" (newval), "c" (msr)); 400193267Sjkim} 401193267Sjkim 402193267Sjkim#else /* !__GNUC__ */ 403193267Sjkim 404193267Sjkimint breakpoint __P((void)); 405245582Sjkimvoid disable_intr __P((void)); 406193267Sjkimvoid enable_intr __P((void)); 407193267Sjkimu_char inb __P((u_int port)); 408245582Sjkimu_long inl __P((u_int port)); 409245582Sjkimvoid insb __P((u_int port, void *addr, size_t cnt)); 410245582Sjkimvoid insl __P((u_int port, void *addr, size_t cnt)); 411245582Sjkimvoid insw __P((u_int port, void *addr, size_t cnt)); 412245582Sjkimvoid invd __P((void)); 413245582Sjkimvoid invlpg __P((u_int addr)); 414245582Sjkimvoid invltlb __P((void)); 415245582Sjkimu_short inw __P((u_int port)); 416245582Sjkimu_int loadandclear __P((u_int *addr)); 417245582Sjkimvoid outb __P((u_int port, u_char data)); 418245582Sjkimvoid outl __P((u_int port, u_long data)); 419245582Sjkimvoid outsb __P((u_int port, void *addr, size_t cnt)); 420245582Sjkimvoid outsl __P((u_int port, void *addr, size_t cnt)); 421245582Sjkimvoid outsw __P((u_int port, void *addr, size_t cnt)); 422245582Sjkimvoid outw __P((u_int port, u_short data)); 423245582Sjkimu_long rcr2 __P((void)); 424245582Sjkimquad_t rdmsr __P((u_int msr)); 425245582Sjkimquad_t rdpmc __P((u_int pmc)); 426245582Sjkimquad_t rdtsc __P((void)); 427245582Sjkimu_long read_eflags __P((void)); 428245582Sjkimvoid setbits __P((volatile unsigned *addr, u_int bits)); 429245582Sjkimvoid wbinvd __P((void)); 430245582Sjkimvoid write_eflags __P((u_long ef)); 431245582Sjkimvoid wrmsr __P((u_int msr, quad_t newval)); 432245582Sjkim 433245582Sjkim#endif /* __GNUC__ */ 434245582Sjkim 435193267Sjkimvoid load_cr0 __P((u_long cr0)); 436193267Sjkimvoid load_cr3 __P((u_long cr3)); 437193267Sjkimvoid load_cr4 __P((u_long cr4)); 438193267Sjkimvoid ltr __P((u_short sel)); 439193267Sjkimu_int rcr0 __P((void)); 440193267Sjkimu_long rcr3 __P((void)); 441193267Sjkimu_long rcr4 __P((void)); 442213800Sjkimvoid i686_pagezero __P((void *addr)); 443213800Sjkim 444193267Sjkim#endif /* !_MACHINE_CPUFUNC_H_ */ 445193267Sjkim