xen-vtx-unstable
annotate linux-2.6.11-xen-sparse/include/asm-xen/asm-i386/system.h @ 5367:70de9cd5f6b6
bitkeeper revision 1.1691.1.2 (42a6ae59pOLKl7oBFyH6Ukzy7yg3EA)
read/write control-reg macros are now same in xenlinux as native linux.
Signed-oiff-by: Keir Fraser <keir@xensource.com>
read/write control-reg macros are now same in xenlinux as native linux.
Signed-oiff-by: Keir Fraser <keir@xensource.com>
author | kaf24@firebug.cl.cam.ac.uk |
---|---|
date | Wed Jun 08 08:37:45 2005 +0000 (2005-06-08) |
parents | 65b28c74cec2 |
children | 5cc07785693b |
rev | line source |
---|---|
cl349@4087 | 1 #ifndef __ASM_SYSTEM_H |
cl349@4087 | 2 #define __ASM_SYSTEM_H |
cl349@4087 | 3 |
cl349@4087 | 4 #include <linux/config.h> |
cl349@4087 | 5 #include <linux/kernel.h> |
cl349@4087 | 6 #include <linux/bitops.h> |
cl349@4087 | 7 #include <asm/synch_bitops.h> |
cl349@4087 | 8 #include <asm/segment.h> |
cl349@4087 | 9 #include <asm/cpufeature.h> |
cl349@4087 | 10 #include <asm-xen/hypervisor.h> |
cl349@4087 | 11 |
cl349@4087 | 12 #ifdef __KERNEL__ |
cl349@4087 | 13 |
cl349@4087 | 14 struct task_struct; /* one of the stranger aspects of C forward declarations.. */ |
cl349@4087 | 15 extern struct task_struct * FASTCALL(__switch_to(struct task_struct *prev, struct task_struct *next)); |
cl349@4087 | 16 |
cl349@4087 | 17 #define switch_to(prev,next,last) do { \ |
cl349@4087 | 18 unsigned long esi,edi; \ |
cl349@4087 | 19 asm volatile("pushfl\n\t" \ |
cl349@4087 | 20 "pushl %%ebp\n\t" \ |
cl349@4087 | 21 "movl %%esp,%0\n\t" /* save ESP */ \ |
cl349@4087 | 22 "movl %5,%%esp\n\t" /* restore ESP */ \ |
cl349@4087 | 23 "movl $1f,%1\n\t" /* save EIP */ \ |
cl349@4087 | 24 "pushl %6\n\t" /* restore EIP */ \ |
cl349@4087 | 25 "jmp __switch_to\n" \ |
cl349@4087 | 26 "1:\t" \ |
cl349@4087 | 27 "popl %%ebp\n\t" \ |
cl349@4087 | 28 "popfl" \ |
cl349@4087 | 29 :"=m" (prev->thread.esp),"=m" (prev->thread.eip), \ |
cl349@4087 | 30 "=a" (last),"=S" (esi),"=D" (edi) \ |
cl349@4087 | 31 :"m" (next->thread.esp),"m" (next->thread.eip), \ |
cl349@4087 | 32 "2" (prev), "d" (next)); \ |
cl349@4087 | 33 } while (0) |
cl349@4087 | 34 |
cl349@4087 | 35 #define _set_base(addr,base) do { unsigned long __pr; \ |
cl349@4087 | 36 __asm__ __volatile__ ("movw %%dx,%1\n\t" \ |
cl349@4087 | 37 "rorl $16,%%edx\n\t" \ |
cl349@4087 | 38 "movb %%dl,%2\n\t" \ |
cl349@4087 | 39 "movb %%dh,%3" \ |
cl349@4087 | 40 :"=&d" (__pr) \ |
cl349@4087 | 41 :"m" (*((addr)+2)), \ |
cl349@4087 | 42 "m" (*((addr)+4)), \ |
cl349@4087 | 43 "m" (*((addr)+7)), \ |
cl349@4087 | 44 "0" (base) \ |
cl349@4087 | 45 ); } while(0) |
cl349@4087 | 46 |
cl349@4087 | 47 #define _set_limit(addr,limit) do { unsigned long __lr; \ |
cl349@4087 | 48 __asm__ __volatile__ ("movw %%dx,%1\n\t" \ |
cl349@4087 | 49 "rorl $16,%%edx\n\t" \ |
cl349@4087 | 50 "movb %2,%%dh\n\t" \ |
cl349@4087 | 51 "andb $0xf0,%%dh\n\t" \ |
cl349@4087 | 52 "orb %%dh,%%dl\n\t" \ |
cl349@4087 | 53 "movb %%dl,%2" \ |
cl349@4087 | 54 :"=&d" (__lr) \ |
cl349@4087 | 55 :"m" (*(addr)), \ |
cl349@4087 | 56 "m" (*((addr)+6)), \ |
cl349@4087 | 57 "0" (limit) \ |
cl349@4087 | 58 ); } while(0) |
cl349@4087 | 59 |
cl349@4087 | 60 #define set_base(ldt,base) _set_base( ((char *)&(ldt)) , (base) ) |
cl349@4087 | 61 #define set_limit(ldt,limit) _set_limit( ((char *)&(ldt)) , ((limit)-1)>>12 ) |
cl349@4087 | 62 |
cl349@4087 | 63 static inline unsigned long _get_base(char * addr) |
cl349@4087 | 64 { |
cl349@4087 | 65 unsigned long __base; |
cl349@4087 | 66 __asm__("movb %3,%%dh\n\t" |
cl349@4087 | 67 "movb %2,%%dl\n\t" |
cl349@4087 | 68 "shll $16,%%edx\n\t" |
cl349@4087 | 69 "movw %1,%%dx" |
cl349@4087 | 70 :"=&d" (__base) |
cl349@4087 | 71 :"m" (*((addr)+2)), |
cl349@4087 | 72 "m" (*((addr)+4)), |
cl349@4087 | 73 "m" (*((addr)+7))); |
cl349@4087 | 74 return __base; |
cl349@4087 | 75 } |
cl349@4087 | 76 |
cl349@4087 | 77 #define get_base(ldt) _get_base( ((char *)&(ldt)) ) |
cl349@4087 | 78 |
cl349@4087 | 79 /* |
cl349@4087 | 80 * Load a segment. Fall back on loading the zero |
cl349@4087 | 81 * segment if something goes wrong.. |
cl349@4087 | 82 */ |
cl349@4087 | 83 #define loadsegment(seg,value) \ |
cl349@4087 | 84 asm volatile("\n" \ |
cl349@4087 | 85 "1:\t" \ |
cl349@4087 | 86 "movl %0,%%" #seg "\n" \ |
cl349@4087 | 87 "2:\n" \ |
cl349@4087 | 88 ".section .fixup,\"ax\"\n" \ |
cl349@4087 | 89 "3:\t" \ |
cl349@4087 | 90 "pushl $0\n\t" \ |
cl349@4087 | 91 "popl %%" #seg "\n\t" \ |
cl349@4087 | 92 "jmp 2b\n" \ |
cl349@4087 | 93 ".previous\n" \ |
cl349@4087 | 94 ".section __ex_table,\"a\"\n\t" \ |
cl349@4087 | 95 ".align 4\n\t" \ |
cl349@4087 | 96 ".long 1b,3b\n" \ |
cl349@4087 | 97 ".previous" \ |
cl349@4087 | 98 : :"m" (*(unsigned int *)&(value))) |
cl349@4087 | 99 |
cl349@4087 | 100 /* |
cl349@4087 | 101 * Save a segment register away |
cl349@4087 | 102 */ |
cl349@4087 | 103 #define savesegment(seg, value) \ |
cl349@4087 | 104 asm volatile("movl %%" #seg ",%0":"=m" (*(int *)&(value))) |
cl349@4087 | 105 |
cl349@4087 | 106 /* |
cl349@4087 | 107 * Clear and set 'TS' bit respectively |
cl349@4087 | 108 */ |
cl349@4112 | 109 #define clts() (HYPERVISOR_fpu_taskswitch(0)) |
kaf24@5367 | 110 #define read_cr0() ({ \ |
kaf24@5367 | 111 unsigned int __dummy; \ |
kaf24@5367 | 112 __asm__( \ |
kaf24@5367 | 113 "movl %%cr0,%0\n\t" \ |
kaf24@5367 | 114 :"=r" (__dummy)); \ |
kaf24@5367 | 115 __dummy; \ |
kaf24@5367 | 116 }) |
cl349@4087 | 117 #define write_cr0(x) \ |
kaf24@5367 | 118 __asm__("movl %0,%%cr0": :"r" (x)); |
kaf24@5367 | 119 |
kaf24@5367 | 120 #define read_cr4() ({ \ |
kaf24@5367 | 121 unsigned int __dummy; \ |
kaf24@5367 | 122 __asm__( \ |
kaf24@5367 | 123 "movl %%cr4,%0\n\t" \ |
kaf24@5367 | 124 :"=r" (__dummy)); \ |
kaf24@5367 | 125 __dummy; \ |
kaf24@5367 | 126 }) |
cl349@4087 | 127 #define write_cr4(x) \ |
kaf24@5367 | 128 __asm__("movl %0,%%cr4": :"r" (x)); |
cl349@4112 | 129 #define stts() (HYPERVISOR_fpu_taskswitch(1)) |
cl349@4087 | 130 |
cl349@4087 | 131 #endif /* __KERNEL__ */ |
cl349@4087 | 132 |
cl349@4112 | 133 #define wbinvd() \ |
cl349@4112 | 134 __asm__ __volatile__ ("wbinvd": : :"memory"); |
cl349@4087 | 135 |
cl349@4087 | 136 static inline unsigned long get_limit(unsigned long segment) |
cl349@4087 | 137 { |
cl349@4087 | 138 unsigned long __limit; |
cl349@4087 | 139 __asm__("lsll %1,%0" |
cl349@4087 | 140 :"=r" (__limit):"r" (segment)); |
cl349@4087 | 141 return __limit+1; |
cl349@4087 | 142 } |
cl349@4087 | 143 |
cl349@4087 | 144 #define nop() __asm__ __volatile__ ("nop") |
cl349@4087 | 145 |
cl349@4087 | 146 #define xchg(ptr,v) ((__typeof__(*(ptr)))__xchg((unsigned long)(v),(ptr),sizeof(*(ptr)))) |
cl349@4087 | 147 |
cl349@4087 | 148 #define tas(ptr) (xchg((ptr),1)) |
cl349@4087 | 149 |
cl349@4087 | 150 struct __xchg_dummy { unsigned long a[100]; }; |
cl349@4087 | 151 #define __xg(x) ((struct __xchg_dummy *)(x)) |
cl349@4087 | 152 |
cl349@4087 | 153 |
cl349@4087 | 154 /* |
cl349@4087 | 155 * The semantics of XCHGCMP8B are a bit strange, this is why |
cl349@4087 | 156 * there is a loop and the loading of %%eax and %%edx has to |
cl349@4087 | 157 * be inside. This inlines well in most cases, the cached |
cl349@4087 | 158 * cost is around ~38 cycles. (in the future we might want |
cl349@4087 | 159 * to do an SIMD/3DNOW!/MMX/FPU 64-bit store here, but that |
cl349@4087 | 160 * might have an implicit FPU-save as a cost, so it's not |
cl349@4087 | 161 * clear which path to go.) |
cl349@4087 | 162 * |
cl349@4087 | 163 * cmpxchg8b must be used with the lock prefix here to allow |
cl349@4087 | 164 * the instruction to be executed atomically, see page 3-102 |
cl349@4087 | 165 * of the instruction set reference 24319102.pdf. We need |
cl349@4087 | 166 * the reader side to see the coherent 64bit value. |
cl349@4087 | 167 */ |
cl349@4087 | 168 static inline void __set_64bit (unsigned long long * ptr, |
cl349@4087 | 169 unsigned int low, unsigned int high) |
cl349@4087 | 170 { |
cl349@4087 | 171 __asm__ __volatile__ ( |
cl349@4087 | 172 "\n1:\t" |
cl349@4087 | 173 "movl (%0), %%eax\n\t" |
cl349@4087 | 174 "movl 4(%0), %%edx\n\t" |
cl349@4087 | 175 "lock cmpxchg8b (%0)\n\t" |
cl349@4087 | 176 "jnz 1b" |
cl349@4087 | 177 : /* no outputs */ |
cl349@4087 | 178 : "D"(ptr), |
cl349@4087 | 179 "b"(low), |
cl349@4087 | 180 "c"(high) |
cl349@4087 | 181 : "ax","dx","memory"); |
cl349@4087 | 182 } |
cl349@4087 | 183 |
cl349@4087 | 184 static inline void __set_64bit_constant (unsigned long long *ptr, |
cl349@4087 | 185 unsigned long long value) |
cl349@4087 | 186 { |
cl349@4087 | 187 __set_64bit(ptr,(unsigned int)(value), (unsigned int)((value)>>32ULL)); |
cl349@4087 | 188 } |
cl349@4087 | 189 #define ll_low(x) *(((unsigned int*)&(x))+0) |
cl349@4087 | 190 #define ll_high(x) *(((unsigned int*)&(x))+1) |
cl349@4087 | 191 |
cl349@4087 | 192 static inline void __set_64bit_var (unsigned long long *ptr, |
cl349@4087 | 193 unsigned long long value) |
cl349@4087 | 194 { |
cl349@4087 | 195 __set_64bit(ptr,ll_low(value), ll_high(value)); |
cl349@4087 | 196 } |
cl349@4087 | 197 |
cl349@4087 | 198 #define set_64bit(ptr,value) \ |
cl349@4087 | 199 (__builtin_constant_p(value) ? \ |
cl349@4087 | 200 __set_64bit_constant(ptr, value) : \ |
cl349@4087 | 201 __set_64bit_var(ptr, value) ) |
cl349@4087 | 202 |
cl349@4087 | 203 #define _set_64bit(ptr,value) \ |
cl349@4087 | 204 (__builtin_constant_p(value) ? \ |
cl349@4087 | 205 __set_64bit(ptr, (unsigned int)(value), (unsigned int)((value)>>32ULL) ) : \ |
cl349@4087 | 206 __set_64bit(ptr, ll_low(value), ll_high(value)) ) |
cl349@4087 | 207 |
cl349@4087 | 208 /* |
cl349@4087 | 209 * Note: no "lock" prefix even on SMP: xchg always implies lock anyway |
cl349@4087 | 210 * Note 2: xchg has side effect, so that attribute volatile is necessary, |
cl349@4087 | 211 * but generally the primitive is invalid, *ptr is output argument. --ANK |
cl349@4087 | 212 */ |
cl349@4087 | 213 static inline unsigned long __xchg(unsigned long x, volatile void * ptr, int size) |
cl349@4087 | 214 { |
cl349@4087 | 215 switch (size) { |
cl349@4087 | 216 case 1: |
cl349@4087 | 217 __asm__ __volatile__("xchgb %b0,%1" |
cl349@4087 | 218 :"=q" (x) |
cl349@4087 | 219 :"m" (*__xg(ptr)), "0" (x) |
cl349@4087 | 220 :"memory"); |
cl349@4087 | 221 break; |
cl349@4087 | 222 case 2: |
cl349@4087 | 223 __asm__ __volatile__("xchgw %w0,%1" |
cl349@4087 | 224 :"=r" (x) |
cl349@4087 | 225 :"m" (*__xg(ptr)), "0" (x) |
cl349@4087 | 226 :"memory"); |
cl349@4087 | 227 break; |
cl349@4087 | 228 case 4: |
cl349@4087 | 229 __asm__ __volatile__("xchgl %0,%1" |
cl349@4087 | 230 :"=r" (x) |
cl349@4087 | 231 :"m" (*__xg(ptr)), "0" (x) |
cl349@4087 | 232 :"memory"); |
cl349@4087 | 233 break; |
cl349@4087 | 234 } |
cl349@4087 | 235 return x; |
cl349@4087 | 236 } |
cl349@4087 | 237 |
cl349@4087 | 238 /* |
cl349@4087 | 239 * Atomic compare and exchange. Compare OLD with MEM, if identical, |
cl349@4087 | 240 * store NEW in MEM. Return the initial value in MEM. Success is |
cl349@4087 | 241 * indicated by comparing RETURN with OLD. |
cl349@4087 | 242 */ |
cl349@4087 | 243 |
cl349@4087 | 244 #ifdef CONFIG_X86_CMPXCHG |
cl349@4087 | 245 #define __HAVE_ARCH_CMPXCHG 1 |
cl349@4087 | 246 #endif |
cl349@4087 | 247 |
cl349@4087 | 248 static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old, |
cl349@4087 | 249 unsigned long new, int size) |
cl349@4087 | 250 { |
cl349@4087 | 251 unsigned long prev; |
cl349@4087 | 252 switch (size) { |
cl349@4087 | 253 case 1: |
cl349@4087 | 254 __asm__ __volatile__(LOCK_PREFIX "cmpxchgb %b1,%2" |
cl349@4087 | 255 : "=a"(prev) |
cl349@4087 | 256 : "q"(new), "m"(*__xg(ptr)), "0"(old) |
cl349@4087 | 257 : "memory"); |
cl349@4087 | 258 return prev; |
cl349@4087 | 259 case 2: |
cl349@4087 | 260 __asm__ __volatile__(LOCK_PREFIX "cmpxchgw %w1,%2" |
cl349@4087 | 261 : "=a"(prev) |
cl349@4087 | 262 : "q"(new), "m"(*__xg(ptr)), "0"(old) |
cl349@4087 | 263 : "memory"); |
cl349@4087 | 264 return prev; |
cl349@4087 | 265 case 4: |
cl349@4087 | 266 __asm__ __volatile__(LOCK_PREFIX "cmpxchgl %1,%2" |
cl349@4087 | 267 : "=a"(prev) |
cl349@4087 | 268 : "q"(new), "m"(*__xg(ptr)), "0"(old) |
cl349@4087 | 269 : "memory"); |
cl349@4087 | 270 return prev; |
cl349@4087 | 271 } |
cl349@4087 | 272 return old; |
cl349@4087 | 273 } |
cl349@4087 | 274 |
cl349@4087 | 275 #define cmpxchg(ptr,o,n)\ |
cl349@4087 | 276 ((__typeof__(*(ptr)))__cmpxchg((ptr),(unsigned long)(o),\ |
cl349@4087 | 277 (unsigned long)(n),sizeof(*(ptr)))) |
cl349@4087 | 278 |
cl349@4087 | 279 #ifdef __KERNEL__ |
cl349@4087 | 280 struct alt_instr { |
cl349@4087 | 281 __u8 *instr; /* original instruction */ |
cl349@4087 | 282 __u8 *replacement; |
cl349@4087 | 283 __u8 cpuid; /* cpuid bit set for replacement */ |
cl349@4087 | 284 __u8 instrlen; /* length of original instruction */ |
cl349@4087 | 285 __u8 replacementlen; /* length of new instruction, <= instrlen */ |
cl349@4087 | 286 __u8 pad; |
cl349@4087 | 287 }; |
cl349@4087 | 288 #endif |
cl349@4087 | 289 |
cl349@4087 | 290 /* |
cl349@4087 | 291 * Alternative instructions for different CPU types or capabilities. |
cl349@4087 | 292 * |
cl349@4087 | 293 * This allows to use optimized instructions even on generic binary |
cl349@4087 | 294 * kernels. |
cl349@4087 | 295 * |
cl349@4087 | 296 * length of oldinstr must be longer or equal the length of newinstr |
cl349@4087 | 297 * It can be padded with nops as needed. |
cl349@4087 | 298 * |
cl349@4087 | 299 * For non barrier like inlines please define new variants |
cl349@4087 | 300 * without volatile and memory clobber. |
cl349@4087 | 301 */ |
cl349@4087 | 302 #define alternative(oldinstr, newinstr, feature) \ |
cl349@4087 | 303 asm volatile ("661:\n\t" oldinstr "\n662:\n" \ |
cl349@4087 | 304 ".section .altinstructions,\"a\"\n" \ |
cl349@4087 | 305 " .align 4\n" \ |
cl349@4087 | 306 " .long 661b\n" /* label */ \ |
cl349@4087 | 307 " .long 663f\n" /* new instruction */ \ |
cl349@4087 | 308 " .byte %c0\n" /* feature bit */ \ |
cl349@4087 | 309 " .byte 662b-661b\n" /* sourcelen */ \ |
cl349@4087 | 310 " .byte 664f-663f\n" /* replacementlen */ \ |
cl349@4087 | 311 ".previous\n" \ |
cl349@4087 | 312 ".section .altinstr_replacement,\"ax\"\n" \ |
cl349@4087 | 313 "663:\n\t" newinstr "\n664:\n" /* replacement */ \ |
cl349@4087 | 314 ".previous" :: "i" (feature) : "memory") |
cl349@4087 | 315 |
cl349@4087 | 316 /* |
cl349@4087 | 317 * Alternative inline assembly with input. |
cl349@4087 | 318 * |
cl349@4087 | 319 * Pecularities: |
cl349@4087 | 320 * No memory clobber here. |
cl349@4087 | 321 * Argument numbers start with 1. |
cl349@4087 | 322 * Best is to use constraints that are fixed size (like (%1) ... "r") |
cl349@4087 | 323 * If you use variable sized constraints like "m" or "g" in the |
cl349@4087 | 324 * replacement maake sure to pad to the worst case length. |
cl349@4087 | 325 */ |
cl349@4087 | 326 #define alternative_input(oldinstr, newinstr, feature, input...) \ |
cl349@4087 | 327 asm volatile ("661:\n\t" oldinstr "\n662:\n" \ |
cl349@4087 | 328 ".section .altinstructions,\"a\"\n" \ |
cl349@4087 | 329 " .align 4\n" \ |
cl349@4087 | 330 " .long 661b\n" /* label */ \ |
cl349@4087 | 331 " .long 663f\n" /* new instruction */ \ |
cl349@4087 | 332 " .byte %c0\n" /* feature bit */ \ |
cl349@4087 | 333 " .byte 662b-661b\n" /* sourcelen */ \ |
cl349@4087 | 334 " .byte 664f-663f\n" /* replacementlen */ \ |
cl349@4087 | 335 ".previous\n" \ |
cl349@4087 | 336 ".section .altinstr_replacement,\"ax\"\n" \ |
cl349@4087 | 337 "663:\n\t" newinstr "\n664:\n" /* replacement */ \ |
cl349@4087 | 338 ".previous" :: "i" (feature), ##input) |
cl349@4087 | 339 |
cl349@4087 | 340 /* |
cl349@4087 | 341 * Force strict CPU ordering. |
cl349@4087 | 342 * And yes, this is required on UP too when we're talking |
cl349@4087 | 343 * to devices. |
cl349@4087 | 344 * |
cl349@4087 | 345 * For now, "wmb()" doesn't actually do anything, as all |
cl349@4087 | 346 * Intel CPU's follow what Intel calls a *Processor Order*, |
cl349@4087 | 347 * in which all writes are seen in the program order even |
cl349@4087 | 348 * outside the CPU. |
cl349@4087 | 349 * |
cl349@4087 | 350 * I expect future Intel CPU's to have a weaker ordering, |
cl349@4087 | 351 * but I'd also expect them to finally get their act together |
cl349@4087 | 352 * and add some real memory barriers if so. |
cl349@4087 | 353 * |
cl349@4087 | 354 * Some non intel clones support out of order store. wmb() ceases to be a |
cl349@4087 | 355 * nop for these. |
cl349@4087 | 356 */ |
cl349@4087 | 357 |
cl349@4087 | 358 |
cl349@4087 | 359 /* |
cl349@4087 | 360 * Actually only lfence would be needed for mb() because all stores done |
cl349@4087 | 361 * by the kernel should be already ordered. But keep a full barrier for now. |
cl349@4087 | 362 */ |
cl349@4087 | 363 |
cl349@4087 | 364 #define mb() alternative("lock; addl $0,0(%%esp)", "mfence", X86_FEATURE_XMM2) |
cl349@4087 | 365 #define rmb() alternative("lock; addl $0,0(%%esp)", "lfence", X86_FEATURE_XMM2) |
cl349@4087 | 366 |
cl349@4087 | 367 /** |
cl349@4087 | 368 * read_barrier_depends - Flush all pending reads that subsequents reads |
cl349@4087 | 369 * depend on. |
cl349@4087 | 370 * |
cl349@4087 | 371 * No data-dependent reads from memory-like regions are ever reordered |
cl349@4087 | 372 * over this barrier. All reads preceding this primitive are guaranteed |
cl349@4087 | 373 * to access memory (but not necessarily other CPUs' caches) before any |
cl349@4087 | 374 * reads following this primitive that depend on the data return by |
cl349@4087 | 375 * any of the preceding reads. This primitive is much lighter weight than |
cl349@4087 | 376 * rmb() on most CPUs, and is never heavier weight than is |
cl349@4087 | 377 * rmb(). |
cl349@4087 | 378 * |
cl349@4087 | 379 * These ordering constraints are respected by both the local CPU |
cl349@4087 | 380 * and the compiler. |
cl349@4087 | 381 * |
cl349@4087 | 382 * Ordering is not guaranteed by anything other than these primitives, |
cl349@4087 | 383 * not even by data dependencies. See the documentation for |
cl349@4087 | 384 * memory_barrier() for examples and URLs to more information. |
cl349@4087 | 385 * |
cl349@4087 | 386 * For example, the following code would force ordering (the initial |
cl349@4087 | 387 * value of "a" is zero, "b" is one, and "p" is "&a"): |
cl349@4087 | 388 * |
cl349@4087 | 389 * <programlisting> |
cl349@4087 | 390 * CPU 0 CPU 1 |
cl349@4087 | 391 * |
cl349@4087 | 392 * b = 2; |
cl349@4087 | 393 * memory_barrier(); |
cl349@4087 | 394 * p = &b; q = p; |
cl349@4087 | 395 * read_barrier_depends(); |
cl349@4087 | 396 * d = *q; |
cl349@4087 | 397 * </programlisting> |
cl349@4087 | 398 * |
cl349@4087 | 399 * because the read of "*q" depends on the read of "p" and these |
cl349@4087 | 400 * two reads are separated by a read_barrier_depends(). However, |
cl349@4087 | 401 * the following code, with the same initial values for "a" and "b": |
cl349@4087 | 402 * |
cl349@4087 | 403 * <programlisting> |
cl349@4087 | 404 * CPU 0 CPU 1 |
cl349@4087 | 405 * |
cl349@4087 | 406 * a = 2; |
cl349@4087 | 407 * memory_barrier(); |
cl349@4087 | 408 * b = 3; y = b; |
cl349@4087 | 409 * read_barrier_depends(); |
cl349@4087 | 410 * x = a; |
cl349@4087 | 411 * </programlisting> |
cl349@4087 | 412 * |
cl349@4087 | 413 * does not enforce ordering, since there is no data dependency between |
cl349@4087 | 414 * the read of "a" and the read of "b". Therefore, on some CPUs, such |
cl349@4087 | 415 * as Alpha, "y" could be set to 3 and "x" to 0. Use rmb() |
cl349@4087 | 416 * in cases like thiswhere there are no data dependencies. |
cl349@4087 | 417 **/ |
cl349@4087 | 418 |
cl349@4087 | 419 #define read_barrier_depends() do { } while(0) |
cl349@4087 | 420 |
cl349@4087 | 421 #ifdef CONFIG_X86_OOSTORE |
cl349@4087 | 422 /* Actually there are no OOO store capable CPUs for now that do SSE, |
cl349@4087 | 423 but make it already an possibility. */ |
cl349@4087 | 424 #define wmb() alternative("lock; addl $0,0(%%esp)", "sfence", X86_FEATURE_XMM) |
cl349@4087 | 425 #else |
cl349@4087 | 426 #define wmb() __asm__ __volatile__ ("": : :"memory") |
cl349@4087 | 427 #endif |
cl349@4087 | 428 |
cl349@4087 | 429 #ifdef CONFIG_SMP |
cl349@4087 | 430 #define smp_mb() mb() |
cl349@4087 | 431 #define smp_rmb() rmb() |
cl349@4087 | 432 #define smp_wmb() wmb() |
cl349@4087 | 433 #define smp_read_barrier_depends() read_barrier_depends() |
cl349@4087 | 434 #define set_mb(var, value) do { xchg(&var, value); } while (0) |
cl349@4087 | 435 #else |
cl349@4087 | 436 #define smp_mb() barrier() |
cl349@4087 | 437 #define smp_rmb() barrier() |
cl349@4087 | 438 #define smp_wmb() barrier() |
cl349@4087 | 439 #define smp_read_barrier_depends() do { } while(0) |
cl349@4087 | 440 #define set_mb(var, value) do { var = value; barrier(); } while (0) |
cl349@4087 | 441 #endif |
cl349@4087 | 442 |
cl349@4087 | 443 #define set_wmb(var, value) do { var = value; wmb(); } while (0) |
cl349@4087 | 444 |
cl349@4087 | 445 /* interrupt control.. */ |
cl349@4087 | 446 |
cl349@4087 | 447 /* |
cl349@4087 | 448 * The use of 'barrier' in the following reflects their use as local-lock |
cl349@4087 | 449 * operations. Reentrancy must be prevented (e.g., __cli()) /before/ following |
cl349@4684 | 450 * critical operations are executed. All critical operations must complete |
cl349@4087 | 451 * /before/ reentrancy is permitted (e.g., __sti()). Alpha architecture also |
cl349@4087 | 452 * includes these barriers, for example. |
cl349@4087 | 453 */ |
cl349@4087 | 454 |
cl349@4087 | 455 #define __cli() \ |
cl349@4087 | 456 do { \ |
cl349@4112 | 457 vcpu_info_t *_vcpu; \ |
cl349@4112 | 458 preempt_disable(); \ |
cl349@4112 | 459 _vcpu = &HYPERVISOR_shared_info->vcpu_data[smp_processor_id()]; \ |
cl349@4112 | 460 _vcpu->evtchn_upcall_mask = 1; \ |
cl349@4112 | 461 preempt_enable_no_resched(); \ |
cl349@4087 | 462 barrier(); \ |
cl349@4087 | 463 } while (0) |
cl349@4087 | 464 |
cl349@4087 | 465 #define __sti() \ |
cl349@4087 | 466 do { \ |
cl349@4112 | 467 vcpu_info_t *_vcpu; \ |
cl349@4087 | 468 barrier(); \ |
cl349@4112 | 469 preempt_disable(); \ |
cl349@4112 | 470 _vcpu = &HYPERVISOR_shared_info->vcpu_data[smp_processor_id()]; \ |
cl349@4112 | 471 _vcpu->evtchn_upcall_mask = 0; \ |
cl349@4087 | 472 barrier(); /* unmask then check (avoid races) */ \ |
cl349@4112 | 473 if ( unlikely(_vcpu->evtchn_upcall_pending) ) \ |
cl349@4112 | 474 force_evtchn_callback(); \ |
cl349@4112 | 475 preempt_enable(); \ |
cl349@4087 | 476 } while (0) |
cl349@4087 | 477 |
cl349@4087 | 478 #define __save_flags(x) \ |
cl349@4087 | 479 do { \ |
cl349@4112 | 480 vcpu_info_t *_vcpu; \ |
cl349@4112 | 481 _vcpu = &HYPERVISOR_shared_info->vcpu_data[smp_processor_id()]; \ |
cl349@4112 | 482 (x) = _vcpu->evtchn_upcall_mask; \ |
cl349@4087 | 483 } while (0) |
cl349@4087 | 484 |
cl349@4087 | 485 #define __restore_flags(x) \ |
cl349@4087 | 486 do { \ |
cl349@4112 | 487 vcpu_info_t *_vcpu; \ |
cl349@4087 | 488 barrier(); \ |
cl349@4112 | 489 preempt_disable(); \ |
cl349@4112 | 490 _vcpu = &HYPERVISOR_shared_info->vcpu_data[smp_processor_id()]; \ |
cl349@4112 | 491 if ((_vcpu->evtchn_upcall_mask = (x)) == 0) { \ |
cl349@4112 | 492 barrier(); /* unmask then check (avoid races) */ \ |
cl349@4112 | 493 if ( unlikely(_vcpu->evtchn_upcall_pending) ) \ |
cl349@4112 | 494 force_evtchn_callback(); \ |
cl349@4112 | 495 preempt_enable(); \ |
cl349@4112 | 496 } else \ |
cl349@4112 | 497 preempt_enable_no_resched(); \ |
cl349@4087 | 498 } while (0) |
cl349@4087 | 499 |
cl349@4112 | 500 #define safe_halt() ((void)0) |
cl349@4087 | 501 |
cl349@4087 | 502 #define __save_and_cli(x) \ |
cl349@4087 | 503 do { \ |
cl349@4112 | 504 vcpu_info_t *_vcpu; \ |
cl349@4112 | 505 preempt_disable(); \ |
cl349@4112 | 506 _vcpu = &HYPERVISOR_shared_info->vcpu_data[smp_processor_id()]; \ |
cl349@4112 | 507 (x) = _vcpu->evtchn_upcall_mask; \ |
cl349@4112 | 508 _vcpu->evtchn_upcall_mask = 1; \ |
cl349@4112 | 509 preempt_enable_no_resched(); \ |
cl349@4087 | 510 barrier(); \ |
cl349@4087 | 511 } while (0) |
cl349@4087 | 512 |
cl349@4087 | 513 #define local_irq_save(x) __save_and_cli(x) |
cl349@4087 | 514 #define local_irq_restore(x) __restore_flags(x) |
cl349@4087 | 515 #define local_save_flags(x) __save_flags(x) |
cl349@4087 | 516 #define local_irq_disable() __cli() |
cl349@4087 | 517 #define local_irq_enable() __sti() |
cl349@4087 | 518 |
cl349@4112 | 519 #define irqs_disabled() \ |
cl349@4112 | 520 HYPERVISOR_shared_info->vcpu_data[smp_processor_id()].evtchn_upcall_mask |
cl349@4087 | 521 |
cl349@4087 | 522 /* |
cl349@4087 | 523 * disable hlt during certain critical i/o operations |
cl349@4087 | 524 */ |
cl349@4087 | 525 #define HAVE_DISABLE_HLT |
cl349@4087 | 526 void disable_hlt(void); |
cl349@4087 | 527 void enable_hlt(void); |
cl349@4087 | 528 |
cl349@4087 | 529 extern int es7000_plat; |
cl349@4087 | 530 void cpu_idle_wait(void); |
cl349@4087 | 531 |
cl349@4087 | 532 #endif |