| 1 |
/* Linux-specific atomic operations for PA Linux. |
| 2 |
Copyright (C) 2008-2025 Free Software Foundation, Inc. |
| 3 |
Based on code contributed by CodeSourcery for ARM EABI Linux. |
| 4 |
Modifications for PA Linux by Helge Deller <[email protected]> |
| 5 |
|
| 6 |
This file is part of GCC. |
| 7 |
|
| 8 |
GCC is free software; you can redistribute it and/or modify it under |
| 9 |
the terms of the GNU General Public License as published by the Free |
| 10 |
Software Foundation; either version 3, or (at your option) any later |
| 11 |
version. |
| 12 |
|
| 13 |
GCC is distributed in the hope that it will be useful, but WITHOUT ANY |
| 14 |
WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 15 |
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 16 |
for more details. |
| 17 |
|
| 18 |
Under Section 7 of GPL version 3, you are granted additional |
| 19 |
permissions described in the GCC Runtime Library Exception, version |
| 20 |
3.1, as published by the Free Software Foundation. |
| 21 |
|
| 22 |
You should have received a copy of the GNU General Public License and |
| 23 |
a copy of the GCC Runtime Library Exception along with this program; |
| 24 |
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see |
| 25 |
<http://www.gnu.org/licenses/>. */ |
| 26 |
|
| 27 |
#define EFAULT 14 |
| 28 |
#define EBUSY 16 |
| 29 |
#define ENOSYS 251 |
| 30 |
|
| 31 |
#define _ASM_EFAULT "-14" |
| 32 |
|
| 33 |
typedef unsigned char u8; |
| 34 |
typedef short unsigned int u16; |
| 35 |
typedef unsigned int u32; |
| 36 |
#ifdef __LP64__ |
| 37 |
typedef long unsigned int u64; |
| 38 |
#else |
| 39 |
typedef long long unsigned int u64; |
| 40 |
#endif |
| 41 |
|
| 42 |
/* PA-RISC 2.0 supports out-of-order execution for loads and stores. |
| 43 |
Thus, we need to synchonize memory accesses. For more info, see: |
| 44 |
"Advanced Performance Features of the 64-bit PA-8000" by Doug Hunt. |
| 45 |
|
| 46 |
We implement byte, short and int versions of each atomic operation |
| 47 |
using the kernel helper defined below. There is no support for |
| 48 |
64-bit operations yet. */ |
| 49 |
|
| 50 |
/* Determine kernel LWS function call (0=32-bit, 1=64-bit userspace). */ |
| 51 |
#define LWS_CAS (sizeof(long) == 4 ? 0 : 1) |
| 52 |
|
| 53 |
/* Kernel helper for compare-and-exchange a 32-bit value. */ |
| 54 |
static inline long |
| 55 |
__kernel_cmpxchg (volatile void *mem, int oldval, int newval) |
| 56 |
{ |
| 57 |
register unsigned long lws_mem asm("r26") = (unsigned long) (mem); |
| 58 |
register int lws_old asm("r25") = oldval; |
| 59 |
register int lws_new asm("r24") = newval; |
| 60 |
register long lws_ret asm("r28"); |
| 61 |
register long lws_errno asm("r21"); |
| 62 |
asm volatile ( "ble 0xb0(%%sr2, %%r0) \n\t" |
| 63 |
"ldi %2, %%r20 \n\t" |
| 64 |
"cmpiclr,<> " _ASM_EFAULT ", %%r21, %%r0\n\t" |
| 65 |
"iitlbp %%r0,(%%sr0, %%r0) \n\t" |
| 66 |
: "=r" (lws_ret), "=r" (lws_errno) |
| 67 |
: "i" (LWS_CAS), "r" (lws_mem), "r" (lws_old), "r" (lws_new) |
| 68 |
: "r1", "r20", "r22", "r23", "r29", "r31", "memory" |
| 69 |
); |
| 70 |
|
| 71 |
/* If the kernel LWS call succeeded (lws_errno == 0), lws_ret contains |
| 72 |
the old value from memory. If this value is equal to OLDVAL, the |
| 73 |
new value was written to memory. If not, return -EBUSY. */ |
| 74 |
if (!lws_errno && lws_ret != oldval) |
| 75 |
return -EBUSY; |
| 76 |
|
| 77 |
return lws_errno; |
| 78 |
} |
| 79 |
|
| 80 |
static inline long |
| 81 |
__kernel_cmpxchg2 (volatile void *mem, const void *oldval, const void *newval, |
| 82 |
int val_size) |
| 83 |
{ |
| 84 |
register unsigned long lws_mem asm("r26") = (unsigned long) (mem); |
| 85 |
register unsigned long lws_old asm("r25") = (unsigned long) oldval; |
| 86 |
register unsigned long lws_new asm("r24") = (unsigned long) newval; |
| 87 |
register int lws_size asm("r23") = val_size; |
| 88 |
register long lws_ret asm("r28"); |
| 89 |
register long lws_errno asm("r21"); |
| 90 |
asm volatile ( "ble 0xb0(%%sr2, %%r0) \n\t" |
| 91 |
"ldi %6, %%r20 \n\t" |
| 92 |
"cmpiclr,<> " _ASM_EFAULT ", %%r21, %%r0\n\t" |
| 93 |
"iitlbp %%r0,(%%sr0, %%r0) \n\t" |
| 94 |
: "=r" (lws_ret), "=r" (lws_errno), "+r" (lws_mem), |
| 95 |
"+r" (lws_old), "+r" (lws_new), "+r" (lws_size) |
| 96 |
: "i" (2) |
| 97 |
: "r1", "r20", "r22", "r29", "r31", "fr4", "memory" |
| 98 |
); |
| 99 |
|
| 100 |
/* If the kernel LWS call is successful, lws_ret contains 0. */ |
| 101 |
if (__builtin_expect (lws_ret == 0, 1)) |
| 102 |
return 0; |
| 103 |
|
| 104 |
/* If the kernel LWS call fails with no error, return -EBUSY */ |
| 105 |
if (__builtin_expect (!lws_errno, 0)) |
| 106 |
return -EBUSY; |
| 107 |
|
| 108 |
return lws_errno; |
| 109 |
} |
| 110 |
#define HIDDEN __attribute__ ((visibility ("hidden"))) |
| 111 |
|
| 112 |
/* Big endian masks */ |
| 113 |
#define INVERT_MASK_1 24 |
| 114 |
#define INVERT_MASK_2 16 |
| 115 |
|
| 116 |
#define MASK_1 0xffu |
| 117 |
#define MASK_2 0xffffu |
| 118 |
|
| 119 |
/* Load value with an atomic processor load if possible. */ |
| 120 |
#define ATOMIC_LOAD(TYPE, WIDTH) \ |
| 121 |
static inline TYPE \ |
| 122 |
atomic_load_##WIDTH (volatile void *ptr) \ |
| 123 |
{ \ |
| 124 |
return *(volatile TYPE *)ptr; \ |
| 125 |
} |
| 126 |
|
| 127 |
#if defined(__LP64__) || defined(__SOFTFP__) |
| 128 |
ATOMIC_LOAD (u64, 8) |
| 129 |
#else |
| 130 |
static inline u64 |
| 131 |
atomic_load_8 (volatile void *ptr) |
| 132 |
{ |
| 133 |
u64 result; |
| 134 |
double tmp; |
| 135 |
|
| 136 |
asm volatile ("{fldds|fldd} 0(%2),%1\n\t" |
| 137 |
"{fstds|fstd} %1,-16(%%sp)\n\t" |
| 138 |
"{ldws|ldw} -16(%%sp),%0\n\t" |
| 139 |
"{ldws|ldw} -12(%%sp),%R0" |
| 140 |
: "=r" (result), "=f" (tmp) : "r" (ptr): "memory"); |
| 141 |
return result; |
| 142 |
} |
| 143 |
#endif |
| 144 |
|
| 145 |
ATOMIC_LOAD (u32, 4) |
| 146 |
ATOMIC_LOAD (u16, 2) |
| 147 |
ATOMIC_LOAD (u8, 1) |
| 148 |
|
| 149 |
#define FETCH_AND_OP_2(OP, PFX_OP, INF_OP, TYPE, WIDTH, INDEX) \ |
| 150 |
TYPE HIDDEN \ |
| 151 |
__sync_fetch_and_##OP##_##WIDTH (volatile void *ptr, TYPE val) \ |
| 152 |
{ \ |
| 153 |
TYPE tmp, newval; \ |
| 154 |
long failure; \ |
| 155 |
\ |
| 156 |
do { \ |
| 157 |
tmp = atomic_load_##WIDTH ((volatile TYPE *)ptr); \ |
| 158 |
newval = PFX_OP (tmp INF_OP val); \ |
| 159 |
failure = __kernel_cmpxchg2 (ptr, &tmp, &newval, INDEX); \ |
| 160 |
} while (failure != 0); \ |
| 161 |
\ |
| 162 |
return tmp; \ |
| 163 |
} |
| 164 |
|
| 165 |
FETCH_AND_OP_2 (add, , +, u64, 8, 3) |
| 166 |
FETCH_AND_OP_2 (sub, , -, u64, 8, 3) |
| 167 |
FETCH_AND_OP_2 (or, , |, u64, 8, 3) |
| 168 |
FETCH_AND_OP_2 (and, , &, u64, 8, 3) |
| 169 |
FETCH_AND_OP_2 (xor, , ^, u64, 8, 3) |
| 170 |
FETCH_AND_OP_2 (nand, ~, &, u64, 8, 3) |
| 171 |
|
| 172 |
FETCH_AND_OP_2 (add, , +, u16, 2, 1) |
| 173 |
FETCH_AND_OP_2 (sub, , -, u16, 2, 1) |
| 174 |
FETCH_AND_OP_2 (or, , |, u16, 2, 1) |
| 175 |
FETCH_AND_OP_2 (and, , &, u16, 2, 1) |
| 176 |
FETCH_AND_OP_2 (xor, , ^, u16, 2, 1) |
| 177 |
FETCH_AND_OP_2 (nand, ~, &, u16, 2, 1) |
| 178 |
|
| 179 |
FETCH_AND_OP_2 (add, , +, u8, 1, 0) |
| 180 |
FETCH_AND_OP_2 (sub, , -, u8, 1, 0) |
| 181 |
FETCH_AND_OP_2 (or, , |, u8, 1, 0) |
| 182 |
FETCH_AND_OP_2 (and, , &, u8, 1, 0) |
| 183 |
FETCH_AND_OP_2 (xor, , ^, u8, 1, 0) |
| 184 |
FETCH_AND_OP_2 (nand, ~, &, u8, 1, 0) |
| 185 |
|
| 186 |
#define OP_AND_FETCH_2(OP, PFX_OP, INF_OP, TYPE, WIDTH, INDEX) \ |
| 187 |
TYPE HIDDEN \ |
| 188 |
__sync_##OP##_and_fetch_##WIDTH (volatile void *ptr, TYPE val) \ |
| 189 |
{ \ |
| 190 |
TYPE tmp, newval; \ |
| 191 |
long failure; \ |
| 192 |
\ |
| 193 |
do { \ |
| 194 |
tmp = atomic_load_##WIDTH ((volatile TYPE *)ptr); \ |
| 195 |
newval = PFX_OP (tmp INF_OP val); \ |
| 196 |
failure = __kernel_cmpxchg2 (ptr, &tmp, &newval, INDEX); \ |
| 197 |
} while (failure != 0); \ |
| 198 |
\ |
| 199 |
return PFX_OP (tmp INF_OP val); \ |
| 200 |
} |
| 201 |
|
| 202 |
OP_AND_FETCH_2 (add, , +, u64, 8, 3) |
| 203 |
OP_AND_FETCH_2 (sub, , -, u64, 8, 3) |
| 204 |
OP_AND_FETCH_2 (or, , |, u64, 8, 3) |
| 205 |
OP_AND_FETCH_2 (and, , &, u64, 8, 3) |
| 206 |
OP_AND_FETCH_2 (xor, , ^, u64, 8, 3) |
| 207 |
OP_AND_FETCH_2 (nand, ~, &, u64, 8, 3) |
| 208 |
|
| 209 |
OP_AND_FETCH_2 (add, , +, u16, 2, 1) |
| 210 |
OP_AND_FETCH_2 (sub, , -, u16, 2, 1) |
| 211 |
OP_AND_FETCH_2 (or, , |, u16, 2, 1) |
| 212 |
OP_AND_FETCH_2 (and, , &, u16, 2, 1) |
| 213 |
OP_AND_FETCH_2 (xor, , ^, u16, 2, 1) |
| 214 |
OP_AND_FETCH_2 (nand, ~, &, u16, 2, 1) |
| 215 |
|
| 216 |
OP_AND_FETCH_2 (add, , +, u8, 1, 0) |
| 217 |
OP_AND_FETCH_2 (sub, , -, u8, 1, 0) |
| 218 |
OP_AND_FETCH_2 (or, , |, u8, 1, 0) |
| 219 |
OP_AND_FETCH_2 (and, , &, u8, 1, 0) |
| 220 |
OP_AND_FETCH_2 (xor, , ^, u8, 1, 0) |
| 221 |
OP_AND_FETCH_2 (nand, ~, &, u8, 1, 0) |
| 222 |
|
| 223 |
#define FETCH_AND_OP_WORD(OP, PFX_OP, INF_OP) \ |
| 224 |
unsigned int HIDDEN \ |
| 225 |
__sync_fetch_and_##OP##_4 (volatile void *ptr, unsigned int val) \ |
| 226 |
{ \ |
| 227 |
unsigned int tmp; \ |
| 228 |
long failure; \ |
| 229 |
\ |
| 230 |
do { \ |
| 231 |
tmp = atomic_load_4 ((volatile unsigned int *)ptr); \ |
| 232 |
failure = __kernel_cmpxchg (ptr, tmp, PFX_OP (tmp INF_OP val)); \ |
| 233 |
} while (failure != 0); \ |
| 234 |
\ |
| 235 |
return tmp; \ |
| 236 |
} |
| 237 |
|
| 238 |
FETCH_AND_OP_WORD (add, , +) |
| 239 |
FETCH_AND_OP_WORD (sub, , -) |
| 240 |
FETCH_AND_OP_WORD (or, , |) |
| 241 |
FETCH_AND_OP_WORD (and, , &) |
| 242 |
FETCH_AND_OP_WORD (xor, , ^) |
| 243 |
FETCH_AND_OP_WORD (nand, ~, &) |
| 244 |
|
| 245 |
#define OP_AND_FETCH_WORD(OP, PFX_OP, INF_OP) \ |
| 246 |
unsigned int HIDDEN \ |
| 247 |
__sync_##OP##_and_fetch_4 (volatile void *ptr, unsigned int val) \ |
| 248 |
{ \ |
| 249 |
unsigned int tmp; \ |
| 250 |
long failure; \ |
| 251 |
\ |
| 252 |
do { \ |
| 253 |
tmp = atomic_load_4 ((volatile unsigned int *)ptr); \ |
| 254 |
failure = __kernel_cmpxchg (ptr, tmp, PFX_OP (tmp INF_OP val)); \ |
| 255 |
} while (failure != 0); \ |
| 256 |
\ |
| 257 |
return PFX_OP (tmp INF_OP val); \ |
| 258 |
} |
| 259 |
|
| 260 |
OP_AND_FETCH_WORD (add, , +) |
| 261 |
OP_AND_FETCH_WORD (sub, , -) |
| 262 |
OP_AND_FETCH_WORD (or, , |) |
| 263 |
OP_AND_FETCH_WORD (and, , &) |
| 264 |
OP_AND_FETCH_WORD (xor, , ^) |
| 265 |
OP_AND_FETCH_WORD (nand, ~, &) |
| 266 |
|
| 267 |
#define COMPARE_AND_SWAP_2(TYPE, WIDTH, INDEX) \ |
| 268 |
TYPE HIDDEN \ |
| 269 |
__sync_val_compare_and_swap_##WIDTH (volatile void *ptr, TYPE oldval, \ |
| 270 |
TYPE newval) \ |
| 271 |
{ \ |
| 272 |
TYPE actual_oldval; \ |
| 273 |
long fail; \ |
| 274 |
\ |
| 275 |
while (1) \ |
| 276 |
{ \ |
| 277 |
actual_oldval = atomic_load_##WIDTH ((volatile TYPE *)ptr); \ |
| 278 |
\ |
| 279 |
if (__builtin_expect (oldval != actual_oldval, 0)) \ |
| 280 |
return actual_oldval; \ |
| 281 |
\ |
| 282 |
fail = __kernel_cmpxchg2 (ptr, &actual_oldval, &newval, INDEX); \ |
| 283 |
\ |
| 284 |
if (__builtin_expect (!fail, 1)) \ |
| 285 |
return actual_oldval; \ |
| 286 |
} \ |
| 287 |
} \ |
| 288 |
\ |
| 289 |
_Bool HIDDEN \ |
| 290 |
__sync_bool_compare_and_swap_##WIDTH (volatile void *ptr, \ |
| 291 |
TYPE oldval, TYPE newval) \ |
| 292 |
{ \ |
| 293 |
long failure = __kernel_cmpxchg2 (ptr, &oldval, &newval, INDEX); \ |
| 294 |
return (failure == 0); \ |
| 295 |
} |
| 296 |
|
| 297 |
COMPARE_AND_SWAP_2 (u64, 8, 3) |
| 298 |
COMPARE_AND_SWAP_2 (u16, 2, 1) |
| 299 |
COMPARE_AND_SWAP_2 (u8, 1, 0) |
| 300 |
|
| 301 |
unsigned int HIDDEN |
| 302 |
__sync_val_compare_and_swap_4 (volatile void *ptr, unsigned int oldval, |
| 303 |
unsigned int newval) |
| 304 |
{ |
| 305 |
long fail; |
| 306 |
unsigned int actual_oldval; |
| 307 |
|
| 308 |
while (1) |
| 309 |
{ |
| 310 |
actual_oldval = atomic_load_4 ((volatile unsigned int *)ptr); |
| 311 |
|
| 312 |
if (__builtin_expect (oldval != actual_oldval, 0)) |
| 313 |
return actual_oldval; |
| 314 |
|
| 315 |
fail = __kernel_cmpxchg (ptr, actual_oldval, newval); |
| 316 |
|
| 317 |
if (__builtin_expect (!fail, 1)) |
| 318 |
return actual_oldval; |
| 319 |
} |
| 320 |
} |
| 321 |
|
| 322 |
_Bool HIDDEN |
| 323 |
__sync_bool_compare_and_swap_4 (volatile void *ptr, unsigned int oldval, |
| 324 |
unsigned int newval) |
| 325 |
{ |
| 326 |
long failure = __kernel_cmpxchg (ptr, oldval, newval); |
| 327 |
return (failure == 0); |
| 328 |
} |
| 329 |
|
| 330 |
#define SYNC_LOCK_TEST_AND_SET_2(TYPE, WIDTH, INDEX) \ |
| 331 |
TYPE HIDDEN \ |
| 332 |
__sync_lock_test_and_set_##WIDTH (volatile void *ptr, TYPE val) \ |
| 333 |
{ \ |
| 334 |
TYPE oldval; \ |
| 335 |
long failure; \ |
| 336 |
\ |
| 337 |
do { \ |
| 338 |
oldval = atomic_load_##WIDTH ((volatile TYPE *)ptr); \ |
| 339 |
failure = __kernel_cmpxchg2 (ptr, &oldval, &val, INDEX); \ |
| 340 |
} while (failure != 0); \ |
| 341 |
\ |
| 342 |
return oldval; \ |
| 343 |
} |
| 344 |
|
| 345 |
SYNC_LOCK_TEST_AND_SET_2 (u64, 8, 3) |
| 346 |
SYNC_LOCK_TEST_AND_SET_2 (u16, 2, 1) |
| 347 |
SYNC_LOCK_TEST_AND_SET_2 (u8, 1, 0) |
| 348 |
|
| 349 |
u32 HIDDEN |
| 350 |
__sync_lock_test_and_set_4 (volatile void *ptr, unsigned int val) |
| 351 |
{ |
| 352 |
long failure; |
| 353 |
unsigned int oldval; |
| 354 |
|
| 355 |
do { |
| 356 |
oldval = atomic_load_4 ((volatile unsigned int *)ptr); |
| 357 |
failure = __kernel_cmpxchg (ptr, oldval, val); |
| 358 |
} while (failure != 0); |
| 359 |
|
| 360 |
return oldval; |
| 361 |
} |
| 362 |
|
| 363 |
#define SYNC_LOCK_RELEASE_1(TYPE, WIDTH, INDEX) \ |
| 364 |
void HIDDEN \ |
| 365 |
__sync_lock_release_##WIDTH (volatile void *ptr) \ |
| 366 |
{ \ |
| 367 |
TYPE oldval, val = 0; \ |
| 368 |
long failure; \ |
| 369 |
\ |
| 370 |
do { \ |
| 371 |
oldval = atomic_load_##WIDTH ((volatile TYPE *)ptr); \ |
| 372 |
failure = __kernel_cmpxchg2 (ptr, &oldval, &val, INDEX); \ |
| 373 |
} while (failure != 0); \ |
| 374 |
} |
| 375 |
|
| 376 |
SYNC_LOCK_RELEASE_1 (u64, 8, 3) |
| 377 |
SYNC_LOCK_RELEASE_1 (u16, 2, 1) |
| 378 |
SYNC_LOCK_RELEASE_1 (u8, 1, 0) |
| 379 |
|
| 380 |
void HIDDEN |
| 381 |
__sync_lock_release_4 (volatile void *ptr) |
| 382 |
{ |
| 383 |
long failure; |
| 384 |
unsigned int oldval; |
| 385 |
|
| 386 |
do { |
| 387 |
oldval = atomic_load_4 ((volatile unsigned int *)ptr); |
| 388 |
failure = __kernel_cmpxchg (ptr, oldval, 0); |
| 389 |
} while (failure != 0); |
| 390 |
} |
| 391 |
|
| 392 |
#ifndef __LP64__ |
| 393 |
#define SYNC_LOCK_LOAD_2(TYPE, WIDTH, INDEX) \ |
| 394 |
TYPE __sync_lock_load_##WIDTH (volatile void *) HIDDEN; \ |
| 395 |
TYPE \ |
| 396 |
__sync_lock_load_##WIDTH (volatile void *ptr) \ |
| 397 |
{ \ |
| 398 |
TYPE oldval; \ |
| 399 |
long failure; \ |
| 400 |
\ |
| 401 |
do { \ |
| 402 |
oldval = atomic_load_##WIDTH ((volatile TYPE *)ptr); \ |
| 403 |
failure = __kernel_cmpxchg2 (ptr, &oldval, &oldval, INDEX); \ |
| 404 |
} while (failure != 0); \ |
| 405 |
\ |
| 406 |
return oldval; \ |
| 407 |
} |
| 408 |
|
| 409 |
SYNC_LOCK_LOAD_2 (u64, 8, 3) |
| 410 |
#endif |