1/* 2 * This file contains the light-weight system call handlers (fsyscall-handlers). 3 * 4 * Copyright (C) 2003 Hewlett-Packard Co 5 * David Mosberger-Tang <davidm@hpl.hp.com> 6 * 7 * 25-Sep-03 davidm Implement fsys_rt_sigprocmask(). 8 * 18-Feb-03 louisk Implement fsys_gettimeofday(). 9 * 28-Feb-03 davidm Fixed several bugs in fsys_gettimeofday(). Tuned it some more, 10 * probably broke it along the way... ;-) 11 * 13-Jul-04 clameter Implement fsys_clock_gettime and revise fsys_gettimeofday to make 12 * it capable of using memory based clocks without falling back to C code. 13 * 08-Feb-07 Fenghua Yu Implement fsys_getcpu. 14 * 15 */ 16 17#include <asm/asmmacro.h> 18#include <asm/errno.h> 19#include <asm/asm-offsets.h> 20#include <asm/percpu.h> 21#include <asm/thread_info.h> 22#include <asm/sal.h> 23#include <asm/signal.h> 24#include <asm/system.h> 25#include <asm/unistd.h> 26 27#include "entry.h" 28 29/* 30 * See Documentation/ia64/fsys.txt for details on fsyscalls. 31 * 32 * On entry to an fsyscall handler: 33 * r10 = 0 (i.e., defaults to "successful syscall return") 34 * r11 = saved ar.pfs (a user-level value) 35 * r15 = system call number 36 * r16 = "current" task pointer (in normal kernel-mode, this is in r13) 37 * r32-r39 = system call arguments 38 * b6 = return address (a user-level value) 39 * ar.pfs = previous frame-state (a user-level value) 40 * PSR.be = cleared to zero (i.e., little-endian byte order is in effect) 41 * all other registers may contain values passed in from user-mode 42 * 43 * On return from an fsyscall handler: 44 * r11 = saved ar.pfs (as passed into the fsyscall handler) 45 * r15 = system call number (as passed into the fsyscall handler) 46 * r32-r39 = system call arguments (as passed into the fsyscall handler) 47 * b6 = return address (as passed into the fsyscall handler) 48 * ar.pfs = previous frame-state (as passed into the fsyscall handler) 49 */ 50 51ENTRY(fsys_ni_syscall) 52 .prologue 53 .altrp b6 54 .body 55 mov r8=ENOSYS 56 mov r10=-1 57 FSYS_RETURN 58END(fsys_ni_syscall) 59 60ENTRY(fsys_getpid) 61 .prologue 62 .altrp b6 63 .body 64 add r9=TI_FLAGS+IA64_TASK_SIZE,r16 65 ;; 66 ld4 r9=[r9] 67 add r8=IA64_TASK_TGID_OFFSET,r16 68 ;; 69 and r9=TIF_ALLWORK_MASK,r9 70 ld4 r8=[r8] // r8 = current->tgid 71 ;; 72 cmp.ne p8,p0=0,r9 73(p8) br.spnt.many fsys_fallback_syscall 74 FSYS_RETURN 75END(fsys_getpid) 76 77ENTRY(fsys_getppid) 78 .prologue 79 .altrp b6 80 .body 81 add r17=IA64_TASK_GROUP_LEADER_OFFSET,r16 82 ;; 83 ld8 r17=[r17] // r17 = current->group_leader 84 add r9=TI_FLAGS+IA64_TASK_SIZE,r16 85 ;; 86 87 ld4 r9=[r9] 88 add r17=IA64_TASK_REAL_PARENT_OFFSET,r17 // r17 = ¤t->group_leader->real_parent 89 ;; 90 and r9=TIF_ALLWORK_MASK,r9 91 921: ld8 r18=[r17] // r18 = current->group_leader->real_parent 93 ;; 94 cmp.ne p8,p0=0,r9 95 add r8=IA64_TASK_TGID_OFFSET,r18 // r8 = ¤t->group_leader->real_parent->tgid 96 ;; 97 98 /* 99 * The .acq is needed to ensure that the read of tgid has returned its data before 100 * we re-check "real_parent". 101 */ 102 ld4.acq r8=[r8] // r8 = current->group_leader->real_parent->tgid 103#ifdef CONFIG_SMP 104 /* 105 * Re-read current->group_leader->real_parent. 106 */ 107 ld8 r19=[r17] // r19 = current->group_leader->real_parent 108(p8) br.spnt.many fsys_fallback_syscall 109 ;; 110 cmp.ne p6,p0=r18,r19 // did real_parent change? 111 mov r19=0 // i must not leak kernel bits... 112(p6) br.cond.spnt.few 1b // yes -> redo the read of tgid and the check 113 ;; 114 mov r17=0 // i must not leak kernel bits... 115 mov r18=0 // i must not leak kernel bits... 116#else 117 mov r17=0 // i must not leak kernel bits... 118 mov r18=0 // i must not leak kernel bits... 119 mov r19=0 // i must not leak kernel bits... 120#endif 121 FSYS_RETURN 122END(fsys_getppid) 123 124ENTRY(fsys_set_tid_address) 125 .prologue 126 .altrp b6 127 .body 128 add r9=TI_FLAGS+IA64_TASK_SIZE,r16 129 ;; 130 ld4 r9=[r9] 131 tnat.z p6,p7=r32 // check argument register for being NaT 132 ;; 133 and r9=TIF_ALLWORK_MASK,r9 134 add r8=IA64_TASK_PID_OFFSET,r16 135 add r18=IA64_TASK_CLEAR_CHILD_TID_OFFSET,r16 136 ;; 137 ld4 r8=[r8] 138 cmp.ne p8,p0=0,r9 139 mov r17=-1 140 ;; 141(p6) st8 [r18]=r32 142(p7) st8 [r18]=r17 143(p8) br.spnt.many fsys_fallback_syscall 144 ;; 145 mov r17=0 // i must not leak kernel bits... 146 mov r18=0 // i must not leak kernel bits... 147 FSYS_RETURN 148END(fsys_set_tid_address) 149 150/* 151 * Ensure that the time interpolator structure is compatible with the asm code 152 */ 153#if IA64_TIME_INTERPOLATOR_SOURCE_OFFSET !=IA64_TIME_INTERPOLATOR_SHIFT_OFFSET != 2 || \ 154 IA64_TIME_INTERPOLATOR_JITTER_OFFSET != 3 || IA64_TIME_INTERPOLATOR_NSEC_OFFSET != 4 155#error fsys_gettimeofday incompatible with changes to struct time_interpolator 156#endif 157#define CLOCK_REALTIME 0 158#define CLOCK_MONOTONIC 1 159#define CLOCK_DIVIDE_BY_1000 0x4000 160#define CLOCK_ADD_MONOTONIC 0x8000 161 162ENTRY(fsys_gettimeofday) 163 .prologue 164 .altrp b6 165 .body 166 mov r31 = r32 167 tnat.nz p6,p0 = r33 // guard against NaT argument 168(p6) br.cond.spnt.few .fail_einval 169 mov r30 = CLOCK_DIVIDE_BY_1000 170 ;; 171.gettime: 172 // Register map 173 // Incoming r31 = pointer to address where to place result 174 // r30 = flags determining how time is processed 175 // r2,r3 = temp r4-r7 preserved 176 // r8 = result nanoseconds 177 // r9 = result seconds 178 // r10 = temporary storage for clock difference 179 // r11 = preserved: saved ar.pfs 180 // r12 = preserved: memory stack 181 // r13 = preserved: thread pointer 182 // r14 = address of mask / mask 183 // r15 = preserved: system call number 184 // r16 = preserved: current task pointer 185 // r17 = wall to monotonic use 186 // r18 = time_interpolator->offset 187 // r19 = address of wall_to_monotonic 188 // r20 = pointer to struct time_interpolator / pointer to time_interpolator->address 189 // r21 = shift factor 190 // r22 = address of time interpolator->last_counter 191 // r23 = address of time_interpolator->last_cycle 192 // r24 = adress of time_interpolator->offset 193 // r25 = last_cycle value 194 // r26 = last_counter value 195 // r27 = pointer to xtime 196 // r28 = sequence number at the beginning of critcal section 197 // r29 = address of seqlock 198 // r30 = time processing flags / memory address 199 // r31 = pointer to result 200 // Predicates 201 // p6,p7 short term use 202 // p8 = timesource ar.itc 203 // p9 = timesource mmio64 204 // p10 = timesource mmio32 205 // p11 = timesource not to be handled by asm code 206 // p12 = memory time source ( = p9 | p10) 207 // p13 = do cmpxchg with time_interpolator_last_cycle 208 // p14 = Divide by 1000 209 // p15 = Add monotonic 210 // 211 // Note that instructions are optimized for McKinley. McKinley can process two 212 // bundles simultaneously and therefore we continuously try to feed the CPU 213 // two bundles and then a stop. 214 tnat.nz p6,p0 = r31 // branch deferred since it does not fit into bundle structure 215 mov pr = r30,0xc000 // Set predicates according to function 216 add r2 = TI_FLAGS+IA64_TASK_SIZE,r16 217 movl r20 = time_interpolator 218 ;; 219 ld8 r20 = [r20] // get pointer to time_interpolator structure 220 movl r29 = xtime_lock 221 ld4 r2 = [r2] // process work pending flags 222 movl r27 = xtime 223 ;; // only one bundle here 224 ld8 r21 = [r20] // first quad with control information 225 and r2 = TIF_ALLWORK_MASK,r2 226(p6) br.cond.spnt.few .fail_einval // deferred branch 227 ;; 228 add r10 = IA64_TIME_INTERPOLATOR_ADDRESS_OFFSET,r20 229 extr r3 = r21,32,32 // time_interpolator->nsec_per_cyc 230 extr r8 = r21,0,16 // time_interpolator->source 231 cmp.ne p6, p0 = 0, r2 // Fallback if work is scheduled 232(p6) br.cond.spnt.many fsys_fallback_syscall 233 ;; 234 cmp.eq p8,p12 = 0,r8 // Check for cpu timer 235 cmp.eq p9,p0 = 1,r8 // MMIO64 ? 236 extr r2 = r21,24,8 // time_interpolator->jitter 237 cmp.eq p10,p0 = 2,r8 // MMIO32 ? 238 cmp.ltu p11,p0 = 2,r8 // function or other clock 239(p11) br.cond.spnt.many fsys_fallback_syscall 240 ;; 241 setf.sig f7 = r3 // Setup for scaling of counter 242(p15) movl r19 = wall_to_monotonic 243(p12) ld8 r30 = [r10] 244 cmp.ne p13,p0 = r2,r0 // need jitter compensation? 245 extr r21 = r21,16,8 // shift factor 246 ;; 247.time_redo: 248 .pred.rel.mutex p8,p9,p10 249 ld4.acq r28 = [r29] // xtime_lock.sequence. Must come first for locking purposes 250(p8) mov r2 = ar.itc // CPU_TIMER. 36 clocks latency!!! 251 add r22 = IA64_TIME_INTERPOLATOR_LAST_COUNTER_OFFSET,r20 252(p9) ld8 r2 = [r30] // readq(ti->address). Could also have latency issues.. 253(p10) ld4 r2 = [r30] // readw(ti->address) 254(p13) add r23 = IA64_TIME_INTERPOLATOR_LAST_CYCLE_OFFSET,r20 255 ;; // could be removed by moving the last add upward 256 ld8 r26 = [r22] // time_interpolator->last_counter 257(p13) ld8 r25 = [r23] // time interpolator->last_cycle 258 add r24 = IA64_TIME_INTERPOLATOR_OFFSET_OFFSET,r20 259(p15) ld8 r17 = [r19],IA64_TIMESPEC_TV_NSEC_OFFSET 260 ld8 r9 = [r27],IA64_TIMESPEC_TV_NSEC_OFFSET 261 add r14 = IA64_TIME_INTERPOLATOR_MASK_OFFSET, r20 262 ;; 263 ld8 r18 = [r24] // time_interpolator->offset 264 ld8 r8 = [r27],-IA64_TIMESPEC_TV_NSEC_OFFSET // xtime.tv_nsec 265(p13) sub r3 = r25,r2 // Diff needed before comparison (thanks davidm) 266 ;; 267 ld8 r14 = [r14] // time_interpolator->mask 268(p13) cmp.gt.unc p6,p7 = r3,r0 // check if it is less than last. p6,p7 cleared 269 sub r10 = r2,r26 // current_counter - last_counter 270 ;; 271(p6) sub r10 = r25,r26 // time we got was less than last_cycle 272(p7) mov ar.ccv = r25 // more than last_cycle. Prep for cmpxchg 273 ;; 274 and r10 = r10,r14 // Apply mask 275 ;; 276 setf.sig f8 = r10 277 nop.i 123 278 ;; 279(p7) cmpxchg8.rel r3 = [r23],r2,ar.ccv 280EX(.fail_efault, probe.w.fault r31, 3) // This takes 5 cycles and we have spare time 281 xmpy.l f8 = f8,f7 // nsec_per_cyc*(counter-last_counter) 282(p15) add r9 = r9,r17 // Add wall to monotonic.secs to result secs 283 ;; 284(p15) ld8 r17 = [r19],-IA64_TIMESPEC_TV_NSEC_OFFSET 285(p7) cmp.ne p7,p0 = r25,r3 // if cmpxchg not successful redo 286 // simulate tbit.nz.or p7,p0 = r28,0 287 and r28 = ~1,r28 // Make sequence even to force retry if odd 288 getf.sig r2 = f8 289 mf 290 add r8 = r8,r18 // Add time interpolator offset 291 ;; 292 ld4 r10 = [r29] // xtime_lock.sequence 293(p15) add r8 = r8, r17 // Add monotonic.nsecs to nsecs 294 shr.u r2 = r2,r21 295 ;; // overloaded 3 bundles! 296 // End critical section. 297 add r8 = r8,r2 // Add xtime.nsecs 298 cmp4.ne.or p7,p0 = r28,r10 299(p7) br.cond.dpnt.few .time_redo // sequence number changed ? 300 // Now r8=tv->tv_nsec and r9=tv->tv_sec 301 mov r10 = r0 302 movl r2 = 1000000000 303 add r23 = IA64_TIMESPEC_TV_NSEC_OFFSET, r31 304(p14) movl r3 = 2361183241434822607 // Prep for / 1000 hack 305 ;; 306.time_normalize: 307 mov r21 = r8 308 cmp.ge p6,p0 = r8,r2 309(p14) shr.u r20 = r8, 3 // We can repeat this if necessary just wasting some time 310 ;; 311(p14) setf.sig f8 = r20 312(p6) sub r8 = r8,r2 313(p6) add r9 = 1,r9 // two nops before the branch. 314(p14) setf.sig f7 = r3 // Chances for repeats are 1 in 10000 for gettod 315(p6) br.cond.dpnt.few .time_normalize 316 ;; 317 // Divided by 8 though shift. Now divide by 125 318 // The compiler was able to do that with a multiply 319 // and a shift and we do the same 320EX(.fail_efault, probe.w.fault r23, 3) // This also costs 5 cycles 321(p14) xmpy.hu f8 = f8, f7 // xmpy has 5 cycles latency so use it... 322 ;; 323 mov r8 = r0 324(p14) getf.sig r2 = f8 325 ;; 326(p14) shr.u r21 = r2, 4 327 ;; 328EX(.fail_efault, st8 [r31] = r9) 329EX(.fail_efault, st8 [r23] = r21) 330 FSYS_RETURN 331.fail_einval: 332 mov r8 = EINVAL 333 mov r10 = -1 334 FSYS_RETURN 335.fail_efault: 336 mov r8 = EFAULT 337 mov r10 = -1 338 FSYS_RETURN 339END(fsys_gettimeofday) 340 341ENTRY(fsys_clock_gettime) 342 .prologue 343 .altrp b6 344 .body 345 cmp4.ltu p6, p0 = CLOCK_MONOTONIC, r32 346 // Fallback if this is not CLOCK_REALTIME or CLOCK_MONOTONIC 347(p6) br.spnt.few fsys_fallback_syscall 348 mov r31 = r33 349 shl r30 = r32,15 350 br.many .gettime 351END(fsys_clock_gettime) 352 353/* 354 * long fsys_rt_sigprocmask (int how, sigset_t *set, sigset_t *oset, size_t sigsetsize). 355 */ 356#if _NSIG_WORDS != 1 357# error Sorry, fsys_rt_sigprocmask() needs to be updated for _NSIG_WORDS != 1. 358#endif 359ENTRY(fsys_rt_sigprocmask) 360 .prologue 361 .altrp b6 362 .body 363 364 add r2=IA64_TASK_BLOCKED_OFFSET,r16 365 add r9=TI_FLAGS+IA64_TASK_SIZE,r16 366 cmp4.ltu p6,p0=SIG_SETMASK,r32 367 368 cmp.ne p15,p0=r0,r34 // oset != NULL? 369 tnat.nz p8,p0=r34 370 add r31=IA64_TASK_SIGHAND_OFFSET,r16 371 ;; 372 ld8 r3=[r2] // read/prefetch current->blocked 373 ld4 r9=[r9] 374 tnat.nz.or p6,p0=r35 375 376 cmp.ne.or p6,p0=_NSIG_WORDS*8,r35 377 tnat.nz.or p6,p0=r32 378(p6) br.spnt.few .fail_einval // fail with EINVAL 379 ;; 380#ifdef CONFIG_SMP 381 ld8 r31=[r31] // r31 <- current->sighand 382#endif 383 and r9=TIF_ALLWORK_MASK,r9 384 tnat.nz.or p8,p0=r33 385 ;; 386 cmp.ne p7,p0=0,r9 387 cmp.eq p6,p0=r0,r33 // set == NULL? 388 add r31=IA64_SIGHAND_SIGLOCK_OFFSET,r31 // r31 <- current->sighand->siglock 389(p8) br.spnt.few .fail_efault // fail with EFAULT 390(p7) br.spnt.many fsys_fallback_syscall // got pending kernel work... 391(p6) br.dpnt.many .store_mask // -> short-circuit to just reading the signal mask 392 393 /* Argh, we actually have to do some work and _update_ the signal mask: */ 394 395EX(.fail_efault, probe.r.fault r33, 3) // verify user has read-access to *set 396EX(.fail_efault, ld8 r14=[r33]) // r14 <- *set 397 mov r17=(1 << (SIGKILL - 1)) | (1 << (SIGSTOP - 1)) 398 ;; 399 400 rsm psr.i // mask interrupt delivery 401 mov ar.ccv=0 402 andcm r14=r14,r17 // filter out SIGKILL & SIGSTOP 403 404#ifdef CONFIG_SMP 405 mov r17=1 406 ;; 407 cmpxchg4.acq r18=[r31],r17,ar.ccv // try to acquire the lock 408 mov r8=EINVAL // default to EINVAL 409 ;; 410 ld8 r3=[r2] // re-read current->blocked now that we hold the lock 411 cmp4.ne p6,p0=r18,r0 412(p6) br.cond.spnt.many .lock_contention 413 ;; 414#else 415 ld8 r3=[r2] // re-read current->blocked now that we hold the lock 416 mov r8=EINVAL // default to EINVAL 417#endif 418 add r18=IA64_TASK_PENDING_OFFSET+IA64_SIGPENDING_SIGNAL_OFFSET,r16 419 add r19=IA64_TASK_SIGNAL_OFFSET,r16 420 cmp4.eq p6,p0=SIG_BLOCK,r32 421 ;; 422 ld8 r19=[r19] // r19 <- current->signal 423 cmp4.eq p7,p0=SIG_UNBLOCK,r32 424 cmp4.eq p8,p0=SIG_SETMASK,r32 425 ;; 426 ld8 r18=[r18] // r18 <- current->pending.signal 427 .pred.rel.mutex p6,p7,p8 428(p6) or r14=r3,r14 // SIG_BLOCK 429(p7) andcm r14=r3,r14 // SIG_UNBLOCK 430 431(p8) mov r14=r14 // SIG_SETMASK 432(p6) mov r8=0 // clear error code 433 // recalc_sigpending() 434 add r17=IA64_SIGNAL_GROUP_STOP_COUNT_OFFSET,r19 435 436 add r19=IA64_SIGNAL_SHARED_PENDING_OFFSET+IA64_SIGPENDING_SIGNAL_OFFSET,r19 437 ;; 438 ld4 r17=[r17] // r17 <- current->signal->group_stop_count 439(p7) mov r8=0 // clear error code 440 441 ld8 r19=[r19] // r19 <- current->signal->shared_pending 442 ;; 443 cmp4.gt p6,p7=r17,r0 // p6/p7 <- (current->signal->group_stop_count > 0)? 444(p8) mov r8=0 // clear error code 445 446 or r18=r18,r19 // r18 <- current->pending | current->signal->shared_pending 447 ;; 448 // r18 <- (current->pending | current->signal->shared_pending) & ~current->blocked: 449 andcm r18=r18,r14 450 add r9=TI_FLAGS+IA64_TASK_SIZE,r16 451 ;; 452 453(p7) cmp.ne.or.andcm p6,p7=r18,r0 // p6/p7 <- signal pending 454 mov r19=0 // i must not leak kernel bits... 455(p6) br.cond.dpnt.many .sig_pending 456 ;; 457 4581: ld4 r17=[r9] // r17 <- current->thread_info->flags 459 ;; 460 mov ar.ccv=r17 461 and r18=~_TIF_SIGPENDING,r17 // r18 <- r17 & ~(1 << TIF_SIGPENDING) 462 ;; 463 464 st8 [r2]=r14 // update current->blocked with new mask 465 cmpxchg4.acq r8=[r9],r18,ar.ccv // current->thread_info->flags <- r18 466 ;; 467 cmp.ne p6,p0=r17,r8 // update failed? 468(p6) br.cond.spnt.few 1b // yes -> retry 469 470#ifdef CONFIG_SMP 471 st4.rel [r31]=r0 // release the lock 472#endif 473 ssm psr.i 474 ;; 475 476 srlz.d // ensure psr.i is set again 477 mov r18=0 // i must not leak kernel bits... 478 479.store_mask: 480EX(.fail_efault, (p15) probe.w.fault r34, 3) // verify user has write-access to *oset 481EX(.fail_efault, (p15) st8 [r34]=r3) 482 mov r2=0 // i must not leak kernel bits... 483 mov r3=0 // i must not leak kernel bits... 484 mov r8=0 // return 0 485 mov r9=0 // i must not leak kernel bits... 486 mov r14=0 // i must not leak kernel bits... 487 mov r17=0 // i must not leak kernel bits... 488 mov r31=0 // i must not leak kernel bits... 489 FSYS_RETURN 490 491.sig_pending: 492#ifdef CONFIG_SMP 493 st4.rel [r31]=r0 // release the lock 494#endif 495 ssm psr.i 496 ;; 497 srlz.d 498 br.sptk.many fsys_fallback_syscall // with signal pending, do the heavy-weight syscall 499 500#ifdef CONFIG_SMP 501.lock_contention: 502 /* Rather than spinning here, fall back on doing a heavy-weight syscall. */ 503 ssm psr.i 504 ;; 505 srlz.d 506 br.sptk.many fsys_fallback_syscall 507#endif 508END(fsys_rt_sigprocmask) 509 510/* 511 * fsys_getcpu doesn't use the third parameter in this implementation. It reads 512 * current_thread_info()->cpu and corresponding node in cpu_to_node_map. 513 */ 514ENTRY(fsys_getcpu) 515 .prologue 516 .altrp b6 517 .body 518 ;; 519 add r2=TI_FLAGS+IA64_TASK_SIZE,r16 520 tnat.nz p6,p0 = r32 // guard against NaT argument 521 add r3=TI_CPU+IA64_TASK_SIZE,r16 522 ;; 523 ld4 r3=[r3] // M r3 = thread_info->cpu 524 ld4 r2=[r2] // M r2 = thread_info->flags 525(p6) br.cond.spnt.few .fail_einval // B 526 ;; 527 tnat.nz p7,p0 = r33 // I guard against NaT argument 528(p7) br.cond.spnt.few .fail_einval // B 529#ifdef CONFIG_NUMA 530 movl r17=cpu_to_node_map 531 ;; 532EX(.fail_efault, probe.w.fault r32, 3) // M This takes 5 cycles 533EX(.fail_efault, probe.w.fault r33, 3) // M This takes 5 cycles 534 shladd r18=r3,1,r17 535 ;; 536 ld2 r20=[r18] // r20 = cpu_to_node_map[cpu] 537 and r2 = TIF_ALLWORK_MASK,r2 538 ;; 539 cmp.ne p8,p0=0,r2 540(p8) br.spnt.many fsys_fallback_syscall 541 ;; 542 ;; 543EX(.fail_efault, st4 [r32] = r3) 544EX(.fail_efault, st2 [r33] = r20) 545 mov r8=0 546 ;; 547#else 548EX(.fail_efault, probe.w.fault r32, 3) // M This takes 5 cycles 549EX(.fail_efault, probe.w.fault r33, 3) // M This takes 5 cycles 550 and r2 = TIF_ALLWORK_MASK,r2 551 ;; 552 cmp.ne p8,p0=0,r2 553(p8) br.spnt.many fsys_fallback_syscall 554 ;; 555EX(.fail_efault, st4 [r32] = r3) 556EX(.fail_efault, st2 [r33] = r0) 557 mov r8=0 558 ;; 559#endif 560 FSYS_RETURN 561END(fsys_getcpu) 562 563ENTRY(fsys_fallback_syscall) 564 .prologue 565 .altrp b6 566 .body 567 /* 568 * We only get here from light-weight syscall handlers. Thus, we already 569 * know that r15 contains a valid syscall number. No need to re-check. 570 */ 571 adds r17=-1024,r15 572 movl r14=sys_call_table 573 ;; 574 rsm psr.i 575 shladd r18=r17,3,r14 576 ;; 577 ld8 r18=[r18] // load normal (heavy-weight) syscall entry-point 578 mov r29=psr // read psr (12 cyc load latency) 579 mov r27=ar.rsc 580 mov r21=ar.fpsr 581 mov r26=ar.pfs 582END(fsys_fallback_syscall) 583 /* FALL THROUGH */ 584GLOBAL_ENTRY(fsys_bubble_down) 585 .prologue 586 .altrp b6 587 .body 588 /* 589 * We get here for syscalls that don't have a lightweight 590 * handler. For those, we need to bubble down into the kernel 591 * and that requires setting up a minimal pt_regs structure, 592 * and initializing the CPU state more or less as if an 593 * interruption had occurred. To make syscall-restarts work, 594 * we setup pt_regs such that cr_iip points to the second 595 * instruction in syscall_via_break. Decrementing the IP 596 * hence will restart the syscall via break and not 597 * decrementing IP will return us to the caller, as usual. 598 * Note that we preserve the value of psr.pp rather than 599 * initializing it from dcr.pp. This makes it possible to 600 * distinguish fsyscall execution from other privileged 601 * execution. 602 * 603 * On entry: 604 * - normal fsyscall handler register usage, except 605 * that we also have: 606 * - r18: address of syscall entry point 607 * - r21: ar.fpsr 608 * - r26: ar.pfs 609 * - r27: ar.rsc 610 * - r29: psr 611 * 612 * We used to clear some PSR bits here but that requires slow 613 * serialization. Fortuntely, that isn't really necessary. 614 * The rationale is as follows: we used to clear bits 615 * ~PSR_PRESERVED_BITS in PSR.L. Since 616 * PSR_PRESERVED_BITS==PSR.{UP,MFL,MFH,PK,DT,PP,SP,RT,IC}, we 617 * ended up clearing PSR.{BE,AC,I,DFL,DFH,DI,DB,SI,TB}. 618 * However, 619 * 620 * PSR.BE : already is turned off in __kernel_syscall_via_epc() 621 * PSR.AC : don't care (kernel normally turns PSR.AC on) 622 * PSR.I : already turned off by the time fsys_bubble_down gets 623 * invoked 624 * PSR.DFL: always 0 (kernel never turns it on) 625 * PSR.DFH: don't care --- kernel never touches f32-f127 on its own 626 * initiative 627 * PSR.DI : always 0 (kernel never turns it on) 628 * PSR.SI : always 0 (kernel never turns it on) 629 * PSR.DB : don't care --- kernel never enables kernel-level 630 * breakpoints 631 * PSR.TB : must be 0 already; if it wasn't zero on entry to 632 * __kernel_syscall_via_epc, the branch to fsys_bubble_down 633 * will trigger a taken branch; the taken-trap-handler then 634 * converts the syscall into a break-based system-call. 635 */ 636 /* 637 * Reading psr.l gives us only bits 0-31, psr.it, and psr.mc. 638 * The rest we have to synthesize. 639 */ 640# define PSR_ONE_BITS ((3 << IA64_PSR_CPL0_BIT) \ 641 | (0x1 << IA64_PSR_RI_BIT) \ 642 | IA64_PSR_BN | IA64_PSR_I) 643 644 invala // M0|1 645 movl r14=ia64_ret_from_syscall // X 646 647 nop.m 0 648 movl r28=__kernel_syscall_via_break // X create cr.iip 649 ;; 650 651 mov r2=r16 // A get task addr to addl-addressable register 652 adds r16=IA64_TASK_THREAD_ON_USTACK_OFFSET,r16 // A 653 mov r31=pr // I0 save pr (2 cyc) 654 ;; 655 st1 [r16]=r0 // M2|3 clear current->thread.on_ustack flag 656 addl r22=IA64_RBS_OFFSET,r2 // A compute base of RBS 657 add r3=TI_FLAGS+IA64_TASK_SIZE,r2 // A 658 ;; 659 ld4 r3=[r3] // M0|1 r3 = current_thread_info()->flags 660 lfetch.fault.excl.nt1 [r22] // M0|1 prefetch register backing-store 661 nop.i 0 662 ;; 663 mov ar.rsc=0 // M2 set enforced lazy mode, pl 0, LE, loadrs=0 664 nop.m 0 665 nop.i 0 666 ;; 667 mov r23=ar.bspstore // M2 (12 cyc) save ar.bspstore 668 mov.m r24=ar.rnat // M2 (5 cyc) read ar.rnat (dual-issues!) 669 nop.i 0 670 ;; 671 mov ar.bspstore=r22 // M2 (6 cyc) switch to kernel RBS 672 movl r8=PSR_ONE_BITS // X 673 ;; 674 mov r25=ar.unat // M2 (5 cyc) save ar.unat 675 mov r19=b6 // I0 save b6 (2 cyc) 676 mov r20=r1 // A save caller's gp in r20 677 ;; 678 or r29=r8,r29 // A construct cr.ipsr value to save 679 mov b6=r18 // I0 copy syscall entry-point to b6 (7 cyc) 680 addl r1=IA64_STK_OFFSET-IA64_PT_REGS_SIZE,r2 // A compute base of memory stack 681 682 mov r18=ar.bsp // M2 save (kernel) ar.bsp (12 cyc) 683 cmp.ne pKStk,pUStk=r0,r0 // A set pKStk <- 0, pUStk <- 1 684 br.call.sptk.many b7=ia64_syscall_setup // B 685 ;; 686 mov ar.rsc=0x3 // M2 set eager mode, pl 0, LE, loadrs=0 687 mov rp=r14 // I0 set the real return addr 688 and r3=_TIF_SYSCALL_TRACEAUDIT,r3 // A 689 ;; 690 ssm psr.i // M2 we're on kernel stacks now, reenable irqs 691 cmp.eq p8,p0=r3,r0 // A 692(p10) br.cond.spnt.many ia64_ret_from_syscall // B return if bad call-frame or r15 is a NaT 693 694 nop.m 0 695(p8) br.call.sptk.many b6=b6 // B (ignore return address) 696 br.cond.spnt ia64_trace_syscall // B 697END(fsys_bubble_down) 698 699 .rodata 700 .align 8 701 .globl fsyscall_table 702 703 data8 fsys_bubble_down 704fsyscall_table: 705 data8 fsys_ni_syscall 706 data8 0 // exit // 1025 707 data8 0 // read 708 data8 0 // write 709 data8 0 // open 710 data8 0 // close 711 data8 0 // creat // 1030 712 data8 0 // link 713 data8 0 // unlink 714 data8 0 // execve 715 data8 0 // chdir 716 data8 0 // fchdir // 1035 717 data8 0 // utimes 718 data8 0 // mknod 719 data8 0 // chmod 720 data8 0 // chown 721 data8 0 // lseek // 1040 722 data8 fsys_getpid // getpid 723 data8 fsys_getppid // getppid 724 data8 0 // mount 725 data8 0 // umount 726 data8 0 // setuid // 1045 727 data8 0 // getuid 728 data8 0 // geteuid 729 data8 0 // ptrace 730 data8 0 // access 731 data8 0 // sync // 1050 732 data8 0 // fsync 733 data8 0 // fdatasync 734 data8 0 // kill 735 data8 0 // rename 736 data8 0 // mkdir // 1055 737 data8 0 // rmdir 738 data8 0 // dup 739 data8 0 // pipe 740 data8 0 // times 741 data8 0 // brk // 1060 742 data8 0 // setgid 743 data8 0 // getgid 744 data8 0 // getegid 745 data8 0 // acct 746 data8 0 // ioctl // 1065 747 data8 0 // fcntl 748 data8 0 // umask 749 data8 0 // chroot 750 data8 0 // ustat 751 data8 0 // dup2 // 1070 752 data8 0 // setreuid 753 data8 0 // setregid 754 data8 0 // getresuid 755 data8 0 // setresuid 756 data8 0 // getresgid // 1075 757 data8 0 // setresgid 758 data8 0 // getgroups 759 data8 0 // setgroups 760 data8 0 // getpgid 761 data8 0 // setpgid // 1080 762 data8 0 // setsid 763 data8 0 // getsid 764 data8 0 // sethostname 765 data8 0 // setrlimit 766 data8 0 // getrlimit // 1085 767 data8 0 // getrusage 768 data8 fsys_gettimeofday // gettimeofday 769 data8 0 // settimeofday 770 data8 0 // select 771 data8 0 // poll // 1090 772 data8 0 // symlink 773 data8 0 // readlink 774 data8 0 // uselib 775 data8 0 // swapon 776 data8 0 // swapoff // 1095 777 data8 0 // reboot 778 data8 0 // truncate 779 data8 0 // ftruncate 780 data8 0 // fchmod 781 data8 0 // fchown // 1100 782 data8 0 // getpriority 783 data8 0 // setpriority 784 data8 0 // statfs 785 data8 0 // fstatfs 786 data8 0 // gettid // 1105 787 data8 0 // semget 788 data8 0 // semop 789 data8 0 // semctl 790 data8 0 // msgget 791 data8 0 // msgsnd // 1110 792 data8 0 // msgrcv 793 data8 0 // msgctl 794 data8 0 // shmget 795 data8 0 // shmat 796 data8 0 // shmdt // 1115 797 data8 0 // shmctl 798 data8 0 // syslog 799 data8 0 // setitimer 800 data8 0 // getitimer 801 data8 0 // 1120 802 data8 0 803 data8 0 804 data8 0 // vhangup 805 data8 0 // lchown 806 data8 0 // remap_file_pages // 1125 807 data8 0 // wait4 808 data8 0 // sysinfo 809 data8 0 // clone 810 data8 0 // setdomainname 811 data8 0 // newuname // 1130 812 data8 0 // adjtimex 813 data8 0 814 data8 0 // init_module 815 data8 0 // delete_module 816 data8 0 // 1135 817 data8 0 818 data8 0 // quotactl 819 data8 0 // bdflush 820 data8 0 // sysfs 821 data8 0 // personality // 1140 822 data8 0 // afs_syscall 823 data8 0 // setfsuid 824 data8 0 // setfsgid 825 data8 0 // getdents 826 data8 0 // flock // 1145 827 data8 0 // readv 828 data8 0 // writev 829 data8 0 // pread64 830 data8 0 // pwrite64 831 data8 0 // sysctl // 1150 832 data8 0 // mmap 833 data8 0 // munmap 834 data8 0 // mlock 835 data8 0 // mlockall 836 data8 0 // mprotect // 1155 837 data8 0 // mremap 838 data8 0 // msync 839 data8 0 // munlock 840 data8 0 // munlockall 841 data8 0 // sched_getparam // 1160 842 data8 0 // sched_setparam 843 data8 0 // sched_getscheduler 844 data8 0 // sched_setscheduler 845 data8 0 // sched_yield 846 data8 0 // sched_get_priority_max // 1165 847 data8 0 // sched_get_priority_min 848 data8 0 // sched_rr_get_interval 849 data8 0 // nanosleep 850 data8 0 // nfsservctl 851 data8 0 // prctl // 1170 852 data8 0 // getpagesize 853 data8 0 // mmap2 854 data8 0 // pciconfig_read 855 data8 0 // pciconfig_write 856 data8 0 // perfmonctl // 1175 857 data8 0 // sigaltstack 858 data8 0 // rt_sigaction 859 data8 0 // rt_sigpending 860 data8 fsys_rt_sigprocmask // rt_sigprocmask 861 data8 0 // rt_sigqueueinfo // 1180 862 data8 0 // rt_sigreturn 863 data8 0 // rt_sigsuspend 864 data8 0 // rt_sigtimedwait 865 data8 0 // getcwd 866 data8 0 // capget // 1185 867 data8 0 // capset 868 data8 0 // sendfile 869 data8 0 870 data8 0 871 data8 0 // socket // 1190 872 data8 0 // bind 873 data8 0 // connect 874 data8 0 // listen 875 data8 0 // accept 876 data8 0 // getsockname // 1195 877 data8 0 // getpeername 878 data8 0 // socketpair 879 data8 0 // send 880 data8 0 // sendto 881 data8 0 // recv // 1200 882 data8 0 // recvfrom 883 data8 0 // shutdown 884 data8 0 // setsockopt 885 data8 0 // getsockopt 886 data8 0 // sendmsg // 1205 887 data8 0 // recvmsg 888 data8 0 // pivot_root 889 data8 0 // mincore 890 data8 0 // madvise 891 data8 0 // newstat // 1210 892 data8 0 // newlstat 893 data8 0 // newfstat 894 data8 0 // clone2 895 data8 0 // getdents64 896 data8 0 // getunwind // 1215 897 data8 0 // readahead 898 data8 0 // setxattr 899 data8 0 // lsetxattr 900 data8 0 // fsetxattr 901 data8 0 // getxattr // 1220 902 data8 0 // lgetxattr 903 data8 0 // fgetxattr 904 data8 0 // listxattr 905 data8 0 // llistxattr 906 data8 0 // flistxattr // 1225 907 data8 0 // removexattr 908 data8 0 // lremovexattr 909 data8 0 // fremovexattr 910 data8 0 // tkill 911 data8 0 // futex // 1230 912 data8 0 // sched_setaffinity 913 data8 0 // sched_getaffinity 914 data8 fsys_set_tid_address // set_tid_address 915 data8 0 // fadvise64_64 916 data8 0 // tgkill // 1235 917 data8 0 // exit_group 918 data8 0 // lookup_dcookie 919 data8 0 // io_setup 920 data8 0 // io_destroy 921 data8 0 // io_getevents // 1240 922 data8 0 // io_submit 923 data8 0 // io_cancel 924 data8 0 // epoll_create 925 data8 0 // epoll_ctl 926 data8 0 // epoll_wait // 1245 927 data8 0 // restart_syscall 928 data8 0 // semtimedop 929 data8 0 // timer_create 930 data8 0 // timer_settime 931 data8 0 // timer_gettime // 1250 932 data8 0 // timer_getoverrun 933 data8 0 // timer_delete 934 data8 0 // clock_settime 935 data8 fsys_clock_gettime // clock_gettime 936 data8 0 // clock_getres // 1255 937 data8 0 // clock_nanosleep 938 data8 0 // fstatfs64 939 data8 0 // statfs64 940 data8 0 // mbind 941 data8 0 // get_mempolicy // 1260 942 data8 0 // set_mempolicy 943 data8 0 // mq_open 944 data8 0 // mq_unlink 945 data8 0 // mq_timedsend 946 data8 0 // mq_timedreceive // 1265 947 data8 0 // mq_notify 948 data8 0 // mq_getsetattr 949 data8 0 // kexec_load 950 data8 0 // vserver 951 data8 0 // waitid // 1270 952 data8 0 // add_key 953 data8 0 // request_key 954 data8 0 // keyctl 955 data8 0 // ioprio_set 956 data8 0 // ioprio_get // 1275 957 data8 0 // move_pages 958 data8 0 // inotify_init 959 data8 0 // inotify_add_watch 960 data8 0 // inotify_rm_watch 961 data8 0 // migrate_pages // 1280 962 data8 0 // openat 963 data8 0 // mkdirat 964 data8 0 // mknodat 965 data8 0 // fchownat 966 data8 0 // futimesat // 1285 967 data8 0 // newfstatat 968 data8 0 // unlinkat 969 data8 0 // renameat 970 data8 0 // linkat 971 data8 0 // symlinkat // 1290 972 data8 0 // readlinkat 973 data8 0 // fchmodat 974 data8 0 // faccessat 975 data8 0 976 data8 0 // 1295 977 data8 0 // unshare 978 data8 0 // splice 979 data8 0 // set_robust_list 980 data8 0 // get_robust_list 981 data8 0 // sync_file_range // 1300 982 data8 0 // tee 983 data8 0 // vmsplice 984 data8 0 985 data8 fsys_getcpu // getcpu // 1304 986 987 // fill in zeros for the remaining entries 988 .zero: 989 .space fsyscall_table + 8*NR_syscalls - .zero, 0 990