projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] ixp2000: fix gcc4 breakage
[linux-2.6-block.git] / arch / ia64 / kernel / entry.S
CommitLineData
1da177e4
LT		 1/*
2 * ia64/kernel/entry.S
3 *
4 * Kernel entry points.
5 *
6 * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co
7 * David Mosberger-Tang <davidm@hpl.hp.com>
8 * Copyright (C) 1999, 2002-2003
9 * Asit Mallick <Asit.K.Mallick@intel.com>
10 * Don Dugger <Don.Dugger@intel.com>
11 * Suresh Siddha <suresh.b.siddha@intel.com>
12 * Fenghua Yu <fenghua.yu@intel.com>
13 * Copyright (C) 1999 VA Linux Systems
14 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
15 */
16/*
17 * ia64_switch_to now places correct virtual mapping in in TR2 for
18 * kernel stack. This allows us to handle interrupts without changing
19 * to physical mode.
20 *
21 * Jonathan Nicklin <nicklin@missioncriticallinux.com>
22 * Patrick O'Rourke <orourke@missioncriticallinux.com>
23 * 11/07/2000
24 */
25/*
26 * Global (preserved) predicate usage on syscall entry/exit path:
27 *
28 * pKStk: See entry.h.
29 * pUStk: See entry.h.
30 * pSys: See entry.h.
31 * pNonSys: !pSys
32 */
33
34#include <linux/config.h>
35
36#include <asm/asmmacro.h>
37#include <asm/cache.h>
38#include <asm/errno.h>
39#include <asm/kregs.h>
39e01cb8 40#include <asm/asm-offsets.h>
1da177e4
LT		 41#include <asm/pgtable.h>
42#include <asm/percpu.h>
43#include <asm/processor.h>
44#include <asm/thread_info.h>
45#include <asm/unistd.h>
46
47#include "minstate.h"
48
49 /*
50 * execve() is special because in case of success, we need to
51 * setup a null register window frame.
52 */
53ENTRY(ia64_execve)
54 /*
55 * Allocate 8 input registers since ptrace() may clobber them
56 */
57 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
58 alloc loc1=ar.pfs,8,2,4,0
59 mov loc0=rp
60 .body
61 mov out0=in0 // filename
62 ;; // stop bit between alloc and call
63 mov out1=in1 // argv
64 mov out2=in2 // envp
65 add out3=16,sp // regs
66 br.call.sptk.many rp=sys_execve
67.ret0:
68#ifdef CONFIG_IA32_SUPPORT
69 /*
70 * Check if we're returning to ia32 mode. If so, we need to restore ia32 registers
71 * from pt_regs.
72 */
73 adds r16=PT(CR_IPSR)+16,sp
74 ;;
75 ld8 r16=[r16]
76#endif
77 cmp4.ge p6,p7=r8,r0
78 mov ar.pfs=loc1 // restore ar.pfs
79 sxt4 r8=r8 // return 64-bit result
80 ;;
81 stf.spill [sp]=f0
82(p6) cmp.ne pKStk,pUStk=r0,r0 // a successful execve() lands us in user-mode...
83 mov rp=loc0
84(p6) mov ar.pfs=r0 // clear ar.pfs on success
85(p7) br.ret.sptk.many rp
86
87 /*
88 * In theory, we'd have to zap this state only to prevent leaking of
89 * security sensitive state (e.g., if current->mm->dumpable is zero). However,
90 * this executes in less than 20 cycles even on Itanium, so it's not worth
91 * optimizing for...).
92 */
93 mov ar.unat=0; mov ar.lc=0
94 mov r4=0; mov f2=f0; mov b1=r0
95 mov r5=0; mov f3=f0; mov b2=r0
96 mov r6=0; mov f4=f0; mov b3=r0
97 mov r7=0; mov f5=f0; mov b4=r0
98 ldf.fill f12=[sp]; mov f13=f0; mov b5=r0
99 ldf.fill f14=[sp]; ldf.fill f15=[sp]; mov f16=f0
100 ldf.fill f17=[sp]; ldf.fill f18=[sp]; mov f19=f0
101 ldf.fill f20=[sp]; ldf.fill f21=[sp]; mov f22=f0
102 ldf.fill f23=[sp]; ldf.fill f24=[sp]; mov f25=f0
103 ldf.fill f26=[sp]; ldf.fill f27=[sp]; mov f28=f0
104 ldf.fill f29=[sp]; ldf.fill f30=[sp]; mov f31=f0
105#ifdef CONFIG_IA32_SUPPORT
106 tbit.nz p6,p0=r16, IA64_PSR_IS_BIT
107 movl loc0=ia64_ret_from_ia32_execve
108 ;;
109(p6) mov rp=loc0
110#endif
111 br.ret.sptk.many rp
112END(ia64_execve)
113
114/*
115 * sys_clone2(u64 flags, u64 ustack_base, u64 ustack_size, u64 parent_tidptr, u64 child_tidptr,
116 * u64 tls)
117 */
118GLOBAL_ENTRY(sys_clone2)
119 /*
120 * Allocate 8 input registers since ptrace() may clobber them
121 */
122 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
123 alloc r16=ar.pfs,8,2,6,0
124 DO_SAVE_SWITCH_STACK
125 adds r2=PT(R16)+IA64_SWITCH_STACK_SIZE+16,sp
126 mov loc0=rp
127 mov loc1=r16 // save ar.pfs across do_fork
128 .body
129 mov out1=in1
130 mov out3=in2
131 tbit.nz p6,p0=in0,CLONE_SETTLS_BIT
132 mov out4=in3 // parent_tidptr: valid only w/CLONE_PARENT_SETTID
133 ;;
134(p6) st8 [r2]=in5 // store TLS in r16 for copy_thread()
135 mov out5=in4 // child_tidptr: valid only w/CLONE_CHILD_SETTID or CLONE_CHILD_CLEARTID
136 adds out2=IA64_SWITCH_STACK_SIZE+16,sp // out2 = &regs
137 mov out0=in0 // out0 = clone_flags
138 br.call.sptk.many rp=do_fork
139.ret1: .restore sp
140 adds sp=IA64_SWITCH_STACK_SIZE,sp // pop the switch stack
141 mov ar.pfs=loc1
142 mov rp=loc0
143 br.ret.sptk.many rp
144END(sys_clone2)
145
146/*
147 * sys_clone(u64 flags, u64 ustack_base, u64 parent_tidptr, u64 child_tidptr, u64 tls)
148 * Deprecated. Use sys_clone2() instead.
149 */
150GLOBAL_ENTRY(sys_clone)
151 /*
152 * Allocate 8 input registers since ptrace() may clobber them
153 */
154 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
155 alloc r16=ar.pfs,8,2,6,0
156 DO_SAVE_SWITCH_STACK
157 adds r2=PT(R16)+IA64_SWITCH_STACK_SIZE+16,sp
158 mov loc0=rp
159 mov loc1=r16 // save ar.pfs across do_fork
160 .body
161 mov out1=in1
162 mov out3=16 // stacksize (compensates for 16-byte scratch area)
163 tbit.nz p6,p0=in0,CLONE_SETTLS_BIT
164 mov out4=in2 // parent_tidptr: valid only w/CLONE_PARENT_SETTID
165 ;;
166(p6) st8 [r2]=in4 // store TLS in r13 (tp)
167 mov out5=in3 // child_tidptr: valid only w/CLONE_CHILD_SETTID or CLONE_CHILD_CLEARTID
168 adds out2=IA64_SWITCH_STACK_SIZE+16,sp // out2 = &regs
169 mov out0=in0 // out0 = clone_flags
170 br.call.sptk.many rp=do_fork
171.ret2: .restore sp
172 adds sp=IA64_SWITCH_STACK_SIZE,sp // pop the switch stack
173 mov ar.pfs=loc1
174 mov rp=loc0
175 br.ret.sptk.many rp
176END(sys_clone)
177
178/*
179 * prev_task <- ia64_switch_to(struct task_struct *next)
180 * With Ingo's new scheduler, interrupts are disabled when this routine gets
181 * called. The code starting at .map relies on this. The rest of the code
182 * doesn't care about the interrupt masking status.
183 */
184GLOBAL_ENTRY(ia64_switch_to)
185 .prologue
186 alloc r16=ar.pfs,1,0,0,0
187 DO_SAVE_SWITCH_STACK
188 .body
189
190 adds r22=IA64_TASK_THREAD_KSP_OFFSET,r13
191 movl r25=init_task
192 mov r27=IA64_KR(CURRENT_STACK)
193 adds r21=IA64_TASK_THREAD_KSP_OFFSET,in0
194 dep r20=0,in0,61,3 // physical address of "next"
195 ;;
196 st8 [r22]=sp // save kernel stack pointer of old task
197 shr.u r26=r20,IA64_GRANULE_SHIFT
198 cmp.eq p7,p6=r25,in0
199 ;;
200 /*
201 * If we've already mapped this task's page, we can skip doing it again.
202 */
203(p6) cmp.eq p7,p6=r26,r27
204(p6) br.cond.dpnt .map
205 ;;
206.done:
1da177e4
LT		 207 ld8 sp=[r21] // load kernel stack pointer of new task
208 mov IA64_KR(CURRENT)=in0 // update "current" application register
209 mov r8=r13 // return pointer to previously running task
210 mov r13=in0 // set "current" pointer
211 ;;
212 DO_LOAD_SWITCH_STACK
213
214#ifdef CONFIG_SMP
215 sync.i // ensure "fc"s done by this CPU are visible on other CPUs
216#endif
217 br.ret.sptk.many rp // boogie on out in new context
218
219.map:
220 rsm psr.ic // interrupts (psr.i) are already disabled here
221 movl r25=PAGE_KERNEL
222 ;;
223 srlz.d
224 or r23=r25,r20 // construct PA | page properties
225 mov r25=IA64_GRANULE_SHIFT<<2
226 ;;
227 mov cr.itir=r25
228 mov cr.ifa=in0 // VA of next task...
229 ;;
230 mov r25=IA64_TR_CURRENT_STACK
231 mov IA64_KR(CURRENT_STACK)=r26 // remember last page we mapped...
232 ;;
233 itr.d dtr[r25]=r23 // wire in new mapping...
02326223
CK		 234 ssm psr.ic // reenable the psr.ic bit
235 ;;
236 srlz.d
1da177e4
LT		 237 br.cond.sptk .done
238END(ia64_switch_to)
239
240/*
241 * Note that interrupts are enabled during save_switch_stack and load_switch_stack. This
242 * means that we may get an interrupt with "sp" pointing to the new kernel stack while
243 * ar.bspstore is still pointing to the old kernel backing store area. Since ar.rsc,
244 * ar.rnat, ar.bsp, and ar.bspstore are all preserved by interrupts, this is not a
245 * problem. Also, we don't need to specify unwind information for preserved registers
246 * that are not modified in save_switch_stack as the right unwind information is already
247 * specified at the call-site of save_switch_stack.
248 */
249
250/*
251 * save_switch_stack:
252 * - r16 holds ar.pfs
253 * - b7 holds address to return to
254 * - rp (b0) holds return address to save
255 */
256GLOBAL_ENTRY(save_switch_stack)
257 .prologue
258 .altrp b7
259 flushrs // flush dirty regs to backing store (must be first in insn group)
260 .save @priunat,r17
261 mov r17=ar.unat // preserve caller's
262 .body
263#ifdef CONFIG_ITANIUM
264 adds r2=16+128,sp
265 adds r3=16+64,sp
266 adds r14=SW(R4)+16,sp
267 ;;
268 st8.spill [r14]=r4,16 // spill r4
269 lfetch.fault.excl.nt1 [r3],128
270 ;;
271 lfetch.fault.excl.nt1 [r2],128
272 lfetch.fault.excl.nt1 [r3],128
273 ;;
274 lfetch.fault.excl [r2]
275 lfetch.fault.excl [r3]
276 adds r15=SW(R5)+16,sp
277#else
278 add r2=16+3*128,sp
279 add r3=16,sp
280 add r14=SW(R4)+16,sp
281 ;;
282 st8.spill [r14]=r4,SW(R6)-SW(R4) // spill r4 and prefetch offset 0x1c0
283 lfetch.fault.excl.nt1 [r3],128 // prefetch offset 0x010
284 ;;
285 lfetch.fault.excl.nt1 [r3],128 // prefetch offset 0x090
286 lfetch.fault.excl.nt1 [r2],128 // prefetch offset 0x190
287 ;;
288 lfetch.fault.excl.nt1 [r3] // prefetch offset 0x110
289 lfetch.fault.excl.nt1 [r2] // prefetch offset 0x210
290 adds r15=SW(R5)+16,sp
291#endif
292 ;;
293 st8.spill [r15]=r5,SW(R7)-SW(R5) // spill r5
294 mov.m ar.rsc=0 // put RSE in mode: enforced lazy, little endian, pl 0
295 add r2=SW(F2)+16,sp // r2 = &sw->f2
296 ;;
297 st8.spill [r14]=r6,SW(B0)-SW(R6) // spill r6
298 mov.m r18=ar.fpsr // preserve fpsr
299 add r3=SW(F3)+16,sp // r3 = &sw->f3
300 ;;
301 stf.spill [r2]=f2,32
302 mov.m r19=ar.rnat
303 mov r21=b0
304
305 stf.spill [r3]=f3,32
306 st8.spill [r15]=r7,SW(B2)-SW(R7) // spill r7
307 mov r22=b1
308 ;;
309 // since we're done with the spills, read and save ar.unat:
310 mov.m r29=ar.unat
311 mov.m r20=ar.bspstore
312 mov r23=b2
313 stf.spill [r2]=f4,32
314 stf.spill [r3]=f5,32
315 mov r24=b3
316 ;;
317 st8 [r14]=r21,SW(B1)-SW(B0) // save b0
318 st8 [r15]=r23,SW(B3)-SW(B2) // save b2
319 mov r25=b4
320 mov r26=b5
321 ;;
322 st8 [r14]=r22,SW(B4)-SW(B1) // save b1
323 st8 [r15]=r24,SW(AR_PFS)-SW(B3) // save b3
324 mov r21=ar.lc // I-unit
325 stf.spill [r2]=f12,32
326 stf.spill [r3]=f13,32
327 ;;
328 st8 [r14]=r25,SW(B5)-SW(B4) // save b4
329 st8 [r15]=r16,SW(AR_LC)-SW(AR_PFS) // save ar.pfs
330 stf.spill [r2]=f14,32
331 stf.spill [r3]=f15,32
332 ;;
333 st8 [r14]=r26 // save b5
334 st8 [r15]=r21 // save ar.lc
335 stf.spill [r2]=f16,32
336 stf.spill [r3]=f17,32
337 ;;
338 stf.spill [r2]=f18,32
339 stf.spill [r3]=f19,32
340 ;;
341 stf.spill [r2]=f20,32
342 stf.spill [r3]=f21,32
343 ;;
344 stf.spill [r2]=f22,32
345 stf.spill [r3]=f23,32
346 ;;
347 stf.spill [r2]=f24,32
348 stf.spill [r3]=f25,32
349 ;;
350 stf.spill [r2]=f26,32
351 stf.spill [r3]=f27,32
352 ;;
353 stf.spill [r2]=f28,32
354 stf.spill [r3]=f29,32
355 ;;
356 stf.spill [r2]=f30,SW(AR_UNAT)-SW(F30)
357 stf.spill [r3]=f31,SW(PR)-SW(F31)
358 add r14=SW(CALLER_UNAT)+16,sp
359 ;;
360 st8 [r2]=r29,SW(AR_RNAT)-SW(AR_UNAT) // save ar.unat
361 st8 [r14]=r17,SW(AR_FPSR)-SW(CALLER_UNAT) // save caller_unat
362 mov r21=pr
363 ;;
364 st8 [r2]=r19,SW(AR_BSPSTORE)-SW(AR_RNAT) // save ar.rnat
365 st8 [r3]=r21 // save predicate registers
366 ;;
367 st8 [r2]=r20 // save ar.bspstore
368 st8 [r14]=r18 // save fpsr
369 mov ar.rsc=3 // put RSE back into eager mode, pl 0
370 br.cond.sptk.many b7
371END(save_switch_stack)
372
373/*
374 * load_switch_stack:
375 * - "invala" MUST be done at call site (normally in DO_LOAD_SWITCH_STACK)
376 * - b7 holds address to return to
377 * - must not touch r8-r11
378 */
379ENTRY(load_switch_stack)
380 .prologue
381 .altrp b7
382
383 .body
384 lfetch.fault.nt1 [sp]
385 adds r2=SW(AR_BSPSTORE)+16,sp
386 adds r3=SW(AR_UNAT)+16,sp
387 mov ar.rsc=0 // put RSE into enforced lazy mode
388 adds r14=SW(CALLER_UNAT)+16,sp
389 adds r15=SW(AR_FPSR)+16,sp
390 ;;
391 ld8 r27=[r2],(SW(B0)-SW(AR_BSPSTORE)) // bspstore
392 ld8 r29=[r3],(SW(B1)-SW(AR_UNAT)) // unat
393 ;;
394 ld8 r21=[r2],16 // restore b0
395 ld8 r22=[r3],16 // restore b1
396 ;;
397 ld8 r23=[r2],16 // restore b2
398 ld8 r24=[r3],16 // restore b3
399 ;;
400 ld8 r25=[r2],16 // restore b4
401 ld8 r26=[r3],16 // restore b5
402 ;;
403 ld8 r16=[r2],(SW(PR)-SW(AR_PFS)) // ar.pfs
404 ld8 r17=[r3],(SW(AR_RNAT)-SW(AR_LC)) // ar.lc
405 ;;
406 ld8 r28=[r2] // restore pr
407 ld8 r30=[r3] // restore rnat
408 ;;
409 ld8 r18=[r14],16 // restore caller's unat
410 ld8 r19=[r15],24 // restore fpsr
411 ;;
412 ldf.fill f2=[r14],32
413 ldf.fill f3=[r15],32
414 ;;
415 ldf.fill f4=[r14],32
416 ldf.fill f5=[r15],32
417 ;;
418 ldf.fill f12=[r14],32
419 ldf.fill f13=[r15],32
420 ;;
421 ldf.fill f14=[r14],32
422 ldf.fill f15=[r15],32
423 ;;
424 ldf.fill f16=[r14],32
425 ldf.fill f17=[r15],32
426 ;;
427 ldf.fill f18=[r14],32
428 ldf.fill f19=[r15],32
429 mov b0=r21
430 ;;
431 ldf.fill f20=[r14],32
432 ldf.fill f21=[r15],32
433 mov b1=r22
434 ;;
435 ldf.fill f22=[r14],32
436 ldf.fill f23=[r15],32
437 mov b2=r23
438 ;;
439 mov ar.bspstore=r27
440 mov ar.unat=r29 // establish unat holding the NaT bits for r4-r7
441 mov b3=r24
442 ;;
443 ldf.fill f24=[r14],32
444 ldf.fill f25=[r15],32
445 mov b4=r25
446 ;;
447 ldf.fill f26=[r14],32
448 ldf.fill f27=[r15],32
449 mov b5=r26
450 ;;
451 ldf.fill f28=[r14],32
452 ldf.fill f29=[r15],32
453 mov ar.pfs=r16
454 ;;
455 ldf.fill f30=[r14],32
456 ldf.fill f31=[r15],24
457 mov ar.lc=r17
458 ;;
459 ld8.fill r4=[r14],16
460 ld8.fill r5=[r15],16
461 mov pr=r28,-1
462 ;;
463 ld8.fill r6=[r14],16
464 ld8.fill r7=[r15],16
465
466 mov ar.unat=r18 // restore caller's unat
467 mov ar.rnat=r30 // must restore after bspstore but before rsc!
468 mov ar.fpsr=r19 // restore fpsr
469 mov ar.rsc=3 // put RSE back into eager mode, pl 0
470 br.cond.sptk.many b7
471END(load_switch_stack)
472
383f2835
CK		 473GLOBAL_ENTRY(prefetch_stack)
474 add r14 = -IA64_SWITCH_STACK_SIZE, sp
475 add r15 = IA64_TASK_THREAD_KSP_OFFSET, in0
476 ;;
477 ld8 r16 = [r15] // load next's stack pointer
478 lfetch.fault.excl [r14], 128
479 ;;
480 lfetch.fault.excl [r14], 128
481 lfetch.fault [r16], 128
482 ;;
483 lfetch.fault.excl [r14], 128
484 lfetch.fault [r16], 128
485 ;;
486 lfetch.fault.excl [r14], 128
487 lfetch.fault [r16], 128
488 ;;
489 lfetch.fault.excl [r14], 128
490 lfetch.fault [r16], 128
491 ;;
492 lfetch.fault [r16], 128
493 br.ret.sptk.many rp
24b8e0cc 494END(prefetch_stack)
383f2835 495
1da177e4
LT		 496GLOBAL_ENTRY(execve)
497 mov r15=__NR_execve // put syscall number in place
498 break __BREAK_SYSCALL
499 br.ret.sptk.many rp
500END(execve)
501
502GLOBAL_ENTRY(clone)
503 mov r15=__NR_clone // put syscall number in place
504 break __BREAK_SYSCALL
505 br.ret.sptk.many rp
506END(clone)
507
508 /*
509 * Invoke a system call, but do some tracing before and after the call.
510 * We MUST preserve the current register frame throughout this routine
511 * because some system calls (such as ia64_execve) directly
512 * manipulate ar.pfs.
513 */
514GLOBAL_ENTRY(ia64_trace_syscall)
515 PT_REGS_UNWIND_INFO(0)
516 /*
517 * We need to preserve the scratch registers f6-f11 in case the system
518 * call is sigreturn.
519 */
520 adds r16=PT(F6)+16,sp
521 adds r17=PT(F7)+16,sp
522 ;;
523 stf.spill [r16]=f6,32
524 stf.spill [r17]=f7,32
525 ;;
526 stf.spill [r16]=f8,32
527 stf.spill [r17]=f9,32
528 ;;
529 stf.spill [r16]=f10
530 stf.spill [r17]=f11
531 br.call.sptk.many rp=syscall_trace_enter // give parent a chance to catch syscall args
532 adds r16=PT(F6)+16,sp
533 adds r17=PT(F7)+16,sp
534 ;;
535 ldf.fill f6=[r16],32
536 ldf.fill f7=[r17],32
537 ;;
538 ldf.fill f8=[r16],32
539 ldf.fill f9=[r17],32
540 ;;
541 ldf.fill f10=[r16]
542 ldf.fill f11=[r17]
543 // the syscall number may have changed, so re-load it and re-calculate the
544 // syscall entry-point:
545 adds r15=PT(R15)+16,sp // r15 = &pt_regs.r15 (syscall #)
546 ;;
547 ld8 r15=[r15]
548 mov r3=NR_syscalls - 1
549 ;;
550 adds r15=-1024,r15
551 movl r16=sys_call_table
552 ;;
553 shladd r20=r15,3,r16 // r20 = sys_call_table + 8*(syscall-1024)
554 cmp.leu p6,p7=r15,r3
555 ;;
556(p6) ld8 r20=[r20] // load address of syscall entry point
557(p7) movl r20=sys_ni_syscall
558 ;;
559 mov b6=r20
560 br.call.sptk.many rp=b6 // do the syscall
561.strace_check_retval:
562 cmp.lt p6,p0=r8,r0 // syscall failed?
563 adds r2=PT(R8)+16,sp // r2 = &pt_regs.r8
564 adds r3=PT(R10)+16,sp // r3 = &pt_regs.r10
565 mov r10=0
566(p6) br.cond.sptk strace_error // syscall failed ->
567 ;; // avoid RAW on r10
568.strace_save_retval:
569.mem.offset 0,0; st8.spill [r2]=r8 // store return value in slot for r8
570.mem.offset 8,0; st8.spill [r3]=r10 // clear error indication in slot for r10
571 br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
6f6d7582
JS		 572.ret3:
573(pUStk) cmp.eq.unc p6,p0=r0,r0 // p6 <- pUStk
574 br.cond.sptk .work_pending_syscall_end
1da177e4
LT		 575
576strace_error:
577 ld8 r3=[r2] // load pt_regs.r8
578 sub r9=0,r8 // negate return value to get errno value
579 ;;
580 cmp.ne p6,p0=r3,r0 // is pt_regs.r8!=0?
581 adds r3=16,r2 // r3=&pt_regs.r10
582 ;;
583(p6) mov r10=-1
584(p6) mov r8=r9
585 br.cond.sptk .strace_save_retval
586END(ia64_trace_syscall)
587
588 /*
589 * When traced and returning from sigreturn, we invoke syscall_trace but then
590 * go straight to ia64_leave_kernel rather than ia64_leave_syscall.
591 */
592GLOBAL_ENTRY(ia64_strace_leave_kernel)
593 PT_REGS_UNWIND_INFO(0)
594{ /*
595 * Some versions of gas generate bad unwind info if the first instruction of a
596 * procedure doesn't go into the first slot of a bundle. This is a workaround.
597 */
598 nop.m 0
599 nop.i 0
600 br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
601}
602.ret4: br.cond.sptk ia64_leave_kernel
603END(ia64_strace_leave_kernel)
604
605GLOBAL_ENTRY(ia64_ret_from_clone)
606 PT_REGS_UNWIND_INFO(0)
607{ /*
608 * Some versions of gas generate bad unwind info if the first instruction of a
609 * procedure doesn't go into the first slot of a bundle. This is a workaround.
610 */
611 nop.m 0
612 nop.i 0
613 /*
614 * We need to call schedule_tail() to complete the scheduling process.
615 * Called by ia64_switch_to() after do_fork()->copy_thread(). r8 contains the
616 * address of the previously executing task.
617 */
618 br.call.sptk.many rp=ia64_invoke_schedule_tail
619}
620.ret8:
621 adds r2=TI_FLAGS+IA64_TASK_SIZE,r13
622 ;;
623 ld4 r2=[r2]
624 ;;
625 mov r8=0
626 and r2=_TIF_SYSCALL_TRACEAUDIT,r2
627 ;;
628 cmp.ne p6,p0=r2,r0
629(p6) br.cond.spnt .strace_check_retval
630 ;; // added stop bits to prevent r8 dependency
631END(ia64_ret_from_clone)
632 // fall through
633GLOBAL_ENTRY(ia64_ret_from_syscall)
634 PT_REGS_UNWIND_INFO(0)
635 cmp.ge p6,p7=r8,r0 // syscall executed successfully?
636 adds r2=PT(R8)+16,sp // r2 = &pt_regs.r8
637 mov r10=r0 // clear error indication in r10
638(p7) br.cond.spnt handle_syscall_error // handle potential syscall failure
639END(ia64_ret_from_syscall)
640 // fall through
641/*
642 * ia64_leave_syscall(): Same as ia64_leave_kernel, except that it doesn't
643 * need to switch to bank 0 and doesn't restore the scratch registers.
644 * To avoid leaking kernel bits, the scratch registers are set to
645 * the following known-to-be-safe values:
646 *
647 * r1: restored (global pointer)
648 * r2: cleared
649 * r3: 1 (when returning to user-level)
650 * r8-r11: restored (syscall return value(s))
651 * r12: restored (user-level stack pointer)
652 * r13: restored (user-level thread pointer)
c03f058f 653 * r14: set to __kernel_syscall_via_epc
1da177e4
LT		 654 * r15: restored (syscall #)
655 * r16-r17: cleared
656 * r18: user-level b6
657 * r19: cleared
658 * r20: user-level ar.fpsr
659 * r21: user-level b0
660 * r22: cleared
661 * r23: user-level ar.bspstore
662 * r24: user-level ar.rnat
663 * r25: user-level ar.unat
664 * r26: user-level ar.pfs
665 * r27: user-level ar.rsc
666 * r28: user-level ip
667 * r29: user-level psr
668 * r30: user-level cfm
669 * r31: user-level pr
670 * f6-f11: cleared
671 * pr: restored (user-level pr)
672 * b0: restored (user-level rp)
673 * b6: restored
c03f058f 674 * b7: set to __kernel_syscall_via_epc
1da177e4
LT		 675 * ar.unat: restored (user-level ar.unat)
676 * ar.pfs: restored (user-level ar.pfs)
677 * ar.rsc: restored (user-level ar.rsc)
678 * ar.rnat: restored (user-level ar.rnat)
679 * ar.bspstore: restored (user-level ar.bspstore)
680 * ar.fpsr: restored (user-level ar.fpsr)
681 * ar.ccv: cleared
682 * ar.csd: cleared
683 * ar.ssd: cleared
684 */
685ENTRY(ia64_leave_syscall)
686 PT_REGS_UNWIND_INFO(0)
687 /*
688 * work.need_resched etc. mustn't get changed by this CPU before it returns to
689 * user- or fsys-mode, hence we disable interrupts early on.
690 *
691 * p6 controls whether current_thread_info()->flags needs to be check for
692 * extra work. We always check for extra work when returning to user-level.
693 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
694 * is 0. After extra work processing has been completed, execution
695 * resumes at .work_processed_syscall with p6 set to 1 if the extra-work-check
696 * needs to be redone.
697 */
698#ifdef CONFIG_PREEMPT
699 rsm psr.i // disable interrupts
700 cmp.eq pLvSys,p0=r0,r0 // pLvSys=1: leave from syscall
701(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
702 ;;
703 .pred.rel.mutex pUStk,pKStk
704(pKStk) ld4 r21=[r20] // r21 <- preempt_count
705(pUStk) mov r21=0 // r21 <- 0
706 ;;
707 cmp.eq p6,p0=r21,r0 // p6 <- pUStk || (preempt_count == 0)
708#else /* !CONFIG_PREEMPT */
709(pUStk) rsm psr.i
710 cmp.eq pLvSys,p0=r0,r0 // pLvSys=1: leave from syscall
711(pUStk) cmp.eq.unc p6,p0=r0,r0 // p6 <- pUStk
712#endif
713.work_processed_syscall:
714 adds r2=PT(LOADRS)+16,r12
715 adds r3=PT(AR_BSPSTORE)+16,r12
716 adds r18=TI_FLAGS+IA64_TASK_SIZE,r13
717 ;;
718(p6) ld4 r31=[r18] // load current_thread_info()->flags
719 ld8 r19=[r2],PT(B6)-PT(LOADRS) // load ar.rsc value for "loadrs"
96e01749 720 nop.i 0
1da177e4 721 ;;
87e522a0 722 mov r16=ar.bsp // M2 get existing backing store pointer
1da177e4
LT		 723 ld8 r18=[r2],PT(R9)-PT(B6) // load b6
724(p6) and r15=TIF_WORK_MASK,r31 // any work other than TIF_SYSCALL_TRACE?
725 ;;
87e522a0 726 ld8 r23=[r3],PT(R11)-PT(AR_BSPSTORE) // load ar.bspstore (may be garbage)
1da177e4
LT		 727(p6) cmp4.ne.unc p6,p0=r15, r0 // any special work pending?
728(p6) br.cond.spnt .work_pending_syscall
729 ;;
730 // start restoring the state saved on the kernel stack (struct pt_regs):
731 ld8 r9=[r2],PT(CR_IPSR)-PT(R9)
732 ld8 r11=[r3],PT(CR_IIP)-PT(R11)
87e522a0 733(pNonSys) break 0 // bug check: we shouldn't be here if pNonSys is TRUE!
1da177e4
LT		 734 ;;
735 invala // M0|1 invalidate ALAT
c03f058f
DMT		 736 rsm psr.i | psr.ic // M2 turn off interrupts and interruption collection
737 cmp.eq p9,p0=r0,r0 // A set p9 to indicate that we should restore cr.ifs
1da177e4 738
c03f058f
DMT		 739 ld8 r29=[r2],16 // M0|1 load cr.ipsr
740 ld8 r28=[r3],16 // M0|1 load cr.iip
741 mov r22=r0 // A clear r22
1da177e4
LT		 742 ;;
743 ld8 r30=[r2],16 // M0|1 load cr.ifs
1da177e4 744 ld8 r25=[r3],16 // M0|1 load ar.unat
87e522a0 745(pUStk) add r14=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13
1da177e4
LT		 746 ;;
747 ld8 r26=[r2],PT(B0)-PT(AR_PFS) // M0|1 load ar.pfs
c03f058f 748(pKStk) mov r22=psr // M2 read PSR now that interrupts are disabled
87e522a0 749 nop 0
1da177e4 750 ;;
c03f058f
DMT		 751 ld8 r21=[r2],PT(AR_RNAT)-PT(B0) // M0|1 load b0
752 ld8 r27=[r3],PT(PR)-PT(AR_RSC) // M0|1 load ar.rsc
753 mov f6=f0 // F clear f6
1da177e4 754 ;;
c03f058f
DMT		 755 ld8 r24=[r2],PT(AR_FPSR)-PT(AR_RNAT) // M0|1 load ar.rnat (may be garbage)
756 ld8 r31=[r3],PT(R1)-PT(PR) // M0|1 load predicates
757 mov f7=f0 // F clear f7
1da177e4 758 ;;
c03f058f
DMT		 759 ld8 r20=[r2],PT(R12)-PT(AR_FPSR) // M0|1 load ar.fpsr
760 ld8.fill r1=[r3],16 // M0|1 load r1
761(pUStk) mov r17=1 // A
1da177e4 762 ;;
c03f058f
DMT		 763(pUStk) st1 [r14]=r17 // M2|3
764 ld8.fill r13=[r3],16 // M0|1
765 mov f8=f0 // F clear f8
1da177e4 766 ;;
c03f058f
DMT		 767 ld8.fill r12=[r2] // M0|1 restore r12 (sp)
768 ld8.fill r15=[r3] // M0|1 restore r15
769 mov b6=r18 // I0 restore b6
30325d17 770
c03f058f
DMT		 771 addl r17=THIS_CPU(ia64_phys_stacked_size_p8),r0 // A
772 mov f9=f0 // F clear f9
773(pKStk) br.cond.dpnt.many skip_rbs_switch // B
87e522a0 774
c03f058f
DMT		 775 srlz.d // M0 ensure interruption collection is off (for cover)
776 shr.u r18=r19,16 // I0|1 get byte size of existing "dirty" partition
777 cover // B add current frame into dirty partition & set cr.ifs
1da177e4 778 ;;
c03f058f
DMT		 779(pUStk) ld4 r17=[r17] // M0|1 r17 = cpu_data->phys_stacked_size_p8
780 mov r19=ar.bsp // M2 get new backing store pointer
781 mov f10=f0 // F clear f10
96e01749
DMT		 782
783 nop.m 0
c03f058f 784 movl r14=__kernel_syscall_via_epc // X
1da177e4 785 ;;
c03f058f
DMT		 786 mov.m ar.csd=r0 // M2 clear ar.csd
787 mov.m ar.ccv=r0 // M2 clear ar.ccv
788 mov b7=r14 // I0 clear b7 (hint with __kernel_syscall_via_epc)
1da177e4 789
c03f058f
DMT		 790 mov.m ar.ssd=r0 // M2 clear ar.ssd
791 mov f11=f0 // F clear f11
792 br.cond.sptk.many rbs_switch // B
1da177e4
LT		 793END(ia64_leave_syscall)
794
795#ifdef CONFIG_IA32_SUPPORT
796GLOBAL_ENTRY(ia64_ret_from_ia32_execve)
797 PT_REGS_UNWIND_INFO(0)
798 adds r2=PT(R8)+16,sp // r2 = &pt_regs.r8
799 adds r3=PT(R10)+16,sp // r3 = &pt_regs.r10
800 ;;
801 .mem.offset 0,0
802 st8.spill [r2]=r8 // store return value in slot for r8 and set unat bit
803 .mem.offset 8,0
804 st8.spill [r3]=r0 // clear error indication in slot for r10 and set unat bit
9df6f705 805END(ia64_ret_from_ia32_execve)
1da177e4
LT		 806 // fall through
807#endif /* CONFIG_IA32_SUPPORT */
808GLOBAL_ENTRY(ia64_leave_kernel)
809 PT_REGS_UNWIND_INFO(0)
810 /*
811 * work.need_resched etc. mustn't get changed by this CPU before it returns to
812 * user- or fsys-mode, hence we disable interrupts early on.
813 *
814 * p6 controls whether current_thread_info()->flags needs to be check for
815 * extra work. We always check for extra work when returning to user-level.
816 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
817 * is 0. After extra work processing has been completed, execution
818 * resumes at .work_processed_syscall with p6 set to 1 if the extra-work-check
819 * needs to be redone.
820 */
821#ifdef CONFIG_PREEMPT
822 rsm psr.i // disable interrupts
823 cmp.eq p0,pLvSys=r0,r0 // pLvSys=0: leave from kernel
824(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
825 ;;
826 .pred.rel.mutex pUStk,pKStk
827(pKStk) ld4 r21=[r20] // r21 <- preempt_count
828(pUStk) mov r21=0 // r21 <- 0
829 ;;
830 cmp.eq p6,p0=r21,r0 // p6 <- pUStk || (preempt_count == 0)
831#else
832(pUStk) rsm psr.i
833 cmp.eq p0,pLvSys=r0,r0 // pLvSys=0: leave from kernel
834(pUStk) cmp.eq.unc p6,p0=r0,r0 // p6 <- pUStk
835#endif
836.work_processed_kernel:
837 adds r17=TI_FLAGS+IA64_TASK_SIZE,r13
838 ;;
839(p6) ld4 r31=[r17] // load current_thread_info()->flags
840 adds r21=PT(PR)+16,r12
841 ;;
842
843 lfetch [r21],PT(CR_IPSR)-PT(PR)
844 adds r2=PT(B6)+16,r12
845 adds r3=PT(R16)+16,r12
846 ;;
847 lfetch [r21]
848 ld8 r28=[r2],8 // load b6
849 adds r29=PT(R24)+16,r12
850
851 ld8.fill r16=[r3],PT(AR_CSD)-PT(R16)
852 adds r30=PT(AR_CCV)+16,r12
853(p6) and r19=TIF_WORK_MASK,r31 // any work other than TIF_SYSCALL_TRACE?
854 ;;
855 ld8.fill r24=[r29]
856 ld8 r15=[r30] // load ar.ccv
857(p6) cmp4.ne.unc p6,p0=r19, r0 // any special work pending?
858 ;;
859 ld8 r29=[r2],16 // load b7
860 ld8 r30=[r3],16 // load ar.csd
861(p6) br.cond.spnt .work_pending
862 ;;
863 ld8 r31=[r2],16 // load ar.ssd
864 ld8.fill r8=[r3],16
865 ;;
866 ld8.fill r9=[r2],16
867 ld8.fill r10=[r3],PT(R17)-PT(R10)
868 ;;
869 ld8.fill r11=[r2],PT(R18)-PT(R11)
870 ld8.fill r17=[r3],16
871 ;;
872 ld8.fill r18=[r2],16
873 ld8.fill r19=[r3],16
874 ;;
875 ld8.fill r20=[r2],16
876 ld8.fill r21=[r3],16
877 mov ar.csd=r30
878 mov ar.ssd=r31
879 ;;
880 rsm psr.i | psr.ic // initiate turning off of interrupt and interruption collection
881 invala // invalidate ALAT
882 ;;
883 ld8.fill r22=[r2],24
884 ld8.fill r23=[r3],24
885 mov b6=r28
886 ;;
887 ld8.fill r25=[r2],16
888 ld8.fill r26=[r3],16
889 mov b7=r29
890 ;;
891 ld8.fill r27=[r2],16
892 ld8.fill r28=[r3],16
893 ;;
894 ld8.fill r29=[r2],16
895 ld8.fill r30=[r3],24
896 ;;
897 ld8.fill r31=[r2],PT(F9)-PT(R31)
898 adds r3=PT(F10)-PT(F6),r3
899 ;;
900 ldf.fill f9=[r2],PT(F6)-PT(F9)
901 ldf.fill f10=[r3],PT(F8)-PT(F10)
902 ;;
903 ldf.fill f6=[r2],PT(F7)-PT(F6)
904 ;;
905 ldf.fill f7=[r2],PT(F11)-PT(F7)
906 ldf.fill f8=[r3],32
907 ;;
e7e965fa 908 srlz.d // ensure that inter. collection is off (VHPT is don't care, since text is pinned)
1da177e4
LT		 909 mov ar.ccv=r15
910 ;;
911 ldf.fill f11=[r2]
912 bsw.0 // switch back to bank 0 (no stop bit required beforehand...)
913 ;;
914(pUStk) mov r18=IA64_KR(CURRENT)// M2 (12 cycle read latency)
915 adds r16=PT(CR_IPSR)+16,r12
916 adds r17=PT(CR_IIP)+16,r12
917
918(pKStk) mov r22=psr // M2 read PSR now that interrupts are disabled
919 nop.i 0
920 nop.i 0
921 ;;
922 ld8 r29=[r16],16 // load cr.ipsr
923 ld8 r28=[r17],16 // load cr.iip
924 ;;
925 ld8 r30=[r16],16 // load cr.ifs
926 ld8 r25=[r17],16 // load ar.unat
927 ;;
928 ld8 r26=[r16],16 // load ar.pfs
929 ld8 r27=[r17],16 // load ar.rsc
930 cmp.eq p9,p0=r0,r0 // set p9 to indicate that we should restore cr.ifs
931 ;;
932 ld8 r24=[r16],16 // load ar.rnat (may be garbage)
933 ld8 r23=[r17],16 // load ar.bspstore (may be garbage)
934 ;;
935 ld8 r31=[r16],16 // load predicates
936 ld8 r21=[r17],16 // load b0
937 ;;
938 ld8 r19=[r16],16 // load ar.rsc value for "loadrs"
939 ld8.fill r1=[r17],16 // load r1
940 ;;
941 ld8.fill r12=[r16],16
942 ld8.fill r13=[r17],16
943(pUStk) adds r18=IA64_TASK_THREAD_ON_USTACK_OFFSET,r18
944 ;;
945 ld8 r20=[r16],16 // ar.fpsr
946 ld8.fill r15=[r17],16
947 ;;
948 ld8.fill r14=[r16],16
949 ld8.fill r2=[r17]
950(pUStk) mov r17=1
951 ;;
952 ld8.fill r3=[r16]
953(pUStk) st1 [r18]=r17 // restore current->thread.on_ustack
954 shr.u r18=r19,16 // get byte size of existing "dirty" partition
955 ;;
956 mov r16=ar.bsp // get existing backing store pointer
957 addl r17=THIS_CPU(ia64_phys_stacked_size_p8),r0
958 ;;
959 ld4 r17=[r17] // r17 = cpu_data->phys_stacked_size_p8
960(pKStk) br.cond.dpnt skip_rbs_switch
961
962 /*
963 * Restore user backing store.
964 *
965 * NOTE: alloc, loadrs, and cover can't be predicated.
966 */
967(pNonSys) br.cond.dpnt dont_preserve_current_frame
1da177e4
LT		 968 cover // add current frame into dirty partition and set cr.ifs
969 ;;
970 mov r19=ar.bsp // get new backing store pointer
87e522a0 971rbs_switch:
1da177e4
LT		 972 sub r16=r16,r18 // krbs = old bsp - size of dirty partition
973 cmp.ne p9,p0=r0,r0 // clear p9 to skip restore of cr.ifs
974 ;;
975 sub r19=r19,r16 // calculate total byte size of dirty partition
976 add r18=64,r18 // don't force in0-in7 into memory...
977 ;;
978 shl r19=r19,16 // shift size of dirty partition into loadrs position
979 ;;
980dont_preserve_current_frame:
981 /*
982 * To prevent leaking bits between the kernel and user-space,
983 * we must clear the stacked registers in the "invalid" partition here.
984 * Not pretty, but at least it's fast (3.34 registers/cycle on Itanium,
985 * 5 registers/cycle on McKinley).
986 */
987# define pRecurse p6
988# define pReturn p7
989#ifdef CONFIG_ITANIUM
990# define Nregs 10
991#else
992# define Nregs 14
993#endif
994 alloc loc0=ar.pfs,2,Nregs-2,2,0
995 shr.u loc1=r18,9 // RNaTslots <= floor(dirtySize / (64*8))
996 sub r17=r17,r18 // r17 = (physStackedSize + 8) - dirtySize
997 ;;
998 mov ar.rsc=r19 // load ar.rsc to be used for "loadrs"
999 shladd in0=loc1,3,r17
1000 mov in1=0
1001 ;;
1002 TEXT_ALIGN(32)
1003rse_clear_invalid:
1004#ifdef CONFIG_ITANIUM
1005 // cycle 0
1006 { .mii
1007 alloc loc0=ar.pfs,2,Nregs-2,2,0
1008 cmp.lt pRecurse,p0=Nregs*8,in0 // if more than Nregs regs left to clear, (re)curse
1009 add out0=-Nregs*8,in0
1010}{ .mfb
1011 add out1=1,in1 // increment recursion count
1012 nop.f 0
1013 nop.b 0 // can't do br.call here because of alloc (WAW on CFM)
1014 ;;
1015}{ .mfi // cycle 1
1016 mov loc1=0
1017 nop.f 0
1018 mov loc2=0
1019}{ .mib
1020 mov loc3=0
1021 mov loc4=0
1022(pRecurse) br.call.sptk.many b0=rse_clear_invalid
1023
1024}{ .mfi // cycle 2
1025 mov loc5=0
1026 nop.f 0
1027 cmp.ne pReturn,p0=r0,in1 // if recursion count != 0, we need to do a br.ret
1028}{ .mib
1029 mov loc6=0
1030 mov loc7=0
1031(pReturn) br.ret.sptk.many b0
1032}
1033#else /* !CONFIG_ITANIUM */
1034 alloc loc0=ar.pfs,2,Nregs-2,2,0
1035 cmp.lt pRecurse,p0=Nregs*8,in0 // if more than Nregs regs left to clear, (re)curse
1036 add out0=-Nregs*8,in0
1037 add out1=1,in1 // increment recursion count
1038 mov loc1=0
1039 mov loc2=0
1040 ;;
1041 mov loc3=0
1042 mov loc4=0
1043 mov loc5=0
1044 mov loc6=0
1045 mov loc7=0
9ec1a7ad 1046(pRecurse) br.call.dptk.few b0=rse_clear_invalid
1da177e4
LT		 1047 ;;
1048 mov loc8=0
1049 mov loc9=0
1050 cmp.ne pReturn,p0=r0,in1 // if recursion count != 0, we need to do a br.ret
1051 mov loc10=0
1052 mov loc11=0
9ec1a7ad 1053(pReturn) br.ret.dptk.many b0
1da177e4
LT		 1054#endif /* !CONFIG_ITANIUM */
1055# undef pRecurse
1056# undef pReturn
1057 ;;
1058 alloc r17=ar.pfs,0,0,0,0 // drop current register frame
1059 ;;
1060 loadrs
1061 ;;
1062skip_rbs_switch:
1063 mov ar.unat=r25 // M2
1064(pKStk) extr.u r22=r22,21,1 // I0 extract current value of psr.pp from r22
1065(pLvSys)mov r19=r0 // A clear r19 for leave_syscall, no-op otherwise
1066 ;;
1067(pUStk) mov ar.bspstore=r23 // M2
1068(pKStk) dep r29=r22,r29,21,1 // I0 update ipsr.pp with psr.pp
1069(pLvSys)mov r16=r0 // A clear r16 for leave_syscall, no-op otherwise
1070 ;;
1071 mov cr.ipsr=r29 // M2
1072 mov ar.pfs=r26 // I0
1073(pLvSys)mov r17=r0 // A clear r17 for leave_syscall, no-op otherwise
1074
1075(p9) mov cr.ifs=r30 // M2
1076 mov b0=r21 // I0
1077(pLvSys)mov r18=r0 // A clear r18 for leave_syscall, no-op otherwise
1078
1079 mov ar.fpsr=r20 // M2
1080 mov cr.iip=r28 // M2
1081 nop 0
1082 ;;
1083(pUStk) mov ar.rnat=r24 // M2 must happen with RSE in lazy mode
1084 nop 0
1085(pLvSys)mov r2=r0
1086
1087 mov ar.rsc=r27 // M2
1088 mov pr=r31,-1 // I0
1089 rfi // B
1090
1091 /*
1092 * On entry:
1093 * r20 = &current->thread_info->pre_count (if CONFIG_PREEMPT)
1094 * r31 = current->thread_info->flags
1095 * On exit:
1096 * p6 = TRUE if work-pending-check needs to be redone
1097 */
1098.work_pending_syscall:
1099 add r2=-8,r2
1100 add r3=-8,r3
1101 ;;
1102 st8 [r2]=r8
1103 st8 [r3]=r10
1104.work_pending:
1da177e4
LT		 1105 tbit.z p6,p0=r31,TIF_NEED_RESCHED // current_thread_info()->need_resched==0?
1106(p6) br.cond.sptk.few .notify
1107#ifdef CONFIG_PREEMPT
1108(pKStk) dep r21=-1,r0,PREEMPT_ACTIVE_BIT,1
1109 ;;
1110(pKStk) st4 [r20]=r21
1111 ssm psr.i // enable interrupts
1112#endif
1113 br.call.spnt.many rp=schedule
1114.ret9: cmp.eq p6,p0=r0,r0 // p6 <- 1
1115 rsm psr.i // disable interrupts
1116 ;;
1117#ifdef CONFIG_PREEMPT
1118(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
1119 ;;
1120(pKStk) st4 [r20]=r0 // preempt_count() <- 0
1121#endif
1122(pLvSys)br.cond.sptk.few .work_pending_syscall_end
1123 br.cond.sptk.many .work_processed_kernel // re-check
1124
1125.notify:
1126(pUStk) br.call.spnt.many rp=notify_resume_user
1127.ret10: cmp.ne p6,p0=r0,r0 // p6 <- 0
1128(pLvSys)br.cond.sptk.few .work_pending_syscall_end
1129 br.cond.sptk.many .work_processed_kernel // don't re-check
1130
1da177e4
LT		 1131.work_pending_syscall_end:
1132 adds r2=PT(R8)+16,r12
1133 adds r3=PT(R10)+16,r12
1134 ;;
1135 ld8 r8=[r2]
1136 ld8 r10=[r3]
1137 br.cond.sptk.many .work_processed_syscall // re-check
1138
1139END(ia64_leave_kernel)
1140
1141ENTRY(handle_syscall_error)
1142 /*
1143 * Some system calls (e.g., ptrace, mmap) can return arbitrary values which could
1144 * lead us to mistake a negative return value as a failed syscall. Those syscall
1145 * must deposit a non-zero value in pt_regs.r8 to indicate an error. If
1146 * pt_regs.r8 is zero, we assume that the call completed successfully.
1147 */
1148 PT_REGS_UNWIND_INFO(0)
1149 ld8 r3=[r2] // load pt_regs.r8
1150 ;;
1151 cmp.eq p6,p7=r3,r0 // is pt_regs.r8==0?
1152 ;;
1153(p7) mov r10=-1
1154(p7) sub r8=0,r8 // negate return value to get errno
1155 br.cond.sptk ia64_leave_syscall
1156END(handle_syscall_error)
1157
1158 /*
1159 * Invoke schedule_tail(task) while preserving in0-in7, which may be needed
1160 * in case a system call gets restarted.
1161 */
1162GLOBAL_ENTRY(ia64_invoke_schedule_tail)
1163 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1164 alloc loc1=ar.pfs,8,2,1,0
1165 mov loc0=rp
1166 mov out0=r8 // Address of previous task
1167 ;;
1168 br.call.sptk.many rp=schedule_tail
1169.ret11: mov ar.pfs=loc1
1170 mov rp=loc0
1171 br.ret.sptk.many rp
1172END(ia64_invoke_schedule_tail)
1173
1174 /*
1175 * Setup stack and call do_notify_resume_user(). Note that pSys and pNonSys need to
1176 * be set up by the caller. We declare 8 input registers so the system call
1177 * args get preserved, in case we need to restart a system call.
1178 */
1179ENTRY(notify_resume_user)
1180 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1181 alloc loc1=ar.pfs,8,2,3,0 // preserve all eight input regs in case of syscall restart!
1182 mov r9=ar.unat
1183 mov loc0=rp // save return address
1184 mov out0=0 // there is no "oldset"
1185 adds out1=8,sp // out1=&sigscratch->ar_pfs
1186(pSys) mov out2=1 // out2==1 => we're in a syscall
1187 ;;
1188(pNonSys) mov out2=0 // out2==0 => not a syscall
1189 .fframe 16
bfd68594 1190 .spillsp ar.unat, 16
1da177e4
LT		 1191 st8 [sp]=r9,-16 // allocate space for ar.unat and save it
1192 st8 [out1]=loc1,-8 // save ar.pfs, out1=&sigscratch
1193 .body
1194 br.call.sptk.many rp=do_notify_resume_user
1195.ret15: .restore sp
1196 adds sp=16,sp // pop scratch stack space
1197 ;;
1198 ld8 r9=[sp] // load new unat from sigscratch->scratch_unat
1199 mov rp=loc0
1200 ;;
1201 mov ar.unat=r9
1202 mov ar.pfs=loc1
1203 br.ret.sptk.many rp
1204END(notify_resume_user)
1205
1206GLOBAL_ENTRY(sys_rt_sigsuspend)
1207 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1208 alloc loc1=ar.pfs,8,2,3,0 // preserve all eight input regs in case of syscall restart!
1209 mov r9=ar.unat
1210 mov loc0=rp // save return address
1211 mov out0=in0 // mask
1212 mov out1=in1 // sigsetsize
1213 adds out2=8,sp // out2=&sigscratch->ar_pfs
1214 ;;
1215 .fframe 16
bfd68594 1216 .spillsp ar.unat, 16
1da177e4
LT		 1217 st8 [sp]=r9,-16 // allocate space for ar.unat and save it
1218 st8 [out2]=loc1,-8 // save ar.pfs, out2=&sigscratch
1219 .body
1220 br.call.sptk.many rp=ia64_rt_sigsuspend
1221.ret17: .restore sp
1222 adds sp=16,sp // pop scratch stack space
1223 ;;
1224 ld8 r9=[sp] // load new unat from sw->caller_unat
1225 mov rp=loc0
1226 ;;
1227 mov ar.unat=r9
1228 mov ar.pfs=loc1
1229 br.ret.sptk.many rp
1230END(sys_rt_sigsuspend)
1231
1232ENTRY(sys_rt_sigreturn)
1233 PT_REGS_UNWIND_INFO(0)
1234 /*
1235 * Allocate 8 input registers since ptrace() may clobber them
1236 */
1237 alloc r2=ar.pfs,8,0,1,0
1238 .prologue
1239 PT_REGS_SAVES(16)
1240 adds sp=-16,sp
1241 .body
1242 cmp.eq pNonSys,pSys=r0,r0 // sigreturn isn't a normal syscall...
1243 ;;
1244 /*
1245 * leave_kernel() restores f6-f11 from pt_regs, but since the streamlined
1246 * syscall-entry path does not save them we save them here instead. Note: we
1247 * don't need to save any other registers that are not saved by the stream-lined
1248 * syscall path, because restore_sigcontext() restores them.
1249 */
1250 adds r16=PT(F6)+32,sp
1251 adds r17=PT(F7)+32,sp
1252 ;;
1253 stf.spill [r16]=f6,32
1254 stf.spill [r17]=f7,32
1255 ;;
1256 stf.spill [r16]=f8,32
1257 stf.spill [r17]=f9,32
1258 ;;
1259 stf.spill [r16]=f10
1260 stf.spill [r17]=f11
1261 adds out0=16,sp // out0 = &sigscratch
1262 br.call.sptk.many rp=ia64_rt_sigreturn
763b3917 1263.ret19: .restore sp,0
1da177e4
LT		 1264 adds sp=16,sp
1265 ;;
1266 ld8 r9=[sp] // load new ar.unat
1267 mov.sptk b7=r8,ia64_leave_kernel
1268 ;;
1269 mov ar.unat=r9
1270 br.many b7
1271END(sys_rt_sigreturn)
1272
1273GLOBAL_ENTRY(ia64_prepare_handle_unaligned)
1274 .prologue
1275 /*
1276 * r16 = fake ar.pfs, we simply need to make sure privilege is still 0
1277 */
1278 mov r16=r0
1279 DO_SAVE_SWITCH_STACK
1280 br.call.sptk.many rp=ia64_handle_unaligned // stack frame setup in ivt
1281.ret21: .body
1282 DO_LOAD_SWITCH_STACK
1283 br.cond.sptk.many rp // goes to ia64_leave_kernel
1284END(ia64_prepare_handle_unaligned)
1285
1286 //
1287 // unw_init_running(void (*callback)(info, arg), void *arg)
1288 //
1289# define EXTRA_FRAME_SIZE ((UNW_FRAME_INFO_SIZE+15)&~15)
1290
1291GLOBAL_ENTRY(unw_init_running)
1292 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1293 alloc loc1=ar.pfs,2,3,3,0
1294 ;;
1295 ld8 loc2=[in0],8
1296 mov loc0=rp
1297 mov r16=loc1
1298 DO_SAVE_SWITCH_STACK
1299 .body
1300
1301 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1302 .fframe IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE
1303 SWITCH_STACK_SAVES(EXTRA_FRAME_SIZE)
1304 adds sp=-EXTRA_FRAME_SIZE,sp
1305 .body
1306 ;;
1307 adds out0=16,sp // &info
1308 mov out1=r13 // current
1309 adds out2=16+EXTRA_FRAME_SIZE,sp // &switch_stack
1310 br.call.sptk.many rp=unw_init_frame_info
13111: adds out0=16,sp // &info
1312 mov b6=loc2
1313 mov loc2=gp // save gp across indirect function call
1314 ;;
1315 ld8 gp=[in0]
1316 mov out1=in1 // arg
1317 br.call.sptk.many rp=b6 // invoke the callback function
13181: mov gp=loc2 // restore gp
1319
1320 // For now, we don't allow changing registers from within
1321 // unw_init_running; if we ever want to allow that, we'd
1322 // have to do a load_switch_stack here:
1323 .restore sp
1324 adds sp=IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE,sp
1325
1326 mov ar.pfs=loc1
1327 mov rp=loc0
1328 br.ret.sptk.many rp
1329END(unw_init_running)
1330
1331 .rodata
1332 .align 8
1333 .globl sys_call_table
1334sys_call_table:
1335 data8 sys_ni_syscall // This must be sys_ni_syscall! See ivt.S.
1336 data8 sys_exit // 1025
1337 data8 sys_read
1338 data8 sys_write
1339 data8 sys_open
1340 data8 sys_close
1341 data8 sys_creat // 1030
1342 data8 sys_link
1343 data8 sys_unlink
1344 data8 ia64_execve
1345 data8 sys_chdir
1346 data8 sys_fchdir // 1035
1347 data8 sys_utimes
1348 data8 sys_mknod
1349 data8 sys_chmod
1350 data8 sys_chown
1351 data8 sys_lseek // 1040
1352 data8 sys_getpid
1353 data8 sys_getppid
1354 data8 sys_mount
1355 data8 sys_umount
1356 data8 sys_setuid // 1045
1357 data8 sys_getuid
1358 data8 sys_geteuid
1359 data8 sys_ptrace
1360 data8 sys_access
1361 data8 sys_sync // 1050
1362 data8 sys_fsync
1363 data8 sys_fdatasync
1364 data8 sys_kill
1365 data8 sys_rename
1366 data8 sys_mkdir // 1055
1367 data8 sys_rmdir
1368 data8 sys_dup
1369 data8 sys_pipe
1370 data8 sys_times
1371 data8 ia64_brk // 1060
1372 data8 sys_setgid
1373 data8 sys_getgid
1374 data8 sys_getegid
1375 data8 sys_acct
1376 data8 sys_ioctl // 1065
1377 data8 sys_fcntl
1378 data8 sys_umask
1379 data8 sys_chroot
1380 data8 sys_ustat
1381 data8 sys_dup2 // 1070
1382 data8 sys_setreuid
1383 data8 sys_setregid
1384 data8 sys_getresuid
1385 data8 sys_setresuid
1386 data8 sys_getresgid // 1075
1387 data8 sys_setresgid
1388 data8 sys_getgroups
1389 data8 sys_setgroups
1390 data8 sys_getpgid
1391 data8 sys_setpgid // 1080
1392 data8 sys_setsid
1393 data8 sys_getsid
1394 data8 sys_sethostname
1395 data8 sys_setrlimit
1396 data8 sys_getrlimit // 1085
1397 data8 sys_getrusage
1398 data8 sys_gettimeofday
1399 data8 sys_settimeofday
1400 data8 sys_select
1401 data8 sys_poll // 1090
1402 data8 sys_symlink
1403 data8 sys_readlink
1404 data8 sys_uselib
1405 data8 sys_swapon
1406 data8 sys_swapoff // 1095
1407 data8 sys_reboot
1408 data8 sys_truncate
1409 data8 sys_ftruncate
1410 data8 sys_fchmod
1411 data8 sys_fchown // 1100
1412 data8 ia64_getpriority
1413 data8 sys_setpriority
1414 data8 sys_statfs
1415 data8 sys_fstatfs
1416 data8 sys_gettid // 1105
1417 data8 sys_semget
1418 data8 sys_semop
1419 data8 sys_semctl
1420 data8 sys_msgget
1421 data8 sys_msgsnd // 1110
1422 data8 sys_msgrcv
1423 data8 sys_msgctl
1424 data8 sys_shmget
7d87e14c 1425 data8 sys_shmat
1da177e4
LT		 1426 data8 sys_shmdt // 1115
1427 data8 sys_shmctl
1428 data8 sys_syslog
1429 data8 sys_setitimer
1430 data8 sys_getitimer
1431 data8 sys_ni_syscall // 1120 /* was: ia64_oldstat */
1432 data8 sys_ni_syscall /* was: ia64_oldlstat */
1433 data8 sys_ni_syscall /* was: ia64_oldfstat */
1434 data8 sys_vhangup
1435 data8 sys_lchown
1436 data8 sys_remap_file_pages // 1125
1437 data8 sys_wait4
1438 data8 sys_sysinfo
1439 data8 sys_clone
1440 data8 sys_setdomainname
1441 data8 sys_newuname // 1130
1442 data8 sys_adjtimex
1443 data8 sys_ni_syscall /* was: ia64_create_module */
1444 data8 sys_init_module
1445 data8 sys_delete_module
1446 data8 sys_ni_syscall // 1135 /* was: sys_get_kernel_syms */
1447 data8 sys_ni_syscall /* was: sys_query_module */
1448 data8 sys_quotactl
1449 data8 sys_bdflush
1450 data8 sys_sysfs
1451 data8 sys_personality // 1140
1452 data8 sys_ni_syscall // sys_afs_syscall
1453 data8 sys_setfsuid
1454 data8 sys_setfsgid
1455 data8 sys_getdents
1456 data8 sys_flock // 1145
1457 data8 sys_readv
1458 data8 sys_writev
1459 data8 sys_pread64
1460 data8 sys_pwrite64
1461 data8 sys_sysctl // 1150
1462 data8 sys_mmap
1463 data8 sys_munmap
1464 data8 sys_mlock
1465 data8 sys_mlockall
1466 data8 sys_mprotect // 1155
1467 data8 ia64_mremap
1468 data8 sys_msync
1469 data8 sys_munlock
1470 data8 sys_munlockall
1471 data8 sys_sched_getparam // 1160
1472 data8 sys_sched_setparam
1473 data8 sys_sched_getscheduler
1474 data8 sys_sched_setscheduler
1475 data8 sys_sched_yield
1476 data8 sys_sched_get_priority_max // 1165
1477 data8 sys_sched_get_priority_min
1478 data8 sys_sched_rr_get_interval
1479 data8 sys_nanosleep
1480 data8 sys_nfsservctl
1481 data8 sys_prctl // 1170
1482 data8 sys_getpagesize
1483 data8 sys_mmap2
1484 data8 sys_pciconfig_read
1485 data8 sys_pciconfig_write
1486 data8 sys_perfmonctl // 1175
1487 data8 sys_sigaltstack
1488 data8 sys_rt_sigaction
1489 data8 sys_rt_sigpending
1490 data8 sys_rt_sigprocmask
1491 data8 sys_rt_sigqueueinfo // 1180
1492 data8 sys_rt_sigreturn
1493 data8 sys_rt_sigsuspend
1494 data8 sys_rt_sigtimedwait
1495 data8 sys_getcwd
1496 data8 sys_capget // 1185
1497 data8 sys_capset
1498 data8 sys_sendfile64
1499 data8 sys_ni_syscall // sys_getpmsg (STREAMS)
1500 data8 sys_ni_syscall // sys_putpmsg (STREAMS)
1501 data8 sys_socket // 1190
1502 data8 sys_bind
1503 data8 sys_connect
1504 data8 sys_listen
1505 data8 sys_accept
1506 data8 sys_getsockname // 1195
1507 data8 sys_getpeername
1508 data8 sys_socketpair
1509 data8 sys_send
1510 data8 sys_sendto
1511 data8 sys_recv // 1200
1512 data8 sys_recvfrom
1513 data8 sys_shutdown
1514 data8 sys_setsockopt
1515 data8 sys_getsockopt
1516 data8 sys_sendmsg // 1205
1517 data8 sys_recvmsg
1518 data8 sys_pivot_root
1519 data8 sys_mincore
1520 data8 sys_madvise
1521 data8 sys_newstat // 1210
1522 data8 sys_newlstat
1523 data8 sys_newfstat
1524 data8 sys_clone2
1525 data8 sys_getdents64
1526 data8 sys_getunwind // 1215
1527 data8 sys_readahead
1528 data8 sys_setxattr
1529 data8 sys_lsetxattr
1530 data8 sys_fsetxattr
1531 data8 sys_getxattr // 1220
1532 data8 sys_lgetxattr
1533 data8 sys_fgetxattr
1534 data8 sys_listxattr
1535 data8 sys_llistxattr
1536 data8 sys_flistxattr // 1225
1537 data8 sys_removexattr
1538 data8 sys_lremovexattr
1539 data8 sys_fremovexattr
1540 data8 sys_tkill
1541 data8 sys_futex // 1230
1542 data8 sys_sched_setaffinity
1543 data8 sys_sched_getaffinity
1544 data8 sys_set_tid_address
1545 data8 sys_fadvise64_64
1546 data8 sys_tgkill // 1235
1547 data8 sys_exit_group
1548 data8 sys_lookup_dcookie
1549 data8 sys_io_setup
1550 data8 sys_io_destroy
1551 data8 sys_io_getevents // 1240
1552 data8 sys_io_submit
1553 data8 sys_io_cancel
1554 data8 sys_epoll_create
1555 data8 sys_epoll_ctl
1556 data8 sys_epoll_wait // 1245
1557 data8 sys_restart_syscall
1558 data8 sys_semtimedop
1559 data8 sys_timer_create
1560 data8 sys_timer_settime
1561 data8 sys_timer_gettime // 1250
1562 data8 sys_timer_getoverrun
1563 data8 sys_timer_delete
1564 data8 sys_clock_settime
1565 data8 sys_clock_gettime
1566 data8 sys_clock_getres // 1255
1567 data8 sys_clock_nanosleep
1568 data8 sys_fstatfs64
1569 data8 sys_statfs64
1570 data8 sys_mbind
1571 data8 sys_get_mempolicy // 1260
1572 data8 sys_set_mempolicy
1573 data8 sys_mq_open
1574 data8 sys_mq_unlink
1575 data8 sys_mq_timedsend
1576 data8 sys_mq_timedreceive // 1265
1577 data8 sys_mq_notify
1578 data8 sys_mq_getsetattr
1579 data8 sys_ni_syscall // reserved for kexec_load
1580 data8 sys_ni_syscall // reserved for vserver
1581 data8 sys_waitid // 1270
1582 data8 sys_add_key
1583 data8 sys_request_key
1584 data8 sys_keyctl
22e2c507
JA		 1585 data8 sys_ioprio_set
1586 data8 sys_ioprio_get // 1275
6cb54819 1587 data8 sys_ni_syscall
d108919b
RL		 1588 data8 sys_inotify_init
1589 data8 sys_inotify_add_watch
1590 data8 sys_inotify_rm_watch
39743889 1591 data8 sys_migrate_pages // 1280
9ed2ad86
CK		 1592 data8 sys_openat
1593 data8 sys_mkdirat
1594 data8 sys_mknodat
1595 data8 sys_fchownat
1596 data8 sys_futimesat // 1285
1597 data8 sys_newfstatat
1598 data8 sys_unlinkat
1599 data8 sys_renameat
1600 data8 sys_linkat
1601 data8 sys_symlinkat // 1290
1602 data8 sys_readlinkat
1603 data8 sys_fchmodat
1604 data8 sys_faccessat
9621a4ef
JD		 1605 data8 sys_ni_syscall // reserved for pselect
1606 data8 sys_ni_syscall // 1295 reserved for ppoll
1607 data8 sys_unshare
1da177e4
LT		 1608
1609 .org sys_call_table + 8*NR_syscalls // guard against failures to increase NR_syscalls
Linux 6.x block layer and io_uring tree(s)