aboutsummaryrefslogtreecommitdiffstats
path: root/arch/ia64/kernel/entry.S
diff options
context:
space:
mode:
Diffstat (limited to 'arch/ia64/kernel/entry.S')
-rw-r--r--arch/ia64/kernel/entry.S1587
1 files changed, 1587 insertions, 0 deletions
diff --git a/arch/ia64/kernel/entry.S b/arch/ia64/kernel/entry.S
new file mode 100644
index 000000000000..0272c010a3ba
--- /dev/null
+++ b/arch/ia64/kernel/entry.S
@@ -0,0 +1,1587 @@
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>
40#include <asm/offsets.h>
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:
207(p6) ssm psr.ic // if we had to map, reenable the psr.ic bit FIRST!!!
208 ;;
209(p6) srlz.d
210 ld8 sp=[r21] // load kernel stack pointer of new task
211 mov IA64_KR(CURRENT)=in0 // update "current" application register
212 mov r8=r13 // return pointer to previously running task
213 mov r13=in0 // set "current" pointer
214 ;;
215 DO_LOAD_SWITCH_STACK
216
217#ifdef CONFIG_SMP
218 sync.i // ensure "fc"s done by this CPU are visible on other CPUs
219#endif
220 br.ret.sptk.many rp // boogie on out in new context
221
222.map:
223 rsm psr.ic // interrupts (psr.i) are already disabled here
224 movl r25=PAGE_KERNEL
225 ;;
226 srlz.d
227 or r23=r25,r20 // construct PA | page properties
228 mov r25=IA64_GRANULE_SHIFT<<2
229 ;;
230 mov cr.itir=r25
231 mov cr.ifa=in0 // VA of next task...
232 ;;
233 mov r25=IA64_TR_CURRENT_STACK
234 mov IA64_KR(CURRENT_STACK)=r26 // remember last page we mapped...
235 ;;
236 itr.d dtr[r25]=r23 // wire in new mapping...
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
473GLOBAL_ENTRY(__ia64_syscall)
474 .regstk 6,0,0,0
475 mov r15=in5 // put syscall number in place
476 break __BREAK_SYSCALL
477 movl r2=errno
478 cmp.eq p6,p7=-1,r10
479 ;;
480(p6) st4 [r2]=r8
481(p6) mov r8=-1
482 br.ret.sptk.many rp
483END(__ia64_syscall)
484
485GLOBAL_ENTRY(execve)
486 mov r15=__NR_execve // put syscall number in place
487 break __BREAK_SYSCALL
488 br.ret.sptk.many rp
489END(execve)
490
491GLOBAL_ENTRY(clone)
492 mov r15=__NR_clone // put syscall number in place
493 break __BREAK_SYSCALL
494 br.ret.sptk.many rp
495END(clone)
496
497 /*
498 * Invoke a system call, but do some tracing before and after the call.
499 * We MUST preserve the current register frame throughout this routine
500 * because some system calls (such as ia64_execve) directly
501 * manipulate ar.pfs.
502 */
503GLOBAL_ENTRY(ia64_trace_syscall)
504 PT_REGS_UNWIND_INFO(0)
505 /*
506 * We need to preserve the scratch registers f6-f11 in case the system
507 * call is sigreturn.
508 */
509 adds r16=PT(F6)+16,sp
510 adds r17=PT(F7)+16,sp
511 ;;
512 stf.spill [r16]=f6,32
513 stf.spill [r17]=f7,32
514 ;;
515 stf.spill [r16]=f8,32
516 stf.spill [r17]=f9,32
517 ;;
518 stf.spill [r16]=f10
519 stf.spill [r17]=f11
520 br.call.sptk.many rp=syscall_trace_enter // give parent a chance to catch syscall args
521 adds r16=PT(F6)+16,sp
522 adds r17=PT(F7)+16,sp
523 ;;
524 ldf.fill f6=[r16],32
525 ldf.fill f7=[r17],32
526 ;;
527 ldf.fill f8=[r16],32
528 ldf.fill f9=[r17],32
529 ;;
530 ldf.fill f10=[r16]
531 ldf.fill f11=[r17]
532 // the syscall number may have changed, so re-load it and re-calculate the
533 // syscall entry-point:
534 adds r15=PT(R15)+16,sp // r15 = &pt_regs.r15 (syscall #)
535 ;;
536 ld8 r15=[r15]
537 mov r3=NR_syscalls - 1
538 ;;
539 adds r15=-1024,r15
540 movl r16=sys_call_table
541 ;;
542 shladd r20=r15,3,r16 // r20 = sys_call_table + 8*(syscall-1024)
543 cmp.leu p6,p7=r15,r3
544 ;;
545(p6) ld8 r20=[r20] // load address of syscall entry point
546(p7) movl r20=sys_ni_syscall
547 ;;
548 mov b6=r20
549 br.call.sptk.many rp=b6 // do the syscall
550.strace_check_retval:
551 cmp.lt p6,p0=r8,r0 // syscall failed?
552 adds r2=PT(R8)+16,sp // r2 = &pt_regs.r8
553 adds r3=PT(R10)+16,sp // r3 = &pt_regs.r10
554 mov r10=0
555(p6) br.cond.sptk strace_error // syscall failed ->
556 ;; // avoid RAW on r10
557.strace_save_retval:
558.mem.offset 0,0; st8.spill [r2]=r8 // store return value in slot for r8
559.mem.offset 8,0; st8.spill [r3]=r10 // clear error indication in slot for r10
560 br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
561.ret3: br.cond.sptk .work_pending_syscall_end
562
563strace_error:
564 ld8 r3=[r2] // load pt_regs.r8
565 sub r9=0,r8 // negate return value to get errno value
566 ;;
567 cmp.ne p6,p0=r3,r0 // is pt_regs.r8!=0?
568 adds r3=16,r2 // r3=&pt_regs.r10
569 ;;
570(p6) mov r10=-1
571(p6) mov r8=r9
572 br.cond.sptk .strace_save_retval
573END(ia64_trace_syscall)
574
575 /*
576 * When traced and returning from sigreturn, we invoke syscall_trace but then
577 * go straight to ia64_leave_kernel rather than ia64_leave_syscall.
578 */
579GLOBAL_ENTRY(ia64_strace_leave_kernel)
580 PT_REGS_UNWIND_INFO(0)
581{ /*
582 * Some versions of gas generate bad unwind info if the first instruction of a
583 * procedure doesn't go into the first slot of a bundle. This is a workaround.
584 */
585 nop.m 0
586 nop.i 0
587 br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
588}
589.ret4: br.cond.sptk ia64_leave_kernel
590END(ia64_strace_leave_kernel)
591
592GLOBAL_ENTRY(ia64_ret_from_clone)
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 /*
601 * We need to call schedule_tail() to complete the scheduling process.
602 * Called by ia64_switch_to() after do_fork()->copy_thread(). r8 contains the
603 * address of the previously executing task.
604 */
605 br.call.sptk.many rp=ia64_invoke_schedule_tail
606}
607.ret8:
608 adds r2=TI_FLAGS+IA64_TASK_SIZE,r13
609 ;;
610 ld4 r2=[r2]
611 ;;
612 mov r8=0
613 and r2=_TIF_SYSCALL_TRACEAUDIT,r2
614 ;;
615 cmp.ne p6,p0=r2,r0
616(p6) br.cond.spnt .strace_check_retval
617 ;; // added stop bits to prevent r8 dependency
618END(ia64_ret_from_clone)
619 // fall through
620GLOBAL_ENTRY(ia64_ret_from_syscall)
621 PT_REGS_UNWIND_INFO(0)
622 cmp.ge p6,p7=r8,r0 // syscall executed successfully?
623 adds r2=PT(R8)+16,sp // r2 = &pt_regs.r8
624 mov r10=r0 // clear error indication in r10
625(p7) br.cond.spnt handle_syscall_error // handle potential syscall failure
626END(ia64_ret_from_syscall)
627 // fall through
628/*
629 * ia64_leave_syscall(): Same as ia64_leave_kernel, except that it doesn't
630 * need to switch to bank 0 and doesn't restore the scratch registers.
631 * To avoid leaking kernel bits, the scratch registers are set to
632 * the following known-to-be-safe values:
633 *
634 * r1: restored (global pointer)
635 * r2: cleared
636 * r3: 1 (when returning to user-level)
637 * r8-r11: restored (syscall return value(s))
638 * r12: restored (user-level stack pointer)
639 * r13: restored (user-level thread pointer)
640 * r14: cleared
641 * r15: restored (syscall #)
642 * r16-r17: cleared
643 * r18: user-level b6
644 * r19: cleared
645 * r20: user-level ar.fpsr
646 * r21: user-level b0
647 * r22: cleared
648 * r23: user-level ar.bspstore
649 * r24: user-level ar.rnat
650 * r25: user-level ar.unat
651 * r26: user-level ar.pfs
652 * r27: user-level ar.rsc
653 * r28: user-level ip
654 * r29: user-level psr
655 * r30: user-level cfm
656 * r31: user-level pr
657 * f6-f11: cleared
658 * pr: restored (user-level pr)
659 * b0: restored (user-level rp)
660 * b6: restored
661 * b7: cleared
662 * ar.unat: restored (user-level ar.unat)
663 * ar.pfs: restored (user-level ar.pfs)
664 * ar.rsc: restored (user-level ar.rsc)
665 * ar.rnat: restored (user-level ar.rnat)
666 * ar.bspstore: restored (user-level ar.bspstore)
667 * ar.fpsr: restored (user-level ar.fpsr)
668 * ar.ccv: cleared
669 * ar.csd: cleared
670 * ar.ssd: cleared
671 */
672ENTRY(ia64_leave_syscall)
673 PT_REGS_UNWIND_INFO(0)
674 /*
675 * work.need_resched etc. mustn't get changed by this CPU before it returns to
676 * user- or fsys-mode, hence we disable interrupts early on.
677 *
678 * p6 controls whether current_thread_info()->flags needs to be check for
679 * extra work. We always check for extra work when returning to user-level.
680 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
681 * is 0. After extra work processing has been completed, execution
682 * resumes at .work_processed_syscall with p6 set to 1 if the extra-work-check
683 * needs to be redone.
684 */
685#ifdef CONFIG_PREEMPT
686 rsm psr.i // disable interrupts
687 cmp.eq pLvSys,p0=r0,r0 // pLvSys=1: leave from syscall
688(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
689 ;;
690 .pred.rel.mutex pUStk,pKStk
691(pKStk) ld4 r21=[r20] // r21 <- preempt_count
692(pUStk) mov r21=0 // r21 <- 0
693 ;;
694 cmp.eq p6,p0=r21,r0 // p6 <- pUStk || (preempt_count == 0)
695#else /* !CONFIG_PREEMPT */
696(pUStk) rsm psr.i
697 cmp.eq pLvSys,p0=r0,r0 // pLvSys=1: leave from syscall
698(pUStk) cmp.eq.unc p6,p0=r0,r0 // p6 <- pUStk
699#endif
700.work_processed_syscall:
701 adds r2=PT(LOADRS)+16,r12
702 adds r3=PT(AR_BSPSTORE)+16,r12
703 adds r18=TI_FLAGS+IA64_TASK_SIZE,r13
704 ;;
705(p6) ld4 r31=[r18] // load current_thread_info()->flags
706 ld8 r19=[r2],PT(B6)-PT(LOADRS) // load ar.rsc value for "loadrs"
707 mov b7=r0 // clear b7
708 ;;
709 ld8 r23=[r3],PT(R11)-PT(AR_BSPSTORE) // load ar.bspstore (may be garbage)
710 ld8 r18=[r2],PT(R9)-PT(B6) // load b6
711(p6) and r15=TIF_WORK_MASK,r31 // any work other than TIF_SYSCALL_TRACE?
712 ;;
713 mov r16=ar.bsp // M2 get existing backing store pointer
714(p6) cmp4.ne.unc p6,p0=r15, r0 // any special work pending?
715(p6) br.cond.spnt .work_pending_syscall
716 ;;
717 // start restoring the state saved on the kernel stack (struct pt_regs):
718 ld8 r9=[r2],PT(CR_IPSR)-PT(R9)
719 ld8 r11=[r3],PT(CR_IIP)-PT(R11)
720 mov f6=f0 // clear f6
721 ;;
722 invala // M0|1 invalidate ALAT
723 rsm psr.i | psr.ic // M2 initiate turning off of interrupt and interruption collection
724 mov f9=f0 // clear f9
725
726 ld8 r29=[r2],16 // load cr.ipsr
727 ld8 r28=[r3],16 // load cr.iip
728 mov f8=f0 // clear f8
729 ;;
730 ld8 r30=[r2],16 // M0|1 load cr.ifs
731 mov.m ar.ssd=r0 // M2 clear ar.ssd
732 cmp.eq p9,p0=r0,r0 // set p9 to indicate that we should restore cr.ifs
733 ;;
734 ld8 r25=[r3],16 // M0|1 load ar.unat
735 mov.m ar.csd=r0 // M2 clear ar.csd
736 mov r22=r0 // clear r22
737 ;;
738 ld8 r26=[r2],PT(B0)-PT(AR_PFS) // M0|1 load ar.pfs
739(pKStk) mov r22=psr // M2 read PSR now that interrupts are disabled
740 mov f10=f0 // clear f10
741 ;;
742 ld8 r21=[r2],PT(AR_RNAT)-PT(B0) // load b0
743 ld8 r27=[r3],PT(PR)-PT(AR_RSC) // load ar.rsc
744 mov f11=f0 // clear f11
745 ;;
746 ld8 r24=[r2],PT(AR_FPSR)-PT(AR_RNAT) // load ar.rnat (may be garbage)
747 ld8 r31=[r3],PT(R1)-PT(PR) // load predicates
748(pUStk) add r14=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13
749 ;;
750 ld8 r20=[r2],PT(R12)-PT(AR_FPSR) // load ar.fpsr
751 ld8.fill r1=[r3],16 // load r1
752(pUStk) mov r17=1
753 ;;
754 srlz.d // M0 ensure interruption collection is off
755 ld8.fill r13=[r3],16
756 mov f7=f0 // clear f7
757 ;;
758 ld8.fill r12=[r2] // restore r12 (sp)
759 ld8.fill r15=[r3] // restore r15
760 addl r3=THIS_CPU(ia64_phys_stacked_size_p8),r0
761 ;;
762(pUStk) ld4 r3=[r3] // r3 = cpu_data->phys_stacked_size_p8
763(pUStk) st1 [r14]=r17
764 mov b6=r18 // I0 restore b6
765 ;;
766 mov r14=r0 // clear r14
767 shr.u r18=r19,16 // I0|1 get byte size of existing "dirty" partition
768(pKStk) br.cond.dpnt.many skip_rbs_switch
769
770 mov.m ar.ccv=r0 // clear ar.ccv
771(pNonSys) br.cond.dpnt.many dont_preserve_current_frame
772 br.cond.sptk.many rbs_switch
773END(ia64_leave_syscall)
774
775#ifdef CONFIG_IA32_SUPPORT
776GLOBAL_ENTRY(ia64_ret_from_ia32_execve)
777 PT_REGS_UNWIND_INFO(0)
778 adds r2=PT(R8)+16,sp // r2 = &pt_regs.r8
779 adds r3=PT(R10)+16,sp // r3 = &pt_regs.r10
780 ;;
781 .mem.offset 0,0
782 st8.spill [r2]=r8 // store return value in slot for r8 and set unat bit
783 .mem.offset 8,0
784 st8.spill [r3]=r0 // clear error indication in slot for r10 and set unat bit
785END(ia64_ret_from_ia32_execve_syscall)
786 // fall through
787#endif /* CONFIG_IA32_SUPPORT */
788GLOBAL_ENTRY(ia64_leave_kernel)
789 PT_REGS_UNWIND_INFO(0)
790 /*
791 * work.need_resched etc. mustn't get changed by this CPU before it returns to
792 * user- or fsys-mode, hence we disable interrupts early on.
793 *
794 * p6 controls whether current_thread_info()->flags needs to be check for
795 * extra work. We always check for extra work when returning to user-level.
796 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
797 * is 0. After extra work processing has been completed, execution
798 * resumes at .work_processed_syscall with p6 set to 1 if the extra-work-check
799 * needs to be redone.
800 */
801#ifdef CONFIG_PREEMPT
802 rsm psr.i // disable interrupts
803 cmp.eq p0,pLvSys=r0,r0 // pLvSys=0: leave from kernel
804(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
805 ;;
806 .pred.rel.mutex pUStk,pKStk
807(pKStk) ld4 r21=[r20] // r21 <- preempt_count
808(pUStk) mov r21=0 // r21 <- 0
809 ;;
810 cmp.eq p6,p0=r21,r0 // p6 <- pUStk || (preempt_count == 0)
811#else
812(pUStk) rsm psr.i
813 cmp.eq p0,pLvSys=r0,r0 // pLvSys=0: leave from kernel
814(pUStk) cmp.eq.unc p6,p0=r0,r0 // p6 <- pUStk
815#endif
816.work_processed_kernel:
817 adds r17=TI_FLAGS+IA64_TASK_SIZE,r13
818 ;;
819(p6) ld4 r31=[r17] // load current_thread_info()->flags
820 adds r21=PT(PR)+16,r12
821 ;;
822
823 lfetch [r21],PT(CR_IPSR)-PT(PR)
824 adds r2=PT(B6)+16,r12
825 adds r3=PT(R16)+16,r12
826 ;;
827 lfetch [r21]
828 ld8 r28=[r2],8 // load b6
829 adds r29=PT(R24)+16,r12
830
831 ld8.fill r16=[r3],PT(AR_CSD)-PT(R16)
832 adds r30=PT(AR_CCV)+16,r12
833(p6) and r19=TIF_WORK_MASK,r31 // any work other than TIF_SYSCALL_TRACE?
834 ;;
835 ld8.fill r24=[r29]
836 ld8 r15=[r30] // load ar.ccv
837(p6) cmp4.ne.unc p6,p0=r19, r0 // any special work pending?
838 ;;
839 ld8 r29=[r2],16 // load b7
840 ld8 r30=[r3],16 // load ar.csd
841(p6) br.cond.spnt .work_pending
842 ;;
843 ld8 r31=[r2],16 // load ar.ssd
844 ld8.fill r8=[r3],16
845 ;;
846 ld8.fill r9=[r2],16
847 ld8.fill r10=[r3],PT(R17)-PT(R10)
848 ;;
849 ld8.fill r11=[r2],PT(R18)-PT(R11)
850 ld8.fill r17=[r3],16
851 ;;
852 ld8.fill r18=[r2],16
853 ld8.fill r19=[r3],16
854 ;;
855 ld8.fill r20=[r2],16
856 ld8.fill r21=[r3],16
857 mov ar.csd=r30
858 mov ar.ssd=r31
859 ;;
860 rsm psr.i | psr.ic // initiate turning off of interrupt and interruption collection
861 invala // invalidate ALAT
862 ;;
863 ld8.fill r22=[r2],24
864 ld8.fill r23=[r3],24
865 mov b6=r28
866 ;;
867 ld8.fill r25=[r2],16
868 ld8.fill r26=[r3],16
869 mov b7=r29
870 ;;
871 ld8.fill r27=[r2],16
872 ld8.fill r28=[r3],16
873 ;;
874 ld8.fill r29=[r2],16
875 ld8.fill r30=[r3],24
876 ;;
877 ld8.fill r31=[r2],PT(F9)-PT(R31)
878 adds r3=PT(F10)-PT(F6),r3
879 ;;
880 ldf.fill f9=[r2],PT(F6)-PT(F9)
881 ldf.fill f10=[r3],PT(F8)-PT(F10)
882 ;;
883 ldf.fill f6=[r2],PT(F7)-PT(F6)
884 ;;
885 ldf.fill f7=[r2],PT(F11)-PT(F7)
886 ldf.fill f8=[r3],32
887 ;;
888 srlz.i // ensure interruption collection is off
889 mov ar.ccv=r15
890 ;;
891 ldf.fill f11=[r2]
892 bsw.0 // switch back to bank 0 (no stop bit required beforehand...)
893 ;;
894(pUStk) mov r18=IA64_KR(CURRENT)// M2 (12 cycle read latency)
895 adds r16=PT(CR_IPSR)+16,r12
896 adds r17=PT(CR_IIP)+16,r12
897
898(pKStk) mov r22=psr // M2 read PSR now that interrupts are disabled
899 nop.i 0
900 nop.i 0
901 ;;
902 ld8 r29=[r16],16 // load cr.ipsr
903 ld8 r28=[r17],16 // load cr.iip
904 ;;
905 ld8 r30=[r16],16 // load cr.ifs
906 ld8 r25=[r17],16 // load ar.unat
907 ;;
908 ld8 r26=[r16],16 // load ar.pfs
909 ld8 r27=[r17],16 // load ar.rsc
910 cmp.eq p9,p0=r0,r0 // set p9 to indicate that we should restore cr.ifs
911 ;;
912 ld8 r24=[r16],16 // load ar.rnat (may be garbage)
913 ld8 r23=[r17],16 // load ar.bspstore (may be garbage)
914 ;;
915 ld8 r31=[r16],16 // load predicates
916 ld8 r21=[r17],16 // load b0
917 ;;
918 ld8 r19=[r16],16 // load ar.rsc value for "loadrs"
919 ld8.fill r1=[r17],16 // load r1
920 ;;
921 ld8.fill r12=[r16],16
922 ld8.fill r13=[r17],16
923(pUStk) adds r18=IA64_TASK_THREAD_ON_USTACK_OFFSET,r18
924 ;;
925 ld8 r20=[r16],16 // ar.fpsr
926 ld8.fill r15=[r17],16
927 ;;
928 ld8.fill r14=[r16],16
929 ld8.fill r2=[r17]
930(pUStk) mov r17=1
931 ;;
932 ld8.fill r3=[r16]
933(pUStk) st1 [r18]=r17 // restore current->thread.on_ustack
934 shr.u r18=r19,16 // get byte size of existing "dirty" partition
935 ;;
936 mov r16=ar.bsp // get existing backing store pointer
937 addl r17=THIS_CPU(ia64_phys_stacked_size_p8),r0
938 ;;
939 ld4 r17=[r17] // r17 = cpu_data->phys_stacked_size_p8
940(pKStk) br.cond.dpnt skip_rbs_switch
941
942 /*
943 * Restore user backing store.
944 *
945 * NOTE: alloc, loadrs, and cover can't be predicated.
946 */
947(pNonSys) br.cond.dpnt dont_preserve_current_frame
948
949rbs_switch:
950 cover // add current frame into dirty partition and set cr.ifs
951 ;;
952 mov r19=ar.bsp // get new backing store pointer
953 sub r16=r16,r18 // krbs = old bsp - size of dirty partition
954 cmp.ne p9,p0=r0,r0 // clear p9 to skip restore of cr.ifs
955 ;;
956 sub r19=r19,r16 // calculate total byte size of dirty partition
957 add r18=64,r18 // don't force in0-in7 into memory...
958 ;;
959 shl r19=r19,16 // shift size of dirty partition into loadrs position
960 ;;
961dont_preserve_current_frame:
962 /*
963 * To prevent leaking bits between the kernel and user-space,
964 * we must clear the stacked registers in the "invalid" partition here.
965 * Not pretty, but at least it's fast (3.34 registers/cycle on Itanium,
966 * 5 registers/cycle on McKinley).
967 */
968# define pRecurse p6
969# define pReturn p7
970#ifdef CONFIG_ITANIUM
971# define Nregs 10
972#else
973# define Nregs 14
974#endif
975 alloc loc0=ar.pfs,2,Nregs-2,2,0
976 shr.u loc1=r18,9 // RNaTslots <= floor(dirtySize / (64*8))
977 sub r17=r17,r18 // r17 = (physStackedSize + 8) - dirtySize
978 ;;
979 mov ar.rsc=r19 // load ar.rsc to be used for "loadrs"
980 shladd in0=loc1,3,r17
981 mov in1=0
982 ;;
983 TEXT_ALIGN(32)
984rse_clear_invalid:
985#ifdef CONFIG_ITANIUM
986 // cycle 0
987 { .mii
988 alloc loc0=ar.pfs,2,Nregs-2,2,0
989 cmp.lt pRecurse,p0=Nregs*8,in0 // if more than Nregs regs left to clear, (re)curse
990 add out0=-Nregs*8,in0
991}{ .mfb
992 add out1=1,in1 // increment recursion count
993 nop.f 0
994 nop.b 0 // can't do br.call here because of alloc (WAW on CFM)
995 ;;
996}{ .mfi // cycle 1
997 mov loc1=0
998 nop.f 0
999 mov loc2=0
1000}{ .mib
1001 mov loc3=0
1002 mov loc4=0
1003(pRecurse) br.call.sptk.many b0=rse_clear_invalid
1004
1005}{ .mfi // cycle 2
1006 mov loc5=0
1007 nop.f 0
1008 cmp.ne pReturn,p0=r0,in1 // if recursion count != 0, we need to do a br.ret
1009}{ .mib
1010 mov loc6=0
1011 mov loc7=0
1012(pReturn) br.ret.sptk.many b0
1013}
1014#else /* !CONFIG_ITANIUM */
1015 alloc loc0=ar.pfs,2,Nregs-2,2,0
1016 cmp.lt pRecurse,p0=Nregs*8,in0 // if more than Nregs regs left to clear, (re)curse
1017 add out0=-Nregs*8,in0
1018 add out1=1,in1 // increment recursion count
1019 mov loc1=0
1020 mov loc2=0
1021 ;;
1022 mov loc3=0
1023 mov loc4=0
1024 mov loc5=0
1025 mov loc6=0
1026 mov loc7=0
1027(pRecurse) br.call.sptk.few b0=rse_clear_invalid
1028 ;;
1029 mov loc8=0
1030 mov loc9=0
1031 cmp.ne pReturn,p0=r0,in1 // if recursion count != 0, we need to do a br.ret
1032 mov loc10=0
1033 mov loc11=0
1034(pReturn) br.ret.sptk.many b0
1035#endif /* !CONFIG_ITANIUM */
1036# undef pRecurse
1037# undef pReturn
1038 ;;
1039 alloc r17=ar.pfs,0,0,0,0 // drop current register frame
1040 ;;
1041 loadrs
1042 ;;
1043skip_rbs_switch:
1044 mov ar.unat=r25 // M2
1045(pKStk) extr.u r22=r22,21,1 // I0 extract current value of psr.pp from r22
1046(pLvSys)mov r19=r0 // A clear r19 for leave_syscall, no-op otherwise
1047 ;;
1048(pUStk) mov ar.bspstore=r23 // M2
1049(pKStk) dep r29=r22,r29,21,1 // I0 update ipsr.pp with psr.pp
1050(pLvSys)mov r16=r0 // A clear r16 for leave_syscall, no-op otherwise
1051 ;;
1052 mov cr.ipsr=r29 // M2
1053 mov ar.pfs=r26 // I0
1054(pLvSys)mov r17=r0 // A clear r17 for leave_syscall, no-op otherwise
1055
1056(p9) mov cr.ifs=r30 // M2
1057 mov b0=r21 // I0
1058(pLvSys)mov r18=r0 // A clear r18 for leave_syscall, no-op otherwise
1059
1060 mov ar.fpsr=r20 // M2
1061 mov cr.iip=r28 // M2
1062 nop 0
1063 ;;
1064(pUStk) mov ar.rnat=r24 // M2 must happen with RSE in lazy mode
1065 nop 0
1066(pLvSys)mov r2=r0
1067
1068 mov ar.rsc=r27 // M2
1069 mov pr=r31,-1 // I0
1070 rfi // B
1071
1072 /*
1073 * On entry:
1074 * r20 = &current->thread_info->pre_count (if CONFIG_PREEMPT)
1075 * r31 = current->thread_info->flags
1076 * On exit:
1077 * p6 = TRUE if work-pending-check needs to be redone
1078 */
1079.work_pending_syscall:
1080 add r2=-8,r2
1081 add r3=-8,r3
1082 ;;
1083 st8 [r2]=r8
1084 st8 [r3]=r10
1085.work_pending:
1086 tbit.nz p6,p0=r31,TIF_SIGDELAYED // signal delayed from MCA/INIT/NMI/PMI context?
1087(p6) br.cond.sptk.few .sigdelayed
1088 ;;
1089 tbit.z p6,p0=r31,TIF_NEED_RESCHED // current_thread_info()->need_resched==0?
1090(p6) br.cond.sptk.few .notify
1091#ifdef CONFIG_PREEMPT
1092(pKStk) dep r21=-1,r0,PREEMPT_ACTIVE_BIT,1
1093 ;;
1094(pKStk) st4 [r20]=r21
1095 ssm psr.i // enable interrupts
1096#endif
1097 br.call.spnt.many rp=schedule
1098.ret9: cmp.eq p6,p0=r0,r0 // p6 <- 1
1099 rsm psr.i // disable interrupts
1100 ;;
1101#ifdef CONFIG_PREEMPT
1102(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
1103 ;;
1104(pKStk) st4 [r20]=r0 // preempt_count() <- 0
1105#endif
1106(pLvSys)br.cond.sptk.few .work_pending_syscall_end
1107 br.cond.sptk.many .work_processed_kernel // re-check
1108
1109.notify:
1110(pUStk) br.call.spnt.many rp=notify_resume_user
1111.ret10: cmp.ne p6,p0=r0,r0 // p6 <- 0
1112(pLvSys)br.cond.sptk.few .work_pending_syscall_end
1113 br.cond.sptk.many .work_processed_kernel // don't re-check
1114
1115// There is a delayed signal that was detected in MCA/INIT/NMI/PMI context where
1116// it could not be delivered. Deliver it now. The signal might be for us and
1117// may set TIF_SIGPENDING, so redrive ia64_leave_* after processing the delayed
1118// signal.
1119
1120.sigdelayed:
1121 br.call.sptk.many rp=do_sigdelayed
1122 cmp.eq p6,p0=r0,r0 // p6 <- 1, always re-check
1123(pLvSys)br.cond.sptk.few .work_pending_syscall_end
1124 br.cond.sptk.many .work_processed_kernel // re-check
1125
1126.work_pending_syscall_end:
1127 adds r2=PT(R8)+16,r12
1128 adds r3=PT(R10)+16,r12
1129 ;;
1130 ld8 r8=[r2]
1131 ld8 r10=[r3]
1132 br.cond.sptk.many .work_processed_syscall // re-check
1133
1134END(ia64_leave_kernel)
1135
1136ENTRY(handle_syscall_error)
1137 /*
1138 * Some system calls (e.g., ptrace, mmap) can return arbitrary values which could
1139 * lead us to mistake a negative return value as a failed syscall. Those syscall
1140 * must deposit a non-zero value in pt_regs.r8 to indicate an error. If
1141 * pt_regs.r8 is zero, we assume that the call completed successfully.
1142 */
1143 PT_REGS_UNWIND_INFO(0)
1144 ld8 r3=[r2] // load pt_regs.r8
1145 ;;
1146 cmp.eq p6,p7=r3,r0 // is pt_regs.r8==0?
1147 ;;
1148(p7) mov r10=-1
1149(p7) sub r8=0,r8 // negate return value to get errno
1150 br.cond.sptk ia64_leave_syscall
1151END(handle_syscall_error)
1152
1153 /*
1154 * Invoke schedule_tail(task) while preserving in0-in7, which may be needed
1155 * in case a system call gets restarted.
1156 */
1157GLOBAL_ENTRY(ia64_invoke_schedule_tail)
1158 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1159 alloc loc1=ar.pfs,8,2,1,0
1160 mov loc0=rp
1161 mov out0=r8 // Address of previous task
1162 ;;
1163 br.call.sptk.many rp=schedule_tail
1164.ret11: mov ar.pfs=loc1
1165 mov rp=loc0
1166 br.ret.sptk.many rp
1167END(ia64_invoke_schedule_tail)
1168
1169 /*
1170 * Setup stack and call do_notify_resume_user(). Note that pSys and pNonSys need to
1171 * be set up by the caller. We declare 8 input registers so the system call
1172 * args get preserved, in case we need to restart a system call.
1173 */
1174ENTRY(notify_resume_user)
1175 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1176 alloc loc1=ar.pfs,8,2,3,0 // preserve all eight input regs in case of syscall restart!
1177 mov r9=ar.unat
1178 mov loc0=rp // save return address
1179 mov out0=0 // there is no "oldset"
1180 adds out1=8,sp // out1=&sigscratch->ar_pfs
1181(pSys) mov out2=1 // out2==1 => we're in a syscall
1182 ;;
1183(pNonSys) mov out2=0 // out2==0 => not a syscall
1184 .fframe 16
1185 .spillpsp ar.unat, 16 // (note that offset is relative to psp+0x10!)
1186 st8 [sp]=r9,-16 // allocate space for ar.unat and save it
1187 st8 [out1]=loc1,-8 // save ar.pfs, out1=&sigscratch
1188 .body
1189 br.call.sptk.many rp=do_notify_resume_user
1190.ret15: .restore sp
1191 adds sp=16,sp // pop scratch stack space
1192 ;;
1193 ld8 r9=[sp] // load new unat from sigscratch->scratch_unat
1194 mov rp=loc0
1195 ;;
1196 mov ar.unat=r9
1197 mov ar.pfs=loc1
1198 br.ret.sptk.many rp
1199END(notify_resume_user)
1200
1201GLOBAL_ENTRY(sys_rt_sigsuspend)
1202 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1203 alloc loc1=ar.pfs,8,2,3,0 // preserve all eight input regs in case of syscall restart!
1204 mov r9=ar.unat
1205 mov loc0=rp // save return address
1206 mov out0=in0 // mask
1207 mov out1=in1 // sigsetsize
1208 adds out2=8,sp // out2=&sigscratch->ar_pfs
1209 ;;
1210 .fframe 16
1211 .spillpsp ar.unat, 16 // (note that offset is relative to psp+0x10!)
1212 st8 [sp]=r9,-16 // allocate space for ar.unat and save it
1213 st8 [out2]=loc1,-8 // save ar.pfs, out2=&sigscratch
1214 .body
1215 br.call.sptk.many rp=ia64_rt_sigsuspend
1216.ret17: .restore sp
1217 adds sp=16,sp // pop scratch stack space
1218 ;;
1219 ld8 r9=[sp] // load new unat from sw->caller_unat
1220 mov rp=loc0
1221 ;;
1222 mov ar.unat=r9
1223 mov ar.pfs=loc1
1224 br.ret.sptk.many rp
1225END(sys_rt_sigsuspend)
1226
1227ENTRY(sys_rt_sigreturn)
1228 PT_REGS_UNWIND_INFO(0)
1229 /*
1230 * Allocate 8 input registers since ptrace() may clobber them
1231 */
1232 alloc r2=ar.pfs,8,0,1,0
1233 .prologue
1234 PT_REGS_SAVES(16)
1235 adds sp=-16,sp
1236 .body
1237 cmp.eq pNonSys,pSys=r0,r0 // sigreturn isn't a normal syscall...
1238 ;;
1239 /*
1240 * leave_kernel() restores f6-f11 from pt_regs, but since the streamlined
1241 * syscall-entry path does not save them we save them here instead. Note: we
1242 * don't need to save any other registers that are not saved by the stream-lined
1243 * syscall path, because restore_sigcontext() restores them.
1244 */
1245 adds r16=PT(F6)+32,sp
1246 adds r17=PT(F7)+32,sp
1247 ;;
1248 stf.spill [r16]=f6,32
1249 stf.spill [r17]=f7,32
1250 ;;
1251 stf.spill [r16]=f8,32
1252 stf.spill [r17]=f9,32
1253 ;;
1254 stf.spill [r16]=f10
1255 stf.spill [r17]=f11
1256 adds out0=16,sp // out0 = &sigscratch
1257 br.call.sptk.many rp=ia64_rt_sigreturn
1258.ret19: .restore sp 0
1259 adds sp=16,sp
1260 ;;
1261 ld8 r9=[sp] // load new ar.unat
1262 mov.sptk b7=r8,ia64_leave_kernel
1263 ;;
1264 mov ar.unat=r9
1265 br.many b7
1266END(sys_rt_sigreturn)
1267
1268GLOBAL_ENTRY(ia64_prepare_handle_unaligned)
1269 .prologue
1270 /*
1271 * r16 = fake ar.pfs, we simply need to make sure privilege is still 0
1272 */
1273 mov r16=r0
1274 DO_SAVE_SWITCH_STACK
1275 br.call.sptk.many rp=ia64_handle_unaligned // stack frame setup in ivt
1276.ret21: .body
1277 DO_LOAD_SWITCH_STACK
1278 br.cond.sptk.many rp // goes to ia64_leave_kernel
1279END(ia64_prepare_handle_unaligned)
1280
1281 //
1282 // unw_init_running(void (*callback)(info, arg), void *arg)
1283 //
1284# define EXTRA_FRAME_SIZE ((UNW_FRAME_INFO_SIZE+15)&~15)
1285
1286GLOBAL_ENTRY(unw_init_running)
1287 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1288 alloc loc1=ar.pfs,2,3,3,0
1289 ;;
1290 ld8 loc2=[in0],8
1291 mov loc0=rp
1292 mov r16=loc1
1293 DO_SAVE_SWITCH_STACK
1294 .body
1295
1296 .prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1297 .fframe IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE
1298 SWITCH_STACK_SAVES(EXTRA_FRAME_SIZE)
1299 adds sp=-EXTRA_FRAME_SIZE,sp
1300 .body
1301 ;;
1302 adds out0=16,sp // &info
1303 mov out1=r13 // current
1304 adds out2=16+EXTRA_FRAME_SIZE,sp // &switch_stack
1305 br.call.sptk.many rp=unw_init_frame_info
13061: adds out0=16,sp // &info
1307 mov b6=loc2
1308 mov loc2=gp // save gp across indirect function call
1309 ;;
1310 ld8 gp=[in0]
1311 mov out1=in1 // arg
1312 br.call.sptk.many rp=b6 // invoke the callback function
13131: mov gp=loc2 // restore gp
1314
1315 // For now, we don't allow changing registers from within
1316 // unw_init_running; if we ever want to allow that, we'd
1317 // have to do a load_switch_stack here:
1318 .restore sp
1319 adds sp=IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE,sp
1320
1321 mov ar.pfs=loc1
1322 mov rp=loc0
1323 br.ret.sptk.many rp
1324END(unw_init_running)
1325
1326 .rodata
1327 .align 8
1328 .globl sys_call_table
1329sys_call_table:
1330 data8 sys_ni_syscall // This must be sys_ni_syscall! See ivt.S.
1331 data8 sys_exit // 1025
1332 data8 sys_read
1333 data8 sys_write
1334 data8 sys_open
1335 data8 sys_close
1336 data8 sys_creat // 1030
1337 data8 sys_link
1338 data8 sys_unlink
1339 data8 ia64_execve
1340 data8 sys_chdir
1341 data8 sys_fchdir // 1035
1342 data8 sys_utimes
1343 data8 sys_mknod
1344 data8 sys_chmod
1345 data8 sys_chown
1346 data8 sys_lseek // 1040
1347 data8 sys_getpid
1348 data8 sys_getppid
1349 data8 sys_mount
1350 data8 sys_umount
1351 data8 sys_setuid // 1045
1352 data8 sys_getuid
1353 data8 sys_geteuid
1354 data8 sys_ptrace
1355 data8 sys_access
1356 data8 sys_sync // 1050
1357 data8 sys_fsync
1358 data8 sys_fdatasync
1359 data8 sys_kill
1360 data8 sys_rename
1361 data8 sys_mkdir // 1055
1362 data8 sys_rmdir
1363 data8 sys_dup
1364 data8 sys_pipe
1365 data8 sys_times
1366 data8 ia64_brk // 1060
1367 data8 sys_setgid
1368 data8 sys_getgid
1369 data8 sys_getegid
1370 data8 sys_acct
1371 data8 sys_ioctl // 1065
1372 data8 sys_fcntl
1373 data8 sys_umask
1374 data8 sys_chroot
1375 data8 sys_ustat
1376 data8 sys_dup2 // 1070
1377 data8 sys_setreuid
1378 data8 sys_setregid
1379 data8 sys_getresuid
1380 data8 sys_setresuid
1381 data8 sys_getresgid // 1075
1382 data8 sys_setresgid
1383 data8 sys_getgroups
1384 data8 sys_setgroups
1385 data8 sys_getpgid
1386 data8 sys_setpgid // 1080
1387 data8 sys_setsid
1388 data8 sys_getsid
1389 data8 sys_sethostname
1390 data8 sys_setrlimit
1391 data8 sys_getrlimit // 1085
1392 data8 sys_getrusage
1393 data8 sys_gettimeofday
1394 data8 sys_settimeofday
1395 data8 sys_select
1396 data8 sys_poll // 1090
1397 data8 sys_symlink
1398 data8 sys_readlink
1399 data8 sys_uselib
1400 data8 sys_swapon
1401 data8 sys_swapoff // 1095
1402 data8 sys_reboot
1403 data8 sys_truncate
1404 data8 sys_ftruncate
1405 data8 sys_fchmod
1406 data8 sys_fchown // 1100
1407 data8 ia64_getpriority
1408 data8 sys_setpriority
1409 data8 sys_statfs
1410 data8 sys_fstatfs
1411 data8 sys_gettid // 1105
1412 data8 sys_semget
1413 data8 sys_semop
1414 data8 sys_semctl
1415 data8 sys_msgget
1416 data8 sys_msgsnd // 1110
1417 data8 sys_msgrcv
1418 data8 sys_msgctl
1419 data8 sys_shmget
1420 data8 ia64_shmat
1421 data8 sys_shmdt // 1115
1422 data8 sys_shmctl
1423 data8 sys_syslog
1424 data8 sys_setitimer
1425 data8 sys_getitimer
1426 data8 sys_ni_syscall // 1120 /* was: ia64_oldstat */
1427 data8 sys_ni_syscall /* was: ia64_oldlstat */
1428 data8 sys_ni_syscall /* was: ia64_oldfstat */
1429 data8 sys_vhangup
1430 data8 sys_lchown
1431 data8 sys_remap_file_pages // 1125
1432 data8 sys_wait4
1433 data8 sys_sysinfo
1434 data8 sys_clone
1435 data8 sys_setdomainname
1436 data8 sys_newuname // 1130
1437 data8 sys_adjtimex
1438 data8 sys_ni_syscall /* was: ia64_create_module */
1439 data8 sys_init_module
1440 data8 sys_delete_module
1441 data8 sys_ni_syscall // 1135 /* was: sys_get_kernel_syms */
1442 data8 sys_ni_syscall /* was: sys_query_module */
1443 data8 sys_quotactl
1444 data8 sys_bdflush
1445 data8 sys_sysfs
1446 data8 sys_personality // 1140
1447 data8 sys_ni_syscall // sys_afs_syscall
1448 data8 sys_setfsuid
1449 data8 sys_setfsgid
1450 data8 sys_getdents
1451 data8 sys_flock // 1145
1452 data8 sys_readv
1453 data8 sys_writev
1454 data8 sys_pread64
1455 data8 sys_pwrite64
1456 data8 sys_sysctl // 1150
1457 data8 sys_mmap
1458 data8 sys_munmap
1459 data8 sys_mlock
1460 data8 sys_mlockall
1461 data8 sys_mprotect // 1155
1462 data8 ia64_mremap
1463 data8 sys_msync
1464 data8 sys_munlock
1465 data8 sys_munlockall
1466 data8 sys_sched_getparam // 1160
1467 data8 sys_sched_setparam
1468 data8 sys_sched_getscheduler
1469 data8 sys_sched_setscheduler
1470 data8 sys_sched_yield
1471 data8 sys_sched_get_priority_max // 1165
1472 data8 sys_sched_get_priority_min
1473 data8 sys_sched_rr_get_interval
1474 data8 sys_nanosleep
1475 data8 sys_nfsservctl
1476 data8 sys_prctl // 1170
1477 data8 sys_getpagesize
1478 data8 sys_mmap2
1479 data8 sys_pciconfig_read
1480 data8 sys_pciconfig_write
1481 data8 sys_perfmonctl // 1175
1482 data8 sys_sigaltstack
1483 data8 sys_rt_sigaction
1484 data8 sys_rt_sigpending
1485 data8 sys_rt_sigprocmask
1486 data8 sys_rt_sigqueueinfo // 1180
1487 data8 sys_rt_sigreturn
1488 data8 sys_rt_sigsuspend
1489 data8 sys_rt_sigtimedwait
1490 data8 sys_getcwd
1491 data8 sys_capget // 1185
1492 data8 sys_capset
1493 data8 sys_sendfile64
1494 data8 sys_ni_syscall // sys_getpmsg (STREAMS)
1495 data8 sys_ni_syscall // sys_putpmsg (STREAMS)
1496 data8 sys_socket // 1190
1497 data8 sys_bind
1498 data8 sys_connect
1499 data8 sys_listen
1500 data8 sys_accept
1501 data8 sys_getsockname // 1195
1502 data8 sys_getpeername
1503 data8 sys_socketpair
1504 data8 sys_send
1505 data8 sys_sendto
1506 data8 sys_recv // 1200
1507 data8 sys_recvfrom
1508 data8 sys_shutdown
1509 data8 sys_setsockopt
1510 data8 sys_getsockopt
1511 data8 sys_sendmsg // 1205
1512 data8 sys_recvmsg
1513 data8 sys_pivot_root
1514 data8 sys_mincore
1515 data8 sys_madvise
1516 data8 sys_newstat // 1210
1517 data8 sys_newlstat
1518 data8 sys_newfstat
1519 data8 sys_clone2
1520 data8 sys_getdents64
1521 data8 sys_getunwind // 1215
1522 data8 sys_readahead
1523 data8 sys_setxattr
1524 data8 sys_lsetxattr
1525 data8 sys_fsetxattr
1526 data8 sys_getxattr // 1220
1527 data8 sys_lgetxattr
1528 data8 sys_fgetxattr
1529 data8 sys_listxattr
1530 data8 sys_llistxattr
1531 data8 sys_flistxattr // 1225
1532 data8 sys_removexattr
1533 data8 sys_lremovexattr
1534 data8 sys_fremovexattr
1535 data8 sys_tkill
1536 data8 sys_futex // 1230
1537 data8 sys_sched_setaffinity
1538 data8 sys_sched_getaffinity
1539 data8 sys_set_tid_address
1540 data8 sys_fadvise64_64
1541 data8 sys_tgkill // 1235
1542 data8 sys_exit_group
1543 data8 sys_lookup_dcookie
1544 data8 sys_io_setup
1545 data8 sys_io_destroy
1546 data8 sys_io_getevents // 1240
1547 data8 sys_io_submit
1548 data8 sys_io_cancel
1549 data8 sys_epoll_create
1550 data8 sys_epoll_ctl
1551 data8 sys_epoll_wait // 1245
1552 data8 sys_restart_syscall
1553 data8 sys_semtimedop
1554 data8 sys_timer_create
1555 data8 sys_timer_settime
1556 data8 sys_timer_gettime // 1250
1557 data8 sys_timer_getoverrun
1558 data8 sys_timer_delete
1559 data8 sys_clock_settime
1560 data8 sys_clock_gettime
1561 data8 sys_clock_getres // 1255
1562 data8 sys_clock_nanosleep
1563 data8 sys_fstatfs64
1564 data8 sys_statfs64
1565 data8 sys_mbind
1566 data8 sys_get_mempolicy // 1260
1567 data8 sys_set_mempolicy
1568 data8 sys_mq_open
1569 data8 sys_mq_unlink
1570 data8 sys_mq_timedsend
1571 data8 sys_mq_timedreceive // 1265
1572 data8 sys_mq_notify
1573 data8 sys_mq_getsetattr
1574 data8 sys_ni_syscall // reserved for kexec_load
1575 data8 sys_ni_syscall // reserved for vserver
1576 data8 sys_waitid // 1270
1577 data8 sys_add_key
1578 data8 sys_request_key
1579 data8 sys_keyctl
1580 data8 sys_ni_syscall
1581 data8 sys_ni_syscall // 1275
1582 data8 sys_ni_syscall
1583 data8 sys_ni_syscall
1584 data8 sys_ni_syscall
1585 data8 sys_ni_syscall
1586
1587 .org sys_call_table + 8*NR_syscalls // guard against failures to increase NR_syscalls