xref: /openbmc/linux/arch/ia64/kernel/entry.S (revision 64c70b1c)
1/*
2 * arch/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
35#include <asm/asmmacro.h>
36#include <asm/cache.h>
37#include <asm/errno.h>
38#include <asm/kregs.h>
39#include <asm/asm-offsets.h>
40#include <asm/pgtable.h>
41#include <asm/percpu.h>
42#include <asm/processor.h>
43#include <asm/thread_info.h>
44#include <asm/unistd.h>
45
46#include "minstate.h"
47
48	/*
49	 * execve() is special because in case of success, we need to
50	 * setup a null register window frame.
51	 */
52ENTRY(ia64_execve)
53	/*
54	 * Allocate 8 input registers since ptrace() may clobber them
55	 */
56	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
57	alloc loc1=ar.pfs,8,2,4,0
58	mov loc0=rp
59	.body
60	mov out0=in0			// filename
61	;;				// stop bit between alloc and call
62	mov out1=in1			// argv
63	mov out2=in2			// envp
64	add out3=16,sp			// regs
65	br.call.sptk.many rp=sys_execve
66.ret0:
67#ifdef CONFIG_IA32_SUPPORT
68	/*
69	 * Check if we're returning to ia32 mode. If so, we need to restore ia32 registers
70	 * from pt_regs.
71	 */
72	adds r16=PT(CR_IPSR)+16,sp
73	;;
74	ld8 r16=[r16]
75#endif
76	cmp4.ge p6,p7=r8,r0
77	mov ar.pfs=loc1			// restore ar.pfs
78	sxt4 r8=r8			// return 64-bit result
79	;;
80	stf.spill [sp]=f0
81(p6)	cmp.ne pKStk,pUStk=r0,r0	// a successful execve() lands us in user-mode...
82	mov rp=loc0
83(p6)	mov ar.pfs=r0			// clear ar.pfs on success
84(p7)	br.ret.sptk.many rp
85
86	/*
87	 * In theory, we'd have to zap this state only to prevent leaking of
88	 * security sensitive state (e.g., if current->mm->dumpable is zero).  However,
89	 * this executes in less than 20 cycles even on Itanium, so it's not worth
90	 * optimizing for...).
91	 */
92	mov ar.unat=0; 		mov ar.lc=0
93	mov r4=0;		mov f2=f0;		mov b1=r0
94	mov r5=0;		mov f3=f0;		mov b2=r0
95	mov r6=0;		mov f4=f0;		mov b3=r0
96	mov r7=0;		mov f5=f0;		mov b4=r0
97	ldf.fill f12=[sp];	mov f13=f0;		mov b5=r0
98	ldf.fill f14=[sp];	ldf.fill f15=[sp];	mov f16=f0
99	ldf.fill f17=[sp];	ldf.fill f18=[sp];	mov f19=f0
100	ldf.fill f20=[sp];	ldf.fill f21=[sp];	mov f22=f0
101	ldf.fill f23=[sp];	ldf.fill f24=[sp];	mov f25=f0
102	ldf.fill f26=[sp];	ldf.fill f27=[sp];	mov f28=f0
103	ldf.fill f29=[sp];	ldf.fill f30=[sp];	mov f31=f0
104#ifdef CONFIG_IA32_SUPPORT
105	tbit.nz p6,p0=r16, IA64_PSR_IS_BIT
106	movl loc0=ia64_ret_from_ia32_execve
107	;;
108(p6)	mov rp=loc0
109#endif
110	br.ret.sptk.many rp
111END(ia64_execve)
112
113/*
114 * sys_clone2(u64 flags, u64 ustack_base, u64 ustack_size, u64 parent_tidptr, u64 child_tidptr,
115 *	      u64 tls)
116 */
117GLOBAL_ENTRY(sys_clone2)
118	/*
119	 * Allocate 8 input registers since ptrace() may clobber them
120	 */
121	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
122	alloc r16=ar.pfs,8,2,6,0
123	DO_SAVE_SWITCH_STACK
124	adds r2=PT(R16)+IA64_SWITCH_STACK_SIZE+16,sp
125	mov loc0=rp
126	mov loc1=r16				// save ar.pfs across do_fork
127	.body
128	mov out1=in1
129	mov out3=in2
130	tbit.nz p6,p0=in0,CLONE_SETTLS_BIT
131	mov out4=in3	// parent_tidptr: valid only w/CLONE_PARENT_SETTID
132	;;
133(p6)	st8 [r2]=in5				// store TLS in r16 for copy_thread()
134	mov out5=in4	// child_tidptr:  valid only w/CLONE_CHILD_SETTID or CLONE_CHILD_CLEARTID
135	adds out2=IA64_SWITCH_STACK_SIZE+16,sp	// out2 = &regs
136	mov out0=in0				// out0 = clone_flags
137	br.call.sptk.many rp=do_fork
138.ret1:	.restore sp
139	adds sp=IA64_SWITCH_STACK_SIZE,sp	// pop the switch stack
140	mov ar.pfs=loc1
141	mov rp=loc0
142	br.ret.sptk.many rp
143END(sys_clone2)
144
145/*
146 * sys_clone(u64 flags, u64 ustack_base, u64 parent_tidptr, u64 child_tidptr, u64 tls)
147 *	Deprecated.  Use sys_clone2() instead.
148 */
149GLOBAL_ENTRY(sys_clone)
150	/*
151	 * Allocate 8 input registers since ptrace() may clobber them
152	 */
153	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
154	alloc r16=ar.pfs,8,2,6,0
155	DO_SAVE_SWITCH_STACK
156	adds r2=PT(R16)+IA64_SWITCH_STACK_SIZE+16,sp
157	mov loc0=rp
158	mov loc1=r16				// save ar.pfs across do_fork
159	.body
160	mov out1=in1
161	mov out3=16				// stacksize (compensates for 16-byte scratch area)
162	tbit.nz p6,p0=in0,CLONE_SETTLS_BIT
163	mov out4=in2	// parent_tidptr: valid only w/CLONE_PARENT_SETTID
164	;;
165(p6)	st8 [r2]=in4				// store TLS in r13 (tp)
166	mov out5=in3	// child_tidptr:  valid only w/CLONE_CHILD_SETTID or CLONE_CHILD_CLEARTID
167	adds out2=IA64_SWITCH_STACK_SIZE+16,sp	// out2 = &regs
168	mov out0=in0				// out0 = clone_flags
169	br.call.sptk.many rp=do_fork
170.ret2:	.restore sp
171	adds sp=IA64_SWITCH_STACK_SIZE,sp	// pop the switch stack
172	mov ar.pfs=loc1
173	mov rp=loc0
174	br.ret.sptk.many rp
175END(sys_clone)
176
177/*
178 * prev_task <- ia64_switch_to(struct task_struct *next)
179 *	With Ingo's new scheduler, interrupts are disabled when this routine gets
180 *	called.  The code starting at .map relies on this.  The rest of the code
181 *	doesn't care about the interrupt masking status.
182 */
183GLOBAL_ENTRY(ia64_switch_to)
184	.prologue
185	alloc r16=ar.pfs,1,0,0,0
186	DO_SAVE_SWITCH_STACK
187	.body
188
189	adds r22=IA64_TASK_THREAD_KSP_OFFSET,r13
190	movl r25=init_task
191	mov r27=IA64_KR(CURRENT_STACK)
192	adds r21=IA64_TASK_THREAD_KSP_OFFSET,in0
193	dep r20=0,in0,61,3		// physical address of "next"
194	;;
195	st8 [r22]=sp			// save kernel stack pointer of old task
196	shr.u r26=r20,IA64_GRANULE_SHIFT
197	cmp.eq p7,p6=r25,in0
198	;;
199	/*
200	 * If we've already mapped this task's page, we can skip doing it again.
201	 */
202(p6)	cmp.eq p7,p6=r26,r27
203(p6)	br.cond.dpnt .map
204	;;
205.done:
206	ld8 sp=[r21]			// load kernel stack pointer of new task
207	mov IA64_KR(CURRENT)=in0	// update "current" application register
208	mov r8=r13			// return pointer to previously running task
209	mov r13=in0			// set "current" pointer
210	;;
211	DO_LOAD_SWITCH_STACK
212
213#ifdef CONFIG_SMP
214	sync.i				// ensure "fc"s done by this CPU are visible on other CPUs
215#endif
216	br.ret.sptk.many rp		// boogie on out in new context
217
218.map:
219	rsm psr.ic			// interrupts (psr.i) are already disabled here
220	movl r25=PAGE_KERNEL
221	;;
222	srlz.d
223	or r23=r25,r20			// construct PA | page properties
224	mov r25=IA64_GRANULE_SHIFT<<2
225	;;
226	mov cr.itir=r25
227	mov cr.ifa=in0			// VA of next task...
228	;;
229	mov r25=IA64_TR_CURRENT_STACK
230	mov IA64_KR(CURRENT_STACK)=r26	// remember last page we mapped...
231	;;
232	itr.d dtr[r25]=r23		// wire in new mapping...
233	ssm psr.ic			// reenable the psr.ic bit
234	;;
235	srlz.d
236	br.cond.sptk .done
237END(ia64_switch_to)
238
239/*
240 * Note that interrupts are enabled during save_switch_stack and load_switch_stack.  This
241 * means that we may get an interrupt with "sp" pointing to the new kernel stack while
242 * ar.bspstore is still pointing to the old kernel backing store area.  Since ar.rsc,
243 * ar.rnat, ar.bsp, and ar.bspstore are all preserved by interrupts, this is not a
244 * problem.  Also, we don't need to specify unwind information for preserved registers
245 * that are not modified in save_switch_stack as the right unwind information is already
246 * specified at the call-site of save_switch_stack.
247 */
248
249/*
250 * save_switch_stack:
251 *	- r16 holds ar.pfs
252 *	- b7 holds address to return to
253 *	- rp (b0) holds return address to save
254 */
255GLOBAL_ENTRY(save_switch_stack)
256	.prologue
257	.altrp b7
258	flushrs			// flush dirty regs to backing store (must be first in insn group)
259	.save @priunat,r17
260	mov r17=ar.unat		// preserve caller's
261	.body
262#ifdef CONFIG_ITANIUM
263	adds r2=16+128,sp
264	adds r3=16+64,sp
265	adds r14=SW(R4)+16,sp
266	;;
267	st8.spill [r14]=r4,16		// spill r4
268	lfetch.fault.excl.nt1 [r3],128
269	;;
270	lfetch.fault.excl.nt1 [r2],128
271	lfetch.fault.excl.nt1 [r3],128
272	;;
273	lfetch.fault.excl [r2]
274	lfetch.fault.excl [r3]
275	adds r15=SW(R5)+16,sp
276#else
277	add r2=16+3*128,sp
278	add r3=16,sp
279	add r14=SW(R4)+16,sp
280	;;
281	st8.spill [r14]=r4,SW(R6)-SW(R4)	// spill r4 and prefetch offset 0x1c0
282	lfetch.fault.excl.nt1 [r3],128	//		prefetch offset 0x010
283	;;
284	lfetch.fault.excl.nt1 [r3],128	//		prefetch offset 0x090
285	lfetch.fault.excl.nt1 [r2],128	//		prefetch offset 0x190
286	;;
287	lfetch.fault.excl.nt1 [r3]	//		prefetch offset 0x110
288	lfetch.fault.excl.nt1 [r2]	//		prefetch offset 0x210
289	adds r15=SW(R5)+16,sp
290#endif
291	;;
292	st8.spill [r15]=r5,SW(R7)-SW(R5)	// spill r5
293	mov.m ar.rsc=0			// put RSE in mode: enforced lazy, little endian, pl 0
294	add r2=SW(F2)+16,sp		// r2 = &sw->f2
295	;;
296	st8.spill [r14]=r6,SW(B0)-SW(R6)	// spill r6
297	mov.m r18=ar.fpsr		// preserve fpsr
298	add r3=SW(F3)+16,sp		// r3 = &sw->f3
299	;;
300	stf.spill [r2]=f2,32
301	mov.m r19=ar.rnat
302	mov r21=b0
303
304	stf.spill [r3]=f3,32
305	st8.spill [r15]=r7,SW(B2)-SW(R7)	// spill r7
306	mov r22=b1
307	;;
308	// since we're done with the spills, read and save ar.unat:
309	mov.m r29=ar.unat
310	mov.m r20=ar.bspstore
311	mov r23=b2
312	stf.spill [r2]=f4,32
313	stf.spill [r3]=f5,32
314	mov r24=b3
315	;;
316	st8 [r14]=r21,SW(B1)-SW(B0)		// save b0
317	st8 [r15]=r23,SW(B3)-SW(B2)		// save b2
318	mov r25=b4
319	mov r26=b5
320	;;
321	st8 [r14]=r22,SW(B4)-SW(B1)		// save b1
322	st8 [r15]=r24,SW(AR_PFS)-SW(B3)		// save b3
323	mov r21=ar.lc		// I-unit
324	stf.spill [r2]=f12,32
325	stf.spill [r3]=f13,32
326	;;
327	st8 [r14]=r25,SW(B5)-SW(B4)		// save b4
328	st8 [r15]=r16,SW(AR_LC)-SW(AR_PFS)	// save ar.pfs
329	stf.spill [r2]=f14,32
330	stf.spill [r3]=f15,32
331	;;
332	st8 [r14]=r26				// save b5
333	st8 [r15]=r21				// save ar.lc
334	stf.spill [r2]=f16,32
335	stf.spill [r3]=f17,32
336	;;
337	stf.spill [r2]=f18,32
338	stf.spill [r3]=f19,32
339	;;
340	stf.spill [r2]=f20,32
341	stf.spill [r3]=f21,32
342	;;
343	stf.spill [r2]=f22,32
344	stf.spill [r3]=f23,32
345	;;
346	stf.spill [r2]=f24,32
347	stf.spill [r3]=f25,32
348	;;
349	stf.spill [r2]=f26,32
350	stf.spill [r3]=f27,32
351	;;
352	stf.spill [r2]=f28,32
353	stf.spill [r3]=f29,32
354	;;
355	stf.spill [r2]=f30,SW(AR_UNAT)-SW(F30)
356	stf.spill [r3]=f31,SW(PR)-SW(F31)
357	add r14=SW(CALLER_UNAT)+16,sp
358	;;
359	st8 [r2]=r29,SW(AR_RNAT)-SW(AR_UNAT)	// save ar.unat
360	st8 [r14]=r17,SW(AR_FPSR)-SW(CALLER_UNAT) // save caller_unat
361	mov r21=pr
362	;;
363	st8 [r2]=r19,SW(AR_BSPSTORE)-SW(AR_RNAT) // save ar.rnat
364	st8 [r3]=r21				// save predicate registers
365	;;
366	st8 [r2]=r20				// save ar.bspstore
367	st8 [r14]=r18				// save fpsr
368	mov ar.rsc=3		// put RSE back into eager mode, pl 0
369	br.cond.sptk.many b7
370END(save_switch_stack)
371
372/*
373 * load_switch_stack:
374 *	- "invala" MUST be done at call site (normally in DO_LOAD_SWITCH_STACK)
375 *	- b7 holds address to return to
376 *	- must not touch r8-r11
377 */
378ENTRY(load_switch_stack)
379	.prologue
380	.altrp b7
381
382	.body
383	lfetch.fault.nt1 [sp]
384	adds r2=SW(AR_BSPSTORE)+16,sp
385	adds r3=SW(AR_UNAT)+16,sp
386	mov ar.rsc=0						// put RSE into enforced lazy mode
387	adds r14=SW(CALLER_UNAT)+16,sp
388	adds r15=SW(AR_FPSR)+16,sp
389	;;
390	ld8 r27=[r2],(SW(B0)-SW(AR_BSPSTORE))	// bspstore
391	ld8 r29=[r3],(SW(B1)-SW(AR_UNAT))	// unat
392	;;
393	ld8 r21=[r2],16		// restore b0
394	ld8 r22=[r3],16		// restore b1
395	;;
396	ld8 r23=[r2],16		// restore b2
397	ld8 r24=[r3],16		// restore b3
398	;;
399	ld8 r25=[r2],16		// restore b4
400	ld8 r26=[r3],16		// restore b5
401	;;
402	ld8 r16=[r2],(SW(PR)-SW(AR_PFS))	// ar.pfs
403	ld8 r17=[r3],(SW(AR_RNAT)-SW(AR_LC))	// ar.lc
404	;;
405	ld8 r28=[r2]		// restore pr
406	ld8 r30=[r3]		// restore rnat
407	;;
408	ld8 r18=[r14],16	// restore caller's unat
409	ld8 r19=[r15],24	// restore fpsr
410	;;
411	ldf.fill f2=[r14],32
412	ldf.fill f3=[r15],32
413	;;
414	ldf.fill f4=[r14],32
415	ldf.fill f5=[r15],32
416	;;
417	ldf.fill f12=[r14],32
418	ldf.fill f13=[r15],32
419	;;
420	ldf.fill f14=[r14],32
421	ldf.fill f15=[r15],32
422	;;
423	ldf.fill f16=[r14],32
424	ldf.fill f17=[r15],32
425	;;
426	ldf.fill f18=[r14],32
427	ldf.fill f19=[r15],32
428	mov b0=r21
429	;;
430	ldf.fill f20=[r14],32
431	ldf.fill f21=[r15],32
432	mov b1=r22
433	;;
434	ldf.fill f22=[r14],32
435	ldf.fill f23=[r15],32
436	mov b2=r23
437	;;
438	mov ar.bspstore=r27
439	mov ar.unat=r29		// establish unat holding the NaT bits for r4-r7
440	mov b3=r24
441	;;
442	ldf.fill f24=[r14],32
443	ldf.fill f25=[r15],32
444	mov b4=r25
445	;;
446	ldf.fill f26=[r14],32
447	ldf.fill f27=[r15],32
448	mov b5=r26
449	;;
450	ldf.fill f28=[r14],32
451	ldf.fill f29=[r15],32
452	mov ar.pfs=r16
453	;;
454	ldf.fill f30=[r14],32
455	ldf.fill f31=[r15],24
456	mov ar.lc=r17
457	;;
458	ld8.fill r4=[r14],16
459	ld8.fill r5=[r15],16
460	mov pr=r28,-1
461	;;
462	ld8.fill r6=[r14],16
463	ld8.fill r7=[r15],16
464
465	mov ar.unat=r18				// restore caller's unat
466	mov ar.rnat=r30				// must restore after bspstore but before rsc!
467	mov ar.fpsr=r19				// restore fpsr
468	mov ar.rsc=3				// put RSE back into eager mode, pl 0
469	br.cond.sptk.many b7
470END(load_switch_stack)
471
472GLOBAL_ENTRY(prefetch_stack)
473	add r14 = -IA64_SWITCH_STACK_SIZE, sp
474	add r15 = IA64_TASK_THREAD_KSP_OFFSET, in0
475	;;
476	ld8 r16 = [r15]				// load next's stack pointer
477	lfetch.fault.excl [r14], 128
478	;;
479	lfetch.fault.excl [r14], 128
480	lfetch.fault [r16], 128
481	;;
482	lfetch.fault.excl [r14], 128
483	lfetch.fault [r16], 128
484	;;
485	lfetch.fault.excl [r14], 128
486	lfetch.fault [r16], 128
487	;;
488	lfetch.fault.excl [r14], 128
489	lfetch.fault [r16], 128
490	;;
491	lfetch.fault [r16], 128
492	br.ret.sptk.many rp
493END(prefetch_stack)
494
495GLOBAL_ENTRY(kernel_execve)
496	mov r15=__NR_execve			// put syscall number in place
497	break __BREAK_SYSCALL
498	br.ret.sptk.many rp
499END(kernel_execve)
500
501GLOBAL_ENTRY(clone)
502	mov r15=__NR_clone			// put syscall number in place
503	break __BREAK_SYSCALL
504	br.ret.sptk.many rp
505END(clone)
506
507	/*
508	 * Invoke a system call, but do some tracing before and after the call.
509	 * We MUST preserve the current register frame throughout this routine
510	 * because some system calls (such as ia64_execve) directly
511	 * manipulate ar.pfs.
512	 */
513GLOBAL_ENTRY(ia64_trace_syscall)
514	PT_REGS_UNWIND_INFO(0)
515	/*
516	 * We need to preserve the scratch registers f6-f11 in case the system
517	 * call is sigreturn.
518	 */
519	adds r16=PT(F6)+16,sp
520	adds r17=PT(F7)+16,sp
521	;;
522 	stf.spill [r16]=f6,32
523 	stf.spill [r17]=f7,32
524	;;
525 	stf.spill [r16]=f8,32
526 	stf.spill [r17]=f9,32
527	;;
528 	stf.spill [r16]=f10
529 	stf.spill [r17]=f11
530	br.call.sptk.many rp=syscall_trace_enter // give parent a chance to catch syscall args
531	adds r16=PT(F6)+16,sp
532	adds r17=PT(F7)+16,sp
533	;;
534	ldf.fill f6=[r16],32
535	ldf.fill f7=[r17],32
536	;;
537	ldf.fill f8=[r16],32
538	ldf.fill f9=[r17],32
539	;;
540	ldf.fill f10=[r16]
541	ldf.fill f11=[r17]
542	// the syscall number may have changed, so re-load it and re-calculate the
543	// syscall entry-point:
544	adds r15=PT(R15)+16,sp			// r15 = &pt_regs.r15 (syscall #)
545	;;
546	ld8 r15=[r15]
547	mov r3=NR_syscalls - 1
548	;;
549	adds r15=-1024,r15
550	movl r16=sys_call_table
551	;;
552	shladd r20=r15,3,r16			// r20 = sys_call_table + 8*(syscall-1024)
553	cmp.leu p6,p7=r15,r3
554	;;
555(p6)	ld8 r20=[r20]				// load address of syscall entry point
556(p7)	movl r20=sys_ni_syscall
557	;;
558	mov b6=r20
559	br.call.sptk.many rp=b6			// do the syscall
560.strace_check_retval:
561	cmp.lt p6,p0=r8,r0			// syscall failed?
562	adds r2=PT(R8)+16,sp			// r2 = &pt_regs.r8
563	adds r3=PT(R10)+16,sp			// r3 = &pt_regs.r10
564	mov r10=0
565(p6)	br.cond.sptk strace_error		// syscall failed ->
566	;;					// avoid RAW on r10
567.strace_save_retval:
568.mem.offset 0,0; st8.spill [r2]=r8		// store return value in slot for r8
569.mem.offset 8,0; st8.spill [r3]=r10		// clear error indication in slot for r10
570	br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
571.ret3:
572(pUStk)	cmp.eq.unc p6,p0=r0,r0			// p6 <- pUStk
573	br.cond.sptk .work_pending_syscall_end
574
575strace_error:
576	ld8 r3=[r2]				// load pt_regs.r8
577	sub r9=0,r8				// negate return value to get errno value
578	;;
579	cmp.ne p6,p0=r3,r0			// is pt_regs.r8!=0?
580	adds r3=16,r2				// r3=&pt_regs.r10
581	;;
582(p6)	mov r10=-1
583(p6)	mov r8=r9
584	br.cond.sptk .strace_save_retval
585END(ia64_trace_syscall)
586
587	/*
588	 * When traced and returning from sigreturn, we invoke syscall_trace but then
589	 * go straight to ia64_leave_kernel rather than ia64_leave_syscall.
590	 */
591GLOBAL_ENTRY(ia64_strace_leave_kernel)
592	PT_REGS_UNWIND_INFO(0)
593{	/*
594	 * Some versions of gas generate bad unwind info if the first instruction of a
595	 * procedure doesn't go into the first slot of a bundle.  This is a workaround.
596	 */
597	nop.m 0
598	nop.i 0
599	br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
600}
601.ret4:	br.cond.sptk ia64_leave_kernel
602END(ia64_strace_leave_kernel)
603
604GLOBAL_ENTRY(ia64_ret_from_clone)
605	PT_REGS_UNWIND_INFO(0)
606{	/*
607	 * Some versions of gas generate bad unwind info if the first instruction of a
608	 * procedure doesn't go into the first slot of a bundle.  This is a workaround.
609	 */
610	nop.m 0
611	nop.i 0
612	/*
613	 * We need to call schedule_tail() to complete the scheduling process.
614	 * Called by ia64_switch_to() after do_fork()->copy_thread().  r8 contains the
615	 * address of the previously executing task.
616	 */
617	br.call.sptk.many rp=ia64_invoke_schedule_tail
618}
619.ret8:
620	adds r2=TI_FLAGS+IA64_TASK_SIZE,r13
621	;;
622	ld4 r2=[r2]
623	;;
624	mov r8=0
625	and r2=_TIF_SYSCALL_TRACEAUDIT,r2
626	;;
627	cmp.ne p6,p0=r2,r0
628(p6)	br.cond.spnt .strace_check_retval
629	;;					// added stop bits to prevent r8 dependency
630END(ia64_ret_from_clone)
631	// fall through
632GLOBAL_ENTRY(ia64_ret_from_syscall)
633	PT_REGS_UNWIND_INFO(0)
634	cmp.ge p6,p7=r8,r0			// syscall executed successfully?
635	adds r2=PT(R8)+16,sp			// r2 = &pt_regs.r8
636	mov r10=r0				// clear error indication in r10
637(p7)	br.cond.spnt handle_syscall_error	// handle potential syscall failure
638END(ia64_ret_from_syscall)
639	// fall through
640/*
641 * ia64_leave_syscall(): Same as ia64_leave_kernel, except that it doesn't
642 *	need to switch to bank 0 and doesn't restore the scratch registers.
643 *	To avoid leaking kernel bits, the scratch registers are set to
644 *	the following known-to-be-safe values:
645 *
646 *		  r1: restored (global pointer)
647 *		  r2: cleared
648 *		  r3: 1 (when returning to user-level)
649 *	      r8-r11: restored (syscall return value(s))
650 *		 r12: restored (user-level stack pointer)
651 *		 r13: restored (user-level thread pointer)
652 *		 r14: set to __kernel_syscall_via_epc
653 *		 r15: restored (syscall #)
654 *	     r16-r17: cleared
655 *		 r18: user-level b6
656 *		 r19: cleared
657 *		 r20: user-level ar.fpsr
658 *		 r21: user-level b0
659 *		 r22: cleared
660 *		 r23: user-level ar.bspstore
661 *		 r24: user-level ar.rnat
662 *		 r25: user-level ar.unat
663 *		 r26: user-level ar.pfs
664 *		 r27: user-level ar.rsc
665 *		 r28: user-level ip
666 *		 r29: user-level psr
667 *		 r30: user-level cfm
668 *		 r31: user-level pr
669 *	      f6-f11: cleared
670 *		  pr: restored (user-level pr)
671 *		  b0: restored (user-level rp)
672 *	          b6: restored
673 *		  b7: set to __kernel_syscall_via_epc
674 *	     ar.unat: restored (user-level ar.unat)
675 *	      ar.pfs: restored (user-level ar.pfs)
676 *	      ar.rsc: restored (user-level ar.rsc)
677 *	     ar.rnat: restored (user-level ar.rnat)
678 *	 ar.bspstore: restored (user-level ar.bspstore)
679 *	     ar.fpsr: restored (user-level ar.fpsr)
680 *	      ar.ccv: cleared
681 *	      ar.csd: cleared
682 *	      ar.ssd: cleared
683 */
684ENTRY(ia64_leave_syscall)
685	PT_REGS_UNWIND_INFO(0)
686	/*
687	 * work.need_resched etc. mustn't get changed by this CPU before it returns to
688	 * user- or fsys-mode, hence we disable interrupts early on.
689	 *
690	 * p6 controls whether current_thread_info()->flags needs to be check for
691	 * extra work.  We always check for extra work when returning to user-level.
692	 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
693	 * is 0.  After extra work processing has been completed, execution
694	 * resumes at .work_processed_syscall with p6 set to 1 if the extra-work-check
695	 * needs to be redone.
696	 */
697#ifdef CONFIG_PREEMPT
698	rsm psr.i				// disable interrupts
699	cmp.eq pLvSys,p0=r0,r0			// pLvSys=1: leave from syscall
700(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
701	;;
702	.pred.rel.mutex pUStk,pKStk
703(pKStk) ld4 r21=[r20]			// r21 <- preempt_count
704(pUStk)	mov r21=0			// r21 <- 0
705	;;
706	cmp.eq p6,p0=r21,r0		// p6 <- pUStk || (preempt_count == 0)
707#else /* !CONFIG_PREEMPT */
708(pUStk)	rsm psr.i
709	cmp.eq pLvSys,p0=r0,r0		// pLvSys=1: leave from syscall
710(pUStk)	cmp.eq.unc p6,p0=r0,r0		// p6 <- pUStk
711#endif
712.work_processed_syscall:
713	adds r2=PT(LOADRS)+16,r12
714	adds r3=PT(AR_BSPSTORE)+16,r12
715	adds r18=TI_FLAGS+IA64_TASK_SIZE,r13
716	;;
717(p6)	ld4 r31=[r18]				// load current_thread_info()->flags
718	ld8 r19=[r2],PT(B6)-PT(LOADRS)		// load ar.rsc value for "loadrs"
719	nop.i 0
720	;;
721	mov r16=ar.bsp				// M2  get existing backing store pointer
722	ld8 r18=[r2],PT(R9)-PT(B6)		// load b6
723(p6)	and r15=TIF_WORK_MASK,r31		// any work other than TIF_SYSCALL_TRACE?
724	;;
725	ld8 r23=[r3],PT(R11)-PT(AR_BSPSTORE)	// load ar.bspstore (may be garbage)
726(p6)	cmp4.ne.unc p6,p0=r15, r0		// any special work pending?
727(p6)	br.cond.spnt .work_pending_syscall
728	;;
729	// start restoring the state saved on the kernel stack (struct pt_regs):
730	ld8 r9=[r2],PT(CR_IPSR)-PT(R9)
731	ld8 r11=[r3],PT(CR_IIP)-PT(R11)
732(pNonSys) break 0		//      bug check: we shouldn't be here if pNonSys is TRUE!
733	;;
734	invala			// M0|1 invalidate ALAT
735	rsm psr.i | psr.ic	// M2   turn off interrupts and interruption collection
736	cmp.eq p9,p0=r0,r0	// A    set p9 to indicate that we should restore cr.ifs
737
738	ld8 r29=[r2],16		// M0|1 load cr.ipsr
739	ld8 r28=[r3],16		// M0|1 load cr.iip
740	mov r22=r0		// A    clear r22
741	;;
742	ld8 r30=[r2],16		// M0|1 load cr.ifs
743	ld8 r25=[r3],16		// M0|1 load ar.unat
744(pUStk) add r14=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13
745	;;
746	ld8 r26=[r2],PT(B0)-PT(AR_PFS)	// M0|1 load ar.pfs
747(pKStk)	mov r22=psr			// M2   read PSR now that interrupts are disabled
748	nop 0
749	;;
750	ld8 r21=[r2],PT(AR_RNAT)-PT(B0) // M0|1 load b0
751	ld8 r27=[r3],PT(PR)-PT(AR_RSC)	// M0|1 load ar.rsc
752	mov f6=f0			// F    clear f6
753	;;
754	ld8 r24=[r2],PT(AR_FPSR)-PT(AR_RNAT)	// M0|1 load ar.rnat (may be garbage)
755	ld8 r31=[r3],PT(R1)-PT(PR)		// M0|1 load predicates
756	mov f7=f0				// F    clear f7
757	;;
758	ld8 r20=[r2],PT(R12)-PT(AR_FPSR)	// M0|1 load ar.fpsr
759	ld8.fill r1=[r3],16			// M0|1 load r1
760(pUStk) mov r17=1				// A
761	;;
762(pUStk) st1 [r14]=r17				// M2|3
763	ld8.fill r13=[r3],16			// M0|1
764	mov f8=f0				// F    clear f8
765	;;
766	ld8.fill r12=[r2]			// M0|1 restore r12 (sp)
767	ld8.fill r15=[r3]			// M0|1 restore r15
768	mov b6=r18				// I0   restore b6
769
770	LOAD_PHYS_STACK_REG_SIZE(r17)
771	mov f9=f0					// F    clear f9
772(pKStk) br.cond.dpnt.many skip_rbs_switch		// B
773
774	srlz.d				// M0   ensure interruption collection is off (for cover)
775	shr.u r18=r19,16		// I0|1 get byte size of existing "dirty" partition
776	cover				// B    add current frame into dirty partition & set cr.ifs
777	;;
778	mov r19=ar.bsp			// M2   get new backing store pointer
779	mov f10=f0			// F    clear f10
780
781	nop.m 0
782	movl r14=__kernel_syscall_via_epc // X
783	;;
784	mov.m ar.csd=r0			// M2   clear ar.csd
785	mov.m ar.ccv=r0			// M2   clear ar.ccv
786	mov b7=r14			// I0   clear b7 (hint with __kernel_syscall_via_epc)
787
788	mov.m ar.ssd=r0			// M2   clear ar.ssd
789	mov f11=f0			// F    clear f11
790	br.cond.sptk.many rbs_switch	// B
791END(ia64_leave_syscall)
792
793#ifdef CONFIG_IA32_SUPPORT
794GLOBAL_ENTRY(ia64_ret_from_ia32_execve)
795	PT_REGS_UNWIND_INFO(0)
796	adds r2=PT(R8)+16,sp			// r2 = &pt_regs.r8
797	adds r3=PT(R10)+16,sp			// r3 = &pt_regs.r10
798	;;
799	.mem.offset 0,0
800	st8.spill [r2]=r8	// store return value in slot for r8 and set unat bit
801	.mem.offset 8,0
802	st8.spill [r3]=r0	// clear error indication in slot for r10 and set unat bit
803END(ia64_ret_from_ia32_execve)
804	// fall through
805#endif /* CONFIG_IA32_SUPPORT */
806GLOBAL_ENTRY(ia64_leave_kernel)
807	PT_REGS_UNWIND_INFO(0)
808	/*
809	 * work.need_resched etc. mustn't get changed by this CPU before it returns to
810	 * user- or fsys-mode, hence we disable interrupts early on.
811	 *
812	 * p6 controls whether current_thread_info()->flags needs to be check for
813	 * extra work.  We always check for extra work when returning to user-level.
814	 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
815	 * is 0.  After extra work processing has been completed, execution
816	 * resumes at .work_processed_syscall with p6 set to 1 if the extra-work-check
817	 * needs to be redone.
818	 */
819#ifdef CONFIG_PREEMPT
820	rsm psr.i				// disable interrupts
821	cmp.eq p0,pLvSys=r0,r0			// pLvSys=0: leave from kernel
822(pKStk)	adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
823	;;
824	.pred.rel.mutex pUStk,pKStk
825(pKStk)	ld4 r21=[r20]			// r21 <- preempt_count
826(pUStk)	mov r21=0			// r21 <- 0
827	;;
828	cmp.eq p6,p0=r21,r0		// p6 <- pUStk || (preempt_count == 0)
829#else
830(pUStk)	rsm psr.i
831	cmp.eq p0,pLvSys=r0,r0		// pLvSys=0: leave from kernel
832(pUStk)	cmp.eq.unc p6,p0=r0,r0		// p6 <- pUStk
833#endif
834.work_processed_kernel:
835	adds r17=TI_FLAGS+IA64_TASK_SIZE,r13
836	;;
837(p6)	ld4 r31=[r17]				// load current_thread_info()->flags
838	adds r21=PT(PR)+16,r12
839	;;
840
841	lfetch [r21],PT(CR_IPSR)-PT(PR)
842	adds r2=PT(B6)+16,r12
843	adds r3=PT(R16)+16,r12
844	;;
845	lfetch [r21]
846	ld8 r28=[r2],8		// load b6
847	adds r29=PT(R24)+16,r12
848
849	ld8.fill r16=[r3],PT(AR_CSD)-PT(R16)
850	adds r30=PT(AR_CCV)+16,r12
851(p6)	and r19=TIF_WORK_MASK,r31		// any work other than TIF_SYSCALL_TRACE?
852	;;
853	ld8.fill r24=[r29]
854	ld8 r15=[r30]		// load ar.ccv
855(p6)	cmp4.ne.unc p6,p0=r19, r0		// any special work pending?
856	;;
857	ld8 r29=[r2],16		// load b7
858	ld8 r30=[r3],16		// load ar.csd
859(p6)	br.cond.spnt .work_pending
860	;;
861	ld8 r31=[r2],16		// load ar.ssd
862	ld8.fill r8=[r3],16
863	;;
864	ld8.fill r9=[r2],16
865	ld8.fill r10=[r3],PT(R17)-PT(R10)
866	;;
867	ld8.fill r11=[r2],PT(R18)-PT(R11)
868	ld8.fill r17=[r3],16
869	;;
870	ld8.fill r18=[r2],16
871	ld8.fill r19=[r3],16
872	;;
873	ld8.fill r20=[r2],16
874	ld8.fill r21=[r3],16
875	mov ar.csd=r30
876	mov ar.ssd=r31
877	;;
878	rsm psr.i | psr.ic	// initiate turning off of interrupt and interruption collection
879	invala			// invalidate ALAT
880	;;
881	ld8.fill r22=[r2],24
882	ld8.fill r23=[r3],24
883	mov b6=r28
884	;;
885	ld8.fill r25=[r2],16
886	ld8.fill r26=[r3],16
887	mov b7=r29
888	;;
889	ld8.fill r27=[r2],16
890	ld8.fill r28=[r3],16
891	;;
892	ld8.fill r29=[r2],16
893	ld8.fill r30=[r3],24
894	;;
895	ld8.fill r31=[r2],PT(F9)-PT(R31)
896	adds r3=PT(F10)-PT(F6),r3
897	;;
898	ldf.fill f9=[r2],PT(F6)-PT(F9)
899	ldf.fill f10=[r3],PT(F8)-PT(F10)
900	;;
901	ldf.fill f6=[r2],PT(F7)-PT(F6)
902	;;
903	ldf.fill f7=[r2],PT(F11)-PT(F7)
904	ldf.fill f8=[r3],32
905	;;
906	srlz.d	// ensure that inter. collection is off (VHPT is don't care, since text is pinned)
907	mov ar.ccv=r15
908	;;
909	ldf.fill f11=[r2]
910	bsw.0			// switch back to bank 0 (no stop bit required beforehand...)
911	;;
912(pUStk)	mov r18=IA64_KR(CURRENT)// M2 (12 cycle read latency)
913	adds r16=PT(CR_IPSR)+16,r12
914	adds r17=PT(CR_IIP)+16,r12
915
916(pKStk)	mov r22=psr		// M2 read PSR now that interrupts are disabled
917	nop.i 0
918	nop.i 0
919	;;
920	ld8 r29=[r16],16	// load cr.ipsr
921	ld8 r28=[r17],16	// load cr.iip
922	;;
923	ld8 r30=[r16],16	// load cr.ifs
924	ld8 r25=[r17],16	// load ar.unat
925	;;
926	ld8 r26=[r16],16	// load ar.pfs
927	ld8 r27=[r17],16	// load ar.rsc
928	cmp.eq p9,p0=r0,r0	// set p9 to indicate that we should restore cr.ifs
929	;;
930	ld8 r24=[r16],16	// load ar.rnat (may be garbage)
931	ld8 r23=[r17],16	// load ar.bspstore (may be garbage)
932	;;
933	ld8 r31=[r16],16	// load predicates
934	ld8 r21=[r17],16	// load b0
935	;;
936	ld8 r19=[r16],16	// load ar.rsc value for "loadrs"
937	ld8.fill r1=[r17],16	// load r1
938	;;
939	ld8.fill r12=[r16],16
940	ld8.fill r13=[r17],16
941(pUStk)	adds r18=IA64_TASK_THREAD_ON_USTACK_OFFSET,r18
942	;;
943	ld8 r20=[r16],16	// ar.fpsr
944	ld8.fill r15=[r17],16
945	;;
946	ld8.fill r14=[r16],16
947	ld8.fill r2=[r17]
948(pUStk)	mov r17=1
949	;;
950	ld8.fill r3=[r16]
951(pUStk)	st1 [r18]=r17		// restore current->thread.on_ustack
952	shr.u r18=r19,16	// get byte size of existing "dirty" partition
953	;;
954	mov r16=ar.bsp		// get existing backing store pointer
955	LOAD_PHYS_STACK_REG_SIZE(r17)
956(pKStk)	br.cond.dpnt skip_rbs_switch
957
958	/*
959	 * Restore user backing store.
960	 *
961	 * NOTE: alloc, loadrs, and cover can't be predicated.
962	 */
963(pNonSys) br.cond.dpnt dont_preserve_current_frame
964	cover				// add current frame into dirty partition and set cr.ifs
965	;;
966	mov r19=ar.bsp			// get new backing store pointer
967rbs_switch:
968	sub r16=r16,r18			// krbs = old bsp - size of dirty partition
969	cmp.ne p9,p0=r0,r0		// clear p9 to skip restore of cr.ifs
970	;;
971	sub r19=r19,r16			// calculate total byte size of dirty partition
972	add r18=64,r18			// don't force in0-in7 into memory...
973	;;
974	shl r19=r19,16			// shift size of dirty partition into loadrs position
975	;;
976dont_preserve_current_frame:
977	/*
978	 * To prevent leaking bits between the kernel and user-space,
979	 * we must clear the stacked registers in the "invalid" partition here.
980	 * Not pretty, but at least it's fast (3.34 registers/cycle on Itanium,
981	 * 5 registers/cycle on McKinley).
982	 */
983#	define pRecurse	p6
984#	define pReturn	p7
985#ifdef CONFIG_ITANIUM
986#	define Nregs	10
987#else
988#	define Nregs	14
989#endif
990	alloc loc0=ar.pfs,2,Nregs-2,2,0
991	shr.u loc1=r18,9		// RNaTslots <= floor(dirtySize / (64*8))
992	sub r17=r17,r18			// r17 = (physStackedSize + 8) - dirtySize
993	;;
994	mov ar.rsc=r19			// load ar.rsc to be used for "loadrs"
995	shladd in0=loc1,3,r17
996	mov in1=0
997	;;
998	TEXT_ALIGN(32)
999rse_clear_invalid:
1000#ifdef CONFIG_ITANIUM
1001	// cycle 0
1002 { .mii
1003	alloc loc0=ar.pfs,2,Nregs-2,2,0
1004	cmp.lt pRecurse,p0=Nregs*8,in0	// if more than Nregs regs left to clear, (re)curse
1005	add out0=-Nregs*8,in0
1006}{ .mfb
1007	add out1=1,in1			// increment recursion count
1008	nop.f 0
1009	nop.b 0				// can't do br.call here because of alloc (WAW on CFM)
1010	;;
1011}{ .mfi	// cycle 1
1012	mov loc1=0
1013	nop.f 0
1014	mov loc2=0
1015}{ .mib
1016	mov loc3=0
1017	mov loc4=0
1018(pRecurse) br.call.sptk.many b0=rse_clear_invalid
1019
1020}{ .mfi	// cycle 2
1021	mov loc5=0
1022	nop.f 0
1023	cmp.ne pReturn,p0=r0,in1	// if recursion count != 0, we need to do a br.ret
1024}{ .mib
1025	mov loc6=0
1026	mov loc7=0
1027(pReturn) br.ret.sptk.many b0
1028}
1029#else /* !CONFIG_ITANIUM */
1030	alloc loc0=ar.pfs,2,Nregs-2,2,0
1031	cmp.lt pRecurse,p0=Nregs*8,in0	// if more than Nregs regs left to clear, (re)curse
1032	add out0=-Nregs*8,in0
1033	add out1=1,in1			// increment recursion count
1034	mov loc1=0
1035	mov loc2=0
1036	;;
1037	mov loc3=0
1038	mov loc4=0
1039	mov loc5=0
1040	mov loc6=0
1041	mov loc7=0
1042(pRecurse) br.call.dptk.few b0=rse_clear_invalid
1043	;;
1044	mov loc8=0
1045	mov loc9=0
1046	cmp.ne pReturn,p0=r0,in1	// if recursion count != 0, we need to do a br.ret
1047	mov loc10=0
1048	mov loc11=0
1049(pReturn) br.ret.dptk.many b0
1050#endif /* !CONFIG_ITANIUM */
1051#	undef pRecurse
1052#	undef pReturn
1053	;;
1054	alloc r17=ar.pfs,0,0,0,0	// drop current register frame
1055	;;
1056	loadrs
1057	;;
1058skip_rbs_switch:
1059	mov ar.unat=r25		// M2
1060(pKStk)	extr.u r22=r22,21,1	// I0 extract current value of psr.pp from r22
1061(pLvSys)mov r19=r0		// A  clear r19 for leave_syscall, no-op otherwise
1062	;;
1063(pUStk)	mov ar.bspstore=r23	// M2
1064(pKStk)	dep r29=r22,r29,21,1	// I0 update ipsr.pp with psr.pp
1065(pLvSys)mov r16=r0		// A  clear r16 for leave_syscall, no-op otherwise
1066	;;
1067	mov cr.ipsr=r29		// M2
1068	mov ar.pfs=r26		// I0
1069(pLvSys)mov r17=r0		// A  clear r17 for leave_syscall, no-op otherwise
1070
1071(p9)	mov cr.ifs=r30		// M2
1072	mov b0=r21		// I0
1073(pLvSys)mov r18=r0		// A  clear r18 for leave_syscall, no-op otherwise
1074
1075	mov ar.fpsr=r20		// M2
1076	mov cr.iip=r28		// M2
1077	nop 0
1078	;;
1079(pUStk)	mov ar.rnat=r24		// M2 must happen with RSE in lazy mode
1080	nop 0
1081(pLvSys)mov r2=r0
1082
1083	mov ar.rsc=r27		// M2
1084	mov pr=r31,-1		// I0
1085	rfi			// B
1086
1087	/*
1088	 * On entry:
1089	 *	r20 = &current->thread_info->pre_count (if CONFIG_PREEMPT)
1090	 *	r31 = current->thread_info->flags
1091	 * On exit:
1092	 *	p6 = TRUE if work-pending-check needs to be redone
1093	 */
1094.work_pending_syscall:
1095	add r2=-8,r2
1096	add r3=-8,r3
1097	;;
1098	st8 [r2]=r8
1099	st8 [r3]=r10
1100.work_pending:
1101	tbit.z p6,p0=r31,TIF_NEED_RESCHED		// current_thread_info()->need_resched==0?
1102(p6)	br.cond.sptk.few .notify
1103#ifdef CONFIG_PREEMPT
1104(pKStk) dep r21=-1,r0,PREEMPT_ACTIVE_BIT,1
1105	;;
1106(pKStk) st4 [r20]=r21
1107	ssm psr.i		// enable interrupts
1108#endif
1109	br.call.spnt.many rp=schedule
1110.ret9:	cmp.eq p6,p0=r0,r0				// p6 <- 1
1111	rsm psr.i		// disable interrupts
1112	;;
1113#ifdef CONFIG_PREEMPT
1114(pKStk)	adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
1115	;;
1116(pKStk)	st4 [r20]=r0		// preempt_count() <- 0
1117#endif
1118(pLvSys)br.cond.sptk.few  .work_pending_syscall_end
1119	br.cond.sptk.many .work_processed_kernel	// re-check
1120
1121.notify:
1122(pUStk)	br.call.spnt.many rp=notify_resume_user
1123.ret10:	cmp.ne p6,p0=r0,r0				// p6 <- 0
1124(pLvSys)br.cond.sptk.few  .work_pending_syscall_end
1125	br.cond.sptk.many .work_processed_kernel	// don't re-check
1126
1127.work_pending_syscall_end:
1128	adds r2=PT(R8)+16,r12
1129	adds r3=PT(R10)+16,r12
1130	;;
1131	ld8 r8=[r2]
1132	ld8 r10=[r3]
1133	br.cond.sptk.many .work_processed_syscall	// re-check
1134
1135END(ia64_leave_kernel)
1136
1137ENTRY(handle_syscall_error)
1138	/*
1139	 * Some system calls (e.g., ptrace, mmap) can return arbitrary values which could
1140	 * lead us to mistake a negative return value as a failed syscall.  Those syscall
1141	 * must deposit a non-zero value in pt_regs.r8 to indicate an error.  If
1142	 * pt_regs.r8 is zero, we assume that the call completed successfully.
1143	 */
1144	PT_REGS_UNWIND_INFO(0)
1145	ld8 r3=[r2]		// load pt_regs.r8
1146	;;
1147	cmp.eq p6,p7=r3,r0	// is pt_regs.r8==0?
1148	;;
1149(p7)	mov r10=-1
1150(p7)	sub r8=0,r8		// negate return value to get errno
1151	br.cond.sptk ia64_leave_syscall
1152END(handle_syscall_error)
1153
1154	/*
1155	 * Invoke schedule_tail(task) while preserving in0-in7, which may be needed
1156	 * in case a system call gets restarted.
1157	 */
1158GLOBAL_ENTRY(ia64_invoke_schedule_tail)
1159	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1160	alloc loc1=ar.pfs,8,2,1,0
1161	mov loc0=rp
1162	mov out0=r8				// Address of previous task
1163	;;
1164	br.call.sptk.many rp=schedule_tail
1165.ret11:	mov ar.pfs=loc1
1166	mov rp=loc0
1167	br.ret.sptk.many rp
1168END(ia64_invoke_schedule_tail)
1169
1170	/*
1171	 * Setup stack and call do_notify_resume_user().  Note that pSys and pNonSys need to
1172	 * be set up by the caller.  We declare 8 input registers so the system call
1173	 * args get preserved, in case we need to restart a system call.
1174	 */
1175ENTRY(notify_resume_user)
1176	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1177	alloc loc1=ar.pfs,8,2,3,0 // preserve all eight input regs in case of syscall restart!
1178	mov r9=ar.unat
1179	mov loc0=rp				// save return address
1180	mov out0=0				// there is no "oldset"
1181	adds out1=8,sp				// out1=&sigscratch->ar_pfs
1182(pSys)	mov out2=1				// out2==1 => we're in a syscall
1183	;;
1184(pNonSys) mov out2=0				// out2==0 => not a syscall
1185	.fframe 16
1186	.spillsp ar.unat, 16
1187	st8 [sp]=r9,-16				// allocate space for ar.unat and save it
1188	st8 [out1]=loc1,-8			// save ar.pfs, out1=&sigscratch
1189	.body
1190	br.call.sptk.many rp=do_notify_resume_user
1191.ret15:	.restore sp
1192	adds sp=16,sp				// pop scratch stack space
1193	;;
1194	ld8 r9=[sp]				// load new unat from sigscratch->scratch_unat
1195	mov rp=loc0
1196	;;
1197	mov ar.unat=r9
1198	mov ar.pfs=loc1
1199	br.ret.sptk.many rp
1200END(notify_resume_user)
1201
1202ENTRY(sys_rt_sigreturn)
1203	PT_REGS_UNWIND_INFO(0)
1204	/*
1205	 * Allocate 8 input registers since ptrace() may clobber them
1206	 */
1207	alloc r2=ar.pfs,8,0,1,0
1208	.prologue
1209	PT_REGS_SAVES(16)
1210	adds sp=-16,sp
1211	.body
1212	cmp.eq pNonSys,pSys=r0,r0		// sigreturn isn't a normal syscall...
1213	;;
1214	/*
1215	 * leave_kernel() restores f6-f11 from pt_regs, but since the streamlined
1216	 * syscall-entry path does not save them we save them here instead.  Note: we
1217	 * don't need to save any other registers that are not saved by the stream-lined
1218	 * syscall path, because restore_sigcontext() restores them.
1219	 */
1220	adds r16=PT(F6)+32,sp
1221	adds r17=PT(F7)+32,sp
1222	;;
1223 	stf.spill [r16]=f6,32
1224 	stf.spill [r17]=f7,32
1225	;;
1226 	stf.spill [r16]=f8,32
1227 	stf.spill [r17]=f9,32
1228	;;
1229 	stf.spill [r16]=f10
1230 	stf.spill [r17]=f11
1231	adds out0=16,sp				// out0 = &sigscratch
1232	br.call.sptk.many rp=ia64_rt_sigreturn
1233.ret19:	.restore sp,0
1234	adds sp=16,sp
1235	;;
1236	ld8 r9=[sp]				// load new ar.unat
1237	mov.sptk b7=r8,ia64_leave_kernel
1238	;;
1239	mov ar.unat=r9
1240	br.many b7
1241END(sys_rt_sigreturn)
1242
1243GLOBAL_ENTRY(ia64_prepare_handle_unaligned)
1244	.prologue
1245	/*
1246	 * r16 = fake ar.pfs, we simply need to make sure privilege is still 0
1247	 */
1248	mov r16=r0
1249	DO_SAVE_SWITCH_STACK
1250	br.call.sptk.many rp=ia64_handle_unaligned	// stack frame setup in ivt
1251.ret21:	.body
1252	DO_LOAD_SWITCH_STACK
1253	br.cond.sptk.many rp				// goes to ia64_leave_kernel
1254END(ia64_prepare_handle_unaligned)
1255
1256	//
1257	// unw_init_running(void (*callback)(info, arg), void *arg)
1258	//
1259#	define EXTRA_FRAME_SIZE	((UNW_FRAME_INFO_SIZE+15)&~15)
1260
1261GLOBAL_ENTRY(unw_init_running)
1262	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1263	alloc loc1=ar.pfs,2,3,3,0
1264	;;
1265	ld8 loc2=[in0],8
1266	mov loc0=rp
1267	mov r16=loc1
1268	DO_SAVE_SWITCH_STACK
1269	.body
1270
1271	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1272	.fframe IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE
1273	SWITCH_STACK_SAVES(EXTRA_FRAME_SIZE)
1274	adds sp=-EXTRA_FRAME_SIZE,sp
1275	.body
1276	;;
1277	adds out0=16,sp				// &info
1278	mov out1=r13				// current
1279	adds out2=16+EXTRA_FRAME_SIZE,sp	// &switch_stack
1280	br.call.sptk.many rp=unw_init_frame_info
12811:	adds out0=16,sp				// &info
1282	mov b6=loc2
1283	mov loc2=gp				// save gp across indirect function call
1284	;;
1285	ld8 gp=[in0]
1286	mov out1=in1				// arg
1287	br.call.sptk.many rp=b6			// invoke the callback function
12881:	mov gp=loc2				// restore gp
1289
1290	// For now, we don't allow changing registers from within
1291	// unw_init_running; if we ever want to allow that, we'd
1292	// have to do a load_switch_stack here:
1293	.restore sp
1294	adds sp=IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE,sp
1295
1296	mov ar.pfs=loc1
1297	mov rp=loc0
1298	br.ret.sptk.many rp
1299END(unw_init_running)
1300
1301	.rodata
1302	.align 8
1303	.globl sys_call_table
1304sys_call_table:
1305	data8 sys_ni_syscall		//  This must be sys_ni_syscall!  See ivt.S.
1306	data8 sys_exit				// 1025
1307	data8 sys_read
1308	data8 sys_write
1309	data8 sys_open
1310	data8 sys_close
1311	data8 sys_creat				// 1030
1312	data8 sys_link
1313	data8 sys_unlink
1314	data8 ia64_execve
1315	data8 sys_chdir
1316	data8 sys_fchdir			// 1035
1317	data8 sys_utimes
1318	data8 sys_mknod
1319	data8 sys_chmod
1320	data8 sys_chown
1321	data8 sys_lseek				// 1040
1322	data8 sys_getpid
1323	data8 sys_getppid
1324	data8 sys_mount
1325	data8 sys_umount
1326	data8 sys_setuid			// 1045
1327	data8 sys_getuid
1328	data8 sys_geteuid
1329	data8 sys_ptrace
1330	data8 sys_access
1331	data8 sys_sync				// 1050
1332	data8 sys_fsync
1333	data8 sys_fdatasync
1334	data8 sys_kill
1335	data8 sys_rename
1336	data8 sys_mkdir				// 1055
1337	data8 sys_rmdir
1338	data8 sys_dup
1339	data8 sys_pipe
1340	data8 sys_times
1341	data8 ia64_brk				// 1060
1342	data8 sys_setgid
1343	data8 sys_getgid
1344	data8 sys_getegid
1345	data8 sys_acct
1346	data8 sys_ioctl				// 1065
1347	data8 sys_fcntl
1348	data8 sys_umask
1349	data8 sys_chroot
1350	data8 sys_ustat
1351	data8 sys_dup2				// 1070
1352	data8 sys_setreuid
1353	data8 sys_setregid
1354	data8 sys_getresuid
1355	data8 sys_setresuid
1356	data8 sys_getresgid			// 1075
1357	data8 sys_setresgid
1358	data8 sys_getgroups
1359	data8 sys_setgroups
1360	data8 sys_getpgid
1361	data8 sys_setpgid			// 1080
1362	data8 sys_setsid
1363	data8 sys_getsid
1364	data8 sys_sethostname
1365	data8 sys_setrlimit
1366	data8 sys_getrlimit			// 1085
1367	data8 sys_getrusage
1368	data8 sys_gettimeofday
1369	data8 sys_settimeofday
1370	data8 sys_select
1371	data8 sys_poll				// 1090
1372	data8 sys_symlink
1373	data8 sys_readlink
1374	data8 sys_uselib
1375	data8 sys_swapon
1376	data8 sys_swapoff			// 1095
1377	data8 sys_reboot
1378	data8 sys_truncate
1379	data8 sys_ftruncate
1380	data8 sys_fchmod
1381	data8 sys_fchown			// 1100
1382	data8 ia64_getpriority
1383	data8 sys_setpriority
1384	data8 sys_statfs
1385	data8 sys_fstatfs
1386	data8 sys_gettid			// 1105
1387	data8 sys_semget
1388	data8 sys_semop
1389	data8 sys_semctl
1390	data8 sys_msgget
1391	data8 sys_msgsnd			// 1110
1392	data8 sys_msgrcv
1393	data8 sys_msgctl
1394	data8 sys_shmget
1395	data8 sys_shmat
1396	data8 sys_shmdt				// 1115
1397	data8 sys_shmctl
1398	data8 sys_syslog
1399	data8 sys_setitimer
1400	data8 sys_getitimer
1401	data8 sys_ni_syscall			// 1120		/* was: ia64_oldstat */
1402	data8 sys_ni_syscall					/* was: ia64_oldlstat */
1403	data8 sys_ni_syscall					/* was: ia64_oldfstat */
1404	data8 sys_vhangup
1405	data8 sys_lchown
1406	data8 sys_remap_file_pages		// 1125
1407	data8 sys_wait4
1408	data8 sys_sysinfo
1409	data8 sys_clone
1410	data8 sys_setdomainname
1411	data8 sys_newuname			// 1130
1412	data8 sys_adjtimex
1413	data8 sys_ni_syscall					/* was: ia64_create_module */
1414	data8 sys_init_module
1415	data8 sys_delete_module
1416	data8 sys_ni_syscall			// 1135		/* was: sys_get_kernel_syms */
1417	data8 sys_ni_syscall					/* was: sys_query_module */
1418	data8 sys_quotactl
1419	data8 sys_bdflush
1420	data8 sys_sysfs
1421	data8 sys_personality			// 1140
1422	data8 sys_ni_syscall		// sys_afs_syscall
1423	data8 sys_setfsuid
1424	data8 sys_setfsgid
1425	data8 sys_getdents
1426	data8 sys_flock				// 1145
1427	data8 sys_readv
1428	data8 sys_writev
1429	data8 sys_pread64
1430	data8 sys_pwrite64
1431	data8 sys_sysctl			// 1150
1432	data8 sys_mmap
1433	data8 sys_munmap
1434	data8 sys_mlock
1435	data8 sys_mlockall
1436	data8 sys_mprotect			// 1155
1437	data8 ia64_mremap
1438	data8 sys_msync
1439	data8 sys_munlock
1440	data8 sys_munlockall
1441	data8 sys_sched_getparam		// 1160
1442	data8 sys_sched_setparam
1443	data8 sys_sched_getscheduler
1444	data8 sys_sched_setscheduler
1445	data8 sys_sched_yield
1446	data8 sys_sched_get_priority_max	// 1165
1447	data8 sys_sched_get_priority_min
1448	data8 sys_sched_rr_get_interval
1449	data8 sys_nanosleep
1450	data8 sys_nfsservctl
1451	data8 sys_prctl				// 1170
1452	data8 sys_getpagesize
1453	data8 sys_mmap2
1454	data8 sys_pciconfig_read
1455	data8 sys_pciconfig_write
1456	data8 sys_perfmonctl			// 1175
1457	data8 sys_sigaltstack
1458	data8 sys_rt_sigaction
1459	data8 sys_rt_sigpending
1460	data8 sys_rt_sigprocmask
1461	data8 sys_rt_sigqueueinfo		// 1180
1462	data8 sys_rt_sigreturn
1463	data8 sys_rt_sigsuspend
1464	data8 sys_rt_sigtimedwait
1465	data8 sys_getcwd
1466	data8 sys_capget			// 1185
1467	data8 sys_capset
1468	data8 sys_sendfile64
1469	data8 sys_ni_syscall		// sys_getpmsg (STREAMS)
1470	data8 sys_ni_syscall		// sys_putpmsg (STREAMS)
1471	data8 sys_socket			// 1190
1472	data8 sys_bind
1473	data8 sys_connect
1474	data8 sys_listen
1475	data8 sys_accept
1476	data8 sys_getsockname			// 1195
1477	data8 sys_getpeername
1478	data8 sys_socketpair
1479	data8 sys_send
1480	data8 sys_sendto
1481	data8 sys_recv				// 1200
1482	data8 sys_recvfrom
1483	data8 sys_shutdown
1484	data8 sys_setsockopt
1485	data8 sys_getsockopt
1486	data8 sys_sendmsg			// 1205
1487	data8 sys_recvmsg
1488	data8 sys_pivot_root
1489	data8 sys_mincore
1490	data8 sys_madvise
1491	data8 sys_newstat			// 1210
1492	data8 sys_newlstat
1493	data8 sys_newfstat
1494	data8 sys_clone2
1495	data8 sys_getdents64
1496	data8 sys_getunwind			// 1215
1497	data8 sys_readahead
1498	data8 sys_setxattr
1499	data8 sys_lsetxattr
1500	data8 sys_fsetxattr
1501	data8 sys_getxattr			// 1220
1502	data8 sys_lgetxattr
1503	data8 sys_fgetxattr
1504	data8 sys_listxattr
1505	data8 sys_llistxattr
1506	data8 sys_flistxattr			// 1225
1507	data8 sys_removexattr
1508	data8 sys_lremovexattr
1509	data8 sys_fremovexattr
1510	data8 sys_tkill
1511	data8 sys_futex				// 1230
1512	data8 sys_sched_setaffinity
1513	data8 sys_sched_getaffinity
1514	data8 sys_set_tid_address
1515	data8 sys_fadvise64_64
1516	data8 sys_tgkill 			// 1235
1517	data8 sys_exit_group
1518	data8 sys_lookup_dcookie
1519	data8 sys_io_setup
1520	data8 sys_io_destroy
1521	data8 sys_io_getevents			// 1240
1522	data8 sys_io_submit
1523	data8 sys_io_cancel
1524	data8 sys_epoll_create
1525	data8 sys_epoll_ctl
1526	data8 sys_epoll_wait			// 1245
1527	data8 sys_restart_syscall
1528	data8 sys_semtimedop
1529	data8 sys_timer_create
1530	data8 sys_timer_settime
1531	data8 sys_timer_gettime			// 1250
1532	data8 sys_timer_getoverrun
1533	data8 sys_timer_delete
1534	data8 sys_clock_settime
1535	data8 sys_clock_gettime
1536	data8 sys_clock_getres			// 1255
1537	data8 sys_clock_nanosleep
1538	data8 sys_fstatfs64
1539	data8 sys_statfs64
1540	data8 sys_mbind
1541	data8 sys_get_mempolicy			// 1260
1542	data8 sys_set_mempolicy
1543	data8 sys_mq_open
1544	data8 sys_mq_unlink
1545	data8 sys_mq_timedsend
1546	data8 sys_mq_timedreceive		// 1265
1547	data8 sys_mq_notify
1548	data8 sys_mq_getsetattr
1549	data8 sys_kexec_load
1550	data8 sys_ni_syscall			// reserved for vserver
1551	data8 sys_waitid			// 1270
1552	data8 sys_add_key
1553	data8 sys_request_key
1554	data8 sys_keyctl
1555	data8 sys_ioprio_set
1556	data8 sys_ioprio_get			// 1275
1557	data8 sys_move_pages
1558	data8 sys_inotify_init
1559	data8 sys_inotify_add_watch
1560	data8 sys_inotify_rm_watch
1561	data8 sys_migrate_pages			// 1280
1562	data8 sys_openat
1563	data8 sys_mkdirat
1564	data8 sys_mknodat
1565	data8 sys_fchownat
1566	data8 sys_futimesat			// 1285
1567	data8 sys_newfstatat
1568	data8 sys_unlinkat
1569	data8 sys_renameat
1570	data8 sys_linkat
1571	data8 sys_symlinkat			// 1290
1572	data8 sys_readlinkat
1573	data8 sys_fchmodat
1574	data8 sys_faccessat
1575	data8 sys_pselect6
1576	data8 sys_ppoll
1577	data8 sys_unshare
1578	data8 sys_splice
1579	data8 sys_set_robust_list
1580	data8 sys_get_robust_list
1581	data8 sys_sync_file_range		// 1300
1582	data8 sys_tee
1583	data8 sys_vmsplice
1584	data8 sys_ni_syscall			// reserved for move_pages
1585	data8 sys_getcpu
1586	data8 sys_epoll_pwait			// 1305
1587	data8 sys_utimensat
1588	data8 sys_signalfd
1589	data8 sys_timerfd
1590	data8 sys_eventfd
1591
1592	.org sys_call_table + 8*NR_syscalls	// guard against failures to increase NR_syscalls
1593