xref: /openbmc/linux/arch/ia64/kernel/entry.S (revision c8ec3743)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * arch/ia64/kernel/entry.S
4 *
5 * Kernel entry points.
6 *
7 * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co
8 *	David Mosberger-Tang <davidm@hpl.hp.com>
9 * Copyright (C) 1999, 2002-2003
10 *	Asit Mallick <Asit.K.Mallick@intel.com>
11 * 	Don Dugger <Don.Dugger@intel.com>
12 *	Suresh Siddha <suresh.b.siddha@intel.com>
13 *	Fenghua Yu <fenghua.yu@intel.com>
14 * Copyright (C) 1999 VA Linux Systems
15 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
16 */
17/*
18 * ia64_switch_to now places correct virtual mapping in in TR2 for
19 * kernel stack. This allows us to handle interrupts without changing
20 * to physical mode.
21 *
22 * Jonathan Nicklin	<nicklin@missioncriticallinux.com>
23 * Patrick O'Rourke	<orourke@missioncriticallinux.com>
24 * 11/07/2000
25 */
26/*
27 * Copyright (c) 2008 Isaku Yamahata <yamahata at valinux co jp>
28 *                    VA Linux Systems Japan K.K.
29 *                    pv_ops.
30 */
31/*
32 * Global (preserved) predicate usage on syscall entry/exit path:
33 *
34 *	pKStk:		See entry.h.
35 *	pUStk:		See entry.h.
36 *	pSys:		See entry.h.
37 *	pNonSys:	!pSys
38 */
39
40
41#include <asm/asmmacro.h>
42#include <asm/cache.h>
43#include <asm/errno.h>
44#include <asm/kregs.h>
45#include <asm/asm-offsets.h>
46#include <asm/pgtable.h>
47#include <asm/percpu.h>
48#include <asm/processor.h>
49#include <asm/thread_info.h>
50#include <asm/unistd.h>
51#include <asm/ftrace.h>
52#include <asm/export.h>
53
54#include "minstate.h"
55
56	/*
57	 * execve() is special because in case of success, we need to
58	 * setup a null register window frame.
59	 */
60ENTRY(ia64_execve)
61	/*
62	 * Allocate 8 input registers since ptrace() may clobber them
63	 */
64	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
65	alloc loc1=ar.pfs,8,2,3,0
66	mov loc0=rp
67	.body
68	mov out0=in0			// filename
69	;;				// stop bit between alloc and call
70	mov out1=in1			// argv
71	mov out2=in2			// envp
72	br.call.sptk.many rp=sys_execve
73.ret0:
74	cmp4.ge p6,p7=r8,r0
75	mov ar.pfs=loc1			// restore ar.pfs
76	sxt4 r8=r8			// return 64-bit result
77	;;
78	stf.spill [sp]=f0
79	mov rp=loc0
80(p6)	mov ar.pfs=r0			// clear ar.pfs on success
81(p7)	br.ret.sptk.many rp
82
83	/*
84	 * In theory, we'd have to zap this state only to prevent leaking of
85	 * security sensitive state (e.g., if current->mm->dumpable is zero).  However,
86	 * this executes in less than 20 cycles even on Itanium, so it's not worth
87	 * optimizing for...).
88	 */
89	mov ar.unat=0; 		mov ar.lc=0
90	mov r4=0;		mov f2=f0;		mov b1=r0
91	mov r5=0;		mov f3=f0;		mov b2=r0
92	mov r6=0;		mov f4=f0;		mov b3=r0
93	mov r7=0;		mov f5=f0;		mov b4=r0
94	ldf.fill f12=[sp];	mov f13=f0;		mov b5=r0
95	ldf.fill f14=[sp];	ldf.fill f15=[sp];	mov f16=f0
96	ldf.fill f17=[sp];	ldf.fill f18=[sp];	mov f19=f0
97	ldf.fill f20=[sp];	ldf.fill f21=[sp];	mov f22=f0
98	ldf.fill f23=[sp];	ldf.fill f24=[sp];	mov f25=f0
99	ldf.fill f26=[sp];	ldf.fill f27=[sp];	mov f28=f0
100	ldf.fill f29=[sp];	ldf.fill f30=[sp];	mov f31=f0
101	br.ret.sptk.many rp
102END(ia64_execve)
103
104/*
105 * sys_clone2(u64 flags, u64 ustack_base, u64 ustack_size, u64 parent_tidptr, u64 child_tidptr,
106 *	      u64 tls)
107 */
108GLOBAL_ENTRY(sys_clone2)
109	/*
110	 * Allocate 8 input registers since ptrace() may clobber them
111	 */
112	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
113	alloc r16=ar.pfs,8,2,6,0
114	DO_SAVE_SWITCH_STACK
115	adds r2=PT(R16)+IA64_SWITCH_STACK_SIZE+16,sp
116	mov loc0=rp
117	mov loc1=r16				// save ar.pfs across do_fork
118	.body
119	mov out1=in1
120	mov out2=in2
121	tbit.nz p6,p0=in0,CLONE_SETTLS_BIT
122	mov out3=in3	// parent_tidptr: valid only w/CLONE_PARENT_SETTID
123	;;
124(p6)	st8 [r2]=in5				// store TLS in r16 for copy_thread()
125	mov out4=in4	// child_tidptr:  valid only w/CLONE_CHILD_SETTID or CLONE_CHILD_CLEARTID
126	mov out0=in0				// out0 = clone_flags
127	br.call.sptk.many rp=do_fork
128.ret1:	.restore sp
129	adds sp=IA64_SWITCH_STACK_SIZE,sp	// pop the switch stack
130	mov ar.pfs=loc1
131	mov rp=loc0
132	br.ret.sptk.many rp
133END(sys_clone2)
134
135/*
136 * sys_clone(u64 flags, u64 ustack_base, u64 parent_tidptr, u64 child_tidptr, u64 tls)
137 *	Deprecated.  Use sys_clone2() instead.
138 */
139GLOBAL_ENTRY(sys_clone)
140	/*
141	 * Allocate 8 input registers since ptrace() may clobber them
142	 */
143	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
144	alloc r16=ar.pfs,8,2,6,0
145	DO_SAVE_SWITCH_STACK
146	adds r2=PT(R16)+IA64_SWITCH_STACK_SIZE+16,sp
147	mov loc0=rp
148	mov loc1=r16				// save ar.pfs across do_fork
149	.body
150	mov out1=in1
151	mov out2=16				// stacksize (compensates for 16-byte scratch area)
152	tbit.nz p6,p0=in0,CLONE_SETTLS_BIT
153	mov out3=in2	// parent_tidptr: valid only w/CLONE_PARENT_SETTID
154	;;
155(p6)	st8 [r2]=in4				// store TLS in r13 (tp)
156	mov out4=in3	// child_tidptr:  valid only w/CLONE_CHILD_SETTID or CLONE_CHILD_CLEARTID
157	mov out0=in0				// out0 = clone_flags
158	br.call.sptk.many rp=do_fork
159.ret2:	.restore sp
160	adds sp=IA64_SWITCH_STACK_SIZE,sp	// pop the switch stack
161	mov ar.pfs=loc1
162	mov rp=loc0
163	br.ret.sptk.many rp
164END(sys_clone)
165
166/*
167 * prev_task <- ia64_switch_to(struct task_struct *next)
168 *	With Ingo's new scheduler, interrupts are disabled when this routine gets
169 *	called.  The code starting at .map relies on this.  The rest of the code
170 *	doesn't care about the interrupt masking status.
171 */
172GLOBAL_ENTRY(ia64_switch_to)
173	.prologue
174	alloc r16=ar.pfs,1,0,0,0
175	DO_SAVE_SWITCH_STACK
176	.body
177
178	adds r22=IA64_TASK_THREAD_KSP_OFFSET,r13
179	movl r25=init_task
180	mov r27=IA64_KR(CURRENT_STACK)
181	adds r21=IA64_TASK_THREAD_KSP_OFFSET,in0
182	dep r20=0,in0,61,3		// physical address of "next"
183	;;
184	st8 [r22]=sp			// save kernel stack pointer of old task
185	shr.u r26=r20,IA64_GRANULE_SHIFT
186	cmp.eq p7,p6=r25,in0
187	;;
188	/*
189	 * If we've already mapped this task's page, we can skip doing it again.
190	 */
191(p6)	cmp.eq p7,p6=r26,r27
192(p6)	br.cond.dpnt .map
193	;;
194.done:
195	ld8 sp=[r21]			// load kernel stack pointer of new task
196	MOV_TO_KR(CURRENT, in0, r8, r9)		// update "current" application register
197	mov r8=r13			// return pointer to previously running task
198	mov r13=in0			// set "current" pointer
199	;;
200	DO_LOAD_SWITCH_STACK
201
202#ifdef CONFIG_SMP
203	sync.i				// ensure "fc"s done by this CPU are visible on other CPUs
204#endif
205	br.ret.sptk.many rp		// boogie on out in new context
206
207.map:
208	RSM_PSR_IC(r25)			// interrupts (psr.i) are already disabled here
209	movl r25=PAGE_KERNEL
210	;;
211	srlz.d
212	or r23=r25,r20			// construct PA | page properties
213	mov r25=IA64_GRANULE_SHIFT<<2
214	;;
215	MOV_TO_ITIR(p0, r25, r8)
216	MOV_TO_IFA(in0, r8)		// VA of next task...
217	;;
218	mov r25=IA64_TR_CURRENT_STACK
219	MOV_TO_KR(CURRENT_STACK, r26, r8, r9)	// remember last page we mapped...
220	;;
221	itr.d dtr[r25]=r23		// wire in new mapping...
222	SSM_PSR_IC_AND_SRLZ_D(r8, r9)	// reenable the psr.ic bit
223	br.cond.sptk .done
224END(ia64_switch_to)
225
226/*
227 * Note that interrupts are enabled during save_switch_stack and load_switch_stack.  This
228 * means that we may get an interrupt with "sp" pointing to the new kernel stack while
229 * ar.bspstore is still pointing to the old kernel backing store area.  Since ar.rsc,
230 * ar.rnat, ar.bsp, and ar.bspstore are all preserved by interrupts, this is not a
231 * problem.  Also, we don't need to specify unwind information for preserved registers
232 * that are not modified in save_switch_stack as the right unwind information is already
233 * specified at the call-site of save_switch_stack.
234 */
235
236/*
237 * save_switch_stack:
238 *	- r16 holds ar.pfs
239 *	- b7 holds address to return to
240 *	- rp (b0) holds return address to save
241 */
242GLOBAL_ENTRY(save_switch_stack)
243	.prologue
244	.altrp b7
245	flushrs			// flush dirty regs to backing store (must be first in insn group)
246	.save @priunat,r17
247	mov r17=ar.unat		// preserve caller's
248	.body
249#ifdef CONFIG_ITANIUM
250	adds r2=16+128,sp
251	adds r3=16+64,sp
252	adds r14=SW(R4)+16,sp
253	;;
254	st8.spill [r14]=r4,16		// spill r4
255	lfetch.fault.excl.nt1 [r3],128
256	;;
257	lfetch.fault.excl.nt1 [r2],128
258	lfetch.fault.excl.nt1 [r3],128
259	;;
260	lfetch.fault.excl [r2]
261	lfetch.fault.excl [r3]
262	adds r15=SW(R5)+16,sp
263#else
264	add r2=16+3*128,sp
265	add r3=16,sp
266	add r14=SW(R4)+16,sp
267	;;
268	st8.spill [r14]=r4,SW(R6)-SW(R4)	// spill r4 and prefetch offset 0x1c0
269	lfetch.fault.excl.nt1 [r3],128	//		prefetch offset 0x010
270	;;
271	lfetch.fault.excl.nt1 [r3],128	//		prefetch offset 0x090
272	lfetch.fault.excl.nt1 [r2],128	//		prefetch offset 0x190
273	;;
274	lfetch.fault.excl.nt1 [r3]	//		prefetch offset 0x110
275	lfetch.fault.excl.nt1 [r2]	//		prefetch offset 0x210
276	adds r15=SW(R5)+16,sp
277#endif
278	;;
279	st8.spill [r15]=r5,SW(R7)-SW(R5)	// spill r5
280	mov.m ar.rsc=0			// put RSE in mode: enforced lazy, little endian, pl 0
281	add r2=SW(F2)+16,sp		// r2 = &sw->f2
282	;;
283	st8.spill [r14]=r6,SW(B0)-SW(R6)	// spill r6
284	mov.m r18=ar.fpsr		// preserve fpsr
285	add r3=SW(F3)+16,sp		// r3 = &sw->f3
286	;;
287	stf.spill [r2]=f2,32
288	mov.m r19=ar.rnat
289	mov r21=b0
290
291	stf.spill [r3]=f3,32
292	st8.spill [r15]=r7,SW(B2)-SW(R7)	// spill r7
293	mov r22=b1
294	;;
295	// since we're done with the spills, read and save ar.unat:
296	mov.m r29=ar.unat
297	mov.m r20=ar.bspstore
298	mov r23=b2
299	stf.spill [r2]=f4,32
300	stf.spill [r3]=f5,32
301	mov r24=b3
302	;;
303	st8 [r14]=r21,SW(B1)-SW(B0)		// save b0
304	st8 [r15]=r23,SW(B3)-SW(B2)		// save b2
305	mov r25=b4
306	mov r26=b5
307	;;
308	st8 [r14]=r22,SW(B4)-SW(B1)		// save b1
309	st8 [r15]=r24,SW(AR_PFS)-SW(B3)		// save b3
310	mov r21=ar.lc		// I-unit
311	stf.spill [r2]=f12,32
312	stf.spill [r3]=f13,32
313	;;
314	st8 [r14]=r25,SW(B5)-SW(B4)		// save b4
315	st8 [r15]=r16,SW(AR_LC)-SW(AR_PFS)	// save ar.pfs
316	stf.spill [r2]=f14,32
317	stf.spill [r3]=f15,32
318	;;
319	st8 [r14]=r26				// save b5
320	st8 [r15]=r21				// save ar.lc
321	stf.spill [r2]=f16,32
322	stf.spill [r3]=f17,32
323	;;
324	stf.spill [r2]=f18,32
325	stf.spill [r3]=f19,32
326	;;
327	stf.spill [r2]=f20,32
328	stf.spill [r3]=f21,32
329	;;
330	stf.spill [r2]=f22,32
331	stf.spill [r3]=f23,32
332	;;
333	stf.spill [r2]=f24,32
334	stf.spill [r3]=f25,32
335	;;
336	stf.spill [r2]=f26,32
337	stf.spill [r3]=f27,32
338	;;
339	stf.spill [r2]=f28,32
340	stf.spill [r3]=f29,32
341	;;
342	stf.spill [r2]=f30,SW(AR_UNAT)-SW(F30)
343	stf.spill [r3]=f31,SW(PR)-SW(F31)
344	add r14=SW(CALLER_UNAT)+16,sp
345	;;
346	st8 [r2]=r29,SW(AR_RNAT)-SW(AR_UNAT)	// save ar.unat
347	st8 [r14]=r17,SW(AR_FPSR)-SW(CALLER_UNAT) // save caller_unat
348	mov r21=pr
349	;;
350	st8 [r2]=r19,SW(AR_BSPSTORE)-SW(AR_RNAT) // save ar.rnat
351	st8 [r3]=r21				// save predicate registers
352	;;
353	st8 [r2]=r20				// save ar.bspstore
354	st8 [r14]=r18				// save fpsr
355	mov ar.rsc=3		// put RSE back into eager mode, pl 0
356	br.cond.sptk.many b7
357END(save_switch_stack)
358
359/*
360 * load_switch_stack:
361 *	- "invala" MUST be done at call site (normally in DO_LOAD_SWITCH_STACK)
362 *	- b7 holds address to return to
363 *	- must not touch r8-r11
364 */
365GLOBAL_ENTRY(load_switch_stack)
366	.prologue
367	.altrp b7
368
369	.body
370	lfetch.fault.nt1 [sp]
371	adds r2=SW(AR_BSPSTORE)+16,sp
372	adds r3=SW(AR_UNAT)+16,sp
373	mov ar.rsc=0						// put RSE into enforced lazy mode
374	adds r14=SW(CALLER_UNAT)+16,sp
375	adds r15=SW(AR_FPSR)+16,sp
376	;;
377	ld8 r27=[r2],(SW(B0)-SW(AR_BSPSTORE))	// bspstore
378	ld8 r29=[r3],(SW(B1)-SW(AR_UNAT))	// unat
379	;;
380	ld8 r21=[r2],16		// restore b0
381	ld8 r22=[r3],16		// restore b1
382	;;
383	ld8 r23=[r2],16		// restore b2
384	ld8 r24=[r3],16		// restore b3
385	;;
386	ld8 r25=[r2],16		// restore b4
387	ld8 r26=[r3],16		// restore b5
388	;;
389	ld8 r16=[r2],(SW(PR)-SW(AR_PFS))	// ar.pfs
390	ld8 r17=[r3],(SW(AR_RNAT)-SW(AR_LC))	// ar.lc
391	;;
392	ld8 r28=[r2]		// restore pr
393	ld8 r30=[r3]		// restore rnat
394	;;
395	ld8 r18=[r14],16	// restore caller's unat
396	ld8 r19=[r15],24	// restore fpsr
397	;;
398	ldf.fill f2=[r14],32
399	ldf.fill f3=[r15],32
400	;;
401	ldf.fill f4=[r14],32
402	ldf.fill f5=[r15],32
403	;;
404	ldf.fill f12=[r14],32
405	ldf.fill f13=[r15],32
406	;;
407	ldf.fill f14=[r14],32
408	ldf.fill f15=[r15],32
409	;;
410	ldf.fill f16=[r14],32
411	ldf.fill f17=[r15],32
412	;;
413	ldf.fill f18=[r14],32
414	ldf.fill f19=[r15],32
415	mov b0=r21
416	;;
417	ldf.fill f20=[r14],32
418	ldf.fill f21=[r15],32
419	mov b1=r22
420	;;
421	ldf.fill f22=[r14],32
422	ldf.fill f23=[r15],32
423	mov b2=r23
424	;;
425	mov ar.bspstore=r27
426	mov ar.unat=r29		// establish unat holding the NaT bits for r4-r7
427	mov b3=r24
428	;;
429	ldf.fill f24=[r14],32
430	ldf.fill f25=[r15],32
431	mov b4=r25
432	;;
433	ldf.fill f26=[r14],32
434	ldf.fill f27=[r15],32
435	mov b5=r26
436	;;
437	ldf.fill f28=[r14],32
438	ldf.fill f29=[r15],32
439	mov ar.pfs=r16
440	;;
441	ldf.fill f30=[r14],32
442	ldf.fill f31=[r15],24
443	mov ar.lc=r17
444	;;
445	ld8.fill r4=[r14],16
446	ld8.fill r5=[r15],16
447	mov pr=r28,-1
448	;;
449	ld8.fill r6=[r14],16
450	ld8.fill r7=[r15],16
451
452	mov ar.unat=r18				// restore caller's unat
453	mov ar.rnat=r30				// must restore after bspstore but before rsc!
454	mov ar.fpsr=r19				// restore fpsr
455	mov ar.rsc=3				// put RSE back into eager mode, pl 0
456	br.cond.sptk.many b7
457END(load_switch_stack)
458
459	/*
460	 * Invoke a system call, but do some tracing before and after the call.
461	 * We MUST preserve the current register frame throughout this routine
462	 * because some system calls (such as ia64_execve) directly
463	 * manipulate ar.pfs.
464	 */
465GLOBAL_ENTRY(ia64_trace_syscall)
466	PT_REGS_UNWIND_INFO(0)
467	/*
468	 * We need to preserve the scratch registers f6-f11 in case the system
469	 * call is sigreturn.
470	 */
471	adds r16=PT(F6)+16,sp
472	adds r17=PT(F7)+16,sp
473	;;
474 	stf.spill [r16]=f6,32
475 	stf.spill [r17]=f7,32
476	;;
477 	stf.spill [r16]=f8,32
478 	stf.spill [r17]=f9,32
479	;;
480 	stf.spill [r16]=f10
481 	stf.spill [r17]=f11
482	br.call.sptk.many rp=syscall_trace_enter // give parent a chance to catch syscall args
483	cmp.lt p6,p0=r8,r0			// check tracehook
484	adds r2=PT(R8)+16,sp			// r2 = &pt_regs.r8
485	adds r3=PT(R10)+16,sp			// r3 = &pt_regs.r10
486	mov r10=0
487(p6)	br.cond.sptk strace_error		// syscall failed ->
488	adds r16=PT(F6)+16,sp
489	adds r17=PT(F7)+16,sp
490	;;
491	ldf.fill f6=[r16],32
492	ldf.fill f7=[r17],32
493	;;
494	ldf.fill f8=[r16],32
495	ldf.fill f9=[r17],32
496	;;
497	ldf.fill f10=[r16]
498	ldf.fill f11=[r17]
499	// the syscall number may have changed, so re-load it and re-calculate the
500	// syscall entry-point:
501	adds r15=PT(R15)+16,sp			// r15 = &pt_regs.r15 (syscall #)
502	;;
503	ld8 r15=[r15]
504	mov r3=NR_syscalls - 1
505	;;
506	adds r15=-1024,r15
507	movl r16=sys_call_table
508	;;
509	shladd r20=r15,3,r16			// r20 = sys_call_table + 8*(syscall-1024)
510	cmp.leu p6,p7=r15,r3
511	;;
512(p6)	ld8 r20=[r20]				// load address of syscall entry point
513(p7)	movl r20=sys_ni_syscall
514	;;
515	mov b6=r20
516	br.call.sptk.many rp=b6			// do the syscall
517.strace_check_retval:
518	cmp.lt p6,p0=r8,r0			// syscall failed?
519	adds r2=PT(R8)+16,sp			// r2 = &pt_regs.r8
520	adds r3=PT(R10)+16,sp			// r3 = &pt_regs.r10
521	mov r10=0
522(p6)	br.cond.sptk strace_error		// syscall failed ->
523	;;					// avoid RAW on r10
524.strace_save_retval:
525.mem.offset 0,0; st8.spill [r2]=r8		// store return value in slot for r8
526.mem.offset 8,0; st8.spill [r3]=r10		// clear error indication in slot for r10
527	br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
528.ret3:
529(pUStk)	cmp.eq.unc p6,p0=r0,r0			// p6 <- pUStk
530(pUStk)	rsm psr.i				// disable interrupts
531	br.cond.sptk ia64_work_pending_syscall_end
532
533strace_error:
534	ld8 r3=[r2]				// load pt_regs.r8
535	sub r9=0,r8				// negate return value to get errno value
536	;;
537	cmp.ne p6,p0=r3,r0			// is pt_regs.r8!=0?
538	adds r3=16,r2				// r3=&pt_regs.r10
539	;;
540(p6)	mov r10=-1
541(p6)	mov r8=r9
542	br.cond.sptk .strace_save_retval
543END(ia64_trace_syscall)
544
545	/*
546	 * When traced and returning from sigreturn, we invoke syscall_trace but then
547	 * go straight to ia64_leave_kernel rather than ia64_leave_syscall.
548	 */
549GLOBAL_ENTRY(ia64_strace_leave_kernel)
550	PT_REGS_UNWIND_INFO(0)
551{	/*
552	 * Some versions of gas generate bad unwind info if the first instruction of a
553	 * procedure doesn't go into the first slot of a bundle.  This is a workaround.
554	 */
555	nop.m 0
556	nop.i 0
557	br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
558}
559.ret4:	br.cond.sptk ia64_leave_kernel
560END(ia64_strace_leave_kernel)
561
562ENTRY(call_payload)
563	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(0)
564	/* call the kernel_thread payload; fn is in r4, arg - in r5 */
565	alloc loc1=ar.pfs,0,3,1,0
566	mov loc0=rp
567	mov loc2=gp
568	mov out0=r5		// arg
569	ld8 r14 = [r4], 8	// fn.address
570	;;
571	mov b6 = r14
572	ld8 gp = [r4]		// fn.gp
573	;;
574	br.call.sptk.many rp=b6	// fn(arg)
575.ret12:	mov gp=loc2
576	mov rp=loc0
577	mov ar.pfs=loc1
578	/* ... and if it has returned, we are going to userland */
579	cmp.ne pKStk,pUStk=r0,r0
580	br.ret.sptk.many rp
581END(call_payload)
582
583GLOBAL_ENTRY(ia64_ret_from_clone)
584	PT_REGS_UNWIND_INFO(0)
585{	/*
586	 * Some versions of gas generate bad unwind info if the first instruction of a
587	 * procedure doesn't go into the first slot of a bundle.  This is a workaround.
588	 */
589	nop.m 0
590	nop.i 0
591	/*
592	 * We need to call schedule_tail() to complete the scheduling process.
593	 * Called by ia64_switch_to() after do_fork()->copy_thread().  r8 contains the
594	 * address of the previously executing task.
595	 */
596	br.call.sptk.many rp=ia64_invoke_schedule_tail
597}
598.ret8:
599(pKStk)	br.call.sptk.many rp=call_payload
600	adds r2=TI_FLAGS+IA64_TASK_SIZE,r13
601	;;
602	ld4 r2=[r2]
603	;;
604	mov r8=0
605	and r2=_TIF_SYSCALL_TRACEAUDIT,r2
606	;;
607	cmp.ne p6,p0=r2,r0
608(p6)	br.cond.spnt .strace_check_retval
609	;;					// added stop bits to prevent r8 dependency
610END(ia64_ret_from_clone)
611	// fall through
612GLOBAL_ENTRY(ia64_ret_from_syscall)
613	PT_REGS_UNWIND_INFO(0)
614	cmp.ge p6,p7=r8,r0			// syscall executed successfully?
615	adds r2=PT(R8)+16,sp			// r2 = &pt_regs.r8
616	mov r10=r0				// clear error indication in r10
617(p7)	br.cond.spnt handle_syscall_error	// handle potential syscall failure
618END(ia64_ret_from_syscall)
619	// fall through
620
621/*
622 * ia64_leave_syscall(): Same as ia64_leave_kernel, except that it doesn't
623 *	need to switch to bank 0 and doesn't restore the scratch registers.
624 *	To avoid leaking kernel bits, the scratch registers are set to
625 *	the following known-to-be-safe values:
626 *
627 *		  r1: restored (global pointer)
628 *		  r2: cleared
629 *		  r3: 1 (when returning to user-level)
630 *	      r8-r11: restored (syscall return value(s))
631 *		 r12: restored (user-level stack pointer)
632 *		 r13: restored (user-level thread pointer)
633 *		 r14: set to __kernel_syscall_via_epc
634 *		 r15: restored (syscall #)
635 *	     r16-r17: cleared
636 *		 r18: user-level b6
637 *		 r19: cleared
638 *		 r20: user-level ar.fpsr
639 *		 r21: user-level b0
640 *		 r22: cleared
641 *		 r23: user-level ar.bspstore
642 *		 r24: user-level ar.rnat
643 *		 r25: user-level ar.unat
644 *		 r26: user-level ar.pfs
645 *		 r27: user-level ar.rsc
646 *		 r28: user-level ip
647 *		 r29: user-level psr
648 *		 r30: user-level cfm
649 *		 r31: user-level pr
650 *	      f6-f11: cleared
651 *		  pr: restored (user-level pr)
652 *		  b0: restored (user-level rp)
653 *	          b6: restored
654 *		  b7: set to __kernel_syscall_via_epc
655 *	     ar.unat: restored (user-level ar.unat)
656 *	      ar.pfs: restored (user-level ar.pfs)
657 *	      ar.rsc: restored (user-level ar.rsc)
658 *	     ar.rnat: restored (user-level ar.rnat)
659 *	 ar.bspstore: restored (user-level ar.bspstore)
660 *	     ar.fpsr: restored (user-level ar.fpsr)
661 *	      ar.ccv: cleared
662 *	      ar.csd: cleared
663 *	      ar.ssd: cleared
664 */
665GLOBAL_ENTRY(ia64_leave_syscall)
666	PT_REGS_UNWIND_INFO(0)
667	/*
668	 * work.need_resched etc. mustn't get changed by this CPU before it returns to
669	 * user- or fsys-mode, hence we disable interrupts early on.
670	 *
671	 * p6 controls whether current_thread_info()->flags needs to be check for
672	 * extra work.  We always check for extra work when returning to user-level.
673	 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
674	 * is 0.  After extra work processing has been completed, execution
675	 * resumes at ia64_work_processed_syscall with p6 set to 1 if the extra-work-check
676	 * needs to be redone.
677	 */
678#ifdef CONFIG_PREEMPT
679	RSM_PSR_I(p0, r2, r18)			// disable interrupts
680	cmp.eq pLvSys,p0=r0,r0			// pLvSys=1: leave from syscall
681(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
682	;;
683	.pred.rel.mutex pUStk,pKStk
684(pKStk) ld4 r21=[r20]			// r21 <- preempt_count
685(pUStk)	mov r21=0			// r21 <- 0
686	;;
687	cmp.eq p6,p0=r21,r0		// p6 <- pUStk || (preempt_count == 0)
688#else /* !CONFIG_PREEMPT */
689	RSM_PSR_I(pUStk, r2, r18)
690	cmp.eq pLvSys,p0=r0,r0		// pLvSys=1: leave from syscall
691(pUStk)	cmp.eq.unc p6,p0=r0,r0		// p6 <- pUStk
692#endif
693.global ia64_work_processed_syscall;
694ia64_work_processed_syscall:
695#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
696	adds r2=PT(LOADRS)+16,r12
697	MOV_FROM_ITC(pUStk, p9, r22, r19)	// fetch time at leave
698	adds r18=TI_FLAGS+IA64_TASK_SIZE,r13
699	;;
700(p6)	ld4 r31=[r18]				// load current_thread_info()->flags
701	ld8 r19=[r2],PT(B6)-PT(LOADRS)		// load ar.rsc value for "loadrs"
702	adds r3=PT(AR_BSPSTORE)+16,r12		// deferred
703	;;
704#else
705	adds r2=PT(LOADRS)+16,r12
706	adds r3=PT(AR_BSPSTORE)+16,r12
707	adds r18=TI_FLAGS+IA64_TASK_SIZE,r13
708	;;
709(p6)	ld4 r31=[r18]				// load current_thread_info()->flags
710	ld8 r19=[r2],PT(B6)-PT(LOADRS)		// load ar.rsc value for "loadrs"
711	nop.i 0
712	;;
713#endif
714	mov r16=ar.bsp				// M2  get existing backing store pointer
715	ld8 r18=[r2],PT(R9)-PT(B6)		// load b6
716(p6)	and r15=TIF_WORK_MASK,r31		// any work other than TIF_SYSCALL_TRACE?
717	;;
718	ld8 r23=[r3],PT(R11)-PT(AR_BSPSTORE)	// load ar.bspstore (may be garbage)
719(p6)	cmp4.ne.unc p6,p0=r15, r0		// any special work pending?
720(p6)	br.cond.spnt .work_pending_syscall
721	;;
722	// start restoring the state saved on the kernel stack (struct pt_regs):
723	ld8 r9=[r2],PT(CR_IPSR)-PT(R9)
724	ld8 r11=[r3],PT(CR_IIP)-PT(R11)
725(pNonSys) break 0		//      bug check: we shouldn't be here if pNonSys is TRUE!
726	;;
727	invala			// M0|1 invalidate ALAT
728	RSM_PSR_I_IC(r28, r29, r30)	// M2   turn off interrupts and interruption collection
729	cmp.eq p9,p0=r0,r0	// A    set p9 to indicate that we should restore cr.ifs
730
731	ld8 r29=[r2],16		// M0|1 load cr.ipsr
732	ld8 r28=[r3],16		// M0|1 load cr.iip
733#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
734(pUStk) add r14=TI_AC_LEAVE+IA64_TASK_SIZE,r13
735	;;
736	ld8 r30=[r2],16		// M0|1 load cr.ifs
737	ld8 r25=[r3],16		// M0|1 load ar.unat
738(pUStk) add r15=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13
739	;;
740#else
741	mov r22=r0		// A    clear r22
742	;;
743	ld8 r30=[r2],16		// M0|1 load cr.ifs
744	ld8 r25=[r3],16		// M0|1 load ar.unat
745(pUStk) add r14=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13
746	;;
747#endif
748	ld8 r26=[r2],PT(B0)-PT(AR_PFS)	// M0|1 load ar.pfs
749	MOV_FROM_PSR(pKStk, r22, r21)	// M2   read PSR now that interrupts are disabled
750	nop 0
751	;;
752	ld8 r21=[r2],PT(AR_RNAT)-PT(B0) // M0|1 load b0
753	ld8 r27=[r3],PT(PR)-PT(AR_RSC)	// M0|1 load ar.rsc
754	mov f6=f0			// F    clear f6
755	;;
756	ld8 r24=[r2],PT(AR_FPSR)-PT(AR_RNAT)	// M0|1 load ar.rnat (may be garbage)
757	ld8 r31=[r3],PT(R1)-PT(PR)		// M0|1 load predicates
758	mov f7=f0				// F    clear f7
759	;;
760	ld8 r20=[r2],PT(R12)-PT(AR_FPSR)	// M0|1 load ar.fpsr
761	ld8.fill r1=[r3],16			// M0|1 load r1
762(pUStk) mov r17=1				// A
763	;;
764#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
765(pUStk) st1 [r15]=r17				// M2|3
766#else
767(pUStk) st1 [r14]=r17				// M2|3
768#endif
769	ld8.fill r13=[r3],16			// M0|1
770	mov f8=f0				// F    clear f8
771	;;
772	ld8.fill r12=[r2]			// M0|1 restore r12 (sp)
773	ld8.fill r15=[r3]			// M0|1 restore r15
774	mov b6=r18				// I0   restore b6
775
776	LOAD_PHYS_STACK_REG_SIZE(r17)
777	mov f9=f0					// F    clear f9
778(pKStk) br.cond.dpnt.many skip_rbs_switch		// B
779
780	srlz.d				// M0   ensure interruption collection is off (for cover)
781	shr.u r18=r19,16		// I0|1 get byte size of existing "dirty" partition
782	COVER				// B    add current frame into dirty partition & set cr.ifs
783	;;
784#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
785	mov r19=ar.bsp			// M2   get new backing store pointer
786	st8 [r14]=r22			// M	save time at leave
787	mov f10=f0			// F    clear f10
788
789	mov r22=r0			// A	clear r22
790	movl r14=__kernel_syscall_via_epc // X
791	;;
792#else
793	mov r19=ar.bsp			// M2   get new backing store pointer
794	mov f10=f0			// F    clear f10
795
796	nop.m 0
797	movl r14=__kernel_syscall_via_epc // X
798	;;
799#endif
800	mov.m ar.csd=r0			// M2   clear ar.csd
801	mov.m ar.ccv=r0			// M2   clear ar.ccv
802	mov b7=r14			// I0   clear b7 (hint with __kernel_syscall_via_epc)
803
804	mov.m ar.ssd=r0			// M2   clear ar.ssd
805	mov f11=f0			// F    clear f11
806	br.cond.sptk.many rbs_switch	// B
807END(ia64_leave_syscall)
808
809GLOBAL_ENTRY(ia64_leave_kernel)
810	PT_REGS_UNWIND_INFO(0)
811	/*
812	 * work.need_resched etc. mustn't get changed by this CPU before it returns to
813	 * user- or fsys-mode, hence we disable interrupts early on.
814	 *
815	 * p6 controls whether current_thread_info()->flags needs to be check for
816	 * extra work.  We always check for extra work when returning to user-level.
817	 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
818	 * is 0.  After extra work processing has been completed, execution
819	 * resumes at .work_processed_syscall with p6 set to 1 if the extra-work-check
820	 * needs to be redone.
821	 */
822#ifdef CONFIG_PREEMPT
823	RSM_PSR_I(p0, r17, r31)			// disable interrupts
824	cmp.eq p0,pLvSys=r0,r0			// pLvSys=0: leave from kernel
825(pKStk)	adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
826	;;
827	.pred.rel.mutex pUStk,pKStk
828(pKStk)	ld4 r21=[r20]			// r21 <- preempt_count
829(pUStk)	mov r21=0			// r21 <- 0
830	;;
831	cmp.eq p6,p0=r21,r0		// p6 <- pUStk || (preempt_count == 0)
832#else
833	RSM_PSR_I(pUStk, r17, r31)
834	cmp.eq p0,pLvSys=r0,r0		// pLvSys=0: leave from kernel
835(pUStk)	cmp.eq.unc p6,p0=r0,r0		// p6 <- pUStk
836#endif
837.work_processed_kernel:
838	adds r17=TI_FLAGS+IA64_TASK_SIZE,r13
839	;;
840(p6)	ld4 r31=[r17]				// load current_thread_info()->flags
841	adds r21=PT(PR)+16,r12
842	;;
843
844	lfetch [r21],PT(CR_IPSR)-PT(PR)
845	adds r2=PT(B6)+16,r12
846	adds r3=PT(R16)+16,r12
847	;;
848	lfetch [r21]
849	ld8 r28=[r2],8		// load b6
850	adds r29=PT(R24)+16,r12
851
852	ld8.fill r16=[r3],PT(AR_CSD)-PT(R16)
853	adds r30=PT(AR_CCV)+16,r12
854(p6)	and r19=TIF_WORK_MASK,r31		// any work other than TIF_SYSCALL_TRACE?
855	;;
856	ld8.fill r24=[r29]
857	ld8 r15=[r30]		// load ar.ccv
858(p6)	cmp4.ne.unc p6,p0=r19, r0		// any special work pending?
859	;;
860	ld8 r29=[r2],16		// load b7
861	ld8 r30=[r3],16		// load ar.csd
862(p6)	br.cond.spnt .work_pending
863	;;
864	ld8 r31=[r2],16		// load ar.ssd
865	ld8.fill r8=[r3],16
866	;;
867	ld8.fill r9=[r2],16
868	ld8.fill r10=[r3],PT(R17)-PT(R10)
869	;;
870	ld8.fill r11=[r2],PT(R18)-PT(R11)
871	ld8.fill r17=[r3],16
872	;;
873	ld8.fill r18=[r2],16
874	ld8.fill r19=[r3],16
875	;;
876	ld8.fill r20=[r2],16
877	ld8.fill r21=[r3],16
878	mov ar.csd=r30
879	mov ar.ssd=r31
880	;;
881	RSM_PSR_I_IC(r23, r22, r25)	// initiate turning off of interrupt and interruption collection
882	invala			// invalidate ALAT
883	;;
884	ld8.fill r22=[r2],24
885	ld8.fill r23=[r3],24
886	mov b6=r28
887	;;
888	ld8.fill r25=[r2],16
889	ld8.fill r26=[r3],16
890	mov b7=r29
891	;;
892	ld8.fill r27=[r2],16
893	ld8.fill r28=[r3],16
894	;;
895	ld8.fill r29=[r2],16
896	ld8.fill r30=[r3],24
897	;;
898	ld8.fill r31=[r2],PT(F9)-PT(R31)
899	adds r3=PT(F10)-PT(F6),r3
900	;;
901	ldf.fill f9=[r2],PT(F6)-PT(F9)
902	ldf.fill f10=[r3],PT(F8)-PT(F10)
903	;;
904	ldf.fill f6=[r2],PT(F7)-PT(F6)
905	;;
906	ldf.fill f7=[r2],PT(F11)-PT(F7)
907	ldf.fill f8=[r3],32
908	;;
909	srlz.d	// ensure that inter. collection is off (VHPT is don't care, since text is pinned)
910	mov ar.ccv=r15
911	;;
912	ldf.fill f11=[r2]
913	BSW_0(r2, r3, r15)	// switch back to bank 0 (no stop bit required beforehand...)
914	;;
915(pUStk)	mov r18=IA64_KR(CURRENT)// M2 (12 cycle read latency)
916	adds r16=PT(CR_IPSR)+16,r12
917	adds r17=PT(CR_IIP)+16,r12
918
919#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
920	.pred.rel.mutex pUStk,pKStk
921	MOV_FROM_PSR(pKStk, r22, r29)	// M2 read PSR now that interrupts are disabled
922	MOV_FROM_ITC(pUStk, p9, r22, r29)	// M  fetch time at leave
923	nop.i 0
924	;;
925#else
926	MOV_FROM_PSR(pKStk, r22, r29)	// M2 read PSR now that interrupts are disabled
927	nop.i 0
928	nop.i 0
929	;;
930#endif
931	ld8 r29=[r16],16	// load cr.ipsr
932	ld8 r28=[r17],16	// load cr.iip
933	;;
934	ld8 r30=[r16],16	// load cr.ifs
935	ld8 r25=[r17],16	// load ar.unat
936	;;
937	ld8 r26=[r16],16	// load ar.pfs
938	ld8 r27=[r17],16	// load ar.rsc
939	cmp.eq p9,p0=r0,r0	// set p9 to indicate that we should restore cr.ifs
940	;;
941	ld8 r24=[r16],16	// load ar.rnat (may be garbage)
942	ld8 r23=[r17],16	// load ar.bspstore (may be garbage)
943	;;
944	ld8 r31=[r16],16	// load predicates
945	ld8 r21=[r17],16	// load b0
946	;;
947	ld8 r19=[r16],16	// load ar.rsc value for "loadrs"
948	ld8.fill r1=[r17],16	// load r1
949	;;
950	ld8.fill r12=[r16],16
951	ld8.fill r13=[r17],16
952#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
953(pUStk)	adds r3=TI_AC_LEAVE+IA64_TASK_SIZE,r18
954#else
955(pUStk)	adds r18=IA64_TASK_THREAD_ON_USTACK_OFFSET,r18
956#endif
957	;;
958	ld8 r20=[r16],16	// ar.fpsr
959	ld8.fill r15=[r17],16
960#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
961(pUStk)	adds r18=IA64_TASK_THREAD_ON_USTACK_OFFSET,r18	// deferred
962#endif
963	;;
964	ld8.fill r14=[r16],16
965	ld8.fill r2=[r17]
966(pUStk)	mov r17=1
967	;;
968#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
969	//  mmi_ :  ld8 st1 shr;;         mmi_ : st8 st1 shr;;
970	//  mib  :  mov add br        ->  mib  : ld8 add br
971	//  bbb_ :  br  nop cover;;       mbb_ : mov br  cover;;
972	//
973	//  no one require bsp in r16 if (pKStk) branch is selected.
974(pUStk)	st8 [r3]=r22		// save time at leave
975(pUStk)	st1 [r18]=r17		// restore current->thread.on_ustack
976	shr.u r18=r19,16	// get byte size of existing "dirty" partition
977	;;
978	ld8.fill r3=[r16]	// deferred
979	LOAD_PHYS_STACK_REG_SIZE(r17)
980(pKStk)	br.cond.dpnt skip_rbs_switch
981	mov r16=ar.bsp		// get existing backing store pointer
982#else
983	ld8.fill r3=[r16]
984(pUStk)	st1 [r18]=r17		// restore current->thread.on_ustack
985	shr.u r18=r19,16	// get byte size of existing "dirty" partition
986	;;
987	mov r16=ar.bsp		// get existing backing store pointer
988	LOAD_PHYS_STACK_REG_SIZE(r17)
989(pKStk)	br.cond.dpnt skip_rbs_switch
990#endif
991
992	/*
993	 * Restore user backing store.
994	 *
995	 * NOTE: alloc, loadrs, and cover can't be predicated.
996	 */
997(pNonSys) br.cond.dpnt dont_preserve_current_frame
998	COVER				// add current frame into dirty partition and set cr.ifs
999	;;
1000	mov r19=ar.bsp			// get new backing store pointer
1001rbs_switch:
1002	sub r16=r16,r18			// krbs = old bsp - size of dirty partition
1003	cmp.ne p9,p0=r0,r0		// clear p9 to skip restore of cr.ifs
1004	;;
1005	sub r19=r19,r16			// calculate total byte size of dirty partition
1006	add r18=64,r18			// don't force in0-in7 into memory...
1007	;;
1008	shl r19=r19,16			// shift size of dirty partition into loadrs position
1009	;;
1010dont_preserve_current_frame:
1011	/*
1012	 * To prevent leaking bits between the kernel and user-space,
1013	 * we must clear the stacked registers in the "invalid" partition here.
1014	 * Not pretty, but at least it's fast (3.34 registers/cycle on Itanium,
1015	 * 5 registers/cycle on McKinley).
1016	 */
1017#	define pRecurse	p6
1018#	define pReturn	p7
1019#ifdef CONFIG_ITANIUM
1020#	define Nregs	10
1021#else
1022#	define Nregs	14
1023#endif
1024	alloc loc0=ar.pfs,2,Nregs-2,2,0
1025	shr.u loc1=r18,9		// RNaTslots <= floor(dirtySize / (64*8))
1026	sub r17=r17,r18			// r17 = (physStackedSize + 8) - dirtySize
1027	;;
1028	mov ar.rsc=r19			// load ar.rsc to be used for "loadrs"
1029	shladd in0=loc1,3,r17
1030	mov in1=0
1031	;;
1032	TEXT_ALIGN(32)
1033rse_clear_invalid:
1034#ifdef CONFIG_ITANIUM
1035	// cycle 0
1036 { .mii
1037	alloc loc0=ar.pfs,2,Nregs-2,2,0
1038	cmp.lt pRecurse,p0=Nregs*8,in0	// if more than Nregs regs left to clear, (re)curse
1039	add out0=-Nregs*8,in0
1040}{ .mfb
1041	add out1=1,in1			// increment recursion count
1042	nop.f 0
1043	nop.b 0				// can't do br.call here because of alloc (WAW on CFM)
1044	;;
1045}{ .mfi	// cycle 1
1046	mov loc1=0
1047	nop.f 0
1048	mov loc2=0
1049}{ .mib
1050	mov loc3=0
1051	mov loc4=0
1052(pRecurse) br.call.sptk.many b0=rse_clear_invalid
1053
1054}{ .mfi	// cycle 2
1055	mov loc5=0
1056	nop.f 0
1057	cmp.ne pReturn,p0=r0,in1	// if recursion count != 0, we need to do a br.ret
1058}{ .mib
1059	mov loc6=0
1060	mov loc7=0
1061(pReturn) br.ret.sptk.many b0
1062}
1063#else /* !CONFIG_ITANIUM */
1064	alloc loc0=ar.pfs,2,Nregs-2,2,0
1065	cmp.lt pRecurse,p0=Nregs*8,in0	// if more than Nregs regs left to clear, (re)curse
1066	add out0=-Nregs*8,in0
1067	add out1=1,in1			// increment recursion count
1068	mov loc1=0
1069	mov loc2=0
1070	;;
1071	mov loc3=0
1072	mov loc4=0
1073	mov loc5=0
1074	mov loc6=0
1075	mov loc7=0
1076(pRecurse) br.call.dptk.few b0=rse_clear_invalid
1077	;;
1078	mov loc8=0
1079	mov loc9=0
1080	cmp.ne pReturn,p0=r0,in1	// if recursion count != 0, we need to do a br.ret
1081	mov loc10=0
1082	mov loc11=0
1083(pReturn) br.ret.dptk.many b0
1084#endif /* !CONFIG_ITANIUM */
1085#	undef pRecurse
1086#	undef pReturn
1087	;;
1088	alloc r17=ar.pfs,0,0,0,0	// drop current register frame
1089	;;
1090	loadrs
1091	;;
1092skip_rbs_switch:
1093	mov ar.unat=r25		// M2
1094(pKStk)	extr.u r22=r22,21,1	// I0 extract current value of psr.pp from r22
1095(pLvSys)mov r19=r0		// A  clear r19 for leave_syscall, no-op otherwise
1096	;;
1097(pUStk)	mov ar.bspstore=r23	// M2
1098(pKStk)	dep r29=r22,r29,21,1	// I0 update ipsr.pp with psr.pp
1099(pLvSys)mov r16=r0		// A  clear r16 for leave_syscall, no-op otherwise
1100	;;
1101	MOV_TO_IPSR(p0, r29, r25)	// M2
1102	mov ar.pfs=r26		// I0
1103(pLvSys)mov r17=r0		// A  clear r17 for leave_syscall, no-op otherwise
1104
1105	MOV_TO_IFS(p9, r30, r25)// M2
1106	mov b0=r21		// I0
1107(pLvSys)mov r18=r0		// A  clear r18 for leave_syscall, no-op otherwise
1108
1109	mov ar.fpsr=r20		// M2
1110	MOV_TO_IIP(r28, r25)	// M2
1111	nop 0
1112	;;
1113(pUStk)	mov ar.rnat=r24		// M2 must happen with RSE in lazy mode
1114	nop 0
1115(pLvSys)mov r2=r0
1116
1117	mov ar.rsc=r27		// M2
1118	mov pr=r31,-1		// I0
1119	RFI			// B
1120
1121	/*
1122	 * On entry:
1123	 *	r20 = &current->thread_info->pre_count (if CONFIG_PREEMPT)
1124	 *	r31 = current->thread_info->flags
1125	 * On exit:
1126	 *	p6 = TRUE if work-pending-check needs to be redone
1127	 *
1128	 * Interrupts are disabled on entry, reenabled depend on work, and
1129	 * disabled on exit.
1130	 */
1131.work_pending_syscall:
1132	add r2=-8,r2
1133	add r3=-8,r3
1134	;;
1135	st8 [r2]=r8
1136	st8 [r3]=r10
1137.work_pending:
1138	tbit.z p6,p0=r31,TIF_NEED_RESCHED	// is resched not needed?
1139(p6)	br.cond.sptk.few .notify
1140	br.call.spnt.many rp=preempt_schedule_irq
1141.ret9:	cmp.eq p6,p0=r0,r0	// p6 <- 1 (re-check)
1142(pLvSys)br.cond.sptk.few  ia64_work_pending_syscall_end
1143	br.cond.sptk.many .work_processed_kernel
1144
1145.notify:
1146(pUStk)	br.call.spnt.many rp=notify_resume_user
1147.ret10:	cmp.ne p6,p0=r0,r0	// p6 <- 0 (don't re-check)
1148(pLvSys)br.cond.sptk.few  ia64_work_pending_syscall_end
1149	br.cond.sptk.many .work_processed_kernel
1150
1151.global ia64_work_pending_syscall_end;
1152ia64_work_pending_syscall_end:
1153	adds r2=PT(R8)+16,r12
1154	adds r3=PT(R10)+16,r12
1155	;;
1156	ld8 r8=[r2]
1157	ld8 r10=[r3]
1158	br.cond.sptk.many ia64_work_processed_syscall
1159END(ia64_leave_kernel)
1160
1161ENTRY(handle_syscall_error)
1162	/*
1163	 * Some system calls (e.g., ptrace, mmap) can return arbitrary values which could
1164	 * lead us to mistake a negative return value as a failed syscall.  Those syscall
1165	 * must deposit a non-zero value in pt_regs.r8 to indicate an error.  If
1166	 * pt_regs.r8 is zero, we assume that the call completed successfully.
1167	 */
1168	PT_REGS_UNWIND_INFO(0)
1169	ld8 r3=[r2]		// load pt_regs.r8
1170	;;
1171	cmp.eq p6,p7=r3,r0	// is pt_regs.r8==0?
1172	;;
1173(p7)	mov r10=-1
1174(p7)	sub r8=0,r8		// negate return value to get errno
1175	br.cond.sptk ia64_leave_syscall
1176END(handle_syscall_error)
1177
1178	/*
1179	 * Invoke schedule_tail(task) while preserving in0-in7, which may be needed
1180	 * in case a system call gets restarted.
1181	 */
1182GLOBAL_ENTRY(ia64_invoke_schedule_tail)
1183	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1184	alloc loc1=ar.pfs,8,2,1,0
1185	mov loc0=rp
1186	mov out0=r8				// Address of previous task
1187	;;
1188	br.call.sptk.many rp=schedule_tail
1189.ret11:	mov ar.pfs=loc1
1190	mov rp=loc0
1191	br.ret.sptk.many rp
1192END(ia64_invoke_schedule_tail)
1193
1194	/*
1195	 * Setup stack and call do_notify_resume_user(), keeping interrupts
1196	 * disabled.
1197	 *
1198	 * Note that pSys and pNonSys need to be set up by the caller.
1199	 * We declare 8 input registers so the system call args get preserved,
1200	 * in case we need to restart a system call.
1201	 */
1202GLOBAL_ENTRY(notify_resume_user)
1203	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1204	alloc loc1=ar.pfs,8,2,3,0 // preserve all eight input regs in case of syscall restart!
1205	mov r9=ar.unat
1206	mov loc0=rp				// save return address
1207	mov out0=0				// there is no "oldset"
1208	adds out1=8,sp				// out1=&sigscratch->ar_pfs
1209(pSys)	mov out2=1				// out2==1 => we're in a syscall
1210	;;
1211(pNonSys) mov out2=0				// out2==0 => not a syscall
1212	.fframe 16
1213	.spillsp ar.unat, 16
1214	st8 [sp]=r9,-16				// allocate space for ar.unat and save it
1215	st8 [out1]=loc1,-8			// save ar.pfs, out1=&sigscratch
1216	.body
1217	br.call.sptk.many rp=do_notify_resume_user
1218.ret15:	.restore sp
1219	adds sp=16,sp				// pop scratch stack space
1220	;;
1221	ld8 r9=[sp]				// load new unat from sigscratch->scratch_unat
1222	mov rp=loc0
1223	;;
1224	mov ar.unat=r9
1225	mov ar.pfs=loc1
1226	br.ret.sptk.many rp
1227END(notify_resume_user)
1228
1229ENTRY(sys_rt_sigreturn)
1230	PT_REGS_UNWIND_INFO(0)
1231	/*
1232	 * Allocate 8 input registers since ptrace() may clobber them
1233	 */
1234	alloc r2=ar.pfs,8,0,1,0
1235	.prologue
1236	PT_REGS_SAVES(16)
1237	adds sp=-16,sp
1238	.body
1239	cmp.eq pNonSys,pSys=r0,r0		// sigreturn isn't a normal syscall...
1240	;;
1241	/*
1242	 * leave_kernel() restores f6-f11 from pt_regs, but since the streamlined
1243	 * syscall-entry path does not save them we save them here instead.  Note: we
1244	 * don't need to save any other registers that are not saved by the stream-lined
1245	 * syscall path, because restore_sigcontext() restores them.
1246	 */
1247	adds r16=PT(F6)+32,sp
1248	adds r17=PT(F7)+32,sp
1249	;;
1250 	stf.spill [r16]=f6,32
1251 	stf.spill [r17]=f7,32
1252	;;
1253 	stf.spill [r16]=f8,32
1254 	stf.spill [r17]=f9,32
1255	;;
1256 	stf.spill [r16]=f10
1257 	stf.spill [r17]=f11
1258	adds out0=16,sp				// out0 = &sigscratch
1259	br.call.sptk.many rp=ia64_rt_sigreturn
1260.ret19:	.restore sp,0
1261	adds sp=16,sp
1262	;;
1263	ld8 r9=[sp]				// load new ar.unat
1264	mov.sptk b7=r8,ia64_leave_kernel
1265	;;
1266	mov ar.unat=r9
1267	br.many b7
1268END(sys_rt_sigreturn)
1269
1270GLOBAL_ENTRY(ia64_prepare_handle_unaligned)
1271	.prologue
1272	/*
1273	 * r16 = fake ar.pfs, we simply need to make sure privilege is still 0
1274	 */
1275	mov r16=r0
1276	DO_SAVE_SWITCH_STACK
1277	br.call.sptk.many rp=ia64_handle_unaligned	// stack frame setup in ivt
1278.ret21:	.body
1279	DO_LOAD_SWITCH_STACK
1280	br.cond.sptk.many rp				// goes to ia64_leave_kernel
1281END(ia64_prepare_handle_unaligned)
1282
1283	//
1284	// unw_init_running(void (*callback)(info, arg), void *arg)
1285	//
1286#	define EXTRA_FRAME_SIZE	((UNW_FRAME_INFO_SIZE+15)&~15)
1287
1288GLOBAL_ENTRY(unw_init_running)
1289	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1290	alloc loc1=ar.pfs,2,3,3,0
1291	;;
1292	ld8 loc2=[in0],8
1293	mov loc0=rp
1294	mov r16=loc1
1295	DO_SAVE_SWITCH_STACK
1296	.body
1297
1298	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1299	.fframe IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE
1300	SWITCH_STACK_SAVES(EXTRA_FRAME_SIZE)
1301	adds sp=-EXTRA_FRAME_SIZE,sp
1302	.body
1303	;;
1304	adds out0=16,sp				// &info
1305	mov out1=r13				// current
1306	adds out2=16+EXTRA_FRAME_SIZE,sp	// &switch_stack
1307	br.call.sptk.many rp=unw_init_frame_info
13081:	adds out0=16,sp				// &info
1309	mov b6=loc2
1310	mov loc2=gp				// save gp across indirect function call
1311	;;
1312	ld8 gp=[in0]
1313	mov out1=in1				// arg
1314	br.call.sptk.many rp=b6			// invoke the callback function
13151:	mov gp=loc2				// restore gp
1316
1317	// For now, we don't allow changing registers from within
1318	// unw_init_running; if we ever want to allow that, we'd
1319	// have to do a load_switch_stack here:
1320	.restore sp
1321	adds sp=IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE,sp
1322
1323	mov ar.pfs=loc1
1324	mov rp=loc0
1325	br.ret.sptk.many rp
1326END(unw_init_running)
1327EXPORT_SYMBOL(unw_init_running)
1328
1329#ifdef CONFIG_FUNCTION_TRACER
1330#ifdef CONFIG_DYNAMIC_FTRACE
1331GLOBAL_ENTRY(_mcount)
1332	br ftrace_stub
1333END(_mcount)
1334EXPORT_SYMBOL(_mcount)
1335
1336.here:
1337	br.ret.sptk.many b0
1338
1339GLOBAL_ENTRY(ftrace_caller)
1340	alloc out0 = ar.pfs, 8, 0, 4, 0
1341	mov out3 = r0
1342	;;
1343	mov out2 = b0
1344	add r3 = 0x20, r3
1345	mov out1 = r1;
1346	br.call.sptk.many b0 = ftrace_patch_gp
1347	//this might be called from module, so we must patch gp
1348ftrace_patch_gp:
1349	movl gp=__gp
1350	mov b0 = r3
1351	;;
1352.global ftrace_call;
1353ftrace_call:
1354{
1355	.mlx
1356	nop.m 0x0
1357	movl r3 = .here;;
1358}
1359	alloc loc0 = ar.pfs, 4, 4, 2, 0
1360	;;
1361	mov loc1 = b0
1362	mov out0 = b0
1363	mov loc2 = r8
1364	mov loc3 = r15
1365	;;
1366	adds out0 = -MCOUNT_INSN_SIZE, out0
1367	mov out1 = in2
1368	mov b6 = r3
1369
1370	br.call.sptk.many b0 = b6
1371	;;
1372	mov ar.pfs = loc0
1373	mov b0 = loc1
1374	mov r8 = loc2
1375	mov r15 = loc3
1376	br ftrace_stub
1377	;;
1378END(ftrace_caller)
1379
1380#else
1381GLOBAL_ENTRY(_mcount)
1382	movl r2 = ftrace_stub
1383	movl r3 = ftrace_trace_function;;
1384	ld8 r3 = [r3];;
1385	ld8 r3 = [r3];;
1386	cmp.eq p7,p0 = r2, r3
1387(p7)	br.sptk.many ftrace_stub
1388	;;
1389
1390	alloc loc0 = ar.pfs, 4, 4, 2, 0
1391	;;
1392	mov loc1 = b0
1393	mov out0 = b0
1394	mov loc2 = r8
1395	mov loc3 = r15
1396	;;
1397	adds out0 = -MCOUNT_INSN_SIZE, out0
1398	mov out1 = in2
1399	mov b6 = r3
1400
1401	br.call.sptk.many b0 = b6
1402	;;
1403	mov ar.pfs = loc0
1404	mov b0 = loc1
1405	mov r8 = loc2
1406	mov r15 = loc3
1407	br ftrace_stub
1408	;;
1409END(_mcount)
1410#endif
1411
1412GLOBAL_ENTRY(ftrace_stub)
1413	mov r3 = b0
1414	movl r2 = _mcount_ret_helper
1415	;;
1416	mov b6 = r2
1417	mov b7 = r3
1418	br.ret.sptk.many b6
1419
1420_mcount_ret_helper:
1421	mov b0 = r42
1422	mov r1 = r41
1423	mov ar.pfs = r40
1424	br b7
1425END(ftrace_stub)
1426
1427#endif /* CONFIG_FUNCTION_TRACER */
1428
1429#define __SYSCALL(nr, entry, nargs) data8 entry
1430	.rodata
1431	.align 8
1432	.globl sys_call_table
1433sys_call_table:
1434#include <asm/syscall_table.h>
1435#undef __SYSCALL
1436