xref: /openbmc/linux/arch/ia64/kernel/entry.S (revision 8c749ce9)
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 * Copyright (c) 2008 Isaku Yamahata <yamahata at valinux co jp>
27 *                    VA Linux Systems Japan K.K.
28 *                    pv_ops.
29 */
30/*
31 * Global (preserved) predicate usage on syscall entry/exit path:
32 *
33 *	pKStk:		See entry.h.
34 *	pUStk:		See entry.h.
35 *	pSys:		See entry.h.
36 *	pNonSys:	!pSys
37 */
38
39
40#include <asm/asmmacro.h>
41#include <asm/cache.h>
42#include <asm/errno.h>
43#include <asm/kregs.h>
44#include <asm/asm-offsets.h>
45#include <asm/pgtable.h>
46#include <asm/percpu.h>
47#include <asm/processor.h>
48#include <asm/thread_info.h>
49#include <asm/unistd.h>
50#include <asm/ftrace.h>
51
52#include "minstate.h"
53
54	/*
55	 * execve() is special because in case of success, we need to
56	 * setup a null register window frame.
57	 */
58ENTRY(ia64_execve)
59	/*
60	 * Allocate 8 input registers since ptrace() may clobber them
61	 */
62	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
63	alloc loc1=ar.pfs,8,2,3,0
64	mov loc0=rp
65	.body
66	mov out0=in0			// filename
67	;;				// stop bit between alloc and call
68	mov out1=in1			// argv
69	mov out2=in2			// envp
70	br.call.sptk.many rp=sys_execve
71.ret0:
72	cmp4.ge p6,p7=r8,r0
73	mov ar.pfs=loc1			// restore ar.pfs
74	sxt4 r8=r8			// return 64-bit result
75	;;
76	stf.spill [sp]=f0
77	mov rp=loc0
78(p6)	mov ar.pfs=r0			// clear ar.pfs on success
79(p7)	br.ret.sptk.many rp
80
81	/*
82	 * In theory, we'd have to zap this state only to prevent leaking of
83	 * security sensitive state (e.g., if current->mm->dumpable is zero).  However,
84	 * this executes in less than 20 cycles even on Itanium, so it's not worth
85	 * optimizing for...).
86	 */
87	mov ar.unat=0; 		mov ar.lc=0
88	mov r4=0;		mov f2=f0;		mov b1=r0
89	mov r5=0;		mov f3=f0;		mov b2=r0
90	mov r6=0;		mov f4=f0;		mov b3=r0
91	mov r7=0;		mov f5=f0;		mov b4=r0
92	ldf.fill f12=[sp];	mov f13=f0;		mov b5=r0
93	ldf.fill f14=[sp];	ldf.fill f15=[sp];	mov f16=f0
94	ldf.fill f17=[sp];	ldf.fill f18=[sp];	mov f19=f0
95	ldf.fill f20=[sp];	ldf.fill f21=[sp];	mov f22=f0
96	ldf.fill f23=[sp];	ldf.fill f24=[sp];	mov f25=f0
97	ldf.fill f26=[sp];	ldf.fill f27=[sp];	mov f28=f0
98	ldf.fill f29=[sp];	ldf.fill f30=[sp];	mov f31=f0
99	br.ret.sptk.many rp
100END(ia64_execve)
101
102/*
103 * sys_clone2(u64 flags, u64 ustack_base, u64 ustack_size, u64 parent_tidptr, u64 child_tidptr,
104 *	      u64 tls)
105 */
106GLOBAL_ENTRY(sys_clone2)
107	/*
108	 * Allocate 8 input registers since ptrace() may clobber them
109	 */
110	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
111	alloc r16=ar.pfs,8,2,6,0
112	DO_SAVE_SWITCH_STACK
113	adds r2=PT(R16)+IA64_SWITCH_STACK_SIZE+16,sp
114	mov loc0=rp
115	mov loc1=r16				// save ar.pfs across do_fork
116	.body
117	mov out1=in1
118	mov out2=in2
119	tbit.nz p6,p0=in0,CLONE_SETTLS_BIT
120	mov out3=in3	// parent_tidptr: valid only w/CLONE_PARENT_SETTID
121	;;
122(p6)	st8 [r2]=in5				// store TLS in r16 for copy_thread()
123	mov out4=in4	// child_tidptr:  valid only w/CLONE_CHILD_SETTID or CLONE_CHILD_CLEARTID
124	mov out0=in0				// out0 = clone_flags
125	br.call.sptk.many rp=do_fork
126.ret1:	.restore sp
127	adds sp=IA64_SWITCH_STACK_SIZE,sp	// pop the switch stack
128	mov ar.pfs=loc1
129	mov rp=loc0
130	br.ret.sptk.many rp
131END(sys_clone2)
132
133/*
134 * sys_clone(u64 flags, u64 ustack_base, u64 parent_tidptr, u64 child_tidptr, u64 tls)
135 *	Deprecated.  Use sys_clone2() instead.
136 */
137GLOBAL_ENTRY(sys_clone)
138	/*
139	 * Allocate 8 input registers since ptrace() may clobber them
140	 */
141	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
142	alloc r16=ar.pfs,8,2,6,0
143	DO_SAVE_SWITCH_STACK
144	adds r2=PT(R16)+IA64_SWITCH_STACK_SIZE+16,sp
145	mov loc0=rp
146	mov loc1=r16				// save ar.pfs across do_fork
147	.body
148	mov out1=in1
149	mov out2=16				// stacksize (compensates for 16-byte scratch area)
150	tbit.nz p6,p0=in0,CLONE_SETTLS_BIT
151	mov out3=in2	// parent_tidptr: valid only w/CLONE_PARENT_SETTID
152	;;
153(p6)	st8 [r2]=in4				// store TLS in r13 (tp)
154	mov out4=in3	// child_tidptr:  valid only w/CLONE_CHILD_SETTID or CLONE_CHILD_CLEARTID
155	mov out0=in0				// out0 = clone_flags
156	br.call.sptk.many rp=do_fork
157.ret2:	.restore sp
158	adds sp=IA64_SWITCH_STACK_SIZE,sp	// pop the switch stack
159	mov ar.pfs=loc1
160	mov rp=loc0
161	br.ret.sptk.many rp
162END(sys_clone)
163
164/*
165 * prev_task <- ia64_switch_to(struct task_struct *next)
166 *	With Ingo's new scheduler, interrupts are disabled when this routine gets
167 *	called.  The code starting at .map relies on this.  The rest of the code
168 *	doesn't care about the interrupt masking status.
169 */
170GLOBAL_ENTRY(ia64_switch_to)
171	.prologue
172	alloc r16=ar.pfs,1,0,0,0
173	DO_SAVE_SWITCH_STACK
174	.body
175
176	adds r22=IA64_TASK_THREAD_KSP_OFFSET,r13
177	movl r25=init_task
178	mov r27=IA64_KR(CURRENT_STACK)
179	adds r21=IA64_TASK_THREAD_KSP_OFFSET,in0
180	dep r20=0,in0,61,3		// physical address of "next"
181	;;
182	st8 [r22]=sp			// save kernel stack pointer of old task
183	shr.u r26=r20,IA64_GRANULE_SHIFT
184	cmp.eq p7,p6=r25,in0
185	;;
186	/*
187	 * If we've already mapped this task's page, we can skip doing it again.
188	 */
189(p6)	cmp.eq p7,p6=r26,r27
190(p6)	br.cond.dpnt .map
191	;;
192.done:
193	ld8 sp=[r21]			// load kernel stack pointer of new task
194	MOV_TO_KR(CURRENT, in0, r8, r9)		// update "current" application register
195	mov r8=r13			// return pointer to previously running task
196	mov r13=in0			// set "current" pointer
197	;;
198	DO_LOAD_SWITCH_STACK
199
200#ifdef CONFIG_SMP
201	sync.i				// ensure "fc"s done by this CPU are visible on other CPUs
202#endif
203	br.ret.sptk.many rp		// boogie on out in new context
204
205.map:
206	RSM_PSR_IC(r25)			// interrupts (psr.i) are already disabled here
207	movl r25=PAGE_KERNEL
208	;;
209	srlz.d
210	or r23=r25,r20			// construct PA | page properties
211	mov r25=IA64_GRANULE_SHIFT<<2
212	;;
213	MOV_TO_ITIR(p0, r25, r8)
214	MOV_TO_IFA(in0, r8)		// VA of next task...
215	;;
216	mov r25=IA64_TR_CURRENT_STACK
217	MOV_TO_KR(CURRENT_STACK, r26, r8, r9)	// remember last page we mapped...
218	;;
219	itr.d dtr[r25]=r23		// wire in new mapping...
220	SSM_PSR_IC_AND_SRLZ_D(r8, r9)	// reenable the psr.ic bit
221	br.cond.sptk .done
222END(ia64_switch_to)
223
224/*
225 * Note that interrupts are enabled during save_switch_stack and load_switch_stack.  This
226 * means that we may get an interrupt with "sp" pointing to the new kernel stack while
227 * ar.bspstore is still pointing to the old kernel backing store area.  Since ar.rsc,
228 * ar.rnat, ar.bsp, and ar.bspstore are all preserved by interrupts, this is not a
229 * problem.  Also, we don't need to specify unwind information for preserved registers
230 * that are not modified in save_switch_stack as the right unwind information is already
231 * specified at the call-site of save_switch_stack.
232 */
233
234/*
235 * save_switch_stack:
236 *	- r16 holds ar.pfs
237 *	- b7 holds address to return to
238 *	- rp (b0) holds return address to save
239 */
240GLOBAL_ENTRY(save_switch_stack)
241	.prologue
242	.altrp b7
243	flushrs			// flush dirty regs to backing store (must be first in insn group)
244	.save @priunat,r17
245	mov r17=ar.unat		// preserve caller's
246	.body
247#ifdef CONFIG_ITANIUM
248	adds r2=16+128,sp
249	adds r3=16+64,sp
250	adds r14=SW(R4)+16,sp
251	;;
252	st8.spill [r14]=r4,16		// spill r4
253	lfetch.fault.excl.nt1 [r3],128
254	;;
255	lfetch.fault.excl.nt1 [r2],128
256	lfetch.fault.excl.nt1 [r3],128
257	;;
258	lfetch.fault.excl [r2]
259	lfetch.fault.excl [r3]
260	adds r15=SW(R5)+16,sp
261#else
262	add r2=16+3*128,sp
263	add r3=16,sp
264	add r14=SW(R4)+16,sp
265	;;
266	st8.spill [r14]=r4,SW(R6)-SW(R4)	// spill r4 and prefetch offset 0x1c0
267	lfetch.fault.excl.nt1 [r3],128	//		prefetch offset 0x010
268	;;
269	lfetch.fault.excl.nt1 [r3],128	//		prefetch offset 0x090
270	lfetch.fault.excl.nt1 [r2],128	//		prefetch offset 0x190
271	;;
272	lfetch.fault.excl.nt1 [r3]	//		prefetch offset 0x110
273	lfetch.fault.excl.nt1 [r2]	//		prefetch offset 0x210
274	adds r15=SW(R5)+16,sp
275#endif
276	;;
277	st8.spill [r15]=r5,SW(R7)-SW(R5)	// spill r5
278	mov.m ar.rsc=0			// put RSE in mode: enforced lazy, little endian, pl 0
279	add r2=SW(F2)+16,sp		// r2 = &sw->f2
280	;;
281	st8.spill [r14]=r6,SW(B0)-SW(R6)	// spill r6
282	mov.m r18=ar.fpsr		// preserve fpsr
283	add r3=SW(F3)+16,sp		// r3 = &sw->f3
284	;;
285	stf.spill [r2]=f2,32
286	mov.m r19=ar.rnat
287	mov r21=b0
288
289	stf.spill [r3]=f3,32
290	st8.spill [r15]=r7,SW(B2)-SW(R7)	// spill r7
291	mov r22=b1
292	;;
293	// since we're done with the spills, read and save ar.unat:
294	mov.m r29=ar.unat
295	mov.m r20=ar.bspstore
296	mov r23=b2
297	stf.spill [r2]=f4,32
298	stf.spill [r3]=f5,32
299	mov r24=b3
300	;;
301	st8 [r14]=r21,SW(B1)-SW(B0)		// save b0
302	st8 [r15]=r23,SW(B3)-SW(B2)		// save b2
303	mov r25=b4
304	mov r26=b5
305	;;
306	st8 [r14]=r22,SW(B4)-SW(B1)		// save b1
307	st8 [r15]=r24,SW(AR_PFS)-SW(B3)		// save b3
308	mov r21=ar.lc		// I-unit
309	stf.spill [r2]=f12,32
310	stf.spill [r3]=f13,32
311	;;
312	st8 [r14]=r25,SW(B5)-SW(B4)		// save b4
313	st8 [r15]=r16,SW(AR_LC)-SW(AR_PFS)	// save ar.pfs
314	stf.spill [r2]=f14,32
315	stf.spill [r3]=f15,32
316	;;
317	st8 [r14]=r26				// save b5
318	st8 [r15]=r21				// save ar.lc
319	stf.spill [r2]=f16,32
320	stf.spill [r3]=f17,32
321	;;
322	stf.spill [r2]=f18,32
323	stf.spill [r3]=f19,32
324	;;
325	stf.spill [r2]=f20,32
326	stf.spill [r3]=f21,32
327	;;
328	stf.spill [r2]=f22,32
329	stf.spill [r3]=f23,32
330	;;
331	stf.spill [r2]=f24,32
332	stf.spill [r3]=f25,32
333	;;
334	stf.spill [r2]=f26,32
335	stf.spill [r3]=f27,32
336	;;
337	stf.spill [r2]=f28,32
338	stf.spill [r3]=f29,32
339	;;
340	stf.spill [r2]=f30,SW(AR_UNAT)-SW(F30)
341	stf.spill [r3]=f31,SW(PR)-SW(F31)
342	add r14=SW(CALLER_UNAT)+16,sp
343	;;
344	st8 [r2]=r29,SW(AR_RNAT)-SW(AR_UNAT)	// save ar.unat
345	st8 [r14]=r17,SW(AR_FPSR)-SW(CALLER_UNAT) // save caller_unat
346	mov r21=pr
347	;;
348	st8 [r2]=r19,SW(AR_BSPSTORE)-SW(AR_RNAT) // save ar.rnat
349	st8 [r3]=r21				// save predicate registers
350	;;
351	st8 [r2]=r20				// save ar.bspstore
352	st8 [r14]=r18				// save fpsr
353	mov ar.rsc=3		// put RSE back into eager mode, pl 0
354	br.cond.sptk.many b7
355END(save_switch_stack)
356
357/*
358 * load_switch_stack:
359 *	- "invala" MUST be done at call site (normally in DO_LOAD_SWITCH_STACK)
360 *	- b7 holds address to return to
361 *	- must not touch r8-r11
362 */
363GLOBAL_ENTRY(load_switch_stack)
364	.prologue
365	.altrp b7
366
367	.body
368	lfetch.fault.nt1 [sp]
369	adds r2=SW(AR_BSPSTORE)+16,sp
370	adds r3=SW(AR_UNAT)+16,sp
371	mov ar.rsc=0						// put RSE into enforced lazy mode
372	adds r14=SW(CALLER_UNAT)+16,sp
373	adds r15=SW(AR_FPSR)+16,sp
374	;;
375	ld8 r27=[r2],(SW(B0)-SW(AR_BSPSTORE))	// bspstore
376	ld8 r29=[r3],(SW(B1)-SW(AR_UNAT))	// unat
377	;;
378	ld8 r21=[r2],16		// restore b0
379	ld8 r22=[r3],16		// restore b1
380	;;
381	ld8 r23=[r2],16		// restore b2
382	ld8 r24=[r3],16		// restore b3
383	;;
384	ld8 r25=[r2],16		// restore b4
385	ld8 r26=[r3],16		// restore b5
386	;;
387	ld8 r16=[r2],(SW(PR)-SW(AR_PFS))	// ar.pfs
388	ld8 r17=[r3],(SW(AR_RNAT)-SW(AR_LC))	// ar.lc
389	;;
390	ld8 r28=[r2]		// restore pr
391	ld8 r30=[r3]		// restore rnat
392	;;
393	ld8 r18=[r14],16	// restore caller's unat
394	ld8 r19=[r15],24	// restore fpsr
395	;;
396	ldf.fill f2=[r14],32
397	ldf.fill f3=[r15],32
398	;;
399	ldf.fill f4=[r14],32
400	ldf.fill f5=[r15],32
401	;;
402	ldf.fill f12=[r14],32
403	ldf.fill f13=[r15],32
404	;;
405	ldf.fill f14=[r14],32
406	ldf.fill f15=[r15],32
407	;;
408	ldf.fill f16=[r14],32
409	ldf.fill f17=[r15],32
410	;;
411	ldf.fill f18=[r14],32
412	ldf.fill f19=[r15],32
413	mov b0=r21
414	;;
415	ldf.fill f20=[r14],32
416	ldf.fill f21=[r15],32
417	mov b1=r22
418	;;
419	ldf.fill f22=[r14],32
420	ldf.fill f23=[r15],32
421	mov b2=r23
422	;;
423	mov ar.bspstore=r27
424	mov ar.unat=r29		// establish unat holding the NaT bits for r4-r7
425	mov b3=r24
426	;;
427	ldf.fill f24=[r14],32
428	ldf.fill f25=[r15],32
429	mov b4=r25
430	;;
431	ldf.fill f26=[r14],32
432	ldf.fill f27=[r15],32
433	mov b5=r26
434	;;
435	ldf.fill f28=[r14],32
436	ldf.fill f29=[r15],32
437	mov ar.pfs=r16
438	;;
439	ldf.fill f30=[r14],32
440	ldf.fill f31=[r15],24
441	mov ar.lc=r17
442	;;
443	ld8.fill r4=[r14],16
444	ld8.fill r5=[r15],16
445	mov pr=r28,-1
446	;;
447	ld8.fill r6=[r14],16
448	ld8.fill r7=[r15],16
449
450	mov ar.unat=r18				// restore caller's unat
451	mov ar.rnat=r30				// must restore after bspstore but before rsc!
452	mov ar.fpsr=r19				// restore fpsr
453	mov ar.rsc=3				// put RSE back into eager mode, pl 0
454	br.cond.sptk.many b7
455END(load_switch_stack)
456
457GLOBAL_ENTRY(prefetch_stack)
458	add r14 = -IA64_SWITCH_STACK_SIZE, sp
459	add r15 = IA64_TASK_THREAD_KSP_OFFSET, in0
460	;;
461	ld8 r16 = [r15]				// load next's stack pointer
462	lfetch.fault.excl [r14], 128
463	;;
464	lfetch.fault.excl [r14], 128
465	lfetch.fault [r16], 128
466	;;
467	lfetch.fault.excl [r14], 128
468	lfetch.fault [r16], 128
469	;;
470	lfetch.fault.excl [r14], 128
471	lfetch.fault [r16], 128
472	;;
473	lfetch.fault.excl [r14], 128
474	lfetch.fault [r16], 128
475	;;
476	lfetch.fault [r16], 128
477	br.ret.sptk.many rp
478END(prefetch_stack)
479
480	/*
481	 * Invoke a system call, but do some tracing before and after the call.
482	 * We MUST preserve the current register frame throughout this routine
483	 * because some system calls (such as ia64_execve) directly
484	 * manipulate ar.pfs.
485	 */
486GLOBAL_ENTRY(ia64_trace_syscall)
487	PT_REGS_UNWIND_INFO(0)
488	/*
489	 * We need to preserve the scratch registers f6-f11 in case the system
490	 * call is sigreturn.
491	 */
492	adds r16=PT(F6)+16,sp
493	adds r17=PT(F7)+16,sp
494	;;
495 	stf.spill [r16]=f6,32
496 	stf.spill [r17]=f7,32
497	;;
498 	stf.spill [r16]=f8,32
499 	stf.spill [r17]=f9,32
500	;;
501 	stf.spill [r16]=f10
502 	stf.spill [r17]=f11
503	br.call.sptk.many rp=syscall_trace_enter // give parent a chance to catch syscall args
504	cmp.lt p6,p0=r8,r0			// check tracehook
505	adds r2=PT(R8)+16,sp			// r2 = &pt_regs.r8
506	adds r3=PT(R10)+16,sp			// r3 = &pt_regs.r10
507	mov r10=0
508(p6)	br.cond.sptk strace_error		// syscall failed ->
509	adds r16=PT(F6)+16,sp
510	adds r17=PT(F7)+16,sp
511	;;
512	ldf.fill f6=[r16],32
513	ldf.fill f7=[r17],32
514	;;
515	ldf.fill f8=[r16],32
516	ldf.fill f9=[r17],32
517	;;
518	ldf.fill f10=[r16]
519	ldf.fill f11=[r17]
520	// the syscall number may have changed, so re-load it and re-calculate the
521	// syscall entry-point:
522	adds r15=PT(R15)+16,sp			// r15 = &pt_regs.r15 (syscall #)
523	;;
524	ld8 r15=[r15]
525	mov r3=NR_syscalls - 1
526	;;
527	adds r15=-1024,r15
528	movl r16=sys_call_table
529	;;
530	shladd r20=r15,3,r16			// r20 = sys_call_table + 8*(syscall-1024)
531	cmp.leu p6,p7=r15,r3
532	;;
533(p6)	ld8 r20=[r20]				// load address of syscall entry point
534(p7)	movl r20=sys_ni_syscall
535	;;
536	mov b6=r20
537	br.call.sptk.many rp=b6			// do the syscall
538.strace_check_retval:
539	cmp.lt p6,p0=r8,r0			// syscall failed?
540	adds r2=PT(R8)+16,sp			// r2 = &pt_regs.r8
541	adds r3=PT(R10)+16,sp			// r3 = &pt_regs.r10
542	mov r10=0
543(p6)	br.cond.sptk strace_error		// syscall failed ->
544	;;					// avoid RAW on r10
545.strace_save_retval:
546.mem.offset 0,0; st8.spill [r2]=r8		// store return value in slot for r8
547.mem.offset 8,0; st8.spill [r3]=r10		// clear error indication in slot for r10
548	br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
549.ret3:
550(pUStk)	cmp.eq.unc p6,p0=r0,r0			// p6 <- pUStk
551(pUStk)	rsm psr.i				// disable interrupts
552	br.cond.sptk ia64_work_pending_syscall_end
553
554strace_error:
555	ld8 r3=[r2]				// load pt_regs.r8
556	sub r9=0,r8				// negate return value to get errno value
557	;;
558	cmp.ne p6,p0=r3,r0			// is pt_regs.r8!=0?
559	adds r3=16,r2				// r3=&pt_regs.r10
560	;;
561(p6)	mov r10=-1
562(p6)	mov r8=r9
563	br.cond.sptk .strace_save_retval
564END(ia64_trace_syscall)
565
566	/*
567	 * When traced and returning from sigreturn, we invoke syscall_trace but then
568	 * go straight to ia64_leave_kernel rather than ia64_leave_syscall.
569	 */
570GLOBAL_ENTRY(ia64_strace_leave_kernel)
571	PT_REGS_UNWIND_INFO(0)
572{	/*
573	 * Some versions of gas generate bad unwind info if the first instruction of a
574	 * procedure doesn't go into the first slot of a bundle.  This is a workaround.
575	 */
576	nop.m 0
577	nop.i 0
578	br.call.sptk.many rp=syscall_trace_leave // give parent a chance to catch return value
579}
580.ret4:	br.cond.sptk ia64_leave_kernel
581END(ia64_strace_leave_kernel)
582
583ENTRY(call_payload)
584	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(0)
585	/* call the kernel_thread payload; fn is in r4, arg - in r5 */
586	alloc loc1=ar.pfs,0,3,1,0
587	mov loc0=rp
588	mov loc2=gp
589	mov out0=r5		// arg
590	ld8 r14 = [r4], 8	// fn.address
591	;;
592	mov b6 = r14
593	ld8 gp = [r4]		// fn.gp
594	;;
595	br.call.sptk.many rp=b6	// fn(arg)
596.ret12:	mov gp=loc2
597	mov rp=loc0
598	mov ar.pfs=loc1
599	/* ... and if it has returned, we are going to userland */
600	cmp.ne pKStk,pUStk=r0,r0
601	br.ret.sptk.many rp
602END(call_payload)
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(pKStk)	br.call.sptk.many rp=call_payload
621	adds r2=TI_FLAGS+IA64_TASK_SIZE,r13
622	;;
623	ld4 r2=[r2]
624	;;
625	mov r8=0
626	and r2=_TIF_SYSCALL_TRACEAUDIT,r2
627	;;
628	cmp.ne p6,p0=r2,r0
629(p6)	br.cond.spnt .strace_check_retval
630	;;					// added stop bits to prevent r8 dependency
631END(ia64_ret_from_clone)
632	// fall through
633GLOBAL_ENTRY(ia64_ret_from_syscall)
634	PT_REGS_UNWIND_INFO(0)
635	cmp.ge p6,p7=r8,r0			// syscall executed successfully?
636	adds r2=PT(R8)+16,sp			// r2 = &pt_regs.r8
637	mov r10=r0				// clear error indication in r10
638(p7)	br.cond.spnt handle_syscall_error	// handle potential syscall failure
639END(ia64_ret_from_syscall)
640	// fall through
641
642/*
643 * ia64_leave_syscall(): Same as ia64_leave_kernel, except that it doesn't
644 *	need to switch to bank 0 and doesn't restore the scratch registers.
645 *	To avoid leaking kernel bits, the scratch registers are set to
646 *	the following known-to-be-safe values:
647 *
648 *		  r1: restored (global pointer)
649 *		  r2: cleared
650 *		  r3: 1 (when returning to user-level)
651 *	      r8-r11: restored (syscall return value(s))
652 *		 r12: restored (user-level stack pointer)
653 *		 r13: restored (user-level thread pointer)
654 *		 r14: set to __kernel_syscall_via_epc
655 *		 r15: restored (syscall #)
656 *	     r16-r17: cleared
657 *		 r18: user-level b6
658 *		 r19: cleared
659 *		 r20: user-level ar.fpsr
660 *		 r21: user-level b0
661 *		 r22: cleared
662 *		 r23: user-level ar.bspstore
663 *		 r24: user-level ar.rnat
664 *		 r25: user-level ar.unat
665 *		 r26: user-level ar.pfs
666 *		 r27: user-level ar.rsc
667 *		 r28: user-level ip
668 *		 r29: user-level psr
669 *		 r30: user-level cfm
670 *		 r31: user-level pr
671 *	      f6-f11: cleared
672 *		  pr: restored (user-level pr)
673 *		  b0: restored (user-level rp)
674 *	          b6: restored
675 *		  b7: set to __kernel_syscall_via_epc
676 *	     ar.unat: restored (user-level ar.unat)
677 *	      ar.pfs: restored (user-level ar.pfs)
678 *	      ar.rsc: restored (user-level ar.rsc)
679 *	     ar.rnat: restored (user-level ar.rnat)
680 *	 ar.bspstore: restored (user-level ar.bspstore)
681 *	     ar.fpsr: restored (user-level ar.fpsr)
682 *	      ar.ccv: cleared
683 *	      ar.csd: cleared
684 *	      ar.ssd: cleared
685 */
686GLOBAL_ENTRY(ia64_leave_syscall)
687	PT_REGS_UNWIND_INFO(0)
688	/*
689	 * work.need_resched etc. mustn't get changed by this CPU before it returns to
690	 * user- or fsys-mode, hence we disable interrupts early on.
691	 *
692	 * p6 controls whether current_thread_info()->flags needs to be check for
693	 * extra work.  We always check for extra work when returning to user-level.
694	 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
695	 * is 0.  After extra work processing has been completed, execution
696	 * resumes at ia64_work_processed_syscall with p6 set to 1 if the extra-work-check
697	 * needs to be redone.
698	 */
699#ifdef CONFIG_PREEMPT
700	RSM_PSR_I(p0, r2, r18)			// disable interrupts
701	cmp.eq pLvSys,p0=r0,r0			// pLvSys=1: leave from syscall
702(pKStk) adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
703	;;
704	.pred.rel.mutex pUStk,pKStk
705(pKStk) ld4 r21=[r20]			// r21 <- preempt_count
706(pUStk)	mov r21=0			// r21 <- 0
707	;;
708	cmp.eq p6,p0=r21,r0		// p6 <- pUStk || (preempt_count == 0)
709#else /* !CONFIG_PREEMPT */
710	RSM_PSR_I(pUStk, r2, r18)
711	cmp.eq pLvSys,p0=r0,r0		// pLvSys=1: leave from syscall
712(pUStk)	cmp.eq.unc p6,p0=r0,r0		// p6 <- pUStk
713#endif
714.global ia64_work_processed_syscall;
715ia64_work_processed_syscall:
716#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
717	adds r2=PT(LOADRS)+16,r12
718	MOV_FROM_ITC(pUStk, p9, r22, r19)	// fetch time at leave
719	adds r18=TI_FLAGS+IA64_TASK_SIZE,r13
720	;;
721(p6)	ld4 r31=[r18]				// load current_thread_info()->flags
722	ld8 r19=[r2],PT(B6)-PT(LOADRS)		// load ar.rsc value for "loadrs"
723	adds r3=PT(AR_BSPSTORE)+16,r12		// deferred
724	;;
725#else
726	adds r2=PT(LOADRS)+16,r12
727	adds r3=PT(AR_BSPSTORE)+16,r12
728	adds r18=TI_FLAGS+IA64_TASK_SIZE,r13
729	;;
730(p6)	ld4 r31=[r18]				// load current_thread_info()->flags
731	ld8 r19=[r2],PT(B6)-PT(LOADRS)		// load ar.rsc value for "loadrs"
732	nop.i 0
733	;;
734#endif
735	mov r16=ar.bsp				// M2  get existing backing store pointer
736	ld8 r18=[r2],PT(R9)-PT(B6)		// load b6
737(p6)	and r15=TIF_WORK_MASK,r31		// any work other than TIF_SYSCALL_TRACE?
738	;;
739	ld8 r23=[r3],PT(R11)-PT(AR_BSPSTORE)	// load ar.bspstore (may be garbage)
740(p6)	cmp4.ne.unc p6,p0=r15, r0		// any special work pending?
741(p6)	br.cond.spnt .work_pending_syscall
742	;;
743	// start restoring the state saved on the kernel stack (struct pt_regs):
744	ld8 r9=[r2],PT(CR_IPSR)-PT(R9)
745	ld8 r11=[r3],PT(CR_IIP)-PT(R11)
746(pNonSys) break 0		//      bug check: we shouldn't be here if pNonSys is TRUE!
747	;;
748	invala			// M0|1 invalidate ALAT
749	RSM_PSR_I_IC(r28, r29, r30)	// M2   turn off interrupts and interruption collection
750	cmp.eq p9,p0=r0,r0	// A    set p9 to indicate that we should restore cr.ifs
751
752	ld8 r29=[r2],16		// M0|1 load cr.ipsr
753	ld8 r28=[r3],16		// M0|1 load cr.iip
754#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
755(pUStk) add r14=TI_AC_LEAVE+IA64_TASK_SIZE,r13
756	;;
757	ld8 r30=[r2],16		// M0|1 load cr.ifs
758	ld8 r25=[r3],16		// M0|1 load ar.unat
759(pUStk) add r15=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13
760	;;
761#else
762	mov r22=r0		// A    clear r22
763	;;
764	ld8 r30=[r2],16		// M0|1 load cr.ifs
765	ld8 r25=[r3],16		// M0|1 load ar.unat
766(pUStk) add r14=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13
767	;;
768#endif
769	ld8 r26=[r2],PT(B0)-PT(AR_PFS)	// M0|1 load ar.pfs
770	MOV_FROM_PSR(pKStk, r22, r21)	// M2   read PSR now that interrupts are disabled
771	nop 0
772	;;
773	ld8 r21=[r2],PT(AR_RNAT)-PT(B0) // M0|1 load b0
774	ld8 r27=[r3],PT(PR)-PT(AR_RSC)	// M0|1 load ar.rsc
775	mov f6=f0			// F    clear f6
776	;;
777	ld8 r24=[r2],PT(AR_FPSR)-PT(AR_RNAT)	// M0|1 load ar.rnat (may be garbage)
778	ld8 r31=[r3],PT(R1)-PT(PR)		// M0|1 load predicates
779	mov f7=f0				// F    clear f7
780	;;
781	ld8 r20=[r2],PT(R12)-PT(AR_FPSR)	// M0|1 load ar.fpsr
782	ld8.fill r1=[r3],16			// M0|1 load r1
783(pUStk) mov r17=1				// A
784	;;
785#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
786(pUStk) st1 [r15]=r17				// M2|3
787#else
788(pUStk) st1 [r14]=r17				// M2|3
789#endif
790	ld8.fill r13=[r3],16			// M0|1
791	mov f8=f0				// F    clear f8
792	;;
793	ld8.fill r12=[r2]			// M0|1 restore r12 (sp)
794	ld8.fill r15=[r3]			// M0|1 restore r15
795	mov b6=r18				// I0   restore b6
796
797	LOAD_PHYS_STACK_REG_SIZE(r17)
798	mov f9=f0					// F    clear f9
799(pKStk) br.cond.dpnt.many skip_rbs_switch		// B
800
801	srlz.d				// M0   ensure interruption collection is off (for cover)
802	shr.u r18=r19,16		// I0|1 get byte size of existing "dirty" partition
803	COVER				// B    add current frame into dirty partition & set cr.ifs
804	;;
805#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
806	mov r19=ar.bsp			// M2   get new backing store pointer
807	st8 [r14]=r22			// M	save time at leave
808	mov f10=f0			// F    clear f10
809
810	mov r22=r0			// A	clear r22
811	movl r14=__kernel_syscall_via_epc // X
812	;;
813#else
814	mov r19=ar.bsp			// M2   get new backing store pointer
815	mov f10=f0			// F    clear f10
816
817	nop.m 0
818	movl r14=__kernel_syscall_via_epc // X
819	;;
820#endif
821	mov.m ar.csd=r0			// M2   clear ar.csd
822	mov.m ar.ccv=r0			// M2   clear ar.ccv
823	mov b7=r14			// I0   clear b7 (hint with __kernel_syscall_via_epc)
824
825	mov.m ar.ssd=r0			// M2   clear ar.ssd
826	mov f11=f0			// F    clear f11
827	br.cond.sptk.many rbs_switch	// B
828END(ia64_leave_syscall)
829
830GLOBAL_ENTRY(ia64_leave_kernel)
831	PT_REGS_UNWIND_INFO(0)
832	/*
833	 * work.need_resched etc. mustn't get changed by this CPU before it returns to
834	 * user- or fsys-mode, hence we disable interrupts early on.
835	 *
836	 * p6 controls whether current_thread_info()->flags needs to be check for
837	 * extra work.  We always check for extra work when returning to user-level.
838	 * With CONFIG_PREEMPT, we also check for extra work when the preempt_count
839	 * is 0.  After extra work processing has been completed, execution
840	 * resumes at .work_processed_syscall with p6 set to 1 if the extra-work-check
841	 * needs to be redone.
842	 */
843#ifdef CONFIG_PREEMPT
844	RSM_PSR_I(p0, r17, r31)			// disable interrupts
845	cmp.eq p0,pLvSys=r0,r0			// pLvSys=0: leave from kernel
846(pKStk)	adds r20=TI_PRE_COUNT+IA64_TASK_SIZE,r13
847	;;
848	.pred.rel.mutex pUStk,pKStk
849(pKStk)	ld4 r21=[r20]			// r21 <- preempt_count
850(pUStk)	mov r21=0			// r21 <- 0
851	;;
852	cmp.eq p6,p0=r21,r0		// p6 <- pUStk || (preempt_count == 0)
853#else
854	RSM_PSR_I(pUStk, r17, r31)
855	cmp.eq p0,pLvSys=r0,r0		// pLvSys=0: leave from kernel
856(pUStk)	cmp.eq.unc p6,p0=r0,r0		// p6 <- pUStk
857#endif
858.work_processed_kernel:
859	adds r17=TI_FLAGS+IA64_TASK_SIZE,r13
860	;;
861(p6)	ld4 r31=[r17]				// load current_thread_info()->flags
862	adds r21=PT(PR)+16,r12
863	;;
864
865	lfetch [r21],PT(CR_IPSR)-PT(PR)
866	adds r2=PT(B6)+16,r12
867	adds r3=PT(R16)+16,r12
868	;;
869	lfetch [r21]
870	ld8 r28=[r2],8		// load b6
871	adds r29=PT(R24)+16,r12
872
873	ld8.fill r16=[r3],PT(AR_CSD)-PT(R16)
874	adds r30=PT(AR_CCV)+16,r12
875(p6)	and r19=TIF_WORK_MASK,r31		// any work other than TIF_SYSCALL_TRACE?
876	;;
877	ld8.fill r24=[r29]
878	ld8 r15=[r30]		// load ar.ccv
879(p6)	cmp4.ne.unc p6,p0=r19, r0		// any special work pending?
880	;;
881	ld8 r29=[r2],16		// load b7
882	ld8 r30=[r3],16		// load ar.csd
883(p6)	br.cond.spnt .work_pending
884	;;
885	ld8 r31=[r2],16		// load ar.ssd
886	ld8.fill r8=[r3],16
887	;;
888	ld8.fill r9=[r2],16
889	ld8.fill r10=[r3],PT(R17)-PT(R10)
890	;;
891	ld8.fill r11=[r2],PT(R18)-PT(R11)
892	ld8.fill r17=[r3],16
893	;;
894	ld8.fill r18=[r2],16
895	ld8.fill r19=[r3],16
896	;;
897	ld8.fill r20=[r2],16
898	ld8.fill r21=[r3],16
899	mov ar.csd=r30
900	mov ar.ssd=r31
901	;;
902	RSM_PSR_I_IC(r23, r22, r25)	// initiate turning off of interrupt and interruption collection
903	invala			// invalidate ALAT
904	;;
905	ld8.fill r22=[r2],24
906	ld8.fill r23=[r3],24
907	mov b6=r28
908	;;
909	ld8.fill r25=[r2],16
910	ld8.fill r26=[r3],16
911	mov b7=r29
912	;;
913	ld8.fill r27=[r2],16
914	ld8.fill r28=[r3],16
915	;;
916	ld8.fill r29=[r2],16
917	ld8.fill r30=[r3],24
918	;;
919	ld8.fill r31=[r2],PT(F9)-PT(R31)
920	adds r3=PT(F10)-PT(F6),r3
921	;;
922	ldf.fill f9=[r2],PT(F6)-PT(F9)
923	ldf.fill f10=[r3],PT(F8)-PT(F10)
924	;;
925	ldf.fill f6=[r2],PT(F7)-PT(F6)
926	;;
927	ldf.fill f7=[r2],PT(F11)-PT(F7)
928	ldf.fill f8=[r3],32
929	;;
930	srlz.d	// ensure that inter. collection is off (VHPT is don't care, since text is pinned)
931	mov ar.ccv=r15
932	;;
933	ldf.fill f11=[r2]
934	BSW_0(r2, r3, r15)	// switch back to bank 0 (no stop bit required beforehand...)
935	;;
936(pUStk)	mov r18=IA64_KR(CURRENT)// M2 (12 cycle read latency)
937	adds r16=PT(CR_IPSR)+16,r12
938	adds r17=PT(CR_IIP)+16,r12
939
940#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
941	.pred.rel.mutex pUStk,pKStk
942	MOV_FROM_PSR(pKStk, r22, r29)	// M2 read PSR now that interrupts are disabled
943	MOV_FROM_ITC(pUStk, p9, r22, r29)	// M  fetch time at leave
944	nop.i 0
945	;;
946#else
947	MOV_FROM_PSR(pKStk, r22, r29)	// M2 read PSR now that interrupts are disabled
948	nop.i 0
949	nop.i 0
950	;;
951#endif
952	ld8 r29=[r16],16	// load cr.ipsr
953	ld8 r28=[r17],16	// load cr.iip
954	;;
955	ld8 r30=[r16],16	// load cr.ifs
956	ld8 r25=[r17],16	// load ar.unat
957	;;
958	ld8 r26=[r16],16	// load ar.pfs
959	ld8 r27=[r17],16	// load ar.rsc
960	cmp.eq p9,p0=r0,r0	// set p9 to indicate that we should restore cr.ifs
961	;;
962	ld8 r24=[r16],16	// load ar.rnat (may be garbage)
963	ld8 r23=[r17],16	// load ar.bspstore (may be garbage)
964	;;
965	ld8 r31=[r16],16	// load predicates
966	ld8 r21=[r17],16	// load b0
967	;;
968	ld8 r19=[r16],16	// load ar.rsc value for "loadrs"
969	ld8.fill r1=[r17],16	// load r1
970	;;
971	ld8.fill r12=[r16],16
972	ld8.fill r13=[r17],16
973#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
974(pUStk)	adds r3=TI_AC_LEAVE+IA64_TASK_SIZE,r18
975#else
976(pUStk)	adds r18=IA64_TASK_THREAD_ON_USTACK_OFFSET,r18
977#endif
978	;;
979	ld8 r20=[r16],16	// ar.fpsr
980	ld8.fill r15=[r17],16
981#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
982(pUStk)	adds r18=IA64_TASK_THREAD_ON_USTACK_OFFSET,r18	// deferred
983#endif
984	;;
985	ld8.fill r14=[r16],16
986	ld8.fill r2=[r17]
987(pUStk)	mov r17=1
988	;;
989#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
990	//  mmi_ :  ld8 st1 shr;;         mmi_ : st8 st1 shr;;
991	//  mib  :  mov add br        ->  mib  : ld8 add br
992	//  bbb_ :  br  nop cover;;       mbb_ : mov br  cover;;
993	//
994	//  no one require bsp in r16 if (pKStk) branch is selected.
995(pUStk)	st8 [r3]=r22		// save time at leave
996(pUStk)	st1 [r18]=r17		// restore current->thread.on_ustack
997	shr.u r18=r19,16	// get byte size of existing "dirty" partition
998	;;
999	ld8.fill r3=[r16]	// deferred
1000	LOAD_PHYS_STACK_REG_SIZE(r17)
1001(pKStk)	br.cond.dpnt skip_rbs_switch
1002	mov r16=ar.bsp		// get existing backing store pointer
1003#else
1004	ld8.fill r3=[r16]
1005(pUStk)	st1 [r18]=r17		// restore current->thread.on_ustack
1006	shr.u r18=r19,16	// get byte size of existing "dirty" partition
1007	;;
1008	mov r16=ar.bsp		// get existing backing store pointer
1009	LOAD_PHYS_STACK_REG_SIZE(r17)
1010(pKStk)	br.cond.dpnt skip_rbs_switch
1011#endif
1012
1013	/*
1014	 * Restore user backing store.
1015	 *
1016	 * NOTE: alloc, loadrs, and cover can't be predicated.
1017	 */
1018(pNonSys) br.cond.dpnt dont_preserve_current_frame
1019	COVER				// add current frame into dirty partition and set cr.ifs
1020	;;
1021	mov r19=ar.bsp			// get new backing store pointer
1022rbs_switch:
1023	sub r16=r16,r18			// krbs = old bsp - size of dirty partition
1024	cmp.ne p9,p0=r0,r0		// clear p9 to skip restore of cr.ifs
1025	;;
1026	sub r19=r19,r16			// calculate total byte size of dirty partition
1027	add r18=64,r18			// don't force in0-in7 into memory...
1028	;;
1029	shl r19=r19,16			// shift size of dirty partition into loadrs position
1030	;;
1031dont_preserve_current_frame:
1032	/*
1033	 * To prevent leaking bits between the kernel and user-space,
1034	 * we must clear the stacked registers in the "invalid" partition here.
1035	 * Not pretty, but at least it's fast (3.34 registers/cycle on Itanium,
1036	 * 5 registers/cycle on McKinley).
1037	 */
1038#	define pRecurse	p6
1039#	define pReturn	p7
1040#ifdef CONFIG_ITANIUM
1041#	define Nregs	10
1042#else
1043#	define Nregs	14
1044#endif
1045	alloc loc0=ar.pfs,2,Nregs-2,2,0
1046	shr.u loc1=r18,9		// RNaTslots <= floor(dirtySize / (64*8))
1047	sub r17=r17,r18			// r17 = (physStackedSize + 8) - dirtySize
1048	;;
1049	mov ar.rsc=r19			// load ar.rsc to be used for "loadrs"
1050	shladd in0=loc1,3,r17
1051	mov in1=0
1052	;;
1053	TEXT_ALIGN(32)
1054rse_clear_invalid:
1055#ifdef CONFIG_ITANIUM
1056	// cycle 0
1057 { .mii
1058	alloc loc0=ar.pfs,2,Nregs-2,2,0
1059	cmp.lt pRecurse,p0=Nregs*8,in0	// if more than Nregs regs left to clear, (re)curse
1060	add out0=-Nregs*8,in0
1061}{ .mfb
1062	add out1=1,in1			// increment recursion count
1063	nop.f 0
1064	nop.b 0				// can't do br.call here because of alloc (WAW on CFM)
1065	;;
1066}{ .mfi	// cycle 1
1067	mov loc1=0
1068	nop.f 0
1069	mov loc2=0
1070}{ .mib
1071	mov loc3=0
1072	mov loc4=0
1073(pRecurse) br.call.sptk.many b0=rse_clear_invalid
1074
1075}{ .mfi	// cycle 2
1076	mov loc5=0
1077	nop.f 0
1078	cmp.ne pReturn,p0=r0,in1	// if recursion count != 0, we need to do a br.ret
1079}{ .mib
1080	mov loc6=0
1081	mov loc7=0
1082(pReturn) br.ret.sptk.many b0
1083}
1084#else /* !CONFIG_ITANIUM */
1085	alloc loc0=ar.pfs,2,Nregs-2,2,0
1086	cmp.lt pRecurse,p0=Nregs*8,in0	// if more than Nregs regs left to clear, (re)curse
1087	add out0=-Nregs*8,in0
1088	add out1=1,in1			// increment recursion count
1089	mov loc1=0
1090	mov loc2=0
1091	;;
1092	mov loc3=0
1093	mov loc4=0
1094	mov loc5=0
1095	mov loc6=0
1096	mov loc7=0
1097(pRecurse) br.call.dptk.few b0=rse_clear_invalid
1098	;;
1099	mov loc8=0
1100	mov loc9=0
1101	cmp.ne pReturn,p0=r0,in1	// if recursion count != 0, we need to do a br.ret
1102	mov loc10=0
1103	mov loc11=0
1104(pReturn) br.ret.dptk.many b0
1105#endif /* !CONFIG_ITANIUM */
1106#	undef pRecurse
1107#	undef pReturn
1108	;;
1109	alloc r17=ar.pfs,0,0,0,0	// drop current register frame
1110	;;
1111	loadrs
1112	;;
1113skip_rbs_switch:
1114	mov ar.unat=r25		// M2
1115(pKStk)	extr.u r22=r22,21,1	// I0 extract current value of psr.pp from r22
1116(pLvSys)mov r19=r0		// A  clear r19 for leave_syscall, no-op otherwise
1117	;;
1118(pUStk)	mov ar.bspstore=r23	// M2
1119(pKStk)	dep r29=r22,r29,21,1	// I0 update ipsr.pp with psr.pp
1120(pLvSys)mov r16=r0		// A  clear r16 for leave_syscall, no-op otherwise
1121	;;
1122	MOV_TO_IPSR(p0, r29, r25)	// M2
1123	mov ar.pfs=r26		// I0
1124(pLvSys)mov r17=r0		// A  clear r17 for leave_syscall, no-op otherwise
1125
1126	MOV_TO_IFS(p9, r30, r25)// M2
1127	mov b0=r21		// I0
1128(pLvSys)mov r18=r0		// A  clear r18 for leave_syscall, no-op otherwise
1129
1130	mov ar.fpsr=r20		// M2
1131	MOV_TO_IIP(r28, r25)	// M2
1132	nop 0
1133	;;
1134(pUStk)	mov ar.rnat=r24		// M2 must happen with RSE in lazy mode
1135	nop 0
1136(pLvSys)mov r2=r0
1137
1138	mov ar.rsc=r27		// M2
1139	mov pr=r31,-1		// I0
1140	RFI			// B
1141
1142	/*
1143	 * On entry:
1144	 *	r20 = &current->thread_info->pre_count (if CONFIG_PREEMPT)
1145	 *	r31 = current->thread_info->flags
1146	 * On exit:
1147	 *	p6 = TRUE if work-pending-check needs to be redone
1148	 *
1149	 * Interrupts are disabled on entry, reenabled depend on work, and
1150	 * disabled on exit.
1151	 */
1152.work_pending_syscall:
1153	add r2=-8,r2
1154	add r3=-8,r3
1155	;;
1156	st8 [r2]=r8
1157	st8 [r3]=r10
1158.work_pending:
1159	tbit.z p6,p0=r31,TIF_NEED_RESCHED	// is resched not needed?
1160(p6)	br.cond.sptk.few .notify
1161	br.call.spnt.many rp=preempt_schedule_irq
1162.ret9:	cmp.eq p6,p0=r0,r0	// p6 <- 1 (re-check)
1163(pLvSys)br.cond.sptk.few  ia64_work_pending_syscall_end
1164	br.cond.sptk.many .work_processed_kernel
1165
1166.notify:
1167(pUStk)	br.call.spnt.many rp=notify_resume_user
1168.ret10:	cmp.ne p6,p0=r0,r0	// p6 <- 0 (don't re-check)
1169(pLvSys)br.cond.sptk.few  ia64_work_pending_syscall_end
1170	br.cond.sptk.many .work_processed_kernel
1171
1172.global ia64_work_pending_syscall_end;
1173ia64_work_pending_syscall_end:
1174	adds r2=PT(R8)+16,r12
1175	adds r3=PT(R10)+16,r12
1176	;;
1177	ld8 r8=[r2]
1178	ld8 r10=[r3]
1179	br.cond.sptk.many ia64_work_processed_syscall
1180END(ia64_leave_kernel)
1181
1182ENTRY(handle_syscall_error)
1183	/*
1184	 * Some system calls (e.g., ptrace, mmap) can return arbitrary values which could
1185	 * lead us to mistake a negative return value as a failed syscall.  Those syscall
1186	 * must deposit a non-zero value in pt_regs.r8 to indicate an error.  If
1187	 * pt_regs.r8 is zero, we assume that the call completed successfully.
1188	 */
1189	PT_REGS_UNWIND_INFO(0)
1190	ld8 r3=[r2]		// load pt_regs.r8
1191	;;
1192	cmp.eq p6,p7=r3,r0	// is pt_regs.r8==0?
1193	;;
1194(p7)	mov r10=-1
1195(p7)	sub r8=0,r8		// negate return value to get errno
1196	br.cond.sptk ia64_leave_syscall
1197END(handle_syscall_error)
1198
1199	/*
1200	 * Invoke schedule_tail(task) while preserving in0-in7, which may be needed
1201	 * in case a system call gets restarted.
1202	 */
1203GLOBAL_ENTRY(ia64_invoke_schedule_tail)
1204	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1205	alloc loc1=ar.pfs,8,2,1,0
1206	mov loc0=rp
1207	mov out0=r8				// Address of previous task
1208	;;
1209	br.call.sptk.many rp=schedule_tail
1210.ret11:	mov ar.pfs=loc1
1211	mov rp=loc0
1212	br.ret.sptk.many rp
1213END(ia64_invoke_schedule_tail)
1214
1215	/*
1216	 * Setup stack and call do_notify_resume_user(), keeping interrupts
1217	 * disabled.
1218	 *
1219	 * Note that pSys and pNonSys need to be set up by the caller.
1220	 * We declare 8 input registers so the system call args get preserved,
1221	 * in case we need to restart a system call.
1222	 */
1223GLOBAL_ENTRY(notify_resume_user)
1224	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(8)
1225	alloc loc1=ar.pfs,8,2,3,0 // preserve all eight input regs in case of syscall restart!
1226	mov r9=ar.unat
1227	mov loc0=rp				// save return address
1228	mov out0=0				// there is no "oldset"
1229	adds out1=8,sp				// out1=&sigscratch->ar_pfs
1230(pSys)	mov out2=1				// out2==1 => we're in a syscall
1231	;;
1232(pNonSys) mov out2=0				// out2==0 => not a syscall
1233	.fframe 16
1234	.spillsp ar.unat, 16
1235	st8 [sp]=r9,-16				// allocate space for ar.unat and save it
1236	st8 [out1]=loc1,-8			// save ar.pfs, out1=&sigscratch
1237	.body
1238	br.call.sptk.many rp=do_notify_resume_user
1239.ret15:	.restore sp
1240	adds sp=16,sp				// pop scratch stack space
1241	;;
1242	ld8 r9=[sp]				// load new unat from sigscratch->scratch_unat
1243	mov rp=loc0
1244	;;
1245	mov ar.unat=r9
1246	mov ar.pfs=loc1
1247	br.ret.sptk.many rp
1248END(notify_resume_user)
1249
1250ENTRY(sys_rt_sigreturn)
1251	PT_REGS_UNWIND_INFO(0)
1252	/*
1253	 * Allocate 8 input registers since ptrace() may clobber them
1254	 */
1255	alloc r2=ar.pfs,8,0,1,0
1256	.prologue
1257	PT_REGS_SAVES(16)
1258	adds sp=-16,sp
1259	.body
1260	cmp.eq pNonSys,pSys=r0,r0		// sigreturn isn't a normal syscall...
1261	;;
1262	/*
1263	 * leave_kernel() restores f6-f11 from pt_regs, but since the streamlined
1264	 * syscall-entry path does not save them we save them here instead.  Note: we
1265	 * don't need to save any other registers that are not saved by the stream-lined
1266	 * syscall path, because restore_sigcontext() restores them.
1267	 */
1268	adds r16=PT(F6)+32,sp
1269	adds r17=PT(F7)+32,sp
1270	;;
1271 	stf.spill [r16]=f6,32
1272 	stf.spill [r17]=f7,32
1273	;;
1274 	stf.spill [r16]=f8,32
1275 	stf.spill [r17]=f9,32
1276	;;
1277 	stf.spill [r16]=f10
1278 	stf.spill [r17]=f11
1279	adds out0=16,sp				// out0 = &sigscratch
1280	br.call.sptk.many rp=ia64_rt_sigreturn
1281.ret19:	.restore sp,0
1282	adds sp=16,sp
1283	;;
1284	ld8 r9=[sp]				// load new ar.unat
1285	mov.sptk b7=r8,ia64_leave_kernel
1286	;;
1287	mov ar.unat=r9
1288	br.many b7
1289END(sys_rt_sigreturn)
1290
1291GLOBAL_ENTRY(ia64_prepare_handle_unaligned)
1292	.prologue
1293	/*
1294	 * r16 = fake ar.pfs, we simply need to make sure privilege is still 0
1295	 */
1296	mov r16=r0
1297	DO_SAVE_SWITCH_STACK
1298	br.call.sptk.many rp=ia64_handle_unaligned	// stack frame setup in ivt
1299.ret21:	.body
1300	DO_LOAD_SWITCH_STACK
1301	br.cond.sptk.many rp				// goes to ia64_leave_kernel
1302END(ia64_prepare_handle_unaligned)
1303
1304	//
1305	// unw_init_running(void (*callback)(info, arg), void *arg)
1306	//
1307#	define EXTRA_FRAME_SIZE	((UNW_FRAME_INFO_SIZE+15)&~15)
1308
1309GLOBAL_ENTRY(unw_init_running)
1310	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1311	alloc loc1=ar.pfs,2,3,3,0
1312	;;
1313	ld8 loc2=[in0],8
1314	mov loc0=rp
1315	mov r16=loc1
1316	DO_SAVE_SWITCH_STACK
1317	.body
1318
1319	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
1320	.fframe IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE
1321	SWITCH_STACK_SAVES(EXTRA_FRAME_SIZE)
1322	adds sp=-EXTRA_FRAME_SIZE,sp
1323	.body
1324	;;
1325	adds out0=16,sp				// &info
1326	mov out1=r13				// current
1327	adds out2=16+EXTRA_FRAME_SIZE,sp	// &switch_stack
1328	br.call.sptk.many rp=unw_init_frame_info
13291:	adds out0=16,sp				// &info
1330	mov b6=loc2
1331	mov loc2=gp				// save gp across indirect function call
1332	;;
1333	ld8 gp=[in0]
1334	mov out1=in1				// arg
1335	br.call.sptk.many rp=b6			// invoke the callback function
13361:	mov gp=loc2				// restore gp
1337
1338	// For now, we don't allow changing registers from within
1339	// unw_init_running; if we ever want to allow that, we'd
1340	// have to do a load_switch_stack here:
1341	.restore sp
1342	adds sp=IA64_SWITCH_STACK_SIZE+EXTRA_FRAME_SIZE,sp
1343
1344	mov ar.pfs=loc1
1345	mov rp=loc0
1346	br.ret.sptk.many rp
1347END(unw_init_running)
1348
1349#ifdef CONFIG_FUNCTION_TRACER
1350#ifdef CONFIG_DYNAMIC_FTRACE
1351GLOBAL_ENTRY(_mcount)
1352	br ftrace_stub
1353END(_mcount)
1354
1355.here:
1356	br.ret.sptk.many b0
1357
1358GLOBAL_ENTRY(ftrace_caller)
1359	alloc out0 = ar.pfs, 8, 0, 4, 0
1360	mov out3 = r0
1361	;;
1362	mov out2 = b0
1363	add r3 = 0x20, r3
1364	mov out1 = r1;
1365	br.call.sptk.many b0 = ftrace_patch_gp
1366	//this might be called from module, so we must patch gp
1367ftrace_patch_gp:
1368	movl gp=__gp
1369	mov b0 = r3
1370	;;
1371.global ftrace_call;
1372ftrace_call:
1373{
1374	.mlx
1375	nop.m 0x0
1376	movl r3 = .here;;
1377}
1378	alloc loc0 = ar.pfs, 4, 4, 2, 0
1379	;;
1380	mov loc1 = b0
1381	mov out0 = b0
1382	mov loc2 = r8
1383	mov loc3 = r15
1384	;;
1385	adds out0 = -MCOUNT_INSN_SIZE, out0
1386	mov out1 = in2
1387	mov b6 = r3
1388
1389	br.call.sptk.many b0 = b6
1390	;;
1391	mov ar.pfs = loc0
1392	mov b0 = loc1
1393	mov r8 = loc2
1394	mov r15 = loc3
1395	br ftrace_stub
1396	;;
1397END(ftrace_caller)
1398
1399#else
1400GLOBAL_ENTRY(_mcount)
1401	movl r2 = ftrace_stub
1402	movl r3 = ftrace_trace_function;;
1403	ld8 r3 = [r3];;
1404	ld8 r3 = [r3];;
1405	cmp.eq p7,p0 = r2, r3
1406(p7)	br.sptk.many ftrace_stub
1407	;;
1408
1409	alloc loc0 = ar.pfs, 4, 4, 2, 0
1410	;;
1411	mov loc1 = b0
1412	mov out0 = b0
1413	mov loc2 = r8
1414	mov loc3 = r15
1415	;;
1416	adds out0 = -MCOUNT_INSN_SIZE, out0
1417	mov out1 = in2
1418	mov b6 = r3
1419
1420	br.call.sptk.many b0 = b6
1421	;;
1422	mov ar.pfs = loc0
1423	mov b0 = loc1
1424	mov r8 = loc2
1425	mov r15 = loc3
1426	br ftrace_stub
1427	;;
1428END(_mcount)
1429#endif
1430
1431GLOBAL_ENTRY(ftrace_stub)
1432	mov r3 = b0
1433	movl r2 = _mcount_ret_helper
1434	;;
1435	mov b6 = r2
1436	mov b7 = r3
1437	br.ret.sptk.many b6
1438
1439_mcount_ret_helper:
1440	mov b0 = r42
1441	mov r1 = r41
1442	mov ar.pfs = r40
1443	br b7
1444END(ftrace_stub)
1445
1446#endif /* CONFIG_FUNCTION_TRACER */
1447
1448	.rodata
1449	.align 8
1450	.globl sys_call_table
1451sys_call_table:
1452	data8 sys_ni_syscall		//  This must be sys_ni_syscall!  See ivt.S.
1453	data8 sys_exit				// 1025
1454	data8 sys_read
1455	data8 sys_write
1456	data8 sys_open
1457	data8 sys_close
1458	data8 sys_creat				// 1030
1459	data8 sys_link
1460	data8 sys_unlink
1461	data8 ia64_execve
1462	data8 sys_chdir
1463	data8 sys_fchdir			// 1035
1464	data8 sys_utimes
1465	data8 sys_mknod
1466	data8 sys_chmod
1467	data8 sys_chown
1468	data8 sys_lseek				// 1040
1469	data8 sys_getpid
1470	data8 sys_getppid
1471	data8 sys_mount
1472	data8 sys_umount
1473	data8 sys_setuid			// 1045
1474	data8 sys_getuid
1475	data8 sys_geteuid
1476	data8 sys_ptrace
1477	data8 sys_access
1478	data8 sys_sync				// 1050
1479	data8 sys_fsync
1480	data8 sys_fdatasync
1481	data8 sys_kill
1482	data8 sys_rename
1483	data8 sys_mkdir				// 1055
1484	data8 sys_rmdir
1485	data8 sys_dup
1486	data8 sys_ia64_pipe
1487	data8 sys_times
1488	data8 ia64_brk				// 1060
1489	data8 sys_setgid
1490	data8 sys_getgid
1491	data8 sys_getegid
1492	data8 sys_acct
1493	data8 sys_ioctl				// 1065
1494	data8 sys_fcntl
1495	data8 sys_umask
1496	data8 sys_chroot
1497	data8 sys_ustat
1498	data8 sys_dup2				// 1070
1499	data8 sys_setreuid
1500	data8 sys_setregid
1501	data8 sys_getresuid
1502	data8 sys_setresuid
1503	data8 sys_getresgid			// 1075
1504	data8 sys_setresgid
1505	data8 sys_getgroups
1506	data8 sys_setgroups
1507	data8 sys_getpgid
1508	data8 sys_setpgid			// 1080
1509	data8 sys_setsid
1510	data8 sys_getsid
1511	data8 sys_sethostname
1512	data8 sys_setrlimit
1513	data8 sys_getrlimit			// 1085
1514	data8 sys_getrusage
1515	data8 sys_gettimeofday
1516	data8 sys_settimeofday
1517	data8 sys_select
1518	data8 sys_poll				// 1090
1519	data8 sys_symlink
1520	data8 sys_readlink
1521	data8 sys_uselib
1522	data8 sys_swapon
1523	data8 sys_swapoff			// 1095
1524	data8 sys_reboot
1525	data8 sys_truncate
1526	data8 sys_ftruncate
1527	data8 sys_fchmod
1528	data8 sys_fchown			// 1100
1529	data8 ia64_getpriority
1530	data8 sys_setpriority
1531	data8 sys_statfs
1532	data8 sys_fstatfs
1533	data8 sys_gettid			// 1105
1534	data8 sys_semget
1535	data8 sys_semop
1536	data8 sys_semctl
1537	data8 sys_msgget
1538	data8 sys_msgsnd			// 1110
1539	data8 sys_msgrcv
1540	data8 sys_msgctl
1541	data8 sys_shmget
1542	data8 sys_shmat
1543	data8 sys_shmdt				// 1115
1544	data8 sys_shmctl
1545	data8 sys_syslog
1546	data8 sys_setitimer
1547	data8 sys_getitimer
1548	data8 sys_ni_syscall			// 1120		/* was: ia64_oldstat */
1549	data8 sys_ni_syscall					/* was: ia64_oldlstat */
1550	data8 sys_ni_syscall					/* was: ia64_oldfstat */
1551	data8 sys_vhangup
1552	data8 sys_lchown
1553	data8 sys_remap_file_pages		// 1125
1554	data8 sys_wait4
1555	data8 sys_sysinfo
1556	data8 sys_clone
1557	data8 sys_setdomainname
1558	data8 sys_newuname			// 1130
1559	data8 sys_adjtimex
1560	data8 sys_ni_syscall					/* was: ia64_create_module */
1561	data8 sys_init_module
1562	data8 sys_delete_module
1563	data8 sys_ni_syscall			// 1135		/* was: sys_get_kernel_syms */
1564	data8 sys_ni_syscall					/* was: sys_query_module */
1565	data8 sys_quotactl
1566	data8 sys_bdflush
1567	data8 sys_sysfs
1568	data8 sys_personality			// 1140
1569	data8 sys_ni_syscall		// sys_afs_syscall
1570	data8 sys_setfsuid
1571	data8 sys_setfsgid
1572	data8 sys_getdents
1573	data8 sys_flock				// 1145
1574	data8 sys_readv
1575	data8 sys_writev
1576	data8 sys_pread64
1577	data8 sys_pwrite64
1578	data8 sys_sysctl			// 1150
1579	data8 sys_mmap
1580	data8 sys_munmap
1581	data8 sys_mlock
1582	data8 sys_mlockall
1583	data8 sys_mprotect			// 1155
1584	data8 ia64_mremap
1585	data8 sys_msync
1586	data8 sys_munlock
1587	data8 sys_munlockall
1588	data8 sys_sched_getparam		// 1160
1589	data8 sys_sched_setparam
1590	data8 sys_sched_getscheduler
1591	data8 sys_sched_setscheduler
1592	data8 sys_sched_yield
1593	data8 sys_sched_get_priority_max	// 1165
1594	data8 sys_sched_get_priority_min
1595	data8 sys_sched_rr_get_interval
1596	data8 sys_nanosleep
1597	data8 sys_ni_syscall			// old nfsservctl
1598	data8 sys_prctl				// 1170
1599	data8 sys_getpagesize
1600	data8 sys_mmap2
1601	data8 sys_pciconfig_read
1602	data8 sys_pciconfig_write
1603	data8 sys_perfmonctl			// 1175
1604	data8 sys_sigaltstack
1605	data8 sys_rt_sigaction
1606	data8 sys_rt_sigpending
1607	data8 sys_rt_sigprocmask
1608	data8 sys_rt_sigqueueinfo		// 1180
1609	data8 sys_rt_sigreturn
1610	data8 sys_rt_sigsuspend
1611	data8 sys_rt_sigtimedwait
1612	data8 sys_getcwd
1613	data8 sys_capget			// 1185
1614	data8 sys_capset
1615	data8 sys_sendfile64
1616	data8 sys_ni_syscall		// sys_getpmsg (STREAMS)
1617	data8 sys_ni_syscall		// sys_putpmsg (STREAMS)
1618	data8 sys_socket			// 1190
1619	data8 sys_bind
1620	data8 sys_connect
1621	data8 sys_listen
1622	data8 sys_accept
1623	data8 sys_getsockname			// 1195
1624	data8 sys_getpeername
1625	data8 sys_socketpair
1626	data8 sys_send
1627	data8 sys_sendto
1628	data8 sys_recv				// 1200
1629	data8 sys_recvfrom
1630	data8 sys_shutdown
1631	data8 sys_setsockopt
1632	data8 sys_getsockopt
1633	data8 sys_sendmsg			// 1205
1634	data8 sys_recvmsg
1635	data8 sys_pivot_root
1636	data8 sys_mincore
1637	data8 sys_madvise
1638	data8 sys_newstat			// 1210
1639	data8 sys_newlstat
1640	data8 sys_newfstat
1641	data8 sys_clone2
1642	data8 sys_getdents64
1643	data8 sys_getunwind			// 1215
1644	data8 sys_readahead
1645	data8 sys_setxattr
1646	data8 sys_lsetxattr
1647	data8 sys_fsetxattr
1648	data8 sys_getxattr			// 1220
1649	data8 sys_lgetxattr
1650	data8 sys_fgetxattr
1651	data8 sys_listxattr
1652	data8 sys_llistxattr
1653	data8 sys_flistxattr			// 1225
1654	data8 sys_removexattr
1655	data8 sys_lremovexattr
1656	data8 sys_fremovexattr
1657	data8 sys_tkill
1658	data8 sys_futex				// 1230
1659	data8 sys_sched_setaffinity
1660	data8 sys_sched_getaffinity
1661	data8 sys_set_tid_address
1662	data8 sys_fadvise64_64
1663	data8 sys_tgkill 			// 1235
1664	data8 sys_exit_group
1665	data8 sys_lookup_dcookie
1666	data8 sys_io_setup
1667	data8 sys_io_destroy
1668	data8 sys_io_getevents			// 1240
1669	data8 sys_io_submit
1670	data8 sys_io_cancel
1671	data8 sys_epoll_create
1672	data8 sys_epoll_ctl
1673	data8 sys_epoll_wait			// 1245
1674	data8 sys_restart_syscall
1675	data8 sys_semtimedop
1676	data8 sys_timer_create
1677	data8 sys_timer_settime
1678	data8 sys_timer_gettime			// 1250
1679	data8 sys_timer_getoverrun
1680	data8 sys_timer_delete
1681	data8 sys_clock_settime
1682	data8 sys_clock_gettime
1683	data8 sys_clock_getres			// 1255
1684	data8 sys_clock_nanosleep
1685	data8 sys_fstatfs64
1686	data8 sys_statfs64
1687	data8 sys_mbind
1688	data8 sys_get_mempolicy			// 1260
1689	data8 sys_set_mempolicy
1690	data8 sys_mq_open
1691	data8 sys_mq_unlink
1692	data8 sys_mq_timedsend
1693	data8 sys_mq_timedreceive		// 1265
1694	data8 sys_mq_notify
1695	data8 sys_mq_getsetattr
1696	data8 sys_kexec_load
1697	data8 sys_ni_syscall			// reserved for vserver
1698	data8 sys_waitid			// 1270
1699	data8 sys_add_key
1700	data8 sys_request_key
1701	data8 sys_keyctl
1702	data8 sys_ioprio_set
1703	data8 sys_ioprio_get			// 1275
1704	data8 sys_move_pages
1705	data8 sys_inotify_init
1706	data8 sys_inotify_add_watch
1707	data8 sys_inotify_rm_watch
1708	data8 sys_migrate_pages			// 1280
1709	data8 sys_openat
1710	data8 sys_mkdirat
1711	data8 sys_mknodat
1712	data8 sys_fchownat
1713	data8 sys_futimesat			// 1285
1714	data8 sys_newfstatat
1715	data8 sys_unlinkat
1716	data8 sys_renameat
1717	data8 sys_linkat
1718	data8 sys_symlinkat			// 1290
1719	data8 sys_readlinkat
1720	data8 sys_fchmodat
1721	data8 sys_faccessat
1722	data8 sys_pselect6
1723	data8 sys_ppoll				// 1295
1724	data8 sys_unshare
1725	data8 sys_splice
1726	data8 sys_set_robust_list
1727	data8 sys_get_robust_list
1728	data8 sys_sync_file_range		// 1300
1729	data8 sys_tee
1730	data8 sys_vmsplice
1731	data8 sys_fallocate
1732	data8 sys_getcpu
1733	data8 sys_epoll_pwait			// 1305
1734	data8 sys_utimensat
1735	data8 sys_signalfd
1736	data8 sys_ni_syscall
1737	data8 sys_eventfd
1738	data8 sys_timerfd_create		// 1310
1739	data8 sys_timerfd_settime
1740	data8 sys_timerfd_gettime
1741	data8 sys_signalfd4
1742	data8 sys_eventfd2
1743	data8 sys_epoll_create1			// 1315
1744	data8 sys_dup3
1745	data8 sys_pipe2
1746	data8 sys_inotify_init1
1747	data8 sys_preadv
1748	data8 sys_pwritev			// 1320
1749	data8 sys_rt_tgsigqueueinfo
1750	data8 sys_recvmmsg
1751	data8 sys_fanotify_init
1752	data8 sys_fanotify_mark
1753	data8 sys_prlimit64			// 1325
1754	data8 sys_name_to_handle_at
1755	data8 sys_open_by_handle_at
1756	data8 sys_clock_adjtime
1757	data8 sys_syncfs
1758	data8 sys_setns				// 1330
1759	data8 sys_sendmmsg
1760	data8 sys_process_vm_readv
1761	data8 sys_process_vm_writev
1762	data8 sys_accept4
1763	data8 sys_finit_module			// 1335
1764	data8 sys_sched_setattr
1765	data8 sys_sched_getattr
1766	data8 sys_renameat2
1767	data8 sys_getrandom
1768	data8 sys_memfd_create			// 1340
1769	data8 sys_bpf
1770	data8 sys_execveat
1771	data8 sys_userfaultfd
1772	data8 sys_membarrier
1773	data8 sys_kcmp				// 1345
1774	data8 sys_mlock2
1775	data8 sys_copy_file_range
1776
1777	.org sys_call_table + 8*NR_syscalls	// guard against failures to increase NR_syscalls
1778