xref: /openbmc/linux/arch/ia64/kernel/fsys.S (revision 93dc544c)
1/*
2 * This file contains the light-weight system call handlers (fsyscall-handlers).
3 *
4 * Copyright (C) 2003 Hewlett-Packard Co
5 * 	David Mosberger-Tang <davidm@hpl.hp.com>
6 *
7 * 25-Sep-03 davidm	Implement fsys_rt_sigprocmask().
8 * 18-Feb-03 louisk	Implement fsys_gettimeofday().
9 * 28-Feb-03 davidm	Fixed several bugs in fsys_gettimeofday().  Tuned it some more,
10 *			probably broke it along the way... ;-)
11 * 13-Jul-04 clameter   Implement fsys_clock_gettime and revise fsys_gettimeofday to make
12 *                      it capable of using memory based clocks without falling back to C code.
13 * 08-Feb-07 Fenghua Yu Implement fsys_getcpu.
14 *
15 */
16
17#include <asm/asmmacro.h>
18#include <asm/errno.h>
19#include <asm/asm-offsets.h>
20#include <asm/percpu.h>
21#include <asm/thread_info.h>
22#include <asm/sal.h>
23#include <asm/signal.h>
24#include <asm/system.h>
25#include <asm/unistd.h>
26
27#include "entry.h"
28
29/*
30 * See Documentation/ia64/fsys.txt for details on fsyscalls.
31 *
32 * On entry to an fsyscall handler:
33 *   r10	= 0 (i.e., defaults to "successful syscall return")
34 *   r11	= saved ar.pfs (a user-level value)
35 *   r15	= system call number
36 *   r16	= "current" task pointer (in normal kernel-mode, this is in r13)
37 *   r32-r39	= system call arguments
38 *   b6		= return address (a user-level value)
39 *   ar.pfs	= previous frame-state (a user-level value)
40 *   PSR.be	= cleared to zero (i.e., little-endian byte order is in effect)
41 *   all other registers may contain values passed in from user-mode
42 *
43 * On return from an fsyscall handler:
44 *   r11	= saved ar.pfs (as passed into the fsyscall handler)
45 *   r15	= system call number (as passed into the fsyscall handler)
46 *   r32-r39	= system call arguments (as passed into the fsyscall handler)
47 *   b6		= return address (as passed into the fsyscall handler)
48 *   ar.pfs	= previous frame-state (as passed into the fsyscall handler)
49 */
50
51ENTRY(fsys_ni_syscall)
52	.prologue
53	.altrp b6
54	.body
55	mov r8=ENOSYS
56	mov r10=-1
57	FSYS_RETURN
58END(fsys_ni_syscall)
59
60ENTRY(fsys_getpid)
61	.prologue
62	.altrp b6
63	.body
64	add r17=IA64_TASK_GROUP_LEADER_OFFSET,r16
65	;;
66	ld8 r17=[r17]				// r17 = current->group_leader
67	add r9=TI_FLAGS+IA64_TASK_SIZE,r16
68	;;
69	ld4 r9=[r9]
70	add r17=IA64_TASK_TGIDLINK_OFFSET,r17
71	;;
72	and r9=TIF_ALLWORK_MASK,r9
73	ld8 r17=[r17]				// r17 = current->group_leader->pids[PIDTYPE_PID].pid
74	;;
75	add r8=IA64_PID_LEVEL_OFFSET,r17
76	;;
77	ld4 r8=[r8]				// r8 = pid->level
78	add r17=IA64_PID_UPID_OFFSET,r17	// r17 = &pid->numbers[0]
79	;;
80	shl r8=r8,IA64_UPID_SHIFT
81	;;
82	add r17=r17,r8				// r17 = &pid->numbers[pid->level]
83	;;
84	ld4 r8=[r17]				// r8 = pid->numbers[pid->level].nr
85	;;
86	mov r17=0
87	;;
88	cmp.ne p8,p0=0,r9
89(p8)	br.spnt.many fsys_fallback_syscall
90	FSYS_RETURN
91END(fsys_getpid)
92
93ENTRY(fsys_getppid)
94	.prologue
95	.altrp b6
96	.body
97	add r17=IA64_TASK_GROUP_LEADER_OFFSET,r16
98	;;
99	ld8 r17=[r17]				// r17 = current->group_leader
100	add r9=TI_FLAGS+IA64_TASK_SIZE,r16
101	;;
102
103	ld4 r9=[r9]
104	add r17=IA64_TASK_REAL_PARENT_OFFSET,r17 // r17 = &current->group_leader->real_parent
105	;;
106	and r9=TIF_ALLWORK_MASK,r9
107
1081:	ld8 r18=[r17]				// r18 = current->group_leader->real_parent
109	;;
110	cmp.ne p8,p0=0,r9
111	add r8=IA64_TASK_TGID_OFFSET,r18	// r8 = &current->group_leader->real_parent->tgid
112	;;
113
114	/*
115	 * The .acq is needed to ensure that the read of tgid has returned its data before
116	 * we re-check "real_parent".
117	 */
118	ld4.acq r8=[r8]				// r8 = current->group_leader->real_parent->tgid
119#ifdef CONFIG_SMP
120	/*
121	 * Re-read current->group_leader->real_parent.
122	 */
123	ld8 r19=[r17]				// r19 = current->group_leader->real_parent
124(p8)	br.spnt.many fsys_fallback_syscall
125	;;
126	cmp.ne p6,p0=r18,r19			// did real_parent change?
127	mov r19=0			// i must not leak kernel bits...
128(p6)	br.cond.spnt.few 1b			// yes -> redo the read of tgid and the check
129	;;
130	mov r17=0			// i must not leak kernel bits...
131	mov r18=0			// i must not leak kernel bits...
132#else
133	mov r17=0			// i must not leak kernel bits...
134	mov r18=0			// i must not leak kernel bits...
135	mov r19=0			// i must not leak kernel bits...
136#endif
137	FSYS_RETURN
138END(fsys_getppid)
139
140ENTRY(fsys_set_tid_address)
141	.prologue
142	.altrp b6
143	.body
144	add r9=TI_FLAGS+IA64_TASK_SIZE,r16
145	add r17=IA64_TASK_TGIDLINK_OFFSET,r16
146	;;
147	ld4 r9=[r9]
148	tnat.z p6,p7=r32		// check argument register for being NaT
149	ld8 r17=[r17]				// r17 = current->pids[PIDTYPE_PID].pid
150	;;
151	and r9=TIF_ALLWORK_MASK,r9
152	add r8=IA64_PID_LEVEL_OFFSET,r17
153	add r18=IA64_TASK_CLEAR_CHILD_TID_OFFSET,r16
154	;;
155	ld4 r8=[r8]				// r8 = pid->level
156	add r17=IA64_PID_UPID_OFFSET,r17	// r17 = &pid->numbers[0]
157	;;
158	shl r8=r8,IA64_UPID_SHIFT
159	;;
160	add r17=r17,r8				// r17 = &pid->numbers[pid->level]
161	;;
162	ld4 r8=[r17]				// r8 = pid->numbers[pid->level].nr
163	;;
164	cmp.ne p8,p0=0,r9
165	mov r17=-1
166	;;
167(p6)	st8 [r18]=r32
168(p7)	st8 [r18]=r17
169(p8)	br.spnt.many fsys_fallback_syscall
170	;;
171	mov r17=0			// i must not leak kernel bits...
172	mov r18=0			// i must not leak kernel bits...
173	FSYS_RETURN
174END(fsys_set_tid_address)
175
176#if IA64_GTOD_LOCK_OFFSET !=0
177#error fsys_gettimeofday incompatible with changes to struct fsyscall_gtod_data_t
178#endif
179#if IA64_ITC_JITTER_OFFSET !=0
180#error fsys_gettimeofday incompatible with changes to struct itc_jitter_data_t
181#endif
182#define CLOCK_REALTIME 0
183#define CLOCK_MONOTONIC 1
184#define CLOCK_DIVIDE_BY_1000 0x4000
185#define CLOCK_ADD_MONOTONIC 0x8000
186
187ENTRY(fsys_gettimeofday)
188	.prologue
189	.altrp b6
190	.body
191	mov r31 = r32
192	tnat.nz p6,p0 = r33		// guard against NaT argument
193(p6)    br.cond.spnt.few .fail_einval
194	mov r30 = CLOCK_DIVIDE_BY_1000
195	;;
196.gettime:
197	// Register map
198	// Incoming r31 = pointer to address where to place result
199	//          r30 = flags determining how time is processed
200	// r2,r3 = temp r4-r7 preserved
201	// r8 = result nanoseconds
202	// r9 = result seconds
203	// r10 = temporary storage for clock difference
204	// r11 = preserved: saved ar.pfs
205	// r12 = preserved: memory stack
206	// r13 = preserved: thread pointer
207	// r14 = address of mask / mask value
208	// r15 = preserved: system call number
209	// r16 = preserved: current task pointer
210	// r17 = (not used)
211	// r18 = (not used)
212	// r19 = address of itc_lastcycle
213	// r20 = struct fsyscall_gtod_data (= address of gtod_lock.sequence)
214	// r21 = address of mmio_ptr
215	// r22 = address of wall_time or monotonic_time
216	// r23 = address of shift / value
217	// r24 = address mult factor / cycle_last value
218	// r25 = itc_lastcycle value
219	// r26 = address clocksource cycle_last
220	// r27 = (not used)
221	// r28 = sequence number at the beginning of critcal section
222	// r29 = address of itc_jitter
223	// r30 = time processing flags / memory address
224	// r31 = pointer to result
225	// Predicates
226	// p6,p7 short term use
227	// p8 = timesource ar.itc
228	// p9 = timesource mmio64
229	// p10 = timesource mmio32 - not used
230	// p11 = timesource not to be handled by asm code
231	// p12 = memory time source ( = p9 | p10) - not used
232	// p13 = do cmpxchg with itc_lastcycle
233	// p14 = Divide by 1000
234	// p15 = Add monotonic
235	//
236	// Note that instructions are optimized for McKinley. McKinley can
237	// process two bundles simultaneously and therefore we continuously
238	// try to feed the CPU two bundles and then a stop.
239
240	add r2 = TI_FLAGS+IA64_TASK_SIZE,r16
241	tnat.nz p6,p0 = r31		// guard against Nat argument
242(p6)	br.cond.spnt.few .fail_einval
243	movl r20 = fsyscall_gtod_data // load fsyscall gettimeofday data address
244	;;
245	ld4 r2 = [r2]			// process work pending flags
246	movl r29 = itc_jitter_data	// itc_jitter
247	add r22 = IA64_GTOD_WALL_TIME_OFFSET,r20	// wall_time
248	add r21 = IA64_CLKSRC_MMIO_OFFSET,r20
249	mov pr = r30,0xc000	// Set predicates according to function
250	;;
251	and r2 = TIF_ALLWORK_MASK,r2
252	add r19 = IA64_ITC_LASTCYCLE_OFFSET,r29
253(p15)	add r22 = IA64_GTOD_MONO_TIME_OFFSET,r20	// monotonic_time
254	;;
255	add r26 = IA64_CLKSRC_CYCLE_LAST_OFFSET,r20	// clksrc_cycle_last
256	cmp.ne p6, p0 = 0, r2	// Fallback if work is scheduled
257(p6)	br.cond.spnt.many fsys_fallback_syscall
258	;;
259	// Begin critical section
260.time_redo:
261	ld4.acq r28 = [r20]	// gtod_lock.sequence, Must take first
262	;;
263	and r28 = ~1,r28	// And make sequence even to force retry if odd
264	;;
265	ld8 r30 = [r21]		// clocksource->mmio_ptr
266	add r24 = IA64_CLKSRC_MULT_OFFSET,r20
267	ld4 r2 = [r29]		// itc_jitter value
268	add r23 = IA64_CLKSRC_SHIFT_OFFSET,r20
269	add r14 = IA64_CLKSRC_MASK_OFFSET,r20
270	;;
271	ld4 r3 = [r24]		// clocksource mult value
272	ld8 r14 = [r14]         // clocksource mask value
273	cmp.eq p8,p9 = 0,r30	// use cpu timer if no mmio_ptr
274	;;
275	setf.sig f7 = r3	// Setup for mult scaling of counter
276(p8)	cmp.ne p13,p0 = r2,r0	// need itc_jitter compensation, set p13
277	ld4 r23 = [r23]		// clocksource shift value
278	ld8 r24 = [r26]		// get clksrc_cycle_last value
279(p9)	cmp.eq p13,p0 = 0,r30	// if mmio_ptr, clear p13 jitter control
280	;;
281	.pred.rel.mutex p8,p9
282(p8)	mov r2 = ar.itc		// CPU_TIMER. 36 clocks latency!!!
283(p9)	ld8 r2 = [r30]		// MMIO_TIMER. Could also have latency issues..
284(p13)	ld8 r25 = [r19]		// get itc_lastcycle value
285	ld8 r9 = [r22],IA64_TIMESPEC_TV_NSEC_OFFSET	// tv_sec
286	;;
287	ld8 r8 = [r22],-IA64_TIMESPEC_TV_NSEC_OFFSET	// tv_nsec
288(p13)	sub r3 = r25,r2		// Diff needed before comparison (thanks davidm)
289	;;
290(p13)	cmp.gt.unc p6,p7 = r3,r0 // check if it is less than last. p6,p7 cleared
291	sub r10 = r2,r24	// current_cycle - last_cycle
292	;;
293(p6)	sub r10 = r25,r24	// time we got was less than last_cycle
294(p7)	mov ar.ccv = r25	// more than last_cycle. Prep for cmpxchg
295	;;
296(p7)	cmpxchg8.rel r3 = [r19],r2,ar.ccv
297	;;
298(p7)	cmp.ne p7,p0 = r25,r3	// if cmpxchg not successful
299	;;
300(p7)	sub r10 = r3,r24	// then use new last_cycle instead
301	;;
302	and r10 = r10,r14	// Apply mask
303	;;
304	setf.sig f8 = r10
305	nop.i 123
306	;;
307	// fault check takes 5 cycles and we have spare time
308EX(.fail_efault, probe.w.fault r31, 3)
309	xmpy.l f8 = f8,f7	// nsec_per_cyc*(counter-last_counter)
310	;;
311	getf.sig r2 = f8
312	mf
313	;;
314	ld4 r10 = [r20]		// gtod_lock.sequence
315	shr.u r2 = r2,r23	// shift by factor
316	;;
317	add r8 = r8,r2		// Add xtime.nsecs
318	cmp4.ne p7,p0 = r28,r10
319(p7)	br.cond.dpnt.few .time_redo	// sequence number changed, redo
320	// End critical section.
321	// Now r8=tv->tv_nsec and r9=tv->tv_sec
322	mov r10 = r0
323	movl r2 = 1000000000
324	add r23 = IA64_TIMESPEC_TV_NSEC_OFFSET, r31
325(p14)	movl r3 = 2361183241434822607	// Prep for / 1000 hack
326	;;
327.time_normalize:
328	mov r21 = r8
329	cmp.ge p6,p0 = r8,r2
330(p14)	shr.u r20 = r8, 3 // We can repeat this if necessary just wasting time
331	;;
332(p14)	setf.sig f8 = r20
333(p6)	sub r8 = r8,r2
334(p6)	add r9 = 1,r9		// two nops before the branch.
335(p14)	setf.sig f7 = r3	// Chances for repeats are 1 in 10000 for gettod
336(p6)	br.cond.dpnt.few .time_normalize
337	;;
338	// Divided by 8 though shift. Now divide by 125
339	// The compiler was able to do that with a multiply
340	// and a shift and we do the same
341EX(.fail_efault, probe.w.fault r23, 3)	// This also costs 5 cycles
342(p14)	xmpy.hu f8 = f8, f7		// xmpy has 5 cycles latency so use it
343	;;
344(p14)	getf.sig r2 = f8
345	;;
346	mov r8 = r0
347(p14)	shr.u r21 = r2, 4
348	;;
349EX(.fail_efault, st8 [r31] = r9)
350EX(.fail_efault, st8 [r23] = r21)
351	FSYS_RETURN
352.fail_einval:
353	mov r8 = EINVAL
354	mov r10 = -1
355	FSYS_RETURN
356.fail_efault:
357	mov r8 = EFAULT
358	mov r10 = -1
359	FSYS_RETURN
360END(fsys_gettimeofday)
361
362ENTRY(fsys_clock_gettime)
363	.prologue
364	.altrp b6
365	.body
366	cmp4.ltu p6, p0 = CLOCK_MONOTONIC, r32
367	// Fallback if this is not CLOCK_REALTIME or CLOCK_MONOTONIC
368(p6)	br.spnt.few fsys_fallback_syscall
369	mov r31 = r33
370	shl r30 = r32,15
371	br.many .gettime
372END(fsys_clock_gettime)
373
374/*
375 * long fsys_rt_sigprocmask (int how, sigset_t *set, sigset_t *oset, size_t sigsetsize).
376 */
377#if _NSIG_WORDS != 1
378# error Sorry, fsys_rt_sigprocmask() needs to be updated for _NSIG_WORDS != 1.
379#endif
380ENTRY(fsys_rt_sigprocmask)
381	.prologue
382	.altrp b6
383	.body
384
385	add r2=IA64_TASK_BLOCKED_OFFSET,r16
386	add r9=TI_FLAGS+IA64_TASK_SIZE,r16
387	cmp4.ltu p6,p0=SIG_SETMASK,r32
388
389	cmp.ne p15,p0=r0,r34			// oset != NULL?
390	tnat.nz p8,p0=r34
391	add r31=IA64_TASK_SIGHAND_OFFSET,r16
392	;;
393	ld8 r3=[r2]				// read/prefetch current->blocked
394	ld4 r9=[r9]
395	tnat.nz.or p6,p0=r35
396
397	cmp.ne.or p6,p0=_NSIG_WORDS*8,r35
398	tnat.nz.or p6,p0=r32
399(p6)	br.spnt.few .fail_einval		// fail with EINVAL
400	;;
401#ifdef CONFIG_SMP
402	ld8 r31=[r31]				// r31 <- current->sighand
403#endif
404	and r9=TIF_ALLWORK_MASK,r9
405	tnat.nz.or p8,p0=r33
406	;;
407	cmp.ne p7,p0=0,r9
408	cmp.eq p6,p0=r0,r33			// set == NULL?
409	add r31=IA64_SIGHAND_SIGLOCK_OFFSET,r31	// r31 <- current->sighand->siglock
410(p8)	br.spnt.few .fail_efault		// fail with EFAULT
411(p7)	br.spnt.many fsys_fallback_syscall	// got pending kernel work...
412(p6)	br.dpnt.many .store_mask		// -> short-circuit to just reading the signal mask
413
414	/* Argh, we actually have to do some work and _update_ the signal mask: */
415
416EX(.fail_efault, probe.r.fault r33, 3)		// verify user has read-access to *set
417EX(.fail_efault, ld8 r14=[r33])			// r14 <- *set
418	mov r17=(1 << (SIGKILL - 1)) | (1 << (SIGSTOP - 1))
419	;;
420
421	rsm psr.i				// mask interrupt delivery
422	mov ar.ccv=0
423	andcm r14=r14,r17			// filter out SIGKILL & SIGSTOP
424
425#ifdef CONFIG_SMP
426	mov r17=1
427	;;
428	cmpxchg4.acq r18=[r31],r17,ar.ccv	// try to acquire the lock
429	mov r8=EINVAL			// default to EINVAL
430	;;
431	ld8 r3=[r2]			// re-read current->blocked now that we hold the lock
432	cmp4.ne p6,p0=r18,r0
433(p6)	br.cond.spnt.many .lock_contention
434	;;
435#else
436	ld8 r3=[r2]			// re-read current->blocked now that we hold the lock
437	mov r8=EINVAL			// default to EINVAL
438#endif
439	add r18=IA64_TASK_PENDING_OFFSET+IA64_SIGPENDING_SIGNAL_OFFSET,r16
440	add r19=IA64_TASK_SIGNAL_OFFSET,r16
441	cmp4.eq p6,p0=SIG_BLOCK,r32
442	;;
443	ld8 r19=[r19]			// r19 <- current->signal
444	cmp4.eq p7,p0=SIG_UNBLOCK,r32
445	cmp4.eq p8,p0=SIG_SETMASK,r32
446	;;
447	ld8 r18=[r18]			// r18 <- current->pending.signal
448	.pred.rel.mutex p6,p7,p8
449(p6)	or r14=r3,r14			// SIG_BLOCK
450(p7)	andcm r14=r3,r14		// SIG_UNBLOCK
451
452(p8)	mov r14=r14			// SIG_SETMASK
453(p6)	mov r8=0			// clear error code
454	// recalc_sigpending()
455	add r17=IA64_SIGNAL_GROUP_STOP_COUNT_OFFSET,r19
456
457	add r19=IA64_SIGNAL_SHARED_PENDING_OFFSET+IA64_SIGPENDING_SIGNAL_OFFSET,r19
458	;;
459	ld4 r17=[r17]		// r17 <- current->signal->group_stop_count
460(p7)	mov r8=0		// clear error code
461
462	ld8 r19=[r19]		// r19 <- current->signal->shared_pending
463	;;
464	cmp4.gt p6,p7=r17,r0	// p6/p7 <- (current->signal->group_stop_count > 0)?
465(p8)	mov r8=0		// clear error code
466
467	or r18=r18,r19		// r18 <- current->pending | current->signal->shared_pending
468	;;
469	// r18 <- (current->pending | current->signal->shared_pending) & ~current->blocked:
470	andcm r18=r18,r14
471	add r9=TI_FLAGS+IA64_TASK_SIZE,r16
472	;;
473
474(p7)	cmp.ne.or.andcm p6,p7=r18,r0		// p6/p7 <- signal pending
475	mov r19=0					// i must not leak kernel bits...
476(p6)	br.cond.dpnt.many .sig_pending
477	;;
478
4791:	ld4 r17=[r9]				// r17 <- current->thread_info->flags
480	;;
481	mov ar.ccv=r17
482	and r18=~_TIF_SIGPENDING,r17		// r18 <- r17 & ~(1 << TIF_SIGPENDING)
483	;;
484
485	st8 [r2]=r14				// update current->blocked with new mask
486	cmpxchg4.acq r8=[r9],r18,ar.ccv		// current->thread_info->flags <- r18
487	;;
488	cmp.ne p6,p0=r17,r8			// update failed?
489(p6)	br.cond.spnt.few 1b			// yes -> retry
490
491#ifdef CONFIG_SMP
492	st4.rel [r31]=r0			// release the lock
493#endif
494	ssm psr.i
495	;;
496
497	srlz.d					// ensure psr.i is set again
498	mov r18=0					// i must not leak kernel bits...
499
500.store_mask:
501EX(.fail_efault, (p15) probe.w.fault r34, 3)	// verify user has write-access to *oset
502EX(.fail_efault, (p15) st8 [r34]=r3)
503	mov r2=0					// i must not leak kernel bits...
504	mov r3=0					// i must not leak kernel bits...
505	mov r8=0				// return 0
506	mov r9=0					// i must not leak kernel bits...
507	mov r14=0					// i must not leak kernel bits...
508	mov r17=0					// i must not leak kernel bits...
509	mov r31=0					// i must not leak kernel bits...
510	FSYS_RETURN
511
512.sig_pending:
513#ifdef CONFIG_SMP
514	st4.rel [r31]=r0			// release the lock
515#endif
516	ssm psr.i
517	;;
518	srlz.d
519	br.sptk.many fsys_fallback_syscall	// with signal pending, do the heavy-weight syscall
520
521#ifdef CONFIG_SMP
522.lock_contention:
523	/* Rather than spinning here, fall back on doing a heavy-weight syscall.  */
524	ssm psr.i
525	;;
526	srlz.d
527	br.sptk.many fsys_fallback_syscall
528#endif
529END(fsys_rt_sigprocmask)
530
531/*
532 * fsys_getcpu doesn't use the third parameter in this implementation. It reads
533 * current_thread_info()->cpu and corresponding node in cpu_to_node_map.
534 */
535ENTRY(fsys_getcpu)
536	.prologue
537	.altrp b6
538	.body
539	;;
540	add r2=TI_FLAGS+IA64_TASK_SIZE,r16
541	tnat.nz p6,p0 = r32			// guard against NaT argument
542	add r3=TI_CPU+IA64_TASK_SIZE,r16
543	;;
544	ld4 r3=[r3]				// M r3 = thread_info->cpu
545	ld4 r2=[r2]				// M r2 = thread_info->flags
546(p6)    br.cond.spnt.few .fail_einval		// B
547	;;
548	tnat.nz p7,p0 = r33			// I guard against NaT argument
549(p7)    br.cond.spnt.few .fail_einval		// B
550#ifdef CONFIG_NUMA
551	movl r17=cpu_to_node_map
552	;;
553EX(.fail_efault, probe.w.fault r32, 3)		// M This takes 5 cycles
554EX(.fail_efault, probe.w.fault r33, 3)		// M This takes 5 cycles
555	shladd r18=r3,1,r17
556	;;
557	ld2 r20=[r18]				// r20 = cpu_to_node_map[cpu]
558	and r2 = TIF_ALLWORK_MASK,r2
559	;;
560	cmp.ne p8,p0=0,r2
561(p8)	br.spnt.many fsys_fallback_syscall
562	;;
563	;;
564EX(.fail_efault, st4 [r32] = r3)
565EX(.fail_efault, st2 [r33] = r20)
566	mov r8=0
567	;;
568#else
569EX(.fail_efault, probe.w.fault r32, 3)		// M This takes 5 cycles
570EX(.fail_efault, probe.w.fault r33, 3)		// M This takes 5 cycles
571	and r2 = TIF_ALLWORK_MASK,r2
572	;;
573	cmp.ne p8,p0=0,r2
574(p8)	br.spnt.many fsys_fallback_syscall
575	;;
576EX(.fail_efault, st4 [r32] = r3)
577EX(.fail_efault, st2 [r33] = r0)
578	mov r8=0
579	;;
580#endif
581	FSYS_RETURN
582END(fsys_getcpu)
583
584ENTRY(fsys_fallback_syscall)
585	.prologue
586	.altrp b6
587	.body
588	/*
589	 * We only get here from light-weight syscall handlers.  Thus, we already
590	 * know that r15 contains a valid syscall number.  No need to re-check.
591	 */
592	adds r17=-1024,r15
593	movl r14=sys_call_table
594	;;
595	rsm psr.i
596	shladd r18=r17,3,r14
597	;;
598	ld8 r18=[r18]				// load normal (heavy-weight) syscall entry-point
599	mov r29=psr				// read psr (12 cyc load latency)
600	mov r27=ar.rsc
601	mov r21=ar.fpsr
602	mov r26=ar.pfs
603END(fsys_fallback_syscall)
604	/* FALL THROUGH */
605GLOBAL_ENTRY(fsys_bubble_down)
606	.prologue
607	.altrp b6
608	.body
609	/*
610	 * We get here for syscalls that don't have a lightweight
611	 * handler.  For those, we need to bubble down into the kernel
612	 * and that requires setting up a minimal pt_regs structure,
613	 * and initializing the CPU state more or less as if an
614	 * interruption had occurred.  To make syscall-restarts work,
615	 * we setup pt_regs such that cr_iip points to the second
616	 * instruction in syscall_via_break.  Decrementing the IP
617	 * hence will restart the syscall via break and not
618	 * decrementing IP will return us to the caller, as usual.
619	 * Note that we preserve the value of psr.pp rather than
620	 * initializing it from dcr.pp.  This makes it possible to
621	 * distinguish fsyscall execution from other privileged
622	 * execution.
623	 *
624	 * On entry:
625	 *	- normal fsyscall handler register usage, except
626	 *	  that we also have:
627	 *	- r18: address of syscall entry point
628	 *	- r21: ar.fpsr
629	 *	- r26: ar.pfs
630	 *	- r27: ar.rsc
631	 *	- r29: psr
632	 *
633	 * We used to clear some PSR bits here but that requires slow
634	 * serialization.  Fortuntely, that isn't really necessary.
635	 * The rationale is as follows: we used to clear bits
636	 * ~PSR_PRESERVED_BITS in PSR.L.  Since
637	 * PSR_PRESERVED_BITS==PSR.{UP,MFL,MFH,PK,DT,PP,SP,RT,IC}, we
638	 * ended up clearing PSR.{BE,AC,I,DFL,DFH,DI,DB,SI,TB}.
639	 * However,
640	 *
641	 * PSR.BE : already is turned off in __kernel_syscall_via_epc()
642	 * PSR.AC : don't care (kernel normally turns PSR.AC on)
643	 * PSR.I  : already turned off by the time fsys_bubble_down gets
644	 *	    invoked
645	 * PSR.DFL: always 0 (kernel never turns it on)
646	 * PSR.DFH: don't care --- kernel never touches f32-f127 on its own
647	 *	    initiative
648	 * PSR.DI : always 0 (kernel never turns it on)
649	 * PSR.SI : always 0 (kernel never turns it on)
650	 * PSR.DB : don't care --- kernel never enables kernel-level
651	 *	    breakpoints
652	 * PSR.TB : must be 0 already; if it wasn't zero on entry to
653	 *          __kernel_syscall_via_epc, the branch to fsys_bubble_down
654	 *          will trigger a taken branch; the taken-trap-handler then
655	 *          converts the syscall into a break-based system-call.
656	 */
657	/*
658	 * Reading psr.l gives us only bits 0-31, psr.it, and psr.mc.
659	 * The rest we have to synthesize.
660	 */
661#	define PSR_ONE_BITS		((3 << IA64_PSR_CPL0_BIT)	\
662					 | (0x1 << IA64_PSR_RI_BIT)	\
663					 | IA64_PSR_BN | IA64_PSR_I)
664
665	invala					// M0|1
666	movl r14=ia64_ret_from_syscall		// X
667
668	nop.m 0
669	movl r28=__kernel_syscall_via_break	// X	create cr.iip
670	;;
671
672	mov r2=r16				// A    get task addr to addl-addressable register
673	adds r16=IA64_TASK_THREAD_ON_USTACK_OFFSET,r16 // A
674	mov r31=pr				// I0   save pr (2 cyc)
675	;;
676	st1 [r16]=r0				// M2|3 clear current->thread.on_ustack flag
677	addl r22=IA64_RBS_OFFSET,r2		// A    compute base of RBS
678	add r3=TI_FLAGS+IA64_TASK_SIZE,r2	// A
679	;;
680	ld4 r3=[r3]				// M0|1 r3 = current_thread_info()->flags
681	lfetch.fault.excl.nt1 [r22]		// M0|1 prefetch register backing-store
682	nop.i 0
683	;;
684	mov ar.rsc=0				// M2   set enforced lazy mode, pl 0, LE, loadrs=0
685#ifdef CONFIG_VIRT_CPU_ACCOUNTING
686	mov.m r30=ar.itc			// M    get cycle for accounting
687#else
688	nop.m 0
689#endif
690	nop.i 0
691	;;
692	mov r23=ar.bspstore			// M2 (12 cyc) save ar.bspstore
693	mov.m r24=ar.rnat			// M2 (5 cyc) read ar.rnat (dual-issues!)
694	nop.i 0
695	;;
696	mov ar.bspstore=r22			// M2 (6 cyc) switch to kernel RBS
697	movl r8=PSR_ONE_BITS			// X
698	;;
699	mov r25=ar.unat				// M2 (5 cyc) save ar.unat
700	mov r19=b6				// I0   save b6 (2 cyc)
701	mov r20=r1				// A    save caller's gp in r20
702	;;
703	or r29=r8,r29				// A    construct cr.ipsr value to save
704	mov b6=r18				// I0   copy syscall entry-point to b6 (7 cyc)
705	addl r1=IA64_STK_OFFSET-IA64_PT_REGS_SIZE,r2 // A compute base of memory stack
706
707	mov r18=ar.bsp				// M2   save (kernel) ar.bsp (12 cyc)
708	cmp.ne pKStk,pUStk=r0,r0		// A    set pKStk <- 0, pUStk <- 1
709	br.call.sptk.many b7=ia64_syscall_setup	// B
710	;;
711#ifdef CONFIG_VIRT_CPU_ACCOUNTING
712	// mov.m r30=ar.itc is called in advance
713	add r16=TI_AC_STAMP+IA64_TASK_SIZE,r2
714	add r17=TI_AC_LEAVE+IA64_TASK_SIZE,r2
715	;;
716	ld8 r18=[r16],TI_AC_STIME-TI_AC_STAMP	// time at last check in kernel
717	ld8 r19=[r17],TI_AC_UTIME-TI_AC_LEAVE	// time at leave kernel
718	;;
719	ld8 r20=[r16],TI_AC_STAMP-TI_AC_STIME	// cumulated stime
720	ld8 r21=[r17]				// cumulated utime
721	sub r22=r19,r18				// stime before leave kernel
722	;;
723	st8 [r16]=r30,TI_AC_STIME-TI_AC_STAMP	// update stamp
724	sub r18=r30,r19				// elapsed time in user mode
725	;;
726	add r20=r20,r22				// sum stime
727	add r21=r21,r18				// sum utime
728	;;
729	st8 [r16]=r20				// update stime
730	st8 [r17]=r21				// update utime
731	;;
732#endif
733	mov ar.rsc=0x3				// M2   set eager mode, pl 0, LE, loadrs=0
734	mov rp=r14				// I0   set the real return addr
735	and r3=_TIF_SYSCALL_TRACEAUDIT,r3	// A
736	;;
737	ssm psr.i				// M2   we're on kernel stacks now, reenable irqs
738	cmp.eq p8,p0=r3,r0			// A
739(p10)	br.cond.spnt.many ia64_ret_from_syscall	// B    return if bad call-frame or r15 is a NaT
740
741	nop.m 0
742(p8)	br.call.sptk.many b6=b6			// B    (ignore return address)
743	br.cond.spnt ia64_trace_syscall		// B
744END(fsys_bubble_down)
745
746	.rodata
747	.align 8
748	.globl fsyscall_table
749
750	data8 fsys_bubble_down
751fsyscall_table:
752	data8 fsys_ni_syscall
753	data8 0				// exit			// 1025
754	data8 0				// read
755	data8 0				// write
756	data8 0				// open
757	data8 0				// close
758	data8 0				// creat		// 1030
759	data8 0				// link
760	data8 0				// unlink
761	data8 0				// execve
762	data8 0				// chdir
763	data8 0				// fchdir		// 1035
764	data8 0				// utimes
765	data8 0				// mknod
766	data8 0				// chmod
767	data8 0				// chown
768	data8 0				// lseek		// 1040
769	data8 fsys_getpid		// getpid
770	data8 fsys_getppid		// getppid
771	data8 0				// mount
772	data8 0				// umount
773	data8 0				// setuid		// 1045
774	data8 0				// getuid
775	data8 0				// geteuid
776	data8 0				// ptrace
777	data8 0				// access
778	data8 0				// sync			// 1050
779	data8 0				// fsync
780	data8 0				// fdatasync
781	data8 0				// kill
782	data8 0				// rename
783	data8 0				// mkdir		// 1055
784	data8 0				// rmdir
785	data8 0				// dup
786	data8 0				// pipe
787	data8 0				// times
788	data8 0				// brk			// 1060
789	data8 0				// setgid
790	data8 0				// getgid
791	data8 0				// getegid
792	data8 0				// acct
793	data8 0				// ioctl		// 1065
794	data8 0				// fcntl
795	data8 0				// umask
796	data8 0				// chroot
797	data8 0				// ustat
798	data8 0				// dup2			// 1070
799	data8 0				// setreuid
800	data8 0				// setregid
801	data8 0				// getresuid
802	data8 0				// setresuid
803	data8 0				// getresgid		// 1075
804	data8 0				// setresgid
805	data8 0				// getgroups
806	data8 0				// setgroups
807	data8 0				// getpgid
808	data8 0				// setpgid		// 1080
809	data8 0				// setsid
810	data8 0				// getsid
811	data8 0				// sethostname
812	data8 0				// setrlimit
813	data8 0				// getrlimit		// 1085
814	data8 0				// getrusage
815	data8 fsys_gettimeofday		// gettimeofday
816	data8 0				// settimeofday
817	data8 0				// select
818	data8 0				// poll			// 1090
819	data8 0				// symlink
820	data8 0				// readlink
821	data8 0				// uselib
822	data8 0				// swapon
823	data8 0				// swapoff		// 1095
824	data8 0				// reboot
825	data8 0				// truncate
826	data8 0				// ftruncate
827	data8 0				// fchmod
828	data8 0				// fchown		// 1100
829	data8 0				// getpriority
830	data8 0				// setpriority
831	data8 0				// statfs
832	data8 0				// fstatfs
833	data8 0				// gettid		// 1105
834	data8 0				// semget
835	data8 0				// semop
836	data8 0				// semctl
837	data8 0				// msgget
838	data8 0				// msgsnd		// 1110
839	data8 0				// msgrcv
840	data8 0				// msgctl
841	data8 0				// shmget
842	data8 0				// shmat
843	data8 0				// shmdt		// 1115
844	data8 0				// shmctl
845	data8 0				// syslog
846	data8 0				// setitimer
847	data8 0				// getitimer
848	data8 0					 		// 1120
849	data8 0
850	data8 0
851	data8 0				// vhangup
852	data8 0				// lchown
853	data8 0				// remap_file_pages	// 1125
854	data8 0				// wait4
855	data8 0				// sysinfo
856	data8 0				// clone
857	data8 0				// setdomainname
858	data8 0				// newuname		// 1130
859	data8 0				// adjtimex
860	data8 0
861	data8 0				// init_module
862	data8 0				// delete_module
863	data8 0							// 1135
864	data8 0
865	data8 0				// quotactl
866	data8 0				// bdflush
867	data8 0				// sysfs
868	data8 0				// personality		// 1140
869	data8 0				// afs_syscall
870	data8 0				// setfsuid
871	data8 0				// setfsgid
872	data8 0				// getdents
873	data8 0				// flock		// 1145
874	data8 0				// readv
875	data8 0				// writev
876	data8 0				// pread64
877	data8 0				// pwrite64
878	data8 0				// sysctl		// 1150
879	data8 0				// mmap
880	data8 0				// munmap
881	data8 0				// mlock
882	data8 0				// mlockall
883	data8 0				// mprotect		// 1155
884	data8 0				// mremap
885	data8 0				// msync
886	data8 0				// munlock
887	data8 0				// munlockall
888	data8 0				// sched_getparam	// 1160
889	data8 0				// sched_setparam
890	data8 0				// sched_getscheduler
891	data8 0				// sched_setscheduler
892	data8 0				// sched_yield
893	data8 0				// sched_get_priority_max	// 1165
894	data8 0				// sched_get_priority_min
895	data8 0				// sched_rr_get_interval
896	data8 0				// nanosleep
897	data8 0				// nfsservctl
898	data8 0				// prctl		// 1170
899	data8 0				// getpagesize
900	data8 0				// mmap2
901	data8 0				// pciconfig_read
902	data8 0				// pciconfig_write
903	data8 0				// perfmonctl		// 1175
904	data8 0				// sigaltstack
905	data8 0				// rt_sigaction
906	data8 0				// rt_sigpending
907	data8 fsys_rt_sigprocmask	// rt_sigprocmask
908	data8 0				// rt_sigqueueinfo	// 1180
909	data8 0				// rt_sigreturn
910	data8 0				// rt_sigsuspend
911	data8 0				// rt_sigtimedwait
912	data8 0				// getcwd
913	data8 0				// capget		// 1185
914	data8 0				// capset
915	data8 0				// sendfile
916	data8 0
917	data8 0
918	data8 0				// socket		// 1190
919	data8 0				// bind
920	data8 0				// connect
921	data8 0				// listen
922	data8 0				// accept
923	data8 0				// getsockname		// 1195
924	data8 0				// getpeername
925	data8 0				// socketpair
926	data8 0				// send
927	data8 0				// sendto
928	data8 0				// recv			// 1200
929	data8 0				// recvfrom
930	data8 0				// shutdown
931	data8 0				// setsockopt
932	data8 0				// getsockopt
933	data8 0				// sendmsg		// 1205
934	data8 0				// recvmsg
935	data8 0				// pivot_root
936	data8 0				// mincore
937	data8 0				// madvise
938	data8 0				// newstat		// 1210
939	data8 0				// newlstat
940	data8 0				// newfstat
941	data8 0				// clone2
942	data8 0				// getdents64
943	data8 0				// getunwind		// 1215
944	data8 0				// readahead
945	data8 0				// setxattr
946	data8 0				// lsetxattr
947	data8 0				// fsetxattr
948	data8 0				// getxattr		// 1220
949	data8 0				// lgetxattr
950	data8 0				// fgetxattr
951	data8 0				// listxattr
952	data8 0				// llistxattr
953	data8 0				// flistxattr		// 1225
954	data8 0				// removexattr
955	data8 0				// lremovexattr
956	data8 0				// fremovexattr
957	data8 0				// tkill
958	data8 0				// futex		// 1230
959	data8 0				// sched_setaffinity
960	data8 0				// sched_getaffinity
961	data8 fsys_set_tid_address	// set_tid_address
962	data8 0				// fadvise64_64
963	data8 0				// tgkill		// 1235
964	data8 0				// exit_group
965	data8 0				// lookup_dcookie
966	data8 0				// io_setup
967	data8 0				// io_destroy
968	data8 0				// io_getevents		// 1240
969	data8 0				// io_submit
970	data8 0				// io_cancel
971	data8 0				// epoll_create
972	data8 0				// epoll_ctl
973	data8 0				// epoll_wait		// 1245
974	data8 0				// restart_syscall
975	data8 0				// semtimedop
976	data8 0				// timer_create
977	data8 0				// timer_settime
978	data8 0				// timer_gettime 	// 1250
979	data8 0				// timer_getoverrun
980	data8 0				// timer_delete
981	data8 0				// clock_settime
982	data8 fsys_clock_gettime	// clock_gettime
983	data8 0				// clock_getres		// 1255
984	data8 0				// clock_nanosleep
985	data8 0				// fstatfs64
986	data8 0				// statfs64
987	data8 0				// mbind
988	data8 0				// get_mempolicy	// 1260
989	data8 0				// set_mempolicy
990	data8 0				// mq_open
991	data8 0				// mq_unlink
992	data8 0				// mq_timedsend
993	data8 0				// mq_timedreceive	// 1265
994	data8 0				// mq_notify
995	data8 0				// mq_getsetattr
996	data8 0				// kexec_load
997	data8 0				// vserver
998	data8 0				// waitid		// 1270
999	data8 0				// add_key
1000	data8 0				// request_key
1001	data8 0				// keyctl
1002	data8 0				// ioprio_set
1003	data8 0				// ioprio_get		// 1275
1004	data8 0				// move_pages
1005	data8 0				// inotify_init
1006	data8 0				// inotify_add_watch
1007	data8 0				// inotify_rm_watch
1008	data8 0				// migrate_pages	// 1280
1009	data8 0				// openat
1010	data8 0				// mkdirat
1011	data8 0				// mknodat
1012	data8 0				// fchownat
1013	data8 0				// futimesat		// 1285
1014	data8 0				// newfstatat
1015	data8 0				// unlinkat
1016	data8 0				// renameat
1017	data8 0				// linkat
1018	data8 0				// symlinkat		// 1290
1019	data8 0				// readlinkat
1020	data8 0				// fchmodat
1021	data8 0				// faccessat
1022	data8 0
1023	data8 0							// 1295
1024	data8 0				// unshare
1025	data8 0				// splice
1026	data8 0				// set_robust_list
1027	data8 0				// get_robust_list
1028	data8 0				// sync_file_range	// 1300
1029	data8 0				// tee
1030	data8 0				// vmsplice
1031	data8 0
1032	data8 fsys_getcpu		// getcpu		// 1304
1033
1034	// fill in zeros for the remaining entries
1035	.zero:
1036	.space fsyscall_table + 8*NR_syscalls - .zero, 0
1037