xref: /openbmc/linux/arch/arm/kernel/entry-armv.S (revision b627b4ed)
1/*
2 *  linux/arch/arm/kernel/entry-armv.S
3 *
4 *  Copyright (C) 1996,1997,1998 Russell King.
5 *  ARM700 fix by Matthew Godbolt (linux-user@willothewisp.demon.co.uk)
6 *  nommu support by Hyok S. Choi (hyok.choi@samsung.com)
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
11 *
12 *  Low-level vector interface routines
13 *
14 *  Note:  there is a StrongARM bug in the STMIA rn, {regs}^ instruction
15 *  that causes it to save wrong values...  Be aware!
16 */
17
18#include <asm/memory.h>
19#include <asm/glue.h>
20#include <asm/vfpmacros.h>
21#include <mach/entry-macro.S>
22#include <asm/thread_notify.h>
23#include <asm/unwind.h>
24
25#include "entry-header.S"
26
27/*
28 * Interrupt handling.  Preserves r7, r8, r9
29 */
30	.macro	irq_handler
31	get_irqnr_preamble r5, lr
321:	get_irqnr_and_base r0, r6, r5, lr
33	movne	r1, sp
34	@
35	@ routine called with r0 = irq number, r1 = struct pt_regs *
36	@
37	adrne	lr, 1b
38	bne	asm_do_IRQ
39
40#ifdef CONFIG_SMP
41	/*
42	 * XXX
43	 *
44	 * this macro assumes that irqstat (r6) and base (r5) are
45	 * preserved from get_irqnr_and_base above
46	 */
47	test_for_ipi r0, r6, r5, lr
48	movne	r0, sp
49	adrne	lr, 1b
50	bne	do_IPI
51
52#ifdef CONFIG_LOCAL_TIMERS
53	test_for_ltirq r0, r6, r5, lr
54	movne	r0, sp
55	adrne	lr, 1b
56	bne	do_local_timer
57#endif
58#endif
59
60	.endm
61
62#ifdef CONFIG_KPROBES
63	.section	.kprobes.text,"ax",%progbits
64#else
65	.text
66#endif
67
68/*
69 * Invalid mode handlers
70 */
71	.macro	inv_entry, reason
72	sub	sp, sp, #S_FRAME_SIZE
73	stmib	sp, {r1 - lr}
74	mov	r1, #\reason
75	.endm
76
77__pabt_invalid:
78	inv_entry BAD_PREFETCH
79	b	common_invalid
80ENDPROC(__pabt_invalid)
81
82__dabt_invalid:
83	inv_entry BAD_DATA
84	b	common_invalid
85ENDPROC(__dabt_invalid)
86
87__irq_invalid:
88	inv_entry BAD_IRQ
89	b	common_invalid
90ENDPROC(__irq_invalid)
91
92__und_invalid:
93	inv_entry BAD_UNDEFINSTR
94
95	@
96	@ XXX fall through to common_invalid
97	@
98
99@
100@ common_invalid - generic code for failed exception (re-entrant version of handlers)
101@
102common_invalid:
103	zero_fp
104
105	ldmia	r0, {r4 - r6}
106	add	r0, sp, #S_PC		@ here for interlock avoidance
107	mov	r7, #-1			@  ""   ""    ""        ""
108	str	r4, [sp]		@ save preserved r0
109	stmia	r0, {r5 - r7}		@ lr_<exception>,
110					@ cpsr_<exception>, "old_r0"
111
112	mov	r0, sp
113	b	bad_mode
114ENDPROC(__und_invalid)
115
116/*
117 * SVC mode handlers
118 */
119
120#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
121#define SPFIX(code...) code
122#else
123#define SPFIX(code...)
124#endif
125
126	.macro	svc_entry, stack_hole=0
127 UNWIND(.fnstart		)
128 UNWIND(.save {r0 - pc}		)
129	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole)
130 SPFIX(	tst	sp, #4		)
131 SPFIX(	bicne	sp, sp, #4	)
132	stmib	sp, {r1 - r12}
133
134	ldmia	r0, {r1 - r3}
135	add	r5, sp, #S_SP		@ here for interlock avoidance
136	mov	r4, #-1			@  ""  ""      ""       ""
137	add	r0, sp, #(S_FRAME_SIZE + \stack_hole)
138 SPFIX(	addne	r0, r0, #4	)
139	str	r1, [sp]		@ save the "real" r0 copied
140					@ from the exception stack
141
142	mov	r1, lr
143
144	@
145	@ We are now ready to fill in the remaining blanks on the stack:
146	@
147	@  r0 - sp_svc
148	@  r1 - lr_svc
149	@  r2 - lr_<exception>, already fixed up for correct return/restart
150	@  r3 - spsr_<exception>
151	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
152	@
153	stmia	r5, {r0 - r4}
154	.endm
155
156	.align	5
157__dabt_svc:
158	svc_entry
159
160	@
161	@ get ready to re-enable interrupts if appropriate
162	@
163	mrs	r9, cpsr
164	tst	r3, #PSR_I_BIT
165	biceq	r9, r9, #PSR_I_BIT
166
167	@
168	@ Call the processor-specific abort handler:
169	@
170	@  r2 - aborted context pc
171	@  r3 - aborted context cpsr
172	@
173	@ The abort handler must return the aborted address in r0, and
174	@ the fault status register in r1.  r9 must be preserved.
175	@
176#ifdef MULTI_DABORT
177	ldr	r4, .LCprocfns
178	mov	lr, pc
179	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
180#else
181	bl	CPU_DABORT_HANDLER
182#endif
183
184	@
185	@ set desired IRQ state, then call main handler
186	@
187	msr	cpsr_c, r9
188	mov	r2, sp
189	bl	do_DataAbort
190
191	@
192	@ IRQs off again before pulling preserved data off the stack
193	@
194	disable_irq
195
196	@
197	@ restore SPSR and restart the instruction
198	@
199	ldr	r0, [sp, #S_PSR]
200	msr	spsr_cxsf, r0
201	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
202 UNWIND(.fnend		)
203ENDPROC(__dabt_svc)
204
205	.align	5
206__irq_svc:
207	svc_entry
208
209#ifdef CONFIG_TRACE_IRQFLAGS
210	bl	trace_hardirqs_off
211#endif
212#ifdef CONFIG_PREEMPT
213	get_thread_info tsk
214	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
215	add	r7, r8, #1			@ increment it
216	str	r7, [tsk, #TI_PREEMPT]
217#endif
218
219	irq_handler
220#ifdef CONFIG_PREEMPT
221	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
222	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
223	teq	r8, #0				@ if preempt count != 0
224	movne	r0, #0				@ force flags to 0
225	tst	r0, #_TIF_NEED_RESCHED
226	blne	svc_preempt
227#endif
228	ldr	r0, [sp, #S_PSR]		@ irqs are already disabled
229	msr	spsr_cxsf, r0
230#ifdef CONFIG_TRACE_IRQFLAGS
231	tst	r0, #PSR_I_BIT
232	bleq	trace_hardirqs_on
233#endif
234	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
235 UNWIND(.fnend		)
236ENDPROC(__irq_svc)
237
238	.ltorg
239
240#ifdef CONFIG_PREEMPT
241svc_preempt:
242	mov	r8, lr
2431:	bl	preempt_schedule_irq		@ irq en/disable is done inside
244	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
245	tst	r0, #_TIF_NEED_RESCHED
246	moveq	pc, r8				@ go again
247	b	1b
248#endif
249
250	.align	5
251__und_svc:
252#ifdef CONFIG_KPROBES
253	@ If a kprobe is about to simulate a "stmdb sp..." instruction,
254	@ it obviously needs free stack space which then will belong to
255	@ the saved context.
256	svc_entry 64
257#else
258	svc_entry
259#endif
260
261	@
262	@ call emulation code, which returns using r9 if it has emulated
263	@ the instruction, or the more conventional lr if we are to treat
264	@ this as a real undefined instruction
265	@
266	@  r0 - instruction
267	@
268	ldr	r0, [r2, #-4]
269	adr	r9, 1f
270	bl	call_fpe
271
272	mov	r0, sp				@ struct pt_regs *regs
273	bl	do_undefinstr
274
275	@
276	@ IRQs off again before pulling preserved data off the stack
277	@
2781:	disable_irq
279
280	@
281	@ restore SPSR and restart the instruction
282	@
283	ldr	lr, [sp, #S_PSR]		@ Get SVC cpsr
284	msr	spsr_cxsf, lr
285	ldmia	sp, {r0 - pc}^			@ Restore SVC registers
286 UNWIND(.fnend		)
287ENDPROC(__und_svc)
288
289	.align	5
290__pabt_svc:
291	svc_entry
292
293	@
294	@ re-enable interrupts if appropriate
295	@
296	mrs	r9, cpsr
297	tst	r3, #PSR_I_BIT
298	biceq	r9, r9, #PSR_I_BIT
299
300	@
301	@ set args, then call main handler
302	@
303	@  r0 - address of faulting instruction
304	@  r1 - pointer to registers on stack
305	@
306#ifdef MULTI_PABORT
307	mov	r0, r2			@ pass address of aborted instruction.
308	ldr	r4, .LCprocfns
309	mov	lr, pc
310	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
311#else
312	CPU_PABORT_HANDLER(r0, r2)
313#endif
314	msr	cpsr_c, r9			@ Maybe enable interrupts
315	mov	r1, sp				@ regs
316	bl	do_PrefetchAbort		@ call abort handler
317
318	@
319	@ IRQs off again before pulling preserved data off the stack
320	@
321	disable_irq
322
323	@
324	@ restore SPSR and restart the instruction
325	@
326	ldr	r0, [sp, #S_PSR]
327	msr	spsr_cxsf, r0
328	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
329 UNWIND(.fnend		)
330ENDPROC(__pabt_svc)
331
332	.align	5
333.LCcralign:
334	.word	cr_alignment
335#ifdef MULTI_DABORT
336.LCprocfns:
337	.word	processor
338#endif
339.LCfp:
340	.word	fp_enter
341
342/*
343 * User mode handlers
344 *
345 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
346 */
347
348#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
349#error "sizeof(struct pt_regs) must be a multiple of 8"
350#endif
351
352	.macro	usr_entry
353 UNWIND(.fnstart	)
354 UNWIND(.cantunwind	)	@ don't unwind the user space
355	sub	sp, sp, #S_FRAME_SIZE
356	stmib	sp, {r1 - r12}
357
358	ldmia	r0, {r1 - r3}
359	add	r0, sp, #S_PC		@ here for interlock avoidance
360	mov	r4, #-1			@  ""  ""     ""        ""
361
362	str	r1, [sp]		@ save the "real" r0 copied
363					@ from the exception stack
364
365	@
366	@ We are now ready to fill in the remaining blanks on the stack:
367	@
368	@  r2 - lr_<exception>, already fixed up for correct return/restart
369	@  r3 - spsr_<exception>
370	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
371	@
372	@ Also, separately save sp_usr and lr_usr
373	@
374	stmia	r0, {r2 - r4}
375	stmdb	r0, {sp, lr}^
376
377	@
378	@ Enable the alignment trap while in kernel mode
379	@
380	alignment_trap r0
381
382	@
383	@ Clear FP to mark the first stack frame
384	@
385	zero_fp
386	.endm
387
388	.macro	kuser_cmpxchg_check
389#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
390#ifndef CONFIG_MMU
391#warning "NPTL on non MMU needs fixing"
392#else
393	@ Make sure our user space atomic helper is restarted
394	@ if it was interrupted in a critical region.  Here we
395	@ perform a quick test inline since it should be false
396	@ 99.9999% of the time.  The rest is done out of line.
397	cmp	r2, #TASK_SIZE
398	blhs	kuser_cmpxchg_fixup
399#endif
400#endif
401	.endm
402
403	.align	5
404__dabt_usr:
405	usr_entry
406	kuser_cmpxchg_check
407
408	@
409	@ Call the processor-specific abort handler:
410	@
411	@  r2 - aborted context pc
412	@  r3 - aborted context cpsr
413	@
414	@ The abort handler must return the aborted address in r0, and
415	@ the fault status register in r1.
416	@
417#ifdef MULTI_DABORT
418	ldr	r4, .LCprocfns
419	mov	lr, pc
420	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
421#else
422	bl	CPU_DABORT_HANDLER
423#endif
424
425	@
426	@ IRQs on, then call the main handler
427	@
428	enable_irq
429	mov	r2, sp
430	adr	lr, ret_from_exception
431	b	do_DataAbort
432 UNWIND(.fnend		)
433ENDPROC(__dabt_usr)
434
435	.align	5
436__irq_usr:
437	usr_entry
438	kuser_cmpxchg_check
439
440#ifdef CONFIG_TRACE_IRQFLAGS
441	bl	trace_hardirqs_off
442#endif
443	get_thread_info tsk
444#ifdef CONFIG_PREEMPT
445	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
446	add	r7, r8, #1			@ increment it
447	str	r7, [tsk, #TI_PREEMPT]
448#endif
449
450	irq_handler
451#ifdef CONFIG_PREEMPT
452	ldr	r0, [tsk, #TI_PREEMPT]
453	str	r8, [tsk, #TI_PREEMPT]
454	teq	r0, r7
455	strne	r0, [r0, -r0]
456#endif
457#ifdef CONFIG_TRACE_IRQFLAGS
458	bl	trace_hardirqs_on
459#endif
460
461	mov	why, #0
462	b	ret_to_user
463 UNWIND(.fnend		)
464ENDPROC(__irq_usr)
465
466	.ltorg
467
468	.align	5
469__und_usr:
470	usr_entry
471
472	@
473	@ fall through to the emulation code, which returns using r9 if
474	@ it has emulated the instruction, or the more conventional lr
475	@ if we are to treat this as a real undefined instruction
476	@
477	@  r0 - instruction
478	@
479	adr	r9, ret_from_exception
480	adr	lr, __und_usr_unknown
481	tst	r3, #PSR_T_BIT			@ Thumb mode?
482	subeq	r4, r2, #4			@ ARM instr at LR - 4
483	subne	r4, r2, #2			@ Thumb instr at LR - 2
4841:	ldreqt	r0, [r4]
485	beq	call_fpe
486	@ Thumb instruction
487#if __LINUX_ARM_ARCH__ >= 7
4882:	ldrht	r5, [r4], #2
489	and	r0, r5, #0xf800			@ mask bits 111x x... .... ....
490	cmp	r0, #0xe800			@ 32bit instruction if xx != 0
491	blo	__und_usr_unknown
4923:	ldrht	r0, [r4]
493	add	r2, r2, #2			@ r2 is PC + 2, make it PC + 4
494	orr	r0, r0, r5, lsl #16
495#else
496	b	__und_usr_unknown
497#endif
498 UNWIND(.fnend		)
499ENDPROC(__und_usr)
500
501	@
502	@ fallthrough to call_fpe
503	@
504
505/*
506 * The out of line fixup for the ldrt above.
507 */
508	.section .fixup, "ax"
5094:	mov	pc, r9
510	.previous
511	.section __ex_table,"a"
512	.long	1b, 4b
513#if __LINUX_ARM_ARCH__ >= 7
514	.long	2b, 4b
515	.long	3b, 4b
516#endif
517	.previous
518
519/*
520 * Check whether the instruction is a co-processor instruction.
521 * If yes, we need to call the relevant co-processor handler.
522 *
523 * Note that we don't do a full check here for the co-processor
524 * instructions; all instructions with bit 27 set are well
525 * defined.  The only instructions that should fault are the
526 * co-processor instructions.  However, we have to watch out
527 * for the ARM6/ARM7 SWI bug.
528 *
529 * NEON is a special case that has to be handled here. Not all
530 * NEON instructions are co-processor instructions, so we have
531 * to make a special case of checking for them. Plus, there's
532 * five groups of them, so we have a table of mask/opcode pairs
533 * to check against, and if any match then we branch off into the
534 * NEON handler code.
535 *
536 * Emulators may wish to make use of the following registers:
537 *  r0  = instruction opcode.
538 *  r2  = PC+4
539 *  r9  = normal "successful" return address
540 *  r10 = this threads thread_info structure.
541 *  lr  = unrecognised instruction return address
542 */
543	@
544	@ Fall-through from Thumb-2 __und_usr
545	@
546#ifdef CONFIG_NEON
547	adr	r6, .LCneon_thumb_opcodes
548	b	2f
549#endif
550call_fpe:
551#ifdef CONFIG_NEON
552	adr	r6, .LCneon_arm_opcodes
5532:
554	ldr	r7, [r6], #4			@ mask value
555	cmp	r7, #0				@ end mask?
556	beq	1f
557	and	r8, r0, r7
558	ldr	r7, [r6], #4			@ opcode bits matching in mask
559	cmp	r8, r7				@ NEON instruction?
560	bne	2b
561	get_thread_info r10
562	mov	r7, #1
563	strb	r7, [r10, #TI_USED_CP + 10]	@ mark CP#10 as used
564	strb	r7, [r10, #TI_USED_CP + 11]	@ mark CP#11 as used
565	b	do_vfp				@ let VFP handler handle this
5661:
567#endif
568	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
569	tstne	r0, #0x04000000			@ bit 26 set on both ARM and Thumb-2
570#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
571	and	r8, r0, #0x0f000000		@ mask out op-code bits
572	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
573#endif
574	moveq	pc, lr
575	get_thread_info r10			@ get current thread
576	and	r8, r0, #0x00000f00		@ mask out CP number
577	mov	r7, #1
578	add	r6, r10, #TI_USED_CP
579	strb	r7, [r6, r8, lsr #8]		@ set appropriate used_cp[]
580#ifdef CONFIG_IWMMXT
581	@ Test if we need to give access to iWMMXt coprocessors
582	ldr	r5, [r10, #TI_FLAGS]
583	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
584	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
585	bcs	iwmmxt_task_enable
586#endif
587	add	pc, pc, r8, lsr #6
588	mov	r0, r0
589
590	mov	pc, lr				@ CP#0
591	b	do_fpe				@ CP#1 (FPE)
592	b	do_fpe				@ CP#2 (FPE)
593	mov	pc, lr				@ CP#3
594#ifdef CONFIG_CRUNCH
595	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
596	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
597	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
598#else
599	mov	pc, lr				@ CP#4
600	mov	pc, lr				@ CP#5
601	mov	pc, lr				@ CP#6
602#endif
603	mov	pc, lr				@ CP#7
604	mov	pc, lr				@ CP#8
605	mov	pc, lr				@ CP#9
606#ifdef CONFIG_VFP
607	b	do_vfp				@ CP#10 (VFP)
608	b	do_vfp				@ CP#11 (VFP)
609#else
610	mov	pc, lr				@ CP#10 (VFP)
611	mov	pc, lr				@ CP#11 (VFP)
612#endif
613	mov	pc, lr				@ CP#12
614	mov	pc, lr				@ CP#13
615	mov	pc, lr				@ CP#14 (Debug)
616	mov	pc, lr				@ CP#15 (Control)
617
618#ifdef CONFIG_NEON
619	.align	6
620
621.LCneon_arm_opcodes:
622	.word	0xfe000000			@ mask
623	.word	0xf2000000			@ opcode
624
625	.word	0xff100000			@ mask
626	.word	0xf4000000			@ opcode
627
628	.word	0x00000000			@ mask
629	.word	0x00000000			@ opcode
630
631.LCneon_thumb_opcodes:
632	.word	0xef000000			@ mask
633	.word	0xef000000			@ opcode
634
635	.word	0xff100000			@ mask
636	.word	0xf9000000			@ opcode
637
638	.word	0x00000000			@ mask
639	.word	0x00000000			@ opcode
640#endif
641
642do_fpe:
643	enable_irq
644	ldr	r4, .LCfp
645	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
646	ldr	pc, [r4]			@ Call FP module USR entry point
647
648/*
649 * The FP module is called with these registers set:
650 *  r0  = instruction
651 *  r2  = PC+4
652 *  r9  = normal "successful" return address
653 *  r10 = FP workspace
654 *  lr  = unrecognised FP instruction return address
655 */
656
657	.data
658ENTRY(fp_enter)
659	.word	no_fp
660	.previous
661
662no_fp:	mov	pc, lr
663
664__und_usr_unknown:
665	enable_irq
666	mov	r0, sp
667	adr	lr, ret_from_exception
668	b	do_undefinstr
669ENDPROC(__und_usr_unknown)
670
671	.align	5
672__pabt_usr:
673	usr_entry
674
675#ifdef MULTI_PABORT
676	mov	r0, r2			@ pass address of aborted instruction.
677	ldr	r4, .LCprocfns
678	mov	lr, pc
679	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
680#else
681	CPU_PABORT_HANDLER(r0, r2)
682#endif
683	enable_irq				@ Enable interrupts
684	mov	r1, sp				@ regs
685	bl	do_PrefetchAbort		@ call abort handler
686 UNWIND(.fnend		)
687	/* fall through */
688/*
689 * This is the return code to user mode for abort handlers
690 */
691ENTRY(ret_from_exception)
692 UNWIND(.fnstart	)
693 UNWIND(.cantunwind	)
694	get_thread_info tsk
695	mov	why, #0
696	b	ret_to_user
697 UNWIND(.fnend		)
698ENDPROC(__pabt_usr)
699ENDPROC(ret_from_exception)
700
701/*
702 * Register switch for ARMv3 and ARMv4 processors
703 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
704 * previous and next are guaranteed not to be the same.
705 */
706ENTRY(__switch_to)
707 UNWIND(.fnstart	)
708 UNWIND(.cantunwind	)
709	add	ip, r1, #TI_CPU_SAVE
710	ldr	r3, [r2, #TI_TP_VALUE]
711	stmia	ip!, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
712#ifdef CONFIG_MMU
713	ldr	r6, [r2, #TI_CPU_DOMAIN]
714#endif
715#if __LINUX_ARM_ARCH__ >= 6
716#ifdef CONFIG_CPU_32v6K
717	clrex
718#else
719	strex	r5, r4, [ip]			@ Clear exclusive monitor
720#endif
721#endif
722#if defined(CONFIG_HAS_TLS_REG)
723	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
724#elif !defined(CONFIG_TLS_REG_EMUL)
725	mov	r4, #0xffff0fff
726	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
727#endif
728#ifdef CONFIG_MMU
729	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
730#endif
731	mov	r5, r0
732	add	r4, r2, #TI_CPU_SAVE
733	ldr	r0, =thread_notify_head
734	mov	r1, #THREAD_NOTIFY_SWITCH
735	bl	atomic_notifier_call_chain
736	mov	r0, r5
737	ldmia	r4, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
738 UNWIND(.fnend		)
739ENDPROC(__switch_to)
740
741	__INIT
742
743/*
744 * User helpers.
745 *
746 * These are segment of kernel provided user code reachable from user space
747 * at a fixed address in kernel memory.  This is used to provide user space
748 * with some operations which require kernel help because of unimplemented
749 * native feature and/or instructions in many ARM CPUs. The idea is for
750 * this code to be executed directly in user mode for best efficiency but
751 * which is too intimate with the kernel counter part to be left to user
752 * libraries.  In fact this code might even differ from one CPU to another
753 * depending on the available  instruction set and restrictions like on
754 * SMP systems.  In other words, the kernel reserves the right to change
755 * this code as needed without warning. Only the entry points and their
756 * results are guaranteed to be stable.
757 *
758 * Each segment is 32-byte aligned and will be moved to the top of the high
759 * vector page.  New segments (if ever needed) must be added in front of
760 * existing ones.  This mechanism should be used only for things that are
761 * really small and justified, and not be abused freely.
762 *
763 * User space is expected to implement those things inline when optimizing
764 * for a processor that has the necessary native support, but only if such
765 * resulting binaries are already to be incompatible with earlier ARM
766 * processors due to the use of unsupported instructions other than what
767 * is provided here.  In other words don't make binaries unable to run on
768 * earlier processors just for the sake of not using these kernel helpers
769 * if your compiled code is not going to use the new instructions for other
770 * purpose.
771 */
772
773	.macro	usr_ret, reg
774#ifdef CONFIG_ARM_THUMB
775	bx	\reg
776#else
777	mov	pc, \reg
778#endif
779	.endm
780
781	.align	5
782	.globl	__kuser_helper_start
783__kuser_helper_start:
784
785/*
786 * Reference prototype:
787 *
788 *	void __kernel_memory_barrier(void)
789 *
790 * Input:
791 *
792 *	lr = return address
793 *
794 * Output:
795 *
796 *	none
797 *
798 * Clobbered:
799 *
800 *	none
801 *
802 * Definition and user space usage example:
803 *
804 *	typedef void (__kernel_dmb_t)(void);
805 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
806 *
807 * Apply any needed memory barrier to preserve consistency with data modified
808 * manually and __kuser_cmpxchg usage.
809 *
810 * This could be used as follows:
811 *
812 * #define __kernel_dmb() \
813 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
814 *	        : : : "r0", "lr","cc" )
815 */
816
817__kuser_memory_barrier:				@ 0xffff0fa0
818
819#if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
820	mcr	p15, 0, r0, c7, c10, 5	@ dmb
821#endif
822	usr_ret	lr
823
824	.align	5
825
826/*
827 * Reference prototype:
828 *
829 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
830 *
831 * Input:
832 *
833 *	r0 = oldval
834 *	r1 = newval
835 *	r2 = ptr
836 *	lr = return address
837 *
838 * Output:
839 *
840 *	r0 = returned value (zero or non-zero)
841 *	C flag = set if r0 == 0, clear if r0 != 0
842 *
843 * Clobbered:
844 *
845 *	r3, ip, flags
846 *
847 * Definition and user space usage example:
848 *
849 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
850 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
851 *
852 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
853 * Return zero if *ptr was changed or non-zero if no exchange happened.
854 * The C flag is also set if *ptr was changed to allow for assembly
855 * optimization in the calling code.
856 *
857 * Notes:
858 *
859 *    - This routine already includes memory barriers as needed.
860 *
861 * For example, a user space atomic_add implementation could look like this:
862 *
863 * #define atomic_add(ptr, val) \
864 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
865 *	   register unsigned int __result asm("r1"); \
866 *	   asm volatile ( \
867 *	       "1: @ atomic_add\n\t" \
868 *	       "ldr	r0, [r2]\n\t" \
869 *	       "mov	r3, #0xffff0fff\n\t" \
870 *	       "add	lr, pc, #4\n\t" \
871 *	       "add	r1, r0, %2\n\t" \
872 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
873 *	       "bcc	1b" \
874 *	       : "=&r" (__result) \
875 *	       : "r" (__ptr), "rIL" (val) \
876 *	       : "r0","r3","ip","lr","cc","memory" ); \
877 *	   __result; })
878 */
879
880__kuser_cmpxchg:				@ 0xffff0fc0
881
882#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
883
884	/*
885	 * Poor you.  No fast solution possible...
886	 * The kernel itself must perform the operation.
887	 * A special ghost syscall is used for that (see traps.c).
888	 */
889	stmfd	sp!, {r7, lr}
890	mov	r7, #0xff00		@ 0xfff0 into r7 for EABI
891	orr	r7, r7, #0xf0
892	swi	#0x9ffff0
893	ldmfd	sp!, {r7, pc}
894
895#elif __LINUX_ARM_ARCH__ < 6
896
897#ifdef CONFIG_MMU
898
899	/*
900	 * The only thing that can break atomicity in this cmpxchg
901	 * implementation is either an IRQ or a data abort exception
902	 * causing another process/thread to be scheduled in the middle
903	 * of the critical sequence.  To prevent this, code is added to
904	 * the IRQ and data abort exception handlers to set the pc back
905	 * to the beginning of the critical section if it is found to be
906	 * within that critical section (see kuser_cmpxchg_fixup).
907	 */
9081:	ldr	r3, [r2]			@ load current val
909	subs	r3, r3, r0			@ compare with oldval
9102:	streq	r1, [r2]			@ store newval if eq
911	rsbs	r0, r3, #0			@ set return val and C flag
912	usr_ret	lr
913
914	.text
915kuser_cmpxchg_fixup:
916	@ Called from kuser_cmpxchg_check macro.
917	@ r2 = address of interrupted insn (must be preserved).
918	@ sp = saved regs. r7 and r8 are clobbered.
919	@ 1b = first critical insn, 2b = last critical insn.
920	@ If r2 >= 1b and r2 <= 2b then saved pc_usr is set to 1b.
921	mov	r7, #0xffff0fff
922	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
923	subs	r8, r2, r7
924	rsbcss	r8, r8, #(2b - 1b)
925	strcs	r7, [sp, #S_PC]
926	mov	pc, lr
927	.previous
928
929#else
930#warning "NPTL on non MMU needs fixing"
931	mov	r0, #-1
932	adds	r0, r0, #0
933	usr_ret	lr
934#endif
935
936#else
937
938#ifdef CONFIG_SMP
939	mcr	p15, 0, r0, c7, c10, 5	@ dmb
940#endif
9411:	ldrex	r3, [r2]
942	subs	r3, r3, r0
943	strexeq	r3, r1, [r2]
944	teqeq	r3, #1
945	beq	1b
946	rsbs	r0, r3, #0
947	/* beware -- each __kuser slot must be 8 instructions max */
948#ifdef CONFIG_SMP
949	b	__kuser_memory_barrier
950#else
951	usr_ret	lr
952#endif
953
954#endif
955
956	.align	5
957
958/*
959 * Reference prototype:
960 *
961 *	int __kernel_get_tls(void)
962 *
963 * Input:
964 *
965 *	lr = return address
966 *
967 * Output:
968 *
969 *	r0 = TLS value
970 *
971 * Clobbered:
972 *
973 *	none
974 *
975 * Definition and user space usage example:
976 *
977 *	typedef int (__kernel_get_tls_t)(void);
978 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
979 *
980 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
981 *
982 * This could be used as follows:
983 *
984 * #define __kernel_get_tls() \
985 *	({ register unsigned int __val asm("r0"); \
986 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
987 *	        : "=r" (__val) : : "lr","cc" ); \
988 *	   __val; })
989 */
990
991__kuser_get_tls:				@ 0xffff0fe0
992
993#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
994	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
995#else
996	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
997#endif
998	usr_ret	lr
999
1000	.rep	5
1001	.word	0			@ pad up to __kuser_helper_version
1002	.endr
1003
1004/*
1005 * Reference declaration:
1006 *
1007 *	extern unsigned int __kernel_helper_version;
1008 *
1009 * Definition and user space usage example:
1010 *
1011 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
1012 *
1013 * User space may read this to determine the curent number of helpers
1014 * available.
1015 */
1016
1017__kuser_helper_version:				@ 0xffff0ffc
1018	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
1019
1020	.globl	__kuser_helper_end
1021__kuser_helper_end:
1022
1023
1024/*
1025 * Vector stubs.
1026 *
1027 * This code is copied to 0xffff0200 so we can use branches in the
1028 * vectors, rather than ldr's.  Note that this code must not
1029 * exceed 0x300 bytes.
1030 *
1031 * Common stub entry macro:
1032 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1033 *
1034 * SP points to a minimal amount of processor-private memory, the address
1035 * of which is copied into r0 for the mode specific abort handler.
1036 */
1037	.macro	vector_stub, name, mode, correction=0
1038	.align	5
1039
1040vector_\name:
1041	.if \correction
1042	sub	lr, lr, #\correction
1043	.endif
1044
1045	@
1046	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
1047	@ (parent CPSR)
1048	@
1049	stmia	sp, {r0, lr}		@ save r0, lr
1050	mrs	lr, spsr
1051	str	lr, [sp, #8]		@ save spsr
1052
1053	@
1054	@ Prepare for SVC32 mode.  IRQs remain disabled.
1055	@
1056	mrs	r0, cpsr
1057	eor	r0, r0, #(\mode ^ SVC_MODE)
1058	msr	spsr_cxsf, r0
1059
1060	@
1061	@ the branch table must immediately follow this code
1062	@
1063	and	lr, lr, #0x0f
1064	mov	r0, sp
1065	ldr	lr, [pc, lr, lsl #2]
1066	movs	pc, lr			@ branch to handler in SVC mode
1067ENDPROC(vector_\name)
1068	.endm
1069
1070	.globl	__stubs_start
1071__stubs_start:
1072/*
1073 * Interrupt dispatcher
1074 */
1075	vector_stub	irq, IRQ_MODE, 4
1076
1077	.long	__irq_usr			@  0  (USR_26 / USR_32)
1078	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
1079	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
1080	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
1081	.long	__irq_invalid			@  4
1082	.long	__irq_invalid			@  5
1083	.long	__irq_invalid			@  6
1084	.long	__irq_invalid			@  7
1085	.long	__irq_invalid			@  8
1086	.long	__irq_invalid			@  9
1087	.long	__irq_invalid			@  a
1088	.long	__irq_invalid			@  b
1089	.long	__irq_invalid			@  c
1090	.long	__irq_invalid			@  d
1091	.long	__irq_invalid			@  e
1092	.long	__irq_invalid			@  f
1093
1094/*
1095 * Data abort dispatcher
1096 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1097 */
1098	vector_stub	dabt, ABT_MODE, 8
1099
1100	.long	__dabt_usr			@  0  (USR_26 / USR_32)
1101	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
1102	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
1103	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
1104	.long	__dabt_invalid			@  4
1105	.long	__dabt_invalid			@  5
1106	.long	__dabt_invalid			@  6
1107	.long	__dabt_invalid			@  7
1108	.long	__dabt_invalid			@  8
1109	.long	__dabt_invalid			@  9
1110	.long	__dabt_invalid			@  a
1111	.long	__dabt_invalid			@  b
1112	.long	__dabt_invalid			@  c
1113	.long	__dabt_invalid			@  d
1114	.long	__dabt_invalid			@  e
1115	.long	__dabt_invalid			@  f
1116
1117/*
1118 * Prefetch abort dispatcher
1119 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1120 */
1121	vector_stub	pabt, ABT_MODE, 4
1122
1123	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1124	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1125	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1126	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1127	.long	__pabt_invalid			@  4
1128	.long	__pabt_invalid			@  5
1129	.long	__pabt_invalid			@  6
1130	.long	__pabt_invalid			@  7
1131	.long	__pabt_invalid			@  8
1132	.long	__pabt_invalid			@  9
1133	.long	__pabt_invalid			@  a
1134	.long	__pabt_invalid			@  b
1135	.long	__pabt_invalid			@  c
1136	.long	__pabt_invalid			@  d
1137	.long	__pabt_invalid			@  e
1138	.long	__pabt_invalid			@  f
1139
1140/*
1141 * Undef instr entry dispatcher
1142 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1143 */
1144	vector_stub	und, UND_MODE
1145
1146	.long	__und_usr			@  0 (USR_26 / USR_32)
1147	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1148	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1149	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1150	.long	__und_invalid			@  4
1151	.long	__und_invalid			@  5
1152	.long	__und_invalid			@  6
1153	.long	__und_invalid			@  7
1154	.long	__und_invalid			@  8
1155	.long	__und_invalid			@  9
1156	.long	__und_invalid			@  a
1157	.long	__und_invalid			@  b
1158	.long	__und_invalid			@  c
1159	.long	__und_invalid			@  d
1160	.long	__und_invalid			@  e
1161	.long	__und_invalid			@  f
1162
1163	.align	5
1164
1165/*=============================================================================
1166 * Undefined FIQs
1167 *-----------------------------------------------------------------------------
1168 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1169 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1170 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1171 * damage alert!  I don't think that we can execute any code in here in any
1172 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1173 * get out of that mode without clobbering one register.
1174 */
1175vector_fiq:
1176	disable_fiq
1177	subs	pc, lr, #4
1178
1179/*=============================================================================
1180 * Address exception handler
1181 *-----------------------------------------------------------------------------
1182 * These aren't too critical.
1183 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1184 */
1185
1186vector_addrexcptn:
1187	b	vector_addrexcptn
1188
1189/*
1190 * We group all the following data together to optimise
1191 * for CPUs with separate I & D caches.
1192 */
1193	.align	5
1194
1195.LCvswi:
1196	.word	vector_swi
1197
1198	.globl	__stubs_end
1199__stubs_end:
1200
1201	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1202
1203	.globl	__vectors_start
1204__vectors_start:
1205	swi	SYS_ERROR0
1206	b	vector_und + stubs_offset
1207	ldr	pc, .LCvswi + stubs_offset
1208	b	vector_pabt + stubs_offset
1209	b	vector_dabt + stubs_offset
1210	b	vector_addrexcptn + stubs_offset
1211	b	vector_irq + stubs_offset
1212	b	vector_fiq + stubs_offset
1213
1214	.globl	__vectors_end
1215__vectors_end:
1216
1217	.data
1218
1219	.globl	cr_alignment
1220	.globl	cr_no_alignment
1221cr_alignment:
1222	.space	4
1223cr_no_alignment:
1224	.space	4
1225