xref: /openbmc/linux/arch/arm/kernel/entry-armv.S (revision 565d76cb)
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-df.h>
20#include <asm/glue-pf.h>
21#include <asm/vfpmacros.h>
22#include <mach/entry-macro.S>
23#include <asm/thread_notify.h>
24#include <asm/unwind.h>
25#include <asm/unistd.h>
26#include <asm/tls.h>
27
28#include "entry-header.S"
29#include <asm/entry-macro-multi.S>
30
31/*
32 * Interrupt handling.  Preserves r7, r8, r9
33 */
34	.macro	irq_handler
35#ifdef CONFIG_MULTI_IRQ_HANDLER
36	ldr	r5, =handle_arch_irq
37	mov	r0, sp
38	ldr	r5, [r5]
39	adr	lr, BSYM(9997f)
40	teq	r5, #0
41	movne	pc, r5
42#endif
43	arch_irq_handler_default
449997:
45	.endm
46
47#ifdef CONFIG_KPROBES
48	.section	.kprobes.text,"ax",%progbits
49#else
50	.text
51#endif
52
53/*
54 * Invalid mode handlers
55 */
56	.macro	inv_entry, reason
57	sub	sp, sp, #S_FRAME_SIZE
58 ARM(	stmib	sp, {r1 - lr}		)
59 THUMB(	stmia	sp, {r0 - r12}		)
60 THUMB(	str	sp, [sp, #S_SP]		)
61 THUMB(	str	lr, [sp, #S_LR]		)
62	mov	r1, #\reason
63	.endm
64
65__pabt_invalid:
66	inv_entry BAD_PREFETCH
67	b	common_invalid
68ENDPROC(__pabt_invalid)
69
70__dabt_invalid:
71	inv_entry BAD_DATA
72	b	common_invalid
73ENDPROC(__dabt_invalid)
74
75__irq_invalid:
76	inv_entry BAD_IRQ
77	b	common_invalid
78ENDPROC(__irq_invalid)
79
80__und_invalid:
81	inv_entry BAD_UNDEFINSTR
82
83	@
84	@ XXX fall through to common_invalid
85	@
86
87@
88@ common_invalid - generic code for failed exception (re-entrant version of handlers)
89@
90common_invalid:
91	zero_fp
92
93	ldmia	r0, {r4 - r6}
94	add	r0, sp, #S_PC		@ here for interlock avoidance
95	mov	r7, #-1			@  ""   ""    ""        ""
96	str	r4, [sp]		@ save preserved r0
97	stmia	r0, {r5 - r7}		@ lr_<exception>,
98					@ cpsr_<exception>, "old_r0"
99
100	mov	r0, sp
101	b	bad_mode
102ENDPROC(__und_invalid)
103
104/*
105 * SVC mode handlers
106 */
107
108#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
109#define SPFIX(code...) code
110#else
111#define SPFIX(code...)
112#endif
113
114	.macro	svc_entry, stack_hole=0
115 UNWIND(.fnstart		)
116 UNWIND(.save {r0 - pc}		)
117	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole - 4)
118#ifdef CONFIG_THUMB2_KERNEL
119 SPFIX(	str	r0, [sp]	)	@ temporarily saved
120 SPFIX(	mov	r0, sp		)
121 SPFIX(	tst	r0, #4		)	@ test original stack alignment
122 SPFIX(	ldr	r0, [sp]	)	@ restored
123#else
124 SPFIX(	tst	sp, #4		)
125#endif
126 SPFIX(	subeq	sp, sp, #4	)
127	stmia	sp, {r1 - r12}
128
129	ldmia	r0, {r1 - r3}
130	add	r5, sp, #S_SP - 4	@ here for interlock avoidance
131	mov	r4, #-1			@  ""  ""      ""       ""
132	add	r0, sp, #(S_FRAME_SIZE + \stack_hole - 4)
133 SPFIX(	addeq	r0, r0, #4	)
134	str	r1, [sp, #-4]!		@ save the "real" r0 copied
135					@ from the exception stack
136
137	mov	r1, lr
138
139	@
140	@ We are now ready to fill in the remaining blanks on the stack:
141	@
142	@  r0 - sp_svc
143	@  r1 - lr_svc
144	@  r2 - lr_<exception>, already fixed up for correct return/restart
145	@  r3 - spsr_<exception>
146	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
147	@
148	stmia	r5, {r0 - r4}
149	.endm
150
151	.align	5
152__dabt_svc:
153	svc_entry
154
155	@
156	@ get ready to re-enable interrupts if appropriate
157	@
158	mrs	r9, cpsr
159	tst	r3, #PSR_I_BIT
160	biceq	r9, r9, #PSR_I_BIT
161
162	@
163	@ Call the processor-specific abort handler:
164	@
165	@  r2 - aborted context pc
166	@  r3 - aborted context cpsr
167	@
168	@ The abort handler must return the aborted address in r0, and
169	@ the fault status register in r1.  r9 must be preserved.
170	@
171#ifdef MULTI_DABORT
172	ldr	r4, .LCprocfns
173	mov	lr, pc
174	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
175#else
176	bl	CPU_DABORT_HANDLER
177#endif
178
179	@
180	@ set desired IRQ state, then call main handler
181	@
182	debug_entry r1
183	msr	cpsr_c, r9
184	mov	r2, sp
185	bl	do_DataAbort
186
187	@
188	@ IRQs off again before pulling preserved data off the stack
189	@
190	disable_irq_notrace
191
192	@
193	@ restore SPSR and restart the instruction
194	@
195	ldr	r2, [sp, #S_PSR]
196	svc_exit r2				@ return from exception
197 UNWIND(.fnend		)
198ENDPROC(__dabt_svc)
199
200	.align	5
201__irq_svc:
202	svc_entry
203
204#ifdef CONFIG_TRACE_IRQFLAGS
205	bl	trace_hardirqs_off
206#endif
207#ifdef CONFIG_PREEMPT
208	get_thread_info tsk
209	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
210	add	r7, r8, #1			@ increment it
211	str	r7, [tsk, #TI_PREEMPT]
212#endif
213
214	irq_handler
215#ifdef CONFIG_PREEMPT
216	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
217	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
218	teq	r8, #0				@ if preempt count != 0
219	movne	r0, #0				@ force flags to 0
220	tst	r0, #_TIF_NEED_RESCHED
221	blne	svc_preempt
222#endif
223	ldr	r4, [sp, #S_PSR]		@ irqs are already disabled
224#ifdef CONFIG_TRACE_IRQFLAGS
225	tst	r4, #PSR_I_BIT
226	bleq	trace_hardirqs_on
227#endif
228	svc_exit r4				@ return from exception
229 UNWIND(.fnend		)
230ENDPROC(__irq_svc)
231
232	.ltorg
233
234#ifdef CONFIG_PREEMPT
235svc_preempt:
236	mov	r8, lr
2371:	bl	preempt_schedule_irq		@ irq en/disable is done inside
238	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
239	tst	r0, #_TIF_NEED_RESCHED
240	moveq	pc, r8				@ go again
241	b	1b
242#endif
243
244	.align	5
245__und_svc:
246#ifdef CONFIG_KPROBES
247	@ If a kprobe is about to simulate a "stmdb sp..." instruction,
248	@ it obviously needs free stack space which then will belong to
249	@ the saved context.
250	svc_entry 64
251#else
252	svc_entry
253#endif
254
255	@
256	@ call emulation code, which returns using r9 if it has emulated
257	@ the instruction, or the more conventional lr if we are to treat
258	@ this as a real undefined instruction
259	@
260	@  r0 - instruction
261	@
262#ifndef	CONFIG_THUMB2_KERNEL
263	ldr	r0, [r2, #-4]
264#else
265	ldrh	r0, [r2, #-2]			@ Thumb instruction at LR - 2
266	and	r9, r0, #0xf800
267	cmp	r9, #0xe800			@ 32-bit instruction if xx >= 0
268	ldrhhs	r9, [r2]			@ bottom 16 bits
269	orrhs	r0, r9, r0, lsl #16
270#endif
271	adr	r9, BSYM(1f)
272	bl	call_fpe
273
274	mov	r0, sp				@ struct pt_regs *regs
275	bl	do_undefinstr
276
277	@
278	@ IRQs off again before pulling preserved data off the stack
279	@
2801:	disable_irq_notrace
281
282	@
283	@ restore SPSR and restart the instruction
284	@
285	ldr	r2, [sp, #S_PSR]		@ Get SVC cpsr
286	svc_exit r2				@ return from exception
287 UNWIND(.fnend		)
288ENDPROC(__und_svc)
289
290	.align	5
291__pabt_svc:
292	svc_entry
293
294	@
295	@ re-enable interrupts if appropriate
296	@
297	mrs	r9, cpsr
298	tst	r3, #PSR_I_BIT
299	biceq	r9, r9, #PSR_I_BIT
300
301	mov	r0, r2			@ pass address of aborted instruction.
302#ifdef MULTI_PABORT
303	ldr	r4, .LCprocfns
304	mov	lr, pc
305	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
306#else
307	bl	CPU_PABORT_HANDLER
308#endif
309	debug_entry r1
310	msr	cpsr_c, r9			@ Maybe enable interrupts
311	mov	r2, sp				@ regs
312	bl	do_PrefetchAbort		@ call abort handler
313
314	@
315	@ IRQs off again before pulling preserved data off the stack
316	@
317	disable_irq_notrace
318
319	@
320	@ restore SPSR and restart the instruction
321	@
322	ldr	r2, [sp, #S_PSR]
323	svc_exit r2				@ return from exception
324 UNWIND(.fnend		)
325ENDPROC(__pabt_svc)
326
327	.align	5
328.LCcralign:
329	.word	cr_alignment
330#ifdef MULTI_DABORT
331.LCprocfns:
332	.word	processor
333#endif
334.LCfp:
335	.word	fp_enter
336
337/*
338 * User mode handlers
339 *
340 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
341 */
342
343#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
344#error "sizeof(struct pt_regs) must be a multiple of 8"
345#endif
346
347	.macro	usr_entry
348 UNWIND(.fnstart	)
349 UNWIND(.cantunwind	)	@ don't unwind the user space
350	sub	sp, sp, #S_FRAME_SIZE
351 ARM(	stmib	sp, {r1 - r12}	)
352 THUMB(	stmia	sp, {r0 - r12}	)
353
354	ldmia	r0, {r1 - r3}
355	add	r0, sp, #S_PC		@ here for interlock avoidance
356	mov	r4, #-1			@  ""  ""     ""        ""
357
358	str	r1, [sp]		@ save the "real" r0 copied
359					@ from the exception stack
360
361	@
362	@ We are now ready to fill in the remaining blanks on the stack:
363	@
364	@  r2 - lr_<exception>, already fixed up for correct return/restart
365	@  r3 - spsr_<exception>
366	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
367	@
368	@ Also, separately save sp_usr and lr_usr
369	@
370	stmia	r0, {r2 - r4}
371 ARM(	stmdb	r0, {sp, lr}^			)
372 THUMB(	store_user_sp_lr r0, r1, S_SP - S_PC	)
373
374	@
375	@ Enable the alignment trap while in kernel mode
376	@
377	alignment_trap r0
378
379	@
380	@ Clear FP to mark the first stack frame
381	@
382	zero_fp
383	.endm
384
385	.macro	kuser_cmpxchg_check
386#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
387#ifndef CONFIG_MMU
388#warning "NPTL on non MMU needs fixing"
389#else
390	@ Make sure our user space atomic helper is restarted
391	@ if it was interrupted in a critical region.  Here we
392	@ perform a quick test inline since it should be false
393	@ 99.9999% of the time.  The rest is done out of line.
394	cmp	r2, #TASK_SIZE
395	blhs	kuser_cmpxchg_fixup
396#endif
397#endif
398	.endm
399
400	.align	5
401__dabt_usr:
402	usr_entry
403	kuser_cmpxchg_check
404
405	@
406	@ Call the processor-specific abort handler:
407	@
408	@  r2 - aborted context pc
409	@  r3 - aborted context cpsr
410	@
411	@ The abort handler must return the aborted address in r0, and
412	@ the fault status register in r1.
413	@
414#ifdef MULTI_DABORT
415	ldr	r4, .LCprocfns
416	mov	lr, pc
417	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
418#else
419	bl	CPU_DABORT_HANDLER
420#endif
421
422	@
423	@ IRQs on, then call the main handler
424	@
425	debug_entry r1
426	enable_irq
427	mov	r2, sp
428	adr	lr, BSYM(ret_from_exception)
429	b	do_DataAbort
430 UNWIND(.fnend		)
431ENDPROC(__dabt_usr)
432
433	.align	5
434__irq_usr:
435	usr_entry
436	kuser_cmpxchg_check
437
438	get_thread_info tsk
439#ifdef CONFIG_PREEMPT
440	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
441	add	r7, r8, #1			@ increment it
442	str	r7, [tsk, #TI_PREEMPT]
443#endif
444
445	irq_handler
446#ifdef CONFIG_PREEMPT
447	ldr	r0, [tsk, #TI_PREEMPT]
448	str	r8, [tsk, #TI_PREEMPT]
449	teq	r0, r7
450 ARM(	strne	r0, [r0, -r0]	)
451 THUMB(	movne	r0, #0		)
452 THUMB(	strne	r0, [r0]	)
453#endif
454
455	mov	why, #0
456	b	ret_to_user
457 UNWIND(.fnend		)
458ENDPROC(__irq_usr)
459
460	.ltorg
461
462	.align	5
463__und_usr:
464	usr_entry
465
466	@
467	@ fall through to the emulation code, which returns using r9 if
468	@ it has emulated the instruction, or the more conventional lr
469	@ if we are to treat this as a real undefined instruction
470	@
471	@  r0 - instruction
472	@
473	adr	r9, BSYM(ret_from_exception)
474	adr	lr, BSYM(__und_usr_unknown)
475	tst	r3, #PSR_T_BIT			@ Thumb mode?
476	itet	eq				@ explicit IT needed for the 1f label
477	subeq	r4, r2, #4			@ ARM instr at LR - 4
478	subne	r4, r2, #2			@ Thumb instr at LR - 2
4791:	ldreqt	r0, [r4]
480#ifdef CONFIG_CPU_ENDIAN_BE8
481	reveq	r0, r0				@ little endian instruction
482#endif
483	beq	call_fpe
484	@ Thumb instruction
485#if __LINUX_ARM_ARCH__ >= 7
4862:
487 ARM(	ldrht	r5, [r4], #2	)
488 THUMB(	ldrht	r5, [r4]	)
489 THUMB(	add	r4, r4, #2	)
490	and	r0, r5, #0xf800			@ mask bits 111x x... .... ....
491	cmp	r0, #0xe800			@ 32bit instruction if xx != 0
492	blo	__und_usr_unknown
4933:	ldrht	r0, [r4]
494	add	r2, r2, #2			@ r2 is PC + 2, make it PC + 4
495	orr	r0, r0, r5, lsl #16
496#else
497	b	__und_usr_unknown
498#endif
499 UNWIND(.fnend		)
500ENDPROC(__und_usr)
501
502	@
503	@ fallthrough to call_fpe
504	@
505
506/*
507 * The out of line fixup for the ldrt above.
508 */
509	.pushsection .fixup, "ax"
5104:	mov	pc, r9
511	.popsection
512	.pushsection __ex_table,"a"
513	.long	1b, 4b
514#if __LINUX_ARM_ARCH__ >= 7
515	.long	2b, 4b
516	.long	3b, 4b
517#endif
518	.popsection
519
520/*
521 * Check whether the instruction is a co-processor instruction.
522 * If yes, we need to call the relevant co-processor handler.
523 *
524 * Note that we don't do a full check here for the co-processor
525 * instructions; all instructions with bit 27 set are well
526 * defined.  The only instructions that should fault are the
527 * co-processor instructions.  However, we have to watch out
528 * for the ARM6/ARM7 SWI bug.
529 *
530 * NEON is a special case that has to be handled here. Not all
531 * NEON instructions are co-processor instructions, so we have
532 * to make a special case of checking for them. Plus, there's
533 * five groups of them, so we have a table of mask/opcode pairs
534 * to check against, and if any match then we branch off into the
535 * NEON handler code.
536 *
537 * Emulators may wish to make use of the following registers:
538 *  r0  = instruction opcode.
539 *  r2  = PC+4
540 *  r9  = normal "successful" return address
541 *  r10 = this threads thread_info structure.
542 *  lr  = unrecognised instruction return address
543 */
544	@
545	@ Fall-through from Thumb-2 __und_usr
546	@
547#ifdef CONFIG_NEON
548	adr	r6, .LCneon_thumb_opcodes
549	b	2f
550#endif
551call_fpe:
552#ifdef CONFIG_NEON
553	adr	r6, .LCneon_arm_opcodes
5542:
555	ldr	r7, [r6], #4			@ mask value
556	cmp	r7, #0				@ end mask?
557	beq	1f
558	and	r8, r0, r7
559	ldr	r7, [r6], #4			@ opcode bits matching in mask
560	cmp	r8, r7				@ NEON instruction?
561	bne	2b
562	get_thread_info r10
563	mov	r7, #1
564	strb	r7, [r10, #TI_USED_CP + 10]	@ mark CP#10 as used
565	strb	r7, [r10, #TI_USED_CP + 11]	@ mark CP#11 as used
566	b	do_vfp				@ let VFP handler handle this
5671:
568#endif
569	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
570	tstne	r0, #0x04000000			@ bit 26 set on both ARM and Thumb-2
571#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
572	and	r8, r0, #0x0f000000		@ mask out op-code bits
573	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
574#endif
575	moveq	pc, lr
576	get_thread_info r10			@ get current thread
577	and	r8, r0, #0x00000f00		@ mask out CP number
578 THUMB(	lsr	r8, r8, #8		)
579	mov	r7, #1
580	add	r6, r10, #TI_USED_CP
581 ARM(	strb	r7, [r6, r8, lsr #8]	)	@ set appropriate used_cp[]
582 THUMB(	strb	r7, [r6, r8]		)	@ set appropriate used_cp[]
583#ifdef CONFIG_IWMMXT
584	@ Test if we need to give access to iWMMXt coprocessors
585	ldr	r5, [r10, #TI_FLAGS]
586	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
587	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
588	bcs	iwmmxt_task_enable
589#endif
590 ARM(	add	pc, pc, r8, lsr #6	)
591 THUMB(	lsl	r8, r8, #2		)
592 THUMB(	add	pc, r8			)
593	nop
594
595	movw_pc	lr				@ CP#0
596	W(b)	do_fpe				@ CP#1 (FPE)
597	W(b)	do_fpe				@ CP#2 (FPE)
598	movw_pc	lr				@ CP#3
599#ifdef CONFIG_CRUNCH
600	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
601	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
602	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
603#else
604	movw_pc	lr				@ CP#4
605	movw_pc	lr				@ CP#5
606	movw_pc	lr				@ CP#6
607#endif
608	movw_pc	lr				@ CP#7
609	movw_pc	lr				@ CP#8
610	movw_pc	lr				@ CP#9
611#ifdef CONFIG_VFP
612	W(b)	do_vfp				@ CP#10 (VFP)
613	W(b)	do_vfp				@ CP#11 (VFP)
614#else
615	movw_pc	lr				@ CP#10 (VFP)
616	movw_pc	lr				@ CP#11 (VFP)
617#endif
618	movw_pc	lr				@ CP#12
619	movw_pc	lr				@ CP#13
620	movw_pc	lr				@ CP#14 (Debug)
621	movw_pc	lr				@ CP#15 (Control)
622
623#ifdef CONFIG_NEON
624	.align	6
625
626.LCneon_arm_opcodes:
627	.word	0xfe000000			@ mask
628	.word	0xf2000000			@ opcode
629
630	.word	0xff100000			@ mask
631	.word	0xf4000000			@ opcode
632
633	.word	0x00000000			@ mask
634	.word	0x00000000			@ opcode
635
636.LCneon_thumb_opcodes:
637	.word	0xef000000			@ mask
638	.word	0xef000000			@ opcode
639
640	.word	0xff100000			@ mask
641	.word	0xf9000000			@ opcode
642
643	.word	0x00000000			@ mask
644	.word	0x00000000			@ opcode
645#endif
646
647do_fpe:
648	enable_irq
649	ldr	r4, .LCfp
650	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
651	ldr	pc, [r4]			@ Call FP module USR entry point
652
653/*
654 * The FP module is called with these registers set:
655 *  r0  = instruction
656 *  r2  = PC+4
657 *  r9  = normal "successful" return address
658 *  r10 = FP workspace
659 *  lr  = unrecognised FP instruction return address
660 */
661
662	.pushsection .data
663ENTRY(fp_enter)
664	.word	no_fp
665	.popsection
666
667ENTRY(no_fp)
668	mov	pc, lr
669ENDPROC(no_fp)
670
671__und_usr_unknown:
672	enable_irq
673	mov	r0, sp
674	adr	lr, BSYM(ret_from_exception)
675	b	do_undefinstr
676ENDPROC(__und_usr_unknown)
677
678	.align	5
679__pabt_usr:
680	usr_entry
681
682	mov	r0, r2			@ pass address of aborted instruction.
683#ifdef MULTI_PABORT
684	ldr	r4, .LCprocfns
685	mov	lr, pc
686	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
687#else
688	bl	CPU_PABORT_HANDLER
689#endif
690	debug_entry r1
691	enable_irq				@ Enable interrupts
692	mov	r2, sp				@ regs
693	bl	do_PrefetchAbort		@ call abort handler
694 UNWIND(.fnend		)
695	/* fall through */
696/*
697 * This is the return code to user mode for abort handlers
698 */
699ENTRY(ret_from_exception)
700 UNWIND(.fnstart	)
701 UNWIND(.cantunwind	)
702	get_thread_info tsk
703	mov	why, #0
704	b	ret_to_user
705 UNWIND(.fnend		)
706ENDPROC(__pabt_usr)
707ENDPROC(ret_from_exception)
708
709/*
710 * Register switch for ARMv3 and ARMv4 processors
711 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
712 * previous and next are guaranteed not to be the same.
713 */
714ENTRY(__switch_to)
715 UNWIND(.fnstart	)
716 UNWIND(.cantunwind	)
717	add	ip, r1, #TI_CPU_SAVE
718	ldr	r3, [r2, #TI_TP_VALUE]
719 ARM(	stmia	ip!, {r4 - sl, fp, sp, lr} )	@ Store most regs on stack
720 THUMB(	stmia	ip!, {r4 - sl, fp}	   )	@ Store most regs on stack
721 THUMB(	str	sp, [ip], #4		   )
722 THUMB(	str	lr, [ip], #4		   )
723#ifdef CONFIG_CPU_USE_DOMAINS
724	ldr	r6, [r2, #TI_CPU_DOMAIN]
725#endif
726	set_tls	r3, r4, r5
727#if defined(CONFIG_CC_STACKPROTECTOR) && !defined(CONFIG_SMP)
728	ldr	r7, [r2, #TI_TASK]
729	ldr	r8, =__stack_chk_guard
730	ldr	r7, [r7, #TSK_STACK_CANARY]
731#endif
732#ifdef CONFIG_CPU_USE_DOMAINS
733	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
734#endif
735	mov	r5, r0
736	add	r4, r2, #TI_CPU_SAVE
737	ldr	r0, =thread_notify_head
738	mov	r1, #THREAD_NOTIFY_SWITCH
739	bl	atomic_notifier_call_chain
740#if defined(CONFIG_CC_STACKPROTECTOR) && !defined(CONFIG_SMP)
741	str	r7, [r8]
742#endif
743 THUMB(	mov	ip, r4			   )
744	mov	r0, r5
745 ARM(	ldmia	r4, {r4 - sl, fp, sp, pc}  )	@ Load all regs saved previously
746 THUMB(	ldmia	ip!, {r4 - sl, fp}	   )	@ Load all regs saved previously
747 THUMB(	ldr	sp, [ip], #4		   )
748 THUMB(	ldr	pc, [ip]		   )
749 UNWIND(.fnend		)
750ENDPROC(__switch_to)
751
752	__INIT
753
754/*
755 * User helpers.
756 *
757 * These are segment of kernel provided user code reachable from user space
758 * at a fixed address in kernel memory.  This is used to provide user space
759 * with some operations which require kernel help because of unimplemented
760 * native feature and/or instructions in many ARM CPUs. The idea is for
761 * this code to be executed directly in user mode for best efficiency but
762 * which is too intimate with the kernel counter part to be left to user
763 * libraries.  In fact this code might even differ from one CPU to another
764 * depending on the available  instruction set and restrictions like on
765 * SMP systems.  In other words, the kernel reserves the right to change
766 * this code as needed without warning. Only the entry points and their
767 * results are guaranteed to be stable.
768 *
769 * Each segment is 32-byte aligned and will be moved to the top of the high
770 * vector page.  New segments (if ever needed) must be added in front of
771 * existing ones.  This mechanism should be used only for things that are
772 * really small and justified, and not be abused freely.
773 *
774 * User space is expected to implement those things inline when optimizing
775 * for a processor that has the necessary native support, but only if such
776 * resulting binaries are already to be incompatible with earlier ARM
777 * processors due to the use of unsupported instructions other than what
778 * is provided here.  In other words don't make binaries unable to run on
779 * earlier processors just for the sake of not using these kernel helpers
780 * if your compiled code is not going to use the new instructions for other
781 * purpose.
782 */
783 THUMB(	.arm	)
784
785	.macro	usr_ret, reg
786#ifdef CONFIG_ARM_THUMB
787	bx	\reg
788#else
789	mov	pc, \reg
790#endif
791	.endm
792
793	.align	5
794	.globl	__kuser_helper_start
795__kuser_helper_start:
796
797/*
798 * Reference prototype:
799 *
800 *	void __kernel_memory_barrier(void)
801 *
802 * Input:
803 *
804 *	lr = return address
805 *
806 * Output:
807 *
808 *	none
809 *
810 * Clobbered:
811 *
812 *	none
813 *
814 * Definition and user space usage example:
815 *
816 *	typedef void (__kernel_dmb_t)(void);
817 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
818 *
819 * Apply any needed memory barrier to preserve consistency with data modified
820 * manually and __kuser_cmpxchg usage.
821 *
822 * This could be used as follows:
823 *
824 * #define __kernel_dmb() \
825 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
826 *	        : : : "r0", "lr","cc" )
827 */
828
829__kuser_memory_barrier:				@ 0xffff0fa0
830	smp_dmb	arm
831	usr_ret	lr
832
833	.align	5
834
835/*
836 * Reference prototype:
837 *
838 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
839 *
840 * Input:
841 *
842 *	r0 = oldval
843 *	r1 = newval
844 *	r2 = ptr
845 *	lr = return address
846 *
847 * Output:
848 *
849 *	r0 = returned value (zero or non-zero)
850 *	C flag = set if r0 == 0, clear if r0 != 0
851 *
852 * Clobbered:
853 *
854 *	r3, ip, flags
855 *
856 * Definition and user space usage example:
857 *
858 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
859 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
860 *
861 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
862 * Return zero if *ptr was changed or non-zero if no exchange happened.
863 * The C flag is also set if *ptr was changed to allow for assembly
864 * optimization in the calling code.
865 *
866 * Notes:
867 *
868 *    - This routine already includes memory barriers as needed.
869 *
870 * For example, a user space atomic_add implementation could look like this:
871 *
872 * #define atomic_add(ptr, val) \
873 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
874 *	   register unsigned int __result asm("r1"); \
875 *	   asm volatile ( \
876 *	       "1: @ atomic_add\n\t" \
877 *	       "ldr	r0, [r2]\n\t" \
878 *	       "mov	r3, #0xffff0fff\n\t" \
879 *	       "add	lr, pc, #4\n\t" \
880 *	       "add	r1, r0, %2\n\t" \
881 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
882 *	       "bcc	1b" \
883 *	       : "=&r" (__result) \
884 *	       : "r" (__ptr), "rIL" (val) \
885 *	       : "r0","r3","ip","lr","cc","memory" ); \
886 *	   __result; })
887 */
888
889__kuser_cmpxchg:				@ 0xffff0fc0
890
891#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
892
893	/*
894	 * Poor you.  No fast solution possible...
895	 * The kernel itself must perform the operation.
896	 * A special ghost syscall is used for that (see traps.c).
897	 */
898	stmfd	sp!, {r7, lr}
899	ldr	r7, 1f			@ it's 20 bits
900	swi	__ARM_NR_cmpxchg
901	ldmfd	sp!, {r7, pc}
9021:	.word	__ARM_NR_cmpxchg
903
904#elif __LINUX_ARM_ARCH__ < 6
905
906#ifdef CONFIG_MMU
907
908	/*
909	 * The only thing that can break atomicity in this cmpxchg
910	 * implementation is either an IRQ or a data abort exception
911	 * causing another process/thread to be scheduled in the middle
912	 * of the critical sequence.  To prevent this, code is added to
913	 * the IRQ and data abort exception handlers to set the pc back
914	 * to the beginning of the critical section if it is found to be
915	 * within that critical section (see kuser_cmpxchg_fixup).
916	 */
9171:	ldr	r3, [r2]			@ load current val
918	subs	r3, r3, r0			@ compare with oldval
9192:	streq	r1, [r2]			@ store newval if eq
920	rsbs	r0, r3, #0			@ set return val and C flag
921	usr_ret	lr
922
923	.text
924kuser_cmpxchg_fixup:
925	@ Called from kuser_cmpxchg_check macro.
926	@ r2 = address of interrupted insn (must be preserved).
927	@ sp = saved regs. r7 and r8 are clobbered.
928	@ 1b = first critical insn, 2b = last critical insn.
929	@ If r2 >= 1b and r2 <= 2b then saved pc_usr is set to 1b.
930	mov	r7, #0xffff0fff
931	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
932	subs	r8, r2, r7
933	rsbcss	r8, r8, #(2b - 1b)
934	strcs	r7, [sp, #S_PC]
935	mov	pc, lr
936	.previous
937
938#else
939#warning "NPTL on non MMU needs fixing"
940	mov	r0, #-1
941	adds	r0, r0, #0
942	usr_ret	lr
943#endif
944
945#else
946
947	smp_dmb	arm
9481:	ldrex	r3, [r2]
949	subs	r3, r3, r0
950	strexeq	r3, r1, [r2]
951	teqeq	r3, #1
952	beq	1b
953	rsbs	r0, r3, #0
954	/* beware -- each __kuser slot must be 8 instructions max */
955	ALT_SMP(b	__kuser_memory_barrier)
956	ALT_UP(usr_ret	lr)
957
958#endif
959
960	.align	5
961
962/*
963 * Reference prototype:
964 *
965 *	int __kernel_get_tls(void)
966 *
967 * Input:
968 *
969 *	lr = return address
970 *
971 * Output:
972 *
973 *	r0 = TLS value
974 *
975 * Clobbered:
976 *
977 *	none
978 *
979 * Definition and user space usage example:
980 *
981 *	typedef int (__kernel_get_tls_t)(void);
982 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
983 *
984 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
985 *
986 * This could be used as follows:
987 *
988 * #define __kernel_get_tls() \
989 *	({ register unsigned int __val asm("r0"); \
990 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
991 *	        : "=r" (__val) : : "lr","cc" ); \
992 *	   __val; })
993 */
994
995__kuser_get_tls:				@ 0xffff0fe0
996	ldr	r0, [pc, #(16 - 8)]	@ read TLS, set in kuser_get_tls_init
997	usr_ret	lr
998	mrc	p15, 0, r0, c13, c0, 3	@ 0xffff0fe8 hardware TLS code
999	.rep	4
1000	.word	0			@ 0xffff0ff0 software TLS value, then
1001	.endr				@ pad up to __kuser_helper_version
1002
1003/*
1004 * Reference declaration:
1005 *
1006 *	extern unsigned int __kernel_helper_version;
1007 *
1008 * Definition and user space usage example:
1009 *
1010 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
1011 *
1012 * User space may read this to determine the curent number of helpers
1013 * available.
1014 */
1015
1016__kuser_helper_version:				@ 0xffff0ffc
1017	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
1018
1019	.globl	__kuser_helper_end
1020__kuser_helper_end:
1021
1022 THUMB(	.thumb	)
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 | PSR_ISETSTATE)
1058	msr	spsr_cxsf, r0
1059
1060	@
1061	@ the branch table must immediately follow this code
1062	@
1063	and	lr, lr, #0x0f
1064 THUMB(	adr	r0, 1f			)
1065 THUMB(	ldr	lr, [r0, lr, lsl #2]	)
1066	mov	r0, sp
1067 ARM(	ldr	lr, [pc, lr, lsl #2]	)
1068	movs	pc, lr			@ branch to handler in SVC mode
1069ENDPROC(vector_\name)
1070
1071	.align	2
1072	@ handler addresses follow this label
10731:
1074	.endm
1075
1076	.globl	__stubs_start
1077__stubs_start:
1078/*
1079 * Interrupt dispatcher
1080 */
1081	vector_stub	irq, IRQ_MODE, 4
1082
1083	.long	__irq_usr			@  0  (USR_26 / USR_32)
1084	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
1085	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
1086	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
1087	.long	__irq_invalid			@  4
1088	.long	__irq_invalid			@  5
1089	.long	__irq_invalid			@  6
1090	.long	__irq_invalid			@  7
1091	.long	__irq_invalid			@  8
1092	.long	__irq_invalid			@  9
1093	.long	__irq_invalid			@  a
1094	.long	__irq_invalid			@  b
1095	.long	__irq_invalid			@  c
1096	.long	__irq_invalid			@  d
1097	.long	__irq_invalid			@  e
1098	.long	__irq_invalid			@  f
1099
1100/*
1101 * Data abort dispatcher
1102 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1103 */
1104	vector_stub	dabt, ABT_MODE, 8
1105
1106	.long	__dabt_usr			@  0  (USR_26 / USR_32)
1107	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
1108	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
1109	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
1110	.long	__dabt_invalid			@  4
1111	.long	__dabt_invalid			@  5
1112	.long	__dabt_invalid			@  6
1113	.long	__dabt_invalid			@  7
1114	.long	__dabt_invalid			@  8
1115	.long	__dabt_invalid			@  9
1116	.long	__dabt_invalid			@  a
1117	.long	__dabt_invalid			@  b
1118	.long	__dabt_invalid			@  c
1119	.long	__dabt_invalid			@  d
1120	.long	__dabt_invalid			@  e
1121	.long	__dabt_invalid			@  f
1122
1123/*
1124 * Prefetch abort dispatcher
1125 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1126 */
1127	vector_stub	pabt, ABT_MODE, 4
1128
1129	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1130	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1131	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1132	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1133	.long	__pabt_invalid			@  4
1134	.long	__pabt_invalid			@  5
1135	.long	__pabt_invalid			@  6
1136	.long	__pabt_invalid			@  7
1137	.long	__pabt_invalid			@  8
1138	.long	__pabt_invalid			@  9
1139	.long	__pabt_invalid			@  a
1140	.long	__pabt_invalid			@  b
1141	.long	__pabt_invalid			@  c
1142	.long	__pabt_invalid			@  d
1143	.long	__pabt_invalid			@  e
1144	.long	__pabt_invalid			@  f
1145
1146/*
1147 * Undef instr entry dispatcher
1148 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1149 */
1150	vector_stub	und, UND_MODE
1151
1152	.long	__und_usr			@  0 (USR_26 / USR_32)
1153	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1154	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1155	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1156	.long	__und_invalid			@  4
1157	.long	__und_invalid			@  5
1158	.long	__und_invalid			@  6
1159	.long	__und_invalid			@  7
1160	.long	__und_invalid			@  8
1161	.long	__und_invalid			@  9
1162	.long	__und_invalid			@  a
1163	.long	__und_invalid			@  b
1164	.long	__und_invalid			@  c
1165	.long	__und_invalid			@  d
1166	.long	__und_invalid			@  e
1167	.long	__und_invalid			@  f
1168
1169	.align	5
1170
1171/*=============================================================================
1172 * Undefined FIQs
1173 *-----------------------------------------------------------------------------
1174 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1175 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1176 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1177 * damage alert!  I don't think that we can execute any code in here in any
1178 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1179 * get out of that mode without clobbering one register.
1180 */
1181vector_fiq:
1182	disable_fiq
1183	subs	pc, lr, #4
1184
1185/*=============================================================================
1186 * Address exception handler
1187 *-----------------------------------------------------------------------------
1188 * These aren't too critical.
1189 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1190 */
1191
1192vector_addrexcptn:
1193	b	vector_addrexcptn
1194
1195/*
1196 * We group all the following data together to optimise
1197 * for CPUs with separate I & D caches.
1198 */
1199	.align	5
1200
1201.LCvswi:
1202	.word	vector_swi
1203
1204	.globl	__stubs_end
1205__stubs_end:
1206
1207	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1208
1209	.globl	__vectors_start
1210__vectors_start:
1211 ARM(	swi	SYS_ERROR0	)
1212 THUMB(	svc	#0		)
1213 THUMB(	nop			)
1214	W(b)	vector_und + stubs_offset
1215	W(ldr)	pc, .LCvswi + stubs_offset
1216	W(b)	vector_pabt + stubs_offset
1217	W(b)	vector_dabt + stubs_offset
1218	W(b)	vector_addrexcptn + stubs_offset
1219	W(b)	vector_irq + stubs_offset
1220	W(b)	vector_fiq + stubs_offset
1221
1222	.globl	__vectors_end
1223__vectors_end:
1224
1225	.data
1226
1227	.globl	cr_alignment
1228	.globl	cr_no_alignment
1229cr_alignment:
1230	.space	4
1231cr_no_alignment:
1232	.space	4
1233
1234#ifdef CONFIG_MULTI_IRQ_HANDLER
1235	.globl	handle_arch_irq
1236handle_arch_irq:
1237	.space	4
1238#endif
1239