xref: /openbmc/linux/arch/arm/kernel/entry-armv.S (revision 615c36f5)
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#include <asm/system.h>
28
29#include "entry-header.S"
30#include <asm/entry-macro-multi.S>
31
32/*
33 * Interrupt handling.
34 */
35	.macro	irq_handler
36#ifdef CONFIG_MULTI_IRQ_HANDLER
37	ldr	r1, =handle_arch_irq
38	mov	r0, sp
39	ldr	r1, [r1]
40	adr	lr, BSYM(9997f)
41	teq	r1, #0
42	movne	pc, r1
43#endif
44	arch_irq_handler_default
459997:
46	.endm
47
48	.macro	pabt_helper
49	@ PABORT handler takes pt_regs in r2, fault address in r4 and psr in r5
50#ifdef MULTI_PABORT
51	ldr	ip, .LCprocfns
52	mov	lr, pc
53	ldr	pc, [ip, #PROCESSOR_PABT_FUNC]
54#else
55	bl	CPU_PABORT_HANDLER
56#endif
57	.endm
58
59	.macro	dabt_helper
60
61	@
62	@ Call the processor-specific abort handler:
63	@
64	@  r2 - pt_regs
65	@  r4 - aborted context pc
66	@  r5 - aborted context psr
67	@
68	@ The abort handler must return the aborted address in r0, and
69	@ the fault status register in r1.  r9 must be preserved.
70	@
71#ifdef MULTI_DABORT
72	ldr	ip, .LCprocfns
73	mov	lr, pc
74	ldr	pc, [ip, #PROCESSOR_DABT_FUNC]
75#else
76	bl	CPU_DABORT_HANDLER
77#endif
78	.endm
79
80#ifdef CONFIG_KPROBES
81	.section	.kprobes.text,"ax",%progbits
82#else
83	.text
84#endif
85
86/*
87 * Invalid mode handlers
88 */
89	.macro	inv_entry, reason
90	sub	sp, sp, #S_FRAME_SIZE
91 ARM(	stmib	sp, {r1 - lr}		)
92 THUMB(	stmia	sp, {r0 - r12}		)
93 THUMB(	str	sp, [sp, #S_SP]		)
94 THUMB(	str	lr, [sp, #S_LR]		)
95	mov	r1, #\reason
96	.endm
97
98__pabt_invalid:
99	inv_entry BAD_PREFETCH
100	b	common_invalid
101ENDPROC(__pabt_invalid)
102
103__dabt_invalid:
104	inv_entry BAD_DATA
105	b	common_invalid
106ENDPROC(__dabt_invalid)
107
108__irq_invalid:
109	inv_entry BAD_IRQ
110	b	common_invalid
111ENDPROC(__irq_invalid)
112
113__und_invalid:
114	inv_entry BAD_UNDEFINSTR
115
116	@
117	@ XXX fall through to common_invalid
118	@
119
120@
121@ common_invalid - generic code for failed exception (re-entrant version of handlers)
122@
123common_invalid:
124	zero_fp
125
126	ldmia	r0, {r4 - r6}
127	add	r0, sp, #S_PC		@ here for interlock avoidance
128	mov	r7, #-1			@  ""   ""    ""        ""
129	str	r4, [sp]		@ save preserved r0
130	stmia	r0, {r5 - r7}		@ lr_<exception>,
131					@ cpsr_<exception>, "old_r0"
132
133	mov	r0, sp
134	b	bad_mode
135ENDPROC(__und_invalid)
136
137/*
138 * SVC mode handlers
139 */
140
141#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
142#define SPFIX(code...) code
143#else
144#define SPFIX(code...)
145#endif
146
147	.macro	svc_entry, stack_hole=0
148 UNWIND(.fnstart		)
149 UNWIND(.save {r0 - pc}		)
150	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole - 4)
151#ifdef CONFIG_THUMB2_KERNEL
152 SPFIX(	str	r0, [sp]	)	@ temporarily saved
153 SPFIX(	mov	r0, sp		)
154 SPFIX(	tst	r0, #4		)	@ test original stack alignment
155 SPFIX(	ldr	r0, [sp]	)	@ restored
156#else
157 SPFIX(	tst	sp, #4		)
158#endif
159 SPFIX(	subeq	sp, sp, #4	)
160	stmia	sp, {r1 - r12}
161
162	ldmia	r0, {r3 - r5}
163	add	r7, sp, #S_SP - 4	@ here for interlock avoidance
164	mov	r6, #-1			@  ""  ""      ""       ""
165	add	r2, sp, #(S_FRAME_SIZE + \stack_hole - 4)
166 SPFIX(	addeq	r2, r2, #4	)
167	str	r3, [sp, #-4]!		@ save the "real" r0 copied
168					@ from the exception stack
169
170	mov	r3, lr
171
172	@
173	@ We are now ready to fill in the remaining blanks on the stack:
174	@
175	@  r2 - sp_svc
176	@  r3 - lr_svc
177	@  r4 - lr_<exception>, already fixed up for correct return/restart
178	@  r5 - spsr_<exception>
179	@  r6 - orig_r0 (see pt_regs definition in ptrace.h)
180	@
181	stmia	r7, {r2 - r6}
182
183#ifdef CONFIG_TRACE_IRQFLAGS
184	bl	trace_hardirqs_off
185#endif
186	.endm
187
188	.align	5
189__dabt_svc:
190	svc_entry
191	mov	r2, sp
192	dabt_helper
193
194	@
195	@ IRQs off again before pulling preserved data off the stack
196	@
197	disable_irq_notrace
198
199#ifdef CONFIG_TRACE_IRQFLAGS
200	tst	r5, #PSR_I_BIT
201	bleq	trace_hardirqs_on
202	tst	r5, #PSR_I_BIT
203	blne	trace_hardirqs_off
204#endif
205	svc_exit r5				@ return from exception
206 UNWIND(.fnend		)
207ENDPROC(__dabt_svc)
208
209	.align	5
210__irq_svc:
211	svc_entry
212	irq_handler
213
214#ifdef CONFIG_PREEMPT
215	get_thread_info tsk
216	ldr	r8, [tsk, #TI_PREEMPT]		@ get 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
224#ifdef CONFIG_TRACE_IRQFLAGS
225	@ The parent context IRQs must have been enabled to get here in
226	@ the first place, so there's no point checking the PSR I bit.
227	bl	trace_hardirqs_on
228#endif
229	svc_exit r5				@ return from exception
230 UNWIND(.fnend		)
231ENDPROC(__irq_svc)
232
233	.ltorg
234
235#ifdef CONFIG_PREEMPT
236svc_preempt:
237	mov	r8, lr
2381:	bl	preempt_schedule_irq		@ irq en/disable is done inside
239	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
240	tst	r0, #_TIF_NEED_RESCHED
241	moveq	pc, r8				@ go again
242	b	1b
243#endif
244
245	.align	5
246__und_svc:
247#ifdef CONFIG_KPROBES
248	@ If a kprobe is about to simulate a "stmdb sp..." instruction,
249	@ it obviously needs free stack space which then will belong to
250	@ the saved context.
251	svc_entry 64
252#else
253	svc_entry
254#endif
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, [r4, #-4]
264#else
265	ldrh	r0, [r4, #-2]			@ Thumb instruction at LR - 2
266	cmp	r0, #0xe800			@ 32-bit instruction if xx >= 0
267	ldrhhs	r9, [r4]			@ bottom 16 bits
268	orrhs	r0, r9, r0, lsl #16
269#endif
270	adr	r9, BSYM(1f)
271	mov	r2, r4
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	r5, [sp, #S_PSR]		@ Get SVC cpsr
286#ifdef CONFIG_TRACE_IRQFLAGS
287	tst	r5, #PSR_I_BIT
288	bleq	trace_hardirqs_on
289	tst	r5, #PSR_I_BIT
290	blne	trace_hardirqs_off
291#endif
292	svc_exit r5				@ return from exception
293 UNWIND(.fnend		)
294ENDPROC(__und_svc)
295
296	.align	5
297__pabt_svc:
298	svc_entry
299	mov	r2, sp				@ regs
300	pabt_helper
301
302	@
303	@ IRQs off again before pulling preserved data off the stack
304	@
305	disable_irq_notrace
306
307#ifdef CONFIG_TRACE_IRQFLAGS
308	tst	r5, #PSR_I_BIT
309	bleq	trace_hardirqs_on
310	tst	r5, #PSR_I_BIT
311	blne	trace_hardirqs_off
312#endif
313	svc_exit r5				@ return from exception
314 UNWIND(.fnend		)
315ENDPROC(__pabt_svc)
316
317	.align	5
318.LCcralign:
319	.word	cr_alignment
320#ifdef MULTI_DABORT
321.LCprocfns:
322	.word	processor
323#endif
324.LCfp:
325	.word	fp_enter
326
327/*
328 * User mode handlers
329 *
330 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
331 */
332
333#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
334#error "sizeof(struct pt_regs) must be a multiple of 8"
335#endif
336
337	.macro	usr_entry
338 UNWIND(.fnstart	)
339 UNWIND(.cantunwind	)	@ don't unwind the user space
340	sub	sp, sp, #S_FRAME_SIZE
341 ARM(	stmib	sp, {r1 - r12}	)
342 THUMB(	stmia	sp, {r0 - r12}	)
343
344	ldmia	r0, {r3 - r5}
345	add	r0, sp, #S_PC		@ here for interlock avoidance
346	mov	r6, #-1			@  ""  ""     ""        ""
347
348	str	r3, [sp]		@ save the "real" r0 copied
349					@ from the exception stack
350
351	@
352	@ We are now ready to fill in the remaining blanks on the stack:
353	@
354	@  r4 - lr_<exception>, already fixed up for correct return/restart
355	@  r5 - spsr_<exception>
356	@  r6 - orig_r0 (see pt_regs definition in ptrace.h)
357	@
358	@ Also, separately save sp_usr and lr_usr
359	@
360	stmia	r0, {r4 - r6}
361 ARM(	stmdb	r0, {sp, lr}^			)
362 THUMB(	store_user_sp_lr r0, r1, S_SP - S_PC	)
363
364	@
365	@ Enable the alignment trap while in kernel mode
366	@
367	alignment_trap r0
368
369	@
370	@ Clear FP to mark the first stack frame
371	@
372	zero_fp
373
374#ifdef CONFIG_IRQSOFF_TRACER
375	bl	trace_hardirqs_off
376#endif
377	.endm
378
379	.macro	kuser_cmpxchg_check
380#if !defined(CONFIG_CPU_32v6K) && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
381#ifndef CONFIG_MMU
382#warning "NPTL on non MMU needs fixing"
383#else
384	@ Make sure our user space atomic helper is restarted
385	@ if it was interrupted in a critical region.  Here we
386	@ perform a quick test inline since it should be false
387	@ 99.9999% of the time.  The rest is done out of line.
388	cmp	r4, #TASK_SIZE
389	blhs	kuser_cmpxchg64_fixup
390#endif
391#endif
392	.endm
393
394	.align	5
395__dabt_usr:
396	usr_entry
397	kuser_cmpxchg_check
398	mov	r2, sp
399	dabt_helper
400	b	ret_from_exception
401 UNWIND(.fnend		)
402ENDPROC(__dabt_usr)
403
404	.align	5
405__irq_usr:
406	usr_entry
407	kuser_cmpxchg_check
408	irq_handler
409	get_thread_info tsk
410	mov	why, #0
411	b	ret_to_user_from_irq
412 UNWIND(.fnend		)
413ENDPROC(__irq_usr)
414
415	.ltorg
416
417	.align	5
418__und_usr:
419	usr_entry
420
421	mov	r2, r4
422	mov	r3, r5
423
424	@
425	@ fall through to the emulation code, which returns using r9 if
426	@ it has emulated the instruction, or the more conventional lr
427	@ if we are to treat this as a real undefined instruction
428	@
429	@  r0 - instruction
430	@
431	adr	r9, BSYM(ret_from_exception)
432	adr	lr, BSYM(__und_usr_unknown)
433	tst	r3, #PSR_T_BIT			@ Thumb mode?
434	itet	eq				@ explicit IT needed for the 1f label
435	subeq	r4, r2, #4			@ ARM instr at LR - 4
436	subne	r4, r2, #2			@ Thumb instr at LR - 2
4371:	ldreqt	r0, [r4]
438#ifdef CONFIG_CPU_ENDIAN_BE8
439	reveq	r0, r0				@ little endian instruction
440#endif
441	beq	call_fpe
442	@ Thumb instruction
443#if CONFIG_ARM_THUMB && __LINUX_ARM_ARCH__ >= 6 && CONFIG_CPU_V7
444/*
445 * Thumb-2 instruction handling.  Note that because pre-v6 and >= v6 platforms
446 * can never be supported in a single kernel, this code is not applicable at
447 * all when __LINUX_ARM_ARCH__ < 6.  This allows simplifying assumptions to be
448 * made about .arch directives.
449 */
450#if __LINUX_ARM_ARCH__ < 7
451/* If the target CPU may not be Thumb-2-capable, a run-time check is needed: */
452#define NEED_CPU_ARCHITECTURE
453	ldr	r5, .LCcpu_architecture
454	ldr	r5, [r5]
455	cmp	r5, #CPU_ARCH_ARMv7
456	blo	__und_usr_unknown
457/*
458 * The following code won't get run unless the running CPU really is v7, so
459 * coding round the lack of ldrht on older arches is pointless.  Temporarily
460 * override the assembler target arch with the minimum required instead:
461 */
462	.arch	armv6t2
463#endif
4642:
465 ARM(	ldrht	r5, [r4], #2	)
466 THUMB(	ldrht	r5, [r4]	)
467 THUMB(	add	r4, r4, #2	)
468	cmp	r5, #0xe800			@ 32bit instruction if xx != 0
469	blo	__und_usr_unknown
4703:	ldrht	r0, [r4]
471	add	r2, r2, #2			@ r2 is PC + 2, make it PC + 4
472	orr	r0, r0, r5, lsl #16
473
474#if __LINUX_ARM_ARCH__ < 7
475/* If the target arch was overridden, change it back: */
476#ifdef CONFIG_CPU_32v6K
477	.arch	armv6k
478#else
479	.arch	armv6
480#endif
481#endif /* __LINUX_ARM_ARCH__ < 7 */
482#else /* !(CONFIG_ARM_THUMB && __LINUX_ARM_ARCH__ >= 6 && CONFIG_CPU_V7) */
483	b	__und_usr_unknown
484#endif
485 UNWIND(.fnend		)
486ENDPROC(__und_usr)
487
488	@
489	@ fallthrough to call_fpe
490	@
491
492/*
493 * The out of line fixup for the ldrt above.
494 */
495	.pushsection .fixup, "ax"
4964:	mov	pc, r9
497	.popsection
498	.pushsection __ex_table,"a"
499	.long	1b, 4b
500#if __LINUX_ARM_ARCH__ >= 7
501	.long	2b, 4b
502	.long	3b, 4b
503#endif
504	.popsection
505
506/*
507 * Check whether the instruction is a co-processor instruction.
508 * If yes, we need to call the relevant co-processor handler.
509 *
510 * Note that we don't do a full check here for the co-processor
511 * instructions; all instructions with bit 27 set are well
512 * defined.  The only instructions that should fault are the
513 * co-processor instructions.  However, we have to watch out
514 * for the ARM6/ARM7 SWI bug.
515 *
516 * NEON is a special case that has to be handled here. Not all
517 * NEON instructions are co-processor instructions, so we have
518 * to make a special case of checking for them. Plus, there's
519 * five groups of them, so we have a table of mask/opcode pairs
520 * to check against, and if any match then we branch off into the
521 * NEON handler code.
522 *
523 * Emulators may wish to make use of the following registers:
524 *  r0  = instruction opcode.
525 *  r2  = PC+4
526 *  r9  = normal "successful" return address
527 *  r10 = this threads thread_info structure.
528 *  lr  = unrecognised instruction return address
529 */
530	@
531	@ Fall-through from Thumb-2 __und_usr
532	@
533#ifdef CONFIG_NEON
534	adr	r6, .LCneon_thumb_opcodes
535	b	2f
536#endif
537call_fpe:
538#ifdef CONFIG_NEON
539	adr	r6, .LCneon_arm_opcodes
5402:
541	ldr	r7, [r6], #4			@ mask value
542	cmp	r7, #0				@ end mask?
543	beq	1f
544	and	r8, r0, r7
545	ldr	r7, [r6], #4			@ opcode bits matching in mask
546	cmp	r8, r7				@ NEON instruction?
547	bne	2b
548	get_thread_info r10
549	mov	r7, #1
550	strb	r7, [r10, #TI_USED_CP + 10]	@ mark CP#10 as used
551	strb	r7, [r10, #TI_USED_CP + 11]	@ mark CP#11 as used
552	b	do_vfp				@ let VFP handler handle this
5531:
554#endif
555	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
556	tstne	r0, #0x04000000			@ bit 26 set on both ARM and Thumb-2
557#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
558	and	r8, r0, #0x0f000000		@ mask out op-code bits
559	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
560#endif
561	moveq	pc, lr
562	get_thread_info r10			@ get current thread
563	and	r8, r0, #0x00000f00		@ mask out CP number
564 THUMB(	lsr	r8, r8, #8		)
565	mov	r7, #1
566	add	r6, r10, #TI_USED_CP
567 ARM(	strb	r7, [r6, r8, lsr #8]	)	@ set appropriate used_cp[]
568 THUMB(	strb	r7, [r6, r8]		)	@ set appropriate used_cp[]
569#ifdef CONFIG_IWMMXT
570	@ Test if we need to give access to iWMMXt coprocessors
571	ldr	r5, [r10, #TI_FLAGS]
572	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
573	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
574	bcs	iwmmxt_task_enable
575#endif
576 ARM(	add	pc, pc, r8, lsr #6	)
577 THUMB(	lsl	r8, r8, #2		)
578 THUMB(	add	pc, r8			)
579	nop
580
581	movw_pc	lr				@ CP#0
582	W(b)	do_fpe				@ CP#1 (FPE)
583	W(b)	do_fpe				@ CP#2 (FPE)
584	movw_pc	lr				@ CP#3
585#ifdef CONFIG_CRUNCH
586	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
587	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
588	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
589#else
590	movw_pc	lr				@ CP#4
591	movw_pc	lr				@ CP#5
592	movw_pc	lr				@ CP#6
593#endif
594	movw_pc	lr				@ CP#7
595	movw_pc	lr				@ CP#8
596	movw_pc	lr				@ CP#9
597#ifdef CONFIG_VFP
598	W(b)	do_vfp				@ CP#10 (VFP)
599	W(b)	do_vfp				@ CP#11 (VFP)
600#else
601	movw_pc	lr				@ CP#10 (VFP)
602	movw_pc	lr				@ CP#11 (VFP)
603#endif
604	movw_pc	lr				@ CP#12
605	movw_pc	lr				@ CP#13
606	movw_pc	lr				@ CP#14 (Debug)
607	movw_pc	lr				@ CP#15 (Control)
608
609#ifdef NEED_CPU_ARCHITECTURE
610	.align	2
611.LCcpu_architecture:
612	.word	__cpu_architecture
613#endif
614
615#ifdef CONFIG_NEON
616	.align	6
617
618.LCneon_arm_opcodes:
619	.word	0xfe000000			@ mask
620	.word	0xf2000000			@ opcode
621
622	.word	0xff100000			@ mask
623	.word	0xf4000000			@ opcode
624
625	.word	0x00000000			@ mask
626	.word	0x00000000			@ opcode
627
628.LCneon_thumb_opcodes:
629	.word	0xef000000			@ mask
630	.word	0xef000000			@ opcode
631
632	.word	0xff100000			@ mask
633	.word	0xf9000000			@ opcode
634
635	.word	0x00000000			@ mask
636	.word	0x00000000			@ opcode
637#endif
638
639do_fpe:
640	enable_irq
641	ldr	r4, .LCfp
642	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
643	ldr	pc, [r4]			@ Call FP module USR entry point
644
645/*
646 * The FP module is called with these registers set:
647 *  r0  = instruction
648 *  r2  = PC+4
649 *  r9  = normal "successful" return address
650 *  r10 = FP workspace
651 *  lr  = unrecognised FP instruction return address
652 */
653
654	.pushsection .data
655ENTRY(fp_enter)
656	.word	no_fp
657	.popsection
658
659ENTRY(no_fp)
660	mov	pc, lr
661ENDPROC(no_fp)
662
663__und_usr_unknown:
664	enable_irq
665	mov	r0, sp
666	adr	lr, BSYM(ret_from_exception)
667	b	do_undefinstr
668ENDPROC(__und_usr_unknown)
669
670	.align	5
671__pabt_usr:
672	usr_entry
673	mov	r2, sp				@ regs
674	pabt_helper
675 UNWIND(.fnend		)
676	/* fall through */
677/*
678 * This is the return code to user mode for abort handlers
679 */
680ENTRY(ret_from_exception)
681 UNWIND(.fnstart	)
682 UNWIND(.cantunwind	)
683	get_thread_info tsk
684	mov	why, #0
685	b	ret_to_user
686 UNWIND(.fnend		)
687ENDPROC(__pabt_usr)
688ENDPROC(ret_from_exception)
689
690/*
691 * Register switch for ARMv3 and ARMv4 processors
692 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
693 * previous and next are guaranteed not to be the same.
694 */
695ENTRY(__switch_to)
696 UNWIND(.fnstart	)
697 UNWIND(.cantunwind	)
698	add	ip, r1, #TI_CPU_SAVE
699	ldr	r3, [r2, #TI_TP_VALUE]
700 ARM(	stmia	ip!, {r4 - sl, fp, sp, lr} )	@ Store most regs on stack
701 THUMB(	stmia	ip!, {r4 - sl, fp}	   )	@ Store most regs on stack
702 THUMB(	str	sp, [ip], #4		   )
703 THUMB(	str	lr, [ip], #4		   )
704#ifdef CONFIG_CPU_USE_DOMAINS
705	ldr	r6, [r2, #TI_CPU_DOMAIN]
706#endif
707	set_tls	r3, r4, r5
708#if defined(CONFIG_CC_STACKPROTECTOR) && !defined(CONFIG_SMP)
709	ldr	r7, [r2, #TI_TASK]
710	ldr	r8, =__stack_chk_guard
711	ldr	r7, [r7, #TSK_STACK_CANARY]
712#endif
713#ifdef CONFIG_CPU_USE_DOMAINS
714	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
715#endif
716	mov	r5, r0
717	add	r4, r2, #TI_CPU_SAVE
718	ldr	r0, =thread_notify_head
719	mov	r1, #THREAD_NOTIFY_SWITCH
720	bl	atomic_notifier_call_chain
721#if defined(CONFIG_CC_STACKPROTECTOR) && !defined(CONFIG_SMP)
722	str	r7, [r8]
723#endif
724 THUMB(	mov	ip, r4			   )
725	mov	r0, r5
726 ARM(	ldmia	r4, {r4 - sl, fp, sp, pc}  )	@ Load all regs saved previously
727 THUMB(	ldmia	ip!, {r4 - sl, fp}	   )	@ Load all regs saved previously
728 THUMB(	ldr	sp, [ip], #4		   )
729 THUMB(	ldr	pc, [ip]		   )
730 UNWIND(.fnend		)
731ENDPROC(__switch_to)
732
733	__INIT
734
735/*
736 * User helpers.
737 *
738 * Each segment is 32-byte aligned and will be moved to the top of the high
739 * vector page.  New segments (if ever needed) must be added in front of
740 * existing ones.  This mechanism should be used only for things that are
741 * really small and justified, and not be abused freely.
742 *
743 * See Documentation/arm/kernel_user_helpers.txt for formal definitions.
744 */
745 THUMB(	.arm	)
746
747	.macro	usr_ret, reg
748#ifdef CONFIG_ARM_THUMB
749	bx	\reg
750#else
751	mov	pc, \reg
752#endif
753	.endm
754
755	.align	5
756	.globl	__kuser_helper_start
757__kuser_helper_start:
758
759/*
760 * Due to the length of some sequences, __kuser_cmpxchg64 spans 2 regular
761 * kuser "slots", therefore 0xffff0f80 is not used as a valid entry point.
762 */
763
764__kuser_cmpxchg64:				@ 0xffff0f60
765
766#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
767
768	/*
769	 * Poor you.  No fast solution possible...
770	 * The kernel itself must perform the operation.
771	 * A special ghost syscall is used for that (see traps.c).
772	 */
773	stmfd	sp!, {r7, lr}
774	ldr	r7, 1f			@ it's 20 bits
775	swi	__ARM_NR_cmpxchg64
776	ldmfd	sp!, {r7, pc}
7771:	.word	__ARM_NR_cmpxchg64
778
779#elif defined(CONFIG_CPU_32v6K)
780
781	stmfd	sp!, {r4, r5, r6, r7}
782	ldrd	r4, r5, [r0]			@ load old val
783	ldrd	r6, r7, [r1]			@ load new val
784	smp_dmb	arm
7851:	ldrexd	r0, r1, [r2]			@ load current val
786	eors	r3, r0, r4			@ compare with oldval (1)
787	eoreqs	r3, r1, r5			@ compare with oldval (2)
788	strexdeq r3, r6, r7, [r2]		@ store newval if eq
789	teqeq	r3, #1				@ success?
790	beq	1b				@ if no then retry
791	smp_dmb	arm
792	rsbs	r0, r3, #0			@ set returned val and C flag
793	ldmfd	sp!, {r4, r5, r6, r7}
794	bx	lr
795
796#elif !defined(CONFIG_SMP)
797
798#ifdef CONFIG_MMU
799
800	/*
801	 * The only thing that can break atomicity in this cmpxchg64
802	 * implementation is either an IRQ or a data abort exception
803	 * causing another process/thread to be scheduled in the middle of
804	 * the critical sequence.  The same strategy as for cmpxchg is used.
805	 */
806	stmfd	sp!, {r4, r5, r6, lr}
807	ldmia	r0, {r4, r5}			@ load old val
808	ldmia	r1, {r6, lr}			@ load new val
8091:	ldmia	r2, {r0, r1}			@ load current val
810	eors	r3, r0, r4			@ compare with oldval (1)
811	eoreqs	r3, r1, r5			@ compare with oldval (2)
8122:	stmeqia	r2, {r6, lr}			@ store newval if eq
813	rsbs	r0, r3, #0			@ set return val and C flag
814	ldmfd	sp!, {r4, r5, r6, pc}
815
816	.text
817kuser_cmpxchg64_fixup:
818	@ Called from kuser_cmpxchg_fixup.
819	@ r4 = address of interrupted insn (must be preserved).
820	@ sp = saved regs. r7 and r8 are clobbered.
821	@ 1b = first critical insn, 2b = last critical insn.
822	@ If r4 >= 1b and r4 <= 2b then saved pc_usr is set to 1b.
823	mov	r7, #0xffff0fff
824	sub	r7, r7, #(0xffff0fff - (0xffff0f60 + (1b - __kuser_cmpxchg64)))
825	subs	r8, r4, r7
826	rsbcss	r8, r8, #(2b - 1b)
827	strcs	r7, [sp, #S_PC]
828#if __LINUX_ARM_ARCH__ < 6
829	bcc	kuser_cmpxchg32_fixup
830#endif
831	mov	pc, lr
832	.previous
833
834#else
835#warning "NPTL on non MMU needs fixing"
836	mov	r0, #-1
837	adds	r0, r0, #0
838	usr_ret	lr
839#endif
840
841#else
842#error "incoherent kernel configuration"
843#endif
844
845	/* pad to next slot */
846	.rept	(16 - (. - __kuser_cmpxchg64)/4)
847	.word	0
848	.endr
849
850	.align	5
851
852__kuser_memory_barrier:				@ 0xffff0fa0
853	smp_dmb	arm
854	usr_ret	lr
855
856	.align	5
857
858__kuser_cmpxchg:				@ 0xffff0fc0
859
860#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
861
862	/*
863	 * Poor you.  No fast solution possible...
864	 * The kernel itself must perform the operation.
865	 * A special ghost syscall is used for that (see traps.c).
866	 */
867	stmfd	sp!, {r7, lr}
868	ldr	r7, 1f			@ it's 20 bits
869	swi	__ARM_NR_cmpxchg
870	ldmfd	sp!, {r7, pc}
8711:	.word	__ARM_NR_cmpxchg
872
873#elif __LINUX_ARM_ARCH__ < 6
874
875#ifdef CONFIG_MMU
876
877	/*
878	 * The only thing that can break atomicity in this cmpxchg
879	 * implementation is either an IRQ or a data abort exception
880	 * causing another process/thread to be scheduled in the middle
881	 * of the critical sequence.  To prevent this, code is added to
882	 * the IRQ and data abort exception handlers to set the pc back
883	 * to the beginning of the critical section if it is found to be
884	 * within that critical section (see kuser_cmpxchg_fixup).
885	 */
8861:	ldr	r3, [r2]			@ load current val
887	subs	r3, r3, r0			@ compare with oldval
8882:	streq	r1, [r2]			@ store newval if eq
889	rsbs	r0, r3, #0			@ set return val and C flag
890	usr_ret	lr
891
892	.text
893kuser_cmpxchg32_fixup:
894	@ Called from kuser_cmpxchg_check macro.
895	@ r4 = address of interrupted insn (must be preserved).
896	@ sp = saved regs. r7 and r8 are clobbered.
897	@ 1b = first critical insn, 2b = last critical insn.
898	@ If r4 >= 1b and r4 <= 2b then saved pc_usr is set to 1b.
899	mov	r7, #0xffff0fff
900	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
901	subs	r8, r4, r7
902	rsbcss	r8, r8, #(2b - 1b)
903	strcs	r7, [sp, #S_PC]
904	mov	pc, lr
905	.previous
906
907#else
908#warning "NPTL on non MMU needs fixing"
909	mov	r0, #-1
910	adds	r0, r0, #0
911	usr_ret	lr
912#endif
913
914#else
915
916	smp_dmb	arm
9171:	ldrex	r3, [r2]
918	subs	r3, r3, r0
919	strexeq	r3, r1, [r2]
920	teqeq	r3, #1
921	beq	1b
922	rsbs	r0, r3, #0
923	/* beware -- each __kuser slot must be 8 instructions max */
924	ALT_SMP(b	__kuser_memory_barrier)
925	ALT_UP(usr_ret	lr)
926
927#endif
928
929	.align	5
930
931__kuser_get_tls:				@ 0xffff0fe0
932	ldr	r0, [pc, #(16 - 8)]	@ read TLS, set in kuser_get_tls_init
933	usr_ret	lr
934	mrc	p15, 0, r0, c13, c0, 3	@ 0xffff0fe8 hardware TLS code
935	.rep	4
936	.word	0			@ 0xffff0ff0 software TLS value, then
937	.endr				@ pad up to __kuser_helper_version
938
939__kuser_helper_version:				@ 0xffff0ffc
940	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
941
942	.globl	__kuser_helper_end
943__kuser_helper_end:
944
945 THUMB(	.thumb	)
946
947/*
948 * Vector stubs.
949 *
950 * This code is copied to 0xffff0200 so we can use branches in the
951 * vectors, rather than ldr's.  Note that this code must not
952 * exceed 0x300 bytes.
953 *
954 * Common stub entry macro:
955 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
956 *
957 * SP points to a minimal amount of processor-private memory, the address
958 * of which is copied into r0 for the mode specific abort handler.
959 */
960	.macro	vector_stub, name, mode, correction=0
961	.align	5
962
963vector_\name:
964	.if \correction
965	sub	lr, lr, #\correction
966	.endif
967
968	@
969	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
970	@ (parent CPSR)
971	@
972	stmia	sp, {r0, lr}		@ save r0, lr
973	mrs	lr, spsr
974	str	lr, [sp, #8]		@ save spsr
975
976	@
977	@ Prepare for SVC32 mode.  IRQs remain disabled.
978	@
979	mrs	r0, cpsr
980	eor	r0, r0, #(\mode ^ SVC_MODE | PSR_ISETSTATE)
981	msr	spsr_cxsf, r0
982
983	@
984	@ the branch table must immediately follow this code
985	@
986	and	lr, lr, #0x0f
987 THUMB(	adr	r0, 1f			)
988 THUMB(	ldr	lr, [r0, lr, lsl #2]	)
989	mov	r0, sp
990 ARM(	ldr	lr, [pc, lr, lsl #2]	)
991	movs	pc, lr			@ branch to handler in SVC mode
992ENDPROC(vector_\name)
993
994	.align	2
995	@ handler addresses follow this label
9961:
997	.endm
998
999	.globl	__stubs_start
1000__stubs_start:
1001/*
1002 * Interrupt dispatcher
1003 */
1004	vector_stub	irq, IRQ_MODE, 4
1005
1006	.long	__irq_usr			@  0  (USR_26 / USR_32)
1007	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
1008	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
1009	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
1010	.long	__irq_invalid			@  4
1011	.long	__irq_invalid			@  5
1012	.long	__irq_invalid			@  6
1013	.long	__irq_invalid			@  7
1014	.long	__irq_invalid			@  8
1015	.long	__irq_invalid			@  9
1016	.long	__irq_invalid			@  a
1017	.long	__irq_invalid			@  b
1018	.long	__irq_invalid			@  c
1019	.long	__irq_invalid			@  d
1020	.long	__irq_invalid			@  e
1021	.long	__irq_invalid			@  f
1022
1023/*
1024 * Data abort dispatcher
1025 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1026 */
1027	vector_stub	dabt, ABT_MODE, 8
1028
1029	.long	__dabt_usr			@  0  (USR_26 / USR_32)
1030	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
1031	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
1032	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
1033	.long	__dabt_invalid			@  4
1034	.long	__dabt_invalid			@  5
1035	.long	__dabt_invalid			@  6
1036	.long	__dabt_invalid			@  7
1037	.long	__dabt_invalid			@  8
1038	.long	__dabt_invalid			@  9
1039	.long	__dabt_invalid			@  a
1040	.long	__dabt_invalid			@  b
1041	.long	__dabt_invalid			@  c
1042	.long	__dabt_invalid			@  d
1043	.long	__dabt_invalid			@  e
1044	.long	__dabt_invalid			@  f
1045
1046/*
1047 * Prefetch abort dispatcher
1048 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1049 */
1050	vector_stub	pabt, ABT_MODE, 4
1051
1052	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1053	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1054	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1055	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1056	.long	__pabt_invalid			@  4
1057	.long	__pabt_invalid			@  5
1058	.long	__pabt_invalid			@  6
1059	.long	__pabt_invalid			@  7
1060	.long	__pabt_invalid			@  8
1061	.long	__pabt_invalid			@  9
1062	.long	__pabt_invalid			@  a
1063	.long	__pabt_invalid			@  b
1064	.long	__pabt_invalid			@  c
1065	.long	__pabt_invalid			@  d
1066	.long	__pabt_invalid			@  e
1067	.long	__pabt_invalid			@  f
1068
1069/*
1070 * Undef instr entry dispatcher
1071 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1072 */
1073	vector_stub	und, UND_MODE
1074
1075	.long	__und_usr			@  0 (USR_26 / USR_32)
1076	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1077	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1078	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1079	.long	__und_invalid			@  4
1080	.long	__und_invalid			@  5
1081	.long	__und_invalid			@  6
1082	.long	__und_invalid			@  7
1083	.long	__und_invalid			@  8
1084	.long	__und_invalid			@  9
1085	.long	__und_invalid			@  a
1086	.long	__und_invalid			@  b
1087	.long	__und_invalid			@  c
1088	.long	__und_invalid			@  d
1089	.long	__und_invalid			@  e
1090	.long	__und_invalid			@  f
1091
1092	.align	5
1093
1094/*=============================================================================
1095 * Undefined FIQs
1096 *-----------------------------------------------------------------------------
1097 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1098 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1099 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1100 * damage alert!  I don't think that we can execute any code in here in any
1101 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1102 * get out of that mode without clobbering one register.
1103 */
1104vector_fiq:
1105	disable_fiq
1106	subs	pc, lr, #4
1107
1108/*=============================================================================
1109 * Address exception handler
1110 *-----------------------------------------------------------------------------
1111 * These aren't too critical.
1112 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1113 */
1114
1115vector_addrexcptn:
1116	b	vector_addrexcptn
1117
1118/*
1119 * We group all the following data together to optimise
1120 * for CPUs with separate I & D caches.
1121 */
1122	.align	5
1123
1124.LCvswi:
1125	.word	vector_swi
1126
1127	.globl	__stubs_end
1128__stubs_end:
1129
1130	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1131
1132	.globl	__vectors_start
1133__vectors_start:
1134 ARM(	swi	SYS_ERROR0	)
1135 THUMB(	svc	#0		)
1136 THUMB(	nop			)
1137	W(b)	vector_und + stubs_offset
1138	W(ldr)	pc, .LCvswi + stubs_offset
1139	W(b)	vector_pabt + stubs_offset
1140	W(b)	vector_dabt + stubs_offset
1141	W(b)	vector_addrexcptn + stubs_offset
1142	W(b)	vector_irq + stubs_offset
1143	W(b)	vector_fiq + stubs_offset
1144
1145	.globl	__vectors_end
1146__vectors_end:
1147
1148	.data
1149
1150	.globl	cr_alignment
1151	.globl	cr_no_alignment
1152cr_alignment:
1153	.space	4
1154cr_no_alignment:
1155	.space	4
1156
1157#ifdef CONFIG_MULTI_IRQ_HANDLER
1158	.globl	handle_arch_irq
1159handle_arch_irq:
1160	.space	4
1161#endif
1162