xref: /openbmc/linux/arch/arm/kernel/entry-armv.S (revision ca79522c)
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/assembler.h>
19#include <asm/memory.h>
20#include <asm/glue-df.h>
21#include <asm/glue-pf.h>
22#include <asm/vfpmacros.h>
23#ifndef CONFIG_MULTI_IRQ_HANDLER
24#include <mach/entry-macro.S>
25#endif
26#include <asm/thread_notify.h>
27#include <asm/unwind.h>
28#include <asm/unistd.h>
29#include <asm/tls.h>
30#include <asm/system_info.h>
31
32#include "entry-header.S"
33#include <asm/entry-macro-multi.S>
34
35/*
36 * Interrupt handling.
37 */
38	.macro	irq_handler
39#ifdef CONFIG_MULTI_IRQ_HANDLER
40	ldr	r1, =handle_arch_irq
41	mov	r0, sp
42	adr	lr, BSYM(9997f)
43	ldr	pc, [r1]
44#else
45	arch_irq_handler_default
46#endif
479997:
48	.endm
49
50	.macro	pabt_helper
51	@ PABORT handler takes pt_regs in r2, fault address in r4 and psr in r5
52#ifdef MULTI_PABORT
53	ldr	ip, .LCprocfns
54	mov	lr, pc
55	ldr	pc, [ip, #PROCESSOR_PABT_FUNC]
56#else
57	bl	CPU_PABORT_HANDLER
58#endif
59	.endm
60
61	.macro	dabt_helper
62
63	@
64	@ Call the processor-specific abort handler:
65	@
66	@  r2 - pt_regs
67	@  r4 - aborted context pc
68	@  r5 - aborted context psr
69	@
70	@ The abort handler must return the aborted address in r0, and
71	@ the fault status register in r1.  r9 must be preserved.
72	@
73#ifdef MULTI_DABORT
74	ldr	ip, .LCprocfns
75	mov	lr, pc
76	ldr	pc, [ip, #PROCESSOR_DABT_FUNC]
77#else
78	bl	CPU_DABORT_HANDLER
79#endif
80	.endm
81
82#ifdef CONFIG_KPROBES
83	.section	.kprobes.text,"ax",%progbits
84#else
85	.text
86#endif
87
88/*
89 * Invalid mode handlers
90 */
91	.macro	inv_entry, reason
92	sub	sp, sp, #S_FRAME_SIZE
93 ARM(	stmib	sp, {r1 - lr}		)
94 THUMB(	stmia	sp, {r0 - r12}		)
95 THUMB(	str	sp, [sp, #S_SP]		)
96 THUMB(	str	lr, [sp, #S_LR]		)
97	mov	r1, #\reason
98	.endm
99
100__pabt_invalid:
101	inv_entry BAD_PREFETCH
102	b	common_invalid
103ENDPROC(__pabt_invalid)
104
105__dabt_invalid:
106	inv_entry BAD_DATA
107	b	common_invalid
108ENDPROC(__dabt_invalid)
109
110__irq_invalid:
111	inv_entry BAD_IRQ
112	b	common_invalid
113ENDPROC(__irq_invalid)
114
115__und_invalid:
116	inv_entry BAD_UNDEFINSTR
117
118	@
119	@ XXX fall through to common_invalid
120	@
121
122@
123@ common_invalid - generic code for failed exception (re-entrant version of handlers)
124@
125common_invalid:
126	zero_fp
127
128	ldmia	r0, {r4 - r6}
129	add	r0, sp, #S_PC		@ here for interlock avoidance
130	mov	r7, #-1			@  ""   ""    ""        ""
131	str	r4, [sp]		@ save preserved r0
132	stmia	r0, {r5 - r7}		@ lr_<exception>,
133					@ cpsr_<exception>, "old_r0"
134
135	mov	r0, sp
136	b	bad_mode
137ENDPROC(__und_invalid)
138
139/*
140 * SVC mode handlers
141 */
142
143#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
144#define SPFIX(code...) code
145#else
146#define SPFIX(code...)
147#endif
148
149	.macro	svc_entry, stack_hole=0
150 UNWIND(.fnstart		)
151 UNWIND(.save {r0 - pc}		)
152	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole - 4)
153#ifdef CONFIG_THUMB2_KERNEL
154 SPFIX(	str	r0, [sp]	)	@ temporarily saved
155 SPFIX(	mov	r0, sp		)
156 SPFIX(	tst	r0, #4		)	@ test original stack alignment
157 SPFIX(	ldr	r0, [sp]	)	@ restored
158#else
159 SPFIX(	tst	sp, #4		)
160#endif
161 SPFIX(	subeq	sp, sp, #4	)
162	stmia	sp, {r1 - r12}
163
164	ldmia	r0, {r3 - r5}
165	add	r7, sp, #S_SP - 4	@ here for interlock avoidance
166	mov	r6, #-1			@  ""  ""      ""       ""
167	add	r2, sp, #(S_FRAME_SIZE + \stack_hole - 4)
168 SPFIX(	addeq	r2, r2, #4	)
169	str	r3, [sp, #-4]!		@ save the "real" r0 copied
170					@ from the exception stack
171
172	mov	r3, lr
173
174	@
175	@ We are now ready to fill in the remaining blanks on the stack:
176	@
177	@  r2 - sp_svc
178	@  r3 - lr_svc
179	@  r4 - lr_<exception>, already fixed up for correct return/restart
180	@  r5 - spsr_<exception>
181	@  r6 - orig_r0 (see pt_regs definition in ptrace.h)
182	@
183	stmia	r7, {r2 - r6}
184
185#ifdef CONFIG_TRACE_IRQFLAGS
186	bl	trace_hardirqs_off
187#endif
188	.endm
189
190	.align	5
191__dabt_svc:
192	svc_entry
193	mov	r2, sp
194	dabt_helper
195	svc_exit r5				@ return from exception
196 UNWIND(.fnend		)
197ENDPROC(__dabt_svc)
198
199	.align	5
200__irq_svc:
201	svc_entry
202	irq_handler
203
204#ifdef CONFIG_PREEMPT
205	get_thread_info tsk
206	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
207	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
208	teq	r8, #0				@ if preempt count != 0
209	movne	r0, #0				@ force flags to 0
210	tst	r0, #_TIF_NEED_RESCHED
211	blne	svc_preempt
212#endif
213
214	svc_exit r5, irq = 1			@ return from exception
215 UNWIND(.fnend		)
216ENDPROC(__irq_svc)
217
218	.ltorg
219
220#ifdef CONFIG_PREEMPT
221svc_preempt:
222	mov	r8, lr
2231:	bl	preempt_schedule_irq		@ irq en/disable is done inside
224	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
225	tst	r0, #_TIF_NEED_RESCHED
226	moveq	pc, r8				@ go again
227	b	1b
228#endif
229
230__und_fault:
231	@ Correct the PC such that it is pointing at the instruction
232	@ which caused the fault.  If the faulting instruction was ARM
233	@ the PC will be pointing at the next instruction, and have to
234	@ subtract 4.  Otherwise, it is Thumb, and the PC will be
235	@ pointing at the second half of the Thumb instruction.  We
236	@ have to subtract 2.
237	ldr	r2, [r0, #S_PC]
238	sub	r2, r2, r1
239	str	r2, [r0, #S_PC]
240	b	do_undefinstr
241ENDPROC(__und_fault)
242
243	.align	5
244__und_svc:
245#ifdef CONFIG_KPROBES
246	@ If a kprobe is about to simulate a "stmdb sp..." instruction,
247	@ it obviously needs free stack space which then will belong to
248	@ the saved context.
249	svc_entry 64
250#else
251	svc_entry
252#endif
253	@
254	@ call emulation code, which returns using r9 if it has emulated
255	@ the instruction, or the more conventional lr if we are to treat
256	@ this as a real undefined instruction
257	@
258	@  r0 - instruction
259	@
260#ifndef CONFIG_THUMB2_KERNEL
261	ldr	r0, [r4, #-4]
262#else
263	mov	r1, #2
264	ldrh	r0, [r4, #-2]			@ Thumb instruction at LR - 2
265	cmp	r0, #0xe800			@ 32-bit instruction if xx >= 0
266	blo	__und_svc_fault
267	ldrh	r9, [r4]			@ bottom 16 bits
268	add	r4, r4, #2
269	str	r4, [sp, #S_PC]
270	orr	r0, r9, r0, lsl #16
271#endif
272	adr	r9, BSYM(__und_svc_finish)
273	mov	r2, r4
274	bl	call_fpe
275
276	mov	r1, #4				@ PC correction to apply
277__und_svc_fault:
278	mov	r0, sp				@ struct pt_regs *regs
279	bl	__und_fault
280
281__und_svc_finish:
282	ldr	r5, [sp, #S_PSR]		@ Get SVC cpsr
283	svc_exit r5				@ return from exception
284 UNWIND(.fnend		)
285ENDPROC(__und_svc)
286
287	.align	5
288__pabt_svc:
289	svc_entry
290	mov	r2, sp				@ regs
291	pabt_helper
292	svc_exit r5				@ return from exception
293 UNWIND(.fnend		)
294ENDPROC(__pabt_svc)
295
296	.align	5
297.LCcralign:
298	.word	cr_alignment
299#ifdef MULTI_DABORT
300.LCprocfns:
301	.word	processor
302#endif
303.LCfp:
304	.word	fp_enter
305
306/*
307 * User mode handlers
308 *
309 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
310 */
311
312#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
313#error "sizeof(struct pt_regs) must be a multiple of 8"
314#endif
315
316	.macro	usr_entry
317 UNWIND(.fnstart	)
318 UNWIND(.cantunwind	)	@ don't unwind the user space
319	sub	sp, sp, #S_FRAME_SIZE
320 ARM(	stmib	sp, {r1 - r12}	)
321 THUMB(	stmia	sp, {r0 - r12}	)
322
323	ldmia	r0, {r3 - r5}
324	add	r0, sp, #S_PC		@ here for interlock avoidance
325	mov	r6, #-1			@  ""  ""     ""        ""
326
327	str	r3, [sp]		@ save the "real" r0 copied
328					@ from the exception stack
329
330	@
331	@ We are now ready to fill in the remaining blanks on the stack:
332	@
333	@  r4 - lr_<exception>, already fixed up for correct return/restart
334	@  r5 - spsr_<exception>
335	@  r6 - orig_r0 (see pt_regs definition in ptrace.h)
336	@
337	@ Also, separately save sp_usr and lr_usr
338	@
339	stmia	r0, {r4 - r6}
340 ARM(	stmdb	r0, {sp, lr}^			)
341 THUMB(	store_user_sp_lr r0, r1, S_SP - S_PC	)
342
343	@
344	@ Enable the alignment trap while in kernel mode
345	@
346	alignment_trap r0
347
348	@
349	@ Clear FP to mark the first stack frame
350	@
351	zero_fp
352
353#ifdef CONFIG_IRQSOFF_TRACER
354	bl	trace_hardirqs_off
355#endif
356	ct_user_exit save = 0
357	.endm
358
359	.macro	kuser_cmpxchg_check
360#if !defined(CONFIG_CPU_32v6K) && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
361#ifndef CONFIG_MMU
362#warning "NPTL on non MMU needs fixing"
363#else
364	@ Make sure our user space atomic helper is restarted
365	@ if it was interrupted in a critical region.  Here we
366	@ perform a quick test inline since it should be false
367	@ 99.9999% of the time.  The rest is done out of line.
368	cmp	r4, #TASK_SIZE
369	blhs	kuser_cmpxchg64_fixup
370#endif
371#endif
372	.endm
373
374	.align	5
375__dabt_usr:
376	usr_entry
377	kuser_cmpxchg_check
378	mov	r2, sp
379	dabt_helper
380	b	ret_from_exception
381 UNWIND(.fnend		)
382ENDPROC(__dabt_usr)
383
384	.align	5
385__irq_usr:
386	usr_entry
387	kuser_cmpxchg_check
388	irq_handler
389	get_thread_info tsk
390	mov	why, #0
391	b	ret_to_user_from_irq
392 UNWIND(.fnend		)
393ENDPROC(__irq_usr)
394
395	.ltorg
396
397	.align	5
398__und_usr:
399	usr_entry
400
401	mov	r2, r4
402	mov	r3, r5
403
404	@ r2 = regs->ARM_pc, which is either 2 or 4 bytes ahead of the
405	@      faulting instruction depending on Thumb mode.
406	@ r3 = regs->ARM_cpsr
407	@
408	@ The emulation code returns using r9 if it has emulated the
409	@ instruction, or the more conventional lr if we are to treat
410	@ this as a real undefined instruction
411	@
412	adr	r9, BSYM(ret_from_exception)
413
414	tst	r3, #PSR_T_BIT			@ Thumb mode?
415	bne	__und_usr_thumb
416	sub	r4, r2, #4			@ ARM instr at LR - 4
4171:	ldrt	r0, [r4]
418#ifdef CONFIG_CPU_ENDIAN_BE8
419	rev	r0, r0				@ little endian instruction
420#endif
421	@ r0 = 32-bit ARM instruction which caused the exception
422	@ r2 = PC value for the following instruction (:= regs->ARM_pc)
423	@ r4 = PC value for the faulting instruction
424	@ lr = 32-bit undefined instruction function
425	adr	lr, BSYM(__und_usr_fault_32)
426	b	call_fpe
427
428__und_usr_thumb:
429	@ Thumb instruction
430	sub	r4, r2, #2			@ First half of thumb instr at LR - 2
431#if CONFIG_ARM_THUMB && __LINUX_ARM_ARCH__ >= 6 && CONFIG_CPU_V7
432/*
433 * Thumb-2 instruction handling.  Note that because pre-v6 and >= v6 platforms
434 * can never be supported in a single kernel, this code is not applicable at
435 * all when __LINUX_ARM_ARCH__ < 6.  This allows simplifying assumptions to be
436 * made about .arch directives.
437 */
438#if __LINUX_ARM_ARCH__ < 7
439/* If the target CPU may not be Thumb-2-capable, a run-time check is needed: */
440#define NEED_CPU_ARCHITECTURE
441	ldr	r5, .LCcpu_architecture
442	ldr	r5, [r5]
443	cmp	r5, #CPU_ARCH_ARMv7
444	blo	__und_usr_fault_16		@ 16bit undefined instruction
445/*
446 * The following code won't get run unless the running CPU really is v7, so
447 * coding round the lack of ldrht on older arches is pointless.  Temporarily
448 * override the assembler target arch with the minimum required instead:
449 */
450	.arch	armv6t2
451#endif
4522:	ldrht	r5, [r4]
453	cmp	r5, #0xe800			@ 32bit instruction if xx != 0
454	blo	__und_usr_fault_16		@ 16bit undefined instruction
4553:	ldrht	r0, [r2]
456	add	r2, r2, #2			@ r2 is PC + 2, make it PC + 4
457	str	r2, [sp, #S_PC]			@ it's a 2x16bit instr, update
458	orr	r0, r0, r5, lsl #16
459	adr	lr, BSYM(__und_usr_fault_32)
460	@ r0 = the two 16-bit Thumb instructions which caused the exception
461	@ r2 = PC value for the following Thumb instruction (:= regs->ARM_pc)
462	@ r4 = PC value for the first 16-bit Thumb instruction
463	@ lr = 32bit undefined instruction function
464
465#if __LINUX_ARM_ARCH__ < 7
466/* If the target arch was overridden, change it back: */
467#ifdef CONFIG_CPU_32v6K
468	.arch	armv6k
469#else
470	.arch	armv6
471#endif
472#endif /* __LINUX_ARM_ARCH__ < 7 */
473#else /* !(CONFIG_ARM_THUMB && __LINUX_ARM_ARCH__ >= 6 && CONFIG_CPU_V7) */
474	b	__und_usr_fault_16
475#endif
476 UNWIND(.fnend)
477ENDPROC(__und_usr)
478
479/*
480 * The out of line fixup for the ldrt instructions above.
481 */
482	.pushsection .fixup, "ax"
483	.align	2
4844:	mov	pc, r9
485	.popsection
486	.pushsection __ex_table,"a"
487	.long	1b, 4b
488#if CONFIG_ARM_THUMB && __LINUX_ARM_ARCH__ >= 6 && CONFIG_CPU_V7
489	.long	2b, 4b
490	.long	3b, 4b
491#endif
492	.popsection
493
494/*
495 * Check whether the instruction is a co-processor instruction.
496 * If yes, we need to call the relevant co-processor handler.
497 *
498 * Note that we don't do a full check here for the co-processor
499 * instructions; all instructions with bit 27 set are well
500 * defined.  The only instructions that should fault are the
501 * co-processor instructions.  However, we have to watch out
502 * for the ARM6/ARM7 SWI bug.
503 *
504 * NEON is a special case that has to be handled here. Not all
505 * NEON instructions are co-processor instructions, so we have
506 * to make a special case of checking for them. Plus, there's
507 * five groups of them, so we have a table of mask/opcode pairs
508 * to check against, and if any match then we branch off into the
509 * NEON handler code.
510 *
511 * Emulators may wish to make use of the following registers:
512 *  r0  = instruction opcode (32-bit ARM or two 16-bit Thumb)
513 *  r2  = PC value to resume execution after successful emulation
514 *  r9  = normal "successful" return address
515 *  r10 = this threads thread_info structure
516 *  lr  = unrecognised instruction return address
517 * IRQs disabled, FIQs enabled.
518 */
519	@
520	@ Fall-through from Thumb-2 __und_usr
521	@
522#ifdef CONFIG_NEON
523	get_thread_info r10			@ get current thread
524	adr	r6, .LCneon_thumb_opcodes
525	b	2f
526#endif
527call_fpe:
528	get_thread_info r10			@ get current thread
529#ifdef CONFIG_NEON
530	adr	r6, .LCneon_arm_opcodes
5312:	ldr	r5, [r6], #4			@ mask value
532	ldr	r7, [r6], #4			@ opcode bits matching in mask
533	cmp	r5, #0				@ end mask?
534	beq	1f
535	and	r8, r0, r5
536	cmp	r8, r7				@ NEON instruction?
537	bne	2b
538	mov	r7, #1
539	strb	r7, [r10, #TI_USED_CP + 10]	@ mark CP#10 as used
540	strb	r7, [r10, #TI_USED_CP + 11]	@ mark CP#11 as used
541	b	do_vfp				@ let VFP handler handle this
5421:
543#endif
544	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
545	tstne	r0, #0x04000000			@ bit 26 set on both ARM and Thumb-2
546	moveq	pc, lr
547	and	r8, r0, #0x00000f00		@ mask out CP number
548 THUMB(	lsr	r8, r8, #8		)
549	mov	r7, #1
550	add	r6, r10, #TI_USED_CP
551 ARM(	strb	r7, [r6, r8, lsr #8]	)	@ set appropriate used_cp[]
552 THUMB(	strb	r7, [r6, r8]		)	@ set appropriate used_cp[]
553#ifdef CONFIG_IWMMXT
554	@ Test if we need to give access to iWMMXt coprocessors
555	ldr	r5, [r10, #TI_FLAGS]
556	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
557	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
558	bcs	iwmmxt_task_enable
559#endif
560 ARM(	add	pc, pc, r8, lsr #6	)
561 THUMB(	lsl	r8, r8, #2		)
562 THUMB(	add	pc, r8			)
563	nop
564
565	movw_pc	lr				@ CP#0
566	W(b)	do_fpe				@ CP#1 (FPE)
567	W(b)	do_fpe				@ CP#2 (FPE)
568	movw_pc	lr				@ CP#3
569#ifdef CONFIG_CRUNCH
570	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
571	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
572	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
573#else
574	movw_pc	lr				@ CP#4
575	movw_pc	lr				@ CP#5
576	movw_pc	lr				@ CP#6
577#endif
578	movw_pc	lr				@ CP#7
579	movw_pc	lr				@ CP#8
580	movw_pc	lr				@ CP#9
581#ifdef CONFIG_VFP
582	W(b)	do_vfp				@ CP#10 (VFP)
583	W(b)	do_vfp				@ CP#11 (VFP)
584#else
585	movw_pc	lr				@ CP#10 (VFP)
586	movw_pc	lr				@ CP#11 (VFP)
587#endif
588	movw_pc	lr				@ CP#12
589	movw_pc	lr				@ CP#13
590	movw_pc	lr				@ CP#14 (Debug)
591	movw_pc	lr				@ CP#15 (Control)
592
593#ifdef NEED_CPU_ARCHITECTURE
594	.align	2
595.LCcpu_architecture:
596	.word	__cpu_architecture
597#endif
598
599#ifdef CONFIG_NEON
600	.align	6
601
602.LCneon_arm_opcodes:
603	.word	0xfe000000			@ mask
604	.word	0xf2000000			@ opcode
605
606	.word	0xff100000			@ mask
607	.word	0xf4000000			@ opcode
608
609	.word	0x00000000			@ mask
610	.word	0x00000000			@ opcode
611
612.LCneon_thumb_opcodes:
613	.word	0xef000000			@ mask
614	.word	0xef000000			@ opcode
615
616	.word	0xff100000			@ mask
617	.word	0xf9000000			@ opcode
618
619	.word	0x00000000			@ mask
620	.word	0x00000000			@ opcode
621#endif
622
623do_fpe:
624	enable_irq
625	ldr	r4, .LCfp
626	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
627	ldr	pc, [r4]			@ Call FP module USR entry point
628
629/*
630 * The FP module is called with these registers set:
631 *  r0  = instruction
632 *  r2  = PC+4
633 *  r9  = normal "successful" return address
634 *  r10 = FP workspace
635 *  lr  = unrecognised FP instruction return address
636 */
637
638	.pushsection .data
639ENTRY(fp_enter)
640	.word	no_fp
641	.popsection
642
643ENTRY(no_fp)
644	mov	pc, lr
645ENDPROC(no_fp)
646
647__und_usr_fault_32:
648	mov	r1, #4
649	b	1f
650__und_usr_fault_16:
651	mov	r1, #2
6521:	enable_irq
653	mov	r0, sp
654	adr	lr, BSYM(ret_from_exception)
655	b	__und_fault
656ENDPROC(__und_usr_fault_32)
657ENDPROC(__und_usr_fault_16)
658
659	.align	5
660__pabt_usr:
661	usr_entry
662	mov	r2, sp				@ regs
663	pabt_helper
664 UNWIND(.fnend		)
665	/* fall through */
666/*
667 * This is the return code to user mode for abort handlers
668 */
669ENTRY(ret_from_exception)
670 UNWIND(.fnstart	)
671 UNWIND(.cantunwind	)
672	get_thread_info tsk
673	mov	why, #0
674	b	ret_to_user
675 UNWIND(.fnend		)
676ENDPROC(__pabt_usr)
677ENDPROC(ret_from_exception)
678
679/*
680 * Register switch for ARMv3 and ARMv4 processors
681 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
682 * previous and next are guaranteed not to be the same.
683 */
684ENTRY(__switch_to)
685 UNWIND(.fnstart	)
686 UNWIND(.cantunwind	)
687	add	ip, r1, #TI_CPU_SAVE
688	ldr	r3, [r2, #TI_TP_VALUE]
689 ARM(	stmia	ip!, {r4 - sl, fp, sp, lr} )	@ Store most regs on stack
690 THUMB(	stmia	ip!, {r4 - sl, fp}	   )	@ Store most regs on stack
691 THUMB(	str	sp, [ip], #4		   )
692 THUMB(	str	lr, [ip], #4		   )
693#ifdef CONFIG_CPU_USE_DOMAINS
694	ldr	r6, [r2, #TI_CPU_DOMAIN]
695#endif
696	set_tls	r3, r4, r5
697#if defined(CONFIG_CC_STACKPROTECTOR) && !defined(CONFIG_SMP)
698	ldr	r7, [r2, #TI_TASK]
699	ldr	r8, =__stack_chk_guard
700	ldr	r7, [r7, #TSK_STACK_CANARY]
701#endif
702#ifdef CONFIG_CPU_USE_DOMAINS
703	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
704#endif
705	mov	r5, r0
706	add	r4, r2, #TI_CPU_SAVE
707	ldr	r0, =thread_notify_head
708	mov	r1, #THREAD_NOTIFY_SWITCH
709	bl	atomic_notifier_call_chain
710#if defined(CONFIG_CC_STACKPROTECTOR) && !defined(CONFIG_SMP)
711	str	r7, [r8]
712#endif
713 THUMB(	mov	ip, r4			   )
714	mov	r0, r5
715 ARM(	ldmia	r4, {r4 - sl, fp, sp, pc}  )	@ Load all regs saved previously
716 THUMB(	ldmia	ip!, {r4 - sl, fp}	   )	@ Load all regs saved previously
717 THUMB(	ldr	sp, [ip], #4		   )
718 THUMB(	ldr	pc, [ip]		   )
719 UNWIND(.fnend		)
720ENDPROC(__switch_to)
721
722	__INIT
723
724/*
725 * User helpers.
726 *
727 * Each segment is 32-byte aligned and will be moved to the top of the high
728 * vector page.  New segments (if ever needed) must be added in front of
729 * existing ones.  This mechanism should be used only for things that are
730 * really small and justified, and not be abused freely.
731 *
732 * See Documentation/arm/kernel_user_helpers.txt for formal definitions.
733 */
734 THUMB(	.arm	)
735
736	.macro	usr_ret, reg
737#ifdef CONFIG_ARM_THUMB
738	bx	\reg
739#else
740	mov	pc, \reg
741#endif
742	.endm
743
744	.align	5
745	.globl	__kuser_helper_start
746__kuser_helper_start:
747
748/*
749 * Due to the length of some sequences, __kuser_cmpxchg64 spans 2 regular
750 * kuser "slots", therefore 0xffff0f80 is not used as a valid entry point.
751 */
752
753__kuser_cmpxchg64:				@ 0xffff0f60
754
755#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
756
757	/*
758	 * Poor you.  No fast solution possible...
759	 * The kernel itself must perform the operation.
760	 * A special ghost syscall is used for that (see traps.c).
761	 */
762	stmfd	sp!, {r7, lr}
763	ldr	r7, 1f			@ it's 20 bits
764	swi	__ARM_NR_cmpxchg64
765	ldmfd	sp!, {r7, pc}
7661:	.word	__ARM_NR_cmpxchg64
767
768#elif defined(CONFIG_CPU_32v6K)
769
770	stmfd	sp!, {r4, r5, r6, r7}
771	ldrd	r4, r5, [r0]			@ load old val
772	ldrd	r6, r7, [r1]			@ load new val
773	smp_dmb	arm
7741:	ldrexd	r0, r1, [r2]			@ load current val
775	eors	r3, r0, r4			@ compare with oldval (1)
776	eoreqs	r3, r1, r5			@ compare with oldval (2)
777	strexdeq r3, r6, r7, [r2]		@ store newval if eq
778	teqeq	r3, #1				@ success?
779	beq	1b				@ if no then retry
780	smp_dmb	arm
781	rsbs	r0, r3, #0			@ set returned val and C flag
782	ldmfd	sp!, {r4, r5, r6, r7}
783	usr_ret	lr
784
785#elif !defined(CONFIG_SMP)
786
787#ifdef CONFIG_MMU
788
789	/*
790	 * The only thing that can break atomicity in this cmpxchg64
791	 * implementation is either an IRQ or a data abort exception
792	 * causing another process/thread to be scheduled in the middle of
793	 * the critical sequence.  The same strategy as for cmpxchg is used.
794	 */
795	stmfd	sp!, {r4, r5, r6, lr}
796	ldmia	r0, {r4, r5}			@ load old val
797	ldmia	r1, {r6, lr}			@ load new val
7981:	ldmia	r2, {r0, r1}			@ load current val
799	eors	r3, r0, r4			@ compare with oldval (1)
800	eoreqs	r3, r1, r5			@ compare with oldval (2)
8012:	stmeqia	r2, {r6, lr}			@ store newval if eq
802	rsbs	r0, r3, #0			@ set return val and C flag
803	ldmfd	sp!, {r4, r5, r6, pc}
804
805	.text
806kuser_cmpxchg64_fixup:
807	@ Called from kuser_cmpxchg_fixup.
808	@ r4 = address of interrupted insn (must be preserved).
809	@ sp = saved regs. r7 and r8 are clobbered.
810	@ 1b = first critical insn, 2b = last critical insn.
811	@ If r4 >= 1b and r4 <= 2b then saved pc_usr is set to 1b.
812	mov	r7, #0xffff0fff
813	sub	r7, r7, #(0xffff0fff - (0xffff0f60 + (1b - __kuser_cmpxchg64)))
814	subs	r8, r4, r7
815	rsbcss	r8, r8, #(2b - 1b)
816	strcs	r7, [sp, #S_PC]
817#if __LINUX_ARM_ARCH__ < 6
818	bcc	kuser_cmpxchg32_fixup
819#endif
820	mov	pc, lr
821	.previous
822
823#else
824#warning "NPTL on non MMU needs fixing"
825	mov	r0, #-1
826	adds	r0, r0, #0
827	usr_ret	lr
828#endif
829
830#else
831#error "incoherent kernel configuration"
832#endif
833
834	/* pad to next slot */
835	.rept	(16 - (. - __kuser_cmpxchg64)/4)
836	.word	0
837	.endr
838
839	.align	5
840
841__kuser_memory_barrier:				@ 0xffff0fa0
842	smp_dmb	arm
843	usr_ret	lr
844
845	.align	5
846
847__kuser_cmpxchg:				@ 0xffff0fc0
848
849#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
850
851	/*
852	 * Poor you.  No fast solution possible...
853	 * The kernel itself must perform the operation.
854	 * A special ghost syscall is used for that (see traps.c).
855	 */
856	stmfd	sp!, {r7, lr}
857	ldr	r7, 1f			@ it's 20 bits
858	swi	__ARM_NR_cmpxchg
859	ldmfd	sp!, {r7, pc}
8601:	.word	__ARM_NR_cmpxchg
861
862#elif __LINUX_ARM_ARCH__ < 6
863
864#ifdef CONFIG_MMU
865
866	/*
867	 * The only thing that can break atomicity in this cmpxchg
868	 * implementation is either an IRQ or a data abort exception
869	 * causing another process/thread to be scheduled in the middle
870	 * of the critical sequence.  To prevent this, code is added to
871	 * the IRQ and data abort exception handlers to set the pc back
872	 * to the beginning of the critical section if it is found to be
873	 * within that critical section (see kuser_cmpxchg_fixup).
874	 */
8751:	ldr	r3, [r2]			@ load current val
876	subs	r3, r3, r0			@ compare with oldval
8772:	streq	r1, [r2]			@ store newval if eq
878	rsbs	r0, r3, #0			@ set return val and C flag
879	usr_ret	lr
880
881	.text
882kuser_cmpxchg32_fixup:
883	@ Called from kuser_cmpxchg_check macro.
884	@ r4 = address of interrupted insn (must be preserved).
885	@ sp = saved regs. r7 and r8 are clobbered.
886	@ 1b = first critical insn, 2b = last critical insn.
887	@ If r4 >= 1b and r4 <= 2b then saved pc_usr is set to 1b.
888	mov	r7, #0xffff0fff
889	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
890	subs	r8, r4, r7
891	rsbcss	r8, r8, #(2b - 1b)
892	strcs	r7, [sp, #S_PC]
893	mov	pc, lr
894	.previous
895
896#else
897#warning "NPTL on non MMU needs fixing"
898	mov	r0, #-1
899	adds	r0, r0, #0
900	usr_ret	lr
901#endif
902
903#else
904
905	smp_dmb	arm
9061:	ldrex	r3, [r2]
907	subs	r3, r3, r0
908	strexeq	r3, r1, [r2]
909	teqeq	r3, #1
910	beq	1b
911	rsbs	r0, r3, #0
912	/* beware -- each __kuser slot must be 8 instructions max */
913	ALT_SMP(b	__kuser_memory_barrier)
914	ALT_UP(usr_ret	lr)
915
916#endif
917
918	.align	5
919
920__kuser_get_tls:				@ 0xffff0fe0
921	ldr	r0, [pc, #(16 - 8)]	@ read TLS, set in kuser_get_tls_init
922	usr_ret	lr
923	mrc	p15, 0, r0, c13, c0, 3	@ 0xffff0fe8 hardware TLS code
924	.rep	4
925	.word	0			@ 0xffff0ff0 software TLS value, then
926	.endr				@ pad up to __kuser_helper_version
927
928__kuser_helper_version:				@ 0xffff0ffc
929	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
930
931	.globl	__kuser_helper_end
932__kuser_helper_end:
933
934 THUMB(	.thumb	)
935
936/*
937 * Vector stubs.
938 *
939 * This code is copied to 0xffff0200 so we can use branches in the
940 * vectors, rather than ldr's.  Note that this code must not
941 * exceed 0x300 bytes.
942 *
943 * Common stub entry macro:
944 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
945 *
946 * SP points to a minimal amount of processor-private memory, the address
947 * of which is copied into r0 for the mode specific abort handler.
948 */
949	.macro	vector_stub, name, mode, correction=0
950	.align	5
951
952vector_\name:
953	.if \correction
954	sub	lr, lr, #\correction
955	.endif
956
957	@
958	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
959	@ (parent CPSR)
960	@
961	stmia	sp, {r0, lr}		@ save r0, lr
962	mrs	lr, spsr
963	str	lr, [sp, #8]		@ save spsr
964
965	@
966	@ Prepare for SVC32 mode.  IRQs remain disabled.
967	@
968	mrs	r0, cpsr
969	eor	r0, r0, #(\mode ^ SVC_MODE | PSR_ISETSTATE)
970	msr	spsr_cxsf, r0
971
972	@
973	@ the branch table must immediately follow this code
974	@
975	and	lr, lr, #0x0f
976 THUMB(	adr	r0, 1f			)
977 THUMB(	ldr	lr, [r0, lr, lsl #2]	)
978	mov	r0, sp
979 ARM(	ldr	lr, [pc, lr, lsl #2]	)
980	movs	pc, lr			@ branch to handler in SVC mode
981ENDPROC(vector_\name)
982
983	.align	2
984	@ handler addresses follow this label
9851:
986	.endm
987
988	.globl	__stubs_start
989__stubs_start:
990/*
991 * Interrupt dispatcher
992 */
993	vector_stub	irq, IRQ_MODE, 4
994
995	.long	__irq_usr			@  0  (USR_26 / USR_32)
996	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
997	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
998	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
999	.long	__irq_invalid			@  4
1000	.long	__irq_invalid			@  5
1001	.long	__irq_invalid			@  6
1002	.long	__irq_invalid			@  7
1003	.long	__irq_invalid			@  8
1004	.long	__irq_invalid			@  9
1005	.long	__irq_invalid			@  a
1006	.long	__irq_invalid			@  b
1007	.long	__irq_invalid			@  c
1008	.long	__irq_invalid			@  d
1009	.long	__irq_invalid			@  e
1010	.long	__irq_invalid			@  f
1011
1012/*
1013 * Data abort dispatcher
1014 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1015 */
1016	vector_stub	dabt, ABT_MODE, 8
1017
1018	.long	__dabt_usr			@  0  (USR_26 / USR_32)
1019	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
1020	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
1021	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
1022	.long	__dabt_invalid			@  4
1023	.long	__dabt_invalid			@  5
1024	.long	__dabt_invalid			@  6
1025	.long	__dabt_invalid			@  7
1026	.long	__dabt_invalid			@  8
1027	.long	__dabt_invalid			@  9
1028	.long	__dabt_invalid			@  a
1029	.long	__dabt_invalid			@  b
1030	.long	__dabt_invalid			@  c
1031	.long	__dabt_invalid			@  d
1032	.long	__dabt_invalid			@  e
1033	.long	__dabt_invalid			@  f
1034
1035/*
1036 * Prefetch abort dispatcher
1037 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1038 */
1039	vector_stub	pabt, ABT_MODE, 4
1040
1041	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1042	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1043	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1044	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1045	.long	__pabt_invalid			@  4
1046	.long	__pabt_invalid			@  5
1047	.long	__pabt_invalid			@  6
1048	.long	__pabt_invalid			@  7
1049	.long	__pabt_invalid			@  8
1050	.long	__pabt_invalid			@  9
1051	.long	__pabt_invalid			@  a
1052	.long	__pabt_invalid			@  b
1053	.long	__pabt_invalid			@  c
1054	.long	__pabt_invalid			@  d
1055	.long	__pabt_invalid			@  e
1056	.long	__pabt_invalid			@  f
1057
1058/*
1059 * Undef instr entry dispatcher
1060 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1061 */
1062	vector_stub	und, UND_MODE
1063
1064	.long	__und_usr			@  0 (USR_26 / USR_32)
1065	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1066	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1067	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1068	.long	__und_invalid			@  4
1069	.long	__und_invalid			@  5
1070	.long	__und_invalid			@  6
1071	.long	__und_invalid			@  7
1072	.long	__und_invalid			@  8
1073	.long	__und_invalid			@  9
1074	.long	__und_invalid			@  a
1075	.long	__und_invalid			@  b
1076	.long	__und_invalid			@  c
1077	.long	__und_invalid			@  d
1078	.long	__und_invalid			@  e
1079	.long	__und_invalid			@  f
1080
1081	.align	5
1082
1083/*=============================================================================
1084 * Undefined FIQs
1085 *-----------------------------------------------------------------------------
1086 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1087 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1088 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1089 * damage alert!  I don't think that we can execute any code in here in any
1090 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1091 * get out of that mode without clobbering one register.
1092 */
1093vector_fiq:
1094	subs	pc, lr, #4
1095
1096/*=============================================================================
1097 * Address exception handler
1098 *-----------------------------------------------------------------------------
1099 * These aren't too critical.
1100 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1101 */
1102
1103vector_addrexcptn:
1104	b	vector_addrexcptn
1105
1106/*
1107 * We group all the following data together to optimise
1108 * for CPUs with separate I & D caches.
1109 */
1110	.align	5
1111
1112.LCvswi:
1113	.word	vector_swi
1114
1115	.globl	__stubs_end
1116__stubs_end:
1117
1118	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1119
1120	.globl	__vectors_start
1121__vectors_start:
1122 ARM(	swi	SYS_ERROR0	)
1123 THUMB(	svc	#0		)
1124 THUMB(	nop			)
1125	W(b)	vector_und + stubs_offset
1126	W(ldr)	pc, .LCvswi + stubs_offset
1127	W(b)	vector_pabt + stubs_offset
1128	W(b)	vector_dabt + stubs_offset
1129	W(b)	vector_addrexcptn + stubs_offset
1130	W(b)	vector_irq + stubs_offset
1131	W(b)	vector_fiq + stubs_offset
1132
1133	.globl	__vectors_end
1134__vectors_end:
1135
1136	.data
1137
1138	.globl	cr_alignment
1139	.globl	cr_no_alignment
1140cr_alignment:
1141	.space	4
1142cr_no_alignment:
1143	.space	4
1144
1145#ifdef CONFIG_MULTI_IRQ_HANDLER
1146	.globl	handle_arch_irq
1147handle_arch_irq:
1148	.space	4
1149#endif
1150