xref: /linux/arch/arm/kernel/entry-armv.S (revision 4c8f1cb266cba4d1052f524d04df839d8f732ace)
1/*
2 *  linux/arch/arm/kernel/entry-armv.S
3 *
4 *  Copyright (C) 1996,1997,1998 Russell King.
5 *  ARM700 fix by Matthew Godbolt (linux-user@willothewisp.demon.co.uk)
6 *  nommu support by Hyok S. Choi (hyok.choi@samsung.com)
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
11 *
12 *  Low-level vector interface routines
13 *
14 *  Note:  there is a StrongARM bug in the STMIA rn, {regs}^ instruction
15 *  that causes it to save wrong values...  Be aware!
16 */
17
18#include <asm/memory.h>
19#include <asm/glue.h>
20#include <asm/vfpmacros.h>
21#include <mach/entry-macro.S>
22#include <asm/thread_notify.h>
23#include <asm/unwind.h>
24
25#include "entry-header.S"
26
27/*
28 * Interrupt handling.  Preserves r7, r8, r9
29 */
30	.macro	irq_handler
31	get_irqnr_preamble r5, lr
321:	get_irqnr_and_base r0, r6, r5, lr
33	movne	r1, sp
34	@
35	@ routine called with r0 = irq number, r1 = struct pt_regs *
36	@
37	adrne	lr, BSYM(1b)
38	bne	asm_do_IRQ
39
40#ifdef CONFIG_SMP
41	/*
42	 * XXX
43	 *
44	 * this macro assumes that irqstat (r6) and base (r5) are
45	 * preserved from get_irqnr_and_base above
46	 */
47	test_for_ipi r0, r6, r5, lr
48	movne	r0, sp
49	adrne	lr, BSYM(1b)
50	bne	do_IPI
51
52#ifdef CONFIG_LOCAL_TIMERS
53	test_for_ltirq r0, r6, r5, lr
54	movne	r0, sp
55	adrne	lr, BSYM(1b)
56	bne	do_local_timer
57#endif
58#endif
59
60	.endm
61
62#ifdef CONFIG_KPROBES
63	.section	.kprobes.text,"ax",%progbits
64#else
65	.text
66#endif
67
68/*
69 * Invalid mode handlers
70 */
71	.macro	inv_entry, reason
72	sub	sp, sp, #S_FRAME_SIZE
73 ARM(	stmib	sp, {r1 - lr}		)
74 THUMB(	stmia	sp, {r0 - r12}		)
75 THUMB(	str	sp, [sp, #S_SP]		)
76 THUMB(	str	lr, [sp, #S_LR]		)
77	mov	r1, #\reason
78	.endm
79
80__pabt_invalid:
81	inv_entry BAD_PREFETCH
82	b	common_invalid
83ENDPROC(__pabt_invalid)
84
85__dabt_invalid:
86	inv_entry BAD_DATA
87	b	common_invalid
88ENDPROC(__dabt_invalid)
89
90__irq_invalid:
91	inv_entry BAD_IRQ
92	b	common_invalid
93ENDPROC(__irq_invalid)
94
95__und_invalid:
96	inv_entry BAD_UNDEFINSTR
97
98	@
99	@ XXX fall through to common_invalid
100	@
101
102@
103@ common_invalid - generic code for failed exception (re-entrant version of handlers)
104@
105common_invalid:
106	zero_fp
107
108	ldmia	r0, {r4 - r6}
109	add	r0, sp, #S_PC		@ here for interlock avoidance
110	mov	r7, #-1			@  ""   ""    ""        ""
111	str	r4, [sp]		@ save preserved r0
112	stmia	r0, {r5 - r7}		@ lr_<exception>,
113					@ cpsr_<exception>, "old_r0"
114
115	mov	r0, sp
116	b	bad_mode
117ENDPROC(__und_invalid)
118
119/*
120 * SVC mode handlers
121 */
122
123#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
124#define SPFIX(code...) code
125#else
126#define SPFIX(code...)
127#endif
128
129	.macro	svc_entry, stack_hole=0
130 UNWIND(.fnstart		)
131 UNWIND(.save {r0 - pc}		)
132	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole - 4)
133#ifdef CONFIG_THUMB2_KERNEL
134 SPFIX(	str	r0, [sp]	)	@ temporarily saved
135 SPFIX(	mov	r0, sp		)
136 SPFIX(	tst	r0, #4		)	@ test original stack alignment
137 SPFIX(	ldr	r0, [sp]	)	@ restored
138#else
139 SPFIX(	tst	sp, #4		)
140#endif
141 SPFIX(	subeq	sp, sp, #4	)
142	stmia	sp, {r1 - r12}
143
144	ldmia	r0, {r1 - r3}
145	add	r5, sp, #S_SP - 4	@ here for interlock avoidance
146	mov	r4, #-1			@  ""  ""      ""       ""
147	add	r0, sp, #(S_FRAME_SIZE + \stack_hole - 4)
148 SPFIX(	addeq	r0, r0, #4	)
149	str	r1, [sp, #-4]!		@ save the "real" r0 copied
150					@ from the exception stack
151
152	mov	r1, lr
153
154	@
155	@ We are now ready to fill in the remaining blanks on the stack:
156	@
157	@  r0 - sp_svc
158	@  r1 - lr_svc
159	@  r2 - lr_<exception>, already fixed up for correct return/restart
160	@  r3 - spsr_<exception>
161	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
162	@
163	stmia	r5, {r0 - r4}
164
165	asm_trace_hardirqs_off
166	.endm
167
168	.align	5
169__dabt_svc:
170	svc_entry
171
172	@
173	@ get ready to re-enable interrupts if appropriate
174	@
175	mrs	r9, cpsr
176	tst	r3, #PSR_I_BIT
177	biceq	r9, r9, #PSR_I_BIT
178
179	@
180	@ Call the processor-specific abort handler:
181	@
182	@  r2 - aborted context pc
183	@  r3 - aborted context cpsr
184	@
185	@ The abort handler must return the aborted address in r0, and
186	@ the fault status register in r1.  r9 must be preserved.
187	@
188#ifdef MULTI_DABORT
189	ldr	r4, .LCprocfns
190	mov	lr, pc
191	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
192#else
193	bl	CPU_DABORT_HANDLER
194#endif
195
196	@
197	@ set desired IRQ state, then call main handler
198	@
199	msr	cpsr_c, r9
200	mov	r2, sp
201	bl	do_DataAbort
202
203	@
204	@ IRQs off again before pulling preserved data off the stack
205	@
206	disable_irq
207
208	@
209	@ restore SPSR and restart the instruction
210	@
211	ldr	r2, [sp, #S_PSR]
212	svc_exit r2				@ return from exception
213 UNWIND(.fnend		)
214ENDPROC(__dabt_svc)
215
216	.align	5
217__irq_svc:
218	svc_entry
219
220#ifdef CONFIG_PREEMPT
221	get_thread_info tsk
222	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
223	add	r7, r8, #1			@ increment it
224	str	r7, [tsk, #TI_PREEMPT]
225#endif
226
227	irq_handler
228#ifdef CONFIG_PREEMPT
229	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
230	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
231	teq	r8, #0				@ if preempt count != 0
232	movne	r0, #0				@ force flags to 0
233	tst	r0, #_TIF_NEED_RESCHED
234	blne	svc_preempt
235#endif
236	ldr	r4, [sp, #S_PSR]		@ irqs are already disabled
237#ifdef CONFIG_TRACE_IRQFLAGS
238	tst	r4, #PSR_I_BIT
239	bleq	trace_hardirqs_on
240#endif
241	svc_exit r4				@ return from exception
242 UNWIND(.fnend		)
243ENDPROC(__irq_svc)
244
245	.ltorg
246
247#ifdef CONFIG_PREEMPT
248svc_preempt:
249	mov	r8, lr
2501:	bl	preempt_schedule_irq		@ irq en/disable is done inside
251	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
252	tst	r0, #_TIF_NEED_RESCHED
253	moveq	pc, r8				@ go again
254	b	1b
255#endif
256
257	.align	5
258__und_svc:
259#ifdef CONFIG_KPROBES
260	@ If a kprobe is about to simulate a "stmdb sp..." instruction,
261	@ it obviously needs free stack space which then will belong to
262	@ the saved context.
263	svc_entry 64
264#else
265	svc_entry
266#endif
267
268	@
269	@ call emulation code, which returns using r9 if it has emulated
270	@ the instruction, or the more conventional lr if we are to treat
271	@ this as a real undefined instruction
272	@
273	@  r0 - instruction
274	@
275#ifndef	CONFIG_THUMB2_KERNEL
276	ldr	r0, [r2, #-4]
277#else
278	ldrh	r0, [r2, #-2]			@ Thumb instruction at LR - 2
279	and	r9, r0, #0xf800
280	cmp	r9, #0xe800			@ 32-bit instruction if xx >= 0
281	ldrhhs	r9, [r2]			@ bottom 16 bits
282	orrhs	r0, r9, r0, lsl #16
283#endif
284	adr	r9, BSYM(1f)
285	bl	call_fpe
286
287	mov	r0, sp				@ struct pt_regs *regs
288	bl	do_undefinstr
289
290	@
291	@ IRQs off again before pulling preserved data off the stack
292	@
2931:	disable_irq
294
295	@
296	@ restore SPSR and restart the instruction
297	@
298	ldr	r2, [sp, #S_PSR]		@ Get SVC cpsr
299	svc_exit r2				@ return from exception
300 UNWIND(.fnend		)
301ENDPROC(__und_svc)
302
303	.align	5
304__pabt_svc:
305	svc_entry
306
307	@
308	@ re-enable interrupts if appropriate
309	@
310	mrs	r9, cpsr
311	tst	r3, #PSR_I_BIT
312	biceq	r9, r9, #PSR_I_BIT
313
314	@
315	@ set args, then call main handler
316	@
317	@  r0 - address of faulting instruction
318	@  r1 - pointer to registers on stack
319	@
320#ifdef MULTI_PABORT
321	mov	r0, r2			@ pass address of aborted instruction.
322	ldr	r4, .LCprocfns
323	mov	lr, pc
324	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
325#else
326	CPU_PABORT_HANDLER(r0, r2)
327#endif
328	msr	cpsr_c, r9			@ Maybe enable interrupts
329	mov	r1, sp				@ regs
330	bl	do_PrefetchAbort		@ call abort handler
331
332	@
333	@ IRQs off again before pulling preserved data off the stack
334	@
335	disable_irq
336
337	@
338	@ restore SPSR and restart the instruction
339	@
340	ldr	r2, [sp, #S_PSR]
341	svc_exit r2				@ return from exception
342 UNWIND(.fnend		)
343ENDPROC(__pabt_svc)
344
345	.align	5
346.LCcralign:
347	.word	cr_alignment
348#ifdef MULTI_DABORT
349.LCprocfns:
350	.word	processor
351#endif
352.LCfp:
353	.word	fp_enter
354
355/*
356 * User mode handlers
357 *
358 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
359 */
360
361#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
362#error "sizeof(struct pt_regs) must be a multiple of 8"
363#endif
364
365	.macro	usr_entry
366 UNWIND(.fnstart	)
367 UNWIND(.cantunwind	)	@ don't unwind the user space
368	sub	sp, sp, #S_FRAME_SIZE
369 ARM(	stmib	sp, {r1 - r12}	)
370 THUMB(	stmia	sp, {r0 - r12}	)
371
372	ldmia	r0, {r1 - r3}
373	add	r0, sp, #S_PC		@ here for interlock avoidance
374	mov	r4, #-1			@  ""  ""     ""        ""
375
376	str	r1, [sp]		@ save the "real" r0 copied
377					@ from the exception stack
378
379	@
380	@ We are now ready to fill in the remaining blanks on the stack:
381	@
382	@  r2 - lr_<exception>, already fixed up for correct return/restart
383	@  r3 - spsr_<exception>
384	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
385	@
386	@ Also, separately save sp_usr and lr_usr
387	@
388	stmia	r0, {r2 - r4}
389 ARM(	stmdb	r0, {sp, lr}^			)
390 THUMB(	store_user_sp_lr r0, r1, S_SP - S_PC	)
391
392	@
393	@ Enable the alignment trap while in kernel mode
394	@
395	alignment_trap r0
396
397	@
398	@ Clear FP to mark the first stack frame
399	@
400	zero_fp
401
402	asm_trace_hardirqs_off
403	.endm
404
405	.macro	kuser_cmpxchg_check
406#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
407#ifndef CONFIG_MMU
408#warning "NPTL on non MMU needs fixing"
409#else
410	@ Make sure our user space atomic helper is restarted
411	@ if it was interrupted in a critical region.  Here we
412	@ perform a quick test inline since it should be false
413	@ 99.9999% of the time.  The rest is done out of line.
414	cmp	r2, #TASK_SIZE
415	blhs	kuser_cmpxchg_fixup
416#endif
417#endif
418	.endm
419
420	.align	5
421__dabt_usr:
422	usr_entry
423	kuser_cmpxchg_check
424
425	@
426	@ Call the processor-specific abort handler:
427	@
428	@  r2 - aborted context pc
429	@  r3 - aborted context cpsr
430	@
431	@ The abort handler must return the aborted address in r0, and
432	@ the fault status register in r1.
433	@
434#ifdef MULTI_DABORT
435	ldr	r4, .LCprocfns
436	mov	lr, pc
437	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
438#else
439	bl	CPU_DABORT_HANDLER
440#endif
441
442	@
443	@ IRQs on, then call the main handler
444	@
445	enable_irq
446	mov	r2, sp
447	adr	lr, BSYM(ret_from_exception)
448	b	do_DataAbort
449 UNWIND(.fnend		)
450ENDPROC(__dabt_usr)
451
452	.align	5
453__irq_usr:
454	usr_entry
455	kuser_cmpxchg_check
456
457	get_thread_info tsk
458#ifdef CONFIG_PREEMPT
459	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
460	add	r7, r8, #1			@ increment it
461	str	r7, [tsk, #TI_PREEMPT]
462#endif
463
464	irq_handler
465#ifdef CONFIG_PREEMPT
466	ldr	r0, [tsk, #TI_PREEMPT]
467	str	r8, [tsk, #TI_PREEMPT]
468	teq	r0, r7
469 ARM(	strne	r0, [r0, -r0]	)
470 THUMB(	movne	r0, #0		)
471 THUMB(	strne	r0, [r0]	)
472#endif
473#ifdef CONFIG_TRACE_IRQFLAGS
474	bl	trace_hardirqs_on
475#endif
476
477	mov	why, #0
478	b	ret_to_user
479 UNWIND(.fnend		)
480ENDPROC(__irq_usr)
481
482	.ltorg
483
484	.align	5
485__und_usr:
486	usr_entry
487
488	@
489	@ fall through to the emulation code, which returns using r9 if
490	@ it has emulated the instruction, or the more conventional lr
491	@ if we are to treat this as a real undefined instruction
492	@
493	@  r0 - instruction
494	@
495	adr	r9, BSYM(ret_from_exception)
496	adr	lr, BSYM(__und_usr_unknown)
497	tst	r3, #PSR_T_BIT			@ Thumb mode?
498	itet	eq				@ explicit IT needed for the 1f label
499	subeq	r4, r2, #4			@ ARM instr at LR - 4
500	subne	r4, r2, #2			@ Thumb instr at LR - 2
5011:	ldreqt	r0, [r4]
502#ifdef CONFIG_CPU_ENDIAN_BE8
503	reveq	r0, r0				@ little endian instruction
504#endif
505	beq	call_fpe
506	@ Thumb instruction
507#if __LINUX_ARM_ARCH__ >= 7
5082:
509 ARM(	ldrht	r5, [r4], #2	)
510 THUMB(	ldrht	r5, [r4]	)
511 THUMB(	add	r4, r4, #2	)
512	and	r0, r5, #0xf800			@ mask bits 111x x... .... ....
513	cmp	r0, #0xe800			@ 32bit instruction if xx != 0
514	blo	__und_usr_unknown
5153:	ldrht	r0, [r4]
516	add	r2, r2, #2			@ r2 is PC + 2, make it PC + 4
517	orr	r0, r0, r5, lsl #16
518#else
519	b	__und_usr_unknown
520#endif
521 UNWIND(.fnend		)
522ENDPROC(__und_usr)
523
524	@
525	@ fallthrough to call_fpe
526	@
527
528/*
529 * The out of line fixup for the ldrt above.
530 */
531	.section .fixup, "ax"
5324:	mov	pc, r9
533	.previous
534	.section __ex_table,"a"
535	.long	1b, 4b
536#if __LINUX_ARM_ARCH__ >= 7
537	.long	2b, 4b
538	.long	3b, 4b
539#endif
540	.previous
541
542/*
543 * Check whether the instruction is a co-processor instruction.
544 * If yes, we need to call the relevant co-processor handler.
545 *
546 * Note that we don't do a full check here for the co-processor
547 * instructions; all instructions with bit 27 set are well
548 * defined.  The only instructions that should fault are the
549 * co-processor instructions.  However, we have to watch out
550 * for the ARM6/ARM7 SWI bug.
551 *
552 * NEON is a special case that has to be handled here. Not all
553 * NEON instructions are co-processor instructions, so we have
554 * to make a special case of checking for them. Plus, there's
555 * five groups of them, so we have a table of mask/opcode pairs
556 * to check against, and if any match then we branch off into the
557 * NEON handler code.
558 *
559 * Emulators may wish to make use of the following registers:
560 *  r0  = instruction opcode.
561 *  r2  = PC+4
562 *  r9  = normal "successful" return address
563 *  r10 = this threads thread_info structure.
564 *  lr  = unrecognised instruction return address
565 */
566	@
567	@ Fall-through from Thumb-2 __und_usr
568	@
569#ifdef CONFIG_NEON
570	adr	r6, .LCneon_thumb_opcodes
571	b	2f
572#endif
573call_fpe:
574#ifdef CONFIG_NEON
575	adr	r6, .LCneon_arm_opcodes
5762:
577	ldr	r7, [r6], #4			@ mask value
578	cmp	r7, #0				@ end mask?
579	beq	1f
580	and	r8, r0, r7
581	ldr	r7, [r6], #4			@ opcode bits matching in mask
582	cmp	r8, r7				@ NEON instruction?
583	bne	2b
584	get_thread_info r10
585	mov	r7, #1
586	strb	r7, [r10, #TI_USED_CP + 10]	@ mark CP#10 as used
587	strb	r7, [r10, #TI_USED_CP + 11]	@ mark CP#11 as used
588	b	do_vfp				@ let VFP handler handle this
5891:
590#endif
591	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
592	tstne	r0, #0x04000000			@ bit 26 set on both ARM and Thumb-2
593#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
594	and	r8, r0, #0x0f000000		@ mask out op-code bits
595	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
596#endif
597	moveq	pc, lr
598	get_thread_info r10			@ get current thread
599	and	r8, r0, #0x00000f00		@ mask out CP number
600 THUMB(	lsr	r8, r8, #8		)
601	mov	r7, #1
602	add	r6, r10, #TI_USED_CP
603 ARM(	strb	r7, [r6, r8, lsr #8]	)	@ set appropriate used_cp[]
604 THUMB(	strb	r7, [r6, r8]		)	@ set appropriate used_cp[]
605#ifdef CONFIG_IWMMXT
606	@ Test if we need to give access to iWMMXt coprocessors
607	ldr	r5, [r10, #TI_FLAGS]
608	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
609	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
610	bcs	iwmmxt_task_enable
611#endif
612 ARM(	add	pc, pc, r8, lsr #6	)
613 THUMB(	lsl	r8, r8, #2		)
614 THUMB(	add	pc, r8			)
615	nop
616
617	W(mov)	pc, lr				@ CP#0
618	W(b)	do_fpe				@ CP#1 (FPE)
619	W(b)	do_fpe				@ CP#2 (FPE)
620	W(mov)	pc, lr				@ CP#3
621#ifdef CONFIG_CRUNCH
622	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
623	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
624	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
625#else
626	W(mov)	pc, lr				@ CP#4
627	W(mov)	pc, lr				@ CP#5
628	W(mov)	pc, lr				@ CP#6
629#endif
630	W(mov)	pc, lr				@ CP#7
631	W(mov)	pc, lr				@ CP#8
632	W(mov)	pc, lr				@ CP#9
633#ifdef CONFIG_VFP
634	W(b)	do_vfp				@ CP#10 (VFP)
635	W(b)	do_vfp				@ CP#11 (VFP)
636#else
637	W(mov)	pc, lr				@ CP#10 (VFP)
638	W(mov)	pc, lr				@ CP#11 (VFP)
639#endif
640	W(mov)	pc, lr				@ CP#12
641	W(mov)	pc, lr				@ CP#13
642	W(mov)	pc, lr				@ CP#14 (Debug)
643	W(mov)	pc, lr				@ CP#15 (Control)
644
645#ifdef CONFIG_NEON
646	.align	6
647
648.LCneon_arm_opcodes:
649	.word	0xfe000000			@ mask
650	.word	0xf2000000			@ opcode
651
652	.word	0xff100000			@ mask
653	.word	0xf4000000			@ opcode
654
655	.word	0x00000000			@ mask
656	.word	0x00000000			@ opcode
657
658.LCneon_thumb_opcodes:
659	.word	0xef000000			@ mask
660	.word	0xef000000			@ opcode
661
662	.word	0xff100000			@ mask
663	.word	0xf9000000			@ opcode
664
665	.word	0x00000000			@ mask
666	.word	0x00000000			@ opcode
667#endif
668
669do_fpe:
670	enable_irq
671	ldr	r4, .LCfp
672	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
673	ldr	pc, [r4]			@ Call FP module USR entry point
674
675/*
676 * The FP module is called with these registers set:
677 *  r0  = instruction
678 *  r2  = PC+4
679 *  r9  = normal "successful" return address
680 *  r10 = FP workspace
681 *  lr  = unrecognised FP instruction return address
682 */
683
684	.data
685ENTRY(fp_enter)
686	.word	no_fp
687	.previous
688
689ENTRY(no_fp)
690	mov	pc, lr
691ENDPROC(no_fp)
692
693__und_usr_unknown:
694	enable_irq
695	mov	r0, sp
696	adr	lr, BSYM(ret_from_exception)
697	b	do_undefinstr
698ENDPROC(__und_usr_unknown)
699
700	.align	5
701__pabt_usr:
702	usr_entry
703
704#ifdef MULTI_PABORT
705	mov	r0, r2			@ pass address of aborted instruction.
706	ldr	r4, .LCprocfns
707	mov	lr, pc
708	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
709#else
710	CPU_PABORT_HANDLER(r0, r2)
711#endif
712	enable_irq				@ Enable interrupts
713	mov	r1, sp				@ regs
714	bl	do_PrefetchAbort		@ call abort handler
715 UNWIND(.fnend		)
716	/* fall through */
717/*
718 * This is the return code to user mode for abort handlers
719 */
720ENTRY(ret_from_exception)
721 UNWIND(.fnstart	)
722 UNWIND(.cantunwind	)
723	get_thread_info tsk
724	mov	why, #0
725	b	ret_to_user
726 UNWIND(.fnend		)
727ENDPROC(__pabt_usr)
728ENDPROC(ret_from_exception)
729
730/*
731 * Register switch for ARMv3 and ARMv4 processors
732 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
733 * previous and next are guaranteed not to be the same.
734 */
735ENTRY(__switch_to)
736 UNWIND(.fnstart	)
737 UNWIND(.cantunwind	)
738	add	ip, r1, #TI_CPU_SAVE
739	ldr	r3, [r2, #TI_TP_VALUE]
740 ARM(	stmia	ip!, {r4 - sl, fp, sp, lr} )	@ Store most regs on stack
741 THUMB(	stmia	ip!, {r4 - sl, fp}	   )	@ Store most regs on stack
742 THUMB(	str	sp, [ip], #4		   )
743 THUMB(	str	lr, [ip], #4		   )
744#ifdef CONFIG_MMU
745	ldr	r6, [r2, #TI_CPU_DOMAIN]
746#endif
747#if defined(CONFIG_HAS_TLS_REG)
748	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
749#elif !defined(CONFIG_TLS_REG_EMUL)
750	mov	r4, #0xffff0fff
751	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
752#endif
753#ifdef CONFIG_MMU
754	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
755#endif
756	mov	r5, r0
757	add	r4, r2, #TI_CPU_SAVE
758	ldr	r0, =thread_notify_head
759	mov	r1, #THREAD_NOTIFY_SWITCH
760	bl	atomic_notifier_call_chain
761 THUMB(	mov	ip, r4			   )
762	mov	r0, r5
763 ARM(	ldmia	r4, {r4 - sl, fp, sp, pc}  )	@ Load all regs saved previously
764 THUMB(	ldmia	ip!, {r4 - sl, fp}	   )	@ Load all regs saved previously
765 THUMB(	ldr	sp, [ip], #4		   )
766 THUMB(	ldr	pc, [ip]		   )
767 UNWIND(.fnend		)
768ENDPROC(__switch_to)
769
770	__INIT
771
772/*
773 * User helpers.
774 *
775 * These are segment of kernel provided user code reachable from user space
776 * at a fixed address in kernel memory.  This is used to provide user space
777 * with some operations which require kernel help because of unimplemented
778 * native feature and/or instructions in many ARM CPUs. The idea is for
779 * this code to be executed directly in user mode for best efficiency but
780 * which is too intimate with the kernel counter part to be left to user
781 * libraries.  In fact this code might even differ from one CPU to another
782 * depending on the available  instruction set and restrictions like on
783 * SMP systems.  In other words, the kernel reserves the right to change
784 * this code as needed without warning. Only the entry points and their
785 * results are guaranteed to be stable.
786 *
787 * Each segment is 32-byte aligned and will be moved to the top of the high
788 * vector page.  New segments (if ever needed) must be added in front of
789 * existing ones.  This mechanism should be used only for things that are
790 * really small and justified, and not be abused freely.
791 *
792 * User space is expected to implement those things inline when optimizing
793 * for a processor that has the necessary native support, but only if such
794 * resulting binaries are already to be incompatible with earlier ARM
795 * processors due to the use of unsupported instructions other than what
796 * is provided here.  In other words don't make binaries unable to run on
797 * earlier processors just for the sake of not using these kernel helpers
798 * if your compiled code is not going to use the new instructions for other
799 * purpose.
800 */
801 THUMB(	.arm	)
802
803	.macro	usr_ret, reg
804#ifdef CONFIG_ARM_THUMB
805	bx	\reg
806#else
807	mov	pc, \reg
808#endif
809	.endm
810
811	.align	5
812	.globl	__kuser_helper_start
813__kuser_helper_start:
814
815/*
816 * Reference prototype:
817 *
818 *	void __kernel_memory_barrier(void)
819 *
820 * Input:
821 *
822 *	lr = return address
823 *
824 * Output:
825 *
826 *	none
827 *
828 * Clobbered:
829 *
830 *	none
831 *
832 * Definition and user space usage example:
833 *
834 *	typedef void (__kernel_dmb_t)(void);
835 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
836 *
837 * Apply any needed memory barrier to preserve consistency with data modified
838 * manually and __kuser_cmpxchg usage.
839 *
840 * This could be used as follows:
841 *
842 * #define __kernel_dmb() \
843 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
844 *	        : : : "r0", "lr","cc" )
845 */
846
847__kuser_memory_barrier:				@ 0xffff0fa0
848	smp_dmb
849	usr_ret	lr
850
851	.align	5
852
853/*
854 * Reference prototype:
855 *
856 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
857 *
858 * Input:
859 *
860 *	r0 = oldval
861 *	r1 = newval
862 *	r2 = ptr
863 *	lr = return address
864 *
865 * Output:
866 *
867 *	r0 = returned value (zero or non-zero)
868 *	C flag = set if r0 == 0, clear if r0 != 0
869 *
870 * Clobbered:
871 *
872 *	r3, ip, flags
873 *
874 * Definition and user space usage example:
875 *
876 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
877 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
878 *
879 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
880 * Return zero if *ptr was changed or non-zero if no exchange happened.
881 * The C flag is also set if *ptr was changed to allow for assembly
882 * optimization in the calling code.
883 *
884 * Notes:
885 *
886 *    - This routine already includes memory barriers as needed.
887 *
888 * For example, a user space atomic_add implementation could look like this:
889 *
890 * #define atomic_add(ptr, val) \
891 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
892 *	   register unsigned int __result asm("r1"); \
893 *	   asm volatile ( \
894 *	       "1: @ atomic_add\n\t" \
895 *	       "ldr	r0, [r2]\n\t" \
896 *	       "mov	r3, #0xffff0fff\n\t" \
897 *	       "add	lr, pc, #4\n\t" \
898 *	       "add	r1, r0, %2\n\t" \
899 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
900 *	       "bcc	1b" \
901 *	       : "=&r" (__result) \
902 *	       : "r" (__ptr), "rIL" (val) \
903 *	       : "r0","r3","ip","lr","cc","memory" ); \
904 *	   __result; })
905 */
906
907__kuser_cmpxchg:				@ 0xffff0fc0
908
909#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
910
911	/*
912	 * Poor you.  No fast solution possible...
913	 * The kernel itself must perform the operation.
914	 * A special ghost syscall is used for that (see traps.c).
915	 */
916	stmfd	sp!, {r7, lr}
917	mov	r7, #0xff00		@ 0xfff0 into r7 for EABI
918	orr	r7, r7, #0xf0
919	swi	#0x9ffff0
920	ldmfd	sp!, {r7, pc}
921
922#elif __LINUX_ARM_ARCH__ < 6
923
924#ifdef CONFIG_MMU
925
926	/*
927	 * The only thing that can break atomicity in this cmpxchg
928	 * implementation is either an IRQ or a data abort exception
929	 * causing another process/thread to be scheduled in the middle
930	 * of the critical sequence.  To prevent this, code is added to
931	 * the IRQ and data abort exception handlers to set the pc back
932	 * to the beginning of the critical section if it is found to be
933	 * within that critical section (see kuser_cmpxchg_fixup).
934	 */
9351:	ldr	r3, [r2]			@ load current val
936	subs	r3, r3, r0			@ compare with oldval
9372:	streq	r1, [r2]			@ store newval if eq
938	rsbs	r0, r3, #0			@ set return val and C flag
939	usr_ret	lr
940
941	.text
942kuser_cmpxchg_fixup:
943	@ Called from kuser_cmpxchg_check macro.
944	@ r2 = address of interrupted insn (must be preserved).
945	@ sp = saved regs. r7 and r8 are clobbered.
946	@ 1b = first critical insn, 2b = last critical insn.
947	@ If r2 >= 1b and r2 <= 2b then saved pc_usr is set to 1b.
948	mov	r7, #0xffff0fff
949	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
950	subs	r8, r2, r7
951	rsbcss	r8, r8, #(2b - 1b)
952	strcs	r7, [sp, #S_PC]
953	mov	pc, lr
954	.previous
955
956#else
957#warning "NPTL on non MMU needs fixing"
958	mov	r0, #-1
959	adds	r0, r0, #0
960	usr_ret	lr
961#endif
962
963#else
964
965#ifdef CONFIG_SMP
966	mcr	p15, 0, r0, c7, c10, 5	@ dmb
967#endif
9681:	ldrex	r3, [r2]
969	subs	r3, r3, r0
970	strexeq	r3, r1, [r2]
971	teqeq	r3, #1
972	beq	1b
973	rsbs	r0, r3, #0
974	/* beware -- each __kuser slot must be 8 instructions max */
975#ifdef CONFIG_SMP
976	b	__kuser_memory_barrier
977#else
978	usr_ret	lr
979#endif
980
981#endif
982
983	.align	5
984
985/*
986 * Reference prototype:
987 *
988 *	int __kernel_get_tls(void)
989 *
990 * Input:
991 *
992 *	lr = return address
993 *
994 * Output:
995 *
996 *	r0 = TLS value
997 *
998 * Clobbered:
999 *
1000 *	none
1001 *
1002 * Definition and user space usage example:
1003 *
1004 *	typedef int (__kernel_get_tls_t)(void);
1005 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
1006 *
1007 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
1008 *
1009 * This could be used as follows:
1010 *
1011 * #define __kernel_get_tls() \
1012 *	({ register unsigned int __val asm("r0"); \
1013 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
1014 *	        : "=r" (__val) : : "lr","cc" ); \
1015 *	   __val; })
1016 */
1017
1018__kuser_get_tls:				@ 0xffff0fe0
1019
1020#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
1021	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
1022#else
1023	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
1024#endif
1025	usr_ret	lr
1026
1027	.rep	5
1028	.word	0			@ pad up to __kuser_helper_version
1029	.endr
1030
1031/*
1032 * Reference declaration:
1033 *
1034 *	extern unsigned int __kernel_helper_version;
1035 *
1036 * Definition and user space usage example:
1037 *
1038 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
1039 *
1040 * User space may read this to determine the curent number of helpers
1041 * available.
1042 */
1043
1044__kuser_helper_version:				@ 0xffff0ffc
1045	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
1046
1047	.globl	__kuser_helper_end
1048__kuser_helper_end:
1049
1050 THUMB(	.thumb	)
1051
1052/*
1053 * Vector stubs.
1054 *
1055 * This code is copied to 0xffff0200 so we can use branches in the
1056 * vectors, rather than ldr's.  Note that this code must not
1057 * exceed 0x300 bytes.
1058 *
1059 * Common stub entry macro:
1060 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1061 *
1062 * SP points to a minimal amount of processor-private memory, the address
1063 * of which is copied into r0 for the mode specific abort handler.
1064 */
1065	.macro	vector_stub, name, mode, correction=0
1066	.align	5
1067
1068vector_\name:
1069	.if \correction
1070	sub	lr, lr, #\correction
1071	.endif
1072
1073	@
1074	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
1075	@ (parent CPSR)
1076	@
1077	stmia	sp, {r0, lr}		@ save r0, lr
1078	mrs	lr, spsr
1079	str	lr, [sp, #8]		@ save spsr
1080
1081	@
1082	@ Prepare for SVC32 mode.  IRQs remain disabled.
1083	@
1084	mrs	r0, cpsr
1085	eor	r0, r0, #(\mode ^ SVC_MODE | PSR_ISETSTATE)
1086	msr	spsr_cxsf, r0
1087
1088	@
1089	@ the branch table must immediately follow this code
1090	@
1091	and	lr, lr, #0x0f
1092 THUMB(	adr	r0, 1f			)
1093 THUMB(	ldr	lr, [r0, lr, lsl #2]	)
1094	mov	r0, sp
1095 ARM(	ldr	lr, [pc, lr, lsl #2]	)
1096	movs	pc, lr			@ branch to handler in SVC mode
1097ENDPROC(vector_\name)
1098
1099	.align	2
1100	@ handler addresses follow this label
11011:
1102	.endm
1103
1104	.globl	__stubs_start
1105__stubs_start:
1106/*
1107 * Interrupt dispatcher
1108 */
1109	vector_stub	irq, IRQ_MODE, 4
1110
1111	.long	__irq_usr			@  0  (USR_26 / USR_32)
1112	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
1113	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
1114	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
1115	.long	__irq_invalid			@  4
1116	.long	__irq_invalid			@  5
1117	.long	__irq_invalid			@  6
1118	.long	__irq_invalid			@  7
1119	.long	__irq_invalid			@  8
1120	.long	__irq_invalid			@  9
1121	.long	__irq_invalid			@  a
1122	.long	__irq_invalid			@  b
1123	.long	__irq_invalid			@  c
1124	.long	__irq_invalid			@  d
1125	.long	__irq_invalid			@  e
1126	.long	__irq_invalid			@  f
1127
1128/*
1129 * Data abort dispatcher
1130 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1131 */
1132	vector_stub	dabt, ABT_MODE, 8
1133
1134	.long	__dabt_usr			@  0  (USR_26 / USR_32)
1135	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
1136	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
1137	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
1138	.long	__dabt_invalid			@  4
1139	.long	__dabt_invalid			@  5
1140	.long	__dabt_invalid			@  6
1141	.long	__dabt_invalid			@  7
1142	.long	__dabt_invalid			@  8
1143	.long	__dabt_invalid			@  9
1144	.long	__dabt_invalid			@  a
1145	.long	__dabt_invalid			@  b
1146	.long	__dabt_invalid			@  c
1147	.long	__dabt_invalid			@  d
1148	.long	__dabt_invalid			@  e
1149	.long	__dabt_invalid			@  f
1150
1151/*
1152 * Prefetch abort dispatcher
1153 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1154 */
1155	vector_stub	pabt, ABT_MODE, 4
1156
1157	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1158	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1159	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1160	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1161	.long	__pabt_invalid			@  4
1162	.long	__pabt_invalid			@  5
1163	.long	__pabt_invalid			@  6
1164	.long	__pabt_invalid			@  7
1165	.long	__pabt_invalid			@  8
1166	.long	__pabt_invalid			@  9
1167	.long	__pabt_invalid			@  a
1168	.long	__pabt_invalid			@  b
1169	.long	__pabt_invalid			@  c
1170	.long	__pabt_invalid			@  d
1171	.long	__pabt_invalid			@  e
1172	.long	__pabt_invalid			@  f
1173
1174/*
1175 * Undef instr entry dispatcher
1176 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1177 */
1178	vector_stub	und, UND_MODE
1179
1180	.long	__und_usr			@  0 (USR_26 / USR_32)
1181	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1182	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1183	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1184	.long	__und_invalid			@  4
1185	.long	__und_invalid			@  5
1186	.long	__und_invalid			@  6
1187	.long	__und_invalid			@  7
1188	.long	__und_invalid			@  8
1189	.long	__und_invalid			@  9
1190	.long	__und_invalid			@  a
1191	.long	__und_invalid			@  b
1192	.long	__und_invalid			@  c
1193	.long	__und_invalid			@  d
1194	.long	__und_invalid			@  e
1195	.long	__und_invalid			@  f
1196
1197	.align	5
1198
1199/*=============================================================================
1200 * Undefined FIQs
1201 *-----------------------------------------------------------------------------
1202 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1203 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1204 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1205 * damage alert!  I don't think that we can execute any code in here in any
1206 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1207 * get out of that mode without clobbering one register.
1208 */
1209vector_fiq:
1210	disable_fiq
1211	subs	pc, lr, #4
1212
1213/*=============================================================================
1214 * Address exception handler
1215 *-----------------------------------------------------------------------------
1216 * These aren't too critical.
1217 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1218 */
1219
1220vector_addrexcptn:
1221	b	vector_addrexcptn
1222
1223/*
1224 * We group all the following data together to optimise
1225 * for CPUs with separate I & D caches.
1226 */
1227	.align	5
1228
1229.LCvswi:
1230	.word	vector_swi
1231
1232	.globl	__stubs_end
1233__stubs_end:
1234
1235	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1236
1237	.globl	__vectors_start
1238__vectors_start:
1239 ARM(	swi	SYS_ERROR0	)
1240 THUMB(	svc	#0		)
1241 THUMB(	nop			)
1242	W(b)	vector_und + stubs_offset
1243	W(ldr)	pc, .LCvswi + stubs_offset
1244	W(b)	vector_pabt + stubs_offset
1245	W(b)	vector_dabt + stubs_offset
1246	W(b)	vector_addrexcptn + stubs_offset
1247	W(b)	vector_irq + stubs_offset
1248	W(b)	vector_fiq + stubs_offset
1249
1250	.globl	__vectors_end
1251__vectors_end:
1252
1253	.data
1254
1255	.globl	cr_alignment
1256	.globl	cr_no_alignment
1257cr_alignment:
1258	.space	4
1259cr_no_alignment:
1260	.space	4
1261