xref: /linux/arch/arm/kernel/entry-armv.S (revision 092e0e7e520a1fca03e13c9f2d157432a8657ff2)
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#include <asm/unistd.h>
25#include <asm/tls.h>
26
27#include "entry-header.S"
28
29/*
30 * Interrupt handling.  Preserves r7, r8, r9
31 */
32	.macro	irq_handler
33	get_irqnr_preamble r5, lr
341:	get_irqnr_and_base r0, r6, r5, lr
35	movne	r1, sp
36	@
37	@ routine called with r0 = irq number, r1 = struct pt_regs *
38	@
39	adrne	lr, BSYM(1b)
40	bne	asm_do_IRQ
41
42#ifdef CONFIG_SMP
43	/*
44	 * XXX
45	 *
46	 * this macro assumes that irqstat (r6) and base (r5) are
47	 * preserved from get_irqnr_and_base above
48	 */
49	ALT_SMP(test_for_ipi r0, r6, r5, lr)
50	ALT_UP_B(9997f)
51	movne	r0, sp
52	adrne	lr, BSYM(1b)
53	bne	do_IPI
54
55#ifdef CONFIG_LOCAL_TIMERS
56	test_for_ltirq r0, r6, r5, lr
57	movne	r0, sp
58	adrne	lr, BSYM(1b)
59	bne	do_local_timer
60#endif
619997:
62#endif
63
64	.endm
65
66#ifdef CONFIG_KPROBES
67	.section	.kprobes.text,"ax",%progbits
68#else
69	.text
70#endif
71
72/*
73 * Invalid mode handlers
74 */
75	.macro	inv_entry, reason
76	sub	sp, sp, #S_FRAME_SIZE
77 ARM(	stmib	sp, {r1 - lr}		)
78 THUMB(	stmia	sp, {r0 - r12}		)
79 THUMB(	str	sp, [sp, #S_SP]		)
80 THUMB(	str	lr, [sp, #S_LR]		)
81	mov	r1, #\reason
82	.endm
83
84__pabt_invalid:
85	inv_entry BAD_PREFETCH
86	b	common_invalid
87ENDPROC(__pabt_invalid)
88
89__dabt_invalid:
90	inv_entry BAD_DATA
91	b	common_invalid
92ENDPROC(__dabt_invalid)
93
94__irq_invalid:
95	inv_entry BAD_IRQ
96	b	common_invalid
97ENDPROC(__irq_invalid)
98
99__und_invalid:
100	inv_entry BAD_UNDEFINSTR
101
102	@
103	@ XXX fall through to common_invalid
104	@
105
106@
107@ common_invalid - generic code for failed exception (re-entrant version of handlers)
108@
109common_invalid:
110	zero_fp
111
112	ldmia	r0, {r4 - r6}
113	add	r0, sp, #S_PC		@ here for interlock avoidance
114	mov	r7, #-1			@  ""   ""    ""        ""
115	str	r4, [sp]		@ save preserved r0
116	stmia	r0, {r5 - r7}		@ lr_<exception>,
117					@ cpsr_<exception>, "old_r0"
118
119	mov	r0, sp
120	b	bad_mode
121ENDPROC(__und_invalid)
122
123/*
124 * SVC mode handlers
125 */
126
127#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
128#define SPFIX(code...) code
129#else
130#define SPFIX(code...)
131#endif
132
133	.macro	svc_entry, stack_hole=0
134 UNWIND(.fnstart		)
135 UNWIND(.save {r0 - pc}		)
136	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole - 4)
137#ifdef CONFIG_THUMB2_KERNEL
138 SPFIX(	str	r0, [sp]	)	@ temporarily saved
139 SPFIX(	mov	r0, sp		)
140 SPFIX(	tst	r0, #4		)	@ test original stack alignment
141 SPFIX(	ldr	r0, [sp]	)	@ restored
142#else
143 SPFIX(	tst	sp, #4		)
144#endif
145 SPFIX(	subeq	sp, sp, #4	)
146	stmia	sp, {r1 - r12}
147
148	ldmia	r0, {r1 - r3}
149	add	r5, sp, #S_SP - 4	@ here for interlock avoidance
150	mov	r4, #-1			@  ""  ""      ""       ""
151	add	r0, sp, #(S_FRAME_SIZE + \stack_hole - 4)
152 SPFIX(	addeq	r0, r0, #4	)
153	str	r1, [sp, #-4]!		@ save the "real" r0 copied
154					@ from the exception stack
155
156	mov	r1, lr
157
158	@
159	@ We are now ready to fill in the remaining blanks on the stack:
160	@
161	@  r0 - sp_svc
162	@  r1 - lr_svc
163	@  r2 - lr_<exception>, already fixed up for correct return/restart
164	@  r3 - spsr_<exception>
165	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
166	@
167	stmia	r5, {r0 - r4}
168	.endm
169
170	.align	5
171__dabt_svc:
172	svc_entry
173
174	@
175	@ get ready to re-enable interrupts if appropriate
176	@
177	mrs	r9, cpsr
178	tst	r3, #PSR_I_BIT
179	biceq	r9, r9, #PSR_I_BIT
180
181	@
182	@ Call the processor-specific abort handler:
183	@
184	@  r2 - aborted context pc
185	@  r3 - aborted context cpsr
186	@
187	@ The abort handler must return the aborted address in r0, and
188	@ the fault status register in r1.  r9 must be preserved.
189	@
190#ifdef MULTI_DABORT
191	ldr	r4, .LCprocfns
192	mov	lr, pc
193	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
194#else
195	bl	CPU_DABORT_HANDLER
196#endif
197
198	@
199	@ set desired IRQ state, then call main handler
200	@
201	msr	cpsr_c, r9
202	mov	r2, sp
203	bl	do_DataAbort
204
205	@
206	@ IRQs off again before pulling preserved data off the stack
207	@
208	disable_irq_notrace
209
210	@
211	@ restore SPSR and restart the instruction
212	@
213	ldr	r2, [sp, #S_PSR]
214	svc_exit r2				@ return from exception
215 UNWIND(.fnend		)
216ENDPROC(__dabt_svc)
217
218	.align	5
219__irq_svc:
220	svc_entry
221
222#ifdef CONFIG_TRACE_IRQFLAGS
223	bl	trace_hardirqs_off
224#endif
225#ifdef CONFIG_PREEMPT
226	get_thread_info tsk
227	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
228	add	r7, r8, #1			@ increment it
229	str	r7, [tsk, #TI_PREEMPT]
230#endif
231
232	irq_handler
233#ifdef CONFIG_PREEMPT
234	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
235	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
236	teq	r8, #0				@ if preempt count != 0
237	movne	r0, #0				@ force flags to 0
238	tst	r0, #_TIF_NEED_RESCHED
239	blne	svc_preempt
240#endif
241	ldr	r4, [sp, #S_PSR]		@ irqs are already disabled
242#ifdef CONFIG_TRACE_IRQFLAGS
243	tst	r4, #PSR_I_BIT
244	bleq	trace_hardirqs_on
245#endif
246	svc_exit r4				@ return from exception
247 UNWIND(.fnend		)
248ENDPROC(__irq_svc)
249
250	.ltorg
251
252#ifdef CONFIG_PREEMPT
253svc_preempt:
254	mov	r8, lr
2551:	bl	preempt_schedule_irq		@ irq en/disable is done inside
256	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
257	tst	r0, #_TIF_NEED_RESCHED
258	moveq	pc, r8				@ go again
259	b	1b
260#endif
261
262	.align	5
263__und_svc:
264#ifdef CONFIG_KPROBES
265	@ If a kprobe is about to simulate a "stmdb sp..." instruction,
266	@ it obviously needs free stack space which then will belong to
267	@ the saved context.
268	svc_entry 64
269#else
270	svc_entry
271#endif
272
273	@
274	@ call emulation code, which returns using r9 if it has emulated
275	@ the instruction, or the more conventional lr if we are to treat
276	@ this as a real undefined instruction
277	@
278	@  r0 - instruction
279	@
280#ifndef	CONFIG_THUMB2_KERNEL
281	ldr	r0, [r2, #-4]
282#else
283	ldrh	r0, [r2, #-2]			@ Thumb instruction at LR - 2
284	and	r9, r0, #0xf800
285	cmp	r9, #0xe800			@ 32-bit instruction if xx >= 0
286	ldrhhs	r9, [r2]			@ bottom 16 bits
287	orrhs	r0, r9, r0, lsl #16
288#endif
289	adr	r9, BSYM(1f)
290	bl	call_fpe
291
292	mov	r0, sp				@ struct pt_regs *regs
293	bl	do_undefinstr
294
295	@
296	@ IRQs off again before pulling preserved data off the stack
297	@
2981:	disable_irq_notrace
299
300	@
301	@ restore SPSR and restart the instruction
302	@
303	ldr	r2, [sp, #S_PSR]		@ Get SVC cpsr
304	svc_exit r2				@ return from exception
305 UNWIND(.fnend		)
306ENDPROC(__und_svc)
307
308	.align	5
309__pabt_svc:
310	svc_entry
311
312	@
313	@ re-enable interrupts if appropriate
314	@
315	mrs	r9, cpsr
316	tst	r3, #PSR_I_BIT
317	biceq	r9, r9, #PSR_I_BIT
318
319	mov	r0, r2			@ pass address of aborted instruction.
320#ifdef MULTI_PABORT
321	ldr	r4, .LCprocfns
322	mov	lr, pc
323	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
324#else
325	bl	CPU_PABORT_HANDLER
326#endif
327	msr	cpsr_c, r9			@ Maybe enable interrupts
328	mov	r2, sp				@ regs
329	bl	do_PrefetchAbort		@ call abort handler
330
331	@
332	@ IRQs off again before pulling preserved data off the stack
333	@
334	disable_irq_notrace
335
336	@
337	@ restore SPSR and restart the instruction
338	@
339	ldr	r2, [sp, #S_PSR]
340	svc_exit r2				@ return from exception
341 UNWIND(.fnend		)
342ENDPROC(__pabt_svc)
343
344	.align	5
345.LCcralign:
346	.word	cr_alignment
347#ifdef MULTI_DABORT
348.LCprocfns:
349	.word	processor
350#endif
351.LCfp:
352	.word	fp_enter
353
354/*
355 * User mode handlers
356 *
357 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
358 */
359
360#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
361#error "sizeof(struct pt_regs) must be a multiple of 8"
362#endif
363
364	.macro	usr_entry
365 UNWIND(.fnstart	)
366 UNWIND(.cantunwind	)	@ don't unwind the user space
367	sub	sp, sp, #S_FRAME_SIZE
368 ARM(	stmib	sp, {r1 - r12}	)
369 THUMB(	stmia	sp, {r0 - r12}	)
370
371	ldmia	r0, {r1 - r3}
372	add	r0, sp, #S_PC		@ here for interlock avoidance
373	mov	r4, #-1			@  ""  ""     ""        ""
374
375	str	r1, [sp]		@ save the "real" r0 copied
376					@ from the exception stack
377
378	@
379	@ We are now ready to fill in the remaining blanks on the stack:
380	@
381	@  r2 - lr_<exception>, already fixed up for correct return/restart
382	@  r3 - spsr_<exception>
383	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
384	@
385	@ Also, separately save sp_usr and lr_usr
386	@
387	stmia	r0, {r2 - r4}
388 ARM(	stmdb	r0, {sp, lr}^			)
389 THUMB(	store_user_sp_lr r0, r1, S_SP - S_PC	)
390
391	@
392	@ Enable the alignment trap while in kernel mode
393	@
394	alignment_trap r0
395
396	@
397	@ Clear FP to mark the first stack frame
398	@
399	zero_fp
400	.endm
401
402	.macro	kuser_cmpxchg_check
403#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
404#ifndef CONFIG_MMU
405#warning "NPTL on non MMU needs fixing"
406#else
407	@ Make sure our user space atomic helper is restarted
408	@ if it was interrupted in a critical region.  Here we
409	@ perform a quick test inline since it should be false
410	@ 99.9999% of the time.  The rest is done out of line.
411	cmp	r2, #TASK_SIZE
412	blhs	kuser_cmpxchg_fixup
413#endif
414#endif
415	.endm
416
417	.align	5
418__dabt_usr:
419	usr_entry
420	kuser_cmpxchg_check
421
422	@
423	@ Call the processor-specific abort handler:
424	@
425	@  r2 - aborted context pc
426	@  r3 - aborted context cpsr
427	@
428	@ The abort handler must return the aborted address in r0, and
429	@ the fault status register in r1.
430	@
431#ifdef MULTI_DABORT
432	ldr	r4, .LCprocfns
433	mov	lr, pc
434	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
435#else
436	bl	CPU_DABORT_HANDLER
437#endif
438
439	@
440	@ IRQs on, then call the main handler
441	@
442	enable_irq
443	mov	r2, sp
444	adr	lr, BSYM(ret_from_exception)
445	b	do_DataAbort
446 UNWIND(.fnend		)
447ENDPROC(__dabt_usr)
448
449	.align	5
450__irq_usr:
451	usr_entry
452	kuser_cmpxchg_check
453
454	get_thread_info tsk
455#ifdef CONFIG_PREEMPT
456	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
457	add	r7, r8, #1			@ increment it
458	str	r7, [tsk, #TI_PREEMPT]
459#endif
460
461	irq_handler
462#ifdef CONFIG_PREEMPT
463	ldr	r0, [tsk, #TI_PREEMPT]
464	str	r8, [tsk, #TI_PREEMPT]
465	teq	r0, r7
466 ARM(	strne	r0, [r0, -r0]	)
467 THUMB(	movne	r0, #0		)
468 THUMB(	strne	r0, [r0]	)
469#endif
470
471	mov	why, #0
472	b	ret_to_user
473 UNWIND(.fnend		)
474ENDPROC(__irq_usr)
475
476	.ltorg
477
478	.align	5
479__und_usr:
480	usr_entry
481
482	@
483	@ fall through to the emulation code, which returns using r9 if
484	@ it has emulated the instruction, or the more conventional lr
485	@ if we are to treat this as a real undefined instruction
486	@
487	@  r0 - instruction
488	@
489	adr	r9, BSYM(ret_from_exception)
490	adr	lr, BSYM(__und_usr_unknown)
491	tst	r3, #PSR_T_BIT			@ Thumb mode?
492	itet	eq				@ explicit IT needed for the 1f label
493	subeq	r4, r2, #4			@ ARM instr at LR - 4
494	subne	r4, r2, #2			@ Thumb instr at LR - 2
4951:	ldreqt	r0, [r4]
496#ifdef CONFIG_CPU_ENDIAN_BE8
497	reveq	r0, r0				@ little endian instruction
498#endif
499	beq	call_fpe
500	@ Thumb instruction
501#if __LINUX_ARM_ARCH__ >= 7
5022:
503 ARM(	ldrht	r5, [r4], #2	)
504 THUMB(	ldrht	r5, [r4]	)
505 THUMB(	add	r4, r4, #2	)
506	and	r0, r5, #0xf800			@ mask bits 111x x... .... ....
507	cmp	r0, #0xe800			@ 32bit instruction if xx != 0
508	blo	__und_usr_unknown
5093:	ldrht	r0, [r4]
510	add	r2, r2, #2			@ r2 is PC + 2, make it PC + 4
511	orr	r0, r0, r5, lsl #16
512#else
513	b	__und_usr_unknown
514#endif
515 UNWIND(.fnend		)
516ENDPROC(__und_usr)
517
518	@
519	@ fallthrough to call_fpe
520	@
521
522/*
523 * The out of line fixup for the ldrt above.
524 */
525	.pushsection .fixup, "ax"
5264:	mov	pc, r9
527	.popsection
528	.pushsection __ex_table,"a"
529	.long	1b, 4b
530#if __LINUX_ARM_ARCH__ >= 7
531	.long	2b, 4b
532	.long	3b, 4b
533#endif
534	.popsection
535
536/*
537 * Check whether the instruction is a co-processor instruction.
538 * If yes, we need to call the relevant co-processor handler.
539 *
540 * Note that we don't do a full check here for the co-processor
541 * instructions; all instructions with bit 27 set are well
542 * defined.  The only instructions that should fault are the
543 * co-processor instructions.  However, we have to watch out
544 * for the ARM6/ARM7 SWI bug.
545 *
546 * NEON is a special case that has to be handled here. Not all
547 * NEON instructions are co-processor instructions, so we have
548 * to make a special case of checking for them. Plus, there's
549 * five groups of them, so we have a table of mask/opcode pairs
550 * to check against, and if any match then we branch off into the
551 * NEON handler code.
552 *
553 * Emulators may wish to make use of the following registers:
554 *  r0  = instruction opcode.
555 *  r2  = PC+4
556 *  r9  = normal "successful" return address
557 *  r10 = this threads thread_info structure.
558 *  lr  = unrecognised instruction return address
559 */
560	@
561	@ Fall-through from Thumb-2 __und_usr
562	@
563#ifdef CONFIG_NEON
564	adr	r6, .LCneon_thumb_opcodes
565	b	2f
566#endif
567call_fpe:
568#ifdef CONFIG_NEON
569	adr	r6, .LCneon_arm_opcodes
5702:
571	ldr	r7, [r6], #4			@ mask value
572	cmp	r7, #0				@ end mask?
573	beq	1f
574	and	r8, r0, r7
575	ldr	r7, [r6], #4			@ opcode bits matching in mask
576	cmp	r8, r7				@ NEON instruction?
577	bne	2b
578	get_thread_info r10
579	mov	r7, #1
580	strb	r7, [r10, #TI_USED_CP + 10]	@ mark CP#10 as used
581	strb	r7, [r10, #TI_USED_CP + 11]	@ mark CP#11 as used
582	b	do_vfp				@ let VFP handler handle this
5831:
584#endif
585	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
586	tstne	r0, #0x04000000			@ bit 26 set on both ARM and Thumb-2
587#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
588	and	r8, r0, #0x0f000000		@ mask out op-code bits
589	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
590#endif
591	moveq	pc, lr
592	get_thread_info r10			@ get current thread
593	and	r8, r0, #0x00000f00		@ mask out CP number
594 THUMB(	lsr	r8, r8, #8		)
595	mov	r7, #1
596	add	r6, r10, #TI_USED_CP
597 ARM(	strb	r7, [r6, r8, lsr #8]	)	@ set appropriate used_cp[]
598 THUMB(	strb	r7, [r6, r8]		)	@ set appropriate used_cp[]
599#ifdef CONFIG_IWMMXT
600	@ Test if we need to give access to iWMMXt coprocessors
601	ldr	r5, [r10, #TI_FLAGS]
602	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
603	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
604	bcs	iwmmxt_task_enable
605#endif
606 ARM(	add	pc, pc, r8, lsr #6	)
607 THUMB(	lsl	r8, r8, #2		)
608 THUMB(	add	pc, r8			)
609	nop
610
611	movw_pc	lr				@ CP#0
612	W(b)	do_fpe				@ CP#1 (FPE)
613	W(b)	do_fpe				@ CP#2 (FPE)
614	movw_pc	lr				@ CP#3
615#ifdef CONFIG_CRUNCH
616	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
617	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
618	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
619#else
620	movw_pc	lr				@ CP#4
621	movw_pc	lr				@ CP#5
622	movw_pc	lr				@ CP#6
623#endif
624	movw_pc	lr				@ CP#7
625	movw_pc	lr				@ CP#8
626	movw_pc	lr				@ CP#9
627#ifdef CONFIG_VFP
628	W(b)	do_vfp				@ CP#10 (VFP)
629	W(b)	do_vfp				@ CP#11 (VFP)
630#else
631	movw_pc	lr				@ CP#10 (VFP)
632	movw_pc	lr				@ CP#11 (VFP)
633#endif
634	movw_pc	lr				@ CP#12
635	movw_pc	lr				@ CP#13
636	movw_pc	lr				@ CP#14 (Debug)
637	movw_pc	lr				@ CP#15 (Control)
638
639#ifdef CONFIG_NEON
640	.align	6
641
642.LCneon_arm_opcodes:
643	.word	0xfe000000			@ mask
644	.word	0xf2000000			@ opcode
645
646	.word	0xff100000			@ mask
647	.word	0xf4000000			@ opcode
648
649	.word	0x00000000			@ mask
650	.word	0x00000000			@ opcode
651
652.LCneon_thumb_opcodes:
653	.word	0xef000000			@ mask
654	.word	0xef000000			@ opcode
655
656	.word	0xff100000			@ mask
657	.word	0xf9000000			@ opcode
658
659	.word	0x00000000			@ mask
660	.word	0x00000000			@ opcode
661#endif
662
663do_fpe:
664	enable_irq
665	ldr	r4, .LCfp
666	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
667	ldr	pc, [r4]			@ Call FP module USR entry point
668
669/*
670 * The FP module is called with these registers set:
671 *  r0  = instruction
672 *  r2  = PC+4
673 *  r9  = normal "successful" return address
674 *  r10 = FP workspace
675 *  lr  = unrecognised FP instruction return address
676 */
677
678	.pushsection .data
679ENTRY(fp_enter)
680	.word	no_fp
681	.popsection
682
683ENTRY(no_fp)
684	mov	pc, lr
685ENDPROC(no_fp)
686
687__und_usr_unknown:
688	enable_irq
689	mov	r0, sp
690	adr	lr, BSYM(ret_from_exception)
691	b	do_undefinstr
692ENDPROC(__und_usr_unknown)
693
694	.align	5
695__pabt_usr:
696	usr_entry
697
698	mov	r0, r2			@ pass address of aborted instruction.
699#ifdef MULTI_PABORT
700	ldr	r4, .LCprocfns
701	mov	lr, pc
702	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
703#else
704	bl	CPU_PABORT_HANDLER
705#endif
706	enable_irq				@ Enable interrupts
707	mov	r2, sp				@ regs
708	bl	do_PrefetchAbort		@ call abort handler
709 UNWIND(.fnend		)
710	/* fall through */
711/*
712 * This is the return code to user mode for abort handlers
713 */
714ENTRY(ret_from_exception)
715 UNWIND(.fnstart	)
716 UNWIND(.cantunwind	)
717	get_thread_info tsk
718	mov	why, #0
719	b	ret_to_user
720 UNWIND(.fnend		)
721ENDPROC(__pabt_usr)
722ENDPROC(ret_from_exception)
723
724/*
725 * Register switch for ARMv3 and ARMv4 processors
726 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
727 * previous and next are guaranteed not to be the same.
728 */
729ENTRY(__switch_to)
730 UNWIND(.fnstart	)
731 UNWIND(.cantunwind	)
732	add	ip, r1, #TI_CPU_SAVE
733	ldr	r3, [r2, #TI_TP_VALUE]
734 ARM(	stmia	ip!, {r4 - sl, fp, sp, lr} )	@ Store most regs on stack
735 THUMB(	stmia	ip!, {r4 - sl, fp}	   )	@ Store most regs on stack
736 THUMB(	str	sp, [ip], #4		   )
737 THUMB(	str	lr, [ip], #4		   )
738#ifdef CONFIG_MMU
739	ldr	r6, [r2, #TI_CPU_DOMAIN]
740#endif
741	set_tls	r3, r4, r5
742#if defined(CONFIG_CC_STACKPROTECTOR) && !defined(CONFIG_SMP)
743	ldr	r7, [r2, #TI_TASK]
744	ldr	r8, =__stack_chk_guard
745	ldr	r7, [r7, #TSK_STACK_CANARY]
746#endif
747#ifdef CONFIG_MMU
748	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
749#endif
750	mov	r5, r0
751	add	r4, r2, #TI_CPU_SAVE
752	ldr	r0, =thread_notify_head
753	mov	r1, #THREAD_NOTIFY_SWITCH
754	bl	atomic_notifier_call_chain
755#if defined(CONFIG_CC_STACKPROTECTOR) && !defined(CONFIG_SMP)
756	str	r7, [r8]
757#endif
758 THUMB(	mov	ip, r4			   )
759	mov	r0, r5
760 ARM(	ldmia	r4, {r4 - sl, fp, sp, pc}  )	@ Load all regs saved previously
761 THUMB(	ldmia	ip!, {r4 - sl, fp}	   )	@ Load all regs saved previously
762 THUMB(	ldr	sp, [ip], #4		   )
763 THUMB(	ldr	pc, [ip]		   )
764 UNWIND(.fnend		)
765ENDPROC(__switch_to)
766
767	__INIT
768
769/*
770 * User helpers.
771 *
772 * These are segment of kernel provided user code reachable from user space
773 * at a fixed address in kernel memory.  This is used to provide user space
774 * with some operations which require kernel help because of unimplemented
775 * native feature and/or instructions in many ARM CPUs. The idea is for
776 * this code to be executed directly in user mode for best efficiency but
777 * which is too intimate with the kernel counter part to be left to user
778 * libraries.  In fact this code might even differ from one CPU to another
779 * depending on the available  instruction set and restrictions like on
780 * SMP systems.  In other words, the kernel reserves the right to change
781 * this code as needed without warning. Only the entry points and their
782 * results are guaranteed to be stable.
783 *
784 * Each segment is 32-byte aligned and will be moved to the top of the high
785 * vector page.  New segments (if ever needed) must be added in front of
786 * existing ones.  This mechanism should be used only for things that are
787 * really small and justified, and not be abused freely.
788 *
789 * User space is expected to implement those things inline when optimizing
790 * for a processor that has the necessary native support, but only if such
791 * resulting binaries are already to be incompatible with earlier ARM
792 * processors due to the use of unsupported instructions other than what
793 * is provided here.  In other words don't make binaries unable to run on
794 * earlier processors just for the sake of not using these kernel helpers
795 * if your compiled code is not going to use the new instructions for other
796 * purpose.
797 */
798 THUMB(	.arm	)
799
800	.macro	usr_ret, reg
801#ifdef CONFIG_ARM_THUMB
802	bx	\reg
803#else
804	mov	pc, \reg
805#endif
806	.endm
807
808	.align	5
809	.globl	__kuser_helper_start
810__kuser_helper_start:
811
812/*
813 * Reference prototype:
814 *
815 *	void __kernel_memory_barrier(void)
816 *
817 * Input:
818 *
819 *	lr = return address
820 *
821 * Output:
822 *
823 *	none
824 *
825 * Clobbered:
826 *
827 *	none
828 *
829 * Definition and user space usage example:
830 *
831 *	typedef void (__kernel_dmb_t)(void);
832 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
833 *
834 * Apply any needed memory barrier to preserve consistency with data modified
835 * manually and __kuser_cmpxchg usage.
836 *
837 * This could be used as follows:
838 *
839 * #define __kernel_dmb() \
840 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
841 *	        : : : "r0", "lr","cc" )
842 */
843
844__kuser_memory_barrier:				@ 0xffff0fa0
845	smp_dmb
846	usr_ret	lr
847
848	.align	5
849
850/*
851 * Reference prototype:
852 *
853 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
854 *
855 * Input:
856 *
857 *	r0 = oldval
858 *	r1 = newval
859 *	r2 = ptr
860 *	lr = return address
861 *
862 * Output:
863 *
864 *	r0 = returned value (zero or non-zero)
865 *	C flag = set if r0 == 0, clear if r0 != 0
866 *
867 * Clobbered:
868 *
869 *	r3, ip, flags
870 *
871 * Definition and user space usage example:
872 *
873 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
874 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
875 *
876 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
877 * Return zero if *ptr was changed or non-zero if no exchange happened.
878 * The C flag is also set if *ptr was changed to allow for assembly
879 * optimization in the calling code.
880 *
881 * Notes:
882 *
883 *    - This routine already includes memory barriers as needed.
884 *
885 * For example, a user space atomic_add implementation could look like this:
886 *
887 * #define atomic_add(ptr, val) \
888 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
889 *	   register unsigned int __result asm("r1"); \
890 *	   asm volatile ( \
891 *	       "1: @ atomic_add\n\t" \
892 *	       "ldr	r0, [r2]\n\t" \
893 *	       "mov	r3, #0xffff0fff\n\t" \
894 *	       "add	lr, pc, #4\n\t" \
895 *	       "add	r1, r0, %2\n\t" \
896 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
897 *	       "bcc	1b" \
898 *	       : "=&r" (__result) \
899 *	       : "r" (__ptr), "rIL" (val) \
900 *	       : "r0","r3","ip","lr","cc","memory" ); \
901 *	   __result; })
902 */
903
904__kuser_cmpxchg:				@ 0xffff0fc0
905
906#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
907
908	/*
909	 * Poor you.  No fast solution possible...
910	 * The kernel itself must perform the operation.
911	 * A special ghost syscall is used for that (see traps.c).
912	 */
913	stmfd	sp!, {r7, lr}
914	ldr	r7, =1f			@ it's 20 bits
915	swi	__ARM_NR_cmpxchg
916	ldmfd	sp!, {r7, pc}
9171:	.word	__ARM_NR_cmpxchg
918
919#elif __LINUX_ARM_ARCH__ < 6
920
921#ifdef CONFIG_MMU
922
923	/*
924	 * The only thing that can break atomicity in this cmpxchg
925	 * implementation is either an IRQ or a data abort exception
926	 * causing another process/thread to be scheduled in the middle
927	 * of the critical sequence.  To prevent this, code is added to
928	 * the IRQ and data abort exception handlers to set the pc back
929	 * to the beginning of the critical section if it is found to be
930	 * within that critical section (see kuser_cmpxchg_fixup).
931	 */
9321:	ldr	r3, [r2]			@ load current val
933	subs	r3, r3, r0			@ compare with oldval
9342:	streq	r1, [r2]			@ store newval if eq
935	rsbs	r0, r3, #0			@ set return val and C flag
936	usr_ret	lr
937
938	.text
939kuser_cmpxchg_fixup:
940	@ Called from kuser_cmpxchg_check macro.
941	@ r2 = address of interrupted insn (must be preserved).
942	@ sp = saved regs. r7 and r8 are clobbered.
943	@ 1b = first critical insn, 2b = last critical insn.
944	@ If r2 >= 1b and r2 <= 2b then saved pc_usr is set to 1b.
945	mov	r7, #0xffff0fff
946	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
947	subs	r8, r2, r7
948	rsbcss	r8, r8, #(2b - 1b)
949	strcs	r7, [sp, #S_PC]
950	mov	pc, lr
951	.previous
952
953#else
954#warning "NPTL on non MMU needs fixing"
955	mov	r0, #-1
956	adds	r0, r0, #0
957	usr_ret	lr
958#endif
959
960#else
961
962	smp_dmb
9631:	ldrex	r3, [r2]
964	subs	r3, r3, r0
965	strexeq	r3, r1, [r2]
966	teqeq	r3, #1
967	beq	1b
968	rsbs	r0, r3, #0
969	/* beware -- each __kuser slot must be 8 instructions max */
970	ALT_SMP(b	__kuser_memory_barrier)
971	ALT_UP(usr_ret	lr)
972
973#endif
974
975	.align	5
976
977/*
978 * Reference prototype:
979 *
980 *	int __kernel_get_tls(void)
981 *
982 * Input:
983 *
984 *	lr = return address
985 *
986 * Output:
987 *
988 *	r0 = TLS value
989 *
990 * Clobbered:
991 *
992 *	none
993 *
994 * Definition and user space usage example:
995 *
996 *	typedef int (__kernel_get_tls_t)(void);
997 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
998 *
999 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
1000 *
1001 * This could be used as follows:
1002 *
1003 * #define __kernel_get_tls() \
1004 *	({ register unsigned int __val asm("r0"); \
1005 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
1006 *	        : "=r" (__val) : : "lr","cc" ); \
1007 *	   __val; })
1008 */
1009
1010__kuser_get_tls:				@ 0xffff0fe0
1011	ldr	r0, [pc, #(16 - 8)]	@ read TLS, set in kuser_get_tls_init
1012	usr_ret	lr
1013	mrc	p15, 0, r0, c13, c0, 3	@ 0xffff0fe8 hardware TLS code
1014	.rep	4
1015	.word	0			@ 0xffff0ff0 software TLS value, then
1016	.endr				@ pad up to __kuser_helper_version
1017
1018/*
1019 * Reference declaration:
1020 *
1021 *	extern unsigned int __kernel_helper_version;
1022 *
1023 * Definition and user space usage example:
1024 *
1025 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
1026 *
1027 * User space may read this to determine the curent number of helpers
1028 * available.
1029 */
1030
1031__kuser_helper_version:				@ 0xffff0ffc
1032	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
1033
1034	.globl	__kuser_helper_end
1035__kuser_helper_end:
1036
1037 THUMB(	.thumb	)
1038
1039/*
1040 * Vector stubs.
1041 *
1042 * This code is copied to 0xffff0200 so we can use branches in the
1043 * vectors, rather than ldr's.  Note that this code must not
1044 * exceed 0x300 bytes.
1045 *
1046 * Common stub entry macro:
1047 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1048 *
1049 * SP points to a minimal amount of processor-private memory, the address
1050 * of which is copied into r0 for the mode specific abort handler.
1051 */
1052	.macro	vector_stub, name, mode, correction=0
1053	.align	5
1054
1055vector_\name:
1056	.if \correction
1057	sub	lr, lr, #\correction
1058	.endif
1059
1060	@
1061	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
1062	@ (parent CPSR)
1063	@
1064	stmia	sp, {r0, lr}		@ save r0, lr
1065	mrs	lr, spsr
1066	str	lr, [sp, #8]		@ save spsr
1067
1068	@
1069	@ Prepare for SVC32 mode.  IRQs remain disabled.
1070	@
1071	mrs	r0, cpsr
1072	eor	r0, r0, #(\mode ^ SVC_MODE | PSR_ISETSTATE)
1073	msr	spsr_cxsf, r0
1074
1075	@
1076	@ the branch table must immediately follow this code
1077	@
1078	and	lr, lr, #0x0f
1079 THUMB(	adr	r0, 1f			)
1080 THUMB(	ldr	lr, [r0, lr, lsl #2]	)
1081	mov	r0, sp
1082 ARM(	ldr	lr, [pc, lr, lsl #2]	)
1083	movs	pc, lr			@ branch to handler in SVC mode
1084ENDPROC(vector_\name)
1085
1086	.align	2
1087	@ handler addresses follow this label
10881:
1089	.endm
1090
1091	.globl	__stubs_start
1092__stubs_start:
1093/*
1094 * Interrupt dispatcher
1095 */
1096	vector_stub	irq, IRQ_MODE, 4
1097
1098	.long	__irq_usr			@  0  (USR_26 / USR_32)
1099	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
1100	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
1101	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
1102	.long	__irq_invalid			@  4
1103	.long	__irq_invalid			@  5
1104	.long	__irq_invalid			@  6
1105	.long	__irq_invalid			@  7
1106	.long	__irq_invalid			@  8
1107	.long	__irq_invalid			@  9
1108	.long	__irq_invalid			@  a
1109	.long	__irq_invalid			@  b
1110	.long	__irq_invalid			@  c
1111	.long	__irq_invalid			@  d
1112	.long	__irq_invalid			@  e
1113	.long	__irq_invalid			@  f
1114
1115/*
1116 * Data abort dispatcher
1117 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1118 */
1119	vector_stub	dabt, ABT_MODE, 8
1120
1121	.long	__dabt_usr			@  0  (USR_26 / USR_32)
1122	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
1123	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
1124	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
1125	.long	__dabt_invalid			@  4
1126	.long	__dabt_invalid			@  5
1127	.long	__dabt_invalid			@  6
1128	.long	__dabt_invalid			@  7
1129	.long	__dabt_invalid			@  8
1130	.long	__dabt_invalid			@  9
1131	.long	__dabt_invalid			@  a
1132	.long	__dabt_invalid			@  b
1133	.long	__dabt_invalid			@  c
1134	.long	__dabt_invalid			@  d
1135	.long	__dabt_invalid			@  e
1136	.long	__dabt_invalid			@  f
1137
1138/*
1139 * Prefetch abort dispatcher
1140 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1141 */
1142	vector_stub	pabt, ABT_MODE, 4
1143
1144	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1145	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1146	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1147	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1148	.long	__pabt_invalid			@  4
1149	.long	__pabt_invalid			@  5
1150	.long	__pabt_invalid			@  6
1151	.long	__pabt_invalid			@  7
1152	.long	__pabt_invalid			@  8
1153	.long	__pabt_invalid			@  9
1154	.long	__pabt_invalid			@  a
1155	.long	__pabt_invalid			@  b
1156	.long	__pabt_invalid			@  c
1157	.long	__pabt_invalid			@  d
1158	.long	__pabt_invalid			@  e
1159	.long	__pabt_invalid			@  f
1160
1161/*
1162 * Undef instr entry dispatcher
1163 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1164 */
1165	vector_stub	und, UND_MODE
1166
1167	.long	__und_usr			@  0 (USR_26 / USR_32)
1168	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1169	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1170	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1171	.long	__und_invalid			@  4
1172	.long	__und_invalid			@  5
1173	.long	__und_invalid			@  6
1174	.long	__und_invalid			@  7
1175	.long	__und_invalid			@  8
1176	.long	__und_invalid			@  9
1177	.long	__und_invalid			@  a
1178	.long	__und_invalid			@  b
1179	.long	__und_invalid			@  c
1180	.long	__und_invalid			@  d
1181	.long	__und_invalid			@  e
1182	.long	__und_invalid			@  f
1183
1184	.align	5
1185
1186/*=============================================================================
1187 * Undefined FIQs
1188 *-----------------------------------------------------------------------------
1189 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1190 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1191 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1192 * damage alert!  I don't think that we can execute any code in here in any
1193 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1194 * get out of that mode without clobbering one register.
1195 */
1196vector_fiq:
1197	disable_fiq
1198	subs	pc, lr, #4
1199
1200/*=============================================================================
1201 * Address exception handler
1202 *-----------------------------------------------------------------------------
1203 * These aren't too critical.
1204 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1205 */
1206
1207vector_addrexcptn:
1208	b	vector_addrexcptn
1209
1210/*
1211 * We group all the following data together to optimise
1212 * for CPUs with separate I & D caches.
1213 */
1214	.align	5
1215
1216.LCvswi:
1217	.word	vector_swi
1218
1219	.globl	__stubs_end
1220__stubs_end:
1221
1222	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1223
1224	.globl	__vectors_start
1225__vectors_start:
1226 ARM(	swi	SYS_ERROR0	)
1227 THUMB(	svc	#0		)
1228 THUMB(	nop			)
1229	W(b)	vector_und + stubs_offset
1230	W(ldr)	pc, .LCvswi + stubs_offset
1231	W(b)	vector_pabt + stubs_offset
1232	W(b)	vector_dabt + stubs_offset
1233	W(b)	vector_addrexcptn + stubs_offset
1234	W(b)	vector_irq + stubs_offset
1235	W(b)	vector_fiq + stubs_offset
1236
1237	.globl	__vectors_end
1238__vectors_end:
1239
1240	.data
1241
1242	.globl	cr_alignment
1243	.globl	cr_no_alignment
1244cr_alignment:
1245	.space	4
1246cr_no_alignment:
1247	.space	4
1248