xref: /openbmc/linux/arch/arm/kernel/entry-armv.S (revision 643d1f7f)
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 <asm/arch/entry-macro.S>
22#include <asm/thread_notify.h>
23
24#include "entry-header.S"
25
26/*
27 * Interrupt handling.  Preserves r7, r8, r9
28 */
29	.macro	irq_handler
30	get_irqnr_preamble r5, lr
311:	get_irqnr_and_base r0, r6, r5, lr
32	movne	r1, sp
33	@
34	@ routine called with r0 = irq number, r1 = struct pt_regs *
35	@
36	adrne	lr, 1b
37	bne	asm_do_IRQ
38
39#ifdef CONFIG_SMP
40	/*
41	 * XXX
42	 *
43	 * this macro assumes that irqstat (r6) and base (r5) are
44	 * preserved from get_irqnr_and_base above
45	 */
46	test_for_ipi r0, r6, r5, lr
47	movne	r0, sp
48	adrne	lr, 1b
49	bne	do_IPI
50
51#ifdef CONFIG_LOCAL_TIMERS
52	test_for_ltirq r0, r6, r5, lr
53	movne	r0, sp
54	adrne	lr, 1b
55	bne	do_local_timer
56#endif
57#endif
58
59	.endm
60
61#ifdef CONFIG_KPROBES
62	.section	.kprobes.text,"ax",%progbits
63#else
64	.text
65#endif
66
67/*
68 * Invalid mode handlers
69 */
70	.macro	inv_entry, reason
71	sub	sp, sp, #S_FRAME_SIZE
72	stmib	sp, {r1 - lr}
73	mov	r1, #\reason
74	.endm
75
76__pabt_invalid:
77	inv_entry BAD_PREFETCH
78	b	common_invalid
79
80__dabt_invalid:
81	inv_entry BAD_DATA
82	b	common_invalid
83
84__irq_invalid:
85	inv_entry BAD_IRQ
86	b	common_invalid
87
88__und_invalid:
89	inv_entry BAD_UNDEFINSTR
90
91	@
92	@ XXX fall through to common_invalid
93	@
94
95@
96@ common_invalid - generic code for failed exception (re-entrant version of handlers)
97@
98common_invalid:
99	zero_fp
100
101	ldmia	r0, {r4 - r6}
102	add	r0, sp, #S_PC		@ here for interlock avoidance
103	mov	r7, #-1			@  ""   ""    ""        ""
104	str	r4, [sp]		@ save preserved r0
105	stmia	r0, {r5 - r7}		@ lr_<exception>,
106					@ cpsr_<exception>, "old_r0"
107
108	mov	r0, sp
109	b	bad_mode
110
111/*
112 * SVC mode handlers
113 */
114
115#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
116#define SPFIX(code...) code
117#else
118#define SPFIX(code...)
119#endif
120
121	.macro	svc_entry, stack_hole=0
122	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole)
123 SPFIX(	tst	sp, #4		)
124 SPFIX(	bicne	sp, sp, #4	)
125	stmib	sp, {r1 - r12}
126
127	ldmia	r0, {r1 - r3}
128	add	r5, sp, #S_SP		@ here for interlock avoidance
129	mov	r4, #-1			@  ""  ""      ""       ""
130	add	r0, sp, #(S_FRAME_SIZE + \stack_hole)
131 SPFIX(	addne	r0, r0, #4	)
132	str	r1, [sp]		@ save the "real" r0 copied
133					@ from the exception stack
134
135	mov	r1, lr
136
137	@
138	@ We are now ready to fill in the remaining blanks on the stack:
139	@
140	@  r0 - sp_svc
141	@  r1 - lr_svc
142	@  r2 - lr_<exception>, already fixed up for correct return/restart
143	@  r3 - spsr_<exception>
144	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
145	@
146	stmia	r5, {r0 - r4}
147	.endm
148
149	.align	5
150__dabt_svc:
151	svc_entry
152
153	@
154	@ get ready to re-enable interrupts if appropriate
155	@
156	mrs	r9, cpsr
157	tst	r3, #PSR_I_BIT
158	biceq	r9, r9, #PSR_I_BIT
159
160	@
161	@ Call the processor-specific abort handler:
162	@
163	@  r2 - aborted context pc
164	@  r3 - aborted context cpsr
165	@
166	@ The abort handler must return the aborted address in r0, and
167	@ the fault status register in r1.  r9 must be preserved.
168	@
169#ifdef MULTI_ABORT
170	ldr	r4, .LCprocfns
171	mov	lr, pc
172	ldr	pc, [r4]
173#else
174	bl	CPU_ABORT_HANDLER
175#endif
176
177	@
178	@ set desired IRQ state, then call main handler
179	@
180	msr	cpsr_c, r9
181	mov	r2, sp
182	bl	do_DataAbort
183
184	@
185	@ IRQs off again before pulling preserved data off the stack
186	@
187	disable_irq
188
189	@
190	@ restore SPSR and restart the instruction
191	@
192	ldr	r0, [sp, #S_PSR]
193	msr	spsr_cxsf, r0
194	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
195
196	.align	5
197__irq_svc:
198	svc_entry
199
200#ifdef CONFIG_TRACE_IRQFLAGS
201	bl	trace_hardirqs_off
202#endif
203#ifdef CONFIG_PREEMPT
204	get_thread_info tsk
205	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
206	add	r7, r8, #1			@ increment it
207	str	r7, [tsk, #TI_PREEMPT]
208#endif
209
210	irq_handler
211#ifdef CONFIG_PREEMPT
212	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
213	tst	r0, #_TIF_NEED_RESCHED
214	blne	svc_preempt
215preempt_return:
216	ldr	r0, [tsk, #TI_PREEMPT]		@ read preempt value
217	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
218	teq	r0, r7
219	strne	r0, [r0, -r0]			@ bug()
220#endif
221	ldr	r0, [sp, #S_PSR]		@ irqs are already disabled
222	msr	spsr_cxsf, r0
223#ifdef CONFIG_TRACE_IRQFLAGS
224	tst	r0, #PSR_I_BIT
225	bleq	trace_hardirqs_on
226#endif
227	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
228
229	.ltorg
230
231#ifdef CONFIG_PREEMPT
232svc_preempt:
233	teq	r8, #0				@ was preempt count = 0
234	ldreq	r6, .LCirq_stat
235	movne	pc, lr				@ no
236	ldr	r0, [r6, #4]			@ local_irq_count
237	ldr	r1, [r6, #8]			@ local_bh_count
238	adds	r0, r0, r1
239	movne	pc, lr
240	mov	r7, #0				@ preempt_schedule_irq
241	str	r7, [tsk, #TI_PREEMPT]		@ expects preempt_count == 0
2421:	bl	preempt_schedule_irq		@ irq en/disable is done inside
243	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
244	tst	r0, #_TIF_NEED_RESCHED
245	beq	preempt_return			@ go again
246	b	1b
247#endif
248
249	.align	5
250__und_svc:
251#ifdef CONFIG_KPROBES
252	@ If a kprobe is about to simulate a "stmdb sp..." instruction,
253	@ it obviously needs free stack space which then will belong to
254	@ the saved context.
255	svc_entry 64
256#else
257	svc_entry
258#endif
259
260	@
261	@ call emulation code, which returns using r9 if it has emulated
262	@ the instruction, or the more conventional lr if we are to treat
263	@ this as a real undefined instruction
264	@
265	@  r0 - instruction
266	@
267	ldr	r0, [r2, #-4]
268	adr	r9, 1f
269	bl	call_fpe
270
271	mov	r0, sp				@ struct pt_regs *regs
272	bl	do_undefinstr
273
274	@
275	@ IRQs off again before pulling preserved data off the stack
276	@
2771:	disable_irq
278
279	@
280	@ restore SPSR and restart the instruction
281	@
282	ldr	lr, [sp, #S_PSR]		@ Get SVC cpsr
283	msr	spsr_cxsf, lr
284	ldmia	sp, {r0 - pc}^			@ Restore SVC registers
285
286	.align	5
287__pabt_svc:
288	svc_entry
289
290	@
291	@ re-enable interrupts if appropriate
292	@
293	mrs	r9, cpsr
294	tst	r3, #PSR_I_BIT
295	biceq	r9, r9, #PSR_I_BIT
296	msr	cpsr_c, r9
297
298	@
299	@ set args, then call main handler
300	@
301	@  r0 - address of faulting instruction
302	@  r1 - pointer to registers on stack
303	@
304	mov	r0, r2				@ address (pc)
305	mov	r1, sp				@ regs
306	bl	do_PrefetchAbort		@ call abort handler
307
308	@
309	@ IRQs off again before pulling preserved data off the stack
310	@
311	disable_irq
312
313	@
314	@ restore SPSR and restart the instruction
315	@
316	ldr	r0, [sp, #S_PSR]
317	msr	spsr_cxsf, r0
318	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
319
320	.align	5
321.LCcralign:
322	.word	cr_alignment
323#ifdef MULTI_ABORT
324.LCprocfns:
325	.word	processor
326#endif
327.LCfp:
328	.word	fp_enter
329#ifdef CONFIG_PREEMPT
330.LCirq_stat:
331	.word	irq_stat
332#endif
333
334/*
335 * User mode handlers
336 *
337 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
338 */
339
340#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
341#error "sizeof(struct pt_regs) must be a multiple of 8"
342#endif
343
344	.macro	usr_entry
345	sub	sp, sp, #S_FRAME_SIZE
346	stmib	sp, {r1 - r12}
347
348	ldmia	r0, {r1 - r3}
349	add	r0, sp, #S_PC		@ here for interlock avoidance
350	mov	r4, #-1			@  ""  ""     ""        ""
351
352	str	r1, [sp]		@ save the "real" r0 copied
353					@ from the exception stack
354
355	@
356	@ We are now ready to fill in the remaining blanks on the stack:
357	@
358	@  r2 - lr_<exception>, already fixed up for correct return/restart
359	@  r3 - spsr_<exception>
360	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
361	@
362	@ Also, separately save sp_usr and lr_usr
363	@
364	stmia	r0, {r2 - r4}
365	stmdb	r0, {sp, lr}^
366
367	@
368	@ Enable the alignment trap while in kernel mode
369	@
370	alignment_trap r0
371
372	@
373	@ Clear FP to mark the first stack frame
374	@
375	zero_fp
376	.endm
377
378	.macro	kuser_cmpxchg_check
379#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
380#ifndef CONFIG_MMU
381#warning "NPTL on non MMU needs fixing"
382#else
383	@ Make sure our user space atomic helper is restarted
384	@ if it was interrupted in a critical region.  Here we
385	@ perform a quick test inline since it should be false
386	@ 99.9999% of the time.  The rest is done out of line.
387	cmp	r2, #TASK_SIZE
388	blhs	kuser_cmpxchg_fixup
389#endif
390#endif
391	.endm
392
393	.align	5
394__dabt_usr:
395	usr_entry
396	kuser_cmpxchg_check
397
398	@
399	@ Call the processor-specific abort handler:
400	@
401	@  r2 - aborted context pc
402	@  r3 - aborted context cpsr
403	@
404	@ The abort handler must return the aborted address in r0, and
405	@ the fault status register in r1.
406	@
407#ifdef MULTI_ABORT
408	ldr	r4, .LCprocfns
409	mov	lr, pc
410	ldr	pc, [r4]
411#else
412	bl	CPU_ABORT_HANDLER
413#endif
414
415	@
416	@ IRQs on, then call the main handler
417	@
418	enable_irq
419	mov	r2, sp
420	adr	lr, ret_from_exception
421	b	do_DataAbort
422
423	.align	5
424__irq_usr:
425	usr_entry
426	kuser_cmpxchg_check
427
428#ifdef CONFIG_TRACE_IRQFLAGS
429	bl	trace_hardirqs_off
430#endif
431	get_thread_info tsk
432#ifdef CONFIG_PREEMPT
433	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
434	add	r7, r8, #1			@ increment it
435	str	r7, [tsk, #TI_PREEMPT]
436#endif
437
438	irq_handler
439#ifdef CONFIG_PREEMPT
440	ldr	r0, [tsk, #TI_PREEMPT]
441	str	r8, [tsk, #TI_PREEMPT]
442	teq	r0, r7
443	strne	r0, [r0, -r0]
444#endif
445#ifdef CONFIG_TRACE_IRQFLAGS
446	bl	trace_hardirqs_on
447#endif
448
449	mov	why, #0
450	b	ret_to_user
451
452	.ltorg
453
454	.align	5
455__und_usr:
456	usr_entry
457
458	tst	r3, #PSR_T_BIT			@ Thumb mode?
459	bne	__und_usr_unknown		@ ignore FP
460	sub	r4, r2, #4
461
462	@
463	@ fall through to the emulation code, which returns using r9 if
464	@ it has emulated the instruction, or the more conventional lr
465	@ if we are to treat this as a real undefined instruction
466	@
467	@  r0 - instruction
468	@
469	adr	r9, ret_from_exception
470	adr	lr, __und_usr_unknown
4711:	ldrt	r0, [r4]
472	@
473	@ fallthrough to call_fpe
474	@
475
476/*
477 * The out of line fixup for the ldrt above.
478 */
479	.section .fixup, "ax"
4802:	mov	pc, r9
481	.previous
482	.section __ex_table,"a"
483	.long	1b, 2b
484	.previous
485
486/*
487 * Check whether the instruction is a co-processor instruction.
488 * If yes, we need to call the relevant co-processor handler.
489 *
490 * Note that we don't do a full check here for the co-processor
491 * instructions; all instructions with bit 27 set are well
492 * defined.  The only instructions that should fault are the
493 * co-processor instructions.  However, we have to watch out
494 * for the ARM6/ARM7 SWI bug.
495 *
496 * NEON is a special case that has to be handled here. Not all
497 * NEON instructions are co-processor instructions, so we have
498 * to make a special case of checking for them. Plus, there's
499 * five groups of them, so we have a table of mask/opcode pairs
500 * to check against, and if any match then we branch off into the
501 * NEON handler code.
502 *
503 * Emulators may wish to make use of the following registers:
504 *  r0  = instruction opcode.
505 *  r2  = PC+4
506 *  r9  = normal "successful" return address
507 *  r10 = this threads thread_info structure.
508 *  lr  = unrecognised instruction return address
509 */
510call_fpe:
511#ifdef CONFIG_NEON
512	adr	r6, .LCneon_opcodes
5132:
514	ldr	r7, [r6], #4			@ mask value
515	cmp	r7, #0				@ end mask?
516	beq	1f
517	and	r8, r0, r7
518	ldr	r7, [r6], #4			@ opcode bits matching in mask
519	cmp	r8, r7				@ NEON instruction?
520	bne	2b
521	get_thread_info r10
522	mov	r7, #1
523	strb	r7, [r10, #TI_USED_CP + 10]	@ mark CP#10 as used
524	strb	r7, [r10, #TI_USED_CP + 11]	@ mark CP#11 as used
525	b	do_vfp				@ let VFP handler handle this
5261:
527#endif
528	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
529#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
530	and	r8, r0, #0x0f000000		@ mask out op-code bits
531	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
532#endif
533	moveq	pc, lr
534	get_thread_info r10			@ get current thread
535	and	r8, r0, #0x00000f00		@ mask out CP number
536	mov	r7, #1
537	add	r6, r10, #TI_USED_CP
538	strb	r7, [r6, r8, lsr #8]		@ set appropriate used_cp[]
539#ifdef CONFIG_IWMMXT
540	@ Test if we need to give access to iWMMXt coprocessors
541	ldr	r5, [r10, #TI_FLAGS]
542	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
543	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
544	bcs	iwmmxt_task_enable
545#endif
546	add	pc, pc, r8, lsr #6
547	mov	r0, r0
548
549	mov	pc, lr				@ CP#0
550	b	do_fpe				@ CP#1 (FPE)
551	b	do_fpe				@ CP#2 (FPE)
552	mov	pc, lr				@ CP#3
553#ifdef CONFIG_CRUNCH
554	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
555	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
556	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
557#else
558	mov	pc, lr				@ CP#4
559	mov	pc, lr				@ CP#5
560	mov	pc, lr				@ CP#6
561#endif
562	mov	pc, lr				@ CP#7
563	mov	pc, lr				@ CP#8
564	mov	pc, lr				@ CP#9
565#ifdef CONFIG_VFP
566	b	do_vfp				@ CP#10 (VFP)
567	b	do_vfp				@ CP#11 (VFP)
568#else
569	mov	pc, lr				@ CP#10 (VFP)
570	mov	pc, lr				@ CP#11 (VFP)
571#endif
572	mov	pc, lr				@ CP#12
573	mov	pc, lr				@ CP#13
574	mov	pc, lr				@ CP#14 (Debug)
575	mov	pc, lr				@ CP#15 (Control)
576
577#ifdef CONFIG_NEON
578	.align	6
579
580.LCneon_opcodes:
581	.word	0xfe000000			@ mask
582	.word	0xf2000000			@ opcode
583
584	.word	0xff100000			@ mask
585	.word	0xf4000000			@ opcode
586
587	.word	0x00000000			@ mask
588	.word	0x00000000			@ opcode
589#endif
590
591do_fpe:
592	enable_irq
593	ldr	r4, .LCfp
594	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
595	ldr	pc, [r4]			@ Call FP module USR entry point
596
597/*
598 * The FP module is called with these registers set:
599 *  r0  = instruction
600 *  r2  = PC+4
601 *  r9  = normal "successful" return address
602 *  r10 = FP workspace
603 *  lr  = unrecognised FP instruction return address
604 */
605
606	.data
607ENTRY(fp_enter)
608	.word	no_fp
609	.previous
610
611no_fp:	mov	pc, lr
612
613__und_usr_unknown:
614	mov	r0, sp
615	adr	lr, ret_from_exception
616	b	do_undefinstr
617
618	.align	5
619__pabt_usr:
620	usr_entry
621
622	enable_irq				@ Enable interrupts
623	mov	r0, r2				@ address (pc)
624	mov	r1, sp				@ regs
625	bl	do_PrefetchAbort		@ call abort handler
626	/* fall through */
627/*
628 * This is the return code to user mode for abort handlers
629 */
630ENTRY(ret_from_exception)
631	get_thread_info tsk
632	mov	why, #0
633	b	ret_to_user
634
635/*
636 * Register switch for ARMv3 and ARMv4 processors
637 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
638 * previous and next are guaranteed not to be the same.
639 */
640ENTRY(__switch_to)
641	add	ip, r1, #TI_CPU_SAVE
642	ldr	r3, [r2, #TI_TP_VALUE]
643	stmia	ip!, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
644#ifdef CONFIG_MMU
645	ldr	r6, [r2, #TI_CPU_DOMAIN]
646#endif
647#if __LINUX_ARM_ARCH__ >= 6
648#ifdef CONFIG_CPU_32v6K
649	clrex
650#else
651	strex	r5, r4, [ip]			@ Clear exclusive monitor
652#endif
653#endif
654#if defined(CONFIG_HAS_TLS_REG)
655	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
656#elif !defined(CONFIG_TLS_REG_EMUL)
657	mov	r4, #0xffff0fff
658	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
659#endif
660#ifdef CONFIG_MMU
661	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
662#endif
663	mov	r5, r0
664	add	r4, r2, #TI_CPU_SAVE
665	ldr	r0, =thread_notify_head
666	mov	r1, #THREAD_NOTIFY_SWITCH
667	bl	atomic_notifier_call_chain
668	mov	r0, r5
669	ldmia	r4, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
670
671	__INIT
672
673/*
674 * User helpers.
675 *
676 * These are segment of kernel provided user code reachable from user space
677 * at a fixed address in kernel memory.  This is used to provide user space
678 * with some operations which require kernel help because of unimplemented
679 * native feature and/or instructions in many ARM CPUs. The idea is for
680 * this code to be executed directly in user mode for best efficiency but
681 * which is too intimate with the kernel counter part to be left to user
682 * libraries.  In fact this code might even differ from one CPU to another
683 * depending on the available  instruction set and restrictions like on
684 * SMP systems.  In other words, the kernel reserves the right to change
685 * this code as needed without warning. Only the entry points and their
686 * results are guaranteed to be stable.
687 *
688 * Each segment is 32-byte aligned and will be moved to the top of the high
689 * vector page.  New segments (if ever needed) must be added in front of
690 * existing ones.  This mechanism should be used only for things that are
691 * really small and justified, and not be abused freely.
692 *
693 * User space is expected to implement those things inline when optimizing
694 * for a processor that has the necessary native support, but only if such
695 * resulting binaries are already to be incompatible with earlier ARM
696 * processors due to the use of unsupported instructions other than what
697 * is provided here.  In other words don't make binaries unable to run on
698 * earlier processors just for the sake of not using these kernel helpers
699 * if your compiled code is not going to use the new instructions for other
700 * purpose.
701 */
702
703	.macro	usr_ret, reg
704#ifdef CONFIG_ARM_THUMB
705	bx	\reg
706#else
707	mov	pc, \reg
708#endif
709	.endm
710
711	.align	5
712	.globl	__kuser_helper_start
713__kuser_helper_start:
714
715/*
716 * Reference prototype:
717 *
718 *	void __kernel_memory_barrier(void)
719 *
720 * Input:
721 *
722 *	lr = return address
723 *
724 * Output:
725 *
726 *	none
727 *
728 * Clobbered:
729 *
730 *	none
731 *
732 * Definition and user space usage example:
733 *
734 *	typedef void (__kernel_dmb_t)(void);
735 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
736 *
737 * Apply any needed memory barrier to preserve consistency with data modified
738 * manually and __kuser_cmpxchg usage.
739 *
740 * This could be used as follows:
741 *
742 * #define __kernel_dmb() \
743 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
744 *	        : : : "r0", "lr","cc" )
745 */
746
747__kuser_memory_barrier:				@ 0xffff0fa0
748
749#if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
750	mcr	p15, 0, r0, c7, c10, 5	@ dmb
751#endif
752	usr_ret	lr
753
754	.align	5
755
756/*
757 * Reference prototype:
758 *
759 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
760 *
761 * Input:
762 *
763 *	r0 = oldval
764 *	r1 = newval
765 *	r2 = ptr
766 *	lr = return address
767 *
768 * Output:
769 *
770 *	r0 = returned value (zero or non-zero)
771 *	C flag = set if r0 == 0, clear if r0 != 0
772 *
773 * Clobbered:
774 *
775 *	r3, ip, flags
776 *
777 * Definition and user space usage example:
778 *
779 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
780 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
781 *
782 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
783 * Return zero if *ptr was changed or non-zero if no exchange happened.
784 * The C flag is also set if *ptr was changed to allow for assembly
785 * optimization in the calling code.
786 *
787 * Notes:
788 *
789 *    - This routine already includes memory barriers as needed.
790 *
791 * For example, a user space atomic_add implementation could look like this:
792 *
793 * #define atomic_add(ptr, val) \
794 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
795 *	   register unsigned int __result asm("r1"); \
796 *	   asm volatile ( \
797 *	       "1: @ atomic_add\n\t" \
798 *	       "ldr	r0, [r2]\n\t" \
799 *	       "mov	r3, #0xffff0fff\n\t" \
800 *	       "add	lr, pc, #4\n\t" \
801 *	       "add	r1, r0, %2\n\t" \
802 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
803 *	       "bcc	1b" \
804 *	       : "=&r" (__result) \
805 *	       : "r" (__ptr), "rIL" (val) \
806 *	       : "r0","r3","ip","lr","cc","memory" ); \
807 *	   __result; })
808 */
809
810__kuser_cmpxchg:				@ 0xffff0fc0
811
812#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
813
814	/*
815	 * Poor you.  No fast solution possible...
816	 * The kernel itself must perform the operation.
817	 * A special ghost syscall is used for that (see traps.c).
818	 */
819	stmfd	sp!, {r7, lr}
820	mov	r7, #0xff00		@ 0xfff0 into r7 for EABI
821	orr	r7, r7, #0xf0
822	swi	#0x9ffff0
823	ldmfd	sp!, {r7, pc}
824
825#elif __LINUX_ARM_ARCH__ < 6
826
827#ifdef CONFIG_MMU
828
829	/*
830	 * The only thing that can break atomicity in this cmpxchg
831	 * implementation is either an IRQ or a data abort exception
832	 * causing another process/thread to be scheduled in the middle
833	 * of the critical sequence.  To prevent this, code is added to
834	 * the IRQ and data abort exception handlers to set the pc back
835	 * to the beginning of the critical section if it is found to be
836	 * within that critical section (see kuser_cmpxchg_fixup).
837	 */
8381:	ldr	r3, [r2]			@ load current val
839	subs	r3, r3, r0			@ compare with oldval
8402:	streq	r1, [r2]			@ store newval if eq
841	rsbs	r0, r3, #0			@ set return val and C flag
842	usr_ret	lr
843
844	.text
845kuser_cmpxchg_fixup:
846	@ Called from kuser_cmpxchg_check macro.
847	@ r2 = address of interrupted insn (must be preserved).
848	@ sp = saved regs. r7 and r8 are clobbered.
849	@ 1b = first critical insn, 2b = last critical insn.
850	@ If r2 >= 1b and r2 <= 2b then saved pc_usr is set to 1b.
851	mov	r7, #0xffff0fff
852	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
853	subs	r8, r2, r7
854	rsbcss	r8, r8, #(2b - 1b)
855	strcs	r7, [sp, #S_PC]
856	mov	pc, lr
857	.previous
858
859#else
860#warning "NPTL on non MMU needs fixing"
861	mov	r0, #-1
862	adds	r0, r0, #0
863	usr_ret	lr
864#endif
865
866#else
867
868#ifdef CONFIG_SMP
869	mcr	p15, 0, r0, c7, c10, 5	@ dmb
870#endif
8711:	ldrex	r3, [r2]
872	subs	r3, r3, r0
873	strexeq	r3, r1, [r2]
874	teqeq	r3, #1
875	beq	1b
876	rsbs	r0, r3, #0
877	/* beware -- each __kuser slot must be 8 instructions max */
878#ifdef CONFIG_SMP
879	b	__kuser_memory_barrier
880#else
881	usr_ret	lr
882#endif
883
884#endif
885
886	.align	5
887
888/*
889 * Reference prototype:
890 *
891 *	int __kernel_get_tls(void)
892 *
893 * Input:
894 *
895 *	lr = return address
896 *
897 * Output:
898 *
899 *	r0 = TLS value
900 *
901 * Clobbered:
902 *
903 *	none
904 *
905 * Definition and user space usage example:
906 *
907 *	typedef int (__kernel_get_tls_t)(void);
908 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
909 *
910 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
911 *
912 * This could be used as follows:
913 *
914 * #define __kernel_get_tls() \
915 *	({ register unsigned int __val asm("r0"); \
916 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
917 *	        : "=r" (__val) : : "lr","cc" ); \
918 *	   __val; })
919 */
920
921__kuser_get_tls:				@ 0xffff0fe0
922
923#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
924	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
925#else
926	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
927#endif
928	usr_ret	lr
929
930	.rep	5
931	.word	0			@ pad up to __kuser_helper_version
932	.endr
933
934/*
935 * Reference declaration:
936 *
937 *	extern unsigned int __kernel_helper_version;
938 *
939 * Definition and user space usage example:
940 *
941 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
942 *
943 * User space may read this to determine the curent number of helpers
944 * available.
945 */
946
947__kuser_helper_version:				@ 0xffff0ffc
948	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
949
950	.globl	__kuser_helper_end
951__kuser_helper_end:
952
953
954/*
955 * Vector stubs.
956 *
957 * This code is copied to 0xffff0200 so we can use branches in the
958 * vectors, rather than ldr's.  Note that this code must not
959 * exceed 0x300 bytes.
960 *
961 * Common stub entry macro:
962 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
963 *
964 * SP points to a minimal amount of processor-private memory, the address
965 * of which is copied into r0 for the mode specific abort handler.
966 */
967	.macro	vector_stub, name, mode, correction=0
968	.align	5
969
970vector_\name:
971	.if \correction
972	sub	lr, lr, #\correction
973	.endif
974
975	@
976	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
977	@ (parent CPSR)
978	@
979	stmia	sp, {r0, lr}		@ save r0, lr
980	mrs	lr, spsr
981	str	lr, [sp, #8]		@ save spsr
982
983	@
984	@ Prepare for SVC32 mode.  IRQs remain disabled.
985	@
986	mrs	r0, cpsr
987	eor	r0, r0, #(\mode ^ SVC_MODE)
988	msr	spsr_cxsf, r0
989
990	@
991	@ the branch table must immediately follow this code
992	@
993	and	lr, lr, #0x0f
994	mov	r0, sp
995	ldr	lr, [pc, lr, lsl #2]
996	movs	pc, lr			@ branch to handler in SVC mode
997	.endm
998
999	.globl	__stubs_start
1000__stubs_start:
1001/*
1002 * Interrupt dispatcher
1003 */
1004	vector_stub	irq, IRQ_MODE, 4
1005
1006	.long	__irq_usr			@  0  (USR_26 / USR_32)
1007	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
1008	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
1009	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
1010	.long	__irq_invalid			@  4
1011	.long	__irq_invalid			@  5
1012	.long	__irq_invalid			@  6
1013	.long	__irq_invalid			@  7
1014	.long	__irq_invalid			@  8
1015	.long	__irq_invalid			@  9
1016	.long	__irq_invalid			@  a
1017	.long	__irq_invalid			@  b
1018	.long	__irq_invalid			@  c
1019	.long	__irq_invalid			@  d
1020	.long	__irq_invalid			@  e
1021	.long	__irq_invalid			@  f
1022
1023/*
1024 * Data abort dispatcher
1025 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1026 */
1027	vector_stub	dabt, ABT_MODE, 8
1028
1029	.long	__dabt_usr			@  0  (USR_26 / USR_32)
1030	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
1031	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
1032	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
1033	.long	__dabt_invalid			@  4
1034	.long	__dabt_invalid			@  5
1035	.long	__dabt_invalid			@  6
1036	.long	__dabt_invalid			@  7
1037	.long	__dabt_invalid			@  8
1038	.long	__dabt_invalid			@  9
1039	.long	__dabt_invalid			@  a
1040	.long	__dabt_invalid			@  b
1041	.long	__dabt_invalid			@  c
1042	.long	__dabt_invalid			@  d
1043	.long	__dabt_invalid			@  e
1044	.long	__dabt_invalid			@  f
1045
1046/*
1047 * Prefetch abort dispatcher
1048 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1049 */
1050	vector_stub	pabt, ABT_MODE, 4
1051
1052	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1053	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1054	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1055	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1056	.long	__pabt_invalid			@  4
1057	.long	__pabt_invalid			@  5
1058	.long	__pabt_invalid			@  6
1059	.long	__pabt_invalid			@  7
1060	.long	__pabt_invalid			@  8
1061	.long	__pabt_invalid			@  9
1062	.long	__pabt_invalid			@  a
1063	.long	__pabt_invalid			@  b
1064	.long	__pabt_invalid			@  c
1065	.long	__pabt_invalid			@  d
1066	.long	__pabt_invalid			@  e
1067	.long	__pabt_invalid			@  f
1068
1069/*
1070 * Undef instr entry dispatcher
1071 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1072 */
1073	vector_stub	und, UND_MODE
1074
1075	.long	__und_usr			@  0 (USR_26 / USR_32)
1076	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1077	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1078	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1079	.long	__und_invalid			@  4
1080	.long	__und_invalid			@  5
1081	.long	__und_invalid			@  6
1082	.long	__und_invalid			@  7
1083	.long	__und_invalid			@  8
1084	.long	__und_invalid			@  9
1085	.long	__und_invalid			@  a
1086	.long	__und_invalid			@  b
1087	.long	__und_invalid			@  c
1088	.long	__und_invalid			@  d
1089	.long	__und_invalid			@  e
1090	.long	__und_invalid			@  f
1091
1092	.align	5
1093
1094/*=============================================================================
1095 * Undefined FIQs
1096 *-----------------------------------------------------------------------------
1097 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1098 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1099 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1100 * damage alert!  I don't think that we can execute any code in here in any
1101 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1102 * get out of that mode without clobbering one register.
1103 */
1104vector_fiq:
1105	disable_fiq
1106	subs	pc, lr, #4
1107
1108/*=============================================================================
1109 * Address exception handler
1110 *-----------------------------------------------------------------------------
1111 * These aren't too critical.
1112 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1113 */
1114
1115vector_addrexcptn:
1116	b	vector_addrexcptn
1117
1118/*
1119 * We group all the following data together to optimise
1120 * for CPUs with separate I & D caches.
1121 */
1122	.align	5
1123
1124.LCvswi:
1125	.word	vector_swi
1126
1127	.globl	__stubs_end
1128__stubs_end:
1129
1130	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1131
1132	.globl	__vectors_start
1133__vectors_start:
1134	swi	SYS_ERROR0
1135	b	vector_und + stubs_offset
1136	ldr	pc, .LCvswi + stubs_offset
1137	b	vector_pabt + stubs_offset
1138	b	vector_dabt + stubs_offset
1139	b	vector_addrexcptn + stubs_offset
1140	b	vector_irq + stubs_offset
1141	b	vector_fiq + stubs_offset
1142
1143	.globl	__vectors_end
1144__vectors_end:
1145
1146	.data
1147
1148	.globl	cr_alignment
1149	.globl	cr_no_alignment
1150cr_alignment:
1151	.space	4
1152cr_no_alignment:
1153	.space	4
1154