xref: /linux/arch/arm/kernel/entry-armv.S (revision 7f3edee81fbd49114c28057512906f169caa0bed)
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 that causes
15 *  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/*
62 * Invalid mode handlers
63 */
64	.macro	inv_entry, reason
65	sub	sp, sp, #S_FRAME_SIZE
66	stmib	sp, {r1 - lr}
67	mov	r1, #\reason
68	.endm
69
70__pabt_invalid:
71	inv_entry BAD_PREFETCH
72	b	common_invalid
73
74__dabt_invalid:
75	inv_entry BAD_DATA
76	b	common_invalid
77
78__irq_invalid:
79	inv_entry BAD_IRQ
80	b	common_invalid
81
82__und_invalid:
83	inv_entry BAD_UNDEFINSTR
84
85	@
86	@ XXX fall through to common_invalid
87	@
88
89@
90@ common_invalid - generic code for failed exception (re-entrant version of handlers)
91@
92common_invalid:
93	zero_fp
94
95	ldmia	r0, {r4 - r6}
96	add	r0, sp, #S_PC		@ here for interlock avoidance
97	mov	r7, #-1			@  ""   ""    ""        ""
98	str	r4, [sp]		@ save preserved r0
99	stmia	r0, {r5 - r7}		@ lr_<exception>,
100					@ cpsr_<exception>, "old_r0"
101
102	mov	r0, sp
103	b	bad_mode
104
105/*
106 * SVC mode handlers
107 */
108
109#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
110#define SPFIX(code...) code
111#else
112#define SPFIX(code...)
113#endif
114
115	.macro	svc_entry
116	sub	sp, sp, #S_FRAME_SIZE
117 SPFIX(	tst	sp, #4		)
118 SPFIX(	bicne	sp, sp, #4	)
119	stmib	sp, {r1 - r12}
120
121	ldmia	r0, {r1 - r3}
122	add	r5, sp, #S_SP		@ here for interlock avoidance
123	mov	r4, #-1			@  ""  ""      ""       ""
124	add	r0, sp, #S_FRAME_SIZE   @  ""  ""      ""       ""
125 SPFIX(	addne	r0, r0, #4	)
126	str	r1, [sp]		@ save the "real" r0 copied
127					@ from the exception stack
128
129	mov	r1, lr
130
131	@
132	@ We are now ready to fill in the remaining blanks on the stack:
133	@
134	@  r0 - sp_svc
135	@  r1 - lr_svc
136	@  r2 - lr_<exception>, already fixed up for correct return/restart
137	@  r3 - spsr_<exception>
138	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
139	@
140	stmia	r5, {r0 - r4}
141	.endm
142
143	.align	5
144__dabt_svc:
145	svc_entry
146
147	@
148	@ get ready to re-enable interrupts if appropriate
149	@
150	mrs	r9, cpsr
151	tst	r3, #PSR_I_BIT
152	biceq	r9, r9, #PSR_I_BIT
153
154	@
155	@ Call the processor-specific abort handler:
156	@
157	@  r2 - aborted context pc
158	@  r3 - aborted context cpsr
159	@
160	@ The abort handler must return the aborted address in r0, and
161	@ the fault status register in r1.  r9 must be preserved.
162	@
163#ifdef MULTI_ABORT
164	ldr	r4, .LCprocfns
165	mov	lr, pc
166	ldr	pc, [r4]
167#else
168	bl	CPU_ABORT_HANDLER
169#endif
170
171	@
172	@ set desired IRQ state, then call main handler
173	@
174	msr	cpsr_c, r9
175	mov	r2, sp
176	bl	do_DataAbort
177
178	@
179	@ IRQs off again before pulling preserved data off the stack
180	@
181	disable_irq
182
183	@
184	@ restore SPSR and restart the instruction
185	@
186	ldr	r0, [sp, #S_PSR]
187	msr	spsr_cxsf, r0
188	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
189
190	.align	5
191__irq_svc:
192	svc_entry
193
194#ifdef CONFIG_TRACE_IRQFLAGS
195	bl	trace_hardirqs_off
196#endif
197#ifdef CONFIG_PREEMPT
198	get_thread_info tsk
199	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
200	add	r7, r8, #1			@ increment it
201	str	r7, [tsk, #TI_PREEMPT]
202#endif
203
204	irq_handler
205#ifdef CONFIG_PREEMPT
206	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
207	tst	r0, #_TIF_NEED_RESCHED
208	blne	svc_preempt
209preempt_return:
210	ldr	r0, [tsk, #TI_PREEMPT]		@ read preempt value
211	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
212	teq	r0, r7
213	strne	r0, [r0, -r0]			@ bug()
214#endif
215	ldr	r0, [sp, #S_PSR]		@ irqs are already disabled
216	msr	spsr_cxsf, r0
217#ifdef CONFIG_TRACE_IRQFLAGS
218	tst	r0, #PSR_I_BIT
219	bleq	trace_hardirqs_on
220#endif
221	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
222
223	.ltorg
224
225#ifdef CONFIG_PREEMPT
226svc_preempt:
227	teq	r8, #0				@ was preempt count = 0
228	ldreq	r6, .LCirq_stat
229	movne	pc, lr				@ no
230	ldr	r0, [r6, #4]			@ local_irq_count
231	ldr	r1, [r6, #8]			@ local_bh_count
232	adds	r0, r0, r1
233	movne	pc, lr
234	mov	r7, #0				@ preempt_schedule_irq
235	str	r7, [tsk, #TI_PREEMPT]		@ expects preempt_count == 0
2361:	bl	preempt_schedule_irq		@ irq en/disable is done inside
237	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
238	tst	r0, #_TIF_NEED_RESCHED
239	beq	preempt_return			@ go again
240	b	1b
241#endif
242
243	.align	5
244__und_svc:
245	svc_entry
246
247	@
248	@ call emulation code, which returns using r9 if it has emulated
249	@ the instruction, or the more conventional lr if we are to treat
250	@ this as a real undefined instruction
251	@
252	@  r0 - instruction
253	@
254	ldr	r0, [r2, #-4]
255	adr	r9, 1f
256	bl	call_fpe
257
258	mov	r0, sp				@ struct pt_regs *regs
259	bl	do_undefinstr
260
261	@
262	@ IRQs off again before pulling preserved data off the stack
263	@
2641:	disable_irq
265
266	@
267	@ restore SPSR and restart the instruction
268	@
269	ldr	lr, [sp, #S_PSR]		@ Get SVC cpsr
270	msr	spsr_cxsf, lr
271	ldmia	sp, {r0 - pc}^			@ Restore SVC registers
272
273	.align	5
274__pabt_svc:
275	svc_entry
276
277	@
278	@ re-enable interrupts if appropriate
279	@
280	mrs	r9, cpsr
281	tst	r3, #PSR_I_BIT
282	biceq	r9, r9, #PSR_I_BIT
283	msr	cpsr_c, r9
284
285	@
286	@ set args, then call main handler
287	@
288	@  r0 - address of faulting instruction
289	@  r1 - pointer to registers on stack
290	@
291	mov	r0, r2				@ address (pc)
292	mov	r1, sp				@ regs
293	bl	do_PrefetchAbort		@ call abort handler
294
295	@
296	@ IRQs off again before pulling preserved data off the stack
297	@
298	disable_irq
299
300	@
301	@ restore SPSR and restart the instruction
302	@
303	ldr	r0, [sp, #S_PSR]
304	msr	spsr_cxsf, r0
305	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
306
307	.align	5
308.LCcralign:
309	.word	cr_alignment
310#ifdef MULTI_ABORT
311.LCprocfns:
312	.word	processor
313#endif
314.LCfp:
315	.word	fp_enter
316#ifdef CONFIG_PREEMPT
317.LCirq_stat:
318	.word	irq_stat
319#endif
320
321/*
322 * User mode handlers
323 *
324 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
325 */
326
327#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
328#error "sizeof(struct pt_regs) must be a multiple of 8"
329#endif
330
331	.macro	usr_entry
332	sub	sp, sp, #S_FRAME_SIZE
333	stmib	sp, {r1 - r12}
334
335	ldmia	r0, {r1 - r3}
336	add	r0, sp, #S_PC		@ here for interlock avoidance
337	mov	r4, #-1			@  ""  ""     ""        ""
338
339	str	r1, [sp]		@ save the "real" r0 copied
340					@ from the exception stack
341
342	@
343	@ We are now ready to fill in the remaining blanks on the stack:
344	@
345	@  r2 - lr_<exception>, already fixed up for correct return/restart
346	@  r3 - spsr_<exception>
347	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
348	@
349	@ Also, separately save sp_usr and lr_usr
350	@
351	stmia	r0, {r2 - r4}
352	stmdb	r0, {sp, lr}^
353
354	@
355	@ Enable the alignment trap while in kernel mode
356	@
357	alignment_trap r0
358
359	@
360	@ Clear FP to mark the first stack frame
361	@
362	zero_fp
363	.endm
364
365	.macro	kuser_cmpxchg_check
366#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
367#ifndef CONFIG_MMU
368#warning "NPTL on non MMU needs fixing"
369#else
370	@ Make sure our user space atomic helper is restarted
371	@ if it was interrupted in a critical region.  Here we
372	@ perform a quick test inline since it should be false
373	@ 99.9999% of the time.  The rest is done out of line.
374	cmp	r2, #TASK_SIZE
375	blhs	kuser_cmpxchg_fixup
376#endif
377#endif
378	.endm
379
380	.align	5
381__dabt_usr:
382	usr_entry
383	kuser_cmpxchg_check
384
385	@
386	@ Call the processor-specific abort handler:
387	@
388	@  r2 - aborted context pc
389	@  r3 - aborted context cpsr
390	@
391	@ The abort handler must return the aborted address in r0, and
392	@ the fault status register in r1.
393	@
394#ifdef MULTI_ABORT
395	ldr	r4, .LCprocfns
396	mov	lr, pc
397	ldr	pc, [r4]
398#else
399	bl	CPU_ABORT_HANDLER
400#endif
401
402	@
403	@ IRQs on, then call the main handler
404	@
405	enable_irq
406	mov	r2, sp
407	adr	lr, ret_from_exception
408	b	do_DataAbort
409
410	.align	5
411__irq_usr:
412	usr_entry
413	kuser_cmpxchg_check
414
415#ifdef CONFIG_TRACE_IRQFLAGS
416	bl	trace_hardirqs_off
417#endif
418	get_thread_info tsk
419#ifdef CONFIG_PREEMPT
420	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
421	add	r7, r8, #1			@ increment it
422	str	r7, [tsk, #TI_PREEMPT]
423#endif
424
425	irq_handler
426#ifdef CONFIG_PREEMPT
427	ldr	r0, [tsk, #TI_PREEMPT]
428	str	r8, [tsk, #TI_PREEMPT]
429	teq	r0, r7
430	strne	r0, [r0, -r0]
431#endif
432#ifdef CONFIG_TRACE_IRQFLAGS
433	bl	trace_hardirqs_on
434#endif
435
436	mov	why, #0
437	b	ret_to_user
438
439	.ltorg
440
441	.align	5
442__und_usr:
443	usr_entry
444
445	tst	r3, #PSR_T_BIT			@ Thumb mode?
446	bne	__und_usr_unknown		@ ignore FP
447	sub	r4, r2, #4
448
449	@
450	@ fall through to the emulation code, which returns using r9 if
451	@ it has emulated the instruction, or the more conventional lr
452	@ if we are to treat this as a real undefined instruction
453	@
454	@  r0 - instruction
455	@
456	adr	r9, ret_from_exception
457	adr	lr, __und_usr_unknown
4581:	ldrt	r0, [r4]
459	@
460	@ fallthrough to call_fpe
461	@
462
463/*
464 * The out of line fixup for the ldrt above.
465 */
466	.section .fixup, "ax"
4672:	mov	pc, r9
468	.previous
469	.section __ex_table,"a"
470	.long	1b, 2b
471	.previous
472
473/*
474 * Check whether the instruction is a co-processor instruction.
475 * If yes, we need to call the relevant co-processor handler.
476 *
477 * Note that we don't do a full check here for the co-processor
478 * instructions; all instructions with bit 27 set are well
479 * defined.  The only instructions that should fault are the
480 * co-processor instructions.  However, we have to watch out
481 * for the ARM6/ARM7 SWI bug.
482 *
483 * Emulators may wish to make use of the following registers:
484 *  r0  = instruction opcode.
485 *  r2  = PC+4
486 *  r9  = normal "successful" return address
487 *  r10 = this threads thread_info structure.
488 *  lr  = unrecognised instruction return address
489 */
490call_fpe:
491	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
492#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
493	and	r8, r0, #0x0f000000		@ mask out op-code bits
494	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
495#endif
496	moveq	pc, lr
497	get_thread_info r10			@ get current thread
498	and	r8, r0, #0x00000f00		@ mask out CP number
499	mov	r7, #1
500	add	r6, r10, #TI_USED_CP
501	strb	r7, [r6, r8, lsr #8]		@ set appropriate used_cp[]
502#ifdef CONFIG_IWMMXT
503	@ Test if we need to give access to iWMMXt coprocessors
504	ldr	r5, [r10, #TI_FLAGS]
505	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
506	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
507	bcs	iwmmxt_task_enable
508#endif
509	add	pc, pc, r8, lsr #6
510	mov	r0, r0
511
512	mov	pc, lr				@ CP#0
513	b	do_fpe				@ CP#1 (FPE)
514	b	do_fpe				@ CP#2 (FPE)
515	mov	pc, lr				@ CP#3
516#ifdef CONFIG_CRUNCH
517	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
518	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
519	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
520#else
521	mov	pc, lr				@ CP#4
522	mov	pc, lr				@ CP#5
523	mov	pc, lr				@ CP#6
524#endif
525	mov	pc, lr				@ CP#7
526	mov	pc, lr				@ CP#8
527	mov	pc, lr				@ CP#9
528#ifdef CONFIG_VFP
529	b	do_vfp				@ CP#10 (VFP)
530	b	do_vfp				@ CP#11 (VFP)
531#else
532	mov	pc, lr				@ CP#10 (VFP)
533	mov	pc, lr				@ CP#11 (VFP)
534#endif
535	mov	pc, lr				@ CP#12
536	mov	pc, lr				@ CP#13
537	mov	pc, lr				@ CP#14 (Debug)
538	mov	pc, lr				@ CP#15 (Control)
539
540do_fpe:
541	enable_irq
542	ldr	r4, .LCfp
543	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
544	ldr	pc, [r4]			@ Call FP module USR entry point
545
546/*
547 * The FP module is called with these registers set:
548 *  r0  = instruction
549 *  r2  = PC+4
550 *  r9  = normal "successful" return address
551 *  r10 = FP workspace
552 *  lr  = unrecognised FP instruction return address
553 */
554
555	.data
556ENTRY(fp_enter)
557	.word	no_fp
558	.text
559
560no_fp:	mov	pc, lr
561
562__und_usr_unknown:
563	mov	r0, sp
564	adr	lr, ret_from_exception
565	b	do_undefinstr
566
567	.align	5
568__pabt_usr:
569	usr_entry
570
571	enable_irq				@ Enable interrupts
572	mov	r0, r2				@ address (pc)
573	mov	r1, sp				@ regs
574	bl	do_PrefetchAbort		@ call abort handler
575	/* fall through */
576/*
577 * This is the return code to user mode for abort handlers
578 */
579ENTRY(ret_from_exception)
580	get_thread_info tsk
581	mov	why, #0
582	b	ret_to_user
583
584/*
585 * Register switch for ARMv3 and ARMv4 processors
586 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
587 * previous and next are guaranteed not to be the same.
588 */
589ENTRY(__switch_to)
590	add	ip, r1, #TI_CPU_SAVE
591	ldr	r3, [r2, #TI_TP_VALUE]
592	stmia	ip!, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
593#ifdef CONFIG_MMU
594	ldr	r6, [r2, #TI_CPU_DOMAIN]
595#endif
596#if __LINUX_ARM_ARCH__ >= 6
597#ifdef CONFIG_CPU_32v6K
598	clrex
599#else
600	strex	r5, r4, [ip]			@ Clear exclusive monitor
601#endif
602#endif
603#if defined(CONFIG_HAS_TLS_REG)
604	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
605#elif !defined(CONFIG_TLS_REG_EMUL)
606	mov	r4, #0xffff0fff
607	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
608#endif
609#ifdef CONFIG_MMU
610	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
611#endif
612	mov	r5, r0
613	add	r4, r2, #TI_CPU_SAVE
614	ldr	r0, =thread_notify_head
615	mov	r1, #THREAD_NOTIFY_SWITCH
616	bl	atomic_notifier_call_chain
617	mov	r0, r5
618	ldmia	r4, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
619
620	__INIT
621
622/*
623 * User helpers.
624 *
625 * These are segment of kernel provided user code reachable from user space
626 * at a fixed address in kernel memory.  This is used to provide user space
627 * with some operations which require kernel help because of unimplemented
628 * native feature and/or instructions in many ARM CPUs. The idea is for
629 * this code to be executed directly in user mode for best efficiency but
630 * which is too intimate with the kernel counter part to be left to user
631 * libraries.  In fact this code might even differ from one CPU to another
632 * depending on the available  instruction set and restrictions like on
633 * SMP systems.  In other words, the kernel reserves the right to change
634 * this code as needed without warning. Only the entry points and their
635 * results are guaranteed to be stable.
636 *
637 * Each segment is 32-byte aligned and will be moved to the top of the high
638 * vector page.  New segments (if ever needed) must be added in front of
639 * existing ones.  This mechanism should be used only for things that are
640 * really small and justified, and not be abused freely.
641 *
642 * User space is expected to implement those things inline when optimizing
643 * for a processor that has the necessary native support, but only if such
644 * resulting binaries are already to be incompatible with earlier ARM
645 * processors due to the use of unsupported instructions other than what
646 * is provided here.  In other words don't make binaries unable to run on
647 * earlier processors just for the sake of not using these kernel helpers
648 * if your compiled code is not going to use the new instructions for other
649 * purpose.
650 */
651
652	.macro	usr_ret, reg
653#ifdef CONFIG_ARM_THUMB
654	bx	\reg
655#else
656	mov	pc, \reg
657#endif
658	.endm
659
660	.align	5
661	.globl	__kuser_helper_start
662__kuser_helper_start:
663
664/*
665 * Reference prototype:
666 *
667 *	void __kernel_memory_barrier(void)
668 *
669 * Input:
670 *
671 *	lr = return address
672 *
673 * Output:
674 *
675 *	none
676 *
677 * Clobbered:
678 *
679 *	none
680 *
681 * Definition and user space usage example:
682 *
683 *	typedef void (__kernel_dmb_t)(void);
684 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
685 *
686 * Apply any needed memory barrier to preserve consistency with data modified
687 * manually and __kuser_cmpxchg usage.
688 *
689 * This could be used as follows:
690 *
691 * #define __kernel_dmb() \
692 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
693 *	        : : : "r0", "lr","cc" )
694 */
695
696__kuser_memory_barrier:				@ 0xffff0fa0
697
698#if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
699	mcr	p15, 0, r0, c7, c10, 5	@ dmb
700#endif
701	usr_ret	lr
702
703	.align	5
704
705/*
706 * Reference prototype:
707 *
708 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
709 *
710 * Input:
711 *
712 *	r0 = oldval
713 *	r1 = newval
714 *	r2 = ptr
715 *	lr = return address
716 *
717 * Output:
718 *
719 *	r0 = returned value (zero or non-zero)
720 *	C flag = set if r0 == 0, clear if r0 != 0
721 *
722 * Clobbered:
723 *
724 *	r3, ip, flags
725 *
726 * Definition and user space usage example:
727 *
728 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
729 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
730 *
731 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
732 * Return zero if *ptr was changed or non-zero if no exchange happened.
733 * The C flag is also set if *ptr was changed to allow for assembly
734 * optimization in the calling code.
735 *
736 * Notes:
737 *
738 *    - This routine already includes memory barriers as needed.
739 *
740 * For example, a user space atomic_add implementation could look like this:
741 *
742 * #define atomic_add(ptr, val) \
743 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
744 *	   register unsigned int __result asm("r1"); \
745 *	   asm volatile ( \
746 *	       "1: @ atomic_add\n\t" \
747 *	       "ldr	r0, [r2]\n\t" \
748 *	       "mov	r3, #0xffff0fff\n\t" \
749 *	       "add	lr, pc, #4\n\t" \
750 *	       "add	r1, r0, %2\n\t" \
751 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
752 *	       "bcc	1b" \
753 *	       : "=&r" (__result) \
754 *	       : "r" (__ptr), "rIL" (val) \
755 *	       : "r0","r3","ip","lr","cc","memory" ); \
756 *	   __result; })
757 */
758
759__kuser_cmpxchg:				@ 0xffff0fc0
760
761#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
762
763	/*
764	 * Poor you.  No fast solution possible...
765	 * The kernel itself must perform the operation.
766	 * A special ghost syscall is used for that (see traps.c).
767	 */
768	stmfd	sp!, {r7, lr}
769	mov	r7, #0xff00		@ 0xfff0 into r7 for EABI
770	orr	r7, r7, #0xf0
771	swi	#0x9ffff0
772	ldmfd	sp!, {r7, pc}
773
774#elif __LINUX_ARM_ARCH__ < 6
775
776#ifdef CONFIG_MMU
777
778	/*
779	 * The only thing that can break atomicity in this cmpxchg
780	 * implementation is either an IRQ or a data abort exception
781	 * causing another process/thread to be scheduled in the middle
782	 * of the critical sequence.  To prevent this, code is added to
783	 * the IRQ and data abort exception handlers to set the pc back
784	 * to the beginning of the critical section if it is found to be
785	 * within that critical section (see kuser_cmpxchg_fixup).
786	 */
7871:	ldr	r3, [r2]			@ load current val
788	subs	r3, r3, r0			@ compare with oldval
7892:	streq	r1, [r2]			@ store newval if eq
790	rsbs	r0, r3, #0			@ set return val and C flag
791	usr_ret	lr
792
793	.text
794kuser_cmpxchg_fixup:
795	@ Called from kuser_cmpxchg_check macro.
796	@ r2 = address of interrupted insn (must be preserved).
797	@ sp = saved regs. r7 and r8 are clobbered.
798	@ 1b = first critical insn, 2b = last critical insn.
799	@ If r2 >= 1b and r2 <= 2b then saved pc_usr is set to 1b.
800	mov	r7, #0xffff0fff
801	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
802	subs	r8, r2, r7
803	rsbcss	r8, r8, #(2b - 1b)
804	strcs	r7, [sp, #S_PC]
805	mov	pc, lr
806	.previous
807
808#else
809#warning "NPTL on non MMU needs fixing"
810	mov	r0, #-1
811	adds	r0, r0, #0
812	usr_ret	lr
813#endif
814
815#else
816
817#ifdef CONFIG_SMP
818	mcr	p15, 0, r0, c7, c10, 5	@ dmb
819#endif
8201:	ldrex	r3, [r2]
821	subs	r3, r3, r0
822	strexeq	r3, r1, [r2]
823	teqeq	r3, #1
824	beq	1b
825	rsbs	r0, r3, #0
826	/* beware -- each __kuser slot must be 8 instructions max */
827#ifdef CONFIG_SMP
828	b	__kuser_memory_barrier
829#else
830	usr_ret	lr
831#endif
832
833#endif
834
835	.align	5
836
837/*
838 * Reference prototype:
839 *
840 *	int __kernel_get_tls(void)
841 *
842 * Input:
843 *
844 *	lr = return address
845 *
846 * Output:
847 *
848 *	r0 = TLS value
849 *
850 * Clobbered:
851 *
852 *	none
853 *
854 * Definition and user space usage example:
855 *
856 *	typedef int (__kernel_get_tls_t)(void);
857 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
858 *
859 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
860 *
861 * This could be used as follows:
862 *
863 * #define __kernel_get_tls() \
864 *	({ register unsigned int __val asm("r0"); \
865 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
866 *	        : "=r" (__val) : : "lr","cc" ); \
867 *	   __val; })
868 */
869
870__kuser_get_tls:				@ 0xffff0fe0
871
872#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
873	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
874#else
875	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
876#endif
877	usr_ret	lr
878
879	.rep	5
880	.word	0			@ pad up to __kuser_helper_version
881	.endr
882
883/*
884 * Reference declaration:
885 *
886 *	extern unsigned int __kernel_helper_version;
887 *
888 * Definition and user space usage example:
889 *
890 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
891 *
892 * User space may read this to determine the curent number of helpers
893 * available.
894 */
895
896__kuser_helper_version:				@ 0xffff0ffc
897	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
898
899	.globl	__kuser_helper_end
900__kuser_helper_end:
901
902
903/*
904 * Vector stubs.
905 *
906 * This code is copied to 0xffff0200 so we can use branches in the
907 * vectors, rather than ldr's.  Note that this code must not
908 * exceed 0x300 bytes.
909 *
910 * Common stub entry macro:
911 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
912 *
913 * SP points to a minimal amount of processor-private memory, the address
914 * of which is copied into r0 for the mode specific abort handler.
915 */
916	.macro	vector_stub, name, mode, correction=0
917	.align	5
918
919vector_\name:
920	.if \correction
921	sub	lr, lr, #\correction
922	.endif
923
924	@
925	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
926	@ (parent CPSR)
927	@
928	stmia	sp, {r0, lr}		@ save r0, lr
929	mrs	lr, spsr
930	str	lr, [sp, #8]		@ save spsr
931
932	@
933	@ Prepare for SVC32 mode.  IRQs remain disabled.
934	@
935	mrs	r0, cpsr
936	eor	r0, r0, #(\mode ^ SVC_MODE)
937	msr	spsr_cxsf, r0
938
939	@
940	@ the branch table must immediately follow this code
941	@
942	and	lr, lr, #0x0f
943	mov	r0, sp
944	ldr	lr, [pc, lr, lsl #2]
945	movs	pc, lr			@ branch to handler in SVC mode
946	.endm
947
948	.globl	__stubs_start
949__stubs_start:
950/*
951 * Interrupt dispatcher
952 */
953	vector_stub	irq, IRQ_MODE, 4
954
955	.long	__irq_usr			@  0  (USR_26 / USR_32)
956	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
957	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
958	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
959	.long	__irq_invalid			@  4
960	.long	__irq_invalid			@  5
961	.long	__irq_invalid			@  6
962	.long	__irq_invalid			@  7
963	.long	__irq_invalid			@  8
964	.long	__irq_invalid			@  9
965	.long	__irq_invalid			@  a
966	.long	__irq_invalid			@  b
967	.long	__irq_invalid			@  c
968	.long	__irq_invalid			@  d
969	.long	__irq_invalid			@  e
970	.long	__irq_invalid			@  f
971
972/*
973 * Data abort dispatcher
974 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
975 */
976	vector_stub	dabt, ABT_MODE, 8
977
978	.long	__dabt_usr			@  0  (USR_26 / USR_32)
979	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
980	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
981	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
982	.long	__dabt_invalid			@  4
983	.long	__dabt_invalid			@  5
984	.long	__dabt_invalid			@  6
985	.long	__dabt_invalid			@  7
986	.long	__dabt_invalid			@  8
987	.long	__dabt_invalid			@  9
988	.long	__dabt_invalid			@  a
989	.long	__dabt_invalid			@  b
990	.long	__dabt_invalid			@  c
991	.long	__dabt_invalid			@  d
992	.long	__dabt_invalid			@  e
993	.long	__dabt_invalid			@  f
994
995/*
996 * Prefetch abort dispatcher
997 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
998 */
999	vector_stub	pabt, ABT_MODE, 4
1000
1001	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1002	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1003	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1004	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1005	.long	__pabt_invalid			@  4
1006	.long	__pabt_invalid			@  5
1007	.long	__pabt_invalid			@  6
1008	.long	__pabt_invalid			@  7
1009	.long	__pabt_invalid			@  8
1010	.long	__pabt_invalid			@  9
1011	.long	__pabt_invalid			@  a
1012	.long	__pabt_invalid			@  b
1013	.long	__pabt_invalid			@  c
1014	.long	__pabt_invalid			@  d
1015	.long	__pabt_invalid			@  e
1016	.long	__pabt_invalid			@  f
1017
1018/*
1019 * Undef instr entry dispatcher
1020 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1021 */
1022	vector_stub	und, UND_MODE
1023
1024	.long	__und_usr			@  0 (USR_26 / USR_32)
1025	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1026	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1027	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1028	.long	__und_invalid			@  4
1029	.long	__und_invalid			@  5
1030	.long	__und_invalid			@  6
1031	.long	__und_invalid			@  7
1032	.long	__und_invalid			@  8
1033	.long	__und_invalid			@  9
1034	.long	__und_invalid			@  a
1035	.long	__und_invalid			@  b
1036	.long	__und_invalid			@  c
1037	.long	__und_invalid			@  d
1038	.long	__und_invalid			@  e
1039	.long	__und_invalid			@  f
1040
1041	.align	5
1042
1043/*=============================================================================
1044 * Undefined FIQs
1045 *-----------------------------------------------------------------------------
1046 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1047 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1048 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1049 * damage alert!  I don't think that we can execute any code in here in any
1050 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1051 * get out of that mode without clobbering one register.
1052 */
1053vector_fiq:
1054	disable_fiq
1055	subs	pc, lr, #4
1056
1057/*=============================================================================
1058 * Address exception handler
1059 *-----------------------------------------------------------------------------
1060 * These aren't too critical.
1061 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1062 */
1063
1064vector_addrexcptn:
1065	b	vector_addrexcptn
1066
1067/*
1068 * We group all the following data together to optimise
1069 * for CPUs with separate I & D caches.
1070 */
1071	.align	5
1072
1073.LCvswi:
1074	.word	vector_swi
1075
1076	.globl	__stubs_end
1077__stubs_end:
1078
1079	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1080
1081	.globl	__vectors_start
1082__vectors_start:
1083	swi	SYS_ERROR0
1084	b	vector_und + stubs_offset
1085	ldr	pc, .LCvswi + stubs_offset
1086	b	vector_pabt + stubs_offset
1087	b	vector_dabt + stubs_offset
1088	b	vector_addrexcptn + stubs_offset
1089	b	vector_irq + stubs_offset
1090	b	vector_fiq + stubs_offset
1091
1092	.globl	__vectors_end
1093__vectors_end:
1094
1095	.data
1096
1097	.globl	cr_alignment
1098	.globl	cr_no_alignment
1099cr_alignment:
1100	.space	4
1101cr_no_alignment:
1102	.space	4
1103