xref: /linux/arch/arm/kernel/entry-armv.S (revision 5e8d780d745c1619aba81fe7166c5a4b5cad2b84)
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#include <linux/config.h>
18
19#include <asm/memory.h>
20#include <asm/glue.h>
21#include <asm/vfpmacros.h>
22#include <asm/arch/entry-macro.S>
23#include <asm/thread_notify.h>
24
25#include "entry-header.S"
26
27/*
28 * Interrupt handling.  Preserves r7, r8, r9
29 */
30	.macro	irq_handler
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	and	r2, r6, #0x1f
104	b	bad_mode
105
106/*
107 * SVC mode handlers
108 */
109
110#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
111#define SPFIX(code...) code
112#else
113#define SPFIX(code...)
114#endif
115
116	.macro	svc_entry
117	sub	sp, sp, #S_FRAME_SIZE
118 SPFIX(	tst	sp, #4		)
119 SPFIX(	bicne	sp, sp, #4	)
120	stmib	sp, {r1 - r12}
121
122	ldmia	r0, {r1 - r3}
123	add	r5, sp, #S_SP		@ here for interlock avoidance
124	mov	r4, #-1			@  ""  ""      ""       ""
125	add	r0, sp, #S_FRAME_SIZE   @  ""  ""      ""       ""
126 SPFIX(	addne	r0, r0, #4	)
127	str	r1, [sp]		@ save the "real" r0 copied
128					@ from the exception stack
129
130	mov	r1, lr
131
132	@
133	@ We are now ready to fill in the remaining blanks on the stack:
134	@
135	@  r0 - sp_svc
136	@  r1 - lr_svc
137	@  r2 - lr_<exception>, already fixed up for correct return/restart
138	@  r3 - spsr_<exception>
139	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
140	@
141	stmia	r5, {r0 - r4}
142	.endm
143
144	.align	5
145__dabt_svc:
146	svc_entry
147
148	@
149	@ get ready to re-enable interrupts if appropriate
150	@
151	mrs	r9, cpsr
152	tst	r3, #PSR_I_BIT
153	biceq	r9, r9, #PSR_I_BIT
154
155	@
156	@ Call the processor-specific abort handler:
157	@
158	@  r2 - aborted context pc
159	@  r3 - aborted context cpsr
160	@
161	@ The abort handler must return the aborted address in r0, and
162	@ the fault status register in r1.  r9 must be preserved.
163	@
164#ifdef MULTI_ABORT
165	ldr	r4, .LCprocfns
166	mov	lr, pc
167	ldr	pc, [r4]
168#else
169	bl	CPU_ABORT_HANDLER
170#endif
171
172	@
173	@ set desired IRQ state, then call main handler
174	@
175	msr	cpsr_c, r9
176	mov	r2, sp
177	bl	do_DataAbort
178
179	@
180	@ IRQs off again before pulling preserved data off the stack
181	@
182	disable_irq
183
184	@
185	@ restore SPSR and restart the instruction
186	@
187	ldr	r0, [sp, #S_PSR]
188	msr	spsr_cxsf, r0
189	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
190
191	.align	5
192__irq_svc:
193	svc_entry
194
195#ifdef CONFIG_PREEMPT
196	get_thread_info tsk
197	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
198	add	r7, r8, #1			@ increment it
199	str	r7, [tsk, #TI_PREEMPT]
200#endif
201
202	irq_handler
203#ifdef CONFIG_PREEMPT
204	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
205	tst	r0, #_TIF_NEED_RESCHED
206	blne	svc_preempt
207preempt_return:
208	ldr	r0, [tsk, #TI_PREEMPT]		@ read preempt value
209	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
210	teq	r0, r7
211	strne	r0, [r0, -r0]			@ bug()
212#endif
213	ldr	r0, [sp, #S_PSR]		@ irqs are already disabled
214	msr	spsr_cxsf, r0
215	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
216
217	.ltorg
218
219#ifdef CONFIG_PREEMPT
220svc_preempt:
221	teq	r8, #0				@ was preempt count = 0
222	ldreq	r6, .LCirq_stat
223	movne	pc, lr				@ no
224	ldr	r0, [r6, #4]			@ local_irq_count
225	ldr	r1, [r6, #8]			@ local_bh_count
226	adds	r0, r0, r1
227	movne	pc, lr
228	mov	r7, #0				@ preempt_schedule_irq
229	str	r7, [tsk, #TI_PREEMPT]		@ expects preempt_count == 0
2301:	bl	preempt_schedule_irq		@ irq en/disable is done inside
231	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
232	tst	r0, #_TIF_NEED_RESCHED
233	beq	preempt_return			@ go again
234	b	1b
235#endif
236
237	.align	5
238__und_svc:
239	svc_entry
240
241	@
242	@ call emulation code, which returns using r9 if it has emulated
243	@ the instruction, or the more conventional lr if we are to treat
244	@ this as a real undefined instruction
245	@
246	@  r0 - instruction
247	@
248	ldr	r0, [r2, #-4]
249	adr	r9, 1f
250	bl	call_fpe
251
252	mov	r0, sp				@ struct pt_regs *regs
253	bl	do_undefinstr
254
255	@
256	@ IRQs off again before pulling preserved data off the stack
257	@
2581:	disable_irq
259
260	@
261	@ restore SPSR and restart the instruction
262	@
263	ldr	lr, [sp, #S_PSR]		@ Get SVC cpsr
264	msr	spsr_cxsf, lr
265	ldmia	sp, {r0 - pc}^			@ Restore SVC registers
266
267	.align	5
268__pabt_svc:
269	svc_entry
270
271	@
272	@ re-enable interrupts if appropriate
273	@
274	mrs	r9, cpsr
275	tst	r3, #PSR_I_BIT
276	biceq	r9, r9, #PSR_I_BIT
277	msr	cpsr_c, r9
278
279	@
280	@ set args, then call main handler
281	@
282	@  r0 - address of faulting instruction
283	@  r1 - pointer to registers on stack
284	@
285	mov	r0, r2				@ address (pc)
286	mov	r1, sp				@ regs
287	bl	do_PrefetchAbort		@ call abort handler
288
289	@
290	@ IRQs off again before pulling preserved data off the stack
291	@
292	disable_irq
293
294	@
295	@ restore SPSR and restart the instruction
296	@
297	ldr	r0, [sp, #S_PSR]
298	msr	spsr_cxsf, r0
299	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
300
301	.align	5
302.LCcralign:
303	.word	cr_alignment
304#ifdef MULTI_ABORT
305.LCprocfns:
306	.word	processor
307#endif
308.LCfp:
309	.word	fp_enter
310#ifdef CONFIG_PREEMPT
311.LCirq_stat:
312	.word	irq_stat
313#endif
314
315/*
316 * User mode handlers
317 *
318 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
319 */
320
321#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
322#error "sizeof(struct pt_regs) must be a multiple of 8"
323#endif
324
325	.macro	usr_entry
326	sub	sp, sp, #S_FRAME_SIZE
327	stmib	sp, {r1 - r12}
328
329	ldmia	r0, {r1 - r3}
330	add	r0, sp, #S_PC		@ here for interlock avoidance
331	mov	r4, #-1			@  ""  ""     ""        ""
332
333	str	r1, [sp]		@ save the "real" r0 copied
334					@ from the exception stack
335
336#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
337#ifndef CONFIG_MMU
338#warning "NPTL on non MMU needs fixing"
339#else
340	@ make sure our user space atomic helper is aborted
341	cmp	r2, #TASK_SIZE
342	bichs	r3, r3, #PSR_Z_BIT
343#endif
344#endif
345
346	@
347	@ We are now ready to fill in the remaining blanks on the stack:
348	@
349	@  r2 - lr_<exception>, already fixed up for correct return/restart
350	@  r3 - spsr_<exception>
351	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
352	@
353	@ Also, separately save sp_usr and lr_usr
354	@
355	stmia	r0, {r2 - r4}
356	stmdb	r0, {sp, lr}^
357
358	@
359	@ Enable the alignment trap while in kernel mode
360	@
361	alignment_trap r0
362
363	@
364	@ Clear FP to mark the first stack frame
365	@
366	zero_fp
367	.endm
368
369	.align	5
370__dabt_usr:
371	usr_entry
372
373	@
374	@ Call the processor-specific abort handler:
375	@
376	@  r2 - aborted context pc
377	@  r3 - aborted context cpsr
378	@
379	@ The abort handler must return the aborted address in r0, and
380	@ the fault status register in r1.
381	@
382#ifdef MULTI_ABORT
383	ldr	r4, .LCprocfns
384	mov	lr, pc
385	ldr	pc, [r4]
386#else
387	bl	CPU_ABORT_HANDLER
388#endif
389
390	@
391	@ IRQs on, then call the main handler
392	@
393	enable_irq
394	mov	r2, sp
395	adr	lr, ret_from_exception
396	b	do_DataAbort
397
398	.align	5
399__irq_usr:
400	usr_entry
401
402	get_thread_info tsk
403#ifdef CONFIG_PREEMPT
404	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
405	add	r7, r8, #1			@ increment it
406	str	r7, [tsk, #TI_PREEMPT]
407#endif
408
409	irq_handler
410#ifdef CONFIG_PREEMPT
411	ldr	r0, [tsk, #TI_PREEMPT]
412	str	r8, [tsk, #TI_PREEMPT]
413	teq	r0, r7
414	strne	r0, [r0, -r0]
415#endif
416
417	mov	why, #0
418	b	ret_to_user
419
420	.ltorg
421
422	.align	5
423__und_usr:
424	usr_entry
425
426	tst	r3, #PSR_T_BIT			@ Thumb mode?
427	bne	fpundefinstr			@ ignore FP
428	sub	r4, r2, #4
429
430	@
431	@ fall through to the emulation code, which returns using r9 if
432	@ it has emulated the instruction, or the more conventional lr
433	@ if we are to treat this as a real undefined instruction
434	@
435	@  r0 - instruction
436	@
4371:	ldrt	r0, [r4]
438	adr	r9, ret_from_exception
439	adr	lr, fpundefinstr
440	@
441	@ fallthrough to call_fpe
442	@
443
444/*
445 * The out of line fixup for the ldrt above.
446 */
447	.section .fixup, "ax"
4482:	mov	pc, r9
449	.previous
450	.section __ex_table,"a"
451	.long	1b, 2b
452	.previous
453
454/*
455 * Check whether the instruction is a co-processor instruction.
456 * If yes, we need to call the relevant co-processor handler.
457 *
458 * Note that we don't do a full check here for the co-processor
459 * instructions; all instructions with bit 27 set are well
460 * defined.  The only instructions that should fault are the
461 * co-processor instructions.  However, we have to watch out
462 * for the ARM6/ARM7 SWI bug.
463 *
464 * Emulators may wish to make use of the following registers:
465 *  r0  = instruction opcode.
466 *  r2  = PC+4
467 *  r10 = this threads thread_info structure.
468 */
469call_fpe:
470	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
471#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
472	and	r8, r0, #0x0f000000		@ mask out op-code bits
473	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
474#endif
475	moveq	pc, lr
476	get_thread_info r10			@ get current thread
477	and	r8, r0, #0x00000f00		@ mask out CP number
478	mov	r7, #1
479	add	r6, r10, #TI_USED_CP
480	strb	r7, [r6, r8, lsr #8]		@ set appropriate used_cp[]
481#ifdef CONFIG_IWMMXT
482	@ Test if we need to give access to iWMMXt coprocessors
483	ldr	r5, [r10, #TI_FLAGS]
484	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
485	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
486	bcs	iwmmxt_task_enable
487#endif
488	add	pc, pc, r8, lsr #6
489	mov	r0, r0
490
491	mov	pc, lr				@ CP#0
492	b	do_fpe				@ CP#1 (FPE)
493	b	do_fpe				@ CP#2 (FPE)
494	mov	pc, lr				@ CP#3
495#ifdef CONFIG_CRUNCH
496	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
497	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
498	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
499#else
500	mov	pc, lr				@ CP#4
501	mov	pc, lr				@ CP#5
502	mov	pc, lr				@ CP#6
503#endif
504	mov	pc, lr				@ CP#7
505	mov	pc, lr				@ CP#8
506	mov	pc, lr				@ CP#9
507#ifdef CONFIG_VFP
508	b	do_vfp				@ CP#10 (VFP)
509	b	do_vfp				@ CP#11 (VFP)
510#else
511	mov	pc, lr				@ CP#10 (VFP)
512	mov	pc, lr				@ CP#11 (VFP)
513#endif
514	mov	pc, lr				@ CP#12
515	mov	pc, lr				@ CP#13
516	mov	pc, lr				@ CP#14 (Debug)
517	mov	pc, lr				@ CP#15 (Control)
518
519do_fpe:
520	enable_irq
521	ldr	r4, .LCfp
522	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
523	ldr	pc, [r4]			@ Call FP module USR entry point
524
525/*
526 * The FP module is called with these registers set:
527 *  r0  = instruction
528 *  r2  = PC+4
529 *  r9  = normal "successful" return address
530 *  r10 = FP workspace
531 *  lr  = unrecognised FP instruction return address
532 */
533
534	.data
535ENTRY(fp_enter)
536	.word	fpundefinstr
537	.text
538
539fpundefinstr:
540	mov	r0, sp
541	adr	lr, ret_from_exception
542	b	do_undefinstr
543
544	.align	5
545__pabt_usr:
546	usr_entry
547
548	enable_irq				@ Enable interrupts
549	mov	r0, r2				@ address (pc)
550	mov	r1, sp				@ regs
551	bl	do_PrefetchAbort		@ call abort handler
552	/* fall through */
553/*
554 * This is the return code to user mode for abort handlers
555 */
556ENTRY(ret_from_exception)
557	get_thread_info tsk
558	mov	why, #0
559	b	ret_to_user
560
561/*
562 * Register switch for ARMv3 and ARMv4 processors
563 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
564 * previous and next are guaranteed not to be the same.
565 */
566ENTRY(__switch_to)
567	add	ip, r1, #TI_CPU_SAVE
568	ldr	r3, [r2, #TI_TP_VALUE]
569	stmia	ip!, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
570#ifdef CONFIG_MMU
571	ldr	r6, [r2, #TI_CPU_DOMAIN]
572#endif
573#if __LINUX_ARM_ARCH__ >= 6
574#ifdef CONFIG_CPU_32v6K
575	clrex
576#else
577	strex	r5, r4, [ip]			@ Clear exclusive monitor
578#endif
579#endif
580#if defined(CONFIG_CPU_XSCALE) && !defined(CONFIG_IWMMXT)
581	mra	r4, r5, acc0
582	stmia   ip, {r4, r5}
583#endif
584#if defined(CONFIG_HAS_TLS_REG)
585	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
586#elif !defined(CONFIG_TLS_REG_EMUL)
587	mov	r4, #0xffff0fff
588	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
589#endif
590#ifdef CONFIG_MMU
591	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
592#endif
593#if defined(CONFIG_IWMMXT)
594	bl	iwmmxt_task_switch
595#elif defined(CONFIG_CPU_XSCALE)
596	add	r4, r2, #TI_CPU_DOMAIN + 40	@ cpu_context_save->extra
597	ldmib	r4, {r4, r5}
598	mar	acc0, r4, r5
599#endif
600	mov	r5, r0
601	add	r4, r2, #TI_CPU_SAVE
602	ldr	r0, =thread_notify_head
603	mov	r1, #THREAD_NOTIFY_SWITCH
604	bl	atomic_notifier_call_chain
605	mov	r0, r5
606	ldmia	r4, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
607
608	__INIT
609
610/*
611 * User helpers.
612 *
613 * These are segment of kernel provided user code reachable from user space
614 * at a fixed address in kernel memory.  This is used to provide user space
615 * with some operations which require kernel help because of unimplemented
616 * native feature and/or instructions in many ARM CPUs. The idea is for
617 * this code to be executed directly in user mode for best efficiency but
618 * which is too intimate with the kernel counter part to be left to user
619 * libraries.  In fact this code might even differ from one CPU to another
620 * depending on the available  instruction set and restrictions like on
621 * SMP systems.  In other words, the kernel reserves the right to change
622 * this code as needed without warning. Only the entry points and their
623 * results are guaranteed to be stable.
624 *
625 * Each segment is 32-byte aligned and will be moved to the top of the high
626 * vector page.  New segments (if ever needed) must be added in front of
627 * existing ones.  This mechanism should be used only for things that are
628 * really small and justified, and not be abused freely.
629 *
630 * User space is expected to implement those things inline when optimizing
631 * for a processor that has the necessary native support, but only if such
632 * resulting binaries are already to be incompatible with earlier ARM
633 * processors due to the use of unsupported instructions other than what
634 * is provided here.  In other words don't make binaries unable to run on
635 * earlier processors just for the sake of not using these kernel helpers
636 * if your compiled code is not going to use the new instructions for other
637 * purpose.
638 */
639
640	.align	5
641	.globl	__kuser_helper_start
642__kuser_helper_start:
643
644/*
645 * Reference prototype:
646 *
647 *	void __kernel_memory_barrier(void)
648 *
649 * Input:
650 *
651 *	lr = return address
652 *
653 * Output:
654 *
655 *	none
656 *
657 * Clobbered:
658 *
659 *	the Z flag might be lost
660 *
661 * Definition and user space usage example:
662 *
663 *	typedef void (__kernel_dmb_t)(void);
664 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
665 *
666 * Apply any needed memory barrier to preserve consistency with data modified
667 * manually and __kuser_cmpxchg usage.
668 *
669 * This could be used as follows:
670 *
671 * #define __kernel_dmb() \
672 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
673 *	        : : : "r0", "lr","cc" )
674 */
675
676__kuser_memory_barrier:				@ 0xffff0fa0
677
678#if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
679	mcr	p15, 0, r0, c7, c10, 5	@ dmb
680#endif
681	mov	pc, lr
682
683	.align	5
684
685/*
686 * Reference prototype:
687 *
688 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
689 *
690 * Input:
691 *
692 *	r0 = oldval
693 *	r1 = newval
694 *	r2 = ptr
695 *	lr = return address
696 *
697 * Output:
698 *
699 *	r0 = returned value (zero or non-zero)
700 *	C flag = set if r0 == 0, clear if r0 != 0
701 *
702 * Clobbered:
703 *
704 *	r3, ip, flags
705 *
706 * Definition and user space usage example:
707 *
708 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
709 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
710 *
711 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
712 * Return zero if *ptr was changed or non-zero if no exchange happened.
713 * The C flag is also set if *ptr was changed to allow for assembly
714 * optimization in the calling code.
715 *
716 * Notes:
717 *
718 *    - This routine already includes memory barriers as needed.
719 *
720 *    - A failure might be transient, i.e. it is possible, although unlikely,
721 *      that "failure" be returned even if *ptr == oldval.
722 *
723 * For example, a user space atomic_add implementation could look like this:
724 *
725 * #define atomic_add(ptr, val) \
726 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
727 *	   register unsigned int __result asm("r1"); \
728 *	   asm volatile ( \
729 *	       "1: @ atomic_add\n\t" \
730 *	       "ldr	r0, [r2]\n\t" \
731 *	       "mov	r3, #0xffff0fff\n\t" \
732 *	       "add	lr, pc, #4\n\t" \
733 *	       "add	r1, r0, %2\n\t" \
734 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
735 *	       "bcc	1b" \
736 *	       : "=&r" (__result) \
737 *	       : "r" (__ptr), "rIL" (val) \
738 *	       : "r0","r3","ip","lr","cc","memory" ); \
739 *	   __result; })
740 */
741
742__kuser_cmpxchg:				@ 0xffff0fc0
743
744#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
745
746	/*
747	 * Poor you.  No fast solution possible...
748	 * The kernel itself must perform the operation.
749	 * A special ghost syscall is used for that (see traps.c).
750	 */
751	stmfd	sp!, {r7, lr}
752	mov	r7, #0xff00		@ 0xfff0 into r7 for EABI
753	orr	r7, r7, #0xf0
754	swi	#0x9ffff0
755	ldmfd	sp!, {r7, pc}
756
757#elif __LINUX_ARM_ARCH__ < 6
758
759	/*
760	 * Theory of operation:
761	 *
762	 * We set the Z flag before loading oldval. If ever an exception
763	 * occurs we can not be sure the loaded value will still be the same
764	 * when the exception returns, therefore the user exception handler
765	 * will clear the Z flag whenever the interrupted user code was
766	 * actually from the kernel address space (see the usr_entry macro).
767	 *
768	 * The post-increment on the str is used to prevent a race with an
769	 * exception happening just after the str instruction which would
770	 * clear the Z flag although the exchange was done.
771	 */
772#ifdef CONFIG_MMU
773	teq	ip, ip			@ set Z flag
774	ldr	ip, [r2]		@ load current val
775	add	r3, r2, #1		@ prepare store ptr
776	teqeq	ip, r0			@ compare with oldval if still allowed
777	streq	r1, [r3, #-1]!		@ store newval if still allowed
778	subs	r0, r2, r3		@ if r2 == r3 the str occured
779#else
780#warning "NPTL on non MMU needs fixing"
781	mov	r0, #-1
782	adds	r0, r0, #0
783#endif
784	mov	pc, lr
785
786#else
787
788#ifdef CONFIG_SMP
789	mcr	p15, 0, r0, c7, c10, 5	@ dmb
790#endif
791	ldrex	r3, [r2]
792	subs	r3, r3, r0
793	strexeq	r3, r1, [r2]
794	rsbs	r0, r3, #0
795#ifdef CONFIG_SMP
796	mcr	p15, 0, r0, c7, c10, 5	@ dmb
797#endif
798	mov	pc, lr
799
800#endif
801
802	.align	5
803
804/*
805 * Reference prototype:
806 *
807 *	int __kernel_get_tls(void)
808 *
809 * Input:
810 *
811 *	lr = return address
812 *
813 * Output:
814 *
815 *	r0 = TLS value
816 *
817 * Clobbered:
818 *
819 *	the Z flag might be lost
820 *
821 * Definition and user space usage example:
822 *
823 *	typedef int (__kernel_get_tls_t)(void);
824 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
825 *
826 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
827 *
828 * This could be used as follows:
829 *
830 * #define __kernel_get_tls() \
831 *	({ register unsigned int __val asm("r0"); \
832 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
833 *	        : "=r" (__val) : : "lr","cc" ); \
834 *	   __val; })
835 */
836
837__kuser_get_tls:				@ 0xffff0fe0
838
839#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
840
841	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
842	mov	pc, lr
843
844#else
845
846	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
847	mov	pc, lr
848
849#endif
850
851	.rep	5
852	.word	0			@ pad up to __kuser_helper_version
853	.endr
854
855/*
856 * Reference declaration:
857 *
858 *	extern unsigned int __kernel_helper_version;
859 *
860 * Definition and user space usage example:
861 *
862 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
863 *
864 * User space may read this to determine the curent number of helpers
865 * available.
866 */
867
868__kuser_helper_version:				@ 0xffff0ffc
869	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
870
871	.globl	__kuser_helper_end
872__kuser_helper_end:
873
874
875/*
876 * Vector stubs.
877 *
878 * This code is copied to 0xffff0200 so we can use branches in the
879 * vectors, rather than ldr's.  Note that this code must not
880 * exceed 0x300 bytes.
881 *
882 * Common stub entry macro:
883 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
884 *
885 * SP points to a minimal amount of processor-private memory, the address
886 * of which is copied into r0 for the mode specific abort handler.
887 */
888	.macro	vector_stub, name, mode, correction=0
889	.align	5
890
891vector_\name:
892	.if \correction
893	sub	lr, lr, #\correction
894	.endif
895
896	@
897	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
898	@ (parent CPSR)
899	@
900	stmia	sp, {r0, lr}		@ save r0, lr
901	mrs	lr, spsr
902	str	lr, [sp, #8]		@ save spsr
903
904	@
905	@ Prepare for SVC32 mode.  IRQs remain disabled.
906	@
907	mrs	r0, cpsr
908	eor	r0, r0, #(\mode ^ SVC_MODE)
909	msr	spsr_cxsf, r0
910
911	@
912	@ the branch table must immediately follow this code
913	@
914	and	lr, lr, #0x0f
915	mov	r0, sp
916	ldr	lr, [pc, lr, lsl #2]
917	movs	pc, lr			@ branch to handler in SVC mode
918	.endm
919
920	.globl	__stubs_start
921__stubs_start:
922/*
923 * Interrupt dispatcher
924 */
925	vector_stub	irq, IRQ_MODE, 4
926
927	.long	__irq_usr			@  0  (USR_26 / USR_32)
928	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
929	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
930	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
931	.long	__irq_invalid			@  4
932	.long	__irq_invalid			@  5
933	.long	__irq_invalid			@  6
934	.long	__irq_invalid			@  7
935	.long	__irq_invalid			@  8
936	.long	__irq_invalid			@  9
937	.long	__irq_invalid			@  a
938	.long	__irq_invalid			@  b
939	.long	__irq_invalid			@  c
940	.long	__irq_invalid			@  d
941	.long	__irq_invalid			@  e
942	.long	__irq_invalid			@  f
943
944/*
945 * Data abort dispatcher
946 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
947 */
948	vector_stub	dabt, ABT_MODE, 8
949
950	.long	__dabt_usr			@  0  (USR_26 / USR_32)
951	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
952	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
953	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
954	.long	__dabt_invalid			@  4
955	.long	__dabt_invalid			@  5
956	.long	__dabt_invalid			@  6
957	.long	__dabt_invalid			@  7
958	.long	__dabt_invalid			@  8
959	.long	__dabt_invalid			@  9
960	.long	__dabt_invalid			@  a
961	.long	__dabt_invalid			@  b
962	.long	__dabt_invalid			@  c
963	.long	__dabt_invalid			@  d
964	.long	__dabt_invalid			@  e
965	.long	__dabt_invalid			@  f
966
967/*
968 * Prefetch abort dispatcher
969 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
970 */
971	vector_stub	pabt, ABT_MODE, 4
972
973	.long	__pabt_usr			@  0 (USR_26 / USR_32)
974	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
975	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
976	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
977	.long	__pabt_invalid			@  4
978	.long	__pabt_invalid			@  5
979	.long	__pabt_invalid			@  6
980	.long	__pabt_invalid			@  7
981	.long	__pabt_invalid			@  8
982	.long	__pabt_invalid			@  9
983	.long	__pabt_invalid			@  a
984	.long	__pabt_invalid			@  b
985	.long	__pabt_invalid			@  c
986	.long	__pabt_invalid			@  d
987	.long	__pabt_invalid			@  e
988	.long	__pabt_invalid			@  f
989
990/*
991 * Undef instr entry dispatcher
992 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
993 */
994	vector_stub	und, UND_MODE
995
996	.long	__und_usr			@  0 (USR_26 / USR_32)
997	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
998	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
999	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1000	.long	__und_invalid			@  4
1001	.long	__und_invalid			@  5
1002	.long	__und_invalid			@  6
1003	.long	__und_invalid			@  7
1004	.long	__und_invalid			@  8
1005	.long	__und_invalid			@  9
1006	.long	__und_invalid			@  a
1007	.long	__und_invalid			@  b
1008	.long	__und_invalid			@  c
1009	.long	__und_invalid			@  d
1010	.long	__und_invalid			@  e
1011	.long	__und_invalid			@  f
1012
1013	.align	5
1014
1015/*=============================================================================
1016 * Undefined FIQs
1017 *-----------------------------------------------------------------------------
1018 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1019 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1020 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1021 * damage alert!  I don't think that we can execute any code in here in any
1022 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1023 * get out of that mode without clobbering one register.
1024 */
1025vector_fiq:
1026	disable_fiq
1027	subs	pc, lr, #4
1028
1029/*=============================================================================
1030 * Address exception handler
1031 *-----------------------------------------------------------------------------
1032 * These aren't too critical.
1033 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1034 */
1035
1036vector_addrexcptn:
1037	b	vector_addrexcptn
1038
1039/*
1040 * We group all the following data together to optimise
1041 * for CPUs with separate I & D caches.
1042 */
1043	.align	5
1044
1045.LCvswi:
1046	.word	vector_swi
1047
1048	.globl	__stubs_end
1049__stubs_end:
1050
1051	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1052
1053	.globl	__vectors_start
1054__vectors_start:
1055	swi	SYS_ERROR0
1056	b	vector_und + stubs_offset
1057	ldr	pc, .LCvswi + stubs_offset
1058	b	vector_pabt + stubs_offset
1059	b	vector_dabt + stubs_offset
1060	b	vector_addrexcptn + stubs_offset
1061	b	vector_irq + stubs_offset
1062	b	vector_fiq + stubs_offset
1063
1064	.globl	__vectors_end
1065__vectors_end:
1066
1067	.data
1068
1069	.globl	cr_alignment
1070	.globl	cr_no_alignment
1071cr_alignment:
1072	.space	4
1073cr_no_alignment:
1074	.space	4
1075