xref: /openbmc/linux/arch/arm/kernel/entry-armv.S (revision 87c2ce3b)
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 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 *
11 *  Low-level vector interface routines
12 *
13 *  Note:  there is a StrongARM bug in the STMIA rn, {regs}^ instruction that causes
14 *  it to save wrong values...  Be aware!
15 */
16#include <linux/config.h>
17
18#include <asm/memory.h>
19#include <asm/glue.h>
20#include <asm/vfpmacros.h>
21#include <asm/arch/entry-macro.S>
22
23#include "entry-header.S"
24
25/*
26 * Interrupt handling.  Preserves r7, r8, r9
27 */
28	.macro	irq_handler
291:	get_irqnr_and_base r0, r6, r5, lr
30	movne	r1, sp
31	@
32	@ routine called with r0 = irq number, r1 = struct pt_regs *
33	@
34	adrne	lr, 1b
35	bne	asm_do_IRQ
36
37#ifdef CONFIG_SMP
38	/*
39	 * XXX
40	 *
41	 * this macro assumes that irqstat (r6) and base (r5) are
42	 * preserved from get_irqnr_and_base above
43	 */
44	test_for_ipi r0, r6, r5, lr
45	movne	r0, sp
46	adrne	lr, 1b
47	bne	do_IPI
48
49#ifdef CONFIG_LOCAL_TIMERS
50	test_for_ltirq r0, r6, r5, lr
51	movne	r0, sp
52	adrne	lr, 1b
53	bne	do_local_timer
54#endif
55#endif
56
57	.endm
58
59/*
60 * Invalid mode handlers
61 */
62	.macro	inv_entry, reason
63	sub	sp, sp, #S_FRAME_SIZE
64	stmib	sp, {r1 - lr}
65	mov	r1, #\reason
66	.endm
67
68__pabt_invalid:
69	inv_entry BAD_PREFETCH
70	b	common_invalid
71
72__dabt_invalid:
73	inv_entry BAD_DATA
74	b	common_invalid
75
76__irq_invalid:
77	inv_entry BAD_IRQ
78	b	common_invalid
79
80__und_invalid:
81	inv_entry BAD_UNDEFINSTR
82
83	@
84	@ XXX fall through to common_invalid
85	@
86
87@
88@ common_invalid - generic code for failed exception (re-entrant version of handlers)
89@
90common_invalid:
91	zero_fp
92
93	ldmia	r0, {r4 - r6}
94	add	r0, sp, #S_PC		@ here for interlock avoidance
95	mov	r7, #-1			@  ""   ""    ""        ""
96	str	r4, [sp]		@ save preserved r0
97	stmia	r0, {r5 - r7}		@ lr_<exception>,
98					@ cpsr_<exception>, "old_r0"
99
100	mov	r0, sp
101	and	r2, r6, #0x1f
102	b	bad_mode
103
104/*
105 * SVC mode handlers
106 */
107	.macro	svc_entry
108	sub	sp, sp, #S_FRAME_SIZE
109	stmib	sp, {r1 - r12}
110
111	ldmia	r0, {r1 - r3}
112	add	r5, sp, #S_SP		@ here for interlock avoidance
113	mov	r4, #-1			@  ""  ""      ""       ""
114	add	r0, sp, #S_FRAME_SIZE   @  ""  ""      ""       ""
115	str	r1, [sp]		@ save the "real" r0 copied
116					@ from the exception stack
117
118	mov	r1, lr
119
120	@
121	@ We are now ready to fill in the remaining blanks on the stack:
122	@
123	@  r0 - sp_svc
124	@  r1 - lr_svc
125	@  r2 - lr_<exception>, already fixed up for correct return/restart
126	@  r3 - spsr_<exception>
127	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
128	@
129	stmia	r5, {r0 - r4}
130	.endm
131
132	.align	5
133__dabt_svc:
134	svc_entry
135
136	@
137	@ get ready to re-enable interrupts if appropriate
138	@
139	mrs	r9, cpsr
140	tst	r3, #PSR_I_BIT
141	biceq	r9, r9, #PSR_I_BIT
142
143	@
144	@ Call the processor-specific abort handler:
145	@
146	@  r2 - aborted context pc
147	@  r3 - aborted context cpsr
148	@
149	@ The abort handler must return the aborted address in r0, and
150	@ the fault status register in r1.  r9 must be preserved.
151	@
152#ifdef MULTI_ABORT
153	ldr	r4, .LCprocfns
154	mov	lr, pc
155	ldr	pc, [r4]
156#else
157	bl	CPU_ABORT_HANDLER
158#endif
159
160	@
161	@ set desired IRQ state, then call main handler
162	@
163	msr	cpsr_c, r9
164	mov	r2, sp
165	bl	do_DataAbort
166
167	@
168	@ IRQs off again before pulling preserved data off the stack
169	@
170	disable_irq
171
172	@
173	@ restore SPSR and restart the instruction
174	@
175	ldr	r0, [sp, #S_PSR]
176	msr	spsr_cxsf, r0
177	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
178
179	.align	5
180__irq_svc:
181	svc_entry
182
183#ifdef CONFIG_PREEMPT
184	get_thread_info tsk
185	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
186	add	r7, r8, #1			@ increment it
187	str	r7, [tsk, #TI_PREEMPT]
188#endif
189
190	irq_handler
191#ifdef CONFIG_PREEMPT
192	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
193	tst	r0, #_TIF_NEED_RESCHED
194	blne	svc_preempt
195preempt_return:
196	ldr	r0, [tsk, #TI_PREEMPT]		@ read preempt value
197	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
198	teq	r0, r7
199	strne	r0, [r0, -r0]			@ bug()
200#endif
201	ldr	r0, [sp, #S_PSR]		@ irqs are already disabled
202	msr	spsr_cxsf, r0
203	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
204
205	.ltorg
206
207#ifdef CONFIG_PREEMPT
208svc_preempt:
209	teq	r8, #0				@ was preempt count = 0
210	ldreq	r6, .LCirq_stat
211	movne	pc, lr				@ no
212	ldr	r0, [r6, #4]			@ local_irq_count
213	ldr	r1, [r6, #8]			@ local_bh_count
214	adds	r0, r0, r1
215	movne	pc, lr
216	mov	r7, #0				@ preempt_schedule_irq
217	str	r7, [tsk, #TI_PREEMPT]		@ expects preempt_count == 0
2181:	bl	preempt_schedule_irq		@ irq en/disable is done inside
219	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
220	tst	r0, #_TIF_NEED_RESCHED
221	beq	preempt_return			@ go again
222	b	1b
223#endif
224
225	.align	5
226__und_svc:
227	svc_entry
228
229	@
230	@ call emulation code, which returns using r9 if it has emulated
231	@ the instruction, or the more conventional lr if we are to treat
232	@ this as a real undefined instruction
233	@
234	@  r0 - instruction
235	@
236	ldr	r0, [r2, #-4]
237	adr	r9, 1f
238	bl	call_fpe
239
240	mov	r0, sp				@ struct pt_regs *regs
241	bl	do_undefinstr
242
243	@
244	@ IRQs off again before pulling preserved data off the stack
245	@
2461:	disable_irq
247
248	@
249	@ restore SPSR and restart the instruction
250	@
251	ldr	lr, [sp, #S_PSR]		@ Get SVC cpsr
252	msr	spsr_cxsf, lr
253	ldmia	sp, {r0 - pc}^			@ Restore SVC registers
254
255	.align	5
256__pabt_svc:
257	svc_entry
258
259	@
260	@ re-enable interrupts if appropriate
261	@
262	mrs	r9, cpsr
263	tst	r3, #PSR_I_BIT
264	biceq	r9, r9, #PSR_I_BIT
265	msr	cpsr_c, r9
266
267	@
268	@ set args, then call main handler
269	@
270	@  r0 - address of faulting instruction
271	@  r1 - pointer to registers on stack
272	@
273	mov	r0, r2				@ address (pc)
274	mov	r1, sp				@ regs
275	bl	do_PrefetchAbort		@ call abort handler
276
277	@
278	@ IRQs off again before pulling preserved data off the stack
279	@
280	disable_irq
281
282	@
283	@ restore SPSR and restart the instruction
284	@
285	ldr	r0, [sp, #S_PSR]
286	msr	spsr_cxsf, r0
287	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
288
289	.align	5
290.LCcralign:
291	.word	cr_alignment
292#ifdef MULTI_ABORT
293.LCprocfns:
294	.word	processor
295#endif
296.LCfp:
297	.word	fp_enter
298#ifdef CONFIG_PREEMPT
299.LCirq_stat:
300	.word	irq_stat
301#endif
302
303/*
304 * User mode handlers
305 */
306	.macro	usr_entry
307	sub	sp, sp, #S_FRAME_SIZE
308	stmib	sp, {r1 - r12}
309
310	ldmia	r0, {r1 - r3}
311	add	r0, sp, #S_PC		@ here for interlock avoidance
312	mov	r4, #-1			@  ""  ""     ""        ""
313
314	str	r1, [sp]		@ save the "real" r0 copied
315					@ from the exception stack
316
317#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
318	@ make sure our user space atomic helper is aborted
319	cmp	r2, #TASK_SIZE
320	bichs	r3, r3, #PSR_Z_BIT
321#endif
322
323	@
324	@ We are now ready to fill in the remaining blanks on the stack:
325	@
326	@  r2 - lr_<exception>, already fixed up for correct return/restart
327	@  r3 - spsr_<exception>
328	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
329	@
330	@ Also, separately save sp_usr and lr_usr
331	@
332	stmia	r0, {r2 - r4}
333	stmdb	r0, {sp, lr}^
334
335	@
336	@ Enable the alignment trap while in kernel mode
337	@
338	alignment_trap r0
339
340	@
341	@ Clear FP to mark the first stack frame
342	@
343	zero_fp
344	.endm
345
346	.align	5
347__dabt_usr:
348	usr_entry
349
350	@
351	@ Call the processor-specific abort handler:
352	@
353	@  r2 - aborted context pc
354	@  r3 - aborted context cpsr
355	@
356	@ The abort handler must return the aborted address in r0, and
357	@ the fault status register in r1.
358	@
359#ifdef MULTI_ABORT
360	ldr	r4, .LCprocfns
361	mov	lr, pc
362	ldr	pc, [r4]
363#else
364	bl	CPU_ABORT_HANDLER
365#endif
366
367	@
368	@ IRQs on, then call the main handler
369	@
370	enable_irq
371	mov	r2, sp
372	adr	lr, ret_from_exception
373	b	do_DataAbort
374
375	.align	5
376__irq_usr:
377	usr_entry
378
379	get_thread_info tsk
380#ifdef CONFIG_PREEMPT
381	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
382	add	r7, r8, #1			@ increment it
383	str	r7, [tsk, #TI_PREEMPT]
384#endif
385
386	irq_handler
387#ifdef CONFIG_PREEMPT
388	ldr	r0, [tsk, #TI_PREEMPT]
389	str	r8, [tsk, #TI_PREEMPT]
390	teq	r0, r7
391	strne	r0, [r0, -r0]
392#endif
393
394	mov	why, #0
395	b	ret_to_user
396
397	.ltorg
398
399	.align	5
400__und_usr:
401	usr_entry
402
403	tst	r3, #PSR_T_BIT			@ Thumb mode?
404	bne	fpundefinstr			@ ignore FP
405	sub	r4, r2, #4
406
407	@
408	@ fall through to the emulation code, which returns using r9 if
409	@ it has emulated the instruction, or the more conventional lr
410	@ if we are to treat this as a real undefined instruction
411	@
412	@  r0 - instruction
413	@
4141:	ldrt	r0, [r4]
415	adr	r9, ret_from_exception
416	adr	lr, fpundefinstr
417	@
418	@ fallthrough to call_fpe
419	@
420
421/*
422 * The out of line fixup for the ldrt above.
423 */
424	.section .fixup, "ax"
4252:	mov	pc, r9
426	.previous
427	.section __ex_table,"a"
428	.long	1b, 2b
429	.previous
430
431/*
432 * Check whether the instruction is a co-processor instruction.
433 * If yes, we need to call the relevant co-processor handler.
434 *
435 * Note that we don't do a full check here for the co-processor
436 * instructions; all instructions with bit 27 set are well
437 * defined.  The only instructions that should fault are the
438 * co-processor instructions.  However, we have to watch out
439 * for the ARM6/ARM7 SWI bug.
440 *
441 * Emulators may wish to make use of the following registers:
442 *  r0  = instruction opcode.
443 *  r2  = PC+4
444 *  r10 = this threads thread_info structure.
445 */
446call_fpe:
447	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
448#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
449	and	r8, r0, #0x0f000000		@ mask out op-code bits
450	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
451#endif
452	moveq	pc, lr
453	get_thread_info r10			@ get current thread
454	and	r8, r0, #0x00000f00		@ mask out CP number
455	mov	r7, #1
456	add	r6, r10, #TI_USED_CP
457	strb	r7, [r6, r8, lsr #8]		@ set appropriate used_cp[]
458#ifdef CONFIG_IWMMXT
459	@ Test if we need to give access to iWMMXt coprocessors
460	ldr	r5, [r10, #TI_FLAGS]
461	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
462	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
463	bcs	iwmmxt_task_enable
464#endif
465	enable_irq
466	add	pc, pc, r8, lsr #6
467	mov	r0, r0
468
469	mov	pc, lr				@ CP#0
470	b	do_fpe				@ CP#1 (FPE)
471	b	do_fpe				@ CP#2 (FPE)
472	mov	pc, lr				@ CP#3
473	mov	pc, lr				@ CP#4
474	mov	pc, lr				@ CP#5
475	mov	pc, lr				@ CP#6
476	mov	pc, lr				@ CP#7
477	mov	pc, lr				@ CP#8
478	mov	pc, lr				@ CP#9
479#ifdef CONFIG_VFP
480	b	do_vfp				@ CP#10 (VFP)
481	b	do_vfp				@ CP#11 (VFP)
482#else
483	mov	pc, lr				@ CP#10 (VFP)
484	mov	pc, lr				@ CP#11 (VFP)
485#endif
486	mov	pc, lr				@ CP#12
487	mov	pc, lr				@ CP#13
488	mov	pc, lr				@ CP#14 (Debug)
489	mov	pc, lr				@ CP#15 (Control)
490
491do_fpe:
492	ldr	r4, .LCfp
493	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
494	ldr	pc, [r4]			@ Call FP module USR entry point
495
496/*
497 * The FP module is called with these registers set:
498 *  r0  = instruction
499 *  r2  = PC+4
500 *  r9  = normal "successful" return address
501 *  r10 = FP workspace
502 *  lr  = unrecognised FP instruction return address
503 */
504
505	.data
506ENTRY(fp_enter)
507	.word	fpundefinstr
508	.text
509
510fpundefinstr:
511	mov	r0, sp
512	adr	lr, ret_from_exception
513	b	do_undefinstr
514
515	.align	5
516__pabt_usr:
517	usr_entry
518
519	enable_irq				@ Enable interrupts
520	mov	r0, r2				@ address (pc)
521	mov	r1, sp				@ regs
522	bl	do_PrefetchAbort		@ call abort handler
523	/* fall through */
524/*
525 * This is the return code to user mode for abort handlers
526 */
527ENTRY(ret_from_exception)
528	get_thread_info tsk
529	mov	why, #0
530	b	ret_to_user
531
532/*
533 * Register switch for ARMv3 and ARMv4 processors
534 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
535 * previous and next are guaranteed not to be the same.
536 */
537ENTRY(__switch_to)
538	add	ip, r1, #TI_CPU_SAVE
539	ldr	r3, [r2, #TI_TP_VALUE]
540	stmia	ip!, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
541	ldr	r6, [r2, #TI_CPU_DOMAIN]!
542#if __LINUX_ARM_ARCH__ >= 6
543#ifdef CONFIG_CPU_MPCORE
544	clrex
545#else
546	strex	r5, r4, [ip]			@ Clear exclusive monitor
547#endif
548#endif
549#if defined(CONFIG_CPU_XSCALE) && !defined(CONFIG_IWMMXT)
550	mra	r4, r5, acc0
551	stmia   ip, {r4, r5}
552#endif
553#if defined(CONFIG_HAS_TLS_REG)
554	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
555#elif !defined(CONFIG_TLS_REG_EMUL)
556	mov	r4, #0xffff0fff
557	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
558#endif
559	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
560#ifdef CONFIG_VFP
561	@ Always disable VFP so we can lazily save/restore the old
562	@ state. This occurs in the context of the previous thread.
563	VFPFMRX	r4, FPEXC
564	bic	r4, r4, #FPEXC_ENABLE
565	VFPFMXR	FPEXC, r4
566#endif
567#if defined(CONFIG_IWMMXT)
568	bl	iwmmxt_task_switch
569#elif defined(CONFIG_CPU_XSCALE)
570	add	r4, r2, #40			@ cpu_context_save->extra
571	ldmib	r4, {r4, r5}
572	mar	acc0, r4, r5
573#endif
574	ldmib	r2, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
575
576	__INIT
577
578/*
579 * User helpers.
580 *
581 * These are segment of kernel provided user code reachable from user space
582 * at a fixed address in kernel memory.  This is used to provide user space
583 * with some operations which require kernel help because of unimplemented
584 * native feature and/or instructions in many ARM CPUs. The idea is for
585 * this code to be executed directly in user mode for best efficiency but
586 * which is too intimate with the kernel counter part to be left to user
587 * libraries.  In fact this code might even differ from one CPU to another
588 * depending on the available  instruction set and restrictions like on
589 * SMP systems.  In other words, the kernel reserves the right to change
590 * this code as needed without warning. Only the entry points and their
591 * results are guaranteed to be stable.
592 *
593 * Each segment is 32-byte aligned and will be moved to the top of the high
594 * vector page.  New segments (if ever needed) must be added in front of
595 * existing ones.  This mechanism should be used only for things that are
596 * really small and justified, and not be abused freely.
597 *
598 * User space is expected to implement those things inline when optimizing
599 * for a processor that has the necessary native support, but only if such
600 * resulting binaries are already to be incompatible with earlier ARM
601 * processors due to the use of unsupported instructions other than what
602 * is provided here.  In other words don't make binaries unable to run on
603 * earlier processors just for the sake of not using these kernel helpers
604 * if your compiled code is not going to use the new instructions for other
605 * purpose.
606 */
607
608	.align	5
609	.globl	__kuser_helper_start
610__kuser_helper_start:
611
612/*
613 * Reference prototype:
614 *
615 *	void __kernel_memory_barrier(void)
616 *
617 * Input:
618 *
619 *	lr = return address
620 *
621 * Output:
622 *
623 *	none
624 *
625 * Clobbered:
626 *
627 *	the Z flag might be lost
628 *
629 * Definition and user space usage example:
630 *
631 *	typedef void (__kernel_dmb_t)(void);
632 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
633 *
634 * Apply any needed memory barrier to preserve consistency with data modified
635 * manually and __kuser_cmpxchg usage.
636 *
637 * This could be used as follows:
638 *
639 * #define __kernel_dmb() \
640 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
641 *	        : : : "lr","cc" )
642 */
643
644__kuser_memory_barrier:				@ 0xffff0fa0
645
646#if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
647	mcr	p15, 0, r0, c7, c10, 5	@ dmb
648#endif
649	mov	pc, lr
650
651	.align	5
652
653/*
654 * Reference prototype:
655 *
656 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
657 *
658 * Input:
659 *
660 *	r0 = oldval
661 *	r1 = newval
662 *	r2 = ptr
663 *	lr = return address
664 *
665 * Output:
666 *
667 *	r0 = returned value (zero or non-zero)
668 *	C flag = set if r0 == 0, clear if r0 != 0
669 *
670 * Clobbered:
671 *
672 *	r3, ip, flags
673 *
674 * Definition and user space usage example:
675 *
676 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
677 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
678 *
679 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
680 * Return zero if *ptr was changed or non-zero if no exchange happened.
681 * The C flag is also set if *ptr was changed to allow for assembly
682 * optimization in the calling code.
683 *
684 * Note: this routine already includes memory barriers as needed.
685 *
686 * For example, a user space atomic_add implementation could look like this:
687 *
688 * #define atomic_add(ptr, val) \
689 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
690 *	   register unsigned int __result asm("r1"); \
691 *	   asm volatile ( \
692 *	       "1: @ atomic_add\n\t" \
693 *	       "ldr	r0, [r2]\n\t" \
694 *	       "mov	r3, #0xffff0fff\n\t" \
695 *	       "add	lr, pc, #4\n\t" \
696 *	       "add	r1, r0, %2\n\t" \
697 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
698 *	       "bcc	1b" \
699 *	       : "=&r" (__result) \
700 *	       : "r" (__ptr), "rIL" (val) \
701 *	       : "r0","r3","ip","lr","cc","memory" ); \
702 *	   __result; })
703 */
704
705__kuser_cmpxchg:				@ 0xffff0fc0
706
707#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
708
709	/*
710	 * Poor you.  No fast solution possible...
711	 * The kernel itself must perform the operation.
712	 * A special ghost syscall is used for that (see traps.c).
713	 */
714	swi	#0x9ffff0
715	mov	pc, lr
716
717#elif __LINUX_ARM_ARCH__ < 6
718
719	/*
720	 * Theory of operation:
721	 *
722	 * We set the Z flag before loading oldval. If ever an exception
723	 * occurs we can not be sure the loaded value will still be the same
724	 * when the exception returns, therefore the user exception handler
725	 * will clear the Z flag whenever the interrupted user code was
726	 * actually from the kernel address space (see the usr_entry macro).
727	 *
728	 * The post-increment on the str is used to prevent a race with an
729	 * exception happening just after the str instruction which would
730	 * clear the Z flag although the exchange was done.
731	 */
732	teq	ip, ip			@ set Z flag
733	ldr	ip, [r2]		@ load current val
734	add	r3, r2, #1		@ prepare store ptr
735	teqeq	ip, r0			@ compare with oldval if still allowed
736	streq	r1, [r3, #-1]!		@ store newval if still allowed
737	subs	r0, r2, r3		@ if r2 == r3 the str occured
738	mov	pc, lr
739
740#else
741
742#ifdef CONFIG_SMP
743	mcr	p15, 0, r0, c7, c10, 5	@ dmb
744#endif
745	ldrex	r3, [r2]
746	subs	r3, r3, r0
747	strexeq	r3, r1, [r2]
748	rsbs	r0, r3, #0
749#ifdef CONFIG_SMP
750	mcr	p15, 0, r0, c7, c10, 5	@ dmb
751#endif
752	mov	pc, lr
753
754#endif
755
756	.align	5
757
758/*
759 * Reference prototype:
760 *
761 *	int __kernel_get_tls(void)
762 *
763 * Input:
764 *
765 *	lr = return address
766 *
767 * Output:
768 *
769 *	r0 = TLS value
770 *
771 * Clobbered:
772 *
773 *	the Z flag might be lost
774 *
775 * Definition and user space usage example:
776 *
777 *	typedef int (__kernel_get_tls_t)(void);
778 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
779 *
780 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
781 *
782 * This could be used as follows:
783 *
784 * #define __kernel_get_tls() \
785 *	({ register unsigned int __val asm("r0"); \
786 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
787 *	        : "=r" (__val) : : "lr","cc" ); \
788 *	   __val; })
789 */
790
791__kuser_get_tls:				@ 0xffff0fe0
792
793#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
794
795	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
796	mov	pc, lr
797
798#else
799
800	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
801	mov	pc, lr
802
803#endif
804
805	.rep	5
806	.word	0			@ pad up to __kuser_helper_version
807	.endr
808
809/*
810 * Reference declaration:
811 *
812 *	extern unsigned int __kernel_helper_version;
813 *
814 * Definition and user space usage example:
815 *
816 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
817 *
818 * User space may read this to determine the curent number of helpers
819 * available.
820 */
821
822__kuser_helper_version:				@ 0xffff0ffc
823	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
824
825	.globl	__kuser_helper_end
826__kuser_helper_end:
827
828
829/*
830 * Vector stubs.
831 *
832 * This code is copied to 0xffff0200 so we can use branches in the
833 * vectors, rather than ldr's.  Note that this code must not
834 * exceed 0x300 bytes.
835 *
836 * Common stub entry macro:
837 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
838 *
839 * SP points to a minimal amount of processor-private memory, the address
840 * of which is copied into r0 for the mode specific abort handler.
841 */
842	.macro	vector_stub, name, mode, correction=0
843	.align	5
844
845vector_\name:
846	.if \correction
847	sub	lr, lr, #\correction
848	.endif
849
850	@
851	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
852	@ (parent CPSR)
853	@
854	stmia	sp, {r0, lr}		@ save r0, lr
855	mrs	lr, spsr
856	str	lr, [sp, #8]		@ save spsr
857
858	@
859	@ Prepare for SVC32 mode.  IRQs remain disabled.
860	@
861	mrs	r0, cpsr
862	eor	r0, r0, #(\mode ^ SVC_MODE)
863	msr	spsr_cxsf, r0
864
865	@
866	@ the branch table must immediately follow this code
867	@
868	and	lr, lr, #0x0f
869	mov	r0, sp
870	ldr	lr, [pc, lr, lsl #2]
871	movs	pc, lr			@ branch to handler in SVC mode
872	.endm
873
874	.globl	__stubs_start
875__stubs_start:
876/*
877 * Interrupt dispatcher
878 */
879	vector_stub	irq, IRQ_MODE, 4
880
881	.long	__irq_usr			@  0  (USR_26 / USR_32)
882	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
883	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
884	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
885	.long	__irq_invalid			@  4
886	.long	__irq_invalid			@  5
887	.long	__irq_invalid			@  6
888	.long	__irq_invalid			@  7
889	.long	__irq_invalid			@  8
890	.long	__irq_invalid			@  9
891	.long	__irq_invalid			@  a
892	.long	__irq_invalid			@  b
893	.long	__irq_invalid			@  c
894	.long	__irq_invalid			@  d
895	.long	__irq_invalid			@  e
896	.long	__irq_invalid			@  f
897
898/*
899 * Data abort dispatcher
900 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
901 */
902	vector_stub	dabt, ABT_MODE, 8
903
904	.long	__dabt_usr			@  0  (USR_26 / USR_32)
905	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
906	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
907	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
908	.long	__dabt_invalid			@  4
909	.long	__dabt_invalid			@  5
910	.long	__dabt_invalid			@  6
911	.long	__dabt_invalid			@  7
912	.long	__dabt_invalid			@  8
913	.long	__dabt_invalid			@  9
914	.long	__dabt_invalid			@  a
915	.long	__dabt_invalid			@  b
916	.long	__dabt_invalid			@  c
917	.long	__dabt_invalid			@  d
918	.long	__dabt_invalid			@  e
919	.long	__dabt_invalid			@  f
920
921/*
922 * Prefetch abort dispatcher
923 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
924 */
925	vector_stub	pabt, ABT_MODE, 4
926
927	.long	__pabt_usr			@  0 (USR_26 / USR_32)
928	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
929	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
930	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
931	.long	__pabt_invalid			@  4
932	.long	__pabt_invalid			@  5
933	.long	__pabt_invalid			@  6
934	.long	__pabt_invalid			@  7
935	.long	__pabt_invalid			@  8
936	.long	__pabt_invalid			@  9
937	.long	__pabt_invalid			@  a
938	.long	__pabt_invalid			@  b
939	.long	__pabt_invalid			@  c
940	.long	__pabt_invalid			@  d
941	.long	__pabt_invalid			@  e
942	.long	__pabt_invalid			@  f
943
944/*
945 * Undef instr entry dispatcher
946 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
947 */
948	vector_stub	und, UND_MODE
949
950	.long	__und_usr			@  0 (USR_26 / USR_32)
951	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
952	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
953	.long	__und_svc			@  3 (SVC_26 / SVC_32)
954	.long	__und_invalid			@  4
955	.long	__und_invalid			@  5
956	.long	__und_invalid			@  6
957	.long	__und_invalid			@  7
958	.long	__und_invalid			@  8
959	.long	__und_invalid			@  9
960	.long	__und_invalid			@  a
961	.long	__und_invalid			@  b
962	.long	__und_invalid			@  c
963	.long	__und_invalid			@  d
964	.long	__und_invalid			@  e
965	.long	__und_invalid			@  f
966
967	.align	5
968
969/*=============================================================================
970 * Undefined FIQs
971 *-----------------------------------------------------------------------------
972 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
973 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
974 * Basically to switch modes, we *HAVE* to clobber one register...  brain
975 * damage alert!  I don't think that we can execute any code in here in any
976 * other mode than FIQ...  Ok you can switch to another mode, but you can't
977 * get out of that mode without clobbering one register.
978 */
979vector_fiq:
980	disable_fiq
981	subs	pc, lr, #4
982
983/*=============================================================================
984 * Address exception handler
985 *-----------------------------------------------------------------------------
986 * These aren't too critical.
987 * (they're not supposed to happen, and won't happen in 32-bit data mode).
988 */
989
990vector_addrexcptn:
991	b	vector_addrexcptn
992
993/*
994 * We group all the following data together to optimise
995 * for CPUs with separate I & D caches.
996 */
997	.align	5
998
999.LCvswi:
1000	.word	vector_swi
1001
1002	.globl	__stubs_end
1003__stubs_end:
1004
1005	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1006
1007	.globl	__vectors_start
1008__vectors_start:
1009	swi	SYS_ERROR0
1010	b	vector_und + stubs_offset
1011	ldr	pc, .LCvswi + stubs_offset
1012	b	vector_pabt + stubs_offset
1013	b	vector_dabt + stubs_offset
1014	b	vector_addrexcptn + stubs_offset
1015	b	vector_irq + stubs_offset
1016	b	vector_fiq + stubs_offset
1017
1018	.globl	__vectors_end
1019__vectors_end:
1020
1021	.data
1022
1023	.globl	cr_alignment
1024	.globl	cr_no_alignment
1025cr_alignment:
1026	.space	4
1027cr_no_alignment:
1028	.space	4
1029