xref: /openbmc/linux/arch/arm/kernel/entry-armv.S (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
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#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
343#ifndef CONFIG_MMU
344#warning "NPTL on non MMU needs fixing"
345#else
346	@ make sure our user space atomic helper is aborted
347	cmp	r2, #TASK_SIZE
348	bichs	r3, r3, #PSR_Z_BIT
349#endif
350#endif
351
352	@
353	@ We are now ready to fill in the remaining blanks on the stack:
354	@
355	@  r2 - lr_<exception>, already fixed up for correct return/restart
356	@  r3 - spsr_<exception>
357	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
358	@
359	@ Also, separately save sp_usr and lr_usr
360	@
361	stmia	r0, {r2 - r4}
362	stmdb	r0, {sp, lr}^
363
364	@
365	@ Enable the alignment trap while in kernel mode
366	@
367	alignment_trap r0
368
369	@
370	@ Clear FP to mark the first stack frame
371	@
372	zero_fp
373	.endm
374
375	.align	5
376__dabt_usr:
377	usr_entry
378
379	@
380	@ Call the processor-specific abort handler:
381	@
382	@  r2 - aborted context pc
383	@  r3 - aborted context cpsr
384	@
385	@ The abort handler must return the aborted address in r0, and
386	@ the fault status register in r1.
387	@
388#ifdef MULTI_ABORT
389	ldr	r4, .LCprocfns
390	mov	lr, pc
391	ldr	pc, [r4]
392#else
393	bl	CPU_ABORT_HANDLER
394#endif
395
396	@
397	@ IRQs on, then call the main handler
398	@
399	enable_irq
400	mov	r2, sp
401	adr	lr, ret_from_exception
402	b	do_DataAbort
403
404	.align	5
405__irq_usr:
406	usr_entry
407
408#ifdef CONFIG_TRACE_IRQFLAGS
409	bl	trace_hardirqs_off
410#endif
411	get_thread_info tsk
412#ifdef CONFIG_PREEMPT
413	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
414	add	r7, r8, #1			@ increment it
415	str	r7, [tsk, #TI_PREEMPT]
416#endif
417
418	irq_handler
419#ifdef CONFIG_PREEMPT
420	ldr	r0, [tsk, #TI_PREEMPT]
421	str	r8, [tsk, #TI_PREEMPT]
422	teq	r0, r7
423	strne	r0, [r0, -r0]
424#endif
425#ifdef CONFIG_TRACE_IRQFLAGS
426	bl	trace_hardirqs_on
427#endif
428
429	mov	why, #0
430	b	ret_to_user
431
432	.ltorg
433
434	.align	5
435__und_usr:
436	usr_entry
437
438	tst	r3, #PSR_T_BIT			@ Thumb mode?
439	bne	__und_usr_unknown		@ ignore FP
440	sub	r4, r2, #4
441
442	@
443	@ fall through to the emulation code, which returns using r9 if
444	@ it has emulated the instruction, or the more conventional lr
445	@ if we are to treat this as a real undefined instruction
446	@
447	@  r0 - instruction
448	@
4491:	ldrt	r0, [r4]
450	adr	r9, ret_from_exception
451	adr	lr, __und_usr_unknown
452	@
453	@ fallthrough to call_fpe
454	@
455
456/*
457 * The out of line fixup for the ldrt above.
458 */
459	.section .fixup, "ax"
4602:	mov	pc, r9
461	.previous
462	.section __ex_table,"a"
463	.long	1b, 2b
464	.previous
465
466/*
467 * Check whether the instruction is a co-processor instruction.
468 * If yes, we need to call the relevant co-processor handler.
469 *
470 * Note that we don't do a full check here for the co-processor
471 * instructions; all instructions with bit 27 set are well
472 * defined.  The only instructions that should fault are the
473 * co-processor instructions.  However, we have to watch out
474 * for the ARM6/ARM7 SWI bug.
475 *
476 * Emulators may wish to make use of the following registers:
477 *  r0  = instruction opcode.
478 *  r2  = PC+4
479 *  r9  = normal "successful" return address
480 *  r10 = this threads thread_info structure.
481 *  lr  = unrecognised instruction return address
482 */
483call_fpe:
484	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
485#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
486	and	r8, r0, #0x0f000000		@ mask out op-code bits
487	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
488#endif
489	moveq	pc, lr
490	get_thread_info r10			@ get current thread
491	and	r8, r0, #0x00000f00		@ mask out CP number
492	mov	r7, #1
493	add	r6, r10, #TI_USED_CP
494	strb	r7, [r6, r8, lsr #8]		@ set appropriate used_cp[]
495#ifdef CONFIG_IWMMXT
496	@ Test if we need to give access to iWMMXt coprocessors
497	ldr	r5, [r10, #TI_FLAGS]
498	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
499	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
500	bcs	iwmmxt_task_enable
501#endif
502	add	pc, pc, r8, lsr #6
503	mov	r0, r0
504
505	mov	pc, lr				@ CP#0
506	b	do_fpe				@ CP#1 (FPE)
507	b	do_fpe				@ CP#2 (FPE)
508	mov	pc, lr				@ CP#3
509#ifdef CONFIG_CRUNCH
510	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
511	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
512	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
513#else
514	mov	pc, lr				@ CP#4
515	mov	pc, lr				@ CP#5
516	mov	pc, lr				@ CP#6
517#endif
518	mov	pc, lr				@ CP#7
519	mov	pc, lr				@ CP#8
520	mov	pc, lr				@ CP#9
521#ifdef CONFIG_VFP
522	b	do_vfp				@ CP#10 (VFP)
523	b	do_vfp				@ CP#11 (VFP)
524#else
525	mov	pc, lr				@ CP#10 (VFP)
526	mov	pc, lr				@ CP#11 (VFP)
527#endif
528	mov	pc, lr				@ CP#12
529	mov	pc, lr				@ CP#13
530	mov	pc, lr				@ CP#14 (Debug)
531	mov	pc, lr				@ CP#15 (Control)
532
533do_fpe:
534	enable_irq
535	ldr	r4, .LCfp
536	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
537	ldr	pc, [r4]			@ Call FP module USR entry point
538
539/*
540 * The FP module is called with these registers set:
541 *  r0  = instruction
542 *  r2  = PC+4
543 *  r9  = normal "successful" return address
544 *  r10 = FP workspace
545 *  lr  = unrecognised FP instruction return address
546 */
547
548	.data
549ENTRY(fp_enter)
550	.word	no_fp
551	.text
552
553no_fp:	mov	pc, lr
554
555__und_usr_unknown:
556	mov	r0, sp
557	adr	lr, ret_from_exception
558	b	do_undefinstr
559
560	.align	5
561__pabt_usr:
562	usr_entry
563
564	enable_irq				@ Enable interrupts
565	mov	r0, r2				@ address (pc)
566	mov	r1, sp				@ regs
567	bl	do_PrefetchAbort		@ call abort handler
568	/* fall through */
569/*
570 * This is the return code to user mode for abort handlers
571 */
572ENTRY(ret_from_exception)
573	get_thread_info tsk
574	mov	why, #0
575	b	ret_to_user
576
577/*
578 * Register switch for ARMv3 and ARMv4 processors
579 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
580 * previous and next are guaranteed not to be the same.
581 */
582ENTRY(__switch_to)
583	add	ip, r1, #TI_CPU_SAVE
584	ldr	r3, [r2, #TI_TP_VALUE]
585	stmia	ip!, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
586#ifdef CONFIG_MMU
587	ldr	r6, [r2, #TI_CPU_DOMAIN]
588#endif
589#if __LINUX_ARM_ARCH__ >= 6
590#ifdef CONFIG_CPU_32v6K
591	clrex
592#else
593	strex	r5, r4, [ip]			@ Clear exclusive monitor
594#endif
595#endif
596#if defined(CONFIG_HAS_TLS_REG)
597	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
598#elif !defined(CONFIG_TLS_REG_EMUL)
599	mov	r4, #0xffff0fff
600	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
601#endif
602#ifdef CONFIG_MMU
603	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
604#endif
605	mov	r5, r0
606	add	r4, r2, #TI_CPU_SAVE
607	ldr	r0, =thread_notify_head
608	mov	r1, #THREAD_NOTIFY_SWITCH
609	bl	atomic_notifier_call_chain
610	mov	r0, r5
611	ldmia	r4, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
612
613	__INIT
614
615/*
616 * User helpers.
617 *
618 * These are segment of kernel provided user code reachable from user space
619 * at a fixed address in kernel memory.  This is used to provide user space
620 * with some operations which require kernel help because of unimplemented
621 * native feature and/or instructions in many ARM CPUs. The idea is for
622 * this code to be executed directly in user mode for best efficiency but
623 * which is too intimate with the kernel counter part to be left to user
624 * libraries.  In fact this code might even differ from one CPU to another
625 * depending on the available  instruction set and restrictions like on
626 * SMP systems.  In other words, the kernel reserves the right to change
627 * this code as needed without warning. Only the entry points and their
628 * results are guaranteed to be stable.
629 *
630 * Each segment is 32-byte aligned and will be moved to the top of the high
631 * vector page.  New segments (if ever needed) must be added in front of
632 * existing ones.  This mechanism should be used only for things that are
633 * really small and justified, and not be abused freely.
634 *
635 * User space is expected to implement those things inline when optimizing
636 * for a processor that has the necessary native support, but only if such
637 * resulting binaries are already to be incompatible with earlier ARM
638 * processors due to the use of unsupported instructions other than what
639 * is provided here.  In other words don't make binaries unable to run on
640 * earlier processors just for the sake of not using these kernel helpers
641 * if your compiled code is not going to use the new instructions for other
642 * purpose.
643 */
644
645	.macro	usr_ret, reg
646#ifdef CONFIG_ARM_THUMB
647	bx	\reg
648#else
649	mov	pc, \reg
650#endif
651	.endm
652
653	.align	5
654	.globl	__kuser_helper_start
655__kuser_helper_start:
656
657/*
658 * Reference prototype:
659 *
660 *	void __kernel_memory_barrier(void)
661 *
662 * Input:
663 *
664 *	lr = return address
665 *
666 * Output:
667 *
668 *	none
669 *
670 * Clobbered:
671 *
672 *	the Z flag might be lost
673 *
674 * Definition and user space usage example:
675 *
676 *	typedef void (__kernel_dmb_t)(void);
677 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
678 *
679 * Apply any needed memory barrier to preserve consistency with data modified
680 * manually and __kuser_cmpxchg usage.
681 *
682 * This could be used as follows:
683 *
684 * #define __kernel_dmb() \
685 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
686 *	        : : : "r0", "lr","cc" )
687 */
688
689__kuser_memory_barrier:				@ 0xffff0fa0
690
691#if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
692	mcr	p15, 0, r0, c7, c10, 5	@ dmb
693#endif
694	usr_ret	lr
695
696	.align	5
697
698/*
699 * Reference prototype:
700 *
701 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
702 *
703 * Input:
704 *
705 *	r0 = oldval
706 *	r1 = newval
707 *	r2 = ptr
708 *	lr = return address
709 *
710 * Output:
711 *
712 *	r0 = returned value (zero or non-zero)
713 *	C flag = set if r0 == 0, clear if r0 != 0
714 *
715 * Clobbered:
716 *
717 *	r3, ip, flags
718 *
719 * Definition and user space usage example:
720 *
721 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
722 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
723 *
724 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
725 * Return zero if *ptr was changed or non-zero if no exchange happened.
726 * The C flag is also set if *ptr was changed to allow for assembly
727 * optimization in the calling code.
728 *
729 * Notes:
730 *
731 *    - This routine already includes memory barriers as needed.
732 *
733 *    - A failure might be transient, i.e. it is possible, although unlikely,
734 *      that "failure" be returned even if *ptr == oldval.
735 *
736 * For example, a user space atomic_add implementation could look like this:
737 *
738 * #define atomic_add(ptr, val) \
739 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
740 *	   register unsigned int __result asm("r1"); \
741 *	   asm volatile ( \
742 *	       "1: @ atomic_add\n\t" \
743 *	       "ldr	r0, [r2]\n\t" \
744 *	       "mov	r3, #0xffff0fff\n\t" \
745 *	       "add	lr, pc, #4\n\t" \
746 *	       "add	r1, r0, %2\n\t" \
747 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
748 *	       "bcc	1b" \
749 *	       : "=&r" (__result) \
750 *	       : "r" (__ptr), "rIL" (val) \
751 *	       : "r0","r3","ip","lr","cc","memory" ); \
752 *	   __result; })
753 */
754
755__kuser_cmpxchg:				@ 0xffff0fc0
756
757#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
758
759	/*
760	 * Poor you.  No fast solution possible...
761	 * The kernel itself must perform the operation.
762	 * A special ghost syscall is used for that (see traps.c).
763	 */
764	stmfd	sp!, {r7, lr}
765	mov	r7, #0xff00		@ 0xfff0 into r7 for EABI
766	orr	r7, r7, #0xf0
767	swi	#0x9ffff0
768	ldmfd	sp!, {r7, pc}
769
770#elif __LINUX_ARM_ARCH__ < 6
771
772	/*
773	 * Theory of operation:
774	 *
775	 * We set the Z flag before loading oldval. If ever an exception
776	 * occurs we can not be sure the loaded value will still be the same
777	 * when the exception returns, therefore the user exception handler
778	 * will clear the Z flag whenever the interrupted user code was
779	 * actually from the kernel address space (see the usr_entry macro).
780	 *
781	 * The post-increment on the str is used to prevent a race with an
782	 * exception happening just after the str instruction which would
783	 * clear the Z flag although the exchange was done.
784	 */
785#ifdef CONFIG_MMU
786	teq	ip, ip			@ set Z flag
787	ldr	ip, [r2]		@ load current val
788	add	r3, r2, #1		@ prepare store ptr
789	teqeq	ip, r0			@ compare with oldval if still allowed
790	streq	r1, [r3, #-1]!		@ store newval if still allowed
791	subs	r0, r2, r3		@ if r2 == r3 the str occured
792#else
793#warning "NPTL on non MMU needs fixing"
794	mov	r0, #-1
795	adds	r0, r0, #0
796#endif
797	usr_ret	lr
798
799#else
800
801#ifdef CONFIG_SMP
802	mcr	p15, 0, r0, c7, c10, 5	@ dmb
803#endif
804	ldrex	r3, [r2]
805	subs	r3, r3, r0
806	strexeq	r3, r1, [r2]
807	rsbs	r0, r3, #0
808#ifdef CONFIG_SMP
809	mcr	p15, 0, r0, c7, c10, 5	@ dmb
810#endif
811	usr_ret	lr
812
813#endif
814
815	.align	5
816
817/*
818 * Reference prototype:
819 *
820 *	int __kernel_get_tls(void)
821 *
822 * Input:
823 *
824 *	lr = return address
825 *
826 * Output:
827 *
828 *	r0 = TLS value
829 *
830 * Clobbered:
831 *
832 *	the Z flag might be lost
833 *
834 * Definition and user space usage example:
835 *
836 *	typedef int (__kernel_get_tls_t)(void);
837 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
838 *
839 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
840 *
841 * This could be used as follows:
842 *
843 * #define __kernel_get_tls() \
844 *	({ register unsigned int __val asm("r0"); \
845 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
846 *	        : "=r" (__val) : : "lr","cc" ); \
847 *	   __val; })
848 */
849
850__kuser_get_tls:				@ 0xffff0fe0
851
852#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
853	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
854#else
855	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
856#endif
857	usr_ret	lr
858
859	.rep	5
860	.word	0			@ pad up to __kuser_helper_version
861	.endr
862
863/*
864 * Reference declaration:
865 *
866 *	extern unsigned int __kernel_helper_version;
867 *
868 * Definition and user space usage example:
869 *
870 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
871 *
872 * User space may read this to determine the curent number of helpers
873 * available.
874 */
875
876__kuser_helper_version:				@ 0xffff0ffc
877	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
878
879	.globl	__kuser_helper_end
880__kuser_helper_end:
881
882
883/*
884 * Vector stubs.
885 *
886 * This code is copied to 0xffff0200 so we can use branches in the
887 * vectors, rather than ldr's.  Note that this code must not
888 * exceed 0x300 bytes.
889 *
890 * Common stub entry macro:
891 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
892 *
893 * SP points to a minimal amount of processor-private memory, the address
894 * of which is copied into r0 for the mode specific abort handler.
895 */
896	.macro	vector_stub, name, mode, correction=0
897	.align	5
898
899vector_\name:
900	.if \correction
901	sub	lr, lr, #\correction
902	.endif
903
904	@
905	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
906	@ (parent CPSR)
907	@
908	stmia	sp, {r0, lr}		@ save r0, lr
909	mrs	lr, spsr
910	str	lr, [sp, #8]		@ save spsr
911
912	@
913	@ Prepare for SVC32 mode.  IRQs remain disabled.
914	@
915	mrs	r0, cpsr
916	eor	r0, r0, #(\mode ^ SVC_MODE)
917	msr	spsr_cxsf, r0
918
919	@
920	@ the branch table must immediately follow this code
921	@
922	and	lr, lr, #0x0f
923	mov	r0, sp
924	ldr	lr, [pc, lr, lsl #2]
925	movs	pc, lr			@ branch to handler in SVC mode
926	.endm
927
928	.globl	__stubs_start
929__stubs_start:
930/*
931 * Interrupt dispatcher
932 */
933	vector_stub	irq, IRQ_MODE, 4
934
935	.long	__irq_usr			@  0  (USR_26 / USR_32)
936	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
937	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
938	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
939	.long	__irq_invalid			@  4
940	.long	__irq_invalid			@  5
941	.long	__irq_invalid			@  6
942	.long	__irq_invalid			@  7
943	.long	__irq_invalid			@  8
944	.long	__irq_invalid			@  9
945	.long	__irq_invalid			@  a
946	.long	__irq_invalid			@  b
947	.long	__irq_invalid			@  c
948	.long	__irq_invalid			@  d
949	.long	__irq_invalid			@  e
950	.long	__irq_invalid			@  f
951
952/*
953 * Data abort dispatcher
954 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
955 */
956	vector_stub	dabt, ABT_MODE, 8
957
958	.long	__dabt_usr			@  0  (USR_26 / USR_32)
959	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
960	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
961	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
962	.long	__dabt_invalid			@  4
963	.long	__dabt_invalid			@  5
964	.long	__dabt_invalid			@  6
965	.long	__dabt_invalid			@  7
966	.long	__dabt_invalid			@  8
967	.long	__dabt_invalid			@  9
968	.long	__dabt_invalid			@  a
969	.long	__dabt_invalid			@  b
970	.long	__dabt_invalid			@  c
971	.long	__dabt_invalid			@  d
972	.long	__dabt_invalid			@  e
973	.long	__dabt_invalid			@  f
974
975/*
976 * Prefetch abort dispatcher
977 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
978 */
979	vector_stub	pabt, ABT_MODE, 4
980
981	.long	__pabt_usr			@  0 (USR_26 / USR_32)
982	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
983	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
984	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
985	.long	__pabt_invalid			@  4
986	.long	__pabt_invalid			@  5
987	.long	__pabt_invalid			@  6
988	.long	__pabt_invalid			@  7
989	.long	__pabt_invalid			@  8
990	.long	__pabt_invalid			@  9
991	.long	__pabt_invalid			@  a
992	.long	__pabt_invalid			@  b
993	.long	__pabt_invalid			@  c
994	.long	__pabt_invalid			@  d
995	.long	__pabt_invalid			@  e
996	.long	__pabt_invalid			@  f
997
998/*
999 * Undef instr entry dispatcher
1000 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1001 */
1002	vector_stub	und, UND_MODE
1003
1004	.long	__und_usr			@  0 (USR_26 / USR_32)
1005	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1006	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1007	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1008	.long	__und_invalid			@  4
1009	.long	__und_invalid			@  5
1010	.long	__und_invalid			@  6
1011	.long	__und_invalid			@  7
1012	.long	__und_invalid			@  8
1013	.long	__und_invalid			@  9
1014	.long	__und_invalid			@  a
1015	.long	__und_invalid			@  b
1016	.long	__und_invalid			@  c
1017	.long	__und_invalid			@  d
1018	.long	__und_invalid			@  e
1019	.long	__und_invalid			@  f
1020
1021	.align	5
1022
1023/*=============================================================================
1024 * Undefined FIQs
1025 *-----------------------------------------------------------------------------
1026 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1027 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1028 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1029 * damage alert!  I don't think that we can execute any code in here in any
1030 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1031 * get out of that mode without clobbering one register.
1032 */
1033vector_fiq:
1034	disable_fiq
1035	subs	pc, lr, #4
1036
1037/*=============================================================================
1038 * Address exception handler
1039 *-----------------------------------------------------------------------------
1040 * These aren't too critical.
1041 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1042 */
1043
1044vector_addrexcptn:
1045	b	vector_addrexcptn
1046
1047/*
1048 * We group all the following data together to optimise
1049 * for CPUs with separate I & D caches.
1050 */
1051	.align	5
1052
1053.LCvswi:
1054	.word	vector_swi
1055
1056	.globl	__stubs_end
1057__stubs_end:
1058
1059	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1060
1061	.globl	__vectors_start
1062__vectors_start:
1063	swi	SYS_ERROR0
1064	b	vector_und + stubs_offset
1065	ldr	pc, .LCvswi + stubs_offset
1066	b	vector_pabt + stubs_offset
1067	b	vector_dabt + stubs_offset
1068	b	vector_addrexcptn + stubs_offset
1069	b	vector_irq + stubs_offset
1070	b	vector_fiq + stubs_offset
1071
1072	.globl	__vectors_end
1073__vectors_end:
1074
1075	.data
1076
1077	.globl	cr_alignment
1078	.globl	cr_no_alignment
1079cr_alignment:
1080	.space	4
1081cr_no_alignment:
1082	.space	4
1083