xref: /linux/arch/arm/kernel/entry-armv.S (revision 14b42963f64b98ab61fa9723c03d71aa5ef4f862)
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
301:	get_irqnr_and_base r0, r6, r5, lr
31	movne	r1, sp
32	@
33	@ routine called with r0 = irq number, r1 = struct pt_regs *
34	@
35	adrne	lr, 1b
36	bne	asm_do_IRQ
37
38#ifdef CONFIG_SMP
39	/*
40	 * XXX
41	 *
42	 * this macro assumes that irqstat (r6) and base (r5) are
43	 * preserved from get_irqnr_and_base above
44	 */
45	test_for_ipi r0, r6, r5, lr
46	movne	r0, sp
47	adrne	lr, 1b
48	bne	do_IPI
49
50#ifdef CONFIG_LOCAL_TIMERS
51	test_for_ltirq r0, r6, r5, lr
52	movne	r0, sp
53	adrne	lr, 1b
54	bne	do_local_timer
55#endif
56#endif
57
58	.endm
59
60/*
61 * Invalid mode handlers
62 */
63	.macro	inv_entry, reason
64	sub	sp, sp, #S_FRAME_SIZE
65	stmib	sp, {r1 - lr}
66	mov	r1, #\reason
67	.endm
68
69__pabt_invalid:
70	inv_entry BAD_PREFETCH
71	b	common_invalid
72
73__dabt_invalid:
74	inv_entry BAD_DATA
75	b	common_invalid
76
77__irq_invalid:
78	inv_entry BAD_IRQ
79	b	common_invalid
80
81__und_invalid:
82	inv_entry BAD_UNDEFINSTR
83
84	@
85	@ XXX fall through to common_invalid
86	@
87
88@
89@ common_invalid - generic code for failed exception (re-entrant version of handlers)
90@
91common_invalid:
92	zero_fp
93
94	ldmia	r0, {r4 - r6}
95	add	r0, sp, #S_PC		@ here for interlock avoidance
96	mov	r7, #-1			@  ""   ""    ""        ""
97	str	r4, [sp]		@ save preserved r0
98	stmia	r0, {r5 - r7}		@ lr_<exception>,
99					@ cpsr_<exception>, "old_r0"
100
101	mov	r0, sp
102	and	r2, r6, #0x1f
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_PREEMPT
195	get_thread_info tsk
196	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
197	add	r7, r8, #1			@ increment it
198	str	r7, [tsk, #TI_PREEMPT]
199#endif
200
201	irq_handler
202#ifdef CONFIG_PREEMPT
203	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
204	tst	r0, #_TIF_NEED_RESCHED
205	blne	svc_preempt
206preempt_return:
207	ldr	r0, [tsk, #TI_PREEMPT]		@ read preempt value
208	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
209	teq	r0, r7
210	strne	r0, [r0, -r0]			@ bug()
211#endif
212	ldr	r0, [sp, #S_PSR]		@ irqs are already disabled
213	msr	spsr_cxsf, r0
214	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
215
216	.ltorg
217
218#ifdef CONFIG_PREEMPT
219svc_preempt:
220	teq	r8, #0				@ was preempt count = 0
221	ldreq	r6, .LCirq_stat
222	movne	pc, lr				@ no
223	ldr	r0, [r6, #4]			@ local_irq_count
224	ldr	r1, [r6, #8]			@ local_bh_count
225	adds	r0, r0, r1
226	movne	pc, lr
227	mov	r7, #0				@ preempt_schedule_irq
228	str	r7, [tsk, #TI_PREEMPT]		@ expects preempt_count == 0
2291:	bl	preempt_schedule_irq		@ irq en/disable is done inside
230	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
231	tst	r0, #_TIF_NEED_RESCHED
232	beq	preempt_return			@ go again
233	b	1b
234#endif
235
236	.align	5
237__und_svc:
238	svc_entry
239
240	@
241	@ call emulation code, which returns using r9 if it has emulated
242	@ the instruction, or the more conventional lr if we are to treat
243	@ this as a real undefined instruction
244	@
245	@  r0 - instruction
246	@
247	ldr	r0, [r2, #-4]
248	adr	r9, 1f
249	bl	call_fpe
250
251	mov	r0, sp				@ struct pt_regs *regs
252	bl	do_undefinstr
253
254	@
255	@ IRQs off again before pulling preserved data off the stack
256	@
2571:	disable_irq
258
259	@
260	@ restore SPSR and restart the instruction
261	@
262	ldr	lr, [sp, #S_PSR]		@ Get SVC cpsr
263	msr	spsr_cxsf, lr
264	ldmia	sp, {r0 - pc}^			@ Restore SVC registers
265
266	.align	5
267__pabt_svc:
268	svc_entry
269
270	@
271	@ re-enable interrupts if appropriate
272	@
273	mrs	r9, cpsr
274	tst	r3, #PSR_I_BIT
275	biceq	r9, r9, #PSR_I_BIT
276	msr	cpsr_c, r9
277
278	@
279	@ set args, then call main handler
280	@
281	@  r0 - address of faulting instruction
282	@  r1 - pointer to registers on stack
283	@
284	mov	r0, r2				@ address (pc)
285	mov	r1, sp				@ regs
286	bl	do_PrefetchAbort		@ call abort handler
287
288	@
289	@ IRQs off again before pulling preserved data off the stack
290	@
291	disable_irq
292
293	@
294	@ restore SPSR and restart the instruction
295	@
296	ldr	r0, [sp, #S_PSR]
297	msr	spsr_cxsf, r0
298	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
299
300	.align	5
301.LCcralign:
302	.word	cr_alignment
303#ifdef MULTI_ABORT
304.LCprocfns:
305	.word	processor
306#endif
307.LCfp:
308	.word	fp_enter
309#ifdef CONFIG_PREEMPT
310.LCirq_stat:
311	.word	irq_stat
312#endif
313
314/*
315 * User mode handlers
316 *
317 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
318 */
319
320#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
321#error "sizeof(struct pt_regs) must be a multiple of 8"
322#endif
323
324	.macro	usr_entry
325	sub	sp, sp, #S_FRAME_SIZE
326	stmib	sp, {r1 - r12}
327
328	ldmia	r0, {r1 - r3}
329	add	r0, sp, #S_PC		@ here for interlock avoidance
330	mov	r4, #-1			@  ""  ""     ""        ""
331
332	str	r1, [sp]		@ save the "real" r0 copied
333					@ from the exception stack
334
335#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
336#ifndef CONFIG_MMU
337#warning "NPTL on non MMU needs fixing"
338#else
339	@ make sure our user space atomic helper is aborted
340	cmp	r2, #TASK_SIZE
341	bichs	r3, r3, #PSR_Z_BIT
342#endif
343#endif
344
345	@
346	@ We are now ready to fill in the remaining blanks on the stack:
347	@
348	@  r2 - lr_<exception>, already fixed up for correct return/restart
349	@  r3 - spsr_<exception>
350	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
351	@
352	@ Also, separately save sp_usr and lr_usr
353	@
354	stmia	r0, {r2 - r4}
355	stmdb	r0, {sp, lr}^
356
357	@
358	@ Enable the alignment trap while in kernel mode
359	@
360	alignment_trap r0
361
362	@
363	@ Clear FP to mark the first stack frame
364	@
365	zero_fp
366	.endm
367
368	.align	5
369__dabt_usr:
370	usr_entry
371
372	@
373	@ Call the processor-specific abort handler:
374	@
375	@  r2 - aborted context pc
376	@  r3 - aborted context cpsr
377	@
378	@ The abort handler must return the aborted address in r0, and
379	@ the fault status register in r1.
380	@
381#ifdef MULTI_ABORT
382	ldr	r4, .LCprocfns
383	mov	lr, pc
384	ldr	pc, [r4]
385#else
386	bl	CPU_ABORT_HANDLER
387#endif
388
389	@
390	@ IRQs on, then call the main handler
391	@
392	enable_irq
393	mov	r2, sp
394	adr	lr, ret_from_exception
395	b	do_DataAbort
396
397	.align	5
398__irq_usr:
399	usr_entry
400
401	get_thread_info tsk
402#ifdef CONFIG_PREEMPT
403	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
404	add	r7, r8, #1			@ increment it
405	str	r7, [tsk, #TI_PREEMPT]
406#endif
407
408	irq_handler
409#ifdef CONFIG_PREEMPT
410	ldr	r0, [tsk, #TI_PREEMPT]
411	str	r8, [tsk, #TI_PREEMPT]
412	teq	r0, r7
413	strne	r0, [r0, -r0]
414#endif
415
416	mov	why, #0
417	b	ret_to_user
418
419	.ltorg
420
421	.align	5
422__und_usr:
423	usr_entry
424
425	tst	r3, #PSR_T_BIT			@ Thumb mode?
426	bne	fpundefinstr			@ ignore FP
427	sub	r4, r2, #4
428
429	@
430	@ fall through to the emulation code, which returns using r9 if
431	@ it has emulated the instruction, or the more conventional lr
432	@ if we are to treat this as a real undefined instruction
433	@
434	@  r0 - instruction
435	@
4361:	ldrt	r0, [r4]
437	adr	r9, ret_from_exception
438	adr	lr, fpundefinstr
439	@
440	@ fallthrough to call_fpe
441	@
442
443/*
444 * The out of line fixup for the ldrt above.
445 */
446	.section .fixup, "ax"
4472:	mov	pc, r9
448	.previous
449	.section __ex_table,"a"
450	.long	1b, 2b
451	.previous
452
453/*
454 * Check whether the instruction is a co-processor instruction.
455 * If yes, we need to call the relevant co-processor handler.
456 *
457 * Note that we don't do a full check here for the co-processor
458 * instructions; all instructions with bit 27 set are well
459 * defined.  The only instructions that should fault are the
460 * co-processor instructions.  However, we have to watch out
461 * for the ARM6/ARM7 SWI bug.
462 *
463 * Emulators may wish to make use of the following registers:
464 *  r0  = instruction opcode.
465 *  r2  = PC+4
466 *  r10 = this threads thread_info structure.
467 */
468call_fpe:
469	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
470#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
471	and	r8, r0, #0x0f000000		@ mask out op-code bits
472	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
473#endif
474	moveq	pc, lr
475	get_thread_info r10			@ get current thread
476	and	r8, r0, #0x00000f00		@ mask out CP number
477	mov	r7, #1
478	add	r6, r10, #TI_USED_CP
479	strb	r7, [r6, r8, lsr #8]		@ set appropriate used_cp[]
480#ifdef CONFIG_IWMMXT
481	@ Test if we need to give access to iWMMXt coprocessors
482	ldr	r5, [r10, #TI_FLAGS]
483	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
484	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
485	bcs	iwmmxt_task_enable
486#endif
487	add	pc, pc, r8, lsr #6
488	mov	r0, r0
489
490	mov	pc, lr				@ CP#0
491	b	do_fpe				@ CP#1 (FPE)
492	b	do_fpe				@ CP#2 (FPE)
493	mov	pc, lr				@ CP#3
494#ifdef CONFIG_CRUNCH
495	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
496	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
497	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
498#else
499	mov	pc, lr				@ CP#4
500	mov	pc, lr				@ CP#5
501	mov	pc, lr				@ CP#6
502#endif
503	mov	pc, lr				@ CP#7
504	mov	pc, lr				@ CP#8
505	mov	pc, lr				@ CP#9
506#ifdef CONFIG_VFP
507	b	do_vfp				@ CP#10 (VFP)
508	b	do_vfp				@ CP#11 (VFP)
509#else
510	mov	pc, lr				@ CP#10 (VFP)
511	mov	pc, lr				@ CP#11 (VFP)
512#endif
513	mov	pc, lr				@ CP#12
514	mov	pc, lr				@ CP#13
515	mov	pc, lr				@ CP#14 (Debug)
516	mov	pc, lr				@ CP#15 (Control)
517
518do_fpe:
519	enable_irq
520	ldr	r4, .LCfp
521	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
522	ldr	pc, [r4]			@ Call FP module USR entry point
523
524/*
525 * The FP module is called with these registers set:
526 *  r0  = instruction
527 *  r2  = PC+4
528 *  r9  = normal "successful" return address
529 *  r10 = FP workspace
530 *  lr  = unrecognised FP instruction return address
531 */
532
533	.data
534ENTRY(fp_enter)
535	.word	fpundefinstr
536	.text
537
538fpundefinstr:
539	mov	r0, sp
540	adr	lr, ret_from_exception
541	b	do_undefinstr
542
543	.align	5
544__pabt_usr:
545	usr_entry
546
547	enable_irq				@ Enable interrupts
548	mov	r0, r2				@ address (pc)
549	mov	r1, sp				@ regs
550	bl	do_PrefetchAbort		@ call abort handler
551	/* fall through */
552/*
553 * This is the return code to user mode for abort handlers
554 */
555ENTRY(ret_from_exception)
556	get_thread_info tsk
557	mov	why, #0
558	b	ret_to_user
559
560/*
561 * Register switch for ARMv3 and ARMv4 processors
562 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
563 * previous and next are guaranteed not to be the same.
564 */
565ENTRY(__switch_to)
566	add	ip, r1, #TI_CPU_SAVE
567	ldr	r3, [r2, #TI_TP_VALUE]
568	stmia	ip!, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
569#ifdef CONFIG_MMU
570	ldr	r6, [r2, #TI_CPU_DOMAIN]
571#endif
572#if __LINUX_ARM_ARCH__ >= 6
573#ifdef CONFIG_CPU_32v6K
574	clrex
575#else
576	strex	r5, r4, [ip]			@ Clear exclusive monitor
577#endif
578#endif
579#if defined(CONFIG_CPU_XSCALE) && !defined(CONFIG_IWMMXT)
580	mra	r4, r5, acc0
581	stmia   ip, {r4, r5}
582#endif
583#if defined(CONFIG_HAS_TLS_REG)
584	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
585#elif !defined(CONFIG_TLS_REG_EMUL)
586	mov	r4, #0xffff0fff
587	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
588#endif
589#ifdef CONFIG_MMU
590	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
591#endif
592#if defined(CONFIG_CPU_XSCALE) && !defined(CONFIG_IWMMXT)
593	add	r4, r2, #TI_CPU_DOMAIN + 40	@ cpu_context_save->extra
594	ldmib	r4, {r4, r5}
595	mar	acc0, r4, r5
596#endif
597	mov	r5, r0
598	add	r4, r2, #TI_CPU_SAVE
599	ldr	r0, =thread_notify_head
600	mov	r1, #THREAD_NOTIFY_SWITCH
601	bl	atomic_notifier_call_chain
602	mov	r0, r5
603	ldmia	r4, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
604
605	__INIT
606
607/*
608 * User helpers.
609 *
610 * These are segment of kernel provided user code reachable from user space
611 * at a fixed address in kernel memory.  This is used to provide user space
612 * with some operations which require kernel help because of unimplemented
613 * native feature and/or instructions in many ARM CPUs. The idea is for
614 * this code to be executed directly in user mode for best efficiency but
615 * which is too intimate with the kernel counter part to be left to user
616 * libraries.  In fact this code might even differ from one CPU to another
617 * depending on the available  instruction set and restrictions like on
618 * SMP systems.  In other words, the kernel reserves the right to change
619 * this code as needed without warning. Only the entry points and their
620 * results are guaranteed to be stable.
621 *
622 * Each segment is 32-byte aligned and will be moved to the top of the high
623 * vector page.  New segments (if ever needed) must be added in front of
624 * existing ones.  This mechanism should be used only for things that are
625 * really small and justified, and not be abused freely.
626 *
627 * User space is expected to implement those things inline when optimizing
628 * for a processor that has the necessary native support, but only if such
629 * resulting binaries are already to be incompatible with earlier ARM
630 * processors due to the use of unsupported instructions other than what
631 * is provided here.  In other words don't make binaries unable to run on
632 * earlier processors just for the sake of not using these kernel helpers
633 * if your compiled code is not going to use the new instructions for other
634 * purpose.
635 */
636
637	.align	5
638	.globl	__kuser_helper_start
639__kuser_helper_start:
640
641/*
642 * Reference prototype:
643 *
644 *	void __kernel_memory_barrier(void)
645 *
646 * Input:
647 *
648 *	lr = return address
649 *
650 * Output:
651 *
652 *	none
653 *
654 * Clobbered:
655 *
656 *	the Z flag might be lost
657 *
658 * Definition and user space usage example:
659 *
660 *	typedef void (__kernel_dmb_t)(void);
661 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
662 *
663 * Apply any needed memory barrier to preserve consistency with data modified
664 * manually and __kuser_cmpxchg usage.
665 *
666 * This could be used as follows:
667 *
668 * #define __kernel_dmb() \
669 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
670 *	        : : : "r0", "lr","cc" )
671 */
672
673__kuser_memory_barrier:				@ 0xffff0fa0
674
675#if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
676	mcr	p15, 0, r0, c7, c10, 5	@ dmb
677#endif
678	mov	pc, lr
679
680	.align	5
681
682/*
683 * Reference prototype:
684 *
685 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
686 *
687 * Input:
688 *
689 *	r0 = oldval
690 *	r1 = newval
691 *	r2 = ptr
692 *	lr = return address
693 *
694 * Output:
695 *
696 *	r0 = returned value (zero or non-zero)
697 *	C flag = set if r0 == 0, clear if r0 != 0
698 *
699 * Clobbered:
700 *
701 *	r3, ip, flags
702 *
703 * Definition and user space usage example:
704 *
705 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
706 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
707 *
708 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
709 * Return zero if *ptr was changed or non-zero if no exchange happened.
710 * The C flag is also set if *ptr was changed to allow for assembly
711 * optimization in the calling code.
712 *
713 * Notes:
714 *
715 *    - This routine already includes memory barriers as needed.
716 *
717 *    - A failure might be transient, i.e. it is possible, although unlikely,
718 *      that "failure" be returned even if *ptr == oldval.
719 *
720 * For example, a user space atomic_add implementation could look like this:
721 *
722 * #define atomic_add(ptr, val) \
723 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
724 *	   register unsigned int __result asm("r1"); \
725 *	   asm volatile ( \
726 *	       "1: @ atomic_add\n\t" \
727 *	       "ldr	r0, [r2]\n\t" \
728 *	       "mov	r3, #0xffff0fff\n\t" \
729 *	       "add	lr, pc, #4\n\t" \
730 *	       "add	r1, r0, %2\n\t" \
731 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
732 *	       "bcc	1b" \
733 *	       : "=&r" (__result) \
734 *	       : "r" (__ptr), "rIL" (val) \
735 *	       : "r0","r3","ip","lr","cc","memory" ); \
736 *	   __result; })
737 */
738
739__kuser_cmpxchg:				@ 0xffff0fc0
740
741#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
742
743	/*
744	 * Poor you.  No fast solution possible...
745	 * The kernel itself must perform the operation.
746	 * A special ghost syscall is used for that (see traps.c).
747	 */
748	stmfd	sp!, {r7, lr}
749	mov	r7, #0xff00		@ 0xfff0 into r7 for EABI
750	orr	r7, r7, #0xf0
751	swi	#0x9ffff0
752	ldmfd	sp!, {r7, pc}
753
754#elif __LINUX_ARM_ARCH__ < 6
755
756	/*
757	 * Theory of operation:
758	 *
759	 * We set the Z flag before loading oldval. If ever an exception
760	 * occurs we can not be sure the loaded value will still be the same
761	 * when the exception returns, therefore the user exception handler
762	 * will clear the Z flag whenever the interrupted user code was
763	 * actually from the kernel address space (see the usr_entry macro).
764	 *
765	 * The post-increment on the str is used to prevent a race with an
766	 * exception happening just after the str instruction which would
767	 * clear the Z flag although the exchange was done.
768	 */
769#ifdef CONFIG_MMU
770	teq	ip, ip			@ set Z flag
771	ldr	ip, [r2]		@ load current val
772	add	r3, r2, #1		@ prepare store ptr
773	teqeq	ip, r0			@ compare with oldval if still allowed
774	streq	r1, [r3, #-1]!		@ store newval if still allowed
775	subs	r0, r2, r3		@ if r2 == r3 the str occured
776#else
777#warning "NPTL on non MMU needs fixing"
778	mov	r0, #-1
779	adds	r0, r0, #0
780#endif
781	mov	pc, lr
782
783#else
784
785#ifdef CONFIG_SMP
786	mcr	p15, 0, r0, c7, c10, 5	@ dmb
787#endif
788	ldrex	r3, [r2]
789	subs	r3, r3, r0
790	strexeq	r3, r1, [r2]
791	rsbs	r0, r3, #0
792#ifdef CONFIG_SMP
793	mcr	p15, 0, r0, c7, c10, 5	@ dmb
794#endif
795	mov	pc, lr
796
797#endif
798
799	.align	5
800
801/*
802 * Reference prototype:
803 *
804 *	int __kernel_get_tls(void)
805 *
806 * Input:
807 *
808 *	lr = return address
809 *
810 * Output:
811 *
812 *	r0 = TLS value
813 *
814 * Clobbered:
815 *
816 *	the Z flag might be lost
817 *
818 * Definition and user space usage example:
819 *
820 *	typedef int (__kernel_get_tls_t)(void);
821 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
822 *
823 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
824 *
825 * This could be used as follows:
826 *
827 * #define __kernel_get_tls() \
828 *	({ register unsigned int __val asm("r0"); \
829 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
830 *	        : "=r" (__val) : : "lr","cc" ); \
831 *	   __val; })
832 */
833
834__kuser_get_tls:				@ 0xffff0fe0
835
836#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
837
838	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
839	mov	pc, lr
840
841#else
842
843	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
844	mov	pc, lr
845
846#endif
847
848	.rep	5
849	.word	0			@ pad up to __kuser_helper_version
850	.endr
851
852/*
853 * Reference declaration:
854 *
855 *	extern unsigned int __kernel_helper_version;
856 *
857 * Definition and user space usage example:
858 *
859 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
860 *
861 * User space may read this to determine the curent number of helpers
862 * available.
863 */
864
865__kuser_helper_version:				@ 0xffff0ffc
866	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
867
868	.globl	__kuser_helper_end
869__kuser_helper_end:
870
871
872/*
873 * Vector stubs.
874 *
875 * This code is copied to 0xffff0200 so we can use branches in the
876 * vectors, rather than ldr's.  Note that this code must not
877 * exceed 0x300 bytes.
878 *
879 * Common stub entry macro:
880 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
881 *
882 * SP points to a minimal amount of processor-private memory, the address
883 * of which is copied into r0 for the mode specific abort handler.
884 */
885	.macro	vector_stub, name, mode, correction=0
886	.align	5
887
888vector_\name:
889	.if \correction
890	sub	lr, lr, #\correction
891	.endif
892
893	@
894	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
895	@ (parent CPSR)
896	@
897	stmia	sp, {r0, lr}		@ save r0, lr
898	mrs	lr, spsr
899	str	lr, [sp, #8]		@ save spsr
900
901	@
902	@ Prepare for SVC32 mode.  IRQs remain disabled.
903	@
904	mrs	r0, cpsr
905	eor	r0, r0, #(\mode ^ SVC_MODE)
906	msr	spsr_cxsf, r0
907
908	@
909	@ the branch table must immediately follow this code
910	@
911	and	lr, lr, #0x0f
912	mov	r0, sp
913	ldr	lr, [pc, lr, lsl #2]
914	movs	pc, lr			@ branch to handler in SVC mode
915	.endm
916
917	.globl	__stubs_start
918__stubs_start:
919/*
920 * Interrupt dispatcher
921 */
922	vector_stub	irq, IRQ_MODE, 4
923
924	.long	__irq_usr			@  0  (USR_26 / USR_32)
925	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
926	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
927	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
928	.long	__irq_invalid			@  4
929	.long	__irq_invalid			@  5
930	.long	__irq_invalid			@  6
931	.long	__irq_invalid			@  7
932	.long	__irq_invalid			@  8
933	.long	__irq_invalid			@  9
934	.long	__irq_invalid			@  a
935	.long	__irq_invalid			@  b
936	.long	__irq_invalid			@  c
937	.long	__irq_invalid			@  d
938	.long	__irq_invalid			@  e
939	.long	__irq_invalid			@  f
940
941/*
942 * Data abort dispatcher
943 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
944 */
945	vector_stub	dabt, ABT_MODE, 8
946
947	.long	__dabt_usr			@  0  (USR_26 / USR_32)
948	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
949	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
950	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
951	.long	__dabt_invalid			@  4
952	.long	__dabt_invalid			@  5
953	.long	__dabt_invalid			@  6
954	.long	__dabt_invalid			@  7
955	.long	__dabt_invalid			@  8
956	.long	__dabt_invalid			@  9
957	.long	__dabt_invalid			@  a
958	.long	__dabt_invalid			@  b
959	.long	__dabt_invalid			@  c
960	.long	__dabt_invalid			@  d
961	.long	__dabt_invalid			@  e
962	.long	__dabt_invalid			@  f
963
964/*
965 * Prefetch abort dispatcher
966 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
967 */
968	vector_stub	pabt, ABT_MODE, 4
969
970	.long	__pabt_usr			@  0 (USR_26 / USR_32)
971	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
972	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
973	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
974	.long	__pabt_invalid			@  4
975	.long	__pabt_invalid			@  5
976	.long	__pabt_invalid			@  6
977	.long	__pabt_invalid			@  7
978	.long	__pabt_invalid			@  8
979	.long	__pabt_invalid			@  9
980	.long	__pabt_invalid			@  a
981	.long	__pabt_invalid			@  b
982	.long	__pabt_invalid			@  c
983	.long	__pabt_invalid			@  d
984	.long	__pabt_invalid			@  e
985	.long	__pabt_invalid			@  f
986
987/*
988 * Undef instr entry dispatcher
989 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
990 */
991	vector_stub	und, UND_MODE
992
993	.long	__und_usr			@  0 (USR_26 / USR_32)
994	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
995	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
996	.long	__und_svc			@  3 (SVC_26 / SVC_32)
997	.long	__und_invalid			@  4
998	.long	__und_invalid			@  5
999	.long	__und_invalid			@  6
1000	.long	__und_invalid			@  7
1001	.long	__und_invalid			@  8
1002	.long	__und_invalid			@  9
1003	.long	__und_invalid			@  a
1004	.long	__und_invalid			@  b
1005	.long	__und_invalid			@  c
1006	.long	__und_invalid			@  d
1007	.long	__und_invalid			@  e
1008	.long	__und_invalid			@  f
1009
1010	.align	5
1011
1012/*=============================================================================
1013 * Undefined FIQs
1014 *-----------------------------------------------------------------------------
1015 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1016 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1017 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1018 * damage alert!  I don't think that we can execute any code in here in any
1019 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1020 * get out of that mode without clobbering one register.
1021 */
1022vector_fiq:
1023	disable_fiq
1024	subs	pc, lr, #4
1025
1026/*=============================================================================
1027 * Address exception handler
1028 *-----------------------------------------------------------------------------
1029 * These aren't too critical.
1030 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1031 */
1032
1033vector_addrexcptn:
1034	b	vector_addrexcptn
1035
1036/*
1037 * We group all the following data together to optimise
1038 * for CPUs with separate I & D caches.
1039 */
1040	.align	5
1041
1042.LCvswi:
1043	.word	vector_swi
1044
1045	.globl	__stubs_end
1046__stubs_end:
1047
1048	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1049
1050	.globl	__vectors_start
1051__vectors_start:
1052	swi	SYS_ERROR0
1053	b	vector_und + stubs_offset
1054	ldr	pc, .LCvswi + stubs_offset
1055	b	vector_pabt + stubs_offset
1056	b	vector_dabt + stubs_offset
1057	b	vector_addrexcptn + stubs_offset
1058	b	vector_irq + stubs_offset
1059	b	vector_fiq + stubs_offset
1060
1061	.globl	__vectors_end
1062__vectors_end:
1063
1064	.data
1065
1066	.globl	cr_alignment
1067	.globl	cr_no_alignment
1068cr_alignment:
1069	.space	4
1070cr_no_alignment:
1071	.space	4
1072