xref: /linux/arch/arm/kernel/entry-armv.S (revision 367b8112fe2ea5c39a7bb4d263dcdd9b612fae18)
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
15 *  that causes 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 <mach/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#ifdef CONFIG_KPROBES
62	.section	.kprobes.text,"ax",%progbits
63#else
64	.text
65#endif
66
67/*
68 * Invalid mode handlers
69 */
70	.macro	inv_entry, reason
71	sub	sp, sp, #S_FRAME_SIZE
72	stmib	sp, {r1 - lr}
73	mov	r1, #\reason
74	.endm
75
76__pabt_invalid:
77	inv_entry BAD_PREFETCH
78	b	common_invalid
79ENDPROC(__pabt_invalid)
80
81__dabt_invalid:
82	inv_entry BAD_DATA
83	b	common_invalid
84ENDPROC(__dabt_invalid)
85
86__irq_invalid:
87	inv_entry BAD_IRQ
88	b	common_invalid
89ENDPROC(__irq_invalid)
90
91__und_invalid:
92	inv_entry BAD_UNDEFINSTR
93
94	@
95	@ XXX fall through to common_invalid
96	@
97
98@
99@ common_invalid - generic code for failed exception (re-entrant version of handlers)
100@
101common_invalid:
102	zero_fp
103
104	ldmia	r0, {r4 - r6}
105	add	r0, sp, #S_PC		@ here for interlock avoidance
106	mov	r7, #-1			@  ""   ""    ""        ""
107	str	r4, [sp]		@ save preserved r0
108	stmia	r0, {r5 - r7}		@ lr_<exception>,
109					@ cpsr_<exception>, "old_r0"
110
111	mov	r0, sp
112	b	bad_mode
113ENDPROC(__und_invalid)
114
115/*
116 * SVC mode handlers
117 */
118
119#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
120#define SPFIX(code...) code
121#else
122#define SPFIX(code...)
123#endif
124
125	.macro	svc_entry, stack_hole=0
126	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole)
127 SPFIX(	tst	sp, #4		)
128 SPFIX(	bicne	sp, sp, #4	)
129	stmib	sp, {r1 - r12}
130
131	ldmia	r0, {r1 - r3}
132	add	r5, sp, #S_SP		@ here for interlock avoidance
133	mov	r4, #-1			@  ""  ""      ""       ""
134	add	r0, sp, #(S_FRAME_SIZE + \stack_hole)
135 SPFIX(	addne	r0, r0, #4	)
136	str	r1, [sp]		@ save the "real" r0 copied
137					@ from the exception stack
138
139	mov	r1, lr
140
141	@
142	@ We are now ready to fill in the remaining blanks on the stack:
143	@
144	@  r0 - sp_svc
145	@  r1 - lr_svc
146	@  r2 - lr_<exception>, already fixed up for correct return/restart
147	@  r3 - spsr_<exception>
148	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
149	@
150	stmia	r5, {r0 - r4}
151	.endm
152
153	.align	5
154__dabt_svc:
155	svc_entry
156
157	@
158	@ get ready to re-enable interrupts if appropriate
159	@
160	mrs	r9, cpsr
161	tst	r3, #PSR_I_BIT
162	biceq	r9, r9, #PSR_I_BIT
163
164	@
165	@ Call the processor-specific abort handler:
166	@
167	@  r2 - aborted context pc
168	@  r3 - aborted context cpsr
169	@
170	@ The abort handler must return the aborted address in r0, and
171	@ the fault status register in r1.  r9 must be preserved.
172	@
173#ifdef MULTI_DABORT
174	ldr	r4, .LCprocfns
175	mov	lr, pc
176	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
177#else
178	bl	CPU_DABORT_HANDLER
179#endif
180
181	@
182	@ set desired IRQ state, then call main handler
183	@
184	msr	cpsr_c, r9
185	mov	r2, sp
186	bl	do_DataAbort
187
188	@
189	@ IRQs off again before pulling preserved data off the stack
190	@
191	disable_irq
192
193	@
194	@ restore SPSR and restart the instruction
195	@
196	ldr	r0, [sp, #S_PSR]
197	msr	spsr_cxsf, r0
198	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
199ENDPROC(__dabt_svc)
200
201	.align	5
202__irq_svc:
203	svc_entry
204
205#ifdef CONFIG_TRACE_IRQFLAGS
206	bl	trace_hardirqs_off
207#endif
208#ifdef CONFIG_PREEMPT
209	get_thread_info tsk
210	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
211	add	r7, r8, #1			@ increment it
212	str	r7, [tsk, #TI_PREEMPT]
213#endif
214
215	irq_handler
216#ifdef CONFIG_PREEMPT
217	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
218	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
219	teq	r8, #0				@ if preempt count != 0
220	movne	r0, #0				@ force flags to 0
221	tst	r0, #_TIF_NEED_RESCHED
222	blne	svc_preempt
223#endif
224	ldr	r0, [sp, #S_PSR]		@ irqs are already disabled
225	msr	spsr_cxsf, r0
226#ifdef CONFIG_TRACE_IRQFLAGS
227	tst	r0, #PSR_I_BIT
228	bleq	trace_hardirqs_on
229#endif
230	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
231ENDPROC(__irq_svc)
232
233	.ltorg
234
235#ifdef CONFIG_PREEMPT
236svc_preempt:
237	mov	r8, lr
2381:	bl	preempt_schedule_irq		@ irq en/disable is done inside
239	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
240	tst	r0, #_TIF_NEED_RESCHED
241	moveq	pc, r8				@ go again
242	b	1b
243#endif
244
245	.align	5
246__und_svc:
247#ifdef CONFIG_KPROBES
248	@ If a kprobe is about to simulate a "stmdb sp..." instruction,
249	@ it obviously needs free stack space which then will belong to
250	@ the saved context.
251	svc_entry 64
252#else
253	svc_entry
254#endif
255
256	@
257	@ call emulation code, which returns using r9 if it has emulated
258	@ the instruction, or the more conventional lr if we are to treat
259	@ this as a real undefined instruction
260	@
261	@  r0 - instruction
262	@
263	ldr	r0, [r2, #-4]
264	adr	r9, 1f
265	bl	call_fpe
266
267	mov	r0, sp				@ struct pt_regs *regs
268	bl	do_undefinstr
269
270	@
271	@ IRQs off again before pulling preserved data off the stack
272	@
2731:	disable_irq
274
275	@
276	@ restore SPSR and restart the instruction
277	@
278	ldr	lr, [sp, #S_PSR]		@ Get SVC cpsr
279	msr	spsr_cxsf, lr
280	ldmia	sp, {r0 - pc}^			@ Restore SVC registers
281ENDPROC(__und_svc)
282
283	.align	5
284__pabt_svc:
285	svc_entry
286
287	@
288	@ re-enable interrupts if appropriate
289	@
290	mrs	r9, cpsr
291	tst	r3, #PSR_I_BIT
292	biceq	r9, r9, #PSR_I_BIT
293
294	@
295	@ set args, then call main handler
296	@
297	@  r0 - address of faulting instruction
298	@  r1 - pointer to registers on stack
299	@
300#ifdef MULTI_PABORT
301	mov	r0, r2			@ pass address of aborted instruction.
302	ldr	r4, .LCprocfns
303	mov	lr, pc
304	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
305#else
306	CPU_PABORT_HANDLER(r0, r2)
307#endif
308	msr	cpsr_c, r9			@ Maybe enable interrupts
309	mov	r1, sp				@ regs
310	bl	do_PrefetchAbort		@ call abort handler
311
312	@
313	@ IRQs off again before pulling preserved data off the stack
314	@
315	disable_irq
316
317	@
318	@ restore SPSR and restart the instruction
319	@
320	ldr	r0, [sp, #S_PSR]
321	msr	spsr_cxsf, r0
322	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
323ENDPROC(__pabt_svc)
324
325	.align	5
326.LCcralign:
327	.word	cr_alignment
328#ifdef MULTI_DABORT
329.LCprocfns:
330	.word	processor
331#endif
332.LCfp:
333	.word	fp_enter
334
335/*
336 * User mode handlers
337 *
338 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
339 */
340
341#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
342#error "sizeof(struct pt_regs) must be a multiple of 8"
343#endif
344
345	.macro	usr_entry
346	sub	sp, sp, #S_FRAME_SIZE
347	stmib	sp, {r1 - r12}
348
349	ldmia	r0, {r1 - r3}
350	add	r0, sp, #S_PC		@ here for interlock avoidance
351	mov	r4, #-1			@  ""  ""     ""        ""
352
353	str	r1, [sp]		@ save the "real" r0 copied
354					@ from the exception stack
355
356	@
357	@ We are now ready to fill in the remaining blanks on the stack:
358	@
359	@  r2 - lr_<exception>, already fixed up for correct return/restart
360	@  r3 - spsr_<exception>
361	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
362	@
363	@ Also, separately save sp_usr and lr_usr
364	@
365	stmia	r0, {r2 - r4}
366	stmdb	r0, {sp, lr}^
367
368	@
369	@ Enable the alignment trap while in kernel mode
370	@
371	alignment_trap r0
372
373	@
374	@ Clear FP to mark the first stack frame
375	@
376	zero_fp
377	.endm
378
379	.macro	kuser_cmpxchg_check
380#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
381#ifndef CONFIG_MMU
382#warning "NPTL on non MMU needs fixing"
383#else
384	@ Make sure our user space atomic helper is restarted
385	@ if it was interrupted in a critical region.  Here we
386	@ perform a quick test inline since it should be false
387	@ 99.9999% of the time.  The rest is done out of line.
388	cmp	r2, #TASK_SIZE
389	blhs	kuser_cmpxchg_fixup
390#endif
391#endif
392	.endm
393
394	.align	5
395__dabt_usr:
396	usr_entry
397	kuser_cmpxchg_check
398
399	@
400	@ Call the processor-specific abort handler:
401	@
402	@  r2 - aborted context pc
403	@  r3 - aborted context cpsr
404	@
405	@ The abort handler must return the aborted address in r0, and
406	@ the fault status register in r1.
407	@
408#ifdef MULTI_DABORT
409	ldr	r4, .LCprocfns
410	mov	lr, pc
411	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
412#else
413	bl	CPU_DABORT_HANDLER
414#endif
415
416	@
417	@ IRQs on, then call the main handler
418	@
419	enable_irq
420	mov	r2, sp
421	adr	lr, ret_from_exception
422	b	do_DataAbort
423ENDPROC(__dabt_usr)
424
425	.align	5
426__irq_usr:
427	usr_entry
428	kuser_cmpxchg_check
429
430#ifdef CONFIG_TRACE_IRQFLAGS
431	bl	trace_hardirqs_off
432#endif
433	get_thread_info tsk
434#ifdef CONFIG_PREEMPT
435	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
436	add	r7, r8, #1			@ increment it
437	str	r7, [tsk, #TI_PREEMPT]
438#endif
439
440	irq_handler
441#ifdef CONFIG_PREEMPT
442	ldr	r0, [tsk, #TI_PREEMPT]
443	str	r8, [tsk, #TI_PREEMPT]
444	teq	r0, r7
445	strne	r0, [r0, -r0]
446#endif
447#ifdef CONFIG_TRACE_IRQFLAGS
448	bl	trace_hardirqs_on
449#endif
450
451	mov	why, #0
452	b	ret_to_user
453ENDPROC(__irq_usr)
454
455	.ltorg
456
457	.align	5
458__und_usr:
459	usr_entry
460
461	@
462	@ fall through to the emulation code, which returns using r9 if
463	@ it has emulated the instruction, or the more conventional lr
464	@ if we are to treat this as a real undefined instruction
465	@
466	@  r0 - instruction
467	@
468	adr	r9, ret_from_exception
469	adr	lr, __und_usr_unknown
470	tst	r3, #PSR_T_BIT			@ Thumb mode?
471	subeq	r4, r2, #4			@ ARM instr at LR - 4
472	subne	r4, r2, #2			@ Thumb instr at LR - 2
4731:	ldreqt	r0, [r4]
474	beq	call_fpe
475	@ Thumb instruction
476#if __LINUX_ARM_ARCH__ >= 7
4772:	ldrht	r5, [r4], #2
478	and	r0, r5, #0xf800			@ mask bits 111x x... .... ....
479	cmp	r0, #0xe800			@ 32bit instruction if xx != 0
480	blo	__und_usr_unknown
4813:	ldrht	r0, [r4]
482	add	r2, r2, #2			@ r2 is PC + 2, make it PC + 4
483	orr	r0, r0, r5, lsl #16
484#else
485	b	__und_usr_unknown
486#endif
487ENDPROC(__und_usr)
488
489	@
490	@ fallthrough to call_fpe
491	@
492
493/*
494 * The out of line fixup for the ldrt above.
495 */
496	.section .fixup, "ax"
4974:	mov	pc, r9
498	.previous
499	.section __ex_table,"a"
500	.long	1b, 4b
501#if __LINUX_ARM_ARCH__ >= 7
502	.long	2b, 4b
503	.long	3b, 4b
504#endif
505	.previous
506
507/*
508 * Check whether the instruction is a co-processor instruction.
509 * If yes, we need to call the relevant co-processor handler.
510 *
511 * Note that we don't do a full check here for the co-processor
512 * instructions; all instructions with bit 27 set are well
513 * defined.  The only instructions that should fault are the
514 * co-processor instructions.  However, we have to watch out
515 * for the ARM6/ARM7 SWI bug.
516 *
517 * NEON is a special case that has to be handled here. Not all
518 * NEON instructions are co-processor instructions, so we have
519 * to make a special case of checking for them. Plus, there's
520 * five groups of them, so we have a table of mask/opcode pairs
521 * to check against, and if any match then we branch off into the
522 * NEON handler code.
523 *
524 * Emulators may wish to make use of the following registers:
525 *  r0  = instruction opcode.
526 *  r2  = PC+4
527 *  r9  = normal "successful" return address
528 *  r10 = this threads thread_info structure.
529 *  lr  = unrecognised instruction return address
530 */
531	@
532	@ Fall-through from Thumb-2 __und_usr
533	@
534#ifdef CONFIG_NEON
535	adr	r6, .LCneon_thumb_opcodes
536	b	2f
537#endif
538call_fpe:
539#ifdef CONFIG_NEON
540	adr	r6, .LCneon_arm_opcodes
5412:
542	ldr	r7, [r6], #4			@ mask value
543	cmp	r7, #0				@ end mask?
544	beq	1f
545	and	r8, r0, r7
546	ldr	r7, [r6], #4			@ opcode bits matching in mask
547	cmp	r8, r7				@ NEON instruction?
548	bne	2b
549	get_thread_info r10
550	mov	r7, #1
551	strb	r7, [r10, #TI_USED_CP + 10]	@ mark CP#10 as used
552	strb	r7, [r10, #TI_USED_CP + 11]	@ mark CP#11 as used
553	b	do_vfp				@ let VFP handler handle this
5541:
555#endif
556	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
557	tstne	r0, #0x04000000			@ bit 26 set on both ARM and Thumb-2
558#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
559	and	r8, r0, #0x0f000000		@ mask out op-code bits
560	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
561#endif
562	moveq	pc, lr
563	get_thread_info r10			@ get current thread
564	and	r8, r0, #0x00000f00		@ mask out CP number
565	mov	r7, #1
566	add	r6, r10, #TI_USED_CP
567	strb	r7, [r6, r8, lsr #8]		@ set appropriate used_cp[]
568#ifdef CONFIG_IWMMXT
569	@ Test if we need to give access to iWMMXt coprocessors
570	ldr	r5, [r10, #TI_FLAGS]
571	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
572	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
573	bcs	iwmmxt_task_enable
574#endif
575	add	pc, pc, r8, lsr #6
576	mov	r0, r0
577
578	mov	pc, lr				@ CP#0
579	b	do_fpe				@ CP#1 (FPE)
580	b	do_fpe				@ CP#2 (FPE)
581	mov	pc, lr				@ CP#3
582#ifdef CONFIG_CRUNCH
583	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
584	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
585	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
586#else
587	mov	pc, lr				@ CP#4
588	mov	pc, lr				@ CP#5
589	mov	pc, lr				@ CP#6
590#endif
591	mov	pc, lr				@ CP#7
592	mov	pc, lr				@ CP#8
593	mov	pc, lr				@ CP#9
594#ifdef CONFIG_VFP
595	b	do_vfp				@ CP#10 (VFP)
596	b	do_vfp				@ CP#11 (VFP)
597#else
598	mov	pc, lr				@ CP#10 (VFP)
599	mov	pc, lr				@ CP#11 (VFP)
600#endif
601	mov	pc, lr				@ CP#12
602	mov	pc, lr				@ CP#13
603	mov	pc, lr				@ CP#14 (Debug)
604	mov	pc, lr				@ CP#15 (Control)
605
606#ifdef CONFIG_NEON
607	.align	6
608
609.LCneon_arm_opcodes:
610	.word	0xfe000000			@ mask
611	.word	0xf2000000			@ opcode
612
613	.word	0xff100000			@ mask
614	.word	0xf4000000			@ opcode
615
616	.word	0x00000000			@ mask
617	.word	0x00000000			@ opcode
618
619.LCneon_thumb_opcodes:
620	.word	0xef000000			@ mask
621	.word	0xef000000			@ opcode
622
623	.word	0xff100000			@ mask
624	.word	0xf9000000			@ opcode
625
626	.word	0x00000000			@ mask
627	.word	0x00000000			@ opcode
628#endif
629
630do_fpe:
631	enable_irq
632	ldr	r4, .LCfp
633	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
634	ldr	pc, [r4]			@ Call FP module USR entry point
635
636/*
637 * The FP module is called with these registers set:
638 *  r0  = instruction
639 *  r2  = PC+4
640 *  r9  = normal "successful" return address
641 *  r10 = FP workspace
642 *  lr  = unrecognised FP instruction return address
643 */
644
645	.data
646ENTRY(fp_enter)
647	.word	no_fp
648	.previous
649
650no_fp:	mov	pc, lr
651
652__und_usr_unknown:
653	mov	r0, sp
654	adr	lr, ret_from_exception
655	b	do_undefinstr
656ENDPROC(__und_usr_unknown)
657
658	.align	5
659__pabt_usr:
660	usr_entry
661
662#ifdef MULTI_PABORT
663	mov	r0, r2			@ pass address of aborted instruction.
664	ldr	r4, .LCprocfns
665	mov	lr, pc
666	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
667#else
668	CPU_PABORT_HANDLER(r0, r2)
669#endif
670	enable_irq				@ Enable interrupts
671	mov	r1, sp				@ regs
672	bl	do_PrefetchAbort		@ call abort handler
673	/* fall through */
674/*
675 * This is the return code to user mode for abort handlers
676 */
677ENTRY(ret_from_exception)
678	get_thread_info tsk
679	mov	why, #0
680	b	ret_to_user
681ENDPROC(__pabt_usr)
682ENDPROC(ret_from_exception)
683
684/*
685 * Register switch for ARMv3 and ARMv4 processors
686 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
687 * previous and next are guaranteed not to be the same.
688 */
689ENTRY(__switch_to)
690	add	ip, r1, #TI_CPU_SAVE
691	ldr	r3, [r2, #TI_TP_VALUE]
692	stmia	ip!, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
693#ifdef CONFIG_MMU
694	ldr	r6, [r2, #TI_CPU_DOMAIN]
695#endif
696#if __LINUX_ARM_ARCH__ >= 6
697#ifdef CONFIG_CPU_32v6K
698	clrex
699#else
700	strex	r5, r4, [ip]			@ Clear exclusive monitor
701#endif
702#endif
703#if defined(CONFIG_HAS_TLS_REG)
704	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
705#elif !defined(CONFIG_TLS_REG_EMUL)
706	mov	r4, #0xffff0fff
707	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
708#endif
709#ifdef CONFIG_MMU
710	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
711#endif
712	mov	r5, r0
713	add	r4, r2, #TI_CPU_SAVE
714	ldr	r0, =thread_notify_head
715	mov	r1, #THREAD_NOTIFY_SWITCH
716	bl	atomic_notifier_call_chain
717	mov	r0, r5
718	ldmia	r4, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
719ENDPROC(__switch_to)
720
721	__INIT
722
723/*
724 * User helpers.
725 *
726 * These are segment of kernel provided user code reachable from user space
727 * at a fixed address in kernel memory.  This is used to provide user space
728 * with some operations which require kernel help because of unimplemented
729 * native feature and/or instructions in many ARM CPUs. The idea is for
730 * this code to be executed directly in user mode for best efficiency but
731 * which is too intimate with the kernel counter part to be left to user
732 * libraries.  In fact this code might even differ from one CPU to another
733 * depending on the available  instruction set and restrictions like on
734 * SMP systems.  In other words, the kernel reserves the right to change
735 * this code as needed without warning. Only the entry points and their
736 * results are guaranteed to be stable.
737 *
738 * Each segment is 32-byte aligned and will be moved to the top of the high
739 * vector page.  New segments (if ever needed) must be added in front of
740 * existing ones.  This mechanism should be used only for things that are
741 * really small and justified, and not be abused freely.
742 *
743 * User space is expected to implement those things inline when optimizing
744 * for a processor that has the necessary native support, but only if such
745 * resulting binaries are already to be incompatible with earlier ARM
746 * processors due to the use of unsupported instructions other than what
747 * is provided here.  In other words don't make binaries unable to run on
748 * earlier processors just for the sake of not using these kernel helpers
749 * if your compiled code is not going to use the new instructions for other
750 * purpose.
751 */
752
753	.macro	usr_ret, reg
754#ifdef CONFIG_ARM_THUMB
755	bx	\reg
756#else
757	mov	pc, \reg
758#endif
759	.endm
760
761	.align	5
762	.globl	__kuser_helper_start
763__kuser_helper_start:
764
765/*
766 * Reference prototype:
767 *
768 *	void __kernel_memory_barrier(void)
769 *
770 * Input:
771 *
772 *	lr = return address
773 *
774 * Output:
775 *
776 *	none
777 *
778 * Clobbered:
779 *
780 *	none
781 *
782 * Definition and user space usage example:
783 *
784 *	typedef void (__kernel_dmb_t)(void);
785 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
786 *
787 * Apply any needed memory barrier to preserve consistency with data modified
788 * manually and __kuser_cmpxchg usage.
789 *
790 * This could be used as follows:
791 *
792 * #define __kernel_dmb() \
793 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
794 *	        : : : "r0", "lr","cc" )
795 */
796
797__kuser_memory_barrier:				@ 0xffff0fa0
798
799#if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
800	mcr	p15, 0, r0, c7, c10, 5	@ dmb
801#endif
802	usr_ret	lr
803
804	.align	5
805
806/*
807 * Reference prototype:
808 *
809 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
810 *
811 * Input:
812 *
813 *	r0 = oldval
814 *	r1 = newval
815 *	r2 = ptr
816 *	lr = return address
817 *
818 * Output:
819 *
820 *	r0 = returned value (zero or non-zero)
821 *	C flag = set if r0 == 0, clear if r0 != 0
822 *
823 * Clobbered:
824 *
825 *	r3, ip, flags
826 *
827 * Definition and user space usage example:
828 *
829 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
830 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
831 *
832 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
833 * Return zero if *ptr was changed or non-zero if no exchange happened.
834 * The C flag is also set if *ptr was changed to allow for assembly
835 * optimization in the calling code.
836 *
837 * Notes:
838 *
839 *    - This routine already includes memory barriers as needed.
840 *
841 * For example, a user space atomic_add implementation could look like this:
842 *
843 * #define atomic_add(ptr, val) \
844 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
845 *	   register unsigned int __result asm("r1"); \
846 *	   asm volatile ( \
847 *	       "1: @ atomic_add\n\t" \
848 *	       "ldr	r0, [r2]\n\t" \
849 *	       "mov	r3, #0xffff0fff\n\t" \
850 *	       "add	lr, pc, #4\n\t" \
851 *	       "add	r1, r0, %2\n\t" \
852 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
853 *	       "bcc	1b" \
854 *	       : "=&r" (__result) \
855 *	       : "r" (__ptr), "rIL" (val) \
856 *	       : "r0","r3","ip","lr","cc","memory" ); \
857 *	   __result; })
858 */
859
860__kuser_cmpxchg:				@ 0xffff0fc0
861
862#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
863
864	/*
865	 * Poor you.  No fast solution possible...
866	 * The kernel itself must perform the operation.
867	 * A special ghost syscall is used for that (see traps.c).
868	 */
869	stmfd	sp!, {r7, lr}
870	mov	r7, #0xff00		@ 0xfff0 into r7 for EABI
871	orr	r7, r7, #0xf0
872	swi	#0x9ffff0
873	ldmfd	sp!, {r7, pc}
874
875#elif __LINUX_ARM_ARCH__ < 6
876
877#ifdef CONFIG_MMU
878
879	/*
880	 * The only thing that can break atomicity in this cmpxchg
881	 * implementation is either an IRQ or a data abort exception
882	 * causing another process/thread to be scheduled in the middle
883	 * of the critical sequence.  To prevent this, code is added to
884	 * the IRQ and data abort exception handlers to set the pc back
885	 * to the beginning of the critical section if it is found to be
886	 * within that critical section (see kuser_cmpxchg_fixup).
887	 */
8881:	ldr	r3, [r2]			@ load current val
889	subs	r3, r3, r0			@ compare with oldval
8902:	streq	r1, [r2]			@ store newval if eq
891	rsbs	r0, r3, #0			@ set return val and C flag
892	usr_ret	lr
893
894	.text
895kuser_cmpxchg_fixup:
896	@ Called from kuser_cmpxchg_check macro.
897	@ r2 = address of interrupted insn (must be preserved).
898	@ sp = saved regs. r7 and r8 are clobbered.
899	@ 1b = first critical insn, 2b = last critical insn.
900	@ If r2 >= 1b and r2 <= 2b then saved pc_usr is set to 1b.
901	mov	r7, #0xffff0fff
902	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
903	subs	r8, r2, r7
904	rsbcss	r8, r8, #(2b - 1b)
905	strcs	r7, [sp, #S_PC]
906	mov	pc, lr
907	.previous
908
909#else
910#warning "NPTL on non MMU needs fixing"
911	mov	r0, #-1
912	adds	r0, r0, #0
913	usr_ret	lr
914#endif
915
916#else
917
918#ifdef CONFIG_SMP
919	mcr	p15, 0, r0, c7, c10, 5	@ dmb
920#endif
9211:	ldrex	r3, [r2]
922	subs	r3, r3, r0
923	strexeq	r3, r1, [r2]
924	teqeq	r3, #1
925	beq	1b
926	rsbs	r0, r3, #0
927	/* beware -- each __kuser slot must be 8 instructions max */
928#ifdef CONFIG_SMP
929	b	__kuser_memory_barrier
930#else
931	usr_ret	lr
932#endif
933
934#endif
935
936	.align	5
937
938/*
939 * Reference prototype:
940 *
941 *	int __kernel_get_tls(void)
942 *
943 * Input:
944 *
945 *	lr = return address
946 *
947 * Output:
948 *
949 *	r0 = TLS value
950 *
951 * Clobbered:
952 *
953 *	none
954 *
955 * Definition and user space usage example:
956 *
957 *	typedef int (__kernel_get_tls_t)(void);
958 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
959 *
960 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
961 *
962 * This could be used as follows:
963 *
964 * #define __kernel_get_tls() \
965 *	({ register unsigned int __val asm("r0"); \
966 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
967 *	        : "=r" (__val) : : "lr","cc" ); \
968 *	   __val; })
969 */
970
971__kuser_get_tls:				@ 0xffff0fe0
972
973#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
974	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
975#else
976	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
977#endif
978	usr_ret	lr
979
980	.rep	5
981	.word	0			@ pad up to __kuser_helper_version
982	.endr
983
984/*
985 * Reference declaration:
986 *
987 *	extern unsigned int __kernel_helper_version;
988 *
989 * Definition and user space usage example:
990 *
991 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
992 *
993 * User space may read this to determine the curent number of helpers
994 * available.
995 */
996
997__kuser_helper_version:				@ 0xffff0ffc
998	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
999
1000	.globl	__kuser_helper_end
1001__kuser_helper_end:
1002
1003
1004/*
1005 * Vector stubs.
1006 *
1007 * This code is copied to 0xffff0200 so we can use branches in the
1008 * vectors, rather than ldr's.  Note that this code must not
1009 * exceed 0x300 bytes.
1010 *
1011 * Common stub entry macro:
1012 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1013 *
1014 * SP points to a minimal amount of processor-private memory, the address
1015 * of which is copied into r0 for the mode specific abort handler.
1016 */
1017	.macro	vector_stub, name, mode, correction=0
1018	.align	5
1019
1020vector_\name:
1021	.if \correction
1022	sub	lr, lr, #\correction
1023	.endif
1024
1025	@
1026	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
1027	@ (parent CPSR)
1028	@
1029	stmia	sp, {r0, lr}		@ save r0, lr
1030	mrs	lr, spsr
1031	str	lr, [sp, #8]		@ save spsr
1032
1033	@
1034	@ Prepare for SVC32 mode.  IRQs remain disabled.
1035	@
1036	mrs	r0, cpsr
1037	eor	r0, r0, #(\mode ^ SVC_MODE)
1038	msr	spsr_cxsf, r0
1039
1040	@
1041	@ the branch table must immediately follow this code
1042	@
1043	and	lr, lr, #0x0f
1044	mov	r0, sp
1045	ldr	lr, [pc, lr, lsl #2]
1046	movs	pc, lr			@ branch to handler in SVC mode
1047ENDPROC(vector_\name)
1048	.endm
1049
1050	.globl	__stubs_start
1051__stubs_start:
1052/*
1053 * Interrupt dispatcher
1054 */
1055	vector_stub	irq, IRQ_MODE, 4
1056
1057	.long	__irq_usr			@  0  (USR_26 / USR_32)
1058	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
1059	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
1060	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
1061	.long	__irq_invalid			@  4
1062	.long	__irq_invalid			@  5
1063	.long	__irq_invalid			@  6
1064	.long	__irq_invalid			@  7
1065	.long	__irq_invalid			@  8
1066	.long	__irq_invalid			@  9
1067	.long	__irq_invalid			@  a
1068	.long	__irq_invalid			@  b
1069	.long	__irq_invalid			@  c
1070	.long	__irq_invalid			@  d
1071	.long	__irq_invalid			@  e
1072	.long	__irq_invalid			@  f
1073
1074/*
1075 * Data abort dispatcher
1076 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1077 */
1078	vector_stub	dabt, ABT_MODE, 8
1079
1080	.long	__dabt_usr			@  0  (USR_26 / USR_32)
1081	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
1082	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
1083	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
1084	.long	__dabt_invalid			@  4
1085	.long	__dabt_invalid			@  5
1086	.long	__dabt_invalid			@  6
1087	.long	__dabt_invalid			@  7
1088	.long	__dabt_invalid			@  8
1089	.long	__dabt_invalid			@  9
1090	.long	__dabt_invalid			@  a
1091	.long	__dabt_invalid			@  b
1092	.long	__dabt_invalid			@  c
1093	.long	__dabt_invalid			@  d
1094	.long	__dabt_invalid			@  e
1095	.long	__dabt_invalid			@  f
1096
1097/*
1098 * Prefetch abort dispatcher
1099 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1100 */
1101	vector_stub	pabt, ABT_MODE, 4
1102
1103	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1104	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1105	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1106	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1107	.long	__pabt_invalid			@  4
1108	.long	__pabt_invalid			@  5
1109	.long	__pabt_invalid			@  6
1110	.long	__pabt_invalid			@  7
1111	.long	__pabt_invalid			@  8
1112	.long	__pabt_invalid			@  9
1113	.long	__pabt_invalid			@  a
1114	.long	__pabt_invalid			@  b
1115	.long	__pabt_invalid			@  c
1116	.long	__pabt_invalid			@  d
1117	.long	__pabt_invalid			@  e
1118	.long	__pabt_invalid			@  f
1119
1120/*
1121 * Undef instr entry dispatcher
1122 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1123 */
1124	vector_stub	und, UND_MODE
1125
1126	.long	__und_usr			@  0 (USR_26 / USR_32)
1127	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1128	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1129	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1130	.long	__und_invalid			@  4
1131	.long	__und_invalid			@  5
1132	.long	__und_invalid			@  6
1133	.long	__und_invalid			@  7
1134	.long	__und_invalid			@  8
1135	.long	__und_invalid			@  9
1136	.long	__und_invalid			@  a
1137	.long	__und_invalid			@  b
1138	.long	__und_invalid			@  c
1139	.long	__und_invalid			@  d
1140	.long	__und_invalid			@  e
1141	.long	__und_invalid			@  f
1142
1143	.align	5
1144
1145/*=============================================================================
1146 * Undefined FIQs
1147 *-----------------------------------------------------------------------------
1148 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1149 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1150 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1151 * damage alert!  I don't think that we can execute any code in here in any
1152 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1153 * get out of that mode without clobbering one register.
1154 */
1155vector_fiq:
1156	disable_fiq
1157	subs	pc, lr, #4
1158
1159/*=============================================================================
1160 * Address exception handler
1161 *-----------------------------------------------------------------------------
1162 * These aren't too critical.
1163 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1164 */
1165
1166vector_addrexcptn:
1167	b	vector_addrexcptn
1168
1169/*
1170 * We group all the following data together to optimise
1171 * for CPUs with separate I & D caches.
1172 */
1173	.align	5
1174
1175.LCvswi:
1176	.word	vector_swi
1177
1178	.globl	__stubs_end
1179__stubs_end:
1180
1181	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1182
1183	.globl	__vectors_start
1184__vectors_start:
1185	swi	SYS_ERROR0
1186	b	vector_und + stubs_offset
1187	ldr	pc, .LCvswi + stubs_offset
1188	b	vector_pabt + stubs_offset
1189	b	vector_dabt + stubs_offset
1190	b	vector_addrexcptn + stubs_offset
1191	b	vector_irq + stubs_offset
1192	b	vector_fiq + stubs_offset
1193
1194	.globl	__vectors_end
1195__vectors_end:
1196
1197	.data
1198
1199	.globl	cr_alignment
1200	.globl	cr_no_alignment
1201cr_alignment:
1202	.space	4
1203cr_no_alignment:
1204	.space	4
1205