xref: /openbmc/linux/arch/arm/kernel/entry-armv.S (revision 545e4006)
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 <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#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
79
80__dabt_invalid:
81	inv_entry BAD_DATA
82	b	common_invalid
83
84__irq_invalid:
85	inv_entry BAD_IRQ
86	b	common_invalid
87
88__und_invalid:
89	inv_entry BAD_UNDEFINSTR
90
91	@
92	@ XXX fall through to common_invalid
93	@
94
95@
96@ common_invalid - generic code for failed exception (re-entrant version of handlers)
97@
98common_invalid:
99	zero_fp
100
101	ldmia	r0, {r4 - r6}
102	add	r0, sp, #S_PC		@ here for interlock avoidance
103	mov	r7, #-1			@  ""   ""    ""        ""
104	str	r4, [sp]		@ save preserved r0
105	stmia	r0, {r5 - r7}		@ lr_<exception>,
106					@ cpsr_<exception>, "old_r0"
107
108	mov	r0, sp
109	b	bad_mode
110
111/*
112 * SVC mode handlers
113 */
114
115#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5)
116#define SPFIX(code...) code
117#else
118#define SPFIX(code...)
119#endif
120
121	.macro	svc_entry, stack_hole=0
122	sub	sp, sp, #(S_FRAME_SIZE + \stack_hole)
123 SPFIX(	tst	sp, #4		)
124 SPFIX(	bicne	sp, sp, #4	)
125	stmib	sp, {r1 - r12}
126
127	ldmia	r0, {r1 - r3}
128	add	r5, sp, #S_SP		@ here for interlock avoidance
129	mov	r4, #-1			@  ""  ""      ""       ""
130	add	r0, sp, #(S_FRAME_SIZE + \stack_hole)
131 SPFIX(	addne	r0, r0, #4	)
132	str	r1, [sp]		@ save the "real" r0 copied
133					@ from the exception stack
134
135	mov	r1, lr
136
137	@
138	@ We are now ready to fill in the remaining blanks on the stack:
139	@
140	@  r0 - sp_svc
141	@  r1 - lr_svc
142	@  r2 - lr_<exception>, already fixed up for correct return/restart
143	@  r3 - spsr_<exception>
144	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
145	@
146	stmia	r5, {r0 - r4}
147	.endm
148
149	.align	5
150__dabt_svc:
151	svc_entry
152
153	@
154	@ get ready to re-enable interrupts if appropriate
155	@
156	mrs	r9, cpsr
157	tst	r3, #PSR_I_BIT
158	biceq	r9, r9, #PSR_I_BIT
159
160	@
161	@ Call the processor-specific abort handler:
162	@
163	@  r2 - aborted context pc
164	@  r3 - aborted context cpsr
165	@
166	@ The abort handler must return the aborted address in r0, and
167	@ the fault status register in r1.  r9 must be preserved.
168	@
169#ifdef MULTI_DABORT
170	ldr	r4, .LCprocfns
171	mov	lr, pc
172	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
173#else
174	bl	CPU_DABORT_HANDLER
175#endif
176
177	@
178	@ set desired IRQ state, then call main handler
179	@
180	msr	cpsr_c, r9
181	mov	r2, sp
182	bl	do_DataAbort
183
184	@
185	@ IRQs off again before pulling preserved data off the stack
186	@
187	disable_irq
188
189	@
190	@ restore SPSR and restart the instruction
191	@
192	ldr	r0, [sp, #S_PSR]
193	msr	spsr_cxsf, r0
194	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
195
196	.align	5
197__irq_svc:
198	svc_entry
199
200#ifdef CONFIG_TRACE_IRQFLAGS
201	bl	trace_hardirqs_off
202#endif
203#ifdef CONFIG_PREEMPT
204	get_thread_info tsk
205	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
206	add	r7, r8, #1			@ increment it
207	str	r7, [tsk, #TI_PREEMPT]
208#endif
209
210	irq_handler
211#ifdef CONFIG_PREEMPT
212	str	r8, [tsk, #TI_PREEMPT]		@ restore preempt count
213	ldr	r0, [tsk, #TI_FLAGS]		@ get flags
214	teq	r8, #0				@ if preempt count != 0
215	movne	r0, #0				@ force flags to 0
216	tst	r0, #_TIF_NEED_RESCHED
217	blne	svc_preempt
218#endif
219	ldr	r0, [sp, #S_PSR]		@ irqs are already disabled
220	msr	spsr_cxsf, r0
221#ifdef CONFIG_TRACE_IRQFLAGS
222	tst	r0, #PSR_I_BIT
223	bleq	trace_hardirqs_on
224#endif
225	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
226
227	.ltorg
228
229#ifdef CONFIG_PREEMPT
230svc_preempt:
231	mov	r8, lr
2321:	bl	preempt_schedule_irq		@ irq en/disable is done inside
233	ldr	r0, [tsk, #TI_FLAGS]		@ get new tasks TI_FLAGS
234	tst	r0, #_TIF_NEED_RESCHED
235	moveq	pc, r8				@ go again
236	b	1b
237#endif
238
239	.align	5
240__und_svc:
241#ifdef CONFIG_KPROBES
242	@ If a kprobe is about to simulate a "stmdb sp..." instruction,
243	@ it obviously needs free stack space which then will belong to
244	@ the saved context.
245	svc_entry 64
246#else
247	svc_entry
248#endif
249
250	@
251	@ call emulation code, which returns using r9 if it has emulated
252	@ the instruction, or the more conventional lr if we are to treat
253	@ this as a real undefined instruction
254	@
255	@  r0 - instruction
256	@
257	ldr	r0, [r2, #-4]
258	adr	r9, 1f
259	bl	call_fpe
260
261	mov	r0, sp				@ struct pt_regs *regs
262	bl	do_undefinstr
263
264	@
265	@ IRQs off again before pulling preserved data off the stack
266	@
2671:	disable_irq
268
269	@
270	@ restore SPSR and restart the instruction
271	@
272	ldr	lr, [sp, #S_PSR]		@ Get SVC cpsr
273	msr	spsr_cxsf, lr
274	ldmia	sp, {r0 - pc}^			@ Restore SVC registers
275
276	.align	5
277__pabt_svc:
278	svc_entry
279
280	@
281	@ re-enable interrupts if appropriate
282	@
283	mrs	r9, cpsr
284	tst	r3, #PSR_I_BIT
285	biceq	r9, r9, #PSR_I_BIT
286
287	@
288	@ set args, then call main handler
289	@
290	@  r0 - address of faulting instruction
291	@  r1 - pointer to registers on stack
292	@
293#ifdef MULTI_PABORT
294	mov	r0, r2			@ pass address of aborted instruction.
295	ldr	r4, .LCprocfns
296	mov	lr, pc
297	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
298#else
299	CPU_PABORT_HANDLER(r0, r2)
300#endif
301	msr	cpsr_c, r9			@ Maybe enable interrupts
302	mov	r1, sp				@ regs
303	bl	do_PrefetchAbort		@ call abort handler
304
305	@
306	@ IRQs off again before pulling preserved data off the stack
307	@
308	disable_irq
309
310	@
311	@ restore SPSR and restart the instruction
312	@
313	ldr	r0, [sp, #S_PSR]
314	msr	spsr_cxsf, r0
315	ldmia	sp, {r0 - pc}^			@ load r0 - pc, cpsr
316
317	.align	5
318.LCcralign:
319	.word	cr_alignment
320#ifdef MULTI_DABORT
321.LCprocfns:
322	.word	processor
323#endif
324.LCfp:
325	.word	fp_enter
326
327/*
328 * User mode handlers
329 *
330 * EABI note: sp_svc is always 64-bit aligned here, so should S_FRAME_SIZE
331 */
332
333#if defined(CONFIG_AEABI) && (__LINUX_ARM_ARCH__ >= 5) && (S_FRAME_SIZE & 7)
334#error "sizeof(struct pt_regs) must be a multiple of 8"
335#endif
336
337	.macro	usr_entry
338	sub	sp, sp, #S_FRAME_SIZE
339	stmib	sp, {r1 - r12}
340
341	ldmia	r0, {r1 - r3}
342	add	r0, sp, #S_PC		@ here for interlock avoidance
343	mov	r4, #-1			@  ""  ""     ""        ""
344
345	str	r1, [sp]		@ save the "real" r0 copied
346					@ from the exception stack
347
348	@
349	@ We are now ready to fill in the remaining blanks on the stack:
350	@
351	@  r2 - lr_<exception>, already fixed up for correct return/restart
352	@  r3 - spsr_<exception>
353	@  r4 - orig_r0 (see pt_regs definition in ptrace.h)
354	@
355	@ Also, separately save sp_usr and lr_usr
356	@
357	stmia	r0, {r2 - r4}
358	stmdb	r0, {sp, lr}^
359
360	@
361	@ Enable the alignment trap while in kernel mode
362	@
363	alignment_trap r0
364
365	@
366	@ Clear FP to mark the first stack frame
367	@
368	zero_fp
369	.endm
370
371	.macro	kuser_cmpxchg_check
372#if __LINUX_ARM_ARCH__ < 6 && !defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
373#ifndef CONFIG_MMU
374#warning "NPTL on non MMU needs fixing"
375#else
376	@ Make sure our user space atomic helper is restarted
377	@ if it was interrupted in a critical region.  Here we
378	@ perform a quick test inline since it should be false
379	@ 99.9999% of the time.  The rest is done out of line.
380	cmp	r2, #TASK_SIZE
381	blhs	kuser_cmpxchg_fixup
382#endif
383#endif
384	.endm
385
386	.align	5
387__dabt_usr:
388	usr_entry
389	kuser_cmpxchg_check
390
391	@
392	@ Call the processor-specific abort handler:
393	@
394	@  r2 - aborted context pc
395	@  r3 - aborted context cpsr
396	@
397	@ The abort handler must return the aborted address in r0, and
398	@ the fault status register in r1.
399	@
400#ifdef MULTI_DABORT
401	ldr	r4, .LCprocfns
402	mov	lr, pc
403	ldr	pc, [r4, #PROCESSOR_DABT_FUNC]
404#else
405	bl	CPU_DABORT_HANDLER
406#endif
407
408	@
409	@ IRQs on, then call the main handler
410	@
411	enable_irq
412	mov	r2, sp
413	adr	lr, ret_from_exception
414	b	do_DataAbort
415
416	.align	5
417__irq_usr:
418	usr_entry
419	kuser_cmpxchg_check
420
421#ifdef CONFIG_TRACE_IRQFLAGS
422	bl	trace_hardirqs_off
423#endif
424	get_thread_info tsk
425#ifdef CONFIG_PREEMPT
426	ldr	r8, [tsk, #TI_PREEMPT]		@ get preempt count
427	add	r7, r8, #1			@ increment it
428	str	r7, [tsk, #TI_PREEMPT]
429#endif
430
431	irq_handler
432#ifdef CONFIG_PREEMPT
433	ldr	r0, [tsk, #TI_PREEMPT]
434	str	r8, [tsk, #TI_PREEMPT]
435	teq	r0, r7
436	strne	r0, [r0, -r0]
437#endif
438#ifdef CONFIG_TRACE_IRQFLAGS
439	bl	trace_hardirqs_on
440#endif
441
442	mov	why, #0
443	b	ret_to_user
444
445	.ltorg
446
447	.align	5
448__und_usr:
449	usr_entry
450
451	@
452	@ fall through to the emulation code, which returns using r9 if
453	@ it has emulated the instruction, or the more conventional lr
454	@ if we are to treat this as a real undefined instruction
455	@
456	@  r0 - instruction
457	@
458	adr	r9, ret_from_exception
459	adr	lr, __und_usr_unknown
460	tst	r3, #PSR_T_BIT			@ Thumb mode?
461	subeq	r4, r2, #4			@ ARM instr at LR - 4
462	subne	r4, r2, #2			@ Thumb instr at LR - 2
4631:	ldreqt	r0, [r4]
464	beq	call_fpe
465	@ Thumb instruction
466#if __LINUX_ARM_ARCH__ >= 7
4672:	ldrht	r5, [r4], #2
468	and	r0, r5, #0xf800			@ mask bits 111x x... .... ....
469	cmp	r0, #0xe800			@ 32bit instruction if xx != 0
470	blo	__und_usr_unknown
4713:	ldrht	r0, [r4]
472	add	r2, r2, #2			@ r2 is PC + 2, make it PC + 4
473	orr	r0, r0, r5, lsl #16
474#else
475	b	__und_usr_unknown
476#endif
477
478	@
479	@ fallthrough to call_fpe
480	@
481
482/*
483 * The out of line fixup for the ldrt above.
484 */
485	.section .fixup, "ax"
4864:	mov	pc, r9
487	.previous
488	.section __ex_table,"a"
489	.long	1b, 4b
490#if __LINUX_ARM_ARCH__ >= 7
491	.long	2b, 4b
492	.long	3b, 4b
493#endif
494	.previous
495
496/*
497 * Check whether the instruction is a co-processor instruction.
498 * If yes, we need to call the relevant co-processor handler.
499 *
500 * Note that we don't do a full check here for the co-processor
501 * instructions; all instructions with bit 27 set are well
502 * defined.  The only instructions that should fault are the
503 * co-processor instructions.  However, we have to watch out
504 * for the ARM6/ARM7 SWI bug.
505 *
506 * NEON is a special case that has to be handled here. Not all
507 * NEON instructions are co-processor instructions, so we have
508 * to make a special case of checking for them. Plus, there's
509 * five groups of them, so we have a table of mask/opcode pairs
510 * to check against, and if any match then we branch off into the
511 * NEON handler code.
512 *
513 * Emulators may wish to make use of the following registers:
514 *  r0  = instruction opcode.
515 *  r2  = PC+4
516 *  r9  = normal "successful" return address
517 *  r10 = this threads thread_info structure.
518 *  lr  = unrecognised instruction return address
519 */
520	@
521	@ Fall-through from Thumb-2 __und_usr
522	@
523#ifdef CONFIG_NEON
524	adr	r6, .LCneon_thumb_opcodes
525	b	2f
526#endif
527call_fpe:
528#ifdef CONFIG_NEON
529	adr	r6, .LCneon_arm_opcodes
5302:
531	ldr	r7, [r6], #4			@ mask value
532	cmp	r7, #0				@ end mask?
533	beq	1f
534	and	r8, r0, r7
535	ldr	r7, [r6], #4			@ opcode bits matching in mask
536	cmp	r8, r7				@ NEON instruction?
537	bne	2b
538	get_thread_info r10
539	mov	r7, #1
540	strb	r7, [r10, #TI_USED_CP + 10]	@ mark CP#10 as used
541	strb	r7, [r10, #TI_USED_CP + 11]	@ mark CP#11 as used
542	b	do_vfp				@ let VFP handler handle this
5431:
544#endif
545	tst	r0, #0x08000000			@ only CDP/CPRT/LDC/STC have bit 27
546	tstne	r0, #0x04000000			@ bit 26 set on both ARM and Thumb-2
547#if defined(CONFIG_CPU_ARM610) || defined(CONFIG_CPU_ARM710)
548	and	r8, r0, #0x0f000000		@ mask out op-code bits
549	teqne	r8, #0x0f000000			@ SWI (ARM6/7 bug)?
550#endif
551	moveq	pc, lr
552	get_thread_info r10			@ get current thread
553	and	r8, r0, #0x00000f00		@ mask out CP number
554	mov	r7, #1
555	add	r6, r10, #TI_USED_CP
556	strb	r7, [r6, r8, lsr #8]		@ set appropriate used_cp[]
557#ifdef CONFIG_IWMMXT
558	@ Test if we need to give access to iWMMXt coprocessors
559	ldr	r5, [r10, #TI_FLAGS]
560	rsbs	r7, r8, #(1 << 8)		@ CP 0 or 1 only
561	movcss	r7, r5, lsr #(TIF_USING_IWMMXT + 1)
562	bcs	iwmmxt_task_enable
563#endif
564	add	pc, pc, r8, lsr #6
565	mov	r0, r0
566
567	mov	pc, lr				@ CP#0
568	b	do_fpe				@ CP#1 (FPE)
569	b	do_fpe				@ CP#2 (FPE)
570	mov	pc, lr				@ CP#3
571#ifdef CONFIG_CRUNCH
572	b	crunch_task_enable		@ CP#4 (MaverickCrunch)
573	b	crunch_task_enable		@ CP#5 (MaverickCrunch)
574	b	crunch_task_enable		@ CP#6 (MaverickCrunch)
575#else
576	mov	pc, lr				@ CP#4
577	mov	pc, lr				@ CP#5
578	mov	pc, lr				@ CP#6
579#endif
580	mov	pc, lr				@ CP#7
581	mov	pc, lr				@ CP#8
582	mov	pc, lr				@ CP#9
583#ifdef CONFIG_VFP
584	b	do_vfp				@ CP#10 (VFP)
585	b	do_vfp				@ CP#11 (VFP)
586#else
587	mov	pc, lr				@ CP#10 (VFP)
588	mov	pc, lr				@ CP#11 (VFP)
589#endif
590	mov	pc, lr				@ CP#12
591	mov	pc, lr				@ CP#13
592	mov	pc, lr				@ CP#14 (Debug)
593	mov	pc, lr				@ CP#15 (Control)
594
595#ifdef CONFIG_NEON
596	.align	6
597
598.LCneon_arm_opcodes:
599	.word	0xfe000000			@ mask
600	.word	0xf2000000			@ opcode
601
602	.word	0xff100000			@ mask
603	.word	0xf4000000			@ opcode
604
605	.word	0x00000000			@ mask
606	.word	0x00000000			@ opcode
607
608.LCneon_thumb_opcodes:
609	.word	0xef000000			@ mask
610	.word	0xef000000			@ opcode
611
612	.word	0xff100000			@ mask
613	.word	0xf9000000			@ opcode
614
615	.word	0x00000000			@ mask
616	.word	0x00000000			@ opcode
617#endif
618
619do_fpe:
620	enable_irq
621	ldr	r4, .LCfp
622	add	r10, r10, #TI_FPSTATE		@ r10 = workspace
623	ldr	pc, [r4]			@ Call FP module USR entry point
624
625/*
626 * The FP module is called with these registers set:
627 *  r0  = instruction
628 *  r2  = PC+4
629 *  r9  = normal "successful" return address
630 *  r10 = FP workspace
631 *  lr  = unrecognised FP instruction return address
632 */
633
634	.data
635ENTRY(fp_enter)
636	.word	no_fp
637	.previous
638
639no_fp:	mov	pc, lr
640
641__und_usr_unknown:
642	mov	r0, sp
643	adr	lr, ret_from_exception
644	b	do_undefinstr
645
646	.align	5
647__pabt_usr:
648	usr_entry
649
650#ifdef MULTI_PABORT
651	mov	r0, r2			@ pass address of aborted instruction.
652	ldr	r4, .LCprocfns
653	mov	lr, pc
654	ldr	pc, [r4, #PROCESSOR_PABT_FUNC]
655#else
656	CPU_PABORT_HANDLER(r0, r2)
657#endif
658	enable_irq				@ Enable interrupts
659	mov	r1, sp				@ regs
660	bl	do_PrefetchAbort		@ call abort handler
661	/* fall through */
662/*
663 * This is the return code to user mode for abort handlers
664 */
665ENTRY(ret_from_exception)
666	get_thread_info tsk
667	mov	why, #0
668	b	ret_to_user
669
670/*
671 * Register switch for ARMv3 and ARMv4 processors
672 * r0 = previous task_struct, r1 = previous thread_info, r2 = next thread_info
673 * previous and next are guaranteed not to be the same.
674 */
675ENTRY(__switch_to)
676	add	ip, r1, #TI_CPU_SAVE
677	ldr	r3, [r2, #TI_TP_VALUE]
678	stmia	ip!, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
679#ifdef CONFIG_MMU
680	ldr	r6, [r2, #TI_CPU_DOMAIN]
681#endif
682#if __LINUX_ARM_ARCH__ >= 6
683#ifdef CONFIG_CPU_32v6K
684	clrex
685#else
686	strex	r5, r4, [ip]			@ Clear exclusive monitor
687#endif
688#endif
689#if defined(CONFIG_HAS_TLS_REG)
690	mcr	p15, 0, r3, c13, c0, 3		@ set TLS register
691#elif !defined(CONFIG_TLS_REG_EMUL)
692	mov	r4, #0xffff0fff
693	str	r3, [r4, #-15]			@ TLS val at 0xffff0ff0
694#endif
695#ifdef CONFIG_MMU
696	mcr	p15, 0, r6, c3, c0, 0		@ Set domain register
697#endif
698	mov	r5, r0
699	add	r4, r2, #TI_CPU_SAVE
700	ldr	r0, =thread_notify_head
701	mov	r1, #THREAD_NOTIFY_SWITCH
702	bl	atomic_notifier_call_chain
703	mov	r0, r5
704	ldmia	r4, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
705
706	__INIT
707
708/*
709 * User helpers.
710 *
711 * These are segment of kernel provided user code reachable from user space
712 * at a fixed address in kernel memory.  This is used to provide user space
713 * with some operations which require kernel help because of unimplemented
714 * native feature and/or instructions in many ARM CPUs. The idea is for
715 * this code to be executed directly in user mode for best efficiency but
716 * which is too intimate with the kernel counter part to be left to user
717 * libraries.  In fact this code might even differ from one CPU to another
718 * depending on the available  instruction set and restrictions like on
719 * SMP systems.  In other words, the kernel reserves the right to change
720 * this code as needed without warning. Only the entry points and their
721 * results are guaranteed to be stable.
722 *
723 * Each segment is 32-byte aligned and will be moved to the top of the high
724 * vector page.  New segments (if ever needed) must be added in front of
725 * existing ones.  This mechanism should be used only for things that are
726 * really small and justified, and not be abused freely.
727 *
728 * User space is expected to implement those things inline when optimizing
729 * for a processor that has the necessary native support, but only if such
730 * resulting binaries are already to be incompatible with earlier ARM
731 * processors due to the use of unsupported instructions other than what
732 * is provided here.  In other words don't make binaries unable to run on
733 * earlier processors just for the sake of not using these kernel helpers
734 * if your compiled code is not going to use the new instructions for other
735 * purpose.
736 */
737
738	.macro	usr_ret, reg
739#ifdef CONFIG_ARM_THUMB
740	bx	\reg
741#else
742	mov	pc, \reg
743#endif
744	.endm
745
746	.align	5
747	.globl	__kuser_helper_start
748__kuser_helper_start:
749
750/*
751 * Reference prototype:
752 *
753 *	void __kernel_memory_barrier(void)
754 *
755 * Input:
756 *
757 *	lr = return address
758 *
759 * Output:
760 *
761 *	none
762 *
763 * Clobbered:
764 *
765 *	none
766 *
767 * Definition and user space usage example:
768 *
769 *	typedef void (__kernel_dmb_t)(void);
770 *	#define __kernel_dmb (*(__kernel_dmb_t *)0xffff0fa0)
771 *
772 * Apply any needed memory barrier to preserve consistency with data modified
773 * manually and __kuser_cmpxchg usage.
774 *
775 * This could be used as follows:
776 *
777 * #define __kernel_dmb() \
778 *         asm volatile ( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #95" \
779 *	        : : : "r0", "lr","cc" )
780 */
781
782__kuser_memory_barrier:				@ 0xffff0fa0
783
784#if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
785	mcr	p15, 0, r0, c7, c10, 5	@ dmb
786#endif
787	usr_ret	lr
788
789	.align	5
790
791/*
792 * Reference prototype:
793 *
794 *	int __kernel_cmpxchg(int oldval, int newval, int *ptr)
795 *
796 * Input:
797 *
798 *	r0 = oldval
799 *	r1 = newval
800 *	r2 = ptr
801 *	lr = return address
802 *
803 * Output:
804 *
805 *	r0 = returned value (zero or non-zero)
806 *	C flag = set if r0 == 0, clear if r0 != 0
807 *
808 * Clobbered:
809 *
810 *	r3, ip, flags
811 *
812 * Definition and user space usage example:
813 *
814 *	typedef int (__kernel_cmpxchg_t)(int oldval, int newval, int *ptr);
815 *	#define __kernel_cmpxchg (*(__kernel_cmpxchg_t *)0xffff0fc0)
816 *
817 * Atomically store newval in *ptr if *ptr is equal to oldval for user space.
818 * Return zero if *ptr was changed or non-zero if no exchange happened.
819 * The C flag is also set if *ptr was changed to allow for assembly
820 * optimization in the calling code.
821 *
822 * Notes:
823 *
824 *    - This routine already includes memory barriers as needed.
825 *
826 * For example, a user space atomic_add implementation could look like this:
827 *
828 * #define atomic_add(ptr, val) \
829 *	({ register unsigned int *__ptr asm("r2") = (ptr); \
830 *	   register unsigned int __result asm("r1"); \
831 *	   asm volatile ( \
832 *	       "1: @ atomic_add\n\t" \
833 *	       "ldr	r0, [r2]\n\t" \
834 *	       "mov	r3, #0xffff0fff\n\t" \
835 *	       "add	lr, pc, #4\n\t" \
836 *	       "add	r1, r0, %2\n\t" \
837 *	       "add	pc, r3, #(0xffff0fc0 - 0xffff0fff)\n\t" \
838 *	       "bcc	1b" \
839 *	       : "=&r" (__result) \
840 *	       : "r" (__ptr), "rIL" (val) \
841 *	       : "r0","r3","ip","lr","cc","memory" ); \
842 *	   __result; })
843 */
844
845__kuser_cmpxchg:				@ 0xffff0fc0
846
847#if defined(CONFIG_NEEDS_SYSCALL_FOR_CMPXCHG)
848
849	/*
850	 * Poor you.  No fast solution possible...
851	 * The kernel itself must perform the operation.
852	 * A special ghost syscall is used for that (see traps.c).
853	 */
854	stmfd	sp!, {r7, lr}
855	mov	r7, #0xff00		@ 0xfff0 into r7 for EABI
856	orr	r7, r7, #0xf0
857	swi	#0x9ffff0
858	ldmfd	sp!, {r7, pc}
859
860#elif __LINUX_ARM_ARCH__ < 6
861
862#ifdef CONFIG_MMU
863
864	/*
865	 * The only thing that can break atomicity in this cmpxchg
866	 * implementation is either an IRQ or a data abort exception
867	 * causing another process/thread to be scheduled in the middle
868	 * of the critical sequence.  To prevent this, code is added to
869	 * the IRQ and data abort exception handlers to set the pc back
870	 * to the beginning of the critical section if it is found to be
871	 * within that critical section (see kuser_cmpxchg_fixup).
872	 */
8731:	ldr	r3, [r2]			@ load current val
874	subs	r3, r3, r0			@ compare with oldval
8752:	streq	r1, [r2]			@ store newval if eq
876	rsbs	r0, r3, #0			@ set return val and C flag
877	usr_ret	lr
878
879	.text
880kuser_cmpxchg_fixup:
881	@ Called from kuser_cmpxchg_check macro.
882	@ r2 = address of interrupted insn (must be preserved).
883	@ sp = saved regs. r7 and r8 are clobbered.
884	@ 1b = first critical insn, 2b = last critical insn.
885	@ If r2 >= 1b and r2 <= 2b then saved pc_usr is set to 1b.
886	mov	r7, #0xffff0fff
887	sub	r7, r7, #(0xffff0fff - (0xffff0fc0 + (1b - __kuser_cmpxchg)))
888	subs	r8, r2, r7
889	rsbcss	r8, r8, #(2b - 1b)
890	strcs	r7, [sp, #S_PC]
891	mov	pc, lr
892	.previous
893
894#else
895#warning "NPTL on non MMU needs fixing"
896	mov	r0, #-1
897	adds	r0, r0, #0
898	usr_ret	lr
899#endif
900
901#else
902
903#ifdef CONFIG_SMP
904	mcr	p15, 0, r0, c7, c10, 5	@ dmb
905#endif
9061:	ldrex	r3, [r2]
907	subs	r3, r3, r0
908	strexeq	r3, r1, [r2]
909	teqeq	r3, #1
910	beq	1b
911	rsbs	r0, r3, #0
912	/* beware -- each __kuser slot must be 8 instructions max */
913#ifdef CONFIG_SMP
914	b	__kuser_memory_barrier
915#else
916	usr_ret	lr
917#endif
918
919#endif
920
921	.align	5
922
923/*
924 * Reference prototype:
925 *
926 *	int __kernel_get_tls(void)
927 *
928 * Input:
929 *
930 *	lr = return address
931 *
932 * Output:
933 *
934 *	r0 = TLS value
935 *
936 * Clobbered:
937 *
938 *	none
939 *
940 * Definition and user space usage example:
941 *
942 *	typedef int (__kernel_get_tls_t)(void);
943 *	#define __kernel_get_tls (*(__kernel_get_tls_t *)0xffff0fe0)
944 *
945 * Get the TLS value as previously set via the __ARM_NR_set_tls syscall.
946 *
947 * This could be used as follows:
948 *
949 * #define __kernel_get_tls() \
950 *	({ register unsigned int __val asm("r0"); \
951 *         asm( "mov r0, #0xffff0fff; mov lr, pc; sub pc, r0, #31" \
952 *	        : "=r" (__val) : : "lr","cc" ); \
953 *	   __val; })
954 */
955
956__kuser_get_tls:				@ 0xffff0fe0
957
958#if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
959	ldr	r0, [pc, #(16 - 8)]		@ TLS stored at 0xffff0ff0
960#else
961	mrc	p15, 0, r0, c13, c0, 3		@ read TLS register
962#endif
963	usr_ret	lr
964
965	.rep	5
966	.word	0			@ pad up to __kuser_helper_version
967	.endr
968
969/*
970 * Reference declaration:
971 *
972 *	extern unsigned int __kernel_helper_version;
973 *
974 * Definition and user space usage example:
975 *
976 *	#define __kernel_helper_version (*(unsigned int *)0xffff0ffc)
977 *
978 * User space may read this to determine the curent number of helpers
979 * available.
980 */
981
982__kuser_helper_version:				@ 0xffff0ffc
983	.word	((__kuser_helper_end - __kuser_helper_start) >> 5)
984
985	.globl	__kuser_helper_end
986__kuser_helper_end:
987
988
989/*
990 * Vector stubs.
991 *
992 * This code is copied to 0xffff0200 so we can use branches in the
993 * vectors, rather than ldr's.  Note that this code must not
994 * exceed 0x300 bytes.
995 *
996 * Common stub entry macro:
997 *   Enter in IRQ mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
998 *
999 * SP points to a minimal amount of processor-private memory, the address
1000 * of which is copied into r0 for the mode specific abort handler.
1001 */
1002	.macro	vector_stub, name, mode, correction=0
1003	.align	5
1004
1005vector_\name:
1006	.if \correction
1007	sub	lr, lr, #\correction
1008	.endif
1009
1010	@
1011	@ Save r0, lr_<exception> (parent PC) and spsr_<exception>
1012	@ (parent CPSR)
1013	@
1014	stmia	sp, {r0, lr}		@ save r0, lr
1015	mrs	lr, spsr
1016	str	lr, [sp, #8]		@ save spsr
1017
1018	@
1019	@ Prepare for SVC32 mode.  IRQs remain disabled.
1020	@
1021	mrs	r0, cpsr
1022	eor	r0, r0, #(\mode ^ SVC_MODE)
1023	msr	spsr_cxsf, r0
1024
1025	@
1026	@ the branch table must immediately follow this code
1027	@
1028	and	lr, lr, #0x0f
1029	mov	r0, sp
1030	ldr	lr, [pc, lr, lsl #2]
1031	movs	pc, lr			@ branch to handler in SVC mode
1032	.endm
1033
1034	.globl	__stubs_start
1035__stubs_start:
1036/*
1037 * Interrupt dispatcher
1038 */
1039	vector_stub	irq, IRQ_MODE, 4
1040
1041	.long	__irq_usr			@  0  (USR_26 / USR_32)
1042	.long	__irq_invalid			@  1  (FIQ_26 / FIQ_32)
1043	.long	__irq_invalid			@  2  (IRQ_26 / IRQ_32)
1044	.long	__irq_svc			@  3  (SVC_26 / SVC_32)
1045	.long	__irq_invalid			@  4
1046	.long	__irq_invalid			@  5
1047	.long	__irq_invalid			@  6
1048	.long	__irq_invalid			@  7
1049	.long	__irq_invalid			@  8
1050	.long	__irq_invalid			@  9
1051	.long	__irq_invalid			@  a
1052	.long	__irq_invalid			@  b
1053	.long	__irq_invalid			@  c
1054	.long	__irq_invalid			@  d
1055	.long	__irq_invalid			@  e
1056	.long	__irq_invalid			@  f
1057
1058/*
1059 * Data abort dispatcher
1060 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1061 */
1062	vector_stub	dabt, ABT_MODE, 8
1063
1064	.long	__dabt_usr			@  0  (USR_26 / USR_32)
1065	.long	__dabt_invalid			@  1  (FIQ_26 / FIQ_32)
1066	.long	__dabt_invalid			@  2  (IRQ_26 / IRQ_32)
1067	.long	__dabt_svc			@  3  (SVC_26 / SVC_32)
1068	.long	__dabt_invalid			@  4
1069	.long	__dabt_invalid			@  5
1070	.long	__dabt_invalid			@  6
1071	.long	__dabt_invalid			@  7
1072	.long	__dabt_invalid			@  8
1073	.long	__dabt_invalid			@  9
1074	.long	__dabt_invalid			@  a
1075	.long	__dabt_invalid			@  b
1076	.long	__dabt_invalid			@  c
1077	.long	__dabt_invalid			@  d
1078	.long	__dabt_invalid			@  e
1079	.long	__dabt_invalid			@  f
1080
1081/*
1082 * Prefetch abort dispatcher
1083 * Enter in ABT mode, spsr = USR CPSR, lr = USR PC
1084 */
1085	vector_stub	pabt, ABT_MODE, 4
1086
1087	.long	__pabt_usr			@  0 (USR_26 / USR_32)
1088	.long	__pabt_invalid			@  1 (FIQ_26 / FIQ_32)
1089	.long	__pabt_invalid			@  2 (IRQ_26 / IRQ_32)
1090	.long	__pabt_svc			@  3 (SVC_26 / SVC_32)
1091	.long	__pabt_invalid			@  4
1092	.long	__pabt_invalid			@  5
1093	.long	__pabt_invalid			@  6
1094	.long	__pabt_invalid			@  7
1095	.long	__pabt_invalid			@  8
1096	.long	__pabt_invalid			@  9
1097	.long	__pabt_invalid			@  a
1098	.long	__pabt_invalid			@  b
1099	.long	__pabt_invalid			@  c
1100	.long	__pabt_invalid			@  d
1101	.long	__pabt_invalid			@  e
1102	.long	__pabt_invalid			@  f
1103
1104/*
1105 * Undef instr entry dispatcher
1106 * Enter in UND mode, spsr = SVC/USR CPSR, lr = SVC/USR PC
1107 */
1108	vector_stub	und, UND_MODE
1109
1110	.long	__und_usr			@  0 (USR_26 / USR_32)
1111	.long	__und_invalid			@  1 (FIQ_26 / FIQ_32)
1112	.long	__und_invalid			@  2 (IRQ_26 / IRQ_32)
1113	.long	__und_svc			@  3 (SVC_26 / SVC_32)
1114	.long	__und_invalid			@  4
1115	.long	__und_invalid			@  5
1116	.long	__und_invalid			@  6
1117	.long	__und_invalid			@  7
1118	.long	__und_invalid			@  8
1119	.long	__und_invalid			@  9
1120	.long	__und_invalid			@  a
1121	.long	__und_invalid			@  b
1122	.long	__und_invalid			@  c
1123	.long	__und_invalid			@  d
1124	.long	__und_invalid			@  e
1125	.long	__und_invalid			@  f
1126
1127	.align	5
1128
1129/*=============================================================================
1130 * Undefined FIQs
1131 *-----------------------------------------------------------------------------
1132 * Enter in FIQ mode, spsr = ANY CPSR, lr = ANY PC
1133 * MUST PRESERVE SVC SPSR, but need to switch to SVC mode to show our msg.
1134 * Basically to switch modes, we *HAVE* to clobber one register...  brain
1135 * damage alert!  I don't think that we can execute any code in here in any
1136 * other mode than FIQ...  Ok you can switch to another mode, but you can't
1137 * get out of that mode without clobbering one register.
1138 */
1139vector_fiq:
1140	disable_fiq
1141	subs	pc, lr, #4
1142
1143/*=============================================================================
1144 * Address exception handler
1145 *-----------------------------------------------------------------------------
1146 * These aren't too critical.
1147 * (they're not supposed to happen, and won't happen in 32-bit data mode).
1148 */
1149
1150vector_addrexcptn:
1151	b	vector_addrexcptn
1152
1153/*
1154 * We group all the following data together to optimise
1155 * for CPUs with separate I & D caches.
1156 */
1157	.align	5
1158
1159.LCvswi:
1160	.word	vector_swi
1161
1162	.globl	__stubs_end
1163__stubs_end:
1164
1165	.equ	stubs_offset, __vectors_start + 0x200 - __stubs_start
1166
1167	.globl	__vectors_start
1168__vectors_start:
1169	swi	SYS_ERROR0
1170	b	vector_und + stubs_offset
1171	ldr	pc, .LCvswi + stubs_offset
1172	b	vector_pabt + stubs_offset
1173	b	vector_dabt + stubs_offset
1174	b	vector_addrexcptn + stubs_offset
1175	b	vector_irq + stubs_offset
1176	b	vector_fiq + stubs_offset
1177
1178	.globl	__vectors_end
1179__vectors_end:
1180
1181	.data
1182
1183	.globl	cr_alignment
1184	.globl	cr_no_alignment
1185cr_alignment:
1186	.space	4
1187cr_no_alignment:
1188	.space	4
1189