xref: /linux/arch/arm/include/asm/assembler.h (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  *  arch/arm/include/asm/assembler.h
4  *
5  *  Copyright (C) 1996-2000 Russell King
6  *
7  *  This file contains arm architecture specific defines
8  *  for the different processors.
9  *
10  *  Do not include any C declarations in this file - it is included by
11  *  assembler source.
12  */
13 #ifndef __ASM_ASSEMBLER_H__
14 #define __ASM_ASSEMBLER_H__
15 
16 #ifndef __ASSEMBLY__
17 #error "Only include this from assembly code"
18 #endif
19 
20 #include <asm/ptrace.h>
21 #include <asm/opcodes-virt.h>
22 #include <asm/asm-offsets.h>
23 #include <asm/page.h>
24 #include <asm/pgtable.h>
25 #include <asm/thread_info.h>
26 #include <asm/uaccess-asm.h>
27 
28 #define IOMEM(x)	(x)
29 
30 /*
31  * Endian independent macros for shifting bytes within registers.
32  */
33 #ifndef __ARMEB__
34 #define lspull          lsr
35 #define lspush          lsl
36 #define get_byte_0      lsl #0
37 #define get_byte_1	lsr #8
38 #define get_byte_2	lsr #16
39 #define get_byte_3	lsr #24
40 #define put_byte_0      lsl #0
41 #define put_byte_1	lsl #8
42 #define put_byte_2	lsl #16
43 #define put_byte_3	lsl #24
44 #else
45 #define lspull          lsl
46 #define lspush          lsr
47 #define get_byte_0	lsr #24
48 #define get_byte_1	lsr #16
49 #define get_byte_2	lsr #8
50 #define get_byte_3      lsl #0
51 #define put_byte_0	lsl #24
52 #define put_byte_1	lsl #16
53 #define put_byte_2	lsl #8
54 #define put_byte_3      lsl #0
55 #endif
56 
57 /* Select code for any configuration running in BE8 mode */
58 #ifdef CONFIG_CPU_ENDIAN_BE8
59 #define ARM_BE8(code...) code
60 #else
61 #define ARM_BE8(code...)
62 #endif
63 
64 /*
65  * Data preload for architectures that support it
66  */
67 #if __LINUX_ARM_ARCH__ >= 5
68 #define PLD(code...)	code
69 #else
70 #define PLD(code...)
71 #endif
72 
73 /*
74  * This can be used to enable code to cacheline align the destination
75  * pointer when bulk writing to memory.  Experiments on StrongARM and
76  * XScale didn't show this a worthwhile thing to do when the cache is not
77  * set to write-allocate (this would need further testing on XScale when WA
78  * is used).
79  *
80  * On Feroceon there is much to gain however, regardless of cache mode.
81  */
82 #ifdef CONFIG_CPU_FEROCEON
83 #define CALGN(code...) code
84 #else
85 #define CALGN(code...)
86 #endif
87 
88 #define IMM12_MASK 0xfff
89 
90 /* the frame pointer used for stack unwinding */
91 ARM(	fpreg	.req	r11	)
92 THUMB(	fpreg	.req	r7	)
93 
94 /*
95  * Enable and disable interrupts
96  */
97 #if __LINUX_ARM_ARCH__ >= 6
98 	.macro	disable_irq_notrace
99 	cpsid	i
100 	.endm
101 
102 	.macro	enable_irq_notrace
103 	cpsie	i
104 	.endm
105 #else
106 	.macro	disable_irq_notrace
107 	msr	cpsr_c, #PSR_I_BIT | SVC_MODE
108 	.endm
109 
110 	.macro	enable_irq_notrace
111 	msr	cpsr_c, #SVC_MODE
112 	.endm
113 #endif
114 
115 #if __LINUX_ARM_ARCH__ < 7
116 	.macro	dsb, args
117 	mcr	p15, 0, r0, c7, c10, 4
118 	.endm
119 
120 	.macro	isb, args
121 	mcr	p15, 0, r0, c7, c5, 4
122 	.endm
123 #endif
124 
125 	.macro asm_trace_hardirqs_off, save=1
126 #if defined(CONFIG_TRACE_IRQFLAGS)
127 	.if \save
128 	stmdb   sp!, {r0-r3, ip, lr}
129 	.endif
130 	bl	trace_hardirqs_off
131 	.if \save
132 	ldmia	sp!, {r0-r3, ip, lr}
133 	.endif
134 #endif
135 	.endm
136 
137 	.macro asm_trace_hardirqs_on, cond=al, save=1
138 #if defined(CONFIG_TRACE_IRQFLAGS)
139 	/*
140 	 * actually the registers should be pushed and pop'd conditionally, but
141 	 * after bl the flags are certainly clobbered
142 	 */
143 	.if \save
144 	stmdb   sp!, {r0-r3, ip, lr}
145 	.endif
146 	bl\cond	trace_hardirqs_on
147 	.if \save
148 	ldmia	sp!, {r0-r3, ip, lr}
149 	.endif
150 #endif
151 	.endm
152 
153 	.macro disable_irq, save=1
154 	disable_irq_notrace
155 	asm_trace_hardirqs_off \save
156 	.endm
157 
158 	.macro enable_irq
159 	asm_trace_hardirqs_on
160 	enable_irq_notrace
161 	.endm
162 /*
163  * Save the current IRQ state and disable IRQs.  Note that this macro
164  * assumes FIQs are enabled, and that the processor is in SVC mode.
165  */
166 	.macro	save_and_disable_irqs, oldcpsr
167 #ifdef CONFIG_CPU_V7M
168 	mrs	\oldcpsr, primask
169 #else
170 	mrs	\oldcpsr, cpsr
171 #endif
172 	disable_irq
173 	.endm
174 
175 	.macro	save_and_disable_irqs_notrace, oldcpsr
176 #ifdef CONFIG_CPU_V7M
177 	mrs	\oldcpsr, primask
178 #else
179 	mrs	\oldcpsr, cpsr
180 #endif
181 	disable_irq_notrace
182 	.endm
183 
184 /*
185  * Restore interrupt state previously stored in a register.  We don't
186  * guarantee that this will preserve the flags.
187  */
188 	.macro	restore_irqs_notrace, oldcpsr
189 #ifdef CONFIG_CPU_V7M
190 	msr	primask, \oldcpsr
191 #else
192 	msr	cpsr_c, \oldcpsr
193 #endif
194 	.endm
195 
196 	.macro restore_irqs, oldcpsr
197 	tst	\oldcpsr, #PSR_I_BIT
198 	asm_trace_hardirqs_on cond=eq
199 	restore_irqs_notrace \oldcpsr
200 	.endm
201 
202 /*
203  * Assembly version of "adr rd, BSYM(sym)".  This should only be used to
204  * reference local symbols in the same assembly file which are to be
205  * resolved by the assembler.  Other usage is undefined.
206  */
207 	.irp	c,,eq,ne,cs,cc,mi,pl,vs,vc,hi,ls,ge,lt,gt,le,hs,lo
208 	.macro	badr\c, rd, sym
209 #ifdef CONFIG_THUMB2_KERNEL
210 	adr\c	\rd, \sym + 1
211 #else
212 	adr\c	\rd, \sym
213 #endif
214 	.endm
215 	.endr
216 
217 /*
218  * Get current thread_info.
219  */
220 	.macro	get_thread_info, rd
221 	/* thread_info is the first member of struct task_struct */
222 	get_current \rd
223 	.endm
224 
225 /*
226  * Increment/decrement the preempt count.
227  */
228 #ifdef CONFIG_PREEMPT_COUNT
229 	.macro	inc_preempt_count, ti, tmp
230 	ldr	\tmp, [\ti, #TI_PREEMPT]	@ get preempt count
231 	add	\tmp, \tmp, #1			@ increment it
232 	str	\tmp, [\ti, #TI_PREEMPT]
233 	.endm
234 
235 	.macro	dec_preempt_count, ti, tmp
236 	ldr	\tmp, [\ti, #TI_PREEMPT]	@ get preempt count
237 	sub	\tmp, \tmp, #1			@ decrement it
238 	str	\tmp, [\ti, #TI_PREEMPT]
239 	.endm
240 #else
241 	.macro	inc_preempt_count, ti, tmp
242 	.endm
243 
244 	.macro	dec_preempt_count, ti, tmp
245 	.endm
246 #endif
247 
248 #define USERL(l, x...)				\
249 9999:	x;					\
250 	.pushsection __ex_table,"a";		\
251 	.align	3;				\
252 	.long	9999b,l;			\
253 	.popsection
254 
255 #define USER(x...)	USERL(9001f, x)
256 
257 #ifdef CONFIG_SMP
258 #define ALT_SMP(instr...)					\
259 9998:	instr
260 /*
261  * Note: if you get assembler errors from ALT_UP() when building with
262  * CONFIG_THUMB2_KERNEL, you almost certainly need to use
263  * ALT_SMP( W(instr) ... )
264  */
265 #define ALT_UP(instr...)					\
266 	.pushsection ".alt.smp.init", "a"			;\
267 	.align	2						;\
268 	.long	9998b - .					;\
269 9997:	instr							;\
270 	.if . - 9997b == 2					;\
271 		nop						;\
272 	.endif							;\
273 	.if . - 9997b != 4					;\
274 		.error "ALT_UP() content must assemble to exactly 4 bytes";\
275 	.endif							;\
276 	.popsection
277 #define ALT_UP_B(label)					\
278 	.pushsection ".alt.smp.init", "a"			;\
279 	.align	2						;\
280 	.long	9998b - .					;\
281 	W(b)	. + (label - 9998b)					;\
282 	.popsection
283 #else
284 #define ALT_SMP(instr...)
285 #define ALT_UP(instr...) instr
286 #define ALT_UP_B(label) b label
287 #endif
288 
289 	/*
290 	 * this_cpu_offset - load the per-CPU offset of this CPU into
291 	 * 		     register 'rd'
292 	 */
293 	.macro		this_cpu_offset, rd:req
294 #ifdef CONFIG_SMP
295 ALT_SMP(mrc		p15, 0, \rd, c13, c0, 4)
296 #ifdef CONFIG_CPU_V6
297 ALT_UP_B(.L1_\@)
298 .L0_\@:
299 	.subsection	1
300 .L1_\@: ldr_va		\rd, __per_cpu_offset
301 	b		.L0_\@
302 	.previous
303 #endif
304 #else
305 	mov		\rd, #0
306 #endif
307 	.endm
308 
309 	/*
310 	 * set_current - store the task pointer of this CPU's current task
311 	 */
312 	.macro		set_current, rn:req, tmp:req
313 #if defined(CONFIG_CURRENT_POINTER_IN_TPIDRURO) || defined(CONFIG_SMP)
314 9998:	mcr		p15, 0, \rn, c13, c0, 3		@ set TPIDRURO register
315 #ifdef CONFIG_CPU_V6
316 ALT_UP_B(.L0_\@)
317 	.subsection	1
318 .L0_\@: str_va		\rn, __current, \tmp
319 	b		.L1_\@
320 	.previous
321 .L1_\@:
322 #endif
323 #else
324 	str_va		\rn, __current, \tmp
325 #endif
326 	.endm
327 
328 	/*
329 	 * get_current - load the task pointer of this CPU's current task
330 	 */
331 	.macro		get_current, rd:req
332 #if defined(CONFIG_CURRENT_POINTER_IN_TPIDRURO) || defined(CONFIG_SMP)
333 9998:	mrc		p15, 0, \rd, c13, c0, 3		@ get TPIDRURO register
334 #ifdef CONFIG_CPU_V6
335 ALT_UP_B(.L0_\@)
336 	.subsection	1
337 .L0_\@: ldr_va		\rd, __current
338 	b		.L1_\@
339 	.previous
340 .L1_\@:
341 #endif
342 #else
343 	ldr_va		\rd, __current
344 #endif
345 	.endm
346 
347 	/*
348 	 * reload_current - reload the task pointer of this CPU's current task
349 	 *		    into the TLS register
350 	 */
351 	.macro		reload_current, t1:req, t2:req
352 #if defined(CONFIG_CURRENT_POINTER_IN_TPIDRURO) || defined(CONFIG_SMP)
353 #ifdef CONFIG_CPU_V6
354 ALT_SMP(nop)
355 ALT_UP_B(.L0_\@)
356 #endif
357 	ldr_this_cpu	\t1, __entry_task, \t1, \t2
358 	mcr		p15, 0, \t1, c13, c0, 3		@ store in TPIDRURO
359 .L0_\@:
360 #endif
361 	.endm
362 
363 /*
364  * Instruction barrier
365  */
366 	.macro	instr_sync
367 #if __LINUX_ARM_ARCH__ >= 7
368 	isb
369 #elif __LINUX_ARM_ARCH__ == 6
370 	mcr	p15, 0, r0, c7, c5, 4
371 #endif
372 	.endm
373 
374 /*
375  * SMP data memory barrier
376  */
377 	.macro	smp_dmb mode
378 #ifdef CONFIG_SMP
379 #if __LINUX_ARM_ARCH__ >= 7
380 	.ifeqs "\mode","arm"
381 	ALT_SMP(dmb	ish)
382 	.else
383 	ALT_SMP(W(dmb)	ish)
384 	.endif
385 #elif __LINUX_ARM_ARCH__ == 6
386 	ALT_SMP(mcr	p15, 0, r0, c7, c10, 5)	@ dmb
387 #else
388 #error Incompatible SMP platform
389 #endif
390 	.ifeqs "\mode","arm"
391 	ALT_UP(nop)
392 	.else
393 	ALT_UP(W(nop))
394 	.endif
395 #endif
396 	.endm
397 
398 /*
399  * Raw SMP data memory barrier
400  */
401 	.macro	__smp_dmb mode
402 #if __LINUX_ARM_ARCH__ >= 7
403 	.ifeqs "\mode","arm"
404 	dmb	ish
405 	.else
406 	W(dmb)	ish
407 	.endif
408 #elif __LINUX_ARM_ARCH__ == 6
409 	mcr	p15, 0, r0, c7, c10, 5	@ dmb
410 #else
411 	.error "Incompatible SMP platform"
412 #endif
413 	.endm
414 
415 #if defined(CONFIG_CPU_V7M)
416 	/*
417 	 * setmode is used to assert to be in svc mode during boot. For v7-M
418 	 * this is done in __v7m_setup, so setmode can be empty here.
419 	 */
420 	.macro	setmode, mode, reg
421 	.endm
422 #elif defined(CONFIG_THUMB2_KERNEL)
423 	.macro	setmode, mode, reg
424 	mov	\reg, #\mode
425 	msr	cpsr_c, \reg
426 	.endm
427 #else
428 	.macro	setmode, mode, reg
429 	msr	cpsr_c, #\mode
430 	.endm
431 #endif
432 
433 /*
434  * Helper macro to enter SVC mode cleanly and mask interrupts. reg is
435  * a scratch register for the macro to overwrite.
436  *
437  * This macro is intended for forcing the CPU into SVC mode at boot time.
438  * you cannot return to the original mode.
439  */
440 .macro safe_svcmode_maskall reg:req
441 #if __LINUX_ARM_ARCH__ >= 6 && !defined(CONFIG_CPU_V7M)
442 	mrs	\reg , cpsr
443 	eor	\reg, \reg, #HYP_MODE
444 	tst	\reg, #MODE_MASK
445 	bic	\reg , \reg , #MODE_MASK
446 	orr	\reg , \reg , #PSR_I_BIT | PSR_F_BIT | SVC_MODE
447 THUMB(	orr	\reg , \reg , #PSR_T_BIT	)
448 	bne	1f
449 	orr	\reg, \reg, #PSR_A_BIT
450 	badr	lr, 2f
451 	msr	spsr_cxsf, \reg
452 	__MSR_ELR_HYP(14)
453 	__ERET
454 1:	msr	cpsr_c, \reg
455 2:
456 #else
457 /*
458  * workaround for possibly broken pre-v6 hardware
459  * (akita, Sharp Zaurus C-1000, PXA270-based)
460  */
461 	setmode	PSR_F_BIT | PSR_I_BIT | SVC_MODE, \reg
462 #endif
463 .endm
464 
465 /*
466  * STRT/LDRT access macros with ARM and Thumb-2 variants
467  */
468 #ifdef CONFIG_THUMB2_KERNEL
469 
470 	.macro	usraccoff, instr, reg, ptr, inc, off, cond, abort, t=TUSER()
471 9999:
472 	.if	\inc == 1
473 	\instr\()b\t\cond\().w \reg, [\ptr, #\off]
474 	.elseif	\inc == 4
475 	\instr\t\cond\().w \reg, [\ptr, #\off]
476 	.else
477 	.error	"Unsupported inc macro argument"
478 	.endif
479 
480 	.pushsection __ex_table,"a"
481 	.align	3
482 	.long	9999b, \abort
483 	.popsection
484 	.endm
485 
486 	.macro	usracc, instr, reg, ptr, inc, cond, rept, abort
487 	@ explicit IT instruction needed because of the label
488 	@ introduced by the USER macro
489 	.ifnc	\cond,al
490 	.if	\rept == 1
491 	itt	\cond
492 	.elseif	\rept == 2
493 	ittt	\cond
494 	.else
495 	.error	"Unsupported rept macro argument"
496 	.endif
497 	.endif
498 
499 	@ Slightly optimised to avoid incrementing the pointer twice
500 	usraccoff \instr, \reg, \ptr, \inc, 0, \cond, \abort
501 	.if	\rept == 2
502 	usraccoff \instr, \reg, \ptr, \inc, \inc, \cond, \abort
503 	.endif
504 
505 	add\cond \ptr, #\rept * \inc
506 	.endm
507 
508 #else	/* !CONFIG_THUMB2_KERNEL */
509 
510 	.macro	usracc, instr, reg, ptr, inc, cond, rept, abort, t=TUSER()
511 	.rept	\rept
512 9999:
513 	.if	\inc == 1
514 	\instr\()b\t\cond \reg, [\ptr], #\inc
515 	.elseif	\inc == 4
516 	\instr\t\cond \reg, [\ptr], #\inc
517 	.else
518 	.error	"Unsupported inc macro argument"
519 	.endif
520 
521 	.pushsection __ex_table,"a"
522 	.align	3
523 	.long	9999b, \abort
524 	.popsection
525 	.endr
526 	.endm
527 
528 #endif	/* CONFIG_THUMB2_KERNEL */
529 
530 	.macro	strusr, reg, ptr, inc, cond=al, rept=1, abort=9001f
531 	usracc	str, \reg, \ptr, \inc, \cond, \rept, \abort
532 	.endm
533 
534 	.macro	ldrusr, reg, ptr, inc, cond=al, rept=1, abort=9001f
535 	usracc	ldr, \reg, \ptr, \inc, \cond, \rept, \abort
536 	.endm
537 
538 /* Utility macro for declaring string literals */
539 	.macro	string name:req, string
540 	.type \name , #object
541 \name:
542 	.asciz "\string"
543 	.size \name , . - \name
544 	.endm
545 
546 	.irp	c,,eq,ne,cs,cc,mi,pl,vs,vc,hi,ls,ge,lt,gt,le,hs,lo
547 	.macro	ret\c, reg
548 #if __LINUX_ARM_ARCH__ < 6
549 	mov\c	pc, \reg
550 #else
551 	.ifeqs	"\reg", "lr"
552 	bx\c	\reg
553 	.else
554 	mov\c	pc, \reg
555 	.endif
556 #endif
557 	.endm
558 	.endr
559 
560 	.macro	ret.w, reg
561 	ret	\reg
562 #ifdef CONFIG_THUMB2_KERNEL
563 	nop
564 #endif
565 	.endm
566 
567 	.macro	bug, msg, line
568 #ifdef CONFIG_THUMB2_KERNEL
569 1:	.inst	0xde02
570 #else
571 1:	.inst	0xe7f001f2
572 #endif
573 #ifdef CONFIG_DEBUG_BUGVERBOSE
574 	.pushsection .rodata.str, "aMS", %progbits, 1
575 2:	.asciz	"\msg"
576 	.popsection
577 	.pushsection __bug_table, "aw"
578 	.align	2
579 	.word	1b, 2b
580 	.hword	\line
581 	.popsection
582 #endif
583 	.endm
584 
585 #ifdef CONFIG_KPROBES
586 #define _ASM_NOKPROBE(entry)				\
587 	.pushsection "_kprobe_blacklist", "aw" ;	\
588 	.balign 4 ;					\
589 	.long entry;					\
590 	.popsection
591 #else
592 #define _ASM_NOKPROBE(entry)
593 #endif
594 
595 	.macro		__adldst_l, op, reg, sym, tmp, c
596 	.if		__LINUX_ARM_ARCH__ < 7
597 	ldr\c		\tmp, .La\@
598 	.subsection	1
599 	.align		2
600 .La\@:	.long		\sym - .Lpc\@
601 	.previous
602 	.else
603 	.ifnb		\c
604  THUMB(	ittt		\c			)
605 	.endif
606 	movw\c		\tmp, #:lower16:\sym - .Lpc\@
607 	movt\c		\tmp, #:upper16:\sym - .Lpc\@
608 	.endif
609 
610 #ifndef CONFIG_THUMB2_KERNEL
611 	.set		.Lpc\@, . + 8			// PC bias
612 	.ifc		\op, add
613 	add\c		\reg, \tmp, pc
614 	.else
615 	\op\c		\reg, [pc, \tmp]
616 	.endif
617 #else
618 .Lb\@:	add\c		\tmp, \tmp, pc
619 	/*
620 	 * In Thumb-2 builds, the PC bias depends on whether we are currently
621 	 * emitting into a .arm or a .thumb section. The size of the add opcode
622 	 * above will be 2 bytes when emitting in Thumb mode and 4 bytes when
623 	 * emitting in ARM mode, so let's use this to account for the bias.
624 	 */
625 	.set		.Lpc\@, . + (. - .Lb\@)
626 
627 	.ifnc		\op, add
628 	\op\c		\reg, [\tmp]
629 	.endif
630 #endif
631 	.endm
632 
633 	/*
634 	 * mov_l - move a constant value or [relocated] address into a register
635 	 */
636 	.macro		mov_l, dst:req, imm:req, cond
637 	.if		__LINUX_ARM_ARCH__ < 7
638 	ldr\cond	\dst, =\imm
639 	.else
640 	movw\cond	\dst, #:lower16:\imm
641 	movt\cond	\dst, #:upper16:\imm
642 	.endif
643 	.endm
644 
645 	/*
646 	 * adr_l - adr pseudo-op with unlimited range
647 	 *
648 	 * @dst: destination register
649 	 * @sym: name of the symbol
650 	 * @cond: conditional opcode suffix
651 	 */
652 	.macro		adr_l, dst:req, sym:req, cond
653 	__adldst_l	add, \dst, \sym, \dst, \cond
654 	.endm
655 
656 	/*
657 	 * ldr_l - ldr <literal> pseudo-op with unlimited range
658 	 *
659 	 * @dst: destination register
660 	 * @sym: name of the symbol
661 	 * @cond: conditional opcode suffix
662 	 */
663 	.macro		ldr_l, dst:req, sym:req, cond
664 	__adldst_l	ldr, \dst, \sym, \dst, \cond
665 	.endm
666 
667 	/*
668 	 * str_l - str <literal> pseudo-op with unlimited range
669 	 *
670 	 * @src: source register
671 	 * @sym: name of the symbol
672 	 * @tmp: mandatory scratch register
673 	 * @cond: conditional opcode suffix
674 	 */
675 	.macro		str_l, src:req, sym:req, tmp:req, cond
676 	__adldst_l	str, \src, \sym, \tmp, \cond
677 	.endm
678 
679 	.macro		__ldst_va, op, reg, tmp, sym, cond, offset
680 #if __LINUX_ARM_ARCH__ >= 7 || \
681     !defined(CONFIG_ARM_HAS_GROUP_RELOCS) || \
682     (defined(MODULE) && defined(CONFIG_ARM_MODULE_PLTS))
683 	mov_l		\tmp, \sym, \cond
684 #else
685 	/*
686 	 * Avoid a literal load, by emitting a sequence of ADD/LDR instructions
687 	 * with the appropriate relocations. The combined sequence has a range
688 	 * of -/+ 256 MiB, which should be sufficient for the core kernel and
689 	 * for modules loaded into the module region.
690 	 */
691 	.globl		\sym
692 	.reloc		.L0_\@, R_ARM_ALU_PC_G0_NC, \sym
693 	.reloc		.L1_\@, R_ARM_ALU_PC_G1_NC, \sym
694 	.reloc		.L2_\@, R_ARM_LDR_PC_G2, \sym
695 .L0_\@: sub\cond	\tmp, pc, #8 - \offset
696 .L1_\@: sub\cond	\tmp, \tmp, #4 - \offset
697 .L2_\@:
698 #endif
699 	\op\cond	\reg, [\tmp, #\offset]
700 	.endm
701 
702 	/*
703 	 * ldr_va - load a 32-bit word from the virtual address of \sym
704 	 */
705 	.macro		ldr_va, rd:req, sym:req, cond, tmp, offset=0
706 	.ifnb		\tmp
707 	__ldst_va	ldr, \rd, \tmp, \sym, \cond, \offset
708 	.else
709 	__ldst_va	ldr, \rd, \rd, \sym, \cond, \offset
710 	.endif
711 	.endm
712 
713 	/*
714 	 * str_va - store a 32-bit word to the virtual address of \sym
715 	 */
716 	.macro		str_va, rn:req, sym:req, tmp:req, cond
717 	__ldst_va	str, \rn, \tmp, \sym, \cond, 0
718 	.endm
719 
720 	/*
721 	 * ldr_this_cpu_armv6 - Load a 32-bit word from the per-CPU variable 'sym',
722 	 *			without using a temp register. Supported in ARM mode
723 	 *			only.
724 	 */
725 	.macro		ldr_this_cpu_armv6, rd:req, sym:req
726 	this_cpu_offset	\rd
727 	.globl		\sym
728 	.reloc		.L0_\@, R_ARM_ALU_PC_G0_NC, \sym
729 	.reloc		.L1_\@, R_ARM_ALU_PC_G1_NC, \sym
730 	.reloc		.L2_\@, R_ARM_LDR_PC_G2, \sym
731 	add		\rd, \rd, pc
732 .L0_\@: sub		\rd, \rd, #4
733 .L1_\@: sub		\rd, \rd, #0
734 .L2_\@: ldr		\rd, [\rd, #4]
735 	.endm
736 
737 	/*
738 	 * ldr_this_cpu - Load a 32-bit word from the per-CPU variable 'sym'
739 	 *		  into register 'rd', which may be the stack pointer,
740 	 *		  using 't1' and 't2' as general temp registers. These
741 	 *		  are permitted to overlap with 'rd' if != sp
742 	 */
743 	.macro		ldr_this_cpu, rd:req, sym:req, t1:req, t2:req
744 #ifndef CONFIG_SMP
745 	ldr_va		\rd, \sym, tmp=\t1
746 #elif __LINUX_ARM_ARCH__ >= 7 || \
747       !defined(CONFIG_ARM_HAS_GROUP_RELOCS) || \
748       (defined(MODULE) && defined(CONFIG_ARM_MODULE_PLTS))
749 	this_cpu_offset	\t1
750 	mov_l		\t2, \sym
751 	ldr		\rd, [\t1, \t2]
752 #else
753 	ldr_this_cpu_armv6 \rd, \sym
754 #endif
755 	.endm
756 
757 	/*
758 	 * rev_l - byte-swap a 32-bit value
759 	 *
760 	 * @val: source/destination register
761 	 * @tmp: scratch register
762 	 */
763 	.macro		rev_l, val:req, tmp:req
764 	.if		__LINUX_ARM_ARCH__ < 6
765 	eor		\tmp, \val, \val, ror #16
766 	bic		\tmp, \tmp, #0x00ff0000
767 	mov		\val, \val, ror #8
768 	eor		\val, \val, \tmp, lsr #8
769 	.else
770 	rev		\val, \val
771 	.endif
772 	.endm
773 
774 	.if		__LINUX_ARM_ARCH__ < 6
775 	.set		.Lrev_l_uses_tmp, 1
776 	.else
777 	.set		.Lrev_l_uses_tmp, 0
778 	.endif
779 
780 	/*
781 	 * bl_r - branch and link to register
782 	 *
783 	 * @dst: target to branch to
784 	 * @c: conditional opcode suffix
785 	 */
786 	.macro		bl_r, dst:req, c
787 	.if		__LINUX_ARM_ARCH__ < 6
788 	mov\c		lr, pc
789 	mov\c		pc, \dst
790 	.else
791 	blx\c		\dst
792 	.endif
793 	.endm
794 
795 #endif /* __ASM_ASSEMBLER_H__ */
796