]> git.ipfire.org Git - thirdparty/linux.git/blob - arch/arm64/include/asm/assembler.h
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 234
[thirdparty/linux.git] / arch / arm64 / include / asm / assembler.h
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3 * Based on arch/arm/include/asm/assembler.h, arch/arm/mm/proc-macros.S
4 *
5 * Copyright (C) 1996-2000 Russell King
6 * Copyright (C) 2012 ARM Ltd.
7 */
8 #ifndef __ASSEMBLY__
9 #error "Only include this from assembly code"
10 #endif
11
12 #ifndef __ASM_ASSEMBLER_H
13 #define __ASM_ASSEMBLER_H
14
15 #include <asm-generic/export.h>
16
17 #include <asm/asm-offsets.h>
18 #include <asm/cpufeature.h>
19 #include <asm/cputype.h>
20 #include <asm/debug-monitors.h>
21 #include <asm/page.h>
22 #include <asm/pgtable-hwdef.h>
23 #include <asm/ptrace.h>
24 #include <asm/thread_info.h>
25
26 .macro save_and_disable_daif, flags
27 mrs \flags, daif
28 msr daifset, #0xf
29 .endm
30
31 .macro disable_daif
32 msr daifset, #0xf
33 .endm
34
35 .macro enable_daif
36 msr daifclr, #0xf
37 .endm
38
39 .macro restore_daif, flags:req
40 msr daif, \flags
41 .endm
42
43 /* Only on aarch64 pstate, PSR_D_BIT is different for aarch32 */
44 .macro inherit_daif, pstate:req, tmp:req
45 and \tmp, \pstate, #(PSR_D_BIT | PSR_A_BIT | PSR_I_BIT | PSR_F_BIT)
46 msr daif, \tmp
47 .endm
48
49 /* IRQ is the lowest priority flag, unconditionally unmask the rest. */
50 .macro enable_da_f
51 msr daifclr, #(8 | 4 | 1)
52 .endm
53
54 /*
55 * Save/restore interrupts.
56 */
57 .macro save_and_disable_irq, flags
58 mrs \flags, daif
59 msr daifset, #2
60 .endm
61
62 .macro restore_irq, flags
63 msr daif, \flags
64 .endm
65
66 .macro enable_dbg
67 msr daifclr, #8
68 .endm
69
70 .macro disable_step_tsk, flgs, tmp
71 tbz \flgs, #TIF_SINGLESTEP, 9990f
72 mrs \tmp, mdscr_el1
73 bic \tmp, \tmp, #DBG_MDSCR_SS
74 msr mdscr_el1, \tmp
75 isb // Synchronise with enable_dbg
76 9990:
77 .endm
78
79 /* call with daif masked */
80 .macro enable_step_tsk, flgs, tmp
81 tbz \flgs, #TIF_SINGLESTEP, 9990f
82 mrs \tmp, mdscr_el1
83 orr \tmp, \tmp, #DBG_MDSCR_SS
84 msr mdscr_el1, \tmp
85 9990:
86 .endm
87
88 /*
89 * SMP data memory barrier
90 */
91 .macro smp_dmb, opt
92 dmb \opt
93 .endm
94
95 /*
96 * RAS Error Synchronization barrier
97 */
98 .macro esb
99 hint #16
100 .endm
101
102 /*
103 * Value prediction barrier
104 */
105 .macro csdb
106 hint #20
107 .endm
108
109 /*
110 * Speculation barrier
111 */
112 .macro sb
113 alternative_if_not ARM64_HAS_SB
114 dsb nsh
115 isb
116 alternative_else
117 SB_BARRIER_INSN
118 nop
119 alternative_endif
120 .endm
121
122 /*
123 * Sanitise a 64-bit bounded index wrt speculation, returning zero if out
124 * of bounds.
125 */
126 .macro mask_nospec64, idx, limit, tmp
127 sub \tmp, \idx, \limit
128 bic \tmp, \tmp, \idx
129 and \idx, \idx, \tmp, asr #63
130 csdb
131 .endm
132
133 /*
134 * NOP sequence
135 */
136 .macro nops, num
137 .rept \num
138 nop
139 .endr
140 .endm
141
142 /*
143 * Emit an entry into the exception table
144 */
145 .macro _asm_extable, from, to
146 .pushsection __ex_table, "a"
147 .align 3
148 .long (\from - .), (\to - .)
149 .popsection
150 .endm
151
152 #define USER(l, x...) \
153 9999: x; \
154 _asm_extable 9999b, l
155
156 /*
157 * Register aliases.
158 */
159 lr .req x30 // link register
160
161 /*
162 * Vector entry
163 */
164 .macro ventry label
165 .align 7
166 b \label
167 .endm
168
169 /*
170 * Select code when configured for BE.
171 */
172 #ifdef CONFIG_CPU_BIG_ENDIAN
173 #define CPU_BE(code...) code
174 #else
175 #define CPU_BE(code...)
176 #endif
177
178 /*
179 * Select code when configured for LE.
180 */
181 #ifdef CONFIG_CPU_BIG_ENDIAN
182 #define CPU_LE(code...)
183 #else
184 #define CPU_LE(code...) code
185 #endif
186
187 /*
188 * Define a macro that constructs a 64-bit value by concatenating two
189 * 32-bit registers. Note that on big endian systems the order of the
190 * registers is swapped.
191 */
192 #ifndef CONFIG_CPU_BIG_ENDIAN
193 .macro regs_to_64, rd, lbits, hbits
194 #else
195 .macro regs_to_64, rd, hbits, lbits
196 #endif
197 orr \rd, \lbits, \hbits, lsl #32
198 .endm
199
200 /*
201 * Pseudo-ops for PC-relative adr/ldr/str <reg>, <symbol> where
202 * <symbol> is within the range +/- 4 GB of the PC.
203 */
204 /*
205 * @dst: destination register (64 bit wide)
206 * @sym: name of the symbol
207 */
208 .macro adr_l, dst, sym
209 adrp \dst, \sym
210 add \dst, \dst, :lo12:\sym
211 .endm
212
213 /*
214 * @dst: destination register (32 or 64 bit wide)
215 * @sym: name of the symbol
216 * @tmp: optional 64-bit scratch register to be used if <dst> is a
217 * 32-bit wide register, in which case it cannot be used to hold
218 * the address
219 */
220 .macro ldr_l, dst, sym, tmp=
221 .ifb \tmp
222 adrp \dst, \sym
223 ldr \dst, [\dst, :lo12:\sym]
224 .else
225 adrp \tmp, \sym
226 ldr \dst, [\tmp, :lo12:\sym]
227 .endif
228 .endm
229
230 /*
231 * @src: source register (32 or 64 bit wide)
232 * @sym: name of the symbol
233 * @tmp: mandatory 64-bit scratch register to calculate the address
234 * while <src> needs to be preserved.
235 */
236 .macro str_l, src, sym, tmp
237 adrp \tmp, \sym
238 str \src, [\tmp, :lo12:\sym]
239 .endm
240
241 /*
242 * @dst: Result of per_cpu(sym, smp_processor_id()) (can be SP)
243 * @sym: The name of the per-cpu variable
244 * @tmp: scratch register
245 */
246 .macro adr_this_cpu, dst, sym, tmp
247 adrp \tmp, \sym
248 add \dst, \tmp, #:lo12:\sym
249 alternative_if_not ARM64_HAS_VIRT_HOST_EXTN
250 mrs \tmp, tpidr_el1
251 alternative_else
252 mrs \tmp, tpidr_el2
253 alternative_endif
254 add \dst, \dst, \tmp
255 .endm
256
257 /*
258 * @dst: Result of READ_ONCE(per_cpu(sym, smp_processor_id()))
259 * @sym: The name of the per-cpu variable
260 * @tmp: scratch register
261 */
262 .macro ldr_this_cpu dst, sym, tmp
263 adr_l \dst, \sym
264 alternative_if_not ARM64_HAS_VIRT_HOST_EXTN
265 mrs \tmp, tpidr_el1
266 alternative_else
267 mrs \tmp, tpidr_el2
268 alternative_endif
269 ldr \dst, [\dst, \tmp]
270 .endm
271
272 /*
273 * vma_vm_mm - get mm pointer from vma pointer (vma->vm_mm)
274 */
275 .macro vma_vm_mm, rd, rn
276 ldr \rd, [\rn, #VMA_VM_MM]
277 .endm
278
279 /*
280 * mmid - get context id from mm pointer (mm->context.id)
281 */
282 .macro mmid, rd, rn
283 ldr \rd, [\rn, #MM_CONTEXT_ID]
284 .endm
285 /*
286 * read_ctr - read CTR_EL0. If the system has mismatched register fields,
287 * provide the system wide safe value from arm64_ftr_reg_ctrel0.sys_val
288 */
289 .macro read_ctr, reg
290 alternative_if_not ARM64_MISMATCHED_CACHE_TYPE
291 mrs \reg, ctr_el0 // read CTR
292 nop
293 alternative_else
294 ldr_l \reg, arm64_ftr_reg_ctrel0 + ARM64_FTR_SYSVAL
295 alternative_endif
296 .endm
297
298
299 /*
300 * raw_dcache_line_size - get the minimum D-cache line size on this CPU
301 * from the CTR register.
302 */
303 .macro raw_dcache_line_size, reg, tmp
304 mrs \tmp, ctr_el0 // read CTR
305 ubfm \tmp, \tmp, #16, #19 // cache line size encoding
306 mov \reg, #4 // bytes per word
307 lsl \reg, \reg, \tmp // actual cache line size
308 .endm
309
310 /*
311 * dcache_line_size - get the safe D-cache line size across all CPUs
312 */
313 .macro dcache_line_size, reg, tmp
314 read_ctr \tmp
315 ubfm \tmp, \tmp, #16, #19 // cache line size encoding
316 mov \reg, #4 // bytes per word
317 lsl \reg, \reg, \tmp // actual cache line size
318 .endm
319
320 /*
321 * raw_icache_line_size - get the minimum I-cache line size on this CPU
322 * from the CTR register.
323 */
324 .macro raw_icache_line_size, reg, tmp
325 mrs \tmp, ctr_el0 // read CTR
326 and \tmp, \tmp, #0xf // cache line size encoding
327 mov \reg, #4 // bytes per word
328 lsl \reg, \reg, \tmp // actual cache line size
329 .endm
330
331 /*
332 * icache_line_size - get the safe I-cache line size across all CPUs
333 */
334 .macro icache_line_size, reg, tmp
335 read_ctr \tmp
336 and \tmp, \tmp, #0xf // cache line size encoding
337 mov \reg, #4 // bytes per word
338 lsl \reg, \reg, \tmp // actual cache line size
339 .endm
340
341 /*
342 * tcr_set_t0sz - update TCR.T0SZ so that we can load the ID map
343 */
344 .macro tcr_set_t0sz, valreg, t0sz
345 bfi \valreg, \t0sz, #TCR_T0SZ_OFFSET, #TCR_TxSZ_WIDTH
346 .endm
347
348 /*
349 * tcr_compute_pa_size - set TCR.(I)PS to the highest supported
350 * ID_AA64MMFR0_EL1.PARange value
351 *
352 * tcr: register with the TCR_ELx value to be updated
353 * pos: IPS or PS bitfield position
354 * tmp{0,1}: temporary registers
355 */
356 .macro tcr_compute_pa_size, tcr, pos, tmp0, tmp1
357 mrs \tmp0, ID_AA64MMFR0_EL1
358 // Narrow PARange to fit the PS field in TCR_ELx
359 ubfx \tmp0, \tmp0, #ID_AA64MMFR0_PARANGE_SHIFT, #3
360 mov \tmp1, #ID_AA64MMFR0_PARANGE_MAX
361 cmp \tmp0, \tmp1
362 csel \tmp0, \tmp1, \tmp0, hi
363 bfi \tcr, \tmp0, \pos, #3
364 .endm
365
366 /*
367 * Macro to perform a data cache maintenance for the interval
368 * [kaddr, kaddr + size)
369 *
370 * op: operation passed to dc instruction
371 * domain: domain used in dsb instruciton
372 * kaddr: starting virtual address of the region
373 * size: size of the region
374 * Corrupts: kaddr, size, tmp1, tmp2
375 */
376 .macro __dcache_op_workaround_clean_cache, op, kaddr
377 alternative_if_not ARM64_WORKAROUND_CLEAN_CACHE
378 dc \op, \kaddr
379 alternative_else
380 dc civac, \kaddr
381 alternative_endif
382 .endm
383
384 .macro dcache_by_line_op op, domain, kaddr, size, tmp1, tmp2
385 dcache_line_size \tmp1, \tmp2
386 add \size, \kaddr, \size
387 sub \tmp2, \tmp1, #1
388 bic \kaddr, \kaddr, \tmp2
389 9998:
390 .ifc \op, cvau
391 __dcache_op_workaround_clean_cache \op, \kaddr
392 .else
393 .ifc \op, cvac
394 __dcache_op_workaround_clean_cache \op, \kaddr
395 .else
396 .ifc \op, cvap
397 sys 3, c7, c12, 1, \kaddr // dc cvap
398 .else
399 .ifc \op, cvadp
400 sys 3, c7, c13, 1, \kaddr // dc cvadp
401 .else
402 dc \op, \kaddr
403 .endif
404 .endif
405 .endif
406 .endif
407 add \kaddr, \kaddr, \tmp1
408 cmp \kaddr, \size
409 b.lo 9998b
410 dsb \domain
411 .endm
412
413 /*
414 * Macro to perform an instruction cache maintenance for the interval
415 * [start, end)
416 *
417 * start, end: virtual addresses describing the region
418 * label: A label to branch to on user fault.
419 * Corrupts: tmp1, tmp2
420 */
421 .macro invalidate_icache_by_line start, end, tmp1, tmp2, label
422 icache_line_size \tmp1, \tmp2
423 sub \tmp2, \tmp1, #1
424 bic \tmp2, \start, \tmp2
425 9997:
426 USER(\label, ic ivau, \tmp2) // invalidate I line PoU
427 add \tmp2, \tmp2, \tmp1
428 cmp \tmp2, \end
429 b.lo 9997b
430 dsb ish
431 isb
432 .endm
433
434 /*
435 * reset_pmuserenr_el0 - reset PMUSERENR_EL0 if PMUv3 present
436 */
437 .macro reset_pmuserenr_el0, tmpreg
438 mrs \tmpreg, id_aa64dfr0_el1
439 sbfx \tmpreg, \tmpreg, #ID_AA64DFR0_PMUVER_SHIFT, #4
440 cmp \tmpreg, #1 // Skip if no PMU present
441 b.lt 9000f
442 msr pmuserenr_el0, xzr // Disable PMU access from EL0
443 9000:
444 .endm
445
446 /*
447 * copy_page - copy src to dest using temp registers t1-t8
448 */
449 .macro copy_page dest:req src:req t1:req t2:req t3:req t4:req t5:req t6:req t7:req t8:req
450 9998: ldp \t1, \t2, [\src]
451 ldp \t3, \t4, [\src, #16]
452 ldp \t5, \t6, [\src, #32]
453 ldp \t7, \t8, [\src, #48]
454 add \src, \src, #64
455 stnp \t1, \t2, [\dest]
456 stnp \t3, \t4, [\dest, #16]
457 stnp \t5, \t6, [\dest, #32]
458 stnp \t7, \t8, [\dest, #48]
459 add \dest, \dest, #64
460 tst \src, #(PAGE_SIZE - 1)
461 b.ne 9998b
462 .endm
463
464 /*
465 * Annotate a function as position independent, i.e., safe to be called before
466 * the kernel virtual mapping is activated.
467 */
468 #define ENDPIPROC(x) \
469 .globl __pi_##x; \
470 .type __pi_##x, %function; \
471 .set __pi_##x, x; \
472 .size __pi_##x, . - x; \
473 ENDPROC(x)
474
475 /*
476 * Annotate a function as being unsuitable for kprobes.
477 */
478 #ifdef CONFIG_KPROBES
479 #define NOKPROBE(x) \
480 .pushsection "_kprobe_blacklist", "aw"; \
481 .quad x; \
482 .popsection;
483 #else
484 #define NOKPROBE(x)
485 #endif
486
487 #ifdef CONFIG_KASAN
488 #define EXPORT_SYMBOL_NOKASAN(name)
489 #else
490 #define EXPORT_SYMBOL_NOKASAN(name) EXPORT_SYMBOL(name)
491 #endif
492
493 /*
494 * Emit a 64-bit absolute little endian symbol reference in a way that
495 * ensures that it will be resolved at build time, even when building a
496 * PIE binary. This requires cooperation from the linker script, which
497 * must emit the lo32/hi32 halves individually.
498 */
499 .macro le64sym, sym
500 .long \sym\()_lo32
501 .long \sym\()_hi32
502 .endm
503
504 /*
505 * mov_q - move an immediate constant into a 64-bit register using
506 * between 2 and 4 movz/movk instructions (depending on the
507 * magnitude and sign of the operand)
508 */
509 .macro mov_q, reg, val
510 .if (((\val) >> 31) == 0 || ((\val) >> 31) == 0x1ffffffff)
511 movz \reg, :abs_g1_s:\val
512 .else
513 .if (((\val) >> 47) == 0 || ((\val) >> 47) == 0x1ffff)
514 movz \reg, :abs_g2_s:\val
515 .else
516 movz \reg, :abs_g3:\val
517 movk \reg, :abs_g2_nc:\val
518 .endif
519 movk \reg, :abs_g1_nc:\val
520 .endif
521 movk \reg, :abs_g0_nc:\val
522 .endm
523
524 /*
525 * Return the current task_struct.
526 */
527 .macro get_current_task, rd
528 mrs \rd, sp_el0
529 .endm
530
531 /*
532 * Offset ttbr1 to allow for 48-bit kernel VAs set with 52-bit PTRS_PER_PGD.
533 * orr is used as it can cover the immediate value (and is idempotent).
534 * In future this may be nop'ed out when dealing with 52-bit kernel VAs.
535 * ttbr: Value of ttbr to set, modified.
536 */
537 .macro offset_ttbr1, ttbr
538 #ifdef CONFIG_ARM64_USER_VA_BITS_52
539 orr \ttbr, \ttbr, #TTBR1_BADDR_4852_OFFSET
540 #endif
541 .endm
542
543 /*
544 * Perform the reverse of offset_ttbr1.
545 * bic is used as it can cover the immediate value and, in future, won't need
546 * to be nop'ed out when dealing with 52-bit kernel VAs.
547 */
548 .macro restore_ttbr1, ttbr
549 #ifdef CONFIG_ARM64_USER_VA_BITS_52
550 bic \ttbr, \ttbr, #TTBR1_BADDR_4852_OFFSET
551 #endif
552 .endm
553
554 /*
555 * Arrange a physical address in a TTBR register, taking care of 52-bit
556 * addresses.
557 *
558 * phys: physical address, preserved
559 * ttbr: returns the TTBR value
560 */
561 .macro phys_to_ttbr, ttbr, phys
562 #ifdef CONFIG_ARM64_PA_BITS_52
563 orr \ttbr, \phys, \phys, lsr #46
564 and \ttbr, \ttbr, #TTBR_BADDR_MASK_52
565 #else
566 mov \ttbr, \phys
567 #endif
568 .endm
569
570 .macro phys_to_pte, pte, phys
571 #ifdef CONFIG_ARM64_PA_BITS_52
572 /*
573 * We assume \phys is 64K aligned and this is guaranteed by only
574 * supporting this configuration with 64K pages.
575 */
576 orr \pte, \phys, \phys, lsr #36
577 and \pte, \pte, #PTE_ADDR_MASK
578 #else
579 mov \pte, \phys
580 #endif
581 .endm
582
583 .macro pte_to_phys, phys, pte
584 #ifdef CONFIG_ARM64_PA_BITS_52
585 ubfiz \phys, \pte, #(48 - 16 - 12), #16
586 bfxil \phys, \pte, #16, #32
587 lsl \phys, \phys, #16
588 #else
589 and \phys, \pte, #PTE_ADDR_MASK
590 #endif
591 .endm
592
593 /*
594 * tcr_clear_errata_bits - Clear TCR bits that trigger an errata on this CPU.
595 */
596 .macro tcr_clear_errata_bits, tcr, tmp1, tmp2
597 #ifdef CONFIG_FUJITSU_ERRATUM_010001
598 mrs \tmp1, midr_el1
599
600 mov_q \tmp2, MIDR_FUJITSU_ERRATUM_010001_MASK
601 and \tmp1, \tmp1, \tmp2
602 mov_q \tmp2, MIDR_FUJITSU_ERRATUM_010001
603 cmp \tmp1, \tmp2
604 b.ne 10f
605
606 mov_q \tmp2, TCR_CLEAR_FUJITSU_ERRATUM_010001
607 bic \tcr, \tcr, \tmp2
608 10:
609 #endif /* CONFIG_FUJITSU_ERRATUM_010001 */
610 .endm
611
612 /**
613 * Errata workaround prior to disable MMU. Insert an ISB immediately prior
614 * to executing the MSR that will change SCTLR_ELn[M] from a value of 1 to 0.
615 */
616 .macro pre_disable_mmu_workaround
617 #ifdef CONFIG_QCOM_FALKOR_ERRATUM_E1041
618 isb
619 #endif
620 .endm
621
622 /*
623 * frame_push - Push @regcount callee saved registers to the stack,
624 * starting at x19, as well as x29/x30, and set x29 to
625 * the new value of sp. Add @extra bytes of stack space
626 * for locals.
627 */
628 .macro frame_push, regcount:req, extra
629 __frame st, \regcount, \extra
630 .endm
631
632 /*
633 * frame_pop - Pop the callee saved registers from the stack that were
634 * pushed in the most recent call to frame_push, as well
635 * as x29/x30 and any extra stack space that may have been
636 * allocated.
637 */
638 .macro frame_pop
639 __frame ld
640 .endm
641
642 .macro __frame_regs, reg1, reg2, op, num
643 .if .Lframe_regcount == \num
644 \op\()r \reg1, [sp, #(\num + 1) * 8]
645 .elseif .Lframe_regcount > \num
646 \op\()p \reg1, \reg2, [sp, #(\num + 1) * 8]
647 .endif
648 .endm
649
650 .macro __frame, op, regcount, extra=0
651 .ifc \op, st
652 .if (\regcount) < 0 || (\regcount) > 10
653 .error "regcount should be in the range [0 ... 10]"
654 .endif
655 .if ((\extra) % 16) != 0
656 .error "extra should be a multiple of 16 bytes"
657 .endif
658 .ifdef .Lframe_regcount
659 .if .Lframe_regcount != -1
660 .error "frame_push/frame_pop may not be nested"
661 .endif
662 .endif
663 .set .Lframe_regcount, \regcount
664 .set .Lframe_extra, \extra
665 .set .Lframe_local_offset, ((\regcount + 3) / 2) * 16
666 stp x29, x30, [sp, #-.Lframe_local_offset - .Lframe_extra]!
667 mov x29, sp
668 .endif
669
670 __frame_regs x19, x20, \op, 1
671 __frame_regs x21, x22, \op, 3
672 __frame_regs x23, x24, \op, 5
673 __frame_regs x25, x26, \op, 7
674 __frame_regs x27, x28, \op, 9
675
676 .ifc \op, ld
677 .if .Lframe_regcount == -1
678 .error "frame_push/frame_pop may not be nested"
679 .endif
680 ldp x29, x30, [sp], #.Lframe_local_offset + .Lframe_extra
681 .set .Lframe_regcount, -1
682 .endif
683 .endm
684
685 /*
686 * Check whether to yield to another runnable task from kernel mode NEON code
687 * (which runs with preemption disabled).
688 *
689 * if_will_cond_yield_neon
690 * // pre-yield patchup code
691 * do_cond_yield_neon
692 * // post-yield patchup code
693 * endif_yield_neon <label>
694 *
695 * where <label> is optional, and marks the point where execution will resume
696 * after a yield has been performed. If omitted, execution resumes right after
697 * the endif_yield_neon invocation. Note that the entire sequence, including
698 * the provided patchup code, will be omitted from the image if CONFIG_PREEMPT
699 * is not defined.
700 *
701 * As a convenience, in the case where no patchup code is required, the above
702 * sequence may be abbreviated to
703 *
704 * cond_yield_neon <label>
705 *
706 * Note that the patchup code does not support assembler directives that change
707 * the output section, any use of such directives is undefined.
708 *
709 * The yield itself consists of the following:
710 * - Check whether the preempt count is exactly 1 and a reschedule is also
711 * needed. If so, calling of preempt_enable() in kernel_neon_end() will
712 * trigger a reschedule. If it is not the case, yielding is pointless.
713 * - Disable and re-enable kernel mode NEON, and branch to the yield fixup
714 * code.
715 *
716 * This macro sequence may clobber all CPU state that is not guaranteed by the
717 * AAPCS to be preserved across an ordinary function call.
718 */
719
720 .macro cond_yield_neon, lbl
721 if_will_cond_yield_neon
722 do_cond_yield_neon
723 endif_yield_neon \lbl
724 .endm
725
726 .macro if_will_cond_yield_neon
727 #ifdef CONFIG_PREEMPT
728 get_current_task x0
729 ldr x0, [x0, #TSK_TI_PREEMPT]
730 sub x0, x0, #PREEMPT_DISABLE_OFFSET
731 cbz x0, .Lyield_\@
732 /* fall through to endif_yield_neon */
733 .subsection 1
734 .Lyield_\@ :
735 #else
736 .section ".discard.cond_yield_neon", "ax"
737 #endif
738 .endm
739
740 .macro do_cond_yield_neon
741 bl kernel_neon_end
742 bl kernel_neon_begin
743 .endm
744
745 .macro endif_yield_neon, lbl
746 .ifnb \lbl
747 b \lbl
748 .else
749 b .Lyield_out_\@
750 .endif
751 .previous
752 .Lyield_out_\@ :
753 .endm
754
755 #endif /* __ASM_ASSEMBLER_H */