xref: /openbmc/linux/arch/arm64/include/asm/assembler.h (revision 2209fda3)
1 /*
2  * Based on arch/arm/include/asm/assembler.h, arch/arm/mm/proc-macros.S
3  *
4  * Copyright (C) 1996-2000 Russell King
5  * Copyright (C) 2012 ARM Ltd.
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18  */
19 #ifndef __ASSEMBLY__
20 #error "Only include this from assembly code"
21 #endif
22 
23 #ifndef __ASM_ASSEMBLER_H
24 #define __ASM_ASSEMBLER_H
25 
26 #include <asm/asm-offsets.h>
27 #include <asm/cpufeature.h>
28 #include <asm/debug-monitors.h>
29 #include <asm/page.h>
30 #include <asm/pgtable-hwdef.h>
31 #include <asm/ptrace.h>
32 #include <asm/thread_info.h>
33 
34 	.macro save_and_disable_daif, flags
35 	mrs	\flags, daif
36 	msr	daifset, #0xf
37 	.endm
38 
39 	.macro disable_daif
40 	msr	daifset, #0xf
41 	.endm
42 
43 	.macro enable_daif
44 	msr	daifclr, #0xf
45 	.endm
46 
47 	.macro	restore_daif, flags:req
48 	msr	daif, \flags
49 	.endm
50 
51 	/* Only on aarch64 pstate, PSR_D_BIT is different for aarch32 */
52 	.macro	inherit_daif, pstate:req, tmp:req
53 	and	\tmp, \pstate, #(PSR_D_BIT | PSR_A_BIT | PSR_I_BIT | PSR_F_BIT)
54 	msr	daif, \tmp
55 	.endm
56 
57 	/* IRQ is the lowest priority flag, unconditionally unmask the rest. */
58 	.macro enable_da_f
59 	msr	daifclr, #(8 | 4 | 1)
60 	.endm
61 
62 /*
63  * Enable and disable interrupts.
64  */
65 	.macro	disable_irq
66 	msr	daifset, #2
67 	.endm
68 
69 	.macro	enable_irq
70 	msr	daifclr, #2
71 	.endm
72 
73 	.macro	save_and_disable_irq, flags
74 	mrs	\flags, daif
75 	msr	daifset, #2
76 	.endm
77 
78 	.macro	restore_irq, flags
79 	msr	daif, \flags
80 	.endm
81 
82 	.macro	enable_dbg
83 	msr	daifclr, #8
84 	.endm
85 
86 	.macro	disable_step_tsk, flgs, tmp
87 	tbz	\flgs, #TIF_SINGLESTEP, 9990f
88 	mrs	\tmp, mdscr_el1
89 	bic	\tmp, \tmp, #DBG_MDSCR_SS
90 	msr	mdscr_el1, \tmp
91 	isb	// Synchronise with enable_dbg
92 9990:
93 	.endm
94 
95 	/* call with daif masked */
96 	.macro	enable_step_tsk, flgs, tmp
97 	tbz	\flgs, #TIF_SINGLESTEP, 9990f
98 	mrs	\tmp, mdscr_el1
99 	orr	\tmp, \tmp, #DBG_MDSCR_SS
100 	msr	mdscr_el1, \tmp
101 9990:
102 	.endm
103 
104 /*
105  * SMP data memory barrier
106  */
107 	.macro	smp_dmb, opt
108 	dmb	\opt
109 	.endm
110 
111 /*
112  * RAS Error Synchronization barrier
113  */
114 	.macro  esb
115 	hint    #16
116 	.endm
117 
118 /*
119  * Value prediction barrier
120  */
121 	.macro	csdb
122 	hint	#20
123 	.endm
124 
125 /*
126  * Sanitise a 64-bit bounded index wrt speculation, returning zero if out
127  * of bounds.
128  */
129 	.macro	mask_nospec64, idx, limit, tmp
130 	sub	\tmp, \idx, \limit
131 	bic	\tmp, \tmp, \idx
132 	and	\idx, \idx, \tmp, asr #63
133 	csdb
134 	.endm
135 
136 /*
137  * NOP sequence
138  */
139 	.macro	nops, num
140 	.rept	\num
141 	nop
142 	.endr
143 	.endm
144 
145 /*
146  * Emit an entry into the exception table
147  */
148 	.macro		_asm_extable, from, to
149 	.pushsection	__ex_table, "a"
150 	.align		3
151 	.long		(\from - .), (\to - .)
152 	.popsection
153 	.endm
154 
155 #define USER(l, x...)				\
156 9999:	x;					\
157 	_asm_extable	9999b, l
158 
159 /*
160  * Register aliases.
161  */
162 lr	.req	x30		// link register
163 
164 /*
165  * Vector entry
166  */
167 	 .macro	ventry	label
168 	.align	7
169 	b	\label
170 	.endm
171 
172 /*
173  * Select code when configured for BE.
174  */
175 #ifdef CONFIG_CPU_BIG_ENDIAN
176 #define CPU_BE(code...) code
177 #else
178 #define CPU_BE(code...)
179 #endif
180 
181 /*
182  * Select code when configured for LE.
183  */
184 #ifdef CONFIG_CPU_BIG_ENDIAN
185 #define CPU_LE(code...)
186 #else
187 #define CPU_LE(code...) code
188 #endif
189 
190 /*
191  * Define a macro that constructs a 64-bit value by concatenating two
192  * 32-bit registers. Note that on big endian systems the order of the
193  * registers is swapped.
194  */
195 #ifndef CONFIG_CPU_BIG_ENDIAN
196 	.macro	regs_to_64, rd, lbits, hbits
197 #else
198 	.macro	regs_to_64, rd, hbits, lbits
199 #endif
200 	orr	\rd, \lbits, \hbits, lsl #32
201 	.endm
202 
203 /*
204  * Pseudo-ops for PC-relative adr/ldr/str <reg>, <symbol> where
205  * <symbol> is within the range +/- 4 GB of the PC.
206  */
207 	/*
208 	 * @dst: destination register (64 bit wide)
209 	 * @sym: name of the symbol
210 	 */
211 	.macro	adr_l, dst, sym
212 	adrp	\dst, \sym
213 	add	\dst, \dst, :lo12:\sym
214 	.endm
215 
216 	/*
217 	 * @dst: destination register (32 or 64 bit wide)
218 	 * @sym: name of the symbol
219 	 * @tmp: optional 64-bit scratch register to be used if <dst> is a
220 	 *       32-bit wide register, in which case it cannot be used to hold
221 	 *       the address
222 	 */
223 	.macro	ldr_l, dst, sym, tmp=
224 	.ifb	\tmp
225 	adrp	\dst, \sym
226 	ldr	\dst, [\dst, :lo12:\sym]
227 	.else
228 	adrp	\tmp, \sym
229 	ldr	\dst, [\tmp, :lo12:\sym]
230 	.endif
231 	.endm
232 
233 	/*
234 	 * @src: source register (32 or 64 bit wide)
235 	 * @sym: name of the symbol
236 	 * @tmp: mandatory 64-bit scratch register to calculate the address
237 	 *       while <src> needs to be preserved.
238 	 */
239 	.macro	str_l, src, sym, tmp
240 	adrp	\tmp, \sym
241 	str	\src, [\tmp, :lo12:\sym]
242 	.endm
243 
244 	/*
245 	 * @dst: Result of per_cpu(sym, smp_processor_id()) (can be SP)
246 	 * @sym: The name of the per-cpu variable
247 	 * @tmp: scratch register
248 	 */
249 	.macro adr_this_cpu, dst, sym, tmp
250 	adrp	\tmp, \sym
251 	add	\dst, \tmp, #:lo12:\sym
252 alternative_if_not ARM64_HAS_VIRT_HOST_EXTN
253 	mrs	\tmp, tpidr_el1
254 alternative_else
255 	mrs	\tmp, tpidr_el2
256 alternative_endif
257 	add	\dst, \dst, \tmp
258 	.endm
259 
260 	/*
261 	 * @dst: Result of READ_ONCE(per_cpu(sym, smp_processor_id()))
262 	 * @sym: The name of the per-cpu variable
263 	 * @tmp: scratch register
264 	 */
265 	.macro ldr_this_cpu dst, sym, tmp
266 	adr_l	\dst, \sym
267 alternative_if_not ARM64_HAS_VIRT_HOST_EXTN
268 	mrs	\tmp, tpidr_el1
269 alternative_else
270 	mrs	\tmp, tpidr_el2
271 alternative_endif
272 	ldr	\dst, [\dst, \tmp]
273 	.endm
274 
275 /*
276  * vma_vm_mm - get mm pointer from vma pointer (vma->vm_mm)
277  */
278 	.macro	vma_vm_mm, rd, rn
279 	ldr	\rd, [\rn, #VMA_VM_MM]
280 	.endm
281 
282 /*
283  * mmid - get context id from mm pointer (mm->context.id)
284  */
285 	.macro	mmid, rd, rn
286 	ldr	\rd, [\rn, #MM_CONTEXT_ID]
287 	.endm
288 /*
289  * read_ctr - read CTR_EL0. If the system has mismatched register fields,
290  * provide the system wide safe value from arm64_ftr_reg_ctrel0.sys_val
291  */
292 	.macro	read_ctr, reg
293 alternative_if_not ARM64_MISMATCHED_CACHE_TYPE
294 	mrs	\reg, ctr_el0			// read CTR
295 	nop
296 alternative_else
297 	ldr_l	\reg, arm64_ftr_reg_ctrel0 + ARM64_FTR_SYSVAL
298 alternative_endif
299 	.endm
300 
301 
302 /*
303  * raw_dcache_line_size - get the minimum D-cache line size on this CPU
304  * from the CTR register.
305  */
306 	.macro	raw_dcache_line_size, reg, tmp
307 	mrs	\tmp, ctr_el0			// read CTR
308 	ubfm	\tmp, \tmp, #16, #19		// cache line size encoding
309 	mov	\reg, #4			// bytes per word
310 	lsl	\reg, \reg, \tmp		// actual cache line size
311 	.endm
312 
313 /*
314  * dcache_line_size - get the safe D-cache line size across all CPUs
315  */
316 	.macro	dcache_line_size, reg, tmp
317 	read_ctr	\tmp
318 	ubfm		\tmp, \tmp, #16, #19	// cache line size encoding
319 	mov		\reg, #4		// bytes per word
320 	lsl		\reg, \reg, \tmp	// actual cache line size
321 	.endm
322 
323 /*
324  * raw_icache_line_size - get the minimum I-cache line size on this CPU
325  * from the CTR register.
326  */
327 	.macro	raw_icache_line_size, reg, tmp
328 	mrs	\tmp, ctr_el0			// read CTR
329 	and	\tmp, \tmp, #0xf		// cache line size encoding
330 	mov	\reg, #4			// bytes per word
331 	lsl	\reg, \reg, \tmp		// actual cache line size
332 	.endm
333 
334 /*
335  * icache_line_size - get the safe I-cache line size across all CPUs
336  */
337 	.macro	icache_line_size, reg, tmp
338 	read_ctr	\tmp
339 	and		\tmp, \tmp, #0xf	// cache line size encoding
340 	mov		\reg, #4		// bytes per word
341 	lsl		\reg, \reg, \tmp	// actual cache line size
342 	.endm
343 
344 /*
345  * tcr_set_idmap_t0sz - update TCR.T0SZ so that we can load the ID map
346  */
347 	.macro	tcr_set_idmap_t0sz, valreg, tmpreg
348 	ldr_l	\tmpreg, idmap_t0sz
349 	bfi	\valreg, \tmpreg, #TCR_T0SZ_OFFSET, #TCR_TxSZ_WIDTH
350 	.endm
351 
352 /*
353  * tcr_compute_pa_size - set TCR.(I)PS to the highest supported
354  * ID_AA64MMFR0_EL1.PARange value
355  *
356  *	tcr:		register with the TCR_ELx value to be updated
357  *	pos:		IPS or PS bitfield position
358  *	tmp{0,1}:	temporary registers
359  */
360 	.macro	tcr_compute_pa_size, tcr, pos, tmp0, tmp1
361 	mrs	\tmp0, ID_AA64MMFR0_EL1
362 	// Narrow PARange to fit the PS field in TCR_ELx
363 	ubfx	\tmp0, \tmp0, #ID_AA64MMFR0_PARANGE_SHIFT, #3
364 	mov	\tmp1, #ID_AA64MMFR0_PARANGE_MAX
365 	cmp	\tmp0, \tmp1
366 	csel	\tmp0, \tmp1, \tmp0, hi
367 	bfi	\tcr, \tmp0, \pos, #3
368 	.endm
369 
370 /*
371  * Macro to perform a data cache maintenance for the interval
372  * [kaddr, kaddr + size)
373  *
374  * 	op:		operation passed to dc instruction
375  * 	domain:		domain used in dsb instruciton
376  * 	kaddr:		starting virtual address of the region
377  * 	size:		size of the region
378  * 	Corrupts:	kaddr, size, tmp1, tmp2
379  */
380 	.macro dcache_by_line_op op, domain, kaddr, size, tmp1, tmp2
381 	dcache_line_size \tmp1, \tmp2
382 	add	\size, \kaddr, \size
383 	sub	\tmp2, \tmp1, #1
384 	bic	\kaddr, \kaddr, \tmp2
385 9998:
386 	.if	(\op == cvau || \op == cvac)
387 alternative_if_not ARM64_WORKAROUND_CLEAN_CACHE
388 	dc	\op, \kaddr
389 alternative_else
390 	dc	civac, \kaddr
391 alternative_endif
392 	.elseif	(\op == cvap)
393 alternative_if ARM64_HAS_DCPOP
394 	sys 3, c7, c12, 1, \kaddr	// dc cvap
395 alternative_else
396 	dc	cvac, \kaddr
397 alternative_endif
398 	.else
399 	dc	\op, \kaddr
400 	.endif
401 	add	\kaddr, \kaddr, \tmp1
402 	cmp	\kaddr, \size
403 	b.lo	9998b
404 	dsb	\domain
405 	.endm
406 
407 /*
408  * Macro to perform an instruction cache maintenance for the interval
409  * [start, end)
410  *
411  * 	start, end:	virtual addresses describing the region
412  *	label:		A label to branch to on user fault.
413  * 	Corrupts:	tmp1, tmp2
414  */
415 	.macro invalidate_icache_by_line start, end, tmp1, tmp2, label
416 	icache_line_size \tmp1, \tmp2
417 	sub	\tmp2, \tmp1, #1
418 	bic	\tmp2, \start, \tmp2
419 9997:
420 USER(\label, ic	ivau, \tmp2)			// invalidate I line PoU
421 	add	\tmp2, \tmp2, \tmp1
422 	cmp	\tmp2, \end
423 	b.lo	9997b
424 	dsb	ish
425 	isb
426 	.endm
427 
428 /*
429  * reset_pmuserenr_el0 - reset PMUSERENR_EL0 if PMUv3 present
430  */
431 	.macro	reset_pmuserenr_el0, tmpreg
432 	mrs	\tmpreg, id_aa64dfr0_el1	// Check ID_AA64DFR0_EL1 PMUVer
433 	sbfx	\tmpreg, \tmpreg, #8, #4
434 	cmp	\tmpreg, #1			// Skip if no PMU present
435 	b.lt	9000f
436 	msr	pmuserenr_el0, xzr		// Disable PMU access from EL0
437 9000:
438 	.endm
439 
440 /*
441  * copy_page - copy src to dest using temp registers t1-t8
442  */
443 	.macro copy_page dest:req src:req t1:req t2:req t3:req t4:req t5:req t6:req t7:req t8:req
444 9998:	ldp	\t1, \t2, [\src]
445 	ldp	\t3, \t4, [\src, #16]
446 	ldp	\t5, \t6, [\src, #32]
447 	ldp	\t7, \t8, [\src, #48]
448 	add	\src, \src, #64
449 	stnp	\t1, \t2, [\dest]
450 	stnp	\t3, \t4, [\dest, #16]
451 	stnp	\t5, \t6, [\dest, #32]
452 	stnp	\t7, \t8, [\dest, #48]
453 	add	\dest, \dest, #64
454 	tst	\src, #(PAGE_SIZE - 1)
455 	b.ne	9998b
456 	.endm
457 
458 /*
459  * Annotate a function as position independent, i.e., safe to be called before
460  * the kernel virtual mapping is activated.
461  */
462 #define ENDPIPROC(x)			\
463 	.globl	__pi_##x;		\
464 	.type 	__pi_##x, %function;	\
465 	.set	__pi_##x, x;		\
466 	.size	__pi_##x, . - x;	\
467 	ENDPROC(x)
468 
469 /*
470  * Annotate a function as being unsuitable for kprobes.
471  */
472 #ifdef CONFIG_KPROBES
473 #define NOKPROBE(x)				\
474 	.pushsection "_kprobe_blacklist", "aw";	\
475 	.quad	x;				\
476 	.popsection;
477 #else
478 #define NOKPROBE(x)
479 #endif
480 	/*
481 	 * Emit a 64-bit absolute little endian symbol reference in a way that
482 	 * ensures that it will be resolved at build time, even when building a
483 	 * PIE binary. This requires cooperation from the linker script, which
484 	 * must emit the lo32/hi32 halves individually.
485 	 */
486 	.macro	le64sym, sym
487 	.long	\sym\()_lo32
488 	.long	\sym\()_hi32
489 	.endm
490 
491 	/*
492 	 * mov_q - move an immediate constant into a 64-bit register using
493 	 *         between 2 and 4 movz/movk instructions (depending on the
494 	 *         magnitude and sign of the operand)
495 	 */
496 	.macro	mov_q, reg, val
497 	.if (((\val) >> 31) == 0 || ((\val) >> 31) == 0x1ffffffff)
498 	movz	\reg, :abs_g1_s:\val
499 	.else
500 	.if (((\val) >> 47) == 0 || ((\val) >> 47) == 0x1ffff)
501 	movz	\reg, :abs_g2_s:\val
502 	.else
503 	movz	\reg, :abs_g3:\val
504 	movk	\reg, :abs_g2_nc:\val
505 	.endif
506 	movk	\reg, :abs_g1_nc:\val
507 	.endif
508 	movk	\reg, :abs_g0_nc:\val
509 	.endm
510 
511 /*
512  * Return the current thread_info.
513  */
514 	.macro	get_thread_info, rd
515 	mrs	\rd, sp_el0
516 	.endm
517 
518 /*
519  * Arrange a physical address in a TTBR register, taking care of 52-bit
520  * addresses.
521  *
522  * 	phys:	physical address, preserved
523  * 	ttbr:	returns the TTBR value
524  */
525 	.macro	phys_to_ttbr, ttbr, phys
526 #ifdef CONFIG_ARM64_PA_BITS_52
527 	orr	\ttbr, \phys, \phys, lsr #46
528 	and	\ttbr, \ttbr, #TTBR_BADDR_MASK_52
529 #else
530 	mov	\ttbr, \phys
531 #endif
532 	.endm
533 
534 	.macro	phys_to_pte, pte, phys
535 #ifdef CONFIG_ARM64_PA_BITS_52
536 	/*
537 	 * We assume \phys is 64K aligned and this is guaranteed by only
538 	 * supporting this configuration with 64K pages.
539 	 */
540 	orr	\pte, \phys, \phys, lsr #36
541 	and	\pte, \pte, #PTE_ADDR_MASK
542 #else
543 	mov	\pte, \phys
544 #endif
545 	.endm
546 
547 	.macro	pte_to_phys, phys, pte
548 #ifdef CONFIG_ARM64_PA_BITS_52
549 	ubfiz	\phys, \pte, #(48 - 16 - 12), #16
550 	bfxil	\phys, \pte, #16, #32
551 	lsl	\phys, \phys, #16
552 #else
553 	and	\phys, \pte, #PTE_ADDR_MASK
554 #endif
555 	.endm
556 
557 /**
558  * Errata workaround prior to disable MMU. Insert an ISB immediately prior
559  * to executing the MSR that will change SCTLR_ELn[M] from a value of 1 to 0.
560  */
561 	.macro pre_disable_mmu_workaround
562 #ifdef CONFIG_QCOM_FALKOR_ERRATUM_E1041
563 	isb
564 #endif
565 	.endm
566 
567 	/*
568 	 * frame_push - Push @regcount callee saved registers to the stack,
569 	 *              starting at x19, as well as x29/x30, and set x29 to
570 	 *              the new value of sp. Add @extra bytes of stack space
571 	 *              for locals.
572 	 */
573 	.macro		frame_push, regcount:req, extra
574 	__frame		st, \regcount, \extra
575 	.endm
576 
577 	/*
578 	 * frame_pop  - Pop the callee saved registers from the stack that were
579 	 *              pushed in the most recent call to frame_push, as well
580 	 *              as x29/x30 and any extra stack space that may have been
581 	 *              allocated.
582 	 */
583 	.macro		frame_pop
584 	__frame		ld
585 	.endm
586 
587 	.macro		__frame_regs, reg1, reg2, op, num
588 	.if		.Lframe_regcount == \num
589 	\op\()r		\reg1, [sp, #(\num + 1) * 8]
590 	.elseif		.Lframe_regcount > \num
591 	\op\()p		\reg1, \reg2, [sp, #(\num + 1) * 8]
592 	.endif
593 	.endm
594 
595 	.macro		__frame, op, regcount, extra=0
596 	.ifc		\op, st
597 	.if		(\regcount) < 0 || (\regcount) > 10
598 	.error		"regcount should be in the range [0 ... 10]"
599 	.endif
600 	.if		((\extra) % 16) != 0
601 	.error		"extra should be a multiple of 16 bytes"
602 	.endif
603 	.ifdef		.Lframe_regcount
604 	.if		.Lframe_regcount != -1
605 	.error		"frame_push/frame_pop may not be nested"
606 	.endif
607 	.endif
608 	.set		.Lframe_regcount, \regcount
609 	.set		.Lframe_extra, \extra
610 	.set		.Lframe_local_offset, ((\regcount + 3) / 2) * 16
611 	stp		x29, x30, [sp, #-.Lframe_local_offset - .Lframe_extra]!
612 	mov		x29, sp
613 	.endif
614 
615 	__frame_regs	x19, x20, \op, 1
616 	__frame_regs	x21, x22, \op, 3
617 	__frame_regs	x23, x24, \op, 5
618 	__frame_regs	x25, x26, \op, 7
619 	__frame_regs	x27, x28, \op, 9
620 
621 	.ifc		\op, ld
622 	.if		.Lframe_regcount == -1
623 	.error		"frame_push/frame_pop may not be nested"
624 	.endif
625 	ldp		x29, x30, [sp], #.Lframe_local_offset + .Lframe_extra
626 	.set		.Lframe_regcount, -1
627 	.endif
628 	.endm
629 
630 /*
631  * Check whether to yield to another runnable task from kernel mode NEON code
632  * (which runs with preemption disabled).
633  *
634  * if_will_cond_yield_neon
635  *        // pre-yield patchup code
636  * do_cond_yield_neon
637  *        // post-yield patchup code
638  * endif_yield_neon    <label>
639  *
640  * where <label> is optional, and marks the point where execution will resume
641  * after a yield has been performed. If omitted, execution resumes right after
642  * the endif_yield_neon invocation. Note that the entire sequence, including
643  * the provided patchup code, will be omitted from the image if CONFIG_PREEMPT
644  * is not defined.
645  *
646  * As a convenience, in the case where no patchup code is required, the above
647  * sequence may be abbreviated to
648  *
649  * cond_yield_neon <label>
650  *
651  * Note that the patchup code does not support assembler directives that change
652  * the output section, any use of such directives is undefined.
653  *
654  * The yield itself consists of the following:
655  * - Check whether the preempt count is exactly 1, in which case disabling
656  *   preemption once will make the task preemptible. If this is not the case,
657  *   yielding is pointless.
658  * - Check whether TIF_NEED_RESCHED is set, and if so, disable and re-enable
659  *   kernel mode NEON (which will trigger a reschedule), and branch to the
660  *   yield fixup code.
661  *
662  * This macro sequence may clobber all CPU state that is not guaranteed by the
663  * AAPCS to be preserved across an ordinary function call.
664  */
665 
666 	.macro		cond_yield_neon, lbl
667 	if_will_cond_yield_neon
668 	do_cond_yield_neon
669 	endif_yield_neon	\lbl
670 	.endm
671 
672 	.macro		if_will_cond_yield_neon
673 #ifdef CONFIG_PREEMPT
674 	get_thread_info	x0
675 	ldr		w1, [x0, #TSK_TI_PREEMPT]
676 	ldr		x0, [x0, #TSK_TI_FLAGS]
677 	cmp		w1, #PREEMPT_DISABLE_OFFSET
678 	csel		x0, x0, xzr, eq
679 	tbnz		x0, #TIF_NEED_RESCHED, .Lyield_\@	// needs rescheduling?
680 	/* fall through to endif_yield_neon */
681 	.subsection	1
682 .Lyield_\@ :
683 #else
684 	.section	".discard.cond_yield_neon", "ax"
685 #endif
686 	.endm
687 
688 	.macro		do_cond_yield_neon
689 	bl		kernel_neon_end
690 	bl		kernel_neon_begin
691 	.endm
692 
693 	.macro		endif_yield_neon, lbl
694 	.ifnb		\lbl
695 	b		\lbl
696 	.else
697 	b		.Lyield_out_\@
698 	.endif
699 	.previous
700 .Lyield_out_\@ :
701 	.endm
702 
703 #endif	/* __ASM_ASSEMBLER_H */
704