xref: /linux/arch/arm64/kernel/entry.S (revision 31354121bf03dac6498a4236928a38490745d601)
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * Low-level exception handling code
4 *
5 * Copyright (C) 2012 ARM Ltd.
6 * Authors:	Catalin Marinas <catalin.marinas@arm.com>
7 *		Will Deacon <will.deacon@arm.com>
8 */
9
10#include <linux/arm-smccc.h>
11#include <linux/init.h>
12#include <linux/linkage.h>
13
14#include <asm/alternative.h>
15#include <asm/assembler.h>
16#include <asm/asm-offsets.h>
17#include <asm/asm_pointer_auth.h>
18#include <asm/bug.h>
19#include <asm/cpufeature.h>
20#include <asm/errno.h>
21#include <asm/esr.h>
22#include <asm/irq.h>
23#include <asm/memory.h>
24#include <asm/mmu.h>
25#include <asm/processor.h>
26#include <asm/ptrace.h>
27#include <asm/scs.h>
28#include <asm/thread_info.h>
29#include <asm/asm-uaccess.h>
30#include <asm/unistd.h>
31
32	.macro	clear_gp_regs
33	.irp	n,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29
34	mov	x\n, xzr
35	.endr
36	.endm
37
38	.macro kernel_ventry, el:req, ht:req, regsize:req, label:req
39	.align 7
40.Lventry_start\@:
41	.if	\el == 0
42	/*
43	 * This must be the first instruction of the EL0 vector entries. It is
44	 * skipped by the trampoline vectors, to trigger the cleanup.
45	 */
46	b	.Lskip_tramp_vectors_cleanup\@
47	.if	\regsize == 64
48	mrs	x30, tpidrro_el0
49	msr	tpidrro_el0, xzr
50	.else
51	mov	x30, xzr
52	.endif
53.Lskip_tramp_vectors_cleanup\@:
54	.endif
55
56	sub	sp, sp, #PT_REGS_SIZE
57#ifdef CONFIG_VMAP_STACK
58	/*
59	 * Test whether the SP has overflowed, without corrupting a GPR.
60	 * Task and IRQ stacks are aligned so that SP & (1 << THREAD_SHIFT)
61	 * should always be zero.
62	 */
63	add	sp, sp, x0			// sp' = sp + x0
64	sub	x0, sp, x0			// x0' = sp' - x0 = (sp + x0) - x0 = sp
65	tbnz	x0, #THREAD_SHIFT, 0f
66	sub	x0, sp, x0			// x0'' = sp' - x0' = (sp + x0) - sp = x0
67	sub	sp, sp, x0			// sp'' = sp' - x0 = (sp + x0) - x0 = sp
68	b	el\el\ht\()_\regsize\()_\label
69
700:
71	/*
72	 * Either we've just detected an overflow, or we've taken an exception
73	 * while on the overflow stack. Either way, we won't return to
74	 * userspace, and can clobber EL0 registers to free up GPRs.
75	 */
76
77	/* Stash the original SP (minus PT_REGS_SIZE) in tpidr_el0. */
78	msr	tpidr_el0, x0
79
80	/* Recover the original x0 value and stash it in tpidrro_el0 */
81	sub	x0, sp, x0
82	msr	tpidrro_el0, x0
83
84	/* Switch to the overflow stack */
85	adr_this_cpu sp, overflow_stack + OVERFLOW_STACK_SIZE, x0
86
87	/*
88	 * Check whether we were already on the overflow stack. This may happen
89	 * after panic() re-enables interrupts.
90	 */
91	mrs	x0, tpidr_el0			// sp of interrupted context
92	sub	x0, sp, x0			// delta with top of overflow stack
93	tst	x0, #~(OVERFLOW_STACK_SIZE - 1)	// within range?
94	b.ne	__bad_stack			// no? -> bad stack pointer
95
96	/* We were already on the overflow stack. Restore sp/x0 and carry on. */
97	sub	sp, sp, x0
98	mrs	x0, tpidrro_el0
99#endif
100	b	el\el\ht\()_\regsize\()_\label
101.org .Lventry_start\@ + 128	// Did we overflow the ventry slot?
102	.endm
103
104	.macro tramp_alias, dst, sym, tmp
105	mov_q	\dst, TRAMP_VALIAS
106	adr_l	\tmp, \sym
107	add	\dst, \dst, \tmp
108	adr_l	\tmp, .entry.tramp.text
109	sub	\dst, \dst, \tmp
110	.endm
111
112	/*
113	 * This macro corrupts x0-x3. It is the caller's duty  to save/restore
114	 * them if required.
115	 */
116	.macro	apply_ssbd, state, tmp1, tmp2
117alternative_cb	ARM64_ALWAYS_SYSTEM, spectre_v4_patch_fw_mitigation_enable
118	b	.L__asm_ssbd_skip\@		// Patched to NOP
119alternative_cb_end
120	ldr_this_cpu	\tmp2, arm64_ssbd_callback_required, \tmp1
121	cbz	\tmp2,	.L__asm_ssbd_skip\@
122	ldr	\tmp2, [tsk, #TSK_TI_FLAGS]
123	tbnz	\tmp2, #TIF_SSBD, .L__asm_ssbd_skip\@
124	mov	w0, #ARM_SMCCC_ARCH_WORKAROUND_2
125	mov	w1, #\state
126alternative_cb	ARM64_ALWAYS_SYSTEM, smccc_patch_fw_mitigation_conduit
127	nop					// Patched to SMC/HVC #0
128alternative_cb_end
129.L__asm_ssbd_skip\@:
130	.endm
131
132	/* Check for MTE asynchronous tag check faults */
133	.macro check_mte_async_tcf, tmp, ti_flags, thread_sctlr
134#ifdef CONFIG_ARM64_MTE
135	.arch_extension lse
136alternative_if_not ARM64_MTE
137	b	1f
138alternative_else_nop_endif
139	/*
140	 * Asynchronous tag check faults are only possible in ASYNC (2) or
141	 * ASYM (3) modes. In each of these modes bit 1 of SCTLR_EL1.TCF0 is
142	 * set, so skip the check if it is unset.
143	 */
144	tbz	\thread_sctlr, #(SCTLR_EL1_TCF0_SHIFT + 1), 1f
145	mrs_s	\tmp, SYS_TFSRE0_EL1
146	tbz	\tmp, #SYS_TFSR_EL1_TF0_SHIFT, 1f
147	/* Asynchronous TCF occurred for TTBR0 access, set the TI flag */
148	mov	\tmp, #_TIF_MTE_ASYNC_FAULT
149	add	\ti_flags, tsk, #TSK_TI_FLAGS
150	stset	\tmp, [\ti_flags]
1511:
152#endif
153	.endm
154
155	/* Clear the MTE asynchronous tag check faults */
156	.macro clear_mte_async_tcf thread_sctlr
157#ifdef CONFIG_ARM64_MTE
158alternative_if ARM64_MTE
159	/* See comment in check_mte_async_tcf above. */
160	tbz	\thread_sctlr, #(SCTLR_EL1_TCF0_SHIFT + 1), 1f
161	dsb	ish
162	msr_s	SYS_TFSRE0_EL1, xzr
1631:
164alternative_else_nop_endif
165#endif
166	.endm
167
168	.macro mte_set_gcr, mte_ctrl, tmp
169#ifdef CONFIG_ARM64_MTE
170	ubfx	\tmp, \mte_ctrl, #MTE_CTRL_GCR_USER_EXCL_SHIFT, #16
171	orr	\tmp, \tmp, #SYS_GCR_EL1_RRND
172	msr_s	SYS_GCR_EL1, \tmp
173#endif
174	.endm
175
176	.macro mte_set_kernel_gcr, tmp, tmp2
177#ifdef CONFIG_KASAN_HW_TAGS
178alternative_cb	ARM64_ALWAYS_SYSTEM, kasan_hw_tags_enable
179	b	1f
180alternative_cb_end
181	mov	\tmp, KERNEL_GCR_EL1
182	msr_s	SYS_GCR_EL1, \tmp
1831:
184#endif
185	.endm
186
187	.macro mte_set_user_gcr, tsk, tmp, tmp2
188#ifdef CONFIG_KASAN_HW_TAGS
189alternative_cb	ARM64_ALWAYS_SYSTEM, kasan_hw_tags_enable
190	b	1f
191alternative_cb_end
192	ldr	\tmp, [\tsk, #THREAD_MTE_CTRL]
193
194	mte_set_gcr \tmp, \tmp2
1951:
196#endif
197	.endm
198
199	.macro	kernel_entry, el, regsize = 64
200	.if	\regsize == 32
201	mov	w0, w0				// zero upper 32 bits of x0
202	.endif
203	stp	x0, x1, [sp, #16 * 0]
204	stp	x2, x3, [sp, #16 * 1]
205	stp	x4, x5, [sp, #16 * 2]
206	stp	x6, x7, [sp, #16 * 3]
207	stp	x8, x9, [sp, #16 * 4]
208	stp	x10, x11, [sp, #16 * 5]
209	stp	x12, x13, [sp, #16 * 6]
210	stp	x14, x15, [sp, #16 * 7]
211	stp	x16, x17, [sp, #16 * 8]
212	stp	x18, x19, [sp, #16 * 9]
213	stp	x20, x21, [sp, #16 * 10]
214	stp	x22, x23, [sp, #16 * 11]
215	stp	x24, x25, [sp, #16 * 12]
216	stp	x26, x27, [sp, #16 * 13]
217	stp	x28, x29, [sp, #16 * 14]
218
219	.if	\el == 0
220	clear_gp_regs
221	mrs	x21, sp_el0
222	ldr_this_cpu	tsk, __entry_task, x20
223	msr	sp_el0, tsk
224
225	/*
226	 * Ensure MDSCR_EL1.SS is clear, since we can unmask debug exceptions
227	 * when scheduling.
228	 */
229	ldr	x19, [tsk, #TSK_TI_FLAGS]
230	disable_step_tsk x19, x20
231
232	/* Check for asynchronous tag check faults in user space */
233	ldr	x0, [tsk, THREAD_SCTLR_USER]
234	check_mte_async_tcf x22, x23, x0
235
236#ifdef CONFIG_ARM64_PTR_AUTH
237alternative_if ARM64_HAS_ADDRESS_AUTH
238	/*
239	 * Enable IA for in-kernel PAC if the task had it disabled. Although
240	 * this could be implemented with an unconditional MRS which would avoid
241	 * a load, this was measured to be slower on Cortex-A75 and Cortex-A76.
242	 *
243	 * Install the kernel IA key only if IA was enabled in the task. If IA
244	 * was disabled on kernel exit then we would have left the kernel IA
245	 * installed so there is no need to install it again.
246	 */
247	tbz	x0, SCTLR_ELx_ENIA_SHIFT, 1f
248	__ptrauth_keys_install_kernel_nosync tsk, x20, x22, x23
249	b	2f
2501:
251	mrs	x0, sctlr_el1
252	orr	x0, x0, SCTLR_ELx_ENIA
253	msr	sctlr_el1, x0
2542:
255alternative_else_nop_endif
256#endif
257
258	apply_ssbd 1, x22, x23
259
260	mte_set_kernel_gcr x22, x23
261
262	/*
263	 * Any non-self-synchronizing system register updates required for
264	 * kernel entry should be placed before this point.
265	 */
266alternative_if ARM64_MTE
267	isb
268	b	1f
269alternative_else_nop_endif
270alternative_if ARM64_HAS_ADDRESS_AUTH
271	isb
272alternative_else_nop_endif
2731:
274
275	scs_load tsk
276	.else
277	add	x21, sp, #PT_REGS_SIZE
278	get_current_task tsk
279	.endif /* \el == 0 */
280	mrs	x22, elr_el1
281	mrs	x23, spsr_el1
282	stp	lr, x21, [sp, #S_LR]
283
284	/*
285	 * For exceptions from EL0, create a final frame record.
286	 * For exceptions from EL1, create a synthetic frame record so the
287	 * interrupted code shows up in the backtrace.
288	 */
289	.if \el == 0
290	stp	xzr, xzr, [sp, #S_STACKFRAME]
291	.else
292	stp	x29, x22, [sp, #S_STACKFRAME]
293	.endif
294	add	x29, sp, #S_STACKFRAME
295
296#ifdef CONFIG_ARM64_SW_TTBR0_PAN
297alternative_if_not ARM64_HAS_PAN
298	bl	__swpan_entry_el\el
299alternative_else_nop_endif
300#endif
301
302	stp	x22, x23, [sp, #S_PC]
303
304	/* Not in a syscall by default (el0_svc overwrites for real syscall) */
305	.if	\el == 0
306	mov	w21, #NO_SYSCALL
307	str	w21, [sp, #S_SYSCALLNO]
308	.endif
309
310#ifdef CONFIG_ARM64_PSEUDO_NMI
311	/* Save pmr */
312alternative_if ARM64_HAS_IRQ_PRIO_MASKING
313	mrs_s	x20, SYS_ICC_PMR_EL1
314	str	x20, [sp, #S_PMR_SAVE]
315	mov	x20, #GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET
316	msr_s	SYS_ICC_PMR_EL1, x20
317alternative_else_nop_endif
318#endif
319
320	/*
321	 * Registers that may be useful after this macro is invoked:
322	 *
323	 * x20 - ICC_PMR_EL1
324	 * x21 - aborted SP
325	 * x22 - aborted PC
326	 * x23 - aborted PSTATE
327	*/
328	.endm
329
330	.macro	kernel_exit, el
331	.if	\el != 0
332	disable_daif
333	.endif
334
335#ifdef CONFIG_ARM64_PSEUDO_NMI
336	/* Restore pmr */
337alternative_if ARM64_HAS_IRQ_PRIO_MASKING
338	ldr	x20, [sp, #S_PMR_SAVE]
339	msr_s	SYS_ICC_PMR_EL1, x20
340	mrs_s	x21, SYS_ICC_CTLR_EL1
341	tbz	x21, #6, .L__skip_pmr_sync\@	// Check for ICC_CTLR_EL1.PMHE
342	dsb	sy				// Ensure priority change is seen by redistributor
343.L__skip_pmr_sync\@:
344alternative_else_nop_endif
345#endif
346
347	ldp	x21, x22, [sp, #S_PC]		// load ELR, SPSR
348
349#ifdef CONFIG_ARM64_SW_TTBR0_PAN
350alternative_if_not ARM64_HAS_PAN
351	bl	__swpan_exit_el\el
352alternative_else_nop_endif
353#endif
354
355	.if	\el == 0
356	ldr	x23, [sp, #S_SP]		// load return stack pointer
357	msr	sp_el0, x23
358	tst	x22, #PSR_MODE32_BIT		// native task?
359	b.eq	3f
360
361#ifdef CONFIG_ARM64_ERRATUM_845719
362alternative_if ARM64_WORKAROUND_845719
363#ifdef CONFIG_PID_IN_CONTEXTIDR
364	mrs	x29, contextidr_el1
365	msr	contextidr_el1, x29
366#else
367	msr contextidr_el1, xzr
368#endif
369alternative_else_nop_endif
370#endif
3713:
372	scs_save tsk
373
374	/* Ignore asynchronous tag check faults in the uaccess routines */
375	ldr	x0, [tsk, THREAD_SCTLR_USER]
376	clear_mte_async_tcf x0
377
378#ifdef CONFIG_ARM64_PTR_AUTH
379alternative_if ARM64_HAS_ADDRESS_AUTH
380	/*
381	 * IA was enabled for in-kernel PAC. Disable it now if needed, or
382	 * alternatively install the user's IA. All other per-task keys and
383	 * SCTLR bits were updated on task switch.
384	 *
385	 * No kernel C function calls after this.
386	 */
387	tbz	x0, SCTLR_ELx_ENIA_SHIFT, 1f
388	__ptrauth_keys_install_user tsk, x0, x1, x2
389	b	2f
3901:
391	mrs	x0, sctlr_el1
392	bic	x0, x0, SCTLR_ELx_ENIA
393	msr	sctlr_el1, x0
3942:
395alternative_else_nop_endif
396#endif
397
398	mte_set_user_gcr tsk, x0, x1
399
400	apply_ssbd 0, x0, x1
401	.endif
402
403	msr	elr_el1, x21			// set up the return data
404	msr	spsr_el1, x22
405	ldp	x0, x1, [sp, #16 * 0]
406	ldp	x2, x3, [sp, #16 * 1]
407	ldp	x4, x5, [sp, #16 * 2]
408	ldp	x6, x7, [sp, #16 * 3]
409	ldp	x8, x9, [sp, #16 * 4]
410	ldp	x10, x11, [sp, #16 * 5]
411	ldp	x12, x13, [sp, #16 * 6]
412	ldp	x14, x15, [sp, #16 * 7]
413	ldp	x16, x17, [sp, #16 * 8]
414	ldp	x18, x19, [sp, #16 * 9]
415	ldp	x20, x21, [sp, #16 * 10]
416	ldp	x22, x23, [sp, #16 * 11]
417	ldp	x24, x25, [sp, #16 * 12]
418	ldp	x26, x27, [sp, #16 * 13]
419	ldp	x28, x29, [sp, #16 * 14]
420
421	.if	\el == 0
422alternative_if_not ARM64_UNMAP_KERNEL_AT_EL0
423	ldr	lr, [sp, #S_LR]
424	add	sp, sp, #PT_REGS_SIZE		// restore sp
425	eret
426alternative_else_nop_endif
427#ifdef CONFIG_UNMAP_KERNEL_AT_EL0
428	bne	4f
429	msr	far_el1, x29
430	tramp_alias	x30, tramp_exit_native, x29
431	br	x30
4324:
433	tramp_alias	x30, tramp_exit_compat, x29
434	br	x30
435#endif
436	.else
437	ldr	lr, [sp, #S_LR]
438	add	sp, sp, #PT_REGS_SIZE		// restore sp
439
440	/* Ensure any device/NC reads complete */
441	alternative_insn nop, "dmb sy", ARM64_WORKAROUND_1508412
442
443	eret
444	.endif
445	sb
446	.endm
447
448#ifdef CONFIG_ARM64_SW_TTBR0_PAN
449	/*
450	 * Set the TTBR0 PAN bit in SPSR. When the exception is taken from
451	 * EL0, there is no need to check the state of TTBR0_EL1 since
452	 * accesses are always enabled.
453	 * Note that the meaning of this bit differs from the ARMv8.1 PAN
454	 * feature as all TTBR0_EL1 accesses are disabled, not just those to
455	 * user mappings.
456	 */
457SYM_CODE_START_LOCAL(__swpan_entry_el1)
458	mrs	x21, ttbr0_el1
459	tst	x21, #TTBR_ASID_MASK		// Check for the reserved ASID
460	orr	x23, x23, #PSR_PAN_BIT		// Set the emulated PAN in the saved SPSR
461	b.eq	1f				// TTBR0 access already disabled
462	and	x23, x23, #~PSR_PAN_BIT		// Clear the emulated PAN in the saved SPSR
463SYM_INNER_LABEL(__swpan_entry_el0, SYM_L_LOCAL)
464	__uaccess_ttbr0_disable x21
4651:	ret
466SYM_CODE_END(__swpan_entry_el1)
467
468	/*
469	 * Restore access to TTBR0_EL1. If returning to EL0, no need for SPSR
470	 * PAN bit checking.
471	 */
472SYM_CODE_START_LOCAL(__swpan_exit_el1)
473	tbnz	x22, #22, 1f			// Skip re-enabling TTBR0 access if the PSR_PAN_BIT is set
474	__uaccess_ttbr0_enable x0, x1
4751:	and	x22, x22, #~PSR_PAN_BIT		// ARMv8.0 CPUs do not understand this bit
476	ret
477SYM_CODE_END(__swpan_exit_el1)
478
479SYM_CODE_START_LOCAL(__swpan_exit_el0)
480	__uaccess_ttbr0_enable x0, x1
481	/*
482	 * Enable errata workarounds only if returning to user. The only
483	 * workaround currently required for TTBR0_EL1 changes are for the
484	 * Cavium erratum 27456 (broadcast TLBI instructions may cause I-cache
485	 * corruption).
486	 */
487	b	post_ttbr_update_workaround
488SYM_CODE_END(__swpan_exit_el0)
489#endif
490
491/* GPRs used by entry code */
492tsk	.req	x28		// current thread_info
493
494	.text
495
496/*
497 * Exception vectors.
498 */
499	.pushsection ".entry.text", "ax"
500
501	.align	11
502SYM_CODE_START(vectors)
503	kernel_ventry	1, t, 64, sync		// Synchronous EL1t
504	kernel_ventry	1, t, 64, irq		// IRQ EL1t
505	kernel_ventry	1, t, 64, fiq		// FIQ EL1t
506	kernel_ventry	1, t, 64, error		// Error EL1t
507
508	kernel_ventry	1, h, 64, sync		// Synchronous EL1h
509	kernel_ventry	1, h, 64, irq		// IRQ EL1h
510	kernel_ventry	1, h, 64, fiq		// FIQ EL1h
511	kernel_ventry	1, h, 64, error		// Error EL1h
512
513	kernel_ventry	0, t, 64, sync		// Synchronous 64-bit EL0
514	kernel_ventry	0, t, 64, irq		// IRQ 64-bit EL0
515	kernel_ventry	0, t, 64, fiq		// FIQ 64-bit EL0
516	kernel_ventry	0, t, 64, error		// Error 64-bit EL0
517
518	kernel_ventry	0, t, 32, sync		// Synchronous 32-bit EL0
519	kernel_ventry	0, t, 32, irq		// IRQ 32-bit EL0
520	kernel_ventry	0, t, 32, fiq		// FIQ 32-bit EL0
521	kernel_ventry	0, t, 32, error		// Error 32-bit EL0
522SYM_CODE_END(vectors)
523
524#ifdef CONFIG_VMAP_STACK
525SYM_CODE_START_LOCAL(__bad_stack)
526	/*
527	 * We detected an overflow in kernel_ventry, which switched to the
528	 * overflow stack. Stash the exception regs, and head to our overflow
529	 * handler.
530	 */
531
532	/* Restore the original x0 value */
533	mrs	x0, tpidrro_el0
534
535	/*
536	 * Store the original GPRs to the new stack. The orginal SP (minus
537	 * PT_REGS_SIZE) was stashed in tpidr_el0 by kernel_ventry.
538	 */
539	sub	sp, sp, #PT_REGS_SIZE
540	kernel_entry 1
541	mrs	x0, tpidr_el0
542	add	x0, x0, #PT_REGS_SIZE
543	str	x0, [sp, #S_SP]
544
545	/* Stash the regs for handle_bad_stack */
546	mov	x0, sp
547
548	/* Time to die */
549	bl	handle_bad_stack
550	ASM_BUG()
551SYM_CODE_END(__bad_stack)
552#endif /* CONFIG_VMAP_STACK */
553
554
555	.macro entry_handler el:req, ht:req, regsize:req, label:req
556SYM_CODE_START_LOCAL(el\el\ht\()_\regsize\()_\label)
557	kernel_entry \el, \regsize
558	mov	x0, sp
559	bl	el\el\ht\()_\regsize\()_\label\()_handler
560	.if \el == 0
561	b	ret_to_user
562	.else
563	b	ret_to_kernel
564	.endif
565SYM_CODE_END(el\el\ht\()_\regsize\()_\label)
566	.endm
567
568/*
569 * Early exception handlers
570 */
571	entry_handler	1, t, 64, sync
572	entry_handler	1, t, 64, irq
573	entry_handler	1, t, 64, fiq
574	entry_handler	1, t, 64, error
575
576	entry_handler	1, h, 64, sync
577	entry_handler	1, h, 64, irq
578	entry_handler	1, h, 64, fiq
579	entry_handler	1, h, 64, error
580
581	entry_handler	0, t, 64, sync
582	entry_handler	0, t, 64, irq
583	entry_handler	0, t, 64, fiq
584	entry_handler	0, t, 64, error
585
586	entry_handler	0, t, 32, sync
587	entry_handler	0, t, 32, irq
588	entry_handler	0, t, 32, fiq
589	entry_handler	0, t, 32, error
590
591SYM_CODE_START_LOCAL(ret_to_kernel)
592	kernel_exit 1
593SYM_CODE_END(ret_to_kernel)
594
595SYM_CODE_START_LOCAL(ret_to_user)
596	ldr	x19, [tsk, #TSK_TI_FLAGS]	// re-check for single-step
597	enable_step_tsk x19, x2
598#ifdef CONFIG_GCC_PLUGIN_STACKLEAK
599	bl	stackleak_erase_on_task_stack
600#endif
601	kernel_exit 0
602SYM_CODE_END(ret_to_user)
603
604	.popsection				// .entry.text
605
606	// Move from tramp_pg_dir to swapper_pg_dir
607	.macro tramp_map_kernel, tmp
608	mrs	\tmp, ttbr1_el1
609	add	\tmp, \tmp, #TRAMP_SWAPPER_OFFSET
610	bic	\tmp, \tmp, #USER_ASID_FLAG
611	msr	ttbr1_el1, \tmp
612#ifdef CONFIG_QCOM_FALKOR_ERRATUM_1003
613alternative_if ARM64_WORKAROUND_QCOM_FALKOR_E1003
614	/* ASID already in \tmp[63:48] */
615	movk	\tmp, #:abs_g2_nc:(TRAMP_VALIAS >> 12)
616	movk	\tmp, #:abs_g1_nc:(TRAMP_VALIAS >> 12)
617	/* 2MB boundary containing the vectors, so we nobble the walk cache */
618	movk	\tmp, #:abs_g0_nc:((TRAMP_VALIAS & ~(SZ_2M - 1)) >> 12)
619	isb
620	tlbi	vae1, \tmp
621	dsb	nsh
622alternative_else_nop_endif
623#endif /* CONFIG_QCOM_FALKOR_ERRATUM_1003 */
624	.endm
625
626	// Move from swapper_pg_dir to tramp_pg_dir
627	.macro tramp_unmap_kernel, tmp
628	mrs	\tmp, ttbr1_el1
629	sub	\tmp, \tmp, #TRAMP_SWAPPER_OFFSET
630	orr	\tmp, \tmp, #USER_ASID_FLAG
631	msr	ttbr1_el1, \tmp
632	/*
633	 * We avoid running the post_ttbr_update_workaround here because
634	 * it's only needed by Cavium ThunderX, which requires KPTI to be
635	 * disabled.
636	 */
637	.endm
638
639	.macro		tramp_data_read_var	dst, var
640#ifdef CONFIG_RELOCATABLE
641	ldr		\dst, .L__tramp_data_\var
642	.ifndef		.L__tramp_data_\var
643	.pushsection	".entry.tramp.rodata", "a", %progbits
644	.align		3
645.L__tramp_data_\var:
646	.quad		\var
647	.popsection
648	.endif
649#else
650	/*
651	 * As !RELOCATABLE implies !RANDOMIZE_BASE the address is always a
652	 * compile time constant (and hence not secret and not worth hiding).
653	 *
654	 * As statically allocated kernel code and data always live in the top
655	 * 47 bits of the address space we can sign-extend bit 47 and avoid an
656	 * instruction to load the upper 16 bits (which must be 0xFFFF).
657	 */
658	movz		\dst, :abs_g2_s:\var
659	movk		\dst, :abs_g1_nc:\var
660	movk		\dst, :abs_g0_nc:\var
661#endif
662	.endm
663
664#define BHB_MITIGATION_NONE	0
665#define BHB_MITIGATION_LOOP	1
666#define BHB_MITIGATION_FW	2
667#define BHB_MITIGATION_INSN	3
668
669	.macro tramp_ventry, vector_start, regsize, kpti, bhb
670	.align	7
6711:
672	.if	\regsize == 64
673	msr	tpidrro_el0, x30	// Restored in kernel_ventry
674	.endif
675
676	.if	\bhb == BHB_MITIGATION_LOOP
677	/*
678	 * This sequence must appear before the first indirect branch. i.e. the
679	 * ret out of tramp_ventry. It appears here because x30 is free.
680	 */
681	__mitigate_spectre_bhb_loop	x30
682	.endif // \bhb == BHB_MITIGATION_LOOP
683
684	.if	\bhb == BHB_MITIGATION_INSN
685	clearbhb
686	isb
687	.endif // \bhb == BHB_MITIGATION_INSN
688
689	.if	\kpti == 1
690	/*
691	 * Defend against branch aliasing attacks by pushing a dummy
692	 * entry onto the return stack and using a RET instruction to
693	 * enter the full-fat kernel vectors.
694	 */
695	bl	2f
696	b	.
6972:
698	tramp_map_kernel	x30
699alternative_insn isb, nop, ARM64_WORKAROUND_QCOM_FALKOR_E1003
700	tramp_data_read_var	x30, vectors
701alternative_if_not ARM64_WORKAROUND_CAVIUM_TX2_219_PRFM
702	prfm	plil1strm, [x30, #(1b - \vector_start)]
703alternative_else_nop_endif
704
705	msr	vbar_el1, x30
706	isb
707	.else
708	adr_l	x30, vectors
709	.endif // \kpti == 1
710
711	.if	\bhb == BHB_MITIGATION_FW
712	/*
713	 * The firmware sequence must appear before the first indirect branch.
714	 * i.e. the ret out of tramp_ventry. But it also needs the stack to be
715	 * mapped to save/restore the registers the SMC clobbers.
716	 */
717	__mitigate_spectre_bhb_fw
718	.endif // \bhb == BHB_MITIGATION_FW
719
720	add	x30, x30, #(1b - \vector_start + 4)
721	ret
722.org 1b + 128	// Did we overflow the ventry slot?
723	.endm
724
725	.macro tramp_exit, regsize = 64
726	tramp_data_read_var	x30, this_cpu_vector
727	get_this_cpu_offset x29
728	ldr	x30, [x30, x29]
729
730	msr	vbar_el1, x30
731	ldr	lr, [sp, #S_LR]
732	tramp_unmap_kernel	x29
733	.if	\regsize == 64
734	mrs	x29, far_el1
735	.endif
736	add	sp, sp, #PT_REGS_SIZE		// restore sp
737	eret
738	sb
739	.endm
740
741	.macro	generate_tramp_vector,	kpti, bhb
742.Lvector_start\@:
743	.space	0x400
744
745	.rept	4
746	tramp_ventry	.Lvector_start\@, 64, \kpti, \bhb
747	.endr
748	.rept	4
749	tramp_ventry	.Lvector_start\@, 32, \kpti, \bhb
750	.endr
751	.endm
752
753#ifdef CONFIG_UNMAP_KERNEL_AT_EL0
754/*
755 * Exception vectors trampoline.
756 * The order must match __bp_harden_el1_vectors and the
757 * arm64_bp_harden_el1_vectors enum.
758 */
759	.pushsection ".entry.tramp.text", "ax"
760	.align	11
761SYM_CODE_START_NOALIGN(tramp_vectors)
762#ifdef CONFIG_MITIGATE_SPECTRE_BRANCH_HISTORY
763	generate_tramp_vector	kpti=1, bhb=BHB_MITIGATION_LOOP
764	generate_tramp_vector	kpti=1, bhb=BHB_MITIGATION_FW
765	generate_tramp_vector	kpti=1, bhb=BHB_MITIGATION_INSN
766#endif /* CONFIG_MITIGATE_SPECTRE_BRANCH_HISTORY */
767	generate_tramp_vector	kpti=1, bhb=BHB_MITIGATION_NONE
768SYM_CODE_END(tramp_vectors)
769
770SYM_CODE_START(tramp_exit_native)
771	tramp_exit
772SYM_CODE_END(tramp_exit_native)
773
774SYM_CODE_START(tramp_exit_compat)
775	tramp_exit	32
776SYM_CODE_END(tramp_exit_compat)
777	.popsection				// .entry.tramp.text
778#endif /* CONFIG_UNMAP_KERNEL_AT_EL0 */
779
780/*
781 * Exception vectors for spectre mitigations on entry from EL1 when
782 * kpti is not in use.
783 */
784	.macro generate_el1_vector, bhb
785.Lvector_start\@:
786	kernel_ventry	1, t, 64, sync		// Synchronous EL1t
787	kernel_ventry	1, t, 64, irq		// IRQ EL1t
788	kernel_ventry	1, t, 64, fiq		// FIQ EL1h
789	kernel_ventry	1, t, 64, error		// Error EL1t
790
791	kernel_ventry	1, h, 64, sync		// Synchronous EL1h
792	kernel_ventry	1, h, 64, irq		// IRQ EL1h
793	kernel_ventry	1, h, 64, fiq		// FIQ EL1h
794	kernel_ventry	1, h, 64, error		// Error EL1h
795
796	.rept	4
797	tramp_ventry	.Lvector_start\@, 64, 0, \bhb
798	.endr
799	.rept 4
800	tramp_ventry	.Lvector_start\@, 32, 0, \bhb
801	.endr
802	.endm
803
804/* The order must match tramp_vecs and the arm64_bp_harden_el1_vectors enum. */
805	.pushsection ".entry.text", "ax"
806	.align	11
807SYM_CODE_START(__bp_harden_el1_vectors)
808#ifdef CONFIG_MITIGATE_SPECTRE_BRANCH_HISTORY
809	generate_el1_vector	bhb=BHB_MITIGATION_LOOP
810	generate_el1_vector	bhb=BHB_MITIGATION_FW
811	generate_el1_vector	bhb=BHB_MITIGATION_INSN
812#endif /* CONFIG_MITIGATE_SPECTRE_BRANCH_HISTORY */
813SYM_CODE_END(__bp_harden_el1_vectors)
814	.popsection
815
816
817/*
818 * Register switch for AArch64. The callee-saved registers need to be saved
819 * and restored. On entry:
820 *   x0 = previous task_struct (must be preserved across the switch)
821 *   x1 = next task_struct
822 * Previous and next are guaranteed not to be the same.
823 *
824 */
825SYM_FUNC_START(cpu_switch_to)
826	mov	x10, #THREAD_CPU_CONTEXT
827	add	x8, x0, x10
828	mov	x9, sp
829	stp	x19, x20, [x8], #16		// store callee-saved registers
830	stp	x21, x22, [x8], #16
831	stp	x23, x24, [x8], #16
832	stp	x25, x26, [x8], #16
833	stp	x27, x28, [x8], #16
834	stp	x29, x9, [x8], #16
835	str	lr, [x8]
836	add	x8, x1, x10
837	ldp	x19, x20, [x8], #16		// restore callee-saved registers
838	ldp	x21, x22, [x8], #16
839	ldp	x23, x24, [x8], #16
840	ldp	x25, x26, [x8], #16
841	ldp	x27, x28, [x8], #16
842	ldp	x29, x9, [x8], #16
843	ldr	lr, [x8]
844	mov	sp, x9
845	msr	sp_el0, x1
846	ptrauth_keys_install_kernel x1, x8, x9, x10
847	scs_save x0
848	scs_load x1
849	ret
850SYM_FUNC_END(cpu_switch_to)
851NOKPROBE(cpu_switch_to)
852
853/*
854 * This is how we return from a fork.
855 */
856SYM_CODE_START(ret_from_fork)
857	bl	schedule_tail
858	cbz	x19, 1f				// not a kernel thread
859	mov	x0, x20
860	blr	x19
8611:	get_current_task tsk
862	mov	x0, sp
863	bl	asm_exit_to_user_mode
864	b	ret_to_user
865SYM_CODE_END(ret_from_fork)
866NOKPROBE(ret_from_fork)
867
868/*
869 * void call_on_irq_stack(struct pt_regs *regs,
870 * 		          void (*func)(struct pt_regs *));
871 *
872 * Calls func(regs) using this CPU's irq stack and shadow irq stack.
873 */
874SYM_FUNC_START(call_on_irq_stack)
875#ifdef CONFIG_SHADOW_CALL_STACK
876	stp	scs_sp, xzr, [sp, #-16]!
877	ldr_this_cpu scs_sp, irq_shadow_call_stack_ptr, x17
878#endif
879	/* Create a frame record to save our LR and SP (implicit in FP) */
880	stp	x29, x30, [sp, #-16]!
881	mov	x29, sp
882
883	ldr_this_cpu x16, irq_stack_ptr, x17
884	mov	x15, #IRQ_STACK_SIZE
885	add	x16, x16, x15
886
887	/* Move to the new stack and call the function there */
888	mov	sp, x16
889	blr	x1
890
891	/*
892	 * Restore the SP from the FP, and restore the FP and LR from the frame
893	 * record.
894	 */
895	mov	sp, x29
896	ldp	x29, x30, [sp], #16
897#ifdef CONFIG_SHADOW_CALL_STACK
898	ldp	scs_sp, xzr, [sp], #16
899#endif
900	ret
901SYM_FUNC_END(call_on_irq_stack)
902NOKPROBE(call_on_irq_stack)
903
904#ifdef CONFIG_ARM_SDE_INTERFACE
905
906#include <asm/sdei.h>
907#include <uapi/linux/arm_sdei.h>
908
909.macro sdei_handler_exit exit_mode
910	/* On success, this call never returns... */
911	cmp	\exit_mode, #SDEI_EXIT_SMC
912	b.ne	99f
913	smc	#0
914	b	.
91599:	hvc	#0
916	b	.
917.endm
918
919#ifdef CONFIG_UNMAP_KERNEL_AT_EL0
920/*
921 * The regular SDEI entry point may have been unmapped along with the rest of
922 * the kernel. This trampoline restores the kernel mapping to make the x1 memory
923 * argument accessible.
924 *
925 * This clobbers x4, __sdei_handler() will restore this from firmware's
926 * copy.
927 */
928.pushsection ".entry.tramp.text", "ax"
929SYM_CODE_START(__sdei_asm_entry_trampoline)
930	mrs	x4, ttbr1_el1
931	tbz	x4, #USER_ASID_BIT, 1f
932
933	tramp_map_kernel tmp=x4
934	isb
935	mov	x4, xzr
936
937	/*
938	 * Remember whether to unmap the kernel on exit.
939	 */
9401:	str	x4, [x1, #(SDEI_EVENT_INTREGS + S_SDEI_TTBR1)]
941	tramp_data_read_var     x4, __sdei_asm_handler
942	br	x4
943SYM_CODE_END(__sdei_asm_entry_trampoline)
944NOKPROBE(__sdei_asm_entry_trampoline)
945
946/*
947 * Make the exit call and restore the original ttbr1_el1
948 *
949 * x0 & x1: setup for the exit API call
950 * x2: exit_mode
951 * x4: struct sdei_registered_event argument from registration time.
952 */
953SYM_CODE_START(__sdei_asm_exit_trampoline)
954	ldr	x4, [x4, #(SDEI_EVENT_INTREGS + S_SDEI_TTBR1)]
955	cbnz	x4, 1f
956
957	tramp_unmap_kernel	tmp=x4
958
9591:	sdei_handler_exit exit_mode=x2
960SYM_CODE_END(__sdei_asm_exit_trampoline)
961NOKPROBE(__sdei_asm_exit_trampoline)
962.popsection		// .entry.tramp.text
963#endif /* CONFIG_UNMAP_KERNEL_AT_EL0 */
964
965/*
966 * Software Delegated Exception entry point.
967 *
968 * x0: Event number
969 * x1: struct sdei_registered_event argument from registration time.
970 * x2: interrupted PC
971 * x3: interrupted PSTATE
972 * x4: maybe clobbered by the trampoline
973 *
974 * Firmware has preserved x0->x17 for us, we must save/restore the rest to
975 * follow SMC-CC. We save (or retrieve) all the registers as the handler may
976 * want them.
977 */
978SYM_CODE_START(__sdei_asm_handler)
979	stp     x2, x3, [x1, #SDEI_EVENT_INTREGS + S_PC]
980	stp     x4, x5, [x1, #SDEI_EVENT_INTREGS + 16 * 2]
981	stp     x6, x7, [x1, #SDEI_EVENT_INTREGS + 16 * 3]
982	stp     x8, x9, [x1, #SDEI_EVENT_INTREGS + 16 * 4]
983	stp     x10, x11, [x1, #SDEI_EVENT_INTREGS + 16 * 5]
984	stp     x12, x13, [x1, #SDEI_EVENT_INTREGS + 16 * 6]
985	stp     x14, x15, [x1, #SDEI_EVENT_INTREGS + 16 * 7]
986	stp     x16, x17, [x1, #SDEI_EVENT_INTREGS + 16 * 8]
987	stp     x18, x19, [x1, #SDEI_EVENT_INTREGS + 16 * 9]
988	stp     x20, x21, [x1, #SDEI_EVENT_INTREGS + 16 * 10]
989	stp     x22, x23, [x1, #SDEI_EVENT_INTREGS + 16 * 11]
990	stp     x24, x25, [x1, #SDEI_EVENT_INTREGS + 16 * 12]
991	stp     x26, x27, [x1, #SDEI_EVENT_INTREGS + 16 * 13]
992	stp     x28, x29, [x1, #SDEI_EVENT_INTREGS + 16 * 14]
993	mov	x4, sp
994	stp     lr, x4, [x1, #SDEI_EVENT_INTREGS + S_LR]
995
996	mov	x19, x1
997
998#if defined(CONFIG_VMAP_STACK) || defined(CONFIG_SHADOW_CALL_STACK)
999	ldrb	w4, [x19, #SDEI_EVENT_PRIORITY]
1000#endif
1001
1002#ifdef CONFIG_VMAP_STACK
1003	/*
1004	 * entry.S may have been using sp as a scratch register, find whether
1005	 * this is a normal or critical event and switch to the appropriate
1006	 * stack for this CPU.
1007	 */
1008	cbnz	w4, 1f
1009	ldr_this_cpu dst=x5, sym=sdei_stack_normal_ptr, tmp=x6
1010	b	2f
10111:	ldr_this_cpu dst=x5, sym=sdei_stack_critical_ptr, tmp=x6
10122:	mov	x6, #SDEI_STACK_SIZE
1013	add	x5, x5, x6
1014	mov	sp, x5
1015#endif
1016
1017#ifdef CONFIG_SHADOW_CALL_STACK
1018	/* Use a separate shadow call stack for normal and critical events */
1019	cbnz	w4, 3f
1020	ldr_this_cpu dst=scs_sp, sym=sdei_shadow_call_stack_normal_ptr, tmp=x6
1021	b	4f
10223:	ldr_this_cpu dst=scs_sp, sym=sdei_shadow_call_stack_critical_ptr, tmp=x6
10234:
1024#endif
1025
1026	/*
1027	 * We may have interrupted userspace, or a guest, or exit-from or
1028	 * return-to either of these. We can't trust sp_el0, restore it.
1029	 */
1030	mrs	x28, sp_el0
1031	ldr_this_cpu	dst=x0, sym=__entry_task, tmp=x1
1032	msr	sp_el0, x0
1033
1034	/* If we interrupted the kernel point to the previous stack/frame. */
1035	and     x0, x3, #0xc
1036	mrs     x1, CurrentEL
1037	cmp     x0, x1
1038	csel	x29, x29, xzr, eq	// fp, or zero
1039	csel	x4, x2, xzr, eq		// elr, or zero
1040
1041	stp	x29, x4, [sp, #-16]!
1042	mov	x29, sp
1043
1044	add	x0, x19, #SDEI_EVENT_INTREGS
1045	mov	x1, x19
1046	bl	__sdei_handler
1047
1048	msr	sp_el0, x28
1049	/* restore regs >x17 that we clobbered */
1050	mov	x4, x19         // keep x4 for __sdei_asm_exit_trampoline
1051	ldp	x28, x29, [x4, #SDEI_EVENT_INTREGS + 16 * 14]
1052	ldp	x18, x19, [x4, #SDEI_EVENT_INTREGS + 16 * 9]
1053	ldp	lr, x1, [x4, #SDEI_EVENT_INTREGS + S_LR]
1054	mov	sp, x1
1055
1056	mov	x1, x0			// address to complete_and_resume
1057	/* x0 = (x0 <= SDEI_EV_FAILED) ?
1058	 * EVENT_COMPLETE:EVENT_COMPLETE_AND_RESUME
1059	 */
1060	cmp	x0, #SDEI_EV_FAILED
1061	mov_q	x2, SDEI_1_0_FN_SDEI_EVENT_COMPLETE
1062	mov_q	x3, SDEI_1_0_FN_SDEI_EVENT_COMPLETE_AND_RESUME
1063	csel	x0, x2, x3, ls
1064
1065	ldr_l	x2, sdei_exit_mode
1066
1067alternative_if_not ARM64_UNMAP_KERNEL_AT_EL0
1068	sdei_handler_exit exit_mode=x2
1069alternative_else_nop_endif
1070
1071#ifdef CONFIG_UNMAP_KERNEL_AT_EL0
1072	tramp_alias	dst=x5, sym=__sdei_asm_exit_trampoline, tmp=x3
1073	br	x5
1074#endif
1075SYM_CODE_END(__sdei_asm_handler)
1076NOKPROBE(__sdei_asm_handler)
1077#endif /* CONFIG_ARM_SDE_INTERFACE */
1078