xref: /openbmc/linux/arch/arm64/kernel/entry.S (revision b755c25f)
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
105	.set	.Lalias\@, TRAMP_VALIAS + \sym - .entry.tramp.text
106	movz	\dst, :abs_g2_s:.Lalias\@
107	movk	\dst, :abs_g1_nc:.Lalias\@
108	movk	\dst, :abs_g0_nc:.Lalias\@
109	.endm
110
111	/*
112	 * This macro corrupts x0-x3. It is the caller's duty  to save/restore
113	 * them if required.
114	 */
115	.macro	apply_ssbd, state, tmp1, tmp2
116alternative_cb	ARM64_ALWAYS_SYSTEM, spectre_v4_patch_fw_mitigation_enable
117	b	.L__asm_ssbd_skip\@		// Patched to NOP
118alternative_cb_end
119	ldr_this_cpu	\tmp2, arm64_ssbd_callback_required, \tmp1
120	cbz	\tmp2,	.L__asm_ssbd_skip\@
121	ldr	\tmp2, [tsk, #TSK_TI_FLAGS]
122	tbnz	\tmp2, #TIF_SSBD, .L__asm_ssbd_skip\@
123	mov	w0, #ARM_SMCCC_ARCH_WORKAROUND_2
124	mov	w1, #\state
125alternative_cb	ARM64_ALWAYS_SYSTEM, smccc_patch_fw_mitigation_conduit
126	nop					// Patched to SMC/HVC #0
127alternative_cb_end
128.L__asm_ssbd_skip\@:
129	.endm
130
131	/* Check for MTE asynchronous tag check faults */
132	.macro check_mte_async_tcf, tmp, ti_flags, thread_sctlr
133#ifdef CONFIG_ARM64_MTE
134	.arch_extension lse
135alternative_if_not ARM64_MTE
136	b	1f
137alternative_else_nop_endif
138	/*
139	 * Asynchronous tag check faults are only possible in ASYNC (2) or
140	 * ASYM (3) modes. In each of these modes bit 1 of SCTLR_EL1.TCF0 is
141	 * set, so skip the check if it is unset.
142	 */
143	tbz	\thread_sctlr, #(SCTLR_EL1_TCF0_SHIFT + 1), 1f
144	mrs_s	\tmp, SYS_TFSRE0_EL1
145	tbz	\tmp, #SYS_TFSR_EL1_TF0_SHIFT, 1f
146	/* Asynchronous TCF occurred for TTBR0 access, set the TI flag */
147	mov	\tmp, #_TIF_MTE_ASYNC_FAULT
148	add	\ti_flags, tsk, #TSK_TI_FLAGS
149	stset	\tmp, [\ti_flags]
1501:
151#endif
152	.endm
153
154	/* Clear the MTE asynchronous tag check faults */
155	.macro clear_mte_async_tcf thread_sctlr
156#ifdef CONFIG_ARM64_MTE
157alternative_if ARM64_MTE
158	/* See comment in check_mte_async_tcf above. */
159	tbz	\thread_sctlr, #(SCTLR_EL1_TCF0_SHIFT + 1), 1f
160	dsb	ish
161	msr_s	SYS_TFSRE0_EL1, xzr
1621:
163alternative_else_nop_endif
164#endif
165	.endm
166
167	.macro mte_set_gcr, mte_ctrl, tmp
168#ifdef CONFIG_ARM64_MTE
169	ubfx	\tmp, \mte_ctrl, #MTE_CTRL_GCR_USER_EXCL_SHIFT, #16
170	orr	\tmp, \tmp, #SYS_GCR_EL1_RRND
171	msr_s	SYS_GCR_EL1, \tmp
172#endif
173	.endm
174
175	.macro mte_set_kernel_gcr, tmp, tmp2
176#ifdef CONFIG_KASAN_HW_TAGS
177alternative_cb	ARM64_ALWAYS_SYSTEM, kasan_hw_tags_enable
178	b	1f
179alternative_cb_end
180	mov	\tmp, KERNEL_GCR_EL1
181	msr_s	SYS_GCR_EL1, \tmp
1821:
183#endif
184	.endm
185
186	.macro mte_set_user_gcr, tsk, tmp, tmp2
187#ifdef CONFIG_KASAN_HW_TAGS
188alternative_cb	ARM64_ALWAYS_SYSTEM, kasan_hw_tags_enable
189	b	1f
190alternative_cb_end
191	ldr	\tmp, [\tsk, #THREAD_MTE_CTRL]
192
193	mte_set_gcr \tmp, \tmp2
1941:
195#endif
196	.endm
197
198	.macro	kernel_entry, el, regsize = 64
199	.if	\el == 0
200	alternative_insn nop, SET_PSTATE_DIT(1), ARM64_HAS_DIT
201	.endif
202	.if	\regsize == 32
203	mov	w0, w0				// zero upper 32 bits of x0
204	.endif
205	stp	x0, x1, [sp, #16 * 0]
206	stp	x2, x3, [sp, #16 * 1]
207	stp	x4, x5, [sp, #16 * 2]
208	stp	x6, x7, [sp, #16 * 3]
209	stp	x8, x9, [sp, #16 * 4]
210	stp	x10, x11, [sp, #16 * 5]
211	stp	x12, x13, [sp, #16 * 6]
212	stp	x14, x15, [sp, #16 * 7]
213	stp	x16, x17, [sp, #16 * 8]
214	stp	x18, x19, [sp, #16 * 9]
215	stp	x20, x21, [sp, #16 * 10]
216	stp	x22, x23, [sp, #16 * 11]
217	stp	x24, x25, [sp, #16 * 12]
218	stp	x26, x27, [sp, #16 * 13]
219	stp	x28, x29, [sp, #16 * 14]
220
221	.if	\el == 0
222	clear_gp_regs
223	mrs	x21, sp_el0
224	ldr_this_cpu	tsk, __entry_task, x20
225	msr	sp_el0, tsk
226
227	/*
228	 * Ensure MDSCR_EL1.SS is clear, since we can unmask debug exceptions
229	 * when scheduling.
230	 */
231	ldr	x19, [tsk, #TSK_TI_FLAGS]
232	disable_step_tsk x19, x20
233
234	/* Check for asynchronous tag check faults in user space */
235	ldr	x0, [tsk, THREAD_SCTLR_USER]
236	check_mte_async_tcf x22, x23, x0
237
238#ifdef CONFIG_ARM64_PTR_AUTH
239alternative_if ARM64_HAS_ADDRESS_AUTH
240	/*
241	 * Enable IA for in-kernel PAC if the task had it disabled. Although
242	 * this could be implemented with an unconditional MRS which would avoid
243	 * a load, this was measured to be slower on Cortex-A75 and Cortex-A76.
244	 *
245	 * Install the kernel IA key only if IA was enabled in the task. If IA
246	 * was disabled on kernel exit then we would have left the kernel IA
247	 * installed so there is no need to install it again.
248	 */
249	tbz	x0, SCTLR_ELx_ENIA_SHIFT, 1f
250	__ptrauth_keys_install_kernel_nosync tsk, x20, x22, x23
251	b	2f
2521:
253	mrs	x0, sctlr_el1
254	orr	x0, x0, SCTLR_ELx_ENIA
255	msr	sctlr_el1, x0
2562:
257alternative_else_nop_endif
258#endif
259
260	apply_ssbd 1, x22, x23
261
262	mte_set_kernel_gcr x22, x23
263
264	/*
265	 * Any non-self-synchronizing system register updates required for
266	 * kernel entry should be placed before this point.
267	 */
268alternative_if ARM64_MTE
269	isb
270	b	1f
271alternative_else_nop_endif
272alternative_if ARM64_HAS_ADDRESS_AUTH
273	isb
274alternative_else_nop_endif
2751:
276
277	scs_load_current
278	.else
279	add	x21, sp, #PT_REGS_SIZE
280	get_current_task tsk
281	.endif /* \el == 0 */
282	mrs	x22, elr_el1
283	mrs	x23, spsr_el1
284	stp	lr, x21, [sp, #S_LR]
285
286	/*
287	 * For exceptions from EL0, create a final frame record.
288	 * For exceptions from EL1, create a synthetic frame record so the
289	 * interrupted code shows up in the backtrace.
290	 */
291	.if \el == 0
292	stp	xzr, xzr, [sp, #S_STACKFRAME]
293	.else
294	stp	x29, x22, [sp, #S_STACKFRAME]
295	.endif
296	add	x29, sp, #S_STACKFRAME
297
298#ifdef CONFIG_ARM64_SW_TTBR0_PAN
299alternative_if_not ARM64_HAS_PAN
300	bl	__swpan_entry_el\el
301alternative_else_nop_endif
302#endif
303
304	stp	x22, x23, [sp, #S_PC]
305
306	/* Not in a syscall by default (el0_svc overwrites for real syscall) */
307	.if	\el == 0
308	mov	w21, #NO_SYSCALL
309	str	w21, [sp, #S_SYSCALLNO]
310	.endif
311
312#ifdef CONFIG_ARM64_PSEUDO_NMI
313alternative_if_not ARM64_HAS_GIC_PRIO_MASKING
314	b	.Lskip_pmr_save\@
315alternative_else_nop_endif
316
317	mrs_s	x20, SYS_ICC_PMR_EL1
318	str	x20, [sp, #S_PMR_SAVE]
319	mov	x20, #GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET
320	msr_s	SYS_ICC_PMR_EL1, x20
321
322.Lskip_pmr_save\@:
323#endif
324
325	/*
326	 * Registers that may be useful after this macro is invoked:
327	 *
328	 * x20 - ICC_PMR_EL1
329	 * x21 - aborted SP
330	 * x22 - aborted PC
331	 * x23 - aborted PSTATE
332	*/
333	.endm
334
335	.macro	kernel_exit, el
336	.if	\el != 0
337	disable_daif
338	.endif
339
340#ifdef CONFIG_ARM64_PSEUDO_NMI
341alternative_if_not ARM64_HAS_GIC_PRIO_MASKING
342	b	.Lskip_pmr_restore\@
343alternative_else_nop_endif
344
345	ldr	x20, [sp, #S_PMR_SAVE]
346	msr_s	SYS_ICC_PMR_EL1, x20
347
348	/* Ensure priority change is seen by redistributor */
349alternative_if_not ARM64_HAS_GIC_PRIO_RELAXED_SYNC
350	dsb	sy
351alternative_else_nop_endif
352
353.Lskip_pmr_restore\@:
354#endif
355
356	ldp	x21, x22, [sp, #S_PC]		// load ELR, SPSR
357
358#ifdef CONFIG_ARM64_SW_TTBR0_PAN
359alternative_if_not ARM64_HAS_PAN
360	bl	__swpan_exit_el\el
361alternative_else_nop_endif
362#endif
363
364	.if	\el == 0
365	ldr	x23, [sp, #S_SP]		// load return stack pointer
366	msr	sp_el0, x23
367	tst	x22, #PSR_MODE32_BIT		// native task?
368	b.eq	3f
369
370#ifdef CONFIG_ARM64_ERRATUM_845719
371alternative_if ARM64_WORKAROUND_845719
372#ifdef CONFIG_PID_IN_CONTEXTIDR
373	mrs	x29, contextidr_el1
374	msr	contextidr_el1, x29
375#else
376	msr contextidr_el1, xzr
377#endif
378alternative_else_nop_endif
379#endif
3803:
381	scs_save tsk
382
383	/* Ignore asynchronous tag check faults in the uaccess routines */
384	ldr	x0, [tsk, THREAD_SCTLR_USER]
385	clear_mte_async_tcf x0
386
387#ifdef CONFIG_ARM64_PTR_AUTH
388alternative_if ARM64_HAS_ADDRESS_AUTH
389	/*
390	 * IA was enabled for in-kernel PAC. Disable it now if needed, or
391	 * alternatively install the user's IA. All other per-task keys and
392	 * SCTLR bits were updated on task switch.
393	 *
394	 * No kernel C function calls after this.
395	 */
396	tbz	x0, SCTLR_ELx_ENIA_SHIFT, 1f
397	__ptrauth_keys_install_user tsk, x0, x1, x2
398	b	2f
3991:
400	mrs	x0, sctlr_el1
401	bic	x0, x0, SCTLR_ELx_ENIA
402	msr	sctlr_el1, x0
4032:
404alternative_else_nop_endif
405#endif
406
407	mte_set_user_gcr tsk, x0, x1
408
409	apply_ssbd 0, x0, x1
410	.endif
411
412	msr	elr_el1, x21			// set up the return data
413	msr	spsr_el1, x22
414	ldp	x0, x1, [sp, #16 * 0]
415	ldp	x2, x3, [sp, #16 * 1]
416	ldp	x4, x5, [sp, #16 * 2]
417	ldp	x6, x7, [sp, #16 * 3]
418	ldp	x8, x9, [sp, #16 * 4]
419	ldp	x10, x11, [sp, #16 * 5]
420	ldp	x12, x13, [sp, #16 * 6]
421	ldp	x14, x15, [sp, #16 * 7]
422	ldp	x16, x17, [sp, #16 * 8]
423	ldp	x18, x19, [sp, #16 * 9]
424	ldp	x20, x21, [sp, #16 * 10]
425	ldp	x22, x23, [sp, #16 * 11]
426	ldp	x24, x25, [sp, #16 * 12]
427	ldp	x26, x27, [sp, #16 * 13]
428	ldp	x28, x29, [sp, #16 * 14]
429
430	.if	\el == 0
431alternative_if_not ARM64_UNMAP_KERNEL_AT_EL0
432	ldr	lr, [sp, #S_LR]
433	add	sp, sp, #PT_REGS_SIZE		// restore sp
434	eret
435alternative_else_nop_endif
436#ifdef CONFIG_UNMAP_KERNEL_AT_EL0
437	msr	far_el1, x29
438
439	ldr_this_cpu	x30, this_cpu_vector, x29
440	tramp_alias	x29, tramp_exit
441	msr		vbar_el1, x30		// install vector table
442	ldr		lr, [sp, #S_LR]		// restore x30
443	add		sp, sp, #PT_REGS_SIZE	// restore sp
444	br		x29
445#endif
446	.else
447	ldr	lr, [sp, #S_LR]
448	add	sp, sp, #PT_REGS_SIZE		// restore sp
449
450	/* Ensure any device/NC reads complete */
451	alternative_insn nop, "dmb sy", ARM64_WORKAROUND_1508412
452
453	eret
454	.endif
455	sb
456	.endm
457
458#ifdef CONFIG_ARM64_SW_TTBR0_PAN
459	/*
460	 * Set the TTBR0 PAN bit in SPSR. When the exception is taken from
461	 * EL0, there is no need to check the state of TTBR0_EL1 since
462	 * accesses are always enabled.
463	 * Note that the meaning of this bit differs from the ARMv8.1 PAN
464	 * feature as all TTBR0_EL1 accesses are disabled, not just those to
465	 * user mappings.
466	 */
467SYM_CODE_START_LOCAL(__swpan_entry_el1)
468	mrs	x21, ttbr0_el1
469	tst	x21, #TTBR_ASID_MASK		// Check for the reserved ASID
470	orr	x23, x23, #PSR_PAN_BIT		// Set the emulated PAN in the saved SPSR
471	b.eq	1f				// TTBR0 access already disabled
472	and	x23, x23, #~PSR_PAN_BIT		// Clear the emulated PAN in the saved SPSR
473SYM_INNER_LABEL(__swpan_entry_el0, SYM_L_LOCAL)
474	__uaccess_ttbr0_disable x21
4751:	ret
476SYM_CODE_END(__swpan_entry_el1)
477
478	/*
479	 * Restore access to TTBR0_EL1. If returning to EL0, no need for SPSR
480	 * PAN bit checking.
481	 */
482SYM_CODE_START_LOCAL(__swpan_exit_el1)
483	tbnz	x22, #22, 1f			// Skip re-enabling TTBR0 access if the PSR_PAN_BIT is set
484	__uaccess_ttbr0_enable x0, x1
4851:	and	x22, x22, #~PSR_PAN_BIT		// ARMv8.0 CPUs do not understand this bit
486	ret
487SYM_CODE_END(__swpan_exit_el1)
488
489SYM_CODE_START_LOCAL(__swpan_exit_el0)
490	__uaccess_ttbr0_enable x0, x1
491	/*
492	 * Enable errata workarounds only if returning to user. The only
493	 * workaround currently required for TTBR0_EL1 changes are for the
494	 * Cavium erratum 27456 (broadcast TLBI instructions may cause I-cache
495	 * corruption).
496	 */
497	b	post_ttbr_update_workaround
498SYM_CODE_END(__swpan_exit_el0)
499#endif
500
501/* GPRs used by entry code */
502tsk	.req	x28		// current thread_info
503
504	.text
505
506/*
507 * Exception vectors.
508 */
509	.pushsection ".entry.text", "ax"
510
511	.align	11
512SYM_CODE_START(vectors)
513	kernel_ventry	1, t, 64, sync		// Synchronous EL1t
514	kernel_ventry	1, t, 64, irq		// IRQ EL1t
515	kernel_ventry	1, t, 64, fiq		// FIQ EL1t
516	kernel_ventry	1, t, 64, error		// Error EL1t
517
518	kernel_ventry	1, h, 64, sync		// Synchronous EL1h
519	kernel_ventry	1, h, 64, irq		// IRQ EL1h
520	kernel_ventry	1, h, 64, fiq		// FIQ EL1h
521	kernel_ventry	1, h, 64, error		// Error EL1h
522
523	kernel_ventry	0, t, 64, sync		// Synchronous 64-bit EL0
524	kernel_ventry	0, t, 64, irq		// IRQ 64-bit EL0
525	kernel_ventry	0, t, 64, fiq		// FIQ 64-bit EL0
526	kernel_ventry	0, t, 64, error		// Error 64-bit EL0
527
528	kernel_ventry	0, t, 32, sync		// Synchronous 32-bit EL0
529	kernel_ventry	0, t, 32, irq		// IRQ 32-bit EL0
530	kernel_ventry	0, t, 32, fiq		// FIQ 32-bit EL0
531	kernel_ventry	0, t, 32, error		// Error 32-bit EL0
532SYM_CODE_END(vectors)
533
534#ifdef CONFIG_VMAP_STACK
535SYM_CODE_START_LOCAL(__bad_stack)
536	/*
537	 * We detected an overflow in kernel_ventry, which switched to the
538	 * overflow stack. Stash the exception regs, and head to our overflow
539	 * handler.
540	 */
541
542	/* Restore the original x0 value */
543	mrs	x0, tpidrro_el0
544
545	/*
546	 * Store the original GPRs to the new stack. The orginal SP (minus
547	 * PT_REGS_SIZE) was stashed in tpidr_el0 by kernel_ventry.
548	 */
549	sub	sp, sp, #PT_REGS_SIZE
550	kernel_entry 1
551	mrs	x0, tpidr_el0
552	add	x0, x0, #PT_REGS_SIZE
553	str	x0, [sp, #S_SP]
554
555	/* Stash the regs for handle_bad_stack */
556	mov	x0, sp
557
558	/* Time to die */
559	bl	handle_bad_stack
560	ASM_BUG()
561SYM_CODE_END(__bad_stack)
562#endif /* CONFIG_VMAP_STACK */
563
564
565	.macro entry_handler el:req, ht:req, regsize:req, label:req
566SYM_CODE_START_LOCAL(el\el\ht\()_\regsize\()_\label)
567	kernel_entry \el, \regsize
568	mov	x0, sp
569	bl	el\el\ht\()_\regsize\()_\label\()_handler
570	.if \el == 0
571	b	ret_to_user
572	.else
573	b	ret_to_kernel
574	.endif
575SYM_CODE_END(el\el\ht\()_\regsize\()_\label)
576	.endm
577
578/*
579 * Early exception handlers
580 */
581	entry_handler	1, t, 64, sync
582	entry_handler	1, t, 64, irq
583	entry_handler	1, t, 64, fiq
584	entry_handler	1, t, 64, error
585
586	entry_handler	1, h, 64, sync
587	entry_handler	1, h, 64, irq
588	entry_handler	1, h, 64, fiq
589	entry_handler	1, h, 64, error
590
591	entry_handler	0, t, 64, sync
592	entry_handler	0, t, 64, irq
593	entry_handler	0, t, 64, fiq
594	entry_handler	0, t, 64, error
595
596	entry_handler	0, t, 32, sync
597	entry_handler	0, t, 32, irq
598	entry_handler	0, t, 32, fiq
599	entry_handler	0, t, 32, error
600
601SYM_CODE_START_LOCAL(ret_to_kernel)
602	kernel_exit 1
603SYM_CODE_END(ret_to_kernel)
604
605SYM_CODE_START_LOCAL(ret_to_user)
606	ldr	x19, [tsk, #TSK_TI_FLAGS]	// re-check for single-step
607	enable_step_tsk x19, x2
608#ifdef CONFIG_GCC_PLUGIN_STACKLEAK
609	bl	stackleak_erase_on_task_stack
610#endif
611	kernel_exit 0
612SYM_CODE_END(ret_to_user)
613
614	.popsection				// .entry.text
615
616	// Move from tramp_pg_dir to swapper_pg_dir
617	.macro tramp_map_kernel, tmp
618	mrs	\tmp, ttbr1_el1
619	add	\tmp, \tmp, #TRAMP_SWAPPER_OFFSET
620	bic	\tmp, \tmp, #USER_ASID_FLAG
621	msr	ttbr1_el1, \tmp
622#ifdef CONFIG_QCOM_FALKOR_ERRATUM_1003
623alternative_if ARM64_WORKAROUND_QCOM_FALKOR_E1003
624	/* ASID already in \tmp[63:48] */
625	movk	\tmp, #:abs_g2_nc:(TRAMP_VALIAS >> 12)
626	movk	\tmp, #:abs_g1_nc:(TRAMP_VALIAS >> 12)
627	/* 2MB boundary containing the vectors, so we nobble the walk cache */
628	movk	\tmp, #:abs_g0_nc:((TRAMP_VALIAS & ~(SZ_2M - 1)) >> 12)
629	isb
630	tlbi	vae1, \tmp
631	dsb	nsh
632alternative_else_nop_endif
633#endif /* CONFIG_QCOM_FALKOR_ERRATUM_1003 */
634	.endm
635
636	// Move from swapper_pg_dir to tramp_pg_dir
637	.macro tramp_unmap_kernel, tmp
638	mrs	\tmp, ttbr1_el1
639	sub	\tmp, \tmp, #TRAMP_SWAPPER_OFFSET
640	orr	\tmp, \tmp, #USER_ASID_FLAG
641	msr	ttbr1_el1, \tmp
642	/*
643	 * We avoid running the post_ttbr_update_workaround here because
644	 * it's only needed by Cavium ThunderX, which requires KPTI to be
645	 * disabled.
646	 */
647	.endm
648
649	.macro		tramp_data_read_var	dst, var
650#ifdef CONFIG_RELOCATABLE
651	ldr		\dst, .L__tramp_data_\var
652	.ifndef		.L__tramp_data_\var
653	.pushsection	".entry.tramp.rodata", "a", %progbits
654	.align		3
655.L__tramp_data_\var:
656	.quad		\var
657	.popsection
658	.endif
659#else
660	/*
661	 * As !RELOCATABLE implies !RANDOMIZE_BASE the address is always a
662	 * compile time constant (and hence not secret and not worth hiding).
663	 *
664	 * As statically allocated kernel code and data always live in the top
665	 * 47 bits of the address space we can sign-extend bit 47 and avoid an
666	 * instruction to load the upper 16 bits (which must be 0xFFFF).
667	 */
668	movz		\dst, :abs_g2_s:\var
669	movk		\dst, :abs_g1_nc:\var
670	movk		\dst, :abs_g0_nc:\var
671#endif
672	.endm
673
674#define BHB_MITIGATION_NONE	0
675#define BHB_MITIGATION_LOOP	1
676#define BHB_MITIGATION_FW	2
677#define BHB_MITIGATION_INSN	3
678
679	.macro tramp_ventry, vector_start, regsize, kpti, bhb
680	.align	7
6811:
682	.if	\regsize == 64
683	msr	tpidrro_el0, x30	// Restored in kernel_ventry
684	.endif
685
686	.if	\bhb == BHB_MITIGATION_LOOP
687	/*
688	 * This sequence must appear before the first indirect branch. i.e. the
689	 * ret out of tramp_ventry. It appears here because x30 is free.
690	 */
691	__mitigate_spectre_bhb_loop	x30
692	.endif // \bhb == BHB_MITIGATION_LOOP
693
694	.if	\bhb == BHB_MITIGATION_INSN
695	clearbhb
696	isb
697	.endif // \bhb == BHB_MITIGATION_INSN
698
699	.if	\kpti == 1
700	/*
701	 * Defend against branch aliasing attacks by pushing a dummy
702	 * entry onto the return stack and using a RET instruction to
703	 * enter the full-fat kernel vectors.
704	 */
705	bl	2f
706	b	.
7072:
708	tramp_map_kernel	x30
709alternative_insn isb, nop, ARM64_WORKAROUND_QCOM_FALKOR_E1003
710	tramp_data_read_var	x30, vectors
711alternative_if_not ARM64_WORKAROUND_CAVIUM_TX2_219_PRFM
712	prfm	plil1strm, [x30, #(1b - \vector_start)]
713alternative_else_nop_endif
714
715	msr	vbar_el1, x30
716	isb
717	.else
718	adr_l	x30, vectors
719	.endif // \kpti == 1
720
721	.if	\bhb == BHB_MITIGATION_FW
722	/*
723	 * The firmware sequence must appear before the first indirect branch.
724	 * i.e. the ret out of tramp_ventry. But it also needs the stack to be
725	 * mapped to save/restore the registers the SMC clobbers.
726	 */
727	__mitigate_spectre_bhb_fw
728	.endif // \bhb == BHB_MITIGATION_FW
729
730	add	x30, x30, #(1b - \vector_start + 4)
731	ret
732.org 1b + 128	// Did we overflow the ventry slot?
733	.endm
734
735	.macro	generate_tramp_vector,	kpti, bhb
736.Lvector_start\@:
737	.space	0x400
738
739	.rept	4
740	tramp_ventry	.Lvector_start\@, 64, \kpti, \bhb
741	.endr
742	.rept	4
743	tramp_ventry	.Lvector_start\@, 32, \kpti, \bhb
744	.endr
745	.endm
746
747#ifdef CONFIG_UNMAP_KERNEL_AT_EL0
748/*
749 * Exception vectors trampoline.
750 * The order must match __bp_harden_el1_vectors and the
751 * arm64_bp_harden_el1_vectors enum.
752 */
753	.pushsection ".entry.tramp.text", "ax"
754	.align	11
755SYM_CODE_START_LOCAL_NOALIGN(tramp_vectors)
756#ifdef CONFIG_MITIGATE_SPECTRE_BRANCH_HISTORY
757	generate_tramp_vector	kpti=1, bhb=BHB_MITIGATION_LOOP
758	generate_tramp_vector	kpti=1, bhb=BHB_MITIGATION_FW
759	generate_tramp_vector	kpti=1, bhb=BHB_MITIGATION_INSN
760#endif /* CONFIG_MITIGATE_SPECTRE_BRANCH_HISTORY */
761	generate_tramp_vector	kpti=1, bhb=BHB_MITIGATION_NONE
762SYM_CODE_END(tramp_vectors)
763
764SYM_CODE_START_LOCAL(tramp_exit)
765	tramp_unmap_kernel	x29
766	mrs		x29, far_el1		// restore x29
767	eret
768	sb
769SYM_CODE_END(tramp_exit)
770	.popsection				// .entry.tramp.text
771#endif /* CONFIG_UNMAP_KERNEL_AT_EL0 */
772
773/*
774 * Exception vectors for spectre mitigations on entry from EL1 when
775 * kpti is not in use.
776 */
777	.macro generate_el1_vector, bhb
778.Lvector_start\@:
779	kernel_ventry	1, t, 64, sync		// Synchronous EL1t
780	kernel_ventry	1, t, 64, irq		// IRQ EL1t
781	kernel_ventry	1, t, 64, fiq		// FIQ EL1h
782	kernel_ventry	1, t, 64, error		// Error EL1t
783
784	kernel_ventry	1, h, 64, sync		// Synchronous EL1h
785	kernel_ventry	1, h, 64, irq		// IRQ EL1h
786	kernel_ventry	1, h, 64, fiq		// FIQ EL1h
787	kernel_ventry	1, h, 64, error		// Error EL1h
788
789	.rept	4
790	tramp_ventry	.Lvector_start\@, 64, 0, \bhb
791	.endr
792	.rept 4
793	tramp_ventry	.Lvector_start\@, 32, 0, \bhb
794	.endr
795	.endm
796
797/* The order must match tramp_vecs and the arm64_bp_harden_el1_vectors enum. */
798	.pushsection ".entry.text", "ax"
799	.align	11
800SYM_CODE_START(__bp_harden_el1_vectors)
801#ifdef CONFIG_MITIGATE_SPECTRE_BRANCH_HISTORY
802	generate_el1_vector	bhb=BHB_MITIGATION_LOOP
803	generate_el1_vector	bhb=BHB_MITIGATION_FW
804	generate_el1_vector	bhb=BHB_MITIGATION_INSN
805#endif /* CONFIG_MITIGATE_SPECTRE_BRANCH_HISTORY */
806SYM_CODE_END(__bp_harden_el1_vectors)
807	.popsection
808
809
810/*
811 * Register switch for AArch64. The callee-saved registers need to be saved
812 * and restored. On entry:
813 *   x0 = previous task_struct (must be preserved across the switch)
814 *   x1 = next task_struct
815 * Previous and next are guaranteed not to be the same.
816 *
817 */
818SYM_FUNC_START(cpu_switch_to)
819	mov	x10, #THREAD_CPU_CONTEXT
820	add	x8, x0, x10
821	mov	x9, sp
822	stp	x19, x20, [x8], #16		// store callee-saved registers
823	stp	x21, x22, [x8], #16
824	stp	x23, x24, [x8], #16
825	stp	x25, x26, [x8], #16
826	stp	x27, x28, [x8], #16
827	stp	x29, x9, [x8], #16
828	str	lr, [x8]
829	add	x8, x1, x10
830	ldp	x19, x20, [x8], #16		// restore callee-saved registers
831	ldp	x21, x22, [x8], #16
832	ldp	x23, x24, [x8], #16
833	ldp	x25, x26, [x8], #16
834	ldp	x27, x28, [x8], #16
835	ldp	x29, x9, [x8], #16
836	ldr	lr, [x8]
837	mov	sp, x9
838	msr	sp_el0, x1
839	ptrauth_keys_install_kernel x1, x8, x9, x10
840	scs_save x0
841	scs_load_current
842	ret
843SYM_FUNC_END(cpu_switch_to)
844NOKPROBE(cpu_switch_to)
845
846/*
847 * This is how we return from a fork.
848 */
849SYM_CODE_START(ret_from_fork)
850	bl	schedule_tail
851	cbz	x19, 1f				// not a kernel thread
852	mov	x0, x20
853	blr	x19
8541:	get_current_task tsk
855	mov	x0, sp
856	bl	asm_exit_to_user_mode
857	b	ret_to_user
858SYM_CODE_END(ret_from_fork)
859NOKPROBE(ret_from_fork)
860
861/*
862 * void call_on_irq_stack(struct pt_regs *regs,
863 * 		          void (*func)(struct pt_regs *));
864 *
865 * Calls func(regs) using this CPU's irq stack and shadow irq stack.
866 */
867SYM_FUNC_START(call_on_irq_stack)
868#ifdef CONFIG_SHADOW_CALL_STACK
869	get_current_task x16
870	scs_save x16
871	ldr_this_cpu scs_sp, irq_shadow_call_stack_ptr, x17
872#endif
873
874	/* Create a frame record to save our LR and SP (implicit in FP) */
875	stp	x29, x30, [sp, #-16]!
876	mov	x29, sp
877
878	ldr_this_cpu x16, irq_stack_ptr, x17
879
880	/* Move to the new stack and call the function there */
881	add	sp, x16, #IRQ_STACK_SIZE
882	blr	x1
883
884	/*
885	 * Restore the SP from the FP, and restore the FP and LR from the frame
886	 * record.
887	 */
888	mov	sp, x29
889	ldp	x29, x30, [sp], #16
890	scs_load_current
891	ret
892SYM_FUNC_END(call_on_irq_stack)
893NOKPROBE(call_on_irq_stack)
894
895#ifdef CONFIG_ARM_SDE_INTERFACE
896
897#include <asm/sdei.h>
898#include <uapi/linux/arm_sdei.h>
899
900.macro sdei_handler_exit exit_mode
901	/* On success, this call never returns... */
902	cmp	\exit_mode, #SDEI_EXIT_SMC
903	b.ne	99f
904	smc	#0
905	b	.
90699:	hvc	#0
907	b	.
908.endm
909
910#ifdef CONFIG_UNMAP_KERNEL_AT_EL0
911/*
912 * The regular SDEI entry point may have been unmapped along with the rest of
913 * the kernel. This trampoline restores the kernel mapping to make the x1 memory
914 * argument accessible.
915 *
916 * This clobbers x4, __sdei_handler() will restore this from firmware's
917 * copy.
918 */
919.pushsection ".entry.tramp.text", "ax"
920SYM_CODE_START(__sdei_asm_entry_trampoline)
921	mrs	x4, ttbr1_el1
922	tbz	x4, #USER_ASID_BIT, 1f
923
924	tramp_map_kernel tmp=x4
925	isb
926	mov	x4, xzr
927
928	/*
929	 * Remember whether to unmap the kernel on exit.
930	 */
9311:	str	x4, [x1, #(SDEI_EVENT_INTREGS + S_SDEI_TTBR1)]
932	tramp_data_read_var     x4, __sdei_asm_handler
933	br	x4
934SYM_CODE_END(__sdei_asm_entry_trampoline)
935NOKPROBE(__sdei_asm_entry_trampoline)
936
937/*
938 * Make the exit call and restore the original ttbr1_el1
939 *
940 * x0 & x1: setup for the exit API call
941 * x2: exit_mode
942 * x4: struct sdei_registered_event argument from registration time.
943 */
944SYM_CODE_START(__sdei_asm_exit_trampoline)
945	ldr	x4, [x4, #(SDEI_EVENT_INTREGS + S_SDEI_TTBR1)]
946	cbnz	x4, 1f
947
948	tramp_unmap_kernel	tmp=x4
949
9501:	sdei_handler_exit exit_mode=x2
951SYM_CODE_END(__sdei_asm_exit_trampoline)
952NOKPROBE(__sdei_asm_exit_trampoline)
953.popsection		// .entry.tramp.text
954#endif /* CONFIG_UNMAP_KERNEL_AT_EL0 */
955
956/*
957 * Software Delegated Exception entry point.
958 *
959 * x0: Event number
960 * x1: struct sdei_registered_event argument from registration time.
961 * x2: interrupted PC
962 * x3: interrupted PSTATE
963 * x4: maybe clobbered by the trampoline
964 *
965 * Firmware has preserved x0->x17 for us, we must save/restore the rest to
966 * follow SMC-CC. We save (or retrieve) all the registers as the handler may
967 * want them.
968 */
969SYM_CODE_START(__sdei_asm_handler)
970	stp     x2, x3, [x1, #SDEI_EVENT_INTREGS + S_PC]
971	stp     x4, x5, [x1, #SDEI_EVENT_INTREGS + 16 * 2]
972	stp     x6, x7, [x1, #SDEI_EVENT_INTREGS + 16 * 3]
973	stp     x8, x9, [x1, #SDEI_EVENT_INTREGS + 16 * 4]
974	stp     x10, x11, [x1, #SDEI_EVENT_INTREGS + 16 * 5]
975	stp     x12, x13, [x1, #SDEI_EVENT_INTREGS + 16 * 6]
976	stp     x14, x15, [x1, #SDEI_EVENT_INTREGS + 16 * 7]
977	stp     x16, x17, [x1, #SDEI_EVENT_INTREGS + 16 * 8]
978	stp     x18, x19, [x1, #SDEI_EVENT_INTREGS + 16 * 9]
979	stp     x20, x21, [x1, #SDEI_EVENT_INTREGS + 16 * 10]
980	stp     x22, x23, [x1, #SDEI_EVENT_INTREGS + 16 * 11]
981	stp     x24, x25, [x1, #SDEI_EVENT_INTREGS + 16 * 12]
982	stp     x26, x27, [x1, #SDEI_EVENT_INTREGS + 16 * 13]
983	stp     x28, x29, [x1, #SDEI_EVENT_INTREGS + 16 * 14]
984	mov	x4, sp
985	stp     lr, x4, [x1, #SDEI_EVENT_INTREGS + S_LR]
986
987	mov	x19, x1
988
989#if defined(CONFIG_VMAP_STACK) || defined(CONFIG_SHADOW_CALL_STACK)
990	ldrb	w4, [x19, #SDEI_EVENT_PRIORITY]
991#endif
992
993#ifdef CONFIG_VMAP_STACK
994	/*
995	 * entry.S may have been using sp as a scratch register, find whether
996	 * this is a normal or critical event and switch to the appropriate
997	 * stack for this CPU.
998	 */
999	cbnz	w4, 1f
1000	ldr_this_cpu dst=x5, sym=sdei_stack_normal_ptr, tmp=x6
1001	b	2f
10021:	ldr_this_cpu dst=x5, sym=sdei_stack_critical_ptr, tmp=x6
10032:	mov	x6, #SDEI_STACK_SIZE
1004	add	x5, x5, x6
1005	mov	sp, x5
1006#endif
1007
1008#ifdef CONFIG_SHADOW_CALL_STACK
1009	/* Use a separate shadow call stack for normal and critical events */
1010	cbnz	w4, 3f
1011	ldr_this_cpu dst=scs_sp, sym=sdei_shadow_call_stack_normal_ptr, tmp=x6
1012	b	4f
10133:	ldr_this_cpu dst=scs_sp, sym=sdei_shadow_call_stack_critical_ptr, tmp=x6
10144:
1015#endif
1016
1017	/*
1018	 * We may have interrupted userspace, or a guest, or exit-from or
1019	 * return-to either of these. We can't trust sp_el0, restore it.
1020	 */
1021	mrs	x28, sp_el0
1022	ldr_this_cpu	dst=x0, sym=__entry_task, tmp=x1
1023	msr	sp_el0, x0
1024
1025	/* If we interrupted the kernel point to the previous stack/frame. */
1026	and     x0, x3, #0xc
1027	mrs     x1, CurrentEL
1028	cmp     x0, x1
1029	csel	x29, x29, xzr, eq	// fp, or zero
1030	csel	x4, x2, xzr, eq		// elr, or zero
1031
1032	stp	x29, x4, [sp, #-16]!
1033	mov	x29, sp
1034
1035	add	x0, x19, #SDEI_EVENT_INTREGS
1036	mov	x1, x19
1037	bl	__sdei_handler
1038
1039	msr	sp_el0, x28
1040	/* restore regs >x17 that we clobbered */
1041	mov	x4, x19         // keep x4 for __sdei_asm_exit_trampoline
1042	ldp	x28, x29, [x4, #SDEI_EVENT_INTREGS + 16 * 14]
1043	ldp	x18, x19, [x4, #SDEI_EVENT_INTREGS + 16 * 9]
1044	ldp	lr, x1, [x4, #SDEI_EVENT_INTREGS + S_LR]
1045	mov	sp, x1
1046
1047	mov	x1, x0			// address to complete_and_resume
1048	/* x0 = (x0 <= SDEI_EV_FAILED) ?
1049	 * EVENT_COMPLETE:EVENT_COMPLETE_AND_RESUME
1050	 */
1051	cmp	x0, #SDEI_EV_FAILED
1052	mov_q	x2, SDEI_1_0_FN_SDEI_EVENT_COMPLETE
1053	mov_q	x3, SDEI_1_0_FN_SDEI_EVENT_COMPLETE_AND_RESUME
1054	csel	x0, x2, x3, ls
1055
1056	ldr_l	x2, sdei_exit_mode
1057
1058alternative_if_not ARM64_UNMAP_KERNEL_AT_EL0
1059	sdei_handler_exit exit_mode=x2
1060alternative_else_nop_endif
1061
1062#ifdef CONFIG_UNMAP_KERNEL_AT_EL0
1063	tramp_alias	dst=x5, sym=__sdei_asm_exit_trampoline
1064	br	x5
1065#endif
1066SYM_CODE_END(__sdei_asm_handler)
1067NOKPROBE(__sdei_asm_handler)
1068#endif /* CONFIG_ARM_SDE_INTERFACE */
1069