xref: /linux/arch/powerpc/kernel/switch.S (revision 306ec721d043bbe5e818d59fbb37c28d999b5d8b)
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2#include <linux/objtool.h>
3#include <asm/asm-offsets.h>
4#include <asm/code-patching-asm.h>
5#include <asm/mmu.h>
6#include <asm/ppc_asm.h>
7#include <asm/kup.h>
8#include <asm/thread_info.h>
9
10.section ".text","ax",@progbits
11
12#ifdef CONFIG_PPC_BOOK3S_64
13/*
14 * Cancel all explict user streams as they will have no use after context
15 * switch and will stop the HW from creating streams itself
16 */
17#define STOP_STREAMS		\
18	DCBT_BOOK3S_STOP_ALL_STREAM_IDS(r6)
19
20#define FLUSH_COUNT_CACHE	\
211:	nop;			\
22	patch_site 1b, patch__call_flush_branch_caches1; \
231:	nop;			\
24	patch_site 1b, patch__call_flush_branch_caches2; \
251:	nop;			\
26	patch_site 1b, patch__call_flush_branch_caches3
27
28.macro nops number
29	.rept \number
30	nop
31	.endr
32.endm
33
34.balign 32
35.global flush_branch_caches
36flush_branch_caches:
37	/* Save LR into r9 */
38	mflr	r9
39
40	// Flush the link stack
41	.rept 64
42	ANNOTATE_INTRA_FUNCTION_CALL
43	bl	.+4
44	.endr
45	b	1f
46	nops	6
47
48	.balign 32
49	/* Restore LR */
501:	mtlr	r9
51
52	// If we're just flushing the link stack, return here
533:	nop
54	patch_site 3b patch__flush_link_stack_return
55
56	li	r9,0x7fff
57	mtctr	r9
58
59	PPC_BCCTR_FLUSH
60
612:	nop
62	patch_site 2b patch__flush_count_cache_return
63
64	nops	3
65
66	.rept 278
67	.balign 32
68	PPC_BCCTR_FLUSH
69	nops	7
70	.endr
71
72	blr
73
74#ifdef CONFIG_PPC_64S_HASH_MMU
75.balign 32
76/*
77 * New stack pointer in r8, old stack pointer in r1, must not clobber r3
78 */
79pin_stack_slb:
80BEGIN_FTR_SECTION
81	clrrdi	r6,r8,28	/* get its ESID */
82	clrrdi	r9,r1,28	/* get current sp ESID */
83FTR_SECTION_ELSE
84	clrrdi	r6,r8,40	/* get its 1T ESID */
85	clrrdi	r9,r1,40	/* get current sp 1T ESID */
86ALT_MMU_FTR_SECTION_END_IFCLR(MMU_FTR_1T_SEGMENT)
87	clrldi.	r0,r6,2		/* is new ESID c00000000? */
88	cmpd	cr1,r6,r9	/* or is new ESID the same as current ESID? */
89	cror	eq,4*cr1+eq,eq
90	beq	2f		/* if yes, don't slbie it */
91
92	/* Bolt in the new stack SLB entry */
93	ld	r7,KSP_VSID(r4)	/* Get new stack's VSID */
94	oris	r0,r6,(SLB_ESID_V)@h
95	ori	r0,r0,(SLB_NUM_BOLTED-1)@l
96BEGIN_FTR_SECTION
97	li	r9,MMU_SEGSIZE_1T	/* insert B field */
98	oris	r6,r6,(MMU_SEGSIZE_1T << SLBIE_SSIZE_SHIFT)@h
99	rldimi	r7,r9,SLB_VSID_SSIZE_SHIFT,0
100END_MMU_FTR_SECTION_IFSET(MMU_FTR_1T_SEGMENT)
101
102	/* Update the last bolted SLB.  No write barriers are needed
103	 * here, provided we only update the current CPU's SLB shadow
104	 * buffer.
105	 */
106	ld	r9,PACA_SLBSHADOWPTR(r13)
107	li	r12,0
108	std	r12,SLBSHADOW_STACKESID(r9)	/* Clear ESID */
109	li	r12,SLBSHADOW_STACKVSID
110	STDX_BE	r7,r12,r9			/* Save VSID */
111	li	r12,SLBSHADOW_STACKESID
112	STDX_BE	r0,r12,r9			/* Save ESID */
113
114	/* No need to check for MMU_FTR_NO_SLBIE_B here, since when
115	 * we have 1TB segments, the only CPUs known to have the errata
116	 * only support less than 1TB of system memory and we'll never
117	 * actually hit this code path.
118	 */
119
120	isync
121	slbie	r6
122BEGIN_FTR_SECTION
123	slbie	r6		/* Workaround POWER5 < DD2.1 issue */
124END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
125	slbmte	r7,r0
126	isync
1272:	blr
128	.size pin_stack_slb,.-pin_stack_slb
129#endif /* CONFIG_PPC_64S_HASH_MMU */
130
131#else
132#define STOP_STREAMS
133#define FLUSH_COUNT_CACHE
134#endif /* CONFIG_PPC_BOOK3S_64 */
135
136/*
137 * do_switch_32/64 have the same calling convention as _switch, i.e., r3,r4
138 * are prev and next thread_struct *, and returns prev task_struct * in r3.
139
140 * This switches the stack, current, and does other task switch housekeeping.
141 */
142.macro do_switch_32
143	tophys(r0,r4)
144	mtspr	SPRN_SPRG_THREAD,r0	/* Update current THREAD phys addr */
145	lwz	r1,KSP(r4)	/* Load new stack pointer */
146
147	/* save the old current 'last' for return value */
148	mr	r3,r2
149	addi	r2,r4,-THREAD	/* Update current */
150.endm
151
152.macro do_switch_64
153	ld	r8,KSP(r4)	/* Load new stack pointer */
154
155	kuap_check_amr r9, r10
156
157	FLUSH_COUNT_CACHE	/* Clobbers r9, ctr */
158
159	STOP_STREAMS		/* Clobbers r6 */
160
161	addi	r3,r3,-THREAD	/* old thread -> task_struct for return value */
162	addi	r6,r4,-THREAD	/* new thread -> task_struct */
163	std	r6,PACACURRENT(r13)	/* Set new task_struct to 'current' */
164#if defined(CONFIG_STACKPROTECTOR)
165	ld	r6, TASK_CANARY(r6)
166	std	r6, PACA_CANARY(r13)
167#endif
168	/* Set new PACAKSAVE */
169	clrrdi	r7,r8,THREAD_SHIFT	/* base of new stack */
170	addi	r7,r7,THREAD_SIZE-SWITCH_FRAME_SIZE
171	std	r7,PACAKSAVE(r13)
172
173#ifdef CONFIG_PPC_64S_HASH_MMU
174BEGIN_MMU_FTR_SECTION
175	bl	pin_stack_slb
176END_MMU_FTR_SECTION_IFCLR(MMU_FTR_TYPE_RADIX)
177#endif
178	/*
179	 * PMU interrupts in radix may come in here. They will use r1, not
180	 * PACAKSAVE, so this stack switch will not cause a problem. They
181	 * will store to the process stack, which may then be migrated to
182	 * another CPU. However the rq lock release on this CPU paired with
183	 * the rq lock acquire on the new CPU before the stack becomes
184	 * active on the new CPU, will order those stores.
185	 */
186	mr	r1,r8		/* start using new stack pointer */
187.endm
188
189/*
190 * This routine switches between two different tasks.  The process
191 * state of one is saved on its kernel stack.  Then the state
192 * of the other is restored from its kernel stack.  The memory
193 * management hardware is updated to the second process's state.
194 * Finally, we can return to the second process.
195 * On entry, r3 points to the THREAD for the current task, r4
196 * points to the THREAD for the new task.
197 *
198 * This routine is always called with interrupts disabled.
199 *
200 * Note: there are two ways to get to the "going out" portion
201 * of this code; either by coming in via the entry (_switch)
202 * or via "fork" which must set up an environment equivalent
203 * to the "_switch" path.  If you change this , you'll have to
204 * change the fork code also.
205 *
206 * The code which creates the new task context is in 'copy_thread'
207 * in arch/ppc/kernel/process.c
208 *
209 * Note: this uses SWITCH_FRAME_SIZE rather than USER_INT_FRAME_SIZE
210 * because we don't need to leave the redzone ABI gap at the top of
211 * the kernel stack.
212 */
213_GLOBAL(_switch)
214	PPC_CREATE_STACK_FRAME(SWITCH_FRAME_SIZE)
215	PPC_STL		r1,KSP(r3)	/* Set old stack pointer */
216	SAVE_NVGPRS(r1)			/* volatiles are caller-saved -- Cort */
217	PPC_STL		r0,_NIP(r1)	/* Return to switch caller */
218	mfcr		r0
219	stw		r0,_CCR(r1)
220
221	/*
222	 * On SMP kernels, care must be taken because a task may be
223	 * scheduled off CPUx and on to CPUy. Memory ordering must be
224	 * considered.
225	 *
226	 * Cacheable stores on CPUx will be visible when the task is
227	 * scheduled on CPUy by virtue of the core scheduler barriers
228	 * (see "Notes on Program-Order guarantees on SMP systems." in
229	 * kernel/sched/core.c).
230	 *
231	 * Uncacheable stores in the case of involuntary preemption must
232	 * be taken care of. The smp_mb__after_spinlock() in __schedule()
233	 * is implemented as hwsync on powerpc, which orders MMIO too. So
234	 * long as there is an hwsync in the context switch path, it will
235	 * be executed on the source CPU after the task has performed
236	 * all MMIO ops on that CPU, and on the destination CPU before the
237	 * task performs any MMIO ops there.
238	 */
239
240	/*
241	 * The kernel context switch path must contain a spin_lock,
242	 * which contains larx/stcx, which will clear any reservation
243	 * of the task being switched.
244	 */
245
246#ifdef CONFIG_PPC32
247	do_switch_32
248#else
249	do_switch_64
250#endif
251
252	lwz	r0,_CCR(r1)
253	mtcrf	0xFF,r0
254	REST_NVGPRS(r1)		/* volatiles are destroyed -- Cort */
255	PPC_LL	r0,_NIP(r1)	/* Return to _switch caller in new task */
256	mtlr	r0
257	addi	r1,r1,SWITCH_FRAME_SIZE
258	blr
259