xref: /linux/arch/powerpc/kvm/book3s_segment.S (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
1/*
2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
5 *
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9 * GNU General Public License for more details.
10 *
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
14 *
15 * Copyright SUSE Linux Products GmbH 2010
16 *
17 * Authors: Alexander Graf <agraf@suse.de>
18 */
19
20/* Real mode helpers */
21
22#include <asm/asm-compat.h>
23#include <asm/feature-fixups.h>
24
25#if defined(CONFIG_PPC_BOOK3S_64)
26
27#define GET_SHADOW_VCPU(reg)    \
28	mr	reg, r13
29
30#elif defined(CONFIG_PPC_BOOK3S_32)
31
32#define GET_SHADOW_VCPU(reg)    			\
33	tophys(reg, r2);       			\
34	lwz     reg, (THREAD + THREAD_KVM_SVCPU)(reg);	\
35	tophys(reg, reg)
36
37#endif
38
39/* Disable for nested KVM */
40#define USE_QUICK_LAST_INST
41
42
43/* Get helper functions for subarch specific functionality */
44
45#if defined(CONFIG_PPC_BOOK3S_64)
46#include "book3s_64_slb.S"
47#elif defined(CONFIG_PPC_BOOK3S_32)
48#include "book3s_32_sr.S"
49#endif
50
51/******************************************************************************
52 *                                                                            *
53 *                               Entry code                                   *
54 *                                                                            *
55 *****************************************************************************/
56
57.global kvmppc_handler_trampoline_enter
58kvmppc_handler_trampoline_enter:
59
60	/* Required state:
61	 *
62	 * MSR = ~IR|DR
63	 * R1 = host R1
64	 * R2 = host R2
65	 * R4 = guest shadow MSR
66	 * R5 = normal host MSR
67	 * R6 = current host MSR (EE, IR, DR off)
68	 * LR = highmem guest exit code
69	 * all other volatile GPRS = free
70	 * SVCPU[CR] = guest CR
71	 * SVCPU[XER] = guest XER
72	 * SVCPU[CTR] = guest CTR
73	 * SVCPU[LR] = guest LR
74	 */
75
76	/* r3 = shadow vcpu */
77	GET_SHADOW_VCPU(r3)
78
79	/* Save guest exit handler address and MSR */
80	mflr	r0
81	PPC_STL	r0, HSTATE_VMHANDLER(r3)
82	PPC_STL	r5, HSTATE_HOST_MSR(r3)
83
84	/* Save R1/R2 in the PACA (64-bit) or shadow_vcpu (32-bit) */
85	PPC_STL	r1, HSTATE_HOST_R1(r3)
86	PPC_STL	r2, HSTATE_HOST_R2(r3)
87
88	/* Activate guest mode, so faults get handled by KVM */
89	li	r11, KVM_GUEST_MODE_GUEST
90	stb	r11, HSTATE_IN_GUEST(r3)
91
92	/* Switch to guest segment. This is subarch specific. */
93	LOAD_GUEST_SEGMENTS
94
95#ifdef CONFIG_PPC_BOOK3S_64
96BEGIN_FTR_SECTION
97	/* Save host FSCR */
98	mfspr	r8, SPRN_FSCR
99	std	r8, HSTATE_HOST_FSCR(r13)
100	/* Set FSCR during guest execution */
101	ld	r9, SVCPU_SHADOW_FSCR(r13)
102	mtspr	SPRN_FSCR, r9
103END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
104
105	/* Some guests may need to have dcbz set to 32 byte length.
106	 *
107	 * Usually we ensure that by patching the guest's instructions
108	 * to trap on dcbz and emulate it in the hypervisor.
109	 *
110	 * If we can, we should tell the CPU to use 32 byte dcbz though,
111	 * because that's a lot faster.
112	 */
113	lbz	r0, HSTATE_RESTORE_HID5(r3)
114	cmpwi	r0, 0
115	beq	no_dcbz32_on
116
117	mfspr   r0,SPRN_HID5
118	ori     r0, r0, 0x80		/* XXX HID5_dcbz32 = 0x80 */
119	mtspr   SPRN_HID5,r0
120no_dcbz32_on:
121
122#endif /* CONFIG_PPC_BOOK3S_64 */
123
124	/* Enter guest */
125
126	PPC_LL	r8, SVCPU_CTR(r3)
127	PPC_LL	r9, SVCPU_LR(r3)
128	lwz	r10, SVCPU_CR(r3)
129	PPC_LL	r11, SVCPU_XER(r3)
130
131	mtctr	r8
132	mtlr	r9
133	mtcr	r10
134	mtxer	r11
135
136	/* Move SRR0 and SRR1 into the respective regs */
137	PPC_LL  r9, SVCPU_PC(r3)
138	/* First clear RI in our current MSR value */
139	li	r0, MSR_RI
140	andc	r6, r6, r0
141
142	PPC_LL	r0, SVCPU_R0(r3)
143	PPC_LL	r1, SVCPU_R1(r3)
144	PPC_LL	r2, SVCPU_R2(r3)
145	PPC_LL	r5, SVCPU_R5(r3)
146	PPC_LL	r7, SVCPU_R7(r3)
147	PPC_LL	r8, SVCPU_R8(r3)
148	PPC_LL	r10, SVCPU_R10(r3)
149	PPC_LL	r11, SVCPU_R11(r3)
150	PPC_LL	r12, SVCPU_R12(r3)
151	PPC_LL	r13, SVCPU_R13(r3)
152
153	MTMSR_EERI(r6)
154	mtsrr0	r9
155	mtsrr1	r4
156
157	PPC_LL	r4, SVCPU_R4(r3)
158	PPC_LL	r6, SVCPU_R6(r3)
159	PPC_LL	r9, SVCPU_R9(r3)
160	PPC_LL	r3, (SVCPU_R3)(r3)
161
162	RFI_TO_GUEST
163kvmppc_handler_trampoline_enter_end:
164
165
166
167/******************************************************************************
168 *                                                                            *
169 *                               Exit code                                    *
170 *                                                                            *
171 *****************************************************************************/
172
173.global kvmppc_interrupt_pr
174kvmppc_interrupt_pr:
175	/* 64-bit entry. Register usage at this point:
176	 *
177	 * SPRG_SCRATCH0   = guest R13
178	 * R12             = (guest CR << 32) | exit handler id
179	 * R13             = PACA
180	 * HSTATE.SCRATCH0 = guest R12
181	 * HSTATE.SCRATCH1 = guest CTR if RELOCATABLE
182	 */
183#ifdef CONFIG_PPC64
184	/* Match 32-bit entry */
185#ifdef CONFIG_RELOCATABLE
186	std	r9, HSTATE_SCRATCH2(r13)
187	ld	r9, HSTATE_SCRATCH1(r13)
188	mtctr	r9
189	ld	r9, HSTATE_SCRATCH2(r13)
190#endif
191	rotldi	r12, r12, 32		  /* Flip R12 halves for stw */
192	stw	r12, HSTATE_SCRATCH1(r13) /* CR is now in the low half */
193	srdi	r12, r12, 32		  /* shift trap into low half */
194#endif
195
196.global kvmppc_handler_trampoline_exit
197kvmppc_handler_trampoline_exit:
198	/* Register usage at this point:
199	 *
200	 * SPRG_SCRATCH0   = guest R13
201	 * R12             = exit handler id
202	 * R13             = shadow vcpu (32-bit) or PACA (64-bit)
203	 * HSTATE.SCRATCH0 = guest R12
204	 * HSTATE.SCRATCH1 = guest CR
205	 */
206
207	/* Save registers */
208
209	PPC_STL	r0, SVCPU_R0(r13)
210	PPC_STL	r1, SVCPU_R1(r13)
211	PPC_STL	r2, SVCPU_R2(r13)
212	PPC_STL	r3, SVCPU_R3(r13)
213	PPC_STL	r4, SVCPU_R4(r13)
214	PPC_STL	r5, SVCPU_R5(r13)
215	PPC_STL	r6, SVCPU_R6(r13)
216	PPC_STL	r7, SVCPU_R7(r13)
217	PPC_STL	r8, SVCPU_R8(r13)
218	PPC_STL	r9, SVCPU_R9(r13)
219	PPC_STL	r10, SVCPU_R10(r13)
220	PPC_STL	r11, SVCPU_R11(r13)
221
222	/* Restore R1/R2 so we can handle faults */
223	PPC_LL	r1, HSTATE_HOST_R1(r13)
224	PPC_LL	r2, HSTATE_HOST_R2(r13)
225
226	/* Save guest PC and MSR */
227#ifdef CONFIG_PPC64
228BEGIN_FTR_SECTION
229	andi.	r0, r12, 0x2
230	cmpwi	cr1, r0, 0
231	beq	1f
232	mfspr	r3,SPRN_HSRR0
233	mfspr	r4,SPRN_HSRR1
234	andi.	r12,r12,0x3ffd
235	b	2f
236END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
237#endif
2381:	mfsrr0	r3
239	mfsrr1	r4
2402:
241	PPC_STL	r3, SVCPU_PC(r13)
242	PPC_STL	r4, SVCPU_SHADOW_SRR1(r13)
243
244	/* Get scratch'ed off registers */
245	GET_SCRATCH0(r9)
246	PPC_LL	r8, HSTATE_SCRATCH0(r13)
247	lwz	r7, HSTATE_SCRATCH1(r13)
248
249	PPC_STL	r9, SVCPU_R13(r13)
250	PPC_STL	r8, SVCPU_R12(r13)
251	stw	r7, SVCPU_CR(r13)
252
253	/* Save more register state  */
254
255	mfxer	r5
256	mfdar	r6
257	mfdsisr	r7
258	mfctr	r8
259	mflr	r9
260
261	PPC_STL	r5, SVCPU_XER(r13)
262	PPC_STL	r6, SVCPU_FAULT_DAR(r13)
263	stw	r7, SVCPU_FAULT_DSISR(r13)
264	PPC_STL	r8, SVCPU_CTR(r13)
265	PPC_STL	r9, SVCPU_LR(r13)
266
267	/*
268	 * In order for us to easily get the last instruction,
269	 * we got the #vmexit at, we exploit the fact that the
270	 * virtual layout is still the same here, so we can just
271	 * ld from the guest's PC address
272	 */
273
274	/* We only load the last instruction when it's safe */
275	cmpwi	r12, BOOK3S_INTERRUPT_DATA_STORAGE
276	beq	ld_last_inst
277	cmpwi	r12, BOOK3S_INTERRUPT_PROGRAM
278	beq	ld_last_inst
279	cmpwi	r12, BOOK3S_INTERRUPT_SYSCALL
280	beq	ld_last_prev_inst
281	cmpwi	r12, BOOK3S_INTERRUPT_ALIGNMENT
282	beq-	ld_last_inst
283#ifdef CONFIG_PPC64
284BEGIN_FTR_SECTION
285	cmpwi	r12, BOOK3S_INTERRUPT_H_EMUL_ASSIST
286	beq-	ld_last_inst
287END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
288BEGIN_FTR_SECTION
289	cmpwi	r12, BOOK3S_INTERRUPT_FAC_UNAVAIL
290	beq-	ld_last_inst
291END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
292#endif
293
294	b	no_ld_last_inst
295
296ld_last_prev_inst:
297	addi	r3, r3, -4
298
299ld_last_inst:
300	/* Save off the guest instruction we're at */
301
302	/* In case lwz faults */
303	li	r0, KVM_INST_FETCH_FAILED
304
305#ifdef USE_QUICK_LAST_INST
306
307	/* Set guest mode to 'jump over instruction' so if lwz faults
308	 * we'll just continue at the next IP. */
309	li	r9, KVM_GUEST_MODE_SKIP
310	stb	r9, HSTATE_IN_GUEST(r13)
311
312	/*    1) enable paging for data */
313	mfmsr	r9
314	ori	r11, r9, MSR_DR			/* Enable paging for data */
315	mtmsr	r11
316	sync
317	/*    2) fetch the instruction */
318	lwz	r0, 0(r3)
319	/*    3) disable paging again */
320	mtmsr	r9
321	sync
322
323#endif
324	stw	r0, SVCPU_LAST_INST(r13)
325
326no_ld_last_inst:
327
328	/* Unset guest mode */
329	li	r9, KVM_GUEST_MODE_NONE
330	stb	r9, HSTATE_IN_GUEST(r13)
331
332	/* Switch back to host MMU */
333	LOAD_HOST_SEGMENTS
334
335#ifdef CONFIG_PPC_BOOK3S_64
336
337	lbz	r5, HSTATE_RESTORE_HID5(r13)
338	cmpwi	r5, 0
339	beq	no_dcbz32_off
340
341	li	r4, 0
342	mfspr   r5,SPRN_HID5
343	rldimi  r5,r4,6,56
344	mtspr   SPRN_HID5,r5
345
346no_dcbz32_off:
347
348BEGIN_FTR_SECTION
349	/* Save guest FSCR on a FAC_UNAVAIL interrupt */
350	cmpwi	r12, BOOK3S_INTERRUPT_FAC_UNAVAIL
351	bne+	no_fscr_save
352	mfspr	r7, SPRN_FSCR
353	std	r7, SVCPU_SHADOW_FSCR(r13)
354no_fscr_save:
355	/* Restore host FSCR */
356	ld	r8, HSTATE_HOST_FSCR(r13)
357	mtspr	SPRN_FSCR, r8
358END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
359
360#endif /* CONFIG_PPC_BOOK3S_64 */
361
362	/*
363	 * For some interrupts, we need to call the real Linux
364	 * handler, so it can do work for us. This has to happen
365	 * as if the interrupt arrived from the kernel though,
366	 * so let's fake it here where most state is restored.
367	 *
368	 * Having set up SRR0/1 with the address where we want
369	 * to continue with relocation on (potentially in module
370	 * space), we either just go straight there with rfi[d],
371	 * or we jump to an interrupt handler if there is an
372	 * interrupt to be handled first.  In the latter case,
373	 * the rfi[d] at the end of the interrupt handler will
374	 * get us back to where we want to continue.
375	 */
376
377	/* Register usage at this point:
378	 *
379	 * R1       = host R1
380	 * R2       = host R2
381	 * R10      = raw exit handler id
382	 * R12      = exit handler id
383	 * R13      = shadow vcpu (32-bit) or PACA (64-bit)
384	 * SVCPU.*  = guest *
385	 *
386	 */
387
388	PPC_LL	r6, HSTATE_HOST_MSR(r13)
389#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
390	/*
391	 * We don't want to change MSR[TS] bits via rfi here.
392	 * The actual TM handling logic will be in host with
393	 * recovered DR/IR bits after HSTATE_VMHANDLER.
394	 * And MSR_TM can be enabled in HOST_MSR so rfid may
395	 * not suppress this change and can lead to exception.
396	 * Manually set MSR to prevent TS state change here.
397	 */
398	mfmsr   r7
399	rldicl  r7, r7, 64 - MSR_TS_S_LG, 62
400	rldimi  r6, r7, MSR_TS_S_LG, 63 - MSR_TS_T_LG
401#endif
402	PPC_LL	r8, HSTATE_VMHANDLER(r13)
403
404#ifdef CONFIG_PPC64
405BEGIN_FTR_SECTION
406	beq	cr1, 1f
407	mtspr	SPRN_HSRR1, r6
408	mtspr	SPRN_HSRR0, r8
409END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
410#endif
4111:	/* Restore host msr -> SRR1 */
412	mtsrr1	r6
413	/* Load highmem handler address */
414	mtsrr0	r8
415
416	/* RFI into the highmem handler, or jump to interrupt handler */
417	cmpwi	r12, BOOK3S_INTERRUPT_EXTERNAL
418	beqa	BOOK3S_INTERRUPT_EXTERNAL
419	cmpwi	r12, BOOK3S_INTERRUPT_DECREMENTER
420	beqa	BOOK3S_INTERRUPT_DECREMENTER
421	cmpwi	r12, BOOK3S_INTERRUPT_PERFMON
422	beqa	BOOK3S_INTERRUPT_PERFMON
423	cmpwi	r12, BOOK3S_INTERRUPT_DOORBELL
424	beqa	BOOK3S_INTERRUPT_DOORBELL
425
426	RFI_TO_KERNEL
427kvmppc_handler_trampoline_exit_end:
428