Commit | Line | Data |
---|---|---|
d2912cb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
de56a948 PM |
2 | /* |
3 | * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com> | |
4 | * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved. | |
5 | * | |
6 | * Authors: | |
7 | * Paul Mackerras <paulus@au1.ibm.com> | |
8 | * Alexander Graf <agraf@suse.de> | |
9 | * Kevin Wolf <mail@kevin-wolf.de> | |
10 | * | |
11 | * Description: KVM functions specific to running on Book 3S | |
12 | * processors in hypervisor mode (specifically POWER7 and later). | |
13 | * | |
14 | * This file is derived from arch/powerpc/kvm/book3s.c, | |
15 | * by Alexander Graf <agraf@suse.de>. | |
de56a948 PM |
16 | */ |
17 | ||
18 | #include <linux/kvm_host.h> | |
4bb817ed | 19 | #include <linux/kernel.h> |
de56a948 PM |
20 | #include <linux/err.h> |
21 | #include <linux/slab.h> | |
22 | #include <linux/preempt.h> | |
174cd4b1 | 23 | #include <linux/sched/signal.h> |
03441a34 | 24 | #include <linux/sched/stat.h> |
de56a948 | 25 | #include <linux/delay.h> |
66b15db6 | 26 | #include <linux/export.h> |
de56a948 PM |
27 | #include <linux/fs.h> |
28 | #include <linux/anon_inodes.h> | |
07f8ab25 | 29 | #include <linux/cpu.h> |
de56a948 | 30 | #include <linux/cpumask.h> |
aa04b4cc PM |
31 | #include <linux/spinlock.h> |
32 | #include <linux/page-flags.h> | |
2c9097e4 | 33 | #include <linux/srcu.h> |
398a76c6 | 34 | #include <linux/miscdevice.h> |
e23a808b | 35 | #include <linux/debugfs.h> |
d3989143 BH |
36 | #include <linux/gfp.h> |
37 | #include <linux/vmalloc.h> | |
38 | #include <linux/highmem.h> | |
39 | #include <linux/hugetlb.h> | |
40 | #include <linux/kvm_irqfd.h> | |
41 | #include <linux/irqbypass.h> | |
42 | #include <linux/module.h> | |
43 | #include <linux/compiler.h> | |
44 | #include <linux/of.h> | |
de56a948 | 45 | |
d6ee76d3 | 46 | #include <asm/ftrace.h> |
de56a948 | 47 | #include <asm/reg.h> |
57900694 | 48 | #include <asm/ppc-opcode.h> |
6de6638b | 49 | #include <asm/asm-prototypes.h> |
4bad7779 | 50 | #include <asm/archrandom.h> |
e303c087 | 51 | #include <asm/debug.h> |
57900694 | 52 | #include <asm/disassemble.h> |
de56a948 PM |
53 | #include <asm/cputable.h> |
54 | #include <asm/cacheflush.h> | |
7c0f6ba6 | 55 | #include <linux/uaccess.h> |
de56a948 PM |
56 | #include <asm/io.h> |
57 | #include <asm/kvm_ppc.h> | |
58 | #include <asm/kvm_book3s.h> | |
59 | #include <asm/mmu_context.h> | |
60 | #include <asm/lppaca.h> | |
61 | #include <asm/processor.h> | |
371fefd6 | 62 | #include <asm/cputhreads.h> |
aa04b4cc | 63 | #include <asm/page.h> |
de1d9248 | 64 | #include <asm/hvcall.h> |
ae3a197e | 65 | #include <asm/switch_to.h> |
512691d4 | 66 | #include <asm/smp.h> |
66feed61 | 67 | #include <asm/dbell.h> |
fd7bacbc | 68 | #include <asm/hmi.h> |
c57875f5 | 69 | #include <asm/pnv-pci.h> |
7a84084c | 70 | #include <asm/mmu.h> |
f725758b PM |
71 | #include <asm/opal.h> |
72 | #include <asm/xics.h> | |
5af50993 | 73 | #include <asm/xive.h> |
c1fe190c | 74 | #include <asm/hw_breakpoint.h> |
ca9f4942 | 75 | #include <asm/kvm_book3s_uvmem.h> |
c3262257 | 76 | #include <asm/ultravisor.h> |
de56a948 | 77 | |
3a167bea AK |
78 | #include "book3s.h" |
79 | ||
3c78f78a SW |
80 | #define CREATE_TRACE_POINTS |
81 | #include "trace_hv.h" | |
82 | ||
de56a948 PM |
83 | /* #define EXIT_DEBUG */ |
84 | /* #define EXIT_DEBUG_SIMPLE */ | |
85 | /* #define EXIT_DEBUG_INT */ | |
86 | ||
913d3ff9 PM |
87 | /* Used to indicate that a guest page fault needs to be handled */ |
88 | #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1) | |
f7af5209 SW |
89 | /* Used to indicate that a guest passthrough interrupt needs to be handled */ |
90 | #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2) | |
913d3ff9 | 91 | |
c7b67670 PM |
92 | /* Used as a "null" value for timebase values */ |
93 | #define TB_NIL (~(u64)0) | |
94 | ||
699a0ea0 PM |
95 | static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1); |
96 | ||
b4deba5c | 97 | static int dynamic_mt_modes = 6; |
57ad583f | 98 | module_param(dynamic_mt_modes, int, 0644); |
b4deba5c | 99 | MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)"); |
ec257165 | 100 | static int target_smt_mode; |
57ad583f | 101 | module_param(target_smt_mode, int, 0644); |
ec257165 | 102 | MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)"); |
9678cdaa | 103 | |
516f7898 PM |
104 | static bool indep_threads_mode = true; |
105 | module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR); | |
106 | MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)"); | |
107 | ||
aa227864 PM |
108 | static bool one_vm_per_core; |
109 | module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR); | |
110 | MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)"); | |
111 | ||
520fe9c6 SW |
112 | #ifdef CONFIG_KVM_XICS |
113 | static struct kernel_param_ops module_param_ops = { | |
114 | .set = param_set_int, | |
115 | .get = param_get_int, | |
116 | }; | |
117 | ||
57ad583f | 118 | module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644); |
644abbb2 SW |
119 | MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization"); |
120 | ||
57ad583f | 121 | module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644); |
520fe9c6 SW |
122 | MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core"); |
123 | #endif | |
124 | ||
aa069a99 PM |
125 | /* If set, guests are allowed to create and control nested guests */ |
126 | static bool nested = true; | |
127 | module_param(nested, bool, S_IRUGO | S_IWUSR); | |
128 | MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)"); | |
129 | ||
130 | static inline bool nesting_enabled(struct kvm *kvm) | |
131 | { | |
132 | return kvm->arch.nested_enable && kvm_is_radix(kvm); | |
133 | } | |
134 | ||
00608e1f PM |
135 | /* If set, the threads on each CPU core have to be in the same MMU mode */ |
136 | static bool no_mixing_hpt_and_radix; | |
137 | ||
32fad281 | 138 | static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu); |
19ccb76a | 139 | |
7aa15842 PM |
140 | /* |
141 | * RWMR values for POWER8. These control the rate at which PURR | |
142 | * and SPURR count and should be set according to the number of | |
143 | * online threads in the vcore being run. | |
144 | */ | |
0abb75b7 NMG |
145 | #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL |
146 | #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL | |
147 | #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL | |
148 | #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL | |
149 | #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL | |
150 | #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL | |
151 | #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL | |
152 | #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL | |
7aa15842 PM |
153 | |
154 | static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = { | |
155 | RWMR_RPA_P8_1THREAD, | |
156 | RWMR_RPA_P8_1THREAD, | |
157 | RWMR_RPA_P8_2THREAD, | |
158 | RWMR_RPA_P8_3THREAD, | |
159 | RWMR_RPA_P8_4THREAD, | |
160 | RWMR_RPA_P8_5THREAD, | |
161 | RWMR_RPA_P8_6THREAD, | |
162 | RWMR_RPA_P8_7THREAD, | |
163 | RWMR_RPA_P8_8THREAD, | |
164 | }; | |
165 | ||
7b5f8272 SJS |
166 | static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc, |
167 | int *ip) | |
168 | { | |
169 | int i = *ip; | |
170 | struct kvm_vcpu *vcpu; | |
171 | ||
172 | while (++i < MAX_SMT_THREADS) { | |
173 | vcpu = READ_ONCE(vc->runnable_threads[i]); | |
174 | if (vcpu) { | |
175 | *ip = i; | |
176 | return vcpu; | |
177 | } | |
178 | } | |
179 | return NULL; | |
180 | } | |
181 | ||
182 | /* Used to traverse the list of runnable threads for a given vcore */ | |
183 | #define for_each_runnable_thread(i, vcpu, vc) \ | |
184 | for (i = -1; (vcpu = next_runnable_thread(vc, &i)); ) | |
185 | ||
66feed61 PM |
186 | static bool kvmppc_ipi_thread(int cpu) |
187 | { | |
1704a81c PM |
188 | unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER); |
189 | ||
f3c18e93 PM |
190 | /* If we're a nested hypervisor, fall back to ordinary IPIs for now */ |
191 | if (kvmhv_on_pseries()) | |
192 | return false; | |
193 | ||
1704a81c PM |
194 | /* On POWER9 we can use msgsnd to IPI any cpu */ |
195 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { | |
196 | msg |= get_hard_smp_processor_id(cpu); | |
197 | smp_mb(); | |
198 | __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); | |
199 | return true; | |
200 | } | |
201 | ||
66feed61 PM |
202 | /* On POWER8 for IPIs to threads in the same core, use msgsnd */ |
203 | if (cpu_has_feature(CPU_FTR_ARCH_207S)) { | |
204 | preempt_disable(); | |
205 | if (cpu_first_thread_sibling(cpu) == | |
206 | cpu_first_thread_sibling(smp_processor_id())) { | |
66feed61 PM |
207 | msg |= cpu_thread_in_core(cpu); |
208 | smp_mb(); | |
209 | __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); | |
210 | preempt_enable(); | |
211 | return true; | |
212 | } | |
213 | preempt_enable(); | |
214 | } | |
215 | ||
216 | #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP) | |
f725758b | 217 | if (cpu >= 0 && cpu < nr_cpu_ids) { |
d2e60075 | 218 | if (paca_ptrs[cpu]->kvm_hstate.xics_phys) { |
f725758b PM |
219 | xics_wake_cpu(cpu); |
220 | return true; | |
221 | } | |
222 | opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY); | |
66feed61 PM |
223 | return true; |
224 | } | |
225 | #endif | |
226 | ||
227 | return false; | |
228 | } | |
229 | ||
3a167bea | 230 | static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu) |
54695c30 | 231 | { |
ec257165 | 232 | int cpu; |
da4ad88c | 233 | struct rcuwait *waitp; |
54695c30 | 234 | |
da4ad88c DB |
235 | waitp = kvm_arch_vcpu_get_wait(vcpu); |
236 | if (rcuwait_wake_up(waitp)) | |
54695c30 | 237 | ++vcpu->stat.halt_wakeup; |
54695c30 | 238 | |
3deda5e5 PM |
239 | cpu = READ_ONCE(vcpu->arch.thread_cpu); |
240 | if (cpu >= 0 && kvmppc_ipi_thread(cpu)) | |
66feed61 | 241 | return; |
54695c30 BH |
242 | |
243 | /* CPU points to the first thread of the core */ | |
ec257165 | 244 | cpu = vcpu->cpu; |
66feed61 PM |
245 | if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu)) |
246 | smp_send_reschedule(cpu); | |
54695c30 BH |
247 | } |
248 | ||
c7b67670 PM |
249 | /* |
250 | * We use the vcpu_load/put functions to measure stolen time. | |
251 | * Stolen time is counted as time when either the vcpu is able to | |
252 | * run as part of a virtual core, but the task running the vcore | |
253 | * is preempted or sleeping, or when the vcpu needs something done | |
254 | * in the kernel by the task running the vcpu, but that task is | |
255 | * preempted or sleeping. Those two things have to be counted | |
256 | * separately, since one of the vcpu tasks will take on the job | |
257 | * of running the core, and the other vcpu tasks in the vcore will | |
258 | * sleep waiting for it to do that, but that sleep shouldn't count | |
259 | * as stolen time. | |
260 | * | |
261 | * Hence we accumulate stolen time when the vcpu can run as part of | |
262 | * a vcore using vc->stolen_tb, and the stolen time when the vcpu | |
263 | * needs its task to do other things in the kernel (for example, | |
264 | * service a page fault) in busy_stolen. We don't accumulate | |
265 | * stolen time for a vcore when it is inactive, or for a vcpu | |
266 | * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of | |
267 | * a misnomer; it means that the vcpu task is not executing in | |
268 | * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in | |
269 | * the kernel. We don't have any way of dividing up that time | |
270 | * between time that the vcpu is genuinely stopped, time that | |
271 | * the task is actively working on behalf of the vcpu, and time | |
272 | * that the task is preempted, so we don't count any of it as | |
273 | * stolen. | |
274 | * | |
275 | * Updates to busy_stolen are protected by arch.tbacct_lock; | |
2711e248 PM |
276 | * updates to vc->stolen_tb are protected by the vcore->stoltb_lock |
277 | * lock. The stolen times are measured in units of timebase ticks. | |
278 | * (Note that the != TB_NIL checks below are purely defensive; | |
279 | * they should never fail.) | |
c7b67670 PM |
280 | */ |
281 | ||
ec257165 PM |
282 | static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc) |
283 | { | |
284 | unsigned long flags; | |
285 | ||
286 | spin_lock_irqsave(&vc->stoltb_lock, flags); | |
287 | vc->preempt_tb = mftb(); | |
288 | spin_unlock_irqrestore(&vc->stoltb_lock, flags); | |
289 | } | |
290 | ||
291 | static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc) | |
292 | { | |
293 | unsigned long flags; | |
294 | ||
295 | spin_lock_irqsave(&vc->stoltb_lock, flags); | |
296 | if (vc->preempt_tb != TB_NIL) { | |
297 | vc->stolen_tb += mftb() - vc->preempt_tb; | |
298 | vc->preempt_tb = TB_NIL; | |
299 | } | |
300 | spin_unlock_irqrestore(&vc->stoltb_lock, flags); | |
301 | } | |
302 | ||
3a167bea | 303 | static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu) |
de56a948 | 304 | { |
0456ec4f | 305 | struct kvmppc_vcore *vc = vcpu->arch.vcore; |
bf3d32e1 | 306 | unsigned long flags; |
0456ec4f | 307 | |
2711e248 PM |
308 | /* |
309 | * We can test vc->runner without taking the vcore lock, | |
310 | * because only this task ever sets vc->runner to this | |
311 | * vcpu, and once it is set to this vcpu, only this task | |
312 | * ever sets it to NULL. | |
313 | */ | |
ec257165 PM |
314 | if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING) |
315 | kvmppc_core_end_stolen(vc); | |
316 | ||
2711e248 | 317 | spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); |
c7b67670 PM |
318 | if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST && |
319 | vcpu->arch.busy_preempt != TB_NIL) { | |
320 | vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt; | |
321 | vcpu->arch.busy_preempt = TB_NIL; | |
322 | } | |
bf3d32e1 | 323 | spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); |
de56a948 PM |
324 | } |
325 | ||
3a167bea | 326 | static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu) |
de56a948 | 327 | { |
0456ec4f | 328 | struct kvmppc_vcore *vc = vcpu->arch.vcore; |
bf3d32e1 | 329 | unsigned long flags; |
0456ec4f | 330 | |
ec257165 PM |
331 | if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING) |
332 | kvmppc_core_start_stolen(vc); | |
333 | ||
2711e248 | 334 | spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); |
c7b67670 PM |
335 | if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST) |
336 | vcpu->arch.busy_preempt = mftb(); | |
bf3d32e1 | 337 | spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); |
de56a948 PM |
338 | } |
339 | ||
5358a963 | 340 | static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr) |
de56a948 PM |
341 | { |
342 | vcpu->arch.pvr = pvr; | |
343 | } | |
344 | ||
5358a963 | 345 | static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat) |
388cc6e1 | 346 | { |
2ee13be3 | 347 | unsigned long host_pcr_bit = 0, guest_pcr_bit = 0; |
388cc6e1 PM |
348 | struct kvmppc_vcore *vc = vcpu->arch.vcore; |
349 | ||
2ee13be3 SJS |
350 | /* We can (emulate) our own architecture version and anything older */ |
351 | if (cpu_has_feature(CPU_FTR_ARCH_300)) | |
352 | host_pcr_bit = PCR_ARCH_300; | |
353 | else if (cpu_has_feature(CPU_FTR_ARCH_207S)) | |
354 | host_pcr_bit = PCR_ARCH_207; | |
355 | else if (cpu_has_feature(CPU_FTR_ARCH_206)) | |
356 | host_pcr_bit = PCR_ARCH_206; | |
357 | else | |
358 | host_pcr_bit = PCR_ARCH_205; | |
359 | ||
360 | /* Determine lowest PCR bit needed to run guest in given PVR level */ | |
361 | guest_pcr_bit = host_pcr_bit; | |
388cc6e1 | 362 | if (arch_compat) { |
388cc6e1 PM |
363 | switch (arch_compat) { |
364 | case PVR_ARCH_205: | |
2ee13be3 | 365 | guest_pcr_bit = PCR_ARCH_205; |
388cc6e1 PM |
366 | break; |
367 | case PVR_ARCH_206: | |
368 | case PVR_ARCH_206p: | |
2ee13be3 | 369 | guest_pcr_bit = PCR_ARCH_206; |
5557ae0e PM |
370 | break; |
371 | case PVR_ARCH_207: | |
2ee13be3 SJS |
372 | guest_pcr_bit = PCR_ARCH_207; |
373 | break; | |
374 | case PVR_ARCH_300: | |
375 | guest_pcr_bit = PCR_ARCH_300; | |
388cc6e1 PM |
376 | break; |
377 | default: | |
378 | return -EINVAL; | |
379 | } | |
380 | } | |
381 | ||
2ee13be3 SJS |
382 | /* Check requested PCR bits don't exceed our capabilities */ |
383 | if (guest_pcr_bit > host_pcr_bit) | |
384 | return -EINVAL; | |
385 | ||
388cc6e1 PM |
386 | spin_lock(&vc->lock); |
387 | vc->arch_compat = arch_compat; | |
13c7bb3c JN |
388 | /* |
389 | * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit | |
390 | * Also set all reserved PCR bits | |
391 | */ | |
392 | vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK; | |
388cc6e1 PM |
393 | spin_unlock(&vc->lock); |
394 | ||
395 | return 0; | |
396 | } | |
397 | ||
5358a963 | 398 | static void kvmppc_dump_regs(struct kvm_vcpu *vcpu) |
de56a948 PM |
399 | { |
400 | int r; | |
401 | ||
402 | pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id); | |
403 | pr_err("pc = %.16lx msr = %.16llx trap = %x\n", | |
173c520a | 404 | vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap); |
de56a948 PM |
405 | for (r = 0; r < 16; ++r) |
406 | pr_err("r%2d = %.16lx r%d = %.16lx\n", | |
407 | r, kvmppc_get_gpr(vcpu, r), | |
408 | r+16, kvmppc_get_gpr(vcpu, r+16)); | |
409 | pr_err("ctr = %.16lx lr = %.16lx\n", | |
173c520a | 410 | vcpu->arch.regs.ctr, vcpu->arch.regs.link); |
de56a948 PM |
411 | pr_err("srr0 = %.16llx srr1 = %.16llx\n", |
412 | vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1); | |
413 | pr_err("sprg0 = %.16llx sprg1 = %.16llx\n", | |
414 | vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1); | |
415 | pr_err("sprg2 = %.16llx sprg3 = %.16llx\n", | |
416 | vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3); | |
fd0944ba PM |
417 | pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n", |
418 | vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr); | |
de56a948 PM |
419 | pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar); |
420 | pr_err("fault dar = %.16lx dsisr = %.8x\n", | |
421 | vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); | |
422 | pr_err("SLB (%d entries):\n", vcpu->arch.slb_max); | |
423 | for (r = 0; r < vcpu->arch.slb_max; ++r) | |
424 | pr_err(" ESID = %.16llx VSID = %.16llx\n", | |
425 | vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv); | |
426 | pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n", | |
a0144e2a | 427 | vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1, |
de56a948 PM |
428 | vcpu->arch.last_inst); |
429 | } | |
430 | ||
5358a963 | 431 | static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id) |
a8606e20 | 432 | { |
5a3f4936 | 433 | return kvm_get_vcpu_by_id(kvm, id); |
a8606e20 PM |
434 | } |
435 | ||
436 | static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa) | |
437 | { | |
f13c13a0 | 438 | vpa->__old_status |= LPPACA_OLD_SHARED_PROC; |
02407552 | 439 | vpa->yield_count = cpu_to_be32(1); |
a8606e20 PM |
440 | } |
441 | ||
55b665b0 PM |
442 | static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v, |
443 | unsigned long addr, unsigned long len) | |
444 | { | |
445 | /* check address is cacheline aligned */ | |
446 | if (addr & (L1_CACHE_BYTES - 1)) | |
447 | return -EINVAL; | |
448 | spin_lock(&vcpu->arch.vpa_update_lock); | |
449 | if (v->next_gpa != addr || v->len != len) { | |
450 | v->next_gpa = addr; | |
451 | v->len = addr ? len : 0; | |
452 | v->update_pending = 1; | |
453 | } | |
454 | spin_unlock(&vcpu->arch.vpa_update_lock); | |
455 | return 0; | |
456 | } | |
457 | ||
2e25aa5f PM |
458 | /* Length for a per-processor buffer is passed in at offset 4 in the buffer */ |
459 | struct reg_vpa { | |
460 | u32 dummy; | |
461 | union { | |
02407552 AG |
462 | __be16 hword; |
463 | __be32 word; | |
2e25aa5f PM |
464 | } length; |
465 | }; | |
466 | ||
467 | static int vpa_is_registered(struct kvmppc_vpa *vpap) | |
468 | { | |
469 | if (vpap->update_pending) | |
470 | return vpap->next_gpa != 0; | |
471 | return vpap->pinned_addr != NULL; | |
472 | } | |
473 | ||
a8606e20 PM |
474 | static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu, |
475 | unsigned long flags, | |
476 | unsigned long vcpuid, unsigned long vpa) | |
477 | { | |
478 | struct kvm *kvm = vcpu->kvm; | |
93e60249 | 479 | unsigned long len, nb; |
a8606e20 PM |
480 | void *va; |
481 | struct kvm_vcpu *tvcpu; | |
2e25aa5f PM |
482 | int err; |
483 | int subfunc; | |
484 | struct kvmppc_vpa *vpap; | |
a8606e20 PM |
485 | |
486 | tvcpu = kvmppc_find_vcpu(kvm, vcpuid); | |
487 | if (!tvcpu) | |
488 | return H_PARAMETER; | |
489 | ||
2e25aa5f PM |
490 | subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK; |
491 | if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL || | |
492 | subfunc == H_VPA_REG_SLB) { | |
493 | /* Registering new area - address must be cache-line aligned */ | |
494 | if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa) | |
a8606e20 | 495 | return H_PARAMETER; |
2e25aa5f PM |
496 | |
497 | /* convert logical addr to kernel addr and read length */ | |
93e60249 PM |
498 | va = kvmppc_pin_guest_page(kvm, vpa, &nb); |
499 | if (va == NULL) | |
b2b2f165 | 500 | return H_PARAMETER; |
2e25aa5f | 501 | if (subfunc == H_VPA_REG_VPA) |
02407552 | 502 | len = be16_to_cpu(((struct reg_vpa *)va)->length.hword); |
a8606e20 | 503 | else |
02407552 | 504 | len = be32_to_cpu(((struct reg_vpa *)va)->length.word); |
c35635ef | 505 | kvmppc_unpin_guest_page(kvm, va, vpa, false); |
2e25aa5f PM |
506 | |
507 | /* Check length */ | |
508 | if (len > nb || len < sizeof(struct reg_vpa)) | |
509 | return H_PARAMETER; | |
510 | } else { | |
511 | vpa = 0; | |
512 | len = 0; | |
513 | } | |
514 | ||
515 | err = H_PARAMETER; | |
516 | vpap = NULL; | |
517 | spin_lock(&tvcpu->arch.vpa_update_lock); | |
518 | ||
519 | switch (subfunc) { | |
520 | case H_VPA_REG_VPA: /* register VPA */ | |
eaac112e NP |
521 | /* |
522 | * The size of our lppaca is 1kB because of the way we align | |
523 | * it for the guest to avoid crossing a 4kB boundary. We only | |
524 | * use 640 bytes of the structure though, so we should accept | |
525 | * clients that set a size of 640. | |
526 | */ | |
499dcd41 NP |
527 | BUILD_BUG_ON(sizeof(struct lppaca) != 640); |
528 | if (len < sizeof(struct lppaca)) | |
a8606e20 | 529 | break; |
2e25aa5f PM |
530 | vpap = &tvcpu->arch.vpa; |
531 | err = 0; | |
532 | break; | |
533 | ||
534 | case H_VPA_REG_DTL: /* register DTL */ | |
535 | if (len < sizeof(struct dtl_entry)) | |
a8606e20 | 536 | break; |
2e25aa5f PM |
537 | len -= len % sizeof(struct dtl_entry); |
538 | ||
539 | /* Check that they have previously registered a VPA */ | |
540 | err = H_RESOURCE; | |
541 | if (!vpa_is_registered(&tvcpu->arch.vpa)) | |
a8606e20 | 542 | break; |
2e25aa5f PM |
543 | |
544 | vpap = &tvcpu->arch.dtl; | |
545 | err = 0; | |
546 | break; | |
547 | ||
548 | case H_VPA_REG_SLB: /* register SLB shadow buffer */ | |
549 | /* Check that they have previously registered a VPA */ | |
550 | err = H_RESOURCE; | |
551 | if (!vpa_is_registered(&tvcpu->arch.vpa)) | |
a8606e20 | 552 | break; |
2e25aa5f PM |
553 | |
554 | vpap = &tvcpu->arch.slb_shadow; | |
555 | err = 0; | |
556 | break; | |
557 | ||
558 | case H_VPA_DEREG_VPA: /* deregister VPA */ | |
559 | /* Check they don't still have a DTL or SLB buf registered */ | |
560 | err = H_RESOURCE; | |
561 | if (vpa_is_registered(&tvcpu->arch.dtl) || | |
562 | vpa_is_registered(&tvcpu->arch.slb_shadow)) | |
a8606e20 | 563 | break; |
2e25aa5f PM |
564 | |
565 | vpap = &tvcpu->arch.vpa; | |
566 | err = 0; | |
567 | break; | |
568 | ||
569 | case H_VPA_DEREG_DTL: /* deregister DTL */ | |
570 | vpap = &tvcpu->arch.dtl; | |
571 | err = 0; | |
572 | break; | |
573 | ||
574 | case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */ | |
575 | vpap = &tvcpu->arch.slb_shadow; | |
576 | err = 0; | |
577 | break; | |
578 | } | |
579 | ||
580 | if (vpap) { | |
581 | vpap->next_gpa = vpa; | |
582 | vpap->len = len; | |
583 | vpap->update_pending = 1; | |
a8606e20 | 584 | } |
93e60249 | 585 | |
2e25aa5f PM |
586 | spin_unlock(&tvcpu->arch.vpa_update_lock); |
587 | ||
93e60249 | 588 | return err; |
a8606e20 PM |
589 | } |
590 | ||
081f323b | 591 | static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap) |
2e25aa5f | 592 | { |
081f323b | 593 | struct kvm *kvm = vcpu->kvm; |
2e25aa5f PM |
594 | void *va; |
595 | unsigned long nb; | |
081f323b | 596 | unsigned long gpa; |
2e25aa5f | 597 | |
081f323b PM |
598 | /* |
599 | * We need to pin the page pointed to by vpap->next_gpa, | |
600 | * but we can't call kvmppc_pin_guest_page under the lock | |
601 | * as it does get_user_pages() and down_read(). So we | |
602 | * have to drop the lock, pin the page, then get the lock | |
603 | * again and check that a new area didn't get registered | |
604 | * in the meantime. | |
605 | */ | |
606 | for (;;) { | |
607 | gpa = vpap->next_gpa; | |
608 | spin_unlock(&vcpu->arch.vpa_update_lock); | |
609 | va = NULL; | |
610 | nb = 0; | |
611 | if (gpa) | |
c35635ef | 612 | va = kvmppc_pin_guest_page(kvm, gpa, &nb); |
081f323b PM |
613 | spin_lock(&vcpu->arch.vpa_update_lock); |
614 | if (gpa == vpap->next_gpa) | |
615 | break; | |
616 | /* sigh... unpin that one and try again */ | |
617 | if (va) | |
c35635ef | 618 | kvmppc_unpin_guest_page(kvm, va, gpa, false); |
081f323b PM |
619 | } |
620 | ||
621 | vpap->update_pending = 0; | |
622 | if (va && nb < vpap->len) { | |
623 | /* | |
624 | * If it's now too short, it must be that userspace | |
625 | * has changed the mappings underlying guest memory, | |
626 | * so unregister the region. | |
627 | */ | |
c35635ef | 628 | kvmppc_unpin_guest_page(kvm, va, gpa, false); |
081f323b | 629 | va = NULL; |
2e25aa5f PM |
630 | } |
631 | if (vpap->pinned_addr) | |
c35635ef PM |
632 | kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa, |
633 | vpap->dirty); | |
634 | vpap->gpa = gpa; | |
2e25aa5f | 635 | vpap->pinned_addr = va; |
c35635ef | 636 | vpap->dirty = false; |
2e25aa5f PM |
637 | if (va) |
638 | vpap->pinned_end = va + vpap->len; | |
639 | } | |
640 | ||
641 | static void kvmppc_update_vpas(struct kvm_vcpu *vcpu) | |
642 | { | |
2f12f034 PM |
643 | if (!(vcpu->arch.vpa.update_pending || |
644 | vcpu->arch.slb_shadow.update_pending || | |
645 | vcpu->arch.dtl.update_pending)) | |
646 | return; | |
647 | ||
2e25aa5f PM |
648 | spin_lock(&vcpu->arch.vpa_update_lock); |
649 | if (vcpu->arch.vpa.update_pending) { | |
081f323b | 650 | kvmppc_update_vpa(vcpu, &vcpu->arch.vpa); |
55b665b0 PM |
651 | if (vcpu->arch.vpa.pinned_addr) |
652 | init_vpa(vcpu, vcpu->arch.vpa.pinned_addr); | |
2e25aa5f PM |
653 | } |
654 | if (vcpu->arch.dtl.update_pending) { | |
081f323b | 655 | kvmppc_update_vpa(vcpu, &vcpu->arch.dtl); |
2e25aa5f PM |
656 | vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr; |
657 | vcpu->arch.dtl_index = 0; | |
658 | } | |
659 | if (vcpu->arch.slb_shadow.update_pending) | |
081f323b | 660 | kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow); |
2e25aa5f PM |
661 | spin_unlock(&vcpu->arch.vpa_update_lock); |
662 | } | |
663 | ||
c7b67670 PM |
664 | /* |
665 | * Return the accumulated stolen time for the vcore up until `now'. | |
666 | * The caller should hold the vcore lock. | |
667 | */ | |
668 | static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now) | |
669 | { | |
670 | u64 p; | |
2711e248 | 671 | unsigned long flags; |
c7b67670 | 672 | |
2711e248 PM |
673 | spin_lock_irqsave(&vc->stoltb_lock, flags); |
674 | p = vc->stolen_tb; | |
c7b67670 | 675 | if (vc->vcore_state != VCORE_INACTIVE && |
2711e248 PM |
676 | vc->preempt_tb != TB_NIL) |
677 | p += now - vc->preempt_tb; | |
678 | spin_unlock_irqrestore(&vc->stoltb_lock, flags); | |
c7b67670 PM |
679 | return p; |
680 | } | |
681 | ||
0456ec4f PM |
682 | static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu, |
683 | struct kvmppc_vcore *vc) | |
684 | { | |
685 | struct dtl_entry *dt; | |
686 | struct lppaca *vpa; | |
c7b67670 PM |
687 | unsigned long stolen; |
688 | unsigned long core_stolen; | |
689 | u64 now; | |
8b24e69f | 690 | unsigned long flags; |
0456ec4f PM |
691 | |
692 | dt = vcpu->arch.dtl_ptr; | |
693 | vpa = vcpu->arch.vpa.pinned_addr; | |
c7b67670 PM |
694 | now = mftb(); |
695 | core_stolen = vcore_stolen_time(vc, now); | |
696 | stolen = core_stolen - vcpu->arch.stolen_logged; | |
697 | vcpu->arch.stolen_logged = core_stolen; | |
8b24e69f | 698 | spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); |
c7b67670 PM |
699 | stolen += vcpu->arch.busy_stolen; |
700 | vcpu->arch.busy_stolen = 0; | |
8b24e69f | 701 | spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); |
0456ec4f PM |
702 | if (!dt || !vpa) |
703 | return; | |
704 | memset(dt, 0, sizeof(struct dtl_entry)); | |
705 | dt->dispatch_reason = 7; | |
02407552 AG |
706 | dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid); |
707 | dt->timebase = cpu_to_be64(now + vc->tb_offset); | |
708 | dt->enqueue_to_dispatch_time = cpu_to_be32(stolen); | |
709 | dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu)); | |
710 | dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr); | |
0456ec4f PM |
711 | ++dt; |
712 | if (dt == vcpu->arch.dtl.pinned_end) | |
713 | dt = vcpu->arch.dtl.pinned_addr; | |
714 | vcpu->arch.dtl_ptr = dt; | |
715 | /* order writing *dt vs. writing vpa->dtl_idx */ | |
716 | smp_wmb(); | |
02407552 | 717 | vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index); |
c35635ef | 718 | vcpu->arch.dtl.dirty = true; |
0456ec4f PM |
719 | } |
720 | ||
1da4e2f4 PM |
721 | /* See if there is a doorbell interrupt pending for a vcpu */ |
722 | static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu) | |
723 | { | |
724 | int thr; | |
725 | struct kvmppc_vcore *vc; | |
726 | ||
57900694 PM |
727 | if (vcpu->arch.doorbell_request) |
728 | return true; | |
729 | /* | |
730 | * Ensure that the read of vcore->dpdes comes after the read | |
731 | * of vcpu->doorbell_request. This barrier matches the | |
6fabc9f2 | 732 | * smp_wmb() in kvmppc_guest_entry_inject(). |
57900694 PM |
733 | */ |
734 | smp_rmb(); | |
1da4e2f4 PM |
735 | vc = vcpu->arch.vcore; |
736 | thr = vcpu->vcpu_id - vc->first_vcpuid; | |
737 | return !!(vc->dpdes & (1 << thr)); | |
738 | } | |
739 | ||
9642382e MN |
740 | static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu) |
741 | { | |
742 | if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207) | |
743 | return true; | |
744 | if ((!vcpu->arch.vcore->arch_compat) && | |
745 | cpu_has_feature(CPU_FTR_ARCH_207S)) | |
746 | return true; | |
747 | return false; | |
748 | } | |
749 | ||
750 | static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags, | |
751 | unsigned long resource, unsigned long value1, | |
752 | unsigned long value2) | |
753 | { | |
754 | switch (resource) { | |
755 | case H_SET_MODE_RESOURCE_SET_CIABR: | |
756 | if (!kvmppc_power8_compatible(vcpu)) | |
757 | return H_P2; | |
758 | if (value2) | |
759 | return H_P4; | |
760 | if (mflags) | |
761 | return H_UNSUPPORTED_FLAG_START; | |
762 | /* Guests can't breakpoint the hypervisor */ | |
763 | if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER) | |
764 | return H_P3; | |
765 | vcpu->arch.ciabr = value1; | |
766 | return H_SUCCESS; | |
767 | case H_SET_MODE_RESOURCE_SET_DAWR: | |
768 | if (!kvmppc_power8_compatible(vcpu)) | |
769 | return H_P2; | |
398e712c MN |
770 | if (!ppc_breakpoint_available()) |
771 | return H_P2; | |
9642382e MN |
772 | if (mflags) |
773 | return H_UNSUPPORTED_FLAG_START; | |
774 | if (value2 & DABRX_HYP) | |
775 | return H_P4; | |
776 | vcpu->arch.dawr = value1; | |
777 | vcpu->arch.dawrx = value2; | |
778 | return H_SUCCESS; | |
55d70042 NP |
779 | case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE: |
780 | /* KVM does not support mflags=2 (AIL=2) */ | |
781 | if (mflags != 0 && mflags != 3) | |
782 | return H_UNSUPPORTED_FLAG_START; | |
783 | return H_TOO_HARD; | |
9642382e MN |
784 | default: |
785 | return H_TOO_HARD; | |
786 | } | |
787 | } | |
788 | ||
2d34d1c3 SJS |
789 | /* Copy guest memory in place - must reside within a single memslot */ |
790 | static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from, | |
791 | unsigned long len) | |
792 | { | |
793 | struct kvm_memory_slot *to_memslot = NULL; | |
794 | struct kvm_memory_slot *from_memslot = NULL; | |
795 | unsigned long to_addr, from_addr; | |
796 | int r; | |
797 | ||
798 | /* Get HPA for from address */ | |
799 | from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT); | |
800 | if (!from_memslot) | |
801 | return -EFAULT; | |
802 | if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages) | |
803 | << PAGE_SHIFT)) | |
804 | return -EINVAL; | |
805 | from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT); | |
806 | if (kvm_is_error_hva(from_addr)) | |
807 | return -EFAULT; | |
808 | from_addr |= (from & (PAGE_SIZE - 1)); | |
809 | ||
810 | /* Get HPA for to address */ | |
811 | to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT); | |
812 | if (!to_memslot) | |
813 | return -EFAULT; | |
814 | if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages) | |
815 | << PAGE_SHIFT)) | |
816 | return -EINVAL; | |
817 | to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT); | |
818 | if (kvm_is_error_hva(to_addr)) | |
819 | return -EFAULT; | |
820 | to_addr |= (to & (PAGE_SIZE - 1)); | |
821 | ||
822 | /* Perform copy */ | |
823 | r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr, | |
824 | len); | |
825 | if (r) | |
826 | return -EFAULT; | |
827 | mark_page_dirty(kvm, to >> PAGE_SHIFT); | |
828 | return 0; | |
829 | } | |
830 | ||
831 | static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags, | |
832 | unsigned long dest, unsigned long src) | |
833 | { | |
834 | u64 pg_sz = SZ_4K; /* 4K page size */ | |
835 | u64 pg_mask = SZ_4K - 1; | |
836 | int ret; | |
837 | ||
838 | /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */ | |
839 | if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE | | |
840 | H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED)) | |
841 | return H_PARAMETER; | |
842 | ||
843 | /* dest (and src if copy_page flag set) must be page aligned */ | |
844 | if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask))) | |
845 | return H_PARAMETER; | |
846 | ||
847 | /* zero and/or copy the page as determined by the flags */ | |
848 | if (flags & H_COPY_PAGE) { | |
849 | ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz); | |
850 | if (ret < 0) | |
851 | return H_PARAMETER; | |
852 | } else if (flags & H_ZERO_PAGE) { | |
853 | ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz); | |
854 | if (ret < 0) | |
855 | return H_PARAMETER; | |
856 | } | |
857 | ||
858 | /* We can ignore the remaining flags */ | |
859 | ||
860 | return H_SUCCESS; | |
861 | } | |
862 | ||
90fd09f8 SB |
863 | static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target) |
864 | { | |
865 | struct kvmppc_vcore *vcore = target->arch.vcore; | |
866 | ||
867 | /* | |
868 | * We expect to have been called by the real mode handler | |
869 | * (kvmppc_rm_h_confer()) which would have directly returned | |
870 | * H_SUCCESS if the source vcore wasn't idle (e.g. if it may | |
871 | * have useful work to do and should not confer) so we don't | |
872 | * recheck that here. | |
873 | */ | |
874 | ||
875 | spin_lock(&vcore->lock); | |
876 | if (target->arch.state == KVMPPC_VCPU_RUNNABLE && | |
ec257165 PM |
877 | vcore->vcore_state != VCORE_INACTIVE && |
878 | vcore->runner) | |
90fd09f8 SB |
879 | target = vcore->runner; |
880 | spin_unlock(&vcore->lock); | |
881 | ||
882 | return kvm_vcpu_yield_to(target); | |
883 | } | |
884 | ||
885 | static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu) | |
886 | { | |
887 | int yield_count = 0; | |
888 | struct lppaca *lppaca; | |
889 | ||
890 | spin_lock(&vcpu->arch.vpa_update_lock); | |
891 | lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr; | |
892 | if (lppaca) | |
ecb6d618 | 893 | yield_count = be32_to_cpu(lppaca->yield_count); |
90fd09f8 SB |
894 | spin_unlock(&vcpu->arch.vpa_update_lock); |
895 | return yield_count; | |
896 | } | |
897 | ||
a8606e20 PM |
898 | int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu) |
899 | { | |
900 | unsigned long req = kvmppc_get_gpr(vcpu, 3); | |
901 | unsigned long target, ret = H_SUCCESS; | |
90fd09f8 | 902 | int yield_count; |
a8606e20 | 903 | struct kvm_vcpu *tvcpu; |
8e591cb7 | 904 | int idx, rc; |
a8606e20 | 905 | |
699a0ea0 PM |
906 | if (req <= MAX_HCALL_OPCODE && |
907 | !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls)) | |
908 | return RESUME_HOST; | |
909 | ||
a8606e20 PM |
910 | switch (req) { |
911 | case H_CEDE: | |
a8606e20 PM |
912 | break; |
913 | case H_PROD: | |
914 | target = kvmppc_get_gpr(vcpu, 4); | |
915 | tvcpu = kvmppc_find_vcpu(vcpu->kvm, target); | |
916 | if (!tvcpu) { | |
917 | ret = H_PARAMETER; | |
918 | break; | |
919 | } | |
920 | tvcpu->arch.prodded = 1; | |
921 | smp_mb(); | |
8464c884 PM |
922 | if (tvcpu->arch.ceded) |
923 | kvmppc_fast_vcpu_kick_hv(tvcpu); | |
a8606e20 PM |
924 | break; |
925 | case H_CONFER: | |
42d7604d PM |
926 | target = kvmppc_get_gpr(vcpu, 4); |
927 | if (target == -1) | |
928 | break; | |
929 | tvcpu = kvmppc_find_vcpu(vcpu->kvm, target); | |
930 | if (!tvcpu) { | |
931 | ret = H_PARAMETER; | |
932 | break; | |
933 | } | |
90fd09f8 SB |
934 | yield_count = kvmppc_get_gpr(vcpu, 5); |
935 | if (kvmppc_get_yield_count(tvcpu) != yield_count) | |
936 | break; | |
937 | kvm_arch_vcpu_yield_to(tvcpu); | |
a8606e20 PM |
938 | break; |
939 | case H_REGISTER_VPA: | |
940 | ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4), | |
941 | kvmppc_get_gpr(vcpu, 5), | |
942 | kvmppc_get_gpr(vcpu, 6)); | |
943 | break; | |
8e591cb7 ME |
944 | case H_RTAS: |
945 | if (list_empty(&vcpu->kvm->arch.rtas_tokens)) | |
946 | return RESUME_HOST; | |
947 | ||
c9438092 | 948 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
8e591cb7 | 949 | rc = kvmppc_rtas_hcall(vcpu); |
c9438092 | 950 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8e591cb7 ME |
951 | |
952 | if (rc == -ENOENT) | |
953 | return RESUME_HOST; | |
954 | else if (rc == 0) | |
955 | break; | |
956 | ||
957 | /* Send the error out to userspace via KVM_RUN */ | |
958 | return rc; | |
99342cf8 DG |
959 | case H_LOGICAL_CI_LOAD: |
960 | ret = kvmppc_h_logical_ci_load(vcpu); | |
961 | if (ret == H_TOO_HARD) | |
962 | return RESUME_HOST; | |
963 | break; | |
964 | case H_LOGICAL_CI_STORE: | |
965 | ret = kvmppc_h_logical_ci_store(vcpu); | |
966 | if (ret == H_TOO_HARD) | |
967 | return RESUME_HOST; | |
968 | break; | |
9642382e MN |
969 | case H_SET_MODE: |
970 | ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4), | |
971 | kvmppc_get_gpr(vcpu, 5), | |
972 | kvmppc_get_gpr(vcpu, 6), | |
973 | kvmppc_get_gpr(vcpu, 7)); | |
974 | if (ret == H_TOO_HARD) | |
975 | return RESUME_HOST; | |
976 | break; | |
bc5ad3f3 BH |
977 | case H_XIRR: |
978 | case H_CPPR: | |
979 | case H_EOI: | |
980 | case H_IPI: | |
8e44ddc3 PM |
981 | case H_IPOLL: |
982 | case H_XIRR_X: | |
bc5ad3f3 | 983 | if (kvmppc_xics_enabled(vcpu)) { |
03f95332 | 984 | if (xics_on_xive()) { |
5af50993 BH |
985 | ret = H_NOT_AVAILABLE; |
986 | return RESUME_GUEST; | |
987 | } | |
bc5ad3f3 BH |
988 | ret = kvmppc_xics_hcall(vcpu, req); |
989 | break; | |
d3695aa4 AK |
990 | } |
991 | return RESUME_HOST; | |
4bad7779 PM |
992 | case H_SET_DABR: |
993 | ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4)); | |
994 | break; | |
995 | case H_SET_XDABR: | |
996 | ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4), | |
997 | kvmppc_get_gpr(vcpu, 5)); | |
998 | break; | |
e40542af | 999 | #ifdef CONFIG_SPAPR_TCE_IOMMU |
4bad7779 PM |
1000 | case H_GET_TCE: |
1001 | ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4), | |
1002 | kvmppc_get_gpr(vcpu, 5)); | |
1003 | if (ret == H_TOO_HARD) | |
1004 | return RESUME_HOST; | |
1005 | break; | |
d3695aa4 AK |
1006 | case H_PUT_TCE: |
1007 | ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4), | |
1008 | kvmppc_get_gpr(vcpu, 5), | |
1009 | kvmppc_get_gpr(vcpu, 6)); | |
1010 | if (ret == H_TOO_HARD) | |
1011 | return RESUME_HOST; | |
1012 | break; | |
1013 | case H_PUT_TCE_INDIRECT: | |
1014 | ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4), | |
1015 | kvmppc_get_gpr(vcpu, 5), | |
1016 | kvmppc_get_gpr(vcpu, 6), | |
1017 | kvmppc_get_gpr(vcpu, 7)); | |
1018 | if (ret == H_TOO_HARD) | |
1019 | return RESUME_HOST; | |
1020 | break; | |
1021 | case H_STUFF_TCE: | |
1022 | ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4), | |
1023 | kvmppc_get_gpr(vcpu, 5), | |
1024 | kvmppc_get_gpr(vcpu, 6), | |
1025 | kvmppc_get_gpr(vcpu, 7)); | |
1026 | if (ret == H_TOO_HARD) | |
1027 | return RESUME_HOST; | |
1028 | break; | |
e40542af | 1029 | #endif |
4bad7779 PM |
1030 | case H_RANDOM: |
1031 | if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4])) | |
1032 | ret = H_HARDWARE; | |
1033 | break; | |
8e3f5fc1 PM |
1034 | |
1035 | case H_SET_PARTITION_TABLE: | |
1036 | ret = H_FUNCTION; | |
aa069a99 | 1037 | if (nesting_enabled(vcpu->kvm)) |
8e3f5fc1 PM |
1038 | ret = kvmhv_set_partition_table(vcpu); |
1039 | break; | |
1040 | case H_ENTER_NESTED: | |
1041 | ret = H_FUNCTION; | |
aa069a99 | 1042 | if (!nesting_enabled(vcpu->kvm)) |
360cae31 PM |
1043 | break; |
1044 | ret = kvmhv_enter_nested_guest(vcpu); | |
1045 | if (ret == H_INTERRUPT) { | |
1046 | kvmppc_set_gpr(vcpu, 3, 0); | |
6c08ec12 | 1047 | vcpu->arch.hcall_needed = 0; |
360cae31 | 1048 | return -EINTR; |
873db2cd SJS |
1049 | } else if (ret == H_TOO_HARD) { |
1050 | kvmppc_set_gpr(vcpu, 3, 0); | |
1051 | vcpu->arch.hcall_needed = 0; | |
1052 | return RESUME_HOST; | |
360cae31 | 1053 | } |
8e3f5fc1 PM |
1054 | break; |
1055 | case H_TLB_INVALIDATE: | |
1056 | ret = H_FUNCTION; | |
aa069a99 PM |
1057 | if (nesting_enabled(vcpu->kvm)) |
1058 | ret = kvmhv_do_nested_tlbie(vcpu); | |
8e3f5fc1 | 1059 | break; |
6ff887b8 SJS |
1060 | case H_COPY_TOFROM_GUEST: |
1061 | ret = H_FUNCTION; | |
1062 | if (nesting_enabled(vcpu->kvm)) | |
1063 | ret = kvmhv_copy_tofrom_guest_nested(vcpu); | |
1064 | break; | |
2d34d1c3 SJS |
1065 | case H_PAGE_INIT: |
1066 | ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4), | |
1067 | kvmppc_get_gpr(vcpu, 5), | |
1068 | kvmppc_get_gpr(vcpu, 6)); | |
1069 | break; | |
ca9f4942 | 1070 | case H_SVM_PAGE_IN: |
8c47b6ff LD |
1071 | ret = H_UNSUPPORTED; |
1072 | if (kvmppc_get_srr1(vcpu) & MSR_S) | |
1073 | ret = kvmppc_h_svm_page_in(vcpu->kvm, | |
1074 | kvmppc_get_gpr(vcpu, 4), | |
1075 | kvmppc_get_gpr(vcpu, 5), | |
1076 | kvmppc_get_gpr(vcpu, 6)); | |
ca9f4942 BR |
1077 | break; |
1078 | case H_SVM_PAGE_OUT: | |
8c47b6ff LD |
1079 | ret = H_UNSUPPORTED; |
1080 | if (kvmppc_get_srr1(vcpu) & MSR_S) | |
1081 | ret = kvmppc_h_svm_page_out(vcpu->kvm, | |
1082 | kvmppc_get_gpr(vcpu, 4), | |
1083 | kvmppc_get_gpr(vcpu, 5), | |
1084 | kvmppc_get_gpr(vcpu, 6)); | |
ca9f4942 BR |
1085 | break; |
1086 | case H_SVM_INIT_START: | |
8c47b6ff LD |
1087 | ret = H_UNSUPPORTED; |
1088 | if (kvmppc_get_srr1(vcpu) & MSR_S) | |
1089 | ret = kvmppc_h_svm_init_start(vcpu->kvm); | |
ca9f4942 BR |
1090 | break; |
1091 | case H_SVM_INIT_DONE: | |
8c47b6ff LD |
1092 | ret = H_UNSUPPORTED; |
1093 | if (kvmppc_get_srr1(vcpu) & MSR_S) | |
1094 | ret = kvmppc_h_svm_init_done(vcpu->kvm); | |
ca9f4942 | 1095 | break; |
3a43970d | 1096 | case H_SVM_INIT_ABORT: |
e3326ae3 LD |
1097 | /* |
1098 | * Even if that call is made by the Ultravisor, the SSR1 value | |
1099 | * is the guest context one, with the secure bit clear as it has | |
1100 | * not yet been secured. So we can't check it here. | |
1101 | * Instead the kvm->arch.secure_guest flag is checked inside | |
1102 | * kvmppc_h_svm_init_abort(). | |
1103 | */ | |
1104 | ret = kvmppc_h_svm_init_abort(vcpu->kvm); | |
3a43970d | 1105 | break; |
ca9f4942 | 1106 | |
a8606e20 PM |
1107 | default: |
1108 | return RESUME_HOST; | |
1109 | } | |
1110 | kvmppc_set_gpr(vcpu, 3, ret); | |
1111 | vcpu->arch.hcall_needed = 0; | |
1112 | return RESUME_GUEST; | |
1113 | } | |
1114 | ||
4bad7779 PM |
1115 | /* |
1116 | * Handle H_CEDE in the nested virtualization case where we haven't | |
1117 | * called the real-mode hcall handlers in book3s_hv_rmhandlers.S. | |
1118 | * This has to be done early, not in kvmppc_pseries_do_hcall(), so | |
1119 | * that the cede logic in kvmppc_run_single_vcpu() works properly. | |
1120 | */ | |
1121 | static void kvmppc_nested_cede(struct kvm_vcpu *vcpu) | |
1122 | { | |
1123 | vcpu->arch.shregs.msr |= MSR_EE; | |
1124 | vcpu->arch.ceded = 1; | |
1125 | smp_mb(); | |
1126 | if (vcpu->arch.prodded) { | |
1127 | vcpu->arch.prodded = 0; | |
1128 | smp_mb(); | |
1129 | vcpu->arch.ceded = 0; | |
1130 | } | |
1131 | } | |
1132 | ||
ae2113a4 PM |
1133 | static int kvmppc_hcall_impl_hv(unsigned long cmd) |
1134 | { | |
1135 | switch (cmd) { | |
1136 | case H_CEDE: | |
1137 | case H_PROD: | |
1138 | case H_CONFER: | |
1139 | case H_REGISTER_VPA: | |
9642382e | 1140 | case H_SET_MODE: |
99342cf8 DG |
1141 | case H_LOGICAL_CI_LOAD: |
1142 | case H_LOGICAL_CI_STORE: | |
ae2113a4 PM |
1143 | #ifdef CONFIG_KVM_XICS |
1144 | case H_XIRR: | |
1145 | case H_CPPR: | |
1146 | case H_EOI: | |
1147 | case H_IPI: | |
1148 | case H_IPOLL: | |
1149 | case H_XIRR_X: | |
1150 | #endif | |
2d34d1c3 | 1151 | case H_PAGE_INIT: |
ae2113a4 PM |
1152 | return 1; |
1153 | } | |
1154 | ||
1155 | /* See if it's in the real-mode table */ | |
1156 | return kvmppc_hcall_impl_hv_realmode(cmd); | |
1157 | } | |
1158 | ||
8c99d345 | 1159 | static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu) |
a59c1d9e MS |
1160 | { |
1161 | u32 last_inst; | |
1162 | ||
1163 | if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) != | |
1164 | EMULATE_DONE) { | |
1165 | /* | |
1166 | * Fetch failed, so return to guest and | |
1167 | * try executing it again. | |
1168 | */ | |
1169 | return RESUME_GUEST; | |
1170 | } | |
1171 | ||
1172 | if (last_inst == KVMPPC_INST_SW_BREAKPOINT) { | |
8c99d345 TZ |
1173 | vcpu->run->exit_reason = KVM_EXIT_DEBUG; |
1174 | vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu); | |
a59c1d9e MS |
1175 | return RESUME_HOST; |
1176 | } else { | |
1177 | kvmppc_core_queue_program(vcpu, SRR1_PROGILL); | |
1178 | return RESUME_GUEST; | |
1179 | } | |
1180 | } | |
1181 | ||
57900694 PM |
1182 | static void do_nothing(void *x) |
1183 | { | |
1184 | } | |
1185 | ||
1186 | static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu) | |
1187 | { | |
1188 | int thr, cpu, pcpu, nthreads; | |
1189 | struct kvm_vcpu *v; | |
1190 | unsigned long dpdes; | |
1191 | ||
1192 | nthreads = vcpu->kvm->arch.emul_smt_mode; | |
1193 | dpdes = 0; | |
1194 | cpu = vcpu->vcpu_id & ~(nthreads - 1); | |
1195 | for (thr = 0; thr < nthreads; ++thr, ++cpu) { | |
1196 | v = kvmppc_find_vcpu(vcpu->kvm, cpu); | |
1197 | if (!v) | |
1198 | continue; | |
1199 | /* | |
1200 | * If the vcpu is currently running on a physical cpu thread, | |
1201 | * interrupt it in order to pull it out of the guest briefly, | |
1202 | * which will update its vcore->dpdes value. | |
1203 | */ | |
1204 | pcpu = READ_ONCE(v->cpu); | |
1205 | if (pcpu >= 0) | |
1206 | smp_call_function_single(pcpu, do_nothing, NULL, 1); | |
1207 | if (kvmppc_doorbell_pending(v)) | |
1208 | dpdes |= 1 << thr; | |
1209 | } | |
1210 | return dpdes; | |
1211 | } | |
1212 | ||
1213 | /* | |
1214 | * On POWER9, emulate doorbell-related instructions in order to | |
1215 | * give the guest the illusion of running on a multi-threaded core. | |
1216 | * The instructions emulated are msgsndp, msgclrp, mfspr TIR, | |
1217 | * and mfspr DPDES. | |
1218 | */ | |
1219 | static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu) | |
1220 | { | |
1221 | u32 inst, rb, thr; | |
1222 | unsigned long arg; | |
1223 | struct kvm *kvm = vcpu->kvm; | |
1224 | struct kvm_vcpu *tvcpu; | |
1225 | ||
57900694 PM |
1226 | if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE) |
1227 | return RESUME_GUEST; | |
1228 | if (get_op(inst) != 31) | |
1229 | return EMULATE_FAIL; | |
1230 | rb = get_rb(inst); | |
1231 | thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1); | |
1232 | switch (get_xop(inst)) { | |
1233 | case OP_31_XOP_MSGSNDP: | |
1234 | arg = kvmppc_get_gpr(vcpu, rb); | |
1235 | if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER) | |
1236 | break; | |
1237 | arg &= 0x3f; | |
1238 | if (arg >= kvm->arch.emul_smt_mode) | |
1239 | break; | |
1240 | tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg); | |
1241 | if (!tvcpu) | |
1242 | break; | |
1243 | if (!tvcpu->arch.doorbell_request) { | |
1244 | tvcpu->arch.doorbell_request = 1; | |
1245 | kvmppc_fast_vcpu_kick_hv(tvcpu); | |
1246 | } | |
1247 | break; | |
1248 | case OP_31_XOP_MSGCLRP: | |
1249 | arg = kvmppc_get_gpr(vcpu, rb); | |
1250 | if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER) | |
1251 | break; | |
1252 | vcpu->arch.vcore->dpdes = 0; | |
1253 | vcpu->arch.doorbell_request = 0; | |
1254 | break; | |
1255 | case OP_31_XOP_MFSPR: | |
1256 | switch (get_sprn(inst)) { | |
1257 | case SPRN_TIR: | |
1258 | arg = thr; | |
1259 | break; | |
1260 | case SPRN_DPDES: | |
1261 | arg = kvmppc_read_dpdes(vcpu); | |
1262 | break; | |
1263 | default: | |
1264 | return EMULATE_FAIL; | |
1265 | } | |
1266 | kvmppc_set_gpr(vcpu, get_rt(inst), arg); | |
1267 | break; | |
1268 | default: | |
1269 | return EMULATE_FAIL; | |
1270 | } | |
1271 | kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4); | |
1272 | return RESUME_GUEST; | |
1273 | } | |
1274 | ||
8c99d345 | 1275 | static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu, |
3a167bea | 1276 | struct task_struct *tsk) |
de56a948 | 1277 | { |
8c99d345 | 1278 | struct kvm_run *run = vcpu->run; |
de56a948 PM |
1279 | int r = RESUME_HOST; |
1280 | ||
1281 | vcpu->stat.sum_exits++; | |
1282 | ||
1c9e3d51 PM |
1283 | /* |
1284 | * This can happen if an interrupt occurs in the last stages | |
1285 | * of guest entry or the first stages of guest exit (i.e. after | |
1286 | * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV | |
1287 | * and before setting it to KVM_GUEST_MODE_HOST_HV). | |
1288 | * That can happen due to a bug, or due to a machine check | |
1289 | * occurring at just the wrong time. | |
1290 | */ | |
1291 | if (vcpu->arch.shregs.msr & MSR_HV) { | |
1292 | printk(KERN_EMERG "KVM trap in HV mode!\n"); | |
1293 | printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n", | |
1294 | vcpu->arch.trap, kvmppc_get_pc(vcpu), | |
1295 | vcpu->arch.shregs.msr); | |
1296 | kvmppc_dump_regs(vcpu); | |
1297 | run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
1298 | run->hw.hardware_exit_reason = vcpu->arch.trap; | |
1299 | return RESUME_HOST; | |
1300 | } | |
de56a948 PM |
1301 | run->exit_reason = KVM_EXIT_UNKNOWN; |
1302 | run->ready_for_interrupt_injection = 1; | |
1303 | switch (vcpu->arch.trap) { | |
1304 | /* We're good on these - the host merely wanted to get our attention */ | |
1305 | case BOOK3S_INTERRUPT_HV_DECREMENTER: | |
1306 | vcpu->stat.dec_exits++; | |
1307 | r = RESUME_GUEST; | |
1308 | break; | |
1309 | case BOOK3S_INTERRUPT_EXTERNAL: | |
5d00f66b | 1310 | case BOOK3S_INTERRUPT_H_DOORBELL: |
84f7139c | 1311 | case BOOK3S_INTERRUPT_H_VIRT: |
de56a948 PM |
1312 | vcpu->stat.ext_intr_exits++; |
1313 | r = RESUME_GUEST; | |
1314 | break; | |
6de6638b | 1315 | /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/ |
dee6f24c | 1316 | case BOOK3S_INTERRUPT_HMI: |
de56a948 | 1317 | case BOOK3S_INTERRUPT_PERFMON: |
6de6638b | 1318 | case BOOK3S_INTERRUPT_SYSTEM_RESET: |
de56a948 PM |
1319 | r = RESUME_GUEST; |
1320 | break; | |
b4072df4 | 1321 | case BOOK3S_INTERRUPT_MACHINE_CHECK: |
884dfb72 | 1322 | /* Print the MCE event to host console. */ |
c0577201 | 1323 | machine_check_print_event_info(&vcpu->arch.mce_evt, false, true); |
884dfb72 PM |
1324 | |
1325 | /* | |
1326 | * If the guest can do FWNMI, exit to userspace so it can | |
1327 | * deliver a FWNMI to the guest. | |
1328 | * Otherwise we synthesize a machine check for the guest | |
1329 | * so that it knows that the machine check occurred. | |
1330 | */ | |
1331 | if (!vcpu->kvm->arch.fwnmi_enabled) { | |
1332 | ulong flags = vcpu->arch.shregs.msr & 0x083c0000; | |
1333 | kvmppc_core_queue_machine_check(vcpu, flags); | |
1334 | r = RESUME_GUEST; | |
1335 | break; | |
1336 | } | |
1337 | ||
e20bbd3d AP |
1338 | /* Exit to guest with KVM_EXIT_NMI as exit reason */ |
1339 | run->exit_reason = KVM_EXIT_NMI; | |
1340 | run->hw.hardware_exit_reason = vcpu->arch.trap; | |
1341 | /* Clear out the old NMI status from run->flags */ | |
1342 | run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK; | |
1343 | /* Now set the NMI status */ | |
1344 | if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED) | |
1345 | run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV; | |
1346 | else | |
1347 | run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV; | |
1348 | ||
1349 | r = RESUME_HOST; | |
b4072df4 | 1350 | break; |
de56a948 PM |
1351 | case BOOK3S_INTERRUPT_PROGRAM: |
1352 | { | |
1353 | ulong flags; | |
1354 | /* | |
1355 | * Normally program interrupts are delivered directly | |
1356 | * to the guest by the hardware, but we can get here | |
1357 | * as a result of a hypervisor emulation interrupt | |
1358 | * (e40) getting turned into a 700 by BML RTAS. | |
1359 | */ | |
1360 | flags = vcpu->arch.shregs.msr & 0x1f0000ull; | |
1361 | kvmppc_core_queue_program(vcpu, flags); | |
1362 | r = RESUME_GUEST; | |
1363 | break; | |
1364 | } | |
1365 | case BOOK3S_INTERRUPT_SYSCALL: | |
1366 | { | |
1367 | /* hcall - punt to userspace */ | |
1368 | int i; | |
1369 | ||
27025a60 LPF |
1370 | /* hypercall with MSR_PR has already been handled in rmode, |
1371 | * and never reaches here. | |
1372 | */ | |
1373 | ||
de56a948 PM |
1374 | run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3); |
1375 | for (i = 0; i < 9; ++i) | |
1376 | run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i); | |
1377 | run->exit_reason = KVM_EXIT_PAPR_HCALL; | |
1378 | vcpu->arch.hcall_needed = 1; | |
1379 | r = RESUME_HOST; | |
1380 | break; | |
1381 | } | |
1382 | /* | |
342d3db7 PM |
1383 | * We get these next two if the guest accesses a page which it thinks |
1384 | * it has mapped but which is not actually present, either because | |
1385 | * it is for an emulated I/O device or because the corresonding | |
1386 | * host page has been paged out. Any other HDSI/HISI interrupts | |
1387 | * have been handled already. | |
de56a948 PM |
1388 | */ |
1389 | case BOOK3S_INTERRUPT_H_DATA_STORAGE: | |
913d3ff9 | 1390 | r = RESUME_PAGE_FAULT; |
de56a948 PM |
1391 | break; |
1392 | case BOOK3S_INTERRUPT_H_INST_STORAGE: | |
913d3ff9 | 1393 | vcpu->arch.fault_dar = kvmppc_get_pc(vcpu); |
32eb150a PM |
1394 | vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr & |
1395 | DSISR_SRR1_MATCH_64S; | |
1396 | if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE) | |
1397 | vcpu->arch.fault_dsisr |= DSISR_ISSTORE; | |
913d3ff9 | 1398 | r = RESUME_PAGE_FAULT; |
de56a948 PM |
1399 | break; |
1400 | /* | |
1401 | * This occurs if the guest executes an illegal instruction. | |
a59c1d9e MS |
1402 | * If the guest debug is disabled, generate a program interrupt |
1403 | * to the guest. If guest debug is enabled, we need to check | |
1404 | * whether the instruction is a software breakpoint instruction. | |
1405 | * Accordingly return to Guest or Host. | |
de56a948 PM |
1406 | */ |
1407 | case BOOK3S_INTERRUPT_H_EMUL_ASSIST: | |
4a157d61 PM |
1408 | if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED) |
1409 | vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ? | |
1410 | swab32(vcpu->arch.emul_inst) : | |
1411 | vcpu->arch.emul_inst; | |
a59c1d9e | 1412 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) { |
8c99d345 | 1413 | r = kvmppc_emulate_debug_inst(vcpu); |
a59c1d9e MS |
1414 | } else { |
1415 | kvmppc_core_queue_program(vcpu, SRR1_PROGILL); | |
1416 | r = RESUME_GUEST; | |
1417 | } | |
bd3048b8 ME |
1418 | break; |
1419 | /* | |
1420 | * This occurs if the guest (kernel or userspace), does something that | |
57900694 PM |
1421 | * is prohibited by HFSCR. |
1422 | * On POWER9, this could be a doorbell instruction that we need | |
1423 | * to emulate. | |
1424 | * Otherwise, we just generate a program interrupt to the guest. | |
bd3048b8 ME |
1425 | */ |
1426 | case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: | |
57900694 | 1427 | r = EMULATE_FAIL; |
36ee41d1 | 1428 | if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) && |
53655ddd | 1429 | cpu_has_feature(CPU_FTR_ARCH_300)) |
57900694 PM |
1430 | r = kvmppc_emulate_doorbell_instr(vcpu); |
1431 | if (r == EMULATE_FAIL) { | |
1432 | kvmppc_core_queue_program(vcpu, SRR1_PROGILL); | |
1433 | r = RESUME_GUEST; | |
1434 | } | |
de56a948 | 1435 | break; |
4bb3c7a0 PM |
1436 | |
1437 | #ifdef CONFIG_PPC_TRANSACTIONAL_MEM | |
1438 | case BOOK3S_INTERRUPT_HV_SOFTPATCH: | |
1439 | /* | |
1440 | * This occurs for various TM-related instructions that | |
1441 | * we need to emulate on POWER9 DD2.2. We have already | |
1442 | * handled the cases where the guest was in real-suspend | |
1443 | * mode and was transitioning to transactional state. | |
1444 | */ | |
1445 | r = kvmhv_p9_tm_emulation(vcpu); | |
1446 | break; | |
1447 | #endif | |
1448 | ||
f7af5209 SW |
1449 | case BOOK3S_INTERRUPT_HV_RM_HARD: |
1450 | r = RESUME_PASSTHROUGH; | |
1451 | break; | |
de56a948 PM |
1452 | default: |
1453 | kvmppc_dump_regs(vcpu); | |
1454 | printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n", | |
1455 | vcpu->arch.trap, kvmppc_get_pc(vcpu), | |
1456 | vcpu->arch.shregs.msr); | |
f3271d4c | 1457 | run->hw.hardware_exit_reason = vcpu->arch.trap; |
de56a948 | 1458 | r = RESUME_HOST; |
de56a948 PM |
1459 | break; |
1460 | } | |
1461 | ||
de56a948 PM |
1462 | return r; |
1463 | } | |
1464 | ||
8c99d345 | 1465 | static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu) |
360cae31 PM |
1466 | { |
1467 | int r; | |
1468 | int srcu_idx; | |
1469 | ||
1470 | vcpu->stat.sum_exits++; | |
1471 | ||
1472 | /* | |
1473 | * This can happen if an interrupt occurs in the last stages | |
1474 | * of guest entry or the first stages of guest exit (i.e. after | |
1475 | * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV | |
1476 | * and before setting it to KVM_GUEST_MODE_HOST_HV). | |
1477 | * That can happen due to a bug, or due to a machine check | |
1478 | * occurring at just the wrong time. | |
1479 | */ | |
1480 | if (vcpu->arch.shregs.msr & MSR_HV) { | |
1481 | pr_emerg("KVM trap in HV mode while nested!\n"); | |
1482 | pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n", | |
1483 | vcpu->arch.trap, kvmppc_get_pc(vcpu), | |
1484 | vcpu->arch.shregs.msr); | |
1485 | kvmppc_dump_regs(vcpu); | |
1486 | return RESUME_HOST; | |
1487 | } | |
1488 | switch (vcpu->arch.trap) { | |
1489 | /* We're good on these - the host merely wanted to get our attention */ | |
1490 | case BOOK3S_INTERRUPT_HV_DECREMENTER: | |
1491 | vcpu->stat.dec_exits++; | |
1492 | r = RESUME_GUEST; | |
1493 | break; | |
1494 | case BOOK3S_INTERRUPT_EXTERNAL: | |
1495 | vcpu->stat.ext_intr_exits++; | |
1496 | r = RESUME_HOST; | |
1497 | break; | |
1498 | case BOOK3S_INTERRUPT_H_DOORBELL: | |
1499 | case BOOK3S_INTERRUPT_H_VIRT: | |
1500 | vcpu->stat.ext_intr_exits++; | |
1501 | r = RESUME_GUEST; | |
1502 | break; | |
1503 | /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/ | |
1504 | case BOOK3S_INTERRUPT_HMI: | |
1505 | case BOOK3S_INTERRUPT_PERFMON: | |
1506 | case BOOK3S_INTERRUPT_SYSTEM_RESET: | |
1507 | r = RESUME_GUEST; | |
1508 | break; | |
1509 | case BOOK3S_INTERRUPT_MACHINE_CHECK: | |
1510 | /* Pass the machine check to the L1 guest */ | |
1511 | r = RESUME_HOST; | |
1512 | /* Print the MCE event to host console. */ | |
c0577201 | 1513 | machine_check_print_event_info(&vcpu->arch.mce_evt, false, true); |
360cae31 PM |
1514 | break; |
1515 | /* | |
1516 | * We get these next two if the guest accesses a page which it thinks | |
1517 | * it has mapped but which is not actually present, either because | |
1518 | * it is for an emulated I/O device or because the corresonding | |
1519 | * host page has been paged out. | |
1520 | */ | |
1521 | case BOOK3S_INTERRUPT_H_DATA_STORAGE: | |
1522 | srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); | |
8c99d345 | 1523 | r = kvmhv_nested_page_fault(vcpu); |
360cae31 PM |
1524 | srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); |
1525 | break; | |
1526 | case BOOK3S_INTERRUPT_H_INST_STORAGE: | |
1527 | vcpu->arch.fault_dar = kvmppc_get_pc(vcpu); | |
1528 | vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) & | |
1529 | DSISR_SRR1_MATCH_64S; | |
1530 | if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE) | |
1531 | vcpu->arch.fault_dsisr |= DSISR_ISSTORE; | |
1532 | srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); | |
8c99d345 | 1533 | r = kvmhv_nested_page_fault(vcpu); |
360cae31 PM |
1534 | srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); |
1535 | break; | |
1536 | ||
1537 | #ifdef CONFIG_PPC_TRANSACTIONAL_MEM | |
1538 | case BOOK3S_INTERRUPT_HV_SOFTPATCH: | |
1539 | /* | |
1540 | * This occurs for various TM-related instructions that | |
1541 | * we need to emulate on POWER9 DD2.2. We have already | |
1542 | * handled the cases where the guest was in real-suspend | |
1543 | * mode and was transitioning to transactional state. | |
1544 | */ | |
1545 | r = kvmhv_p9_tm_emulation(vcpu); | |
1546 | break; | |
1547 | #endif | |
1548 | ||
1549 | case BOOK3S_INTERRUPT_HV_RM_HARD: | |
1550 | vcpu->arch.trap = 0; | |
1551 | r = RESUME_GUEST; | |
03f95332 | 1552 | if (!xics_on_xive()) |
360cae31 PM |
1553 | kvmppc_xics_rm_complete(vcpu, 0); |
1554 | break; | |
1555 | default: | |
1556 | r = RESUME_HOST; | |
1557 | break; | |
1558 | } | |
1559 | ||
1560 | return r; | |
1561 | } | |
1562 | ||
3a167bea AK |
1563 | static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu, |
1564 | struct kvm_sregs *sregs) | |
de56a948 PM |
1565 | { |
1566 | int i; | |
1567 | ||
de56a948 | 1568 | memset(sregs, 0, sizeof(struct kvm_sregs)); |
87916442 | 1569 | sregs->pvr = vcpu->arch.pvr; |
de56a948 PM |
1570 | for (i = 0; i < vcpu->arch.slb_max; i++) { |
1571 | sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige; | |
1572 | sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv; | |
1573 | } | |
1574 | ||
1575 | return 0; | |
1576 | } | |
1577 | ||
3a167bea AK |
1578 | static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu, |
1579 | struct kvm_sregs *sregs) | |
de56a948 PM |
1580 | { |
1581 | int i, j; | |
1582 | ||
9333e6c4 PM |
1583 | /* Only accept the same PVR as the host's, since we can't spoof it */ |
1584 | if (sregs->pvr != vcpu->arch.pvr) | |
1585 | return -EINVAL; | |
de56a948 PM |
1586 | |
1587 | j = 0; | |
1588 | for (i = 0; i < vcpu->arch.slb_nr; i++) { | |
1589 | if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) { | |
1590 | vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe; | |
1591 | vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv; | |
1592 | ++j; | |
1593 | } | |
1594 | } | |
1595 | vcpu->arch.slb_max = j; | |
1596 | ||
1597 | return 0; | |
1598 | } | |
1599 | ||
a0840240 AK |
1600 | static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr, |
1601 | bool preserve_top32) | |
a0144e2a | 1602 | { |
8f902b00 | 1603 | struct kvm *kvm = vcpu->kvm; |
a0144e2a PM |
1604 | struct kvmppc_vcore *vc = vcpu->arch.vcore; |
1605 | u64 mask; | |
1606 | ||
1607 | spin_lock(&vc->lock); | |
d682916a AB |
1608 | /* |
1609 | * If ILE (interrupt little-endian) has changed, update the | |
1610 | * MSR_LE bit in the intr_msr for each vcpu in this vcore. | |
1611 | */ | |
1612 | if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) { | |
d682916a AB |
1613 | struct kvm_vcpu *vcpu; |
1614 | int i; | |
1615 | ||
d682916a AB |
1616 | kvm_for_each_vcpu(i, vcpu, kvm) { |
1617 | if (vcpu->arch.vcore != vc) | |
1618 | continue; | |
1619 | if (new_lpcr & LPCR_ILE) | |
1620 | vcpu->arch.intr_msr |= MSR_LE; | |
1621 | else | |
1622 | vcpu->arch.intr_msr &= ~MSR_LE; | |
1623 | } | |
d682916a AB |
1624 | } |
1625 | ||
a0144e2a PM |
1626 | /* |
1627 | * Userspace can only modify DPFD (default prefetch depth), | |
1628 | * ILE (interrupt little-endian) and TC (translation control). | |
8cf4ecc0 | 1629 | * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.). |
a0144e2a PM |
1630 | */ |
1631 | mask = LPCR_DPFD | LPCR_ILE | LPCR_TC; | |
e0622bd9 PM |
1632 | if (cpu_has_feature(CPU_FTR_ARCH_207S)) |
1633 | mask |= LPCR_AIL; | |
1bc3fe81 PM |
1634 | /* |
1635 | * On POWER9, allow userspace to enable large decrementer for the | |
1636 | * guest, whether or not the host has it enabled. | |
1637 | */ | |
1638 | if (cpu_has_feature(CPU_FTR_ARCH_300)) | |
1639 | mask |= LPCR_LD; | |
a0840240 AK |
1640 | |
1641 | /* Broken 32-bit version of LPCR must not clear top bits */ | |
1642 | if (preserve_top32) | |
1643 | mask &= 0xFFFFFFFF; | |
a0144e2a PM |
1644 | vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask); |
1645 | spin_unlock(&vc->lock); | |
1646 | } | |
1647 | ||
3a167bea AK |
1648 | static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id, |
1649 | union kvmppc_one_reg *val) | |
31f3438e | 1650 | { |
a136a8bd PM |
1651 | int r = 0; |
1652 | long int i; | |
31f3438e | 1653 | |
a136a8bd | 1654 | switch (id) { |
a59c1d9e MS |
1655 | case KVM_REG_PPC_DEBUG_INST: |
1656 | *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT); | |
1657 | break; | |
31f3438e | 1658 | case KVM_REG_PPC_HIOR: |
a136a8bd PM |
1659 | *val = get_reg_val(id, 0); |
1660 | break; | |
1661 | case KVM_REG_PPC_DABR: | |
1662 | *val = get_reg_val(id, vcpu->arch.dabr); | |
1663 | break; | |
8563bf52 PM |
1664 | case KVM_REG_PPC_DABRX: |
1665 | *val = get_reg_val(id, vcpu->arch.dabrx); | |
1666 | break; | |
a136a8bd PM |
1667 | case KVM_REG_PPC_DSCR: |
1668 | *val = get_reg_val(id, vcpu->arch.dscr); | |
1669 | break; | |
1670 | case KVM_REG_PPC_PURR: | |
1671 | *val = get_reg_val(id, vcpu->arch.purr); | |
1672 | break; | |
1673 | case KVM_REG_PPC_SPURR: | |
1674 | *val = get_reg_val(id, vcpu->arch.spurr); | |
1675 | break; | |
1676 | case KVM_REG_PPC_AMR: | |
1677 | *val = get_reg_val(id, vcpu->arch.amr); | |
1678 | break; | |
1679 | case KVM_REG_PPC_UAMOR: | |
1680 | *val = get_reg_val(id, vcpu->arch.uamor); | |
1681 | break; | |
b005255e | 1682 | case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS: |
a136a8bd PM |
1683 | i = id - KVM_REG_PPC_MMCR0; |
1684 | *val = get_reg_val(id, vcpu->arch.mmcr[i]); | |
1685 | break; | |
1686 | case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8: | |
1687 | i = id - KVM_REG_PPC_PMC1; | |
1688 | *val = get_reg_val(id, vcpu->arch.pmc[i]); | |
31f3438e | 1689 | break; |
b005255e MN |
1690 | case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2: |
1691 | i = id - KVM_REG_PPC_SPMC1; | |
1692 | *val = get_reg_val(id, vcpu->arch.spmc[i]); | |
1693 | break; | |
14941789 PM |
1694 | case KVM_REG_PPC_SIAR: |
1695 | *val = get_reg_val(id, vcpu->arch.siar); | |
1696 | break; | |
1697 | case KVM_REG_PPC_SDAR: | |
1698 | *val = get_reg_val(id, vcpu->arch.sdar); | |
1699 | break; | |
b005255e MN |
1700 | case KVM_REG_PPC_SIER: |
1701 | *val = get_reg_val(id, vcpu->arch.sier); | |
a8bd19ef | 1702 | break; |
b005255e MN |
1703 | case KVM_REG_PPC_IAMR: |
1704 | *val = get_reg_val(id, vcpu->arch.iamr); | |
1705 | break; | |
b005255e MN |
1706 | case KVM_REG_PPC_PSPB: |
1707 | *val = get_reg_val(id, vcpu->arch.pspb); | |
1708 | break; | |
b005255e | 1709 | case KVM_REG_PPC_DPDES: |
ff42df49 PM |
1710 | /* |
1711 | * On POWER9, where we are emulating msgsndp etc., | |
1712 | * we return 1 bit for each vcpu, which can come from | |
1713 | * either vcore->dpdes or doorbell_request. | |
1714 | * On POWER8, doorbell_request is 0. | |
1715 | */ | |
1716 | *val = get_reg_val(id, vcpu->arch.vcore->dpdes | | |
1717 | vcpu->arch.doorbell_request); | |
b005255e | 1718 | break; |
88b02cf9 PM |
1719 | case KVM_REG_PPC_VTB: |
1720 | *val = get_reg_val(id, vcpu->arch.vcore->vtb); | |
1721 | break; | |
b005255e MN |
1722 | case KVM_REG_PPC_DAWR: |
1723 | *val = get_reg_val(id, vcpu->arch.dawr); | |
1724 | break; | |
1725 | case KVM_REG_PPC_DAWRX: | |
1726 | *val = get_reg_val(id, vcpu->arch.dawrx); | |
1727 | break; | |
1728 | case KVM_REG_PPC_CIABR: | |
1729 | *val = get_reg_val(id, vcpu->arch.ciabr); | |
1730 | break; | |
b005255e MN |
1731 | case KVM_REG_PPC_CSIGR: |
1732 | *val = get_reg_val(id, vcpu->arch.csigr); | |
1733 | break; | |
1734 | case KVM_REG_PPC_TACR: | |
1735 | *val = get_reg_val(id, vcpu->arch.tacr); | |
1736 | break; | |
1737 | case KVM_REG_PPC_TCSCR: | |
1738 | *val = get_reg_val(id, vcpu->arch.tcscr); | |
1739 | break; | |
1740 | case KVM_REG_PPC_PID: | |
1741 | *val = get_reg_val(id, vcpu->arch.pid); | |
1742 | break; | |
1743 | case KVM_REG_PPC_ACOP: | |
1744 | *val = get_reg_val(id, vcpu->arch.acop); | |
1745 | break; | |
1746 | case KVM_REG_PPC_WORT: | |
1747 | *val = get_reg_val(id, vcpu->arch.wort); | |
a8bd19ef | 1748 | break; |
e9cf1e08 PM |
1749 | case KVM_REG_PPC_TIDR: |
1750 | *val = get_reg_val(id, vcpu->arch.tid); | |
1751 | break; | |
1752 | case KVM_REG_PPC_PSSCR: | |
1753 | *val = get_reg_val(id, vcpu->arch.psscr); | |
1754 | break; | |
55b665b0 PM |
1755 | case KVM_REG_PPC_VPA_ADDR: |
1756 | spin_lock(&vcpu->arch.vpa_update_lock); | |
1757 | *val = get_reg_val(id, vcpu->arch.vpa.next_gpa); | |
1758 | spin_unlock(&vcpu->arch.vpa_update_lock); | |
1759 | break; | |
1760 | case KVM_REG_PPC_VPA_SLB: | |
1761 | spin_lock(&vcpu->arch.vpa_update_lock); | |
1762 | val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa; | |
1763 | val->vpaval.length = vcpu->arch.slb_shadow.len; | |
1764 | spin_unlock(&vcpu->arch.vpa_update_lock); | |
1765 | break; | |
1766 | case KVM_REG_PPC_VPA_DTL: | |
1767 | spin_lock(&vcpu->arch.vpa_update_lock); | |
1768 | val->vpaval.addr = vcpu->arch.dtl.next_gpa; | |
1769 | val->vpaval.length = vcpu->arch.dtl.len; | |
1770 | spin_unlock(&vcpu->arch.vpa_update_lock); | |
1771 | break; | |
93b0f4dc PM |
1772 | case KVM_REG_PPC_TB_OFFSET: |
1773 | *val = get_reg_val(id, vcpu->arch.vcore->tb_offset); | |
1774 | break; | |
a0144e2a | 1775 | case KVM_REG_PPC_LPCR: |
a0840240 | 1776 | case KVM_REG_PPC_LPCR_64: |
a0144e2a PM |
1777 | *val = get_reg_val(id, vcpu->arch.vcore->lpcr); |
1778 | break; | |
4b8473c9 PM |
1779 | case KVM_REG_PPC_PPR: |
1780 | *val = get_reg_val(id, vcpu->arch.ppr); | |
1781 | break; | |
a7d80d01 MN |
1782 | #ifdef CONFIG_PPC_TRANSACTIONAL_MEM |
1783 | case KVM_REG_PPC_TFHAR: | |
1784 | *val = get_reg_val(id, vcpu->arch.tfhar); | |
1785 | break; | |
1786 | case KVM_REG_PPC_TFIAR: | |
1787 | *val = get_reg_val(id, vcpu->arch.tfiar); | |
1788 | break; | |
1789 | case KVM_REG_PPC_TEXASR: | |
1790 | *val = get_reg_val(id, vcpu->arch.texasr); | |
1791 | break; | |
1792 | case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31: | |
1793 | i = id - KVM_REG_PPC_TM_GPR0; | |
1794 | *val = get_reg_val(id, vcpu->arch.gpr_tm[i]); | |
1795 | break; | |
1796 | case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63: | |
1797 | { | |
1798 | int j; | |
1799 | i = id - KVM_REG_PPC_TM_VSR0; | |
1800 | if (i < 32) | |
1801 | for (j = 0; j < TS_FPRWIDTH; j++) | |
1802 | val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j]; | |
1803 | else { | |
1804 | if (cpu_has_feature(CPU_FTR_ALTIVEC)) | |
1805 | val->vval = vcpu->arch.vr_tm.vr[i-32]; | |
1806 | else | |
1807 | r = -ENXIO; | |
1808 | } | |
1809 | break; | |
1810 | } | |
1811 | case KVM_REG_PPC_TM_CR: | |
1812 | *val = get_reg_val(id, vcpu->arch.cr_tm); | |
1813 | break; | |
0d808df0 PM |
1814 | case KVM_REG_PPC_TM_XER: |
1815 | *val = get_reg_val(id, vcpu->arch.xer_tm); | |
1816 | break; | |
a7d80d01 MN |
1817 | case KVM_REG_PPC_TM_LR: |
1818 | *val = get_reg_val(id, vcpu->arch.lr_tm); | |
1819 | break; | |
1820 | case KVM_REG_PPC_TM_CTR: | |
1821 | *val = get_reg_val(id, vcpu->arch.ctr_tm); | |
1822 | break; | |
1823 | case KVM_REG_PPC_TM_FPSCR: | |
1824 | *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr); | |
1825 | break; | |
1826 | case KVM_REG_PPC_TM_AMR: | |
1827 | *val = get_reg_val(id, vcpu->arch.amr_tm); | |
1828 | break; | |
1829 | case KVM_REG_PPC_TM_PPR: | |
1830 | *val = get_reg_val(id, vcpu->arch.ppr_tm); | |
1831 | break; | |
1832 | case KVM_REG_PPC_TM_VRSAVE: | |
1833 | *val = get_reg_val(id, vcpu->arch.vrsave_tm); | |
1834 | break; | |
1835 | case KVM_REG_PPC_TM_VSCR: | |
1836 | if (cpu_has_feature(CPU_FTR_ALTIVEC)) | |
1837 | *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]); | |
1838 | else | |
1839 | r = -ENXIO; | |
1840 | break; | |
1841 | case KVM_REG_PPC_TM_DSCR: | |
1842 | *val = get_reg_val(id, vcpu->arch.dscr_tm); | |
1843 | break; | |
1844 | case KVM_REG_PPC_TM_TAR: | |
1845 | *val = get_reg_val(id, vcpu->arch.tar_tm); | |
1846 | break; | |
1847 | #endif | |
388cc6e1 PM |
1848 | case KVM_REG_PPC_ARCH_COMPAT: |
1849 | *val = get_reg_val(id, vcpu->arch.vcore->arch_compat); | |
1850 | break; | |
5855564c PM |
1851 | case KVM_REG_PPC_DEC_EXPIRY: |
1852 | *val = get_reg_val(id, vcpu->arch.dec_expires + | |
1853 | vcpu->arch.vcore->tb_offset); | |
1854 | break; | |
a1f15826 PM |
1855 | case KVM_REG_PPC_ONLINE: |
1856 | *val = get_reg_val(id, vcpu->arch.online); | |
1857 | break; | |
30323418 PM |
1858 | case KVM_REG_PPC_PTCR: |
1859 | *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr); | |
1860 | break; | |
31f3438e | 1861 | default: |
a136a8bd | 1862 | r = -EINVAL; |
31f3438e PM |
1863 | break; |
1864 | } | |
1865 | ||
1866 | return r; | |
1867 | } | |
1868 | ||
3a167bea AK |
1869 | static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id, |
1870 | union kvmppc_one_reg *val) | |
31f3438e | 1871 | { |
a136a8bd PM |
1872 | int r = 0; |
1873 | long int i; | |
55b665b0 | 1874 | unsigned long addr, len; |
31f3438e | 1875 | |
a136a8bd | 1876 | switch (id) { |
31f3438e | 1877 | case KVM_REG_PPC_HIOR: |
31f3438e | 1878 | /* Only allow this to be set to zero */ |
a136a8bd | 1879 | if (set_reg_val(id, *val)) |
31f3438e PM |
1880 | r = -EINVAL; |
1881 | break; | |
a136a8bd PM |
1882 | case KVM_REG_PPC_DABR: |
1883 | vcpu->arch.dabr = set_reg_val(id, *val); | |
1884 | break; | |
8563bf52 PM |
1885 | case KVM_REG_PPC_DABRX: |
1886 | vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP; | |
1887 | break; | |
a136a8bd PM |
1888 | case KVM_REG_PPC_DSCR: |
1889 | vcpu->arch.dscr = set_reg_val(id, *val); | |
1890 | break; | |
1891 | case KVM_REG_PPC_PURR: | |
1892 | vcpu->arch.purr = set_reg_val(id, *val); | |
1893 | break; | |
1894 | case KVM_REG_PPC_SPURR: | |
1895 | vcpu->arch.spurr = set_reg_val(id, *val); | |
1896 | break; | |
1897 | case KVM_REG_PPC_AMR: | |
1898 | vcpu->arch.amr = set_reg_val(id, *val); | |
1899 | break; | |
1900 | case KVM_REG_PPC_UAMOR: | |
1901 | vcpu->arch.uamor = set_reg_val(id, *val); | |
1902 | break; | |
b005255e | 1903 | case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS: |
a136a8bd PM |
1904 | i = id - KVM_REG_PPC_MMCR0; |
1905 | vcpu->arch.mmcr[i] = set_reg_val(id, *val); | |
1906 | break; | |
1907 | case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8: | |
1908 | i = id - KVM_REG_PPC_PMC1; | |
1909 | vcpu->arch.pmc[i] = set_reg_val(id, *val); | |
1910 | break; | |
b005255e MN |
1911 | case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2: |
1912 | i = id - KVM_REG_PPC_SPMC1; | |
1913 | vcpu->arch.spmc[i] = set_reg_val(id, *val); | |
1914 | break; | |
14941789 PM |
1915 | case KVM_REG_PPC_SIAR: |
1916 | vcpu->arch.siar = set_reg_val(id, *val); | |
1917 | break; | |
1918 | case KVM_REG_PPC_SDAR: | |
1919 | vcpu->arch.sdar = set_reg_val(id, *val); | |
1920 | break; | |
b005255e MN |
1921 | case KVM_REG_PPC_SIER: |
1922 | vcpu->arch.sier = set_reg_val(id, *val); | |
a8bd19ef | 1923 | break; |
b005255e MN |
1924 | case KVM_REG_PPC_IAMR: |
1925 | vcpu->arch.iamr = set_reg_val(id, *val); | |
1926 | break; | |
b005255e MN |
1927 | case KVM_REG_PPC_PSPB: |
1928 | vcpu->arch.pspb = set_reg_val(id, *val); | |
1929 | break; | |
b005255e MN |
1930 | case KVM_REG_PPC_DPDES: |
1931 | vcpu->arch.vcore->dpdes = set_reg_val(id, *val); | |
1932 | break; | |
88b02cf9 PM |
1933 | case KVM_REG_PPC_VTB: |
1934 | vcpu->arch.vcore->vtb = set_reg_val(id, *val); | |
1935 | break; | |
b005255e MN |
1936 | case KVM_REG_PPC_DAWR: |
1937 | vcpu->arch.dawr = set_reg_val(id, *val); | |
1938 | break; | |
1939 | case KVM_REG_PPC_DAWRX: | |
1940 | vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP; | |
1941 | break; | |
1942 | case KVM_REG_PPC_CIABR: | |
1943 | vcpu->arch.ciabr = set_reg_val(id, *val); | |
1944 | /* Don't allow setting breakpoints in hypervisor code */ | |
1945 | if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER) | |
1946 | vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */ | |
1947 | break; | |
b005255e MN |
1948 | case KVM_REG_PPC_CSIGR: |
1949 | vcpu->arch.csigr = set_reg_val(id, *val); | |
1950 | break; | |
1951 | case KVM_REG_PPC_TACR: | |
1952 | vcpu->arch.tacr = set_reg_val(id, *val); | |
1953 | break; | |
1954 | case KVM_REG_PPC_TCSCR: | |
1955 | vcpu->arch.tcscr = set_reg_val(id, *val); | |
1956 | break; | |
1957 | case KVM_REG_PPC_PID: | |
1958 | vcpu->arch.pid = set_reg_val(id, *val); | |
1959 | break; | |
1960 | case KVM_REG_PPC_ACOP: | |
1961 | vcpu->arch.acop = set_reg_val(id, *val); | |
1962 | break; | |
1963 | case KVM_REG_PPC_WORT: | |
1964 | vcpu->arch.wort = set_reg_val(id, *val); | |
a8bd19ef | 1965 | break; |
e9cf1e08 PM |
1966 | case KVM_REG_PPC_TIDR: |
1967 | vcpu->arch.tid = set_reg_val(id, *val); | |
1968 | break; | |
1969 | case KVM_REG_PPC_PSSCR: | |
1970 | vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS; | |
1971 | break; | |
55b665b0 PM |
1972 | case KVM_REG_PPC_VPA_ADDR: |
1973 | addr = set_reg_val(id, *val); | |
1974 | r = -EINVAL; | |
1975 | if (!addr && (vcpu->arch.slb_shadow.next_gpa || | |
1976 | vcpu->arch.dtl.next_gpa)) | |
1977 | break; | |
1978 | r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca)); | |
1979 | break; | |
1980 | case KVM_REG_PPC_VPA_SLB: | |
1981 | addr = val->vpaval.addr; | |
1982 | len = val->vpaval.length; | |
1983 | r = -EINVAL; | |
1984 | if (addr && !vcpu->arch.vpa.next_gpa) | |
1985 | break; | |
1986 | r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len); | |
1987 | break; | |
1988 | case KVM_REG_PPC_VPA_DTL: | |
1989 | addr = val->vpaval.addr; | |
1990 | len = val->vpaval.length; | |
1991 | r = -EINVAL; | |
9f8c8c78 PM |
1992 | if (addr && (len < sizeof(struct dtl_entry) || |
1993 | !vcpu->arch.vpa.next_gpa)) | |
55b665b0 PM |
1994 | break; |
1995 | len -= len % sizeof(struct dtl_entry); | |
1996 | r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len); | |
1997 | break; | |
93b0f4dc PM |
1998 | case KVM_REG_PPC_TB_OFFSET: |
1999 | /* round up to multiple of 2^24 */ | |
2000 | vcpu->arch.vcore->tb_offset = | |
2001 | ALIGN(set_reg_val(id, *val), 1UL << 24); | |
2002 | break; | |
a0144e2a | 2003 | case KVM_REG_PPC_LPCR: |
a0840240 AK |
2004 | kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true); |
2005 | break; | |
2006 | case KVM_REG_PPC_LPCR_64: | |
2007 | kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false); | |
a0144e2a | 2008 | break; |
4b8473c9 PM |
2009 | case KVM_REG_PPC_PPR: |
2010 | vcpu->arch.ppr = set_reg_val(id, *val); | |
2011 | break; | |
a7d80d01 MN |
2012 | #ifdef CONFIG_PPC_TRANSACTIONAL_MEM |
2013 | case KVM_REG_PPC_TFHAR: | |
2014 | vcpu->arch.tfhar = set_reg_val(id, *val); | |
2015 | break; | |
2016 | case KVM_REG_PPC_TFIAR: | |
2017 | vcpu->arch.tfiar = set_reg_val(id, *val); | |
2018 | break; | |
2019 | case KVM_REG_PPC_TEXASR: | |
2020 | vcpu->arch.texasr = set_reg_val(id, *val); | |
2021 | break; | |
2022 | case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31: | |
2023 | i = id - KVM_REG_PPC_TM_GPR0; | |
2024 | vcpu->arch.gpr_tm[i] = set_reg_val(id, *val); | |
2025 | break; | |
2026 | case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63: | |
2027 | { | |
2028 | int j; | |
2029 | i = id - KVM_REG_PPC_TM_VSR0; | |
2030 | if (i < 32) | |
2031 | for (j = 0; j < TS_FPRWIDTH; j++) | |
2032 | vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j]; | |
2033 | else | |
2034 | if (cpu_has_feature(CPU_FTR_ALTIVEC)) | |
2035 | vcpu->arch.vr_tm.vr[i-32] = val->vval; | |
2036 | else | |
2037 | r = -ENXIO; | |
2038 | break; | |
2039 | } | |
2040 | case KVM_REG_PPC_TM_CR: | |
2041 | vcpu->arch.cr_tm = set_reg_val(id, *val); | |
2042 | break; | |
0d808df0 PM |
2043 | case KVM_REG_PPC_TM_XER: |
2044 | vcpu->arch.xer_tm = set_reg_val(id, *val); | |
2045 | break; | |
a7d80d01 MN |
2046 | case KVM_REG_PPC_TM_LR: |
2047 | vcpu->arch.lr_tm = set_reg_val(id, *val); | |
2048 | break; | |
2049 | case KVM_REG_PPC_TM_CTR: | |
2050 | vcpu->arch.ctr_tm = set_reg_val(id, *val); | |
2051 | break; | |
2052 | case KVM_REG_PPC_TM_FPSCR: | |
2053 | vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val); | |
2054 | break; | |
2055 | case KVM_REG_PPC_TM_AMR: | |
2056 | vcpu->arch.amr_tm = set_reg_val(id, *val); | |
2057 | break; | |
2058 | case KVM_REG_PPC_TM_PPR: | |
2059 | vcpu->arch.ppr_tm = set_reg_val(id, *val); | |
2060 | break; | |
2061 | case KVM_REG_PPC_TM_VRSAVE: | |
2062 | vcpu->arch.vrsave_tm = set_reg_val(id, *val); | |
2063 | break; | |
2064 | case KVM_REG_PPC_TM_VSCR: | |
2065 | if (cpu_has_feature(CPU_FTR_ALTIVEC)) | |
2066 | vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val); | |
2067 | else | |
2068 | r = - ENXIO; | |
2069 | break; | |
2070 | case KVM_REG_PPC_TM_DSCR: | |
2071 | vcpu->arch.dscr_tm = set_reg_val(id, *val); | |
2072 | break; | |
2073 | case KVM_REG_PPC_TM_TAR: | |
2074 | vcpu->arch.tar_tm = set_reg_val(id, *val); | |
2075 | break; | |
2076 | #endif | |
388cc6e1 PM |
2077 | case KVM_REG_PPC_ARCH_COMPAT: |
2078 | r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val)); | |
2079 | break; | |
5855564c PM |
2080 | case KVM_REG_PPC_DEC_EXPIRY: |
2081 | vcpu->arch.dec_expires = set_reg_val(id, *val) - | |
2082 | vcpu->arch.vcore->tb_offset; | |
2083 | break; | |
a1f15826 | 2084 | case KVM_REG_PPC_ONLINE: |
7aa15842 PM |
2085 | i = set_reg_val(id, *val); |
2086 | if (i && !vcpu->arch.online) | |
2087 | atomic_inc(&vcpu->arch.vcore->online_count); | |
2088 | else if (!i && vcpu->arch.online) | |
2089 | atomic_dec(&vcpu->arch.vcore->online_count); | |
2090 | vcpu->arch.online = i; | |
a1f15826 | 2091 | break; |
30323418 PM |
2092 | case KVM_REG_PPC_PTCR: |
2093 | vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val); | |
2094 | break; | |
31f3438e | 2095 | default: |
a136a8bd | 2096 | r = -EINVAL; |
31f3438e PM |
2097 | break; |
2098 | } | |
2099 | ||
2100 | return r; | |
2101 | } | |
2102 | ||
45c940ba PM |
2103 | /* |
2104 | * On POWER9, threads are independent and can be in different partitions. | |
2105 | * Therefore we consider each thread to be a subcore. | |
2106 | * There is a restriction that all threads have to be in the same | |
2107 | * MMU mode (radix or HPT), unfortunately, but since we only support | |
2108 | * HPT guests on a HPT host so far, that isn't an impediment yet. | |
2109 | */ | |
516f7898 | 2110 | static int threads_per_vcore(struct kvm *kvm) |
45c940ba | 2111 | { |
516f7898 | 2112 | if (kvm->arch.threads_indep) |
45c940ba PM |
2113 | return 1; |
2114 | return threads_per_subcore; | |
2115 | } | |
2116 | ||
1e175d2e | 2117 | static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id) |
de9bdd1a SS |
2118 | { |
2119 | struct kvmppc_vcore *vcore; | |
2120 | ||
2121 | vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL); | |
2122 | ||
2123 | if (vcore == NULL) | |
2124 | return NULL; | |
2125 | ||
de9bdd1a | 2126 | spin_lock_init(&vcore->lock); |
2711e248 | 2127 | spin_lock_init(&vcore->stoltb_lock); |
da4ad88c | 2128 | rcuwait_init(&vcore->wait); |
de9bdd1a SS |
2129 | vcore->preempt_tb = TB_NIL; |
2130 | vcore->lpcr = kvm->arch.lpcr; | |
1e175d2e | 2131 | vcore->first_vcpuid = id; |
de9bdd1a | 2132 | vcore->kvm = kvm; |
ec257165 | 2133 | INIT_LIST_HEAD(&vcore->preempt_list); |
de9bdd1a SS |
2134 | |
2135 | return vcore; | |
2136 | } | |
2137 | ||
b6c295df PM |
2138 | #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING |
2139 | static struct debugfs_timings_element { | |
2140 | const char *name; | |
2141 | size_t offset; | |
2142 | } timings[] = { | |
2143 | {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)}, | |
2144 | {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)}, | |
2145 | {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)}, | |
2146 | {"guest", offsetof(struct kvm_vcpu, arch.guest_time)}, | |
2147 | {"cede", offsetof(struct kvm_vcpu, arch.cede_time)}, | |
2148 | }; | |
2149 | ||
4bb817ed | 2150 | #define N_TIMINGS (ARRAY_SIZE(timings)) |
b6c295df PM |
2151 | |
2152 | struct debugfs_timings_state { | |
2153 | struct kvm_vcpu *vcpu; | |
2154 | unsigned int buflen; | |
2155 | char buf[N_TIMINGS * 100]; | |
2156 | }; | |
2157 | ||
2158 | static int debugfs_timings_open(struct inode *inode, struct file *file) | |
2159 | { | |
2160 | struct kvm_vcpu *vcpu = inode->i_private; | |
2161 | struct debugfs_timings_state *p; | |
2162 | ||
2163 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2164 | if (!p) | |
2165 | return -ENOMEM; | |
2166 | ||
2167 | kvm_get_kvm(vcpu->kvm); | |
2168 | p->vcpu = vcpu; | |
2169 | file->private_data = p; | |
2170 | ||
2171 | return nonseekable_open(inode, file); | |
2172 | } | |
2173 | ||
2174 | static int debugfs_timings_release(struct inode *inode, struct file *file) | |
2175 | { | |
2176 | struct debugfs_timings_state *p = file->private_data; | |
2177 | ||
2178 | kvm_put_kvm(p->vcpu->kvm); | |
2179 | kfree(p); | |
2180 | return 0; | |
2181 | } | |
2182 | ||
2183 | static ssize_t debugfs_timings_read(struct file *file, char __user *buf, | |
2184 | size_t len, loff_t *ppos) | |
2185 | { | |
2186 | struct debugfs_timings_state *p = file->private_data; | |
2187 | struct kvm_vcpu *vcpu = p->vcpu; | |
2188 | char *s, *buf_end; | |
2189 | struct kvmhv_tb_accumulator tb; | |
2190 | u64 count; | |
2191 | loff_t pos; | |
2192 | ssize_t n; | |
2193 | int i, loops; | |
2194 | bool ok; | |
2195 | ||
2196 | if (!p->buflen) { | |
2197 | s = p->buf; | |
2198 | buf_end = s + sizeof(p->buf); | |
2199 | for (i = 0; i < N_TIMINGS; ++i) { | |
2200 | struct kvmhv_tb_accumulator *acc; | |
2201 | ||
2202 | acc = (struct kvmhv_tb_accumulator *) | |
2203 | ((unsigned long)vcpu + timings[i].offset); | |
2204 | ok = false; | |
2205 | for (loops = 0; loops < 1000; ++loops) { | |
2206 | count = acc->seqcount; | |
2207 | if (!(count & 1)) { | |
2208 | smp_rmb(); | |
2209 | tb = *acc; | |
2210 | smp_rmb(); | |
2211 | if (count == acc->seqcount) { | |
2212 | ok = true; | |
2213 | break; | |
2214 | } | |
2215 | } | |
2216 | udelay(1); | |
2217 | } | |
2218 | if (!ok) | |
2219 | snprintf(s, buf_end - s, "%s: stuck\n", | |
2220 | timings[i].name); | |
2221 | else | |
2222 | snprintf(s, buf_end - s, | |
2223 | "%s: %llu %llu %llu %llu\n", | |
2224 | timings[i].name, count / 2, | |
2225 | tb_to_ns(tb.tb_total), | |
2226 | tb_to_ns(tb.tb_min), | |
2227 | tb_to_ns(tb.tb_max)); | |
2228 | s += strlen(s); | |
2229 | } | |
2230 | p->buflen = s - p->buf; | |
2231 | } | |
2232 | ||
2233 | pos = *ppos; | |
2234 | if (pos >= p->buflen) | |
2235 | return 0; | |
2236 | if (len > p->buflen - pos) | |
2237 | len = p->buflen - pos; | |
2238 | n = copy_to_user(buf, p->buf + pos, len); | |
2239 | if (n) { | |
2240 | if (n == len) | |
2241 | return -EFAULT; | |
2242 | len -= n; | |
2243 | } | |
2244 | *ppos = pos + len; | |
2245 | return len; | |
2246 | } | |
2247 | ||
2248 | static ssize_t debugfs_timings_write(struct file *file, const char __user *buf, | |
2249 | size_t len, loff_t *ppos) | |
2250 | { | |
2251 | return -EACCES; | |
2252 | } | |
2253 | ||
2254 | static const struct file_operations debugfs_timings_ops = { | |
2255 | .owner = THIS_MODULE, | |
2256 | .open = debugfs_timings_open, | |
2257 | .release = debugfs_timings_release, | |
2258 | .read = debugfs_timings_read, | |
2259 | .write = debugfs_timings_write, | |
2260 | .llseek = generic_file_llseek, | |
2261 | }; | |
2262 | ||
2263 | /* Create a debugfs directory for the vcpu */ | |
2264 | static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id) | |
2265 | { | |
2266 | char buf[16]; | |
2267 | struct kvm *kvm = vcpu->kvm; | |
2268 | ||
2269 | snprintf(buf, sizeof(buf), "vcpu%u", id); | |
b6c295df | 2270 | vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir); |
c4fd527f GKH |
2271 | debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu, |
2272 | &debugfs_timings_ops); | |
b6c295df PM |
2273 | } |
2274 | ||
2275 | #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */ | |
2276 | static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id) | |
2277 | { | |
2278 | } | |
2279 | #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */ | |
2280 | ||
ff030fdf | 2281 | static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu) |
de56a948 | 2282 | { |
3c313524 | 2283 | int err; |
371fefd6 PM |
2284 | int core; |
2285 | struct kvmppc_vcore *vcore; | |
ff030fdf SC |
2286 | struct kvm *kvm; |
2287 | unsigned int id; | |
de56a948 | 2288 | |
ff030fdf SC |
2289 | kvm = vcpu->kvm; |
2290 | id = vcpu->vcpu_id; | |
de56a948 PM |
2291 | |
2292 | vcpu->arch.shared = &vcpu->arch.shregs; | |
5deb8e7a AG |
2293 | #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE |
2294 | /* | |
2295 | * The shared struct is never shared on HV, | |
2296 | * so we can always use host endianness | |
2297 | */ | |
2298 | #ifdef __BIG_ENDIAN__ | |
2299 | vcpu->arch.shared_big_endian = true; | |
2300 | #else | |
2301 | vcpu->arch.shared_big_endian = false; | |
2302 | #endif | |
2303 | #endif | |
de56a948 PM |
2304 | vcpu->arch.mmcr[0] = MMCR0_FC; |
2305 | vcpu->arch.ctrl = CTRL_RUNLATCH; | |
2306 | /* default to host PVR, since we can't spoof it */ | |
3a167bea | 2307 | kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR)); |
2e25aa5f | 2308 | spin_lock_init(&vcpu->arch.vpa_update_lock); |
c7b67670 PM |
2309 | spin_lock_init(&vcpu->arch.tbacct_lock); |
2310 | vcpu->arch.busy_preempt = TB_NIL; | |
d682916a | 2311 | vcpu->arch.intr_msr = MSR_SF | MSR_ME; |
de56a948 | 2312 | |
769377f7 PM |
2313 | /* |
2314 | * Set the default HFSCR for the guest from the host value. | |
2315 | * This value is only used on POWER9. | |
57900694 | 2316 | * On POWER9, we want to virtualize the doorbell facility, so we |
f3c99f97 PM |
2317 | * don't set the HFSCR_MSGP bit, and that causes those instructions |
2318 | * to trap and then we emulate them. | |
769377f7 | 2319 | */ |
f3c99f97 PM |
2320 | vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB | |
2321 | HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP; | |
2322 | if (cpu_has_feature(CPU_FTR_HVMODE)) { | |
2323 | vcpu->arch.hfscr &= mfspr(SPRN_HFSCR); | |
2324 | if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) | |
2325 | vcpu->arch.hfscr |= HFSCR_TM; | |
2326 | } | |
2327 | if (cpu_has_feature(CPU_FTR_TM_COMP)) | |
4bb3c7a0 | 2328 | vcpu->arch.hfscr |= HFSCR_TM; |
769377f7 | 2329 | |
de56a948 PM |
2330 | kvmppc_mmu_book3s_hv_init(vcpu); |
2331 | ||
8455d79e | 2332 | vcpu->arch.state = KVMPPC_VCPU_NOTREADY; |
371fefd6 PM |
2333 | |
2334 | init_waitqueue_head(&vcpu->arch.cpu_run); | |
2335 | ||
2336 | mutex_lock(&kvm->lock); | |
3c313524 PM |
2337 | vcore = NULL; |
2338 | err = -EINVAL; | |
1e175d2e | 2339 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { |
b5c6f760 PM |
2340 | if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) { |
2341 | pr_devel("KVM: VCPU ID too high\n"); | |
2342 | core = KVM_MAX_VCORES; | |
2343 | } else { | |
2344 | BUG_ON(kvm->arch.smt_mode != 1); | |
2345 | core = kvmppc_pack_vcpu_id(kvm, id); | |
2346 | } | |
1e175d2e SB |
2347 | } else { |
2348 | core = id / kvm->arch.smt_mode; | |
2349 | } | |
3c313524 PM |
2350 | if (core < KVM_MAX_VCORES) { |
2351 | vcore = kvm->arch.vcores[core]; | |
1e175d2e SB |
2352 | if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) { |
2353 | pr_devel("KVM: collision on id %u", id); | |
2354 | vcore = NULL; | |
2355 | } else if (!vcore) { | |
0d4ee88d PM |
2356 | /* |
2357 | * Take mmu_setup_lock for mutual exclusion | |
2358 | * with kvmppc_update_lpcr(). | |
2359 | */ | |
3c313524 | 2360 | err = -ENOMEM; |
1e175d2e SB |
2361 | vcore = kvmppc_vcore_create(kvm, |
2362 | id & ~(kvm->arch.smt_mode - 1)); | |
0d4ee88d | 2363 | mutex_lock(&kvm->arch.mmu_setup_lock); |
3c313524 PM |
2364 | kvm->arch.vcores[core] = vcore; |
2365 | kvm->arch.online_vcores++; | |
0d4ee88d | 2366 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
3c313524 | 2367 | } |
371fefd6 PM |
2368 | } |
2369 | mutex_unlock(&kvm->lock); | |
2370 | ||
2371 | if (!vcore) | |
ff030fdf | 2372 | return err; |
371fefd6 PM |
2373 | |
2374 | spin_lock(&vcore->lock); | |
2375 | ++vcore->num_threads; | |
371fefd6 PM |
2376 | spin_unlock(&vcore->lock); |
2377 | vcpu->arch.vcore = vcore; | |
e0b7ec05 | 2378 | vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid; |
ec257165 | 2379 | vcpu->arch.thread_cpu = -1; |
a29ebeaf | 2380 | vcpu->arch.prev_cpu = -1; |
371fefd6 | 2381 | |
af8f38b3 AG |
2382 | vcpu->arch.cpu_type = KVM_CPU_3S_64; |
2383 | kvmppc_sanity_check(vcpu); | |
2384 | ||
b6c295df PM |
2385 | debugfs_vcpu_init(vcpu, id); |
2386 | ||
c50bfbdc | 2387 | return 0; |
de56a948 PM |
2388 | } |
2389 | ||
3c313524 PM |
2390 | static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode, |
2391 | unsigned long flags) | |
2392 | { | |
2393 | int err; | |
57900694 | 2394 | int esmt = 0; |
3c313524 PM |
2395 | |
2396 | if (flags) | |
2397 | return -EINVAL; | |
2398 | if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode)) | |
2399 | return -EINVAL; | |
2400 | if (!cpu_has_feature(CPU_FTR_ARCH_300)) { | |
2401 | /* | |
2402 | * On POWER8 (or POWER7), the threading mode is "strict", | |
2403 | * so we pack smt_mode vcpus per vcore. | |
2404 | */ | |
2405 | if (smt_mode > threads_per_subcore) | |
2406 | return -EINVAL; | |
2407 | } else { | |
2408 | /* | |
2409 | * On POWER9, the threading mode is "loose", | |
2410 | * so each vcpu gets its own vcore. | |
2411 | */ | |
57900694 | 2412 | esmt = smt_mode; |
3c313524 PM |
2413 | smt_mode = 1; |
2414 | } | |
2415 | mutex_lock(&kvm->lock); | |
2416 | err = -EBUSY; | |
2417 | if (!kvm->arch.online_vcores) { | |
2418 | kvm->arch.smt_mode = smt_mode; | |
57900694 | 2419 | kvm->arch.emul_smt_mode = esmt; |
3c313524 PM |
2420 | err = 0; |
2421 | } | |
2422 | mutex_unlock(&kvm->lock); | |
2423 | ||
2424 | return err; | |
2425 | } | |
2426 | ||
c35635ef PM |
2427 | static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa) |
2428 | { | |
2429 | if (vpa->pinned_addr) | |
2430 | kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa, | |
2431 | vpa->dirty); | |
2432 | } | |
2433 | ||
3a167bea | 2434 | static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu) |
de56a948 | 2435 | { |
2e25aa5f | 2436 | spin_lock(&vcpu->arch.vpa_update_lock); |
c35635ef PM |
2437 | unpin_vpa(vcpu->kvm, &vcpu->arch.dtl); |
2438 | unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow); | |
2439 | unpin_vpa(vcpu->kvm, &vcpu->arch.vpa); | |
2e25aa5f | 2440 | spin_unlock(&vcpu->arch.vpa_update_lock); |
de56a948 PM |
2441 | } |
2442 | ||
3a167bea AK |
2443 | static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu) |
2444 | { | |
2445 | /* Indicate we want to get back into the guest */ | |
2446 | return 1; | |
2447 | } | |
2448 | ||
19ccb76a | 2449 | static void kvmppc_set_timer(struct kvm_vcpu *vcpu) |
371fefd6 | 2450 | { |
19ccb76a | 2451 | unsigned long dec_nsec, now; |
371fefd6 | 2452 | |
19ccb76a PM |
2453 | now = get_tb(); |
2454 | if (now > vcpu->arch.dec_expires) { | |
2455 | /* decrementer has already gone negative */ | |
2456 | kvmppc_core_queue_dec(vcpu); | |
7e28e60e | 2457 | kvmppc_core_prepare_to_enter(vcpu); |
19ccb76a | 2458 | return; |
371fefd6 | 2459 | } |
c43befca | 2460 | dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now); |
8b0e1953 | 2461 | hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL); |
19ccb76a | 2462 | vcpu->arch.timer_running = 1; |
371fefd6 PM |
2463 | } |
2464 | ||
8b24e69f | 2465 | extern int __kvmppc_vcore_entry(void); |
de56a948 | 2466 | |
371fefd6 PM |
2467 | static void kvmppc_remove_runnable(struct kvmppc_vcore *vc, |
2468 | struct kvm_vcpu *vcpu) | |
de56a948 | 2469 | { |
c7b67670 PM |
2470 | u64 now; |
2471 | ||
371fefd6 PM |
2472 | if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) |
2473 | return; | |
bf3d32e1 | 2474 | spin_lock_irq(&vcpu->arch.tbacct_lock); |
c7b67670 PM |
2475 | now = mftb(); |
2476 | vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) - | |
2477 | vcpu->arch.stolen_logged; | |
2478 | vcpu->arch.busy_preempt = now; | |
2479 | vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; | |
bf3d32e1 | 2480 | spin_unlock_irq(&vcpu->arch.tbacct_lock); |
371fefd6 | 2481 | --vc->n_runnable; |
7b5f8272 | 2482 | WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL); |
371fefd6 PM |
2483 | } |
2484 | ||
f0888f70 PM |
2485 | static int kvmppc_grab_hwthread(int cpu) |
2486 | { | |
2487 | struct paca_struct *tpaca; | |
b754c739 | 2488 | long timeout = 10000; |
f0888f70 | 2489 | |
d2e60075 | 2490 | tpaca = paca_ptrs[cpu]; |
f0888f70 PM |
2491 | |
2492 | /* Ensure the thread won't go into the kernel if it wakes */ | |
7b444c67 | 2493 | tpaca->kvm_hstate.kvm_vcpu = NULL; |
b4deba5c | 2494 | tpaca->kvm_hstate.kvm_vcore = NULL; |
5d5b99cd PM |
2495 | tpaca->kvm_hstate.napping = 0; |
2496 | smp_wmb(); | |
2497 | tpaca->kvm_hstate.hwthread_req = 1; | |
f0888f70 PM |
2498 | |
2499 | /* | |
2500 | * If the thread is already executing in the kernel (e.g. handling | |
2501 | * a stray interrupt), wait for it to get back to nap mode. | |
2502 | * The smp_mb() is to ensure that our setting of hwthread_req | |
2503 | * is visible before we look at hwthread_state, so if this | |
2504 | * races with the code at system_reset_pSeries and the thread | |
2505 | * misses our setting of hwthread_req, we are sure to see its | |
2506 | * setting of hwthread_state, and vice versa. | |
2507 | */ | |
2508 | smp_mb(); | |
2509 | while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) { | |
2510 | if (--timeout <= 0) { | |
2511 | pr_err("KVM: couldn't grab cpu %d\n", cpu); | |
2512 | return -EBUSY; | |
2513 | } | |
2514 | udelay(1); | |
2515 | } | |
2516 | return 0; | |
2517 | } | |
2518 | ||
2519 | static void kvmppc_release_hwthread(int cpu) | |
2520 | { | |
2521 | struct paca_struct *tpaca; | |
2522 | ||
d2e60075 | 2523 | tpaca = paca_ptrs[cpu]; |
31a4d448 | 2524 | tpaca->kvm_hstate.hwthread_req = 0; |
f0888f70 | 2525 | tpaca->kvm_hstate.kvm_vcpu = NULL; |
b4deba5c PM |
2526 | tpaca->kvm_hstate.kvm_vcore = NULL; |
2527 | tpaca->kvm_hstate.kvm_split_mode = NULL; | |
f0888f70 PM |
2528 | } |
2529 | ||
a29ebeaf PM |
2530 | static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu) |
2531 | { | |
9d0b048d SJS |
2532 | struct kvm_nested_guest *nested = vcpu->arch.nested; |
2533 | cpumask_t *cpu_in_guest; | |
a29ebeaf PM |
2534 | int i; |
2535 | ||
2536 | cpu = cpu_first_thread_sibling(cpu); | |
9d0b048d SJS |
2537 | if (nested) { |
2538 | cpumask_set_cpu(cpu, &nested->need_tlb_flush); | |
2539 | cpu_in_guest = &nested->cpu_in_guest; | |
2540 | } else { | |
2541 | cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush); | |
2542 | cpu_in_guest = &kvm->arch.cpu_in_guest; | |
2543 | } | |
a29ebeaf PM |
2544 | /* |
2545 | * Make sure setting of bit in need_tlb_flush precedes | |
2546 | * testing of cpu_in_guest bits. The matching barrier on | |
2547 | * the other side is the first smp_mb() in kvmppc_run_core(). | |
2548 | */ | |
2549 | smp_mb(); | |
2550 | for (i = 0; i < threads_per_core; ++i) | |
9d0b048d | 2551 | if (cpumask_test_cpu(cpu + i, cpu_in_guest)) |
a29ebeaf PM |
2552 | smp_call_function_single(cpu + i, do_nothing, NULL, 1); |
2553 | } | |
2554 | ||
8b24e69f PM |
2555 | static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu) |
2556 | { | |
9d0b048d | 2557 | struct kvm_nested_guest *nested = vcpu->arch.nested; |
8b24e69f | 2558 | struct kvm *kvm = vcpu->kvm; |
9d0b048d SJS |
2559 | int prev_cpu; |
2560 | ||
2561 | if (!cpu_has_feature(CPU_FTR_HVMODE)) | |
2562 | return; | |
2563 | ||
2564 | if (nested) | |
2565 | prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id]; | |
2566 | else | |
2567 | prev_cpu = vcpu->arch.prev_cpu; | |
8b24e69f PM |
2568 | |
2569 | /* | |
2570 | * With radix, the guest can do TLB invalidations itself, | |
2571 | * and it could choose to use the local form (tlbiel) if | |
2572 | * it is invalidating a translation that has only ever been | |
2573 | * used on one vcpu. However, that doesn't mean it has | |
2574 | * only ever been used on one physical cpu, since vcpus | |
2575 | * can move around between pcpus. To cope with this, when | |
2576 | * a vcpu moves from one pcpu to another, we need to tell | |
2577 | * any vcpus running on the same core as this vcpu previously | |
2578 | * ran to flush the TLB. The TLB is shared between threads, | |
2579 | * so we use a single bit in .need_tlb_flush for all 4 threads. | |
2580 | */ | |
9d0b048d SJS |
2581 | if (prev_cpu != pcpu) { |
2582 | if (prev_cpu >= 0 && | |
2583 | cpu_first_thread_sibling(prev_cpu) != | |
8b24e69f | 2584 | cpu_first_thread_sibling(pcpu)) |
9d0b048d SJS |
2585 | radix_flush_cpu(kvm, prev_cpu, vcpu); |
2586 | if (nested) | |
2587 | nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu; | |
2588 | else | |
2589 | vcpu->arch.prev_cpu = pcpu; | |
2590 | } | |
2591 | } | |
2592 | ||
b4deba5c | 2593 | static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc) |
371fefd6 PM |
2594 | { |
2595 | int cpu; | |
2596 | struct paca_struct *tpaca; | |
a29ebeaf | 2597 | struct kvm *kvm = vc->kvm; |
371fefd6 | 2598 | |
b4deba5c PM |
2599 | cpu = vc->pcpu; |
2600 | if (vcpu) { | |
2601 | if (vcpu->arch.timer_running) { | |
2602 | hrtimer_try_to_cancel(&vcpu->arch.dec_timer); | |
2603 | vcpu->arch.timer_running = 0; | |
2604 | } | |
2605 | cpu += vcpu->arch.ptid; | |
898b25b2 | 2606 | vcpu->cpu = vc->pcpu; |
b4deba5c | 2607 | vcpu->arch.thread_cpu = cpu; |
a29ebeaf | 2608 | cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest); |
19ccb76a | 2609 | } |
d2e60075 | 2610 | tpaca = paca_ptrs[cpu]; |
5d5b99cd | 2611 | tpaca->kvm_hstate.kvm_vcpu = vcpu; |
898b25b2 | 2612 | tpaca->kvm_hstate.ptid = cpu - vc->pcpu; |
4bb3c7a0 | 2613 | tpaca->kvm_hstate.fake_suspend = 0; |
ec257165 | 2614 | /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */ |
371fefd6 | 2615 | smp_wmb(); |
898b25b2 | 2616 | tpaca->kvm_hstate.kvm_vcore = vc; |
5d5b99cd | 2617 | if (cpu != smp_processor_id()) |
66feed61 | 2618 | kvmppc_ipi_thread(cpu); |
371fefd6 | 2619 | } |
de56a948 | 2620 | |
516f7898 | 2621 | static void kvmppc_wait_for_nap(int n_threads) |
371fefd6 | 2622 | { |
5d5b99cd PM |
2623 | int cpu = smp_processor_id(); |
2624 | int i, loops; | |
371fefd6 | 2625 | |
45c940ba PM |
2626 | if (n_threads <= 1) |
2627 | return; | |
5d5b99cd PM |
2628 | for (loops = 0; loops < 1000000; ++loops) { |
2629 | /* | |
2630 | * Check if all threads are finished. | |
b4deba5c | 2631 | * We set the vcore pointer when starting a thread |
5d5b99cd | 2632 | * and the thread clears it when finished, so we look |
b4deba5c | 2633 | * for any threads that still have a non-NULL vcore ptr. |
5d5b99cd | 2634 | */ |
45c940ba | 2635 | for (i = 1; i < n_threads; ++i) |
d2e60075 | 2636 | if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore) |
5d5b99cd | 2637 | break; |
45c940ba | 2638 | if (i == n_threads) { |
5d5b99cd PM |
2639 | HMT_medium(); |
2640 | return; | |
371fefd6 | 2641 | } |
5d5b99cd | 2642 | HMT_low(); |
371fefd6 PM |
2643 | } |
2644 | HMT_medium(); | |
45c940ba | 2645 | for (i = 1; i < n_threads; ++i) |
d2e60075 | 2646 | if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore) |
5d5b99cd | 2647 | pr_err("KVM: CPU %d seems to be stuck\n", cpu + i); |
371fefd6 PM |
2648 | } |
2649 | ||
2650 | /* | |
2651 | * Check that we are on thread 0 and that any other threads in | |
7b444c67 PM |
2652 | * this core are off-line. Then grab the threads so they can't |
2653 | * enter the kernel. | |
371fefd6 PM |
2654 | */ |
2655 | static int on_primary_thread(void) | |
2656 | { | |
2657 | int cpu = smp_processor_id(); | |
3102f784 | 2658 | int thr; |
371fefd6 | 2659 | |
3102f784 ME |
2660 | /* Are we on a primary subcore? */ |
2661 | if (cpu_thread_in_subcore(cpu)) | |
371fefd6 | 2662 | return 0; |
3102f784 ME |
2663 | |
2664 | thr = 0; | |
2665 | while (++thr < threads_per_subcore) | |
371fefd6 PM |
2666 | if (cpu_online(cpu + thr)) |
2667 | return 0; | |
7b444c67 PM |
2668 | |
2669 | /* Grab all hw threads so they can't go into the kernel */ | |
3102f784 | 2670 | for (thr = 1; thr < threads_per_subcore; ++thr) { |
7b444c67 PM |
2671 | if (kvmppc_grab_hwthread(cpu + thr)) { |
2672 | /* Couldn't grab one; let the others go */ | |
2673 | do { | |
2674 | kvmppc_release_hwthread(cpu + thr); | |
2675 | } while (--thr > 0); | |
2676 | return 0; | |
2677 | } | |
2678 | } | |
371fefd6 PM |
2679 | return 1; |
2680 | } | |
2681 | ||
ec257165 PM |
2682 | /* |
2683 | * A list of virtual cores for each physical CPU. | |
2684 | * These are vcores that could run but their runner VCPU tasks are | |
2685 | * (or may be) preempted. | |
2686 | */ | |
2687 | struct preempted_vcore_list { | |
2688 | struct list_head list; | |
2689 | spinlock_t lock; | |
2690 | }; | |
2691 | ||
2692 | static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores); | |
2693 | ||
2694 | static void init_vcore_lists(void) | |
2695 | { | |
2696 | int cpu; | |
2697 | ||
2698 | for_each_possible_cpu(cpu) { | |
2699 | struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu); | |
2700 | spin_lock_init(&lp->lock); | |
2701 | INIT_LIST_HEAD(&lp->list); | |
2702 | } | |
2703 | } | |
2704 | ||
2705 | static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc) | |
2706 | { | |
2707 | struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores); | |
2708 | ||
2709 | vc->vcore_state = VCORE_PREEMPT; | |
2710 | vc->pcpu = smp_processor_id(); | |
516f7898 | 2711 | if (vc->num_threads < threads_per_vcore(vc->kvm)) { |
ec257165 PM |
2712 | spin_lock(&lp->lock); |
2713 | list_add_tail(&vc->preempt_list, &lp->list); | |
2714 | spin_unlock(&lp->lock); | |
2715 | } | |
2716 | ||
2717 | /* Start accumulating stolen time */ | |
2718 | kvmppc_core_start_stolen(vc); | |
2719 | } | |
2720 | ||
2721 | static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc) | |
2722 | { | |
402813fe | 2723 | struct preempted_vcore_list *lp; |
ec257165 PM |
2724 | |
2725 | kvmppc_core_end_stolen(vc); | |
2726 | if (!list_empty(&vc->preempt_list)) { | |
402813fe | 2727 | lp = &per_cpu(preempted_vcores, vc->pcpu); |
ec257165 PM |
2728 | spin_lock(&lp->lock); |
2729 | list_del_init(&vc->preempt_list); | |
2730 | spin_unlock(&lp->lock); | |
2731 | } | |
2732 | vc->vcore_state = VCORE_INACTIVE; | |
2733 | } | |
2734 | ||
b4deba5c PM |
2735 | /* |
2736 | * This stores information about the virtual cores currently | |
2737 | * assigned to a physical core. | |
2738 | */ | |
ec257165 | 2739 | struct core_info { |
b4deba5c PM |
2740 | int n_subcores; |
2741 | int max_subcore_threads; | |
ec257165 | 2742 | int total_threads; |
b4deba5c | 2743 | int subcore_threads[MAX_SUBCORES]; |
898b25b2 | 2744 | struct kvmppc_vcore *vc[MAX_SUBCORES]; |
ec257165 PM |
2745 | }; |
2746 | ||
b4deba5c PM |
2747 | /* |
2748 | * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7 | |
516f7898 | 2749 | * respectively in 2-way micro-threading (split-core) mode on POWER8. |
b4deba5c PM |
2750 | */ |
2751 | static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 }; | |
2752 | ||
ec257165 PM |
2753 | static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc) |
2754 | { | |
2755 | memset(cip, 0, sizeof(*cip)); | |
b4deba5c PM |
2756 | cip->n_subcores = 1; |
2757 | cip->max_subcore_threads = vc->num_threads; | |
ec257165 | 2758 | cip->total_threads = vc->num_threads; |
b4deba5c | 2759 | cip->subcore_threads[0] = vc->num_threads; |
898b25b2 | 2760 | cip->vc[0] = vc; |
b4deba5c PM |
2761 | } |
2762 | ||
2763 | static bool subcore_config_ok(int n_subcores, int n_threads) | |
2764 | { | |
516f7898 | 2765 | /* |
00608e1f PM |
2766 | * POWER9 "SMT4" cores are permanently in what is effectively a 4-way |
2767 | * split-core mode, with one thread per subcore. | |
516f7898 PM |
2768 | */ |
2769 | if (cpu_has_feature(CPU_FTR_ARCH_300)) | |
2770 | return n_subcores <= 4 && n_threads == 1; | |
2771 | ||
2772 | /* On POWER8, can only dynamically split if unsplit to begin with */ | |
b4deba5c PM |
2773 | if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS) |
2774 | return false; | |
2775 | if (n_subcores > MAX_SUBCORES) | |
2776 | return false; | |
2777 | if (n_subcores > 1) { | |
2778 | if (!(dynamic_mt_modes & 2)) | |
2779 | n_subcores = 4; | |
2780 | if (n_subcores > 2 && !(dynamic_mt_modes & 4)) | |
2781 | return false; | |
2782 | } | |
2783 | ||
2784 | return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS; | |
ec257165 PM |
2785 | } |
2786 | ||
898b25b2 | 2787 | static void init_vcore_to_run(struct kvmppc_vcore *vc) |
ec257165 | 2788 | { |
ec257165 PM |
2789 | vc->entry_exit_map = 0; |
2790 | vc->in_guest = 0; | |
2791 | vc->napping_threads = 0; | |
2792 | vc->conferring_threads = 0; | |
57b8daa7 | 2793 | vc->tb_offset_applied = 0; |
ec257165 PM |
2794 | } |
2795 | ||
b4deba5c PM |
2796 | static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip) |
2797 | { | |
2798 | int n_threads = vc->num_threads; | |
2799 | int sub; | |
2800 | ||
2801 | if (!cpu_has_feature(CPU_FTR_ARCH_207S)) | |
2802 | return false; | |
2803 | ||
aa227864 PM |
2804 | /* In one_vm_per_core mode, require all vcores to be from the same vm */ |
2805 | if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm) | |
2806 | return false; | |
2807 | ||
00608e1f PM |
2808 | /* Some POWER9 chips require all threads to be in the same MMU mode */ |
2809 | if (no_mixing_hpt_and_radix && | |
c0101509 PM |
2810 | kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm)) |
2811 | return false; | |
2812 | ||
b4deba5c PM |
2813 | if (n_threads < cip->max_subcore_threads) |
2814 | n_threads = cip->max_subcore_threads; | |
b009031f | 2815 | if (!subcore_config_ok(cip->n_subcores + 1, n_threads)) |
b4deba5c | 2816 | return false; |
b009031f | 2817 | cip->max_subcore_threads = n_threads; |
b4deba5c PM |
2818 | |
2819 | sub = cip->n_subcores; | |
2820 | ++cip->n_subcores; | |
2821 | cip->total_threads += vc->num_threads; | |
2822 | cip->subcore_threads[sub] = vc->num_threads; | |
898b25b2 PM |
2823 | cip->vc[sub] = vc; |
2824 | init_vcore_to_run(vc); | |
2825 | list_del_init(&vc->preempt_list); | |
b4deba5c PM |
2826 | |
2827 | return true; | |
2828 | } | |
2829 | ||
b4deba5c PM |
2830 | /* |
2831 | * Work out whether it is possible to piggyback the execution of | |
2832 | * vcore *pvc onto the execution of the other vcores described in *cip. | |
2833 | */ | |
2834 | static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip, | |
2835 | int target_threads) | |
2836 | { | |
b4deba5c PM |
2837 | if (cip->total_threads + pvc->num_threads > target_threads) |
2838 | return false; | |
b4deba5c | 2839 | |
b009031f | 2840 | return can_dynamic_split(pvc, cip); |
b4deba5c PM |
2841 | } |
2842 | ||
d911f0be PM |
2843 | static void prepare_threads(struct kvmppc_vcore *vc) |
2844 | { | |
7b5f8272 SJS |
2845 | int i; |
2846 | struct kvm_vcpu *vcpu; | |
d911f0be | 2847 | |
7b5f8272 | 2848 | for_each_runnable_thread(i, vcpu, vc) { |
d911f0be PM |
2849 | if (signal_pending(vcpu->arch.run_task)) |
2850 | vcpu->arch.ret = -EINTR; | |
2851 | else if (vcpu->arch.vpa.update_pending || | |
2852 | vcpu->arch.slb_shadow.update_pending || | |
2853 | vcpu->arch.dtl.update_pending) | |
2854 | vcpu->arch.ret = RESUME_GUEST; | |
2855 | else | |
2856 | continue; | |
2857 | kvmppc_remove_runnable(vc, vcpu); | |
2858 | wake_up(&vcpu->arch.cpu_run); | |
2859 | } | |
2860 | } | |
2861 | ||
ec257165 PM |
2862 | static void collect_piggybacks(struct core_info *cip, int target_threads) |
2863 | { | |
2864 | struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores); | |
2865 | struct kvmppc_vcore *pvc, *vcnext; | |
2866 | ||
2867 | spin_lock(&lp->lock); | |
2868 | list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) { | |
2869 | if (!spin_trylock(&pvc->lock)) | |
2870 | continue; | |
2871 | prepare_threads(pvc); | |
d28eafc5 | 2872 | if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) { |
ec257165 PM |
2873 | list_del_init(&pvc->preempt_list); |
2874 | if (pvc->runner == NULL) { | |
2875 | pvc->vcore_state = VCORE_INACTIVE; | |
2876 | kvmppc_core_end_stolen(pvc); | |
2877 | } | |
2878 | spin_unlock(&pvc->lock); | |
2879 | continue; | |
2880 | } | |
2881 | if (!can_piggyback(pvc, cip, target_threads)) { | |
2882 | spin_unlock(&pvc->lock); | |
2883 | continue; | |
2884 | } | |
2885 | kvmppc_core_end_stolen(pvc); | |
2886 | pvc->vcore_state = VCORE_PIGGYBACK; | |
2887 | if (cip->total_threads >= target_threads) | |
2888 | break; | |
2889 | } | |
2890 | spin_unlock(&lp->lock); | |
2891 | } | |
2892 | ||
d28eafc5 | 2893 | static bool recheck_signals_and_mmu(struct core_info *cip) |
8b24e69f PM |
2894 | { |
2895 | int sub, i; | |
2896 | struct kvm_vcpu *vcpu; | |
d28eafc5 | 2897 | struct kvmppc_vcore *vc; |
8b24e69f | 2898 | |
d28eafc5 PM |
2899 | for (sub = 0; sub < cip->n_subcores; ++sub) { |
2900 | vc = cip->vc[sub]; | |
2901 | if (!vc->kvm->arch.mmu_ready) | |
2902 | return true; | |
2903 | for_each_runnable_thread(i, vcpu, vc) | |
8b24e69f PM |
2904 | if (signal_pending(vcpu->arch.run_task)) |
2905 | return true; | |
d28eafc5 | 2906 | } |
8b24e69f PM |
2907 | return false; |
2908 | } | |
2909 | ||
ec257165 | 2910 | static void post_guest_process(struct kvmppc_vcore *vc, bool is_master) |
25fedfca | 2911 | { |
7b5f8272 | 2912 | int still_running = 0, i; |
25fedfca PM |
2913 | u64 now; |
2914 | long ret; | |
7b5f8272 | 2915 | struct kvm_vcpu *vcpu; |
25fedfca | 2916 | |
ec257165 | 2917 | spin_lock(&vc->lock); |
25fedfca | 2918 | now = get_tb(); |
7b5f8272 | 2919 | for_each_runnable_thread(i, vcpu, vc) { |
53655ddd PM |
2920 | /* |
2921 | * It's safe to unlock the vcore in the loop here, because | |
2922 | * for_each_runnable_thread() is safe against removal of | |
2923 | * the vcpu, and the vcore state is VCORE_EXITING here, | |
2924 | * so any vcpus becoming runnable will have their arch.trap | |
2925 | * set to zero and can't actually run in the guest. | |
2926 | */ | |
2927 | spin_unlock(&vc->lock); | |
25fedfca PM |
2928 | /* cancel pending dec exception if dec is positive */ |
2929 | if (now < vcpu->arch.dec_expires && | |
2930 | kvmppc_core_pending_dec(vcpu)) | |
2931 | kvmppc_core_dequeue_dec(vcpu); | |
2932 | ||
2933 | trace_kvm_guest_exit(vcpu); | |
2934 | ||
2935 | ret = RESUME_GUEST; | |
2936 | if (vcpu->arch.trap) | |
8c99d345 | 2937 | ret = kvmppc_handle_exit_hv(vcpu, |
25fedfca PM |
2938 | vcpu->arch.run_task); |
2939 | ||
2940 | vcpu->arch.ret = ret; | |
2941 | vcpu->arch.trap = 0; | |
2942 | ||
53655ddd | 2943 | spin_lock(&vc->lock); |
ec257165 PM |
2944 | if (is_kvmppc_resume_guest(vcpu->arch.ret)) { |
2945 | if (vcpu->arch.pending_exceptions) | |
2946 | kvmppc_core_prepare_to_enter(vcpu); | |
2947 | if (vcpu->arch.ceded) | |
25fedfca | 2948 | kvmppc_set_timer(vcpu); |
ec257165 PM |
2949 | else |
2950 | ++still_running; | |
2951 | } else { | |
25fedfca PM |
2952 | kvmppc_remove_runnable(vc, vcpu); |
2953 | wake_up(&vcpu->arch.cpu_run); | |
2954 | } | |
2955 | } | |
ec257165 | 2956 | if (!is_master) { |
563a1e93 | 2957 | if (still_running > 0) { |
ec257165 | 2958 | kvmppc_vcore_preempt(vc); |
563a1e93 PM |
2959 | } else if (vc->runner) { |
2960 | vc->vcore_state = VCORE_PREEMPT; | |
2961 | kvmppc_core_start_stolen(vc); | |
2962 | } else { | |
2963 | vc->vcore_state = VCORE_INACTIVE; | |
2964 | } | |
ec257165 PM |
2965 | if (vc->n_runnable > 0 && vc->runner == NULL) { |
2966 | /* make sure there's a candidate runner awake */ | |
7b5f8272 SJS |
2967 | i = -1; |
2968 | vcpu = next_runnable_thread(vc, &i); | |
ec257165 PM |
2969 | wake_up(&vcpu->arch.cpu_run); |
2970 | } | |
2971 | } | |
2972 | spin_unlock(&vc->lock); | |
25fedfca PM |
2973 | } |
2974 | ||
b8e6a87c SW |
2975 | /* |
2976 | * Clear core from the list of active host cores as we are about to | |
2977 | * enter the guest. Only do this if it is the primary thread of the | |
2978 | * core (not if a subcore) that is entering the guest. | |
2979 | */ | |
3f7cd919 | 2980 | static inline int kvmppc_clear_host_core(unsigned int cpu) |
b8e6a87c SW |
2981 | { |
2982 | int core; | |
2983 | ||
2984 | if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu)) | |
3f7cd919 | 2985 | return 0; |
b8e6a87c SW |
2986 | /* |
2987 | * Memory barrier can be omitted here as we will do a smp_wmb() | |
2988 | * later in kvmppc_start_thread and we need ensure that state is | |
2989 | * visible to other CPUs only after we enter guest. | |
2990 | */ | |
2991 | core = cpu >> threads_shift; | |
2992 | kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0; | |
3f7cd919 | 2993 | return 0; |
b8e6a87c SW |
2994 | } |
2995 | ||
2996 | /* | |
2997 | * Advertise this core as an active host core since we exited the guest | |
2998 | * Only need to do this if it is the primary thread of the core that is | |
2999 | * exiting. | |
3000 | */ | |
3f7cd919 | 3001 | static inline int kvmppc_set_host_core(unsigned int cpu) |
b8e6a87c SW |
3002 | { |
3003 | int core; | |
3004 | ||
3005 | if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu)) | |
3f7cd919 | 3006 | return 0; |
b8e6a87c SW |
3007 | |
3008 | /* | |
3009 | * Memory barrier can be omitted here because we do a spin_unlock | |
3010 | * immediately after this which provides the memory barrier. | |
3011 | */ | |
3012 | core = cpu >> threads_shift; | |
3013 | kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1; | |
3f7cd919 | 3014 | return 0; |
b8e6a87c SW |
3015 | } |
3016 | ||
8b24e69f PM |
3017 | static void set_irq_happened(int trap) |
3018 | { | |
3019 | switch (trap) { | |
3020 | case BOOK3S_INTERRUPT_EXTERNAL: | |
3021 | local_paca->irq_happened |= PACA_IRQ_EE; | |
3022 | break; | |
3023 | case BOOK3S_INTERRUPT_H_DOORBELL: | |
3024 | local_paca->irq_happened |= PACA_IRQ_DBELL; | |
3025 | break; | |
3026 | case BOOK3S_INTERRUPT_HMI: | |
3027 | local_paca->irq_happened |= PACA_IRQ_HMI; | |
3028 | break; | |
6de6638b NP |
3029 | case BOOK3S_INTERRUPT_SYSTEM_RESET: |
3030 | replay_system_reset(); | |
3031 | break; | |
8b24e69f PM |
3032 | } |
3033 | } | |
3034 | ||
371fefd6 PM |
3035 | /* |
3036 | * Run a set of guest threads on a physical core. | |
3037 | * Called with vc->lock held. | |
3038 | */ | |
66feed61 | 3039 | static noinline void kvmppc_run_core(struct kvmppc_vcore *vc) |
371fefd6 | 3040 | { |
7b5f8272 | 3041 | struct kvm_vcpu *vcpu; |
d911f0be | 3042 | int i; |
2c9097e4 | 3043 | int srcu_idx; |
ec257165 | 3044 | struct core_info core_info; |
898b25b2 | 3045 | struct kvmppc_vcore *pvc; |
b4deba5c PM |
3046 | struct kvm_split_mode split_info, *sip; |
3047 | int split, subcore_size, active; | |
3048 | int sub; | |
3049 | bool thr0_done; | |
3050 | unsigned long cmd_bit, stat_bit; | |
ec257165 PM |
3051 | int pcpu, thr; |
3052 | int target_threads; | |
45c940ba | 3053 | int controlled_threads; |
8b24e69f | 3054 | int trap; |
516f7898 | 3055 | bool is_power8; |
c0101509 | 3056 | bool hpt_on_radix; |
371fefd6 | 3057 | |
d911f0be PM |
3058 | /* |
3059 | * Remove from the list any threads that have a signal pending | |
3060 | * or need a VPA update done | |
3061 | */ | |
3062 | prepare_threads(vc); | |
3063 | ||
3064 | /* if the runner is no longer runnable, let the caller pick a new one */ | |
3065 | if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE) | |
3066 | return; | |
081f323b PM |
3067 | |
3068 | /* | |
d911f0be | 3069 | * Initialize *vc. |
081f323b | 3070 | */ |
898b25b2 | 3071 | init_vcore_to_run(vc); |
2711e248 | 3072 | vc->preempt_tb = TB_NIL; |
081f323b | 3073 | |
45c940ba PM |
3074 | /* |
3075 | * Number of threads that we will be controlling: the same as | |
3076 | * the number of threads per subcore, except on POWER9, | |
3077 | * where it's 1 because the threads are (mostly) independent. | |
3078 | */ | |
516f7898 | 3079 | controlled_threads = threads_per_vcore(vc->kvm); |
45c940ba | 3080 | |
7b444c67 | 3081 | /* |
3102f784 ME |
3082 | * Make sure we are running on primary threads, and that secondary |
3083 | * threads are offline. Also check if the number of threads in this | |
3084 | * guest are greater than the current system threads per guest. | |
c0101509 | 3085 | * On POWER9, we need to be not in independent-threads mode if |
00608e1f PM |
3086 | * this is a HPT guest on a radix host machine where the |
3087 | * CPU threads may not be in different MMU modes. | |
7b444c67 | 3088 | */ |
00608e1f PM |
3089 | hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() && |
3090 | !kvm_is_radix(vc->kvm); | |
c0101509 PM |
3091 | if (((controlled_threads > 1) && |
3092 | ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) || | |
3093 | (hpt_on_radix && vc->kvm->arch.threads_indep)) { | |
7b5f8272 | 3094 | for_each_runnable_thread(i, vcpu, vc) { |
7b444c67 | 3095 | vcpu->arch.ret = -EBUSY; |
25fedfca PM |
3096 | kvmppc_remove_runnable(vc, vcpu); |
3097 | wake_up(&vcpu->arch.cpu_run); | |
3098 | } | |
7b444c67 PM |
3099 | goto out; |
3100 | } | |
3101 | ||
ec257165 PM |
3102 | /* |
3103 | * See if we could run any other vcores on the physical core | |
3104 | * along with this one. | |
3105 | */ | |
3106 | init_core_info(&core_info, vc); | |
3107 | pcpu = smp_processor_id(); | |
45c940ba | 3108 | target_threads = controlled_threads; |
ec257165 PM |
3109 | if (target_smt_mode && target_smt_mode < target_threads) |
3110 | target_threads = target_smt_mode; | |
3111 | if (vc->num_threads < target_threads) | |
3112 | collect_piggybacks(&core_info, target_threads); | |
3102f784 | 3113 | |
8b24e69f PM |
3114 | /* |
3115 | * On radix, arrange for TLB flushing if necessary. | |
3116 | * This has to be done before disabling interrupts since | |
3117 | * it uses smp_call_function(). | |
3118 | */ | |
3119 | pcpu = smp_processor_id(); | |
3120 | if (kvm_is_radix(vc->kvm)) { | |
3121 | for (sub = 0; sub < core_info.n_subcores; ++sub) | |
3122 | for_each_runnable_thread(i, vcpu, core_info.vc[sub]) | |
3123 | kvmppc_prepare_radix_vcpu(vcpu, pcpu); | |
3124 | } | |
3125 | ||
3126 | /* | |
3127 | * Hard-disable interrupts, and check resched flag and signals. | |
3128 | * If we need to reschedule or deliver a signal, clean up | |
3129 | * and return without going into the guest(s). | |
072df813 | 3130 | * If the mmu_ready flag has been cleared, don't go into the |
38c53af8 | 3131 | * guest because that means a HPT resize operation is in progress. |
8b24e69f PM |
3132 | */ |
3133 | local_irq_disable(); | |
3134 | hard_irq_disable(); | |
3135 | if (lazy_irq_pending() || need_resched() || | |
d28eafc5 | 3136 | recheck_signals_and_mmu(&core_info)) { |
8b24e69f PM |
3137 | local_irq_enable(); |
3138 | vc->vcore_state = VCORE_INACTIVE; | |
3139 | /* Unlock all except the primary vcore */ | |
3140 | for (sub = 1; sub < core_info.n_subcores; ++sub) { | |
3141 | pvc = core_info.vc[sub]; | |
3142 | /* Put back on to the preempted vcores list */ | |
3143 | kvmppc_vcore_preempt(pvc); | |
3144 | spin_unlock(&pvc->lock); | |
3145 | } | |
3146 | for (i = 0; i < controlled_threads; ++i) | |
3147 | kvmppc_release_hwthread(pcpu + i); | |
3148 | return; | |
3149 | } | |
3150 | ||
3151 | kvmppc_clear_host_core(pcpu); | |
3152 | ||
b4deba5c PM |
3153 | /* Decide on micro-threading (split-core) mode */ |
3154 | subcore_size = threads_per_subcore; | |
3155 | cmd_bit = stat_bit = 0; | |
3156 | split = core_info.n_subcores; | |
3157 | sip = NULL; | |
516f7898 PM |
3158 | is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S) |
3159 | && !cpu_has_feature(CPU_FTR_ARCH_300); | |
3160 | ||
c0101509 | 3161 | if (split > 1 || hpt_on_radix) { |
b4deba5c PM |
3162 | sip = &split_info; |
3163 | memset(&split_info, 0, sizeof(split_info)); | |
b4deba5c | 3164 | for (sub = 0; sub < core_info.n_subcores; ++sub) |
898b25b2 | 3165 | split_info.vc[sub] = core_info.vc[sub]; |
516f7898 PM |
3166 | |
3167 | if (is_power8) { | |
3168 | if (split == 2 && (dynamic_mt_modes & 2)) { | |
3169 | cmd_bit = HID0_POWER8_1TO2LPAR; | |
3170 | stat_bit = HID0_POWER8_2LPARMODE; | |
3171 | } else { | |
3172 | split = 4; | |
3173 | cmd_bit = HID0_POWER8_1TO4LPAR; | |
3174 | stat_bit = HID0_POWER8_4LPARMODE; | |
3175 | } | |
3176 | subcore_size = MAX_SMT_THREADS / split; | |
3177 | split_info.rpr = mfspr(SPRN_RPR); | |
3178 | split_info.pmmar = mfspr(SPRN_PMMAR); | |
3179 | split_info.ldbar = mfspr(SPRN_LDBAR); | |
3180 | split_info.subcore_size = subcore_size; | |
3181 | } else { | |
3182 | split_info.subcore_size = 1; | |
c0101509 PM |
3183 | if (hpt_on_radix) { |
3184 | /* Use the split_info for LPCR/LPIDR changes */ | |
3185 | split_info.lpcr_req = vc->lpcr; | |
3186 | split_info.lpidr_req = vc->kvm->arch.lpid; | |
3187 | split_info.host_lpcr = vc->kvm->arch.host_lpcr; | |
3188 | split_info.do_set = 1; | |
3189 | } | |
516f7898 PM |
3190 | } |
3191 | ||
b4deba5c PM |
3192 | /* order writes to split_info before kvm_split_mode pointer */ |
3193 | smp_wmb(); | |
3194 | } | |
c0101509 PM |
3195 | |
3196 | for (thr = 0; thr < controlled_threads; ++thr) { | |
d2e60075 NP |
3197 | struct paca_struct *paca = paca_ptrs[pcpu + thr]; |
3198 | ||
3199 | paca->kvm_hstate.tid = thr; | |
3200 | paca->kvm_hstate.napping = 0; | |
3201 | paca->kvm_hstate.kvm_split_mode = sip; | |
c0101509 | 3202 | } |
b4deba5c | 3203 | |
516f7898 | 3204 | /* Initiate micro-threading (split-core) on POWER8 if required */ |
b4deba5c PM |
3205 | if (cmd_bit) { |
3206 | unsigned long hid0 = mfspr(SPRN_HID0); | |
3207 | ||
3208 | hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS; | |
3209 | mb(); | |
3210 | mtspr(SPRN_HID0, hid0); | |
3211 | isync(); | |
3212 | for (;;) { | |
3213 | hid0 = mfspr(SPRN_HID0); | |
3214 | if (hid0 & stat_bit) | |
3215 | break; | |
3216 | cpu_relax(); | |
ec257165 | 3217 | } |
2e25aa5f | 3218 | } |
3102f784 | 3219 | |
7aa15842 PM |
3220 | /* |
3221 | * On POWER8, set RWMR register. | |
3222 | * Since it only affects PURR and SPURR, it doesn't affect | |
3223 | * the host, so we don't save/restore the host value. | |
3224 | */ | |
3225 | if (is_power8) { | |
3226 | unsigned long rwmr_val = RWMR_RPA_P8_8THREAD; | |
3227 | int n_online = atomic_read(&vc->online_count); | |
3228 | ||
3229 | /* | |
3230 | * Use the 8-thread value if we're doing split-core | |
3231 | * or if the vcore's online count looks bogus. | |
3232 | */ | |
3233 | if (split == 1 && threads_per_subcore == MAX_SMT_THREADS && | |
3234 | n_online >= 1 && n_online <= MAX_SMT_THREADS) | |
3235 | rwmr_val = p8_rwmr_values[n_online]; | |
3236 | mtspr(SPRN_RWMR, rwmr_val); | |
3237 | } | |
3238 | ||
b4deba5c PM |
3239 | /* Start all the threads */ |
3240 | active = 0; | |
3241 | for (sub = 0; sub < core_info.n_subcores; ++sub) { | |
516f7898 | 3242 | thr = is_power8 ? subcore_thread_map[sub] : sub; |
b4deba5c PM |
3243 | thr0_done = false; |
3244 | active |= 1 << thr; | |
898b25b2 PM |
3245 | pvc = core_info.vc[sub]; |
3246 | pvc->pcpu = pcpu + thr; | |
3247 | for_each_runnable_thread(i, vcpu, pvc) { | |
3248 | kvmppc_start_thread(vcpu, pvc); | |
3249 | kvmppc_create_dtl_entry(vcpu, pvc); | |
3250 | trace_kvm_guest_enter(vcpu); | |
3251 | if (!vcpu->arch.ptid) | |
3252 | thr0_done = true; | |
3253 | active |= 1 << (thr + vcpu->arch.ptid); | |
b4deba5c | 3254 | } |
898b25b2 PM |
3255 | /* |
3256 | * We need to start the first thread of each subcore | |
3257 | * even if it doesn't have a vcpu. | |
3258 | */ | |
3259 | if (!thr0_done) | |
3260 | kvmppc_start_thread(NULL, pvc); | |
2e25aa5f | 3261 | } |
371fefd6 | 3262 | |
7f235328 GS |
3263 | /* |
3264 | * Ensure that split_info.do_nap is set after setting | |
3265 | * the vcore pointer in the PACA of the secondaries. | |
3266 | */ | |
3267 | smp_mb(); | |
7f235328 | 3268 | |
b4deba5c PM |
3269 | /* |
3270 | * When doing micro-threading, poke the inactive threads as well. | |
3271 | * This gets them to the nap instruction after kvm_do_nap, | |
3272 | * which reduces the time taken to unsplit later. | |
c0101509 PM |
3273 | * For POWER9 HPT guest on radix host, we need all the secondary |
3274 | * threads woken up so they can do the LPCR/LPIDR change. | |
b4deba5c | 3275 | */ |
c0101509 | 3276 | if (cmd_bit || hpt_on_radix) { |
516f7898 | 3277 | split_info.do_nap = 1; /* ask secondaries to nap when done */ |
b4deba5c PM |
3278 | for (thr = 1; thr < threads_per_subcore; ++thr) |
3279 | if (!(active & (1 << thr))) | |
3280 | kvmppc_ipi_thread(pcpu + thr); | |
516f7898 | 3281 | } |
e0b7ec05 | 3282 | |
2f12f034 | 3283 | vc->vcore_state = VCORE_RUNNING; |
19ccb76a | 3284 | preempt_disable(); |
3c78f78a SW |
3285 | |
3286 | trace_kvmppc_run_core(vc, 0); | |
3287 | ||
b4deba5c | 3288 | for (sub = 0; sub < core_info.n_subcores; ++sub) |
898b25b2 | 3289 | spin_unlock(&core_info.vc[sub]->lock); |
de56a948 | 3290 | |
61bd0f66 | 3291 | guest_enter_irqoff(); |
2c9097e4 | 3292 | |
e0b7ec05 | 3293 | srcu_idx = srcu_read_lock(&vc->kvm->srcu); |
2c9097e4 | 3294 | |
a4bc64d3 NR |
3295 | this_cpu_disable_ftrace(); |
3296 | ||
3309bec8 AK |
3297 | /* |
3298 | * Interrupts will be enabled once we get into the guest, | |
3299 | * so tell lockdep that we're about to enable interrupts. | |
3300 | */ | |
3301 | trace_hardirqs_on(); | |
3302 | ||
8b24e69f | 3303 | trap = __kvmppc_vcore_entry(); |
de56a948 | 3304 | |
3309bec8 AK |
3305 | trace_hardirqs_off(); |
3306 | ||
a4bc64d3 NR |
3307 | this_cpu_enable_ftrace(); |
3308 | ||
ec257165 PM |
3309 | srcu_read_unlock(&vc->kvm->srcu, srcu_idx); |
3310 | ||
8b24e69f PM |
3311 | set_irq_happened(trap); |
3312 | ||
ec257165 | 3313 | spin_lock(&vc->lock); |
371fefd6 | 3314 | /* prevent other vcpu threads from doing kvmppc_start_thread() now */ |
19ccb76a | 3315 | vc->vcore_state = VCORE_EXITING; |
371fefd6 | 3316 | |
19ccb76a | 3317 | /* wait for secondary threads to finish writing their state to memory */ |
516f7898 | 3318 | kvmppc_wait_for_nap(controlled_threads); |
b4deba5c PM |
3319 | |
3320 | /* Return to whole-core mode if we split the core earlier */ | |
516f7898 | 3321 | if (cmd_bit) { |
b4deba5c PM |
3322 | unsigned long hid0 = mfspr(SPRN_HID0); |
3323 | unsigned long loops = 0; | |
3324 | ||
3325 | hid0 &= ~HID0_POWER8_DYNLPARDIS; | |
3326 | stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE; | |
3327 | mb(); | |
3328 | mtspr(SPRN_HID0, hid0); | |
3329 | isync(); | |
3330 | for (;;) { | |
3331 | hid0 = mfspr(SPRN_HID0); | |
3332 | if (!(hid0 & stat_bit)) | |
3333 | break; | |
3334 | cpu_relax(); | |
3335 | ++loops; | |
3336 | } | |
c0101509 PM |
3337 | } else if (hpt_on_radix) { |
3338 | /* Wait for all threads to have seen final sync */ | |
3339 | for (thr = 1; thr < controlled_threads; ++thr) { | |
d2e60075 NP |
3340 | struct paca_struct *paca = paca_ptrs[pcpu + thr]; |
3341 | ||
3342 | while (paca->kvm_hstate.kvm_split_mode) { | |
c0101509 PM |
3343 | HMT_low(); |
3344 | barrier(); | |
3345 | } | |
3346 | HMT_medium(); | |
3347 | } | |
b4deba5c | 3348 | } |
c0101509 | 3349 | split_info.do_nap = 0; |
b4deba5c | 3350 | |
8b24e69f PM |
3351 | kvmppc_set_host_core(pcpu); |
3352 | ||
3353 | local_irq_enable(); | |
61bd0f66 | 3354 | guest_exit(); |
8b24e69f | 3355 | |
b4deba5c | 3356 | /* Let secondaries go back to the offline loop */ |
45c940ba | 3357 | for (i = 0; i < controlled_threads; ++i) { |
b4deba5c PM |
3358 | kvmppc_release_hwthread(pcpu + i); |
3359 | if (sip && sip->napped[i]) | |
3360 | kvmppc_ipi_thread(pcpu + i); | |
a29ebeaf | 3361 | cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest); |
b4deba5c PM |
3362 | } |
3363 | ||
371fefd6 | 3364 | spin_unlock(&vc->lock); |
2c9097e4 | 3365 | |
371fefd6 PM |
3366 | /* make sure updates to secondary vcpu structs are visible now */ |
3367 | smp_mb(); | |
de56a948 | 3368 | |
36ee41d1 PM |
3369 | preempt_enable(); |
3370 | ||
898b25b2 PM |
3371 | for (sub = 0; sub < core_info.n_subcores; ++sub) { |
3372 | pvc = core_info.vc[sub]; | |
3373 | post_guest_process(pvc, pvc == vc); | |
3374 | } | |
de56a948 | 3375 | |
913d3ff9 | 3376 | spin_lock(&vc->lock); |
de56a948 PM |
3377 | |
3378 | out: | |
19ccb76a | 3379 | vc->vcore_state = VCORE_INACTIVE; |
3c78f78a | 3380 | trace_kvmppc_run_core(vc, 1); |
371fefd6 PM |
3381 | } |
3382 | ||
95a6432c PM |
3383 | /* |
3384 | * Load up hypervisor-mode registers on P9. | |
3385 | */ | |
360cae31 PM |
3386 | static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit, |
3387 | unsigned long lpcr) | |
95a6432c PM |
3388 | { |
3389 | struct kvmppc_vcore *vc = vcpu->arch.vcore; | |
3390 | s64 hdec; | |
3391 | u64 tb, purr, spurr; | |
3392 | int trap; | |
3393 | unsigned long host_hfscr = mfspr(SPRN_HFSCR); | |
3394 | unsigned long host_ciabr = mfspr(SPRN_CIABR); | |
09f82b06 RB |
3395 | unsigned long host_dawr = mfspr(SPRN_DAWR0); |
3396 | unsigned long host_dawrx = mfspr(SPRN_DAWRX0); | |
95a6432c PM |
3397 | unsigned long host_psscr = mfspr(SPRN_PSSCR); |
3398 | unsigned long host_pidr = mfspr(SPRN_PID); | |
3399 | ||
3400 | hdec = time_limit - mftb(); | |
3401 | if (hdec < 0) | |
3402 | return BOOK3S_INTERRUPT_HV_DECREMENTER; | |
3403 | mtspr(SPRN_HDEC, hdec); | |
3404 | ||
3405 | if (vc->tb_offset) { | |
3406 | u64 new_tb = mftb() + vc->tb_offset; | |
3407 | mtspr(SPRN_TBU40, new_tb); | |
3408 | tb = mftb(); | |
3409 | if ((tb & 0xffffff) < (new_tb & 0xffffff)) | |
3410 | mtspr(SPRN_TBU40, new_tb + 0x1000000); | |
3411 | vc->tb_offset_applied = vc->tb_offset; | |
3412 | } | |
3413 | ||
3414 | if (vc->pcr) | |
13c7bb3c | 3415 | mtspr(SPRN_PCR, vc->pcr | PCR_MASK); |
95a6432c PM |
3416 | mtspr(SPRN_DPDES, vc->dpdes); |
3417 | mtspr(SPRN_VTB, vc->vtb); | |
3418 | ||
3419 | local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR); | |
3420 | local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR); | |
3421 | mtspr(SPRN_PURR, vcpu->arch.purr); | |
3422 | mtspr(SPRN_SPURR, vcpu->arch.spurr); | |
3423 | ||
c1fe190c | 3424 | if (dawr_enabled()) { |
09f82b06 RB |
3425 | mtspr(SPRN_DAWR0, vcpu->arch.dawr); |
3426 | mtspr(SPRN_DAWRX0, vcpu->arch.dawrx); | |
95a6432c PM |
3427 | } |
3428 | mtspr(SPRN_CIABR, vcpu->arch.ciabr); | |
3429 | mtspr(SPRN_IC, vcpu->arch.ic); | |
3430 | mtspr(SPRN_PID, vcpu->arch.pid); | |
3431 | ||
3432 | mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC | | |
3433 | (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG)); | |
3434 | ||
3435 | mtspr(SPRN_HFSCR, vcpu->arch.hfscr); | |
3436 | ||
3437 | mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0); | |
3438 | mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1); | |
3439 | mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2); | |
3440 | mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3); | |
3441 | ||
3442 | mtspr(SPRN_AMOR, ~0UL); | |
3443 | ||
360cae31 | 3444 | mtspr(SPRN_LPCR, lpcr); |
95a6432c PM |
3445 | isync(); |
3446 | ||
3447 | kvmppc_xive_push_vcpu(vcpu); | |
3448 | ||
3449 | mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0); | |
3450 | mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1); | |
3451 | ||
3452 | trap = __kvmhv_vcpu_entry_p9(vcpu); | |
3453 | ||
3454 | /* Advance host PURR/SPURR by the amount used by guest */ | |
3455 | purr = mfspr(SPRN_PURR); | |
3456 | spurr = mfspr(SPRN_SPURR); | |
3457 | mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr + | |
3458 | purr - vcpu->arch.purr); | |
3459 | mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr + | |
3460 | spurr - vcpu->arch.spurr); | |
3461 | vcpu->arch.purr = purr; | |
3462 | vcpu->arch.spurr = spurr; | |
3463 | ||
3464 | vcpu->arch.ic = mfspr(SPRN_IC); | |
3465 | vcpu->arch.pid = mfspr(SPRN_PID); | |
3466 | vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS; | |
3467 | ||
3468 | vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0); | |
3469 | vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1); | |
3470 | vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2); | |
3471 | vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3); | |
3472 | ||
7cb9eb10 SJS |
3473 | /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */ |
3474 | mtspr(SPRN_PSSCR, host_psscr | | |
3475 | (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG)); | |
95a6432c PM |
3476 | mtspr(SPRN_HFSCR, host_hfscr); |
3477 | mtspr(SPRN_CIABR, host_ciabr); | |
09f82b06 RB |
3478 | mtspr(SPRN_DAWR0, host_dawr); |
3479 | mtspr(SPRN_DAWRX0, host_dawrx); | |
95a6432c PM |
3480 | mtspr(SPRN_PID, host_pidr); |
3481 | ||
3482 | /* | |
3483 | * Since this is radix, do a eieio; tlbsync; ptesync sequence in | |
3484 | * case we interrupted the guest between a tlbie and a ptesync. | |
3485 | */ | |
3486 | asm volatile("eieio; tlbsync; ptesync"); | |
3487 | ||
3488 | mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid); /* restore host LPID */ | |
3489 | isync(); | |
3490 | ||
3491 | vc->dpdes = mfspr(SPRN_DPDES); | |
3492 | vc->vtb = mfspr(SPRN_VTB); | |
3493 | mtspr(SPRN_DPDES, 0); | |
3494 | if (vc->pcr) | |
13c7bb3c | 3495 | mtspr(SPRN_PCR, PCR_MASK); |
95a6432c PM |
3496 | |
3497 | if (vc->tb_offset_applied) { | |
3498 | u64 new_tb = mftb() - vc->tb_offset_applied; | |
3499 | mtspr(SPRN_TBU40, new_tb); | |
3500 | tb = mftb(); | |
3501 | if ((tb & 0xffffff) < (new_tb & 0xffffff)) | |
3502 | mtspr(SPRN_TBU40, new_tb + 0x1000000); | |
3503 | vc->tb_offset_applied = 0; | |
3504 | } | |
3505 | ||
3506 | mtspr(SPRN_HDEC, 0x7fffffff); | |
3507 | mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr); | |
3508 | ||
3509 | return trap; | |
3510 | } | |
3511 | ||
3512 | /* | |
3513 | * Virtual-mode guest entry for POWER9 and later when the host and | |
3514 | * guest are both using the radix MMU. The LPIDR has already been set. | |
3515 | */ | |
360cae31 PM |
3516 | int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit, |
3517 | unsigned long lpcr) | |
95a6432c PM |
3518 | { |
3519 | struct kvmppc_vcore *vc = vcpu->arch.vcore; | |
3520 | unsigned long host_dscr = mfspr(SPRN_DSCR); | |
3521 | unsigned long host_tidr = mfspr(SPRN_TIDR); | |
3522 | unsigned long host_iamr = mfspr(SPRN_IAMR); | |
d976f680 | 3523 | unsigned long host_amr = mfspr(SPRN_AMR); |
95a6432c PM |
3524 | s64 dec; |
3525 | u64 tb; | |
3526 | int trap, save_pmu; | |
3527 | ||
3528 | dec = mfspr(SPRN_DEC); | |
3529 | tb = mftb(); | |
3530 | if (dec < 512) | |
3531 | return BOOK3S_INTERRUPT_HV_DECREMENTER; | |
3532 | local_paca->kvm_hstate.dec_expires = dec + tb; | |
3533 | if (local_paca->kvm_hstate.dec_expires < time_limit) | |
3534 | time_limit = local_paca->kvm_hstate.dec_expires; | |
3535 | ||
3536 | vcpu->arch.ceded = 0; | |
3537 | ||
3538 | kvmhv_save_host_pmu(); /* saves it to PACA kvm_hstate */ | |
3539 | ||
3540 | kvmppc_subcore_enter_guest(); | |
3541 | ||
3542 | vc->entry_exit_map = 1; | |
3543 | vc->in_guest = 1; | |
3544 | ||
3545 | if (vcpu->arch.vpa.pinned_addr) { | |
3546 | struct lppaca *lp = vcpu->arch.vpa.pinned_addr; | |
3547 | u32 yield_count = be32_to_cpu(lp->yield_count) + 1; | |
3548 | lp->yield_count = cpu_to_be32(yield_count); | |
3549 | vcpu->arch.vpa.dirty = 1; | |
3550 | } | |
3551 | ||
3552 | if (cpu_has_feature(CPU_FTR_TM) || | |
3553 | cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) | |
3554 | kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true); | |
3555 | ||
3556 | kvmhv_load_guest_pmu(vcpu); | |
3557 | ||
3558 | msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX); | |
3559 | load_fp_state(&vcpu->arch.fp); | |
3560 | #ifdef CONFIG_ALTIVEC | |
3561 | load_vr_state(&vcpu->arch.vr); | |
3562 | #endif | |
44b198ae | 3563 | mtspr(SPRN_VRSAVE, vcpu->arch.vrsave); |
95a6432c PM |
3564 | |
3565 | mtspr(SPRN_DSCR, vcpu->arch.dscr); | |
3566 | mtspr(SPRN_IAMR, vcpu->arch.iamr); | |
3567 | mtspr(SPRN_PSPB, vcpu->arch.pspb); | |
3568 | mtspr(SPRN_FSCR, vcpu->arch.fscr); | |
3569 | mtspr(SPRN_TAR, vcpu->arch.tar); | |
3570 | mtspr(SPRN_EBBHR, vcpu->arch.ebbhr); | |
3571 | mtspr(SPRN_EBBRR, vcpu->arch.ebbrr); | |
3572 | mtspr(SPRN_BESCR, vcpu->arch.bescr); | |
3573 | mtspr(SPRN_WORT, vcpu->arch.wort); | |
3574 | mtspr(SPRN_TIDR, vcpu->arch.tid); | |
3575 | mtspr(SPRN_DAR, vcpu->arch.shregs.dar); | |
3576 | mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr); | |
3577 | mtspr(SPRN_AMR, vcpu->arch.amr); | |
3578 | mtspr(SPRN_UAMOR, vcpu->arch.uamor); | |
3579 | ||
3580 | if (!(vcpu->arch.ctrl & 1)) | |
3581 | mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1); | |
3582 | ||
3583 | mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb()); | |
3584 | ||
360cae31 | 3585 | if (kvmhv_on_pseries()) { |
c8b4083d SJS |
3586 | /* |
3587 | * We need to save and restore the guest visible part of the | |
3588 | * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor | |
3589 | * doesn't do this for us. Note only required if pseries since | |
3590 | * this is done in kvmhv_load_hv_regs_and_go() below otherwise. | |
3591 | */ | |
3592 | unsigned long host_psscr; | |
360cae31 PM |
3593 | /* call our hypervisor to load up HV regs and go */ |
3594 | struct hv_guest_state hvregs; | |
3595 | ||
c8b4083d SJS |
3596 | host_psscr = mfspr(SPRN_PSSCR_PR); |
3597 | mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr); | |
360cae31 PM |
3598 | kvmhv_save_hv_regs(vcpu, &hvregs); |
3599 | hvregs.lpcr = lpcr; | |
3600 | vcpu->arch.regs.msr = vcpu->arch.shregs.msr; | |
3601 | hvregs.version = HV_GUEST_STATE_VERSION; | |
3602 | if (vcpu->arch.nested) { | |
3603 | hvregs.lpid = vcpu->arch.nested->shadow_lpid; | |
3604 | hvregs.vcpu_token = vcpu->arch.nested_vcpu_id; | |
3605 | } else { | |
3606 | hvregs.lpid = vcpu->kvm->arch.lpid; | |
3607 | hvregs.vcpu_token = vcpu->vcpu_id; | |
3608 | } | |
3609 | hvregs.hdec_expiry = time_limit; | |
3610 | trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs), | |
3611 | __pa(&vcpu->arch.regs)); | |
3612 | kvmhv_restore_hv_return_state(vcpu, &hvregs); | |
3613 | vcpu->arch.shregs.msr = vcpu->arch.regs.msr; | |
3614 | vcpu->arch.shregs.dar = mfspr(SPRN_DAR); | |
3615 | vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR); | |
c8b4083d SJS |
3616 | vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR); |
3617 | mtspr(SPRN_PSSCR_PR, host_psscr); | |
4bad7779 PM |
3618 | |
3619 | /* H_CEDE has to be handled now, not later */ | |
3620 | if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested && | |
3621 | kvmppc_get_gpr(vcpu, 3) == H_CEDE) { | |
3622 | kvmppc_nested_cede(vcpu); | |
1f50cc17 | 3623 | kvmppc_set_gpr(vcpu, 3, 0); |
4bad7779 PM |
3624 | trap = 0; |
3625 | } | |
360cae31 PM |
3626 | } else { |
3627 | trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr); | |
95a6432c PM |
3628 | } |
3629 | ||
95a6432c PM |
3630 | vcpu->arch.slb_max = 0; |
3631 | dec = mfspr(SPRN_DEC); | |
86953770 SJS |
3632 | if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */ |
3633 | dec = (s32) dec; | |
95a6432c PM |
3634 | tb = mftb(); |
3635 | vcpu->arch.dec_expires = dec + tb; | |
3636 | vcpu->cpu = -1; | |
3637 | vcpu->arch.thread_cpu = -1; | |
3638 | vcpu->arch.ctrl = mfspr(SPRN_CTRLF); | |
3639 | ||
3640 | vcpu->arch.iamr = mfspr(SPRN_IAMR); | |
3641 | vcpu->arch.pspb = mfspr(SPRN_PSPB); | |
3642 | vcpu->arch.fscr = mfspr(SPRN_FSCR); | |
3643 | vcpu->arch.tar = mfspr(SPRN_TAR); | |
3644 | vcpu->arch.ebbhr = mfspr(SPRN_EBBHR); | |
3645 | vcpu->arch.ebbrr = mfspr(SPRN_EBBRR); | |
3646 | vcpu->arch.bescr = mfspr(SPRN_BESCR); | |
3647 | vcpu->arch.wort = mfspr(SPRN_WORT); | |
3648 | vcpu->arch.tid = mfspr(SPRN_TIDR); | |
3649 | vcpu->arch.amr = mfspr(SPRN_AMR); | |
3650 | vcpu->arch.uamor = mfspr(SPRN_UAMOR); | |
3651 | vcpu->arch.dscr = mfspr(SPRN_DSCR); | |
3652 | ||
3653 | mtspr(SPRN_PSPB, 0); | |
3654 | mtspr(SPRN_WORT, 0); | |
95a6432c PM |
3655 | mtspr(SPRN_UAMOR, 0); |
3656 | mtspr(SPRN_DSCR, host_dscr); | |
3657 | mtspr(SPRN_TIDR, host_tidr); | |
3658 | mtspr(SPRN_IAMR, host_iamr); | |
3659 | mtspr(SPRN_PSPB, 0); | |
3660 | ||
d976f680 ME |
3661 | if (host_amr != vcpu->arch.amr) |
3662 | mtspr(SPRN_AMR, host_amr); | |
3663 | ||
95a6432c PM |
3664 | msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX); |
3665 | store_fp_state(&vcpu->arch.fp); | |
3666 | #ifdef CONFIG_ALTIVEC | |
3667 | store_vr_state(&vcpu->arch.vr); | |
3668 | #endif | |
44b198ae | 3669 | vcpu->arch.vrsave = mfspr(SPRN_VRSAVE); |
95a6432c PM |
3670 | |
3671 | if (cpu_has_feature(CPU_FTR_TM) || | |
3672 | cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) | |
3673 | kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true); | |
3674 | ||
3675 | save_pmu = 1; | |
3676 | if (vcpu->arch.vpa.pinned_addr) { | |
3677 | struct lppaca *lp = vcpu->arch.vpa.pinned_addr; | |
3678 | u32 yield_count = be32_to_cpu(lp->yield_count) + 1; | |
3679 | lp->yield_count = cpu_to_be32(yield_count); | |
3680 | vcpu->arch.vpa.dirty = 1; | |
3681 | save_pmu = lp->pmcregs_in_use; | |
3682 | } | |
63279eeb SJS |
3683 | /* Must save pmu if this guest is capable of running nested guests */ |
3684 | save_pmu |= nesting_enabled(vcpu->kvm); | |
95a6432c PM |
3685 | |
3686 | kvmhv_save_guest_pmu(vcpu, save_pmu); | |
3687 | ||
3688 | vc->entry_exit_map = 0x101; | |
3689 | vc->in_guest = 0; | |
3690 | ||
3691 | mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb()); | |
d724c9e5 | 3692 | mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso); |
95a6432c PM |
3693 | |
3694 | kvmhv_load_host_pmu(); | |
3695 | ||
3696 | kvmppc_subcore_exit_guest(); | |
3697 | ||
3698 | return trap; | |
3699 | } | |
3700 | ||
19ccb76a PM |
3701 | /* |
3702 | * Wait for some other vcpu thread to execute us, and | |
3703 | * wake us up when we need to handle something in the host. | |
3704 | */ | |
ec257165 PM |
3705 | static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc, |
3706 | struct kvm_vcpu *vcpu, int wait_state) | |
371fefd6 | 3707 | { |
371fefd6 PM |
3708 | DEFINE_WAIT(wait); |
3709 | ||
19ccb76a | 3710 | prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state); |
ec257165 PM |
3711 | if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { |
3712 | spin_unlock(&vc->lock); | |
19ccb76a | 3713 | schedule(); |
ec257165 PM |
3714 | spin_lock(&vc->lock); |
3715 | } | |
19ccb76a PM |
3716 | finish_wait(&vcpu->arch.cpu_run, &wait); |
3717 | } | |
3718 | ||
0cda69dd SJS |
3719 | static void grow_halt_poll_ns(struct kvmppc_vcore *vc) |
3720 | { | |
7fa08e71 NW |
3721 | if (!halt_poll_ns_grow) |
3722 | return; | |
3723 | ||
dee339b5 NW |
3724 | vc->halt_poll_ns *= halt_poll_ns_grow; |
3725 | if (vc->halt_poll_ns < halt_poll_ns_grow_start) | |
49113d36 | 3726 | vc->halt_poll_ns = halt_poll_ns_grow_start; |
0cda69dd SJS |
3727 | } |
3728 | ||
3729 | static void shrink_halt_poll_ns(struct kvmppc_vcore *vc) | |
3730 | { | |
3731 | if (halt_poll_ns_shrink == 0) | |
3732 | vc->halt_poll_ns = 0; | |
3733 | else | |
3734 | vc->halt_poll_ns /= halt_poll_ns_shrink; | |
3735 | } | |
3736 | ||
ee3308a2 PM |
3737 | #ifdef CONFIG_KVM_XICS |
3738 | static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu) | |
3739 | { | |
03f95332 | 3740 | if (!xics_on_xive()) |
ee3308a2 | 3741 | return false; |
2267ea76 | 3742 | return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr < |
ee3308a2 PM |
3743 | vcpu->arch.xive_saved_state.cppr; |
3744 | } | |
3745 | #else | |
3746 | static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu) | |
3747 | { | |
3748 | return false; | |
3749 | } | |
3750 | #endif /* CONFIG_KVM_XICS */ | |
3751 | ||
1da4e2f4 PM |
3752 | static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu) |
3753 | { | |
3754 | if (vcpu->arch.pending_exceptions || vcpu->arch.prodded || | |
ee3308a2 | 3755 | kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu)) |
1da4e2f4 PM |
3756 | return true; |
3757 | ||
3758 | return false; | |
3759 | } | |
3760 | ||
908a0935 SJS |
3761 | /* |
3762 | * Check to see if any of the runnable vcpus on the vcore have pending | |
0cda69dd SJS |
3763 | * exceptions or are no longer ceded |
3764 | */ | |
3765 | static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc) | |
3766 | { | |
3767 | struct kvm_vcpu *vcpu; | |
3768 | int i; | |
3769 | ||
3770 | for_each_runnable_thread(i, vcpu, vc) { | |
1da4e2f4 | 3771 | if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu)) |
0cda69dd SJS |
3772 | return 1; |
3773 | } | |
3774 | ||
3775 | return 0; | |
3776 | } | |
3777 | ||
19ccb76a PM |
3778 | /* |
3779 | * All the vcpus in this vcore are idle, so wait for a decrementer | |
3780 | * or external interrupt to one of the vcpus. vc->lock is held. | |
3781 | */ | |
3782 | static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc) | |
3783 | { | |
2a27f514 | 3784 | ktime_t cur, start_poll, start_wait; |
0cda69dd | 3785 | int do_sleep = 1; |
0cda69dd | 3786 | u64 block_ns; |
1bc5d59c | 3787 | |
0cda69dd | 3788 | /* Poll for pending exceptions and ceded state */ |
2a27f514 | 3789 | cur = start_poll = ktime_get(); |
0cda69dd | 3790 | if (vc->halt_poll_ns) { |
2a27f514 SJS |
3791 | ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns); |
3792 | ++vc->runner->stat.halt_attempted_poll; | |
1bc5d59c | 3793 | |
0cda69dd SJS |
3794 | vc->vcore_state = VCORE_POLLING; |
3795 | spin_unlock(&vc->lock); | |
3796 | ||
3797 | do { | |
3798 | if (kvmppc_vcore_check_block(vc)) { | |
3799 | do_sleep = 0; | |
3800 | break; | |
3801 | } | |
3802 | cur = ktime_get(); | |
3803 | } while (single_task_running() && ktime_before(cur, stop)); | |
3804 | ||
3805 | spin_lock(&vc->lock); | |
3806 | vc->vcore_state = VCORE_INACTIVE; | |
3807 | ||
2a27f514 SJS |
3808 | if (!do_sleep) { |
3809 | ++vc->runner->stat.halt_successful_poll; | |
0cda69dd | 3810 | goto out; |
2a27f514 | 3811 | } |
1bc5d59c SW |
3812 | } |
3813 | ||
da4ad88c DB |
3814 | prepare_to_rcuwait(&vc->wait); |
3815 | set_current_state(TASK_INTERRUPTIBLE); | |
0cda69dd | 3816 | if (kvmppc_vcore_check_block(vc)) { |
da4ad88c | 3817 | finish_rcuwait(&vc->wait); |
0cda69dd | 3818 | do_sleep = 0; |
2a27f514 SJS |
3819 | /* If we polled, count this as a successful poll */ |
3820 | if (vc->halt_poll_ns) | |
3821 | ++vc->runner->stat.halt_successful_poll; | |
0cda69dd | 3822 | goto out; |
1bc5d59c SW |
3823 | } |
3824 | ||
2a27f514 SJS |
3825 | start_wait = ktime_get(); |
3826 | ||
19ccb76a | 3827 | vc->vcore_state = VCORE_SLEEPING; |
3c78f78a | 3828 | trace_kvmppc_vcore_blocked(vc, 0); |
19ccb76a | 3829 | spin_unlock(&vc->lock); |
913d3ff9 | 3830 | schedule(); |
da4ad88c | 3831 | finish_rcuwait(&vc->wait); |
19ccb76a PM |
3832 | spin_lock(&vc->lock); |
3833 | vc->vcore_state = VCORE_INACTIVE; | |
3c78f78a | 3834 | trace_kvmppc_vcore_blocked(vc, 1); |
2a27f514 | 3835 | ++vc->runner->stat.halt_successful_wait; |
0cda69dd SJS |
3836 | |
3837 | cur = ktime_get(); | |
3838 | ||
3839 | out: | |
2a27f514 SJS |
3840 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll); |
3841 | ||
3842 | /* Attribute wait time */ | |
3843 | if (do_sleep) { | |
3844 | vc->runner->stat.halt_wait_ns += | |
3845 | ktime_to_ns(cur) - ktime_to_ns(start_wait); | |
3846 | /* Attribute failed poll time */ | |
3847 | if (vc->halt_poll_ns) | |
3848 | vc->runner->stat.halt_poll_fail_ns += | |
3849 | ktime_to_ns(start_wait) - | |
3850 | ktime_to_ns(start_poll); | |
3851 | } else { | |
3852 | /* Attribute successful poll time */ | |
3853 | if (vc->halt_poll_ns) | |
3854 | vc->runner->stat.halt_poll_success_ns += | |
3855 | ktime_to_ns(cur) - | |
3856 | ktime_to_ns(start_poll); | |
3857 | } | |
0cda69dd SJS |
3858 | |
3859 | /* Adjust poll time */ | |
307d93e4 | 3860 | if (halt_poll_ns) { |
0cda69dd SJS |
3861 | if (block_ns <= vc->halt_poll_ns) |
3862 | ; | |
3863 | /* We slept and blocked for longer than the max halt time */ | |
307d93e4 | 3864 | else if (vc->halt_poll_ns && block_ns > halt_poll_ns) |
0cda69dd SJS |
3865 | shrink_halt_poll_ns(vc); |
3866 | /* We slept and our poll time is too small */ | |
307d93e4 SJS |
3867 | else if (vc->halt_poll_ns < halt_poll_ns && |
3868 | block_ns < halt_poll_ns) | |
0cda69dd | 3869 | grow_halt_poll_ns(vc); |
e03f3921 SJS |
3870 | if (vc->halt_poll_ns > halt_poll_ns) |
3871 | vc->halt_poll_ns = halt_poll_ns; | |
0cda69dd SJS |
3872 | } else |
3873 | vc->halt_poll_ns = 0; | |
3874 | ||
3875 | trace_kvmppc_vcore_wakeup(do_sleep, block_ns); | |
19ccb76a | 3876 | } |
371fefd6 | 3877 | |
360cae31 PM |
3878 | /* |
3879 | * This never fails for a radix guest, as none of the operations it does | |
3880 | * for a radix guest can fail or have a way to report failure. | |
3881 | * kvmhv_run_single_vcpu() relies on this fact. | |
3882 | */ | |
432953b4 PM |
3883 | static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu) |
3884 | { | |
3885 | int r = 0; | |
3886 | struct kvm *kvm = vcpu->kvm; | |
3887 | ||
0d4ee88d | 3888 | mutex_lock(&kvm->arch.mmu_setup_lock); |
432953b4 PM |
3889 | if (!kvm->arch.mmu_ready) { |
3890 | if (!kvm_is_radix(kvm)) | |
3891 | r = kvmppc_hv_setup_htab_rma(vcpu); | |
3892 | if (!r) { | |
3893 | if (cpu_has_feature(CPU_FTR_ARCH_300)) | |
3894 | kvmppc_setup_partition_table(kvm); | |
3895 | kvm->arch.mmu_ready = 1; | |
3896 | } | |
3897 | } | |
0d4ee88d | 3898 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
432953b4 PM |
3899 | return r; |
3900 | } | |
3901 | ||
8c99d345 | 3902 | static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu) |
19ccb76a | 3903 | { |
8c99d345 | 3904 | struct kvm_run *run = vcpu->run; |
38c53af8 | 3905 | int n_ceded, i, r; |
19ccb76a | 3906 | struct kvmppc_vcore *vc; |
7b5f8272 | 3907 | struct kvm_vcpu *v; |
9e368f29 | 3908 | |
3c78f78a SW |
3909 | trace_kvmppc_run_vcpu_enter(vcpu); |
3910 | ||
8c99d345 | 3911 | run->exit_reason = 0; |
371fefd6 PM |
3912 | vcpu->arch.ret = RESUME_GUEST; |
3913 | vcpu->arch.trap = 0; | |
2f12f034 | 3914 | kvmppc_update_vpas(vcpu); |
371fefd6 | 3915 | |
371fefd6 PM |
3916 | /* |
3917 | * Synchronize with other threads in this virtual core | |
3918 | */ | |
3919 | vc = vcpu->arch.vcore; | |
3920 | spin_lock(&vc->lock); | |
19ccb76a | 3921 | vcpu->arch.ceded = 0; |
371fefd6 | 3922 | vcpu->arch.run_task = current; |
c7b67670 | 3923 | vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb()); |
19ccb76a | 3924 | vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; |
c7b67670 | 3925 | vcpu->arch.busy_preempt = TB_NIL; |
7b5f8272 | 3926 | WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu); |
371fefd6 PM |
3927 | ++vc->n_runnable; |
3928 | ||
19ccb76a PM |
3929 | /* |
3930 | * This happens the first time this is called for a vcpu. | |
3931 | * If the vcore is already running, we may be able to start | |
3932 | * this thread straight away and have it join in. | |
3933 | */ | |
8455d79e | 3934 | if (!signal_pending(current)) { |
c0093f1a PM |
3935 | if ((vc->vcore_state == VCORE_PIGGYBACK || |
3936 | vc->vcore_state == VCORE_RUNNING) && | |
ec257165 | 3937 | !VCORE_IS_EXITING(vc)) { |
2f12f034 | 3938 | kvmppc_create_dtl_entry(vcpu, vc); |
b4deba5c | 3939 | kvmppc_start_thread(vcpu, vc); |
3c78f78a | 3940 | trace_kvm_guest_enter(vcpu); |
8455d79e | 3941 | } else if (vc->vcore_state == VCORE_SLEEPING) { |
da4ad88c | 3942 | rcuwait_wake_up(&vc->wait); |
371fefd6 PM |
3943 | } |
3944 | ||
8455d79e | 3945 | } |
371fefd6 | 3946 | |
19ccb76a PM |
3947 | while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && |
3948 | !signal_pending(current)) { | |
072df813 PM |
3949 | /* See if the MMU is ready to go */ |
3950 | if (!vcpu->kvm->arch.mmu_ready) { | |
38c53af8 | 3951 | spin_unlock(&vc->lock); |
432953b4 | 3952 | r = kvmhv_setup_mmu(vcpu); |
38c53af8 PM |
3953 | spin_lock(&vc->lock); |
3954 | if (r) { | |
8c99d345 TZ |
3955 | run->exit_reason = KVM_EXIT_FAIL_ENTRY; |
3956 | run->fail_entry. | |
432953b4 | 3957 | hardware_entry_failure_reason = 0; |
38c53af8 PM |
3958 | vcpu->arch.ret = r; |
3959 | break; | |
3960 | } | |
3961 | } | |
3962 | ||
ec257165 PM |
3963 | if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL) |
3964 | kvmppc_vcore_end_preempt(vc); | |
3965 | ||
8455d79e | 3966 | if (vc->vcore_state != VCORE_INACTIVE) { |
ec257165 | 3967 | kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE); |
19ccb76a PM |
3968 | continue; |
3969 | } | |
7b5f8272 | 3970 | for_each_runnable_thread(i, v, vc) { |
7e28e60e | 3971 | kvmppc_core_prepare_to_enter(v); |
19ccb76a PM |
3972 | if (signal_pending(v->arch.run_task)) { |
3973 | kvmppc_remove_runnable(vc, v); | |
3974 | v->stat.signal_exits++; | |
2610a57f | 3975 | v->run->exit_reason = KVM_EXIT_INTR; |
19ccb76a PM |
3976 | v->arch.ret = -EINTR; |
3977 | wake_up(&v->arch.cpu_run); | |
3978 | } | |
3979 | } | |
8455d79e PM |
3980 | if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) |
3981 | break; | |
8455d79e | 3982 | n_ceded = 0; |
7b5f8272 | 3983 | for_each_runnable_thread(i, v, vc) { |
1da4e2f4 | 3984 | if (!kvmppc_vcpu_woken(v)) |
8455d79e | 3985 | n_ceded += v->arch.ceded; |
4619ac88 PM |
3986 | else |
3987 | v->arch.ceded = 0; | |
3988 | } | |
25fedfca PM |
3989 | vc->runner = vcpu; |
3990 | if (n_ceded == vc->n_runnable) { | |
8455d79e | 3991 | kvmppc_vcore_blocked(vc); |
c56dadf3 | 3992 | } else if (need_resched()) { |
ec257165 | 3993 | kvmppc_vcore_preempt(vc); |
25fedfca PM |
3994 | /* Let something else run */ |
3995 | cond_resched_lock(&vc->lock); | |
ec257165 PM |
3996 | if (vc->vcore_state == VCORE_PREEMPT) |
3997 | kvmppc_vcore_end_preempt(vc); | |
25fedfca | 3998 | } else { |
8455d79e | 3999 | kvmppc_run_core(vc); |
25fedfca | 4000 | } |
0456ec4f | 4001 | vc->runner = NULL; |
19ccb76a | 4002 | } |
371fefd6 | 4003 | |
8455d79e PM |
4004 | while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && |
4005 | (vc->vcore_state == VCORE_RUNNING || | |
5fc3e64f PM |
4006 | vc->vcore_state == VCORE_EXITING || |
4007 | vc->vcore_state == VCORE_PIGGYBACK)) | |
ec257165 | 4008 | kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE); |
8455d79e | 4009 | |
5fc3e64f PM |
4010 | if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL) |
4011 | kvmppc_vcore_end_preempt(vc); | |
4012 | ||
8455d79e PM |
4013 | if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { |
4014 | kvmppc_remove_runnable(vc, vcpu); | |
4015 | vcpu->stat.signal_exits++; | |
8c99d345 | 4016 | run->exit_reason = KVM_EXIT_INTR; |
8455d79e PM |
4017 | vcpu->arch.ret = -EINTR; |
4018 | } | |
4019 | ||
4020 | if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) { | |
4021 | /* Wake up some vcpu to run the core */ | |
7b5f8272 SJS |
4022 | i = -1; |
4023 | v = next_runnable_thread(vc, &i); | |
8455d79e | 4024 | wake_up(&v->arch.cpu_run); |
371fefd6 PM |
4025 | } |
4026 | ||
8c99d345 | 4027 | trace_kvmppc_run_vcpu_exit(vcpu); |
371fefd6 | 4028 | spin_unlock(&vc->lock); |
371fefd6 | 4029 | return vcpu->arch.ret; |
de56a948 PM |
4030 | } |
4031 | ||
8c99d345 | 4032 | int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit, |
360cae31 | 4033 | unsigned long lpcr) |
95a6432c | 4034 | { |
8c99d345 | 4035 | struct kvm_run *run = vcpu->run; |
9d0b048d | 4036 | int trap, r, pcpu; |
70ea13f6 | 4037 | int srcu_idx, lpid; |
95a6432c PM |
4038 | struct kvmppc_vcore *vc; |
4039 | struct kvm *kvm = vcpu->kvm; | |
360cae31 | 4040 | struct kvm_nested_guest *nested = vcpu->arch.nested; |
95a6432c PM |
4041 | |
4042 | trace_kvmppc_run_vcpu_enter(vcpu); | |
4043 | ||
8c99d345 | 4044 | run->exit_reason = 0; |
95a6432c PM |
4045 | vcpu->arch.ret = RESUME_GUEST; |
4046 | vcpu->arch.trap = 0; | |
4047 | ||
4048 | vc = vcpu->arch.vcore; | |
4049 | vcpu->arch.ceded = 0; | |
4050 | vcpu->arch.run_task = current; | |
95a6432c PM |
4051 | vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb()); |
4052 | vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; | |
4053 | vcpu->arch.busy_preempt = TB_NIL; | |
4054 | vcpu->arch.last_inst = KVM_INST_FETCH_FAILED; | |
4055 | vc->runnable_threads[0] = vcpu; | |
4056 | vc->n_runnable = 1; | |
4057 | vc->runner = vcpu; | |
4058 | ||
4059 | /* See if the MMU is ready to go */ | |
360cae31 PM |
4060 | if (!kvm->arch.mmu_ready) |
4061 | kvmhv_setup_mmu(vcpu); | |
95a6432c PM |
4062 | |
4063 | if (need_resched()) | |
4064 | cond_resched(); | |
4065 | ||
4066 | kvmppc_update_vpas(vcpu); | |
4067 | ||
4068 | init_vcore_to_run(vc); | |
4069 | vc->preempt_tb = TB_NIL; | |
4070 | ||
4071 | preempt_disable(); | |
4072 | pcpu = smp_processor_id(); | |
4073 | vc->pcpu = pcpu; | |
4074 | kvmppc_prepare_radix_vcpu(vcpu, pcpu); | |
4075 | ||
4076 | local_irq_disable(); | |
4077 | hard_irq_disable(); | |
4078 | if (signal_pending(current)) | |
4079 | goto sigpend; | |
4080 | if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready) | |
4081 | goto out; | |
4082 | ||
360cae31 PM |
4083 | if (!nested) { |
4084 | kvmppc_core_prepare_to_enter(vcpu); | |
4085 | if (vcpu->arch.doorbell_request) { | |
4086 | vc->dpdes = 1; | |
4087 | smp_wmb(); | |
4088 | vcpu->arch.doorbell_request = 0; | |
4089 | } | |
4090 | if (test_bit(BOOK3S_IRQPRIO_EXTERNAL, | |
4091 | &vcpu->arch.pending_exceptions)) | |
4092 | lpcr |= LPCR_MER; | |
4093 | } else if (vcpu->arch.pending_exceptions || | |
4094 | vcpu->arch.doorbell_request || | |
4095 | xive_interrupt_pending(vcpu)) { | |
4096 | vcpu->arch.ret = RESUME_HOST; | |
4097 | goto out; | |
4098 | } | |
95a6432c PM |
4099 | |
4100 | kvmppc_clear_host_core(pcpu); | |
4101 | ||
4102 | local_paca->kvm_hstate.tid = 0; | |
4103 | local_paca->kvm_hstate.napping = 0; | |
4104 | local_paca->kvm_hstate.kvm_split_mode = NULL; | |
4105 | kvmppc_start_thread(vcpu, vc); | |
4106 | kvmppc_create_dtl_entry(vcpu, vc); | |
4107 | trace_kvm_guest_enter(vcpu); | |
4108 | ||
4109 | vc->vcore_state = VCORE_RUNNING; | |
4110 | trace_kvmppc_run_core(vc, 0); | |
4111 | ||
70ea13f6 PM |
4112 | if (cpu_has_feature(CPU_FTR_HVMODE)) { |
4113 | lpid = nested ? nested->shadow_lpid : kvm->arch.lpid; | |
4114 | mtspr(SPRN_LPID, lpid); | |
4115 | isync(); | |
4116 | kvmppc_check_need_tlb_flush(kvm, pcpu, nested); | |
4117 | } | |
95a6432c | 4118 | |
95a6432c PM |
4119 | guest_enter_irqoff(); |
4120 | ||
4121 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
4122 | ||
4123 | this_cpu_disable_ftrace(); | |
4124 | ||
1b28d553 PM |
4125 | /* Tell lockdep that we're about to enable interrupts */ |
4126 | trace_hardirqs_on(); | |
4127 | ||
360cae31 | 4128 | trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr); |
95a6432c PM |
4129 | vcpu->arch.trap = trap; |
4130 | ||
1b28d553 PM |
4131 | trace_hardirqs_off(); |
4132 | ||
95a6432c PM |
4133 | this_cpu_enable_ftrace(); |
4134 | ||
4135 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
4136 | ||
f3c99f97 PM |
4137 | if (cpu_has_feature(CPU_FTR_HVMODE)) { |
4138 | mtspr(SPRN_LPID, kvm->arch.host_lpid); | |
4139 | isync(); | |
4140 | } | |
95a6432c | 4141 | |
95a6432c PM |
4142 | set_irq_happened(trap); |
4143 | ||
4144 | kvmppc_set_host_core(pcpu); | |
4145 | ||
4146 | local_irq_enable(); | |
4147 | guest_exit(); | |
4148 | ||
4149 | cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest); | |
4150 | ||
4151 | preempt_enable(); | |
4152 | ||
3c25ab35 SJS |
4153 | /* |
4154 | * cancel pending decrementer exception if DEC is now positive, or if | |
4155 | * entering a nested guest in which case the decrementer is now owned | |
4156 | * by L2 and the L1 decrementer is provided in hdec_expires | |
4157 | */ | |
4158 | if (kvmppc_core_pending_dec(vcpu) && | |
4159 | ((get_tb() < vcpu->arch.dec_expires) || | |
4160 | (trap == BOOK3S_INTERRUPT_SYSCALL && | |
4161 | kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED))) | |
95a6432c PM |
4162 | kvmppc_core_dequeue_dec(vcpu); |
4163 | ||
4164 | trace_kvm_guest_exit(vcpu); | |
4165 | r = RESUME_GUEST; | |
360cae31 PM |
4166 | if (trap) { |
4167 | if (!nested) | |
8c99d345 | 4168 | r = kvmppc_handle_exit_hv(vcpu, current); |
360cae31 | 4169 | else |
8c99d345 | 4170 | r = kvmppc_handle_nested_exit(vcpu); |
360cae31 | 4171 | } |
95a6432c PM |
4172 | vcpu->arch.ret = r; |
4173 | ||
4174 | if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded && | |
4175 | !kvmppc_vcpu_woken(vcpu)) { | |
4176 | kvmppc_set_timer(vcpu); | |
4177 | while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) { | |
4178 | if (signal_pending(current)) { | |
4179 | vcpu->stat.signal_exits++; | |
8c99d345 | 4180 | run->exit_reason = KVM_EXIT_INTR; |
95a6432c PM |
4181 | vcpu->arch.ret = -EINTR; |
4182 | break; | |
4183 | } | |
4184 | spin_lock(&vc->lock); | |
4185 | kvmppc_vcore_blocked(vc); | |
4186 | spin_unlock(&vc->lock); | |
4187 | } | |
4188 | } | |
4189 | vcpu->arch.ceded = 0; | |
4190 | ||
4191 | vc->vcore_state = VCORE_INACTIVE; | |
4192 | trace_kvmppc_run_core(vc, 1); | |
4193 | ||
4194 | done: | |
4195 | kvmppc_remove_runnable(vc, vcpu); | |
8c99d345 | 4196 | trace_kvmppc_run_vcpu_exit(vcpu); |
95a6432c PM |
4197 | |
4198 | return vcpu->arch.ret; | |
4199 | ||
4200 | sigpend: | |
4201 | vcpu->stat.signal_exits++; | |
8c99d345 | 4202 | run->exit_reason = KVM_EXIT_INTR; |
95a6432c PM |
4203 | vcpu->arch.ret = -EINTR; |
4204 | out: | |
4205 | local_irq_enable(); | |
4206 | preempt_enable(); | |
4207 | goto done; | |
4208 | } | |
4209 | ||
8c99d345 | 4210 | static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu) |
a8606e20 | 4211 | { |
8c99d345 | 4212 | struct kvm_run *run = vcpu->run; |
a8606e20 | 4213 | int r; |
913d3ff9 | 4214 | int srcu_idx; |
ca8efa1d | 4215 | unsigned long ebb_regs[3] = {}; /* shut up GCC */ |
4c3bb4cc PM |
4216 | unsigned long user_tar = 0; |
4217 | unsigned int user_vrsave; | |
1b151ce4 | 4218 | struct kvm *kvm; |
a8606e20 | 4219 | |
af8f38b3 AG |
4220 | if (!vcpu->arch.sane) { |
4221 | run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
4222 | return -EINVAL; | |
4223 | } | |
4224 | ||
46a704f8 PM |
4225 | /* |
4226 | * Don't allow entry with a suspended transaction, because | |
4227 | * the guest entry/exit code will lose it. | |
4228 | * If the guest has TM enabled, save away their TM-related SPRs | |
4229 | * (they will get restored by the TM unavailable interrupt). | |
4230 | */ | |
4231 | #ifdef CONFIG_PPC_TRANSACTIONAL_MEM | |
4232 | if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs && | |
4233 | (current->thread.regs->msr & MSR_TM)) { | |
4234 | if (MSR_TM_ACTIVE(current->thread.regs->msr)) { | |
4235 | run->exit_reason = KVM_EXIT_FAIL_ENTRY; | |
4236 | run->fail_entry.hardware_entry_failure_reason = 0; | |
4237 | return -EINVAL; | |
4238 | } | |
e4705715 PM |
4239 | /* Enable TM so we can read the TM SPRs */ |
4240 | mtmsr(mfmsr() | MSR_TM); | |
46a704f8 PM |
4241 | current->thread.tm_tfhar = mfspr(SPRN_TFHAR); |
4242 | current->thread.tm_tfiar = mfspr(SPRN_TFIAR); | |
4243 | current->thread.tm_texasr = mfspr(SPRN_TEXASR); | |
4244 | current->thread.regs->msr &= ~MSR_TM; | |
4245 | } | |
4246 | #endif | |
4247 | ||
7aa15842 PM |
4248 | /* |
4249 | * Force online to 1 for the sake of old userspace which doesn't | |
4250 | * set it. | |
4251 | */ | |
4252 | if (!vcpu->arch.online) { | |
4253 | atomic_inc(&vcpu->arch.vcore->online_count); | |
4254 | vcpu->arch.online = 1; | |
4255 | } | |
4256 | ||
25051b5a SW |
4257 | kvmppc_core_prepare_to_enter(vcpu); |
4258 | ||
19ccb76a PM |
4259 | /* No need to go into the guest when all we'll do is come back out */ |
4260 | if (signal_pending(current)) { | |
4261 | run->exit_reason = KVM_EXIT_INTR; | |
4262 | return -EINTR; | |
4263 | } | |
4264 | ||
1b151ce4 PM |
4265 | kvm = vcpu->kvm; |
4266 | atomic_inc(&kvm->arch.vcpus_running); | |
4267 | /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */ | |
32fad281 PM |
4268 | smp_mb(); |
4269 | ||
579e633e AB |
4270 | flush_all_to_thread(current); |
4271 | ||
4c3bb4cc | 4272 | /* Save userspace EBB and other register values */ |
ca8efa1d PM |
4273 | if (cpu_has_feature(CPU_FTR_ARCH_207S)) { |
4274 | ebb_regs[0] = mfspr(SPRN_EBBHR); | |
4275 | ebb_regs[1] = mfspr(SPRN_EBBRR); | |
4276 | ebb_regs[2] = mfspr(SPRN_BESCR); | |
4c3bb4cc | 4277 | user_tar = mfspr(SPRN_TAR); |
ca8efa1d | 4278 | } |
4c3bb4cc | 4279 | user_vrsave = mfspr(SPRN_VRSAVE); |
ca8efa1d | 4280 | |
da4ad88c | 4281 | vcpu->arch.waitp = &vcpu->arch.vcore->wait; |
8a9c8925 | 4282 | vcpu->arch.pgdir = kvm->mm->pgd; |
c7b67670 | 4283 | vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; |
19ccb76a | 4284 | |
a8606e20 | 4285 | do { |
8d9fcacf PM |
4286 | /* |
4287 | * The early POWER9 chips that can't mix radix and HPT threads | |
4288 | * on the same core also need the workaround for the problem | |
4289 | * where the TLB would prefetch entries in the guest exit path | |
4290 | * for radix guests using the guest PIDR value and LPID 0. | |
4291 | * The workaround is in the old path (kvmppc_run_vcpu()) | |
4292 | * but not the new path (kvmhv_run_single_vcpu()). | |
4293 | */ | |
4294 | if (kvm->arch.threads_indep && kvm_is_radix(kvm) && | |
4295 | !no_mixing_hpt_and_radix) | |
8c99d345 | 4296 | r = kvmhv_run_single_vcpu(vcpu, ~(u64)0, |
360cae31 | 4297 | vcpu->arch.vcore->lpcr); |
95a6432c | 4298 | else |
8c99d345 | 4299 | r = kvmppc_run_vcpu(vcpu); |
a8606e20 PM |
4300 | |
4301 | if (run->exit_reason == KVM_EXIT_PAPR_HCALL && | |
4302 | !(vcpu->arch.shregs.msr & MSR_PR)) { | |
3c78f78a | 4303 | trace_kvm_hcall_enter(vcpu); |
a8606e20 | 4304 | r = kvmppc_pseries_do_hcall(vcpu); |
3c78f78a | 4305 | trace_kvm_hcall_exit(vcpu, r); |
7e28e60e | 4306 | kvmppc_core_prepare_to_enter(vcpu); |
913d3ff9 | 4307 | } else if (r == RESUME_PAGE_FAULT) { |
432953b4 | 4308 | srcu_idx = srcu_read_lock(&kvm->srcu); |
8c99d345 | 4309 | r = kvmppc_book3s_hv_page_fault(vcpu, |
913d3ff9 | 4310 | vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); |
432953b4 | 4311 | srcu_read_unlock(&kvm->srcu, srcu_idx); |
5af50993 | 4312 | } else if (r == RESUME_PASSTHROUGH) { |
03f95332 | 4313 | if (WARN_ON(xics_on_xive())) |
5af50993 BH |
4314 | r = H_SUCCESS; |
4315 | else | |
4316 | r = kvmppc_xics_rm_complete(vcpu, 0); | |
4317 | } | |
e59d24e6 | 4318 | } while (is_kvmppc_resume_guest(r)); |
32fad281 | 4319 | |
4c3bb4cc | 4320 | /* Restore userspace EBB and other register values */ |
ca8efa1d PM |
4321 | if (cpu_has_feature(CPU_FTR_ARCH_207S)) { |
4322 | mtspr(SPRN_EBBHR, ebb_regs[0]); | |
4323 | mtspr(SPRN_EBBRR, ebb_regs[1]); | |
4324 | mtspr(SPRN_BESCR, ebb_regs[2]); | |
4c3bb4cc PM |
4325 | mtspr(SPRN_TAR, user_tar); |
4326 | mtspr(SPRN_FSCR, current->thread.fscr); | |
ca8efa1d | 4327 | } |
4c3bb4cc | 4328 | mtspr(SPRN_VRSAVE, user_vrsave); |
ca8efa1d | 4329 | |
c7b67670 | 4330 | vcpu->arch.state = KVMPPC_VCPU_NOTREADY; |
432953b4 | 4331 | atomic_dec(&kvm->arch.vcpus_running); |
a8606e20 PM |
4332 | return r; |
4333 | } | |
4334 | ||
5b74716e | 4335 | static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps, |
8dc6cca5 | 4336 | int shift, int sllp) |
5b74716e | 4337 | { |
8dc6cca5 PM |
4338 | (*sps)->page_shift = shift; |
4339 | (*sps)->slb_enc = sllp; | |
4340 | (*sps)->enc[0].page_shift = shift; | |
4341 | (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift); | |
1f365bb0 | 4342 | /* |
8dc6cca5 | 4343 | * Add 16MB MPSS support (may get filtered out by userspace) |
1f365bb0 | 4344 | */ |
8dc6cca5 PM |
4345 | if (shift != 24) { |
4346 | int penc = kvmppc_pgsize_lp_encoding(shift, 24); | |
4347 | if (penc != -1) { | |
4348 | (*sps)->enc[1].page_shift = 24; | |
4349 | (*sps)->enc[1].pte_enc = penc; | |
4350 | } | |
1f365bb0 | 4351 | } |
5b74716e BH |
4352 | (*sps)++; |
4353 | } | |
4354 | ||
3a167bea AK |
4355 | static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm, |
4356 | struct kvm_ppc_smmu_info *info) | |
5b74716e BH |
4357 | { |
4358 | struct kvm_ppc_one_seg_page_size *sps; | |
4359 | ||
e3bfed1d PM |
4360 | /* |
4361 | * POWER7, POWER8 and POWER9 all support 32 storage keys for data. | |
4362 | * POWER7 doesn't support keys for instruction accesses, | |
4363 | * POWER8 and POWER9 do. | |
4364 | */ | |
4365 | info->data_keys = 32; | |
4366 | info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0; | |
4367 | ||
8dc6cca5 PM |
4368 | /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */ |
4369 | info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS; | |
4370 | info->slb_size = 32; | |
5b74716e BH |
4371 | |
4372 | /* We only support these sizes for now, and no muti-size segments */ | |
4373 | sps = &info->sps[0]; | |
8dc6cca5 PM |
4374 | kvmppc_add_seg_page_size(&sps, 12, 0); |
4375 | kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01); | |
4376 | kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L); | |
5b74716e | 4377 | |
901f8c3f PM |
4378 | /* If running as a nested hypervisor, we don't support HPT guests */ |
4379 | if (kvmhv_on_pseries()) | |
4380 | info->flags |= KVM_PPC_NO_HASH; | |
4381 | ||
5b74716e BH |
4382 | return 0; |
4383 | } | |
4384 | ||
82ed3616 PM |
4385 | /* |
4386 | * Get (and clear) the dirty memory log for a memory slot. | |
4387 | */ | |
3a167bea AK |
4388 | static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm, |
4389 | struct kvm_dirty_log *log) | |
82ed3616 | 4390 | { |
9f6b8029 | 4391 | struct kvm_memslots *slots; |
82ed3616 | 4392 | struct kvm_memory_slot *memslot; |
8f7b79b8 | 4393 | int i, r; |
82ed3616 | 4394 | unsigned long n; |
e641a317 | 4395 | unsigned long *buf, *p; |
8f7b79b8 | 4396 | struct kvm_vcpu *vcpu; |
82ed3616 PM |
4397 | |
4398 | mutex_lock(&kvm->slots_lock); | |
4399 | ||
4400 | r = -EINVAL; | |
bbacc0c1 | 4401 | if (log->slot >= KVM_USER_MEM_SLOTS) |
82ed3616 PM |
4402 | goto out; |
4403 | ||
9f6b8029 PB |
4404 | slots = kvm_memslots(kvm); |
4405 | memslot = id_to_memslot(slots, log->slot); | |
82ed3616 | 4406 | r = -ENOENT; |
0577d1ab | 4407 | if (!memslot || !memslot->dirty_bitmap) |
82ed3616 PM |
4408 | goto out; |
4409 | ||
8f7b79b8 | 4410 | /* |
e641a317 PM |
4411 | * Use second half of bitmap area because both HPT and radix |
4412 | * accumulate bits in the first half. | |
8f7b79b8 | 4413 | */ |
82ed3616 | 4414 | n = kvm_dirty_bitmap_bytes(memslot); |
8f7b79b8 PM |
4415 | buf = memslot->dirty_bitmap + n / sizeof(long); |
4416 | memset(buf, 0, n); | |
82ed3616 | 4417 | |
8f7b79b8 PM |
4418 | if (kvm_is_radix(kvm)) |
4419 | r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf); | |
4420 | else | |
4421 | r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf); | |
82ed3616 PM |
4422 | if (r) |
4423 | goto out; | |
4424 | ||
e641a317 PM |
4425 | /* |
4426 | * We accumulate dirty bits in the first half of the | |
4427 | * memslot's dirty_bitmap area, for when pages are paged | |
4428 | * out or modified by the host directly. Pick up these | |
4429 | * bits and add them to the map. | |
4430 | */ | |
4431 | p = memslot->dirty_bitmap; | |
4432 | for (i = 0; i < n / sizeof(long); ++i) | |
4433 | buf[i] |= xchg(&p[i], 0); | |
4434 | ||
8f7b79b8 PM |
4435 | /* Harvest dirty bits from VPA and DTL updates */ |
4436 | /* Note: we never modify the SLB shadow buffer areas */ | |
4437 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
4438 | spin_lock(&vcpu->arch.vpa_update_lock); | |
4439 | kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf); | |
4440 | kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf); | |
4441 | spin_unlock(&vcpu->arch.vpa_update_lock); | |
4442 | } | |
4443 | ||
82ed3616 | 4444 | r = -EFAULT; |
8f7b79b8 | 4445 | if (copy_to_user(log->dirty_bitmap, buf, n)) |
82ed3616 PM |
4446 | goto out; |
4447 | ||
4448 | r = 0; | |
4449 | out: | |
4450 | mutex_unlock(&kvm->slots_lock); | |
4451 | return r; | |
4452 | } | |
4453 | ||
e96c81ee | 4454 | static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot) |
a66b48c3 | 4455 | { |
e96c81ee SC |
4456 | vfree(slot->arch.rmap); |
4457 | slot->arch.rmap = NULL; | |
a66b48c3 PM |
4458 | } |
4459 | ||
82307e67 SC |
4460 | static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm, |
4461 | struct kvm_memory_slot *slot, | |
4462 | const struct kvm_userspace_memory_region *mem, | |
4463 | enum kvm_mr_change change) | |
a66b48c3 | 4464 | { |
82307e67 | 4465 | unsigned long npages = mem->memory_size >> PAGE_SHIFT; |
aa04b4cc | 4466 | |
82307e67 SC |
4467 | if (change == KVM_MR_CREATE) { |
4468 | slot->arch.rmap = vzalloc(array_size(npages, | |
4469 | sizeof(*slot->arch.rmap))); | |
4470 | if (!slot->arch.rmap) | |
4471 | return -ENOMEM; | |
4472 | } | |
aa04b4cc | 4473 | |
a66b48c3 | 4474 | return 0; |
c77162de PM |
4475 | } |
4476 | ||
3a167bea | 4477 | static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm, |
09170a49 | 4478 | const struct kvm_userspace_memory_region *mem, |
f36f3f28 | 4479 | const struct kvm_memory_slot *old, |
f032b734 BR |
4480 | const struct kvm_memory_slot *new, |
4481 | enum kvm_mr_change change) | |
c77162de | 4482 | { |
dfe49dbd | 4483 | unsigned long npages = mem->memory_size >> PAGE_SHIFT; |
dfe49dbd | 4484 | |
a56ee9f8 YX |
4485 | /* |
4486 | * If we are making a new memslot, it might make | |
4487 | * some address that was previously cached as emulated | |
4488 | * MMIO be no longer emulated MMIO, so invalidate | |
4489 | * all the caches of emulated MMIO translations. | |
4490 | */ | |
4491 | if (npages) | |
4492 | atomic64_inc(&kvm->arch.mmio_update); | |
5af3e9d0 PM |
4493 | |
4494 | /* | |
4495 | * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels | |
4496 | * have already called kvm_arch_flush_shadow_memslot() to | |
4497 | * flush shadow mappings. For KVM_MR_CREATE we have no | |
4498 | * previous mappings. So the only case to handle is | |
4499 | * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit | |
4500 | * has been changed. | |
4501 | * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES | |
4502 | * to get rid of any THP PTEs in the partition-scoped page tables | |
4503 | * so we can track dirtiness at the page level; we flush when | |
4504 | * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to | |
4505 | * using THP PTEs. | |
4506 | */ | |
4507 | if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) && | |
4508 | ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES)) | |
4509 | kvmppc_radix_flush_memslot(kvm, old); | |
c3262257 BR |
4510 | /* |
4511 | * If UV hasn't yet called H_SVM_INIT_START, don't register memslots. | |
4512 | */ | |
4513 | if (!kvm->arch.secure_guest) | |
4514 | return; | |
4515 | ||
4516 | switch (change) { | |
4517 | case KVM_MR_CREATE: | |
4518 | if (kvmppc_uvmem_slot_init(kvm, new)) | |
4519 | return; | |
4520 | uv_register_mem_slot(kvm->arch.lpid, | |
4521 | new->base_gfn << PAGE_SHIFT, | |
4522 | new->npages * PAGE_SIZE, | |
4523 | 0, new->id); | |
4524 | break; | |
4525 | case KVM_MR_DELETE: | |
4526 | uv_unregister_mem_slot(kvm->arch.lpid, old->id); | |
4527 | kvmppc_uvmem_slot_free(kvm, old); | |
4528 | break; | |
4529 | default: | |
4530 | /* TODO: Handle KVM_MR_MOVE */ | |
4531 | break; | |
4532 | } | |
c77162de PM |
4533 | } |
4534 | ||
a0144e2a PM |
4535 | /* |
4536 | * Update LPCR values in kvm->arch and in vcores. | |
0d4ee88d PM |
4537 | * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion |
4538 | * of kvm->arch.lpcr update). | |
a0144e2a PM |
4539 | */ |
4540 | void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask) | |
4541 | { | |
4542 | long int i; | |
4543 | u32 cores_done = 0; | |
4544 | ||
4545 | if ((kvm->arch.lpcr & mask) == lpcr) | |
4546 | return; | |
4547 | ||
4548 | kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr; | |
4549 | ||
4550 | for (i = 0; i < KVM_MAX_VCORES; ++i) { | |
4551 | struct kvmppc_vcore *vc = kvm->arch.vcores[i]; | |
4552 | if (!vc) | |
4553 | continue; | |
4554 | spin_lock(&vc->lock); | |
4555 | vc->lpcr = (vc->lpcr & ~mask) | lpcr; | |
4556 | spin_unlock(&vc->lock); | |
4557 | if (++cores_done >= kvm->arch.online_vcores) | |
4558 | break; | |
4559 | } | |
4560 | } | |
4561 | ||
ded13fc1 | 4562 | void kvmppc_setup_partition_table(struct kvm *kvm) |
7a84084c PM |
4563 | { |
4564 | unsigned long dw0, dw1; | |
4565 | ||
8cf4ecc0 PM |
4566 | if (!kvm_is_radix(kvm)) { |
4567 | /* PS field - page size for VRMA */ | |
4568 | dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) | | |
4569 | ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1); | |
4570 | /* HTABSIZE and HTABORG fields */ | |
4571 | dw0 |= kvm->arch.sdr1; | |
7a84084c | 4572 | |
8cf4ecc0 PM |
4573 | /* Second dword as set by userspace */ |
4574 | dw1 = kvm->arch.process_table; | |
4575 | } else { | |
4576 | dw0 = PATB_HR | radix__get_tree_size() | | |
4577 | __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE; | |
4578 | dw1 = PATB_GR | kvm->arch.process_table; | |
4579 | } | |
8e3f5fc1 | 4580 | kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1); |
7a84084c PM |
4581 | } |
4582 | ||
1b151ce4 PM |
4583 | /* |
4584 | * Set up HPT (hashed page table) and RMA (real-mode area). | |
0d4ee88d | 4585 | * Must be called with kvm->arch.mmu_setup_lock held. |
1b151ce4 | 4586 | */ |
32fad281 | 4587 | static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu) |
c77162de PM |
4588 | { |
4589 | int err = 0; | |
4590 | struct kvm *kvm = vcpu->kvm; | |
c77162de PM |
4591 | unsigned long hva; |
4592 | struct kvm_memory_slot *memslot; | |
4593 | struct vm_area_struct *vma; | |
a0144e2a | 4594 | unsigned long lpcr = 0, senc; |
c77162de | 4595 | unsigned long psize, porder; |
2c9097e4 | 4596 | int srcu_idx; |
c77162de | 4597 | |
32fad281 | 4598 | /* Allocate hashed page table (if not done already) and reset it */ |
3f9d4f5a | 4599 | if (!kvm->arch.hpt.virt) { |
aae0777f DG |
4600 | int order = KVM_DEFAULT_HPT_ORDER; |
4601 | struct kvm_hpt_info info; | |
4602 | ||
4603 | err = kvmppc_allocate_hpt(&info, order); | |
4604 | /* If we get here, it means userspace didn't specify a | |
4605 | * size explicitly. So, try successively smaller | |
4606 | * sizes if the default failed. */ | |
4607 | while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER) | |
4608 | err = kvmppc_allocate_hpt(&info, order); | |
4609 | ||
4610 | if (err < 0) { | |
32fad281 PM |
4611 | pr_err("KVM: Couldn't alloc HPT\n"); |
4612 | goto out; | |
4613 | } | |
aae0777f DG |
4614 | |
4615 | kvmppc_set_hpt(kvm, &info); | |
32fad281 PM |
4616 | } |
4617 | ||
c77162de | 4618 | /* Look up the memslot for guest physical address 0 */ |
2c9097e4 | 4619 | srcu_idx = srcu_read_lock(&kvm->srcu); |
c77162de | 4620 | memslot = gfn_to_memslot(kvm, 0); |
aa04b4cc | 4621 | |
c77162de PM |
4622 | /* We must have some memory at 0 by now */ |
4623 | err = -EINVAL; | |
4624 | if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) | |
2c9097e4 | 4625 | goto out_srcu; |
c77162de PM |
4626 | |
4627 | /* Look up the VMA for the start of this memory slot */ | |
4628 | hva = memslot->userspace_addr; | |
d8ed45c5 | 4629 | mmap_read_lock(kvm->mm); |
8a9c8925 | 4630 | vma = find_vma(kvm->mm, hva); |
c77162de PM |
4631 | if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO)) |
4632 | goto up_out; | |
4633 | ||
4634 | psize = vma_kernel_pagesize(vma); | |
c77162de | 4635 | |
d8ed45c5 | 4636 | mmap_read_unlock(kvm->mm); |
c77162de | 4637 | |
c17b98cf | 4638 | /* We can handle 4k, 64k or 16M pages in the VRMA */ |
debd574f PM |
4639 | if (psize >= 0x1000000) |
4640 | psize = 0x1000000; | |
4641 | else if (psize >= 0x10000) | |
4642 | psize = 0x10000; | |
4643 | else | |
4644 | psize = 0x1000; | |
4645 | porder = __ilog2(psize); | |
c77162de | 4646 | |
c17b98cf PM |
4647 | senc = slb_pgsize_encoding(psize); |
4648 | kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T | | |
4649 | (VRMA_VSID << SLB_VSID_SHIFT_1T); | |
c17b98cf PM |
4650 | /* Create HPTEs in the hash page table for the VRMA */ |
4651 | kvmppc_map_vrma(vcpu, memslot, porder); | |
aa04b4cc | 4652 | |
7a84084c PM |
4653 | /* Update VRMASD field in the LPCR */ |
4654 | if (!cpu_has_feature(CPU_FTR_ARCH_300)) { | |
4655 | /* the -4 is to account for senc values starting at 0x10 */ | |
4656 | lpcr = senc << (LPCR_VRMASD_SH - 4); | |
4657 | kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD); | |
7a84084c | 4658 | } |
a0144e2a | 4659 | |
1b151ce4 | 4660 | /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */ |
c77162de | 4661 | smp_wmb(); |
c77162de | 4662 | err = 0; |
2c9097e4 PM |
4663 | out_srcu: |
4664 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
c77162de | 4665 | out: |
c77162de | 4666 | return err; |
b2b2f165 | 4667 | |
c77162de | 4668 | up_out: |
d8ed45c5 | 4669 | mmap_read_unlock(kvm->mm); |
505d6421 | 4670 | goto out_srcu; |
de56a948 PM |
4671 | } |
4672 | ||
0d4ee88d PM |
4673 | /* |
4674 | * Must be called with kvm->arch.mmu_setup_lock held and | |
4675 | * mmu_ready = 0 and no vcpus running. | |
4676 | */ | |
18c3640c PM |
4677 | int kvmppc_switch_mmu_to_hpt(struct kvm *kvm) |
4678 | { | |
aa069a99 | 4679 | if (nesting_enabled(kvm)) |
8e3f5fc1 | 4680 | kvmhv_release_all_nested(kvm); |
234ff0b7 PM |
4681 | kvmppc_rmap_reset(kvm); |
4682 | kvm->arch.process_table = 0; | |
4683 | /* Mutual exclusion with kvm_unmap_hva_range etc. */ | |
4684 | spin_lock(&kvm->mmu_lock); | |
4685 | kvm->arch.radix = 0; | |
4686 | spin_unlock(&kvm->mmu_lock); | |
18c3640c PM |
4687 | kvmppc_free_radix(kvm); |
4688 | kvmppc_update_lpcr(kvm, LPCR_VPM1, | |
4689 | LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR); | |
18c3640c PM |
4690 | return 0; |
4691 | } | |
4692 | ||
0d4ee88d PM |
4693 | /* |
4694 | * Must be called with kvm->arch.mmu_setup_lock held and | |
4695 | * mmu_ready = 0 and no vcpus running. | |
4696 | */ | |
18c3640c PM |
4697 | int kvmppc_switch_mmu_to_radix(struct kvm *kvm) |
4698 | { | |
4699 | int err; | |
4700 | ||
4701 | err = kvmppc_init_vm_radix(kvm); | |
4702 | if (err) | |
4703 | return err; | |
234ff0b7 PM |
4704 | kvmppc_rmap_reset(kvm); |
4705 | /* Mutual exclusion with kvm_unmap_hva_range etc. */ | |
4706 | spin_lock(&kvm->mmu_lock); | |
4707 | kvm->arch.radix = 1; | |
4708 | spin_unlock(&kvm->mmu_lock); | |
18c3640c PM |
4709 | kvmppc_free_hpt(&kvm->arch.hpt); |
4710 | kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR, | |
4711 | LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR); | |
18c3640c PM |
4712 | return 0; |
4713 | } | |
4714 | ||
79b6c247 SW |
4715 | #ifdef CONFIG_KVM_XICS |
4716 | /* | |
4717 | * Allocate a per-core structure for managing state about which cores are | |
4718 | * running in the host versus the guest and for exchanging data between | |
4719 | * real mode KVM and CPU running in the host. | |
4720 | * This is only done for the first VM. | |
4721 | * The allocated structure stays even if all VMs have stopped. | |
4722 | * It is only freed when the kvm-hv module is unloaded. | |
4723 | * It's OK for this routine to fail, we just don't support host | |
4724 | * core operations like redirecting H_IPI wakeups. | |
4725 | */ | |
4726 | void kvmppc_alloc_host_rm_ops(void) | |
4727 | { | |
4728 | struct kvmppc_host_rm_ops *ops; | |
4729 | unsigned long l_ops; | |
4730 | int cpu, core; | |
4731 | int size; | |
4732 | ||
4733 | /* Not the first time here ? */ | |
4734 | if (kvmppc_host_rm_ops_hv != NULL) | |
4735 | return; | |
4736 | ||
4737 | ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL); | |
4738 | if (!ops) | |
4739 | return; | |
4740 | ||
4741 | size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core); | |
4742 | ops->rm_core = kzalloc(size, GFP_KERNEL); | |
4743 | ||
4744 | if (!ops->rm_core) { | |
4745 | kfree(ops); | |
4746 | return; | |
4747 | } | |
4748 | ||
419af25f | 4749 | cpus_read_lock(); |
6f3bb809 | 4750 | |
79b6c247 SW |
4751 | for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) { |
4752 | if (!cpu_online(cpu)) | |
4753 | continue; | |
4754 | ||
4755 | core = cpu >> threads_shift; | |
4756 | ops->rm_core[core].rm_state.in_host = 1; | |
4757 | } | |
4758 | ||
0c2a6606 SW |
4759 | ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv; |
4760 | ||
79b6c247 SW |
4761 | /* |
4762 | * Make the contents of the kvmppc_host_rm_ops structure visible | |
4763 | * to other CPUs before we assign it to the global variable. | |
4764 | * Do an atomic assignment (no locks used here), but if someone | |
4765 | * beats us to it, just free our copy and return. | |
4766 | */ | |
4767 | smp_wmb(); | |
4768 | l_ops = (unsigned long) ops; | |
4769 | ||
4770 | if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) { | |
419af25f | 4771 | cpus_read_unlock(); |
79b6c247 SW |
4772 | kfree(ops->rm_core); |
4773 | kfree(ops); | |
6f3bb809 | 4774 | return; |
79b6c247 | 4775 | } |
6f3bb809 | 4776 | |
419af25f SAS |
4777 | cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE, |
4778 | "ppc/kvm_book3s:prepare", | |
4779 | kvmppc_set_host_core, | |
4780 | kvmppc_clear_host_core); | |
4781 | cpus_read_unlock(); | |
79b6c247 SW |
4782 | } |
4783 | ||
4784 | void kvmppc_free_host_rm_ops(void) | |
4785 | { | |
4786 | if (kvmppc_host_rm_ops_hv) { | |
3f7cd919 | 4787 | cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE); |
79b6c247 SW |
4788 | kfree(kvmppc_host_rm_ops_hv->rm_core); |
4789 | kfree(kvmppc_host_rm_ops_hv); | |
4790 | kvmppc_host_rm_ops_hv = NULL; | |
4791 | } | |
4792 | } | |
4793 | #endif | |
4794 | ||
3a167bea | 4795 | static int kvmppc_core_init_vm_hv(struct kvm *kvm) |
de56a948 | 4796 | { |
32fad281 | 4797 | unsigned long lpcr, lpid; |
e23a808b | 4798 | char buf[32]; |
8cf4ecc0 | 4799 | int ret; |
de56a948 | 4800 | |
ca9f4942 BR |
4801 | mutex_init(&kvm->arch.uvmem_lock); |
4802 | INIT_LIST_HEAD(&kvm->arch.uvmem_pfns); | |
0d4ee88d PM |
4803 | mutex_init(&kvm->arch.mmu_setup_lock); |
4804 | ||
32fad281 PM |
4805 | /* Allocate the guest's logical partition ID */ |
4806 | ||
4807 | lpid = kvmppc_alloc_lpid(); | |
5d226ae5 | 4808 | if ((long)lpid < 0) |
32fad281 PM |
4809 | return -ENOMEM; |
4810 | kvm->arch.lpid = lpid; | |
de56a948 | 4811 | |
79b6c247 SW |
4812 | kvmppc_alloc_host_rm_ops(); |
4813 | ||
8e3f5fc1 PM |
4814 | kvmhv_vm_nested_init(kvm); |
4815 | ||
1b400ba0 PM |
4816 | /* |
4817 | * Since we don't flush the TLB when tearing down a VM, | |
4818 | * and this lpid might have previously been used, | |
4819 | * make sure we flush on each core before running the new VM. | |
7c5b06ca PM |
4820 | * On POWER9, the tlbie in mmu_partition_table_set_entry() |
4821 | * does this flush for us. | |
1b400ba0 | 4822 | */ |
7c5b06ca PM |
4823 | if (!cpu_has_feature(CPU_FTR_ARCH_300)) |
4824 | cpumask_setall(&kvm->arch.need_tlb_flush); | |
1b400ba0 | 4825 | |
699a0ea0 PM |
4826 | /* Start out with the default set of hcalls enabled */ |
4827 | memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls, | |
4828 | sizeof(kvm->arch.enabled_hcalls)); | |
4829 | ||
7a84084c PM |
4830 | if (!cpu_has_feature(CPU_FTR_ARCH_300)) |
4831 | kvm->arch.host_sdr1 = mfspr(SPRN_SDR1); | |
aa04b4cc | 4832 | |
c17b98cf | 4833 | /* Init LPCR for virtual RMA mode */ |
f3c99f97 PM |
4834 | if (cpu_has_feature(CPU_FTR_HVMODE)) { |
4835 | kvm->arch.host_lpid = mfspr(SPRN_LPID); | |
4836 | kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR); | |
4837 | lpcr &= LPCR_PECE | LPCR_LPES; | |
4838 | } else { | |
4839 | lpcr = 0; | |
4840 | } | |
c17b98cf PM |
4841 | lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE | |
4842 | LPCR_VPM0 | LPCR_VPM1; | |
4843 | kvm->arch.vrma_slb_v = SLB_VSID_B_1T | | |
4844 | (VRMA_VSID << SLB_VSID_SHIFT_1T); | |
4845 | /* On POWER8 turn on online bit to enable PURR/SPURR */ | |
4846 | if (cpu_has_feature(CPU_FTR_ARCH_207S)) | |
4847 | lpcr |= LPCR_ONL; | |
84f7139c PM |
4848 | /* |
4849 | * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed) | |
4850 | * Set HVICE bit to enable hypervisor virtualization interrupts. | |
5af50993 BH |
4851 | * Set HEIC to prevent OS interrupts to go to hypervisor (should |
4852 | * be unnecessary but better safe than sorry in case we re-enable | |
4853 | * EE in HV mode with this LPCR still set) | |
84f7139c PM |
4854 | */ |
4855 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { | |
7a84084c | 4856 | lpcr &= ~LPCR_VPM0; |
5af50993 BH |
4857 | lpcr |= LPCR_HVICE | LPCR_HEIC; |
4858 | ||
4859 | /* | |
4860 | * If xive is enabled, we route 0x500 interrupts directly | |
4861 | * to the guest. | |
4862 | */ | |
03f95332 | 4863 | if (xics_on_xive()) |
5af50993 | 4864 | lpcr |= LPCR_LPES; |
84f7139c PM |
4865 | } |
4866 | ||
8cf4ecc0 | 4867 | /* |
18c3640c | 4868 | * If the host uses radix, the guest starts out as radix. |
8cf4ecc0 PM |
4869 | */ |
4870 | if (radix_enabled()) { | |
4871 | kvm->arch.radix = 1; | |
1b151ce4 | 4872 | kvm->arch.mmu_ready = 1; |
8cf4ecc0 PM |
4873 | lpcr &= ~LPCR_VPM1; |
4874 | lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR; | |
4875 | ret = kvmppc_init_vm_radix(kvm); | |
4876 | if (ret) { | |
4877 | kvmppc_free_lpid(kvm->arch.lpid); | |
4878 | return ret; | |
4879 | } | |
4880 | kvmppc_setup_partition_table(kvm); | |
4881 | } | |
4882 | ||
9e368f29 | 4883 | kvm->arch.lpcr = lpcr; |
aa04b4cc | 4884 | |
5e985969 DG |
4885 | /* Initialization for future HPT resizes */ |
4886 | kvm->arch.resize_hpt = NULL; | |
4887 | ||
7c5b06ca PM |
4888 | /* |
4889 | * Work out how many sets the TLB has, for the use of | |
4890 | * the TLB invalidation loop in book3s_hv_rmhandlers.S. | |
4891 | */ | |
18c3640c | 4892 | if (radix_enabled()) |
8cf4ecc0 PM |
4893 | kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */ |
4894 | else if (cpu_has_feature(CPU_FTR_ARCH_300)) | |
7c5b06ca PM |
4895 | kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */ |
4896 | else if (cpu_has_feature(CPU_FTR_ARCH_207S)) | |
4897 | kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */ | |
4898 | else | |
4899 | kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */ | |
4900 | ||
512691d4 | 4901 | /* |
441c19c8 ME |
4902 | * Track that we now have a HV mode VM active. This blocks secondary |
4903 | * CPU threads from coming online. | |
516f7898 PM |
4904 | * On POWER9, we only need to do this if the "indep_threads_mode" |
4905 | * module parameter has been set to N. | |
512691d4 | 4906 | */ |
360cae31 PM |
4907 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { |
4908 | if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) { | |
4909 | pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n"); | |
4910 | kvm->arch.threads_indep = true; | |
4911 | } else { | |
4912 | kvm->arch.threads_indep = indep_threads_mode; | |
4913 | } | |
4914 | } | |
516f7898 | 4915 | if (!kvm->arch.threads_indep) |
8cf4ecc0 | 4916 | kvm_hv_vm_activated(); |
512691d4 | 4917 | |
3c313524 PM |
4918 | /* |
4919 | * Initialize smt_mode depending on processor. | |
4920 | * POWER8 and earlier have to use "strict" threading, where | |
4921 | * all vCPUs in a vcore have to run on the same (sub)core, | |
4922 | * whereas on POWER9 the threads can each run a different | |
4923 | * guest. | |
4924 | */ | |
4925 | if (!cpu_has_feature(CPU_FTR_ARCH_300)) | |
4926 | kvm->arch.smt_mode = threads_per_subcore; | |
4927 | else | |
4928 | kvm->arch.smt_mode = 1; | |
57900694 | 4929 | kvm->arch.emul_smt_mode = 1; |
3c313524 | 4930 | |
e23a808b PM |
4931 | /* |
4932 | * Create a debugfs directory for the VM | |
4933 | */ | |
4934 | snprintf(buf, sizeof(buf), "vm%d", current->pid); | |
4935 | kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir); | |
929f45e3 | 4936 | kvmppc_mmu_debugfs_init(kvm); |
9a94d3ee PM |
4937 | if (radix_enabled()) |
4938 | kvmhv_radix_debugfs_init(kvm); | |
e23a808b | 4939 | |
54738c09 | 4940 | return 0; |
de56a948 PM |
4941 | } |
4942 | ||
f1378b1c PM |
4943 | static void kvmppc_free_vcores(struct kvm *kvm) |
4944 | { | |
4945 | long int i; | |
4946 | ||
23316316 | 4947 | for (i = 0; i < KVM_MAX_VCORES; ++i) |
f1378b1c PM |
4948 | kfree(kvm->arch.vcores[i]); |
4949 | kvm->arch.online_vcores = 0; | |
4950 | } | |
4951 | ||
3a167bea | 4952 | static void kvmppc_core_destroy_vm_hv(struct kvm *kvm) |
de56a948 | 4953 | { |
e23a808b PM |
4954 | debugfs_remove_recursive(kvm->arch.debugfs_dir); |
4955 | ||
516f7898 | 4956 | if (!kvm->arch.threads_indep) |
8cf4ecc0 | 4957 | kvm_hv_vm_deactivated(); |
512691d4 | 4958 | |
f1378b1c | 4959 | kvmppc_free_vcores(kvm); |
aa04b4cc | 4960 | |
8cf4ecc0 | 4961 | |
5a319350 PM |
4962 | if (kvm_is_radix(kvm)) |
4963 | kvmppc_free_radix(kvm); | |
4964 | else | |
aae0777f | 4965 | kvmppc_free_hpt(&kvm->arch.hpt); |
c57875f5 | 4966 | |
89329c0b SJS |
4967 | /* Perform global invalidation and return lpid to the pool */ |
4968 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { | |
aa069a99 | 4969 | if (nesting_enabled(kvm)) |
8e3f5fc1 | 4970 | kvmhv_release_all_nested(kvm); |
89329c0b | 4971 | kvm->arch.process_table = 0; |
d89c69f4 PM |
4972 | if (kvm->arch.secure_guest) |
4973 | uv_svm_terminate(kvm->arch.lpid); | |
8e3f5fc1 | 4974 | kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0); |
89329c0b | 4975 | } |
ca9f4942 | 4976 | |
89329c0b SJS |
4977 | kvmppc_free_lpid(kvm->arch.lpid); |
4978 | ||
c57875f5 | 4979 | kvmppc_free_pimap(kvm); |
de56a948 PM |
4980 | } |
4981 | ||
3a167bea | 4982 | /* We don't need to emulate any privileged instructions or dcbz */ |
8c99d345 | 4983 | static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu, |
3a167bea | 4984 | unsigned int inst, int *advance) |
de56a948 | 4985 | { |
3a167bea | 4986 | return EMULATE_FAIL; |
de56a948 PM |
4987 | } |
4988 | ||
3a167bea AK |
4989 | static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn, |
4990 | ulong spr_val) | |
de56a948 PM |
4991 | { |
4992 | return EMULATE_FAIL; | |
4993 | } | |
4994 | ||
3a167bea AK |
4995 | static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn, |
4996 | ulong *spr_val) | |
de56a948 PM |
4997 | { |
4998 | return EMULATE_FAIL; | |
4999 | } | |
5000 | ||
3a167bea | 5001 | static int kvmppc_core_check_processor_compat_hv(void) |
de56a948 | 5002 | { |
de760db4 PM |
5003 | if (cpu_has_feature(CPU_FTR_HVMODE) && |
5004 | cpu_has_feature(CPU_FTR_ARCH_206)) | |
5005 | return 0; | |
50de596d | 5006 | |
de760db4 PM |
5007 | /* POWER9 in radix mode is capable of being a nested hypervisor. */ |
5008 | if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled()) | |
5009 | return 0; | |
5010 | ||
5011 | return -EIO; | |
de56a948 PM |
5012 | } |
5013 | ||
8daaafc8 SW |
5014 | #ifdef CONFIG_KVM_XICS |
5015 | ||
5016 | void kvmppc_free_pimap(struct kvm *kvm) | |
5017 | { | |
5018 | kfree(kvm->arch.pimap); | |
5019 | } | |
5020 | ||
c57875f5 | 5021 | static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void) |
8daaafc8 SW |
5022 | { |
5023 | return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL); | |
5024 | } | |
c57875f5 SW |
5025 | |
5026 | static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi) | |
5027 | { | |
5028 | struct irq_desc *desc; | |
5029 | struct kvmppc_irq_map *irq_map; | |
5030 | struct kvmppc_passthru_irqmap *pimap; | |
5031 | struct irq_chip *chip; | |
5af50993 | 5032 | int i, rc = 0; |
c57875f5 | 5033 | |
644abbb2 SW |
5034 | if (!kvm_irq_bypass) |
5035 | return 1; | |
5036 | ||
c57875f5 SW |
5037 | desc = irq_to_desc(host_irq); |
5038 | if (!desc) | |
5039 | return -EIO; | |
5040 | ||
5041 | mutex_lock(&kvm->lock); | |
5042 | ||
5043 | pimap = kvm->arch.pimap; | |
5044 | if (pimap == NULL) { | |
5045 | /* First call, allocate structure to hold IRQ map */ | |
5046 | pimap = kvmppc_alloc_pimap(); | |
5047 | if (pimap == NULL) { | |
5048 | mutex_unlock(&kvm->lock); | |
5049 | return -ENOMEM; | |
5050 | } | |
5051 | kvm->arch.pimap = pimap; | |
5052 | } | |
5053 | ||
5054 | /* | |
5055 | * For now, we only support interrupts for which the EOI operation | |
5056 | * is an OPAL call followed by a write to XIRR, since that's | |
5af50993 | 5057 | * what our real-mode EOI code does, or a XIVE interrupt |
c57875f5 SW |
5058 | */ |
5059 | chip = irq_data_get_irq_chip(&desc->irq_data); | |
5af50993 | 5060 | if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) { |
c57875f5 SW |
5061 | pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n", |
5062 | host_irq, guest_gsi); | |
5063 | mutex_unlock(&kvm->lock); | |
5064 | return -ENOENT; | |
5065 | } | |
5066 | ||
5067 | /* | |
5068 | * See if we already have an entry for this guest IRQ number. | |
5069 | * If it's mapped to a hardware IRQ number, that's an error, | |
5070 | * otherwise re-use this entry. | |
5071 | */ | |
5072 | for (i = 0; i < pimap->n_mapped; i++) { | |
5073 | if (guest_gsi == pimap->mapped[i].v_hwirq) { | |
5074 | if (pimap->mapped[i].r_hwirq) { | |
5075 | mutex_unlock(&kvm->lock); | |
5076 | return -EINVAL; | |
5077 | } | |
5078 | break; | |
5079 | } | |
5080 | } | |
5081 | ||
5082 | if (i == KVMPPC_PIRQ_MAPPED) { | |
5083 | mutex_unlock(&kvm->lock); | |
5084 | return -EAGAIN; /* table is full */ | |
5085 | } | |
5086 | ||
5087 | irq_map = &pimap->mapped[i]; | |
5088 | ||
5089 | irq_map->v_hwirq = guest_gsi; | |
c57875f5 SW |
5090 | irq_map->desc = desc; |
5091 | ||
e3c13e56 SW |
5092 | /* |
5093 | * Order the above two stores before the next to serialize with | |
5094 | * the KVM real mode handler. | |
5095 | */ | |
5096 | smp_wmb(); | |
5097 | irq_map->r_hwirq = desc->irq_data.hwirq; | |
5098 | ||
c57875f5 SW |
5099 | if (i == pimap->n_mapped) |
5100 | pimap->n_mapped++; | |
5101 | ||
03f95332 | 5102 | if (xics_on_xive()) |
5af50993 BH |
5103 | rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc); |
5104 | else | |
5105 | kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq); | |
5106 | if (rc) | |
5107 | irq_map->r_hwirq = 0; | |
5d375199 | 5108 | |
c57875f5 SW |
5109 | mutex_unlock(&kvm->lock); |
5110 | ||
5111 | return 0; | |
5112 | } | |
5113 | ||
5114 | static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi) | |
5115 | { | |
5116 | struct irq_desc *desc; | |
5117 | struct kvmppc_passthru_irqmap *pimap; | |
5af50993 | 5118 | int i, rc = 0; |
c57875f5 | 5119 | |
644abbb2 SW |
5120 | if (!kvm_irq_bypass) |
5121 | return 0; | |
5122 | ||
c57875f5 SW |
5123 | desc = irq_to_desc(host_irq); |
5124 | if (!desc) | |
5125 | return -EIO; | |
5126 | ||
5127 | mutex_lock(&kvm->lock); | |
a1c52e1c ME |
5128 | if (!kvm->arch.pimap) |
5129 | goto unlock; | |
c57875f5 | 5130 | |
c57875f5 SW |
5131 | pimap = kvm->arch.pimap; |
5132 | ||
5133 | for (i = 0; i < pimap->n_mapped; i++) { | |
5134 | if (guest_gsi == pimap->mapped[i].v_hwirq) | |
5135 | break; | |
5136 | } | |
5137 | ||
5138 | if (i == pimap->n_mapped) { | |
5139 | mutex_unlock(&kvm->lock); | |
5140 | return -ENODEV; | |
5141 | } | |
5142 | ||
03f95332 | 5143 | if (xics_on_xive()) |
5af50993 BH |
5144 | rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc); |
5145 | else | |
5146 | kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq); | |
5d375199 | 5147 | |
5af50993 | 5148 | /* invalidate the entry (what do do on error from the above ?) */ |
c57875f5 SW |
5149 | pimap->mapped[i].r_hwirq = 0; |
5150 | ||
5151 | /* | |
5152 | * We don't free this structure even when the count goes to | |
5153 | * zero. The structure is freed when we destroy the VM. | |
5154 | */ | |
a1c52e1c | 5155 | unlock: |
c57875f5 | 5156 | mutex_unlock(&kvm->lock); |
5af50993 | 5157 | return rc; |
c57875f5 SW |
5158 | } |
5159 | ||
5160 | static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons, | |
5161 | struct irq_bypass_producer *prod) | |
5162 | { | |
5163 | int ret = 0; | |
5164 | struct kvm_kernel_irqfd *irqfd = | |
5165 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
5166 | ||
5167 | irqfd->producer = prod; | |
5168 | ||
5169 | ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi); | |
5170 | if (ret) | |
5171 | pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n", | |
5172 | prod->irq, irqfd->gsi, ret); | |
5173 | ||
5174 | return ret; | |
5175 | } | |
5176 | ||
5177 | static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons, | |
5178 | struct irq_bypass_producer *prod) | |
5179 | { | |
5180 | int ret; | |
5181 | struct kvm_kernel_irqfd *irqfd = | |
5182 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
5183 | ||
5184 | irqfd->producer = NULL; | |
5185 | ||
5186 | /* | |
5187 | * When producer of consumer is unregistered, we change back to | |
5188 | * default external interrupt handling mode - KVM real mode | |
5189 | * will switch back to host. | |
5190 | */ | |
5191 | ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi); | |
5192 | if (ret) | |
5193 | pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n", | |
5194 | prod->irq, irqfd->gsi, ret); | |
5195 | } | |
8daaafc8 SW |
5196 | #endif |
5197 | ||
3a167bea AK |
5198 | static long kvm_arch_vm_ioctl_hv(struct file *filp, |
5199 | unsigned int ioctl, unsigned long arg) | |
5200 | { | |
5201 | struct kvm *kvm __maybe_unused = filp->private_data; | |
5202 | void __user *argp = (void __user *)arg; | |
5203 | long r; | |
5204 | ||
5205 | switch (ioctl) { | |
5206 | ||
3a167bea AK |
5207 | case KVM_PPC_ALLOCATE_HTAB: { |
5208 | u32 htab_order; | |
5209 | ||
5210 | r = -EFAULT; | |
5211 | if (get_user(htab_order, (u32 __user *)argp)) | |
5212 | break; | |
f98a8bf9 | 5213 | r = kvmppc_alloc_reset_hpt(kvm, htab_order); |
3a167bea AK |
5214 | if (r) |
5215 | break; | |
3a167bea AK |
5216 | r = 0; |
5217 | break; | |
5218 | } | |
5219 | ||
5220 | case KVM_PPC_GET_HTAB_FD: { | |
5221 | struct kvm_get_htab_fd ghf; | |
5222 | ||
5223 | r = -EFAULT; | |
5224 | if (copy_from_user(&ghf, argp, sizeof(ghf))) | |
5225 | break; | |
5226 | r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf); | |
5227 | break; | |
5228 | } | |
5229 | ||
5e985969 DG |
5230 | case KVM_PPC_RESIZE_HPT_PREPARE: { |
5231 | struct kvm_ppc_resize_hpt rhpt; | |
5232 | ||
5233 | r = -EFAULT; | |
5234 | if (copy_from_user(&rhpt, argp, sizeof(rhpt))) | |
5235 | break; | |
5236 | ||
5237 | r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt); | |
5238 | break; | |
5239 | } | |
5240 | ||
5241 | case KVM_PPC_RESIZE_HPT_COMMIT: { | |
5242 | struct kvm_ppc_resize_hpt rhpt; | |
5243 | ||
5244 | r = -EFAULT; | |
5245 | if (copy_from_user(&rhpt, argp, sizeof(rhpt))) | |
5246 | break; | |
5247 | ||
5248 | r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt); | |
5249 | break; | |
5250 | } | |
5251 | ||
3a167bea AK |
5252 | default: |
5253 | r = -ENOTTY; | |
5254 | } | |
5255 | ||
5256 | return r; | |
5257 | } | |
5258 | ||
699a0ea0 PM |
5259 | /* |
5260 | * List of hcall numbers to enable by default. | |
5261 | * For compatibility with old userspace, we enable by default | |
5262 | * all hcalls that were implemented before the hcall-enabling | |
5263 | * facility was added. Note this list should not include H_RTAS. | |
5264 | */ | |
5265 | static unsigned int default_hcall_list[] = { | |
5266 | H_REMOVE, | |
5267 | H_ENTER, | |
5268 | H_READ, | |
5269 | H_PROTECT, | |
5270 | H_BULK_REMOVE, | |
5271 | H_GET_TCE, | |
5272 | H_PUT_TCE, | |
5273 | H_SET_DABR, | |
5274 | H_SET_XDABR, | |
5275 | H_CEDE, | |
5276 | H_PROD, | |
5277 | H_CONFER, | |
5278 | H_REGISTER_VPA, | |
5279 | #ifdef CONFIG_KVM_XICS | |
5280 | H_EOI, | |
5281 | H_CPPR, | |
5282 | H_IPI, | |
5283 | H_IPOLL, | |
5284 | H_XIRR, | |
5285 | H_XIRR_X, | |
5286 | #endif | |
5287 | 0 | |
5288 | }; | |
5289 | ||
5290 | static void init_default_hcalls(void) | |
5291 | { | |
5292 | int i; | |
ae2113a4 | 5293 | unsigned int hcall; |
699a0ea0 | 5294 | |
ae2113a4 PM |
5295 | for (i = 0; default_hcall_list[i]; ++i) { |
5296 | hcall = default_hcall_list[i]; | |
5297 | WARN_ON(!kvmppc_hcall_impl_hv(hcall)); | |
5298 | __set_bit(hcall / 4, default_enabled_hcalls); | |
5299 | } | |
699a0ea0 PM |
5300 | } |
5301 | ||
c9270132 PM |
5302 | static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg) |
5303 | { | |
468808bd | 5304 | unsigned long lpcr; |
8cf4ecc0 | 5305 | int radix; |
18c3640c | 5306 | int err; |
468808bd PM |
5307 | |
5308 | /* If not on a POWER9, reject it */ | |
5309 | if (!cpu_has_feature(CPU_FTR_ARCH_300)) | |
5310 | return -ENODEV; | |
5311 | ||
5312 | /* If any unknown flags set, reject it */ | |
5313 | if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE)) | |
5314 | return -EINVAL; | |
5315 | ||
468808bd | 5316 | /* GR (guest radix) bit in process_table field must match */ |
18c3640c | 5317 | radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX); |
8cf4ecc0 | 5318 | if (!!(cfg->process_table & PATB_GR) != radix) |
468808bd PM |
5319 | return -EINVAL; |
5320 | ||
5321 | /* Process table size field must be reasonable, i.e. <= 24 */ | |
5322 | if ((cfg->process_table & PRTS_MASK) > 24) | |
5323 | return -EINVAL; | |
5324 | ||
18c3640c PM |
5325 | /* We can change a guest to/from radix now, if the host is radix */ |
5326 | if (radix && !radix_enabled()) | |
5327 | return -EINVAL; | |
5328 | ||
de760db4 PM |
5329 | /* If we're a nested hypervisor, we currently only support radix */ |
5330 | if (kvmhv_on_pseries() && !radix) | |
5331 | return -EINVAL; | |
5332 | ||
0d4ee88d | 5333 | mutex_lock(&kvm->arch.mmu_setup_lock); |
18c3640c PM |
5334 | if (radix != kvm_is_radix(kvm)) { |
5335 | if (kvm->arch.mmu_ready) { | |
5336 | kvm->arch.mmu_ready = 0; | |
5337 | /* order mmu_ready vs. vcpus_running */ | |
5338 | smp_mb(); | |
5339 | if (atomic_read(&kvm->arch.vcpus_running)) { | |
5340 | kvm->arch.mmu_ready = 1; | |
5341 | err = -EBUSY; | |
5342 | goto out_unlock; | |
5343 | } | |
5344 | } | |
5345 | if (radix) | |
5346 | err = kvmppc_switch_mmu_to_radix(kvm); | |
5347 | else | |
5348 | err = kvmppc_switch_mmu_to_hpt(kvm); | |
5349 | if (err) | |
5350 | goto out_unlock; | |
5351 | } | |
5352 | ||
468808bd PM |
5353 | kvm->arch.process_table = cfg->process_table; |
5354 | kvmppc_setup_partition_table(kvm); | |
5355 | ||
5356 | lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0; | |
5357 | kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE); | |
18c3640c | 5358 | err = 0; |
468808bd | 5359 | |
18c3640c | 5360 | out_unlock: |
0d4ee88d | 5361 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
18c3640c | 5362 | return err; |
c9270132 PM |
5363 | } |
5364 | ||
aa069a99 PM |
5365 | static int kvmhv_enable_nested(struct kvm *kvm) |
5366 | { | |
5367 | if (!nested) | |
5368 | return -EPERM; | |
8d9fcacf | 5369 | if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix) |
aa069a99 PM |
5370 | return -ENODEV; |
5371 | ||
5372 | /* kvm == NULL means the caller is testing if the capability exists */ | |
5373 | if (kvm) | |
5374 | kvm->arch.nested_enable = true; | |
5375 | return 0; | |
5376 | } | |
5377 | ||
dceadcf9 SJS |
5378 | static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr, |
5379 | int size) | |
5380 | { | |
5381 | int rc = -EINVAL; | |
5382 | ||
5383 | if (kvmhv_vcpu_is_radix(vcpu)) { | |
5384 | rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size); | |
5385 | ||
5386 | if (rc > 0) | |
5387 | rc = -EINVAL; | |
5388 | } | |
5389 | ||
5390 | /* For now quadrants are the only way to access nested guest memory */ | |
5391 | if (rc && vcpu->arch.nested) | |
5392 | rc = -EAGAIN; | |
5393 | ||
5394 | return rc; | |
5395 | } | |
5396 | ||
5397 | static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr, | |
5398 | int size) | |
5399 | { | |
5400 | int rc = -EINVAL; | |
5401 | ||
5402 | if (kvmhv_vcpu_is_radix(vcpu)) { | |
5403 | rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size); | |
5404 | ||
5405 | if (rc > 0) | |
5406 | rc = -EINVAL; | |
5407 | } | |
5408 | ||
5409 | /* For now quadrants are the only way to access nested guest memory */ | |
5410 | if (rc && vcpu->arch.nested) | |
5411 | rc = -EAGAIN; | |
5412 | ||
5413 | return rc; | |
5414 | } | |
5415 | ||
22945688 BR |
5416 | static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa) |
5417 | { | |
5418 | unpin_vpa(kvm, vpa); | |
5419 | vpa->gpa = 0; | |
5420 | vpa->pinned_addr = NULL; | |
5421 | vpa->dirty = false; | |
5422 | vpa->update_pending = 0; | |
5423 | } | |
5424 | ||
9a5788c6 PM |
5425 | /* |
5426 | * Enable a guest to become a secure VM, or test whether | |
5427 | * that could be enabled. | |
5428 | * Called when the KVM_CAP_PPC_SECURE_GUEST capability is | |
5429 | * tested (kvm == NULL) or enabled (kvm != NULL). | |
5430 | */ | |
5431 | static int kvmhv_enable_svm(struct kvm *kvm) | |
5432 | { | |
5433 | if (!kvmppc_uvmem_available()) | |
5434 | return -EINVAL; | |
5435 | if (kvm) | |
5436 | kvm->arch.svm_enabled = 1; | |
5437 | return 0; | |
5438 | } | |
5439 | ||
22945688 BR |
5440 | /* |
5441 | * IOCTL handler to turn off secure mode of guest | |
5442 | * | |
5443 | * - Release all device pages | |
5444 | * - Issue ucall to terminate the guest on the UV side | |
5445 | * - Unpin the VPA pages. | |
5446 | * - Reinit the partition scoped page tables | |
5447 | */ | |
5448 | static int kvmhv_svm_off(struct kvm *kvm) | |
5449 | { | |
5450 | struct kvm_vcpu *vcpu; | |
5451 | int mmu_was_ready; | |
5452 | int srcu_idx; | |
5453 | int ret = 0; | |
5454 | int i; | |
5455 | ||
5456 | if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START)) | |
5457 | return ret; | |
5458 | ||
5459 | mutex_lock(&kvm->arch.mmu_setup_lock); | |
5460 | mmu_was_ready = kvm->arch.mmu_ready; | |
5461 | if (kvm->arch.mmu_ready) { | |
5462 | kvm->arch.mmu_ready = 0; | |
5463 | /* order mmu_ready vs. vcpus_running */ | |
5464 | smp_mb(); | |
5465 | if (atomic_read(&kvm->arch.vcpus_running)) { | |
5466 | kvm->arch.mmu_ready = 1; | |
5467 | ret = -EBUSY; | |
5468 | goto out; | |
5469 | } | |
5470 | } | |
5471 | ||
5472 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5473 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
5474 | struct kvm_memory_slot *memslot; | |
5475 | struct kvm_memslots *slots = __kvm_memslots(kvm, i); | |
5476 | ||
5477 | if (!slots) | |
5478 | continue; | |
5479 | ||
5480 | kvm_for_each_memslot(memslot, slots) { | |
ce477a7a | 5481 | kvmppc_uvmem_drop_pages(memslot, kvm, true); |
22945688 BR |
5482 | uv_unregister_mem_slot(kvm->arch.lpid, memslot->id); |
5483 | } | |
5484 | } | |
5485 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5486 | ||
5487 | ret = uv_svm_terminate(kvm->arch.lpid); | |
5488 | if (ret != U_SUCCESS) { | |
5489 | ret = -EINVAL; | |
5490 | goto out; | |
5491 | } | |
5492 | ||
5493 | /* | |
5494 | * When secure guest is reset, all the guest pages are sent | |
5495 | * to UV via UV_PAGE_IN before the non-boot vcpus get a | |
5496 | * chance to run and unpin their VPA pages. Unpinning of all | |
5497 | * VPA pages is done here explicitly so that VPA pages | |
5498 | * can be migrated to the secure side. | |
5499 | * | |
5500 | * This is required to for the secure SMP guest to reboot | |
5501 | * correctly. | |
5502 | */ | |
5503 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
5504 | spin_lock(&vcpu->arch.vpa_update_lock); | |
5505 | unpin_vpa_reset(kvm, &vcpu->arch.dtl); | |
5506 | unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow); | |
5507 | unpin_vpa_reset(kvm, &vcpu->arch.vpa); | |
5508 | spin_unlock(&vcpu->arch.vpa_update_lock); | |
5509 | } | |
5510 | ||
5511 | kvmppc_setup_partition_table(kvm); | |
5512 | kvm->arch.secure_guest = 0; | |
5513 | kvm->arch.mmu_ready = mmu_was_ready; | |
5514 | out: | |
5515 | mutex_unlock(&kvm->arch.mmu_setup_lock); | |
5516 | return ret; | |
5517 | } | |
5518 | ||
cbbc58d4 | 5519 | static struct kvmppc_ops kvm_ops_hv = { |
3a167bea AK |
5520 | .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv, |
5521 | .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv, | |
5522 | .get_one_reg = kvmppc_get_one_reg_hv, | |
5523 | .set_one_reg = kvmppc_set_one_reg_hv, | |
5524 | .vcpu_load = kvmppc_core_vcpu_load_hv, | |
5525 | .vcpu_put = kvmppc_core_vcpu_put_hv, | |
87a45e07 | 5526 | .inject_interrupt = kvmppc_inject_interrupt_hv, |
3a167bea AK |
5527 | .set_msr = kvmppc_set_msr_hv, |
5528 | .vcpu_run = kvmppc_vcpu_run_hv, | |
5529 | .vcpu_create = kvmppc_core_vcpu_create_hv, | |
5530 | .vcpu_free = kvmppc_core_vcpu_free_hv, | |
5531 | .check_requests = kvmppc_core_check_requests_hv, | |
5532 | .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv, | |
5533 | .flush_memslot = kvmppc_core_flush_memslot_hv, | |
5534 | .prepare_memory_region = kvmppc_core_prepare_memory_region_hv, | |
5535 | .commit_memory_region = kvmppc_core_commit_memory_region_hv, | |
3a167bea AK |
5536 | .unmap_hva_range = kvm_unmap_hva_range_hv, |
5537 | .age_hva = kvm_age_hva_hv, | |
5538 | .test_age_hva = kvm_test_age_hva_hv, | |
5539 | .set_spte_hva = kvm_set_spte_hva_hv, | |
3a167bea | 5540 | .free_memslot = kvmppc_core_free_memslot_hv, |
3a167bea AK |
5541 | .init_vm = kvmppc_core_init_vm_hv, |
5542 | .destroy_vm = kvmppc_core_destroy_vm_hv, | |
3a167bea AK |
5543 | .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv, |
5544 | .emulate_op = kvmppc_core_emulate_op_hv, | |
5545 | .emulate_mtspr = kvmppc_core_emulate_mtspr_hv, | |
5546 | .emulate_mfspr = kvmppc_core_emulate_mfspr_hv, | |
5547 | .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv, | |
5548 | .arch_vm_ioctl = kvm_arch_vm_ioctl_hv, | |
ae2113a4 | 5549 | .hcall_implemented = kvmppc_hcall_impl_hv, |
c57875f5 SW |
5550 | #ifdef CONFIG_KVM_XICS |
5551 | .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv, | |
5552 | .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv, | |
5553 | #endif | |
c9270132 PM |
5554 | .configure_mmu = kvmhv_configure_mmu, |
5555 | .get_rmmu_info = kvmhv_get_rmmu_info, | |
3c313524 | 5556 | .set_smt_mode = kvmhv_set_smt_mode, |
aa069a99 | 5557 | .enable_nested = kvmhv_enable_nested, |
dceadcf9 SJS |
5558 | .load_from_eaddr = kvmhv_load_from_eaddr, |
5559 | .store_to_eaddr = kvmhv_store_to_eaddr, | |
9a5788c6 | 5560 | .enable_svm = kvmhv_enable_svm, |
22945688 | 5561 | .svm_off = kvmhv_svm_off, |
3a167bea AK |
5562 | }; |
5563 | ||
fd7bacbc MS |
5564 | static int kvm_init_subcore_bitmap(void) |
5565 | { | |
5566 | int i, j; | |
5567 | int nr_cores = cpu_nr_cores(); | |
5568 | struct sibling_subcore_state *sibling_subcore_state; | |
5569 | ||
5570 | for (i = 0; i < nr_cores; i++) { | |
5571 | int first_cpu = i * threads_per_core; | |
5572 | int node = cpu_to_node(first_cpu); | |
5573 | ||
5574 | /* Ignore if it is already allocated. */ | |
d2e60075 | 5575 | if (paca_ptrs[first_cpu]->sibling_subcore_state) |
fd7bacbc MS |
5576 | continue; |
5577 | ||
5578 | sibling_subcore_state = | |
08434ab4 | 5579 | kzalloc_node(sizeof(struct sibling_subcore_state), |
fd7bacbc MS |
5580 | GFP_KERNEL, node); |
5581 | if (!sibling_subcore_state) | |
5582 | return -ENOMEM; | |
5583 | ||
fd7bacbc MS |
5584 | |
5585 | for (j = 0; j < threads_per_core; j++) { | |
5586 | int cpu = first_cpu + j; | |
5587 | ||
d2e60075 NP |
5588 | paca_ptrs[cpu]->sibling_subcore_state = |
5589 | sibling_subcore_state; | |
fd7bacbc MS |
5590 | } |
5591 | } | |
5592 | return 0; | |
5593 | } | |
5594 | ||
5a319350 PM |
5595 | static int kvmppc_radix_possible(void) |
5596 | { | |
5597 | return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled(); | |
5598 | } | |
5599 | ||
3a167bea | 5600 | static int kvmppc_book3s_init_hv(void) |
de56a948 PM |
5601 | { |
5602 | int r; | |
2275d7b5 NP |
5603 | |
5604 | if (!tlbie_capable) { | |
5605 | pr_err("KVM-HV: Host does not support TLBIE\n"); | |
5606 | return -ENODEV; | |
5607 | } | |
5608 | ||
cbbc58d4 AK |
5609 | /* |
5610 | * FIXME!! Do we need to check on all cpus ? | |
5611 | */ | |
5612 | r = kvmppc_core_check_processor_compat_hv(); | |
5613 | if (r < 0) | |
739e2425 | 5614 | return -ENODEV; |
de56a948 | 5615 | |
8e3f5fc1 PM |
5616 | r = kvmhv_nested_init(); |
5617 | if (r) | |
5618 | return r; | |
5619 | ||
fd7bacbc MS |
5620 | r = kvm_init_subcore_bitmap(); |
5621 | if (r) | |
5622 | return r; | |
5623 | ||
f725758b PM |
5624 | /* |
5625 | * We need a way of accessing the XICS interrupt controller, | |
d2e60075 | 5626 | * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or |
f725758b PM |
5627 | * indirectly, via OPAL. |
5628 | */ | |
5629 | #ifdef CONFIG_SMP | |
03f95332 | 5630 | if (!xics_on_xive() && !kvmhv_on_pseries() && |
f3c18e93 | 5631 | !local_paca->kvm_hstate.xics_phys) { |
f725758b PM |
5632 | struct device_node *np; |
5633 | ||
5634 | np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc"); | |
5635 | if (!np) { | |
5636 | pr_err("KVM-HV: Cannot determine method for accessing XICS\n"); | |
5637 | return -ENODEV; | |
5638 | } | |
51eaa08f NMG |
5639 | /* presence of intc confirmed - node can be dropped again */ |
5640 | of_node_put(np); | |
f725758b PM |
5641 | } |
5642 | #endif | |
5643 | ||
cbbc58d4 AK |
5644 | kvm_ops_hv.owner = THIS_MODULE; |
5645 | kvmppc_hv_ops = &kvm_ops_hv; | |
de56a948 | 5646 | |
699a0ea0 PM |
5647 | init_default_hcalls(); |
5648 | ||
ec257165 PM |
5649 | init_vcore_lists(); |
5650 | ||
cbbc58d4 | 5651 | r = kvmppc_mmu_hv_init(); |
5a319350 PM |
5652 | if (r) |
5653 | return r; | |
5654 | ||
5655 | if (kvmppc_radix_possible()) | |
5656 | r = kvmppc_radix_init(); | |
00608e1f PM |
5657 | |
5658 | /* | |
5659 | * POWER9 chips before version 2.02 can't have some threads in | |
5660 | * HPT mode and some in radix mode on the same core. | |
5661 | */ | |
5662 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { | |
5663 | unsigned int pvr = mfspr(SPRN_PVR); | |
5664 | if ((pvr >> 16) == PVR_POWER9 && | |
5665 | (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) || | |
5666 | ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101))) | |
5667 | no_mixing_hpt_and_radix = true; | |
5668 | } | |
5669 | ||
ca9f4942 BR |
5670 | r = kvmppc_uvmem_init(); |
5671 | if (r < 0) | |
5672 | pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r); | |
5673 | ||
de56a948 PM |
5674 | return r; |
5675 | } | |
5676 | ||
3a167bea | 5677 | static void kvmppc_book3s_exit_hv(void) |
de56a948 | 5678 | { |
ca9f4942 | 5679 | kvmppc_uvmem_free(); |
79b6c247 | 5680 | kvmppc_free_host_rm_ops(); |
5a319350 PM |
5681 | if (kvmppc_radix_possible()) |
5682 | kvmppc_radix_exit(); | |
cbbc58d4 | 5683 | kvmppc_hv_ops = NULL; |
8e3f5fc1 | 5684 | kvmhv_nested_exit(); |
de56a948 PM |
5685 | } |
5686 | ||
3a167bea AK |
5687 | module_init(kvmppc_book3s_init_hv); |
5688 | module_exit(kvmppc_book3s_exit_hv); | |
2ba9f0d8 | 5689 | MODULE_LICENSE("GPL"); |
398a76c6 AG |
5690 | MODULE_ALIAS_MISCDEV(KVM_MINOR); |
5691 | MODULE_ALIAS("devname:kvm"); |