]>
Commit | Line | Data |
---|---|---|
de56a948 PM |
1 | /* |
2 | * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com> | |
3 | * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved. | |
4 | * | |
5 | * Authors: | |
6 | * Paul Mackerras <paulus@au1.ibm.com> | |
7 | * Alexander Graf <agraf@suse.de> | |
8 | * Kevin Wolf <mail@kevin-wolf.de> | |
9 | * | |
10 | * Description: KVM functions specific to running on Book 3S | |
11 | * processors in hypervisor mode (specifically POWER7 and later). | |
12 | * | |
13 | * This file is derived from arch/powerpc/kvm/book3s.c, | |
14 | * by Alexander Graf <agraf@suse.de>. | |
15 | * | |
16 | * This program is free software; you can redistribute it and/or modify | |
17 | * it under the terms of the GNU General Public License, version 2, as | |
18 | * published by the Free Software Foundation. | |
19 | */ | |
20 | ||
21 | #include <linux/kvm_host.h> | |
22 | #include <linux/err.h> | |
23 | #include <linux/slab.h> | |
24 | #include <linux/preempt.h> | |
25 | #include <linux/sched.h> | |
26 | #include <linux/delay.h> | |
66b15db6 | 27 | #include <linux/export.h> |
de56a948 PM |
28 | #include <linux/fs.h> |
29 | #include <linux/anon_inodes.h> | |
30 | #include <linux/cpumask.h> | |
aa04b4cc PM |
31 | #include <linux/spinlock.h> |
32 | #include <linux/page-flags.h> | |
de56a948 PM |
33 | |
34 | #include <asm/reg.h> | |
35 | #include <asm/cputable.h> | |
36 | #include <asm/cacheflush.h> | |
37 | #include <asm/tlbflush.h> | |
38 | #include <asm/uaccess.h> | |
39 | #include <asm/io.h> | |
40 | #include <asm/kvm_ppc.h> | |
41 | #include <asm/kvm_book3s.h> | |
42 | #include <asm/mmu_context.h> | |
43 | #include <asm/lppaca.h> | |
44 | #include <asm/processor.h> | |
371fefd6 | 45 | #include <asm/cputhreads.h> |
aa04b4cc | 46 | #include <asm/page.h> |
de1d9248 | 47 | #include <asm/hvcall.h> |
ae3a197e | 48 | #include <asm/switch_to.h> |
de56a948 | 49 | #include <linux/gfp.h> |
de56a948 PM |
50 | #include <linux/vmalloc.h> |
51 | #include <linux/highmem.h> | |
c77162de | 52 | #include <linux/hugetlb.h> |
de56a948 PM |
53 | |
54 | /* #define EXIT_DEBUG */ | |
55 | /* #define EXIT_DEBUG_SIMPLE */ | |
56 | /* #define EXIT_DEBUG_INT */ | |
57 | ||
19ccb76a | 58 | static void kvmppc_end_cede(struct kvm_vcpu *vcpu); |
c77162de | 59 | static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu); |
19ccb76a | 60 | |
de56a948 PM |
61 | void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
62 | { | |
63 | local_paca->kvm_hstate.kvm_vcpu = vcpu; | |
371fefd6 | 64 | local_paca->kvm_hstate.kvm_vcore = vcpu->arch.vcore; |
de56a948 PM |
65 | } |
66 | ||
67 | void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu) | |
68 | { | |
69 | } | |
70 | ||
de56a948 PM |
71 | void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr) |
72 | { | |
73 | vcpu->arch.shregs.msr = msr; | |
19ccb76a | 74 | kvmppc_end_cede(vcpu); |
de56a948 PM |
75 | } |
76 | ||
77 | void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr) | |
78 | { | |
79 | vcpu->arch.pvr = pvr; | |
80 | } | |
81 | ||
82 | void kvmppc_dump_regs(struct kvm_vcpu *vcpu) | |
83 | { | |
84 | int r; | |
85 | ||
86 | pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id); | |
87 | pr_err("pc = %.16lx msr = %.16llx trap = %x\n", | |
88 | vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap); | |
89 | for (r = 0; r < 16; ++r) | |
90 | pr_err("r%2d = %.16lx r%d = %.16lx\n", | |
91 | r, kvmppc_get_gpr(vcpu, r), | |
92 | r+16, kvmppc_get_gpr(vcpu, r+16)); | |
93 | pr_err("ctr = %.16lx lr = %.16lx\n", | |
94 | vcpu->arch.ctr, vcpu->arch.lr); | |
95 | pr_err("srr0 = %.16llx srr1 = %.16llx\n", | |
96 | vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1); | |
97 | pr_err("sprg0 = %.16llx sprg1 = %.16llx\n", | |
98 | vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1); | |
99 | pr_err("sprg2 = %.16llx sprg3 = %.16llx\n", | |
100 | vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3); | |
101 | pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n", | |
102 | vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr); | |
103 | pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar); | |
104 | pr_err("fault dar = %.16lx dsisr = %.8x\n", | |
105 | vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); | |
106 | pr_err("SLB (%d entries):\n", vcpu->arch.slb_max); | |
107 | for (r = 0; r < vcpu->arch.slb_max; ++r) | |
108 | pr_err(" ESID = %.16llx VSID = %.16llx\n", | |
109 | vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv); | |
110 | pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n", | |
aa04b4cc | 111 | vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1, |
de56a948 PM |
112 | vcpu->arch.last_inst); |
113 | } | |
114 | ||
a8606e20 PM |
115 | struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id) |
116 | { | |
117 | int r; | |
118 | struct kvm_vcpu *v, *ret = NULL; | |
119 | ||
120 | mutex_lock(&kvm->lock); | |
121 | kvm_for_each_vcpu(r, v, kvm) { | |
122 | if (v->vcpu_id == id) { | |
123 | ret = v; | |
124 | break; | |
125 | } | |
126 | } | |
127 | mutex_unlock(&kvm->lock); | |
128 | return ret; | |
129 | } | |
130 | ||
131 | static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa) | |
132 | { | |
133 | vpa->shared_proc = 1; | |
134 | vpa->yield_count = 1; | |
135 | } | |
136 | ||
137 | static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu, | |
138 | unsigned long flags, | |
139 | unsigned long vcpuid, unsigned long vpa) | |
140 | { | |
141 | struct kvm *kvm = vcpu->kvm; | |
93e60249 | 142 | unsigned long len, nb; |
a8606e20 PM |
143 | void *va; |
144 | struct kvm_vcpu *tvcpu; | |
93e60249 | 145 | int err = H_PARAMETER; |
a8606e20 PM |
146 | |
147 | tvcpu = kvmppc_find_vcpu(kvm, vcpuid); | |
148 | if (!tvcpu) | |
149 | return H_PARAMETER; | |
150 | ||
151 | flags >>= 63 - 18; | |
152 | flags &= 7; | |
153 | if (flags == 0 || flags == 4) | |
154 | return H_PARAMETER; | |
155 | if (flags < 4) { | |
156 | if (vpa & 0x7f) | |
157 | return H_PARAMETER; | |
93e60249 PM |
158 | if (flags >= 2 && !tvcpu->arch.vpa) |
159 | return H_RESOURCE; | |
a8606e20 | 160 | /* registering new area; convert logical addr to real */ |
93e60249 PM |
161 | va = kvmppc_pin_guest_page(kvm, vpa, &nb); |
162 | if (va == NULL) | |
b2b2f165 | 163 | return H_PARAMETER; |
a8606e20 PM |
164 | if (flags <= 1) |
165 | len = *(unsigned short *)(va + 4); | |
166 | else | |
167 | len = *(unsigned int *)(va + 4); | |
93e60249 PM |
168 | if (len > nb) |
169 | goto out_unpin; | |
a8606e20 PM |
170 | switch (flags) { |
171 | case 1: /* register VPA */ | |
172 | if (len < 640) | |
93e60249 PM |
173 | goto out_unpin; |
174 | if (tvcpu->arch.vpa) | |
175 | kvmppc_unpin_guest_page(kvm, vcpu->arch.vpa); | |
a8606e20 PM |
176 | tvcpu->arch.vpa = va; |
177 | init_vpa(vcpu, va); | |
178 | break; | |
179 | case 2: /* register DTL */ | |
180 | if (len < 48) | |
93e60249 | 181 | goto out_unpin; |
a8606e20 | 182 | len -= len % 48; |
93e60249 PM |
183 | if (tvcpu->arch.dtl) |
184 | kvmppc_unpin_guest_page(kvm, vcpu->arch.dtl); | |
a8606e20 PM |
185 | tvcpu->arch.dtl = va; |
186 | tvcpu->arch.dtl_end = va + len; | |
187 | break; | |
188 | case 3: /* register SLB shadow buffer */ | |
93e60249 PM |
189 | if (len < 16) |
190 | goto out_unpin; | |
191 | if (tvcpu->arch.slb_shadow) | |
192 | kvmppc_unpin_guest_page(kvm, vcpu->arch.slb_shadow); | |
a8606e20 PM |
193 | tvcpu->arch.slb_shadow = va; |
194 | break; | |
195 | } | |
196 | } else { | |
197 | switch (flags) { | |
198 | case 5: /* unregister VPA */ | |
199 | if (tvcpu->arch.slb_shadow || tvcpu->arch.dtl) | |
200 | return H_RESOURCE; | |
93e60249 PM |
201 | if (!tvcpu->arch.vpa) |
202 | break; | |
203 | kvmppc_unpin_guest_page(kvm, tvcpu->arch.vpa); | |
a8606e20 PM |
204 | tvcpu->arch.vpa = NULL; |
205 | break; | |
206 | case 6: /* unregister DTL */ | |
93e60249 PM |
207 | if (!tvcpu->arch.dtl) |
208 | break; | |
209 | kvmppc_unpin_guest_page(kvm, tvcpu->arch.dtl); | |
a8606e20 PM |
210 | tvcpu->arch.dtl = NULL; |
211 | break; | |
212 | case 7: /* unregister SLB shadow buffer */ | |
93e60249 PM |
213 | if (!tvcpu->arch.slb_shadow) |
214 | break; | |
215 | kvmppc_unpin_guest_page(kvm, tvcpu->arch.slb_shadow); | |
a8606e20 PM |
216 | tvcpu->arch.slb_shadow = NULL; |
217 | break; | |
218 | } | |
219 | } | |
220 | return H_SUCCESS; | |
93e60249 PM |
221 | |
222 | out_unpin: | |
223 | kvmppc_unpin_guest_page(kvm, va); | |
224 | return err; | |
a8606e20 PM |
225 | } |
226 | ||
227 | int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu) | |
228 | { | |
229 | unsigned long req = kvmppc_get_gpr(vcpu, 3); | |
230 | unsigned long target, ret = H_SUCCESS; | |
231 | struct kvm_vcpu *tvcpu; | |
232 | ||
233 | switch (req) { | |
c77162de PM |
234 | case H_ENTER: |
235 | ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4), | |
236 | kvmppc_get_gpr(vcpu, 5), | |
237 | kvmppc_get_gpr(vcpu, 6), | |
238 | kvmppc_get_gpr(vcpu, 7)); | |
239 | break; | |
a8606e20 | 240 | case H_CEDE: |
a8606e20 PM |
241 | break; |
242 | case H_PROD: | |
243 | target = kvmppc_get_gpr(vcpu, 4); | |
244 | tvcpu = kvmppc_find_vcpu(vcpu->kvm, target); | |
245 | if (!tvcpu) { | |
246 | ret = H_PARAMETER; | |
247 | break; | |
248 | } | |
249 | tvcpu->arch.prodded = 1; | |
250 | smp_mb(); | |
251 | if (vcpu->arch.ceded) { | |
252 | if (waitqueue_active(&vcpu->wq)) { | |
253 | wake_up_interruptible(&vcpu->wq); | |
254 | vcpu->stat.halt_wakeup++; | |
255 | } | |
256 | } | |
257 | break; | |
258 | case H_CONFER: | |
259 | break; | |
260 | case H_REGISTER_VPA: | |
261 | ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4), | |
262 | kvmppc_get_gpr(vcpu, 5), | |
263 | kvmppc_get_gpr(vcpu, 6)); | |
264 | break; | |
265 | default: | |
266 | return RESUME_HOST; | |
267 | } | |
268 | kvmppc_set_gpr(vcpu, 3, ret); | |
269 | vcpu->arch.hcall_needed = 0; | |
270 | return RESUME_GUEST; | |
271 | } | |
272 | ||
de56a948 PM |
273 | static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu, |
274 | struct task_struct *tsk) | |
275 | { | |
276 | int r = RESUME_HOST; | |
277 | ||
278 | vcpu->stat.sum_exits++; | |
279 | ||
280 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
281 | run->ready_for_interrupt_injection = 1; | |
282 | switch (vcpu->arch.trap) { | |
283 | /* We're good on these - the host merely wanted to get our attention */ | |
284 | case BOOK3S_INTERRUPT_HV_DECREMENTER: | |
285 | vcpu->stat.dec_exits++; | |
286 | r = RESUME_GUEST; | |
287 | break; | |
288 | case BOOK3S_INTERRUPT_EXTERNAL: | |
289 | vcpu->stat.ext_intr_exits++; | |
290 | r = RESUME_GUEST; | |
291 | break; | |
292 | case BOOK3S_INTERRUPT_PERFMON: | |
293 | r = RESUME_GUEST; | |
294 | break; | |
295 | case BOOK3S_INTERRUPT_PROGRAM: | |
296 | { | |
297 | ulong flags; | |
298 | /* | |
299 | * Normally program interrupts are delivered directly | |
300 | * to the guest by the hardware, but we can get here | |
301 | * as a result of a hypervisor emulation interrupt | |
302 | * (e40) getting turned into a 700 by BML RTAS. | |
303 | */ | |
304 | flags = vcpu->arch.shregs.msr & 0x1f0000ull; | |
305 | kvmppc_core_queue_program(vcpu, flags); | |
306 | r = RESUME_GUEST; | |
307 | break; | |
308 | } | |
309 | case BOOK3S_INTERRUPT_SYSCALL: | |
310 | { | |
311 | /* hcall - punt to userspace */ | |
312 | int i; | |
313 | ||
314 | if (vcpu->arch.shregs.msr & MSR_PR) { | |
315 | /* sc 1 from userspace - reflect to guest syscall */ | |
316 | kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL); | |
317 | r = RESUME_GUEST; | |
318 | break; | |
319 | } | |
320 | run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3); | |
321 | for (i = 0; i < 9; ++i) | |
322 | run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i); | |
323 | run->exit_reason = KVM_EXIT_PAPR_HCALL; | |
324 | vcpu->arch.hcall_needed = 1; | |
325 | r = RESUME_HOST; | |
326 | break; | |
327 | } | |
328 | /* | |
342d3db7 PM |
329 | * We get these next two if the guest accesses a page which it thinks |
330 | * it has mapped but which is not actually present, either because | |
331 | * it is for an emulated I/O device or because the corresonding | |
332 | * host page has been paged out. Any other HDSI/HISI interrupts | |
333 | * have been handled already. | |
de56a948 PM |
334 | */ |
335 | case BOOK3S_INTERRUPT_H_DATA_STORAGE: | |
697d3899 PM |
336 | r = kvmppc_book3s_hv_page_fault(run, vcpu, |
337 | vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); | |
de56a948 PM |
338 | break; |
339 | case BOOK3S_INTERRUPT_H_INST_STORAGE: | |
342d3db7 PM |
340 | r = kvmppc_book3s_hv_page_fault(run, vcpu, |
341 | kvmppc_get_pc(vcpu), 0); | |
de56a948 PM |
342 | break; |
343 | /* | |
344 | * This occurs if the guest executes an illegal instruction. | |
345 | * We just generate a program interrupt to the guest, since | |
346 | * we don't emulate any guest instructions at this stage. | |
347 | */ | |
348 | case BOOK3S_INTERRUPT_H_EMUL_ASSIST: | |
349 | kvmppc_core_queue_program(vcpu, 0x80000); | |
350 | r = RESUME_GUEST; | |
351 | break; | |
352 | default: | |
353 | kvmppc_dump_regs(vcpu); | |
354 | printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n", | |
355 | vcpu->arch.trap, kvmppc_get_pc(vcpu), | |
356 | vcpu->arch.shregs.msr); | |
357 | r = RESUME_HOST; | |
358 | BUG(); | |
359 | break; | |
360 | } | |
361 | ||
de56a948 PM |
362 | return r; |
363 | } | |
364 | ||
365 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, | |
366 | struct kvm_sregs *sregs) | |
367 | { | |
368 | int i; | |
369 | ||
370 | sregs->pvr = vcpu->arch.pvr; | |
371 | ||
372 | memset(sregs, 0, sizeof(struct kvm_sregs)); | |
373 | for (i = 0; i < vcpu->arch.slb_max; i++) { | |
374 | sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige; | |
375 | sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv; | |
376 | } | |
377 | ||
378 | return 0; | |
379 | } | |
380 | ||
381 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, | |
382 | struct kvm_sregs *sregs) | |
383 | { | |
384 | int i, j; | |
385 | ||
386 | kvmppc_set_pvr(vcpu, sregs->pvr); | |
387 | ||
388 | j = 0; | |
389 | for (i = 0; i < vcpu->arch.slb_nr; i++) { | |
390 | if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) { | |
391 | vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe; | |
392 | vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv; | |
393 | ++j; | |
394 | } | |
395 | } | |
396 | vcpu->arch.slb_max = j; | |
397 | ||
398 | return 0; | |
399 | } | |
400 | ||
31f3438e PM |
401 | int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) |
402 | { | |
403 | int r = -EINVAL; | |
404 | ||
405 | switch (reg->id) { | |
406 | case KVM_REG_PPC_HIOR: | |
407 | r = put_user(0, (u64 __user *)reg->addr); | |
408 | break; | |
409 | default: | |
410 | break; | |
411 | } | |
412 | ||
413 | return r; | |
414 | } | |
415 | ||
416 | int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) | |
417 | { | |
418 | int r = -EINVAL; | |
419 | ||
420 | switch (reg->id) { | |
421 | case KVM_REG_PPC_HIOR: | |
422 | { | |
423 | u64 hior; | |
424 | /* Only allow this to be set to zero */ | |
425 | r = get_user(hior, (u64 __user *)reg->addr); | |
426 | if (!r && (hior != 0)) | |
427 | r = -EINVAL; | |
428 | break; | |
429 | } | |
430 | default: | |
431 | break; | |
432 | } | |
433 | ||
434 | return r; | |
435 | } | |
436 | ||
de56a948 PM |
437 | int kvmppc_core_check_processor_compat(void) |
438 | { | |
9e368f29 | 439 | if (cpu_has_feature(CPU_FTR_HVMODE)) |
de56a948 PM |
440 | return 0; |
441 | return -EIO; | |
442 | } | |
443 | ||
444 | struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id) | |
445 | { | |
446 | struct kvm_vcpu *vcpu; | |
371fefd6 PM |
447 | int err = -EINVAL; |
448 | int core; | |
449 | struct kvmppc_vcore *vcore; | |
de56a948 | 450 | |
371fefd6 PM |
451 | core = id / threads_per_core; |
452 | if (core >= KVM_MAX_VCORES) | |
453 | goto out; | |
454 | ||
455 | err = -ENOMEM; | |
6b75e6bf | 456 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
de56a948 PM |
457 | if (!vcpu) |
458 | goto out; | |
459 | ||
460 | err = kvm_vcpu_init(vcpu, kvm, id); | |
461 | if (err) | |
462 | goto free_vcpu; | |
463 | ||
464 | vcpu->arch.shared = &vcpu->arch.shregs; | |
465 | vcpu->arch.last_cpu = -1; | |
466 | vcpu->arch.mmcr[0] = MMCR0_FC; | |
467 | vcpu->arch.ctrl = CTRL_RUNLATCH; | |
468 | /* default to host PVR, since we can't spoof it */ | |
469 | vcpu->arch.pvr = mfspr(SPRN_PVR); | |
470 | kvmppc_set_pvr(vcpu, vcpu->arch.pvr); | |
471 | ||
de56a948 PM |
472 | kvmppc_mmu_book3s_hv_init(vcpu); |
473 | ||
371fefd6 | 474 | /* |
19ccb76a | 475 | * We consider the vcpu stopped until we see the first run ioctl for it. |
371fefd6 | 476 | */ |
19ccb76a | 477 | vcpu->arch.state = KVMPPC_VCPU_STOPPED; |
371fefd6 PM |
478 | |
479 | init_waitqueue_head(&vcpu->arch.cpu_run); | |
480 | ||
481 | mutex_lock(&kvm->lock); | |
482 | vcore = kvm->arch.vcores[core]; | |
483 | if (!vcore) { | |
484 | vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL); | |
485 | if (vcore) { | |
486 | INIT_LIST_HEAD(&vcore->runnable_threads); | |
487 | spin_lock_init(&vcore->lock); | |
19ccb76a | 488 | init_waitqueue_head(&vcore->wq); |
371fefd6 PM |
489 | } |
490 | kvm->arch.vcores[core] = vcore; | |
491 | } | |
492 | mutex_unlock(&kvm->lock); | |
493 | ||
494 | if (!vcore) | |
495 | goto free_vcpu; | |
496 | ||
497 | spin_lock(&vcore->lock); | |
498 | ++vcore->num_threads; | |
371fefd6 PM |
499 | spin_unlock(&vcore->lock); |
500 | vcpu->arch.vcore = vcore; | |
501 | ||
af8f38b3 AG |
502 | vcpu->arch.cpu_type = KVM_CPU_3S_64; |
503 | kvmppc_sanity_check(vcpu); | |
504 | ||
de56a948 PM |
505 | return vcpu; |
506 | ||
507 | free_vcpu: | |
6b75e6bf | 508 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
de56a948 PM |
509 | out: |
510 | return ERR_PTR(err); | |
511 | } | |
512 | ||
513 | void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu) | |
514 | { | |
93e60249 PM |
515 | if (vcpu->arch.dtl) |
516 | kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.dtl); | |
517 | if (vcpu->arch.slb_shadow) | |
518 | kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.slb_shadow); | |
519 | if (vcpu->arch.vpa) | |
520 | kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.vpa); | |
de56a948 | 521 | kvm_vcpu_uninit(vcpu); |
6b75e6bf | 522 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
de56a948 PM |
523 | } |
524 | ||
19ccb76a | 525 | static void kvmppc_set_timer(struct kvm_vcpu *vcpu) |
371fefd6 | 526 | { |
19ccb76a | 527 | unsigned long dec_nsec, now; |
371fefd6 | 528 | |
19ccb76a PM |
529 | now = get_tb(); |
530 | if (now > vcpu->arch.dec_expires) { | |
531 | /* decrementer has already gone negative */ | |
532 | kvmppc_core_queue_dec(vcpu); | |
7e28e60e | 533 | kvmppc_core_prepare_to_enter(vcpu); |
19ccb76a | 534 | return; |
371fefd6 | 535 | } |
19ccb76a PM |
536 | dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC |
537 | / tb_ticks_per_sec; | |
538 | hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec), | |
539 | HRTIMER_MODE_REL); | |
540 | vcpu->arch.timer_running = 1; | |
371fefd6 PM |
541 | } |
542 | ||
19ccb76a | 543 | static void kvmppc_end_cede(struct kvm_vcpu *vcpu) |
371fefd6 | 544 | { |
19ccb76a PM |
545 | vcpu->arch.ceded = 0; |
546 | if (vcpu->arch.timer_running) { | |
547 | hrtimer_try_to_cancel(&vcpu->arch.dec_timer); | |
548 | vcpu->arch.timer_running = 0; | |
549 | } | |
371fefd6 PM |
550 | } |
551 | ||
de56a948 | 552 | extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu); |
371fefd6 | 553 | extern void xics_wake_cpu(int cpu); |
de56a948 | 554 | |
371fefd6 PM |
555 | static void kvmppc_remove_runnable(struct kvmppc_vcore *vc, |
556 | struct kvm_vcpu *vcpu) | |
de56a948 | 557 | { |
371fefd6 | 558 | struct kvm_vcpu *v; |
de56a948 | 559 | |
371fefd6 PM |
560 | if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) |
561 | return; | |
562 | vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; | |
563 | --vc->n_runnable; | |
19ccb76a | 564 | ++vc->n_busy; |
371fefd6 PM |
565 | /* decrement the physical thread id of each following vcpu */ |
566 | v = vcpu; | |
567 | list_for_each_entry_continue(v, &vc->runnable_threads, arch.run_list) | |
568 | --v->arch.ptid; | |
569 | list_del(&vcpu->arch.run_list); | |
570 | } | |
571 | ||
f0888f70 PM |
572 | static int kvmppc_grab_hwthread(int cpu) |
573 | { | |
574 | struct paca_struct *tpaca; | |
575 | long timeout = 1000; | |
576 | ||
577 | tpaca = &paca[cpu]; | |
578 | ||
579 | /* Ensure the thread won't go into the kernel if it wakes */ | |
580 | tpaca->kvm_hstate.hwthread_req = 1; | |
581 | ||
582 | /* | |
583 | * If the thread is already executing in the kernel (e.g. handling | |
584 | * a stray interrupt), wait for it to get back to nap mode. | |
585 | * The smp_mb() is to ensure that our setting of hwthread_req | |
586 | * is visible before we look at hwthread_state, so if this | |
587 | * races with the code at system_reset_pSeries and the thread | |
588 | * misses our setting of hwthread_req, we are sure to see its | |
589 | * setting of hwthread_state, and vice versa. | |
590 | */ | |
591 | smp_mb(); | |
592 | while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) { | |
593 | if (--timeout <= 0) { | |
594 | pr_err("KVM: couldn't grab cpu %d\n", cpu); | |
595 | return -EBUSY; | |
596 | } | |
597 | udelay(1); | |
598 | } | |
599 | return 0; | |
600 | } | |
601 | ||
602 | static void kvmppc_release_hwthread(int cpu) | |
603 | { | |
604 | struct paca_struct *tpaca; | |
605 | ||
606 | tpaca = &paca[cpu]; | |
607 | tpaca->kvm_hstate.hwthread_req = 0; | |
608 | tpaca->kvm_hstate.kvm_vcpu = NULL; | |
609 | } | |
610 | ||
371fefd6 PM |
611 | static void kvmppc_start_thread(struct kvm_vcpu *vcpu) |
612 | { | |
613 | int cpu; | |
614 | struct paca_struct *tpaca; | |
615 | struct kvmppc_vcore *vc = vcpu->arch.vcore; | |
616 | ||
19ccb76a PM |
617 | if (vcpu->arch.timer_running) { |
618 | hrtimer_try_to_cancel(&vcpu->arch.dec_timer); | |
619 | vcpu->arch.timer_running = 0; | |
620 | } | |
371fefd6 PM |
621 | cpu = vc->pcpu + vcpu->arch.ptid; |
622 | tpaca = &paca[cpu]; | |
623 | tpaca->kvm_hstate.kvm_vcpu = vcpu; | |
624 | tpaca->kvm_hstate.kvm_vcore = vc; | |
19ccb76a PM |
625 | tpaca->kvm_hstate.napping = 0; |
626 | vcpu->cpu = vc->pcpu; | |
371fefd6 | 627 | smp_wmb(); |
251da038 | 628 | #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP) |
371fefd6 | 629 | if (vcpu->arch.ptid) { |
f0888f70 | 630 | kvmppc_grab_hwthread(cpu); |
371fefd6 PM |
631 | xics_wake_cpu(cpu); |
632 | ++vc->n_woken; | |
de56a948 | 633 | } |
371fefd6 PM |
634 | #endif |
635 | } | |
de56a948 | 636 | |
371fefd6 PM |
637 | static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc) |
638 | { | |
639 | int i; | |
640 | ||
641 | HMT_low(); | |
642 | i = 0; | |
643 | while (vc->nap_count < vc->n_woken) { | |
644 | if (++i >= 1000000) { | |
645 | pr_err("kvmppc_wait_for_nap timeout %d %d\n", | |
646 | vc->nap_count, vc->n_woken); | |
647 | break; | |
648 | } | |
649 | cpu_relax(); | |
650 | } | |
651 | HMT_medium(); | |
652 | } | |
653 | ||
654 | /* | |
655 | * Check that we are on thread 0 and that any other threads in | |
656 | * this core are off-line. | |
657 | */ | |
658 | static int on_primary_thread(void) | |
659 | { | |
660 | int cpu = smp_processor_id(); | |
661 | int thr = cpu_thread_in_core(cpu); | |
662 | ||
663 | if (thr) | |
664 | return 0; | |
665 | while (++thr < threads_per_core) | |
666 | if (cpu_online(cpu + thr)) | |
667 | return 0; | |
668 | return 1; | |
669 | } | |
670 | ||
671 | /* | |
672 | * Run a set of guest threads on a physical core. | |
673 | * Called with vc->lock held. | |
674 | */ | |
675 | static int kvmppc_run_core(struct kvmppc_vcore *vc) | |
676 | { | |
19ccb76a | 677 | struct kvm_vcpu *vcpu, *vcpu0, *vnext; |
371fefd6 PM |
678 | long ret; |
679 | u64 now; | |
f0888f70 | 680 | int ptid, i; |
371fefd6 PM |
681 | |
682 | /* don't start if any threads have a signal pending */ | |
683 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
684 | if (signal_pending(vcpu->arch.run_task)) | |
685 | return 0; | |
de56a948 PM |
686 | |
687 | /* | |
688 | * Make sure we are running on thread 0, and that | |
689 | * secondary threads are offline. | |
690 | * XXX we should also block attempts to bring any | |
691 | * secondary threads online. | |
692 | */ | |
371fefd6 PM |
693 | if (threads_per_core > 1 && !on_primary_thread()) { |
694 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
695 | vcpu->arch.ret = -EBUSY; | |
696 | goto out; | |
de56a948 PM |
697 | } |
698 | ||
19ccb76a PM |
699 | /* |
700 | * Assign physical thread IDs, first to non-ceded vcpus | |
701 | * and then to ceded ones. | |
702 | */ | |
703 | ptid = 0; | |
704 | vcpu0 = NULL; | |
705 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) { | |
706 | if (!vcpu->arch.ceded) { | |
707 | if (!ptid) | |
708 | vcpu0 = vcpu; | |
709 | vcpu->arch.ptid = ptid++; | |
710 | } | |
711 | } | |
712 | if (!vcpu0) | |
713 | return 0; /* nothing to run */ | |
714 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
715 | if (vcpu->arch.ceded) | |
716 | vcpu->arch.ptid = ptid++; | |
717 | ||
371fefd6 PM |
718 | vc->n_woken = 0; |
719 | vc->nap_count = 0; | |
720 | vc->entry_exit_count = 0; | |
19ccb76a | 721 | vc->vcore_state = VCORE_RUNNING; |
371fefd6 PM |
722 | vc->in_guest = 0; |
723 | vc->pcpu = smp_processor_id(); | |
19ccb76a | 724 | vc->napping_threads = 0; |
371fefd6 PM |
725 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) |
726 | kvmppc_start_thread(vcpu); | |
f0888f70 PM |
727 | /* Grab any remaining hw threads so they can't go into the kernel */ |
728 | for (i = ptid; i < threads_per_core; ++i) | |
729 | kvmppc_grab_hwthread(vc->pcpu + i); | |
371fefd6 | 730 | |
19ccb76a | 731 | preempt_disable(); |
371fefd6 | 732 | spin_unlock(&vc->lock); |
de56a948 | 733 | |
371fefd6 | 734 | kvm_guest_enter(); |
19ccb76a | 735 | __kvmppc_vcore_entry(NULL, vcpu0); |
f0888f70 PM |
736 | for (i = 0; i < threads_per_core; ++i) |
737 | kvmppc_release_hwthread(vc->pcpu + i); | |
de56a948 | 738 | |
371fefd6 | 739 | spin_lock(&vc->lock); |
19ccb76a PM |
740 | /* disable sending of IPIs on virtual external irqs */ |
741 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
742 | vcpu->cpu = -1; | |
743 | /* wait for secondary threads to finish writing their state to memory */ | |
371fefd6 PM |
744 | if (vc->nap_count < vc->n_woken) |
745 | kvmppc_wait_for_nap(vc); | |
746 | /* prevent other vcpu threads from doing kvmppc_start_thread() now */ | |
19ccb76a | 747 | vc->vcore_state = VCORE_EXITING; |
371fefd6 PM |
748 | spin_unlock(&vc->lock); |
749 | ||
750 | /* make sure updates to secondary vcpu structs are visible now */ | |
751 | smp_mb(); | |
de56a948 PM |
752 | kvm_guest_exit(); |
753 | ||
754 | preempt_enable(); | |
755 | kvm_resched(vcpu); | |
756 | ||
757 | now = get_tb(); | |
371fefd6 PM |
758 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) { |
759 | /* cancel pending dec exception if dec is positive */ | |
760 | if (now < vcpu->arch.dec_expires && | |
761 | kvmppc_core_pending_dec(vcpu)) | |
762 | kvmppc_core_dequeue_dec(vcpu); | |
19ccb76a PM |
763 | |
764 | ret = RESUME_GUEST; | |
765 | if (vcpu->arch.trap) | |
766 | ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu, | |
767 | vcpu->arch.run_task); | |
768 | ||
371fefd6 PM |
769 | vcpu->arch.ret = ret; |
770 | vcpu->arch.trap = 0; | |
19ccb76a PM |
771 | |
772 | if (vcpu->arch.ceded) { | |
773 | if (ret != RESUME_GUEST) | |
774 | kvmppc_end_cede(vcpu); | |
775 | else | |
776 | kvmppc_set_timer(vcpu); | |
777 | } | |
371fefd6 | 778 | } |
de56a948 | 779 | |
371fefd6 | 780 | spin_lock(&vc->lock); |
de56a948 | 781 | out: |
19ccb76a | 782 | vc->vcore_state = VCORE_INACTIVE; |
371fefd6 PM |
783 | list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads, |
784 | arch.run_list) { | |
785 | if (vcpu->arch.ret != RESUME_GUEST) { | |
786 | kvmppc_remove_runnable(vc, vcpu); | |
787 | wake_up(&vcpu->arch.cpu_run); | |
788 | } | |
789 | } | |
790 | ||
791 | return 1; | |
792 | } | |
793 | ||
19ccb76a PM |
794 | /* |
795 | * Wait for some other vcpu thread to execute us, and | |
796 | * wake us up when we need to handle something in the host. | |
797 | */ | |
798 | static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state) | |
371fefd6 | 799 | { |
371fefd6 PM |
800 | DEFINE_WAIT(wait); |
801 | ||
19ccb76a PM |
802 | prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state); |
803 | if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) | |
804 | schedule(); | |
805 | finish_wait(&vcpu->arch.cpu_run, &wait); | |
806 | } | |
807 | ||
808 | /* | |
809 | * All the vcpus in this vcore are idle, so wait for a decrementer | |
810 | * or external interrupt to one of the vcpus. vc->lock is held. | |
811 | */ | |
812 | static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc) | |
813 | { | |
814 | DEFINE_WAIT(wait); | |
815 | struct kvm_vcpu *v; | |
816 | int all_idle = 1; | |
817 | ||
818 | prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE); | |
819 | vc->vcore_state = VCORE_SLEEPING; | |
820 | spin_unlock(&vc->lock); | |
821 | list_for_each_entry(v, &vc->runnable_threads, arch.run_list) { | |
822 | if (!v->arch.ceded || v->arch.pending_exceptions) { | |
823 | all_idle = 0; | |
824 | break; | |
825 | } | |
371fefd6 | 826 | } |
19ccb76a PM |
827 | if (all_idle) |
828 | schedule(); | |
829 | finish_wait(&vc->wq, &wait); | |
830 | spin_lock(&vc->lock); | |
831 | vc->vcore_state = VCORE_INACTIVE; | |
832 | } | |
371fefd6 | 833 | |
19ccb76a PM |
834 | static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu) |
835 | { | |
836 | int n_ceded; | |
837 | int prev_state; | |
838 | struct kvmppc_vcore *vc; | |
839 | struct kvm_vcpu *v, *vn; | |
9e368f29 | 840 | |
371fefd6 PM |
841 | kvm_run->exit_reason = 0; |
842 | vcpu->arch.ret = RESUME_GUEST; | |
843 | vcpu->arch.trap = 0; | |
844 | ||
371fefd6 PM |
845 | /* |
846 | * Synchronize with other threads in this virtual core | |
847 | */ | |
848 | vc = vcpu->arch.vcore; | |
849 | spin_lock(&vc->lock); | |
19ccb76a | 850 | vcpu->arch.ceded = 0; |
371fefd6 PM |
851 | vcpu->arch.run_task = current; |
852 | vcpu->arch.kvm_run = kvm_run; | |
19ccb76a PM |
853 | prev_state = vcpu->arch.state; |
854 | vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; | |
371fefd6 PM |
855 | list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads); |
856 | ++vc->n_runnable; | |
857 | ||
19ccb76a PM |
858 | /* |
859 | * This happens the first time this is called for a vcpu. | |
860 | * If the vcore is already running, we may be able to start | |
861 | * this thread straight away and have it join in. | |
862 | */ | |
863 | if (prev_state == KVMPPC_VCPU_STOPPED) { | |
864 | if (vc->vcore_state == VCORE_RUNNING && | |
865 | VCORE_EXIT_COUNT(vc) == 0) { | |
866 | vcpu->arch.ptid = vc->n_runnable - 1; | |
867 | kvmppc_start_thread(vcpu); | |
371fefd6 PM |
868 | } |
869 | ||
19ccb76a PM |
870 | } else if (prev_state == KVMPPC_VCPU_BUSY_IN_HOST) |
871 | --vc->n_busy; | |
371fefd6 | 872 | |
19ccb76a PM |
873 | while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && |
874 | !signal_pending(current)) { | |
875 | if (vc->n_busy || vc->vcore_state != VCORE_INACTIVE) { | |
876 | spin_unlock(&vc->lock); | |
877 | kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE); | |
878 | spin_lock(&vc->lock); | |
879 | continue; | |
880 | } | |
881 | n_ceded = 0; | |
882 | list_for_each_entry(v, &vc->runnable_threads, arch.run_list) | |
883 | n_ceded += v->arch.ceded; | |
884 | if (n_ceded == vc->n_runnable) | |
885 | kvmppc_vcore_blocked(vc); | |
886 | else | |
887 | kvmppc_run_core(vc); | |
888 | ||
889 | list_for_each_entry_safe(v, vn, &vc->runnable_threads, | |
890 | arch.run_list) { | |
7e28e60e | 891 | kvmppc_core_prepare_to_enter(v); |
19ccb76a PM |
892 | if (signal_pending(v->arch.run_task)) { |
893 | kvmppc_remove_runnable(vc, v); | |
894 | v->stat.signal_exits++; | |
895 | v->arch.kvm_run->exit_reason = KVM_EXIT_INTR; | |
896 | v->arch.ret = -EINTR; | |
897 | wake_up(&v->arch.cpu_run); | |
898 | } | |
899 | } | |
900 | } | |
371fefd6 | 901 | |
19ccb76a PM |
902 | if (signal_pending(current)) { |
903 | if (vc->vcore_state == VCORE_RUNNING || | |
904 | vc->vcore_state == VCORE_EXITING) { | |
905 | spin_unlock(&vc->lock); | |
906 | kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE); | |
907 | spin_lock(&vc->lock); | |
908 | } | |
909 | if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { | |
910 | kvmppc_remove_runnable(vc, vcpu); | |
911 | vcpu->stat.signal_exits++; | |
912 | kvm_run->exit_reason = KVM_EXIT_INTR; | |
913 | vcpu->arch.ret = -EINTR; | |
914 | } | |
371fefd6 PM |
915 | } |
916 | ||
371fefd6 | 917 | spin_unlock(&vc->lock); |
371fefd6 | 918 | return vcpu->arch.ret; |
de56a948 PM |
919 | } |
920 | ||
a8606e20 PM |
921 | int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu) |
922 | { | |
923 | int r; | |
924 | ||
af8f38b3 AG |
925 | if (!vcpu->arch.sane) { |
926 | run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
927 | return -EINVAL; | |
928 | } | |
929 | ||
25051b5a SW |
930 | kvmppc_core_prepare_to_enter(vcpu); |
931 | ||
19ccb76a PM |
932 | /* No need to go into the guest when all we'll do is come back out */ |
933 | if (signal_pending(current)) { | |
934 | run->exit_reason = KVM_EXIT_INTR; | |
935 | return -EINTR; | |
936 | } | |
937 | ||
c77162de PM |
938 | /* On the first time here, set up VRMA or RMA */ |
939 | if (!vcpu->kvm->arch.rma_setup_done) { | |
940 | r = kvmppc_hv_setup_rma(vcpu); | |
941 | if (r) | |
942 | return r; | |
943 | } | |
19ccb76a PM |
944 | |
945 | flush_fp_to_thread(current); | |
946 | flush_altivec_to_thread(current); | |
947 | flush_vsx_to_thread(current); | |
948 | vcpu->arch.wqp = &vcpu->arch.vcore->wq; | |
342d3db7 | 949 | vcpu->arch.pgdir = current->mm->pgd; |
19ccb76a | 950 | |
a8606e20 PM |
951 | do { |
952 | r = kvmppc_run_vcpu(run, vcpu); | |
953 | ||
954 | if (run->exit_reason == KVM_EXIT_PAPR_HCALL && | |
955 | !(vcpu->arch.shregs.msr & MSR_PR)) { | |
956 | r = kvmppc_pseries_do_hcall(vcpu); | |
7e28e60e | 957 | kvmppc_core_prepare_to_enter(vcpu); |
a8606e20 PM |
958 | } |
959 | } while (r == RESUME_GUEST); | |
960 | return r; | |
961 | } | |
962 | ||
54738c09 DG |
963 | static long kvmppc_stt_npages(unsigned long window_size) |
964 | { | |
965 | return ALIGN((window_size >> SPAPR_TCE_SHIFT) | |
966 | * sizeof(u64), PAGE_SIZE) / PAGE_SIZE; | |
967 | } | |
968 | ||
969 | static void release_spapr_tce_table(struct kvmppc_spapr_tce_table *stt) | |
970 | { | |
971 | struct kvm *kvm = stt->kvm; | |
972 | int i; | |
973 | ||
974 | mutex_lock(&kvm->lock); | |
975 | list_del(&stt->list); | |
976 | for (i = 0; i < kvmppc_stt_npages(stt->window_size); i++) | |
977 | __free_page(stt->pages[i]); | |
978 | kfree(stt); | |
979 | mutex_unlock(&kvm->lock); | |
980 | ||
981 | kvm_put_kvm(kvm); | |
982 | } | |
983 | ||
984 | static int kvm_spapr_tce_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
985 | { | |
986 | struct kvmppc_spapr_tce_table *stt = vma->vm_file->private_data; | |
987 | struct page *page; | |
988 | ||
989 | if (vmf->pgoff >= kvmppc_stt_npages(stt->window_size)) | |
990 | return VM_FAULT_SIGBUS; | |
991 | ||
992 | page = stt->pages[vmf->pgoff]; | |
993 | get_page(page); | |
994 | vmf->page = page; | |
995 | return 0; | |
996 | } | |
997 | ||
998 | static const struct vm_operations_struct kvm_spapr_tce_vm_ops = { | |
999 | .fault = kvm_spapr_tce_fault, | |
1000 | }; | |
1001 | ||
1002 | static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma) | |
1003 | { | |
1004 | vma->vm_ops = &kvm_spapr_tce_vm_ops; | |
1005 | return 0; | |
1006 | } | |
1007 | ||
1008 | static int kvm_spapr_tce_release(struct inode *inode, struct file *filp) | |
1009 | { | |
1010 | struct kvmppc_spapr_tce_table *stt = filp->private_data; | |
1011 | ||
1012 | release_spapr_tce_table(stt); | |
1013 | return 0; | |
1014 | } | |
1015 | ||
1016 | static struct file_operations kvm_spapr_tce_fops = { | |
1017 | .mmap = kvm_spapr_tce_mmap, | |
1018 | .release = kvm_spapr_tce_release, | |
1019 | }; | |
1020 | ||
1021 | long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm, | |
1022 | struct kvm_create_spapr_tce *args) | |
1023 | { | |
1024 | struct kvmppc_spapr_tce_table *stt = NULL; | |
1025 | long npages; | |
1026 | int ret = -ENOMEM; | |
1027 | int i; | |
1028 | ||
1029 | /* Check this LIOBN hasn't been previously allocated */ | |
1030 | list_for_each_entry(stt, &kvm->arch.spapr_tce_tables, list) { | |
1031 | if (stt->liobn == args->liobn) | |
1032 | return -EBUSY; | |
1033 | } | |
1034 | ||
1035 | npages = kvmppc_stt_npages(args->window_size); | |
1036 | ||
1037 | stt = kzalloc(sizeof(*stt) + npages* sizeof(struct page *), | |
1038 | GFP_KERNEL); | |
1039 | if (!stt) | |
1040 | goto fail; | |
1041 | ||
1042 | stt->liobn = args->liobn; | |
1043 | stt->window_size = args->window_size; | |
1044 | stt->kvm = kvm; | |
1045 | ||
1046 | for (i = 0; i < npages; i++) { | |
1047 | stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
1048 | if (!stt->pages[i]) | |
1049 | goto fail; | |
1050 | } | |
1051 | ||
1052 | kvm_get_kvm(kvm); | |
1053 | ||
1054 | mutex_lock(&kvm->lock); | |
1055 | list_add(&stt->list, &kvm->arch.spapr_tce_tables); | |
1056 | ||
1057 | mutex_unlock(&kvm->lock); | |
1058 | ||
1059 | return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops, | |
1060 | stt, O_RDWR); | |
1061 | ||
1062 | fail: | |
1063 | if (stt) { | |
1064 | for (i = 0; i < npages; i++) | |
1065 | if (stt->pages[i]) | |
1066 | __free_page(stt->pages[i]); | |
1067 | ||
1068 | kfree(stt); | |
1069 | } | |
1070 | return ret; | |
1071 | } | |
1072 | ||
aa04b4cc | 1073 | /* Work out RMLS (real mode limit selector) field value for a given RMA size. |
9e368f29 | 1074 | Assumes POWER7 or PPC970. */ |
aa04b4cc PM |
1075 | static inline int lpcr_rmls(unsigned long rma_size) |
1076 | { | |
1077 | switch (rma_size) { | |
1078 | case 32ul << 20: /* 32 MB */ | |
9e368f29 PM |
1079 | if (cpu_has_feature(CPU_FTR_ARCH_206)) |
1080 | return 8; /* only supported on POWER7 */ | |
1081 | return -1; | |
aa04b4cc PM |
1082 | case 64ul << 20: /* 64 MB */ |
1083 | return 3; | |
1084 | case 128ul << 20: /* 128 MB */ | |
1085 | return 7; | |
1086 | case 256ul << 20: /* 256 MB */ | |
1087 | return 4; | |
1088 | case 1ul << 30: /* 1 GB */ | |
1089 | return 2; | |
1090 | case 16ul << 30: /* 16 GB */ | |
1091 | return 1; | |
1092 | case 256ul << 30: /* 256 GB */ | |
1093 | return 0; | |
1094 | default: | |
1095 | return -1; | |
1096 | } | |
1097 | } | |
1098 | ||
1099 | static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
1100 | { | |
b4e70611 | 1101 | struct kvmppc_linear_info *ri = vma->vm_file->private_data; |
aa04b4cc PM |
1102 | struct page *page; |
1103 | ||
1104 | if (vmf->pgoff >= ri->npages) | |
1105 | return VM_FAULT_SIGBUS; | |
1106 | ||
1107 | page = pfn_to_page(ri->base_pfn + vmf->pgoff); | |
1108 | get_page(page); | |
1109 | vmf->page = page; | |
1110 | return 0; | |
1111 | } | |
1112 | ||
1113 | static const struct vm_operations_struct kvm_rma_vm_ops = { | |
1114 | .fault = kvm_rma_fault, | |
1115 | }; | |
1116 | ||
1117 | static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma) | |
1118 | { | |
1119 | vma->vm_flags |= VM_RESERVED; | |
1120 | vma->vm_ops = &kvm_rma_vm_ops; | |
1121 | return 0; | |
1122 | } | |
1123 | ||
1124 | static int kvm_rma_release(struct inode *inode, struct file *filp) | |
1125 | { | |
b4e70611 | 1126 | struct kvmppc_linear_info *ri = filp->private_data; |
aa04b4cc PM |
1127 | |
1128 | kvm_release_rma(ri); | |
1129 | return 0; | |
1130 | } | |
1131 | ||
1132 | static struct file_operations kvm_rma_fops = { | |
1133 | .mmap = kvm_rma_mmap, | |
1134 | .release = kvm_rma_release, | |
1135 | }; | |
1136 | ||
1137 | long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret) | |
1138 | { | |
b4e70611 | 1139 | struct kvmppc_linear_info *ri; |
aa04b4cc PM |
1140 | long fd; |
1141 | ||
1142 | ri = kvm_alloc_rma(); | |
1143 | if (!ri) | |
1144 | return -ENOMEM; | |
1145 | ||
1146 | fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR); | |
1147 | if (fd < 0) | |
1148 | kvm_release_rma(ri); | |
1149 | ||
1150 | ret->rma_size = ri->npages << PAGE_SHIFT; | |
1151 | return fd; | |
1152 | } | |
1153 | ||
82ed3616 PM |
1154 | /* |
1155 | * Get (and clear) the dirty memory log for a memory slot. | |
1156 | */ | |
1157 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) | |
1158 | { | |
1159 | struct kvm_memory_slot *memslot; | |
1160 | int r; | |
1161 | unsigned long n; | |
1162 | ||
1163 | mutex_lock(&kvm->slots_lock); | |
1164 | ||
1165 | r = -EINVAL; | |
1166 | if (log->slot >= KVM_MEMORY_SLOTS) | |
1167 | goto out; | |
1168 | ||
1169 | memslot = id_to_memslot(kvm->memslots, log->slot); | |
1170 | r = -ENOENT; | |
1171 | if (!memslot->dirty_bitmap) | |
1172 | goto out; | |
1173 | ||
1174 | n = kvm_dirty_bitmap_bytes(memslot); | |
1175 | memset(memslot->dirty_bitmap, 0, n); | |
1176 | ||
1177 | r = kvmppc_hv_get_dirty_log(kvm, memslot); | |
1178 | if (r) | |
1179 | goto out; | |
1180 | ||
1181 | r = -EFAULT; | |
1182 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
1183 | goto out; | |
1184 | ||
1185 | r = 0; | |
1186 | out: | |
1187 | mutex_unlock(&kvm->slots_lock); | |
1188 | return r; | |
1189 | } | |
1190 | ||
da9d1d7f PM |
1191 | static unsigned long slb_pgsize_encoding(unsigned long psize) |
1192 | { | |
1193 | unsigned long senc = 0; | |
1194 | ||
1195 | if (psize > 0x1000) { | |
1196 | senc = SLB_VSID_L; | |
1197 | if (psize == 0x10000) | |
1198 | senc |= SLB_VSID_LP_01; | |
1199 | } | |
1200 | return senc; | |
1201 | } | |
1202 | ||
de56a948 PM |
1203 | int kvmppc_core_prepare_memory_region(struct kvm *kvm, |
1204 | struct kvm_userspace_memory_region *mem) | |
1205 | { | |
c77162de | 1206 | unsigned long npages; |
b2b2f165 | 1207 | unsigned long *phys; |
aa04b4cc | 1208 | |
b2b2f165 | 1209 | /* Allocate a slot_phys array */ |
b2b2f165 | 1210 | phys = kvm->arch.slot_phys[mem->slot]; |
342d3db7 PM |
1211 | if (!kvm->arch.using_mmu_notifiers && !phys) { |
1212 | npages = mem->memory_size >> PAGE_SHIFT; | |
b2b2f165 PM |
1213 | phys = vzalloc(npages * sizeof(unsigned long)); |
1214 | if (!phys) | |
1215 | return -ENOMEM; | |
1216 | kvm->arch.slot_phys[mem->slot] = phys; | |
1217 | kvm->arch.slot_npages[mem->slot] = npages; | |
1218 | } | |
aa04b4cc | 1219 | |
c77162de PM |
1220 | return 0; |
1221 | } | |
aa04b4cc | 1222 | |
c77162de PM |
1223 | static void unpin_slot(struct kvm *kvm, int slot_id) |
1224 | { | |
1225 | unsigned long *physp; | |
1226 | unsigned long j, npages, pfn; | |
1227 | struct page *page; | |
1228 | ||
1229 | physp = kvm->arch.slot_phys[slot_id]; | |
1230 | npages = kvm->arch.slot_npages[slot_id]; | |
1231 | if (physp) { | |
1232 | spin_lock(&kvm->arch.slot_phys_lock); | |
1233 | for (j = 0; j < npages; j++) { | |
1234 | if (!(physp[j] & KVMPPC_GOT_PAGE)) | |
1235 | continue; | |
1236 | pfn = physp[j] >> PAGE_SHIFT; | |
1237 | page = pfn_to_page(pfn); | |
da9d1d7f PM |
1238 | if (PageHuge(page)) |
1239 | page = compound_head(page); | |
c77162de PM |
1240 | SetPageDirty(page); |
1241 | put_page(page); | |
9e368f29 | 1242 | } |
c77162de PM |
1243 | kvm->arch.slot_phys[slot_id] = NULL; |
1244 | spin_unlock(&kvm->arch.slot_phys_lock); | |
1245 | vfree(physp); | |
aa04b4cc | 1246 | } |
c77162de PM |
1247 | } |
1248 | ||
1249 | void kvmppc_core_commit_memory_region(struct kvm *kvm, | |
1250 | struct kvm_userspace_memory_region *mem) | |
1251 | { | |
1252 | } | |
1253 | ||
1254 | static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu) | |
1255 | { | |
1256 | int err = 0; | |
1257 | struct kvm *kvm = vcpu->kvm; | |
b4e70611 | 1258 | struct kvmppc_linear_info *ri = NULL; |
c77162de PM |
1259 | unsigned long hva; |
1260 | struct kvm_memory_slot *memslot; | |
1261 | struct vm_area_struct *vma; | |
da9d1d7f | 1262 | unsigned long lpcr, senc; |
c77162de PM |
1263 | unsigned long psize, porder; |
1264 | unsigned long rma_size; | |
1265 | unsigned long rmls; | |
1266 | unsigned long *physp; | |
da9d1d7f | 1267 | unsigned long i, npages; |
c77162de PM |
1268 | |
1269 | mutex_lock(&kvm->lock); | |
1270 | if (kvm->arch.rma_setup_done) | |
1271 | goto out; /* another vcpu beat us to it */ | |
aa04b4cc | 1272 | |
c77162de PM |
1273 | /* Look up the memslot for guest physical address 0 */ |
1274 | memslot = gfn_to_memslot(kvm, 0); | |
aa04b4cc | 1275 | |
c77162de PM |
1276 | /* We must have some memory at 0 by now */ |
1277 | err = -EINVAL; | |
1278 | if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) | |
1279 | goto out; | |
1280 | ||
1281 | /* Look up the VMA for the start of this memory slot */ | |
1282 | hva = memslot->userspace_addr; | |
1283 | down_read(¤t->mm->mmap_sem); | |
1284 | vma = find_vma(current->mm, hva); | |
1285 | if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO)) | |
1286 | goto up_out; | |
1287 | ||
1288 | psize = vma_kernel_pagesize(vma); | |
da9d1d7f | 1289 | porder = __ilog2(psize); |
c77162de PM |
1290 | |
1291 | /* Is this one of our preallocated RMAs? */ | |
1292 | if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops && | |
1293 | hva == vma->vm_start) | |
1294 | ri = vma->vm_file->private_data; | |
1295 | ||
1296 | up_read(¤t->mm->mmap_sem); | |
1297 | ||
1298 | if (!ri) { | |
1299 | /* On POWER7, use VRMA; on PPC970, give up */ | |
1300 | err = -EPERM; | |
1301 | if (cpu_has_feature(CPU_FTR_ARCH_201)) { | |
1302 | pr_err("KVM: CPU requires an RMO\n"); | |
1303 | goto out; | |
1304 | } | |
1305 | ||
da9d1d7f PM |
1306 | /* We can handle 4k, 64k or 16M pages in the VRMA */ |
1307 | err = -EINVAL; | |
1308 | if (!(psize == 0x1000 || psize == 0x10000 || | |
1309 | psize == 0x1000000)) | |
1310 | goto out; | |
1311 | ||
c77162de | 1312 | /* Update VRMASD field in the LPCR */ |
da9d1d7f | 1313 | senc = slb_pgsize_encoding(psize); |
697d3899 PM |
1314 | kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T | |
1315 | (VRMA_VSID << SLB_VSID_SHIFT_1T); | |
da9d1d7f PM |
1316 | lpcr = kvm->arch.lpcr & ~LPCR_VRMASD; |
1317 | lpcr |= senc << (LPCR_VRMASD_SH - 4); | |
c77162de PM |
1318 | kvm->arch.lpcr = lpcr; |
1319 | ||
1320 | /* Create HPTEs in the hash page table for the VRMA */ | |
da9d1d7f | 1321 | kvmppc_map_vrma(vcpu, memslot, porder); |
c77162de PM |
1322 | |
1323 | } else { | |
1324 | /* Set up to use an RMO region */ | |
1325 | rma_size = ri->npages; | |
1326 | if (rma_size > memslot->npages) | |
1327 | rma_size = memslot->npages; | |
1328 | rma_size <<= PAGE_SHIFT; | |
aa04b4cc | 1329 | rmls = lpcr_rmls(rma_size); |
c77162de | 1330 | err = -EINVAL; |
aa04b4cc | 1331 | if (rmls < 0) { |
c77162de PM |
1332 | pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size); |
1333 | goto out; | |
aa04b4cc PM |
1334 | } |
1335 | atomic_inc(&ri->use_count); | |
1336 | kvm->arch.rma = ri; | |
9e368f29 PM |
1337 | |
1338 | /* Update LPCR and RMOR */ | |
1339 | lpcr = kvm->arch.lpcr; | |
1340 | if (cpu_has_feature(CPU_FTR_ARCH_201)) { | |
1341 | /* PPC970; insert RMLS value (split field) in HID4 */ | |
1342 | lpcr &= ~((1ul << HID4_RMLS0_SH) | | |
1343 | (3ul << HID4_RMLS2_SH)); | |
1344 | lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) | | |
1345 | ((rmls & 3) << HID4_RMLS2_SH); | |
1346 | /* RMOR is also in HID4 */ | |
1347 | lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff) | |
1348 | << HID4_RMOR_SH; | |
1349 | } else { | |
1350 | /* POWER7 */ | |
1351 | lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L); | |
1352 | lpcr |= rmls << LPCR_RMLS_SH; | |
1353 | kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT; | |
1354 | } | |
aa04b4cc | 1355 | kvm->arch.lpcr = lpcr; |
c77162de | 1356 | pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n", |
aa04b4cc | 1357 | ri->base_pfn << PAGE_SHIFT, rma_size, lpcr); |
aa04b4cc | 1358 | |
c77162de | 1359 | /* Initialize phys addrs of pages in RMO */ |
da9d1d7f PM |
1360 | npages = ri->npages; |
1361 | porder = __ilog2(npages); | |
c77162de PM |
1362 | physp = kvm->arch.slot_phys[memslot->id]; |
1363 | spin_lock(&kvm->arch.slot_phys_lock); | |
1364 | for (i = 0; i < npages; ++i) | |
da9d1d7f | 1365 | physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) + porder; |
c77162de | 1366 | spin_unlock(&kvm->arch.slot_phys_lock); |
aa04b4cc PM |
1367 | } |
1368 | ||
c77162de PM |
1369 | /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */ |
1370 | smp_wmb(); | |
1371 | kvm->arch.rma_setup_done = 1; | |
1372 | err = 0; | |
1373 | out: | |
1374 | mutex_unlock(&kvm->lock); | |
1375 | return err; | |
b2b2f165 | 1376 | |
c77162de PM |
1377 | up_out: |
1378 | up_read(¤t->mm->mmap_sem); | |
1379 | goto out; | |
de56a948 PM |
1380 | } |
1381 | ||
1382 | int kvmppc_core_init_vm(struct kvm *kvm) | |
1383 | { | |
1384 | long r; | |
aa04b4cc | 1385 | unsigned long lpcr; |
de56a948 PM |
1386 | |
1387 | /* Allocate hashed page table */ | |
1388 | r = kvmppc_alloc_hpt(kvm); | |
54738c09 DG |
1389 | if (r) |
1390 | return r; | |
de56a948 | 1391 | |
54738c09 | 1392 | INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables); |
aa04b4cc | 1393 | |
aa04b4cc | 1394 | kvm->arch.rma = NULL; |
aa04b4cc | 1395 | |
9e368f29 | 1396 | kvm->arch.host_sdr1 = mfspr(SPRN_SDR1); |
aa04b4cc | 1397 | |
9e368f29 PM |
1398 | if (cpu_has_feature(CPU_FTR_ARCH_201)) { |
1399 | /* PPC970; HID4 is effectively the LPCR */ | |
1400 | unsigned long lpid = kvm->arch.lpid; | |
1401 | kvm->arch.host_lpid = 0; | |
1402 | kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4); | |
1403 | lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH)); | |
1404 | lpcr |= ((lpid >> 4) << HID4_LPID1_SH) | | |
1405 | ((lpid & 0xf) << HID4_LPID5_SH); | |
1406 | } else { | |
1407 | /* POWER7; init LPCR for virtual RMA mode */ | |
1408 | kvm->arch.host_lpid = mfspr(SPRN_LPID); | |
1409 | kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR); | |
1410 | lpcr &= LPCR_PECE | LPCR_LPES; | |
1411 | lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE | | |
697d3899 PM |
1412 | LPCR_VPM0 | LPCR_VPM1; |
1413 | kvm->arch.vrma_slb_v = SLB_VSID_B_1T | | |
1414 | (VRMA_VSID << SLB_VSID_SHIFT_1T); | |
9e368f29 PM |
1415 | } |
1416 | kvm->arch.lpcr = lpcr; | |
aa04b4cc | 1417 | |
342d3db7 | 1418 | kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206); |
c77162de | 1419 | spin_lock_init(&kvm->arch.slot_phys_lock); |
54738c09 | 1420 | return 0; |
de56a948 PM |
1421 | } |
1422 | ||
1423 | void kvmppc_core_destroy_vm(struct kvm *kvm) | |
1424 | { | |
aa04b4cc PM |
1425 | unsigned long i; |
1426 | ||
342d3db7 PM |
1427 | if (!kvm->arch.using_mmu_notifiers) |
1428 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) | |
1429 | unpin_slot(kvm, i); | |
b2b2f165 | 1430 | |
aa04b4cc PM |
1431 | if (kvm->arch.rma) { |
1432 | kvm_release_rma(kvm->arch.rma); | |
1433 | kvm->arch.rma = NULL; | |
1434 | } | |
1435 | ||
de56a948 | 1436 | kvmppc_free_hpt(kvm); |
54738c09 | 1437 | WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables)); |
de56a948 PM |
1438 | } |
1439 | ||
1440 | /* These are stubs for now */ | |
1441 | void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end) | |
1442 | { | |
1443 | } | |
1444 | ||
1445 | /* We don't need to emulate any privileged instructions or dcbz */ | |
1446 | int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu, | |
1447 | unsigned int inst, int *advance) | |
1448 | { | |
1449 | return EMULATE_FAIL; | |
1450 | } | |
1451 | ||
1452 | int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, int rs) | |
1453 | { | |
1454 | return EMULATE_FAIL; | |
1455 | } | |
1456 | ||
1457 | int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, int rt) | |
1458 | { | |
1459 | return EMULATE_FAIL; | |
1460 | } | |
1461 | ||
1462 | static int kvmppc_book3s_hv_init(void) | |
1463 | { | |
1464 | int r; | |
1465 | ||
1466 | r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1467 | ||
1468 | if (r) | |
1469 | return r; | |
1470 | ||
1471 | r = kvmppc_mmu_hv_init(); | |
1472 | ||
1473 | return r; | |
1474 | } | |
1475 | ||
1476 | static void kvmppc_book3s_hv_exit(void) | |
1477 | { | |
1478 | kvm_exit(); | |
1479 | } | |
1480 | ||
1481 | module_init(kvmppc_book3s_hv_init); | |
1482 | module_exit(kvmppc_book3s_hv_exit); |