]>
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> |
de56a948 PM |
48 | #include <linux/gfp.h> |
49 | #include <linux/sched.h> | |
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 | /* | |
697d3899 PM |
329 | * We get this if the guest accesses a page which it thinks |
330 | * it has mapped but which is not actually present, because | |
331 | * it is for an emulated I/O device. | |
332 | * Any other HDSI interrupt has been handled already. | |
de56a948 PM |
333 | */ |
334 | case BOOK3S_INTERRUPT_H_DATA_STORAGE: | |
697d3899 PM |
335 | r = kvmppc_book3s_hv_page_fault(run, vcpu, |
336 | vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); | |
de56a948 PM |
337 | break; |
338 | case BOOK3S_INTERRUPT_H_INST_STORAGE: | |
339 | kvmppc_inject_interrupt(vcpu, BOOK3S_INTERRUPT_INST_STORAGE, | |
697d3899 | 340 | vcpu->arch.shregs.msr & 0x58000000); |
de56a948 PM |
341 | r = RESUME_GUEST; |
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 | ||
401 | int kvmppc_core_check_processor_compat(void) | |
402 | { | |
9e368f29 | 403 | if (cpu_has_feature(CPU_FTR_HVMODE)) |
de56a948 PM |
404 | return 0; |
405 | return -EIO; | |
406 | } | |
407 | ||
408 | struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id) | |
409 | { | |
410 | struct kvm_vcpu *vcpu; | |
371fefd6 PM |
411 | int err = -EINVAL; |
412 | int core; | |
413 | struct kvmppc_vcore *vcore; | |
de56a948 | 414 | |
371fefd6 PM |
415 | core = id / threads_per_core; |
416 | if (core >= KVM_MAX_VCORES) | |
417 | goto out; | |
418 | ||
419 | err = -ENOMEM; | |
de56a948 PM |
420 | vcpu = kzalloc(sizeof(struct kvm_vcpu), GFP_KERNEL); |
421 | if (!vcpu) | |
422 | goto out; | |
423 | ||
424 | err = kvm_vcpu_init(vcpu, kvm, id); | |
425 | if (err) | |
426 | goto free_vcpu; | |
427 | ||
428 | vcpu->arch.shared = &vcpu->arch.shregs; | |
429 | vcpu->arch.last_cpu = -1; | |
430 | vcpu->arch.mmcr[0] = MMCR0_FC; | |
431 | vcpu->arch.ctrl = CTRL_RUNLATCH; | |
432 | /* default to host PVR, since we can't spoof it */ | |
433 | vcpu->arch.pvr = mfspr(SPRN_PVR); | |
434 | kvmppc_set_pvr(vcpu, vcpu->arch.pvr); | |
435 | ||
de56a948 PM |
436 | kvmppc_mmu_book3s_hv_init(vcpu); |
437 | ||
371fefd6 | 438 | /* |
19ccb76a | 439 | * We consider the vcpu stopped until we see the first run ioctl for it. |
371fefd6 | 440 | */ |
19ccb76a | 441 | vcpu->arch.state = KVMPPC_VCPU_STOPPED; |
371fefd6 PM |
442 | |
443 | init_waitqueue_head(&vcpu->arch.cpu_run); | |
444 | ||
445 | mutex_lock(&kvm->lock); | |
446 | vcore = kvm->arch.vcores[core]; | |
447 | if (!vcore) { | |
448 | vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL); | |
449 | if (vcore) { | |
450 | INIT_LIST_HEAD(&vcore->runnable_threads); | |
451 | spin_lock_init(&vcore->lock); | |
19ccb76a | 452 | init_waitqueue_head(&vcore->wq); |
371fefd6 PM |
453 | } |
454 | kvm->arch.vcores[core] = vcore; | |
455 | } | |
456 | mutex_unlock(&kvm->lock); | |
457 | ||
458 | if (!vcore) | |
459 | goto free_vcpu; | |
460 | ||
461 | spin_lock(&vcore->lock); | |
462 | ++vcore->num_threads; | |
371fefd6 PM |
463 | spin_unlock(&vcore->lock); |
464 | vcpu->arch.vcore = vcore; | |
465 | ||
af8f38b3 AG |
466 | vcpu->arch.cpu_type = KVM_CPU_3S_64; |
467 | kvmppc_sanity_check(vcpu); | |
468 | ||
de56a948 PM |
469 | return vcpu; |
470 | ||
471 | free_vcpu: | |
472 | kfree(vcpu); | |
473 | out: | |
474 | return ERR_PTR(err); | |
475 | } | |
476 | ||
477 | void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu) | |
478 | { | |
93e60249 PM |
479 | if (vcpu->arch.dtl) |
480 | kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.dtl); | |
481 | if (vcpu->arch.slb_shadow) | |
482 | kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.slb_shadow); | |
483 | if (vcpu->arch.vpa) | |
484 | kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.vpa); | |
de56a948 PM |
485 | kvm_vcpu_uninit(vcpu); |
486 | kfree(vcpu); | |
487 | } | |
488 | ||
19ccb76a | 489 | static void kvmppc_set_timer(struct kvm_vcpu *vcpu) |
371fefd6 | 490 | { |
19ccb76a | 491 | unsigned long dec_nsec, now; |
371fefd6 | 492 | |
19ccb76a PM |
493 | now = get_tb(); |
494 | if (now > vcpu->arch.dec_expires) { | |
495 | /* decrementer has already gone negative */ | |
496 | kvmppc_core_queue_dec(vcpu); | |
7e28e60e | 497 | kvmppc_core_prepare_to_enter(vcpu); |
19ccb76a | 498 | return; |
371fefd6 | 499 | } |
19ccb76a PM |
500 | dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC |
501 | / tb_ticks_per_sec; | |
502 | hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec), | |
503 | HRTIMER_MODE_REL); | |
504 | vcpu->arch.timer_running = 1; | |
371fefd6 PM |
505 | } |
506 | ||
19ccb76a | 507 | static void kvmppc_end_cede(struct kvm_vcpu *vcpu) |
371fefd6 | 508 | { |
19ccb76a PM |
509 | vcpu->arch.ceded = 0; |
510 | if (vcpu->arch.timer_running) { | |
511 | hrtimer_try_to_cancel(&vcpu->arch.dec_timer); | |
512 | vcpu->arch.timer_running = 0; | |
513 | } | |
371fefd6 PM |
514 | } |
515 | ||
de56a948 | 516 | extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu); |
371fefd6 | 517 | extern void xics_wake_cpu(int cpu); |
de56a948 | 518 | |
371fefd6 PM |
519 | static void kvmppc_remove_runnable(struct kvmppc_vcore *vc, |
520 | struct kvm_vcpu *vcpu) | |
de56a948 | 521 | { |
371fefd6 | 522 | struct kvm_vcpu *v; |
de56a948 | 523 | |
371fefd6 PM |
524 | if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) |
525 | return; | |
526 | vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; | |
527 | --vc->n_runnable; | |
19ccb76a | 528 | ++vc->n_busy; |
371fefd6 PM |
529 | /* decrement the physical thread id of each following vcpu */ |
530 | v = vcpu; | |
531 | list_for_each_entry_continue(v, &vc->runnable_threads, arch.run_list) | |
532 | --v->arch.ptid; | |
533 | list_del(&vcpu->arch.run_list); | |
534 | } | |
535 | ||
536 | static void kvmppc_start_thread(struct kvm_vcpu *vcpu) | |
537 | { | |
538 | int cpu; | |
539 | struct paca_struct *tpaca; | |
540 | struct kvmppc_vcore *vc = vcpu->arch.vcore; | |
541 | ||
19ccb76a PM |
542 | if (vcpu->arch.timer_running) { |
543 | hrtimer_try_to_cancel(&vcpu->arch.dec_timer); | |
544 | vcpu->arch.timer_running = 0; | |
545 | } | |
371fefd6 PM |
546 | cpu = vc->pcpu + vcpu->arch.ptid; |
547 | tpaca = &paca[cpu]; | |
548 | tpaca->kvm_hstate.kvm_vcpu = vcpu; | |
549 | tpaca->kvm_hstate.kvm_vcore = vc; | |
19ccb76a PM |
550 | tpaca->kvm_hstate.napping = 0; |
551 | vcpu->cpu = vc->pcpu; | |
371fefd6 | 552 | smp_wmb(); |
251da038 | 553 | #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP) |
371fefd6 PM |
554 | if (vcpu->arch.ptid) { |
555 | tpaca->cpu_start = 0x80; | |
371fefd6 PM |
556 | wmb(); |
557 | xics_wake_cpu(cpu); | |
558 | ++vc->n_woken; | |
de56a948 | 559 | } |
371fefd6 PM |
560 | #endif |
561 | } | |
de56a948 | 562 | |
371fefd6 PM |
563 | static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc) |
564 | { | |
565 | int i; | |
566 | ||
567 | HMT_low(); | |
568 | i = 0; | |
569 | while (vc->nap_count < vc->n_woken) { | |
570 | if (++i >= 1000000) { | |
571 | pr_err("kvmppc_wait_for_nap timeout %d %d\n", | |
572 | vc->nap_count, vc->n_woken); | |
573 | break; | |
574 | } | |
575 | cpu_relax(); | |
576 | } | |
577 | HMT_medium(); | |
578 | } | |
579 | ||
580 | /* | |
581 | * Check that we are on thread 0 and that any other threads in | |
582 | * this core are off-line. | |
583 | */ | |
584 | static int on_primary_thread(void) | |
585 | { | |
586 | int cpu = smp_processor_id(); | |
587 | int thr = cpu_thread_in_core(cpu); | |
588 | ||
589 | if (thr) | |
590 | return 0; | |
591 | while (++thr < threads_per_core) | |
592 | if (cpu_online(cpu + thr)) | |
593 | return 0; | |
594 | return 1; | |
595 | } | |
596 | ||
597 | /* | |
598 | * Run a set of guest threads on a physical core. | |
599 | * Called with vc->lock held. | |
600 | */ | |
601 | static int kvmppc_run_core(struct kvmppc_vcore *vc) | |
602 | { | |
19ccb76a | 603 | struct kvm_vcpu *vcpu, *vcpu0, *vnext; |
371fefd6 PM |
604 | long ret; |
605 | u64 now; | |
19ccb76a | 606 | int ptid; |
371fefd6 PM |
607 | |
608 | /* don't start if any threads have a signal pending */ | |
609 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
610 | if (signal_pending(vcpu->arch.run_task)) | |
611 | return 0; | |
de56a948 PM |
612 | |
613 | /* | |
614 | * Make sure we are running on thread 0, and that | |
615 | * secondary threads are offline. | |
616 | * XXX we should also block attempts to bring any | |
617 | * secondary threads online. | |
618 | */ | |
371fefd6 PM |
619 | if (threads_per_core > 1 && !on_primary_thread()) { |
620 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
621 | vcpu->arch.ret = -EBUSY; | |
622 | goto out; | |
de56a948 PM |
623 | } |
624 | ||
19ccb76a PM |
625 | /* |
626 | * Assign physical thread IDs, first to non-ceded vcpus | |
627 | * and then to ceded ones. | |
628 | */ | |
629 | ptid = 0; | |
630 | vcpu0 = NULL; | |
631 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) { | |
632 | if (!vcpu->arch.ceded) { | |
633 | if (!ptid) | |
634 | vcpu0 = vcpu; | |
635 | vcpu->arch.ptid = ptid++; | |
636 | } | |
637 | } | |
638 | if (!vcpu0) | |
639 | return 0; /* nothing to run */ | |
640 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
641 | if (vcpu->arch.ceded) | |
642 | vcpu->arch.ptid = ptid++; | |
643 | ||
371fefd6 PM |
644 | vc->n_woken = 0; |
645 | vc->nap_count = 0; | |
646 | vc->entry_exit_count = 0; | |
19ccb76a | 647 | vc->vcore_state = VCORE_RUNNING; |
371fefd6 PM |
648 | vc->in_guest = 0; |
649 | vc->pcpu = smp_processor_id(); | |
19ccb76a | 650 | vc->napping_threads = 0; |
371fefd6 PM |
651 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) |
652 | kvmppc_start_thread(vcpu); | |
371fefd6 | 653 | |
19ccb76a | 654 | preempt_disable(); |
371fefd6 | 655 | spin_unlock(&vc->lock); |
de56a948 | 656 | |
371fefd6 | 657 | kvm_guest_enter(); |
19ccb76a | 658 | __kvmppc_vcore_entry(NULL, vcpu0); |
de56a948 | 659 | |
371fefd6 | 660 | spin_lock(&vc->lock); |
19ccb76a PM |
661 | /* disable sending of IPIs on virtual external irqs */ |
662 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
663 | vcpu->cpu = -1; | |
664 | /* wait for secondary threads to finish writing their state to memory */ | |
371fefd6 PM |
665 | if (vc->nap_count < vc->n_woken) |
666 | kvmppc_wait_for_nap(vc); | |
667 | /* prevent other vcpu threads from doing kvmppc_start_thread() now */ | |
19ccb76a | 668 | vc->vcore_state = VCORE_EXITING; |
371fefd6 PM |
669 | spin_unlock(&vc->lock); |
670 | ||
671 | /* make sure updates to secondary vcpu structs are visible now */ | |
672 | smp_mb(); | |
de56a948 PM |
673 | kvm_guest_exit(); |
674 | ||
675 | preempt_enable(); | |
676 | kvm_resched(vcpu); | |
677 | ||
678 | now = get_tb(); | |
371fefd6 PM |
679 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) { |
680 | /* cancel pending dec exception if dec is positive */ | |
681 | if (now < vcpu->arch.dec_expires && | |
682 | kvmppc_core_pending_dec(vcpu)) | |
683 | kvmppc_core_dequeue_dec(vcpu); | |
19ccb76a PM |
684 | |
685 | ret = RESUME_GUEST; | |
686 | if (vcpu->arch.trap) | |
687 | ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu, | |
688 | vcpu->arch.run_task); | |
689 | ||
371fefd6 PM |
690 | vcpu->arch.ret = ret; |
691 | vcpu->arch.trap = 0; | |
19ccb76a PM |
692 | |
693 | if (vcpu->arch.ceded) { | |
694 | if (ret != RESUME_GUEST) | |
695 | kvmppc_end_cede(vcpu); | |
696 | else | |
697 | kvmppc_set_timer(vcpu); | |
698 | } | |
371fefd6 | 699 | } |
de56a948 | 700 | |
371fefd6 | 701 | spin_lock(&vc->lock); |
de56a948 | 702 | out: |
19ccb76a | 703 | vc->vcore_state = VCORE_INACTIVE; |
371fefd6 PM |
704 | list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads, |
705 | arch.run_list) { | |
706 | if (vcpu->arch.ret != RESUME_GUEST) { | |
707 | kvmppc_remove_runnable(vc, vcpu); | |
708 | wake_up(&vcpu->arch.cpu_run); | |
709 | } | |
710 | } | |
711 | ||
712 | return 1; | |
713 | } | |
714 | ||
19ccb76a PM |
715 | /* |
716 | * Wait for some other vcpu thread to execute us, and | |
717 | * wake us up when we need to handle something in the host. | |
718 | */ | |
719 | static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state) | |
371fefd6 | 720 | { |
371fefd6 PM |
721 | DEFINE_WAIT(wait); |
722 | ||
19ccb76a PM |
723 | prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state); |
724 | if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) | |
725 | schedule(); | |
726 | finish_wait(&vcpu->arch.cpu_run, &wait); | |
727 | } | |
728 | ||
729 | /* | |
730 | * All the vcpus in this vcore are idle, so wait for a decrementer | |
731 | * or external interrupt to one of the vcpus. vc->lock is held. | |
732 | */ | |
733 | static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc) | |
734 | { | |
735 | DEFINE_WAIT(wait); | |
736 | struct kvm_vcpu *v; | |
737 | int all_idle = 1; | |
738 | ||
739 | prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE); | |
740 | vc->vcore_state = VCORE_SLEEPING; | |
741 | spin_unlock(&vc->lock); | |
742 | list_for_each_entry(v, &vc->runnable_threads, arch.run_list) { | |
743 | if (!v->arch.ceded || v->arch.pending_exceptions) { | |
744 | all_idle = 0; | |
745 | break; | |
746 | } | |
371fefd6 | 747 | } |
19ccb76a PM |
748 | if (all_idle) |
749 | schedule(); | |
750 | finish_wait(&vc->wq, &wait); | |
751 | spin_lock(&vc->lock); | |
752 | vc->vcore_state = VCORE_INACTIVE; | |
753 | } | |
371fefd6 | 754 | |
19ccb76a PM |
755 | static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu) |
756 | { | |
757 | int n_ceded; | |
758 | int prev_state; | |
759 | struct kvmppc_vcore *vc; | |
760 | struct kvm_vcpu *v, *vn; | |
9e368f29 | 761 | |
371fefd6 PM |
762 | kvm_run->exit_reason = 0; |
763 | vcpu->arch.ret = RESUME_GUEST; | |
764 | vcpu->arch.trap = 0; | |
765 | ||
371fefd6 PM |
766 | /* |
767 | * Synchronize with other threads in this virtual core | |
768 | */ | |
769 | vc = vcpu->arch.vcore; | |
770 | spin_lock(&vc->lock); | |
19ccb76a | 771 | vcpu->arch.ceded = 0; |
371fefd6 PM |
772 | vcpu->arch.run_task = current; |
773 | vcpu->arch.kvm_run = kvm_run; | |
19ccb76a PM |
774 | prev_state = vcpu->arch.state; |
775 | vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; | |
371fefd6 PM |
776 | list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads); |
777 | ++vc->n_runnable; | |
778 | ||
19ccb76a PM |
779 | /* |
780 | * This happens the first time this is called for a vcpu. | |
781 | * If the vcore is already running, we may be able to start | |
782 | * this thread straight away and have it join in. | |
783 | */ | |
784 | if (prev_state == KVMPPC_VCPU_STOPPED) { | |
785 | if (vc->vcore_state == VCORE_RUNNING && | |
786 | VCORE_EXIT_COUNT(vc) == 0) { | |
787 | vcpu->arch.ptid = vc->n_runnable - 1; | |
788 | kvmppc_start_thread(vcpu); | |
371fefd6 PM |
789 | } |
790 | ||
19ccb76a PM |
791 | } else if (prev_state == KVMPPC_VCPU_BUSY_IN_HOST) |
792 | --vc->n_busy; | |
371fefd6 | 793 | |
19ccb76a PM |
794 | while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && |
795 | !signal_pending(current)) { | |
796 | if (vc->n_busy || vc->vcore_state != VCORE_INACTIVE) { | |
797 | spin_unlock(&vc->lock); | |
798 | kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE); | |
799 | spin_lock(&vc->lock); | |
800 | continue; | |
801 | } | |
802 | n_ceded = 0; | |
803 | list_for_each_entry(v, &vc->runnable_threads, arch.run_list) | |
804 | n_ceded += v->arch.ceded; | |
805 | if (n_ceded == vc->n_runnable) | |
806 | kvmppc_vcore_blocked(vc); | |
807 | else | |
808 | kvmppc_run_core(vc); | |
809 | ||
810 | list_for_each_entry_safe(v, vn, &vc->runnable_threads, | |
811 | arch.run_list) { | |
7e28e60e | 812 | kvmppc_core_prepare_to_enter(v); |
19ccb76a PM |
813 | if (signal_pending(v->arch.run_task)) { |
814 | kvmppc_remove_runnable(vc, v); | |
815 | v->stat.signal_exits++; | |
816 | v->arch.kvm_run->exit_reason = KVM_EXIT_INTR; | |
817 | v->arch.ret = -EINTR; | |
818 | wake_up(&v->arch.cpu_run); | |
819 | } | |
820 | } | |
821 | } | |
371fefd6 | 822 | |
19ccb76a PM |
823 | if (signal_pending(current)) { |
824 | if (vc->vcore_state == VCORE_RUNNING || | |
825 | vc->vcore_state == VCORE_EXITING) { | |
826 | spin_unlock(&vc->lock); | |
827 | kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE); | |
828 | spin_lock(&vc->lock); | |
829 | } | |
830 | if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { | |
831 | kvmppc_remove_runnable(vc, vcpu); | |
832 | vcpu->stat.signal_exits++; | |
833 | kvm_run->exit_reason = KVM_EXIT_INTR; | |
834 | vcpu->arch.ret = -EINTR; | |
835 | } | |
371fefd6 PM |
836 | } |
837 | ||
371fefd6 | 838 | spin_unlock(&vc->lock); |
371fefd6 | 839 | return vcpu->arch.ret; |
de56a948 PM |
840 | } |
841 | ||
a8606e20 PM |
842 | int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu) |
843 | { | |
844 | int r; | |
845 | ||
af8f38b3 AG |
846 | if (!vcpu->arch.sane) { |
847 | run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
848 | return -EINVAL; | |
849 | } | |
850 | ||
25051b5a SW |
851 | kvmppc_core_prepare_to_enter(vcpu); |
852 | ||
19ccb76a PM |
853 | /* No need to go into the guest when all we'll do is come back out */ |
854 | if (signal_pending(current)) { | |
855 | run->exit_reason = KVM_EXIT_INTR; | |
856 | return -EINTR; | |
857 | } | |
858 | ||
c77162de PM |
859 | /* On the first time here, set up VRMA or RMA */ |
860 | if (!vcpu->kvm->arch.rma_setup_done) { | |
861 | r = kvmppc_hv_setup_rma(vcpu); | |
862 | if (r) | |
863 | return r; | |
864 | } | |
19ccb76a PM |
865 | |
866 | flush_fp_to_thread(current); | |
867 | flush_altivec_to_thread(current); | |
868 | flush_vsx_to_thread(current); | |
869 | vcpu->arch.wqp = &vcpu->arch.vcore->wq; | |
870 | ||
a8606e20 PM |
871 | do { |
872 | r = kvmppc_run_vcpu(run, vcpu); | |
873 | ||
874 | if (run->exit_reason == KVM_EXIT_PAPR_HCALL && | |
875 | !(vcpu->arch.shregs.msr & MSR_PR)) { | |
876 | r = kvmppc_pseries_do_hcall(vcpu); | |
7e28e60e | 877 | kvmppc_core_prepare_to_enter(vcpu); |
a8606e20 PM |
878 | } |
879 | } while (r == RESUME_GUEST); | |
880 | return r; | |
881 | } | |
882 | ||
54738c09 DG |
883 | static long kvmppc_stt_npages(unsigned long window_size) |
884 | { | |
885 | return ALIGN((window_size >> SPAPR_TCE_SHIFT) | |
886 | * sizeof(u64), PAGE_SIZE) / PAGE_SIZE; | |
887 | } | |
888 | ||
889 | static void release_spapr_tce_table(struct kvmppc_spapr_tce_table *stt) | |
890 | { | |
891 | struct kvm *kvm = stt->kvm; | |
892 | int i; | |
893 | ||
894 | mutex_lock(&kvm->lock); | |
895 | list_del(&stt->list); | |
896 | for (i = 0; i < kvmppc_stt_npages(stt->window_size); i++) | |
897 | __free_page(stt->pages[i]); | |
898 | kfree(stt); | |
899 | mutex_unlock(&kvm->lock); | |
900 | ||
901 | kvm_put_kvm(kvm); | |
902 | } | |
903 | ||
904 | static int kvm_spapr_tce_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
905 | { | |
906 | struct kvmppc_spapr_tce_table *stt = vma->vm_file->private_data; | |
907 | struct page *page; | |
908 | ||
909 | if (vmf->pgoff >= kvmppc_stt_npages(stt->window_size)) | |
910 | return VM_FAULT_SIGBUS; | |
911 | ||
912 | page = stt->pages[vmf->pgoff]; | |
913 | get_page(page); | |
914 | vmf->page = page; | |
915 | return 0; | |
916 | } | |
917 | ||
918 | static const struct vm_operations_struct kvm_spapr_tce_vm_ops = { | |
919 | .fault = kvm_spapr_tce_fault, | |
920 | }; | |
921 | ||
922 | static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma) | |
923 | { | |
924 | vma->vm_ops = &kvm_spapr_tce_vm_ops; | |
925 | return 0; | |
926 | } | |
927 | ||
928 | static int kvm_spapr_tce_release(struct inode *inode, struct file *filp) | |
929 | { | |
930 | struct kvmppc_spapr_tce_table *stt = filp->private_data; | |
931 | ||
932 | release_spapr_tce_table(stt); | |
933 | return 0; | |
934 | } | |
935 | ||
936 | static struct file_operations kvm_spapr_tce_fops = { | |
937 | .mmap = kvm_spapr_tce_mmap, | |
938 | .release = kvm_spapr_tce_release, | |
939 | }; | |
940 | ||
941 | long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm, | |
942 | struct kvm_create_spapr_tce *args) | |
943 | { | |
944 | struct kvmppc_spapr_tce_table *stt = NULL; | |
945 | long npages; | |
946 | int ret = -ENOMEM; | |
947 | int i; | |
948 | ||
949 | /* Check this LIOBN hasn't been previously allocated */ | |
950 | list_for_each_entry(stt, &kvm->arch.spapr_tce_tables, list) { | |
951 | if (stt->liobn == args->liobn) | |
952 | return -EBUSY; | |
953 | } | |
954 | ||
955 | npages = kvmppc_stt_npages(args->window_size); | |
956 | ||
957 | stt = kzalloc(sizeof(*stt) + npages* sizeof(struct page *), | |
958 | GFP_KERNEL); | |
959 | if (!stt) | |
960 | goto fail; | |
961 | ||
962 | stt->liobn = args->liobn; | |
963 | stt->window_size = args->window_size; | |
964 | stt->kvm = kvm; | |
965 | ||
966 | for (i = 0; i < npages; i++) { | |
967 | stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
968 | if (!stt->pages[i]) | |
969 | goto fail; | |
970 | } | |
971 | ||
972 | kvm_get_kvm(kvm); | |
973 | ||
974 | mutex_lock(&kvm->lock); | |
975 | list_add(&stt->list, &kvm->arch.spapr_tce_tables); | |
976 | ||
977 | mutex_unlock(&kvm->lock); | |
978 | ||
979 | return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops, | |
980 | stt, O_RDWR); | |
981 | ||
982 | fail: | |
983 | if (stt) { | |
984 | for (i = 0; i < npages; i++) | |
985 | if (stt->pages[i]) | |
986 | __free_page(stt->pages[i]); | |
987 | ||
988 | kfree(stt); | |
989 | } | |
990 | return ret; | |
991 | } | |
992 | ||
aa04b4cc | 993 | /* Work out RMLS (real mode limit selector) field value for a given RMA size. |
9e368f29 | 994 | Assumes POWER7 or PPC970. */ |
aa04b4cc PM |
995 | static inline int lpcr_rmls(unsigned long rma_size) |
996 | { | |
997 | switch (rma_size) { | |
998 | case 32ul << 20: /* 32 MB */ | |
9e368f29 PM |
999 | if (cpu_has_feature(CPU_FTR_ARCH_206)) |
1000 | return 8; /* only supported on POWER7 */ | |
1001 | return -1; | |
aa04b4cc PM |
1002 | case 64ul << 20: /* 64 MB */ |
1003 | return 3; | |
1004 | case 128ul << 20: /* 128 MB */ | |
1005 | return 7; | |
1006 | case 256ul << 20: /* 256 MB */ | |
1007 | return 4; | |
1008 | case 1ul << 30: /* 1 GB */ | |
1009 | return 2; | |
1010 | case 16ul << 30: /* 16 GB */ | |
1011 | return 1; | |
1012 | case 256ul << 30: /* 256 GB */ | |
1013 | return 0; | |
1014 | default: | |
1015 | return -1; | |
1016 | } | |
1017 | } | |
1018 | ||
1019 | static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
1020 | { | |
1021 | struct kvmppc_rma_info *ri = vma->vm_file->private_data; | |
1022 | struct page *page; | |
1023 | ||
1024 | if (vmf->pgoff >= ri->npages) | |
1025 | return VM_FAULT_SIGBUS; | |
1026 | ||
1027 | page = pfn_to_page(ri->base_pfn + vmf->pgoff); | |
1028 | get_page(page); | |
1029 | vmf->page = page; | |
1030 | return 0; | |
1031 | } | |
1032 | ||
1033 | static const struct vm_operations_struct kvm_rma_vm_ops = { | |
1034 | .fault = kvm_rma_fault, | |
1035 | }; | |
1036 | ||
1037 | static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma) | |
1038 | { | |
1039 | vma->vm_flags |= VM_RESERVED; | |
1040 | vma->vm_ops = &kvm_rma_vm_ops; | |
1041 | return 0; | |
1042 | } | |
1043 | ||
1044 | static int kvm_rma_release(struct inode *inode, struct file *filp) | |
1045 | { | |
1046 | struct kvmppc_rma_info *ri = filp->private_data; | |
1047 | ||
1048 | kvm_release_rma(ri); | |
1049 | return 0; | |
1050 | } | |
1051 | ||
1052 | static struct file_operations kvm_rma_fops = { | |
1053 | .mmap = kvm_rma_mmap, | |
1054 | .release = kvm_rma_release, | |
1055 | }; | |
1056 | ||
1057 | long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret) | |
1058 | { | |
1059 | struct kvmppc_rma_info *ri; | |
1060 | long fd; | |
1061 | ||
1062 | ri = kvm_alloc_rma(); | |
1063 | if (!ri) | |
1064 | return -ENOMEM; | |
1065 | ||
1066 | fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR); | |
1067 | if (fd < 0) | |
1068 | kvm_release_rma(ri); | |
1069 | ||
1070 | ret->rma_size = ri->npages << PAGE_SHIFT; | |
1071 | return fd; | |
1072 | } | |
1073 | ||
da9d1d7f PM |
1074 | static unsigned long slb_pgsize_encoding(unsigned long psize) |
1075 | { | |
1076 | unsigned long senc = 0; | |
1077 | ||
1078 | if (psize > 0x1000) { | |
1079 | senc = SLB_VSID_L; | |
1080 | if (psize == 0x10000) | |
1081 | senc |= SLB_VSID_LP_01; | |
1082 | } | |
1083 | return senc; | |
1084 | } | |
1085 | ||
de56a948 PM |
1086 | int kvmppc_core_prepare_memory_region(struct kvm *kvm, |
1087 | struct kvm_userspace_memory_region *mem) | |
1088 | { | |
c77162de | 1089 | unsigned long npages; |
b2b2f165 | 1090 | unsigned long *phys; |
aa04b4cc | 1091 | |
b2b2f165 | 1092 | /* Allocate a slot_phys array */ |
da9d1d7f | 1093 | npages = mem->memory_size >> PAGE_SHIFT; |
b2b2f165 PM |
1094 | phys = kvm->arch.slot_phys[mem->slot]; |
1095 | if (!phys) { | |
1096 | phys = vzalloc(npages * sizeof(unsigned long)); | |
1097 | if (!phys) | |
1098 | return -ENOMEM; | |
1099 | kvm->arch.slot_phys[mem->slot] = phys; | |
1100 | kvm->arch.slot_npages[mem->slot] = npages; | |
1101 | } | |
aa04b4cc | 1102 | |
c77162de PM |
1103 | return 0; |
1104 | } | |
aa04b4cc | 1105 | |
c77162de PM |
1106 | static void unpin_slot(struct kvm *kvm, int slot_id) |
1107 | { | |
1108 | unsigned long *physp; | |
1109 | unsigned long j, npages, pfn; | |
1110 | struct page *page; | |
1111 | ||
1112 | physp = kvm->arch.slot_phys[slot_id]; | |
1113 | npages = kvm->arch.slot_npages[slot_id]; | |
1114 | if (physp) { | |
1115 | spin_lock(&kvm->arch.slot_phys_lock); | |
1116 | for (j = 0; j < npages; j++) { | |
1117 | if (!(physp[j] & KVMPPC_GOT_PAGE)) | |
1118 | continue; | |
1119 | pfn = physp[j] >> PAGE_SHIFT; | |
1120 | page = pfn_to_page(pfn); | |
da9d1d7f PM |
1121 | if (PageHuge(page)) |
1122 | page = compound_head(page); | |
c77162de PM |
1123 | SetPageDirty(page); |
1124 | put_page(page); | |
9e368f29 | 1125 | } |
c77162de PM |
1126 | kvm->arch.slot_phys[slot_id] = NULL; |
1127 | spin_unlock(&kvm->arch.slot_phys_lock); | |
1128 | vfree(physp); | |
aa04b4cc | 1129 | } |
c77162de PM |
1130 | } |
1131 | ||
1132 | void kvmppc_core_commit_memory_region(struct kvm *kvm, | |
1133 | struct kvm_userspace_memory_region *mem) | |
1134 | { | |
1135 | } | |
1136 | ||
1137 | static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu) | |
1138 | { | |
1139 | int err = 0; | |
1140 | struct kvm *kvm = vcpu->kvm; | |
1141 | struct kvmppc_rma_info *ri = NULL; | |
1142 | unsigned long hva; | |
1143 | struct kvm_memory_slot *memslot; | |
1144 | struct vm_area_struct *vma; | |
da9d1d7f | 1145 | unsigned long lpcr, senc; |
c77162de PM |
1146 | unsigned long psize, porder; |
1147 | unsigned long rma_size; | |
1148 | unsigned long rmls; | |
1149 | unsigned long *physp; | |
da9d1d7f | 1150 | unsigned long i, npages; |
c77162de PM |
1151 | |
1152 | mutex_lock(&kvm->lock); | |
1153 | if (kvm->arch.rma_setup_done) | |
1154 | goto out; /* another vcpu beat us to it */ | |
aa04b4cc | 1155 | |
c77162de PM |
1156 | /* Look up the memslot for guest physical address 0 */ |
1157 | memslot = gfn_to_memslot(kvm, 0); | |
aa04b4cc | 1158 | |
c77162de PM |
1159 | /* We must have some memory at 0 by now */ |
1160 | err = -EINVAL; | |
1161 | if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) | |
1162 | goto out; | |
1163 | ||
1164 | /* Look up the VMA for the start of this memory slot */ | |
1165 | hva = memslot->userspace_addr; | |
1166 | down_read(¤t->mm->mmap_sem); | |
1167 | vma = find_vma(current->mm, hva); | |
1168 | if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO)) | |
1169 | goto up_out; | |
1170 | ||
1171 | psize = vma_kernel_pagesize(vma); | |
da9d1d7f | 1172 | porder = __ilog2(psize); |
c77162de PM |
1173 | |
1174 | /* Is this one of our preallocated RMAs? */ | |
1175 | if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops && | |
1176 | hva == vma->vm_start) | |
1177 | ri = vma->vm_file->private_data; | |
1178 | ||
1179 | up_read(¤t->mm->mmap_sem); | |
1180 | ||
1181 | if (!ri) { | |
1182 | /* On POWER7, use VRMA; on PPC970, give up */ | |
1183 | err = -EPERM; | |
1184 | if (cpu_has_feature(CPU_FTR_ARCH_201)) { | |
1185 | pr_err("KVM: CPU requires an RMO\n"); | |
1186 | goto out; | |
1187 | } | |
1188 | ||
da9d1d7f PM |
1189 | /* We can handle 4k, 64k or 16M pages in the VRMA */ |
1190 | err = -EINVAL; | |
1191 | if (!(psize == 0x1000 || psize == 0x10000 || | |
1192 | psize == 0x1000000)) | |
1193 | goto out; | |
1194 | ||
c77162de | 1195 | /* Update VRMASD field in the LPCR */ |
da9d1d7f | 1196 | senc = slb_pgsize_encoding(psize); |
697d3899 PM |
1197 | kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T | |
1198 | (VRMA_VSID << SLB_VSID_SHIFT_1T); | |
da9d1d7f PM |
1199 | lpcr = kvm->arch.lpcr & ~LPCR_VRMASD; |
1200 | lpcr |= senc << (LPCR_VRMASD_SH - 4); | |
c77162de PM |
1201 | kvm->arch.lpcr = lpcr; |
1202 | ||
1203 | /* Create HPTEs in the hash page table for the VRMA */ | |
da9d1d7f | 1204 | kvmppc_map_vrma(vcpu, memslot, porder); |
c77162de PM |
1205 | |
1206 | } else { | |
1207 | /* Set up to use an RMO region */ | |
1208 | rma_size = ri->npages; | |
1209 | if (rma_size > memslot->npages) | |
1210 | rma_size = memslot->npages; | |
1211 | rma_size <<= PAGE_SHIFT; | |
aa04b4cc | 1212 | rmls = lpcr_rmls(rma_size); |
c77162de | 1213 | err = -EINVAL; |
aa04b4cc | 1214 | if (rmls < 0) { |
c77162de PM |
1215 | pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size); |
1216 | goto out; | |
aa04b4cc PM |
1217 | } |
1218 | atomic_inc(&ri->use_count); | |
1219 | kvm->arch.rma = ri; | |
9e368f29 PM |
1220 | |
1221 | /* Update LPCR and RMOR */ | |
1222 | lpcr = kvm->arch.lpcr; | |
1223 | if (cpu_has_feature(CPU_FTR_ARCH_201)) { | |
1224 | /* PPC970; insert RMLS value (split field) in HID4 */ | |
1225 | lpcr &= ~((1ul << HID4_RMLS0_SH) | | |
1226 | (3ul << HID4_RMLS2_SH)); | |
1227 | lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) | | |
1228 | ((rmls & 3) << HID4_RMLS2_SH); | |
1229 | /* RMOR is also in HID4 */ | |
1230 | lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff) | |
1231 | << HID4_RMOR_SH; | |
1232 | } else { | |
1233 | /* POWER7 */ | |
1234 | lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L); | |
1235 | lpcr |= rmls << LPCR_RMLS_SH; | |
1236 | kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT; | |
1237 | } | |
aa04b4cc | 1238 | kvm->arch.lpcr = lpcr; |
c77162de | 1239 | pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n", |
aa04b4cc | 1240 | ri->base_pfn << PAGE_SHIFT, rma_size, lpcr); |
aa04b4cc | 1241 | |
c77162de | 1242 | /* Initialize phys addrs of pages in RMO */ |
da9d1d7f PM |
1243 | npages = ri->npages; |
1244 | porder = __ilog2(npages); | |
c77162de PM |
1245 | physp = kvm->arch.slot_phys[memslot->id]; |
1246 | spin_lock(&kvm->arch.slot_phys_lock); | |
1247 | for (i = 0; i < npages; ++i) | |
da9d1d7f | 1248 | physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) + porder; |
c77162de | 1249 | spin_unlock(&kvm->arch.slot_phys_lock); |
aa04b4cc PM |
1250 | } |
1251 | ||
c77162de PM |
1252 | /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */ |
1253 | smp_wmb(); | |
1254 | kvm->arch.rma_setup_done = 1; | |
1255 | err = 0; | |
1256 | out: | |
1257 | mutex_unlock(&kvm->lock); | |
1258 | return err; | |
b2b2f165 | 1259 | |
c77162de PM |
1260 | up_out: |
1261 | up_read(¤t->mm->mmap_sem); | |
1262 | goto out; | |
de56a948 PM |
1263 | } |
1264 | ||
1265 | int kvmppc_core_init_vm(struct kvm *kvm) | |
1266 | { | |
1267 | long r; | |
aa04b4cc | 1268 | unsigned long lpcr; |
de56a948 PM |
1269 | |
1270 | /* Allocate hashed page table */ | |
1271 | r = kvmppc_alloc_hpt(kvm); | |
54738c09 DG |
1272 | if (r) |
1273 | return r; | |
de56a948 | 1274 | |
54738c09 | 1275 | INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables); |
aa04b4cc | 1276 | |
aa04b4cc | 1277 | kvm->arch.rma = NULL; |
aa04b4cc | 1278 | |
9e368f29 | 1279 | kvm->arch.host_sdr1 = mfspr(SPRN_SDR1); |
aa04b4cc | 1280 | |
9e368f29 PM |
1281 | if (cpu_has_feature(CPU_FTR_ARCH_201)) { |
1282 | /* PPC970; HID4 is effectively the LPCR */ | |
1283 | unsigned long lpid = kvm->arch.lpid; | |
1284 | kvm->arch.host_lpid = 0; | |
1285 | kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4); | |
1286 | lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH)); | |
1287 | lpcr |= ((lpid >> 4) << HID4_LPID1_SH) | | |
1288 | ((lpid & 0xf) << HID4_LPID5_SH); | |
1289 | } else { | |
1290 | /* POWER7; init LPCR for virtual RMA mode */ | |
1291 | kvm->arch.host_lpid = mfspr(SPRN_LPID); | |
1292 | kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR); | |
1293 | lpcr &= LPCR_PECE | LPCR_LPES; | |
1294 | lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE | | |
697d3899 PM |
1295 | LPCR_VPM0 | LPCR_VPM1; |
1296 | kvm->arch.vrma_slb_v = SLB_VSID_B_1T | | |
1297 | (VRMA_VSID << SLB_VSID_SHIFT_1T); | |
9e368f29 PM |
1298 | } |
1299 | kvm->arch.lpcr = lpcr; | |
aa04b4cc | 1300 | |
c77162de | 1301 | spin_lock_init(&kvm->arch.slot_phys_lock); |
54738c09 | 1302 | return 0; |
de56a948 PM |
1303 | } |
1304 | ||
1305 | void kvmppc_core_destroy_vm(struct kvm *kvm) | |
1306 | { | |
aa04b4cc PM |
1307 | unsigned long i; |
1308 | ||
b2b2f165 PM |
1309 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
1310 | unpin_slot(kvm, i); | |
1311 | ||
aa04b4cc PM |
1312 | if (kvm->arch.rma) { |
1313 | kvm_release_rma(kvm->arch.rma); | |
1314 | kvm->arch.rma = NULL; | |
1315 | } | |
1316 | ||
de56a948 | 1317 | kvmppc_free_hpt(kvm); |
54738c09 | 1318 | WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables)); |
de56a948 PM |
1319 | } |
1320 | ||
1321 | /* These are stubs for now */ | |
1322 | void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end) | |
1323 | { | |
1324 | } | |
1325 | ||
1326 | /* We don't need to emulate any privileged instructions or dcbz */ | |
1327 | int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu, | |
1328 | unsigned int inst, int *advance) | |
1329 | { | |
1330 | return EMULATE_FAIL; | |
1331 | } | |
1332 | ||
1333 | int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, int rs) | |
1334 | { | |
1335 | return EMULATE_FAIL; | |
1336 | } | |
1337 | ||
1338 | int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, int rt) | |
1339 | { | |
1340 | return EMULATE_FAIL; | |
1341 | } | |
1342 | ||
1343 | static int kvmppc_book3s_hv_init(void) | |
1344 | { | |
1345 | int r; | |
1346 | ||
1347 | r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1348 | ||
1349 | if (r) | |
1350 | return r; | |
1351 | ||
1352 | r = kvmppc_mmu_hv_init(); | |
1353 | ||
1354 | return r; | |
1355 | } | |
1356 | ||
1357 | static void kvmppc_book3s_hv_exit(void) | |
1358 | { | |
1359 | kvm_exit(); | |
1360 | } | |
1361 | ||
1362 | module_init(kvmppc_book3s_hv_init); | |
1363 | module_exit(kvmppc_book3s_hv_exit); |