]>
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 | 48 | #include <linux/gfp.h> |
de56a948 PM |
49 | #include <linux/vmalloc.h> |
50 | #include <linux/highmem.h> | |
51 | ||
aa04b4cc PM |
52 | /* |
53 | * For now, limit memory to 64GB and require it to be large pages. | |
54 | * This value is chosen because it makes the ram_pginfo array be | |
55 | * 64kB in size, which is about as large as we want to be trying | |
56 | * to allocate with kmalloc. | |
57 | */ | |
58 | #define MAX_MEM_ORDER 36 | |
59 | ||
60 | #define LARGE_PAGE_ORDER 24 /* 16MB pages */ | |
61 | ||
de56a948 PM |
62 | /* #define EXIT_DEBUG */ |
63 | /* #define EXIT_DEBUG_SIMPLE */ | |
64 | /* #define EXIT_DEBUG_INT */ | |
65 | ||
19ccb76a PM |
66 | static void kvmppc_end_cede(struct kvm_vcpu *vcpu); |
67 | ||
de56a948 PM |
68 | void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
69 | { | |
70 | local_paca->kvm_hstate.kvm_vcpu = vcpu; | |
371fefd6 | 71 | local_paca->kvm_hstate.kvm_vcore = vcpu->arch.vcore; |
de56a948 PM |
72 | } |
73 | ||
74 | void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu) | |
75 | { | |
76 | } | |
77 | ||
de56a948 PM |
78 | void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr) |
79 | { | |
80 | vcpu->arch.shregs.msr = msr; | |
19ccb76a | 81 | kvmppc_end_cede(vcpu); |
de56a948 PM |
82 | } |
83 | ||
84 | void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr) | |
85 | { | |
86 | vcpu->arch.pvr = pvr; | |
87 | } | |
88 | ||
89 | void kvmppc_dump_regs(struct kvm_vcpu *vcpu) | |
90 | { | |
91 | int r; | |
92 | ||
93 | pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id); | |
94 | pr_err("pc = %.16lx msr = %.16llx trap = %x\n", | |
95 | vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap); | |
96 | for (r = 0; r < 16; ++r) | |
97 | pr_err("r%2d = %.16lx r%d = %.16lx\n", | |
98 | r, kvmppc_get_gpr(vcpu, r), | |
99 | r+16, kvmppc_get_gpr(vcpu, r+16)); | |
100 | pr_err("ctr = %.16lx lr = %.16lx\n", | |
101 | vcpu->arch.ctr, vcpu->arch.lr); | |
102 | pr_err("srr0 = %.16llx srr1 = %.16llx\n", | |
103 | vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1); | |
104 | pr_err("sprg0 = %.16llx sprg1 = %.16llx\n", | |
105 | vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1); | |
106 | pr_err("sprg2 = %.16llx sprg3 = %.16llx\n", | |
107 | vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3); | |
108 | pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n", | |
109 | vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr); | |
110 | pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar); | |
111 | pr_err("fault dar = %.16lx dsisr = %.8x\n", | |
112 | vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); | |
113 | pr_err("SLB (%d entries):\n", vcpu->arch.slb_max); | |
114 | for (r = 0; r < vcpu->arch.slb_max; ++r) | |
115 | pr_err(" ESID = %.16llx VSID = %.16llx\n", | |
116 | vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv); | |
117 | pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n", | |
aa04b4cc | 118 | vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1, |
de56a948 PM |
119 | vcpu->arch.last_inst); |
120 | } | |
121 | ||
a8606e20 PM |
122 | struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id) |
123 | { | |
124 | int r; | |
125 | struct kvm_vcpu *v, *ret = NULL; | |
126 | ||
127 | mutex_lock(&kvm->lock); | |
128 | kvm_for_each_vcpu(r, v, kvm) { | |
129 | if (v->vcpu_id == id) { | |
130 | ret = v; | |
131 | break; | |
132 | } | |
133 | } | |
134 | mutex_unlock(&kvm->lock); | |
135 | return ret; | |
136 | } | |
137 | ||
138 | static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa) | |
139 | { | |
140 | vpa->shared_proc = 1; | |
141 | vpa->yield_count = 1; | |
142 | } | |
143 | ||
144 | static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu, | |
145 | unsigned long flags, | |
146 | unsigned long vcpuid, unsigned long vpa) | |
147 | { | |
148 | struct kvm *kvm = vcpu->kvm; | |
149 | unsigned long pg_index, ra, len; | |
150 | unsigned long pg_offset; | |
151 | void *va; | |
152 | struct kvm_vcpu *tvcpu; | |
153 | ||
154 | tvcpu = kvmppc_find_vcpu(kvm, vcpuid); | |
155 | if (!tvcpu) | |
156 | return H_PARAMETER; | |
157 | ||
158 | flags >>= 63 - 18; | |
159 | flags &= 7; | |
160 | if (flags == 0 || flags == 4) | |
161 | return H_PARAMETER; | |
162 | if (flags < 4) { | |
163 | if (vpa & 0x7f) | |
164 | return H_PARAMETER; | |
165 | /* registering new area; convert logical addr to real */ | |
166 | pg_index = vpa >> kvm->arch.ram_porder; | |
167 | pg_offset = vpa & (kvm->arch.ram_psize - 1); | |
168 | if (pg_index >= kvm->arch.ram_npages) | |
169 | return H_PARAMETER; | |
170 | if (kvm->arch.ram_pginfo[pg_index].pfn == 0) | |
171 | return H_PARAMETER; | |
172 | ra = kvm->arch.ram_pginfo[pg_index].pfn << PAGE_SHIFT; | |
173 | ra |= pg_offset; | |
174 | va = __va(ra); | |
175 | if (flags <= 1) | |
176 | len = *(unsigned short *)(va + 4); | |
177 | else | |
178 | len = *(unsigned int *)(va + 4); | |
179 | if (pg_offset + len > kvm->arch.ram_psize) | |
180 | return H_PARAMETER; | |
181 | switch (flags) { | |
182 | case 1: /* register VPA */ | |
183 | if (len < 640) | |
184 | return H_PARAMETER; | |
185 | tvcpu->arch.vpa = va; | |
186 | init_vpa(vcpu, va); | |
187 | break; | |
188 | case 2: /* register DTL */ | |
189 | if (len < 48) | |
190 | return H_PARAMETER; | |
191 | if (!tvcpu->arch.vpa) | |
192 | return H_RESOURCE; | |
193 | len -= len % 48; | |
194 | tvcpu->arch.dtl = va; | |
195 | tvcpu->arch.dtl_end = va + len; | |
196 | break; | |
197 | case 3: /* register SLB shadow buffer */ | |
198 | if (len < 8) | |
199 | return H_PARAMETER; | |
200 | if (!tvcpu->arch.vpa) | |
201 | return H_RESOURCE; | |
202 | tvcpu->arch.slb_shadow = va; | |
203 | len = (len - 16) / 16; | |
204 | tvcpu->arch.slb_shadow = va; | |
205 | break; | |
206 | } | |
207 | } else { | |
208 | switch (flags) { | |
209 | case 5: /* unregister VPA */ | |
210 | if (tvcpu->arch.slb_shadow || tvcpu->arch.dtl) | |
211 | return H_RESOURCE; | |
212 | tvcpu->arch.vpa = NULL; | |
213 | break; | |
214 | case 6: /* unregister DTL */ | |
215 | tvcpu->arch.dtl = NULL; | |
216 | break; | |
217 | case 7: /* unregister SLB shadow buffer */ | |
218 | tvcpu->arch.slb_shadow = NULL; | |
219 | break; | |
220 | } | |
221 | } | |
222 | return H_SUCCESS; | |
223 | } | |
224 | ||
225 | int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu) | |
226 | { | |
227 | unsigned long req = kvmppc_get_gpr(vcpu, 3); | |
228 | unsigned long target, ret = H_SUCCESS; | |
229 | struct kvm_vcpu *tvcpu; | |
230 | ||
231 | switch (req) { | |
232 | case H_CEDE: | |
a8606e20 PM |
233 | break; |
234 | case H_PROD: | |
235 | target = kvmppc_get_gpr(vcpu, 4); | |
236 | tvcpu = kvmppc_find_vcpu(vcpu->kvm, target); | |
237 | if (!tvcpu) { | |
238 | ret = H_PARAMETER; | |
239 | break; | |
240 | } | |
241 | tvcpu->arch.prodded = 1; | |
242 | smp_mb(); | |
243 | if (vcpu->arch.ceded) { | |
244 | if (waitqueue_active(&vcpu->wq)) { | |
245 | wake_up_interruptible(&vcpu->wq); | |
246 | vcpu->stat.halt_wakeup++; | |
247 | } | |
248 | } | |
249 | break; | |
250 | case H_CONFER: | |
251 | break; | |
252 | case H_REGISTER_VPA: | |
253 | ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4), | |
254 | kvmppc_get_gpr(vcpu, 5), | |
255 | kvmppc_get_gpr(vcpu, 6)); | |
256 | break; | |
257 | default: | |
258 | return RESUME_HOST; | |
259 | } | |
260 | kvmppc_set_gpr(vcpu, 3, ret); | |
261 | vcpu->arch.hcall_needed = 0; | |
262 | return RESUME_GUEST; | |
263 | } | |
264 | ||
de56a948 PM |
265 | static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu, |
266 | struct task_struct *tsk) | |
267 | { | |
268 | int r = RESUME_HOST; | |
269 | ||
270 | vcpu->stat.sum_exits++; | |
271 | ||
272 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
273 | run->ready_for_interrupt_injection = 1; | |
274 | switch (vcpu->arch.trap) { | |
275 | /* We're good on these - the host merely wanted to get our attention */ | |
276 | case BOOK3S_INTERRUPT_HV_DECREMENTER: | |
277 | vcpu->stat.dec_exits++; | |
278 | r = RESUME_GUEST; | |
279 | break; | |
280 | case BOOK3S_INTERRUPT_EXTERNAL: | |
281 | vcpu->stat.ext_intr_exits++; | |
282 | r = RESUME_GUEST; | |
283 | break; | |
284 | case BOOK3S_INTERRUPT_PERFMON: | |
285 | r = RESUME_GUEST; | |
286 | break; | |
287 | case BOOK3S_INTERRUPT_PROGRAM: | |
288 | { | |
289 | ulong flags; | |
290 | /* | |
291 | * Normally program interrupts are delivered directly | |
292 | * to the guest by the hardware, but we can get here | |
293 | * as a result of a hypervisor emulation interrupt | |
294 | * (e40) getting turned into a 700 by BML RTAS. | |
295 | */ | |
296 | flags = vcpu->arch.shregs.msr & 0x1f0000ull; | |
297 | kvmppc_core_queue_program(vcpu, flags); | |
298 | r = RESUME_GUEST; | |
299 | break; | |
300 | } | |
301 | case BOOK3S_INTERRUPT_SYSCALL: | |
302 | { | |
303 | /* hcall - punt to userspace */ | |
304 | int i; | |
305 | ||
306 | if (vcpu->arch.shregs.msr & MSR_PR) { | |
307 | /* sc 1 from userspace - reflect to guest syscall */ | |
308 | kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL); | |
309 | r = RESUME_GUEST; | |
310 | break; | |
311 | } | |
312 | run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3); | |
313 | for (i = 0; i < 9; ++i) | |
314 | run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i); | |
315 | run->exit_reason = KVM_EXIT_PAPR_HCALL; | |
316 | vcpu->arch.hcall_needed = 1; | |
317 | r = RESUME_HOST; | |
318 | break; | |
319 | } | |
320 | /* | |
321 | * We get these next two if the guest does a bad real-mode access, | |
322 | * as we have enabled VRMA (virtualized real mode area) mode in the | |
323 | * LPCR. We just generate an appropriate DSI/ISI to the guest. | |
324 | */ | |
325 | case BOOK3S_INTERRUPT_H_DATA_STORAGE: | |
326 | vcpu->arch.shregs.dsisr = vcpu->arch.fault_dsisr; | |
327 | vcpu->arch.shregs.dar = vcpu->arch.fault_dar; | |
328 | kvmppc_inject_interrupt(vcpu, BOOK3S_INTERRUPT_DATA_STORAGE, 0); | |
329 | r = RESUME_GUEST; | |
330 | break; | |
331 | case BOOK3S_INTERRUPT_H_INST_STORAGE: | |
332 | kvmppc_inject_interrupt(vcpu, BOOK3S_INTERRUPT_INST_STORAGE, | |
333 | 0x08000000); | |
334 | r = RESUME_GUEST; | |
335 | break; | |
336 | /* | |
337 | * This occurs if the guest executes an illegal instruction. | |
338 | * We just generate a program interrupt to the guest, since | |
339 | * we don't emulate any guest instructions at this stage. | |
340 | */ | |
341 | case BOOK3S_INTERRUPT_H_EMUL_ASSIST: | |
342 | kvmppc_core_queue_program(vcpu, 0x80000); | |
343 | r = RESUME_GUEST; | |
344 | break; | |
345 | default: | |
346 | kvmppc_dump_regs(vcpu); | |
347 | printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n", | |
348 | vcpu->arch.trap, kvmppc_get_pc(vcpu), | |
349 | vcpu->arch.shregs.msr); | |
350 | r = RESUME_HOST; | |
351 | BUG(); | |
352 | break; | |
353 | } | |
354 | ||
de56a948 PM |
355 | return r; |
356 | } | |
357 | ||
358 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, | |
359 | struct kvm_sregs *sregs) | |
360 | { | |
361 | int i; | |
362 | ||
363 | sregs->pvr = vcpu->arch.pvr; | |
364 | ||
365 | memset(sregs, 0, sizeof(struct kvm_sregs)); | |
366 | for (i = 0; i < vcpu->arch.slb_max; i++) { | |
367 | sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige; | |
368 | sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv; | |
369 | } | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, | |
375 | struct kvm_sregs *sregs) | |
376 | { | |
377 | int i, j; | |
378 | ||
379 | kvmppc_set_pvr(vcpu, sregs->pvr); | |
380 | ||
381 | j = 0; | |
382 | for (i = 0; i < vcpu->arch.slb_nr; i++) { | |
383 | if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) { | |
384 | vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe; | |
385 | vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv; | |
386 | ++j; | |
387 | } | |
388 | } | |
389 | vcpu->arch.slb_max = j; | |
390 | ||
391 | return 0; | |
392 | } | |
393 | ||
394 | int kvmppc_core_check_processor_compat(void) | |
395 | { | |
9e368f29 | 396 | if (cpu_has_feature(CPU_FTR_HVMODE)) |
de56a948 PM |
397 | return 0; |
398 | return -EIO; | |
399 | } | |
400 | ||
401 | struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id) | |
402 | { | |
403 | struct kvm_vcpu *vcpu; | |
371fefd6 PM |
404 | int err = -EINVAL; |
405 | int core; | |
406 | struct kvmppc_vcore *vcore; | |
de56a948 | 407 | |
371fefd6 PM |
408 | core = id / threads_per_core; |
409 | if (core >= KVM_MAX_VCORES) | |
410 | goto out; | |
411 | ||
412 | err = -ENOMEM; | |
de56a948 PM |
413 | vcpu = kzalloc(sizeof(struct kvm_vcpu), GFP_KERNEL); |
414 | if (!vcpu) | |
415 | goto out; | |
416 | ||
417 | err = kvm_vcpu_init(vcpu, kvm, id); | |
418 | if (err) | |
419 | goto free_vcpu; | |
420 | ||
421 | vcpu->arch.shared = &vcpu->arch.shregs; | |
422 | vcpu->arch.last_cpu = -1; | |
423 | vcpu->arch.mmcr[0] = MMCR0_FC; | |
424 | vcpu->arch.ctrl = CTRL_RUNLATCH; | |
425 | /* default to host PVR, since we can't spoof it */ | |
426 | vcpu->arch.pvr = mfspr(SPRN_PVR); | |
427 | kvmppc_set_pvr(vcpu, vcpu->arch.pvr); | |
428 | ||
de56a948 PM |
429 | kvmppc_mmu_book3s_hv_init(vcpu); |
430 | ||
371fefd6 | 431 | /* |
19ccb76a | 432 | * We consider the vcpu stopped until we see the first run ioctl for it. |
371fefd6 | 433 | */ |
19ccb76a | 434 | vcpu->arch.state = KVMPPC_VCPU_STOPPED; |
371fefd6 PM |
435 | |
436 | init_waitqueue_head(&vcpu->arch.cpu_run); | |
437 | ||
438 | mutex_lock(&kvm->lock); | |
439 | vcore = kvm->arch.vcores[core]; | |
440 | if (!vcore) { | |
441 | vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL); | |
442 | if (vcore) { | |
443 | INIT_LIST_HEAD(&vcore->runnable_threads); | |
444 | spin_lock_init(&vcore->lock); | |
19ccb76a | 445 | init_waitqueue_head(&vcore->wq); |
371fefd6 PM |
446 | } |
447 | kvm->arch.vcores[core] = vcore; | |
448 | } | |
449 | mutex_unlock(&kvm->lock); | |
450 | ||
451 | if (!vcore) | |
452 | goto free_vcpu; | |
453 | ||
454 | spin_lock(&vcore->lock); | |
455 | ++vcore->num_threads; | |
371fefd6 PM |
456 | spin_unlock(&vcore->lock); |
457 | vcpu->arch.vcore = vcore; | |
458 | ||
af8f38b3 AG |
459 | vcpu->arch.cpu_type = KVM_CPU_3S_64; |
460 | kvmppc_sanity_check(vcpu); | |
461 | ||
de56a948 PM |
462 | return vcpu; |
463 | ||
464 | free_vcpu: | |
465 | kfree(vcpu); | |
466 | out: | |
467 | return ERR_PTR(err); | |
468 | } | |
469 | ||
470 | void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu) | |
471 | { | |
472 | kvm_vcpu_uninit(vcpu); | |
473 | kfree(vcpu); | |
474 | } | |
475 | ||
19ccb76a | 476 | static void kvmppc_set_timer(struct kvm_vcpu *vcpu) |
371fefd6 | 477 | { |
19ccb76a | 478 | unsigned long dec_nsec, now; |
371fefd6 | 479 | |
19ccb76a PM |
480 | now = get_tb(); |
481 | if (now > vcpu->arch.dec_expires) { | |
482 | /* decrementer has already gone negative */ | |
483 | kvmppc_core_queue_dec(vcpu); | |
484 | kvmppc_core_deliver_interrupts(vcpu); | |
485 | return; | |
371fefd6 | 486 | } |
19ccb76a PM |
487 | dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC |
488 | / tb_ticks_per_sec; | |
489 | hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec), | |
490 | HRTIMER_MODE_REL); | |
491 | vcpu->arch.timer_running = 1; | |
371fefd6 PM |
492 | } |
493 | ||
19ccb76a | 494 | static void kvmppc_end_cede(struct kvm_vcpu *vcpu) |
371fefd6 | 495 | { |
19ccb76a PM |
496 | vcpu->arch.ceded = 0; |
497 | if (vcpu->arch.timer_running) { | |
498 | hrtimer_try_to_cancel(&vcpu->arch.dec_timer); | |
499 | vcpu->arch.timer_running = 0; | |
500 | } | |
371fefd6 PM |
501 | } |
502 | ||
de56a948 | 503 | extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu); |
371fefd6 | 504 | extern void xics_wake_cpu(int cpu); |
de56a948 | 505 | |
371fefd6 PM |
506 | static void kvmppc_remove_runnable(struct kvmppc_vcore *vc, |
507 | struct kvm_vcpu *vcpu) | |
de56a948 | 508 | { |
371fefd6 | 509 | struct kvm_vcpu *v; |
de56a948 | 510 | |
371fefd6 PM |
511 | if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) |
512 | return; | |
513 | vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; | |
514 | --vc->n_runnable; | |
19ccb76a | 515 | ++vc->n_busy; |
371fefd6 PM |
516 | /* decrement the physical thread id of each following vcpu */ |
517 | v = vcpu; | |
518 | list_for_each_entry_continue(v, &vc->runnable_threads, arch.run_list) | |
519 | --v->arch.ptid; | |
520 | list_del(&vcpu->arch.run_list); | |
521 | } | |
522 | ||
523 | static void kvmppc_start_thread(struct kvm_vcpu *vcpu) | |
524 | { | |
525 | int cpu; | |
526 | struct paca_struct *tpaca; | |
527 | struct kvmppc_vcore *vc = vcpu->arch.vcore; | |
528 | ||
19ccb76a PM |
529 | if (vcpu->arch.timer_running) { |
530 | hrtimer_try_to_cancel(&vcpu->arch.dec_timer); | |
531 | vcpu->arch.timer_running = 0; | |
532 | } | |
371fefd6 PM |
533 | cpu = vc->pcpu + vcpu->arch.ptid; |
534 | tpaca = &paca[cpu]; | |
535 | tpaca->kvm_hstate.kvm_vcpu = vcpu; | |
536 | tpaca->kvm_hstate.kvm_vcore = vc; | |
19ccb76a PM |
537 | tpaca->kvm_hstate.napping = 0; |
538 | vcpu->cpu = vc->pcpu; | |
371fefd6 | 539 | smp_wmb(); |
251da038 | 540 | #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP) |
371fefd6 PM |
541 | if (vcpu->arch.ptid) { |
542 | tpaca->cpu_start = 0x80; | |
371fefd6 PM |
543 | wmb(); |
544 | xics_wake_cpu(cpu); | |
545 | ++vc->n_woken; | |
de56a948 | 546 | } |
371fefd6 PM |
547 | #endif |
548 | } | |
de56a948 | 549 | |
371fefd6 PM |
550 | static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc) |
551 | { | |
552 | int i; | |
553 | ||
554 | HMT_low(); | |
555 | i = 0; | |
556 | while (vc->nap_count < vc->n_woken) { | |
557 | if (++i >= 1000000) { | |
558 | pr_err("kvmppc_wait_for_nap timeout %d %d\n", | |
559 | vc->nap_count, vc->n_woken); | |
560 | break; | |
561 | } | |
562 | cpu_relax(); | |
563 | } | |
564 | HMT_medium(); | |
565 | } | |
566 | ||
567 | /* | |
568 | * Check that we are on thread 0 and that any other threads in | |
569 | * this core are off-line. | |
570 | */ | |
571 | static int on_primary_thread(void) | |
572 | { | |
573 | int cpu = smp_processor_id(); | |
574 | int thr = cpu_thread_in_core(cpu); | |
575 | ||
576 | if (thr) | |
577 | return 0; | |
578 | while (++thr < threads_per_core) | |
579 | if (cpu_online(cpu + thr)) | |
580 | return 0; | |
581 | return 1; | |
582 | } | |
583 | ||
584 | /* | |
585 | * Run a set of guest threads on a physical core. | |
586 | * Called with vc->lock held. | |
587 | */ | |
588 | static int kvmppc_run_core(struct kvmppc_vcore *vc) | |
589 | { | |
19ccb76a | 590 | struct kvm_vcpu *vcpu, *vcpu0, *vnext; |
371fefd6 PM |
591 | long ret; |
592 | u64 now; | |
19ccb76a | 593 | int ptid; |
371fefd6 PM |
594 | |
595 | /* don't start if any threads have a signal pending */ | |
596 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
597 | if (signal_pending(vcpu->arch.run_task)) | |
598 | return 0; | |
de56a948 PM |
599 | |
600 | /* | |
601 | * Make sure we are running on thread 0, and that | |
602 | * secondary threads are offline. | |
603 | * XXX we should also block attempts to bring any | |
604 | * secondary threads online. | |
605 | */ | |
371fefd6 PM |
606 | if (threads_per_core > 1 && !on_primary_thread()) { |
607 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
608 | vcpu->arch.ret = -EBUSY; | |
609 | goto out; | |
de56a948 PM |
610 | } |
611 | ||
19ccb76a PM |
612 | /* |
613 | * Assign physical thread IDs, first to non-ceded vcpus | |
614 | * and then to ceded ones. | |
615 | */ | |
616 | ptid = 0; | |
617 | vcpu0 = NULL; | |
618 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) { | |
619 | if (!vcpu->arch.ceded) { | |
620 | if (!ptid) | |
621 | vcpu0 = vcpu; | |
622 | vcpu->arch.ptid = ptid++; | |
623 | } | |
624 | } | |
625 | if (!vcpu0) | |
626 | return 0; /* nothing to run */ | |
627 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
628 | if (vcpu->arch.ceded) | |
629 | vcpu->arch.ptid = ptid++; | |
630 | ||
371fefd6 PM |
631 | vc->n_woken = 0; |
632 | vc->nap_count = 0; | |
633 | vc->entry_exit_count = 0; | |
19ccb76a | 634 | vc->vcore_state = VCORE_RUNNING; |
371fefd6 PM |
635 | vc->in_guest = 0; |
636 | vc->pcpu = smp_processor_id(); | |
19ccb76a | 637 | vc->napping_threads = 0; |
371fefd6 PM |
638 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) |
639 | kvmppc_start_thread(vcpu); | |
371fefd6 | 640 | |
19ccb76a | 641 | preempt_disable(); |
371fefd6 | 642 | spin_unlock(&vc->lock); |
de56a948 | 643 | |
371fefd6 | 644 | kvm_guest_enter(); |
19ccb76a | 645 | __kvmppc_vcore_entry(NULL, vcpu0); |
de56a948 | 646 | |
371fefd6 | 647 | spin_lock(&vc->lock); |
19ccb76a PM |
648 | /* disable sending of IPIs on virtual external irqs */ |
649 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) | |
650 | vcpu->cpu = -1; | |
651 | /* wait for secondary threads to finish writing their state to memory */ | |
371fefd6 PM |
652 | if (vc->nap_count < vc->n_woken) |
653 | kvmppc_wait_for_nap(vc); | |
654 | /* prevent other vcpu threads from doing kvmppc_start_thread() now */ | |
19ccb76a | 655 | vc->vcore_state = VCORE_EXITING; |
371fefd6 PM |
656 | spin_unlock(&vc->lock); |
657 | ||
658 | /* make sure updates to secondary vcpu structs are visible now */ | |
659 | smp_mb(); | |
de56a948 PM |
660 | kvm_guest_exit(); |
661 | ||
662 | preempt_enable(); | |
663 | kvm_resched(vcpu); | |
664 | ||
665 | now = get_tb(); | |
371fefd6 PM |
666 | list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) { |
667 | /* cancel pending dec exception if dec is positive */ | |
668 | if (now < vcpu->arch.dec_expires && | |
669 | kvmppc_core_pending_dec(vcpu)) | |
670 | kvmppc_core_dequeue_dec(vcpu); | |
19ccb76a PM |
671 | |
672 | ret = RESUME_GUEST; | |
673 | if (vcpu->arch.trap) | |
674 | ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu, | |
675 | vcpu->arch.run_task); | |
676 | ||
371fefd6 PM |
677 | vcpu->arch.ret = ret; |
678 | vcpu->arch.trap = 0; | |
19ccb76a PM |
679 | |
680 | if (vcpu->arch.ceded) { | |
681 | if (ret != RESUME_GUEST) | |
682 | kvmppc_end_cede(vcpu); | |
683 | else | |
684 | kvmppc_set_timer(vcpu); | |
685 | } | |
371fefd6 | 686 | } |
de56a948 | 687 | |
371fefd6 | 688 | spin_lock(&vc->lock); |
de56a948 | 689 | out: |
19ccb76a | 690 | vc->vcore_state = VCORE_INACTIVE; |
371fefd6 PM |
691 | list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads, |
692 | arch.run_list) { | |
693 | if (vcpu->arch.ret != RESUME_GUEST) { | |
694 | kvmppc_remove_runnable(vc, vcpu); | |
695 | wake_up(&vcpu->arch.cpu_run); | |
696 | } | |
697 | } | |
698 | ||
699 | return 1; | |
700 | } | |
701 | ||
19ccb76a PM |
702 | /* |
703 | * Wait for some other vcpu thread to execute us, and | |
704 | * wake us up when we need to handle something in the host. | |
705 | */ | |
706 | static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state) | |
371fefd6 | 707 | { |
371fefd6 PM |
708 | DEFINE_WAIT(wait); |
709 | ||
19ccb76a PM |
710 | prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state); |
711 | if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) | |
712 | schedule(); | |
713 | finish_wait(&vcpu->arch.cpu_run, &wait); | |
714 | } | |
715 | ||
716 | /* | |
717 | * All the vcpus in this vcore are idle, so wait for a decrementer | |
718 | * or external interrupt to one of the vcpus. vc->lock is held. | |
719 | */ | |
720 | static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc) | |
721 | { | |
722 | DEFINE_WAIT(wait); | |
723 | struct kvm_vcpu *v; | |
724 | int all_idle = 1; | |
725 | ||
726 | prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE); | |
727 | vc->vcore_state = VCORE_SLEEPING; | |
728 | spin_unlock(&vc->lock); | |
729 | list_for_each_entry(v, &vc->runnable_threads, arch.run_list) { | |
730 | if (!v->arch.ceded || v->arch.pending_exceptions) { | |
731 | all_idle = 0; | |
732 | break; | |
733 | } | |
371fefd6 | 734 | } |
19ccb76a PM |
735 | if (all_idle) |
736 | schedule(); | |
737 | finish_wait(&vc->wq, &wait); | |
738 | spin_lock(&vc->lock); | |
739 | vc->vcore_state = VCORE_INACTIVE; | |
740 | } | |
371fefd6 | 741 | |
19ccb76a PM |
742 | static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu) |
743 | { | |
744 | int n_ceded; | |
745 | int prev_state; | |
746 | struct kvmppc_vcore *vc; | |
747 | struct kvm_vcpu *v, *vn; | |
9e368f29 | 748 | |
371fefd6 PM |
749 | kvm_run->exit_reason = 0; |
750 | vcpu->arch.ret = RESUME_GUEST; | |
751 | vcpu->arch.trap = 0; | |
752 | ||
371fefd6 PM |
753 | /* |
754 | * Synchronize with other threads in this virtual core | |
755 | */ | |
756 | vc = vcpu->arch.vcore; | |
757 | spin_lock(&vc->lock); | |
19ccb76a | 758 | vcpu->arch.ceded = 0; |
371fefd6 PM |
759 | vcpu->arch.run_task = current; |
760 | vcpu->arch.kvm_run = kvm_run; | |
19ccb76a PM |
761 | prev_state = vcpu->arch.state; |
762 | vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; | |
371fefd6 PM |
763 | list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads); |
764 | ++vc->n_runnable; | |
765 | ||
19ccb76a PM |
766 | /* |
767 | * This happens the first time this is called for a vcpu. | |
768 | * If the vcore is already running, we may be able to start | |
769 | * this thread straight away and have it join in. | |
770 | */ | |
771 | if (prev_state == KVMPPC_VCPU_STOPPED) { | |
772 | if (vc->vcore_state == VCORE_RUNNING && | |
773 | VCORE_EXIT_COUNT(vc) == 0) { | |
774 | vcpu->arch.ptid = vc->n_runnable - 1; | |
775 | kvmppc_start_thread(vcpu); | |
371fefd6 PM |
776 | } |
777 | ||
19ccb76a PM |
778 | } else if (prev_state == KVMPPC_VCPU_BUSY_IN_HOST) |
779 | --vc->n_busy; | |
371fefd6 | 780 | |
19ccb76a PM |
781 | while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && |
782 | !signal_pending(current)) { | |
783 | if (vc->n_busy || vc->vcore_state != VCORE_INACTIVE) { | |
784 | spin_unlock(&vc->lock); | |
785 | kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE); | |
786 | spin_lock(&vc->lock); | |
787 | continue; | |
788 | } | |
789 | n_ceded = 0; | |
790 | list_for_each_entry(v, &vc->runnable_threads, arch.run_list) | |
791 | n_ceded += v->arch.ceded; | |
792 | if (n_ceded == vc->n_runnable) | |
793 | kvmppc_vcore_blocked(vc); | |
794 | else | |
795 | kvmppc_run_core(vc); | |
796 | ||
797 | list_for_each_entry_safe(v, vn, &vc->runnable_threads, | |
798 | arch.run_list) { | |
799 | kvmppc_core_deliver_interrupts(v); | |
800 | if (signal_pending(v->arch.run_task)) { | |
801 | kvmppc_remove_runnable(vc, v); | |
802 | v->stat.signal_exits++; | |
803 | v->arch.kvm_run->exit_reason = KVM_EXIT_INTR; | |
804 | v->arch.ret = -EINTR; | |
805 | wake_up(&v->arch.cpu_run); | |
806 | } | |
807 | } | |
808 | } | |
371fefd6 | 809 | |
19ccb76a PM |
810 | if (signal_pending(current)) { |
811 | if (vc->vcore_state == VCORE_RUNNING || | |
812 | vc->vcore_state == VCORE_EXITING) { | |
813 | spin_unlock(&vc->lock); | |
814 | kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE); | |
815 | spin_lock(&vc->lock); | |
816 | } | |
817 | if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { | |
818 | kvmppc_remove_runnable(vc, vcpu); | |
819 | vcpu->stat.signal_exits++; | |
820 | kvm_run->exit_reason = KVM_EXIT_INTR; | |
821 | vcpu->arch.ret = -EINTR; | |
822 | } | |
371fefd6 PM |
823 | } |
824 | ||
371fefd6 | 825 | spin_unlock(&vc->lock); |
371fefd6 | 826 | return vcpu->arch.ret; |
de56a948 PM |
827 | } |
828 | ||
a8606e20 PM |
829 | int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu) |
830 | { | |
831 | int r; | |
832 | ||
af8f38b3 AG |
833 | if (!vcpu->arch.sane) { |
834 | run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
835 | return -EINVAL; | |
836 | } | |
837 | ||
19ccb76a PM |
838 | /* No need to go into the guest when all we'll do is come back out */ |
839 | if (signal_pending(current)) { | |
840 | run->exit_reason = KVM_EXIT_INTR; | |
841 | return -EINTR; | |
842 | } | |
843 | ||
844 | /* On PPC970, check that we have an RMA region */ | |
845 | if (!vcpu->kvm->arch.rma && cpu_has_feature(CPU_FTR_ARCH_201)) | |
846 | return -EPERM; | |
847 | ||
848 | flush_fp_to_thread(current); | |
849 | flush_altivec_to_thread(current); | |
850 | flush_vsx_to_thread(current); | |
851 | vcpu->arch.wqp = &vcpu->arch.vcore->wq; | |
852 | ||
a8606e20 PM |
853 | do { |
854 | r = kvmppc_run_vcpu(run, vcpu); | |
855 | ||
856 | if (run->exit_reason == KVM_EXIT_PAPR_HCALL && | |
857 | !(vcpu->arch.shregs.msr & MSR_PR)) { | |
858 | r = kvmppc_pseries_do_hcall(vcpu); | |
859 | kvmppc_core_deliver_interrupts(vcpu); | |
860 | } | |
861 | } while (r == RESUME_GUEST); | |
862 | return r; | |
863 | } | |
864 | ||
54738c09 DG |
865 | static long kvmppc_stt_npages(unsigned long window_size) |
866 | { | |
867 | return ALIGN((window_size >> SPAPR_TCE_SHIFT) | |
868 | * sizeof(u64), PAGE_SIZE) / PAGE_SIZE; | |
869 | } | |
870 | ||
871 | static void release_spapr_tce_table(struct kvmppc_spapr_tce_table *stt) | |
872 | { | |
873 | struct kvm *kvm = stt->kvm; | |
874 | int i; | |
875 | ||
876 | mutex_lock(&kvm->lock); | |
877 | list_del(&stt->list); | |
878 | for (i = 0; i < kvmppc_stt_npages(stt->window_size); i++) | |
879 | __free_page(stt->pages[i]); | |
880 | kfree(stt); | |
881 | mutex_unlock(&kvm->lock); | |
882 | ||
883 | kvm_put_kvm(kvm); | |
884 | } | |
885 | ||
886 | static int kvm_spapr_tce_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
887 | { | |
888 | struct kvmppc_spapr_tce_table *stt = vma->vm_file->private_data; | |
889 | struct page *page; | |
890 | ||
891 | if (vmf->pgoff >= kvmppc_stt_npages(stt->window_size)) | |
892 | return VM_FAULT_SIGBUS; | |
893 | ||
894 | page = stt->pages[vmf->pgoff]; | |
895 | get_page(page); | |
896 | vmf->page = page; | |
897 | return 0; | |
898 | } | |
899 | ||
900 | static const struct vm_operations_struct kvm_spapr_tce_vm_ops = { | |
901 | .fault = kvm_spapr_tce_fault, | |
902 | }; | |
903 | ||
904 | static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma) | |
905 | { | |
906 | vma->vm_ops = &kvm_spapr_tce_vm_ops; | |
907 | return 0; | |
908 | } | |
909 | ||
910 | static int kvm_spapr_tce_release(struct inode *inode, struct file *filp) | |
911 | { | |
912 | struct kvmppc_spapr_tce_table *stt = filp->private_data; | |
913 | ||
914 | release_spapr_tce_table(stt); | |
915 | return 0; | |
916 | } | |
917 | ||
918 | static struct file_operations kvm_spapr_tce_fops = { | |
919 | .mmap = kvm_spapr_tce_mmap, | |
920 | .release = kvm_spapr_tce_release, | |
921 | }; | |
922 | ||
923 | long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm, | |
924 | struct kvm_create_spapr_tce *args) | |
925 | { | |
926 | struct kvmppc_spapr_tce_table *stt = NULL; | |
927 | long npages; | |
928 | int ret = -ENOMEM; | |
929 | int i; | |
930 | ||
931 | /* Check this LIOBN hasn't been previously allocated */ | |
932 | list_for_each_entry(stt, &kvm->arch.spapr_tce_tables, list) { | |
933 | if (stt->liobn == args->liobn) | |
934 | return -EBUSY; | |
935 | } | |
936 | ||
937 | npages = kvmppc_stt_npages(args->window_size); | |
938 | ||
939 | stt = kzalloc(sizeof(*stt) + npages* sizeof(struct page *), | |
940 | GFP_KERNEL); | |
941 | if (!stt) | |
942 | goto fail; | |
943 | ||
944 | stt->liobn = args->liobn; | |
945 | stt->window_size = args->window_size; | |
946 | stt->kvm = kvm; | |
947 | ||
948 | for (i = 0; i < npages; i++) { | |
949 | stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
950 | if (!stt->pages[i]) | |
951 | goto fail; | |
952 | } | |
953 | ||
954 | kvm_get_kvm(kvm); | |
955 | ||
956 | mutex_lock(&kvm->lock); | |
957 | list_add(&stt->list, &kvm->arch.spapr_tce_tables); | |
958 | ||
959 | mutex_unlock(&kvm->lock); | |
960 | ||
961 | return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops, | |
962 | stt, O_RDWR); | |
963 | ||
964 | fail: | |
965 | if (stt) { | |
966 | for (i = 0; i < npages; i++) | |
967 | if (stt->pages[i]) | |
968 | __free_page(stt->pages[i]); | |
969 | ||
970 | kfree(stt); | |
971 | } | |
972 | return ret; | |
973 | } | |
974 | ||
aa04b4cc | 975 | /* Work out RMLS (real mode limit selector) field value for a given RMA size. |
9e368f29 | 976 | Assumes POWER7 or PPC970. */ |
aa04b4cc PM |
977 | static inline int lpcr_rmls(unsigned long rma_size) |
978 | { | |
979 | switch (rma_size) { | |
980 | case 32ul << 20: /* 32 MB */ | |
9e368f29 PM |
981 | if (cpu_has_feature(CPU_FTR_ARCH_206)) |
982 | return 8; /* only supported on POWER7 */ | |
983 | return -1; | |
aa04b4cc PM |
984 | case 64ul << 20: /* 64 MB */ |
985 | return 3; | |
986 | case 128ul << 20: /* 128 MB */ | |
987 | return 7; | |
988 | case 256ul << 20: /* 256 MB */ | |
989 | return 4; | |
990 | case 1ul << 30: /* 1 GB */ | |
991 | return 2; | |
992 | case 16ul << 30: /* 16 GB */ | |
993 | return 1; | |
994 | case 256ul << 30: /* 256 GB */ | |
995 | return 0; | |
996 | default: | |
997 | return -1; | |
998 | } | |
999 | } | |
1000 | ||
1001 | static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
1002 | { | |
1003 | struct kvmppc_rma_info *ri = vma->vm_file->private_data; | |
1004 | struct page *page; | |
1005 | ||
1006 | if (vmf->pgoff >= ri->npages) | |
1007 | return VM_FAULT_SIGBUS; | |
1008 | ||
1009 | page = pfn_to_page(ri->base_pfn + vmf->pgoff); | |
1010 | get_page(page); | |
1011 | vmf->page = page; | |
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | static const struct vm_operations_struct kvm_rma_vm_ops = { | |
1016 | .fault = kvm_rma_fault, | |
1017 | }; | |
1018 | ||
1019 | static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma) | |
1020 | { | |
1021 | vma->vm_flags |= VM_RESERVED; | |
1022 | vma->vm_ops = &kvm_rma_vm_ops; | |
1023 | return 0; | |
1024 | } | |
1025 | ||
1026 | static int kvm_rma_release(struct inode *inode, struct file *filp) | |
1027 | { | |
1028 | struct kvmppc_rma_info *ri = filp->private_data; | |
1029 | ||
1030 | kvm_release_rma(ri); | |
1031 | return 0; | |
1032 | } | |
1033 | ||
1034 | static struct file_operations kvm_rma_fops = { | |
1035 | .mmap = kvm_rma_mmap, | |
1036 | .release = kvm_rma_release, | |
1037 | }; | |
1038 | ||
1039 | long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret) | |
1040 | { | |
1041 | struct kvmppc_rma_info *ri; | |
1042 | long fd; | |
1043 | ||
1044 | ri = kvm_alloc_rma(); | |
1045 | if (!ri) | |
1046 | return -ENOMEM; | |
1047 | ||
1048 | fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR); | |
1049 | if (fd < 0) | |
1050 | kvm_release_rma(ri); | |
1051 | ||
1052 | ret->rma_size = ri->npages << PAGE_SHIFT; | |
1053 | return fd; | |
1054 | } | |
1055 | ||
1056 | static struct page *hva_to_page(unsigned long addr) | |
1057 | { | |
1058 | struct page *page[1]; | |
1059 | int npages; | |
1060 | ||
1061 | might_sleep(); | |
1062 | ||
1063 | npages = get_user_pages_fast(addr, 1, 1, page); | |
1064 | ||
1065 | if (unlikely(npages != 1)) | |
1066 | return 0; | |
1067 | ||
1068 | return page[0]; | |
1069 | } | |
1070 | ||
de56a948 PM |
1071 | int kvmppc_core_prepare_memory_region(struct kvm *kvm, |
1072 | struct kvm_userspace_memory_region *mem) | |
1073 | { | |
aa04b4cc PM |
1074 | unsigned long psize, porder; |
1075 | unsigned long i, npages, totalpages; | |
1076 | unsigned long pg_ix; | |
1077 | struct kvmppc_pginfo *pginfo; | |
1078 | unsigned long hva; | |
1079 | struct kvmppc_rma_info *ri = NULL; | |
1080 | struct page *page; | |
1081 | ||
1082 | /* For now, only allow 16MB pages */ | |
1083 | porder = LARGE_PAGE_ORDER; | |
1084 | psize = 1ul << porder; | |
1085 | if ((mem->memory_size & (psize - 1)) || | |
1086 | (mem->guest_phys_addr & (psize - 1))) { | |
1087 | pr_err("bad memory_size=%llx @ %llx\n", | |
1088 | mem->memory_size, mem->guest_phys_addr); | |
1089 | return -EINVAL; | |
1090 | } | |
1091 | ||
1092 | npages = mem->memory_size >> porder; | |
1093 | totalpages = (mem->guest_phys_addr + mem->memory_size) >> porder; | |
1094 | ||
1095 | /* More memory than we have space to track? */ | |
1096 | if (totalpages > (1ul << (MAX_MEM_ORDER - LARGE_PAGE_ORDER))) | |
1097 | return -EINVAL; | |
1098 | ||
1099 | /* Do we already have an RMA registered? */ | |
1100 | if (mem->guest_phys_addr == 0 && kvm->arch.rma) | |
1101 | return -EINVAL; | |
1102 | ||
1103 | if (totalpages > kvm->arch.ram_npages) | |
1104 | kvm->arch.ram_npages = totalpages; | |
1105 | ||
1106 | /* Is this one of our preallocated RMAs? */ | |
1107 | if (mem->guest_phys_addr == 0) { | |
1108 | struct vm_area_struct *vma; | |
1109 | ||
1110 | down_read(¤t->mm->mmap_sem); | |
1111 | vma = find_vma(current->mm, mem->userspace_addr); | |
1112 | if (vma && vma->vm_file && | |
1113 | vma->vm_file->f_op == &kvm_rma_fops && | |
1114 | mem->userspace_addr == vma->vm_start) | |
1115 | ri = vma->vm_file->private_data; | |
1116 | up_read(¤t->mm->mmap_sem); | |
9e368f29 PM |
1117 | if (!ri && cpu_has_feature(CPU_FTR_ARCH_201)) { |
1118 | pr_err("CPU requires an RMO\n"); | |
1119 | return -EINVAL; | |
1120 | } | |
aa04b4cc PM |
1121 | } |
1122 | ||
1123 | if (ri) { | |
1124 | unsigned long rma_size; | |
1125 | unsigned long lpcr; | |
1126 | long rmls; | |
1127 | ||
1128 | rma_size = ri->npages << PAGE_SHIFT; | |
1129 | if (rma_size > mem->memory_size) | |
1130 | rma_size = mem->memory_size; | |
1131 | rmls = lpcr_rmls(rma_size); | |
1132 | if (rmls < 0) { | |
1133 | pr_err("Can't use RMA of 0x%lx bytes\n", rma_size); | |
1134 | return -EINVAL; | |
1135 | } | |
1136 | atomic_inc(&ri->use_count); | |
1137 | kvm->arch.rma = ri; | |
1138 | kvm->arch.n_rma_pages = rma_size >> porder; | |
9e368f29 PM |
1139 | |
1140 | /* Update LPCR and RMOR */ | |
1141 | lpcr = kvm->arch.lpcr; | |
1142 | if (cpu_has_feature(CPU_FTR_ARCH_201)) { | |
1143 | /* PPC970; insert RMLS value (split field) in HID4 */ | |
1144 | lpcr &= ~((1ul << HID4_RMLS0_SH) | | |
1145 | (3ul << HID4_RMLS2_SH)); | |
1146 | lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) | | |
1147 | ((rmls & 3) << HID4_RMLS2_SH); | |
1148 | /* RMOR is also in HID4 */ | |
1149 | lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff) | |
1150 | << HID4_RMOR_SH; | |
1151 | } else { | |
1152 | /* POWER7 */ | |
1153 | lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L); | |
1154 | lpcr |= rmls << LPCR_RMLS_SH; | |
1155 | kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT; | |
1156 | } | |
aa04b4cc | 1157 | kvm->arch.lpcr = lpcr; |
aa04b4cc PM |
1158 | pr_info("Using RMO at %lx size %lx (LPCR = %lx)\n", |
1159 | ri->base_pfn << PAGE_SHIFT, rma_size, lpcr); | |
1160 | } | |
1161 | ||
1162 | pg_ix = mem->guest_phys_addr >> porder; | |
1163 | pginfo = kvm->arch.ram_pginfo + pg_ix; | |
1164 | for (i = 0; i < npages; ++i, ++pg_ix) { | |
1165 | if (ri && pg_ix < kvm->arch.n_rma_pages) { | |
1166 | pginfo[i].pfn = ri->base_pfn + | |
1167 | (pg_ix << (porder - PAGE_SHIFT)); | |
1168 | continue; | |
1169 | } | |
1170 | hva = mem->userspace_addr + (i << porder); | |
1171 | page = hva_to_page(hva); | |
1172 | if (!page) { | |
1173 | pr_err("oops, no pfn for hva %lx\n", hva); | |
1174 | goto err; | |
1175 | } | |
1176 | /* Check it's a 16MB page */ | |
1177 | if (!PageHead(page) || | |
1178 | compound_order(page) != (LARGE_PAGE_ORDER - PAGE_SHIFT)) { | |
1179 | pr_err("page at %lx isn't 16MB (o=%d)\n", | |
1180 | hva, compound_order(page)); | |
1181 | goto err; | |
1182 | } | |
1183 | pginfo[i].pfn = page_to_pfn(page); | |
1184 | } | |
1185 | ||
de56a948 | 1186 | return 0; |
aa04b4cc PM |
1187 | |
1188 | err: | |
1189 | return -EINVAL; | |
de56a948 PM |
1190 | } |
1191 | ||
1192 | void kvmppc_core_commit_memory_region(struct kvm *kvm, | |
1193 | struct kvm_userspace_memory_region *mem) | |
1194 | { | |
aa04b4cc PM |
1195 | if (mem->guest_phys_addr == 0 && mem->memory_size != 0 && |
1196 | !kvm->arch.rma) | |
de56a948 PM |
1197 | kvmppc_map_vrma(kvm, mem); |
1198 | } | |
1199 | ||
1200 | int kvmppc_core_init_vm(struct kvm *kvm) | |
1201 | { | |
1202 | long r; | |
aa04b4cc PM |
1203 | unsigned long npages = 1ul << (MAX_MEM_ORDER - LARGE_PAGE_ORDER); |
1204 | long err = -ENOMEM; | |
1205 | unsigned long lpcr; | |
de56a948 PM |
1206 | |
1207 | /* Allocate hashed page table */ | |
1208 | r = kvmppc_alloc_hpt(kvm); | |
54738c09 DG |
1209 | if (r) |
1210 | return r; | |
de56a948 | 1211 | |
54738c09 | 1212 | INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables); |
aa04b4cc PM |
1213 | |
1214 | kvm->arch.ram_pginfo = kzalloc(npages * sizeof(struct kvmppc_pginfo), | |
1215 | GFP_KERNEL); | |
1216 | if (!kvm->arch.ram_pginfo) { | |
1217 | pr_err("kvmppc_core_init_vm: couldn't alloc %lu bytes\n", | |
1218 | npages * sizeof(struct kvmppc_pginfo)); | |
1219 | goto out_free; | |
1220 | } | |
1221 | ||
1222 | kvm->arch.ram_npages = 0; | |
1223 | kvm->arch.ram_psize = 1ul << LARGE_PAGE_ORDER; | |
1224 | kvm->arch.ram_porder = LARGE_PAGE_ORDER; | |
1225 | kvm->arch.rma = NULL; | |
1226 | kvm->arch.n_rma_pages = 0; | |
1227 | ||
9e368f29 | 1228 | kvm->arch.host_sdr1 = mfspr(SPRN_SDR1); |
aa04b4cc | 1229 | |
9e368f29 PM |
1230 | if (cpu_has_feature(CPU_FTR_ARCH_201)) { |
1231 | /* PPC970; HID4 is effectively the LPCR */ | |
1232 | unsigned long lpid = kvm->arch.lpid; | |
1233 | kvm->arch.host_lpid = 0; | |
1234 | kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4); | |
1235 | lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH)); | |
1236 | lpcr |= ((lpid >> 4) << HID4_LPID1_SH) | | |
1237 | ((lpid & 0xf) << HID4_LPID5_SH); | |
1238 | } else { | |
1239 | /* POWER7; init LPCR for virtual RMA mode */ | |
1240 | kvm->arch.host_lpid = mfspr(SPRN_LPID); | |
1241 | kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR); | |
1242 | lpcr &= LPCR_PECE | LPCR_LPES; | |
1243 | lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE | | |
1244 | LPCR_VPM0 | LPCR_VRMA_L; | |
1245 | } | |
1246 | kvm->arch.lpcr = lpcr; | |
aa04b4cc | 1247 | |
54738c09 | 1248 | return 0; |
aa04b4cc PM |
1249 | |
1250 | out_free: | |
1251 | kvmppc_free_hpt(kvm); | |
1252 | return err; | |
de56a948 PM |
1253 | } |
1254 | ||
1255 | void kvmppc_core_destroy_vm(struct kvm *kvm) | |
1256 | { | |
aa04b4cc PM |
1257 | struct kvmppc_pginfo *pginfo; |
1258 | unsigned long i; | |
1259 | ||
1260 | if (kvm->arch.ram_pginfo) { | |
1261 | pginfo = kvm->arch.ram_pginfo; | |
1262 | kvm->arch.ram_pginfo = NULL; | |
1263 | for (i = kvm->arch.n_rma_pages; i < kvm->arch.ram_npages; ++i) | |
1264 | if (pginfo[i].pfn) | |
1265 | put_page(pfn_to_page(pginfo[i].pfn)); | |
1266 | kfree(pginfo); | |
1267 | } | |
1268 | if (kvm->arch.rma) { | |
1269 | kvm_release_rma(kvm->arch.rma); | |
1270 | kvm->arch.rma = NULL; | |
1271 | } | |
1272 | ||
de56a948 | 1273 | kvmppc_free_hpt(kvm); |
54738c09 | 1274 | WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables)); |
de56a948 PM |
1275 | } |
1276 | ||
1277 | /* These are stubs for now */ | |
1278 | void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end) | |
1279 | { | |
1280 | } | |
1281 | ||
1282 | /* We don't need to emulate any privileged instructions or dcbz */ | |
1283 | int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu, | |
1284 | unsigned int inst, int *advance) | |
1285 | { | |
1286 | return EMULATE_FAIL; | |
1287 | } | |
1288 | ||
1289 | int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, int rs) | |
1290 | { | |
1291 | return EMULATE_FAIL; | |
1292 | } | |
1293 | ||
1294 | int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, int rt) | |
1295 | { | |
1296 | return EMULATE_FAIL; | |
1297 | } | |
1298 | ||
1299 | static int kvmppc_book3s_hv_init(void) | |
1300 | { | |
1301 | int r; | |
1302 | ||
1303 | r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1304 | ||
1305 | if (r) | |
1306 | return r; | |
1307 | ||
1308 | r = kvmppc_mmu_hv_init(); | |
1309 | ||
1310 | return r; | |
1311 | } | |
1312 | ||
1313 | static void kvmppc_book3s_hv_exit(void) | |
1314 | { | |
1315 | kvm_exit(); | |
1316 | } | |
1317 | ||
1318 | module_init(kvmppc_book3s_hv_init); | |
1319 | module_exit(kvmppc_book3s_hv_exit); |