projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
powerpc/kvm/book3s: Add support for H_IPOLL and H_XIRR_X in XICS emulation
[deliverable/linux.git] / arch / powerpc / kvm / book3s_hv.c
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>
27 #include <linux/export.h>
28 #include <linux/fs.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34
35 #include <asm/reg.h>
36 #include <asm/cputable.h>
37 #include <asm/cacheflush.h>
38 #include <asm/tlbflush.h>
39 #include <asm/uaccess.h>
40 #include <asm/io.h>
41 #include <asm/kvm_ppc.h>
42 #include <asm/kvm_book3s.h>
43 #include <asm/mmu_context.h>
44 #include <asm/lppaca.h>
45 #include <asm/processor.h>
46 #include <asm/cputhreads.h>
47 #include <asm/page.h>
48 #include <asm/hvcall.h>
49 #include <asm/switch_to.h>
50 #include <asm/smp.h>
51 #include <linux/gfp.h>
52 #include <linux/vmalloc.h>
53 #include <linux/highmem.h>
54 #include <linux/hugetlb.h>
55
56 /* #define EXIT_DEBUG */
57 /* #define EXIT_DEBUG_SIMPLE */
58 /* #define EXIT_DEBUG_INT */
59
60 /* Used to indicate that a guest page fault needs to be handled */
61 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
62
63 /* Used as a "null" value for timebase values */
64 #define TB_NIL (~(u64)0)
65
66 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
67 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
68
69 void kvmppc_fast_vcpu_kick(struct kvm_vcpu *vcpu)
70 {
71 int me;
72 int cpu = vcpu->cpu;
73 wait_queue_head_t *wqp;
74
75 wqp = kvm_arch_vcpu_wq(vcpu);
76 if (waitqueue_active(wqp)) {
77 wake_up_interruptible(wqp);
78 ++vcpu->stat.halt_wakeup;
79 }
80
81 me = get_cpu();
82
83 /* CPU points to the first thread of the core */
84 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
85 int real_cpu = cpu + vcpu->arch.ptid;
86 if (paca[real_cpu].kvm_hstate.xics_phys)
87 xics_wake_cpu(real_cpu);
88 else if (cpu_online(cpu))
89 smp_send_reschedule(cpu);
90 }
91 put_cpu();
92 }
93
94 /*
95 * We use the vcpu_load/put functions to measure stolen time.
96 * Stolen time is counted as time when either the vcpu is able to
97 * run as part of a virtual core, but the task running the vcore
98 * is preempted or sleeping, or when the vcpu needs something done
99 * in the kernel by the task running the vcpu, but that task is
100 * preempted or sleeping. Those two things have to be counted
101 * separately, since one of the vcpu tasks will take on the job
102 * of running the core, and the other vcpu tasks in the vcore will
103 * sleep waiting for it to do that, but that sleep shouldn't count
104 * as stolen time.
105 *
106 * Hence we accumulate stolen time when the vcpu can run as part of
107 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
108 * needs its task to do other things in the kernel (for example,
109 * service a page fault) in busy_stolen. We don't accumulate
110 * stolen time for a vcore when it is inactive, or for a vcpu
111 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
112 * a misnomer; it means that the vcpu task is not executing in
113 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
114 * the kernel. We don't have any way of dividing up that time
115 * between time that the vcpu is genuinely stopped, time that
116 * the task is actively working on behalf of the vcpu, and time
117 * that the task is preempted, so we don't count any of it as
118 * stolen.
119 *
120 * Updates to busy_stolen are protected by arch.tbacct_lock;
121 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
122 * of the vcpu that has taken responsibility for running the vcore
123 * (i.e. vc->runner). The stolen times are measured in units of
124 * timebase ticks. (Note that the != TB_NIL checks below are
125 * purely defensive; they should never fail.)
126 */
127
128 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
129 {
130 struct kvmppc_vcore *vc = vcpu->arch.vcore;
131
132 spin_lock(&vcpu->arch.tbacct_lock);
133 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
134 vc->preempt_tb != TB_NIL) {
135 vc->stolen_tb += mftb() - vc->preempt_tb;
136 vc->preempt_tb = TB_NIL;
137 }
138 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
139 vcpu->arch.busy_preempt != TB_NIL) {
140 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
141 vcpu->arch.busy_preempt = TB_NIL;
142 }
143 spin_unlock(&vcpu->arch.tbacct_lock);
144 }
145
146 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
147 {
148 struct kvmppc_vcore *vc = vcpu->arch.vcore;
149
150 spin_lock(&vcpu->arch.tbacct_lock);
151 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
152 vc->preempt_tb = mftb();
153 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
154 vcpu->arch.busy_preempt = mftb();
155 spin_unlock(&vcpu->arch.tbacct_lock);
156 }
157
158 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
159 {
160 vcpu->arch.shregs.msr = msr;
161 kvmppc_end_cede(vcpu);
162 }
163
164 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
165 {
166 vcpu->arch.pvr = pvr;
167 }
168
169 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
170 {
171 int r;
172
173 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
174 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
175 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
176 for (r = 0; r < 16; ++r)
177 pr_err("r%2d = %.16lx r%d = %.16lx\n",
178 r, kvmppc_get_gpr(vcpu, r),
179 r+16, kvmppc_get_gpr(vcpu, r+16));
180 pr_err("ctr = %.16lx lr = %.16lx\n",
181 vcpu->arch.ctr, vcpu->arch.lr);
182 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
183 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
184 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
185 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
186 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
187 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
188 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
189 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
190 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
191 pr_err("fault dar = %.16lx dsisr = %.8x\n",
192 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
193 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
194 for (r = 0; r < vcpu->arch.slb_max; ++r)
195 pr_err(" ESID = %.16llx VSID = %.16llx\n",
196 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
197 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
198 vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
199 vcpu->arch.last_inst);
200 }
201
202 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
203 {
204 int r;
205 struct kvm_vcpu *v, *ret = NULL;
206
207 mutex_lock(&kvm->lock);
208 kvm_for_each_vcpu(r, v, kvm) {
209 if (v->vcpu_id == id) {
210 ret = v;
211 break;
212 }
213 }
214 mutex_unlock(&kvm->lock);
215 return ret;
216 }
217
218 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
219 {
220 vpa->shared_proc = 1;
221 vpa->yield_count = 1;
222 }
223
224 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
225 unsigned long addr, unsigned long len)
226 {
227 /* check address is cacheline aligned */
228 if (addr & (L1_CACHE_BYTES - 1))
229 return -EINVAL;
230 spin_lock(&vcpu->arch.vpa_update_lock);
231 if (v->next_gpa != addr || v->len != len) {
232 v->next_gpa = addr;
233 v->len = addr ? len : 0;
234 v->update_pending = 1;
235 }
236 spin_unlock(&vcpu->arch.vpa_update_lock);
237 return 0;
238 }
239
240 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
241 struct reg_vpa {
242 u32 dummy;
243 union {
244 u16 hword;
245 u32 word;
246 } length;
247 };
248
249 static int vpa_is_registered(struct kvmppc_vpa *vpap)
250 {
251 if (vpap->update_pending)
252 return vpap->next_gpa != 0;
253 return vpap->pinned_addr != NULL;
254 }
255
256 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
257 unsigned long flags,
258 unsigned long vcpuid, unsigned long vpa)
259 {
260 struct kvm *kvm = vcpu->kvm;
261 unsigned long len, nb;
262 void *va;
263 struct kvm_vcpu *tvcpu;
264 int err;
265 int subfunc;
266 struct kvmppc_vpa *vpap;
267
268 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
269 if (!tvcpu)
270 return H_PARAMETER;
271
272 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
273 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
274 subfunc == H_VPA_REG_SLB) {
275 /* Registering new area - address must be cache-line aligned */
276 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
277 return H_PARAMETER;
278
279 /* convert logical addr to kernel addr and read length */
280 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
281 if (va == NULL)
282 return H_PARAMETER;
283 if (subfunc == H_VPA_REG_VPA)
284 len = ((struct reg_vpa *)va)->length.hword;
285 else
286 len = ((struct reg_vpa *)va)->length.word;
287 kvmppc_unpin_guest_page(kvm, va, vpa, false);
288
289 /* Check length */
290 if (len > nb || len < sizeof(struct reg_vpa))
291 return H_PARAMETER;
292 } else {
293 vpa = 0;
294 len = 0;
295 }
296
297 err = H_PARAMETER;
298 vpap = NULL;
299 spin_lock(&tvcpu->arch.vpa_update_lock);
300
301 switch (subfunc) {
302 case H_VPA_REG_VPA: /* register VPA */
303 if (len < sizeof(struct lppaca))
304 break;
305 vpap = &tvcpu->arch.vpa;
306 err = 0;
307 break;
308
309 case H_VPA_REG_DTL: /* register DTL */
310 if (len < sizeof(struct dtl_entry))
311 break;
312 len -= len % sizeof(struct dtl_entry);
313
314 /* Check that they have previously registered a VPA */
315 err = H_RESOURCE;
316 if (!vpa_is_registered(&tvcpu->arch.vpa))
317 break;
318
319 vpap = &tvcpu->arch.dtl;
320 err = 0;
321 break;
322
323 case H_VPA_REG_SLB: /* register SLB shadow buffer */
324 /* Check that they have previously registered a VPA */
325 err = H_RESOURCE;
326 if (!vpa_is_registered(&tvcpu->arch.vpa))
327 break;
328
329 vpap = &tvcpu->arch.slb_shadow;
330 err = 0;
331 break;
332
333 case H_VPA_DEREG_VPA: /* deregister VPA */
334 /* Check they don't still have a DTL or SLB buf registered */
335 err = H_RESOURCE;
336 if (vpa_is_registered(&tvcpu->arch.dtl) ||
337 vpa_is_registered(&tvcpu->arch.slb_shadow))
338 break;
339
340 vpap = &tvcpu->arch.vpa;
341 err = 0;
342 break;
343
344 case H_VPA_DEREG_DTL: /* deregister DTL */
345 vpap = &tvcpu->arch.dtl;
346 err = 0;
347 break;
348
349 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
350 vpap = &tvcpu->arch.slb_shadow;
351 err = 0;
352 break;
353 }
354
355 if (vpap) {
356 vpap->next_gpa = vpa;
357 vpap->len = len;
358 vpap->update_pending = 1;
359 }
360
361 spin_unlock(&tvcpu->arch.vpa_update_lock);
362
363 return err;
364 }
365
366 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
367 {
368 struct kvm *kvm = vcpu->kvm;
369 void *va;
370 unsigned long nb;
371 unsigned long gpa;
372
373 /*
374 * We need to pin the page pointed to by vpap->next_gpa,
375 * but we can't call kvmppc_pin_guest_page under the lock
376 * as it does get_user_pages() and down_read(). So we
377 * have to drop the lock, pin the page, then get the lock
378 * again and check that a new area didn't get registered
379 * in the meantime.
380 */
381 for (;;) {
382 gpa = vpap->next_gpa;
383 spin_unlock(&vcpu->arch.vpa_update_lock);
384 va = NULL;
385 nb = 0;
386 if (gpa)
387 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
388 spin_lock(&vcpu->arch.vpa_update_lock);
389 if (gpa == vpap->next_gpa)
390 break;
391 /* sigh... unpin that one and try again */
392 if (va)
393 kvmppc_unpin_guest_page(kvm, va, gpa, false);
394 }
395
396 vpap->update_pending = 0;
397 if (va && nb < vpap->len) {
398 /*
399 * If it's now too short, it must be that userspace
400 * has changed the mappings underlying guest memory,
401 * so unregister the region.
402 */
403 kvmppc_unpin_guest_page(kvm, va, gpa, false);
404 va = NULL;
405 }
406 if (vpap->pinned_addr)
407 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
408 vpap->dirty);
409 vpap->gpa = gpa;
410 vpap->pinned_addr = va;
411 vpap->dirty = false;
412 if (va)
413 vpap->pinned_end = va + vpap->len;
414 }
415
416 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
417 {
418 if (!(vcpu->arch.vpa.update_pending ||
419 vcpu->arch.slb_shadow.update_pending ||
420 vcpu->arch.dtl.update_pending))
421 return;
422
423 spin_lock(&vcpu->arch.vpa_update_lock);
424 if (vcpu->arch.vpa.update_pending) {
425 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
426 if (vcpu->arch.vpa.pinned_addr)
427 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
428 }
429 if (vcpu->arch.dtl.update_pending) {
430 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
431 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
432 vcpu->arch.dtl_index = 0;
433 }
434 if (vcpu->arch.slb_shadow.update_pending)
435 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
436 spin_unlock(&vcpu->arch.vpa_update_lock);
437 }
438
439 /*
440 * Return the accumulated stolen time for the vcore up until `now'.
441 * The caller should hold the vcore lock.
442 */
443 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
444 {
445 u64 p;
446
447 /*
448 * If we are the task running the vcore, then since we hold
449 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
450 * can't be updated, so we don't need the tbacct_lock.
451 * If the vcore is inactive, it can't become active (since we
452 * hold the vcore lock), so the vcpu load/put functions won't
453 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
454 */
455 if (vc->vcore_state != VCORE_INACTIVE &&
456 vc->runner->arch.run_task != current) {
457 spin_lock(&vc->runner->arch.tbacct_lock);
458 p = vc->stolen_tb;
459 if (vc->preempt_tb != TB_NIL)
460 p += now - vc->preempt_tb;
461 spin_unlock(&vc->runner->arch.tbacct_lock);
462 } else {
463 p = vc->stolen_tb;
464 }
465 return p;
466 }
467
468 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
469 struct kvmppc_vcore *vc)
470 {
471 struct dtl_entry *dt;
472 struct lppaca *vpa;
473 unsigned long stolen;
474 unsigned long core_stolen;
475 u64 now;
476
477 dt = vcpu->arch.dtl_ptr;
478 vpa = vcpu->arch.vpa.pinned_addr;
479 now = mftb();
480 core_stolen = vcore_stolen_time(vc, now);
481 stolen = core_stolen - vcpu->arch.stolen_logged;
482 vcpu->arch.stolen_logged = core_stolen;
483 spin_lock(&vcpu->arch.tbacct_lock);
484 stolen += vcpu->arch.busy_stolen;
485 vcpu->arch.busy_stolen = 0;
486 spin_unlock(&vcpu->arch.tbacct_lock);
487 if (!dt || !vpa)
488 return;
489 memset(dt, 0, sizeof(struct dtl_entry));
490 dt->dispatch_reason = 7;
491 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
492 dt->timebase = now;
493 dt->enqueue_to_dispatch_time = stolen;
494 dt->srr0 = kvmppc_get_pc(vcpu);
495 dt->srr1 = vcpu->arch.shregs.msr;
496 ++dt;
497 if (dt == vcpu->arch.dtl.pinned_end)
498 dt = vcpu->arch.dtl.pinned_addr;
499 vcpu->arch.dtl_ptr = dt;
500 /* order writing *dt vs. writing vpa->dtl_idx */
501 smp_wmb();
502 vpa->dtl_idx = ++vcpu->arch.dtl_index;
503 vcpu->arch.dtl.dirty = true;
504 }
505
506 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
507 {
508 unsigned long req = kvmppc_get_gpr(vcpu, 3);
509 unsigned long target, ret = H_SUCCESS;
510 struct kvm_vcpu *tvcpu;
511 int idx, rc;
512
513 switch (req) {
514 case H_ENTER:
515 idx = srcu_read_lock(&vcpu->kvm->srcu);
516 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
517 kvmppc_get_gpr(vcpu, 5),
518 kvmppc_get_gpr(vcpu, 6),
519 kvmppc_get_gpr(vcpu, 7));
520 srcu_read_unlock(&vcpu->kvm->srcu, idx);
521 break;
522 case H_CEDE:
523 break;
524 case H_PROD:
525 target = kvmppc_get_gpr(vcpu, 4);
526 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
527 if (!tvcpu) {
528 ret = H_PARAMETER;
529 break;
530 }
531 tvcpu->arch.prodded = 1;
532 smp_mb();
533 if (vcpu->arch.ceded) {
534 if (waitqueue_active(&vcpu->wq)) {
535 wake_up_interruptible(&vcpu->wq);
536 vcpu->stat.halt_wakeup++;
537 }
538 }
539 break;
540 case H_CONFER:
541 break;
542 case H_REGISTER_VPA:
543 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
544 kvmppc_get_gpr(vcpu, 5),
545 kvmppc_get_gpr(vcpu, 6));
546 break;
547 case H_RTAS:
548 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
549 return RESUME_HOST;
550
551 rc = kvmppc_rtas_hcall(vcpu);
552
553 if (rc == -ENOENT)
554 return RESUME_HOST;
555 else if (rc == 0)
556 break;
557
558 /* Send the error out to userspace via KVM_RUN */
559 return rc;
560
561 case H_XIRR:
562 case H_CPPR:
563 case H_EOI:
564 case H_IPI:
565 case H_IPOLL:
566 case H_XIRR_X:
567 if (kvmppc_xics_enabled(vcpu)) {
568 ret = kvmppc_xics_hcall(vcpu, req);
569 break;
570 } /* fallthrough */
571 default:
572 return RESUME_HOST;
573 }
574 kvmppc_set_gpr(vcpu, 3, ret);
575 vcpu->arch.hcall_needed = 0;
576 return RESUME_GUEST;
577 }
578
579 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
580 struct task_struct *tsk)
581 {
582 int r = RESUME_HOST;
583
584 vcpu->stat.sum_exits++;
585
586 run->exit_reason = KVM_EXIT_UNKNOWN;
587 run->ready_for_interrupt_injection = 1;
588 switch (vcpu->arch.trap) {
589 /* We're good on these - the host merely wanted to get our attention */
590 case BOOK3S_INTERRUPT_HV_DECREMENTER:
591 vcpu->stat.dec_exits++;
592 r = RESUME_GUEST;
593 break;
594 case BOOK3S_INTERRUPT_EXTERNAL:
595 vcpu->stat.ext_intr_exits++;
596 r = RESUME_GUEST;
597 break;
598 case BOOK3S_INTERRUPT_PERFMON:
599 r = RESUME_GUEST;
600 break;
601 case BOOK3S_INTERRUPT_MACHINE_CHECK:
602 /*
603 * Deliver a machine check interrupt to the guest.
604 * We have to do this, even if the host has handled the
605 * machine check, because machine checks use SRR0/1 and
606 * the interrupt might have trashed guest state in them.
607 */
608 kvmppc_book3s_queue_irqprio(vcpu,
609 BOOK3S_INTERRUPT_MACHINE_CHECK);
610 r = RESUME_GUEST;
611 break;
612 case BOOK3S_INTERRUPT_PROGRAM:
613 {
614 ulong flags;
615 /*
616 * Normally program interrupts are delivered directly
617 * to the guest by the hardware, but we can get here
618 * as a result of a hypervisor emulation interrupt
619 * (e40) getting turned into a 700 by BML RTAS.
620 */
621 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
622 kvmppc_core_queue_program(vcpu, flags);
623 r = RESUME_GUEST;
624 break;
625 }
626 case BOOK3S_INTERRUPT_SYSCALL:
627 {
628 /* hcall - punt to userspace */
629 int i;
630
631 if (vcpu->arch.shregs.msr & MSR_PR) {
632 /* sc 1 from userspace - reflect to guest syscall */
633 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
634 r = RESUME_GUEST;
635 break;
636 }
637 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
638 for (i = 0; i < 9; ++i)
639 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
640 run->exit_reason = KVM_EXIT_PAPR_HCALL;
641 vcpu->arch.hcall_needed = 1;
642 r = RESUME_HOST;
643 break;
644 }
645 /*
646 * We get these next two if the guest accesses a page which it thinks
647 * it has mapped but which is not actually present, either because
648 * it is for an emulated I/O device or because the corresonding
649 * host page has been paged out. Any other HDSI/HISI interrupts
650 * have been handled already.
651 */
652 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
653 r = RESUME_PAGE_FAULT;
654 break;
655 case BOOK3S_INTERRUPT_H_INST_STORAGE:
656 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
657 vcpu->arch.fault_dsisr = 0;
658 r = RESUME_PAGE_FAULT;
659 break;
660 /*
661 * This occurs if the guest executes an illegal instruction.
662 * We just generate a program interrupt to the guest, since
663 * we don't emulate any guest instructions at this stage.
664 */
665 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
666 kvmppc_core_queue_program(vcpu, 0x80000);
667 r = RESUME_GUEST;
668 break;
669 default:
670 kvmppc_dump_regs(vcpu);
671 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
672 vcpu->arch.trap, kvmppc_get_pc(vcpu),
673 vcpu->arch.shregs.msr);
674 r = RESUME_HOST;
675 BUG();
676 break;
677 }
678
679 return r;
680 }
681
682 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
683 struct kvm_sregs *sregs)
684 {
685 int i;
686
687 sregs->pvr = vcpu->arch.pvr;
688
689 memset(sregs, 0, sizeof(struct kvm_sregs));
690 for (i = 0; i < vcpu->arch.slb_max; i++) {
691 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
692 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
693 }
694
695 return 0;
696 }
697
698 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
699 struct kvm_sregs *sregs)
700 {
701 int i, j;
702
703 kvmppc_set_pvr(vcpu, sregs->pvr);
704
705 j = 0;
706 for (i = 0; i < vcpu->arch.slb_nr; i++) {
707 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
708 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
709 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
710 ++j;
711 }
712 }
713 vcpu->arch.slb_max = j;
714
715 return 0;
716 }
717
718 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
719 {
720 int r = 0;
721 long int i;
722
723 switch (id) {
724 case KVM_REG_PPC_HIOR:
725 *val = get_reg_val(id, 0);
726 break;
727 case KVM_REG_PPC_DABR:
728 *val = get_reg_val(id, vcpu->arch.dabr);
729 break;
730 case KVM_REG_PPC_DSCR:
731 *val = get_reg_val(id, vcpu->arch.dscr);
732 break;
733 case KVM_REG_PPC_PURR:
734 *val = get_reg_val(id, vcpu->arch.purr);
735 break;
736 case KVM_REG_PPC_SPURR:
737 *val = get_reg_val(id, vcpu->arch.spurr);
738 break;
739 case KVM_REG_PPC_AMR:
740 *val = get_reg_val(id, vcpu->arch.amr);
741 break;
742 case KVM_REG_PPC_UAMOR:
743 *val = get_reg_val(id, vcpu->arch.uamor);
744 break;
745 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
746 i = id - KVM_REG_PPC_MMCR0;
747 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
748 break;
749 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
750 i = id - KVM_REG_PPC_PMC1;
751 *val = get_reg_val(id, vcpu->arch.pmc[i]);
752 break;
753 #ifdef CONFIG_VSX
754 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
755 if (cpu_has_feature(CPU_FTR_VSX)) {
756 /* VSX => FP reg i is stored in arch.vsr[2*i] */
757 long int i = id - KVM_REG_PPC_FPR0;
758 *val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
759 } else {
760 /* let generic code handle it */
761 r = -EINVAL;
762 }
763 break;
764 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
765 if (cpu_has_feature(CPU_FTR_VSX)) {
766 long int i = id - KVM_REG_PPC_VSR0;
767 val->vsxval[0] = vcpu->arch.vsr[2 * i];
768 val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
769 } else {
770 r = -ENXIO;
771 }
772 break;
773 #endif /* CONFIG_VSX */
774 case KVM_REG_PPC_VPA_ADDR:
775 spin_lock(&vcpu->arch.vpa_update_lock);
776 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
777 spin_unlock(&vcpu->arch.vpa_update_lock);
778 break;
779 case KVM_REG_PPC_VPA_SLB:
780 spin_lock(&vcpu->arch.vpa_update_lock);
781 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
782 val->vpaval.length = vcpu->arch.slb_shadow.len;
783 spin_unlock(&vcpu->arch.vpa_update_lock);
784 break;
785 case KVM_REG_PPC_VPA_DTL:
786 spin_lock(&vcpu->arch.vpa_update_lock);
787 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
788 val->vpaval.length = vcpu->arch.dtl.len;
789 spin_unlock(&vcpu->arch.vpa_update_lock);
790 break;
791 default:
792 r = -EINVAL;
793 break;
794 }
795
796 return r;
797 }
798
799 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
800 {
801 int r = 0;
802 long int i;
803 unsigned long addr, len;
804
805 switch (id) {
806 case KVM_REG_PPC_HIOR:
807 /* Only allow this to be set to zero */
808 if (set_reg_val(id, *val))
809 r = -EINVAL;
810 break;
811 case KVM_REG_PPC_DABR:
812 vcpu->arch.dabr = set_reg_val(id, *val);
813 break;
814 case KVM_REG_PPC_DSCR:
815 vcpu->arch.dscr = set_reg_val(id, *val);
816 break;
817 case KVM_REG_PPC_PURR:
818 vcpu->arch.purr = set_reg_val(id, *val);
819 break;
820 case KVM_REG_PPC_SPURR:
821 vcpu->arch.spurr = set_reg_val(id, *val);
822 break;
823 case KVM_REG_PPC_AMR:
824 vcpu->arch.amr = set_reg_val(id, *val);
825 break;
826 case KVM_REG_PPC_UAMOR:
827 vcpu->arch.uamor = set_reg_val(id, *val);
828 break;
829 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
830 i = id - KVM_REG_PPC_MMCR0;
831 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
832 break;
833 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
834 i = id - KVM_REG_PPC_PMC1;
835 vcpu->arch.pmc[i] = set_reg_val(id, *val);
836 break;
837 #ifdef CONFIG_VSX
838 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
839 if (cpu_has_feature(CPU_FTR_VSX)) {
840 /* VSX => FP reg i is stored in arch.vsr[2*i] */
841 long int i = id - KVM_REG_PPC_FPR0;
842 vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
843 } else {
844 /* let generic code handle it */
845 r = -EINVAL;
846 }
847 break;
848 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
849 if (cpu_has_feature(CPU_FTR_VSX)) {
850 long int i = id - KVM_REG_PPC_VSR0;
851 vcpu->arch.vsr[2 * i] = val->vsxval[0];
852 vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
853 } else {
854 r = -ENXIO;
855 }
856 break;
857 #endif /* CONFIG_VSX */
858 case KVM_REG_PPC_VPA_ADDR:
859 addr = set_reg_val(id, *val);
860 r = -EINVAL;
861 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
862 vcpu->arch.dtl.next_gpa))
863 break;
864 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
865 break;
866 case KVM_REG_PPC_VPA_SLB:
867 addr = val->vpaval.addr;
868 len = val->vpaval.length;
869 r = -EINVAL;
870 if (addr && !vcpu->arch.vpa.next_gpa)
871 break;
872 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
873 break;
874 case KVM_REG_PPC_VPA_DTL:
875 addr = val->vpaval.addr;
876 len = val->vpaval.length;
877 r = -EINVAL;
878 if (addr && (len < sizeof(struct dtl_entry) ||
879 !vcpu->arch.vpa.next_gpa))
880 break;
881 len -= len % sizeof(struct dtl_entry);
882 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
883 break;
884 default:
885 r = -EINVAL;
886 break;
887 }
888
889 return r;
890 }
891
892 int kvmppc_core_check_processor_compat(void)
893 {
894 if (cpu_has_feature(CPU_FTR_HVMODE))
895 return 0;
896 return -EIO;
897 }
898
899 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
900 {
901 struct kvm_vcpu *vcpu;
902 int err = -EINVAL;
903 int core;
904 struct kvmppc_vcore *vcore;
905
906 core = id / threads_per_core;
907 if (core >= KVM_MAX_VCORES)
908 goto out;
909
910 err = -ENOMEM;
911 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
912 if (!vcpu)
913 goto out;
914
915 err = kvm_vcpu_init(vcpu, kvm, id);
916 if (err)
917 goto free_vcpu;
918
919 vcpu->arch.shared = &vcpu->arch.shregs;
920 vcpu->arch.mmcr[0] = MMCR0_FC;
921 vcpu->arch.ctrl = CTRL_RUNLATCH;
922 /* default to host PVR, since we can't spoof it */
923 vcpu->arch.pvr = mfspr(SPRN_PVR);
924 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
925 spin_lock_init(&vcpu->arch.vpa_update_lock);
926 spin_lock_init(&vcpu->arch.tbacct_lock);
927 vcpu->arch.busy_preempt = TB_NIL;
928
929 kvmppc_mmu_book3s_hv_init(vcpu);
930
931 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
932
933 init_waitqueue_head(&vcpu->arch.cpu_run);
934
935 mutex_lock(&kvm->lock);
936 vcore = kvm->arch.vcores[core];
937 if (!vcore) {
938 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
939 if (vcore) {
940 INIT_LIST_HEAD(&vcore->runnable_threads);
941 spin_lock_init(&vcore->lock);
942 init_waitqueue_head(&vcore->wq);
943 vcore->preempt_tb = TB_NIL;
944 }
945 kvm->arch.vcores[core] = vcore;
946 kvm->arch.online_vcores++;
947 }
948 mutex_unlock(&kvm->lock);
949
950 if (!vcore)
951 goto free_vcpu;
952
953 spin_lock(&vcore->lock);
954 ++vcore->num_threads;
955 spin_unlock(&vcore->lock);
956 vcpu->arch.vcore = vcore;
957
958 vcpu->arch.cpu_type = KVM_CPU_3S_64;
959 kvmppc_sanity_check(vcpu);
960
961 return vcpu;
962
963 free_vcpu:
964 kmem_cache_free(kvm_vcpu_cache, vcpu);
965 out:
966 return ERR_PTR(err);
967 }
968
969 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
970 {
971 if (vpa->pinned_addr)
972 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
973 vpa->dirty);
974 }
975
976 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
977 {
978 spin_lock(&vcpu->arch.vpa_update_lock);
979 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
980 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
981 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
982 spin_unlock(&vcpu->arch.vpa_update_lock);
983 kvm_vcpu_uninit(vcpu);
984 kmem_cache_free(kvm_vcpu_cache, vcpu);
985 }
986
987 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
988 {
989 unsigned long dec_nsec, now;
990
991 now = get_tb();
992 if (now > vcpu->arch.dec_expires) {
993 /* decrementer has already gone negative */
994 kvmppc_core_queue_dec(vcpu);
995 kvmppc_core_prepare_to_enter(vcpu);
996 return;
997 }
998 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
999 / tb_ticks_per_sec;
1000 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1001 HRTIMER_MODE_REL);
1002 vcpu->arch.timer_running = 1;
1003 }
1004
1005 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1006 {
1007 vcpu->arch.ceded = 0;
1008 if (vcpu->arch.timer_running) {
1009 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1010 vcpu->arch.timer_running = 0;
1011 }
1012 }
1013
1014 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1015
1016 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1017 struct kvm_vcpu *vcpu)
1018 {
1019 u64 now;
1020
1021 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1022 return;
1023 spin_lock(&vcpu->arch.tbacct_lock);
1024 now = mftb();
1025 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1026 vcpu->arch.stolen_logged;
1027 vcpu->arch.busy_preempt = now;
1028 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1029 spin_unlock(&vcpu->arch.tbacct_lock);
1030 --vc->n_runnable;
1031 list_del(&vcpu->arch.run_list);
1032 }
1033
1034 static int kvmppc_grab_hwthread(int cpu)
1035 {
1036 struct paca_struct *tpaca;
1037 long timeout = 1000;
1038
1039 tpaca = &paca[cpu];
1040
1041 /* Ensure the thread won't go into the kernel if it wakes */
1042 tpaca->kvm_hstate.hwthread_req = 1;
1043 tpaca->kvm_hstate.kvm_vcpu = NULL;
1044
1045 /*
1046 * If the thread is already executing in the kernel (e.g. handling
1047 * a stray interrupt), wait for it to get back to nap mode.
1048 * The smp_mb() is to ensure that our setting of hwthread_req
1049 * is visible before we look at hwthread_state, so if this
1050 * races with the code at system_reset_pSeries and the thread
1051 * misses our setting of hwthread_req, we are sure to see its
1052 * setting of hwthread_state, and vice versa.
1053 */
1054 smp_mb();
1055 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1056 if (--timeout <= 0) {
1057 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1058 return -EBUSY;
1059 }
1060 udelay(1);
1061 }
1062 return 0;
1063 }
1064
1065 static void kvmppc_release_hwthread(int cpu)
1066 {
1067 struct paca_struct *tpaca;
1068
1069 tpaca = &paca[cpu];
1070 tpaca->kvm_hstate.hwthread_req = 0;
1071 tpaca->kvm_hstate.kvm_vcpu = NULL;
1072 }
1073
1074 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1075 {
1076 int cpu;
1077 struct paca_struct *tpaca;
1078 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1079
1080 if (vcpu->arch.timer_running) {
1081 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1082 vcpu->arch.timer_running = 0;
1083 }
1084 cpu = vc->pcpu + vcpu->arch.ptid;
1085 tpaca = &paca[cpu];
1086 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1087 tpaca->kvm_hstate.kvm_vcore = vc;
1088 tpaca->kvm_hstate.napping = 0;
1089 vcpu->cpu = vc->pcpu;
1090 smp_wmb();
1091 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1092 if (vcpu->arch.ptid) {
1093 xics_wake_cpu(cpu);
1094 ++vc->n_woken;
1095 }
1096 #endif
1097 }
1098
1099 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1100 {
1101 int i;
1102
1103 HMT_low();
1104 i = 0;
1105 while (vc->nap_count < vc->n_woken) {
1106 if (++i >= 1000000) {
1107 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1108 vc->nap_count, vc->n_woken);
1109 break;
1110 }
1111 cpu_relax();
1112 }
1113 HMT_medium();
1114 }
1115
1116 /*
1117 * Check that we are on thread 0 and that any other threads in
1118 * this core are off-line. Then grab the threads so they can't
1119 * enter the kernel.
1120 */
1121 static int on_primary_thread(void)
1122 {
1123 int cpu = smp_processor_id();
1124 int thr = cpu_thread_in_core(cpu);
1125
1126 if (thr)
1127 return 0;
1128 while (++thr < threads_per_core)
1129 if (cpu_online(cpu + thr))
1130 return 0;
1131
1132 /* Grab all hw threads so they can't go into the kernel */
1133 for (thr = 1; thr < threads_per_core; ++thr) {
1134 if (kvmppc_grab_hwthread(cpu + thr)) {
1135 /* Couldn't grab one; let the others go */
1136 do {
1137 kvmppc_release_hwthread(cpu + thr);
1138 } while (--thr > 0);
1139 return 0;
1140 }
1141 }
1142 return 1;
1143 }
1144
1145 /*
1146 * Run a set of guest threads on a physical core.
1147 * Called with vc->lock held.
1148 */
1149 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1150 {
1151 struct kvm_vcpu *vcpu, *vcpu0, *vnext;
1152 long ret;
1153 u64 now;
1154 int ptid, i, need_vpa_update;
1155 int srcu_idx;
1156 struct kvm_vcpu *vcpus_to_update[threads_per_core];
1157
1158 /* don't start if any threads have a signal pending */
1159 need_vpa_update = 0;
1160 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1161 if (signal_pending(vcpu->arch.run_task))
1162 return;
1163 if (vcpu->arch.vpa.update_pending ||
1164 vcpu->arch.slb_shadow.update_pending ||
1165 vcpu->arch.dtl.update_pending)
1166 vcpus_to_update[need_vpa_update++] = vcpu;
1167 }
1168
1169 /*
1170 * Initialize *vc, in particular vc->vcore_state, so we can
1171 * drop the vcore lock if necessary.
1172 */
1173 vc->n_woken = 0;
1174 vc->nap_count = 0;
1175 vc->entry_exit_count = 0;
1176 vc->vcore_state = VCORE_STARTING;
1177 vc->in_guest = 0;
1178 vc->napping_threads = 0;
1179
1180 /*
1181 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1182 * which can't be called with any spinlocks held.
1183 */
1184 if (need_vpa_update) {
1185 spin_unlock(&vc->lock);
1186 for (i = 0; i < need_vpa_update; ++i)
1187 kvmppc_update_vpas(vcpus_to_update[i]);
1188 spin_lock(&vc->lock);
1189 }
1190
1191 /*
1192 * Assign physical thread IDs, first to non-ceded vcpus
1193 * and then to ceded ones.
1194 */
1195 ptid = 0;
1196 vcpu0 = NULL;
1197 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1198 if (!vcpu->arch.ceded) {
1199 if (!ptid)
1200 vcpu0 = vcpu;
1201 vcpu->arch.ptid = ptid++;
1202 }
1203 }
1204 if (!vcpu0)
1205 goto out; /* nothing to run; should never happen */
1206 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1207 if (vcpu->arch.ceded)
1208 vcpu->arch.ptid = ptid++;
1209
1210 /*
1211 * Make sure we are running on thread 0, and that
1212 * secondary threads are offline.
1213 */
1214 if (threads_per_core > 1 && !on_primary_thread()) {
1215 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1216 vcpu->arch.ret = -EBUSY;
1217 goto out;
1218 }
1219
1220 vc->pcpu = smp_processor_id();
1221 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1222 kvmppc_start_thread(vcpu);
1223 kvmppc_create_dtl_entry(vcpu, vc);
1224 }
1225
1226 vc->vcore_state = VCORE_RUNNING;
1227 preempt_disable();
1228 spin_unlock(&vc->lock);
1229
1230 kvm_guest_enter();
1231
1232 srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1233
1234 __kvmppc_vcore_entry(NULL, vcpu0);
1235
1236 spin_lock(&vc->lock);
1237 /* disable sending of IPIs on virtual external irqs */
1238 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1239 vcpu->cpu = -1;
1240 /* wait for secondary threads to finish writing their state to memory */
1241 if (vc->nap_count < vc->n_woken)
1242 kvmppc_wait_for_nap(vc);
1243 for (i = 0; i < threads_per_core; ++i)
1244 kvmppc_release_hwthread(vc->pcpu + i);
1245 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1246 vc->vcore_state = VCORE_EXITING;
1247 spin_unlock(&vc->lock);
1248
1249 srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1250
1251 /* make sure updates to secondary vcpu structs are visible now */
1252 smp_mb();
1253 kvm_guest_exit();
1254
1255 preempt_enable();
1256 kvm_resched(vcpu);
1257
1258 spin_lock(&vc->lock);
1259 now = get_tb();
1260 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1261 /* cancel pending dec exception if dec is positive */
1262 if (now < vcpu->arch.dec_expires &&
1263 kvmppc_core_pending_dec(vcpu))
1264 kvmppc_core_dequeue_dec(vcpu);
1265
1266 ret = RESUME_GUEST;
1267 if (vcpu->arch.trap)
1268 ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
1269 vcpu->arch.run_task);
1270
1271 vcpu->arch.ret = ret;
1272 vcpu->arch.trap = 0;
1273
1274 if (vcpu->arch.ceded) {
1275 if (ret != RESUME_GUEST)
1276 kvmppc_end_cede(vcpu);
1277 else
1278 kvmppc_set_timer(vcpu);
1279 }
1280 }
1281
1282 out:
1283 vc->vcore_state = VCORE_INACTIVE;
1284 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1285 arch.run_list) {
1286 if (vcpu->arch.ret != RESUME_GUEST) {
1287 kvmppc_remove_runnable(vc, vcpu);
1288 wake_up(&vcpu->arch.cpu_run);
1289 }
1290 }
1291 }
1292
1293 /*
1294 * Wait for some other vcpu thread to execute us, and
1295 * wake us up when we need to handle something in the host.
1296 */
1297 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1298 {
1299 DEFINE_WAIT(wait);
1300
1301 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1302 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1303 schedule();
1304 finish_wait(&vcpu->arch.cpu_run, &wait);
1305 }
1306
1307 /*
1308 * All the vcpus in this vcore are idle, so wait for a decrementer
1309 * or external interrupt to one of the vcpus. vc->lock is held.
1310 */
1311 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1312 {
1313 DEFINE_WAIT(wait);
1314
1315 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1316 vc->vcore_state = VCORE_SLEEPING;
1317 spin_unlock(&vc->lock);
1318 schedule();
1319 finish_wait(&vc->wq, &wait);
1320 spin_lock(&vc->lock);
1321 vc->vcore_state = VCORE_INACTIVE;
1322 }
1323
1324 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1325 {
1326 int n_ceded;
1327 struct kvmppc_vcore *vc;
1328 struct kvm_vcpu *v, *vn;
1329
1330 kvm_run->exit_reason = 0;
1331 vcpu->arch.ret = RESUME_GUEST;
1332 vcpu->arch.trap = 0;
1333 kvmppc_update_vpas(vcpu);
1334
1335 /*
1336 * Synchronize with other threads in this virtual core
1337 */
1338 vc = vcpu->arch.vcore;
1339 spin_lock(&vc->lock);
1340 vcpu->arch.ceded = 0;
1341 vcpu->arch.run_task = current;
1342 vcpu->arch.kvm_run = kvm_run;
1343 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1344 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1345 vcpu->arch.busy_preempt = TB_NIL;
1346 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1347 ++vc->n_runnable;
1348
1349 /*
1350 * This happens the first time this is called for a vcpu.
1351 * If the vcore is already running, we may be able to start
1352 * this thread straight away and have it join in.
1353 */
1354 if (!signal_pending(current)) {
1355 if (vc->vcore_state == VCORE_RUNNING &&
1356 VCORE_EXIT_COUNT(vc) == 0) {
1357 vcpu->arch.ptid = vc->n_runnable - 1;
1358 kvmppc_create_dtl_entry(vcpu, vc);
1359 kvmppc_start_thread(vcpu);
1360 } else if (vc->vcore_state == VCORE_SLEEPING) {
1361 wake_up(&vc->wq);
1362 }
1363
1364 }
1365
1366 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1367 !signal_pending(current)) {
1368 if (vc->vcore_state != VCORE_INACTIVE) {
1369 spin_unlock(&vc->lock);
1370 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1371 spin_lock(&vc->lock);
1372 continue;
1373 }
1374 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1375 arch.run_list) {
1376 kvmppc_core_prepare_to_enter(v);
1377 if (signal_pending(v->arch.run_task)) {
1378 kvmppc_remove_runnable(vc, v);
1379 v->stat.signal_exits++;
1380 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1381 v->arch.ret = -EINTR;
1382 wake_up(&v->arch.cpu_run);
1383 }
1384 }
1385 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1386 break;
1387 vc->runner = vcpu;
1388 n_ceded = 0;
1389 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1390 if (!v->arch.pending_exceptions)
1391 n_ceded += v->arch.ceded;
1392 else
1393 v->arch.ceded = 0;
1394 }
1395 if (n_ceded == vc->n_runnable)
1396 kvmppc_vcore_blocked(vc);
1397 else
1398 kvmppc_run_core(vc);
1399 vc->runner = NULL;
1400 }
1401
1402 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1403 (vc->vcore_state == VCORE_RUNNING ||
1404 vc->vcore_state == VCORE_EXITING)) {
1405 spin_unlock(&vc->lock);
1406 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1407 spin_lock(&vc->lock);
1408 }
1409
1410 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1411 kvmppc_remove_runnable(vc, vcpu);
1412 vcpu->stat.signal_exits++;
1413 kvm_run->exit_reason = KVM_EXIT_INTR;
1414 vcpu->arch.ret = -EINTR;
1415 }
1416
1417 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1418 /* Wake up some vcpu to run the core */
1419 v = list_first_entry(&vc->runnable_threads,
1420 struct kvm_vcpu, arch.run_list);
1421 wake_up(&v->arch.cpu_run);
1422 }
1423
1424 spin_unlock(&vc->lock);
1425 return vcpu->arch.ret;
1426 }
1427
1428 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1429 {
1430 int r;
1431 int srcu_idx;
1432
1433 if (!vcpu->arch.sane) {
1434 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1435 return -EINVAL;
1436 }
1437
1438 kvmppc_core_prepare_to_enter(vcpu);
1439
1440 /* No need to go into the guest when all we'll do is come back out */
1441 if (signal_pending(current)) {
1442 run->exit_reason = KVM_EXIT_INTR;
1443 return -EINTR;
1444 }
1445
1446 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1447 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1448 smp_mb();
1449
1450 /* On the first time here, set up HTAB and VRMA or RMA */
1451 if (!vcpu->kvm->arch.rma_setup_done) {
1452 r = kvmppc_hv_setup_htab_rma(vcpu);
1453 if (r)
1454 goto out;
1455 }
1456
1457 flush_fp_to_thread(current);
1458 flush_altivec_to_thread(current);
1459 flush_vsx_to_thread(current);
1460 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1461 vcpu->arch.pgdir = current->mm->pgd;
1462 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1463
1464 do {
1465 r = kvmppc_run_vcpu(run, vcpu);
1466
1467 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1468 !(vcpu->arch.shregs.msr & MSR_PR)) {
1469 r = kvmppc_pseries_do_hcall(vcpu);
1470 kvmppc_core_prepare_to_enter(vcpu);
1471 } else if (r == RESUME_PAGE_FAULT) {
1472 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1473 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1474 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1475 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1476 }
1477 } while (r == RESUME_GUEST);
1478
1479 out:
1480 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1481 atomic_dec(&vcpu->kvm->arch.vcpus_running);
1482 return r;
1483 }
1484
1485
1486 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1487 Assumes POWER7 or PPC970. */
1488 static inline int lpcr_rmls(unsigned long rma_size)
1489 {
1490 switch (rma_size) {
1491 case 32ul << 20: /* 32 MB */
1492 if (cpu_has_feature(CPU_FTR_ARCH_206))
1493 return 8; /* only supported on POWER7 */
1494 return -1;
1495 case 64ul << 20: /* 64 MB */
1496 return 3;
1497 case 128ul << 20: /* 128 MB */
1498 return 7;
1499 case 256ul << 20: /* 256 MB */
1500 return 4;
1501 case 1ul << 30: /* 1 GB */
1502 return 2;
1503 case 16ul << 30: /* 16 GB */
1504 return 1;
1505 case 256ul << 30: /* 256 GB */
1506 return 0;
1507 default:
1508 return -1;
1509 }
1510 }
1511
1512 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1513 {
1514 struct kvmppc_linear_info *ri = vma->vm_file->private_data;
1515 struct page *page;
1516
1517 if (vmf->pgoff >= ri->npages)
1518 return VM_FAULT_SIGBUS;
1519
1520 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1521 get_page(page);
1522 vmf->page = page;
1523 return 0;
1524 }
1525
1526 static const struct vm_operations_struct kvm_rma_vm_ops = {
1527 .fault = kvm_rma_fault,
1528 };
1529
1530 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1531 {
1532 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1533 vma->vm_ops = &kvm_rma_vm_ops;
1534 return 0;
1535 }
1536
1537 static int kvm_rma_release(struct inode *inode, struct file *filp)
1538 {
1539 struct kvmppc_linear_info *ri = filp->private_data;
1540
1541 kvm_release_rma(ri);
1542 return 0;
1543 }
1544
1545 static const struct file_operations kvm_rma_fops = {
1546 .mmap = kvm_rma_mmap,
1547 .release = kvm_rma_release,
1548 };
1549
1550 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1551 {
1552 struct kvmppc_linear_info *ri;
1553 long fd;
1554
1555 ri = kvm_alloc_rma();
1556 if (!ri)
1557 return -ENOMEM;
1558
1559 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1560 if (fd < 0)
1561 kvm_release_rma(ri);
1562
1563 ret->rma_size = ri->npages << PAGE_SHIFT;
1564 return fd;
1565 }
1566
1567 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1568 int linux_psize)
1569 {
1570 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1571
1572 if (!def->shift)
1573 return;
1574 (*sps)->page_shift = def->shift;
1575 (*sps)->slb_enc = def->sllp;
1576 (*sps)->enc[0].page_shift = def->shift;
1577 /*
1578 * Only return base page encoding. We don't want to return
1579 * all the supporting pte_enc, because our H_ENTER doesn't
1580 * support MPSS yet. Once they do, we can start passing all
1581 * support pte_enc here
1582 */
1583 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1584 (*sps)++;
1585 }
1586
1587 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1588 {
1589 struct kvm_ppc_one_seg_page_size *sps;
1590
1591 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1592 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1593 info->flags |= KVM_PPC_1T_SEGMENTS;
1594 info->slb_size = mmu_slb_size;
1595
1596 /* We only support these sizes for now, and no muti-size segments */
1597 sps = &info->sps[0];
1598 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1599 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1600 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1601
1602 return 0;
1603 }
1604
1605 /*
1606 * Get (and clear) the dirty memory log for a memory slot.
1607 */
1608 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1609 {
1610 struct kvm_memory_slot *memslot;
1611 int r;
1612 unsigned long n;
1613
1614 mutex_lock(&kvm->slots_lock);
1615
1616 r = -EINVAL;
1617 if (log->slot >= KVM_USER_MEM_SLOTS)
1618 goto out;
1619
1620 memslot = id_to_memslot(kvm->memslots, log->slot);
1621 r = -ENOENT;
1622 if (!memslot->dirty_bitmap)
1623 goto out;
1624
1625 n = kvm_dirty_bitmap_bytes(memslot);
1626 memset(memslot->dirty_bitmap, 0, n);
1627
1628 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1629 if (r)
1630 goto out;
1631
1632 r = -EFAULT;
1633 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1634 goto out;
1635
1636 r = 0;
1637 out:
1638 mutex_unlock(&kvm->slots_lock);
1639 return r;
1640 }
1641
1642 static void unpin_slot(struct kvm_memory_slot *memslot)
1643 {
1644 unsigned long *physp;
1645 unsigned long j, npages, pfn;
1646 struct page *page;
1647
1648 physp = memslot->arch.slot_phys;
1649 npages = memslot->npages;
1650 if (!physp)
1651 return;
1652 for (j = 0; j < npages; j++) {
1653 if (!(physp[j] & KVMPPC_GOT_PAGE))
1654 continue;
1655 pfn = physp[j] >> PAGE_SHIFT;
1656 page = pfn_to_page(pfn);
1657 SetPageDirty(page);
1658 put_page(page);
1659 }
1660 }
1661
1662 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1663 struct kvm_memory_slot *dont)
1664 {
1665 if (!dont || free->arch.rmap != dont->arch.rmap) {
1666 vfree(free->arch.rmap);
1667 free->arch.rmap = NULL;
1668 }
1669 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1670 unpin_slot(free);
1671 vfree(free->arch.slot_phys);
1672 free->arch.slot_phys = NULL;
1673 }
1674 }
1675
1676 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1677 unsigned long npages)
1678 {
1679 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1680 if (!slot->arch.rmap)
1681 return -ENOMEM;
1682 slot->arch.slot_phys = NULL;
1683
1684 return 0;
1685 }
1686
1687 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1688 struct kvm_memory_slot *memslot,
1689 struct kvm_userspace_memory_region *mem)
1690 {
1691 unsigned long *phys;
1692
1693 /* Allocate a slot_phys array if needed */
1694 phys = memslot->arch.slot_phys;
1695 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1696 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1697 if (!phys)
1698 return -ENOMEM;
1699 memslot->arch.slot_phys = phys;
1700 }
1701
1702 return 0;
1703 }
1704
1705 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1706 struct kvm_userspace_memory_region *mem,
1707 const struct kvm_memory_slot *old)
1708 {
1709 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1710 struct kvm_memory_slot *memslot;
1711
1712 if (npages && old->npages) {
1713 /*
1714 * If modifying a memslot, reset all the rmap dirty bits.
1715 * If this is a new memslot, we don't need to do anything
1716 * since the rmap array starts out as all zeroes,
1717 * i.e. no pages are dirty.
1718 */
1719 memslot = id_to_memslot(kvm->memslots, mem->slot);
1720 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1721 }
1722 }
1723
1724 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1725 {
1726 int err = 0;
1727 struct kvm *kvm = vcpu->kvm;
1728 struct kvmppc_linear_info *ri = NULL;
1729 unsigned long hva;
1730 struct kvm_memory_slot *memslot;
1731 struct vm_area_struct *vma;
1732 unsigned long lpcr, senc;
1733 unsigned long psize, porder;
1734 unsigned long rma_size;
1735 unsigned long rmls;
1736 unsigned long *physp;
1737 unsigned long i, npages;
1738 int srcu_idx;
1739
1740 mutex_lock(&kvm->lock);
1741 if (kvm->arch.rma_setup_done)
1742 goto out; /* another vcpu beat us to it */
1743
1744 /* Allocate hashed page table (if not done already) and reset it */
1745 if (!kvm->arch.hpt_virt) {
1746 err = kvmppc_alloc_hpt(kvm, NULL);
1747 if (err) {
1748 pr_err("KVM: Couldn't alloc HPT\n");
1749 goto out;
1750 }
1751 }
1752
1753 /* Look up the memslot for guest physical address 0 */
1754 srcu_idx = srcu_read_lock(&kvm->srcu);
1755 memslot = gfn_to_memslot(kvm, 0);
1756
1757 /* We must have some memory at 0 by now */
1758 err = -EINVAL;
1759 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1760 goto out_srcu;
1761
1762 /* Look up the VMA for the start of this memory slot */
1763 hva = memslot->userspace_addr;
1764 down_read(&current->mm->mmap_sem);
1765 vma = find_vma(current->mm, hva);
1766 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1767 goto up_out;
1768
1769 psize = vma_kernel_pagesize(vma);
1770 porder = __ilog2(psize);
1771
1772 /* Is this one of our preallocated RMAs? */
1773 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1774 hva == vma->vm_start)
1775 ri = vma->vm_file->private_data;
1776
1777 up_read(&current->mm->mmap_sem);
1778
1779 if (!ri) {
1780 /* On POWER7, use VRMA; on PPC970, give up */
1781 err = -EPERM;
1782 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1783 pr_err("KVM: CPU requires an RMO\n");
1784 goto out_srcu;
1785 }
1786
1787 /* We can handle 4k, 64k or 16M pages in the VRMA */
1788 err = -EINVAL;
1789 if (!(psize == 0x1000 || psize == 0x10000 ||
1790 psize == 0x1000000))
1791 goto out_srcu;
1792
1793 /* Update VRMASD field in the LPCR */
1794 senc = slb_pgsize_encoding(psize);
1795 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1796 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1797 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1798 lpcr |= senc << (LPCR_VRMASD_SH - 4);
1799 kvm->arch.lpcr = lpcr;
1800
1801 /* Create HPTEs in the hash page table for the VRMA */
1802 kvmppc_map_vrma(vcpu, memslot, porder);
1803
1804 } else {
1805 /* Set up to use an RMO region */
1806 rma_size = ri->npages;
1807 if (rma_size > memslot->npages)
1808 rma_size = memslot->npages;
1809 rma_size <<= PAGE_SHIFT;
1810 rmls = lpcr_rmls(rma_size);
1811 err = -EINVAL;
1812 if (rmls < 0) {
1813 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1814 goto out_srcu;
1815 }
1816 atomic_inc(&ri->use_count);
1817 kvm->arch.rma = ri;
1818
1819 /* Update LPCR and RMOR */
1820 lpcr = kvm->arch.lpcr;
1821 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1822 /* PPC970; insert RMLS value (split field) in HID4 */
1823 lpcr &= ~((1ul << HID4_RMLS0_SH) |
1824 (3ul << HID4_RMLS2_SH));
1825 lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1826 ((rmls & 3) << HID4_RMLS2_SH);
1827 /* RMOR is also in HID4 */
1828 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1829 << HID4_RMOR_SH;
1830 } else {
1831 /* POWER7 */
1832 lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1833 lpcr |= rmls << LPCR_RMLS_SH;
1834 kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1835 }
1836 kvm->arch.lpcr = lpcr;
1837 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1838 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1839
1840 /* Initialize phys addrs of pages in RMO */
1841 npages = ri->npages;
1842 porder = __ilog2(npages);
1843 physp = memslot->arch.slot_phys;
1844 if (physp) {
1845 if (npages > memslot->npages)
1846 npages = memslot->npages;
1847 spin_lock(&kvm->arch.slot_phys_lock);
1848 for (i = 0; i < npages; ++i)
1849 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1850 porder;
1851 spin_unlock(&kvm->arch.slot_phys_lock);
1852 }
1853 }
1854
1855 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1856 smp_wmb();
1857 kvm->arch.rma_setup_done = 1;
1858 err = 0;
1859 out_srcu:
1860 srcu_read_unlock(&kvm->srcu, srcu_idx);
1861 out:
1862 mutex_unlock(&kvm->lock);
1863 return err;
1864
1865 up_out:
1866 up_read(&current->mm->mmap_sem);
1867 goto out;
1868 }
1869
1870 int kvmppc_core_init_vm(struct kvm *kvm)
1871 {
1872 unsigned long lpcr, lpid;
1873
1874 /* Allocate the guest's logical partition ID */
1875
1876 lpid = kvmppc_alloc_lpid();
1877 if (lpid < 0)
1878 return -ENOMEM;
1879 kvm->arch.lpid = lpid;
1880
1881 /*
1882 * Since we don't flush the TLB when tearing down a VM,
1883 * and this lpid might have previously been used,
1884 * make sure we flush on each core before running the new VM.
1885 */
1886 cpumask_setall(&kvm->arch.need_tlb_flush);
1887
1888 INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1889 INIT_LIST_HEAD(&kvm->arch.rtas_tokens);
1890
1891 kvm->arch.rma = NULL;
1892
1893 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1894
1895 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1896 /* PPC970; HID4 is effectively the LPCR */
1897 kvm->arch.host_lpid = 0;
1898 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1899 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1900 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1901 ((lpid & 0xf) << HID4_LPID5_SH);
1902 } else {
1903 /* POWER7; init LPCR for virtual RMA mode */
1904 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1905 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1906 lpcr &= LPCR_PECE | LPCR_LPES;
1907 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1908 LPCR_VPM0 | LPCR_VPM1;
1909 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1910 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1911 }
1912 kvm->arch.lpcr = lpcr;
1913
1914 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1915 spin_lock_init(&kvm->arch.slot_phys_lock);
1916
1917 /*
1918 * Don't allow secondary CPU threads to come online
1919 * while any KVM VMs exist.
1920 */
1921 inhibit_secondary_onlining();
1922
1923 return 0;
1924 }
1925
1926 void kvmppc_core_destroy_vm(struct kvm *kvm)
1927 {
1928 uninhibit_secondary_onlining();
1929
1930 if (kvm->arch.rma) {
1931 kvm_release_rma(kvm->arch.rma);
1932 kvm->arch.rma = NULL;
1933 }
1934
1935 kvmppc_rtas_tokens_free(kvm);
1936
1937 kvmppc_free_hpt(kvm);
1938 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1939 }
1940
1941 /* These are stubs for now */
1942 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1943 {
1944 }
1945
1946 /* We don't need to emulate any privileged instructions or dcbz */
1947 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1948 unsigned int inst, int *advance)
1949 {
1950 return EMULATE_FAIL;
1951 }
1952
1953 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
1954 {
1955 return EMULATE_FAIL;
1956 }
1957
1958 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
1959 {
1960 return EMULATE_FAIL;
1961 }
1962
1963 static int kvmppc_book3s_hv_init(void)
1964 {
1965 int r;
1966
1967 r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1968
1969 if (r)
1970 return r;
1971
1972 r = kvmppc_mmu_hv_init();
1973
1974 return r;
1975 }
1976
1977 static void kvmppc_book3s_hv_exit(void)
1978 {
1979 kvm_exit();
1980 }
1981
1982 module_init(kvmppc_book3s_hv_init);
1983 module_exit(kvmppc_book3s_hv_exit);
Linux kernel - Deliverables
This page took 0.184892 seconds and 5 git commands to generate.