?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux/fpc-iii.git] / arch / powerpc / kvm / book3s_hv.c
blob17fc9496b6aca0d088bf6d6b13d38dca19121a13
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 #include <linux/miscdevice.h>
35
36 #include <asm/reg.h>
37 #include <asm/cputable.h>
38 #include <asm/cacheflush.h>
39 #include <asm/tlbflush.h>
40 #include <asm/uaccess.h>
41 #include <asm/io.h>
42 #include <asm/kvm_ppc.h>
43 #include <asm/kvm_book3s.h>
44 #include <asm/mmu_context.h>
45 #include <asm/lppaca.h>
46 #include <asm/processor.h>
47 #include <asm/cputhreads.h>
48 #include <asm/page.h>
49 #include <asm/hvcall.h>
50 #include <asm/switch_to.h>
51 #include <asm/smp.h>
52 #include <linux/gfp.h>
53 #include <linux/vmalloc.h>
54 #include <linux/highmem.h>
55 #include <linux/hugetlb.h>
56 #include <linux/module.h>
57
58 #include "book3s.h"
59
60 /* #define EXIT_DEBUG */
61 /* #define EXIT_DEBUG_SIMPLE */
62 /* #define EXIT_DEBUG_INT */
63
64 /* Used to indicate that a guest page fault needs to be handled */
65 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
66
67 /* Used as a "null" value for timebase values */
68 #define TB_NIL (~(u64)0)
69
70 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
71 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
72
73 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
74 {
75 int me;
76 int cpu = vcpu->cpu;
77 wait_queue_head_t *wqp;
78
79 wqp = kvm_arch_vcpu_wq(vcpu);
80 if (waitqueue_active(wqp)) {
81 wake_up_interruptible(wqp);
82 ++vcpu->stat.halt_wakeup;
83 }
84
85 me = get_cpu();
86
87 /* CPU points to the first thread of the core */
88 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
89 #ifdef CONFIG_KVM_XICS
90 int real_cpu = cpu + vcpu->arch.ptid;
91 if (paca[real_cpu].kvm_hstate.xics_phys)
92 xics_wake_cpu(real_cpu);
93 else
94 #endif
95 if (cpu_online(cpu))
96 smp_send_reschedule(cpu);
97 }
98 put_cpu();
99 }
100
101 /*
102 * We use the vcpu_load/put functions to measure stolen time.
103 * Stolen time is counted as time when either the vcpu is able to
104 * run as part of a virtual core, but the task running the vcore
105 * is preempted or sleeping, or when the vcpu needs something done
106 * in the kernel by the task running the vcpu, but that task is
107 * preempted or sleeping. Those two things have to be counted
108 * separately, since one of the vcpu tasks will take on the job
109 * of running the core, and the other vcpu tasks in the vcore will
110 * sleep waiting for it to do that, but that sleep shouldn't count
111 * as stolen time.
112 *
113 * Hence we accumulate stolen time when the vcpu can run as part of
114 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
115 * needs its task to do other things in the kernel (for example,
116 * service a page fault) in busy_stolen. We don't accumulate
117 * stolen time for a vcore when it is inactive, or for a vcpu
118 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
119 * a misnomer; it means that the vcpu task is not executing in
120 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
121 * the kernel. We don't have any way of dividing up that time
122 * between time that the vcpu is genuinely stopped, time that
123 * the task is actively working on behalf of the vcpu, and time
124 * that the task is preempted, so we don't count any of it as
125 * stolen.
126 *
127 * Updates to busy_stolen are protected by arch.tbacct_lock;
128 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
129 * of the vcpu that has taken responsibility for running the vcore
130 * (i.e. vc->runner). The stolen times are measured in units of
131 * timebase ticks. (Note that the != TB_NIL checks below are
132 * purely defensive; they should never fail.)
133 */
134
135 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
136 {
137 struct kvmppc_vcore *vc = vcpu->arch.vcore;
138 unsigned long flags;
139
140 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
141 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
142 vc->preempt_tb != TB_NIL) {
143 vc->stolen_tb += mftb() - vc->preempt_tb;
144 vc->preempt_tb = TB_NIL;
145 }
146 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
147 vcpu->arch.busy_preempt != TB_NIL) {
148 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
149 vcpu->arch.busy_preempt = TB_NIL;
150 }
151 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
152 }
153
154 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
155 {
156 struct kvmppc_vcore *vc = vcpu->arch.vcore;
157 unsigned long flags;
158
159 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
160 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
161 vc->preempt_tb = mftb();
162 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
163 vcpu->arch.busy_preempt = mftb();
164 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
165 }
166
167 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
168 {
169 vcpu->arch.shregs.msr = msr;
170 kvmppc_end_cede(vcpu);
171 }
172
173 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
174 {
175 vcpu->arch.pvr = pvr;
176 }
177
178 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
179 {
180 unsigned long pcr = 0;
181 struct kvmppc_vcore *vc = vcpu->arch.vcore;
182
183 if (arch_compat) {
184 if (!cpu_has_feature(CPU_FTR_ARCH_206))
185 return -EINVAL; /* 970 has no compat mode support */
186
187 switch (arch_compat) {
188 case PVR_ARCH_205:
189 /*
190 * If an arch bit is set in PCR, all the defined
191 * higher-order arch bits also have to be set.
192 */
193 pcr = PCR_ARCH_206 | PCR_ARCH_205;
194 break;
195 case PVR_ARCH_206:
196 case PVR_ARCH_206p:
197 pcr = PCR_ARCH_206;
198 break;
199 case PVR_ARCH_207:
200 break;
201 default:
202 return -EINVAL;
203 }
204
205 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
206 /* POWER7 can't emulate POWER8 */
207 if (!(pcr & PCR_ARCH_206))
208 return -EINVAL;
209 pcr &= ~PCR_ARCH_206;
210 }
211 }
212
213 spin_lock(&vc->lock);
214 vc->arch_compat = arch_compat;
215 vc->pcr = pcr;
216 spin_unlock(&vc->lock);
217
218 return 0;
219 }
220
221 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
222 {
223 int r;
224
225 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
226 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
227 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
228 for (r = 0; r < 16; ++r)
229 pr_err("r%2d = %.16lx r%d = %.16lx\n",
230 r, kvmppc_get_gpr(vcpu, r),
231 r+16, kvmppc_get_gpr(vcpu, r+16));
232 pr_err("ctr = %.16lx lr = %.16lx\n",
233 vcpu->arch.ctr, vcpu->arch.lr);
234 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
235 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
236 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
237 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
238 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
239 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
240 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
241 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
242 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
243 pr_err("fault dar = %.16lx dsisr = %.8x\n",
244 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
245 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
246 for (r = 0; r < vcpu->arch.slb_max; ++r)
247 pr_err(" ESID = %.16llx VSID = %.16llx\n",
248 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
249 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
250 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
251 vcpu->arch.last_inst);
252 }
253
254 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
255 {
256 int r;
257 struct kvm_vcpu *v, *ret = NULL;
258
259 mutex_lock(&kvm->lock);
260 kvm_for_each_vcpu(r, v, kvm) {
261 if (v->vcpu_id == id) {
262 ret = v;
263 break;
264 }
265 }
266 mutex_unlock(&kvm->lock);
267 return ret;
268 }
269
270 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
271 {
272 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
273 vpa->yield_count = 1;
274 }
275
276 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
277 unsigned long addr, unsigned long len)
278 {
279 /* check address is cacheline aligned */
280 if (addr & (L1_CACHE_BYTES - 1))
281 return -EINVAL;
282 spin_lock(&vcpu->arch.vpa_update_lock);
283 if (v->next_gpa != addr || v->len != len) {
284 v->next_gpa = addr;
285 v->len = addr ? len : 0;
286 v->update_pending = 1;
287 }
288 spin_unlock(&vcpu->arch.vpa_update_lock);
289 return 0;
290 }
291
292 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
293 struct reg_vpa {
294 u32 dummy;
295 union {
296 u16 hword;
297 u32 word;
298 } length;
299 };
300
301 static int vpa_is_registered(struct kvmppc_vpa *vpap)
302 {
303 if (vpap->update_pending)
304 return vpap->next_gpa != 0;
305 return vpap->pinned_addr != NULL;
306 }
307
308 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
309 unsigned long flags,
310 unsigned long vcpuid, unsigned long vpa)
311 {
312 struct kvm *kvm = vcpu->kvm;
313 unsigned long len, nb;
314 void *va;
315 struct kvm_vcpu *tvcpu;
316 int err;
317 int subfunc;
318 struct kvmppc_vpa *vpap;
319
320 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
321 if (!tvcpu)
322 return H_PARAMETER;
323
324 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
325 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
326 subfunc == H_VPA_REG_SLB) {
327 /* Registering new area - address must be cache-line aligned */
328 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
329 return H_PARAMETER;
330
331 /* convert logical addr to kernel addr and read length */
332 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
333 if (va == NULL)
334 return H_PARAMETER;
335 if (subfunc == H_VPA_REG_VPA)
336 len = ((struct reg_vpa *)va)->length.hword;
337 else
338 len = ((struct reg_vpa *)va)->length.word;
339 kvmppc_unpin_guest_page(kvm, va, vpa, false);
340
341 /* Check length */
342 if (len > nb || len < sizeof(struct reg_vpa))
343 return H_PARAMETER;
344 } else {
345 vpa = 0;
346 len = 0;
347 }
348
349 err = H_PARAMETER;
350 vpap = NULL;
351 spin_lock(&tvcpu->arch.vpa_update_lock);
352
353 switch (subfunc) {
354 case H_VPA_REG_VPA: /* register VPA */
355 if (len < sizeof(struct lppaca))
356 break;
357 vpap = &tvcpu->arch.vpa;
358 err = 0;
359 break;
360
361 case H_VPA_REG_DTL: /* register DTL */
362 if (len < sizeof(struct dtl_entry))
363 break;
364 len -= len % sizeof(struct dtl_entry);
365
366 /* Check that they have previously registered a VPA */
367 err = H_RESOURCE;
368 if (!vpa_is_registered(&tvcpu->arch.vpa))
369 break;
370
371 vpap = &tvcpu->arch.dtl;
372 err = 0;
373 break;
374
375 case H_VPA_REG_SLB: /* register SLB shadow buffer */
376 /* Check that they have previously registered a VPA */
377 err = H_RESOURCE;
378 if (!vpa_is_registered(&tvcpu->arch.vpa))
379 break;
380
381 vpap = &tvcpu->arch.slb_shadow;
382 err = 0;
383 break;
384
385 case H_VPA_DEREG_VPA: /* deregister VPA */
386 /* Check they don't still have a DTL or SLB buf registered */
387 err = H_RESOURCE;
388 if (vpa_is_registered(&tvcpu->arch.dtl) ||
389 vpa_is_registered(&tvcpu->arch.slb_shadow))
390 break;
391
392 vpap = &tvcpu->arch.vpa;
393 err = 0;
394 break;
395
396 case H_VPA_DEREG_DTL: /* deregister DTL */
397 vpap = &tvcpu->arch.dtl;
398 err = 0;
399 break;
400
401 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
402 vpap = &tvcpu->arch.slb_shadow;
403 err = 0;
404 break;
405 }
406
407 if (vpap) {
408 vpap->next_gpa = vpa;
409 vpap->len = len;
410 vpap->update_pending = 1;
411 }
412
413 spin_unlock(&tvcpu->arch.vpa_update_lock);
414
415 return err;
416 }
417
418 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
419 {
420 struct kvm *kvm = vcpu->kvm;
421 void *va;
422 unsigned long nb;
423 unsigned long gpa;
424
425 /*
426 * We need to pin the page pointed to by vpap->next_gpa,
427 * but we can't call kvmppc_pin_guest_page under the lock
428 * as it does get_user_pages() and down_read(). So we
429 * have to drop the lock, pin the page, then get the lock
430 * again and check that a new area didn't get registered
431 * in the meantime.
432 */
433 for (;;) {
434 gpa = vpap->next_gpa;
435 spin_unlock(&vcpu->arch.vpa_update_lock);
436 va = NULL;
437 nb = 0;
438 if (gpa)
439 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
440 spin_lock(&vcpu->arch.vpa_update_lock);
441 if (gpa == vpap->next_gpa)
442 break;
443 /* sigh... unpin that one and try again */
444 if (va)
445 kvmppc_unpin_guest_page(kvm, va, gpa, false);
446 }
447
448 vpap->update_pending = 0;
449 if (va && nb < vpap->len) {
450 /*
451 * If it's now too short, it must be that userspace
452 * has changed the mappings underlying guest memory,
453 * so unregister the region.
454 */
455 kvmppc_unpin_guest_page(kvm, va, gpa, false);
456 va = NULL;
457 }
458 if (vpap->pinned_addr)
459 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
460 vpap->dirty);
461 vpap->gpa = gpa;
462 vpap->pinned_addr = va;
463 vpap->dirty = false;
464 if (va)
465 vpap->pinned_end = va + vpap->len;
466 }
467
468 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
469 {
470 if (!(vcpu->arch.vpa.update_pending ||
471 vcpu->arch.slb_shadow.update_pending ||
472 vcpu->arch.dtl.update_pending))
473 return;
474
475 spin_lock(&vcpu->arch.vpa_update_lock);
476 if (vcpu->arch.vpa.update_pending) {
477 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
478 if (vcpu->arch.vpa.pinned_addr)
479 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
480 }
481 if (vcpu->arch.dtl.update_pending) {
482 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
483 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
484 vcpu->arch.dtl_index = 0;
485 }
486 if (vcpu->arch.slb_shadow.update_pending)
487 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
488 spin_unlock(&vcpu->arch.vpa_update_lock);
489 }
490
491 /*
492 * Return the accumulated stolen time for the vcore up until `now'.
493 * The caller should hold the vcore lock.
494 */
495 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
496 {
497 u64 p;
498
499 /*
500 * If we are the task running the vcore, then since we hold
501 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
502 * can't be updated, so we don't need the tbacct_lock.
503 * If the vcore is inactive, it can't become active (since we
504 * hold the vcore lock), so the vcpu load/put functions won't
505 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
506 */
507 if (vc->vcore_state != VCORE_INACTIVE &&
508 vc->runner->arch.run_task != current) {
509 spin_lock_irq(&vc->runner->arch.tbacct_lock);
510 p = vc->stolen_tb;
511 if (vc->preempt_tb != TB_NIL)
512 p += now - vc->preempt_tb;
513 spin_unlock_irq(&vc->runner->arch.tbacct_lock);
514 } else {
515 p = vc->stolen_tb;
516 }
517 return p;
518 }
519
520 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
521 struct kvmppc_vcore *vc)
522 {
523 struct dtl_entry *dt;
524 struct lppaca *vpa;
525 unsigned long stolen;
526 unsigned long core_stolen;
527 u64 now;
528
529 dt = vcpu->arch.dtl_ptr;
530 vpa = vcpu->arch.vpa.pinned_addr;
531 now = mftb();
532 core_stolen = vcore_stolen_time(vc, now);
533 stolen = core_stolen - vcpu->arch.stolen_logged;
534 vcpu->arch.stolen_logged = core_stolen;
535 spin_lock_irq(&vcpu->arch.tbacct_lock);
536 stolen += vcpu->arch.busy_stolen;
537 vcpu->arch.busy_stolen = 0;
538 spin_unlock_irq(&vcpu->arch.tbacct_lock);
539 if (!dt || !vpa)
540 return;
541 memset(dt, 0, sizeof(struct dtl_entry));
542 dt->dispatch_reason = 7;
543 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
544 dt->timebase = now + vc->tb_offset;
545 dt->enqueue_to_dispatch_time = stolen;
546 dt->srr0 = kvmppc_get_pc(vcpu);
547 dt->srr1 = vcpu->arch.shregs.msr;
548 ++dt;
549 if (dt == vcpu->arch.dtl.pinned_end)
550 dt = vcpu->arch.dtl.pinned_addr;
551 vcpu->arch.dtl_ptr = dt;
552 /* order writing *dt vs. writing vpa->dtl_idx */
553 smp_wmb();
554 vpa->dtl_idx = ++vcpu->arch.dtl_index;
555 vcpu->arch.dtl.dirty = true;
556 }
557
558 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
559 {
560 unsigned long req = kvmppc_get_gpr(vcpu, 3);
561 unsigned long target, ret = H_SUCCESS;
562 struct kvm_vcpu *tvcpu;
563 int idx, rc;
564
565 switch (req) {
566 case H_ENTER:
567 idx = srcu_read_lock(&vcpu->kvm->srcu);
568 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
569 kvmppc_get_gpr(vcpu, 5),
570 kvmppc_get_gpr(vcpu, 6),
571 kvmppc_get_gpr(vcpu, 7));
572 srcu_read_unlock(&vcpu->kvm->srcu, idx);
573 break;
574 case H_CEDE:
575 break;
576 case H_PROD:
577 target = kvmppc_get_gpr(vcpu, 4);
578 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
579 if (!tvcpu) {
580 ret = H_PARAMETER;
581 break;
582 }
583 tvcpu->arch.prodded = 1;
584 smp_mb();
585 if (vcpu->arch.ceded) {
586 if (waitqueue_active(&vcpu->wq)) {
587 wake_up_interruptible(&vcpu->wq);
588 vcpu->stat.halt_wakeup++;
589 }
590 }
591 break;
592 case H_CONFER:
593 target = kvmppc_get_gpr(vcpu, 4);
594 if (target == -1)
595 break;
596 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
597 if (!tvcpu) {
598 ret = H_PARAMETER;
599 break;
600 }
601 kvm_vcpu_yield_to(tvcpu);
602 break;
603 case H_REGISTER_VPA:
604 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
605 kvmppc_get_gpr(vcpu, 5),
606 kvmppc_get_gpr(vcpu, 6));
607 break;
608 case H_RTAS:
609 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
610 return RESUME_HOST;
611
612 idx = srcu_read_lock(&vcpu->kvm->srcu);
613 rc = kvmppc_rtas_hcall(vcpu);
614 srcu_read_unlock(&vcpu->kvm->srcu, idx);
615
616 if (rc == -ENOENT)
617 return RESUME_HOST;
618 else if (rc == 0)
619 break;
620
621 /* Send the error out to userspace via KVM_RUN */
622 return rc;
623
624 case H_XIRR:
625 case H_CPPR:
626 case H_EOI:
627 case H_IPI:
628 case H_IPOLL:
629 case H_XIRR_X:
630 if (kvmppc_xics_enabled(vcpu)) {
631 ret = kvmppc_xics_hcall(vcpu, req);
632 break;
633 } /* fallthrough */
634 default:
635 return RESUME_HOST;
636 }
637 kvmppc_set_gpr(vcpu, 3, ret);
638 vcpu->arch.hcall_needed = 0;
639 return RESUME_GUEST;
640 }
641
642 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
643 struct task_struct *tsk)
644 {
645 int r = RESUME_HOST;
646
647 vcpu->stat.sum_exits++;
648
649 run->exit_reason = KVM_EXIT_UNKNOWN;
650 run->ready_for_interrupt_injection = 1;
651 switch (vcpu->arch.trap) {
652 /* We're good on these - the host merely wanted to get our attention */
653 case BOOK3S_INTERRUPT_HV_DECREMENTER:
654 vcpu->stat.dec_exits++;
655 r = RESUME_GUEST;
656 break;
657 case BOOK3S_INTERRUPT_EXTERNAL:
658 case BOOK3S_INTERRUPT_H_DOORBELL:
659 vcpu->stat.ext_intr_exits++;
660 r = RESUME_GUEST;
661 break;
662 case BOOK3S_INTERRUPT_PERFMON:
663 r = RESUME_GUEST;
664 break;
665 case BOOK3S_INTERRUPT_MACHINE_CHECK:
666 /*
667 * Deliver a machine check interrupt to the guest.
668 * We have to do this, even if the host has handled the
669 * machine check, because machine checks use SRR0/1 and
670 * the interrupt might have trashed guest state in them.
671 */
672 kvmppc_book3s_queue_irqprio(vcpu,
673 BOOK3S_INTERRUPT_MACHINE_CHECK);
674 r = RESUME_GUEST;
675 break;
676 case BOOK3S_INTERRUPT_PROGRAM:
677 {
678 ulong flags;
679 /*
680 * Normally program interrupts are delivered directly
681 * to the guest by the hardware, but we can get here
682 * as a result of a hypervisor emulation interrupt
683 * (e40) getting turned into a 700 by BML RTAS.
684 */
685 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
686 kvmppc_core_queue_program(vcpu, flags);
687 r = RESUME_GUEST;
688 break;
689 }
690 case BOOK3S_INTERRUPT_SYSCALL:
691 {
692 /* hcall - punt to userspace */
693 int i;
694
695 /* hypercall with MSR_PR has already been handled in rmode,
696 * and never reaches here.
697 */
698
699 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
700 for (i = 0; i < 9; ++i)
701 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
702 run->exit_reason = KVM_EXIT_PAPR_HCALL;
703 vcpu->arch.hcall_needed = 1;
704 r = RESUME_HOST;
705 break;
706 }
707 /*
708 * We get these next two if the guest accesses a page which it thinks
709 * it has mapped but which is not actually present, either because
710 * it is for an emulated I/O device or because the corresonding
711 * host page has been paged out. Any other HDSI/HISI interrupts
712 * have been handled already.
713 */
714 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
715 r = RESUME_PAGE_FAULT;
716 break;
717 case BOOK3S_INTERRUPT_H_INST_STORAGE:
718 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
719 vcpu->arch.fault_dsisr = 0;
720 r = RESUME_PAGE_FAULT;
721 break;
722 /*
723 * This occurs if the guest executes an illegal instruction.
724 * We just generate a program interrupt to the guest, since
725 * we don't emulate any guest instructions at this stage.
726 */
727 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
728 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
729 r = RESUME_GUEST;
730 break;
731 /*
732 * This occurs if the guest (kernel or userspace), does something that
733 * is prohibited by HFSCR. We just generate a program interrupt to
734 * the guest.
735 */
736 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
737 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
738 r = RESUME_GUEST;
739 break;
740 default:
741 kvmppc_dump_regs(vcpu);
742 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
743 vcpu->arch.trap, kvmppc_get_pc(vcpu),
744 vcpu->arch.shregs.msr);
745 run->hw.hardware_exit_reason = vcpu->arch.trap;
746 r = RESUME_HOST;
747 break;
748 }
749
750 return r;
751 }
752
753 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
754 struct kvm_sregs *sregs)
755 {
756 int i;
757
758 memset(sregs, 0, sizeof(struct kvm_sregs));
759 sregs->pvr = vcpu->arch.pvr;
760 for (i = 0; i < vcpu->arch.slb_max; i++) {
761 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
762 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
763 }
764
765 return 0;
766 }
767
768 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
769 struct kvm_sregs *sregs)
770 {
771 int i, j;
772
773 kvmppc_set_pvr_hv(vcpu, sregs->pvr);
774
775 j = 0;
776 for (i = 0; i < vcpu->arch.slb_nr; i++) {
777 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
778 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
779 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
780 ++j;
781 }
782 }
783 vcpu->arch.slb_max = j;
784
785 return 0;
786 }
787
788 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
789 {
790 struct kvmppc_vcore *vc = vcpu->arch.vcore;
791 u64 mask;
792
793 spin_lock(&vc->lock);
794 /*
795 * If ILE (interrupt little-endian) has changed, update the
796 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
797 */
798 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
799 struct kvm *kvm = vcpu->kvm;
800 struct kvm_vcpu *vcpu;
801 int i;
802
803 mutex_lock(&kvm->lock);
804 kvm_for_each_vcpu(i, vcpu, kvm) {
805 if (vcpu->arch.vcore != vc)
806 continue;
807 if (new_lpcr & LPCR_ILE)
808 vcpu->arch.intr_msr |= MSR_LE;
809 else
810 vcpu->arch.intr_msr &= ~MSR_LE;
811 }
812 mutex_unlock(&kvm->lock);
813 }
814
815 /*
816 * Userspace can only modify DPFD (default prefetch depth),
817 * ILE (interrupt little-endian) and TC (translation control).
818 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
819 */
820 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
821 if (cpu_has_feature(CPU_FTR_ARCH_207S))
822 mask |= LPCR_AIL;
823 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
824 spin_unlock(&vc->lock);
825 }
826
827 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
828 union kvmppc_one_reg *val)
829 {
830 int r = 0;
831 long int i;
832
833 switch (id) {
834 case KVM_REG_PPC_HIOR:
835 *val = get_reg_val(id, 0);
836 break;
837 case KVM_REG_PPC_DABR:
838 *val = get_reg_val(id, vcpu->arch.dabr);
839 break;
840 case KVM_REG_PPC_DABRX:
841 *val = get_reg_val(id, vcpu->arch.dabrx);
842 break;
843 case KVM_REG_PPC_DSCR:
844 *val = get_reg_val(id, vcpu->arch.dscr);
845 break;
846 case KVM_REG_PPC_PURR:
847 *val = get_reg_val(id, vcpu->arch.purr);
848 break;
849 case KVM_REG_PPC_SPURR:
850 *val = get_reg_val(id, vcpu->arch.spurr);
851 break;
852 case KVM_REG_PPC_AMR:
853 *val = get_reg_val(id, vcpu->arch.amr);
854 break;
855 case KVM_REG_PPC_UAMOR:
856 *val = get_reg_val(id, vcpu->arch.uamor);
857 break;
858 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
859 i = id - KVM_REG_PPC_MMCR0;
860 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
861 break;
862 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
863 i = id - KVM_REG_PPC_PMC1;
864 *val = get_reg_val(id, vcpu->arch.pmc[i]);
865 break;
866 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
867 i = id - KVM_REG_PPC_SPMC1;
868 *val = get_reg_val(id, vcpu->arch.spmc[i]);
869 break;
870 case KVM_REG_PPC_SIAR:
871 *val = get_reg_val(id, vcpu->arch.siar);
872 break;
873 case KVM_REG_PPC_SDAR:
874 *val = get_reg_val(id, vcpu->arch.sdar);
875 break;
876 case KVM_REG_PPC_SIER:
877 *val = get_reg_val(id, vcpu->arch.sier);
878 break;
879 case KVM_REG_PPC_IAMR:
880 *val = get_reg_val(id, vcpu->arch.iamr);
881 break;
882 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
883 case KVM_REG_PPC_TFHAR:
884 *val = get_reg_val(id, vcpu->arch.tfhar);
885 break;
886 case KVM_REG_PPC_TFIAR:
887 *val = get_reg_val(id, vcpu->arch.tfiar);
888 break;
889 case KVM_REG_PPC_TEXASR:
890 *val = get_reg_val(id, vcpu->arch.texasr);
891 break;
892 #endif
893 case KVM_REG_PPC_FSCR:
894 *val = get_reg_val(id, vcpu->arch.fscr);
895 break;
896 case KVM_REG_PPC_PSPB:
897 *val = get_reg_val(id, vcpu->arch.pspb);
898 break;
899 case KVM_REG_PPC_EBBHR:
900 *val = get_reg_val(id, vcpu->arch.ebbhr);
901 break;
902 case KVM_REG_PPC_EBBRR:
903 *val = get_reg_val(id, vcpu->arch.ebbrr);
904 break;
905 case KVM_REG_PPC_BESCR:
906 *val = get_reg_val(id, vcpu->arch.bescr);
907 break;
908 case KVM_REG_PPC_TAR:
909 *val = get_reg_val(id, vcpu->arch.tar);
910 break;
911 case KVM_REG_PPC_DPDES:
912 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
913 break;
914 case KVM_REG_PPC_DAWR:
915 *val = get_reg_val(id, vcpu->arch.dawr);
916 break;
917 case KVM_REG_PPC_DAWRX:
918 *val = get_reg_val(id, vcpu->arch.dawrx);
919 break;
920 case KVM_REG_PPC_CIABR:
921 *val = get_reg_val(id, vcpu->arch.ciabr);
922 break;
923 case KVM_REG_PPC_IC:
924 *val = get_reg_val(id, vcpu->arch.ic);
925 break;
926 case KVM_REG_PPC_VTB:
927 *val = get_reg_val(id, vcpu->arch.vtb);
928 break;
929 case KVM_REG_PPC_CSIGR:
930 *val = get_reg_val(id, vcpu->arch.csigr);
931 break;
932 case KVM_REG_PPC_TACR:
933 *val = get_reg_val(id, vcpu->arch.tacr);
934 break;
935 case KVM_REG_PPC_TCSCR:
936 *val = get_reg_val(id, vcpu->arch.tcscr);
937 break;
938 case KVM_REG_PPC_PID:
939 *val = get_reg_val(id, vcpu->arch.pid);
940 break;
941 case KVM_REG_PPC_ACOP:
942 *val = get_reg_val(id, vcpu->arch.acop);
943 break;
944 case KVM_REG_PPC_WORT:
945 *val = get_reg_val(id, vcpu->arch.wort);
946 break;
947 case KVM_REG_PPC_VPA_ADDR:
948 spin_lock(&vcpu->arch.vpa_update_lock);
949 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
950 spin_unlock(&vcpu->arch.vpa_update_lock);
951 break;
952 case KVM_REG_PPC_VPA_SLB:
953 spin_lock(&vcpu->arch.vpa_update_lock);
954 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
955 val->vpaval.length = vcpu->arch.slb_shadow.len;
956 spin_unlock(&vcpu->arch.vpa_update_lock);
957 break;
958 case KVM_REG_PPC_VPA_DTL:
959 spin_lock(&vcpu->arch.vpa_update_lock);
960 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
961 val->vpaval.length = vcpu->arch.dtl.len;
962 spin_unlock(&vcpu->arch.vpa_update_lock);
963 break;
964 case KVM_REG_PPC_TB_OFFSET:
965 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
966 break;
967 case KVM_REG_PPC_LPCR:
968 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
969 break;
970 case KVM_REG_PPC_PPR:
971 *val = get_reg_val(id, vcpu->arch.ppr);
972 break;
973 case KVM_REG_PPC_ARCH_COMPAT:
974 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
975 break;
976 default:
977 r = -EINVAL;
978 break;
979 }
980
981 return r;
982 }
983
984 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
985 union kvmppc_one_reg *val)
986 {
987 int r = 0;
988 long int i;
989 unsigned long addr, len;
990
991 switch (id) {
992 case KVM_REG_PPC_HIOR:
993 /* Only allow this to be set to zero */
994 if (set_reg_val(id, *val))
995 r = -EINVAL;
996 break;
997 case KVM_REG_PPC_DABR:
998 vcpu->arch.dabr = set_reg_val(id, *val);
999 break;
1000 case KVM_REG_PPC_DABRX:
1001 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1002 break;
1003 case KVM_REG_PPC_DSCR:
1004 vcpu->arch.dscr = set_reg_val(id, *val);
1005 break;
1006 case KVM_REG_PPC_PURR:
1007 vcpu->arch.purr = set_reg_val(id, *val);
1008 break;
1009 case KVM_REG_PPC_SPURR:
1010 vcpu->arch.spurr = set_reg_val(id, *val);
1011 break;
1012 case KVM_REG_PPC_AMR:
1013 vcpu->arch.amr = set_reg_val(id, *val);
1014 break;
1015 case KVM_REG_PPC_UAMOR:
1016 vcpu->arch.uamor = set_reg_val(id, *val);
1017 break;
1018 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1019 i = id - KVM_REG_PPC_MMCR0;
1020 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1021 break;
1022 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1023 i = id - KVM_REG_PPC_PMC1;
1024 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1025 break;
1026 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1027 i = id - KVM_REG_PPC_SPMC1;
1028 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1029 break;
1030 case KVM_REG_PPC_SIAR:
1031 vcpu->arch.siar = set_reg_val(id, *val);
1032 break;
1033 case KVM_REG_PPC_SDAR:
1034 vcpu->arch.sdar = set_reg_val(id, *val);
1035 break;
1036 case KVM_REG_PPC_SIER:
1037 vcpu->arch.sier = set_reg_val(id, *val);
1038 break;
1039 case KVM_REG_PPC_IAMR:
1040 vcpu->arch.iamr = set_reg_val(id, *val);
1041 break;
1042 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1043 case KVM_REG_PPC_TFHAR:
1044 vcpu->arch.tfhar = set_reg_val(id, *val);
1045 break;
1046 case KVM_REG_PPC_TFIAR:
1047 vcpu->arch.tfiar = set_reg_val(id, *val);
1048 break;
1049 case KVM_REG_PPC_TEXASR:
1050 vcpu->arch.texasr = set_reg_val(id, *val);
1051 break;
1052 #endif
1053 case KVM_REG_PPC_FSCR:
1054 vcpu->arch.fscr = set_reg_val(id, *val);
1055 break;
1056 case KVM_REG_PPC_PSPB:
1057 vcpu->arch.pspb = set_reg_val(id, *val);
1058 break;
1059 case KVM_REG_PPC_EBBHR:
1060 vcpu->arch.ebbhr = set_reg_val(id, *val);
1061 break;
1062 case KVM_REG_PPC_EBBRR:
1063 vcpu->arch.ebbrr = set_reg_val(id, *val);
1064 break;
1065 case KVM_REG_PPC_BESCR:
1066 vcpu->arch.bescr = set_reg_val(id, *val);
1067 break;
1068 case KVM_REG_PPC_TAR:
1069 vcpu->arch.tar = set_reg_val(id, *val);
1070 break;
1071 case KVM_REG_PPC_DPDES:
1072 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1073 break;
1074 case KVM_REG_PPC_DAWR:
1075 vcpu->arch.dawr = set_reg_val(id, *val);
1076 break;
1077 case KVM_REG_PPC_DAWRX:
1078 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1079 break;
1080 case KVM_REG_PPC_CIABR:
1081 vcpu->arch.ciabr = set_reg_val(id, *val);
1082 /* Don't allow setting breakpoints in hypervisor code */
1083 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1084 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1085 break;
1086 case KVM_REG_PPC_IC:
1087 vcpu->arch.ic = set_reg_val(id, *val);
1088 break;
1089 case KVM_REG_PPC_VTB:
1090 vcpu->arch.vtb = set_reg_val(id, *val);
1091 break;
1092 case KVM_REG_PPC_CSIGR:
1093 vcpu->arch.csigr = set_reg_val(id, *val);
1094 break;
1095 case KVM_REG_PPC_TACR:
1096 vcpu->arch.tacr = set_reg_val(id, *val);
1097 break;
1098 case KVM_REG_PPC_TCSCR:
1099 vcpu->arch.tcscr = set_reg_val(id, *val);
1100 break;
1101 case KVM_REG_PPC_PID:
1102 vcpu->arch.pid = set_reg_val(id, *val);
1103 break;
1104 case KVM_REG_PPC_ACOP:
1105 vcpu->arch.acop = set_reg_val(id, *val);
1106 break;
1107 case KVM_REG_PPC_WORT:
1108 vcpu->arch.wort = set_reg_val(id, *val);
1109 break;
1110 case KVM_REG_PPC_VPA_ADDR:
1111 addr = set_reg_val(id, *val);
1112 r = -EINVAL;
1113 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1114 vcpu->arch.dtl.next_gpa))
1115 break;
1116 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1117 break;
1118 case KVM_REG_PPC_VPA_SLB:
1119 addr = val->vpaval.addr;
1120 len = val->vpaval.length;
1121 r = -EINVAL;
1122 if (addr && !vcpu->arch.vpa.next_gpa)
1123 break;
1124 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1125 break;
1126 case KVM_REG_PPC_VPA_DTL:
1127 addr = val->vpaval.addr;
1128 len = val->vpaval.length;
1129 r = -EINVAL;
1130 if (addr && (len < sizeof(struct dtl_entry) ||
1131 !vcpu->arch.vpa.next_gpa))
1132 break;
1133 len -= len % sizeof(struct dtl_entry);
1134 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1135 break;
1136 case KVM_REG_PPC_TB_OFFSET:
1137 /* round up to multiple of 2^24 */
1138 vcpu->arch.vcore->tb_offset =
1139 ALIGN(set_reg_val(id, *val), 1UL << 24);
1140 break;
1141 case KVM_REG_PPC_LPCR:
1142 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1143 break;
1144 case KVM_REG_PPC_PPR:
1145 vcpu->arch.ppr = set_reg_val(id, *val);
1146 break;
1147 case KVM_REG_PPC_ARCH_COMPAT:
1148 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1149 break;
1150 default:
1151 r = -EINVAL;
1152 break;
1153 }
1154
1155 return r;
1156 }
1157
1158 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1159 unsigned int id)
1160 {
1161 struct kvm_vcpu *vcpu;
1162 int err = -EINVAL;
1163 int core;
1164 struct kvmppc_vcore *vcore;
1165
1166 core = id / threads_per_core;
1167 if (core >= KVM_MAX_VCORES)
1168 goto out;
1169
1170 err = -ENOMEM;
1171 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1172 if (!vcpu)
1173 goto out;
1174
1175 err = kvm_vcpu_init(vcpu, kvm, id);
1176 if (err)
1177 goto free_vcpu;
1178
1179 vcpu->arch.shared = &vcpu->arch.shregs;
1180 vcpu->arch.mmcr[0] = MMCR0_FC;
1181 vcpu->arch.ctrl = CTRL_RUNLATCH;
1182 /* default to host PVR, since we can't spoof it */
1183 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1184 spin_lock_init(&vcpu->arch.vpa_update_lock);
1185 spin_lock_init(&vcpu->arch.tbacct_lock);
1186 vcpu->arch.busy_preempt = TB_NIL;
1187 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1188
1189 kvmppc_mmu_book3s_hv_init(vcpu);
1190
1191 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1192
1193 init_waitqueue_head(&vcpu->arch.cpu_run);
1194
1195 mutex_lock(&kvm->lock);
1196 vcore = kvm->arch.vcores[core];
1197 if (!vcore) {
1198 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1199 if (vcore) {
1200 INIT_LIST_HEAD(&vcore->runnable_threads);
1201 spin_lock_init(&vcore->lock);
1202 init_waitqueue_head(&vcore->wq);
1203 vcore->preempt_tb = TB_NIL;
1204 vcore->lpcr = kvm->arch.lpcr;
1205 vcore->first_vcpuid = core * threads_per_core;
1206 vcore->kvm = kvm;
1207 }
1208 kvm->arch.vcores[core] = vcore;
1209 kvm->arch.online_vcores++;
1210 }
1211 mutex_unlock(&kvm->lock);
1212
1213 if (!vcore)
1214 goto free_vcpu;
1215
1216 spin_lock(&vcore->lock);
1217 ++vcore->num_threads;
1218 spin_unlock(&vcore->lock);
1219 vcpu->arch.vcore = vcore;
1220 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1221
1222 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1223 kvmppc_sanity_check(vcpu);
1224
1225 return vcpu;
1226
1227 free_vcpu:
1228 kmem_cache_free(kvm_vcpu_cache, vcpu);
1229 out:
1230 return ERR_PTR(err);
1231 }
1232
1233 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1234 {
1235 if (vpa->pinned_addr)
1236 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1237 vpa->dirty);
1238 }
1239
1240 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1241 {
1242 spin_lock(&vcpu->arch.vpa_update_lock);
1243 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1244 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1245 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1246 spin_unlock(&vcpu->arch.vpa_update_lock);
1247 kvm_vcpu_uninit(vcpu);
1248 kmem_cache_free(kvm_vcpu_cache, vcpu);
1249 }
1250
1251 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1252 {
1253 /* Indicate we want to get back into the guest */
1254 return 1;
1255 }
1256
1257 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1258 {
1259 unsigned long dec_nsec, now;
1260
1261 now = get_tb();
1262 if (now > vcpu->arch.dec_expires) {
1263 /* decrementer has already gone negative */
1264 kvmppc_core_queue_dec(vcpu);
1265 kvmppc_core_prepare_to_enter(vcpu);
1266 return;
1267 }
1268 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1269 / tb_ticks_per_sec;
1270 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1271 HRTIMER_MODE_REL);
1272 vcpu->arch.timer_running = 1;
1273 }
1274
1275 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1276 {
1277 vcpu->arch.ceded = 0;
1278 if (vcpu->arch.timer_running) {
1279 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1280 vcpu->arch.timer_running = 0;
1281 }
1282 }
1283
1284 extern void __kvmppc_vcore_entry(void);
1285
1286 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1287 struct kvm_vcpu *vcpu)
1288 {
1289 u64 now;
1290
1291 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1292 return;
1293 spin_lock_irq(&vcpu->arch.tbacct_lock);
1294 now = mftb();
1295 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1296 vcpu->arch.stolen_logged;
1297 vcpu->arch.busy_preempt = now;
1298 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1299 spin_unlock_irq(&vcpu->arch.tbacct_lock);
1300 --vc->n_runnable;
1301 list_del(&vcpu->arch.run_list);
1302 }
1303
1304 static int kvmppc_grab_hwthread(int cpu)
1305 {
1306 struct paca_struct *tpaca;
1307 long timeout = 1000;
1308
1309 tpaca = &paca[cpu];
1310
1311 /* Ensure the thread won't go into the kernel if it wakes */
1312 tpaca->kvm_hstate.hwthread_req = 1;
1313 tpaca->kvm_hstate.kvm_vcpu = NULL;
1314
1315 /*
1316 * If the thread is already executing in the kernel (e.g. handling
1317 * a stray interrupt), wait for it to get back to nap mode.
1318 * The smp_mb() is to ensure that our setting of hwthread_req
1319 * is visible before we look at hwthread_state, so if this
1320 * races with the code at system_reset_pSeries and the thread
1321 * misses our setting of hwthread_req, we are sure to see its
1322 * setting of hwthread_state, and vice versa.
1323 */
1324 smp_mb();
1325 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1326 if (--timeout <= 0) {
1327 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1328 return -EBUSY;
1329 }
1330 udelay(1);
1331 }
1332 return 0;
1333 }
1334
1335 static void kvmppc_release_hwthread(int cpu)
1336 {
1337 struct paca_struct *tpaca;
1338
1339 tpaca = &paca[cpu];
1340 tpaca->kvm_hstate.hwthread_req = 0;
1341 tpaca->kvm_hstate.kvm_vcpu = NULL;
1342 }
1343
1344 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1345 {
1346 int cpu;
1347 struct paca_struct *tpaca;
1348 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1349
1350 if (vcpu->arch.timer_running) {
1351 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1352 vcpu->arch.timer_running = 0;
1353 }
1354 cpu = vc->pcpu + vcpu->arch.ptid;
1355 tpaca = &paca[cpu];
1356 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1357 tpaca->kvm_hstate.kvm_vcore = vc;
1358 tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1359 vcpu->cpu = vc->pcpu;
1360 smp_wmb();
1361 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1362 if (cpu != smp_processor_id()) {
1363 #ifdef CONFIG_KVM_XICS
1364 xics_wake_cpu(cpu);
1365 #endif
1366 if (vcpu->arch.ptid)
1367 ++vc->n_woken;
1368 }
1369 #endif
1370 }
1371
1372 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1373 {
1374 int i;
1375
1376 HMT_low();
1377 i = 0;
1378 while (vc->nap_count < vc->n_woken) {
1379 if (++i >= 1000000) {
1380 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1381 vc->nap_count, vc->n_woken);
1382 break;
1383 }
1384 cpu_relax();
1385 }
1386 HMT_medium();
1387 }
1388
1389 /*
1390 * Check that we are on thread 0 and that any other threads in
1391 * this core are off-line. Then grab the threads so they can't
1392 * enter the kernel.
1393 */
1394 static int on_primary_thread(void)
1395 {
1396 int cpu = smp_processor_id();
1397 int thr = cpu_thread_in_core(cpu);
1398
1399 if (thr)
1400 return 0;
1401 while (++thr < threads_per_core)
1402 if (cpu_online(cpu + thr))
1403 return 0;
1404
1405 /* Grab all hw threads so they can't go into the kernel */
1406 for (thr = 1; thr < threads_per_core; ++thr) {
1407 if (kvmppc_grab_hwthread(cpu + thr)) {
1408 /* Couldn't grab one; let the others go */
1409 do {
1410 kvmppc_release_hwthread(cpu + thr);
1411 } while (--thr > 0);
1412 return 0;
1413 }
1414 }
1415 return 1;
1416 }
1417
1418 /*
1419 * Run a set of guest threads on a physical core.
1420 * Called with vc->lock held.
1421 */
1422 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1423 {
1424 struct kvm_vcpu *vcpu, *vnext;
1425 long ret;
1426 u64 now;
1427 int i, need_vpa_update;
1428 int srcu_idx;
1429 struct kvm_vcpu *vcpus_to_update[threads_per_core];
1430
1431 /* don't start if any threads have a signal pending */
1432 need_vpa_update = 0;
1433 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1434 if (signal_pending(vcpu->arch.run_task))
1435 return;
1436 if (vcpu->arch.vpa.update_pending ||
1437 vcpu->arch.slb_shadow.update_pending ||
1438 vcpu->arch.dtl.update_pending)
1439 vcpus_to_update[need_vpa_update++] = vcpu;
1440 }
1441
1442 /*
1443 * Initialize *vc, in particular vc->vcore_state, so we can
1444 * drop the vcore lock if necessary.
1445 */
1446 vc->n_woken = 0;
1447 vc->nap_count = 0;
1448 vc->entry_exit_count = 0;
1449 vc->vcore_state = VCORE_STARTING;
1450 vc->in_guest = 0;
1451 vc->napping_threads = 0;
1452
1453 /*
1454 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1455 * which can't be called with any spinlocks held.
1456 */
1457 if (need_vpa_update) {
1458 spin_unlock(&vc->lock);
1459 for (i = 0; i < need_vpa_update; ++i)
1460 kvmppc_update_vpas(vcpus_to_update[i]);
1461 spin_lock(&vc->lock);
1462 }
1463
1464 /*
1465 * Make sure we are running on thread 0, and that
1466 * secondary threads are offline.
1467 */
1468 if (threads_per_core > 1 && !on_primary_thread()) {
1469 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1470 vcpu->arch.ret = -EBUSY;
1471 goto out;
1472 }
1473
1474 vc->pcpu = smp_processor_id();
1475 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1476 kvmppc_start_thread(vcpu);
1477 kvmppc_create_dtl_entry(vcpu, vc);
1478 }
1479
1480 /* Set this explicitly in case thread 0 doesn't have a vcpu */
1481 get_paca()->kvm_hstate.kvm_vcore = vc;
1482 get_paca()->kvm_hstate.ptid = 0;
1483
1484 vc->vcore_state = VCORE_RUNNING;
1485 preempt_disable();
1486 spin_unlock(&vc->lock);
1487
1488 kvm_guest_enter();
1489
1490 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1491
1492 __kvmppc_vcore_entry();
1493
1494 spin_lock(&vc->lock);
1495 /* disable sending of IPIs on virtual external irqs */
1496 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1497 vcpu->cpu = -1;
1498 /* wait for secondary threads to finish writing their state to memory */
1499 if (vc->nap_count < vc->n_woken)
1500 kvmppc_wait_for_nap(vc);
1501 for (i = 0; i < threads_per_core; ++i)
1502 kvmppc_release_hwthread(vc->pcpu + i);
1503 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1504 vc->vcore_state = VCORE_EXITING;
1505 spin_unlock(&vc->lock);
1506
1507 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1508
1509 /* make sure updates to secondary vcpu structs are visible now */
1510 smp_mb();
1511 kvm_guest_exit();
1512
1513 preempt_enable();
1514 cond_resched();
1515
1516 spin_lock(&vc->lock);
1517 now = get_tb();
1518 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1519 /* cancel pending dec exception if dec is positive */
1520 if (now < vcpu->arch.dec_expires &&
1521 kvmppc_core_pending_dec(vcpu))
1522 kvmppc_core_dequeue_dec(vcpu);
1523
1524 ret = RESUME_GUEST;
1525 if (vcpu->arch.trap)
1526 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1527 vcpu->arch.run_task);
1528
1529 vcpu->arch.ret = ret;
1530 vcpu->arch.trap = 0;
1531
1532 if (vcpu->arch.ceded) {
1533 if (ret != RESUME_GUEST)
1534 kvmppc_end_cede(vcpu);
1535 else
1536 kvmppc_set_timer(vcpu);
1537 }
1538 }
1539
1540 out:
1541 vc->vcore_state = VCORE_INACTIVE;
1542 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1543 arch.run_list) {
1544 if (vcpu->arch.ret != RESUME_GUEST) {
1545 kvmppc_remove_runnable(vc, vcpu);
1546 wake_up(&vcpu->arch.cpu_run);
1547 }
1548 }
1549 }
1550
1551 /*
1552 * Wait for some other vcpu thread to execute us, and
1553 * wake us up when we need to handle something in the host.
1554 */
1555 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1556 {
1557 DEFINE_WAIT(wait);
1558
1559 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1560 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1561 schedule();
1562 finish_wait(&vcpu->arch.cpu_run, &wait);
1563 }
1564
1565 /*
1566 * All the vcpus in this vcore are idle, so wait for a decrementer
1567 * or external interrupt to one of the vcpus. vc->lock is held.
1568 */
1569 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1570 {
1571 DEFINE_WAIT(wait);
1572
1573 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1574 vc->vcore_state = VCORE_SLEEPING;
1575 spin_unlock(&vc->lock);
1576 schedule();
1577 finish_wait(&vc->wq, &wait);
1578 spin_lock(&vc->lock);
1579 vc->vcore_state = VCORE_INACTIVE;
1580 }
1581
1582 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1583 {
1584 int n_ceded;
1585 struct kvmppc_vcore *vc;
1586 struct kvm_vcpu *v, *vn;
1587
1588 kvm_run->exit_reason = 0;
1589 vcpu->arch.ret = RESUME_GUEST;
1590 vcpu->arch.trap = 0;
1591 kvmppc_update_vpas(vcpu);
1592
1593 /*
1594 * Synchronize with other threads in this virtual core
1595 */
1596 vc = vcpu->arch.vcore;
1597 spin_lock(&vc->lock);
1598 vcpu->arch.ceded = 0;
1599 vcpu->arch.run_task = current;
1600 vcpu->arch.kvm_run = kvm_run;
1601 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1602 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1603 vcpu->arch.busy_preempt = TB_NIL;
1604 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1605 ++vc->n_runnable;
1606
1607 /*
1608 * This happens the first time this is called for a vcpu.
1609 * If the vcore is already running, we may be able to start
1610 * this thread straight away and have it join in.
1611 */
1612 if (!signal_pending(current)) {
1613 if (vc->vcore_state == VCORE_RUNNING &&
1614 VCORE_EXIT_COUNT(vc) == 0) {
1615 kvmppc_create_dtl_entry(vcpu, vc);
1616 kvmppc_start_thread(vcpu);
1617 } else if (vc->vcore_state == VCORE_SLEEPING) {
1618 wake_up(&vc->wq);
1619 }
1620
1621 }
1622
1623 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1624 !signal_pending(current)) {
1625 if (vc->vcore_state != VCORE_INACTIVE) {
1626 spin_unlock(&vc->lock);
1627 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1628 spin_lock(&vc->lock);
1629 continue;
1630 }
1631 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1632 arch.run_list) {
1633 kvmppc_core_prepare_to_enter(v);
1634 if (signal_pending(v->arch.run_task)) {
1635 kvmppc_remove_runnable(vc, v);
1636 v->stat.signal_exits++;
1637 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1638 v->arch.ret = -EINTR;
1639 wake_up(&v->arch.cpu_run);
1640 }
1641 }
1642 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1643 break;
1644 vc->runner = vcpu;
1645 n_ceded = 0;
1646 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1647 if (!v->arch.pending_exceptions)
1648 n_ceded += v->arch.ceded;
1649 else
1650 v->arch.ceded = 0;
1651 }
1652 if (n_ceded == vc->n_runnable)
1653 kvmppc_vcore_blocked(vc);
1654 else
1655 kvmppc_run_core(vc);
1656 vc->runner = NULL;
1657 }
1658
1659 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1660 (vc->vcore_state == VCORE_RUNNING ||
1661 vc->vcore_state == VCORE_EXITING)) {
1662 spin_unlock(&vc->lock);
1663 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1664 spin_lock(&vc->lock);
1665 }
1666
1667 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1668 kvmppc_remove_runnable(vc, vcpu);
1669 vcpu->stat.signal_exits++;
1670 kvm_run->exit_reason = KVM_EXIT_INTR;
1671 vcpu->arch.ret = -EINTR;
1672 }
1673
1674 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1675 /* Wake up some vcpu to run the core */
1676 v = list_first_entry(&vc->runnable_threads,
1677 struct kvm_vcpu, arch.run_list);
1678 wake_up(&v->arch.cpu_run);
1679 }
1680
1681 spin_unlock(&vc->lock);
1682 return vcpu->arch.ret;
1683 }
1684
1685 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1686 {
1687 int r;
1688 int srcu_idx;
1689
1690 if (!vcpu->arch.sane) {
1691 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1692 return -EINVAL;
1693 }
1694
1695 kvmppc_core_prepare_to_enter(vcpu);
1696
1697 /* No need to go into the guest when all we'll do is come back out */
1698 if (signal_pending(current)) {
1699 run->exit_reason = KVM_EXIT_INTR;
1700 return -EINTR;
1701 }
1702
1703 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1704 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1705 smp_mb();
1706
1707 /* On the first time here, set up HTAB and VRMA or RMA */
1708 if (!vcpu->kvm->arch.rma_setup_done) {
1709 r = kvmppc_hv_setup_htab_rma(vcpu);
1710 if (r)
1711 goto out;
1712 }
1713
1714 flush_fp_to_thread(current);
1715 flush_altivec_to_thread(current);
1716 flush_vsx_to_thread(current);
1717 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1718 vcpu->arch.pgdir = current->mm->pgd;
1719 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1720
1721 do {
1722 r = kvmppc_run_vcpu(run, vcpu);
1723
1724 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1725 !(vcpu->arch.shregs.msr & MSR_PR)) {
1726 r = kvmppc_pseries_do_hcall(vcpu);
1727 kvmppc_core_prepare_to_enter(vcpu);
1728 } else if (r == RESUME_PAGE_FAULT) {
1729 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1730 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1731 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1732 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1733 }
1734 } while (r == RESUME_GUEST);
1735
1736 out:
1737 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1738 atomic_dec(&vcpu->kvm->arch.vcpus_running);
1739 return r;
1740 }
1741
1742
1743 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1744 Assumes POWER7 or PPC970. */
1745 static inline int lpcr_rmls(unsigned long rma_size)
1746 {
1747 switch (rma_size) {
1748 case 32ul << 20: /* 32 MB */
1749 if (cpu_has_feature(CPU_FTR_ARCH_206))
1750 return 8; /* only supported on POWER7 */
1751 return -1;
1752 case 64ul << 20: /* 64 MB */
1753 return 3;
1754 case 128ul << 20: /* 128 MB */
1755 return 7;
1756 case 256ul << 20: /* 256 MB */
1757 return 4;
1758 case 1ul << 30: /* 1 GB */
1759 return 2;
1760 case 16ul << 30: /* 16 GB */
1761 return 1;
1762 case 256ul << 30: /* 256 GB */
1763 return 0;
1764 default:
1765 return -1;
1766 }
1767 }
1768
1769 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1770 {
1771 struct page *page;
1772 struct kvm_rma_info *ri = vma->vm_file->private_data;
1773
1774 if (vmf->pgoff >= kvm_rma_pages)
1775 return VM_FAULT_SIGBUS;
1776
1777 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1778 get_page(page);
1779 vmf->page = page;
1780 return 0;
1781 }
1782
1783 static const struct vm_operations_struct kvm_rma_vm_ops = {
1784 .fault = kvm_rma_fault,
1785 };
1786
1787 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1788 {
1789 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1790 vma->vm_ops = &kvm_rma_vm_ops;
1791 return 0;
1792 }
1793
1794 static int kvm_rma_release(struct inode *inode, struct file *filp)
1795 {
1796 struct kvm_rma_info *ri = filp->private_data;
1797
1798 kvm_release_rma(ri);
1799 return 0;
1800 }
1801
1802 static const struct file_operations kvm_rma_fops = {
1803 .mmap = kvm_rma_mmap,
1804 .release = kvm_rma_release,
1805 };
1806
1807 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1808 struct kvm_allocate_rma *ret)
1809 {
1810 long fd;
1811 struct kvm_rma_info *ri;
1812 /*
1813 * Only do this on PPC970 in HV mode
1814 */
1815 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1816 !cpu_has_feature(CPU_FTR_ARCH_201))
1817 return -EINVAL;
1818
1819 if (!kvm_rma_pages)
1820 return -EINVAL;
1821
1822 ri = kvm_alloc_rma();
1823 if (!ri)
1824 return -ENOMEM;
1825
1826 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1827 if (fd < 0)
1828 kvm_release_rma(ri);
1829
1830 ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1831 return fd;
1832 }
1833
1834 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1835 int linux_psize)
1836 {
1837 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1838
1839 if (!def->shift)
1840 return;
1841 (*sps)->page_shift = def->shift;
1842 (*sps)->slb_enc = def->sllp;
1843 (*sps)->enc[0].page_shift = def->shift;
1844 /*
1845 * Only return base page encoding. We don't want to return
1846 * all the supporting pte_enc, because our H_ENTER doesn't
1847 * support MPSS yet. Once they do, we can start passing all
1848 * support pte_enc here
1849 */
1850 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1851 (*sps)++;
1852 }
1853
1854 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1855 struct kvm_ppc_smmu_info *info)
1856 {
1857 struct kvm_ppc_one_seg_page_size *sps;
1858
1859 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1860 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1861 info->flags |= KVM_PPC_1T_SEGMENTS;
1862 info->slb_size = mmu_slb_size;
1863
1864 /* We only support these sizes for now, and no muti-size segments */
1865 sps = &info->sps[0];
1866 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1867 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1868 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1869
1870 return 0;
1871 }
1872
1873 /*
1874 * Get (and clear) the dirty memory log for a memory slot.
1875 */
1876 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1877 struct kvm_dirty_log *log)
1878 {
1879 struct kvm_memory_slot *memslot;
1880 int r;
1881 unsigned long n;
1882
1883 mutex_lock(&kvm->slots_lock);
1884
1885 r = -EINVAL;
1886 if (log->slot >= KVM_USER_MEM_SLOTS)
1887 goto out;
1888
1889 memslot = id_to_memslot(kvm->memslots, log->slot);
1890 r = -ENOENT;
1891 if (!memslot->dirty_bitmap)
1892 goto out;
1893
1894 n = kvm_dirty_bitmap_bytes(memslot);
1895 memset(memslot->dirty_bitmap, 0, n);
1896
1897 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1898 if (r)
1899 goto out;
1900
1901 r = -EFAULT;
1902 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1903 goto out;
1904
1905 r = 0;
1906 out:
1907 mutex_unlock(&kvm->slots_lock);
1908 return r;
1909 }
1910
1911 static void unpin_slot(struct kvm_memory_slot *memslot)
1912 {
1913 unsigned long *physp;
1914 unsigned long j, npages, pfn;
1915 struct page *page;
1916
1917 physp = memslot->arch.slot_phys;
1918 npages = memslot->npages;
1919 if (!physp)
1920 return;
1921 for (j = 0; j < npages; j++) {
1922 if (!(physp[j] & KVMPPC_GOT_PAGE))
1923 continue;
1924 pfn = physp[j] >> PAGE_SHIFT;
1925 page = pfn_to_page(pfn);
1926 SetPageDirty(page);
1927 put_page(page);
1928 }
1929 }
1930
1931 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
1932 struct kvm_memory_slot *dont)
1933 {
1934 if (!dont || free->arch.rmap != dont->arch.rmap) {
1935 vfree(free->arch.rmap);
1936 free->arch.rmap = NULL;
1937 }
1938 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1939 unpin_slot(free);
1940 vfree(free->arch.slot_phys);
1941 free->arch.slot_phys = NULL;
1942 }
1943 }
1944
1945 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
1946 unsigned long npages)
1947 {
1948 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1949 if (!slot->arch.rmap)
1950 return -ENOMEM;
1951 slot->arch.slot_phys = NULL;
1952
1953 return 0;
1954 }
1955
1956 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
1957 struct kvm_memory_slot *memslot,
1958 struct kvm_userspace_memory_region *mem)
1959 {
1960 unsigned long *phys;
1961
1962 /* Allocate a slot_phys array if needed */
1963 phys = memslot->arch.slot_phys;
1964 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1965 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1966 if (!phys)
1967 return -ENOMEM;
1968 memslot->arch.slot_phys = phys;
1969 }
1970
1971 return 0;
1972 }
1973
1974 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
1975 struct kvm_userspace_memory_region *mem,
1976 const struct kvm_memory_slot *old)
1977 {
1978 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1979 struct kvm_memory_slot *memslot;
1980
1981 if (npages && old->npages) {
1982 /*
1983 * If modifying a memslot, reset all the rmap dirty bits.
1984 * If this is a new memslot, we don't need to do anything
1985 * since the rmap array starts out as all zeroes,
1986 * i.e. no pages are dirty.
1987 */
1988 memslot = id_to_memslot(kvm->memslots, mem->slot);
1989 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1990 }
1991 }
1992
1993 /*
1994 * Update LPCR values in kvm->arch and in vcores.
1995 * Caller must hold kvm->lock.
1996 */
1997 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
1998 {
1999 long int i;
2000 u32 cores_done = 0;
2001
2002 if ((kvm->arch.lpcr & mask) == lpcr)
2003 return;
2004
2005 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2006
2007 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2008 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2009 if (!vc)
2010 continue;
2011 spin_lock(&vc->lock);
2012 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2013 spin_unlock(&vc->lock);
2014 if (++cores_done >= kvm->arch.online_vcores)
2015 break;
2016 }
2017 }
2018
2019 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2020 {
2021 return;
2022 }
2023
2024 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2025 {
2026 int err = 0;
2027 struct kvm *kvm = vcpu->kvm;
2028 struct kvm_rma_info *ri = NULL;
2029 unsigned long hva;
2030 struct kvm_memory_slot *memslot;
2031 struct vm_area_struct *vma;
2032 unsigned long lpcr = 0, senc;
2033 unsigned long lpcr_mask = 0;
2034 unsigned long psize, porder;
2035 unsigned long rma_size;
2036 unsigned long rmls;
2037 unsigned long *physp;
2038 unsigned long i, npages;
2039 int srcu_idx;
2040
2041 mutex_lock(&kvm->lock);
2042 if (kvm->arch.rma_setup_done)
2043 goto out; /* another vcpu beat us to it */
2044
2045 /* Allocate hashed page table (if not done already) and reset it */
2046 if (!kvm->arch.hpt_virt) {
2047 err = kvmppc_alloc_hpt(kvm, NULL);
2048 if (err) {
2049 pr_err("KVM: Couldn't alloc HPT\n");
2050 goto out;
2051 }
2052 }
2053
2054 /* Look up the memslot for guest physical address 0 */
2055 srcu_idx = srcu_read_lock(&kvm->srcu);
2056 memslot = gfn_to_memslot(kvm, 0);
2057
2058 /* We must have some memory at 0 by now */
2059 err = -EINVAL;
2060 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2061 goto out_srcu;
2062
2063 /* Look up the VMA for the start of this memory slot */
2064 hva = memslot->userspace_addr;
2065 down_read(&current->mm->mmap_sem);
2066 vma = find_vma(current->mm, hva);
2067 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2068 goto up_out;
2069
2070 psize = vma_kernel_pagesize(vma);
2071 porder = __ilog2(psize);
2072
2073 /* Is this one of our preallocated RMAs? */
2074 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2075 hva == vma->vm_start)
2076 ri = vma->vm_file->private_data;
2077
2078 up_read(&current->mm->mmap_sem);
2079
2080 if (!ri) {
2081 /* On POWER7, use VRMA; on PPC970, give up */
2082 err = -EPERM;
2083 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2084 pr_err("KVM: CPU requires an RMO\n");
2085 goto out_srcu;
2086 }
2087
2088 /* We can handle 4k, 64k or 16M pages in the VRMA */
2089 err = -EINVAL;
2090 if (!(psize == 0x1000 || psize == 0x10000 ||
2091 psize == 0x1000000))
2092 goto out_srcu;
2093
2094 /* Update VRMASD field in the LPCR */
2095 senc = slb_pgsize_encoding(psize);
2096 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2097 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2098 lpcr_mask = LPCR_VRMASD;
2099 /* the -4 is to account for senc values starting at 0x10 */
2100 lpcr = senc << (LPCR_VRMASD_SH - 4);
2101
2102 /* Create HPTEs in the hash page table for the VRMA */
2103 kvmppc_map_vrma(vcpu, memslot, porder);
2104
2105 } else {
2106 /* Set up to use an RMO region */
2107 rma_size = kvm_rma_pages;
2108 if (rma_size > memslot->npages)
2109 rma_size = memslot->npages;
2110 rma_size <<= PAGE_SHIFT;
2111 rmls = lpcr_rmls(rma_size);
2112 err = -EINVAL;
2113 if ((long)rmls < 0) {
2114 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2115 goto out_srcu;
2116 }
2117 atomic_inc(&ri->use_count);
2118 kvm->arch.rma = ri;
2119
2120 /* Update LPCR and RMOR */
2121 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2122 /* PPC970; insert RMLS value (split field) in HID4 */
2123 lpcr_mask = (1ul << HID4_RMLS0_SH) |
2124 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2125 lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
2126 ((rmls & 3) << HID4_RMLS2_SH);
2127 /* RMOR is also in HID4 */
2128 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2129 << HID4_RMOR_SH;
2130 } else {
2131 /* POWER7 */
2132 lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2133 lpcr = rmls << LPCR_RMLS_SH;
2134 kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
2135 }
2136 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
2137 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
2138
2139 /* Initialize phys addrs of pages in RMO */
2140 npages = kvm_rma_pages;
2141 porder = __ilog2(npages);
2142 physp = memslot->arch.slot_phys;
2143 if (physp) {
2144 if (npages > memslot->npages)
2145 npages = memslot->npages;
2146 spin_lock(&kvm->arch.slot_phys_lock);
2147 for (i = 0; i < npages; ++i)
2148 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2149 porder;
2150 spin_unlock(&kvm->arch.slot_phys_lock);
2151 }
2152 }
2153
2154 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2155
2156 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2157 smp_wmb();
2158 kvm->arch.rma_setup_done = 1;
2159 err = 0;
2160 out_srcu:
2161 srcu_read_unlock(&kvm->srcu, srcu_idx);
2162 out:
2163 mutex_unlock(&kvm->lock);
2164 return err;
2165
2166 up_out:
2167 up_read(&current->mm->mmap_sem);
2168 goto out_srcu;
2169 }
2170
2171 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2172 {
2173 unsigned long lpcr, lpid;
2174
2175 /* Allocate the guest's logical partition ID */
2176
2177 lpid = kvmppc_alloc_lpid();
2178 if ((long)lpid < 0)
2179 return -ENOMEM;
2180 kvm->arch.lpid = lpid;
2181
2182 /*
2183 * Since we don't flush the TLB when tearing down a VM,
2184 * and this lpid might have previously been used,
2185 * make sure we flush on each core before running the new VM.
2186 */
2187 cpumask_setall(&kvm->arch.need_tlb_flush);
2188
2189 kvm->arch.rma = NULL;
2190
2191 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2192
2193 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2194 /* PPC970; HID4 is effectively the LPCR */
2195 kvm->arch.host_lpid = 0;
2196 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2197 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2198 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2199 ((lpid & 0xf) << HID4_LPID5_SH);
2200 } else {
2201 /* POWER7; init LPCR for virtual RMA mode */
2202 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2203 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2204 lpcr &= LPCR_PECE | LPCR_LPES;
2205 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2206 LPCR_VPM0 | LPCR_VPM1;
2207 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2208 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2209 /* On POWER8 turn on online bit to enable PURR/SPURR */
2210 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2211 lpcr |= LPCR_ONL;
2212 }
2213 kvm->arch.lpcr = lpcr;
2214
2215 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2216 spin_lock_init(&kvm->arch.slot_phys_lock);
2217
2218 /*
2219 * Don't allow secondary CPU threads to come online
2220 * while any KVM VMs exist.
2221 */
2222 inhibit_secondary_onlining();
2223
2224 return 0;
2225 }
2226
2227 static void kvmppc_free_vcores(struct kvm *kvm)
2228 {
2229 long int i;
2230
2231 for (i = 0; i < KVM_MAX_VCORES; ++i)
2232 kfree(kvm->arch.vcores[i]);
2233 kvm->arch.online_vcores = 0;
2234 }
2235
2236 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2237 {
2238 uninhibit_secondary_onlining();
2239
2240 kvmppc_free_vcores(kvm);
2241 if (kvm->arch.rma) {
2242 kvm_release_rma(kvm->arch.rma);
2243 kvm->arch.rma = NULL;
2244 }
2245
2246 kvmppc_free_hpt(kvm);
2247 }
2248
2249 /* We don't need to emulate any privileged instructions or dcbz */
2250 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2251 unsigned int inst, int *advance)
2252 {
2253 return EMULATE_FAIL;
2254 }
2255
2256 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2257 ulong spr_val)
2258 {
2259 return EMULATE_FAIL;
2260 }
2261
2262 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2263 ulong *spr_val)
2264 {
2265 return EMULATE_FAIL;
2266 }
2267
2268 static int kvmppc_core_check_processor_compat_hv(void)
2269 {
2270 if (!cpu_has_feature(CPU_FTR_HVMODE))
2271 return -EIO;
2272 return 0;
2273 }
2274
2275 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2276 unsigned int ioctl, unsigned long arg)
2277 {
2278 struct kvm *kvm __maybe_unused = filp->private_data;
2279 void __user *argp = (void __user *)arg;
2280 long r;
2281
2282 switch (ioctl) {
2283
2284 case KVM_ALLOCATE_RMA: {
2285 struct kvm_allocate_rma rma;
2286 struct kvm *kvm = filp->private_data;
2287
2288 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2289 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2290 r = -EFAULT;
2291 break;
2292 }
2293
2294 case KVM_PPC_ALLOCATE_HTAB: {
2295 u32 htab_order;
2296
2297 r = -EFAULT;
2298 if (get_user(htab_order, (u32 __user *)argp))
2299 break;
2300 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2301 if (r)
2302 break;
2303 r = -EFAULT;
2304 if (put_user(htab_order, (u32 __user *)argp))
2305 break;
2306 r = 0;
2307 break;
2308 }
2309
2310 case KVM_PPC_GET_HTAB_FD: {
2311 struct kvm_get_htab_fd ghf;
2312
2313 r = -EFAULT;
2314 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2315 break;
2316 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2317 break;
2318 }
2319
2320 default:
2321 r = -ENOTTY;
2322 }
2323
2324 return r;
2325 }
2326
2327 static struct kvmppc_ops kvm_ops_hv = {
2328 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2329 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2330 .get_one_reg = kvmppc_get_one_reg_hv,
2331 .set_one_reg = kvmppc_set_one_reg_hv,
2332 .vcpu_load = kvmppc_core_vcpu_load_hv,
2333 .vcpu_put = kvmppc_core_vcpu_put_hv,
2334 .set_msr = kvmppc_set_msr_hv,
2335 .vcpu_run = kvmppc_vcpu_run_hv,
2336 .vcpu_create = kvmppc_core_vcpu_create_hv,
2337 .vcpu_free = kvmppc_core_vcpu_free_hv,
2338 .check_requests = kvmppc_core_check_requests_hv,
2339 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
2340 .flush_memslot = kvmppc_core_flush_memslot_hv,
2341 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2342 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
2343 .unmap_hva = kvm_unmap_hva_hv,
2344 .unmap_hva_range = kvm_unmap_hva_range_hv,
2345 .age_hva = kvm_age_hva_hv,
2346 .test_age_hva = kvm_test_age_hva_hv,
2347 .set_spte_hva = kvm_set_spte_hva_hv,
2348 .mmu_destroy = kvmppc_mmu_destroy_hv,
2349 .free_memslot = kvmppc_core_free_memslot_hv,
2350 .create_memslot = kvmppc_core_create_memslot_hv,
2351 .init_vm = kvmppc_core_init_vm_hv,
2352 .destroy_vm = kvmppc_core_destroy_vm_hv,
2353 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2354 .emulate_op = kvmppc_core_emulate_op_hv,
2355 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2356 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2357 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2358 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
2359 };
2360
2361 static int kvmppc_book3s_init_hv(void)
2362 {
2363 int r;
2364 /*
2365 * FIXME!! Do we need to check on all cpus ?
2366 */
2367 r = kvmppc_core_check_processor_compat_hv();
2368 if (r < 0)
2369 return r;
2370
2371 kvm_ops_hv.owner = THIS_MODULE;
2372 kvmppc_hv_ops = &kvm_ops_hv;
2373
2374 r = kvmppc_mmu_hv_init();
2375 return r;
2376 }
2377
2378 static void kvmppc_book3s_exit_hv(void)
2379 {
2380 kvmppc_hv_ops = NULL;
2381 }
2382
2383 module_init(kvmppc_book3s_init_hv);
2384 module_exit(kvmppc_book3s_exit_hv);
2385 MODULE_LICENSE("GPL");
2386 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2387 MODULE_ALIAS("devname:kvm");
Linux with added FPC-III support