projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / powerpc / kvm / book3s_hv.c
CommitLineData
de56a948
PM		 1/*
2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4 *
5 * Authors:
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
9 *
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
12 *
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
15 *
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License, version 2, as
18 * published by the Free Software Foundation.
19 */
20
21#include <linux/kvm_host.h>
22#include <linux/err.h>
23#include <linux/slab.h>
24#include <linux/preempt.h>
25#include <linux/sched.h>
26#include <linux/delay.h>
66b15db6 27#include <linux/export.h>
de56a948
PM		 28#include <linux/fs.h>
29#include <linux/anon_inodes.h>
30#include <linux/cpumask.h>
aa04b4cc
PM		 31#include <linux/spinlock.h>
32#include <linux/page-flags.h>
2c9097e4 33#include <linux/srcu.h>
de56a948
PM		 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>
371fefd6 46#include <asm/cputhreads.h>
aa04b4cc 47#include <asm/page.h>
de1d9248 48#include <asm/hvcall.h>
ae3a197e 49#include <asm/switch_to.h>
512691d4 50#include <asm/smp.h>
de56a948 51#include <linux/gfp.h>
de56a948
PM		 52#include <linux/vmalloc.h>
53#include <linux/highmem.h>
c77162de 54#include <linux/hugetlb.h>
de56a948
PM		 55
56/* #define EXIT_DEBUG */
57/* #define EXIT_DEBUG_SIMPLE */
58/* #define EXIT_DEBUG_INT */
59
913d3ff9
PM		 60/* Used to indicate that a guest page fault needs to be handled */
61#define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
62
c7b67670
PM		 63/* Used as a "null" value for timebase values */
64#define TB_NIL (~(u64)0)
65
19ccb76a 66static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
32fad281 67static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
19ccb76a 68
c7b67670
PM		 69/*
70 * We use the vcpu_load/put functions to measure stolen time.
71 * Stolen time is counted as time when either the vcpu is able to
72 * run as part of a virtual core, but the task running the vcore
73 * is preempted or sleeping, or when the vcpu needs something done
74 * in the kernel by the task running the vcpu, but that task is
75 * preempted or sleeping. Those two things have to be counted
76 * separately, since one of the vcpu tasks will take on the job
77 * of running the core, and the other vcpu tasks in the vcore will
78 * sleep waiting for it to do that, but that sleep shouldn't count
79 * as stolen time.
80 *
81 * Hence we accumulate stolen time when the vcpu can run as part of
82 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
83 * needs its task to do other things in the kernel (for example,
84 * service a page fault) in busy_stolen. We don't accumulate
85 * stolen time for a vcore when it is inactive, or for a vcpu
86 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
87 * a misnomer; it means that the vcpu task is not executing in
88 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
89 * the kernel. We don't have any way of dividing up that time
90 * between time that the vcpu is genuinely stopped, time that
91 * the task is actively working on behalf of the vcpu, and time
92 * that the task is preempted, so we don't count any of it as
93 * stolen.
94 *
95 * Updates to busy_stolen are protected by arch.tbacct_lock;
96 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
97 * of the vcpu that has taken responsibility for running the vcore
98 * (i.e. vc->runner). The stolen times are measured in units of
99 * timebase ticks. (Note that the != TB_NIL checks below are
100 * purely defensive; they should never fail.)
101 */
102
de56a948
PM		 103void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
104{
0456ec4f
PM		 105 struct kvmppc_vcore *vc = vcpu->arch.vcore;
106
c7b67670
PM		 107 spin_lock(&vcpu->arch.tbacct_lock);
108 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
109 vc->preempt_tb != TB_NIL) {
0456ec4f 110 vc->stolen_tb += mftb() - vc->preempt_tb;
c7b67670
PM		 111 vc->preempt_tb = TB_NIL;
112 }
113 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
114 vcpu->arch.busy_preempt != TB_NIL) {
115 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
116 vcpu->arch.busy_preempt = TB_NIL;
117 }
118 spin_unlock(&vcpu->arch.tbacct_lock);
de56a948
PM		 119}
120
121void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
122{
0456ec4f
PM		 123 struct kvmppc_vcore *vc = vcpu->arch.vcore;
124
c7b67670 125 spin_lock(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 126 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
127 vc->preempt_tb = mftb();
c7b67670
PM		 128 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
129 vcpu->arch.busy_preempt = mftb();
130 spin_unlock(&vcpu->arch.tbacct_lock);
de56a948
PM		 131}
132
de56a948
PM		 133void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
134{
135 vcpu->arch.shregs.msr = msr;
19ccb76a 136 kvmppc_end_cede(vcpu);
de56a948
PM		 137}
138
139void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
140{
141 vcpu->arch.pvr = pvr;
142}
143
144void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
145{
146 int r;
147
148 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
149 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
150 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
151 for (r = 0; r < 16; ++r)
152 pr_err("r%2d = %.16lx r%d = %.16lx\n",
153 r, kvmppc_get_gpr(vcpu, r),
154 r+16, kvmppc_get_gpr(vcpu, r+16));
155 pr_err("ctr = %.16lx lr = %.16lx\n",
156 vcpu->arch.ctr, vcpu->arch.lr);
157 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
158 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
159 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
160 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
161 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
162 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
163 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
164 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
165 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
166 pr_err("fault dar = %.16lx dsisr = %.8x\n",
167 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
168 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
169 for (r = 0; r < vcpu->arch.slb_max; ++r)
170 pr_err(" ESID = %.16llx VSID = %.16llx\n",
171 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
172 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
aa04b4cc 173 vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
de56a948
PM		 174 vcpu->arch.last_inst);
175}
176
a8606e20
PM		 177struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
178{
179 int r;
180 struct kvm_vcpu *v, *ret = NULL;
181
182 mutex_lock(&kvm->lock);
183 kvm_for_each_vcpu(r, v, kvm) {
184 if (v->vcpu_id == id) {
185 ret = v;
186 break;
187 }
188 }
189 mutex_unlock(&kvm->lock);
190 return ret;
191}
192
193static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
194{
195 vpa->shared_proc = 1;
196 vpa->yield_count = 1;
197}
198
55b665b0
PM		 199static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
200 unsigned long addr, unsigned long len)
201{
202 /* check address is cacheline aligned */
203 if (addr & (L1_CACHE_BYTES - 1))
204 return -EINVAL;
205 spin_lock(&vcpu->arch.vpa_update_lock);
206 if (v->next_gpa != addr || v->len != len) {
207 v->next_gpa = addr;
208 v->len = addr ? len : 0;
209 v->update_pending = 1;
210 }
211 spin_unlock(&vcpu->arch.vpa_update_lock);
212 return 0;
213}
214
2e25aa5f
PM		 215/* Length for a per-processor buffer is passed in at offset 4 in the buffer */
216struct reg_vpa {
217 u32 dummy;
218 union {
219 u16 hword;
220 u32 word;
221 } length;
222};
223
224static int vpa_is_registered(struct kvmppc_vpa *vpap)
225{
226 if (vpap->update_pending)
227 return vpap->next_gpa != 0;
228 return vpap->pinned_addr != NULL;
229}
230
a8606e20
PM		 231static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
232 unsigned long flags,
233 unsigned long vcpuid, unsigned long vpa)
234{
235 struct kvm *kvm = vcpu->kvm;
93e60249 236 unsigned long len, nb;
a8606e20
PM		 237 void *va;
238 struct kvm_vcpu *tvcpu;
2e25aa5f
PM		 239 int err;
240 int subfunc;
241 struct kvmppc_vpa *vpap;
a8606e20
PM		 242
243 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
244 if (!tvcpu)
245 return H_PARAMETER;
246
2e25aa5f
PM		 247 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
248 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
249 subfunc == H_VPA_REG_SLB) {
250 /* Registering new area - address must be cache-line aligned */
251 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
a8606e20 252 return H_PARAMETER;
2e25aa5f
PM		 253
254 /* convert logical addr to kernel addr and read length */
93e60249
PM		 255 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
256 if (va == NULL)
b2b2f165 257 return H_PARAMETER;
2e25aa5f
PM		 258 if (subfunc == H_VPA_REG_VPA)
259 len = ((struct reg_vpa *)va)->length.hword;
a8606e20 260 else
2e25aa5f
PM		 261 len = ((struct reg_vpa *)va)->length.word;
262 kvmppc_unpin_guest_page(kvm, va);
263
264 /* Check length */
265 if (len > nb || len < sizeof(struct reg_vpa))
266 return H_PARAMETER;
267 } else {
268 vpa = 0;
269 len = 0;
270 }
271
272 err = H_PARAMETER;
273 vpap = NULL;
274 spin_lock(&tvcpu->arch.vpa_update_lock);
275
276 switch (subfunc) {
277 case H_VPA_REG_VPA: /* register VPA */
278 if (len < sizeof(struct lppaca))
a8606e20 279 break;
2e25aa5f
PM		 280 vpap = &tvcpu->arch.vpa;
281 err = 0;
282 break;
283
284 case H_VPA_REG_DTL: /* register DTL */
285 if (len < sizeof(struct dtl_entry))
a8606e20 286 break;
2e25aa5f
PM		 287 len -= len % sizeof(struct dtl_entry);
288
289 /* Check that they have previously registered a VPA */
290 err = H_RESOURCE;
291 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 292 break;
2e25aa5f
PM		 293
294 vpap = &tvcpu->arch.dtl;
295 err = 0;
296 break;
297
298 case H_VPA_REG_SLB: /* register SLB shadow buffer */
299 /* Check that they have previously registered a VPA */
300 err = H_RESOURCE;
301 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 302 break;
2e25aa5f
PM		 303
304 vpap = &tvcpu->arch.slb_shadow;
305 err = 0;
306 break;
307
308 case H_VPA_DEREG_VPA: /* deregister VPA */
309 /* Check they don't still have a DTL or SLB buf registered */
310 err = H_RESOURCE;
311 if (vpa_is_registered(&tvcpu->arch.dtl) ||
312 vpa_is_registered(&tvcpu->arch.slb_shadow))
a8606e20 313 break;
2e25aa5f
PM		 314
315 vpap = &tvcpu->arch.vpa;
316 err = 0;
317 break;
318
319 case H_VPA_DEREG_DTL: /* deregister DTL */
320 vpap = &tvcpu->arch.dtl;
321 err = 0;
322 break;
323
324 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
325 vpap = &tvcpu->arch.slb_shadow;
326 err = 0;
327 break;
328 }
329
330 if (vpap) {
331 vpap->next_gpa = vpa;
332 vpap->len = len;
333 vpap->update_pending = 1;
a8606e20 334 }
93e60249 335
2e25aa5f
PM		 336 spin_unlock(&tvcpu->arch.vpa_update_lock);
337
93e60249 338 return err;
a8606e20
PM		 339}
340
081f323b 341static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
2e25aa5f 342{
081f323b 343 struct kvm *kvm = vcpu->kvm;
2e25aa5f
PM		 344 void *va;
345 unsigned long nb;
081f323b 346 unsigned long gpa;
2e25aa5f 347
081f323b
PM		 348 /*
349 * We need to pin the page pointed to by vpap->next_gpa,
350 * but we can't call kvmppc_pin_guest_page under the lock
351 * as it does get_user_pages() and down_read(). So we
352 * have to drop the lock, pin the page, then get the lock
353 * again and check that a new area didn't get registered
354 * in the meantime.
355 */
356 for (;;) {
357 gpa = vpap->next_gpa;
358 spin_unlock(&vcpu->arch.vpa_update_lock);
359 va = NULL;
360 nb = 0;
361 if (gpa)
362 va = kvmppc_pin_guest_page(kvm, vpap->next_gpa, &nb);
363 spin_lock(&vcpu->arch.vpa_update_lock);
364 if (gpa == vpap->next_gpa)
365 break;
366 /* sigh... unpin that one and try again */
367 if (va)
2e25aa5f 368 kvmppc_unpin_guest_page(kvm, va);
081f323b
PM		 369 }
370
371 vpap->update_pending = 0;
372 if (va && nb < vpap->len) {
373 /*
374 * If it's now too short, it must be that userspace
375 * has changed the mappings underlying guest memory,
376 * so unregister the region.
377 */
378 kvmppc_unpin_guest_page(kvm, va);
379 va = NULL;
2e25aa5f
PM		 380 }
381 if (vpap->pinned_addr)
382 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr);
383 vpap->pinned_addr = va;
384 if (va)
385 vpap->pinned_end = va + vpap->len;
386}
387
388static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
389{
2f12f034
PM		 390 if (!(vcpu->arch.vpa.update_pending ||
391 vcpu->arch.slb_shadow.update_pending ||
392 vcpu->arch.dtl.update_pending))
393 return;
394
2e25aa5f
PM		 395 spin_lock(&vcpu->arch.vpa_update_lock);
396 if (vcpu->arch.vpa.update_pending) {
081f323b 397 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
55b665b0
PM		 398 if (vcpu->arch.vpa.pinned_addr)
399 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
2e25aa5f
PM		 400 }
401 if (vcpu->arch.dtl.update_pending) {
081f323b 402 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
2e25aa5f
PM		 403 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
404 vcpu->arch.dtl_index = 0;
405 }
406 if (vcpu->arch.slb_shadow.update_pending)
081f323b 407 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
2e25aa5f
PM		 408 spin_unlock(&vcpu->arch.vpa_update_lock);
409}
410
c7b67670
PM		 411/*
412 * Return the accumulated stolen time for the vcore up until `now'.
413 * The caller should hold the vcore lock.
414 */
415static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
416{
417 u64 p;
418
419 /*
420 * If we are the task running the vcore, then since we hold
421 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
422 * can't be updated, so we don't need the tbacct_lock.
423 * If the vcore is inactive, it can't become active (since we
424 * hold the vcore lock), so the vcpu load/put functions won't
425 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
426 */
427 if (vc->vcore_state != VCORE_INACTIVE &&
428 vc->runner->arch.run_task != current) {
429 spin_lock(&vc->runner->arch.tbacct_lock);
430 p = vc->stolen_tb;
431 if (vc->preempt_tb != TB_NIL)
432 p += now - vc->preempt_tb;
433 spin_unlock(&vc->runner->arch.tbacct_lock);
434 } else {
435 p = vc->stolen_tb;
436 }
437 return p;
438}
439
0456ec4f
PM		 440static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
441 struct kvmppc_vcore *vc)
442{
443 struct dtl_entry *dt;
444 struct lppaca *vpa;
c7b67670
PM		 445 unsigned long stolen;
446 unsigned long core_stolen;
447 u64 now;
0456ec4f
PM		 448
449 dt = vcpu->arch.dtl_ptr;
450 vpa = vcpu->arch.vpa.pinned_addr;
c7b67670
PM		 451 now = mftb();
452 core_stolen = vcore_stolen_time(vc, now);
453 stolen = core_stolen - vcpu->arch.stolen_logged;
454 vcpu->arch.stolen_logged = core_stolen;
455 spin_lock(&vcpu->arch.tbacct_lock);
456 stolen += vcpu->arch.busy_stolen;
457 vcpu->arch.busy_stolen = 0;
458 spin_unlock(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 459 if (!dt || !vpa)
460 return;
461 memset(dt, 0, sizeof(struct dtl_entry));
462 dt->dispatch_reason = 7;
463 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
c7b67670
PM		 464 dt->timebase = now;
465 dt->enqueue_to_dispatch_time = stolen;
0456ec4f
PM		 466 dt->srr0 = kvmppc_get_pc(vcpu);
467 dt->srr1 = vcpu->arch.shregs.msr;
468 ++dt;
469 if (dt == vcpu->arch.dtl.pinned_end)
470 dt = vcpu->arch.dtl.pinned_addr;
471 vcpu->arch.dtl_ptr = dt;
472 /* order writing *dt vs. writing vpa->dtl_idx */
473 smp_wmb();
474 vpa->dtl_idx = ++vcpu->arch.dtl_index;
475}
476
a8606e20
PM		 477int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
478{
479 unsigned long req = kvmppc_get_gpr(vcpu, 3);
480 unsigned long target, ret = H_SUCCESS;
481 struct kvm_vcpu *tvcpu;
2c9097e4 482 int idx;
a8606e20
PM		 483
484 switch (req) {
c77162de 485 case H_ENTER:
2c9097e4 486 idx = srcu_read_lock(&vcpu->kvm->srcu);
c77162de
PM		 487 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
488 kvmppc_get_gpr(vcpu, 5),
489 kvmppc_get_gpr(vcpu, 6),
490 kvmppc_get_gpr(vcpu, 7));
2c9097e4 491 srcu_read_unlock(&vcpu->kvm->srcu, idx);
c77162de 492 break;
a8606e20 493 case H_CEDE:
a8606e20
PM		 494 break;
495 case H_PROD:
496 target = kvmppc_get_gpr(vcpu, 4);
497 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
498 if (!tvcpu) {
499 ret = H_PARAMETER;
500 break;
501 }
502 tvcpu->arch.prodded = 1;
503 smp_mb();
504 if (vcpu->arch.ceded) {
505 if (waitqueue_active(&vcpu->wq)) {
506 wake_up_interruptible(&vcpu->wq);
507 vcpu->stat.halt_wakeup++;
508 }
509 }
510 break;
511 case H_CONFER:
512 break;
513 case H_REGISTER_VPA:
514 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
515 kvmppc_get_gpr(vcpu, 5),
516 kvmppc_get_gpr(vcpu, 6));
517 break;
518 default:
519 return RESUME_HOST;
520 }
521 kvmppc_set_gpr(vcpu, 3, ret);
522 vcpu->arch.hcall_needed = 0;
523 return RESUME_GUEST;
524}
525
de56a948
PM		 526static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
527 struct task_struct *tsk)
528{
529 int r = RESUME_HOST;
530
531 vcpu->stat.sum_exits++;
532
533 run->exit_reason = KVM_EXIT_UNKNOWN;
534 run->ready_for_interrupt_injection = 1;
535 switch (vcpu->arch.trap) {
536 /* We're good on these - the host merely wanted to get our attention */
537 case BOOK3S_INTERRUPT_HV_DECREMENTER:
538 vcpu->stat.dec_exits++;
539 r = RESUME_GUEST;
540 break;
541 case BOOK3S_INTERRUPT_EXTERNAL:
542 vcpu->stat.ext_intr_exits++;
543 r = RESUME_GUEST;
544 break;
545 case BOOK3S_INTERRUPT_PERFMON:
546 r = RESUME_GUEST;
547 break;
548 case BOOK3S_INTERRUPT_PROGRAM:
549 {
550 ulong flags;
551 /*
552 * Normally program interrupts are delivered directly
553 * to the guest by the hardware, but we can get here
554 * as a result of a hypervisor emulation interrupt
555 * (e40) getting turned into a 700 by BML RTAS.
556 */
557 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
558 kvmppc_core_queue_program(vcpu, flags);
559 r = RESUME_GUEST;
560 break;
561 }
562 case BOOK3S_INTERRUPT_SYSCALL:
563 {
564 /* hcall - punt to userspace */
565 int i;
566
567 if (vcpu->arch.shregs.msr & MSR_PR) {
568 /* sc 1 from userspace - reflect to guest syscall */
569 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
570 r = RESUME_GUEST;
571 break;
572 }
573 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
574 for (i = 0; i < 9; ++i)
575 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
576 run->exit_reason = KVM_EXIT_PAPR_HCALL;
577 vcpu->arch.hcall_needed = 1;
578 r = RESUME_HOST;
579 break;
580 }
581 /*
342d3db7
PM		 582 * We get these next two if the guest accesses a page which it thinks
583 * it has mapped but which is not actually present, either because
584 * it is for an emulated I/O device or because the corresonding
585 * host page has been paged out. Any other HDSI/HISI interrupts
586 * have been handled already.
de56a948
PM		 587 */
588 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
913d3ff9 589 r = RESUME_PAGE_FAULT;
de56a948
PM		 590 break;
591 case BOOK3S_INTERRUPT_H_INST_STORAGE:
913d3ff9
PM		 592 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
593 vcpu->arch.fault_dsisr = 0;
594 r = RESUME_PAGE_FAULT;
de56a948
PM		 595 break;
596 /*
597 * This occurs if the guest executes an illegal instruction.
598 * We just generate a program interrupt to the guest, since
599 * we don't emulate any guest instructions at this stage.
600 */
601 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
602 kvmppc_core_queue_program(vcpu, 0x80000);
603 r = RESUME_GUEST;
604 break;
605 default:
606 kvmppc_dump_regs(vcpu);
607 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
608 vcpu->arch.trap, kvmppc_get_pc(vcpu),
609 vcpu->arch.shregs.msr);
610 r = RESUME_HOST;
611 BUG();
612 break;
613 }
614
de56a948
PM		 615 return r;
616}
617
618int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
619 struct kvm_sregs *sregs)
620{
621 int i;
622
623 sregs->pvr = vcpu->arch.pvr;
624
625 memset(sregs, 0, sizeof(struct kvm_sregs));
626 for (i = 0; i < vcpu->arch.slb_max; i++) {
627 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
628 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
629 }
630
631 return 0;
632}
633
634int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
635 struct kvm_sregs *sregs)
636{
637 int i, j;
638
639 kvmppc_set_pvr(vcpu, sregs->pvr);
640
641 j = 0;
642 for (i = 0; i < vcpu->arch.slb_nr; i++) {
643 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
644 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
645 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
646 ++j;
647 }
648 }
649 vcpu->arch.slb_max = j;
650
651 return 0;
652}
653
a136a8bd 654int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
31f3438e 655{
a136a8bd
PM		 656 int r = 0;
657 long int i;
31f3438e 658
a136a8bd 659 switch (id) {
31f3438e 660 case KVM_REG_PPC_HIOR:
a136a8bd
PM		 661 *val = get_reg_val(id, 0);
662 break;
663 case KVM_REG_PPC_DABR:
664 *val = get_reg_val(id, vcpu->arch.dabr);
665 break;
666 case KVM_REG_PPC_DSCR:
667 *val = get_reg_val(id, vcpu->arch.dscr);
668 break;
669 case KVM_REG_PPC_PURR:
670 *val = get_reg_val(id, vcpu->arch.purr);
671 break;
672 case KVM_REG_PPC_SPURR:
673 *val = get_reg_val(id, vcpu->arch.spurr);
674 break;
675 case KVM_REG_PPC_AMR:
676 *val = get_reg_val(id, vcpu->arch.amr);
677 break;
678 case KVM_REG_PPC_UAMOR:
679 *val = get_reg_val(id, vcpu->arch.uamor);
680 break;
681 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
682 i = id - KVM_REG_PPC_MMCR0;
683 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
684 break;
685 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
686 i = id - KVM_REG_PPC_PMC1;
687 *val = get_reg_val(id, vcpu->arch.pmc[i]);
31f3438e 688 break;
a8bd19ef
PM		 689#ifdef CONFIG_VSX
690 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
691 if (cpu_has_feature(CPU_FTR_VSX)) {
692 /* VSX => FP reg i is stored in arch.vsr[2*i] */
693 long int i = id - KVM_REG_PPC_FPR0;
694 *val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
695 } else {
696 /* let generic code handle it */
697 r = -EINVAL;
698 }
699 break;
700 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
701 if (cpu_has_feature(CPU_FTR_VSX)) {
702 long int i = id - KVM_REG_PPC_VSR0;
703 val->vsxval[0] = vcpu->arch.vsr[2 * i];
704 val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
705 } else {
706 r = -ENXIO;
707 }
708 break;
709#endif /* CONFIG_VSX */
55b665b0
PM		 710 case KVM_REG_PPC_VPA_ADDR:
711 spin_lock(&vcpu->arch.vpa_update_lock);
712 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
713 spin_unlock(&vcpu->arch.vpa_update_lock);
714 break;
715 case KVM_REG_PPC_VPA_SLB:
716 spin_lock(&vcpu->arch.vpa_update_lock);
717 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
718 val->vpaval.length = vcpu->arch.slb_shadow.len;
719 spin_unlock(&vcpu->arch.vpa_update_lock);
720 break;
721 case KVM_REG_PPC_VPA_DTL:
722 spin_lock(&vcpu->arch.vpa_update_lock);
723 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
724 val->vpaval.length = vcpu->arch.dtl.len;
725 spin_unlock(&vcpu->arch.vpa_update_lock);
726 break;
31f3438e 727 default:
a136a8bd 728 r = -EINVAL;
31f3438e
PM		 729 break;
730 }
731
732 return r;
733}
734
a136a8bd 735int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
31f3438e 736{
a136a8bd
PM		 737 int r = 0;
738 long int i;
55b665b0 739 unsigned long addr, len;
31f3438e 740
a136a8bd 741 switch (id) {
31f3438e 742 case KVM_REG_PPC_HIOR:
31f3438e 743 /* Only allow this to be set to zero */
a136a8bd 744 if (set_reg_val(id, *val))
31f3438e
PM		 745 r = -EINVAL;
746 break;
a136a8bd
PM		 747 case KVM_REG_PPC_DABR:
748 vcpu->arch.dabr = set_reg_val(id, *val);
749 break;
750 case KVM_REG_PPC_DSCR:
751 vcpu->arch.dscr = set_reg_val(id, *val);
752 break;
753 case KVM_REG_PPC_PURR:
754 vcpu->arch.purr = set_reg_val(id, *val);
755 break;
756 case KVM_REG_PPC_SPURR:
757 vcpu->arch.spurr = set_reg_val(id, *val);
758 break;
759 case KVM_REG_PPC_AMR:
760 vcpu->arch.amr = set_reg_val(id, *val);
761 break;
762 case KVM_REG_PPC_UAMOR:
763 vcpu->arch.uamor = set_reg_val(id, *val);
764 break;
765 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
766 i = id - KVM_REG_PPC_MMCR0;
767 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
768 break;
769 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
770 i = id - KVM_REG_PPC_PMC1;
771 vcpu->arch.pmc[i] = set_reg_val(id, *val);
772 break;
a8bd19ef
PM		 773#ifdef CONFIG_VSX
774 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
775 if (cpu_has_feature(CPU_FTR_VSX)) {
776 /* VSX => FP reg i is stored in arch.vsr[2*i] */
777 long int i = id - KVM_REG_PPC_FPR0;
778 vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
779 } else {
780 /* let generic code handle it */
781 r = -EINVAL;
782 }
783 break;
784 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
785 if (cpu_has_feature(CPU_FTR_VSX)) {
786 long int i = id - KVM_REG_PPC_VSR0;
787 vcpu->arch.vsr[2 * i] = val->vsxval[0];
788 vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
789 } else {
790 r = -ENXIO;
791 }
792 break;
793#endif /* CONFIG_VSX */
55b665b0
PM		 794 case KVM_REG_PPC_VPA_ADDR:
795 addr = set_reg_val(id, *val);
796 r = -EINVAL;
797 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
798 vcpu->arch.dtl.next_gpa))
799 break;
800 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
801 break;
802 case KVM_REG_PPC_VPA_SLB:
803 addr = val->vpaval.addr;
804 len = val->vpaval.length;
805 r = -EINVAL;
806 if (addr && !vcpu->arch.vpa.next_gpa)
807 break;
808 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
809 break;
810 case KVM_REG_PPC_VPA_DTL:
811 addr = val->vpaval.addr;
812 len = val->vpaval.length;
813 r = -EINVAL;
9f8c8c78
PM		 814 if (addr && (len < sizeof(struct dtl_entry) ||
815 !vcpu->arch.vpa.next_gpa))
55b665b0
PM		 816 break;
817 len -= len % sizeof(struct dtl_entry);
818 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
819 break;
31f3438e 820 default:
a136a8bd 821 r = -EINVAL;
31f3438e
PM		 822 break;
823 }
824
825 return r;
826}
827
de56a948
PM		 828int kvmppc_core_check_processor_compat(void)
829{
9e368f29 830 if (cpu_has_feature(CPU_FTR_HVMODE))
de56a948
PM		 831 return 0;
832 return -EIO;
833}
834
835struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
836{
837 struct kvm_vcpu *vcpu;
371fefd6
PM		 838 int err = -EINVAL;
839 int core;
840 struct kvmppc_vcore *vcore;
de56a948 841
371fefd6
PM		 842 core = id / threads_per_core;
843 if (core >= KVM_MAX_VCORES)
844 goto out;
845
846 err = -ENOMEM;
6b75e6bf 847 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
de56a948
PM		 848 if (!vcpu)
849 goto out;
850
851 err = kvm_vcpu_init(vcpu, kvm, id);
852 if (err)
853 goto free_vcpu;
854
855 vcpu->arch.shared = &vcpu->arch.shregs;
de56a948
PM		 856 vcpu->arch.mmcr[0] = MMCR0_FC;
857 vcpu->arch.ctrl = CTRL_RUNLATCH;
858 /* default to host PVR, since we can't spoof it */
859 vcpu->arch.pvr = mfspr(SPRN_PVR);
860 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
2e25aa5f 861 spin_lock_init(&vcpu->arch.vpa_update_lock);
c7b67670
PM		 862 spin_lock_init(&vcpu->arch.tbacct_lock);
863 vcpu->arch.busy_preempt = TB_NIL;
de56a948 864
de56a948
PM		 865 kvmppc_mmu_book3s_hv_init(vcpu);
866
8455d79e 867 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
371fefd6
PM		 868
869 init_waitqueue_head(&vcpu->arch.cpu_run);
870
871 mutex_lock(&kvm->lock);
872 vcore = kvm->arch.vcores[core];
873 if (!vcore) {
874 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
875 if (vcore) {
876 INIT_LIST_HEAD(&vcore->runnable_threads);
877 spin_lock_init(&vcore->lock);
19ccb76a 878 init_waitqueue_head(&vcore->wq);
c7b67670 879 vcore->preempt_tb = TB_NIL;
371fefd6
PM		 880 }
881 kvm->arch.vcores[core] = vcore;
1b400ba0 882 kvm->arch.online_vcores++;
371fefd6
PM		 883 }
884 mutex_unlock(&kvm->lock);
885
886 if (!vcore)
887 goto free_vcpu;
888
889 spin_lock(&vcore->lock);
890 ++vcore->num_threads;
371fefd6
PM		 891 spin_unlock(&vcore->lock);
892 vcpu->arch.vcore = vcore;
893
af8f38b3
AG		 894 vcpu->arch.cpu_type = KVM_CPU_3S_64;
895 kvmppc_sanity_check(vcpu);
896
de56a948
PM		 897 return vcpu;
898
899free_vcpu:
6b75e6bf 900 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 901out:
902 return ERR_PTR(err);
903}
904
905void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
906{
2e25aa5f
PM		 907 spin_lock(&vcpu->arch.vpa_update_lock);
908 if (vcpu->arch.dtl.pinned_addr)
909 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.dtl.pinned_addr);
910 if (vcpu->arch.slb_shadow.pinned_addr)
911 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.slb_shadow.pinned_addr);
912 if (vcpu->arch.vpa.pinned_addr)
913 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.vpa.pinned_addr);
914 spin_unlock(&vcpu->arch.vpa_update_lock);
de56a948 915 kvm_vcpu_uninit(vcpu);
6b75e6bf 916 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 917}
918
19ccb76a 919static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
371fefd6 920{
19ccb76a 921 unsigned long dec_nsec, now;
371fefd6 922
19ccb76a
PM		 923 now = get_tb();
924 if (now > vcpu->arch.dec_expires) {
925 /* decrementer has already gone negative */
926 kvmppc_core_queue_dec(vcpu);
7e28e60e 927 kvmppc_core_prepare_to_enter(vcpu);
19ccb76a 928 return;
371fefd6 929 }
19ccb76a
PM		 930 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
931 / tb_ticks_per_sec;
932 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
933 HRTIMER_MODE_REL);
934 vcpu->arch.timer_running = 1;
371fefd6
PM		 935}
936
19ccb76a 937static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
371fefd6 938{
19ccb76a
PM		 939 vcpu->arch.ceded = 0;
940 if (vcpu->arch.timer_running) {
941 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
942 vcpu->arch.timer_running = 0;
943 }
371fefd6
PM		 944}
945
de56a948 946extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
371fefd6 947extern void xics_wake_cpu(int cpu);
de56a948 948
371fefd6
PM		 949static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
950 struct kvm_vcpu *vcpu)
de56a948 951{
c7b67670
PM		 952 u64 now;
953
371fefd6
PM		 954 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
955 return;
c7b67670
PM		 956 spin_lock(&vcpu->arch.tbacct_lock);
957 now = mftb();
958 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
959 vcpu->arch.stolen_logged;
960 vcpu->arch.busy_preempt = now;
961 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
962 spin_unlock(&vcpu->arch.tbacct_lock);
371fefd6 963 --vc->n_runnable;
371fefd6
PM		 964 list_del(&vcpu->arch.run_list);
965}
966
f0888f70
PM		 967static int kvmppc_grab_hwthread(int cpu)
968{
969 struct paca_struct *tpaca;
970 long timeout = 1000;
971
972 tpaca = &paca[cpu];
973
974 /* Ensure the thread won't go into the kernel if it wakes */
975 tpaca->kvm_hstate.hwthread_req = 1;
7b444c67 976 tpaca->kvm_hstate.kvm_vcpu = NULL;
f0888f70
PM		 977
978 /*
979 * If the thread is already executing in the kernel (e.g. handling
980 * a stray interrupt), wait for it to get back to nap mode.
981 * The smp_mb() is to ensure that our setting of hwthread_req
982 * is visible before we look at hwthread_state, so if this
983 * races with the code at system_reset_pSeries and the thread
984 * misses our setting of hwthread_req, we are sure to see its
985 * setting of hwthread_state, and vice versa.
986 */
987 smp_mb();
988 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
989 if (--timeout <= 0) {
990 pr_err("KVM: couldn't grab cpu %d\n", cpu);
991 return -EBUSY;
992 }
993 udelay(1);
994 }
995 return 0;
996}
997
998static void kvmppc_release_hwthread(int cpu)
999{
1000 struct paca_struct *tpaca;
1001
1002 tpaca = &paca[cpu];
1003 tpaca->kvm_hstate.hwthread_req = 0;
1004 tpaca->kvm_hstate.kvm_vcpu = NULL;
1005}
1006
371fefd6
PM		 1007static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1008{
1009 int cpu;
1010 struct paca_struct *tpaca;
1011 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1012
19ccb76a
PM		 1013 if (vcpu->arch.timer_running) {
1014 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1015 vcpu->arch.timer_running = 0;
1016 }
371fefd6
PM		 1017 cpu = vc->pcpu + vcpu->arch.ptid;
1018 tpaca = &paca[cpu];
1019 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1020 tpaca->kvm_hstate.kvm_vcore = vc;
19ccb76a
PM		 1021 tpaca->kvm_hstate.napping = 0;
1022 vcpu->cpu = vc->pcpu;
371fefd6 1023 smp_wmb();
251da038 1024#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
371fefd6 1025 if (vcpu->arch.ptid) {
371fefd6
PM		 1026 xics_wake_cpu(cpu);
1027 ++vc->n_woken;
de56a948 1028 }
371fefd6
PM		 1029#endif
1030}
de56a948 1031
371fefd6
PM		 1032static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1033{
1034 int i;
1035
1036 HMT_low();
1037 i = 0;
1038 while (vc->nap_count < vc->n_woken) {
1039 if (++i >= 1000000) {
1040 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1041 vc->nap_count, vc->n_woken);
1042 break;
1043 }
1044 cpu_relax();
1045 }
1046 HMT_medium();
1047}
1048
1049/*
1050 * Check that we are on thread 0 and that any other threads in
7b444c67
PM		 1051 * this core are off-line. Then grab the threads so they can't
1052 * enter the kernel.
371fefd6
PM		 1053 */
1054static int on_primary_thread(void)
1055{
1056 int cpu = smp_processor_id();
1057 int thr = cpu_thread_in_core(cpu);
1058
1059 if (thr)
1060 return 0;
1061 while (++thr < threads_per_core)
1062 if (cpu_online(cpu + thr))
1063 return 0;
7b444c67
PM		 1064
1065 /* Grab all hw threads so they can't go into the kernel */
1066 for (thr = 1; thr < threads_per_core; ++thr) {
1067 if (kvmppc_grab_hwthread(cpu + thr)) {
1068 /* Couldn't grab one; let the others go */
1069 do {
1070 kvmppc_release_hwthread(cpu + thr);
1071 } while (--thr > 0);
1072 return 0;
1073 }
1074 }
371fefd6
PM		 1075 return 1;
1076}
1077
1078/*
1079 * Run a set of guest threads on a physical core.
1080 * Called with vc->lock held.
1081 */
913d3ff9 1082static void kvmppc_run_core(struct kvmppc_vcore *vc)
371fefd6 1083{
19ccb76a 1084 struct kvm_vcpu *vcpu, *vcpu0, *vnext;
371fefd6
PM		 1085 long ret;
1086 u64 now;
081f323b 1087 int ptid, i, need_vpa_update;
2c9097e4 1088 int srcu_idx;
913d3ff9 1089 struct kvm_vcpu *vcpus_to_update[threads_per_core];
371fefd6
PM		 1090
1091 /* don't start if any threads have a signal pending */
081f323b
PM		 1092 need_vpa_update = 0;
1093 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1094 if (signal_pending(vcpu->arch.run_task))
913d3ff9
PM		 1095 return;
1096 if (vcpu->arch.vpa.update_pending ||
1097 vcpu->arch.slb_shadow.update_pending ||
1098 vcpu->arch.dtl.update_pending)
1099 vcpus_to_update[need_vpa_update++] = vcpu;
081f323b
PM		 1100 }
1101
1102 /*
1103 * Initialize *vc, in particular vc->vcore_state, so we can
1104 * drop the vcore lock if necessary.
1105 */
1106 vc->n_woken = 0;
1107 vc->nap_count = 0;
1108 vc->entry_exit_count = 0;
2f12f034 1109 vc->vcore_state = VCORE_STARTING;
081f323b
PM		 1110 vc->in_guest = 0;
1111 vc->napping_threads = 0;
1112
1113 /*
1114 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1115 * which can't be called with any spinlocks held.
1116 */
1117 if (need_vpa_update) {
1118 spin_unlock(&vc->lock);
913d3ff9
PM		 1119 for (i = 0; i < need_vpa_update; ++i)
1120 kvmppc_update_vpas(vcpus_to_update[i]);
081f323b
PM		 1121 spin_lock(&vc->lock);
1122 }
de56a948 1123
19ccb76a
PM		 1124 /*
1125 * Assign physical thread IDs, first to non-ceded vcpus
1126 * and then to ceded ones.
1127 */
1128 ptid = 0;
1129 vcpu0 = NULL;
1130 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1131 if (!vcpu->arch.ceded) {
1132 if (!ptid)
1133 vcpu0 = vcpu;
1134 vcpu->arch.ptid = ptid++;
1135 }
1136 }
c7b67670
PM		 1137 if (!vcpu0)
1138 goto out; /* nothing to run; should never happen */
19ccb76a
PM		 1139 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1140 if (vcpu->arch.ceded)
1141 vcpu->arch.ptid = ptid++;
1142
7b444c67
PM		 1143 /*
1144 * Make sure we are running on thread 0, and that
1145 * secondary threads are offline.
1146 */
1147 if (threads_per_core > 1 && !on_primary_thread()) {
1148 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1149 vcpu->arch.ret = -EBUSY;
1150 goto out;
1151 }
1152
371fefd6 1153 vc->pcpu = smp_processor_id();
2e25aa5f 1154 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1155 kvmppc_start_thread(vcpu);
0456ec4f 1156 kvmppc_create_dtl_entry(vcpu, vc);
2e25aa5f 1157 }
371fefd6 1158
2f12f034 1159 vc->vcore_state = VCORE_RUNNING;
19ccb76a 1160 preempt_disable();
371fefd6 1161 spin_unlock(&vc->lock);
de56a948 1162
371fefd6 1163 kvm_guest_enter();
2c9097e4
PM		 1164
1165 srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1166
19ccb76a 1167 __kvmppc_vcore_entry(NULL, vcpu0);
de56a948 1168
371fefd6 1169 spin_lock(&vc->lock);
19ccb76a
PM		 1170 /* disable sending of IPIs on virtual external irqs */
1171 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1172 vcpu->cpu = -1;
1173 /* wait for secondary threads to finish writing their state to memory */
371fefd6
PM		 1174 if (vc->nap_count < vc->n_woken)
1175 kvmppc_wait_for_nap(vc);
2f12f034
PM		 1176 for (i = 0; i < threads_per_core; ++i)
1177 kvmppc_release_hwthread(vc->pcpu + i);
371fefd6 1178 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
19ccb76a 1179 vc->vcore_state = VCORE_EXITING;
371fefd6
PM		 1180 spin_unlock(&vc->lock);
1181
2c9097e4
PM		 1182 srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1183
371fefd6
PM		 1184 /* make sure updates to secondary vcpu structs are visible now */
1185 smp_mb();
de56a948
PM		 1186 kvm_guest_exit();
1187
1188 preempt_enable();
1189 kvm_resched(vcpu);
1190
913d3ff9 1191 spin_lock(&vc->lock);
de56a948 1192 now = get_tb();
371fefd6
PM		 1193 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1194 /* cancel pending dec exception if dec is positive */
1195 if (now < vcpu->arch.dec_expires &&
1196 kvmppc_core_pending_dec(vcpu))
1197 kvmppc_core_dequeue_dec(vcpu);
19ccb76a
PM		 1198
1199 ret = RESUME_GUEST;
1200 if (vcpu->arch.trap)
1201 ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
1202 vcpu->arch.run_task);
1203
371fefd6
PM		 1204 vcpu->arch.ret = ret;
1205 vcpu->arch.trap = 0;
19ccb76a
PM		 1206
1207 if (vcpu->arch.ceded) {
1208 if (ret != RESUME_GUEST)
1209 kvmppc_end_cede(vcpu);
1210 else
1211 kvmppc_set_timer(vcpu);
1212 }
371fefd6 1213 }
de56a948
PM		 1214
1215 out:
19ccb76a 1216 vc->vcore_state = VCORE_INACTIVE;
371fefd6
PM		 1217 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1218 arch.run_list) {
1219 if (vcpu->arch.ret != RESUME_GUEST) {
1220 kvmppc_remove_runnable(vc, vcpu);
1221 wake_up(&vcpu->arch.cpu_run);
1222 }
1223 }
371fefd6
PM		 1224}
1225
19ccb76a
PM		 1226/*
1227 * Wait for some other vcpu thread to execute us, and
1228 * wake us up when we need to handle something in the host.
1229 */
1230static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
371fefd6 1231{
371fefd6
PM		 1232 DEFINE_WAIT(wait);
1233
19ccb76a
PM		 1234 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1235 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1236 schedule();
1237 finish_wait(&vcpu->arch.cpu_run, &wait);
1238}
1239
1240/*
1241 * All the vcpus in this vcore are idle, so wait for a decrementer
1242 * or external interrupt to one of the vcpus. vc->lock is held.
1243 */
1244static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1245{
1246 DEFINE_WAIT(wait);
19ccb76a
PM		 1247
1248 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1249 vc->vcore_state = VCORE_SLEEPING;
1250 spin_unlock(&vc->lock);
913d3ff9 1251 schedule();
19ccb76a
PM		 1252 finish_wait(&vc->wq, &wait);
1253 spin_lock(&vc->lock);
1254 vc->vcore_state = VCORE_INACTIVE;
1255}
371fefd6 1256
19ccb76a
PM		 1257static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1258{
1259 int n_ceded;
19ccb76a
PM		 1260 struct kvmppc_vcore *vc;
1261 struct kvm_vcpu *v, *vn;
9e368f29 1262
371fefd6
PM		 1263 kvm_run->exit_reason = 0;
1264 vcpu->arch.ret = RESUME_GUEST;
1265 vcpu->arch.trap = 0;
2f12f034 1266 kvmppc_update_vpas(vcpu);
371fefd6 1267
371fefd6
PM		 1268 /*
1269 * Synchronize with other threads in this virtual core
1270 */
1271 vc = vcpu->arch.vcore;
1272 spin_lock(&vc->lock);
19ccb76a 1273 vcpu->arch.ceded = 0;
371fefd6
PM		 1274 vcpu->arch.run_task = current;
1275 vcpu->arch.kvm_run = kvm_run;
c7b67670 1276 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
19ccb76a 1277 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
c7b67670 1278 vcpu->arch.busy_preempt = TB_NIL;
371fefd6
PM		 1279 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1280 ++vc->n_runnable;
1281
19ccb76a
PM		 1282 /*
1283 * This happens the first time this is called for a vcpu.
1284 * If the vcore is already running, we may be able to start
1285 * this thread straight away and have it join in.
1286 */
8455d79e 1287 if (!signal_pending(current)) {
19ccb76a
PM		 1288 if (vc->vcore_state == VCORE_RUNNING &&
1289 VCORE_EXIT_COUNT(vc) == 0) {
1290 vcpu->arch.ptid = vc->n_runnable - 1;
2f12f034 1291 kvmppc_create_dtl_entry(vcpu, vc);
19ccb76a 1292 kvmppc_start_thread(vcpu);
8455d79e
PM		 1293 } else if (vc->vcore_state == VCORE_SLEEPING) {
1294 wake_up(&vc->wq);
371fefd6
PM		 1295 }
1296
8455d79e 1297 }
371fefd6 1298
19ccb76a
PM		 1299 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1300 !signal_pending(current)) {
8455d79e 1301 if (vc->vcore_state != VCORE_INACTIVE) {
19ccb76a
PM		 1302 spin_unlock(&vc->lock);
1303 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1304 spin_lock(&vc->lock);
1305 continue;
1306 }
19ccb76a
PM		 1307 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1308 arch.run_list) {
7e28e60e 1309 kvmppc_core_prepare_to_enter(v);
19ccb76a
PM		 1310 if (signal_pending(v->arch.run_task)) {
1311 kvmppc_remove_runnable(vc, v);
1312 v->stat.signal_exits++;
1313 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1314 v->arch.ret = -EINTR;
1315 wake_up(&v->arch.cpu_run);
1316 }
1317 }
8455d79e
PM		 1318 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1319 break;
1320 vc->runner = vcpu;
1321 n_ceded = 0;
1322 list_for_each_entry(v, &vc->runnable_threads, arch.run_list)
1323 if (!v->arch.pending_exceptions)
1324 n_ceded += v->arch.ceded;
1325 if (n_ceded == vc->n_runnable)
1326 kvmppc_vcore_blocked(vc);
1327 else
1328 kvmppc_run_core(vc);
0456ec4f 1329 vc->runner = NULL;
19ccb76a 1330 }
371fefd6 1331
8455d79e
PM		 1332 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1333 (vc->vcore_state == VCORE_RUNNING ||
1334 vc->vcore_state == VCORE_EXITING)) {
1335 spin_unlock(&vc->lock);
1336 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1337 spin_lock(&vc->lock);
1338 }
1339
1340 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1341 kvmppc_remove_runnable(vc, vcpu);
1342 vcpu->stat.signal_exits++;
1343 kvm_run->exit_reason = KVM_EXIT_INTR;
1344 vcpu->arch.ret = -EINTR;
1345 }
1346
1347 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1348 /* Wake up some vcpu to run the core */
1349 v = list_first_entry(&vc->runnable_threads,
1350 struct kvm_vcpu, arch.run_list);
1351 wake_up(&v->arch.cpu_run);
371fefd6
PM		 1352 }
1353
371fefd6 1354 spin_unlock(&vc->lock);
371fefd6 1355 return vcpu->arch.ret;
de56a948
PM		 1356}
1357
a8606e20
PM		 1358int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1359{
1360 int r;
913d3ff9 1361 int srcu_idx;
a8606e20 1362
af8f38b3
AG		 1363 if (!vcpu->arch.sane) {
1364 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1365 return -EINVAL;
1366 }
1367
25051b5a
SW		 1368 kvmppc_core_prepare_to_enter(vcpu);
1369
19ccb76a
PM		 1370 /* No need to go into the guest when all we'll do is come back out */
1371 if (signal_pending(current)) {
1372 run->exit_reason = KVM_EXIT_INTR;
1373 return -EINTR;
1374 }
1375
32fad281
PM		 1376 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1377 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1378 smp_mb();
1379
1380 /* On the first time here, set up HTAB and VRMA or RMA */
c77162de 1381 if (!vcpu->kvm->arch.rma_setup_done) {
32fad281 1382 r = kvmppc_hv_setup_htab_rma(vcpu);
c77162de 1383 if (r)
32fad281 1384 goto out;
c77162de 1385 }
19ccb76a
PM		 1386
1387 flush_fp_to_thread(current);
1388 flush_altivec_to_thread(current);
1389 flush_vsx_to_thread(current);
1390 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
342d3db7 1391 vcpu->arch.pgdir = current->mm->pgd;
c7b67670 1392 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
19ccb76a 1393
a8606e20
PM		 1394 do {
1395 r = kvmppc_run_vcpu(run, vcpu);
1396
1397 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1398 !(vcpu->arch.shregs.msr & MSR_PR)) {
1399 r = kvmppc_pseries_do_hcall(vcpu);
7e28e60e 1400 kvmppc_core_prepare_to_enter(vcpu);
913d3ff9
PM		 1401 } else if (r == RESUME_PAGE_FAULT) {
1402 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1403 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1404 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1405 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
a8606e20
PM		 1406 }
1407 } while (r == RESUME_GUEST);
32fad281
PM		 1408
1409 out:
c7b67670 1410 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
32fad281 1411 atomic_dec(&vcpu->kvm->arch.vcpus_running);
a8606e20
PM		 1412 return r;
1413}
1414
54738c09 1415
aa04b4cc 1416/* Work out RMLS (real mode limit selector) field value for a given RMA size.
9e368f29 1417 Assumes POWER7 or PPC970. */
aa04b4cc
PM		 1418static inline int lpcr_rmls(unsigned long rma_size)
1419{
1420 switch (rma_size) {
1421 case 32ul << 20: /* 32 MB */
9e368f29
PM		 1422 if (cpu_has_feature(CPU_FTR_ARCH_206))
1423 return 8; /* only supported on POWER7 */
1424 return -1;
aa04b4cc
PM		 1425 case 64ul << 20: /* 64 MB */
1426 return 3;
1427 case 128ul << 20: /* 128 MB */
1428 return 7;
1429 case 256ul << 20: /* 256 MB */
1430 return 4;
1431 case 1ul << 30: /* 1 GB */
1432 return 2;
1433 case 16ul << 30: /* 16 GB */
1434 return 1;
1435 case 256ul << 30: /* 256 GB */
1436 return 0;
1437 default:
1438 return -1;
1439 }
1440}
1441
1442static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1443{
b4e70611 1444 struct kvmppc_linear_info *ri = vma->vm_file->private_data;
aa04b4cc
PM		 1445 struct page *page;
1446
1447 if (vmf->pgoff >= ri->npages)
1448 return VM_FAULT_SIGBUS;
1449
1450 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1451 get_page(page);
1452 vmf->page = page;
1453 return 0;
1454}
1455
1456static const struct vm_operations_struct kvm_rma_vm_ops = {
1457 .fault = kvm_rma_fault,
1458};
1459
1460static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1461{
314e51b9 1462 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
aa04b4cc
PM		 1463 vma->vm_ops = &kvm_rma_vm_ops;
1464 return 0;
1465}
1466
1467static int kvm_rma_release(struct inode *inode, struct file *filp)
1468{
b4e70611 1469 struct kvmppc_linear_info *ri = filp->private_data;
aa04b4cc
PM		 1470
1471 kvm_release_rma(ri);
1472 return 0;
1473}
1474
1475static struct file_operations kvm_rma_fops = {
1476 .mmap = kvm_rma_mmap,
1477 .release = kvm_rma_release,
1478};
1479
1480long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1481{
b4e70611 1482 struct kvmppc_linear_info *ri;
aa04b4cc
PM		 1483 long fd;
1484
1485 ri = kvm_alloc_rma();
1486 if (!ri)
1487 return -ENOMEM;
1488
1489 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1490 if (fd < 0)
1491 kvm_release_rma(ri);
1492
1493 ret->rma_size = ri->npages << PAGE_SHIFT;
1494 return fd;
1495}
1496
5b74716e
BH		 1497static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1498 int linux_psize)
1499{
1500 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1501
1502 if (!def->shift)
1503 return;
1504 (*sps)->page_shift = def->shift;
1505 (*sps)->slb_enc = def->sllp;
1506 (*sps)->enc[0].page_shift = def->shift;
1507 (*sps)->enc[0].pte_enc = def->penc;
1508 (*sps)++;
1509}
1510
1511int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1512{
1513 struct kvm_ppc_one_seg_page_size *sps;
1514
1515 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1516 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1517 info->flags |= KVM_PPC_1T_SEGMENTS;
1518 info->slb_size = mmu_slb_size;
1519
1520 /* We only support these sizes for now, and no muti-size segments */
1521 sps = &info->sps[0];
1522 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1523 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1524 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1525
1526 return 0;
1527}
1528
82ed3616
PM		 1529/*
1530 * Get (and clear) the dirty memory log for a memory slot.
1531 */
1532int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1533{
1534 struct kvm_memory_slot *memslot;
1535 int r;
1536 unsigned long n;
1537
1538 mutex_lock(&kvm->slots_lock);
1539
1540 r = -EINVAL;
1541 if (log->slot >= KVM_MEMORY_SLOTS)
1542 goto out;
1543
1544 memslot = id_to_memslot(kvm->memslots, log->slot);
1545 r = -ENOENT;
1546 if (!memslot->dirty_bitmap)
1547 goto out;
1548
1549 n = kvm_dirty_bitmap_bytes(memslot);
1550 memset(memslot->dirty_bitmap, 0, n);
1551
dfe49dbd 1552 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
82ed3616
PM		 1553 if (r)
1554 goto out;
1555
1556 r = -EFAULT;
1557 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1558 goto out;
1559
1560 r = 0;
1561out:
1562 mutex_unlock(&kvm->slots_lock);
1563 return r;
1564}
1565
a66b48c3 1566static void unpin_slot(struct kvm_memory_slot *memslot)
de56a948 1567{
a66b48c3
PM		 1568 unsigned long *physp;
1569 unsigned long j, npages, pfn;
1570 struct page *page;
aa04b4cc 1571
a66b48c3
PM		 1572 physp = memslot->arch.slot_phys;
1573 npages = memslot->npages;
1574 if (!physp)
1575 return;
1576 for (j = 0; j < npages; j++) {
1577 if (!(physp[j] & KVMPPC_GOT_PAGE))
1578 continue;
1579 pfn = physp[j] >> PAGE_SHIFT;
1580 page = pfn_to_page(pfn);
1581 SetPageDirty(page);
1582 put_page(page);
1583 }
1584}
1585
1586void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1587 struct kvm_memory_slot *dont)
1588{
1589 if (!dont || free->arch.rmap != dont->arch.rmap) {
1590 vfree(free->arch.rmap);
1591 free->arch.rmap = NULL;
b2b2f165 1592 }
a66b48c3
PM		 1593 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1594 unpin_slot(free);
1595 vfree(free->arch.slot_phys);
1596 free->arch.slot_phys = NULL;
1597 }
1598}
1599
1600int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1601 unsigned long npages)
1602{
1603 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1604 if (!slot->arch.rmap)
1605 return -ENOMEM;
1606 slot->arch.slot_phys = NULL;
aa04b4cc 1607
c77162de
PM		 1608 return 0;
1609}
aa04b4cc 1610
a66b48c3
PM		 1611int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1612 struct kvm_memory_slot *memslot,
1613 struct kvm_userspace_memory_region *mem)
c77162de 1614{
a66b48c3 1615 unsigned long *phys;
c77162de 1616
a66b48c3
PM		 1617 /* Allocate a slot_phys array if needed */
1618 phys = memslot->arch.slot_phys;
1619 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1620 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1621 if (!phys)
1622 return -ENOMEM;
1623 memslot->arch.slot_phys = phys;
aa04b4cc 1624 }
a66b48c3
PM		 1625
1626 return 0;
c77162de
PM		 1627}
1628
1629void kvmppc_core_commit_memory_region(struct kvm *kvm,
dfe49dbd
PM		 1630 struct kvm_userspace_memory_region *mem,
1631 struct kvm_memory_slot old)
c77162de 1632{
dfe49dbd
PM		 1633 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1634 struct kvm_memory_slot *memslot;
1635
1636 if (npages && old.npages) {
1637 /*
1638 * If modifying a memslot, reset all the rmap dirty bits.
1639 * If this is a new memslot, we don't need to do anything
1640 * since the rmap array starts out as all zeroes,
1641 * i.e. no pages are dirty.
1642 */
1643 memslot = id_to_memslot(kvm->memslots, mem->slot);
1644 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1645 }
c77162de
PM		 1646}
1647
32fad281 1648static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
c77162de
PM		 1649{
1650 int err = 0;
1651 struct kvm *kvm = vcpu->kvm;
b4e70611 1652 struct kvmppc_linear_info *ri = NULL;
c77162de
PM		 1653 unsigned long hva;
1654 struct kvm_memory_slot *memslot;
1655 struct vm_area_struct *vma;
da9d1d7f 1656 unsigned long lpcr, senc;
c77162de
PM		 1657 unsigned long psize, porder;
1658 unsigned long rma_size;
1659 unsigned long rmls;
1660 unsigned long *physp;
da9d1d7f 1661 unsigned long i, npages;
2c9097e4 1662 int srcu_idx;
c77162de
PM		 1663
1664 mutex_lock(&kvm->lock);
1665 if (kvm->arch.rma_setup_done)
1666 goto out; /* another vcpu beat us to it */
aa04b4cc 1667
32fad281
PM		 1668 /* Allocate hashed page table (if not done already) and reset it */
1669 if (!kvm->arch.hpt_virt) {
1670 err = kvmppc_alloc_hpt(kvm, NULL);
1671 if (err) {
1672 pr_err("KVM: Couldn't alloc HPT\n");
1673 goto out;
1674 }
1675 }
1676
c77162de 1677 /* Look up the memslot for guest physical address 0 */
2c9097e4 1678 srcu_idx = srcu_read_lock(&kvm->srcu);
c77162de 1679 memslot = gfn_to_memslot(kvm, 0);
aa04b4cc 1680
c77162de
PM		 1681 /* We must have some memory at 0 by now */
1682 err = -EINVAL;
1683 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2c9097e4 1684 goto out_srcu;
c77162de
PM		 1685
1686 /* Look up the VMA for the start of this memory slot */
1687 hva = memslot->userspace_addr;
1688 down_read(&current->mm->mmap_sem);
1689 vma = find_vma(current->mm, hva);
1690 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1691 goto up_out;
1692
1693 psize = vma_kernel_pagesize(vma);
da9d1d7f 1694 porder = __ilog2(psize);
c77162de
PM		 1695
1696 /* Is this one of our preallocated RMAs? */
1697 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1698 hva == vma->vm_start)
1699 ri = vma->vm_file->private_data;
1700
1701 up_read(&current->mm->mmap_sem);
1702
1703 if (!ri) {
1704 /* On POWER7, use VRMA; on PPC970, give up */
1705 err = -EPERM;
1706 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1707 pr_err("KVM: CPU requires an RMO\n");
2c9097e4 1708 goto out_srcu;
c77162de
PM		 1709 }
1710
da9d1d7f
PM		 1711 /* We can handle 4k, 64k or 16M pages in the VRMA */
1712 err = -EINVAL;
1713 if (!(psize == 0x1000 || psize == 0x10000 ||
1714 psize == 0x1000000))
2c9097e4 1715 goto out_srcu;
da9d1d7f 1716
c77162de 1717 /* Update VRMASD field in the LPCR */
da9d1d7f 1718 senc = slb_pgsize_encoding(psize);
697d3899
PM		 1719 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1720 (VRMA_VSID << SLB_VSID_SHIFT_1T);
da9d1d7f
PM		 1721 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1722 lpcr |= senc << (LPCR_VRMASD_SH - 4);
c77162de
PM		 1723 kvm->arch.lpcr = lpcr;
1724
1725 /* Create HPTEs in the hash page table for the VRMA */
da9d1d7f 1726 kvmppc_map_vrma(vcpu, memslot, porder);
c77162de
PM		 1727
1728 } else {
1729 /* Set up to use an RMO region */
1730 rma_size = ri->npages;
1731 if (rma_size > memslot->npages)
1732 rma_size = memslot->npages;
1733 rma_size <<= PAGE_SHIFT;
aa04b4cc 1734 rmls = lpcr_rmls(rma_size);
c77162de 1735 err = -EINVAL;
aa04b4cc 1736 if (rmls < 0) {
c77162de 1737 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2c9097e4 1738 goto out_srcu;
aa04b4cc
PM		 1739 }
1740 atomic_inc(&ri->use_count);
1741 kvm->arch.rma = ri;
9e368f29
PM		 1742
1743 /* Update LPCR and RMOR */
1744 lpcr = kvm->arch.lpcr;
1745 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1746 /* PPC970; insert RMLS value (split field) in HID4 */
1747 lpcr &= ~((1ul << HID4_RMLS0_SH) |
1748 (3ul << HID4_RMLS2_SH));
1749 lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1750 ((rmls & 3) << HID4_RMLS2_SH);
1751 /* RMOR is also in HID4 */
1752 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1753 << HID4_RMOR_SH;
1754 } else {
1755 /* POWER7 */
1756 lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1757 lpcr |= rmls << LPCR_RMLS_SH;
1758 kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1759 }
aa04b4cc 1760 kvm->arch.lpcr = lpcr;
c77162de 1761 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
aa04b4cc 1762 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
aa04b4cc 1763
c77162de 1764 /* Initialize phys addrs of pages in RMO */
da9d1d7f
PM		 1765 npages = ri->npages;
1766 porder = __ilog2(npages);
a66b48c3
PM		 1767 physp = memslot->arch.slot_phys;
1768 if (physp) {
1769 if (npages > memslot->npages)
1770 npages = memslot->npages;
1771 spin_lock(&kvm->arch.slot_phys_lock);
1772 for (i = 0; i < npages; ++i)
1773 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1774 porder;
1775 spin_unlock(&kvm->arch.slot_phys_lock);
1776 }
aa04b4cc
PM		 1777 }
1778
c77162de
PM		 1779 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1780 smp_wmb();
1781 kvm->arch.rma_setup_done = 1;
1782 err = 0;
2c9097e4
PM		 1783 out_srcu:
1784 srcu_read_unlock(&kvm->srcu, srcu_idx);
c77162de
PM		 1785 out:
1786 mutex_unlock(&kvm->lock);
1787 return err;
b2b2f165 1788
c77162de
PM		 1789 up_out:
1790 up_read(&current->mm->mmap_sem);
1791 goto out;
de56a948
PM		 1792}
1793
1794int kvmppc_core_init_vm(struct kvm *kvm)
1795{
32fad281 1796 unsigned long lpcr, lpid;
de56a948 1797
32fad281
PM		 1798 /* Allocate the guest's logical partition ID */
1799
1800 lpid = kvmppc_alloc_lpid();
1801 if (lpid < 0)
1802 return -ENOMEM;
1803 kvm->arch.lpid = lpid;
de56a948 1804
1b400ba0
PM		 1805 /*
1806 * Since we don't flush the TLB when tearing down a VM,
1807 * and this lpid might have previously been used,
1808 * make sure we flush on each core before running the new VM.
1809 */
1810 cpumask_setall(&kvm->arch.need_tlb_flush);
1811
54738c09 1812 INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
aa04b4cc 1813
aa04b4cc 1814 kvm->arch.rma = NULL;
aa04b4cc 1815
9e368f29 1816 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
aa04b4cc 1817
9e368f29
PM		 1818 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1819 /* PPC970; HID4 is effectively the LPCR */
9e368f29
PM		 1820 kvm->arch.host_lpid = 0;
1821 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1822 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1823 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1824 ((lpid & 0xf) << HID4_LPID5_SH);
1825 } else {
1826 /* POWER7; init LPCR for virtual RMA mode */
1827 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1828 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1829 lpcr &= LPCR_PECE | LPCR_LPES;
1830 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
697d3899
PM		 1831 LPCR_VPM0 | LPCR_VPM1;
1832 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1833 (VRMA_VSID << SLB_VSID_SHIFT_1T);
9e368f29
PM		 1834 }
1835 kvm->arch.lpcr = lpcr;
aa04b4cc 1836
342d3db7 1837 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
c77162de 1838 spin_lock_init(&kvm->arch.slot_phys_lock);
512691d4
PM		 1839
1840 /*
1841 * Don't allow secondary CPU threads to come online
1842 * while any KVM VMs exist.
1843 */
1844 inhibit_secondary_onlining();
1845
54738c09 1846 return 0;
de56a948
PM		 1847}
1848
1849void kvmppc_core_destroy_vm(struct kvm *kvm)
1850{
512691d4
PM		 1851 uninhibit_secondary_onlining();
1852
aa04b4cc
PM		 1853 if (kvm->arch.rma) {
1854 kvm_release_rma(kvm->arch.rma);
1855 kvm->arch.rma = NULL;
1856 }
1857
de56a948 1858 kvmppc_free_hpt(kvm);
54738c09 1859 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
de56a948
PM		 1860}
1861
1862/* These are stubs for now */
1863void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1864{
1865}
1866
1867/* We don't need to emulate any privileged instructions or dcbz */
1868int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1869 unsigned int inst, int *advance)
1870{
1871 return EMULATE_FAIL;
1872}
1873
54771e62 1874int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
de56a948
PM		 1875{
1876 return EMULATE_FAIL;
1877}
1878
54771e62 1879int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
de56a948
PM		 1880{
1881 return EMULATE_FAIL;
1882}
1883
1884static int kvmppc_book3s_hv_init(void)
1885{
1886 int r;
1887
1888 r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1889
1890 if (r)
1891 return r;
1892
1893 r = kvmppc_mmu_hv_init();
1894
1895 return r;
1896}
1897
1898static void kvmppc_book3s_hv_exit(void)
1899{
1900 kvm_exit();
1901}
1902
1903module_init(kvmppc_book3s_hv_init);
1904module_exit(kvmppc_book3s_hv_exit);
kernel source for telekom puls (alcatel/mediatek)