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/cpu.h>
31 #include <linux/cpumask.h>
32 #include <linux/spinlock.h>
33 #include <linux/page-flags.h>
34 #include <linux/srcu.h>
35 #include <linux/miscdevice.h>
36 #include <linux/debugfs.h>
37
38 #include <asm/reg.h>
39 #include <asm/cputable.h>
40 #include <asm/cacheflush.h>
41 #include <asm/tlbflush.h>
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/kvm_ppc.h>
45 #include <asm/kvm_book3s.h>
46 #include <asm/mmu_context.h>
47 #include <asm/lppaca.h>
48 #include <asm/processor.h>
49 #include <asm/cputhreads.h>
50 #include <asm/page.h>
51 #include <asm/hvcall.h>
52 #include <asm/switch_to.h>
53 #include <asm/smp.h>
54 #include <asm/dbell.h>
55 #include <asm/hmi.h>
56 #include <linux/gfp.h>
57 #include <linux/vmalloc.h>
58 #include <linux/highmem.h>
59 #include <linux/hugetlb.h>
60 #include <linux/module.h>
61
62 #include "book3s.h"
63
64 #define CREATE_TRACE_POINTS
65 #include "trace_hv.h"
66
67 /* #define EXIT_DEBUG */
68 /* #define EXIT_DEBUG_SIMPLE */
69 /* #define EXIT_DEBUG_INT */
70
71 /* Used to indicate that a guest page fault needs to be handled */
72 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
73
74 /* Used as a "null" value for timebase values */
75 #define TB_NIL (~(u64)0)
76
77 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
78
79 static int dynamic_mt_modes = 6;
80 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
81 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
82 static int target_smt_mode;
83 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
84 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
85
86 #ifdef CONFIG_KVM_XICS
87 static struct kernel_param_ops module_param_ops = {
88 .set = param_set_int,
89 .get = param_get_int,
90 };
91
92 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
93 S_IRUGO | S_IWUSR);
94 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
95 #endif
96
97 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
98 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
99
100 static bool kvmppc_ipi_thread(int cpu)
101 {
102 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
103 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
104 preempt_disable();
105 if (cpu_first_thread_sibling(cpu) ==
106 cpu_first_thread_sibling(smp_processor_id())) {
107 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
108 msg |= cpu_thread_in_core(cpu);
109 smp_mb();
110 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
111 preempt_enable();
112 return true;
113 }
114 preempt_enable();
115 }
116
117 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
118 if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
119 xics_wake_cpu(cpu);
120 return true;
121 }
122 #endif
123
124 return false;
125 }
126
127 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
128 {
129 int cpu;
130 struct swait_queue_head *wqp;
131
132 wqp = kvm_arch_vcpu_wq(vcpu);
133 if (swait_active(wqp)) {
134 swake_up(wqp);
135 ++vcpu->stat.halt_wakeup;
136 }
137
138 if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
139 return;
140
141 /* CPU points to the first thread of the core */
142 cpu = vcpu->cpu;
143 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
144 smp_send_reschedule(cpu);
145 }
146
147 /*
148 * We use the vcpu_load/put functions to measure stolen time.
149 * Stolen time is counted as time when either the vcpu is able to
150 * run as part of a virtual core, but the task running the vcore
151 * is preempted or sleeping, or when the vcpu needs something done
152 * in the kernel by the task running the vcpu, but that task is
153 * preempted or sleeping. Those two things have to be counted
154 * separately, since one of the vcpu tasks will take on the job
155 * of running the core, and the other vcpu tasks in the vcore will
156 * sleep waiting for it to do that, but that sleep shouldn't count
157 * as stolen time.
158 *
159 * Hence we accumulate stolen time when the vcpu can run as part of
160 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
161 * needs its task to do other things in the kernel (for example,
162 * service a page fault) in busy_stolen. We don't accumulate
163 * stolen time for a vcore when it is inactive, or for a vcpu
164 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
165 * a misnomer; it means that the vcpu task is not executing in
166 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
167 * the kernel. We don't have any way of dividing up that time
168 * between time that the vcpu is genuinely stopped, time that
169 * the task is actively working on behalf of the vcpu, and time
170 * that the task is preempted, so we don't count any of it as
171 * stolen.
172 *
173 * Updates to busy_stolen are protected by arch.tbacct_lock;
174 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
175 * lock. The stolen times are measured in units of timebase ticks.
176 * (Note that the != TB_NIL checks below are purely defensive;
177 * they should never fail.)
178 */
179
180 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
181 {
182 unsigned long flags;
183
184 spin_lock_irqsave(&vc->stoltb_lock, flags);
185 vc->preempt_tb = mftb();
186 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
187 }
188
189 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
190 {
191 unsigned long flags;
192
193 spin_lock_irqsave(&vc->stoltb_lock, flags);
194 if (vc->preempt_tb != TB_NIL) {
195 vc->stolen_tb += mftb() - vc->preempt_tb;
196 vc->preempt_tb = TB_NIL;
197 }
198 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
199 }
200
201 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
202 {
203 struct kvmppc_vcore *vc = vcpu->arch.vcore;
204 unsigned long flags;
205
206 /*
207 * We can test vc->runner without taking the vcore lock,
208 * because only this task ever sets vc->runner to this
209 * vcpu, and once it is set to this vcpu, only this task
210 * ever sets it to NULL.
211 */
212 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
213 kvmppc_core_end_stolen(vc);
214
215 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
216 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
217 vcpu->arch.busy_preempt != TB_NIL) {
218 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
219 vcpu->arch.busy_preempt = TB_NIL;
220 }
221 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
222 }
223
224 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
225 {
226 struct kvmppc_vcore *vc = vcpu->arch.vcore;
227 unsigned long flags;
228
229 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
230 kvmppc_core_start_stolen(vc);
231
232 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
233 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
234 vcpu->arch.busy_preempt = mftb();
235 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
236 }
237
238 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
239 {
240 /*
241 * Check for illegal transactional state bit combination
242 * and if we find it, force the TS field to a safe state.
243 */
244 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
245 msr &= ~MSR_TS_MASK;
246 vcpu->arch.shregs.msr = msr;
247 kvmppc_end_cede(vcpu);
248 }
249
250 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
251 {
252 vcpu->arch.pvr = pvr;
253 }
254
255 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
256 {
257 unsigned long pcr = 0;
258 struct kvmppc_vcore *vc = vcpu->arch.vcore;
259
260 if (arch_compat) {
261 switch (arch_compat) {
262 case PVR_ARCH_205:
263 /*
264 * If an arch bit is set in PCR, all the defined
265 * higher-order arch bits also have to be set.
266 */
267 pcr = PCR_ARCH_206 | PCR_ARCH_205;
268 break;
269 case PVR_ARCH_206:
270 case PVR_ARCH_206p:
271 pcr = PCR_ARCH_206;
272 break;
273 case PVR_ARCH_207:
274 break;
275 default:
276 return -EINVAL;
277 }
278
279 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
280 /* POWER7 can't emulate POWER8 */
281 if (!(pcr & PCR_ARCH_206))
282 return -EINVAL;
283 pcr &= ~PCR_ARCH_206;
284 }
285 }
286
287 spin_lock(&vc->lock);
288 vc->arch_compat = arch_compat;
289 vc->pcr = pcr;
290 spin_unlock(&vc->lock);
291
292 return 0;
293 }
294
295 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
296 {
297 int r;
298
299 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
300 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
301 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
302 for (r = 0; r < 16; ++r)
303 pr_err("r%2d = %.16lx r%d = %.16lx\n",
304 r, kvmppc_get_gpr(vcpu, r),
305 r+16, kvmppc_get_gpr(vcpu, r+16));
306 pr_err("ctr = %.16lx lr = %.16lx\n",
307 vcpu->arch.ctr, vcpu->arch.lr);
308 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
309 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
310 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
311 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
312 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
313 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
314 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
315 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
316 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
317 pr_err("fault dar = %.16lx dsisr = %.8x\n",
318 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
319 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
320 for (r = 0; r < vcpu->arch.slb_max; ++r)
321 pr_err(" ESID = %.16llx VSID = %.16llx\n",
322 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
323 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
324 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
325 vcpu->arch.last_inst);
326 }
327
328 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
329 {
330 struct kvm_vcpu *ret;
331
332 mutex_lock(&kvm->lock);
333 ret = kvm_get_vcpu_by_id(kvm, id);
334 mutex_unlock(&kvm->lock);
335 return ret;
336 }
337
338 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
339 {
340 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
341 vpa->yield_count = cpu_to_be32(1);
342 }
343
344 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
345 unsigned long addr, unsigned long len)
346 {
347 /* check address is cacheline aligned */
348 if (addr & (L1_CACHE_BYTES - 1))
349 return -EINVAL;
350 spin_lock(&vcpu->arch.vpa_update_lock);
351 if (v->next_gpa != addr || v->len != len) {
352 v->next_gpa = addr;
353 v->len = addr ? len : 0;
354 v->update_pending = 1;
355 }
356 spin_unlock(&vcpu->arch.vpa_update_lock);
357 return 0;
358 }
359
360 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
361 struct reg_vpa {
362 u32 dummy;
363 union {
364 __be16 hword;
365 __be32 word;
366 } length;
367 };
368
369 static int vpa_is_registered(struct kvmppc_vpa *vpap)
370 {
371 if (vpap->update_pending)
372 return vpap->next_gpa != 0;
373 return vpap->pinned_addr != NULL;
374 }
375
376 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
377 unsigned long flags,
378 unsigned long vcpuid, unsigned long vpa)
379 {
380 struct kvm *kvm = vcpu->kvm;
381 unsigned long len, nb;
382 void *va;
383 struct kvm_vcpu *tvcpu;
384 int err;
385 int subfunc;
386 struct kvmppc_vpa *vpap;
387
388 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
389 if (!tvcpu)
390 return H_PARAMETER;
391
392 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
393 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
394 subfunc == H_VPA_REG_SLB) {
395 /* Registering new area - address must be cache-line aligned */
396 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
397 return H_PARAMETER;
398
399 /* convert logical addr to kernel addr and read length */
400 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
401 if (va == NULL)
402 return H_PARAMETER;
403 if (subfunc == H_VPA_REG_VPA)
404 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
405 else
406 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
407 kvmppc_unpin_guest_page(kvm, va, vpa, false);
408
409 /* Check length */
410 if (len > nb || len < sizeof(struct reg_vpa))
411 return H_PARAMETER;
412 } else {
413 vpa = 0;
414 len = 0;
415 }
416
417 err = H_PARAMETER;
418 vpap = NULL;
419 spin_lock(&tvcpu->arch.vpa_update_lock);
420
421 switch (subfunc) {
422 case H_VPA_REG_VPA: /* register VPA */
423 if (len < sizeof(struct lppaca))
424 break;
425 vpap = &tvcpu->arch.vpa;
426 err = 0;
427 break;
428
429 case H_VPA_REG_DTL: /* register DTL */
430 if (len < sizeof(struct dtl_entry))
431 break;
432 len -= len % sizeof(struct dtl_entry);
433
434 /* Check that they have previously registered a VPA */
435 err = H_RESOURCE;
436 if (!vpa_is_registered(&tvcpu->arch.vpa))
437 break;
438
439 vpap = &tvcpu->arch.dtl;
440 err = 0;
441 break;
442
443 case H_VPA_REG_SLB: /* register SLB shadow buffer */
444 /* Check that they have previously registered a VPA */
445 err = H_RESOURCE;
446 if (!vpa_is_registered(&tvcpu->arch.vpa))
447 break;
448
449 vpap = &tvcpu->arch.slb_shadow;
450 err = 0;
451 break;
452
453 case H_VPA_DEREG_VPA: /* deregister VPA */
454 /* Check they don't still have a DTL or SLB buf registered */
455 err = H_RESOURCE;
456 if (vpa_is_registered(&tvcpu->arch.dtl) ||
457 vpa_is_registered(&tvcpu->arch.slb_shadow))
458 break;
459
460 vpap = &tvcpu->arch.vpa;
461 err = 0;
462 break;
463
464 case H_VPA_DEREG_DTL: /* deregister DTL */
465 vpap = &tvcpu->arch.dtl;
466 err = 0;
467 break;
468
469 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
470 vpap = &tvcpu->arch.slb_shadow;
471 err = 0;
472 break;
473 }
474
475 if (vpap) {
476 vpap->next_gpa = vpa;
477 vpap->len = len;
478 vpap->update_pending = 1;
479 }
480
481 spin_unlock(&tvcpu->arch.vpa_update_lock);
482
483 return err;
484 }
485
486 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
487 {
488 struct kvm *kvm = vcpu->kvm;
489 void *va;
490 unsigned long nb;
491 unsigned long gpa;
492
493 /*
494 * We need to pin the page pointed to by vpap->next_gpa,
495 * but we can't call kvmppc_pin_guest_page under the lock
496 * as it does get_user_pages() and down_read(). So we
497 * have to drop the lock, pin the page, then get the lock
498 * again and check that a new area didn't get registered
499 * in the meantime.
500 */
501 for (;;) {
502 gpa = vpap->next_gpa;
503 spin_unlock(&vcpu->arch.vpa_update_lock);
504 va = NULL;
505 nb = 0;
506 if (gpa)
507 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
508 spin_lock(&vcpu->arch.vpa_update_lock);
509 if (gpa == vpap->next_gpa)
510 break;
511 /* sigh... unpin that one and try again */
512 if (va)
513 kvmppc_unpin_guest_page(kvm, va, gpa, false);
514 }
515
516 vpap->update_pending = 0;
517 if (va && nb < vpap->len) {
518 /*
519 * If it's now too short, it must be that userspace
520 * has changed the mappings underlying guest memory,
521 * so unregister the region.
522 */
523 kvmppc_unpin_guest_page(kvm, va, gpa, false);
524 va = NULL;
525 }
526 if (vpap->pinned_addr)
527 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
528 vpap->dirty);
529 vpap->gpa = gpa;
530 vpap->pinned_addr = va;
531 vpap->dirty = false;
532 if (va)
533 vpap->pinned_end = va + vpap->len;
534 }
535
536 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
537 {
538 if (!(vcpu->arch.vpa.update_pending ||
539 vcpu->arch.slb_shadow.update_pending ||
540 vcpu->arch.dtl.update_pending))
541 return;
542
543 spin_lock(&vcpu->arch.vpa_update_lock);
544 if (vcpu->arch.vpa.update_pending) {
545 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
546 if (vcpu->arch.vpa.pinned_addr)
547 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
548 }
549 if (vcpu->arch.dtl.update_pending) {
550 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
551 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
552 vcpu->arch.dtl_index = 0;
553 }
554 if (vcpu->arch.slb_shadow.update_pending)
555 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
556 spin_unlock(&vcpu->arch.vpa_update_lock);
557 }
558
559 /*
560 * Return the accumulated stolen time for the vcore up until `now'.
561 * The caller should hold the vcore lock.
562 */
563 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
564 {
565 u64 p;
566 unsigned long flags;
567
568 spin_lock_irqsave(&vc->stoltb_lock, flags);
569 p = vc->stolen_tb;
570 if (vc->vcore_state != VCORE_INACTIVE &&
571 vc->preempt_tb != TB_NIL)
572 p += now - vc->preempt_tb;
573 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
574 return p;
575 }
576
577 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
578 struct kvmppc_vcore *vc)
579 {
580 struct dtl_entry *dt;
581 struct lppaca *vpa;
582 unsigned long stolen;
583 unsigned long core_stolen;
584 u64 now;
585
586 dt = vcpu->arch.dtl_ptr;
587 vpa = vcpu->arch.vpa.pinned_addr;
588 now = mftb();
589 core_stolen = vcore_stolen_time(vc, now);
590 stolen = core_stolen - vcpu->arch.stolen_logged;
591 vcpu->arch.stolen_logged = core_stolen;
592 spin_lock_irq(&vcpu->arch.tbacct_lock);
593 stolen += vcpu->arch.busy_stolen;
594 vcpu->arch.busy_stolen = 0;
595 spin_unlock_irq(&vcpu->arch.tbacct_lock);
596 if (!dt || !vpa)
597 return;
598 memset(dt, 0, sizeof(struct dtl_entry));
599 dt->dispatch_reason = 7;
600 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
601 dt->timebase = cpu_to_be64(now + vc->tb_offset);
602 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
603 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
604 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
605 ++dt;
606 if (dt == vcpu->arch.dtl.pinned_end)
607 dt = vcpu->arch.dtl.pinned_addr;
608 vcpu->arch.dtl_ptr = dt;
609 /* order writing *dt vs. writing vpa->dtl_idx */
610 smp_wmb();
611 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
612 vcpu->arch.dtl.dirty = true;
613 }
614
615 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
616 {
617 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
618 return true;
619 if ((!vcpu->arch.vcore->arch_compat) &&
620 cpu_has_feature(CPU_FTR_ARCH_207S))
621 return true;
622 return false;
623 }
624
625 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
626 unsigned long resource, unsigned long value1,
627 unsigned long value2)
628 {
629 switch (resource) {
630 case H_SET_MODE_RESOURCE_SET_CIABR:
631 if (!kvmppc_power8_compatible(vcpu))
632 return H_P2;
633 if (value2)
634 return H_P4;
635 if (mflags)
636 return H_UNSUPPORTED_FLAG_START;
637 /* Guests can't breakpoint the hypervisor */
638 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
639 return H_P3;
640 vcpu->arch.ciabr = value1;
641 return H_SUCCESS;
642 case H_SET_MODE_RESOURCE_SET_DAWR:
643 if (!kvmppc_power8_compatible(vcpu))
644 return H_P2;
645 if (mflags)
646 return H_UNSUPPORTED_FLAG_START;
647 if (value2 & DABRX_HYP)
648 return H_P4;
649 vcpu->arch.dawr = value1;
650 vcpu->arch.dawrx = value2;
651 return H_SUCCESS;
652 default:
653 return H_TOO_HARD;
654 }
655 }
656
657 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
658 {
659 struct kvmppc_vcore *vcore = target->arch.vcore;
660
661 /*
662 * We expect to have been called by the real mode handler
663 * (kvmppc_rm_h_confer()) which would have directly returned
664 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
665 * have useful work to do and should not confer) so we don't
666 * recheck that here.
667 */
668
669 spin_lock(&vcore->lock);
670 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
671 vcore->vcore_state != VCORE_INACTIVE &&
672 vcore->runner)
673 target = vcore->runner;
674 spin_unlock(&vcore->lock);
675
676 return kvm_vcpu_yield_to(target);
677 }
678
679 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
680 {
681 int yield_count = 0;
682 struct lppaca *lppaca;
683
684 spin_lock(&vcpu->arch.vpa_update_lock);
685 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
686 if (lppaca)
687 yield_count = be32_to_cpu(lppaca->yield_count);
688 spin_unlock(&vcpu->arch.vpa_update_lock);
689 return yield_count;
690 }
691
692 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
693 {
694 unsigned long req = kvmppc_get_gpr(vcpu, 3);
695 unsigned long target, ret = H_SUCCESS;
696 int yield_count;
697 struct kvm_vcpu *tvcpu;
698 int idx, rc;
699
700 if (req <= MAX_HCALL_OPCODE &&
701 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
702 return RESUME_HOST;
703
704 switch (req) {
705 case H_CEDE:
706 break;
707 case H_PROD:
708 target = kvmppc_get_gpr(vcpu, 4);
709 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
710 if (!tvcpu) {
711 ret = H_PARAMETER;
712 break;
713 }
714 tvcpu->arch.prodded = 1;
715 smp_mb();
716 if (vcpu->arch.ceded) {
717 if (swait_active(&vcpu->wq)) {
718 swake_up(&vcpu->wq);
719 vcpu->stat.halt_wakeup++;
720 }
721 }
722 break;
723 case H_CONFER:
724 target = kvmppc_get_gpr(vcpu, 4);
725 if (target == -1)
726 break;
727 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
728 if (!tvcpu) {
729 ret = H_PARAMETER;
730 break;
731 }
732 yield_count = kvmppc_get_gpr(vcpu, 5);
733 if (kvmppc_get_yield_count(tvcpu) != yield_count)
734 break;
735 kvm_arch_vcpu_yield_to(tvcpu);
736 break;
737 case H_REGISTER_VPA:
738 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
739 kvmppc_get_gpr(vcpu, 5),
740 kvmppc_get_gpr(vcpu, 6));
741 break;
742 case H_RTAS:
743 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
744 return RESUME_HOST;
745
746 idx = srcu_read_lock(&vcpu->kvm->srcu);
747 rc = kvmppc_rtas_hcall(vcpu);
748 srcu_read_unlock(&vcpu->kvm->srcu, idx);
749
750 if (rc == -ENOENT)
751 return RESUME_HOST;
752 else if (rc == 0)
753 break;
754
755 /* Send the error out to userspace via KVM_RUN */
756 return rc;
757 case H_LOGICAL_CI_LOAD:
758 ret = kvmppc_h_logical_ci_load(vcpu);
759 if (ret == H_TOO_HARD)
760 return RESUME_HOST;
761 break;
762 case H_LOGICAL_CI_STORE:
763 ret = kvmppc_h_logical_ci_store(vcpu);
764 if (ret == H_TOO_HARD)
765 return RESUME_HOST;
766 break;
767 case H_SET_MODE:
768 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
769 kvmppc_get_gpr(vcpu, 5),
770 kvmppc_get_gpr(vcpu, 6),
771 kvmppc_get_gpr(vcpu, 7));
772 if (ret == H_TOO_HARD)
773 return RESUME_HOST;
774 break;
775 case H_XIRR:
776 case H_CPPR:
777 case H_EOI:
778 case H_IPI:
779 case H_IPOLL:
780 case H_XIRR_X:
781 if (kvmppc_xics_enabled(vcpu)) {
782 ret = kvmppc_xics_hcall(vcpu, req);
783 break;
784 }
785 return RESUME_HOST;
786 case H_PUT_TCE:
787 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
788 kvmppc_get_gpr(vcpu, 5),
789 kvmppc_get_gpr(vcpu, 6));
790 if (ret == H_TOO_HARD)
791 return RESUME_HOST;
792 break;
793 case H_PUT_TCE_INDIRECT:
794 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
795 kvmppc_get_gpr(vcpu, 5),
796 kvmppc_get_gpr(vcpu, 6),
797 kvmppc_get_gpr(vcpu, 7));
798 if (ret == H_TOO_HARD)
799 return RESUME_HOST;
800 break;
801 case H_STUFF_TCE:
802 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
803 kvmppc_get_gpr(vcpu, 5),
804 kvmppc_get_gpr(vcpu, 6),
805 kvmppc_get_gpr(vcpu, 7));
806 if (ret == H_TOO_HARD)
807 return RESUME_HOST;
808 break;
809 default:
810 return RESUME_HOST;
811 }
812 kvmppc_set_gpr(vcpu, 3, ret);
813 vcpu->arch.hcall_needed = 0;
814 return RESUME_GUEST;
815 }
816
817 static int kvmppc_hcall_impl_hv(unsigned long cmd)
818 {
819 switch (cmd) {
820 case H_CEDE:
821 case H_PROD:
822 case H_CONFER:
823 case H_REGISTER_VPA:
824 case H_SET_MODE:
825 case H_LOGICAL_CI_LOAD:
826 case H_LOGICAL_CI_STORE:
827 #ifdef CONFIG_KVM_XICS
828 case H_XIRR:
829 case H_CPPR:
830 case H_EOI:
831 case H_IPI:
832 case H_IPOLL:
833 case H_XIRR_X:
834 #endif
835 return 1;
836 }
837
838 /* See if it's in the real-mode table */
839 return kvmppc_hcall_impl_hv_realmode(cmd);
840 }
841
842 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
843 struct kvm_vcpu *vcpu)
844 {
845 u32 last_inst;
846
847 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
848 EMULATE_DONE) {
849 /*
850 * Fetch failed, so return to guest and
851 * try executing it again.
852 */
853 return RESUME_GUEST;
854 }
855
856 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
857 run->exit_reason = KVM_EXIT_DEBUG;
858 run->debug.arch.address = kvmppc_get_pc(vcpu);
859 return RESUME_HOST;
860 } else {
861 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
862 return RESUME_GUEST;
863 }
864 }
865
866 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
867 struct task_struct *tsk)
868 {
869 int r = RESUME_HOST;
870
871 vcpu->stat.sum_exits++;
872
873 /*
874 * This can happen if an interrupt occurs in the last stages
875 * of guest entry or the first stages of guest exit (i.e. after
876 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
877 * and before setting it to KVM_GUEST_MODE_HOST_HV).
878 * That can happen due to a bug, or due to a machine check
879 * occurring at just the wrong time.
880 */
881 if (vcpu->arch.shregs.msr & MSR_HV) {
882 printk(KERN_EMERG "KVM trap in HV mode!\n");
883 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
884 vcpu->arch.trap, kvmppc_get_pc(vcpu),
885 vcpu->arch.shregs.msr);
886 kvmppc_dump_regs(vcpu);
887 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
888 run->hw.hardware_exit_reason = vcpu->arch.trap;
889 return RESUME_HOST;
890 }
891 run->exit_reason = KVM_EXIT_UNKNOWN;
892 run->ready_for_interrupt_injection = 1;
893 switch (vcpu->arch.trap) {
894 /* We're good on these - the host merely wanted to get our attention */
895 case BOOK3S_INTERRUPT_HV_DECREMENTER:
896 vcpu->stat.dec_exits++;
897 r = RESUME_GUEST;
898 break;
899 case BOOK3S_INTERRUPT_EXTERNAL:
900 case BOOK3S_INTERRUPT_H_DOORBELL:
901 vcpu->stat.ext_intr_exits++;
902 r = RESUME_GUEST;
903 break;
904 /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
905 case BOOK3S_INTERRUPT_HMI:
906 case BOOK3S_INTERRUPT_PERFMON:
907 r = RESUME_GUEST;
908 break;
909 case BOOK3S_INTERRUPT_MACHINE_CHECK:
910 /*
911 * Deliver a machine check interrupt to the guest.
912 * We have to do this, even if the host has handled the
913 * machine check, because machine checks use SRR0/1 and
914 * the interrupt might have trashed guest state in them.
915 */
916 kvmppc_book3s_queue_irqprio(vcpu,
917 BOOK3S_INTERRUPT_MACHINE_CHECK);
918 r = RESUME_GUEST;
919 break;
920 case BOOK3S_INTERRUPT_PROGRAM:
921 {
922 ulong flags;
923 /*
924 * Normally program interrupts are delivered directly
925 * to the guest by the hardware, but we can get here
926 * as a result of a hypervisor emulation interrupt
927 * (e40) getting turned into a 700 by BML RTAS.
928 */
929 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
930 kvmppc_core_queue_program(vcpu, flags);
931 r = RESUME_GUEST;
932 break;
933 }
934 case BOOK3S_INTERRUPT_SYSCALL:
935 {
936 /* hcall - punt to userspace */
937 int i;
938
939 /* hypercall with MSR_PR has already been handled in rmode,
940 * and never reaches here.
941 */
942
943 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
944 for (i = 0; i < 9; ++i)
945 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
946 run->exit_reason = KVM_EXIT_PAPR_HCALL;
947 vcpu->arch.hcall_needed = 1;
948 r = RESUME_HOST;
949 break;
950 }
951 /*
952 * We get these next two if the guest accesses a page which it thinks
953 * it has mapped but which is not actually present, either because
954 * it is for an emulated I/O device or because the corresonding
955 * host page has been paged out. Any other HDSI/HISI interrupts
956 * have been handled already.
957 */
958 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
959 r = RESUME_PAGE_FAULT;
960 break;
961 case BOOK3S_INTERRUPT_H_INST_STORAGE:
962 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
963 vcpu->arch.fault_dsisr = 0;
964 r = RESUME_PAGE_FAULT;
965 break;
966 /*
967 * This occurs if the guest executes an illegal instruction.
968 * If the guest debug is disabled, generate a program interrupt
969 * to the guest. If guest debug is enabled, we need to check
970 * whether the instruction is a software breakpoint instruction.
971 * Accordingly return to Guest or Host.
972 */
973 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
974 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
975 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
976 swab32(vcpu->arch.emul_inst) :
977 vcpu->arch.emul_inst;
978 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
979 r = kvmppc_emulate_debug_inst(run, vcpu);
980 } else {
981 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
982 r = RESUME_GUEST;
983 }
984 break;
985 /*
986 * This occurs if the guest (kernel or userspace), does something that
987 * is prohibited by HFSCR. We just generate a program interrupt to
988 * the guest.
989 */
990 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
991 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
992 r = RESUME_GUEST;
993 break;
994 default:
995 kvmppc_dump_regs(vcpu);
996 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
997 vcpu->arch.trap, kvmppc_get_pc(vcpu),
998 vcpu->arch.shregs.msr);
999 run->hw.hardware_exit_reason = vcpu->arch.trap;
1000 r = RESUME_HOST;
1001 break;
1002 }
1003
1004 return r;
1005 }
1006
1007 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1008 struct kvm_sregs *sregs)
1009 {
1010 int i;
1011
1012 memset(sregs, 0, sizeof(struct kvm_sregs));
1013 sregs->pvr = vcpu->arch.pvr;
1014 for (i = 0; i < vcpu->arch.slb_max; i++) {
1015 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1016 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1017 }
1018
1019 return 0;
1020 }
1021
1022 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1023 struct kvm_sregs *sregs)
1024 {
1025 int i, j;
1026
1027 /* Only accept the same PVR as the host's, since we can't spoof it */
1028 if (sregs->pvr != vcpu->arch.pvr)
1029 return -EINVAL;
1030
1031 j = 0;
1032 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1033 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1034 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1035 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1036 ++j;
1037 }
1038 }
1039 vcpu->arch.slb_max = j;
1040
1041 return 0;
1042 }
1043
1044 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1045 bool preserve_top32)
1046 {
1047 struct kvm *kvm = vcpu->kvm;
1048 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1049 u64 mask;
1050
1051 mutex_lock(&kvm->lock);
1052 spin_lock(&vc->lock);
1053 /*
1054 * If ILE (interrupt little-endian) has changed, update the
1055 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1056 */
1057 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1058 struct kvm_vcpu *vcpu;
1059 int i;
1060
1061 kvm_for_each_vcpu(i, vcpu, kvm) {
1062 if (vcpu->arch.vcore != vc)
1063 continue;
1064 if (new_lpcr & LPCR_ILE)
1065 vcpu->arch.intr_msr |= MSR_LE;
1066 else
1067 vcpu->arch.intr_msr &= ~MSR_LE;
1068 }
1069 }
1070
1071 /*
1072 * Userspace can only modify DPFD (default prefetch depth),
1073 * ILE (interrupt little-endian) and TC (translation control).
1074 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1075 */
1076 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1077 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1078 mask |= LPCR_AIL;
1079
1080 /* Broken 32-bit version of LPCR must not clear top bits */
1081 if (preserve_top32)
1082 mask &= 0xFFFFFFFF;
1083 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1084 spin_unlock(&vc->lock);
1085 mutex_unlock(&kvm->lock);
1086 }
1087
1088 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1089 union kvmppc_one_reg *val)
1090 {
1091 int r = 0;
1092 long int i;
1093
1094 switch (id) {
1095 case KVM_REG_PPC_DEBUG_INST:
1096 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1097 break;
1098 case KVM_REG_PPC_HIOR:
1099 *val = get_reg_val(id, 0);
1100 break;
1101 case KVM_REG_PPC_DABR:
1102 *val = get_reg_val(id, vcpu->arch.dabr);
1103 break;
1104 case KVM_REG_PPC_DABRX:
1105 *val = get_reg_val(id, vcpu->arch.dabrx);
1106 break;
1107 case KVM_REG_PPC_DSCR:
1108 *val = get_reg_val(id, vcpu->arch.dscr);
1109 break;
1110 case KVM_REG_PPC_PURR:
1111 *val = get_reg_val(id, vcpu->arch.purr);
1112 break;
1113 case KVM_REG_PPC_SPURR:
1114 *val = get_reg_val(id, vcpu->arch.spurr);
1115 break;
1116 case KVM_REG_PPC_AMR:
1117 *val = get_reg_val(id, vcpu->arch.amr);
1118 break;
1119 case KVM_REG_PPC_UAMOR:
1120 *val = get_reg_val(id, vcpu->arch.uamor);
1121 break;
1122 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1123 i = id - KVM_REG_PPC_MMCR0;
1124 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1125 break;
1126 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1127 i = id - KVM_REG_PPC_PMC1;
1128 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1129 break;
1130 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1131 i = id - KVM_REG_PPC_SPMC1;
1132 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1133 break;
1134 case KVM_REG_PPC_SIAR:
1135 *val = get_reg_val(id, vcpu->arch.siar);
1136 break;
1137 case KVM_REG_PPC_SDAR:
1138 *val = get_reg_val(id, vcpu->arch.sdar);
1139 break;
1140 case KVM_REG_PPC_SIER:
1141 *val = get_reg_val(id, vcpu->arch.sier);
1142 break;
1143 case KVM_REG_PPC_IAMR:
1144 *val = get_reg_val(id, vcpu->arch.iamr);
1145 break;
1146 case KVM_REG_PPC_PSPB:
1147 *val = get_reg_val(id, vcpu->arch.pspb);
1148 break;
1149 case KVM_REG_PPC_DPDES:
1150 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1151 break;
1152 case KVM_REG_PPC_DAWR:
1153 *val = get_reg_val(id, vcpu->arch.dawr);
1154 break;
1155 case KVM_REG_PPC_DAWRX:
1156 *val = get_reg_val(id, vcpu->arch.dawrx);
1157 break;
1158 case KVM_REG_PPC_CIABR:
1159 *val = get_reg_val(id, vcpu->arch.ciabr);
1160 break;
1161 case KVM_REG_PPC_CSIGR:
1162 *val = get_reg_val(id, vcpu->arch.csigr);
1163 break;
1164 case KVM_REG_PPC_TACR:
1165 *val = get_reg_val(id, vcpu->arch.tacr);
1166 break;
1167 case KVM_REG_PPC_TCSCR:
1168 *val = get_reg_val(id, vcpu->arch.tcscr);
1169 break;
1170 case KVM_REG_PPC_PID:
1171 *val = get_reg_val(id, vcpu->arch.pid);
1172 break;
1173 case KVM_REG_PPC_ACOP:
1174 *val = get_reg_val(id, vcpu->arch.acop);
1175 break;
1176 case KVM_REG_PPC_WORT:
1177 *val = get_reg_val(id, vcpu->arch.wort);
1178 break;
1179 case KVM_REG_PPC_VPA_ADDR:
1180 spin_lock(&vcpu->arch.vpa_update_lock);
1181 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1182 spin_unlock(&vcpu->arch.vpa_update_lock);
1183 break;
1184 case KVM_REG_PPC_VPA_SLB:
1185 spin_lock(&vcpu->arch.vpa_update_lock);
1186 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1187 val->vpaval.length = vcpu->arch.slb_shadow.len;
1188 spin_unlock(&vcpu->arch.vpa_update_lock);
1189 break;
1190 case KVM_REG_PPC_VPA_DTL:
1191 spin_lock(&vcpu->arch.vpa_update_lock);
1192 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1193 val->vpaval.length = vcpu->arch.dtl.len;
1194 spin_unlock(&vcpu->arch.vpa_update_lock);
1195 break;
1196 case KVM_REG_PPC_TB_OFFSET:
1197 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1198 break;
1199 case KVM_REG_PPC_LPCR:
1200 case KVM_REG_PPC_LPCR_64:
1201 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1202 break;
1203 case KVM_REG_PPC_PPR:
1204 *val = get_reg_val(id, vcpu->arch.ppr);
1205 break;
1206 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1207 case KVM_REG_PPC_TFHAR:
1208 *val = get_reg_val(id, vcpu->arch.tfhar);
1209 break;
1210 case KVM_REG_PPC_TFIAR:
1211 *val = get_reg_val(id, vcpu->arch.tfiar);
1212 break;
1213 case KVM_REG_PPC_TEXASR:
1214 *val = get_reg_val(id, vcpu->arch.texasr);
1215 break;
1216 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1217 i = id - KVM_REG_PPC_TM_GPR0;
1218 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1219 break;
1220 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1221 {
1222 int j;
1223 i = id - KVM_REG_PPC_TM_VSR0;
1224 if (i < 32)
1225 for (j = 0; j < TS_FPRWIDTH; j++)
1226 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1227 else {
1228 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1229 val->vval = vcpu->arch.vr_tm.vr[i-32];
1230 else
1231 r = -ENXIO;
1232 }
1233 break;
1234 }
1235 case KVM_REG_PPC_TM_CR:
1236 *val = get_reg_val(id, vcpu->arch.cr_tm);
1237 break;
1238 case KVM_REG_PPC_TM_XER:
1239 *val = get_reg_val(id, vcpu->arch.xer_tm);
1240 break;
1241 case KVM_REG_PPC_TM_LR:
1242 *val = get_reg_val(id, vcpu->arch.lr_tm);
1243 break;
1244 case KVM_REG_PPC_TM_CTR:
1245 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1246 break;
1247 case KVM_REG_PPC_TM_FPSCR:
1248 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1249 break;
1250 case KVM_REG_PPC_TM_AMR:
1251 *val = get_reg_val(id, vcpu->arch.amr_tm);
1252 break;
1253 case KVM_REG_PPC_TM_PPR:
1254 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1255 break;
1256 case KVM_REG_PPC_TM_VRSAVE:
1257 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1258 break;
1259 case KVM_REG_PPC_TM_VSCR:
1260 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1261 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1262 else
1263 r = -ENXIO;
1264 break;
1265 case KVM_REG_PPC_TM_DSCR:
1266 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1267 break;
1268 case KVM_REG_PPC_TM_TAR:
1269 *val = get_reg_val(id, vcpu->arch.tar_tm);
1270 break;
1271 #endif
1272 case KVM_REG_PPC_ARCH_COMPAT:
1273 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1274 break;
1275 default:
1276 r = -EINVAL;
1277 break;
1278 }
1279
1280 return r;
1281 }
1282
1283 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1284 union kvmppc_one_reg *val)
1285 {
1286 int r = 0;
1287 long int i;
1288 unsigned long addr, len;
1289
1290 switch (id) {
1291 case KVM_REG_PPC_HIOR:
1292 /* Only allow this to be set to zero */
1293 if (set_reg_val(id, *val))
1294 r = -EINVAL;
1295 break;
1296 case KVM_REG_PPC_DABR:
1297 vcpu->arch.dabr = set_reg_val(id, *val);
1298 break;
1299 case KVM_REG_PPC_DABRX:
1300 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1301 break;
1302 case KVM_REG_PPC_DSCR:
1303 vcpu->arch.dscr = set_reg_val(id, *val);
1304 break;
1305 case KVM_REG_PPC_PURR:
1306 vcpu->arch.purr = set_reg_val(id, *val);
1307 break;
1308 case KVM_REG_PPC_SPURR:
1309 vcpu->arch.spurr = set_reg_val(id, *val);
1310 break;
1311 case KVM_REG_PPC_AMR:
1312 vcpu->arch.amr = set_reg_val(id, *val);
1313 break;
1314 case KVM_REG_PPC_UAMOR:
1315 vcpu->arch.uamor = set_reg_val(id, *val);
1316 break;
1317 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1318 i = id - KVM_REG_PPC_MMCR0;
1319 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1320 break;
1321 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1322 i = id - KVM_REG_PPC_PMC1;
1323 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1324 break;
1325 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1326 i = id - KVM_REG_PPC_SPMC1;
1327 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1328 break;
1329 case KVM_REG_PPC_SIAR:
1330 vcpu->arch.siar = set_reg_val(id, *val);
1331 break;
1332 case KVM_REG_PPC_SDAR:
1333 vcpu->arch.sdar = set_reg_val(id, *val);
1334 break;
1335 case KVM_REG_PPC_SIER:
1336 vcpu->arch.sier = set_reg_val(id, *val);
1337 break;
1338 case KVM_REG_PPC_IAMR:
1339 vcpu->arch.iamr = set_reg_val(id, *val);
1340 break;
1341 case KVM_REG_PPC_PSPB:
1342 vcpu->arch.pspb = set_reg_val(id, *val);
1343 break;
1344 case KVM_REG_PPC_DPDES:
1345 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1346 break;
1347 case KVM_REG_PPC_DAWR:
1348 vcpu->arch.dawr = set_reg_val(id, *val);
1349 break;
1350 case KVM_REG_PPC_DAWRX:
1351 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1352 break;
1353 case KVM_REG_PPC_CIABR:
1354 vcpu->arch.ciabr = set_reg_val(id, *val);
1355 /* Don't allow setting breakpoints in hypervisor code */
1356 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1357 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1358 break;
1359 case KVM_REG_PPC_CSIGR:
1360 vcpu->arch.csigr = set_reg_val(id, *val);
1361 break;
1362 case KVM_REG_PPC_TACR:
1363 vcpu->arch.tacr = set_reg_val(id, *val);
1364 break;
1365 case KVM_REG_PPC_TCSCR:
1366 vcpu->arch.tcscr = set_reg_val(id, *val);
1367 break;
1368 case KVM_REG_PPC_PID:
1369 vcpu->arch.pid = set_reg_val(id, *val);
1370 break;
1371 case KVM_REG_PPC_ACOP:
1372 vcpu->arch.acop = set_reg_val(id, *val);
1373 break;
1374 case KVM_REG_PPC_WORT:
1375 vcpu->arch.wort = set_reg_val(id, *val);
1376 break;
1377 case KVM_REG_PPC_VPA_ADDR:
1378 addr = set_reg_val(id, *val);
1379 r = -EINVAL;
1380 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1381 vcpu->arch.dtl.next_gpa))
1382 break;
1383 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1384 break;
1385 case KVM_REG_PPC_VPA_SLB:
1386 addr = val->vpaval.addr;
1387 len = val->vpaval.length;
1388 r = -EINVAL;
1389 if (addr && !vcpu->arch.vpa.next_gpa)
1390 break;
1391 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1392 break;
1393 case KVM_REG_PPC_VPA_DTL:
1394 addr = val->vpaval.addr;
1395 len = val->vpaval.length;
1396 r = -EINVAL;
1397 if (addr && (len < sizeof(struct dtl_entry) ||
1398 !vcpu->arch.vpa.next_gpa))
1399 break;
1400 len -= len % sizeof(struct dtl_entry);
1401 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1402 break;
1403 case KVM_REG_PPC_TB_OFFSET:
1404 /* round up to multiple of 2^24 */
1405 vcpu->arch.vcore->tb_offset =
1406 ALIGN(set_reg_val(id, *val), 1UL << 24);
1407 break;
1408 case KVM_REG_PPC_LPCR:
1409 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1410 break;
1411 case KVM_REG_PPC_LPCR_64:
1412 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1413 break;
1414 case KVM_REG_PPC_PPR:
1415 vcpu->arch.ppr = set_reg_val(id, *val);
1416 break;
1417 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1418 case KVM_REG_PPC_TFHAR:
1419 vcpu->arch.tfhar = set_reg_val(id, *val);
1420 break;
1421 case KVM_REG_PPC_TFIAR:
1422 vcpu->arch.tfiar = set_reg_val(id, *val);
1423 break;
1424 case KVM_REG_PPC_TEXASR:
1425 vcpu->arch.texasr = set_reg_val(id, *val);
1426 break;
1427 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1428 i = id - KVM_REG_PPC_TM_GPR0;
1429 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1430 break;
1431 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1432 {
1433 int j;
1434 i = id - KVM_REG_PPC_TM_VSR0;
1435 if (i < 32)
1436 for (j = 0; j < TS_FPRWIDTH; j++)
1437 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1438 else
1439 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1440 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1441 else
1442 r = -ENXIO;
1443 break;
1444 }
1445 case KVM_REG_PPC_TM_CR:
1446 vcpu->arch.cr_tm = set_reg_val(id, *val);
1447 break;
1448 case KVM_REG_PPC_TM_XER:
1449 vcpu->arch.xer_tm = set_reg_val(id, *val);
1450 break;
1451 case KVM_REG_PPC_TM_LR:
1452 vcpu->arch.lr_tm = set_reg_val(id, *val);
1453 break;
1454 case KVM_REG_PPC_TM_CTR:
1455 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1456 break;
1457 case KVM_REG_PPC_TM_FPSCR:
1458 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1459 break;
1460 case KVM_REG_PPC_TM_AMR:
1461 vcpu->arch.amr_tm = set_reg_val(id, *val);
1462 break;
1463 case KVM_REG_PPC_TM_PPR:
1464 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1465 break;
1466 case KVM_REG_PPC_TM_VRSAVE:
1467 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1468 break;
1469 case KVM_REG_PPC_TM_VSCR:
1470 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1471 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1472 else
1473 r = - ENXIO;
1474 break;
1475 case KVM_REG_PPC_TM_DSCR:
1476 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1477 break;
1478 case KVM_REG_PPC_TM_TAR:
1479 vcpu->arch.tar_tm = set_reg_val(id, *val);
1480 break;
1481 #endif
1482 case KVM_REG_PPC_ARCH_COMPAT:
1483 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1484 break;
1485 default:
1486 r = -EINVAL;
1487 break;
1488 }
1489
1490 return r;
1491 }
1492
1493 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1494 {
1495 struct kvmppc_vcore *vcore;
1496
1497 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1498
1499 if (vcore == NULL)
1500 return NULL;
1501
1502 INIT_LIST_HEAD(&vcore->runnable_threads);
1503 spin_lock_init(&vcore->lock);
1504 spin_lock_init(&vcore->stoltb_lock);
1505 init_swait_queue_head(&vcore->wq);
1506 vcore->preempt_tb = TB_NIL;
1507 vcore->lpcr = kvm->arch.lpcr;
1508 vcore->first_vcpuid = core * threads_per_subcore;
1509 vcore->kvm = kvm;
1510 INIT_LIST_HEAD(&vcore->preempt_list);
1511
1512 return vcore;
1513 }
1514
1515 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1516 static struct debugfs_timings_element {
1517 const char *name;
1518 size_t offset;
1519 } timings[] = {
1520 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1521 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1522 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1523 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1524 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1525 };
1526
1527 #define N_TIMINGS (sizeof(timings) / sizeof(timings[0]))
1528
1529 struct debugfs_timings_state {
1530 struct kvm_vcpu *vcpu;
1531 unsigned int buflen;
1532 char buf[N_TIMINGS * 100];
1533 };
1534
1535 static int debugfs_timings_open(struct inode *inode, struct file *file)
1536 {
1537 struct kvm_vcpu *vcpu = inode->i_private;
1538 struct debugfs_timings_state *p;
1539
1540 p = kzalloc(sizeof(*p), GFP_KERNEL);
1541 if (!p)
1542 return -ENOMEM;
1543
1544 kvm_get_kvm(vcpu->kvm);
1545 p->vcpu = vcpu;
1546 file->private_data = p;
1547
1548 return nonseekable_open(inode, file);
1549 }
1550
1551 static int debugfs_timings_release(struct inode *inode, struct file *file)
1552 {
1553 struct debugfs_timings_state *p = file->private_data;
1554
1555 kvm_put_kvm(p->vcpu->kvm);
1556 kfree(p);
1557 return 0;
1558 }
1559
1560 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1561 size_t len, loff_t *ppos)
1562 {
1563 struct debugfs_timings_state *p = file->private_data;
1564 struct kvm_vcpu *vcpu = p->vcpu;
1565 char *s, *buf_end;
1566 struct kvmhv_tb_accumulator tb;
1567 u64 count;
1568 loff_t pos;
1569 ssize_t n;
1570 int i, loops;
1571 bool ok;
1572
1573 if (!p->buflen) {
1574 s = p->buf;
1575 buf_end = s + sizeof(p->buf);
1576 for (i = 0; i < N_TIMINGS; ++i) {
1577 struct kvmhv_tb_accumulator *acc;
1578
1579 acc = (struct kvmhv_tb_accumulator *)
1580 ((unsigned long)vcpu + timings[i].offset);
1581 ok = false;
1582 for (loops = 0; loops < 1000; ++loops) {
1583 count = acc->seqcount;
1584 if (!(count & 1)) {
1585 smp_rmb();
1586 tb = *acc;
1587 smp_rmb();
1588 if (count == acc->seqcount) {
1589 ok = true;
1590 break;
1591 }
1592 }
1593 udelay(1);
1594 }
1595 if (!ok)
1596 snprintf(s, buf_end - s, "%s: stuck\n",
1597 timings[i].name);
1598 else
1599 snprintf(s, buf_end - s,
1600 "%s: %llu %llu %llu %llu\n",
1601 timings[i].name, count / 2,
1602 tb_to_ns(tb.tb_total),
1603 tb_to_ns(tb.tb_min),
1604 tb_to_ns(tb.tb_max));
1605 s += strlen(s);
1606 }
1607 p->buflen = s - p->buf;
1608 }
1609
1610 pos = *ppos;
1611 if (pos >= p->buflen)
1612 return 0;
1613 if (len > p->buflen - pos)
1614 len = p->buflen - pos;
1615 n = copy_to_user(buf, p->buf + pos, len);
1616 if (n) {
1617 if (n == len)
1618 return -EFAULT;
1619 len -= n;
1620 }
1621 *ppos = pos + len;
1622 return len;
1623 }
1624
1625 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1626 size_t len, loff_t *ppos)
1627 {
1628 return -EACCES;
1629 }
1630
1631 static const struct file_operations debugfs_timings_ops = {
1632 .owner = THIS_MODULE,
1633 .open = debugfs_timings_open,
1634 .release = debugfs_timings_release,
1635 .read = debugfs_timings_read,
1636 .write = debugfs_timings_write,
1637 .llseek = generic_file_llseek,
1638 };
1639
1640 /* Create a debugfs directory for the vcpu */
1641 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1642 {
1643 char buf[16];
1644 struct kvm *kvm = vcpu->kvm;
1645
1646 snprintf(buf, sizeof(buf), "vcpu%u", id);
1647 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1648 return;
1649 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1650 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1651 return;
1652 vcpu->arch.debugfs_timings =
1653 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1654 vcpu, &debugfs_timings_ops);
1655 }
1656
1657 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1658 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1659 {
1660 }
1661 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1662
1663 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1664 unsigned int id)
1665 {
1666 struct kvm_vcpu *vcpu;
1667 int err = -EINVAL;
1668 int core;
1669 struct kvmppc_vcore *vcore;
1670
1671 core = id / threads_per_subcore;
1672 if (core >= KVM_MAX_VCORES)
1673 goto out;
1674
1675 err = -ENOMEM;
1676 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1677 if (!vcpu)
1678 goto out;
1679
1680 err = kvm_vcpu_init(vcpu, kvm, id);
1681 if (err)
1682 goto free_vcpu;
1683
1684 vcpu->arch.shared = &vcpu->arch.shregs;
1685 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1686 /*
1687 * The shared struct is never shared on HV,
1688 * so we can always use host endianness
1689 */
1690 #ifdef __BIG_ENDIAN__
1691 vcpu->arch.shared_big_endian = true;
1692 #else
1693 vcpu->arch.shared_big_endian = false;
1694 #endif
1695 #endif
1696 vcpu->arch.mmcr[0] = MMCR0_FC;
1697 vcpu->arch.ctrl = CTRL_RUNLATCH;
1698 /* default to host PVR, since we can't spoof it */
1699 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1700 spin_lock_init(&vcpu->arch.vpa_update_lock);
1701 spin_lock_init(&vcpu->arch.tbacct_lock);
1702 vcpu->arch.busy_preempt = TB_NIL;
1703 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1704
1705 kvmppc_mmu_book3s_hv_init(vcpu);
1706
1707 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1708
1709 init_waitqueue_head(&vcpu->arch.cpu_run);
1710
1711 mutex_lock(&kvm->lock);
1712 vcore = kvm->arch.vcores[core];
1713 if (!vcore) {
1714 vcore = kvmppc_vcore_create(kvm, core);
1715 kvm->arch.vcores[core] = vcore;
1716 kvm->arch.online_vcores++;
1717 }
1718 mutex_unlock(&kvm->lock);
1719
1720 if (!vcore)
1721 goto free_vcpu;
1722
1723 spin_lock(&vcore->lock);
1724 ++vcore->num_threads;
1725 spin_unlock(&vcore->lock);
1726 vcpu->arch.vcore = vcore;
1727 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1728 vcpu->arch.thread_cpu = -1;
1729
1730 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1731 kvmppc_sanity_check(vcpu);
1732
1733 debugfs_vcpu_init(vcpu, id);
1734
1735 return vcpu;
1736
1737 free_vcpu:
1738 kmem_cache_free(kvm_vcpu_cache, vcpu);
1739 out:
1740 return ERR_PTR(err);
1741 }
1742
1743 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1744 {
1745 if (vpa->pinned_addr)
1746 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1747 vpa->dirty);
1748 }
1749
1750 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1751 {
1752 spin_lock(&vcpu->arch.vpa_update_lock);
1753 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1754 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1755 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1756 spin_unlock(&vcpu->arch.vpa_update_lock);
1757 kvm_vcpu_uninit(vcpu);
1758 kmem_cache_free(kvm_vcpu_cache, vcpu);
1759 }
1760
1761 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1762 {
1763 /* Indicate we want to get back into the guest */
1764 return 1;
1765 }
1766
1767 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1768 {
1769 unsigned long dec_nsec, now;
1770
1771 now = get_tb();
1772 if (now > vcpu->arch.dec_expires) {
1773 /* decrementer has already gone negative */
1774 kvmppc_core_queue_dec(vcpu);
1775 kvmppc_core_prepare_to_enter(vcpu);
1776 return;
1777 }
1778 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1779 / tb_ticks_per_sec;
1780 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1781 HRTIMER_MODE_REL);
1782 vcpu->arch.timer_running = 1;
1783 }
1784
1785 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1786 {
1787 vcpu->arch.ceded = 0;
1788 if (vcpu->arch.timer_running) {
1789 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1790 vcpu->arch.timer_running = 0;
1791 }
1792 }
1793
1794 extern void __kvmppc_vcore_entry(void);
1795
1796 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1797 struct kvm_vcpu *vcpu)
1798 {
1799 u64 now;
1800
1801 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1802 return;
1803 spin_lock_irq(&vcpu->arch.tbacct_lock);
1804 now = mftb();
1805 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1806 vcpu->arch.stolen_logged;
1807 vcpu->arch.busy_preempt = now;
1808 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1809 spin_unlock_irq(&vcpu->arch.tbacct_lock);
1810 --vc->n_runnable;
1811 list_del(&vcpu->arch.run_list);
1812 }
1813
1814 static int kvmppc_grab_hwthread(int cpu)
1815 {
1816 struct paca_struct *tpaca;
1817 long timeout = 10000;
1818
1819 tpaca = &paca[cpu];
1820
1821 /* Ensure the thread won't go into the kernel if it wakes */
1822 tpaca->kvm_hstate.kvm_vcpu = NULL;
1823 tpaca->kvm_hstate.kvm_vcore = NULL;
1824 tpaca->kvm_hstate.napping = 0;
1825 smp_wmb();
1826 tpaca->kvm_hstate.hwthread_req = 1;
1827
1828 /*
1829 * If the thread is already executing in the kernel (e.g. handling
1830 * a stray interrupt), wait for it to get back to nap mode.
1831 * The smp_mb() is to ensure that our setting of hwthread_req
1832 * is visible before we look at hwthread_state, so if this
1833 * races with the code at system_reset_pSeries and the thread
1834 * misses our setting of hwthread_req, we are sure to see its
1835 * setting of hwthread_state, and vice versa.
1836 */
1837 smp_mb();
1838 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1839 if (--timeout <= 0) {
1840 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1841 return -EBUSY;
1842 }
1843 udelay(1);
1844 }
1845 return 0;
1846 }
1847
1848 static void kvmppc_release_hwthread(int cpu)
1849 {
1850 struct paca_struct *tpaca;
1851
1852 tpaca = &paca[cpu];
1853 tpaca->kvm_hstate.hwthread_req = 0;
1854 tpaca->kvm_hstate.kvm_vcpu = NULL;
1855 tpaca->kvm_hstate.kvm_vcore = NULL;
1856 tpaca->kvm_hstate.kvm_split_mode = NULL;
1857 }
1858
1859 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1860 {
1861 int cpu;
1862 struct paca_struct *tpaca;
1863 struct kvmppc_vcore *mvc = vc->master_vcore;
1864
1865 cpu = vc->pcpu;
1866 if (vcpu) {
1867 if (vcpu->arch.timer_running) {
1868 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1869 vcpu->arch.timer_running = 0;
1870 }
1871 cpu += vcpu->arch.ptid;
1872 vcpu->cpu = mvc->pcpu;
1873 vcpu->arch.thread_cpu = cpu;
1874 }
1875 tpaca = &paca[cpu];
1876 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1877 tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1878 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1879 smp_wmb();
1880 tpaca->kvm_hstate.kvm_vcore = mvc;
1881 if (cpu != smp_processor_id())
1882 kvmppc_ipi_thread(cpu);
1883 }
1884
1885 static void kvmppc_wait_for_nap(void)
1886 {
1887 int cpu = smp_processor_id();
1888 int i, loops;
1889
1890 for (loops = 0; loops < 1000000; ++loops) {
1891 /*
1892 * Check if all threads are finished.
1893 * We set the vcore pointer when starting a thread
1894 * and the thread clears it when finished, so we look
1895 * for any threads that still have a non-NULL vcore ptr.
1896 */
1897 for (i = 1; i < threads_per_subcore; ++i)
1898 if (paca[cpu + i].kvm_hstate.kvm_vcore)
1899 break;
1900 if (i == threads_per_subcore) {
1901 HMT_medium();
1902 return;
1903 }
1904 HMT_low();
1905 }
1906 HMT_medium();
1907 for (i = 1; i < threads_per_subcore; ++i)
1908 if (paca[cpu + i].kvm_hstate.kvm_vcore)
1909 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1910 }
1911
1912 /*
1913 * Check that we are on thread 0 and that any other threads in
1914 * this core are off-line. Then grab the threads so they can't
1915 * enter the kernel.
1916 */
1917 static int on_primary_thread(void)
1918 {
1919 int cpu = smp_processor_id();
1920 int thr;
1921
1922 /* Are we on a primary subcore? */
1923 if (cpu_thread_in_subcore(cpu))
1924 return 0;
1925
1926 thr = 0;
1927 while (++thr < threads_per_subcore)
1928 if (cpu_online(cpu + thr))
1929 return 0;
1930
1931 /* Grab all hw threads so they can't go into the kernel */
1932 for (thr = 1; thr < threads_per_subcore; ++thr) {
1933 if (kvmppc_grab_hwthread(cpu + thr)) {
1934 /* Couldn't grab one; let the others go */
1935 do {
1936 kvmppc_release_hwthread(cpu + thr);
1937 } while (--thr > 0);
1938 return 0;
1939 }
1940 }
1941 return 1;
1942 }
1943
1944 /*
1945 * A list of virtual cores for each physical CPU.
1946 * These are vcores that could run but their runner VCPU tasks are
1947 * (or may be) preempted.
1948 */
1949 struct preempted_vcore_list {
1950 struct list_head list;
1951 spinlock_t lock;
1952 };
1953
1954 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1955
1956 static void init_vcore_lists(void)
1957 {
1958 int cpu;
1959
1960 for_each_possible_cpu(cpu) {
1961 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1962 spin_lock_init(&lp->lock);
1963 INIT_LIST_HEAD(&lp->list);
1964 }
1965 }
1966
1967 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1968 {
1969 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1970
1971 vc->vcore_state = VCORE_PREEMPT;
1972 vc->pcpu = smp_processor_id();
1973 if (vc->num_threads < threads_per_subcore) {
1974 spin_lock(&lp->lock);
1975 list_add_tail(&vc->preempt_list, &lp->list);
1976 spin_unlock(&lp->lock);
1977 }
1978
1979 /* Start accumulating stolen time */
1980 kvmppc_core_start_stolen(vc);
1981 }
1982
1983 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1984 {
1985 struct preempted_vcore_list *lp;
1986
1987 kvmppc_core_end_stolen(vc);
1988 if (!list_empty(&vc->preempt_list)) {
1989 lp = &per_cpu(preempted_vcores, vc->pcpu);
1990 spin_lock(&lp->lock);
1991 list_del_init(&vc->preempt_list);
1992 spin_unlock(&lp->lock);
1993 }
1994 vc->vcore_state = VCORE_INACTIVE;
1995 }
1996
1997 /*
1998 * This stores information about the virtual cores currently
1999 * assigned to a physical core.
2000 */
2001 struct core_info {
2002 int n_subcores;
2003 int max_subcore_threads;
2004 int total_threads;
2005 int subcore_threads[MAX_SUBCORES];
2006 struct kvm *subcore_vm[MAX_SUBCORES];
2007 struct list_head vcs[MAX_SUBCORES];
2008 };
2009
2010 /*
2011 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2012 * respectively in 2-way micro-threading (split-core) mode.
2013 */
2014 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2015
2016 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2017 {
2018 int sub;
2019
2020 memset(cip, 0, sizeof(*cip));
2021 cip->n_subcores = 1;
2022 cip->max_subcore_threads = vc->num_threads;
2023 cip->total_threads = vc->num_threads;
2024 cip->subcore_threads[0] = vc->num_threads;
2025 cip->subcore_vm[0] = vc->kvm;
2026 for (sub = 0; sub < MAX_SUBCORES; ++sub)
2027 INIT_LIST_HEAD(&cip->vcs[sub]);
2028 list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2029 }
2030
2031 static bool subcore_config_ok(int n_subcores, int n_threads)
2032 {
2033 /* Can only dynamically split if unsplit to begin with */
2034 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2035 return false;
2036 if (n_subcores > MAX_SUBCORES)
2037 return false;
2038 if (n_subcores > 1) {
2039 if (!(dynamic_mt_modes & 2))
2040 n_subcores = 4;
2041 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2042 return false;
2043 }
2044
2045 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2046 }
2047
2048 static void init_master_vcore(struct kvmppc_vcore *vc)
2049 {
2050 vc->master_vcore = vc;
2051 vc->entry_exit_map = 0;
2052 vc->in_guest = 0;
2053 vc->napping_threads = 0;
2054 vc->conferring_threads = 0;
2055 }
2056
2057 /*
2058 * See if the existing subcores can be split into 3 (or fewer) subcores
2059 * of at most two threads each, so we can fit in another vcore. This
2060 * assumes there are at most two subcores and at most 6 threads in total.
2061 */
2062 static bool can_split_piggybacked_subcores(struct core_info *cip)
2063 {
2064 int sub, new_sub;
2065 int large_sub = -1;
2066 int thr;
2067 int n_subcores = cip->n_subcores;
2068 struct kvmppc_vcore *vc, *vcnext;
2069 struct kvmppc_vcore *master_vc = NULL;
2070
2071 for (sub = 0; sub < cip->n_subcores; ++sub) {
2072 if (cip->subcore_threads[sub] <= 2)
2073 continue;
2074 if (large_sub >= 0)
2075 return false;
2076 large_sub = sub;
2077 vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2078 preempt_list);
2079 if (vc->num_threads > 2)
2080 return false;
2081 n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2082 }
2083 if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2084 return false;
2085
2086 /*
2087 * Seems feasible, so go through and move vcores to new subcores.
2088 * Note that when we have two or more vcores in one subcore,
2089 * all those vcores must have only one thread each.
2090 */
2091 new_sub = cip->n_subcores;
2092 thr = 0;
2093 sub = large_sub;
2094 list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2095 if (thr >= 2) {
2096 list_del(&vc->preempt_list);
2097 list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2098 /* vc->num_threads must be 1 */
2099 if (++cip->subcore_threads[new_sub] == 1) {
2100 cip->subcore_vm[new_sub] = vc->kvm;
2101 init_master_vcore(vc);
2102 master_vc = vc;
2103 ++cip->n_subcores;
2104 } else {
2105 vc->master_vcore = master_vc;
2106 ++new_sub;
2107 }
2108 }
2109 thr += vc->num_threads;
2110 }
2111 cip->subcore_threads[large_sub] = 2;
2112 cip->max_subcore_threads = 2;
2113
2114 return true;
2115 }
2116
2117 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2118 {
2119 int n_threads = vc->num_threads;
2120 int sub;
2121
2122 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2123 return false;
2124
2125 if (n_threads < cip->max_subcore_threads)
2126 n_threads = cip->max_subcore_threads;
2127 if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2128 cip->max_subcore_threads = n_threads;
2129 } else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2130 vc->num_threads <= 2) {
2131 /*
2132 * We may be able to fit another subcore in by
2133 * splitting an existing subcore with 3 or 4
2134 * threads into two 2-thread subcores, or one
2135 * with 5 or 6 threads into three subcores.
2136 * We can only do this if those subcores have
2137 * piggybacked virtual cores.
2138 */
2139 if (!can_split_piggybacked_subcores(cip))
2140 return false;
2141 } else {
2142 return false;
2143 }
2144
2145 sub = cip->n_subcores;
2146 ++cip->n_subcores;
2147 cip->total_threads += vc->num_threads;
2148 cip->subcore_threads[sub] = vc->num_threads;
2149 cip->subcore_vm[sub] = vc->kvm;
2150 init_master_vcore(vc);
2151 list_del(&vc->preempt_list);
2152 list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2153
2154 return true;
2155 }
2156
2157 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2158 struct core_info *cip, int sub)
2159 {
2160 struct kvmppc_vcore *vc;
2161 int n_thr;
2162
2163 vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2164 preempt_list);
2165
2166 /* require same VM and same per-core reg values */
2167 if (pvc->kvm != vc->kvm ||
2168 pvc->tb_offset != vc->tb_offset ||
2169 pvc->pcr != vc->pcr ||
2170 pvc->lpcr != vc->lpcr)
2171 return false;
2172
2173 /* P8 guest with > 1 thread per core would see wrong TIR value */
2174 if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2175 (vc->num_threads > 1 || pvc->num_threads > 1))
2176 return false;
2177
2178 n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2179 if (n_thr > cip->max_subcore_threads) {
2180 if (!subcore_config_ok(cip->n_subcores, n_thr))
2181 return false;
2182 cip->max_subcore_threads = n_thr;
2183 }
2184
2185 cip->total_threads += pvc->num_threads;
2186 cip->subcore_threads[sub] = n_thr;
2187 pvc->master_vcore = vc;
2188 list_del(&pvc->preempt_list);
2189 list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2190
2191 return true;
2192 }
2193
2194 /*
2195 * Work out whether it is possible to piggyback the execution of
2196 * vcore *pvc onto the execution of the other vcores described in *cip.
2197 */
2198 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2199 int target_threads)
2200 {
2201 int sub;
2202
2203 if (cip->total_threads + pvc->num_threads > target_threads)
2204 return false;
2205 for (sub = 0; sub < cip->n_subcores; ++sub)
2206 if (cip->subcore_threads[sub] &&
2207 can_piggyback_subcore(pvc, cip, sub))
2208 return true;
2209
2210 if (can_dynamic_split(pvc, cip))
2211 return true;
2212
2213 return false;
2214 }
2215
2216 static void prepare_threads(struct kvmppc_vcore *vc)
2217 {
2218 struct kvm_vcpu *vcpu, *vnext;
2219
2220 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2221 arch.run_list) {
2222 if (signal_pending(vcpu->arch.run_task))
2223 vcpu->arch.ret = -EINTR;
2224 else if (vcpu->arch.vpa.update_pending ||
2225 vcpu->arch.slb_shadow.update_pending ||
2226 vcpu->arch.dtl.update_pending)
2227 vcpu->arch.ret = RESUME_GUEST;
2228 else
2229 continue;
2230 kvmppc_remove_runnable(vc, vcpu);
2231 wake_up(&vcpu->arch.cpu_run);
2232 }
2233 }
2234
2235 static void collect_piggybacks(struct core_info *cip, int target_threads)
2236 {
2237 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2238 struct kvmppc_vcore *pvc, *vcnext;
2239
2240 spin_lock(&lp->lock);
2241 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2242 if (!spin_trylock(&pvc->lock))
2243 continue;
2244 prepare_threads(pvc);
2245 if (!pvc->n_runnable) {
2246 list_del_init(&pvc->preempt_list);
2247 if (pvc->runner == NULL) {
2248 pvc->vcore_state = VCORE_INACTIVE;
2249 kvmppc_core_end_stolen(pvc);
2250 }
2251 spin_unlock(&pvc->lock);
2252 continue;
2253 }
2254 if (!can_piggyback(pvc, cip, target_threads)) {
2255 spin_unlock(&pvc->lock);
2256 continue;
2257 }
2258 kvmppc_core_end_stolen(pvc);
2259 pvc->vcore_state = VCORE_PIGGYBACK;
2260 if (cip->total_threads >= target_threads)
2261 break;
2262 }
2263 spin_unlock(&lp->lock);
2264 }
2265
2266 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2267 {
2268 int still_running = 0;
2269 u64 now;
2270 long ret;
2271 struct kvm_vcpu *vcpu, *vnext;
2272
2273 spin_lock(&vc->lock);
2274 now = get_tb();
2275 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2276 arch.run_list) {
2277 /* cancel pending dec exception if dec is positive */
2278 if (now < vcpu->arch.dec_expires &&
2279 kvmppc_core_pending_dec(vcpu))
2280 kvmppc_core_dequeue_dec(vcpu);
2281
2282 trace_kvm_guest_exit(vcpu);
2283
2284 ret = RESUME_GUEST;
2285 if (vcpu->arch.trap)
2286 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2287 vcpu->arch.run_task);
2288
2289 vcpu->arch.ret = ret;
2290 vcpu->arch.trap = 0;
2291
2292 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2293 if (vcpu->arch.pending_exceptions)
2294 kvmppc_core_prepare_to_enter(vcpu);
2295 if (vcpu->arch.ceded)
2296 kvmppc_set_timer(vcpu);
2297 else
2298 ++still_running;
2299 } else {
2300 kvmppc_remove_runnable(vc, vcpu);
2301 wake_up(&vcpu->arch.cpu_run);
2302 }
2303 }
2304 list_del_init(&vc->preempt_list);
2305 if (!is_master) {
2306 if (still_running > 0) {
2307 kvmppc_vcore_preempt(vc);
2308 } else if (vc->runner) {
2309 vc->vcore_state = VCORE_PREEMPT;
2310 kvmppc_core_start_stolen(vc);
2311 } else {
2312 vc->vcore_state = VCORE_INACTIVE;
2313 }
2314 if (vc->n_runnable > 0 && vc->runner == NULL) {
2315 /* make sure there's a candidate runner awake */
2316 vcpu = list_first_entry(&vc->runnable_threads,
2317 struct kvm_vcpu, arch.run_list);
2318 wake_up(&vcpu->arch.cpu_run);
2319 }
2320 }
2321 spin_unlock(&vc->lock);
2322 }
2323
2324 /*
2325 * Clear core from the list of active host cores as we are about to
2326 * enter the guest. Only do this if it is the primary thread of the
2327 * core (not if a subcore) that is entering the guest.
2328 */
2329 static inline void kvmppc_clear_host_core(int cpu)
2330 {
2331 int core;
2332
2333 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2334 return;
2335 /*
2336 * Memory barrier can be omitted here as we will do a smp_wmb()
2337 * later in kvmppc_start_thread and we need ensure that state is
2338 * visible to other CPUs only after we enter guest.
2339 */
2340 core = cpu >> threads_shift;
2341 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2342 }
2343
2344 /*
2345 * Advertise this core as an active host core since we exited the guest
2346 * Only need to do this if it is the primary thread of the core that is
2347 * exiting.
2348 */
2349 static inline void kvmppc_set_host_core(int cpu)
2350 {
2351 int core;
2352
2353 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2354 return;
2355
2356 /*
2357 * Memory barrier can be omitted here because we do a spin_unlock
2358 * immediately after this which provides the memory barrier.
2359 */
2360 core = cpu >> threads_shift;
2361 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2362 }
2363
2364 /*
2365 * Run a set of guest threads on a physical core.
2366 * Called with vc->lock held.
2367 */
2368 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2369 {
2370 struct kvm_vcpu *vcpu, *vnext;
2371 int i;
2372 int srcu_idx;
2373 struct core_info core_info;
2374 struct kvmppc_vcore *pvc, *vcnext;
2375 struct kvm_split_mode split_info, *sip;
2376 int split, subcore_size, active;
2377 int sub;
2378 bool thr0_done;
2379 unsigned long cmd_bit, stat_bit;
2380 int pcpu, thr;
2381 int target_threads;
2382
2383 /*
2384 * Remove from the list any threads that have a signal pending
2385 * or need a VPA update done
2386 */
2387 prepare_threads(vc);
2388
2389 /* if the runner is no longer runnable, let the caller pick a new one */
2390 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2391 return;
2392
2393 /*
2394 * Initialize *vc.
2395 */
2396 init_master_vcore(vc);
2397 vc->preempt_tb = TB_NIL;
2398
2399 /*
2400 * Make sure we are running on primary threads, and that secondary
2401 * threads are offline. Also check if the number of threads in this
2402 * guest are greater than the current system threads per guest.
2403 */
2404 if ((threads_per_core > 1) &&
2405 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2406 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2407 arch.run_list) {
2408 vcpu->arch.ret = -EBUSY;
2409 kvmppc_remove_runnable(vc, vcpu);
2410 wake_up(&vcpu->arch.cpu_run);
2411 }
2412 goto out;
2413 }
2414
2415 /*
2416 * See if we could run any other vcores on the physical core
2417 * along with this one.
2418 */
2419 init_core_info(&core_info, vc);
2420 pcpu = smp_processor_id();
2421 target_threads = threads_per_subcore;
2422 if (target_smt_mode && target_smt_mode < target_threads)
2423 target_threads = target_smt_mode;
2424 if (vc->num_threads < target_threads)
2425 collect_piggybacks(&core_info, target_threads);
2426
2427 /* Decide on micro-threading (split-core) mode */
2428 subcore_size = threads_per_subcore;
2429 cmd_bit = stat_bit = 0;
2430 split = core_info.n_subcores;
2431 sip = NULL;
2432 if (split > 1) {
2433 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2434 if (split == 2 && (dynamic_mt_modes & 2)) {
2435 cmd_bit = HID0_POWER8_1TO2LPAR;
2436 stat_bit = HID0_POWER8_2LPARMODE;
2437 } else {
2438 split = 4;
2439 cmd_bit = HID0_POWER8_1TO4LPAR;
2440 stat_bit = HID0_POWER8_4LPARMODE;
2441 }
2442 subcore_size = MAX_SMT_THREADS / split;
2443 sip = &split_info;
2444 memset(&split_info, 0, sizeof(split_info));
2445 split_info.rpr = mfspr(SPRN_RPR);
2446 split_info.pmmar = mfspr(SPRN_PMMAR);
2447 split_info.ldbar = mfspr(SPRN_LDBAR);
2448 split_info.subcore_size = subcore_size;
2449 for (sub = 0; sub < core_info.n_subcores; ++sub)
2450 split_info.master_vcs[sub] =
2451 list_first_entry(&core_info.vcs[sub],
2452 struct kvmppc_vcore, preempt_list);
2453 /* order writes to split_info before kvm_split_mode pointer */
2454 smp_wmb();
2455 }
2456 pcpu = smp_processor_id();
2457 for (thr = 0; thr < threads_per_subcore; ++thr)
2458 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2459
2460 /* Initiate micro-threading (split-core) if required */
2461 if (cmd_bit) {
2462 unsigned long hid0 = mfspr(SPRN_HID0);
2463
2464 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2465 mb();
2466 mtspr(SPRN_HID0, hid0);
2467 isync();
2468 for (;;) {
2469 hid0 = mfspr(SPRN_HID0);
2470 if (hid0 & stat_bit)
2471 break;
2472 cpu_relax();
2473 }
2474 }
2475
2476 kvmppc_clear_host_core(pcpu);
2477
2478 /* Start all the threads */
2479 active = 0;
2480 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2481 thr = subcore_thread_map[sub];
2482 thr0_done = false;
2483 active |= 1 << thr;
2484 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2485 pvc->pcpu = pcpu + thr;
2486 list_for_each_entry(vcpu, &pvc->runnable_threads,
2487 arch.run_list) {
2488 kvmppc_start_thread(vcpu, pvc);
2489 kvmppc_create_dtl_entry(vcpu, pvc);
2490 trace_kvm_guest_enter(vcpu);
2491 if (!vcpu->arch.ptid)
2492 thr0_done = true;
2493 active |= 1 << (thr + vcpu->arch.ptid);
2494 }
2495 /*
2496 * We need to start the first thread of each subcore
2497 * even if it doesn't have a vcpu.
2498 */
2499 if (pvc->master_vcore == pvc && !thr0_done)
2500 kvmppc_start_thread(NULL, pvc);
2501 thr += pvc->num_threads;
2502 }
2503 }
2504
2505 /*
2506 * Ensure that split_info.do_nap is set after setting
2507 * the vcore pointer in the PACA of the secondaries.
2508 */
2509 smp_mb();
2510 if (cmd_bit)
2511 split_info.do_nap = 1; /* ask secondaries to nap when done */
2512
2513 /*
2514 * When doing micro-threading, poke the inactive threads as well.
2515 * This gets them to the nap instruction after kvm_do_nap,
2516 * which reduces the time taken to unsplit later.
2517 */
2518 if (split > 1)
2519 for (thr = 1; thr < threads_per_subcore; ++thr)
2520 if (!(active & (1 << thr)))
2521 kvmppc_ipi_thread(pcpu + thr);
2522
2523 vc->vcore_state = VCORE_RUNNING;
2524 preempt_disable();
2525
2526 trace_kvmppc_run_core(vc, 0);
2527
2528 for (sub = 0; sub < core_info.n_subcores; ++sub)
2529 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2530 spin_unlock(&pvc->lock);
2531
2532 guest_enter();
2533
2534 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2535
2536 __kvmppc_vcore_entry();
2537
2538 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2539
2540 spin_lock(&vc->lock);
2541 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2542 vc->vcore_state = VCORE_EXITING;
2543
2544 /* wait for secondary threads to finish writing their state to memory */
2545 kvmppc_wait_for_nap();
2546
2547 /* Return to whole-core mode if we split the core earlier */
2548 if (split > 1) {
2549 unsigned long hid0 = mfspr(SPRN_HID0);
2550 unsigned long loops = 0;
2551
2552 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2553 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2554 mb();
2555 mtspr(SPRN_HID0, hid0);
2556 isync();
2557 for (;;) {
2558 hid0 = mfspr(SPRN_HID0);
2559 if (!(hid0 & stat_bit))
2560 break;
2561 cpu_relax();
2562 ++loops;
2563 }
2564 split_info.do_nap = 0;
2565 }
2566
2567 /* Let secondaries go back to the offline loop */
2568 for (i = 0; i < threads_per_subcore; ++i) {
2569 kvmppc_release_hwthread(pcpu + i);
2570 if (sip && sip->napped[i])
2571 kvmppc_ipi_thread(pcpu + i);
2572 }
2573
2574 kvmppc_set_host_core(pcpu);
2575
2576 spin_unlock(&vc->lock);
2577
2578 /* make sure updates to secondary vcpu structs are visible now */
2579 smp_mb();
2580 guest_exit();
2581
2582 for (sub = 0; sub < core_info.n_subcores; ++sub)
2583 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2584 preempt_list)
2585 post_guest_process(pvc, pvc == vc);
2586
2587 spin_lock(&vc->lock);
2588 preempt_enable();
2589
2590 out:
2591 vc->vcore_state = VCORE_INACTIVE;
2592 trace_kvmppc_run_core(vc, 1);
2593 }
2594
2595 /*
2596 * Wait for some other vcpu thread to execute us, and
2597 * wake us up when we need to handle something in the host.
2598 */
2599 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2600 struct kvm_vcpu *vcpu, int wait_state)
2601 {
2602 DEFINE_WAIT(wait);
2603
2604 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2605 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2606 spin_unlock(&vc->lock);
2607 schedule();
2608 spin_lock(&vc->lock);
2609 }
2610 finish_wait(&vcpu->arch.cpu_run, &wait);
2611 }
2612
2613 /*
2614 * All the vcpus in this vcore are idle, so wait for a decrementer
2615 * or external interrupt to one of the vcpus. vc->lock is held.
2616 */
2617 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2618 {
2619 struct kvm_vcpu *vcpu;
2620 int do_sleep = 1;
2621 DECLARE_SWAITQUEUE(wait);
2622
2623 prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2624
2625 /*
2626 * Check one last time for pending exceptions and ceded state after
2627 * we put ourselves on the wait queue
2628 */
2629 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2630 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2631 do_sleep = 0;
2632 break;
2633 }
2634 }
2635
2636 if (!do_sleep) {
2637 finish_swait(&vc->wq, &wait);
2638 return;
2639 }
2640
2641 vc->vcore_state = VCORE_SLEEPING;
2642 trace_kvmppc_vcore_blocked(vc, 0);
2643 spin_unlock(&vc->lock);
2644 schedule();
2645 finish_swait(&vc->wq, &wait);
2646 spin_lock(&vc->lock);
2647 vc->vcore_state = VCORE_INACTIVE;
2648 trace_kvmppc_vcore_blocked(vc, 1);
2649 }
2650
2651 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2652 {
2653 int n_ceded;
2654 struct kvmppc_vcore *vc;
2655 struct kvm_vcpu *v, *vn;
2656
2657 trace_kvmppc_run_vcpu_enter(vcpu);
2658
2659 kvm_run->exit_reason = 0;
2660 vcpu->arch.ret = RESUME_GUEST;
2661 vcpu->arch.trap = 0;
2662 kvmppc_update_vpas(vcpu);
2663
2664 /*
2665 * Synchronize with other threads in this virtual core
2666 */
2667 vc = vcpu->arch.vcore;
2668 spin_lock(&vc->lock);
2669 vcpu->arch.ceded = 0;
2670 vcpu->arch.run_task = current;
2671 vcpu->arch.kvm_run = kvm_run;
2672 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2673 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2674 vcpu->arch.busy_preempt = TB_NIL;
2675 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2676 ++vc->n_runnable;
2677
2678 /*
2679 * This happens the first time this is called for a vcpu.
2680 * If the vcore is already running, we may be able to start
2681 * this thread straight away and have it join in.
2682 */
2683 if (!signal_pending(current)) {
2684 if (vc->vcore_state == VCORE_PIGGYBACK) {
2685 struct kvmppc_vcore *mvc = vc->master_vcore;
2686 if (spin_trylock(&mvc->lock)) {
2687 if (mvc->vcore_state == VCORE_RUNNING &&
2688 !VCORE_IS_EXITING(mvc)) {
2689 kvmppc_create_dtl_entry(vcpu, vc);
2690 kvmppc_start_thread(vcpu, vc);
2691 trace_kvm_guest_enter(vcpu);
2692 }
2693 spin_unlock(&mvc->lock);
2694 }
2695 } else if (vc->vcore_state == VCORE_RUNNING &&
2696 !VCORE_IS_EXITING(vc)) {
2697 kvmppc_create_dtl_entry(vcpu, vc);
2698 kvmppc_start_thread(vcpu, vc);
2699 trace_kvm_guest_enter(vcpu);
2700 } else if (vc->vcore_state == VCORE_SLEEPING) {
2701 swake_up(&vc->wq);
2702 }
2703
2704 }
2705
2706 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2707 !signal_pending(current)) {
2708 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2709 kvmppc_vcore_end_preempt(vc);
2710
2711 if (vc->vcore_state != VCORE_INACTIVE) {
2712 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2713 continue;
2714 }
2715 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2716 arch.run_list) {
2717 kvmppc_core_prepare_to_enter(v);
2718 if (signal_pending(v->arch.run_task)) {
2719 kvmppc_remove_runnable(vc, v);
2720 v->stat.signal_exits++;
2721 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2722 v->arch.ret = -EINTR;
2723 wake_up(&v->arch.cpu_run);
2724 }
2725 }
2726 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2727 break;
2728 n_ceded = 0;
2729 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2730 if (!v->arch.pending_exceptions)
2731 n_ceded += v->arch.ceded;
2732 else
2733 v->arch.ceded = 0;
2734 }
2735 vc->runner = vcpu;
2736 if (n_ceded == vc->n_runnable) {
2737 kvmppc_vcore_blocked(vc);
2738 } else if (need_resched()) {
2739 kvmppc_vcore_preempt(vc);
2740 /* Let something else run */
2741 cond_resched_lock(&vc->lock);
2742 if (vc->vcore_state == VCORE_PREEMPT)
2743 kvmppc_vcore_end_preempt(vc);
2744 } else {
2745 kvmppc_run_core(vc);
2746 }
2747 vc->runner = NULL;
2748 }
2749
2750 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2751 (vc->vcore_state == VCORE_RUNNING ||
2752 vc->vcore_state == VCORE_EXITING ||
2753 vc->vcore_state == VCORE_PIGGYBACK))
2754 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2755
2756 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2757 kvmppc_vcore_end_preempt(vc);
2758
2759 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2760 kvmppc_remove_runnable(vc, vcpu);
2761 vcpu->stat.signal_exits++;
2762 kvm_run->exit_reason = KVM_EXIT_INTR;
2763 vcpu->arch.ret = -EINTR;
2764 }
2765
2766 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2767 /* Wake up some vcpu to run the core */
2768 v = list_first_entry(&vc->runnable_threads,
2769 struct kvm_vcpu, arch.run_list);
2770 wake_up(&v->arch.cpu_run);
2771 }
2772
2773 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2774 spin_unlock(&vc->lock);
2775 return vcpu->arch.ret;
2776 }
2777
2778 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2779 {
2780 int r;
2781 int srcu_idx;
2782
2783 if (!vcpu->arch.sane) {
2784 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2785 return -EINVAL;
2786 }
2787
2788 kvmppc_core_prepare_to_enter(vcpu);
2789
2790 /* No need to go into the guest when all we'll do is come back out */
2791 if (signal_pending(current)) {
2792 run->exit_reason = KVM_EXIT_INTR;
2793 return -EINTR;
2794 }
2795
2796 atomic_inc(&vcpu->kvm->arch.vcpus_running);
2797 /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2798 smp_mb();
2799
2800 /* On the first time here, set up HTAB and VRMA */
2801 if (!vcpu->kvm->arch.hpte_setup_done) {
2802 r = kvmppc_hv_setup_htab_rma(vcpu);
2803 if (r)
2804 goto out;
2805 }
2806
2807 flush_all_to_thread(current);
2808
2809 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2810 vcpu->arch.pgdir = current->mm->pgd;
2811 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2812
2813 do {
2814 r = kvmppc_run_vcpu(run, vcpu);
2815
2816 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2817 !(vcpu->arch.shregs.msr & MSR_PR)) {
2818 trace_kvm_hcall_enter(vcpu);
2819 r = kvmppc_pseries_do_hcall(vcpu);
2820 trace_kvm_hcall_exit(vcpu, r);
2821 kvmppc_core_prepare_to_enter(vcpu);
2822 } else if (r == RESUME_PAGE_FAULT) {
2823 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2824 r = kvmppc_book3s_hv_page_fault(run, vcpu,
2825 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2826 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2827 }
2828 } while (is_kvmppc_resume_guest(r));
2829
2830 out:
2831 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2832 atomic_dec(&vcpu->kvm->arch.vcpus_running);
2833 return r;
2834 }
2835
2836 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2837 int linux_psize)
2838 {
2839 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2840
2841 if (!def->shift)
2842 return;
2843 (*sps)->page_shift = def->shift;
2844 (*sps)->slb_enc = def->sllp;
2845 (*sps)->enc[0].page_shift = def->shift;
2846 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2847 /*
2848 * Add 16MB MPSS support if host supports it
2849 */
2850 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2851 (*sps)->enc[1].page_shift = 24;
2852 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2853 }
2854 (*sps)++;
2855 }
2856
2857 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2858 struct kvm_ppc_smmu_info *info)
2859 {
2860 struct kvm_ppc_one_seg_page_size *sps;
2861
2862 info->flags = KVM_PPC_PAGE_SIZES_REAL;
2863 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2864 info->flags |= KVM_PPC_1T_SEGMENTS;
2865 info->slb_size = mmu_slb_size;
2866
2867 /* We only support these sizes for now, and no muti-size segments */
2868 sps = &info->sps[0];
2869 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2870 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2871 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2872
2873 return 0;
2874 }
2875
2876 /*
2877 * Get (and clear) the dirty memory log for a memory slot.
2878 */
2879 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2880 struct kvm_dirty_log *log)
2881 {
2882 struct kvm_memslots *slots;
2883 struct kvm_memory_slot *memslot;
2884 int r;
2885 unsigned long n;
2886
2887 mutex_lock(&kvm->slots_lock);
2888
2889 r = -EINVAL;
2890 if (log->slot >= KVM_USER_MEM_SLOTS)
2891 goto out;
2892
2893 slots = kvm_memslots(kvm);
2894 memslot = id_to_memslot(slots, log->slot);
2895 r = -ENOENT;
2896 if (!memslot->dirty_bitmap)
2897 goto out;
2898
2899 n = kvm_dirty_bitmap_bytes(memslot);
2900 memset(memslot->dirty_bitmap, 0, n);
2901
2902 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2903 if (r)
2904 goto out;
2905
2906 r = -EFAULT;
2907 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2908 goto out;
2909
2910 r = 0;
2911 out:
2912 mutex_unlock(&kvm->slots_lock);
2913 return r;
2914 }
2915
2916 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2917 struct kvm_memory_slot *dont)
2918 {
2919 if (!dont || free->arch.rmap != dont->arch.rmap) {
2920 vfree(free->arch.rmap);
2921 free->arch.rmap = NULL;
2922 }
2923 }
2924
2925 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2926 unsigned long npages)
2927 {
2928 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2929 if (!slot->arch.rmap)
2930 return -ENOMEM;
2931
2932 return 0;
2933 }
2934
2935 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2936 struct kvm_memory_slot *memslot,
2937 const struct kvm_userspace_memory_region *mem)
2938 {
2939 return 0;
2940 }
2941
2942 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2943 const struct kvm_userspace_memory_region *mem,
2944 const struct kvm_memory_slot *old,
2945 const struct kvm_memory_slot *new)
2946 {
2947 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2948 struct kvm_memslots *slots;
2949 struct kvm_memory_slot *memslot;
2950
2951 if (npages && old->npages) {
2952 /*
2953 * If modifying a memslot, reset all the rmap dirty bits.
2954 * If this is a new memslot, we don't need to do anything
2955 * since the rmap array starts out as all zeroes,
2956 * i.e. no pages are dirty.
2957 */
2958 slots = kvm_memslots(kvm);
2959 memslot = id_to_memslot(slots, mem->slot);
2960 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2961 }
2962 }
2963
2964 /*
2965 * Update LPCR values in kvm->arch and in vcores.
2966 * Caller must hold kvm->lock.
2967 */
2968 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2969 {
2970 long int i;
2971 u32 cores_done = 0;
2972
2973 if ((kvm->arch.lpcr & mask) == lpcr)
2974 return;
2975
2976 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2977
2978 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2979 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2980 if (!vc)
2981 continue;
2982 spin_lock(&vc->lock);
2983 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2984 spin_unlock(&vc->lock);
2985 if (++cores_done >= kvm->arch.online_vcores)
2986 break;
2987 }
2988 }
2989
2990 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2991 {
2992 return;
2993 }
2994
2995 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2996 {
2997 int err = 0;
2998 struct kvm *kvm = vcpu->kvm;
2999 unsigned long hva;
3000 struct kvm_memory_slot *memslot;
3001 struct vm_area_struct *vma;
3002 unsigned long lpcr = 0, senc;
3003 unsigned long psize, porder;
3004 int srcu_idx;
3005
3006 mutex_lock(&kvm->lock);
3007 if (kvm->arch.hpte_setup_done)
3008 goto out; /* another vcpu beat us to it */
3009
3010 /* Allocate hashed page table (if not done already) and reset it */
3011 if (!kvm->arch.hpt_virt) {
3012 err = kvmppc_alloc_hpt(kvm, NULL);
3013 if (err) {
3014 pr_err("KVM: Couldn't alloc HPT\n");
3015 goto out;
3016 }
3017 }
3018
3019 /* Look up the memslot for guest physical address 0 */
3020 srcu_idx = srcu_read_lock(&kvm->srcu);
3021 memslot = gfn_to_memslot(kvm, 0);
3022
3023 /* We must have some memory at 0 by now */
3024 err = -EINVAL;
3025 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3026 goto out_srcu;
3027
3028 /* Look up the VMA for the start of this memory slot */
3029 hva = memslot->userspace_addr;
3030 down_read(¤t->mm->mmap_sem);
3031 vma = find_vma(current->mm, hva);
3032 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3033 goto up_out;
3034
3035 psize = vma_kernel_pagesize(vma);
3036 porder = __ilog2(psize);
3037
3038 up_read(¤t->mm->mmap_sem);
3039
3040 /* We can handle 4k, 64k or 16M pages in the VRMA */
3041 err = -EINVAL;
3042 if (!(psize == 0x1000 || psize == 0x10000 ||
3043 psize == 0x1000000))
3044 goto out_srcu;
3045
3046 /* Update VRMASD field in the LPCR */
3047 senc = slb_pgsize_encoding(psize);
3048 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3049 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3050 /* the -4 is to account for senc values starting at 0x10 */
3051 lpcr = senc << (LPCR_VRMASD_SH - 4);
3052
3053 /* Create HPTEs in the hash page table for the VRMA */
3054 kvmppc_map_vrma(vcpu, memslot, porder);
3055
3056 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3057
3058 /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3059 smp_wmb();
3060 kvm->arch.hpte_setup_done = 1;
3061 err = 0;
3062 out_srcu:
3063 srcu_read_unlock(&kvm->srcu, srcu_idx);
3064 out:
3065 mutex_unlock(&kvm->lock);
3066 return err;
3067
3068 up_out:
3069 up_read(¤t->mm->mmap_sem);
3070 goto out_srcu;
3071 }
3072
3073 #ifdef CONFIG_KVM_XICS
3074 static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3075 void *hcpu)
3076 {
3077 unsigned long cpu = (long)hcpu;
3078
3079 switch (action) {
3080 case CPU_UP_PREPARE:
3081 case CPU_UP_PREPARE_FROZEN:
3082 kvmppc_set_host_core(cpu);
3083 break;
3084
3085 #ifdef CONFIG_HOTPLUG_CPU
3086 case CPU_DEAD:
3087 case CPU_DEAD_FROZEN:
3088 case CPU_UP_CANCELED:
3089 case CPU_UP_CANCELED_FROZEN:
3090 kvmppc_clear_host_core(cpu);
3091 break;
3092 #endif
3093 default:
3094 break;
3095 }
3096
3097 return NOTIFY_OK;
3098 }
3099
3100 static struct notifier_block kvmppc_cpu_notifier = {
3101 .notifier_call = kvmppc_cpu_notify,
3102 };
3103
3104 /*
3105 * Allocate a per-core structure for managing state about which cores are
3106 * running in the host versus the guest and for exchanging data between
3107 * real mode KVM and CPU running in the host.
3108 * This is only done for the first VM.
3109 * The allocated structure stays even if all VMs have stopped.
3110 * It is only freed when the kvm-hv module is unloaded.
3111 * It's OK for this routine to fail, we just don't support host
3112 * core operations like redirecting H_IPI wakeups.
3113 */
3114 void kvmppc_alloc_host_rm_ops(void)
3115 {
3116 struct kvmppc_host_rm_ops *ops;
3117 unsigned long l_ops;
3118 int cpu, core;
3119 int size;
3120
3121 /* Not the first time here ? */
3122 if (kvmppc_host_rm_ops_hv != NULL)
3123 return;
3124
3125 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3126 if (!ops)
3127 return;
3128
3129 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3130 ops->rm_core = kzalloc(size, GFP_KERNEL);
3131
3132 if (!ops->rm_core) {
3133 kfree(ops);
3134 return;
3135 }
3136
3137 get_online_cpus();
3138
3139 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3140 if (!cpu_online(cpu))
3141 continue;
3142
3143 core = cpu >> threads_shift;
3144 ops->rm_core[core].rm_state.in_host = 1;
3145 }
3146
3147 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3148
3149 /*
3150 * Make the contents of the kvmppc_host_rm_ops structure visible
3151 * to other CPUs before we assign it to the global variable.
3152 * Do an atomic assignment (no locks used here), but if someone
3153 * beats us to it, just free our copy and return.
3154 */
3155 smp_wmb();
3156 l_ops = (unsigned long) ops;
3157
3158 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3159 put_online_cpus();
3160 kfree(ops->rm_core);
3161 kfree(ops);
3162 return;
3163 }
3164
3165 register_cpu_notifier(&kvmppc_cpu_notifier);
3166
3167 put_online_cpus();
3168 }
3169
3170 void kvmppc_free_host_rm_ops(void)
3171 {
3172 if (kvmppc_host_rm_ops_hv) {
3173 unregister_cpu_notifier(&kvmppc_cpu_notifier);
3174 kfree(kvmppc_host_rm_ops_hv->rm_core);
3175 kfree(kvmppc_host_rm_ops_hv);
3176 kvmppc_host_rm_ops_hv = NULL;
3177 }
3178 }
3179 #endif
3180
3181 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3182 {
3183 unsigned long lpcr, lpid;
3184 char buf[32];
3185
3186 /* Allocate the guest's logical partition ID */
3187
3188 lpid = kvmppc_alloc_lpid();
3189 if ((long)lpid < 0)
3190 return -ENOMEM;
3191 kvm->arch.lpid = lpid;
3192
3193 kvmppc_alloc_host_rm_ops();
3194
3195 /*
3196 * Since we don't flush the TLB when tearing down a VM,
3197 * and this lpid might have previously been used,
3198 * make sure we flush on each core before running the new VM.
3199 */
3200 cpumask_setall(&kvm->arch.need_tlb_flush);
3201
3202 /* Start out with the default set of hcalls enabled */
3203 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3204 sizeof(kvm->arch.enabled_hcalls));
3205
3206 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3207
3208 /* Init LPCR for virtual RMA mode */
3209 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3210 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3211 lpcr &= LPCR_PECE | LPCR_LPES;
3212 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3213 LPCR_VPM0 | LPCR_VPM1;
3214 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3215 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3216 /* On POWER8 turn on online bit to enable PURR/SPURR */
3217 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3218 lpcr |= LPCR_ONL;
3219 kvm->arch.lpcr = lpcr;
3220
3221 /*
3222 * Track that we now have a HV mode VM active. This blocks secondary
3223 * CPU threads from coming online.
3224 */
3225 kvm_hv_vm_activated();
3226
3227 /*
3228 * Create a debugfs directory for the VM
3229 */
3230 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3231 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3232 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3233 kvmppc_mmu_debugfs_init(kvm);
3234
3235 return 0;
3236 }
3237
3238 static void kvmppc_free_vcores(struct kvm *kvm)
3239 {
3240 long int i;
3241
3242 for (i = 0; i < KVM_MAX_VCORES; ++i)
3243 kfree(kvm->arch.vcores[i]);
3244 kvm->arch.online_vcores = 0;
3245 }
3246
3247 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3248 {
3249 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3250
3251 kvm_hv_vm_deactivated();
3252
3253 kvmppc_free_vcores(kvm);
3254
3255 kvmppc_free_hpt(kvm);
3256 }
3257
3258 /* We don't need to emulate any privileged instructions or dcbz */
3259 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3260 unsigned int inst, int *advance)
3261 {
3262 return EMULATE_FAIL;
3263 }
3264
3265 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3266 ulong spr_val)
3267 {
3268 return EMULATE_FAIL;
3269 }
3270
3271 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3272 ulong *spr_val)
3273 {
3274 return EMULATE_FAIL;
3275 }
3276
3277 static int kvmppc_core_check_processor_compat_hv(void)
3278 {
3279 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3280 !cpu_has_feature(CPU_FTR_ARCH_206))
3281 return -EIO;
3282 /*
3283 * Disable KVM for Power9, untill the required bits merged.
3284 */
3285 if (cpu_has_feature(CPU_FTR_ARCH_300))
3286 return -EIO;
3287
3288 return 0;
3289 }
3290
3291 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3292 unsigned int ioctl, unsigned long arg)
3293 {
3294 struct kvm *kvm __maybe_unused = filp->private_data;
3295 void __user *argp = (void __user *)arg;
3296 long r;
3297
3298 switch (ioctl) {
3299
3300 case KVM_PPC_ALLOCATE_HTAB: {
3301 u32 htab_order;
3302
3303 r = -EFAULT;
3304 if (get_user(htab_order, (u32 __user *)argp))
3305 break;
3306 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3307 if (r)
3308 break;
3309 r = -EFAULT;
3310 if (put_user(htab_order, (u32 __user *)argp))
3311 break;
3312 r = 0;
3313 break;
3314 }
3315
3316 case KVM_PPC_GET_HTAB_FD: {
3317 struct kvm_get_htab_fd ghf;
3318
3319 r = -EFAULT;
3320 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3321 break;
3322 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3323 break;
3324 }
3325
3326 default:
3327 r = -ENOTTY;
3328 }
3329
3330 return r;
3331 }
3332
3333 /*
3334 * List of hcall numbers to enable by default.
3335 * For compatibility with old userspace, we enable by default
3336 * all hcalls that were implemented before the hcall-enabling
3337 * facility was added. Note this list should not include H_RTAS.
3338 */
3339 static unsigned int default_hcall_list[] = {
3340 H_REMOVE,
3341 H_ENTER,
3342 H_READ,
3343 H_PROTECT,
3344 H_BULK_REMOVE,
3345 H_GET_TCE,
3346 H_PUT_TCE,
3347 H_SET_DABR,
3348 H_SET_XDABR,
3349 H_CEDE,
3350 H_PROD,
3351 H_CONFER,
3352 H_REGISTER_VPA,
3353 #ifdef CONFIG_KVM_XICS
3354 H_EOI,
3355 H_CPPR,
3356 H_IPI,
3357 H_IPOLL,
3358 H_XIRR,
3359 H_XIRR_X,
3360 #endif
3361 0
3362 };
3363
3364 static void init_default_hcalls(void)
3365 {
3366 int i;
3367 unsigned int hcall;
3368
3369 for (i = 0; default_hcall_list[i]; ++i) {
3370 hcall = default_hcall_list[i];
3371 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3372 __set_bit(hcall / 4, default_enabled_hcalls);
3373 }
3374 }
3375
3376 static struct kvmppc_ops kvm_ops_hv = {
3377 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3378 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3379 .get_one_reg = kvmppc_get_one_reg_hv,
3380 .set_one_reg = kvmppc_set_one_reg_hv,
3381 .vcpu_load = kvmppc_core_vcpu_load_hv,
3382 .vcpu_put = kvmppc_core_vcpu_put_hv,
3383 .set_msr = kvmppc_set_msr_hv,
3384 .vcpu_run = kvmppc_vcpu_run_hv,
3385 .vcpu_create = kvmppc_core_vcpu_create_hv,
3386 .vcpu_free = kvmppc_core_vcpu_free_hv,
3387 .check_requests = kvmppc_core_check_requests_hv,
3388 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
3389 .flush_memslot = kvmppc_core_flush_memslot_hv,
3390 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3391 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
3392 .unmap_hva = kvm_unmap_hva_hv,
3393 .unmap_hva_range = kvm_unmap_hva_range_hv,
3394 .age_hva = kvm_age_hva_hv,
3395 .test_age_hva = kvm_test_age_hva_hv,
3396 .set_spte_hva = kvm_set_spte_hva_hv,
3397 .mmu_destroy = kvmppc_mmu_destroy_hv,
3398 .free_memslot = kvmppc_core_free_memslot_hv,
3399 .create_memslot = kvmppc_core_create_memslot_hv,
3400 .init_vm = kvmppc_core_init_vm_hv,
3401 .destroy_vm = kvmppc_core_destroy_vm_hv,
3402 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3403 .emulate_op = kvmppc_core_emulate_op_hv,
3404 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3405 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3406 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3407 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
3408 .hcall_implemented = kvmppc_hcall_impl_hv,
3409 };
3410
3411 static int kvm_init_subcore_bitmap(void)
3412 {
3413 int i, j;
3414 int nr_cores = cpu_nr_cores();
3415 struct sibling_subcore_state *sibling_subcore_state;
3416
3417 for (i = 0; i < nr_cores; i++) {
3418 int first_cpu = i * threads_per_core;
3419 int node = cpu_to_node(first_cpu);
3420
3421 /* Ignore if it is already allocated. */
3422 if (paca[first_cpu].sibling_subcore_state)
3423 continue;
3424
3425 sibling_subcore_state =
3426 kmalloc_node(sizeof(struct sibling_subcore_state),
3427 GFP_KERNEL, node);
3428 if (!sibling_subcore_state)
3429 return -ENOMEM;
3430
3431 memset(sibling_subcore_state, 0,
3432 sizeof(struct sibling_subcore_state));
3433
3434 for (j = 0; j < threads_per_core; j++) {
3435 int cpu = first_cpu + j;
3436
3437 paca[cpu].sibling_subcore_state = sibling_subcore_state;
3438 }
3439 }
3440 return 0;
3441 }
3442
3443 static int kvmppc_book3s_init_hv(void)
3444 {
3445 int r;
3446 /*
3447 * FIXME!! Do we need to check on all cpus ?
3448 */
3449 r = kvmppc_core_check_processor_compat_hv();
3450 if (r < 0)
3451 return -ENODEV;
3452
3453 r = kvm_init_subcore_bitmap();
3454 if (r)
3455 return r;
3456
3457 kvm_ops_hv.owner = THIS_MODULE;
3458 kvmppc_hv_ops = &kvm_ops_hv;
3459
3460 init_default_hcalls();
3461
3462 init_vcore_lists();
3463
3464 r = kvmppc_mmu_hv_init();
3465 return r;
3466 }
3467
3468 static void kvmppc_book3s_exit_hv(void)
3469 {
3470 kvmppc_free_host_rm_ops();
3471 kvmppc_hv_ops = NULL;
3472 }
3473
3474 module_init(kvmppc_book3s_init_hv);
3475 module_exit(kvmppc_book3s_exit_hv);
3476 MODULE_LICENSE("GPL");
3477 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3478 MODULE_ALIAS("devname:kvm");
3479
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.