2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
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.
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/signal.h>
26 #include <linux/sched/stat.h>
27 #include <linux/delay.h>
28 #include <linux/export.h>
30 #include <linux/anon_inodes.h>
31 #include <linux/cpu.h>
32 #include <linux/cpumask.h>
33 #include <linux/spinlock.h>
34 #include <linux/page-flags.h>
35 #include <linux/srcu.h>
36 #include <linux/miscdevice.h>
37 #include <linux/debugfs.h>
38 #include <linux/gfp.h>
39 #include <linux/vmalloc.h>
40 #include <linux/highmem.h>
41 #include <linux/hugetlb.h>
42 #include <linux/kvm_irqfd.h>
43 #include <linux/irqbypass.h>
44 #include <linux/module.h>
45 #include <linux/compiler.h>
49 #include <asm/ppc-opcode.h>
50 #include <asm/disassemble.h>
51 #include <asm/cputable.h>
52 #include <asm/cacheflush.h>
53 #include <asm/tlbflush.h>
54 #include <linux/uaccess.h>
56 #include <asm/kvm_ppc.h>
57 #include <asm/kvm_book3s.h>
58 #include <asm/mmu_context.h>
59 #include <asm/lppaca.h>
60 #include <asm/processor.h>
61 #include <asm/cputhreads.h>
63 #include <asm/hvcall.h>
64 #include <asm/switch_to.h>
66 #include <asm/dbell.h>
68 #include <asm/pnv-pci.h>
76 #define CREATE_TRACE_POINTS
79 /* #define EXIT_DEBUG */
80 /* #define EXIT_DEBUG_SIMPLE */
81 /* #define EXIT_DEBUG_INT */
83 /* Used to indicate that a guest page fault needs to be handled */
84 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
85 /* Used to indicate that a guest passthrough interrupt needs to be handled */
86 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
88 /* Used as a "null" value for timebase values */
89 #define TB_NIL (~(u64)0)
91 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
93 static int dynamic_mt_modes = 6;
94 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
95 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
96 static int target_smt_mode;
97 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
98 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
100 #ifdef CONFIG_KVM_XICS
101 static struct kernel_param_ops module_param_ops = {
102 .set = param_set_int,
103 .get = param_get_int,
106 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
108 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
110 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
112 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
115 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
116 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
118 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
122 struct kvm_vcpu *vcpu;
124 while (++i < MAX_SMT_THREADS) {
125 vcpu = READ_ONCE(vc->runnable_threads[i]);
134 /* Used to traverse the list of runnable threads for a given vcore */
135 #define for_each_runnable_thread(i, vcpu, vc) \
136 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
138 static bool kvmppc_ipi_thread(int cpu)
140 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
142 /* On POWER9 we can use msgsnd to IPI any cpu */
143 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
144 msg |= get_hard_smp_processor_id(cpu);
146 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
150 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
151 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
153 if (cpu_first_thread_sibling(cpu) ==
154 cpu_first_thread_sibling(smp_processor_id())) {
155 msg |= cpu_thread_in_core(cpu);
157 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
164 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
165 if (cpu >= 0 && cpu < nr_cpu_ids) {
166 if (paca[cpu].kvm_hstate.xics_phys) {
170 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
178 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
181 struct swait_queue_head *wqp;
183 wqp = kvm_arch_vcpu_wq(vcpu);
184 if (swq_has_sleeper(wqp)) {
186 ++vcpu->stat.halt_wakeup;
189 cpu = READ_ONCE(vcpu->arch.thread_cpu);
190 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
193 /* CPU points to the first thread of the core */
195 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
196 smp_send_reschedule(cpu);
200 * We use the vcpu_load/put functions to measure stolen time.
201 * Stolen time is counted as time when either the vcpu is able to
202 * run as part of a virtual core, but the task running the vcore
203 * is preempted or sleeping, or when the vcpu needs something done
204 * in the kernel by the task running the vcpu, but that task is
205 * preempted or sleeping. Those two things have to be counted
206 * separately, since one of the vcpu tasks will take on the job
207 * of running the core, and the other vcpu tasks in the vcore will
208 * sleep waiting for it to do that, but that sleep shouldn't count
211 * Hence we accumulate stolen time when the vcpu can run as part of
212 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
213 * needs its task to do other things in the kernel (for example,
214 * service a page fault) in busy_stolen. We don't accumulate
215 * stolen time for a vcore when it is inactive, or for a vcpu
216 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
217 * a misnomer; it means that the vcpu task is not executing in
218 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
219 * the kernel. We don't have any way of dividing up that time
220 * between time that the vcpu is genuinely stopped, time that
221 * the task is actively working on behalf of the vcpu, and time
222 * that the task is preempted, so we don't count any of it as
225 * Updates to busy_stolen are protected by arch.tbacct_lock;
226 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
227 * lock. The stolen times are measured in units of timebase ticks.
228 * (Note that the != TB_NIL checks below are purely defensive;
229 * they should never fail.)
232 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
236 spin_lock_irqsave(&vc->stoltb_lock, flags);
237 vc->preempt_tb = mftb();
238 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
241 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
245 spin_lock_irqsave(&vc->stoltb_lock, flags);
246 if (vc->preempt_tb != TB_NIL) {
247 vc->stolen_tb += mftb() - vc->preempt_tb;
248 vc->preempt_tb = TB_NIL;
250 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
253 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
255 struct kvmppc_vcore *vc = vcpu->arch.vcore;
259 * We can test vc->runner without taking the vcore lock,
260 * because only this task ever sets vc->runner to this
261 * vcpu, and once it is set to this vcpu, only this task
262 * ever sets it to NULL.
264 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
265 kvmppc_core_end_stolen(vc);
267 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
268 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
269 vcpu->arch.busy_preempt != TB_NIL) {
270 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
271 vcpu->arch.busy_preempt = TB_NIL;
273 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
276 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
278 struct kvmppc_vcore *vc = vcpu->arch.vcore;
281 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
282 kvmppc_core_start_stolen(vc);
284 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
285 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
286 vcpu->arch.busy_preempt = mftb();
287 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
290 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
293 * Check for illegal transactional state bit combination
294 * and if we find it, force the TS field to a safe state.
296 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
298 vcpu->arch.shregs.msr = msr;
299 kvmppc_end_cede(vcpu);
302 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
304 vcpu->arch.pvr = pvr;
307 /* Dummy value used in computing PCR value below */
308 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
310 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
312 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
313 struct kvmppc_vcore *vc = vcpu->arch.vcore;
315 /* We can (emulate) our own architecture version and anything older */
316 if (cpu_has_feature(CPU_FTR_ARCH_300))
317 host_pcr_bit = PCR_ARCH_300;
318 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
319 host_pcr_bit = PCR_ARCH_207;
320 else if (cpu_has_feature(CPU_FTR_ARCH_206))
321 host_pcr_bit = PCR_ARCH_206;
323 host_pcr_bit = PCR_ARCH_205;
325 /* Determine lowest PCR bit needed to run guest in given PVR level */
326 guest_pcr_bit = host_pcr_bit;
328 switch (arch_compat) {
330 guest_pcr_bit = PCR_ARCH_205;
334 guest_pcr_bit = PCR_ARCH_206;
337 guest_pcr_bit = PCR_ARCH_207;
340 guest_pcr_bit = PCR_ARCH_300;
347 /* Check requested PCR bits don't exceed our capabilities */
348 if (guest_pcr_bit > host_pcr_bit)
351 spin_lock(&vc->lock);
352 vc->arch_compat = arch_compat;
353 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
354 vc->pcr = host_pcr_bit - guest_pcr_bit;
355 spin_unlock(&vc->lock);
360 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
364 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
365 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
366 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
367 for (r = 0; r < 16; ++r)
368 pr_err("r%2d = %.16lx r%d = %.16lx\n",
369 r, kvmppc_get_gpr(vcpu, r),
370 r+16, kvmppc_get_gpr(vcpu, r+16));
371 pr_err("ctr = %.16lx lr = %.16lx\n",
372 vcpu->arch.ctr, vcpu->arch.lr);
373 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
374 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
375 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
376 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
377 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
378 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
379 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
380 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
381 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
382 pr_err("fault dar = %.16lx dsisr = %.8x\n",
383 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
384 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
385 for (r = 0; r < vcpu->arch.slb_max; ++r)
386 pr_err(" ESID = %.16llx VSID = %.16llx\n",
387 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
388 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
389 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
390 vcpu->arch.last_inst);
393 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
395 struct kvm_vcpu *ret;
397 mutex_lock(&kvm->lock);
398 ret = kvm_get_vcpu_by_id(kvm, id);
399 mutex_unlock(&kvm->lock);
403 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
405 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
406 vpa->yield_count = cpu_to_be32(1);
409 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
410 unsigned long addr, unsigned long len)
412 /* check address is cacheline aligned */
413 if (addr & (L1_CACHE_BYTES - 1))
415 spin_lock(&vcpu->arch.vpa_update_lock);
416 if (v->next_gpa != addr || v->len != len) {
418 v->len = addr ? len : 0;
419 v->update_pending = 1;
421 spin_unlock(&vcpu->arch.vpa_update_lock);
425 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
434 static int vpa_is_registered(struct kvmppc_vpa *vpap)
436 if (vpap->update_pending)
437 return vpap->next_gpa != 0;
438 return vpap->pinned_addr != NULL;
441 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
443 unsigned long vcpuid, unsigned long vpa)
445 struct kvm *kvm = vcpu->kvm;
446 unsigned long len, nb;
448 struct kvm_vcpu *tvcpu;
451 struct kvmppc_vpa *vpap;
453 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
457 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
458 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
459 subfunc == H_VPA_REG_SLB) {
460 /* Registering new area - address must be cache-line aligned */
461 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
464 /* convert logical addr to kernel addr and read length */
465 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
468 if (subfunc == H_VPA_REG_VPA)
469 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
471 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
472 kvmppc_unpin_guest_page(kvm, va, vpa, false);
475 if (len > nb || len < sizeof(struct reg_vpa))
484 spin_lock(&tvcpu->arch.vpa_update_lock);
487 case H_VPA_REG_VPA: /* register VPA */
489 * The size of our lppaca is 1kB because of the way we align
490 * it for the guest to avoid crossing a 4kB boundary. We only
491 * use 640 bytes of the structure though, so we should accept
492 * clients that set a size of 640.
496 vpap = &tvcpu->arch.vpa;
500 case H_VPA_REG_DTL: /* register DTL */
501 if (len < sizeof(struct dtl_entry))
503 len -= len % sizeof(struct dtl_entry);
505 /* Check that they have previously registered a VPA */
507 if (!vpa_is_registered(&tvcpu->arch.vpa))
510 vpap = &tvcpu->arch.dtl;
514 case H_VPA_REG_SLB: /* register SLB shadow buffer */
515 /* Check that they have previously registered a VPA */
517 if (!vpa_is_registered(&tvcpu->arch.vpa))
520 vpap = &tvcpu->arch.slb_shadow;
524 case H_VPA_DEREG_VPA: /* deregister VPA */
525 /* Check they don't still have a DTL or SLB buf registered */
527 if (vpa_is_registered(&tvcpu->arch.dtl) ||
528 vpa_is_registered(&tvcpu->arch.slb_shadow))
531 vpap = &tvcpu->arch.vpa;
535 case H_VPA_DEREG_DTL: /* deregister DTL */
536 vpap = &tvcpu->arch.dtl;
540 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
541 vpap = &tvcpu->arch.slb_shadow;
547 vpap->next_gpa = vpa;
549 vpap->update_pending = 1;
552 spin_unlock(&tvcpu->arch.vpa_update_lock);
557 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
559 struct kvm *kvm = vcpu->kvm;
565 * We need to pin the page pointed to by vpap->next_gpa,
566 * but we can't call kvmppc_pin_guest_page under the lock
567 * as it does get_user_pages() and down_read(). So we
568 * have to drop the lock, pin the page, then get the lock
569 * again and check that a new area didn't get registered
573 gpa = vpap->next_gpa;
574 spin_unlock(&vcpu->arch.vpa_update_lock);
578 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
579 spin_lock(&vcpu->arch.vpa_update_lock);
580 if (gpa == vpap->next_gpa)
582 /* sigh... unpin that one and try again */
584 kvmppc_unpin_guest_page(kvm, va, gpa, false);
587 vpap->update_pending = 0;
588 if (va && nb < vpap->len) {
590 * If it's now too short, it must be that userspace
591 * has changed the mappings underlying guest memory,
592 * so unregister the region.
594 kvmppc_unpin_guest_page(kvm, va, gpa, false);
597 if (vpap->pinned_addr)
598 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
601 vpap->pinned_addr = va;
604 vpap->pinned_end = va + vpap->len;
607 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
609 if (!(vcpu->arch.vpa.update_pending ||
610 vcpu->arch.slb_shadow.update_pending ||
611 vcpu->arch.dtl.update_pending))
614 spin_lock(&vcpu->arch.vpa_update_lock);
615 if (vcpu->arch.vpa.update_pending) {
616 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
617 if (vcpu->arch.vpa.pinned_addr)
618 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
620 if (vcpu->arch.dtl.update_pending) {
621 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
622 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
623 vcpu->arch.dtl_index = 0;
625 if (vcpu->arch.slb_shadow.update_pending)
626 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
627 spin_unlock(&vcpu->arch.vpa_update_lock);
631 * Return the accumulated stolen time for the vcore up until `now'.
632 * The caller should hold the vcore lock.
634 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
639 spin_lock_irqsave(&vc->stoltb_lock, flags);
641 if (vc->vcore_state != VCORE_INACTIVE &&
642 vc->preempt_tb != TB_NIL)
643 p += now - vc->preempt_tb;
644 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
648 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
649 struct kvmppc_vcore *vc)
651 struct dtl_entry *dt;
653 unsigned long stolen;
654 unsigned long core_stolen;
658 dt = vcpu->arch.dtl_ptr;
659 vpa = vcpu->arch.vpa.pinned_addr;
661 core_stolen = vcore_stolen_time(vc, now);
662 stolen = core_stolen - vcpu->arch.stolen_logged;
663 vcpu->arch.stolen_logged = core_stolen;
664 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
665 stolen += vcpu->arch.busy_stolen;
666 vcpu->arch.busy_stolen = 0;
667 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
670 memset(dt, 0, sizeof(struct dtl_entry));
671 dt->dispatch_reason = 7;
672 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
673 dt->timebase = cpu_to_be64(now + vc->tb_offset);
674 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
675 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
676 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
678 if (dt == vcpu->arch.dtl.pinned_end)
679 dt = vcpu->arch.dtl.pinned_addr;
680 vcpu->arch.dtl_ptr = dt;
681 /* order writing *dt vs. writing vpa->dtl_idx */
683 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
684 vcpu->arch.dtl.dirty = true;
687 /* See if there is a doorbell interrupt pending for a vcpu */
688 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
691 struct kvmppc_vcore *vc;
693 if (vcpu->arch.doorbell_request)
696 * Ensure that the read of vcore->dpdes comes after the read
697 * of vcpu->doorbell_request. This barrier matches the
698 * lwsync in book3s_hv_rmhandlers.S just before the
699 * fast_guest_return label.
702 vc = vcpu->arch.vcore;
703 thr = vcpu->vcpu_id - vc->first_vcpuid;
704 return !!(vc->dpdes & (1 << thr));
707 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
709 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
711 if ((!vcpu->arch.vcore->arch_compat) &&
712 cpu_has_feature(CPU_FTR_ARCH_207S))
717 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
718 unsigned long resource, unsigned long value1,
719 unsigned long value2)
722 case H_SET_MODE_RESOURCE_SET_CIABR:
723 if (!kvmppc_power8_compatible(vcpu))
728 return H_UNSUPPORTED_FLAG_START;
729 /* Guests can't breakpoint the hypervisor */
730 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
732 vcpu->arch.ciabr = value1;
734 case H_SET_MODE_RESOURCE_SET_DAWR:
735 if (!kvmppc_power8_compatible(vcpu))
738 return H_UNSUPPORTED_FLAG_START;
739 if (value2 & DABRX_HYP)
741 vcpu->arch.dawr = value1;
742 vcpu->arch.dawrx = value2;
749 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
751 struct kvmppc_vcore *vcore = target->arch.vcore;
754 * We expect to have been called by the real mode handler
755 * (kvmppc_rm_h_confer()) which would have directly returned
756 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
757 * have useful work to do and should not confer) so we don't
761 spin_lock(&vcore->lock);
762 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
763 vcore->vcore_state != VCORE_INACTIVE &&
765 target = vcore->runner;
766 spin_unlock(&vcore->lock);
768 return kvm_vcpu_yield_to(target);
771 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
774 struct lppaca *lppaca;
776 spin_lock(&vcpu->arch.vpa_update_lock);
777 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
779 yield_count = be32_to_cpu(lppaca->yield_count);
780 spin_unlock(&vcpu->arch.vpa_update_lock);
784 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
786 unsigned long req = kvmppc_get_gpr(vcpu, 3);
787 unsigned long target, ret = H_SUCCESS;
789 struct kvm_vcpu *tvcpu;
792 if (req <= MAX_HCALL_OPCODE &&
793 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
800 target = kvmppc_get_gpr(vcpu, 4);
801 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
806 tvcpu->arch.prodded = 1;
808 if (tvcpu->arch.ceded)
809 kvmppc_fast_vcpu_kick_hv(tvcpu);
812 target = kvmppc_get_gpr(vcpu, 4);
815 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
820 yield_count = kvmppc_get_gpr(vcpu, 5);
821 if (kvmppc_get_yield_count(tvcpu) != yield_count)
823 kvm_arch_vcpu_yield_to(tvcpu);
826 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
827 kvmppc_get_gpr(vcpu, 5),
828 kvmppc_get_gpr(vcpu, 6));
831 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
834 idx = srcu_read_lock(&vcpu->kvm->srcu);
835 rc = kvmppc_rtas_hcall(vcpu);
836 srcu_read_unlock(&vcpu->kvm->srcu, idx);
843 /* Send the error out to userspace via KVM_RUN */
845 case H_LOGICAL_CI_LOAD:
846 ret = kvmppc_h_logical_ci_load(vcpu);
847 if (ret == H_TOO_HARD)
850 case H_LOGICAL_CI_STORE:
851 ret = kvmppc_h_logical_ci_store(vcpu);
852 if (ret == H_TOO_HARD)
856 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
857 kvmppc_get_gpr(vcpu, 5),
858 kvmppc_get_gpr(vcpu, 6),
859 kvmppc_get_gpr(vcpu, 7));
860 if (ret == H_TOO_HARD)
869 if (kvmppc_xics_enabled(vcpu)) {
870 if (xive_enabled()) {
871 ret = H_NOT_AVAILABLE;
874 ret = kvmppc_xics_hcall(vcpu, req);
879 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
880 kvmppc_get_gpr(vcpu, 5),
881 kvmppc_get_gpr(vcpu, 6));
882 if (ret == H_TOO_HARD)
885 case H_PUT_TCE_INDIRECT:
886 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
887 kvmppc_get_gpr(vcpu, 5),
888 kvmppc_get_gpr(vcpu, 6),
889 kvmppc_get_gpr(vcpu, 7));
890 if (ret == H_TOO_HARD)
894 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
895 kvmppc_get_gpr(vcpu, 5),
896 kvmppc_get_gpr(vcpu, 6),
897 kvmppc_get_gpr(vcpu, 7));
898 if (ret == H_TOO_HARD)
904 kvmppc_set_gpr(vcpu, 3, ret);
905 vcpu->arch.hcall_needed = 0;
909 static int kvmppc_hcall_impl_hv(unsigned long cmd)
917 case H_LOGICAL_CI_LOAD:
918 case H_LOGICAL_CI_STORE:
919 #ifdef CONFIG_KVM_XICS
930 /* See if it's in the real-mode table */
931 return kvmppc_hcall_impl_hv_realmode(cmd);
934 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
935 struct kvm_vcpu *vcpu)
939 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
942 * Fetch failed, so return to guest and
943 * try executing it again.
948 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
949 run->exit_reason = KVM_EXIT_DEBUG;
950 run->debug.arch.address = kvmppc_get_pc(vcpu);
953 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
958 static void do_nothing(void *x)
962 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
964 int thr, cpu, pcpu, nthreads;
968 nthreads = vcpu->kvm->arch.emul_smt_mode;
970 cpu = vcpu->vcpu_id & ~(nthreads - 1);
971 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
972 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
976 * If the vcpu is currently running on a physical cpu thread,
977 * interrupt it in order to pull it out of the guest briefly,
978 * which will update its vcore->dpdes value.
980 pcpu = READ_ONCE(v->cpu);
982 smp_call_function_single(pcpu, do_nothing, NULL, 1);
983 if (kvmppc_doorbell_pending(v))
990 * On POWER9, emulate doorbell-related instructions in order to
991 * give the guest the illusion of running on a multi-threaded core.
992 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
995 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
999 struct kvm *kvm = vcpu->kvm;
1000 struct kvm_vcpu *tvcpu;
1002 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1003 return EMULATE_FAIL;
1004 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1005 return RESUME_GUEST;
1006 if (get_op(inst) != 31)
1007 return EMULATE_FAIL;
1009 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1010 switch (get_xop(inst)) {
1011 case OP_31_XOP_MSGSNDP:
1012 arg = kvmppc_get_gpr(vcpu, rb);
1013 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1016 if (arg >= kvm->arch.emul_smt_mode)
1018 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1021 if (!tvcpu->arch.doorbell_request) {
1022 tvcpu->arch.doorbell_request = 1;
1023 kvmppc_fast_vcpu_kick_hv(tvcpu);
1026 case OP_31_XOP_MSGCLRP:
1027 arg = kvmppc_get_gpr(vcpu, rb);
1028 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1030 vcpu->arch.vcore->dpdes = 0;
1031 vcpu->arch.doorbell_request = 0;
1033 case OP_31_XOP_MFSPR:
1034 switch (get_sprn(inst)) {
1039 arg = kvmppc_read_dpdes(vcpu);
1042 return EMULATE_FAIL;
1044 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1047 return EMULATE_FAIL;
1049 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1050 return RESUME_GUEST;
1053 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1054 struct task_struct *tsk)
1056 int r = RESUME_HOST;
1058 vcpu->stat.sum_exits++;
1061 * This can happen if an interrupt occurs in the last stages
1062 * of guest entry or the first stages of guest exit (i.e. after
1063 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1064 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1065 * That can happen due to a bug, or due to a machine check
1066 * occurring at just the wrong time.
1068 if (vcpu->arch.shregs.msr & MSR_HV) {
1069 printk(KERN_EMERG "KVM trap in HV mode!\n");
1070 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1071 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1072 vcpu->arch.shregs.msr);
1073 kvmppc_dump_regs(vcpu);
1074 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1075 run->hw.hardware_exit_reason = vcpu->arch.trap;
1078 run->exit_reason = KVM_EXIT_UNKNOWN;
1079 run->ready_for_interrupt_injection = 1;
1080 switch (vcpu->arch.trap) {
1081 /* We're good on these - the host merely wanted to get our attention */
1082 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1083 vcpu->stat.dec_exits++;
1086 case BOOK3S_INTERRUPT_EXTERNAL:
1087 case BOOK3S_INTERRUPT_H_DOORBELL:
1088 case BOOK3S_INTERRUPT_H_VIRT:
1089 vcpu->stat.ext_intr_exits++;
1092 /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
1093 case BOOK3S_INTERRUPT_HMI:
1094 case BOOK3S_INTERRUPT_PERFMON:
1097 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1098 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1099 run->exit_reason = KVM_EXIT_NMI;
1100 run->hw.hardware_exit_reason = vcpu->arch.trap;
1101 /* Clear out the old NMI status from run->flags */
1102 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1103 /* Now set the NMI status */
1104 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1105 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1107 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1110 /* Print the MCE event to host console. */
1111 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1113 case BOOK3S_INTERRUPT_PROGRAM:
1117 * Normally program interrupts are delivered directly
1118 * to the guest by the hardware, but we can get here
1119 * as a result of a hypervisor emulation interrupt
1120 * (e40) getting turned into a 700 by BML RTAS.
1122 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1123 kvmppc_core_queue_program(vcpu, flags);
1127 case BOOK3S_INTERRUPT_SYSCALL:
1129 /* hcall - punt to userspace */
1132 /* hypercall with MSR_PR has already been handled in rmode,
1133 * and never reaches here.
1136 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1137 for (i = 0; i < 9; ++i)
1138 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1139 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1140 vcpu->arch.hcall_needed = 1;
1145 * We get these next two if the guest accesses a page which it thinks
1146 * it has mapped but which is not actually present, either because
1147 * it is for an emulated I/O device or because the corresonding
1148 * host page has been paged out. Any other HDSI/HISI interrupts
1149 * have been handled already.
1151 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1152 r = RESUME_PAGE_FAULT;
1154 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1155 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1156 vcpu->arch.fault_dsisr = 0;
1157 r = RESUME_PAGE_FAULT;
1160 * This occurs if the guest executes an illegal instruction.
1161 * If the guest debug is disabled, generate a program interrupt
1162 * to the guest. If guest debug is enabled, we need to check
1163 * whether the instruction is a software breakpoint instruction.
1164 * Accordingly return to Guest or Host.
1166 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1167 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1168 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1169 swab32(vcpu->arch.emul_inst) :
1170 vcpu->arch.emul_inst;
1171 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1172 r = kvmppc_emulate_debug_inst(run, vcpu);
1174 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1179 * This occurs if the guest (kernel or userspace), does something that
1180 * is prohibited by HFSCR.
1181 * On POWER9, this could be a doorbell instruction that we need
1183 * Otherwise, we just generate a program interrupt to the guest.
1185 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1187 if ((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG)
1188 r = kvmppc_emulate_doorbell_instr(vcpu);
1189 if (r == EMULATE_FAIL) {
1190 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1194 case BOOK3S_INTERRUPT_HV_RM_HARD:
1195 r = RESUME_PASSTHROUGH;
1198 kvmppc_dump_regs(vcpu);
1199 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1200 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1201 vcpu->arch.shregs.msr);
1202 run->hw.hardware_exit_reason = vcpu->arch.trap;
1210 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1211 struct kvm_sregs *sregs)
1215 memset(sregs, 0, sizeof(struct kvm_sregs));
1216 sregs->pvr = vcpu->arch.pvr;
1217 for (i = 0; i < vcpu->arch.slb_max; i++) {
1218 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1219 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1225 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1226 struct kvm_sregs *sregs)
1230 /* Only accept the same PVR as the host's, since we can't spoof it */
1231 if (sregs->pvr != vcpu->arch.pvr)
1235 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1236 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1237 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1238 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1242 vcpu->arch.slb_max = j;
1247 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1248 bool preserve_top32)
1250 struct kvm *kvm = vcpu->kvm;
1251 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1254 mutex_lock(&kvm->lock);
1255 spin_lock(&vc->lock);
1257 * If ILE (interrupt little-endian) has changed, update the
1258 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1260 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1261 struct kvm_vcpu *vcpu;
1264 kvm_for_each_vcpu(i, vcpu, kvm) {
1265 if (vcpu->arch.vcore != vc)
1267 if (new_lpcr & LPCR_ILE)
1268 vcpu->arch.intr_msr |= MSR_LE;
1270 vcpu->arch.intr_msr &= ~MSR_LE;
1275 * Userspace can only modify DPFD (default prefetch depth),
1276 * ILE (interrupt little-endian) and TC (translation control).
1277 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1279 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1280 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1283 * On POWER9, allow userspace to enable large decrementer for the
1284 * guest, whether or not the host has it enabled.
1286 if (cpu_has_feature(CPU_FTR_ARCH_300))
1289 /* Broken 32-bit version of LPCR must not clear top bits */
1292 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1293 spin_unlock(&vc->lock);
1294 mutex_unlock(&kvm->lock);
1297 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1298 union kvmppc_one_reg *val)
1304 case KVM_REG_PPC_DEBUG_INST:
1305 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1307 case KVM_REG_PPC_HIOR:
1308 *val = get_reg_val(id, 0);
1310 case KVM_REG_PPC_DABR:
1311 *val = get_reg_val(id, vcpu->arch.dabr);
1313 case KVM_REG_PPC_DABRX:
1314 *val = get_reg_val(id, vcpu->arch.dabrx);
1316 case KVM_REG_PPC_DSCR:
1317 *val = get_reg_val(id, vcpu->arch.dscr);
1319 case KVM_REG_PPC_PURR:
1320 *val = get_reg_val(id, vcpu->arch.purr);
1322 case KVM_REG_PPC_SPURR:
1323 *val = get_reg_val(id, vcpu->arch.spurr);
1325 case KVM_REG_PPC_AMR:
1326 *val = get_reg_val(id, vcpu->arch.amr);
1328 case KVM_REG_PPC_UAMOR:
1329 *val = get_reg_val(id, vcpu->arch.uamor);
1331 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1332 i = id - KVM_REG_PPC_MMCR0;
1333 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1335 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1336 i = id - KVM_REG_PPC_PMC1;
1337 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1339 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1340 i = id - KVM_REG_PPC_SPMC1;
1341 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1343 case KVM_REG_PPC_SIAR:
1344 *val = get_reg_val(id, vcpu->arch.siar);
1346 case KVM_REG_PPC_SDAR:
1347 *val = get_reg_val(id, vcpu->arch.sdar);
1349 case KVM_REG_PPC_SIER:
1350 *val = get_reg_val(id, vcpu->arch.sier);
1352 case KVM_REG_PPC_IAMR:
1353 *val = get_reg_val(id, vcpu->arch.iamr);
1355 case KVM_REG_PPC_PSPB:
1356 *val = get_reg_val(id, vcpu->arch.pspb);
1358 case KVM_REG_PPC_DPDES:
1359 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1361 case KVM_REG_PPC_VTB:
1362 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1364 case KVM_REG_PPC_DAWR:
1365 *val = get_reg_val(id, vcpu->arch.dawr);
1367 case KVM_REG_PPC_DAWRX:
1368 *val = get_reg_val(id, vcpu->arch.dawrx);
1370 case KVM_REG_PPC_CIABR:
1371 *val = get_reg_val(id, vcpu->arch.ciabr);
1373 case KVM_REG_PPC_CSIGR:
1374 *val = get_reg_val(id, vcpu->arch.csigr);
1376 case KVM_REG_PPC_TACR:
1377 *val = get_reg_val(id, vcpu->arch.tacr);
1379 case KVM_REG_PPC_TCSCR:
1380 *val = get_reg_val(id, vcpu->arch.tcscr);
1382 case KVM_REG_PPC_PID:
1383 *val = get_reg_val(id, vcpu->arch.pid);
1385 case KVM_REG_PPC_ACOP:
1386 *val = get_reg_val(id, vcpu->arch.acop);
1388 case KVM_REG_PPC_WORT:
1389 *val = get_reg_val(id, vcpu->arch.wort);
1391 case KVM_REG_PPC_TIDR:
1392 *val = get_reg_val(id, vcpu->arch.tid);
1394 case KVM_REG_PPC_PSSCR:
1395 *val = get_reg_val(id, vcpu->arch.psscr);
1397 case KVM_REG_PPC_VPA_ADDR:
1398 spin_lock(&vcpu->arch.vpa_update_lock);
1399 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1400 spin_unlock(&vcpu->arch.vpa_update_lock);
1402 case KVM_REG_PPC_VPA_SLB:
1403 spin_lock(&vcpu->arch.vpa_update_lock);
1404 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1405 val->vpaval.length = vcpu->arch.slb_shadow.len;
1406 spin_unlock(&vcpu->arch.vpa_update_lock);
1408 case KVM_REG_PPC_VPA_DTL:
1409 spin_lock(&vcpu->arch.vpa_update_lock);
1410 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1411 val->vpaval.length = vcpu->arch.dtl.len;
1412 spin_unlock(&vcpu->arch.vpa_update_lock);
1414 case KVM_REG_PPC_TB_OFFSET:
1415 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1417 case KVM_REG_PPC_LPCR:
1418 case KVM_REG_PPC_LPCR_64:
1419 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1421 case KVM_REG_PPC_PPR:
1422 *val = get_reg_val(id, vcpu->arch.ppr);
1424 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1425 case KVM_REG_PPC_TFHAR:
1426 *val = get_reg_val(id, vcpu->arch.tfhar);
1428 case KVM_REG_PPC_TFIAR:
1429 *val = get_reg_val(id, vcpu->arch.tfiar);
1431 case KVM_REG_PPC_TEXASR:
1432 *val = get_reg_val(id, vcpu->arch.texasr);
1434 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1435 i = id - KVM_REG_PPC_TM_GPR0;
1436 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1438 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1441 i = id - KVM_REG_PPC_TM_VSR0;
1443 for (j = 0; j < TS_FPRWIDTH; j++)
1444 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1446 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1447 val->vval = vcpu->arch.vr_tm.vr[i-32];
1453 case KVM_REG_PPC_TM_CR:
1454 *val = get_reg_val(id, vcpu->arch.cr_tm);
1456 case KVM_REG_PPC_TM_XER:
1457 *val = get_reg_val(id, vcpu->arch.xer_tm);
1459 case KVM_REG_PPC_TM_LR:
1460 *val = get_reg_val(id, vcpu->arch.lr_tm);
1462 case KVM_REG_PPC_TM_CTR:
1463 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1465 case KVM_REG_PPC_TM_FPSCR:
1466 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1468 case KVM_REG_PPC_TM_AMR:
1469 *val = get_reg_val(id, vcpu->arch.amr_tm);
1471 case KVM_REG_PPC_TM_PPR:
1472 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1474 case KVM_REG_PPC_TM_VRSAVE:
1475 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1477 case KVM_REG_PPC_TM_VSCR:
1478 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1479 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1483 case KVM_REG_PPC_TM_DSCR:
1484 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1486 case KVM_REG_PPC_TM_TAR:
1487 *val = get_reg_val(id, vcpu->arch.tar_tm);
1490 case KVM_REG_PPC_ARCH_COMPAT:
1491 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1501 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1502 union kvmppc_one_reg *val)
1506 unsigned long addr, len;
1509 case KVM_REG_PPC_HIOR:
1510 /* Only allow this to be set to zero */
1511 if (set_reg_val(id, *val))
1514 case KVM_REG_PPC_DABR:
1515 vcpu->arch.dabr = set_reg_val(id, *val);
1517 case KVM_REG_PPC_DABRX:
1518 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1520 case KVM_REG_PPC_DSCR:
1521 vcpu->arch.dscr = set_reg_val(id, *val);
1523 case KVM_REG_PPC_PURR:
1524 vcpu->arch.purr = set_reg_val(id, *val);
1526 case KVM_REG_PPC_SPURR:
1527 vcpu->arch.spurr = set_reg_val(id, *val);
1529 case KVM_REG_PPC_AMR:
1530 vcpu->arch.amr = set_reg_val(id, *val);
1532 case KVM_REG_PPC_UAMOR:
1533 vcpu->arch.uamor = set_reg_val(id, *val);
1535 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1536 i = id - KVM_REG_PPC_MMCR0;
1537 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1539 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1540 i = id - KVM_REG_PPC_PMC1;
1541 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1543 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1544 i = id - KVM_REG_PPC_SPMC1;
1545 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1547 case KVM_REG_PPC_SIAR:
1548 vcpu->arch.siar = set_reg_val(id, *val);
1550 case KVM_REG_PPC_SDAR:
1551 vcpu->arch.sdar = set_reg_val(id, *val);
1553 case KVM_REG_PPC_SIER:
1554 vcpu->arch.sier = set_reg_val(id, *val);
1556 case KVM_REG_PPC_IAMR:
1557 vcpu->arch.iamr = set_reg_val(id, *val);
1559 case KVM_REG_PPC_PSPB:
1560 vcpu->arch.pspb = set_reg_val(id, *val);
1562 case KVM_REG_PPC_DPDES:
1563 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1565 case KVM_REG_PPC_VTB:
1566 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1568 case KVM_REG_PPC_DAWR:
1569 vcpu->arch.dawr = set_reg_val(id, *val);
1571 case KVM_REG_PPC_DAWRX:
1572 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1574 case KVM_REG_PPC_CIABR:
1575 vcpu->arch.ciabr = set_reg_val(id, *val);
1576 /* Don't allow setting breakpoints in hypervisor code */
1577 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1578 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1580 case KVM_REG_PPC_CSIGR:
1581 vcpu->arch.csigr = set_reg_val(id, *val);
1583 case KVM_REG_PPC_TACR:
1584 vcpu->arch.tacr = set_reg_val(id, *val);
1586 case KVM_REG_PPC_TCSCR:
1587 vcpu->arch.tcscr = set_reg_val(id, *val);
1589 case KVM_REG_PPC_PID:
1590 vcpu->arch.pid = set_reg_val(id, *val);
1592 case KVM_REG_PPC_ACOP:
1593 vcpu->arch.acop = set_reg_val(id, *val);
1595 case KVM_REG_PPC_WORT:
1596 vcpu->arch.wort = set_reg_val(id, *val);
1598 case KVM_REG_PPC_TIDR:
1599 vcpu->arch.tid = set_reg_val(id, *val);
1601 case KVM_REG_PPC_PSSCR:
1602 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1604 case KVM_REG_PPC_VPA_ADDR:
1605 addr = set_reg_val(id, *val);
1607 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1608 vcpu->arch.dtl.next_gpa))
1610 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1612 case KVM_REG_PPC_VPA_SLB:
1613 addr = val->vpaval.addr;
1614 len = val->vpaval.length;
1616 if (addr && !vcpu->arch.vpa.next_gpa)
1618 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1620 case KVM_REG_PPC_VPA_DTL:
1621 addr = val->vpaval.addr;
1622 len = val->vpaval.length;
1624 if (addr && (len < sizeof(struct dtl_entry) ||
1625 !vcpu->arch.vpa.next_gpa))
1627 len -= len % sizeof(struct dtl_entry);
1628 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1630 case KVM_REG_PPC_TB_OFFSET:
1632 * POWER9 DD1 has an erratum where writing TBU40 causes
1633 * the timebase to lose ticks. So we don't let the
1634 * timebase offset be changed on P9 DD1. (It is
1635 * initialized to zero.)
1637 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1639 /* round up to multiple of 2^24 */
1640 vcpu->arch.vcore->tb_offset =
1641 ALIGN(set_reg_val(id, *val), 1UL << 24);
1643 case KVM_REG_PPC_LPCR:
1644 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1646 case KVM_REG_PPC_LPCR_64:
1647 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1649 case KVM_REG_PPC_PPR:
1650 vcpu->arch.ppr = set_reg_val(id, *val);
1652 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1653 case KVM_REG_PPC_TFHAR:
1654 vcpu->arch.tfhar = set_reg_val(id, *val);
1656 case KVM_REG_PPC_TFIAR:
1657 vcpu->arch.tfiar = set_reg_val(id, *val);
1659 case KVM_REG_PPC_TEXASR:
1660 vcpu->arch.texasr = set_reg_val(id, *val);
1662 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1663 i = id - KVM_REG_PPC_TM_GPR0;
1664 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1666 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1669 i = id - KVM_REG_PPC_TM_VSR0;
1671 for (j = 0; j < TS_FPRWIDTH; j++)
1672 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1674 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1675 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1680 case KVM_REG_PPC_TM_CR:
1681 vcpu->arch.cr_tm = set_reg_val(id, *val);
1683 case KVM_REG_PPC_TM_XER:
1684 vcpu->arch.xer_tm = set_reg_val(id, *val);
1686 case KVM_REG_PPC_TM_LR:
1687 vcpu->arch.lr_tm = set_reg_val(id, *val);
1689 case KVM_REG_PPC_TM_CTR:
1690 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1692 case KVM_REG_PPC_TM_FPSCR:
1693 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1695 case KVM_REG_PPC_TM_AMR:
1696 vcpu->arch.amr_tm = set_reg_val(id, *val);
1698 case KVM_REG_PPC_TM_PPR:
1699 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1701 case KVM_REG_PPC_TM_VRSAVE:
1702 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1704 case KVM_REG_PPC_TM_VSCR:
1705 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1706 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1710 case KVM_REG_PPC_TM_DSCR:
1711 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1713 case KVM_REG_PPC_TM_TAR:
1714 vcpu->arch.tar_tm = set_reg_val(id, *val);
1717 case KVM_REG_PPC_ARCH_COMPAT:
1718 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1729 * On POWER9, threads are independent and can be in different partitions.
1730 * Therefore we consider each thread to be a subcore.
1731 * There is a restriction that all threads have to be in the same
1732 * MMU mode (radix or HPT), unfortunately, but since we only support
1733 * HPT guests on a HPT host so far, that isn't an impediment yet.
1735 static int threads_per_vcore(void)
1737 if (cpu_has_feature(CPU_FTR_ARCH_300))
1739 return threads_per_subcore;
1742 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1744 struct kvmppc_vcore *vcore;
1746 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1751 spin_lock_init(&vcore->lock);
1752 spin_lock_init(&vcore->stoltb_lock);
1753 init_swait_queue_head(&vcore->wq);
1754 vcore->preempt_tb = TB_NIL;
1755 vcore->lpcr = kvm->arch.lpcr;
1756 vcore->first_vcpuid = core * kvm->arch.smt_mode;
1758 INIT_LIST_HEAD(&vcore->preempt_list);
1763 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1764 static struct debugfs_timings_element {
1768 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1769 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1770 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1771 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1772 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1775 #define N_TIMINGS (sizeof(timings) / sizeof(timings[0]))
1777 struct debugfs_timings_state {
1778 struct kvm_vcpu *vcpu;
1779 unsigned int buflen;
1780 char buf[N_TIMINGS * 100];
1783 static int debugfs_timings_open(struct inode *inode, struct file *file)
1785 struct kvm_vcpu *vcpu = inode->i_private;
1786 struct debugfs_timings_state *p;
1788 p = kzalloc(sizeof(*p), GFP_KERNEL);
1792 kvm_get_kvm(vcpu->kvm);
1794 file->private_data = p;
1796 return nonseekable_open(inode, file);
1799 static int debugfs_timings_release(struct inode *inode, struct file *file)
1801 struct debugfs_timings_state *p = file->private_data;
1803 kvm_put_kvm(p->vcpu->kvm);
1808 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1809 size_t len, loff_t *ppos)
1811 struct debugfs_timings_state *p = file->private_data;
1812 struct kvm_vcpu *vcpu = p->vcpu;
1814 struct kvmhv_tb_accumulator tb;
1823 buf_end = s + sizeof(p->buf);
1824 for (i = 0; i < N_TIMINGS; ++i) {
1825 struct kvmhv_tb_accumulator *acc;
1827 acc = (struct kvmhv_tb_accumulator *)
1828 ((unsigned long)vcpu + timings[i].offset);
1830 for (loops = 0; loops < 1000; ++loops) {
1831 count = acc->seqcount;
1836 if (count == acc->seqcount) {
1844 snprintf(s, buf_end - s, "%s: stuck\n",
1847 snprintf(s, buf_end - s,
1848 "%s: %llu %llu %llu %llu\n",
1849 timings[i].name, count / 2,
1850 tb_to_ns(tb.tb_total),
1851 tb_to_ns(tb.tb_min),
1852 tb_to_ns(tb.tb_max));
1855 p->buflen = s - p->buf;
1859 if (pos >= p->buflen)
1861 if (len > p->buflen - pos)
1862 len = p->buflen - pos;
1863 n = copy_to_user(buf, p->buf + pos, len);
1873 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1874 size_t len, loff_t *ppos)
1879 static const struct file_operations debugfs_timings_ops = {
1880 .owner = THIS_MODULE,
1881 .open = debugfs_timings_open,
1882 .release = debugfs_timings_release,
1883 .read = debugfs_timings_read,
1884 .write = debugfs_timings_write,
1885 .llseek = generic_file_llseek,
1888 /* Create a debugfs directory for the vcpu */
1889 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1892 struct kvm *kvm = vcpu->kvm;
1894 snprintf(buf, sizeof(buf), "vcpu%u", id);
1895 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1897 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1898 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1900 vcpu->arch.debugfs_timings =
1901 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1902 vcpu, &debugfs_timings_ops);
1905 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1906 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1909 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1911 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1914 struct kvm_vcpu *vcpu;
1917 struct kvmppc_vcore *vcore;
1920 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1924 err = kvm_vcpu_init(vcpu, kvm, id);
1928 vcpu->arch.shared = &vcpu->arch.shregs;
1929 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1931 * The shared struct is never shared on HV,
1932 * so we can always use host endianness
1934 #ifdef __BIG_ENDIAN__
1935 vcpu->arch.shared_big_endian = true;
1937 vcpu->arch.shared_big_endian = false;
1940 vcpu->arch.mmcr[0] = MMCR0_FC;
1941 vcpu->arch.ctrl = CTRL_RUNLATCH;
1942 /* default to host PVR, since we can't spoof it */
1943 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1944 spin_lock_init(&vcpu->arch.vpa_update_lock);
1945 spin_lock_init(&vcpu->arch.tbacct_lock);
1946 vcpu->arch.busy_preempt = TB_NIL;
1947 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1950 * Set the default HFSCR for the guest from the host value.
1951 * This value is only used on POWER9.
1952 * On POWER9 DD1, TM doesn't work, so we make sure to
1953 * prevent the guest from using it.
1954 * On POWER9, we want to virtualize the doorbell facility, so we
1955 * turn off the HFSCR bit, which causes those instructions to trap.
1957 vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1958 if (!cpu_has_feature(CPU_FTR_TM))
1959 vcpu->arch.hfscr &= ~HFSCR_TM;
1960 if (cpu_has_feature(CPU_FTR_ARCH_300))
1961 vcpu->arch.hfscr &= ~HFSCR_MSGP;
1963 kvmppc_mmu_book3s_hv_init(vcpu);
1965 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1967 init_waitqueue_head(&vcpu->arch.cpu_run);
1969 mutex_lock(&kvm->lock);
1972 core = id / kvm->arch.smt_mode;
1973 if (core < KVM_MAX_VCORES) {
1974 vcore = kvm->arch.vcores[core];
1977 vcore = kvmppc_vcore_create(kvm, core);
1978 kvm->arch.vcores[core] = vcore;
1979 kvm->arch.online_vcores++;
1982 mutex_unlock(&kvm->lock);
1987 spin_lock(&vcore->lock);
1988 ++vcore->num_threads;
1989 spin_unlock(&vcore->lock);
1990 vcpu->arch.vcore = vcore;
1991 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1992 vcpu->arch.thread_cpu = -1;
1993 vcpu->arch.prev_cpu = -1;
1995 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1996 kvmppc_sanity_check(vcpu);
1998 debugfs_vcpu_init(vcpu, id);
2003 kmem_cache_free(kvm_vcpu_cache, vcpu);
2005 return ERR_PTR(err);
2008 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2009 unsigned long flags)
2016 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2018 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2020 * On POWER8 (or POWER7), the threading mode is "strict",
2021 * so we pack smt_mode vcpus per vcore.
2023 if (smt_mode > threads_per_subcore)
2027 * On POWER9, the threading mode is "loose",
2028 * so each vcpu gets its own vcore.
2033 mutex_lock(&kvm->lock);
2035 if (!kvm->arch.online_vcores) {
2036 kvm->arch.smt_mode = smt_mode;
2037 kvm->arch.emul_smt_mode = esmt;
2040 mutex_unlock(&kvm->lock);
2045 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2047 if (vpa->pinned_addr)
2048 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2052 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2054 spin_lock(&vcpu->arch.vpa_update_lock);
2055 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2056 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2057 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2058 spin_unlock(&vcpu->arch.vpa_update_lock);
2059 kvm_vcpu_uninit(vcpu);
2060 kmem_cache_free(kvm_vcpu_cache, vcpu);
2063 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2065 /* Indicate we want to get back into the guest */
2069 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2071 unsigned long dec_nsec, now;
2074 if (now > vcpu->arch.dec_expires) {
2075 /* decrementer has already gone negative */
2076 kvmppc_core_queue_dec(vcpu);
2077 kvmppc_core_prepare_to_enter(vcpu);
2080 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2082 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2083 vcpu->arch.timer_running = 1;
2086 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2088 vcpu->arch.ceded = 0;
2089 if (vcpu->arch.timer_running) {
2090 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2091 vcpu->arch.timer_running = 0;
2095 extern int __kvmppc_vcore_entry(void);
2097 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2098 struct kvm_vcpu *vcpu)
2102 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2104 spin_lock_irq(&vcpu->arch.tbacct_lock);
2106 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2107 vcpu->arch.stolen_logged;
2108 vcpu->arch.busy_preempt = now;
2109 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2110 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2112 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2115 static int kvmppc_grab_hwthread(int cpu)
2117 struct paca_struct *tpaca;
2118 long timeout = 10000;
2121 * ISA v3.0 idle routines do not set hwthread_state or test
2122 * hwthread_req, so they can not grab idle threads.
2124 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2125 WARN(1, "KVM: can not control sibling threads\n");
2131 /* Ensure the thread won't go into the kernel if it wakes */
2132 tpaca->kvm_hstate.kvm_vcpu = NULL;
2133 tpaca->kvm_hstate.kvm_vcore = NULL;
2134 tpaca->kvm_hstate.napping = 0;
2136 tpaca->kvm_hstate.hwthread_req = 1;
2139 * If the thread is already executing in the kernel (e.g. handling
2140 * a stray interrupt), wait for it to get back to nap mode.
2141 * The smp_mb() is to ensure that our setting of hwthread_req
2142 * is visible before we look at hwthread_state, so if this
2143 * races with the code at system_reset_pSeries and the thread
2144 * misses our setting of hwthread_req, we are sure to see its
2145 * setting of hwthread_state, and vice versa.
2148 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2149 if (--timeout <= 0) {
2150 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2158 static void kvmppc_release_hwthread(int cpu)
2160 struct paca_struct *tpaca;
2163 tpaca->kvm_hstate.kvm_vcpu = NULL;
2164 tpaca->kvm_hstate.kvm_vcore = NULL;
2165 tpaca->kvm_hstate.kvm_split_mode = NULL;
2166 if (!cpu_has_feature(CPU_FTR_ARCH_300))
2167 tpaca->kvm_hstate.hwthread_req = 0;
2171 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2175 cpu = cpu_first_thread_sibling(cpu);
2176 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2178 * Make sure setting of bit in need_tlb_flush precedes
2179 * testing of cpu_in_guest bits. The matching barrier on
2180 * the other side is the first smp_mb() in kvmppc_run_core().
2183 for (i = 0; i < threads_per_core; ++i)
2184 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2185 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2188 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2190 struct kvm *kvm = vcpu->kvm;
2193 * With radix, the guest can do TLB invalidations itself,
2194 * and it could choose to use the local form (tlbiel) if
2195 * it is invalidating a translation that has only ever been
2196 * used on one vcpu. However, that doesn't mean it has
2197 * only ever been used on one physical cpu, since vcpus
2198 * can move around between pcpus. To cope with this, when
2199 * a vcpu moves from one pcpu to another, we need to tell
2200 * any vcpus running on the same core as this vcpu previously
2201 * ran to flush the TLB. The TLB is shared between threads,
2202 * so we use a single bit in .need_tlb_flush for all 4 threads.
2204 if (vcpu->arch.prev_cpu != pcpu) {
2205 if (vcpu->arch.prev_cpu >= 0 &&
2206 cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2207 cpu_first_thread_sibling(pcpu))
2208 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2209 vcpu->arch.prev_cpu = pcpu;
2213 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2216 struct paca_struct *tpaca;
2217 struct kvm *kvm = vc->kvm;
2221 if (vcpu->arch.timer_running) {
2222 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2223 vcpu->arch.timer_running = 0;
2225 cpu += vcpu->arch.ptid;
2226 vcpu->cpu = vc->pcpu;
2227 vcpu->arch.thread_cpu = cpu;
2228 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2231 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2232 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2233 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2235 tpaca->kvm_hstate.kvm_vcore = vc;
2236 if (cpu != smp_processor_id())
2237 kvmppc_ipi_thread(cpu);
2240 static void kvmppc_wait_for_nap(void)
2242 int cpu = smp_processor_id();
2244 int n_threads = threads_per_vcore();
2248 for (loops = 0; loops < 1000000; ++loops) {
2250 * Check if all threads are finished.
2251 * We set the vcore pointer when starting a thread
2252 * and the thread clears it when finished, so we look
2253 * for any threads that still have a non-NULL vcore ptr.
2255 for (i = 1; i < n_threads; ++i)
2256 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2258 if (i == n_threads) {
2265 for (i = 1; i < n_threads; ++i)
2266 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2267 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2271 * Check that we are on thread 0 and that any other threads in
2272 * this core are off-line. Then grab the threads so they can't
2275 static int on_primary_thread(void)
2277 int cpu = smp_processor_id();
2280 /* Are we on a primary subcore? */
2281 if (cpu_thread_in_subcore(cpu))
2285 while (++thr < threads_per_subcore)
2286 if (cpu_online(cpu + thr))
2289 /* Grab all hw threads so they can't go into the kernel */
2290 for (thr = 1; thr < threads_per_subcore; ++thr) {
2291 if (kvmppc_grab_hwthread(cpu + thr)) {
2292 /* Couldn't grab one; let the others go */
2294 kvmppc_release_hwthread(cpu + thr);
2295 } while (--thr > 0);
2303 * A list of virtual cores for each physical CPU.
2304 * These are vcores that could run but their runner VCPU tasks are
2305 * (or may be) preempted.
2307 struct preempted_vcore_list {
2308 struct list_head list;
2312 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2314 static void init_vcore_lists(void)
2318 for_each_possible_cpu(cpu) {
2319 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2320 spin_lock_init(&lp->lock);
2321 INIT_LIST_HEAD(&lp->list);
2325 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2327 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2329 vc->vcore_state = VCORE_PREEMPT;
2330 vc->pcpu = smp_processor_id();
2331 if (vc->num_threads < threads_per_vcore()) {
2332 spin_lock(&lp->lock);
2333 list_add_tail(&vc->preempt_list, &lp->list);
2334 spin_unlock(&lp->lock);
2337 /* Start accumulating stolen time */
2338 kvmppc_core_start_stolen(vc);
2341 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2343 struct preempted_vcore_list *lp;
2345 kvmppc_core_end_stolen(vc);
2346 if (!list_empty(&vc->preempt_list)) {
2347 lp = &per_cpu(preempted_vcores, vc->pcpu);
2348 spin_lock(&lp->lock);
2349 list_del_init(&vc->preempt_list);
2350 spin_unlock(&lp->lock);
2352 vc->vcore_state = VCORE_INACTIVE;
2356 * This stores information about the virtual cores currently
2357 * assigned to a physical core.
2361 int max_subcore_threads;
2363 int subcore_threads[MAX_SUBCORES];
2364 struct kvmppc_vcore *vc[MAX_SUBCORES];
2368 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2369 * respectively in 2-way micro-threading (split-core) mode.
2371 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2373 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2375 memset(cip, 0, sizeof(*cip));
2376 cip->n_subcores = 1;
2377 cip->max_subcore_threads = vc->num_threads;
2378 cip->total_threads = vc->num_threads;
2379 cip->subcore_threads[0] = vc->num_threads;
2383 static bool subcore_config_ok(int n_subcores, int n_threads)
2385 /* Can only dynamically split if unsplit to begin with */
2386 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2388 if (n_subcores > MAX_SUBCORES)
2390 if (n_subcores > 1) {
2391 if (!(dynamic_mt_modes & 2))
2393 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2397 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2400 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2402 vc->entry_exit_map = 0;
2404 vc->napping_threads = 0;
2405 vc->conferring_threads = 0;
2408 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2410 int n_threads = vc->num_threads;
2413 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2416 if (n_threads < cip->max_subcore_threads)
2417 n_threads = cip->max_subcore_threads;
2418 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2420 cip->max_subcore_threads = n_threads;
2422 sub = cip->n_subcores;
2424 cip->total_threads += vc->num_threads;
2425 cip->subcore_threads[sub] = vc->num_threads;
2427 init_vcore_to_run(vc);
2428 list_del_init(&vc->preempt_list);
2434 * Work out whether it is possible to piggyback the execution of
2435 * vcore *pvc onto the execution of the other vcores described in *cip.
2437 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2440 if (cip->total_threads + pvc->num_threads > target_threads)
2443 return can_dynamic_split(pvc, cip);
2446 static void prepare_threads(struct kvmppc_vcore *vc)
2449 struct kvm_vcpu *vcpu;
2451 for_each_runnable_thread(i, vcpu, vc) {
2452 if (signal_pending(vcpu->arch.run_task))
2453 vcpu->arch.ret = -EINTR;
2454 else if (vcpu->arch.vpa.update_pending ||
2455 vcpu->arch.slb_shadow.update_pending ||
2456 vcpu->arch.dtl.update_pending)
2457 vcpu->arch.ret = RESUME_GUEST;
2460 kvmppc_remove_runnable(vc, vcpu);
2461 wake_up(&vcpu->arch.cpu_run);
2465 static void collect_piggybacks(struct core_info *cip, int target_threads)
2467 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2468 struct kvmppc_vcore *pvc, *vcnext;
2470 spin_lock(&lp->lock);
2471 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2472 if (!spin_trylock(&pvc->lock))
2474 prepare_threads(pvc);
2475 if (!pvc->n_runnable) {
2476 list_del_init(&pvc->preempt_list);
2477 if (pvc->runner == NULL) {
2478 pvc->vcore_state = VCORE_INACTIVE;
2479 kvmppc_core_end_stolen(pvc);
2481 spin_unlock(&pvc->lock);
2484 if (!can_piggyback(pvc, cip, target_threads)) {
2485 spin_unlock(&pvc->lock);
2488 kvmppc_core_end_stolen(pvc);
2489 pvc->vcore_state = VCORE_PIGGYBACK;
2490 if (cip->total_threads >= target_threads)
2493 spin_unlock(&lp->lock);
2496 static bool recheck_signals(struct core_info *cip)
2499 struct kvm_vcpu *vcpu;
2501 for (sub = 0; sub < cip->n_subcores; ++sub)
2502 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2503 if (signal_pending(vcpu->arch.run_task))
2508 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2510 int still_running = 0, i;
2513 struct kvm_vcpu *vcpu;
2515 spin_lock(&vc->lock);
2517 for_each_runnable_thread(i, vcpu, vc) {
2518 /* cancel pending dec exception if dec is positive */
2519 if (now < vcpu->arch.dec_expires &&
2520 kvmppc_core_pending_dec(vcpu))
2521 kvmppc_core_dequeue_dec(vcpu);
2523 trace_kvm_guest_exit(vcpu);
2526 if (vcpu->arch.trap)
2527 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2528 vcpu->arch.run_task);
2530 vcpu->arch.ret = ret;
2531 vcpu->arch.trap = 0;
2533 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2534 if (vcpu->arch.pending_exceptions)
2535 kvmppc_core_prepare_to_enter(vcpu);
2536 if (vcpu->arch.ceded)
2537 kvmppc_set_timer(vcpu);
2541 kvmppc_remove_runnable(vc, vcpu);
2542 wake_up(&vcpu->arch.cpu_run);
2546 if (still_running > 0) {
2547 kvmppc_vcore_preempt(vc);
2548 } else if (vc->runner) {
2549 vc->vcore_state = VCORE_PREEMPT;
2550 kvmppc_core_start_stolen(vc);
2552 vc->vcore_state = VCORE_INACTIVE;
2554 if (vc->n_runnable > 0 && vc->runner == NULL) {
2555 /* make sure there's a candidate runner awake */
2557 vcpu = next_runnable_thread(vc, &i);
2558 wake_up(&vcpu->arch.cpu_run);
2561 spin_unlock(&vc->lock);
2565 * Clear core from the list of active host cores as we are about to
2566 * enter the guest. Only do this if it is the primary thread of the
2567 * core (not if a subcore) that is entering the guest.
2569 static inline int kvmppc_clear_host_core(unsigned int cpu)
2573 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2576 * Memory barrier can be omitted here as we will do a smp_wmb()
2577 * later in kvmppc_start_thread and we need ensure that state is
2578 * visible to other CPUs only after we enter guest.
2580 core = cpu >> threads_shift;
2581 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2586 * Advertise this core as an active host core since we exited the guest
2587 * Only need to do this if it is the primary thread of the core that is
2590 static inline int kvmppc_set_host_core(unsigned int cpu)
2594 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2598 * Memory barrier can be omitted here because we do a spin_unlock
2599 * immediately after this which provides the memory barrier.
2601 core = cpu >> threads_shift;
2602 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2606 static void set_irq_happened(int trap)
2609 case BOOK3S_INTERRUPT_EXTERNAL:
2610 local_paca->irq_happened |= PACA_IRQ_EE;
2612 case BOOK3S_INTERRUPT_H_DOORBELL:
2613 local_paca->irq_happened |= PACA_IRQ_DBELL;
2615 case BOOK3S_INTERRUPT_HMI:
2616 local_paca->irq_happened |= PACA_IRQ_HMI;
2622 * Run a set of guest threads on a physical core.
2623 * Called with vc->lock held.
2625 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2627 struct kvm_vcpu *vcpu;
2630 struct core_info core_info;
2631 struct kvmppc_vcore *pvc;
2632 struct kvm_split_mode split_info, *sip;
2633 int split, subcore_size, active;
2636 unsigned long cmd_bit, stat_bit;
2639 int controlled_threads;
2643 * Remove from the list any threads that have a signal pending
2644 * or need a VPA update done
2646 prepare_threads(vc);
2648 /* if the runner is no longer runnable, let the caller pick a new one */
2649 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2655 init_vcore_to_run(vc);
2656 vc->preempt_tb = TB_NIL;
2659 * Number of threads that we will be controlling: the same as
2660 * the number of threads per subcore, except on POWER9,
2661 * where it's 1 because the threads are (mostly) independent.
2663 controlled_threads = threads_per_vcore();
2666 * Make sure we are running on primary threads, and that secondary
2667 * threads are offline. Also check if the number of threads in this
2668 * guest are greater than the current system threads per guest.
2670 if ((controlled_threads > 1) &&
2671 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2672 for_each_runnable_thread(i, vcpu, vc) {
2673 vcpu->arch.ret = -EBUSY;
2674 kvmppc_remove_runnable(vc, vcpu);
2675 wake_up(&vcpu->arch.cpu_run);
2681 * See if we could run any other vcores on the physical core
2682 * along with this one.
2684 init_core_info(&core_info, vc);
2685 pcpu = smp_processor_id();
2686 target_threads = controlled_threads;
2687 if (target_smt_mode && target_smt_mode < target_threads)
2688 target_threads = target_smt_mode;
2689 if (vc->num_threads < target_threads)
2690 collect_piggybacks(&core_info, target_threads);
2693 * On radix, arrange for TLB flushing if necessary.
2694 * This has to be done before disabling interrupts since
2695 * it uses smp_call_function().
2697 pcpu = smp_processor_id();
2698 if (kvm_is_radix(vc->kvm)) {
2699 for (sub = 0; sub < core_info.n_subcores; ++sub)
2700 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2701 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2705 * Hard-disable interrupts, and check resched flag and signals.
2706 * If we need to reschedule or deliver a signal, clean up
2707 * and return without going into the guest(s).
2708 * If the hpte_setup_done flag has been cleared, don't go into the
2709 * guest because that means a HPT resize operation is in progress.
2711 local_irq_disable();
2713 if (lazy_irq_pending() || need_resched() ||
2714 recheck_signals(&core_info) ||
2715 (!kvm_is_radix(vc->kvm) && !vc->kvm->arch.hpte_setup_done)) {
2717 vc->vcore_state = VCORE_INACTIVE;
2718 /* Unlock all except the primary vcore */
2719 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2720 pvc = core_info.vc[sub];
2721 /* Put back on to the preempted vcores list */
2722 kvmppc_vcore_preempt(pvc);
2723 spin_unlock(&pvc->lock);
2725 for (i = 0; i < controlled_threads; ++i)
2726 kvmppc_release_hwthread(pcpu + i);
2730 kvmppc_clear_host_core(pcpu);
2732 /* Decide on micro-threading (split-core) mode */
2733 subcore_size = threads_per_subcore;
2734 cmd_bit = stat_bit = 0;
2735 split = core_info.n_subcores;
2738 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2739 if (split == 2 && (dynamic_mt_modes & 2)) {
2740 cmd_bit = HID0_POWER8_1TO2LPAR;
2741 stat_bit = HID0_POWER8_2LPARMODE;
2744 cmd_bit = HID0_POWER8_1TO4LPAR;
2745 stat_bit = HID0_POWER8_4LPARMODE;
2747 subcore_size = MAX_SMT_THREADS / split;
2749 memset(&split_info, 0, sizeof(split_info));
2750 split_info.rpr = mfspr(SPRN_RPR);
2751 split_info.pmmar = mfspr(SPRN_PMMAR);
2752 split_info.ldbar = mfspr(SPRN_LDBAR);
2753 split_info.subcore_size = subcore_size;
2754 for (sub = 0; sub < core_info.n_subcores; ++sub)
2755 split_info.vc[sub] = core_info.vc[sub];
2756 /* order writes to split_info before kvm_split_mode pointer */
2759 for (thr = 0; thr < controlled_threads; ++thr)
2760 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2762 /* Initiate micro-threading (split-core) if required */
2764 unsigned long hid0 = mfspr(SPRN_HID0);
2766 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2768 mtspr(SPRN_HID0, hid0);
2771 hid0 = mfspr(SPRN_HID0);
2772 if (hid0 & stat_bit)
2778 /* Start all the threads */
2780 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2781 thr = subcore_thread_map[sub];
2784 pvc = core_info.vc[sub];
2785 pvc->pcpu = pcpu + thr;
2786 for_each_runnable_thread(i, vcpu, pvc) {
2787 kvmppc_start_thread(vcpu, pvc);
2788 kvmppc_create_dtl_entry(vcpu, pvc);
2789 trace_kvm_guest_enter(vcpu);
2790 if (!vcpu->arch.ptid)
2792 active |= 1 << (thr + vcpu->arch.ptid);
2795 * We need to start the first thread of each subcore
2796 * even if it doesn't have a vcpu.
2799 kvmppc_start_thread(NULL, pvc);
2800 thr += pvc->num_threads;
2804 * Ensure that split_info.do_nap is set after setting
2805 * the vcore pointer in the PACA of the secondaries.
2809 split_info.do_nap = 1; /* ask secondaries to nap when done */
2812 * When doing micro-threading, poke the inactive threads as well.
2813 * This gets them to the nap instruction after kvm_do_nap,
2814 * which reduces the time taken to unsplit later.
2817 for (thr = 1; thr < threads_per_subcore; ++thr)
2818 if (!(active & (1 << thr)))
2819 kvmppc_ipi_thread(pcpu + thr);
2821 vc->vcore_state = VCORE_RUNNING;
2824 trace_kvmppc_run_core(vc, 0);
2826 for (sub = 0; sub < core_info.n_subcores; ++sub)
2827 spin_unlock(&core_info.vc[sub]->lock);
2830 * Interrupts will be enabled once we get into the guest,
2831 * so tell lockdep that we're about to enable interrupts.
2833 trace_hardirqs_on();
2837 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2839 trap = __kvmppc_vcore_entry();
2841 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2845 trace_hardirqs_off();
2846 set_irq_happened(trap);
2848 spin_lock(&vc->lock);
2849 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2850 vc->vcore_state = VCORE_EXITING;
2852 /* wait for secondary threads to finish writing their state to memory */
2853 kvmppc_wait_for_nap();
2855 /* Return to whole-core mode if we split the core earlier */
2857 unsigned long hid0 = mfspr(SPRN_HID0);
2858 unsigned long loops = 0;
2860 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2861 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2863 mtspr(SPRN_HID0, hid0);
2866 hid0 = mfspr(SPRN_HID0);
2867 if (!(hid0 & stat_bit))
2872 split_info.do_nap = 0;
2875 kvmppc_set_host_core(pcpu);
2879 /* Let secondaries go back to the offline loop */
2880 for (i = 0; i < controlled_threads; ++i) {
2881 kvmppc_release_hwthread(pcpu + i);
2882 if (sip && sip->napped[i])
2883 kvmppc_ipi_thread(pcpu + i);
2884 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2887 spin_unlock(&vc->lock);
2889 /* make sure updates to secondary vcpu structs are visible now */
2892 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2893 pvc = core_info.vc[sub];
2894 post_guest_process(pvc, pvc == vc);
2897 spin_lock(&vc->lock);
2901 vc->vcore_state = VCORE_INACTIVE;
2902 trace_kvmppc_run_core(vc, 1);
2906 * Wait for some other vcpu thread to execute us, and
2907 * wake us up when we need to handle something in the host.
2909 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2910 struct kvm_vcpu *vcpu, int wait_state)
2914 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2915 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2916 spin_unlock(&vc->lock);
2918 spin_lock(&vc->lock);
2920 finish_wait(&vcpu->arch.cpu_run, &wait);
2923 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2926 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2927 vc->halt_poll_ns = 10000;
2929 vc->halt_poll_ns *= halt_poll_ns_grow;
2932 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2934 if (halt_poll_ns_shrink == 0)
2935 vc->halt_poll_ns = 0;
2937 vc->halt_poll_ns /= halt_poll_ns_shrink;
2940 #ifdef CONFIG_KVM_XICS
2941 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2943 if (!xive_enabled())
2945 return vcpu->arch.xive_saved_state.pipr <
2946 vcpu->arch.xive_saved_state.cppr;
2949 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2953 #endif /* CONFIG_KVM_XICS */
2955 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
2957 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
2958 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
2965 * Check to see if any of the runnable vcpus on the vcore have pending
2966 * exceptions or are no longer ceded
2968 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2970 struct kvm_vcpu *vcpu;
2973 for_each_runnable_thread(i, vcpu, vc) {
2974 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
2982 * All the vcpus in this vcore are idle, so wait for a decrementer
2983 * or external interrupt to one of the vcpus. vc->lock is held.
2985 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2987 ktime_t cur, start_poll, start_wait;
2990 DECLARE_SWAITQUEUE(wait);
2992 /* Poll for pending exceptions and ceded state */
2993 cur = start_poll = ktime_get();
2994 if (vc->halt_poll_ns) {
2995 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2996 ++vc->runner->stat.halt_attempted_poll;
2998 vc->vcore_state = VCORE_POLLING;
2999 spin_unlock(&vc->lock);
3002 if (kvmppc_vcore_check_block(vc)) {
3007 } while (single_task_running() && ktime_before(cur, stop));
3009 spin_lock(&vc->lock);
3010 vc->vcore_state = VCORE_INACTIVE;
3013 ++vc->runner->stat.halt_successful_poll;
3018 prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3020 if (kvmppc_vcore_check_block(vc)) {
3021 finish_swait(&vc->wq, &wait);
3023 /* If we polled, count this as a successful poll */
3024 if (vc->halt_poll_ns)
3025 ++vc->runner->stat.halt_successful_poll;
3029 start_wait = ktime_get();
3031 vc->vcore_state = VCORE_SLEEPING;
3032 trace_kvmppc_vcore_blocked(vc, 0);
3033 spin_unlock(&vc->lock);
3035 finish_swait(&vc->wq, &wait);
3036 spin_lock(&vc->lock);
3037 vc->vcore_state = VCORE_INACTIVE;
3038 trace_kvmppc_vcore_blocked(vc, 1);
3039 ++vc->runner->stat.halt_successful_wait;
3044 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3046 /* Attribute wait time */
3048 vc->runner->stat.halt_wait_ns +=
3049 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3050 /* Attribute failed poll time */
3051 if (vc->halt_poll_ns)
3052 vc->runner->stat.halt_poll_fail_ns +=
3053 ktime_to_ns(start_wait) -
3054 ktime_to_ns(start_poll);
3056 /* Attribute successful poll time */
3057 if (vc->halt_poll_ns)
3058 vc->runner->stat.halt_poll_success_ns +=
3060 ktime_to_ns(start_poll);
3063 /* Adjust poll time */
3065 if (block_ns <= vc->halt_poll_ns)
3067 /* We slept and blocked for longer than the max halt time */
3068 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3069 shrink_halt_poll_ns(vc);
3070 /* We slept and our poll time is too small */
3071 else if (vc->halt_poll_ns < halt_poll_ns &&
3072 block_ns < halt_poll_ns)
3073 grow_halt_poll_ns(vc);
3074 if (vc->halt_poll_ns > halt_poll_ns)
3075 vc->halt_poll_ns = halt_poll_ns;
3077 vc->halt_poll_ns = 0;
3079 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3082 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3085 struct kvmppc_vcore *vc;
3088 trace_kvmppc_run_vcpu_enter(vcpu);
3090 kvm_run->exit_reason = 0;
3091 vcpu->arch.ret = RESUME_GUEST;
3092 vcpu->arch.trap = 0;
3093 kvmppc_update_vpas(vcpu);
3096 * Synchronize with other threads in this virtual core
3098 vc = vcpu->arch.vcore;
3099 spin_lock(&vc->lock);
3100 vcpu->arch.ceded = 0;
3101 vcpu->arch.run_task = current;
3102 vcpu->arch.kvm_run = kvm_run;
3103 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3104 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3105 vcpu->arch.busy_preempt = TB_NIL;
3106 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3110 * This happens the first time this is called for a vcpu.
3111 * If the vcore is already running, we may be able to start
3112 * this thread straight away and have it join in.
3114 if (!signal_pending(current)) {
3115 if (vc->vcore_state == VCORE_PIGGYBACK) {
3116 if (spin_trylock(&vc->lock)) {
3117 if (vc->vcore_state == VCORE_RUNNING &&
3118 !VCORE_IS_EXITING(vc)) {
3119 kvmppc_create_dtl_entry(vcpu, vc);
3120 kvmppc_start_thread(vcpu, vc);
3121 trace_kvm_guest_enter(vcpu);
3123 spin_unlock(&vc->lock);
3125 } else if (vc->vcore_state == VCORE_RUNNING &&
3126 !VCORE_IS_EXITING(vc)) {
3127 kvmppc_create_dtl_entry(vcpu, vc);
3128 kvmppc_start_thread(vcpu, vc);
3129 trace_kvm_guest_enter(vcpu);
3130 } else if (vc->vcore_state == VCORE_SLEEPING) {
3136 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3137 !signal_pending(current)) {
3138 /* See if the HPT and VRMA are ready to go */
3139 if (!kvm_is_radix(vcpu->kvm) &&
3140 !vcpu->kvm->arch.hpte_setup_done) {
3141 spin_unlock(&vc->lock);
3142 r = kvmppc_hv_setup_htab_rma(vcpu);
3143 spin_lock(&vc->lock);
3145 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3146 kvm_run->fail_entry.hardware_entry_failure_reason = 0;
3152 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3153 kvmppc_vcore_end_preempt(vc);
3155 if (vc->vcore_state != VCORE_INACTIVE) {
3156 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3159 for_each_runnable_thread(i, v, vc) {
3160 kvmppc_core_prepare_to_enter(v);
3161 if (signal_pending(v->arch.run_task)) {
3162 kvmppc_remove_runnable(vc, v);
3163 v->stat.signal_exits++;
3164 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3165 v->arch.ret = -EINTR;
3166 wake_up(&v->arch.cpu_run);
3169 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3172 for_each_runnable_thread(i, v, vc) {
3173 if (!kvmppc_vcpu_woken(v))
3174 n_ceded += v->arch.ceded;
3179 if (n_ceded == vc->n_runnable) {
3180 kvmppc_vcore_blocked(vc);
3181 } else if (need_resched()) {
3182 kvmppc_vcore_preempt(vc);
3183 /* Let something else run */
3184 cond_resched_lock(&vc->lock);
3185 if (vc->vcore_state == VCORE_PREEMPT)
3186 kvmppc_vcore_end_preempt(vc);
3188 kvmppc_run_core(vc);
3193 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3194 (vc->vcore_state == VCORE_RUNNING ||
3195 vc->vcore_state == VCORE_EXITING ||
3196 vc->vcore_state == VCORE_PIGGYBACK))
3197 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3199 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3200 kvmppc_vcore_end_preempt(vc);
3202 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3203 kvmppc_remove_runnable(vc, vcpu);
3204 vcpu->stat.signal_exits++;
3205 kvm_run->exit_reason = KVM_EXIT_INTR;
3206 vcpu->arch.ret = -EINTR;
3209 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3210 /* Wake up some vcpu to run the core */
3212 v = next_runnable_thread(vc, &i);
3213 wake_up(&v->arch.cpu_run);
3216 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3217 spin_unlock(&vc->lock);
3218 return vcpu->arch.ret;
3221 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3225 unsigned long ebb_regs[3] = {}; /* shut up GCC */
3226 unsigned long user_tar = 0;
3227 unsigned int user_vrsave;
3229 if (!vcpu->arch.sane) {
3230 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3235 * Don't allow entry with a suspended transaction, because
3236 * the guest entry/exit code will lose it.
3237 * If the guest has TM enabled, save away their TM-related SPRs
3238 * (they will get restored by the TM unavailable interrupt).
3240 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3241 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3242 (current->thread.regs->msr & MSR_TM)) {
3243 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3244 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3245 run->fail_entry.hardware_entry_failure_reason = 0;
3248 /* Enable TM so we can read the TM SPRs */
3249 mtmsr(mfmsr() | MSR_TM);
3250 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3251 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3252 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3253 current->thread.regs->msr &= ~MSR_TM;
3257 kvmppc_core_prepare_to_enter(vcpu);
3259 /* No need to go into the guest when all we'll do is come back out */
3260 if (signal_pending(current)) {
3261 run->exit_reason = KVM_EXIT_INTR;
3265 atomic_inc(&vcpu->kvm->arch.vcpus_running);
3266 /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
3269 flush_all_to_thread(current);
3271 /* Save userspace EBB and other register values */
3272 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3273 ebb_regs[0] = mfspr(SPRN_EBBHR);
3274 ebb_regs[1] = mfspr(SPRN_EBBRR);
3275 ebb_regs[2] = mfspr(SPRN_BESCR);
3276 user_tar = mfspr(SPRN_TAR);
3278 user_vrsave = mfspr(SPRN_VRSAVE);
3280 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3281 vcpu->arch.pgdir = current->mm->pgd;
3282 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3285 r = kvmppc_run_vcpu(run, vcpu);
3287 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3288 !(vcpu->arch.shregs.msr & MSR_PR)) {
3289 trace_kvm_hcall_enter(vcpu);
3290 r = kvmppc_pseries_do_hcall(vcpu);
3291 trace_kvm_hcall_exit(vcpu, r);
3292 kvmppc_core_prepare_to_enter(vcpu);
3293 } else if (r == RESUME_PAGE_FAULT) {
3294 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3295 r = kvmppc_book3s_hv_page_fault(run, vcpu,
3296 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3297 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3298 } else if (r == RESUME_PASSTHROUGH) {
3299 if (WARN_ON(xive_enabled()))
3302 r = kvmppc_xics_rm_complete(vcpu, 0);
3304 } while (is_kvmppc_resume_guest(r));
3306 /* Restore userspace EBB and other register values */
3307 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3308 mtspr(SPRN_EBBHR, ebb_regs[0]);
3309 mtspr(SPRN_EBBRR, ebb_regs[1]);
3310 mtspr(SPRN_BESCR, ebb_regs[2]);
3311 mtspr(SPRN_TAR, user_tar);
3312 mtspr(SPRN_FSCR, current->thread.fscr);
3314 mtspr(SPRN_VRSAVE, user_vrsave);
3316 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3317 atomic_dec(&vcpu->kvm->arch.vcpus_running);
3321 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3324 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
3328 (*sps)->page_shift = def->shift;
3329 (*sps)->slb_enc = def->sllp;
3330 (*sps)->enc[0].page_shift = def->shift;
3331 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
3333 * Add 16MB MPSS support if host supports it
3335 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
3336 (*sps)->enc[1].page_shift = 24;
3337 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
3342 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3343 struct kvm_ppc_smmu_info *info)
3345 struct kvm_ppc_one_seg_page_size *sps;
3348 * Since we don't yet support HPT guests on a radix host,
3349 * return an error if the host uses radix.
3351 if (radix_enabled())
3355 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3356 * POWER7 doesn't support keys for instruction accesses,
3357 * POWER8 and POWER9 do.
3359 info->data_keys = 32;
3360 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3362 info->flags = KVM_PPC_PAGE_SIZES_REAL;
3363 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
3364 info->flags |= KVM_PPC_1T_SEGMENTS;
3365 info->slb_size = mmu_slb_size;
3367 /* We only support these sizes for now, and no muti-size segments */
3368 sps = &info->sps[0];
3369 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
3370 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
3371 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
3377 * Get (and clear) the dirty memory log for a memory slot.
3379 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3380 struct kvm_dirty_log *log)
3382 struct kvm_memslots *slots;
3383 struct kvm_memory_slot *memslot;
3387 struct kvm_vcpu *vcpu;
3389 mutex_lock(&kvm->slots_lock);
3392 if (log->slot >= KVM_USER_MEM_SLOTS)
3395 slots = kvm_memslots(kvm);
3396 memslot = id_to_memslot(slots, log->slot);
3398 if (!memslot->dirty_bitmap)
3402 * Use second half of bitmap area because radix accumulates
3403 * bits in the first half.
3405 n = kvm_dirty_bitmap_bytes(memslot);
3406 buf = memslot->dirty_bitmap + n / sizeof(long);
3409 if (kvm_is_radix(kvm))
3410 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3412 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3416 /* Harvest dirty bits from VPA and DTL updates */
3417 /* Note: we never modify the SLB shadow buffer areas */
3418 kvm_for_each_vcpu(i, vcpu, kvm) {
3419 spin_lock(&vcpu->arch.vpa_update_lock);
3420 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3421 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3422 spin_unlock(&vcpu->arch.vpa_update_lock);
3426 if (copy_to_user(log->dirty_bitmap, buf, n))
3431 mutex_unlock(&kvm->slots_lock);
3435 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3436 struct kvm_memory_slot *dont)
3438 if (!dont || free->arch.rmap != dont->arch.rmap) {
3439 vfree(free->arch.rmap);
3440 free->arch.rmap = NULL;
3444 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3445 unsigned long npages)
3448 * For now, if radix_enabled() then we only support radix guests,
3449 * and in that case we don't need the rmap array.
3451 if (radix_enabled()) {
3452 slot->arch.rmap = NULL;
3456 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3457 if (!slot->arch.rmap)
3463 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3464 struct kvm_memory_slot *memslot,
3465 const struct kvm_userspace_memory_region *mem)
3470 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3471 const struct kvm_userspace_memory_region *mem,
3472 const struct kvm_memory_slot *old,
3473 const struct kvm_memory_slot *new)
3475 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3476 struct kvm_memslots *slots;
3477 struct kvm_memory_slot *memslot;
3480 * If we are making a new memslot, it might make
3481 * some address that was previously cached as emulated
3482 * MMIO be no longer emulated MMIO, so invalidate
3483 * all the caches of emulated MMIO translations.
3486 atomic64_inc(&kvm->arch.mmio_update);
3488 if (npages && old->npages && !kvm_is_radix(kvm)) {
3490 * If modifying a memslot, reset all the rmap dirty bits.
3491 * If this is a new memslot, we don't need to do anything
3492 * since the rmap array starts out as all zeroes,
3493 * i.e. no pages are dirty.
3495 slots = kvm_memslots(kvm);
3496 memslot = id_to_memslot(slots, mem->slot);
3497 kvmppc_hv_get_dirty_log_hpt(kvm, memslot, NULL);
3502 * Update LPCR values in kvm->arch and in vcores.
3503 * Caller must hold kvm->lock.
3505 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3510 if ((kvm->arch.lpcr & mask) == lpcr)
3513 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3515 for (i = 0; i < KVM_MAX_VCORES; ++i) {
3516 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3519 spin_lock(&vc->lock);
3520 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3521 spin_unlock(&vc->lock);
3522 if (++cores_done >= kvm->arch.online_vcores)
3527 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3532 static void kvmppc_setup_partition_table(struct kvm *kvm)
3534 unsigned long dw0, dw1;
3536 if (!kvm_is_radix(kvm)) {
3537 /* PS field - page size for VRMA */
3538 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3539 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3540 /* HTABSIZE and HTABORG fields */
3541 dw0 |= kvm->arch.sdr1;
3543 /* Second dword as set by userspace */
3544 dw1 = kvm->arch.process_table;
3546 dw0 = PATB_HR | radix__get_tree_size() |
3547 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3548 dw1 = PATB_GR | kvm->arch.process_table;
3551 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3554 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3557 struct kvm *kvm = vcpu->kvm;
3559 struct kvm_memory_slot *memslot;
3560 struct vm_area_struct *vma;
3561 unsigned long lpcr = 0, senc;
3562 unsigned long psize, porder;
3565 mutex_lock(&kvm->lock);
3566 if (kvm->arch.hpte_setup_done)
3567 goto out; /* another vcpu beat us to it */
3569 /* Allocate hashed page table (if not done already) and reset it */
3570 if (!kvm->arch.hpt.virt) {
3571 int order = KVM_DEFAULT_HPT_ORDER;
3572 struct kvm_hpt_info info;
3574 err = kvmppc_allocate_hpt(&info, order);
3575 /* If we get here, it means userspace didn't specify a
3576 * size explicitly. So, try successively smaller
3577 * sizes if the default failed. */
3578 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3579 err = kvmppc_allocate_hpt(&info, order);
3582 pr_err("KVM: Couldn't alloc HPT\n");
3586 kvmppc_set_hpt(kvm, &info);
3589 /* Look up the memslot for guest physical address 0 */
3590 srcu_idx = srcu_read_lock(&kvm->srcu);
3591 memslot = gfn_to_memslot(kvm, 0);
3593 /* We must have some memory at 0 by now */
3595 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3598 /* Look up the VMA for the start of this memory slot */
3599 hva = memslot->userspace_addr;
3600 down_read(¤t->mm->mmap_sem);
3601 vma = find_vma(current->mm, hva);
3602 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3605 psize = vma_kernel_pagesize(vma);
3606 porder = __ilog2(psize);
3608 up_read(¤t->mm->mmap_sem);
3610 /* We can handle 4k, 64k or 16M pages in the VRMA */
3612 if (!(psize == 0x1000 || psize == 0x10000 ||
3613 psize == 0x1000000))
3616 senc = slb_pgsize_encoding(psize);
3617 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3618 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3619 /* Create HPTEs in the hash page table for the VRMA */
3620 kvmppc_map_vrma(vcpu, memslot, porder);
3622 /* Update VRMASD field in the LPCR */
3623 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3624 /* the -4 is to account for senc values starting at 0x10 */
3625 lpcr = senc << (LPCR_VRMASD_SH - 4);
3626 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3628 kvmppc_setup_partition_table(kvm);
3631 /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3633 kvm->arch.hpte_setup_done = 1;
3636 srcu_read_unlock(&kvm->srcu, srcu_idx);
3638 mutex_unlock(&kvm->lock);
3642 up_read(¤t->mm->mmap_sem);
3646 #ifdef CONFIG_KVM_XICS
3648 * Allocate a per-core structure for managing state about which cores are
3649 * running in the host versus the guest and for exchanging data between
3650 * real mode KVM and CPU running in the host.
3651 * This is only done for the first VM.
3652 * The allocated structure stays even if all VMs have stopped.
3653 * It is only freed when the kvm-hv module is unloaded.
3654 * It's OK for this routine to fail, we just don't support host
3655 * core operations like redirecting H_IPI wakeups.
3657 void kvmppc_alloc_host_rm_ops(void)
3659 struct kvmppc_host_rm_ops *ops;
3660 unsigned long l_ops;
3664 /* Not the first time here ? */
3665 if (kvmppc_host_rm_ops_hv != NULL)
3668 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3672 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3673 ops->rm_core = kzalloc(size, GFP_KERNEL);
3675 if (!ops->rm_core) {
3682 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3683 if (!cpu_online(cpu))
3686 core = cpu >> threads_shift;
3687 ops->rm_core[core].rm_state.in_host = 1;
3690 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3693 * Make the contents of the kvmppc_host_rm_ops structure visible
3694 * to other CPUs before we assign it to the global variable.
3695 * Do an atomic assignment (no locks used here), but if someone
3696 * beats us to it, just free our copy and return.
3699 l_ops = (unsigned long) ops;
3701 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3703 kfree(ops->rm_core);
3708 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3709 "ppc/kvm_book3s:prepare",
3710 kvmppc_set_host_core,
3711 kvmppc_clear_host_core);
3715 void kvmppc_free_host_rm_ops(void)
3717 if (kvmppc_host_rm_ops_hv) {
3718 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3719 kfree(kvmppc_host_rm_ops_hv->rm_core);
3720 kfree(kvmppc_host_rm_ops_hv);
3721 kvmppc_host_rm_ops_hv = NULL;
3726 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3728 unsigned long lpcr, lpid;
3732 /* Allocate the guest's logical partition ID */
3734 lpid = kvmppc_alloc_lpid();
3737 kvm->arch.lpid = lpid;
3739 kvmppc_alloc_host_rm_ops();
3742 * Since we don't flush the TLB when tearing down a VM,
3743 * and this lpid might have previously been used,
3744 * make sure we flush on each core before running the new VM.
3745 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3746 * does this flush for us.
3748 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3749 cpumask_setall(&kvm->arch.need_tlb_flush);
3751 /* Start out with the default set of hcalls enabled */
3752 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3753 sizeof(kvm->arch.enabled_hcalls));
3755 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3756 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3758 /* Init LPCR for virtual RMA mode */
3759 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3760 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3761 lpcr &= LPCR_PECE | LPCR_LPES;
3762 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3763 LPCR_VPM0 | LPCR_VPM1;
3764 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3765 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3766 /* On POWER8 turn on online bit to enable PURR/SPURR */
3767 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3770 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3771 * Set HVICE bit to enable hypervisor virtualization interrupts.
3772 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3773 * be unnecessary but better safe than sorry in case we re-enable
3774 * EE in HV mode with this LPCR still set)
3776 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3778 lpcr |= LPCR_HVICE | LPCR_HEIC;
3781 * If xive is enabled, we route 0x500 interrupts directly
3789 * For now, if the host uses radix, the guest must be radix.
3791 if (radix_enabled()) {
3792 kvm->arch.radix = 1;
3794 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3795 ret = kvmppc_init_vm_radix(kvm);
3797 kvmppc_free_lpid(kvm->arch.lpid);
3800 kvmppc_setup_partition_table(kvm);
3803 kvm->arch.lpcr = lpcr;
3805 /* Initialization for future HPT resizes */
3806 kvm->arch.resize_hpt = NULL;
3809 * Work out how many sets the TLB has, for the use of
3810 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3812 if (kvm_is_radix(kvm))
3813 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
3814 else if (cpu_has_feature(CPU_FTR_ARCH_300))
3815 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
3816 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3817 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
3819 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
3822 * Track that we now have a HV mode VM active. This blocks secondary
3823 * CPU threads from coming online.
3824 * On POWER9, we only need to do this for HPT guests on a radix
3825 * host, which is not yet supported.
3827 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3828 kvm_hv_vm_activated();
3831 * Initialize smt_mode depending on processor.
3832 * POWER8 and earlier have to use "strict" threading, where
3833 * all vCPUs in a vcore have to run on the same (sub)core,
3834 * whereas on POWER9 the threads can each run a different
3837 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3838 kvm->arch.smt_mode = threads_per_subcore;
3840 kvm->arch.smt_mode = 1;
3841 kvm->arch.emul_smt_mode = 1;
3844 * Create a debugfs directory for the VM
3846 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3847 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3848 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3849 kvmppc_mmu_debugfs_init(kvm);
3854 static void kvmppc_free_vcores(struct kvm *kvm)
3858 for (i = 0; i < KVM_MAX_VCORES; ++i)
3859 kfree(kvm->arch.vcores[i]);
3860 kvm->arch.online_vcores = 0;
3863 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3865 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3867 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3868 kvm_hv_vm_deactivated();
3870 kvmppc_free_vcores(kvm);
3872 kvmppc_free_lpid(kvm->arch.lpid);
3874 if (kvm_is_radix(kvm))
3875 kvmppc_free_radix(kvm);
3877 kvmppc_free_hpt(&kvm->arch.hpt);
3879 kvmppc_free_pimap(kvm);
3882 /* We don't need to emulate any privileged instructions or dcbz */
3883 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3884 unsigned int inst, int *advance)
3886 return EMULATE_FAIL;
3889 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3892 return EMULATE_FAIL;
3895 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3898 return EMULATE_FAIL;
3901 static int kvmppc_core_check_processor_compat_hv(void)
3903 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3904 !cpu_has_feature(CPU_FTR_ARCH_206))
3910 #ifdef CONFIG_KVM_XICS
3912 void kvmppc_free_pimap(struct kvm *kvm)
3914 kfree(kvm->arch.pimap);
3917 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3919 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3922 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3924 struct irq_desc *desc;
3925 struct kvmppc_irq_map *irq_map;
3926 struct kvmppc_passthru_irqmap *pimap;
3927 struct irq_chip *chip;
3930 if (!kvm_irq_bypass)
3933 desc = irq_to_desc(host_irq);
3937 mutex_lock(&kvm->lock);
3939 pimap = kvm->arch.pimap;
3940 if (pimap == NULL) {
3941 /* First call, allocate structure to hold IRQ map */
3942 pimap = kvmppc_alloc_pimap();
3943 if (pimap == NULL) {
3944 mutex_unlock(&kvm->lock);
3947 kvm->arch.pimap = pimap;
3951 * For now, we only support interrupts for which the EOI operation
3952 * is an OPAL call followed by a write to XIRR, since that's
3953 * what our real-mode EOI code does, or a XIVE interrupt
3955 chip = irq_data_get_irq_chip(&desc->irq_data);
3956 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
3957 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3958 host_irq, guest_gsi);
3959 mutex_unlock(&kvm->lock);
3964 * See if we already have an entry for this guest IRQ number.
3965 * If it's mapped to a hardware IRQ number, that's an error,
3966 * otherwise re-use this entry.
3968 for (i = 0; i < pimap->n_mapped; i++) {
3969 if (guest_gsi == pimap->mapped[i].v_hwirq) {
3970 if (pimap->mapped[i].r_hwirq) {
3971 mutex_unlock(&kvm->lock);
3978 if (i == KVMPPC_PIRQ_MAPPED) {
3979 mutex_unlock(&kvm->lock);
3980 return -EAGAIN; /* table is full */
3983 irq_map = &pimap->mapped[i];
3985 irq_map->v_hwirq = guest_gsi;
3986 irq_map->desc = desc;
3989 * Order the above two stores before the next to serialize with
3990 * the KVM real mode handler.
3993 irq_map->r_hwirq = desc->irq_data.hwirq;
3995 if (i == pimap->n_mapped)
3999 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4001 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4003 irq_map->r_hwirq = 0;
4005 mutex_unlock(&kvm->lock);
4010 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4012 struct irq_desc *desc;
4013 struct kvmppc_passthru_irqmap *pimap;
4016 if (!kvm_irq_bypass)
4019 desc = irq_to_desc(host_irq);
4023 mutex_lock(&kvm->lock);
4024 if (!kvm->arch.pimap)
4027 pimap = kvm->arch.pimap;
4029 for (i = 0; i < pimap->n_mapped; i++) {
4030 if (guest_gsi == pimap->mapped[i].v_hwirq)
4034 if (i == pimap->n_mapped) {
4035 mutex_unlock(&kvm->lock);
4040 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4042 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4044 /* invalidate the entry (what do do on error from the above ?) */
4045 pimap->mapped[i].r_hwirq = 0;
4048 * We don't free this structure even when the count goes to
4049 * zero. The structure is freed when we destroy the VM.
4052 mutex_unlock(&kvm->lock);
4056 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4057 struct irq_bypass_producer *prod)
4060 struct kvm_kernel_irqfd *irqfd =
4061 container_of(cons, struct kvm_kernel_irqfd, consumer);
4063 irqfd->producer = prod;
4065 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4067 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4068 prod->irq, irqfd->gsi, ret);
4073 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4074 struct irq_bypass_producer *prod)
4077 struct kvm_kernel_irqfd *irqfd =
4078 container_of(cons, struct kvm_kernel_irqfd, consumer);
4080 irqfd->producer = NULL;
4083 * When producer of consumer is unregistered, we change back to
4084 * default external interrupt handling mode - KVM real mode
4085 * will switch back to host.
4087 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4089 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4090 prod->irq, irqfd->gsi, ret);
4094 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4095 unsigned int ioctl, unsigned long arg)
4097 struct kvm *kvm __maybe_unused = filp->private_data;
4098 void __user *argp = (void __user *)arg;
4103 case KVM_PPC_ALLOCATE_HTAB: {
4107 if (get_user(htab_order, (u32 __user *)argp))
4109 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4116 case KVM_PPC_GET_HTAB_FD: {
4117 struct kvm_get_htab_fd ghf;
4120 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4122 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4126 case KVM_PPC_RESIZE_HPT_PREPARE: {
4127 struct kvm_ppc_resize_hpt rhpt;
4130 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4133 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4137 case KVM_PPC_RESIZE_HPT_COMMIT: {
4138 struct kvm_ppc_resize_hpt rhpt;
4141 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4144 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4156 * List of hcall numbers to enable by default.
4157 * For compatibility with old userspace, we enable by default
4158 * all hcalls that were implemented before the hcall-enabling
4159 * facility was added. Note this list should not include H_RTAS.
4161 static unsigned int default_hcall_list[] = {
4175 #ifdef CONFIG_KVM_XICS
4186 static void init_default_hcalls(void)
4191 for (i = 0; default_hcall_list[i]; ++i) {
4192 hcall = default_hcall_list[i];
4193 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4194 __set_bit(hcall / 4, default_enabled_hcalls);
4198 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4203 /* If not on a POWER9, reject it */
4204 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4207 /* If any unknown flags set, reject it */
4208 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4211 /* We can't change a guest to/from radix yet */
4212 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4213 if (radix != kvm_is_radix(kvm))
4216 /* GR (guest radix) bit in process_table field must match */
4217 if (!!(cfg->process_table & PATB_GR) != radix)
4220 /* Process table size field must be reasonable, i.e. <= 24 */
4221 if ((cfg->process_table & PRTS_MASK) > 24)
4224 mutex_lock(&kvm->lock);
4225 kvm->arch.process_table = cfg->process_table;
4226 kvmppc_setup_partition_table(kvm);
4228 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4229 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4230 mutex_unlock(&kvm->lock);
4235 static struct kvmppc_ops kvm_ops_hv = {
4236 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4237 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4238 .get_one_reg = kvmppc_get_one_reg_hv,
4239 .set_one_reg = kvmppc_set_one_reg_hv,
4240 .vcpu_load = kvmppc_core_vcpu_load_hv,
4241 .vcpu_put = kvmppc_core_vcpu_put_hv,
4242 .set_msr = kvmppc_set_msr_hv,
4243 .vcpu_run = kvmppc_vcpu_run_hv,
4244 .vcpu_create = kvmppc_core_vcpu_create_hv,
4245 .vcpu_free = kvmppc_core_vcpu_free_hv,
4246 .check_requests = kvmppc_core_check_requests_hv,
4247 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
4248 .flush_memslot = kvmppc_core_flush_memslot_hv,
4249 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4250 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
4251 .unmap_hva = kvm_unmap_hva_hv,
4252 .unmap_hva_range = kvm_unmap_hva_range_hv,
4253 .age_hva = kvm_age_hva_hv,
4254 .test_age_hva = kvm_test_age_hva_hv,
4255 .set_spte_hva = kvm_set_spte_hva_hv,
4256 .mmu_destroy = kvmppc_mmu_destroy_hv,
4257 .free_memslot = kvmppc_core_free_memslot_hv,
4258 .create_memslot = kvmppc_core_create_memslot_hv,
4259 .init_vm = kvmppc_core_init_vm_hv,
4260 .destroy_vm = kvmppc_core_destroy_vm_hv,
4261 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4262 .emulate_op = kvmppc_core_emulate_op_hv,
4263 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4264 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4265 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4266 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
4267 .hcall_implemented = kvmppc_hcall_impl_hv,
4268 #ifdef CONFIG_KVM_XICS
4269 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4270 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4272 .configure_mmu = kvmhv_configure_mmu,
4273 .get_rmmu_info = kvmhv_get_rmmu_info,
4274 .set_smt_mode = kvmhv_set_smt_mode,
4277 static int kvm_init_subcore_bitmap(void)
4280 int nr_cores = cpu_nr_cores();
4281 struct sibling_subcore_state *sibling_subcore_state;
4283 for (i = 0; i < nr_cores; i++) {
4284 int first_cpu = i * threads_per_core;
4285 int node = cpu_to_node(first_cpu);
4287 /* Ignore if it is already allocated. */
4288 if (paca[first_cpu].sibling_subcore_state)
4291 sibling_subcore_state =
4292 kmalloc_node(sizeof(struct sibling_subcore_state),
4294 if (!sibling_subcore_state)
4297 memset(sibling_subcore_state, 0,
4298 sizeof(struct sibling_subcore_state));
4300 for (j = 0; j < threads_per_core; j++) {
4301 int cpu = first_cpu + j;
4303 paca[cpu].sibling_subcore_state = sibling_subcore_state;
4309 static int kvmppc_radix_possible(void)
4311 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4314 static int kvmppc_book3s_init_hv(void)
4318 * FIXME!! Do we need to check on all cpus ?
4320 r = kvmppc_core_check_processor_compat_hv();
4324 r = kvm_init_subcore_bitmap();
4329 * We need a way of accessing the XICS interrupt controller,
4330 * either directly, via paca[cpu].kvm_hstate.xics_phys, or
4331 * indirectly, via OPAL.
4334 if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4335 struct device_node *np;
4337 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4339 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4345 kvm_ops_hv.owner = THIS_MODULE;
4346 kvmppc_hv_ops = &kvm_ops_hv;
4348 init_default_hcalls();
4352 r = kvmppc_mmu_hv_init();
4356 if (kvmppc_radix_possible())
4357 r = kvmppc_radix_init();
4361 static void kvmppc_book3s_exit_hv(void)
4363 kvmppc_free_host_rm_ops();
4364 if (kvmppc_radix_possible())
4365 kvmppc_radix_exit();
4366 kvmppc_hv_ops = NULL;
4369 module_init(kvmppc_book3s_init_hv);
4370 module_exit(kvmppc_book3s_exit_hv);
4371 MODULE_LICENSE("GPL");
4372 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4373 MODULE_ALIAS("devname:kvm");