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/kernel.h>
23 #include <linux/err.h>
24 #include <linux/slab.h>
25 #include <linux/preempt.h>
26 #include <linux/sched/signal.h>
27 #include <linux/sched/stat.h>
28 #include <linux/delay.h>
29 #include <linux/export.h>
31 #include <linux/anon_inodes.h>
32 #include <linux/cpu.h>
33 #include <linux/cpumask.h>
34 #include <linux/spinlock.h>
35 #include <linux/page-flags.h>
36 #include <linux/srcu.h>
37 #include <linux/miscdevice.h>
38 #include <linux/debugfs.h>
39 #include <linux/gfp.h>
40 #include <linux/vmalloc.h>
41 #include <linux/highmem.h>
42 #include <linux/hugetlb.h>
43 #include <linux/kvm_irqfd.h>
44 #include <linux/irqbypass.h>
45 #include <linux/module.h>
46 #include <linux/compiler.h>
49 #include <asm/ftrace.h>
51 #include <asm/ppc-opcode.h>
52 #include <asm/asm-prototypes.h>
53 #include <asm/debug.h>
54 #include <asm/disassemble.h>
55 #include <asm/cputable.h>
56 #include <asm/cacheflush.h>
57 #include <linux/uaccess.h>
59 #include <asm/kvm_ppc.h>
60 #include <asm/kvm_book3s.h>
61 #include <asm/mmu_context.h>
62 #include <asm/lppaca.h>
63 #include <asm/processor.h>
64 #include <asm/cputhreads.h>
66 #include <asm/hvcall.h>
67 #include <asm/switch_to.h>
69 #include <asm/dbell.h>
71 #include <asm/pnv-pci.h>
79 #define CREATE_TRACE_POINTS
82 /* #define EXIT_DEBUG */
83 /* #define EXIT_DEBUG_SIMPLE */
84 /* #define EXIT_DEBUG_INT */
86 /* Used to indicate that a guest page fault needs to be handled */
87 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
88 /* Used to indicate that a guest passthrough interrupt needs to be handled */
89 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
91 /* Used as a "null" value for timebase values */
92 #define TB_NIL (~(u64)0)
94 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
96 static int dynamic_mt_modes = 6;
97 module_param(dynamic_mt_modes, int, 0644);
98 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
99 static int target_smt_mode;
100 module_param(target_smt_mode, int, 0644);
101 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
103 static bool indep_threads_mode = true;
104 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
105 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
107 #ifdef CONFIG_KVM_XICS
108 static struct kernel_param_ops module_param_ops = {
109 .set = param_set_int,
110 .get = param_get_int,
113 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
114 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
116 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
117 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
120 /* If set, the threads on each CPU core have to be in the same MMU mode */
121 static bool no_mixing_hpt_and_radix;
123 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
124 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
127 * RWMR values for POWER8. These control the rate at which PURR
128 * and SPURR count and should be set according to the number of
129 * online threads in the vcore being run.
131 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL
132 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL
133 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL
134 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL
135 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL
136 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL
137 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL
138 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL
140 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
152 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
156 struct kvm_vcpu *vcpu;
158 while (++i < MAX_SMT_THREADS) {
159 vcpu = READ_ONCE(vc->runnable_threads[i]);
168 /* Used to traverse the list of runnable threads for a given vcore */
169 #define for_each_runnable_thread(i, vcpu, vc) \
170 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
172 static bool kvmppc_ipi_thread(int cpu)
174 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
176 /* On POWER9 we can use msgsnd to IPI any cpu */
177 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
178 msg |= get_hard_smp_processor_id(cpu);
180 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
184 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
185 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
187 if (cpu_first_thread_sibling(cpu) ==
188 cpu_first_thread_sibling(smp_processor_id())) {
189 msg |= cpu_thread_in_core(cpu);
191 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
198 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
199 if (cpu >= 0 && cpu < nr_cpu_ids) {
200 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
204 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
212 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
215 struct swait_queue_head *wqp;
217 wqp = kvm_arch_vcpu_wq(vcpu);
218 if (swq_has_sleeper(wqp)) {
220 ++vcpu->stat.halt_wakeup;
223 cpu = READ_ONCE(vcpu->arch.thread_cpu);
224 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
227 /* CPU points to the first thread of the core */
229 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
230 smp_send_reschedule(cpu);
234 * We use the vcpu_load/put functions to measure stolen time.
235 * Stolen time is counted as time when either the vcpu is able to
236 * run as part of a virtual core, but the task running the vcore
237 * is preempted or sleeping, or when the vcpu needs something done
238 * in the kernel by the task running the vcpu, but that task is
239 * preempted or sleeping. Those two things have to be counted
240 * separately, since one of the vcpu tasks will take on the job
241 * of running the core, and the other vcpu tasks in the vcore will
242 * sleep waiting for it to do that, but that sleep shouldn't count
245 * Hence we accumulate stolen time when the vcpu can run as part of
246 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
247 * needs its task to do other things in the kernel (for example,
248 * service a page fault) in busy_stolen. We don't accumulate
249 * stolen time for a vcore when it is inactive, or for a vcpu
250 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
251 * a misnomer; it means that the vcpu task is not executing in
252 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
253 * the kernel. We don't have any way of dividing up that time
254 * between time that the vcpu is genuinely stopped, time that
255 * the task is actively working on behalf of the vcpu, and time
256 * that the task is preempted, so we don't count any of it as
259 * Updates to busy_stolen are protected by arch.tbacct_lock;
260 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
261 * lock. The stolen times are measured in units of timebase ticks.
262 * (Note that the != TB_NIL checks below are purely defensive;
263 * they should never fail.)
266 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
270 spin_lock_irqsave(&vc->stoltb_lock, flags);
271 vc->preempt_tb = mftb();
272 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
275 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
279 spin_lock_irqsave(&vc->stoltb_lock, flags);
280 if (vc->preempt_tb != TB_NIL) {
281 vc->stolen_tb += mftb() - vc->preempt_tb;
282 vc->preempt_tb = TB_NIL;
284 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
287 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
289 struct kvmppc_vcore *vc = vcpu->arch.vcore;
293 * We can test vc->runner without taking the vcore lock,
294 * because only this task ever sets vc->runner to this
295 * vcpu, and once it is set to this vcpu, only this task
296 * ever sets it to NULL.
298 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
299 kvmppc_core_end_stolen(vc);
301 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
302 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
303 vcpu->arch.busy_preempt != TB_NIL) {
304 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
305 vcpu->arch.busy_preempt = TB_NIL;
307 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
310 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
312 struct kvmppc_vcore *vc = vcpu->arch.vcore;
315 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
316 kvmppc_core_start_stolen(vc);
318 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
319 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
320 vcpu->arch.busy_preempt = mftb();
321 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
324 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
327 * Check for illegal transactional state bit combination
328 * and if we find it, force the TS field to a safe state.
330 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
332 vcpu->arch.shregs.msr = msr;
333 kvmppc_end_cede(vcpu);
336 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
338 vcpu->arch.pvr = pvr;
341 /* Dummy value used in computing PCR value below */
342 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
344 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
346 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
347 struct kvmppc_vcore *vc = vcpu->arch.vcore;
349 /* We can (emulate) our own architecture version and anything older */
350 if (cpu_has_feature(CPU_FTR_ARCH_300))
351 host_pcr_bit = PCR_ARCH_300;
352 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
353 host_pcr_bit = PCR_ARCH_207;
354 else if (cpu_has_feature(CPU_FTR_ARCH_206))
355 host_pcr_bit = PCR_ARCH_206;
357 host_pcr_bit = PCR_ARCH_205;
359 /* Determine lowest PCR bit needed to run guest in given PVR level */
360 guest_pcr_bit = host_pcr_bit;
362 switch (arch_compat) {
364 guest_pcr_bit = PCR_ARCH_205;
368 guest_pcr_bit = PCR_ARCH_206;
371 guest_pcr_bit = PCR_ARCH_207;
374 guest_pcr_bit = PCR_ARCH_300;
381 /* Check requested PCR bits don't exceed our capabilities */
382 if (guest_pcr_bit > host_pcr_bit)
385 spin_lock(&vc->lock);
386 vc->arch_compat = arch_compat;
387 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
388 vc->pcr = host_pcr_bit - guest_pcr_bit;
389 spin_unlock(&vc->lock);
394 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
398 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
399 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
400 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
401 for (r = 0; r < 16; ++r)
402 pr_err("r%2d = %.16lx r%d = %.16lx\n",
403 r, kvmppc_get_gpr(vcpu, r),
404 r+16, kvmppc_get_gpr(vcpu, r+16));
405 pr_err("ctr = %.16lx lr = %.16lx\n",
406 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
407 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
408 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
409 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
410 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
411 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
412 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
413 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n",
414 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
415 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
416 pr_err("fault dar = %.16lx dsisr = %.8x\n",
417 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
418 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
419 for (r = 0; r < vcpu->arch.slb_max; ++r)
420 pr_err(" ESID = %.16llx VSID = %.16llx\n",
421 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
422 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
423 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
424 vcpu->arch.last_inst);
427 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
429 return kvm_get_vcpu_by_id(kvm, id);
432 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
434 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
435 vpa->yield_count = cpu_to_be32(1);
438 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
439 unsigned long addr, unsigned long len)
441 /* check address is cacheline aligned */
442 if (addr & (L1_CACHE_BYTES - 1))
444 spin_lock(&vcpu->arch.vpa_update_lock);
445 if (v->next_gpa != addr || v->len != len) {
447 v->len = addr ? len : 0;
448 v->update_pending = 1;
450 spin_unlock(&vcpu->arch.vpa_update_lock);
454 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
463 static int vpa_is_registered(struct kvmppc_vpa *vpap)
465 if (vpap->update_pending)
466 return vpap->next_gpa != 0;
467 return vpap->pinned_addr != NULL;
470 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
472 unsigned long vcpuid, unsigned long vpa)
474 struct kvm *kvm = vcpu->kvm;
475 unsigned long len, nb;
477 struct kvm_vcpu *tvcpu;
480 struct kvmppc_vpa *vpap;
482 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
486 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
487 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
488 subfunc == H_VPA_REG_SLB) {
489 /* Registering new area - address must be cache-line aligned */
490 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
493 /* convert logical addr to kernel addr and read length */
494 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
497 if (subfunc == H_VPA_REG_VPA)
498 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
500 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
501 kvmppc_unpin_guest_page(kvm, va, vpa, false);
504 if (len > nb || len < sizeof(struct reg_vpa))
513 spin_lock(&tvcpu->arch.vpa_update_lock);
516 case H_VPA_REG_VPA: /* register VPA */
518 * The size of our lppaca is 1kB because of the way we align
519 * it for the guest to avoid crossing a 4kB boundary. We only
520 * use 640 bytes of the structure though, so we should accept
521 * clients that set a size of 640.
523 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
524 if (len < sizeof(struct lppaca))
526 vpap = &tvcpu->arch.vpa;
530 case H_VPA_REG_DTL: /* register DTL */
531 if (len < sizeof(struct dtl_entry))
533 len -= len % sizeof(struct dtl_entry);
535 /* Check that they have previously registered a VPA */
537 if (!vpa_is_registered(&tvcpu->arch.vpa))
540 vpap = &tvcpu->arch.dtl;
544 case H_VPA_REG_SLB: /* register SLB shadow buffer */
545 /* Check that they have previously registered a VPA */
547 if (!vpa_is_registered(&tvcpu->arch.vpa))
550 vpap = &tvcpu->arch.slb_shadow;
554 case H_VPA_DEREG_VPA: /* deregister VPA */
555 /* Check they don't still have a DTL or SLB buf registered */
557 if (vpa_is_registered(&tvcpu->arch.dtl) ||
558 vpa_is_registered(&tvcpu->arch.slb_shadow))
561 vpap = &tvcpu->arch.vpa;
565 case H_VPA_DEREG_DTL: /* deregister DTL */
566 vpap = &tvcpu->arch.dtl;
570 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
571 vpap = &tvcpu->arch.slb_shadow;
577 vpap->next_gpa = vpa;
579 vpap->update_pending = 1;
582 spin_unlock(&tvcpu->arch.vpa_update_lock);
587 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
589 struct kvm *kvm = vcpu->kvm;
595 * We need to pin the page pointed to by vpap->next_gpa,
596 * but we can't call kvmppc_pin_guest_page under the lock
597 * as it does get_user_pages() and down_read(). So we
598 * have to drop the lock, pin the page, then get the lock
599 * again and check that a new area didn't get registered
603 gpa = vpap->next_gpa;
604 spin_unlock(&vcpu->arch.vpa_update_lock);
608 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
609 spin_lock(&vcpu->arch.vpa_update_lock);
610 if (gpa == vpap->next_gpa)
612 /* sigh... unpin that one and try again */
614 kvmppc_unpin_guest_page(kvm, va, gpa, false);
617 vpap->update_pending = 0;
618 if (va && nb < vpap->len) {
620 * If it's now too short, it must be that userspace
621 * has changed the mappings underlying guest memory,
622 * so unregister the region.
624 kvmppc_unpin_guest_page(kvm, va, gpa, false);
627 if (vpap->pinned_addr)
628 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
631 vpap->pinned_addr = va;
634 vpap->pinned_end = va + vpap->len;
637 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
639 if (!(vcpu->arch.vpa.update_pending ||
640 vcpu->arch.slb_shadow.update_pending ||
641 vcpu->arch.dtl.update_pending))
644 spin_lock(&vcpu->arch.vpa_update_lock);
645 if (vcpu->arch.vpa.update_pending) {
646 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
647 if (vcpu->arch.vpa.pinned_addr)
648 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
650 if (vcpu->arch.dtl.update_pending) {
651 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
652 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
653 vcpu->arch.dtl_index = 0;
655 if (vcpu->arch.slb_shadow.update_pending)
656 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
657 spin_unlock(&vcpu->arch.vpa_update_lock);
661 * Return the accumulated stolen time for the vcore up until `now'.
662 * The caller should hold the vcore lock.
664 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
669 spin_lock_irqsave(&vc->stoltb_lock, flags);
671 if (vc->vcore_state != VCORE_INACTIVE &&
672 vc->preempt_tb != TB_NIL)
673 p += now - vc->preempt_tb;
674 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
678 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
679 struct kvmppc_vcore *vc)
681 struct dtl_entry *dt;
683 unsigned long stolen;
684 unsigned long core_stolen;
688 dt = vcpu->arch.dtl_ptr;
689 vpa = vcpu->arch.vpa.pinned_addr;
691 core_stolen = vcore_stolen_time(vc, now);
692 stolen = core_stolen - vcpu->arch.stolen_logged;
693 vcpu->arch.stolen_logged = core_stolen;
694 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
695 stolen += vcpu->arch.busy_stolen;
696 vcpu->arch.busy_stolen = 0;
697 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
700 memset(dt, 0, sizeof(struct dtl_entry));
701 dt->dispatch_reason = 7;
702 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
703 dt->timebase = cpu_to_be64(now + vc->tb_offset);
704 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
705 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
706 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
708 if (dt == vcpu->arch.dtl.pinned_end)
709 dt = vcpu->arch.dtl.pinned_addr;
710 vcpu->arch.dtl_ptr = dt;
711 /* order writing *dt vs. writing vpa->dtl_idx */
713 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
714 vcpu->arch.dtl.dirty = true;
717 /* See if there is a doorbell interrupt pending for a vcpu */
718 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
721 struct kvmppc_vcore *vc;
723 if (vcpu->arch.doorbell_request)
726 * Ensure that the read of vcore->dpdes comes after the read
727 * of vcpu->doorbell_request. This barrier matches the
728 * lwsync in book3s_hv_rmhandlers.S just before the
729 * fast_guest_return label.
732 vc = vcpu->arch.vcore;
733 thr = vcpu->vcpu_id - vc->first_vcpuid;
734 return !!(vc->dpdes & (1 << thr));
737 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
739 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
741 if ((!vcpu->arch.vcore->arch_compat) &&
742 cpu_has_feature(CPU_FTR_ARCH_207S))
747 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
748 unsigned long resource, unsigned long value1,
749 unsigned long value2)
752 case H_SET_MODE_RESOURCE_SET_CIABR:
753 if (!kvmppc_power8_compatible(vcpu))
758 return H_UNSUPPORTED_FLAG_START;
759 /* Guests can't breakpoint the hypervisor */
760 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
762 vcpu->arch.ciabr = value1;
764 case H_SET_MODE_RESOURCE_SET_DAWR:
765 if (!kvmppc_power8_compatible(vcpu))
767 if (!ppc_breakpoint_available())
770 return H_UNSUPPORTED_FLAG_START;
771 if (value2 & DABRX_HYP)
773 vcpu->arch.dawr = value1;
774 vcpu->arch.dawrx = value2;
781 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
783 struct kvmppc_vcore *vcore = target->arch.vcore;
786 * We expect to have been called by the real mode handler
787 * (kvmppc_rm_h_confer()) which would have directly returned
788 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
789 * have useful work to do and should not confer) so we don't
793 spin_lock(&vcore->lock);
794 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
795 vcore->vcore_state != VCORE_INACTIVE &&
797 target = vcore->runner;
798 spin_unlock(&vcore->lock);
800 return kvm_vcpu_yield_to(target);
803 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
806 struct lppaca *lppaca;
808 spin_lock(&vcpu->arch.vpa_update_lock);
809 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
811 yield_count = be32_to_cpu(lppaca->yield_count);
812 spin_unlock(&vcpu->arch.vpa_update_lock);
816 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
818 unsigned long req = kvmppc_get_gpr(vcpu, 3);
819 unsigned long target, ret = H_SUCCESS;
821 struct kvm_vcpu *tvcpu;
824 if (req <= MAX_HCALL_OPCODE &&
825 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
832 target = kvmppc_get_gpr(vcpu, 4);
833 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
838 tvcpu->arch.prodded = 1;
840 if (tvcpu->arch.ceded)
841 kvmppc_fast_vcpu_kick_hv(tvcpu);
844 target = kvmppc_get_gpr(vcpu, 4);
847 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
852 yield_count = kvmppc_get_gpr(vcpu, 5);
853 if (kvmppc_get_yield_count(tvcpu) != yield_count)
855 kvm_arch_vcpu_yield_to(tvcpu);
858 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
859 kvmppc_get_gpr(vcpu, 5),
860 kvmppc_get_gpr(vcpu, 6));
863 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
866 idx = srcu_read_lock(&vcpu->kvm->srcu);
867 rc = kvmppc_rtas_hcall(vcpu);
868 srcu_read_unlock(&vcpu->kvm->srcu, idx);
875 /* Send the error out to userspace via KVM_RUN */
877 case H_LOGICAL_CI_LOAD:
878 ret = kvmppc_h_logical_ci_load(vcpu);
879 if (ret == H_TOO_HARD)
882 case H_LOGICAL_CI_STORE:
883 ret = kvmppc_h_logical_ci_store(vcpu);
884 if (ret == H_TOO_HARD)
888 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
889 kvmppc_get_gpr(vcpu, 5),
890 kvmppc_get_gpr(vcpu, 6),
891 kvmppc_get_gpr(vcpu, 7));
892 if (ret == H_TOO_HARD)
901 if (kvmppc_xics_enabled(vcpu)) {
902 if (xive_enabled()) {
903 ret = H_NOT_AVAILABLE;
906 ret = kvmppc_xics_hcall(vcpu, req);
911 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
912 kvmppc_get_gpr(vcpu, 5),
913 kvmppc_get_gpr(vcpu, 6));
914 if (ret == H_TOO_HARD)
917 case H_PUT_TCE_INDIRECT:
918 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
919 kvmppc_get_gpr(vcpu, 5),
920 kvmppc_get_gpr(vcpu, 6),
921 kvmppc_get_gpr(vcpu, 7));
922 if (ret == H_TOO_HARD)
926 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
927 kvmppc_get_gpr(vcpu, 5),
928 kvmppc_get_gpr(vcpu, 6),
929 kvmppc_get_gpr(vcpu, 7));
930 if (ret == H_TOO_HARD)
936 kvmppc_set_gpr(vcpu, 3, ret);
937 vcpu->arch.hcall_needed = 0;
941 static int kvmppc_hcall_impl_hv(unsigned long cmd)
949 case H_LOGICAL_CI_LOAD:
950 case H_LOGICAL_CI_STORE:
951 #ifdef CONFIG_KVM_XICS
962 /* See if it's in the real-mode table */
963 return kvmppc_hcall_impl_hv_realmode(cmd);
966 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
967 struct kvm_vcpu *vcpu)
971 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
974 * Fetch failed, so return to guest and
975 * try executing it again.
980 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
981 run->exit_reason = KVM_EXIT_DEBUG;
982 run->debug.arch.address = kvmppc_get_pc(vcpu);
985 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
990 static void do_nothing(void *x)
994 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
996 int thr, cpu, pcpu, nthreads;
1000 nthreads = vcpu->kvm->arch.emul_smt_mode;
1002 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1003 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1004 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1008 * If the vcpu is currently running on a physical cpu thread,
1009 * interrupt it in order to pull it out of the guest briefly,
1010 * which will update its vcore->dpdes value.
1012 pcpu = READ_ONCE(v->cpu);
1014 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1015 if (kvmppc_doorbell_pending(v))
1022 * On POWER9, emulate doorbell-related instructions in order to
1023 * give the guest the illusion of running on a multi-threaded core.
1024 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1027 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1031 struct kvm *kvm = vcpu->kvm;
1032 struct kvm_vcpu *tvcpu;
1034 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1035 return RESUME_GUEST;
1036 if (get_op(inst) != 31)
1037 return EMULATE_FAIL;
1039 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1040 switch (get_xop(inst)) {
1041 case OP_31_XOP_MSGSNDP:
1042 arg = kvmppc_get_gpr(vcpu, rb);
1043 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1046 if (arg >= kvm->arch.emul_smt_mode)
1048 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1051 if (!tvcpu->arch.doorbell_request) {
1052 tvcpu->arch.doorbell_request = 1;
1053 kvmppc_fast_vcpu_kick_hv(tvcpu);
1056 case OP_31_XOP_MSGCLRP:
1057 arg = kvmppc_get_gpr(vcpu, rb);
1058 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1060 vcpu->arch.vcore->dpdes = 0;
1061 vcpu->arch.doorbell_request = 0;
1063 case OP_31_XOP_MFSPR:
1064 switch (get_sprn(inst)) {
1069 arg = kvmppc_read_dpdes(vcpu);
1072 return EMULATE_FAIL;
1074 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1077 return EMULATE_FAIL;
1079 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1080 return RESUME_GUEST;
1083 /* Called with vcpu->arch.vcore->lock held */
1084 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1085 struct task_struct *tsk)
1087 int r = RESUME_HOST;
1089 vcpu->stat.sum_exits++;
1092 * This can happen if an interrupt occurs in the last stages
1093 * of guest entry or the first stages of guest exit (i.e. after
1094 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1095 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1096 * That can happen due to a bug, or due to a machine check
1097 * occurring at just the wrong time.
1099 if (vcpu->arch.shregs.msr & MSR_HV) {
1100 printk(KERN_EMERG "KVM trap in HV mode!\n");
1101 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1102 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1103 vcpu->arch.shregs.msr);
1104 kvmppc_dump_regs(vcpu);
1105 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1106 run->hw.hardware_exit_reason = vcpu->arch.trap;
1109 run->exit_reason = KVM_EXIT_UNKNOWN;
1110 run->ready_for_interrupt_injection = 1;
1111 switch (vcpu->arch.trap) {
1112 /* We're good on these - the host merely wanted to get our attention */
1113 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1114 vcpu->stat.dec_exits++;
1117 case BOOK3S_INTERRUPT_EXTERNAL:
1118 case BOOK3S_INTERRUPT_H_DOORBELL:
1119 case BOOK3S_INTERRUPT_H_VIRT:
1120 vcpu->stat.ext_intr_exits++;
1123 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1124 case BOOK3S_INTERRUPT_HMI:
1125 case BOOK3S_INTERRUPT_PERFMON:
1126 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1129 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1130 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1131 run->exit_reason = KVM_EXIT_NMI;
1132 run->hw.hardware_exit_reason = vcpu->arch.trap;
1133 /* Clear out the old NMI status from run->flags */
1134 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1135 /* Now set the NMI status */
1136 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1137 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1139 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1142 /* Print the MCE event to host console. */
1143 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1145 case BOOK3S_INTERRUPT_PROGRAM:
1149 * Normally program interrupts are delivered directly
1150 * to the guest by the hardware, but we can get here
1151 * as a result of a hypervisor emulation interrupt
1152 * (e40) getting turned into a 700 by BML RTAS.
1154 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1155 kvmppc_core_queue_program(vcpu, flags);
1159 case BOOK3S_INTERRUPT_SYSCALL:
1161 /* hcall - punt to userspace */
1164 /* hypercall with MSR_PR has already been handled in rmode,
1165 * and never reaches here.
1168 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1169 for (i = 0; i < 9; ++i)
1170 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1171 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1172 vcpu->arch.hcall_needed = 1;
1177 * We get these next two if the guest accesses a page which it thinks
1178 * it has mapped but which is not actually present, either because
1179 * it is for an emulated I/O device or because the corresonding
1180 * host page has been paged out. Any other HDSI/HISI interrupts
1181 * have been handled already.
1183 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1184 r = RESUME_PAGE_FAULT;
1186 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1187 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1188 vcpu->arch.fault_dsisr = 0;
1189 r = RESUME_PAGE_FAULT;
1192 * This occurs if the guest executes an illegal instruction.
1193 * If the guest debug is disabled, generate a program interrupt
1194 * to the guest. If guest debug is enabled, we need to check
1195 * whether the instruction is a software breakpoint instruction.
1196 * Accordingly return to Guest or Host.
1198 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1199 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1200 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1201 swab32(vcpu->arch.emul_inst) :
1202 vcpu->arch.emul_inst;
1203 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1204 /* Need vcore unlocked to call kvmppc_get_last_inst */
1205 spin_unlock(&vcpu->arch.vcore->lock);
1206 r = kvmppc_emulate_debug_inst(run, vcpu);
1207 spin_lock(&vcpu->arch.vcore->lock);
1209 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1214 * This occurs if the guest (kernel or userspace), does something that
1215 * is prohibited by HFSCR.
1216 * On POWER9, this could be a doorbell instruction that we need
1218 * Otherwise, we just generate a program interrupt to the guest.
1220 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1222 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1223 cpu_has_feature(CPU_FTR_ARCH_300)) {
1224 /* Need vcore unlocked to call kvmppc_get_last_inst */
1225 spin_unlock(&vcpu->arch.vcore->lock);
1226 r = kvmppc_emulate_doorbell_instr(vcpu);
1227 spin_lock(&vcpu->arch.vcore->lock);
1229 if (r == EMULATE_FAIL) {
1230 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1235 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1236 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1238 * This occurs for various TM-related instructions that
1239 * we need to emulate on POWER9 DD2.2. We have already
1240 * handled the cases where the guest was in real-suspend
1241 * mode and was transitioning to transactional state.
1243 r = kvmhv_p9_tm_emulation(vcpu);
1247 case BOOK3S_INTERRUPT_HV_RM_HARD:
1248 r = RESUME_PASSTHROUGH;
1251 kvmppc_dump_regs(vcpu);
1252 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1253 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1254 vcpu->arch.shregs.msr);
1255 run->hw.hardware_exit_reason = vcpu->arch.trap;
1263 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1264 struct kvm_sregs *sregs)
1268 memset(sregs, 0, sizeof(struct kvm_sregs));
1269 sregs->pvr = vcpu->arch.pvr;
1270 for (i = 0; i < vcpu->arch.slb_max; i++) {
1271 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1272 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1278 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1279 struct kvm_sregs *sregs)
1283 /* Only accept the same PVR as the host's, since we can't spoof it */
1284 if (sregs->pvr != vcpu->arch.pvr)
1288 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1289 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1290 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1291 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1295 vcpu->arch.slb_max = j;
1300 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1301 bool preserve_top32)
1303 struct kvm *kvm = vcpu->kvm;
1304 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1307 spin_lock(&vc->lock);
1309 * If ILE (interrupt little-endian) has changed, update the
1310 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1312 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1313 struct kvm_vcpu *vcpu;
1316 kvm_for_each_vcpu(i, vcpu, kvm) {
1317 if (vcpu->arch.vcore != vc)
1319 if (new_lpcr & LPCR_ILE)
1320 vcpu->arch.intr_msr |= MSR_LE;
1322 vcpu->arch.intr_msr &= ~MSR_LE;
1327 * Userspace can only modify DPFD (default prefetch depth),
1328 * ILE (interrupt little-endian) and TC (translation control).
1329 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1331 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1332 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1335 * On POWER9, allow userspace to enable large decrementer for the
1336 * guest, whether or not the host has it enabled.
1338 if (cpu_has_feature(CPU_FTR_ARCH_300))
1341 /* Broken 32-bit version of LPCR must not clear top bits */
1344 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1345 spin_unlock(&vc->lock);
1348 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1349 union kvmppc_one_reg *val)
1355 case KVM_REG_PPC_DEBUG_INST:
1356 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1358 case KVM_REG_PPC_HIOR:
1359 *val = get_reg_val(id, 0);
1361 case KVM_REG_PPC_DABR:
1362 *val = get_reg_val(id, vcpu->arch.dabr);
1364 case KVM_REG_PPC_DABRX:
1365 *val = get_reg_val(id, vcpu->arch.dabrx);
1367 case KVM_REG_PPC_DSCR:
1368 *val = get_reg_val(id, vcpu->arch.dscr);
1370 case KVM_REG_PPC_PURR:
1371 *val = get_reg_val(id, vcpu->arch.purr);
1373 case KVM_REG_PPC_SPURR:
1374 *val = get_reg_val(id, vcpu->arch.spurr);
1376 case KVM_REG_PPC_AMR:
1377 *val = get_reg_val(id, vcpu->arch.amr);
1379 case KVM_REG_PPC_UAMOR:
1380 *val = get_reg_val(id, vcpu->arch.uamor);
1382 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1383 i = id - KVM_REG_PPC_MMCR0;
1384 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1386 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1387 i = id - KVM_REG_PPC_PMC1;
1388 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1390 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1391 i = id - KVM_REG_PPC_SPMC1;
1392 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1394 case KVM_REG_PPC_SIAR:
1395 *val = get_reg_val(id, vcpu->arch.siar);
1397 case KVM_REG_PPC_SDAR:
1398 *val = get_reg_val(id, vcpu->arch.sdar);
1400 case KVM_REG_PPC_SIER:
1401 *val = get_reg_val(id, vcpu->arch.sier);
1403 case KVM_REG_PPC_IAMR:
1404 *val = get_reg_val(id, vcpu->arch.iamr);
1406 case KVM_REG_PPC_PSPB:
1407 *val = get_reg_val(id, vcpu->arch.pspb);
1409 case KVM_REG_PPC_DPDES:
1411 * On POWER9, where we are emulating msgsndp etc.,
1412 * we return 1 bit for each vcpu, which can come from
1413 * either vcore->dpdes or doorbell_request.
1414 * On POWER8, doorbell_request is 0.
1416 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1417 vcpu->arch.doorbell_request);
1419 case KVM_REG_PPC_VTB:
1420 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1422 case KVM_REG_PPC_DAWR:
1423 *val = get_reg_val(id, vcpu->arch.dawr);
1425 case KVM_REG_PPC_DAWRX:
1426 *val = get_reg_val(id, vcpu->arch.dawrx);
1428 case KVM_REG_PPC_CIABR:
1429 *val = get_reg_val(id, vcpu->arch.ciabr);
1431 case KVM_REG_PPC_CSIGR:
1432 *val = get_reg_val(id, vcpu->arch.csigr);
1434 case KVM_REG_PPC_TACR:
1435 *val = get_reg_val(id, vcpu->arch.tacr);
1437 case KVM_REG_PPC_TCSCR:
1438 *val = get_reg_val(id, vcpu->arch.tcscr);
1440 case KVM_REG_PPC_PID:
1441 *val = get_reg_val(id, vcpu->arch.pid);
1443 case KVM_REG_PPC_ACOP:
1444 *val = get_reg_val(id, vcpu->arch.acop);
1446 case KVM_REG_PPC_WORT:
1447 *val = get_reg_val(id, vcpu->arch.wort);
1449 case KVM_REG_PPC_TIDR:
1450 *val = get_reg_val(id, vcpu->arch.tid);
1452 case KVM_REG_PPC_PSSCR:
1453 *val = get_reg_val(id, vcpu->arch.psscr);
1455 case KVM_REG_PPC_VPA_ADDR:
1456 spin_lock(&vcpu->arch.vpa_update_lock);
1457 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1458 spin_unlock(&vcpu->arch.vpa_update_lock);
1460 case KVM_REG_PPC_VPA_SLB:
1461 spin_lock(&vcpu->arch.vpa_update_lock);
1462 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1463 val->vpaval.length = vcpu->arch.slb_shadow.len;
1464 spin_unlock(&vcpu->arch.vpa_update_lock);
1466 case KVM_REG_PPC_VPA_DTL:
1467 spin_lock(&vcpu->arch.vpa_update_lock);
1468 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1469 val->vpaval.length = vcpu->arch.dtl.len;
1470 spin_unlock(&vcpu->arch.vpa_update_lock);
1472 case KVM_REG_PPC_TB_OFFSET:
1473 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1475 case KVM_REG_PPC_LPCR:
1476 case KVM_REG_PPC_LPCR_64:
1477 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1479 case KVM_REG_PPC_PPR:
1480 *val = get_reg_val(id, vcpu->arch.ppr);
1482 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1483 case KVM_REG_PPC_TFHAR:
1484 *val = get_reg_val(id, vcpu->arch.tfhar);
1486 case KVM_REG_PPC_TFIAR:
1487 *val = get_reg_val(id, vcpu->arch.tfiar);
1489 case KVM_REG_PPC_TEXASR:
1490 *val = get_reg_val(id, vcpu->arch.texasr);
1492 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1493 i = id - KVM_REG_PPC_TM_GPR0;
1494 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1496 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1499 i = id - KVM_REG_PPC_TM_VSR0;
1501 for (j = 0; j < TS_FPRWIDTH; j++)
1502 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1504 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1505 val->vval = vcpu->arch.vr_tm.vr[i-32];
1511 case KVM_REG_PPC_TM_CR:
1512 *val = get_reg_val(id, vcpu->arch.cr_tm);
1514 case KVM_REG_PPC_TM_XER:
1515 *val = get_reg_val(id, vcpu->arch.xer_tm);
1517 case KVM_REG_PPC_TM_LR:
1518 *val = get_reg_val(id, vcpu->arch.lr_tm);
1520 case KVM_REG_PPC_TM_CTR:
1521 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1523 case KVM_REG_PPC_TM_FPSCR:
1524 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1526 case KVM_REG_PPC_TM_AMR:
1527 *val = get_reg_val(id, vcpu->arch.amr_tm);
1529 case KVM_REG_PPC_TM_PPR:
1530 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1532 case KVM_REG_PPC_TM_VRSAVE:
1533 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1535 case KVM_REG_PPC_TM_VSCR:
1536 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1537 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1541 case KVM_REG_PPC_TM_DSCR:
1542 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1544 case KVM_REG_PPC_TM_TAR:
1545 *val = get_reg_val(id, vcpu->arch.tar_tm);
1548 case KVM_REG_PPC_ARCH_COMPAT:
1549 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1551 case KVM_REG_PPC_DEC_EXPIRY:
1552 *val = get_reg_val(id, vcpu->arch.dec_expires +
1553 vcpu->arch.vcore->tb_offset);
1555 case KVM_REG_PPC_ONLINE:
1556 *val = get_reg_val(id, vcpu->arch.online);
1566 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1567 union kvmppc_one_reg *val)
1571 unsigned long addr, len;
1574 case KVM_REG_PPC_HIOR:
1575 /* Only allow this to be set to zero */
1576 if (set_reg_val(id, *val))
1579 case KVM_REG_PPC_DABR:
1580 vcpu->arch.dabr = set_reg_val(id, *val);
1582 case KVM_REG_PPC_DABRX:
1583 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1585 case KVM_REG_PPC_DSCR:
1586 vcpu->arch.dscr = set_reg_val(id, *val);
1588 case KVM_REG_PPC_PURR:
1589 vcpu->arch.purr = set_reg_val(id, *val);
1591 case KVM_REG_PPC_SPURR:
1592 vcpu->arch.spurr = set_reg_val(id, *val);
1594 case KVM_REG_PPC_AMR:
1595 vcpu->arch.amr = set_reg_val(id, *val);
1597 case KVM_REG_PPC_UAMOR:
1598 vcpu->arch.uamor = set_reg_val(id, *val);
1600 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1601 i = id - KVM_REG_PPC_MMCR0;
1602 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1604 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1605 i = id - KVM_REG_PPC_PMC1;
1606 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1608 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1609 i = id - KVM_REG_PPC_SPMC1;
1610 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1612 case KVM_REG_PPC_SIAR:
1613 vcpu->arch.siar = set_reg_val(id, *val);
1615 case KVM_REG_PPC_SDAR:
1616 vcpu->arch.sdar = set_reg_val(id, *val);
1618 case KVM_REG_PPC_SIER:
1619 vcpu->arch.sier = set_reg_val(id, *val);
1621 case KVM_REG_PPC_IAMR:
1622 vcpu->arch.iamr = set_reg_val(id, *val);
1624 case KVM_REG_PPC_PSPB:
1625 vcpu->arch.pspb = set_reg_val(id, *val);
1627 case KVM_REG_PPC_DPDES:
1628 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1630 case KVM_REG_PPC_VTB:
1631 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1633 case KVM_REG_PPC_DAWR:
1634 vcpu->arch.dawr = set_reg_val(id, *val);
1636 case KVM_REG_PPC_DAWRX:
1637 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1639 case KVM_REG_PPC_CIABR:
1640 vcpu->arch.ciabr = set_reg_val(id, *val);
1641 /* Don't allow setting breakpoints in hypervisor code */
1642 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1643 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1645 case KVM_REG_PPC_CSIGR:
1646 vcpu->arch.csigr = set_reg_val(id, *val);
1648 case KVM_REG_PPC_TACR:
1649 vcpu->arch.tacr = set_reg_val(id, *val);
1651 case KVM_REG_PPC_TCSCR:
1652 vcpu->arch.tcscr = set_reg_val(id, *val);
1654 case KVM_REG_PPC_PID:
1655 vcpu->arch.pid = set_reg_val(id, *val);
1657 case KVM_REG_PPC_ACOP:
1658 vcpu->arch.acop = set_reg_val(id, *val);
1660 case KVM_REG_PPC_WORT:
1661 vcpu->arch.wort = set_reg_val(id, *val);
1663 case KVM_REG_PPC_TIDR:
1664 vcpu->arch.tid = set_reg_val(id, *val);
1666 case KVM_REG_PPC_PSSCR:
1667 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1669 case KVM_REG_PPC_VPA_ADDR:
1670 addr = set_reg_val(id, *val);
1672 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1673 vcpu->arch.dtl.next_gpa))
1675 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1677 case KVM_REG_PPC_VPA_SLB:
1678 addr = val->vpaval.addr;
1679 len = val->vpaval.length;
1681 if (addr && !vcpu->arch.vpa.next_gpa)
1683 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1685 case KVM_REG_PPC_VPA_DTL:
1686 addr = val->vpaval.addr;
1687 len = val->vpaval.length;
1689 if (addr && (len < sizeof(struct dtl_entry) ||
1690 !vcpu->arch.vpa.next_gpa))
1692 len -= len % sizeof(struct dtl_entry);
1693 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1695 case KVM_REG_PPC_TB_OFFSET:
1696 /* round up to multiple of 2^24 */
1697 vcpu->arch.vcore->tb_offset =
1698 ALIGN(set_reg_val(id, *val), 1UL << 24);
1700 case KVM_REG_PPC_LPCR:
1701 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1703 case KVM_REG_PPC_LPCR_64:
1704 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1706 case KVM_REG_PPC_PPR:
1707 vcpu->arch.ppr = set_reg_val(id, *val);
1709 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1710 case KVM_REG_PPC_TFHAR:
1711 vcpu->arch.tfhar = set_reg_val(id, *val);
1713 case KVM_REG_PPC_TFIAR:
1714 vcpu->arch.tfiar = set_reg_val(id, *val);
1716 case KVM_REG_PPC_TEXASR:
1717 vcpu->arch.texasr = set_reg_val(id, *val);
1719 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1720 i = id - KVM_REG_PPC_TM_GPR0;
1721 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1723 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1726 i = id - KVM_REG_PPC_TM_VSR0;
1728 for (j = 0; j < TS_FPRWIDTH; j++)
1729 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1731 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1732 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1737 case KVM_REG_PPC_TM_CR:
1738 vcpu->arch.cr_tm = set_reg_val(id, *val);
1740 case KVM_REG_PPC_TM_XER:
1741 vcpu->arch.xer_tm = set_reg_val(id, *val);
1743 case KVM_REG_PPC_TM_LR:
1744 vcpu->arch.lr_tm = set_reg_val(id, *val);
1746 case KVM_REG_PPC_TM_CTR:
1747 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1749 case KVM_REG_PPC_TM_FPSCR:
1750 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1752 case KVM_REG_PPC_TM_AMR:
1753 vcpu->arch.amr_tm = set_reg_val(id, *val);
1755 case KVM_REG_PPC_TM_PPR:
1756 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1758 case KVM_REG_PPC_TM_VRSAVE:
1759 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1761 case KVM_REG_PPC_TM_VSCR:
1762 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1763 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1767 case KVM_REG_PPC_TM_DSCR:
1768 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1770 case KVM_REG_PPC_TM_TAR:
1771 vcpu->arch.tar_tm = set_reg_val(id, *val);
1774 case KVM_REG_PPC_ARCH_COMPAT:
1775 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1777 case KVM_REG_PPC_DEC_EXPIRY:
1778 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1779 vcpu->arch.vcore->tb_offset;
1781 case KVM_REG_PPC_ONLINE:
1782 i = set_reg_val(id, *val);
1783 if (i && !vcpu->arch.online)
1784 atomic_inc(&vcpu->arch.vcore->online_count);
1785 else if (!i && vcpu->arch.online)
1786 atomic_dec(&vcpu->arch.vcore->online_count);
1787 vcpu->arch.online = i;
1798 * On POWER9, threads are independent and can be in different partitions.
1799 * Therefore we consider each thread to be a subcore.
1800 * There is a restriction that all threads have to be in the same
1801 * MMU mode (radix or HPT), unfortunately, but since we only support
1802 * HPT guests on a HPT host so far, that isn't an impediment yet.
1804 static int threads_per_vcore(struct kvm *kvm)
1806 if (kvm->arch.threads_indep)
1808 return threads_per_subcore;
1811 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
1813 struct kvmppc_vcore *vcore;
1815 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1820 spin_lock_init(&vcore->lock);
1821 spin_lock_init(&vcore->stoltb_lock);
1822 init_swait_queue_head(&vcore->wq);
1823 vcore->preempt_tb = TB_NIL;
1824 vcore->lpcr = kvm->arch.lpcr;
1825 vcore->first_vcpuid = id;
1827 INIT_LIST_HEAD(&vcore->preempt_list);
1832 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1833 static struct debugfs_timings_element {
1837 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1838 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1839 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1840 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1841 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1844 #define N_TIMINGS (ARRAY_SIZE(timings))
1846 struct debugfs_timings_state {
1847 struct kvm_vcpu *vcpu;
1848 unsigned int buflen;
1849 char buf[N_TIMINGS * 100];
1852 static int debugfs_timings_open(struct inode *inode, struct file *file)
1854 struct kvm_vcpu *vcpu = inode->i_private;
1855 struct debugfs_timings_state *p;
1857 p = kzalloc(sizeof(*p), GFP_KERNEL);
1861 kvm_get_kvm(vcpu->kvm);
1863 file->private_data = p;
1865 return nonseekable_open(inode, file);
1868 static int debugfs_timings_release(struct inode *inode, struct file *file)
1870 struct debugfs_timings_state *p = file->private_data;
1872 kvm_put_kvm(p->vcpu->kvm);
1877 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1878 size_t len, loff_t *ppos)
1880 struct debugfs_timings_state *p = file->private_data;
1881 struct kvm_vcpu *vcpu = p->vcpu;
1883 struct kvmhv_tb_accumulator tb;
1892 buf_end = s + sizeof(p->buf);
1893 for (i = 0; i < N_TIMINGS; ++i) {
1894 struct kvmhv_tb_accumulator *acc;
1896 acc = (struct kvmhv_tb_accumulator *)
1897 ((unsigned long)vcpu + timings[i].offset);
1899 for (loops = 0; loops < 1000; ++loops) {
1900 count = acc->seqcount;
1905 if (count == acc->seqcount) {
1913 snprintf(s, buf_end - s, "%s: stuck\n",
1916 snprintf(s, buf_end - s,
1917 "%s: %llu %llu %llu %llu\n",
1918 timings[i].name, count / 2,
1919 tb_to_ns(tb.tb_total),
1920 tb_to_ns(tb.tb_min),
1921 tb_to_ns(tb.tb_max));
1924 p->buflen = s - p->buf;
1928 if (pos >= p->buflen)
1930 if (len > p->buflen - pos)
1931 len = p->buflen - pos;
1932 n = copy_to_user(buf, p->buf + pos, len);
1942 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1943 size_t len, loff_t *ppos)
1948 static const struct file_operations debugfs_timings_ops = {
1949 .owner = THIS_MODULE,
1950 .open = debugfs_timings_open,
1951 .release = debugfs_timings_release,
1952 .read = debugfs_timings_read,
1953 .write = debugfs_timings_write,
1954 .llseek = generic_file_llseek,
1957 /* Create a debugfs directory for the vcpu */
1958 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1961 struct kvm *kvm = vcpu->kvm;
1963 snprintf(buf, sizeof(buf), "vcpu%u", id);
1964 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1966 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1967 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1969 vcpu->arch.debugfs_timings =
1970 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1971 vcpu, &debugfs_timings_ops);
1974 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1975 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1978 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1980 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1983 struct kvm_vcpu *vcpu;
1986 struct kvmppc_vcore *vcore;
1989 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1993 err = kvm_vcpu_init(vcpu, kvm, id);
1997 vcpu->arch.shared = &vcpu->arch.shregs;
1998 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2000 * The shared struct is never shared on HV,
2001 * so we can always use host endianness
2003 #ifdef __BIG_ENDIAN__
2004 vcpu->arch.shared_big_endian = true;
2006 vcpu->arch.shared_big_endian = false;
2009 vcpu->arch.mmcr[0] = MMCR0_FC;
2010 vcpu->arch.ctrl = CTRL_RUNLATCH;
2011 /* default to host PVR, since we can't spoof it */
2012 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2013 spin_lock_init(&vcpu->arch.vpa_update_lock);
2014 spin_lock_init(&vcpu->arch.tbacct_lock);
2015 vcpu->arch.busy_preempt = TB_NIL;
2016 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2019 * Set the default HFSCR for the guest from the host value.
2020 * This value is only used on POWER9.
2021 * On POWER9, we want to virtualize the doorbell facility, so we
2022 * turn off the HFSCR bit, which causes those instructions to trap.
2024 vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
2025 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2026 vcpu->arch.hfscr |= HFSCR_TM;
2027 else if (!cpu_has_feature(CPU_FTR_TM_COMP))
2028 vcpu->arch.hfscr &= ~HFSCR_TM;
2029 if (cpu_has_feature(CPU_FTR_ARCH_300))
2030 vcpu->arch.hfscr &= ~HFSCR_MSGP;
2032 kvmppc_mmu_book3s_hv_init(vcpu);
2034 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2036 init_waitqueue_head(&vcpu->arch.cpu_run);
2038 mutex_lock(&kvm->lock);
2041 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2042 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2043 pr_devel("KVM: VCPU ID too high\n");
2044 core = KVM_MAX_VCORES;
2046 BUG_ON(kvm->arch.smt_mode != 1);
2047 core = kvmppc_pack_vcpu_id(kvm, id);
2050 core = id / kvm->arch.smt_mode;
2052 if (core < KVM_MAX_VCORES) {
2053 vcore = kvm->arch.vcores[core];
2054 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2055 pr_devel("KVM: collision on id %u", id);
2057 } else if (!vcore) {
2059 vcore = kvmppc_vcore_create(kvm,
2060 id & ~(kvm->arch.smt_mode - 1));
2061 kvm->arch.vcores[core] = vcore;
2062 kvm->arch.online_vcores++;
2065 mutex_unlock(&kvm->lock);
2070 spin_lock(&vcore->lock);
2071 ++vcore->num_threads;
2072 spin_unlock(&vcore->lock);
2073 vcpu->arch.vcore = vcore;
2074 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2075 vcpu->arch.thread_cpu = -1;
2076 vcpu->arch.prev_cpu = -1;
2078 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2079 kvmppc_sanity_check(vcpu);
2081 debugfs_vcpu_init(vcpu, id);
2086 kvm_vcpu_uninit(vcpu);
2088 kmem_cache_free(kvm_vcpu_cache, vcpu);
2090 return ERR_PTR(err);
2093 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2094 unsigned long flags)
2101 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2103 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2105 * On POWER8 (or POWER7), the threading mode is "strict",
2106 * so we pack smt_mode vcpus per vcore.
2108 if (smt_mode > threads_per_subcore)
2112 * On POWER9, the threading mode is "loose",
2113 * so each vcpu gets its own vcore.
2118 mutex_lock(&kvm->lock);
2120 if (!kvm->arch.online_vcores) {
2121 kvm->arch.smt_mode = smt_mode;
2122 kvm->arch.emul_smt_mode = esmt;
2125 mutex_unlock(&kvm->lock);
2130 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2132 if (vpa->pinned_addr)
2133 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2137 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2139 spin_lock(&vcpu->arch.vpa_update_lock);
2140 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2141 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2142 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2143 spin_unlock(&vcpu->arch.vpa_update_lock);
2144 kvm_vcpu_uninit(vcpu);
2145 kmem_cache_free(kvm_vcpu_cache, vcpu);
2148 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2150 /* Indicate we want to get back into the guest */
2154 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2156 unsigned long dec_nsec, now;
2159 if (now > vcpu->arch.dec_expires) {
2160 /* decrementer has already gone negative */
2161 kvmppc_core_queue_dec(vcpu);
2162 kvmppc_core_prepare_to_enter(vcpu);
2165 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2167 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2168 vcpu->arch.timer_running = 1;
2171 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2173 vcpu->arch.ceded = 0;
2174 if (vcpu->arch.timer_running) {
2175 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2176 vcpu->arch.timer_running = 0;
2180 extern int __kvmppc_vcore_entry(void);
2182 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2183 struct kvm_vcpu *vcpu)
2187 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2189 spin_lock_irq(&vcpu->arch.tbacct_lock);
2191 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2192 vcpu->arch.stolen_logged;
2193 vcpu->arch.busy_preempt = now;
2194 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2195 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2197 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2200 static int kvmppc_grab_hwthread(int cpu)
2202 struct paca_struct *tpaca;
2203 long timeout = 10000;
2205 tpaca = paca_ptrs[cpu];
2207 /* Ensure the thread won't go into the kernel if it wakes */
2208 tpaca->kvm_hstate.kvm_vcpu = NULL;
2209 tpaca->kvm_hstate.kvm_vcore = NULL;
2210 tpaca->kvm_hstate.napping = 0;
2212 tpaca->kvm_hstate.hwthread_req = 1;
2215 * If the thread is already executing in the kernel (e.g. handling
2216 * a stray interrupt), wait for it to get back to nap mode.
2217 * The smp_mb() is to ensure that our setting of hwthread_req
2218 * is visible before we look at hwthread_state, so if this
2219 * races with the code at system_reset_pSeries and the thread
2220 * misses our setting of hwthread_req, we are sure to see its
2221 * setting of hwthread_state, and vice versa.
2224 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2225 if (--timeout <= 0) {
2226 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2234 static void kvmppc_release_hwthread(int cpu)
2236 struct paca_struct *tpaca;
2238 tpaca = paca_ptrs[cpu];
2239 tpaca->kvm_hstate.hwthread_req = 0;
2240 tpaca->kvm_hstate.kvm_vcpu = NULL;
2241 tpaca->kvm_hstate.kvm_vcore = NULL;
2242 tpaca->kvm_hstate.kvm_split_mode = NULL;
2245 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2249 cpu = cpu_first_thread_sibling(cpu);
2250 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2252 * Make sure setting of bit in need_tlb_flush precedes
2253 * testing of cpu_in_guest bits. The matching barrier on
2254 * the other side is the first smp_mb() in kvmppc_run_core().
2257 for (i = 0; i < threads_per_core; ++i)
2258 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2259 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2262 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2264 struct kvm *kvm = vcpu->kvm;
2267 * With radix, the guest can do TLB invalidations itself,
2268 * and it could choose to use the local form (tlbiel) if
2269 * it is invalidating a translation that has only ever been
2270 * used on one vcpu. However, that doesn't mean it has
2271 * only ever been used on one physical cpu, since vcpus
2272 * can move around between pcpus. To cope with this, when
2273 * a vcpu moves from one pcpu to another, we need to tell
2274 * any vcpus running on the same core as this vcpu previously
2275 * ran to flush the TLB. The TLB is shared between threads,
2276 * so we use a single bit in .need_tlb_flush for all 4 threads.
2278 if (vcpu->arch.prev_cpu != pcpu) {
2279 if (vcpu->arch.prev_cpu >= 0 &&
2280 cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2281 cpu_first_thread_sibling(pcpu))
2282 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2283 vcpu->arch.prev_cpu = pcpu;
2287 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2290 struct paca_struct *tpaca;
2291 struct kvm *kvm = vc->kvm;
2295 if (vcpu->arch.timer_running) {
2296 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2297 vcpu->arch.timer_running = 0;
2299 cpu += vcpu->arch.ptid;
2300 vcpu->cpu = vc->pcpu;
2301 vcpu->arch.thread_cpu = cpu;
2302 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2304 tpaca = paca_ptrs[cpu];
2305 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2306 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2307 tpaca->kvm_hstate.fake_suspend = 0;
2308 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2310 tpaca->kvm_hstate.kvm_vcore = vc;
2311 if (cpu != smp_processor_id())
2312 kvmppc_ipi_thread(cpu);
2315 static void kvmppc_wait_for_nap(int n_threads)
2317 int cpu = smp_processor_id();
2322 for (loops = 0; loops < 1000000; ++loops) {
2324 * Check if all threads are finished.
2325 * We set the vcore pointer when starting a thread
2326 * and the thread clears it when finished, so we look
2327 * for any threads that still have a non-NULL vcore ptr.
2329 for (i = 1; i < n_threads; ++i)
2330 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2332 if (i == n_threads) {
2339 for (i = 1; i < n_threads; ++i)
2340 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2341 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2345 * Check that we are on thread 0 and that any other threads in
2346 * this core are off-line. Then grab the threads so they can't
2349 static int on_primary_thread(void)
2351 int cpu = smp_processor_id();
2354 /* Are we on a primary subcore? */
2355 if (cpu_thread_in_subcore(cpu))
2359 while (++thr < threads_per_subcore)
2360 if (cpu_online(cpu + thr))
2363 /* Grab all hw threads so they can't go into the kernel */
2364 for (thr = 1; thr < threads_per_subcore; ++thr) {
2365 if (kvmppc_grab_hwthread(cpu + thr)) {
2366 /* Couldn't grab one; let the others go */
2368 kvmppc_release_hwthread(cpu + thr);
2369 } while (--thr > 0);
2377 * A list of virtual cores for each physical CPU.
2378 * These are vcores that could run but their runner VCPU tasks are
2379 * (or may be) preempted.
2381 struct preempted_vcore_list {
2382 struct list_head list;
2386 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2388 static void init_vcore_lists(void)
2392 for_each_possible_cpu(cpu) {
2393 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2394 spin_lock_init(&lp->lock);
2395 INIT_LIST_HEAD(&lp->list);
2399 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2401 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2403 vc->vcore_state = VCORE_PREEMPT;
2404 vc->pcpu = smp_processor_id();
2405 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2406 spin_lock(&lp->lock);
2407 list_add_tail(&vc->preempt_list, &lp->list);
2408 spin_unlock(&lp->lock);
2411 /* Start accumulating stolen time */
2412 kvmppc_core_start_stolen(vc);
2415 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2417 struct preempted_vcore_list *lp;
2419 kvmppc_core_end_stolen(vc);
2420 if (!list_empty(&vc->preempt_list)) {
2421 lp = &per_cpu(preempted_vcores, vc->pcpu);
2422 spin_lock(&lp->lock);
2423 list_del_init(&vc->preempt_list);
2424 spin_unlock(&lp->lock);
2426 vc->vcore_state = VCORE_INACTIVE;
2430 * This stores information about the virtual cores currently
2431 * assigned to a physical core.
2435 int max_subcore_threads;
2437 int subcore_threads[MAX_SUBCORES];
2438 struct kvmppc_vcore *vc[MAX_SUBCORES];
2442 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2443 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2445 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2447 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2449 memset(cip, 0, sizeof(*cip));
2450 cip->n_subcores = 1;
2451 cip->max_subcore_threads = vc->num_threads;
2452 cip->total_threads = vc->num_threads;
2453 cip->subcore_threads[0] = vc->num_threads;
2457 static bool subcore_config_ok(int n_subcores, int n_threads)
2460 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2461 * split-core mode, with one thread per subcore.
2463 if (cpu_has_feature(CPU_FTR_ARCH_300))
2464 return n_subcores <= 4 && n_threads == 1;
2466 /* On POWER8, can only dynamically split if unsplit to begin with */
2467 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2469 if (n_subcores > MAX_SUBCORES)
2471 if (n_subcores > 1) {
2472 if (!(dynamic_mt_modes & 2))
2474 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2478 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2481 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2483 vc->entry_exit_map = 0;
2485 vc->napping_threads = 0;
2486 vc->conferring_threads = 0;
2487 vc->tb_offset_applied = 0;
2490 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2492 int n_threads = vc->num_threads;
2495 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2498 /* Some POWER9 chips require all threads to be in the same MMU mode */
2499 if (no_mixing_hpt_and_radix &&
2500 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2503 if (n_threads < cip->max_subcore_threads)
2504 n_threads = cip->max_subcore_threads;
2505 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2507 cip->max_subcore_threads = n_threads;
2509 sub = cip->n_subcores;
2511 cip->total_threads += vc->num_threads;
2512 cip->subcore_threads[sub] = vc->num_threads;
2514 init_vcore_to_run(vc);
2515 list_del_init(&vc->preempt_list);
2521 * Work out whether it is possible to piggyback the execution of
2522 * vcore *pvc onto the execution of the other vcores described in *cip.
2524 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2527 if (cip->total_threads + pvc->num_threads > target_threads)
2530 return can_dynamic_split(pvc, cip);
2533 static void prepare_threads(struct kvmppc_vcore *vc)
2536 struct kvm_vcpu *vcpu;
2538 for_each_runnable_thread(i, vcpu, vc) {
2539 if (signal_pending(vcpu->arch.run_task))
2540 vcpu->arch.ret = -EINTR;
2541 else if (vcpu->arch.vpa.update_pending ||
2542 vcpu->arch.slb_shadow.update_pending ||
2543 vcpu->arch.dtl.update_pending)
2544 vcpu->arch.ret = RESUME_GUEST;
2547 kvmppc_remove_runnable(vc, vcpu);
2548 wake_up(&vcpu->arch.cpu_run);
2552 static void collect_piggybacks(struct core_info *cip, int target_threads)
2554 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2555 struct kvmppc_vcore *pvc, *vcnext;
2557 spin_lock(&lp->lock);
2558 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2559 if (!spin_trylock(&pvc->lock))
2561 prepare_threads(pvc);
2562 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2563 list_del_init(&pvc->preempt_list);
2564 if (pvc->runner == NULL) {
2565 pvc->vcore_state = VCORE_INACTIVE;
2566 kvmppc_core_end_stolen(pvc);
2568 spin_unlock(&pvc->lock);
2571 if (!can_piggyback(pvc, cip, target_threads)) {
2572 spin_unlock(&pvc->lock);
2575 kvmppc_core_end_stolen(pvc);
2576 pvc->vcore_state = VCORE_PIGGYBACK;
2577 if (cip->total_threads >= target_threads)
2580 spin_unlock(&lp->lock);
2583 static bool recheck_signals_and_mmu(struct core_info *cip)
2586 struct kvm_vcpu *vcpu;
2587 struct kvmppc_vcore *vc;
2589 for (sub = 0; sub < cip->n_subcores; ++sub) {
2591 if (!vc->kvm->arch.mmu_ready)
2593 for_each_runnable_thread(i, vcpu, vc)
2594 if (signal_pending(vcpu->arch.run_task))
2600 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2602 int still_running = 0, i;
2605 struct kvm_vcpu *vcpu;
2607 spin_lock(&vc->lock);
2609 for_each_runnable_thread(i, vcpu, vc) {
2610 /* cancel pending dec exception if dec is positive */
2611 if (now < vcpu->arch.dec_expires &&
2612 kvmppc_core_pending_dec(vcpu))
2613 kvmppc_core_dequeue_dec(vcpu);
2615 trace_kvm_guest_exit(vcpu);
2618 if (vcpu->arch.trap)
2619 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2620 vcpu->arch.run_task);
2622 vcpu->arch.ret = ret;
2623 vcpu->arch.trap = 0;
2625 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2626 if (vcpu->arch.pending_exceptions)
2627 kvmppc_core_prepare_to_enter(vcpu);
2628 if (vcpu->arch.ceded)
2629 kvmppc_set_timer(vcpu);
2633 kvmppc_remove_runnable(vc, vcpu);
2634 wake_up(&vcpu->arch.cpu_run);
2638 if (still_running > 0) {
2639 kvmppc_vcore_preempt(vc);
2640 } else if (vc->runner) {
2641 vc->vcore_state = VCORE_PREEMPT;
2642 kvmppc_core_start_stolen(vc);
2644 vc->vcore_state = VCORE_INACTIVE;
2646 if (vc->n_runnable > 0 && vc->runner == NULL) {
2647 /* make sure there's a candidate runner awake */
2649 vcpu = next_runnable_thread(vc, &i);
2650 wake_up(&vcpu->arch.cpu_run);
2653 spin_unlock(&vc->lock);
2657 * Clear core from the list of active host cores as we are about to
2658 * enter the guest. Only do this if it is the primary thread of the
2659 * core (not if a subcore) that is entering the guest.
2661 static inline int kvmppc_clear_host_core(unsigned int cpu)
2665 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2668 * Memory barrier can be omitted here as we will do a smp_wmb()
2669 * later in kvmppc_start_thread and we need ensure that state is
2670 * visible to other CPUs only after we enter guest.
2672 core = cpu >> threads_shift;
2673 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2678 * Advertise this core as an active host core since we exited the guest
2679 * Only need to do this if it is the primary thread of the core that is
2682 static inline int kvmppc_set_host_core(unsigned int cpu)
2686 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2690 * Memory barrier can be omitted here because we do a spin_unlock
2691 * immediately after this which provides the memory barrier.
2693 core = cpu >> threads_shift;
2694 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2698 static void set_irq_happened(int trap)
2701 case BOOK3S_INTERRUPT_EXTERNAL:
2702 local_paca->irq_happened |= PACA_IRQ_EE;
2704 case BOOK3S_INTERRUPT_H_DOORBELL:
2705 local_paca->irq_happened |= PACA_IRQ_DBELL;
2707 case BOOK3S_INTERRUPT_HMI:
2708 local_paca->irq_happened |= PACA_IRQ_HMI;
2710 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2711 replay_system_reset();
2717 * Run a set of guest threads on a physical core.
2718 * Called with vc->lock held.
2720 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2722 struct kvm_vcpu *vcpu;
2725 struct core_info core_info;
2726 struct kvmppc_vcore *pvc;
2727 struct kvm_split_mode split_info, *sip;
2728 int split, subcore_size, active;
2731 unsigned long cmd_bit, stat_bit;
2734 int controlled_threads;
2740 * Remove from the list any threads that have a signal pending
2741 * or need a VPA update done
2743 prepare_threads(vc);
2745 /* if the runner is no longer runnable, let the caller pick a new one */
2746 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2752 init_vcore_to_run(vc);
2753 vc->preempt_tb = TB_NIL;
2756 * Number of threads that we will be controlling: the same as
2757 * the number of threads per subcore, except on POWER9,
2758 * where it's 1 because the threads are (mostly) independent.
2760 controlled_threads = threads_per_vcore(vc->kvm);
2763 * Make sure we are running on primary threads, and that secondary
2764 * threads are offline. Also check if the number of threads in this
2765 * guest are greater than the current system threads per guest.
2766 * On POWER9, we need to be not in independent-threads mode if
2767 * this is a HPT guest on a radix host machine where the
2768 * CPU threads may not be in different MMU modes.
2770 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
2771 !kvm_is_radix(vc->kvm);
2772 if (((controlled_threads > 1) &&
2773 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2774 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2775 for_each_runnable_thread(i, vcpu, vc) {
2776 vcpu->arch.ret = -EBUSY;
2777 kvmppc_remove_runnable(vc, vcpu);
2778 wake_up(&vcpu->arch.cpu_run);
2784 * See if we could run any other vcores on the physical core
2785 * along with this one.
2787 init_core_info(&core_info, vc);
2788 pcpu = smp_processor_id();
2789 target_threads = controlled_threads;
2790 if (target_smt_mode && target_smt_mode < target_threads)
2791 target_threads = target_smt_mode;
2792 if (vc->num_threads < target_threads)
2793 collect_piggybacks(&core_info, target_threads);
2796 * On radix, arrange for TLB flushing if necessary.
2797 * This has to be done before disabling interrupts since
2798 * it uses smp_call_function().
2800 pcpu = smp_processor_id();
2801 if (kvm_is_radix(vc->kvm)) {
2802 for (sub = 0; sub < core_info.n_subcores; ++sub)
2803 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2804 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2808 * Hard-disable interrupts, and check resched flag and signals.
2809 * If we need to reschedule or deliver a signal, clean up
2810 * and return without going into the guest(s).
2811 * If the mmu_ready flag has been cleared, don't go into the
2812 * guest because that means a HPT resize operation is in progress.
2814 local_irq_disable();
2816 if (lazy_irq_pending() || need_resched() ||
2817 recheck_signals_and_mmu(&core_info)) {
2819 vc->vcore_state = VCORE_INACTIVE;
2820 /* Unlock all except the primary vcore */
2821 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2822 pvc = core_info.vc[sub];
2823 /* Put back on to the preempted vcores list */
2824 kvmppc_vcore_preempt(pvc);
2825 spin_unlock(&pvc->lock);
2827 for (i = 0; i < controlled_threads; ++i)
2828 kvmppc_release_hwthread(pcpu + i);
2832 kvmppc_clear_host_core(pcpu);
2834 /* Decide on micro-threading (split-core) mode */
2835 subcore_size = threads_per_subcore;
2836 cmd_bit = stat_bit = 0;
2837 split = core_info.n_subcores;
2839 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2840 && !cpu_has_feature(CPU_FTR_ARCH_300);
2842 if (split > 1 || hpt_on_radix) {
2844 memset(&split_info, 0, sizeof(split_info));
2845 for (sub = 0; sub < core_info.n_subcores; ++sub)
2846 split_info.vc[sub] = core_info.vc[sub];
2849 if (split == 2 && (dynamic_mt_modes & 2)) {
2850 cmd_bit = HID0_POWER8_1TO2LPAR;
2851 stat_bit = HID0_POWER8_2LPARMODE;
2854 cmd_bit = HID0_POWER8_1TO4LPAR;
2855 stat_bit = HID0_POWER8_4LPARMODE;
2857 subcore_size = MAX_SMT_THREADS / split;
2858 split_info.rpr = mfspr(SPRN_RPR);
2859 split_info.pmmar = mfspr(SPRN_PMMAR);
2860 split_info.ldbar = mfspr(SPRN_LDBAR);
2861 split_info.subcore_size = subcore_size;
2863 split_info.subcore_size = 1;
2865 /* Use the split_info for LPCR/LPIDR changes */
2866 split_info.lpcr_req = vc->lpcr;
2867 split_info.lpidr_req = vc->kvm->arch.lpid;
2868 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2869 split_info.do_set = 1;
2873 /* order writes to split_info before kvm_split_mode pointer */
2877 for (thr = 0; thr < controlled_threads; ++thr) {
2878 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2880 paca->kvm_hstate.tid = thr;
2881 paca->kvm_hstate.napping = 0;
2882 paca->kvm_hstate.kvm_split_mode = sip;
2885 /* Initiate micro-threading (split-core) on POWER8 if required */
2887 unsigned long hid0 = mfspr(SPRN_HID0);
2889 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2891 mtspr(SPRN_HID0, hid0);
2894 hid0 = mfspr(SPRN_HID0);
2895 if (hid0 & stat_bit)
2902 * On POWER8, set RWMR register.
2903 * Since it only affects PURR and SPURR, it doesn't affect
2904 * the host, so we don't save/restore the host value.
2907 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
2908 int n_online = atomic_read(&vc->online_count);
2911 * Use the 8-thread value if we're doing split-core
2912 * or if the vcore's online count looks bogus.
2914 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
2915 n_online >= 1 && n_online <= MAX_SMT_THREADS)
2916 rwmr_val = p8_rwmr_values[n_online];
2917 mtspr(SPRN_RWMR, rwmr_val);
2920 /* Start all the threads */
2922 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2923 thr = is_power8 ? subcore_thread_map[sub] : sub;
2926 pvc = core_info.vc[sub];
2927 pvc->pcpu = pcpu + thr;
2928 for_each_runnable_thread(i, vcpu, pvc) {
2929 kvmppc_start_thread(vcpu, pvc);
2930 kvmppc_create_dtl_entry(vcpu, pvc);
2931 trace_kvm_guest_enter(vcpu);
2932 if (!vcpu->arch.ptid)
2934 active |= 1 << (thr + vcpu->arch.ptid);
2937 * We need to start the first thread of each subcore
2938 * even if it doesn't have a vcpu.
2941 kvmppc_start_thread(NULL, pvc);
2945 * Ensure that split_info.do_nap is set after setting
2946 * the vcore pointer in the PACA of the secondaries.
2951 * When doing micro-threading, poke the inactive threads as well.
2952 * This gets them to the nap instruction after kvm_do_nap,
2953 * which reduces the time taken to unsplit later.
2954 * For POWER9 HPT guest on radix host, we need all the secondary
2955 * threads woken up so they can do the LPCR/LPIDR change.
2957 if (cmd_bit || hpt_on_radix) {
2958 split_info.do_nap = 1; /* ask secondaries to nap when done */
2959 for (thr = 1; thr < threads_per_subcore; ++thr)
2960 if (!(active & (1 << thr)))
2961 kvmppc_ipi_thread(pcpu + thr);
2964 vc->vcore_state = VCORE_RUNNING;
2967 trace_kvmppc_run_core(vc, 0);
2969 for (sub = 0; sub < core_info.n_subcores; ++sub)
2970 spin_unlock(&core_info.vc[sub]->lock);
2972 if (kvm_is_radix(vc->kvm)) {
2976 * Do we need to flush the process scoped TLB for the LPAR?
2978 * On POWER9, individual threads can come in here, but the
2979 * TLB is shared between the 4 threads in a core, hence
2980 * invalidating on one thread invalidates for all.
2981 * Thus we make all 4 threads use the same bit here.
2983 * Hash must be flushed in realmode in order to use tlbiel.
2985 mtspr(SPRN_LPID, vc->kvm->arch.lpid);
2988 if (cpu_has_feature(CPU_FTR_ARCH_300))
2991 if (cpumask_test_cpu(tmp, &vc->kvm->arch.need_tlb_flush)) {
2992 radix__local_flush_tlb_lpid_guest(vc->kvm->arch.lpid);
2993 /* Clear the bit after the TLB flush */
2994 cpumask_clear_cpu(tmp, &vc->kvm->arch.need_tlb_flush);
2998 guest_enter_irqoff();
3000 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3002 this_cpu_disable_ftrace();
3005 * Interrupts will be enabled once we get into the guest,
3006 * so tell lockdep that we're about to enable interrupts.
3008 trace_hardirqs_on();
3010 trap = __kvmppc_vcore_entry();
3012 trace_hardirqs_off();
3014 this_cpu_enable_ftrace();
3016 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3018 set_irq_happened(trap);
3020 spin_lock(&vc->lock);
3021 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3022 vc->vcore_state = VCORE_EXITING;
3024 /* wait for secondary threads to finish writing their state to memory */
3025 kvmppc_wait_for_nap(controlled_threads);
3027 /* Return to whole-core mode if we split the core earlier */
3029 unsigned long hid0 = mfspr(SPRN_HID0);
3030 unsigned long loops = 0;
3032 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3033 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3035 mtspr(SPRN_HID0, hid0);
3038 hid0 = mfspr(SPRN_HID0);
3039 if (!(hid0 & stat_bit))
3044 } else if (hpt_on_radix) {
3045 /* Wait for all threads to have seen final sync */
3046 for (thr = 1; thr < controlled_threads; ++thr) {
3047 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3049 while (paca->kvm_hstate.kvm_split_mode) {
3056 split_info.do_nap = 0;
3058 kvmppc_set_host_core(pcpu);
3063 /* Let secondaries go back to the offline loop */
3064 for (i = 0; i < controlled_threads; ++i) {
3065 kvmppc_release_hwthread(pcpu + i);
3066 if (sip && sip->napped[i])
3067 kvmppc_ipi_thread(pcpu + i);
3068 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3071 spin_unlock(&vc->lock);
3073 /* make sure updates to secondary vcpu structs are visible now */
3078 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3079 pvc = core_info.vc[sub];
3080 post_guest_process(pvc, pvc == vc);
3083 spin_lock(&vc->lock);
3086 vc->vcore_state = VCORE_INACTIVE;
3087 trace_kvmppc_run_core(vc, 1);
3091 * Wait for some other vcpu thread to execute us, and
3092 * wake us up when we need to handle something in the host.
3094 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3095 struct kvm_vcpu *vcpu, int wait_state)
3099 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3100 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3101 spin_unlock(&vc->lock);
3103 spin_lock(&vc->lock);
3105 finish_wait(&vcpu->arch.cpu_run, &wait);
3108 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3111 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3112 vc->halt_poll_ns = 10000;
3114 vc->halt_poll_ns *= halt_poll_ns_grow;
3117 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3119 if (halt_poll_ns_shrink == 0)
3120 vc->halt_poll_ns = 0;
3122 vc->halt_poll_ns /= halt_poll_ns_shrink;
3125 #ifdef CONFIG_KVM_XICS
3126 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3128 if (!xive_enabled())
3130 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3131 vcpu->arch.xive_saved_state.cppr;
3134 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3138 #endif /* CONFIG_KVM_XICS */
3140 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3142 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3143 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3150 * Check to see if any of the runnable vcpus on the vcore have pending
3151 * exceptions or are no longer ceded
3153 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3155 struct kvm_vcpu *vcpu;
3158 for_each_runnable_thread(i, vcpu, vc) {
3159 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3167 * All the vcpus in this vcore are idle, so wait for a decrementer
3168 * or external interrupt to one of the vcpus. vc->lock is held.
3170 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3172 ktime_t cur, start_poll, start_wait;
3175 DECLARE_SWAITQUEUE(wait);
3177 /* Poll for pending exceptions and ceded state */
3178 cur = start_poll = ktime_get();
3179 if (vc->halt_poll_ns) {
3180 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3181 ++vc->runner->stat.halt_attempted_poll;
3183 vc->vcore_state = VCORE_POLLING;
3184 spin_unlock(&vc->lock);
3187 if (kvmppc_vcore_check_block(vc)) {
3192 } while (single_task_running() && ktime_before(cur, stop));
3194 spin_lock(&vc->lock);
3195 vc->vcore_state = VCORE_INACTIVE;
3198 ++vc->runner->stat.halt_successful_poll;
3203 prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3205 if (kvmppc_vcore_check_block(vc)) {
3206 finish_swait(&vc->wq, &wait);
3208 /* If we polled, count this as a successful poll */
3209 if (vc->halt_poll_ns)
3210 ++vc->runner->stat.halt_successful_poll;
3214 start_wait = ktime_get();
3216 vc->vcore_state = VCORE_SLEEPING;
3217 trace_kvmppc_vcore_blocked(vc, 0);
3218 spin_unlock(&vc->lock);
3220 finish_swait(&vc->wq, &wait);
3221 spin_lock(&vc->lock);
3222 vc->vcore_state = VCORE_INACTIVE;
3223 trace_kvmppc_vcore_blocked(vc, 1);
3224 ++vc->runner->stat.halt_successful_wait;
3229 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3231 /* Attribute wait time */
3233 vc->runner->stat.halt_wait_ns +=
3234 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3235 /* Attribute failed poll time */
3236 if (vc->halt_poll_ns)
3237 vc->runner->stat.halt_poll_fail_ns +=
3238 ktime_to_ns(start_wait) -
3239 ktime_to_ns(start_poll);
3241 /* Attribute successful poll time */
3242 if (vc->halt_poll_ns)
3243 vc->runner->stat.halt_poll_success_ns +=
3245 ktime_to_ns(start_poll);
3248 /* Adjust poll time */
3250 if (block_ns <= vc->halt_poll_ns)
3252 /* We slept and blocked for longer than the max halt time */
3253 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3254 shrink_halt_poll_ns(vc);
3255 /* We slept and our poll time is too small */
3256 else if (vc->halt_poll_ns < halt_poll_ns &&
3257 block_ns < halt_poll_ns)
3258 grow_halt_poll_ns(vc);
3259 if (vc->halt_poll_ns > halt_poll_ns)
3260 vc->halt_poll_ns = halt_poll_ns;
3262 vc->halt_poll_ns = 0;
3264 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3267 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3270 struct kvm *kvm = vcpu->kvm;
3272 mutex_lock(&kvm->lock);
3273 if (!kvm->arch.mmu_ready) {
3274 if (!kvm_is_radix(kvm))
3275 r = kvmppc_hv_setup_htab_rma(vcpu);
3277 if (cpu_has_feature(CPU_FTR_ARCH_300))
3278 kvmppc_setup_partition_table(kvm);
3279 kvm->arch.mmu_ready = 1;
3282 mutex_unlock(&kvm->lock);
3286 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3289 struct kvmppc_vcore *vc;
3292 trace_kvmppc_run_vcpu_enter(vcpu);
3294 kvm_run->exit_reason = 0;
3295 vcpu->arch.ret = RESUME_GUEST;
3296 vcpu->arch.trap = 0;
3297 kvmppc_update_vpas(vcpu);
3300 * Synchronize with other threads in this virtual core
3302 vc = vcpu->arch.vcore;
3303 spin_lock(&vc->lock);
3304 vcpu->arch.ceded = 0;
3305 vcpu->arch.run_task = current;
3306 vcpu->arch.kvm_run = kvm_run;
3307 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3308 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3309 vcpu->arch.busy_preempt = TB_NIL;
3310 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3314 * This happens the first time this is called for a vcpu.
3315 * If the vcore is already running, we may be able to start
3316 * this thread straight away and have it join in.
3318 if (!signal_pending(current)) {
3319 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3320 vc->vcore_state == VCORE_RUNNING) &&
3321 !VCORE_IS_EXITING(vc)) {
3322 kvmppc_create_dtl_entry(vcpu, vc);
3323 kvmppc_start_thread(vcpu, vc);
3324 trace_kvm_guest_enter(vcpu);
3325 } else if (vc->vcore_state == VCORE_SLEEPING) {
3326 swake_up_one(&vc->wq);
3331 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3332 !signal_pending(current)) {
3333 /* See if the MMU is ready to go */
3334 if (!vcpu->kvm->arch.mmu_ready) {
3335 spin_unlock(&vc->lock);
3336 r = kvmhv_setup_mmu(vcpu);
3337 spin_lock(&vc->lock);
3339 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3340 kvm_run->fail_entry.
3341 hardware_entry_failure_reason = 0;
3347 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3348 kvmppc_vcore_end_preempt(vc);
3350 if (vc->vcore_state != VCORE_INACTIVE) {
3351 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3354 for_each_runnable_thread(i, v, vc) {
3355 kvmppc_core_prepare_to_enter(v);
3356 if (signal_pending(v->arch.run_task)) {
3357 kvmppc_remove_runnable(vc, v);
3358 v->stat.signal_exits++;
3359 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3360 v->arch.ret = -EINTR;
3361 wake_up(&v->arch.cpu_run);
3364 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3367 for_each_runnable_thread(i, v, vc) {
3368 if (!kvmppc_vcpu_woken(v))
3369 n_ceded += v->arch.ceded;
3374 if (n_ceded == vc->n_runnable) {
3375 kvmppc_vcore_blocked(vc);
3376 } else if (need_resched()) {
3377 kvmppc_vcore_preempt(vc);
3378 /* Let something else run */
3379 cond_resched_lock(&vc->lock);
3380 if (vc->vcore_state == VCORE_PREEMPT)
3381 kvmppc_vcore_end_preempt(vc);
3383 kvmppc_run_core(vc);
3388 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3389 (vc->vcore_state == VCORE_RUNNING ||
3390 vc->vcore_state == VCORE_EXITING ||
3391 vc->vcore_state == VCORE_PIGGYBACK))
3392 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3394 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3395 kvmppc_vcore_end_preempt(vc);
3397 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3398 kvmppc_remove_runnable(vc, vcpu);
3399 vcpu->stat.signal_exits++;
3400 kvm_run->exit_reason = KVM_EXIT_INTR;
3401 vcpu->arch.ret = -EINTR;
3404 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3405 /* Wake up some vcpu to run the core */
3407 v = next_runnable_thread(vc, &i);
3408 wake_up(&v->arch.cpu_run);
3411 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3412 spin_unlock(&vc->lock);
3413 return vcpu->arch.ret;
3416 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3420 unsigned long ebb_regs[3] = {}; /* shut up GCC */
3421 unsigned long user_tar = 0;
3422 unsigned int user_vrsave;
3425 if (!vcpu->arch.sane) {
3426 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3431 * Don't allow entry with a suspended transaction, because
3432 * the guest entry/exit code will lose it.
3433 * If the guest has TM enabled, save away their TM-related SPRs
3434 * (they will get restored by the TM unavailable interrupt).
3436 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3437 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3438 (current->thread.regs->msr & MSR_TM)) {
3439 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3440 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3441 run->fail_entry.hardware_entry_failure_reason = 0;
3444 /* Enable TM so we can read the TM SPRs */
3445 mtmsr(mfmsr() | MSR_TM);
3446 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3447 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3448 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3449 current->thread.regs->msr &= ~MSR_TM;
3454 * Force online to 1 for the sake of old userspace which doesn't
3457 if (!vcpu->arch.online) {
3458 atomic_inc(&vcpu->arch.vcore->online_count);
3459 vcpu->arch.online = 1;
3462 kvmppc_core_prepare_to_enter(vcpu);
3464 /* No need to go into the guest when all we'll do is come back out */
3465 if (signal_pending(current)) {
3466 run->exit_reason = KVM_EXIT_INTR;
3471 atomic_inc(&kvm->arch.vcpus_running);
3472 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3475 flush_all_to_thread(current);
3477 /* Save userspace EBB and other register values */
3478 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3479 ebb_regs[0] = mfspr(SPRN_EBBHR);
3480 ebb_regs[1] = mfspr(SPRN_EBBRR);
3481 ebb_regs[2] = mfspr(SPRN_BESCR);
3482 user_tar = mfspr(SPRN_TAR);
3484 user_vrsave = mfspr(SPRN_VRSAVE);
3486 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3487 vcpu->arch.pgdir = current->mm->pgd;
3488 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3491 r = kvmppc_run_vcpu(run, vcpu);
3493 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3494 !(vcpu->arch.shregs.msr & MSR_PR)) {
3495 trace_kvm_hcall_enter(vcpu);
3496 r = kvmppc_pseries_do_hcall(vcpu);
3497 trace_kvm_hcall_exit(vcpu, r);
3498 kvmppc_core_prepare_to_enter(vcpu);
3499 } else if (r == RESUME_PAGE_FAULT) {
3500 srcu_idx = srcu_read_lock(&kvm->srcu);
3501 r = kvmppc_book3s_hv_page_fault(run, vcpu,
3502 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3503 srcu_read_unlock(&kvm->srcu, srcu_idx);
3504 } else if (r == RESUME_PASSTHROUGH) {
3505 if (WARN_ON(xive_enabled()))
3508 r = kvmppc_xics_rm_complete(vcpu, 0);
3510 } while (is_kvmppc_resume_guest(r));
3512 /* Restore userspace EBB and other register values */
3513 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3514 mtspr(SPRN_EBBHR, ebb_regs[0]);
3515 mtspr(SPRN_EBBRR, ebb_regs[1]);
3516 mtspr(SPRN_BESCR, ebb_regs[2]);
3517 mtspr(SPRN_TAR, user_tar);
3518 mtspr(SPRN_FSCR, current->thread.fscr);
3520 mtspr(SPRN_VRSAVE, user_vrsave);
3522 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3523 atomic_dec(&kvm->arch.vcpus_running);
3527 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3528 int shift, int sllp)
3530 (*sps)->page_shift = shift;
3531 (*sps)->slb_enc = sllp;
3532 (*sps)->enc[0].page_shift = shift;
3533 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3535 * Add 16MB MPSS support (may get filtered out by userspace)
3538 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3540 (*sps)->enc[1].page_shift = 24;
3541 (*sps)->enc[1].pte_enc = penc;
3547 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3548 struct kvm_ppc_smmu_info *info)
3550 struct kvm_ppc_one_seg_page_size *sps;
3553 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3554 * POWER7 doesn't support keys for instruction accesses,
3555 * POWER8 and POWER9 do.
3557 info->data_keys = 32;
3558 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3560 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3561 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3562 info->slb_size = 32;
3564 /* We only support these sizes for now, and no muti-size segments */
3565 sps = &info->sps[0];
3566 kvmppc_add_seg_page_size(&sps, 12, 0);
3567 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3568 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3574 * Get (and clear) the dirty memory log for a memory slot.
3576 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3577 struct kvm_dirty_log *log)
3579 struct kvm_memslots *slots;
3580 struct kvm_memory_slot *memslot;
3583 unsigned long *buf, *p;
3584 struct kvm_vcpu *vcpu;
3586 mutex_lock(&kvm->slots_lock);
3589 if (log->slot >= KVM_USER_MEM_SLOTS)
3592 slots = kvm_memslots(kvm);
3593 memslot = id_to_memslot(slots, log->slot);
3595 if (!memslot->dirty_bitmap)
3599 * Use second half of bitmap area because both HPT and radix
3600 * accumulate bits in the first half.
3602 n = kvm_dirty_bitmap_bytes(memslot);
3603 buf = memslot->dirty_bitmap + n / sizeof(long);
3606 if (kvm_is_radix(kvm))
3607 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3609 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3614 * We accumulate dirty bits in the first half of the
3615 * memslot's dirty_bitmap area, for when pages are paged
3616 * out or modified by the host directly. Pick up these
3617 * bits and add them to the map.
3619 p = memslot->dirty_bitmap;
3620 for (i = 0; i < n / sizeof(long); ++i)
3621 buf[i] |= xchg(&p[i], 0);
3623 /* Harvest dirty bits from VPA and DTL updates */
3624 /* Note: we never modify the SLB shadow buffer areas */
3625 kvm_for_each_vcpu(i, vcpu, kvm) {
3626 spin_lock(&vcpu->arch.vpa_update_lock);
3627 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3628 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3629 spin_unlock(&vcpu->arch.vpa_update_lock);
3633 if (copy_to_user(log->dirty_bitmap, buf, n))
3638 mutex_unlock(&kvm->slots_lock);
3642 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3643 struct kvm_memory_slot *dont)
3645 if (!dont || free->arch.rmap != dont->arch.rmap) {
3646 vfree(free->arch.rmap);
3647 free->arch.rmap = NULL;
3651 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3652 unsigned long npages)
3654 slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
3655 if (!slot->arch.rmap)
3661 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3662 struct kvm_memory_slot *memslot,
3663 const struct kvm_userspace_memory_region *mem)
3668 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3669 const struct kvm_userspace_memory_region *mem,
3670 const struct kvm_memory_slot *old,
3671 const struct kvm_memory_slot *new)
3673 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3676 * If we are making a new memslot, it might make
3677 * some address that was previously cached as emulated
3678 * MMIO be no longer emulated MMIO, so invalidate
3679 * all the caches of emulated MMIO translations.
3682 atomic64_inc(&kvm->arch.mmio_update);
3686 * Update LPCR values in kvm->arch and in vcores.
3687 * Caller must hold kvm->lock.
3689 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3694 if ((kvm->arch.lpcr & mask) == lpcr)
3697 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3699 for (i = 0; i < KVM_MAX_VCORES; ++i) {
3700 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3703 spin_lock(&vc->lock);
3704 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3705 spin_unlock(&vc->lock);
3706 if (++cores_done >= kvm->arch.online_vcores)
3711 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3716 void kvmppc_setup_partition_table(struct kvm *kvm)
3718 unsigned long dw0, dw1;
3720 if (!kvm_is_radix(kvm)) {
3721 /* PS field - page size for VRMA */
3722 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3723 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3724 /* HTABSIZE and HTABORG fields */
3725 dw0 |= kvm->arch.sdr1;
3727 /* Second dword as set by userspace */
3728 dw1 = kvm->arch.process_table;
3730 dw0 = PATB_HR | radix__get_tree_size() |
3731 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3732 dw1 = PATB_GR | kvm->arch.process_table;
3735 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3739 * Set up HPT (hashed page table) and RMA (real-mode area).
3740 * Must be called with kvm->lock held.
3742 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3745 struct kvm *kvm = vcpu->kvm;
3747 struct kvm_memory_slot *memslot;
3748 struct vm_area_struct *vma;
3749 unsigned long lpcr = 0, senc;
3750 unsigned long psize, porder;
3753 /* Allocate hashed page table (if not done already) and reset it */
3754 if (!kvm->arch.hpt.virt) {
3755 int order = KVM_DEFAULT_HPT_ORDER;
3756 struct kvm_hpt_info info;
3758 err = kvmppc_allocate_hpt(&info, order);
3759 /* If we get here, it means userspace didn't specify a
3760 * size explicitly. So, try successively smaller
3761 * sizes if the default failed. */
3762 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3763 err = kvmppc_allocate_hpt(&info, order);
3766 pr_err("KVM: Couldn't alloc HPT\n");
3770 kvmppc_set_hpt(kvm, &info);
3773 /* Look up the memslot for guest physical address 0 */
3774 srcu_idx = srcu_read_lock(&kvm->srcu);
3775 memslot = gfn_to_memslot(kvm, 0);
3777 /* We must have some memory at 0 by now */
3779 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3782 /* Look up the VMA for the start of this memory slot */
3783 hva = memslot->userspace_addr;
3784 down_read(¤t->mm->mmap_sem);
3785 vma = find_vma(current->mm, hva);
3786 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3789 psize = vma_kernel_pagesize(vma);
3791 up_read(¤t->mm->mmap_sem);
3793 /* We can handle 4k, 64k or 16M pages in the VRMA */
3794 if (psize >= 0x1000000)
3796 else if (psize >= 0x10000)
3800 porder = __ilog2(psize);
3802 senc = slb_pgsize_encoding(psize);
3803 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3804 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3805 /* Create HPTEs in the hash page table for the VRMA */
3806 kvmppc_map_vrma(vcpu, memslot, porder);
3808 /* Update VRMASD field in the LPCR */
3809 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3810 /* the -4 is to account for senc values starting at 0x10 */
3811 lpcr = senc << (LPCR_VRMASD_SH - 4);
3812 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3815 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
3819 srcu_read_unlock(&kvm->srcu, srcu_idx);
3824 up_read(¤t->mm->mmap_sem);
3828 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3829 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
3831 kvmppc_rmap_reset(kvm);
3832 kvm->arch.process_table = 0;
3833 /* Mutual exclusion with kvm_unmap_hva_range etc. */
3834 spin_lock(&kvm->mmu_lock);
3835 kvm->arch.radix = 0;
3836 spin_unlock(&kvm->mmu_lock);
3837 kvmppc_free_radix(kvm);
3838 kvmppc_update_lpcr(kvm, LPCR_VPM1,
3839 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3843 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3844 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
3848 err = kvmppc_init_vm_radix(kvm);
3852 kvmppc_rmap_reset(kvm);
3853 /* Mutual exclusion with kvm_unmap_hva_range etc. */
3854 spin_lock(&kvm->mmu_lock);
3855 kvm->arch.radix = 1;
3856 spin_unlock(&kvm->mmu_lock);
3857 kvmppc_free_hpt(&kvm->arch.hpt);
3858 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
3859 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3863 #ifdef CONFIG_KVM_XICS
3865 * Allocate a per-core structure for managing state about which cores are
3866 * running in the host versus the guest and for exchanging data between
3867 * real mode KVM and CPU running in the host.
3868 * This is only done for the first VM.
3869 * The allocated structure stays even if all VMs have stopped.
3870 * It is only freed when the kvm-hv module is unloaded.
3871 * It's OK for this routine to fail, we just don't support host
3872 * core operations like redirecting H_IPI wakeups.
3874 void kvmppc_alloc_host_rm_ops(void)
3876 struct kvmppc_host_rm_ops *ops;
3877 unsigned long l_ops;
3881 /* Not the first time here ? */
3882 if (kvmppc_host_rm_ops_hv != NULL)
3885 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3889 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3890 ops->rm_core = kzalloc(size, GFP_KERNEL);
3892 if (!ops->rm_core) {
3899 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3900 if (!cpu_online(cpu))
3903 core = cpu >> threads_shift;
3904 ops->rm_core[core].rm_state.in_host = 1;
3907 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3910 * Make the contents of the kvmppc_host_rm_ops structure visible
3911 * to other CPUs before we assign it to the global variable.
3912 * Do an atomic assignment (no locks used here), but if someone
3913 * beats us to it, just free our copy and return.
3916 l_ops = (unsigned long) ops;
3918 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3920 kfree(ops->rm_core);
3925 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3926 "ppc/kvm_book3s:prepare",
3927 kvmppc_set_host_core,
3928 kvmppc_clear_host_core);
3932 void kvmppc_free_host_rm_ops(void)
3934 if (kvmppc_host_rm_ops_hv) {
3935 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3936 kfree(kvmppc_host_rm_ops_hv->rm_core);
3937 kfree(kvmppc_host_rm_ops_hv);
3938 kvmppc_host_rm_ops_hv = NULL;
3943 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3945 unsigned long lpcr, lpid;
3949 /* Allocate the guest's logical partition ID */
3951 lpid = kvmppc_alloc_lpid();
3954 kvm->arch.lpid = lpid;
3956 kvmppc_alloc_host_rm_ops();
3959 * Since we don't flush the TLB when tearing down a VM,
3960 * and this lpid might have previously been used,
3961 * make sure we flush on each core before running the new VM.
3962 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3963 * does this flush for us.
3965 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3966 cpumask_setall(&kvm->arch.need_tlb_flush);
3968 /* Start out with the default set of hcalls enabled */
3969 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3970 sizeof(kvm->arch.enabled_hcalls));
3972 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3973 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3975 /* Init LPCR for virtual RMA mode */
3976 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3977 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3978 lpcr &= LPCR_PECE | LPCR_LPES;
3979 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3980 LPCR_VPM0 | LPCR_VPM1;
3981 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3982 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3983 /* On POWER8 turn on online bit to enable PURR/SPURR */
3984 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3987 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3988 * Set HVICE bit to enable hypervisor virtualization interrupts.
3989 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3990 * be unnecessary but better safe than sorry in case we re-enable
3991 * EE in HV mode with this LPCR still set)
3993 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3995 lpcr |= LPCR_HVICE | LPCR_HEIC;
3998 * If xive is enabled, we route 0x500 interrupts directly
4006 * If the host uses radix, the guest starts out as radix.
4008 if (radix_enabled()) {
4009 kvm->arch.radix = 1;
4010 kvm->arch.mmu_ready = 1;
4012 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4013 ret = kvmppc_init_vm_radix(kvm);
4015 kvmppc_free_lpid(kvm->arch.lpid);
4018 kvmppc_setup_partition_table(kvm);
4021 kvm->arch.lpcr = lpcr;
4023 /* Initialization for future HPT resizes */
4024 kvm->arch.resize_hpt = NULL;
4027 * Work out how many sets the TLB has, for the use of
4028 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4030 if (radix_enabled())
4031 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
4032 else if (cpu_has_feature(CPU_FTR_ARCH_300))
4033 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
4034 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4035 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
4037 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
4040 * Track that we now have a HV mode VM active. This blocks secondary
4041 * CPU threads from coming online.
4042 * On POWER9, we only need to do this if the "indep_threads_mode"
4043 * module parameter has been set to N.
4045 if (cpu_has_feature(CPU_FTR_ARCH_300))
4046 kvm->arch.threads_indep = indep_threads_mode;
4047 if (!kvm->arch.threads_indep)
4048 kvm_hv_vm_activated();
4051 * Initialize smt_mode depending on processor.
4052 * POWER8 and earlier have to use "strict" threading, where
4053 * all vCPUs in a vcore have to run on the same (sub)core,
4054 * whereas on POWER9 the threads can each run a different
4057 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4058 kvm->arch.smt_mode = threads_per_subcore;
4060 kvm->arch.smt_mode = 1;
4061 kvm->arch.emul_smt_mode = 1;
4064 * Create a debugfs directory for the VM
4066 snprintf(buf, sizeof(buf), "vm%d", current->pid);
4067 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4068 kvmppc_mmu_debugfs_init(kvm);
4073 static void kvmppc_free_vcores(struct kvm *kvm)
4077 for (i = 0; i < KVM_MAX_VCORES; ++i)
4078 kfree(kvm->arch.vcores[i]);
4079 kvm->arch.online_vcores = 0;
4082 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4084 debugfs_remove_recursive(kvm->arch.debugfs_dir);
4086 if (!kvm->arch.threads_indep)
4087 kvm_hv_vm_deactivated();
4089 kvmppc_free_vcores(kvm);
4091 kvmppc_free_lpid(kvm->arch.lpid);
4093 if (kvm_is_radix(kvm))
4094 kvmppc_free_radix(kvm);
4096 kvmppc_free_hpt(&kvm->arch.hpt);
4098 kvmppc_free_pimap(kvm);
4101 /* We don't need to emulate any privileged instructions or dcbz */
4102 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4103 unsigned int inst, int *advance)
4105 return EMULATE_FAIL;
4108 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4111 return EMULATE_FAIL;
4114 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4117 return EMULATE_FAIL;
4120 static int kvmppc_core_check_processor_compat_hv(void)
4122 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
4123 !cpu_has_feature(CPU_FTR_ARCH_206))
4129 #ifdef CONFIG_KVM_XICS
4131 void kvmppc_free_pimap(struct kvm *kvm)
4133 kfree(kvm->arch.pimap);
4136 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4138 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4141 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4143 struct irq_desc *desc;
4144 struct kvmppc_irq_map *irq_map;
4145 struct kvmppc_passthru_irqmap *pimap;
4146 struct irq_chip *chip;
4149 if (!kvm_irq_bypass)
4152 desc = irq_to_desc(host_irq);
4156 mutex_lock(&kvm->lock);
4158 pimap = kvm->arch.pimap;
4159 if (pimap == NULL) {
4160 /* First call, allocate structure to hold IRQ map */
4161 pimap = kvmppc_alloc_pimap();
4162 if (pimap == NULL) {
4163 mutex_unlock(&kvm->lock);
4166 kvm->arch.pimap = pimap;
4170 * For now, we only support interrupts for which the EOI operation
4171 * is an OPAL call followed by a write to XIRR, since that's
4172 * what our real-mode EOI code does, or a XIVE interrupt
4174 chip = irq_data_get_irq_chip(&desc->irq_data);
4175 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4176 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4177 host_irq, guest_gsi);
4178 mutex_unlock(&kvm->lock);
4183 * See if we already have an entry for this guest IRQ number.
4184 * If it's mapped to a hardware IRQ number, that's an error,
4185 * otherwise re-use this entry.
4187 for (i = 0; i < pimap->n_mapped; i++) {
4188 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4189 if (pimap->mapped[i].r_hwirq) {
4190 mutex_unlock(&kvm->lock);
4197 if (i == KVMPPC_PIRQ_MAPPED) {
4198 mutex_unlock(&kvm->lock);
4199 return -EAGAIN; /* table is full */
4202 irq_map = &pimap->mapped[i];
4204 irq_map->v_hwirq = guest_gsi;
4205 irq_map->desc = desc;
4208 * Order the above two stores before the next to serialize with
4209 * the KVM real mode handler.
4212 irq_map->r_hwirq = desc->irq_data.hwirq;
4214 if (i == pimap->n_mapped)
4218 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4220 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4222 irq_map->r_hwirq = 0;
4224 mutex_unlock(&kvm->lock);
4229 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4231 struct irq_desc *desc;
4232 struct kvmppc_passthru_irqmap *pimap;
4235 if (!kvm_irq_bypass)
4238 desc = irq_to_desc(host_irq);
4242 mutex_lock(&kvm->lock);
4243 if (!kvm->arch.pimap)
4246 pimap = kvm->arch.pimap;
4248 for (i = 0; i < pimap->n_mapped; i++) {
4249 if (guest_gsi == pimap->mapped[i].v_hwirq)
4253 if (i == pimap->n_mapped) {
4254 mutex_unlock(&kvm->lock);
4259 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4261 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4263 /* invalidate the entry (what do do on error from the above ?) */
4264 pimap->mapped[i].r_hwirq = 0;
4267 * We don't free this structure even when the count goes to
4268 * zero. The structure is freed when we destroy the VM.
4271 mutex_unlock(&kvm->lock);
4275 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4276 struct irq_bypass_producer *prod)
4279 struct kvm_kernel_irqfd *irqfd =
4280 container_of(cons, struct kvm_kernel_irqfd, consumer);
4282 irqfd->producer = prod;
4284 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4286 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4287 prod->irq, irqfd->gsi, ret);
4292 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4293 struct irq_bypass_producer *prod)
4296 struct kvm_kernel_irqfd *irqfd =
4297 container_of(cons, struct kvm_kernel_irqfd, consumer);
4299 irqfd->producer = NULL;
4302 * When producer of consumer is unregistered, we change back to
4303 * default external interrupt handling mode - KVM real mode
4304 * will switch back to host.
4306 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4308 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4309 prod->irq, irqfd->gsi, ret);
4313 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4314 unsigned int ioctl, unsigned long arg)
4316 struct kvm *kvm __maybe_unused = filp->private_data;
4317 void __user *argp = (void __user *)arg;
4322 case KVM_PPC_ALLOCATE_HTAB: {
4326 if (get_user(htab_order, (u32 __user *)argp))
4328 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4335 case KVM_PPC_GET_HTAB_FD: {
4336 struct kvm_get_htab_fd ghf;
4339 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4341 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4345 case KVM_PPC_RESIZE_HPT_PREPARE: {
4346 struct kvm_ppc_resize_hpt rhpt;
4349 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4352 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4356 case KVM_PPC_RESIZE_HPT_COMMIT: {
4357 struct kvm_ppc_resize_hpt rhpt;
4360 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4363 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4375 * List of hcall numbers to enable by default.
4376 * For compatibility with old userspace, we enable by default
4377 * all hcalls that were implemented before the hcall-enabling
4378 * facility was added. Note this list should not include H_RTAS.
4380 static unsigned int default_hcall_list[] = {
4394 #ifdef CONFIG_KVM_XICS
4405 static void init_default_hcalls(void)
4410 for (i = 0; default_hcall_list[i]; ++i) {
4411 hcall = default_hcall_list[i];
4412 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4413 __set_bit(hcall / 4, default_enabled_hcalls);
4417 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4423 /* If not on a POWER9, reject it */
4424 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4427 /* If any unknown flags set, reject it */
4428 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4431 /* GR (guest radix) bit in process_table field must match */
4432 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4433 if (!!(cfg->process_table & PATB_GR) != radix)
4436 /* Process table size field must be reasonable, i.e. <= 24 */
4437 if ((cfg->process_table & PRTS_MASK) > 24)
4440 /* We can change a guest to/from radix now, if the host is radix */
4441 if (radix && !radix_enabled())
4444 mutex_lock(&kvm->lock);
4445 if (radix != kvm_is_radix(kvm)) {
4446 if (kvm->arch.mmu_ready) {
4447 kvm->arch.mmu_ready = 0;
4448 /* order mmu_ready vs. vcpus_running */
4450 if (atomic_read(&kvm->arch.vcpus_running)) {
4451 kvm->arch.mmu_ready = 1;
4457 err = kvmppc_switch_mmu_to_radix(kvm);
4459 err = kvmppc_switch_mmu_to_hpt(kvm);
4464 kvm->arch.process_table = cfg->process_table;
4465 kvmppc_setup_partition_table(kvm);
4467 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4468 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4472 mutex_unlock(&kvm->lock);
4476 static struct kvmppc_ops kvm_ops_hv = {
4477 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4478 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4479 .get_one_reg = kvmppc_get_one_reg_hv,
4480 .set_one_reg = kvmppc_set_one_reg_hv,
4481 .vcpu_load = kvmppc_core_vcpu_load_hv,
4482 .vcpu_put = kvmppc_core_vcpu_put_hv,
4483 .set_msr = kvmppc_set_msr_hv,
4484 .vcpu_run = kvmppc_vcpu_run_hv,
4485 .vcpu_create = kvmppc_core_vcpu_create_hv,
4486 .vcpu_free = kvmppc_core_vcpu_free_hv,
4487 .check_requests = kvmppc_core_check_requests_hv,
4488 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
4489 .flush_memslot = kvmppc_core_flush_memslot_hv,
4490 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4491 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
4492 .unmap_hva_range = kvm_unmap_hva_range_hv,
4493 .age_hva = kvm_age_hva_hv,
4494 .test_age_hva = kvm_test_age_hva_hv,
4495 .set_spte_hva = kvm_set_spte_hva_hv,
4496 .mmu_destroy = kvmppc_mmu_destroy_hv,
4497 .free_memslot = kvmppc_core_free_memslot_hv,
4498 .create_memslot = kvmppc_core_create_memslot_hv,
4499 .init_vm = kvmppc_core_init_vm_hv,
4500 .destroy_vm = kvmppc_core_destroy_vm_hv,
4501 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4502 .emulate_op = kvmppc_core_emulate_op_hv,
4503 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4504 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4505 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4506 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
4507 .hcall_implemented = kvmppc_hcall_impl_hv,
4508 #ifdef CONFIG_KVM_XICS
4509 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4510 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4512 .configure_mmu = kvmhv_configure_mmu,
4513 .get_rmmu_info = kvmhv_get_rmmu_info,
4514 .set_smt_mode = kvmhv_set_smt_mode,
4517 static int kvm_init_subcore_bitmap(void)
4520 int nr_cores = cpu_nr_cores();
4521 struct sibling_subcore_state *sibling_subcore_state;
4523 for (i = 0; i < nr_cores; i++) {
4524 int first_cpu = i * threads_per_core;
4525 int node = cpu_to_node(first_cpu);
4527 /* Ignore if it is already allocated. */
4528 if (paca_ptrs[first_cpu]->sibling_subcore_state)
4531 sibling_subcore_state =
4532 kmalloc_node(sizeof(struct sibling_subcore_state),
4534 if (!sibling_subcore_state)
4537 memset(sibling_subcore_state, 0,
4538 sizeof(struct sibling_subcore_state));
4540 for (j = 0; j < threads_per_core; j++) {
4541 int cpu = first_cpu + j;
4543 paca_ptrs[cpu]->sibling_subcore_state =
4544 sibling_subcore_state;
4550 static int kvmppc_radix_possible(void)
4552 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4555 static int kvmppc_book3s_init_hv(void)
4559 * FIXME!! Do we need to check on all cpus ?
4561 r = kvmppc_core_check_processor_compat_hv();
4565 r = kvm_init_subcore_bitmap();
4570 * We need a way of accessing the XICS interrupt controller,
4571 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
4572 * indirectly, via OPAL.
4575 if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4576 struct device_node *np;
4578 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4580 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4583 /* presence of intc confirmed - node can be dropped again */
4588 kvm_ops_hv.owner = THIS_MODULE;
4589 kvmppc_hv_ops = &kvm_ops_hv;
4591 init_default_hcalls();
4595 r = kvmppc_mmu_hv_init();
4599 if (kvmppc_radix_possible())
4600 r = kvmppc_radix_init();
4603 * POWER9 chips before version 2.02 can't have some threads in
4604 * HPT mode and some in radix mode on the same core.
4606 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4607 unsigned int pvr = mfspr(SPRN_PVR);
4608 if ((pvr >> 16) == PVR_POWER9 &&
4609 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
4610 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
4611 no_mixing_hpt_and_radix = true;
4617 static void kvmppc_book3s_exit_hv(void)
4619 kvmppc_free_host_rm_ops();
4620 if (kvmppc_radix_possible())
4621 kvmppc_radix_exit();
4622 kvmppc_hv_ops = NULL;
4625 module_init(kvmppc_book3s_init_hv);
4626 module_exit(kvmppc_book3s_exit_hv);
4627 MODULE_LICENSE("GPL");
4628 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4629 MODULE_ALIAS("devname:kvm");