1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
7 * Paul Mackerras <paulus@au1.ibm.com>
8 * Alexander Graf <agraf@suse.de>
9 * Kevin Wolf <mail@kevin-wolf.de>
11 * Description: KVM functions specific to running on Book 3S
12 * processors in hypervisor mode (specifically POWER7 and later).
14 * This file is derived from arch/powerpc/kvm/book3s.c,
15 * by Alexander Graf <agraf@suse.de>.
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 #include <linux/err.h>
21 #include <linux/slab.h>
22 #include <linux/preempt.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/stat.h>
25 #include <linux/delay.h>
26 #include <linux/export.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpu.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 #include <linux/gfp.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/hugetlb.h>
40 #include <linux/kvm_irqfd.h>
41 #include <linux/irqbypass.h>
42 #include <linux/module.h>
43 #include <linux/compiler.h>
46 #include <asm/ftrace.h>
48 #include <asm/ppc-opcode.h>
49 #include <asm/asm-prototypes.h>
50 #include <asm/archrandom.h>
51 #include <asm/debug.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <linux/uaccess.h>
57 #include <asm/kvm_ppc.h>
58 #include <asm/kvm_book3s.h>
59 #include <asm/mmu_context.h>
60 #include <asm/lppaca.h>
61 #include <asm/processor.h>
62 #include <asm/cputhreads.h>
64 #include <asm/hvcall.h>
65 #include <asm/switch_to.h>
67 #include <asm/dbell.h>
69 #include <asm/pnv-pci.h>
74 #include <asm/hw_breakpoint.h>
78 #define CREATE_TRACE_POINTS
81 /* #define EXIT_DEBUG */
82 /* #define EXIT_DEBUG_SIMPLE */
83 /* #define EXIT_DEBUG_INT */
85 /* Used to indicate that a guest page fault needs to be handled */
86 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
87 /* Used to indicate that a guest passthrough interrupt needs to be handled */
88 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
90 /* Used as a "null" value for timebase values */
91 #define TB_NIL (~(u64)0)
93 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
95 static int dynamic_mt_modes = 6;
96 module_param(dynamic_mt_modes, int, 0644);
97 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
98 static int target_smt_mode;
99 module_param(target_smt_mode, int, 0644);
100 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
102 static bool indep_threads_mode = true;
103 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
106 static bool one_vm_per_core;
107 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
108 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
110 #ifdef CONFIG_KVM_XICS
111 static struct kernel_param_ops module_param_ops = {
112 .set = param_set_int,
113 .get = param_get_int,
116 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
117 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
119 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
120 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
123 /* If set, guests are allowed to create and control nested guests */
124 static bool nested = true;
125 module_param(nested, bool, S_IRUGO | S_IWUSR);
126 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
128 static inline bool nesting_enabled(struct kvm *kvm)
130 return kvm->arch.nested_enable && kvm_is_radix(kvm);
133 /* If set, the threads on each CPU core have to be in the same MMU mode */
134 static bool no_mixing_hpt_and_radix;
136 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
139 * RWMR values for POWER8. These control the rate at which PURR
140 * and SPURR count and should be set according to the number of
141 * online threads in the vcore being run.
143 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL
144 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL
145 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL
146 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL
147 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL
148 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL
149 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL
150 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL
152 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
164 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
168 struct kvm_vcpu *vcpu;
170 while (++i < MAX_SMT_THREADS) {
171 vcpu = READ_ONCE(vc->runnable_threads[i]);
180 /* Used to traverse the list of runnable threads for a given vcore */
181 #define for_each_runnable_thread(i, vcpu, vc) \
182 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
184 static bool kvmppc_ipi_thread(int cpu)
186 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
188 /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
189 if (kvmhv_on_pseries())
192 /* On POWER9 we can use msgsnd to IPI any cpu */
193 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
194 msg |= get_hard_smp_processor_id(cpu);
196 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
200 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
201 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
203 if (cpu_first_thread_sibling(cpu) ==
204 cpu_first_thread_sibling(smp_processor_id())) {
205 msg |= cpu_thread_in_core(cpu);
207 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
214 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
215 if (cpu >= 0 && cpu < nr_cpu_ids) {
216 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
220 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
228 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
231 struct swait_queue_head *wqp;
233 wqp = kvm_arch_vcpu_wq(vcpu);
234 if (swq_has_sleeper(wqp)) {
236 ++vcpu->stat.halt_wakeup;
239 cpu = READ_ONCE(vcpu->arch.thread_cpu);
240 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
243 /* CPU points to the first thread of the core */
245 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
246 smp_send_reschedule(cpu);
250 * We use the vcpu_load/put functions to measure stolen time.
251 * Stolen time is counted as time when either the vcpu is able to
252 * run as part of a virtual core, but the task running the vcore
253 * is preempted or sleeping, or when the vcpu needs something done
254 * in the kernel by the task running the vcpu, but that task is
255 * preempted or sleeping. Those two things have to be counted
256 * separately, since one of the vcpu tasks will take on the job
257 * of running the core, and the other vcpu tasks in the vcore will
258 * sleep waiting for it to do that, but that sleep shouldn't count
261 * Hence we accumulate stolen time when the vcpu can run as part of
262 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
263 * needs its task to do other things in the kernel (for example,
264 * service a page fault) in busy_stolen. We don't accumulate
265 * stolen time for a vcore when it is inactive, or for a vcpu
266 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
267 * a misnomer; it means that the vcpu task is not executing in
268 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
269 * the kernel. We don't have any way of dividing up that time
270 * between time that the vcpu is genuinely stopped, time that
271 * the task is actively working on behalf of the vcpu, and time
272 * that the task is preempted, so we don't count any of it as
275 * Updates to busy_stolen are protected by arch.tbacct_lock;
276 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
277 * lock. The stolen times are measured in units of timebase ticks.
278 * (Note that the != TB_NIL checks below are purely defensive;
279 * they should never fail.)
282 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
286 spin_lock_irqsave(&vc->stoltb_lock, flags);
287 vc->preempt_tb = mftb();
288 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
291 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
295 spin_lock_irqsave(&vc->stoltb_lock, flags);
296 if (vc->preempt_tb != TB_NIL) {
297 vc->stolen_tb += mftb() - vc->preempt_tb;
298 vc->preempt_tb = TB_NIL;
300 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
303 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
305 struct kvmppc_vcore *vc = vcpu->arch.vcore;
309 * We can test vc->runner without taking the vcore lock,
310 * because only this task ever sets vc->runner to this
311 * vcpu, and once it is set to this vcpu, only this task
312 * ever sets it to NULL.
314 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
315 kvmppc_core_end_stolen(vc);
317 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
318 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
319 vcpu->arch.busy_preempt != TB_NIL) {
320 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
321 vcpu->arch.busy_preempt = TB_NIL;
323 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
326 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
328 struct kvmppc_vcore *vc = vcpu->arch.vcore;
331 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
332 kvmppc_core_start_stolen(vc);
334 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
335 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
336 vcpu->arch.busy_preempt = mftb();
337 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
340 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
342 vcpu->arch.pvr = pvr;
345 /* Dummy value used in computing PCR value below */
346 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
348 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
350 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
351 struct kvmppc_vcore *vc = vcpu->arch.vcore;
353 /* We can (emulate) our own architecture version and anything older */
354 if (cpu_has_feature(CPU_FTR_ARCH_300))
355 host_pcr_bit = PCR_ARCH_300;
356 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
357 host_pcr_bit = PCR_ARCH_207;
358 else if (cpu_has_feature(CPU_FTR_ARCH_206))
359 host_pcr_bit = PCR_ARCH_206;
361 host_pcr_bit = PCR_ARCH_205;
363 /* Determine lowest PCR bit needed to run guest in given PVR level */
364 guest_pcr_bit = host_pcr_bit;
366 switch (arch_compat) {
368 guest_pcr_bit = PCR_ARCH_205;
372 guest_pcr_bit = PCR_ARCH_206;
375 guest_pcr_bit = PCR_ARCH_207;
378 guest_pcr_bit = PCR_ARCH_300;
385 /* Check requested PCR bits don't exceed our capabilities */
386 if (guest_pcr_bit > host_pcr_bit)
389 spin_lock(&vc->lock);
390 vc->arch_compat = arch_compat;
392 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
393 * Also set all reserved PCR bits
395 vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
396 spin_unlock(&vc->lock);
401 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
405 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
406 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
407 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
408 for (r = 0; r < 16; ++r)
409 pr_err("r%2d = %.16lx r%d = %.16lx\n",
410 r, kvmppc_get_gpr(vcpu, r),
411 r+16, kvmppc_get_gpr(vcpu, r+16));
412 pr_err("ctr = %.16lx lr = %.16lx\n",
413 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
414 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
415 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
416 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
417 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
418 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
419 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
420 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n",
421 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
422 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
423 pr_err("fault dar = %.16lx dsisr = %.8x\n",
424 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
425 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
426 for (r = 0; r < vcpu->arch.slb_max; ++r)
427 pr_err(" ESID = %.16llx VSID = %.16llx\n",
428 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
429 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
430 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
431 vcpu->arch.last_inst);
434 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
436 return kvm_get_vcpu_by_id(kvm, id);
439 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
441 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
442 vpa->yield_count = cpu_to_be32(1);
445 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
446 unsigned long addr, unsigned long len)
448 /* check address is cacheline aligned */
449 if (addr & (L1_CACHE_BYTES - 1))
451 spin_lock(&vcpu->arch.vpa_update_lock);
452 if (v->next_gpa != addr || v->len != len) {
454 v->len = addr ? len : 0;
455 v->update_pending = 1;
457 spin_unlock(&vcpu->arch.vpa_update_lock);
461 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
470 static int vpa_is_registered(struct kvmppc_vpa *vpap)
472 if (vpap->update_pending)
473 return vpap->next_gpa != 0;
474 return vpap->pinned_addr != NULL;
477 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
479 unsigned long vcpuid, unsigned long vpa)
481 struct kvm *kvm = vcpu->kvm;
482 unsigned long len, nb;
484 struct kvm_vcpu *tvcpu;
487 struct kvmppc_vpa *vpap;
489 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
493 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
494 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
495 subfunc == H_VPA_REG_SLB) {
496 /* Registering new area - address must be cache-line aligned */
497 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
500 /* convert logical addr to kernel addr and read length */
501 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
504 if (subfunc == H_VPA_REG_VPA)
505 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
507 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
508 kvmppc_unpin_guest_page(kvm, va, vpa, false);
511 if (len > nb || len < sizeof(struct reg_vpa))
520 spin_lock(&tvcpu->arch.vpa_update_lock);
523 case H_VPA_REG_VPA: /* register VPA */
525 * The size of our lppaca is 1kB because of the way we align
526 * it for the guest to avoid crossing a 4kB boundary. We only
527 * use 640 bytes of the structure though, so we should accept
528 * clients that set a size of 640.
530 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
531 if (len < sizeof(struct lppaca))
533 vpap = &tvcpu->arch.vpa;
537 case H_VPA_REG_DTL: /* register DTL */
538 if (len < sizeof(struct dtl_entry))
540 len -= len % sizeof(struct dtl_entry);
542 /* Check that they have previously registered a VPA */
544 if (!vpa_is_registered(&tvcpu->arch.vpa))
547 vpap = &tvcpu->arch.dtl;
551 case H_VPA_REG_SLB: /* register SLB shadow buffer */
552 /* Check that they have previously registered a VPA */
554 if (!vpa_is_registered(&tvcpu->arch.vpa))
557 vpap = &tvcpu->arch.slb_shadow;
561 case H_VPA_DEREG_VPA: /* deregister VPA */
562 /* Check they don't still have a DTL or SLB buf registered */
564 if (vpa_is_registered(&tvcpu->arch.dtl) ||
565 vpa_is_registered(&tvcpu->arch.slb_shadow))
568 vpap = &tvcpu->arch.vpa;
572 case H_VPA_DEREG_DTL: /* deregister DTL */
573 vpap = &tvcpu->arch.dtl;
577 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
578 vpap = &tvcpu->arch.slb_shadow;
584 vpap->next_gpa = vpa;
586 vpap->update_pending = 1;
589 spin_unlock(&tvcpu->arch.vpa_update_lock);
594 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
596 struct kvm *kvm = vcpu->kvm;
602 * We need to pin the page pointed to by vpap->next_gpa,
603 * but we can't call kvmppc_pin_guest_page under the lock
604 * as it does get_user_pages() and down_read(). So we
605 * have to drop the lock, pin the page, then get the lock
606 * again and check that a new area didn't get registered
610 gpa = vpap->next_gpa;
611 spin_unlock(&vcpu->arch.vpa_update_lock);
615 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
616 spin_lock(&vcpu->arch.vpa_update_lock);
617 if (gpa == vpap->next_gpa)
619 /* sigh... unpin that one and try again */
621 kvmppc_unpin_guest_page(kvm, va, gpa, false);
624 vpap->update_pending = 0;
625 if (va && nb < vpap->len) {
627 * If it's now too short, it must be that userspace
628 * has changed the mappings underlying guest memory,
629 * so unregister the region.
631 kvmppc_unpin_guest_page(kvm, va, gpa, false);
634 if (vpap->pinned_addr)
635 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
638 vpap->pinned_addr = va;
641 vpap->pinned_end = va + vpap->len;
644 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
646 if (!(vcpu->arch.vpa.update_pending ||
647 vcpu->arch.slb_shadow.update_pending ||
648 vcpu->arch.dtl.update_pending))
651 spin_lock(&vcpu->arch.vpa_update_lock);
652 if (vcpu->arch.vpa.update_pending) {
653 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
654 if (vcpu->arch.vpa.pinned_addr)
655 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
657 if (vcpu->arch.dtl.update_pending) {
658 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
659 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
660 vcpu->arch.dtl_index = 0;
662 if (vcpu->arch.slb_shadow.update_pending)
663 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
664 spin_unlock(&vcpu->arch.vpa_update_lock);
668 * Return the accumulated stolen time for the vcore up until `now'.
669 * The caller should hold the vcore lock.
671 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
676 spin_lock_irqsave(&vc->stoltb_lock, flags);
678 if (vc->vcore_state != VCORE_INACTIVE &&
679 vc->preempt_tb != TB_NIL)
680 p += now - vc->preempt_tb;
681 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
685 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
686 struct kvmppc_vcore *vc)
688 struct dtl_entry *dt;
690 unsigned long stolen;
691 unsigned long core_stolen;
695 dt = vcpu->arch.dtl_ptr;
696 vpa = vcpu->arch.vpa.pinned_addr;
698 core_stolen = vcore_stolen_time(vc, now);
699 stolen = core_stolen - vcpu->arch.stolen_logged;
700 vcpu->arch.stolen_logged = core_stolen;
701 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
702 stolen += vcpu->arch.busy_stolen;
703 vcpu->arch.busy_stolen = 0;
704 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
707 memset(dt, 0, sizeof(struct dtl_entry));
708 dt->dispatch_reason = 7;
709 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
710 dt->timebase = cpu_to_be64(now + vc->tb_offset);
711 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
712 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
713 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
715 if (dt == vcpu->arch.dtl.pinned_end)
716 dt = vcpu->arch.dtl.pinned_addr;
717 vcpu->arch.dtl_ptr = dt;
718 /* order writing *dt vs. writing vpa->dtl_idx */
720 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
721 vcpu->arch.dtl.dirty = true;
724 /* See if there is a doorbell interrupt pending for a vcpu */
725 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
728 struct kvmppc_vcore *vc;
730 if (vcpu->arch.doorbell_request)
733 * Ensure that the read of vcore->dpdes comes after the read
734 * of vcpu->doorbell_request. This barrier matches the
735 * smp_wmb() in kvmppc_guest_entry_inject().
738 vc = vcpu->arch.vcore;
739 thr = vcpu->vcpu_id - vc->first_vcpuid;
740 return !!(vc->dpdes & (1 << thr));
743 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
745 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
747 if ((!vcpu->arch.vcore->arch_compat) &&
748 cpu_has_feature(CPU_FTR_ARCH_207S))
753 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
754 unsigned long resource, unsigned long value1,
755 unsigned long value2)
758 case H_SET_MODE_RESOURCE_SET_CIABR:
759 if (!kvmppc_power8_compatible(vcpu))
764 return H_UNSUPPORTED_FLAG_START;
765 /* Guests can't breakpoint the hypervisor */
766 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
768 vcpu->arch.ciabr = value1;
770 case H_SET_MODE_RESOURCE_SET_DAWR:
771 if (!kvmppc_power8_compatible(vcpu))
773 if (!ppc_breakpoint_available())
776 return H_UNSUPPORTED_FLAG_START;
777 if (value2 & DABRX_HYP)
779 vcpu->arch.dawr = value1;
780 vcpu->arch.dawrx = value2;
782 case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
783 /* KVM does not support mflags=2 (AIL=2) */
784 if (mflags != 0 && mflags != 3)
785 return H_UNSUPPORTED_FLAG_START;
792 /* Copy guest memory in place - must reside within a single memslot */
793 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
796 struct kvm_memory_slot *to_memslot = NULL;
797 struct kvm_memory_slot *from_memslot = NULL;
798 unsigned long to_addr, from_addr;
801 /* Get HPA for from address */
802 from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
805 if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
808 from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
809 if (kvm_is_error_hva(from_addr))
811 from_addr |= (from & (PAGE_SIZE - 1));
813 /* Get HPA for to address */
814 to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
817 if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
820 to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
821 if (kvm_is_error_hva(to_addr))
823 to_addr |= (to & (PAGE_SIZE - 1));
826 r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
830 mark_page_dirty(kvm, to >> PAGE_SHIFT);
834 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
835 unsigned long dest, unsigned long src)
837 u64 pg_sz = SZ_4K; /* 4K page size */
838 u64 pg_mask = SZ_4K - 1;
841 /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
842 if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
843 H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
846 /* dest (and src if copy_page flag set) must be page aligned */
847 if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
850 /* zero and/or copy the page as determined by the flags */
851 if (flags & H_COPY_PAGE) {
852 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
855 } else if (flags & H_ZERO_PAGE) {
856 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
861 /* We can ignore the remaining flags */
866 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
868 struct kvmppc_vcore *vcore = target->arch.vcore;
871 * We expect to have been called by the real mode handler
872 * (kvmppc_rm_h_confer()) which would have directly returned
873 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
874 * have useful work to do and should not confer) so we don't
878 spin_lock(&vcore->lock);
879 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
880 vcore->vcore_state != VCORE_INACTIVE &&
882 target = vcore->runner;
883 spin_unlock(&vcore->lock);
885 return kvm_vcpu_yield_to(target);
888 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
891 struct lppaca *lppaca;
893 spin_lock(&vcpu->arch.vpa_update_lock);
894 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
896 yield_count = be32_to_cpu(lppaca->yield_count);
897 spin_unlock(&vcpu->arch.vpa_update_lock);
901 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
903 unsigned long req = kvmppc_get_gpr(vcpu, 3);
904 unsigned long target, ret = H_SUCCESS;
906 struct kvm_vcpu *tvcpu;
909 if (req <= MAX_HCALL_OPCODE &&
910 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
917 target = kvmppc_get_gpr(vcpu, 4);
918 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
923 tvcpu->arch.prodded = 1;
925 if (tvcpu->arch.ceded)
926 kvmppc_fast_vcpu_kick_hv(tvcpu);
929 target = kvmppc_get_gpr(vcpu, 4);
932 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
937 yield_count = kvmppc_get_gpr(vcpu, 5);
938 if (kvmppc_get_yield_count(tvcpu) != yield_count)
940 kvm_arch_vcpu_yield_to(tvcpu);
943 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
944 kvmppc_get_gpr(vcpu, 5),
945 kvmppc_get_gpr(vcpu, 6));
948 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
951 idx = srcu_read_lock(&vcpu->kvm->srcu);
952 rc = kvmppc_rtas_hcall(vcpu);
953 srcu_read_unlock(&vcpu->kvm->srcu, idx);
960 /* Send the error out to userspace via KVM_RUN */
962 case H_LOGICAL_CI_LOAD:
963 ret = kvmppc_h_logical_ci_load(vcpu);
964 if (ret == H_TOO_HARD)
967 case H_LOGICAL_CI_STORE:
968 ret = kvmppc_h_logical_ci_store(vcpu);
969 if (ret == H_TOO_HARD)
973 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
974 kvmppc_get_gpr(vcpu, 5),
975 kvmppc_get_gpr(vcpu, 6),
976 kvmppc_get_gpr(vcpu, 7));
977 if (ret == H_TOO_HARD)
986 if (kvmppc_xics_enabled(vcpu)) {
987 if (xics_on_xive()) {
988 ret = H_NOT_AVAILABLE;
991 ret = kvmppc_xics_hcall(vcpu, req);
996 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
999 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1000 kvmppc_get_gpr(vcpu, 5));
1002 #ifdef CONFIG_SPAPR_TCE_IOMMU
1004 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1005 kvmppc_get_gpr(vcpu, 5));
1006 if (ret == H_TOO_HARD)
1010 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1011 kvmppc_get_gpr(vcpu, 5),
1012 kvmppc_get_gpr(vcpu, 6));
1013 if (ret == H_TOO_HARD)
1016 case H_PUT_TCE_INDIRECT:
1017 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1018 kvmppc_get_gpr(vcpu, 5),
1019 kvmppc_get_gpr(vcpu, 6),
1020 kvmppc_get_gpr(vcpu, 7));
1021 if (ret == H_TOO_HARD)
1025 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1026 kvmppc_get_gpr(vcpu, 5),
1027 kvmppc_get_gpr(vcpu, 6),
1028 kvmppc_get_gpr(vcpu, 7));
1029 if (ret == H_TOO_HARD)
1034 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1038 case H_SET_PARTITION_TABLE:
1040 if (nesting_enabled(vcpu->kvm))
1041 ret = kvmhv_set_partition_table(vcpu);
1043 case H_ENTER_NESTED:
1045 if (!nesting_enabled(vcpu->kvm))
1047 ret = kvmhv_enter_nested_guest(vcpu);
1048 if (ret == H_INTERRUPT) {
1049 kvmppc_set_gpr(vcpu, 3, 0);
1050 vcpu->arch.hcall_needed = 0;
1052 } else if (ret == H_TOO_HARD) {
1053 kvmppc_set_gpr(vcpu, 3, 0);
1054 vcpu->arch.hcall_needed = 0;
1058 case H_TLB_INVALIDATE:
1060 if (nesting_enabled(vcpu->kvm))
1061 ret = kvmhv_do_nested_tlbie(vcpu);
1063 case H_COPY_TOFROM_GUEST:
1065 if (nesting_enabled(vcpu->kvm))
1066 ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1069 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1070 kvmppc_get_gpr(vcpu, 5),
1071 kvmppc_get_gpr(vcpu, 6));
1076 kvmppc_set_gpr(vcpu, 3, ret);
1077 vcpu->arch.hcall_needed = 0;
1078 return RESUME_GUEST;
1082 * Handle H_CEDE in the nested virtualization case where we haven't
1083 * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1084 * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1085 * that the cede logic in kvmppc_run_single_vcpu() works properly.
1087 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1089 vcpu->arch.shregs.msr |= MSR_EE;
1090 vcpu->arch.ceded = 1;
1092 if (vcpu->arch.prodded) {
1093 vcpu->arch.prodded = 0;
1095 vcpu->arch.ceded = 0;
1099 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1105 case H_REGISTER_VPA:
1107 case H_LOGICAL_CI_LOAD:
1108 case H_LOGICAL_CI_STORE:
1109 #ifdef CONFIG_KVM_XICS
1121 /* See if it's in the real-mode table */
1122 return kvmppc_hcall_impl_hv_realmode(cmd);
1125 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1126 struct kvm_vcpu *vcpu)
1130 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1133 * Fetch failed, so return to guest and
1134 * try executing it again.
1136 return RESUME_GUEST;
1139 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1140 run->exit_reason = KVM_EXIT_DEBUG;
1141 run->debug.arch.address = kvmppc_get_pc(vcpu);
1144 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1145 return RESUME_GUEST;
1149 static void do_nothing(void *x)
1153 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1155 int thr, cpu, pcpu, nthreads;
1157 unsigned long dpdes;
1159 nthreads = vcpu->kvm->arch.emul_smt_mode;
1161 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1162 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1163 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1167 * If the vcpu is currently running on a physical cpu thread,
1168 * interrupt it in order to pull it out of the guest briefly,
1169 * which will update its vcore->dpdes value.
1171 pcpu = READ_ONCE(v->cpu);
1173 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1174 if (kvmppc_doorbell_pending(v))
1181 * On POWER9, emulate doorbell-related instructions in order to
1182 * give the guest the illusion of running on a multi-threaded core.
1183 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1186 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1190 struct kvm *kvm = vcpu->kvm;
1191 struct kvm_vcpu *tvcpu;
1193 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1194 return RESUME_GUEST;
1195 if (get_op(inst) != 31)
1196 return EMULATE_FAIL;
1198 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1199 switch (get_xop(inst)) {
1200 case OP_31_XOP_MSGSNDP:
1201 arg = kvmppc_get_gpr(vcpu, rb);
1202 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1205 if (arg >= kvm->arch.emul_smt_mode)
1207 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1210 if (!tvcpu->arch.doorbell_request) {
1211 tvcpu->arch.doorbell_request = 1;
1212 kvmppc_fast_vcpu_kick_hv(tvcpu);
1215 case OP_31_XOP_MSGCLRP:
1216 arg = kvmppc_get_gpr(vcpu, rb);
1217 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1219 vcpu->arch.vcore->dpdes = 0;
1220 vcpu->arch.doorbell_request = 0;
1222 case OP_31_XOP_MFSPR:
1223 switch (get_sprn(inst)) {
1228 arg = kvmppc_read_dpdes(vcpu);
1231 return EMULATE_FAIL;
1233 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1236 return EMULATE_FAIL;
1238 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1239 return RESUME_GUEST;
1242 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1243 struct task_struct *tsk)
1245 int r = RESUME_HOST;
1247 vcpu->stat.sum_exits++;
1250 * This can happen if an interrupt occurs in the last stages
1251 * of guest entry or the first stages of guest exit (i.e. after
1252 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1253 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1254 * That can happen due to a bug, or due to a machine check
1255 * occurring at just the wrong time.
1257 if (vcpu->arch.shregs.msr & MSR_HV) {
1258 printk(KERN_EMERG "KVM trap in HV mode!\n");
1259 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1260 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1261 vcpu->arch.shregs.msr);
1262 kvmppc_dump_regs(vcpu);
1263 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1264 run->hw.hardware_exit_reason = vcpu->arch.trap;
1267 run->exit_reason = KVM_EXIT_UNKNOWN;
1268 run->ready_for_interrupt_injection = 1;
1269 switch (vcpu->arch.trap) {
1270 /* We're good on these - the host merely wanted to get our attention */
1271 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1272 vcpu->stat.dec_exits++;
1275 case BOOK3S_INTERRUPT_EXTERNAL:
1276 case BOOK3S_INTERRUPT_H_DOORBELL:
1277 case BOOK3S_INTERRUPT_H_VIRT:
1278 vcpu->stat.ext_intr_exits++;
1281 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1282 case BOOK3S_INTERRUPT_HMI:
1283 case BOOK3S_INTERRUPT_PERFMON:
1284 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1287 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1288 /* Print the MCE event to host console. */
1289 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1292 * If the guest can do FWNMI, exit to userspace so it can
1293 * deliver a FWNMI to the guest.
1294 * Otherwise we synthesize a machine check for the guest
1295 * so that it knows that the machine check occurred.
1297 if (!vcpu->kvm->arch.fwnmi_enabled) {
1298 ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1299 kvmppc_core_queue_machine_check(vcpu, flags);
1304 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1305 run->exit_reason = KVM_EXIT_NMI;
1306 run->hw.hardware_exit_reason = vcpu->arch.trap;
1307 /* Clear out the old NMI status from run->flags */
1308 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1309 /* Now set the NMI status */
1310 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1311 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1313 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1317 case BOOK3S_INTERRUPT_PROGRAM:
1321 * Normally program interrupts are delivered directly
1322 * to the guest by the hardware, but we can get here
1323 * as a result of a hypervisor emulation interrupt
1324 * (e40) getting turned into a 700 by BML RTAS.
1326 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1327 kvmppc_core_queue_program(vcpu, flags);
1331 case BOOK3S_INTERRUPT_SYSCALL:
1333 /* hcall - punt to userspace */
1336 /* hypercall with MSR_PR has already been handled in rmode,
1337 * and never reaches here.
1340 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1341 for (i = 0; i < 9; ++i)
1342 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1343 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1344 vcpu->arch.hcall_needed = 1;
1349 * We get these next two if the guest accesses a page which it thinks
1350 * it has mapped but which is not actually present, either because
1351 * it is for an emulated I/O device or because the corresonding
1352 * host page has been paged out. Any other HDSI/HISI interrupts
1353 * have been handled already.
1355 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1356 r = RESUME_PAGE_FAULT;
1358 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1359 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1360 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1361 DSISR_SRR1_MATCH_64S;
1362 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1363 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1364 r = RESUME_PAGE_FAULT;
1367 * This occurs if the guest executes an illegal instruction.
1368 * If the guest debug is disabled, generate a program interrupt
1369 * to the guest. If guest debug is enabled, we need to check
1370 * whether the instruction is a software breakpoint instruction.
1371 * Accordingly return to Guest or Host.
1373 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1374 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1375 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1376 swab32(vcpu->arch.emul_inst) :
1377 vcpu->arch.emul_inst;
1378 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1379 r = kvmppc_emulate_debug_inst(run, vcpu);
1381 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1386 * This occurs if the guest (kernel or userspace), does something that
1387 * is prohibited by HFSCR.
1388 * On POWER9, this could be a doorbell instruction that we need
1390 * Otherwise, we just generate a program interrupt to the guest.
1392 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1394 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1395 cpu_has_feature(CPU_FTR_ARCH_300))
1396 r = kvmppc_emulate_doorbell_instr(vcpu);
1397 if (r == EMULATE_FAIL) {
1398 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1403 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1404 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1406 * This occurs for various TM-related instructions that
1407 * we need to emulate on POWER9 DD2.2. We have already
1408 * handled the cases where the guest was in real-suspend
1409 * mode and was transitioning to transactional state.
1411 r = kvmhv_p9_tm_emulation(vcpu);
1415 case BOOK3S_INTERRUPT_HV_RM_HARD:
1416 r = RESUME_PASSTHROUGH;
1419 kvmppc_dump_regs(vcpu);
1420 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1421 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1422 vcpu->arch.shregs.msr);
1423 run->hw.hardware_exit_reason = vcpu->arch.trap;
1431 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1436 vcpu->stat.sum_exits++;
1439 * This can happen if an interrupt occurs in the last stages
1440 * of guest entry or the first stages of guest exit (i.e. after
1441 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1442 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1443 * That can happen due to a bug, or due to a machine check
1444 * occurring at just the wrong time.
1446 if (vcpu->arch.shregs.msr & MSR_HV) {
1447 pr_emerg("KVM trap in HV mode while nested!\n");
1448 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1449 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1450 vcpu->arch.shregs.msr);
1451 kvmppc_dump_regs(vcpu);
1454 switch (vcpu->arch.trap) {
1455 /* We're good on these - the host merely wanted to get our attention */
1456 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1457 vcpu->stat.dec_exits++;
1460 case BOOK3S_INTERRUPT_EXTERNAL:
1461 vcpu->stat.ext_intr_exits++;
1464 case BOOK3S_INTERRUPT_H_DOORBELL:
1465 case BOOK3S_INTERRUPT_H_VIRT:
1466 vcpu->stat.ext_intr_exits++;
1469 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1470 case BOOK3S_INTERRUPT_HMI:
1471 case BOOK3S_INTERRUPT_PERFMON:
1472 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1475 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1476 /* Pass the machine check to the L1 guest */
1478 /* Print the MCE event to host console. */
1479 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1482 * We get these next two if the guest accesses a page which it thinks
1483 * it has mapped but which is not actually present, either because
1484 * it is for an emulated I/O device or because the corresonding
1485 * host page has been paged out.
1487 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1488 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1489 r = kvmhv_nested_page_fault(run, vcpu);
1490 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1492 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1493 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1494 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1495 DSISR_SRR1_MATCH_64S;
1496 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1497 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1498 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1499 r = kvmhv_nested_page_fault(run, vcpu);
1500 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1503 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1504 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1506 * This occurs for various TM-related instructions that
1507 * we need to emulate on POWER9 DD2.2. We have already
1508 * handled the cases where the guest was in real-suspend
1509 * mode and was transitioning to transactional state.
1511 r = kvmhv_p9_tm_emulation(vcpu);
1515 case BOOK3S_INTERRUPT_HV_RM_HARD:
1516 vcpu->arch.trap = 0;
1518 if (!xics_on_xive())
1519 kvmppc_xics_rm_complete(vcpu, 0);
1529 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1530 struct kvm_sregs *sregs)
1534 memset(sregs, 0, sizeof(struct kvm_sregs));
1535 sregs->pvr = vcpu->arch.pvr;
1536 for (i = 0; i < vcpu->arch.slb_max; i++) {
1537 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1538 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1544 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1545 struct kvm_sregs *sregs)
1549 /* Only accept the same PVR as the host's, since we can't spoof it */
1550 if (sregs->pvr != vcpu->arch.pvr)
1554 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1555 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1556 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1557 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1561 vcpu->arch.slb_max = j;
1566 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1567 bool preserve_top32)
1569 struct kvm *kvm = vcpu->kvm;
1570 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1573 spin_lock(&vc->lock);
1575 * If ILE (interrupt little-endian) has changed, update the
1576 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1578 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1579 struct kvm_vcpu *vcpu;
1582 kvm_for_each_vcpu(i, vcpu, kvm) {
1583 if (vcpu->arch.vcore != vc)
1585 if (new_lpcr & LPCR_ILE)
1586 vcpu->arch.intr_msr |= MSR_LE;
1588 vcpu->arch.intr_msr &= ~MSR_LE;
1593 * Userspace can only modify DPFD (default prefetch depth),
1594 * ILE (interrupt little-endian) and TC (translation control).
1595 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1597 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1598 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1601 * On POWER9, allow userspace to enable large decrementer for the
1602 * guest, whether or not the host has it enabled.
1604 if (cpu_has_feature(CPU_FTR_ARCH_300))
1607 /* Broken 32-bit version of LPCR must not clear top bits */
1610 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1611 spin_unlock(&vc->lock);
1614 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1615 union kvmppc_one_reg *val)
1621 case KVM_REG_PPC_DEBUG_INST:
1622 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1624 case KVM_REG_PPC_HIOR:
1625 *val = get_reg_val(id, 0);
1627 case KVM_REG_PPC_DABR:
1628 *val = get_reg_val(id, vcpu->arch.dabr);
1630 case KVM_REG_PPC_DABRX:
1631 *val = get_reg_val(id, vcpu->arch.dabrx);
1633 case KVM_REG_PPC_DSCR:
1634 *val = get_reg_val(id, vcpu->arch.dscr);
1636 case KVM_REG_PPC_PURR:
1637 *val = get_reg_val(id, vcpu->arch.purr);
1639 case KVM_REG_PPC_SPURR:
1640 *val = get_reg_val(id, vcpu->arch.spurr);
1642 case KVM_REG_PPC_AMR:
1643 *val = get_reg_val(id, vcpu->arch.amr);
1645 case KVM_REG_PPC_UAMOR:
1646 *val = get_reg_val(id, vcpu->arch.uamor);
1648 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1649 i = id - KVM_REG_PPC_MMCR0;
1650 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1652 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1653 i = id - KVM_REG_PPC_PMC1;
1654 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1656 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1657 i = id - KVM_REG_PPC_SPMC1;
1658 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1660 case KVM_REG_PPC_SIAR:
1661 *val = get_reg_val(id, vcpu->arch.siar);
1663 case KVM_REG_PPC_SDAR:
1664 *val = get_reg_val(id, vcpu->arch.sdar);
1666 case KVM_REG_PPC_SIER:
1667 *val = get_reg_val(id, vcpu->arch.sier);
1669 case KVM_REG_PPC_IAMR:
1670 *val = get_reg_val(id, vcpu->arch.iamr);
1672 case KVM_REG_PPC_PSPB:
1673 *val = get_reg_val(id, vcpu->arch.pspb);
1675 case KVM_REG_PPC_DPDES:
1677 * On POWER9, where we are emulating msgsndp etc.,
1678 * we return 1 bit for each vcpu, which can come from
1679 * either vcore->dpdes or doorbell_request.
1680 * On POWER8, doorbell_request is 0.
1682 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1683 vcpu->arch.doorbell_request);
1685 case KVM_REG_PPC_VTB:
1686 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1688 case KVM_REG_PPC_DAWR:
1689 *val = get_reg_val(id, vcpu->arch.dawr);
1691 case KVM_REG_PPC_DAWRX:
1692 *val = get_reg_val(id, vcpu->arch.dawrx);
1694 case KVM_REG_PPC_CIABR:
1695 *val = get_reg_val(id, vcpu->arch.ciabr);
1697 case KVM_REG_PPC_CSIGR:
1698 *val = get_reg_val(id, vcpu->arch.csigr);
1700 case KVM_REG_PPC_TACR:
1701 *val = get_reg_val(id, vcpu->arch.tacr);
1703 case KVM_REG_PPC_TCSCR:
1704 *val = get_reg_val(id, vcpu->arch.tcscr);
1706 case KVM_REG_PPC_PID:
1707 *val = get_reg_val(id, vcpu->arch.pid);
1709 case KVM_REG_PPC_ACOP:
1710 *val = get_reg_val(id, vcpu->arch.acop);
1712 case KVM_REG_PPC_WORT:
1713 *val = get_reg_val(id, vcpu->arch.wort);
1715 case KVM_REG_PPC_TIDR:
1716 *val = get_reg_val(id, vcpu->arch.tid);
1718 case KVM_REG_PPC_PSSCR:
1719 *val = get_reg_val(id, vcpu->arch.psscr);
1721 case KVM_REG_PPC_VPA_ADDR:
1722 spin_lock(&vcpu->arch.vpa_update_lock);
1723 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1724 spin_unlock(&vcpu->arch.vpa_update_lock);
1726 case KVM_REG_PPC_VPA_SLB:
1727 spin_lock(&vcpu->arch.vpa_update_lock);
1728 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1729 val->vpaval.length = vcpu->arch.slb_shadow.len;
1730 spin_unlock(&vcpu->arch.vpa_update_lock);
1732 case KVM_REG_PPC_VPA_DTL:
1733 spin_lock(&vcpu->arch.vpa_update_lock);
1734 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1735 val->vpaval.length = vcpu->arch.dtl.len;
1736 spin_unlock(&vcpu->arch.vpa_update_lock);
1738 case KVM_REG_PPC_TB_OFFSET:
1739 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1741 case KVM_REG_PPC_LPCR:
1742 case KVM_REG_PPC_LPCR_64:
1743 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1745 case KVM_REG_PPC_PPR:
1746 *val = get_reg_val(id, vcpu->arch.ppr);
1748 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1749 case KVM_REG_PPC_TFHAR:
1750 *val = get_reg_val(id, vcpu->arch.tfhar);
1752 case KVM_REG_PPC_TFIAR:
1753 *val = get_reg_val(id, vcpu->arch.tfiar);
1755 case KVM_REG_PPC_TEXASR:
1756 *val = get_reg_val(id, vcpu->arch.texasr);
1758 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1759 i = id - KVM_REG_PPC_TM_GPR0;
1760 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1762 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1765 i = id - KVM_REG_PPC_TM_VSR0;
1767 for (j = 0; j < TS_FPRWIDTH; j++)
1768 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1770 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1771 val->vval = vcpu->arch.vr_tm.vr[i-32];
1777 case KVM_REG_PPC_TM_CR:
1778 *val = get_reg_val(id, vcpu->arch.cr_tm);
1780 case KVM_REG_PPC_TM_XER:
1781 *val = get_reg_val(id, vcpu->arch.xer_tm);
1783 case KVM_REG_PPC_TM_LR:
1784 *val = get_reg_val(id, vcpu->arch.lr_tm);
1786 case KVM_REG_PPC_TM_CTR:
1787 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1789 case KVM_REG_PPC_TM_FPSCR:
1790 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1792 case KVM_REG_PPC_TM_AMR:
1793 *val = get_reg_val(id, vcpu->arch.amr_tm);
1795 case KVM_REG_PPC_TM_PPR:
1796 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1798 case KVM_REG_PPC_TM_VRSAVE:
1799 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1801 case KVM_REG_PPC_TM_VSCR:
1802 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1803 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1807 case KVM_REG_PPC_TM_DSCR:
1808 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1810 case KVM_REG_PPC_TM_TAR:
1811 *val = get_reg_val(id, vcpu->arch.tar_tm);
1814 case KVM_REG_PPC_ARCH_COMPAT:
1815 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1817 case KVM_REG_PPC_DEC_EXPIRY:
1818 *val = get_reg_val(id, vcpu->arch.dec_expires +
1819 vcpu->arch.vcore->tb_offset);
1821 case KVM_REG_PPC_ONLINE:
1822 *val = get_reg_val(id, vcpu->arch.online);
1824 case KVM_REG_PPC_PTCR:
1825 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1835 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1836 union kvmppc_one_reg *val)
1840 unsigned long addr, len;
1843 case KVM_REG_PPC_HIOR:
1844 /* Only allow this to be set to zero */
1845 if (set_reg_val(id, *val))
1848 case KVM_REG_PPC_DABR:
1849 vcpu->arch.dabr = set_reg_val(id, *val);
1851 case KVM_REG_PPC_DABRX:
1852 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1854 case KVM_REG_PPC_DSCR:
1855 vcpu->arch.dscr = set_reg_val(id, *val);
1857 case KVM_REG_PPC_PURR:
1858 vcpu->arch.purr = set_reg_val(id, *val);
1860 case KVM_REG_PPC_SPURR:
1861 vcpu->arch.spurr = set_reg_val(id, *val);
1863 case KVM_REG_PPC_AMR:
1864 vcpu->arch.amr = set_reg_val(id, *val);
1866 case KVM_REG_PPC_UAMOR:
1867 vcpu->arch.uamor = set_reg_val(id, *val);
1869 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1870 i = id - KVM_REG_PPC_MMCR0;
1871 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1873 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1874 i = id - KVM_REG_PPC_PMC1;
1875 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1877 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1878 i = id - KVM_REG_PPC_SPMC1;
1879 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1881 case KVM_REG_PPC_SIAR:
1882 vcpu->arch.siar = set_reg_val(id, *val);
1884 case KVM_REG_PPC_SDAR:
1885 vcpu->arch.sdar = set_reg_val(id, *val);
1887 case KVM_REG_PPC_SIER:
1888 vcpu->arch.sier = set_reg_val(id, *val);
1890 case KVM_REG_PPC_IAMR:
1891 vcpu->arch.iamr = set_reg_val(id, *val);
1893 case KVM_REG_PPC_PSPB:
1894 vcpu->arch.pspb = set_reg_val(id, *val);
1896 case KVM_REG_PPC_DPDES:
1897 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1899 case KVM_REG_PPC_VTB:
1900 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1902 case KVM_REG_PPC_DAWR:
1903 vcpu->arch.dawr = set_reg_val(id, *val);
1905 case KVM_REG_PPC_DAWRX:
1906 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1908 case KVM_REG_PPC_CIABR:
1909 vcpu->arch.ciabr = set_reg_val(id, *val);
1910 /* Don't allow setting breakpoints in hypervisor code */
1911 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1912 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1914 case KVM_REG_PPC_CSIGR:
1915 vcpu->arch.csigr = set_reg_val(id, *val);
1917 case KVM_REG_PPC_TACR:
1918 vcpu->arch.tacr = set_reg_val(id, *val);
1920 case KVM_REG_PPC_TCSCR:
1921 vcpu->arch.tcscr = set_reg_val(id, *val);
1923 case KVM_REG_PPC_PID:
1924 vcpu->arch.pid = set_reg_val(id, *val);
1926 case KVM_REG_PPC_ACOP:
1927 vcpu->arch.acop = set_reg_val(id, *val);
1929 case KVM_REG_PPC_WORT:
1930 vcpu->arch.wort = set_reg_val(id, *val);
1932 case KVM_REG_PPC_TIDR:
1933 vcpu->arch.tid = set_reg_val(id, *val);
1935 case KVM_REG_PPC_PSSCR:
1936 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1938 case KVM_REG_PPC_VPA_ADDR:
1939 addr = set_reg_val(id, *val);
1941 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1942 vcpu->arch.dtl.next_gpa))
1944 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1946 case KVM_REG_PPC_VPA_SLB:
1947 addr = val->vpaval.addr;
1948 len = val->vpaval.length;
1950 if (addr && !vcpu->arch.vpa.next_gpa)
1952 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1954 case KVM_REG_PPC_VPA_DTL:
1955 addr = val->vpaval.addr;
1956 len = val->vpaval.length;
1958 if (addr && (len < sizeof(struct dtl_entry) ||
1959 !vcpu->arch.vpa.next_gpa))
1961 len -= len % sizeof(struct dtl_entry);
1962 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1964 case KVM_REG_PPC_TB_OFFSET:
1965 /* round up to multiple of 2^24 */
1966 vcpu->arch.vcore->tb_offset =
1967 ALIGN(set_reg_val(id, *val), 1UL << 24);
1969 case KVM_REG_PPC_LPCR:
1970 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1972 case KVM_REG_PPC_LPCR_64:
1973 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1975 case KVM_REG_PPC_PPR:
1976 vcpu->arch.ppr = set_reg_val(id, *val);
1978 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1979 case KVM_REG_PPC_TFHAR:
1980 vcpu->arch.tfhar = set_reg_val(id, *val);
1982 case KVM_REG_PPC_TFIAR:
1983 vcpu->arch.tfiar = set_reg_val(id, *val);
1985 case KVM_REG_PPC_TEXASR:
1986 vcpu->arch.texasr = set_reg_val(id, *val);
1988 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1989 i = id - KVM_REG_PPC_TM_GPR0;
1990 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1992 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1995 i = id - KVM_REG_PPC_TM_VSR0;
1997 for (j = 0; j < TS_FPRWIDTH; j++)
1998 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2000 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2001 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2006 case KVM_REG_PPC_TM_CR:
2007 vcpu->arch.cr_tm = set_reg_val(id, *val);
2009 case KVM_REG_PPC_TM_XER:
2010 vcpu->arch.xer_tm = set_reg_val(id, *val);
2012 case KVM_REG_PPC_TM_LR:
2013 vcpu->arch.lr_tm = set_reg_val(id, *val);
2015 case KVM_REG_PPC_TM_CTR:
2016 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2018 case KVM_REG_PPC_TM_FPSCR:
2019 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2021 case KVM_REG_PPC_TM_AMR:
2022 vcpu->arch.amr_tm = set_reg_val(id, *val);
2024 case KVM_REG_PPC_TM_PPR:
2025 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2027 case KVM_REG_PPC_TM_VRSAVE:
2028 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2030 case KVM_REG_PPC_TM_VSCR:
2031 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2032 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2036 case KVM_REG_PPC_TM_DSCR:
2037 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2039 case KVM_REG_PPC_TM_TAR:
2040 vcpu->arch.tar_tm = set_reg_val(id, *val);
2043 case KVM_REG_PPC_ARCH_COMPAT:
2044 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2046 case KVM_REG_PPC_DEC_EXPIRY:
2047 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2048 vcpu->arch.vcore->tb_offset;
2050 case KVM_REG_PPC_ONLINE:
2051 i = set_reg_val(id, *val);
2052 if (i && !vcpu->arch.online)
2053 atomic_inc(&vcpu->arch.vcore->online_count);
2054 else if (!i && vcpu->arch.online)
2055 atomic_dec(&vcpu->arch.vcore->online_count);
2056 vcpu->arch.online = i;
2058 case KVM_REG_PPC_PTCR:
2059 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2070 * On POWER9, threads are independent and can be in different partitions.
2071 * Therefore we consider each thread to be a subcore.
2072 * There is a restriction that all threads have to be in the same
2073 * MMU mode (radix or HPT), unfortunately, but since we only support
2074 * HPT guests on a HPT host so far, that isn't an impediment yet.
2076 static int threads_per_vcore(struct kvm *kvm)
2078 if (kvm->arch.threads_indep)
2080 return threads_per_subcore;
2083 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2085 struct kvmppc_vcore *vcore;
2087 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2092 spin_lock_init(&vcore->lock);
2093 spin_lock_init(&vcore->stoltb_lock);
2094 init_swait_queue_head(&vcore->wq);
2095 vcore->preempt_tb = TB_NIL;
2096 vcore->lpcr = kvm->arch.lpcr;
2097 vcore->first_vcpuid = id;
2099 INIT_LIST_HEAD(&vcore->preempt_list);
2104 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2105 static struct debugfs_timings_element {
2109 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2110 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2111 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2112 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2113 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2116 #define N_TIMINGS (ARRAY_SIZE(timings))
2118 struct debugfs_timings_state {
2119 struct kvm_vcpu *vcpu;
2120 unsigned int buflen;
2121 char buf[N_TIMINGS * 100];
2124 static int debugfs_timings_open(struct inode *inode, struct file *file)
2126 struct kvm_vcpu *vcpu = inode->i_private;
2127 struct debugfs_timings_state *p;
2129 p = kzalloc(sizeof(*p), GFP_KERNEL);
2133 kvm_get_kvm(vcpu->kvm);
2135 file->private_data = p;
2137 return nonseekable_open(inode, file);
2140 static int debugfs_timings_release(struct inode *inode, struct file *file)
2142 struct debugfs_timings_state *p = file->private_data;
2144 kvm_put_kvm(p->vcpu->kvm);
2149 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2150 size_t len, loff_t *ppos)
2152 struct debugfs_timings_state *p = file->private_data;
2153 struct kvm_vcpu *vcpu = p->vcpu;
2155 struct kvmhv_tb_accumulator tb;
2164 buf_end = s + sizeof(p->buf);
2165 for (i = 0; i < N_TIMINGS; ++i) {
2166 struct kvmhv_tb_accumulator *acc;
2168 acc = (struct kvmhv_tb_accumulator *)
2169 ((unsigned long)vcpu + timings[i].offset);
2171 for (loops = 0; loops < 1000; ++loops) {
2172 count = acc->seqcount;
2177 if (count == acc->seqcount) {
2185 snprintf(s, buf_end - s, "%s: stuck\n",
2188 snprintf(s, buf_end - s,
2189 "%s: %llu %llu %llu %llu\n",
2190 timings[i].name, count / 2,
2191 tb_to_ns(tb.tb_total),
2192 tb_to_ns(tb.tb_min),
2193 tb_to_ns(tb.tb_max));
2196 p->buflen = s - p->buf;
2200 if (pos >= p->buflen)
2202 if (len > p->buflen - pos)
2203 len = p->buflen - pos;
2204 n = copy_to_user(buf, p->buf + pos, len);
2214 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2215 size_t len, loff_t *ppos)
2220 static const struct file_operations debugfs_timings_ops = {
2221 .owner = THIS_MODULE,
2222 .open = debugfs_timings_open,
2223 .release = debugfs_timings_release,
2224 .read = debugfs_timings_read,
2225 .write = debugfs_timings_write,
2226 .llseek = generic_file_llseek,
2229 /* Create a debugfs directory for the vcpu */
2230 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2233 struct kvm *kvm = vcpu->kvm;
2235 snprintf(buf, sizeof(buf), "vcpu%u", id);
2236 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2238 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2239 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2241 vcpu->arch.debugfs_timings =
2242 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2243 vcpu, &debugfs_timings_ops);
2246 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2247 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2250 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2252 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2255 struct kvm_vcpu *vcpu;
2258 struct kvmppc_vcore *vcore;
2261 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2265 err = kvm_vcpu_init(vcpu, kvm, id);
2269 vcpu->arch.shared = &vcpu->arch.shregs;
2270 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2272 * The shared struct is never shared on HV,
2273 * so we can always use host endianness
2275 #ifdef __BIG_ENDIAN__
2276 vcpu->arch.shared_big_endian = true;
2278 vcpu->arch.shared_big_endian = false;
2281 vcpu->arch.mmcr[0] = MMCR0_FC;
2282 vcpu->arch.ctrl = CTRL_RUNLATCH;
2283 /* default to host PVR, since we can't spoof it */
2284 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2285 spin_lock_init(&vcpu->arch.vpa_update_lock);
2286 spin_lock_init(&vcpu->arch.tbacct_lock);
2287 vcpu->arch.busy_preempt = TB_NIL;
2288 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2291 * Set the default HFSCR for the guest from the host value.
2292 * This value is only used on POWER9.
2293 * On POWER9, we want to virtualize the doorbell facility, so we
2294 * don't set the HFSCR_MSGP bit, and that causes those instructions
2295 * to trap and then we emulate them.
2297 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2298 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2299 if (cpu_has_feature(CPU_FTR_HVMODE)) {
2300 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2301 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2302 vcpu->arch.hfscr |= HFSCR_TM;
2304 if (cpu_has_feature(CPU_FTR_TM_COMP))
2305 vcpu->arch.hfscr |= HFSCR_TM;
2307 kvmppc_mmu_book3s_hv_init(vcpu);
2309 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2311 init_waitqueue_head(&vcpu->arch.cpu_run);
2313 mutex_lock(&kvm->lock);
2316 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2317 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2318 pr_devel("KVM: VCPU ID too high\n");
2319 core = KVM_MAX_VCORES;
2321 BUG_ON(kvm->arch.smt_mode != 1);
2322 core = kvmppc_pack_vcpu_id(kvm, id);
2325 core = id / kvm->arch.smt_mode;
2327 if (core < KVM_MAX_VCORES) {
2328 vcore = kvm->arch.vcores[core];
2329 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2330 pr_devel("KVM: collision on id %u", id);
2332 } else if (!vcore) {
2334 * Take mmu_setup_lock for mutual exclusion
2335 * with kvmppc_update_lpcr().
2338 vcore = kvmppc_vcore_create(kvm,
2339 id & ~(kvm->arch.smt_mode - 1));
2340 mutex_lock(&kvm->arch.mmu_setup_lock);
2341 kvm->arch.vcores[core] = vcore;
2342 kvm->arch.online_vcores++;
2343 mutex_unlock(&kvm->arch.mmu_setup_lock);
2346 mutex_unlock(&kvm->lock);
2351 spin_lock(&vcore->lock);
2352 ++vcore->num_threads;
2353 spin_unlock(&vcore->lock);
2354 vcpu->arch.vcore = vcore;
2355 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2356 vcpu->arch.thread_cpu = -1;
2357 vcpu->arch.prev_cpu = -1;
2359 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2360 kvmppc_sanity_check(vcpu);
2362 debugfs_vcpu_init(vcpu, id);
2367 kmem_cache_free(kvm_vcpu_cache, vcpu);
2369 return ERR_PTR(err);
2372 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2373 unsigned long flags)
2380 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2382 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2384 * On POWER8 (or POWER7), the threading mode is "strict",
2385 * so we pack smt_mode vcpus per vcore.
2387 if (smt_mode > threads_per_subcore)
2391 * On POWER9, the threading mode is "loose",
2392 * so each vcpu gets its own vcore.
2397 mutex_lock(&kvm->lock);
2399 if (!kvm->arch.online_vcores) {
2400 kvm->arch.smt_mode = smt_mode;
2401 kvm->arch.emul_smt_mode = esmt;
2404 mutex_unlock(&kvm->lock);
2409 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2411 if (vpa->pinned_addr)
2412 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2416 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2418 spin_lock(&vcpu->arch.vpa_update_lock);
2419 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2420 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2421 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2422 spin_unlock(&vcpu->arch.vpa_update_lock);
2423 kvm_vcpu_uninit(vcpu);
2424 kmem_cache_free(kvm_vcpu_cache, vcpu);
2427 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2429 /* Indicate we want to get back into the guest */
2433 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2435 unsigned long dec_nsec, now;
2438 if (now > vcpu->arch.dec_expires) {
2439 /* decrementer has already gone negative */
2440 kvmppc_core_queue_dec(vcpu);
2441 kvmppc_core_prepare_to_enter(vcpu);
2444 dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2445 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2446 vcpu->arch.timer_running = 1;
2449 extern int __kvmppc_vcore_entry(void);
2451 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2452 struct kvm_vcpu *vcpu)
2456 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2458 spin_lock_irq(&vcpu->arch.tbacct_lock);
2460 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2461 vcpu->arch.stolen_logged;
2462 vcpu->arch.busy_preempt = now;
2463 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2464 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2466 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2469 static int kvmppc_grab_hwthread(int cpu)
2471 struct paca_struct *tpaca;
2472 long timeout = 10000;
2474 tpaca = paca_ptrs[cpu];
2476 /* Ensure the thread won't go into the kernel if it wakes */
2477 tpaca->kvm_hstate.kvm_vcpu = NULL;
2478 tpaca->kvm_hstate.kvm_vcore = NULL;
2479 tpaca->kvm_hstate.napping = 0;
2481 tpaca->kvm_hstate.hwthread_req = 1;
2484 * If the thread is already executing in the kernel (e.g. handling
2485 * a stray interrupt), wait for it to get back to nap mode.
2486 * The smp_mb() is to ensure that our setting of hwthread_req
2487 * is visible before we look at hwthread_state, so if this
2488 * races with the code at system_reset_pSeries and the thread
2489 * misses our setting of hwthread_req, we are sure to see its
2490 * setting of hwthread_state, and vice versa.
2493 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2494 if (--timeout <= 0) {
2495 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2503 static void kvmppc_release_hwthread(int cpu)
2505 struct paca_struct *tpaca;
2507 tpaca = paca_ptrs[cpu];
2508 tpaca->kvm_hstate.hwthread_req = 0;
2509 tpaca->kvm_hstate.kvm_vcpu = NULL;
2510 tpaca->kvm_hstate.kvm_vcore = NULL;
2511 tpaca->kvm_hstate.kvm_split_mode = NULL;
2514 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2516 struct kvm_nested_guest *nested = vcpu->arch.nested;
2517 cpumask_t *cpu_in_guest;
2520 cpu = cpu_first_thread_sibling(cpu);
2522 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2523 cpu_in_guest = &nested->cpu_in_guest;
2525 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2526 cpu_in_guest = &kvm->arch.cpu_in_guest;
2529 * Make sure setting of bit in need_tlb_flush precedes
2530 * testing of cpu_in_guest bits. The matching barrier on
2531 * the other side is the first smp_mb() in kvmppc_run_core().
2534 for (i = 0; i < threads_per_core; ++i)
2535 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2536 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2539 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2541 struct kvm_nested_guest *nested = vcpu->arch.nested;
2542 struct kvm *kvm = vcpu->kvm;
2545 if (!cpu_has_feature(CPU_FTR_HVMODE))
2549 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2551 prev_cpu = vcpu->arch.prev_cpu;
2554 * With radix, the guest can do TLB invalidations itself,
2555 * and it could choose to use the local form (tlbiel) if
2556 * it is invalidating a translation that has only ever been
2557 * used on one vcpu. However, that doesn't mean it has
2558 * only ever been used on one physical cpu, since vcpus
2559 * can move around between pcpus. To cope with this, when
2560 * a vcpu moves from one pcpu to another, we need to tell
2561 * any vcpus running on the same core as this vcpu previously
2562 * ran to flush the TLB. The TLB is shared between threads,
2563 * so we use a single bit in .need_tlb_flush for all 4 threads.
2565 if (prev_cpu != pcpu) {
2566 if (prev_cpu >= 0 &&
2567 cpu_first_thread_sibling(prev_cpu) !=
2568 cpu_first_thread_sibling(pcpu))
2569 radix_flush_cpu(kvm, prev_cpu, vcpu);
2571 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2573 vcpu->arch.prev_cpu = pcpu;
2577 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2580 struct paca_struct *tpaca;
2581 struct kvm *kvm = vc->kvm;
2585 if (vcpu->arch.timer_running) {
2586 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2587 vcpu->arch.timer_running = 0;
2589 cpu += vcpu->arch.ptid;
2590 vcpu->cpu = vc->pcpu;
2591 vcpu->arch.thread_cpu = cpu;
2592 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2594 tpaca = paca_ptrs[cpu];
2595 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2596 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2597 tpaca->kvm_hstate.fake_suspend = 0;
2598 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2600 tpaca->kvm_hstate.kvm_vcore = vc;
2601 if (cpu != smp_processor_id())
2602 kvmppc_ipi_thread(cpu);
2605 static void kvmppc_wait_for_nap(int n_threads)
2607 int cpu = smp_processor_id();
2612 for (loops = 0; loops < 1000000; ++loops) {
2614 * Check if all threads are finished.
2615 * We set the vcore pointer when starting a thread
2616 * and the thread clears it when finished, so we look
2617 * for any threads that still have a non-NULL vcore ptr.
2619 for (i = 1; i < n_threads; ++i)
2620 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2622 if (i == n_threads) {
2629 for (i = 1; i < n_threads; ++i)
2630 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2631 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2635 * Check that we are on thread 0 and that any other threads in
2636 * this core are off-line. Then grab the threads so they can't
2639 static int on_primary_thread(void)
2641 int cpu = smp_processor_id();
2644 /* Are we on a primary subcore? */
2645 if (cpu_thread_in_subcore(cpu))
2649 while (++thr < threads_per_subcore)
2650 if (cpu_online(cpu + thr))
2653 /* Grab all hw threads so they can't go into the kernel */
2654 for (thr = 1; thr < threads_per_subcore; ++thr) {
2655 if (kvmppc_grab_hwthread(cpu + thr)) {
2656 /* Couldn't grab one; let the others go */
2658 kvmppc_release_hwthread(cpu + thr);
2659 } while (--thr > 0);
2667 * A list of virtual cores for each physical CPU.
2668 * These are vcores that could run but their runner VCPU tasks are
2669 * (or may be) preempted.
2671 struct preempted_vcore_list {
2672 struct list_head list;
2676 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2678 static void init_vcore_lists(void)
2682 for_each_possible_cpu(cpu) {
2683 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2684 spin_lock_init(&lp->lock);
2685 INIT_LIST_HEAD(&lp->list);
2689 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2691 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2693 vc->vcore_state = VCORE_PREEMPT;
2694 vc->pcpu = smp_processor_id();
2695 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2696 spin_lock(&lp->lock);
2697 list_add_tail(&vc->preempt_list, &lp->list);
2698 spin_unlock(&lp->lock);
2701 /* Start accumulating stolen time */
2702 kvmppc_core_start_stolen(vc);
2705 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2707 struct preempted_vcore_list *lp;
2709 kvmppc_core_end_stolen(vc);
2710 if (!list_empty(&vc->preempt_list)) {
2711 lp = &per_cpu(preempted_vcores, vc->pcpu);
2712 spin_lock(&lp->lock);
2713 list_del_init(&vc->preempt_list);
2714 spin_unlock(&lp->lock);
2716 vc->vcore_state = VCORE_INACTIVE;
2720 * This stores information about the virtual cores currently
2721 * assigned to a physical core.
2725 int max_subcore_threads;
2727 int subcore_threads[MAX_SUBCORES];
2728 struct kvmppc_vcore *vc[MAX_SUBCORES];
2732 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2733 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2735 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2737 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2739 memset(cip, 0, sizeof(*cip));
2740 cip->n_subcores = 1;
2741 cip->max_subcore_threads = vc->num_threads;
2742 cip->total_threads = vc->num_threads;
2743 cip->subcore_threads[0] = vc->num_threads;
2747 static bool subcore_config_ok(int n_subcores, int n_threads)
2750 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2751 * split-core mode, with one thread per subcore.
2753 if (cpu_has_feature(CPU_FTR_ARCH_300))
2754 return n_subcores <= 4 && n_threads == 1;
2756 /* On POWER8, can only dynamically split if unsplit to begin with */
2757 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2759 if (n_subcores > MAX_SUBCORES)
2761 if (n_subcores > 1) {
2762 if (!(dynamic_mt_modes & 2))
2764 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2768 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2771 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2773 vc->entry_exit_map = 0;
2775 vc->napping_threads = 0;
2776 vc->conferring_threads = 0;
2777 vc->tb_offset_applied = 0;
2780 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2782 int n_threads = vc->num_threads;
2785 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2788 /* In one_vm_per_core mode, require all vcores to be from the same vm */
2789 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2792 /* Some POWER9 chips require all threads to be in the same MMU mode */
2793 if (no_mixing_hpt_and_radix &&
2794 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2797 if (n_threads < cip->max_subcore_threads)
2798 n_threads = cip->max_subcore_threads;
2799 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2801 cip->max_subcore_threads = n_threads;
2803 sub = cip->n_subcores;
2805 cip->total_threads += vc->num_threads;
2806 cip->subcore_threads[sub] = vc->num_threads;
2808 init_vcore_to_run(vc);
2809 list_del_init(&vc->preempt_list);
2815 * Work out whether it is possible to piggyback the execution of
2816 * vcore *pvc onto the execution of the other vcores described in *cip.
2818 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2821 if (cip->total_threads + pvc->num_threads > target_threads)
2824 return can_dynamic_split(pvc, cip);
2827 static void prepare_threads(struct kvmppc_vcore *vc)
2830 struct kvm_vcpu *vcpu;
2832 for_each_runnable_thread(i, vcpu, vc) {
2833 if (signal_pending(vcpu->arch.run_task))
2834 vcpu->arch.ret = -EINTR;
2835 else if (vcpu->arch.vpa.update_pending ||
2836 vcpu->arch.slb_shadow.update_pending ||
2837 vcpu->arch.dtl.update_pending)
2838 vcpu->arch.ret = RESUME_GUEST;
2841 kvmppc_remove_runnable(vc, vcpu);
2842 wake_up(&vcpu->arch.cpu_run);
2846 static void collect_piggybacks(struct core_info *cip, int target_threads)
2848 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2849 struct kvmppc_vcore *pvc, *vcnext;
2851 spin_lock(&lp->lock);
2852 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2853 if (!spin_trylock(&pvc->lock))
2855 prepare_threads(pvc);
2856 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2857 list_del_init(&pvc->preempt_list);
2858 if (pvc->runner == NULL) {
2859 pvc->vcore_state = VCORE_INACTIVE;
2860 kvmppc_core_end_stolen(pvc);
2862 spin_unlock(&pvc->lock);
2865 if (!can_piggyback(pvc, cip, target_threads)) {
2866 spin_unlock(&pvc->lock);
2869 kvmppc_core_end_stolen(pvc);
2870 pvc->vcore_state = VCORE_PIGGYBACK;
2871 if (cip->total_threads >= target_threads)
2874 spin_unlock(&lp->lock);
2877 static bool recheck_signals_and_mmu(struct core_info *cip)
2880 struct kvm_vcpu *vcpu;
2881 struct kvmppc_vcore *vc;
2883 for (sub = 0; sub < cip->n_subcores; ++sub) {
2885 if (!vc->kvm->arch.mmu_ready)
2887 for_each_runnable_thread(i, vcpu, vc)
2888 if (signal_pending(vcpu->arch.run_task))
2894 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2896 int still_running = 0, i;
2899 struct kvm_vcpu *vcpu;
2901 spin_lock(&vc->lock);
2903 for_each_runnable_thread(i, vcpu, vc) {
2905 * It's safe to unlock the vcore in the loop here, because
2906 * for_each_runnable_thread() is safe against removal of
2907 * the vcpu, and the vcore state is VCORE_EXITING here,
2908 * so any vcpus becoming runnable will have their arch.trap
2909 * set to zero and can't actually run in the guest.
2911 spin_unlock(&vc->lock);
2912 /* cancel pending dec exception if dec is positive */
2913 if (now < vcpu->arch.dec_expires &&
2914 kvmppc_core_pending_dec(vcpu))
2915 kvmppc_core_dequeue_dec(vcpu);
2917 trace_kvm_guest_exit(vcpu);
2920 if (vcpu->arch.trap)
2921 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2922 vcpu->arch.run_task);
2924 vcpu->arch.ret = ret;
2925 vcpu->arch.trap = 0;
2927 spin_lock(&vc->lock);
2928 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2929 if (vcpu->arch.pending_exceptions)
2930 kvmppc_core_prepare_to_enter(vcpu);
2931 if (vcpu->arch.ceded)
2932 kvmppc_set_timer(vcpu);
2936 kvmppc_remove_runnable(vc, vcpu);
2937 wake_up(&vcpu->arch.cpu_run);
2941 if (still_running > 0) {
2942 kvmppc_vcore_preempt(vc);
2943 } else if (vc->runner) {
2944 vc->vcore_state = VCORE_PREEMPT;
2945 kvmppc_core_start_stolen(vc);
2947 vc->vcore_state = VCORE_INACTIVE;
2949 if (vc->n_runnable > 0 && vc->runner == NULL) {
2950 /* make sure there's a candidate runner awake */
2952 vcpu = next_runnable_thread(vc, &i);
2953 wake_up(&vcpu->arch.cpu_run);
2956 spin_unlock(&vc->lock);
2960 * Clear core from the list of active host cores as we are about to
2961 * enter the guest. Only do this if it is the primary thread of the
2962 * core (not if a subcore) that is entering the guest.
2964 static inline int kvmppc_clear_host_core(unsigned int cpu)
2968 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2971 * Memory barrier can be omitted here as we will do a smp_wmb()
2972 * later in kvmppc_start_thread and we need ensure that state is
2973 * visible to other CPUs only after we enter guest.
2975 core = cpu >> threads_shift;
2976 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2981 * Advertise this core as an active host core since we exited the guest
2982 * Only need to do this if it is the primary thread of the core that is
2985 static inline int kvmppc_set_host_core(unsigned int cpu)
2989 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2993 * Memory barrier can be omitted here because we do a spin_unlock
2994 * immediately after this which provides the memory barrier.
2996 core = cpu >> threads_shift;
2997 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3001 static void set_irq_happened(int trap)
3004 case BOOK3S_INTERRUPT_EXTERNAL:
3005 local_paca->irq_happened |= PACA_IRQ_EE;
3007 case BOOK3S_INTERRUPT_H_DOORBELL:
3008 local_paca->irq_happened |= PACA_IRQ_DBELL;
3010 case BOOK3S_INTERRUPT_HMI:
3011 local_paca->irq_happened |= PACA_IRQ_HMI;
3013 case BOOK3S_INTERRUPT_SYSTEM_RESET:
3014 replay_system_reset();
3020 * Run a set of guest threads on a physical core.
3021 * Called with vc->lock held.
3023 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3025 struct kvm_vcpu *vcpu;
3028 struct core_info core_info;
3029 struct kvmppc_vcore *pvc;
3030 struct kvm_split_mode split_info, *sip;
3031 int split, subcore_size, active;
3034 unsigned long cmd_bit, stat_bit;
3037 int controlled_threads;
3043 * Remove from the list any threads that have a signal pending
3044 * or need a VPA update done
3046 prepare_threads(vc);
3048 /* if the runner is no longer runnable, let the caller pick a new one */
3049 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3055 init_vcore_to_run(vc);
3056 vc->preempt_tb = TB_NIL;
3059 * Number of threads that we will be controlling: the same as
3060 * the number of threads per subcore, except on POWER9,
3061 * where it's 1 because the threads are (mostly) independent.
3063 controlled_threads = threads_per_vcore(vc->kvm);
3066 * Make sure we are running on primary threads, and that secondary
3067 * threads are offline. Also check if the number of threads in this
3068 * guest are greater than the current system threads per guest.
3069 * On POWER9, we need to be not in independent-threads mode if
3070 * this is a HPT guest on a radix host machine where the
3071 * CPU threads may not be in different MMU modes.
3073 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3074 !kvm_is_radix(vc->kvm);
3075 if (((controlled_threads > 1) &&
3076 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3077 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3078 for_each_runnable_thread(i, vcpu, vc) {
3079 vcpu->arch.ret = -EBUSY;
3080 kvmppc_remove_runnable(vc, vcpu);
3081 wake_up(&vcpu->arch.cpu_run);
3087 * See if we could run any other vcores on the physical core
3088 * along with this one.
3090 init_core_info(&core_info, vc);
3091 pcpu = smp_processor_id();
3092 target_threads = controlled_threads;
3093 if (target_smt_mode && target_smt_mode < target_threads)
3094 target_threads = target_smt_mode;
3095 if (vc->num_threads < target_threads)
3096 collect_piggybacks(&core_info, target_threads);
3099 * On radix, arrange for TLB flushing if necessary.
3100 * This has to be done before disabling interrupts since
3101 * it uses smp_call_function().
3103 pcpu = smp_processor_id();
3104 if (kvm_is_radix(vc->kvm)) {
3105 for (sub = 0; sub < core_info.n_subcores; ++sub)
3106 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3107 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3111 * Hard-disable interrupts, and check resched flag and signals.
3112 * If we need to reschedule or deliver a signal, clean up
3113 * and return without going into the guest(s).
3114 * If the mmu_ready flag has been cleared, don't go into the
3115 * guest because that means a HPT resize operation is in progress.
3117 local_irq_disable();
3119 if (lazy_irq_pending() || need_resched() ||
3120 recheck_signals_and_mmu(&core_info)) {
3122 vc->vcore_state = VCORE_INACTIVE;
3123 /* Unlock all except the primary vcore */
3124 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3125 pvc = core_info.vc[sub];
3126 /* Put back on to the preempted vcores list */
3127 kvmppc_vcore_preempt(pvc);
3128 spin_unlock(&pvc->lock);
3130 for (i = 0; i < controlled_threads; ++i)
3131 kvmppc_release_hwthread(pcpu + i);
3135 kvmppc_clear_host_core(pcpu);
3137 /* Decide on micro-threading (split-core) mode */
3138 subcore_size = threads_per_subcore;
3139 cmd_bit = stat_bit = 0;
3140 split = core_info.n_subcores;
3142 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3143 && !cpu_has_feature(CPU_FTR_ARCH_300);
3145 if (split > 1 || hpt_on_radix) {
3147 memset(&split_info, 0, sizeof(split_info));
3148 for (sub = 0; sub < core_info.n_subcores; ++sub)
3149 split_info.vc[sub] = core_info.vc[sub];
3152 if (split == 2 && (dynamic_mt_modes & 2)) {
3153 cmd_bit = HID0_POWER8_1TO2LPAR;
3154 stat_bit = HID0_POWER8_2LPARMODE;
3157 cmd_bit = HID0_POWER8_1TO4LPAR;
3158 stat_bit = HID0_POWER8_4LPARMODE;
3160 subcore_size = MAX_SMT_THREADS / split;
3161 split_info.rpr = mfspr(SPRN_RPR);
3162 split_info.pmmar = mfspr(SPRN_PMMAR);
3163 split_info.ldbar = mfspr(SPRN_LDBAR);
3164 split_info.subcore_size = subcore_size;
3166 split_info.subcore_size = 1;
3168 /* Use the split_info for LPCR/LPIDR changes */
3169 split_info.lpcr_req = vc->lpcr;
3170 split_info.lpidr_req = vc->kvm->arch.lpid;
3171 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3172 split_info.do_set = 1;
3176 /* order writes to split_info before kvm_split_mode pointer */
3180 for (thr = 0; thr < controlled_threads; ++thr) {
3181 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3183 paca->kvm_hstate.tid = thr;
3184 paca->kvm_hstate.napping = 0;
3185 paca->kvm_hstate.kvm_split_mode = sip;
3188 /* Initiate micro-threading (split-core) on POWER8 if required */
3190 unsigned long hid0 = mfspr(SPRN_HID0);
3192 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3194 mtspr(SPRN_HID0, hid0);
3197 hid0 = mfspr(SPRN_HID0);
3198 if (hid0 & stat_bit)
3205 * On POWER8, set RWMR register.
3206 * Since it only affects PURR and SPURR, it doesn't affect
3207 * the host, so we don't save/restore the host value.
3210 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3211 int n_online = atomic_read(&vc->online_count);
3214 * Use the 8-thread value if we're doing split-core
3215 * or if the vcore's online count looks bogus.
3217 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3218 n_online >= 1 && n_online <= MAX_SMT_THREADS)
3219 rwmr_val = p8_rwmr_values[n_online];
3220 mtspr(SPRN_RWMR, rwmr_val);
3223 /* Start all the threads */
3225 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3226 thr = is_power8 ? subcore_thread_map[sub] : sub;
3229 pvc = core_info.vc[sub];
3230 pvc->pcpu = pcpu + thr;
3231 for_each_runnable_thread(i, vcpu, pvc) {
3232 kvmppc_start_thread(vcpu, pvc);
3233 kvmppc_create_dtl_entry(vcpu, pvc);
3234 trace_kvm_guest_enter(vcpu);
3235 if (!vcpu->arch.ptid)
3237 active |= 1 << (thr + vcpu->arch.ptid);
3240 * We need to start the first thread of each subcore
3241 * even if it doesn't have a vcpu.
3244 kvmppc_start_thread(NULL, pvc);
3248 * Ensure that split_info.do_nap is set after setting
3249 * the vcore pointer in the PACA of the secondaries.
3254 * When doing micro-threading, poke the inactive threads as well.
3255 * This gets them to the nap instruction after kvm_do_nap,
3256 * which reduces the time taken to unsplit later.
3257 * For POWER9 HPT guest on radix host, we need all the secondary
3258 * threads woken up so they can do the LPCR/LPIDR change.
3260 if (cmd_bit || hpt_on_radix) {
3261 split_info.do_nap = 1; /* ask secondaries to nap when done */
3262 for (thr = 1; thr < threads_per_subcore; ++thr)
3263 if (!(active & (1 << thr)))
3264 kvmppc_ipi_thread(pcpu + thr);
3267 vc->vcore_state = VCORE_RUNNING;
3270 trace_kvmppc_run_core(vc, 0);
3272 for (sub = 0; sub < core_info.n_subcores; ++sub)
3273 spin_unlock(&core_info.vc[sub]->lock);
3275 guest_enter_irqoff();
3277 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3279 this_cpu_disable_ftrace();
3282 * Interrupts will be enabled once we get into the guest,
3283 * so tell lockdep that we're about to enable interrupts.
3285 trace_hardirqs_on();
3287 trap = __kvmppc_vcore_entry();
3289 trace_hardirqs_off();
3291 this_cpu_enable_ftrace();
3293 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3295 set_irq_happened(trap);
3297 spin_lock(&vc->lock);
3298 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3299 vc->vcore_state = VCORE_EXITING;
3301 /* wait for secondary threads to finish writing their state to memory */
3302 kvmppc_wait_for_nap(controlled_threads);
3304 /* Return to whole-core mode if we split the core earlier */
3306 unsigned long hid0 = mfspr(SPRN_HID0);
3307 unsigned long loops = 0;
3309 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3310 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3312 mtspr(SPRN_HID0, hid0);
3315 hid0 = mfspr(SPRN_HID0);
3316 if (!(hid0 & stat_bit))
3321 } else if (hpt_on_radix) {
3322 /* Wait for all threads to have seen final sync */
3323 for (thr = 1; thr < controlled_threads; ++thr) {
3324 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3326 while (paca->kvm_hstate.kvm_split_mode) {
3333 split_info.do_nap = 0;
3335 kvmppc_set_host_core(pcpu);
3340 /* Let secondaries go back to the offline loop */
3341 for (i = 0; i < controlled_threads; ++i) {
3342 kvmppc_release_hwthread(pcpu + i);
3343 if (sip && sip->napped[i])
3344 kvmppc_ipi_thread(pcpu + i);
3345 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3348 spin_unlock(&vc->lock);
3350 /* make sure updates to secondary vcpu structs are visible now */
3355 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3356 pvc = core_info.vc[sub];
3357 post_guest_process(pvc, pvc == vc);
3360 spin_lock(&vc->lock);
3363 vc->vcore_state = VCORE_INACTIVE;
3364 trace_kvmppc_run_core(vc, 1);
3368 * Load up hypervisor-mode registers on P9.
3370 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3373 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3375 u64 tb, purr, spurr;
3377 unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3378 unsigned long host_ciabr = mfspr(SPRN_CIABR);
3379 unsigned long host_dawr = mfspr(SPRN_DAWR);
3380 unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3381 unsigned long host_psscr = mfspr(SPRN_PSSCR);
3382 unsigned long host_pidr = mfspr(SPRN_PID);
3384 hdec = time_limit - mftb();
3386 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3387 mtspr(SPRN_HDEC, hdec);
3389 if (vc->tb_offset) {
3390 u64 new_tb = mftb() + vc->tb_offset;
3391 mtspr(SPRN_TBU40, new_tb);
3393 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3394 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3395 vc->tb_offset_applied = vc->tb_offset;
3399 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3400 mtspr(SPRN_DPDES, vc->dpdes);
3401 mtspr(SPRN_VTB, vc->vtb);
3403 local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3404 local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3405 mtspr(SPRN_PURR, vcpu->arch.purr);
3406 mtspr(SPRN_SPURR, vcpu->arch.spurr);
3408 if (dawr_enabled()) {
3409 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3410 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3412 mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3413 mtspr(SPRN_IC, vcpu->arch.ic);
3414 mtspr(SPRN_PID, vcpu->arch.pid);
3416 mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3417 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3419 mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3421 mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3422 mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3423 mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3424 mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3426 mtspr(SPRN_AMOR, ~0UL);
3428 mtspr(SPRN_LPCR, lpcr);
3431 kvmppc_xive_push_vcpu(vcpu);
3433 mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3434 mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3436 trap = __kvmhv_vcpu_entry_p9(vcpu);
3438 /* Advance host PURR/SPURR by the amount used by guest */
3439 purr = mfspr(SPRN_PURR);
3440 spurr = mfspr(SPRN_SPURR);
3441 mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3442 purr - vcpu->arch.purr);
3443 mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3444 spurr - vcpu->arch.spurr);
3445 vcpu->arch.purr = purr;
3446 vcpu->arch.spurr = spurr;
3448 vcpu->arch.ic = mfspr(SPRN_IC);
3449 vcpu->arch.pid = mfspr(SPRN_PID);
3450 vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3452 vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3453 vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3454 vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3455 vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3457 /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3458 mtspr(SPRN_PSSCR, host_psscr |
3459 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3460 mtspr(SPRN_HFSCR, host_hfscr);
3461 mtspr(SPRN_CIABR, host_ciabr);
3462 mtspr(SPRN_DAWR, host_dawr);
3463 mtspr(SPRN_DAWRX, host_dawrx);
3464 mtspr(SPRN_PID, host_pidr);
3467 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3468 * case we interrupted the guest between a tlbie and a ptesync.
3470 asm volatile("eieio; tlbsync; ptesync");
3472 mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid); /* restore host LPID */
3475 vc->dpdes = mfspr(SPRN_DPDES);
3476 vc->vtb = mfspr(SPRN_VTB);
3477 mtspr(SPRN_DPDES, 0);
3479 mtspr(SPRN_PCR, PCR_MASK);
3481 if (vc->tb_offset_applied) {
3482 u64 new_tb = mftb() - vc->tb_offset_applied;
3483 mtspr(SPRN_TBU40, new_tb);
3485 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3486 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3487 vc->tb_offset_applied = 0;
3490 mtspr(SPRN_HDEC, 0x7fffffff);
3491 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3497 * Virtual-mode guest entry for POWER9 and later when the host and
3498 * guest are both using the radix MMU. The LPIDR has already been set.
3500 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3503 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3504 unsigned long host_dscr = mfspr(SPRN_DSCR);
3505 unsigned long host_tidr = mfspr(SPRN_TIDR);
3506 unsigned long host_iamr = mfspr(SPRN_IAMR);
3507 unsigned long host_amr = mfspr(SPRN_AMR);
3512 dec = mfspr(SPRN_DEC);
3515 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3516 local_paca->kvm_hstate.dec_expires = dec + tb;
3517 if (local_paca->kvm_hstate.dec_expires < time_limit)
3518 time_limit = local_paca->kvm_hstate.dec_expires;
3520 vcpu->arch.ceded = 0;
3522 kvmhv_save_host_pmu(); /* saves it to PACA kvm_hstate */
3524 kvmppc_subcore_enter_guest();
3526 vc->entry_exit_map = 1;
3529 if (vcpu->arch.vpa.pinned_addr) {
3530 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3531 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3532 lp->yield_count = cpu_to_be32(yield_count);
3533 vcpu->arch.vpa.dirty = 1;
3536 if (cpu_has_feature(CPU_FTR_TM) ||
3537 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3538 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3540 kvmhv_load_guest_pmu(vcpu);
3542 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3543 load_fp_state(&vcpu->arch.fp);
3544 #ifdef CONFIG_ALTIVEC
3545 load_vr_state(&vcpu->arch.vr);
3547 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3549 mtspr(SPRN_DSCR, vcpu->arch.dscr);
3550 mtspr(SPRN_IAMR, vcpu->arch.iamr);
3551 mtspr(SPRN_PSPB, vcpu->arch.pspb);
3552 mtspr(SPRN_FSCR, vcpu->arch.fscr);
3553 mtspr(SPRN_TAR, vcpu->arch.tar);
3554 mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3555 mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3556 mtspr(SPRN_BESCR, vcpu->arch.bescr);
3557 mtspr(SPRN_WORT, vcpu->arch.wort);
3558 mtspr(SPRN_TIDR, vcpu->arch.tid);
3559 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3560 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3561 mtspr(SPRN_AMR, vcpu->arch.amr);
3562 mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3564 if (!(vcpu->arch.ctrl & 1))
3565 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3567 mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3569 if (kvmhv_on_pseries()) {
3571 * We need to save and restore the guest visible part of the
3572 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3573 * doesn't do this for us. Note only required if pseries since
3574 * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3576 unsigned long host_psscr;
3577 /* call our hypervisor to load up HV regs and go */
3578 struct hv_guest_state hvregs;
3580 host_psscr = mfspr(SPRN_PSSCR_PR);
3581 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3582 kvmhv_save_hv_regs(vcpu, &hvregs);
3584 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3585 hvregs.version = HV_GUEST_STATE_VERSION;
3586 if (vcpu->arch.nested) {
3587 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3588 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3590 hvregs.lpid = vcpu->kvm->arch.lpid;
3591 hvregs.vcpu_token = vcpu->vcpu_id;
3593 hvregs.hdec_expiry = time_limit;
3594 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3595 __pa(&vcpu->arch.regs));
3596 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3597 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3598 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3599 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3600 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3601 mtspr(SPRN_PSSCR_PR, host_psscr);
3603 /* H_CEDE has to be handled now, not later */
3604 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3605 kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3606 kvmppc_nested_cede(vcpu);
3610 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3613 vcpu->arch.slb_max = 0;
3614 dec = mfspr(SPRN_DEC);
3615 if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3618 vcpu->arch.dec_expires = dec + tb;
3620 vcpu->arch.thread_cpu = -1;
3621 vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3623 vcpu->arch.iamr = mfspr(SPRN_IAMR);
3624 vcpu->arch.pspb = mfspr(SPRN_PSPB);
3625 vcpu->arch.fscr = mfspr(SPRN_FSCR);
3626 vcpu->arch.tar = mfspr(SPRN_TAR);
3627 vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3628 vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3629 vcpu->arch.bescr = mfspr(SPRN_BESCR);
3630 vcpu->arch.wort = mfspr(SPRN_WORT);
3631 vcpu->arch.tid = mfspr(SPRN_TIDR);
3632 vcpu->arch.amr = mfspr(SPRN_AMR);
3633 vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3634 vcpu->arch.dscr = mfspr(SPRN_DSCR);
3636 mtspr(SPRN_PSPB, 0);
3637 mtspr(SPRN_WORT, 0);
3638 mtspr(SPRN_UAMOR, 0);
3639 mtspr(SPRN_DSCR, host_dscr);
3640 mtspr(SPRN_TIDR, host_tidr);
3641 mtspr(SPRN_IAMR, host_iamr);
3642 mtspr(SPRN_PSPB, 0);
3644 if (host_amr != vcpu->arch.amr)
3645 mtspr(SPRN_AMR, host_amr);
3647 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3648 store_fp_state(&vcpu->arch.fp);
3649 #ifdef CONFIG_ALTIVEC
3650 store_vr_state(&vcpu->arch.vr);
3652 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3654 if (cpu_has_feature(CPU_FTR_TM) ||
3655 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3656 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3659 if (vcpu->arch.vpa.pinned_addr) {
3660 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3661 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3662 lp->yield_count = cpu_to_be32(yield_count);
3663 vcpu->arch.vpa.dirty = 1;
3664 save_pmu = lp->pmcregs_in_use;
3666 /* Must save pmu if this guest is capable of running nested guests */
3667 save_pmu |= nesting_enabled(vcpu->kvm);
3669 kvmhv_save_guest_pmu(vcpu, save_pmu);
3671 vc->entry_exit_map = 0x101;
3674 mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3675 mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3677 kvmhv_load_host_pmu();
3679 kvmppc_subcore_exit_guest();
3685 * Wait for some other vcpu thread to execute us, and
3686 * wake us up when we need to handle something in the host.
3688 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3689 struct kvm_vcpu *vcpu, int wait_state)
3693 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3694 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3695 spin_unlock(&vc->lock);
3697 spin_lock(&vc->lock);
3699 finish_wait(&vcpu->arch.cpu_run, &wait);
3702 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3704 if (!halt_poll_ns_grow)
3707 vc->halt_poll_ns *= halt_poll_ns_grow;
3708 if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3709 vc->halt_poll_ns = halt_poll_ns_grow_start;
3712 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3714 if (halt_poll_ns_shrink == 0)
3715 vc->halt_poll_ns = 0;
3717 vc->halt_poll_ns /= halt_poll_ns_shrink;
3720 #ifdef CONFIG_KVM_XICS
3721 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3723 if (!xics_on_xive())
3725 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3726 vcpu->arch.xive_saved_state.cppr;
3729 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3733 #endif /* CONFIG_KVM_XICS */
3735 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3737 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3738 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3745 * Check to see if any of the runnable vcpus on the vcore have pending
3746 * exceptions or are no longer ceded
3748 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3750 struct kvm_vcpu *vcpu;
3753 for_each_runnable_thread(i, vcpu, vc) {
3754 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3762 * All the vcpus in this vcore are idle, so wait for a decrementer
3763 * or external interrupt to one of the vcpus. vc->lock is held.
3765 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3767 ktime_t cur, start_poll, start_wait;
3770 DECLARE_SWAITQUEUE(wait);
3772 /* Poll for pending exceptions and ceded state */
3773 cur = start_poll = ktime_get();
3774 if (vc->halt_poll_ns) {
3775 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3776 ++vc->runner->stat.halt_attempted_poll;
3778 vc->vcore_state = VCORE_POLLING;
3779 spin_unlock(&vc->lock);
3782 if (kvmppc_vcore_check_block(vc)) {
3787 } while (single_task_running() && ktime_before(cur, stop));
3789 spin_lock(&vc->lock);
3790 vc->vcore_state = VCORE_INACTIVE;
3793 ++vc->runner->stat.halt_successful_poll;
3798 prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3800 if (kvmppc_vcore_check_block(vc)) {
3801 finish_swait(&vc->wq, &wait);
3803 /* If we polled, count this as a successful poll */
3804 if (vc->halt_poll_ns)
3805 ++vc->runner->stat.halt_successful_poll;
3809 start_wait = ktime_get();
3811 vc->vcore_state = VCORE_SLEEPING;
3812 trace_kvmppc_vcore_blocked(vc, 0);
3813 spin_unlock(&vc->lock);
3815 finish_swait(&vc->wq, &wait);
3816 spin_lock(&vc->lock);
3817 vc->vcore_state = VCORE_INACTIVE;
3818 trace_kvmppc_vcore_blocked(vc, 1);
3819 ++vc->runner->stat.halt_successful_wait;
3824 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3826 /* Attribute wait time */
3828 vc->runner->stat.halt_wait_ns +=
3829 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3830 /* Attribute failed poll time */
3831 if (vc->halt_poll_ns)
3832 vc->runner->stat.halt_poll_fail_ns +=
3833 ktime_to_ns(start_wait) -
3834 ktime_to_ns(start_poll);
3836 /* Attribute successful poll time */
3837 if (vc->halt_poll_ns)
3838 vc->runner->stat.halt_poll_success_ns +=
3840 ktime_to_ns(start_poll);
3843 /* Adjust poll time */
3845 if (block_ns <= vc->halt_poll_ns)
3847 /* We slept and blocked for longer than the max halt time */
3848 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3849 shrink_halt_poll_ns(vc);
3850 /* We slept and our poll time is too small */
3851 else if (vc->halt_poll_ns < halt_poll_ns &&
3852 block_ns < halt_poll_ns)
3853 grow_halt_poll_ns(vc);
3854 if (vc->halt_poll_ns > halt_poll_ns)
3855 vc->halt_poll_ns = halt_poll_ns;
3857 vc->halt_poll_ns = 0;
3859 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3863 * This never fails for a radix guest, as none of the operations it does
3864 * for a radix guest can fail or have a way to report failure.
3865 * kvmhv_run_single_vcpu() relies on this fact.
3867 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3870 struct kvm *kvm = vcpu->kvm;
3872 mutex_lock(&kvm->arch.mmu_setup_lock);
3873 if (!kvm->arch.mmu_ready) {
3874 if (!kvm_is_radix(kvm))
3875 r = kvmppc_hv_setup_htab_rma(vcpu);
3877 if (cpu_has_feature(CPU_FTR_ARCH_300))
3878 kvmppc_setup_partition_table(kvm);
3879 kvm->arch.mmu_ready = 1;
3882 mutex_unlock(&kvm->arch.mmu_setup_lock);
3886 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3889 struct kvmppc_vcore *vc;
3892 trace_kvmppc_run_vcpu_enter(vcpu);
3894 kvm_run->exit_reason = 0;
3895 vcpu->arch.ret = RESUME_GUEST;
3896 vcpu->arch.trap = 0;
3897 kvmppc_update_vpas(vcpu);
3900 * Synchronize with other threads in this virtual core
3902 vc = vcpu->arch.vcore;
3903 spin_lock(&vc->lock);
3904 vcpu->arch.ceded = 0;
3905 vcpu->arch.run_task = current;
3906 vcpu->arch.kvm_run = kvm_run;
3907 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3908 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3909 vcpu->arch.busy_preempt = TB_NIL;
3910 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3914 * This happens the first time this is called for a vcpu.
3915 * If the vcore is already running, we may be able to start
3916 * this thread straight away and have it join in.
3918 if (!signal_pending(current)) {
3919 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3920 vc->vcore_state == VCORE_RUNNING) &&
3921 !VCORE_IS_EXITING(vc)) {
3922 kvmppc_create_dtl_entry(vcpu, vc);
3923 kvmppc_start_thread(vcpu, vc);
3924 trace_kvm_guest_enter(vcpu);
3925 } else if (vc->vcore_state == VCORE_SLEEPING) {
3926 swake_up_one(&vc->wq);
3931 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3932 !signal_pending(current)) {
3933 /* See if the MMU is ready to go */
3934 if (!vcpu->kvm->arch.mmu_ready) {
3935 spin_unlock(&vc->lock);
3936 r = kvmhv_setup_mmu(vcpu);
3937 spin_lock(&vc->lock);
3939 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3940 kvm_run->fail_entry.
3941 hardware_entry_failure_reason = 0;
3947 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3948 kvmppc_vcore_end_preempt(vc);
3950 if (vc->vcore_state != VCORE_INACTIVE) {
3951 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3954 for_each_runnable_thread(i, v, vc) {
3955 kvmppc_core_prepare_to_enter(v);
3956 if (signal_pending(v->arch.run_task)) {
3957 kvmppc_remove_runnable(vc, v);
3958 v->stat.signal_exits++;
3959 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3960 v->arch.ret = -EINTR;
3961 wake_up(&v->arch.cpu_run);
3964 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3967 for_each_runnable_thread(i, v, vc) {
3968 if (!kvmppc_vcpu_woken(v))
3969 n_ceded += v->arch.ceded;
3974 if (n_ceded == vc->n_runnable) {
3975 kvmppc_vcore_blocked(vc);
3976 } else if (need_resched()) {
3977 kvmppc_vcore_preempt(vc);
3978 /* Let something else run */
3979 cond_resched_lock(&vc->lock);
3980 if (vc->vcore_state == VCORE_PREEMPT)
3981 kvmppc_vcore_end_preempt(vc);
3983 kvmppc_run_core(vc);
3988 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3989 (vc->vcore_state == VCORE_RUNNING ||
3990 vc->vcore_state == VCORE_EXITING ||
3991 vc->vcore_state == VCORE_PIGGYBACK))
3992 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3994 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3995 kvmppc_vcore_end_preempt(vc);
3997 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3998 kvmppc_remove_runnable(vc, vcpu);
3999 vcpu->stat.signal_exits++;
4000 kvm_run->exit_reason = KVM_EXIT_INTR;
4001 vcpu->arch.ret = -EINTR;
4004 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4005 /* Wake up some vcpu to run the core */
4007 v = next_runnable_thread(vc, &i);
4008 wake_up(&v->arch.cpu_run);
4011 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4012 spin_unlock(&vc->lock);
4013 return vcpu->arch.ret;
4016 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4017 struct kvm_vcpu *vcpu, u64 time_limit,
4022 struct kvmppc_vcore *vc;
4023 struct kvm *kvm = vcpu->kvm;
4024 struct kvm_nested_guest *nested = vcpu->arch.nested;
4026 trace_kvmppc_run_vcpu_enter(vcpu);
4028 kvm_run->exit_reason = 0;
4029 vcpu->arch.ret = RESUME_GUEST;
4030 vcpu->arch.trap = 0;
4032 vc = vcpu->arch.vcore;
4033 vcpu->arch.ceded = 0;
4034 vcpu->arch.run_task = current;
4035 vcpu->arch.kvm_run = kvm_run;
4036 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4037 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4038 vcpu->arch.busy_preempt = TB_NIL;
4039 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4040 vc->runnable_threads[0] = vcpu;
4044 /* See if the MMU is ready to go */
4045 if (!kvm->arch.mmu_ready)
4046 kvmhv_setup_mmu(vcpu);
4051 kvmppc_update_vpas(vcpu);
4053 init_vcore_to_run(vc);
4054 vc->preempt_tb = TB_NIL;
4057 pcpu = smp_processor_id();
4059 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4061 local_irq_disable();
4063 if (signal_pending(current))
4065 if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4069 kvmppc_core_prepare_to_enter(vcpu);
4070 if (vcpu->arch.doorbell_request) {
4073 vcpu->arch.doorbell_request = 0;
4075 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4076 &vcpu->arch.pending_exceptions))
4078 } else if (vcpu->arch.pending_exceptions ||
4079 vcpu->arch.doorbell_request ||
4080 xive_interrupt_pending(vcpu)) {
4081 vcpu->arch.ret = RESUME_HOST;
4085 kvmppc_clear_host_core(pcpu);
4087 local_paca->kvm_hstate.tid = 0;
4088 local_paca->kvm_hstate.napping = 0;
4089 local_paca->kvm_hstate.kvm_split_mode = NULL;
4090 kvmppc_start_thread(vcpu, vc);
4091 kvmppc_create_dtl_entry(vcpu, vc);
4092 trace_kvm_guest_enter(vcpu);
4094 vc->vcore_state = VCORE_RUNNING;
4095 trace_kvmppc_run_core(vc, 0);
4097 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4098 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4099 mtspr(SPRN_LPID, lpid);
4101 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4104 guest_enter_irqoff();
4106 srcu_idx = srcu_read_lock(&kvm->srcu);
4108 this_cpu_disable_ftrace();
4110 /* Tell lockdep that we're about to enable interrupts */
4111 trace_hardirqs_on();
4113 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4114 vcpu->arch.trap = trap;
4116 trace_hardirqs_off();
4118 this_cpu_enable_ftrace();
4120 srcu_read_unlock(&kvm->srcu, srcu_idx);
4122 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4123 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4127 set_irq_happened(trap);
4129 kvmppc_set_host_core(pcpu);
4134 cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4139 * cancel pending decrementer exception if DEC is now positive, or if
4140 * entering a nested guest in which case the decrementer is now owned
4141 * by L2 and the L1 decrementer is provided in hdec_expires
4143 if (kvmppc_core_pending_dec(vcpu) &&
4144 ((get_tb() < vcpu->arch.dec_expires) ||
4145 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4146 kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4147 kvmppc_core_dequeue_dec(vcpu);
4149 trace_kvm_guest_exit(vcpu);
4153 r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4155 r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4159 if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4160 !kvmppc_vcpu_woken(vcpu)) {
4161 kvmppc_set_timer(vcpu);
4162 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4163 if (signal_pending(current)) {
4164 vcpu->stat.signal_exits++;
4165 kvm_run->exit_reason = KVM_EXIT_INTR;
4166 vcpu->arch.ret = -EINTR;
4169 spin_lock(&vc->lock);
4170 kvmppc_vcore_blocked(vc);
4171 spin_unlock(&vc->lock);
4174 vcpu->arch.ceded = 0;
4176 vc->vcore_state = VCORE_INACTIVE;
4177 trace_kvmppc_run_core(vc, 1);
4180 kvmppc_remove_runnable(vc, vcpu);
4181 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4183 return vcpu->arch.ret;
4186 vcpu->stat.signal_exits++;
4187 kvm_run->exit_reason = KVM_EXIT_INTR;
4188 vcpu->arch.ret = -EINTR;
4195 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4199 unsigned long ebb_regs[3] = {}; /* shut up GCC */
4200 unsigned long user_tar = 0;
4201 unsigned int user_vrsave;
4204 if (!vcpu->arch.sane) {
4205 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4210 * Don't allow entry with a suspended transaction, because
4211 * the guest entry/exit code will lose it.
4212 * If the guest has TM enabled, save away their TM-related SPRs
4213 * (they will get restored by the TM unavailable interrupt).
4215 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4216 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4217 (current->thread.regs->msr & MSR_TM)) {
4218 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4219 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4220 run->fail_entry.hardware_entry_failure_reason = 0;
4223 /* Enable TM so we can read the TM SPRs */
4224 mtmsr(mfmsr() | MSR_TM);
4225 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4226 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4227 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4228 current->thread.regs->msr &= ~MSR_TM;
4233 * Force online to 1 for the sake of old userspace which doesn't
4236 if (!vcpu->arch.online) {
4237 atomic_inc(&vcpu->arch.vcore->online_count);
4238 vcpu->arch.online = 1;
4241 kvmppc_core_prepare_to_enter(vcpu);
4243 /* No need to go into the guest when all we'll do is come back out */
4244 if (signal_pending(current)) {
4245 run->exit_reason = KVM_EXIT_INTR;
4250 atomic_inc(&kvm->arch.vcpus_running);
4251 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4254 flush_all_to_thread(current);
4256 /* Save userspace EBB and other register values */
4257 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4258 ebb_regs[0] = mfspr(SPRN_EBBHR);
4259 ebb_regs[1] = mfspr(SPRN_EBBRR);
4260 ebb_regs[2] = mfspr(SPRN_BESCR);
4261 user_tar = mfspr(SPRN_TAR);
4263 user_vrsave = mfspr(SPRN_VRSAVE);
4265 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4266 vcpu->arch.pgdir = current->mm->pgd;
4267 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4271 * The early POWER9 chips that can't mix radix and HPT threads
4272 * on the same core also need the workaround for the problem
4273 * where the TLB would prefetch entries in the guest exit path
4274 * for radix guests using the guest PIDR value and LPID 0.
4275 * The workaround is in the old path (kvmppc_run_vcpu())
4276 * but not the new path (kvmhv_run_single_vcpu()).
4278 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4279 !no_mixing_hpt_and_radix)
4280 r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4281 vcpu->arch.vcore->lpcr);
4283 r = kvmppc_run_vcpu(run, vcpu);
4285 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4286 !(vcpu->arch.shregs.msr & MSR_PR)) {
4287 trace_kvm_hcall_enter(vcpu);
4288 r = kvmppc_pseries_do_hcall(vcpu);
4289 trace_kvm_hcall_exit(vcpu, r);
4290 kvmppc_core_prepare_to_enter(vcpu);
4291 } else if (r == RESUME_PAGE_FAULT) {
4292 srcu_idx = srcu_read_lock(&kvm->srcu);
4293 r = kvmppc_book3s_hv_page_fault(run, vcpu,
4294 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4295 srcu_read_unlock(&kvm->srcu, srcu_idx);
4296 } else if (r == RESUME_PASSTHROUGH) {
4297 if (WARN_ON(xics_on_xive()))
4300 r = kvmppc_xics_rm_complete(vcpu, 0);
4302 } while (is_kvmppc_resume_guest(r));
4304 /* Restore userspace EBB and other register values */
4305 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4306 mtspr(SPRN_EBBHR, ebb_regs[0]);
4307 mtspr(SPRN_EBBRR, ebb_regs[1]);
4308 mtspr(SPRN_BESCR, ebb_regs[2]);
4309 mtspr(SPRN_TAR, user_tar);
4310 mtspr(SPRN_FSCR, current->thread.fscr);
4312 mtspr(SPRN_VRSAVE, user_vrsave);
4314 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4315 atomic_dec(&kvm->arch.vcpus_running);
4319 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4320 int shift, int sllp)
4322 (*sps)->page_shift = shift;
4323 (*sps)->slb_enc = sllp;
4324 (*sps)->enc[0].page_shift = shift;
4325 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4327 * Add 16MB MPSS support (may get filtered out by userspace)
4330 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4332 (*sps)->enc[1].page_shift = 24;
4333 (*sps)->enc[1].pte_enc = penc;
4339 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4340 struct kvm_ppc_smmu_info *info)
4342 struct kvm_ppc_one_seg_page_size *sps;
4345 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4346 * POWER7 doesn't support keys for instruction accesses,
4347 * POWER8 and POWER9 do.
4349 info->data_keys = 32;
4350 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4352 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4353 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4354 info->slb_size = 32;
4356 /* We only support these sizes for now, and no muti-size segments */
4357 sps = &info->sps[0];
4358 kvmppc_add_seg_page_size(&sps, 12, 0);
4359 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4360 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4362 /* If running as a nested hypervisor, we don't support HPT guests */
4363 if (kvmhv_on_pseries())
4364 info->flags |= KVM_PPC_NO_HASH;
4370 * Get (and clear) the dirty memory log for a memory slot.
4372 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4373 struct kvm_dirty_log *log)
4375 struct kvm_memslots *slots;
4376 struct kvm_memory_slot *memslot;
4379 unsigned long *buf, *p;
4380 struct kvm_vcpu *vcpu;
4382 mutex_lock(&kvm->slots_lock);
4385 if (log->slot >= KVM_USER_MEM_SLOTS)
4388 slots = kvm_memslots(kvm);
4389 memslot = id_to_memslot(slots, log->slot);
4391 if (!memslot->dirty_bitmap)
4395 * Use second half of bitmap area because both HPT and radix
4396 * accumulate bits in the first half.
4398 n = kvm_dirty_bitmap_bytes(memslot);
4399 buf = memslot->dirty_bitmap + n / sizeof(long);
4402 if (kvm_is_radix(kvm))
4403 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4405 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4410 * We accumulate dirty bits in the first half of the
4411 * memslot's dirty_bitmap area, for when pages are paged
4412 * out or modified by the host directly. Pick up these
4413 * bits and add them to the map.
4415 p = memslot->dirty_bitmap;
4416 for (i = 0; i < n / sizeof(long); ++i)
4417 buf[i] |= xchg(&p[i], 0);
4419 /* Harvest dirty bits from VPA and DTL updates */
4420 /* Note: we never modify the SLB shadow buffer areas */
4421 kvm_for_each_vcpu(i, vcpu, kvm) {
4422 spin_lock(&vcpu->arch.vpa_update_lock);
4423 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4424 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4425 spin_unlock(&vcpu->arch.vpa_update_lock);
4429 if (copy_to_user(log->dirty_bitmap, buf, n))
4434 mutex_unlock(&kvm->slots_lock);
4438 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4439 struct kvm_memory_slot *dont)
4441 if (!dont || free->arch.rmap != dont->arch.rmap) {
4442 vfree(free->arch.rmap);
4443 free->arch.rmap = NULL;
4447 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4448 unsigned long npages)
4450 slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4451 if (!slot->arch.rmap)
4457 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4458 struct kvm_memory_slot *memslot,
4459 const struct kvm_userspace_memory_region *mem)
4464 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4465 const struct kvm_userspace_memory_region *mem,
4466 const struct kvm_memory_slot *old,
4467 const struct kvm_memory_slot *new,
4468 enum kvm_mr_change change)
4470 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4473 * If we are making a new memslot, it might make
4474 * some address that was previously cached as emulated
4475 * MMIO be no longer emulated MMIO, so invalidate
4476 * all the caches of emulated MMIO translations.
4479 atomic64_inc(&kvm->arch.mmio_update);
4482 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4483 * have already called kvm_arch_flush_shadow_memslot() to
4484 * flush shadow mappings. For KVM_MR_CREATE we have no
4485 * previous mappings. So the only case to handle is
4486 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4488 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4489 * to get rid of any THP PTEs in the partition-scoped page tables
4490 * so we can track dirtiness at the page level; we flush when
4491 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4494 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4495 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4496 kvmppc_radix_flush_memslot(kvm, old);
4500 * Update LPCR values in kvm->arch and in vcores.
4501 * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4502 * of kvm->arch.lpcr update).
4504 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4509 if ((kvm->arch.lpcr & mask) == lpcr)
4512 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4514 for (i = 0; i < KVM_MAX_VCORES; ++i) {
4515 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4518 spin_lock(&vc->lock);
4519 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4520 spin_unlock(&vc->lock);
4521 if (++cores_done >= kvm->arch.online_vcores)
4526 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4531 void kvmppc_setup_partition_table(struct kvm *kvm)
4533 unsigned long dw0, dw1;
4535 if (!kvm_is_radix(kvm)) {
4536 /* PS field - page size for VRMA */
4537 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4538 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4539 /* HTABSIZE and HTABORG fields */
4540 dw0 |= kvm->arch.sdr1;
4542 /* Second dword as set by userspace */
4543 dw1 = kvm->arch.process_table;
4545 dw0 = PATB_HR | radix__get_tree_size() |
4546 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4547 dw1 = PATB_GR | kvm->arch.process_table;
4549 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4553 * Set up HPT (hashed page table) and RMA (real-mode area).
4554 * Must be called with kvm->arch.mmu_setup_lock held.
4556 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4559 struct kvm *kvm = vcpu->kvm;
4561 struct kvm_memory_slot *memslot;
4562 struct vm_area_struct *vma;
4563 unsigned long lpcr = 0, senc;
4564 unsigned long psize, porder;
4567 /* Allocate hashed page table (if not done already) and reset it */
4568 if (!kvm->arch.hpt.virt) {
4569 int order = KVM_DEFAULT_HPT_ORDER;
4570 struct kvm_hpt_info info;
4572 err = kvmppc_allocate_hpt(&info, order);
4573 /* If we get here, it means userspace didn't specify a
4574 * size explicitly. So, try successively smaller
4575 * sizes if the default failed. */
4576 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4577 err = kvmppc_allocate_hpt(&info, order);
4580 pr_err("KVM: Couldn't alloc HPT\n");
4584 kvmppc_set_hpt(kvm, &info);
4587 /* Look up the memslot for guest physical address 0 */
4588 srcu_idx = srcu_read_lock(&kvm->srcu);
4589 memslot = gfn_to_memslot(kvm, 0);
4591 /* We must have some memory at 0 by now */
4593 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4596 /* Look up the VMA for the start of this memory slot */
4597 hva = memslot->userspace_addr;
4598 down_read(¤t->mm->mmap_sem);
4599 vma = find_vma(current->mm, hva);
4600 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4603 psize = vma_kernel_pagesize(vma);
4605 up_read(¤t->mm->mmap_sem);
4607 /* We can handle 4k, 64k or 16M pages in the VRMA */
4608 if (psize >= 0x1000000)
4610 else if (psize >= 0x10000)
4614 porder = __ilog2(psize);
4616 senc = slb_pgsize_encoding(psize);
4617 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4618 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4619 /* Create HPTEs in the hash page table for the VRMA */
4620 kvmppc_map_vrma(vcpu, memslot, porder);
4622 /* Update VRMASD field in the LPCR */
4623 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4624 /* the -4 is to account for senc values starting at 0x10 */
4625 lpcr = senc << (LPCR_VRMASD_SH - 4);
4626 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4629 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4633 srcu_read_unlock(&kvm->srcu, srcu_idx);
4638 up_read(¤t->mm->mmap_sem);
4643 * Must be called with kvm->arch.mmu_setup_lock held and
4644 * mmu_ready = 0 and no vcpus running.
4646 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4648 if (nesting_enabled(kvm))
4649 kvmhv_release_all_nested(kvm);
4650 kvmppc_rmap_reset(kvm);
4651 kvm->arch.process_table = 0;
4652 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4653 spin_lock(&kvm->mmu_lock);
4654 kvm->arch.radix = 0;
4655 spin_unlock(&kvm->mmu_lock);
4656 kvmppc_free_radix(kvm);
4657 kvmppc_update_lpcr(kvm, LPCR_VPM1,
4658 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4663 * Must be called with kvm->arch.mmu_setup_lock held and
4664 * mmu_ready = 0 and no vcpus running.
4666 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4670 err = kvmppc_init_vm_radix(kvm);
4673 kvmppc_rmap_reset(kvm);
4674 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4675 spin_lock(&kvm->mmu_lock);
4676 kvm->arch.radix = 1;
4677 spin_unlock(&kvm->mmu_lock);
4678 kvmppc_free_hpt(&kvm->arch.hpt);
4679 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4680 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4684 #ifdef CONFIG_KVM_XICS
4686 * Allocate a per-core structure for managing state about which cores are
4687 * running in the host versus the guest and for exchanging data between
4688 * real mode KVM and CPU running in the host.
4689 * This is only done for the first VM.
4690 * The allocated structure stays even if all VMs have stopped.
4691 * It is only freed when the kvm-hv module is unloaded.
4692 * It's OK for this routine to fail, we just don't support host
4693 * core operations like redirecting H_IPI wakeups.
4695 void kvmppc_alloc_host_rm_ops(void)
4697 struct kvmppc_host_rm_ops *ops;
4698 unsigned long l_ops;
4702 /* Not the first time here ? */
4703 if (kvmppc_host_rm_ops_hv != NULL)
4706 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4710 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4711 ops->rm_core = kzalloc(size, GFP_KERNEL);
4713 if (!ops->rm_core) {
4720 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4721 if (!cpu_online(cpu))
4724 core = cpu >> threads_shift;
4725 ops->rm_core[core].rm_state.in_host = 1;
4728 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4731 * Make the contents of the kvmppc_host_rm_ops structure visible
4732 * to other CPUs before we assign it to the global variable.
4733 * Do an atomic assignment (no locks used here), but if someone
4734 * beats us to it, just free our copy and return.
4737 l_ops = (unsigned long) ops;
4739 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4741 kfree(ops->rm_core);
4746 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4747 "ppc/kvm_book3s:prepare",
4748 kvmppc_set_host_core,
4749 kvmppc_clear_host_core);
4753 void kvmppc_free_host_rm_ops(void)
4755 if (kvmppc_host_rm_ops_hv) {
4756 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4757 kfree(kvmppc_host_rm_ops_hv->rm_core);
4758 kfree(kvmppc_host_rm_ops_hv);
4759 kvmppc_host_rm_ops_hv = NULL;
4764 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4766 unsigned long lpcr, lpid;
4770 mutex_init(&kvm->arch.mmu_setup_lock);
4772 /* Allocate the guest's logical partition ID */
4774 lpid = kvmppc_alloc_lpid();
4777 kvm->arch.lpid = lpid;
4779 kvmppc_alloc_host_rm_ops();
4781 kvmhv_vm_nested_init(kvm);
4784 * Since we don't flush the TLB when tearing down a VM,
4785 * and this lpid might have previously been used,
4786 * make sure we flush on each core before running the new VM.
4787 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4788 * does this flush for us.
4790 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4791 cpumask_setall(&kvm->arch.need_tlb_flush);
4793 /* Start out with the default set of hcalls enabled */
4794 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4795 sizeof(kvm->arch.enabled_hcalls));
4797 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4798 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4800 /* Init LPCR for virtual RMA mode */
4801 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4802 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4803 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4804 lpcr &= LPCR_PECE | LPCR_LPES;
4808 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4809 LPCR_VPM0 | LPCR_VPM1;
4810 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4811 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4812 /* On POWER8 turn on online bit to enable PURR/SPURR */
4813 if (cpu_has_feature(CPU_FTR_ARCH_207S))
4816 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4817 * Set HVICE bit to enable hypervisor virtualization interrupts.
4818 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4819 * be unnecessary but better safe than sorry in case we re-enable
4820 * EE in HV mode with this LPCR still set)
4822 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4824 lpcr |= LPCR_HVICE | LPCR_HEIC;
4827 * If xive is enabled, we route 0x500 interrupts directly
4835 * If the host uses radix, the guest starts out as radix.
4837 if (radix_enabled()) {
4838 kvm->arch.radix = 1;
4839 kvm->arch.mmu_ready = 1;
4841 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4842 ret = kvmppc_init_vm_radix(kvm);
4844 kvmppc_free_lpid(kvm->arch.lpid);
4847 kvmppc_setup_partition_table(kvm);
4850 kvm->arch.lpcr = lpcr;
4852 /* Initialization for future HPT resizes */
4853 kvm->arch.resize_hpt = NULL;
4856 * Work out how many sets the TLB has, for the use of
4857 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4859 if (radix_enabled())
4860 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
4861 else if (cpu_has_feature(CPU_FTR_ARCH_300))
4862 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
4863 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4864 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
4866 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
4869 * Track that we now have a HV mode VM active. This blocks secondary
4870 * CPU threads from coming online.
4871 * On POWER9, we only need to do this if the "indep_threads_mode"
4872 * module parameter has been set to N.
4874 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4875 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4876 pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4877 kvm->arch.threads_indep = true;
4879 kvm->arch.threads_indep = indep_threads_mode;
4882 if (!kvm->arch.threads_indep)
4883 kvm_hv_vm_activated();
4886 * Initialize smt_mode depending on processor.
4887 * POWER8 and earlier have to use "strict" threading, where
4888 * all vCPUs in a vcore have to run on the same (sub)core,
4889 * whereas on POWER9 the threads can each run a different
4892 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4893 kvm->arch.smt_mode = threads_per_subcore;
4895 kvm->arch.smt_mode = 1;
4896 kvm->arch.emul_smt_mode = 1;
4899 * Create a debugfs directory for the VM
4901 snprintf(buf, sizeof(buf), "vm%d", current->pid);
4902 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4903 kvmppc_mmu_debugfs_init(kvm);
4904 if (radix_enabled())
4905 kvmhv_radix_debugfs_init(kvm);
4910 static void kvmppc_free_vcores(struct kvm *kvm)
4914 for (i = 0; i < KVM_MAX_VCORES; ++i)
4915 kfree(kvm->arch.vcores[i]);
4916 kvm->arch.online_vcores = 0;
4919 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4921 debugfs_remove_recursive(kvm->arch.debugfs_dir);
4923 if (!kvm->arch.threads_indep)
4924 kvm_hv_vm_deactivated();
4926 kvmppc_free_vcores(kvm);
4929 if (kvm_is_radix(kvm))
4930 kvmppc_free_radix(kvm);
4932 kvmppc_free_hpt(&kvm->arch.hpt);
4934 /* Perform global invalidation and return lpid to the pool */
4935 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4936 if (nesting_enabled(kvm))
4937 kvmhv_release_all_nested(kvm);
4938 kvm->arch.process_table = 0;
4939 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4941 kvmppc_free_lpid(kvm->arch.lpid);
4943 kvmppc_free_pimap(kvm);
4946 /* We don't need to emulate any privileged instructions or dcbz */
4947 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4948 unsigned int inst, int *advance)
4950 return EMULATE_FAIL;
4953 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4956 return EMULATE_FAIL;
4959 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4962 return EMULATE_FAIL;
4965 static int kvmppc_core_check_processor_compat_hv(void)
4967 if (cpu_has_feature(CPU_FTR_HVMODE) &&
4968 cpu_has_feature(CPU_FTR_ARCH_206))
4971 /* POWER9 in radix mode is capable of being a nested hypervisor. */
4972 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4978 #ifdef CONFIG_KVM_XICS
4980 void kvmppc_free_pimap(struct kvm *kvm)
4982 kfree(kvm->arch.pimap);
4985 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4987 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4990 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4992 struct irq_desc *desc;
4993 struct kvmppc_irq_map *irq_map;
4994 struct kvmppc_passthru_irqmap *pimap;
4995 struct irq_chip *chip;
4998 if (!kvm_irq_bypass)
5001 desc = irq_to_desc(host_irq);
5005 mutex_lock(&kvm->lock);
5007 pimap = kvm->arch.pimap;
5008 if (pimap == NULL) {
5009 /* First call, allocate structure to hold IRQ map */
5010 pimap = kvmppc_alloc_pimap();
5011 if (pimap == NULL) {
5012 mutex_unlock(&kvm->lock);
5015 kvm->arch.pimap = pimap;
5019 * For now, we only support interrupts for which the EOI operation
5020 * is an OPAL call followed by a write to XIRR, since that's
5021 * what our real-mode EOI code does, or a XIVE interrupt
5023 chip = irq_data_get_irq_chip(&desc->irq_data);
5024 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5025 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5026 host_irq, guest_gsi);
5027 mutex_unlock(&kvm->lock);
5032 * See if we already have an entry for this guest IRQ number.
5033 * If it's mapped to a hardware IRQ number, that's an error,
5034 * otherwise re-use this entry.
5036 for (i = 0; i < pimap->n_mapped; i++) {
5037 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5038 if (pimap->mapped[i].r_hwirq) {
5039 mutex_unlock(&kvm->lock);
5046 if (i == KVMPPC_PIRQ_MAPPED) {
5047 mutex_unlock(&kvm->lock);
5048 return -EAGAIN; /* table is full */
5051 irq_map = &pimap->mapped[i];
5053 irq_map->v_hwirq = guest_gsi;
5054 irq_map->desc = desc;
5057 * Order the above two stores before the next to serialize with
5058 * the KVM real mode handler.
5061 irq_map->r_hwirq = desc->irq_data.hwirq;
5063 if (i == pimap->n_mapped)
5067 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5069 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5071 irq_map->r_hwirq = 0;
5073 mutex_unlock(&kvm->lock);
5078 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5080 struct irq_desc *desc;
5081 struct kvmppc_passthru_irqmap *pimap;
5084 if (!kvm_irq_bypass)
5087 desc = irq_to_desc(host_irq);
5091 mutex_lock(&kvm->lock);
5092 if (!kvm->arch.pimap)
5095 pimap = kvm->arch.pimap;
5097 for (i = 0; i < pimap->n_mapped; i++) {
5098 if (guest_gsi == pimap->mapped[i].v_hwirq)
5102 if (i == pimap->n_mapped) {
5103 mutex_unlock(&kvm->lock);
5108 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5110 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5112 /* invalidate the entry (what do do on error from the above ?) */
5113 pimap->mapped[i].r_hwirq = 0;
5116 * We don't free this structure even when the count goes to
5117 * zero. The structure is freed when we destroy the VM.
5120 mutex_unlock(&kvm->lock);
5124 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5125 struct irq_bypass_producer *prod)
5128 struct kvm_kernel_irqfd *irqfd =
5129 container_of(cons, struct kvm_kernel_irqfd, consumer);
5131 irqfd->producer = prod;
5133 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5135 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5136 prod->irq, irqfd->gsi, ret);
5141 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5142 struct irq_bypass_producer *prod)
5145 struct kvm_kernel_irqfd *irqfd =
5146 container_of(cons, struct kvm_kernel_irqfd, consumer);
5148 irqfd->producer = NULL;
5151 * When producer of consumer is unregistered, we change back to
5152 * default external interrupt handling mode - KVM real mode
5153 * will switch back to host.
5155 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5157 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5158 prod->irq, irqfd->gsi, ret);
5162 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5163 unsigned int ioctl, unsigned long arg)
5165 struct kvm *kvm __maybe_unused = filp->private_data;
5166 void __user *argp = (void __user *)arg;
5171 case KVM_PPC_ALLOCATE_HTAB: {
5175 if (get_user(htab_order, (u32 __user *)argp))
5177 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5184 case KVM_PPC_GET_HTAB_FD: {
5185 struct kvm_get_htab_fd ghf;
5188 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5190 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5194 case KVM_PPC_RESIZE_HPT_PREPARE: {
5195 struct kvm_ppc_resize_hpt rhpt;
5198 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5201 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5205 case KVM_PPC_RESIZE_HPT_COMMIT: {
5206 struct kvm_ppc_resize_hpt rhpt;
5209 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5212 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5224 * List of hcall numbers to enable by default.
5225 * For compatibility with old userspace, we enable by default
5226 * all hcalls that were implemented before the hcall-enabling
5227 * facility was added. Note this list should not include H_RTAS.
5229 static unsigned int default_hcall_list[] = {
5243 #ifdef CONFIG_KVM_XICS
5254 static void init_default_hcalls(void)
5259 for (i = 0; default_hcall_list[i]; ++i) {
5260 hcall = default_hcall_list[i];
5261 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5262 __set_bit(hcall / 4, default_enabled_hcalls);
5266 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5272 /* If not on a POWER9, reject it */
5273 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5276 /* If any unknown flags set, reject it */
5277 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5280 /* GR (guest radix) bit in process_table field must match */
5281 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5282 if (!!(cfg->process_table & PATB_GR) != radix)
5285 /* Process table size field must be reasonable, i.e. <= 24 */
5286 if ((cfg->process_table & PRTS_MASK) > 24)
5289 /* We can change a guest to/from radix now, if the host is radix */
5290 if (radix && !radix_enabled())
5293 /* If we're a nested hypervisor, we currently only support radix */
5294 if (kvmhv_on_pseries() && !radix)
5297 mutex_lock(&kvm->arch.mmu_setup_lock);
5298 if (radix != kvm_is_radix(kvm)) {
5299 if (kvm->arch.mmu_ready) {
5300 kvm->arch.mmu_ready = 0;
5301 /* order mmu_ready vs. vcpus_running */
5303 if (atomic_read(&kvm->arch.vcpus_running)) {
5304 kvm->arch.mmu_ready = 1;
5310 err = kvmppc_switch_mmu_to_radix(kvm);
5312 err = kvmppc_switch_mmu_to_hpt(kvm);
5317 kvm->arch.process_table = cfg->process_table;
5318 kvmppc_setup_partition_table(kvm);
5320 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5321 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5325 mutex_unlock(&kvm->arch.mmu_setup_lock);
5329 static int kvmhv_enable_nested(struct kvm *kvm)
5333 if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5336 /* kvm == NULL means the caller is testing if the capability exists */
5338 kvm->arch.nested_enable = true;
5342 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5347 if (kvmhv_vcpu_is_radix(vcpu)) {
5348 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5354 /* For now quadrants are the only way to access nested guest memory */
5355 if (rc && vcpu->arch.nested)
5361 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5366 if (kvmhv_vcpu_is_radix(vcpu)) {
5367 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5373 /* For now quadrants are the only way to access nested guest memory */
5374 if (rc && vcpu->arch.nested)
5380 static struct kvmppc_ops kvm_ops_hv = {
5381 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5382 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5383 .get_one_reg = kvmppc_get_one_reg_hv,
5384 .set_one_reg = kvmppc_set_one_reg_hv,
5385 .vcpu_load = kvmppc_core_vcpu_load_hv,
5386 .vcpu_put = kvmppc_core_vcpu_put_hv,
5387 .inject_interrupt = kvmppc_inject_interrupt_hv,
5388 .set_msr = kvmppc_set_msr_hv,
5389 .vcpu_run = kvmppc_vcpu_run_hv,
5390 .vcpu_create = kvmppc_core_vcpu_create_hv,
5391 .vcpu_free = kvmppc_core_vcpu_free_hv,
5392 .check_requests = kvmppc_core_check_requests_hv,
5393 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
5394 .flush_memslot = kvmppc_core_flush_memslot_hv,
5395 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5396 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
5397 .unmap_hva_range = kvm_unmap_hva_range_hv,
5398 .age_hva = kvm_age_hva_hv,
5399 .test_age_hva = kvm_test_age_hva_hv,
5400 .set_spte_hva = kvm_set_spte_hva_hv,
5401 .mmu_destroy = kvmppc_mmu_destroy_hv,
5402 .free_memslot = kvmppc_core_free_memslot_hv,
5403 .create_memslot = kvmppc_core_create_memslot_hv,
5404 .init_vm = kvmppc_core_init_vm_hv,
5405 .destroy_vm = kvmppc_core_destroy_vm_hv,
5406 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5407 .emulate_op = kvmppc_core_emulate_op_hv,
5408 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5409 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5410 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5411 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
5412 .hcall_implemented = kvmppc_hcall_impl_hv,
5413 #ifdef CONFIG_KVM_XICS
5414 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5415 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5417 .configure_mmu = kvmhv_configure_mmu,
5418 .get_rmmu_info = kvmhv_get_rmmu_info,
5419 .set_smt_mode = kvmhv_set_smt_mode,
5420 .enable_nested = kvmhv_enable_nested,
5421 .load_from_eaddr = kvmhv_load_from_eaddr,
5422 .store_to_eaddr = kvmhv_store_to_eaddr,
5425 static int kvm_init_subcore_bitmap(void)
5428 int nr_cores = cpu_nr_cores();
5429 struct sibling_subcore_state *sibling_subcore_state;
5431 for (i = 0; i < nr_cores; i++) {
5432 int first_cpu = i * threads_per_core;
5433 int node = cpu_to_node(first_cpu);
5435 /* Ignore if it is already allocated. */
5436 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5439 sibling_subcore_state =
5440 kzalloc_node(sizeof(struct sibling_subcore_state),
5442 if (!sibling_subcore_state)
5446 for (j = 0; j < threads_per_core; j++) {
5447 int cpu = first_cpu + j;
5449 paca_ptrs[cpu]->sibling_subcore_state =
5450 sibling_subcore_state;
5456 static int kvmppc_radix_possible(void)
5458 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5461 static int kvmppc_book3s_init_hv(void)
5465 if (!tlbie_capable) {
5466 pr_err("KVM-HV: Host does not support TLBIE\n");
5471 * FIXME!! Do we need to check on all cpus ?
5473 r = kvmppc_core_check_processor_compat_hv();
5477 r = kvmhv_nested_init();
5481 r = kvm_init_subcore_bitmap();
5486 * We need a way of accessing the XICS interrupt controller,
5487 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5488 * indirectly, via OPAL.
5491 if (!xics_on_xive() && !kvmhv_on_pseries() &&
5492 !local_paca->kvm_hstate.xics_phys) {
5493 struct device_node *np;
5495 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5497 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5500 /* presence of intc confirmed - node can be dropped again */
5505 kvm_ops_hv.owner = THIS_MODULE;
5506 kvmppc_hv_ops = &kvm_ops_hv;
5508 init_default_hcalls();
5512 r = kvmppc_mmu_hv_init();
5516 if (kvmppc_radix_possible())
5517 r = kvmppc_radix_init();
5520 * POWER9 chips before version 2.02 can't have some threads in
5521 * HPT mode and some in radix mode on the same core.
5523 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5524 unsigned int pvr = mfspr(SPRN_PVR);
5525 if ((pvr >> 16) == PVR_POWER9 &&
5526 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5527 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5528 no_mixing_hpt_and_radix = true;
5534 static void kvmppc_book3s_exit_hv(void)
5536 kvmppc_free_host_rm_ops();
5537 if (kvmppc_radix_possible())
5538 kvmppc_radix_exit();
5539 kvmppc_hv_ops = NULL;
5540 kvmhv_nested_exit();
5543 module_init(kvmppc_book3s_init_hv);
5544 module_exit(kvmppc_book3s_exit_hv);
5545 MODULE_LICENSE("GPL");
5546 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5547 MODULE_ALIAS("devname:kvm");