~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/powerpc/kvm/book3s_hv.c

Version: ~ [ linux-5.7 ] ~ [ linux-5.6.15 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.43 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.125 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.182 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.225 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.225 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.84 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
  3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
  4  *
  5  * Authors:
  6  *    Paul Mackerras <paulus@au1.ibm.com>
  7  *    Alexander Graf <agraf@suse.de>
  8  *    Kevin Wolf <mail@kevin-wolf.de>
  9  *
 10  * Description: KVM functions specific to running on Book 3S
 11  * processors in hypervisor mode (specifically POWER7 and later).
 12  *
 13  * This file is derived from arch/powerpc/kvm/book3s.c,
 14  * by Alexander Graf <agraf@suse.de>.
 15  *
 16  * This program is free software; you can redistribute it and/or modify
 17  * it under the terms of the GNU General Public License, version 2, as
 18  * published by the Free Software Foundation.
 19  */
 20 
 21 #include <linux/kvm_host.h>
 22 #include <linux/kernel.h>
 23 #include <linux/err.h>
 24 #include <linux/slab.h>
 25 #include <linux/preempt.h>
 26 #include <linux/sched/signal.h>
 27 #include <linux/sched/stat.h>
 28 #include <linux/delay.h>
 29 #include <linux/export.h>
 30 #include <linux/fs.h>
 31 #include <linux/anon_inodes.h>
 32 #include <linux/cpu.h>
 33 #include <linux/cpumask.h>
 34 #include <linux/spinlock.h>
 35 #include <linux/page-flags.h>
 36 #include <linux/srcu.h>
 37 #include <linux/miscdevice.h>
 38 #include <linux/debugfs.h>
 39 #include <linux/gfp.h>
 40 #include <linux/vmalloc.h>
 41 #include <linux/highmem.h>
 42 #include <linux/hugetlb.h>
 43 #include <linux/kvm_irqfd.h>
 44 #include <linux/irqbypass.h>
 45 #include <linux/module.h>
 46 #include <linux/compiler.h>
 47 #include <linux/of.h>
 48 
 49 #include <asm/ftrace.h>
 50 #include <asm/reg.h>
 51 #include <asm/ppc-opcode.h>
 52 #include <asm/asm-prototypes.h>
 53 #include <asm/archrandom.h>
 54 #include <asm/debug.h>
 55 #include <asm/disassemble.h>
 56 #include <asm/cputable.h>
 57 #include <asm/cacheflush.h>
 58 #include <linux/uaccess.h>
 59 #include <asm/io.h>
 60 #include <asm/kvm_ppc.h>
 61 #include <asm/kvm_book3s.h>
 62 #include <asm/mmu_context.h>
 63 #include <asm/lppaca.h>
 64 #include <asm/processor.h>
 65 #include <asm/cputhreads.h>
 66 #include <asm/page.h>
 67 #include <asm/hvcall.h>
 68 #include <asm/switch_to.h>
 69 #include <asm/smp.h>
 70 #include <asm/dbell.h>
 71 #include <asm/hmi.h>
 72 #include <asm/pnv-pci.h>
 73 #include <asm/mmu.h>
 74 #include <asm/opal.h>
 75 #include <asm/xics.h>
 76 #include <asm/xive.h>
 77 
 78 #include "book3s.h"
 79 
 80 #define CREATE_TRACE_POINTS
 81 #include "trace_hv.h"
 82 
 83 /* #define EXIT_DEBUG */
 84 /* #define EXIT_DEBUG_SIMPLE */
 85 /* #define EXIT_DEBUG_INT */
 86 
 87 /* Used to indicate that a guest page fault needs to be handled */
 88 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
 89 /* Used to indicate that a guest passthrough interrupt needs to be handled */
 90 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
 91 
 92 /* Used as a "null" value for timebase values */
 93 #define TB_NIL  (~(u64)0)
 94 
 95 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
 96 
 97 static int dynamic_mt_modes = 6;
 98 module_param(dynamic_mt_modes, int, 0644);
 99 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
100 static int target_smt_mode;
101 module_param(target_smt_mode, int, 0644);
102 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
103 
104 static bool indep_threads_mode = true;
105 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
106 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
107 
108 static bool one_vm_per_core;
109 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
110 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
111 
112 #ifdef CONFIG_KVM_XICS
113 static struct kernel_param_ops module_param_ops = {
114         .set = param_set_int,
115         .get = param_get_int,
116 };
117 
118 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
119 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
120 
121 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
122 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
123 #endif
124 
125 /* If set, guests are allowed to create and control nested guests */
126 static bool nested = true;
127 module_param(nested, bool, S_IRUGO | S_IWUSR);
128 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
129 
130 static inline bool nesting_enabled(struct kvm *kvm)
131 {
132         return kvm->arch.nested_enable && kvm_is_radix(kvm);
133 }
134 
135 /* If set, the threads on each CPU core have to be in the same MMU mode */
136 static bool no_mixing_hpt_and_radix;
137 
138 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
139 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
140 
141 /*
142  * RWMR values for POWER8.  These control the rate at which PURR
143  * and SPURR count and should be set according to the number of
144  * online threads in the vcore being run.
145  */
146 #define RWMR_RPA_P8_1THREAD     0x164520C62609AECAUL
147 #define RWMR_RPA_P8_2THREAD     0x7FFF2908450D8DA9UL
148 #define RWMR_RPA_P8_3THREAD     0x164520C62609AECAUL
149 #define RWMR_RPA_P8_4THREAD     0x199A421245058DA9UL
150 #define RWMR_RPA_P8_5THREAD     0x164520C62609AECAUL
151 #define RWMR_RPA_P8_6THREAD     0x164520C62609AECAUL
152 #define RWMR_RPA_P8_7THREAD     0x164520C62609AECAUL
153 #define RWMR_RPA_P8_8THREAD     0x164520C62609AECAUL
154 
155 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
156         RWMR_RPA_P8_1THREAD,
157         RWMR_RPA_P8_1THREAD,
158         RWMR_RPA_P8_2THREAD,
159         RWMR_RPA_P8_3THREAD,
160         RWMR_RPA_P8_4THREAD,
161         RWMR_RPA_P8_5THREAD,
162         RWMR_RPA_P8_6THREAD,
163         RWMR_RPA_P8_7THREAD,
164         RWMR_RPA_P8_8THREAD,
165 };
166 
167 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
168                 int *ip)
169 {
170         int i = *ip;
171         struct kvm_vcpu *vcpu;
172 
173         while (++i < MAX_SMT_THREADS) {
174                 vcpu = READ_ONCE(vc->runnable_threads[i]);
175                 if (vcpu) {
176                         *ip = i;
177                         return vcpu;
178                 }
179         }
180         return NULL;
181 }
182 
183 /* Used to traverse the list of runnable threads for a given vcore */
184 #define for_each_runnable_thread(i, vcpu, vc) \
185         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
186 
187 static bool kvmppc_ipi_thread(int cpu)
188 {
189         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
190 
191         /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
192         if (kvmhv_on_pseries())
193                 return false;
194 
195         /* On POWER9 we can use msgsnd to IPI any cpu */
196         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
197                 msg |= get_hard_smp_processor_id(cpu);
198                 smp_mb();
199                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
200                 return true;
201         }
202 
203         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
204         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
205                 preempt_disable();
206                 if (cpu_first_thread_sibling(cpu) ==
207                     cpu_first_thread_sibling(smp_processor_id())) {
208                         msg |= cpu_thread_in_core(cpu);
209                         smp_mb();
210                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
211                         preempt_enable();
212                         return true;
213                 }
214                 preempt_enable();
215         }
216 
217 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
218         if (cpu >= 0 && cpu < nr_cpu_ids) {
219                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
220                         xics_wake_cpu(cpu);
221                         return true;
222                 }
223                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
224                 return true;
225         }
226 #endif
227 
228         return false;
229 }
230 
231 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
232 {
233         int cpu;
234         struct swait_queue_head *wqp;
235 
236         wqp = kvm_arch_vcpu_wq(vcpu);
237         if (swq_has_sleeper(wqp)) {
238                 swake_up_one(wqp);
239                 ++vcpu->stat.halt_wakeup;
240         }
241 
242         cpu = READ_ONCE(vcpu->arch.thread_cpu);
243         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
244                 return;
245 
246         /* CPU points to the first thread of the core */
247         cpu = vcpu->cpu;
248         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
249                 smp_send_reschedule(cpu);
250 }
251 
252 /*
253  * We use the vcpu_load/put functions to measure stolen time.
254  * Stolen time is counted as time when either the vcpu is able to
255  * run as part of a virtual core, but the task running the vcore
256  * is preempted or sleeping, or when the vcpu needs something done
257  * in the kernel by the task running the vcpu, but that task is
258  * preempted or sleeping.  Those two things have to be counted
259  * separately, since one of the vcpu tasks will take on the job
260  * of running the core, and the other vcpu tasks in the vcore will
261  * sleep waiting for it to do that, but that sleep shouldn't count
262  * as stolen time.
263  *
264  * Hence we accumulate stolen time when the vcpu can run as part of
265  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
266  * needs its task to do other things in the kernel (for example,
267  * service a page fault) in busy_stolen.  We don't accumulate
268  * stolen time for a vcore when it is inactive, or for a vcpu
269  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
270  * a misnomer; it means that the vcpu task is not executing in
271  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
272  * the kernel.  We don't have any way of dividing up that time
273  * between time that the vcpu is genuinely stopped, time that
274  * the task is actively working on behalf of the vcpu, and time
275  * that the task is preempted, so we don't count any of it as
276  * stolen.
277  *
278  * Updates to busy_stolen are protected by arch.tbacct_lock;
279  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
280  * lock.  The stolen times are measured in units of timebase ticks.
281  * (Note that the != TB_NIL checks below are purely defensive;
282  * they should never fail.)
283  */
284 
285 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
286 {
287         unsigned long flags;
288 
289         spin_lock_irqsave(&vc->stoltb_lock, flags);
290         vc->preempt_tb = mftb();
291         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
292 }
293 
294 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
295 {
296         unsigned long flags;
297 
298         spin_lock_irqsave(&vc->stoltb_lock, flags);
299         if (vc->preempt_tb != TB_NIL) {
300                 vc->stolen_tb += mftb() - vc->preempt_tb;
301                 vc->preempt_tb = TB_NIL;
302         }
303         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
304 }
305 
306 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
307 {
308         struct kvmppc_vcore *vc = vcpu->arch.vcore;
309         unsigned long flags;
310 
311         /*
312          * We can test vc->runner without taking the vcore lock,
313          * because only this task ever sets vc->runner to this
314          * vcpu, and once it is set to this vcpu, only this task
315          * ever sets it to NULL.
316          */
317         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
318                 kvmppc_core_end_stolen(vc);
319 
320         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
321         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
322             vcpu->arch.busy_preempt != TB_NIL) {
323                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
324                 vcpu->arch.busy_preempt = TB_NIL;
325         }
326         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
327 }
328 
329 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
330 {
331         struct kvmppc_vcore *vc = vcpu->arch.vcore;
332         unsigned long flags;
333 
334         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
335                 kvmppc_core_start_stolen(vc);
336 
337         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
338         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
339                 vcpu->arch.busy_preempt = mftb();
340         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
341 }
342 
343 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
344 {
345         /*
346          * Check for illegal transactional state bit combination
347          * and if we find it, force the TS field to a safe state.
348          */
349         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
350                 msr &= ~MSR_TS_MASK;
351         vcpu->arch.shregs.msr = msr;
352         kvmppc_end_cede(vcpu);
353 }
354 
355 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
356 {
357         vcpu->arch.pvr = pvr;
358 }
359 
360 /* Dummy value used in computing PCR value below */
361 #define PCR_ARCH_300    (PCR_ARCH_207 << 1)
362 
363 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
364 {
365         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
366         struct kvmppc_vcore *vc = vcpu->arch.vcore;
367 
368         /* We can (emulate) our own architecture version and anything older */
369         if (cpu_has_feature(CPU_FTR_ARCH_300))
370                 host_pcr_bit = PCR_ARCH_300;
371         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
372                 host_pcr_bit = PCR_ARCH_207;
373         else if (cpu_has_feature(CPU_FTR_ARCH_206))
374                 host_pcr_bit = PCR_ARCH_206;
375         else
376                 host_pcr_bit = PCR_ARCH_205;
377 
378         /* Determine lowest PCR bit needed to run guest in given PVR level */
379         guest_pcr_bit = host_pcr_bit;
380         if (arch_compat) {
381                 switch (arch_compat) {
382                 case PVR_ARCH_205:
383                         guest_pcr_bit = PCR_ARCH_205;
384                         break;
385                 case PVR_ARCH_206:
386                 case PVR_ARCH_206p:
387                         guest_pcr_bit = PCR_ARCH_206;
388                         break;
389                 case PVR_ARCH_207:
390                         guest_pcr_bit = PCR_ARCH_207;
391                         break;
392                 case PVR_ARCH_300:
393                         guest_pcr_bit = PCR_ARCH_300;
394                         break;
395                 default:
396                         return -EINVAL;
397                 }
398         }
399 
400         /* Check requested PCR bits don't exceed our capabilities */
401         if (guest_pcr_bit > host_pcr_bit)
402                 return -EINVAL;
403 
404         spin_lock(&vc->lock);
405         vc->arch_compat = arch_compat;
406         /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
407         vc->pcr = host_pcr_bit - guest_pcr_bit;
408         spin_unlock(&vc->lock);
409 
410         return 0;
411 }
412 
413 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
414 {
415         int r;
416 
417         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
418         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
419                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
420         for (r = 0; r < 16; ++r)
421                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
422                        r, kvmppc_get_gpr(vcpu, r),
423                        r+16, kvmppc_get_gpr(vcpu, r+16));
424         pr_err("ctr = %.16lx  lr  = %.16lx\n",
425                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
426         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
427                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
428         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
429                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
430         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
431                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
432         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
433                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
434         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
435         pr_err("fault dar = %.16lx dsisr = %.8x\n",
436                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
437         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
438         for (r = 0; r < vcpu->arch.slb_max; ++r)
439                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
440                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
441         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
442                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
443                vcpu->arch.last_inst);
444 }
445 
446 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
447 {
448         struct kvm_vcpu *ret;
449 
450         mutex_lock(&kvm->lock);
451         ret = kvm_get_vcpu_by_id(kvm, id);
452         mutex_unlock(&kvm->lock);
453         return ret;
454 }
455 
456 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
457 {
458         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
459         vpa->yield_count = cpu_to_be32(1);
460 }
461 
462 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
463                    unsigned long addr, unsigned long len)
464 {
465         /* check address is cacheline aligned */
466         if (addr & (L1_CACHE_BYTES - 1))
467                 return -EINVAL;
468         spin_lock(&vcpu->arch.vpa_update_lock);
469         if (v->next_gpa != addr || v->len != len) {
470                 v->next_gpa = addr;
471                 v->len = addr ? len : 0;
472                 v->update_pending = 1;
473         }
474         spin_unlock(&vcpu->arch.vpa_update_lock);
475         return 0;
476 }
477 
478 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
479 struct reg_vpa {
480         u32 dummy;
481         union {
482                 __be16 hword;
483                 __be32 word;
484         } length;
485 };
486 
487 static int vpa_is_registered(struct kvmppc_vpa *vpap)
488 {
489         if (vpap->update_pending)
490                 return vpap->next_gpa != 0;
491         return vpap->pinned_addr != NULL;
492 }
493 
494 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
495                                        unsigned long flags,
496                                        unsigned long vcpuid, unsigned long vpa)
497 {
498         struct kvm *kvm = vcpu->kvm;
499         unsigned long len, nb;
500         void *va;
501         struct kvm_vcpu *tvcpu;
502         int err;
503         int subfunc;
504         struct kvmppc_vpa *vpap;
505 
506         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
507         if (!tvcpu)
508                 return H_PARAMETER;
509 
510         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
511         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
512             subfunc == H_VPA_REG_SLB) {
513                 /* Registering new area - address must be cache-line aligned */
514                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
515                         return H_PARAMETER;
516 
517                 /* convert logical addr to kernel addr and read length */
518                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
519                 if (va == NULL)
520                         return H_PARAMETER;
521                 if (subfunc == H_VPA_REG_VPA)
522                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
523                 else
524                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
525                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
526 
527                 /* Check length */
528                 if (len > nb || len < sizeof(struct reg_vpa))
529                         return H_PARAMETER;
530         } else {
531                 vpa = 0;
532                 len = 0;
533         }
534 
535         err = H_PARAMETER;
536         vpap = NULL;
537         spin_lock(&tvcpu->arch.vpa_update_lock);
538 
539         switch (subfunc) {
540         case H_VPA_REG_VPA:             /* register VPA */
541                 /*
542                  * The size of our lppaca is 1kB because of the way we align
543                  * it for the guest to avoid crossing a 4kB boundary. We only
544                  * use 640 bytes of the structure though, so we should accept
545                  * clients that set a size of 640.
546                  */
547                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
548                 if (len < sizeof(struct lppaca))
549                         break;
550                 vpap = &tvcpu->arch.vpa;
551                 err = 0;
552                 break;
553 
554         case H_VPA_REG_DTL:             /* register DTL */
555                 if (len < sizeof(struct dtl_entry))
556                         break;
557                 len -= len % sizeof(struct dtl_entry);
558 
559                 /* Check that they have previously registered a VPA */
560                 err = H_RESOURCE;
561                 if (!vpa_is_registered(&tvcpu->arch.vpa))
562                         break;
563 
564                 vpap = &tvcpu->arch.dtl;
565                 err = 0;
566                 break;
567 
568         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
569                 /* Check that they have previously registered a VPA */
570                 err = H_RESOURCE;
571                 if (!vpa_is_registered(&tvcpu->arch.vpa))
572                         break;
573 
574                 vpap = &tvcpu->arch.slb_shadow;
575                 err = 0;
576                 break;
577 
578         case H_VPA_DEREG_VPA:           /* deregister VPA */
579                 /* Check they don't still have a DTL or SLB buf registered */
580                 err = H_RESOURCE;
581                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
582                     vpa_is_registered(&tvcpu->arch.slb_shadow))
583                         break;
584 
585                 vpap = &tvcpu->arch.vpa;
586                 err = 0;
587                 break;
588 
589         case H_VPA_DEREG_DTL:           /* deregister DTL */
590                 vpap = &tvcpu->arch.dtl;
591                 err = 0;
592                 break;
593 
594         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
595                 vpap = &tvcpu->arch.slb_shadow;
596                 err = 0;
597                 break;
598         }
599 
600         if (vpap) {
601                 vpap->next_gpa = vpa;
602                 vpap->len = len;
603                 vpap->update_pending = 1;
604         }
605 
606         spin_unlock(&tvcpu->arch.vpa_update_lock);
607 
608         return err;
609 }
610 
611 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
612 {
613         struct kvm *kvm = vcpu->kvm;
614         void *va;
615         unsigned long nb;
616         unsigned long gpa;
617 
618         /*
619          * We need to pin the page pointed to by vpap->next_gpa,
620          * but we can't call kvmppc_pin_guest_page under the lock
621          * as it does get_user_pages() and down_read().  So we
622          * have to drop the lock, pin the page, then get the lock
623          * again and check that a new area didn't get registered
624          * in the meantime.
625          */
626         for (;;) {
627                 gpa = vpap->next_gpa;
628                 spin_unlock(&vcpu->arch.vpa_update_lock);
629                 va = NULL;
630                 nb = 0;
631                 if (gpa)
632                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
633                 spin_lock(&vcpu->arch.vpa_update_lock);
634                 if (gpa == vpap->next_gpa)
635                         break;
636                 /* sigh... unpin that one and try again */
637                 if (va)
638                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
639         }
640 
641         vpap->update_pending = 0;
642         if (va && nb < vpap->len) {
643                 /*
644                  * If it's now too short, it must be that userspace
645                  * has changed the mappings underlying guest memory,
646                  * so unregister the region.
647                  */
648                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
649                 va = NULL;
650         }
651         if (vpap->pinned_addr)
652                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
653                                         vpap->dirty);
654         vpap->gpa = gpa;
655         vpap->pinned_addr = va;
656         vpap->dirty = false;
657         if (va)
658                 vpap->pinned_end = va + vpap->len;
659 }
660 
661 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
662 {
663         if (!(vcpu->arch.vpa.update_pending ||
664               vcpu->arch.slb_shadow.update_pending ||
665               vcpu->arch.dtl.update_pending))
666                 return;
667 
668         spin_lock(&vcpu->arch.vpa_update_lock);
669         if (vcpu->arch.vpa.update_pending) {
670                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
671                 if (vcpu->arch.vpa.pinned_addr)
672                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
673         }
674         if (vcpu->arch.dtl.update_pending) {
675                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
676                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
677                 vcpu->arch.dtl_index = 0;
678         }
679         if (vcpu->arch.slb_shadow.update_pending)
680                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
681         spin_unlock(&vcpu->arch.vpa_update_lock);
682 }
683 
684 /*
685  * Return the accumulated stolen time for the vcore up until `now'.
686  * The caller should hold the vcore lock.
687  */
688 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
689 {
690         u64 p;
691         unsigned long flags;
692 
693         spin_lock_irqsave(&vc->stoltb_lock, flags);
694         p = vc->stolen_tb;
695         if (vc->vcore_state != VCORE_INACTIVE &&
696             vc->preempt_tb != TB_NIL)
697                 p += now - vc->preempt_tb;
698         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
699         return p;
700 }
701 
702 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
703                                     struct kvmppc_vcore *vc)
704 {
705         struct dtl_entry *dt;
706         struct lppaca *vpa;
707         unsigned long stolen;
708         unsigned long core_stolen;
709         u64 now;
710         unsigned long flags;
711 
712         dt = vcpu->arch.dtl_ptr;
713         vpa = vcpu->arch.vpa.pinned_addr;
714         now = mftb();
715         core_stolen = vcore_stolen_time(vc, now);
716         stolen = core_stolen - vcpu->arch.stolen_logged;
717         vcpu->arch.stolen_logged = core_stolen;
718         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
719         stolen += vcpu->arch.busy_stolen;
720         vcpu->arch.busy_stolen = 0;
721         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
722         if (!dt || !vpa)
723                 return;
724         memset(dt, 0, sizeof(struct dtl_entry));
725         dt->dispatch_reason = 7;
726         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
727         dt->timebase = cpu_to_be64(now + vc->tb_offset);
728         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
729         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
730         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
731         ++dt;
732         if (dt == vcpu->arch.dtl.pinned_end)
733                 dt = vcpu->arch.dtl.pinned_addr;
734         vcpu->arch.dtl_ptr = dt;
735         /* order writing *dt vs. writing vpa->dtl_idx */
736         smp_wmb();
737         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
738         vcpu->arch.dtl.dirty = true;
739 }
740 
741 /* See if there is a doorbell interrupt pending for a vcpu */
742 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
743 {
744         int thr;
745         struct kvmppc_vcore *vc;
746 
747         if (vcpu->arch.doorbell_request)
748                 return true;
749         /*
750          * Ensure that the read of vcore->dpdes comes after the read
751          * of vcpu->doorbell_request.  This barrier matches the
752          * smb_wmb() in kvmppc_guest_entry_inject().
753          */
754         smp_rmb();
755         vc = vcpu->arch.vcore;
756         thr = vcpu->vcpu_id - vc->first_vcpuid;
757         return !!(vc->dpdes & (1 << thr));
758 }
759 
760 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
761 {
762         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
763                 return true;
764         if ((!vcpu->arch.vcore->arch_compat) &&
765             cpu_has_feature(CPU_FTR_ARCH_207S))
766                 return true;
767         return false;
768 }
769 
770 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
771                              unsigned long resource, unsigned long value1,
772                              unsigned long value2)
773 {
774         switch (resource) {
775         case H_SET_MODE_RESOURCE_SET_CIABR:
776                 if (!kvmppc_power8_compatible(vcpu))
777                         return H_P2;
778                 if (value2)
779                         return H_P4;
780                 if (mflags)
781                         return H_UNSUPPORTED_FLAG_START;
782                 /* Guests can't breakpoint the hypervisor */
783                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
784                         return H_P3;
785                 vcpu->arch.ciabr  = value1;
786                 return H_SUCCESS;
787         case H_SET_MODE_RESOURCE_SET_DAWR:
788                 if (!kvmppc_power8_compatible(vcpu))
789                         return H_P2;
790                 if (!ppc_breakpoint_available())
791                         return H_P2;
792                 if (mflags)
793                         return H_UNSUPPORTED_FLAG_START;
794                 if (value2 & DABRX_HYP)
795                         return H_P4;
796                 vcpu->arch.dawr  = value1;
797                 vcpu->arch.dawrx = value2;
798                 return H_SUCCESS;
799         default:
800                 return H_TOO_HARD;
801         }
802 }
803 
804 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
805 {
806         struct kvmppc_vcore *vcore = target->arch.vcore;
807 
808         /*
809          * We expect to have been called by the real mode handler
810          * (kvmppc_rm_h_confer()) which would have directly returned
811          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
812          * have useful work to do and should not confer) so we don't
813          * recheck that here.
814          */
815 
816         spin_lock(&vcore->lock);
817         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
818             vcore->vcore_state != VCORE_INACTIVE &&
819             vcore->runner)
820                 target = vcore->runner;
821         spin_unlock(&vcore->lock);
822 
823         return kvm_vcpu_yield_to(target);
824 }
825 
826 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
827 {
828         int yield_count = 0;
829         struct lppaca *lppaca;
830 
831         spin_lock(&vcpu->arch.vpa_update_lock);
832         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
833         if (lppaca)
834                 yield_count = be32_to_cpu(lppaca->yield_count);
835         spin_unlock(&vcpu->arch.vpa_update_lock);
836         return yield_count;
837 }
838 
839 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
840 {
841         unsigned long req = kvmppc_get_gpr(vcpu, 3);
842         unsigned long target, ret = H_SUCCESS;
843         int yield_count;
844         struct kvm_vcpu *tvcpu;
845         int idx, rc;
846 
847         if (req <= MAX_HCALL_OPCODE &&
848             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
849                 return RESUME_HOST;
850 
851         switch (req) {
852         case H_CEDE:
853                 break;
854         case H_PROD:
855                 target = kvmppc_get_gpr(vcpu, 4);
856                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
857                 if (!tvcpu) {
858                         ret = H_PARAMETER;
859                         break;
860                 }
861                 tvcpu->arch.prodded = 1;
862                 smp_mb();
863                 if (tvcpu->arch.ceded)
864                         kvmppc_fast_vcpu_kick_hv(tvcpu);
865                 break;
866         case H_CONFER:
867                 target = kvmppc_get_gpr(vcpu, 4);
868                 if (target == -1)
869                         break;
870                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
871                 if (!tvcpu) {
872                         ret = H_PARAMETER;
873                         break;
874                 }
875                 yield_count = kvmppc_get_gpr(vcpu, 5);
876                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
877                         break;
878                 kvm_arch_vcpu_yield_to(tvcpu);
879                 break;
880         case H_REGISTER_VPA:
881                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
882                                         kvmppc_get_gpr(vcpu, 5),
883                                         kvmppc_get_gpr(vcpu, 6));
884                 break;
885         case H_RTAS:
886                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
887                         return RESUME_HOST;
888 
889                 idx = srcu_read_lock(&vcpu->kvm->srcu);
890                 rc = kvmppc_rtas_hcall(vcpu);
891                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
892 
893                 if (rc == -ENOENT)
894                         return RESUME_HOST;
895                 else if (rc == 0)
896                         break;
897 
898                 /* Send the error out to userspace via KVM_RUN */
899                 return rc;
900         case H_LOGICAL_CI_LOAD:
901                 ret = kvmppc_h_logical_ci_load(vcpu);
902                 if (ret == H_TOO_HARD)
903                         return RESUME_HOST;
904                 break;
905         case H_LOGICAL_CI_STORE:
906                 ret = kvmppc_h_logical_ci_store(vcpu);
907                 if (ret == H_TOO_HARD)
908                         return RESUME_HOST;
909                 break;
910         case H_SET_MODE:
911                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
912                                         kvmppc_get_gpr(vcpu, 5),
913                                         kvmppc_get_gpr(vcpu, 6),
914                                         kvmppc_get_gpr(vcpu, 7));
915                 if (ret == H_TOO_HARD)
916                         return RESUME_HOST;
917                 break;
918         case H_XIRR:
919         case H_CPPR:
920         case H_EOI:
921         case H_IPI:
922         case H_IPOLL:
923         case H_XIRR_X:
924                 if (kvmppc_xics_enabled(vcpu)) {
925                         if (xive_enabled()) {
926                                 ret = H_NOT_AVAILABLE;
927                                 return RESUME_GUEST;
928                         }
929                         ret = kvmppc_xics_hcall(vcpu, req);
930                         break;
931                 }
932                 return RESUME_HOST;
933         case H_SET_DABR:
934                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
935                 break;
936         case H_SET_XDABR:
937                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
938                                                 kvmppc_get_gpr(vcpu, 5));
939                 break;
940         case H_GET_TCE:
941                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
942                                                 kvmppc_get_gpr(vcpu, 5));
943                 if (ret == H_TOO_HARD)
944                         return RESUME_HOST;
945                 break;
946         case H_PUT_TCE:
947                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
948                                                 kvmppc_get_gpr(vcpu, 5),
949                                                 kvmppc_get_gpr(vcpu, 6));
950                 if (ret == H_TOO_HARD)
951                         return RESUME_HOST;
952                 break;
953         case H_PUT_TCE_INDIRECT:
954                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
955                                                 kvmppc_get_gpr(vcpu, 5),
956                                                 kvmppc_get_gpr(vcpu, 6),
957                                                 kvmppc_get_gpr(vcpu, 7));
958                 if (ret == H_TOO_HARD)
959                         return RESUME_HOST;
960                 break;
961         case H_STUFF_TCE:
962                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
963                                                 kvmppc_get_gpr(vcpu, 5),
964                                                 kvmppc_get_gpr(vcpu, 6),
965                                                 kvmppc_get_gpr(vcpu, 7));
966                 if (ret == H_TOO_HARD)
967                         return RESUME_HOST;
968                 break;
969         case H_RANDOM:
970                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
971                         ret = H_HARDWARE;
972                 break;
973 
974         case H_SET_PARTITION_TABLE:
975                 ret = H_FUNCTION;
976                 if (nesting_enabled(vcpu->kvm))
977                         ret = kvmhv_set_partition_table(vcpu);
978                 break;
979         case H_ENTER_NESTED:
980                 ret = H_FUNCTION;
981                 if (!nesting_enabled(vcpu->kvm))
982                         break;
983                 ret = kvmhv_enter_nested_guest(vcpu);
984                 if (ret == H_INTERRUPT) {
985                         kvmppc_set_gpr(vcpu, 3, 0);
986                         vcpu->arch.hcall_needed = 0;
987                         return -EINTR;
988                 } else if (ret == H_TOO_HARD) {
989                         kvmppc_set_gpr(vcpu, 3, 0);
990                         vcpu->arch.hcall_needed = 0;
991                         return RESUME_HOST;
992                 }
993                 break;
994         case H_TLB_INVALIDATE:
995                 ret = H_FUNCTION;
996                 if (nesting_enabled(vcpu->kvm))
997                         ret = kvmhv_do_nested_tlbie(vcpu);
998                 break;
999         case H_COPY_TOFROM_GUEST:
1000                 ret = H_FUNCTION;
1001                 if (nesting_enabled(vcpu->kvm))
1002                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1003                 break;
1004         default:
1005                 return RESUME_HOST;
1006         }
1007         kvmppc_set_gpr(vcpu, 3, ret);
1008         vcpu->arch.hcall_needed = 0;
1009         return RESUME_GUEST;
1010 }
1011 
1012 /*
1013  * Handle H_CEDE in the nested virtualization case where we haven't
1014  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1015  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1016  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1017  */
1018 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1019 {
1020         vcpu->arch.shregs.msr |= MSR_EE;
1021         vcpu->arch.ceded = 1;
1022         smp_mb();
1023         if (vcpu->arch.prodded) {
1024                 vcpu->arch.prodded = 0;
1025                 smp_mb();
1026                 vcpu->arch.ceded = 0;
1027         }
1028 }
1029 
1030 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1031 {
1032         switch (cmd) {
1033         case H_CEDE:
1034         case H_PROD:
1035         case H_CONFER:
1036         case H_REGISTER_VPA:
1037         case H_SET_MODE:
1038         case H_LOGICAL_CI_LOAD:
1039         case H_LOGICAL_CI_STORE:
1040 #ifdef CONFIG_KVM_XICS
1041         case H_XIRR:
1042         case H_CPPR:
1043         case H_EOI:
1044         case H_IPI:
1045         case H_IPOLL:
1046         case H_XIRR_X:
1047 #endif
1048                 return 1;
1049         }
1050 
1051         /* See if it's in the real-mode table */
1052         return kvmppc_hcall_impl_hv_realmode(cmd);
1053 }
1054 
1055 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1056                                         struct kvm_vcpu *vcpu)
1057 {
1058         u32 last_inst;
1059 
1060         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1061                                         EMULATE_DONE) {
1062                 /*
1063                  * Fetch failed, so return to guest and
1064                  * try executing it again.
1065                  */
1066                 return RESUME_GUEST;
1067         }
1068 
1069         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1070                 run->exit_reason = KVM_EXIT_DEBUG;
1071                 run->debug.arch.address = kvmppc_get_pc(vcpu);
1072                 return RESUME_HOST;
1073         } else {
1074                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1075                 return RESUME_GUEST;
1076         }
1077 }
1078 
1079 static void do_nothing(void *x)
1080 {
1081 }
1082 
1083 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1084 {
1085         int thr, cpu, pcpu, nthreads;
1086         struct kvm_vcpu *v;
1087         unsigned long dpdes;
1088 
1089         nthreads = vcpu->kvm->arch.emul_smt_mode;
1090         dpdes = 0;
1091         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1092         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1093                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1094                 if (!v)
1095                         continue;
1096                 /*
1097                  * If the vcpu is currently running on a physical cpu thread,
1098                  * interrupt it in order to pull it out of the guest briefly,
1099                  * which will update its vcore->dpdes value.
1100                  */
1101                 pcpu = READ_ONCE(v->cpu);
1102                 if (pcpu >= 0)
1103                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1104                 if (kvmppc_doorbell_pending(v))
1105                         dpdes |= 1 << thr;
1106         }
1107         return dpdes;
1108 }
1109 
1110 /*
1111  * On POWER9, emulate doorbell-related instructions in order to
1112  * give the guest the illusion of running on a multi-threaded core.
1113  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1114  * and mfspr DPDES.
1115  */
1116 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1117 {
1118         u32 inst, rb, thr;
1119         unsigned long arg;
1120         struct kvm *kvm = vcpu->kvm;
1121         struct kvm_vcpu *tvcpu;
1122 
1123         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1124                 return RESUME_GUEST;
1125         if (get_op(inst) != 31)
1126                 return EMULATE_FAIL;
1127         rb = get_rb(inst);
1128         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1129         switch (get_xop(inst)) {
1130         case OP_31_XOP_MSGSNDP:
1131                 arg = kvmppc_get_gpr(vcpu, rb);
1132                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1133                         break;
1134                 arg &= 0x3f;
1135                 if (arg >= kvm->arch.emul_smt_mode)
1136                         break;
1137                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1138                 if (!tvcpu)
1139                         break;
1140                 if (!tvcpu->arch.doorbell_request) {
1141                         tvcpu->arch.doorbell_request = 1;
1142                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1143                 }
1144                 break;
1145         case OP_31_XOP_MSGCLRP:
1146                 arg = kvmppc_get_gpr(vcpu, rb);
1147                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1148                         break;
1149                 vcpu->arch.vcore->dpdes = 0;
1150                 vcpu->arch.doorbell_request = 0;
1151                 break;
1152         case OP_31_XOP_MFSPR:
1153                 switch (get_sprn(inst)) {
1154                 case SPRN_TIR:
1155                         arg = thr;
1156                         break;
1157                 case SPRN_DPDES:
1158                         arg = kvmppc_read_dpdes(vcpu);
1159                         break;
1160                 default:
1161                         return EMULATE_FAIL;
1162                 }
1163                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1164                 break;
1165         default:
1166                 return EMULATE_FAIL;
1167         }
1168         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1169         return RESUME_GUEST;
1170 }
1171 
1172 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1173                                  struct task_struct *tsk)
1174 {
1175         int r = RESUME_HOST;
1176 
1177         vcpu->stat.sum_exits++;
1178 
1179         /*
1180          * This can happen if an interrupt occurs in the last stages
1181          * of guest entry or the first stages of guest exit (i.e. after
1182          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1183          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1184          * That can happen due to a bug, or due to a machine check
1185          * occurring at just the wrong time.
1186          */
1187         if (vcpu->arch.shregs.msr & MSR_HV) {
1188                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1189                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1190                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1191                         vcpu->arch.shregs.msr);
1192                 kvmppc_dump_regs(vcpu);
1193                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1194                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1195                 return RESUME_HOST;
1196         }
1197         run->exit_reason = KVM_EXIT_UNKNOWN;
1198         run->ready_for_interrupt_injection = 1;
1199         switch (vcpu->arch.trap) {
1200         /* We're good on these - the host merely wanted to get our attention */
1201         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1202                 vcpu->stat.dec_exits++;
1203                 r = RESUME_GUEST;
1204                 break;
1205         case BOOK3S_INTERRUPT_EXTERNAL:
1206         case BOOK3S_INTERRUPT_H_DOORBELL:
1207         case BOOK3S_INTERRUPT_H_VIRT:
1208                 vcpu->stat.ext_intr_exits++;
1209                 r = RESUME_GUEST;
1210                 break;
1211         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1212         case BOOK3S_INTERRUPT_HMI:
1213         case BOOK3S_INTERRUPT_PERFMON:
1214         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1215                 r = RESUME_GUEST;
1216                 break;
1217         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1218                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1219                 run->exit_reason = KVM_EXIT_NMI;
1220                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1221                 /* Clear out the old NMI status from run->flags */
1222                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1223                 /* Now set the NMI status */
1224                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1225                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1226                 else
1227                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1228 
1229                 r = RESUME_HOST;
1230                 /* Print the MCE event to host console. */
1231                 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1232                 break;
1233         case BOOK3S_INTERRUPT_PROGRAM:
1234         {
1235                 ulong flags;
1236                 /*
1237                  * Normally program interrupts are delivered directly
1238                  * to the guest by the hardware, but we can get here
1239                  * as a result of a hypervisor emulation interrupt
1240                  * (e40) getting turned into a 700 by BML RTAS.
1241                  */
1242                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1243                 kvmppc_core_queue_program(vcpu, flags);
1244                 r = RESUME_GUEST;
1245                 break;
1246         }
1247         case BOOK3S_INTERRUPT_SYSCALL:
1248         {
1249                 /* hcall - punt to userspace */
1250                 int i;
1251 
1252                 /* hypercall with MSR_PR has already been handled in rmode,
1253                  * and never reaches here.
1254                  */
1255 
1256                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1257                 for (i = 0; i < 9; ++i)
1258                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1259                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1260                 vcpu->arch.hcall_needed = 1;
1261                 r = RESUME_HOST;
1262                 break;
1263         }
1264         /*
1265          * We get these next two if the guest accesses a page which it thinks
1266          * it has mapped but which is not actually present, either because
1267          * it is for an emulated I/O device or because the corresonding
1268          * host page has been paged out.  Any other HDSI/HISI interrupts
1269          * have been handled already.
1270          */
1271         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1272                 r = RESUME_PAGE_FAULT;
1273                 break;
1274         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1275                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1276                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1277                         DSISR_SRR1_MATCH_64S;
1278                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1279                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1280                 r = RESUME_PAGE_FAULT;
1281                 break;
1282         /*
1283          * This occurs if the guest executes an illegal instruction.
1284          * If the guest debug is disabled, generate a program interrupt
1285          * to the guest. If guest debug is enabled, we need to check
1286          * whether the instruction is a software breakpoint instruction.
1287          * Accordingly return to Guest or Host.
1288          */
1289         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1290                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1291                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1292                                 swab32(vcpu->arch.emul_inst) :
1293                                 vcpu->arch.emul_inst;
1294                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1295                         r = kvmppc_emulate_debug_inst(run, vcpu);
1296                 } else {
1297                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1298                         r = RESUME_GUEST;
1299                 }
1300                 break;
1301         /*
1302          * This occurs if the guest (kernel or userspace), does something that
1303          * is prohibited by HFSCR.
1304          * On POWER9, this could be a doorbell instruction that we need
1305          * to emulate.
1306          * Otherwise, we just generate a program interrupt to the guest.
1307          */
1308         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1309                 r = EMULATE_FAIL;
1310                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1311                     cpu_has_feature(CPU_FTR_ARCH_300))
1312                         r = kvmppc_emulate_doorbell_instr(vcpu);
1313                 if (r == EMULATE_FAIL) {
1314                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1315                         r = RESUME_GUEST;
1316                 }
1317                 break;
1318 
1319 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1320         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1321                 /*
1322                  * This occurs for various TM-related instructions that
1323                  * we need to emulate on POWER9 DD2.2.  We have already
1324                  * handled the cases where the guest was in real-suspend
1325                  * mode and was transitioning to transactional state.
1326                  */
1327                 r = kvmhv_p9_tm_emulation(vcpu);
1328                 break;
1329 #endif
1330 
1331         case BOOK3S_INTERRUPT_HV_RM_HARD:
1332                 r = RESUME_PASSTHROUGH;
1333                 break;
1334         default:
1335                 kvmppc_dump_regs(vcpu);
1336                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1337                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1338                         vcpu->arch.shregs.msr);
1339                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1340                 r = RESUME_HOST;
1341                 break;
1342         }
1343 
1344         return r;
1345 }
1346 
1347 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1348 {
1349         int r;
1350         int srcu_idx;
1351 
1352         vcpu->stat.sum_exits++;
1353 
1354         /*
1355          * This can happen if an interrupt occurs in the last stages
1356          * of guest entry or the first stages of guest exit (i.e. after
1357          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1358          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1359          * That can happen due to a bug, or due to a machine check
1360          * occurring at just the wrong time.
1361          */
1362         if (vcpu->arch.shregs.msr & MSR_HV) {
1363                 pr_emerg("KVM trap in HV mode while nested!\n");
1364                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1365                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1366                          vcpu->arch.shregs.msr);
1367                 kvmppc_dump_regs(vcpu);
1368                 return RESUME_HOST;
1369         }
1370         switch (vcpu->arch.trap) {
1371         /* We're good on these - the host merely wanted to get our attention */
1372         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1373                 vcpu->stat.dec_exits++;
1374                 r = RESUME_GUEST;
1375                 break;
1376         case BOOK3S_INTERRUPT_EXTERNAL:
1377                 vcpu->stat.ext_intr_exits++;
1378                 r = RESUME_HOST;
1379                 break;
1380         case BOOK3S_INTERRUPT_H_DOORBELL:
1381         case BOOK3S_INTERRUPT_H_VIRT:
1382                 vcpu->stat.ext_intr_exits++;
1383                 r = RESUME_GUEST;
1384                 break;
1385         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1386         case BOOK3S_INTERRUPT_HMI:
1387         case BOOK3S_INTERRUPT_PERFMON:
1388         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1389                 r = RESUME_GUEST;
1390                 break;
1391         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1392                 /* Pass the machine check to the L1 guest */
1393                 r = RESUME_HOST;
1394                 /* Print the MCE event to host console. */
1395                 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1396                 break;
1397         /*
1398          * We get these next two if the guest accesses a page which it thinks
1399          * it has mapped but which is not actually present, either because
1400          * it is for an emulated I/O device or because the corresonding
1401          * host page has been paged out.
1402          */
1403         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1404                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1405                 r = kvmhv_nested_page_fault(run, vcpu);
1406                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1407                 break;
1408         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1409                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1410                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1411                                          DSISR_SRR1_MATCH_64S;
1412                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1413                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1414                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1415                 r = kvmhv_nested_page_fault(run, vcpu);
1416                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1417                 break;
1418 
1419 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1420         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1421                 /*
1422                  * This occurs for various TM-related instructions that
1423                  * we need to emulate on POWER9 DD2.2.  We have already
1424                  * handled the cases where the guest was in real-suspend
1425                  * mode and was transitioning to transactional state.
1426                  */
1427                 r = kvmhv_p9_tm_emulation(vcpu);
1428                 break;
1429 #endif
1430 
1431         case BOOK3S_INTERRUPT_HV_RM_HARD:
1432                 vcpu->arch.trap = 0;
1433                 r = RESUME_GUEST;
1434                 if (!xive_enabled())
1435                         kvmppc_xics_rm_complete(vcpu, 0);
1436                 break;
1437         default:
1438                 r = RESUME_HOST;
1439                 break;
1440         }
1441 
1442         return r;
1443 }
1444 
1445 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1446                                             struct kvm_sregs *sregs)
1447 {
1448         int i;
1449 
1450         memset(sregs, 0, sizeof(struct kvm_sregs));
1451         sregs->pvr = vcpu->arch.pvr;
1452         for (i = 0; i < vcpu->arch.slb_max; i++) {
1453                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1454                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1455         }
1456 
1457         return 0;
1458 }
1459 
1460 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1461                                             struct kvm_sregs *sregs)
1462 {
1463         int i, j;
1464 
1465         /* Only accept the same PVR as the host's, since we can't spoof it */
1466         if (sregs->pvr != vcpu->arch.pvr)
1467                 return -EINVAL;
1468 
1469         j = 0;
1470         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1471                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1472                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1473                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1474                         ++j;
1475                 }
1476         }
1477         vcpu->arch.slb_max = j;
1478 
1479         return 0;
1480 }
1481 
1482 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1483                 bool preserve_top32)
1484 {
1485         struct kvm *kvm = vcpu->kvm;
1486         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1487         u64 mask;
1488 
1489         mutex_lock(&kvm->lock);
1490         spin_lock(&vc->lock);
1491         /*
1492          * If ILE (interrupt little-endian) has changed, update the
1493          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1494          */
1495         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1496                 struct kvm_vcpu *vcpu;
1497                 int i;
1498 
1499                 kvm_for_each_vcpu(i, vcpu, kvm) {
1500                         if (vcpu->arch.vcore != vc)
1501                                 continue;
1502                         if (new_lpcr & LPCR_ILE)
1503                                 vcpu->arch.intr_msr |= MSR_LE;
1504                         else
1505                                 vcpu->arch.intr_msr &= ~MSR_LE;
1506                 }
1507         }
1508 
1509         /*
1510          * Userspace can only modify DPFD (default prefetch depth),
1511          * ILE (interrupt little-endian) and TC (translation control).
1512          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1513          */
1514         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1515         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1516                 mask |= LPCR_AIL;
1517         /*
1518          * On POWER9, allow userspace to enable large decrementer for the
1519          * guest, whether or not the host has it enabled.
1520          */
1521         if (cpu_has_feature(CPU_FTR_ARCH_300))
1522                 mask |= LPCR_LD;
1523 
1524         /* Broken 32-bit version of LPCR must not clear top bits */
1525         if (preserve_top32)
1526                 mask &= 0xFFFFFFFF;
1527         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1528         spin_unlock(&vc->lock);
1529         mutex_unlock(&kvm->lock);
1530 }
1531 
1532 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1533                                  union kvmppc_one_reg *val)
1534 {
1535         int r = 0;
1536         long int i;
1537 
1538         switch (id) {
1539         case KVM_REG_PPC_DEBUG_INST:
1540                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1541                 break;
1542         case KVM_REG_PPC_HIOR:
1543                 *val = get_reg_val(id, 0);
1544                 break;
1545         case KVM_REG_PPC_DABR:
1546                 *val = get_reg_val(id, vcpu->arch.dabr);
1547                 break;
1548         case KVM_REG_PPC_DABRX:
1549                 *val = get_reg_val(id, vcpu->arch.dabrx);
1550                 break;
1551         case KVM_REG_PPC_DSCR:
1552                 *val = get_reg_val(id, vcpu->arch.dscr);
1553                 break;
1554         case KVM_REG_PPC_PURR:
1555                 *val = get_reg_val(id, vcpu->arch.purr);
1556                 break;
1557         case KVM_REG_PPC_SPURR:
1558                 *val = get_reg_val(id, vcpu->arch.spurr);
1559                 break;
1560         case KVM_REG_PPC_AMR:
1561                 *val = get_reg_val(id, vcpu->arch.amr);
1562                 break;
1563         case KVM_REG_PPC_UAMOR:
1564                 *val = get_reg_val(id, vcpu->arch.uamor);
1565                 break;
1566         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1567                 i = id - KVM_REG_PPC_MMCR0;
1568                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1569                 break;
1570         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1571                 i = id - KVM_REG_PPC_PMC1;
1572                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1573                 break;
1574         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1575                 i = id - KVM_REG_PPC_SPMC1;
1576                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1577                 break;
1578         case KVM_REG_PPC_SIAR:
1579                 *val = get_reg_val(id, vcpu->arch.siar);
1580                 break;
1581         case KVM_REG_PPC_SDAR:
1582                 *val = get_reg_val(id, vcpu->arch.sdar);
1583                 break;
1584         case KVM_REG_PPC_SIER:
1585                 *val = get_reg_val(id, vcpu->arch.sier);
1586                 break;
1587         case KVM_REG_PPC_IAMR:
1588                 *val = get_reg_val(id, vcpu->arch.iamr);
1589                 break;
1590         case KVM_REG_PPC_PSPB:
1591                 *val = get_reg_val(id, vcpu->arch.pspb);
1592                 break;
1593         case KVM_REG_PPC_DPDES:
1594                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1595                 break;
1596         case KVM_REG_PPC_VTB:
1597                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1598                 break;
1599         case KVM_REG_PPC_DAWR:
1600                 *val = get_reg_val(id, vcpu->arch.dawr);
1601                 break;
1602         case KVM_REG_PPC_DAWRX:
1603                 *val = get_reg_val(id, vcpu->arch.dawrx);
1604                 break;
1605         case KVM_REG_PPC_CIABR:
1606                 *val = get_reg_val(id, vcpu->arch.ciabr);
1607                 break;
1608         case KVM_REG_PPC_CSIGR:
1609                 *val = get_reg_val(id, vcpu->arch.csigr);
1610                 break;
1611         case KVM_REG_PPC_TACR:
1612                 *val = get_reg_val(id, vcpu->arch.tacr);
1613                 break;
1614         case KVM_REG_PPC_TCSCR:
1615                 *val = get_reg_val(id, vcpu->arch.tcscr);
1616                 break;
1617         case KVM_REG_PPC_PID:
1618                 *val = get_reg_val(id, vcpu->arch.pid);
1619                 break;
1620         case KVM_REG_PPC_ACOP:
1621                 *val = get_reg_val(id, vcpu->arch.acop);
1622                 break;
1623         case KVM_REG_PPC_WORT:
1624                 *val = get_reg_val(id, vcpu->arch.wort);
1625                 break;
1626         case KVM_REG_PPC_TIDR:
1627                 *val = get_reg_val(id, vcpu->arch.tid);
1628                 break;
1629         case KVM_REG_PPC_PSSCR:
1630                 *val = get_reg_val(id, vcpu->arch.psscr);
1631                 break;
1632         case KVM_REG_PPC_VPA_ADDR:
1633                 spin_lock(&vcpu->arch.vpa_update_lock);
1634                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1635                 spin_unlock(&vcpu->arch.vpa_update_lock);
1636                 break;
1637         case KVM_REG_PPC_VPA_SLB:
1638                 spin_lock(&vcpu->arch.vpa_update_lock);
1639                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1640                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1641                 spin_unlock(&vcpu->arch.vpa_update_lock);
1642                 break;
1643         case KVM_REG_PPC_VPA_DTL:
1644                 spin_lock(&vcpu->arch.vpa_update_lock);
1645                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1646                 val->vpaval.length = vcpu->arch.dtl.len;
1647                 spin_unlock(&vcpu->arch.vpa_update_lock);
1648                 break;
1649         case KVM_REG_PPC_TB_OFFSET:
1650                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1651                 break;
1652         case KVM_REG_PPC_LPCR:
1653         case KVM_REG_PPC_LPCR_64:
1654                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1655                 break;
1656         case KVM_REG_PPC_PPR:
1657                 *val = get_reg_val(id, vcpu->arch.ppr);
1658                 break;
1659 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1660         case KVM_REG_PPC_TFHAR:
1661                 *val = get_reg_val(id, vcpu->arch.tfhar);
1662                 break;
1663         case KVM_REG_PPC_TFIAR:
1664                 *val = get_reg_val(id, vcpu->arch.tfiar);
1665                 break;
1666         case KVM_REG_PPC_TEXASR:
1667                 *val = get_reg_val(id, vcpu->arch.texasr);
1668                 break;
1669         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1670                 i = id - KVM_REG_PPC_TM_GPR0;
1671                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1672                 break;
1673         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1674         {
1675                 int j;
1676                 i = id - KVM_REG_PPC_TM_VSR0;
1677                 if (i < 32)
1678                         for (j = 0; j < TS_FPRWIDTH; j++)
1679                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1680                 else {
1681                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1682                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1683                         else
1684                                 r = -ENXIO;
1685                 }
1686                 break;
1687         }
1688         case KVM_REG_PPC_TM_CR:
1689                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1690                 break;
1691         case KVM_REG_PPC_TM_XER:
1692                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1693                 break;
1694         case KVM_REG_PPC_TM_LR:
1695                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1696                 break;
1697         case KVM_REG_PPC_TM_CTR:
1698                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1699                 break;
1700         case KVM_REG_PPC_TM_FPSCR:
1701                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1702                 break;
1703         case KVM_REG_PPC_TM_AMR:
1704                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1705                 break;
1706         case KVM_REG_PPC_TM_PPR:
1707                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1708                 break;
1709         case KVM_REG_PPC_TM_VRSAVE:
1710                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1711                 break;
1712         case KVM_REG_PPC_TM_VSCR:
1713                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1714                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1715                 else
1716                         r = -ENXIO;
1717                 break;
1718         case KVM_REG_PPC_TM_DSCR:
1719                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1720                 break;
1721         case KVM_REG_PPC_TM_TAR:
1722                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1723                 break;
1724 #endif
1725         case KVM_REG_PPC_ARCH_COMPAT:
1726                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1727                 break;
1728         case KVM_REG_PPC_DEC_EXPIRY:
1729                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1730                                    vcpu->arch.vcore->tb_offset);
1731                 break;
1732         case KVM_REG_PPC_ONLINE:
1733                 *val = get_reg_val(id, vcpu->arch.online);
1734                 break;
1735         case KVM_REG_PPC_PTCR:
1736                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1737                 break;
1738         default:
1739                 r = -EINVAL;
1740                 break;
1741         }
1742 
1743         return r;
1744 }
1745 
1746 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1747                                  union kvmppc_one_reg *val)
1748 {
1749         int r = 0;
1750         long int i;
1751         unsigned long addr, len;
1752 
1753         switch (id) {
1754         case KVM_REG_PPC_HIOR:
1755                 /* Only allow this to be set to zero */
1756                 if (set_reg_val(id, *val))
1757                         r = -EINVAL;
1758                 break;
1759         case KVM_REG_PPC_DABR:
1760                 vcpu->arch.dabr = set_reg_val(id, *val);
1761                 break;
1762         case KVM_REG_PPC_DABRX:
1763                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1764                 break;
1765         case KVM_REG_PPC_DSCR:
1766                 vcpu->arch.dscr = set_reg_val(id, *val);
1767                 break;
1768         case KVM_REG_PPC_PURR:
1769                 vcpu->arch.purr = set_reg_val(id, *val);
1770                 break;
1771         case KVM_REG_PPC_SPURR:
1772                 vcpu->arch.spurr = set_reg_val(id, *val);
1773                 break;
1774         case KVM_REG_PPC_AMR:
1775                 vcpu->arch.amr = set_reg_val(id, *val);
1776                 break;
1777         case KVM_REG_PPC_UAMOR:
1778                 vcpu->arch.uamor = set_reg_val(id, *val);
1779                 break;
1780         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1781                 i = id - KVM_REG_PPC_MMCR0;
1782                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1783                 break;
1784         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1785                 i = id - KVM_REG_PPC_PMC1;
1786                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1787                 break;
1788         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1789                 i = id - KVM_REG_PPC_SPMC1;
1790                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1791                 break;
1792         case KVM_REG_PPC_SIAR:
1793                 vcpu->arch.siar = set_reg_val(id, *val);
1794                 break;
1795         case KVM_REG_PPC_SDAR:
1796                 vcpu->arch.sdar = set_reg_val(id, *val);
1797                 break;
1798         case KVM_REG_PPC_SIER:
1799                 vcpu->arch.sier = set_reg_val(id, *val);
1800                 break;
1801         case KVM_REG_PPC_IAMR:
1802                 vcpu->arch.iamr = set_reg_val(id, *val);
1803                 break;
1804         case KVM_REG_PPC_PSPB:
1805                 vcpu->arch.pspb = set_reg_val(id, *val);
1806                 break;
1807         case KVM_REG_PPC_DPDES:
1808                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1809                 break;
1810         case KVM_REG_PPC_VTB:
1811                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1812                 break;
1813         case KVM_REG_PPC_DAWR:
1814                 vcpu->arch.dawr = set_reg_val(id, *val);
1815                 break;
1816         case KVM_REG_PPC_DAWRX:
1817                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1818                 break;
1819         case KVM_REG_PPC_CIABR:
1820                 vcpu->arch.ciabr = set_reg_val(id, *val);
1821                 /* Don't allow setting breakpoints in hypervisor code */
1822                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1823                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1824                 break;
1825         case KVM_REG_PPC_CSIGR:
1826                 vcpu->arch.csigr = set_reg_val(id, *val);
1827                 break;
1828         case KVM_REG_PPC_TACR:
1829                 vcpu->arch.tacr = set_reg_val(id, *val);
1830                 break;
1831         case KVM_REG_PPC_TCSCR:
1832                 vcpu->arch.tcscr = set_reg_val(id, *val);
1833                 break;
1834         case KVM_REG_PPC_PID:
1835                 vcpu->arch.pid = set_reg_val(id, *val);
1836                 break;
1837         case KVM_REG_PPC_ACOP:
1838                 vcpu->arch.acop = set_reg_val(id, *val);
1839                 break;
1840         case KVM_REG_PPC_WORT:
1841                 vcpu->arch.wort = set_reg_val(id, *val);
1842                 break;
1843         case KVM_REG_PPC_TIDR:
1844                 vcpu->arch.tid = set_reg_val(id, *val);
1845                 break;
1846         case KVM_REG_PPC_PSSCR:
1847                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1848                 break;
1849         case KVM_REG_PPC_VPA_ADDR:
1850                 addr = set_reg_val(id, *val);
1851                 r = -EINVAL;
1852                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1853                               vcpu->arch.dtl.next_gpa))
1854                         break;
1855                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1856                 break;
1857         case KVM_REG_PPC_VPA_SLB:
1858                 addr = val->vpaval.addr;
1859                 len = val->vpaval.length;
1860                 r = -EINVAL;
1861                 if (addr && !vcpu->arch.vpa.next_gpa)
1862                         break;
1863                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1864                 break;
1865         case KVM_REG_PPC_VPA_DTL:
1866                 addr = val->vpaval.addr;
1867                 len = val->vpaval.length;
1868                 r = -EINVAL;
1869                 if (addr && (len < sizeof(struct dtl_entry) ||
1870                              !vcpu->arch.vpa.next_gpa))
1871                         break;
1872                 len -= len % sizeof(struct dtl_entry);
1873                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1874                 break;
1875         case KVM_REG_PPC_TB_OFFSET:
1876                 /* round up to multiple of 2^24 */
1877                 vcpu->arch.vcore->tb_offset =
1878                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1879                 break;
1880         case KVM_REG_PPC_LPCR:
1881                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1882                 break;
1883         case KVM_REG_PPC_LPCR_64:
1884                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1885                 break;
1886         case KVM_REG_PPC_PPR:
1887                 vcpu->arch.ppr = set_reg_val(id, *val);
1888                 break;
1889 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1890         case KVM_REG_PPC_TFHAR:
1891                 vcpu->arch.tfhar = set_reg_val(id, *val);
1892                 break;
1893         case KVM_REG_PPC_TFIAR:
1894                 vcpu->arch.tfiar = set_reg_val(id, *val);
1895                 break;
1896         case KVM_REG_PPC_TEXASR:
1897                 vcpu->arch.texasr = set_reg_val(id, *val);
1898                 break;
1899         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1900                 i = id - KVM_REG_PPC_TM_GPR0;
1901                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1902                 break;
1903         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1904         {
1905                 int j;
1906                 i = id - KVM_REG_PPC_TM_VSR0;
1907                 if (i < 32)
1908                         for (j = 0; j < TS_FPRWIDTH; j++)
1909                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1910                 else
1911                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1912                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1913                         else
1914                                 r = -ENXIO;
1915                 break;
1916         }
1917         case KVM_REG_PPC_TM_CR:
1918                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1919                 break;
1920         case KVM_REG_PPC_TM_XER:
1921                 vcpu->arch.xer_tm = set_reg_val(id, *val);
1922                 break;
1923         case KVM_REG_PPC_TM_LR:
1924                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1925                 break;
1926         case KVM_REG_PPC_TM_CTR:
1927                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1928                 break;
1929         case KVM_REG_PPC_TM_FPSCR:
1930                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1931                 break;
1932         case KVM_REG_PPC_TM_AMR:
1933                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1934                 break;
1935         case KVM_REG_PPC_TM_PPR:
1936                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1937                 break;
1938         case KVM_REG_PPC_TM_VRSAVE:
1939                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1940                 break;
1941         case KVM_REG_PPC_TM_VSCR:
1942                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1943                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1944                 else
1945                         r = - ENXIO;
1946                 break;
1947         case KVM_REG_PPC_TM_DSCR:
1948                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1949                 break;
1950         case KVM_REG_PPC_TM_TAR:
1951                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1952                 break;
1953 #endif
1954         case KVM_REG_PPC_ARCH_COMPAT:
1955                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1956                 break;
1957         case KVM_REG_PPC_DEC_EXPIRY:
1958                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1959                         vcpu->arch.vcore->tb_offset;
1960                 break;
1961         case KVM_REG_PPC_ONLINE:
1962                 i = set_reg_val(id, *val);
1963                 if (i && !vcpu->arch.online)
1964                         atomic_inc(&vcpu->arch.vcore->online_count);
1965                 else if (!i && vcpu->arch.online)
1966                         atomic_dec(&vcpu->arch.vcore->online_count);
1967                 vcpu->arch.online = i;
1968                 break;
1969         case KVM_REG_PPC_PTCR:
1970                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
1971                 break;
1972         default:
1973                 r = -EINVAL;
1974                 break;
1975         }
1976 
1977         return r;
1978 }
1979 
1980 /*
1981  * On POWER9, threads are independent and can be in different partitions.
1982  * Therefore we consider each thread to be a subcore.
1983  * There is a restriction that all threads have to be in the same
1984  * MMU mode (radix or HPT), unfortunately, but since we only support
1985  * HPT guests on a HPT host so far, that isn't an impediment yet.
1986  */
1987 static int threads_per_vcore(struct kvm *kvm)
1988 {
1989         if (kvm->arch.threads_indep)
1990                 return 1;
1991         return threads_per_subcore;
1992 }
1993 
1994 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
1995 {
1996         struct kvmppc_vcore *vcore;
1997 
1998         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1999 
2000         if (vcore == NULL)
2001                 return NULL;
2002 
2003         spin_lock_init(&vcore->lock);
2004         spin_lock_init(&vcore->stoltb_lock);
2005         init_swait_queue_head(&vcore->wq);
2006         vcore->preempt_tb = TB_NIL;
2007         vcore->lpcr = kvm->arch.lpcr;
2008         vcore->first_vcpuid = id;
2009         vcore->kvm = kvm;
2010         INIT_LIST_HEAD(&vcore->preempt_list);
2011 
2012         return vcore;
2013 }
2014 
2015 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2016 static struct debugfs_timings_element {
2017         const char *name;
2018         size_t offset;
2019 } timings[] = {
2020         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2021         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2022         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2023         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2024         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2025 };
2026 
2027 #define N_TIMINGS       (ARRAY_SIZE(timings))
2028 
2029 struct debugfs_timings_state {
2030         struct kvm_vcpu *vcpu;
2031         unsigned int    buflen;
2032         char            buf[N_TIMINGS * 100];
2033 };
2034 
2035 static int debugfs_timings_open(struct inode *inode, struct file *file)
2036 {
2037         struct kvm_vcpu *vcpu = inode->i_private;
2038         struct debugfs_timings_state *p;
2039 
2040         p = kzalloc(sizeof(*p), GFP_KERNEL);
2041         if (!p)
2042                 return -ENOMEM;
2043 
2044         kvm_get_kvm(vcpu->kvm);
2045         p->vcpu = vcpu;
2046         file->private_data = p;
2047 
2048         return nonseekable_open(inode, file);
2049 }
2050 
2051 static int debugfs_timings_release(struct inode *inode, struct file *file)
2052 {
2053         struct debugfs_timings_state *p = file->private_data;
2054 
2055         kvm_put_kvm(p->vcpu->kvm);
2056         kfree(p);
2057         return 0;
2058 }
2059 
2060 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2061                                     size_t len, loff_t *ppos)
2062 {
2063         struct debugfs_timings_state *p = file->private_data;
2064         struct kvm_vcpu *vcpu = p->vcpu;
2065         char *s, *buf_end;
2066         struct kvmhv_tb_accumulator tb;
2067         u64 count;
2068         loff_t pos;
2069         ssize_t n;
2070         int i, loops;
2071         bool ok;
2072 
2073         if (!p->buflen) {
2074                 s = p->buf;
2075                 buf_end = s + sizeof(p->buf);
2076                 for (i = 0; i < N_TIMINGS; ++i) {
2077                         struct kvmhv_tb_accumulator *acc;
2078 
2079                         acc = (struct kvmhv_tb_accumulator *)
2080                                 ((unsigned long)vcpu + timings[i].offset);
2081                         ok = false;
2082                         for (loops = 0; loops < 1000; ++loops) {
2083                                 count = acc->seqcount;
2084                                 if (!(count & 1)) {
2085                                         smp_rmb();
2086                                         tb = *acc;
2087                                         smp_rmb();
2088                                         if (count == acc->seqcount) {
2089                                                 ok = true;
2090                                                 break;
2091                                         }
2092                                 }
2093                                 udelay(1);
2094                         }
2095                         if (!ok)
2096                                 snprintf(s, buf_end - s, "%s: stuck\n",
2097                                         timings[i].name);
2098                         else
2099                                 snprintf(s, buf_end - s,
2100                                         "%s: %llu %llu %llu %llu\n",
2101                                         timings[i].name, count / 2,
2102                                         tb_to_ns(tb.tb_total),
2103                                         tb_to_ns(tb.tb_min),
2104                                         tb_to_ns(tb.tb_max));
2105                         s += strlen(s);
2106                 }
2107                 p->buflen = s - p->buf;
2108         }
2109 
2110         pos = *ppos;
2111         if (pos >= p->buflen)
2112                 return 0;
2113         if (len > p->buflen - pos)
2114                 len = p->buflen - pos;
2115         n = copy_to_user(buf, p->buf + pos, len);
2116         if (n) {
2117                 if (n == len)
2118                         return -EFAULT;
2119                 len -= n;
2120         }
2121         *ppos = pos + len;
2122         return len;
2123 }
2124 
2125 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2126                                      size_t len, loff_t *ppos)
2127 {
2128         return -EACCES;
2129 }
2130 
2131 static const struct file_operations debugfs_timings_ops = {
2132         .owner   = THIS_MODULE,
2133         .open    = debugfs_timings_open,
2134         .release = debugfs_timings_release,
2135         .read    = debugfs_timings_read,
2136         .write   = debugfs_timings_write,
2137         .llseek  = generic_file_llseek,
2138 };
2139 
2140 /* Create a debugfs directory for the vcpu */
2141 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2142 {
2143         char buf[16];
2144         struct kvm *kvm = vcpu->kvm;
2145 
2146         snprintf(buf, sizeof(buf), "vcpu%u", id);
2147         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2148                 return;
2149         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2150         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2151                 return;
2152         vcpu->arch.debugfs_timings =
2153                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2154                                     vcpu, &debugfs_timings_ops);
2155 }
2156 
2157 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2158 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2159 {
2160 }
2161 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2162 
2163 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2164                                                    unsigned int id)
2165 {
2166         struct kvm_vcpu *vcpu;
2167         int err;
2168         int core;
2169         struct kvmppc_vcore *vcore;
2170 
2171         err = -ENOMEM;
2172         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2173         if (!vcpu)
2174                 goto out;
2175 
2176         err = kvm_vcpu_init(vcpu, kvm, id);
2177         if (err)
2178                 goto free_vcpu;
2179 
2180         vcpu->arch.shared = &vcpu->arch.shregs;
2181 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2182         /*
2183          * The shared struct is never shared on HV,
2184          * so we can always use host endianness
2185          */
2186 #ifdef __BIG_ENDIAN__
2187         vcpu->arch.shared_big_endian = true;
2188 #else
2189         vcpu->arch.shared_big_endian = false;
2190 #endif
2191 #endif
2192         vcpu->arch.mmcr[0] = MMCR0_FC;
2193         vcpu->arch.ctrl = CTRL_RUNLATCH;
2194         /* default to host PVR, since we can't spoof it */
2195         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2196         spin_lock_init(&vcpu->arch.vpa_update_lock);
2197         spin_lock_init(&vcpu->arch.tbacct_lock);
2198         vcpu->arch.busy_preempt = TB_NIL;
2199         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2200 
2201         /*
2202          * Set the default HFSCR for the guest from the host value.
2203          * This value is only used on POWER9.
2204          * On POWER9, we want to virtualize the doorbell facility, so we
2205          * don't set the HFSCR_MSGP bit, and that causes those instructions
2206          * to trap and then we emulate them.
2207          */
2208         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2209                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2210         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2211                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2212                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2213                         vcpu->arch.hfscr |= HFSCR_TM;
2214         }
2215         if (cpu_has_feature(CPU_FTR_TM_COMP))
2216                 vcpu->arch.hfscr |= HFSCR_TM;
2217 
2218         kvmppc_mmu_book3s_hv_init(vcpu);
2219 
2220         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2221 
2222         init_waitqueue_head(&vcpu->arch.cpu_run);
2223 
2224         mutex_lock(&kvm->lock);
2225         vcore = NULL;
2226         err = -EINVAL;
2227         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2228                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2229                         pr_devel("KVM: VCPU ID too high\n");
2230                         core = KVM_MAX_VCORES;
2231                 } else {
2232                         BUG_ON(kvm->arch.smt_mode != 1);
2233                         core = kvmppc_pack_vcpu_id(kvm, id);
2234                 }
2235         } else {
2236                 core = id / kvm->arch.smt_mode;
2237         }
2238         if (core < KVM_MAX_VCORES) {
2239                 vcore = kvm->arch.vcores[core];
2240                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2241                         pr_devel("KVM: collision on id %u", id);
2242                         vcore = NULL;
2243                 } else if (!vcore) {
2244                         err = -ENOMEM;
2245                         vcore = kvmppc_vcore_create(kvm,
2246                                         id & ~(kvm->arch.smt_mode - 1));
2247                         kvm->arch.vcores[core] = vcore;
2248                         kvm->arch.online_vcores++;
2249                 }
2250         }
2251         mutex_unlock(&kvm->lock);
2252 
2253         if (!vcore)
2254                 goto free_vcpu;
2255 
2256         spin_lock(&vcore->lock);
2257         ++vcore->num_threads;
2258         spin_unlock(&vcore->lock);
2259         vcpu->arch.vcore = vcore;
2260         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2261         vcpu->arch.thread_cpu = -1;
2262         vcpu->arch.prev_cpu = -1;
2263 
2264         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2265         kvmppc_sanity_check(vcpu);
2266 
2267         debugfs_vcpu_init(vcpu, id);
2268 
2269         return vcpu;
2270 
2271 free_vcpu:
2272         kmem_cache_free(kvm_vcpu_cache, vcpu);
2273 out:
2274         return ERR_PTR(err);
2275 }
2276 
2277 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2278                               unsigned long flags)
2279 {
2280         int err;
2281         int esmt = 0;
2282 
2283         if (flags)
2284                 return -EINVAL;
2285         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2286                 return -EINVAL;
2287         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2288                 /*
2289                  * On POWER8 (or POWER7), the threading mode is "strict",
2290                  * so we pack smt_mode vcpus per vcore.
2291                  */
2292                 if (smt_mode > threads_per_subcore)
2293                         return -EINVAL;
2294         } else {
2295                 /*
2296                  * On POWER9, the threading mode is "loose",
2297                  * so each vcpu gets its own vcore.
2298                  */
2299                 esmt = smt_mode;
2300                 smt_mode = 1;
2301         }
2302         mutex_lock(&kvm->lock);
2303         err = -EBUSY;
2304         if (!kvm->arch.online_vcores) {
2305                 kvm->arch.smt_mode = smt_mode;
2306                 kvm->arch.emul_smt_mode = esmt;
2307                 err = 0;
2308         }
2309         mutex_unlock(&kvm->lock);
2310 
2311         return err;
2312 }
2313 
2314 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2315 {
2316         if (vpa->pinned_addr)
2317                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2318                                         vpa->dirty);
2319 }
2320 
2321 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2322 {
2323         spin_lock(&vcpu->arch.vpa_update_lock);
2324         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2325         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2326         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2327         spin_unlock(&vcpu->arch.vpa_update_lock);
2328         kvm_vcpu_uninit(vcpu);
2329         kmem_cache_free(kvm_vcpu_cache, vcpu);
2330 }
2331 
2332 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2333 {
2334         /* Indicate we want to get back into the guest */
2335         return 1;
2336 }
2337 
2338 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2339 {
2340         unsigned long dec_nsec, now;
2341 
2342         now = get_tb();
2343         if (now > vcpu->arch.dec_expires) {
2344                 /* decrementer has already gone negative */
2345                 kvmppc_core_queue_dec(vcpu);
2346                 kvmppc_core_prepare_to_enter(vcpu);
2347                 return;
2348         }
2349         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2350         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2351         vcpu->arch.timer_running = 1;
2352 }
2353 
2354 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2355 {
2356         vcpu->arch.ceded = 0;
2357         if (vcpu->arch.timer_running) {
2358                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2359                 vcpu->arch.timer_running = 0;
2360         }
2361 }
2362 
2363 extern int __kvmppc_vcore_entry(void);
2364 
2365 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2366                                    struct kvm_vcpu *vcpu)
2367 {
2368         u64 now;
2369 
2370         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2371                 return;
2372         spin_lock_irq(&vcpu->arch.tbacct_lock);
2373         now = mftb();
2374         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2375                 vcpu->arch.stolen_logged;
2376         vcpu->arch.busy_preempt = now;
2377         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2378         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2379         --vc->n_runnable;
2380         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2381 }
2382 
2383 static int kvmppc_grab_hwthread(int cpu)
2384 {
2385         struct paca_struct *tpaca;
2386         long timeout = 10000;
2387 
2388         tpaca = paca_ptrs[cpu];
2389 
2390         /* Ensure the thread won't go into the kernel if it wakes */
2391         tpaca->kvm_hstate.kvm_vcpu = NULL;
2392         tpaca->kvm_hstate.kvm_vcore = NULL;
2393         tpaca->kvm_hstate.napping = 0;
2394         smp_wmb();
2395         tpaca->kvm_hstate.hwthread_req = 1;
2396 
2397         /*
2398          * If the thread is already executing in the kernel (e.g. handling
2399          * a stray interrupt), wait for it to get back to nap mode.
2400          * The smp_mb() is to ensure that our setting of hwthread_req
2401          * is visible before we look at hwthread_state, so if this
2402          * races with the code at system_reset_pSeries and the thread
2403          * misses our setting of hwthread_req, we are sure to see its
2404          * setting of hwthread_state, and vice versa.
2405          */
2406         smp_mb();
2407         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2408                 if (--timeout <= 0) {
2409                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2410                         return -EBUSY;
2411                 }
2412                 udelay(1);
2413         }
2414         return 0;
2415 }
2416 
2417 static void kvmppc_release_hwthread(int cpu)
2418 {
2419         struct paca_struct *tpaca;
2420 
2421         tpaca = paca_ptrs[cpu];
2422         tpaca->kvm_hstate.hwthread_req = 0;
2423         tpaca->kvm_hstate.kvm_vcpu = NULL;
2424         tpaca->kvm_hstate.kvm_vcore = NULL;
2425         tpaca->kvm_hstate.kvm_split_mode = NULL;
2426 }
2427 
2428 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2429 {
2430         struct kvm_nested_guest *nested = vcpu->arch.nested;
2431         cpumask_t *cpu_in_guest;
2432         int i;
2433 
2434         cpu = cpu_first_thread_sibling(cpu);
2435         if (nested) {
2436                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2437                 cpu_in_guest = &nested->cpu_in_guest;
2438         } else {
2439                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2440                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2441         }
2442         /*
2443          * Make sure setting of bit in need_tlb_flush precedes
2444          * testing of cpu_in_guest bits.  The matching barrier on
2445          * the other side is the first smp_mb() in kvmppc_run_core().
2446          */
2447         smp_mb();
2448         for (i = 0; i < threads_per_core; ++i)
2449                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2450                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2451 }
2452 
2453 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2454 {
2455         struct kvm_nested_guest *nested = vcpu->arch.nested;
2456         struct kvm *kvm = vcpu->kvm;
2457         int prev_cpu;
2458 
2459         if (!cpu_has_feature(CPU_FTR_HVMODE))
2460                 return;
2461 
2462         if (nested)
2463                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2464         else
2465                 prev_cpu = vcpu->arch.prev_cpu;
2466 
2467         /*
2468          * With radix, the guest can do TLB invalidations itself,
2469          * and it could choose to use the local form (tlbiel) if
2470          * it is invalidating a translation that has only ever been
2471          * used on one vcpu.  However, that doesn't mean it has
2472          * only ever been used on one physical cpu, since vcpus
2473          * can move around between pcpus.  To cope with this, when
2474          * a vcpu moves from one pcpu to another, we need to tell
2475          * any vcpus running on the same core as this vcpu previously
2476          * ran to flush the TLB.  The TLB is shared between threads,
2477          * so we use a single bit in .need_tlb_flush for all 4 threads.
2478          */
2479         if (prev_cpu != pcpu) {
2480                 if (prev_cpu >= 0 &&
2481                     cpu_first_thread_sibling(prev_cpu) !=
2482                     cpu_first_thread_sibling(pcpu))
2483                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2484                 if (nested)
2485                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2486                 else
2487                         vcpu->arch.prev_cpu = pcpu;
2488         }
2489 }
2490 
2491 static void kvmppc_radix_check_need_tlb_flush(struct kvm *kvm, int pcpu,
2492                                               struct kvm_nested_guest *nested)
2493 {
2494         cpumask_t *need_tlb_flush;
2495         int lpid;
2496 
2497         if (!cpu_has_feature(CPU_FTR_HVMODE))
2498                 return;
2499 
2500         if (cpu_has_feature(CPU_FTR_ARCH_300))
2501                 pcpu &= ~0x3UL;
2502 
2503         if (nested) {
2504                 lpid = nested->shadow_lpid;
2505                 need_tlb_flush = &nested->need_tlb_flush;
2506         } else {
2507                 lpid = kvm->arch.lpid;
2508                 need_tlb_flush = &kvm->arch.need_tlb_flush;
2509         }
2510 
2511         mtspr(SPRN_LPID, lpid);
2512         isync();
2513         smp_mb();
2514 
2515         if (cpumask_test_cpu(pcpu, need_tlb_flush)) {
2516                 radix__local_flush_tlb_lpid_guest(lpid);
2517                 /* Clear the bit after the TLB flush */
2518                 cpumask_clear_cpu(pcpu, need_tlb_flush);
2519         }
2520 }
2521 
2522 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2523 {
2524         int cpu;
2525         struct paca_struct *tpaca;
2526         struct kvm *kvm = vc->kvm;
2527 
2528         cpu = vc->pcpu;
2529         if (vcpu) {
2530                 if (vcpu->arch.timer_running) {
2531                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2532                         vcpu->arch.timer_running = 0;
2533                 }
2534                 cpu += vcpu->arch.ptid;
2535                 vcpu->cpu = vc->pcpu;
2536                 vcpu->arch.thread_cpu = cpu;
2537                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2538         }
2539         tpaca = paca_ptrs[cpu];
2540         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2541         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2542         tpaca->kvm_hstate.fake_suspend = 0;
2543         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2544         smp_wmb();
2545         tpaca->kvm_hstate.kvm_vcore = vc;
2546         if (cpu != smp_processor_id())
2547                 kvmppc_ipi_thread(cpu);
2548 }
2549 
2550 static void kvmppc_wait_for_nap(int n_threads)
2551 {
2552         int cpu = smp_processor_id();
2553         int i, loops;
2554 
2555         if (n_threads <= 1)
2556                 return;
2557         for (loops = 0; loops < 1000000; ++loops) {
2558                 /*
2559                  * Check if all threads are finished.
2560                  * We set the vcore pointer when starting a thread
2561                  * and the thread clears it when finished, so we look
2562                  * for any threads that still have a non-NULL vcore ptr.
2563                  */
2564                 for (i = 1; i < n_threads; ++i)
2565                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2566                                 break;
2567                 if (i == n_threads) {
2568                         HMT_medium();
2569                         return;
2570                 }
2571                 HMT_low();
2572         }
2573         HMT_medium();
2574         for (i = 1; i < n_threads; ++i)
2575                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2576                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2577 }
2578 
2579 /*
2580  * Check that we are on thread 0 and that any other threads in
2581  * this core are off-line.  Then grab the threads so they can't
2582  * enter the kernel.
2583  */
2584 static int on_primary_thread(void)
2585 {
2586         int cpu = smp_processor_id();
2587         int thr;
2588 
2589         /* Are we on a primary subcore? */
2590         if (cpu_thread_in_subcore(cpu))
2591                 return 0;
2592 
2593         thr = 0;
2594         while (++thr < threads_per_subcore)
2595                 if (cpu_online(cpu + thr))
2596                         return 0;
2597 
2598         /* Grab all hw threads so they can't go into the kernel */
2599         for (thr = 1; thr < threads_per_subcore; ++thr) {
2600                 if (kvmppc_grab_hwthread(cpu + thr)) {
2601                         /* Couldn't grab one; let the others go */
2602                         do {
2603                                 kvmppc_release_hwthread(cpu + thr);
2604                         } while (--thr > 0);
2605                         return 0;
2606                 }
2607         }
2608         return 1;
2609 }
2610 
2611 /*
2612  * A list of virtual cores for each physical CPU.
2613  * These are vcores that could run but their runner VCPU tasks are
2614  * (or may be) preempted.
2615  */
2616 struct preempted_vcore_list {
2617         struct list_head        list;
2618         spinlock_t              lock;
2619 };
2620 
2621 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2622 
2623 static void init_vcore_lists(void)
2624 {
2625         int cpu;
2626 
2627         for_each_possible_cpu(cpu) {
2628                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2629                 spin_lock_init(&lp->lock);
2630                 INIT_LIST_HEAD(&lp->list);
2631         }
2632 }
2633 
2634 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2635 {
2636         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2637 
2638         vc->vcore_state = VCORE_PREEMPT;
2639         vc->pcpu = smp_processor_id();
2640         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2641                 spin_lock(&lp->lock);
2642                 list_add_tail(&vc->preempt_list, &lp->list);
2643                 spin_unlock(&lp->lock);
2644         }
2645 
2646         /* Start accumulating stolen time */
2647         kvmppc_core_start_stolen(vc);
2648 }
2649 
2650 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2651 {
2652         struct preempted_vcore_list *lp;
2653 
2654         kvmppc_core_end_stolen(vc);
2655         if (!list_empty(&vc->preempt_list)) {
2656                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2657                 spin_lock(&lp->lock);
2658                 list_del_init(&vc->preempt_list);
2659                 spin_unlock(&lp->lock);
2660         }
2661         vc->vcore_state = VCORE_INACTIVE;
2662 }
2663 
2664 /*
2665  * This stores information about the virtual cores currently
2666  * assigned to a physical core.
2667  */
2668 struct core_info {
2669         int             n_subcores;
2670         int             max_subcore_threads;
2671         int             total_threads;
2672         int             subcore_threads[MAX_SUBCORES];
2673         struct kvmppc_vcore *vc[MAX_SUBCORES];
2674 };
2675 
2676 /*
2677  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2678  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2679  */
2680 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2681 
2682 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2683 {
2684         memset(cip, 0, sizeof(*cip));
2685         cip->n_subcores = 1;
2686         cip->max_subcore_threads = vc->num_threads;
2687         cip->total_threads = vc->num_threads;
2688         cip->subcore_threads[0] = vc->num_threads;
2689         cip->vc[0] = vc;
2690 }
2691 
2692 static bool subcore_config_ok(int n_subcores, int n_threads)
2693 {
2694         /*
2695          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2696          * split-core mode, with one thread per subcore.
2697          */
2698         if (cpu_has_feature(CPU_FTR_ARCH_300))
2699                 return n_subcores <= 4 && n_threads == 1;
2700 
2701         /* On POWER8, can only dynamically split if unsplit to begin with */
2702         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2703                 return false;
2704         if (n_subcores > MAX_SUBCORES)
2705                 return false;
2706         if (n_subcores > 1) {
2707                 if (!(dynamic_mt_modes & 2))
2708                         n_subcores = 4;
2709                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2710                         return false;
2711         }
2712 
2713         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2714 }
2715 
2716 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2717 {
2718         vc->entry_exit_map = 0;
2719         vc->in_guest = 0;
2720         vc->napping_threads = 0;
2721         vc->conferring_threads = 0;
2722         vc->tb_offset_applied = 0;
2723 }
2724 
2725 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2726 {
2727         int n_threads = vc->num_threads;
2728         int sub;
2729 
2730         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2731                 return false;
2732 
2733         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2734         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2735                 return false;
2736 
2737         /* Some POWER9 chips require all threads to be in the same MMU mode */
2738         if (no_mixing_hpt_and_radix &&
2739             kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2740                 return false;
2741 
2742         if (n_threads < cip->max_subcore_threads)
2743                 n_threads = cip->max_subcore_threads;
2744         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2745                 return false;
2746         cip->max_subcore_threads = n_threads;
2747 
2748         sub = cip->n_subcores;
2749         ++cip->n_subcores;
2750         cip->total_threads += vc->num_threads;
2751         cip->subcore_threads[sub] = vc->num_threads;
2752         cip->vc[sub] = vc;
2753         init_vcore_to_run(vc);
2754         list_del_init(&vc->preempt_list);
2755 
2756         return true;
2757 }
2758 
2759 /*
2760  * Work out whether it is possible to piggyback the execution of
2761  * vcore *pvc onto the execution of the other vcores described in *cip.
2762  */
2763 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2764                           int target_threads)
2765 {
2766         if (cip->total_threads + pvc->num_threads > target_threads)
2767                 return false;
2768 
2769         return can_dynamic_split(pvc, cip);
2770 }
2771 
2772 static void prepare_threads(struct kvmppc_vcore *vc)
2773 {
2774         int i;
2775         struct kvm_vcpu *vcpu;
2776 
2777         for_each_runnable_thread(i, vcpu, vc) {
2778                 if (signal_pending(vcpu->arch.run_task))
2779                         vcpu->arch.ret = -EINTR;
2780                 else if (vcpu->arch.vpa.update_pending ||
2781                          vcpu->arch.slb_shadow.update_pending ||
2782                          vcpu->arch.dtl.update_pending)
2783                         vcpu->arch.ret = RESUME_GUEST;
2784                 else
2785                         continue;
2786                 kvmppc_remove_runnable(vc, vcpu);
2787                 wake_up(&vcpu->arch.cpu_run);
2788         }
2789 }
2790 
2791 static void collect_piggybacks(struct core_info *cip, int target_threads)
2792 {
2793         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2794         struct kvmppc_vcore *pvc, *vcnext;
2795 
2796         spin_lock(&lp->lock);
2797         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2798                 if (!spin_trylock(&pvc->lock))
2799                         continue;
2800                 prepare_threads(pvc);
2801                 if (!pvc->n_runnable) {
2802                         list_del_init(&pvc->preempt_list);
2803                         if (pvc->runner == NULL) {
2804                                 pvc->vcore_state = VCORE_INACTIVE;
2805                                 kvmppc_core_end_stolen(pvc);
2806                         }
2807                         spin_unlock(&pvc->lock);
2808                         continue;
2809                 }
2810                 if (!can_piggyback(pvc, cip, target_threads)) {
2811                         spin_unlock(&pvc->lock);
2812                         continue;
2813                 }
2814                 kvmppc_core_end_stolen(pvc);
2815                 pvc->vcore_state = VCORE_PIGGYBACK;
2816                 if (cip->total_threads >= target_threads)
2817                         break;
2818         }
2819         spin_unlock(&lp->lock);
2820 }
2821 
2822 static bool recheck_signals(struct core_info *cip)
2823 {
2824         int sub, i;
2825         struct kvm_vcpu *vcpu;
2826 
2827         for (sub = 0; sub < cip->n_subcores; ++sub)
2828                 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2829                         if (signal_pending(vcpu->arch.run_task))
2830                                 return true;
2831         return false;
2832 }
2833 
2834 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2835 {
2836         int still_running = 0, i;
2837         u64 now;
2838         long ret;
2839         struct kvm_vcpu *vcpu;
2840 
2841         spin_lock(&vc->lock);
2842         now = get_tb();
2843         for_each_runnable_thread(i, vcpu, vc) {
2844                 /*
2845                  * It's safe to unlock the vcore in the loop here, because
2846                  * for_each_runnable_thread() is safe against removal of
2847                  * the vcpu, and the vcore state is VCORE_EXITING here,
2848                  * so any vcpus becoming runnable will have their arch.trap
2849                  * set to zero and can't actually run in the guest.
2850                  */
2851                 spin_unlock(&vc->lock);
2852                 /* cancel pending dec exception if dec is positive */
2853                 if (now < vcpu->arch.dec_expires &&
2854                     kvmppc_core_pending_dec(vcpu))
2855                         kvmppc_core_dequeue_dec(vcpu);
2856 
2857                 trace_kvm_guest_exit(vcpu);
2858 
2859                 ret = RESUME_GUEST;
2860                 if (vcpu->arch.trap)
2861                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2862                                                     vcpu->arch.run_task);
2863 
2864                 vcpu->arch.ret = ret;
2865                 vcpu->arch.trap = 0;
2866 
2867                 spin_lock(&vc->lock);
2868                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2869                         if (vcpu->arch.pending_exceptions)
2870                                 kvmppc_core_prepare_to_enter(vcpu);
2871                         if (vcpu->arch.ceded)
2872                                 kvmppc_set_timer(vcpu);
2873                         else
2874                                 ++still_running;
2875                 } else {
2876                         kvmppc_remove_runnable(vc, vcpu);
2877                         wake_up(&vcpu->arch.cpu_run);
2878                 }
2879         }
2880         if (!is_master) {
2881                 if (still_running > 0) {
2882                         kvmppc_vcore_preempt(vc);
2883                 } else if (vc->runner) {
2884                         vc->vcore_state = VCORE_PREEMPT;
2885                         kvmppc_core_start_stolen(vc);
2886                 } else {
2887                         vc->vcore_state = VCORE_INACTIVE;
2888                 }
2889                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2890                         /* make sure there's a candidate runner awake */
2891                         i = -1;
2892                         vcpu = next_runnable_thread(vc, &i);
2893                         wake_up(&vcpu->arch.cpu_run);
2894                 }
2895         }
2896         spin_unlock(&vc->lock);
2897 }
2898 
2899 /*
2900  * Clear core from the list of active host cores as we are about to
2901  * enter the guest. Only do this if it is the primary thread of the
2902  * core (not if a subcore) that is entering the guest.
2903  */
2904 static inline int kvmppc_clear_host_core(unsigned int cpu)
2905 {
2906         int core;
2907 
2908         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2909                 return 0;
2910         /*
2911          * Memory barrier can be omitted here as we will do a smp_wmb()
2912          * later in kvmppc_start_thread and we need ensure that state is
2913          * visible to other CPUs only after we enter guest.
2914          */
2915         core = cpu >> threads_shift;
2916         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2917         return 0;
2918 }
2919 
2920 /*
2921  * Advertise this core as an active host core since we exited the guest
2922  * Only need to do this if it is the primary thread of the core that is
2923  * exiting.
2924  */
2925 static inline int kvmppc_set_host_core(unsigned int cpu)
2926 {
2927         int core;
2928 
2929         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2930                 return 0;
2931 
2932         /*
2933          * Memory barrier can be omitted here because we do a spin_unlock
2934          * immediately after this which provides the memory barrier.
2935          */
2936         core = cpu >> threads_shift;
2937         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2938         return 0;
2939 }
2940 
2941 static void set_irq_happened(int trap)
2942 {
2943         switch (trap) {
2944         case BOOK3S_INTERRUPT_EXTERNAL:
2945                 local_paca->irq_happened |= PACA_IRQ_EE;
2946                 break;
2947         case BOOK3S_INTERRUPT_H_DOORBELL:
2948                 local_paca->irq_happened |= PACA_IRQ_DBELL;
2949                 break;
2950         case BOOK3S_INTERRUPT_HMI:
2951                 local_paca->irq_happened |= PACA_IRQ_HMI;
2952                 break;
2953         case BOOK3S_INTERRUPT_SYSTEM_RESET:
2954                 replay_system_reset();
2955                 break;
2956         }
2957 }
2958 
2959 /*
2960  * Run a set of guest threads on a physical core.
2961  * Called with vc->lock held.
2962  */
2963 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2964 {
2965         struct kvm_vcpu *vcpu;
2966         int i;
2967         int srcu_idx;
2968         struct core_info core_info;
2969         struct kvmppc_vcore *pvc;
2970         struct kvm_split_mode split_info, *sip;
2971         int split, subcore_size, active;
2972         int sub;
2973         bool thr0_done;
2974         unsigned long cmd_bit, stat_bit;
2975         int pcpu, thr;
2976         int target_threads;
2977         int controlled_threads;
2978         int trap;
2979         bool is_power8;
2980         bool hpt_on_radix;
2981 
2982         /*
2983          * Remove from the list any threads that have a signal pending
2984          * or need a VPA update done
2985          */
2986         prepare_threads(vc);
2987 
2988         /* if the runner is no longer runnable, let the caller pick a new one */
2989         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2990                 return;
2991 
2992         /*
2993          * Initialize *vc.
2994          */
2995         init_vcore_to_run(vc);
2996         vc->preempt_tb = TB_NIL;
2997 
2998         /*
2999          * Number of threads that we will be controlling: the same as
3000          * the number of threads per subcore, except on POWER9,
3001          * where it's 1 because the threads are (mostly) independent.
3002          */
3003         controlled_threads = threads_per_vcore(vc->kvm);
3004 
3005         /*
3006          * Make sure we are running on primary threads, and that secondary
3007          * threads are offline.  Also check if the number of threads in this
3008          * guest are greater than the current system threads per guest.
3009          * On POWER9, we need to be not in independent-threads mode if
3010          * this is a HPT guest on a radix host machine where the
3011          * CPU threads may not be in different MMU modes.
3012          */
3013         hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3014                 !kvm_is_radix(vc->kvm);
3015         if (((controlled_threads > 1) &&
3016              ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3017             (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3018                 for_each_runnable_thread(i, vcpu, vc) {
3019                         vcpu->arch.ret = -EBUSY;
3020                         kvmppc_remove_runnable(vc, vcpu);
3021                         wake_up(&vcpu->arch.cpu_run);
3022                 }
3023                 goto out;
3024         }
3025 
3026         /*
3027          * See if we could run any other vcores on the physical core
3028          * along with this one.
3029          */
3030         init_core_info(&core_info, vc);
3031         pcpu = smp_processor_id();
3032         target_threads = controlled_threads;
3033         if (target_smt_mode && target_smt_mode < target_threads)
3034                 target_threads = target_smt_mode;
3035         if (vc->num_threads < target_threads)
3036                 collect_piggybacks(&core_info, target_threads);
3037 
3038         /*
3039          * On radix, arrange for TLB flushing if necessary.
3040          * This has to be done before disabling interrupts since
3041          * it uses smp_call_function().
3042          */
3043         pcpu = smp_processor_id();
3044         if (kvm_is_radix(vc->kvm)) {
3045                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3046                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3047                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3048         }
3049 
3050         /*
3051          * Hard-disable interrupts, and check resched flag and signals.
3052          * If we need to reschedule or deliver a signal, clean up
3053          * and return without going into the guest(s).
3054          * If the mmu_ready flag has been cleared, don't go into the
3055          * guest because that means a HPT resize operation is in progress.
3056          */
3057         local_irq_disable();
3058         hard_irq_disable();
3059         if (lazy_irq_pending() || need_resched() ||
3060             recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3061                 local_irq_enable();
3062                 vc->vcore_state = VCORE_INACTIVE;
3063                 /* Unlock all except the primary vcore */
3064                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3065                         pvc = core_info.vc[sub];
3066                         /* Put back on to the preempted vcores list */
3067                         kvmppc_vcore_preempt(pvc);
3068                         spin_unlock(&pvc->lock);
3069                 }
3070                 for (i = 0; i < controlled_threads; ++i)
3071                         kvmppc_release_hwthread(pcpu + i);
3072                 return;
3073         }
3074 
3075         kvmppc_clear_host_core(pcpu);
3076 
3077         /* Decide on micro-threading (split-core) mode */
3078         subcore_size = threads_per_subcore;
3079         cmd_bit = stat_bit = 0;
3080         split = core_info.n_subcores;
3081         sip = NULL;
3082         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3083                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3084 
3085         if (split > 1 || hpt_on_radix) {
3086                 sip = &split_info;
3087                 memset(&split_info, 0, sizeof(split_info));
3088                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3089                         split_info.vc[sub] = core_info.vc[sub];
3090 
3091                 if (is_power8) {
3092                         if (split == 2 && (dynamic_mt_modes & 2)) {
3093                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3094                                 stat_bit = HID0_POWER8_2LPARMODE;
3095                         } else {
3096                                 split = 4;
3097                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3098                                 stat_bit = HID0_POWER8_4LPARMODE;
3099                         }
3100                         subcore_size = MAX_SMT_THREADS / split;
3101                         split_info.rpr = mfspr(SPRN_RPR);
3102                         split_info.pmmar = mfspr(SPRN_PMMAR);
3103                         split_info.ldbar = mfspr(SPRN_LDBAR);
3104                         split_info.subcore_size = subcore_size;
3105                 } else {
3106                         split_info.subcore_size = 1;
3107                         if (hpt_on_radix) {
3108                                 /* Use the split_info for LPCR/LPIDR changes */
3109                                 split_info.lpcr_req = vc->lpcr;
3110                                 split_info.lpidr_req = vc->kvm->arch.lpid;
3111                                 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3112                                 split_info.do_set = 1;
3113                         }
3114                 }
3115 
3116                 /* order writes to split_info before kvm_split_mode pointer */
3117                 smp_wmb();
3118         }
3119 
3120         for (thr = 0; thr < controlled_threads; ++thr) {
3121                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3122 
3123                 paca->kvm_hstate.tid = thr;
3124                 paca->kvm_hstate.napping = 0;
3125                 paca->kvm_hstate.kvm_split_mode = sip;
3126         }
3127 
3128         /* Initiate micro-threading (split-core) on POWER8 if required */
3129         if (cmd_bit) {
3130                 unsigned long hid0 = mfspr(SPRN_HID0);
3131 
3132                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3133                 mb();
3134                 mtspr(SPRN_HID0, hid0);
3135                 isync();
3136                 for (;;) {
3137                         hid0 = mfspr(SPRN_HID0);
3138                         if (hid0 & stat_bit)
3139                                 break;
3140                         cpu_relax();
3141                 }
3142         }
3143 
3144         /*
3145          * On POWER8, set RWMR register.
3146          * Since it only affects PURR and SPURR, it doesn't affect
3147          * the host, so we don't save/restore the host value.
3148          */
3149         if (is_power8) {
3150                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3151                 int n_online = atomic_read(&vc->online_count);
3152 
3153                 /*
3154                  * Use the 8-thread value if we're doing split-core
3155                  * or if the vcore's online count looks bogus.
3156                  */
3157                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3158                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3159                         rwmr_val = p8_rwmr_values[n_online];
3160                 mtspr(SPRN_RWMR, rwmr_val);
3161         }
3162 
3163         /* Start all the threads */
3164         active = 0;
3165         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3166                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3167                 thr0_done = false;
3168                 active |= 1 << thr;
3169                 pvc = core_info.vc[sub];
3170                 pvc->pcpu = pcpu + thr;
3171                 for_each_runnable_thread(i, vcpu, pvc) {
3172                         kvmppc_start_thread(vcpu, pvc);
3173                         kvmppc_create_dtl_entry(vcpu, pvc);
3174                         trace_kvm_guest_enter(vcpu);
3175                         if (!vcpu->arch.ptid)
3176                                 thr0_done = true;
3177                         active |= 1 << (thr + vcpu->arch.ptid);
3178                 }
3179                 /*
3180                  * We need to start the first thread of each subcore
3181                  * even if it doesn't have a vcpu.
3182                  */
3183                 if (!thr0_done)
3184                         kvmppc_start_thread(NULL, pvc);
3185         }
3186 
3187         /*
3188          * Ensure that split_info.do_nap is set after setting
3189          * the vcore pointer in the PACA of the secondaries.
3190          */
3191         smp_mb();
3192 
3193         /*
3194          * When doing micro-threading, poke the inactive threads as well.
3195          * This gets them to the nap instruction after kvm_do_nap,
3196          * which reduces the time taken to unsplit later.
3197          * For POWER9 HPT guest on radix host, we need all the secondary
3198          * threads woken up so they can do the LPCR/LPIDR change.
3199          */
3200         if (cmd_bit || hpt_on_radix) {
3201                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3202                 for (thr = 1; thr < threads_per_subcore; ++thr)
3203                         if (!(active & (1 << thr)))
3204                                 kvmppc_ipi_thread(pcpu + thr);
3205         }
3206 
3207         vc->vcore_state = VCORE_RUNNING;
3208         preempt_disable();
3209 
3210         trace_kvmppc_run_core(vc, 0);
3211 
3212         for (sub = 0; sub < core_info.n_subcores; ++sub)
3213                 spin_unlock(&core_info.vc[sub]->lock);
3214 
3215         if (kvm_is_radix(vc->kvm)) {
3216                 /*
3217                  * Do we need to flush the process scoped TLB for the LPAR?
3218                  *
3219                  * On POWER9, individual threads can come in here, but the
3220                  * TLB is shared between the 4 threads in a core, hence
3221                  * invalidating on one thread invalidates for all.
3222                  * Thus we make all 4 threads use the same bit here.
3223                  *
3224                  * Hash must be flushed in realmode in order to use tlbiel.
3225                  */
3226                 kvmppc_radix_check_need_tlb_flush(vc->kvm, pcpu, NULL);
3227         }
3228 
3229         /*
3230          * Interrupts will be enabled once we get into the guest,
3231          * so tell lockdep that we're about to enable interrupts.
3232          */
3233         trace_hardirqs_on();
3234 
3235         guest_enter_irqoff();
3236 
3237         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3238 
3239         this_cpu_disable_ftrace();
3240 
3241         trap = __kvmppc_vcore_entry();
3242 
3243         this_cpu_enable_ftrace();
3244 
3245         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3246 
3247         trace_hardirqs_off();
3248         set_irq_happened(trap);
3249 
3250         spin_lock(&vc->lock);
3251         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3252         vc->vcore_state = VCORE_EXITING;
3253 
3254         /* wait for secondary threads to finish writing their state to memory */
3255         kvmppc_wait_for_nap(controlled_threads);
3256 
3257         /* Return to whole-core mode if we split the core earlier */
3258         if (cmd_bit) {
3259                 unsigned long hid0 = mfspr(SPRN_HID0);
3260                 unsigned long loops = 0;
3261 
3262                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3263                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3264                 mb();
3265                 mtspr(SPRN_HID0, hid0);
3266                 isync();
3267                 for (;;) {
3268                         hid0 = mfspr(SPRN_HID0);
3269                         if (!(hid0 & stat_bit))
3270                                 break;
3271                         cpu_relax();
3272                         ++loops;
3273                 }
3274         } else if (hpt_on_radix) {
3275                 /* Wait for all threads to have seen final sync */
3276                 for (thr = 1; thr < controlled_threads; ++thr) {
3277                         struct paca_struct *paca = paca_ptrs[pcpu + thr];
3278 
3279                         while (paca->kvm_hstate.kvm_split_mode) {
3280                                 HMT_low();
3281                                 barrier();
3282                         }
3283                         HMT_medium();
3284                 }
3285         }
3286         split_info.do_nap = 0;
3287 
3288         kvmppc_set_host_core(pcpu);
3289 
3290         local_irq_enable();
3291         guest_exit();
3292 
3293         /* Let secondaries go back to the offline loop */
3294         for (i = 0; i < controlled_threads; ++i) {
3295                 kvmppc_release_hwthread(pcpu + i);
3296                 if (sip && sip->napped[i])
3297                         kvmppc_ipi_thread(pcpu + i);
3298                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3299         }
3300 
3301         spin_unlock(&vc->lock);
3302 
3303         /* make sure updates to secondary vcpu structs are visible now */
3304         smp_mb();
3305 
3306         preempt_enable();
3307 
3308         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3309                 pvc = core_info.vc[sub];
3310                 post_guest_process(pvc, pvc == vc);
3311         }
3312 
3313         spin_lock(&vc->lock);
3314 
3315  out:
3316         vc->vcore_state = VCORE_INACTIVE;
3317         trace_kvmppc_run_core(vc, 1);
3318 }
3319 
3320 /*
3321  * Load up hypervisor-mode registers on P9.
3322  */
3323 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3324                                      unsigned long lpcr)
3325 {
3326         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3327         s64 hdec;
3328         u64 tb, purr, spurr;
3329         int trap;
3330         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3331         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3332         unsigned long host_dawr = mfspr(SPRN_DAWR);
3333         unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3334         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3335         unsigned long host_pidr = mfspr(SPRN_PID);
3336 
3337         hdec = time_limit - mftb();
3338         if (hdec < 0)
3339                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3340         mtspr(SPRN_HDEC, hdec);
3341 
3342         if (vc->tb_offset) {
3343                 u64 new_tb = mftb() + vc->tb_offset;
3344                 mtspr(SPRN_TBU40, new_tb);
3345                 tb = mftb();
3346                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3347                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3348                 vc->tb_offset_applied = vc->tb_offset;
3349         }
3350 
3351         if (vc->pcr)
3352                 mtspr(SPRN_PCR, vc->pcr);
3353         mtspr(SPRN_DPDES, vc->dpdes);
3354         mtspr(SPRN_VTB, vc->vtb);
3355 
3356         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3357         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3358         mtspr(SPRN_PURR, vcpu->arch.purr);
3359         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3360 
3361         if (cpu_has_feature(CPU_FTR_DAWR)) {
3362                 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3363                 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3364         }
3365         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3366         mtspr(SPRN_IC, vcpu->arch.ic);
3367         mtspr(SPRN_PID, vcpu->arch.pid);
3368 
3369         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3370               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3371 
3372         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3373 
3374         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3375         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3376         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3377         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3378 
3379         mtspr(SPRN_AMOR, ~0UL);
3380 
3381         mtspr(SPRN_LPCR, lpcr);
3382         isync();
3383 
3384         kvmppc_xive_push_vcpu(vcpu);
3385 
3386         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3387         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3388 
3389         trap = __kvmhv_vcpu_entry_p9(vcpu);
3390 
3391         /* Advance host PURR/SPURR by the amount used by guest */
3392         purr = mfspr(SPRN_PURR);
3393         spurr = mfspr(SPRN_SPURR);
3394         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3395               purr - vcpu->arch.purr);
3396         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3397               spurr - vcpu->arch.spurr);
3398         vcpu->arch.purr = purr;
3399         vcpu->arch.spurr = spurr;
3400 
3401         vcpu->arch.ic = mfspr(SPRN_IC);
3402         vcpu->arch.pid = mfspr(SPRN_PID);
3403         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3404 
3405         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3406         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3407         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3408         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3409 
3410         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3411         mtspr(SPRN_PSSCR, host_psscr |
3412               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3413         mtspr(SPRN_HFSCR, host_hfscr);
3414         mtspr(SPRN_CIABR, host_ciabr);
3415         mtspr(SPRN_DAWR, host_dawr);
3416         mtspr(SPRN_DAWRX, host_dawrx);
3417         mtspr(SPRN_PID, host_pidr);
3418 
3419         /*
3420          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3421          * case we interrupted the guest between a tlbie and a ptesync.
3422          */
3423         asm volatile("eieio; tlbsync; ptesync");
3424 
3425         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3426         isync();
3427 
3428         vc->dpdes = mfspr(SPRN_DPDES);
3429         vc->vtb = mfspr(SPRN_VTB);
3430         mtspr(SPRN_DPDES, 0);
3431         if (vc->pcr)
3432                 mtspr(SPRN_PCR, 0);
3433 
3434         if (vc->tb_offset_applied) {
3435                 u64 new_tb = mftb() - vc->tb_offset_applied;
3436                 mtspr(SPRN_TBU40, new_tb);
3437                 tb = mftb();
3438                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3439                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3440                 vc->tb_offset_applied = 0;
3441         }
3442 
3443         mtspr(SPRN_HDEC, 0x7fffffff);
3444         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3445 
3446         return trap;
3447 }
3448 
3449 /*
3450  * Virtual-mode guest entry for POWER9 and later when the host and
3451  * guest are both using the radix MMU.  The LPIDR has already been set.
3452  */
3453 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3454                          unsigned long lpcr)
3455 {
3456         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3457         unsigned long host_dscr = mfspr(SPRN_DSCR);
3458         unsigned long host_tidr = mfspr(SPRN_TIDR);
3459         unsigned long host_iamr = mfspr(SPRN_IAMR);
3460         s64 dec;
3461         u64 tb;
3462         int trap, save_pmu;
3463 
3464         dec = mfspr(SPRN_DEC);
3465         tb = mftb();
3466         if (dec < 512)
3467                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3468         local_paca->kvm_hstate.dec_expires = dec + tb;
3469         if (local_paca->kvm_hstate.dec_expires < time_limit)
3470                 time_limit = local_paca->kvm_hstate.dec_expires;
3471 
3472         vcpu->arch.ceded = 0;
3473 
3474         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3475 
3476         kvmppc_subcore_enter_guest();
3477 
3478         vc->entry_exit_map = 1;
3479         vc->in_guest = 1;
3480 
3481         if (vcpu->arch.vpa.pinned_addr) {
3482                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3483                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3484                 lp->yield_count = cpu_to_be32(yield_count);
3485                 vcpu->arch.vpa.dirty = 1;
3486         }
3487 
3488         if (cpu_has_feature(CPU_FTR_TM) ||
3489             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3490                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3491 
3492         kvmhv_load_guest_pmu(vcpu);
3493 
3494         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3495         load_fp_state(&vcpu->arch.fp);
3496 #ifdef CONFIG_ALTIVEC
3497         load_vr_state(&vcpu->arch.vr);
3498 #endif
3499 
3500         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3501         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3502         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3503         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3504         mtspr(SPRN_TAR, vcpu->arch.tar);
3505         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3506         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3507         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3508         mtspr(SPRN_WORT, vcpu->arch.wort);
3509         mtspr(SPRN_TIDR, vcpu->arch.tid);
3510         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3511         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3512         mtspr(SPRN_AMR, vcpu->arch.amr);
3513         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3514 
3515         if (!(vcpu->arch.ctrl & 1))
3516                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3517 
3518         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3519 
3520         if (kvmhv_on_pseries()) {
3521                 /* call our hypervisor to load up HV regs and go */
3522                 struct hv_guest_state hvregs;
3523 
3524                 kvmhv_save_hv_regs(vcpu, &hvregs);
3525                 hvregs.lpcr = lpcr;
3526                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3527                 hvregs.version = HV_GUEST_STATE_VERSION;
3528                 if (vcpu->arch.nested) {
3529                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3530                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3531                 } else {
3532                         hvregs.lpid = vcpu->kvm->arch.lpid;
3533                         hvregs.vcpu_token = vcpu->vcpu_id;
3534                 }
3535                 hvregs.hdec_expiry = time_limit;
3536                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3537                                           __pa(&vcpu->arch.regs));
3538                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3539                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3540                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3541                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3542 
3543                 /* H_CEDE has to be handled now, not later */
3544                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3545                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3546                         kvmppc_nested_cede(vcpu);
3547                         trap = 0;
3548                 }
3549         } else {
3550                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3551         }
3552 
3553         vcpu->arch.slb_max = 0;
3554         dec = mfspr(SPRN_DEC);
3555         tb = mftb();
3556         vcpu->arch.dec_expires = dec + tb;
3557         vcpu->cpu = -1;
3558         vcpu->arch.thread_cpu = -1;
3559         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3560 
3561         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3562         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3563         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3564         vcpu->arch.tar = mfspr(SPRN_TAR);
3565         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3566         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3567         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3568         vcpu->arch.wort = mfspr(SPRN_WORT);
3569         vcpu->arch.tid = mfspr(SPRN_TIDR);
3570         vcpu->arch.amr = mfspr(SPRN_AMR);
3571         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3572         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3573 
3574         mtspr(SPRN_PSPB, 0);
3575         mtspr(SPRN_WORT, 0);
3576         mtspr(SPRN_AMR, 0);
3577         mtspr(SPRN_UAMOR, 0);
3578         mtspr(SPRN_DSCR, host_dscr);
3579         mtspr(SPRN_TIDR, host_tidr);
3580         mtspr(SPRN_IAMR, host_iamr);
3581         mtspr(SPRN_PSPB, 0);
3582 
3583         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3584         store_fp_state(&vcpu->arch.fp);
3585 #ifdef CONFIG_ALTIVEC
3586         store_vr_state(&vcpu->arch.vr);
3587 #endif
3588 
3589         if (cpu_has_feature(CPU_FTR_TM) ||
3590             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3591                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3592 
3593         save_pmu = 1;
3594         if (vcpu->arch.vpa.pinned_addr) {
3595                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3596                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3597                 lp->yield_count = cpu_to_be32(yield_count);
3598                 vcpu->arch.vpa.dirty = 1;
3599                 save_pmu = lp->pmcregs_in_use;
3600         }
3601 
3602         kvmhv_save_guest_pmu(vcpu, save_pmu);
3603 
3604         vc->entry_exit_map = 0x101;
3605         vc->in_guest = 0;
3606 
3607         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3608 
3609         kvmhv_load_host_pmu();
3610 
3611         kvmppc_subcore_exit_guest();
3612 
3613         return trap;
3614 }
3615 
3616 /*
3617  * Wait for some other vcpu thread to execute us, and
3618  * wake us up when we need to handle something in the host.
3619  */
3620 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3621                                  struct kvm_vcpu *vcpu, int wait_state)
3622 {
3623         DEFINE_WAIT(wait);
3624 
3625         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3626         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3627                 spin_unlock(&vc->lock);
3628                 schedule();
3629                 spin_lock(&vc->lock);
3630         }
3631         finish_wait(&vcpu->arch.cpu_run, &wait);
3632 }
3633 
3634 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3635 {
3636         /* 10us base */
3637         if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3638                 vc->halt_poll_ns = 10000;
3639         else
3640                 vc->halt_poll_ns *= halt_poll_ns_grow;
3641 }
3642 
3643 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3644 {
3645         if (halt_poll_ns_shrink == 0)
3646                 vc->halt_poll_ns = 0;
3647         else
3648                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3649 }
3650 
3651 #ifdef CONFIG_KVM_XICS
3652 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3653 {
3654         if (!xive_enabled())
3655                 return false;
3656         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3657                 vcpu->arch.xive_saved_state.cppr;
3658 }
3659 #else
3660 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3661 {
3662         return false;
3663 }
3664 #endif /* CONFIG_KVM_XICS */
3665 
3666 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3667 {
3668         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3669             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3670                 return true;
3671 
3672         return false;
3673 }
3674 
3675 /*
3676  * Check to see if any of the runnable vcpus on the vcore have pending
3677  * exceptions or are no longer ceded
3678  */
3679 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3680 {
3681         struct kvm_vcpu *vcpu;
3682         int i;
3683 
3684         for_each_runnable_thread(i, vcpu, vc) {
3685                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3686                         return 1;
3687         }
3688 
3689         return 0;
3690 }
3691 
3692 /*
3693  * All the vcpus in this vcore are idle, so wait for a decrementer
3694  * or external interrupt to one of the vcpus.  vc->lock is held.
3695  */
3696 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3697 {
3698         ktime_t cur, start_poll, start_wait;
3699         int do_sleep = 1;
3700         u64 block_ns;
3701         DECLARE_SWAITQUEUE(wait);
3702 
3703         /* Poll for pending exceptions and ceded state */
3704         cur = start_poll = ktime_get();
3705         if (vc->halt_poll_ns) {
3706                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3707                 ++vc->runner->stat.halt_attempted_poll;
3708 
3709                 vc->vcore_state = VCORE_POLLING;
3710                 spin_unlock(&vc->lock);
3711 
3712                 do {
3713                         if (kvmppc_vcore_check_block(vc)) {
3714                                 do_sleep = 0;
3715                                 break;
3716                         }
3717                         cur = ktime_get();
3718                 } while (single_task_running() && ktime_before(cur, stop));
3719 
3720                 spin_lock(&vc->lock);
3721                 vc->vcore_state = VCORE_INACTIVE;
3722 
3723                 if (!do_sleep) {
3724                         ++vc->runner->stat.halt_successful_poll;
3725                         goto out;
3726                 }
3727         }
3728 
3729         prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3730 
3731         if (kvmppc_vcore_check_block(vc)) {
3732                 finish_swait(&vc->wq, &wait);
3733                 do_sleep = 0;
3734                 /* If we polled, count this as a successful poll */
3735                 if (vc->halt_poll_ns)
3736                         ++vc->runner->stat.halt_successful_poll;
3737                 goto out;
3738         }
3739 
3740         start_wait = ktime_get();
3741 
3742         vc->vcore_state = VCORE_SLEEPING;
3743         trace_kvmppc_vcore_blocked(vc, 0);
3744         spin_unlock(&vc->lock);
3745         schedule();
3746         finish_swait(&vc->wq, &wait);
3747         spin_lock(&vc->lock);
3748         vc->vcore_state = VCORE_INACTIVE;
3749         trace_kvmppc_vcore_blocked(vc, 1);
3750         ++vc->runner->stat.halt_successful_wait;
3751 
3752         cur = ktime_get();
3753 
3754 out:
3755         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3756 
3757         /* Attribute wait time */
3758         if (do_sleep) {
3759                 vc->runner->stat.halt_wait_ns +=
3760                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3761                 /* Attribute failed poll time */
3762                 if (vc->halt_poll_ns)
3763                         vc->runner->stat.halt_poll_fail_ns +=
3764                                 ktime_to_ns(start_wait) -
3765                                 ktime_to_ns(start_poll);
3766         } else {
3767                 /* Attribute successful poll time */
3768                 if (vc->halt_poll_ns)
3769                         vc->runner->stat.halt_poll_success_ns +=
3770                                 ktime_to_ns(cur) -
3771                                 ktime_to_ns(start_poll);
3772         }
3773 
3774         /* Adjust poll time */
3775         if (halt_poll_ns) {
3776                 if (block_ns <= vc->halt_poll_ns)
3777                         ;
3778                 /* We slept and blocked for longer than the max halt time */
3779                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3780                         shrink_halt_poll_ns(vc);
3781                 /* We slept and our poll time is too small */
3782                 else if (vc->halt_poll_ns < halt_poll_ns &&
3783                                 block_ns < halt_poll_ns)
3784                         grow_halt_poll_ns(vc);
3785                 if (vc->halt_poll_ns > halt_poll_ns)
3786                         vc->halt_poll_ns = halt_poll_ns;
3787         } else
3788                 vc->halt_poll_ns = 0;
3789 
3790         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3791 }
3792 
3793 /*
3794  * This never fails for a radix guest, as none of the operations it does
3795  * for a radix guest can fail or have a way to report failure.
3796  * kvmhv_run_single_vcpu() relies on this fact.
3797  */
3798 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3799 {
3800         int r = 0;
3801         struct kvm *kvm = vcpu->kvm;
3802 
3803         mutex_lock(&kvm->lock);
3804         if (!kvm->arch.mmu_ready) {
3805                 if (!kvm_is_radix(kvm))
3806                         r = kvmppc_hv_setup_htab_rma(vcpu);
3807                 if (!r) {
3808                         if (cpu_has_feature(CPU_FTR_ARCH_300))
3809                                 kvmppc_setup_partition_table(kvm);
3810                         kvm->arch.mmu_ready = 1;
3811                 }
3812         }
3813         mutex_unlock(&kvm->lock);
3814         return r;
3815 }
3816 
3817 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3818 {
3819         int n_ceded, i, r;
3820         struct kvmppc_vcore *vc;
3821         struct kvm_vcpu *v;
3822 
3823         trace_kvmppc_run_vcpu_enter(vcpu);
3824 
3825         kvm_run->exit_reason = 0;
3826         vcpu->arch.ret = RESUME_GUEST;
3827         vcpu->arch.trap = 0;
3828         kvmppc_update_vpas(vcpu);
3829 
3830         /*
3831          * Synchronize with other threads in this virtual core
3832          */
3833         vc = vcpu->arch.vcore;
3834         spin_lock(&vc->lock);
3835         vcpu->arch.ceded = 0;
3836         vcpu->arch.run_task = current;
3837         vcpu->arch.kvm_run = kvm_run;
3838         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3839         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3840         vcpu->arch.busy_preempt = TB_NIL;
3841         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3842         ++vc->n_runnable;
3843 
3844         /*
3845          * This happens the first time this is called for a vcpu.
3846          * If the vcore is already running, we may be able to start
3847          * this thread straight away and have it join in.
3848          */
3849         if (!signal_pending(current)) {
3850                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3851                      vc->vcore_state == VCORE_RUNNING) &&
3852                            !VCORE_IS_EXITING(vc)) {
3853                         kvmppc_create_dtl_entry(vcpu, vc);
3854                         kvmppc_start_thread(vcpu, vc);
3855                         trace_kvm_guest_enter(vcpu);
3856                 } else if (vc->vcore_state == VCORE_SLEEPING) {
3857                         swake_up_one(&vc->wq);
3858                 }
3859 
3860         }
3861 
3862         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3863                !signal_pending(current)) {
3864                 /* See if the MMU is ready to go */
3865                 if (!vcpu->kvm->arch.mmu_ready) {
3866                         spin_unlock(&vc->lock);
3867                         r = kvmhv_setup_mmu(vcpu);
3868                         spin_lock(&vc->lock);
3869                         if (r) {
3870                                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3871                                 kvm_run->fail_entry.
3872                                         hardware_entry_failure_reason = 0;
3873                                 vcpu->arch.ret = r;
3874                                 break;
3875                         }
3876                 }
3877 
3878                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3879                         kvmppc_vcore_end_preempt(vc);
3880 
3881                 if (vc->vcore_state != VCORE_INACTIVE) {
3882                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3883                         continue;
3884                 }
3885                 for_each_runnable_thread(i, v, vc) {
3886                         kvmppc_core_prepare_to_enter(v);
3887                         if (signal_pending(v->arch.run_task)) {
3888                                 kvmppc_remove_runnable(vc, v);
3889                                 v->stat.signal_exits++;
3890                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3891                                 v->arch.ret = -EINTR;
3892                                 wake_up(&v->arch.cpu_run);
3893                         }
3894                 }
3895                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3896                         break;
3897                 n_ceded = 0;
3898                 for_each_runnable_thread(i, v, vc) {
3899                         if (!kvmppc_vcpu_woken(v))
3900                                 n_ceded += v->arch.ceded;
3901                         else
3902                                 v->arch.ceded = 0;
3903                 }
3904                 vc->runner = vcpu;
3905                 if (n_ceded == vc->n_runnable) {
3906                         kvmppc_vcore_blocked(vc);
3907                 } else if (need_resched()) {
3908                         kvmppc_vcore_preempt(vc);
3909                         /* Let something else run */
3910                         cond_resched_lock(&vc->lock);
3911                         if (vc->vcore_state == VCORE_PREEMPT)
3912                                 kvmppc_vcore_end_preempt(vc);
3913                 } else {
3914                         kvmppc_run_core(vc);
3915                 }
3916                 vc->runner = NULL;
3917         }
3918 
3919         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3920                (vc->vcore_state == VCORE_RUNNING ||
3921                 vc->vcore_state == VCORE_EXITING ||
3922                 vc->vcore_state == VCORE_PIGGYBACK))
3923                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3924 
3925         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3926                 kvmppc_vcore_end_preempt(vc);
3927 
3928         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3929                 kvmppc_remove_runnable(vc, vcpu);
3930                 vcpu->stat.signal_exits++;
3931                 kvm_run->exit_reason = KVM_EXIT_INTR;
3932                 vcpu->arch.ret = -EINTR;
3933         }
3934 
3935         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3936                 /* Wake up some vcpu to run the core */
3937                 i = -1;
3938                 v = next_runnable_thread(vc, &i);
3939                 wake_up(&v->arch.cpu_run);
3940         }
3941 
3942         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3943         spin_unlock(&vc->lock);
3944         return vcpu->arch.ret;
3945 }
3946 
3947 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
3948                           struct kvm_vcpu *vcpu, u64 time_limit,
3949                           unsigned long lpcr)
3950 {
3951         int trap, r, pcpu;
3952         int srcu_idx;
3953         struct kvmppc_vcore *vc;
3954         struct kvm *kvm = vcpu->kvm;
3955         struct kvm_nested_guest *nested = vcpu->arch.nested;
3956 
3957         trace_kvmppc_run_vcpu_enter(vcpu);
3958 
3959         kvm_run->exit_reason = 0;
3960         vcpu->arch.ret = RESUME_GUEST;
3961         vcpu->arch.trap = 0;
3962 
3963         vc = vcpu->arch.vcore;
3964         vcpu->arch.ceded = 0;
3965         vcpu->arch.run_task = current;
3966         vcpu->arch.kvm_run = kvm_run;
3967         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3968         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3969         vcpu->arch.busy_preempt = TB_NIL;
3970         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
3971         vc->runnable_threads[0] = vcpu;
3972         vc->n_runnable = 1;
3973         vc->runner = vcpu;
3974 
3975         /* See if the MMU is ready to go */
3976         if (!kvm->arch.mmu_ready)
3977                 kvmhv_setup_mmu(vcpu);
3978 
3979         if (need_resched())
3980                 cond_resched();
3981 
3982         kvmppc_update_vpas(vcpu);
3983 
3984         init_vcore_to_run(vc);
3985         vc->preempt_tb = TB_NIL;
3986 
3987         preempt_disable();
3988         pcpu = smp_processor_id();
3989         vc->pcpu = pcpu;
3990         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3991 
3992         local_irq_disable();
3993         hard_irq_disable();
3994         if (signal_pending(current))
3995                 goto sigpend;
3996         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
3997                 goto out;
3998 
3999         if (!nested) {
4000                 kvmppc_core_prepare_to_enter(vcpu);
4001                 if (vcpu->arch.doorbell_request) {
4002                         vc->dpdes = 1;
4003                         smp_wmb();
4004                         vcpu->arch.doorbell_request = 0;
4005                 }
4006                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4007                              &vcpu->arch.pending_exceptions))
4008                         lpcr |= LPCR_MER;
4009         } else if (vcpu->arch.pending_exceptions ||
4010                    vcpu->arch.doorbell_request ||
4011                    xive_interrupt_pending(vcpu)) {
4012                 vcpu->arch.ret = RESUME_HOST;
4013                 goto out;
4014         }
4015 
4016         kvmppc_clear_host_core(pcpu);
4017 
4018         local_paca->kvm_hstate.tid = 0;
4019         local_paca->kvm_hstate.napping = 0;
4020         local_paca->kvm_hstate.kvm_split_mode = NULL;
4021         kvmppc_start_thread(vcpu, vc);
4022         kvmppc_create_dtl_entry(vcpu, vc);
4023         trace_kvm_guest_enter(vcpu);
4024 
4025         vc->vcore_state = VCORE_RUNNING;
4026         trace_kvmppc_run_core(vc, 0);
4027 
4028         if (cpu_has_feature(CPU_FTR_HVMODE))
4029                 kvmppc_radix_check_need_tlb_flush(kvm, pcpu, nested);
4030 
4031         trace_hardirqs_on();
4032         guest_enter_irqoff();
4033 
4034         srcu_idx = srcu_read_lock(&kvm->srcu);
4035 
4036         this_cpu_disable_ftrace();
4037 
4038         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4039         vcpu->arch.trap = trap;
4040 
4041         this_cpu_enable_ftrace();
4042 
4043         srcu_read_unlock(&kvm->srcu, srcu_idx);
4044 
4045         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4046                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4047                 isync();
4048         }
4049 
4050         trace_hardirqs_off();
4051         set_irq_happened(trap);
4052 
4053         kvmppc_set_host_core(pcpu);
4054 
4055         local_irq_enable();
4056         guest_exit();
4057 
4058         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4059 
4060         preempt_enable();
4061 
4062         /* cancel pending decrementer exception if DEC is now positive */
4063         if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
4064                 kvmppc_core_dequeue_dec(vcpu);
4065 
4066         trace_kvm_guest_exit(vcpu);
4067         r = RESUME_GUEST;
4068         if (trap) {
4069                 if (!nested)
4070                         r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4071                 else
4072                         r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4073         }
4074         vcpu->arch.ret = r;
4075 
4076         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4077             !kvmppc_vcpu_woken(vcpu)) {
4078                 kvmppc_set_timer(vcpu);
4079                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4080                         if (signal_pending(current)) {
4081                                 vcpu->stat.signal_exits++;
4082                                 kvm_run->exit_reason = KVM_EXIT_INTR;
4083                                 vcpu->arch.ret = -EINTR;
4084                                 break;
4085                         }
4086                         spin_lock(&vc->lock);
4087                         kvmppc_vcore_blocked(vc);
4088                         spin_unlock(&vc->lock);
4089                 }
4090         }
4091         vcpu->arch.ceded = 0;
4092 
4093         vc->vcore_state = VCORE_INACTIVE;
4094         trace_kvmppc_run_core(vc, 1);
4095 
4096  done:
4097         kvmppc_remove_runnable(vc, vcpu);
4098         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4099 
4100         return vcpu->arch.ret;
4101 
4102  sigpend:
4103         vcpu->stat.signal_exits++;
4104         kvm_run->exit_reason = KVM_EXIT_INTR;
4105         vcpu->arch.ret = -EINTR;
4106  out:
4107         local_irq_enable();
4108         preempt_enable();
4109         goto done;
4110 }
4111 
4112 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4113 {
4114         int r;
4115         int srcu_idx;
4116         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4117         unsigned long user_tar = 0;
4118         unsigned int user_vrsave;
4119         struct kvm *kvm;
4120 
4121         if (!vcpu->arch.sane) {
4122                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4123                 return -EINVAL;
4124         }
4125 
4126         /*
4127          * Don't allow entry with a suspended transaction, because
4128          * the guest entry/exit code will lose it.
4129          * If the guest has TM enabled, save away their TM-related SPRs
4130          * (they will get restored by the TM unavailable interrupt).
4131          */
4132 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4133         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4134             (current->thread.regs->msr & MSR_TM)) {
4135                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4136                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4137                         run->fail_entry.hardware_entry_failure_reason = 0;
4138                         return -EINVAL;
4139                 }
4140                 /* Enable TM so we can read the TM SPRs */
4141                 mtmsr(mfmsr() | MSR_TM);
4142                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4143                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4144                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4145                 current->thread.regs->msr &= ~MSR_TM;
4146         }
4147 #endif
4148 
4149         /*
4150          * Force online to 1 for the sake of old userspace which doesn't
4151          * set it.
4152          */
4153         if (!vcpu->arch.online) {
4154                 atomic_inc(&vcpu->arch.vcore->online_count);
4155                 vcpu->arch.online = 1;
4156         }
4157 
4158         kvmppc_core_prepare_to_enter(vcpu);
4159 
4160         /* No need to go into the guest when all we'll do is come back out */
4161         if (signal_pending(current)) {
4162                 run->exit_reason = KVM_EXIT_INTR;
4163                 return -EINTR;
4164         }
4165 
4166         kvm = vcpu->kvm;
4167         atomic_inc(&kvm->arch.vcpus_running);
4168         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4169         smp_mb();
4170 
4171         flush_all_to_thread(current);
4172 
4173         /* Save userspace EBB and other register values */
4174         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4175                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4176                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4177                 ebb_regs[2] = mfspr(SPRN_BESCR);
4178                 user_tar = mfspr(SPRN_TAR);
4179         }
4180         user_vrsave = mfspr(SPRN_VRSAVE);
4181 
4182         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4183         vcpu->arch.pgdir = current->mm->pgd;
4184         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4185 
4186         do {
4187                 /*
4188                  * The early POWER9 chips that can't mix radix and HPT threads
4189                  * on the same core also need the workaround for the problem
4190                  * where the TLB would prefetch entries in the guest exit path
4191                  * for radix guests using the guest PIDR value and LPID 0.
4192                  * The workaround is in the old path (kvmppc_run_vcpu())
4193                  * but not the new path (kvmhv_run_single_vcpu()).
4194                  */
4195                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4196                     !no_mixing_hpt_and_radix)
4197                         r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4198                                                   vcpu->arch.vcore->lpcr);
4199                 else
4200                         r = kvmppc_run_vcpu(run, vcpu);
4201 
4202                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4203                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4204                         trace_kvm_hcall_enter(vcpu);
4205                         r = kvmppc_pseries_do_hcall(vcpu);
4206                         trace_kvm_hcall_exit(vcpu, r);
4207                         kvmppc_core_prepare_to_enter(vcpu);
4208                 } else if (r == RESUME_PAGE_FAULT) {
4209                         srcu_idx = srcu_read_lock(&kvm->srcu);
4210                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
4211                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4212                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4213                 } else if (r == RESUME_PASSTHROUGH) {
4214                         if (WARN_ON(xive_enabled()))
4215                                 r = H_SUCCESS;
4216                         else
4217                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4218                 }
4219         } while (is_kvmppc_resume_guest(r));
4220 
4221         /* Restore userspace EBB and other register values */
4222         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4223                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4224                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4225                 mtspr(SPRN_BESCR, ebb_regs[2]);
4226                 mtspr(SPRN_TAR, user_tar);
4227                 mtspr(SPRN_FSCR, current->thread.fscr);
4228         }
4229         mtspr(SPRN_VRSAVE, user_vrsave);
4230 
4231         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4232         atomic_dec(&kvm->arch.vcpus_running);
4233         return r;
4234 }
4235 
4236 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4237                                      int shift, int sllp)
4238 {
4239         (*sps)->page_shift = shift;
4240         (*sps)->slb_enc = sllp;
4241         (*sps)->enc[0].page_shift = shift;
4242         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4243         /*
4244          * Add 16MB MPSS support (may get filtered out by userspace)
4245          */
4246         if (shift != 24) {
4247                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4248                 if (penc != -1) {
4249                         (*sps)->enc[1].page_shift = 24;
4250                         (*sps)->enc[1].pte_enc = penc;
4251                 }
4252         }
4253         (*sps)++;
4254 }
4255 
4256 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4257                                          struct kvm_ppc_smmu_info *info)
4258 {
4259         struct kvm_ppc_one_seg_page_size *sps;
4260 
4261         /*
4262          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4263          * POWER7 doesn't support keys for instruction accesses,
4264          * POWER8 and POWER9 do.
4265          */
4266         info->data_keys = 32;
4267         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4268 
4269         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4270         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4271         info->slb_size = 32;
4272 
4273         /* We only support these sizes for now, and no muti-size segments */
4274         sps = &info->sps[0];
4275         kvmppc_add_seg_page_size(&sps, 12, 0);
4276         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4277         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4278 
4279         /* If running as a nested hypervisor, we don't support HPT guests */
4280         if (kvmhv_on_pseries())
4281                 info->flags |= KVM_PPC_NO_HASH;
4282 
4283         return 0;
4284 }
4285 
4286 /*
4287  * Get (and clear) the dirty memory log for a memory slot.
4288  */
4289 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4290                                          struct kvm_dirty_log *log)
4291 {
4292         struct kvm_memslots *slots;
4293         struct kvm_memory_slot *memslot;
4294         int i, r;
4295         unsigned long n;
4296         unsigned long *buf, *p;
4297         struct kvm_vcpu *vcpu;
4298 
4299         mutex_lock(&kvm->slots_lock);
4300 
4301         r = -EINVAL;
4302         if (log->slot >= KVM_USER_MEM_SLOTS)
4303                 goto out;
4304 
4305         slots = kvm_memslots(kvm);
4306         memslot = id_to_memslot(slots, log->slot);
4307         r = -ENOENT;
4308         if (!memslot->dirty_bitmap)
4309                 goto out;
4310 
4311         /*
4312          * Use second half of bitmap area because both HPT and radix
4313          * accumulate bits in the first half.
4314          */
4315         n = kvm_dirty_bitmap_bytes(memslot);
4316         buf = memslot->dirty_bitmap + n / sizeof(long);
4317         memset(buf, 0, n);
4318 
4319         if (kvm_is_radix(kvm))
4320                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4321         else
4322                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4323         if (r)
4324                 goto out;
4325 
4326         /*
4327          * We accumulate dirty bits in the first half of the
4328          * memslot's dirty_bitmap area, for when pages are paged
4329          * out or modified by the host directly.  Pick up these
4330          * bits and add them to the map.
4331          */
4332         p = memslot->dirty_bitmap;
4333         for (i = 0; i < n / sizeof(long); ++i)
4334                 buf[i] |= xchg(&p[i], 0);
4335 
4336         /* Harvest dirty bits from VPA and DTL updates */
4337         /* Note: we never modify the SLB shadow buffer areas */
4338         kvm_for_each_vcpu(i, vcpu, kvm) {
4339                 spin_lock(&vcpu->arch.vpa_update_lock);
4340                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4341                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4342                 spin_unlock(&vcpu->arch.vpa_update_lock);
4343         }
4344 
4345         r = -EFAULT;
4346         if (copy_to_user(log->dirty_bitmap, buf, n))
4347                 goto out;
4348 
4349         r = 0;
4350 out:
4351         mutex_unlock(&kvm->slots_lock);
4352         return r;
4353 }
4354 
4355 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4356                                         struct kvm_memory_slot *dont)
4357 {
4358         if (!dont || free->arch.rmap != dont->arch.rmap) {
4359                 vfree(free->arch.rmap);
4360                 free->arch.rmap = NULL;
4361         }
4362 }
4363 
4364 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4365                                          unsigned long npages)
4366 {
4367         slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4368         if (!slot->arch.rmap)
4369                 return -ENOMEM;
4370 
4371         return 0;
4372 }
4373 
4374 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4375                                         struct kvm_memory_slot *memslot,
4376                                         const struct kvm_userspace_memory_region *mem)
4377 {
4378         return 0;
4379 }
4380 
4381 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4382                                 const struct kvm_userspace_memory_region *mem,
4383                                 const struct kvm_memory_slot *old,
4384                                 const struct kvm_memory_slot *new,
4385                                 enum kvm_mr_change change)
4386 {
4387         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4388 
4389         /*
4390          * If we are making a new memslot, it might make
4391          * some address that was previously cached as emulated
4392          * MMIO be no longer emulated MMIO, so invalidate
4393          * all the caches of emulated MMIO translations.
4394          */
4395         if (npages)
4396                 atomic64_inc(&kvm->arch.mmio_update);
4397 
4398         /*
4399          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4400          * have already called kvm_arch_flush_shadow_memslot() to
4401          * flush shadow mappings.  For KVM_MR_CREATE we have no
4402          * previous mappings.  So the only case to handle is
4403          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4404          * has been changed.
4405          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4406          * to get rid of any THP PTEs in the partition-scoped page tables
4407          * so we can track dirtiness at the page level; we flush when
4408          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4409          * using THP PTEs.
4410          */
4411         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4412             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4413                 kvmppc_radix_flush_memslot(kvm, old);
4414 }
4415 
4416 /*
4417  * Update LPCR values in kvm->arch and in vcores.
4418  * Caller must hold kvm->lock.
4419  */
4420 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4421 {
4422         long int i;
4423         u32 cores_done = 0;
4424 
4425         if ((kvm->arch.lpcr & mask) == lpcr)
4426                 return;
4427 
4428         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4429 
4430         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4431                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4432                 if (!vc)
4433                         continue;
4434                 spin_lock(&vc->lock);
4435                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4436                 spin_unlock(&vc->lock);
4437                 if (++cores_done >= kvm->arch.online_vcores)
4438                         break;
4439         }
4440 }
4441 
4442 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4443 {
4444         return;
4445 }
4446 
4447 void kvmppc_setup_partition_table(struct kvm *kvm)
4448 {
4449         unsigned long dw0, dw1;
4450 
4451         if (!kvm_is_radix(kvm)) {
4452                 /* PS field - page size for VRMA */
4453                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4454                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4455                 /* HTABSIZE and HTABORG fields */
4456                 dw0 |= kvm->arch.sdr1;
4457 
4458                 /* Second dword as set by userspace */
4459                 dw1 = kvm->arch.process_table;
4460         } else {
4461                 dw0 = PATB_HR | radix__get_tree_size() |
4462                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4463                 dw1 = PATB_GR | kvm->arch.process_table;
4464         }
4465         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4466 }
4467 
4468 /*
4469  * Set up HPT (hashed page table) and RMA (real-mode area).
4470  * Must be called with kvm->lock held.
4471  */
4472 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4473 {
4474         int err = 0;
4475         struct kvm *kvm = vcpu->kvm;
4476         unsigned long hva;
4477         struct kvm_memory_slot *memslot;
4478         struct vm_area_struct *vma;
4479         unsigned long lpcr = 0, senc;
4480         unsigned long psize, porder;
4481         int srcu_idx;
4482 
4483         /* Allocate hashed page table (if not done already) and reset it */
4484         if (!kvm->arch.hpt.virt) {
4485                 int order = KVM_DEFAULT_HPT_ORDER;
4486                 struct kvm_hpt_info info;
4487 
4488                 err = kvmppc_allocate_hpt(&info, order);
4489                 /* If we get here, it means userspace didn't specify a
4490                  * size explicitly.  So, try successively smaller
4491                  * sizes if the default failed. */
4492                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4493                         err  = kvmppc_allocate_hpt(&info, order);
4494 
4495                 if (err < 0) {
4496                         pr_err("KVM: Couldn't alloc HPT\n");
4497                         goto out;
4498                 }
4499 
4500                 kvmppc_set_hpt(kvm, &info);
4501         }
4502 
4503         /* Look up the memslot for guest physical address 0 */
4504         srcu_idx = srcu_read_lock(&kvm->srcu);
4505         memslot = gfn_to_memslot(kvm, 0);
4506 
4507         /* We must have some memory at 0 by now */
4508         err = -EINVAL;
4509         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4510                 goto out_srcu;
4511 
4512         /* Look up the VMA for the start of this memory slot */
4513         hva = memslot->userspace_addr;
4514         down_read(&current->mm->mmap_sem);
4515         vma = find_vma(current->mm, hva);
4516         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4517                 goto up_out;
4518 
4519         psize = vma_kernel_pagesize(vma);
4520 
4521         up_read(&current->mm->mmap_sem);
4522 
4523         /* We can handle 4k, 64k or 16M pages in the VRMA */
4524         if (psize >= 0x1000000)
4525                 psize = 0x1000000;
4526         else if (psize >= 0x10000)
4527                 psize = 0x10000;
4528         else
4529                 psize = 0x1000;
4530         porder = __ilog2(psize);
4531 
4532         senc = slb_pgsize_encoding(psize);
4533         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4534                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4535         /* Create HPTEs in the hash page table for the VRMA */
4536         kvmppc_map_vrma(vcpu, memslot, porder);
4537 
4538         /* Update VRMASD field in the LPCR */
4539         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4540                 /* the -4 is to account for senc values starting at 0x10 */
4541                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4542                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4543         }
4544 
4545         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4546         smp_wmb();
4547         err = 0;
4548  out_srcu:
4549         srcu_read_unlock(&kvm->srcu, srcu_idx);
4550  out:
4551         return err;
4552 
4553  up_out:
4554         up_read(&current->mm->mmap_sem);
4555         goto out_srcu;
4556 }
4557 
4558 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4559 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4560 {
4561         if (nesting_enabled(kvm))
4562                 kvmhv_release_all_nested(kvm);
4563         kvmppc_rmap_reset(kvm);
4564         kvm->arch.process_table = 0;
4565         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4566         spin_lock(&kvm->mmu_lock);
4567         kvm->arch.radix = 0;
4568         spin_unlock(&kvm->mmu_lock);
4569         kvmppc_free_radix(kvm);
4570         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4571                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4572         return 0;
4573 }
4574 
4575 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4576 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4577 {
4578         int err;
4579 
4580         err = kvmppc_init_vm_radix(kvm);
4581         if (err)
4582                 return err;
4583         kvmppc_rmap_reset(kvm);
4584         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4585         spin_lock(&kvm->mmu_lock);
4586         kvm->arch.radix = 1;
4587         spin_unlock(&kvm->mmu_lock);
4588         kvmppc_free_hpt(&kvm->arch.hpt);
4589         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4590                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4591         return 0;
4592 }
4593 
4594 #ifdef CONFIG_KVM_XICS
4595 /*
4596  * Allocate a per-core structure for managing state about which cores are
4597  * running in the host versus the guest and for exchanging data between
4598  * real mode KVM and CPU running in the host.
4599  * This is only done for the first VM.
4600  * The allocated structure stays even if all VMs have stopped.
4601  * It is only freed when the kvm-hv module is unloaded.
4602  * It's OK for this routine to fail, we just don't support host
4603  * core operations like redirecting H_IPI wakeups.
4604  */
4605 void kvmppc_alloc_host_rm_ops(void)
4606 {
4607         struct kvmppc_host_rm_ops *ops;
4608         unsigned long l_ops;
4609         int cpu, core;
4610         int size;
4611 
4612         /* Not the first time here ? */
4613         if (kvmppc_host_rm_ops_hv != NULL)
4614                 return;
4615 
4616         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4617         if (!ops)
4618                 return;
4619 
4620         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4621         ops->rm_core = kzalloc(size, GFP_KERNEL);
4622 
4623         if (!ops->rm_core) {
4624                 kfree(ops);
4625                 return;
4626         }
4627 
4628         cpus_read_lock();
4629 
4630         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4631                 if (!cpu_online(cpu))
4632                         continue;
4633 
4634                 core = cpu >> threads_shift;
4635                 ops->rm_core[core].rm_state.in_host = 1;
4636         }
4637 
4638         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4639 
4640         /*
4641          * Make the contents of the kvmppc_host_rm_ops structure visible
4642          * to other CPUs before we assign it to the global variable.
4643          * Do an atomic assignment (no locks used here), but if someone
4644          * beats us to it, just free our copy and return.
4645          */
4646         smp_wmb();
4647         l_ops = (unsigned long) ops;
4648 
4649         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4650                 cpus_read_unlock();
4651                 kfree(ops->rm_core);
4652                 kfree(ops);
4653                 return;
4654         }
4655 
4656         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4657                                              "ppc/kvm_book3s:prepare",
4658                                              kvmppc_set_host_core,
4659                                              kvmppc_clear_host_core);
4660         cpus_read_unlock();
4661 }
4662 
4663 void kvmppc_free_host_rm_ops(void)
4664 {
4665         if (kvmppc_host_rm_ops_hv) {
4666                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4667                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4668                 kfree(kvmppc_host_rm_ops_hv);
4669                 kvmppc_host_rm_ops_hv = NULL;
4670         }
4671 }
4672 #endif
4673 
4674 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4675 {
4676         unsigned long lpcr, lpid;
4677         char buf[32];
4678         int ret;
4679 
4680         /* Allocate the guest's logical partition ID */
4681 
4682         lpid = kvmppc_alloc_lpid();
4683         if ((long)lpid < 0)
4684                 return -ENOMEM;
4685         kvm->arch.lpid = lpid;
4686 
4687         kvmppc_alloc_host_rm_ops();
4688 
4689         kvmhv_vm_nested_init(kvm);
4690 
4691         /*
4692          * Since we don't flush the TLB when tearing down a VM,
4693          * and this lpid might have previously been used,
4694          * make sure we flush on each core before running the new VM.
4695          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4696          * does this flush for us.
4697          */
4698         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4699                 cpumask_setall(&kvm->arch.need_tlb_flush);
4700 
4701         /* Start out with the default set of hcalls enabled */
4702         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4703                sizeof(kvm->arch.enabled_hcalls));
4704 
4705         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4706                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4707 
4708         /* Init LPCR for virtual RMA mode */
4709         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4710                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4711                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4712                 lpcr &= LPCR_PECE | LPCR_LPES;
4713         } else {
4714                 lpcr = 0;
4715         }
4716         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4717                 LPCR_VPM0 | LPCR_VPM1;
4718         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4719                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4720         /* On POWER8 turn on online bit to enable PURR/SPURR */
4721         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4722                 lpcr |= LPCR_ONL;
4723         /*
4724          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4725          * Set HVICE bit to enable hypervisor virtualization interrupts.
4726          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4727          * be unnecessary but better safe than sorry in case we re-enable
4728          * EE in HV mode with this LPCR still set)
4729          */
4730         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4731                 lpcr &= ~LPCR_VPM0;
4732                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4733 
4734                 /*
4735                  * If xive is enabled, we route 0x500 interrupts directly
4736                  * to the guest.
4737                  */
4738                 if (xive_enabled())
4739                         lpcr |= LPCR_LPES;
4740         }
4741 
4742         /*
4743          * If the host uses radix, the guest starts out as radix.
4744          */
4745         if (radix_enabled()) {
4746                 kvm->arch.radix = 1;
4747                 kvm->arch.mmu_ready = 1;
4748                 lpcr &= ~LPCR_VPM1;
4749                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4750                 ret = kvmppc_init_vm_radix(kvm);
4751                 if (ret) {
4752                         kvmppc_free_lpid(kvm->arch.lpid);
4753                         return ret;
4754                 }
4755                 kvmppc_setup_partition_table(kvm);
4756         }
4757 
4758         kvm->arch.lpcr = lpcr;
4759 
4760         /* Initialization for future HPT resizes */
4761         kvm->arch.resize_hpt = NULL;
4762 
4763         /*
4764          * Work out how many sets the TLB has, for the use of
4765          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4766          */
4767         if (radix_enabled())
4768                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
4769         else if (cpu_has_feature(CPU_FTR_ARCH_300))
4770                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
4771         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4772                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
4773         else
4774                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
4775 
4776         /*
4777          * Track that we now have a HV mode VM active. This blocks secondary
4778          * CPU threads from coming online.
4779          * On POWER9, we only need to do this if the "indep_threads_mode"
4780          * module parameter has been set to N.
4781          */
4782         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4783                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4784                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4785                         kvm->arch.threads_indep = true;
4786                 } else {
4787                         kvm->arch.threads_indep = indep_threads_mode;
4788                 }
4789         }
4790         if (!kvm->arch.threads_indep)
4791                 kvm_hv_vm_activated();
4792 
4793         /*
4794          * Initialize smt_mode depending on processor.
4795          * POWER8 and earlier have to use "strict" threading, where
4796          * all vCPUs in a vcore have to run on the same (sub)core,
4797          * whereas on POWER9 the threads can each run a different
4798          * guest.
4799          */
4800         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4801                 kvm->arch.smt_mode = threads_per_subcore;
4802         else
4803                 kvm->arch.smt_mode = 1;
4804         kvm->arch.emul_smt_mode = 1;
4805 
4806         /*
4807          * Create a debugfs directory for the VM
4808          */
4809         snprintf(buf, sizeof(buf), "vm%d", current->pid);
4810         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4811         kvmppc_mmu_debugfs_init(kvm);
4812         if (radix_enabled())
4813                 kvmhv_radix_debugfs_init(kvm);
4814 
4815         return 0;
4816 }
4817 
4818 static void kvmppc_free_vcores(struct kvm *kvm)
4819 {
4820         long int i;
4821 
4822         for (i = 0; i < KVM_MAX_VCORES; ++i)
4823                 kfree(kvm->arch.vcores[i]);
4824         kvm->arch.online_vcores = 0;
4825 }
4826 
4827 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4828 {
4829         debugfs_remove_recursive(kvm->arch.debugfs_dir);
4830 
4831         if (!kvm->arch.threads_indep)
4832                 kvm_hv_vm_deactivated();
4833 
4834         kvmppc_free_vcores(kvm);
4835 
4836 
4837         if (kvm_is_radix(kvm))
4838                 kvmppc_free_radix(kvm);
4839         else
4840                 kvmppc_free_hpt(&kvm->arch.hpt);
4841 
4842         /* Perform global invalidation and return lpid to the pool */
4843         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4844                 if (nesting_enabled(kvm))
4845                         kvmhv_release_all_nested(kvm);
4846                 kvm->arch.process_table = 0;
4847                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4848         }
4849         kvmppc_free_lpid(kvm->arch.lpid);
4850 
4851         kvmppc_free_pimap(kvm);
4852 }
4853 
4854 /* We don't need to emulate any privileged instructions or dcbz */
4855 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4856                                      unsigned int inst, int *advance)
4857 {
4858         return EMULATE_FAIL;
4859 }
4860 
4861 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4862                                         ulong spr_val)
4863 {
4864         return EMULATE_FAIL;
4865 }
4866 
4867 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4868                                         ulong *spr_val)
4869 {
4870         return EMULATE_FAIL;
4871 }
4872 
4873 static int kvmppc_core_check_processor_compat_hv(void)
4874 {
4875         if (cpu_has_feature(CPU_FTR_HVMODE) &&
4876             cpu_has_feature(CPU_FTR_ARCH_206))
4877                 return 0;
4878 
4879         /* POWER9 in radix mode is capable of being a nested hypervisor. */
4880         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4881                 return 0;
4882 
4883         return -EIO;
4884 }
4885 
4886 #ifdef CONFIG_KVM_XICS
4887 
4888 void kvmppc_free_pimap(struct kvm *kvm)
4889 {
4890         kfree(kvm->arch.pimap);
4891 }
4892 
4893 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4894 {
4895         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4896 }
4897 
4898 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4899 {
4900         struct irq_desc *desc;
4901         struct kvmppc_irq_map *irq_map;
4902         struct kvmppc_passthru_irqmap *pimap;
4903         struct irq_chip *chip;
4904         int i, rc = 0;
4905 
4906         if (!kvm_irq_bypass)
4907                 return 1;
4908 
4909         desc = irq_to_desc(host_irq);
4910         if (!desc)
4911                 return -EIO;
4912 
4913         mutex_lock(&kvm->lock);
4914 
4915         pimap = kvm->arch.pimap;
4916         if (pimap == NULL) {
4917                 /* First call, allocate structure to hold IRQ map */
4918                 pimap = kvmppc_alloc_pimap();
4919                 if (pimap == NULL) {
4920                         mutex_unlock(&kvm->lock);
4921                         return -ENOMEM;
4922                 }
4923                 kvm->arch.pimap = pimap;
4924         }
4925 
4926         /*
4927          * For now, we only support interrupts for which the EOI operation
4928          * is an OPAL call followed by a write to XIRR, since that's
4929          * what our real-mode EOI code does, or a XIVE interrupt
4930          */
4931         chip = irq_data_get_irq_chip(&desc->irq_data);
4932         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4933                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4934                         host_irq, guest_gsi);
4935                 mutex_unlock(&kvm->lock);
4936                 return -ENOENT;
4937         }
4938 
4939         /*
4940          * See if we already have an entry for this guest IRQ number.
4941          * If it's mapped to a hardware IRQ number, that's an error,
4942          * otherwise re-use this entry.
4943          */
4944         for (i = 0; i < pimap->n_mapped; i++) {
4945                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4946                         if (pimap->mapped[i].r_hwirq) {
4947                                 mutex_unlock(&kvm->lock);
4948                                 return -EINVAL;
4949                         }
4950                         break;
4951                 }
4952         }
4953 
4954         if (i == KVMPPC_PIRQ_MAPPED) {
4955                 mutex_unlock(&kvm->lock);
4956                 return -EAGAIN;         /* table is full */
4957         }
4958 
4959         irq_map = &pimap->mapped[i];
4960 
4961         irq_map->v_hwirq = guest_gsi;
4962         irq_map->desc = desc;
4963 
4964         /*
4965          * Order the above two stores before the next to serialize with
4966          * the KVM real mode handler.
4967          */
4968         smp_wmb();
4969         irq_map->r_hwirq = desc->irq_data.hwirq;
4970 
4971         if (i == pimap->n_mapped)
4972                 pimap->n_mapped++;
4973 
4974         if (xive_enabled())
4975                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4976         else
4977                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4978         if (rc)
4979                 irq_map->r_hwirq = 0;
4980 
4981         mutex_unlock(&kvm->lock);
4982 
4983         return 0;
4984 }
4985 
4986 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4987 {
4988         struct irq_desc *desc;
4989         struct kvmppc_passthru_irqmap *pimap;
4990         int i, rc = 0;
4991 
4992         if (!kvm_irq_bypass)
4993                 return 0;
4994 
4995         desc = irq_to_desc(host_irq);
4996         if (!desc)
4997                 return -EIO;
4998 
4999         mutex_lock(&kvm->lock);
5000         if (!kvm->arch.pimap)
5001                 goto unlock;
5002 
5003         pimap = kvm->arch.pimap;
5004 
5005         for (i = 0; i < pimap->n_mapped; i++) {
5006                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5007                         break;
5008         }
5009 
5010         if (i == pimap->n_mapped) {
5011                 mutex_unlock(&kvm->lock);
5012                 return -ENODEV;
5013         }
5014 
5015         if (xive_enabled())
5016                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5017         else
5018                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5019 
5020         /* invalidate the entry (what do do on error from the above ?) */
5021         pimap->mapped[i].r_hwirq = 0;
5022 
5023         /*
5024          * We don't free this structure even when the count goes to
5025          * zero. The structure is freed when we destroy the VM.
5026          */
5027  unlock:
5028         mutex_unlock(&kvm->lock);
5029         return rc;
5030 }
5031 
5032 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5033                                              struct irq_bypass_producer *prod)
5034 {
5035         int ret = 0;
5036         struct kvm_kernel_irqfd *irqfd =
5037                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5038 
5039         irqfd->producer = prod;
5040 
5041         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5042         if (ret)
5043                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5044                         prod->irq, irqfd->gsi, ret);
5045 
5046         return ret;
5047 }
5048 
5049 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5050                                               struct irq_bypass_producer *prod)
5051 {
5052         int ret;
5053         struct kvm_kernel_irqfd *irqfd =
5054                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5055 
5056         irqfd->producer = NULL;
5057 
5058         /*
5059          * When producer of consumer is unregistered, we change back to
5060          * default external interrupt handling mode - KVM real mode
5061          * will switch back to host.
5062          */
5063         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5064         if (ret)
5065                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5066                         prod->irq, irqfd->gsi, ret);
5067 }
5068 #endif
5069 
5070 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5071                                  unsigned int ioctl, unsigned long arg)
5072 {
5073         struct kvm *kvm __maybe_unused = filp->private_data;
5074         void __user *argp = (void __user *)arg;
5075         long r;
5076 
5077         switch (ioctl) {
5078 
5079         case KVM_PPC_ALLOCATE_HTAB: {
5080                 u32 htab_order;
5081 
5082                 r = -EFAULT;
5083                 if (get_user(htab_order, (u32 __user *)argp))
5084                         break;
5085                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5086                 if (r)
5087                         break;
5088                 r = 0;
5089                 break;
5090         }
5091 
5092         case KVM_PPC_GET_HTAB_FD: {
5093                 struct kvm_get_htab_fd ghf;
5094 
5095                 r = -EFAULT;
5096                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5097                         break;
5098                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5099                 break;
5100         }
5101 
5102         case KVM_PPC_RESIZE_HPT_PREPARE: {
5103                 struct kvm_ppc_resize_hpt rhpt;
5104 
5105                 r = -EFAULT;
5106                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5107                         break;
5108 
5109                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5110                 break;
5111         }
5112 
5113         case KVM_PPC_RESIZE_HPT_COMMIT: {
5114                 struct kvm_ppc_resize_hpt rhpt;
5115 
5116                 r = -EFAULT;
5117                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5118                         break;
5119 
5120                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5121                 break;
5122         }
5123 
5124         default:
5125                 r = -ENOTTY;
5126         }
5127 
5128         return r;
5129 }
5130 
5131 /*
5132  * List of hcall numbers to enable by default.
5133  * For compatibility with old userspace, we enable by default
5134  * all hcalls that were implemented before the hcall-enabling
5135  * facility was added.  Note this list should not include H_RTAS.
5136  */
5137 static unsigned int default_hcall_list[] = {
5138         H_REMOVE,
5139         H_ENTER,
5140         H_READ,
5141         H_PROTECT,
5142         H_BULK_REMOVE,
5143         H_GET_TCE,
5144         H_PUT_TCE,
5145         H_SET_DABR,
5146         H_SET_XDABR,
5147         H_CEDE,
5148         H_PROD,
5149         H_CONFER,
5150         H_REGISTER_VPA,
5151 #ifdef CONFIG_KVM_XICS
5152         H_EOI,
5153         H_CPPR,
5154         H_IPI,
5155         H_IPOLL,
5156         H_XIRR,
5157         H_XIRR_X,
5158 #endif
5159         0
5160 };
5161 
5162 static void init_default_hcalls(void)
5163 {
5164         int i;
5165         unsigned int hcall;
5166 
5167         for (i = 0; default_hcall_list[i]; ++i) {
5168                 hcall = default_hcall_list[i];
5169                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5170                 __set_bit(hcall / 4, default_enabled_hcalls);
5171         }
5172 }
5173 
5174 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5175 {
5176         unsigned long lpcr;
5177         int radix;
5178         int err;
5179 
5180         /* If not on a POWER9, reject it */
5181         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5182                 return -ENODEV;
5183 
5184         /* If any unknown flags set, reject it */
5185         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5186                 return -EINVAL;
5187 
5188         /* GR (guest radix) bit in process_table field must match */
5189         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5190         if (!!(cfg->process_table & PATB_GR) != radix)
5191                 return -EINVAL;
5192 
5193         /* Process table size field must be reasonable, i.e. <= 24 */
5194         if ((cfg->process_table & PRTS_MASK) > 24)
5195                 return -EINVAL;
5196 
5197         /* We can change a guest to/from radix now, if the host is radix */
5198         if (radix && !radix_enabled())
5199                 return -EINVAL;
5200 
5201         /* If we're a nested hypervisor, we currently only support radix */
5202         if (kvmhv_on_pseries() && !radix)
5203                 return -EINVAL;
5204 
5205         mutex_lock(&kvm->lock);
5206         if (radix != kvm_is_radix(kvm)) {
5207                 if (kvm->arch.mmu_ready) {
5208                         kvm->arch.mmu_ready = 0;
5209                         /* order mmu_ready vs. vcpus_running */
5210                         smp_mb();
5211                         if (atomic_read(&kvm->arch.vcpus_running)) {
5212                                 kvm->arch.mmu_ready = 1;
5213                                 err = -EBUSY;
5214                                 goto out_unlock;
5215                         }
5216                 }
5217                 if (radix)
5218                         err = kvmppc_switch_mmu_to_radix(kvm);
5219                 else
5220                         err = kvmppc_switch_mmu_to_hpt(kvm);
5221                 if (err)
5222                         goto out_unlock;
5223         }
5224 
5225         kvm->arch.process_table = cfg->process_table;
5226         kvmppc_setup_partition_table(kvm);
5227 
5228         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5229         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5230         err = 0;
5231 
5232  out_unlock:
5233         mutex_unlock(&kvm->lock);
5234         return err;
5235 }
5236 
5237 static int kvmhv_enable_nested(struct kvm *kvm)
5238 {
5239         if (!nested)
5240                 return -EPERM;
5241         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5242                 return -ENODEV;
5243 
5244         /* kvm == NULL means the caller is testing if the capability exists */
5245         if (kvm)
5246                 kvm->arch.nested_enable = true;
5247         return 0;
5248 }
5249 
5250 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5251                                  int size)
5252 {
5253         int rc = -EINVAL;
5254 
5255         if (kvmhv_vcpu_is_radix(vcpu)) {
5256                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5257 
5258                 if (rc > 0)
5259                         rc = -EINVAL;
5260         }
5261 
5262         /* For now quadrants are the only way to access nested guest memory */
5263         if (rc && vcpu->arch.nested)
5264                 rc = -EAGAIN;
5265 
5266         return rc;
5267 }
5268 
5269 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5270                                 int size)
5271 {
5272         int rc = -EINVAL;
5273 
5274         if (kvmhv_vcpu_is_radix(vcpu)) {
5275                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5276 
5277                 if (rc > 0)
5278                         rc = -EINVAL;
5279         }
5280 
5281         /* For now quadrants are the only way to access nested guest memory */
5282         if (rc && vcpu->arch.nested)
5283                 rc = -EAGAIN;
5284 
5285         return rc;
5286 }
5287 
5288 static struct kvmppc_ops kvm_ops_hv = {
5289         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5290         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5291         .get_one_reg = kvmppc_get_one_reg_hv,
5292         .set_one_reg = kvmppc_set_one_reg_hv,
5293         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5294         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5295         .set_msr     = kvmppc_set_msr_hv,
5296         .vcpu_run    = kvmppc_vcpu_run_hv,
5297         .vcpu_create = kvmppc_core_vcpu_create_hv,
5298         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5299         .check_requests = kvmppc_core_check_requests_hv,
5300         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5301         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5302         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5303         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5304         .unmap_hva_range = kvm_unmap_hva_range_hv,
5305         .age_hva  = kvm_age_hva_hv,
5306         .test_age_hva = kvm_test_age_hva_hv,
5307         .set_spte_hva = kvm_set_spte_hva_hv,
5308         .mmu_destroy  = kvmppc_mmu_destroy_hv,
5309         .free_memslot = kvmppc_core_free_memslot_hv,
5310         .create_memslot = kvmppc_core_create_memslot_hv,
5311         .init_vm =  kvmppc_core_init_vm_hv,
5312         .destroy_vm = kvmppc_core_destroy_vm_hv,
5313         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5314         .emulate_op = kvmppc_core_emulate_op_hv,
5315         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5316         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5317         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5318         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5319         .hcall_implemented = kvmppc_hcall_impl_hv,
5320 #ifdef CONFIG_KVM_XICS
5321         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5322         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5323 #endif
5324         .configure_mmu = kvmhv_configure_mmu,
5325         .get_rmmu_info = kvmhv_get_rmmu_info,
5326         .set_smt_mode = kvmhv_set_smt_mode,
5327         .enable_nested = kvmhv_enable_nested,
5328         .load_from_eaddr = kvmhv_load_from_eaddr,
5329         .store_to_eaddr = kvmhv_store_to_eaddr,
5330 };
5331 
5332 static int kvm_init_subcore_bitmap(void)
5333 {
5334         int i, j;
5335         int nr_cores = cpu_nr_cores();
5336         struct sibling_subcore_state *sibling_subcore_state;
5337 
5338         for (i = 0; i < nr_cores; i++) {
5339                 int first_cpu = i * threads_per_core;
5340                 int node = cpu_to_node(first_cpu);
5341 
5342                 /* Ignore if it is already allocated. */
5343                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5344                         continue;
5345 
5346                 sibling_subcore_state =
5347                         kmalloc_node(sizeof(struct sibling_subcore_state),
5348                                                         GFP_KERNEL, node);
5349                 if (!sibling_subcore_state)
5350                         return -ENOMEM;
5351 
5352                 memset(sibling_subcore_state, 0,
5353                                 sizeof(struct sibling_subcore_state));
5354 
5355                 for (j = 0; j < threads_per_core; j++) {
5356                         int cpu = first_cpu + j;
5357 
5358                         paca_ptrs[cpu]->sibling_subcore_state =
5359                                                 sibling_subcore_state;
5360                 }
5361         }
5362         return 0;
5363 }
5364 
5365 static int kvmppc_radix_possible(void)
5366 {
5367         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5368 }
5369 
5370 static int kvmppc_book3s_init_hv(void)
5371 {
5372         int r;
5373         /*
5374          * FIXME!! Do we need to check on all cpus ?
5375          */
5376         r = kvmppc_core_check_processor_compat_hv();
5377         if (r < 0)
5378                 return -ENODEV;
5379 
5380         r = kvmhv_nested_init();
5381         if (r)
5382                 return r;
5383 
5384         r = kvm_init_subcore_bitmap();
5385         if (r)
5386                 return r;
5387 
5388         /*
5389          * We need a way of accessing the XICS interrupt controller,
5390          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5391          * indirectly, via OPAL.
5392          */
5393 #ifdef CONFIG_SMP
5394         if (!xive_enabled() && !kvmhv_on_pseries() &&
5395             !local_paca->kvm_hstate.xics_phys) {
5396                 struct device_node *np;
5397 
5398                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5399                 if (!np) {
5400                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5401                         return -ENODEV;
5402                 }
5403                 /* presence of intc confirmed - node can be dropped again */
5404                 of_node_put(np);
5405         }
5406 #endif
5407 
5408         kvm_ops_hv.owner = THIS_MODULE;
5409         kvmppc_hv_ops = &kvm_ops_hv;
5410 
5411         init_default_hcalls();
5412 
5413         init_vcore_lists();
5414 
5415         r = kvmppc_mmu_hv_init();
5416         if (r)
5417                 return r;
5418 
5419         if (kvmppc_radix_possible())
5420                 r = kvmppc_radix_init();
5421 
5422         /*
5423          * POWER9 chips before version 2.02 can't have some threads in
5424          * HPT mode and some in radix mode on the same core.
5425          */
5426         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5427                 unsigned int pvr = mfspr(SPRN_PVR);
5428                 if ((pvr >> 16) == PVR_POWER9 &&
5429                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5430                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5431                         no_mixing_hpt_and_radix = true;
5432         }
5433 
5434         return r;
5435 }
5436 
5437 static void kvmppc_book3s_exit_hv(void)
5438 {
5439         kvmppc_free_host_rm_ops();
5440         if (kvmppc_radix_possible())
5441                 kvmppc_radix_exit();
5442         kvmppc_hv_ops = NULL;
5443         kvmhv_nested_exit();
5444 }
5445 
5446 module_init(kvmppc_book3s_init_hv);
5447 module_exit(kvmppc_book3s_exit_hv);
5448 MODULE_LICENSE("GPL");
5449 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5450 MODULE_ALIAS("devname:kvm");
5451 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp