xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision bb26cfd9)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
5  *
6  * Authors:
7  *    Paul Mackerras <paulus@au1.ibm.com>
8  *    Alexander Graf <agraf@suse.de>
9  *    Kevin Wolf <mail@kevin-wolf.de>
10  *
11  * Description: KVM functions specific to running on Book 3S
12  * processors in hypervisor mode (specifically POWER7 and later).
13  *
14  * This file is derived from arch/powerpc/kvm/book3s.c,
15  * by Alexander Graf <agraf@suse.de>.
16  */
17 
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 #include <linux/err.h>
21 #include <linux/slab.h>
22 #include <linux/preempt.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/stat.h>
25 #include <linux/delay.h>
26 #include <linux/export.h>
27 #include <linux/fs.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpu.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 #include <linux/gfp.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/hugetlb.h>
40 #include <linux/kvm_irqfd.h>
41 #include <linux/irqbypass.h>
42 #include <linux/module.h>
43 #include <linux/compiler.h>
44 #include <linux/of.h>
45 #include <linux/irqdomain.h>
46 
47 #include <asm/ftrace.h>
48 #include <asm/reg.h>
49 #include <asm/ppc-opcode.h>
50 #include <asm/asm-prototypes.h>
51 #include <asm/archrandom.h>
52 #include <asm/debug.h>
53 #include <asm/disassemble.h>
54 #include <asm/cputable.h>
55 #include <asm/cacheflush.h>
56 #include <linux/uaccess.h>
57 #include <asm/interrupt.h>
58 #include <asm/io.h>
59 #include <asm/kvm_ppc.h>
60 #include <asm/kvm_book3s.h>
61 #include <asm/mmu_context.h>
62 #include <asm/lppaca.h>
63 #include <asm/pmc.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 #include <asm/hw_breakpoint.h>
78 #include <asm/kvm_book3s_uvmem.h>
79 #include <asm/ultravisor.h>
80 #include <asm/dtl.h>
81 #include <asm/plpar_wrappers.h>
82 
83 #include "book3s.h"
84 #include "book3s_hv.h"
85 
86 #define CREATE_TRACE_POINTS
87 #include "trace_hv.h"
88 
89 /* #define EXIT_DEBUG */
90 /* #define EXIT_DEBUG_SIMPLE */
91 /* #define EXIT_DEBUG_INT */
92 
93 /* Used to indicate that a guest page fault needs to be handled */
94 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
95 /* Used to indicate that a guest passthrough interrupt needs to be handled */
96 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
97 
98 /* Used as a "null" value for timebase values */
99 #define TB_NIL	(~(u64)0)
100 
101 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
102 
103 static int dynamic_mt_modes = 6;
104 module_param(dynamic_mt_modes, int, 0644);
105 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
106 static int target_smt_mode;
107 module_param(target_smt_mode, int, 0644);
108 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
109 
110 static bool one_vm_per_core;
111 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
112 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
113 
114 #ifdef CONFIG_KVM_XICS
115 static const struct kernel_param_ops module_param_ops = {
116 	.set = param_set_int,
117 	.get = param_get_int,
118 };
119 
120 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
121 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
122 
123 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
124 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
125 #endif
126 
127 /* If set, guests are allowed to create and control nested guests */
128 static bool nested = true;
129 module_param(nested, bool, S_IRUGO | S_IWUSR);
130 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
131 
132 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
133 
134 /*
135  * RWMR values for POWER8.  These control the rate at which PURR
136  * and SPURR count and should be set according to the number of
137  * online threads in the vcore being run.
138  */
139 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
140 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
141 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
142 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
143 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
144 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
145 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
146 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
147 
148 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
149 	RWMR_RPA_P8_1THREAD,
150 	RWMR_RPA_P8_1THREAD,
151 	RWMR_RPA_P8_2THREAD,
152 	RWMR_RPA_P8_3THREAD,
153 	RWMR_RPA_P8_4THREAD,
154 	RWMR_RPA_P8_5THREAD,
155 	RWMR_RPA_P8_6THREAD,
156 	RWMR_RPA_P8_7THREAD,
157 	RWMR_RPA_P8_8THREAD,
158 };
159 
160 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
161 		int *ip)
162 {
163 	int i = *ip;
164 	struct kvm_vcpu *vcpu;
165 
166 	while (++i < MAX_SMT_THREADS) {
167 		vcpu = READ_ONCE(vc->runnable_threads[i]);
168 		if (vcpu) {
169 			*ip = i;
170 			return vcpu;
171 		}
172 	}
173 	return NULL;
174 }
175 
176 /* Used to traverse the list of runnable threads for a given vcore */
177 #define for_each_runnable_thread(i, vcpu, vc) \
178 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
179 
180 static bool kvmppc_ipi_thread(int cpu)
181 {
182 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
183 
184 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
185 	if (kvmhv_on_pseries())
186 		return false;
187 
188 	/* On POWER9 we can use msgsnd to IPI any cpu */
189 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
190 		msg |= get_hard_smp_processor_id(cpu);
191 		smp_mb();
192 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
193 		return true;
194 	}
195 
196 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
197 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
198 		preempt_disable();
199 		if (cpu_first_thread_sibling(cpu) ==
200 		    cpu_first_thread_sibling(smp_processor_id())) {
201 			msg |= cpu_thread_in_core(cpu);
202 			smp_mb();
203 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
204 			preempt_enable();
205 			return true;
206 		}
207 		preempt_enable();
208 	}
209 
210 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
211 	if (cpu >= 0 && cpu < nr_cpu_ids) {
212 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
213 			xics_wake_cpu(cpu);
214 			return true;
215 		}
216 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
217 		return true;
218 	}
219 #endif
220 
221 	return false;
222 }
223 
224 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
225 {
226 	int cpu;
227 	struct rcuwait *waitp;
228 
229 	/*
230 	 * rcuwait_wake_up contains smp_mb() which orders prior stores that
231 	 * create pending work vs below loads of cpu fields. The other side
232 	 * is the barrier in vcpu run that orders setting the cpu fields vs
233 	 * testing for pending work.
234 	 */
235 
236 	waitp = kvm_arch_vcpu_get_wait(vcpu);
237 	if (rcuwait_wake_up(waitp))
238 		++vcpu->stat.generic.halt_wakeup;
239 
240 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
241 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
242 		return;
243 
244 	/* CPU points to the first thread of the core */
245 	cpu = vcpu->cpu;
246 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
247 		smp_send_reschedule(cpu);
248 }
249 
250 /*
251  * We use the vcpu_load/put functions to measure stolen time.
252  * Stolen time is counted as time when either the vcpu is able to
253  * run as part of a virtual core, but the task running the vcore
254  * is preempted or sleeping, or when the vcpu needs something done
255  * in the kernel by the task running the vcpu, but that task is
256  * preempted or sleeping.  Those two things have to be counted
257  * separately, since one of the vcpu tasks will take on the job
258  * of running the core, and the other vcpu tasks in the vcore will
259  * sleep waiting for it to do that, but that sleep shouldn't count
260  * as stolen time.
261  *
262  * Hence we accumulate stolen time when the vcpu can run as part of
263  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
264  * needs its task to do other things in the kernel (for example,
265  * service a page fault) in busy_stolen.  We don't accumulate
266  * stolen time for a vcore when it is inactive, or for a vcpu
267  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
268  * a misnomer; it means that the vcpu task is not executing in
269  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
270  * the kernel.  We don't have any way of dividing up that time
271  * between time that the vcpu is genuinely stopped, time that
272  * the task is actively working on behalf of the vcpu, and time
273  * that the task is preempted, so we don't count any of it as
274  * stolen.
275  *
276  * Updates to busy_stolen are protected by arch.tbacct_lock;
277  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
278  * lock.  The stolen times are measured in units of timebase ticks.
279  * (Note that the != TB_NIL checks below are purely defensive;
280  * they should never fail.)
281  */
282 
283 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb)
284 {
285 	unsigned long flags;
286 
287 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
288 
289 	spin_lock_irqsave(&vc->stoltb_lock, flags);
290 	vc->preempt_tb = tb;
291 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
292 }
293 
294 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb)
295 {
296 	unsigned long flags;
297 
298 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
299 
300 	spin_lock_irqsave(&vc->stoltb_lock, flags);
301 	if (vc->preempt_tb != TB_NIL) {
302 		vc->stolen_tb += tb - vc->preempt_tb;
303 		vc->preempt_tb = TB_NIL;
304 	}
305 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
306 }
307 
308 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
309 {
310 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
311 	unsigned long flags;
312 	u64 now;
313 
314 	if (cpu_has_feature(CPU_FTR_ARCH_300))
315 		return;
316 
317 	now = mftb();
318 
319 	/*
320 	 * We can test vc->runner without taking the vcore lock,
321 	 * because only this task ever sets vc->runner to this
322 	 * vcpu, and once it is set to this vcpu, only this task
323 	 * ever sets it to NULL.
324 	 */
325 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
326 		kvmppc_core_end_stolen(vc, now);
327 
328 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
329 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
330 	    vcpu->arch.busy_preempt != TB_NIL) {
331 		vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt;
332 		vcpu->arch.busy_preempt = TB_NIL;
333 	}
334 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
335 }
336 
337 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
338 {
339 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
340 	unsigned long flags;
341 	u64 now;
342 
343 	if (cpu_has_feature(CPU_FTR_ARCH_300))
344 		return;
345 
346 	now = mftb();
347 
348 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
349 		kvmppc_core_start_stolen(vc, now);
350 
351 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
352 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
353 		vcpu->arch.busy_preempt = now;
354 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
355 }
356 
357 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
358 {
359 	vcpu->arch.pvr = pvr;
360 }
361 
362 /* Dummy value used in computing PCR value below */
363 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
364 
365 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
366 {
367 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
368 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
369 
370 	/* We can (emulate) our own architecture version and anything older */
371 	if (cpu_has_feature(CPU_FTR_ARCH_31))
372 		host_pcr_bit = PCR_ARCH_31;
373 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
374 		host_pcr_bit = PCR_ARCH_300;
375 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
376 		host_pcr_bit = PCR_ARCH_207;
377 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
378 		host_pcr_bit = PCR_ARCH_206;
379 	else
380 		host_pcr_bit = PCR_ARCH_205;
381 
382 	/* Determine lowest PCR bit needed to run guest in given PVR level */
383 	guest_pcr_bit = host_pcr_bit;
384 	if (arch_compat) {
385 		switch (arch_compat) {
386 		case PVR_ARCH_205:
387 			guest_pcr_bit = PCR_ARCH_205;
388 			break;
389 		case PVR_ARCH_206:
390 		case PVR_ARCH_206p:
391 			guest_pcr_bit = PCR_ARCH_206;
392 			break;
393 		case PVR_ARCH_207:
394 			guest_pcr_bit = PCR_ARCH_207;
395 			break;
396 		case PVR_ARCH_300:
397 			guest_pcr_bit = PCR_ARCH_300;
398 			break;
399 		case PVR_ARCH_31:
400 			guest_pcr_bit = PCR_ARCH_31;
401 			break;
402 		default:
403 			return -EINVAL;
404 		}
405 	}
406 
407 	/* Check requested PCR bits don't exceed our capabilities */
408 	if (guest_pcr_bit > host_pcr_bit)
409 		return -EINVAL;
410 
411 	spin_lock(&vc->lock);
412 	vc->arch_compat = arch_compat;
413 	/*
414 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
415 	 * Also set all reserved PCR bits
416 	 */
417 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
418 	spin_unlock(&vc->lock);
419 
420 	return 0;
421 }
422 
423 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
424 {
425 	int r;
426 
427 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
428 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
429 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
430 	for (r = 0; r < 16; ++r)
431 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
432 		       r, kvmppc_get_gpr(vcpu, r),
433 		       r+16, kvmppc_get_gpr(vcpu, r+16));
434 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
435 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
436 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
437 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
438 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
439 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
440 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
441 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
442 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
443 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
444 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
445 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
446 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
447 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
448 	for (r = 0; r < vcpu->arch.slb_max; ++r)
449 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
450 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
451 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
452 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
453 	       vcpu->arch.last_inst);
454 }
455 
456 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
457 {
458 	return kvm_get_vcpu_by_id(kvm, id);
459 }
460 
461 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
462 {
463 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
464 	vpa->yield_count = cpu_to_be32(1);
465 }
466 
467 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
468 		   unsigned long addr, unsigned long len)
469 {
470 	/* check address is cacheline aligned */
471 	if (addr & (L1_CACHE_BYTES - 1))
472 		return -EINVAL;
473 	spin_lock(&vcpu->arch.vpa_update_lock);
474 	if (v->next_gpa != addr || v->len != len) {
475 		v->next_gpa = addr;
476 		v->len = addr ? len : 0;
477 		v->update_pending = 1;
478 	}
479 	spin_unlock(&vcpu->arch.vpa_update_lock);
480 	return 0;
481 }
482 
483 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
484 struct reg_vpa {
485 	u32 dummy;
486 	union {
487 		__be16 hword;
488 		__be32 word;
489 	} length;
490 };
491 
492 static int vpa_is_registered(struct kvmppc_vpa *vpap)
493 {
494 	if (vpap->update_pending)
495 		return vpap->next_gpa != 0;
496 	return vpap->pinned_addr != NULL;
497 }
498 
499 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
500 				       unsigned long flags,
501 				       unsigned long vcpuid, unsigned long vpa)
502 {
503 	struct kvm *kvm = vcpu->kvm;
504 	unsigned long len, nb;
505 	void *va;
506 	struct kvm_vcpu *tvcpu;
507 	int err;
508 	int subfunc;
509 	struct kvmppc_vpa *vpap;
510 
511 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
512 	if (!tvcpu)
513 		return H_PARAMETER;
514 
515 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
516 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
517 	    subfunc == H_VPA_REG_SLB) {
518 		/* Registering new area - address must be cache-line aligned */
519 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
520 			return H_PARAMETER;
521 
522 		/* convert logical addr to kernel addr and read length */
523 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
524 		if (va == NULL)
525 			return H_PARAMETER;
526 		if (subfunc == H_VPA_REG_VPA)
527 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
528 		else
529 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
530 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
531 
532 		/* Check length */
533 		if (len > nb || len < sizeof(struct reg_vpa))
534 			return H_PARAMETER;
535 	} else {
536 		vpa = 0;
537 		len = 0;
538 	}
539 
540 	err = H_PARAMETER;
541 	vpap = NULL;
542 	spin_lock(&tvcpu->arch.vpa_update_lock);
543 
544 	switch (subfunc) {
545 	case H_VPA_REG_VPA:		/* register VPA */
546 		/*
547 		 * The size of our lppaca is 1kB because of the way we align
548 		 * it for the guest to avoid crossing a 4kB boundary. We only
549 		 * use 640 bytes of the structure though, so we should accept
550 		 * clients that set a size of 640.
551 		 */
552 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
553 		if (len < sizeof(struct lppaca))
554 			break;
555 		vpap = &tvcpu->arch.vpa;
556 		err = 0;
557 		break;
558 
559 	case H_VPA_REG_DTL:		/* register DTL */
560 		if (len < sizeof(struct dtl_entry))
561 			break;
562 		len -= len % sizeof(struct dtl_entry);
563 
564 		/* Check that they have previously registered a VPA */
565 		err = H_RESOURCE;
566 		if (!vpa_is_registered(&tvcpu->arch.vpa))
567 			break;
568 
569 		vpap = &tvcpu->arch.dtl;
570 		err = 0;
571 		break;
572 
573 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
574 		/* Check that they have previously registered a VPA */
575 		err = H_RESOURCE;
576 		if (!vpa_is_registered(&tvcpu->arch.vpa))
577 			break;
578 
579 		vpap = &tvcpu->arch.slb_shadow;
580 		err = 0;
581 		break;
582 
583 	case H_VPA_DEREG_VPA:		/* deregister VPA */
584 		/* Check they don't still have a DTL or SLB buf registered */
585 		err = H_RESOURCE;
586 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
587 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
588 			break;
589 
590 		vpap = &tvcpu->arch.vpa;
591 		err = 0;
592 		break;
593 
594 	case H_VPA_DEREG_DTL:		/* deregister DTL */
595 		vpap = &tvcpu->arch.dtl;
596 		err = 0;
597 		break;
598 
599 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
600 		vpap = &tvcpu->arch.slb_shadow;
601 		err = 0;
602 		break;
603 	}
604 
605 	if (vpap) {
606 		vpap->next_gpa = vpa;
607 		vpap->len = len;
608 		vpap->update_pending = 1;
609 	}
610 
611 	spin_unlock(&tvcpu->arch.vpa_update_lock);
612 
613 	return err;
614 }
615 
616 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
617 {
618 	struct kvm *kvm = vcpu->kvm;
619 	void *va;
620 	unsigned long nb;
621 	unsigned long gpa;
622 
623 	/*
624 	 * We need to pin the page pointed to by vpap->next_gpa,
625 	 * but we can't call kvmppc_pin_guest_page under the lock
626 	 * as it does get_user_pages() and down_read().  So we
627 	 * have to drop the lock, pin the page, then get the lock
628 	 * again and check that a new area didn't get registered
629 	 * in the meantime.
630 	 */
631 	for (;;) {
632 		gpa = vpap->next_gpa;
633 		spin_unlock(&vcpu->arch.vpa_update_lock);
634 		va = NULL;
635 		nb = 0;
636 		if (gpa)
637 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
638 		spin_lock(&vcpu->arch.vpa_update_lock);
639 		if (gpa == vpap->next_gpa)
640 			break;
641 		/* sigh... unpin that one and try again */
642 		if (va)
643 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
644 	}
645 
646 	vpap->update_pending = 0;
647 	if (va && nb < vpap->len) {
648 		/*
649 		 * If it's now too short, it must be that userspace
650 		 * has changed the mappings underlying guest memory,
651 		 * so unregister the region.
652 		 */
653 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
654 		va = NULL;
655 	}
656 	if (vpap->pinned_addr)
657 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
658 					vpap->dirty);
659 	vpap->gpa = gpa;
660 	vpap->pinned_addr = va;
661 	vpap->dirty = false;
662 	if (va)
663 		vpap->pinned_end = va + vpap->len;
664 }
665 
666 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
667 {
668 	if (!(vcpu->arch.vpa.update_pending ||
669 	      vcpu->arch.slb_shadow.update_pending ||
670 	      vcpu->arch.dtl.update_pending))
671 		return;
672 
673 	spin_lock(&vcpu->arch.vpa_update_lock);
674 	if (vcpu->arch.vpa.update_pending) {
675 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
676 		if (vcpu->arch.vpa.pinned_addr)
677 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
678 	}
679 	if (vcpu->arch.dtl.update_pending) {
680 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
681 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
682 		vcpu->arch.dtl_index = 0;
683 	}
684 	if (vcpu->arch.slb_shadow.update_pending)
685 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
686 	spin_unlock(&vcpu->arch.vpa_update_lock);
687 }
688 
689 /*
690  * Return the accumulated stolen time for the vcore up until `now'.
691  * The caller should hold the vcore lock.
692  */
693 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
694 {
695 	u64 p;
696 	unsigned long flags;
697 
698 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
699 
700 	spin_lock_irqsave(&vc->stoltb_lock, flags);
701 	p = vc->stolen_tb;
702 	if (vc->vcore_state != VCORE_INACTIVE &&
703 	    vc->preempt_tb != TB_NIL)
704 		p += now - vc->preempt_tb;
705 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
706 	return p;
707 }
708 
709 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
710 					unsigned int pcpu, u64 now,
711 					unsigned long stolen)
712 {
713 	struct dtl_entry *dt;
714 	struct lppaca *vpa;
715 
716 	dt = vcpu->arch.dtl_ptr;
717 	vpa = vcpu->arch.vpa.pinned_addr;
718 
719 	if (!dt || !vpa)
720 		return;
721 
722 	dt->dispatch_reason = 7;
723 	dt->preempt_reason = 0;
724 	dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid);
725 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
726 	dt->ready_to_enqueue_time = 0;
727 	dt->waiting_to_ready_time = 0;
728 	dt->timebase = cpu_to_be64(now);
729 	dt->fault_addr = 0;
730 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
731 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
732 
733 	++dt;
734 	if (dt == vcpu->arch.dtl.pinned_end)
735 		dt = vcpu->arch.dtl.pinned_addr;
736 	vcpu->arch.dtl_ptr = dt;
737 	/* order writing *dt vs. writing vpa->dtl_idx */
738 	smp_wmb();
739 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
740 	vcpu->arch.dtl.dirty = true;
741 }
742 
743 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
744 				    struct kvmppc_vcore *vc)
745 {
746 	unsigned long stolen;
747 	unsigned long core_stolen;
748 	u64 now;
749 	unsigned long flags;
750 
751 	now = mftb();
752 
753 	core_stolen = vcore_stolen_time(vc, now);
754 	stolen = core_stolen - vcpu->arch.stolen_logged;
755 	vcpu->arch.stolen_logged = core_stolen;
756 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
757 	stolen += vcpu->arch.busy_stolen;
758 	vcpu->arch.busy_stolen = 0;
759 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
760 
761 	__kvmppc_create_dtl_entry(vcpu, vc->pcpu, now + vc->tb_offset, stolen);
762 }
763 
764 /* See if there is a doorbell interrupt pending for a vcpu */
765 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
766 {
767 	int thr;
768 	struct kvmppc_vcore *vc;
769 
770 	if (vcpu->arch.doorbell_request)
771 		return true;
772 	if (cpu_has_feature(CPU_FTR_ARCH_300))
773 		return false;
774 	/*
775 	 * Ensure that the read of vcore->dpdes comes after the read
776 	 * of vcpu->doorbell_request.  This barrier matches the
777 	 * smp_wmb() in kvmppc_guest_entry_inject().
778 	 */
779 	smp_rmb();
780 	vc = vcpu->arch.vcore;
781 	thr = vcpu->vcpu_id - vc->first_vcpuid;
782 	return !!(vc->dpdes & (1 << thr));
783 }
784 
785 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
786 {
787 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
788 		return true;
789 	if ((!vcpu->arch.vcore->arch_compat) &&
790 	    cpu_has_feature(CPU_FTR_ARCH_207S))
791 		return true;
792 	return false;
793 }
794 
795 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
796 			     unsigned long resource, unsigned long value1,
797 			     unsigned long value2)
798 {
799 	switch (resource) {
800 	case H_SET_MODE_RESOURCE_SET_CIABR:
801 		if (!kvmppc_power8_compatible(vcpu))
802 			return H_P2;
803 		if (value2)
804 			return H_P4;
805 		if (mflags)
806 			return H_UNSUPPORTED_FLAG_START;
807 		/* Guests can't breakpoint the hypervisor */
808 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
809 			return H_P3;
810 		vcpu->arch.ciabr  = value1;
811 		return H_SUCCESS;
812 	case H_SET_MODE_RESOURCE_SET_DAWR0:
813 		if (!kvmppc_power8_compatible(vcpu))
814 			return H_P2;
815 		if (!ppc_breakpoint_available())
816 			return H_P2;
817 		if (mflags)
818 			return H_UNSUPPORTED_FLAG_START;
819 		if (value2 & DABRX_HYP)
820 			return H_P4;
821 		vcpu->arch.dawr0  = value1;
822 		vcpu->arch.dawrx0 = value2;
823 		return H_SUCCESS;
824 	case H_SET_MODE_RESOURCE_SET_DAWR1:
825 		if (!kvmppc_power8_compatible(vcpu))
826 			return H_P2;
827 		if (!ppc_breakpoint_available())
828 			return H_P2;
829 		if (!cpu_has_feature(CPU_FTR_DAWR1))
830 			return H_P2;
831 		if (!vcpu->kvm->arch.dawr1_enabled)
832 			return H_FUNCTION;
833 		if (mflags)
834 			return H_UNSUPPORTED_FLAG_START;
835 		if (value2 & DABRX_HYP)
836 			return H_P4;
837 		vcpu->arch.dawr1  = value1;
838 		vcpu->arch.dawrx1 = value2;
839 		return H_SUCCESS;
840 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
841 		/*
842 		 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
843 		 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
844 		 */
845 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
846 				kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
847 			return H_UNSUPPORTED_FLAG_START;
848 		return H_TOO_HARD;
849 	default:
850 		return H_TOO_HARD;
851 	}
852 }
853 
854 /* Copy guest memory in place - must reside within a single memslot */
855 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
856 				  unsigned long len)
857 {
858 	struct kvm_memory_slot *to_memslot = NULL;
859 	struct kvm_memory_slot *from_memslot = NULL;
860 	unsigned long to_addr, from_addr;
861 	int r;
862 
863 	/* Get HPA for from address */
864 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
865 	if (!from_memslot)
866 		return -EFAULT;
867 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
868 			     << PAGE_SHIFT))
869 		return -EINVAL;
870 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
871 	if (kvm_is_error_hva(from_addr))
872 		return -EFAULT;
873 	from_addr |= (from & (PAGE_SIZE - 1));
874 
875 	/* Get HPA for to address */
876 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
877 	if (!to_memslot)
878 		return -EFAULT;
879 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
880 			   << PAGE_SHIFT))
881 		return -EINVAL;
882 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
883 	if (kvm_is_error_hva(to_addr))
884 		return -EFAULT;
885 	to_addr |= (to & (PAGE_SIZE - 1));
886 
887 	/* Perform copy */
888 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
889 			     len);
890 	if (r)
891 		return -EFAULT;
892 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
893 	return 0;
894 }
895 
896 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
897 			       unsigned long dest, unsigned long src)
898 {
899 	u64 pg_sz = SZ_4K;		/* 4K page size */
900 	u64 pg_mask = SZ_4K - 1;
901 	int ret;
902 
903 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
904 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
905 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
906 		return H_PARAMETER;
907 
908 	/* dest (and src if copy_page flag set) must be page aligned */
909 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
910 		return H_PARAMETER;
911 
912 	/* zero and/or copy the page as determined by the flags */
913 	if (flags & H_COPY_PAGE) {
914 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
915 		if (ret < 0)
916 			return H_PARAMETER;
917 	} else if (flags & H_ZERO_PAGE) {
918 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
919 		if (ret < 0)
920 			return H_PARAMETER;
921 	}
922 
923 	/* We can ignore the remaining flags */
924 
925 	return H_SUCCESS;
926 }
927 
928 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
929 {
930 	struct kvmppc_vcore *vcore = target->arch.vcore;
931 
932 	/*
933 	 * We expect to have been called by the real mode handler
934 	 * (kvmppc_rm_h_confer()) which would have directly returned
935 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
936 	 * have useful work to do and should not confer) so we don't
937 	 * recheck that here.
938 	 *
939 	 * In the case of the P9 single vcpu per vcore case, the real
940 	 * mode handler is not called but no other threads are in the
941 	 * source vcore.
942 	 */
943 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
944 		spin_lock(&vcore->lock);
945 		if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
946 		    vcore->vcore_state != VCORE_INACTIVE &&
947 		    vcore->runner)
948 			target = vcore->runner;
949 		spin_unlock(&vcore->lock);
950 	}
951 
952 	return kvm_vcpu_yield_to(target);
953 }
954 
955 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
956 {
957 	int yield_count = 0;
958 	struct lppaca *lppaca;
959 
960 	spin_lock(&vcpu->arch.vpa_update_lock);
961 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
962 	if (lppaca)
963 		yield_count = be32_to_cpu(lppaca->yield_count);
964 	spin_unlock(&vcpu->arch.vpa_update_lock);
965 	return yield_count;
966 }
967 
968 /*
969  * H_RPT_INVALIDATE hcall handler for nested guests.
970  *
971  * Handles only nested process-scoped invalidation requests in L0.
972  */
973 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
974 {
975 	unsigned long type = kvmppc_get_gpr(vcpu, 6);
976 	unsigned long pid, pg_sizes, start, end;
977 
978 	/*
979 	 * The partition-scoped invalidations aren't handled here in L0.
980 	 */
981 	if (type & H_RPTI_TYPE_NESTED)
982 		return RESUME_HOST;
983 
984 	pid = kvmppc_get_gpr(vcpu, 4);
985 	pg_sizes = kvmppc_get_gpr(vcpu, 7);
986 	start = kvmppc_get_gpr(vcpu, 8);
987 	end = kvmppc_get_gpr(vcpu, 9);
988 
989 	do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
990 				type, pg_sizes, start, end);
991 
992 	kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
993 	return RESUME_GUEST;
994 }
995 
996 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
997 				    unsigned long id, unsigned long target,
998 				    unsigned long type, unsigned long pg_sizes,
999 				    unsigned long start, unsigned long end)
1000 {
1001 	if (!kvm_is_radix(vcpu->kvm))
1002 		return H_UNSUPPORTED;
1003 
1004 	if (end < start)
1005 		return H_P5;
1006 
1007 	/*
1008 	 * Partition-scoped invalidation for nested guests.
1009 	 */
1010 	if (type & H_RPTI_TYPE_NESTED) {
1011 		if (!nesting_enabled(vcpu->kvm))
1012 			return H_FUNCTION;
1013 
1014 		/* Support only cores as target */
1015 		if (target != H_RPTI_TARGET_CMMU)
1016 			return H_P2;
1017 
1018 		return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
1019 					       start, end);
1020 	}
1021 
1022 	/*
1023 	 * Process-scoped invalidation for L1 guests.
1024 	 */
1025 	do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
1026 				type, pg_sizes, start, end);
1027 	return H_SUCCESS;
1028 }
1029 
1030 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
1031 {
1032 	struct kvm *kvm = vcpu->kvm;
1033 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
1034 	unsigned long target, ret = H_SUCCESS;
1035 	int yield_count;
1036 	struct kvm_vcpu *tvcpu;
1037 	int idx, rc;
1038 
1039 	if (req <= MAX_HCALL_OPCODE &&
1040 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1041 		return RESUME_HOST;
1042 
1043 	switch (req) {
1044 	case H_REMOVE:
1045 		ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1046 					kvmppc_get_gpr(vcpu, 5),
1047 					kvmppc_get_gpr(vcpu, 6));
1048 		if (ret == H_TOO_HARD)
1049 			return RESUME_HOST;
1050 		break;
1051 	case H_ENTER:
1052 		ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1053 					kvmppc_get_gpr(vcpu, 5),
1054 					kvmppc_get_gpr(vcpu, 6),
1055 					kvmppc_get_gpr(vcpu, 7));
1056 		if (ret == H_TOO_HARD)
1057 			return RESUME_HOST;
1058 		break;
1059 	case H_READ:
1060 		ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1061 					kvmppc_get_gpr(vcpu, 5));
1062 		if (ret == H_TOO_HARD)
1063 			return RESUME_HOST;
1064 		break;
1065 	case H_CLEAR_MOD:
1066 		ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1067 					kvmppc_get_gpr(vcpu, 5));
1068 		if (ret == H_TOO_HARD)
1069 			return RESUME_HOST;
1070 		break;
1071 	case H_CLEAR_REF:
1072 		ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1073 					kvmppc_get_gpr(vcpu, 5));
1074 		if (ret == H_TOO_HARD)
1075 			return RESUME_HOST;
1076 		break;
1077 	case H_PROTECT:
1078 		ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1079 					kvmppc_get_gpr(vcpu, 5),
1080 					kvmppc_get_gpr(vcpu, 6));
1081 		if (ret == H_TOO_HARD)
1082 			return RESUME_HOST;
1083 		break;
1084 	case H_BULK_REMOVE:
1085 		ret = kvmppc_h_bulk_remove(vcpu);
1086 		if (ret == H_TOO_HARD)
1087 			return RESUME_HOST;
1088 		break;
1089 
1090 	case H_CEDE:
1091 		break;
1092 	case H_PROD:
1093 		target = kvmppc_get_gpr(vcpu, 4);
1094 		tvcpu = kvmppc_find_vcpu(kvm, target);
1095 		if (!tvcpu) {
1096 			ret = H_PARAMETER;
1097 			break;
1098 		}
1099 		tvcpu->arch.prodded = 1;
1100 		smp_mb(); /* This orders prodded store vs ceded load */
1101 		if (tvcpu->arch.ceded)
1102 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1103 		break;
1104 	case H_CONFER:
1105 		target = kvmppc_get_gpr(vcpu, 4);
1106 		if (target == -1)
1107 			break;
1108 		tvcpu = kvmppc_find_vcpu(kvm, target);
1109 		if (!tvcpu) {
1110 			ret = H_PARAMETER;
1111 			break;
1112 		}
1113 		yield_count = kvmppc_get_gpr(vcpu, 5);
1114 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
1115 			break;
1116 		kvm_arch_vcpu_yield_to(tvcpu);
1117 		break;
1118 	case H_REGISTER_VPA:
1119 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1120 					kvmppc_get_gpr(vcpu, 5),
1121 					kvmppc_get_gpr(vcpu, 6));
1122 		break;
1123 	case H_RTAS:
1124 		if (list_empty(&kvm->arch.rtas_tokens))
1125 			return RESUME_HOST;
1126 
1127 		idx = srcu_read_lock(&kvm->srcu);
1128 		rc = kvmppc_rtas_hcall(vcpu);
1129 		srcu_read_unlock(&kvm->srcu, idx);
1130 
1131 		if (rc == -ENOENT)
1132 			return RESUME_HOST;
1133 		else if (rc == 0)
1134 			break;
1135 
1136 		/* Send the error out to userspace via KVM_RUN */
1137 		return rc;
1138 	case H_LOGICAL_CI_LOAD:
1139 		ret = kvmppc_h_logical_ci_load(vcpu);
1140 		if (ret == H_TOO_HARD)
1141 			return RESUME_HOST;
1142 		break;
1143 	case H_LOGICAL_CI_STORE:
1144 		ret = kvmppc_h_logical_ci_store(vcpu);
1145 		if (ret == H_TOO_HARD)
1146 			return RESUME_HOST;
1147 		break;
1148 	case H_SET_MODE:
1149 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1150 					kvmppc_get_gpr(vcpu, 5),
1151 					kvmppc_get_gpr(vcpu, 6),
1152 					kvmppc_get_gpr(vcpu, 7));
1153 		if (ret == H_TOO_HARD)
1154 			return RESUME_HOST;
1155 		break;
1156 	case H_XIRR:
1157 	case H_CPPR:
1158 	case H_EOI:
1159 	case H_IPI:
1160 	case H_IPOLL:
1161 	case H_XIRR_X:
1162 		if (kvmppc_xics_enabled(vcpu)) {
1163 			if (xics_on_xive()) {
1164 				ret = H_NOT_AVAILABLE;
1165 				return RESUME_GUEST;
1166 			}
1167 			ret = kvmppc_xics_hcall(vcpu, req);
1168 			break;
1169 		}
1170 		return RESUME_HOST;
1171 	case H_SET_DABR:
1172 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1173 		break;
1174 	case H_SET_XDABR:
1175 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1176 						kvmppc_get_gpr(vcpu, 5));
1177 		break;
1178 #ifdef CONFIG_SPAPR_TCE_IOMMU
1179 	case H_GET_TCE:
1180 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1181 						kvmppc_get_gpr(vcpu, 5));
1182 		if (ret == H_TOO_HARD)
1183 			return RESUME_HOST;
1184 		break;
1185 	case H_PUT_TCE:
1186 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1187 						kvmppc_get_gpr(vcpu, 5),
1188 						kvmppc_get_gpr(vcpu, 6));
1189 		if (ret == H_TOO_HARD)
1190 			return RESUME_HOST;
1191 		break;
1192 	case H_PUT_TCE_INDIRECT:
1193 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1194 						kvmppc_get_gpr(vcpu, 5),
1195 						kvmppc_get_gpr(vcpu, 6),
1196 						kvmppc_get_gpr(vcpu, 7));
1197 		if (ret == H_TOO_HARD)
1198 			return RESUME_HOST;
1199 		break;
1200 	case H_STUFF_TCE:
1201 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1202 						kvmppc_get_gpr(vcpu, 5),
1203 						kvmppc_get_gpr(vcpu, 6),
1204 						kvmppc_get_gpr(vcpu, 7));
1205 		if (ret == H_TOO_HARD)
1206 			return RESUME_HOST;
1207 		break;
1208 #endif
1209 	case H_RANDOM:
1210 		if (!arch_get_random_seed_longs(&vcpu->arch.regs.gpr[4], 1))
1211 			ret = H_HARDWARE;
1212 		break;
1213 	case H_RPT_INVALIDATE:
1214 		ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1215 					      kvmppc_get_gpr(vcpu, 5),
1216 					      kvmppc_get_gpr(vcpu, 6),
1217 					      kvmppc_get_gpr(vcpu, 7),
1218 					      kvmppc_get_gpr(vcpu, 8),
1219 					      kvmppc_get_gpr(vcpu, 9));
1220 		break;
1221 
1222 	case H_SET_PARTITION_TABLE:
1223 		ret = H_FUNCTION;
1224 		if (nesting_enabled(kvm))
1225 			ret = kvmhv_set_partition_table(vcpu);
1226 		break;
1227 	case H_ENTER_NESTED:
1228 		ret = H_FUNCTION;
1229 		if (!nesting_enabled(kvm))
1230 			break;
1231 		ret = kvmhv_enter_nested_guest(vcpu);
1232 		if (ret == H_INTERRUPT) {
1233 			kvmppc_set_gpr(vcpu, 3, 0);
1234 			vcpu->arch.hcall_needed = 0;
1235 			return -EINTR;
1236 		} else if (ret == H_TOO_HARD) {
1237 			kvmppc_set_gpr(vcpu, 3, 0);
1238 			vcpu->arch.hcall_needed = 0;
1239 			return RESUME_HOST;
1240 		}
1241 		break;
1242 	case H_TLB_INVALIDATE:
1243 		ret = H_FUNCTION;
1244 		if (nesting_enabled(kvm))
1245 			ret = kvmhv_do_nested_tlbie(vcpu);
1246 		break;
1247 	case H_COPY_TOFROM_GUEST:
1248 		ret = H_FUNCTION;
1249 		if (nesting_enabled(kvm))
1250 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1251 		break;
1252 	case H_PAGE_INIT:
1253 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1254 					 kvmppc_get_gpr(vcpu, 5),
1255 					 kvmppc_get_gpr(vcpu, 6));
1256 		break;
1257 	case H_SVM_PAGE_IN:
1258 		ret = H_UNSUPPORTED;
1259 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1260 			ret = kvmppc_h_svm_page_in(kvm,
1261 						   kvmppc_get_gpr(vcpu, 4),
1262 						   kvmppc_get_gpr(vcpu, 5),
1263 						   kvmppc_get_gpr(vcpu, 6));
1264 		break;
1265 	case H_SVM_PAGE_OUT:
1266 		ret = H_UNSUPPORTED;
1267 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1268 			ret = kvmppc_h_svm_page_out(kvm,
1269 						    kvmppc_get_gpr(vcpu, 4),
1270 						    kvmppc_get_gpr(vcpu, 5),
1271 						    kvmppc_get_gpr(vcpu, 6));
1272 		break;
1273 	case H_SVM_INIT_START:
1274 		ret = H_UNSUPPORTED;
1275 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1276 			ret = kvmppc_h_svm_init_start(kvm);
1277 		break;
1278 	case H_SVM_INIT_DONE:
1279 		ret = H_UNSUPPORTED;
1280 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1281 			ret = kvmppc_h_svm_init_done(kvm);
1282 		break;
1283 	case H_SVM_INIT_ABORT:
1284 		/*
1285 		 * Even if that call is made by the Ultravisor, the SSR1 value
1286 		 * is the guest context one, with the secure bit clear as it has
1287 		 * not yet been secured. So we can't check it here.
1288 		 * Instead the kvm->arch.secure_guest flag is checked inside
1289 		 * kvmppc_h_svm_init_abort().
1290 		 */
1291 		ret = kvmppc_h_svm_init_abort(kvm);
1292 		break;
1293 
1294 	default:
1295 		return RESUME_HOST;
1296 	}
1297 	WARN_ON_ONCE(ret == H_TOO_HARD);
1298 	kvmppc_set_gpr(vcpu, 3, ret);
1299 	vcpu->arch.hcall_needed = 0;
1300 	return RESUME_GUEST;
1301 }
1302 
1303 /*
1304  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1305  * handlers in book3s_hv_rmhandlers.S.
1306  *
1307  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1308  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1309  */
1310 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1311 {
1312 	vcpu->arch.shregs.msr |= MSR_EE;
1313 	vcpu->arch.ceded = 1;
1314 	smp_mb();
1315 	if (vcpu->arch.prodded) {
1316 		vcpu->arch.prodded = 0;
1317 		smp_mb();
1318 		vcpu->arch.ceded = 0;
1319 	}
1320 }
1321 
1322 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1323 {
1324 	switch (cmd) {
1325 	case H_CEDE:
1326 	case H_PROD:
1327 	case H_CONFER:
1328 	case H_REGISTER_VPA:
1329 	case H_SET_MODE:
1330 #ifdef CONFIG_SPAPR_TCE_IOMMU
1331 	case H_GET_TCE:
1332 	case H_PUT_TCE:
1333 	case H_PUT_TCE_INDIRECT:
1334 	case H_STUFF_TCE:
1335 #endif
1336 	case H_LOGICAL_CI_LOAD:
1337 	case H_LOGICAL_CI_STORE:
1338 #ifdef CONFIG_KVM_XICS
1339 	case H_XIRR:
1340 	case H_CPPR:
1341 	case H_EOI:
1342 	case H_IPI:
1343 	case H_IPOLL:
1344 	case H_XIRR_X:
1345 #endif
1346 	case H_PAGE_INIT:
1347 	case H_RPT_INVALIDATE:
1348 		return 1;
1349 	}
1350 
1351 	/* See if it's in the real-mode table */
1352 	return kvmppc_hcall_impl_hv_realmode(cmd);
1353 }
1354 
1355 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1356 {
1357 	u32 last_inst;
1358 
1359 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1360 					EMULATE_DONE) {
1361 		/*
1362 		 * Fetch failed, so return to guest and
1363 		 * try executing it again.
1364 		 */
1365 		return RESUME_GUEST;
1366 	}
1367 
1368 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1369 		vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1370 		vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1371 		return RESUME_HOST;
1372 	} else {
1373 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1374 		return RESUME_GUEST;
1375 	}
1376 }
1377 
1378 static void do_nothing(void *x)
1379 {
1380 }
1381 
1382 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1383 {
1384 	int thr, cpu, pcpu, nthreads;
1385 	struct kvm_vcpu *v;
1386 	unsigned long dpdes;
1387 
1388 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1389 	dpdes = 0;
1390 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1391 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1392 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1393 		if (!v)
1394 			continue;
1395 		/*
1396 		 * If the vcpu is currently running on a physical cpu thread,
1397 		 * interrupt it in order to pull it out of the guest briefly,
1398 		 * which will update its vcore->dpdes value.
1399 		 */
1400 		pcpu = READ_ONCE(v->cpu);
1401 		if (pcpu >= 0)
1402 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1403 		if (kvmppc_doorbell_pending(v))
1404 			dpdes |= 1 << thr;
1405 	}
1406 	return dpdes;
1407 }
1408 
1409 /*
1410  * On POWER9, emulate doorbell-related instructions in order to
1411  * give the guest the illusion of running on a multi-threaded core.
1412  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1413  * and mfspr DPDES.
1414  */
1415 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1416 {
1417 	u32 inst, rb, thr;
1418 	unsigned long arg;
1419 	struct kvm *kvm = vcpu->kvm;
1420 	struct kvm_vcpu *tvcpu;
1421 
1422 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1423 		return RESUME_GUEST;
1424 	if (get_op(inst) != 31)
1425 		return EMULATE_FAIL;
1426 	rb = get_rb(inst);
1427 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1428 	switch (get_xop(inst)) {
1429 	case OP_31_XOP_MSGSNDP:
1430 		arg = kvmppc_get_gpr(vcpu, rb);
1431 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1432 			break;
1433 		arg &= 0x7f;
1434 		if (arg >= kvm->arch.emul_smt_mode)
1435 			break;
1436 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1437 		if (!tvcpu)
1438 			break;
1439 		if (!tvcpu->arch.doorbell_request) {
1440 			tvcpu->arch.doorbell_request = 1;
1441 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1442 		}
1443 		break;
1444 	case OP_31_XOP_MSGCLRP:
1445 		arg = kvmppc_get_gpr(vcpu, rb);
1446 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1447 			break;
1448 		vcpu->arch.vcore->dpdes = 0;
1449 		vcpu->arch.doorbell_request = 0;
1450 		break;
1451 	case OP_31_XOP_MFSPR:
1452 		switch (get_sprn(inst)) {
1453 		case SPRN_TIR:
1454 			arg = thr;
1455 			break;
1456 		case SPRN_DPDES:
1457 			arg = kvmppc_read_dpdes(vcpu);
1458 			break;
1459 		default:
1460 			return EMULATE_FAIL;
1461 		}
1462 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1463 		break;
1464 	default:
1465 		return EMULATE_FAIL;
1466 	}
1467 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1468 	return RESUME_GUEST;
1469 }
1470 
1471 /*
1472  * If the lppaca had pmcregs_in_use clear when we exited the guest, then
1473  * HFSCR_PM is cleared for next entry. If the guest then tries to access
1474  * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM
1475  * back in the guest HFSCR will cause the next entry to load the PMU SPRs and
1476  * allow the guest access to continue.
1477  */
1478 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu)
1479 {
1480 	if (!(vcpu->arch.hfscr_permitted & HFSCR_PM))
1481 		return EMULATE_FAIL;
1482 
1483 	vcpu->arch.hfscr |= HFSCR_PM;
1484 
1485 	return RESUME_GUEST;
1486 }
1487 
1488 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu)
1489 {
1490 	if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB))
1491 		return EMULATE_FAIL;
1492 
1493 	vcpu->arch.hfscr |= HFSCR_EBB;
1494 
1495 	return RESUME_GUEST;
1496 }
1497 
1498 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu)
1499 {
1500 	if (!(vcpu->arch.hfscr_permitted & HFSCR_TM))
1501 		return EMULATE_FAIL;
1502 
1503 	vcpu->arch.hfscr |= HFSCR_TM;
1504 
1505 	return RESUME_GUEST;
1506 }
1507 
1508 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1509 				 struct task_struct *tsk)
1510 {
1511 	struct kvm_run *run = vcpu->run;
1512 	int r = RESUME_HOST;
1513 
1514 	vcpu->stat.sum_exits++;
1515 
1516 	/*
1517 	 * This can happen if an interrupt occurs in the last stages
1518 	 * of guest entry or the first stages of guest exit (i.e. after
1519 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1520 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1521 	 * That can happen due to a bug, or due to a machine check
1522 	 * occurring at just the wrong time.
1523 	 */
1524 	if (vcpu->arch.shregs.msr & MSR_HV) {
1525 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1526 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1527 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1528 			vcpu->arch.shregs.msr);
1529 		kvmppc_dump_regs(vcpu);
1530 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1531 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1532 		return RESUME_HOST;
1533 	}
1534 	run->exit_reason = KVM_EXIT_UNKNOWN;
1535 	run->ready_for_interrupt_injection = 1;
1536 	switch (vcpu->arch.trap) {
1537 	/* We're good on these - the host merely wanted to get our attention */
1538 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1539 		WARN_ON_ONCE(1); /* Should never happen */
1540 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1541 		fallthrough;
1542 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1543 		vcpu->stat.dec_exits++;
1544 		r = RESUME_GUEST;
1545 		break;
1546 	case BOOK3S_INTERRUPT_EXTERNAL:
1547 	case BOOK3S_INTERRUPT_H_DOORBELL:
1548 	case BOOK3S_INTERRUPT_H_VIRT:
1549 		vcpu->stat.ext_intr_exits++;
1550 		r = RESUME_GUEST;
1551 		break;
1552 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1553 	case BOOK3S_INTERRUPT_HMI:
1554 	case BOOK3S_INTERRUPT_PERFMON:
1555 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1556 		r = RESUME_GUEST;
1557 		break;
1558 	case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1559 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1560 					      DEFAULT_RATELIMIT_BURST);
1561 		/*
1562 		 * Print the MCE event to host console. Ratelimit so the guest
1563 		 * can't flood the host log.
1564 		 */
1565 		if (__ratelimit(&rs))
1566 			machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1567 
1568 		/*
1569 		 * If the guest can do FWNMI, exit to userspace so it can
1570 		 * deliver a FWNMI to the guest.
1571 		 * Otherwise we synthesize a machine check for the guest
1572 		 * so that it knows that the machine check occurred.
1573 		 */
1574 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1575 			ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1576 			kvmppc_core_queue_machine_check(vcpu, flags);
1577 			r = RESUME_GUEST;
1578 			break;
1579 		}
1580 
1581 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1582 		run->exit_reason = KVM_EXIT_NMI;
1583 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1584 		/* Clear out the old NMI status from run->flags */
1585 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1586 		/* Now set the NMI status */
1587 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1588 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1589 		else
1590 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1591 
1592 		r = RESUME_HOST;
1593 		break;
1594 	}
1595 	case BOOK3S_INTERRUPT_PROGRAM:
1596 	{
1597 		ulong flags;
1598 		/*
1599 		 * Normally program interrupts are delivered directly
1600 		 * to the guest by the hardware, but we can get here
1601 		 * as a result of a hypervisor emulation interrupt
1602 		 * (e40) getting turned into a 700 by BML RTAS.
1603 		 */
1604 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1605 		kvmppc_core_queue_program(vcpu, flags);
1606 		r = RESUME_GUEST;
1607 		break;
1608 	}
1609 	case BOOK3S_INTERRUPT_SYSCALL:
1610 	{
1611 		int i;
1612 
1613 		if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) {
1614 			/*
1615 			 * Guest userspace executed sc 1. This can only be
1616 			 * reached by the P9 path because the old path
1617 			 * handles this case in realmode hcall handlers.
1618 			 */
1619 			if (!kvmhv_vcpu_is_radix(vcpu)) {
1620 				/*
1621 				 * A guest could be running PR KVM, so this
1622 				 * may be a PR KVM hcall. It must be reflected
1623 				 * to the guest kernel as a sc interrupt.
1624 				 */
1625 				kvmppc_core_queue_syscall(vcpu);
1626 			} else {
1627 				/*
1628 				 * Radix guests can not run PR KVM or nested HV
1629 				 * hash guests which might run PR KVM, so this
1630 				 * is always a privilege fault. Send a program
1631 				 * check to guest kernel.
1632 				 */
1633 				kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1634 			}
1635 			r = RESUME_GUEST;
1636 			break;
1637 		}
1638 
1639 		/*
1640 		 * hcall - gather args and set exit_reason. This will next be
1641 		 * handled by kvmppc_pseries_do_hcall which may be able to deal
1642 		 * with it and resume guest, or may punt to userspace.
1643 		 */
1644 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1645 		for (i = 0; i < 9; ++i)
1646 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1647 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1648 		vcpu->arch.hcall_needed = 1;
1649 		r = RESUME_HOST;
1650 		break;
1651 	}
1652 	/*
1653 	 * We get these next two if the guest accesses a page which it thinks
1654 	 * it has mapped but which is not actually present, either because
1655 	 * it is for an emulated I/O device or because the corresonding
1656 	 * host page has been paged out.
1657 	 *
1658 	 * Any other HDSI/HISI interrupts have been handled already for P7/8
1659 	 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1660 	 * fault handling is done here.
1661 	 */
1662 	case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1663 		unsigned long vsid;
1664 		long err;
1665 
1666 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
1667 		    unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) {
1668 			r = RESUME_GUEST; /* Just retry if it's the canary */
1669 			break;
1670 		}
1671 
1672 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1673 			/*
1674 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1675 			 * already attempted to handle this in rmhandlers. The
1676 			 * hash fault handling below is v3 only (it uses ASDR
1677 			 * via fault_gpa).
1678 			 */
1679 			r = RESUME_PAGE_FAULT;
1680 			break;
1681 		}
1682 
1683 		if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1684 			kvmppc_core_queue_data_storage(vcpu,
1685 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1686 			r = RESUME_GUEST;
1687 			break;
1688 		}
1689 
1690 		if (!(vcpu->arch.shregs.msr & MSR_DR))
1691 			vsid = vcpu->kvm->arch.vrma_slb_v;
1692 		else
1693 			vsid = vcpu->arch.fault_gpa;
1694 
1695 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1696 				vsid, vcpu->arch.fault_dsisr, true);
1697 		if (err == 0) {
1698 			r = RESUME_GUEST;
1699 		} else if (err == -1 || err == -2) {
1700 			r = RESUME_PAGE_FAULT;
1701 		} else {
1702 			kvmppc_core_queue_data_storage(vcpu,
1703 				vcpu->arch.fault_dar, err);
1704 			r = RESUME_GUEST;
1705 		}
1706 		break;
1707 	}
1708 	case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1709 		unsigned long vsid;
1710 		long err;
1711 
1712 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1713 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1714 			DSISR_SRR1_MATCH_64S;
1715 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1716 			/*
1717 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1718 			 * already attempted to handle this in rmhandlers. The
1719 			 * hash fault handling below is v3 only (it uses ASDR
1720 			 * via fault_gpa).
1721 			 */
1722 			if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1723 				vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1724 			r = RESUME_PAGE_FAULT;
1725 			break;
1726 		}
1727 
1728 		if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1729 			kvmppc_core_queue_inst_storage(vcpu,
1730 				vcpu->arch.fault_dsisr);
1731 			r = RESUME_GUEST;
1732 			break;
1733 		}
1734 
1735 		if (!(vcpu->arch.shregs.msr & MSR_IR))
1736 			vsid = vcpu->kvm->arch.vrma_slb_v;
1737 		else
1738 			vsid = vcpu->arch.fault_gpa;
1739 
1740 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1741 				vsid, vcpu->arch.fault_dsisr, false);
1742 		if (err == 0) {
1743 			r = RESUME_GUEST;
1744 		} else if (err == -1) {
1745 			r = RESUME_PAGE_FAULT;
1746 		} else {
1747 			kvmppc_core_queue_inst_storage(vcpu, err);
1748 			r = RESUME_GUEST;
1749 		}
1750 		break;
1751 	}
1752 
1753 	/*
1754 	 * This occurs if the guest executes an illegal instruction.
1755 	 * If the guest debug is disabled, generate a program interrupt
1756 	 * to the guest. If guest debug is enabled, we need to check
1757 	 * whether the instruction is a software breakpoint instruction.
1758 	 * Accordingly return to Guest or Host.
1759 	 */
1760 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1761 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1762 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1763 				swab32(vcpu->arch.emul_inst) :
1764 				vcpu->arch.emul_inst;
1765 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1766 			r = kvmppc_emulate_debug_inst(vcpu);
1767 		} else {
1768 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1769 			r = RESUME_GUEST;
1770 		}
1771 		break;
1772 
1773 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1774 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1775 		/*
1776 		 * This occurs for various TM-related instructions that
1777 		 * we need to emulate on POWER9 DD2.2.  We have already
1778 		 * handled the cases where the guest was in real-suspend
1779 		 * mode and was transitioning to transactional state.
1780 		 */
1781 		r = kvmhv_p9_tm_emulation(vcpu);
1782 		if (r != -1)
1783 			break;
1784 		fallthrough; /* go to facility unavailable handler */
1785 #endif
1786 
1787 	/*
1788 	 * This occurs if the guest (kernel or userspace), does something that
1789 	 * is prohibited by HFSCR.
1790 	 * On POWER9, this could be a doorbell instruction that we need
1791 	 * to emulate.
1792 	 * Otherwise, we just generate a program interrupt to the guest.
1793 	 */
1794 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1795 		u64 cause = vcpu->arch.hfscr >> 56;
1796 
1797 		r = EMULATE_FAIL;
1798 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1799 			if (cause == FSCR_MSGP_LG)
1800 				r = kvmppc_emulate_doorbell_instr(vcpu);
1801 			if (cause == FSCR_PM_LG)
1802 				r = kvmppc_pmu_unavailable(vcpu);
1803 			if (cause == FSCR_EBB_LG)
1804 				r = kvmppc_ebb_unavailable(vcpu);
1805 			if (cause == FSCR_TM_LG)
1806 				r = kvmppc_tm_unavailable(vcpu);
1807 		}
1808 		if (r == EMULATE_FAIL) {
1809 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1810 			r = RESUME_GUEST;
1811 		}
1812 		break;
1813 	}
1814 
1815 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1816 		r = RESUME_PASSTHROUGH;
1817 		break;
1818 	default:
1819 		kvmppc_dump_regs(vcpu);
1820 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1821 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1822 			vcpu->arch.shregs.msr);
1823 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1824 		r = RESUME_HOST;
1825 		break;
1826 	}
1827 
1828 	return r;
1829 }
1830 
1831 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1832 {
1833 	int r;
1834 	int srcu_idx;
1835 
1836 	vcpu->stat.sum_exits++;
1837 
1838 	/*
1839 	 * This can happen if an interrupt occurs in the last stages
1840 	 * of guest entry or the first stages of guest exit (i.e. after
1841 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1842 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1843 	 * That can happen due to a bug, or due to a machine check
1844 	 * occurring at just the wrong time.
1845 	 */
1846 	if (vcpu->arch.shregs.msr & MSR_HV) {
1847 		pr_emerg("KVM trap in HV mode while nested!\n");
1848 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1849 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1850 			 vcpu->arch.shregs.msr);
1851 		kvmppc_dump_regs(vcpu);
1852 		return RESUME_HOST;
1853 	}
1854 	switch (vcpu->arch.trap) {
1855 	/* We're good on these - the host merely wanted to get our attention */
1856 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1857 		vcpu->stat.dec_exits++;
1858 		r = RESUME_GUEST;
1859 		break;
1860 	case BOOK3S_INTERRUPT_EXTERNAL:
1861 		vcpu->stat.ext_intr_exits++;
1862 		r = RESUME_HOST;
1863 		break;
1864 	case BOOK3S_INTERRUPT_H_DOORBELL:
1865 	case BOOK3S_INTERRUPT_H_VIRT:
1866 		vcpu->stat.ext_intr_exits++;
1867 		r = RESUME_GUEST;
1868 		break;
1869 	/* These need to go to the nested HV */
1870 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1871 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1872 		vcpu->stat.dec_exits++;
1873 		r = RESUME_HOST;
1874 		break;
1875 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1876 	case BOOK3S_INTERRUPT_HMI:
1877 	case BOOK3S_INTERRUPT_PERFMON:
1878 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1879 		r = RESUME_GUEST;
1880 		break;
1881 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1882 	{
1883 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1884 					      DEFAULT_RATELIMIT_BURST);
1885 		/* Pass the machine check to the L1 guest */
1886 		r = RESUME_HOST;
1887 		/* Print the MCE event to host console. */
1888 		if (__ratelimit(&rs))
1889 			machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1890 		break;
1891 	}
1892 	/*
1893 	 * We get these next two if the guest accesses a page which it thinks
1894 	 * it has mapped but which is not actually present, either because
1895 	 * it is for an emulated I/O device or because the corresonding
1896 	 * host page has been paged out.
1897 	 */
1898 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1899 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1900 		r = kvmhv_nested_page_fault(vcpu);
1901 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1902 		break;
1903 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1904 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1905 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1906 					 DSISR_SRR1_MATCH_64S;
1907 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1908 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1909 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1910 		r = kvmhv_nested_page_fault(vcpu);
1911 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1912 		break;
1913 
1914 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1915 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1916 		/*
1917 		 * This occurs for various TM-related instructions that
1918 		 * we need to emulate on POWER9 DD2.2.  We have already
1919 		 * handled the cases where the guest was in real-suspend
1920 		 * mode and was transitioning to transactional state.
1921 		 */
1922 		r = kvmhv_p9_tm_emulation(vcpu);
1923 		if (r != -1)
1924 			break;
1925 		fallthrough; /* go to facility unavailable handler */
1926 #endif
1927 
1928 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1929 		u64 cause = vcpu->arch.hfscr >> 56;
1930 
1931 		/*
1932 		 * Only pass HFU interrupts to the L1 if the facility is
1933 		 * permitted but disabled by the L1's HFSCR, otherwise
1934 		 * the interrupt does not make sense to the L1 so turn
1935 		 * it into a HEAI.
1936 		 */
1937 		if (!(vcpu->arch.hfscr_permitted & (1UL << cause)) ||
1938 				(vcpu->arch.nested_hfscr & (1UL << cause))) {
1939 			vcpu->arch.trap = BOOK3S_INTERRUPT_H_EMUL_ASSIST;
1940 
1941 			/*
1942 			 * If the fetch failed, return to guest and
1943 			 * try executing it again.
1944 			 */
1945 			r = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1946 						 &vcpu->arch.emul_inst);
1947 			if (r != EMULATE_DONE)
1948 				r = RESUME_GUEST;
1949 			else
1950 				r = RESUME_HOST;
1951 		} else {
1952 			r = RESUME_HOST;
1953 		}
1954 
1955 		break;
1956 	}
1957 
1958 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1959 		vcpu->arch.trap = 0;
1960 		r = RESUME_GUEST;
1961 		if (!xics_on_xive())
1962 			kvmppc_xics_rm_complete(vcpu, 0);
1963 		break;
1964 	case BOOK3S_INTERRUPT_SYSCALL:
1965 	{
1966 		unsigned long req = kvmppc_get_gpr(vcpu, 3);
1967 
1968 		/*
1969 		 * The H_RPT_INVALIDATE hcalls issued by nested
1970 		 * guests for process-scoped invalidations when
1971 		 * GTSE=0, are handled here in L0.
1972 		 */
1973 		if (req == H_RPT_INVALIDATE) {
1974 			r = kvmppc_nested_h_rpt_invalidate(vcpu);
1975 			break;
1976 		}
1977 
1978 		r = RESUME_HOST;
1979 		break;
1980 	}
1981 	default:
1982 		r = RESUME_HOST;
1983 		break;
1984 	}
1985 
1986 	return r;
1987 }
1988 
1989 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1990 					    struct kvm_sregs *sregs)
1991 {
1992 	int i;
1993 
1994 	memset(sregs, 0, sizeof(struct kvm_sregs));
1995 	sregs->pvr = vcpu->arch.pvr;
1996 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1997 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1998 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1999 	}
2000 
2001 	return 0;
2002 }
2003 
2004 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
2005 					    struct kvm_sregs *sregs)
2006 {
2007 	int i, j;
2008 
2009 	/* Only accept the same PVR as the host's, since we can't spoof it */
2010 	if (sregs->pvr != vcpu->arch.pvr)
2011 		return -EINVAL;
2012 
2013 	j = 0;
2014 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
2015 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
2016 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
2017 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
2018 			++j;
2019 		}
2020 	}
2021 	vcpu->arch.slb_max = j;
2022 
2023 	return 0;
2024 }
2025 
2026 /*
2027  * Enforce limits on guest LPCR values based on hardware availability,
2028  * guest configuration, and possibly hypervisor support and security
2029  * concerns.
2030  */
2031 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
2032 {
2033 	/* LPCR_TC only applies to HPT guests */
2034 	if (kvm_is_radix(kvm))
2035 		lpcr &= ~LPCR_TC;
2036 
2037 	/* On POWER8 and above, userspace can modify AIL */
2038 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2039 		lpcr &= ~LPCR_AIL;
2040 	if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
2041 		lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
2042 	/*
2043 	 * On some POWER9s we force AIL off for radix guests to prevent
2044 	 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
2045 	 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
2046 	 * be cached, which the host TLB management does not expect.
2047 	 */
2048 	if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
2049 		lpcr &= ~LPCR_AIL;
2050 
2051 	/*
2052 	 * On POWER9, allow userspace to enable large decrementer for the
2053 	 * guest, whether or not the host has it enabled.
2054 	 */
2055 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
2056 		lpcr &= ~LPCR_LD;
2057 
2058 	return lpcr;
2059 }
2060 
2061 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
2062 {
2063 	if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
2064 		WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
2065 			  lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
2066 	}
2067 }
2068 
2069 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
2070 		bool preserve_top32)
2071 {
2072 	struct kvm *kvm = vcpu->kvm;
2073 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
2074 	u64 mask;
2075 
2076 	spin_lock(&vc->lock);
2077 
2078 	/*
2079 	 * Userspace can only modify
2080 	 * DPFD (default prefetch depth), ILE (interrupt little-endian),
2081 	 * TC (translation control), AIL (alternate interrupt location),
2082 	 * LD (large decrementer).
2083 	 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
2084 	 */
2085 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
2086 
2087 	/* Broken 32-bit version of LPCR must not clear top bits */
2088 	if (preserve_top32)
2089 		mask &= 0xFFFFFFFF;
2090 
2091 	new_lpcr = kvmppc_filter_lpcr_hv(kvm,
2092 			(vc->lpcr & ~mask) | (new_lpcr & mask));
2093 
2094 	/*
2095 	 * If ILE (interrupt little-endian) has changed, update the
2096 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
2097 	 */
2098 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
2099 		struct kvm_vcpu *vcpu;
2100 		unsigned long i;
2101 
2102 		kvm_for_each_vcpu(i, vcpu, kvm) {
2103 			if (vcpu->arch.vcore != vc)
2104 				continue;
2105 			if (new_lpcr & LPCR_ILE)
2106 				vcpu->arch.intr_msr |= MSR_LE;
2107 			else
2108 				vcpu->arch.intr_msr &= ~MSR_LE;
2109 		}
2110 	}
2111 
2112 	vc->lpcr = new_lpcr;
2113 
2114 	spin_unlock(&vc->lock);
2115 }
2116 
2117 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2118 				 union kvmppc_one_reg *val)
2119 {
2120 	int r = 0;
2121 	long int i;
2122 
2123 	switch (id) {
2124 	case KVM_REG_PPC_DEBUG_INST:
2125 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2126 		break;
2127 	case KVM_REG_PPC_HIOR:
2128 		*val = get_reg_val(id, 0);
2129 		break;
2130 	case KVM_REG_PPC_DABR:
2131 		*val = get_reg_val(id, vcpu->arch.dabr);
2132 		break;
2133 	case KVM_REG_PPC_DABRX:
2134 		*val = get_reg_val(id, vcpu->arch.dabrx);
2135 		break;
2136 	case KVM_REG_PPC_DSCR:
2137 		*val = get_reg_val(id, vcpu->arch.dscr);
2138 		break;
2139 	case KVM_REG_PPC_PURR:
2140 		*val = get_reg_val(id, vcpu->arch.purr);
2141 		break;
2142 	case KVM_REG_PPC_SPURR:
2143 		*val = get_reg_val(id, vcpu->arch.spurr);
2144 		break;
2145 	case KVM_REG_PPC_AMR:
2146 		*val = get_reg_val(id, vcpu->arch.amr);
2147 		break;
2148 	case KVM_REG_PPC_UAMOR:
2149 		*val = get_reg_val(id, vcpu->arch.uamor);
2150 		break;
2151 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2152 		i = id - KVM_REG_PPC_MMCR0;
2153 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
2154 		break;
2155 	case KVM_REG_PPC_MMCR2:
2156 		*val = get_reg_val(id, vcpu->arch.mmcr[2]);
2157 		break;
2158 	case KVM_REG_PPC_MMCRA:
2159 		*val = get_reg_val(id, vcpu->arch.mmcra);
2160 		break;
2161 	case KVM_REG_PPC_MMCRS:
2162 		*val = get_reg_val(id, vcpu->arch.mmcrs);
2163 		break;
2164 	case KVM_REG_PPC_MMCR3:
2165 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
2166 		break;
2167 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2168 		i = id - KVM_REG_PPC_PMC1;
2169 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
2170 		break;
2171 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2172 		i = id - KVM_REG_PPC_SPMC1;
2173 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
2174 		break;
2175 	case KVM_REG_PPC_SIAR:
2176 		*val = get_reg_val(id, vcpu->arch.siar);
2177 		break;
2178 	case KVM_REG_PPC_SDAR:
2179 		*val = get_reg_val(id, vcpu->arch.sdar);
2180 		break;
2181 	case KVM_REG_PPC_SIER:
2182 		*val = get_reg_val(id, vcpu->arch.sier[0]);
2183 		break;
2184 	case KVM_REG_PPC_SIER2:
2185 		*val = get_reg_val(id, vcpu->arch.sier[1]);
2186 		break;
2187 	case KVM_REG_PPC_SIER3:
2188 		*val = get_reg_val(id, vcpu->arch.sier[2]);
2189 		break;
2190 	case KVM_REG_PPC_IAMR:
2191 		*val = get_reg_val(id, vcpu->arch.iamr);
2192 		break;
2193 	case KVM_REG_PPC_PSPB:
2194 		*val = get_reg_val(id, vcpu->arch.pspb);
2195 		break;
2196 	case KVM_REG_PPC_DPDES:
2197 		/*
2198 		 * On POWER9, where we are emulating msgsndp etc.,
2199 		 * we return 1 bit for each vcpu, which can come from
2200 		 * either vcore->dpdes or doorbell_request.
2201 		 * On POWER8, doorbell_request is 0.
2202 		 */
2203 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2204 			*val = get_reg_val(id, vcpu->arch.doorbell_request);
2205 		else
2206 			*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
2207 		break;
2208 	case KVM_REG_PPC_VTB:
2209 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
2210 		break;
2211 	case KVM_REG_PPC_DAWR:
2212 		*val = get_reg_val(id, vcpu->arch.dawr0);
2213 		break;
2214 	case KVM_REG_PPC_DAWRX:
2215 		*val = get_reg_val(id, vcpu->arch.dawrx0);
2216 		break;
2217 	case KVM_REG_PPC_DAWR1:
2218 		*val = get_reg_val(id, vcpu->arch.dawr1);
2219 		break;
2220 	case KVM_REG_PPC_DAWRX1:
2221 		*val = get_reg_val(id, vcpu->arch.dawrx1);
2222 		break;
2223 	case KVM_REG_PPC_CIABR:
2224 		*val = get_reg_val(id, vcpu->arch.ciabr);
2225 		break;
2226 	case KVM_REG_PPC_CSIGR:
2227 		*val = get_reg_val(id, vcpu->arch.csigr);
2228 		break;
2229 	case KVM_REG_PPC_TACR:
2230 		*val = get_reg_val(id, vcpu->arch.tacr);
2231 		break;
2232 	case KVM_REG_PPC_TCSCR:
2233 		*val = get_reg_val(id, vcpu->arch.tcscr);
2234 		break;
2235 	case KVM_REG_PPC_PID:
2236 		*val = get_reg_val(id, vcpu->arch.pid);
2237 		break;
2238 	case KVM_REG_PPC_ACOP:
2239 		*val = get_reg_val(id, vcpu->arch.acop);
2240 		break;
2241 	case KVM_REG_PPC_WORT:
2242 		*val = get_reg_val(id, vcpu->arch.wort);
2243 		break;
2244 	case KVM_REG_PPC_TIDR:
2245 		*val = get_reg_val(id, vcpu->arch.tid);
2246 		break;
2247 	case KVM_REG_PPC_PSSCR:
2248 		*val = get_reg_val(id, vcpu->arch.psscr);
2249 		break;
2250 	case KVM_REG_PPC_VPA_ADDR:
2251 		spin_lock(&vcpu->arch.vpa_update_lock);
2252 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2253 		spin_unlock(&vcpu->arch.vpa_update_lock);
2254 		break;
2255 	case KVM_REG_PPC_VPA_SLB:
2256 		spin_lock(&vcpu->arch.vpa_update_lock);
2257 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2258 		val->vpaval.length = vcpu->arch.slb_shadow.len;
2259 		spin_unlock(&vcpu->arch.vpa_update_lock);
2260 		break;
2261 	case KVM_REG_PPC_VPA_DTL:
2262 		spin_lock(&vcpu->arch.vpa_update_lock);
2263 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2264 		val->vpaval.length = vcpu->arch.dtl.len;
2265 		spin_unlock(&vcpu->arch.vpa_update_lock);
2266 		break;
2267 	case KVM_REG_PPC_TB_OFFSET:
2268 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
2269 		break;
2270 	case KVM_REG_PPC_LPCR:
2271 	case KVM_REG_PPC_LPCR_64:
2272 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
2273 		break;
2274 	case KVM_REG_PPC_PPR:
2275 		*val = get_reg_val(id, vcpu->arch.ppr);
2276 		break;
2277 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2278 	case KVM_REG_PPC_TFHAR:
2279 		*val = get_reg_val(id, vcpu->arch.tfhar);
2280 		break;
2281 	case KVM_REG_PPC_TFIAR:
2282 		*val = get_reg_val(id, vcpu->arch.tfiar);
2283 		break;
2284 	case KVM_REG_PPC_TEXASR:
2285 		*val = get_reg_val(id, vcpu->arch.texasr);
2286 		break;
2287 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2288 		i = id - KVM_REG_PPC_TM_GPR0;
2289 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2290 		break;
2291 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2292 	{
2293 		int j;
2294 		i = id - KVM_REG_PPC_TM_VSR0;
2295 		if (i < 32)
2296 			for (j = 0; j < TS_FPRWIDTH; j++)
2297 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2298 		else {
2299 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2300 				val->vval = vcpu->arch.vr_tm.vr[i-32];
2301 			else
2302 				r = -ENXIO;
2303 		}
2304 		break;
2305 	}
2306 	case KVM_REG_PPC_TM_CR:
2307 		*val = get_reg_val(id, vcpu->arch.cr_tm);
2308 		break;
2309 	case KVM_REG_PPC_TM_XER:
2310 		*val = get_reg_val(id, vcpu->arch.xer_tm);
2311 		break;
2312 	case KVM_REG_PPC_TM_LR:
2313 		*val = get_reg_val(id, vcpu->arch.lr_tm);
2314 		break;
2315 	case KVM_REG_PPC_TM_CTR:
2316 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
2317 		break;
2318 	case KVM_REG_PPC_TM_FPSCR:
2319 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2320 		break;
2321 	case KVM_REG_PPC_TM_AMR:
2322 		*val = get_reg_val(id, vcpu->arch.amr_tm);
2323 		break;
2324 	case KVM_REG_PPC_TM_PPR:
2325 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
2326 		break;
2327 	case KVM_REG_PPC_TM_VRSAVE:
2328 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
2329 		break;
2330 	case KVM_REG_PPC_TM_VSCR:
2331 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2332 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2333 		else
2334 			r = -ENXIO;
2335 		break;
2336 	case KVM_REG_PPC_TM_DSCR:
2337 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
2338 		break;
2339 	case KVM_REG_PPC_TM_TAR:
2340 		*val = get_reg_val(id, vcpu->arch.tar_tm);
2341 		break;
2342 #endif
2343 	case KVM_REG_PPC_ARCH_COMPAT:
2344 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
2345 		break;
2346 	case KVM_REG_PPC_DEC_EXPIRY:
2347 		*val = get_reg_val(id, vcpu->arch.dec_expires);
2348 		break;
2349 	case KVM_REG_PPC_ONLINE:
2350 		*val = get_reg_val(id, vcpu->arch.online);
2351 		break;
2352 	case KVM_REG_PPC_PTCR:
2353 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2354 		break;
2355 	default:
2356 		r = -EINVAL;
2357 		break;
2358 	}
2359 
2360 	return r;
2361 }
2362 
2363 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2364 				 union kvmppc_one_reg *val)
2365 {
2366 	int r = 0;
2367 	long int i;
2368 	unsigned long addr, len;
2369 
2370 	switch (id) {
2371 	case KVM_REG_PPC_HIOR:
2372 		/* Only allow this to be set to zero */
2373 		if (set_reg_val(id, *val))
2374 			r = -EINVAL;
2375 		break;
2376 	case KVM_REG_PPC_DABR:
2377 		vcpu->arch.dabr = set_reg_val(id, *val);
2378 		break;
2379 	case KVM_REG_PPC_DABRX:
2380 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2381 		break;
2382 	case KVM_REG_PPC_DSCR:
2383 		vcpu->arch.dscr = set_reg_val(id, *val);
2384 		break;
2385 	case KVM_REG_PPC_PURR:
2386 		vcpu->arch.purr = set_reg_val(id, *val);
2387 		break;
2388 	case KVM_REG_PPC_SPURR:
2389 		vcpu->arch.spurr = set_reg_val(id, *val);
2390 		break;
2391 	case KVM_REG_PPC_AMR:
2392 		vcpu->arch.amr = set_reg_val(id, *val);
2393 		break;
2394 	case KVM_REG_PPC_UAMOR:
2395 		vcpu->arch.uamor = set_reg_val(id, *val);
2396 		break;
2397 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2398 		i = id - KVM_REG_PPC_MMCR0;
2399 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2400 		break;
2401 	case KVM_REG_PPC_MMCR2:
2402 		vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2403 		break;
2404 	case KVM_REG_PPC_MMCRA:
2405 		vcpu->arch.mmcra = set_reg_val(id, *val);
2406 		break;
2407 	case KVM_REG_PPC_MMCRS:
2408 		vcpu->arch.mmcrs = set_reg_val(id, *val);
2409 		break;
2410 	case KVM_REG_PPC_MMCR3:
2411 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
2412 		break;
2413 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2414 		i = id - KVM_REG_PPC_PMC1;
2415 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
2416 		break;
2417 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2418 		i = id - KVM_REG_PPC_SPMC1;
2419 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
2420 		break;
2421 	case KVM_REG_PPC_SIAR:
2422 		vcpu->arch.siar = set_reg_val(id, *val);
2423 		break;
2424 	case KVM_REG_PPC_SDAR:
2425 		vcpu->arch.sdar = set_reg_val(id, *val);
2426 		break;
2427 	case KVM_REG_PPC_SIER:
2428 		vcpu->arch.sier[0] = set_reg_val(id, *val);
2429 		break;
2430 	case KVM_REG_PPC_SIER2:
2431 		vcpu->arch.sier[1] = set_reg_val(id, *val);
2432 		break;
2433 	case KVM_REG_PPC_SIER3:
2434 		vcpu->arch.sier[2] = set_reg_val(id, *val);
2435 		break;
2436 	case KVM_REG_PPC_IAMR:
2437 		vcpu->arch.iamr = set_reg_val(id, *val);
2438 		break;
2439 	case KVM_REG_PPC_PSPB:
2440 		vcpu->arch.pspb = set_reg_val(id, *val);
2441 		break;
2442 	case KVM_REG_PPC_DPDES:
2443 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2444 			vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1;
2445 		else
2446 			vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2447 		break;
2448 	case KVM_REG_PPC_VTB:
2449 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2450 		break;
2451 	case KVM_REG_PPC_DAWR:
2452 		vcpu->arch.dawr0 = set_reg_val(id, *val);
2453 		break;
2454 	case KVM_REG_PPC_DAWRX:
2455 		vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2456 		break;
2457 	case KVM_REG_PPC_DAWR1:
2458 		vcpu->arch.dawr1 = set_reg_val(id, *val);
2459 		break;
2460 	case KVM_REG_PPC_DAWRX1:
2461 		vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2462 		break;
2463 	case KVM_REG_PPC_CIABR:
2464 		vcpu->arch.ciabr = set_reg_val(id, *val);
2465 		/* Don't allow setting breakpoints in hypervisor code */
2466 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2467 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
2468 		break;
2469 	case KVM_REG_PPC_CSIGR:
2470 		vcpu->arch.csigr = set_reg_val(id, *val);
2471 		break;
2472 	case KVM_REG_PPC_TACR:
2473 		vcpu->arch.tacr = set_reg_val(id, *val);
2474 		break;
2475 	case KVM_REG_PPC_TCSCR:
2476 		vcpu->arch.tcscr = set_reg_val(id, *val);
2477 		break;
2478 	case KVM_REG_PPC_PID:
2479 		vcpu->arch.pid = set_reg_val(id, *val);
2480 		break;
2481 	case KVM_REG_PPC_ACOP:
2482 		vcpu->arch.acop = set_reg_val(id, *val);
2483 		break;
2484 	case KVM_REG_PPC_WORT:
2485 		vcpu->arch.wort = set_reg_val(id, *val);
2486 		break;
2487 	case KVM_REG_PPC_TIDR:
2488 		vcpu->arch.tid = set_reg_val(id, *val);
2489 		break;
2490 	case KVM_REG_PPC_PSSCR:
2491 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2492 		break;
2493 	case KVM_REG_PPC_VPA_ADDR:
2494 		addr = set_reg_val(id, *val);
2495 		r = -EINVAL;
2496 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2497 			      vcpu->arch.dtl.next_gpa))
2498 			break;
2499 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2500 		break;
2501 	case KVM_REG_PPC_VPA_SLB:
2502 		addr = val->vpaval.addr;
2503 		len = val->vpaval.length;
2504 		r = -EINVAL;
2505 		if (addr && !vcpu->arch.vpa.next_gpa)
2506 			break;
2507 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2508 		break;
2509 	case KVM_REG_PPC_VPA_DTL:
2510 		addr = val->vpaval.addr;
2511 		len = val->vpaval.length;
2512 		r = -EINVAL;
2513 		if (addr && (len < sizeof(struct dtl_entry) ||
2514 			     !vcpu->arch.vpa.next_gpa))
2515 			break;
2516 		len -= len % sizeof(struct dtl_entry);
2517 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2518 		break;
2519 	case KVM_REG_PPC_TB_OFFSET:
2520 		/* round up to multiple of 2^24 */
2521 		vcpu->arch.vcore->tb_offset =
2522 			ALIGN(set_reg_val(id, *val), 1UL << 24);
2523 		break;
2524 	case KVM_REG_PPC_LPCR:
2525 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2526 		break;
2527 	case KVM_REG_PPC_LPCR_64:
2528 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2529 		break;
2530 	case KVM_REG_PPC_PPR:
2531 		vcpu->arch.ppr = set_reg_val(id, *val);
2532 		break;
2533 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2534 	case KVM_REG_PPC_TFHAR:
2535 		vcpu->arch.tfhar = set_reg_val(id, *val);
2536 		break;
2537 	case KVM_REG_PPC_TFIAR:
2538 		vcpu->arch.tfiar = set_reg_val(id, *val);
2539 		break;
2540 	case KVM_REG_PPC_TEXASR:
2541 		vcpu->arch.texasr = set_reg_val(id, *val);
2542 		break;
2543 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2544 		i = id - KVM_REG_PPC_TM_GPR0;
2545 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2546 		break;
2547 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2548 	{
2549 		int j;
2550 		i = id - KVM_REG_PPC_TM_VSR0;
2551 		if (i < 32)
2552 			for (j = 0; j < TS_FPRWIDTH; j++)
2553 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2554 		else
2555 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2556 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2557 			else
2558 				r = -ENXIO;
2559 		break;
2560 	}
2561 	case KVM_REG_PPC_TM_CR:
2562 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2563 		break;
2564 	case KVM_REG_PPC_TM_XER:
2565 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2566 		break;
2567 	case KVM_REG_PPC_TM_LR:
2568 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2569 		break;
2570 	case KVM_REG_PPC_TM_CTR:
2571 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2572 		break;
2573 	case KVM_REG_PPC_TM_FPSCR:
2574 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2575 		break;
2576 	case KVM_REG_PPC_TM_AMR:
2577 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2578 		break;
2579 	case KVM_REG_PPC_TM_PPR:
2580 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2581 		break;
2582 	case KVM_REG_PPC_TM_VRSAVE:
2583 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2584 		break;
2585 	case KVM_REG_PPC_TM_VSCR:
2586 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2587 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2588 		else
2589 			r = - ENXIO;
2590 		break;
2591 	case KVM_REG_PPC_TM_DSCR:
2592 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2593 		break;
2594 	case KVM_REG_PPC_TM_TAR:
2595 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2596 		break;
2597 #endif
2598 	case KVM_REG_PPC_ARCH_COMPAT:
2599 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2600 		break;
2601 	case KVM_REG_PPC_DEC_EXPIRY:
2602 		vcpu->arch.dec_expires = set_reg_val(id, *val);
2603 		break;
2604 	case KVM_REG_PPC_ONLINE:
2605 		i = set_reg_val(id, *val);
2606 		if (i && !vcpu->arch.online)
2607 			atomic_inc(&vcpu->arch.vcore->online_count);
2608 		else if (!i && vcpu->arch.online)
2609 			atomic_dec(&vcpu->arch.vcore->online_count);
2610 		vcpu->arch.online = i;
2611 		break;
2612 	case KVM_REG_PPC_PTCR:
2613 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2614 		break;
2615 	default:
2616 		r = -EINVAL;
2617 		break;
2618 	}
2619 
2620 	return r;
2621 }
2622 
2623 /*
2624  * On POWER9, threads are independent and can be in different partitions.
2625  * Therefore we consider each thread to be a subcore.
2626  * There is a restriction that all threads have to be in the same
2627  * MMU mode (radix or HPT), unfortunately, but since we only support
2628  * HPT guests on a HPT host so far, that isn't an impediment yet.
2629  */
2630 static int threads_per_vcore(struct kvm *kvm)
2631 {
2632 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2633 		return 1;
2634 	return threads_per_subcore;
2635 }
2636 
2637 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2638 {
2639 	struct kvmppc_vcore *vcore;
2640 
2641 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2642 
2643 	if (vcore == NULL)
2644 		return NULL;
2645 
2646 	spin_lock_init(&vcore->lock);
2647 	spin_lock_init(&vcore->stoltb_lock);
2648 	rcuwait_init(&vcore->wait);
2649 	vcore->preempt_tb = TB_NIL;
2650 	vcore->lpcr = kvm->arch.lpcr;
2651 	vcore->first_vcpuid = id;
2652 	vcore->kvm = kvm;
2653 	INIT_LIST_HEAD(&vcore->preempt_list);
2654 
2655 	return vcore;
2656 }
2657 
2658 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2659 static struct debugfs_timings_element {
2660 	const char *name;
2661 	size_t offset;
2662 } timings[] = {
2663 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2664 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2665 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2666 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2667 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2668 };
2669 
2670 #define N_TIMINGS	(ARRAY_SIZE(timings))
2671 
2672 struct debugfs_timings_state {
2673 	struct kvm_vcpu	*vcpu;
2674 	unsigned int	buflen;
2675 	char		buf[N_TIMINGS * 100];
2676 };
2677 
2678 static int debugfs_timings_open(struct inode *inode, struct file *file)
2679 {
2680 	struct kvm_vcpu *vcpu = inode->i_private;
2681 	struct debugfs_timings_state *p;
2682 
2683 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2684 	if (!p)
2685 		return -ENOMEM;
2686 
2687 	kvm_get_kvm(vcpu->kvm);
2688 	p->vcpu = vcpu;
2689 	file->private_data = p;
2690 
2691 	return nonseekable_open(inode, file);
2692 }
2693 
2694 static int debugfs_timings_release(struct inode *inode, struct file *file)
2695 {
2696 	struct debugfs_timings_state *p = file->private_data;
2697 
2698 	kvm_put_kvm(p->vcpu->kvm);
2699 	kfree(p);
2700 	return 0;
2701 }
2702 
2703 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2704 				    size_t len, loff_t *ppos)
2705 {
2706 	struct debugfs_timings_state *p = file->private_data;
2707 	struct kvm_vcpu *vcpu = p->vcpu;
2708 	char *s, *buf_end;
2709 	struct kvmhv_tb_accumulator tb;
2710 	u64 count;
2711 	loff_t pos;
2712 	ssize_t n;
2713 	int i, loops;
2714 	bool ok;
2715 
2716 	if (!p->buflen) {
2717 		s = p->buf;
2718 		buf_end = s + sizeof(p->buf);
2719 		for (i = 0; i < N_TIMINGS; ++i) {
2720 			struct kvmhv_tb_accumulator *acc;
2721 
2722 			acc = (struct kvmhv_tb_accumulator *)
2723 				((unsigned long)vcpu + timings[i].offset);
2724 			ok = false;
2725 			for (loops = 0; loops < 1000; ++loops) {
2726 				count = acc->seqcount;
2727 				if (!(count & 1)) {
2728 					smp_rmb();
2729 					tb = *acc;
2730 					smp_rmb();
2731 					if (count == acc->seqcount) {
2732 						ok = true;
2733 						break;
2734 					}
2735 				}
2736 				udelay(1);
2737 			}
2738 			if (!ok)
2739 				snprintf(s, buf_end - s, "%s: stuck\n",
2740 					timings[i].name);
2741 			else
2742 				snprintf(s, buf_end - s,
2743 					"%s: %llu %llu %llu %llu\n",
2744 					timings[i].name, count / 2,
2745 					tb_to_ns(tb.tb_total),
2746 					tb_to_ns(tb.tb_min),
2747 					tb_to_ns(tb.tb_max));
2748 			s += strlen(s);
2749 		}
2750 		p->buflen = s - p->buf;
2751 	}
2752 
2753 	pos = *ppos;
2754 	if (pos >= p->buflen)
2755 		return 0;
2756 	if (len > p->buflen - pos)
2757 		len = p->buflen - pos;
2758 	n = copy_to_user(buf, p->buf + pos, len);
2759 	if (n) {
2760 		if (n == len)
2761 			return -EFAULT;
2762 		len -= n;
2763 	}
2764 	*ppos = pos + len;
2765 	return len;
2766 }
2767 
2768 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2769 				     size_t len, loff_t *ppos)
2770 {
2771 	return -EACCES;
2772 }
2773 
2774 static const struct file_operations debugfs_timings_ops = {
2775 	.owner	 = THIS_MODULE,
2776 	.open	 = debugfs_timings_open,
2777 	.release = debugfs_timings_release,
2778 	.read	 = debugfs_timings_read,
2779 	.write	 = debugfs_timings_write,
2780 	.llseek	 = generic_file_llseek,
2781 };
2782 
2783 /* Create a debugfs directory for the vcpu */
2784 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2785 {
2786 	debugfs_create_file("timings", 0444, debugfs_dentry, vcpu,
2787 			    &debugfs_timings_ops);
2788 	return 0;
2789 }
2790 
2791 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2792 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2793 {
2794 	return 0;
2795 }
2796 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2797 
2798 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2799 {
2800 	int err;
2801 	int core;
2802 	struct kvmppc_vcore *vcore;
2803 	struct kvm *kvm;
2804 	unsigned int id;
2805 
2806 	kvm = vcpu->kvm;
2807 	id = vcpu->vcpu_id;
2808 
2809 	vcpu->arch.shared = &vcpu->arch.shregs;
2810 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2811 	/*
2812 	 * The shared struct is never shared on HV,
2813 	 * so we can always use host endianness
2814 	 */
2815 #ifdef __BIG_ENDIAN__
2816 	vcpu->arch.shared_big_endian = true;
2817 #else
2818 	vcpu->arch.shared_big_endian = false;
2819 #endif
2820 #endif
2821 	vcpu->arch.mmcr[0] = MMCR0_FC;
2822 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
2823 		vcpu->arch.mmcr[0] |= MMCR0_PMCCEXT;
2824 		vcpu->arch.mmcra = MMCRA_BHRB_DISABLE;
2825 	}
2826 
2827 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2828 	/* default to host PVR, since we can't spoof it */
2829 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2830 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2831 	spin_lock_init(&vcpu->arch.tbacct_lock);
2832 	vcpu->arch.busy_preempt = TB_NIL;
2833 	vcpu->arch.shregs.msr = MSR_ME;
2834 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2835 
2836 	/*
2837 	 * Set the default HFSCR for the guest from the host value.
2838 	 * This value is only used on POWER9.
2839 	 * On POWER9, we want to virtualize the doorbell facility, so we
2840 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2841 	 * to trap and then we emulate them.
2842 	 */
2843 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2844 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2845 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2846 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2847 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2848 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2849 			vcpu->arch.hfscr |= HFSCR_TM;
2850 #endif
2851 	}
2852 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2853 		vcpu->arch.hfscr |= HFSCR_TM;
2854 
2855 	vcpu->arch.hfscr_permitted = vcpu->arch.hfscr;
2856 
2857 	/*
2858 	 * PM, EBB, TM are demand-faulted so start with it clear.
2859 	 */
2860 	vcpu->arch.hfscr &= ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM);
2861 
2862 	kvmppc_mmu_book3s_hv_init(vcpu);
2863 
2864 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2865 
2866 	init_waitqueue_head(&vcpu->arch.cpu_run);
2867 
2868 	mutex_lock(&kvm->lock);
2869 	vcore = NULL;
2870 	err = -EINVAL;
2871 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2872 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2873 			pr_devel("KVM: VCPU ID too high\n");
2874 			core = KVM_MAX_VCORES;
2875 		} else {
2876 			BUG_ON(kvm->arch.smt_mode != 1);
2877 			core = kvmppc_pack_vcpu_id(kvm, id);
2878 		}
2879 	} else {
2880 		core = id / kvm->arch.smt_mode;
2881 	}
2882 	if (core < KVM_MAX_VCORES) {
2883 		vcore = kvm->arch.vcores[core];
2884 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2885 			pr_devel("KVM: collision on id %u", id);
2886 			vcore = NULL;
2887 		} else if (!vcore) {
2888 			/*
2889 			 * Take mmu_setup_lock for mutual exclusion
2890 			 * with kvmppc_update_lpcr().
2891 			 */
2892 			err = -ENOMEM;
2893 			vcore = kvmppc_vcore_create(kvm,
2894 					id & ~(kvm->arch.smt_mode - 1));
2895 			mutex_lock(&kvm->arch.mmu_setup_lock);
2896 			kvm->arch.vcores[core] = vcore;
2897 			kvm->arch.online_vcores++;
2898 			mutex_unlock(&kvm->arch.mmu_setup_lock);
2899 		}
2900 	}
2901 	mutex_unlock(&kvm->lock);
2902 
2903 	if (!vcore)
2904 		return err;
2905 
2906 	spin_lock(&vcore->lock);
2907 	++vcore->num_threads;
2908 	spin_unlock(&vcore->lock);
2909 	vcpu->arch.vcore = vcore;
2910 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2911 	vcpu->arch.thread_cpu = -1;
2912 	vcpu->arch.prev_cpu = -1;
2913 
2914 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2915 	kvmppc_sanity_check(vcpu);
2916 
2917 	return 0;
2918 }
2919 
2920 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2921 			      unsigned long flags)
2922 {
2923 	int err;
2924 	int esmt = 0;
2925 
2926 	if (flags)
2927 		return -EINVAL;
2928 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2929 		return -EINVAL;
2930 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2931 		/*
2932 		 * On POWER8 (or POWER7), the threading mode is "strict",
2933 		 * so we pack smt_mode vcpus per vcore.
2934 		 */
2935 		if (smt_mode > threads_per_subcore)
2936 			return -EINVAL;
2937 	} else {
2938 		/*
2939 		 * On POWER9, the threading mode is "loose",
2940 		 * so each vcpu gets its own vcore.
2941 		 */
2942 		esmt = smt_mode;
2943 		smt_mode = 1;
2944 	}
2945 	mutex_lock(&kvm->lock);
2946 	err = -EBUSY;
2947 	if (!kvm->arch.online_vcores) {
2948 		kvm->arch.smt_mode = smt_mode;
2949 		kvm->arch.emul_smt_mode = esmt;
2950 		err = 0;
2951 	}
2952 	mutex_unlock(&kvm->lock);
2953 
2954 	return err;
2955 }
2956 
2957 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2958 {
2959 	if (vpa->pinned_addr)
2960 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2961 					vpa->dirty);
2962 }
2963 
2964 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2965 {
2966 	spin_lock(&vcpu->arch.vpa_update_lock);
2967 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2968 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2969 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2970 	spin_unlock(&vcpu->arch.vpa_update_lock);
2971 }
2972 
2973 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2974 {
2975 	/* Indicate we want to get back into the guest */
2976 	return 1;
2977 }
2978 
2979 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2980 {
2981 	unsigned long dec_nsec, now;
2982 
2983 	now = get_tb();
2984 	if (now > kvmppc_dec_expires_host_tb(vcpu)) {
2985 		/* decrementer has already gone negative */
2986 		kvmppc_core_queue_dec(vcpu);
2987 		kvmppc_core_prepare_to_enter(vcpu);
2988 		return;
2989 	}
2990 	dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now);
2991 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2992 	vcpu->arch.timer_running = 1;
2993 }
2994 
2995 extern int __kvmppc_vcore_entry(void);
2996 
2997 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2998 				   struct kvm_vcpu *vcpu, u64 tb)
2999 {
3000 	u64 now;
3001 
3002 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3003 		return;
3004 	spin_lock_irq(&vcpu->arch.tbacct_lock);
3005 	now = tb;
3006 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
3007 		vcpu->arch.stolen_logged;
3008 	vcpu->arch.busy_preempt = now;
3009 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3010 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
3011 	--vc->n_runnable;
3012 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
3013 }
3014 
3015 static int kvmppc_grab_hwthread(int cpu)
3016 {
3017 	struct paca_struct *tpaca;
3018 	long timeout = 10000;
3019 
3020 	tpaca = paca_ptrs[cpu];
3021 
3022 	/* Ensure the thread won't go into the kernel if it wakes */
3023 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3024 	tpaca->kvm_hstate.kvm_vcore = NULL;
3025 	tpaca->kvm_hstate.napping = 0;
3026 	smp_wmb();
3027 	tpaca->kvm_hstate.hwthread_req = 1;
3028 
3029 	/*
3030 	 * If the thread is already executing in the kernel (e.g. handling
3031 	 * a stray interrupt), wait for it to get back to nap mode.
3032 	 * The smp_mb() is to ensure that our setting of hwthread_req
3033 	 * is visible before we look at hwthread_state, so if this
3034 	 * races with the code at system_reset_pSeries and the thread
3035 	 * misses our setting of hwthread_req, we are sure to see its
3036 	 * setting of hwthread_state, and vice versa.
3037 	 */
3038 	smp_mb();
3039 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
3040 		if (--timeout <= 0) {
3041 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
3042 			return -EBUSY;
3043 		}
3044 		udelay(1);
3045 	}
3046 	return 0;
3047 }
3048 
3049 static void kvmppc_release_hwthread(int cpu)
3050 {
3051 	struct paca_struct *tpaca;
3052 
3053 	tpaca = paca_ptrs[cpu];
3054 	tpaca->kvm_hstate.hwthread_req = 0;
3055 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3056 	tpaca->kvm_hstate.kvm_vcore = NULL;
3057 	tpaca->kvm_hstate.kvm_split_mode = NULL;
3058 }
3059 
3060 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest);
3061 
3062 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
3063 {
3064 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3065 	cpumask_t *need_tlb_flush;
3066 	int i;
3067 
3068 	if (nested)
3069 		need_tlb_flush = &nested->need_tlb_flush;
3070 	else
3071 		need_tlb_flush = &kvm->arch.need_tlb_flush;
3072 
3073 	cpu = cpu_first_tlb_thread_sibling(cpu);
3074 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3075 					i += cpu_tlb_thread_sibling_step())
3076 		cpumask_set_cpu(i, need_tlb_flush);
3077 
3078 	/*
3079 	 * Make sure setting of bit in need_tlb_flush precedes testing of
3080 	 * cpu_in_guest. The matching barrier on the other side is hwsync
3081 	 * when switching to guest MMU mode, which happens between
3082 	 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit
3083 	 * being tested.
3084 	 */
3085 	smp_mb();
3086 
3087 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3088 					i += cpu_tlb_thread_sibling_step()) {
3089 		struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i);
3090 
3091 		if (running == kvm)
3092 			smp_call_function_single(i, do_nothing, NULL, 1);
3093 	}
3094 }
3095 
3096 static void do_migrate_away_vcpu(void *arg)
3097 {
3098 	struct kvm_vcpu *vcpu = arg;
3099 	struct kvm *kvm = vcpu->kvm;
3100 
3101 	/*
3102 	 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
3103 	 * ptesync sequence on the old CPU before migrating to a new one, in
3104 	 * case we interrupted the guest between a tlbie ; eieio ;
3105 	 * tlbsync; ptesync sequence.
3106 	 *
3107 	 * Otherwise, ptesync is sufficient for ordering tlbiel sequences.
3108 	 */
3109 	if (kvm->arch.lpcr & LPCR_GTSE)
3110 		asm volatile("eieio; tlbsync; ptesync");
3111 	else
3112 		asm volatile("ptesync");
3113 }
3114 
3115 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
3116 {
3117 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3118 	struct kvm *kvm = vcpu->kvm;
3119 	int prev_cpu;
3120 
3121 	if (!cpu_has_feature(CPU_FTR_HVMODE))
3122 		return;
3123 
3124 	if (nested)
3125 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
3126 	else
3127 		prev_cpu = vcpu->arch.prev_cpu;
3128 
3129 	/*
3130 	 * With radix, the guest can do TLB invalidations itself,
3131 	 * and it could choose to use the local form (tlbiel) if
3132 	 * it is invalidating a translation that has only ever been
3133 	 * used on one vcpu.  However, that doesn't mean it has
3134 	 * only ever been used on one physical cpu, since vcpus
3135 	 * can move around between pcpus.  To cope with this, when
3136 	 * a vcpu moves from one pcpu to another, we need to tell
3137 	 * any vcpus running on the same core as this vcpu previously
3138 	 * ran to flush the TLB.
3139 	 */
3140 	if (prev_cpu != pcpu) {
3141 		if (prev_cpu >= 0) {
3142 			if (cpu_first_tlb_thread_sibling(prev_cpu) !=
3143 			    cpu_first_tlb_thread_sibling(pcpu))
3144 				radix_flush_cpu(kvm, prev_cpu, vcpu);
3145 
3146 			smp_call_function_single(prev_cpu,
3147 					do_migrate_away_vcpu, vcpu, 1);
3148 		}
3149 		if (nested)
3150 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3151 		else
3152 			vcpu->arch.prev_cpu = pcpu;
3153 	}
3154 }
3155 
3156 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3157 {
3158 	int cpu;
3159 	struct paca_struct *tpaca;
3160 
3161 	cpu = vc->pcpu;
3162 	if (vcpu) {
3163 		if (vcpu->arch.timer_running) {
3164 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3165 			vcpu->arch.timer_running = 0;
3166 		}
3167 		cpu += vcpu->arch.ptid;
3168 		vcpu->cpu = vc->pcpu;
3169 		vcpu->arch.thread_cpu = cpu;
3170 	}
3171 	tpaca = paca_ptrs[cpu];
3172 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
3173 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3174 	tpaca->kvm_hstate.fake_suspend = 0;
3175 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3176 	smp_wmb();
3177 	tpaca->kvm_hstate.kvm_vcore = vc;
3178 	if (cpu != smp_processor_id())
3179 		kvmppc_ipi_thread(cpu);
3180 }
3181 
3182 static void kvmppc_wait_for_nap(int n_threads)
3183 {
3184 	int cpu = smp_processor_id();
3185 	int i, loops;
3186 
3187 	if (n_threads <= 1)
3188 		return;
3189 	for (loops = 0; loops < 1000000; ++loops) {
3190 		/*
3191 		 * Check if all threads are finished.
3192 		 * We set the vcore pointer when starting a thread
3193 		 * and the thread clears it when finished, so we look
3194 		 * for any threads that still have a non-NULL vcore ptr.
3195 		 */
3196 		for (i = 1; i < n_threads; ++i)
3197 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3198 				break;
3199 		if (i == n_threads) {
3200 			HMT_medium();
3201 			return;
3202 		}
3203 		HMT_low();
3204 	}
3205 	HMT_medium();
3206 	for (i = 1; i < n_threads; ++i)
3207 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3208 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3209 }
3210 
3211 /*
3212  * Check that we are on thread 0 and that any other threads in
3213  * this core are off-line.  Then grab the threads so they can't
3214  * enter the kernel.
3215  */
3216 static int on_primary_thread(void)
3217 {
3218 	int cpu = smp_processor_id();
3219 	int thr;
3220 
3221 	/* Are we on a primary subcore? */
3222 	if (cpu_thread_in_subcore(cpu))
3223 		return 0;
3224 
3225 	thr = 0;
3226 	while (++thr < threads_per_subcore)
3227 		if (cpu_online(cpu + thr))
3228 			return 0;
3229 
3230 	/* Grab all hw threads so they can't go into the kernel */
3231 	for (thr = 1; thr < threads_per_subcore; ++thr) {
3232 		if (kvmppc_grab_hwthread(cpu + thr)) {
3233 			/* Couldn't grab one; let the others go */
3234 			do {
3235 				kvmppc_release_hwthread(cpu + thr);
3236 			} while (--thr > 0);
3237 			return 0;
3238 		}
3239 	}
3240 	return 1;
3241 }
3242 
3243 /*
3244  * A list of virtual cores for each physical CPU.
3245  * These are vcores that could run but their runner VCPU tasks are
3246  * (or may be) preempted.
3247  */
3248 struct preempted_vcore_list {
3249 	struct list_head	list;
3250 	spinlock_t		lock;
3251 };
3252 
3253 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3254 
3255 static void init_vcore_lists(void)
3256 {
3257 	int cpu;
3258 
3259 	for_each_possible_cpu(cpu) {
3260 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3261 		spin_lock_init(&lp->lock);
3262 		INIT_LIST_HEAD(&lp->list);
3263 	}
3264 }
3265 
3266 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3267 {
3268 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3269 
3270 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3271 
3272 	vc->vcore_state = VCORE_PREEMPT;
3273 	vc->pcpu = smp_processor_id();
3274 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3275 		spin_lock(&lp->lock);
3276 		list_add_tail(&vc->preempt_list, &lp->list);
3277 		spin_unlock(&lp->lock);
3278 	}
3279 
3280 	/* Start accumulating stolen time */
3281 	kvmppc_core_start_stolen(vc, mftb());
3282 }
3283 
3284 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3285 {
3286 	struct preempted_vcore_list *lp;
3287 
3288 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3289 
3290 	kvmppc_core_end_stolen(vc, mftb());
3291 	if (!list_empty(&vc->preempt_list)) {
3292 		lp = &per_cpu(preempted_vcores, vc->pcpu);
3293 		spin_lock(&lp->lock);
3294 		list_del_init(&vc->preempt_list);
3295 		spin_unlock(&lp->lock);
3296 	}
3297 	vc->vcore_state = VCORE_INACTIVE;
3298 }
3299 
3300 /*
3301  * This stores information about the virtual cores currently
3302  * assigned to a physical core.
3303  */
3304 struct core_info {
3305 	int		n_subcores;
3306 	int		max_subcore_threads;
3307 	int		total_threads;
3308 	int		subcore_threads[MAX_SUBCORES];
3309 	struct kvmppc_vcore *vc[MAX_SUBCORES];
3310 };
3311 
3312 /*
3313  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3314  * respectively in 2-way micro-threading (split-core) mode on POWER8.
3315  */
3316 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3317 
3318 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3319 {
3320 	memset(cip, 0, sizeof(*cip));
3321 	cip->n_subcores = 1;
3322 	cip->max_subcore_threads = vc->num_threads;
3323 	cip->total_threads = vc->num_threads;
3324 	cip->subcore_threads[0] = vc->num_threads;
3325 	cip->vc[0] = vc;
3326 }
3327 
3328 static bool subcore_config_ok(int n_subcores, int n_threads)
3329 {
3330 	/*
3331 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3332 	 * split-core mode, with one thread per subcore.
3333 	 */
3334 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3335 		return n_subcores <= 4 && n_threads == 1;
3336 
3337 	/* On POWER8, can only dynamically split if unsplit to begin with */
3338 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3339 		return false;
3340 	if (n_subcores > MAX_SUBCORES)
3341 		return false;
3342 	if (n_subcores > 1) {
3343 		if (!(dynamic_mt_modes & 2))
3344 			n_subcores = 4;
3345 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3346 			return false;
3347 	}
3348 
3349 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3350 }
3351 
3352 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3353 {
3354 	vc->entry_exit_map = 0;
3355 	vc->in_guest = 0;
3356 	vc->napping_threads = 0;
3357 	vc->conferring_threads = 0;
3358 	vc->tb_offset_applied = 0;
3359 }
3360 
3361 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3362 {
3363 	int n_threads = vc->num_threads;
3364 	int sub;
3365 
3366 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3367 		return false;
3368 
3369 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
3370 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3371 		return false;
3372 
3373 	if (n_threads < cip->max_subcore_threads)
3374 		n_threads = cip->max_subcore_threads;
3375 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3376 		return false;
3377 	cip->max_subcore_threads = n_threads;
3378 
3379 	sub = cip->n_subcores;
3380 	++cip->n_subcores;
3381 	cip->total_threads += vc->num_threads;
3382 	cip->subcore_threads[sub] = vc->num_threads;
3383 	cip->vc[sub] = vc;
3384 	init_vcore_to_run(vc);
3385 	list_del_init(&vc->preempt_list);
3386 
3387 	return true;
3388 }
3389 
3390 /*
3391  * Work out whether it is possible to piggyback the execution of
3392  * vcore *pvc onto the execution of the other vcores described in *cip.
3393  */
3394 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3395 			  int target_threads)
3396 {
3397 	if (cip->total_threads + pvc->num_threads > target_threads)
3398 		return false;
3399 
3400 	return can_dynamic_split(pvc, cip);
3401 }
3402 
3403 static void prepare_threads(struct kvmppc_vcore *vc)
3404 {
3405 	int i;
3406 	struct kvm_vcpu *vcpu;
3407 
3408 	for_each_runnable_thread(i, vcpu, vc) {
3409 		if (signal_pending(vcpu->arch.run_task))
3410 			vcpu->arch.ret = -EINTR;
3411 		else if (vcpu->arch.vpa.update_pending ||
3412 			 vcpu->arch.slb_shadow.update_pending ||
3413 			 vcpu->arch.dtl.update_pending)
3414 			vcpu->arch.ret = RESUME_GUEST;
3415 		else
3416 			continue;
3417 		kvmppc_remove_runnable(vc, vcpu, mftb());
3418 		wake_up(&vcpu->arch.cpu_run);
3419 	}
3420 }
3421 
3422 static void collect_piggybacks(struct core_info *cip, int target_threads)
3423 {
3424 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3425 	struct kvmppc_vcore *pvc, *vcnext;
3426 
3427 	spin_lock(&lp->lock);
3428 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3429 		if (!spin_trylock(&pvc->lock))
3430 			continue;
3431 		prepare_threads(pvc);
3432 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3433 			list_del_init(&pvc->preempt_list);
3434 			if (pvc->runner == NULL) {
3435 				pvc->vcore_state = VCORE_INACTIVE;
3436 				kvmppc_core_end_stolen(pvc, mftb());
3437 			}
3438 			spin_unlock(&pvc->lock);
3439 			continue;
3440 		}
3441 		if (!can_piggyback(pvc, cip, target_threads)) {
3442 			spin_unlock(&pvc->lock);
3443 			continue;
3444 		}
3445 		kvmppc_core_end_stolen(pvc, mftb());
3446 		pvc->vcore_state = VCORE_PIGGYBACK;
3447 		if (cip->total_threads >= target_threads)
3448 			break;
3449 	}
3450 	spin_unlock(&lp->lock);
3451 }
3452 
3453 static bool recheck_signals_and_mmu(struct core_info *cip)
3454 {
3455 	int sub, i;
3456 	struct kvm_vcpu *vcpu;
3457 	struct kvmppc_vcore *vc;
3458 
3459 	for (sub = 0; sub < cip->n_subcores; ++sub) {
3460 		vc = cip->vc[sub];
3461 		if (!vc->kvm->arch.mmu_ready)
3462 			return true;
3463 		for_each_runnable_thread(i, vcpu, vc)
3464 			if (signal_pending(vcpu->arch.run_task))
3465 				return true;
3466 	}
3467 	return false;
3468 }
3469 
3470 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3471 {
3472 	int still_running = 0, i;
3473 	u64 now;
3474 	long ret;
3475 	struct kvm_vcpu *vcpu;
3476 
3477 	spin_lock(&vc->lock);
3478 	now = get_tb();
3479 	for_each_runnable_thread(i, vcpu, vc) {
3480 		/*
3481 		 * It's safe to unlock the vcore in the loop here, because
3482 		 * for_each_runnable_thread() is safe against removal of
3483 		 * the vcpu, and the vcore state is VCORE_EXITING here,
3484 		 * so any vcpus becoming runnable will have their arch.trap
3485 		 * set to zero and can't actually run in the guest.
3486 		 */
3487 		spin_unlock(&vc->lock);
3488 		/* cancel pending dec exception if dec is positive */
3489 		if (now < kvmppc_dec_expires_host_tb(vcpu) &&
3490 		    kvmppc_core_pending_dec(vcpu))
3491 			kvmppc_core_dequeue_dec(vcpu);
3492 
3493 		trace_kvm_guest_exit(vcpu);
3494 
3495 		ret = RESUME_GUEST;
3496 		if (vcpu->arch.trap)
3497 			ret = kvmppc_handle_exit_hv(vcpu,
3498 						    vcpu->arch.run_task);
3499 
3500 		vcpu->arch.ret = ret;
3501 		vcpu->arch.trap = 0;
3502 
3503 		spin_lock(&vc->lock);
3504 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3505 			if (vcpu->arch.pending_exceptions)
3506 				kvmppc_core_prepare_to_enter(vcpu);
3507 			if (vcpu->arch.ceded)
3508 				kvmppc_set_timer(vcpu);
3509 			else
3510 				++still_running;
3511 		} else {
3512 			kvmppc_remove_runnable(vc, vcpu, mftb());
3513 			wake_up(&vcpu->arch.cpu_run);
3514 		}
3515 	}
3516 	if (!is_master) {
3517 		if (still_running > 0) {
3518 			kvmppc_vcore_preempt(vc);
3519 		} else if (vc->runner) {
3520 			vc->vcore_state = VCORE_PREEMPT;
3521 			kvmppc_core_start_stolen(vc, mftb());
3522 		} else {
3523 			vc->vcore_state = VCORE_INACTIVE;
3524 		}
3525 		if (vc->n_runnable > 0 && vc->runner == NULL) {
3526 			/* make sure there's a candidate runner awake */
3527 			i = -1;
3528 			vcpu = next_runnable_thread(vc, &i);
3529 			wake_up(&vcpu->arch.cpu_run);
3530 		}
3531 	}
3532 	spin_unlock(&vc->lock);
3533 }
3534 
3535 /*
3536  * Clear core from the list of active host cores as we are about to
3537  * enter the guest. Only do this if it is the primary thread of the
3538  * core (not if a subcore) that is entering the guest.
3539  */
3540 static inline int kvmppc_clear_host_core(unsigned int cpu)
3541 {
3542 	int core;
3543 
3544 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3545 		return 0;
3546 	/*
3547 	 * Memory barrier can be omitted here as we will do a smp_wmb()
3548 	 * later in kvmppc_start_thread and we need ensure that state is
3549 	 * visible to other CPUs only after we enter guest.
3550 	 */
3551 	core = cpu >> threads_shift;
3552 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3553 	return 0;
3554 }
3555 
3556 /*
3557  * Advertise this core as an active host core since we exited the guest
3558  * Only need to do this if it is the primary thread of the core that is
3559  * exiting.
3560  */
3561 static inline int kvmppc_set_host_core(unsigned int cpu)
3562 {
3563 	int core;
3564 
3565 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3566 		return 0;
3567 
3568 	/*
3569 	 * Memory barrier can be omitted here because we do a spin_unlock
3570 	 * immediately after this which provides the memory barrier.
3571 	 */
3572 	core = cpu >> threads_shift;
3573 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3574 	return 0;
3575 }
3576 
3577 static void set_irq_happened(int trap)
3578 {
3579 	switch (trap) {
3580 	case BOOK3S_INTERRUPT_EXTERNAL:
3581 		local_paca->irq_happened |= PACA_IRQ_EE;
3582 		break;
3583 	case BOOK3S_INTERRUPT_H_DOORBELL:
3584 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3585 		break;
3586 	case BOOK3S_INTERRUPT_HMI:
3587 		local_paca->irq_happened |= PACA_IRQ_HMI;
3588 		break;
3589 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3590 		replay_system_reset();
3591 		break;
3592 	}
3593 }
3594 
3595 /*
3596  * Run a set of guest threads on a physical core.
3597  * Called with vc->lock held.
3598  */
3599 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3600 {
3601 	struct kvm_vcpu *vcpu;
3602 	int i;
3603 	int srcu_idx;
3604 	struct core_info core_info;
3605 	struct kvmppc_vcore *pvc;
3606 	struct kvm_split_mode split_info, *sip;
3607 	int split, subcore_size, active;
3608 	int sub;
3609 	bool thr0_done;
3610 	unsigned long cmd_bit, stat_bit;
3611 	int pcpu, thr;
3612 	int target_threads;
3613 	int controlled_threads;
3614 	int trap;
3615 	bool is_power8;
3616 
3617 	if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3618 		return;
3619 
3620 	/*
3621 	 * Remove from the list any threads that have a signal pending
3622 	 * or need a VPA update done
3623 	 */
3624 	prepare_threads(vc);
3625 
3626 	/* if the runner is no longer runnable, let the caller pick a new one */
3627 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3628 		return;
3629 
3630 	/*
3631 	 * Initialize *vc.
3632 	 */
3633 	init_vcore_to_run(vc);
3634 	vc->preempt_tb = TB_NIL;
3635 
3636 	/*
3637 	 * Number of threads that we will be controlling: the same as
3638 	 * the number of threads per subcore, except on POWER9,
3639 	 * where it's 1 because the threads are (mostly) independent.
3640 	 */
3641 	controlled_threads = threads_per_vcore(vc->kvm);
3642 
3643 	/*
3644 	 * Make sure we are running on primary threads, and that secondary
3645 	 * threads are offline.  Also check if the number of threads in this
3646 	 * guest are greater than the current system threads per guest.
3647 	 */
3648 	if ((controlled_threads > 1) &&
3649 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3650 		for_each_runnable_thread(i, vcpu, vc) {
3651 			vcpu->arch.ret = -EBUSY;
3652 			kvmppc_remove_runnable(vc, vcpu, mftb());
3653 			wake_up(&vcpu->arch.cpu_run);
3654 		}
3655 		goto out;
3656 	}
3657 
3658 	/*
3659 	 * See if we could run any other vcores on the physical core
3660 	 * along with this one.
3661 	 */
3662 	init_core_info(&core_info, vc);
3663 	pcpu = smp_processor_id();
3664 	target_threads = controlled_threads;
3665 	if (target_smt_mode && target_smt_mode < target_threads)
3666 		target_threads = target_smt_mode;
3667 	if (vc->num_threads < target_threads)
3668 		collect_piggybacks(&core_info, target_threads);
3669 
3670 	/*
3671 	 * Hard-disable interrupts, and check resched flag and signals.
3672 	 * If we need to reschedule or deliver a signal, clean up
3673 	 * and return without going into the guest(s).
3674 	 * If the mmu_ready flag has been cleared, don't go into the
3675 	 * guest because that means a HPT resize operation is in progress.
3676 	 */
3677 	local_irq_disable();
3678 	hard_irq_disable();
3679 	if (lazy_irq_pending() || need_resched() ||
3680 	    recheck_signals_and_mmu(&core_info)) {
3681 		local_irq_enable();
3682 		vc->vcore_state = VCORE_INACTIVE;
3683 		/* Unlock all except the primary vcore */
3684 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3685 			pvc = core_info.vc[sub];
3686 			/* Put back on to the preempted vcores list */
3687 			kvmppc_vcore_preempt(pvc);
3688 			spin_unlock(&pvc->lock);
3689 		}
3690 		for (i = 0; i < controlled_threads; ++i)
3691 			kvmppc_release_hwthread(pcpu + i);
3692 		return;
3693 	}
3694 
3695 	kvmppc_clear_host_core(pcpu);
3696 
3697 	/* Decide on micro-threading (split-core) mode */
3698 	subcore_size = threads_per_subcore;
3699 	cmd_bit = stat_bit = 0;
3700 	split = core_info.n_subcores;
3701 	sip = NULL;
3702 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3703 
3704 	if (split > 1) {
3705 		sip = &split_info;
3706 		memset(&split_info, 0, sizeof(split_info));
3707 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3708 			split_info.vc[sub] = core_info.vc[sub];
3709 
3710 		if (is_power8) {
3711 			if (split == 2 && (dynamic_mt_modes & 2)) {
3712 				cmd_bit = HID0_POWER8_1TO2LPAR;
3713 				stat_bit = HID0_POWER8_2LPARMODE;
3714 			} else {
3715 				split = 4;
3716 				cmd_bit = HID0_POWER8_1TO4LPAR;
3717 				stat_bit = HID0_POWER8_4LPARMODE;
3718 			}
3719 			subcore_size = MAX_SMT_THREADS / split;
3720 			split_info.rpr = mfspr(SPRN_RPR);
3721 			split_info.pmmar = mfspr(SPRN_PMMAR);
3722 			split_info.ldbar = mfspr(SPRN_LDBAR);
3723 			split_info.subcore_size = subcore_size;
3724 		} else {
3725 			split_info.subcore_size = 1;
3726 		}
3727 
3728 		/* order writes to split_info before kvm_split_mode pointer */
3729 		smp_wmb();
3730 	}
3731 
3732 	for (thr = 0; thr < controlled_threads; ++thr) {
3733 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3734 
3735 		paca->kvm_hstate.napping = 0;
3736 		paca->kvm_hstate.kvm_split_mode = sip;
3737 	}
3738 
3739 	/* Initiate micro-threading (split-core) on POWER8 if required */
3740 	if (cmd_bit) {
3741 		unsigned long hid0 = mfspr(SPRN_HID0);
3742 
3743 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3744 		mb();
3745 		mtspr(SPRN_HID0, hid0);
3746 		isync();
3747 		for (;;) {
3748 			hid0 = mfspr(SPRN_HID0);
3749 			if (hid0 & stat_bit)
3750 				break;
3751 			cpu_relax();
3752 		}
3753 	}
3754 
3755 	/*
3756 	 * On POWER8, set RWMR register.
3757 	 * Since it only affects PURR and SPURR, it doesn't affect
3758 	 * the host, so we don't save/restore the host value.
3759 	 */
3760 	if (is_power8) {
3761 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3762 		int n_online = atomic_read(&vc->online_count);
3763 
3764 		/*
3765 		 * Use the 8-thread value if we're doing split-core
3766 		 * or if the vcore's online count looks bogus.
3767 		 */
3768 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3769 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3770 			rwmr_val = p8_rwmr_values[n_online];
3771 		mtspr(SPRN_RWMR, rwmr_val);
3772 	}
3773 
3774 	/* Start all the threads */
3775 	active = 0;
3776 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3777 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3778 		thr0_done = false;
3779 		active |= 1 << thr;
3780 		pvc = core_info.vc[sub];
3781 		pvc->pcpu = pcpu + thr;
3782 		for_each_runnable_thread(i, vcpu, pvc) {
3783 			/*
3784 			 * XXX: is kvmppc_start_thread called too late here?
3785 			 * It updates vcpu->cpu and vcpu->arch.thread_cpu
3786 			 * which are used by kvmppc_fast_vcpu_kick_hv(), but
3787 			 * kick is called after new exceptions become available
3788 			 * and exceptions are checked earlier than here, by
3789 			 * kvmppc_core_prepare_to_enter.
3790 			 */
3791 			kvmppc_start_thread(vcpu, pvc);
3792 			kvmppc_create_dtl_entry(vcpu, pvc);
3793 			trace_kvm_guest_enter(vcpu);
3794 			if (!vcpu->arch.ptid)
3795 				thr0_done = true;
3796 			active |= 1 << (thr + vcpu->arch.ptid);
3797 		}
3798 		/*
3799 		 * We need to start the first thread of each subcore
3800 		 * even if it doesn't have a vcpu.
3801 		 */
3802 		if (!thr0_done)
3803 			kvmppc_start_thread(NULL, pvc);
3804 	}
3805 
3806 	/*
3807 	 * Ensure that split_info.do_nap is set after setting
3808 	 * the vcore pointer in the PACA of the secondaries.
3809 	 */
3810 	smp_mb();
3811 
3812 	/*
3813 	 * When doing micro-threading, poke the inactive threads as well.
3814 	 * This gets them to the nap instruction after kvm_do_nap,
3815 	 * which reduces the time taken to unsplit later.
3816 	 */
3817 	if (cmd_bit) {
3818 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3819 		for (thr = 1; thr < threads_per_subcore; ++thr)
3820 			if (!(active & (1 << thr)))
3821 				kvmppc_ipi_thread(pcpu + thr);
3822 	}
3823 
3824 	vc->vcore_state = VCORE_RUNNING;
3825 	preempt_disable();
3826 
3827 	trace_kvmppc_run_core(vc, 0);
3828 
3829 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3830 		spin_unlock(&core_info.vc[sub]->lock);
3831 
3832 	guest_enter_irqoff();
3833 
3834 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3835 
3836 	this_cpu_disable_ftrace();
3837 
3838 	/*
3839 	 * Interrupts will be enabled once we get into the guest,
3840 	 * so tell lockdep that we're about to enable interrupts.
3841 	 */
3842 	trace_hardirqs_on();
3843 
3844 	trap = __kvmppc_vcore_entry();
3845 
3846 	trace_hardirqs_off();
3847 
3848 	this_cpu_enable_ftrace();
3849 
3850 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3851 
3852 	set_irq_happened(trap);
3853 
3854 	spin_lock(&vc->lock);
3855 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3856 	vc->vcore_state = VCORE_EXITING;
3857 
3858 	/* wait for secondary threads to finish writing their state to memory */
3859 	kvmppc_wait_for_nap(controlled_threads);
3860 
3861 	/* Return to whole-core mode if we split the core earlier */
3862 	if (cmd_bit) {
3863 		unsigned long hid0 = mfspr(SPRN_HID0);
3864 		unsigned long loops = 0;
3865 
3866 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3867 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3868 		mb();
3869 		mtspr(SPRN_HID0, hid0);
3870 		isync();
3871 		for (;;) {
3872 			hid0 = mfspr(SPRN_HID0);
3873 			if (!(hid0 & stat_bit))
3874 				break;
3875 			cpu_relax();
3876 			++loops;
3877 		}
3878 		split_info.do_nap = 0;
3879 	}
3880 
3881 	kvmppc_set_host_core(pcpu);
3882 
3883 	context_tracking_guest_exit();
3884 	if (!vtime_accounting_enabled_this_cpu()) {
3885 		local_irq_enable();
3886 		/*
3887 		 * Service IRQs here before vtime_account_guest_exit() so any
3888 		 * ticks that occurred while running the guest are accounted to
3889 		 * the guest. If vtime accounting is enabled, accounting uses
3890 		 * TB rather than ticks, so it can be done without enabling
3891 		 * interrupts here, which has the problem that it accounts
3892 		 * interrupt processing overhead to the host.
3893 		 */
3894 		local_irq_disable();
3895 	}
3896 	vtime_account_guest_exit();
3897 
3898 	local_irq_enable();
3899 
3900 	/* Let secondaries go back to the offline loop */
3901 	for (i = 0; i < controlled_threads; ++i) {
3902 		kvmppc_release_hwthread(pcpu + i);
3903 		if (sip && sip->napped[i])
3904 			kvmppc_ipi_thread(pcpu + i);
3905 	}
3906 
3907 	spin_unlock(&vc->lock);
3908 
3909 	/* make sure updates to secondary vcpu structs are visible now */
3910 	smp_mb();
3911 
3912 	preempt_enable();
3913 
3914 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3915 		pvc = core_info.vc[sub];
3916 		post_guest_process(pvc, pvc == vc);
3917 	}
3918 
3919 	spin_lock(&vc->lock);
3920 
3921  out:
3922 	vc->vcore_state = VCORE_INACTIVE;
3923 	trace_kvmppc_run_core(vc, 1);
3924 }
3925 
3926 static inline bool hcall_is_xics(unsigned long req)
3927 {
3928 	return req == H_EOI || req == H_CPPR || req == H_IPI ||
3929 		req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
3930 }
3931 
3932 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu)
3933 {
3934 	struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3935 	if (lp) {
3936 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3937 		lp->yield_count = cpu_to_be32(yield_count);
3938 		vcpu->arch.vpa.dirty = 1;
3939 	}
3940 }
3941 
3942 /* call our hypervisor to load up HV regs and go */
3943 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
3944 {
3945 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3946 	unsigned long host_psscr;
3947 	unsigned long msr;
3948 	struct hv_guest_state hvregs;
3949 	struct p9_host_os_sprs host_os_sprs;
3950 	s64 dec;
3951 	int trap;
3952 
3953 	msr = mfmsr();
3954 
3955 	save_p9_host_os_sprs(&host_os_sprs);
3956 
3957 	/*
3958 	 * We need to save and restore the guest visible part of the
3959 	 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3960 	 * doesn't do this for us. Note only required if pseries since
3961 	 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
3962 	 */
3963 	host_psscr = mfspr(SPRN_PSSCR_PR);
3964 
3965 	kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
3966 	if (lazy_irq_pending())
3967 		return 0;
3968 
3969 	if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
3970 		msr = mfmsr(); /* TM restore can update msr */
3971 
3972 	if (vcpu->arch.psscr != host_psscr)
3973 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3974 
3975 	kvmhv_save_hv_regs(vcpu, &hvregs);
3976 	hvregs.lpcr = lpcr;
3977 	hvregs.amor = ~0;
3978 	vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3979 	hvregs.version = HV_GUEST_STATE_VERSION;
3980 	if (vcpu->arch.nested) {
3981 		hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3982 		hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3983 	} else {
3984 		hvregs.lpid = vcpu->kvm->arch.lpid;
3985 		hvregs.vcpu_token = vcpu->vcpu_id;
3986 	}
3987 	hvregs.hdec_expiry = time_limit;
3988 
3989 	/*
3990 	 * When setting DEC, we must always deal with irq_work_raise
3991 	 * via NMI vs setting DEC. The problem occurs right as we
3992 	 * switch into guest mode if a NMI hits and sets pending work
3993 	 * and sets DEC, then that will apply to the guest and not
3994 	 * bring us back to the host.
3995 	 *
3996 	 * irq_work_raise could check a flag (or possibly LPCR[HDICE]
3997 	 * for example) and set HDEC to 1? That wouldn't solve the
3998 	 * nested hv case which needs to abort the hcall or zero the
3999 	 * time limit.
4000 	 *
4001 	 * XXX: Another day's problem.
4002 	 */
4003 	mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb);
4004 
4005 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
4006 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
4007 	switch_pmu_to_guest(vcpu, &host_os_sprs);
4008 	trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
4009 				  __pa(&vcpu->arch.regs));
4010 	kvmhv_restore_hv_return_state(vcpu, &hvregs);
4011 	switch_pmu_to_host(vcpu, &host_os_sprs);
4012 	vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
4013 	vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
4014 	vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
4015 	vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
4016 
4017 	store_vcpu_state(vcpu);
4018 
4019 	dec = mfspr(SPRN_DEC);
4020 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4021 		dec = (s32) dec;
4022 	*tb = mftb();
4023 	vcpu->arch.dec_expires = dec + (*tb + vc->tb_offset);
4024 
4025 	timer_rearm_host_dec(*tb);
4026 
4027 	restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4028 	if (vcpu->arch.psscr != host_psscr)
4029 		mtspr(SPRN_PSSCR_PR, host_psscr);
4030 
4031 	return trap;
4032 }
4033 
4034 /*
4035  * Guest entry for POWER9 and later CPUs.
4036  */
4037 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
4038 			 unsigned long lpcr, u64 *tb)
4039 {
4040 	struct kvm *kvm = vcpu->kvm;
4041 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4042 	u64 next_timer;
4043 	int trap;
4044 
4045 	next_timer = timer_get_next_tb();
4046 	if (*tb >= next_timer)
4047 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
4048 	if (next_timer < time_limit)
4049 		time_limit = next_timer;
4050 	else if (*tb >= time_limit) /* nested time limit */
4051 		return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER;
4052 
4053 	vcpu->arch.ceded = 0;
4054 
4055 	vcpu_vpa_increment_dispatch(vcpu);
4056 
4057 	if (kvmhv_on_pseries()) {
4058 		trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb);
4059 
4060 		/* H_CEDE has to be handled now, not later */
4061 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !nested &&
4062 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
4063 			kvmppc_cede(vcpu);
4064 			kvmppc_set_gpr(vcpu, 3, 0);
4065 			trap = 0;
4066 		}
4067 
4068 	} else if (nested) {
4069 		__this_cpu_write(cpu_in_guest, kvm);
4070 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4071 		__this_cpu_write(cpu_in_guest, NULL);
4072 
4073 	} else {
4074 		kvmppc_xive_push_vcpu(vcpu);
4075 
4076 		__this_cpu_write(cpu_in_guest, kvm);
4077 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4078 		__this_cpu_write(cpu_in_guest, NULL);
4079 
4080 		if (trap == BOOK3S_INTERRUPT_SYSCALL &&
4081 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4082 			unsigned long req = kvmppc_get_gpr(vcpu, 3);
4083 
4084 			/*
4085 			 * XIVE rearm and XICS hcalls must be handled
4086 			 * before xive context is pulled (is this
4087 			 * true?)
4088 			 */
4089 			if (req == H_CEDE) {
4090 				/* H_CEDE has to be handled now */
4091 				kvmppc_cede(vcpu);
4092 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4093 					/*
4094 					 * Pending escalation so abort
4095 					 * the cede.
4096 					 */
4097 					vcpu->arch.ceded = 0;
4098 				}
4099 				kvmppc_set_gpr(vcpu, 3, 0);
4100 				trap = 0;
4101 
4102 			} else if (req == H_ENTER_NESTED) {
4103 				/*
4104 				 * L2 should not run with the L1
4105 				 * context so rearm and pull it.
4106 				 */
4107 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4108 					/*
4109 					 * Pending escalation so abort
4110 					 * H_ENTER_NESTED.
4111 					 */
4112 					kvmppc_set_gpr(vcpu, 3, 0);
4113 					trap = 0;
4114 				}
4115 
4116 			} else if (hcall_is_xics(req)) {
4117 				int ret;
4118 
4119 				ret = kvmppc_xive_xics_hcall(vcpu, req);
4120 				if (ret != H_TOO_HARD) {
4121 					kvmppc_set_gpr(vcpu, 3, ret);
4122 					trap = 0;
4123 				}
4124 			}
4125 		}
4126 		kvmppc_xive_pull_vcpu(vcpu);
4127 
4128 		if (kvm_is_radix(kvm))
4129 			vcpu->arch.slb_max = 0;
4130 	}
4131 
4132 	vcpu_vpa_increment_dispatch(vcpu);
4133 
4134 	return trap;
4135 }
4136 
4137 /*
4138  * Wait for some other vcpu thread to execute us, and
4139  * wake us up when we need to handle something in the host.
4140  */
4141 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4142 				 struct kvm_vcpu *vcpu, int wait_state)
4143 {
4144 	DEFINE_WAIT(wait);
4145 
4146 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4147 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4148 		spin_unlock(&vc->lock);
4149 		schedule();
4150 		spin_lock(&vc->lock);
4151 	}
4152 	finish_wait(&vcpu->arch.cpu_run, &wait);
4153 }
4154 
4155 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4156 {
4157 	if (!halt_poll_ns_grow)
4158 		return;
4159 
4160 	vc->halt_poll_ns *= halt_poll_ns_grow;
4161 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4162 		vc->halt_poll_ns = halt_poll_ns_grow_start;
4163 }
4164 
4165 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4166 {
4167 	if (halt_poll_ns_shrink == 0)
4168 		vc->halt_poll_ns = 0;
4169 	else
4170 		vc->halt_poll_ns /= halt_poll_ns_shrink;
4171 }
4172 
4173 #ifdef CONFIG_KVM_XICS
4174 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4175 {
4176 	if (!xics_on_xive())
4177 		return false;
4178 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4179 		vcpu->arch.xive_saved_state.cppr;
4180 }
4181 #else
4182 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4183 {
4184 	return false;
4185 }
4186 #endif /* CONFIG_KVM_XICS */
4187 
4188 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4189 {
4190 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4191 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4192 		return true;
4193 
4194 	return false;
4195 }
4196 
4197 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu)
4198 {
4199 	if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4200 		return true;
4201 	return false;
4202 }
4203 
4204 /*
4205  * Check to see if any of the runnable vcpus on the vcore have pending
4206  * exceptions or are no longer ceded
4207  */
4208 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4209 {
4210 	struct kvm_vcpu *vcpu;
4211 	int i;
4212 
4213 	for_each_runnable_thread(i, vcpu, vc) {
4214 		if (kvmppc_vcpu_check_block(vcpu))
4215 			return 1;
4216 	}
4217 
4218 	return 0;
4219 }
4220 
4221 /*
4222  * All the vcpus in this vcore are idle, so wait for a decrementer
4223  * or external interrupt to one of the vcpus.  vc->lock is held.
4224  */
4225 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4226 {
4227 	ktime_t cur, start_poll, start_wait;
4228 	int do_sleep = 1;
4229 	u64 block_ns;
4230 
4231 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
4232 
4233 	/* Poll for pending exceptions and ceded state */
4234 	cur = start_poll = ktime_get();
4235 	if (vc->halt_poll_ns) {
4236 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4237 		++vc->runner->stat.generic.halt_attempted_poll;
4238 
4239 		vc->vcore_state = VCORE_POLLING;
4240 		spin_unlock(&vc->lock);
4241 
4242 		do {
4243 			if (kvmppc_vcore_check_block(vc)) {
4244 				do_sleep = 0;
4245 				break;
4246 			}
4247 			cur = ktime_get();
4248 		} while (kvm_vcpu_can_poll(cur, stop));
4249 
4250 		spin_lock(&vc->lock);
4251 		vc->vcore_state = VCORE_INACTIVE;
4252 
4253 		if (!do_sleep) {
4254 			++vc->runner->stat.generic.halt_successful_poll;
4255 			goto out;
4256 		}
4257 	}
4258 
4259 	prepare_to_rcuwait(&vc->wait);
4260 	set_current_state(TASK_INTERRUPTIBLE);
4261 	if (kvmppc_vcore_check_block(vc)) {
4262 		finish_rcuwait(&vc->wait);
4263 		do_sleep = 0;
4264 		/* If we polled, count this as a successful poll */
4265 		if (vc->halt_poll_ns)
4266 			++vc->runner->stat.generic.halt_successful_poll;
4267 		goto out;
4268 	}
4269 
4270 	start_wait = ktime_get();
4271 
4272 	vc->vcore_state = VCORE_SLEEPING;
4273 	trace_kvmppc_vcore_blocked(vc->runner, 0);
4274 	spin_unlock(&vc->lock);
4275 	schedule();
4276 	finish_rcuwait(&vc->wait);
4277 	spin_lock(&vc->lock);
4278 	vc->vcore_state = VCORE_INACTIVE;
4279 	trace_kvmppc_vcore_blocked(vc->runner, 1);
4280 	++vc->runner->stat.halt_successful_wait;
4281 
4282 	cur = ktime_get();
4283 
4284 out:
4285 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4286 
4287 	/* Attribute wait time */
4288 	if (do_sleep) {
4289 		vc->runner->stat.generic.halt_wait_ns +=
4290 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
4291 		KVM_STATS_LOG_HIST_UPDATE(
4292 				vc->runner->stat.generic.halt_wait_hist,
4293 				ktime_to_ns(cur) - ktime_to_ns(start_wait));
4294 		/* Attribute failed poll time */
4295 		if (vc->halt_poll_ns) {
4296 			vc->runner->stat.generic.halt_poll_fail_ns +=
4297 				ktime_to_ns(start_wait) -
4298 				ktime_to_ns(start_poll);
4299 			KVM_STATS_LOG_HIST_UPDATE(
4300 				vc->runner->stat.generic.halt_poll_fail_hist,
4301 				ktime_to_ns(start_wait) -
4302 				ktime_to_ns(start_poll));
4303 		}
4304 	} else {
4305 		/* Attribute successful poll time */
4306 		if (vc->halt_poll_ns) {
4307 			vc->runner->stat.generic.halt_poll_success_ns +=
4308 				ktime_to_ns(cur) -
4309 				ktime_to_ns(start_poll);
4310 			KVM_STATS_LOG_HIST_UPDATE(
4311 				vc->runner->stat.generic.halt_poll_success_hist,
4312 				ktime_to_ns(cur) - ktime_to_ns(start_poll));
4313 		}
4314 	}
4315 
4316 	/* Adjust poll time */
4317 	if (halt_poll_ns) {
4318 		if (block_ns <= vc->halt_poll_ns)
4319 			;
4320 		/* We slept and blocked for longer than the max halt time */
4321 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4322 			shrink_halt_poll_ns(vc);
4323 		/* We slept and our poll time is too small */
4324 		else if (vc->halt_poll_ns < halt_poll_ns &&
4325 				block_ns < halt_poll_ns)
4326 			grow_halt_poll_ns(vc);
4327 		if (vc->halt_poll_ns > halt_poll_ns)
4328 			vc->halt_poll_ns = halt_poll_ns;
4329 	} else
4330 		vc->halt_poll_ns = 0;
4331 
4332 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4333 }
4334 
4335 /*
4336  * This never fails for a radix guest, as none of the operations it does
4337  * for a radix guest can fail or have a way to report failure.
4338  */
4339 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4340 {
4341 	int r = 0;
4342 	struct kvm *kvm = vcpu->kvm;
4343 
4344 	mutex_lock(&kvm->arch.mmu_setup_lock);
4345 	if (!kvm->arch.mmu_ready) {
4346 		if (!kvm_is_radix(kvm))
4347 			r = kvmppc_hv_setup_htab_rma(vcpu);
4348 		if (!r) {
4349 			if (cpu_has_feature(CPU_FTR_ARCH_300))
4350 				kvmppc_setup_partition_table(kvm);
4351 			kvm->arch.mmu_ready = 1;
4352 		}
4353 	}
4354 	mutex_unlock(&kvm->arch.mmu_setup_lock);
4355 	return r;
4356 }
4357 
4358 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4359 {
4360 	struct kvm_run *run = vcpu->run;
4361 	int n_ceded, i, r;
4362 	struct kvmppc_vcore *vc;
4363 	struct kvm_vcpu *v;
4364 
4365 	trace_kvmppc_run_vcpu_enter(vcpu);
4366 
4367 	run->exit_reason = 0;
4368 	vcpu->arch.ret = RESUME_GUEST;
4369 	vcpu->arch.trap = 0;
4370 	kvmppc_update_vpas(vcpu);
4371 
4372 	/*
4373 	 * Synchronize with other threads in this virtual core
4374 	 */
4375 	vc = vcpu->arch.vcore;
4376 	spin_lock(&vc->lock);
4377 	vcpu->arch.ceded = 0;
4378 	vcpu->arch.run_task = current;
4379 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4380 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4381 	vcpu->arch.busy_preempt = TB_NIL;
4382 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4383 	++vc->n_runnable;
4384 
4385 	/*
4386 	 * This happens the first time this is called for a vcpu.
4387 	 * If the vcore is already running, we may be able to start
4388 	 * this thread straight away and have it join in.
4389 	 */
4390 	if (!signal_pending(current)) {
4391 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
4392 		     vc->vcore_state == VCORE_RUNNING) &&
4393 			   !VCORE_IS_EXITING(vc)) {
4394 			kvmppc_create_dtl_entry(vcpu, vc);
4395 			kvmppc_start_thread(vcpu, vc);
4396 			trace_kvm_guest_enter(vcpu);
4397 		} else if (vc->vcore_state == VCORE_SLEEPING) {
4398 		        rcuwait_wake_up(&vc->wait);
4399 		}
4400 
4401 	}
4402 
4403 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4404 	       !signal_pending(current)) {
4405 		/* See if the MMU is ready to go */
4406 		if (!vcpu->kvm->arch.mmu_ready) {
4407 			spin_unlock(&vc->lock);
4408 			r = kvmhv_setup_mmu(vcpu);
4409 			spin_lock(&vc->lock);
4410 			if (r) {
4411 				run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4412 				run->fail_entry.
4413 					hardware_entry_failure_reason = 0;
4414 				vcpu->arch.ret = r;
4415 				break;
4416 			}
4417 		}
4418 
4419 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4420 			kvmppc_vcore_end_preempt(vc);
4421 
4422 		if (vc->vcore_state != VCORE_INACTIVE) {
4423 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4424 			continue;
4425 		}
4426 		for_each_runnable_thread(i, v, vc) {
4427 			kvmppc_core_prepare_to_enter(v);
4428 			if (signal_pending(v->arch.run_task)) {
4429 				kvmppc_remove_runnable(vc, v, mftb());
4430 				v->stat.signal_exits++;
4431 				v->run->exit_reason = KVM_EXIT_INTR;
4432 				v->arch.ret = -EINTR;
4433 				wake_up(&v->arch.cpu_run);
4434 			}
4435 		}
4436 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4437 			break;
4438 		n_ceded = 0;
4439 		for_each_runnable_thread(i, v, vc) {
4440 			if (!kvmppc_vcpu_woken(v))
4441 				n_ceded += v->arch.ceded;
4442 			else
4443 				v->arch.ceded = 0;
4444 		}
4445 		vc->runner = vcpu;
4446 		if (n_ceded == vc->n_runnable) {
4447 			kvmppc_vcore_blocked(vc);
4448 		} else if (need_resched()) {
4449 			kvmppc_vcore_preempt(vc);
4450 			/* Let something else run */
4451 			cond_resched_lock(&vc->lock);
4452 			if (vc->vcore_state == VCORE_PREEMPT)
4453 				kvmppc_vcore_end_preempt(vc);
4454 		} else {
4455 			kvmppc_run_core(vc);
4456 		}
4457 		vc->runner = NULL;
4458 	}
4459 
4460 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4461 	       (vc->vcore_state == VCORE_RUNNING ||
4462 		vc->vcore_state == VCORE_EXITING ||
4463 		vc->vcore_state == VCORE_PIGGYBACK))
4464 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4465 
4466 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4467 		kvmppc_vcore_end_preempt(vc);
4468 
4469 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4470 		kvmppc_remove_runnable(vc, vcpu, mftb());
4471 		vcpu->stat.signal_exits++;
4472 		run->exit_reason = KVM_EXIT_INTR;
4473 		vcpu->arch.ret = -EINTR;
4474 	}
4475 
4476 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4477 		/* Wake up some vcpu to run the core */
4478 		i = -1;
4479 		v = next_runnable_thread(vc, &i);
4480 		wake_up(&v->arch.cpu_run);
4481 	}
4482 
4483 	trace_kvmppc_run_vcpu_exit(vcpu);
4484 	spin_unlock(&vc->lock);
4485 	return vcpu->arch.ret;
4486 }
4487 
4488 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4489 			  unsigned long lpcr)
4490 {
4491 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
4492 	struct kvm_run *run = vcpu->run;
4493 	int trap, r, pcpu;
4494 	int srcu_idx;
4495 	struct kvmppc_vcore *vc;
4496 	struct kvm *kvm = vcpu->kvm;
4497 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4498 	unsigned long flags;
4499 	u64 tb;
4500 
4501 	trace_kvmppc_run_vcpu_enter(vcpu);
4502 
4503 	run->exit_reason = 0;
4504 	vcpu->arch.ret = RESUME_GUEST;
4505 	vcpu->arch.trap = 0;
4506 
4507 	vc = vcpu->arch.vcore;
4508 	vcpu->arch.ceded = 0;
4509 	vcpu->arch.run_task = current;
4510 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4511 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4512 
4513 	/* See if the MMU is ready to go */
4514 	if (unlikely(!kvm->arch.mmu_ready)) {
4515 		r = kvmhv_setup_mmu(vcpu);
4516 		if (r) {
4517 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4518 			run->fail_entry.hardware_entry_failure_reason = 0;
4519 			vcpu->arch.ret = r;
4520 			return r;
4521 		}
4522 	}
4523 
4524 	if (need_resched())
4525 		cond_resched();
4526 
4527 	kvmppc_update_vpas(vcpu);
4528 
4529 	preempt_disable();
4530 	pcpu = smp_processor_id();
4531 	if (kvm_is_radix(kvm))
4532 		kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4533 
4534 	/* flags save not required, but irq_pmu has no disable/enable API */
4535 	powerpc_local_irq_pmu_save(flags);
4536 
4537 	if (signal_pending(current))
4538 		goto sigpend;
4539 	if (need_resched() || !kvm->arch.mmu_ready)
4540 		goto out;
4541 
4542 	vcpu->cpu = pcpu;
4543 	vcpu->arch.thread_cpu = pcpu;
4544 	vc->pcpu = pcpu;
4545 	local_paca->kvm_hstate.kvm_vcpu = vcpu;
4546 	local_paca->kvm_hstate.ptid = 0;
4547 	local_paca->kvm_hstate.fake_suspend = 0;
4548 
4549 	/*
4550 	 * Orders set cpu/thread_cpu vs testing for pending interrupts and
4551 	 * doorbells below. The other side is when these fields are set vs
4552 	 * kvmppc_fast_vcpu_kick_hv reading the cpu/thread_cpu fields to
4553 	 * kick a vCPU to notice the pending interrupt.
4554 	 */
4555 	smp_mb();
4556 
4557 	if (!nested) {
4558 		kvmppc_core_prepare_to_enter(vcpu);
4559 		if (vcpu->arch.shregs.msr & MSR_EE) {
4560 			if (xive_interrupt_pending(vcpu))
4561 				kvmppc_inject_interrupt_hv(vcpu,
4562 						BOOK3S_INTERRUPT_EXTERNAL, 0);
4563 		} else if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4564 			     &vcpu->arch.pending_exceptions)) {
4565 			lpcr |= LPCR_MER;
4566 		}
4567 	} else if (vcpu->arch.pending_exceptions ||
4568 		   vcpu->arch.doorbell_request ||
4569 		   xive_interrupt_pending(vcpu)) {
4570 		vcpu->arch.ret = RESUME_HOST;
4571 		goto out;
4572 	}
4573 
4574 	if (vcpu->arch.timer_running) {
4575 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
4576 		vcpu->arch.timer_running = 0;
4577 	}
4578 
4579 	tb = mftb();
4580 
4581 	__kvmppc_create_dtl_entry(vcpu, pcpu, tb + vc->tb_offset, 0);
4582 
4583 	trace_kvm_guest_enter(vcpu);
4584 
4585 	guest_enter_irqoff();
4586 
4587 	srcu_idx = srcu_read_lock(&kvm->srcu);
4588 
4589 	this_cpu_disable_ftrace();
4590 
4591 	/* Tell lockdep that we're about to enable interrupts */
4592 	trace_hardirqs_on();
4593 
4594 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb);
4595 	vcpu->arch.trap = trap;
4596 
4597 	trace_hardirqs_off();
4598 
4599 	this_cpu_enable_ftrace();
4600 
4601 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4602 
4603 	set_irq_happened(trap);
4604 
4605 	context_tracking_guest_exit();
4606 	if (!vtime_accounting_enabled_this_cpu()) {
4607 		local_irq_enable();
4608 		/*
4609 		 * Service IRQs here before vtime_account_guest_exit() so any
4610 		 * ticks that occurred while running the guest are accounted to
4611 		 * the guest. If vtime accounting is enabled, accounting uses
4612 		 * TB rather than ticks, so it can be done without enabling
4613 		 * interrupts here, which has the problem that it accounts
4614 		 * interrupt processing overhead to the host.
4615 		 */
4616 		local_irq_disable();
4617 	}
4618 	vtime_account_guest_exit();
4619 
4620 	vcpu->cpu = -1;
4621 	vcpu->arch.thread_cpu = -1;
4622 
4623 	powerpc_local_irq_pmu_restore(flags);
4624 
4625 	preempt_enable();
4626 
4627 	/*
4628 	 * cancel pending decrementer exception if DEC is now positive, or if
4629 	 * entering a nested guest in which case the decrementer is now owned
4630 	 * by L2 and the L1 decrementer is provided in hdec_expires
4631 	 */
4632 	if (kvmppc_core_pending_dec(vcpu) &&
4633 			((tb < kvmppc_dec_expires_host_tb(vcpu)) ||
4634 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4635 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4636 		kvmppc_core_dequeue_dec(vcpu);
4637 
4638 	trace_kvm_guest_exit(vcpu);
4639 	r = RESUME_GUEST;
4640 	if (trap) {
4641 		if (!nested)
4642 			r = kvmppc_handle_exit_hv(vcpu, current);
4643 		else
4644 			r = kvmppc_handle_nested_exit(vcpu);
4645 	}
4646 	vcpu->arch.ret = r;
4647 
4648 	if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) {
4649 		kvmppc_set_timer(vcpu);
4650 
4651 		prepare_to_rcuwait(wait);
4652 		for (;;) {
4653 			set_current_state(TASK_INTERRUPTIBLE);
4654 			if (signal_pending(current)) {
4655 				vcpu->stat.signal_exits++;
4656 				run->exit_reason = KVM_EXIT_INTR;
4657 				vcpu->arch.ret = -EINTR;
4658 				break;
4659 			}
4660 
4661 			if (kvmppc_vcpu_check_block(vcpu))
4662 				break;
4663 
4664 			trace_kvmppc_vcore_blocked(vcpu, 0);
4665 			schedule();
4666 			trace_kvmppc_vcore_blocked(vcpu, 1);
4667 		}
4668 		finish_rcuwait(wait);
4669 	}
4670 	vcpu->arch.ceded = 0;
4671 
4672  done:
4673 	trace_kvmppc_run_vcpu_exit(vcpu);
4674 
4675 	return vcpu->arch.ret;
4676 
4677  sigpend:
4678 	vcpu->stat.signal_exits++;
4679 	run->exit_reason = KVM_EXIT_INTR;
4680 	vcpu->arch.ret = -EINTR;
4681  out:
4682 	vcpu->cpu = -1;
4683 	vcpu->arch.thread_cpu = -1;
4684 	powerpc_local_irq_pmu_restore(flags);
4685 	preempt_enable();
4686 	goto done;
4687 }
4688 
4689 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4690 {
4691 	struct kvm_run *run = vcpu->run;
4692 	int r;
4693 	int srcu_idx;
4694 	struct kvm *kvm;
4695 	unsigned long msr;
4696 
4697 	if (!vcpu->arch.sane) {
4698 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4699 		return -EINVAL;
4700 	}
4701 
4702 	/* No need to go into the guest when all we'll do is come back out */
4703 	if (signal_pending(current)) {
4704 		run->exit_reason = KVM_EXIT_INTR;
4705 		return -EINTR;
4706 	}
4707 
4708 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4709 	/*
4710 	 * Don't allow entry with a suspended transaction, because
4711 	 * the guest entry/exit code will lose it.
4712 	 */
4713 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4714 	    (current->thread.regs->msr & MSR_TM)) {
4715 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4716 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4717 			run->fail_entry.hardware_entry_failure_reason = 0;
4718 			return -EINVAL;
4719 		}
4720 	}
4721 #endif
4722 
4723 	/*
4724 	 * Force online to 1 for the sake of old userspace which doesn't
4725 	 * set it.
4726 	 */
4727 	if (!vcpu->arch.online) {
4728 		atomic_inc(&vcpu->arch.vcore->online_count);
4729 		vcpu->arch.online = 1;
4730 	}
4731 
4732 	kvmppc_core_prepare_to_enter(vcpu);
4733 
4734 	kvm = vcpu->kvm;
4735 	atomic_inc(&kvm->arch.vcpus_running);
4736 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4737 	smp_mb();
4738 
4739 	msr = 0;
4740 	if (IS_ENABLED(CONFIG_PPC_FPU))
4741 		msr |= MSR_FP;
4742 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
4743 		msr |= MSR_VEC;
4744 	if (cpu_has_feature(CPU_FTR_VSX))
4745 		msr |= MSR_VSX;
4746 	if ((cpu_has_feature(CPU_FTR_TM) ||
4747 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
4748 			(vcpu->arch.hfscr & HFSCR_TM))
4749 		msr |= MSR_TM;
4750 	msr = msr_check_and_set(msr);
4751 
4752 	kvmppc_save_user_regs();
4753 
4754 	kvmppc_save_current_sprs();
4755 
4756 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4757 		vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4758 	vcpu->arch.pgdir = kvm->mm->pgd;
4759 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4760 
4761 	do {
4762 		if (cpu_has_feature(CPU_FTR_ARCH_300))
4763 			r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4764 						  vcpu->arch.vcore->lpcr);
4765 		else
4766 			r = kvmppc_run_vcpu(vcpu);
4767 
4768 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4769 			if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) {
4770 				/*
4771 				 * These should have been caught reflected
4772 				 * into the guest by now. Final sanity check:
4773 				 * don't allow userspace to execute hcalls in
4774 				 * the hypervisor.
4775 				 */
4776 				r = RESUME_GUEST;
4777 				continue;
4778 			}
4779 			trace_kvm_hcall_enter(vcpu);
4780 			r = kvmppc_pseries_do_hcall(vcpu);
4781 			trace_kvm_hcall_exit(vcpu, r);
4782 			kvmppc_core_prepare_to_enter(vcpu);
4783 		} else if (r == RESUME_PAGE_FAULT) {
4784 			srcu_idx = srcu_read_lock(&kvm->srcu);
4785 			r = kvmppc_book3s_hv_page_fault(vcpu,
4786 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4787 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4788 		} else if (r == RESUME_PASSTHROUGH) {
4789 			if (WARN_ON(xics_on_xive()))
4790 				r = H_SUCCESS;
4791 			else
4792 				r = kvmppc_xics_rm_complete(vcpu, 0);
4793 		}
4794 	} while (is_kvmppc_resume_guest(r));
4795 
4796 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4797 	atomic_dec(&kvm->arch.vcpus_running);
4798 
4799 	srr_regs_clobbered();
4800 
4801 	return r;
4802 }
4803 
4804 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4805 				     int shift, int sllp)
4806 {
4807 	(*sps)->page_shift = shift;
4808 	(*sps)->slb_enc = sllp;
4809 	(*sps)->enc[0].page_shift = shift;
4810 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4811 	/*
4812 	 * Add 16MB MPSS support (may get filtered out by userspace)
4813 	 */
4814 	if (shift != 24) {
4815 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4816 		if (penc != -1) {
4817 			(*sps)->enc[1].page_shift = 24;
4818 			(*sps)->enc[1].pte_enc = penc;
4819 		}
4820 	}
4821 	(*sps)++;
4822 }
4823 
4824 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4825 					 struct kvm_ppc_smmu_info *info)
4826 {
4827 	struct kvm_ppc_one_seg_page_size *sps;
4828 
4829 	/*
4830 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4831 	 * POWER7 doesn't support keys for instruction accesses,
4832 	 * POWER8 and POWER9 do.
4833 	 */
4834 	info->data_keys = 32;
4835 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4836 
4837 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4838 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4839 	info->slb_size = 32;
4840 
4841 	/* We only support these sizes for now, and no muti-size segments */
4842 	sps = &info->sps[0];
4843 	kvmppc_add_seg_page_size(&sps, 12, 0);
4844 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4845 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4846 
4847 	/* If running as a nested hypervisor, we don't support HPT guests */
4848 	if (kvmhv_on_pseries())
4849 		info->flags |= KVM_PPC_NO_HASH;
4850 
4851 	return 0;
4852 }
4853 
4854 /*
4855  * Get (and clear) the dirty memory log for a memory slot.
4856  */
4857 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4858 					 struct kvm_dirty_log *log)
4859 {
4860 	struct kvm_memslots *slots;
4861 	struct kvm_memory_slot *memslot;
4862 	int r;
4863 	unsigned long n, i;
4864 	unsigned long *buf, *p;
4865 	struct kvm_vcpu *vcpu;
4866 
4867 	mutex_lock(&kvm->slots_lock);
4868 
4869 	r = -EINVAL;
4870 	if (log->slot >= KVM_USER_MEM_SLOTS)
4871 		goto out;
4872 
4873 	slots = kvm_memslots(kvm);
4874 	memslot = id_to_memslot(slots, log->slot);
4875 	r = -ENOENT;
4876 	if (!memslot || !memslot->dirty_bitmap)
4877 		goto out;
4878 
4879 	/*
4880 	 * Use second half of bitmap area because both HPT and radix
4881 	 * accumulate bits in the first half.
4882 	 */
4883 	n = kvm_dirty_bitmap_bytes(memslot);
4884 	buf = memslot->dirty_bitmap + n / sizeof(long);
4885 	memset(buf, 0, n);
4886 
4887 	if (kvm_is_radix(kvm))
4888 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4889 	else
4890 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4891 	if (r)
4892 		goto out;
4893 
4894 	/*
4895 	 * We accumulate dirty bits in the first half of the
4896 	 * memslot's dirty_bitmap area, for when pages are paged
4897 	 * out or modified by the host directly.  Pick up these
4898 	 * bits and add them to the map.
4899 	 */
4900 	p = memslot->dirty_bitmap;
4901 	for (i = 0; i < n / sizeof(long); ++i)
4902 		buf[i] |= xchg(&p[i], 0);
4903 
4904 	/* Harvest dirty bits from VPA and DTL updates */
4905 	/* Note: we never modify the SLB shadow buffer areas */
4906 	kvm_for_each_vcpu(i, vcpu, kvm) {
4907 		spin_lock(&vcpu->arch.vpa_update_lock);
4908 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4909 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4910 		spin_unlock(&vcpu->arch.vpa_update_lock);
4911 	}
4912 
4913 	r = -EFAULT;
4914 	if (copy_to_user(log->dirty_bitmap, buf, n))
4915 		goto out;
4916 
4917 	r = 0;
4918 out:
4919 	mutex_unlock(&kvm->slots_lock);
4920 	return r;
4921 }
4922 
4923 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4924 {
4925 	vfree(slot->arch.rmap);
4926 	slot->arch.rmap = NULL;
4927 }
4928 
4929 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4930 				const struct kvm_memory_slot *old,
4931 				struct kvm_memory_slot *new,
4932 				enum kvm_mr_change change)
4933 {
4934 	if (change == KVM_MR_CREATE) {
4935 		unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap));
4936 
4937 		if ((size >> PAGE_SHIFT) > totalram_pages())
4938 			return -ENOMEM;
4939 
4940 		new->arch.rmap = vzalloc(size);
4941 		if (!new->arch.rmap)
4942 			return -ENOMEM;
4943 	} else if (change != KVM_MR_DELETE) {
4944 		new->arch.rmap = old->arch.rmap;
4945 	}
4946 
4947 	return 0;
4948 }
4949 
4950 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4951 				struct kvm_memory_slot *old,
4952 				const struct kvm_memory_slot *new,
4953 				enum kvm_mr_change change)
4954 {
4955 	/*
4956 	 * If we are creating or modifying a memslot, it might make
4957 	 * some address that was previously cached as emulated
4958 	 * MMIO be no longer emulated MMIO, so invalidate
4959 	 * all the caches of emulated MMIO translations.
4960 	 */
4961 	if (change != KVM_MR_DELETE)
4962 		atomic64_inc(&kvm->arch.mmio_update);
4963 
4964 	/*
4965 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4966 	 * have already called kvm_arch_flush_shadow_memslot() to
4967 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
4968 	 * previous mappings.  So the only case to handle is
4969 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4970 	 * has been changed.
4971 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4972 	 * to get rid of any THP PTEs in the partition-scoped page tables
4973 	 * so we can track dirtiness at the page level; we flush when
4974 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4975 	 * using THP PTEs.
4976 	 */
4977 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4978 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4979 		kvmppc_radix_flush_memslot(kvm, old);
4980 	/*
4981 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4982 	 */
4983 	if (!kvm->arch.secure_guest)
4984 		return;
4985 
4986 	switch (change) {
4987 	case KVM_MR_CREATE:
4988 		/*
4989 		 * @TODO kvmppc_uvmem_memslot_create() can fail and
4990 		 * return error. Fix this.
4991 		 */
4992 		kvmppc_uvmem_memslot_create(kvm, new);
4993 		break;
4994 	case KVM_MR_DELETE:
4995 		kvmppc_uvmem_memslot_delete(kvm, old);
4996 		break;
4997 	default:
4998 		/* TODO: Handle KVM_MR_MOVE */
4999 		break;
5000 	}
5001 }
5002 
5003 /*
5004  * Update LPCR values in kvm->arch and in vcores.
5005  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
5006  * of kvm->arch.lpcr update).
5007  */
5008 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
5009 {
5010 	long int i;
5011 	u32 cores_done = 0;
5012 
5013 	if ((kvm->arch.lpcr & mask) == lpcr)
5014 		return;
5015 
5016 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
5017 
5018 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
5019 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
5020 		if (!vc)
5021 			continue;
5022 
5023 		spin_lock(&vc->lock);
5024 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
5025 		verify_lpcr(kvm, vc->lpcr);
5026 		spin_unlock(&vc->lock);
5027 		if (++cores_done >= kvm->arch.online_vcores)
5028 			break;
5029 	}
5030 }
5031 
5032 void kvmppc_setup_partition_table(struct kvm *kvm)
5033 {
5034 	unsigned long dw0, dw1;
5035 
5036 	if (!kvm_is_radix(kvm)) {
5037 		/* PS field - page size for VRMA */
5038 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
5039 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
5040 		/* HTABSIZE and HTABORG fields */
5041 		dw0 |= kvm->arch.sdr1;
5042 
5043 		/* Second dword as set by userspace */
5044 		dw1 = kvm->arch.process_table;
5045 	} else {
5046 		dw0 = PATB_HR | radix__get_tree_size() |
5047 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
5048 		dw1 = PATB_GR | kvm->arch.process_table;
5049 	}
5050 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
5051 }
5052 
5053 /*
5054  * Set up HPT (hashed page table) and RMA (real-mode area).
5055  * Must be called with kvm->arch.mmu_setup_lock held.
5056  */
5057 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
5058 {
5059 	int err = 0;
5060 	struct kvm *kvm = vcpu->kvm;
5061 	unsigned long hva;
5062 	struct kvm_memory_slot *memslot;
5063 	struct vm_area_struct *vma;
5064 	unsigned long lpcr = 0, senc;
5065 	unsigned long psize, porder;
5066 	int srcu_idx;
5067 
5068 	/* Allocate hashed page table (if not done already) and reset it */
5069 	if (!kvm->arch.hpt.virt) {
5070 		int order = KVM_DEFAULT_HPT_ORDER;
5071 		struct kvm_hpt_info info;
5072 
5073 		err = kvmppc_allocate_hpt(&info, order);
5074 		/* If we get here, it means userspace didn't specify a
5075 		 * size explicitly.  So, try successively smaller
5076 		 * sizes if the default failed. */
5077 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5078 			err  = kvmppc_allocate_hpt(&info, order);
5079 
5080 		if (err < 0) {
5081 			pr_err("KVM: Couldn't alloc HPT\n");
5082 			goto out;
5083 		}
5084 
5085 		kvmppc_set_hpt(kvm, &info);
5086 	}
5087 
5088 	/* Look up the memslot for guest physical address 0 */
5089 	srcu_idx = srcu_read_lock(&kvm->srcu);
5090 	memslot = gfn_to_memslot(kvm, 0);
5091 
5092 	/* We must have some memory at 0 by now */
5093 	err = -EINVAL;
5094 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5095 		goto out_srcu;
5096 
5097 	/* Look up the VMA for the start of this memory slot */
5098 	hva = memslot->userspace_addr;
5099 	mmap_read_lock(kvm->mm);
5100 	vma = vma_lookup(kvm->mm, hva);
5101 	if (!vma || (vma->vm_flags & VM_IO))
5102 		goto up_out;
5103 
5104 	psize = vma_kernel_pagesize(vma);
5105 
5106 	mmap_read_unlock(kvm->mm);
5107 
5108 	/* We can handle 4k, 64k or 16M pages in the VRMA */
5109 	if (psize >= 0x1000000)
5110 		psize = 0x1000000;
5111 	else if (psize >= 0x10000)
5112 		psize = 0x10000;
5113 	else
5114 		psize = 0x1000;
5115 	porder = __ilog2(psize);
5116 
5117 	senc = slb_pgsize_encoding(psize);
5118 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5119 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5120 	/* Create HPTEs in the hash page table for the VRMA */
5121 	kvmppc_map_vrma(vcpu, memslot, porder);
5122 
5123 	/* Update VRMASD field in the LPCR */
5124 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5125 		/* the -4 is to account for senc values starting at 0x10 */
5126 		lpcr = senc << (LPCR_VRMASD_SH - 4);
5127 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5128 	}
5129 
5130 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5131 	smp_wmb();
5132 	err = 0;
5133  out_srcu:
5134 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5135  out:
5136 	return err;
5137 
5138  up_out:
5139 	mmap_read_unlock(kvm->mm);
5140 	goto out_srcu;
5141 }
5142 
5143 /*
5144  * Must be called with kvm->arch.mmu_setup_lock held and
5145  * mmu_ready = 0 and no vcpus running.
5146  */
5147 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5148 {
5149 	unsigned long lpcr, lpcr_mask;
5150 
5151 	if (nesting_enabled(kvm))
5152 		kvmhv_release_all_nested(kvm);
5153 	kvmppc_rmap_reset(kvm);
5154 	kvm->arch.process_table = 0;
5155 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5156 	spin_lock(&kvm->mmu_lock);
5157 	kvm->arch.radix = 0;
5158 	spin_unlock(&kvm->mmu_lock);
5159 	kvmppc_free_radix(kvm);
5160 
5161 	lpcr = LPCR_VPM1;
5162 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5163 	if (cpu_has_feature(CPU_FTR_ARCH_31))
5164 		lpcr_mask |= LPCR_HAIL;
5165 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5166 
5167 	return 0;
5168 }
5169 
5170 /*
5171  * Must be called with kvm->arch.mmu_setup_lock held and
5172  * mmu_ready = 0 and no vcpus running.
5173  */
5174 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5175 {
5176 	unsigned long lpcr, lpcr_mask;
5177 	int err;
5178 
5179 	err = kvmppc_init_vm_radix(kvm);
5180 	if (err)
5181 		return err;
5182 	kvmppc_rmap_reset(kvm);
5183 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5184 	spin_lock(&kvm->mmu_lock);
5185 	kvm->arch.radix = 1;
5186 	spin_unlock(&kvm->mmu_lock);
5187 	kvmppc_free_hpt(&kvm->arch.hpt);
5188 
5189 	lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5190 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5191 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5192 		lpcr_mask |= LPCR_HAIL;
5193 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5194 				(kvm->arch.host_lpcr & LPCR_HAIL))
5195 			lpcr |= LPCR_HAIL;
5196 	}
5197 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5198 
5199 	return 0;
5200 }
5201 
5202 #ifdef CONFIG_KVM_XICS
5203 /*
5204  * Allocate a per-core structure for managing state about which cores are
5205  * running in the host versus the guest and for exchanging data between
5206  * real mode KVM and CPU running in the host.
5207  * This is only done for the first VM.
5208  * The allocated structure stays even if all VMs have stopped.
5209  * It is only freed when the kvm-hv module is unloaded.
5210  * It's OK for this routine to fail, we just don't support host
5211  * core operations like redirecting H_IPI wakeups.
5212  */
5213 void kvmppc_alloc_host_rm_ops(void)
5214 {
5215 	struct kvmppc_host_rm_ops *ops;
5216 	unsigned long l_ops;
5217 	int cpu, core;
5218 	int size;
5219 
5220 	if (cpu_has_feature(CPU_FTR_ARCH_300))
5221 		return;
5222 
5223 	/* Not the first time here ? */
5224 	if (kvmppc_host_rm_ops_hv != NULL)
5225 		return;
5226 
5227 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5228 	if (!ops)
5229 		return;
5230 
5231 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5232 	ops->rm_core = kzalloc(size, GFP_KERNEL);
5233 
5234 	if (!ops->rm_core) {
5235 		kfree(ops);
5236 		return;
5237 	}
5238 
5239 	cpus_read_lock();
5240 
5241 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5242 		if (!cpu_online(cpu))
5243 			continue;
5244 
5245 		core = cpu >> threads_shift;
5246 		ops->rm_core[core].rm_state.in_host = 1;
5247 	}
5248 
5249 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5250 
5251 	/*
5252 	 * Make the contents of the kvmppc_host_rm_ops structure visible
5253 	 * to other CPUs before we assign it to the global variable.
5254 	 * Do an atomic assignment (no locks used here), but if someone
5255 	 * beats us to it, just free our copy and return.
5256 	 */
5257 	smp_wmb();
5258 	l_ops = (unsigned long) ops;
5259 
5260 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5261 		cpus_read_unlock();
5262 		kfree(ops->rm_core);
5263 		kfree(ops);
5264 		return;
5265 	}
5266 
5267 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5268 					     "ppc/kvm_book3s:prepare",
5269 					     kvmppc_set_host_core,
5270 					     kvmppc_clear_host_core);
5271 	cpus_read_unlock();
5272 }
5273 
5274 void kvmppc_free_host_rm_ops(void)
5275 {
5276 	if (kvmppc_host_rm_ops_hv) {
5277 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5278 		kfree(kvmppc_host_rm_ops_hv->rm_core);
5279 		kfree(kvmppc_host_rm_ops_hv);
5280 		kvmppc_host_rm_ops_hv = NULL;
5281 	}
5282 }
5283 #endif
5284 
5285 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5286 {
5287 	unsigned long lpcr, lpid;
5288 	int ret;
5289 
5290 	mutex_init(&kvm->arch.uvmem_lock);
5291 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5292 	mutex_init(&kvm->arch.mmu_setup_lock);
5293 
5294 	/* Allocate the guest's logical partition ID */
5295 
5296 	lpid = kvmppc_alloc_lpid();
5297 	if ((long)lpid < 0)
5298 		return -ENOMEM;
5299 	kvm->arch.lpid = lpid;
5300 
5301 	kvmppc_alloc_host_rm_ops();
5302 
5303 	kvmhv_vm_nested_init(kvm);
5304 
5305 	/*
5306 	 * Since we don't flush the TLB when tearing down a VM,
5307 	 * and this lpid might have previously been used,
5308 	 * make sure we flush on each core before running the new VM.
5309 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5310 	 * does this flush for us.
5311 	 */
5312 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5313 		cpumask_setall(&kvm->arch.need_tlb_flush);
5314 
5315 	/* Start out with the default set of hcalls enabled */
5316 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5317 	       sizeof(kvm->arch.enabled_hcalls));
5318 
5319 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5320 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5321 
5322 	/* Init LPCR for virtual RMA mode */
5323 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
5324 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
5325 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5326 		lpcr &= LPCR_PECE | LPCR_LPES;
5327 	} else {
5328 		/*
5329 		 * The L2 LPES mode will be set by the L0 according to whether
5330 		 * or not it needs to take external interrupts in HV mode.
5331 		 */
5332 		lpcr = 0;
5333 	}
5334 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5335 		LPCR_VPM0 | LPCR_VPM1;
5336 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5337 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5338 	/* On POWER8 turn on online bit to enable PURR/SPURR */
5339 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
5340 		lpcr |= LPCR_ONL;
5341 	/*
5342 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5343 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
5344 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5345 	 * be unnecessary but better safe than sorry in case we re-enable
5346 	 * EE in HV mode with this LPCR still set)
5347 	 */
5348 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5349 		lpcr &= ~LPCR_VPM0;
5350 		lpcr |= LPCR_HVICE | LPCR_HEIC;
5351 
5352 		/*
5353 		 * If xive is enabled, we route 0x500 interrupts directly
5354 		 * to the guest.
5355 		 */
5356 		if (xics_on_xive())
5357 			lpcr |= LPCR_LPES;
5358 	}
5359 
5360 	/*
5361 	 * If the host uses radix, the guest starts out as radix.
5362 	 */
5363 	if (radix_enabled()) {
5364 		kvm->arch.radix = 1;
5365 		kvm->arch.mmu_ready = 1;
5366 		lpcr &= ~LPCR_VPM1;
5367 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5368 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5369 		    cpu_has_feature(CPU_FTR_ARCH_31) &&
5370 		    (kvm->arch.host_lpcr & LPCR_HAIL))
5371 			lpcr |= LPCR_HAIL;
5372 		ret = kvmppc_init_vm_radix(kvm);
5373 		if (ret) {
5374 			kvmppc_free_lpid(kvm->arch.lpid);
5375 			return ret;
5376 		}
5377 		kvmppc_setup_partition_table(kvm);
5378 	}
5379 
5380 	verify_lpcr(kvm, lpcr);
5381 	kvm->arch.lpcr = lpcr;
5382 
5383 	/* Initialization for future HPT resizes */
5384 	kvm->arch.resize_hpt = NULL;
5385 
5386 	/*
5387 	 * Work out how many sets the TLB has, for the use of
5388 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5389 	 */
5390 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5391 		/*
5392 		 * P10 will flush all the congruence class with a single tlbiel
5393 		 */
5394 		kvm->arch.tlb_sets = 1;
5395 	} else if (radix_enabled())
5396 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
5397 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
5398 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
5399 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5400 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
5401 	else
5402 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
5403 
5404 	/*
5405 	 * Track that we now have a HV mode VM active. This blocks secondary
5406 	 * CPU threads from coming online.
5407 	 */
5408 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5409 		kvm_hv_vm_activated();
5410 
5411 	/*
5412 	 * Initialize smt_mode depending on processor.
5413 	 * POWER8 and earlier have to use "strict" threading, where
5414 	 * all vCPUs in a vcore have to run on the same (sub)core,
5415 	 * whereas on POWER9 the threads can each run a different
5416 	 * guest.
5417 	 */
5418 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5419 		kvm->arch.smt_mode = threads_per_subcore;
5420 	else
5421 		kvm->arch.smt_mode = 1;
5422 	kvm->arch.emul_smt_mode = 1;
5423 
5424 	return 0;
5425 }
5426 
5427 static int kvmppc_arch_create_vm_debugfs_hv(struct kvm *kvm)
5428 {
5429 	kvmppc_mmu_debugfs_init(kvm);
5430 	if (radix_enabled())
5431 		kvmhv_radix_debugfs_init(kvm);
5432 	return 0;
5433 }
5434 
5435 static void kvmppc_free_vcores(struct kvm *kvm)
5436 {
5437 	long int i;
5438 
5439 	for (i = 0; i < KVM_MAX_VCORES; ++i)
5440 		kfree(kvm->arch.vcores[i]);
5441 	kvm->arch.online_vcores = 0;
5442 }
5443 
5444 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5445 {
5446 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5447 		kvm_hv_vm_deactivated();
5448 
5449 	kvmppc_free_vcores(kvm);
5450 
5451 
5452 	if (kvm_is_radix(kvm))
5453 		kvmppc_free_radix(kvm);
5454 	else
5455 		kvmppc_free_hpt(&kvm->arch.hpt);
5456 
5457 	/* Perform global invalidation and return lpid to the pool */
5458 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5459 		if (nesting_enabled(kvm))
5460 			kvmhv_release_all_nested(kvm);
5461 		kvm->arch.process_table = 0;
5462 		if (kvm->arch.secure_guest)
5463 			uv_svm_terminate(kvm->arch.lpid);
5464 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5465 	}
5466 
5467 	kvmppc_free_lpid(kvm->arch.lpid);
5468 
5469 	kvmppc_free_pimap(kvm);
5470 }
5471 
5472 /* We don't need to emulate any privileged instructions or dcbz */
5473 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5474 				     unsigned int inst, int *advance)
5475 {
5476 	return EMULATE_FAIL;
5477 }
5478 
5479 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5480 					ulong spr_val)
5481 {
5482 	return EMULATE_FAIL;
5483 }
5484 
5485 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5486 					ulong *spr_val)
5487 {
5488 	return EMULATE_FAIL;
5489 }
5490 
5491 static int kvmppc_core_check_processor_compat_hv(void)
5492 {
5493 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5494 	    cpu_has_feature(CPU_FTR_ARCH_206))
5495 		return 0;
5496 
5497 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5498 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5499 		return 0;
5500 
5501 	return -EIO;
5502 }
5503 
5504 #ifdef CONFIG_KVM_XICS
5505 
5506 void kvmppc_free_pimap(struct kvm *kvm)
5507 {
5508 	kfree(kvm->arch.pimap);
5509 }
5510 
5511 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5512 {
5513 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5514 }
5515 
5516 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5517 {
5518 	struct irq_desc *desc;
5519 	struct kvmppc_irq_map *irq_map;
5520 	struct kvmppc_passthru_irqmap *pimap;
5521 	struct irq_chip *chip;
5522 	int i, rc = 0;
5523 	struct irq_data *host_data;
5524 
5525 	if (!kvm_irq_bypass)
5526 		return 1;
5527 
5528 	desc = irq_to_desc(host_irq);
5529 	if (!desc)
5530 		return -EIO;
5531 
5532 	mutex_lock(&kvm->lock);
5533 
5534 	pimap = kvm->arch.pimap;
5535 	if (pimap == NULL) {
5536 		/* First call, allocate structure to hold IRQ map */
5537 		pimap = kvmppc_alloc_pimap();
5538 		if (pimap == NULL) {
5539 			mutex_unlock(&kvm->lock);
5540 			return -ENOMEM;
5541 		}
5542 		kvm->arch.pimap = pimap;
5543 	}
5544 
5545 	/*
5546 	 * For now, we only support interrupts for which the EOI operation
5547 	 * is an OPAL call followed by a write to XIRR, since that's
5548 	 * what our real-mode EOI code does, or a XIVE interrupt
5549 	 */
5550 	chip = irq_data_get_irq_chip(&desc->irq_data);
5551 	if (!chip || !is_pnv_opal_msi(chip)) {
5552 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5553 			host_irq, guest_gsi);
5554 		mutex_unlock(&kvm->lock);
5555 		return -ENOENT;
5556 	}
5557 
5558 	/*
5559 	 * See if we already have an entry for this guest IRQ number.
5560 	 * If it's mapped to a hardware IRQ number, that's an error,
5561 	 * otherwise re-use this entry.
5562 	 */
5563 	for (i = 0; i < pimap->n_mapped; i++) {
5564 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5565 			if (pimap->mapped[i].r_hwirq) {
5566 				mutex_unlock(&kvm->lock);
5567 				return -EINVAL;
5568 			}
5569 			break;
5570 		}
5571 	}
5572 
5573 	if (i == KVMPPC_PIRQ_MAPPED) {
5574 		mutex_unlock(&kvm->lock);
5575 		return -EAGAIN;		/* table is full */
5576 	}
5577 
5578 	irq_map = &pimap->mapped[i];
5579 
5580 	irq_map->v_hwirq = guest_gsi;
5581 	irq_map->desc = desc;
5582 
5583 	/*
5584 	 * Order the above two stores before the next to serialize with
5585 	 * the KVM real mode handler.
5586 	 */
5587 	smp_wmb();
5588 
5589 	/*
5590 	 * The 'host_irq' number is mapped in the PCI-MSI domain but
5591 	 * the underlying calls, which will EOI the interrupt in real
5592 	 * mode, need an HW IRQ number mapped in the XICS IRQ domain.
5593 	 */
5594 	host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq);
5595 	irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
5596 
5597 	if (i == pimap->n_mapped)
5598 		pimap->n_mapped++;
5599 
5600 	if (xics_on_xive())
5601 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
5602 	else
5603 		kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
5604 	if (rc)
5605 		irq_map->r_hwirq = 0;
5606 
5607 	mutex_unlock(&kvm->lock);
5608 
5609 	return 0;
5610 }
5611 
5612 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5613 {
5614 	struct irq_desc *desc;
5615 	struct kvmppc_passthru_irqmap *pimap;
5616 	int i, rc = 0;
5617 
5618 	if (!kvm_irq_bypass)
5619 		return 0;
5620 
5621 	desc = irq_to_desc(host_irq);
5622 	if (!desc)
5623 		return -EIO;
5624 
5625 	mutex_lock(&kvm->lock);
5626 	if (!kvm->arch.pimap)
5627 		goto unlock;
5628 
5629 	pimap = kvm->arch.pimap;
5630 
5631 	for (i = 0; i < pimap->n_mapped; i++) {
5632 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5633 			break;
5634 	}
5635 
5636 	if (i == pimap->n_mapped) {
5637 		mutex_unlock(&kvm->lock);
5638 		return -ENODEV;
5639 	}
5640 
5641 	if (xics_on_xive())
5642 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
5643 	else
5644 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5645 
5646 	/* invalidate the entry (what do do on error from the above ?) */
5647 	pimap->mapped[i].r_hwirq = 0;
5648 
5649 	/*
5650 	 * We don't free this structure even when the count goes to
5651 	 * zero. The structure is freed when we destroy the VM.
5652 	 */
5653  unlock:
5654 	mutex_unlock(&kvm->lock);
5655 	return rc;
5656 }
5657 
5658 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5659 					     struct irq_bypass_producer *prod)
5660 {
5661 	int ret = 0;
5662 	struct kvm_kernel_irqfd *irqfd =
5663 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5664 
5665 	irqfd->producer = prod;
5666 
5667 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5668 	if (ret)
5669 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5670 			prod->irq, irqfd->gsi, ret);
5671 
5672 	return ret;
5673 }
5674 
5675 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5676 					      struct irq_bypass_producer *prod)
5677 {
5678 	int ret;
5679 	struct kvm_kernel_irqfd *irqfd =
5680 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5681 
5682 	irqfd->producer = NULL;
5683 
5684 	/*
5685 	 * When producer of consumer is unregistered, we change back to
5686 	 * default external interrupt handling mode - KVM real mode
5687 	 * will switch back to host.
5688 	 */
5689 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5690 	if (ret)
5691 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5692 			prod->irq, irqfd->gsi, ret);
5693 }
5694 #endif
5695 
5696 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5697 				 unsigned int ioctl, unsigned long arg)
5698 {
5699 	struct kvm *kvm __maybe_unused = filp->private_data;
5700 	void __user *argp = (void __user *)arg;
5701 	long r;
5702 
5703 	switch (ioctl) {
5704 
5705 	case KVM_PPC_ALLOCATE_HTAB: {
5706 		u32 htab_order;
5707 
5708 		/* If we're a nested hypervisor, we currently only support radix */
5709 		if (kvmhv_on_pseries()) {
5710 			r = -EOPNOTSUPP;
5711 			break;
5712 		}
5713 
5714 		r = -EFAULT;
5715 		if (get_user(htab_order, (u32 __user *)argp))
5716 			break;
5717 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5718 		if (r)
5719 			break;
5720 		r = 0;
5721 		break;
5722 	}
5723 
5724 	case KVM_PPC_GET_HTAB_FD: {
5725 		struct kvm_get_htab_fd ghf;
5726 
5727 		r = -EFAULT;
5728 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5729 			break;
5730 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5731 		break;
5732 	}
5733 
5734 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5735 		struct kvm_ppc_resize_hpt rhpt;
5736 
5737 		r = -EFAULT;
5738 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5739 			break;
5740 
5741 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5742 		break;
5743 	}
5744 
5745 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5746 		struct kvm_ppc_resize_hpt rhpt;
5747 
5748 		r = -EFAULT;
5749 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5750 			break;
5751 
5752 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5753 		break;
5754 	}
5755 
5756 	default:
5757 		r = -ENOTTY;
5758 	}
5759 
5760 	return r;
5761 }
5762 
5763 /*
5764  * List of hcall numbers to enable by default.
5765  * For compatibility with old userspace, we enable by default
5766  * all hcalls that were implemented before the hcall-enabling
5767  * facility was added.  Note this list should not include H_RTAS.
5768  */
5769 static unsigned int default_hcall_list[] = {
5770 	H_REMOVE,
5771 	H_ENTER,
5772 	H_READ,
5773 	H_PROTECT,
5774 	H_BULK_REMOVE,
5775 #ifdef CONFIG_SPAPR_TCE_IOMMU
5776 	H_GET_TCE,
5777 	H_PUT_TCE,
5778 #endif
5779 	H_SET_DABR,
5780 	H_SET_XDABR,
5781 	H_CEDE,
5782 	H_PROD,
5783 	H_CONFER,
5784 	H_REGISTER_VPA,
5785 #ifdef CONFIG_KVM_XICS
5786 	H_EOI,
5787 	H_CPPR,
5788 	H_IPI,
5789 	H_IPOLL,
5790 	H_XIRR,
5791 	H_XIRR_X,
5792 #endif
5793 	0
5794 };
5795 
5796 static void init_default_hcalls(void)
5797 {
5798 	int i;
5799 	unsigned int hcall;
5800 
5801 	for (i = 0; default_hcall_list[i]; ++i) {
5802 		hcall = default_hcall_list[i];
5803 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5804 		__set_bit(hcall / 4, default_enabled_hcalls);
5805 	}
5806 }
5807 
5808 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5809 {
5810 	unsigned long lpcr;
5811 	int radix;
5812 	int err;
5813 
5814 	/* If not on a POWER9, reject it */
5815 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5816 		return -ENODEV;
5817 
5818 	/* If any unknown flags set, reject it */
5819 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5820 		return -EINVAL;
5821 
5822 	/* GR (guest radix) bit in process_table field must match */
5823 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5824 	if (!!(cfg->process_table & PATB_GR) != radix)
5825 		return -EINVAL;
5826 
5827 	/* Process table size field must be reasonable, i.e. <= 24 */
5828 	if ((cfg->process_table & PRTS_MASK) > 24)
5829 		return -EINVAL;
5830 
5831 	/* We can change a guest to/from radix now, if the host is radix */
5832 	if (radix && !radix_enabled())
5833 		return -EINVAL;
5834 
5835 	/* If we're a nested hypervisor, we currently only support radix */
5836 	if (kvmhv_on_pseries() && !radix)
5837 		return -EINVAL;
5838 
5839 	mutex_lock(&kvm->arch.mmu_setup_lock);
5840 	if (radix != kvm_is_radix(kvm)) {
5841 		if (kvm->arch.mmu_ready) {
5842 			kvm->arch.mmu_ready = 0;
5843 			/* order mmu_ready vs. vcpus_running */
5844 			smp_mb();
5845 			if (atomic_read(&kvm->arch.vcpus_running)) {
5846 				kvm->arch.mmu_ready = 1;
5847 				err = -EBUSY;
5848 				goto out_unlock;
5849 			}
5850 		}
5851 		if (radix)
5852 			err = kvmppc_switch_mmu_to_radix(kvm);
5853 		else
5854 			err = kvmppc_switch_mmu_to_hpt(kvm);
5855 		if (err)
5856 			goto out_unlock;
5857 	}
5858 
5859 	kvm->arch.process_table = cfg->process_table;
5860 	kvmppc_setup_partition_table(kvm);
5861 
5862 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5863 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5864 	err = 0;
5865 
5866  out_unlock:
5867 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5868 	return err;
5869 }
5870 
5871 static int kvmhv_enable_nested(struct kvm *kvm)
5872 {
5873 	if (!nested)
5874 		return -EPERM;
5875 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5876 		return -ENODEV;
5877 	if (!radix_enabled())
5878 		return -ENODEV;
5879 
5880 	/* kvm == NULL means the caller is testing if the capability exists */
5881 	if (kvm)
5882 		kvm->arch.nested_enable = true;
5883 	return 0;
5884 }
5885 
5886 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5887 				 int size)
5888 {
5889 	int rc = -EINVAL;
5890 
5891 	if (kvmhv_vcpu_is_radix(vcpu)) {
5892 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5893 
5894 		if (rc > 0)
5895 			rc = -EINVAL;
5896 	}
5897 
5898 	/* For now quadrants are the only way to access nested guest memory */
5899 	if (rc && vcpu->arch.nested)
5900 		rc = -EAGAIN;
5901 
5902 	return rc;
5903 }
5904 
5905 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5906 				int size)
5907 {
5908 	int rc = -EINVAL;
5909 
5910 	if (kvmhv_vcpu_is_radix(vcpu)) {
5911 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5912 
5913 		if (rc > 0)
5914 			rc = -EINVAL;
5915 	}
5916 
5917 	/* For now quadrants are the only way to access nested guest memory */
5918 	if (rc && vcpu->arch.nested)
5919 		rc = -EAGAIN;
5920 
5921 	return rc;
5922 }
5923 
5924 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5925 {
5926 	unpin_vpa(kvm, vpa);
5927 	vpa->gpa = 0;
5928 	vpa->pinned_addr = NULL;
5929 	vpa->dirty = false;
5930 	vpa->update_pending = 0;
5931 }
5932 
5933 /*
5934  * Enable a guest to become a secure VM, or test whether
5935  * that could be enabled.
5936  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5937  * tested (kvm == NULL) or enabled (kvm != NULL).
5938  */
5939 static int kvmhv_enable_svm(struct kvm *kvm)
5940 {
5941 	if (!kvmppc_uvmem_available())
5942 		return -EINVAL;
5943 	if (kvm)
5944 		kvm->arch.svm_enabled = 1;
5945 	return 0;
5946 }
5947 
5948 /*
5949  *  IOCTL handler to turn off secure mode of guest
5950  *
5951  * - Release all device pages
5952  * - Issue ucall to terminate the guest on the UV side
5953  * - Unpin the VPA pages.
5954  * - Reinit the partition scoped page tables
5955  */
5956 static int kvmhv_svm_off(struct kvm *kvm)
5957 {
5958 	struct kvm_vcpu *vcpu;
5959 	int mmu_was_ready;
5960 	int srcu_idx;
5961 	int ret = 0;
5962 	unsigned long i;
5963 
5964 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5965 		return ret;
5966 
5967 	mutex_lock(&kvm->arch.mmu_setup_lock);
5968 	mmu_was_ready = kvm->arch.mmu_ready;
5969 	if (kvm->arch.mmu_ready) {
5970 		kvm->arch.mmu_ready = 0;
5971 		/* order mmu_ready vs. vcpus_running */
5972 		smp_mb();
5973 		if (atomic_read(&kvm->arch.vcpus_running)) {
5974 			kvm->arch.mmu_ready = 1;
5975 			ret = -EBUSY;
5976 			goto out;
5977 		}
5978 	}
5979 
5980 	srcu_idx = srcu_read_lock(&kvm->srcu);
5981 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5982 		struct kvm_memory_slot *memslot;
5983 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5984 		int bkt;
5985 
5986 		if (!slots)
5987 			continue;
5988 
5989 		kvm_for_each_memslot(memslot, bkt, slots) {
5990 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
5991 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5992 		}
5993 	}
5994 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5995 
5996 	ret = uv_svm_terminate(kvm->arch.lpid);
5997 	if (ret != U_SUCCESS) {
5998 		ret = -EINVAL;
5999 		goto out;
6000 	}
6001 
6002 	/*
6003 	 * When secure guest is reset, all the guest pages are sent
6004 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
6005 	 * chance to run and unpin their VPA pages. Unpinning of all
6006 	 * VPA pages is done here explicitly so that VPA pages
6007 	 * can be migrated to the secure side.
6008 	 *
6009 	 * This is required to for the secure SMP guest to reboot
6010 	 * correctly.
6011 	 */
6012 	kvm_for_each_vcpu(i, vcpu, kvm) {
6013 		spin_lock(&vcpu->arch.vpa_update_lock);
6014 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
6015 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
6016 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
6017 		spin_unlock(&vcpu->arch.vpa_update_lock);
6018 	}
6019 
6020 	kvmppc_setup_partition_table(kvm);
6021 	kvm->arch.secure_guest = 0;
6022 	kvm->arch.mmu_ready = mmu_was_ready;
6023 out:
6024 	mutex_unlock(&kvm->arch.mmu_setup_lock);
6025 	return ret;
6026 }
6027 
6028 static int kvmhv_enable_dawr1(struct kvm *kvm)
6029 {
6030 	if (!cpu_has_feature(CPU_FTR_DAWR1))
6031 		return -ENODEV;
6032 
6033 	/* kvm == NULL means the caller is testing if the capability exists */
6034 	if (kvm)
6035 		kvm->arch.dawr1_enabled = true;
6036 	return 0;
6037 }
6038 
6039 static bool kvmppc_hash_v3_possible(void)
6040 {
6041 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6042 		return false;
6043 
6044 	if (!cpu_has_feature(CPU_FTR_HVMODE))
6045 		return false;
6046 
6047 	/*
6048 	 * POWER9 chips before version 2.02 can't have some threads in
6049 	 * HPT mode and some in radix mode on the same core.
6050 	 */
6051 	if (radix_enabled()) {
6052 		unsigned int pvr = mfspr(SPRN_PVR);
6053 		if ((pvr >> 16) == PVR_POWER9 &&
6054 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
6055 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
6056 			return false;
6057 	}
6058 
6059 	return true;
6060 }
6061 
6062 static struct kvmppc_ops kvm_ops_hv = {
6063 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
6064 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
6065 	.get_one_reg = kvmppc_get_one_reg_hv,
6066 	.set_one_reg = kvmppc_set_one_reg_hv,
6067 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
6068 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
6069 	.inject_interrupt = kvmppc_inject_interrupt_hv,
6070 	.set_msr     = kvmppc_set_msr_hv,
6071 	.vcpu_run    = kvmppc_vcpu_run_hv,
6072 	.vcpu_create = kvmppc_core_vcpu_create_hv,
6073 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
6074 	.check_requests = kvmppc_core_check_requests_hv,
6075 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
6076 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
6077 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
6078 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
6079 	.unmap_gfn_range = kvm_unmap_gfn_range_hv,
6080 	.age_gfn = kvm_age_gfn_hv,
6081 	.test_age_gfn = kvm_test_age_gfn_hv,
6082 	.set_spte_gfn = kvm_set_spte_gfn_hv,
6083 	.free_memslot = kvmppc_core_free_memslot_hv,
6084 	.init_vm =  kvmppc_core_init_vm_hv,
6085 	.destroy_vm = kvmppc_core_destroy_vm_hv,
6086 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
6087 	.emulate_op = kvmppc_core_emulate_op_hv,
6088 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
6089 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
6090 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
6091 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
6092 	.hcall_implemented = kvmppc_hcall_impl_hv,
6093 #ifdef CONFIG_KVM_XICS
6094 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
6095 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
6096 #endif
6097 	.configure_mmu = kvmhv_configure_mmu,
6098 	.get_rmmu_info = kvmhv_get_rmmu_info,
6099 	.set_smt_mode = kvmhv_set_smt_mode,
6100 	.enable_nested = kvmhv_enable_nested,
6101 	.load_from_eaddr = kvmhv_load_from_eaddr,
6102 	.store_to_eaddr = kvmhv_store_to_eaddr,
6103 	.enable_svm = kvmhv_enable_svm,
6104 	.svm_off = kvmhv_svm_off,
6105 	.enable_dawr1 = kvmhv_enable_dawr1,
6106 	.hash_v3_possible = kvmppc_hash_v3_possible,
6107 	.create_vcpu_debugfs = kvmppc_arch_create_vcpu_debugfs_hv,
6108 	.create_vm_debugfs = kvmppc_arch_create_vm_debugfs_hv,
6109 };
6110 
6111 static int kvm_init_subcore_bitmap(void)
6112 {
6113 	int i, j;
6114 	int nr_cores = cpu_nr_cores();
6115 	struct sibling_subcore_state *sibling_subcore_state;
6116 
6117 	for (i = 0; i < nr_cores; i++) {
6118 		int first_cpu = i * threads_per_core;
6119 		int node = cpu_to_node(first_cpu);
6120 
6121 		/* Ignore if it is already allocated. */
6122 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
6123 			continue;
6124 
6125 		sibling_subcore_state =
6126 			kzalloc_node(sizeof(struct sibling_subcore_state),
6127 							GFP_KERNEL, node);
6128 		if (!sibling_subcore_state)
6129 			return -ENOMEM;
6130 
6131 
6132 		for (j = 0; j < threads_per_core; j++) {
6133 			int cpu = first_cpu + j;
6134 
6135 			paca_ptrs[cpu]->sibling_subcore_state =
6136 						sibling_subcore_state;
6137 		}
6138 	}
6139 	return 0;
6140 }
6141 
6142 static int kvmppc_radix_possible(void)
6143 {
6144 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6145 }
6146 
6147 static int kvmppc_book3s_init_hv(void)
6148 {
6149 	int r;
6150 
6151 	if (!tlbie_capable) {
6152 		pr_err("KVM-HV: Host does not support TLBIE\n");
6153 		return -ENODEV;
6154 	}
6155 
6156 	/*
6157 	 * FIXME!! Do we need to check on all cpus ?
6158 	 */
6159 	r = kvmppc_core_check_processor_compat_hv();
6160 	if (r < 0)
6161 		return -ENODEV;
6162 
6163 	r = kvmhv_nested_init();
6164 	if (r)
6165 		return r;
6166 
6167 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
6168 		r = kvm_init_subcore_bitmap();
6169 		if (r)
6170 			goto err;
6171 	}
6172 
6173 	/*
6174 	 * We need a way of accessing the XICS interrupt controller,
6175 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6176 	 * indirectly, via OPAL.
6177 	 */
6178 #ifdef CONFIG_SMP
6179 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
6180 	    !local_paca->kvm_hstate.xics_phys) {
6181 		struct device_node *np;
6182 
6183 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6184 		if (!np) {
6185 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6186 			r = -ENODEV;
6187 			goto err;
6188 		}
6189 		/* presence of intc confirmed - node can be dropped again */
6190 		of_node_put(np);
6191 	}
6192 #endif
6193 
6194 	init_default_hcalls();
6195 
6196 	init_vcore_lists();
6197 
6198 	r = kvmppc_mmu_hv_init();
6199 	if (r)
6200 		goto err;
6201 
6202 	if (kvmppc_radix_possible()) {
6203 		r = kvmppc_radix_init();
6204 		if (r)
6205 			goto err;
6206 	}
6207 
6208 	r = kvmppc_uvmem_init();
6209 	if (r < 0) {
6210 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6211 		return r;
6212 	}
6213 
6214 	kvm_ops_hv.owner = THIS_MODULE;
6215 	kvmppc_hv_ops = &kvm_ops_hv;
6216 
6217 	return 0;
6218 
6219 err:
6220 	kvmhv_nested_exit();
6221 	kvmppc_radix_exit();
6222 
6223 	return r;
6224 }
6225 
6226 static void kvmppc_book3s_exit_hv(void)
6227 {
6228 	kvmppc_uvmem_free();
6229 	kvmppc_free_host_rm_ops();
6230 	if (kvmppc_radix_possible())
6231 		kvmppc_radix_exit();
6232 	kvmppc_hv_ops = NULL;
6233 	kvmhv_nested_exit();
6234 }
6235 
6236 module_init(kvmppc_book3s_init_hv);
6237 module_exit(kvmppc_book3s_exit_hv);
6238 MODULE_LICENSE("GPL");
6239 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6240 MODULE_ALIAS("devname:kvm");
6241