xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 86db9f28)
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 
46 #include <asm/ftrace.h>
47 #include <asm/reg.h>
48 #include <asm/ppc-opcode.h>
49 #include <asm/asm-prototypes.h>
50 #include <asm/archrandom.h>
51 #include <asm/debug.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <linux/uaccess.h>
56 #include <asm/io.h>
57 #include <asm/kvm_ppc.h>
58 #include <asm/kvm_book3s.h>
59 #include <asm/mmu_context.h>
60 #include <asm/lppaca.h>
61 #include <asm/processor.h>
62 #include <asm/cputhreads.h>
63 #include <asm/page.h>
64 #include <asm/hvcall.h>
65 #include <asm/switch_to.h>
66 #include <asm/smp.h>
67 #include <asm/dbell.h>
68 #include <asm/hmi.h>
69 #include <asm/pnv-pci.h>
70 #include <asm/mmu.h>
71 #include <asm/opal.h>
72 #include <asm/xics.h>
73 #include <asm/xive.h>
74 #include <asm/hw_breakpoint.h>
75 
76 #include "book3s.h"
77 
78 #define CREATE_TRACE_POINTS
79 #include "trace_hv.h"
80 
81 /* #define EXIT_DEBUG */
82 /* #define EXIT_DEBUG_SIMPLE */
83 /* #define EXIT_DEBUG_INT */
84 
85 /* Used to indicate that a guest page fault needs to be handled */
86 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
87 /* Used to indicate that a guest passthrough interrupt needs to be handled */
88 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
89 
90 /* Used as a "null" value for timebase values */
91 #define TB_NIL	(~(u64)0)
92 
93 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
94 
95 static int dynamic_mt_modes = 6;
96 module_param(dynamic_mt_modes, int, 0644);
97 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
98 static int target_smt_mode;
99 module_param(target_smt_mode, int, 0644);
100 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
101 
102 static bool indep_threads_mode = true;
103 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
105 
106 static bool one_vm_per_core;
107 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
108 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
109 
110 #ifdef CONFIG_KVM_XICS
111 static struct kernel_param_ops module_param_ops = {
112 	.set = param_set_int,
113 	.get = param_get_int,
114 };
115 
116 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
117 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
118 
119 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
120 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
121 #endif
122 
123 /* If set, guests are allowed to create and control nested guests */
124 static bool nested = true;
125 module_param(nested, bool, S_IRUGO | S_IWUSR);
126 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
127 
128 static inline bool nesting_enabled(struct kvm *kvm)
129 {
130 	return kvm->arch.nested_enable && kvm_is_radix(kvm);
131 }
132 
133 /* If set, the threads on each CPU core have to be in the same MMU mode */
134 static bool no_mixing_hpt_and_radix;
135 
136 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
137 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
138 
139 /*
140  * RWMR values for POWER8.  These control the rate at which PURR
141  * and SPURR count and should be set according to the number of
142  * online threads in the vcore being run.
143  */
144 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
145 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
146 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
147 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
148 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
149 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
150 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
151 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
152 
153 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
154 	RWMR_RPA_P8_1THREAD,
155 	RWMR_RPA_P8_1THREAD,
156 	RWMR_RPA_P8_2THREAD,
157 	RWMR_RPA_P8_3THREAD,
158 	RWMR_RPA_P8_4THREAD,
159 	RWMR_RPA_P8_5THREAD,
160 	RWMR_RPA_P8_6THREAD,
161 	RWMR_RPA_P8_7THREAD,
162 	RWMR_RPA_P8_8THREAD,
163 };
164 
165 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
166 		int *ip)
167 {
168 	int i = *ip;
169 	struct kvm_vcpu *vcpu;
170 
171 	while (++i < MAX_SMT_THREADS) {
172 		vcpu = READ_ONCE(vc->runnable_threads[i]);
173 		if (vcpu) {
174 			*ip = i;
175 			return vcpu;
176 		}
177 	}
178 	return NULL;
179 }
180 
181 /* Used to traverse the list of runnable threads for a given vcore */
182 #define for_each_runnable_thread(i, vcpu, vc) \
183 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
184 
185 static bool kvmppc_ipi_thread(int cpu)
186 {
187 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
188 
189 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
190 	if (kvmhv_on_pseries())
191 		return false;
192 
193 	/* On POWER9 we can use msgsnd to IPI any cpu */
194 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
195 		msg |= get_hard_smp_processor_id(cpu);
196 		smp_mb();
197 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
198 		return true;
199 	}
200 
201 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
202 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
203 		preempt_disable();
204 		if (cpu_first_thread_sibling(cpu) ==
205 		    cpu_first_thread_sibling(smp_processor_id())) {
206 			msg |= cpu_thread_in_core(cpu);
207 			smp_mb();
208 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
209 			preempt_enable();
210 			return true;
211 		}
212 		preempt_enable();
213 	}
214 
215 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
216 	if (cpu >= 0 && cpu < nr_cpu_ids) {
217 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
218 			xics_wake_cpu(cpu);
219 			return true;
220 		}
221 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
222 		return true;
223 	}
224 #endif
225 
226 	return false;
227 }
228 
229 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
230 {
231 	int cpu;
232 	struct swait_queue_head *wqp;
233 
234 	wqp = kvm_arch_vcpu_wq(vcpu);
235 	if (swq_has_sleeper(wqp)) {
236 		swake_up_one(wqp);
237 		++vcpu->stat.halt_wakeup;
238 	}
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)
284 {
285 	unsigned long flags;
286 
287 	spin_lock_irqsave(&vc->stoltb_lock, flags);
288 	vc->preempt_tb = mftb();
289 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
290 }
291 
292 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
293 {
294 	unsigned long flags;
295 
296 	spin_lock_irqsave(&vc->stoltb_lock, flags);
297 	if (vc->preempt_tb != TB_NIL) {
298 		vc->stolen_tb += mftb() - vc->preempt_tb;
299 		vc->preempt_tb = TB_NIL;
300 	}
301 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
302 }
303 
304 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
305 {
306 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
307 	unsigned long flags;
308 
309 	/*
310 	 * We can test vc->runner without taking the vcore lock,
311 	 * because only this task ever sets vc->runner to this
312 	 * vcpu, and once it is set to this vcpu, only this task
313 	 * ever sets it to NULL.
314 	 */
315 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
316 		kvmppc_core_end_stolen(vc);
317 
318 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
319 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
320 	    vcpu->arch.busy_preempt != TB_NIL) {
321 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
322 		vcpu->arch.busy_preempt = TB_NIL;
323 	}
324 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
325 }
326 
327 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
328 {
329 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
330 	unsigned long flags;
331 
332 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
333 		kvmppc_core_start_stolen(vc);
334 
335 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
336 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
337 		vcpu->arch.busy_preempt = mftb();
338 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
339 }
340 
341 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
342 {
343 	/*
344 	 * Check for illegal transactional state bit combination
345 	 * and if we find it, force the TS field to a safe state.
346 	 */
347 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
348 		msr &= ~MSR_TS_MASK;
349 	vcpu->arch.shregs.msr = msr;
350 	kvmppc_end_cede(vcpu);
351 }
352 
353 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
354 {
355 	vcpu->arch.pvr = pvr;
356 }
357 
358 /* Dummy value used in computing PCR value below */
359 #define PCR_ARCH_300	(PCR_ARCH_207 << 1)
360 
361 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
362 {
363 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
364 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
365 
366 	/* We can (emulate) our own architecture version and anything older */
367 	if (cpu_has_feature(CPU_FTR_ARCH_300))
368 		host_pcr_bit = PCR_ARCH_300;
369 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
370 		host_pcr_bit = PCR_ARCH_207;
371 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
372 		host_pcr_bit = PCR_ARCH_206;
373 	else
374 		host_pcr_bit = PCR_ARCH_205;
375 
376 	/* Determine lowest PCR bit needed to run guest in given PVR level */
377 	guest_pcr_bit = host_pcr_bit;
378 	if (arch_compat) {
379 		switch (arch_compat) {
380 		case PVR_ARCH_205:
381 			guest_pcr_bit = PCR_ARCH_205;
382 			break;
383 		case PVR_ARCH_206:
384 		case PVR_ARCH_206p:
385 			guest_pcr_bit = PCR_ARCH_206;
386 			break;
387 		case PVR_ARCH_207:
388 			guest_pcr_bit = PCR_ARCH_207;
389 			break;
390 		case PVR_ARCH_300:
391 			guest_pcr_bit = PCR_ARCH_300;
392 			break;
393 		default:
394 			return -EINVAL;
395 		}
396 	}
397 
398 	/* Check requested PCR bits don't exceed our capabilities */
399 	if (guest_pcr_bit > host_pcr_bit)
400 		return -EINVAL;
401 
402 	spin_lock(&vc->lock);
403 	vc->arch_compat = arch_compat;
404 	/*
405 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
406 	 * Also set all reserved PCR bits
407 	 */
408 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
409 	spin_unlock(&vc->lock);
410 
411 	return 0;
412 }
413 
414 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
415 {
416 	int r;
417 
418 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
419 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
420 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
421 	for (r = 0; r < 16; ++r)
422 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
423 		       r, kvmppc_get_gpr(vcpu, r),
424 		       r+16, kvmppc_get_gpr(vcpu, r+16));
425 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
426 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
427 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
428 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
429 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
430 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
431 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
432 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
433 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
434 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
435 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
436 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
437 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
438 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
439 	for (r = 0; r < vcpu->arch.slb_max; ++r)
440 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
441 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
442 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
443 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
444 	       vcpu->arch.last_inst);
445 }
446 
447 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
448 {
449 	return kvm_get_vcpu_by_id(kvm, id);
450 }
451 
452 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
453 {
454 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
455 	vpa->yield_count = cpu_to_be32(1);
456 }
457 
458 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
459 		   unsigned long addr, unsigned long len)
460 {
461 	/* check address is cacheline aligned */
462 	if (addr & (L1_CACHE_BYTES - 1))
463 		return -EINVAL;
464 	spin_lock(&vcpu->arch.vpa_update_lock);
465 	if (v->next_gpa != addr || v->len != len) {
466 		v->next_gpa = addr;
467 		v->len = addr ? len : 0;
468 		v->update_pending = 1;
469 	}
470 	spin_unlock(&vcpu->arch.vpa_update_lock);
471 	return 0;
472 }
473 
474 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
475 struct reg_vpa {
476 	u32 dummy;
477 	union {
478 		__be16 hword;
479 		__be32 word;
480 	} length;
481 };
482 
483 static int vpa_is_registered(struct kvmppc_vpa *vpap)
484 {
485 	if (vpap->update_pending)
486 		return vpap->next_gpa != 0;
487 	return vpap->pinned_addr != NULL;
488 }
489 
490 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
491 				       unsigned long flags,
492 				       unsigned long vcpuid, unsigned long vpa)
493 {
494 	struct kvm *kvm = vcpu->kvm;
495 	unsigned long len, nb;
496 	void *va;
497 	struct kvm_vcpu *tvcpu;
498 	int err;
499 	int subfunc;
500 	struct kvmppc_vpa *vpap;
501 
502 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
503 	if (!tvcpu)
504 		return H_PARAMETER;
505 
506 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
507 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
508 	    subfunc == H_VPA_REG_SLB) {
509 		/* Registering new area - address must be cache-line aligned */
510 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
511 			return H_PARAMETER;
512 
513 		/* convert logical addr to kernel addr and read length */
514 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
515 		if (va == NULL)
516 			return H_PARAMETER;
517 		if (subfunc == H_VPA_REG_VPA)
518 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
519 		else
520 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
521 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
522 
523 		/* Check length */
524 		if (len > nb || len < sizeof(struct reg_vpa))
525 			return H_PARAMETER;
526 	} else {
527 		vpa = 0;
528 		len = 0;
529 	}
530 
531 	err = H_PARAMETER;
532 	vpap = NULL;
533 	spin_lock(&tvcpu->arch.vpa_update_lock);
534 
535 	switch (subfunc) {
536 	case H_VPA_REG_VPA:		/* register VPA */
537 		/*
538 		 * The size of our lppaca is 1kB because of the way we align
539 		 * it for the guest to avoid crossing a 4kB boundary. We only
540 		 * use 640 bytes of the structure though, so we should accept
541 		 * clients that set a size of 640.
542 		 */
543 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
544 		if (len < sizeof(struct lppaca))
545 			break;
546 		vpap = &tvcpu->arch.vpa;
547 		err = 0;
548 		break;
549 
550 	case H_VPA_REG_DTL:		/* register DTL */
551 		if (len < sizeof(struct dtl_entry))
552 			break;
553 		len -= len % sizeof(struct dtl_entry);
554 
555 		/* Check that they have previously registered a VPA */
556 		err = H_RESOURCE;
557 		if (!vpa_is_registered(&tvcpu->arch.vpa))
558 			break;
559 
560 		vpap = &tvcpu->arch.dtl;
561 		err = 0;
562 		break;
563 
564 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
565 		/* Check that they have previously registered a VPA */
566 		err = H_RESOURCE;
567 		if (!vpa_is_registered(&tvcpu->arch.vpa))
568 			break;
569 
570 		vpap = &tvcpu->arch.slb_shadow;
571 		err = 0;
572 		break;
573 
574 	case H_VPA_DEREG_VPA:		/* deregister VPA */
575 		/* Check they don't still have a DTL or SLB buf registered */
576 		err = H_RESOURCE;
577 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
578 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
579 			break;
580 
581 		vpap = &tvcpu->arch.vpa;
582 		err = 0;
583 		break;
584 
585 	case H_VPA_DEREG_DTL:		/* deregister DTL */
586 		vpap = &tvcpu->arch.dtl;
587 		err = 0;
588 		break;
589 
590 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
591 		vpap = &tvcpu->arch.slb_shadow;
592 		err = 0;
593 		break;
594 	}
595 
596 	if (vpap) {
597 		vpap->next_gpa = vpa;
598 		vpap->len = len;
599 		vpap->update_pending = 1;
600 	}
601 
602 	spin_unlock(&tvcpu->arch.vpa_update_lock);
603 
604 	return err;
605 }
606 
607 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
608 {
609 	struct kvm *kvm = vcpu->kvm;
610 	void *va;
611 	unsigned long nb;
612 	unsigned long gpa;
613 
614 	/*
615 	 * We need to pin the page pointed to by vpap->next_gpa,
616 	 * but we can't call kvmppc_pin_guest_page under the lock
617 	 * as it does get_user_pages() and down_read().  So we
618 	 * have to drop the lock, pin the page, then get the lock
619 	 * again and check that a new area didn't get registered
620 	 * in the meantime.
621 	 */
622 	for (;;) {
623 		gpa = vpap->next_gpa;
624 		spin_unlock(&vcpu->arch.vpa_update_lock);
625 		va = NULL;
626 		nb = 0;
627 		if (gpa)
628 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
629 		spin_lock(&vcpu->arch.vpa_update_lock);
630 		if (gpa == vpap->next_gpa)
631 			break;
632 		/* sigh... unpin that one and try again */
633 		if (va)
634 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
635 	}
636 
637 	vpap->update_pending = 0;
638 	if (va && nb < vpap->len) {
639 		/*
640 		 * If it's now too short, it must be that userspace
641 		 * has changed the mappings underlying guest memory,
642 		 * so unregister the region.
643 		 */
644 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
645 		va = NULL;
646 	}
647 	if (vpap->pinned_addr)
648 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
649 					vpap->dirty);
650 	vpap->gpa = gpa;
651 	vpap->pinned_addr = va;
652 	vpap->dirty = false;
653 	if (va)
654 		vpap->pinned_end = va + vpap->len;
655 }
656 
657 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
658 {
659 	if (!(vcpu->arch.vpa.update_pending ||
660 	      vcpu->arch.slb_shadow.update_pending ||
661 	      vcpu->arch.dtl.update_pending))
662 		return;
663 
664 	spin_lock(&vcpu->arch.vpa_update_lock);
665 	if (vcpu->arch.vpa.update_pending) {
666 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
667 		if (vcpu->arch.vpa.pinned_addr)
668 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
669 	}
670 	if (vcpu->arch.dtl.update_pending) {
671 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
672 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
673 		vcpu->arch.dtl_index = 0;
674 	}
675 	if (vcpu->arch.slb_shadow.update_pending)
676 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
677 	spin_unlock(&vcpu->arch.vpa_update_lock);
678 }
679 
680 /*
681  * Return the accumulated stolen time for the vcore up until `now'.
682  * The caller should hold the vcore lock.
683  */
684 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
685 {
686 	u64 p;
687 	unsigned long flags;
688 
689 	spin_lock_irqsave(&vc->stoltb_lock, flags);
690 	p = vc->stolen_tb;
691 	if (vc->vcore_state != VCORE_INACTIVE &&
692 	    vc->preempt_tb != TB_NIL)
693 		p += now - vc->preempt_tb;
694 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
695 	return p;
696 }
697 
698 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
699 				    struct kvmppc_vcore *vc)
700 {
701 	struct dtl_entry *dt;
702 	struct lppaca *vpa;
703 	unsigned long stolen;
704 	unsigned long core_stolen;
705 	u64 now;
706 	unsigned long flags;
707 
708 	dt = vcpu->arch.dtl_ptr;
709 	vpa = vcpu->arch.vpa.pinned_addr;
710 	now = mftb();
711 	core_stolen = vcore_stolen_time(vc, now);
712 	stolen = core_stolen - vcpu->arch.stolen_logged;
713 	vcpu->arch.stolen_logged = core_stolen;
714 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
715 	stolen += vcpu->arch.busy_stolen;
716 	vcpu->arch.busy_stolen = 0;
717 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
718 	if (!dt || !vpa)
719 		return;
720 	memset(dt, 0, sizeof(struct dtl_entry));
721 	dt->dispatch_reason = 7;
722 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
723 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
724 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
725 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
726 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
727 	++dt;
728 	if (dt == vcpu->arch.dtl.pinned_end)
729 		dt = vcpu->arch.dtl.pinned_addr;
730 	vcpu->arch.dtl_ptr = dt;
731 	/* order writing *dt vs. writing vpa->dtl_idx */
732 	smp_wmb();
733 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
734 	vcpu->arch.dtl.dirty = true;
735 }
736 
737 /* See if there is a doorbell interrupt pending for a vcpu */
738 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
739 {
740 	int thr;
741 	struct kvmppc_vcore *vc;
742 
743 	if (vcpu->arch.doorbell_request)
744 		return true;
745 	/*
746 	 * Ensure that the read of vcore->dpdes comes after the read
747 	 * of vcpu->doorbell_request.  This barrier matches the
748 	 * smp_wmb() in kvmppc_guest_entry_inject().
749 	 */
750 	smp_rmb();
751 	vc = vcpu->arch.vcore;
752 	thr = vcpu->vcpu_id - vc->first_vcpuid;
753 	return !!(vc->dpdes & (1 << thr));
754 }
755 
756 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
757 {
758 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
759 		return true;
760 	if ((!vcpu->arch.vcore->arch_compat) &&
761 	    cpu_has_feature(CPU_FTR_ARCH_207S))
762 		return true;
763 	return false;
764 }
765 
766 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
767 			     unsigned long resource, unsigned long value1,
768 			     unsigned long value2)
769 {
770 	switch (resource) {
771 	case H_SET_MODE_RESOURCE_SET_CIABR:
772 		if (!kvmppc_power8_compatible(vcpu))
773 			return H_P2;
774 		if (value2)
775 			return H_P4;
776 		if (mflags)
777 			return H_UNSUPPORTED_FLAG_START;
778 		/* Guests can't breakpoint the hypervisor */
779 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
780 			return H_P3;
781 		vcpu->arch.ciabr  = value1;
782 		return H_SUCCESS;
783 	case H_SET_MODE_RESOURCE_SET_DAWR:
784 		if (!kvmppc_power8_compatible(vcpu))
785 			return H_P2;
786 		if (!ppc_breakpoint_available())
787 			return H_P2;
788 		if (mflags)
789 			return H_UNSUPPORTED_FLAG_START;
790 		if (value2 & DABRX_HYP)
791 			return H_P4;
792 		vcpu->arch.dawr  = value1;
793 		vcpu->arch.dawrx = value2;
794 		return H_SUCCESS;
795 	default:
796 		return H_TOO_HARD;
797 	}
798 }
799 
800 /* Copy guest memory in place - must reside within a single memslot */
801 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
802 				  unsigned long len)
803 {
804 	struct kvm_memory_slot *to_memslot = NULL;
805 	struct kvm_memory_slot *from_memslot = NULL;
806 	unsigned long to_addr, from_addr;
807 	int r;
808 
809 	/* Get HPA for from address */
810 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
811 	if (!from_memslot)
812 		return -EFAULT;
813 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
814 			     << PAGE_SHIFT))
815 		return -EINVAL;
816 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
817 	if (kvm_is_error_hva(from_addr))
818 		return -EFAULT;
819 	from_addr |= (from & (PAGE_SIZE - 1));
820 
821 	/* Get HPA for to address */
822 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
823 	if (!to_memslot)
824 		return -EFAULT;
825 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
826 			   << PAGE_SHIFT))
827 		return -EINVAL;
828 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
829 	if (kvm_is_error_hva(to_addr))
830 		return -EFAULT;
831 	to_addr |= (to & (PAGE_SIZE - 1));
832 
833 	/* Perform copy */
834 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
835 			     len);
836 	if (r)
837 		return -EFAULT;
838 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
839 	return 0;
840 }
841 
842 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
843 			       unsigned long dest, unsigned long src)
844 {
845 	u64 pg_sz = SZ_4K;		/* 4K page size */
846 	u64 pg_mask = SZ_4K - 1;
847 	int ret;
848 
849 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
850 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
851 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
852 		return H_PARAMETER;
853 
854 	/* dest (and src if copy_page flag set) must be page aligned */
855 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
856 		return H_PARAMETER;
857 
858 	/* zero and/or copy the page as determined by the flags */
859 	if (flags & H_COPY_PAGE) {
860 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
861 		if (ret < 0)
862 			return H_PARAMETER;
863 	} else if (flags & H_ZERO_PAGE) {
864 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
865 		if (ret < 0)
866 			return H_PARAMETER;
867 	}
868 
869 	/* We can ignore the remaining flags */
870 
871 	return H_SUCCESS;
872 }
873 
874 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
875 {
876 	struct kvmppc_vcore *vcore = target->arch.vcore;
877 
878 	/*
879 	 * We expect to have been called by the real mode handler
880 	 * (kvmppc_rm_h_confer()) which would have directly returned
881 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
882 	 * have useful work to do and should not confer) so we don't
883 	 * recheck that here.
884 	 */
885 
886 	spin_lock(&vcore->lock);
887 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
888 	    vcore->vcore_state != VCORE_INACTIVE &&
889 	    vcore->runner)
890 		target = vcore->runner;
891 	spin_unlock(&vcore->lock);
892 
893 	return kvm_vcpu_yield_to(target);
894 }
895 
896 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
897 {
898 	int yield_count = 0;
899 	struct lppaca *lppaca;
900 
901 	spin_lock(&vcpu->arch.vpa_update_lock);
902 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
903 	if (lppaca)
904 		yield_count = be32_to_cpu(lppaca->yield_count);
905 	spin_unlock(&vcpu->arch.vpa_update_lock);
906 	return yield_count;
907 }
908 
909 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
910 {
911 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
912 	unsigned long target, ret = H_SUCCESS;
913 	int yield_count;
914 	struct kvm_vcpu *tvcpu;
915 	int idx, rc;
916 
917 	if (req <= MAX_HCALL_OPCODE &&
918 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
919 		return RESUME_HOST;
920 
921 	switch (req) {
922 	case H_CEDE:
923 		break;
924 	case H_PROD:
925 		target = kvmppc_get_gpr(vcpu, 4);
926 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
927 		if (!tvcpu) {
928 			ret = H_PARAMETER;
929 			break;
930 		}
931 		tvcpu->arch.prodded = 1;
932 		smp_mb();
933 		if (tvcpu->arch.ceded)
934 			kvmppc_fast_vcpu_kick_hv(tvcpu);
935 		break;
936 	case H_CONFER:
937 		target = kvmppc_get_gpr(vcpu, 4);
938 		if (target == -1)
939 			break;
940 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
941 		if (!tvcpu) {
942 			ret = H_PARAMETER;
943 			break;
944 		}
945 		yield_count = kvmppc_get_gpr(vcpu, 5);
946 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
947 			break;
948 		kvm_arch_vcpu_yield_to(tvcpu);
949 		break;
950 	case H_REGISTER_VPA:
951 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
952 					kvmppc_get_gpr(vcpu, 5),
953 					kvmppc_get_gpr(vcpu, 6));
954 		break;
955 	case H_RTAS:
956 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
957 			return RESUME_HOST;
958 
959 		idx = srcu_read_lock(&vcpu->kvm->srcu);
960 		rc = kvmppc_rtas_hcall(vcpu);
961 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
962 
963 		if (rc == -ENOENT)
964 			return RESUME_HOST;
965 		else if (rc == 0)
966 			break;
967 
968 		/* Send the error out to userspace via KVM_RUN */
969 		return rc;
970 	case H_LOGICAL_CI_LOAD:
971 		ret = kvmppc_h_logical_ci_load(vcpu);
972 		if (ret == H_TOO_HARD)
973 			return RESUME_HOST;
974 		break;
975 	case H_LOGICAL_CI_STORE:
976 		ret = kvmppc_h_logical_ci_store(vcpu);
977 		if (ret == H_TOO_HARD)
978 			return RESUME_HOST;
979 		break;
980 	case H_SET_MODE:
981 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
982 					kvmppc_get_gpr(vcpu, 5),
983 					kvmppc_get_gpr(vcpu, 6),
984 					kvmppc_get_gpr(vcpu, 7));
985 		if (ret == H_TOO_HARD)
986 			return RESUME_HOST;
987 		break;
988 	case H_XIRR:
989 	case H_CPPR:
990 	case H_EOI:
991 	case H_IPI:
992 	case H_IPOLL:
993 	case H_XIRR_X:
994 		if (kvmppc_xics_enabled(vcpu)) {
995 			if (xics_on_xive()) {
996 				ret = H_NOT_AVAILABLE;
997 				return RESUME_GUEST;
998 			}
999 			ret = kvmppc_xics_hcall(vcpu, req);
1000 			break;
1001 		}
1002 		return RESUME_HOST;
1003 	case H_SET_DABR:
1004 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1005 		break;
1006 	case H_SET_XDABR:
1007 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1008 						kvmppc_get_gpr(vcpu, 5));
1009 		break;
1010 #ifdef CONFIG_SPAPR_TCE_IOMMU
1011 	case H_GET_TCE:
1012 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1013 						kvmppc_get_gpr(vcpu, 5));
1014 		if (ret == H_TOO_HARD)
1015 			return RESUME_HOST;
1016 		break;
1017 	case H_PUT_TCE:
1018 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1019 						kvmppc_get_gpr(vcpu, 5),
1020 						kvmppc_get_gpr(vcpu, 6));
1021 		if (ret == H_TOO_HARD)
1022 			return RESUME_HOST;
1023 		break;
1024 	case H_PUT_TCE_INDIRECT:
1025 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1026 						kvmppc_get_gpr(vcpu, 5),
1027 						kvmppc_get_gpr(vcpu, 6),
1028 						kvmppc_get_gpr(vcpu, 7));
1029 		if (ret == H_TOO_HARD)
1030 			return RESUME_HOST;
1031 		break;
1032 	case H_STUFF_TCE:
1033 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1034 						kvmppc_get_gpr(vcpu, 5),
1035 						kvmppc_get_gpr(vcpu, 6),
1036 						kvmppc_get_gpr(vcpu, 7));
1037 		if (ret == H_TOO_HARD)
1038 			return RESUME_HOST;
1039 		break;
1040 #endif
1041 	case H_RANDOM:
1042 		if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1043 			ret = H_HARDWARE;
1044 		break;
1045 
1046 	case H_SET_PARTITION_TABLE:
1047 		ret = H_FUNCTION;
1048 		if (nesting_enabled(vcpu->kvm))
1049 			ret = kvmhv_set_partition_table(vcpu);
1050 		break;
1051 	case H_ENTER_NESTED:
1052 		ret = H_FUNCTION;
1053 		if (!nesting_enabled(vcpu->kvm))
1054 			break;
1055 		ret = kvmhv_enter_nested_guest(vcpu);
1056 		if (ret == H_INTERRUPT) {
1057 			kvmppc_set_gpr(vcpu, 3, 0);
1058 			vcpu->arch.hcall_needed = 0;
1059 			return -EINTR;
1060 		} else if (ret == H_TOO_HARD) {
1061 			kvmppc_set_gpr(vcpu, 3, 0);
1062 			vcpu->arch.hcall_needed = 0;
1063 			return RESUME_HOST;
1064 		}
1065 		break;
1066 	case H_TLB_INVALIDATE:
1067 		ret = H_FUNCTION;
1068 		if (nesting_enabled(vcpu->kvm))
1069 			ret = kvmhv_do_nested_tlbie(vcpu);
1070 		break;
1071 	case H_COPY_TOFROM_GUEST:
1072 		ret = H_FUNCTION;
1073 		if (nesting_enabled(vcpu->kvm))
1074 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1075 		break;
1076 	case H_PAGE_INIT:
1077 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1078 					 kvmppc_get_gpr(vcpu, 5),
1079 					 kvmppc_get_gpr(vcpu, 6));
1080 		break;
1081 	default:
1082 		return RESUME_HOST;
1083 	}
1084 	kvmppc_set_gpr(vcpu, 3, ret);
1085 	vcpu->arch.hcall_needed = 0;
1086 	return RESUME_GUEST;
1087 }
1088 
1089 /*
1090  * Handle H_CEDE in the nested virtualization case where we haven't
1091  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1092  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1093  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1094  */
1095 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1096 {
1097 	vcpu->arch.shregs.msr |= MSR_EE;
1098 	vcpu->arch.ceded = 1;
1099 	smp_mb();
1100 	if (vcpu->arch.prodded) {
1101 		vcpu->arch.prodded = 0;
1102 		smp_mb();
1103 		vcpu->arch.ceded = 0;
1104 	}
1105 }
1106 
1107 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1108 {
1109 	switch (cmd) {
1110 	case H_CEDE:
1111 	case H_PROD:
1112 	case H_CONFER:
1113 	case H_REGISTER_VPA:
1114 	case H_SET_MODE:
1115 	case H_LOGICAL_CI_LOAD:
1116 	case H_LOGICAL_CI_STORE:
1117 #ifdef CONFIG_KVM_XICS
1118 	case H_XIRR:
1119 	case H_CPPR:
1120 	case H_EOI:
1121 	case H_IPI:
1122 	case H_IPOLL:
1123 	case H_XIRR_X:
1124 #endif
1125 	case H_PAGE_INIT:
1126 		return 1;
1127 	}
1128 
1129 	/* See if it's in the real-mode table */
1130 	return kvmppc_hcall_impl_hv_realmode(cmd);
1131 }
1132 
1133 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1134 					struct kvm_vcpu *vcpu)
1135 {
1136 	u32 last_inst;
1137 
1138 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1139 					EMULATE_DONE) {
1140 		/*
1141 		 * Fetch failed, so return to guest and
1142 		 * try executing it again.
1143 		 */
1144 		return RESUME_GUEST;
1145 	}
1146 
1147 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1148 		run->exit_reason = KVM_EXIT_DEBUG;
1149 		run->debug.arch.address = kvmppc_get_pc(vcpu);
1150 		return RESUME_HOST;
1151 	} else {
1152 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1153 		return RESUME_GUEST;
1154 	}
1155 }
1156 
1157 static void do_nothing(void *x)
1158 {
1159 }
1160 
1161 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1162 {
1163 	int thr, cpu, pcpu, nthreads;
1164 	struct kvm_vcpu *v;
1165 	unsigned long dpdes;
1166 
1167 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1168 	dpdes = 0;
1169 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1170 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1171 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1172 		if (!v)
1173 			continue;
1174 		/*
1175 		 * If the vcpu is currently running on a physical cpu thread,
1176 		 * interrupt it in order to pull it out of the guest briefly,
1177 		 * which will update its vcore->dpdes value.
1178 		 */
1179 		pcpu = READ_ONCE(v->cpu);
1180 		if (pcpu >= 0)
1181 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1182 		if (kvmppc_doorbell_pending(v))
1183 			dpdes |= 1 << thr;
1184 	}
1185 	return dpdes;
1186 }
1187 
1188 /*
1189  * On POWER9, emulate doorbell-related instructions in order to
1190  * give the guest the illusion of running on a multi-threaded core.
1191  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1192  * and mfspr DPDES.
1193  */
1194 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1195 {
1196 	u32 inst, rb, thr;
1197 	unsigned long arg;
1198 	struct kvm *kvm = vcpu->kvm;
1199 	struct kvm_vcpu *tvcpu;
1200 
1201 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1202 		return RESUME_GUEST;
1203 	if (get_op(inst) != 31)
1204 		return EMULATE_FAIL;
1205 	rb = get_rb(inst);
1206 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1207 	switch (get_xop(inst)) {
1208 	case OP_31_XOP_MSGSNDP:
1209 		arg = kvmppc_get_gpr(vcpu, rb);
1210 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1211 			break;
1212 		arg &= 0x3f;
1213 		if (arg >= kvm->arch.emul_smt_mode)
1214 			break;
1215 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1216 		if (!tvcpu)
1217 			break;
1218 		if (!tvcpu->arch.doorbell_request) {
1219 			tvcpu->arch.doorbell_request = 1;
1220 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1221 		}
1222 		break;
1223 	case OP_31_XOP_MSGCLRP:
1224 		arg = kvmppc_get_gpr(vcpu, rb);
1225 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1226 			break;
1227 		vcpu->arch.vcore->dpdes = 0;
1228 		vcpu->arch.doorbell_request = 0;
1229 		break;
1230 	case OP_31_XOP_MFSPR:
1231 		switch (get_sprn(inst)) {
1232 		case SPRN_TIR:
1233 			arg = thr;
1234 			break;
1235 		case SPRN_DPDES:
1236 			arg = kvmppc_read_dpdes(vcpu);
1237 			break;
1238 		default:
1239 			return EMULATE_FAIL;
1240 		}
1241 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1242 		break;
1243 	default:
1244 		return EMULATE_FAIL;
1245 	}
1246 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1247 	return RESUME_GUEST;
1248 }
1249 
1250 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1251 				 struct task_struct *tsk)
1252 {
1253 	int r = RESUME_HOST;
1254 
1255 	vcpu->stat.sum_exits++;
1256 
1257 	/*
1258 	 * This can happen if an interrupt occurs in the last stages
1259 	 * of guest entry or the first stages of guest exit (i.e. after
1260 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1261 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1262 	 * That can happen due to a bug, or due to a machine check
1263 	 * occurring at just the wrong time.
1264 	 */
1265 	if (vcpu->arch.shregs.msr & MSR_HV) {
1266 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1267 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1268 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1269 			vcpu->arch.shregs.msr);
1270 		kvmppc_dump_regs(vcpu);
1271 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1272 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1273 		return RESUME_HOST;
1274 	}
1275 	run->exit_reason = KVM_EXIT_UNKNOWN;
1276 	run->ready_for_interrupt_injection = 1;
1277 	switch (vcpu->arch.trap) {
1278 	/* We're good on these - the host merely wanted to get our attention */
1279 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1280 		vcpu->stat.dec_exits++;
1281 		r = RESUME_GUEST;
1282 		break;
1283 	case BOOK3S_INTERRUPT_EXTERNAL:
1284 	case BOOK3S_INTERRUPT_H_DOORBELL:
1285 	case BOOK3S_INTERRUPT_H_VIRT:
1286 		vcpu->stat.ext_intr_exits++;
1287 		r = RESUME_GUEST;
1288 		break;
1289 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1290 	case BOOK3S_INTERRUPT_HMI:
1291 	case BOOK3S_INTERRUPT_PERFMON:
1292 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1293 		r = RESUME_GUEST;
1294 		break;
1295 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1296 		/* Print the MCE event to host console. */
1297 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1298 
1299 		/*
1300 		 * If the guest can do FWNMI, exit to userspace so it can
1301 		 * deliver a FWNMI to the guest.
1302 		 * Otherwise we synthesize a machine check for the guest
1303 		 * so that it knows that the machine check occurred.
1304 		 */
1305 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1306 			ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1307 			kvmppc_core_queue_machine_check(vcpu, flags);
1308 			r = RESUME_GUEST;
1309 			break;
1310 		}
1311 
1312 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1313 		run->exit_reason = KVM_EXIT_NMI;
1314 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1315 		/* Clear out the old NMI status from run->flags */
1316 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1317 		/* Now set the NMI status */
1318 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1319 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1320 		else
1321 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1322 
1323 		r = RESUME_HOST;
1324 		break;
1325 	case BOOK3S_INTERRUPT_PROGRAM:
1326 	{
1327 		ulong flags;
1328 		/*
1329 		 * Normally program interrupts are delivered directly
1330 		 * to the guest by the hardware, but we can get here
1331 		 * as a result of a hypervisor emulation interrupt
1332 		 * (e40) getting turned into a 700 by BML RTAS.
1333 		 */
1334 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1335 		kvmppc_core_queue_program(vcpu, flags);
1336 		r = RESUME_GUEST;
1337 		break;
1338 	}
1339 	case BOOK3S_INTERRUPT_SYSCALL:
1340 	{
1341 		/* hcall - punt to userspace */
1342 		int i;
1343 
1344 		/* hypercall with MSR_PR has already been handled in rmode,
1345 		 * and never reaches here.
1346 		 */
1347 
1348 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1349 		for (i = 0; i < 9; ++i)
1350 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1351 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1352 		vcpu->arch.hcall_needed = 1;
1353 		r = RESUME_HOST;
1354 		break;
1355 	}
1356 	/*
1357 	 * We get these next two if the guest accesses a page which it thinks
1358 	 * it has mapped but which is not actually present, either because
1359 	 * it is for an emulated I/O device or because the corresonding
1360 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1361 	 * have been handled already.
1362 	 */
1363 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1364 		r = RESUME_PAGE_FAULT;
1365 		break;
1366 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1367 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1368 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1369 			DSISR_SRR1_MATCH_64S;
1370 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1371 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1372 		r = RESUME_PAGE_FAULT;
1373 		break;
1374 	/*
1375 	 * This occurs if the guest executes an illegal instruction.
1376 	 * If the guest debug is disabled, generate a program interrupt
1377 	 * to the guest. If guest debug is enabled, we need to check
1378 	 * whether the instruction is a software breakpoint instruction.
1379 	 * Accordingly return to Guest or Host.
1380 	 */
1381 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1382 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1383 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1384 				swab32(vcpu->arch.emul_inst) :
1385 				vcpu->arch.emul_inst;
1386 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1387 			r = kvmppc_emulate_debug_inst(run, vcpu);
1388 		} else {
1389 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1390 			r = RESUME_GUEST;
1391 		}
1392 		break;
1393 	/*
1394 	 * This occurs if the guest (kernel or userspace), does something that
1395 	 * is prohibited by HFSCR.
1396 	 * On POWER9, this could be a doorbell instruction that we need
1397 	 * to emulate.
1398 	 * Otherwise, we just generate a program interrupt to the guest.
1399 	 */
1400 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1401 		r = EMULATE_FAIL;
1402 		if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1403 		    cpu_has_feature(CPU_FTR_ARCH_300))
1404 			r = kvmppc_emulate_doorbell_instr(vcpu);
1405 		if (r == EMULATE_FAIL) {
1406 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1407 			r = RESUME_GUEST;
1408 		}
1409 		break;
1410 
1411 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1412 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1413 		/*
1414 		 * This occurs for various TM-related instructions that
1415 		 * we need to emulate on POWER9 DD2.2.  We have already
1416 		 * handled the cases where the guest was in real-suspend
1417 		 * mode and was transitioning to transactional state.
1418 		 */
1419 		r = kvmhv_p9_tm_emulation(vcpu);
1420 		break;
1421 #endif
1422 
1423 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1424 		r = RESUME_PASSTHROUGH;
1425 		break;
1426 	default:
1427 		kvmppc_dump_regs(vcpu);
1428 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1429 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1430 			vcpu->arch.shregs.msr);
1431 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1432 		r = RESUME_HOST;
1433 		break;
1434 	}
1435 
1436 	return r;
1437 }
1438 
1439 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1440 {
1441 	int r;
1442 	int srcu_idx;
1443 
1444 	vcpu->stat.sum_exits++;
1445 
1446 	/*
1447 	 * This can happen if an interrupt occurs in the last stages
1448 	 * of guest entry or the first stages of guest exit (i.e. after
1449 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1450 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1451 	 * That can happen due to a bug, or due to a machine check
1452 	 * occurring at just the wrong time.
1453 	 */
1454 	if (vcpu->arch.shregs.msr & MSR_HV) {
1455 		pr_emerg("KVM trap in HV mode while nested!\n");
1456 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1457 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1458 			 vcpu->arch.shregs.msr);
1459 		kvmppc_dump_regs(vcpu);
1460 		return RESUME_HOST;
1461 	}
1462 	switch (vcpu->arch.trap) {
1463 	/* We're good on these - the host merely wanted to get our attention */
1464 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1465 		vcpu->stat.dec_exits++;
1466 		r = RESUME_GUEST;
1467 		break;
1468 	case BOOK3S_INTERRUPT_EXTERNAL:
1469 		vcpu->stat.ext_intr_exits++;
1470 		r = RESUME_HOST;
1471 		break;
1472 	case BOOK3S_INTERRUPT_H_DOORBELL:
1473 	case BOOK3S_INTERRUPT_H_VIRT:
1474 		vcpu->stat.ext_intr_exits++;
1475 		r = RESUME_GUEST;
1476 		break;
1477 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1478 	case BOOK3S_INTERRUPT_HMI:
1479 	case BOOK3S_INTERRUPT_PERFMON:
1480 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1481 		r = RESUME_GUEST;
1482 		break;
1483 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1484 		/* Pass the machine check to the L1 guest */
1485 		r = RESUME_HOST;
1486 		/* Print the MCE event to host console. */
1487 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1488 		break;
1489 	/*
1490 	 * We get these next two if the guest accesses a page which it thinks
1491 	 * it has mapped but which is not actually present, either because
1492 	 * it is for an emulated I/O device or because the corresonding
1493 	 * host page has been paged out.
1494 	 */
1495 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1496 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1497 		r = kvmhv_nested_page_fault(run, vcpu);
1498 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1499 		break;
1500 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1501 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1502 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1503 					 DSISR_SRR1_MATCH_64S;
1504 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1505 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1506 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1507 		r = kvmhv_nested_page_fault(run, vcpu);
1508 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1509 		break;
1510 
1511 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1512 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1513 		/*
1514 		 * This occurs for various TM-related instructions that
1515 		 * we need to emulate on POWER9 DD2.2.  We have already
1516 		 * handled the cases where the guest was in real-suspend
1517 		 * mode and was transitioning to transactional state.
1518 		 */
1519 		r = kvmhv_p9_tm_emulation(vcpu);
1520 		break;
1521 #endif
1522 
1523 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1524 		vcpu->arch.trap = 0;
1525 		r = RESUME_GUEST;
1526 		if (!xics_on_xive())
1527 			kvmppc_xics_rm_complete(vcpu, 0);
1528 		break;
1529 	default:
1530 		r = RESUME_HOST;
1531 		break;
1532 	}
1533 
1534 	return r;
1535 }
1536 
1537 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1538 					    struct kvm_sregs *sregs)
1539 {
1540 	int i;
1541 
1542 	memset(sregs, 0, sizeof(struct kvm_sregs));
1543 	sregs->pvr = vcpu->arch.pvr;
1544 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1545 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1546 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1547 	}
1548 
1549 	return 0;
1550 }
1551 
1552 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1553 					    struct kvm_sregs *sregs)
1554 {
1555 	int i, j;
1556 
1557 	/* Only accept the same PVR as the host's, since we can't spoof it */
1558 	if (sregs->pvr != vcpu->arch.pvr)
1559 		return -EINVAL;
1560 
1561 	j = 0;
1562 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1563 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1564 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1565 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1566 			++j;
1567 		}
1568 	}
1569 	vcpu->arch.slb_max = j;
1570 
1571 	return 0;
1572 }
1573 
1574 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1575 		bool preserve_top32)
1576 {
1577 	struct kvm *kvm = vcpu->kvm;
1578 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1579 	u64 mask;
1580 
1581 	spin_lock(&vc->lock);
1582 	/*
1583 	 * If ILE (interrupt little-endian) has changed, update the
1584 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1585 	 */
1586 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1587 		struct kvm_vcpu *vcpu;
1588 		int i;
1589 
1590 		kvm_for_each_vcpu(i, vcpu, kvm) {
1591 			if (vcpu->arch.vcore != vc)
1592 				continue;
1593 			if (new_lpcr & LPCR_ILE)
1594 				vcpu->arch.intr_msr |= MSR_LE;
1595 			else
1596 				vcpu->arch.intr_msr &= ~MSR_LE;
1597 		}
1598 	}
1599 
1600 	/*
1601 	 * Userspace can only modify DPFD (default prefetch depth),
1602 	 * ILE (interrupt little-endian) and TC (translation control).
1603 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1604 	 */
1605 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1606 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1607 		mask |= LPCR_AIL;
1608 	/*
1609 	 * On POWER9, allow userspace to enable large decrementer for the
1610 	 * guest, whether or not the host has it enabled.
1611 	 */
1612 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1613 		mask |= LPCR_LD;
1614 
1615 	/* Broken 32-bit version of LPCR must not clear top bits */
1616 	if (preserve_top32)
1617 		mask &= 0xFFFFFFFF;
1618 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1619 	spin_unlock(&vc->lock);
1620 }
1621 
1622 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1623 				 union kvmppc_one_reg *val)
1624 {
1625 	int r = 0;
1626 	long int i;
1627 
1628 	switch (id) {
1629 	case KVM_REG_PPC_DEBUG_INST:
1630 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1631 		break;
1632 	case KVM_REG_PPC_HIOR:
1633 		*val = get_reg_val(id, 0);
1634 		break;
1635 	case KVM_REG_PPC_DABR:
1636 		*val = get_reg_val(id, vcpu->arch.dabr);
1637 		break;
1638 	case KVM_REG_PPC_DABRX:
1639 		*val = get_reg_val(id, vcpu->arch.dabrx);
1640 		break;
1641 	case KVM_REG_PPC_DSCR:
1642 		*val = get_reg_val(id, vcpu->arch.dscr);
1643 		break;
1644 	case KVM_REG_PPC_PURR:
1645 		*val = get_reg_val(id, vcpu->arch.purr);
1646 		break;
1647 	case KVM_REG_PPC_SPURR:
1648 		*val = get_reg_val(id, vcpu->arch.spurr);
1649 		break;
1650 	case KVM_REG_PPC_AMR:
1651 		*val = get_reg_val(id, vcpu->arch.amr);
1652 		break;
1653 	case KVM_REG_PPC_UAMOR:
1654 		*val = get_reg_val(id, vcpu->arch.uamor);
1655 		break;
1656 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1657 		i = id - KVM_REG_PPC_MMCR0;
1658 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1659 		break;
1660 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1661 		i = id - KVM_REG_PPC_PMC1;
1662 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1663 		break;
1664 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1665 		i = id - KVM_REG_PPC_SPMC1;
1666 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1667 		break;
1668 	case KVM_REG_PPC_SIAR:
1669 		*val = get_reg_val(id, vcpu->arch.siar);
1670 		break;
1671 	case KVM_REG_PPC_SDAR:
1672 		*val = get_reg_val(id, vcpu->arch.sdar);
1673 		break;
1674 	case KVM_REG_PPC_SIER:
1675 		*val = get_reg_val(id, vcpu->arch.sier);
1676 		break;
1677 	case KVM_REG_PPC_IAMR:
1678 		*val = get_reg_val(id, vcpu->arch.iamr);
1679 		break;
1680 	case KVM_REG_PPC_PSPB:
1681 		*val = get_reg_val(id, vcpu->arch.pspb);
1682 		break;
1683 	case KVM_REG_PPC_DPDES:
1684 		/*
1685 		 * On POWER9, where we are emulating msgsndp etc.,
1686 		 * we return 1 bit for each vcpu, which can come from
1687 		 * either vcore->dpdes or doorbell_request.
1688 		 * On POWER8, doorbell_request is 0.
1689 		 */
1690 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1691 				   vcpu->arch.doorbell_request);
1692 		break;
1693 	case KVM_REG_PPC_VTB:
1694 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1695 		break;
1696 	case KVM_REG_PPC_DAWR:
1697 		*val = get_reg_val(id, vcpu->arch.dawr);
1698 		break;
1699 	case KVM_REG_PPC_DAWRX:
1700 		*val = get_reg_val(id, vcpu->arch.dawrx);
1701 		break;
1702 	case KVM_REG_PPC_CIABR:
1703 		*val = get_reg_val(id, vcpu->arch.ciabr);
1704 		break;
1705 	case KVM_REG_PPC_CSIGR:
1706 		*val = get_reg_val(id, vcpu->arch.csigr);
1707 		break;
1708 	case KVM_REG_PPC_TACR:
1709 		*val = get_reg_val(id, vcpu->arch.tacr);
1710 		break;
1711 	case KVM_REG_PPC_TCSCR:
1712 		*val = get_reg_val(id, vcpu->arch.tcscr);
1713 		break;
1714 	case KVM_REG_PPC_PID:
1715 		*val = get_reg_val(id, vcpu->arch.pid);
1716 		break;
1717 	case KVM_REG_PPC_ACOP:
1718 		*val = get_reg_val(id, vcpu->arch.acop);
1719 		break;
1720 	case KVM_REG_PPC_WORT:
1721 		*val = get_reg_val(id, vcpu->arch.wort);
1722 		break;
1723 	case KVM_REG_PPC_TIDR:
1724 		*val = get_reg_val(id, vcpu->arch.tid);
1725 		break;
1726 	case KVM_REG_PPC_PSSCR:
1727 		*val = get_reg_val(id, vcpu->arch.psscr);
1728 		break;
1729 	case KVM_REG_PPC_VPA_ADDR:
1730 		spin_lock(&vcpu->arch.vpa_update_lock);
1731 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1732 		spin_unlock(&vcpu->arch.vpa_update_lock);
1733 		break;
1734 	case KVM_REG_PPC_VPA_SLB:
1735 		spin_lock(&vcpu->arch.vpa_update_lock);
1736 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1737 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1738 		spin_unlock(&vcpu->arch.vpa_update_lock);
1739 		break;
1740 	case KVM_REG_PPC_VPA_DTL:
1741 		spin_lock(&vcpu->arch.vpa_update_lock);
1742 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1743 		val->vpaval.length = vcpu->arch.dtl.len;
1744 		spin_unlock(&vcpu->arch.vpa_update_lock);
1745 		break;
1746 	case KVM_REG_PPC_TB_OFFSET:
1747 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1748 		break;
1749 	case KVM_REG_PPC_LPCR:
1750 	case KVM_REG_PPC_LPCR_64:
1751 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1752 		break;
1753 	case KVM_REG_PPC_PPR:
1754 		*val = get_reg_val(id, vcpu->arch.ppr);
1755 		break;
1756 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1757 	case KVM_REG_PPC_TFHAR:
1758 		*val = get_reg_val(id, vcpu->arch.tfhar);
1759 		break;
1760 	case KVM_REG_PPC_TFIAR:
1761 		*val = get_reg_val(id, vcpu->arch.tfiar);
1762 		break;
1763 	case KVM_REG_PPC_TEXASR:
1764 		*val = get_reg_val(id, vcpu->arch.texasr);
1765 		break;
1766 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1767 		i = id - KVM_REG_PPC_TM_GPR0;
1768 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1769 		break;
1770 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1771 	{
1772 		int j;
1773 		i = id - KVM_REG_PPC_TM_VSR0;
1774 		if (i < 32)
1775 			for (j = 0; j < TS_FPRWIDTH; j++)
1776 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1777 		else {
1778 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1779 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1780 			else
1781 				r = -ENXIO;
1782 		}
1783 		break;
1784 	}
1785 	case KVM_REG_PPC_TM_CR:
1786 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1787 		break;
1788 	case KVM_REG_PPC_TM_XER:
1789 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1790 		break;
1791 	case KVM_REG_PPC_TM_LR:
1792 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1793 		break;
1794 	case KVM_REG_PPC_TM_CTR:
1795 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1796 		break;
1797 	case KVM_REG_PPC_TM_FPSCR:
1798 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1799 		break;
1800 	case KVM_REG_PPC_TM_AMR:
1801 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1802 		break;
1803 	case KVM_REG_PPC_TM_PPR:
1804 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1805 		break;
1806 	case KVM_REG_PPC_TM_VRSAVE:
1807 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1808 		break;
1809 	case KVM_REG_PPC_TM_VSCR:
1810 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1811 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1812 		else
1813 			r = -ENXIO;
1814 		break;
1815 	case KVM_REG_PPC_TM_DSCR:
1816 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1817 		break;
1818 	case KVM_REG_PPC_TM_TAR:
1819 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1820 		break;
1821 #endif
1822 	case KVM_REG_PPC_ARCH_COMPAT:
1823 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1824 		break;
1825 	case KVM_REG_PPC_DEC_EXPIRY:
1826 		*val = get_reg_val(id, vcpu->arch.dec_expires +
1827 				   vcpu->arch.vcore->tb_offset);
1828 		break;
1829 	case KVM_REG_PPC_ONLINE:
1830 		*val = get_reg_val(id, vcpu->arch.online);
1831 		break;
1832 	case KVM_REG_PPC_PTCR:
1833 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1834 		break;
1835 	default:
1836 		r = -EINVAL;
1837 		break;
1838 	}
1839 
1840 	return r;
1841 }
1842 
1843 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1844 				 union kvmppc_one_reg *val)
1845 {
1846 	int r = 0;
1847 	long int i;
1848 	unsigned long addr, len;
1849 
1850 	switch (id) {
1851 	case KVM_REG_PPC_HIOR:
1852 		/* Only allow this to be set to zero */
1853 		if (set_reg_val(id, *val))
1854 			r = -EINVAL;
1855 		break;
1856 	case KVM_REG_PPC_DABR:
1857 		vcpu->arch.dabr = set_reg_val(id, *val);
1858 		break;
1859 	case KVM_REG_PPC_DABRX:
1860 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1861 		break;
1862 	case KVM_REG_PPC_DSCR:
1863 		vcpu->arch.dscr = set_reg_val(id, *val);
1864 		break;
1865 	case KVM_REG_PPC_PURR:
1866 		vcpu->arch.purr = set_reg_val(id, *val);
1867 		break;
1868 	case KVM_REG_PPC_SPURR:
1869 		vcpu->arch.spurr = set_reg_val(id, *val);
1870 		break;
1871 	case KVM_REG_PPC_AMR:
1872 		vcpu->arch.amr = set_reg_val(id, *val);
1873 		break;
1874 	case KVM_REG_PPC_UAMOR:
1875 		vcpu->arch.uamor = set_reg_val(id, *val);
1876 		break;
1877 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1878 		i = id - KVM_REG_PPC_MMCR0;
1879 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1880 		break;
1881 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1882 		i = id - KVM_REG_PPC_PMC1;
1883 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1884 		break;
1885 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1886 		i = id - KVM_REG_PPC_SPMC1;
1887 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1888 		break;
1889 	case KVM_REG_PPC_SIAR:
1890 		vcpu->arch.siar = set_reg_val(id, *val);
1891 		break;
1892 	case KVM_REG_PPC_SDAR:
1893 		vcpu->arch.sdar = set_reg_val(id, *val);
1894 		break;
1895 	case KVM_REG_PPC_SIER:
1896 		vcpu->arch.sier = set_reg_val(id, *val);
1897 		break;
1898 	case KVM_REG_PPC_IAMR:
1899 		vcpu->arch.iamr = set_reg_val(id, *val);
1900 		break;
1901 	case KVM_REG_PPC_PSPB:
1902 		vcpu->arch.pspb = set_reg_val(id, *val);
1903 		break;
1904 	case KVM_REG_PPC_DPDES:
1905 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1906 		break;
1907 	case KVM_REG_PPC_VTB:
1908 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1909 		break;
1910 	case KVM_REG_PPC_DAWR:
1911 		vcpu->arch.dawr = set_reg_val(id, *val);
1912 		break;
1913 	case KVM_REG_PPC_DAWRX:
1914 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1915 		break;
1916 	case KVM_REG_PPC_CIABR:
1917 		vcpu->arch.ciabr = set_reg_val(id, *val);
1918 		/* Don't allow setting breakpoints in hypervisor code */
1919 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1920 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1921 		break;
1922 	case KVM_REG_PPC_CSIGR:
1923 		vcpu->arch.csigr = set_reg_val(id, *val);
1924 		break;
1925 	case KVM_REG_PPC_TACR:
1926 		vcpu->arch.tacr = set_reg_val(id, *val);
1927 		break;
1928 	case KVM_REG_PPC_TCSCR:
1929 		vcpu->arch.tcscr = set_reg_val(id, *val);
1930 		break;
1931 	case KVM_REG_PPC_PID:
1932 		vcpu->arch.pid = set_reg_val(id, *val);
1933 		break;
1934 	case KVM_REG_PPC_ACOP:
1935 		vcpu->arch.acop = set_reg_val(id, *val);
1936 		break;
1937 	case KVM_REG_PPC_WORT:
1938 		vcpu->arch.wort = set_reg_val(id, *val);
1939 		break;
1940 	case KVM_REG_PPC_TIDR:
1941 		vcpu->arch.tid = set_reg_val(id, *val);
1942 		break;
1943 	case KVM_REG_PPC_PSSCR:
1944 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1945 		break;
1946 	case KVM_REG_PPC_VPA_ADDR:
1947 		addr = set_reg_val(id, *val);
1948 		r = -EINVAL;
1949 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1950 			      vcpu->arch.dtl.next_gpa))
1951 			break;
1952 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1953 		break;
1954 	case KVM_REG_PPC_VPA_SLB:
1955 		addr = val->vpaval.addr;
1956 		len = val->vpaval.length;
1957 		r = -EINVAL;
1958 		if (addr && !vcpu->arch.vpa.next_gpa)
1959 			break;
1960 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1961 		break;
1962 	case KVM_REG_PPC_VPA_DTL:
1963 		addr = val->vpaval.addr;
1964 		len = val->vpaval.length;
1965 		r = -EINVAL;
1966 		if (addr && (len < sizeof(struct dtl_entry) ||
1967 			     !vcpu->arch.vpa.next_gpa))
1968 			break;
1969 		len -= len % sizeof(struct dtl_entry);
1970 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1971 		break;
1972 	case KVM_REG_PPC_TB_OFFSET:
1973 		/* round up to multiple of 2^24 */
1974 		vcpu->arch.vcore->tb_offset =
1975 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1976 		break;
1977 	case KVM_REG_PPC_LPCR:
1978 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1979 		break;
1980 	case KVM_REG_PPC_LPCR_64:
1981 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1982 		break;
1983 	case KVM_REG_PPC_PPR:
1984 		vcpu->arch.ppr = set_reg_val(id, *val);
1985 		break;
1986 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1987 	case KVM_REG_PPC_TFHAR:
1988 		vcpu->arch.tfhar = set_reg_val(id, *val);
1989 		break;
1990 	case KVM_REG_PPC_TFIAR:
1991 		vcpu->arch.tfiar = set_reg_val(id, *val);
1992 		break;
1993 	case KVM_REG_PPC_TEXASR:
1994 		vcpu->arch.texasr = set_reg_val(id, *val);
1995 		break;
1996 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1997 		i = id - KVM_REG_PPC_TM_GPR0;
1998 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1999 		break;
2000 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2001 	{
2002 		int j;
2003 		i = id - KVM_REG_PPC_TM_VSR0;
2004 		if (i < 32)
2005 			for (j = 0; j < TS_FPRWIDTH; j++)
2006 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2007 		else
2008 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2009 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2010 			else
2011 				r = -ENXIO;
2012 		break;
2013 	}
2014 	case KVM_REG_PPC_TM_CR:
2015 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2016 		break;
2017 	case KVM_REG_PPC_TM_XER:
2018 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2019 		break;
2020 	case KVM_REG_PPC_TM_LR:
2021 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2022 		break;
2023 	case KVM_REG_PPC_TM_CTR:
2024 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2025 		break;
2026 	case KVM_REG_PPC_TM_FPSCR:
2027 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2028 		break;
2029 	case KVM_REG_PPC_TM_AMR:
2030 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2031 		break;
2032 	case KVM_REG_PPC_TM_PPR:
2033 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2034 		break;
2035 	case KVM_REG_PPC_TM_VRSAVE:
2036 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2037 		break;
2038 	case KVM_REG_PPC_TM_VSCR:
2039 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2040 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2041 		else
2042 			r = - ENXIO;
2043 		break;
2044 	case KVM_REG_PPC_TM_DSCR:
2045 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2046 		break;
2047 	case KVM_REG_PPC_TM_TAR:
2048 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2049 		break;
2050 #endif
2051 	case KVM_REG_PPC_ARCH_COMPAT:
2052 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2053 		break;
2054 	case KVM_REG_PPC_DEC_EXPIRY:
2055 		vcpu->arch.dec_expires = set_reg_val(id, *val) -
2056 			vcpu->arch.vcore->tb_offset;
2057 		break;
2058 	case KVM_REG_PPC_ONLINE:
2059 		i = set_reg_val(id, *val);
2060 		if (i && !vcpu->arch.online)
2061 			atomic_inc(&vcpu->arch.vcore->online_count);
2062 		else if (!i && vcpu->arch.online)
2063 			atomic_dec(&vcpu->arch.vcore->online_count);
2064 		vcpu->arch.online = i;
2065 		break;
2066 	case KVM_REG_PPC_PTCR:
2067 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2068 		break;
2069 	default:
2070 		r = -EINVAL;
2071 		break;
2072 	}
2073 
2074 	return r;
2075 }
2076 
2077 /*
2078  * On POWER9, threads are independent and can be in different partitions.
2079  * Therefore we consider each thread to be a subcore.
2080  * There is a restriction that all threads have to be in the same
2081  * MMU mode (radix or HPT), unfortunately, but since we only support
2082  * HPT guests on a HPT host so far, that isn't an impediment yet.
2083  */
2084 static int threads_per_vcore(struct kvm *kvm)
2085 {
2086 	if (kvm->arch.threads_indep)
2087 		return 1;
2088 	return threads_per_subcore;
2089 }
2090 
2091 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2092 {
2093 	struct kvmppc_vcore *vcore;
2094 
2095 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2096 
2097 	if (vcore == NULL)
2098 		return NULL;
2099 
2100 	spin_lock_init(&vcore->lock);
2101 	spin_lock_init(&vcore->stoltb_lock);
2102 	init_swait_queue_head(&vcore->wq);
2103 	vcore->preempt_tb = TB_NIL;
2104 	vcore->lpcr = kvm->arch.lpcr;
2105 	vcore->first_vcpuid = id;
2106 	vcore->kvm = kvm;
2107 	INIT_LIST_HEAD(&vcore->preempt_list);
2108 
2109 	return vcore;
2110 }
2111 
2112 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2113 static struct debugfs_timings_element {
2114 	const char *name;
2115 	size_t offset;
2116 } timings[] = {
2117 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2118 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2119 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2120 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2121 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2122 };
2123 
2124 #define N_TIMINGS	(ARRAY_SIZE(timings))
2125 
2126 struct debugfs_timings_state {
2127 	struct kvm_vcpu	*vcpu;
2128 	unsigned int	buflen;
2129 	char		buf[N_TIMINGS * 100];
2130 };
2131 
2132 static int debugfs_timings_open(struct inode *inode, struct file *file)
2133 {
2134 	struct kvm_vcpu *vcpu = inode->i_private;
2135 	struct debugfs_timings_state *p;
2136 
2137 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2138 	if (!p)
2139 		return -ENOMEM;
2140 
2141 	kvm_get_kvm(vcpu->kvm);
2142 	p->vcpu = vcpu;
2143 	file->private_data = p;
2144 
2145 	return nonseekable_open(inode, file);
2146 }
2147 
2148 static int debugfs_timings_release(struct inode *inode, struct file *file)
2149 {
2150 	struct debugfs_timings_state *p = file->private_data;
2151 
2152 	kvm_put_kvm(p->vcpu->kvm);
2153 	kfree(p);
2154 	return 0;
2155 }
2156 
2157 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2158 				    size_t len, loff_t *ppos)
2159 {
2160 	struct debugfs_timings_state *p = file->private_data;
2161 	struct kvm_vcpu *vcpu = p->vcpu;
2162 	char *s, *buf_end;
2163 	struct kvmhv_tb_accumulator tb;
2164 	u64 count;
2165 	loff_t pos;
2166 	ssize_t n;
2167 	int i, loops;
2168 	bool ok;
2169 
2170 	if (!p->buflen) {
2171 		s = p->buf;
2172 		buf_end = s + sizeof(p->buf);
2173 		for (i = 0; i < N_TIMINGS; ++i) {
2174 			struct kvmhv_tb_accumulator *acc;
2175 
2176 			acc = (struct kvmhv_tb_accumulator *)
2177 				((unsigned long)vcpu + timings[i].offset);
2178 			ok = false;
2179 			for (loops = 0; loops < 1000; ++loops) {
2180 				count = acc->seqcount;
2181 				if (!(count & 1)) {
2182 					smp_rmb();
2183 					tb = *acc;
2184 					smp_rmb();
2185 					if (count == acc->seqcount) {
2186 						ok = true;
2187 						break;
2188 					}
2189 				}
2190 				udelay(1);
2191 			}
2192 			if (!ok)
2193 				snprintf(s, buf_end - s, "%s: stuck\n",
2194 					timings[i].name);
2195 			else
2196 				snprintf(s, buf_end - s,
2197 					"%s: %llu %llu %llu %llu\n",
2198 					timings[i].name, count / 2,
2199 					tb_to_ns(tb.tb_total),
2200 					tb_to_ns(tb.tb_min),
2201 					tb_to_ns(tb.tb_max));
2202 			s += strlen(s);
2203 		}
2204 		p->buflen = s - p->buf;
2205 	}
2206 
2207 	pos = *ppos;
2208 	if (pos >= p->buflen)
2209 		return 0;
2210 	if (len > p->buflen - pos)
2211 		len = p->buflen - pos;
2212 	n = copy_to_user(buf, p->buf + pos, len);
2213 	if (n) {
2214 		if (n == len)
2215 			return -EFAULT;
2216 		len -= n;
2217 	}
2218 	*ppos = pos + len;
2219 	return len;
2220 }
2221 
2222 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2223 				     size_t len, loff_t *ppos)
2224 {
2225 	return -EACCES;
2226 }
2227 
2228 static const struct file_operations debugfs_timings_ops = {
2229 	.owner	 = THIS_MODULE,
2230 	.open	 = debugfs_timings_open,
2231 	.release = debugfs_timings_release,
2232 	.read	 = debugfs_timings_read,
2233 	.write	 = debugfs_timings_write,
2234 	.llseek	 = generic_file_llseek,
2235 };
2236 
2237 /* Create a debugfs directory for the vcpu */
2238 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2239 {
2240 	char buf[16];
2241 	struct kvm *kvm = vcpu->kvm;
2242 
2243 	snprintf(buf, sizeof(buf), "vcpu%u", id);
2244 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2245 		return;
2246 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2247 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2248 		return;
2249 	vcpu->arch.debugfs_timings =
2250 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2251 				    vcpu, &debugfs_timings_ops);
2252 }
2253 
2254 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2255 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2256 {
2257 }
2258 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2259 
2260 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2261 						   unsigned int id)
2262 {
2263 	struct kvm_vcpu *vcpu;
2264 	int err;
2265 	int core;
2266 	struct kvmppc_vcore *vcore;
2267 
2268 	err = -ENOMEM;
2269 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2270 	if (!vcpu)
2271 		goto out;
2272 
2273 	err = kvm_vcpu_init(vcpu, kvm, id);
2274 	if (err)
2275 		goto free_vcpu;
2276 
2277 	vcpu->arch.shared = &vcpu->arch.shregs;
2278 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2279 	/*
2280 	 * The shared struct is never shared on HV,
2281 	 * so we can always use host endianness
2282 	 */
2283 #ifdef __BIG_ENDIAN__
2284 	vcpu->arch.shared_big_endian = true;
2285 #else
2286 	vcpu->arch.shared_big_endian = false;
2287 #endif
2288 #endif
2289 	vcpu->arch.mmcr[0] = MMCR0_FC;
2290 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2291 	/* default to host PVR, since we can't spoof it */
2292 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2293 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2294 	spin_lock_init(&vcpu->arch.tbacct_lock);
2295 	vcpu->arch.busy_preempt = TB_NIL;
2296 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2297 
2298 	/*
2299 	 * Set the default HFSCR for the guest from the host value.
2300 	 * This value is only used on POWER9.
2301 	 * On POWER9, we want to virtualize the doorbell facility, so we
2302 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2303 	 * to trap and then we emulate them.
2304 	 */
2305 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2306 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2307 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2308 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2309 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2310 			vcpu->arch.hfscr |= HFSCR_TM;
2311 	}
2312 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2313 		vcpu->arch.hfscr |= HFSCR_TM;
2314 
2315 	kvmppc_mmu_book3s_hv_init(vcpu);
2316 
2317 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2318 
2319 	init_waitqueue_head(&vcpu->arch.cpu_run);
2320 
2321 	mutex_lock(&kvm->lock);
2322 	vcore = NULL;
2323 	err = -EINVAL;
2324 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2325 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2326 			pr_devel("KVM: VCPU ID too high\n");
2327 			core = KVM_MAX_VCORES;
2328 		} else {
2329 			BUG_ON(kvm->arch.smt_mode != 1);
2330 			core = kvmppc_pack_vcpu_id(kvm, id);
2331 		}
2332 	} else {
2333 		core = id / kvm->arch.smt_mode;
2334 	}
2335 	if (core < KVM_MAX_VCORES) {
2336 		vcore = kvm->arch.vcores[core];
2337 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2338 			pr_devel("KVM: collision on id %u", id);
2339 			vcore = NULL;
2340 		} else if (!vcore) {
2341 			/*
2342 			 * Take mmu_setup_lock for mutual exclusion
2343 			 * with kvmppc_update_lpcr().
2344 			 */
2345 			err = -ENOMEM;
2346 			vcore = kvmppc_vcore_create(kvm,
2347 					id & ~(kvm->arch.smt_mode - 1));
2348 			mutex_lock(&kvm->arch.mmu_setup_lock);
2349 			kvm->arch.vcores[core] = vcore;
2350 			kvm->arch.online_vcores++;
2351 			mutex_unlock(&kvm->arch.mmu_setup_lock);
2352 		}
2353 	}
2354 	mutex_unlock(&kvm->lock);
2355 
2356 	if (!vcore)
2357 		goto free_vcpu;
2358 
2359 	spin_lock(&vcore->lock);
2360 	++vcore->num_threads;
2361 	spin_unlock(&vcore->lock);
2362 	vcpu->arch.vcore = vcore;
2363 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2364 	vcpu->arch.thread_cpu = -1;
2365 	vcpu->arch.prev_cpu = -1;
2366 
2367 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2368 	kvmppc_sanity_check(vcpu);
2369 
2370 	debugfs_vcpu_init(vcpu, id);
2371 
2372 	return vcpu;
2373 
2374 free_vcpu:
2375 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2376 out:
2377 	return ERR_PTR(err);
2378 }
2379 
2380 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2381 			      unsigned long flags)
2382 {
2383 	int err;
2384 	int esmt = 0;
2385 
2386 	if (flags)
2387 		return -EINVAL;
2388 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2389 		return -EINVAL;
2390 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2391 		/*
2392 		 * On POWER8 (or POWER7), the threading mode is "strict",
2393 		 * so we pack smt_mode vcpus per vcore.
2394 		 */
2395 		if (smt_mode > threads_per_subcore)
2396 			return -EINVAL;
2397 	} else {
2398 		/*
2399 		 * On POWER9, the threading mode is "loose",
2400 		 * so each vcpu gets its own vcore.
2401 		 */
2402 		esmt = smt_mode;
2403 		smt_mode = 1;
2404 	}
2405 	mutex_lock(&kvm->lock);
2406 	err = -EBUSY;
2407 	if (!kvm->arch.online_vcores) {
2408 		kvm->arch.smt_mode = smt_mode;
2409 		kvm->arch.emul_smt_mode = esmt;
2410 		err = 0;
2411 	}
2412 	mutex_unlock(&kvm->lock);
2413 
2414 	return err;
2415 }
2416 
2417 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2418 {
2419 	if (vpa->pinned_addr)
2420 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2421 					vpa->dirty);
2422 }
2423 
2424 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2425 {
2426 	spin_lock(&vcpu->arch.vpa_update_lock);
2427 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2428 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2429 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2430 	spin_unlock(&vcpu->arch.vpa_update_lock);
2431 	kvm_vcpu_uninit(vcpu);
2432 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2433 }
2434 
2435 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2436 {
2437 	/* Indicate we want to get back into the guest */
2438 	return 1;
2439 }
2440 
2441 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2442 {
2443 	unsigned long dec_nsec, now;
2444 
2445 	now = get_tb();
2446 	if (now > vcpu->arch.dec_expires) {
2447 		/* decrementer has already gone negative */
2448 		kvmppc_core_queue_dec(vcpu);
2449 		kvmppc_core_prepare_to_enter(vcpu);
2450 		return;
2451 	}
2452 	dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2453 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2454 	vcpu->arch.timer_running = 1;
2455 }
2456 
2457 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2458 {
2459 	vcpu->arch.ceded = 0;
2460 	if (vcpu->arch.timer_running) {
2461 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2462 		vcpu->arch.timer_running = 0;
2463 	}
2464 }
2465 
2466 extern int __kvmppc_vcore_entry(void);
2467 
2468 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2469 				   struct kvm_vcpu *vcpu)
2470 {
2471 	u64 now;
2472 
2473 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2474 		return;
2475 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2476 	now = mftb();
2477 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2478 		vcpu->arch.stolen_logged;
2479 	vcpu->arch.busy_preempt = now;
2480 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2481 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2482 	--vc->n_runnable;
2483 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2484 }
2485 
2486 static int kvmppc_grab_hwthread(int cpu)
2487 {
2488 	struct paca_struct *tpaca;
2489 	long timeout = 10000;
2490 
2491 	tpaca = paca_ptrs[cpu];
2492 
2493 	/* Ensure the thread won't go into the kernel if it wakes */
2494 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2495 	tpaca->kvm_hstate.kvm_vcore = NULL;
2496 	tpaca->kvm_hstate.napping = 0;
2497 	smp_wmb();
2498 	tpaca->kvm_hstate.hwthread_req = 1;
2499 
2500 	/*
2501 	 * If the thread is already executing in the kernel (e.g. handling
2502 	 * a stray interrupt), wait for it to get back to nap mode.
2503 	 * The smp_mb() is to ensure that our setting of hwthread_req
2504 	 * is visible before we look at hwthread_state, so if this
2505 	 * races with the code at system_reset_pSeries and the thread
2506 	 * misses our setting of hwthread_req, we are sure to see its
2507 	 * setting of hwthread_state, and vice versa.
2508 	 */
2509 	smp_mb();
2510 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2511 		if (--timeout <= 0) {
2512 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2513 			return -EBUSY;
2514 		}
2515 		udelay(1);
2516 	}
2517 	return 0;
2518 }
2519 
2520 static void kvmppc_release_hwthread(int cpu)
2521 {
2522 	struct paca_struct *tpaca;
2523 
2524 	tpaca = paca_ptrs[cpu];
2525 	tpaca->kvm_hstate.hwthread_req = 0;
2526 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2527 	tpaca->kvm_hstate.kvm_vcore = NULL;
2528 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2529 }
2530 
2531 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2532 {
2533 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2534 	cpumask_t *cpu_in_guest;
2535 	int i;
2536 
2537 	cpu = cpu_first_thread_sibling(cpu);
2538 	if (nested) {
2539 		cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2540 		cpu_in_guest = &nested->cpu_in_guest;
2541 	} else {
2542 		cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2543 		cpu_in_guest = &kvm->arch.cpu_in_guest;
2544 	}
2545 	/*
2546 	 * Make sure setting of bit in need_tlb_flush precedes
2547 	 * testing of cpu_in_guest bits.  The matching barrier on
2548 	 * the other side is the first smp_mb() in kvmppc_run_core().
2549 	 */
2550 	smp_mb();
2551 	for (i = 0; i < threads_per_core; ++i)
2552 		if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2553 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2554 }
2555 
2556 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2557 {
2558 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2559 	struct kvm *kvm = vcpu->kvm;
2560 	int prev_cpu;
2561 
2562 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2563 		return;
2564 
2565 	if (nested)
2566 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2567 	else
2568 		prev_cpu = vcpu->arch.prev_cpu;
2569 
2570 	/*
2571 	 * With radix, the guest can do TLB invalidations itself,
2572 	 * and it could choose to use the local form (tlbiel) if
2573 	 * it is invalidating a translation that has only ever been
2574 	 * used on one vcpu.  However, that doesn't mean it has
2575 	 * only ever been used on one physical cpu, since vcpus
2576 	 * can move around between pcpus.  To cope with this, when
2577 	 * a vcpu moves from one pcpu to another, we need to tell
2578 	 * any vcpus running on the same core as this vcpu previously
2579 	 * ran to flush the TLB.  The TLB is shared between threads,
2580 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2581 	 */
2582 	if (prev_cpu != pcpu) {
2583 		if (prev_cpu >= 0 &&
2584 		    cpu_first_thread_sibling(prev_cpu) !=
2585 		    cpu_first_thread_sibling(pcpu))
2586 			radix_flush_cpu(kvm, prev_cpu, vcpu);
2587 		if (nested)
2588 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2589 		else
2590 			vcpu->arch.prev_cpu = pcpu;
2591 	}
2592 }
2593 
2594 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2595 {
2596 	int cpu;
2597 	struct paca_struct *tpaca;
2598 	struct kvm *kvm = vc->kvm;
2599 
2600 	cpu = vc->pcpu;
2601 	if (vcpu) {
2602 		if (vcpu->arch.timer_running) {
2603 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2604 			vcpu->arch.timer_running = 0;
2605 		}
2606 		cpu += vcpu->arch.ptid;
2607 		vcpu->cpu = vc->pcpu;
2608 		vcpu->arch.thread_cpu = cpu;
2609 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2610 	}
2611 	tpaca = paca_ptrs[cpu];
2612 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2613 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2614 	tpaca->kvm_hstate.fake_suspend = 0;
2615 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2616 	smp_wmb();
2617 	tpaca->kvm_hstate.kvm_vcore = vc;
2618 	if (cpu != smp_processor_id())
2619 		kvmppc_ipi_thread(cpu);
2620 }
2621 
2622 static void kvmppc_wait_for_nap(int n_threads)
2623 {
2624 	int cpu = smp_processor_id();
2625 	int i, loops;
2626 
2627 	if (n_threads <= 1)
2628 		return;
2629 	for (loops = 0; loops < 1000000; ++loops) {
2630 		/*
2631 		 * Check if all threads are finished.
2632 		 * We set the vcore pointer when starting a thread
2633 		 * and the thread clears it when finished, so we look
2634 		 * for any threads that still have a non-NULL vcore ptr.
2635 		 */
2636 		for (i = 1; i < n_threads; ++i)
2637 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2638 				break;
2639 		if (i == n_threads) {
2640 			HMT_medium();
2641 			return;
2642 		}
2643 		HMT_low();
2644 	}
2645 	HMT_medium();
2646 	for (i = 1; i < n_threads; ++i)
2647 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2648 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2649 }
2650 
2651 /*
2652  * Check that we are on thread 0 and that any other threads in
2653  * this core are off-line.  Then grab the threads so they can't
2654  * enter the kernel.
2655  */
2656 static int on_primary_thread(void)
2657 {
2658 	int cpu = smp_processor_id();
2659 	int thr;
2660 
2661 	/* Are we on a primary subcore? */
2662 	if (cpu_thread_in_subcore(cpu))
2663 		return 0;
2664 
2665 	thr = 0;
2666 	while (++thr < threads_per_subcore)
2667 		if (cpu_online(cpu + thr))
2668 			return 0;
2669 
2670 	/* Grab all hw threads so they can't go into the kernel */
2671 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2672 		if (kvmppc_grab_hwthread(cpu + thr)) {
2673 			/* Couldn't grab one; let the others go */
2674 			do {
2675 				kvmppc_release_hwthread(cpu + thr);
2676 			} while (--thr > 0);
2677 			return 0;
2678 		}
2679 	}
2680 	return 1;
2681 }
2682 
2683 /*
2684  * A list of virtual cores for each physical CPU.
2685  * These are vcores that could run but their runner VCPU tasks are
2686  * (or may be) preempted.
2687  */
2688 struct preempted_vcore_list {
2689 	struct list_head	list;
2690 	spinlock_t		lock;
2691 };
2692 
2693 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2694 
2695 static void init_vcore_lists(void)
2696 {
2697 	int cpu;
2698 
2699 	for_each_possible_cpu(cpu) {
2700 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2701 		spin_lock_init(&lp->lock);
2702 		INIT_LIST_HEAD(&lp->list);
2703 	}
2704 }
2705 
2706 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2707 {
2708 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2709 
2710 	vc->vcore_state = VCORE_PREEMPT;
2711 	vc->pcpu = smp_processor_id();
2712 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2713 		spin_lock(&lp->lock);
2714 		list_add_tail(&vc->preempt_list, &lp->list);
2715 		spin_unlock(&lp->lock);
2716 	}
2717 
2718 	/* Start accumulating stolen time */
2719 	kvmppc_core_start_stolen(vc);
2720 }
2721 
2722 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2723 {
2724 	struct preempted_vcore_list *lp;
2725 
2726 	kvmppc_core_end_stolen(vc);
2727 	if (!list_empty(&vc->preempt_list)) {
2728 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2729 		spin_lock(&lp->lock);
2730 		list_del_init(&vc->preempt_list);
2731 		spin_unlock(&lp->lock);
2732 	}
2733 	vc->vcore_state = VCORE_INACTIVE;
2734 }
2735 
2736 /*
2737  * This stores information about the virtual cores currently
2738  * assigned to a physical core.
2739  */
2740 struct core_info {
2741 	int		n_subcores;
2742 	int		max_subcore_threads;
2743 	int		total_threads;
2744 	int		subcore_threads[MAX_SUBCORES];
2745 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2746 };
2747 
2748 /*
2749  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2750  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2751  */
2752 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2753 
2754 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2755 {
2756 	memset(cip, 0, sizeof(*cip));
2757 	cip->n_subcores = 1;
2758 	cip->max_subcore_threads = vc->num_threads;
2759 	cip->total_threads = vc->num_threads;
2760 	cip->subcore_threads[0] = vc->num_threads;
2761 	cip->vc[0] = vc;
2762 }
2763 
2764 static bool subcore_config_ok(int n_subcores, int n_threads)
2765 {
2766 	/*
2767 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2768 	 * split-core mode, with one thread per subcore.
2769 	 */
2770 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2771 		return n_subcores <= 4 && n_threads == 1;
2772 
2773 	/* On POWER8, can only dynamically split if unsplit to begin with */
2774 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2775 		return false;
2776 	if (n_subcores > MAX_SUBCORES)
2777 		return false;
2778 	if (n_subcores > 1) {
2779 		if (!(dynamic_mt_modes & 2))
2780 			n_subcores = 4;
2781 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2782 			return false;
2783 	}
2784 
2785 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2786 }
2787 
2788 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2789 {
2790 	vc->entry_exit_map = 0;
2791 	vc->in_guest = 0;
2792 	vc->napping_threads = 0;
2793 	vc->conferring_threads = 0;
2794 	vc->tb_offset_applied = 0;
2795 }
2796 
2797 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2798 {
2799 	int n_threads = vc->num_threads;
2800 	int sub;
2801 
2802 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2803 		return false;
2804 
2805 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
2806 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2807 		return false;
2808 
2809 	/* Some POWER9 chips require all threads to be in the same MMU mode */
2810 	if (no_mixing_hpt_and_radix &&
2811 	    kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2812 		return false;
2813 
2814 	if (n_threads < cip->max_subcore_threads)
2815 		n_threads = cip->max_subcore_threads;
2816 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2817 		return false;
2818 	cip->max_subcore_threads = n_threads;
2819 
2820 	sub = cip->n_subcores;
2821 	++cip->n_subcores;
2822 	cip->total_threads += vc->num_threads;
2823 	cip->subcore_threads[sub] = vc->num_threads;
2824 	cip->vc[sub] = vc;
2825 	init_vcore_to_run(vc);
2826 	list_del_init(&vc->preempt_list);
2827 
2828 	return true;
2829 }
2830 
2831 /*
2832  * Work out whether it is possible to piggyback the execution of
2833  * vcore *pvc onto the execution of the other vcores described in *cip.
2834  */
2835 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2836 			  int target_threads)
2837 {
2838 	if (cip->total_threads + pvc->num_threads > target_threads)
2839 		return false;
2840 
2841 	return can_dynamic_split(pvc, cip);
2842 }
2843 
2844 static void prepare_threads(struct kvmppc_vcore *vc)
2845 {
2846 	int i;
2847 	struct kvm_vcpu *vcpu;
2848 
2849 	for_each_runnable_thread(i, vcpu, vc) {
2850 		if (signal_pending(vcpu->arch.run_task))
2851 			vcpu->arch.ret = -EINTR;
2852 		else if (vcpu->arch.vpa.update_pending ||
2853 			 vcpu->arch.slb_shadow.update_pending ||
2854 			 vcpu->arch.dtl.update_pending)
2855 			vcpu->arch.ret = RESUME_GUEST;
2856 		else
2857 			continue;
2858 		kvmppc_remove_runnable(vc, vcpu);
2859 		wake_up(&vcpu->arch.cpu_run);
2860 	}
2861 }
2862 
2863 static void collect_piggybacks(struct core_info *cip, int target_threads)
2864 {
2865 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2866 	struct kvmppc_vcore *pvc, *vcnext;
2867 
2868 	spin_lock(&lp->lock);
2869 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2870 		if (!spin_trylock(&pvc->lock))
2871 			continue;
2872 		prepare_threads(pvc);
2873 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2874 			list_del_init(&pvc->preempt_list);
2875 			if (pvc->runner == NULL) {
2876 				pvc->vcore_state = VCORE_INACTIVE;
2877 				kvmppc_core_end_stolen(pvc);
2878 			}
2879 			spin_unlock(&pvc->lock);
2880 			continue;
2881 		}
2882 		if (!can_piggyback(pvc, cip, target_threads)) {
2883 			spin_unlock(&pvc->lock);
2884 			continue;
2885 		}
2886 		kvmppc_core_end_stolen(pvc);
2887 		pvc->vcore_state = VCORE_PIGGYBACK;
2888 		if (cip->total_threads >= target_threads)
2889 			break;
2890 	}
2891 	spin_unlock(&lp->lock);
2892 }
2893 
2894 static bool recheck_signals_and_mmu(struct core_info *cip)
2895 {
2896 	int sub, i;
2897 	struct kvm_vcpu *vcpu;
2898 	struct kvmppc_vcore *vc;
2899 
2900 	for (sub = 0; sub < cip->n_subcores; ++sub) {
2901 		vc = cip->vc[sub];
2902 		if (!vc->kvm->arch.mmu_ready)
2903 			return true;
2904 		for_each_runnable_thread(i, vcpu, vc)
2905 			if (signal_pending(vcpu->arch.run_task))
2906 				return true;
2907 	}
2908 	return false;
2909 }
2910 
2911 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2912 {
2913 	int still_running = 0, i;
2914 	u64 now;
2915 	long ret;
2916 	struct kvm_vcpu *vcpu;
2917 
2918 	spin_lock(&vc->lock);
2919 	now = get_tb();
2920 	for_each_runnable_thread(i, vcpu, vc) {
2921 		/*
2922 		 * It's safe to unlock the vcore in the loop here, because
2923 		 * for_each_runnable_thread() is safe against removal of
2924 		 * the vcpu, and the vcore state is VCORE_EXITING here,
2925 		 * so any vcpus becoming runnable will have their arch.trap
2926 		 * set to zero and can't actually run in the guest.
2927 		 */
2928 		spin_unlock(&vc->lock);
2929 		/* cancel pending dec exception if dec is positive */
2930 		if (now < vcpu->arch.dec_expires &&
2931 		    kvmppc_core_pending_dec(vcpu))
2932 			kvmppc_core_dequeue_dec(vcpu);
2933 
2934 		trace_kvm_guest_exit(vcpu);
2935 
2936 		ret = RESUME_GUEST;
2937 		if (vcpu->arch.trap)
2938 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2939 						    vcpu->arch.run_task);
2940 
2941 		vcpu->arch.ret = ret;
2942 		vcpu->arch.trap = 0;
2943 
2944 		spin_lock(&vc->lock);
2945 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2946 			if (vcpu->arch.pending_exceptions)
2947 				kvmppc_core_prepare_to_enter(vcpu);
2948 			if (vcpu->arch.ceded)
2949 				kvmppc_set_timer(vcpu);
2950 			else
2951 				++still_running;
2952 		} else {
2953 			kvmppc_remove_runnable(vc, vcpu);
2954 			wake_up(&vcpu->arch.cpu_run);
2955 		}
2956 	}
2957 	if (!is_master) {
2958 		if (still_running > 0) {
2959 			kvmppc_vcore_preempt(vc);
2960 		} else if (vc->runner) {
2961 			vc->vcore_state = VCORE_PREEMPT;
2962 			kvmppc_core_start_stolen(vc);
2963 		} else {
2964 			vc->vcore_state = VCORE_INACTIVE;
2965 		}
2966 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2967 			/* make sure there's a candidate runner awake */
2968 			i = -1;
2969 			vcpu = next_runnable_thread(vc, &i);
2970 			wake_up(&vcpu->arch.cpu_run);
2971 		}
2972 	}
2973 	spin_unlock(&vc->lock);
2974 }
2975 
2976 /*
2977  * Clear core from the list of active host cores as we are about to
2978  * enter the guest. Only do this if it is the primary thread of the
2979  * core (not if a subcore) that is entering the guest.
2980  */
2981 static inline int kvmppc_clear_host_core(unsigned int cpu)
2982 {
2983 	int core;
2984 
2985 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2986 		return 0;
2987 	/*
2988 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2989 	 * later in kvmppc_start_thread and we need ensure that state is
2990 	 * visible to other CPUs only after we enter guest.
2991 	 */
2992 	core = cpu >> threads_shift;
2993 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2994 	return 0;
2995 }
2996 
2997 /*
2998  * Advertise this core as an active host core since we exited the guest
2999  * Only need to do this if it is the primary thread of the core that is
3000  * exiting.
3001  */
3002 static inline int kvmppc_set_host_core(unsigned int cpu)
3003 {
3004 	int core;
3005 
3006 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3007 		return 0;
3008 
3009 	/*
3010 	 * Memory barrier can be omitted here because we do a spin_unlock
3011 	 * immediately after this which provides the memory barrier.
3012 	 */
3013 	core = cpu >> threads_shift;
3014 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3015 	return 0;
3016 }
3017 
3018 static void set_irq_happened(int trap)
3019 {
3020 	switch (trap) {
3021 	case BOOK3S_INTERRUPT_EXTERNAL:
3022 		local_paca->irq_happened |= PACA_IRQ_EE;
3023 		break;
3024 	case BOOK3S_INTERRUPT_H_DOORBELL:
3025 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3026 		break;
3027 	case BOOK3S_INTERRUPT_HMI:
3028 		local_paca->irq_happened |= PACA_IRQ_HMI;
3029 		break;
3030 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3031 		replay_system_reset();
3032 		break;
3033 	}
3034 }
3035 
3036 /*
3037  * Run a set of guest threads on a physical core.
3038  * Called with vc->lock held.
3039  */
3040 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3041 {
3042 	struct kvm_vcpu *vcpu;
3043 	int i;
3044 	int srcu_idx;
3045 	struct core_info core_info;
3046 	struct kvmppc_vcore *pvc;
3047 	struct kvm_split_mode split_info, *sip;
3048 	int split, subcore_size, active;
3049 	int sub;
3050 	bool thr0_done;
3051 	unsigned long cmd_bit, stat_bit;
3052 	int pcpu, thr;
3053 	int target_threads;
3054 	int controlled_threads;
3055 	int trap;
3056 	bool is_power8;
3057 	bool hpt_on_radix;
3058 
3059 	/*
3060 	 * Remove from the list any threads that have a signal pending
3061 	 * or need a VPA update done
3062 	 */
3063 	prepare_threads(vc);
3064 
3065 	/* if the runner is no longer runnable, let the caller pick a new one */
3066 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3067 		return;
3068 
3069 	/*
3070 	 * Initialize *vc.
3071 	 */
3072 	init_vcore_to_run(vc);
3073 	vc->preempt_tb = TB_NIL;
3074 
3075 	/*
3076 	 * Number of threads that we will be controlling: the same as
3077 	 * the number of threads per subcore, except on POWER9,
3078 	 * where it's 1 because the threads are (mostly) independent.
3079 	 */
3080 	controlled_threads = threads_per_vcore(vc->kvm);
3081 
3082 	/*
3083 	 * Make sure we are running on primary threads, and that secondary
3084 	 * threads are offline.  Also check if the number of threads in this
3085 	 * guest are greater than the current system threads per guest.
3086 	 * On POWER9, we need to be not in independent-threads mode if
3087 	 * this is a HPT guest on a radix host machine where the
3088 	 * CPU threads may not be in different MMU modes.
3089 	 */
3090 	hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3091 		!kvm_is_radix(vc->kvm);
3092 	if (((controlled_threads > 1) &&
3093 	     ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3094 	    (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3095 		for_each_runnable_thread(i, vcpu, vc) {
3096 			vcpu->arch.ret = -EBUSY;
3097 			kvmppc_remove_runnable(vc, vcpu);
3098 			wake_up(&vcpu->arch.cpu_run);
3099 		}
3100 		goto out;
3101 	}
3102 
3103 	/*
3104 	 * See if we could run any other vcores on the physical core
3105 	 * along with this one.
3106 	 */
3107 	init_core_info(&core_info, vc);
3108 	pcpu = smp_processor_id();
3109 	target_threads = controlled_threads;
3110 	if (target_smt_mode && target_smt_mode < target_threads)
3111 		target_threads = target_smt_mode;
3112 	if (vc->num_threads < target_threads)
3113 		collect_piggybacks(&core_info, target_threads);
3114 
3115 	/*
3116 	 * On radix, arrange for TLB flushing if necessary.
3117 	 * This has to be done before disabling interrupts since
3118 	 * it uses smp_call_function().
3119 	 */
3120 	pcpu = smp_processor_id();
3121 	if (kvm_is_radix(vc->kvm)) {
3122 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3123 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3124 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3125 	}
3126 
3127 	/*
3128 	 * Hard-disable interrupts, and check resched flag and signals.
3129 	 * If we need to reschedule or deliver a signal, clean up
3130 	 * and return without going into the guest(s).
3131 	 * If the mmu_ready flag has been cleared, don't go into the
3132 	 * guest because that means a HPT resize operation is in progress.
3133 	 */
3134 	local_irq_disable();
3135 	hard_irq_disable();
3136 	if (lazy_irq_pending() || need_resched() ||
3137 	    recheck_signals_and_mmu(&core_info)) {
3138 		local_irq_enable();
3139 		vc->vcore_state = VCORE_INACTIVE;
3140 		/* Unlock all except the primary vcore */
3141 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3142 			pvc = core_info.vc[sub];
3143 			/* Put back on to the preempted vcores list */
3144 			kvmppc_vcore_preempt(pvc);
3145 			spin_unlock(&pvc->lock);
3146 		}
3147 		for (i = 0; i < controlled_threads; ++i)
3148 			kvmppc_release_hwthread(pcpu + i);
3149 		return;
3150 	}
3151 
3152 	kvmppc_clear_host_core(pcpu);
3153 
3154 	/* Decide on micro-threading (split-core) mode */
3155 	subcore_size = threads_per_subcore;
3156 	cmd_bit = stat_bit = 0;
3157 	split = core_info.n_subcores;
3158 	sip = NULL;
3159 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3160 		&& !cpu_has_feature(CPU_FTR_ARCH_300);
3161 
3162 	if (split > 1 || hpt_on_radix) {
3163 		sip = &split_info;
3164 		memset(&split_info, 0, sizeof(split_info));
3165 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3166 			split_info.vc[sub] = core_info.vc[sub];
3167 
3168 		if (is_power8) {
3169 			if (split == 2 && (dynamic_mt_modes & 2)) {
3170 				cmd_bit = HID0_POWER8_1TO2LPAR;
3171 				stat_bit = HID0_POWER8_2LPARMODE;
3172 			} else {
3173 				split = 4;
3174 				cmd_bit = HID0_POWER8_1TO4LPAR;
3175 				stat_bit = HID0_POWER8_4LPARMODE;
3176 			}
3177 			subcore_size = MAX_SMT_THREADS / split;
3178 			split_info.rpr = mfspr(SPRN_RPR);
3179 			split_info.pmmar = mfspr(SPRN_PMMAR);
3180 			split_info.ldbar = mfspr(SPRN_LDBAR);
3181 			split_info.subcore_size = subcore_size;
3182 		} else {
3183 			split_info.subcore_size = 1;
3184 			if (hpt_on_radix) {
3185 				/* Use the split_info for LPCR/LPIDR changes */
3186 				split_info.lpcr_req = vc->lpcr;
3187 				split_info.lpidr_req = vc->kvm->arch.lpid;
3188 				split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3189 				split_info.do_set = 1;
3190 			}
3191 		}
3192 
3193 		/* order writes to split_info before kvm_split_mode pointer */
3194 		smp_wmb();
3195 	}
3196 
3197 	for (thr = 0; thr < controlled_threads; ++thr) {
3198 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3199 
3200 		paca->kvm_hstate.tid = thr;
3201 		paca->kvm_hstate.napping = 0;
3202 		paca->kvm_hstate.kvm_split_mode = sip;
3203 	}
3204 
3205 	/* Initiate micro-threading (split-core) on POWER8 if required */
3206 	if (cmd_bit) {
3207 		unsigned long hid0 = mfspr(SPRN_HID0);
3208 
3209 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3210 		mb();
3211 		mtspr(SPRN_HID0, hid0);
3212 		isync();
3213 		for (;;) {
3214 			hid0 = mfspr(SPRN_HID0);
3215 			if (hid0 & stat_bit)
3216 				break;
3217 			cpu_relax();
3218 		}
3219 	}
3220 
3221 	/*
3222 	 * On POWER8, set RWMR register.
3223 	 * Since it only affects PURR and SPURR, it doesn't affect
3224 	 * the host, so we don't save/restore the host value.
3225 	 */
3226 	if (is_power8) {
3227 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3228 		int n_online = atomic_read(&vc->online_count);
3229 
3230 		/*
3231 		 * Use the 8-thread value if we're doing split-core
3232 		 * or if the vcore's online count looks bogus.
3233 		 */
3234 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3235 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3236 			rwmr_val = p8_rwmr_values[n_online];
3237 		mtspr(SPRN_RWMR, rwmr_val);
3238 	}
3239 
3240 	/* Start all the threads */
3241 	active = 0;
3242 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3243 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3244 		thr0_done = false;
3245 		active |= 1 << thr;
3246 		pvc = core_info.vc[sub];
3247 		pvc->pcpu = pcpu + thr;
3248 		for_each_runnable_thread(i, vcpu, pvc) {
3249 			kvmppc_start_thread(vcpu, pvc);
3250 			kvmppc_create_dtl_entry(vcpu, pvc);
3251 			trace_kvm_guest_enter(vcpu);
3252 			if (!vcpu->arch.ptid)
3253 				thr0_done = true;
3254 			active |= 1 << (thr + vcpu->arch.ptid);
3255 		}
3256 		/*
3257 		 * We need to start the first thread of each subcore
3258 		 * even if it doesn't have a vcpu.
3259 		 */
3260 		if (!thr0_done)
3261 			kvmppc_start_thread(NULL, pvc);
3262 	}
3263 
3264 	/*
3265 	 * Ensure that split_info.do_nap is set after setting
3266 	 * the vcore pointer in the PACA of the secondaries.
3267 	 */
3268 	smp_mb();
3269 
3270 	/*
3271 	 * When doing micro-threading, poke the inactive threads as well.
3272 	 * This gets them to the nap instruction after kvm_do_nap,
3273 	 * which reduces the time taken to unsplit later.
3274 	 * For POWER9 HPT guest on radix host, we need all the secondary
3275 	 * threads woken up so they can do the LPCR/LPIDR change.
3276 	 */
3277 	if (cmd_bit || hpt_on_radix) {
3278 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3279 		for (thr = 1; thr < threads_per_subcore; ++thr)
3280 			if (!(active & (1 << thr)))
3281 				kvmppc_ipi_thread(pcpu + thr);
3282 	}
3283 
3284 	vc->vcore_state = VCORE_RUNNING;
3285 	preempt_disable();
3286 
3287 	trace_kvmppc_run_core(vc, 0);
3288 
3289 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3290 		spin_unlock(&core_info.vc[sub]->lock);
3291 
3292 	guest_enter_irqoff();
3293 
3294 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3295 
3296 	this_cpu_disable_ftrace();
3297 
3298 	/*
3299 	 * Interrupts will be enabled once we get into the guest,
3300 	 * so tell lockdep that we're about to enable interrupts.
3301 	 */
3302 	trace_hardirqs_on();
3303 
3304 	trap = __kvmppc_vcore_entry();
3305 
3306 	trace_hardirqs_off();
3307 
3308 	this_cpu_enable_ftrace();
3309 
3310 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3311 
3312 	set_irq_happened(trap);
3313 
3314 	spin_lock(&vc->lock);
3315 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3316 	vc->vcore_state = VCORE_EXITING;
3317 
3318 	/* wait for secondary threads to finish writing their state to memory */
3319 	kvmppc_wait_for_nap(controlled_threads);
3320 
3321 	/* Return to whole-core mode if we split the core earlier */
3322 	if (cmd_bit) {
3323 		unsigned long hid0 = mfspr(SPRN_HID0);
3324 		unsigned long loops = 0;
3325 
3326 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3327 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3328 		mb();
3329 		mtspr(SPRN_HID0, hid0);
3330 		isync();
3331 		for (;;) {
3332 			hid0 = mfspr(SPRN_HID0);
3333 			if (!(hid0 & stat_bit))
3334 				break;
3335 			cpu_relax();
3336 			++loops;
3337 		}
3338 	} else if (hpt_on_radix) {
3339 		/* Wait for all threads to have seen final sync */
3340 		for (thr = 1; thr < controlled_threads; ++thr) {
3341 			struct paca_struct *paca = paca_ptrs[pcpu + thr];
3342 
3343 			while (paca->kvm_hstate.kvm_split_mode) {
3344 				HMT_low();
3345 				barrier();
3346 			}
3347 			HMT_medium();
3348 		}
3349 	}
3350 	split_info.do_nap = 0;
3351 
3352 	kvmppc_set_host_core(pcpu);
3353 
3354 	local_irq_enable();
3355 	guest_exit();
3356 
3357 	/* Let secondaries go back to the offline loop */
3358 	for (i = 0; i < controlled_threads; ++i) {
3359 		kvmppc_release_hwthread(pcpu + i);
3360 		if (sip && sip->napped[i])
3361 			kvmppc_ipi_thread(pcpu + i);
3362 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3363 	}
3364 
3365 	spin_unlock(&vc->lock);
3366 
3367 	/* make sure updates to secondary vcpu structs are visible now */
3368 	smp_mb();
3369 
3370 	preempt_enable();
3371 
3372 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3373 		pvc = core_info.vc[sub];
3374 		post_guest_process(pvc, pvc == vc);
3375 	}
3376 
3377 	spin_lock(&vc->lock);
3378 
3379  out:
3380 	vc->vcore_state = VCORE_INACTIVE;
3381 	trace_kvmppc_run_core(vc, 1);
3382 }
3383 
3384 /*
3385  * Load up hypervisor-mode registers on P9.
3386  */
3387 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3388 				     unsigned long lpcr)
3389 {
3390 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3391 	s64 hdec;
3392 	u64 tb, purr, spurr;
3393 	int trap;
3394 	unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3395 	unsigned long host_ciabr = mfspr(SPRN_CIABR);
3396 	unsigned long host_dawr = mfspr(SPRN_DAWR);
3397 	unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3398 	unsigned long host_psscr = mfspr(SPRN_PSSCR);
3399 	unsigned long host_pidr = mfspr(SPRN_PID);
3400 
3401 	hdec = time_limit - mftb();
3402 	if (hdec < 0)
3403 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3404 	mtspr(SPRN_HDEC, hdec);
3405 
3406 	if (vc->tb_offset) {
3407 		u64 new_tb = mftb() + vc->tb_offset;
3408 		mtspr(SPRN_TBU40, new_tb);
3409 		tb = mftb();
3410 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3411 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3412 		vc->tb_offset_applied = vc->tb_offset;
3413 	}
3414 
3415 	if (vc->pcr)
3416 		mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3417 	mtspr(SPRN_DPDES, vc->dpdes);
3418 	mtspr(SPRN_VTB, vc->vtb);
3419 
3420 	local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3421 	local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3422 	mtspr(SPRN_PURR, vcpu->arch.purr);
3423 	mtspr(SPRN_SPURR, vcpu->arch.spurr);
3424 
3425 	if (dawr_enabled()) {
3426 		mtspr(SPRN_DAWR, vcpu->arch.dawr);
3427 		mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3428 	}
3429 	mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3430 	mtspr(SPRN_IC, vcpu->arch.ic);
3431 	mtspr(SPRN_PID, vcpu->arch.pid);
3432 
3433 	mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3434 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3435 
3436 	mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3437 
3438 	mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3439 	mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3440 	mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3441 	mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3442 
3443 	mtspr(SPRN_AMOR, ~0UL);
3444 
3445 	mtspr(SPRN_LPCR, lpcr);
3446 	isync();
3447 
3448 	kvmppc_xive_push_vcpu(vcpu);
3449 
3450 	mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3451 	mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3452 
3453 	trap = __kvmhv_vcpu_entry_p9(vcpu);
3454 
3455 	/* Advance host PURR/SPURR by the amount used by guest */
3456 	purr = mfspr(SPRN_PURR);
3457 	spurr = mfspr(SPRN_SPURR);
3458 	mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3459 	      purr - vcpu->arch.purr);
3460 	mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3461 	      spurr - vcpu->arch.spurr);
3462 	vcpu->arch.purr = purr;
3463 	vcpu->arch.spurr = spurr;
3464 
3465 	vcpu->arch.ic = mfspr(SPRN_IC);
3466 	vcpu->arch.pid = mfspr(SPRN_PID);
3467 	vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3468 
3469 	vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3470 	vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3471 	vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3472 	vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3473 
3474 	/* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3475 	mtspr(SPRN_PSSCR, host_psscr |
3476 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3477 	mtspr(SPRN_HFSCR, host_hfscr);
3478 	mtspr(SPRN_CIABR, host_ciabr);
3479 	mtspr(SPRN_DAWR, host_dawr);
3480 	mtspr(SPRN_DAWRX, host_dawrx);
3481 	mtspr(SPRN_PID, host_pidr);
3482 
3483 	/*
3484 	 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3485 	 * case we interrupted the guest between a tlbie and a ptesync.
3486 	 */
3487 	asm volatile("eieio; tlbsync; ptesync");
3488 
3489 	mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);	/* restore host LPID */
3490 	isync();
3491 
3492 	vc->dpdes = mfspr(SPRN_DPDES);
3493 	vc->vtb = mfspr(SPRN_VTB);
3494 	mtspr(SPRN_DPDES, 0);
3495 	if (vc->pcr)
3496 		mtspr(SPRN_PCR, PCR_MASK);
3497 
3498 	if (vc->tb_offset_applied) {
3499 		u64 new_tb = mftb() - vc->tb_offset_applied;
3500 		mtspr(SPRN_TBU40, new_tb);
3501 		tb = mftb();
3502 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3503 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3504 		vc->tb_offset_applied = 0;
3505 	}
3506 
3507 	mtspr(SPRN_HDEC, 0x7fffffff);
3508 	mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3509 
3510 	return trap;
3511 }
3512 
3513 /*
3514  * Virtual-mode guest entry for POWER9 and later when the host and
3515  * guest are both using the radix MMU.  The LPIDR has already been set.
3516  */
3517 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3518 			 unsigned long lpcr)
3519 {
3520 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3521 	unsigned long host_dscr = mfspr(SPRN_DSCR);
3522 	unsigned long host_tidr = mfspr(SPRN_TIDR);
3523 	unsigned long host_iamr = mfspr(SPRN_IAMR);
3524 	unsigned long host_amr = mfspr(SPRN_AMR);
3525 	s64 dec;
3526 	u64 tb;
3527 	int trap, save_pmu;
3528 
3529 	dec = mfspr(SPRN_DEC);
3530 	tb = mftb();
3531 	if (dec < 512)
3532 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3533 	local_paca->kvm_hstate.dec_expires = dec + tb;
3534 	if (local_paca->kvm_hstate.dec_expires < time_limit)
3535 		time_limit = local_paca->kvm_hstate.dec_expires;
3536 
3537 	vcpu->arch.ceded = 0;
3538 
3539 	kvmhv_save_host_pmu();		/* saves it to PACA kvm_hstate */
3540 
3541 	kvmppc_subcore_enter_guest();
3542 
3543 	vc->entry_exit_map = 1;
3544 	vc->in_guest = 1;
3545 
3546 	if (vcpu->arch.vpa.pinned_addr) {
3547 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3548 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3549 		lp->yield_count = cpu_to_be32(yield_count);
3550 		vcpu->arch.vpa.dirty = 1;
3551 	}
3552 
3553 	if (cpu_has_feature(CPU_FTR_TM) ||
3554 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3555 		kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3556 
3557 	kvmhv_load_guest_pmu(vcpu);
3558 
3559 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3560 	load_fp_state(&vcpu->arch.fp);
3561 #ifdef CONFIG_ALTIVEC
3562 	load_vr_state(&vcpu->arch.vr);
3563 #endif
3564 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3565 
3566 	mtspr(SPRN_DSCR, vcpu->arch.dscr);
3567 	mtspr(SPRN_IAMR, vcpu->arch.iamr);
3568 	mtspr(SPRN_PSPB, vcpu->arch.pspb);
3569 	mtspr(SPRN_FSCR, vcpu->arch.fscr);
3570 	mtspr(SPRN_TAR, vcpu->arch.tar);
3571 	mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3572 	mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3573 	mtspr(SPRN_BESCR, vcpu->arch.bescr);
3574 	mtspr(SPRN_WORT, vcpu->arch.wort);
3575 	mtspr(SPRN_TIDR, vcpu->arch.tid);
3576 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3577 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3578 	mtspr(SPRN_AMR, vcpu->arch.amr);
3579 	mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3580 
3581 	if (!(vcpu->arch.ctrl & 1))
3582 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3583 
3584 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3585 
3586 	if (kvmhv_on_pseries()) {
3587 		/*
3588 		 * We need to save and restore the guest visible part of the
3589 		 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3590 		 * doesn't do this for us. Note only required if pseries since
3591 		 * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3592 		 */
3593 		unsigned long host_psscr;
3594 		/* call our hypervisor to load up HV regs and go */
3595 		struct hv_guest_state hvregs;
3596 
3597 		host_psscr = mfspr(SPRN_PSSCR_PR);
3598 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3599 		kvmhv_save_hv_regs(vcpu, &hvregs);
3600 		hvregs.lpcr = lpcr;
3601 		vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3602 		hvregs.version = HV_GUEST_STATE_VERSION;
3603 		if (vcpu->arch.nested) {
3604 			hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3605 			hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3606 		} else {
3607 			hvregs.lpid = vcpu->kvm->arch.lpid;
3608 			hvregs.vcpu_token = vcpu->vcpu_id;
3609 		}
3610 		hvregs.hdec_expiry = time_limit;
3611 		trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3612 					  __pa(&vcpu->arch.regs));
3613 		kvmhv_restore_hv_return_state(vcpu, &hvregs);
3614 		vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3615 		vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3616 		vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3617 		vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3618 		mtspr(SPRN_PSSCR_PR, host_psscr);
3619 
3620 		/* H_CEDE has to be handled now, not later */
3621 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3622 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3623 			kvmppc_nested_cede(vcpu);
3624 			trap = 0;
3625 		}
3626 	} else {
3627 		trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3628 	}
3629 
3630 	vcpu->arch.slb_max = 0;
3631 	dec = mfspr(SPRN_DEC);
3632 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3633 		dec = (s32) dec;
3634 	tb = mftb();
3635 	vcpu->arch.dec_expires = dec + tb;
3636 	vcpu->cpu = -1;
3637 	vcpu->arch.thread_cpu = -1;
3638 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3639 
3640 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
3641 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
3642 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
3643 	vcpu->arch.tar = mfspr(SPRN_TAR);
3644 	vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3645 	vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3646 	vcpu->arch.bescr = mfspr(SPRN_BESCR);
3647 	vcpu->arch.wort = mfspr(SPRN_WORT);
3648 	vcpu->arch.tid = mfspr(SPRN_TIDR);
3649 	vcpu->arch.amr = mfspr(SPRN_AMR);
3650 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3651 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
3652 
3653 	mtspr(SPRN_PSPB, 0);
3654 	mtspr(SPRN_WORT, 0);
3655 	mtspr(SPRN_UAMOR, 0);
3656 	mtspr(SPRN_DSCR, host_dscr);
3657 	mtspr(SPRN_TIDR, host_tidr);
3658 	mtspr(SPRN_IAMR, host_iamr);
3659 	mtspr(SPRN_PSPB, 0);
3660 
3661 	if (host_amr != vcpu->arch.amr)
3662 		mtspr(SPRN_AMR, host_amr);
3663 
3664 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3665 	store_fp_state(&vcpu->arch.fp);
3666 #ifdef CONFIG_ALTIVEC
3667 	store_vr_state(&vcpu->arch.vr);
3668 #endif
3669 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3670 
3671 	if (cpu_has_feature(CPU_FTR_TM) ||
3672 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3673 		kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3674 
3675 	save_pmu = 1;
3676 	if (vcpu->arch.vpa.pinned_addr) {
3677 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3678 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3679 		lp->yield_count = cpu_to_be32(yield_count);
3680 		vcpu->arch.vpa.dirty = 1;
3681 		save_pmu = lp->pmcregs_in_use;
3682 	}
3683 	/* Must save pmu if this guest is capable of running nested guests */
3684 	save_pmu |= nesting_enabled(vcpu->kvm);
3685 
3686 	kvmhv_save_guest_pmu(vcpu, save_pmu);
3687 
3688 	vc->entry_exit_map = 0x101;
3689 	vc->in_guest = 0;
3690 
3691 	mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3692 	mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3693 
3694 	kvmhv_load_host_pmu();
3695 
3696 	kvmppc_subcore_exit_guest();
3697 
3698 	return trap;
3699 }
3700 
3701 /*
3702  * Wait for some other vcpu thread to execute us, and
3703  * wake us up when we need to handle something in the host.
3704  */
3705 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3706 				 struct kvm_vcpu *vcpu, int wait_state)
3707 {
3708 	DEFINE_WAIT(wait);
3709 
3710 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3711 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3712 		spin_unlock(&vc->lock);
3713 		schedule();
3714 		spin_lock(&vc->lock);
3715 	}
3716 	finish_wait(&vcpu->arch.cpu_run, &wait);
3717 }
3718 
3719 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3720 {
3721 	if (!halt_poll_ns_grow)
3722 		return;
3723 
3724 	vc->halt_poll_ns *= halt_poll_ns_grow;
3725 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3726 		vc->halt_poll_ns = halt_poll_ns_grow_start;
3727 }
3728 
3729 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3730 {
3731 	if (halt_poll_ns_shrink == 0)
3732 		vc->halt_poll_ns = 0;
3733 	else
3734 		vc->halt_poll_ns /= halt_poll_ns_shrink;
3735 }
3736 
3737 #ifdef CONFIG_KVM_XICS
3738 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3739 {
3740 	if (!xics_on_xive())
3741 		return false;
3742 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3743 		vcpu->arch.xive_saved_state.cppr;
3744 }
3745 #else
3746 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3747 {
3748 	return false;
3749 }
3750 #endif /* CONFIG_KVM_XICS */
3751 
3752 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3753 {
3754 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3755 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3756 		return true;
3757 
3758 	return false;
3759 }
3760 
3761 /*
3762  * Check to see if any of the runnable vcpus on the vcore have pending
3763  * exceptions or are no longer ceded
3764  */
3765 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3766 {
3767 	struct kvm_vcpu *vcpu;
3768 	int i;
3769 
3770 	for_each_runnable_thread(i, vcpu, vc) {
3771 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3772 			return 1;
3773 	}
3774 
3775 	return 0;
3776 }
3777 
3778 /*
3779  * All the vcpus in this vcore are idle, so wait for a decrementer
3780  * or external interrupt to one of the vcpus.  vc->lock is held.
3781  */
3782 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3783 {
3784 	ktime_t cur, start_poll, start_wait;
3785 	int do_sleep = 1;
3786 	u64 block_ns;
3787 	DECLARE_SWAITQUEUE(wait);
3788 
3789 	/* Poll for pending exceptions and ceded state */
3790 	cur = start_poll = ktime_get();
3791 	if (vc->halt_poll_ns) {
3792 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3793 		++vc->runner->stat.halt_attempted_poll;
3794 
3795 		vc->vcore_state = VCORE_POLLING;
3796 		spin_unlock(&vc->lock);
3797 
3798 		do {
3799 			if (kvmppc_vcore_check_block(vc)) {
3800 				do_sleep = 0;
3801 				break;
3802 			}
3803 			cur = ktime_get();
3804 		} while (single_task_running() && ktime_before(cur, stop));
3805 
3806 		spin_lock(&vc->lock);
3807 		vc->vcore_state = VCORE_INACTIVE;
3808 
3809 		if (!do_sleep) {
3810 			++vc->runner->stat.halt_successful_poll;
3811 			goto out;
3812 		}
3813 	}
3814 
3815 	prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3816 
3817 	if (kvmppc_vcore_check_block(vc)) {
3818 		finish_swait(&vc->wq, &wait);
3819 		do_sleep = 0;
3820 		/* If we polled, count this as a successful poll */
3821 		if (vc->halt_poll_ns)
3822 			++vc->runner->stat.halt_successful_poll;
3823 		goto out;
3824 	}
3825 
3826 	start_wait = ktime_get();
3827 
3828 	vc->vcore_state = VCORE_SLEEPING;
3829 	trace_kvmppc_vcore_blocked(vc, 0);
3830 	spin_unlock(&vc->lock);
3831 	schedule();
3832 	finish_swait(&vc->wq, &wait);
3833 	spin_lock(&vc->lock);
3834 	vc->vcore_state = VCORE_INACTIVE;
3835 	trace_kvmppc_vcore_blocked(vc, 1);
3836 	++vc->runner->stat.halt_successful_wait;
3837 
3838 	cur = ktime_get();
3839 
3840 out:
3841 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3842 
3843 	/* Attribute wait time */
3844 	if (do_sleep) {
3845 		vc->runner->stat.halt_wait_ns +=
3846 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3847 		/* Attribute failed poll time */
3848 		if (vc->halt_poll_ns)
3849 			vc->runner->stat.halt_poll_fail_ns +=
3850 				ktime_to_ns(start_wait) -
3851 				ktime_to_ns(start_poll);
3852 	} else {
3853 		/* Attribute successful poll time */
3854 		if (vc->halt_poll_ns)
3855 			vc->runner->stat.halt_poll_success_ns +=
3856 				ktime_to_ns(cur) -
3857 				ktime_to_ns(start_poll);
3858 	}
3859 
3860 	/* Adjust poll time */
3861 	if (halt_poll_ns) {
3862 		if (block_ns <= vc->halt_poll_ns)
3863 			;
3864 		/* We slept and blocked for longer than the max halt time */
3865 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3866 			shrink_halt_poll_ns(vc);
3867 		/* We slept and our poll time is too small */
3868 		else if (vc->halt_poll_ns < halt_poll_ns &&
3869 				block_ns < halt_poll_ns)
3870 			grow_halt_poll_ns(vc);
3871 		if (vc->halt_poll_ns > halt_poll_ns)
3872 			vc->halt_poll_ns = halt_poll_ns;
3873 	} else
3874 		vc->halt_poll_ns = 0;
3875 
3876 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3877 }
3878 
3879 /*
3880  * This never fails for a radix guest, as none of the operations it does
3881  * for a radix guest can fail or have a way to report failure.
3882  * kvmhv_run_single_vcpu() relies on this fact.
3883  */
3884 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3885 {
3886 	int r = 0;
3887 	struct kvm *kvm = vcpu->kvm;
3888 
3889 	mutex_lock(&kvm->arch.mmu_setup_lock);
3890 	if (!kvm->arch.mmu_ready) {
3891 		if (!kvm_is_radix(kvm))
3892 			r = kvmppc_hv_setup_htab_rma(vcpu);
3893 		if (!r) {
3894 			if (cpu_has_feature(CPU_FTR_ARCH_300))
3895 				kvmppc_setup_partition_table(kvm);
3896 			kvm->arch.mmu_ready = 1;
3897 		}
3898 	}
3899 	mutex_unlock(&kvm->arch.mmu_setup_lock);
3900 	return r;
3901 }
3902 
3903 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3904 {
3905 	int n_ceded, i, r;
3906 	struct kvmppc_vcore *vc;
3907 	struct kvm_vcpu *v;
3908 
3909 	trace_kvmppc_run_vcpu_enter(vcpu);
3910 
3911 	kvm_run->exit_reason = 0;
3912 	vcpu->arch.ret = RESUME_GUEST;
3913 	vcpu->arch.trap = 0;
3914 	kvmppc_update_vpas(vcpu);
3915 
3916 	/*
3917 	 * Synchronize with other threads in this virtual core
3918 	 */
3919 	vc = vcpu->arch.vcore;
3920 	spin_lock(&vc->lock);
3921 	vcpu->arch.ceded = 0;
3922 	vcpu->arch.run_task = current;
3923 	vcpu->arch.kvm_run = kvm_run;
3924 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3925 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3926 	vcpu->arch.busy_preempt = TB_NIL;
3927 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3928 	++vc->n_runnable;
3929 
3930 	/*
3931 	 * This happens the first time this is called for a vcpu.
3932 	 * If the vcore is already running, we may be able to start
3933 	 * this thread straight away and have it join in.
3934 	 */
3935 	if (!signal_pending(current)) {
3936 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
3937 		     vc->vcore_state == VCORE_RUNNING) &&
3938 			   !VCORE_IS_EXITING(vc)) {
3939 			kvmppc_create_dtl_entry(vcpu, vc);
3940 			kvmppc_start_thread(vcpu, vc);
3941 			trace_kvm_guest_enter(vcpu);
3942 		} else if (vc->vcore_state == VCORE_SLEEPING) {
3943 			swake_up_one(&vc->wq);
3944 		}
3945 
3946 	}
3947 
3948 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3949 	       !signal_pending(current)) {
3950 		/* See if the MMU is ready to go */
3951 		if (!vcpu->kvm->arch.mmu_ready) {
3952 			spin_unlock(&vc->lock);
3953 			r = kvmhv_setup_mmu(vcpu);
3954 			spin_lock(&vc->lock);
3955 			if (r) {
3956 				kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3957 				kvm_run->fail_entry.
3958 					hardware_entry_failure_reason = 0;
3959 				vcpu->arch.ret = r;
3960 				break;
3961 			}
3962 		}
3963 
3964 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3965 			kvmppc_vcore_end_preempt(vc);
3966 
3967 		if (vc->vcore_state != VCORE_INACTIVE) {
3968 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3969 			continue;
3970 		}
3971 		for_each_runnable_thread(i, v, vc) {
3972 			kvmppc_core_prepare_to_enter(v);
3973 			if (signal_pending(v->arch.run_task)) {
3974 				kvmppc_remove_runnable(vc, v);
3975 				v->stat.signal_exits++;
3976 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3977 				v->arch.ret = -EINTR;
3978 				wake_up(&v->arch.cpu_run);
3979 			}
3980 		}
3981 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3982 			break;
3983 		n_ceded = 0;
3984 		for_each_runnable_thread(i, v, vc) {
3985 			if (!kvmppc_vcpu_woken(v))
3986 				n_ceded += v->arch.ceded;
3987 			else
3988 				v->arch.ceded = 0;
3989 		}
3990 		vc->runner = vcpu;
3991 		if (n_ceded == vc->n_runnable) {
3992 			kvmppc_vcore_blocked(vc);
3993 		} else if (need_resched()) {
3994 			kvmppc_vcore_preempt(vc);
3995 			/* Let something else run */
3996 			cond_resched_lock(&vc->lock);
3997 			if (vc->vcore_state == VCORE_PREEMPT)
3998 				kvmppc_vcore_end_preempt(vc);
3999 		} else {
4000 			kvmppc_run_core(vc);
4001 		}
4002 		vc->runner = NULL;
4003 	}
4004 
4005 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4006 	       (vc->vcore_state == VCORE_RUNNING ||
4007 		vc->vcore_state == VCORE_EXITING ||
4008 		vc->vcore_state == VCORE_PIGGYBACK))
4009 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4010 
4011 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4012 		kvmppc_vcore_end_preempt(vc);
4013 
4014 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4015 		kvmppc_remove_runnable(vc, vcpu);
4016 		vcpu->stat.signal_exits++;
4017 		kvm_run->exit_reason = KVM_EXIT_INTR;
4018 		vcpu->arch.ret = -EINTR;
4019 	}
4020 
4021 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4022 		/* Wake up some vcpu to run the core */
4023 		i = -1;
4024 		v = next_runnable_thread(vc, &i);
4025 		wake_up(&v->arch.cpu_run);
4026 	}
4027 
4028 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4029 	spin_unlock(&vc->lock);
4030 	return vcpu->arch.ret;
4031 }
4032 
4033 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4034 			  struct kvm_vcpu *vcpu, u64 time_limit,
4035 			  unsigned long lpcr)
4036 {
4037 	int trap, r, pcpu;
4038 	int srcu_idx, lpid;
4039 	struct kvmppc_vcore *vc;
4040 	struct kvm *kvm = vcpu->kvm;
4041 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4042 
4043 	trace_kvmppc_run_vcpu_enter(vcpu);
4044 
4045 	kvm_run->exit_reason = 0;
4046 	vcpu->arch.ret = RESUME_GUEST;
4047 	vcpu->arch.trap = 0;
4048 
4049 	vc = vcpu->arch.vcore;
4050 	vcpu->arch.ceded = 0;
4051 	vcpu->arch.run_task = current;
4052 	vcpu->arch.kvm_run = kvm_run;
4053 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4054 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4055 	vcpu->arch.busy_preempt = TB_NIL;
4056 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4057 	vc->runnable_threads[0] = vcpu;
4058 	vc->n_runnable = 1;
4059 	vc->runner = vcpu;
4060 
4061 	/* See if the MMU is ready to go */
4062 	if (!kvm->arch.mmu_ready)
4063 		kvmhv_setup_mmu(vcpu);
4064 
4065 	if (need_resched())
4066 		cond_resched();
4067 
4068 	kvmppc_update_vpas(vcpu);
4069 
4070 	init_vcore_to_run(vc);
4071 	vc->preempt_tb = TB_NIL;
4072 
4073 	preempt_disable();
4074 	pcpu = smp_processor_id();
4075 	vc->pcpu = pcpu;
4076 	kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4077 
4078 	local_irq_disable();
4079 	hard_irq_disable();
4080 	if (signal_pending(current))
4081 		goto sigpend;
4082 	if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4083 		goto out;
4084 
4085 	if (!nested) {
4086 		kvmppc_core_prepare_to_enter(vcpu);
4087 		if (vcpu->arch.doorbell_request) {
4088 			vc->dpdes = 1;
4089 			smp_wmb();
4090 			vcpu->arch.doorbell_request = 0;
4091 		}
4092 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4093 			     &vcpu->arch.pending_exceptions))
4094 			lpcr |= LPCR_MER;
4095 	} else if (vcpu->arch.pending_exceptions ||
4096 		   vcpu->arch.doorbell_request ||
4097 		   xive_interrupt_pending(vcpu)) {
4098 		vcpu->arch.ret = RESUME_HOST;
4099 		goto out;
4100 	}
4101 
4102 	kvmppc_clear_host_core(pcpu);
4103 
4104 	local_paca->kvm_hstate.tid = 0;
4105 	local_paca->kvm_hstate.napping = 0;
4106 	local_paca->kvm_hstate.kvm_split_mode = NULL;
4107 	kvmppc_start_thread(vcpu, vc);
4108 	kvmppc_create_dtl_entry(vcpu, vc);
4109 	trace_kvm_guest_enter(vcpu);
4110 
4111 	vc->vcore_state = VCORE_RUNNING;
4112 	trace_kvmppc_run_core(vc, 0);
4113 
4114 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4115 		lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4116 		mtspr(SPRN_LPID, lpid);
4117 		isync();
4118 		kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4119 	}
4120 
4121 	guest_enter_irqoff();
4122 
4123 	srcu_idx = srcu_read_lock(&kvm->srcu);
4124 
4125 	this_cpu_disable_ftrace();
4126 
4127 	/* Tell lockdep that we're about to enable interrupts */
4128 	trace_hardirqs_on();
4129 
4130 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4131 	vcpu->arch.trap = trap;
4132 
4133 	trace_hardirqs_off();
4134 
4135 	this_cpu_enable_ftrace();
4136 
4137 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4138 
4139 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4140 		mtspr(SPRN_LPID, kvm->arch.host_lpid);
4141 		isync();
4142 	}
4143 
4144 	set_irq_happened(trap);
4145 
4146 	kvmppc_set_host_core(pcpu);
4147 
4148 	local_irq_enable();
4149 	guest_exit();
4150 
4151 	cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4152 
4153 	preempt_enable();
4154 
4155 	/*
4156 	 * cancel pending decrementer exception if DEC is now positive, or if
4157 	 * entering a nested guest in which case the decrementer is now owned
4158 	 * by L2 and the L1 decrementer is provided in hdec_expires
4159 	 */
4160 	if (kvmppc_core_pending_dec(vcpu) &&
4161 			((get_tb() < vcpu->arch.dec_expires) ||
4162 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4163 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4164 		kvmppc_core_dequeue_dec(vcpu);
4165 
4166 	trace_kvm_guest_exit(vcpu);
4167 	r = RESUME_GUEST;
4168 	if (trap) {
4169 		if (!nested)
4170 			r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4171 		else
4172 			r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4173 	}
4174 	vcpu->arch.ret = r;
4175 
4176 	if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4177 	    !kvmppc_vcpu_woken(vcpu)) {
4178 		kvmppc_set_timer(vcpu);
4179 		while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4180 			if (signal_pending(current)) {
4181 				vcpu->stat.signal_exits++;
4182 				kvm_run->exit_reason = KVM_EXIT_INTR;
4183 				vcpu->arch.ret = -EINTR;
4184 				break;
4185 			}
4186 			spin_lock(&vc->lock);
4187 			kvmppc_vcore_blocked(vc);
4188 			spin_unlock(&vc->lock);
4189 		}
4190 	}
4191 	vcpu->arch.ceded = 0;
4192 
4193 	vc->vcore_state = VCORE_INACTIVE;
4194 	trace_kvmppc_run_core(vc, 1);
4195 
4196  done:
4197 	kvmppc_remove_runnable(vc, vcpu);
4198 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4199 
4200 	return vcpu->arch.ret;
4201 
4202  sigpend:
4203 	vcpu->stat.signal_exits++;
4204 	kvm_run->exit_reason = KVM_EXIT_INTR;
4205 	vcpu->arch.ret = -EINTR;
4206  out:
4207 	local_irq_enable();
4208 	preempt_enable();
4209 	goto done;
4210 }
4211 
4212 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4213 {
4214 	int r;
4215 	int srcu_idx;
4216 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
4217 	unsigned long user_tar = 0;
4218 	unsigned int user_vrsave;
4219 	struct kvm *kvm;
4220 
4221 	if (!vcpu->arch.sane) {
4222 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4223 		return -EINVAL;
4224 	}
4225 
4226 	/*
4227 	 * Don't allow entry with a suspended transaction, because
4228 	 * the guest entry/exit code will lose it.
4229 	 * If the guest has TM enabled, save away their TM-related SPRs
4230 	 * (they will get restored by the TM unavailable interrupt).
4231 	 */
4232 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4233 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4234 	    (current->thread.regs->msr & MSR_TM)) {
4235 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4236 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4237 			run->fail_entry.hardware_entry_failure_reason = 0;
4238 			return -EINVAL;
4239 		}
4240 		/* Enable TM so we can read the TM SPRs */
4241 		mtmsr(mfmsr() | MSR_TM);
4242 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4243 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4244 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4245 		current->thread.regs->msr &= ~MSR_TM;
4246 	}
4247 #endif
4248 
4249 	/*
4250 	 * Force online to 1 for the sake of old userspace which doesn't
4251 	 * set it.
4252 	 */
4253 	if (!vcpu->arch.online) {
4254 		atomic_inc(&vcpu->arch.vcore->online_count);
4255 		vcpu->arch.online = 1;
4256 	}
4257 
4258 	kvmppc_core_prepare_to_enter(vcpu);
4259 
4260 	/* No need to go into the guest when all we'll do is come back out */
4261 	if (signal_pending(current)) {
4262 		run->exit_reason = KVM_EXIT_INTR;
4263 		return -EINTR;
4264 	}
4265 
4266 	kvm = vcpu->kvm;
4267 	atomic_inc(&kvm->arch.vcpus_running);
4268 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4269 	smp_mb();
4270 
4271 	flush_all_to_thread(current);
4272 
4273 	/* Save userspace EBB and other register values */
4274 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4275 		ebb_regs[0] = mfspr(SPRN_EBBHR);
4276 		ebb_regs[1] = mfspr(SPRN_EBBRR);
4277 		ebb_regs[2] = mfspr(SPRN_BESCR);
4278 		user_tar = mfspr(SPRN_TAR);
4279 	}
4280 	user_vrsave = mfspr(SPRN_VRSAVE);
4281 
4282 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4283 	vcpu->arch.pgdir = current->mm->pgd;
4284 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4285 
4286 	do {
4287 		/*
4288 		 * The early POWER9 chips that can't mix radix and HPT threads
4289 		 * on the same core also need the workaround for the problem
4290 		 * where the TLB would prefetch entries in the guest exit path
4291 		 * for radix guests using the guest PIDR value and LPID 0.
4292 		 * The workaround is in the old path (kvmppc_run_vcpu())
4293 		 * but not the new path (kvmhv_run_single_vcpu()).
4294 		 */
4295 		if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4296 		    !no_mixing_hpt_and_radix)
4297 			r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4298 						  vcpu->arch.vcore->lpcr);
4299 		else
4300 			r = kvmppc_run_vcpu(run, vcpu);
4301 
4302 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4303 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4304 			trace_kvm_hcall_enter(vcpu);
4305 			r = kvmppc_pseries_do_hcall(vcpu);
4306 			trace_kvm_hcall_exit(vcpu, r);
4307 			kvmppc_core_prepare_to_enter(vcpu);
4308 		} else if (r == RESUME_PAGE_FAULT) {
4309 			srcu_idx = srcu_read_lock(&kvm->srcu);
4310 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
4311 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4312 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4313 		} else if (r == RESUME_PASSTHROUGH) {
4314 			if (WARN_ON(xics_on_xive()))
4315 				r = H_SUCCESS;
4316 			else
4317 				r = kvmppc_xics_rm_complete(vcpu, 0);
4318 		}
4319 	} while (is_kvmppc_resume_guest(r));
4320 
4321 	/* Restore userspace EBB and other register values */
4322 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4323 		mtspr(SPRN_EBBHR, ebb_regs[0]);
4324 		mtspr(SPRN_EBBRR, ebb_regs[1]);
4325 		mtspr(SPRN_BESCR, ebb_regs[2]);
4326 		mtspr(SPRN_TAR, user_tar);
4327 		mtspr(SPRN_FSCR, current->thread.fscr);
4328 	}
4329 	mtspr(SPRN_VRSAVE, user_vrsave);
4330 
4331 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4332 	atomic_dec(&kvm->arch.vcpus_running);
4333 	return r;
4334 }
4335 
4336 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4337 				     int shift, int sllp)
4338 {
4339 	(*sps)->page_shift = shift;
4340 	(*sps)->slb_enc = sllp;
4341 	(*sps)->enc[0].page_shift = shift;
4342 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4343 	/*
4344 	 * Add 16MB MPSS support (may get filtered out by userspace)
4345 	 */
4346 	if (shift != 24) {
4347 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4348 		if (penc != -1) {
4349 			(*sps)->enc[1].page_shift = 24;
4350 			(*sps)->enc[1].pte_enc = penc;
4351 		}
4352 	}
4353 	(*sps)++;
4354 }
4355 
4356 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4357 					 struct kvm_ppc_smmu_info *info)
4358 {
4359 	struct kvm_ppc_one_seg_page_size *sps;
4360 
4361 	/*
4362 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4363 	 * POWER7 doesn't support keys for instruction accesses,
4364 	 * POWER8 and POWER9 do.
4365 	 */
4366 	info->data_keys = 32;
4367 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4368 
4369 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4370 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4371 	info->slb_size = 32;
4372 
4373 	/* We only support these sizes for now, and no muti-size segments */
4374 	sps = &info->sps[0];
4375 	kvmppc_add_seg_page_size(&sps, 12, 0);
4376 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4377 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4378 
4379 	/* If running as a nested hypervisor, we don't support HPT guests */
4380 	if (kvmhv_on_pseries())
4381 		info->flags |= KVM_PPC_NO_HASH;
4382 
4383 	return 0;
4384 }
4385 
4386 /*
4387  * Get (and clear) the dirty memory log for a memory slot.
4388  */
4389 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4390 					 struct kvm_dirty_log *log)
4391 {
4392 	struct kvm_memslots *slots;
4393 	struct kvm_memory_slot *memslot;
4394 	int i, r;
4395 	unsigned long n;
4396 	unsigned long *buf, *p;
4397 	struct kvm_vcpu *vcpu;
4398 
4399 	mutex_lock(&kvm->slots_lock);
4400 
4401 	r = -EINVAL;
4402 	if (log->slot >= KVM_USER_MEM_SLOTS)
4403 		goto out;
4404 
4405 	slots = kvm_memslots(kvm);
4406 	memslot = id_to_memslot(slots, log->slot);
4407 	r = -ENOENT;
4408 	if (!memslot->dirty_bitmap)
4409 		goto out;
4410 
4411 	/*
4412 	 * Use second half of bitmap area because both HPT and radix
4413 	 * accumulate bits in the first half.
4414 	 */
4415 	n = kvm_dirty_bitmap_bytes(memslot);
4416 	buf = memslot->dirty_bitmap + n / sizeof(long);
4417 	memset(buf, 0, n);
4418 
4419 	if (kvm_is_radix(kvm))
4420 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4421 	else
4422 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4423 	if (r)
4424 		goto out;
4425 
4426 	/*
4427 	 * We accumulate dirty bits in the first half of the
4428 	 * memslot's dirty_bitmap area, for when pages are paged
4429 	 * out or modified by the host directly.  Pick up these
4430 	 * bits and add them to the map.
4431 	 */
4432 	p = memslot->dirty_bitmap;
4433 	for (i = 0; i < n / sizeof(long); ++i)
4434 		buf[i] |= xchg(&p[i], 0);
4435 
4436 	/* Harvest dirty bits from VPA and DTL updates */
4437 	/* Note: we never modify the SLB shadow buffer areas */
4438 	kvm_for_each_vcpu(i, vcpu, kvm) {
4439 		spin_lock(&vcpu->arch.vpa_update_lock);
4440 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4441 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4442 		spin_unlock(&vcpu->arch.vpa_update_lock);
4443 	}
4444 
4445 	r = -EFAULT;
4446 	if (copy_to_user(log->dirty_bitmap, buf, n))
4447 		goto out;
4448 
4449 	r = 0;
4450 out:
4451 	mutex_unlock(&kvm->slots_lock);
4452 	return r;
4453 }
4454 
4455 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4456 					struct kvm_memory_slot *dont)
4457 {
4458 	if (!dont || free->arch.rmap != dont->arch.rmap) {
4459 		vfree(free->arch.rmap);
4460 		free->arch.rmap = NULL;
4461 	}
4462 }
4463 
4464 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4465 					 unsigned long npages)
4466 {
4467 	slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4468 	if (!slot->arch.rmap)
4469 		return -ENOMEM;
4470 
4471 	return 0;
4472 }
4473 
4474 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4475 					struct kvm_memory_slot *memslot,
4476 					const struct kvm_userspace_memory_region *mem)
4477 {
4478 	return 0;
4479 }
4480 
4481 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4482 				const struct kvm_userspace_memory_region *mem,
4483 				const struct kvm_memory_slot *old,
4484 				const struct kvm_memory_slot *new,
4485 				enum kvm_mr_change change)
4486 {
4487 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4488 
4489 	/*
4490 	 * If we are making a new memslot, it might make
4491 	 * some address that was previously cached as emulated
4492 	 * MMIO be no longer emulated MMIO, so invalidate
4493 	 * all the caches of emulated MMIO translations.
4494 	 */
4495 	if (npages)
4496 		atomic64_inc(&kvm->arch.mmio_update);
4497 
4498 	/*
4499 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4500 	 * have already called kvm_arch_flush_shadow_memslot() to
4501 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
4502 	 * previous mappings.  So the only case to handle is
4503 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4504 	 * has been changed.
4505 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4506 	 * to get rid of any THP PTEs in the partition-scoped page tables
4507 	 * so we can track dirtiness at the page level; we flush when
4508 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4509 	 * using THP PTEs.
4510 	 */
4511 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4512 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4513 		kvmppc_radix_flush_memslot(kvm, old);
4514 }
4515 
4516 /*
4517  * Update LPCR values in kvm->arch and in vcores.
4518  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4519  * of kvm->arch.lpcr update).
4520  */
4521 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4522 {
4523 	long int i;
4524 	u32 cores_done = 0;
4525 
4526 	if ((kvm->arch.lpcr & mask) == lpcr)
4527 		return;
4528 
4529 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4530 
4531 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4532 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4533 		if (!vc)
4534 			continue;
4535 		spin_lock(&vc->lock);
4536 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4537 		spin_unlock(&vc->lock);
4538 		if (++cores_done >= kvm->arch.online_vcores)
4539 			break;
4540 	}
4541 }
4542 
4543 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4544 {
4545 	return;
4546 }
4547 
4548 void kvmppc_setup_partition_table(struct kvm *kvm)
4549 {
4550 	unsigned long dw0, dw1;
4551 
4552 	if (!kvm_is_radix(kvm)) {
4553 		/* PS field - page size for VRMA */
4554 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4555 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4556 		/* HTABSIZE and HTABORG fields */
4557 		dw0 |= kvm->arch.sdr1;
4558 
4559 		/* Second dword as set by userspace */
4560 		dw1 = kvm->arch.process_table;
4561 	} else {
4562 		dw0 = PATB_HR | radix__get_tree_size() |
4563 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4564 		dw1 = PATB_GR | kvm->arch.process_table;
4565 	}
4566 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4567 }
4568 
4569 /*
4570  * Set up HPT (hashed page table) and RMA (real-mode area).
4571  * Must be called with kvm->arch.mmu_setup_lock held.
4572  */
4573 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4574 {
4575 	int err = 0;
4576 	struct kvm *kvm = vcpu->kvm;
4577 	unsigned long hva;
4578 	struct kvm_memory_slot *memslot;
4579 	struct vm_area_struct *vma;
4580 	unsigned long lpcr = 0, senc;
4581 	unsigned long psize, porder;
4582 	int srcu_idx;
4583 
4584 	/* Allocate hashed page table (if not done already) and reset it */
4585 	if (!kvm->arch.hpt.virt) {
4586 		int order = KVM_DEFAULT_HPT_ORDER;
4587 		struct kvm_hpt_info info;
4588 
4589 		err = kvmppc_allocate_hpt(&info, order);
4590 		/* If we get here, it means userspace didn't specify a
4591 		 * size explicitly.  So, try successively smaller
4592 		 * sizes if the default failed. */
4593 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4594 			err  = kvmppc_allocate_hpt(&info, order);
4595 
4596 		if (err < 0) {
4597 			pr_err("KVM: Couldn't alloc HPT\n");
4598 			goto out;
4599 		}
4600 
4601 		kvmppc_set_hpt(kvm, &info);
4602 	}
4603 
4604 	/* Look up the memslot for guest physical address 0 */
4605 	srcu_idx = srcu_read_lock(&kvm->srcu);
4606 	memslot = gfn_to_memslot(kvm, 0);
4607 
4608 	/* We must have some memory at 0 by now */
4609 	err = -EINVAL;
4610 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4611 		goto out_srcu;
4612 
4613 	/* Look up the VMA for the start of this memory slot */
4614 	hva = memslot->userspace_addr;
4615 	down_read(&current->mm->mmap_sem);
4616 	vma = find_vma(current->mm, hva);
4617 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4618 		goto up_out;
4619 
4620 	psize = vma_kernel_pagesize(vma);
4621 
4622 	up_read(&current->mm->mmap_sem);
4623 
4624 	/* We can handle 4k, 64k or 16M pages in the VRMA */
4625 	if (psize >= 0x1000000)
4626 		psize = 0x1000000;
4627 	else if (psize >= 0x10000)
4628 		psize = 0x10000;
4629 	else
4630 		psize = 0x1000;
4631 	porder = __ilog2(psize);
4632 
4633 	senc = slb_pgsize_encoding(psize);
4634 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4635 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4636 	/* Create HPTEs in the hash page table for the VRMA */
4637 	kvmppc_map_vrma(vcpu, memslot, porder);
4638 
4639 	/* Update VRMASD field in the LPCR */
4640 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4641 		/* the -4 is to account for senc values starting at 0x10 */
4642 		lpcr = senc << (LPCR_VRMASD_SH - 4);
4643 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4644 	}
4645 
4646 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4647 	smp_wmb();
4648 	err = 0;
4649  out_srcu:
4650 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4651  out:
4652 	return err;
4653 
4654  up_out:
4655 	up_read(&current->mm->mmap_sem);
4656 	goto out_srcu;
4657 }
4658 
4659 /*
4660  * Must be called with kvm->arch.mmu_setup_lock held and
4661  * mmu_ready = 0 and no vcpus running.
4662  */
4663 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4664 {
4665 	if (nesting_enabled(kvm))
4666 		kvmhv_release_all_nested(kvm);
4667 	kvmppc_rmap_reset(kvm);
4668 	kvm->arch.process_table = 0;
4669 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4670 	spin_lock(&kvm->mmu_lock);
4671 	kvm->arch.radix = 0;
4672 	spin_unlock(&kvm->mmu_lock);
4673 	kvmppc_free_radix(kvm);
4674 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
4675 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4676 	return 0;
4677 }
4678 
4679 /*
4680  * Must be called with kvm->arch.mmu_setup_lock held and
4681  * mmu_ready = 0 and no vcpus running.
4682  */
4683 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4684 {
4685 	int err;
4686 
4687 	err = kvmppc_init_vm_radix(kvm);
4688 	if (err)
4689 		return err;
4690 	kvmppc_rmap_reset(kvm);
4691 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4692 	spin_lock(&kvm->mmu_lock);
4693 	kvm->arch.radix = 1;
4694 	spin_unlock(&kvm->mmu_lock);
4695 	kvmppc_free_hpt(&kvm->arch.hpt);
4696 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4697 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4698 	return 0;
4699 }
4700 
4701 #ifdef CONFIG_KVM_XICS
4702 /*
4703  * Allocate a per-core structure for managing state about which cores are
4704  * running in the host versus the guest and for exchanging data between
4705  * real mode KVM and CPU running in the host.
4706  * This is only done for the first VM.
4707  * The allocated structure stays even if all VMs have stopped.
4708  * It is only freed when the kvm-hv module is unloaded.
4709  * It's OK for this routine to fail, we just don't support host
4710  * core operations like redirecting H_IPI wakeups.
4711  */
4712 void kvmppc_alloc_host_rm_ops(void)
4713 {
4714 	struct kvmppc_host_rm_ops *ops;
4715 	unsigned long l_ops;
4716 	int cpu, core;
4717 	int size;
4718 
4719 	/* Not the first time here ? */
4720 	if (kvmppc_host_rm_ops_hv != NULL)
4721 		return;
4722 
4723 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4724 	if (!ops)
4725 		return;
4726 
4727 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4728 	ops->rm_core = kzalloc(size, GFP_KERNEL);
4729 
4730 	if (!ops->rm_core) {
4731 		kfree(ops);
4732 		return;
4733 	}
4734 
4735 	cpus_read_lock();
4736 
4737 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4738 		if (!cpu_online(cpu))
4739 			continue;
4740 
4741 		core = cpu >> threads_shift;
4742 		ops->rm_core[core].rm_state.in_host = 1;
4743 	}
4744 
4745 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4746 
4747 	/*
4748 	 * Make the contents of the kvmppc_host_rm_ops structure visible
4749 	 * to other CPUs before we assign it to the global variable.
4750 	 * Do an atomic assignment (no locks used here), but if someone
4751 	 * beats us to it, just free our copy and return.
4752 	 */
4753 	smp_wmb();
4754 	l_ops = (unsigned long) ops;
4755 
4756 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4757 		cpus_read_unlock();
4758 		kfree(ops->rm_core);
4759 		kfree(ops);
4760 		return;
4761 	}
4762 
4763 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4764 					     "ppc/kvm_book3s:prepare",
4765 					     kvmppc_set_host_core,
4766 					     kvmppc_clear_host_core);
4767 	cpus_read_unlock();
4768 }
4769 
4770 void kvmppc_free_host_rm_ops(void)
4771 {
4772 	if (kvmppc_host_rm_ops_hv) {
4773 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4774 		kfree(kvmppc_host_rm_ops_hv->rm_core);
4775 		kfree(kvmppc_host_rm_ops_hv);
4776 		kvmppc_host_rm_ops_hv = NULL;
4777 	}
4778 }
4779 #endif
4780 
4781 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4782 {
4783 	unsigned long lpcr, lpid;
4784 	char buf[32];
4785 	int ret;
4786 
4787 	mutex_init(&kvm->arch.mmu_setup_lock);
4788 
4789 	/* Allocate the guest's logical partition ID */
4790 
4791 	lpid = kvmppc_alloc_lpid();
4792 	if ((long)lpid < 0)
4793 		return -ENOMEM;
4794 	kvm->arch.lpid = lpid;
4795 
4796 	kvmppc_alloc_host_rm_ops();
4797 
4798 	kvmhv_vm_nested_init(kvm);
4799 
4800 	/*
4801 	 * Since we don't flush the TLB when tearing down a VM,
4802 	 * and this lpid might have previously been used,
4803 	 * make sure we flush on each core before running the new VM.
4804 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4805 	 * does this flush for us.
4806 	 */
4807 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4808 		cpumask_setall(&kvm->arch.need_tlb_flush);
4809 
4810 	/* Start out with the default set of hcalls enabled */
4811 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4812 	       sizeof(kvm->arch.enabled_hcalls));
4813 
4814 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4815 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4816 
4817 	/* Init LPCR for virtual RMA mode */
4818 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4819 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
4820 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4821 		lpcr &= LPCR_PECE | LPCR_LPES;
4822 	} else {
4823 		lpcr = 0;
4824 	}
4825 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4826 		LPCR_VPM0 | LPCR_VPM1;
4827 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4828 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4829 	/* On POWER8 turn on online bit to enable PURR/SPURR */
4830 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
4831 		lpcr |= LPCR_ONL;
4832 	/*
4833 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4834 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
4835 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4836 	 * be unnecessary but better safe than sorry in case we re-enable
4837 	 * EE in HV mode with this LPCR still set)
4838 	 */
4839 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4840 		lpcr &= ~LPCR_VPM0;
4841 		lpcr |= LPCR_HVICE | LPCR_HEIC;
4842 
4843 		/*
4844 		 * If xive is enabled, we route 0x500 interrupts directly
4845 		 * to the guest.
4846 		 */
4847 		if (xics_on_xive())
4848 			lpcr |= LPCR_LPES;
4849 	}
4850 
4851 	/*
4852 	 * If the host uses radix, the guest starts out as radix.
4853 	 */
4854 	if (radix_enabled()) {
4855 		kvm->arch.radix = 1;
4856 		kvm->arch.mmu_ready = 1;
4857 		lpcr &= ~LPCR_VPM1;
4858 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4859 		ret = kvmppc_init_vm_radix(kvm);
4860 		if (ret) {
4861 			kvmppc_free_lpid(kvm->arch.lpid);
4862 			return ret;
4863 		}
4864 		kvmppc_setup_partition_table(kvm);
4865 	}
4866 
4867 	kvm->arch.lpcr = lpcr;
4868 
4869 	/* Initialization for future HPT resizes */
4870 	kvm->arch.resize_hpt = NULL;
4871 
4872 	/*
4873 	 * Work out how many sets the TLB has, for the use of
4874 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4875 	 */
4876 	if (radix_enabled())
4877 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
4878 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
4879 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
4880 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4881 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
4882 	else
4883 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
4884 
4885 	/*
4886 	 * Track that we now have a HV mode VM active. This blocks secondary
4887 	 * CPU threads from coming online.
4888 	 * On POWER9, we only need to do this if the "indep_threads_mode"
4889 	 * module parameter has been set to N.
4890 	 */
4891 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4892 		if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4893 			pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4894 			kvm->arch.threads_indep = true;
4895 		} else {
4896 			kvm->arch.threads_indep = indep_threads_mode;
4897 		}
4898 	}
4899 	if (!kvm->arch.threads_indep)
4900 		kvm_hv_vm_activated();
4901 
4902 	/*
4903 	 * Initialize smt_mode depending on processor.
4904 	 * POWER8 and earlier have to use "strict" threading, where
4905 	 * all vCPUs in a vcore have to run on the same (sub)core,
4906 	 * whereas on POWER9 the threads can each run a different
4907 	 * guest.
4908 	 */
4909 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4910 		kvm->arch.smt_mode = threads_per_subcore;
4911 	else
4912 		kvm->arch.smt_mode = 1;
4913 	kvm->arch.emul_smt_mode = 1;
4914 
4915 	/*
4916 	 * Create a debugfs directory for the VM
4917 	 */
4918 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
4919 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4920 	kvmppc_mmu_debugfs_init(kvm);
4921 	if (radix_enabled())
4922 		kvmhv_radix_debugfs_init(kvm);
4923 
4924 	return 0;
4925 }
4926 
4927 static void kvmppc_free_vcores(struct kvm *kvm)
4928 {
4929 	long int i;
4930 
4931 	for (i = 0; i < KVM_MAX_VCORES; ++i)
4932 		kfree(kvm->arch.vcores[i]);
4933 	kvm->arch.online_vcores = 0;
4934 }
4935 
4936 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4937 {
4938 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
4939 
4940 	if (!kvm->arch.threads_indep)
4941 		kvm_hv_vm_deactivated();
4942 
4943 	kvmppc_free_vcores(kvm);
4944 
4945 
4946 	if (kvm_is_radix(kvm))
4947 		kvmppc_free_radix(kvm);
4948 	else
4949 		kvmppc_free_hpt(&kvm->arch.hpt);
4950 
4951 	/* Perform global invalidation and return lpid to the pool */
4952 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4953 		if (nesting_enabled(kvm))
4954 			kvmhv_release_all_nested(kvm);
4955 		kvm->arch.process_table = 0;
4956 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4957 	}
4958 	kvmppc_free_lpid(kvm->arch.lpid);
4959 
4960 	kvmppc_free_pimap(kvm);
4961 }
4962 
4963 /* We don't need to emulate any privileged instructions or dcbz */
4964 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4965 				     unsigned int inst, int *advance)
4966 {
4967 	return EMULATE_FAIL;
4968 }
4969 
4970 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4971 					ulong spr_val)
4972 {
4973 	return EMULATE_FAIL;
4974 }
4975 
4976 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4977 					ulong *spr_val)
4978 {
4979 	return EMULATE_FAIL;
4980 }
4981 
4982 static int kvmppc_core_check_processor_compat_hv(void)
4983 {
4984 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
4985 	    cpu_has_feature(CPU_FTR_ARCH_206))
4986 		return 0;
4987 
4988 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
4989 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4990 		return 0;
4991 
4992 	return -EIO;
4993 }
4994 
4995 #ifdef CONFIG_KVM_XICS
4996 
4997 void kvmppc_free_pimap(struct kvm *kvm)
4998 {
4999 	kfree(kvm->arch.pimap);
5000 }
5001 
5002 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5003 {
5004 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5005 }
5006 
5007 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5008 {
5009 	struct irq_desc *desc;
5010 	struct kvmppc_irq_map *irq_map;
5011 	struct kvmppc_passthru_irqmap *pimap;
5012 	struct irq_chip *chip;
5013 	int i, rc = 0;
5014 
5015 	if (!kvm_irq_bypass)
5016 		return 1;
5017 
5018 	desc = irq_to_desc(host_irq);
5019 	if (!desc)
5020 		return -EIO;
5021 
5022 	mutex_lock(&kvm->lock);
5023 
5024 	pimap = kvm->arch.pimap;
5025 	if (pimap == NULL) {
5026 		/* First call, allocate structure to hold IRQ map */
5027 		pimap = kvmppc_alloc_pimap();
5028 		if (pimap == NULL) {
5029 			mutex_unlock(&kvm->lock);
5030 			return -ENOMEM;
5031 		}
5032 		kvm->arch.pimap = pimap;
5033 	}
5034 
5035 	/*
5036 	 * For now, we only support interrupts for which the EOI operation
5037 	 * is an OPAL call followed by a write to XIRR, since that's
5038 	 * what our real-mode EOI code does, or a XIVE interrupt
5039 	 */
5040 	chip = irq_data_get_irq_chip(&desc->irq_data);
5041 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5042 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5043 			host_irq, guest_gsi);
5044 		mutex_unlock(&kvm->lock);
5045 		return -ENOENT;
5046 	}
5047 
5048 	/*
5049 	 * See if we already have an entry for this guest IRQ number.
5050 	 * If it's mapped to a hardware IRQ number, that's an error,
5051 	 * otherwise re-use this entry.
5052 	 */
5053 	for (i = 0; i < pimap->n_mapped; i++) {
5054 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5055 			if (pimap->mapped[i].r_hwirq) {
5056 				mutex_unlock(&kvm->lock);
5057 				return -EINVAL;
5058 			}
5059 			break;
5060 		}
5061 	}
5062 
5063 	if (i == KVMPPC_PIRQ_MAPPED) {
5064 		mutex_unlock(&kvm->lock);
5065 		return -EAGAIN;		/* table is full */
5066 	}
5067 
5068 	irq_map = &pimap->mapped[i];
5069 
5070 	irq_map->v_hwirq = guest_gsi;
5071 	irq_map->desc = desc;
5072 
5073 	/*
5074 	 * Order the above two stores before the next to serialize with
5075 	 * the KVM real mode handler.
5076 	 */
5077 	smp_wmb();
5078 	irq_map->r_hwirq = desc->irq_data.hwirq;
5079 
5080 	if (i == pimap->n_mapped)
5081 		pimap->n_mapped++;
5082 
5083 	if (xics_on_xive())
5084 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5085 	else
5086 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5087 	if (rc)
5088 		irq_map->r_hwirq = 0;
5089 
5090 	mutex_unlock(&kvm->lock);
5091 
5092 	return 0;
5093 }
5094 
5095 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5096 {
5097 	struct irq_desc *desc;
5098 	struct kvmppc_passthru_irqmap *pimap;
5099 	int i, rc = 0;
5100 
5101 	if (!kvm_irq_bypass)
5102 		return 0;
5103 
5104 	desc = irq_to_desc(host_irq);
5105 	if (!desc)
5106 		return -EIO;
5107 
5108 	mutex_lock(&kvm->lock);
5109 	if (!kvm->arch.pimap)
5110 		goto unlock;
5111 
5112 	pimap = kvm->arch.pimap;
5113 
5114 	for (i = 0; i < pimap->n_mapped; i++) {
5115 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5116 			break;
5117 	}
5118 
5119 	if (i == pimap->n_mapped) {
5120 		mutex_unlock(&kvm->lock);
5121 		return -ENODEV;
5122 	}
5123 
5124 	if (xics_on_xive())
5125 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5126 	else
5127 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5128 
5129 	/* invalidate the entry (what do do on error from the above ?) */
5130 	pimap->mapped[i].r_hwirq = 0;
5131 
5132 	/*
5133 	 * We don't free this structure even when the count goes to
5134 	 * zero. The structure is freed when we destroy the VM.
5135 	 */
5136  unlock:
5137 	mutex_unlock(&kvm->lock);
5138 	return rc;
5139 }
5140 
5141 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5142 					     struct irq_bypass_producer *prod)
5143 {
5144 	int ret = 0;
5145 	struct kvm_kernel_irqfd *irqfd =
5146 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5147 
5148 	irqfd->producer = prod;
5149 
5150 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5151 	if (ret)
5152 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5153 			prod->irq, irqfd->gsi, ret);
5154 
5155 	return ret;
5156 }
5157 
5158 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5159 					      struct irq_bypass_producer *prod)
5160 {
5161 	int ret;
5162 	struct kvm_kernel_irqfd *irqfd =
5163 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5164 
5165 	irqfd->producer = NULL;
5166 
5167 	/*
5168 	 * When producer of consumer is unregistered, we change back to
5169 	 * default external interrupt handling mode - KVM real mode
5170 	 * will switch back to host.
5171 	 */
5172 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5173 	if (ret)
5174 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5175 			prod->irq, irqfd->gsi, ret);
5176 }
5177 #endif
5178 
5179 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5180 				 unsigned int ioctl, unsigned long arg)
5181 {
5182 	struct kvm *kvm __maybe_unused = filp->private_data;
5183 	void __user *argp = (void __user *)arg;
5184 	long r;
5185 
5186 	switch (ioctl) {
5187 
5188 	case KVM_PPC_ALLOCATE_HTAB: {
5189 		u32 htab_order;
5190 
5191 		r = -EFAULT;
5192 		if (get_user(htab_order, (u32 __user *)argp))
5193 			break;
5194 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5195 		if (r)
5196 			break;
5197 		r = 0;
5198 		break;
5199 	}
5200 
5201 	case KVM_PPC_GET_HTAB_FD: {
5202 		struct kvm_get_htab_fd ghf;
5203 
5204 		r = -EFAULT;
5205 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5206 			break;
5207 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5208 		break;
5209 	}
5210 
5211 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5212 		struct kvm_ppc_resize_hpt rhpt;
5213 
5214 		r = -EFAULT;
5215 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5216 			break;
5217 
5218 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5219 		break;
5220 	}
5221 
5222 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5223 		struct kvm_ppc_resize_hpt rhpt;
5224 
5225 		r = -EFAULT;
5226 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5227 			break;
5228 
5229 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5230 		break;
5231 	}
5232 
5233 	default:
5234 		r = -ENOTTY;
5235 	}
5236 
5237 	return r;
5238 }
5239 
5240 /*
5241  * List of hcall numbers to enable by default.
5242  * For compatibility with old userspace, we enable by default
5243  * all hcalls that were implemented before the hcall-enabling
5244  * facility was added.  Note this list should not include H_RTAS.
5245  */
5246 static unsigned int default_hcall_list[] = {
5247 	H_REMOVE,
5248 	H_ENTER,
5249 	H_READ,
5250 	H_PROTECT,
5251 	H_BULK_REMOVE,
5252 	H_GET_TCE,
5253 	H_PUT_TCE,
5254 	H_SET_DABR,
5255 	H_SET_XDABR,
5256 	H_CEDE,
5257 	H_PROD,
5258 	H_CONFER,
5259 	H_REGISTER_VPA,
5260 #ifdef CONFIG_KVM_XICS
5261 	H_EOI,
5262 	H_CPPR,
5263 	H_IPI,
5264 	H_IPOLL,
5265 	H_XIRR,
5266 	H_XIRR_X,
5267 #endif
5268 	0
5269 };
5270 
5271 static void init_default_hcalls(void)
5272 {
5273 	int i;
5274 	unsigned int hcall;
5275 
5276 	for (i = 0; default_hcall_list[i]; ++i) {
5277 		hcall = default_hcall_list[i];
5278 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5279 		__set_bit(hcall / 4, default_enabled_hcalls);
5280 	}
5281 }
5282 
5283 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5284 {
5285 	unsigned long lpcr;
5286 	int radix;
5287 	int err;
5288 
5289 	/* If not on a POWER9, reject it */
5290 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5291 		return -ENODEV;
5292 
5293 	/* If any unknown flags set, reject it */
5294 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5295 		return -EINVAL;
5296 
5297 	/* GR (guest radix) bit in process_table field must match */
5298 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5299 	if (!!(cfg->process_table & PATB_GR) != radix)
5300 		return -EINVAL;
5301 
5302 	/* Process table size field must be reasonable, i.e. <= 24 */
5303 	if ((cfg->process_table & PRTS_MASK) > 24)
5304 		return -EINVAL;
5305 
5306 	/* We can change a guest to/from radix now, if the host is radix */
5307 	if (radix && !radix_enabled())
5308 		return -EINVAL;
5309 
5310 	/* If we're a nested hypervisor, we currently only support radix */
5311 	if (kvmhv_on_pseries() && !radix)
5312 		return -EINVAL;
5313 
5314 	mutex_lock(&kvm->arch.mmu_setup_lock);
5315 	if (radix != kvm_is_radix(kvm)) {
5316 		if (kvm->arch.mmu_ready) {
5317 			kvm->arch.mmu_ready = 0;
5318 			/* order mmu_ready vs. vcpus_running */
5319 			smp_mb();
5320 			if (atomic_read(&kvm->arch.vcpus_running)) {
5321 				kvm->arch.mmu_ready = 1;
5322 				err = -EBUSY;
5323 				goto out_unlock;
5324 			}
5325 		}
5326 		if (radix)
5327 			err = kvmppc_switch_mmu_to_radix(kvm);
5328 		else
5329 			err = kvmppc_switch_mmu_to_hpt(kvm);
5330 		if (err)
5331 			goto out_unlock;
5332 	}
5333 
5334 	kvm->arch.process_table = cfg->process_table;
5335 	kvmppc_setup_partition_table(kvm);
5336 
5337 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5338 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5339 	err = 0;
5340 
5341  out_unlock:
5342 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5343 	return err;
5344 }
5345 
5346 static int kvmhv_enable_nested(struct kvm *kvm)
5347 {
5348 	if (!nested)
5349 		return -EPERM;
5350 	if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5351 		return -ENODEV;
5352 
5353 	/* kvm == NULL means the caller is testing if the capability exists */
5354 	if (kvm)
5355 		kvm->arch.nested_enable = true;
5356 	return 0;
5357 }
5358 
5359 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5360 				 int size)
5361 {
5362 	int rc = -EINVAL;
5363 
5364 	if (kvmhv_vcpu_is_radix(vcpu)) {
5365 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5366 
5367 		if (rc > 0)
5368 			rc = -EINVAL;
5369 	}
5370 
5371 	/* For now quadrants are the only way to access nested guest memory */
5372 	if (rc && vcpu->arch.nested)
5373 		rc = -EAGAIN;
5374 
5375 	return rc;
5376 }
5377 
5378 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5379 				int size)
5380 {
5381 	int rc = -EINVAL;
5382 
5383 	if (kvmhv_vcpu_is_radix(vcpu)) {
5384 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5385 
5386 		if (rc > 0)
5387 			rc = -EINVAL;
5388 	}
5389 
5390 	/* For now quadrants are the only way to access nested guest memory */
5391 	if (rc && vcpu->arch.nested)
5392 		rc = -EAGAIN;
5393 
5394 	return rc;
5395 }
5396 
5397 static struct kvmppc_ops kvm_ops_hv = {
5398 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5399 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5400 	.get_one_reg = kvmppc_get_one_reg_hv,
5401 	.set_one_reg = kvmppc_set_one_reg_hv,
5402 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
5403 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
5404 	.set_msr     = kvmppc_set_msr_hv,
5405 	.vcpu_run    = kvmppc_vcpu_run_hv,
5406 	.vcpu_create = kvmppc_core_vcpu_create_hv,
5407 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
5408 	.check_requests = kvmppc_core_check_requests_hv,
5409 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5410 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
5411 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5412 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5413 	.unmap_hva_range = kvm_unmap_hva_range_hv,
5414 	.age_hva  = kvm_age_hva_hv,
5415 	.test_age_hva = kvm_test_age_hva_hv,
5416 	.set_spte_hva = kvm_set_spte_hva_hv,
5417 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
5418 	.free_memslot = kvmppc_core_free_memslot_hv,
5419 	.create_memslot = kvmppc_core_create_memslot_hv,
5420 	.init_vm =  kvmppc_core_init_vm_hv,
5421 	.destroy_vm = kvmppc_core_destroy_vm_hv,
5422 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5423 	.emulate_op = kvmppc_core_emulate_op_hv,
5424 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5425 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5426 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5427 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5428 	.hcall_implemented = kvmppc_hcall_impl_hv,
5429 #ifdef CONFIG_KVM_XICS
5430 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5431 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5432 #endif
5433 	.configure_mmu = kvmhv_configure_mmu,
5434 	.get_rmmu_info = kvmhv_get_rmmu_info,
5435 	.set_smt_mode = kvmhv_set_smt_mode,
5436 	.enable_nested = kvmhv_enable_nested,
5437 	.load_from_eaddr = kvmhv_load_from_eaddr,
5438 	.store_to_eaddr = kvmhv_store_to_eaddr,
5439 };
5440 
5441 static int kvm_init_subcore_bitmap(void)
5442 {
5443 	int i, j;
5444 	int nr_cores = cpu_nr_cores();
5445 	struct sibling_subcore_state *sibling_subcore_state;
5446 
5447 	for (i = 0; i < nr_cores; i++) {
5448 		int first_cpu = i * threads_per_core;
5449 		int node = cpu_to_node(first_cpu);
5450 
5451 		/* Ignore if it is already allocated. */
5452 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
5453 			continue;
5454 
5455 		sibling_subcore_state =
5456 			kzalloc_node(sizeof(struct sibling_subcore_state),
5457 							GFP_KERNEL, node);
5458 		if (!sibling_subcore_state)
5459 			return -ENOMEM;
5460 
5461 
5462 		for (j = 0; j < threads_per_core; j++) {
5463 			int cpu = first_cpu + j;
5464 
5465 			paca_ptrs[cpu]->sibling_subcore_state =
5466 						sibling_subcore_state;
5467 		}
5468 	}
5469 	return 0;
5470 }
5471 
5472 static int kvmppc_radix_possible(void)
5473 {
5474 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5475 }
5476 
5477 static int kvmppc_book3s_init_hv(void)
5478 {
5479 	int r;
5480 
5481 	if (!tlbie_capable) {
5482 		pr_err("KVM-HV: Host does not support TLBIE\n");
5483 		return -ENODEV;
5484 	}
5485 
5486 	/*
5487 	 * FIXME!! Do we need to check on all cpus ?
5488 	 */
5489 	r = kvmppc_core_check_processor_compat_hv();
5490 	if (r < 0)
5491 		return -ENODEV;
5492 
5493 	r = kvmhv_nested_init();
5494 	if (r)
5495 		return r;
5496 
5497 	r = kvm_init_subcore_bitmap();
5498 	if (r)
5499 		return r;
5500 
5501 	/*
5502 	 * We need a way of accessing the XICS interrupt controller,
5503 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5504 	 * indirectly, via OPAL.
5505 	 */
5506 #ifdef CONFIG_SMP
5507 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
5508 	    !local_paca->kvm_hstate.xics_phys) {
5509 		struct device_node *np;
5510 
5511 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5512 		if (!np) {
5513 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5514 			return -ENODEV;
5515 		}
5516 		/* presence of intc confirmed - node can be dropped again */
5517 		of_node_put(np);
5518 	}
5519 #endif
5520 
5521 	kvm_ops_hv.owner = THIS_MODULE;
5522 	kvmppc_hv_ops = &kvm_ops_hv;
5523 
5524 	init_default_hcalls();
5525 
5526 	init_vcore_lists();
5527 
5528 	r = kvmppc_mmu_hv_init();
5529 	if (r)
5530 		return r;
5531 
5532 	if (kvmppc_radix_possible())
5533 		r = kvmppc_radix_init();
5534 
5535 	/*
5536 	 * POWER9 chips before version 2.02 can't have some threads in
5537 	 * HPT mode and some in radix mode on the same core.
5538 	 */
5539 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5540 		unsigned int pvr = mfspr(SPRN_PVR);
5541 		if ((pvr >> 16) == PVR_POWER9 &&
5542 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5543 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5544 			no_mixing_hpt_and_radix = true;
5545 	}
5546 
5547 	return r;
5548 }
5549 
5550 static void kvmppc_book3s_exit_hv(void)
5551 {
5552 	kvmppc_free_host_rm_ops();
5553 	if (kvmppc_radix_possible())
5554 		kvmppc_radix_exit();
5555 	kvmppc_hv_ops = NULL;
5556 	kvmhv_nested_exit();
5557 }
5558 
5559 module_init(kvmppc_book3s_init_hv);
5560 module_exit(kvmppc_book3s_exit_hv);
5561 MODULE_LICENSE("GPL");
5562 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5563 MODULE_ALIAS("devname:kvm");
5564