xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 4fc4dca8)
1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Paul Mackerras <paulus@au1.ibm.com>
7  *    Alexander Graf <agraf@suse.de>
8  *    Kevin Wolf <mail@kevin-wolf.de>
9  *
10  * Description: KVM functions specific to running on Book 3S
11  * processors in hypervisor mode (specifically POWER7 and later).
12  *
13  * This file is derived from arch/powerpc/kvm/book3s.c,
14  * by Alexander Graf <agraf@suse.de>.
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License, version 2, as
18  * published by the Free Software Foundation.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/kernel.h>
23 #include <linux/err.h>
24 #include <linux/slab.h>
25 #include <linux/preempt.h>
26 #include <linux/sched/signal.h>
27 #include <linux/sched/stat.h>
28 #include <linux/delay.h>
29 #include <linux/export.h>
30 #include <linux/fs.h>
31 #include <linux/anon_inodes.h>
32 #include <linux/cpu.h>
33 #include <linux/cpumask.h>
34 #include <linux/spinlock.h>
35 #include <linux/page-flags.h>
36 #include <linux/srcu.h>
37 #include <linux/miscdevice.h>
38 #include <linux/debugfs.h>
39 #include <linux/gfp.h>
40 #include <linux/vmalloc.h>
41 #include <linux/highmem.h>
42 #include <linux/hugetlb.h>
43 #include <linux/kvm_irqfd.h>
44 #include <linux/irqbypass.h>
45 #include <linux/module.h>
46 #include <linux/compiler.h>
47 #include <linux/of.h>
48 
49 #include <asm/ftrace.h>
50 #include <asm/reg.h>
51 #include <asm/ppc-opcode.h>
52 #include <asm/asm-prototypes.h>
53 #include <asm/archrandom.h>
54 #include <asm/debug.h>
55 #include <asm/disassemble.h>
56 #include <asm/cputable.h>
57 #include <asm/cacheflush.h>
58 #include <linux/uaccess.h>
59 #include <asm/io.h>
60 #include <asm/kvm_ppc.h>
61 #include <asm/kvm_book3s.h>
62 #include <asm/mmu_context.h>
63 #include <asm/lppaca.h>
64 #include <asm/processor.h>
65 #include <asm/cputhreads.h>
66 #include <asm/page.h>
67 #include <asm/hvcall.h>
68 #include <asm/switch_to.h>
69 #include <asm/smp.h>
70 #include <asm/dbell.h>
71 #include <asm/hmi.h>
72 #include <asm/pnv-pci.h>
73 #include <asm/mmu.h>
74 #include <asm/opal.h>
75 #include <asm/xics.h>
76 #include <asm/xive.h>
77 #include <asm/hw_breakpoint.h>
78 
79 #include "book3s.h"
80 
81 #define CREATE_TRACE_POINTS
82 #include "trace_hv.h"
83 
84 /* #define EXIT_DEBUG */
85 /* #define EXIT_DEBUG_SIMPLE */
86 /* #define EXIT_DEBUG_INT */
87 
88 /* Used to indicate that a guest page fault needs to be handled */
89 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
90 /* Used to indicate that a guest passthrough interrupt needs to be handled */
91 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
92 
93 /* Used as a "null" value for timebase values */
94 #define TB_NIL	(~(u64)0)
95 
96 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
97 
98 static int dynamic_mt_modes = 6;
99 module_param(dynamic_mt_modes, int, 0644);
100 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
101 static int target_smt_mode;
102 module_param(target_smt_mode, int, 0644);
103 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
104 
105 static bool indep_threads_mode = true;
106 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
107 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
108 
109 static bool one_vm_per_core;
110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
112 
113 #ifdef CONFIG_KVM_XICS
114 static struct kernel_param_ops module_param_ops = {
115 	.set = param_set_int,
116 	.get = param_get_int,
117 };
118 
119 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
120 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
121 
122 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
123 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
124 #endif
125 
126 /* If set, guests are allowed to create and control nested guests */
127 static bool nested = true;
128 module_param(nested, bool, S_IRUGO | S_IWUSR);
129 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
130 
131 static inline bool nesting_enabled(struct kvm *kvm)
132 {
133 	return kvm->arch.nested_enable && kvm_is_radix(kvm);
134 }
135 
136 /* If set, the threads on each CPU core have to be in the same MMU mode */
137 static bool no_mixing_hpt_and_radix;
138 
139 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
140 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
141 
142 /*
143  * RWMR values for POWER8.  These control the rate at which PURR
144  * and SPURR count and should be set according to the number of
145  * online threads in the vcore being run.
146  */
147 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
148 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
149 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
150 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
151 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
152 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
153 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
154 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
155 
156 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
157 	RWMR_RPA_P8_1THREAD,
158 	RWMR_RPA_P8_1THREAD,
159 	RWMR_RPA_P8_2THREAD,
160 	RWMR_RPA_P8_3THREAD,
161 	RWMR_RPA_P8_4THREAD,
162 	RWMR_RPA_P8_5THREAD,
163 	RWMR_RPA_P8_6THREAD,
164 	RWMR_RPA_P8_7THREAD,
165 	RWMR_RPA_P8_8THREAD,
166 };
167 
168 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
169 		int *ip)
170 {
171 	int i = *ip;
172 	struct kvm_vcpu *vcpu;
173 
174 	while (++i < MAX_SMT_THREADS) {
175 		vcpu = READ_ONCE(vc->runnable_threads[i]);
176 		if (vcpu) {
177 			*ip = i;
178 			return vcpu;
179 		}
180 	}
181 	return NULL;
182 }
183 
184 /* Used to traverse the list of runnable threads for a given vcore */
185 #define for_each_runnable_thread(i, vcpu, vc) \
186 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
187 
188 static bool kvmppc_ipi_thread(int cpu)
189 {
190 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
191 
192 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
193 	if (kvmhv_on_pseries())
194 		return false;
195 
196 	/* On POWER9 we can use msgsnd to IPI any cpu */
197 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
198 		msg |= get_hard_smp_processor_id(cpu);
199 		smp_mb();
200 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
201 		return true;
202 	}
203 
204 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
205 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
206 		preempt_disable();
207 		if (cpu_first_thread_sibling(cpu) ==
208 		    cpu_first_thread_sibling(smp_processor_id())) {
209 			msg |= cpu_thread_in_core(cpu);
210 			smp_mb();
211 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
212 			preempt_enable();
213 			return true;
214 		}
215 		preempt_enable();
216 	}
217 
218 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
219 	if (cpu >= 0 && cpu < nr_cpu_ids) {
220 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
221 			xics_wake_cpu(cpu);
222 			return true;
223 		}
224 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
225 		return true;
226 	}
227 #endif
228 
229 	return false;
230 }
231 
232 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
233 {
234 	int cpu;
235 	struct swait_queue_head *wqp;
236 
237 	wqp = kvm_arch_vcpu_wq(vcpu);
238 	if (swq_has_sleeper(wqp)) {
239 		swake_up_one(wqp);
240 		++vcpu->stat.halt_wakeup;
241 	}
242 
243 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
244 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
245 		return;
246 
247 	/* CPU points to the first thread of the core */
248 	cpu = vcpu->cpu;
249 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
250 		smp_send_reschedule(cpu);
251 }
252 
253 /*
254  * We use the vcpu_load/put functions to measure stolen time.
255  * Stolen time is counted as time when either the vcpu is able to
256  * run as part of a virtual core, but the task running the vcore
257  * is preempted or sleeping, or when the vcpu needs something done
258  * in the kernel by the task running the vcpu, but that task is
259  * preempted or sleeping.  Those two things have to be counted
260  * separately, since one of the vcpu tasks will take on the job
261  * of running the core, and the other vcpu tasks in the vcore will
262  * sleep waiting for it to do that, but that sleep shouldn't count
263  * as stolen time.
264  *
265  * Hence we accumulate stolen time when the vcpu can run as part of
266  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
267  * needs its task to do other things in the kernel (for example,
268  * service a page fault) in busy_stolen.  We don't accumulate
269  * stolen time for a vcore when it is inactive, or for a vcpu
270  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
271  * a misnomer; it means that the vcpu task is not executing in
272  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
273  * the kernel.  We don't have any way of dividing up that time
274  * between time that the vcpu is genuinely stopped, time that
275  * the task is actively working on behalf of the vcpu, and time
276  * that the task is preempted, so we don't count any of it as
277  * stolen.
278  *
279  * Updates to busy_stolen are protected by arch.tbacct_lock;
280  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
281  * lock.  The stolen times are measured in units of timebase ticks.
282  * (Note that the != TB_NIL checks below are purely defensive;
283  * they should never fail.)
284  */
285 
286 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
287 {
288 	unsigned long flags;
289 
290 	spin_lock_irqsave(&vc->stoltb_lock, flags);
291 	vc->preempt_tb = mftb();
292 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
293 }
294 
295 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
296 {
297 	unsigned long flags;
298 
299 	spin_lock_irqsave(&vc->stoltb_lock, flags);
300 	if (vc->preempt_tb != TB_NIL) {
301 		vc->stolen_tb += mftb() - vc->preempt_tb;
302 		vc->preempt_tb = TB_NIL;
303 	}
304 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
305 }
306 
307 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
308 {
309 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
310 	unsigned long flags;
311 
312 	/*
313 	 * We can test vc->runner without taking the vcore lock,
314 	 * because only this task ever sets vc->runner to this
315 	 * vcpu, and once it is set to this vcpu, only this task
316 	 * ever sets it to NULL.
317 	 */
318 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
319 		kvmppc_core_end_stolen(vc);
320 
321 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
322 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
323 	    vcpu->arch.busy_preempt != TB_NIL) {
324 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
325 		vcpu->arch.busy_preempt = TB_NIL;
326 	}
327 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
328 }
329 
330 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
331 {
332 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
333 	unsigned long flags;
334 
335 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
336 		kvmppc_core_start_stolen(vc);
337 
338 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
339 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
340 		vcpu->arch.busy_preempt = mftb();
341 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
342 }
343 
344 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
345 {
346 	/*
347 	 * Check for illegal transactional state bit combination
348 	 * and if we find it, force the TS field to a safe state.
349 	 */
350 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
351 		msr &= ~MSR_TS_MASK;
352 	vcpu->arch.shregs.msr = msr;
353 	kvmppc_end_cede(vcpu);
354 }
355 
356 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
357 {
358 	vcpu->arch.pvr = pvr;
359 }
360 
361 /* Dummy value used in computing PCR value below */
362 #define PCR_ARCH_300	(PCR_ARCH_207 << 1)
363 
364 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
365 {
366 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
367 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
368 
369 	/* We can (emulate) our own architecture version and anything older */
370 	if (cpu_has_feature(CPU_FTR_ARCH_300))
371 		host_pcr_bit = PCR_ARCH_300;
372 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
373 		host_pcr_bit = PCR_ARCH_207;
374 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
375 		host_pcr_bit = PCR_ARCH_206;
376 	else
377 		host_pcr_bit = PCR_ARCH_205;
378 
379 	/* Determine lowest PCR bit needed to run guest in given PVR level */
380 	guest_pcr_bit = host_pcr_bit;
381 	if (arch_compat) {
382 		switch (arch_compat) {
383 		case PVR_ARCH_205:
384 			guest_pcr_bit = PCR_ARCH_205;
385 			break;
386 		case PVR_ARCH_206:
387 		case PVR_ARCH_206p:
388 			guest_pcr_bit = PCR_ARCH_206;
389 			break;
390 		case PVR_ARCH_207:
391 			guest_pcr_bit = PCR_ARCH_207;
392 			break;
393 		case PVR_ARCH_300:
394 			guest_pcr_bit = PCR_ARCH_300;
395 			break;
396 		default:
397 			return -EINVAL;
398 		}
399 	}
400 
401 	/* Check requested PCR bits don't exceed our capabilities */
402 	if (guest_pcr_bit > host_pcr_bit)
403 		return -EINVAL;
404 
405 	spin_lock(&vc->lock);
406 	vc->arch_compat = arch_compat;
407 	/* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
408 	vc->pcr = host_pcr_bit - guest_pcr_bit;
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 	struct kvm_vcpu *ret;
450 
451 	mutex_lock(&kvm->lock);
452 	ret = kvm_get_vcpu_by_id(kvm, id);
453 	mutex_unlock(&kvm->lock);
454 	return ret;
455 }
456 
457 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
458 {
459 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
460 	vpa->yield_count = cpu_to_be32(1);
461 }
462 
463 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
464 		   unsigned long addr, unsigned long len)
465 {
466 	/* check address is cacheline aligned */
467 	if (addr & (L1_CACHE_BYTES - 1))
468 		return -EINVAL;
469 	spin_lock(&vcpu->arch.vpa_update_lock);
470 	if (v->next_gpa != addr || v->len != len) {
471 		v->next_gpa = addr;
472 		v->len = addr ? len : 0;
473 		v->update_pending = 1;
474 	}
475 	spin_unlock(&vcpu->arch.vpa_update_lock);
476 	return 0;
477 }
478 
479 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
480 struct reg_vpa {
481 	u32 dummy;
482 	union {
483 		__be16 hword;
484 		__be32 word;
485 	} length;
486 };
487 
488 static int vpa_is_registered(struct kvmppc_vpa *vpap)
489 {
490 	if (vpap->update_pending)
491 		return vpap->next_gpa != 0;
492 	return vpap->pinned_addr != NULL;
493 }
494 
495 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
496 				       unsigned long flags,
497 				       unsigned long vcpuid, unsigned long vpa)
498 {
499 	struct kvm *kvm = vcpu->kvm;
500 	unsigned long len, nb;
501 	void *va;
502 	struct kvm_vcpu *tvcpu;
503 	int err;
504 	int subfunc;
505 	struct kvmppc_vpa *vpap;
506 
507 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
508 	if (!tvcpu)
509 		return H_PARAMETER;
510 
511 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
512 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
513 	    subfunc == H_VPA_REG_SLB) {
514 		/* Registering new area - address must be cache-line aligned */
515 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
516 			return H_PARAMETER;
517 
518 		/* convert logical addr to kernel addr and read length */
519 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
520 		if (va == NULL)
521 			return H_PARAMETER;
522 		if (subfunc == H_VPA_REG_VPA)
523 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
524 		else
525 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
526 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
527 
528 		/* Check length */
529 		if (len > nb || len < sizeof(struct reg_vpa))
530 			return H_PARAMETER;
531 	} else {
532 		vpa = 0;
533 		len = 0;
534 	}
535 
536 	err = H_PARAMETER;
537 	vpap = NULL;
538 	spin_lock(&tvcpu->arch.vpa_update_lock);
539 
540 	switch (subfunc) {
541 	case H_VPA_REG_VPA:		/* register VPA */
542 		/*
543 		 * The size of our lppaca is 1kB because of the way we align
544 		 * it for the guest to avoid crossing a 4kB boundary. We only
545 		 * use 640 bytes of the structure though, so we should accept
546 		 * clients that set a size of 640.
547 		 */
548 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
549 		if (len < sizeof(struct lppaca))
550 			break;
551 		vpap = &tvcpu->arch.vpa;
552 		err = 0;
553 		break;
554 
555 	case H_VPA_REG_DTL:		/* register DTL */
556 		if (len < sizeof(struct dtl_entry))
557 			break;
558 		len -= len % sizeof(struct dtl_entry);
559 
560 		/* Check that they have previously registered a VPA */
561 		err = H_RESOURCE;
562 		if (!vpa_is_registered(&tvcpu->arch.vpa))
563 			break;
564 
565 		vpap = &tvcpu->arch.dtl;
566 		err = 0;
567 		break;
568 
569 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
570 		/* Check that they have previously registered a VPA */
571 		err = H_RESOURCE;
572 		if (!vpa_is_registered(&tvcpu->arch.vpa))
573 			break;
574 
575 		vpap = &tvcpu->arch.slb_shadow;
576 		err = 0;
577 		break;
578 
579 	case H_VPA_DEREG_VPA:		/* deregister VPA */
580 		/* Check they don't still have a DTL or SLB buf registered */
581 		err = H_RESOURCE;
582 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
583 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
584 			break;
585 
586 		vpap = &tvcpu->arch.vpa;
587 		err = 0;
588 		break;
589 
590 	case H_VPA_DEREG_DTL:		/* deregister DTL */
591 		vpap = &tvcpu->arch.dtl;
592 		err = 0;
593 		break;
594 
595 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
596 		vpap = &tvcpu->arch.slb_shadow;
597 		err = 0;
598 		break;
599 	}
600 
601 	if (vpap) {
602 		vpap->next_gpa = vpa;
603 		vpap->len = len;
604 		vpap->update_pending = 1;
605 	}
606 
607 	spin_unlock(&tvcpu->arch.vpa_update_lock);
608 
609 	return err;
610 }
611 
612 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
613 {
614 	struct kvm *kvm = vcpu->kvm;
615 	void *va;
616 	unsigned long nb;
617 	unsigned long gpa;
618 
619 	/*
620 	 * We need to pin the page pointed to by vpap->next_gpa,
621 	 * but we can't call kvmppc_pin_guest_page under the lock
622 	 * as it does get_user_pages() and down_read().  So we
623 	 * have to drop the lock, pin the page, then get the lock
624 	 * again and check that a new area didn't get registered
625 	 * in the meantime.
626 	 */
627 	for (;;) {
628 		gpa = vpap->next_gpa;
629 		spin_unlock(&vcpu->arch.vpa_update_lock);
630 		va = NULL;
631 		nb = 0;
632 		if (gpa)
633 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
634 		spin_lock(&vcpu->arch.vpa_update_lock);
635 		if (gpa == vpap->next_gpa)
636 			break;
637 		/* sigh... unpin that one and try again */
638 		if (va)
639 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
640 	}
641 
642 	vpap->update_pending = 0;
643 	if (va && nb < vpap->len) {
644 		/*
645 		 * If it's now too short, it must be that userspace
646 		 * has changed the mappings underlying guest memory,
647 		 * so unregister the region.
648 		 */
649 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
650 		va = NULL;
651 	}
652 	if (vpap->pinned_addr)
653 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
654 					vpap->dirty);
655 	vpap->gpa = gpa;
656 	vpap->pinned_addr = va;
657 	vpap->dirty = false;
658 	if (va)
659 		vpap->pinned_end = va + vpap->len;
660 }
661 
662 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
663 {
664 	if (!(vcpu->arch.vpa.update_pending ||
665 	      vcpu->arch.slb_shadow.update_pending ||
666 	      vcpu->arch.dtl.update_pending))
667 		return;
668 
669 	spin_lock(&vcpu->arch.vpa_update_lock);
670 	if (vcpu->arch.vpa.update_pending) {
671 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
672 		if (vcpu->arch.vpa.pinned_addr)
673 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
674 	}
675 	if (vcpu->arch.dtl.update_pending) {
676 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
677 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
678 		vcpu->arch.dtl_index = 0;
679 	}
680 	if (vcpu->arch.slb_shadow.update_pending)
681 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
682 	spin_unlock(&vcpu->arch.vpa_update_lock);
683 }
684 
685 /*
686  * Return the accumulated stolen time for the vcore up until `now'.
687  * The caller should hold the vcore lock.
688  */
689 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
690 {
691 	u64 p;
692 	unsigned long flags;
693 
694 	spin_lock_irqsave(&vc->stoltb_lock, flags);
695 	p = vc->stolen_tb;
696 	if (vc->vcore_state != VCORE_INACTIVE &&
697 	    vc->preempt_tb != TB_NIL)
698 		p += now - vc->preempt_tb;
699 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
700 	return p;
701 }
702 
703 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
704 				    struct kvmppc_vcore *vc)
705 {
706 	struct dtl_entry *dt;
707 	struct lppaca *vpa;
708 	unsigned long stolen;
709 	unsigned long core_stolen;
710 	u64 now;
711 	unsigned long flags;
712 
713 	dt = vcpu->arch.dtl_ptr;
714 	vpa = vcpu->arch.vpa.pinned_addr;
715 	now = mftb();
716 	core_stolen = vcore_stolen_time(vc, now);
717 	stolen = core_stolen - vcpu->arch.stolen_logged;
718 	vcpu->arch.stolen_logged = core_stolen;
719 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
720 	stolen += vcpu->arch.busy_stolen;
721 	vcpu->arch.busy_stolen = 0;
722 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
723 	if (!dt || !vpa)
724 		return;
725 	memset(dt, 0, sizeof(struct dtl_entry));
726 	dt->dispatch_reason = 7;
727 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
728 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
729 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
730 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
731 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
732 	++dt;
733 	if (dt == vcpu->arch.dtl.pinned_end)
734 		dt = vcpu->arch.dtl.pinned_addr;
735 	vcpu->arch.dtl_ptr = dt;
736 	/* order writing *dt vs. writing vpa->dtl_idx */
737 	smp_wmb();
738 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
739 	vcpu->arch.dtl.dirty = true;
740 }
741 
742 /* See if there is a doorbell interrupt pending for a vcpu */
743 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
744 {
745 	int thr;
746 	struct kvmppc_vcore *vc;
747 
748 	if (vcpu->arch.doorbell_request)
749 		return true;
750 	/*
751 	 * Ensure that the read of vcore->dpdes comes after the read
752 	 * of vcpu->doorbell_request.  This barrier matches the
753 	 * smp_wmb() in kvmppc_guest_entry_inject().
754 	 */
755 	smp_rmb();
756 	vc = vcpu->arch.vcore;
757 	thr = vcpu->vcpu_id - vc->first_vcpuid;
758 	return !!(vc->dpdes & (1 << thr));
759 }
760 
761 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
762 {
763 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
764 		return true;
765 	if ((!vcpu->arch.vcore->arch_compat) &&
766 	    cpu_has_feature(CPU_FTR_ARCH_207S))
767 		return true;
768 	return false;
769 }
770 
771 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
772 			     unsigned long resource, unsigned long value1,
773 			     unsigned long value2)
774 {
775 	switch (resource) {
776 	case H_SET_MODE_RESOURCE_SET_CIABR:
777 		if (!kvmppc_power8_compatible(vcpu))
778 			return H_P2;
779 		if (value2)
780 			return H_P4;
781 		if (mflags)
782 			return H_UNSUPPORTED_FLAG_START;
783 		/* Guests can't breakpoint the hypervisor */
784 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
785 			return H_P3;
786 		vcpu->arch.ciabr  = value1;
787 		return H_SUCCESS;
788 	case H_SET_MODE_RESOURCE_SET_DAWR:
789 		if (!kvmppc_power8_compatible(vcpu))
790 			return H_P2;
791 		if (!ppc_breakpoint_available())
792 			return H_P2;
793 		if (mflags)
794 			return H_UNSUPPORTED_FLAG_START;
795 		if (value2 & DABRX_HYP)
796 			return H_P4;
797 		vcpu->arch.dawr  = value1;
798 		vcpu->arch.dawrx = value2;
799 		return H_SUCCESS;
800 	default:
801 		return H_TOO_HARD;
802 	}
803 }
804 
805 /* Copy guest memory in place - must reside within a single memslot */
806 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
807 				  unsigned long len)
808 {
809 	struct kvm_memory_slot *to_memslot = NULL;
810 	struct kvm_memory_slot *from_memslot = NULL;
811 	unsigned long to_addr, from_addr;
812 	int r;
813 
814 	/* Get HPA for from address */
815 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
816 	if (!from_memslot)
817 		return -EFAULT;
818 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
819 			     << PAGE_SHIFT))
820 		return -EINVAL;
821 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
822 	if (kvm_is_error_hva(from_addr))
823 		return -EFAULT;
824 	from_addr |= (from & (PAGE_SIZE - 1));
825 
826 	/* Get HPA for to address */
827 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
828 	if (!to_memslot)
829 		return -EFAULT;
830 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
831 			   << PAGE_SHIFT))
832 		return -EINVAL;
833 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
834 	if (kvm_is_error_hva(to_addr))
835 		return -EFAULT;
836 	to_addr |= (to & (PAGE_SIZE - 1));
837 
838 	/* Perform copy */
839 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
840 			     len);
841 	if (r)
842 		return -EFAULT;
843 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
844 	return 0;
845 }
846 
847 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
848 			       unsigned long dest, unsigned long src)
849 {
850 	u64 pg_sz = SZ_4K;		/* 4K page size */
851 	u64 pg_mask = SZ_4K - 1;
852 	int ret;
853 
854 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
855 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
856 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
857 		return H_PARAMETER;
858 
859 	/* dest (and src if copy_page flag set) must be page aligned */
860 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
861 		return H_PARAMETER;
862 
863 	/* zero and/or copy the page as determined by the flags */
864 	if (flags & H_COPY_PAGE) {
865 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
866 		if (ret < 0)
867 			return H_PARAMETER;
868 	} else if (flags & H_ZERO_PAGE) {
869 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
870 		if (ret < 0)
871 			return H_PARAMETER;
872 	}
873 
874 	/* We can ignore the remaining flags */
875 
876 	return H_SUCCESS;
877 }
878 
879 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
880 {
881 	struct kvmppc_vcore *vcore = target->arch.vcore;
882 
883 	/*
884 	 * We expect to have been called by the real mode handler
885 	 * (kvmppc_rm_h_confer()) which would have directly returned
886 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
887 	 * have useful work to do and should not confer) so we don't
888 	 * recheck that here.
889 	 */
890 
891 	spin_lock(&vcore->lock);
892 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
893 	    vcore->vcore_state != VCORE_INACTIVE &&
894 	    vcore->runner)
895 		target = vcore->runner;
896 	spin_unlock(&vcore->lock);
897 
898 	return kvm_vcpu_yield_to(target);
899 }
900 
901 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
902 {
903 	int yield_count = 0;
904 	struct lppaca *lppaca;
905 
906 	spin_lock(&vcpu->arch.vpa_update_lock);
907 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
908 	if (lppaca)
909 		yield_count = be32_to_cpu(lppaca->yield_count);
910 	spin_unlock(&vcpu->arch.vpa_update_lock);
911 	return yield_count;
912 }
913 
914 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
915 {
916 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
917 	unsigned long target, ret = H_SUCCESS;
918 	int yield_count;
919 	struct kvm_vcpu *tvcpu;
920 	int idx, rc;
921 
922 	if (req <= MAX_HCALL_OPCODE &&
923 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
924 		return RESUME_HOST;
925 
926 	switch (req) {
927 	case H_CEDE:
928 		break;
929 	case H_PROD:
930 		target = kvmppc_get_gpr(vcpu, 4);
931 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
932 		if (!tvcpu) {
933 			ret = H_PARAMETER;
934 			break;
935 		}
936 		tvcpu->arch.prodded = 1;
937 		smp_mb();
938 		if (tvcpu->arch.ceded)
939 			kvmppc_fast_vcpu_kick_hv(tvcpu);
940 		break;
941 	case H_CONFER:
942 		target = kvmppc_get_gpr(vcpu, 4);
943 		if (target == -1)
944 			break;
945 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
946 		if (!tvcpu) {
947 			ret = H_PARAMETER;
948 			break;
949 		}
950 		yield_count = kvmppc_get_gpr(vcpu, 5);
951 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
952 			break;
953 		kvm_arch_vcpu_yield_to(tvcpu);
954 		break;
955 	case H_REGISTER_VPA:
956 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
957 					kvmppc_get_gpr(vcpu, 5),
958 					kvmppc_get_gpr(vcpu, 6));
959 		break;
960 	case H_RTAS:
961 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
962 			return RESUME_HOST;
963 
964 		idx = srcu_read_lock(&vcpu->kvm->srcu);
965 		rc = kvmppc_rtas_hcall(vcpu);
966 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
967 
968 		if (rc == -ENOENT)
969 			return RESUME_HOST;
970 		else if (rc == 0)
971 			break;
972 
973 		/* Send the error out to userspace via KVM_RUN */
974 		return rc;
975 	case H_LOGICAL_CI_LOAD:
976 		ret = kvmppc_h_logical_ci_load(vcpu);
977 		if (ret == H_TOO_HARD)
978 			return RESUME_HOST;
979 		break;
980 	case H_LOGICAL_CI_STORE:
981 		ret = kvmppc_h_logical_ci_store(vcpu);
982 		if (ret == H_TOO_HARD)
983 			return RESUME_HOST;
984 		break;
985 	case H_SET_MODE:
986 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
987 					kvmppc_get_gpr(vcpu, 5),
988 					kvmppc_get_gpr(vcpu, 6),
989 					kvmppc_get_gpr(vcpu, 7));
990 		if (ret == H_TOO_HARD)
991 			return RESUME_HOST;
992 		break;
993 	case H_XIRR:
994 	case H_CPPR:
995 	case H_EOI:
996 	case H_IPI:
997 	case H_IPOLL:
998 	case H_XIRR_X:
999 		if (kvmppc_xics_enabled(vcpu)) {
1000 			if (xics_on_xive()) {
1001 				ret = H_NOT_AVAILABLE;
1002 				return RESUME_GUEST;
1003 			}
1004 			ret = kvmppc_xics_hcall(vcpu, req);
1005 			break;
1006 		}
1007 		return RESUME_HOST;
1008 	case H_SET_DABR:
1009 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1010 		break;
1011 	case H_SET_XDABR:
1012 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1013 						kvmppc_get_gpr(vcpu, 5));
1014 		break;
1015 #ifdef CONFIG_SPAPR_TCE_IOMMU
1016 	case H_GET_TCE:
1017 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1018 						kvmppc_get_gpr(vcpu, 5));
1019 		if (ret == H_TOO_HARD)
1020 			return RESUME_HOST;
1021 		break;
1022 	case H_PUT_TCE:
1023 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1024 						kvmppc_get_gpr(vcpu, 5),
1025 						kvmppc_get_gpr(vcpu, 6));
1026 		if (ret == H_TOO_HARD)
1027 			return RESUME_HOST;
1028 		break;
1029 	case H_PUT_TCE_INDIRECT:
1030 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1031 						kvmppc_get_gpr(vcpu, 5),
1032 						kvmppc_get_gpr(vcpu, 6),
1033 						kvmppc_get_gpr(vcpu, 7));
1034 		if (ret == H_TOO_HARD)
1035 			return RESUME_HOST;
1036 		break;
1037 	case H_STUFF_TCE:
1038 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1039 						kvmppc_get_gpr(vcpu, 5),
1040 						kvmppc_get_gpr(vcpu, 6),
1041 						kvmppc_get_gpr(vcpu, 7));
1042 		if (ret == H_TOO_HARD)
1043 			return RESUME_HOST;
1044 		break;
1045 #endif
1046 	case H_RANDOM:
1047 		if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1048 			ret = H_HARDWARE;
1049 		break;
1050 
1051 	case H_SET_PARTITION_TABLE:
1052 		ret = H_FUNCTION;
1053 		if (nesting_enabled(vcpu->kvm))
1054 			ret = kvmhv_set_partition_table(vcpu);
1055 		break;
1056 	case H_ENTER_NESTED:
1057 		ret = H_FUNCTION;
1058 		if (!nesting_enabled(vcpu->kvm))
1059 			break;
1060 		ret = kvmhv_enter_nested_guest(vcpu);
1061 		if (ret == H_INTERRUPT) {
1062 			kvmppc_set_gpr(vcpu, 3, 0);
1063 			vcpu->arch.hcall_needed = 0;
1064 			return -EINTR;
1065 		} else if (ret == H_TOO_HARD) {
1066 			kvmppc_set_gpr(vcpu, 3, 0);
1067 			vcpu->arch.hcall_needed = 0;
1068 			return RESUME_HOST;
1069 		}
1070 		break;
1071 	case H_TLB_INVALIDATE:
1072 		ret = H_FUNCTION;
1073 		if (nesting_enabled(vcpu->kvm))
1074 			ret = kvmhv_do_nested_tlbie(vcpu);
1075 		break;
1076 	case H_COPY_TOFROM_GUEST:
1077 		ret = H_FUNCTION;
1078 		if (nesting_enabled(vcpu->kvm))
1079 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1080 		break;
1081 	case H_PAGE_INIT:
1082 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1083 					 kvmppc_get_gpr(vcpu, 5),
1084 					 kvmppc_get_gpr(vcpu, 6));
1085 		break;
1086 	default:
1087 		return RESUME_HOST;
1088 	}
1089 	kvmppc_set_gpr(vcpu, 3, ret);
1090 	vcpu->arch.hcall_needed = 0;
1091 	return RESUME_GUEST;
1092 }
1093 
1094 /*
1095  * Handle H_CEDE in the nested virtualization case where we haven't
1096  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1097  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1098  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1099  */
1100 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1101 {
1102 	vcpu->arch.shregs.msr |= MSR_EE;
1103 	vcpu->arch.ceded = 1;
1104 	smp_mb();
1105 	if (vcpu->arch.prodded) {
1106 		vcpu->arch.prodded = 0;
1107 		smp_mb();
1108 		vcpu->arch.ceded = 0;
1109 	}
1110 }
1111 
1112 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1113 {
1114 	switch (cmd) {
1115 	case H_CEDE:
1116 	case H_PROD:
1117 	case H_CONFER:
1118 	case H_REGISTER_VPA:
1119 	case H_SET_MODE:
1120 	case H_LOGICAL_CI_LOAD:
1121 	case H_LOGICAL_CI_STORE:
1122 #ifdef CONFIG_KVM_XICS
1123 	case H_XIRR:
1124 	case H_CPPR:
1125 	case H_EOI:
1126 	case H_IPI:
1127 	case H_IPOLL:
1128 	case H_XIRR_X:
1129 #endif
1130 	case H_PAGE_INIT:
1131 		return 1;
1132 	}
1133 
1134 	/* See if it's in the real-mode table */
1135 	return kvmppc_hcall_impl_hv_realmode(cmd);
1136 }
1137 
1138 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1139 					struct kvm_vcpu *vcpu)
1140 {
1141 	u32 last_inst;
1142 
1143 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1144 					EMULATE_DONE) {
1145 		/*
1146 		 * Fetch failed, so return to guest and
1147 		 * try executing it again.
1148 		 */
1149 		return RESUME_GUEST;
1150 	}
1151 
1152 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1153 		run->exit_reason = KVM_EXIT_DEBUG;
1154 		run->debug.arch.address = kvmppc_get_pc(vcpu);
1155 		return RESUME_HOST;
1156 	} else {
1157 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1158 		return RESUME_GUEST;
1159 	}
1160 }
1161 
1162 static void do_nothing(void *x)
1163 {
1164 }
1165 
1166 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1167 {
1168 	int thr, cpu, pcpu, nthreads;
1169 	struct kvm_vcpu *v;
1170 	unsigned long dpdes;
1171 
1172 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1173 	dpdes = 0;
1174 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1175 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1176 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1177 		if (!v)
1178 			continue;
1179 		/*
1180 		 * If the vcpu is currently running on a physical cpu thread,
1181 		 * interrupt it in order to pull it out of the guest briefly,
1182 		 * which will update its vcore->dpdes value.
1183 		 */
1184 		pcpu = READ_ONCE(v->cpu);
1185 		if (pcpu >= 0)
1186 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1187 		if (kvmppc_doorbell_pending(v))
1188 			dpdes |= 1 << thr;
1189 	}
1190 	return dpdes;
1191 }
1192 
1193 /*
1194  * On POWER9, emulate doorbell-related instructions in order to
1195  * give the guest the illusion of running on a multi-threaded core.
1196  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1197  * and mfspr DPDES.
1198  */
1199 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1200 {
1201 	u32 inst, rb, thr;
1202 	unsigned long arg;
1203 	struct kvm *kvm = vcpu->kvm;
1204 	struct kvm_vcpu *tvcpu;
1205 
1206 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1207 		return RESUME_GUEST;
1208 	if (get_op(inst) != 31)
1209 		return EMULATE_FAIL;
1210 	rb = get_rb(inst);
1211 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1212 	switch (get_xop(inst)) {
1213 	case OP_31_XOP_MSGSNDP:
1214 		arg = kvmppc_get_gpr(vcpu, rb);
1215 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1216 			break;
1217 		arg &= 0x3f;
1218 		if (arg >= kvm->arch.emul_smt_mode)
1219 			break;
1220 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1221 		if (!tvcpu)
1222 			break;
1223 		if (!tvcpu->arch.doorbell_request) {
1224 			tvcpu->arch.doorbell_request = 1;
1225 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1226 		}
1227 		break;
1228 	case OP_31_XOP_MSGCLRP:
1229 		arg = kvmppc_get_gpr(vcpu, rb);
1230 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1231 			break;
1232 		vcpu->arch.vcore->dpdes = 0;
1233 		vcpu->arch.doorbell_request = 0;
1234 		break;
1235 	case OP_31_XOP_MFSPR:
1236 		switch (get_sprn(inst)) {
1237 		case SPRN_TIR:
1238 			arg = thr;
1239 			break;
1240 		case SPRN_DPDES:
1241 			arg = kvmppc_read_dpdes(vcpu);
1242 			break;
1243 		default:
1244 			return EMULATE_FAIL;
1245 		}
1246 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1247 		break;
1248 	default:
1249 		return EMULATE_FAIL;
1250 	}
1251 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1252 	return RESUME_GUEST;
1253 }
1254 
1255 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1256 				 struct task_struct *tsk)
1257 {
1258 	int r = RESUME_HOST;
1259 
1260 	vcpu->stat.sum_exits++;
1261 
1262 	/*
1263 	 * This can happen if an interrupt occurs in the last stages
1264 	 * of guest entry or the first stages of guest exit (i.e. after
1265 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1266 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1267 	 * That can happen due to a bug, or due to a machine check
1268 	 * occurring at just the wrong time.
1269 	 */
1270 	if (vcpu->arch.shregs.msr & MSR_HV) {
1271 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1272 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1273 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1274 			vcpu->arch.shregs.msr);
1275 		kvmppc_dump_regs(vcpu);
1276 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1277 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1278 		return RESUME_HOST;
1279 	}
1280 	run->exit_reason = KVM_EXIT_UNKNOWN;
1281 	run->ready_for_interrupt_injection = 1;
1282 	switch (vcpu->arch.trap) {
1283 	/* We're good on these - the host merely wanted to get our attention */
1284 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1285 		vcpu->stat.dec_exits++;
1286 		r = RESUME_GUEST;
1287 		break;
1288 	case BOOK3S_INTERRUPT_EXTERNAL:
1289 	case BOOK3S_INTERRUPT_H_DOORBELL:
1290 	case BOOK3S_INTERRUPT_H_VIRT:
1291 		vcpu->stat.ext_intr_exits++;
1292 		r = RESUME_GUEST;
1293 		break;
1294 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1295 	case BOOK3S_INTERRUPT_HMI:
1296 	case BOOK3S_INTERRUPT_PERFMON:
1297 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1298 		r = RESUME_GUEST;
1299 		break;
1300 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1301 		/* Print the MCE event to host console. */
1302 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1303 
1304 		/*
1305 		 * If the guest can do FWNMI, exit to userspace so it can
1306 		 * deliver a FWNMI to the guest.
1307 		 * Otherwise we synthesize a machine check for the guest
1308 		 * so that it knows that the machine check occurred.
1309 		 */
1310 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1311 			ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1312 			kvmppc_core_queue_machine_check(vcpu, flags);
1313 			r = RESUME_GUEST;
1314 			break;
1315 		}
1316 
1317 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1318 		run->exit_reason = KVM_EXIT_NMI;
1319 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1320 		/* Clear out the old NMI status from run->flags */
1321 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1322 		/* Now set the NMI status */
1323 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1324 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1325 		else
1326 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1327 
1328 		r = RESUME_HOST;
1329 		break;
1330 	case BOOK3S_INTERRUPT_PROGRAM:
1331 	{
1332 		ulong flags;
1333 		/*
1334 		 * Normally program interrupts are delivered directly
1335 		 * to the guest by the hardware, but we can get here
1336 		 * as a result of a hypervisor emulation interrupt
1337 		 * (e40) getting turned into a 700 by BML RTAS.
1338 		 */
1339 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1340 		kvmppc_core_queue_program(vcpu, flags);
1341 		r = RESUME_GUEST;
1342 		break;
1343 	}
1344 	case BOOK3S_INTERRUPT_SYSCALL:
1345 	{
1346 		/* hcall - punt to userspace */
1347 		int i;
1348 
1349 		/* hypercall with MSR_PR has already been handled in rmode,
1350 		 * and never reaches here.
1351 		 */
1352 
1353 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1354 		for (i = 0; i < 9; ++i)
1355 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1356 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1357 		vcpu->arch.hcall_needed = 1;
1358 		r = RESUME_HOST;
1359 		break;
1360 	}
1361 	/*
1362 	 * We get these next two if the guest accesses a page which it thinks
1363 	 * it has mapped but which is not actually present, either because
1364 	 * it is for an emulated I/O device or because the corresonding
1365 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1366 	 * have been handled already.
1367 	 */
1368 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1369 		r = RESUME_PAGE_FAULT;
1370 		break;
1371 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1372 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1373 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1374 			DSISR_SRR1_MATCH_64S;
1375 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1376 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1377 		r = RESUME_PAGE_FAULT;
1378 		break;
1379 	/*
1380 	 * This occurs if the guest executes an illegal instruction.
1381 	 * If the guest debug is disabled, generate a program interrupt
1382 	 * to the guest. If guest debug is enabled, we need to check
1383 	 * whether the instruction is a software breakpoint instruction.
1384 	 * Accordingly return to Guest or Host.
1385 	 */
1386 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1387 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1388 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1389 				swab32(vcpu->arch.emul_inst) :
1390 				vcpu->arch.emul_inst;
1391 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1392 			r = kvmppc_emulate_debug_inst(run, vcpu);
1393 		} else {
1394 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1395 			r = RESUME_GUEST;
1396 		}
1397 		break;
1398 	/*
1399 	 * This occurs if the guest (kernel or userspace), does something that
1400 	 * is prohibited by HFSCR.
1401 	 * On POWER9, this could be a doorbell instruction that we need
1402 	 * to emulate.
1403 	 * Otherwise, we just generate a program interrupt to the guest.
1404 	 */
1405 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1406 		r = EMULATE_FAIL;
1407 		if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1408 		    cpu_has_feature(CPU_FTR_ARCH_300))
1409 			r = kvmppc_emulate_doorbell_instr(vcpu);
1410 		if (r == EMULATE_FAIL) {
1411 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1412 			r = RESUME_GUEST;
1413 		}
1414 		break;
1415 
1416 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1417 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1418 		/*
1419 		 * This occurs for various TM-related instructions that
1420 		 * we need to emulate on POWER9 DD2.2.  We have already
1421 		 * handled the cases where the guest was in real-suspend
1422 		 * mode and was transitioning to transactional state.
1423 		 */
1424 		r = kvmhv_p9_tm_emulation(vcpu);
1425 		break;
1426 #endif
1427 
1428 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1429 		r = RESUME_PASSTHROUGH;
1430 		break;
1431 	default:
1432 		kvmppc_dump_regs(vcpu);
1433 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1434 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1435 			vcpu->arch.shregs.msr);
1436 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1437 		r = RESUME_HOST;
1438 		break;
1439 	}
1440 
1441 	return r;
1442 }
1443 
1444 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1445 {
1446 	int r;
1447 	int srcu_idx;
1448 
1449 	vcpu->stat.sum_exits++;
1450 
1451 	/*
1452 	 * This can happen if an interrupt occurs in the last stages
1453 	 * of guest entry or the first stages of guest exit (i.e. after
1454 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1455 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1456 	 * That can happen due to a bug, or due to a machine check
1457 	 * occurring at just the wrong time.
1458 	 */
1459 	if (vcpu->arch.shregs.msr & MSR_HV) {
1460 		pr_emerg("KVM trap in HV mode while nested!\n");
1461 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1462 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1463 			 vcpu->arch.shregs.msr);
1464 		kvmppc_dump_regs(vcpu);
1465 		return RESUME_HOST;
1466 	}
1467 	switch (vcpu->arch.trap) {
1468 	/* We're good on these - the host merely wanted to get our attention */
1469 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1470 		vcpu->stat.dec_exits++;
1471 		r = RESUME_GUEST;
1472 		break;
1473 	case BOOK3S_INTERRUPT_EXTERNAL:
1474 		vcpu->stat.ext_intr_exits++;
1475 		r = RESUME_HOST;
1476 		break;
1477 	case BOOK3S_INTERRUPT_H_DOORBELL:
1478 	case BOOK3S_INTERRUPT_H_VIRT:
1479 		vcpu->stat.ext_intr_exits++;
1480 		r = RESUME_GUEST;
1481 		break;
1482 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1483 	case BOOK3S_INTERRUPT_HMI:
1484 	case BOOK3S_INTERRUPT_PERFMON:
1485 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1486 		r = RESUME_GUEST;
1487 		break;
1488 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1489 		/* Pass the machine check to the L1 guest */
1490 		r = RESUME_HOST;
1491 		/* Print the MCE event to host console. */
1492 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1493 		break;
1494 	/*
1495 	 * We get these next two if the guest accesses a page which it thinks
1496 	 * it has mapped but which is not actually present, either because
1497 	 * it is for an emulated I/O device or because the corresonding
1498 	 * host page has been paged out.
1499 	 */
1500 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1501 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1502 		r = kvmhv_nested_page_fault(run, vcpu);
1503 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1504 		break;
1505 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1506 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1507 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1508 					 DSISR_SRR1_MATCH_64S;
1509 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1510 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1511 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1512 		r = kvmhv_nested_page_fault(run, vcpu);
1513 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1514 		break;
1515 
1516 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1517 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1518 		/*
1519 		 * This occurs for various TM-related instructions that
1520 		 * we need to emulate on POWER9 DD2.2.  We have already
1521 		 * handled the cases where the guest was in real-suspend
1522 		 * mode and was transitioning to transactional state.
1523 		 */
1524 		r = kvmhv_p9_tm_emulation(vcpu);
1525 		break;
1526 #endif
1527 
1528 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1529 		vcpu->arch.trap = 0;
1530 		r = RESUME_GUEST;
1531 		if (!xics_on_xive())
1532 			kvmppc_xics_rm_complete(vcpu, 0);
1533 		break;
1534 	default:
1535 		r = RESUME_HOST;
1536 		break;
1537 	}
1538 
1539 	return r;
1540 }
1541 
1542 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1543 					    struct kvm_sregs *sregs)
1544 {
1545 	int i;
1546 
1547 	memset(sregs, 0, sizeof(struct kvm_sregs));
1548 	sregs->pvr = vcpu->arch.pvr;
1549 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1550 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1551 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1552 	}
1553 
1554 	return 0;
1555 }
1556 
1557 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1558 					    struct kvm_sregs *sregs)
1559 {
1560 	int i, j;
1561 
1562 	/* Only accept the same PVR as the host's, since we can't spoof it */
1563 	if (sregs->pvr != vcpu->arch.pvr)
1564 		return -EINVAL;
1565 
1566 	j = 0;
1567 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1568 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1569 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1570 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1571 			++j;
1572 		}
1573 	}
1574 	vcpu->arch.slb_max = j;
1575 
1576 	return 0;
1577 }
1578 
1579 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1580 		bool preserve_top32)
1581 {
1582 	struct kvm *kvm = vcpu->kvm;
1583 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1584 	u64 mask;
1585 
1586 	mutex_lock(&kvm->lock);
1587 	spin_lock(&vc->lock);
1588 	/*
1589 	 * If ILE (interrupt little-endian) has changed, update the
1590 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1591 	 */
1592 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1593 		struct kvm_vcpu *vcpu;
1594 		int i;
1595 
1596 		kvm_for_each_vcpu(i, vcpu, kvm) {
1597 			if (vcpu->arch.vcore != vc)
1598 				continue;
1599 			if (new_lpcr & LPCR_ILE)
1600 				vcpu->arch.intr_msr |= MSR_LE;
1601 			else
1602 				vcpu->arch.intr_msr &= ~MSR_LE;
1603 		}
1604 	}
1605 
1606 	/*
1607 	 * Userspace can only modify DPFD (default prefetch depth),
1608 	 * ILE (interrupt little-endian) and TC (translation control).
1609 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1610 	 */
1611 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1612 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1613 		mask |= LPCR_AIL;
1614 	/*
1615 	 * On POWER9, allow userspace to enable large decrementer for the
1616 	 * guest, whether or not the host has it enabled.
1617 	 */
1618 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1619 		mask |= LPCR_LD;
1620 
1621 	/* Broken 32-bit version of LPCR must not clear top bits */
1622 	if (preserve_top32)
1623 		mask &= 0xFFFFFFFF;
1624 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1625 	spin_unlock(&vc->lock);
1626 	mutex_unlock(&kvm->lock);
1627 }
1628 
1629 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1630 				 union kvmppc_one_reg *val)
1631 {
1632 	int r = 0;
1633 	long int i;
1634 
1635 	switch (id) {
1636 	case KVM_REG_PPC_DEBUG_INST:
1637 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1638 		break;
1639 	case KVM_REG_PPC_HIOR:
1640 		*val = get_reg_val(id, 0);
1641 		break;
1642 	case KVM_REG_PPC_DABR:
1643 		*val = get_reg_val(id, vcpu->arch.dabr);
1644 		break;
1645 	case KVM_REG_PPC_DABRX:
1646 		*val = get_reg_val(id, vcpu->arch.dabrx);
1647 		break;
1648 	case KVM_REG_PPC_DSCR:
1649 		*val = get_reg_val(id, vcpu->arch.dscr);
1650 		break;
1651 	case KVM_REG_PPC_PURR:
1652 		*val = get_reg_val(id, vcpu->arch.purr);
1653 		break;
1654 	case KVM_REG_PPC_SPURR:
1655 		*val = get_reg_val(id, vcpu->arch.spurr);
1656 		break;
1657 	case KVM_REG_PPC_AMR:
1658 		*val = get_reg_val(id, vcpu->arch.amr);
1659 		break;
1660 	case KVM_REG_PPC_UAMOR:
1661 		*val = get_reg_val(id, vcpu->arch.uamor);
1662 		break;
1663 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1664 		i = id - KVM_REG_PPC_MMCR0;
1665 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1666 		break;
1667 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1668 		i = id - KVM_REG_PPC_PMC1;
1669 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1670 		break;
1671 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1672 		i = id - KVM_REG_PPC_SPMC1;
1673 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1674 		break;
1675 	case KVM_REG_PPC_SIAR:
1676 		*val = get_reg_val(id, vcpu->arch.siar);
1677 		break;
1678 	case KVM_REG_PPC_SDAR:
1679 		*val = get_reg_val(id, vcpu->arch.sdar);
1680 		break;
1681 	case KVM_REG_PPC_SIER:
1682 		*val = get_reg_val(id, vcpu->arch.sier);
1683 		break;
1684 	case KVM_REG_PPC_IAMR:
1685 		*val = get_reg_val(id, vcpu->arch.iamr);
1686 		break;
1687 	case KVM_REG_PPC_PSPB:
1688 		*val = get_reg_val(id, vcpu->arch.pspb);
1689 		break;
1690 	case KVM_REG_PPC_DPDES:
1691 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
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 			err = -ENOMEM;
2342 			vcore = kvmppc_vcore_create(kvm,
2343 					id & ~(kvm->arch.smt_mode - 1));
2344 			kvm->arch.vcores[core] = vcore;
2345 			kvm->arch.online_vcores++;
2346 		}
2347 	}
2348 	mutex_unlock(&kvm->lock);
2349 
2350 	if (!vcore)
2351 		goto free_vcpu;
2352 
2353 	spin_lock(&vcore->lock);
2354 	++vcore->num_threads;
2355 	spin_unlock(&vcore->lock);
2356 	vcpu->arch.vcore = vcore;
2357 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2358 	vcpu->arch.thread_cpu = -1;
2359 	vcpu->arch.prev_cpu = -1;
2360 
2361 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2362 	kvmppc_sanity_check(vcpu);
2363 
2364 	debugfs_vcpu_init(vcpu, id);
2365 
2366 	return vcpu;
2367 
2368 free_vcpu:
2369 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2370 out:
2371 	return ERR_PTR(err);
2372 }
2373 
2374 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2375 			      unsigned long flags)
2376 {
2377 	int err;
2378 	int esmt = 0;
2379 
2380 	if (flags)
2381 		return -EINVAL;
2382 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2383 		return -EINVAL;
2384 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2385 		/*
2386 		 * On POWER8 (or POWER7), the threading mode is "strict",
2387 		 * so we pack smt_mode vcpus per vcore.
2388 		 */
2389 		if (smt_mode > threads_per_subcore)
2390 			return -EINVAL;
2391 	} else {
2392 		/*
2393 		 * On POWER9, the threading mode is "loose",
2394 		 * so each vcpu gets its own vcore.
2395 		 */
2396 		esmt = smt_mode;
2397 		smt_mode = 1;
2398 	}
2399 	mutex_lock(&kvm->lock);
2400 	err = -EBUSY;
2401 	if (!kvm->arch.online_vcores) {
2402 		kvm->arch.smt_mode = smt_mode;
2403 		kvm->arch.emul_smt_mode = esmt;
2404 		err = 0;
2405 	}
2406 	mutex_unlock(&kvm->lock);
2407 
2408 	return err;
2409 }
2410 
2411 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2412 {
2413 	if (vpa->pinned_addr)
2414 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2415 					vpa->dirty);
2416 }
2417 
2418 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2419 {
2420 	spin_lock(&vcpu->arch.vpa_update_lock);
2421 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2422 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2423 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2424 	spin_unlock(&vcpu->arch.vpa_update_lock);
2425 	kvm_vcpu_uninit(vcpu);
2426 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2427 }
2428 
2429 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2430 {
2431 	/* Indicate we want to get back into the guest */
2432 	return 1;
2433 }
2434 
2435 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2436 {
2437 	unsigned long dec_nsec, now;
2438 
2439 	now = get_tb();
2440 	if (now > vcpu->arch.dec_expires) {
2441 		/* decrementer has already gone negative */
2442 		kvmppc_core_queue_dec(vcpu);
2443 		kvmppc_core_prepare_to_enter(vcpu);
2444 		return;
2445 	}
2446 	dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2447 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2448 	vcpu->arch.timer_running = 1;
2449 }
2450 
2451 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2452 {
2453 	vcpu->arch.ceded = 0;
2454 	if (vcpu->arch.timer_running) {
2455 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2456 		vcpu->arch.timer_running = 0;
2457 	}
2458 }
2459 
2460 extern int __kvmppc_vcore_entry(void);
2461 
2462 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2463 				   struct kvm_vcpu *vcpu)
2464 {
2465 	u64 now;
2466 
2467 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2468 		return;
2469 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2470 	now = mftb();
2471 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2472 		vcpu->arch.stolen_logged;
2473 	vcpu->arch.busy_preempt = now;
2474 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2475 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2476 	--vc->n_runnable;
2477 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2478 }
2479 
2480 static int kvmppc_grab_hwthread(int cpu)
2481 {
2482 	struct paca_struct *tpaca;
2483 	long timeout = 10000;
2484 
2485 	tpaca = paca_ptrs[cpu];
2486 
2487 	/* Ensure the thread won't go into the kernel if it wakes */
2488 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2489 	tpaca->kvm_hstate.kvm_vcore = NULL;
2490 	tpaca->kvm_hstate.napping = 0;
2491 	smp_wmb();
2492 	tpaca->kvm_hstate.hwthread_req = 1;
2493 
2494 	/*
2495 	 * If the thread is already executing in the kernel (e.g. handling
2496 	 * a stray interrupt), wait for it to get back to nap mode.
2497 	 * The smp_mb() is to ensure that our setting of hwthread_req
2498 	 * is visible before we look at hwthread_state, so if this
2499 	 * races with the code at system_reset_pSeries and the thread
2500 	 * misses our setting of hwthread_req, we are sure to see its
2501 	 * setting of hwthread_state, and vice versa.
2502 	 */
2503 	smp_mb();
2504 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2505 		if (--timeout <= 0) {
2506 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2507 			return -EBUSY;
2508 		}
2509 		udelay(1);
2510 	}
2511 	return 0;
2512 }
2513 
2514 static void kvmppc_release_hwthread(int cpu)
2515 {
2516 	struct paca_struct *tpaca;
2517 
2518 	tpaca = paca_ptrs[cpu];
2519 	tpaca->kvm_hstate.hwthread_req = 0;
2520 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2521 	tpaca->kvm_hstate.kvm_vcore = NULL;
2522 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2523 }
2524 
2525 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2526 {
2527 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2528 	cpumask_t *cpu_in_guest;
2529 	int i;
2530 
2531 	cpu = cpu_first_thread_sibling(cpu);
2532 	if (nested) {
2533 		cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2534 		cpu_in_guest = &nested->cpu_in_guest;
2535 	} else {
2536 		cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2537 		cpu_in_guest = &kvm->arch.cpu_in_guest;
2538 	}
2539 	/*
2540 	 * Make sure setting of bit in need_tlb_flush precedes
2541 	 * testing of cpu_in_guest bits.  The matching barrier on
2542 	 * the other side is the first smp_mb() in kvmppc_run_core().
2543 	 */
2544 	smp_mb();
2545 	for (i = 0; i < threads_per_core; ++i)
2546 		if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2547 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2548 }
2549 
2550 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2551 {
2552 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2553 	struct kvm *kvm = vcpu->kvm;
2554 	int prev_cpu;
2555 
2556 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2557 		return;
2558 
2559 	if (nested)
2560 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2561 	else
2562 		prev_cpu = vcpu->arch.prev_cpu;
2563 
2564 	/*
2565 	 * With radix, the guest can do TLB invalidations itself,
2566 	 * and it could choose to use the local form (tlbiel) if
2567 	 * it is invalidating a translation that has only ever been
2568 	 * used on one vcpu.  However, that doesn't mean it has
2569 	 * only ever been used on one physical cpu, since vcpus
2570 	 * can move around between pcpus.  To cope with this, when
2571 	 * a vcpu moves from one pcpu to another, we need to tell
2572 	 * any vcpus running on the same core as this vcpu previously
2573 	 * ran to flush the TLB.  The TLB is shared between threads,
2574 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2575 	 */
2576 	if (prev_cpu != pcpu) {
2577 		if (prev_cpu >= 0 &&
2578 		    cpu_first_thread_sibling(prev_cpu) !=
2579 		    cpu_first_thread_sibling(pcpu))
2580 			radix_flush_cpu(kvm, prev_cpu, vcpu);
2581 		if (nested)
2582 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2583 		else
2584 			vcpu->arch.prev_cpu = pcpu;
2585 	}
2586 }
2587 
2588 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2589 {
2590 	int cpu;
2591 	struct paca_struct *tpaca;
2592 	struct kvm *kvm = vc->kvm;
2593 
2594 	cpu = vc->pcpu;
2595 	if (vcpu) {
2596 		if (vcpu->arch.timer_running) {
2597 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2598 			vcpu->arch.timer_running = 0;
2599 		}
2600 		cpu += vcpu->arch.ptid;
2601 		vcpu->cpu = vc->pcpu;
2602 		vcpu->arch.thread_cpu = cpu;
2603 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2604 	}
2605 	tpaca = paca_ptrs[cpu];
2606 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2607 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2608 	tpaca->kvm_hstate.fake_suspend = 0;
2609 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2610 	smp_wmb();
2611 	tpaca->kvm_hstate.kvm_vcore = vc;
2612 	if (cpu != smp_processor_id())
2613 		kvmppc_ipi_thread(cpu);
2614 }
2615 
2616 static void kvmppc_wait_for_nap(int n_threads)
2617 {
2618 	int cpu = smp_processor_id();
2619 	int i, loops;
2620 
2621 	if (n_threads <= 1)
2622 		return;
2623 	for (loops = 0; loops < 1000000; ++loops) {
2624 		/*
2625 		 * Check if all threads are finished.
2626 		 * We set the vcore pointer when starting a thread
2627 		 * and the thread clears it when finished, so we look
2628 		 * for any threads that still have a non-NULL vcore ptr.
2629 		 */
2630 		for (i = 1; i < n_threads; ++i)
2631 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2632 				break;
2633 		if (i == n_threads) {
2634 			HMT_medium();
2635 			return;
2636 		}
2637 		HMT_low();
2638 	}
2639 	HMT_medium();
2640 	for (i = 1; i < n_threads; ++i)
2641 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2642 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2643 }
2644 
2645 /*
2646  * Check that we are on thread 0 and that any other threads in
2647  * this core are off-line.  Then grab the threads so they can't
2648  * enter the kernel.
2649  */
2650 static int on_primary_thread(void)
2651 {
2652 	int cpu = smp_processor_id();
2653 	int thr;
2654 
2655 	/* Are we on a primary subcore? */
2656 	if (cpu_thread_in_subcore(cpu))
2657 		return 0;
2658 
2659 	thr = 0;
2660 	while (++thr < threads_per_subcore)
2661 		if (cpu_online(cpu + thr))
2662 			return 0;
2663 
2664 	/* Grab all hw threads so they can't go into the kernel */
2665 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2666 		if (kvmppc_grab_hwthread(cpu + thr)) {
2667 			/* Couldn't grab one; let the others go */
2668 			do {
2669 				kvmppc_release_hwthread(cpu + thr);
2670 			} while (--thr > 0);
2671 			return 0;
2672 		}
2673 	}
2674 	return 1;
2675 }
2676 
2677 /*
2678  * A list of virtual cores for each physical CPU.
2679  * These are vcores that could run but their runner VCPU tasks are
2680  * (or may be) preempted.
2681  */
2682 struct preempted_vcore_list {
2683 	struct list_head	list;
2684 	spinlock_t		lock;
2685 };
2686 
2687 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2688 
2689 static void init_vcore_lists(void)
2690 {
2691 	int cpu;
2692 
2693 	for_each_possible_cpu(cpu) {
2694 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2695 		spin_lock_init(&lp->lock);
2696 		INIT_LIST_HEAD(&lp->list);
2697 	}
2698 }
2699 
2700 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2701 {
2702 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2703 
2704 	vc->vcore_state = VCORE_PREEMPT;
2705 	vc->pcpu = smp_processor_id();
2706 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2707 		spin_lock(&lp->lock);
2708 		list_add_tail(&vc->preempt_list, &lp->list);
2709 		spin_unlock(&lp->lock);
2710 	}
2711 
2712 	/* Start accumulating stolen time */
2713 	kvmppc_core_start_stolen(vc);
2714 }
2715 
2716 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2717 {
2718 	struct preempted_vcore_list *lp;
2719 
2720 	kvmppc_core_end_stolen(vc);
2721 	if (!list_empty(&vc->preempt_list)) {
2722 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2723 		spin_lock(&lp->lock);
2724 		list_del_init(&vc->preempt_list);
2725 		spin_unlock(&lp->lock);
2726 	}
2727 	vc->vcore_state = VCORE_INACTIVE;
2728 }
2729 
2730 /*
2731  * This stores information about the virtual cores currently
2732  * assigned to a physical core.
2733  */
2734 struct core_info {
2735 	int		n_subcores;
2736 	int		max_subcore_threads;
2737 	int		total_threads;
2738 	int		subcore_threads[MAX_SUBCORES];
2739 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2740 };
2741 
2742 /*
2743  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2744  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2745  */
2746 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2747 
2748 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2749 {
2750 	memset(cip, 0, sizeof(*cip));
2751 	cip->n_subcores = 1;
2752 	cip->max_subcore_threads = vc->num_threads;
2753 	cip->total_threads = vc->num_threads;
2754 	cip->subcore_threads[0] = vc->num_threads;
2755 	cip->vc[0] = vc;
2756 }
2757 
2758 static bool subcore_config_ok(int n_subcores, int n_threads)
2759 {
2760 	/*
2761 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2762 	 * split-core mode, with one thread per subcore.
2763 	 */
2764 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2765 		return n_subcores <= 4 && n_threads == 1;
2766 
2767 	/* On POWER8, can only dynamically split if unsplit to begin with */
2768 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2769 		return false;
2770 	if (n_subcores > MAX_SUBCORES)
2771 		return false;
2772 	if (n_subcores > 1) {
2773 		if (!(dynamic_mt_modes & 2))
2774 			n_subcores = 4;
2775 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2776 			return false;
2777 	}
2778 
2779 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2780 }
2781 
2782 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2783 {
2784 	vc->entry_exit_map = 0;
2785 	vc->in_guest = 0;
2786 	vc->napping_threads = 0;
2787 	vc->conferring_threads = 0;
2788 	vc->tb_offset_applied = 0;
2789 }
2790 
2791 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2792 {
2793 	int n_threads = vc->num_threads;
2794 	int sub;
2795 
2796 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2797 		return false;
2798 
2799 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
2800 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2801 		return false;
2802 
2803 	/* Some POWER9 chips require all threads to be in the same MMU mode */
2804 	if (no_mixing_hpt_and_radix &&
2805 	    kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2806 		return false;
2807 
2808 	if (n_threads < cip->max_subcore_threads)
2809 		n_threads = cip->max_subcore_threads;
2810 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2811 		return false;
2812 	cip->max_subcore_threads = n_threads;
2813 
2814 	sub = cip->n_subcores;
2815 	++cip->n_subcores;
2816 	cip->total_threads += vc->num_threads;
2817 	cip->subcore_threads[sub] = vc->num_threads;
2818 	cip->vc[sub] = vc;
2819 	init_vcore_to_run(vc);
2820 	list_del_init(&vc->preempt_list);
2821 
2822 	return true;
2823 }
2824 
2825 /*
2826  * Work out whether it is possible to piggyback the execution of
2827  * vcore *pvc onto the execution of the other vcores described in *cip.
2828  */
2829 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2830 			  int target_threads)
2831 {
2832 	if (cip->total_threads + pvc->num_threads > target_threads)
2833 		return false;
2834 
2835 	return can_dynamic_split(pvc, cip);
2836 }
2837 
2838 static void prepare_threads(struct kvmppc_vcore *vc)
2839 {
2840 	int i;
2841 	struct kvm_vcpu *vcpu;
2842 
2843 	for_each_runnable_thread(i, vcpu, vc) {
2844 		if (signal_pending(vcpu->arch.run_task))
2845 			vcpu->arch.ret = -EINTR;
2846 		else if (vcpu->arch.vpa.update_pending ||
2847 			 vcpu->arch.slb_shadow.update_pending ||
2848 			 vcpu->arch.dtl.update_pending)
2849 			vcpu->arch.ret = RESUME_GUEST;
2850 		else
2851 			continue;
2852 		kvmppc_remove_runnable(vc, vcpu);
2853 		wake_up(&vcpu->arch.cpu_run);
2854 	}
2855 }
2856 
2857 static void collect_piggybacks(struct core_info *cip, int target_threads)
2858 {
2859 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2860 	struct kvmppc_vcore *pvc, *vcnext;
2861 
2862 	spin_lock(&lp->lock);
2863 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2864 		if (!spin_trylock(&pvc->lock))
2865 			continue;
2866 		prepare_threads(pvc);
2867 		if (!pvc->n_runnable) {
2868 			list_del_init(&pvc->preempt_list);
2869 			if (pvc->runner == NULL) {
2870 				pvc->vcore_state = VCORE_INACTIVE;
2871 				kvmppc_core_end_stolen(pvc);
2872 			}
2873 			spin_unlock(&pvc->lock);
2874 			continue;
2875 		}
2876 		if (!can_piggyback(pvc, cip, target_threads)) {
2877 			spin_unlock(&pvc->lock);
2878 			continue;
2879 		}
2880 		kvmppc_core_end_stolen(pvc);
2881 		pvc->vcore_state = VCORE_PIGGYBACK;
2882 		if (cip->total_threads >= target_threads)
2883 			break;
2884 	}
2885 	spin_unlock(&lp->lock);
2886 }
2887 
2888 static bool recheck_signals(struct core_info *cip)
2889 {
2890 	int sub, i;
2891 	struct kvm_vcpu *vcpu;
2892 
2893 	for (sub = 0; sub < cip->n_subcores; ++sub)
2894 		for_each_runnable_thread(i, vcpu, cip->vc[sub])
2895 			if (signal_pending(vcpu->arch.run_task))
2896 				return true;
2897 	return false;
2898 }
2899 
2900 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2901 {
2902 	int still_running = 0, i;
2903 	u64 now;
2904 	long ret;
2905 	struct kvm_vcpu *vcpu;
2906 
2907 	spin_lock(&vc->lock);
2908 	now = get_tb();
2909 	for_each_runnable_thread(i, vcpu, vc) {
2910 		/*
2911 		 * It's safe to unlock the vcore in the loop here, because
2912 		 * for_each_runnable_thread() is safe against removal of
2913 		 * the vcpu, and the vcore state is VCORE_EXITING here,
2914 		 * so any vcpus becoming runnable will have their arch.trap
2915 		 * set to zero and can't actually run in the guest.
2916 		 */
2917 		spin_unlock(&vc->lock);
2918 		/* cancel pending dec exception if dec is positive */
2919 		if (now < vcpu->arch.dec_expires &&
2920 		    kvmppc_core_pending_dec(vcpu))
2921 			kvmppc_core_dequeue_dec(vcpu);
2922 
2923 		trace_kvm_guest_exit(vcpu);
2924 
2925 		ret = RESUME_GUEST;
2926 		if (vcpu->arch.trap)
2927 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2928 						    vcpu->arch.run_task);
2929 
2930 		vcpu->arch.ret = ret;
2931 		vcpu->arch.trap = 0;
2932 
2933 		spin_lock(&vc->lock);
2934 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2935 			if (vcpu->arch.pending_exceptions)
2936 				kvmppc_core_prepare_to_enter(vcpu);
2937 			if (vcpu->arch.ceded)
2938 				kvmppc_set_timer(vcpu);
2939 			else
2940 				++still_running;
2941 		} else {
2942 			kvmppc_remove_runnable(vc, vcpu);
2943 			wake_up(&vcpu->arch.cpu_run);
2944 		}
2945 	}
2946 	if (!is_master) {
2947 		if (still_running > 0) {
2948 			kvmppc_vcore_preempt(vc);
2949 		} else if (vc->runner) {
2950 			vc->vcore_state = VCORE_PREEMPT;
2951 			kvmppc_core_start_stolen(vc);
2952 		} else {
2953 			vc->vcore_state = VCORE_INACTIVE;
2954 		}
2955 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2956 			/* make sure there's a candidate runner awake */
2957 			i = -1;
2958 			vcpu = next_runnable_thread(vc, &i);
2959 			wake_up(&vcpu->arch.cpu_run);
2960 		}
2961 	}
2962 	spin_unlock(&vc->lock);
2963 }
2964 
2965 /*
2966  * Clear core from the list of active host cores as we are about to
2967  * enter the guest. Only do this if it is the primary thread of the
2968  * core (not if a subcore) that is entering the guest.
2969  */
2970 static inline int kvmppc_clear_host_core(unsigned int cpu)
2971 {
2972 	int core;
2973 
2974 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2975 		return 0;
2976 	/*
2977 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2978 	 * later in kvmppc_start_thread and we need ensure that state is
2979 	 * visible to other CPUs only after we enter guest.
2980 	 */
2981 	core = cpu >> threads_shift;
2982 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2983 	return 0;
2984 }
2985 
2986 /*
2987  * Advertise this core as an active host core since we exited the guest
2988  * Only need to do this if it is the primary thread of the core that is
2989  * exiting.
2990  */
2991 static inline int kvmppc_set_host_core(unsigned int cpu)
2992 {
2993 	int core;
2994 
2995 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2996 		return 0;
2997 
2998 	/*
2999 	 * Memory barrier can be omitted here because we do a spin_unlock
3000 	 * immediately after this which provides the memory barrier.
3001 	 */
3002 	core = cpu >> threads_shift;
3003 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3004 	return 0;
3005 }
3006 
3007 static void set_irq_happened(int trap)
3008 {
3009 	switch (trap) {
3010 	case BOOK3S_INTERRUPT_EXTERNAL:
3011 		local_paca->irq_happened |= PACA_IRQ_EE;
3012 		break;
3013 	case BOOK3S_INTERRUPT_H_DOORBELL:
3014 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3015 		break;
3016 	case BOOK3S_INTERRUPT_HMI:
3017 		local_paca->irq_happened |= PACA_IRQ_HMI;
3018 		break;
3019 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3020 		replay_system_reset();
3021 		break;
3022 	}
3023 }
3024 
3025 /*
3026  * Run a set of guest threads on a physical core.
3027  * Called with vc->lock held.
3028  */
3029 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3030 {
3031 	struct kvm_vcpu *vcpu;
3032 	int i;
3033 	int srcu_idx;
3034 	struct core_info core_info;
3035 	struct kvmppc_vcore *pvc;
3036 	struct kvm_split_mode split_info, *sip;
3037 	int split, subcore_size, active;
3038 	int sub;
3039 	bool thr0_done;
3040 	unsigned long cmd_bit, stat_bit;
3041 	int pcpu, thr;
3042 	int target_threads;
3043 	int controlled_threads;
3044 	int trap;
3045 	bool is_power8;
3046 	bool hpt_on_radix;
3047 
3048 	/*
3049 	 * Remove from the list any threads that have a signal pending
3050 	 * or need a VPA update done
3051 	 */
3052 	prepare_threads(vc);
3053 
3054 	/* if the runner is no longer runnable, let the caller pick a new one */
3055 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3056 		return;
3057 
3058 	/*
3059 	 * Initialize *vc.
3060 	 */
3061 	init_vcore_to_run(vc);
3062 	vc->preempt_tb = TB_NIL;
3063 
3064 	/*
3065 	 * Number of threads that we will be controlling: the same as
3066 	 * the number of threads per subcore, except on POWER9,
3067 	 * where it's 1 because the threads are (mostly) independent.
3068 	 */
3069 	controlled_threads = threads_per_vcore(vc->kvm);
3070 
3071 	/*
3072 	 * Make sure we are running on primary threads, and that secondary
3073 	 * threads are offline.  Also check if the number of threads in this
3074 	 * guest are greater than the current system threads per guest.
3075 	 * On POWER9, we need to be not in independent-threads mode if
3076 	 * this is a HPT guest on a radix host machine where the
3077 	 * CPU threads may not be in different MMU modes.
3078 	 */
3079 	hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3080 		!kvm_is_radix(vc->kvm);
3081 	if (((controlled_threads > 1) &&
3082 	     ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3083 	    (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3084 		for_each_runnable_thread(i, vcpu, vc) {
3085 			vcpu->arch.ret = -EBUSY;
3086 			kvmppc_remove_runnable(vc, vcpu);
3087 			wake_up(&vcpu->arch.cpu_run);
3088 		}
3089 		goto out;
3090 	}
3091 
3092 	/*
3093 	 * See if we could run any other vcores on the physical core
3094 	 * along with this one.
3095 	 */
3096 	init_core_info(&core_info, vc);
3097 	pcpu = smp_processor_id();
3098 	target_threads = controlled_threads;
3099 	if (target_smt_mode && target_smt_mode < target_threads)
3100 		target_threads = target_smt_mode;
3101 	if (vc->num_threads < target_threads)
3102 		collect_piggybacks(&core_info, target_threads);
3103 
3104 	/*
3105 	 * On radix, arrange for TLB flushing if necessary.
3106 	 * This has to be done before disabling interrupts since
3107 	 * it uses smp_call_function().
3108 	 */
3109 	pcpu = smp_processor_id();
3110 	if (kvm_is_radix(vc->kvm)) {
3111 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3112 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3113 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3114 	}
3115 
3116 	/*
3117 	 * Hard-disable interrupts, and check resched flag and signals.
3118 	 * If we need to reschedule or deliver a signal, clean up
3119 	 * and return without going into the guest(s).
3120 	 * If the mmu_ready flag has been cleared, don't go into the
3121 	 * guest because that means a HPT resize operation is in progress.
3122 	 */
3123 	local_irq_disable();
3124 	hard_irq_disable();
3125 	if (lazy_irq_pending() || need_resched() ||
3126 	    recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3127 		local_irq_enable();
3128 		vc->vcore_state = VCORE_INACTIVE;
3129 		/* Unlock all except the primary vcore */
3130 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3131 			pvc = core_info.vc[sub];
3132 			/* Put back on to the preempted vcores list */
3133 			kvmppc_vcore_preempt(pvc);
3134 			spin_unlock(&pvc->lock);
3135 		}
3136 		for (i = 0; i < controlled_threads; ++i)
3137 			kvmppc_release_hwthread(pcpu + i);
3138 		return;
3139 	}
3140 
3141 	kvmppc_clear_host_core(pcpu);
3142 
3143 	/* Decide on micro-threading (split-core) mode */
3144 	subcore_size = threads_per_subcore;
3145 	cmd_bit = stat_bit = 0;
3146 	split = core_info.n_subcores;
3147 	sip = NULL;
3148 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3149 		&& !cpu_has_feature(CPU_FTR_ARCH_300);
3150 
3151 	if (split > 1 || hpt_on_radix) {
3152 		sip = &split_info;
3153 		memset(&split_info, 0, sizeof(split_info));
3154 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3155 			split_info.vc[sub] = core_info.vc[sub];
3156 
3157 		if (is_power8) {
3158 			if (split == 2 && (dynamic_mt_modes & 2)) {
3159 				cmd_bit = HID0_POWER8_1TO2LPAR;
3160 				stat_bit = HID0_POWER8_2LPARMODE;
3161 			} else {
3162 				split = 4;
3163 				cmd_bit = HID0_POWER8_1TO4LPAR;
3164 				stat_bit = HID0_POWER8_4LPARMODE;
3165 			}
3166 			subcore_size = MAX_SMT_THREADS / split;
3167 			split_info.rpr = mfspr(SPRN_RPR);
3168 			split_info.pmmar = mfspr(SPRN_PMMAR);
3169 			split_info.ldbar = mfspr(SPRN_LDBAR);
3170 			split_info.subcore_size = subcore_size;
3171 		} else {
3172 			split_info.subcore_size = 1;
3173 			if (hpt_on_radix) {
3174 				/* Use the split_info for LPCR/LPIDR changes */
3175 				split_info.lpcr_req = vc->lpcr;
3176 				split_info.lpidr_req = vc->kvm->arch.lpid;
3177 				split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3178 				split_info.do_set = 1;
3179 			}
3180 		}
3181 
3182 		/* order writes to split_info before kvm_split_mode pointer */
3183 		smp_wmb();
3184 	}
3185 
3186 	for (thr = 0; thr < controlled_threads; ++thr) {
3187 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3188 
3189 		paca->kvm_hstate.tid = thr;
3190 		paca->kvm_hstate.napping = 0;
3191 		paca->kvm_hstate.kvm_split_mode = sip;
3192 	}
3193 
3194 	/* Initiate micro-threading (split-core) on POWER8 if required */
3195 	if (cmd_bit) {
3196 		unsigned long hid0 = mfspr(SPRN_HID0);
3197 
3198 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3199 		mb();
3200 		mtspr(SPRN_HID0, hid0);
3201 		isync();
3202 		for (;;) {
3203 			hid0 = mfspr(SPRN_HID0);
3204 			if (hid0 & stat_bit)
3205 				break;
3206 			cpu_relax();
3207 		}
3208 	}
3209 
3210 	/*
3211 	 * On POWER8, set RWMR register.
3212 	 * Since it only affects PURR and SPURR, it doesn't affect
3213 	 * the host, so we don't save/restore the host value.
3214 	 */
3215 	if (is_power8) {
3216 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3217 		int n_online = atomic_read(&vc->online_count);
3218 
3219 		/*
3220 		 * Use the 8-thread value if we're doing split-core
3221 		 * or if the vcore's online count looks bogus.
3222 		 */
3223 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3224 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3225 			rwmr_val = p8_rwmr_values[n_online];
3226 		mtspr(SPRN_RWMR, rwmr_val);
3227 	}
3228 
3229 	/* Start all the threads */
3230 	active = 0;
3231 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3232 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3233 		thr0_done = false;
3234 		active |= 1 << thr;
3235 		pvc = core_info.vc[sub];
3236 		pvc->pcpu = pcpu + thr;
3237 		for_each_runnable_thread(i, vcpu, pvc) {
3238 			kvmppc_start_thread(vcpu, pvc);
3239 			kvmppc_create_dtl_entry(vcpu, pvc);
3240 			trace_kvm_guest_enter(vcpu);
3241 			if (!vcpu->arch.ptid)
3242 				thr0_done = true;
3243 			active |= 1 << (thr + vcpu->arch.ptid);
3244 		}
3245 		/*
3246 		 * We need to start the first thread of each subcore
3247 		 * even if it doesn't have a vcpu.
3248 		 */
3249 		if (!thr0_done)
3250 			kvmppc_start_thread(NULL, pvc);
3251 	}
3252 
3253 	/*
3254 	 * Ensure that split_info.do_nap is set after setting
3255 	 * the vcore pointer in the PACA of the secondaries.
3256 	 */
3257 	smp_mb();
3258 
3259 	/*
3260 	 * When doing micro-threading, poke the inactive threads as well.
3261 	 * This gets them to the nap instruction after kvm_do_nap,
3262 	 * which reduces the time taken to unsplit later.
3263 	 * For POWER9 HPT guest on radix host, we need all the secondary
3264 	 * threads woken up so they can do the LPCR/LPIDR change.
3265 	 */
3266 	if (cmd_bit || hpt_on_radix) {
3267 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3268 		for (thr = 1; thr < threads_per_subcore; ++thr)
3269 			if (!(active & (1 << thr)))
3270 				kvmppc_ipi_thread(pcpu + thr);
3271 	}
3272 
3273 	vc->vcore_state = VCORE_RUNNING;
3274 	preempt_disable();
3275 
3276 	trace_kvmppc_run_core(vc, 0);
3277 
3278 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3279 		spin_unlock(&core_info.vc[sub]->lock);
3280 
3281 	guest_enter_irqoff();
3282 
3283 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3284 
3285 	this_cpu_disable_ftrace();
3286 
3287 	/*
3288 	 * Interrupts will be enabled once we get into the guest,
3289 	 * so tell lockdep that we're about to enable interrupts.
3290 	 */
3291 	trace_hardirqs_on();
3292 
3293 	trap = __kvmppc_vcore_entry();
3294 
3295 	trace_hardirqs_off();
3296 
3297 	this_cpu_enable_ftrace();
3298 
3299 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3300 
3301 	set_irq_happened(trap);
3302 
3303 	spin_lock(&vc->lock);
3304 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3305 	vc->vcore_state = VCORE_EXITING;
3306 
3307 	/* wait for secondary threads to finish writing their state to memory */
3308 	kvmppc_wait_for_nap(controlled_threads);
3309 
3310 	/* Return to whole-core mode if we split the core earlier */
3311 	if (cmd_bit) {
3312 		unsigned long hid0 = mfspr(SPRN_HID0);
3313 		unsigned long loops = 0;
3314 
3315 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3316 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3317 		mb();
3318 		mtspr(SPRN_HID0, hid0);
3319 		isync();
3320 		for (;;) {
3321 			hid0 = mfspr(SPRN_HID0);
3322 			if (!(hid0 & stat_bit))
3323 				break;
3324 			cpu_relax();
3325 			++loops;
3326 		}
3327 	} else if (hpt_on_radix) {
3328 		/* Wait for all threads to have seen final sync */
3329 		for (thr = 1; thr < controlled_threads; ++thr) {
3330 			struct paca_struct *paca = paca_ptrs[pcpu + thr];
3331 
3332 			while (paca->kvm_hstate.kvm_split_mode) {
3333 				HMT_low();
3334 				barrier();
3335 			}
3336 			HMT_medium();
3337 		}
3338 	}
3339 	split_info.do_nap = 0;
3340 
3341 	kvmppc_set_host_core(pcpu);
3342 
3343 	local_irq_enable();
3344 	guest_exit();
3345 
3346 	/* Let secondaries go back to the offline loop */
3347 	for (i = 0; i < controlled_threads; ++i) {
3348 		kvmppc_release_hwthread(pcpu + i);
3349 		if (sip && sip->napped[i])
3350 			kvmppc_ipi_thread(pcpu + i);
3351 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3352 	}
3353 
3354 	spin_unlock(&vc->lock);
3355 
3356 	/* make sure updates to secondary vcpu structs are visible now */
3357 	smp_mb();
3358 
3359 	preempt_enable();
3360 
3361 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3362 		pvc = core_info.vc[sub];
3363 		post_guest_process(pvc, pvc == vc);
3364 	}
3365 
3366 	spin_lock(&vc->lock);
3367 
3368  out:
3369 	vc->vcore_state = VCORE_INACTIVE;
3370 	trace_kvmppc_run_core(vc, 1);
3371 }
3372 
3373 /*
3374  * Load up hypervisor-mode registers on P9.
3375  */
3376 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3377 				     unsigned long lpcr)
3378 {
3379 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3380 	s64 hdec;
3381 	u64 tb, purr, spurr;
3382 	int trap;
3383 	unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3384 	unsigned long host_ciabr = mfspr(SPRN_CIABR);
3385 	unsigned long host_dawr = mfspr(SPRN_DAWR);
3386 	unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3387 	unsigned long host_psscr = mfspr(SPRN_PSSCR);
3388 	unsigned long host_pidr = mfspr(SPRN_PID);
3389 
3390 	hdec = time_limit - mftb();
3391 	if (hdec < 0)
3392 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3393 	mtspr(SPRN_HDEC, hdec);
3394 
3395 	if (vc->tb_offset) {
3396 		u64 new_tb = mftb() + vc->tb_offset;
3397 		mtspr(SPRN_TBU40, new_tb);
3398 		tb = mftb();
3399 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3400 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3401 		vc->tb_offset_applied = vc->tb_offset;
3402 	}
3403 
3404 	if (vc->pcr)
3405 		mtspr(SPRN_PCR, vc->pcr);
3406 	mtspr(SPRN_DPDES, vc->dpdes);
3407 	mtspr(SPRN_VTB, vc->vtb);
3408 
3409 	local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3410 	local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3411 	mtspr(SPRN_PURR, vcpu->arch.purr);
3412 	mtspr(SPRN_SPURR, vcpu->arch.spurr);
3413 
3414 	if (dawr_enabled()) {
3415 		mtspr(SPRN_DAWR, vcpu->arch.dawr);
3416 		mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3417 	}
3418 	mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3419 	mtspr(SPRN_IC, vcpu->arch.ic);
3420 	mtspr(SPRN_PID, vcpu->arch.pid);
3421 
3422 	mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3423 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3424 
3425 	mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3426 
3427 	mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3428 	mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3429 	mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3430 	mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3431 
3432 	mtspr(SPRN_AMOR, ~0UL);
3433 
3434 	mtspr(SPRN_LPCR, lpcr);
3435 	isync();
3436 
3437 	kvmppc_xive_push_vcpu(vcpu);
3438 
3439 	mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3440 	mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3441 
3442 	trap = __kvmhv_vcpu_entry_p9(vcpu);
3443 
3444 	/* Advance host PURR/SPURR by the amount used by guest */
3445 	purr = mfspr(SPRN_PURR);
3446 	spurr = mfspr(SPRN_SPURR);
3447 	mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3448 	      purr - vcpu->arch.purr);
3449 	mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3450 	      spurr - vcpu->arch.spurr);
3451 	vcpu->arch.purr = purr;
3452 	vcpu->arch.spurr = spurr;
3453 
3454 	vcpu->arch.ic = mfspr(SPRN_IC);
3455 	vcpu->arch.pid = mfspr(SPRN_PID);
3456 	vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3457 
3458 	vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3459 	vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3460 	vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3461 	vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3462 
3463 	/* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3464 	mtspr(SPRN_PSSCR, host_psscr |
3465 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3466 	mtspr(SPRN_HFSCR, host_hfscr);
3467 	mtspr(SPRN_CIABR, host_ciabr);
3468 	mtspr(SPRN_DAWR, host_dawr);
3469 	mtspr(SPRN_DAWRX, host_dawrx);
3470 	mtspr(SPRN_PID, host_pidr);
3471 
3472 	/*
3473 	 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3474 	 * case we interrupted the guest between a tlbie and a ptesync.
3475 	 */
3476 	asm volatile("eieio; tlbsync; ptesync");
3477 
3478 	mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);	/* restore host LPID */
3479 	isync();
3480 
3481 	vc->dpdes = mfspr(SPRN_DPDES);
3482 	vc->vtb = mfspr(SPRN_VTB);
3483 	mtspr(SPRN_DPDES, 0);
3484 	if (vc->pcr)
3485 		mtspr(SPRN_PCR, 0);
3486 
3487 	if (vc->tb_offset_applied) {
3488 		u64 new_tb = mftb() - vc->tb_offset_applied;
3489 		mtspr(SPRN_TBU40, new_tb);
3490 		tb = mftb();
3491 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3492 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3493 		vc->tb_offset_applied = 0;
3494 	}
3495 
3496 	mtspr(SPRN_HDEC, 0x7fffffff);
3497 	mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3498 
3499 	return trap;
3500 }
3501 
3502 /*
3503  * Virtual-mode guest entry for POWER9 and later when the host and
3504  * guest are both using the radix MMU.  The LPIDR has already been set.
3505  */
3506 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3507 			 unsigned long lpcr)
3508 {
3509 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3510 	unsigned long host_dscr = mfspr(SPRN_DSCR);
3511 	unsigned long host_tidr = mfspr(SPRN_TIDR);
3512 	unsigned long host_iamr = mfspr(SPRN_IAMR);
3513 	unsigned long host_amr = mfspr(SPRN_AMR);
3514 	s64 dec;
3515 	u64 tb;
3516 	int trap, save_pmu;
3517 
3518 	dec = mfspr(SPRN_DEC);
3519 	tb = mftb();
3520 	if (dec < 512)
3521 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3522 	local_paca->kvm_hstate.dec_expires = dec + tb;
3523 	if (local_paca->kvm_hstate.dec_expires < time_limit)
3524 		time_limit = local_paca->kvm_hstate.dec_expires;
3525 
3526 	vcpu->arch.ceded = 0;
3527 
3528 	kvmhv_save_host_pmu();		/* saves it to PACA kvm_hstate */
3529 
3530 	kvmppc_subcore_enter_guest();
3531 
3532 	vc->entry_exit_map = 1;
3533 	vc->in_guest = 1;
3534 
3535 	if (vcpu->arch.vpa.pinned_addr) {
3536 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3537 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3538 		lp->yield_count = cpu_to_be32(yield_count);
3539 		vcpu->arch.vpa.dirty = 1;
3540 	}
3541 
3542 	if (cpu_has_feature(CPU_FTR_TM) ||
3543 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3544 		kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3545 
3546 	kvmhv_load_guest_pmu(vcpu);
3547 
3548 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3549 	load_fp_state(&vcpu->arch.fp);
3550 #ifdef CONFIG_ALTIVEC
3551 	load_vr_state(&vcpu->arch.vr);
3552 #endif
3553 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3554 
3555 	mtspr(SPRN_DSCR, vcpu->arch.dscr);
3556 	mtspr(SPRN_IAMR, vcpu->arch.iamr);
3557 	mtspr(SPRN_PSPB, vcpu->arch.pspb);
3558 	mtspr(SPRN_FSCR, vcpu->arch.fscr);
3559 	mtspr(SPRN_TAR, vcpu->arch.tar);
3560 	mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3561 	mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3562 	mtspr(SPRN_BESCR, vcpu->arch.bescr);
3563 	mtspr(SPRN_WORT, vcpu->arch.wort);
3564 	mtspr(SPRN_TIDR, vcpu->arch.tid);
3565 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3566 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3567 	mtspr(SPRN_AMR, vcpu->arch.amr);
3568 	mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3569 
3570 	if (!(vcpu->arch.ctrl & 1))
3571 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3572 
3573 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3574 
3575 	if (kvmhv_on_pseries()) {
3576 		/* call our hypervisor to load up HV regs and go */
3577 		struct hv_guest_state hvregs;
3578 
3579 		kvmhv_save_hv_regs(vcpu, &hvregs);
3580 		hvregs.lpcr = lpcr;
3581 		vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3582 		hvregs.version = HV_GUEST_STATE_VERSION;
3583 		if (vcpu->arch.nested) {
3584 			hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3585 			hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3586 		} else {
3587 			hvregs.lpid = vcpu->kvm->arch.lpid;
3588 			hvregs.vcpu_token = vcpu->vcpu_id;
3589 		}
3590 		hvregs.hdec_expiry = time_limit;
3591 		trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3592 					  __pa(&vcpu->arch.regs));
3593 		kvmhv_restore_hv_return_state(vcpu, &hvregs);
3594 		vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3595 		vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3596 		vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3597 
3598 		/* H_CEDE has to be handled now, not later */
3599 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3600 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3601 			kvmppc_nested_cede(vcpu);
3602 			trap = 0;
3603 		}
3604 	} else {
3605 		trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3606 	}
3607 
3608 	vcpu->arch.slb_max = 0;
3609 	dec = mfspr(SPRN_DEC);
3610 	tb = mftb();
3611 	vcpu->arch.dec_expires = dec + tb;
3612 	vcpu->cpu = -1;
3613 	vcpu->arch.thread_cpu = -1;
3614 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3615 
3616 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
3617 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
3618 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
3619 	vcpu->arch.tar = mfspr(SPRN_TAR);
3620 	vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3621 	vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3622 	vcpu->arch.bescr = mfspr(SPRN_BESCR);
3623 	vcpu->arch.wort = mfspr(SPRN_WORT);
3624 	vcpu->arch.tid = mfspr(SPRN_TIDR);
3625 	vcpu->arch.amr = mfspr(SPRN_AMR);
3626 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3627 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
3628 
3629 	mtspr(SPRN_PSPB, 0);
3630 	mtspr(SPRN_WORT, 0);
3631 	mtspr(SPRN_UAMOR, 0);
3632 	mtspr(SPRN_DSCR, host_dscr);
3633 	mtspr(SPRN_TIDR, host_tidr);
3634 	mtspr(SPRN_IAMR, host_iamr);
3635 	mtspr(SPRN_PSPB, 0);
3636 
3637 	if (host_amr != vcpu->arch.amr)
3638 		mtspr(SPRN_AMR, host_amr);
3639 
3640 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3641 	store_fp_state(&vcpu->arch.fp);
3642 #ifdef CONFIG_ALTIVEC
3643 	store_vr_state(&vcpu->arch.vr);
3644 #endif
3645 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3646 
3647 	if (cpu_has_feature(CPU_FTR_TM) ||
3648 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3649 		kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3650 
3651 	save_pmu = 1;
3652 	if (vcpu->arch.vpa.pinned_addr) {
3653 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3654 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3655 		lp->yield_count = cpu_to_be32(yield_count);
3656 		vcpu->arch.vpa.dirty = 1;
3657 		save_pmu = lp->pmcregs_in_use;
3658 	}
3659 
3660 	kvmhv_save_guest_pmu(vcpu, save_pmu);
3661 
3662 	vc->entry_exit_map = 0x101;
3663 	vc->in_guest = 0;
3664 
3665 	mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3666 
3667 	kvmhv_load_host_pmu();
3668 
3669 	kvmppc_subcore_exit_guest();
3670 
3671 	return trap;
3672 }
3673 
3674 /*
3675  * Wait for some other vcpu thread to execute us, and
3676  * wake us up when we need to handle something in the host.
3677  */
3678 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3679 				 struct kvm_vcpu *vcpu, int wait_state)
3680 {
3681 	DEFINE_WAIT(wait);
3682 
3683 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3684 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3685 		spin_unlock(&vc->lock);
3686 		schedule();
3687 		spin_lock(&vc->lock);
3688 	}
3689 	finish_wait(&vcpu->arch.cpu_run, &wait);
3690 }
3691 
3692 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3693 {
3694 	if (!halt_poll_ns_grow)
3695 		return;
3696 
3697 	vc->halt_poll_ns *= halt_poll_ns_grow;
3698 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3699 		vc->halt_poll_ns = halt_poll_ns_grow_start;
3700 }
3701 
3702 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3703 {
3704 	if (halt_poll_ns_shrink == 0)
3705 		vc->halt_poll_ns = 0;
3706 	else
3707 		vc->halt_poll_ns /= halt_poll_ns_shrink;
3708 }
3709 
3710 #ifdef CONFIG_KVM_XICS
3711 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3712 {
3713 	if (!xics_on_xive())
3714 		return false;
3715 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3716 		vcpu->arch.xive_saved_state.cppr;
3717 }
3718 #else
3719 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3720 {
3721 	return false;
3722 }
3723 #endif /* CONFIG_KVM_XICS */
3724 
3725 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3726 {
3727 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3728 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3729 		return true;
3730 
3731 	return false;
3732 }
3733 
3734 /*
3735  * Check to see if any of the runnable vcpus on the vcore have pending
3736  * exceptions or are no longer ceded
3737  */
3738 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3739 {
3740 	struct kvm_vcpu *vcpu;
3741 	int i;
3742 
3743 	for_each_runnable_thread(i, vcpu, vc) {
3744 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3745 			return 1;
3746 	}
3747 
3748 	return 0;
3749 }
3750 
3751 /*
3752  * All the vcpus in this vcore are idle, so wait for a decrementer
3753  * or external interrupt to one of the vcpus.  vc->lock is held.
3754  */
3755 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3756 {
3757 	ktime_t cur, start_poll, start_wait;
3758 	int do_sleep = 1;
3759 	u64 block_ns;
3760 	DECLARE_SWAITQUEUE(wait);
3761 
3762 	/* Poll for pending exceptions and ceded state */
3763 	cur = start_poll = ktime_get();
3764 	if (vc->halt_poll_ns) {
3765 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3766 		++vc->runner->stat.halt_attempted_poll;
3767 
3768 		vc->vcore_state = VCORE_POLLING;
3769 		spin_unlock(&vc->lock);
3770 
3771 		do {
3772 			if (kvmppc_vcore_check_block(vc)) {
3773 				do_sleep = 0;
3774 				break;
3775 			}
3776 			cur = ktime_get();
3777 		} while (single_task_running() && ktime_before(cur, stop));
3778 
3779 		spin_lock(&vc->lock);
3780 		vc->vcore_state = VCORE_INACTIVE;
3781 
3782 		if (!do_sleep) {
3783 			++vc->runner->stat.halt_successful_poll;
3784 			goto out;
3785 		}
3786 	}
3787 
3788 	prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3789 
3790 	if (kvmppc_vcore_check_block(vc)) {
3791 		finish_swait(&vc->wq, &wait);
3792 		do_sleep = 0;
3793 		/* If we polled, count this as a successful poll */
3794 		if (vc->halt_poll_ns)
3795 			++vc->runner->stat.halt_successful_poll;
3796 		goto out;
3797 	}
3798 
3799 	start_wait = ktime_get();
3800 
3801 	vc->vcore_state = VCORE_SLEEPING;
3802 	trace_kvmppc_vcore_blocked(vc, 0);
3803 	spin_unlock(&vc->lock);
3804 	schedule();
3805 	finish_swait(&vc->wq, &wait);
3806 	spin_lock(&vc->lock);
3807 	vc->vcore_state = VCORE_INACTIVE;
3808 	trace_kvmppc_vcore_blocked(vc, 1);
3809 	++vc->runner->stat.halt_successful_wait;
3810 
3811 	cur = ktime_get();
3812 
3813 out:
3814 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3815 
3816 	/* Attribute wait time */
3817 	if (do_sleep) {
3818 		vc->runner->stat.halt_wait_ns +=
3819 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3820 		/* Attribute failed poll time */
3821 		if (vc->halt_poll_ns)
3822 			vc->runner->stat.halt_poll_fail_ns +=
3823 				ktime_to_ns(start_wait) -
3824 				ktime_to_ns(start_poll);
3825 	} else {
3826 		/* Attribute successful poll time */
3827 		if (vc->halt_poll_ns)
3828 			vc->runner->stat.halt_poll_success_ns +=
3829 				ktime_to_ns(cur) -
3830 				ktime_to_ns(start_poll);
3831 	}
3832 
3833 	/* Adjust poll time */
3834 	if (halt_poll_ns) {
3835 		if (block_ns <= vc->halt_poll_ns)
3836 			;
3837 		/* We slept and blocked for longer than the max halt time */
3838 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3839 			shrink_halt_poll_ns(vc);
3840 		/* We slept and our poll time is too small */
3841 		else if (vc->halt_poll_ns < halt_poll_ns &&
3842 				block_ns < halt_poll_ns)
3843 			grow_halt_poll_ns(vc);
3844 		if (vc->halt_poll_ns > halt_poll_ns)
3845 			vc->halt_poll_ns = halt_poll_ns;
3846 	} else
3847 		vc->halt_poll_ns = 0;
3848 
3849 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3850 }
3851 
3852 /*
3853  * This never fails for a radix guest, as none of the operations it does
3854  * for a radix guest can fail or have a way to report failure.
3855  * kvmhv_run_single_vcpu() relies on this fact.
3856  */
3857 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3858 {
3859 	int r = 0;
3860 	struct kvm *kvm = vcpu->kvm;
3861 
3862 	mutex_lock(&kvm->lock);
3863 	if (!kvm->arch.mmu_ready) {
3864 		if (!kvm_is_radix(kvm))
3865 			r = kvmppc_hv_setup_htab_rma(vcpu);
3866 		if (!r) {
3867 			if (cpu_has_feature(CPU_FTR_ARCH_300))
3868 				kvmppc_setup_partition_table(kvm);
3869 			kvm->arch.mmu_ready = 1;
3870 		}
3871 	}
3872 	mutex_unlock(&kvm->lock);
3873 	return r;
3874 }
3875 
3876 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3877 {
3878 	int n_ceded, i, r;
3879 	struct kvmppc_vcore *vc;
3880 	struct kvm_vcpu *v;
3881 
3882 	trace_kvmppc_run_vcpu_enter(vcpu);
3883 
3884 	kvm_run->exit_reason = 0;
3885 	vcpu->arch.ret = RESUME_GUEST;
3886 	vcpu->arch.trap = 0;
3887 	kvmppc_update_vpas(vcpu);
3888 
3889 	/*
3890 	 * Synchronize with other threads in this virtual core
3891 	 */
3892 	vc = vcpu->arch.vcore;
3893 	spin_lock(&vc->lock);
3894 	vcpu->arch.ceded = 0;
3895 	vcpu->arch.run_task = current;
3896 	vcpu->arch.kvm_run = kvm_run;
3897 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3898 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3899 	vcpu->arch.busy_preempt = TB_NIL;
3900 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3901 	++vc->n_runnable;
3902 
3903 	/*
3904 	 * This happens the first time this is called for a vcpu.
3905 	 * If the vcore is already running, we may be able to start
3906 	 * this thread straight away and have it join in.
3907 	 */
3908 	if (!signal_pending(current)) {
3909 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
3910 		     vc->vcore_state == VCORE_RUNNING) &&
3911 			   !VCORE_IS_EXITING(vc)) {
3912 			kvmppc_create_dtl_entry(vcpu, vc);
3913 			kvmppc_start_thread(vcpu, vc);
3914 			trace_kvm_guest_enter(vcpu);
3915 		} else if (vc->vcore_state == VCORE_SLEEPING) {
3916 			swake_up_one(&vc->wq);
3917 		}
3918 
3919 	}
3920 
3921 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3922 	       !signal_pending(current)) {
3923 		/* See if the MMU is ready to go */
3924 		if (!vcpu->kvm->arch.mmu_ready) {
3925 			spin_unlock(&vc->lock);
3926 			r = kvmhv_setup_mmu(vcpu);
3927 			spin_lock(&vc->lock);
3928 			if (r) {
3929 				kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3930 				kvm_run->fail_entry.
3931 					hardware_entry_failure_reason = 0;
3932 				vcpu->arch.ret = r;
3933 				break;
3934 			}
3935 		}
3936 
3937 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3938 			kvmppc_vcore_end_preempt(vc);
3939 
3940 		if (vc->vcore_state != VCORE_INACTIVE) {
3941 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3942 			continue;
3943 		}
3944 		for_each_runnable_thread(i, v, vc) {
3945 			kvmppc_core_prepare_to_enter(v);
3946 			if (signal_pending(v->arch.run_task)) {
3947 				kvmppc_remove_runnable(vc, v);
3948 				v->stat.signal_exits++;
3949 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3950 				v->arch.ret = -EINTR;
3951 				wake_up(&v->arch.cpu_run);
3952 			}
3953 		}
3954 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3955 			break;
3956 		n_ceded = 0;
3957 		for_each_runnable_thread(i, v, vc) {
3958 			if (!kvmppc_vcpu_woken(v))
3959 				n_ceded += v->arch.ceded;
3960 			else
3961 				v->arch.ceded = 0;
3962 		}
3963 		vc->runner = vcpu;
3964 		if (n_ceded == vc->n_runnable) {
3965 			kvmppc_vcore_blocked(vc);
3966 		} else if (need_resched()) {
3967 			kvmppc_vcore_preempt(vc);
3968 			/* Let something else run */
3969 			cond_resched_lock(&vc->lock);
3970 			if (vc->vcore_state == VCORE_PREEMPT)
3971 				kvmppc_vcore_end_preempt(vc);
3972 		} else {
3973 			kvmppc_run_core(vc);
3974 		}
3975 		vc->runner = NULL;
3976 	}
3977 
3978 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3979 	       (vc->vcore_state == VCORE_RUNNING ||
3980 		vc->vcore_state == VCORE_EXITING ||
3981 		vc->vcore_state == VCORE_PIGGYBACK))
3982 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3983 
3984 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3985 		kvmppc_vcore_end_preempt(vc);
3986 
3987 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3988 		kvmppc_remove_runnable(vc, vcpu);
3989 		vcpu->stat.signal_exits++;
3990 		kvm_run->exit_reason = KVM_EXIT_INTR;
3991 		vcpu->arch.ret = -EINTR;
3992 	}
3993 
3994 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3995 		/* Wake up some vcpu to run the core */
3996 		i = -1;
3997 		v = next_runnable_thread(vc, &i);
3998 		wake_up(&v->arch.cpu_run);
3999 	}
4000 
4001 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4002 	spin_unlock(&vc->lock);
4003 	return vcpu->arch.ret;
4004 }
4005 
4006 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4007 			  struct kvm_vcpu *vcpu, u64 time_limit,
4008 			  unsigned long lpcr)
4009 {
4010 	int trap, r, pcpu;
4011 	int srcu_idx, lpid;
4012 	struct kvmppc_vcore *vc;
4013 	struct kvm *kvm = vcpu->kvm;
4014 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4015 
4016 	trace_kvmppc_run_vcpu_enter(vcpu);
4017 
4018 	kvm_run->exit_reason = 0;
4019 	vcpu->arch.ret = RESUME_GUEST;
4020 	vcpu->arch.trap = 0;
4021 
4022 	vc = vcpu->arch.vcore;
4023 	vcpu->arch.ceded = 0;
4024 	vcpu->arch.run_task = current;
4025 	vcpu->arch.kvm_run = kvm_run;
4026 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4027 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4028 	vcpu->arch.busy_preempt = TB_NIL;
4029 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4030 	vc->runnable_threads[0] = vcpu;
4031 	vc->n_runnable = 1;
4032 	vc->runner = vcpu;
4033 
4034 	/* See if the MMU is ready to go */
4035 	if (!kvm->arch.mmu_ready)
4036 		kvmhv_setup_mmu(vcpu);
4037 
4038 	if (need_resched())
4039 		cond_resched();
4040 
4041 	kvmppc_update_vpas(vcpu);
4042 
4043 	init_vcore_to_run(vc);
4044 	vc->preempt_tb = TB_NIL;
4045 
4046 	preempt_disable();
4047 	pcpu = smp_processor_id();
4048 	vc->pcpu = pcpu;
4049 	kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4050 
4051 	local_irq_disable();
4052 	hard_irq_disable();
4053 	if (signal_pending(current))
4054 		goto sigpend;
4055 	if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4056 		goto out;
4057 
4058 	if (!nested) {
4059 		kvmppc_core_prepare_to_enter(vcpu);
4060 		if (vcpu->arch.doorbell_request) {
4061 			vc->dpdes = 1;
4062 			smp_wmb();
4063 			vcpu->arch.doorbell_request = 0;
4064 		}
4065 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4066 			     &vcpu->arch.pending_exceptions))
4067 			lpcr |= LPCR_MER;
4068 	} else if (vcpu->arch.pending_exceptions ||
4069 		   vcpu->arch.doorbell_request ||
4070 		   xive_interrupt_pending(vcpu)) {
4071 		vcpu->arch.ret = RESUME_HOST;
4072 		goto out;
4073 	}
4074 
4075 	kvmppc_clear_host_core(pcpu);
4076 
4077 	local_paca->kvm_hstate.tid = 0;
4078 	local_paca->kvm_hstate.napping = 0;
4079 	local_paca->kvm_hstate.kvm_split_mode = NULL;
4080 	kvmppc_start_thread(vcpu, vc);
4081 	kvmppc_create_dtl_entry(vcpu, vc);
4082 	trace_kvm_guest_enter(vcpu);
4083 
4084 	vc->vcore_state = VCORE_RUNNING;
4085 	trace_kvmppc_run_core(vc, 0);
4086 
4087 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4088 		lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4089 		mtspr(SPRN_LPID, lpid);
4090 		isync();
4091 		kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4092 	}
4093 
4094 	trace_hardirqs_on();
4095 	guest_enter_irqoff();
4096 
4097 	srcu_idx = srcu_read_lock(&kvm->srcu);
4098 
4099 	this_cpu_disable_ftrace();
4100 
4101 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4102 	vcpu->arch.trap = trap;
4103 
4104 	this_cpu_enable_ftrace();
4105 
4106 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4107 
4108 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4109 		mtspr(SPRN_LPID, kvm->arch.host_lpid);
4110 		isync();
4111 	}
4112 
4113 	trace_hardirqs_off();
4114 	set_irq_happened(trap);
4115 
4116 	kvmppc_set_host_core(pcpu);
4117 
4118 	local_irq_enable();
4119 	guest_exit();
4120 
4121 	cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4122 
4123 	preempt_enable();
4124 
4125 	/* cancel pending decrementer exception if DEC is now positive */
4126 	if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
4127 		kvmppc_core_dequeue_dec(vcpu);
4128 
4129 	trace_kvm_guest_exit(vcpu);
4130 	r = RESUME_GUEST;
4131 	if (trap) {
4132 		if (!nested)
4133 			r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4134 		else
4135 			r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4136 	}
4137 	vcpu->arch.ret = r;
4138 
4139 	if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4140 	    !kvmppc_vcpu_woken(vcpu)) {
4141 		kvmppc_set_timer(vcpu);
4142 		while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4143 			if (signal_pending(current)) {
4144 				vcpu->stat.signal_exits++;
4145 				kvm_run->exit_reason = KVM_EXIT_INTR;
4146 				vcpu->arch.ret = -EINTR;
4147 				break;
4148 			}
4149 			spin_lock(&vc->lock);
4150 			kvmppc_vcore_blocked(vc);
4151 			spin_unlock(&vc->lock);
4152 		}
4153 	}
4154 	vcpu->arch.ceded = 0;
4155 
4156 	vc->vcore_state = VCORE_INACTIVE;
4157 	trace_kvmppc_run_core(vc, 1);
4158 
4159  done:
4160 	kvmppc_remove_runnable(vc, vcpu);
4161 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4162 
4163 	return vcpu->arch.ret;
4164 
4165  sigpend:
4166 	vcpu->stat.signal_exits++;
4167 	kvm_run->exit_reason = KVM_EXIT_INTR;
4168 	vcpu->arch.ret = -EINTR;
4169  out:
4170 	local_irq_enable();
4171 	preempt_enable();
4172 	goto done;
4173 }
4174 
4175 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4176 {
4177 	int r;
4178 	int srcu_idx;
4179 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
4180 	unsigned long user_tar = 0;
4181 	unsigned int user_vrsave;
4182 	struct kvm *kvm;
4183 
4184 	if (!vcpu->arch.sane) {
4185 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4186 		return -EINVAL;
4187 	}
4188 
4189 	/*
4190 	 * Don't allow entry with a suspended transaction, because
4191 	 * the guest entry/exit code will lose it.
4192 	 * If the guest has TM enabled, save away their TM-related SPRs
4193 	 * (they will get restored by the TM unavailable interrupt).
4194 	 */
4195 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4196 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4197 	    (current->thread.regs->msr & MSR_TM)) {
4198 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4199 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4200 			run->fail_entry.hardware_entry_failure_reason = 0;
4201 			return -EINVAL;
4202 		}
4203 		/* Enable TM so we can read the TM SPRs */
4204 		mtmsr(mfmsr() | MSR_TM);
4205 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4206 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4207 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4208 		current->thread.regs->msr &= ~MSR_TM;
4209 	}
4210 #endif
4211 
4212 	/*
4213 	 * Force online to 1 for the sake of old userspace which doesn't
4214 	 * set it.
4215 	 */
4216 	if (!vcpu->arch.online) {
4217 		atomic_inc(&vcpu->arch.vcore->online_count);
4218 		vcpu->arch.online = 1;
4219 	}
4220 
4221 	kvmppc_core_prepare_to_enter(vcpu);
4222 
4223 	/* No need to go into the guest when all we'll do is come back out */
4224 	if (signal_pending(current)) {
4225 		run->exit_reason = KVM_EXIT_INTR;
4226 		return -EINTR;
4227 	}
4228 
4229 	kvm = vcpu->kvm;
4230 	atomic_inc(&kvm->arch.vcpus_running);
4231 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4232 	smp_mb();
4233 
4234 	flush_all_to_thread(current);
4235 
4236 	/* Save userspace EBB and other register values */
4237 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4238 		ebb_regs[0] = mfspr(SPRN_EBBHR);
4239 		ebb_regs[1] = mfspr(SPRN_EBBRR);
4240 		ebb_regs[2] = mfspr(SPRN_BESCR);
4241 		user_tar = mfspr(SPRN_TAR);
4242 	}
4243 	user_vrsave = mfspr(SPRN_VRSAVE);
4244 
4245 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4246 	vcpu->arch.pgdir = current->mm->pgd;
4247 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4248 
4249 	do {
4250 		/*
4251 		 * The early POWER9 chips that can't mix radix and HPT threads
4252 		 * on the same core also need the workaround for the problem
4253 		 * where the TLB would prefetch entries in the guest exit path
4254 		 * for radix guests using the guest PIDR value and LPID 0.
4255 		 * The workaround is in the old path (kvmppc_run_vcpu())
4256 		 * but not the new path (kvmhv_run_single_vcpu()).
4257 		 */
4258 		if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4259 		    !no_mixing_hpt_and_radix)
4260 			r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4261 						  vcpu->arch.vcore->lpcr);
4262 		else
4263 			r = kvmppc_run_vcpu(run, vcpu);
4264 
4265 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4266 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4267 			trace_kvm_hcall_enter(vcpu);
4268 			r = kvmppc_pseries_do_hcall(vcpu);
4269 			trace_kvm_hcall_exit(vcpu, r);
4270 			kvmppc_core_prepare_to_enter(vcpu);
4271 		} else if (r == RESUME_PAGE_FAULT) {
4272 			srcu_idx = srcu_read_lock(&kvm->srcu);
4273 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
4274 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4275 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4276 		} else if (r == RESUME_PASSTHROUGH) {
4277 			if (WARN_ON(xics_on_xive()))
4278 				r = H_SUCCESS;
4279 			else
4280 				r = kvmppc_xics_rm_complete(vcpu, 0);
4281 		}
4282 	} while (is_kvmppc_resume_guest(r));
4283 
4284 	/* Restore userspace EBB and other register values */
4285 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4286 		mtspr(SPRN_EBBHR, ebb_regs[0]);
4287 		mtspr(SPRN_EBBRR, ebb_regs[1]);
4288 		mtspr(SPRN_BESCR, ebb_regs[2]);
4289 		mtspr(SPRN_TAR, user_tar);
4290 		mtspr(SPRN_FSCR, current->thread.fscr);
4291 	}
4292 	mtspr(SPRN_VRSAVE, user_vrsave);
4293 
4294 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4295 	atomic_dec(&kvm->arch.vcpus_running);
4296 	return r;
4297 }
4298 
4299 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4300 				     int shift, int sllp)
4301 {
4302 	(*sps)->page_shift = shift;
4303 	(*sps)->slb_enc = sllp;
4304 	(*sps)->enc[0].page_shift = shift;
4305 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4306 	/*
4307 	 * Add 16MB MPSS support (may get filtered out by userspace)
4308 	 */
4309 	if (shift != 24) {
4310 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4311 		if (penc != -1) {
4312 			(*sps)->enc[1].page_shift = 24;
4313 			(*sps)->enc[1].pte_enc = penc;
4314 		}
4315 	}
4316 	(*sps)++;
4317 }
4318 
4319 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4320 					 struct kvm_ppc_smmu_info *info)
4321 {
4322 	struct kvm_ppc_one_seg_page_size *sps;
4323 
4324 	/*
4325 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4326 	 * POWER7 doesn't support keys for instruction accesses,
4327 	 * POWER8 and POWER9 do.
4328 	 */
4329 	info->data_keys = 32;
4330 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4331 
4332 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4333 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4334 	info->slb_size = 32;
4335 
4336 	/* We only support these sizes for now, and no muti-size segments */
4337 	sps = &info->sps[0];
4338 	kvmppc_add_seg_page_size(&sps, 12, 0);
4339 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4340 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4341 
4342 	/* If running as a nested hypervisor, we don't support HPT guests */
4343 	if (kvmhv_on_pseries())
4344 		info->flags |= KVM_PPC_NO_HASH;
4345 
4346 	return 0;
4347 }
4348 
4349 /*
4350  * Get (and clear) the dirty memory log for a memory slot.
4351  */
4352 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4353 					 struct kvm_dirty_log *log)
4354 {
4355 	struct kvm_memslots *slots;
4356 	struct kvm_memory_slot *memslot;
4357 	int i, r;
4358 	unsigned long n;
4359 	unsigned long *buf, *p;
4360 	struct kvm_vcpu *vcpu;
4361 
4362 	mutex_lock(&kvm->slots_lock);
4363 
4364 	r = -EINVAL;
4365 	if (log->slot >= KVM_USER_MEM_SLOTS)
4366 		goto out;
4367 
4368 	slots = kvm_memslots(kvm);
4369 	memslot = id_to_memslot(slots, log->slot);
4370 	r = -ENOENT;
4371 	if (!memslot->dirty_bitmap)
4372 		goto out;
4373 
4374 	/*
4375 	 * Use second half of bitmap area because both HPT and radix
4376 	 * accumulate bits in the first half.
4377 	 */
4378 	n = kvm_dirty_bitmap_bytes(memslot);
4379 	buf = memslot->dirty_bitmap + n / sizeof(long);
4380 	memset(buf, 0, n);
4381 
4382 	if (kvm_is_radix(kvm))
4383 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4384 	else
4385 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4386 	if (r)
4387 		goto out;
4388 
4389 	/*
4390 	 * We accumulate dirty bits in the first half of the
4391 	 * memslot's dirty_bitmap area, for when pages are paged
4392 	 * out or modified by the host directly.  Pick up these
4393 	 * bits and add them to the map.
4394 	 */
4395 	p = memslot->dirty_bitmap;
4396 	for (i = 0; i < n / sizeof(long); ++i)
4397 		buf[i] |= xchg(&p[i], 0);
4398 
4399 	/* Harvest dirty bits from VPA and DTL updates */
4400 	/* Note: we never modify the SLB shadow buffer areas */
4401 	kvm_for_each_vcpu(i, vcpu, kvm) {
4402 		spin_lock(&vcpu->arch.vpa_update_lock);
4403 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4404 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4405 		spin_unlock(&vcpu->arch.vpa_update_lock);
4406 	}
4407 
4408 	r = -EFAULT;
4409 	if (copy_to_user(log->dirty_bitmap, buf, n))
4410 		goto out;
4411 
4412 	r = 0;
4413 out:
4414 	mutex_unlock(&kvm->slots_lock);
4415 	return r;
4416 }
4417 
4418 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4419 					struct kvm_memory_slot *dont)
4420 {
4421 	if (!dont || free->arch.rmap != dont->arch.rmap) {
4422 		vfree(free->arch.rmap);
4423 		free->arch.rmap = NULL;
4424 	}
4425 }
4426 
4427 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4428 					 unsigned long npages)
4429 {
4430 	slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4431 	if (!slot->arch.rmap)
4432 		return -ENOMEM;
4433 
4434 	return 0;
4435 }
4436 
4437 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4438 					struct kvm_memory_slot *memslot,
4439 					const struct kvm_userspace_memory_region *mem)
4440 {
4441 	return 0;
4442 }
4443 
4444 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4445 				const struct kvm_userspace_memory_region *mem,
4446 				const struct kvm_memory_slot *old,
4447 				const struct kvm_memory_slot *new,
4448 				enum kvm_mr_change change)
4449 {
4450 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4451 
4452 	/*
4453 	 * If we are making a new memslot, it might make
4454 	 * some address that was previously cached as emulated
4455 	 * MMIO be no longer emulated MMIO, so invalidate
4456 	 * all the caches of emulated MMIO translations.
4457 	 */
4458 	if (npages)
4459 		atomic64_inc(&kvm->arch.mmio_update);
4460 
4461 	/*
4462 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4463 	 * have already called kvm_arch_flush_shadow_memslot() to
4464 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
4465 	 * previous mappings.  So the only case to handle is
4466 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4467 	 * has been changed.
4468 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4469 	 * to get rid of any THP PTEs in the partition-scoped page tables
4470 	 * so we can track dirtiness at the page level; we flush when
4471 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4472 	 * using THP PTEs.
4473 	 */
4474 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4475 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4476 		kvmppc_radix_flush_memslot(kvm, old);
4477 }
4478 
4479 /*
4480  * Update LPCR values in kvm->arch and in vcores.
4481  * Caller must hold kvm->lock.
4482  */
4483 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4484 {
4485 	long int i;
4486 	u32 cores_done = 0;
4487 
4488 	if ((kvm->arch.lpcr & mask) == lpcr)
4489 		return;
4490 
4491 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4492 
4493 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4494 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4495 		if (!vc)
4496 			continue;
4497 		spin_lock(&vc->lock);
4498 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4499 		spin_unlock(&vc->lock);
4500 		if (++cores_done >= kvm->arch.online_vcores)
4501 			break;
4502 	}
4503 }
4504 
4505 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4506 {
4507 	return;
4508 }
4509 
4510 void kvmppc_setup_partition_table(struct kvm *kvm)
4511 {
4512 	unsigned long dw0, dw1;
4513 
4514 	if (!kvm_is_radix(kvm)) {
4515 		/* PS field - page size for VRMA */
4516 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4517 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4518 		/* HTABSIZE and HTABORG fields */
4519 		dw0 |= kvm->arch.sdr1;
4520 
4521 		/* Second dword as set by userspace */
4522 		dw1 = kvm->arch.process_table;
4523 	} else {
4524 		dw0 = PATB_HR | radix__get_tree_size() |
4525 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4526 		dw1 = PATB_GR | kvm->arch.process_table;
4527 	}
4528 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4529 }
4530 
4531 /*
4532  * Set up HPT (hashed page table) and RMA (real-mode area).
4533  * Must be called with kvm->lock held.
4534  */
4535 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4536 {
4537 	int err = 0;
4538 	struct kvm *kvm = vcpu->kvm;
4539 	unsigned long hva;
4540 	struct kvm_memory_slot *memslot;
4541 	struct vm_area_struct *vma;
4542 	unsigned long lpcr = 0, senc;
4543 	unsigned long psize, porder;
4544 	int srcu_idx;
4545 
4546 	/* Allocate hashed page table (if not done already) and reset it */
4547 	if (!kvm->arch.hpt.virt) {
4548 		int order = KVM_DEFAULT_HPT_ORDER;
4549 		struct kvm_hpt_info info;
4550 
4551 		err = kvmppc_allocate_hpt(&info, order);
4552 		/* If we get here, it means userspace didn't specify a
4553 		 * size explicitly.  So, try successively smaller
4554 		 * sizes if the default failed. */
4555 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4556 			err  = kvmppc_allocate_hpt(&info, order);
4557 
4558 		if (err < 0) {
4559 			pr_err("KVM: Couldn't alloc HPT\n");
4560 			goto out;
4561 		}
4562 
4563 		kvmppc_set_hpt(kvm, &info);
4564 	}
4565 
4566 	/* Look up the memslot for guest physical address 0 */
4567 	srcu_idx = srcu_read_lock(&kvm->srcu);
4568 	memslot = gfn_to_memslot(kvm, 0);
4569 
4570 	/* We must have some memory at 0 by now */
4571 	err = -EINVAL;
4572 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4573 		goto out_srcu;
4574 
4575 	/* Look up the VMA for the start of this memory slot */
4576 	hva = memslot->userspace_addr;
4577 	down_read(&current->mm->mmap_sem);
4578 	vma = find_vma(current->mm, hva);
4579 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4580 		goto up_out;
4581 
4582 	psize = vma_kernel_pagesize(vma);
4583 
4584 	up_read(&current->mm->mmap_sem);
4585 
4586 	/* We can handle 4k, 64k or 16M pages in the VRMA */
4587 	if (psize >= 0x1000000)
4588 		psize = 0x1000000;
4589 	else if (psize >= 0x10000)
4590 		psize = 0x10000;
4591 	else
4592 		psize = 0x1000;
4593 	porder = __ilog2(psize);
4594 
4595 	senc = slb_pgsize_encoding(psize);
4596 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4597 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4598 	/* Create HPTEs in the hash page table for the VRMA */
4599 	kvmppc_map_vrma(vcpu, memslot, porder);
4600 
4601 	/* Update VRMASD field in the LPCR */
4602 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4603 		/* the -4 is to account for senc values starting at 0x10 */
4604 		lpcr = senc << (LPCR_VRMASD_SH - 4);
4605 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4606 	}
4607 
4608 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4609 	smp_wmb();
4610 	err = 0;
4611  out_srcu:
4612 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4613  out:
4614 	return err;
4615 
4616  up_out:
4617 	up_read(&current->mm->mmap_sem);
4618 	goto out_srcu;
4619 }
4620 
4621 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4622 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4623 {
4624 	if (nesting_enabled(kvm))
4625 		kvmhv_release_all_nested(kvm);
4626 	kvmppc_rmap_reset(kvm);
4627 	kvm->arch.process_table = 0;
4628 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4629 	spin_lock(&kvm->mmu_lock);
4630 	kvm->arch.radix = 0;
4631 	spin_unlock(&kvm->mmu_lock);
4632 	kvmppc_free_radix(kvm);
4633 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
4634 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4635 	return 0;
4636 }
4637 
4638 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4639 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4640 {
4641 	int err;
4642 
4643 	err = kvmppc_init_vm_radix(kvm);
4644 	if (err)
4645 		return err;
4646 	kvmppc_rmap_reset(kvm);
4647 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4648 	spin_lock(&kvm->mmu_lock);
4649 	kvm->arch.radix = 1;
4650 	spin_unlock(&kvm->mmu_lock);
4651 	kvmppc_free_hpt(&kvm->arch.hpt);
4652 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4653 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4654 	return 0;
4655 }
4656 
4657 #ifdef CONFIG_KVM_XICS
4658 /*
4659  * Allocate a per-core structure for managing state about which cores are
4660  * running in the host versus the guest and for exchanging data between
4661  * real mode KVM and CPU running in the host.
4662  * This is only done for the first VM.
4663  * The allocated structure stays even if all VMs have stopped.
4664  * It is only freed when the kvm-hv module is unloaded.
4665  * It's OK for this routine to fail, we just don't support host
4666  * core operations like redirecting H_IPI wakeups.
4667  */
4668 void kvmppc_alloc_host_rm_ops(void)
4669 {
4670 	struct kvmppc_host_rm_ops *ops;
4671 	unsigned long l_ops;
4672 	int cpu, core;
4673 	int size;
4674 
4675 	/* Not the first time here ? */
4676 	if (kvmppc_host_rm_ops_hv != NULL)
4677 		return;
4678 
4679 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4680 	if (!ops)
4681 		return;
4682 
4683 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4684 	ops->rm_core = kzalloc(size, GFP_KERNEL);
4685 
4686 	if (!ops->rm_core) {
4687 		kfree(ops);
4688 		return;
4689 	}
4690 
4691 	cpus_read_lock();
4692 
4693 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4694 		if (!cpu_online(cpu))
4695 			continue;
4696 
4697 		core = cpu >> threads_shift;
4698 		ops->rm_core[core].rm_state.in_host = 1;
4699 	}
4700 
4701 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4702 
4703 	/*
4704 	 * Make the contents of the kvmppc_host_rm_ops structure visible
4705 	 * to other CPUs before we assign it to the global variable.
4706 	 * Do an atomic assignment (no locks used here), but if someone
4707 	 * beats us to it, just free our copy and return.
4708 	 */
4709 	smp_wmb();
4710 	l_ops = (unsigned long) ops;
4711 
4712 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4713 		cpus_read_unlock();
4714 		kfree(ops->rm_core);
4715 		kfree(ops);
4716 		return;
4717 	}
4718 
4719 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4720 					     "ppc/kvm_book3s:prepare",
4721 					     kvmppc_set_host_core,
4722 					     kvmppc_clear_host_core);
4723 	cpus_read_unlock();
4724 }
4725 
4726 void kvmppc_free_host_rm_ops(void)
4727 {
4728 	if (kvmppc_host_rm_ops_hv) {
4729 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4730 		kfree(kvmppc_host_rm_ops_hv->rm_core);
4731 		kfree(kvmppc_host_rm_ops_hv);
4732 		kvmppc_host_rm_ops_hv = NULL;
4733 	}
4734 }
4735 #endif
4736 
4737 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4738 {
4739 	unsigned long lpcr, lpid;
4740 	char buf[32];
4741 	int ret;
4742 
4743 	/* Allocate the guest's logical partition ID */
4744 
4745 	lpid = kvmppc_alloc_lpid();
4746 	if ((long)lpid < 0)
4747 		return -ENOMEM;
4748 	kvm->arch.lpid = lpid;
4749 
4750 	kvmppc_alloc_host_rm_ops();
4751 
4752 	kvmhv_vm_nested_init(kvm);
4753 
4754 	/*
4755 	 * Since we don't flush the TLB when tearing down a VM,
4756 	 * and this lpid might have previously been used,
4757 	 * make sure we flush on each core before running the new VM.
4758 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4759 	 * does this flush for us.
4760 	 */
4761 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4762 		cpumask_setall(&kvm->arch.need_tlb_flush);
4763 
4764 	/* Start out with the default set of hcalls enabled */
4765 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4766 	       sizeof(kvm->arch.enabled_hcalls));
4767 
4768 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4769 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4770 
4771 	/* Init LPCR for virtual RMA mode */
4772 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4773 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
4774 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4775 		lpcr &= LPCR_PECE | LPCR_LPES;
4776 	} else {
4777 		lpcr = 0;
4778 	}
4779 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4780 		LPCR_VPM0 | LPCR_VPM1;
4781 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4782 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4783 	/* On POWER8 turn on online bit to enable PURR/SPURR */
4784 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
4785 		lpcr |= LPCR_ONL;
4786 	/*
4787 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4788 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
4789 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4790 	 * be unnecessary but better safe than sorry in case we re-enable
4791 	 * EE in HV mode with this LPCR still set)
4792 	 */
4793 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4794 		lpcr &= ~LPCR_VPM0;
4795 		lpcr |= LPCR_HVICE | LPCR_HEIC;
4796 
4797 		/*
4798 		 * If xive is enabled, we route 0x500 interrupts directly
4799 		 * to the guest.
4800 		 */
4801 		if (xics_on_xive())
4802 			lpcr |= LPCR_LPES;
4803 	}
4804 
4805 	/*
4806 	 * If the host uses radix, the guest starts out as radix.
4807 	 */
4808 	if (radix_enabled()) {
4809 		kvm->arch.radix = 1;
4810 		kvm->arch.mmu_ready = 1;
4811 		lpcr &= ~LPCR_VPM1;
4812 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4813 		ret = kvmppc_init_vm_radix(kvm);
4814 		if (ret) {
4815 			kvmppc_free_lpid(kvm->arch.lpid);
4816 			return ret;
4817 		}
4818 		kvmppc_setup_partition_table(kvm);
4819 	}
4820 
4821 	kvm->arch.lpcr = lpcr;
4822 
4823 	/* Initialization for future HPT resizes */
4824 	kvm->arch.resize_hpt = NULL;
4825 
4826 	/*
4827 	 * Work out how many sets the TLB has, for the use of
4828 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4829 	 */
4830 	if (radix_enabled())
4831 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
4832 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
4833 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
4834 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4835 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
4836 	else
4837 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
4838 
4839 	/*
4840 	 * Track that we now have a HV mode VM active. This blocks secondary
4841 	 * CPU threads from coming online.
4842 	 * On POWER9, we only need to do this if the "indep_threads_mode"
4843 	 * module parameter has been set to N.
4844 	 */
4845 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4846 		if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4847 			pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4848 			kvm->arch.threads_indep = true;
4849 		} else {
4850 			kvm->arch.threads_indep = indep_threads_mode;
4851 		}
4852 	}
4853 	if (!kvm->arch.threads_indep)
4854 		kvm_hv_vm_activated();
4855 
4856 	/*
4857 	 * Initialize smt_mode depending on processor.
4858 	 * POWER8 and earlier have to use "strict" threading, where
4859 	 * all vCPUs in a vcore have to run on the same (sub)core,
4860 	 * whereas on POWER9 the threads can each run a different
4861 	 * guest.
4862 	 */
4863 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4864 		kvm->arch.smt_mode = threads_per_subcore;
4865 	else
4866 		kvm->arch.smt_mode = 1;
4867 	kvm->arch.emul_smt_mode = 1;
4868 
4869 	/*
4870 	 * Create a debugfs directory for the VM
4871 	 */
4872 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
4873 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4874 	kvmppc_mmu_debugfs_init(kvm);
4875 	if (radix_enabled())
4876 		kvmhv_radix_debugfs_init(kvm);
4877 
4878 	return 0;
4879 }
4880 
4881 static void kvmppc_free_vcores(struct kvm *kvm)
4882 {
4883 	long int i;
4884 
4885 	for (i = 0; i < KVM_MAX_VCORES; ++i)
4886 		kfree(kvm->arch.vcores[i]);
4887 	kvm->arch.online_vcores = 0;
4888 }
4889 
4890 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4891 {
4892 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
4893 
4894 	if (!kvm->arch.threads_indep)
4895 		kvm_hv_vm_deactivated();
4896 
4897 	kvmppc_free_vcores(kvm);
4898 
4899 
4900 	if (kvm_is_radix(kvm))
4901 		kvmppc_free_radix(kvm);
4902 	else
4903 		kvmppc_free_hpt(&kvm->arch.hpt);
4904 
4905 	/* Perform global invalidation and return lpid to the pool */
4906 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4907 		if (nesting_enabled(kvm))
4908 			kvmhv_release_all_nested(kvm);
4909 		kvm->arch.process_table = 0;
4910 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4911 	}
4912 	kvmppc_free_lpid(kvm->arch.lpid);
4913 
4914 	kvmppc_free_pimap(kvm);
4915 }
4916 
4917 /* We don't need to emulate any privileged instructions or dcbz */
4918 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4919 				     unsigned int inst, int *advance)
4920 {
4921 	return EMULATE_FAIL;
4922 }
4923 
4924 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4925 					ulong spr_val)
4926 {
4927 	return EMULATE_FAIL;
4928 }
4929 
4930 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4931 					ulong *spr_val)
4932 {
4933 	return EMULATE_FAIL;
4934 }
4935 
4936 static int kvmppc_core_check_processor_compat_hv(void)
4937 {
4938 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
4939 	    cpu_has_feature(CPU_FTR_ARCH_206))
4940 		return 0;
4941 
4942 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
4943 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4944 		return 0;
4945 
4946 	return -EIO;
4947 }
4948 
4949 #ifdef CONFIG_KVM_XICS
4950 
4951 void kvmppc_free_pimap(struct kvm *kvm)
4952 {
4953 	kfree(kvm->arch.pimap);
4954 }
4955 
4956 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4957 {
4958 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4959 }
4960 
4961 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4962 {
4963 	struct irq_desc *desc;
4964 	struct kvmppc_irq_map *irq_map;
4965 	struct kvmppc_passthru_irqmap *pimap;
4966 	struct irq_chip *chip;
4967 	int i, rc = 0;
4968 
4969 	if (!kvm_irq_bypass)
4970 		return 1;
4971 
4972 	desc = irq_to_desc(host_irq);
4973 	if (!desc)
4974 		return -EIO;
4975 
4976 	mutex_lock(&kvm->lock);
4977 
4978 	pimap = kvm->arch.pimap;
4979 	if (pimap == NULL) {
4980 		/* First call, allocate structure to hold IRQ map */
4981 		pimap = kvmppc_alloc_pimap();
4982 		if (pimap == NULL) {
4983 			mutex_unlock(&kvm->lock);
4984 			return -ENOMEM;
4985 		}
4986 		kvm->arch.pimap = pimap;
4987 	}
4988 
4989 	/*
4990 	 * For now, we only support interrupts for which the EOI operation
4991 	 * is an OPAL call followed by a write to XIRR, since that's
4992 	 * what our real-mode EOI code does, or a XIVE interrupt
4993 	 */
4994 	chip = irq_data_get_irq_chip(&desc->irq_data);
4995 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4996 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4997 			host_irq, guest_gsi);
4998 		mutex_unlock(&kvm->lock);
4999 		return -ENOENT;
5000 	}
5001 
5002 	/*
5003 	 * See if we already have an entry for this guest IRQ number.
5004 	 * If it's mapped to a hardware IRQ number, that's an error,
5005 	 * otherwise re-use this entry.
5006 	 */
5007 	for (i = 0; i < pimap->n_mapped; i++) {
5008 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5009 			if (pimap->mapped[i].r_hwirq) {
5010 				mutex_unlock(&kvm->lock);
5011 				return -EINVAL;
5012 			}
5013 			break;
5014 		}
5015 	}
5016 
5017 	if (i == KVMPPC_PIRQ_MAPPED) {
5018 		mutex_unlock(&kvm->lock);
5019 		return -EAGAIN;		/* table is full */
5020 	}
5021 
5022 	irq_map = &pimap->mapped[i];
5023 
5024 	irq_map->v_hwirq = guest_gsi;
5025 	irq_map->desc = desc;
5026 
5027 	/*
5028 	 * Order the above two stores before the next to serialize with
5029 	 * the KVM real mode handler.
5030 	 */
5031 	smp_wmb();
5032 	irq_map->r_hwirq = desc->irq_data.hwirq;
5033 
5034 	if (i == pimap->n_mapped)
5035 		pimap->n_mapped++;
5036 
5037 	if (xics_on_xive())
5038 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5039 	else
5040 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5041 	if (rc)
5042 		irq_map->r_hwirq = 0;
5043 
5044 	mutex_unlock(&kvm->lock);
5045 
5046 	return 0;
5047 }
5048 
5049 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5050 {
5051 	struct irq_desc *desc;
5052 	struct kvmppc_passthru_irqmap *pimap;
5053 	int i, rc = 0;
5054 
5055 	if (!kvm_irq_bypass)
5056 		return 0;
5057 
5058 	desc = irq_to_desc(host_irq);
5059 	if (!desc)
5060 		return -EIO;
5061 
5062 	mutex_lock(&kvm->lock);
5063 	if (!kvm->arch.pimap)
5064 		goto unlock;
5065 
5066 	pimap = kvm->arch.pimap;
5067 
5068 	for (i = 0; i < pimap->n_mapped; i++) {
5069 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5070 			break;
5071 	}
5072 
5073 	if (i == pimap->n_mapped) {
5074 		mutex_unlock(&kvm->lock);
5075 		return -ENODEV;
5076 	}
5077 
5078 	if (xics_on_xive())
5079 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5080 	else
5081 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5082 
5083 	/* invalidate the entry (what do do on error from the above ?) */
5084 	pimap->mapped[i].r_hwirq = 0;
5085 
5086 	/*
5087 	 * We don't free this structure even when the count goes to
5088 	 * zero. The structure is freed when we destroy the VM.
5089 	 */
5090  unlock:
5091 	mutex_unlock(&kvm->lock);
5092 	return rc;
5093 }
5094 
5095 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5096 					     struct irq_bypass_producer *prod)
5097 {
5098 	int ret = 0;
5099 	struct kvm_kernel_irqfd *irqfd =
5100 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5101 
5102 	irqfd->producer = prod;
5103 
5104 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5105 	if (ret)
5106 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5107 			prod->irq, irqfd->gsi, ret);
5108 
5109 	return ret;
5110 }
5111 
5112 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5113 					      struct irq_bypass_producer *prod)
5114 {
5115 	int ret;
5116 	struct kvm_kernel_irqfd *irqfd =
5117 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5118 
5119 	irqfd->producer = NULL;
5120 
5121 	/*
5122 	 * When producer of consumer is unregistered, we change back to
5123 	 * default external interrupt handling mode - KVM real mode
5124 	 * will switch back to host.
5125 	 */
5126 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5127 	if (ret)
5128 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5129 			prod->irq, irqfd->gsi, ret);
5130 }
5131 #endif
5132 
5133 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5134 				 unsigned int ioctl, unsigned long arg)
5135 {
5136 	struct kvm *kvm __maybe_unused = filp->private_data;
5137 	void __user *argp = (void __user *)arg;
5138 	long r;
5139 
5140 	switch (ioctl) {
5141 
5142 	case KVM_PPC_ALLOCATE_HTAB: {
5143 		u32 htab_order;
5144 
5145 		r = -EFAULT;
5146 		if (get_user(htab_order, (u32 __user *)argp))
5147 			break;
5148 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5149 		if (r)
5150 			break;
5151 		r = 0;
5152 		break;
5153 	}
5154 
5155 	case KVM_PPC_GET_HTAB_FD: {
5156 		struct kvm_get_htab_fd ghf;
5157 
5158 		r = -EFAULT;
5159 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5160 			break;
5161 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5162 		break;
5163 	}
5164 
5165 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5166 		struct kvm_ppc_resize_hpt rhpt;
5167 
5168 		r = -EFAULT;
5169 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5170 			break;
5171 
5172 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5173 		break;
5174 	}
5175 
5176 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5177 		struct kvm_ppc_resize_hpt rhpt;
5178 
5179 		r = -EFAULT;
5180 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5181 			break;
5182 
5183 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5184 		break;
5185 	}
5186 
5187 	default:
5188 		r = -ENOTTY;
5189 	}
5190 
5191 	return r;
5192 }
5193 
5194 /*
5195  * List of hcall numbers to enable by default.
5196  * For compatibility with old userspace, we enable by default
5197  * all hcalls that were implemented before the hcall-enabling
5198  * facility was added.  Note this list should not include H_RTAS.
5199  */
5200 static unsigned int default_hcall_list[] = {
5201 	H_REMOVE,
5202 	H_ENTER,
5203 	H_READ,
5204 	H_PROTECT,
5205 	H_BULK_REMOVE,
5206 	H_GET_TCE,
5207 	H_PUT_TCE,
5208 	H_SET_DABR,
5209 	H_SET_XDABR,
5210 	H_CEDE,
5211 	H_PROD,
5212 	H_CONFER,
5213 	H_REGISTER_VPA,
5214 #ifdef CONFIG_KVM_XICS
5215 	H_EOI,
5216 	H_CPPR,
5217 	H_IPI,
5218 	H_IPOLL,
5219 	H_XIRR,
5220 	H_XIRR_X,
5221 #endif
5222 	0
5223 };
5224 
5225 static void init_default_hcalls(void)
5226 {
5227 	int i;
5228 	unsigned int hcall;
5229 
5230 	for (i = 0; default_hcall_list[i]; ++i) {
5231 		hcall = default_hcall_list[i];
5232 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5233 		__set_bit(hcall / 4, default_enabled_hcalls);
5234 	}
5235 }
5236 
5237 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5238 {
5239 	unsigned long lpcr;
5240 	int radix;
5241 	int err;
5242 
5243 	/* If not on a POWER9, reject it */
5244 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5245 		return -ENODEV;
5246 
5247 	/* If any unknown flags set, reject it */
5248 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5249 		return -EINVAL;
5250 
5251 	/* GR (guest radix) bit in process_table field must match */
5252 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5253 	if (!!(cfg->process_table & PATB_GR) != radix)
5254 		return -EINVAL;
5255 
5256 	/* Process table size field must be reasonable, i.e. <= 24 */
5257 	if ((cfg->process_table & PRTS_MASK) > 24)
5258 		return -EINVAL;
5259 
5260 	/* We can change a guest to/from radix now, if the host is radix */
5261 	if (radix && !radix_enabled())
5262 		return -EINVAL;
5263 
5264 	/* If we're a nested hypervisor, we currently only support radix */
5265 	if (kvmhv_on_pseries() && !radix)
5266 		return -EINVAL;
5267 
5268 	mutex_lock(&kvm->lock);
5269 	if (radix != kvm_is_radix(kvm)) {
5270 		if (kvm->arch.mmu_ready) {
5271 			kvm->arch.mmu_ready = 0;
5272 			/* order mmu_ready vs. vcpus_running */
5273 			smp_mb();
5274 			if (atomic_read(&kvm->arch.vcpus_running)) {
5275 				kvm->arch.mmu_ready = 1;
5276 				err = -EBUSY;
5277 				goto out_unlock;
5278 			}
5279 		}
5280 		if (radix)
5281 			err = kvmppc_switch_mmu_to_radix(kvm);
5282 		else
5283 			err = kvmppc_switch_mmu_to_hpt(kvm);
5284 		if (err)
5285 			goto out_unlock;
5286 	}
5287 
5288 	kvm->arch.process_table = cfg->process_table;
5289 	kvmppc_setup_partition_table(kvm);
5290 
5291 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5292 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5293 	err = 0;
5294 
5295  out_unlock:
5296 	mutex_unlock(&kvm->lock);
5297 	return err;
5298 }
5299 
5300 static int kvmhv_enable_nested(struct kvm *kvm)
5301 {
5302 	if (!nested)
5303 		return -EPERM;
5304 	if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5305 		return -ENODEV;
5306 
5307 	/* kvm == NULL means the caller is testing if the capability exists */
5308 	if (kvm)
5309 		kvm->arch.nested_enable = true;
5310 	return 0;
5311 }
5312 
5313 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5314 				 int size)
5315 {
5316 	int rc = -EINVAL;
5317 
5318 	if (kvmhv_vcpu_is_radix(vcpu)) {
5319 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5320 
5321 		if (rc > 0)
5322 			rc = -EINVAL;
5323 	}
5324 
5325 	/* For now quadrants are the only way to access nested guest memory */
5326 	if (rc && vcpu->arch.nested)
5327 		rc = -EAGAIN;
5328 
5329 	return rc;
5330 }
5331 
5332 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5333 				int size)
5334 {
5335 	int rc = -EINVAL;
5336 
5337 	if (kvmhv_vcpu_is_radix(vcpu)) {
5338 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5339 
5340 		if (rc > 0)
5341 			rc = -EINVAL;
5342 	}
5343 
5344 	/* For now quadrants are the only way to access nested guest memory */
5345 	if (rc && vcpu->arch.nested)
5346 		rc = -EAGAIN;
5347 
5348 	return rc;
5349 }
5350 
5351 static struct kvmppc_ops kvm_ops_hv = {
5352 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5353 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5354 	.get_one_reg = kvmppc_get_one_reg_hv,
5355 	.set_one_reg = kvmppc_set_one_reg_hv,
5356 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
5357 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
5358 	.set_msr     = kvmppc_set_msr_hv,
5359 	.vcpu_run    = kvmppc_vcpu_run_hv,
5360 	.vcpu_create = kvmppc_core_vcpu_create_hv,
5361 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
5362 	.check_requests = kvmppc_core_check_requests_hv,
5363 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5364 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
5365 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5366 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5367 	.unmap_hva_range = kvm_unmap_hva_range_hv,
5368 	.age_hva  = kvm_age_hva_hv,
5369 	.test_age_hva = kvm_test_age_hva_hv,
5370 	.set_spte_hva = kvm_set_spte_hva_hv,
5371 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
5372 	.free_memslot = kvmppc_core_free_memslot_hv,
5373 	.create_memslot = kvmppc_core_create_memslot_hv,
5374 	.init_vm =  kvmppc_core_init_vm_hv,
5375 	.destroy_vm = kvmppc_core_destroy_vm_hv,
5376 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5377 	.emulate_op = kvmppc_core_emulate_op_hv,
5378 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5379 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5380 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5381 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5382 	.hcall_implemented = kvmppc_hcall_impl_hv,
5383 #ifdef CONFIG_KVM_XICS
5384 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5385 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5386 #endif
5387 	.configure_mmu = kvmhv_configure_mmu,
5388 	.get_rmmu_info = kvmhv_get_rmmu_info,
5389 	.set_smt_mode = kvmhv_set_smt_mode,
5390 	.enable_nested = kvmhv_enable_nested,
5391 	.load_from_eaddr = kvmhv_load_from_eaddr,
5392 	.store_to_eaddr = kvmhv_store_to_eaddr,
5393 };
5394 
5395 static int kvm_init_subcore_bitmap(void)
5396 {
5397 	int i, j;
5398 	int nr_cores = cpu_nr_cores();
5399 	struct sibling_subcore_state *sibling_subcore_state;
5400 
5401 	for (i = 0; i < nr_cores; i++) {
5402 		int first_cpu = i * threads_per_core;
5403 		int node = cpu_to_node(first_cpu);
5404 
5405 		/* Ignore if it is already allocated. */
5406 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
5407 			continue;
5408 
5409 		sibling_subcore_state =
5410 			kzalloc_node(sizeof(struct sibling_subcore_state),
5411 							GFP_KERNEL, node);
5412 		if (!sibling_subcore_state)
5413 			return -ENOMEM;
5414 
5415 
5416 		for (j = 0; j < threads_per_core; j++) {
5417 			int cpu = first_cpu + j;
5418 
5419 			paca_ptrs[cpu]->sibling_subcore_state =
5420 						sibling_subcore_state;
5421 		}
5422 	}
5423 	return 0;
5424 }
5425 
5426 static int kvmppc_radix_possible(void)
5427 {
5428 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5429 }
5430 
5431 static int kvmppc_book3s_init_hv(void)
5432 {
5433 	int r;
5434 	/*
5435 	 * FIXME!! Do we need to check on all cpus ?
5436 	 */
5437 	r = kvmppc_core_check_processor_compat_hv();
5438 	if (r < 0)
5439 		return -ENODEV;
5440 
5441 	r = kvmhv_nested_init();
5442 	if (r)
5443 		return r;
5444 
5445 	r = kvm_init_subcore_bitmap();
5446 	if (r)
5447 		return r;
5448 
5449 	/*
5450 	 * We need a way of accessing the XICS interrupt controller,
5451 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5452 	 * indirectly, via OPAL.
5453 	 */
5454 #ifdef CONFIG_SMP
5455 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
5456 	    !local_paca->kvm_hstate.xics_phys) {
5457 		struct device_node *np;
5458 
5459 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5460 		if (!np) {
5461 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5462 			return -ENODEV;
5463 		}
5464 		/* presence of intc confirmed - node can be dropped again */
5465 		of_node_put(np);
5466 	}
5467 #endif
5468 
5469 	kvm_ops_hv.owner = THIS_MODULE;
5470 	kvmppc_hv_ops = &kvm_ops_hv;
5471 
5472 	init_default_hcalls();
5473 
5474 	init_vcore_lists();
5475 
5476 	r = kvmppc_mmu_hv_init();
5477 	if (r)
5478 		return r;
5479 
5480 	if (kvmppc_radix_possible())
5481 		r = kvmppc_radix_init();
5482 
5483 	/*
5484 	 * POWER9 chips before version 2.02 can't have some threads in
5485 	 * HPT mode and some in radix mode on the same core.
5486 	 */
5487 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5488 		unsigned int pvr = mfspr(SPRN_PVR);
5489 		if ((pvr >> 16) == PVR_POWER9 &&
5490 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5491 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5492 			no_mixing_hpt_and_radix = true;
5493 	}
5494 
5495 	return r;
5496 }
5497 
5498 static void kvmppc_book3s_exit_hv(void)
5499 {
5500 	kvmppc_free_host_rm_ops();
5501 	if (kvmppc_radix_possible())
5502 		kvmppc_radix_exit();
5503 	kvmppc_hv_ops = NULL;
5504 	kvmhv_nested_exit();
5505 }
5506 
5507 module_init(kvmppc_book3s_init_hv);
5508 module_exit(kvmppc_book3s_exit_hv);
5509 MODULE_LICENSE("GPL");
5510 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5511 MODULE_ALIAS("devname:kvm");
5512