xref: /linux/arch/powerpc/kvm/book3s_hv.c (revision bdd1a21b52557ea8f61d0a5dc2f77151b576eb70)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
5  *
6  * Authors:
7  *    Paul Mackerras <paulus@au1.ibm.com>
8  *    Alexander Graf <agraf@suse.de>
9  *    Kevin Wolf <mail@kevin-wolf.de>
10  *
11  * Description: KVM functions specific to running on Book 3S
12  * processors in hypervisor mode (specifically POWER7 and later).
13  *
14  * This file is derived from arch/powerpc/kvm/book3s.c,
15  * by Alexander Graf <agraf@suse.de>.
16  */
17 
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 #include <linux/err.h>
21 #include <linux/slab.h>
22 #include <linux/preempt.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/stat.h>
25 #include <linux/delay.h>
26 #include <linux/export.h>
27 #include <linux/fs.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpu.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 #include <linux/gfp.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/hugetlb.h>
40 #include <linux/kvm_irqfd.h>
41 #include <linux/irqbypass.h>
42 #include <linux/module.h>
43 #include <linux/compiler.h>
44 #include <linux/of.h>
45 
46 #include <asm/ftrace.h>
47 #include <asm/reg.h>
48 #include <asm/ppc-opcode.h>
49 #include <asm/asm-prototypes.h>
50 #include <asm/archrandom.h>
51 #include <asm/debug.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <linux/uaccess.h>
56 #include <asm/interrupt.h>
57 #include <asm/io.h>
58 #include <asm/kvm_ppc.h>
59 #include <asm/kvm_book3s.h>
60 #include <asm/mmu_context.h>
61 #include <asm/lppaca.h>
62 #include <asm/processor.h>
63 #include <asm/cputhreads.h>
64 #include <asm/page.h>
65 #include <asm/hvcall.h>
66 #include <asm/switch_to.h>
67 #include <asm/smp.h>
68 #include <asm/dbell.h>
69 #include <asm/hmi.h>
70 #include <asm/pnv-pci.h>
71 #include <asm/mmu.h>
72 #include <asm/opal.h>
73 #include <asm/xics.h>
74 #include <asm/xive.h>
75 #include <asm/hw_breakpoint.h>
76 #include <asm/kvm_book3s_uvmem.h>
77 #include <asm/ultravisor.h>
78 #include <asm/dtl.h>
79 #include <asm/plpar_wrappers.h>
80 
81 #include "book3s.h"
82 
83 #define CREATE_TRACE_POINTS
84 #include "trace_hv.h"
85 
86 /* #define EXIT_DEBUG */
87 /* #define EXIT_DEBUG_SIMPLE */
88 /* #define EXIT_DEBUG_INT */
89 
90 /* Used to indicate that a guest page fault needs to be handled */
91 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
92 /* Used to indicate that a guest passthrough interrupt needs to be handled */
93 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
94 
95 /* Used as a "null" value for timebase values */
96 #define TB_NIL	(~(u64)0)
97 
98 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
99 
100 static int dynamic_mt_modes = 6;
101 module_param(dynamic_mt_modes, int, 0644);
102 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
103 static int target_smt_mode;
104 module_param(target_smt_mode, int, 0644);
105 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
106 
107 static bool one_vm_per_core;
108 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
109 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
110 
111 #ifdef CONFIG_KVM_XICS
112 static const struct kernel_param_ops module_param_ops = {
113 	.set = param_set_int,
114 	.get = param_get_int,
115 };
116 
117 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
118 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
119 
120 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
121 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
122 #endif
123 
124 /* If set, guests are allowed to create and control nested guests */
125 static bool nested = true;
126 module_param(nested, bool, S_IRUGO | S_IWUSR);
127 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
128 
129 static inline bool nesting_enabled(struct kvm *kvm)
130 {
131 	return kvm->arch.nested_enable && kvm_is_radix(kvm);
132 }
133 
134 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
135 
136 /*
137  * RWMR values for POWER8.  These control the rate at which PURR
138  * and SPURR count and should be set according to the number of
139  * online threads in the vcore being run.
140  */
141 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
142 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
143 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
144 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
145 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
146 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
147 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
148 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
149 
150 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
151 	RWMR_RPA_P8_1THREAD,
152 	RWMR_RPA_P8_1THREAD,
153 	RWMR_RPA_P8_2THREAD,
154 	RWMR_RPA_P8_3THREAD,
155 	RWMR_RPA_P8_4THREAD,
156 	RWMR_RPA_P8_5THREAD,
157 	RWMR_RPA_P8_6THREAD,
158 	RWMR_RPA_P8_7THREAD,
159 	RWMR_RPA_P8_8THREAD,
160 };
161 
162 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
163 		int *ip)
164 {
165 	int i = *ip;
166 	struct kvm_vcpu *vcpu;
167 
168 	while (++i < MAX_SMT_THREADS) {
169 		vcpu = READ_ONCE(vc->runnable_threads[i]);
170 		if (vcpu) {
171 			*ip = i;
172 			return vcpu;
173 		}
174 	}
175 	return NULL;
176 }
177 
178 /* Used to traverse the list of runnable threads for a given vcore */
179 #define for_each_runnable_thread(i, vcpu, vc) \
180 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
181 
182 static bool kvmppc_ipi_thread(int cpu)
183 {
184 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
185 
186 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
187 	if (kvmhv_on_pseries())
188 		return false;
189 
190 	/* On POWER9 we can use msgsnd to IPI any cpu */
191 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
192 		msg |= get_hard_smp_processor_id(cpu);
193 		smp_mb();
194 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
195 		return true;
196 	}
197 
198 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
199 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
200 		preempt_disable();
201 		if (cpu_first_thread_sibling(cpu) ==
202 		    cpu_first_thread_sibling(smp_processor_id())) {
203 			msg |= cpu_thread_in_core(cpu);
204 			smp_mb();
205 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
206 			preempt_enable();
207 			return true;
208 		}
209 		preempt_enable();
210 	}
211 
212 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
213 	if (cpu >= 0 && cpu < nr_cpu_ids) {
214 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
215 			xics_wake_cpu(cpu);
216 			return true;
217 		}
218 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
219 		return true;
220 	}
221 #endif
222 
223 	return false;
224 }
225 
226 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
227 {
228 	int cpu;
229 	struct rcuwait *waitp;
230 
231 	waitp = kvm_arch_vcpu_get_wait(vcpu);
232 	if (rcuwait_wake_up(waitp))
233 		++vcpu->stat.generic.halt_wakeup;
234 
235 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
236 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
237 		return;
238 
239 	/* CPU points to the first thread of the core */
240 	cpu = vcpu->cpu;
241 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
242 		smp_send_reschedule(cpu);
243 }
244 
245 /*
246  * We use the vcpu_load/put functions to measure stolen time.
247  * Stolen time is counted as time when either the vcpu is able to
248  * run as part of a virtual core, but the task running the vcore
249  * is preempted or sleeping, or when the vcpu needs something done
250  * in the kernel by the task running the vcpu, but that task is
251  * preempted or sleeping.  Those two things have to be counted
252  * separately, since one of the vcpu tasks will take on the job
253  * of running the core, and the other vcpu tasks in the vcore will
254  * sleep waiting for it to do that, but that sleep shouldn't count
255  * as stolen time.
256  *
257  * Hence we accumulate stolen time when the vcpu can run as part of
258  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
259  * needs its task to do other things in the kernel (for example,
260  * service a page fault) in busy_stolen.  We don't accumulate
261  * stolen time for a vcore when it is inactive, or for a vcpu
262  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
263  * a misnomer; it means that the vcpu task is not executing in
264  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
265  * the kernel.  We don't have any way of dividing up that time
266  * between time that the vcpu is genuinely stopped, time that
267  * the task is actively working on behalf of the vcpu, and time
268  * that the task is preempted, so we don't count any of it as
269  * stolen.
270  *
271  * Updates to busy_stolen are protected by arch.tbacct_lock;
272  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
273  * lock.  The stolen times are measured in units of timebase ticks.
274  * (Note that the != TB_NIL checks below are purely defensive;
275  * they should never fail.)
276  */
277 
278 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
279 {
280 	unsigned long flags;
281 
282 	spin_lock_irqsave(&vc->stoltb_lock, flags);
283 	vc->preempt_tb = mftb();
284 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
285 }
286 
287 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
288 {
289 	unsigned long flags;
290 
291 	spin_lock_irqsave(&vc->stoltb_lock, flags);
292 	if (vc->preempt_tb != TB_NIL) {
293 		vc->stolen_tb += mftb() - vc->preempt_tb;
294 		vc->preempt_tb = TB_NIL;
295 	}
296 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
297 }
298 
299 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
300 {
301 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
302 	unsigned long flags;
303 
304 	/*
305 	 * We can test vc->runner without taking the vcore lock,
306 	 * because only this task ever sets vc->runner to this
307 	 * vcpu, and once it is set to this vcpu, only this task
308 	 * ever sets it to NULL.
309 	 */
310 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
311 		kvmppc_core_end_stolen(vc);
312 
313 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
314 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
315 	    vcpu->arch.busy_preempt != TB_NIL) {
316 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
317 		vcpu->arch.busy_preempt = TB_NIL;
318 	}
319 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
320 }
321 
322 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
323 {
324 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
325 	unsigned long flags;
326 
327 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
328 		kvmppc_core_start_stolen(vc);
329 
330 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
331 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
332 		vcpu->arch.busy_preempt = mftb();
333 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
334 }
335 
336 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
337 {
338 	vcpu->arch.pvr = pvr;
339 }
340 
341 /* Dummy value used in computing PCR value below */
342 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
343 
344 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
345 {
346 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
347 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
348 
349 	/* We can (emulate) our own architecture version and anything older */
350 	if (cpu_has_feature(CPU_FTR_ARCH_31))
351 		host_pcr_bit = PCR_ARCH_31;
352 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
353 		host_pcr_bit = PCR_ARCH_300;
354 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
355 		host_pcr_bit = PCR_ARCH_207;
356 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
357 		host_pcr_bit = PCR_ARCH_206;
358 	else
359 		host_pcr_bit = PCR_ARCH_205;
360 
361 	/* Determine lowest PCR bit needed to run guest in given PVR level */
362 	guest_pcr_bit = host_pcr_bit;
363 	if (arch_compat) {
364 		switch (arch_compat) {
365 		case PVR_ARCH_205:
366 			guest_pcr_bit = PCR_ARCH_205;
367 			break;
368 		case PVR_ARCH_206:
369 		case PVR_ARCH_206p:
370 			guest_pcr_bit = PCR_ARCH_206;
371 			break;
372 		case PVR_ARCH_207:
373 			guest_pcr_bit = PCR_ARCH_207;
374 			break;
375 		case PVR_ARCH_300:
376 			guest_pcr_bit = PCR_ARCH_300;
377 			break;
378 		case PVR_ARCH_31:
379 			guest_pcr_bit = PCR_ARCH_31;
380 			break;
381 		default:
382 			return -EINVAL;
383 		}
384 	}
385 
386 	/* Check requested PCR bits don't exceed our capabilities */
387 	if (guest_pcr_bit > host_pcr_bit)
388 		return -EINVAL;
389 
390 	spin_lock(&vc->lock);
391 	vc->arch_compat = arch_compat;
392 	/*
393 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
394 	 * Also set all reserved PCR bits
395 	 */
396 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
397 	spin_unlock(&vc->lock);
398 
399 	return 0;
400 }
401 
402 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
403 {
404 	int r;
405 
406 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
407 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
408 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
409 	for (r = 0; r < 16; ++r)
410 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
411 		       r, kvmppc_get_gpr(vcpu, r),
412 		       r+16, kvmppc_get_gpr(vcpu, r+16));
413 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
414 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
415 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
416 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
417 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
418 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
419 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
420 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
421 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
422 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
423 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
424 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
425 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
426 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
427 	for (r = 0; r < vcpu->arch.slb_max; ++r)
428 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
429 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
430 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
431 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
432 	       vcpu->arch.last_inst);
433 }
434 
435 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
436 {
437 	return kvm_get_vcpu_by_id(kvm, id);
438 }
439 
440 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
441 {
442 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
443 	vpa->yield_count = cpu_to_be32(1);
444 }
445 
446 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
447 		   unsigned long addr, unsigned long len)
448 {
449 	/* check address is cacheline aligned */
450 	if (addr & (L1_CACHE_BYTES - 1))
451 		return -EINVAL;
452 	spin_lock(&vcpu->arch.vpa_update_lock);
453 	if (v->next_gpa != addr || v->len != len) {
454 		v->next_gpa = addr;
455 		v->len = addr ? len : 0;
456 		v->update_pending = 1;
457 	}
458 	spin_unlock(&vcpu->arch.vpa_update_lock);
459 	return 0;
460 }
461 
462 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
463 struct reg_vpa {
464 	u32 dummy;
465 	union {
466 		__be16 hword;
467 		__be32 word;
468 	} length;
469 };
470 
471 static int vpa_is_registered(struct kvmppc_vpa *vpap)
472 {
473 	if (vpap->update_pending)
474 		return vpap->next_gpa != 0;
475 	return vpap->pinned_addr != NULL;
476 }
477 
478 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
479 				       unsigned long flags,
480 				       unsigned long vcpuid, unsigned long vpa)
481 {
482 	struct kvm *kvm = vcpu->kvm;
483 	unsigned long len, nb;
484 	void *va;
485 	struct kvm_vcpu *tvcpu;
486 	int err;
487 	int subfunc;
488 	struct kvmppc_vpa *vpap;
489 
490 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
491 	if (!tvcpu)
492 		return H_PARAMETER;
493 
494 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
495 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
496 	    subfunc == H_VPA_REG_SLB) {
497 		/* Registering new area - address must be cache-line aligned */
498 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
499 			return H_PARAMETER;
500 
501 		/* convert logical addr to kernel addr and read length */
502 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
503 		if (va == NULL)
504 			return H_PARAMETER;
505 		if (subfunc == H_VPA_REG_VPA)
506 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
507 		else
508 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
509 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
510 
511 		/* Check length */
512 		if (len > nb || len < sizeof(struct reg_vpa))
513 			return H_PARAMETER;
514 	} else {
515 		vpa = 0;
516 		len = 0;
517 	}
518 
519 	err = H_PARAMETER;
520 	vpap = NULL;
521 	spin_lock(&tvcpu->arch.vpa_update_lock);
522 
523 	switch (subfunc) {
524 	case H_VPA_REG_VPA:		/* register VPA */
525 		/*
526 		 * The size of our lppaca is 1kB because of the way we align
527 		 * it for the guest to avoid crossing a 4kB boundary. We only
528 		 * use 640 bytes of the structure though, so we should accept
529 		 * clients that set a size of 640.
530 		 */
531 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
532 		if (len < sizeof(struct lppaca))
533 			break;
534 		vpap = &tvcpu->arch.vpa;
535 		err = 0;
536 		break;
537 
538 	case H_VPA_REG_DTL:		/* register DTL */
539 		if (len < sizeof(struct dtl_entry))
540 			break;
541 		len -= len % sizeof(struct dtl_entry);
542 
543 		/* Check that they have previously registered a VPA */
544 		err = H_RESOURCE;
545 		if (!vpa_is_registered(&tvcpu->arch.vpa))
546 			break;
547 
548 		vpap = &tvcpu->arch.dtl;
549 		err = 0;
550 		break;
551 
552 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
553 		/* Check that they have previously registered a VPA */
554 		err = H_RESOURCE;
555 		if (!vpa_is_registered(&tvcpu->arch.vpa))
556 			break;
557 
558 		vpap = &tvcpu->arch.slb_shadow;
559 		err = 0;
560 		break;
561 
562 	case H_VPA_DEREG_VPA:		/* deregister VPA */
563 		/* Check they don't still have a DTL or SLB buf registered */
564 		err = H_RESOURCE;
565 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
566 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
567 			break;
568 
569 		vpap = &tvcpu->arch.vpa;
570 		err = 0;
571 		break;
572 
573 	case H_VPA_DEREG_DTL:		/* deregister DTL */
574 		vpap = &tvcpu->arch.dtl;
575 		err = 0;
576 		break;
577 
578 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
579 		vpap = &tvcpu->arch.slb_shadow;
580 		err = 0;
581 		break;
582 	}
583 
584 	if (vpap) {
585 		vpap->next_gpa = vpa;
586 		vpap->len = len;
587 		vpap->update_pending = 1;
588 	}
589 
590 	spin_unlock(&tvcpu->arch.vpa_update_lock);
591 
592 	return err;
593 }
594 
595 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
596 {
597 	struct kvm *kvm = vcpu->kvm;
598 	void *va;
599 	unsigned long nb;
600 	unsigned long gpa;
601 
602 	/*
603 	 * We need to pin the page pointed to by vpap->next_gpa,
604 	 * but we can't call kvmppc_pin_guest_page under the lock
605 	 * as it does get_user_pages() and down_read().  So we
606 	 * have to drop the lock, pin the page, then get the lock
607 	 * again and check that a new area didn't get registered
608 	 * in the meantime.
609 	 */
610 	for (;;) {
611 		gpa = vpap->next_gpa;
612 		spin_unlock(&vcpu->arch.vpa_update_lock);
613 		va = NULL;
614 		nb = 0;
615 		if (gpa)
616 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
617 		spin_lock(&vcpu->arch.vpa_update_lock);
618 		if (gpa == vpap->next_gpa)
619 			break;
620 		/* sigh... unpin that one and try again */
621 		if (va)
622 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
623 	}
624 
625 	vpap->update_pending = 0;
626 	if (va && nb < vpap->len) {
627 		/*
628 		 * If it's now too short, it must be that userspace
629 		 * has changed the mappings underlying guest memory,
630 		 * so unregister the region.
631 		 */
632 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
633 		va = NULL;
634 	}
635 	if (vpap->pinned_addr)
636 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
637 					vpap->dirty);
638 	vpap->gpa = gpa;
639 	vpap->pinned_addr = va;
640 	vpap->dirty = false;
641 	if (va)
642 		vpap->pinned_end = va + vpap->len;
643 }
644 
645 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
646 {
647 	if (!(vcpu->arch.vpa.update_pending ||
648 	      vcpu->arch.slb_shadow.update_pending ||
649 	      vcpu->arch.dtl.update_pending))
650 		return;
651 
652 	spin_lock(&vcpu->arch.vpa_update_lock);
653 	if (vcpu->arch.vpa.update_pending) {
654 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
655 		if (vcpu->arch.vpa.pinned_addr)
656 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
657 	}
658 	if (vcpu->arch.dtl.update_pending) {
659 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
660 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
661 		vcpu->arch.dtl_index = 0;
662 	}
663 	if (vcpu->arch.slb_shadow.update_pending)
664 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
665 	spin_unlock(&vcpu->arch.vpa_update_lock);
666 }
667 
668 /*
669  * Return the accumulated stolen time for the vcore up until `now'.
670  * The caller should hold the vcore lock.
671  */
672 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
673 {
674 	u64 p;
675 	unsigned long flags;
676 
677 	spin_lock_irqsave(&vc->stoltb_lock, flags);
678 	p = vc->stolen_tb;
679 	if (vc->vcore_state != VCORE_INACTIVE &&
680 	    vc->preempt_tb != TB_NIL)
681 		p += now - vc->preempt_tb;
682 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
683 	return p;
684 }
685 
686 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
687 				    struct kvmppc_vcore *vc)
688 {
689 	struct dtl_entry *dt;
690 	struct lppaca *vpa;
691 	unsigned long stolen;
692 	unsigned long core_stolen;
693 	u64 now;
694 	unsigned long flags;
695 
696 	dt = vcpu->arch.dtl_ptr;
697 	vpa = vcpu->arch.vpa.pinned_addr;
698 	now = mftb();
699 	core_stolen = vcore_stolen_time(vc, now);
700 	stolen = core_stolen - vcpu->arch.stolen_logged;
701 	vcpu->arch.stolen_logged = core_stolen;
702 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
703 	stolen += vcpu->arch.busy_stolen;
704 	vcpu->arch.busy_stolen = 0;
705 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
706 	if (!dt || !vpa)
707 		return;
708 	memset(dt, 0, sizeof(struct dtl_entry));
709 	dt->dispatch_reason = 7;
710 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
711 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
712 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
713 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
714 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
715 	++dt;
716 	if (dt == vcpu->arch.dtl.pinned_end)
717 		dt = vcpu->arch.dtl.pinned_addr;
718 	vcpu->arch.dtl_ptr = dt;
719 	/* order writing *dt vs. writing vpa->dtl_idx */
720 	smp_wmb();
721 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
722 	vcpu->arch.dtl.dirty = true;
723 }
724 
725 /* See if there is a doorbell interrupt pending for a vcpu */
726 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
727 {
728 	int thr;
729 	struct kvmppc_vcore *vc;
730 
731 	if (vcpu->arch.doorbell_request)
732 		return true;
733 	/*
734 	 * Ensure that the read of vcore->dpdes comes after the read
735 	 * of vcpu->doorbell_request.  This barrier matches the
736 	 * smp_wmb() in kvmppc_guest_entry_inject().
737 	 */
738 	smp_rmb();
739 	vc = vcpu->arch.vcore;
740 	thr = vcpu->vcpu_id - vc->first_vcpuid;
741 	return !!(vc->dpdes & (1 << thr));
742 }
743 
744 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
745 {
746 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
747 		return true;
748 	if ((!vcpu->arch.vcore->arch_compat) &&
749 	    cpu_has_feature(CPU_FTR_ARCH_207S))
750 		return true;
751 	return false;
752 }
753 
754 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
755 			     unsigned long resource, unsigned long value1,
756 			     unsigned long value2)
757 {
758 	switch (resource) {
759 	case H_SET_MODE_RESOURCE_SET_CIABR:
760 		if (!kvmppc_power8_compatible(vcpu))
761 			return H_P2;
762 		if (value2)
763 			return H_P4;
764 		if (mflags)
765 			return H_UNSUPPORTED_FLAG_START;
766 		/* Guests can't breakpoint the hypervisor */
767 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
768 			return H_P3;
769 		vcpu->arch.ciabr  = value1;
770 		return H_SUCCESS;
771 	case H_SET_MODE_RESOURCE_SET_DAWR0:
772 		if (!kvmppc_power8_compatible(vcpu))
773 			return H_P2;
774 		if (!ppc_breakpoint_available())
775 			return H_P2;
776 		if (mflags)
777 			return H_UNSUPPORTED_FLAG_START;
778 		if (value2 & DABRX_HYP)
779 			return H_P4;
780 		vcpu->arch.dawr0  = value1;
781 		vcpu->arch.dawrx0 = value2;
782 		return H_SUCCESS;
783 	case H_SET_MODE_RESOURCE_SET_DAWR1:
784 		if (!kvmppc_power8_compatible(vcpu))
785 			return H_P2;
786 		if (!ppc_breakpoint_available())
787 			return H_P2;
788 		if (!cpu_has_feature(CPU_FTR_DAWR1))
789 			return H_P2;
790 		if (!vcpu->kvm->arch.dawr1_enabled)
791 			return H_FUNCTION;
792 		if (mflags)
793 			return H_UNSUPPORTED_FLAG_START;
794 		if (value2 & DABRX_HYP)
795 			return H_P4;
796 		vcpu->arch.dawr1  = value1;
797 		vcpu->arch.dawrx1 = value2;
798 		return H_SUCCESS;
799 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
800 		/*
801 		 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
802 		 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
803 		 */
804 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
805 				kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
806 			return H_UNSUPPORTED_FLAG_START;
807 		return H_TOO_HARD;
808 	default:
809 		return H_TOO_HARD;
810 	}
811 }
812 
813 /* Copy guest memory in place - must reside within a single memslot */
814 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
815 				  unsigned long len)
816 {
817 	struct kvm_memory_slot *to_memslot = NULL;
818 	struct kvm_memory_slot *from_memslot = NULL;
819 	unsigned long to_addr, from_addr;
820 	int r;
821 
822 	/* Get HPA for from address */
823 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
824 	if (!from_memslot)
825 		return -EFAULT;
826 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
827 			     << PAGE_SHIFT))
828 		return -EINVAL;
829 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
830 	if (kvm_is_error_hva(from_addr))
831 		return -EFAULT;
832 	from_addr |= (from & (PAGE_SIZE - 1));
833 
834 	/* Get HPA for to address */
835 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
836 	if (!to_memslot)
837 		return -EFAULT;
838 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
839 			   << PAGE_SHIFT))
840 		return -EINVAL;
841 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
842 	if (kvm_is_error_hva(to_addr))
843 		return -EFAULT;
844 	to_addr |= (to & (PAGE_SIZE - 1));
845 
846 	/* Perform copy */
847 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
848 			     len);
849 	if (r)
850 		return -EFAULT;
851 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
852 	return 0;
853 }
854 
855 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
856 			       unsigned long dest, unsigned long src)
857 {
858 	u64 pg_sz = SZ_4K;		/* 4K page size */
859 	u64 pg_mask = SZ_4K - 1;
860 	int ret;
861 
862 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
863 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
864 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
865 		return H_PARAMETER;
866 
867 	/* dest (and src if copy_page flag set) must be page aligned */
868 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
869 		return H_PARAMETER;
870 
871 	/* zero and/or copy the page as determined by the flags */
872 	if (flags & H_COPY_PAGE) {
873 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
874 		if (ret < 0)
875 			return H_PARAMETER;
876 	} else if (flags & H_ZERO_PAGE) {
877 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
878 		if (ret < 0)
879 			return H_PARAMETER;
880 	}
881 
882 	/* We can ignore the remaining flags */
883 
884 	return H_SUCCESS;
885 }
886 
887 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
888 {
889 	struct kvmppc_vcore *vcore = target->arch.vcore;
890 
891 	/*
892 	 * We expect to have been called by the real mode handler
893 	 * (kvmppc_rm_h_confer()) which would have directly returned
894 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
895 	 * have useful work to do and should not confer) so we don't
896 	 * recheck that here.
897 	 *
898 	 * In the case of the P9 single vcpu per vcore case, the real
899 	 * mode handler is not called but no other threads are in the
900 	 * source vcore.
901 	 */
902 
903 	spin_lock(&vcore->lock);
904 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
905 	    vcore->vcore_state != VCORE_INACTIVE &&
906 	    vcore->runner)
907 		target = vcore->runner;
908 	spin_unlock(&vcore->lock);
909 
910 	return kvm_vcpu_yield_to(target);
911 }
912 
913 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
914 {
915 	int yield_count = 0;
916 	struct lppaca *lppaca;
917 
918 	spin_lock(&vcpu->arch.vpa_update_lock);
919 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
920 	if (lppaca)
921 		yield_count = be32_to_cpu(lppaca->yield_count);
922 	spin_unlock(&vcpu->arch.vpa_update_lock);
923 	return yield_count;
924 }
925 
926 /*
927  * H_RPT_INVALIDATE hcall handler for nested guests.
928  *
929  * Handles only nested process-scoped invalidation requests in L0.
930  */
931 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
932 {
933 	unsigned long type = kvmppc_get_gpr(vcpu, 6);
934 	unsigned long pid, pg_sizes, start, end;
935 
936 	/*
937 	 * The partition-scoped invalidations aren't handled here in L0.
938 	 */
939 	if (type & H_RPTI_TYPE_NESTED)
940 		return RESUME_HOST;
941 
942 	pid = kvmppc_get_gpr(vcpu, 4);
943 	pg_sizes = kvmppc_get_gpr(vcpu, 7);
944 	start = kvmppc_get_gpr(vcpu, 8);
945 	end = kvmppc_get_gpr(vcpu, 9);
946 
947 	do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
948 				type, pg_sizes, start, end);
949 
950 	kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
951 	return RESUME_GUEST;
952 }
953 
954 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
955 				    unsigned long id, unsigned long target,
956 				    unsigned long type, unsigned long pg_sizes,
957 				    unsigned long start, unsigned long end)
958 {
959 	if (!kvm_is_radix(vcpu->kvm))
960 		return H_UNSUPPORTED;
961 
962 	if (end < start)
963 		return H_P5;
964 
965 	/*
966 	 * Partition-scoped invalidation for nested guests.
967 	 */
968 	if (type & H_RPTI_TYPE_NESTED) {
969 		if (!nesting_enabled(vcpu->kvm))
970 			return H_FUNCTION;
971 
972 		/* Support only cores as target */
973 		if (target != H_RPTI_TARGET_CMMU)
974 			return H_P2;
975 
976 		return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
977 					       start, end);
978 	}
979 
980 	/*
981 	 * Process-scoped invalidation for L1 guests.
982 	 */
983 	do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
984 				type, pg_sizes, start, end);
985 	return H_SUCCESS;
986 }
987 
988 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
989 {
990 	struct kvm *kvm = vcpu->kvm;
991 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
992 	unsigned long target, ret = H_SUCCESS;
993 	int yield_count;
994 	struct kvm_vcpu *tvcpu;
995 	int idx, rc;
996 
997 	if (req <= MAX_HCALL_OPCODE &&
998 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
999 		return RESUME_HOST;
1000 
1001 	switch (req) {
1002 	case H_REMOVE:
1003 		ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1004 					kvmppc_get_gpr(vcpu, 5),
1005 					kvmppc_get_gpr(vcpu, 6));
1006 		if (ret == H_TOO_HARD)
1007 			return RESUME_HOST;
1008 		break;
1009 	case H_ENTER:
1010 		ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1011 					kvmppc_get_gpr(vcpu, 5),
1012 					kvmppc_get_gpr(vcpu, 6),
1013 					kvmppc_get_gpr(vcpu, 7));
1014 		if (ret == H_TOO_HARD)
1015 			return RESUME_HOST;
1016 		break;
1017 	case H_READ:
1018 		ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1019 					kvmppc_get_gpr(vcpu, 5));
1020 		if (ret == H_TOO_HARD)
1021 			return RESUME_HOST;
1022 		break;
1023 	case H_CLEAR_MOD:
1024 		ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1025 					kvmppc_get_gpr(vcpu, 5));
1026 		if (ret == H_TOO_HARD)
1027 			return RESUME_HOST;
1028 		break;
1029 	case H_CLEAR_REF:
1030 		ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1031 					kvmppc_get_gpr(vcpu, 5));
1032 		if (ret == H_TOO_HARD)
1033 			return RESUME_HOST;
1034 		break;
1035 	case H_PROTECT:
1036 		ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1037 					kvmppc_get_gpr(vcpu, 5),
1038 					kvmppc_get_gpr(vcpu, 6));
1039 		if (ret == H_TOO_HARD)
1040 			return RESUME_HOST;
1041 		break;
1042 	case H_BULK_REMOVE:
1043 		ret = kvmppc_h_bulk_remove(vcpu);
1044 		if (ret == H_TOO_HARD)
1045 			return RESUME_HOST;
1046 		break;
1047 
1048 	case H_CEDE:
1049 		break;
1050 	case H_PROD:
1051 		target = kvmppc_get_gpr(vcpu, 4);
1052 		tvcpu = kvmppc_find_vcpu(kvm, target);
1053 		if (!tvcpu) {
1054 			ret = H_PARAMETER;
1055 			break;
1056 		}
1057 		tvcpu->arch.prodded = 1;
1058 		smp_mb();
1059 		if (tvcpu->arch.ceded)
1060 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1061 		break;
1062 	case H_CONFER:
1063 		target = kvmppc_get_gpr(vcpu, 4);
1064 		if (target == -1)
1065 			break;
1066 		tvcpu = kvmppc_find_vcpu(kvm, target);
1067 		if (!tvcpu) {
1068 			ret = H_PARAMETER;
1069 			break;
1070 		}
1071 		yield_count = kvmppc_get_gpr(vcpu, 5);
1072 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
1073 			break;
1074 		kvm_arch_vcpu_yield_to(tvcpu);
1075 		break;
1076 	case H_REGISTER_VPA:
1077 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1078 					kvmppc_get_gpr(vcpu, 5),
1079 					kvmppc_get_gpr(vcpu, 6));
1080 		break;
1081 	case H_RTAS:
1082 		if (list_empty(&kvm->arch.rtas_tokens))
1083 			return RESUME_HOST;
1084 
1085 		idx = srcu_read_lock(&kvm->srcu);
1086 		rc = kvmppc_rtas_hcall(vcpu);
1087 		srcu_read_unlock(&kvm->srcu, idx);
1088 
1089 		if (rc == -ENOENT)
1090 			return RESUME_HOST;
1091 		else if (rc == 0)
1092 			break;
1093 
1094 		/* Send the error out to userspace via KVM_RUN */
1095 		return rc;
1096 	case H_LOGICAL_CI_LOAD:
1097 		ret = kvmppc_h_logical_ci_load(vcpu);
1098 		if (ret == H_TOO_HARD)
1099 			return RESUME_HOST;
1100 		break;
1101 	case H_LOGICAL_CI_STORE:
1102 		ret = kvmppc_h_logical_ci_store(vcpu);
1103 		if (ret == H_TOO_HARD)
1104 			return RESUME_HOST;
1105 		break;
1106 	case H_SET_MODE:
1107 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1108 					kvmppc_get_gpr(vcpu, 5),
1109 					kvmppc_get_gpr(vcpu, 6),
1110 					kvmppc_get_gpr(vcpu, 7));
1111 		if (ret == H_TOO_HARD)
1112 			return RESUME_HOST;
1113 		break;
1114 	case H_XIRR:
1115 	case H_CPPR:
1116 	case H_EOI:
1117 	case H_IPI:
1118 	case H_IPOLL:
1119 	case H_XIRR_X:
1120 		if (kvmppc_xics_enabled(vcpu)) {
1121 			if (xics_on_xive()) {
1122 				ret = H_NOT_AVAILABLE;
1123 				return RESUME_GUEST;
1124 			}
1125 			ret = kvmppc_xics_hcall(vcpu, req);
1126 			break;
1127 		}
1128 		return RESUME_HOST;
1129 	case H_SET_DABR:
1130 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1131 		break;
1132 	case H_SET_XDABR:
1133 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1134 						kvmppc_get_gpr(vcpu, 5));
1135 		break;
1136 #ifdef CONFIG_SPAPR_TCE_IOMMU
1137 	case H_GET_TCE:
1138 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1139 						kvmppc_get_gpr(vcpu, 5));
1140 		if (ret == H_TOO_HARD)
1141 			return RESUME_HOST;
1142 		break;
1143 	case H_PUT_TCE:
1144 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1145 						kvmppc_get_gpr(vcpu, 5),
1146 						kvmppc_get_gpr(vcpu, 6));
1147 		if (ret == H_TOO_HARD)
1148 			return RESUME_HOST;
1149 		break;
1150 	case H_PUT_TCE_INDIRECT:
1151 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1152 						kvmppc_get_gpr(vcpu, 5),
1153 						kvmppc_get_gpr(vcpu, 6),
1154 						kvmppc_get_gpr(vcpu, 7));
1155 		if (ret == H_TOO_HARD)
1156 			return RESUME_HOST;
1157 		break;
1158 	case H_STUFF_TCE:
1159 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1160 						kvmppc_get_gpr(vcpu, 5),
1161 						kvmppc_get_gpr(vcpu, 6),
1162 						kvmppc_get_gpr(vcpu, 7));
1163 		if (ret == H_TOO_HARD)
1164 			return RESUME_HOST;
1165 		break;
1166 #endif
1167 	case H_RANDOM:
1168 		if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1169 			ret = H_HARDWARE;
1170 		break;
1171 	case H_RPT_INVALIDATE:
1172 		ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1173 					      kvmppc_get_gpr(vcpu, 5),
1174 					      kvmppc_get_gpr(vcpu, 6),
1175 					      kvmppc_get_gpr(vcpu, 7),
1176 					      kvmppc_get_gpr(vcpu, 8),
1177 					      kvmppc_get_gpr(vcpu, 9));
1178 		break;
1179 
1180 	case H_SET_PARTITION_TABLE:
1181 		ret = H_FUNCTION;
1182 		if (nesting_enabled(kvm))
1183 			ret = kvmhv_set_partition_table(vcpu);
1184 		break;
1185 	case H_ENTER_NESTED:
1186 		ret = H_FUNCTION;
1187 		if (!nesting_enabled(kvm))
1188 			break;
1189 		ret = kvmhv_enter_nested_guest(vcpu);
1190 		if (ret == H_INTERRUPT) {
1191 			kvmppc_set_gpr(vcpu, 3, 0);
1192 			vcpu->arch.hcall_needed = 0;
1193 			return -EINTR;
1194 		} else if (ret == H_TOO_HARD) {
1195 			kvmppc_set_gpr(vcpu, 3, 0);
1196 			vcpu->arch.hcall_needed = 0;
1197 			return RESUME_HOST;
1198 		}
1199 		break;
1200 	case H_TLB_INVALIDATE:
1201 		ret = H_FUNCTION;
1202 		if (nesting_enabled(kvm))
1203 			ret = kvmhv_do_nested_tlbie(vcpu);
1204 		break;
1205 	case H_COPY_TOFROM_GUEST:
1206 		ret = H_FUNCTION;
1207 		if (nesting_enabled(kvm))
1208 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1209 		break;
1210 	case H_PAGE_INIT:
1211 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1212 					 kvmppc_get_gpr(vcpu, 5),
1213 					 kvmppc_get_gpr(vcpu, 6));
1214 		break;
1215 	case H_SVM_PAGE_IN:
1216 		ret = H_UNSUPPORTED;
1217 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1218 			ret = kvmppc_h_svm_page_in(kvm,
1219 						   kvmppc_get_gpr(vcpu, 4),
1220 						   kvmppc_get_gpr(vcpu, 5),
1221 						   kvmppc_get_gpr(vcpu, 6));
1222 		break;
1223 	case H_SVM_PAGE_OUT:
1224 		ret = H_UNSUPPORTED;
1225 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1226 			ret = kvmppc_h_svm_page_out(kvm,
1227 						    kvmppc_get_gpr(vcpu, 4),
1228 						    kvmppc_get_gpr(vcpu, 5),
1229 						    kvmppc_get_gpr(vcpu, 6));
1230 		break;
1231 	case H_SVM_INIT_START:
1232 		ret = H_UNSUPPORTED;
1233 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1234 			ret = kvmppc_h_svm_init_start(kvm);
1235 		break;
1236 	case H_SVM_INIT_DONE:
1237 		ret = H_UNSUPPORTED;
1238 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1239 			ret = kvmppc_h_svm_init_done(kvm);
1240 		break;
1241 	case H_SVM_INIT_ABORT:
1242 		/*
1243 		 * Even if that call is made by the Ultravisor, the SSR1 value
1244 		 * is the guest context one, with the secure bit clear as it has
1245 		 * not yet been secured. So we can't check it here.
1246 		 * Instead the kvm->arch.secure_guest flag is checked inside
1247 		 * kvmppc_h_svm_init_abort().
1248 		 */
1249 		ret = kvmppc_h_svm_init_abort(kvm);
1250 		break;
1251 
1252 	default:
1253 		return RESUME_HOST;
1254 	}
1255 	WARN_ON_ONCE(ret == H_TOO_HARD);
1256 	kvmppc_set_gpr(vcpu, 3, ret);
1257 	vcpu->arch.hcall_needed = 0;
1258 	return RESUME_GUEST;
1259 }
1260 
1261 /*
1262  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1263  * handlers in book3s_hv_rmhandlers.S.
1264  *
1265  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1266  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1267  */
1268 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1269 {
1270 	vcpu->arch.shregs.msr |= MSR_EE;
1271 	vcpu->arch.ceded = 1;
1272 	smp_mb();
1273 	if (vcpu->arch.prodded) {
1274 		vcpu->arch.prodded = 0;
1275 		smp_mb();
1276 		vcpu->arch.ceded = 0;
1277 	}
1278 }
1279 
1280 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1281 {
1282 	switch (cmd) {
1283 	case H_CEDE:
1284 	case H_PROD:
1285 	case H_CONFER:
1286 	case H_REGISTER_VPA:
1287 	case H_SET_MODE:
1288 	case H_LOGICAL_CI_LOAD:
1289 	case H_LOGICAL_CI_STORE:
1290 #ifdef CONFIG_KVM_XICS
1291 	case H_XIRR:
1292 	case H_CPPR:
1293 	case H_EOI:
1294 	case H_IPI:
1295 	case H_IPOLL:
1296 	case H_XIRR_X:
1297 #endif
1298 	case H_PAGE_INIT:
1299 	case H_RPT_INVALIDATE:
1300 		return 1;
1301 	}
1302 
1303 	/* See if it's in the real-mode table */
1304 	return kvmppc_hcall_impl_hv_realmode(cmd);
1305 }
1306 
1307 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1308 {
1309 	u32 last_inst;
1310 
1311 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1312 					EMULATE_DONE) {
1313 		/*
1314 		 * Fetch failed, so return to guest and
1315 		 * try executing it again.
1316 		 */
1317 		return RESUME_GUEST;
1318 	}
1319 
1320 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1321 		vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1322 		vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1323 		return RESUME_HOST;
1324 	} else {
1325 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1326 		return RESUME_GUEST;
1327 	}
1328 }
1329 
1330 static void do_nothing(void *x)
1331 {
1332 }
1333 
1334 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1335 {
1336 	int thr, cpu, pcpu, nthreads;
1337 	struct kvm_vcpu *v;
1338 	unsigned long dpdes;
1339 
1340 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1341 	dpdes = 0;
1342 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1343 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1344 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1345 		if (!v)
1346 			continue;
1347 		/*
1348 		 * If the vcpu is currently running on a physical cpu thread,
1349 		 * interrupt it in order to pull it out of the guest briefly,
1350 		 * which will update its vcore->dpdes value.
1351 		 */
1352 		pcpu = READ_ONCE(v->cpu);
1353 		if (pcpu >= 0)
1354 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1355 		if (kvmppc_doorbell_pending(v))
1356 			dpdes |= 1 << thr;
1357 	}
1358 	return dpdes;
1359 }
1360 
1361 /*
1362  * On POWER9, emulate doorbell-related instructions in order to
1363  * give the guest the illusion of running on a multi-threaded core.
1364  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1365  * and mfspr DPDES.
1366  */
1367 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1368 {
1369 	u32 inst, rb, thr;
1370 	unsigned long arg;
1371 	struct kvm *kvm = vcpu->kvm;
1372 	struct kvm_vcpu *tvcpu;
1373 
1374 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1375 		return RESUME_GUEST;
1376 	if (get_op(inst) != 31)
1377 		return EMULATE_FAIL;
1378 	rb = get_rb(inst);
1379 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1380 	switch (get_xop(inst)) {
1381 	case OP_31_XOP_MSGSNDP:
1382 		arg = kvmppc_get_gpr(vcpu, rb);
1383 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1384 			break;
1385 		arg &= 0x7f;
1386 		if (arg >= kvm->arch.emul_smt_mode)
1387 			break;
1388 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1389 		if (!tvcpu)
1390 			break;
1391 		if (!tvcpu->arch.doorbell_request) {
1392 			tvcpu->arch.doorbell_request = 1;
1393 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1394 		}
1395 		break;
1396 	case OP_31_XOP_MSGCLRP:
1397 		arg = kvmppc_get_gpr(vcpu, rb);
1398 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1399 			break;
1400 		vcpu->arch.vcore->dpdes = 0;
1401 		vcpu->arch.doorbell_request = 0;
1402 		break;
1403 	case OP_31_XOP_MFSPR:
1404 		switch (get_sprn(inst)) {
1405 		case SPRN_TIR:
1406 			arg = thr;
1407 			break;
1408 		case SPRN_DPDES:
1409 			arg = kvmppc_read_dpdes(vcpu);
1410 			break;
1411 		default:
1412 			return EMULATE_FAIL;
1413 		}
1414 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1415 		break;
1416 	default:
1417 		return EMULATE_FAIL;
1418 	}
1419 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1420 	return RESUME_GUEST;
1421 }
1422 
1423 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1424 				 struct task_struct *tsk)
1425 {
1426 	struct kvm_run *run = vcpu->run;
1427 	int r = RESUME_HOST;
1428 
1429 	vcpu->stat.sum_exits++;
1430 
1431 	/*
1432 	 * This can happen if an interrupt occurs in the last stages
1433 	 * of guest entry or the first stages of guest exit (i.e. after
1434 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1435 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1436 	 * That can happen due to a bug, or due to a machine check
1437 	 * occurring at just the wrong time.
1438 	 */
1439 	if (vcpu->arch.shregs.msr & MSR_HV) {
1440 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1441 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1442 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1443 			vcpu->arch.shregs.msr);
1444 		kvmppc_dump_regs(vcpu);
1445 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1446 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1447 		return RESUME_HOST;
1448 	}
1449 	run->exit_reason = KVM_EXIT_UNKNOWN;
1450 	run->ready_for_interrupt_injection = 1;
1451 	switch (vcpu->arch.trap) {
1452 	/* We're good on these - the host merely wanted to get our attention */
1453 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1454 		vcpu->stat.dec_exits++;
1455 		r = RESUME_GUEST;
1456 		break;
1457 	case BOOK3S_INTERRUPT_EXTERNAL:
1458 	case BOOK3S_INTERRUPT_H_DOORBELL:
1459 	case BOOK3S_INTERRUPT_H_VIRT:
1460 		vcpu->stat.ext_intr_exits++;
1461 		r = RESUME_GUEST;
1462 		break;
1463 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1464 	case BOOK3S_INTERRUPT_HMI:
1465 	case BOOK3S_INTERRUPT_PERFMON:
1466 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1467 		r = RESUME_GUEST;
1468 		break;
1469 	case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1470 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1471 					      DEFAULT_RATELIMIT_BURST);
1472 		/*
1473 		 * Print the MCE event to host console. Ratelimit so the guest
1474 		 * can't flood the host log.
1475 		 */
1476 		if (__ratelimit(&rs))
1477 			machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1478 
1479 		/*
1480 		 * If the guest can do FWNMI, exit to userspace so it can
1481 		 * deliver a FWNMI to the guest.
1482 		 * Otherwise we synthesize a machine check for the guest
1483 		 * so that it knows that the machine check occurred.
1484 		 */
1485 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1486 			ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1487 			kvmppc_core_queue_machine_check(vcpu, flags);
1488 			r = RESUME_GUEST;
1489 			break;
1490 		}
1491 
1492 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1493 		run->exit_reason = KVM_EXIT_NMI;
1494 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1495 		/* Clear out the old NMI status from run->flags */
1496 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1497 		/* Now set the NMI status */
1498 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1499 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1500 		else
1501 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1502 
1503 		r = RESUME_HOST;
1504 		break;
1505 	}
1506 	case BOOK3S_INTERRUPT_PROGRAM:
1507 	{
1508 		ulong flags;
1509 		/*
1510 		 * Normally program interrupts are delivered directly
1511 		 * to the guest by the hardware, but we can get here
1512 		 * as a result of a hypervisor emulation interrupt
1513 		 * (e40) getting turned into a 700 by BML RTAS.
1514 		 */
1515 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1516 		kvmppc_core_queue_program(vcpu, flags);
1517 		r = RESUME_GUEST;
1518 		break;
1519 	}
1520 	case BOOK3S_INTERRUPT_SYSCALL:
1521 	{
1522 		int i;
1523 
1524 		if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) {
1525 			/*
1526 			 * Guest userspace executed sc 1. This can only be
1527 			 * reached by the P9 path because the old path
1528 			 * handles this case in realmode hcall handlers.
1529 			 */
1530 			if (!kvmhv_vcpu_is_radix(vcpu)) {
1531 				/*
1532 				 * A guest could be running PR KVM, so this
1533 				 * may be a PR KVM hcall. It must be reflected
1534 				 * to the guest kernel as a sc interrupt.
1535 				 */
1536 				kvmppc_core_queue_syscall(vcpu);
1537 			} else {
1538 				/*
1539 				 * Radix guests can not run PR KVM or nested HV
1540 				 * hash guests which might run PR KVM, so this
1541 				 * is always a privilege fault. Send a program
1542 				 * check to guest kernel.
1543 				 */
1544 				kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1545 			}
1546 			r = RESUME_GUEST;
1547 			break;
1548 		}
1549 
1550 		/*
1551 		 * hcall - gather args and set exit_reason. This will next be
1552 		 * handled by kvmppc_pseries_do_hcall which may be able to deal
1553 		 * with it and resume guest, or may punt to userspace.
1554 		 */
1555 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1556 		for (i = 0; i < 9; ++i)
1557 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1558 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1559 		vcpu->arch.hcall_needed = 1;
1560 		r = RESUME_HOST;
1561 		break;
1562 	}
1563 	/*
1564 	 * We get these next two if the guest accesses a page which it thinks
1565 	 * it has mapped but which is not actually present, either because
1566 	 * it is for an emulated I/O device or because the corresonding
1567 	 * host page has been paged out.
1568 	 *
1569 	 * Any other HDSI/HISI interrupts have been handled already for P7/8
1570 	 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1571 	 * fault handling is done here.
1572 	 */
1573 	case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1574 		unsigned long vsid;
1575 		long err;
1576 
1577 		if (vcpu->arch.fault_dsisr == HDSISR_CANARY) {
1578 			r = RESUME_GUEST; /* Just retry if it's the canary */
1579 			break;
1580 		}
1581 
1582 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1583 			/*
1584 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1585 			 * already attempted to handle this in rmhandlers. The
1586 			 * hash fault handling below is v3 only (it uses ASDR
1587 			 * via fault_gpa).
1588 			 */
1589 			r = RESUME_PAGE_FAULT;
1590 			break;
1591 		}
1592 
1593 		if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1594 			kvmppc_core_queue_data_storage(vcpu,
1595 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1596 			r = RESUME_GUEST;
1597 			break;
1598 		}
1599 
1600 		if (!(vcpu->arch.shregs.msr & MSR_DR))
1601 			vsid = vcpu->kvm->arch.vrma_slb_v;
1602 		else
1603 			vsid = vcpu->arch.fault_gpa;
1604 
1605 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1606 				vsid, vcpu->arch.fault_dsisr, true);
1607 		if (err == 0) {
1608 			r = RESUME_GUEST;
1609 		} else if (err == -1 || err == -2) {
1610 			r = RESUME_PAGE_FAULT;
1611 		} else {
1612 			kvmppc_core_queue_data_storage(vcpu,
1613 				vcpu->arch.fault_dar, err);
1614 			r = RESUME_GUEST;
1615 		}
1616 		break;
1617 	}
1618 	case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1619 		unsigned long vsid;
1620 		long err;
1621 
1622 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1623 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1624 			DSISR_SRR1_MATCH_64S;
1625 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1626 			/*
1627 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1628 			 * already attempted to handle this in rmhandlers. The
1629 			 * hash fault handling below is v3 only (it uses ASDR
1630 			 * via fault_gpa).
1631 			 */
1632 			if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1633 				vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1634 			r = RESUME_PAGE_FAULT;
1635 			break;
1636 		}
1637 
1638 		if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1639 			kvmppc_core_queue_inst_storage(vcpu,
1640 				vcpu->arch.fault_dsisr);
1641 			r = RESUME_GUEST;
1642 			break;
1643 		}
1644 
1645 		if (!(vcpu->arch.shregs.msr & MSR_IR))
1646 			vsid = vcpu->kvm->arch.vrma_slb_v;
1647 		else
1648 			vsid = vcpu->arch.fault_gpa;
1649 
1650 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1651 				vsid, vcpu->arch.fault_dsisr, false);
1652 		if (err == 0) {
1653 			r = RESUME_GUEST;
1654 		} else if (err == -1) {
1655 			r = RESUME_PAGE_FAULT;
1656 		} else {
1657 			kvmppc_core_queue_inst_storage(vcpu, err);
1658 			r = RESUME_GUEST;
1659 		}
1660 		break;
1661 	}
1662 
1663 	/*
1664 	 * This occurs if the guest executes an illegal instruction.
1665 	 * If the guest debug is disabled, generate a program interrupt
1666 	 * to the guest. If guest debug is enabled, we need to check
1667 	 * whether the instruction is a software breakpoint instruction.
1668 	 * Accordingly return to Guest or Host.
1669 	 */
1670 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1671 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1672 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1673 				swab32(vcpu->arch.emul_inst) :
1674 				vcpu->arch.emul_inst;
1675 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1676 			r = kvmppc_emulate_debug_inst(vcpu);
1677 		} else {
1678 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1679 			r = RESUME_GUEST;
1680 		}
1681 		break;
1682 	/*
1683 	 * This occurs if the guest (kernel or userspace), does something that
1684 	 * is prohibited by HFSCR.
1685 	 * On POWER9, this could be a doorbell instruction that we need
1686 	 * to emulate.
1687 	 * Otherwise, we just generate a program interrupt to the guest.
1688 	 */
1689 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1690 		r = EMULATE_FAIL;
1691 		if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1692 		    cpu_has_feature(CPU_FTR_ARCH_300))
1693 			r = kvmppc_emulate_doorbell_instr(vcpu);
1694 		if (r == EMULATE_FAIL) {
1695 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1696 			r = RESUME_GUEST;
1697 		}
1698 		break;
1699 
1700 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1701 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1702 		/*
1703 		 * This occurs for various TM-related instructions that
1704 		 * we need to emulate on POWER9 DD2.2.  We have already
1705 		 * handled the cases where the guest was in real-suspend
1706 		 * mode and was transitioning to transactional state.
1707 		 */
1708 		r = kvmhv_p9_tm_emulation(vcpu);
1709 		break;
1710 #endif
1711 
1712 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1713 		r = RESUME_PASSTHROUGH;
1714 		break;
1715 	default:
1716 		kvmppc_dump_regs(vcpu);
1717 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1718 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1719 			vcpu->arch.shregs.msr);
1720 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1721 		r = RESUME_HOST;
1722 		break;
1723 	}
1724 
1725 	return r;
1726 }
1727 
1728 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1729 {
1730 	int r;
1731 	int srcu_idx;
1732 
1733 	vcpu->stat.sum_exits++;
1734 
1735 	/*
1736 	 * This can happen if an interrupt occurs in the last stages
1737 	 * of guest entry or the first stages of guest exit (i.e. after
1738 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1739 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1740 	 * That can happen due to a bug, or due to a machine check
1741 	 * occurring at just the wrong time.
1742 	 */
1743 	if (vcpu->arch.shregs.msr & MSR_HV) {
1744 		pr_emerg("KVM trap in HV mode while nested!\n");
1745 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1746 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1747 			 vcpu->arch.shregs.msr);
1748 		kvmppc_dump_regs(vcpu);
1749 		return RESUME_HOST;
1750 	}
1751 	switch (vcpu->arch.trap) {
1752 	/* We're good on these - the host merely wanted to get our attention */
1753 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1754 		vcpu->stat.dec_exits++;
1755 		r = RESUME_GUEST;
1756 		break;
1757 	case BOOK3S_INTERRUPT_EXTERNAL:
1758 		vcpu->stat.ext_intr_exits++;
1759 		r = RESUME_HOST;
1760 		break;
1761 	case BOOK3S_INTERRUPT_H_DOORBELL:
1762 	case BOOK3S_INTERRUPT_H_VIRT:
1763 		vcpu->stat.ext_intr_exits++;
1764 		r = RESUME_GUEST;
1765 		break;
1766 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1767 	case BOOK3S_INTERRUPT_HMI:
1768 	case BOOK3S_INTERRUPT_PERFMON:
1769 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1770 		r = RESUME_GUEST;
1771 		break;
1772 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1773 	{
1774 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1775 					      DEFAULT_RATELIMIT_BURST);
1776 		/* Pass the machine check to the L1 guest */
1777 		r = RESUME_HOST;
1778 		/* Print the MCE event to host console. */
1779 		if (__ratelimit(&rs))
1780 			machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1781 		break;
1782 	}
1783 	/*
1784 	 * We get these next two if the guest accesses a page which it thinks
1785 	 * it has mapped but which is not actually present, either because
1786 	 * it is for an emulated I/O device or because the corresonding
1787 	 * host page has been paged out.
1788 	 */
1789 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1790 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1791 		r = kvmhv_nested_page_fault(vcpu);
1792 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1793 		break;
1794 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1795 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1796 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1797 					 DSISR_SRR1_MATCH_64S;
1798 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1799 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1800 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1801 		r = kvmhv_nested_page_fault(vcpu);
1802 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1803 		break;
1804 
1805 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1806 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1807 		/*
1808 		 * This occurs for various TM-related instructions that
1809 		 * we need to emulate on POWER9 DD2.2.  We have already
1810 		 * handled the cases where the guest was in real-suspend
1811 		 * mode and was transitioning to transactional state.
1812 		 */
1813 		r = kvmhv_p9_tm_emulation(vcpu);
1814 		break;
1815 #endif
1816 
1817 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1818 		vcpu->arch.trap = 0;
1819 		r = RESUME_GUEST;
1820 		if (!xics_on_xive())
1821 			kvmppc_xics_rm_complete(vcpu, 0);
1822 		break;
1823 	case BOOK3S_INTERRUPT_SYSCALL:
1824 	{
1825 		unsigned long req = kvmppc_get_gpr(vcpu, 3);
1826 
1827 		/*
1828 		 * The H_RPT_INVALIDATE hcalls issued by nested
1829 		 * guests for process-scoped invalidations when
1830 		 * GTSE=0, are handled here in L0.
1831 		 */
1832 		if (req == H_RPT_INVALIDATE) {
1833 			r = kvmppc_nested_h_rpt_invalidate(vcpu);
1834 			break;
1835 		}
1836 
1837 		r = RESUME_HOST;
1838 		break;
1839 	}
1840 	default:
1841 		r = RESUME_HOST;
1842 		break;
1843 	}
1844 
1845 	return r;
1846 }
1847 
1848 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1849 					    struct kvm_sregs *sregs)
1850 {
1851 	int i;
1852 
1853 	memset(sregs, 0, sizeof(struct kvm_sregs));
1854 	sregs->pvr = vcpu->arch.pvr;
1855 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1856 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1857 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1858 	}
1859 
1860 	return 0;
1861 }
1862 
1863 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1864 					    struct kvm_sregs *sregs)
1865 {
1866 	int i, j;
1867 
1868 	/* Only accept the same PVR as the host's, since we can't spoof it */
1869 	if (sregs->pvr != vcpu->arch.pvr)
1870 		return -EINVAL;
1871 
1872 	j = 0;
1873 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1874 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1875 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1876 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1877 			++j;
1878 		}
1879 	}
1880 	vcpu->arch.slb_max = j;
1881 
1882 	return 0;
1883 }
1884 
1885 /*
1886  * Enforce limits on guest LPCR values based on hardware availability,
1887  * guest configuration, and possibly hypervisor support and security
1888  * concerns.
1889  */
1890 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
1891 {
1892 	/* LPCR_TC only applies to HPT guests */
1893 	if (kvm_is_radix(kvm))
1894 		lpcr &= ~LPCR_TC;
1895 
1896 	/* On POWER8 and above, userspace can modify AIL */
1897 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
1898 		lpcr &= ~LPCR_AIL;
1899 	if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
1900 		lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
1901 	/*
1902 	 * On some POWER9s we force AIL off for radix guests to prevent
1903 	 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
1904 	 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
1905 	 * be cached, which the host TLB management does not expect.
1906 	 */
1907 	if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
1908 		lpcr &= ~LPCR_AIL;
1909 
1910 	/*
1911 	 * On POWER9, allow userspace to enable large decrementer for the
1912 	 * guest, whether or not the host has it enabled.
1913 	 */
1914 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1915 		lpcr &= ~LPCR_LD;
1916 
1917 	return lpcr;
1918 }
1919 
1920 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
1921 {
1922 	if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
1923 		WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
1924 			  lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
1925 	}
1926 }
1927 
1928 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1929 		bool preserve_top32)
1930 {
1931 	struct kvm *kvm = vcpu->kvm;
1932 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1933 	u64 mask;
1934 
1935 	spin_lock(&vc->lock);
1936 
1937 	/*
1938 	 * Userspace can only modify
1939 	 * DPFD (default prefetch depth), ILE (interrupt little-endian),
1940 	 * TC (translation control), AIL (alternate interrupt location),
1941 	 * LD (large decrementer).
1942 	 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
1943 	 */
1944 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
1945 
1946 	/* Broken 32-bit version of LPCR must not clear top bits */
1947 	if (preserve_top32)
1948 		mask &= 0xFFFFFFFF;
1949 
1950 	new_lpcr = kvmppc_filter_lpcr_hv(kvm,
1951 			(vc->lpcr & ~mask) | (new_lpcr & mask));
1952 
1953 	/*
1954 	 * If ILE (interrupt little-endian) has changed, update the
1955 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1956 	 */
1957 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1958 		struct kvm_vcpu *vcpu;
1959 		int i;
1960 
1961 		kvm_for_each_vcpu(i, vcpu, kvm) {
1962 			if (vcpu->arch.vcore != vc)
1963 				continue;
1964 			if (new_lpcr & LPCR_ILE)
1965 				vcpu->arch.intr_msr |= MSR_LE;
1966 			else
1967 				vcpu->arch.intr_msr &= ~MSR_LE;
1968 		}
1969 	}
1970 
1971 	vc->lpcr = new_lpcr;
1972 
1973 	spin_unlock(&vc->lock);
1974 }
1975 
1976 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1977 				 union kvmppc_one_reg *val)
1978 {
1979 	int r = 0;
1980 	long int i;
1981 
1982 	switch (id) {
1983 	case KVM_REG_PPC_DEBUG_INST:
1984 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1985 		break;
1986 	case KVM_REG_PPC_HIOR:
1987 		*val = get_reg_val(id, 0);
1988 		break;
1989 	case KVM_REG_PPC_DABR:
1990 		*val = get_reg_val(id, vcpu->arch.dabr);
1991 		break;
1992 	case KVM_REG_PPC_DABRX:
1993 		*val = get_reg_val(id, vcpu->arch.dabrx);
1994 		break;
1995 	case KVM_REG_PPC_DSCR:
1996 		*val = get_reg_val(id, vcpu->arch.dscr);
1997 		break;
1998 	case KVM_REG_PPC_PURR:
1999 		*val = get_reg_val(id, vcpu->arch.purr);
2000 		break;
2001 	case KVM_REG_PPC_SPURR:
2002 		*val = get_reg_val(id, vcpu->arch.spurr);
2003 		break;
2004 	case KVM_REG_PPC_AMR:
2005 		*val = get_reg_val(id, vcpu->arch.amr);
2006 		break;
2007 	case KVM_REG_PPC_UAMOR:
2008 		*val = get_reg_val(id, vcpu->arch.uamor);
2009 		break;
2010 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2011 		i = id - KVM_REG_PPC_MMCR0;
2012 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
2013 		break;
2014 	case KVM_REG_PPC_MMCR2:
2015 		*val = get_reg_val(id, vcpu->arch.mmcr[2]);
2016 		break;
2017 	case KVM_REG_PPC_MMCRA:
2018 		*val = get_reg_val(id, vcpu->arch.mmcra);
2019 		break;
2020 	case KVM_REG_PPC_MMCRS:
2021 		*val = get_reg_val(id, vcpu->arch.mmcrs);
2022 		break;
2023 	case KVM_REG_PPC_MMCR3:
2024 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
2025 		break;
2026 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2027 		i = id - KVM_REG_PPC_PMC1;
2028 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
2029 		break;
2030 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2031 		i = id - KVM_REG_PPC_SPMC1;
2032 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
2033 		break;
2034 	case KVM_REG_PPC_SIAR:
2035 		*val = get_reg_val(id, vcpu->arch.siar);
2036 		break;
2037 	case KVM_REG_PPC_SDAR:
2038 		*val = get_reg_val(id, vcpu->arch.sdar);
2039 		break;
2040 	case KVM_REG_PPC_SIER:
2041 		*val = get_reg_val(id, vcpu->arch.sier[0]);
2042 		break;
2043 	case KVM_REG_PPC_SIER2:
2044 		*val = get_reg_val(id, vcpu->arch.sier[1]);
2045 		break;
2046 	case KVM_REG_PPC_SIER3:
2047 		*val = get_reg_val(id, vcpu->arch.sier[2]);
2048 		break;
2049 	case KVM_REG_PPC_IAMR:
2050 		*val = get_reg_val(id, vcpu->arch.iamr);
2051 		break;
2052 	case KVM_REG_PPC_PSPB:
2053 		*val = get_reg_val(id, vcpu->arch.pspb);
2054 		break;
2055 	case KVM_REG_PPC_DPDES:
2056 		/*
2057 		 * On POWER9, where we are emulating msgsndp etc.,
2058 		 * we return 1 bit for each vcpu, which can come from
2059 		 * either vcore->dpdes or doorbell_request.
2060 		 * On POWER8, doorbell_request is 0.
2061 		 */
2062 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes |
2063 				   vcpu->arch.doorbell_request);
2064 		break;
2065 	case KVM_REG_PPC_VTB:
2066 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
2067 		break;
2068 	case KVM_REG_PPC_DAWR:
2069 		*val = get_reg_val(id, vcpu->arch.dawr0);
2070 		break;
2071 	case KVM_REG_PPC_DAWRX:
2072 		*val = get_reg_val(id, vcpu->arch.dawrx0);
2073 		break;
2074 	case KVM_REG_PPC_DAWR1:
2075 		*val = get_reg_val(id, vcpu->arch.dawr1);
2076 		break;
2077 	case KVM_REG_PPC_DAWRX1:
2078 		*val = get_reg_val(id, vcpu->arch.dawrx1);
2079 		break;
2080 	case KVM_REG_PPC_CIABR:
2081 		*val = get_reg_val(id, vcpu->arch.ciabr);
2082 		break;
2083 	case KVM_REG_PPC_CSIGR:
2084 		*val = get_reg_val(id, vcpu->arch.csigr);
2085 		break;
2086 	case KVM_REG_PPC_TACR:
2087 		*val = get_reg_val(id, vcpu->arch.tacr);
2088 		break;
2089 	case KVM_REG_PPC_TCSCR:
2090 		*val = get_reg_val(id, vcpu->arch.tcscr);
2091 		break;
2092 	case KVM_REG_PPC_PID:
2093 		*val = get_reg_val(id, vcpu->arch.pid);
2094 		break;
2095 	case KVM_REG_PPC_ACOP:
2096 		*val = get_reg_val(id, vcpu->arch.acop);
2097 		break;
2098 	case KVM_REG_PPC_WORT:
2099 		*val = get_reg_val(id, vcpu->arch.wort);
2100 		break;
2101 	case KVM_REG_PPC_TIDR:
2102 		*val = get_reg_val(id, vcpu->arch.tid);
2103 		break;
2104 	case KVM_REG_PPC_PSSCR:
2105 		*val = get_reg_val(id, vcpu->arch.psscr);
2106 		break;
2107 	case KVM_REG_PPC_VPA_ADDR:
2108 		spin_lock(&vcpu->arch.vpa_update_lock);
2109 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2110 		spin_unlock(&vcpu->arch.vpa_update_lock);
2111 		break;
2112 	case KVM_REG_PPC_VPA_SLB:
2113 		spin_lock(&vcpu->arch.vpa_update_lock);
2114 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2115 		val->vpaval.length = vcpu->arch.slb_shadow.len;
2116 		spin_unlock(&vcpu->arch.vpa_update_lock);
2117 		break;
2118 	case KVM_REG_PPC_VPA_DTL:
2119 		spin_lock(&vcpu->arch.vpa_update_lock);
2120 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2121 		val->vpaval.length = vcpu->arch.dtl.len;
2122 		spin_unlock(&vcpu->arch.vpa_update_lock);
2123 		break;
2124 	case KVM_REG_PPC_TB_OFFSET:
2125 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
2126 		break;
2127 	case KVM_REG_PPC_LPCR:
2128 	case KVM_REG_PPC_LPCR_64:
2129 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
2130 		break;
2131 	case KVM_REG_PPC_PPR:
2132 		*val = get_reg_val(id, vcpu->arch.ppr);
2133 		break;
2134 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2135 	case KVM_REG_PPC_TFHAR:
2136 		*val = get_reg_val(id, vcpu->arch.tfhar);
2137 		break;
2138 	case KVM_REG_PPC_TFIAR:
2139 		*val = get_reg_val(id, vcpu->arch.tfiar);
2140 		break;
2141 	case KVM_REG_PPC_TEXASR:
2142 		*val = get_reg_val(id, vcpu->arch.texasr);
2143 		break;
2144 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2145 		i = id - KVM_REG_PPC_TM_GPR0;
2146 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2147 		break;
2148 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2149 	{
2150 		int j;
2151 		i = id - KVM_REG_PPC_TM_VSR0;
2152 		if (i < 32)
2153 			for (j = 0; j < TS_FPRWIDTH; j++)
2154 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2155 		else {
2156 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2157 				val->vval = vcpu->arch.vr_tm.vr[i-32];
2158 			else
2159 				r = -ENXIO;
2160 		}
2161 		break;
2162 	}
2163 	case KVM_REG_PPC_TM_CR:
2164 		*val = get_reg_val(id, vcpu->arch.cr_tm);
2165 		break;
2166 	case KVM_REG_PPC_TM_XER:
2167 		*val = get_reg_val(id, vcpu->arch.xer_tm);
2168 		break;
2169 	case KVM_REG_PPC_TM_LR:
2170 		*val = get_reg_val(id, vcpu->arch.lr_tm);
2171 		break;
2172 	case KVM_REG_PPC_TM_CTR:
2173 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
2174 		break;
2175 	case KVM_REG_PPC_TM_FPSCR:
2176 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2177 		break;
2178 	case KVM_REG_PPC_TM_AMR:
2179 		*val = get_reg_val(id, vcpu->arch.amr_tm);
2180 		break;
2181 	case KVM_REG_PPC_TM_PPR:
2182 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
2183 		break;
2184 	case KVM_REG_PPC_TM_VRSAVE:
2185 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
2186 		break;
2187 	case KVM_REG_PPC_TM_VSCR:
2188 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2189 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2190 		else
2191 			r = -ENXIO;
2192 		break;
2193 	case KVM_REG_PPC_TM_DSCR:
2194 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
2195 		break;
2196 	case KVM_REG_PPC_TM_TAR:
2197 		*val = get_reg_val(id, vcpu->arch.tar_tm);
2198 		break;
2199 #endif
2200 	case KVM_REG_PPC_ARCH_COMPAT:
2201 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
2202 		break;
2203 	case KVM_REG_PPC_DEC_EXPIRY:
2204 		*val = get_reg_val(id, vcpu->arch.dec_expires +
2205 				   vcpu->arch.vcore->tb_offset);
2206 		break;
2207 	case KVM_REG_PPC_ONLINE:
2208 		*val = get_reg_val(id, vcpu->arch.online);
2209 		break;
2210 	case KVM_REG_PPC_PTCR:
2211 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2212 		break;
2213 	default:
2214 		r = -EINVAL;
2215 		break;
2216 	}
2217 
2218 	return r;
2219 }
2220 
2221 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2222 				 union kvmppc_one_reg *val)
2223 {
2224 	int r = 0;
2225 	long int i;
2226 	unsigned long addr, len;
2227 
2228 	switch (id) {
2229 	case KVM_REG_PPC_HIOR:
2230 		/* Only allow this to be set to zero */
2231 		if (set_reg_val(id, *val))
2232 			r = -EINVAL;
2233 		break;
2234 	case KVM_REG_PPC_DABR:
2235 		vcpu->arch.dabr = set_reg_val(id, *val);
2236 		break;
2237 	case KVM_REG_PPC_DABRX:
2238 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2239 		break;
2240 	case KVM_REG_PPC_DSCR:
2241 		vcpu->arch.dscr = set_reg_val(id, *val);
2242 		break;
2243 	case KVM_REG_PPC_PURR:
2244 		vcpu->arch.purr = set_reg_val(id, *val);
2245 		break;
2246 	case KVM_REG_PPC_SPURR:
2247 		vcpu->arch.spurr = set_reg_val(id, *val);
2248 		break;
2249 	case KVM_REG_PPC_AMR:
2250 		vcpu->arch.amr = set_reg_val(id, *val);
2251 		break;
2252 	case KVM_REG_PPC_UAMOR:
2253 		vcpu->arch.uamor = set_reg_val(id, *val);
2254 		break;
2255 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2256 		i = id - KVM_REG_PPC_MMCR0;
2257 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2258 		break;
2259 	case KVM_REG_PPC_MMCR2:
2260 		vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2261 		break;
2262 	case KVM_REG_PPC_MMCRA:
2263 		vcpu->arch.mmcra = set_reg_val(id, *val);
2264 		break;
2265 	case KVM_REG_PPC_MMCRS:
2266 		vcpu->arch.mmcrs = set_reg_val(id, *val);
2267 		break;
2268 	case KVM_REG_PPC_MMCR3:
2269 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
2270 		break;
2271 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2272 		i = id - KVM_REG_PPC_PMC1;
2273 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
2274 		break;
2275 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2276 		i = id - KVM_REG_PPC_SPMC1;
2277 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
2278 		break;
2279 	case KVM_REG_PPC_SIAR:
2280 		vcpu->arch.siar = set_reg_val(id, *val);
2281 		break;
2282 	case KVM_REG_PPC_SDAR:
2283 		vcpu->arch.sdar = set_reg_val(id, *val);
2284 		break;
2285 	case KVM_REG_PPC_SIER:
2286 		vcpu->arch.sier[0] = set_reg_val(id, *val);
2287 		break;
2288 	case KVM_REG_PPC_SIER2:
2289 		vcpu->arch.sier[1] = set_reg_val(id, *val);
2290 		break;
2291 	case KVM_REG_PPC_SIER3:
2292 		vcpu->arch.sier[2] = set_reg_val(id, *val);
2293 		break;
2294 	case KVM_REG_PPC_IAMR:
2295 		vcpu->arch.iamr = set_reg_val(id, *val);
2296 		break;
2297 	case KVM_REG_PPC_PSPB:
2298 		vcpu->arch.pspb = set_reg_val(id, *val);
2299 		break;
2300 	case KVM_REG_PPC_DPDES:
2301 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2302 		break;
2303 	case KVM_REG_PPC_VTB:
2304 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2305 		break;
2306 	case KVM_REG_PPC_DAWR:
2307 		vcpu->arch.dawr0 = set_reg_val(id, *val);
2308 		break;
2309 	case KVM_REG_PPC_DAWRX:
2310 		vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2311 		break;
2312 	case KVM_REG_PPC_DAWR1:
2313 		vcpu->arch.dawr1 = set_reg_val(id, *val);
2314 		break;
2315 	case KVM_REG_PPC_DAWRX1:
2316 		vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2317 		break;
2318 	case KVM_REG_PPC_CIABR:
2319 		vcpu->arch.ciabr = set_reg_val(id, *val);
2320 		/* Don't allow setting breakpoints in hypervisor code */
2321 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2322 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
2323 		break;
2324 	case KVM_REG_PPC_CSIGR:
2325 		vcpu->arch.csigr = set_reg_val(id, *val);
2326 		break;
2327 	case KVM_REG_PPC_TACR:
2328 		vcpu->arch.tacr = set_reg_val(id, *val);
2329 		break;
2330 	case KVM_REG_PPC_TCSCR:
2331 		vcpu->arch.tcscr = set_reg_val(id, *val);
2332 		break;
2333 	case KVM_REG_PPC_PID:
2334 		vcpu->arch.pid = set_reg_val(id, *val);
2335 		break;
2336 	case KVM_REG_PPC_ACOP:
2337 		vcpu->arch.acop = set_reg_val(id, *val);
2338 		break;
2339 	case KVM_REG_PPC_WORT:
2340 		vcpu->arch.wort = set_reg_val(id, *val);
2341 		break;
2342 	case KVM_REG_PPC_TIDR:
2343 		vcpu->arch.tid = set_reg_val(id, *val);
2344 		break;
2345 	case KVM_REG_PPC_PSSCR:
2346 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2347 		break;
2348 	case KVM_REG_PPC_VPA_ADDR:
2349 		addr = set_reg_val(id, *val);
2350 		r = -EINVAL;
2351 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2352 			      vcpu->arch.dtl.next_gpa))
2353 			break;
2354 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2355 		break;
2356 	case KVM_REG_PPC_VPA_SLB:
2357 		addr = val->vpaval.addr;
2358 		len = val->vpaval.length;
2359 		r = -EINVAL;
2360 		if (addr && !vcpu->arch.vpa.next_gpa)
2361 			break;
2362 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2363 		break;
2364 	case KVM_REG_PPC_VPA_DTL:
2365 		addr = val->vpaval.addr;
2366 		len = val->vpaval.length;
2367 		r = -EINVAL;
2368 		if (addr && (len < sizeof(struct dtl_entry) ||
2369 			     !vcpu->arch.vpa.next_gpa))
2370 			break;
2371 		len -= len % sizeof(struct dtl_entry);
2372 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2373 		break;
2374 	case KVM_REG_PPC_TB_OFFSET:
2375 		/* round up to multiple of 2^24 */
2376 		vcpu->arch.vcore->tb_offset =
2377 			ALIGN(set_reg_val(id, *val), 1UL << 24);
2378 		break;
2379 	case KVM_REG_PPC_LPCR:
2380 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2381 		break;
2382 	case KVM_REG_PPC_LPCR_64:
2383 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2384 		break;
2385 	case KVM_REG_PPC_PPR:
2386 		vcpu->arch.ppr = set_reg_val(id, *val);
2387 		break;
2388 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2389 	case KVM_REG_PPC_TFHAR:
2390 		vcpu->arch.tfhar = set_reg_val(id, *val);
2391 		break;
2392 	case KVM_REG_PPC_TFIAR:
2393 		vcpu->arch.tfiar = set_reg_val(id, *val);
2394 		break;
2395 	case KVM_REG_PPC_TEXASR:
2396 		vcpu->arch.texasr = set_reg_val(id, *val);
2397 		break;
2398 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2399 		i = id - KVM_REG_PPC_TM_GPR0;
2400 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2401 		break;
2402 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2403 	{
2404 		int j;
2405 		i = id - KVM_REG_PPC_TM_VSR0;
2406 		if (i < 32)
2407 			for (j = 0; j < TS_FPRWIDTH; j++)
2408 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2409 		else
2410 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2411 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2412 			else
2413 				r = -ENXIO;
2414 		break;
2415 	}
2416 	case KVM_REG_PPC_TM_CR:
2417 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2418 		break;
2419 	case KVM_REG_PPC_TM_XER:
2420 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2421 		break;
2422 	case KVM_REG_PPC_TM_LR:
2423 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2424 		break;
2425 	case KVM_REG_PPC_TM_CTR:
2426 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2427 		break;
2428 	case KVM_REG_PPC_TM_FPSCR:
2429 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2430 		break;
2431 	case KVM_REG_PPC_TM_AMR:
2432 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2433 		break;
2434 	case KVM_REG_PPC_TM_PPR:
2435 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2436 		break;
2437 	case KVM_REG_PPC_TM_VRSAVE:
2438 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2439 		break;
2440 	case KVM_REG_PPC_TM_VSCR:
2441 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2442 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2443 		else
2444 			r = - ENXIO;
2445 		break;
2446 	case KVM_REG_PPC_TM_DSCR:
2447 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2448 		break;
2449 	case KVM_REG_PPC_TM_TAR:
2450 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2451 		break;
2452 #endif
2453 	case KVM_REG_PPC_ARCH_COMPAT:
2454 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2455 		break;
2456 	case KVM_REG_PPC_DEC_EXPIRY:
2457 		vcpu->arch.dec_expires = set_reg_val(id, *val) -
2458 			vcpu->arch.vcore->tb_offset;
2459 		break;
2460 	case KVM_REG_PPC_ONLINE:
2461 		i = set_reg_val(id, *val);
2462 		if (i && !vcpu->arch.online)
2463 			atomic_inc(&vcpu->arch.vcore->online_count);
2464 		else if (!i && vcpu->arch.online)
2465 			atomic_dec(&vcpu->arch.vcore->online_count);
2466 		vcpu->arch.online = i;
2467 		break;
2468 	case KVM_REG_PPC_PTCR:
2469 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2470 		break;
2471 	default:
2472 		r = -EINVAL;
2473 		break;
2474 	}
2475 
2476 	return r;
2477 }
2478 
2479 /*
2480  * On POWER9, threads are independent and can be in different partitions.
2481  * Therefore we consider each thread to be a subcore.
2482  * There is a restriction that all threads have to be in the same
2483  * MMU mode (radix or HPT), unfortunately, but since we only support
2484  * HPT guests on a HPT host so far, that isn't an impediment yet.
2485  */
2486 static int threads_per_vcore(struct kvm *kvm)
2487 {
2488 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2489 		return 1;
2490 	return threads_per_subcore;
2491 }
2492 
2493 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2494 {
2495 	struct kvmppc_vcore *vcore;
2496 
2497 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2498 
2499 	if (vcore == NULL)
2500 		return NULL;
2501 
2502 	spin_lock_init(&vcore->lock);
2503 	spin_lock_init(&vcore->stoltb_lock);
2504 	rcuwait_init(&vcore->wait);
2505 	vcore->preempt_tb = TB_NIL;
2506 	vcore->lpcr = kvm->arch.lpcr;
2507 	vcore->first_vcpuid = id;
2508 	vcore->kvm = kvm;
2509 	INIT_LIST_HEAD(&vcore->preempt_list);
2510 
2511 	return vcore;
2512 }
2513 
2514 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2515 static struct debugfs_timings_element {
2516 	const char *name;
2517 	size_t offset;
2518 } timings[] = {
2519 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2520 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2521 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2522 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2523 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2524 };
2525 
2526 #define N_TIMINGS	(ARRAY_SIZE(timings))
2527 
2528 struct debugfs_timings_state {
2529 	struct kvm_vcpu	*vcpu;
2530 	unsigned int	buflen;
2531 	char		buf[N_TIMINGS * 100];
2532 };
2533 
2534 static int debugfs_timings_open(struct inode *inode, struct file *file)
2535 {
2536 	struct kvm_vcpu *vcpu = inode->i_private;
2537 	struct debugfs_timings_state *p;
2538 
2539 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2540 	if (!p)
2541 		return -ENOMEM;
2542 
2543 	kvm_get_kvm(vcpu->kvm);
2544 	p->vcpu = vcpu;
2545 	file->private_data = p;
2546 
2547 	return nonseekable_open(inode, file);
2548 }
2549 
2550 static int debugfs_timings_release(struct inode *inode, struct file *file)
2551 {
2552 	struct debugfs_timings_state *p = file->private_data;
2553 
2554 	kvm_put_kvm(p->vcpu->kvm);
2555 	kfree(p);
2556 	return 0;
2557 }
2558 
2559 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2560 				    size_t len, loff_t *ppos)
2561 {
2562 	struct debugfs_timings_state *p = file->private_data;
2563 	struct kvm_vcpu *vcpu = p->vcpu;
2564 	char *s, *buf_end;
2565 	struct kvmhv_tb_accumulator tb;
2566 	u64 count;
2567 	loff_t pos;
2568 	ssize_t n;
2569 	int i, loops;
2570 	bool ok;
2571 
2572 	if (!p->buflen) {
2573 		s = p->buf;
2574 		buf_end = s + sizeof(p->buf);
2575 		for (i = 0; i < N_TIMINGS; ++i) {
2576 			struct kvmhv_tb_accumulator *acc;
2577 
2578 			acc = (struct kvmhv_tb_accumulator *)
2579 				((unsigned long)vcpu + timings[i].offset);
2580 			ok = false;
2581 			for (loops = 0; loops < 1000; ++loops) {
2582 				count = acc->seqcount;
2583 				if (!(count & 1)) {
2584 					smp_rmb();
2585 					tb = *acc;
2586 					smp_rmb();
2587 					if (count == acc->seqcount) {
2588 						ok = true;
2589 						break;
2590 					}
2591 				}
2592 				udelay(1);
2593 			}
2594 			if (!ok)
2595 				snprintf(s, buf_end - s, "%s: stuck\n",
2596 					timings[i].name);
2597 			else
2598 				snprintf(s, buf_end - s,
2599 					"%s: %llu %llu %llu %llu\n",
2600 					timings[i].name, count / 2,
2601 					tb_to_ns(tb.tb_total),
2602 					tb_to_ns(tb.tb_min),
2603 					tb_to_ns(tb.tb_max));
2604 			s += strlen(s);
2605 		}
2606 		p->buflen = s - p->buf;
2607 	}
2608 
2609 	pos = *ppos;
2610 	if (pos >= p->buflen)
2611 		return 0;
2612 	if (len > p->buflen - pos)
2613 		len = p->buflen - pos;
2614 	n = copy_to_user(buf, p->buf + pos, len);
2615 	if (n) {
2616 		if (n == len)
2617 			return -EFAULT;
2618 		len -= n;
2619 	}
2620 	*ppos = pos + len;
2621 	return len;
2622 }
2623 
2624 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2625 				     size_t len, loff_t *ppos)
2626 {
2627 	return -EACCES;
2628 }
2629 
2630 static const struct file_operations debugfs_timings_ops = {
2631 	.owner	 = THIS_MODULE,
2632 	.open	 = debugfs_timings_open,
2633 	.release = debugfs_timings_release,
2634 	.read	 = debugfs_timings_read,
2635 	.write	 = debugfs_timings_write,
2636 	.llseek	 = generic_file_llseek,
2637 };
2638 
2639 /* Create a debugfs directory for the vcpu */
2640 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2641 {
2642 	char buf[16];
2643 	struct kvm *kvm = vcpu->kvm;
2644 
2645 	snprintf(buf, sizeof(buf), "vcpu%u", id);
2646 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2647 	debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2648 			    &debugfs_timings_ops);
2649 }
2650 
2651 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2652 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2653 {
2654 }
2655 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2656 
2657 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2658 {
2659 	int err;
2660 	int core;
2661 	struct kvmppc_vcore *vcore;
2662 	struct kvm *kvm;
2663 	unsigned int id;
2664 
2665 	kvm = vcpu->kvm;
2666 	id = vcpu->vcpu_id;
2667 
2668 	vcpu->arch.shared = &vcpu->arch.shregs;
2669 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2670 	/*
2671 	 * The shared struct is never shared on HV,
2672 	 * so we can always use host endianness
2673 	 */
2674 #ifdef __BIG_ENDIAN__
2675 	vcpu->arch.shared_big_endian = true;
2676 #else
2677 	vcpu->arch.shared_big_endian = false;
2678 #endif
2679 #endif
2680 	vcpu->arch.mmcr[0] = MMCR0_FC;
2681 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2682 	/* default to host PVR, since we can't spoof it */
2683 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2684 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2685 	spin_lock_init(&vcpu->arch.tbacct_lock);
2686 	vcpu->arch.busy_preempt = TB_NIL;
2687 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2688 
2689 	/*
2690 	 * Set the default HFSCR for the guest from the host value.
2691 	 * This value is only used on POWER9.
2692 	 * On POWER9, we want to virtualize the doorbell facility, so we
2693 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2694 	 * to trap and then we emulate them.
2695 	 */
2696 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2697 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2698 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2699 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2700 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2701 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2702 			vcpu->arch.hfscr |= HFSCR_TM;
2703 #endif
2704 	}
2705 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2706 		vcpu->arch.hfscr |= HFSCR_TM;
2707 
2708 	kvmppc_mmu_book3s_hv_init(vcpu);
2709 
2710 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2711 
2712 	init_waitqueue_head(&vcpu->arch.cpu_run);
2713 
2714 	mutex_lock(&kvm->lock);
2715 	vcore = NULL;
2716 	err = -EINVAL;
2717 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2718 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2719 			pr_devel("KVM: VCPU ID too high\n");
2720 			core = KVM_MAX_VCORES;
2721 		} else {
2722 			BUG_ON(kvm->arch.smt_mode != 1);
2723 			core = kvmppc_pack_vcpu_id(kvm, id);
2724 		}
2725 	} else {
2726 		core = id / kvm->arch.smt_mode;
2727 	}
2728 	if (core < KVM_MAX_VCORES) {
2729 		vcore = kvm->arch.vcores[core];
2730 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2731 			pr_devel("KVM: collision on id %u", id);
2732 			vcore = NULL;
2733 		} else if (!vcore) {
2734 			/*
2735 			 * Take mmu_setup_lock for mutual exclusion
2736 			 * with kvmppc_update_lpcr().
2737 			 */
2738 			err = -ENOMEM;
2739 			vcore = kvmppc_vcore_create(kvm,
2740 					id & ~(kvm->arch.smt_mode - 1));
2741 			mutex_lock(&kvm->arch.mmu_setup_lock);
2742 			kvm->arch.vcores[core] = vcore;
2743 			kvm->arch.online_vcores++;
2744 			mutex_unlock(&kvm->arch.mmu_setup_lock);
2745 		}
2746 	}
2747 	mutex_unlock(&kvm->lock);
2748 
2749 	if (!vcore)
2750 		return err;
2751 
2752 	spin_lock(&vcore->lock);
2753 	++vcore->num_threads;
2754 	spin_unlock(&vcore->lock);
2755 	vcpu->arch.vcore = vcore;
2756 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2757 	vcpu->arch.thread_cpu = -1;
2758 	vcpu->arch.prev_cpu = -1;
2759 
2760 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2761 	kvmppc_sanity_check(vcpu);
2762 
2763 	debugfs_vcpu_init(vcpu, id);
2764 
2765 	return 0;
2766 }
2767 
2768 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2769 			      unsigned long flags)
2770 {
2771 	int err;
2772 	int esmt = 0;
2773 
2774 	if (flags)
2775 		return -EINVAL;
2776 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2777 		return -EINVAL;
2778 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2779 		/*
2780 		 * On POWER8 (or POWER7), the threading mode is "strict",
2781 		 * so we pack smt_mode vcpus per vcore.
2782 		 */
2783 		if (smt_mode > threads_per_subcore)
2784 			return -EINVAL;
2785 	} else {
2786 		/*
2787 		 * On POWER9, the threading mode is "loose",
2788 		 * so each vcpu gets its own vcore.
2789 		 */
2790 		esmt = smt_mode;
2791 		smt_mode = 1;
2792 	}
2793 	mutex_lock(&kvm->lock);
2794 	err = -EBUSY;
2795 	if (!kvm->arch.online_vcores) {
2796 		kvm->arch.smt_mode = smt_mode;
2797 		kvm->arch.emul_smt_mode = esmt;
2798 		err = 0;
2799 	}
2800 	mutex_unlock(&kvm->lock);
2801 
2802 	return err;
2803 }
2804 
2805 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2806 {
2807 	if (vpa->pinned_addr)
2808 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2809 					vpa->dirty);
2810 }
2811 
2812 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2813 {
2814 	spin_lock(&vcpu->arch.vpa_update_lock);
2815 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2816 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2817 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2818 	spin_unlock(&vcpu->arch.vpa_update_lock);
2819 }
2820 
2821 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2822 {
2823 	/* Indicate we want to get back into the guest */
2824 	return 1;
2825 }
2826 
2827 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2828 {
2829 	unsigned long dec_nsec, now;
2830 
2831 	now = get_tb();
2832 	if (now > vcpu->arch.dec_expires) {
2833 		/* decrementer has already gone negative */
2834 		kvmppc_core_queue_dec(vcpu);
2835 		kvmppc_core_prepare_to_enter(vcpu);
2836 		return;
2837 	}
2838 	dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2839 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2840 	vcpu->arch.timer_running = 1;
2841 }
2842 
2843 extern int __kvmppc_vcore_entry(void);
2844 
2845 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2846 				   struct kvm_vcpu *vcpu)
2847 {
2848 	u64 now;
2849 
2850 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2851 		return;
2852 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2853 	now = mftb();
2854 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2855 		vcpu->arch.stolen_logged;
2856 	vcpu->arch.busy_preempt = now;
2857 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2858 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2859 	--vc->n_runnable;
2860 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2861 }
2862 
2863 static int kvmppc_grab_hwthread(int cpu)
2864 {
2865 	struct paca_struct *tpaca;
2866 	long timeout = 10000;
2867 
2868 	tpaca = paca_ptrs[cpu];
2869 
2870 	/* Ensure the thread won't go into the kernel if it wakes */
2871 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2872 	tpaca->kvm_hstate.kvm_vcore = NULL;
2873 	tpaca->kvm_hstate.napping = 0;
2874 	smp_wmb();
2875 	tpaca->kvm_hstate.hwthread_req = 1;
2876 
2877 	/*
2878 	 * If the thread is already executing in the kernel (e.g. handling
2879 	 * a stray interrupt), wait for it to get back to nap mode.
2880 	 * The smp_mb() is to ensure that our setting of hwthread_req
2881 	 * is visible before we look at hwthread_state, so if this
2882 	 * races with the code at system_reset_pSeries and the thread
2883 	 * misses our setting of hwthread_req, we are sure to see its
2884 	 * setting of hwthread_state, and vice versa.
2885 	 */
2886 	smp_mb();
2887 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2888 		if (--timeout <= 0) {
2889 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2890 			return -EBUSY;
2891 		}
2892 		udelay(1);
2893 	}
2894 	return 0;
2895 }
2896 
2897 static void kvmppc_release_hwthread(int cpu)
2898 {
2899 	struct paca_struct *tpaca;
2900 
2901 	tpaca = paca_ptrs[cpu];
2902 	tpaca->kvm_hstate.hwthread_req = 0;
2903 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2904 	tpaca->kvm_hstate.kvm_vcore = NULL;
2905 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2906 }
2907 
2908 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2909 {
2910 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2911 	cpumask_t *cpu_in_guest;
2912 	int i;
2913 
2914 	cpu = cpu_first_tlb_thread_sibling(cpu);
2915 	if (nested) {
2916 		cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2917 		cpu_in_guest = &nested->cpu_in_guest;
2918 	} else {
2919 		cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2920 		cpu_in_guest = &kvm->arch.cpu_in_guest;
2921 	}
2922 	/*
2923 	 * Make sure setting of bit in need_tlb_flush precedes
2924 	 * testing of cpu_in_guest bits.  The matching barrier on
2925 	 * the other side is the first smp_mb() in kvmppc_run_core().
2926 	 */
2927 	smp_mb();
2928 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
2929 					i += cpu_tlb_thread_sibling_step())
2930 		if (cpumask_test_cpu(i, cpu_in_guest))
2931 			smp_call_function_single(i, do_nothing, NULL, 1);
2932 }
2933 
2934 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2935 {
2936 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2937 	struct kvm *kvm = vcpu->kvm;
2938 	int prev_cpu;
2939 
2940 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2941 		return;
2942 
2943 	if (nested)
2944 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2945 	else
2946 		prev_cpu = vcpu->arch.prev_cpu;
2947 
2948 	/*
2949 	 * With radix, the guest can do TLB invalidations itself,
2950 	 * and it could choose to use the local form (tlbiel) if
2951 	 * it is invalidating a translation that has only ever been
2952 	 * used on one vcpu.  However, that doesn't mean it has
2953 	 * only ever been used on one physical cpu, since vcpus
2954 	 * can move around between pcpus.  To cope with this, when
2955 	 * a vcpu moves from one pcpu to another, we need to tell
2956 	 * any vcpus running on the same core as this vcpu previously
2957 	 * ran to flush the TLB.  The TLB is shared between threads,
2958 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2959 	 */
2960 	if (prev_cpu != pcpu) {
2961 		if (prev_cpu >= 0 &&
2962 		    cpu_first_tlb_thread_sibling(prev_cpu) !=
2963 		    cpu_first_tlb_thread_sibling(pcpu))
2964 			radix_flush_cpu(kvm, prev_cpu, vcpu);
2965 		if (nested)
2966 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2967 		else
2968 			vcpu->arch.prev_cpu = pcpu;
2969 	}
2970 }
2971 
2972 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2973 {
2974 	int cpu;
2975 	struct paca_struct *tpaca;
2976 	struct kvm *kvm = vc->kvm;
2977 
2978 	cpu = vc->pcpu;
2979 	if (vcpu) {
2980 		if (vcpu->arch.timer_running) {
2981 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2982 			vcpu->arch.timer_running = 0;
2983 		}
2984 		cpu += vcpu->arch.ptid;
2985 		vcpu->cpu = vc->pcpu;
2986 		vcpu->arch.thread_cpu = cpu;
2987 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2988 	}
2989 	tpaca = paca_ptrs[cpu];
2990 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2991 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2992 	tpaca->kvm_hstate.fake_suspend = 0;
2993 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2994 	smp_wmb();
2995 	tpaca->kvm_hstate.kvm_vcore = vc;
2996 	if (cpu != smp_processor_id())
2997 		kvmppc_ipi_thread(cpu);
2998 }
2999 
3000 static void kvmppc_wait_for_nap(int n_threads)
3001 {
3002 	int cpu = smp_processor_id();
3003 	int i, loops;
3004 
3005 	if (n_threads <= 1)
3006 		return;
3007 	for (loops = 0; loops < 1000000; ++loops) {
3008 		/*
3009 		 * Check if all threads are finished.
3010 		 * We set the vcore pointer when starting a thread
3011 		 * and the thread clears it when finished, so we look
3012 		 * for any threads that still have a non-NULL vcore ptr.
3013 		 */
3014 		for (i = 1; i < n_threads; ++i)
3015 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3016 				break;
3017 		if (i == n_threads) {
3018 			HMT_medium();
3019 			return;
3020 		}
3021 		HMT_low();
3022 	}
3023 	HMT_medium();
3024 	for (i = 1; i < n_threads; ++i)
3025 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3026 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3027 }
3028 
3029 /*
3030  * Check that we are on thread 0 and that any other threads in
3031  * this core are off-line.  Then grab the threads so they can't
3032  * enter the kernel.
3033  */
3034 static int on_primary_thread(void)
3035 {
3036 	int cpu = smp_processor_id();
3037 	int thr;
3038 
3039 	/* Are we on a primary subcore? */
3040 	if (cpu_thread_in_subcore(cpu))
3041 		return 0;
3042 
3043 	thr = 0;
3044 	while (++thr < threads_per_subcore)
3045 		if (cpu_online(cpu + thr))
3046 			return 0;
3047 
3048 	/* Grab all hw threads so they can't go into the kernel */
3049 	for (thr = 1; thr < threads_per_subcore; ++thr) {
3050 		if (kvmppc_grab_hwthread(cpu + thr)) {
3051 			/* Couldn't grab one; let the others go */
3052 			do {
3053 				kvmppc_release_hwthread(cpu + thr);
3054 			} while (--thr > 0);
3055 			return 0;
3056 		}
3057 	}
3058 	return 1;
3059 }
3060 
3061 /*
3062  * A list of virtual cores for each physical CPU.
3063  * These are vcores that could run but their runner VCPU tasks are
3064  * (or may be) preempted.
3065  */
3066 struct preempted_vcore_list {
3067 	struct list_head	list;
3068 	spinlock_t		lock;
3069 };
3070 
3071 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3072 
3073 static void init_vcore_lists(void)
3074 {
3075 	int cpu;
3076 
3077 	for_each_possible_cpu(cpu) {
3078 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3079 		spin_lock_init(&lp->lock);
3080 		INIT_LIST_HEAD(&lp->list);
3081 	}
3082 }
3083 
3084 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3085 {
3086 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3087 
3088 	vc->vcore_state = VCORE_PREEMPT;
3089 	vc->pcpu = smp_processor_id();
3090 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3091 		spin_lock(&lp->lock);
3092 		list_add_tail(&vc->preempt_list, &lp->list);
3093 		spin_unlock(&lp->lock);
3094 	}
3095 
3096 	/* Start accumulating stolen time */
3097 	kvmppc_core_start_stolen(vc);
3098 }
3099 
3100 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3101 {
3102 	struct preempted_vcore_list *lp;
3103 
3104 	kvmppc_core_end_stolen(vc);
3105 	if (!list_empty(&vc->preempt_list)) {
3106 		lp = &per_cpu(preempted_vcores, vc->pcpu);
3107 		spin_lock(&lp->lock);
3108 		list_del_init(&vc->preempt_list);
3109 		spin_unlock(&lp->lock);
3110 	}
3111 	vc->vcore_state = VCORE_INACTIVE;
3112 }
3113 
3114 /*
3115  * This stores information about the virtual cores currently
3116  * assigned to a physical core.
3117  */
3118 struct core_info {
3119 	int		n_subcores;
3120 	int		max_subcore_threads;
3121 	int		total_threads;
3122 	int		subcore_threads[MAX_SUBCORES];
3123 	struct kvmppc_vcore *vc[MAX_SUBCORES];
3124 };
3125 
3126 /*
3127  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3128  * respectively in 2-way micro-threading (split-core) mode on POWER8.
3129  */
3130 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3131 
3132 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3133 {
3134 	memset(cip, 0, sizeof(*cip));
3135 	cip->n_subcores = 1;
3136 	cip->max_subcore_threads = vc->num_threads;
3137 	cip->total_threads = vc->num_threads;
3138 	cip->subcore_threads[0] = vc->num_threads;
3139 	cip->vc[0] = vc;
3140 }
3141 
3142 static bool subcore_config_ok(int n_subcores, int n_threads)
3143 {
3144 	/*
3145 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3146 	 * split-core mode, with one thread per subcore.
3147 	 */
3148 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3149 		return n_subcores <= 4 && n_threads == 1;
3150 
3151 	/* On POWER8, can only dynamically split if unsplit to begin with */
3152 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3153 		return false;
3154 	if (n_subcores > MAX_SUBCORES)
3155 		return false;
3156 	if (n_subcores > 1) {
3157 		if (!(dynamic_mt_modes & 2))
3158 			n_subcores = 4;
3159 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3160 			return false;
3161 	}
3162 
3163 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3164 }
3165 
3166 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3167 {
3168 	vc->entry_exit_map = 0;
3169 	vc->in_guest = 0;
3170 	vc->napping_threads = 0;
3171 	vc->conferring_threads = 0;
3172 	vc->tb_offset_applied = 0;
3173 }
3174 
3175 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3176 {
3177 	int n_threads = vc->num_threads;
3178 	int sub;
3179 
3180 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3181 		return false;
3182 
3183 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
3184 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3185 		return false;
3186 
3187 	if (n_threads < cip->max_subcore_threads)
3188 		n_threads = cip->max_subcore_threads;
3189 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3190 		return false;
3191 	cip->max_subcore_threads = n_threads;
3192 
3193 	sub = cip->n_subcores;
3194 	++cip->n_subcores;
3195 	cip->total_threads += vc->num_threads;
3196 	cip->subcore_threads[sub] = vc->num_threads;
3197 	cip->vc[sub] = vc;
3198 	init_vcore_to_run(vc);
3199 	list_del_init(&vc->preempt_list);
3200 
3201 	return true;
3202 }
3203 
3204 /*
3205  * Work out whether it is possible to piggyback the execution of
3206  * vcore *pvc onto the execution of the other vcores described in *cip.
3207  */
3208 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3209 			  int target_threads)
3210 {
3211 	if (cip->total_threads + pvc->num_threads > target_threads)
3212 		return false;
3213 
3214 	return can_dynamic_split(pvc, cip);
3215 }
3216 
3217 static void prepare_threads(struct kvmppc_vcore *vc)
3218 {
3219 	int i;
3220 	struct kvm_vcpu *vcpu;
3221 
3222 	for_each_runnable_thread(i, vcpu, vc) {
3223 		if (signal_pending(vcpu->arch.run_task))
3224 			vcpu->arch.ret = -EINTR;
3225 		else if (vcpu->arch.vpa.update_pending ||
3226 			 vcpu->arch.slb_shadow.update_pending ||
3227 			 vcpu->arch.dtl.update_pending)
3228 			vcpu->arch.ret = RESUME_GUEST;
3229 		else
3230 			continue;
3231 		kvmppc_remove_runnable(vc, vcpu);
3232 		wake_up(&vcpu->arch.cpu_run);
3233 	}
3234 }
3235 
3236 static void collect_piggybacks(struct core_info *cip, int target_threads)
3237 {
3238 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3239 	struct kvmppc_vcore *pvc, *vcnext;
3240 
3241 	spin_lock(&lp->lock);
3242 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3243 		if (!spin_trylock(&pvc->lock))
3244 			continue;
3245 		prepare_threads(pvc);
3246 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3247 			list_del_init(&pvc->preempt_list);
3248 			if (pvc->runner == NULL) {
3249 				pvc->vcore_state = VCORE_INACTIVE;
3250 				kvmppc_core_end_stolen(pvc);
3251 			}
3252 			spin_unlock(&pvc->lock);
3253 			continue;
3254 		}
3255 		if (!can_piggyback(pvc, cip, target_threads)) {
3256 			spin_unlock(&pvc->lock);
3257 			continue;
3258 		}
3259 		kvmppc_core_end_stolen(pvc);
3260 		pvc->vcore_state = VCORE_PIGGYBACK;
3261 		if (cip->total_threads >= target_threads)
3262 			break;
3263 	}
3264 	spin_unlock(&lp->lock);
3265 }
3266 
3267 static bool recheck_signals_and_mmu(struct core_info *cip)
3268 {
3269 	int sub, i;
3270 	struct kvm_vcpu *vcpu;
3271 	struct kvmppc_vcore *vc;
3272 
3273 	for (sub = 0; sub < cip->n_subcores; ++sub) {
3274 		vc = cip->vc[sub];
3275 		if (!vc->kvm->arch.mmu_ready)
3276 			return true;
3277 		for_each_runnable_thread(i, vcpu, vc)
3278 			if (signal_pending(vcpu->arch.run_task))
3279 				return true;
3280 	}
3281 	return false;
3282 }
3283 
3284 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3285 {
3286 	int still_running = 0, i;
3287 	u64 now;
3288 	long ret;
3289 	struct kvm_vcpu *vcpu;
3290 
3291 	spin_lock(&vc->lock);
3292 	now = get_tb();
3293 	for_each_runnable_thread(i, vcpu, vc) {
3294 		/*
3295 		 * It's safe to unlock the vcore in the loop here, because
3296 		 * for_each_runnable_thread() is safe against removal of
3297 		 * the vcpu, and the vcore state is VCORE_EXITING here,
3298 		 * so any vcpus becoming runnable will have their arch.trap
3299 		 * set to zero and can't actually run in the guest.
3300 		 */
3301 		spin_unlock(&vc->lock);
3302 		/* cancel pending dec exception if dec is positive */
3303 		if (now < vcpu->arch.dec_expires &&
3304 		    kvmppc_core_pending_dec(vcpu))
3305 			kvmppc_core_dequeue_dec(vcpu);
3306 
3307 		trace_kvm_guest_exit(vcpu);
3308 
3309 		ret = RESUME_GUEST;
3310 		if (vcpu->arch.trap)
3311 			ret = kvmppc_handle_exit_hv(vcpu,
3312 						    vcpu->arch.run_task);
3313 
3314 		vcpu->arch.ret = ret;
3315 		vcpu->arch.trap = 0;
3316 
3317 		spin_lock(&vc->lock);
3318 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3319 			if (vcpu->arch.pending_exceptions)
3320 				kvmppc_core_prepare_to_enter(vcpu);
3321 			if (vcpu->arch.ceded)
3322 				kvmppc_set_timer(vcpu);
3323 			else
3324 				++still_running;
3325 		} else {
3326 			kvmppc_remove_runnable(vc, vcpu);
3327 			wake_up(&vcpu->arch.cpu_run);
3328 		}
3329 	}
3330 	if (!is_master) {
3331 		if (still_running > 0) {
3332 			kvmppc_vcore_preempt(vc);
3333 		} else if (vc->runner) {
3334 			vc->vcore_state = VCORE_PREEMPT;
3335 			kvmppc_core_start_stolen(vc);
3336 		} else {
3337 			vc->vcore_state = VCORE_INACTIVE;
3338 		}
3339 		if (vc->n_runnable > 0 && vc->runner == NULL) {
3340 			/* make sure there's a candidate runner awake */
3341 			i = -1;
3342 			vcpu = next_runnable_thread(vc, &i);
3343 			wake_up(&vcpu->arch.cpu_run);
3344 		}
3345 	}
3346 	spin_unlock(&vc->lock);
3347 }
3348 
3349 /*
3350  * Clear core from the list of active host cores as we are about to
3351  * enter the guest. Only do this if it is the primary thread of the
3352  * core (not if a subcore) that is entering the guest.
3353  */
3354 static inline int kvmppc_clear_host_core(unsigned int cpu)
3355 {
3356 	int core;
3357 
3358 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3359 		return 0;
3360 	/*
3361 	 * Memory barrier can be omitted here as we will do a smp_wmb()
3362 	 * later in kvmppc_start_thread and we need ensure that state is
3363 	 * visible to other CPUs only after we enter guest.
3364 	 */
3365 	core = cpu >> threads_shift;
3366 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3367 	return 0;
3368 }
3369 
3370 /*
3371  * Advertise this core as an active host core since we exited the guest
3372  * Only need to do this if it is the primary thread of the core that is
3373  * exiting.
3374  */
3375 static inline int kvmppc_set_host_core(unsigned int cpu)
3376 {
3377 	int core;
3378 
3379 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3380 		return 0;
3381 
3382 	/*
3383 	 * Memory barrier can be omitted here because we do a spin_unlock
3384 	 * immediately after this which provides the memory barrier.
3385 	 */
3386 	core = cpu >> threads_shift;
3387 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3388 	return 0;
3389 }
3390 
3391 static void set_irq_happened(int trap)
3392 {
3393 	switch (trap) {
3394 	case BOOK3S_INTERRUPT_EXTERNAL:
3395 		local_paca->irq_happened |= PACA_IRQ_EE;
3396 		break;
3397 	case BOOK3S_INTERRUPT_H_DOORBELL:
3398 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3399 		break;
3400 	case BOOK3S_INTERRUPT_HMI:
3401 		local_paca->irq_happened |= PACA_IRQ_HMI;
3402 		break;
3403 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3404 		replay_system_reset();
3405 		break;
3406 	}
3407 }
3408 
3409 /*
3410  * Run a set of guest threads on a physical core.
3411  * Called with vc->lock held.
3412  */
3413 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3414 {
3415 	struct kvm_vcpu *vcpu;
3416 	int i;
3417 	int srcu_idx;
3418 	struct core_info core_info;
3419 	struct kvmppc_vcore *pvc;
3420 	struct kvm_split_mode split_info, *sip;
3421 	int split, subcore_size, active;
3422 	int sub;
3423 	bool thr0_done;
3424 	unsigned long cmd_bit, stat_bit;
3425 	int pcpu, thr;
3426 	int target_threads;
3427 	int controlled_threads;
3428 	int trap;
3429 	bool is_power8;
3430 
3431 	if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3432 		return;
3433 
3434 	/*
3435 	 * Remove from the list any threads that have a signal pending
3436 	 * or need a VPA update done
3437 	 */
3438 	prepare_threads(vc);
3439 
3440 	/* if the runner is no longer runnable, let the caller pick a new one */
3441 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3442 		return;
3443 
3444 	/*
3445 	 * Initialize *vc.
3446 	 */
3447 	init_vcore_to_run(vc);
3448 	vc->preempt_tb = TB_NIL;
3449 
3450 	/*
3451 	 * Number of threads that we will be controlling: the same as
3452 	 * the number of threads per subcore, except on POWER9,
3453 	 * where it's 1 because the threads are (mostly) independent.
3454 	 */
3455 	controlled_threads = threads_per_vcore(vc->kvm);
3456 
3457 	/*
3458 	 * Make sure we are running on primary threads, and that secondary
3459 	 * threads are offline.  Also check if the number of threads in this
3460 	 * guest are greater than the current system threads per guest.
3461 	 */
3462 	if ((controlled_threads > 1) &&
3463 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3464 		for_each_runnable_thread(i, vcpu, vc) {
3465 			vcpu->arch.ret = -EBUSY;
3466 			kvmppc_remove_runnable(vc, vcpu);
3467 			wake_up(&vcpu->arch.cpu_run);
3468 		}
3469 		goto out;
3470 	}
3471 
3472 	/*
3473 	 * See if we could run any other vcores on the physical core
3474 	 * along with this one.
3475 	 */
3476 	init_core_info(&core_info, vc);
3477 	pcpu = smp_processor_id();
3478 	target_threads = controlled_threads;
3479 	if (target_smt_mode && target_smt_mode < target_threads)
3480 		target_threads = target_smt_mode;
3481 	if (vc->num_threads < target_threads)
3482 		collect_piggybacks(&core_info, target_threads);
3483 
3484 	/*
3485 	 * Hard-disable interrupts, and check resched flag and signals.
3486 	 * If we need to reschedule or deliver a signal, clean up
3487 	 * and return without going into the guest(s).
3488 	 * If the mmu_ready flag has been cleared, don't go into the
3489 	 * guest because that means a HPT resize operation is in progress.
3490 	 */
3491 	local_irq_disable();
3492 	hard_irq_disable();
3493 	if (lazy_irq_pending() || need_resched() ||
3494 	    recheck_signals_and_mmu(&core_info)) {
3495 		local_irq_enable();
3496 		vc->vcore_state = VCORE_INACTIVE;
3497 		/* Unlock all except the primary vcore */
3498 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3499 			pvc = core_info.vc[sub];
3500 			/* Put back on to the preempted vcores list */
3501 			kvmppc_vcore_preempt(pvc);
3502 			spin_unlock(&pvc->lock);
3503 		}
3504 		for (i = 0; i < controlled_threads; ++i)
3505 			kvmppc_release_hwthread(pcpu + i);
3506 		return;
3507 	}
3508 
3509 	kvmppc_clear_host_core(pcpu);
3510 
3511 	/* Decide on micro-threading (split-core) mode */
3512 	subcore_size = threads_per_subcore;
3513 	cmd_bit = stat_bit = 0;
3514 	split = core_info.n_subcores;
3515 	sip = NULL;
3516 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3517 
3518 	if (split > 1) {
3519 		sip = &split_info;
3520 		memset(&split_info, 0, sizeof(split_info));
3521 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3522 			split_info.vc[sub] = core_info.vc[sub];
3523 
3524 		if (is_power8) {
3525 			if (split == 2 && (dynamic_mt_modes & 2)) {
3526 				cmd_bit = HID0_POWER8_1TO2LPAR;
3527 				stat_bit = HID0_POWER8_2LPARMODE;
3528 			} else {
3529 				split = 4;
3530 				cmd_bit = HID0_POWER8_1TO4LPAR;
3531 				stat_bit = HID0_POWER8_4LPARMODE;
3532 			}
3533 			subcore_size = MAX_SMT_THREADS / split;
3534 			split_info.rpr = mfspr(SPRN_RPR);
3535 			split_info.pmmar = mfspr(SPRN_PMMAR);
3536 			split_info.ldbar = mfspr(SPRN_LDBAR);
3537 			split_info.subcore_size = subcore_size;
3538 		} else {
3539 			split_info.subcore_size = 1;
3540 		}
3541 
3542 		/* order writes to split_info before kvm_split_mode pointer */
3543 		smp_wmb();
3544 	}
3545 
3546 	for (thr = 0; thr < controlled_threads; ++thr) {
3547 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3548 
3549 		paca->kvm_hstate.napping = 0;
3550 		paca->kvm_hstate.kvm_split_mode = sip;
3551 	}
3552 
3553 	/* Initiate micro-threading (split-core) on POWER8 if required */
3554 	if (cmd_bit) {
3555 		unsigned long hid0 = mfspr(SPRN_HID0);
3556 
3557 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3558 		mb();
3559 		mtspr(SPRN_HID0, hid0);
3560 		isync();
3561 		for (;;) {
3562 			hid0 = mfspr(SPRN_HID0);
3563 			if (hid0 & stat_bit)
3564 				break;
3565 			cpu_relax();
3566 		}
3567 	}
3568 
3569 	/*
3570 	 * On POWER8, set RWMR register.
3571 	 * Since it only affects PURR and SPURR, it doesn't affect
3572 	 * the host, so we don't save/restore the host value.
3573 	 */
3574 	if (is_power8) {
3575 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3576 		int n_online = atomic_read(&vc->online_count);
3577 
3578 		/*
3579 		 * Use the 8-thread value if we're doing split-core
3580 		 * or if the vcore's online count looks bogus.
3581 		 */
3582 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3583 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3584 			rwmr_val = p8_rwmr_values[n_online];
3585 		mtspr(SPRN_RWMR, rwmr_val);
3586 	}
3587 
3588 	/* Start all the threads */
3589 	active = 0;
3590 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3591 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3592 		thr0_done = false;
3593 		active |= 1 << thr;
3594 		pvc = core_info.vc[sub];
3595 		pvc->pcpu = pcpu + thr;
3596 		for_each_runnable_thread(i, vcpu, pvc) {
3597 			kvmppc_start_thread(vcpu, pvc);
3598 			kvmppc_create_dtl_entry(vcpu, pvc);
3599 			trace_kvm_guest_enter(vcpu);
3600 			if (!vcpu->arch.ptid)
3601 				thr0_done = true;
3602 			active |= 1 << (thr + vcpu->arch.ptid);
3603 		}
3604 		/*
3605 		 * We need to start the first thread of each subcore
3606 		 * even if it doesn't have a vcpu.
3607 		 */
3608 		if (!thr0_done)
3609 			kvmppc_start_thread(NULL, pvc);
3610 	}
3611 
3612 	/*
3613 	 * Ensure that split_info.do_nap is set after setting
3614 	 * the vcore pointer in the PACA of the secondaries.
3615 	 */
3616 	smp_mb();
3617 
3618 	/*
3619 	 * When doing micro-threading, poke the inactive threads as well.
3620 	 * This gets them to the nap instruction after kvm_do_nap,
3621 	 * which reduces the time taken to unsplit later.
3622 	 */
3623 	if (cmd_bit) {
3624 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3625 		for (thr = 1; thr < threads_per_subcore; ++thr)
3626 			if (!(active & (1 << thr)))
3627 				kvmppc_ipi_thread(pcpu + thr);
3628 	}
3629 
3630 	vc->vcore_state = VCORE_RUNNING;
3631 	preempt_disable();
3632 
3633 	trace_kvmppc_run_core(vc, 0);
3634 
3635 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3636 		spin_unlock(&core_info.vc[sub]->lock);
3637 
3638 	guest_enter_irqoff();
3639 
3640 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3641 
3642 	this_cpu_disable_ftrace();
3643 
3644 	/*
3645 	 * Interrupts will be enabled once we get into the guest,
3646 	 * so tell lockdep that we're about to enable interrupts.
3647 	 */
3648 	trace_hardirqs_on();
3649 
3650 	trap = __kvmppc_vcore_entry();
3651 
3652 	trace_hardirqs_off();
3653 
3654 	this_cpu_enable_ftrace();
3655 
3656 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3657 
3658 	set_irq_happened(trap);
3659 
3660 	spin_lock(&vc->lock);
3661 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3662 	vc->vcore_state = VCORE_EXITING;
3663 
3664 	/* wait for secondary threads to finish writing their state to memory */
3665 	kvmppc_wait_for_nap(controlled_threads);
3666 
3667 	/* Return to whole-core mode if we split the core earlier */
3668 	if (cmd_bit) {
3669 		unsigned long hid0 = mfspr(SPRN_HID0);
3670 		unsigned long loops = 0;
3671 
3672 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3673 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3674 		mb();
3675 		mtspr(SPRN_HID0, hid0);
3676 		isync();
3677 		for (;;) {
3678 			hid0 = mfspr(SPRN_HID0);
3679 			if (!(hid0 & stat_bit))
3680 				break;
3681 			cpu_relax();
3682 			++loops;
3683 		}
3684 		split_info.do_nap = 0;
3685 	}
3686 
3687 	kvmppc_set_host_core(pcpu);
3688 
3689 	guest_exit_irqoff();
3690 
3691 	local_irq_enable();
3692 
3693 	/* Let secondaries go back to the offline loop */
3694 	for (i = 0; i < controlled_threads; ++i) {
3695 		kvmppc_release_hwthread(pcpu + i);
3696 		if (sip && sip->napped[i])
3697 			kvmppc_ipi_thread(pcpu + i);
3698 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3699 	}
3700 
3701 	spin_unlock(&vc->lock);
3702 
3703 	/* make sure updates to secondary vcpu structs are visible now */
3704 	smp_mb();
3705 
3706 	preempt_enable();
3707 
3708 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3709 		pvc = core_info.vc[sub];
3710 		post_guest_process(pvc, pvc == vc);
3711 	}
3712 
3713 	spin_lock(&vc->lock);
3714 
3715  out:
3716 	vc->vcore_state = VCORE_INACTIVE;
3717 	trace_kvmppc_run_core(vc, 1);
3718 }
3719 
3720 static void load_spr_state(struct kvm_vcpu *vcpu)
3721 {
3722 	mtspr(SPRN_DSCR, vcpu->arch.dscr);
3723 	mtspr(SPRN_IAMR, vcpu->arch.iamr);
3724 	mtspr(SPRN_PSPB, vcpu->arch.pspb);
3725 	mtspr(SPRN_FSCR, vcpu->arch.fscr);
3726 	mtspr(SPRN_TAR, vcpu->arch.tar);
3727 	mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3728 	mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3729 	mtspr(SPRN_BESCR, vcpu->arch.bescr);
3730 	mtspr(SPRN_WORT, vcpu->arch.wort);
3731 	mtspr(SPRN_TIDR, vcpu->arch.tid);
3732 	mtspr(SPRN_AMR, vcpu->arch.amr);
3733 	mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3734 
3735 	/*
3736 	 * DAR, DSISR, and for nested HV, SPRGs must be set with MSR[RI]
3737 	 * clear (or hstate set appropriately to catch those registers
3738 	 * being clobbered if we take a MCE or SRESET), so those are done
3739 	 * later.
3740 	 */
3741 
3742 	if (!(vcpu->arch.ctrl & 1))
3743 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3744 }
3745 
3746 static void store_spr_state(struct kvm_vcpu *vcpu)
3747 {
3748 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3749 
3750 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
3751 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
3752 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
3753 	vcpu->arch.tar = mfspr(SPRN_TAR);
3754 	vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3755 	vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3756 	vcpu->arch.bescr = mfspr(SPRN_BESCR);
3757 	vcpu->arch.wort = mfspr(SPRN_WORT);
3758 	vcpu->arch.tid = mfspr(SPRN_TIDR);
3759 	vcpu->arch.amr = mfspr(SPRN_AMR);
3760 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3761 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
3762 }
3763 
3764 /*
3765  * Privileged (non-hypervisor) host registers to save.
3766  */
3767 struct p9_host_os_sprs {
3768 	unsigned long dscr;
3769 	unsigned long tidr;
3770 	unsigned long iamr;
3771 	unsigned long amr;
3772 	unsigned long fscr;
3773 };
3774 
3775 static void save_p9_host_os_sprs(struct p9_host_os_sprs *host_os_sprs)
3776 {
3777 	host_os_sprs->dscr = mfspr(SPRN_DSCR);
3778 	host_os_sprs->tidr = mfspr(SPRN_TIDR);
3779 	host_os_sprs->iamr = mfspr(SPRN_IAMR);
3780 	host_os_sprs->amr = mfspr(SPRN_AMR);
3781 	host_os_sprs->fscr = mfspr(SPRN_FSCR);
3782 }
3783 
3784 /* vcpu guest regs must already be saved */
3785 static void restore_p9_host_os_sprs(struct kvm_vcpu *vcpu,
3786 				    struct p9_host_os_sprs *host_os_sprs)
3787 {
3788 	mtspr(SPRN_PSPB, 0);
3789 	mtspr(SPRN_WORT, 0);
3790 	mtspr(SPRN_UAMOR, 0);
3791 
3792 	mtspr(SPRN_DSCR, host_os_sprs->dscr);
3793 	mtspr(SPRN_TIDR, host_os_sprs->tidr);
3794 	mtspr(SPRN_IAMR, host_os_sprs->iamr);
3795 
3796 	if (host_os_sprs->amr != vcpu->arch.amr)
3797 		mtspr(SPRN_AMR, host_os_sprs->amr);
3798 
3799 	if (host_os_sprs->fscr != vcpu->arch.fscr)
3800 		mtspr(SPRN_FSCR, host_os_sprs->fscr);
3801 
3802 	/* Save guest CTRL register, set runlatch to 1 */
3803 	if (!(vcpu->arch.ctrl & 1))
3804 		mtspr(SPRN_CTRLT, 1);
3805 }
3806 
3807 static inline bool hcall_is_xics(unsigned long req)
3808 {
3809 	return req == H_EOI || req == H_CPPR || req == H_IPI ||
3810 		req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
3811 }
3812 
3813 /*
3814  * Guest entry for POWER9 and later CPUs.
3815  */
3816 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3817 			 unsigned long lpcr)
3818 {
3819 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3820 	struct p9_host_os_sprs host_os_sprs;
3821 	s64 dec;
3822 	u64 tb;
3823 	int trap, save_pmu;
3824 
3825 	WARN_ON_ONCE(vcpu->arch.ceded);
3826 
3827 	dec = mfspr(SPRN_DEC);
3828 	tb = mftb();
3829 	if (dec < 0)
3830 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3831 	local_paca->kvm_hstate.dec_expires = dec + tb;
3832 	if (local_paca->kvm_hstate.dec_expires < time_limit)
3833 		time_limit = local_paca->kvm_hstate.dec_expires;
3834 
3835 	save_p9_host_os_sprs(&host_os_sprs);
3836 
3837 	kvmhv_save_host_pmu();		/* saves it to PACA kvm_hstate */
3838 
3839 	kvmppc_subcore_enter_guest();
3840 
3841 	vc->entry_exit_map = 1;
3842 	vc->in_guest = 1;
3843 
3844 	if (vcpu->arch.vpa.pinned_addr) {
3845 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3846 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3847 		lp->yield_count = cpu_to_be32(yield_count);
3848 		vcpu->arch.vpa.dirty = 1;
3849 	}
3850 
3851 	if (cpu_has_feature(CPU_FTR_TM) ||
3852 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3853 		kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3854 
3855 	kvmhv_load_guest_pmu(vcpu);
3856 
3857 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3858 	load_fp_state(&vcpu->arch.fp);
3859 #ifdef CONFIG_ALTIVEC
3860 	load_vr_state(&vcpu->arch.vr);
3861 #endif
3862 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3863 
3864 	load_spr_state(vcpu);
3865 
3866 	/*
3867 	 * When setting DEC, we must always deal with irq_work_raise via NMI vs
3868 	 * setting DEC. The problem occurs right as we switch into guest mode
3869 	 * if a NMI hits and sets pending work and sets DEC, then that will
3870 	 * apply to the guest and not bring us back to the host.
3871 	 *
3872 	 * irq_work_raise could check a flag (or possibly LPCR[HDICE] for
3873 	 * example) and set HDEC to 1? That wouldn't solve the nested hv
3874 	 * case which needs to abort the hcall or zero the time limit.
3875 	 *
3876 	 * XXX: Another day's problem.
3877 	 */
3878 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3879 
3880 	if (kvmhv_on_pseries()) {
3881 		/*
3882 		 * We need to save and restore the guest visible part of the
3883 		 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3884 		 * doesn't do this for us. Note only required if pseries since
3885 		 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
3886 		 */
3887 		unsigned long host_psscr;
3888 		/* call our hypervisor to load up HV regs and go */
3889 		struct hv_guest_state hvregs;
3890 
3891 		host_psscr = mfspr(SPRN_PSSCR_PR);
3892 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3893 		kvmhv_save_hv_regs(vcpu, &hvregs);
3894 		hvregs.lpcr = lpcr;
3895 		vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3896 		hvregs.version = HV_GUEST_STATE_VERSION;
3897 		if (vcpu->arch.nested) {
3898 			hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3899 			hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3900 		} else {
3901 			hvregs.lpid = vcpu->kvm->arch.lpid;
3902 			hvregs.vcpu_token = vcpu->vcpu_id;
3903 		}
3904 		hvregs.hdec_expiry = time_limit;
3905 		mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3906 		mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3907 		trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3908 					  __pa(&vcpu->arch.regs));
3909 		kvmhv_restore_hv_return_state(vcpu, &hvregs);
3910 		vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3911 		vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3912 		vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3913 		vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3914 		mtspr(SPRN_PSSCR_PR, host_psscr);
3915 
3916 		/* H_CEDE has to be handled now, not later */
3917 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3918 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3919 			kvmppc_cede(vcpu);
3920 			kvmppc_set_gpr(vcpu, 3, 0);
3921 			trap = 0;
3922 		}
3923 	} else {
3924 		kvmppc_xive_push_vcpu(vcpu);
3925 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr);
3926 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3927 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
3928 			unsigned long req = kvmppc_get_gpr(vcpu, 3);
3929 
3930 			/* H_CEDE has to be handled now, not later */
3931 			if (req == H_CEDE) {
3932 				kvmppc_cede(vcpu);
3933 				kvmppc_xive_rearm_escalation(vcpu); /* may un-cede */
3934 				kvmppc_set_gpr(vcpu, 3, 0);
3935 				trap = 0;
3936 
3937 			/* XICS hcalls must be handled before xive is pulled */
3938 			} else if (hcall_is_xics(req)) {
3939 				int ret;
3940 
3941 				ret = kvmppc_xive_xics_hcall(vcpu, req);
3942 				if (ret != H_TOO_HARD) {
3943 					kvmppc_set_gpr(vcpu, 3, ret);
3944 					trap = 0;
3945 				}
3946 			}
3947 		}
3948 		kvmppc_xive_pull_vcpu(vcpu);
3949 
3950 		if (kvm_is_radix(vcpu->kvm))
3951 			vcpu->arch.slb_max = 0;
3952 	}
3953 
3954 	dec = mfspr(SPRN_DEC);
3955 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3956 		dec = (s32) dec;
3957 	tb = mftb();
3958 	vcpu->arch.dec_expires = dec + tb;
3959 	vcpu->cpu = -1;
3960 	vcpu->arch.thread_cpu = -1;
3961 
3962 	store_spr_state(vcpu);
3963 
3964 	restore_p9_host_os_sprs(vcpu, &host_os_sprs);
3965 
3966 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3967 	store_fp_state(&vcpu->arch.fp);
3968 #ifdef CONFIG_ALTIVEC
3969 	store_vr_state(&vcpu->arch.vr);
3970 #endif
3971 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3972 
3973 	if (cpu_has_feature(CPU_FTR_TM) ||
3974 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3975 		kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3976 
3977 	save_pmu = 1;
3978 	if (vcpu->arch.vpa.pinned_addr) {
3979 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3980 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3981 		lp->yield_count = cpu_to_be32(yield_count);
3982 		vcpu->arch.vpa.dirty = 1;
3983 		save_pmu = lp->pmcregs_in_use;
3984 	}
3985 	/* Must save pmu if this guest is capable of running nested guests */
3986 	save_pmu |= nesting_enabled(vcpu->kvm);
3987 
3988 	kvmhv_save_guest_pmu(vcpu, save_pmu);
3989 
3990 	vc->entry_exit_map = 0x101;
3991 	vc->in_guest = 0;
3992 
3993 	mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3994 	/* We may have raced with new irq work */
3995 	if (test_irq_work_pending())
3996 		set_dec(1);
3997 	mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3998 
3999 	kvmhv_load_host_pmu();
4000 
4001 	kvmppc_subcore_exit_guest();
4002 
4003 	return trap;
4004 }
4005 
4006 /*
4007  * Wait for some other vcpu thread to execute us, and
4008  * wake us up when we need to handle something in the host.
4009  */
4010 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4011 				 struct kvm_vcpu *vcpu, int wait_state)
4012 {
4013 	DEFINE_WAIT(wait);
4014 
4015 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4016 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4017 		spin_unlock(&vc->lock);
4018 		schedule();
4019 		spin_lock(&vc->lock);
4020 	}
4021 	finish_wait(&vcpu->arch.cpu_run, &wait);
4022 }
4023 
4024 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4025 {
4026 	if (!halt_poll_ns_grow)
4027 		return;
4028 
4029 	vc->halt_poll_ns *= halt_poll_ns_grow;
4030 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4031 		vc->halt_poll_ns = halt_poll_ns_grow_start;
4032 }
4033 
4034 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4035 {
4036 	if (halt_poll_ns_shrink == 0)
4037 		vc->halt_poll_ns = 0;
4038 	else
4039 		vc->halt_poll_ns /= halt_poll_ns_shrink;
4040 }
4041 
4042 #ifdef CONFIG_KVM_XICS
4043 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4044 {
4045 	if (!xics_on_xive())
4046 		return false;
4047 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4048 		vcpu->arch.xive_saved_state.cppr;
4049 }
4050 #else
4051 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4052 {
4053 	return false;
4054 }
4055 #endif /* CONFIG_KVM_XICS */
4056 
4057 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4058 {
4059 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4060 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4061 		return true;
4062 
4063 	return false;
4064 }
4065 
4066 /*
4067  * Check to see if any of the runnable vcpus on the vcore have pending
4068  * exceptions or are no longer ceded
4069  */
4070 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4071 {
4072 	struct kvm_vcpu *vcpu;
4073 	int i;
4074 
4075 	for_each_runnable_thread(i, vcpu, vc) {
4076 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4077 			return 1;
4078 	}
4079 
4080 	return 0;
4081 }
4082 
4083 /*
4084  * All the vcpus in this vcore are idle, so wait for a decrementer
4085  * or external interrupt to one of the vcpus.  vc->lock is held.
4086  */
4087 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4088 {
4089 	ktime_t cur, start_poll, start_wait;
4090 	int do_sleep = 1;
4091 	u64 block_ns;
4092 
4093 	/* Poll for pending exceptions and ceded state */
4094 	cur = start_poll = ktime_get();
4095 	if (vc->halt_poll_ns) {
4096 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4097 		++vc->runner->stat.generic.halt_attempted_poll;
4098 
4099 		vc->vcore_state = VCORE_POLLING;
4100 		spin_unlock(&vc->lock);
4101 
4102 		do {
4103 			if (kvmppc_vcore_check_block(vc)) {
4104 				do_sleep = 0;
4105 				break;
4106 			}
4107 			cur = ktime_get();
4108 		} while (kvm_vcpu_can_poll(cur, stop));
4109 
4110 		spin_lock(&vc->lock);
4111 		vc->vcore_state = VCORE_INACTIVE;
4112 
4113 		if (!do_sleep) {
4114 			++vc->runner->stat.generic.halt_successful_poll;
4115 			goto out;
4116 		}
4117 	}
4118 
4119 	prepare_to_rcuwait(&vc->wait);
4120 	set_current_state(TASK_INTERRUPTIBLE);
4121 	if (kvmppc_vcore_check_block(vc)) {
4122 		finish_rcuwait(&vc->wait);
4123 		do_sleep = 0;
4124 		/* If we polled, count this as a successful poll */
4125 		if (vc->halt_poll_ns)
4126 			++vc->runner->stat.generic.halt_successful_poll;
4127 		goto out;
4128 	}
4129 
4130 	start_wait = ktime_get();
4131 
4132 	vc->vcore_state = VCORE_SLEEPING;
4133 	trace_kvmppc_vcore_blocked(vc, 0);
4134 	spin_unlock(&vc->lock);
4135 	schedule();
4136 	finish_rcuwait(&vc->wait);
4137 	spin_lock(&vc->lock);
4138 	vc->vcore_state = VCORE_INACTIVE;
4139 	trace_kvmppc_vcore_blocked(vc, 1);
4140 	++vc->runner->stat.halt_successful_wait;
4141 
4142 	cur = ktime_get();
4143 
4144 out:
4145 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4146 
4147 	/* Attribute wait time */
4148 	if (do_sleep) {
4149 		vc->runner->stat.halt_wait_ns +=
4150 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
4151 		/* Attribute failed poll time */
4152 		if (vc->halt_poll_ns)
4153 			vc->runner->stat.generic.halt_poll_fail_ns +=
4154 				ktime_to_ns(start_wait) -
4155 				ktime_to_ns(start_poll);
4156 	} else {
4157 		/* Attribute successful poll time */
4158 		if (vc->halt_poll_ns)
4159 			vc->runner->stat.generic.halt_poll_success_ns +=
4160 				ktime_to_ns(cur) -
4161 				ktime_to_ns(start_poll);
4162 	}
4163 
4164 	/* Adjust poll time */
4165 	if (halt_poll_ns) {
4166 		if (block_ns <= vc->halt_poll_ns)
4167 			;
4168 		/* We slept and blocked for longer than the max halt time */
4169 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4170 			shrink_halt_poll_ns(vc);
4171 		/* We slept and our poll time is too small */
4172 		else if (vc->halt_poll_ns < halt_poll_ns &&
4173 				block_ns < halt_poll_ns)
4174 			grow_halt_poll_ns(vc);
4175 		if (vc->halt_poll_ns > halt_poll_ns)
4176 			vc->halt_poll_ns = halt_poll_ns;
4177 	} else
4178 		vc->halt_poll_ns = 0;
4179 
4180 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4181 }
4182 
4183 /*
4184  * This never fails for a radix guest, as none of the operations it does
4185  * for a radix guest can fail or have a way to report failure.
4186  */
4187 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4188 {
4189 	int r = 0;
4190 	struct kvm *kvm = vcpu->kvm;
4191 
4192 	mutex_lock(&kvm->arch.mmu_setup_lock);
4193 	if (!kvm->arch.mmu_ready) {
4194 		if (!kvm_is_radix(kvm))
4195 			r = kvmppc_hv_setup_htab_rma(vcpu);
4196 		if (!r) {
4197 			if (cpu_has_feature(CPU_FTR_ARCH_300))
4198 				kvmppc_setup_partition_table(kvm);
4199 			kvm->arch.mmu_ready = 1;
4200 		}
4201 	}
4202 	mutex_unlock(&kvm->arch.mmu_setup_lock);
4203 	return r;
4204 }
4205 
4206 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4207 {
4208 	struct kvm_run *run = vcpu->run;
4209 	int n_ceded, i, r;
4210 	struct kvmppc_vcore *vc;
4211 	struct kvm_vcpu *v;
4212 
4213 	trace_kvmppc_run_vcpu_enter(vcpu);
4214 
4215 	run->exit_reason = 0;
4216 	vcpu->arch.ret = RESUME_GUEST;
4217 	vcpu->arch.trap = 0;
4218 	kvmppc_update_vpas(vcpu);
4219 
4220 	/*
4221 	 * Synchronize with other threads in this virtual core
4222 	 */
4223 	vc = vcpu->arch.vcore;
4224 	spin_lock(&vc->lock);
4225 	vcpu->arch.ceded = 0;
4226 	vcpu->arch.run_task = current;
4227 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4228 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4229 	vcpu->arch.busy_preempt = TB_NIL;
4230 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4231 	++vc->n_runnable;
4232 
4233 	/*
4234 	 * This happens the first time this is called for a vcpu.
4235 	 * If the vcore is already running, we may be able to start
4236 	 * this thread straight away and have it join in.
4237 	 */
4238 	if (!signal_pending(current)) {
4239 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
4240 		     vc->vcore_state == VCORE_RUNNING) &&
4241 			   !VCORE_IS_EXITING(vc)) {
4242 			kvmppc_create_dtl_entry(vcpu, vc);
4243 			kvmppc_start_thread(vcpu, vc);
4244 			trace_kvm_guest_enter(vcpu);
4245 		} else if (vc->vcore_state == VCORE_SLEEPING) {
4246 		        rcuwait_wake_up(&vc->wait);
4247 		}
4248 
4249 	}
4250 
4251 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4252 	       !signal_pending(current)) {
4253 		/* See if the MMU is ready to go */
4254 		if (!vcpu->kvm->arch.mmu_ready) {
4255 			spin_unlock(&vc->lock);
4256 			r = kvmhv_setup_mmu(vcpu);
4257 			spin_lock(&vc->lock);
4258 			if (r) {
4259 				run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4260 				run->fail_entry.
4261 					hardware_entry_failure_reason = 0;
4262 				vcpu->arch.ret = r;
4263 				break;
4264 			}
4265 		}
4266 
4267 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4268 			kvmppc_vcore_end_preempt(vc);
4269 
4270 		if (vc->vcore_state != VCORE_INACTIVE) {
4271 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4272 			continue;
4273 		}
4274 		for_each_runnable_thread(i, v, vc) {
4275 			kvmppc_core_prepare_to_enter(v);
4276 			if (signal_pending(v->arch.run_task)) {
4277 				kvmppc_remove_runnable(vc, v);
4278 				v->stat.signal_exits++;
4279 				v->run->exit_reason = KVM_EXIT_INTR;
4280 				v->arch.ret = -EINTR;
4281 				wake_up(&v->arch.cpu_run);
4282 			}
4283 		}
4284 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4285 			break;
4286 		n_ceded = 0;
4287 		for_each_runnable_thread(i, v, vc) {
4288 			if (!kvmppc_vcpu_woken(v))
4289 				n_ceded += v->arch.ceded;
4290 			else
4291 				v->arch.ceded = 0;
4292 		}
4293 		vc->runner = vcpu;
4294 		if (n_ceded == vc->n_runnable) {
4295 			kvmppc_vcore_blocked(vc);
4296 		} else if (need_resched()) {
4297 			kvmppc_vcore_preempt(vc);
4298 			/* Let something else run */
4299 			cond_resched_lock(&vc->lock);
4300 			if (vc->vcore_state == VCORE_PREEMPT)
4301 				kvmppc_vcore_end_preempt(vc);
4302 		} else {
4303 			kvmppc_run_core(vc);
4304 		}
4305 		vc->runner = NULL;
4306 	}
4307 
4308 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4309 	       (vc->vcore_state == VCORE_RUNNING ||
4310 		vc->vcore_state == VCORE_EXITING ||
4311 		vc->vcore_state == VCORE_PIGGYBACK))
4312 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4313 
4314 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4315 		kvmppc_vcore_end_preempt(vc);
4316 
4317 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4318 		kvmppc_remove_runnable(vc, vcpu);
4319 		vcpu->stat.signal_exits++;
4320 		run->exit_reason = KVM_EXIT_INTR;
4321 		vcpu->arch.ret = -EINTR;
4322 	}
4323 
4324 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4325 		/* Wake up some vcpu to run the core */
4326 		i = -1;
4327 		v = next_runnable_thread(vc, &i);
4328 		wake_up(&v->arch.cpu_run);
4329 	}
4330 
4331 	trace_kvmppc_run_vcpu_exit(vcpu);
4332 	spin_unlock(&vc->lock);
4333 	return vcpu->arch.ret;
4334 }
4335 
4336 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4337 			  unsigned long lpcr)
4338 {
4339 	struct kvm_run *run = vcpu->run;
4340 	int trap, r, pcpu;
4341 	int srcu_idx;
4342 	struct kvmppc_vcore *vc;
4343 	struct kvm *kvm = vcpu->kvm;
4344 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4345 
4346 	trace_kvmppc_run_vcpu_enter(vcpu);
4347 
4348 	run->exit_reason = 0;
4349 	vcpu->arch.ret = RESUME_GUEST;
4350 	vcpu->arch.trap = 0;
4351 
4352 	vc = vcpu->arch.vcore;
4353 	vcpu->arch.ceded = 0;
4354 	vcpu->arch.run_task = current;
4355 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4356 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4357 	vcpu->arch.busy_preempt = TB_NIL;
4358 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4359 	vc->runnable_threads[0] = vcpu;
4360 	vc->n_runnable = 1;
4361 	vc->runner = vcpu;
4362 
4363 	/* See if the MMU is ready to go */
4364 	if (!kvm->arch.mmu_ready) {
4365 		r = kvmhv_setup_mmu(vcpu);
4366 		if (r) {
4367 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4368 			run->fail_entry.hardware_entry_failure_reason = 0;
4369 			vcpu->arch.ret = r;
4370 			return r;
4371 		}
4372 	}
4373 
4374 	if (need_resched())
4375 		cond_resched();
4376 
4377 	kvmppc_update_vpas(vcpu);
4378 
4379 	init_vcore_to_run(vc);
4380 	vc->preempt_tb = TB_NIL;
4381 
4382 	preempt_disable();
4383 	pcpu = smp_processor_id();
4384 	vc->pcpu = pcpu;
4385 	if (kvm_is_radix(kvm))
4386 		kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4387 
4388 	local_irq_disable();
4389 	hard_irq_disable();
4390 	if (signal_pending(current))
4391 		goto sigpend;
4392 	if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4393 		goto out;
4394 
4395 	if (!nested) {
4396 		kvmppc_core_prepare_to_enter(vcpu);
4397 		if (vcpu->arch.doorbell_request) {
4398 			vc->dpdes = 1;
4399 			smp_wmb();
4400 			vcpu->arch.doorbell_request = 0;
4401 		}
4402 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4403 			     &vcpu->arch.pending_exceptions))
4404 			lpcr |= LPCR_MER;
4405 	} else if (vcpu->arch.pending_exceptions ||
4406 		   vcpu->arch.doorbell_request ||
4407 		   xive_interrupt_pending(vcpu)) {
4408 		vcpu->arch.ret = RESUME_HOST;
4409 		goto out;
4410 	}
4411 
4412 	kvmppc_clear_host_core(pcpu);
4413 
4414 	local_paca->kvm_hstate.napping = 0;
4415 	local_paca->kvm_hstate.kvm_split_mode = NULL;
4416 	kvmppc_start_thread(vcpu, vc);
4417 	kvmppc_create_dtl_entry(vcpu, vc);
4418 	trace_kvm_guest_enter(vcpu);
4419 
4420 	vc->vcore_state = VCORE_RUNNING;
4421 	trace_kvmppc_run_core(vc, 0);
4422 
4423 	guest_enter_irqoff();
4424 
4425 	srcu_idx = srcu_read_lock(&kvm->srcu);
4426 
4427 	this_cpu_disable_ftrace();
4428 
4429 	/* Tell lockdep that we're about to enable interrupts */
4430 	trace_hardirqs_on();
4431 
4432 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4433 	vcpu->arch.trap = trap;
4434 
4435 	trace_hardirqs_off();
4436 
4437 	this_cpu_enable_ftrace();
4438 
4439 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4440 
4441 	set_irq_happened(trap);
4442 
4443 	kvmppc_set_host_core(pcpu);
4444 
4445 	guest_exit_irqoff();
4446 
4447 	local_irq_enable();
4448 
4449 	cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4450 
4451 	preempt_enable();
4452 
4453 	/*
4454 	 * cancel pending decrementer exception if DEC is now positive, or if
4455 	 * entering a nested guest in which case the decrementer is now owned
4456 	 * by L2 and the L1 decrementer is provided in hdec_expires
4457 	 */
4458 	if (kvmppc_core_pending_dec(vcpu) &&
4459 			((get_tb() < vcpu->arch.dec_expires) ||
4460 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4461 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4462 		kvmppc_core_dequeue_dec(vcpu);
4463 
4464 	trace_kvm_guest_exit(vcpu);
4465 	r = RESUME_GUEST;
4466 	if (trap) {
4467 		if (!nested)
4468 			r = kvmppc_handle_exit_hv(vcpu, current);
4469 		else
4470 			r = kvmppc_handle_nested_exit(vcpu);
4471 	}
4472 	vcpu->arch.ret = r;
4473 
4474 	if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4475 	    !kvmppc_vcpu_woken(vcpu)) {
4476 		kvmppc_set_timer(vcpu);
4477 		while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4478 			if (signal_pending(current)) {
4479 				vcpu->stat.signal_exits++;
4480 				run->exit_reason = KVM_EXIT_INTR;
4481 				vcpu->arch.ret = -EINTR;
4482 				break;
4483 			}
4484 			spin_lock(&vc->lock);
4485 			kvmppc_vcore_blocked(vc);
4486 			spin_unlock(&vc->lock);
4487 		}
4488 	}
4489 	vcpu->arch.ceded = 0;
4490 
4491 	vc->vcore_state = VCORE_INACTIVE;
4492 	trace_kvmppc_run_core(vc, 1);
4493 
4494  done:
4495 	kvmppc_remove_runnable(vc, vcpu);
4496 	trace_kvmppc_run_vcpu_exit(vcpu);
4497 
4498 	return vcpu->arch.ret;
4499 
4500  sigpend:
4501 	vcpu->stat.signal_exits++;
4502 	run->exit_reason = KVM_EXIT_INTR;
4503 	vcpu->arch.ret = -EINTR;
4504  out:
4505 	local_irq_enable();
4506 	preempt_enable();
4507 	goto done;
4508 }
4509 
4510 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4511 {
4512 	struct kvm_run *run = vcpu->run;
4513 	int r;
4514 	int srcu_idx;
4515 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
4516 	unsigned long user_tar = 0;
4517 	unsigned int user_vrsave;
4518 	struct kvm *kvm;
4519 
4520 	if (!vcpu->arch.sane) {
4521 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4522 		return -EINVAL;
4523 	}
4524 
4525 	/*
4526 	 * Don't allow entry with a suspended transaction, because
4527 	 * the guest entry/exit code will lose it.
4528 	 * If the guest has TM enabled, save away their TM-related SPRs
4529 	 * (they will get restored by the TM unavailable interrupt).
4530 	 */
4531 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4532 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4533 	    (current->thread.regs->msr & MSR_TM)) {
4534 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4535 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4536 			run->fail_entry.hardware_entry_failure_reason = 0;
4537 			return -EINVAL;
4538 		}
4539 		/* Enable TM so we can read the TM SPRs */
4540 		mtmsr(mfmsr() | MSR_TM);
4541 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4542 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4543 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4544 		current->thread.regs->msr &= ~MSR_TM;
4545 	}
4546 #endif
4547 
4548 	/*
4549 	 * Force online to 1 for the sake of old userspace which doesn't
4550 	 * set it.
4551 	 */
4552 	if (!vcpu->arch.online) {
4553 		atomic_inc(&vcpu->arch.vcore->online_count);
4554 		vcpu->arch.online = 1;
4555 	}
4556 
4557 	kvmppc_core_prepare_to_enter(vcpu);
4558 
4559 	/* No need to go into the guest when all we'll do is come back out */
4560 	if (signal_pending(current)) {
4561 		run->exit_reason = KVM_EXIT_INTR;
4562 		return -EINTR;
4563 	}
4564 
4565 	kvm = vcpu->kvm;
4566 	atomic_inc(&kvm->arch.vcpus_running);
4567 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4568 	smp_mb();
4569 
4570 	flush_all_to_thread(current);
4571 
4572 	/* Save userspace EBB and other register values */
4573 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4574 		ebb_regs[0] = mfspr(SPRN_EBBHR);
4575 		ebb_regs[1] = mfspr(SPRN_EBBRR);
4576 		ebb_regs[2] = mfspr(SPRN_BESCR);
4577 		user_tar = mfspr(SPRN_TAR);
4578 	}
4579 	user_vrsave = mfspr(SPRN_VRSAVE);
4580 
4581 	vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4582 	vcpu->arch.pgdir = kvm->mm->pgd;
4583 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4584 
4585 	do {
4586 		if (cpu_has_feature(CPU_FTR_ARCH_300))
4587 			r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4588 						  vcpu->arch.vcore->lpcr);
4589 		else
4590 			r = kvmppc_run_vcpu(vcpu);
4591 
4592 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4593 			if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) {
4594 				/*
4595 				 * These should have been caught reflected
4596 				 * into the guest by now. Final sanity check:
4597 				 * don't allow userspace to execute hcalls in
4598 				 * the hypervisor.
4599 				 */
4600 				r = RESUME_GUEST;
4601 				continue;
4602 			}
4603 			trace_kvm_hcall_enter(vcpu);
4604 			r = kvmppc_pseries_do_hcall(vcpu);
4605 			trace_kvm_hcall_exit(vcpu, r);
4606 			kvmppc_core_prepare_to_enter(vcpu);
4607 		} else if (r == RESUME_PAGE_FAULT) {
4608 			srcu_idx = srcu_read_lock(&kvm->srcu);
4609 			r = kvmppc_book3s_hv_page_fault(vcpu,
4610 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4611 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4612 		} else if (r == RESUME_PASSTHROUGH) {
4613 			if (WARN_ON(xics_on_xive()))
4614 				r = H_SUCCESS;
4615 			else
4616 				r = kvmppc_xics_rm_complete(vcpu, 0);
4617 		}
4618 	} while (is_kvmppc_resume_guest(r));
4619 
4620 	/* Restore userspace EBB and other register values */
4621 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4622 		mtspr(SPRN_EBBHR, ebb_regs[0]);
4623 		mtspr(SPRN_EBBRR, ebb_regs[1]);
4624 		mtspr(SPRN_BESCR, ebb_regs[2]);
4625 		mtspr(SPRN_TAR, user_tar);
4626 	}
4627 	mtspr(SPRN_VRSAVE, user_vrsave);
4628 
4629 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4630 	atomic_dec(&kvm->arch.vcpus_running);
4631 
4632 	srr_regs_clobbered();
4633 
4634 	return r;
4635 }
4636 
4637 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4638 				     int shift, int sllp)
4639 {
4640 	(*sps)->page_shift = shift;
4641 	(*sps)->slb_enc = sllp;
4642 	(*sps)->enc[0].page_shift = shift;
4643 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4644 	/*
4645 	 * Add 16MB MPSS support (may get filtered out by userspace)
4646 	 */
4647 	if (shift != 24) {
4648 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4649 		if (penc != -1) {
4650 			(*sps)->enc[1].page_shift = 24;
4651 			(*sps)->enc[1].pte_enc = penc;
4652 		}
4653 	}
4654 	(*sps)++;
4655 }
4656 
4657 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4658 					 struct kvm_ppc_smmu_info *info)
4659 {
4660 	struct kvm_ppc_one_seg_page_size *sps;
4661 
4662 	/*
4663 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4664 	 * POWER7 doesn't support keys for instruction accesses,
4665 	 * POWER8 and POWER9 do.
4666 	 */
4667 	info->data_keys = 32;
4668 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4669 
4670 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4671 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4672 	info->slb_size = 32;
4673 
4674 	/* We only support these sizes for now, and no muti-size segments */
4675 	sps = &info->sps[0];
4676 	kvmppc_add_seg_page_size(&sps, 12, 0);
4677 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4678 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4679 
4680 	/* If running as a nested hypervisor, we don't support HPT guests */
4681 	if (kvmhv_on_pseries())
4682 		info->flags |= KVM_PPC_NO_HASH;
4683 
4684 	return 0;
4685 }
4686 
4687 /*
4688  * Get (and clear) the dirty memory log for a memory slot.
4689  */
4690 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4691 					 struct kvm_dirty_log *log)
4692 {
4693 	struct kvm_memslots *slots;
4694 	struct kvm_memory_slot *memslot;
4695 	int i, r;
4696 	unsigned long n;
4697 	unsigned long *buf, *p;
4698 	struct kvm_vcpu *vcpu;
4699 
4700 	mutex_lock(&kvm->slots_lock);
4701 
4702 	r = -EINVAL;
4703 	if (log->slot >= KVM_USER_MEM_SLOTS)
4704 		goto out;
4705 
4706 	slots = kvm_memslots(kvm);
4707 	memslot = id_to_memslot(slots, log->slot);
4708 	r = -ENOENT;
4709 	if (!memslot || !memslot->dirty_bitmap)
4710 		goto out;
4711 
4712 	/*
4713 	 * Use second half of bitmap area because both HPT and radix
4714 	 * accumulate bits in the first half.
4715 	 */
4716 	n = kvm_dirty_bitmap_bytes(memslot);
4717 	buf = memslot->dirty_bitmap + n / sizeof(long);
4718 	memset(buf, 0, n);
4719 
4720 	if (kvm_is_radix(kvm))
4721 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4722 	else
4723 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4724 	if (r)
4725 		goto out;
4726 
4727 	/*
4728 	 * We accumulate dirty bits in the first half of the
4729 	 * memslot's dirty_bitmap area, for when pages are paged
4730 	 * out or modified by the host directly.  Pick up these
4731 	 * bits and add them to the map.
4732 	 */
4733 	p = memslot->dirty_bitmap;
4734 	for (i = 0; i < n / sizeof(long); ++i)
4735 		buf[i] |= xchg(&p[i], 0);
4736 
4737 	/* Harvest dirty bits from VPA and DTL updates */
4738 	/* Note: we never modify the SLB shadow buffer areas */
4739 	kvm_for_each_vcpu(i, vcpu, kvm) {
4740 		spin_lock(&vcpu->arch.vpa_update_lock);
4741 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4742 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4743 		spin_unlock(&vcpu->arch.vpa_update_lock);
4744 	}
4745 
4746 	r = -EFAULT;
4747 	if (copy_to_user(log->dirty_bitmap, buf, n))
4748 		goto out;
4749 
4750 	r = 0;
4751 out:
4752 	mutex_unlock(&kvm->slots_lock);
4753 	return r;
4754 }
4755 
4756 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4757 {
4758 	vfree(slot->arch.rmap);
4759 	slot->arch.rmap = NULL;
4760 }
4761 
4762 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4763 					struct kvm_memory_slot *slot,
4764 					const struct kvm_userspace_memory_region *mem,
4765 					enum kvm_mr_change change)
4766 {
4767 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4768 
4769 	if (change == KVM_MR_CREATE) {
4770 		slot->arch.rmap = vzalloc(array_size(npages,
4771 					  sizeof(*slot->arch.rmap)));
4772 		if (!slot->arch.rmap)
4773 			return -ENOMEM;
4774 	}
4775 
4776 	return 0;
4777 }
4778 
4779 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4780 				const struct kvm_userspace_memory_region *mem,
4781 				const struct kvm_memory_slot *old,
4782 				const struct kvm_memory_slot *new,
4783 				enum kvm_mr_change change)
4784 {
4785 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4786 
4787 	/*
4788 	 * If we are making a new memslot, it might make
4789 	 * some address that was previously cached as emulated
4790 	 * MMIO be no longer emulated MMIO, so invalidate
4791 	 * all the caches of emulated MMIO translations.
4792 	 */
4793 	if (npages)
4794 		atomic64_inc(&kvm->arch.mmio_update);
4795 
4796 	/*
4797 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4798 	 * have already called kvm_arch_flush_shadow_memslot() to
4799 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
4800 	 * previous mappings.  So the only case to handle is
4801 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4802 	 * has been changed.
4803 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4804 	 * to get rid of any THP PTEs in the partition-scoped page tables
4805 	 * so we can track dirtiness at the page level; we flush when
4806 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4807 	 * using THP PTEs.
4808 	 */
4809 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4810 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4811 		kvmppc_radix_flush_memslot(kvm, old);
4812 	/*
4813 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4814 	 */
4815 	if (!kvm->arch.secure_guest)
4816 		return;
4817 
4818 	switch (change) {
4819 	case KVM_MR_CREATE:
4820 		/*
4821 		 * @TODO kvmppc_uvmem_memslot_create() can fail and
4822 		 * return error. Fix this.
4823 		 */
4824 		kvmppc_uvmem_memslot_create(kvm, new);
4825 		break;
4826 	case KVM_MR_DELETE:
4827 		kvmppc_uvmem_memslot_delete(kvm, old);
4828 		break;
4829 	default:
4830 		/* TODO: Handle KVM_MR_MOVE */
4831 		break;
4832 	}
4833 }
4834 
4835 /*
4836  * Update LPCR values in kvm->arch and in vcores.
4837  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4838  * of kvm->arch.lpcr update).
4839  */
4840 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4841 {
4842 	long int i;
4843 	u32 cores_done = 0;
4844 
4845 	if ((kvm->arch.lpcr & mask) == lpcr)
4846 		return;
4847 
4848 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4849 
4850 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4851 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4852 		if (!vc)
4853 			continue;
4854 
4855 		spin_lock(&vc->lock);
4856 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4857 		verify_lpcr(kvm, vc->lpcr);
4858 		spin_unlock(&vc->lock);
4859 		if (++cores_done >= kvm->arch.online_vcores)
4860 			break;
4861 	}
4862 }
4863 
4864 void kvmppc_setup_partition_table(struct kvm *kvm)
4865 {
4866 	unsigned long dw0, dw1;
4867 
4868 	if (!kvm_is_radix(kvm)) {
4869 		/* PS field - page size for VRMA */
4870 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4871 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4872 		/* HTABSIZE and HTABORG fields */
4873 		dw0 |= kvm->arch.sdr1;
4874 
4875 		/* Second dword as set by userspace */
4876 		dw1 = kvm->arch.process_table;
4877 	} else {
4878 		dw0 = PATB_HR | radix__get_tree_size() |
4879 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4880 		dw1 = PATB_GR | kvm->arch.process_table;
4881 	}
4882 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4883 }
4884 
4885 /*
4886  * Set up HPT (hashed page table) and RMA (real-mode area).
4887  * Must be called with kvm->arch.mmu_setup_lock held.
4888  */
4889 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4890 {
4891 	int err = 0;
4892 	struct kvm *kvm = vcpu->kvm;
4893 	unsigned long hva;
4894 	struct kvm_memory_slot *memslot;
4895 	struct vm_area_struct *vma;
4896 	unsigned long lpcr = 0, senc;
4897 	unsigned long psize, porder;
4898 	int srcu_idx;
4899 
4900 	/* Allocate hashed page table (if not done already) and reset it */
4901 	if (!kvm->arch.hpt.virt) {
4902 		int order = KVM_DEFAULT_HPT_ORDER;
4903 		struct kvm_hpt_info info;
4904 
4905 		err = kvmppc_allocate_hpt(&info, order);
4906 		/* If we get here, it means userspace didn't specify a
4907 		 * size explicitly.  So, try successively smaller
4908 		 * sizes if the default failed. */
4909 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4910 			err  = kvmppc_allocate_hpt(&info, order);
4911 
4912 		if (err < 0) {
4913 			pr_err("KVM: Couldn't alloc HPT\n");
4914 			goto out;
4915 		}
4916 
4917 		kvmppc_set_hpt(kvm, &info);
4918 	}
4919 
4920 	/* Look up the memslot for guest physical address 0 */
4921 	srcu_idx = srcu_read_lock(&kvm->srcu);
4922 	memslot = gfn_to_memslot(kvm, 0);
4923 
4924 	/* We must have some memory at 0 by now */
4925 	err = -EINVAL;
4926 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4927 		goto out_srcu;
4928 
4929 	/* Look up the VMA for the start of this memory slot */
4930 	hva = memslot->userspace_addr;
4931 	mmap_read_lock(kvm->mm);
4932 	vma = vma_lookup(kvm->mm, hva);
4933 	if (!vma || (vma->vm_flags & VM_IO))
4934 		goto up_out;
4935 
4936 	psize = vma_kernel_pagesize(vma);
4937 
4938 	mmap_read_unlock(kvm->mm);
4939 
4940 	/* We can handle 4k, 64k or 16M pages in the VRMA */
4941 	if (psize >= 0x1000000)
4942 		psize = 0x1000000;
4943 	else if (psize >= 0x10000)
4944 		psize = 0x10000;
4945 	else
4946 		psize = 0x1000;
4947 	porder = __ilog2(psize);
4948 
4949 	senc = slb_pgsize_encoding(psize);
4950 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4951 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4952 	/* Create HPTEs in the hash page table for the VRMA */
4953 	kvmppc_map_vrma(vcpu, memslot, porder);
4954 
4955 	/* Update VRMASD field in the LPCR */
4956 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4957 		/* the -4 is to account for senc values starting at 0x10 */
4958 		lpcr = senc << (LPCR_VRMASD_SH - 4);
4959 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4960 	}
4961 
4962 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4963 	smp_wmb();
4964 	err = 0;
4965  out_srcu:
4966 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4967  out:
4968 	return err;
4969 
4970  up_out:
4971 	mmap_read_unlock(kvm->mm);
4972 	goto out_srcu;
4973 }
4974 
4975 /*
4976  * Must be called with kvm->arch.mmu_setup_lock held and
4977  * mmu_ready = 0 and no vcpus running.
4978  */
4979 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4980 {
4981 	if (nesting_enabled(kvm))
4982 		kvmhv_release_all_nested(kvm);
4983 	kvmppc_rmap_reset(kvm);
4984 	kvm->arch.process_table = 0;
4985 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
4986 	spin_lock(&kvm->mmu_lock);
4987 	kvm->arch.radix = 0;
4988 	spin_unlock(&kvm->mmu_lock);
4989 	kvmppc_free_radix(kvm);
4990 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
4991 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4992 	return 0;
4993 }
4994 
4995 /*
4996  * Must be called with kvm->arch.mmu_setup_lock held and
4997  * mmu_ready = 0 and no vcpus running.
4998  */
4999 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5000 {
5001 	int err;
5002 
5003 	err = kvmppc_init_vm_radix(kvm);
5004 	if (err)
5005 		return err;
5006 	kvmppc_rmap_reset(kvm);
5007 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5008 	spin_lock(&kvm->mmu_lock);
5009 	kvm->arch.radix = 1;
5010 	spin_unlock(&kvm->mmu_lock);
5011 	kvmppc_free_hpt(&kvm->arch.hpt);
5012 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
5013 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
5014 	return 0;
5015 }
5016 
5017 #ifdef CONFIG_KVM_XICS
5018 /*
5019  * Allocate a per-core structure for managing state about which cores are
5020  * running in the host versus the guest and for exchanging data between
5021  * real mode KVM and CPU running in the host.
5022  * This is only done for the first VM.
5023  * The allocated structure stays even if all VMs have stopped.
5024  * It is only freed when the kvm-hv module is unloaded.
5025  * It's OK for this routine to fail, we just don't support host
5026  * core operations like redirecting H_IPI wakeups.
5027  */
5028 void kvmppc_alloc_host_rm_ops(void)
5029 {
5030 	struct kvmppc_host_rm_ops *ops;
5031 	unsigned long l_ops;
5032 	int cpu, core;
5033 	int size;
5034 
5035 	/* Not the first time here ? */
5036 	if (kvmppc_host_rm_ops_hv != NULL)
5037 		return;
5038 
5039 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5040 	if (!ops)
5041 		return;
5042 
5043 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5044 	ops->rm_core = kzalloc(size, GFP_KERNEL);
5045 
5046 	if (!ops->rm_core) {
5047 		kfree(ops);
5048 		return;
5049 	}
5050 
5051 	cpus_read_lock();
5052 
5053 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5054 		if (!cpu_online(cpu))
5055 			continue;
5056 
5057 		core = cpu >> threads_shift;
5058 		ops->rm_core[core].rm_state.in_host = 1;
5059 	}
5060 
5061 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5062 
5063 	/*
5064 	 * Make the contents of the kvmppc_host_rm_ops structure visible
5065 	 * to other CPUs before we assign it to the global variable.
5066 	 * Do an atomic assignment (no locks used here), but if someone
5067 	 * beats us to it, just free our copy and return.
5068 	 */
5069 	smp_wmb();
5070 	l_ops = (unsigned long) ops;
5071 
5072 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5073 		cpus_read_unlock();
5074 		kfree(ops->rm_core);
5075 		kfree(ops);
5076 		return;
5077 	}
5078 
5079 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5080 					     "ppc/kvm_book3s:prepare",
5081 					     kvmppc_set_host_core,
5082 					     kvmppc_clear_host_core);
5083 	cpus_read_unlock();
5084 }
5085 
5086 void kvmppc_free_host_rm_ops(void)
5087 {
5088 	if (kvmppc_host_rm_ops_hv) {
5089 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5090 		kfree(kvmppc_host_rm_ops_hv->rm_core);
5091 		kfree(kvmppc_host_rm_ops_hv);
5092 		kvmppc_host_rm_ops_hv = NULL;
5093 	}
5094 }
5095 #endif
5096 
5097 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5098 {
5099 	unsigned long lpcr, lpid;
5100 	char buf[32];
5101 	int ret;
5102 
5103 	mutex_init(&kvm->arch.uvmem_lock);
5104 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5105 	mutex_init(&kvm->arch.mmu_setup_lock);
5106 
5107 	/* Allocate the guest's logical partition ID */
5108 
5109 	lpid = kvmppc_alloc_lpid();
5110 	if ((long)lpid < 0)
5111 		return -ENOMEM;
5112 	kvm->arch.lpid = lpid;
5113 
5114 	kvmppc_alloc_host_rm_ops();
5115 
5116 	kvmhv_vm_nested_init(kvm);
5117 
5118 	/*
5119 	 * Since we don't flush the TLB when tearing down a VM,
5120 	 * and this lpid might have previously been used,
5121 	 * make sure we flush on each core before running the new VM.
5122 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5123 	 * does this flush for us.
5124 	 */
5125 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5126 		cpumask_setall(&kvm->arch.need_tlb_flush);
5127 
5128 	/* Start out with the default set of hcalls enabled */
5129 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5130 	       sizeof(kvm->arch.enabled_hcalls));
5131 
5132 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5133 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5134 
5135 	/* Init LPCR for virtual RMA mode */
5136 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
5137 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
5138 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5139 		lpcr &= LPCR_PECE | LPCR_LPES;
5140 	} else {
5141 		lpcr = 0;
5142 	}
5143 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5144 		LPCR_VPM0 | LPCR_VPM1;
5145 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5146 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5147 	/* On POWER8 turn on online bit to enable PURR/SPURR */
5148 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
5149 		lpcr |= LPCR_ONL;
5150 	/*
5151 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5152 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
5153 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5154 	 * be unnecessary but better safe than sorry in case we re-enable
5155 	 * EE in HV mode with this LPCR still set)
5156 	 */
5157 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5158 		lpcr &= ~LPCR_VPM0;
5159 		lpcr |= LPCR_HVICE | LPCR_HEIC;
5160 
5161 		/*
5162 		 * If xive is enabled, we route 0x500 interrupts directly
5163 		 * to the guest.
5164 		 */
5165 		if (xics_on_xive())
5166 			lpcr |= LPCR_LPES;
5167 	}
5168 
5169 	/*
5170 	 * If the host uses radix, the guest starts out as radix.
5171 	 */
5172 	if (radix_enabled()) {
5173 		kvm->arch.radix = 1;
5174 		kvm->arch.mmu_ready = 1;
5175 		lpcr &= ~LPCR_VPM1;
5176 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5177 		ret = kvmppc_init_vm_radix(kvm);
5178 		if (ret) {
5179 			kvmppc_free_lpid(kvm->arch.lpid);
5180 			return ret;
5181 		}
5182 		kvmppc_setup_partition_table(kvm);
5183 	}
5184 
5185 	verify_lpcr(kvm, lpcr);
5186 	kvm->arch.lpcr = lpcr;
5187 
5188 	/* Initialization for future HPT resizes */
5189 	kvm->arch.resize_hpt = NULL;
5190 
5191 	/*
5192 	 * Work out how many sets the TLB has, for the use of
5193 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5194 	 */
5195 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5196 		/*
5197 		 * P10 will flush all the congruence class with a single tlbiel
5198 		 */
5199 		kvm->arch.tlb_sets = 1;
5200 	} else if (radix_enabled())
5201 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
5202 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
5203 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
5204 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5205 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
5206 	else
5207 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
5208 
5209 	/*
5210 	 * Track that we now have a HV mode VM active. This blocks secondary
5211 	 * CPU threads from coming online.
5212 	 */
5213 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5214 		kvm_hv_vm_activated();
5215 
5216 	/*
5217 	 * Initialize smt_mode depending on processor.
5218 	 * POWER8 and earlier have to use "strict" threading, where
5219 	 * all vCPUs in a vcore have to run on the same (sub)core,
5220 	 * whereas on POWER9 the threads can each run a different
5221 	 * guest.
5222 	 */
5223 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5224 		kvm->arch.smt_mode = threads_per_subcore;
5225 	else
5226 		kvm->arch.smt_mode = 1;
5227 	kvm->arch.emul_smt_mode = 1;
5228 
5229 	/*
5230 	 * Create a debugfs directory for the VM
5231 	 */
5232 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
5233 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5234 	kvmppc_mmu_debugfs_init(kvm);
5235 	if (radix_enabled())
5236 		kvmhv_radix_debugfs_init(kvm);
5237 
5238 	return 0;
5239 }
5240 
5241 static void kvmppc_free_vcores(struct kvm *kvm)
5242 {
5243 	long int i;
5244 
5245 	for (i = 0; i < KVM_MAX_VCORES; ++i)
5246 		kfree(kvm->arch.vcores[i]);
5247 	kvm->arch.online_vcores = 0;
5248 }
5249 
5250 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5251 {
5252 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
5253 
5254 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5255 		kvm_hv_vm_deactivated();
5256 
5257 	kvmppc_free_vcores(kvm);
5258 
5259 
5260 	if (kvm_is_radix(kvm))
5261 		kvmppc_free_radix(kvm);
5262 	else
5263 		kvmppc_free_hpt(&kvm->arch.hpt);
5264 
5265 	/* Perform global invalidation and return lpid to the pool */
5266 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5267 		if (nesting_enabled(kvm))
5268 			kvmhv_release_all_nested(kvm);
5269 		kvm->arch.process_table = 0;
5270 		if (kvm->arch.secure_guest)
5271 			uv_svm_terminate(kvm->arch.lpid);
5272 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5273 	}
5274 
5275 	kvmppc_free_lpid(kvm->arch.lpid);
5276 
5277 	kvmppc_free_pimap(kvm);
5278 }
5279 
5280 /* We don't need to emulate any privileged instructions or dcbz */
5281 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5282 				     unsigned int inst, int *advance)
5283 {
5284 	return EMULATE_FAIL;
5285 }
5286 
5287 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5288 					ulong spr_val)
5289 {
5290 	return EMULATE_FAIL;
5291 }
5292 
5293 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5294 					ulong *spr_val)
5295 {
5296 	return EMULATE_FAIL;
5297 }
5298 
5299 static int kvmppc_core_check_processor_compat_hv(void)
5300 {
5301 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5302 	    cpu_has_feature(CPU_FTR_ARCH_206))
5303 		return 0;
5304 
5305 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5306 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5307 		return 0;
5308 
5309 	return -EIO;
5310 }
5311 
5312 #ifdef CONFIG_KVM_XICS
5313 
5314 void kvmppc_free_pimap(struct kvm *kvm)
5315 {
5316 	kfree(kvm->arch.pimap);
5317 }
5318 
5319 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5320 {
5321 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5322 }
5323 
5324 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5325 {
5326 	struct irq_desc *desc;
5327 	struct kvmppc_irq_map *irq_map;
5328 	struct kvmppc_passthru_irqmap *pimap;
5329 	struct irq_chip *chip;
5330 	int i, rc = 0;
5331 
5332 	if (!kvm_irq_bypass)
5333 		return 1;
5334 
5335 	desc = irq_to_desc(host_irq);
5336 	if (!desc)
5337 		return -EIO;
5338 
5339 	mutex_lock(&kvm->lock);
5340 
5341 	pimap = kvm->arch.pimap;
5342 	if (pimap == NULL) {
5343 		/* First call, allocate structure to hold IRQ map */
5344 		pimap = kvmppc_alloc_pimap();
5345 		if (pimap == NULL) {
5346 			mutex_unlock(&kvm->lock);
5347 			return -ENOMEM;
5348 		}
5349 		kvm->arch.pimap = pimap;
5350 	}
5351 
5352 	/*
5353 	 * For now, we only support interrupts for which the EOI operation
5354 	 * is an OPAL call followed by a write to XIRR, since that's
5355 	 * what our real-mode EOI code does, or a XIVE interrupt
5356 	 */
5357 	chip = irq_data_get_irq_chip(&desc->irq_data);
5358 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5359 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5360 			host_irq, guest_gsi);
5361 		mutex_unlock(&kvm->lock);
5362 		return -ENOENT;
5363 	}
5364 
5365 	/*
5366 	 * See if we already have an entry for this guest IRQ number.
5367 	 * If it's mapped to a hardware IRQ number, that's an error,
5368 	 * otherwise re-use this entry.
5369 	 */
5370 	for (i = 0; i < pimap->n_mapped; i++) {
5371 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5372 			if (pimap->mapped[i].r_hwirq) {
5373 				mutex_unlock(&kvm->lock);
5374 				return -EINVAL;
5375 			}
5376 			break;
5377 		}
5378 	}
5379 
5380 	if (i == KVMPPC_PIRQ_MAPPED) {
5381 		mutex_unlock(&kvm->lock);
5382 		return -EAGAIN;		/* table is full */
5383 	}
5384 
5385 	irq_map = &pimap->mapped[i];
5386 
5387 	irq_map->v_hwirq = guest_gsi;
5388 	irq_map->desc = desc;
5389 
5390 	/*
5391 	 * Order the above two stores before the next to serialize with
5392 	 * the KVM real mode handler.
5393 	 */
5394 	smp_wmb();
5395 	irq_map->r_hwirq = desc->irq_data.hwirq;
5396 
5397 	if (i == pimap->n_mapped)
5398 		pimap->n_mapped++;
5399 
5400 	if (xics_on_xive())
5401 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5402 	else
5403 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5404 	if (rc)
5405 		irq_map->r_hwirq = 0;
5406 
5407 	mutex_unlock(&kvm->lock);
5408 
5409 	return 0;
5410 }
5411 
5412 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5413 {
5414 	struct irq_desc *desc;
5415 	struct kvmppc_passthru_irqmap *pimap;
5416 	int i, rc = 0;
5417 
5418 	if (!kvm_irq_bypass)
5419 		return 0;
5420 
5421 	desc = irq_to_desc(host_irq);
5422 	if (!desc)
5423 		return -EIO;
5424 
5425 	mutex_lock(&kvm->lock);
5426 	if (!kvm->arch.pimap)
5427 		goto unlock;
5428 
5429 	pimap = kvm->arch.pimap;
5430 
5431 	for (i = 0; i < pimap->n_mapped; i++) {
5432 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5433 			break;
5434 	}
5435 
5436 	if (i == pimap->n_mapped) {
5437 		mutex_unlock(&kvm->lock);
5438 		return -ENODEV;
5439 	}
5440 
5441 	if (xics_on_xive())
5442 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5443 	else
5444 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5445 
5446 	/* invalidate the entry (what do do on error from the above ?) */
5447 	pimap->mapped[i].r_hwirq = 0;
5448 
5449 	/*
5450 	 * We don't free this structure even when the count goes to
5451 	 * zero. The structure is freed when we destroy the VM.
5452 	 */
5453  unlock:
5454 	mutex_unlock(&kvm->lock);
5455 	return rc;
5456 }
5457 
5458 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5459 					     struct irq_bypass_producer *prod)
5460 {
5461 	int ret = 0;
5462 	struct kvm_kernel_irqfd *irqfd =
5463 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5464 
5465 	irqfd->producer = prod;
5466 
5467 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5468 	if (ret)
5469 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5470 			prod->irq, irqfd->gsi, ret);
5471 
5472 	return ret;
5473 }
5474 
5475 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5476 					      struct irq_bypass_producer *prod)
5477 {
5478 	int ret;
5479 	struct kvm_kernel_irqfd *irqfd =
5480 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5481 
5482 	irqfd->producer = NULL;
5483 
5484 	/*
5485 	 * When producer of consumer is unregistered, we change back to
5486 	 * default external interrupt handling mode - KVM real mode
5487 	 * will switch back to host.
5488 	 */
5489 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5490 	if (ret)
5491 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5492 			prod->irq, irqfd->gsi, ret);
5493 }
5494 #endif
5495 
5496 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5497 				 unsigned int ioctl, unsigned long arg)
5498 {
5499 	struct kvm *kvm __maybe_unused = filp->private_data;
5500 	void __user *argp = (void __user *)arg;
5501 	long r;
5502 
5503 	switch (ioctl) {
5504 
5505 	case KVM_PPC_ALLOCATE_HTAB: {
5506 		u32 htab_order;
5507 
5508 		/* If we're a nested hypervisor, we currently only support radix */
5509 		if (kvmhv_on_pseries()) {
5510 			r = -EOPNOTSUPP;
5511 			break;
5512 		}
5513 
5514 		r = -EFAULT;
5515 		if (get_user(htab_order, (u32 __user *)argp))
5516 			break;
5517 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5518 		if (r)
5519 			break;
5520 		r = 0;
5521 		break;
5522 	}
5523 
5524 	case KVM_PPC_GET_HTAB_FD: {
5525 		struct kvm_get_htab_fd ghf;
5526 
5527 		r = -EFAULT;
5528 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5529 			break;
5530 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5531 		break;
5532 	}
5533 
5534 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5535 		struct kvm_ppc_resize_hpt rhpt;
5536 
5537 		r = -EFAULT;
5538 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5539 			break;
5540 
5541 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5542 		break;
5543 	}
5544 
5545 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5546 		struct kvm_ppc_resize_hpt rhpt;
5547 
5548 		r = -EFAULT;
5549 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5550 			break;
5551 
5552 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5553 		break;
5554 	}
5555 
5556 	default:
5557 		r = -ENOTTY;
5558 	}
5559 
5560 	return r;
5561 }
5562 
5563 /*
5564  * List of hcall numbers to enable by default.
5565  * For compatibility with old userspace, we enable by default
5566  * all hcalls that were implemented before the hcall-enabling
5567  * facility was added.  Note this list should not include H_RTAS.
5568  */
5569 static unsigned int default_hcall_list[] = {
5570 	H_REMOVE,
5571 	H_ENTER,
5572 	H_READ,
5573 	H_PROTECT,
5574 	H_BULK_REMOVE,
5575 #ifdef CONFIG_SPAPR_TCE_IOMMU
5576 	H_GET_TCE,
5577 	H_PUT_TCE,
5578 #endif
5579 	H_SET_DABR,
5580 	H_SET_XDABR,
5581 	H_CEDE,
5582 	H_PROD,
5583 	H_CONFER,
5584 	H_REGISTER_VPA,
5585 #ifdef CONFIG_KVM_XICS
5586 	H_EOI,
5587 	H_CPPR,
5588 	H_IPI,
5589 	H_IPOLL,
5590 	H_XIRR,
5591 	H_XIRR_X,
5592 #endif
5593 	0
5594 };
5595 
5596 static void init_default_hcalls(void)
5597 {
5598 	int i;
5599 	unsigned int hcall;
5600 
5601 	for (i = 0; default_hcall_list[i]; ++i) {
5602 		hcall = default_hcall_list[i];
5603 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5604 		__set_bit(hcall / 4, default_enabled_hcalls);
5605 	}
5606 }
5607 
5608 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5609 {
5610 	unsigned long lpcr;
5611 	int radix;
5612 	int err;
5613 
5614 	/* If not on a POWER9, reject it */
5615 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5616 		return -ENODEV;
5617 
5618 	/* If any unknown flags set, reject it */
5619 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5620 		return -EINVAL;
5621 
5622 	/* GR (guest radix) bit in process_table field must match */
5623 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5624 	if (!!(cfg->process_table & PATB_GR) != radix)
5625 		return -EINVAL;
5626 
5627 	/* Process table size field must be reasonable, i.e. <= 24 */
5628 	if ((cfg->process_table & PRTS_MASK) > 24)
5629 		return -EINVAL;
5630 
5631 	/* We can change a guest to/from radix now, if the host is radix */
5632 	if (radix && !radix_enabled())
5633 		return -EINVAL;
5634 
5635 	/* If we're a nested hypervisor, we currently only support radix */
5636 	if (kvmhv_on_pseries() && !radix)
5637 		return -EINVAL;
5638 
5639 	mutex_lock(&kvm->arch.mmu_setup_lock);
5640 	if (radix != kvm_is_radix(kvm)) {
5641 		if (kvm->arch.mmu_ready) {
5642 			kvm->arch.mmu_ready = 0;
5643 			/* order mmu_ready vs. vcpus_running */
5644 			smp_mb();
5645 			if (atomic_read(&kvm->arch.vcpus_running)) {
5646 				kvm->arch.mmu_ready = 1;
5647 				err = -EBUSY;
5648 				goto out_unlock;
5649 			}
5650 		}
5651 		if (radix)
5652 			err = kvmppc_switch_mmu_to_radix(kvm);
5653 		else
5654 			err = kvmppc_switch_mmu_to_hpt(kvm);
5655 		if (err)
5656 			goto out_unlock;
5657 	}
5658 
5659 	kvm->arch.process_table = cfg->process_table;
5660 	kvmppc_setup_partition_table(kvm);
5661 
5662 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5663 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5664 	err = 0;
5665 
5666  out_unlock:
5667 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5668 	return err;
5669 }
5670 
5671 static int kvmhv_enable_nested(struct kvm *kvm)
5672 {
5673 	if (!nested)
5674 		return -EPERM;
5675 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5676 		return -ENODEV;
5677 	if (!radix_enabled())
5678 		return -ENODEV;
5679 
5680 	/* kvm == NULL means the caller is testing if the capability exists */
5681 	if (kvm)
5682 		kvm->arch.nested_enable = true;
5683 	return 0;
5684 }
5685 
5686 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5687 				 int size)
5688 {
5689 	int rc = -EINVAL;
5690 
5691 	if (kvmhv_vcpu_is_radix(vcpu)) {
5692 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5693 
5694 		if (rc > 0)
5695 			rc = -EINVAL;
5696 	}
5697 
5698 	/* For now quadrants are the only way to access nested guest memory */
5699 	if (rc && vcpu->arch.nested)
5700 		rc = -EAGAIN;
5701 
5702 	return rc;
5703 }
5704 
5705 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5706 				int size)
5707 {
5708 	int rc = -EINVAL;
5709 
5710 	if (kvmhv_vcpu_is_radix(vcpu)) {
5711 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5712 
5713 		if (rc > 0)
5714 			rc = -EINVAL;
5715 	}
5716 
5717 	/* For now quadrants are the only way to access nested guest memory */
5718 	if (rc && vcpu->arch.nested)
5719 		rc = -EAGAIN;
5720 
5721 	return rc;
5722 }
5723 
5724 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5725 {
5726 	unpin_vpa(kvm, vpa);
5727 	vpa->gpa = 0;
5728 	vpa->pinned_addr = NULL;
5729 	vpa->dirty = false;
5730 	vpa->update_pending = 0;
5731 }
5732 
5733 /*
5734  * Enable a guest to become a secure VM, or test whether
5735  * that could be enabled.
5736  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5737  * tested (kvm == NULL) or enabled (kvm != NULL).
5738  */
5739 static int kvmhv_enable_svm(struct kvm *kvm)
5740 {
5741 	if (!kvmppc_uvmem_available())
5742 		return -EINVAL;
5743 	if (kvm)
5744 		kvm->arch.svm_enabled = 1;
5745 	return 0;
5746 }
5747 
5748 /*
5749  *  IOCTL handler to turn off secure mode of guest
5750  *
5751  * - Release all device pages
5752  * - Issue ucall to terminate the guest on the UV side
5753  * - Unpin the VPA pages.
5754  * - Reinit the partition scoped page tables
5755  */
5756 static int kvmhv_svm_off(struct kvm *kvm)
5757 {
5758 	struct kvm_vcpu *vcpu;
5759 	int mmu_was_ready;
5760 	int srcu_idx;
5761 	int ret = 0;
5762 	int i;
5763 
5764 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5765 		return ret;
5766 
5767 	mutex_lock(&kvm->arch.mmu_setup_lock);
5768 	mmu_was_ready = kvm->arch.mmu_ready;
5769 	if (kvm->arch.mmu_ready) {
5770 		kvm->arch.mmu_ready = 0;
5771 		/* order mmu_ready vs. vcpus_running */
5772 		smp_mb();
5773 		if (atomic_read(&kvm->arch.vcpus_running)) {
5774 			kvm->arch.mmu_ready = 1;
5775 			ret = -EBUSY;
5776 			goto out;
5777 		}
5778 	}
5779 
5780 	srcu_idx = srcu_read_lock(&kvm->srcu);
5781 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5782 		struct kvm_memory_slot *memslot;
5783 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5784 
5785 		if (!slots)
5786 			continue;
5787 
5788 		kvm_for_each_memslot(memslot, slots) {
5789 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
5790 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5791 		}
5792 	}
5793 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5794 
5795 	ret = uv_svm_terminate(kvm->arch.lpid);
5796 	if (ret != U_SUCCESS) {
5797 		ret = -EINVAL;
5798 		goto out;
5799 	}
5800 
5801 	/*
5802 	 * When secure guest is reset, all the guest pages are sent
5803 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
5804 	 * chance to run and unpin their VPA pages. Unpinning of all
5805 	 * VPA pages is done here explicitly so that VPA pages
5806 	 * can be migrated to the secure side.
5807 	 *
5808 	 * This is required to for the secure SMP guest to reboot
5809 	 * correctly.
5810 	 */
5811 	kvm_for_each_vcpu(i, vcpu, kvm) {
5812 		spin_lock(&vcpu->arch.vpa_update_lock);
5813 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5814 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5815 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5816 		spin_unlock(&vcpu->arch.vpa_update_lock);
5817 	}
5818 
5819 	kvmppc_setup_partition_table(kvm);
5820 	kvm->arch.secure_guest = 0;
5821 	kvm->arch.mmu_ready = mmu_was_ready;
5822 out:
5823 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5824 	return ret;
5825 }
5826 
5827 static int kvmhv_enable_dawr1(struct kvm *kvm)
5828 {
5829 	if (!cpu_has_feature(CPU_FTR_DAWR1))
5830 		return -ENODEV;
5831 
5832 	/* kvm == NULL means the caller is testing if the capability exists */
5833 	if (kvm)
5834 		kvm->arch.dawr1_enabled = true;
5835 	return 0;
5836 }
5837 
5838 static bool kvmppc_hash_v3_possible(void)
5839 {
5840 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5841 		return false;
5842 
5843 	if (!cpu_has_feature(CPU_FTR_HVMODE))
5844 		return false;
5845 
5846 	/*
5847 	 * POWER9 chips before version 2.02 can't have some threads in
5848 	 * HPT mode and some in radix mode on the same core.
5849 	 */
5850 	if (radix_enabled()) {
5851 		unsigned int pvr = mfspr(SPRN_PVR);
5852 		if ((pvr >> 16) == PVR_POWER9 &&
5853 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5854 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5855 			return false;
5856 	}
5857 
5858 	return true;
5859 }
5860 
5861 static struct kvmppc_ops kvm_ops_hv = {
5862 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5863 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5864 	.get_one_reg = kvmppc_get_one_reg_hv,
5865 	.set_one_reg = kvmppc_set_one_reg_hv,
5866 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
5867 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
5868 	.inject_interrupt = kvmppc_inject_interrupt_hv,
5869 	.set_msr     = kvmppc_set_msr_hv,
5870 	.vcpu_run    = kvmppc_vcpu_run_hv,
5871 	.vcpu_create = kvmppc_core_vcpu_create_hv,
5872 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
5873 	.check_requests = kvmppc_core_check_requests_hv,
5874 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5875 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
5876 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5877 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5878 	.unmap_gfn_range = kvm_unmap_gfn_range_hv,
5879 	.age_gfn = kvm_age_gfn_hv,
5880 	.test_age_gfn = kvm_test_age_gfn_hv,
5881 	.set_spte_gfn = kvm_set_spte_gfn_hv,
5882 	.free_memslot = kvmppc_core_free_memslot_hv,
5883 	.init_vm =  kvmppc_core_init_vm_hv,
5884 	.destroy_vm = kvmppc_core_destroy_vm_hv,
5885 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5886 	.emulate_op = kvmppc_core_emulate_op_hv,
5887 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5888 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5889 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5890 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5891 	.hcall_implemented = kvmppc_hcall_impl_hv,
5892 #ifdef CONFIG_KVM_XICS
5893 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5894 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5895 #endif
5896 	.configure_mmu = kvmhv_configure_mmu,
5897 	.get_rmmu_info = kvmhv_get_rmmu_info,
5898 	.set_smt_mode = kvmhv_set_smt_mode,
5899 	.enable_nested = kvmhv_enable_nested,
5900 	.load_from_eaddr = kvmhv_load_from_eaddr,
5901 	.store_to_eaddr = kvmhv_store_to_eaddr,
5902 	.enable_svm = kvmhv_enable_svm,
5903 	.svm_off = kvmhv_svm_off,
5904 	.enable_dawr1 = kvmhv_enable_dawr1,
5905 	.hash_v3_possible = kvmppc_hash_v3_possible,
5906 };
5907 
5908 static int kvm_init_subcore_bitmap(void)
5909 {
5910 	int i, j;
5911 	int nr_cores = cpu_nr_cores();
5912 	struct sibling_subcore_state *sibling_subcore_state;
5913 
5914 	for (i = 0; i < nr_cores; i++) {
5915 		int first_cpu = i * threads_per_core;
5916 		int node = cpu_to_node(first_cpu);
5917 
5918 		/* Ignore if it is already allocated. */
5919 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
5920 			continue;
5921 
5922 		sibling_subcore_state =
5923 			kzalloc_node(sizeof(struct sibling_subcore_state),
5924 							GFP_KERNEL, node);
5925 		if (!sibling_subcore_state)
5926 			return -ENOMEM;
5927 
5928 
5929 		for (j = 0; j < threads_per_core; j++) {
5930 			int cpu = first_cpu + j;
5931 
5932 			paca_ptrs[cpu]->sibling_subcore_state =
5933 						sibling_subcore_state;
5934 		}
5935 	}
5936 	return 0;
5937 }
5938 
5939 static int kvmppc_radix_possible(void)
5940 {
5941 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5942 }
5943 
5944 static int kvmppc_book3s_init_hv(void)
5945 {
5946 	int r;
5947 
5948 	if (!tlbie_capable) {
5949 		pr_err("KVM-HV: Host does not support TLBIE\n");
5950 		return -ENODEV;
5951 	}
5952 
5953 	/*
5954 	 * FIXME!! Do we need to check on all cpus ?
5955 	 */
5956 	r = kvmppc_core_check_processor_compat_hv();
5957 	if (r < 0)
5958 		return -ENODEV;
5959 
5960 	r = kvmhv_nested_init();
5961 	if (r)
5962 		return r;
5963 
5964 	r = kvm_init_subcore_bitmap();
5965 	if (r)
5966 		return r;
5967 
5968 	/*
5969 	 * We need a way of accessing the XICS interrupt controller,
5970 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5971 	 * indirectly, via OPAL.
5972 	 */
5973 #ifdef CONFIG_SMP
5974 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
5975 	    !local_paca->kvm_hstate.xics_phys) {
5976 		struct device_node *np;
5977 
5978 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5979 		if (!np) {
5980 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5981 			return -ENODEV;
5982 		}
5983 		/* presence of intc confirmed - node can be dropped again */
5984 		of_node_put(np);
5985 	}
5986 #endif
5987 
5988 	kvm_ops_hv.owner = THIS_MODULE;
5989 	kvmppc_hv_ops = &kvm_ops_hv;
5990 
5991 	init_default_hcalls();
5992 
5993 	init_vcore_lists();
5994 
5995 	r = kvmppc_mmu_hv_init();
5996 	if (r)
5997 		return r;
5998 
5999 	if (kvmppc_radix_possible())
6000 		r = kvmppc_radix_init();
6001 
6002 	r = kvmppc_uvmem_init();
6003 	if (r < 0)
6004 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6005 
6006 	return r;
6007 }
6008 
6009 static void kvmppc_book3s_exit_hv(void)
6010 {
6011 	kvmppc_uvmem_free();
6012 	kvmppc_free_host_rm_ops();
6013 	if (kvmppc_radix_possible())
6014 		kvmppc_radix_exit();
6015 	kvmppc_hv_ops = NULL;
6016 	kvmhv_nested_exit();
6017 }
6018 
6019 module_init(kvmppc_book3s_init_hv);
6020 module_exit(kvmppc_book3s_exit_hv);
6021 MODULE_LICENSE("GPL");
6022 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6023 MODULE_ALIAS("devname:kvm");
6024