xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision d15cb3da)
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/pmc.h>
63 #include <asm/processor.h>
64 #include <asm/cputhreads.h>
65 #include <asm/page.h>
66 #include <asm/hvcall.h>
67 #include <asm/switch_to.h>
68 #include <asm/smp.h>
69 #include <asm/dbell.h>
70 #include <asm/hmi.h>
71 #include <asm/pnv-pci.h>
72 #include <asm/mmu.h>
73 #include <asm/opal.h>
74 #include <asm/xics.h>
75 #include <asm/xive.h>
76 #include <asm/hw_breakpoint.h>
77 #include <asm/kvm_book3s_uvmem.h>
78 #include <asm/ultravisor.h>
79 #include <asm/dtl.h>
80 #include <asm/plpar_wrappers.h>
81 
82 #include "book3s.h"
83 #include "book3s_hv.h"
84 
85 #define CREATE_TRACE_POINTS
86 #include "trace_hv.h"
87 
88 /* #define EXIT_DEBUG */
89 /* #define EXIT_DEBUG_SIMPLE */
90 /* #define EXIT_DEBUG_INT */
91 
92 /* Used to indicate that a guest page fault needs to be handled */
93 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
94 /* Used to indicate that a guest passthrough interrupt needs to be handled */
95 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
96 
97 /* Used as a "null" value for timebase values */
98 #define TB_NIL	(~(u64)0)
99 
100 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
101 
102 static int dynamic_mt_modes = 6;
103 module_param(dynamic_mt_modes, int, 0644);
104 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
105 static int target_smt_mode;
106 module_param(target_smt_mode, int, 0644);
107 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
108 
109 static bool one_vm_per_core;
110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
112 
113 #ifdef CONFIG_KVM_XICS
114 static const struct kernel_param_ops module_param_ops = {
115 	.set = param_set_int,
116 	.get = param_get_int,
117 };
118 
119 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
120 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
121 
122 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
123 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
124 #endif
125 
126 /* If set, guests are allowed to create and control nested guests */
127 static bool nested = true;
128 module_param(nested, bool, S_IRUGO | S_IWUSR);
129 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
130 
131 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
132 
133 /*
134  * RWMR values for POWER8.  These control the rate at which PURR
135  * and SPURR count and should be set according to the number of
136  * online threads in the vcore being run.
137  */
138 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
139 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
140 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
141 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
142 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
143 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
144 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
145 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
146 
147 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
148 	RWMR_RPA_P8_1THREAD,
149 	RWMR_RPA_P8_1THREAD,
150 	RWMR_RPA_P8_2THREAD,
151 	RWMR_RPA_P8_3THREAD,
152 	RWMR_RPA_P8_4THREAD,
153 	RWMR_RPA_P8_5THREAD,
154 	RWMR_RPA_P8_6THREAD,
155 	RWMR_RPA_P8_7THREAD,
156 	RWMR_RPA_P8_8THREAD,
157 };
158 
159 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
160 		int *ip)
161 {
162 	int i = *ip;
163 	struct kvm_vcpu *vcpu;
164 
165 	while (++i < MAX_SMT_THREADS) {
166 		vcpu = READ_ONCE(vc->runnable_threads[i]);
167 		if (vcpu) {
168 			*ip = i;
169 			return vcpu;
170 		}
171 	}
172 	return NULL;
173 }
174 
175 /* Used to traverse the list of runnable threads for a given vcore */
176 #define for_each_runnable_thread(i, vcpu, vc) \
177 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
178 
179 static bool kvmppc_ipi_thread(int cpu)
180 {
181 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
182 
183 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
184 	if (kvmhv_on_pseries())
185 		return false;
186 
187 	/* On POWER9 we can use msgsnd to IPI any cpu */
188 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
189 		msg |= get_hard_smp_processor_id(cpu);
190 		smp_mb();
191 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
192 		return true;
193 	}
194 
195 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
196 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
197 		preempt_disable();
198 		if (cpu_first_thread_sibling(cpu) ==
199 		    cpu_first_thread_sibling(smp_processor_id())) {
200 			msg |= cpu_thread_in_core(cpu);
201 			smp_mb();
202 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
203 			preempt_enable();
204 			return true;
205 		}
206 		preempt_enable();
207 	}
208 
209 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
210 	if (cpu >= 0 && cpu < nr_cpu_ids) {
211 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
212 			xics_wake_cpu(cpu);
213 			return true;
214 		}
215 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
216 		return true;
217 	}
218 #endif
219 
220 	return false;
221 }
222 
223 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
224 {
225 	int cpu;
226 	struct rcuwait *waitp;
227 
228 	waitp = kvm_arch_vcpu_get_wait(vcpu);
229 	if (rcuwait_wake_up(waitp))
230 		++vcpu->stat.generic.halt_wakeup;
231 
232 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
233 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
234 		return;
235 
236 	/* CPU points to the first thread of the core */
237 	cpu = vcpu->cpu;
238 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
239 		smp_send_reschedule(cpu);
240 }
241 
242 /*
243  * We use the vcpu_load/put functions to measure stolen time.
244  * Stolen time is counted as time when either the vcpu is able to
245  * run as part of a virtual core, but the task running the vcore
246  * is preempted or sleeping, or when the vcpu needs something done
247  * in the kernel by the task running the vcpu, but that task is
248  * preempted or sleeping.  Those two things have to be counted
249  * separately, since one of the vcpu tasks will take on the job
250  * of running the core, and the other vcpu tasks in the vcore will
251  * sleep waiting for it to do that, but that sleep shouldn't count
252  * as stolen time.
253  *
254  * Hence we accumulate stolen time when the vcpu can run as part of
255  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
256  * needs its task to do other things in the kernel (for example,
257  * service a page fault) in busy_stolen.  We don't accumulate
258  * stolen time for a vcore when it is inactive, or for a vcpu
259  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
260  * a misnomer; it means that the vcpu task is not executing in
261  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
262  * the kernel.  We don't have any way of dividing up that time
263  * between time that the vcpu is genuinely stopped, time that
264  * the task is actively working on behalf of the vcpu, and time
265  * that the task is preempted, so we don't count any of it as
266  * stolen.
267  *
268  * Updates to busy_stolen are protected by arch.tbacct_lock;
269  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
270  * lock.  The stolen times are measured in units of timebase ticks.
271  * (Note that the != TB_NIL checks below are purely defensive;
272  * they should never fail.)
273  */
274 
275 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb)
276 {
277 	unsigned long flags;
278 
279 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
280 
281 	spin_lock_irqsave(&vc->stoltb_lock, flags);
282 	vc->preempt_tb = tb;
283 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
284 }
285 
286 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb)
287 {
288 	unsigned long flags;
289 
290 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
291 
292 	spin_lock_irqsave(&vc->stoltb_lock, flags);
293 	if (vc->preempt_tb != TB_NIL) {
294 		vc->stolen_tb += tb - vc->preempt_tb;
295 		vc->preempt_tb = TB_NIL;
296 	}
297 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
298 }
299 
300 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
301 {
302 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
303 	unsigned long flags;
304 	u64 now;
305 
306 	if (cpu_has_feature(CPU_FTR_ARCH_300))
307 		return;
308 
309 	now = mftb();
310 
311 	/*
312 	 * We can test vc->runner without taking the vcore lock,
313 	 * because only this task ever sets vc->runner to this
314 	 * vcpu, and once it is set to this vcpu, only this task
315 	 * ever sets it to NULL.
316 	 */
317 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
318 		kvmppc_core_end_stolen(vc, now);
319 
320 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
321 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
322 	    vcpu->arch.busy_preempt != TB_NIL) {
323 		vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt;
324 		vcpu->arch.busy_preempt = TB_NIL;
325 	}
326 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
327 }
328 
329 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
330 {
331 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
332 	unsigned long flags;
333 	u64 now;
334 
335 	if (cpu_has_feature(CPU_FTR_ARCH_300))
336 		return;
337 
338 	now = mftb();
339 
340 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
341 		kvmppc_core_start_stolen(vc, now);
342 
343 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
344 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
345 		vcpu->arch.busy_preempt = now;
346 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
347 }
348 
349 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
350 {
351 	vcpu->arch.pvr = pvr;
352 }
353 
354 /* Dummy value used in computing PCR value below */
355 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
356 
357 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
358 {
359 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
360 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
361 
362 	/* We can (emulate) our own architecture version and anything older */
363 	if (cpu_has_feature(CPU_FTR_ARCH_31))
364 		host_pcr_bit = PCR_ARCH_31;
365 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
366 		host_pcr_bit = PCR_ARCH_300;
367 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
368 		host_pcr_bit = PCR_ARCH_207;
369 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
370 		host_pcr_bit = PCR_ARCH_206;
371 	else
372 		host_pcr_bit = PCR_ARCH_205;
373 
374 	/* Determine lowest PCR bit needed to run guest in given PVR level */
375 	guest_pcr_bit = host_pcr_bit;
376 	if (arch_compat) {
377 		switch (arch_compat) {
378 		case PVR_ARCH_205:
379 			guest_pcr_bit = PCR_ARCH_205;
380 			break;
381 		case PVR_ARCH_206:
382 		case PVR_ARCH_206p:
383 			guest_pcr_bit = PCR_ARCH_206;
384 			break;
385 		case PVR_ARCH_207:
386 			guest_pcr_bit = PCR_ARCH_207;
387 			break;
388 		case PVR_ARCH_300:
389 			guest_pcr_bit = PCR_ARCH_300;
390 			break;
391 		case PVR_ARCH_31:
392 			guest_pcr_bit = PCR_ARCH_31;
393 			break;
394 		default:
395 			return -EINVAL;
396 		}
397 	}
398 
399 	/* Check requested PCR bits don't exceed our capabilities */
400 	if (guest_pcr_bit > host_pcr_bit)
401 		return -EINVAL;
402 
403 	spin_lock(&vc->lock);
404 	vc->arch_compat = arch_compat;
405 	/*
406 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
407 	 * Also set all reserved PCR bits
408 	 */
409 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
410 	spin_unlock(&vc->lock);
411 
412 	return 0;
413 }
414 
415 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
416 {
417 	int r;
418 
419 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
420 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
421 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
422 	for (r = 0; r < 16; ++r)
423 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
424 		       r, kvmppc_get_gpr(vcpu, r),
425 		       r+16, kvmppc_get_gpr(vcpu, r+16));
426 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
427 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
428 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
429 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
430 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
431 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
432 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
433 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
434 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
435 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
436 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
437 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
438 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
439 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
440 	for (r = 0; r < vcpu->arch.slb_max; ++r)
441 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
442 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
443 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
444 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
445 	       vcpu->arch.last_inst);
446 }
447 
448 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
449 {
450 	return kvm_get_vcpu_by_id(kvm, id);
451 }
452 
453 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
454 {
455 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
456 	vpa->yield_count = cpu_to_be32(1);
457 }
458 
459 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
460 		   unsigned long addr, unsigned long len)
461 {
462 	/* check address is cacheline aligned */
463 	if (addr & (L1_CACHE_BYTES - 1))
464 		return -EINVAL;
465 	spin_lock(&vcpu->arch.vpa_update_lock);
466 	if (v->next_gpa != addr || v->len != len) {
467 		v->next_gpa = addr;
468 		v->len = addr ? len : 0;
469 		v->update_pending = 1;
470 	}
471 	spin_unlock(&vcpu->arch.vpa_update_lock);
472 	return 0;
473 }
474 
475 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
476 struct reg_vpa {
477 	u32 dummy;
478 	union {
479 		__be16 hword;
480 		__be32 word;
481 	} length;
482 };
483 
484 static int vpa_is_registered(struct kvmppc_vpa *vpap)
485 {
486 	if (vpap->update_pending)
487 		return vpap->next_gpa != 0;
488 	return vpap->pinned_addr != NULL;
489 }
490 
491 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
492 				       unsigned long flags,
493 				       unsigned long vcpuid, unsigned long vpa)
494 {
495 	struct kvm *kvm = vcpu->kvm;
496 	unsigned long len, nb;
497 	void *va;
498 	struct kvm_vcpu *tvcpu;
499 	int err;
500 	int subfunc;
501 	struct kvmppc_vpa *vpap;
502 
503 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
504 	if (!tvcpu)
505 		return H_PARAMETER;
506 
507 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
508 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
509 	    subfunc == H_VPA_REG_SLB) {
510 		/* Registering new area - address must be cache-line aligned */
511 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
512 			return H_PARAMETER;
513 
514 		/* convert logical addr to kernel addr and read length */
515 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
516 		if (va == NULL)
517 			return H_PARAMETER;
518 		if (subfunc == H_VPA_REG_VPA)
519 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
520 		else
521 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
522 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
523 
524 		/* Check length */
525 		if (len > nb || len < sizeof(struct reg_vpa))
526 			return H_PARAMETER;
527 	} else {
528 		vpa = 0;
529 		len = 0;
530 	}
531 
532 	err = H_PARAMETER;
533 	vpap = NULL;
534 	spin_lock(&tvcpu->arch.vpa_update_lock);
535 
536 	switch (subfunc) {
537 	case H_VPA_REG_VPA:		/* register VPA */
538 		/*
539 		 * The size of our lppaca is 1kB because of the way we align
540 		 * it for the guest to avoid crossing a 4kB boundary. We only
541 		 * use 640 bytes of the structure though, so we should accept
542 		 * clients that set a size of 640.
543 		 */
544 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
545 		if (len < sizeof(struct lppaca))
546 			break;
547 		vpap = &tvcpu->arch.vpa;
548 		err = 0;
549 		break;
550 
551 	case H_VPA_REG_DTL:		/* register DTL */
552 		if (len < sizeof(struct dtl_entry))
553 			break;
554 		len -= len % sizeof(struct dtl_entry);
555 
556 		/* Check that they have previously registered a VPA */
557 		err = H_RESOURCE;
558 		if (!vpa_is_registered(&tvcpu->arch.vpa))
559 			break;
560 
561 		vpap = &tvcpu->arch.dtl;
562 		err = 0;
563 		break;
564 
565 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
566 		/* Check that they have previously registered a VPA */
567 		err = H_RESOURCE;
568 		if (!vpa_is_registered(&tvcpu->arch.vpa))
569 			break;
570 
571 		vpap = &tvcpu->arch.slb_shadow;
572 		err = 0;
573 		break;
574 
575 	case H_VPA_DEREG_VPA:		/* deregister VPA */
576 		/* Check they don't still have a DTL or SLB buf registered */
577 		err = H_RESOURCE;
578 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
579 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
580 			break;
581 
582 		vpap = &tvcpu->arch.vpa;
583 		err = 0;
584 		break;
585 
586 	case H_VPA_DEREG_DTL:		/* deregister DTL */
587 		vpap = &tvcpu->arch.dtl;
588 		err = 0;
589 		break;
590 
591 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
592 		vpap = &tvcpu->arch.slb_shadow;
593 		err = 0;
594 		break;
595 	}
596 
597 	if (vpap) {
598 		vpap->next_gpa = vpa;
599 		vpap->len = len;
600 		vpap->update_pending = 1;
601 	}
602 
603 	spin_unlock(&tvcpu->arch.vpa_update_lock);
604 
605 	return err;
606 }
607 
608 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
609 {
610 	struct kvm *kvm = vcpu->kvm;
611 	void *va;
612 	unsigned long nb;
613 	unsigned long gpa;
614 
615 	/*
616 	 * We need to pin the page pointed to by vpap->next_gpa,
617 	 * but we can't call kvmppc_pin_guest_page under the lock
618 	 * as it does get_user_pages() and down_read().  So we
619 	 * have to drop the lock, pin the page, then get the lock
620 	 * again and check that a new area didn't get registered
621 	 * in the meantime.
622 	 */
623 	for (;;) {
624 		gpa = vpap->next_gpa;
625 		spin_unlock(&vcpu->arch.vpa_update_lock);
626 		va = NULL;
627 		nb = 0;
628 		if (gpa)
629 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
630 		spin_lock(&vcpu->arch.vpa_update_lock);
631 		if (gpa == vpap->next_gpa)
632 			break;
633 		/* sigh... unpin that one and try again */
634 		if (va)
635 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
636 	}
637 
638 	vpap->update_pending = 0;
639 	if (va && nb < vpap->len) {
640 		/*
641 		 * If it's now too short, it must be that userspace
642 		 * has changed the mappings underlying guest memory,
643 		 * so unregister the region.
644 		 */
645 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
646 		va = NULL;
647 	}
648 	if (vpap->pinned_addr)
649 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
650 					vpap->dirty);
651 	vpap->gpa = gpa;
652 	vpap->pinned_addr = va;
653 	vpap->dirty = false;
654 	if (va)
655 		vpap->pinned_end = va + vpap->len;
656 }
657 
658 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
659 {
660 	if (!(vcpu->arch.vpa.update_pending ||
661 	      vcpu->arch.slb_shadow.update_pending ||
662 	      vcpu->arch.dtl.update_pending))
663 		return;
664 
665 	spin_lock(&vcpu->arch.vpa_update_lock);
666 	if (vcpu->arch.vpa.update_pending) {
667 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
668 		if (vcpu->arch.vpa.pinned_addr)
669 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
670 	}
671 	if (vcpu->arch.dtl.update_pending) {
672 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
673 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
674 		vcpu->arch.dtl_index = 0;
675 	}
676 	if (vcpu->arch.slb_shadow.update_pending)
677 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
678 	spin_unlock(&vcpu->arch.vpa_update_lock);
679 }
680 
681 /*
682  * Return the accumulated stolen time for the vcore up until `now'.
683  * The caller should hold the vcore lock.
684  */
685 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
686 {
687 	u64 p;
688 	unsigned long flags;
689 
690 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
691 
692 	spin_lock_irqsave(&vc->stoltb_lock, flags);
693 	p = vc->stolen_tb;
694 	if (vc->vcore_state != VCORE_INACTIVE &&
695 	    vc->preempt_tb != TB_NIL)
696 		p += now - vc->preempt_tb;
697 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
698 	return p;
699 }
700 
701 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
702 					unsigned int pcpu, u64 now,
703 					unsigned long stolen)
704 {
705 	struct dtl_entry *dt;
706 	struct lppaca *vpa;
707 
708 	dt = vcpu->arch.dtl_ptr;
709 	vpa = vcpu->arch.vpa.pinned_addr;
710 
711 	if (!dt || !vpa)
712 		return;
713 
714 	dt->dispatch_reason = 7;
715 	dt->preempt_reason = 0;
716 	dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid);
717 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
718 	dt->ready_to_enqueue_time = 0;
719 	dt->waiting_to_ready_time = 0;
720 	dt->timebase = cpu_to_be64(now);
721 	dt->fault_addr = 0;
722 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
723 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
724 
725 	++dt;
726 	if (dt == vcpu->arch.dtl.pinned_end)
727 		dt = vcpu->arch.dtl.pinned_addr;
728 	vcpu->arch.dtl_ptr = dt;
729 	/* order writing *dt vs. writing vpa->dtl_idx */
730 	smp_wmb();
731 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
732 	vcpu->arch.dtl.dirty = true;
733 }
734 
735 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
736 				    struct kvmppc_vcore *vc)
737 {
738 	unsigned long stolen;
739 	unsigned long core_stolen;
740 	u64 now;
741 	unsigned long flags;
742 
743 	now = mftb();
744 
745 	core_stolen = vcore_stolen_time(vc, now);
746 	stolen = core_stolen - vcpu->arch.stolen_logged;
747 	vcpu->arch.stolen_logged = core_stolen;
748 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
749 	stolen += vcpu->arch.busy_stolen;
750 	vcpu->arch.busy_stolen = 0;
751 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
752 
753 	__kvmppc_create_dtl_entry(vcpu, vc->pcpu, now + vc->tb_offset, stolen);
754 }
755 
756 /* See if there is a doorbell interrupt pending for a vcpu */
757 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
758 {
759 	int thr;
760 	struct kvmppc_vcore *vc;
761 
762 	if (vcpu->arch.doorbell_request)
763 		return true;
764 	if (cpu_has_feature(CPU_FTR_ARCH_300))
765 		return false;
766 	/*
767 	 * Ensure that the read of vcore->dpdes comes after the read
768 	 * of vcpu->doorbell_request.  This barrier matches the
769 	 * smp_wmb() in kvmppc_guest_entry_inject().
770 	 */
771 	smp_rmb();
772 	vc = vcpu->arch.vcore;
773 	thr = vcpu->vcpu_id - vc->first_vcpuid;
774 	return !!(vc->dpdes & (1 << thr));
775 }
776 
777 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
778 {
779 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
780 		return true;
781 	if ((!vcpu->arch.vcore->arch_compat) &&
782 	    cpu_has_feature(CPU_FTR_ARCH_207S))
783 		return true;
784 	return false;
785 }
786 
787 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
788 			     unsigned long resource, unsigned long value1,
789 			     unsigned long value2)
790 {
791 	switch (resource) {
792 	case H_SET_MODE_RESOURCE_SET_CIABR:
793 		if (!kvmppc_power8_compatible(vcpu))
794 			return H_P2;
795 		if (value2)
796 			return H_P4;
797 		if (mflags)
798 			return H_UNSUPPORTED_FLAG_START;
799 		/* Guests can't breakpoint the hypervisor */
800 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
801 			return H_P3;
802 		vcpu->arch.ciabr  = value1;
803 		return H_SUCCESS;
804 	case H_SET_MODE_RESOURCE_SET_DAWR0:
805 		if (!kvmppc_power8_compatible(vcpu))
806 			return H_P2;
807 		if (!ppc_breakpoint_available())
808 			return H_P2;
809 		if (mflags)
810 			return H_UNSUPPORTED_FLAG_START;
811 		if (value2 & DABRX_HYP)
812 			return H_P4;
813 		vcpu->arch.dawr0  = value1;
814 		vcpu->arch.dawrx0 = value2;
815 		return H_SUCCESS;
816 	case H_SET_MODE_RESOURCE_SET_DAWR1:
817 		if (!kvmppc_power8_compatible(vcpu))
818 			return H_P2;
819 		if (!ppc_breakpoint_available())
820 			return H_P2;
821 		if (!cpu_has_feature(CPU_FTR_DAWR1))
822 			return H_P2;
823 		if (!vcpu->kvm->arch.dawr1_enabled)
824 			return H_FUNCTION;
825 		if (mflags)
826 			return H_UNSUPPORTED_FLAG_START;
827 		if (value2 & DABRX_HYP)
828 			return H_P4;
829 		vcpu->arch.dawr1  = value1;
830 		vcpu->arch.dawrx1 = value2;
831 		return H_SUCCESS;
832 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
833 		/*
834 		 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
835 		 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
836 		 */
837 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
838 				kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
839 			return H_UNSUPPORTED_FLAG_START;
840 		return H_TOO_HARD;
841 	default:
842 		return H_TOO_HARD;
843 	}
844 }
845 
846 /* Copy guest memory in place - must reside within a single memslot */
847 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
848 				  unsigned long len)
849 {
850 	struct kvm_memory_slot *to_memslot = NULL;
851 	struct kvm_memory_slot *from_memslot = NULL;
852 	unsigned long to_addr, from_addr;
853 	int r;
854 
855 	/* Get HPA for from address */
856 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
857 	if (!from_memslot)
858 		return -EFAULT;
859 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
860 			     << PAGE_SHIFT))
861 		return -EINVAL;
862 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
863 	if (kvm_is_error_hva(from_addr))
864 		return -EFAULT;
865 	from_addr |= (from & (PAGE_SIZE - 1));
866 
867 	/* Get HPA for to address */
868 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
869 	if (!to_memslot)
870 		return -EFAULT;
871 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
872 			   << PAGE_SHIFT))
873 		return -EINVAL;
874 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
875 	if (kvm_is_error_hva(to_addr))
876 		return -EFAULT;
877 	to_addr |= (to & (PAGE_SIZE - 1));
878 
879 	/* Perform copy */
880 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
881 			     len);
882 	if (r)
883 		return -EFAULT;
884 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
885 	return 0;
886 }
887 
888 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
889 			       unsigned long dest, unsigned long src)
890 {
891 	u64 pg_sz = SZ_4K;		/* 4K page size */
892 	u64 pg_mask = SZ_4K - 1;
893 	int ret;
894 
895 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
896 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
897 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
898 		return H_PARAMETER;
899 
900 	/* dest (and src if copy_page flag set) must be page aligned */
901 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
902 		return H_PARAMETER;
903 
904 	/* zero and/or copy the page as determined by the flags */
905 	if (flags & H_COPY_PAGE) {
906 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
907 		if (ret < 0)
908 			return H_PARAMETER;
909 	} else if (flags & H_ZERO_PAGE) {
910 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
911 		if (ret < 0)
912 			return H_PARAMETER;
913 	}
914 
915 	/* We can ignore the remaining flags */
916 
917 	return H_SUCCESS;
918 }
919 
920 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
921 {
922 	struct kvmppc_vcore *vcore = target->arch.vcore;
923 
924 	/*
925 	 * We expect to have been called by the real mode handler
926 	 * (kvmppc_rm_h_confer()) which would have directly returned
927 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
928 	 * have useful work to do and should not confer) so we don't
929 	 * recheck that here.
930 	 *
931 	 * In the case of the P9 single vcpu per vcore case, the real
932 	 * mode handler is not called but no other threads are in the
933 	 * source vcore.
934 	 */
935 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
936 		spin_lock(&vcore->lock);
937 		if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
938 		    vcore->vcore_state != VCORE_INACTIVE &&
939 		    vcore->runner)
940 			target = vcore->runner;
941 		spin_unlock(&vcore->lock);
942 	}
943 
944 	return kvm_vcpu_yield_to(target);
945 }
946 
947 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
948 {
949 	int yield_count = 0;
950 	struct lppaca *lppaca;
951 
952 	spin_lock(&vcpu->arch.vpa_update_lock);
953 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
954 	if (lppaca)
955 		yield_count = be32_to_cpu(lppaca->yield_count);
956 	spin_unlock(&vcpu->arch.vpa_update_lock);
957 	return yield_count;
958 }
959 
960 /*
961  * H_RPT_INVALIDATE hcall handler for nested guests.
962  *
963  * Handles only nested process-scoped invalidation requests in L0.
964  */
965 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
966 {
967 	unsigned long type = kvmppc_get_gpr(vcpu, 6);
968 	unsigned long pid, pg_sizes, start, end;
969 
970 	/*
971 	 * The partition-scoped invalidations aren't handled here in L0.
972 	 */
973 	if (type & H_RPTI_TYPE_NESTED)
974 		return RESUME_HOST;
975 
976 	pid = kvmppc_get_gpr(vcpu, 4);
977 	pg_sizes = kvmppc_get_gpr(vcpu, 7);
978 	start = kvmppc_get_gpr(vcpu, 8);
979 	end = kvmppc_get_gpr(vcpu, 9);
980 
981 	do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
982 				type, pg_sizes, start, end);
983 
984 	kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
985 	return RESUME_GUEST;
986 }
987 
988 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
989 				    unsigned long id, unsigned long target,
990 				    unsigned long type, unsigned long pg_sizes,
991 				    unsigned long start, unsigned long end)
992 {
993 	if (!kvm_is_radix(vcpu->kvm))
994 		return H_UNSUPPORTED;
995 
996 	if (end < start)
997 		return H_P5;
998 
999 	/*
1000 	 * Partition-scoped invalidation for nested guests.
1001 	 */
1002 	if (type & H_RPTI_TYPE_NESTED) {
1003 		if (!nesting_enabled(vcpu->kvm))
1004 			return H_FUNCTION;
1005 
1006 		/* Support only cores as target */
1007 		if (target != H_RPTI_TARGET_CMMU)
1008 			return H_P2;
1009 
1010 		return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
1011 					       start, end);
1012 	}
1013 
1014 	/*
1015 	 * Process-scoped invalidation for L1 guests.
1016 	 */
1017 	do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
1018 				type, pg_sizes, start, end);
1019 	return H_SUCCESS;
1020 }
1021 
1022 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
1023 {
1024 	struct kvm *kvm = vcpu->kvm;
1025 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
1026 	unsigned long target, ret = H_SUCCESS;
1027 	int yield_count;
1028 	struct kvm_vcpu *tvcpu;
1029 	int idx, rc;
1030 
1031 	if (req <= MAX_HCALL_OPCODE &&
1032 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1033 		return RESUME_HOST;
1034 
1035 	switch (req) {
1036 	case H_REMOVE:
1037 		ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1038 					kvmppc_get_gpr(vcpu, 5),
1039 					kvmppc_get_gpr(vcpu, 6));
1040 		if (ret == H_TOO_HARD)
1041 			return RESUME_HOST;
1042 		break;
1043 	case H_ENTER:
1044 		ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1045 					kvmppc_get_gpr(vcpu, 5),
1046 					kvmppc_get_gpr(vcpu, 6),
1047 					kvmppc_get_gpr(vcpu, 7));
1048 		if (ret == H_TOO_HARD)
1049 			return RESUME_HOST;
1050 		break;
1051 	case H_READ:
1052 		ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1053 					kvmppc_get_gpr(vcpu, 5));
1054 		if (ret == H_TOO_HARD)
1055 			return RESUME_HOST;
1056 		break;
1057 	case H_CLEAR_MOD:
1058 		ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1059 					kvmppc_get_gpr(vcpu, 5));
1060 		if (ret == H_TOO_HARD)
1061 			return RESUME_HOST;
1062 		break;
1063 	case H_CLEAR_REF:
1064 		ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1065 					kvmppc_get_gpr(vcpu, 5));
1066 		if (ret == H_TOO_HARD)
1067 			return RESUME_HOST;
1068 		break;
1069 	case H_PROTECT:
1070 		ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1071 					kvmppc_get_gpr(vcpu, 5),
1072 					kvmppc_get_gpr(vcpu, 6));
1073 		if (ret == H_TOO_HARD)
1074 			return RESUME_HOST;
1075 		break;
1076 	case H_BULK_REMOVE:
1077 		ret = kvmppc_h_bulk_remove(vcpu);
1078 		if (ret == H_TOO_HARD)
1079 			return RESUME_HOST;
1080 		break;
1081 
1082 	case H_CEDE:
1083 		break;
1084 	case H_PROD:
1085 		target = kvmppc_get_gpr(vcpu, 4);
1086 		tvcpu = kvmppc_find_vcpu(kvm, target);
1087 		if (!tvcpu) {
1088 			ret = H_PARAMETER;
1089 			break;
1090 		}
1091 		tvcpu->arch.prodded = 1;
1092 		smp_mb();
1093 		if (tvcpu->arch.ceded)
1094 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1095 		break;
1096 	case H_CONFER:
1097 		target = kvmppc_get_gpr(vcpu, 4);
1098 		if (target == -1)
1099 			break;
1100 		tvcpu = kvmppc_find_vcpu(kvm, target);
1101 		if (!tvcpu) {
1102 			ret = H_PARAMETER;
1103 			break;
1104 		}
1105 		yield_count = kvmppc_get_gpr(vcpu, 5);
1106 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
1107 			break;
1108 		kvm_arch_vcpu_yield_to(tvcpu);
1109 		break;
1110 	case H_REGISTER_VPA:
1111 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1112 					kvmppc_get_gpr(vcpu, 5),
1113 					kvmppc_get_gpr(vcpu, 6));
1114 		break;
1115 	case H_RTAS:
1116 		if (list_empty(&kvm->arch.rtas_tokens))
1117 			return RESUME_HOST;
1118 
1119 		idx = srcu_read_lock(&kvm->srcu);
1120 		rc = kvmppc_rtas_hcall(vcpu);
1121 		srcu_read_unlock(&kvm->srcu, idx);
1122 
1123 		if (rc == -ENOENT)
1124 			return RESUME_HOST;
1125 		else if (rc == 0)
1126 			break;
1127 
1128 		/* Send the error out to userspace via KVM_RUN */
1129 		return rc;
1130 	case H_LOGICAL_CI_LOAD:
1131 		ret = kvmppc_h_logical_ci_load(vcpu);
1132 		if (ret == H_TOO_HARD)
1133 			return RESUME_HOST;
1134 		break;
1135 	case H_LOGICAL_CI_STORE:
1136 		ret = kvmppc_h_logical_ci_store(vcpu);
1137 		if (ret == H_TOO_HARD)
1138 			return RESUME_HOST;
1139 		break;
1140 	case H_SET_MODE:
1141 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1142 					kvmppc_get_gpr(vcpu, 5),
1143 					kvmppc_get_gpr(vcpu, 6),
1144 					kvmppc_get_gpr(vcpu, 7));
1145 		if (ret == H_TOO_HARD)
1146 			return RESUME_HOST;
1147 		break;
1148 	case H_XIRR:
1149 	case H_CPPR:
1150 	case H_EOI:
1151 	case H_IPI:
1152 	case H_IPOLL:
1153 	case H_XIRR_X:
1154 		if (kvmppc_xics_enabled(vcpu)) {
1155 			if (xics_on_xive()) {
1156 				ret = H_NOT_AVAILABLE;
1157 				return RESUME_GUEST;
1158 			}
1159 			ret = kvmppc_xics_hcall(vcpu, req);
1160 			break;
1161 		}
1162 		return RESUME_HOST;
1163 	case H_SET_DABR:
1164 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1165 		break;
1166 	case H_SET_XDABR:
1167 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1168 						kvmppc_get_gpr(vcpu, 5));
1169 		break;
1170 #ifdef CONFIG_SPAPR_TCE_IOMMU
1171 	case H_GET_TCE:
1172 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1173 						kvmppc_get_gpr(vcpu, 5));
1174 		if (ret == H_TOO_HARD)
1175 			return RESUME_HOST;
1176 		break;
1177 	case H_PUT_TCE:
1178 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1179 						kvmppc_get_gpr(vcpu, 5),
1180 						kvmppc_get_gpr(vcpu, 6));
1181 		if (ret == H_TOO_HARD)
1182 			return RESUME_HOST;
1183 		break;
1184 	case H_PUT_TCE_INDIRECT:
1185 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1186 						kvmppc_get_gpr(vcpu, 5),
1187 						kvmppc_get_gpr(vcpu, 6),
1188 						kvmppc_get_gpr(vcpu, 7));
1189 		if (ret == H_TOO_HARD)
1190 			return RESUME_HOST;
1191 		break;
1192 	case H_STUFF_TCE:
1193 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1194 						kvmppc_get_gpr(vcpu, 5),
1195 						kvmppc_get_gpr(vcpu, 6),
1196 						kvmppc_get_gpr(vcpu, 7));
1197 		if (ret == H_TOO_HARD)
1198 			return RESUME_HOST;
1199 		break;
1200 #endif
1201 	case H_RANDOM:
1202 		if (!arch_get_random_seed_long(&vcpu->arch.regs.gpr[4]))
1203 			ret = H_HARDWARE;
1204 		break;
1205 	case H_RPT_INVALIDATE:
1206 		ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1207 					      kvmppc_get_gpr(vcpu, 5),
1208 					      kvmppc_get_gpr(vcpu, 6),
1209 					      kvmppc_get_gpr(vcpu, 7),
1210 					      kvmppc_get_gpr(vcpu, 8),
1211 					      kvmppc_get_gpr(vcpu, 9));
1212 		break;
1213 
1214 	case H_SET_PARTITION_TABLE:
1215 		ret = H_FUNCTION;
1216 		if (nesting_enabled(kvm))
1217 			ret = kvmhv_set_partition_table(vcpu);
1218 		break;
1219 	case H_ENTER_NESTED:
1220 		ret = H_FUNCTION;
1221 		if (!nesting_enabled(kvm))
1222 			break;
1223 		ret = kvmhv_enter_nested_guest(vcpu);
1224 		if (ret == H_INTERRUPT) {
1225 			kvmppc_set_gpr(vcpu, 3, 0);
1226 			vcpu->arch.hcall_needed = 0;
1227 			return -EINTR;
1228 		} else if (ret == H_TOO_HARD) {
1229 			kvmppc_set_gpr(vcpu, 3, 0);
1230 			vcpu->arch.hcall_needed = 0;
1231 			return RESUME_HOST;
1232 		}
1233 		break;
1234 	case H_TLB_INVALIDATE:
1235 		ret = H_FUNCTION;
1236 		if (nesting_enabled(kvm))
1237 			ret = kvmhv_do_nested_tlbie(vcpu);
1238 		break;
1239 	case H_COPY_TOFROM_GUEST:
1240 		ret = H_FUNCTION;
1241 		if (nesting_enabled(kvm))
1242 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1243 		break;
1244 	case H_PAGE_INIT:
1245 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1246 					 kvmppc_get_gpr(vcpu, 5),
1247 					 kvmppc_get_gpr(vcpu, 6));
1248 		break;
1249 	case H_SVM_PAGE_IN:
1250 		ret = H_UNSUPPORTED;
1251 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1252 			ret = kvmppc_h_svm_page_in(kvm,
1253 						   kvmppc_get_gpr(vcpu, 4),
1254 						   kvmppc_get_gpr(vcpu, 5),
1255 						   kvmppc_get_gpr(vcpu, 6));
1256 		break;
1257 	case H_SVM_PAGE_OUT:
1258 		ret = H_UNSUPPORTED;
1259 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1260 			ret = kvmppc_h_svm_page_out(kvm,
1261 						    kvmppc_get_gpr(vcpu, 4),
1262 						    kvmppc_get_gpr(vcpu, 5),
1263 						    kvmppc_get_gpr(vcpu, 6));
1264 		break;
1265 	case H_SVM_INIT_START:
1266 		ret = H_UNSUPPORTED;
1267 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1268 			ret = kvmppc_h_svm_init_start(kvm);
1269 		break;
1270 	case H_SVM_INIT_DONE:
1271 		ret = H_UNSUPPORTED;
1272 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1273 			ret = kvmppc_h_svm_init_done(kvm);
1274 		break;
1275 	case H_SVM_INIT_ABORT:
1276 		/*
1277 		 * Even if that call is made by the Ultravisor, the SSR1 value
1278 		 * is the guest context one, with the secure bit clear as it has
1279 		 * not yet been secured. So we can't check it here.
1280 		 * Instead the kvm->arch.secure_guest flag is checked inside
1281 		 * kvmppc_h_svm_init_abort().
1282 		 */
1283 		ret = kvmppc_h_svm_init_abort(kvm);
1284 		break;
1285 
1286 	default:
1287 		return RESUME_HOST;
1288 	}
1289 	WARN_ON_ONCE(ret == H_TOO_HARD);
1290 	kvmppc_set_gpr(vcpu, 3, ret);
1291 	vcpu->arch.hcall_needed = 0;
1292 	return RESUME_GUEST;
1293 }
1294 
1295 /*
1296  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1297  * handlers in book3s_hv_rmhandlers.S.
1298  *
1299  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1300  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1301  */
1302 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1303 {
1304 	vcpu->arch.shregs.msr |= MSR_EE;
1305 	vcpu->arch.ceded = 1;
1306 	smp_mb();
1307 	if (vcpu->arch.prodded) {
1308 		vcpu->arch.prodded = 0;
1309 		smp_mb();
1310 		vcpu->arch.ceded = 0;
1311 	}
1312 }
1313 
1314 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1315 {
1316 	switch (cmd) {
1317 	case H_CEDE:
1318 	case H_PROD:
1319 	case H_CONFER:
1320 	case H_REGISTER_VPA:
1321 	case H_SET_MODE:
1322 	case H_LOGICAL_CI_LOAD:
1323 	case H_LOGICAL_CI_STORE:
1324 #ifdef CONFIG_KVM_XICS
1325 	case H_XIRR:
1326 	case H_CPPR:
1327 	case H_EOI:
1328 	case H_IPI:
1329 	case H_IPOLL:
1330 	case H_XIRR_X:
1331 #endif
1332 	case H_PAGE_INIT:
1333 	case H_RPT_INVALIDATE:
1334 		return 1;
1335 	}
1336 
1337 	/* See if it's in the real-mode table */
1338 	return kvmppc_hcall_impl_hv_realmode(cmd);
1339 }
1340 
1341 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1342 {
1343 	u32 last_inst;
1344 
1345 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1346 					EMULATE_DONE) {
1347 		/*
1348 		 * Fetch failed, so return to guest and
1349 		 * try executing it again.
1350 		 */
1351 		return RESUME_GUEST;
1352 	}
1353 
1354 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1355 		vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1356 		vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1357 		return RESUME_HOST;
1358 	} else {
1359 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1360 		return RESUME_GUEST;
1361 	}
1362 }
1363 
1364 static void do_nothing(void *x)
1365 {
1366 }
1367 
1368 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1369 {
1370 	int thr, cpu, pcpu, nthreads;
1371 	struct kvm_vcpu *v;
1372 	unsigned long dpdes;
1373 
1374 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1375 	dpdes = 0;
1376 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1377 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1378 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1379 		if (!v)
1380 			continue;
1381 		/*
1382 		 * If the vcpu is currently running on a physical cpu thread,
1383 		 * interrupt it in order to pull it out of the guest briefly,
1384 		 * which will update its vcore->dpdes value.
1385 		 */
1386 		pcpu = READ_ONCE(v->cpu);
1387 		if (pcpu >= 0)
1388 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1389 		if (kvmppc_doorbell_pending(v))
1390 			dpdes |= 1 << thr;
1391 	}
1392 	return dpdes;
1393 }
1394 
1395 /*
1396  * On POWER9, emulate doorbell-related instructions in order to
1397  * give the guest the illusion of running on a multi-threaded core.
1398  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1399  * and mfspr DPDES.
1400  */
1401 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1402 {
1403 	u32 inst, rb, thr;
1404 	unsigned long arg;
1405 	struct kvm *kvm = vcpu->kvm;
1406 	struct kvm_vcpu *tvcpu;
1407 
1408 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1409 		return RESUME_GUEST;
1410 	if (get_op(inst) != 31)
1411 		return EMULATE_FAIL;
1412 	rb = get_rb(inst);
1413 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1414 	switch (get_xop(inst)) {
1415 	case OP_31_XOP_MSGSNDP:
1416 		arg = kvmppc_get_gpr(vcpu, rb);
1417 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1418 			break;
1419 		arg &= 0x7f;
1420 		if (arg >= kvm->arch.emul_smt_mode)
1421 			break;
1422 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1423 		if (!tvcpu)
1424 			break;
1425 		if (!tvcpu->arch.doorbell_request) {
1426 			tvcpu->arch.doorbell_request = 1;
1427 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1428 		}
1429 		break;
1430 	case OP_31_XOP_MSGCLRP:
1431 		arg = kvmppc_get_gpr(vcpu, rb);
1432 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1433 			break;
1434 		vcpu->arch.vcore->dpdes = 0;
1435 		vcpu->arch.doorbell_request = 0;
1436 		break;
1437 	case OP_31_XOP_MFSPR:
1438 		switch (get_sprn(inst)) {
1439 		case SPRN_TIR:
1440 			arg = thr;
1441 			break;
1442 		case SPRN_DPDES:
1443 			arg = kvmppc_read_dpdes(vcpu);
1444 			break;
1445 		default:
1446 			return EMULATE_FAIL;
1447 		}
1448 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1449 		break;
1450 	default:
1451 		return EMULATE_FAIL;
1452 	}
1453 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1454 	return RESUME_GUEST;
1455 }
1456 
1457 /*
1458  * If the lppaca had pmcregs_in_use clear when we exited the guest, then
1459  * HFSCR_PM is cleared for next entry. If the guest then tries to access
1460  * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM
1461  * back in the guest HFSCR will cause the next entry to load the PMU SPRs and
1462  * allow the guest access to continue.
1463  */
1464 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu)
1465 {
1466 	if (!(vcpu->arch.hfscr_permitted & HFSCR_PM))
1467 		return EMULATE_FAIL;
1468 
1469 	vcpu->arch.hfscr |= HFSCR_PM;
1470 
1471 	return RESUME_GUEST;
1472 }
1473 
1474 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu)
1475 {
1476 	if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB))
1477 		return EMULATE_FAIL;
1478 
1479 	vcpu->arch.hfscr |= HFSCR_EBB;
1480 
1481 	return RESUME_GUEST;
1482 }
1483 
1484 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu)
1485 {
1486 	if (!(vcpu->arch.hfscr_permitted & HFSCR_TM))
1487 		return EMULATE_FAIL;
1488 
1489 	vcpu->arch.hfscr |= HFSCR_TM;
1490 
1491 	return RESUME_GUEST;
1492 }
1493 
1494 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1495 				 struct task_struct *tsk)
1496 {
1497 	struct kvm_run *run = vcpu->run;
1498 	int r = RESUME_HOST;
1499 
1500 	vcpu->stat.sum_exits++;
1501 
1502 	/*
1503 	 * This can happen if an interrupt occurs in the last stages
1504 	 * of guest entry or the first stages of guest exit (i.e. after
1505 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1506 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1507 	 * That can happen due to a bug, or due to a machine check
1508 	 * occurring at just the wrong time.
1509 	 */
1510 	if (vcpu->arch.shregs.msr & MSR_HV) {
1511 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1512 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1513 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1514 			vcpu->arch.shregs.msr);
1515 		kvmppc_dump_regs(vcpu);
1516 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1517 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1518 		return RESUME_HOST;
1519 	}
1520 	run->exit_reason = KVM_EXIT_UNKNOWN;
1521 	run->ready_for_interrupt_injection = 1;
1522 	switch (vcpu->arch.trap) {
1523 	/* We're good on these - the host merely wanted to get our attention */
1524 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1525 		WARN_ON_ONCE(1); /* Should never happen */
1526 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1527 		fallthrough;
1528 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1529 		vcpu->stat.dec_exits++;
1530 		r = RESUME_GUEST;
1531 		break;
1532 	case BOOK3S_INTERRUPT_EXTERNAL:
1533 	case BOOK3S_INTERRUPT_H_DOORBELL:
1534 	case BOOK3S_INTERRUPT_H_VIRT:
1535 		vcpu->stat.ext_intr_exits++;
1536 		r = RESUME_GUEST;
1537 		break;
1538 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1539 	case BOOK3S_INTERRUPT_HMI:
1540 	case BOOK3S_INTERRUPT_PERFMON:
1541 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1542 		r = RESUME_GUEST;
1543 		break;
1544 	case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1545 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1546 					      DEFAULT_RATELIMIT_BURST);
1547 		/*
1548 		 * Print the MCE event to host console. Ratelimit so the guest
1549 		 * can't flood the host log.
1550 		 */
1551 		if (__ratelimit(&rs))
1552 			machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1553 
1554 		/*
1555 		 * If the guest can do FWNMI, exit to userspace so it can
1556 		 * deliver a FWNMI to the guest.
1557 		 * Otherwise we synthesize a machine check for the guest
1558 		 * so that it knows that the machine check occurred.
1559 		 */
1560 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1561 			ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1562 			kvmppc_core_queue_machine_check(vcpu, flags);
1563 			r = RESUME_GUEST;
1564 			break;
1565 		}
1566 
1567 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1568 		run->exit_reason = KVM_EXIT_NMI;
1569 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1570 		/* Clear out the old NMI status from run->flags */
1571 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1572 		/* Now set the NMI status */
1573 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1574 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1575 		else
1576 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1577 
1578 		r = RESUME_HOST;
1579 		break;
1580 	}
1581 	case BOOK3S_INTERRUPT_PROGRAM:
1582 	{
1583 		ulong flags;
1584 		/*
1585 		 * Normally program interrupts are delivered directly
1586 		 * to the guest by the hardware, but we can get here
1587 		 * as a result of a hypervisor emulation interrupt
1588 		 * (e40) getting turned into a 700 by BML RTAS.
1589 		 */
1590 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1591 		kvmppc_core_queue_program(vcpu, flags);
1592 		r = RESUME_GUEST;
1593 		break;
1594 	}
1595 	case BOOK3S_INTERRUPT_SYSCALL:
1596 	{
1597 		int i;
1598 
1599 		if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) {
1600 			/*
1601 			 * Guest userspace executed sc 1. This can only be
1602 			 * reached by the P9 path because the old path
1603 			 * handles this case in realmode hcall handlers.
1604 			 */
1605 			if (!kvmhv_vcpu_is_radix(vcpu)) {
1606 				/*
1607 				 * A guest could be running PR KVM, so this
1608 				 * may be a PR KVM hcall. It must be reflected
1609 				 * to the guest kernel as a sc interrupt.
1610 				 */
1611 				kvmppc_core_queue_syscall(vcpu);
1612 			} else {
1613 				/*
1614 				 * Radix guests can not run PR KVM or nested HV
1615 				 * hash guests which might run PR KVM, so this
1616 				 * is always a privilege fault. Send a program
1617 				 * check to guest kernel.
1618 				 */
1619 				kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1620 			}
1621 			r = RESUME_GUEST;
1622 			break;
1623 		}
1624 
1625 		/*
1626 		 * hcall - gather args and set exit_reason. This will next be
1627 		 * handled by kvmppc_pseries_do_hcall which may be able to deal
1628 		 * with it and resume guest, or may punt to userspace.
1629 		 */
1630 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1631 		for (i = 0; i < 9; ++i)
1632 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1633 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1634 		vcpu->arch.hcall_needed = 1;
1635 		r = RESUME_HOST;
1636 		break;
1637 	}
1638 	/*
1639 	 * We get these next two if the guest accesses a page which it thinks
1640 	 * it has mapped but which is not actually present, either because
1641 	 * it is for an emulated I/O device or because the corresonding
1642 	 * host page has been paged out.
1643 	 *
1644 	 * Any other HDSI/HISI interrupts have been handled already for P7/8
1645 	 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1646 	 * fault handling is done here.
1647 	 */
1648 	case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1649 		unsigned long vsid;
1650 		long err;
1651 
1652 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
1653 		    unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) {
1654 			r = RESUME_GUEST; /* Just retry if it's the canary */
1655 			break;
1656 		}
1657 
1658 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1659 			/*
1660 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1661 			 * already attempted to handle this in rmhandlers. The
1662 			 * hash fault handling below is v3 only (it uses ASDR
1663 			 * via fault_gpa).
1664 			 */
1665 			r = RESUME_PAGE_FAULT;
1666 			break;
1667 		}
1668 
1669 		if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1670 			kvmppc_core_queue_data_storage(vcpu,
1671 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1672 			r = RESUME_GUEST;
1673 			break;
1674 		}
1675 
1676 		if (!(vcpu->arch.shregs.msr & MSR_DR))
1677 			vsid = vcpu->kvm->arch.vrma_slb_v;
1678 		else
1679 			vsid = vcpu->arch.fault_gpa;
1680 
1681 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1682 				vsid, vcpu->arch.fault_dsisr, true);
1683 		if (err == 0) {
1684 			r = RESUME_GUEST;
1685 		} else if (err == -1 || err == -2) {
1686 			r = RESUME_PAGE_FAULT;
1687 		} else {
1688 			kvmppc_core_queue_data_storage(vcpu,
1689 				vcpu->arch.fault_dar, err);
1690 			r = RESUME_GUEST;
1691 		}
1692 		break;
1693 	}
1694 	case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1695 		unsigned long vsid;
1696 		long err;
1697 
1698 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1699 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1700 			DSISR_SRR1_MATCH_64S;
1701 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1702 			/*
1703 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1704 			 * already attempted to handle this in rmhandlers. The
1705 			 * hash fault handling below is v3 only (it uses ASDR
1706 			 * via fault_gpa).
1707 			 */
1708 			if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1709 				vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1710 			r = RESUME_PAGE_FAULT;
1711 			break;
1712 		}
1713 
1714 		if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1715 			kvmppc_core_queue_inst_storage(vcpu,
1716 				vcpu->arch.fault_dsisr);
1717 			r = RESUME_GUEST;
1718 			break;
1719 		}
1720 
1721 		if (!(vcpu->arch.shregs.msr & MSR_IR))
1722 			vsid = vcpu->kvm->arch.vrma_slb_v;
1723 		else
1724 			vsid = vcpu->arch.fault_gpa;
1725 
1726 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1727 				vsid, vcpu->arch.fault_dsisr, false);
1728 		if (err == 0) {
1729 			r = RESUME_GUEST;
1730 		} else if (err == -1) {
1731 			r = RESUME_PAGE_FAULT;
1732 		} else {
1733 			kvmppc_core_queue_inst_storage(vcpu, err);
1734 			r = RESUME_GUEST;
1735 		}
1736 		break;
1737 	}
1738 
1739 	/*
1740 	 * This occurs if the guest executes an illegal instruction.
1741 	 * If the guest debug is disabled, generate a program interrupt
1742 	 * to the guest. If guest debug is enabled, we need to check
1743 	 * whether the instruction is a software breakpoint instruction.
1744 	 * Accordingly return to Guest or Host.
1745 	 */
1746 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1747 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1748 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1749 				swab32(vcpu->arch.emul_inst) :
1750 				vcpu->arch.emul_inst;
1751 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1752 			r = kvmppc_emulate_debug_inst(vcpu);
1753 		} else {
1754 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1755 			r = RESUME_GUEST;
1756 		}
1757 		break;
1758 
1759 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1760 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1761 		/*
1762 		 * This occurs for various TM-related instructions that
1763 		 * we need to emulate on POWER9 DD2.2.  We have already
1764 		 * handled the cases where the guest was in real-suspend
1765 		 * mode and was transitioning to transactional state.
1766 		 */
1767 		r = kvmhv_p9_tm_emulation(vcpu);
1768 		if (r != -1)
1769 			break;
1770 		fallthrough; /* go to facility unavailable handler */
1771 #endif
1772 
1773 	/*
1774 	 * This occurs if the guest (kernel or userspace), does something that
1775 	 * is prohibited by HFSCR.
1776 	 * On POWER9, this could be a doorbell instruction that we need
1777 	 * to emulate.
1778 	 * Otherwise, we just generate a program interrupt to the guest.
1779 	 */
1780 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1781 		u64 cause = vcpu->arch.hfscr >> 56;
1782 
1783 		r = EMULATE_FAIL;
1784 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1785 			if (cause == FSCR_MSGP_LG)
1786 				r = kvmppc_emulate_doorbell_instr(vcpu);
1787 			if (cause == FSCR_PM_LG)
1788 				r = kvmppc_pmu_unavailable(vcpu);
1789 			if (cause == FSCR_EBB_LG)
1790 				r = kvmppc_ebb_unavailable(vcpu);
1791 			if (cause == FSCR_TM_LG)
1792 				r = kvmppc_tm_unavailable(vcpu);
1793 		}
1794 		if (r == EMULATE_FAIL) {
1795 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1796 			r = RESUME_GUEST;
1797 		}
1798 		break;
1799 	}
1800 
1801 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1802 		r = RESUME_PASSTHROUGH;
1803 		break;
1804 	default:
1805 		kvmppc_dump_regs(vcpu);
1806 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1807 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1808 			vcpu->arch.shregs.msr);
1809 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1810 		r = RESUME_HOST;
1811 		break;
1812 	}
1813 
1814 	return r;
1815 }
1816 
1817 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1818 {
1819 	int r;
1820 	int srcu_idx;
1821 
1822 	vcpu->stat.sum_exits++;
1823 
1824 	/*
1825 	 * This can happen if an interrupt occurs in the last stages
1826 	 * of guest entry or the first stages of guest exit (i.e. after
1827 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1828 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1829 	 * That can happen due to a bug, or due to a machine check
1830 	 * occurring at just the wrong time.
1831 	 */
1832 	if (vcpu->arch.shregs.msr & MSR_HV) {
1833 		pr_emerg("KVM trap in HV mode while nested!\n");
1834 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1835 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1836 			 vcpu->arch.shregs.msr);
1837 		kvmppc_dump_regs(vcpu);
1838 		return RESUME_HOST;
1839 	}
1840 	switch (vcpu->arch.trap) {
1841 	/* We're good on these - the host merely wanted to get our attention */
1842 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1843 		vcpu->stat.dec_exits++;
1844 		r = RESUME_GUEST;
1845 		break;
1846 	case BOOK3S_INTERRUPT_EXTERNAL:
1847 		vcpu->stat.ext_intr_exits++;
1848 		r = RESUME_HOST;
1849 		break;
1850 	case BOOK3S_INTERRUPT_H_DOORBELL:
1851 	case BOOK3S_INTERRUPT_H_VIRT:
1852 		vcpu->stat.ext_intr_exits++;
1853 		r = RESUME_GUEST;
1854 		break;
1855 	/* These need to go to the nested HV */
1856 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1857 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1858 		vcpu->stat.dec_exits++;
1859 		r = RESUME_HOST;
1860 		break;
1861 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1862 	case BOOK3S_INTERRUPT_HMI:
1863 	case BOOK3S_INTERRUPT_PERFMON:
1864 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1865 		r = RESUME_GUEST;
1866 		break;
1867 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1868 	{
1869 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1870 					      DEFAULT_RATELIMIT_BURST);
1871 		/* Pass the machine check to the L1 guest */
1872 		r = RESUME_HOST;
1873 		/* Print the MCE event to host console. */
1874 		if (__ratelimit(&rs))
1875 			machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1876 		break;
1877 	}
1878 	/*
1879 	 * We get these next two if the guest accesses a page which it thinks
1880 	 * it has mapped but which is not actually present, either because
1881 	 * it is for an emulated I/O device or because the corresonding
1882 	 * host page has been paged out.
1883 	 */
1884 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1885 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1886 		r = kvmhv_nested_page_fault(vcpu);
1887 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1888 		break;
1889 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1890 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1891 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1892 					 DSISR_SRR1_MATCH_64S;
1893 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1894 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1895 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1896 		r = kvmhv_nested_page_fault(vcpu);
1897 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1898 		break;
1899 
1900 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1901 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1902 		/*
1903 		 * This occurs for various TM-related instructions that
1904 		 * we need to emulate on POWER9 DD2.2.  We have already
1905 		 * handled the cases where the guest was in real-suspend
1906 		 * mode and was transitioning to transactional state.
1907 		 */
1908 		r = kvmhv_p9_tm_emulation(vcpu);
1909 		if (r != -1)
1910 			break;
1911 		fallthrough; /* go to facility unavailable handler */
1912 #endif
1913 
1914 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1915 		u64 cause = vcpu->arch.hfscr >> 56;
1916 
1917 		/*
1918 		 * Only pass HFU interrupts to the L1 if the facility is
1919 		 * permitted but disabled by the L1's HFSCR, otherwise
1920 		 * the interrupt does not make sense to the L1 so turn
1921 		 * it into a HEAI.
1922 		 */
1923 		if (!(vcpu->arch.hfscr_permitted & (1UL << cause)) ||
1924 				(vcpu->arch.nested_hfscr & (1UL << cause))) {
1925 			vcpu->arch.trap = BOOK3S_INTERRUPT_H_EMUL_ASSIST;
1926 
1927 			/*
1928 			 * If the fetch failed, return to guest and
1929 			 * try executing it again.
1930 			 */
1931 			r = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1932 						 &vcpu->arch.emul_inst);
1933 			if (r != EMULATE_DONE)
1934 				r = RESUME_GUEST;
1935 			else
1936 				r = RESUME_HOST;
1937 		} else {
1938 			r = RESUME_HOST;
1939 		}
1940 
1941 		break;
1942 	}
1943 
1944 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1945 		vcpu->arch.trap = 0;
1946 		r = RESUME_GUEST;
1947 		if (!xics_on_xive())
1948 			kvmppc_xics_rm_complete(vcpu, 0);
1949 		break;
1950 	case BOOK3S_INTERRUPT_SYSCALL:
1951 	{
1952 		unsigned long req = kvmppc_get_gpr(vcpu, 3);
1953 
1954 		/*
1955 		 * The H_RPT_INVALIDATE hcalls issued by nested
1956 		 * guests for process-scoped invalidations when
1957 		 * GTSE=0, are handled here in L0.
1958 		 */
1959 		if (req == H_RPT_INVALIDATE) {
1960 			r = kvmppc_nested_h_rpt_invalidate(vcpu);
1961 			break;
1962 		}
1963 
1964 		r = RESUME_HOST;
1965 		break;
1966 	}
1967 	default:
1968 		r = RESUME_HOST;
1969 		break;
1970 	}
1971 
1972 	return r;
1973 }
1974 
1975 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1976 					    struct kvm_sregs *sregs)
1977 {
1978 	int i;
1979 
1980 	memset(sregs, 0, sizeof(struct kvm_sregs));
1981 	sregs->pvr = vcpu->arch.pvr;
1982 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1983 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1984 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1985 	}
1986 
1987 	return 0;
1988 }
1989 
1990 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1991 					    struct kvm_sregs *sregs)
1992 {
1993 	int i, j;
1994 
1995 	/* Only accept the same PVR as the host's, since we can't spoof it */
1996 	if (sregs->pvr != vcpu->arch.pvr)
1997 		return -EINVAL;
1998 
1999 	j = 0;
2000 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
2001 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
2002 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
2003 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
2004 			++j;
2005 		}
2006 	}
2007 	vcpu->arch.slb_max = j;
2008 
2009 	return 0;
2010 }
2011 
2012 /*
2013  * Enforce limits on guest LPCR values based on hardware availability,
2014  * guest configuration, and possibly hypervisor support and security
2015  * concerns.
2016  */
2017 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
2018 {
2019 	/* LPCR_TC only applies to HPT guests */
2020 	if (kvm_is_radix(kvm))
2021 		lpcr &= ~LPCR_TC;
2022 
2023 	/* On POWER8 and above, userspace can modify AIL */
2024 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2025 		lpcr &= ~LPCR_AIL;
2026 	if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
2027 		lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
2028 	/*
2029 	 * On some POWER9s we force AIL off for radix guests to prevent
2030 	 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
2031 	 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
2032 	 * be cached, which the host TLB management does not expect.
2033 	 */
2034 	if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
2035 		lpcr &= ~LPCR_AIL;
2036 
2037 	/*
2038 	 * On POWER9, allow userspace to enable large decrementer for the
2039 	 * guest, whether or not the host has it enabled.
2040 	 */
2041 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
2042 		lpcr &= ~LPCR_LD;
2043 
2044 	return lpcr;
2045 }
2046 
2047 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
2048 {
2049 	if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
2050 		WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
2051 			  lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
2052 	}
2053 }
2054 
2055 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
2056 		bool preserve_top32)
2057 {
2058 	struct kvm *kvm = vcpu->kvm;
2059 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
2060 	u64 mask;
2061 
2062 	spin_lock(&vc->lock);
2063 
2064 	/*
2065 	 * Userspace can only modify
2066 	 * DPFD (default prefetch depth), ILE (interrupt little-endian),
2067 	 * TC (translation control), AIL (alternate interrupt location),
2068 	 * LD (large decrementer).
2069 	 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
2070 	 */
2071 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
2072 
2073 	/* Broken 32-bit version of LPCR must not clear top bits */
2074 	if (preserve_top32)
2075 		mask &= 0xFFFFFFFF;
2076 
2077 	new_lpcr = kvmppc_filter_lpcr_hv(kvm,
2078 			(vc->lpcr & ~mask) | (new_lpcr & mask));
2079 
2080 	/*
2081 	 * If ILE (interrupt little-endian) has changed, update the
2082 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
2083 	 */
2084 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
2085 		struct kvm_vcpu *vcpu;
2086 		unsigned long i;
2087 
2088 		kvm_for_each_vcpu(i, vcpu, kvm) {
2089 			if (vcpu->arch.vcore != vc)
2090 				continue;
2091 			if (new_lpcr & LPCR_ILE)
2092 				vcpu->arch.intr_msr |= MSR_LE;
2093 			else
2094 				vcpu->arch.intr_msr &= ~MSR_LE;
2095 		}
2096 	}
2097 
2098 	vc->lpcr = new_lpcr;
2099 
2100 	spin_unlock(&vc->lock);
2101 }
2102 
2103 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2104 				 union kvmppc_one_reg *val)
2105 {
2106 	int r = 0;
2107 	long int i;
2108 
2109 	switch (id) {
2110 	case KVM_REG_PPC_DEBUG_INST:
2111 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2112 		break;
2113 	case KVM_REG_PPC_HIOR:
2114 		*val = get_reg_val(id, 0);
2115 		break;
2116 	case KVM_REG_PPC_DABR:
2117 		*val = get_reg_val(id, vcpu->arch.dabr);
2118 		break;
2119 	case KVM_REG_PPC_DABRX:
2120 		*val = get_reg_val(id, vcpu->arch.dabrx);
2121 		break;
2122 	case KVM_REG_PPC_DSCR:
2123 		*val = get_reg_val(id, vcpu->arch.dscr);
2124 		break;
2125 	case KVM_REG_PPC_PURR:
2126 		*val = get_reg_val(id, vcpu->arch.purr);
2127 		break;
2128 	case KVM_REG_PPC_SPURR:
2129 		*val = get_reg_val(id, vcpu->arch.spurr);
2130 		break;
2131 	case KVM_REG_PPC_AMR:
2132 		*val = get_reg_val(id, vcpu->arch.amr);
2133 		break;
2134 	case KVM_REG_PPC_UAMOR:
2135 		*val = get_reg_val(id, vcpu->arch.uamor);
2136 		break;
2137 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2138 		i = id - KVM_REG_PPC_MMCR0;
2139 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
2140 		break;
2141 	case KVM_REG_PPC_MMCR2:
2142 		*val = get_reg_val(id, vcpu->arch.mmcr[2]);
2143 		break;
2144 	case KVM_REG_PPC_MMCRA:
2145 		*val = get_reg_val(id, vcpu->arch.mmcra);
2146 		break;
2147 	case KVM_REG_PPC_MMCRS:
2148 		*val = get_reg_val(id, vcpu->arch.mmcrs);
2149 		break;
2150 	case KVM_REG_PPC_MMCR3:
2151 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
2152 		break;
2153 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2154 		i = id - KVM_REG_PPC_PMC1;
2155 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
2156 		break;
2157 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2158 		i = id - KVM_REG_PPC_SPMC1;
2159 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
2160 		break;
2161 	case KVM_REG_PPC_SIAR:
2162 		*val = get_reg_val(id, vcpu->arch.siar);
2163 		break;
2164 	case KVM_REG_PPC_SDAR:
2165 		*val = get_reg_val(id, vcpu->arch.sdar);
2166 		break;
2167 	case KVM_REG_PPC_SIER:
2168 		*val = get_reg_val(id, vcpu->arch.sier[0]);
2169 		break;
2170 	case KVM_REG_PPC_SIER2:
2171 		*val = get_reg_val(id, vcpu->arch.sier[1]);
2172 		break;
2173 	case KVM_REG_PPC_SIER3:
2174 		*val = get_reg_val(id, vcpu->arch.sier[2]);
2175 		break;
2176 	case KVM_REG_PPC_IAMR:
2177 		*val = get_reg_val(id, vcpu->arch.iamr);
2178 		break;
2179 	case KVM_REG_PPC_PSPB:
2180 		*val = get_reg_val(id, vcpu->arch.pspb);
2181 		break;
2182 	case KVM_REG_PPC_DPDES:
2183 		/*
2184 		 * On POWER9, where we are emulating msgsndp etc.,
2185 		 * we return 1 bit for each vcpu, which can come from
2186 		 * either vcore->dpdes or doorbell_request.
2187 		 * On POWER8, doorbell_request is 0.
2188 		 */
2189 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2190 			*val = get_reg_val(id, vcpu->arch.doorbell_request);
2191 		else
2192 			*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
2193 		break;
2194 	case KVM_REG_PPC_VTB:
2195 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
2196 		break;
2197 	case KVM_REG_PPC_DAWR:
2198 		*val = get_reg_val(id, vcpu->arch.dawr0);
2199 		break;
2200 	case KVM_REG_PPC_DAWRX:
2201 		*val = get_reg_val(id, vcpu->arch.dawrx0);
2202 		break;
2203 	case KVM_REG_PPC_DAWR1:
2204 		*val = get_reg_val(id, vcpu->arch.dawr1);
2205 		break;
2206 	case KVM_REG_PPC_DAWRX1:
2207 		*val = get_reg_val(id, vcpu->arch.dawrx1);
2208 		break;
2209 	case KVM_REG_PPC_CIABR:
2210 		*val = get_reg_val(id, vcpu->arch.ciabr);
2211 		break;
2212 	case KVM_REG_PPC_CSIGR:
2213 		*val = get_reg_val(id, vcpu->arch.csigr);
2214 		break;
2215 	case KVM_REG_PPC_TACR:
2216 		*val = get_reg_val(id, vcpu->arch.tacr);
2217 		break;
2218 	case KVM_REG_PPC_TCSCR:
2219 		*val = get_reg_val(id, vcpu->arch.tcscr);
2220 		break;
2221 	case KVM_REG_PPC_PID:
2222 		*val = get_reg_val(id, vcpu->arch.pid);
2223 		break;
2224 	case KVM_REG_PPC_ACOP:
2225 		*val = get_reg_val(id, vcpu->arch.acop);
2226 		break;
2227 	case KVM_REG_PPC_WORT:
2228 		*val = get_reg_val(id, vcpu->arch.wort);
2229 		break;
2230 	case KVM_REG_PPC_TIDR:
2231 		*val = get_reg_val(id, vcpu->arch.tid);
2232 		break;
2233 	case KVM_REG_PPC_PSSCR:
2234 		*val = get_reg_val(id, vcpu->arch.psscr);
2235 		break;
2236 	case KVM_REG_PPC_VPA_ADDR:
2237 		spin_lock(&vcpu->arch.vpa_update_lock);
2238 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2239 		spin_unlock(&vcpu->arch.vpa_update_lock);
2240 		break;
2241 	case KVM_REG_PPC_VPA_SLB:
2242 		spin_lock(&vcpu->arch.vpa_update_lock);
2243 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2244 		val->vpaval.length = vcpu->arch.slb_shadow.len;
2245 		spin_unlock(&vcpu->arch.vpa_update_lock);
2246 		break;
2247 	case KVM_REG_PPC_VPA_DTL:
2248 		spin_lock(&vcpu->arch.vpa_update_lock);
2249 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2250 		val->vpaval.length = vcpu->arch.dtl.len;
2251 		spin_unlock(&vcpu->arch.vpa_update_lock);
2252 		break;
2253 	case KVM_REG_PPC_TB_OFFSET:
2254 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
2255 		break;
2256 	case KVM_REG_PPC_LPCR:
2257 	case KVM_REG_PPC_LPCR_64:
2258 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
2259 		break;
2260 	case KVM_REG_PPC_PPR:
2261 		*val = get_reg_val(id, vcpu->arch.ppr);
2262 		break;
2263 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2264 	case KVM_REG_PPC_TFHAR:
2265 		*val = get_reg_val(id, vcpu->arch.tfhar);
2266 		break;
2267 	case KVM_REG_PPC_TFIAR:
2268 		*val = get_reg_val(id, vcpu->arch.tfiar);
2269 		break;
2270 	case KVM_REG_PPC_TEXASR:
2271 		*val = get_reg_val(id, vcpu->arch.texasr);
2272 		break;
2273 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2274 		i = id - KVM_REG_PPC_TM_GPR0;
2275 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2276 		break;
2277 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2278 	{
2279 		int j;
2280 		i = id - KVM_REG_PPC_TM_VSR0;
2281 		if (i < 32)
2282 			for (j = 0; j < TS_FPRWIDTH; j++)
2283 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2284 		else {
2285 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2286 				val->vval = vcpu->arch.vr_tm.vr[i-32];
2287 			else
2288 				r = -ENXIO;
2289 		}
2290 		break;
2291 	}
2292 	case KVM_REG_PPC_TM_CR:
2293 		*val = get_reg_val(id, vcpu->arch.cr_tm);
2294 		break;
2295 	case KVM_REG_PPC_TM_XER:
2296 		*val = get_reg_val(id, vcpu->arch.xer_tm);
2297 		break;
2298 	case KVM_REG_PPC_TM_LR:
2299 		*val = get_reg_val(id, vcpu->arch.lr_tm);
2300 		break;
2301 	case KVM_REG_PPC_TM_CTR:
2302 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
2303 		break;
2304 	case KVM_REG_PPC_TM_FPSCR:
2305 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2306 		break;
2307 	case KVM_REG_PPC_TM_AMR:
2308 		*val = get_reg_val(id, vcpu->arch.amr_tm);
2309 		break;
2310 	case KVM_REG_PPC_TM_PPR:
2311 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
2312 		break;
2313 	case KVM_REG_PPC_TM_VRSAVE:
2314 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
2315 		break;
2316 	case KVM_REG_PPC_TM_VSCR:
2317 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2318 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2319 		else
2320 			r = -ENXIO;
2321 		break;
2322 	case KVM_REG_PPC_TM_DSCR:
2323 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
2324 		break;
2325 	case KVM_REG_PPC_TM_TAR:
2326 		*val = get_reg_val(id, vcpu->arch.tar_tm);
2327 		break;
2328 #endif
2329 	case KVM_REG_PPC_ARCH_COMPAT:
2330 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
2331 		break;
2332 	case KVM_REG_PPC_DEC_EXPIRY:
2333 		*val = get_reg_val(id, vcpu->arch.dec_expires);
2334 		break;
2335 	case KVM_REG_PPC_ONLINE:
2336 		*val = get_reg_val(id, vcpu->arch.online);
2337 		break;
2338 	case KVM_REG_PPC_PTCR:
2339 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2340 		break;
2341 	default:
2342 		r = -EINVAL;
2343 		break;
2344 	}
2345 
2346 	return r;
2347 }
2348 
2349 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2350 				 union kvmppc_one_reg *val)
2351 {
2352 	int r = 0;
2353 	long int i;
2354 	unsigned long addr, len;
2355 
2356 	switch (id) {
2357 	case KVM_REG_PPC_HIOR:
2358 		/* Only allow this to be set to zero */
2359 		if (set_reg_val(id, *val))
2360 			r = -EINVAL;
2361 		break;
2362 	case KVM_REG_PPC_DABR:
2363 		vcpu->arch.dabr = set_reg_val(id, *val);
2364 		break;
2365 	case KVM_REG_PPC_DABRX:
2366 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2367 		break;
2368 	case KVM_REG_PPC_DSCR:
2369 		vcpu->arch.dscr = set_reg_val(id, *val);
2370 		break;
2371 	case KVM_REG_PPC_PURR:
2372 		vcpu->arch.purr = set_reg_val(id, *val);
2373 		break;
2374 	case KVM_REG_PPC_SPURR:
2375 		vcpu->arch.spurr = set_reg_val(id, *val);
2376 		break;
2377 	case KVM_REG_PPC_AMR:
2378 		vcpu->arch.amr = set_reg_val(id, *val);
2379 		break;
2380 	case KVM_REG_PPC_UAMOR:
2381 		vcpu->arch.uamor = set_reg_val(id, *val);
2382 		break;
2383 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2384 		i = id - KVM_REG_PPC_MMCR0;
2385 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2386 		break;
2387 	case KVM_REG_PPC_MMCR2:
2388 		vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2389 		break;
2390 	case KVM_REG_PPC_MMCRA:
2391 		vcpu->arch.mmcra = set_reg_val(id, *val);
2392 		break;
2393 	case KVM_REG_PPC_MMCRS:
2394 		vcpu->arch.mmcrs = set_reg_val(id, *val);
2395 		break;
2396 	case KVM_REG_PPC_MMCR3:
2397 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
2398 		break;
2399 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2400 		i = id - KVM_REG_PPC_PMC1;
2401 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
2402 		break;
2403 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2404 		i = id - KVM_REG_PPC_SPMC1;
2405 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
2406 		break;
2407 	case KVM_REG_PPC_SIAR:
2408 		vcpu->arch.siar = set_reg_val(id, *val);
2409 		break;
2410 	case KVM_REG_PPC_SDAR:
2411 		vcpu->arch.sdar = set_reg_val(id, *val);
2412 		break;
2413 	case KVM_REG_PPC_SIER:
2414 		vcpu->arch.sier[0] = set_reg_val(id, *val);
2415 		break;
2416 	case KVM_REG_PPC_SIER2:
2417 		vcpu->arch.sier[1] = set_reg_val(id, *val);
2418 		break;
2419 	case KVM_REG_PPC_SIER3:
2420 		vcpu->arch.sier[2] = set_reg_val(id, *val);
2421 		break;
2422 	case KVM_REG_PPC_IAMR:
2423 		vcpu->arch.iamr = set_reg_val(id, *val);
2424 		break;
2425 	case KVM_REG_PPC_PSPB:
2426 		vcpu->arch.pspb = set_reg_val(id, *val);
2427 		break;
2428 	case KVM_REG_PPC_DPDES:
2429 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2430 			vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1;
2431 		else
2432 			vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2433 		break;
2434 	case KVM_REG_PPC_VTB:
2435 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2436 		break;
2437 	case KVM_REG_PPC_DAWR:
2438 		vcpu->arch.dawr0 = set_reg_val(id, *val);
2439 		break;
2440 	case KVM_REG_PPC_DAWRX:
2441 		vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2442 		break;
2443 	case KVM_REG_PPC_DAWR1:
2444 		vcpu->arch.dawr1 = set_reg_val(id, *val);
2445 		break;
2446 	case KVM_REG_PPC_DAWRX1:
2447 		vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2448 		break;
2449 	case KVM_REG_PPC_CIABR:
2450 		vcpu->arch.ciabr = set_reg_val(id, *val);
2451 		/* Don't allow setting breakpoints in hypervisor code */
2452 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2453 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
2454 		break;
2455 	case KVM_REG_PPC_CSIGR:
2456 		vcpu->arch.csigr = set_reg_val(id, *val);
2457 		break;
2458 	case KVM_REG_PPC_TACR:
2459 		vcpu->arch.tacr = set_reg_val(id, *val);
2460 		break;
2461 	case KVM_REG_PPC_TCSCR:
2462 		vcpu->arch.tcscr = set_reg_val(id, *val);
2463 		break;
2464 	case KVM_REG_PPC_PID:
2465 		vcpu->arch.pid = set_reg_val(id, *val);
2466 		break;
2467 	case KVM_REG_PPC_ACOP:
2468 		vcpu->arch.acop = set_reg_val(id, *val);
2469 		break;
2470 	case KVM_REG_PPC_WORT:
2471 		vcpu->arch.wort = set_reg_val(id, *val);
2472 		break;
2473 	case KVM_REG_PPC_TIDR:
2474 		vcpu->arch.tid = set_reg_val(id, *val);
2475 		break;
2476 	case KVM_REG_PPC_PSSCR:
2477 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2478 		break;
2479 	case KVM_REG_PPC_VPA_ADDR:
2480 		addr = set_reg_val(id, *val);
2481 		r = -EINVAL;
2482 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2483 			      vcpu->arch.dtl.next_gpa))
2484 			break;
2485 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2486 		break;
2487 	case KVM_REG_PPC_VPA_SLB:
2488 		addr = val->vpaval.addr;
2489 		len = val->vpaval.length;
2490 		r = -EINVAL;
2491 		if (addr && !vcpu->arch.vpa.next_gpa)
2492 			break;
2493 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2494 		break;
2495 	case KVM_REG_PPC_VPA_DTL:
2496 		addr = val->vpaval.addr;
2497 		len = val->vpaval.length;
2498 		r = -EINVAL;
2499 		if (addr && (len < sizeof(struct dtl_entry) ||
2500 			     !vcpu->arch.vpa.next_gpa))
2501 			break;
2502 		len -= len % sizeof(struct dtl_entry);
2503 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2504 		break;
2505 	case KVM_REG_PPC_TB_OFFSET:
2506 		/* round up to multiple of 2^24 */
2507 		vcpu->arch.vcore->tb_offset =
2508 			ALIGN(set_reg_val(id, *val), 1UL << 24);
2509 		break;
2510 	case KVM_REG_PPC_LPCR:
2511 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2512 		break;
2513 	case KVM_REG_PPC_LPCR_64:
2514 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2515 		break;
2516 	case KVM_REG_PPC_PPR:
2517 		vcpu->arch.ppr = set_reg_val(id, *val);
2518 		break;
2519 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2520 	case KVM_REG_PPC_TFHAR:
2521 		vcpu->arch.tfhar = set_reg_val(id, *val);
2522 		break;
2523 	case KVM_REG_PPC_TFIAR:
2524 		vcpu->arch.tfiar = set_reg_val(id, *val);
2525 		break;
2526 	case KVM_REG_PPC_TEXASR:
2527 		vcpu->arch.texasr = set_reg_val(id, *val);
2528 		break;
2529 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2530 		i = id - KVM_REG_PPC_TM_GPR0;
2531 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2532 		break;
2533 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2534 	{
2535 		int j;
2536 		i = id - KVM_REG_PPC_TM_VSR0;
2537 		if (i < 32)
2538 			for (j = 0; j < TS_FPRWIDTH; j++)
2539 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2540 		else
2541 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2542 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2543 			else
2544 				r = -ENXIO;
2545 		break;
2546 	}
2547 	case KVM_REG_PPC_TM_CR:
2548 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2549 		break;
2550 	case KVM_REG_PPC_TM_XER:
2551 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2552 		break;
2553 	case KVM_REG_PPC_TM_LR:
2554 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2555 		break;
2556 	case KVM_REG_PPC_TM_CTR:
2557 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2558 		break;
2559 	case KVM_REG_PPC_TM_FPSCR:
2560 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2561 		break;
2562 	case KVM_REG_PPC_TM_AMR:
2563 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2564 		break;
2565 	case KVM_REG_PPC_TM_PPR:
2566 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2567 		break;
2568 	case KVM_REG_PPC_TM_VRSAVE:
2569 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2570 		break;
2571 	case KVM_REG_PPC_TM_VSCR:
2572 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2573 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2574 		else
2575 			r = - ENXIO;
2576 		break;
2577 	case KVM_REG_PPC_TM_DSCR:
2578 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2579 		break;
2580 	case KVM_REG_PPC_TM_TAR:
2581 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2582 		break;
2583 #endif
2584 	case KVM_REG_PPC_ARCH_COMPAT:
2585 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2586 		break;
2587 	case KVM_REG_PPC_DEC_EXPIRY:
2588 		vcpu->arch.dec_expires = set_reg_val(id, *val);
2589 		break;
2590 	case KVM_REG_PPC_ONLINE:
2591 		i = set_reg_val(id, *val);
2592 		if (i && !vcpu->arch.online)
2593 			atomic_inc(&vcpu->arch.vcore->online_count);
2594 		else if (!i && vcpu->arch.online)
2595 			atomic_dec(&vcpu->arch.vcore->online_count);
2596 		vcpu->arch.online = i;
2597 		break;
2598 	case KVM_REG_PPC_PTCR:
2599 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2600 		break;
2601 	default:
2602 		r = -EINVAL;
2603 		break;
2604 	}
2605 
2606 	return r;
2607 }
2608 
2609 /*
2610  * On POWER9, threads are independent and can be in different partitions.
2611  * Therefore we consider each thread to be a subcore.
2612  * There is a restriction that all threads have to be in the same
2613  * MMU mode (radix or HPT), unfortunately, but since we only support
2614  * HPT guests on a HPT host so far, that isn't an impediment yet.
2615  */
2616 static int threads_per_vcore(struct kvm *kvm)
2617 {
2618 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2619 		return 1;
2620 	return threads_per_subcore;
2621 }
2622 
2623 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2624 {
2625 	struct kvmppc_vcore *vcore;
2626 
2627 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2628 
2629 	if (vcore == NULL)
2630 		return NULL;
2631 
2632 	spin_lock_init(&vcore->lock);
2633 	spin_lock_init(&vcore->stoltb_lock);
2634 	rcuwait_init(&vcore->wait);
2635 	vcore->preempt_tb = TB_NIL;
2636 	vcore->lpcr = kvm->arch.lpcr;
2637 	vcore->first_vcpuid = id;
2638 	vcore->kvm = kvm;
2639 	INIT_LIST_HEAD(&vcore->preempt_list);
2640 
2641 	return vcore;
2642 }
2643 
2644 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2645 static struct debugfs_timings_element {
2646 	const char *name;
2647 	size_t offset;
2648 } timings[] = {
2649 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2650 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2651 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2652 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2653 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2654 };
2655 
2656 #define N_TIMINGS	(ARRAY_SIZE(timings))
2657 
2658 struct debugfs_timings_state {
2659 	struct kvm_vcpu	*vcpu;
2660 	unsigned int	buflen;
2661 	char		buf[N_TIMINGS * 100];
2662 };
2663 
2664 static int debugfs_timings_open(struct inode *inode, struct file *file)
2665 {
2666 	struct kvm_vcpu *vcpu = inode->i_private;
2667 	struct debugfs_timings_state *p;
2668 
2669 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2670 	if (!p)
2671 		return -ENOMEM;
2672 
2673 	kvm_get_kvm(vcpu->kvm);
2674 	p->vcpu = vcpu;
2675 	file->private_data = p;
2676 
2677 	return nonseekable_open(inode, file);
2678 }
2679 
2680 static int debugfs_timings_release(struct inode *inode, struct file *file)
2681 {
2682 	struct debugfs_timings_state *p = file->private_data;
2683 
2684 	kvm_put_kvm(p->vcpu->kvm);
2685 	kfree(p);
2686 	return 0;
2687 }
2688 
2689 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2690 				    size_t len, loff_t *ppos)
2691 {
2692 	struct debugfs_timings_state *p = file->private_data;
2693 	struct kvm_vcpu *vcpu = p->vcpu;
2694 	char *s, *buf_end;
2695 	struct kvmhv_tb_accumulator tb;
2696 	u64 count;
2697 	loff_t pos;
2698 	ssize_t n;
2699 	int i, loops;
2700 	bool ok;
2701 
2702 	if (!p->buflen) {
2703 		s = p->buf;
2704 		buf_end = s + sizeof(p->buf);
2705 		for (i = 0; i < N_TIMINGS; ++i) {
2706 			struct kvmhv_tb_accumulator *acc;
2707 
2708 			acc = (struct kvmhv_tb_accumulator *)
2709 				((unsigned long)vcpu + timings[i].offset);
2710 			ok = false;
2711 			for (loops = 0; loops < 1000; ++loops) {
2712 				count = acc->seqcount;
2713 				if (!(count & 1)) {
2714 					smp_rmb();
2715 					tb = *acc;
2716 					smp_rmb();
2717 					if (count == acc->seqcount) {
2718 						ok = true;
2719 						break;
2720 					}
2721 				}
2722 				udelay(1);
2723 			}
2724 			if (!ok)
2725 				snprintf(s, buf_end - s, "%s: stuck\n",
2726 					timings[i].name);
2727 			else
2728 				snprintf(s, buf_end - s,
2729 					"%s: %llu %llu %llu %llu\n",
2730 					timings[i].name, count / 2,
2731 					tb_to_ns(tb.tb_total),
2732 					tb_to_ns(tb.tb_min),
2733 					tb_to_ns(tb.tb_max));
2734 			s += strlen(s);
2735 		}
2736 		p->buflen = s - p->buf;
2737 	}
2738 
2739 	pos = *ppos;
2740 	if (pos >= p->buflen)
2741 		return 0;
2742 	if (len > p->buflen - pos)
2743 		len = p->buflen - pos;
2744 	n = copy_to_user(buf, p->buf + pos, len);
2745 	if (n) {
2746 		if (n == len)
2747 			return -EFAULT;
2748 		len -= n;
2749 	}
2750 	*ppos = pos + len;
2751 	return len;
2752 }
2753 
2754 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2755 				     size_t len, loff_t *ppos)
2756 {
2757 	return -EACCES;
2758 }
2759 
2760 static const struct file_operations debugfs_timings_ops = {
2761 	.owner	 = THIS_MODULE,
2762 	.open	 = debugfs_timings_open,
2763 	.release = debugfs_timings_release,
2764 	.read	 = debugfs_timings_read,
2765 	.write	 = debugfs_timings_write,
2766 	.llseek	 = generic_file_llseek,
2767 };
2768 
2769 /* Create a debugfs directory for the vcpu */
2770 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2771 {
2772 	char buf[16];
2773 	struct kvm *kvm = vcpu->kvm;
2774 
2775 	snprintf(buf, sizeof(buf), "vcpu%u", id);
2776 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2777 	debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2778 			    &debugfs_timings_ops);
2779 }
2780 
2781 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2782 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2783 {
2784 }
2785 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2786 
2787 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2788 {
2789 	int err;
2790 	int core;
2791 	struct kvmppc_vcore *vcore;
2792 	struct kvm *kvm;
2793 	unsigned int id;
2794 
2795 	kvm = vcpu->kvm;
2796 	id = vcpu->vcpu_id;
2797 
2798 	vcpu->arch.shared = &vcpu->arch.shregs;
2799 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2800 	/*
2801 	 * The shared struct is never shared on HV,
2802 	 * so we can always use host endianness
2803 	 */
2804 #ifdef __BIG_ENDIAN__
2805 	vcpu->arch.shared_big_endian = true;
2806 #else
2807 	vcpu->arch.shared_big_endian = false;
2808 #endif
2809 #endif
2810 	vcpu->arch.mmcr[0] = MMCR0_FC;
2811 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
2812 		vcpu->arch.mmcr[0] |= MMCR0_PMCCEXT;
2813 		vcpu->arch.mmcra = MMCRA_BHRB_DISABLE;
2814 	}
2815 
2816 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2817 	/* default to host PVR, since we can't spoof it */
2818 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2819 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2820 	spin_lock_init(&vcpu->arch.tbacct_lock);
2821 	vcpu->arch.busy_preempt = TB_NIL;
2822 	vcpu->arch.shregs.msr = MSR_ME;
2823 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2824 
2825 	/*
2826 	 * Set the default HFSCR for the guest from the host value.
2827 	 * This value is only used on POWER9.
2828 	 * On POWER9, we want to virtualize the doorbell facility, so we
2829 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2830 	 * to trap and then we emulate them.
2831 	 */
2832 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2833 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2834 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2835 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2836 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2837 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2838 			vcpu->arch.hfscr |= HFSCR_TM;
2839 #endif
2840 	}
2841 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2842 		vcpu->arch.hfscr |= HFSCR_TM;
2843 
2844 	vcpu->arch.hfscr_permitted = vcpu->arch.hfscr;
2845 
2846 	/*
2847 	 * PM, EBB, TM are demand-faulted so start with it clear.
2848 	 */
2849 	vcpu->arch.hfscr &= ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM);
2850 
2851 	kvmppc_mmu_book3s_hv_init(vcpu);
2852 
2853 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2854 
2855 	init_waitqueue_head(&vcpu->arch.cpu_run);
2856 
2857 	mutex_lock(&kvm->lock);
2858 	vcore = NULL;
2859 	err = -EINVAL;
2860 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2861 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2862 			pr_devel("KVM: VCPU ID too high\n");
2863 			core = KVM_MAX_VCORES;
2864 		} else {
2865 			BUG_ON(kvm->arch.smt_mode != 1);
2866 			core = kvmppc_pack_vcpu_id(kvm, id);
2867 		}
2868 	} else {
2869 		core = id / kvm->arch.smt_mode;
2870 	}
2871 	if (core < KVM_MAX_VCORES) {
2872 		vcore = kvm->arch.vcores[core];
2873 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2874 			pr_devel("KVM: collision on id %u", id);
2875 			vcore = NULL;
2876 		} else if (!vcore) {
2877 			/*
2878 			 * Take mmu_setup_lock for mutual exclusion
2879 			 * with kvmppc_update_lpcr().
2880 			 */
2881 			err = -ENOMEM;
2882 			vcore = kvmppc_vcore_create(kvm,
2883 					id & ~(kvm->arch.smt_mode - 1));
2884 			mutex_lock(&kvm->arch.mmu_setup_lock);
2885 			kvm->arch.vcores[core] = vcore;
2886 			kvm->arch.online_vcores++;
2887 			mutex_unlock(&kvm->arch.mmu_setup_lock);
2888 		}
2889 	}
2890 	mutex_unlock(&kvm->lock);
2891 
2892 	if (!vcore)
2893 		return err;
2894 
2895 	spin_lock(&vcore->lock);
2896 	++vcore->num_threads;
2897 	spin_unlock(&vcore->lock);
2898 	vcpu->arch.vcore = vcore;
2899 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2900 	vcpu->arch.thread_cpu = -1;
2901 	vcpu->arch.prev_cpu = -1;
2902 
2903 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2904 	kvmppc_sanity_check(vcpu);
2905 
2906 	debugfs_vcpu_init(vcpu, id);
2907 
2908 	return 0;
2909 }
2910 
2911 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2912 			      unsigned long flags)
2913 {
2914 	int err;
2915 	int esmt = 0;
2916 
2917 	if (flags)
2918 		return -EINVAL;
2919 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2920 		return -EINVAL;
2921 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2922 		/*
2923 		 * On POWER8 (or POWER7), the threading mode is "strict",
2924 		 * so we pack smt_mode vcpus per vcore.
2925 		 */
2926 		if (smt_mode > threads_per_subcore)
2927 			return -EINVAL;
2928 	} else {
2929 		/*
2930 		 * On POWER9, the threading mode is "loose",
2931 		 * so each vcpu gets its own vcore.
2932 		 */
2933 		esmt = smt_mode;
2934 		smt_mode = 1;
2935 	}
2936 	mutex_lock(&kvm->lock);
2937 	err = -EBUSY;
2938 	if (!kvm->arch.online_vcores) {
2939 		kvm->arch.smt_mode = smt_mode;
2940 		kvm->arch.emul_smt_mode = esmt;
2941 		err = 0;
2942 	}
2943 	mutex_unlock(&kvm->lock);
2944 
2945 	return err;
2946 }
2947 
2948 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2949 {
2950 	if (vpa->pinned_addr)
2951 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2952 					vpa->dirty);
2953 }
2954 
2955 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2956 {
2957 	spin_lock(&vcpu->arch.vpa_update_lock);
2958 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2959 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2960 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2961 	spin_unlock(&vcpu->arch.vpa_update_lock);
2962 }
2963 
2964 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2965 {
2966 	/* Indicate we want to get back into the guest */
2967 	return 1;
2968 }
2969 
2970 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2971 {
2972 	unsigned long dec_nsec, now;
2973 
2974 	now = get_tb();
2975 	if (now > kvmppc_dec_expires_host_tb(vcpu)) {
2976 		/* decrementer has already gone negative */
2977 		kvmppc_core_queue_dec(vcpu);
2978 		kvmppc_core_prepare_to_enter(vcpu);
2979 		return;
2980 	}
2981 	dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now);
2982 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2983 	vcpu->arch.timer_running = 1;
2984 }
2985 
2986 extern int __kvmppc_vcore_entry(void);
2987 
2988 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2989 				   struct kvm_vcpu *vcpu, u64 tb)
2990 {
2991 	u64 now;
2992 
2993 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2994 		return;
2995 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2996 	now = tb;
2997 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2998 		vcpu->arch.stolen_logged;
2999 	vcpu->arch.busy_preempt = now;
3000 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3001 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
3002 	--vc->n_runnable;
3003 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
3004 }
3005 
3006 static int kvmppc_grab_hwthread(int cpu)
3007 {
3008 	struct paca_struct *tpaca;
3009 	long timeout = 10000;
3010 
3011 	tpaca = paca_ptrs[cpu];
3012 
3013 	/* Ensure the thread won't go into the kernel if it wakes */
3014 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3015 	tpaca->kvm_hstate.kvm_vcore = NULL;
3016 	tpaca->kvm_hstate.napping = 0;
3017 	smp_wmb();
3018 	tpaca->kvm_hstate.hwthread_req = 1;
3019 
3020 	/*
3021 	 * If the thread is already executing in the kernel (e.g. handling
3022 	 * a stray interrupt), wait for it to get back to nap mode.
3023 	 * The smp_mb() is to ensure that our setting of hwthread_req
3024 	 * is visible before we look at hwthread_state, so if this
3025 	 * races with the code at system_reset_pSeries and the thread
3026 	 * misses our setting of hwthread_req, we are sure to see its
3027 	 * setting of hwthread_state, and vice versa.
3028 	 */
3029 	smp_mb();
3030 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
3031 		if (--timeout <= 0) {
3032 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
3033 			return -EBUSY;
3034 		}
3035 		udelay(1);
3036 	}
3037 	return 0;
3038 }
3039 
3040 static void kvmppc_release_hwthread(int cpu)
3041 {
3042 	struct paca_struct *tpaca;
3043 
3044 	tpaca = paca_ptrs[cpu];
3045 	tpaca->kvm_hstate.hwthread_req = 0;
3046 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3047 	tpaca->kvm_hstate.kvm_vcore = NULL;
3048 	tpaca->kvm_hstate.kvm_split_mode = NULL;
3049 }
3050 
3051 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest);
3052 
3053 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
3054 {
3055 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3056 	cpumask_t *need_tlb_flush;
3057 	int i;
3058 
3059 	if (nested)
3060 		need_tlb_flush = &nested->need_tlb_flush;
3061 	else
3062 		need_tlb_flush = &kvm->arch.need_tlb_flush;
3063 
3064 	cpu = cpu_first_tlb_thread_sibling(cpu);
3065 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3066 					i += cpu_tlb_thread_sibling_step())
3067 		cpumask_set_cpu(i, need_tlb_flush);
3068 
3069 	/*
3070 	 * Make sure setting of bit in need_tlb_flush precedes testing of
3071 	 * cpu_in_guest. The matching barrier on the other side is hwsync
3072 	 * when switching to guest MMU mode, which happens between
3073 	 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit
3074 	 * being tested.
3075 	 */
3076 	smp_mb();
3077 
3078 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3079 					i += cpu_tlb_thread_sibling_step()) {
3080 		struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i);
3081 
3082 		if (running == kvm)
3083 			smp_call_function_single(i, do_nothing, NULL, 1);
3084 	}
3085 }
3086 
3087 static void do_migrate_away_vcpu(void *arg)
3088 {
3089 	struct kvm_vcpu *vcpu = arg;
3090 	struct kvm *kvm = vcpu->kvm;
3091 
3092 	/*
3093 	 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
3094 	 * ptesync sequence on the old CPU before migrating to a new one, in
3095 	 * case we interrupted the guest between a tlbie ; eieio ;
3096 	 * tlbsync; ptesync sequence.
3097 	 *
3098 	 * Otherwise, ptesync is sufficient for ordering tlbiel sequences.
3099 	 */
3100 	if (kvm->arch.lpcr & LPCR_GTSE)
3101 		asm volatile("eieio; tlbsync; ptesync");
3102 	else
3103 		asm volatile("ptesync");
3104 }
3105 
3106 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
3107 {
3108 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3109 	struct kvm *kvm = vcpu->kvm;
3110 	int prev_cpu;
3111 
3112 	if (!cpu_has_feature(CPU_FTR_HVMODE))
3113 		return;
3114 
3115 	if (nested)
3116 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
3117 	else
3118 		prev_cpu = vcpu->arch.prev_cpu;
3119 
3120 	/*
3121 	 * With radix, the guest can do TLB invalidations itself,
3122 	 * and it could choose to use the local form (tlbiel) if
3123 	 * it is invalidating a translation that has only ever been
3124 	 * used on one vcpu.  However, that doesn't mean it has
3125 	 * only ever been used on one physical cpu, since vcpus
3126 	 * can move around between pcpus.  To cope with this, when
3127 	 * a vcpu moves from one pcpu to another, we need to tell
3128 	 * any vcpus running on the same core as this vcpu previously
3129 	 * ran to flush the TLB.
3130 	 */
3131 	if (prev_cpu != pcpu) {
3132 		if (prev_cpu >= 0) {
3133 			if (cpu_first_tlb_thread_sibling(prev_cpu) !=
3134 			    cpu_first_tlb_thread_sibling(pcpu))
3135 				radix_flush_cpu(kvm, prev_cpu, vcpu);
3136 
3137 			smp_call_function_single(prev_cpu,
3138 					do_migrate_away_vcpu, vcpu, 1);
3139 		}
3140 		if (nested)
3141 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3142 		else
3143 			vcpu->arch.prev_cpu = pcpu;
3144 	}
3145 }
3146 
3147 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3148 {
3149 	int cpu;
3150 	struct paca_struct *tpaca;
3151 
3152 	cpu = vc->pcpu;
3153 	if (vcpu) {
3154 		if (vcpu->arch.timer_running) {
3155 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3156 			vcpu->arch.timer_running = 0;
3157 		}
3158 		cpu += vcpu->arch.ptid;
3159 		vcpu->cpu = vc->pcpu;
3160 		vcpu->arch.thread_cpu = cpu;
3161 	}
3162 	tpaca = paca_ptrs[cpu];
3163 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
3164 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3165 	tpaca->kvm_hstate.fake_suspend = 0;
3166 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3167 	smp_wmb();
3168 	tpaca->kvm_hstate.kvm_vcore = vc;
3169 	if (cpu != smp_processor_id())
3170 		kvmppc_ipi_thread(cpu);
3171 }
3172 
3173 static void kvmppc_wait_for_nap(int n_threads)
3174 {
3175 	int cpu = smp_processor_id();
3176 	int i, loops;
3177 
3178 	if (n_threads <= 1)
3179 		return;
3180 	for (loops = 0; loops < 1000000; ++loops) {
3181 		/*
3182 		 * Check if all threads are finished.
3183 		 * We set the vcore pointer when starting a thread
3184 		 * and the thread clears it when finished, so we look
3185 		 * for any threads that still have a non-NULL vcore ptr.
3186 		 */
3187 		for (i = 1; i < n_threads; ++i)
3188 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3189 				break;
3190 		if (i == n_threads) {
3191 			HMT_medium();
3192 			return;
3193 		}
3194 		HMT_low();
3195 	}
3196 	HMT_medium();
3197 	for (i = 1; i < n_threads; ++i)
3198 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3199 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3200 }
3201 
3202 /*
3203  * Check that we are on thread 0 and that any other threads in
3204  * this core are off-line.  Then grab the threads so they can't
3205  * enter the kernel.
3206  */
3207 static int on_primary_thread(void)
3208 {
3209 	int cpu = smp_processor_id();
3210 	int thr;
3211 
3212 	/* Are we on a primary subcore? */
3213 	if (cpu_thread_in_subcore(cpu))
3214 		return 0;
3215 
3216 	thr = 0;
3217 	while (++thr < threads_per_subcore)
3218 		if (cpu_online(cpu + thr))
3219 			return 0;
3220 
3221 	/* Grab all hw threads so they can't go into the kernel */
3222 	for (thr = 1; thr < threads_per_subcore; ++thr) {
3223 		if (kvmppc_grab_hwthread(cpu + thr)) {
3224 			/* Couldn't grab one; let the others go */
3225 			do {
3226 				kvmppc_release_hwthread(cpu + thr);
3227 			} while (--thr > 0);
3228 			return 0;
3229 		}
3230 	}
3231 	return 1;
3232 }
3233 
3234 /*
3235  * A list of virtual cores for each physical CPU.
3236  * These are vcores that could run but their runner VCPU tasks are
3237  * (or may be) preempted.
3238  */
3239 struct preempted_vcore_list {
3240 	struct list_head	list;
3241 	spinlock_t		lock;
3242 };
3243 
3244 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3245 
3246 static void init_vcore_lists(void)
3247 {
3248 	int cpu;
3249 
3250 	for_each_possible_cpu(cpu) {
3251 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3252 		spin_lock_init(&lp->lock);
3253 		INIT_LIST_HEAD(&lp->list);
3254 	}
3255 }
3256 
3257 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3258 {
3259 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3260 
3261 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3262 
3263 	vc->vcore_state = VCORE_PREEMPT;
3264 	vc->pcpu = smp_processor_id();
3265 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3266 		spin_lock(&lp->lock);
3267 		list_add_tail(&vc->preempt_list, &lp->list);
3268 		spin_unlock(&lp->lock);
3269 	}
3270 
3271 	/* Start accumulating stolen time */
3272 	kvmppc_core_start_stolen(vc, mftb());
3273 }
3274 
3275 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3276 {
3277 	struct preempted_vcore_list *lp;
3278 
3279 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3280 
3281 	kvmppc_core_end_stolen(vc, mftb());
3282 	if (!list_empty(&vc->preempt_list)) {
3283 		lp = &per_cpu(preempted_vcores, vc->pcpu);
3284 		spin_lock(&lp->lock);
3285 		list_del_init(&vc->preempt_list);
3286 		spin_unlock(&lp->lock);
3287 	}
3288 	vc->vcore_state = VCORE_INACTIVE;
3289 }
3290 
3291 /*
3292  * This stores information about the virtual cores currently
3293  * assigned to a physical core.
3294  */
3295 struct core_info {
3296 	int		n_subcores;
3297 	int		max_subcore_threads;
3298 	int		total_threads;
3299 	int		subcore_threads[MAX_SUBCORES];
3300 	struct kvmppc_vcore *vc[MAX_SUBCORES];
3301 };
3302 
3303 /*
3304  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3305  * respectively in 2-way micro-threading (split-core) mode on POWER8.
3306  */
3307 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3308 
3309 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3310 {
3311 	memset(cip, 0, sizeof(*cip));
3312 	cip->n_subcores = 1;
3313 	cip->max_subcore_threads = vc->num_threads;
3314 	cip->total_threads = vc->num_threads;
3315 	cip->subcore_threads[0] = vc->num_threads;
3316 	cip->vc[0] = vc;
3317 }
3318 
3319 static bool subcore_config_ok(int n_subcores, int n_threads)
3320 {
3321 	/*
3322 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3323 	 * split-core mode, with one thread per subcore.
3324 	 */
3325 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3326 		return n_subcores <= 4 && n_threads == 1;
3327 
3328 	/* On POWER8, can only dynamically split if unsplit to begin with */
3329 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3330 		return false;
3331 	if (n_subcores > MAX_SUBCORES)
3332 		return false;
3333 	if (n_subcores > 1) {
3334 		if (!(dynamic_mt_modes & 2))
3335 			n_subcores = 4;
3336 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3337 			return false;
3338 	}
3339 
3340 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3341 }
3342 
3343 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3344 {
3345 	vc->entry_exit_map = 0;
3346 	vc->in_guest = 0;
3347 	vc->napping_threads = 0;
3348 	vc->conferring_threads = 0;
3349 	vc->tb_offset_applied = 0;
3350 }
3351 
3352 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3353 {
3354 	int n_threads = vc->num_threads;
3355 	int sub;
3356 
3357 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3358 		return false;
3359 
3360 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
3361 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3362 		return false;
3363 
3364 	if (n_threads < cip->max_subcore_threads)
3365 		n_threads = cip->max_subcore_threads;
3366 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3367 		return false;
3368 	cip->max_subcore_threads = n_threads;
3369 
3370 	sub = cip->n_subcores;
3371 	++cip->n_subcores;
3372 	cip->total_threads += vc->num_threads;
3373 	cip->subcore_threads[sub] = vc->num_threads;
3374 	cip->vc[sub] = vc;
3375 	init_vcore_to_run(vc);
3376 	list_del_init(&vc->preempt_list);
3377 
3378 	return true;
3379 }
3380 
3381 /*
3382  * Work out whether it is possible to piggyback the execution of
3383  * vcore *pvc onto the execution of the other vcores described in *cip.
3384  */
3385 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3386 			  int target_threads)
3387 {
3388 	if (cip->total_threads + pvc->num_threads > target_threads)
3389 		return false;
3390 
3391 	return can_dynamic_split(pvc, cip);
3392 }
3393 
3394 static void prepare_threads(struct kvmppc_vcore *vc)
3395 {
3396 	int i;
3397 	struct kvm_vcpu *vcpu;
3398 
3399 	for_each_runnable_thread(i, vcpu, vc) {
3400 		if (signal_pending(vcpu->arch.run_task))
3401 			vcpu->arch.ret = -EINTR;
3402 		else if (vcpu->arch.vpa.update_pending ||
3403 			 vcpu->arch.slb_shadow.update_pending ||
3404 			 vcpu->arch.dtl.update_pending)
3405 			vcpu->arch.ret = RESUME_GUEST;
3406 		else
3407 			continue;
3408 		kvmppc_remove_runnable(vc, vcpu, mftb());
3409 		wake_up(&vcpu->arch.cpu_run);
3410 	}
3411 }
3412 
3413 static void collect_piggybacks(struct core_info *cip, int target_threads)
3414 {
3415 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3416 	struct kvmppc_vcore *pvc, *vcnext;
3417 
3418 	spin_lock(&lp->lock);
3419 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3420 		if (!spin_trylock(&pvc->lock))
3421 			continue;
3422 		prepare_threads(pvc);
3423 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3424 			list_del_init(&pvc->preempt_list);
3425 			if (pvc->runner == NULL) {
3426 				pvc->vcore_state = VCORE_INACTIVE;
3427 				kvmppc_core_end_stolen(pvc, mftb());
3428 			}
3429 			spin_unlock(&pvc->lock);
3430 			continue;
3431 		}
3432 		if (!can_piggyback(pvc, cip, target_threads)) {
3433 			spin_unlock(&pvc->lock);
3434 			continue;
3435 		}
3436 		kvmppc_core_end_stolen(pvc, mftb());
3437 		pvc->vcore_state = VCORE_PIGGYBACK;
3438 		if (cip->total_threads >= target_threads)
3439 			break;
3440 	}
3441 	spin_unlock(&lp->lock);
3442 }
3443 
3444 static bool recheck_signals_and_mmu(struct core_info *cip)
3445 {
3446 	int sub, i;
3447 	struct kvm_vcpu *vcpu;
3448 	struct kvmppc_vcore *vc;
3449 
3450 	for (sub = 0; sub < cip->n_subcores; ++sub) {
3451 		vc = cip->vc[sub];
3452 		if (!vc->kvm->arch.mmu_ready)
3453 			return true;
3454 		for_each_runnable_thread(i, vcpu, vc)
3455 			if (signal_pending(vcpu->arch.run_task))
3456 				return true;
3457 	}
3458 	return false;
3459 }
3460 
3461 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3462 {
3463 	int still_running = 0, i;
3464 	u64 now;
3465 	long ret;
3466 	struct kvm_vcpu *vcpu;
3467 
3468 	spin_lock(&vc->lock);
3469 	now = get_tb();
3470 	for_each_runnable_thread(i, vcpu, vc) {
3471 		/*
3472 		 * It's safe to unlock the vcore in the loop here, because
3473 		 * for_each_runnable_thread() is safe against removal of
3474 		 * the vcpu, and the vcore state is VCORE_EXITING here,
3475 		 * so any vcpus becoming runnable will have their arch.trap
3476 		 * set to zero and can't actually run in the guest.
3477 		 */
3478 		spin_unlock(&vc->lock);
3479 		/* cancel pending dec exception if dec is positive */
3480 		if (now < kvmppc_dec_expires_host_tb(vcpu) &&
3481 		    kvmppc_core_pending_dec(vcpu))
3482 			kvmppc_core_dequeue_dec(vcpu);
3483 
3484 		trace_kvm_guest_exit(vcpu);
3485 
3486 		ret = RESUME_GUEST;
3487 		if (vcpu->arch.trap)
3488 			ret = kvmppc_handle_exit_hv(vcpu,
3489 						    vcpu->arch.run_task);
3490 
3491 		vcpu->arch.ret = ret;
3492 		vcpu->arch.trap = 0;
3493 
3494 		spin_lock(&vc->lock);
3495 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3496 			if (vcpu->arch.pending_exceptions)
3497 				kvmppc_core_prepare_to_enter(vcpu);
3498 			if (vcpu->arch.ceded)
3499 				kvmppc_set_timer(vcpu);
3500 			else
3501 				++still_running;
3502 		} else {
3503 			kvmppc_remove_runnable(vc, vcpu, mftb());
3504 			wake_up(&vcpu->arch.cpu_run);
3505 		}
3506 	}
3507 	if (!is_master) {
3508 		if (still_running > 0) {
3509 			kvmppc_vcore_preempt(vc);
3510 		} else if (vc->runner) {
3511 			vc->vcore_state = VCORE_PREEMPT;
3512 			kvmppc_core_start_stolen(vc, mftb());
3513 		} else {
3514 			vc->vcore_state = VCORE_INACTIVE;
3515 		}
3516 		if (vc->n_runnable > 0 && vc->runner == NULL) {
3517 			/* make sure there's a candidate runner awake */
3518 			i = -1;
3519 			vcpu = next_runnable_thread(vc, &i);
3520 			wake_up(&vcpu->arch.cpu_run);
3521 		}
3522 	}
3523 	spin_unlock(&vc->lock);
3524 }
3525 
3526 /*
3527  * Clear core from the list of active host cores as we are about to
3528  * enter the guest. Only do this if it is the primary thread of the
3529  * core (not if a subcore) that is entering the guest.
3530  */
3531 static inline int kvmppc_clear_host_core(unsigned int cpu)
3532 {
3533 	int core;
3534 
3535 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3536 		return 0;
3537 	/*
3538 	 * Memory barrier can be omitted here as we will do a smp_wmb()
3539 	 * later in kvmppc_start_thread and we need ensure that state is
3540 	 * visible to other CPUs only after we enter guest.
3541 	 */
3542 	core = cpu >> threads_shift;
3543 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3544 	return 0;
3545 }
3546 
3547 /*
3548  * Advertise this core as an active host core since we exited the guest
3549  * Only need to do this if it is the primary thread of the core that is
3550  * exiting.
3551  */
3552 static inline int kvmppc_set_host_core(unsigned int cpu)
3553 {
3554 	int core;
3555 
3556 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3557 		return 0;
3558 
3559 	/*
3560 	 * Memory barrier can be omitted here because we do a spin_unlock
3561 	 * immediately after this which provides the memory barrier.
3562 	 */
3563 	core = cpu >> threads_shift;
3564 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3565 	return 0;
3566 }
3567 
3568 static void set_irq_happened(int trap)
3569 {
3570 	switch (trap) {
3571 	case BOOK3S_INTERRUPT_EXTERNAL:
3572 		local_paca->irq_happened |= PACA_IRQ_EE;
3573 		break;
3574 	case BOOK3S_INTERRUPT_H_DOORBELL:
3575 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3576 		break;
3577 	case BOOK3S_INTERRUPT_HMI:
3578 		local_paca->irq_happened |= PACA_IRQ_HMI;
3579 		break;
3580 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3581 		replay_system_reset();
3582 		break;
3583 	}
3584 }
3585 
3586 /*
3587  * Run a set of guest threads on a physical core.
3588  * Called with vc->lock held.
3589  */
3590 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3591 {
3592 	struct kvm_vcpu *vcpu;
3593 	int i;
3594 	int srcu_idx;
3595 	struct core_info core_info;
3596 	struct kvmppc_vcore *pvc;
3597 	struct kvm_split_mode split_info, *sip;
3598 	int split, subcore_size, active;
3599 	int sub;
3600 	bool thr0_done;
3601 	unsigned long cmd_bit, stat_bit;
3602 	int pcpu, thr;
3603 	int target_threads;
3604 	int controlled_threads;
3605 	int trap;
3606 	bool is_power8;
3607 
3608 	if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3609 		return;
3610 
3611 	/*
3612 	 * Remove from the list any threads that have a signal pending
3613 	 * or need a VPA update done
3614 	 */
3615 	prepare_threads(vc);
3616 
3617 	/* if the runner is no longer runnable, let the caller pick a new one */
3618 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3619 		return;
3620 
3621 	/*
3622 	 * Initialize *vc.
3623 	 */
3624 	init_vcore_to_run(vc);
3625 	vc->preempt_tb = TB_NIL;
3626 
3627 	/*
3628 	 * Number of threads that we will be controlling: the same as
3629 	 * the number of threads per subcore, except on POWER9,
3630 	 * where it's 1 because the threads are (mostly) independent.
3631 	 */
3632 	controlled_threads = threads_per_vcore(vc->kvm);
3633 
3634 	/*
3635 	 * Make sure we are running on primary threads, and that secondary
3636 	 * threads are offline.  Also check if the number of threads in this
3637 	 * guest are greater than the current system threads per guest.
3638 	 */
3639 	if ((controlled_threads > 1) &&
3640 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3641 		for_each_runnable_thread(i, vcpu, vc) {
3642 			vcpu->arch.ret = -EBUSY;
3643 			kvmppc_remove_runnable(vc, vcpu, mftb());
3644 			wake_up(&vcpu->arch.cpu_run);
3645 		}
3646 		goto out;
3647 	}
3648 
3649 	/*
3650 	 * See if we could run any other vcores on the physical core
3651 	 * along with this one.
3652 	 */
3653 	init_core_info(&core_info, vc);
3654 	pcpu = smp_processor_id();
3655 	target_threads = controlled_threads;
3656 	if (target_smt_mode && target_smt_mode < target_threads)
3657 		target_threads = target_smt_mode;
3658 	if (vc->num_threads < target_threads)
3659 		collect_piggybacks(&core_info, target_threads);
3660 
3661 	/*
3662 	 * Hard-disable interrupts, and check resched flag and signals.
3663 	 * If we need to reschedule or deliver a signal, clean up
3664 	 * and return without going into the guest(s).
3665 	 * If the mmu_ready flag has been cleared, don't go into the
3666 	 * guest because that means a HPT resize operation is in progress.
3667 	 */
3668 	local_irq_disable();
3669 	hard_irq_disable();
3670 	if (lazy_irq_pending() || need_resched() ||
3671 	    recheck_signals_and_mmu(&core_info)) {
3672 		local_irq_enable();
3673 		vc->vcore_state = VCORE_INACTIVE;
3674 		/* Unlock all except the primary vcore */
3675 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3676 			pvc = core_info.vc[sub];
3677 			/* Put back on to the preempted vcores list */
3678 			kvmppc_vcore_preempt(pvc);
3679 			spin_unlock(&pvc->lock);
3680 		}
3681 		for (i = 0; i < controlled_threads; ++i)
3682 			kvmppc_release_hwthread(pcpu + i);
3683 		return;
3684 	}
3685 
3686 	kvmppc_clear_host_core(pcpu);
3687 
3688 	/* Decide on micro-threading (split-core) mode */
3689 	subcore_size = threads_per_subcore;
3690 	cmd_bit = stat_bit = 0;
3691 	split = core_info.n_subcores;
3692 	sip = NULL;
3693 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3694 
3695 	if (split > 1) {
3696 		sip = &split_info;
3697 		memset(&split_info, 0, sizeof(split_info));
3698 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3699 			split_info.vc[sub] = core_info.vc[sub];
3700 
3701 		if (is_power8) {
3702 			if (split == 2 && (dynamic_mt_modes & 2)) {
3703 				cmd_bit = HID0_POWER8_1TO2LPAR;
3704 				stat_bit = HID0_POWER8_2LPARMODE;
3705 			} else {
3706 				split = 4;
3707 				cmd_bit = HID0_POWER8_1TO4LPAR;
3708 				stat_bit = HID0_POWER8_4LPARMODE;
3709 			}
3710 			subcore_size = MAX_SMT_THREADS / split;
3711 			split_info.rpr = mfspr(SPRN_RPR);
3712 			split_info.pmmar = mfspr(SPRN_PMMAR);
3713 			split_info.ldbar = mfspr(SPRN_LDBAR);
3714 			split_info.subcore_size = subcore_size;
3715 		} else {
3716 			split_info.subcore_size = 1;
3717 		}
3718 
3719 		/* order writes to split_info before kvm_split_mode pointer */
3720 		smp_wmb();
3721 	}
3722 
3723 	for (thr = 0; thr < controlled_threads; ++thr) {
3724 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3725 
3726 		paca->kvm_hstate.napping = 0;
3727 		paca->kvm_hstate.kvm_split_mode = sip;
3728 	}
3729 
3730 	/* Initiate micro-threading (split-core) on POWER8 if required */
3731 	if (cmd_bit) {
3732 		unsigned long hid0 = mfspr(SPRN_HID0);
3733 
3734 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3735 		mb();
3736 		mtspr(SPRN_HID0, hid0);
3737 		isync();
3738 		for (;;) {
3739 			hid0 = mfspr(SPRN_HID0);
3740 			if (hid0 & stat_bit)
3741 				break;
3742 			cpu_relax();
3743 		}
3744 	}
3745 
3746 	/*
3747 	 * On POWER8, set RWMR register.
3748 	 * Since it only affects PURR and SPURR, it doesn't affect
3749 	 * the host, so we don't save/restore the host value.
3750 	 */
3751 	if (is_power8) {
3752 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3753 		int n_online = atomic_read(&vc->online_count);
3754 
3755 		/*
3756 		 * Use the 8-thread value if we're doing split-core
3757 		 * or if the vcore's online count looks bogus.
3758 		 */
3759 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3760 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3761 			rwmr_val = p8_rwmr_values[n_online];
3762 		mtspr(SPRN_RWMR, rwmr_val);
3763 	}
3764 
3765 	/* Start all the threads */
3766 	active = 0;
3767 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3768 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3769 		thr0_done = false;
3770 		active |= 1 << thr;
3771 		pvc = core_info.vc[sub];
3772 		pvc->pcpu = pcpu + thr;
3773 		for_each_runnable_thread(i, vcpu, pvc) {
3774 			kvmppc_start_thread(vcpu, pvc);
3775 			kvmppc_create_dtl_entry(vcpu, pvc);
3776 			trace_kvm_guest_enter(vcpu);
3777 			if (!vcpu->arch.ptid)
3778 				thr0_done = true;
3779 			active |= 1 << (thr + vcpu->arch.ptid);
3780 		}
3781 		/*
3782 		 * We need to start the first thread of each subcore
3783 		 * even if it doesn't have a vcpu.
3784 		 */
3785 		if (!thr0_done)
3786 			kvmppc_start_thread(NULL, pvc);
3787 	}
3788 
3789 	/*
3790 	 * Ensure that split_info.do_nap is set after setting
3791 	 * the vcore pointer in the PACA of the secondaries.
3792 	 */
3793 	smp_mb();
3794 
3795 	/*
3796 	 * When doing micro-threading, poke the inactive threads as well.
3797 	 * This gets them to the nap instruction after kvm_do_nap,
3798 	 * which reduces the time taken to unsplit later.
3799 	 */
3800 	if (cmd_bit) {
3801 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3802 		for (thr = 1; thr < threads_per_subcore; ++thr)
3803 			if (!(active & (1 << thr)))
3804 				kvmppc_ipi_thread(pcpu + thr);
3805 	}
3806 
3807 	vc->vcore_state = VCORE_RUNNING;
3808 	preempt_disable();
3809 
3810 	trace_kvmppc_run_core(vc, 0);
3811 
3812 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3813 		spin_unlock(&core_info.vc[sub]->lock);
3814 
3815 	guest_enter_irqoff();
3816 
3817 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3818 
3819 	this_cpu_disable_ftrace();
3820 
3821 	/*
3822 	 * Interrupts will be enabled once we get into the guest,
3823 	 * so tell lockdep that we're about to enable interrupts.
3824 	 */
3825 	trace_hardirqs_on();
3826 
3827 	trap = __kvmppc_vcore_entry();
3828 
3829 	trace_hardirqs_off();
3830 
3831 	this_cpu_enable_ftrace();
3832 
3833 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3834 
3835 	set_irq_happened(trap);
3836 
3837 	spin_lock(&vc->lock);
3838 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3839 	vc->vcore_state = VCORE_EXITING;
3840 
3841 	/* wait for secondary threads to finish writing their state to memory */
3842 	kvmppc_wait_for_nap(controlled_threads);
3843 
3844 	/* Return to whole-core mode if we split the core earlier */
3845 	if (cmd_bit) {
3846 		unsigned long hid0 = mfspr(SPRN_HID0);
3847 		unsigned long loops = 0;
3848 
3849 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3850 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3851 		mb();
3852 		mtspr(SPRN_HID0, hid0);
3853 		isync();
3854 		for (;;) {
3855 			hid0 = mfspr(SPRN_HID0);
3856 			if (!(hid0 & stat_bit))
3857 				break;
3858 			cpu_relax();
3859 			++loops;
3860 		}
3861 		split_info.do_nap = 0;
3862 	}
3863 
3864 	kvmppc_set_host_core(pcpu);
3865 
3866 	context_tracking_guest_exit();
3867 	if (!vtime_accounting_enabled_this_cpu()) {
3868 		local_irq_enable();
3869 		/*
3870 		 * Service IRQs here before vtime_account_guest_exit() so any
3871 		 * ticks that occurred while running the guest are accounted to
3872 		 * the guest. If vtime accounting is enabled, accounting uses
3873 		 * TB rather than ticks, so it can be done without enabling
3874 		 * interrupts here, which has the problem that it accounts
3875 		 * interrupt processing overhead to the host.
3876 		 */
3877 		local_irq_disable();
3878 	}
3879 	vtime_account_guest_exit();
3880 
3881 	local_irq_enable();
3882 
3883 	/* Let secondaries go back to the offline loop */
3884 	for (i = 0; i < controlled_threads; ++i) {
3885 		kvmppc_release_hwthread(pcpu + i);
3886 		if (sip && sip->napped[i])
3887 			kvmppc_ipi_thread(pcpu + i);
3888 	}
3889 
3890 	spin_unlock(&vc->lock);
3891 
3892 	/* make sure updates to secondary vcpu structs are visible now */
3893 	smp_mb();
3894 
3895 	preempt_enable();
3896 
3897 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3898 		pvc = core_info.vc[sub];
3899 		post_guest_process(pvc, pvc == vc);
3900 	}
3901 
3902 	spin_lock(&vc->lock);
3903 
3904  out:
3905 	vc->vcore_state = VCORE_INACTIVE;
3906 	trace_kvmppc_run_core(vc, 1);
3907 }
3908 
3909 static inline bool hcall_is_xics(unsigned long req)
3910 {
3911 	return req == H_EOI || req == H_CPPR || req == H_IPI ||
3912 		req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
3913 }
3914 
3915 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu)
3916 {
3917 	struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3918 	if (lp) {
3919 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3920 		lp->yield_count = cpu_to_be32(yield_count);
3921 		vcpu->arch.vpa.dirty = 1;
3922 	}
3923 }
3924 
3925 /* call our hypervisor to load up HV regs and go */
3926 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
3927 {
3928 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3929 	unsigned long host_psscr;
3930 	unsigned long msr;
3931 	struct hv_guest_state hvregs;
3932 	struct p9_host_os_sprs host_os_sprs;
3933 	s64 dec;
3934 	int trap;
3935 
3936 	msr = mfmsr();
3937 
3938 	save_p9_host_os_sprs(&host_os_sprs);
3939 
3940 	/*
3941 	 * We need to save and restore the guest visible part of the
3942 	 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3943 	 * doesn't do this for us. Note only required if pseries since
3944 	 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
3945 	 */
3946 	host_psscr = mfspr(SPRN_PSSCR_PR);
3947 
3948 	kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
3949 	if (lazy_irq_pending())
3950 		return 0;
3951 
3952 	if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
3953 		msr = mfmsr(); /* TM restore can update msr */
3954 
3955 	if (vcpu->arch.psscr != host_psscr)
3956 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3957 
3958 	kvmhv_save_hv_regs(vcpu, &hvregs);
3959 	hvregs.lpcr = lpcr;
3960 	vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3961 	hvregs.version = HV_GUEST_STATE_VERSION;
3962 	if (vcpu->arch.nested) {
3963 		hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3964 		hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3965 	} else {
3966 		hvregs.lpid = vcpu->kvm->arch.lpid;
3967 		hvregs.vcpu_token = vcpu->vcpu_id;
3968 	}
3969 	hvregs.hdec_expiry = time_limit;
3970 
3971 	/*
3972 	 * When setting DEC, we must always deal with irq_work_raise
3973 	 * via NMI vs setting DEC. The problem occurs right as we
3974 	 * switch into guest mode if a NMI hits and sets pending work
3975 	 * and sets DEC, then that will apply to the guest and not
3976 	 * bring us back to the host.
3977 	 *
3978 	 * irq_work_raise could check a flag (or possibly LPCR[HDICE]
3979 	 * for example) and set HDEC to 1? That wouldn't solve the
3980 	 * nested hv case which needs to abort the hcall or zero the
3981 	 * time limit.
3982 	 *
3983 	 * XXX: Another day's problem.
3984 	 */
3985 	mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb);
3986 
3987 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3988 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3989 	switch_pmu_to_guest(vcpu, &host_os_sprs);
3990 	trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3991 				  __pa(&vcpu->arch.regs));
3992 	kvmhv_restore_hv_return_state(vcpu, &hvregs);
3993 	switch_pmu_to_host(vcpu, &host_os_sprs);
3994 	vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3995 	vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3996 	vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3997 	vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3998 
3999 	store_vcpu_state(vcpu);
4000 
4001 	dec = mfspr(SPRN_DEC);
4002 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4003 		dec = (s32) dec;
4004 	*tb = mftb();
4005 	vcpu->arch.dec_expires = dec + (*tb + vc->tb_offset);
4006 
4007 	timer_rearm_host_dec(*tb);
4008 
4009 	restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4010 	if (vcpu->arch.psscr != host_psscr)
4011 		mtspr(SPRN_PSSCR_PR, host_psscr);
4012 
4013 	return trap;
4014 }
4015 
4016 /*
4017  * Guest entry for POWER9 and later CPUs.
4018  */
4019 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
4020 			 unsigned long lpcr, u64 *tb)
4021 {
4022 	u64 next_timer;
4023 	int trap;
4024 
4025 	next_timer = timer_get_next_tb();
4026 	if (*tb >= next_timer)
4027 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
4028 	if (next_timer < time_limit)
4029 		time_limit = next_timer;
4030 	else if (*tb >= time_limit) /* nested time limit */
4031 		return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER;
4032 
4033 	vcpu->arch.ceded = 0;
4034 
4035 	vcpu_vpa_increment_dispatch(vcpu);
4036 
4037 	if (kvmhv_on_pseries()) {
4038 		trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb);
4039 
4040 		/* H_CEDE has to be handled now, not later */
4041 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
4042 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
4043 			kvmppc_cede(vcpu);
4044 			kvmppc_set_gpr(vcpu, 3, 0);
4045 			trap = 0;
4046 		}
4047 
4048 	} else {
4049 		struct kvm *kvm = vcpu->kvm;
4050 
4051 		kvmppc_xive_push_vcpu(vcpu);
4052 
4053 		__this_cpu_write(cpu_in_guest, kvm);
4054 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4055 		__this_cpu_write(cpu_in_guest, NULL);
4056 
4057 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
4058 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4059 			unsigned long req = kvmppc_get_gpr(vcpu, 3);
4060 
4061 			/* H_CEDE has to be handled now, not later */
4062 			if (req == H_CEDE) {
4063 				kvmppc_cede(vcpu);
4064 				kvmppc_xive_rearm_escalation(vcpu); /* may un-cede */
4065 				kvmppc_set_gpr(vcpu, 3, 0);
4066 				trap = 0;
4067 
4068 			/* XICS hcalls must be handled before xive is pulled */
4069 			} else if (hcall_is_xics(req)) {
4070 				int ret;
4071 
4072 				ret = kvmppc_xive_xics_hcall(vcpu, req);
4073 				if (ret != H_TOO_HARD) {
4074 					kvmppc_set_gpr(vcpu, 3, ret);
4075 					trap = 0;
4076 				}
4077 			}
4078 		}
4079 		kvmppc_xive_pull_vcpu(vcpu);
4080 
4081 		if (kvm_is_radix(kvm))
4082 			vcpu->arch.slb_max = 0;
4083 	}
4084 
4085 	vcpu_vpa_increment_dispatch(vcpu);
4086 
4087 	return trap;
4088 }
4089 
4090 /*
4091  * Wait for some other vcpu thread to execute us, and
4092  * wake us up when we need to handle something in the host.
4093  */
4094 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4095 				 struct kvm_vcpu *vcpu, int wait_state)
4096 {
4097 	DEFINE_WAIT(wait);
4098 
4099 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4100 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4101 		spin_unlock(&vc->lock);
4102 		schedule();
4103 		spin_lock(&vc->lock);
4104 	}
4105 	finish_wait(&vcpu->arch.cpu_run, &wait);
4106 }
4107 
4108 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4109 {
4110 	if (!halt_poll_ns_grow)
4111 		return;
4112 
4113 	vc->halt_poll_ns *= halt_poll_ns_grow;
4114 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4115 		vc->halt_poll_ns = halt_poll_ns_grow_start;
4116 }
4117 
4118 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4119 {
4120 	if (halt_poll_ns_shrink == 0)
4121 		vc->halt_poll_ns = 0;
4122 	else
4123 		vc->halt_poll_ns /= halt_poll_ns_shrink;
4124 }
4125 
4126 #ifdef CONFIG_KVM_XICS
4127 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4128 {
4129 	if (!xics_on_xive())
4130 		return false;
4131 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4132 		vcpu->arch.xive_saved_state.cppr;
4133 }
4134 #else
4135 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4136 {
4137 	return false;
4138 }
4139 #endif /* CONFIG_KVM_XICS */
4140 
4141 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4142 {
4143 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4144 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4145 		return true;
4146 
4147 	return false;
4148 }
4149 
4150 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu)
4151 {
4152 	if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4153 		return true;
4154 	return false;
4155 }
4156 
4157 /*
4158  * Check to see if any of the runnable vcpus on the vcore have pending
4159  * exceptions or are no longer ceded
4160  */
4161 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4162 {
4163 	struct kvm_vcpu *vcpu;
4164 	int i;
4165 
4166 	for_each_runnable_thread(i, vcpu, vc) {
4167 		if (kvmppc_vcpu_check_block(vcpu))
4168 			return 1;
4169 	}
4170 
4171 	return 0;
4172 }
4173 
4174 /*
4175  * All the vcpus in this vcore are idle, so wait for a decrementer
4176  * or external interrupt to one of the vcpus.  vc->lock is held.
4177  */
4178 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4179 {
4180 	ktime_t cur, start_poll, start_wait;
4181 	int do_sleep = 1;
4182 	u64 block_ns;
4183 
4184 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
4185 
4186 	/* Poll for pending exceptions and ceded state */
4187 	cur = start_poll = ktime_get();
4188 	if (vc->halt_poll_ns) {
4189 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4190 		++vc->runner->stat.generic.halt_attempted_poll;
4191 
4192 		vc->vcore_state = VCORE_POLLING;
4193 		spin_unlock(&vc->lock);
4194 
4195 		do {
4196 			if (kvmppc_vcore_check_block(vc)) {
4197 				do_sleep = 0;
4198 				break;
4199 			}
4200 			cur = ktime_get();
4201 		} while (kvm_vcpu_can_poll(cur, stop));
4202 
4203 		spin_lock(&vc->lock);
4204 		vc->vcore_state = VCORE_INACTIVE;
4205 
4206 		if (!do_sleep) {
4207 			++vc->runner->stat.generic.halt_successful_poll;
4208 			goto out;
4209 		}
4210 	}
4211 
4212 	prepare_to_rcuwait(&vc->wait);
4213 	set_current_state(TASK_INTERRUPTIBLE);
4214 	if (kvmppc_vcore_check_block(vc)) {
4215 		finish_rcuwait(&vc->wait);
4216 		do_sleep = 0;
4217 		/* If we polled, count this as a successful poll */
4218 		if (vc->halt_poll_ns)
4219 			++vc->runner->stat.generic.halt_successful_poll;
4220 		goto out;
4221 	}
4222 
4223 	start_wait = ktime_get();
4224 
4225 	vc->vcore_state = VCORE_SLEEPING;
4226 	trace_kvmppc_vcore_blocked(vc, 0);
4227 	spin_unlock(&vc->lock);
4228 	schedule();
4229 	finish_rcuwait(&vc->wait);
4230 	spin_lock(&vc->lock);
4231 	vc->vcore_state = VCORE_INACTIVE;
4232 	trace_kvmppc_vcore_blocked(vc, 1);
4233 	++vc->runner->stat.halt_successful_wait;
4234 
4235 	cur = ktime_get();
4236 
4237 out:
4238 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4239 
4240 	/* Attribute wait time */
4241 	if (do_sleep) {
4242 		vc->runner->stat.generic.halt_wait_ns +=
4243 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
4244 		KVM_STATS_LOG_HIST_UPDATE(
4245 				vc->runner->stat.generic.halt_wait_hist,
4246 				ktime_to_ns(cur) - ktime_to_ns(start_wait));
4247 		/* Attribute failed poll time */
4248 		if (vc->halt_poll_ns) {
4249 			vc->runner->stat.generic.halt_poll_fail_ns +=
4250 				ktime_to_ns(start_wait) -
4251 				ktime_to_ns(start_poll);
4252 			KVM_STATS_LOG_HIST_UPDATE(
4253 				vc->runner->stat.generic.halt_poll_fail_hist,
4254 				ktime_to_ns(start_wait) -
4255 				ktime_to_ns(start_poll));
4256 		}
4257 	} else {
4258 		/* Attribute successful poll time */
4259 		if (vc->halt_poll_ns) {
4260 			vc->runner->stat.generic.halt_poll_success_ns +=
4261 				ktime_to_ns(cur) -
4262 				ktime_to_ns(start_poll);
4263 			KVM_STATS_LOG_HIST_UPDATE(
4264 				vc->runner->stat.generic.halt_poll_success_hist,
4265 				ktime_to_ns(cur) - ktime_to_ns(start_poll));
4266 		}
4267 	}
4268 
4269 	/* Adjust poll time */
4270 	if (halt_poll_ns) {
4271 		if (block_ns <= vc->halt_poll_ns)
4272 			;
4273 		/* We slept and blocked for longer than the max halt time */
4274 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4275 			shrink_halt_poll_ns(vc);
4276 		/* We slept and our poll time is too small */
4277 		else if (vc->halt_poll_ns < halt_poll_ns &&
4278 				block_ns < halt_poll_ns)
4279 			grow_halt_poll_ns(vc);
4280 		if (vc->halt_poll_ns > halt_poll_ns)
4281 			vc->halt_poll_ns = halt_poll_ns;
4282 	} else
4283 		vc->halt_poll_ns = 0;
4284 
4285 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4286 }
4287 
4288 /*
4289  * This never fails for a radix guest, as none of the operations it does
4290  * for a radix guest can fail or have a way to report failure.
4291  */
4292 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4293 {
4294 	int r = 0;
4295 	struct kvm *kvm = vcpu->kvm;
4296 
4297 	mutex_lock(&kvm->arch.mmu_setup_lock);
4298 	if (!kvm->arch.mmu_ready) {
4299 		if (!kvm_is_radix(kvm))
4300 			r = kvmppc_hv_setup_htab_rma(vcpu);
4301 		if (!r) {
4302 			if (cpu_has_feature(CPU_FTR_ARCH_300))
4303 				kvmppc_setup_partition_table(kvm);
4304 			kvm->arch.mmu_ready = 1;
4305 		}
4306 	}
4307 	mutex_unlock(&kvm->arch.mmu_setup_lock);
4308 	return r;
4309 }
4310 
4311 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4312 {
4313 	struct kvm_run *run = vcpu->run;
4314 	int n_ceded, i, r;
4315 	struct kvmppc_vcore *vc;
4316 	struct kvm_vcpu *v;
4317 
4318 	trace_kvmppc_run_vcpu_enter(vcpu);
4319 
4320 	run->exit_reason = 0;
4321 	vcpu->arch.ret = RESUME_GUEST;
4322 	vcpu->arch.trap = 0;
4323 	kvmppc_update_vpas(vcpu);
4324 
4325 	/*
4326 	 * Synchronize with other threads in this virtual core
4327 	 */
4328 	vc = vcpu->arch.vcore;
4329 	spin_lock(&vc->lock);
4330 	vcpu->arch.ceded = 0;
4331 	vcpu->arch.run_task = current;
4332 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4333 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4334 	vcpu->arch.busy_preempt = TB_NIL;
4335 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4336 	++vc->n_runnable;
4337 
4338 	/*
4339 	 * This happens the first time this is called for a vcpu.
4340 	 * If the vcore is already running, we may be able to start
4341 	 * this thread straight away and have it join in.
4342 	 */
4343 	if (!signal_pending(current)) {
4344 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
4345 		     vc->vcore_state == VCORE_RUNNING) &&
4346 			   !VCORE_IS_EXITING(vc)) {
4347 			kvmppc_create_dtl_entry(vcpu, vc);
4348 			kvmppc_start_thread(vcpu, vc);
4349 			trace_kvm_guest_enter(vcpu);
4350 		} else if (vc->vcore_state == VCORE_SLEEPING) {
4351 		        rcuwait_wake_up(&vc->wait);
4352 		}
4353 
4354 	}
4355 
4356 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4357 	       !signal_pending(current)) {
4358 		/* See if the MMU is ready to go */
4359 		if (!vcpu->kvm->arch.mmu_ready) {
4360 			spin_unlock(&vc->lock);
4361 			r = kvmhv_setup_mmu(vcpu);
4362 			spin_lock(&vc->lock);
4363 			if (r) {
4364 				run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4365 				run->fail_entry.
4366 					hardware_entry_failure_reason = 0;
4367 				vcpu->arch.ret = r;
4368 				break;
4369 			}
4370 		}
4371 
4372 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4373 			kvmppc_vcore_end_preempt(vc);
4374 
4375 		if (vc->vcore_state != VCORE_INACTIVE) {
4376 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4377 			continue;
4378 		}
4379 		for_each_runnable_thread(i, v, vc) {
4380 			kvmppc_core_prepare_to_enter(v);
4381 			if (signal_pending(v->arch.run_task)) {
4382 				kvmppc_remove_runnable(vc, v, mftb());
4383 				v->stat.signal_exits++;
4384 				v->run->exit_reason = KVM_EXIT_INTR;
4385 				v->arch.ret = -EINTR;
4386 				wake_up(&v->arch.cpu_run);
4387 			}
4388 		}
4389 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4390 			break;
4391 		n_ceded = 0;
4392 		for_each_runnable_thread(i, v, vc) {
4393 			if (!kvmppc_vcpu_woken(v))
4394 				n_ceded += v->arch.ceded;
4395 			else
4396 				v->arch.ceded = 0;
4397 		}
4398 		vc->runner = vcpu;
4399 		if (n_ceded == vc->n_runnable) {
4400 			kvmppc_vcore_blocked(vc);
4401 		} else if (need_resched()) {
4402 			kvmppc_vcore_preempt(vc);
4403 			/* Let something else run */
4404 			cond_resched_lock(&vc->lock);
4405 			if (vc->vcore_state == VCORE_PREEMPT)
4406 				kvmppc_vcore_end_preempt(vc);
4407 		} else {
4408 			kvmppc_run_core(vc);
4409 		}
4410 		vc->runner = NULL;
4411 	}
4412 
4413 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4414 	       (vc->vcore_state == VCORE_RUNNING ||
4415 		vc->vcore_state == VCORE_EXITING ||
4416 		vc->vcore_state == VCORE_PIGGYBACK))
4417 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4418 
4419 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4420 		kvmppc_vcore_end_preempt(vc);
4421 
4422 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4423 		kvmppc_remove_runnable(vc, vcpu, mftb());
4424 		vcpu->stat.signal_exits++;
4425 		run->exit_reason = KVM_EXIT_INTR;
4426 		vcpu->arch.ret = -EINTR;
4427 	}
4428 
4429 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4430 		/* Wake up some vcpu to run the core */
4431 		i = -1;
4432 		v = next_runnable_thread(vc, &i);
4433 		wake_up(&v->arch.cpu_run);
4434 	}
4435 
4436 	trace_kvmppc_run_vcpu_exit(vcpu);
4437 	spin_unlock(&vc->lock);
4438 	return vcpu->arch.ret;
4439 }
4440 
4441 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4442 			  unsigned long lpcr)
4443 {
4444 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
4445 	struct kvm_run *run = vcpu->run;
4446 	int trap, r, pcpu;
4447 	int srcu_idx;
4448 	struct kvmppc_vcore *vc;
4449 	struct kvm *kvm = vcpu->kvm;
4450 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4451 	unsigned long flags;
4452 	u64 tb;
4453 
4454 	trace_kvmppc_run_vcpu_enter(vcpu);
4455 
4456 	run->exit_reason = 0;
4457 	vcpu->arch.ret = RESUME_GUEST;
4458 	vcpu->arch.trap = 0;
4459 
4460 	vc = vcpu->arch.vcore;
4461 	vcpu->arch.ceded = 0;
4462 	vcpu->arch.run_task = current;
4463 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4464 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4465 
4466 	/* See if the MMU is ready to go */
4467 	if (unlikely(!kvm->arch.mmu_ready)) {
4468 		r = kvmhv_setup_mmu(vcpu);
4469 		if (r) {
4470 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4471 			run->fail_entry.hardware_entry_failure_reason = 0;
4472 			vcpu->arch.ret = r;
4473 			return r;
4474 		}
4475 	}
4476 
4477 	if (need_resched())
4478 		cond_resched();
4479 
4480 	kvmppc_update_vpas(vcpu);
4481 
4482 	preempt_disable();
4483 	pcpu = smp_processor_id();
4484 	if (kvm_is_radix(kvm))
4485 		kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4486 
4487 	/* flags save not required, but irq_pmu has no disable/enable API */
4488 	powerpc_local_irq_pmu_save(flags);
4489 
4490 	if (signal_pending(current))
4491 		goto sigpend;
4492 	if (need_resched() || !kvm->arch.mmu_ready)
4493 		goto out;
4494 
4495 	if (!nested) {
4496 		kvmppc_core_prepare_to_enter(vcpu);
4497 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4498 			     &vcpu->arch.pending_exceptions))
4499 			lpcr |= LPCR_MER;
4500 	} else if (vcpu->arch.pending_exceptions ||
4501 		   vcpu->arch.doorbell_request ||
4502 		   xive_interrupt_pending(vcpu)) {
4503 		vcpu->arch.ret = RESUME_HOST;
4504 		goto out;
4505 	}
4506 
4507 	if (vcpu->arch.timer_running) {
4508 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
4509 		vcpu->arch.timer_running = 0;
4510 	}
4511 
4512 	tb = mftb();
4513 
4514 	vcpu->cpu = pcpu;
4515 	vcpu->arch.thread_cpu = pcpu;
4516 	vc->pcpu = pcpu;
4517 	local_paca->kvm_hstate.kvm_vcpu = vcpu;
4518 	local_paca->kvm_hstate.ptid = 0;
4519 	local_paca->kvm_hstate.fake_suspend = 0;
4520 
4521 	__kvmppc_create_dtl_entry(vcpu, pcpu, tb + vc->tb_offset, 0);
4522 
4523 	trace_kvm_guest_enter(vcpu);
4524 
4525 	guest_enter_irqoff();
4526 
4527 	srcu_idx = srcu_read_lock(&kvm->srcu);
4528 
4529 	this_cpu_disable_ftrace();
4530 
4531 	/* Tell lockdep that we're about to enable interrupts */
4532 	trace_hardirqs_on();
4533 
4534 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb);
4535 	vcpu->arch.trap = trap;
4536 
4537 	trace_hardirqs_off();
4538 
4539 	this_cpu_enable_ftrace();
4540 
4541 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4542 
4543 	set_irq_happened(trap);
4544 
4545 	context_tracking_guest_exit();
4546 	if (!vtime_accounting_enabled_this_cpu()) {
4547 		local_irq_enable();
4548 		/*
4549 		 * Service IRQs here before vtime_account_guest_exit() so any
4550 		 * ticks that occurred while running the guest are accounted to
4551 		 * the guest. If vtime accounting is enabled, accounting uses
4552 		 * TB rather than ticks, so it can be done without enabling
4553 		 * interrupts here, which has the problem that it accounts
4554 		 * interrupt processing overhead to the host.
4555 		 */
4556 		local_irq_disable();
4557 	}
4558 	vtime_account_guest_exit();
4559 
4560 	vcpu->cpu = -1;
4561 	vcpu->arch.thread_cpu = -1;
4562 
4563 	powerpc_local_irq_pmu_restore(flags);
4564 
4565 	preempt_enable();
4566 
4567 	/*
4568 	 * cancel pending decrementer exception if DEC is now positive, or if
4569 	 * entering a nested guest in which case the decrementer is now owned
4570 	 * by L2 and the L1 decrementer is provided in hdec_expires
4571 	 */
4572 	if (kvmppc_core_pending_dec(vcpu) &&
4573 			((tb < kvmppc_dec_expires_host_tb(vcpu)) ||
4574 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4575 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4576 		kvmppc_core_dequeue_dec(vcpu);
4577 
4578 	trace_kvm_guest_exit(vcpu);
4579 	r = RESUME_GUEST;
4580 	if (trap) {
4581 		if (!nested)
4582 			r = kvmppc_handle_exit_hv(vcpu, current);
4583 		else
4584 			r = kvmppc_handle_nested_exit(vcpu);
4585 	}
4586 	vcpu->arch.ret = r;
4587 
4588 	if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) {
4589 		kvmppc_set_timer(vcpu);
4590 
4591 		prepare_to_rcuwait(wait);
4592 		for (;;) {
4593 			set_current_state(TASK_INTERRUPTIBLE);
4594 			if (signal_pending(current)) {
4595 				vcpu->stat.signal_exits++;
4596 				run->exit_reason = KVM_EXIT_INTR;
4597 				vcpu->arch.ret = -EINTR;
4598 				break;
4599 			}
4600 
4601 			if (kvmppc_vcpu_check_block(vcpu))
4602 				break;
4603 
4604 			trace_kvmppc_vcore_blocked(vc, 0);
4605 			schedule();
4606 			trace_kvmppc_vcore_blocked(vc, 1);
4607 		}
4608 		finish_rcuwait(wait);
4609 	}
4610 	vcpu->arch.ceded = 0;
4611 
4612  done:
4613 	trace_kvmppc_run_vcpu_exit(vcpu);
4614 
4615 	return vcpu->arch.ret;
4616 
4617  sigpend:
4618 	vcpu->stat.signal_exits++;
4619 	run->exit_reason = KVM_EXIT_INTR;
4620 	vcpu->arch.ret = -EINTR;
4621  out:
4622 	powerpc_local_irq_pmu_restore(flags);
4623 	preempt_enable();
4624 	goto done;
4625 }
4626 
4627 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4628 {
4629 	struct kvm_run *run = vcpu->run;
4630 	int r;
4631 	int srcu_idx;
4632 	struct kvm *kvm;
4633 	unsigned long msr;
4634 
4635 	if (!vcpu->arch.sane) {
4636 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4637 		return -EINVAL;
4638 	}
4639 
4640 	/* No need to go into the guest when all we'll do is come back out */
4641 	if (signal_pending(current)) {
4642 		run->exit_reason = KVM_EXIT_INTR;
4643 		return -EINTR;
4644 	}
4645 
4646 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4647 	/*
4648 	 * Don't allow entry with a suspended transaction, because
4649 	 * the guest entry/exit code will lose it.
4650 	 */
4651 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4652 	    (current->thread.regs->msr & MSR_TM)) {
4653 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4654 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4655 			run->fail_entry.hardware_entry_failure_reason = 0;
4656 			return -EINVAL;
4657 		}
4658 	}
4659 #endif
4660 
4661 	/*
4662 	 * Force online to 1 for the sake of old userspace which doesn't
4663 	 * set it.
4664 	 */
4665 	if (!vcpu->arch.online) {
4666 		atomic_inc(&vcpu->arch.vcore->online_count);
4667 		vcpu->arch.online = 1;
4668 	}
4669 
4670 	kvmppc_core_prepare_to_enter(vcpu);
4671 
4672 	kvm = vcpu->kvm;
4673 	atomic_inc(&kvm->arch.vcpus_running);
4674 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4675 	smp_mb();
4676 
4677 	msr = 0;
4678 	if (IS_ENABLED(CONFIG_PPC_FPU))
4679 		msr |= MSR_FP;
4680 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
4681 		msr |= MSR_VEC;
4682 	if (cpu_has_feature(CPU_FTR_VSX))
4683 		msr |= MSR_VSX;
4684 	if ((cpu_has_feature(CPU_FTR_TM) ||
4685 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
4686 			(vcpu->arch.hfscr & HFSCR_TM))
4687 		msr |= MSR_TM;
4688 	msr = msr_check_and_set(msr);
4689 
4690 	kvmppc_save_user_regs();
4691 
4692 	kvmppc_save_current_sprs();
4693 
4694 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4695 		vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4696 	vcpu->arch.pgdir = kvm->mm->pgd;
4697 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4698 
4699 	do {
4700 		if (cpu_has_feature(CPU_FTR_ARCH_300))
4701 			r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4702 						  vcpu->arch.vcore->lpcr);
4703 		else
4704 			r = kvmppc_run_vcpu(vcpu);
4705 
4706 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4707 			if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) {
4708 				/*
4709 				 * These should have been caught reflected
4710 				 * into the guest by now. Final sanity check:
4711 				 * don't allow userspace to execute hcalls in
4712 				 * the hypervisor.
4713 				 */
4714 				r = RESUME_GUEST;
4715 				continue;
4716 			}
4717 			trace_kvm_hcall_enter(vcpu);
4718 			r = kvmppc_pseries_do_hcall(vcpu);
4719 			trace_kvm_hcall_exit(vcpu, r);
4720 			kvmppc_core_prepare_to_enter(vcpu);
4721 		} else if (r == RESUME_PAGE_FAULT) {
4722 			srcu_idx = srcu_read_lock(&kvm->srcu);
4723 			r = kvmppc_book3s_hv_page_fault(vcpu,
4724 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4725 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4726 		} else if (r == RESUME_PASSTHROUGH) {
4727 			if (WARN_ON(xics_on_xive()))
4728 				r = H_SUCCESS;
4729 			else
4730 				r = kvmppc_xics_rm_complete(vcpu, 0);
4731 		}
4732 	} while (is_kvmppc_resume_guest(r));
4733 
4734 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4735 	atomic_dec(&kvm->arch.vcpus_running);
4736 
4737 	srr_regs_clobbered();
4738 
4739 	return r;
4740 }
4741 
4742 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4743 				     int shift, int sllp)
4744 {
4745 	(*sps)->page_shift = shift;
4746 	(*sps)->slb_enc = sllp;
4747 	(*sps)->enc[0].page_shift = shift;
4748 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4749 	/*
4750 	 * Add 16MB MPSS support (may get filtered out by userspace)
4751 	 */
4752 	if (shift != 24) {
4753 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4754 		if (penc != -1) {
4755 			(*sps)->enc[1].page_shift = 24;
4756 			(*sps)->enc[1].pte_enc = penc;
4757 		}
4758 	}
4759 	(*sps)++;
4760 }
4761 
4762 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4763 					 struct kvm_ppc_smmu_info *info)
4764 {
4765 	struct kvm_ppc_one_seg_page_size *sps;
4766 
4767 	/*
4768 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4769 	 * POWER7 doesn't support keys for instruction accesses,
4770 	 * POWER8 and POWER9 do.
4771 	 */
4772 	info->data_keys = 32;
4773 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4774 
4775 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4776 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4777 	info->slb_size = 32;
4778 
4779 	/* We only support these sizes for now, and no muti-size segments */
4780 	sps = &info->sps[0];
4781 	kvmppc_add_seg_page_size(&sps, 12, 0);
4782 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4783 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4784 
4785 	/* If running as a nested hypervisor, we don't support HPT guests */
4786 	if (kvmhv_on_pseries())
4787 		info->flags |= KVM_PPC_NO_HASH;
4788 
4789 	return 0;
4790 }
4791 
4792 /*
4793  * Get (and clear) the dirty memory log for a memory slot.
4794  */
4795 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4796 					 struct kvm_dirty_log *log)
4797 {
4798 	struct kvm_memslots *slots;
4799 	struct kvm_memory_slot *memslot;
4800 	int r;
4801 	unsigned long n, i;
4802 	unsigned long *buf, *p;
4803 	struct kvm_vcpu *vcpu;
4804 
4805 	mutex_lock(&kvm->slots_lock);
4806 
4807 	r = -EINVAL;
4808 	if (log->slot >= KVM_USER_MEM_SLOTS)
4809 		goto out;
4810 
4811 	slots = kvm_memslots(kvm);
4812 	memslot = id_to_memslot(slots, log->slot);
4813 	r = -ENOENT;
4814 	if (!memslot || !memslot->dirty_bitmap)
4815 		goto out;
4816 
4817 	/*
4818 	 * Use second half of bitmap area because both HPT and radix
4819 	 * accumulate bits in the first half.
4820 	 */
4821 	n = kvm_dirty_bitmap_bytes(memslot);
4822 	buf = memslot->dirty_bitmap + n / sizeof(long);
4823 	memset(buf, 0, n);
4824 
4825 	if (kvm_is_radix(kvm))
4826 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4827 	else
4828 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4829 	if (r)
4830 		goto out;
4831 
4832 	/*
4833 	 * We accumulate dirty bits in the first half of the
4834 	 * memslot's dirty_bitmap area, for when pages are paged
4835 	 * out or modified by the host directly.  Pick up these
4836 	 * bits and add them to the map.
4837 	 */
4838 	p = memslot->dirty_bitmap;
4839 	for (i = 0; i < n / sizeof(long); ++i)
4840 		buf[i] |= xchg(&p[i], 0);
4841 
4842 	/* Harvest dirty bits from VPA and DTL updates */
4843 	/* Note: we never modify the SLB shadow buffer areas */
4844 	kvm_for_each_vcpu(i, vcpu, kvm) {
4845 		spin_lock(&vcpu->arch.vpa_update_lock);
4846 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4847 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4848 		spin_unlock(&vcpu->arch.vpa_update_lock);
4849 	}
4850 
4851 	r = -EFAULT;
4852 	if (copy_to_user(log->dirty_bitmap, buf, n))
4853 		goto out;
4854 
4855 	r = 0;
4856 out:
4857 	mutex_unlock(&kvm->slots_lock);
4858 	return r;
4859 }
4860 
4861 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4862 {
4863 	vfree(slot->arch.rmap);
4864 	slot->arch.rmap = NULL;
4865 }
4866 
4867 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4868 				const struct kvm_memory_slot *old,
4869 				struct kvm_memory_slot *new,
4870 				enum kvm_mr_change change)
4871 {
4872 	if (change == KVM_MR_CREATE) {
4873 		unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap));
4874 
4875 		if ((size >> PAGE_SHIFT) > totalram_pages())
4876 			return -ENOMEM;
4877 
4878 		new->arch.rmap = vzalloc(size);
4879 		if (!new->arch.rmap)
4880 			return -ENOMEM;
4881 	} else if (change != KVM_MR_DELETE) {
4882 		new->arch.rmap = old->arch.rmap;
4883 	}
4884 
4885 	return 0;
4886 }
4887 
4888 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4889 				struct kvm_memory_slot *old,
4890 				const struct kvm_memory_slot *new,
4891 				enum kvm_mr_change change)
4892 {
4893 	/*
4894 	 * If we are creating or modifying a memslot, it might make
4895 	 * some address that was previously cached as emulated
4896 	 * MMIO be no longer emulated MMIO, so invalidate
4897 	 * all the caches of emulated MMIO translations.
4898 	 */
4899 	if (change != KVM_MR_DELETE)
4900 		atomic64_inc(&kvm->arch.mmio_update);
4901 
4902 	/*
4903 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4904 	 * have already called kvm_arch_flush_shadow_memslot() to
4905 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
4906 	 * previous mappings.  So the only case to handle is
4907 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4908 	 * has been changed.
4909 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4910 	 * to get rid of any THP PTEs in the partition-scoped page tables
4911 	 * so we can track dirtiness at the page level; we flush when
4912 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4913 	 * using THP PTEs.
4914 	 */
4915 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4916 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4917 		kvmppc_radix_flush_memslot(kvm, old);
4918 	/*
4919 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4920 	 */
4921 	if (!kvm->arch.secure_guest)
4922 		return;
4923 
4924 	switch (change) {
4925 	case KVM_MR_CREATE:
4926 		/*
4927 		 * @TODO kvmppc_uvmem_memslot_create() can fail and
4928 		 * return error. Fix this.
4929 		 */
4930 		kvmppc_uvmem_memslot_create(kvm, new);
4931 		break;
4932 	case KVM_MR_DELETE:
4933 		kvmppc_uvmem_memslot_delete(kvm, old);
4934 		break;
4935 	default:
4936 		/* TODO: Handle KVM_MR_MOVE */
4937 		break;
4938 	}
4939 }
4940 
4941 /*
4942  * Update LPCR values in kvm->arch and in vcores.
4943  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4944  * of kvm->arch.lpcr update).
4945  */
4946 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4947 {
4948 	long int i;
4949 	u32 cores_done = 0;
4950 
4951 	if ((kvm->arch.lpcr & mask) == lpcr)
4952 		return;
4953 
4954 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4955 
4956 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4957 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4958 		if (!vc)
4959 			continue;
4960 
4961 		spin_lock(&vc->lock);
4962 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4963 		verify_lpcr(kvm, vc->lpcr);
4964 		spin_unlock(&vc->lock);
4965 		if (++cores_done >= kvm->arch.online_vcores)
4966 			break;
4967 	}
4968 }
4969 
4970 void kvmppc_setup_partition_table(struct kvm *kvm)
4971 {
4972 	unsigned long dw0, dw1;
4973 
4974 	if (!kvm_is_radix(kvm)) {
4975 		/* PS field - page size for VRMA */
4976 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4977 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4978 		/* HTABSIZE and HTABORG fields */
4979 		dw0 |= kvm->arch.sdr1;
4980 
4981 		/* Second dword as set by userspace */
4982 		dw1 = kvm->arch.process_table;
4983 	} else {
4984 		dw0 = PATB_HR | radix__get_tree_size() |
4985 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4986 		dw1 = PATB_GR | kvm->arch.process_table;
4987 	}
4988 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4989 }
4990 
4991 /*
4992  * Set up HPT (hashed page table) and RMA (real-mode area).
4993  * Must be called with kvm->arch.mmu_setup_lock held.
4994  */
4995 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4996 {
4997 	int err = 0;
4998 	struct kvm *kvm = vcpu->kvm;
4999 	unsigned long hva;
5000 	struct kvm_memory_slot *memslot;
5001 	struct vm_area_struct *vma;
5002 	unsigned long lpcr = 0, senc;
5003 	unsigned long psize, porder;
5004 	int srcu_idx;
5005 
5006 	/* Allocate hashed page table (if not done already) and reset it */
5007 	if (!kvm->arch.hpt.virt) {
5008 		int order = KVM_DEFAULT_HPT_ORDER;
5009 		struct kvm_hpt_info info;
5010 
5011 		err = kvmppc_allocate_hpt(&info, order);
5012 		/* If we get here, it means userspace didn't specify a
5013 		 * size explicitly.  So, try successively smaller
5014 		 * sizes if the default failed. */
5015 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5016 			err  = kvmppc_allocate_hpt(&info, order);
5017 
5018 		if (err < 0) {
5019 			pr_err("KVM: Couldn't alloc HPT\n");
5020 			goto out;
5021 		}
5022 
5023 		kvmppc_set_hpt(kvm, &info);
5024 	}
5025 
5026 	/* Look up the memslot for guest physical address 0 */
5027 	srcu_idx = srcu_read_lock(&kvm->srcu);
5028 	memslot = gfn_to_memslot(kvm, 0);
5029 
5030 	/* We must have some memory at 0 by now */
5031 	err = -EINVAL;
5032 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5033 		goto out_srcu;
5034 
5035 	/* Look up the VMA for the start of this memory slot */
5036 	hva = memslot->userspace_addr;
5037 	mmap_read_lock(kvm->mm);
5038 	vma = vma_lookup(kvm->mm, hva);
5039 	if (!vma || (vma->vm_flags & VM_IO))
5040 		goto up_out;
5041 
5042 	psize = vma_kernel_pagesize(vma);
5043 
5044 	mmap_read_unlock(kvm->mm);
5045 
5046 	/* We can handle 4k, 64k or 16M pages in the VRMA */
5047 	if (psize >= 0x1000000)
5048 		psize = 0x1000000;
5049 	else if (psize >= 0x10000)
5050 		psize = 0x10000;
5051 	else
5052 		psize = 0x1000;
5053 	porder = __ilog2(psize);
5054 
5055 	senc = slb_pgsize_encoding(psize);
5056 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5057 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5058 	/* Create HPTEs in the hash page table for the VRMA */
5059 	kvmppc_map_vrma(vcpu, memslot, porder);
5060 
5061 	/* Update VRMASD field in the LPCR */
5062 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5063 		/* the -4 is to account for senc values starting at 0x10 */
5064 		lpcr = senc << (LPCR_VRMASD_SH - 4);
5065 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5066 	}
5067 
5068 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5069 	smp_wmb();
5070 	err = 0;
5071  out_srcu:
5072 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5073  out:
5074 	return err;
5075 
5076  up_out:
5077 	mmap_read_unlock(kvm->mm);
5078 	goto out_srcu;
5079 }
5080 
5081 /*
5082  * Must be called with kvm->arch.mmu_setup_lock held and
5083  * mmu_ready = 0 and no vcpus running.
5084  */
5085 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5086 {
5087 	unsigned long lpcr, lpcr_mask;
5088 
5089 	if (nesting_enabled(kvm))
5090 		kvmhv_release_all_nested(kvm);
5091 	kvmppc_rmap_reset(kvm);
5092 	kvm->arch.process_table = 0;
5093 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5094 	spin_lock(&kvm->mmu_lock);
5095 	kvm->arch.radix = 0;
5096 	spin_unlock(&kvm->mmu_lock);
5097 	kvmppc_free_radix(kvm);
5098 
5099 	lpcr = LPCR_VPM1;
5100 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5101 	if (cpu_has_feature(CPU_FTR_ARCH_31))
5102 		lpcr_mask |= LPCR_HAIL;
5103 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5104 
5105 	return 0;
5106 }
5107 
5108 /*
5109  * Must be called with kvm->arch.mmu_setup_lock held and
5110  * mmu_ready = 0 and no vcpus running.
5111  */
5112 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5113 {
5114 	unsigned long lpcr, lpcr_mask;
5115 	int err;
5116 
5117 	err = kvmppc_init_vm_radix(kvm);
5118 	if (err)
5119 		return err;
5120 	kvmppc_rmap_reset(kvm);
5121 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5122 	spin_lock(&kvm->mmu_lock);
5123 	kvm->arch.radix = 1;
5124 	spin_unlock(&kvm->mmu_lock);
5125 	kvmppc_free_hpt(&kvm->arch.hpt);
5126 
5127 	lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5128 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5129 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5130 		lpcr_mask |= LPCR_HAIL;
5131 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5132 				(kvm->arch.host_lpcr & LPCR_HAIL))
5133 			lpcr |= LPCR_HAIL;
5134 	}
5135 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5136 
5137 	return 0;
5138 }
5139 
5140 #ifdef CONFIG_KVM_XICS
5141 /*
5142  * Allocate a per-core structure for managing state about which cores are
5143  * running in the host versus the guest and for exchanging data between
5144  * real mode KVM and CPU running in the host.
5145  * This is only done for the first VM.
5146  * The allocated structure stays even if all VMs have stopped.
5147  * It is only freed when the kvm-hv module is unloaded.
5148  * It's OK for this routine to fail, we just don't support host
5149  * core operations like redirecting H_IPI wakeups.
5150  */
5151 void kvmppc_alloc_host_rm_ops(void)
5152 {
5153 	struct kvmppc_host_rm_ops *ops;
5154 	unsigned long l_ops;
5155 	int cpu, core;
5156 	int size;
5157 
5158 	if (cpu_has_feature(CPU_FTR_ARCH_300))
5159 		return;
5160 
5161 	/* Not the first time here ? */
5162 	if (kvmppc_host_rm_ops_hv != NULL)
5163 		return;
5164 
5165 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5166 	if (!ops)
5167 		return;
5168 
5169 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5170 	ops->rm_core = kzalloc(size, GFP_KERNEL);
5171 
5172 	if (!ops->rm_core) {
5173 		kfree(ops);
5174 		return;
5175 	}
5176 
5177 	cpus_read_lock();
5178 
5179 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5180 		if (!cpu_online(cpu))
5181 			continue;
5182 
5183 		core = cpu >> threads_shift;
5184 		ops->rm_core[core].rm_state.in_host = 1;
5185 	}
5186 
5187 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5188 
5189 	/*
5190 	 * Make the contents of the kvmppc_host_rm_ops structure visible
5191 	 * to other CPUs before we assign it to the global variable.
5192 	 * Do an atomic assignment (no locks used here), but if someone
5193 	 * beats us to it, just free our copy and return.
5194 	 */
5195 	smp_wmb();
5196 	l_ops = (unsigned long) ops;
5197 
5198 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5199 		cpus_read_unlock();
5200 		kfree(ops->rm_core);
5201 		kfree(ops);
5202 		return;
5203 	}
5204 
5205 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5206 					     "ppc/kvm_book3s:prepare",
5207 					     kvmppc_set_host_core,
5208 					     kvmppc_clear_host_core);
5209 	cpus_read_unlock();
5210 }
5211 
5212 void kvmppc_free_host_rm_ops(void)
5213 {
5214 	if (kvmppc_host_rm_ops_hv) {
5215 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5216 		kfree(kvmppc_host_rm_ops_hv->rm_core);
5217 		kfree(kvmppc_host_rm_ops_hv);
5218 		kvmppc_host_rm_ops_hv = NULL;
5219 	}
5220 }
5221 #endif
5222 
5223 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5224 {
5225 	unsigned long lpcr, lpid;
5226 	char buf[32];
5227 	int ret;
5228 
5229 	mutex_init(&kvm->arch.uvmem_lock);
5230 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5231 	mutex_init(&kvm->arch.mmu_setup_lock);
5232 
5233 	/* Allocate the guest's logical partition ID */
5234 
5235 	lpid = kvmppc_alloc_lpid();
5236 	if ((long)lpid < 0)
5237 		return -ENOMEM;
5238 	kvm->arch.lpid = lpid;
5239 
5240 	kvmppc_alloc_host_rm_ops();
5241 
5242 	kvmhv_vm_nested_init(kvm);
5243 
5244 	/*
5245 	 * Since we don't flush the TLB when tearing down a VM,
5246 	 * and this lpid might have previously been used,
5247 	 * make sure we flush on each core before running the new VM.
5248 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5249 	 * does this flush for us.
5250 	 */
5251 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5252 		cpumask_setall(&kvm->arch.need_tlb_flush);
5253 
5254 	/* Start out with the default set of hcalls enabled */
5255 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5256 	       sizeof(kvm->arch.enabled_hcalls));
5257 
5258 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5259 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5260 
5261 	/* Init LPCR for virtual RMA mode */
5262 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
5263 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
5264 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5265 		lpcr &= LPCR_PECE | LPCR_LPES;
5266 	} else {
5267 		lpcr = 0;
5268 	}
5269 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5270 		LPCR_VPM0 | LPCR_VPM1;
5271 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5272 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5273 	/* On POWER8 turn on online bit to enable PURR/SPURR */
5274 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
5275 		lpcr |= LPCR_ONL;
5276 	/*
5277 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5278 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
5279 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5280 	 * be unnecessary but better safe than sorry in case we re-enable
5281 	 * EE in HV mode with this LPCR still set)
5282 	 */
5283 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5284 		lpcr &= ~LPCR_VPM0;
5285 		lpcr |= LPCR_HVICE | LPCR_HEIC;
5286 
5287 		/*
5288 		 * If xive is enabled, we route 0x500 interrupts directly
5289 		 * to the guest.
5290 		 */
5291 		if (xics_on_xive())
5292 			lpcr |= LPCR_LPES;
5293 	}
5294 
5295 	/*
5296 	 * If the host uses radix, the guest starts out as radix.
5297 	 */
5298 	if (radix_enabled()) {
5299 		kvm->arch.radix = 1;
5300 		kvm->arch.mmu_ready = 1;
5301 		lpcr &= ~LPCR_VPM1;
5302 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5303 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5304 		    cpu_has_feature(CPU_FTR_ARCH_31) &&
5305 		    (kvm->arch.host_lpcr & LPCR_HAIL))
5306 			lpcr |= LPCR_HAIL;
5307 		ret = kvmppc_init_vm_radix(kvm);
5308 		if (ret) {
5309 			kvmppc_free_lpid(kvm->arch.lpid);
5310 			return ret;
5311 		}
5312 		kvmppc_setup_partition_table(kvm);
5313 	}
5314 
5315 	verify_lpcr(kvm, lpcr);
5316 	kvm->arch.lpcr = lpcr;
5317 
5318 	/* Initialization for future HPT resizes */
5319 	kvm->arch.resize_hpt = NULL;
5320 
5321 	/*
5322 	 * Work out how many sets the TLB has, for the use of
5323 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5324 	 */
5325 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5326 		/*
5327 		 * P10 will flush all the congruence class with a single tlbiel
5328 		 */
5329 		kvm->arch.tlb_sets = 1;
5330 	} else if (radix_enabled())
5331 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
5332 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
5333 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
5334 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5335 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
5336 	else
5337 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
5338 
5339 	/*
5340 	 * Track that we now have a HV mode VM active. This blocks secondary
5341 	 * CPU threads from coming online.
5342 	 */
5343 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5344 		kvm_hv_vm_activated();
5345 
5346 	/*
5347 	 * Initialize smt_mode depending on processor.
5348 	 * POWER8 and earlier have to use "strict" threading, where
5349 	 * all vCPUs in a vcore have to run on the same (sub)core,
5350 	 * whereas on POWER9 the threads can each run a different
5351 	 * guest.
5352 	 */
5353 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5354 		kvm->arch.smt_mode = threads_per_subcore;
5355 	else
5356 		kvm->arch.smt_mode = 1;
5357 	kvm->arch.emul_smt_mode = 1;
5358 
5359 	/*
5360 	 * Create a debugfs directory for the VM
5361 	 */
5362 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
5363 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5364 	kvmppc_mmu_debugfs_init(kvm);
5365 	if (radix_enabled())
5366 		kvmhv_radix_debugfs_init(kvm);
5367 
5368 	return 0;
5369 }
5370 
5371 static void kvmppc_free_vcores(struct kvm *kvm)
5372 {
5373 	long int i;
5374 
5375 	for (i = 0; i < KVM_MAX_VCORES; ++i)
5376 		kfree(kvm->arch.vcores[i]);
5377 	kvm->arch.online_vcores = 0;
5378 }
5379 
5380 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5381 {
5382 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
5383 
5384 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5385 		kvm_hv_vm_deactivated();
5386 
5387 	kvmppc_free_vcores(kvm);
5388 
5389 
5390 	if (kvm_is_radix(kvm))
5391 		kvmppc_free_radix(kvm);
5392 	else
5393 		kvmppc_free_hpt(&kvm->arch.hpt);
5394 
5395 	/* Perform global invalidation and return lpid to the pool */
5396 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5397 		if (nesting_enabled(kvm))
5398 			kvmhv_release_all_nested(kvm);
5399 		kvm->arch.process_table = 0;
5400 		if (kvm->arch.secure_guest)
5401 			uv_svm_terminate(kvm->arch.lpid);
5402 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5403 	}
5404 
5405 	kvmppc_free_lpid(kvm->arch.lpid);
5406 
5407 	kvmppc_free_pimap(kvm);
5408 }
5409 
5410 /* We don't need to emulate any privileged instructions or dcbz */
5411 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5412 				     unsigned int inst, int *advance)
5413 {
5414 	return EMULATE_FAIL;
5415 }
5416 
5417 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5418 					ulong spr_val)
5419 {
5420 	return EMULATE_FAIL;
5421 }
5422 
5423 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5424 					ulong *spr_val)
5425 {
5426 	return EMULATE_FAIL;
5427 }
5428 
5429 static int kvmppc_core_check_processor_compat_hv(void)
5430 {
5431 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5432 	    cpu_has_feature(CPU_FTR_ARCH_206))
5433 		return 0;
5434 
5435 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5436 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5437 		return 0;
5438 
5439 	return -EIO;
5440 }
5441 
5442 #ifdef CONFIG_KVM_XICS
5443 
5444 void kvmppc_free_pimap(struct kvm *kvm)
5445 {
5446 	kfree(kvm->arch.pimap);
5447 }
5448 
5449 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5450 {
5451 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5452 }
5453 
5454 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5455 {
5456 	struct irq_desc *desc;
5457 	struct kvmppc_irq_map *irq_map;
5458 	struct kvmppc_passthru_irqmap *pimap;
5459 	struct irq_chip *chip;
5460 	int i, rc = 0;
5461 	struct irq_data *host_data;
5462 
5463 	if (!kvm_irq_bypass)
5464 		return 1;
5465 
5466 	desc = irq_to_desc(host_irq);
5467 	if (!desc)
5468 		return -EIO;
5469 
5470 	mutex_lock(&kvm->lock);
5471 
5472 	pimap = kvm->arch.pimap;
5473 	if (pimap == NULL) {
5474 		/* First call, allocate structure to hold IRQ map */
5475 		pimap = kvmppc_alloc_pimap();
5476 		if (pimap == NULL) {
5477 			mutex_unlock(&kvm->lock);
5478 			return -ENOMEM;
5479 		}
5480 		kvm->arch.pimap = pimap;
5481 	}
5482 
5483 	/*
5484 	 * For now, we only support interrupts for which the EOI operation
5485 	 * is an OPAL call followed by a write to XIRR, since that's
5486 	 * what our real-mode EOI code does, or a XIVE interrupt
5487 	 */
5488 	chip = irq_data_get_irq_chip(&desc->irq_data);
5489 	if (!chip || !is_pnv_opal_msi(chip)) {
5490 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5491 			host_irq, guest_gsi);
5492 		mutex_unlock(&kvm->lock);
5493 		return -ENOENT;
5494 	}
5495 
5496 	/*
5497 	 * See if we already have an entry for this guest IRQ number.
5498 	 * If it's mapped to a hardware IRQ number, that's an error,
5499 	 * otherwise re-use this entry.
5500 	 */
5501 	for (i = 0; i < pimap->n_mapped; i++) {
5502 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5503 			if (pimap->mapped[i].r_hwirq) {
5504 				mutex_unlock(&kvm->lock);
5505 				return -EINVAL;
5506 			}
5507 			break;
5508 		}
5509 	}
5510 
5511 	if (i == KVMPPC_PIRQ_MAPPED) {
5512 		mutex_unlock(&kvm->lock);
5513 		return -EAGAIN;		/* table is full */
5514 	}
5515 
5516 	irq_map = &pimap->mapped[i];
5517 
5518 	irq_map->v_hwirq = guest_gsi;
5519 	irq_map->desc = desc;
5520 
5521 	/*
5522 	 * Order the above two stores before the next to serialize with
5523 	 * the KVM real mode handler.
5524 	 */
5525 	smp_wmb();
5526 
5527 	/*
5528 	 * The 'host_irq' number is mapped in the PCI-MSI domain but
5529 	 * the underlying calls, which will EOI the interrupt in real
5530 	 * mode, need an HW IRQ number mapped in the XICS IRQ domain.
5531 	 */
5532 	host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq);
5533 	irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
5534 
5535 	if (i == pimap->n_mapped)
5536 		pimap->n_mapped++;
5537 
5538 	if (xics_on_xive())
5539 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
5540 	else
5541 		kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
5542 	if (rc)
5543 		irq_map->r_hwirq = 0;
5544 
5545 	mutex_unlock(&kvm->lock);
5546 
5547 	return 0;
5548 }
5549 
5550 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5551 {
5552 	struct irq_desc *desc;
5553 	struct kvmppc_passthru_irqmap *pimap;
5554 	int i, rc = 0;
5555 
5556 	if (!kvm_irq_bypass)
5557 		return 0;
5558 
5559 	desc = irq_to_desc(host_irq);
5560 	if (!desc)
5561 		return -EIO;
5562 
5563 	mutex_lock(&kvm->lock);
5564 	if (!kvm->arch.pimap)
5565 		goto unlock;
5566 
5567 	pimap = kvm->arch.pimap;
5568 
5569 	for (i = 0; i < pimap->n_mapped; i++) {
5570 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5571 			break;
5572 	}
5573 
5574 	if (i == pimap->n_mapped) {
5575 		mutex_unlock(&kvm->lock);
5576 		return -ENODEV;
5577 	}
5578 
5579 	if (xics_on_xive())
5580 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
5581 	else
5582 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5583 
5584 	/* invalidate the entry (what do do on error from the above ?) */
5585 	pimap->mapped[i].r_hwirq = 0;
5586 
5587 	/*
5588 	 * We don't free this structure even when the count goes to
5589 	 * zero. The structure is freed when we destroy the VM.
5590 	 */
5591  unlock:
5592 	mutex_unlock(&kvm->lock);
5593 	return rc;
5594 }
5595 
5596 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5597 					     struct irq_bypass_producer *prod)
5598 {
5599 	int ret = 0;
5600 	struct kvm_kernel_irqfd *irqfd =
5601 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5602 
5603 	irqfd->producer = prod;
5604 
5605 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5606 	if (ret)
5607 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5608 			prod->irq, irqfd->gsi, ret);
5609 
5610 	return ret;
5611 }
5612 
5613 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5614 					      struct irq_bypass_producer *prod)
5615 {
5616 	int ret;
5617 	struct kvm_kernel_irqfd *irqfd =
5618 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5619 
5620 	irqfd->producer = NULL;
5621 
5622 	/*
5623 	 * When producer of consumer is unregistered, we change back to
5624 	 * default external interrupt handling mode - KVM real mode
5625 	 * will switch back to host.
5626 	 */
5627 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5628 	if (ret)
5629 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5630 			prod->irq, irqfd->gsi, ret);
5631 }
5632 #endif
5633 
5634 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5635 				 unsigned int ioctl, unsigned long arg)
5636 {
5637 	struct kvm *kvm __maybe_unused = filp->private_data;
5638 	void __user *argp = (void __user *)arg;
5639 	long r;
5640 
5641 	switch (ioctl) {
5642 
5643 	case KVM_PPC_ALLOCATE_HTAB: {
5644 		u32 htab_order;
5645 
5646 		/* If we're a nested hypervisor, we currently only support radix */
5647 		if (kvmhv_on_pseries()) {
5648 			r = -EOPNOTSUPP;
5649 			break;
5650 		}
5651 
5652 		r = -EFAULT;
5653 		if (get_user(htab_order, (u32 __user *)argp))
5654 			break;
5655 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5656 		if (r)
5657 			break;
5658 		r = 0;
5659 		break;
5660 	}
5661 
5662 	case KVM_PPC_GET_HTAB_FD: {
5663 		struct kvm_get_htab_fd ghf;
5664 
5665 		r = -EFAULT;
5666 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5667 			break;
5668 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5669 		break;
5670 	}
5671 
5672 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5673 		struct kvm_ppc_resize_hpt rhpt;
5674 
5675 		r = -EFAULT;
5676 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5677 			break;
5678 
5679 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5680 		break;
5681 	}
5682 
5683 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5684 		struct kvm_ppc_resize_hpt rhpt;
5685 
5686 		r = -EFAULT;
5687 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5688 			break;
5689 
5690 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5691 		break;
5692 	}
5693 
5694 	default:
5695 		r = -ENOTTY;
5696 	}
5697 
5698 	return r;
5699 }
5700 
5701 /*
5702  * List of hcall numbers to enable by default.
5703  * For compatibility with old userspace, we enable by default
5704  * all hcalls that were implemented before the hcall-enabling
5705  * facility was added.  Note this list should not include H_RTAS.
5706  */
5707 static unsigned int default_hcall_list[] = {
5708 	H_REMOVE,
5709 	H_ENTER,
5710 	H_READ,
5711 	H_PROTECT,
5712 	H_BULK_REMOVE,
5713 #ifdef CONFIG_SPAPR_TCE_IOMMU
5714 	H_GET_TCE,
5715 	H_PUT_TCE,
5716 #endif
5717 	H_SET_DABR,
5718 	H_SET_XDABR,
5719 	H_CEDE,
5720 	H_PROD,
5721 	H_CONFER,
5722 	H_REGISTER_VPA,
5723 #ifdef CONFIG_KVM_XICS
5724 	H_EOI,
5725 	H_CPPR,
5726 	H_IPI,
5727 	H_IPOLL,
5728 	H_XIRR,
5729 	H_XIRR_X,
5730 #endif
5731 	0
5732 };
5733 
5734 static void init_default_hcalls(void)
5735 {
5736 	int i;
5737 	unsigned int hcall;
5738 
5739 	for (i = 0; default_hcall_list[i]; ++i) {
5740 		hcall = default_hcall_list[i];
5741 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5742 		__set_bit(hcall / 4, default_enabled_hcalls);
5743 	}
5744 }
5745 
5746 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5747 {
5748 	unsigned long lpcr;
5749 	int radix;
5750 	int err;
5751 
5752 	/* If not on a POWER9, reject it */
5753 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5754 		return -ENODEV;
5755 
5756 	/* If any unknown flags set, reject it */
5757 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5758 		return -EINVAL;
5759 
5760 	/* GR (guest radix) bit in process_table field must match */
5761 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5762 	if (!!(cfg->process_table & PATB_GR) != radix)
5763 		return -EINVAL;
5764 
5765 	/* Process table size field must be reasonable, i.e. <= 24 */
5766 	if ((cfg->process_table & PRTS_MASK) > 24)
5767 		return -EINVAL;
5768 
5769 	/* We can change a guest to/from radix now, if the host is radix */
5770 	if (radix && !radix_enabled())
5771 		return -EINVAL;
5772 
5773 	/* If we're a nested hypervisor, we currently only support radix */
5774 	if (kvmhv_on_pseries() && !radix)
5775 		return -EINVAL;
5776 
5777 	mutex_lock(&kvm->arch.mmu_setup_lock);
5778 	if (radix != kvm_is_radix(kvm)) {
5779 		if (kvm->arch.mmu_ready) {
5780 			kvm->arch.mmu_ready = 0;
5781 			/* order mmu_ready vs. vcpus_running */
5782 			smp_mb();
5783 			if (atomic_read(&kvm->arch.vcpus_running)) {
5784 				kvm->arch.mmu_ready = 1;
5785 				err = -EBUSY;
5786 				goto out_unlock;
5787 			}
5788 		}
5789 		if (radix)
5790 			err = kvmppc_switch_mmu_to_radix(kvm);
5791 		else
5792 			err = kvmppc_switch_mmu_to_hpt(kvm);
5793 		if (err)
5794 			goto out_unlock;
5795 	}
5796 
5797 	kvm->arch.process_table = cfg->process_table;
5798 	kvmppc_setup_partition_table(kvm);
5799 
5800 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5801 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5802 	err = 0;
5803 
5804  out_unlock:
5805 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5806 	return err;
5807 }
5808 
5809 static int kvmhv_enable_nested(struct kvm *kvm)
5810 {
5811 	if (!nested)
5812 		return -EPERM;
5813 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5814 		return -ENODEV;
5815 	if (!radix_enabled())
5816 		return -ENODEV;
5817 
5818 	/* kvm == NULL means the caller is testing if the capability exists */
5819 	if (kvm)
5820 		kvm->arch.nested_enable = true;
5821 	return 0;
5822 }
5823 
5824 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5825 				 int size)
5826 {
5827 	int rc = -EINVAL;
5828 
5829 	if (kvmhv_vcpu_is_radix(vcpu)) {
5830 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5831 
5832 		if (rc > 0)
5833 			rc = -EINVAL;
5834 	}
5835 
5836 	/* For now quadrants are the only way to access nested guest memory */
5837 	if (rc && vcpu->arch.nested)
5838 		rc = -EAGAIN;
5839 
5840 	return rc;
5841 }
5842 
5843 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5844 				int size)
5845 {
5846 	int rc = -EINVAL;
5847 
5848 	if (kvmhv_vcpu_is_radix(vcpu)) {
5849 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5850 
5851 		if (rc > 0)
5852 			rc = -EINVAL;
5853 	}
5854 
5855 	/* For now quadrants are the only way to access nested guest memory */
5856 	if (rc && vcpu->arch.nested)
5857 		rc = -EAGAIN;
5858 
5859 	return rc;
5860 }
5861 
5862 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5863 {
5864 	unpin_vpa(kvm, vpa);
5865 	vpa->gpa = 0;
5866 	vpa->pinned_addr = NULL;
5867 	vpa->dirty = false;
5868 	vpa->update_pending = 0;
5869 }
5870 
5871 /*
5872  * Enable a guest to become a secure VM, or test whether
5873  * that could be enabled.
5874  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5875  * tested (kvm == NULL) or enabled (kvm != NULL).
5876  */
5877 static int kvmhv_enable_svm(struct kvm *kvm)
5878 {
5879 	if (!kvmppc_uvmem_available())
5880 		return -EINVAL;
5881 	if (kvm)
5882 		kvm->arch.svm_enabled = 1;
5883 	return 0;
5884 }
5885 
5886 /*
5887  *  IOCTL handler to turn off secure mode of guest
5888  *
5889  * - Release all device pages
5890  * - Issue ucall to terminate the guest on the UV side
5891  * - Unpin the VPA pages.
5892  * - Reinit the partition scoped page tables
5893  */
5894 static int kvmhv_svm_off(struct kvm *kvm)
5895 {
5896 	struct kvm_vcpu *vcpu;
5897 	int mmu_was_ready;
5898 	int srcu_idx;
5899 	int ret = 0;
5900 	unsigned long i;
5901 
5902 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5903 		return ret;
5904 
5905 	mutex_lock(&kvm->arch.mmu_setup_lock);
5906 	mmu_was_ready = kvm->arch.mmu_ready;
5907 	if (kvm->arch.mmu_ready) {
5908 		kvm->arch.mmu_ready = 0;
5909 		/* order mmu_ready vs. vcpus_running */
5910 		smp_mb();
5911 		if (atomic_read(&kvm->arch.vcpus_running)) {
5912 			kvm->arch.mmu_ready = 1;
5913 			ret = -EBUSY;
5914 			goto out;
5915 		}
5916 	}
5917 
5918 	srcu_idx = srcu_read_lock(&kvm->srcu);
5919 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5920 		struct kvm_memory_slot *memslot;
5921 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5922 		int bkt;
5923 
5924 		if (!slots)
5925 			continue;
5926 
5927 		kvm_for_each_memslot(memslot, bkt, slots) {
5928 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
5929 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5930 		}
5931 	}
5932 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5933 
5934 	ret = uv_svm_terminate(kvm->arch.lpid);
5935 	if (ret != U_SUCCESS) {
5936 		ret = -EINVAL;
5937 		goto out;
5938 	}
5939 
5940 	/*
5941 	 * When secure guest is reset, all the guest pages are sent
5942 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
5943 	 * chance to run and unpin their VPA pages. Unpinning of all
5944 	 * VPA pages is done here explicitly so that VPA pages
5945 	 * can be migrated to the secure side.
5946 	 *
5947 	 * This is required to for the secure SMP guest to reboot
5948 	 * correctly.
5949 	 */
5950 	kvm_for_each_vcpu(i, vcpu, kvm) {
5951 		spin_lock(&vcpu->arch.vpa_update_lock);
5952 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5953 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5954 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5955 		spin_unlock(&vcpu->arch.vpa_update_lock);
5956 	}
5957 
5958 	kvmppc_setup_partition_table(kvm);
5959 	kvm->arch.secure_guest = 0;
5960 	kvm->arch.mmu_ready = mmu_was_ready;
5961 out:
5962 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5963 	return ret;
5964 }
5965 
5966 static int kvmhv_enable_dawr1(struct kvm *kvm)
5967 {
5968 	if (!cpu_has_feature(CPU_FTR_DAWR1))
5969 		return -ENODEV;
5970 
5971 	/* kvm == NULL means the caller is testing if the capability exists */
5972 	if (kvm)
5973 		kvm->arch.dawr1_enabled = true;
5974 	return 0;
5975 }
5976 
5977 static bool kvmppc_hash_v3_possible(void)
5978 {
5979 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5980 		return false;
5981 
5982 	if (!cpu_has_feature(CPU_FTR_HVMODE))
5983 		return false;
5984 
5985 	/*
5986 	 * POWER9 chips before version 2.02 can't have some threads in
5987 	 * HPT mode and some in radix mode on the same core.
5988 	 */
5989 	if (radix_enabled()) {
5990 		unsigned int pvr = mfspr(SPRN_PVR);
5991 		if ((pvr >> 16) == PVR_POWER9 &&
5992 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5993 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5994 			return false;
5995 	}
5996 
5997 	return true;
5998 }
5999 
6000 static struct kvmppc_ops kvm_ops_hv = {
6001 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
6002 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
6003 	.get_one_reg = kvmppc_get_one_reg_hv,
6004 	.set_one_reg = kvmppc_set_one_reg_hv,
6005 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
6006 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
6007 	.inject_interrupt = kvmppc_inject_interrupt_hv,
6008 	.set_msr     = kvmppc_set_msr_hv,
6009 	.vcpu_run    = kvmppc_vcpu_run_hv,
6010 	.vcpu_create = kvmppc_core_vcpu_create_hv,
6011 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
6012 	.check_requests = kvmppc_core_check_requests_hv,
6013 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
6014 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
6015 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
6016 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
6017 	.unmap_gfn_range = kvm_unmap_gfn_range_hv,
6018 	.age_gfn = kvm_age_gfn_hv,
6019 	.test_age_gfn = kvm_test_age_gfn_hv,
6020 	.set_spte_gfn = kvm_set_spte_gfn_hv,
6021 	.free_memslot = kvmppc_core_free_memslot_hv,
6022 	.init_vm =  kvmppc_core_init_vm_hv,
6023 	.destroy_vm = kvmppc_core_destroy_vm_hv,
6024 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
6025 	.emulate_op = kvmppc_core_emulate_op_hv,
6026 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
6027 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
6028 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
6029 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
6030 	.hcall_implemented = kvmppc_hcall_impl_hv,
6031 #ifdef CONFIG_KVM_XICS
6032 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
6033 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
6034 #endif
6035 	.configure_mmu = kvmhv_configure_mmu,
6036 	.get_rmmu_info = kvmhv_get_rmmu_info,
6037 	.set_smt_mode = kvmhv_set_smt_mode,
6038 	.enable_nested = kvmhv_enable_nested,
6039 	.load_from_eaddr = kvmhv_load_from_eaddr,
6040 	.store_to_eaddr = kvmhv_store_to_eaddr,
6041 	.enable_svm = kvmhv_enable_svm,
6042 	.svm_off = kvmhv_svm_off,
6043 	.enable_dawr1 = kvmhv_enable_dawr1,
6044 	.hash_v3_possible = kvmppc_hash_v3_possible,
6045 };
6046 
6047 static int kvm_init_subcore_bitmap(void)
6048 {
6049 	int i, j;
6050 	int nr_cores = cpu_nr_cores();
6051 	struct sibling_subcore_state *sibling_subcore_state;
6052 
6053 	for (i = 0; i < nr_cores; i++) {
6054 		int first_cpu = i * threads_per_core;
6055 		int node = cpu_to_node(first_cpu);
6056 
6057 		/* Ignore if it is already allocated. */
6058 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
6059 			continue;
6060 
6061 		sibling_subcore_state =
6062 			kzalloc_node(sizeof(struct sibling_subcore_state),
6063 							GFP_KERNEL, node);
6064 		if (!sibling_subcore_state)
6065 			return -ENOMEM;
6066 
6067 
6068 		for (j = 0; j < threads_per_core; j++) {
6069 			int cpu = first_cpu + j;
6070 
6071 			paca_ptrs[cpu]->sibling_subcore_state =
6072 						sibling_subcore_state;
6073 		}
6074 	}
6075 	return 0;
6076 }
6077 
6078 static int kvmppc_radix_possible(void)
6079 {
6080 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6081 }
6082 
6083 static int kvmppc_book3s_init_hv(void)
6084 {
6085 	int r;
6086 
6087 	if (!tlbie_capable) {
6088 		pr_err("KVM-HV: Host does not support TLBIE\n");
6089 		return -ENODEV;
6090 	}
6091 
6092 	/*
6093 	 * FIXME!! Do we need to check on all cpus ?
6094 	 */
6095 	r = kvmppc_core_check_processor_compat_hv();
6096 	if (r < 0)
6097 		return -ENODEV;
6098 
6099 	r = kvmhv_nested_init();
6100 	if (r)
6101 		return r;
6102 
6103 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
6104 		r = kvm_init_subcore_bitmap();
6105 		if (r)
6106 			return r;
6107 	}
6108 
6109 	/*
6110 	 * We need a way of accessing the XICS interrupt controller,
6111 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6112 	 * indirectly, via OPAL.
6113 	 */
6114 #ifdef CONFIG_SMP
6115 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
6116 	    !local_paca->kvm_hstate.xics_phys) {
6117 		struct device_node *np;
6118 
6119 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6120 		if (!np) {
6121 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6122 			return -ENODEV;
6123 		}
6124 		/* presence of intc confirmed - node can be dropped again */
6125 		of_node_put(np);
6126 	}
6127 #endif
6128 
6129 	kvm_ops_hv.owner = THIS_MODULE;
6130 	kvmppc_hv_ops = &kvm_ops_hv;
6131 
6132 	init_default_hcalls();
6133 
6134 	init_vcore_lists();
6135 
6136 	r = kvmppc_mmu_hv_init();
6137 	if (r)
6138 		return r;
6139 
6140 	if (kvmppc_radix_possible())
6141 		r = kvmppc_radix_init();
6142 
6143 	r = kvmppc_uvmem_init();
6144 	if (r < 0)
6145 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6146 
6147 	return r;
6148 }
6149 
6150 static void kvmppc_book3s_exit_hv(void)
6151 {
6152 	kvmppc_uvmem_free();
6153 	kvmppc_free_host_rm_ops();
6154 	if (kvmppc_radix_possible())
6155 		kvmppc_radix_exit();
6156 	kvmppc_hv_ops = NULL;
6157 	kvmhv_nested_exit();
6158 }
6159 
6160 module_init(kvmppc_book3s_init_hv);
6161 module_exit(kvmppc_book3s_exit_hv);
6162 MODULE_LICENSE("GPL");
6163 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6164 MODULE_ALIAS("devname:kvm");
6165