xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 36fe4655)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
5  *
6  * Authors:
7  *    Paul Mackerras <paulus@au1.ibm.com>
8  *    Alexander Graf <agraf@suse.de>
9  *    Kevin Wolf <mail@kevin-wolf.de>
10  *
11  * Description: KVM functions specific to running on Book 3S
12  * processors in hypervisor mode (specifically POWER7 and later).
13  *
14  * This file is derived from arch/powerpc/kvm/book3s.c,
15  * by Alexander Graf <agraf@suse.de>.
16  */
17 
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 #include <linux/err.h>
21 #include <linux/slab.h>
22 #include <linux/preempt.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/stat.h>
25 #include <linux/delay.h>
26 #include <linux/export.h>
27 #include <linux/fs.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpu.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 #include <linux/gfp.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/hugetlb.h>
40 #include <linux/kvm_irqfd.h>
41 #include <linux/irqbypass.h>
42 #include <linux/module.h>
43 #include <linux/compiler.h>
44 #include <linux/of.h>
45 
46 #include <asm/ftrace.h>
47 #include <asm/reg.h>
48 #include <asm/ppc-opcode.h>
49 #include <asm/asm-prototypes.h>
50 #include <asm/archrandom.h>
51 #include <asm/debug.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <linux/uaccess.h>
56 #include <asm/io.h>
57 #include <asm/kvm_ppc.h>
58 #include <asm/kvm_book3s.h>
59 #include <asm/mmu_context.h>
60 #include <asm/lppaca.h>
61 #include <asm/processor.h>
62 #include <asm/cputhreads.h>
63 #include <asm/page.h>
64 #include <asm/hvcall.h>
65 #include <asm/switch_to.h>
66 #include <asm/smp.h>
67 #include <asm/dbell.h>
68 #include <asm/hmi.h>
69 #include <asm/pnv-pci.h>
70 #include <asm/mmu.h>
71 #include <asm/opal.h>
72 #include <asm/xics.h>
73 #include <asm/xive.h>
74 #include <asm/hw_breakpoint.h>
75 #include <asm/kvm_book3s_uvmem.h>
76 #include <asm/ultravisor.h>
77 #include <asm/dtl.h>
78 
79 #include "book3s.h"
80 
81 #define CREATE_TRACE_POINTS
82 #include "trace_hv.h"
83 
84 /* #define EXIT_DEBUG */
85 /* #define EXIT_DEBUG_SIMPLE */
86 /* #define EXIT_DEBUG_INT */
87 
88 /* Used to indicate that a guest page fault needs to be handled */
89 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
90 /* Used to indicate that a guest passthrough interrupt needs to be handled */
91 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
92 
93 /* Used as a "null" value for timebase values */
94 #define TB_NIL	(~(u64)0)
95 
96 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
97 
98 static int dynamic_mt_modes = 6;
99 module_param(dynamic_mt_modes, int, 0644);
100 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
101 static int target_smt_mode;
102 module_param(target_smt_mode, int, 0644);
103 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
104 
105 static bool indep_threads_mode = true;
106 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
107 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
108 
109 static bool one_vm_per_core;
110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
112 
113 #ifdef CONFIG_KVM_XICS
114 static 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 inline bool nesting_enabled(struct kvm *kvm)
132 {
133 	return kvm->arch.nested_enable && kvm_is_radix(kvm);
134 }
135 
136 /* If set, the threads on each CPU core have to be in the same MMU mode */
137 static bool no_mixing_hpt_and_radix;
138 
139 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
140 
141 /*
142  * RWMR values for POWER8.  These control the rate at which PURR
143  * and SPURR count and should be set according to the number of
144  * online threads in the vcore being run.
145  */
146 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
147 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
148 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
149 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
150 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
151 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
152 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
153 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
154 
155 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
156 	RWMR_RPA_P8_1THREAD,
157 	RWMR_RPA_P8_1THREAD,
158 	RWMR_RPA_P8_2THREAD,
159 	RWMR_RPA_P8_3THREAD,
160 	RWMR_RPA_P8_4THREAD,
161 	RWMR_RPA_P8_5THREAD,
162 	RWMR_RPA_P8_6THREAD,
163 	RWMR_RPA_P8_7THREAD,
164 	RWMR_RPA_P8_8THREAD,
165 };
166 
167 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
168 		int *ip)
169 {
170 	int i = *ip;
171 	struct kvm_vcpu *vcpu;
172 
173 	while (++i < MAX_SMT_THREADS) {
174 		vcpu = READ_ONCE(vc->runnable_threads[i]);
175 		if (vcpu) {
176 			*ip = i;
177 			return vcpu;
178 		}
179 	}
180 	return NULL;
181 }
182 
183 /* Used to traverse the list of runnable threads for a given vcore */
184 #define for_each_runnable_thread(i, vcpu, vc) \
185 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
186 
187 static bool kvmppc_ipi_thread(int cpu)
188 {
189 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
190 
191 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
192 	if (kvmhv_on_pseries())
193 		return false;
194 
195 	/* On POWER9 we can use msgsnd to IPI any cpu */
196 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
197 		msg |= get_hard_smp_processor_id(cpu);
198 		smp_mb();
199 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
200 		return true;
201 	}
202 
203 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
204 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
205 		preempt_disable();
206 		if (cpu_first_thread_sibling(cpu) ==
207 		    cpu_first_thread_sibling(smp_processor_id())) {
208 			msg |= cpu_thread_in_core(cpu);
209 			smp_mb();
210 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
211 			preempt_enable();
212 			return true;
213 		}
214 		preempt_enable();
215 	}
216 
217 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
218 	if (cpu >= 0 && cpu < nr_cpu_ids) {
219 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
220 			xics_wake_cpu(cpu);
221 			return true;
222 		}
223 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
224 		return true;
225 	}
226 #endif
227 
228 	return false;
229 }
230 
231 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
232 {
233 	int cpu;
234 	struct rcuwait *waitp;
235 
236 	waitp = kvm_arch_vcpu_get_wait(vcpu);
237 	if (rcuwait_wake_up(waitp))
238 		++vcpu->stat.halt_wakeup;
239 
240 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
241 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
242 		return;
243 
244 	/* CPU points to the first thread of the core */
245 	cpu = vcpu->cpu;
246 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
247 		smp_send_reschedule(cpu);
248 }
249 
250 /*
251  * We use the vcpu_load/put functions to measure stolen time.
252  * Stolen time is counted as time when either the vcpu is able to
253  * run as part of a virtual core, but the task running the vcore
254  * is preempted or sleeping, or when the vcpu needs something done
255  * in the kernel by the task running the vcpu, but that task is
256  * preempted or sleeping.  Those two things have to be counted
257  * separately, since one of the vcpu tasks will take on the job
258  * of running the core, and the other vcpu tasks in the vcore will
259  * sleep waiting for it to do that, but that sleep shouldn't count
260  * as stolen time.
261  *
262  * Hence we accumulate stolen time when the vcpu can run as part of
263  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
264  * needs its task to do other things in the kernel (for example,
265  * service a page fault) in busy_stolen.  We don't accumulate
266  * stolen time for a vcore when it is inactive, or for a vcpu
267  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
268  * a misnomer; it means that the vcpu task is not executing in
269  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
270  * the kernel.  We don't have any way of dividing up that time
271  * between time that the vcpu is genuinely stopped, time that
272  * the task is actively working on behalf of the vcpu, and time
273  * that the task is preempted, so we don't count any of it as
274  * stolen.
275  *
276  * Updates to busy_stolen are protected by arch.tbacct_lock;
277  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
278  * lock.  The stolen times are measured in units of timebase ticks.
279  * (Note that the != TB_NIL checks below are purely defensive;
280  * they should never fail.)
281  */
282 
283 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
284 {
285 	unsigned long flags;
286 
287 	spin_lock_irqsave(&vc->stoltb_lock, flags);
288 	vc->preempt_tb = mftb();
289 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
290 }
291 
292 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
293 {
294 	unsigned long flags;
295 
296 	spin_lock_irqsave(&vc->stoltb_lock, flags);
297 	if (vc->preempt_tb != TB_NIL) {
298 		vc->stolen_tb += mftb() - vc->preempt_tb;
299 		vc->preempt_tb = TB_NIL;
300 	}
301 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
302 }
303 
304 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
305 {
306 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
307 	unsigned long flags;
308 
309 	/*
310 	 * We can test vc->runner without taking the vcore lock,
311 	 * because only this task ever sets vc->runner to this
312 	 * vcpu, and once it is set to this vcpu, only this task
313 	 * ever sets it to NULL.
314 	 */
315 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
316 		kvmppc_core_end_stolen(vc);
317 
318 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
319 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
320 	    vcpu->arch.busy_preempt != TB_NIL) {
321 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
322 		vcpu->arch.busy_preempt = TB_NIL;
323 	}
324 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
325 }
326 
327 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
328 {
329 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
330 	unsigned long flags;
331 
332 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
333 		kvmppc_core_start_stolen(vc);
334 
335 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
336 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
337 		vcpu->arch.busy_preempt = mftb();
338 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
339 }
340 
341 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
342 {
343 	vcpu->arch.pvr = pvr;
344 }
345 
346 /* Dummy value used in computing PCR value below */
347 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
348 
349 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
350 {
351 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
352 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
353 
354 	/* We can (emulate) our own architecture version and anything older */
355 	if (cpu_has_feature(CPU_FTR_ARCH_31))
356 		host_pcr_bit = PCR_ARCH_31;
357 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
358 		host_pcr_bit = PCR_ARCH_300;
359 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
360 		host_pcr_bit = PCR_ARCH_207;
361 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
362 		host_pcr_bit = PCR_ARCH_206;
363 	else
364 		host_pcr_bit = PCR_ARCH_205;
365 
366 	/* Determine lowest PCR bit needed to run guest in given PVR level */
367 	guest_pcr_bit = host_pcr_bit;
368 	if (arch_compat) {
369 		switch (arch_compat) {
370 		case PVR_ARCH_205:
371 			guest_pcr_bit = PCR_ARCH_205;
372 			break;
373 		case PVR_ARCH_206:
374 		case PVR_ARCH_206p:
375 			guest_pcr_bit = PCR_ARCH_206;
376 			break;
377 		case PVR_ARCH_207:
378 			guest_pcr_bit = PCR_ARCH_207;
379 			break;
380 		case PVR_ARCH_300:
381 			guest_pcr_bit = PCR_ARCH_300;
382 			break;
383 		case PVR_ARCH_31:
384 			guest_pcr_bit = PCR_ARCH_31;
385 			break;
386 		default:
387 			return -EINVAL;
388 		}
389 	}
390 
391 	/* Check requested PCR bits don't exceed our capabilities */
392 	if (guest_pcr_bit > host_pcr_bit)
393 		return -EINVAL;
394 
395 	spin_lock(&vc->lock);
396 	vc->arch_compat = arch_compat;
397 	/*
398 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
399 	 * Also set all reserved PCR bits
400 	 */
401 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
402 	spin_unlock(&vc->lock);
403 
404 	return 0;
405 }
406 
407 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
408 {
409 	int r;
410 
411 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
412 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
413 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
414 	for (r = 0; r < 16; ++r)
415 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
416 		       r, kvmppc_get_gpr(vcpu, r),
417 		       r+16, kvmppc_get_gpr(vcpu, r+16));
418 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
419 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
420 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
421 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
422 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
423 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
424 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
425 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
426 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
427 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
428 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
429 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
430 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
431 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
432 	for (r = 0; r < vcpu->arch.slb_max; ++r)
433 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
434 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
435 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
436 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
437 	       vcpu->arch.last_inst);
438 }
439 
440 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
441 {
442 	return kvm_get_vcpu_by_id(kvm, id);
443 }
444 
445 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
446 {
447 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
448 	vpa->yield_count = cpu_to_be32(1);
449 }
450 
451 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
452 		   unsigned long addr, unsigned long len)
453 {
454 	/* check address is cacheline aligned */
455 	if (addr & (L1_CACHE_BYTES - 1))
456 		return -EINVAL;
457 	spin_lock(&vcpu->arch.vpa_update_lock);
458 	if (v->next_gpa != addr || v->len != len) {
459 		v->next_gpa = addr;
460 		v->len = addr ? len : 0;
461 		v->update_pending = 1;
462 	}
463 	spin_unlock(&vcpu->arch.vpa_update_lock);
464 	return 0;
465 }
466 
467 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
468 struct reg_vpa {
469 	u32 dummy;
470 	union {
471 		__be16 hword;
472 		__be32 word;
473 	} length;
474 };
475 
476 static int vpa_is_registered(struct kvmppc_vpa *vpap)
477 {
478 	if (vpap->update_pending)
479 		return vpap->next_gpa != 0;
480 	return vpap->pinned_addr != NULL;
481 }
482 
483 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
484 				       unsigned long flags,
485 				       unsigned long vcpuid, unsigned long vpa)
486 {
487 	struct kvm *kvm = vcpu->kvm;
488 	unsigned long len, nb;
489 	void *va;
490 	struct kvm_vcpu *tvcpu;
491 	int err;
492 	int subfunc;
493 	struct kvmppc_vpa *vpap;
494 
495 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
496 	if (!tvcpu)
497 		return H_PARAMETER;
498 
499 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
500 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
501 	    subfunc == H_VPA_REG_SLB) {
502 		/* Registering new area - address must be cache-line aligned */
503 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
504 			return H_PARAMETER;
505 
506 		/* convert logical addr to kernel addr and read length */
507 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
508 		if (va == NULL)
509 			return H_PARAMETER;
510 		if (subfunc == H_VPA_REG_VPA)
511 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
512 		else
513 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
514 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
515 
516 		/* Check length */
517 		if (len > nb || len < sizeof(struct reg_vpa))
518 			return H_PARAMETER;
519 	} else {
520 		vpa = 0;
521 		len = 0;
522 	}
523 
524 	err = H_PARAMETER;
525 	vpap = NULL;
526 	spin_lock(&tvcpu->arch.vpa_update_lock);
527 
528 	switch (subfunc) {
529 	case H_VPA_REG_VPA:		/* register VPA */
530 		/*
531 		 * The size of our lppaca is 1kB because of the way we align
532 		 * it for the guest to avoid crossing a 4kB boundary. We only
533 		 * use 640 bytes of the structure though, so we should accept
534 		 * clients that set a size of 640.
535 		 */
536 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
537 		if (len < sizeof(struct lppaca))
538 			break;
539 		vpap = &tvcpu->arch.vpa;
540 		err = 0;
541 		break;
542 
543 	case H_VPA_REG_DTL:		/* register DTL */
544 		if (len < sizeof(struct dtl_entry))
545 			break;
546 		len -= len % sizeof(struct dtl_entry);
547 
548 		/* Check that they have previously registered a VPA */
549 		err = H_RESOURCE;
550 		if (!vpa_is_registered(&tvcpu->arch.vpa))
551 			break;
552 
553 		vpap = &tvcpu->arch.dtl;
554 		err = 0;
555 		break;
556 
557 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
558 		/* Check that they have previously registered a VPA */
559 		err = H_RESOURCE;
560 		if (!vpa_is_registered(&tvcpu->arch.vpa))
561 			break;
562 
563 		vpap = &tvcpu->arch.slb_shadow;
564 		err = 0;
565 		break;
566 
567 	case H_VPA_DEREG_VPA:		/* deregister VPA */
568 		/* Check they don't still have a DTL or SLB buf registered */
569 		err = H_RESOURCE;
570 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
571 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
572 			break;
573 
574 		vpap = &tvcpu->arch.vpa;
575 		err = 0;
576 		break;
577 
578 	case H_VPA_DEREG_DTL:		/* deregister DTL */
579 		vpap = &tvcpu->arch.dtl;
580 		err = 0;
581 		break;
582 
583 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
584 		vpap = &tvcpu->arch.slb_shadow;
585 		err = 0;
586 		break;
587 	}
588 
589 	if (vpap) {
590 		vpap->next_gpa = vpa;
591 		vpap->len = len;
592 		vpap->update_pending = 1;
593 	}
594 
595 	spin_unlock(&tvcpu->arch.vpa_update_lock);
596 
597 	return err;
598 }
599 
600 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
601 {
602 	struct kvm *kvm = vcpu->kvm;
603 	void *va;
604 	unsigned long nb;
605 	unsigned long gpa;
606 
607 	/*
608 	 * We need to pin the page pointed to by vpap->next_gpa,
609 	 * but we can't call kvmppc_pin_guest_page under the lock
610 	 * as it does get_user_pages() and down_read().  So we
611 	 * have to drop the lock, pin the page, then get the lock
612 	 * again and check that a new area didn't get registered
613 	 * in the meantime.
614 	 */
615 	for (;;) {
616 		gpa = vpap->next_gpa;
617 		spin_unlock(&vcpu->arch.vpa_update_lock);
618 		va = NULL;
619 		nb = 0;
620 		if (gpa)
621 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
622 		spin_lock(&vcpu->arch.vpa_update_lock);
623 		if (gpa == vpap->next_gpa)
624 			break;
625 		/* sigh... unpin that one and try again */
626 		if (va)
627 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
628 	}
629 
630 	vpap->update_pending = 0;
631 	if (va && nb < vpap->len) {
632 		/*
633 		 * If it's now too short, it must be that userspace
634 		 * has changed the mappings underlying guest memory,
635 		 * so unregister the region.
636 		 */
637 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
638 		va = NULL;
639 	}
640 	if (vpap->pinned_addr)
641 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
642 					vpap->dirty);
643 	vpap->gpa = gpa;
644 	vpap->pinned_addr = va;
645 	vpap->dirty = false;
646 	if (va)
647 		vpap->pinned_end = va + vpap->len;
648 }
649 
650 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
651 {
652 	if (!(vcpu->arch.vpa.update_pending ||
653 	      vcpu->arch.slb_shadow.update_pending ||
654 	      vcpu->arch.dtl.update_pending))
655 		return;
656 
657 	spin_lock(&vcpu->arch.vpa_update_lock);
658 	if (vcpu->arch.vpa.update_pending) {
659 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
660 		if (vcpu->arch.vpa.pinned_addr)
661 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
662 	}
663 	if (vcpu->arch.dtl.update_pending) {
664 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
665 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
666 		vcpu->arch.dtl_index = 0;
667 	}
668 	if (vcpu->arch.slb_shadow.update_pending)
669 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
670 	spin_unlock(&vcpu->arch.vpa_update_lock);
671 }
672 
673 /*
674  * Return the accumulated stolen time for the vcore up until `now'.
675  * The caller should hold the vcore lock.
676  */
677 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
678 {
679 	u64 p;
680 	unsigned long flags;
681 
682 	spin_lock_irqsave(&vc->stoltb_lock, flags);
683 	p = vc->stolen_tb;
684 	if (vc->vcore_state != VCORE_INACTIVE &&
685 	    vc->preempt_tb != TB_NIL)
686 		p += now - vc->preempt_tb;
687 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
688 	return p;
689 }
690 
691 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
692 				    struct kvmppc_vcore *vc)
693 {
694 	struct dtl_entry *dt;
695 	struct lppaca *vpa;
696 	unsigned long stolen;
697 	unsigned long core_stolen;
698 	u64 now;
699 	unsigned long flags;
700 
701 	dt = vcpu->arch.dtl_ptr;
702 	vpa = vcpu->arch.vpa.pinned_addr;
703 	now = mftb();
704 	core_stolen = vcore_stolen_time(vc, now);
705 	stolen = core_stolen - vcpu->arch.stolen_logged;
706 	vcpu->arch.stolen_logged = core_stolen;
707 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
708 	stolen += vcpu->arch.busy_stolen;
709 	vcpu->arch.busy_stolen = 0;
710 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
711 	if (!dt || !vpa)
712 		return;
713 	memset(dt, 0, sizeof(struct dtl_entry));
714 	dt->dispatch_reason = 7;
715 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
716 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
717 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
718 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
719 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
720 	++dt;
721 	if (dt == vcpu->arch.dtl.pinned_end)
722 		dt = vcpu->arch.dtl.pinned_addr;
723 	vcpu->arch.dtl_ptr = dt;
724 	/* order writing *dt vs. writing vpa->dtl_idx */
725 	smp_wmb();
726 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
727 	vcpu->arch.dtl.dirty = true;
728 }
729 
730 /* See if there is a doorbell interrupt pending for a vcpu */
731 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
732 {
733 	int thr;
734 	struct kvmppc_vcore *vc;
735 
736 	if (vcpu->arch.doorbell_request)
737 		return true;
738 	/*
739 	 * Ensure that the read of vcore->dpdes comes after the read
740 	 * of vcpu->doorbell_request.  This barrier matches the
741 	 * smp_wmb() in kvmppc_guest_entry_inject().
742 	 */
743 	smp_rmb();
744 	vc = vcpu->arch.vcore;
745 	thr = vcpu->vcpu_id - vc->first_vcpuid;
746 	return !!(vc->dpdes & (1 << thr));
747 }
748 
749 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
750 {
751 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
752 		return true;
753 	if ((!vcpu->arch.vcore->arch_compat) &&
754 	    cpu_has_feature(CPU_FTR_ARCH_207S))
755 		return true;
756 	return false;
757 }
758 
759 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
760 			     unsigned long resource, unsigned long value1,
761 			     unsigned long value2)
762 {
763 	switch (resource) {
764 	case H_SET_MODE_RESOURCE_SET_CIABR:
765 		if (!kvmppc_power8_compatible(vcpu))
766 			return H_P2;
767 		if (value2)
768 			return H_P4;
769 		if (mflags)
770 			return H_UNSUPPORTED_FLAG_START;
771 		/* Guests can't breakpoint the hypervisor */
772 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
773 			return H_P3;
774 		vcpu->arch.ciabr  = value1;
775 		return H_SUCCESS;
776 	case H_SET_MODE_RESOURCE_SET_DAWR0:
777 		if (!kvmppc_power8_compatible(vcpu))
778 			return H_P2;
779 		if (!ppc_breakpoint_available())
780 			return H_P2;
781 		if (mflags)
782 			return H_UNSUPPORTED_FLAG_START;
783 		if (value2 & DABRX_HYP)
784 			return H_P4;
785 		vcpu->arch.dawr  = value1;
786 		vcpu->arch.dawrx = value2;
787 		return H_SUCCESS;
788 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
789 		/* KVM does not support mflags=2 (AIL=2) */
790 		if (mflags != 0 && mflags != 3)
791 			return H_UNSUPPORTED_FLAG_START;
792 		return H_TOO_HARD;
793 	default:
794 		return H_TOO_HARD;
795 	}
796 }
797 
798 /* Copy guest memory in place - must reside within a single memslot */
799 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
800 				  unsigned long len)
801 {
802 	struct kvm_memory_slot *to_memslot = NULL;
803 	struct kvm_memory_slot *from_memslot = NULL;
804 	unsigned long to_addr, from_addr;
805 	int r;
806 
807 	/* Get HPA for from address */
808 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
809 	if (!from_memslot)
810 		return -EFAULT;
811 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
812 			     << PAGE_SHIFT))
813 		return -EINVAL;
814 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
815 	if (kvm_is_error_hva(from_addr))
816 		return -EFAULT;
817 	from_addr |= (from & (PAGE_SIZE - 1));
818 
819 	/* Get HPA for to address */
820 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
821 	if (!to_memslot)
822 		return -EFAULT;
823 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
824 			   << PAGE_SHIFT))
825 		return -EINVAL;
826 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
827 	if (kvm_is_error_hva(to_addr))
828 		return -EFAULT;
829 	to_addr |= (to & (PAGE_SIZE - 1));
830 
831 	/* Perform copy */
832 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
833 			     len);
834 	if (r)
835 		return -EFAULT;
836 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
837 	return 0;
838 }
839 
840 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
841 			       unsigned long dest, unsigned long src)
842 {
843 	u64 pg_sz = SZ_4K;		/* 4K page size */
844 	u64 pg_mask = SZ_4K - 1;
845 	int ret;
846 
847 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
848 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
849 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
850 		return H_PARAMETER;
851 
852 	/* dest (and src if copy_page flag set) must be page aligned */
853 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
854 		return H_PARAMETER;
855 
856 	/* zero and/or copy the page as determined by the flags */
857 	if (flags & H_COPY_PAGE) {
858 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
859 		if (ret < 0)
860 			return H_PARAMETER;
861 	} else if (flags & H_ZERO_PAGE) {
862 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
863 		if (ret < 0)
864 			return H_PARAMETER;
865 	}
866 
867 	/* We can ignore the remaining flags */
868 
869 	return H_SUCCESS;
870 }
871 
872 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
873 {
874 	struct kvmppc_vcore *vcore = target->arch.vcore;
875 
876 	/*
877 	 * We expect to have been called by the real mode handler
878 	 * (kvmppc_rm_h_confer()) which would have directly returned
879 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
880 	 * have useful work to do and should not confer) so we don't
881 	 * recheck that here.
882 	 */
883 
884 	spin_lock(&vcore->lock);
885 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
886 	    vcore->vcore_state != VCORE_INACTIVE &&
887 	    vcore->runner)
888 		target = vcore->runner;
889 	spin_unlock(&vcore->lock);
890 
891 	return kvm_vcpu_yield_to(target);
892 }
893 
894 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
895 {
896 	int yield_count = 0;
897 	struct lppaca *lppaca;
898 
899 	spin_lock(&vcpu->arch.vpa_update_lock);
900 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
901 	if (lppaca)
902 		yield_count = be32_to_cpu(lppaca->yield_count);
903 	spin_unlock(&vcpu->arch.vpa_update_lock);
904 	return yield_count;
905 }
906 
907 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
908 {
909 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
910 	unsigned long target, ret = H_SUCCESS;
911 	int yield_count;
912 	struct kvm_vcpu *tvcpu;
913 	int idx, rc;
914 
915 	if (req <= MAX_HCALL_OPCODE &&
916 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
917 		return RESUME_HOST;
918 
919 	switch (req) {
920 	case H_CEDE:
921 		break;
922 	case H_PROD:
923 		target = kvmppc_get_gpr(vcpu, 4);
924 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
925 		if (!tvcpu) {
926 			ret = H_PARAMETER;
927 			break;
928 		}
929 		tvcpu->arch.prodded = 1;
930 		smp_mb();
931 		if (tvcpu->arch.ceded)
932 			kvmppc_fast_vcpu_kick_hv(tvcpu);
933 		break;
934 	case H_CONFER:
935 		target = kvmppc_get_gpr(vcpu, 4);
936 		if (target == -1)
937 			break;
938 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
939 		if (!tvcpu) {
940 			ret = H_PARAMETER;
941 			break;
942 		}
943 		yield_count = kvmppc_get_gpr(vcpu, 5);
944 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
945 			break;
946 		kvm_arch_vcpu_yield_to(tvcpu);
947 		break;
948 	case H_REGISTER_VPA:
949 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
950 					kvmppc_get_gpr(vcpu, 5),
951 					kvmppc_get_gpr(vcpu, 6));
952 		break;
953 	case H_RTAS:
954 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
955 			return RESUME_HOST;
956 
957 		idx = srcu_read_lock(&vcpu->kvm->srcu);
958 		rc = kvmppc_rtas_hcall(vcpu);
959 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
960 
961 		if (rc == -ENOENT)
962 			return RESUME_HOST;
963 		else if (rc == 0)
964 			break;
965 
966 		/* Send the error out to userspace via KVM_RUN */
967 		return rc;
968 	case H_LOGICAL_CI_LOAD:
969 		ret = kvmppc_h_logical_ci_load(vcpu);
970 		if (ret == H_TOO_HARD)
971 			return RESUME_HOST;
972 		break;
973 	case H_LOGICAL_CI_STORE:
974 		ret = kvmppc_h_logical_ci_store(vcpu);
975 		if (ret == H_TOO_HARD)
976 			return RESUME_HOST;
977 		break;
978 	case H_SET_MODE:
979 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
980 					kvmppc_get_gpr(vcpu, 5),
981 					kvmppc_get_gpr(vcpu, 6),
982 					kvmppc_get_gpr(vcpu, 7));
983 		if (ret == H_TOO_HARD)
984 			return RESUME_HOST;
985 		break;
986 	case H_XIRR:
987 	case H_CPPR:
988 	case H_EOI:
989 	case H_IPI:
990 	case H_IPOLL:
991 	case H_XIRR_X:
992 		if (kvmppc_xics_enabled(vcpu)) {
993 			if (xics_on_xive()) {
994 				ret = H_NOT_AVAILABLE;
995 				return RESUME_GUEST;
996 			}
997 			ret = kvmppc_xics_hcall(vcpu, req);
998 			break;
999 		}
1000 		return RESUME_HOST;
1001 	case H_SET_DABR:
1002 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1003 		break;
1004 	case H_SET_XDABR:
1005 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1006 						kvmppc_get_gpr(vcpu, 5));
1007 		break;
1008 #ifdef CONFIG_SPAPR_TCE_IOMMU
1009 	case H_GET_TCE:
1010 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1011 						kvmppc_get_gpr(vcpu, 5));
1012 		if (ret == H_TOO_HARD)
1013 			return RESUME_HOST;
1014 		break;
1015 	case H_PUT_TCE:
1016 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1017 						kvmppc_get_gpr(vcpu, 5),
1018 						kvmppc_get_gpr(vcpu, 6));
1019 		if (ret == H_TOO_HARD)
1020 			return RESUME_HOST;
1021 		break;
1022 	case H_PUT_TCE_INDIRECT:
1023 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1024 						kvmppc_get_gpr(vcpu, 5),
1025 						kvmppc_get_gpr(vcpu, 6),
1026 						kvmppc_get_gpr(vcpu, 7));
1027 		if (ret == H_TOO_HARD)
1028 			return RESUME_HOST;
1029 		break;
1030 	case H_STUFF_TCE:
1031 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1032 						kvmppc_get_gpr(vcpu, 5),
1033 						kvmppc_get_gpr(vcpu, 6),
1034 						kvmppc_get_gpr(vcpu, 7));
1035 		if (ret == H_TOO_HARD)
1036 			return RESUME_HOST;
1037 		break;
1038 #endif
1039 	case H_RANDOM:
1040 		if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1041 			ret = H_HARDWARE;
1042 		break;
1043 
1044 	case H_SET_PARTITION_TABLE:
1045 		ret = H_FUNCTION;
1046 		if (nesting_enabled(vcpu->kvm))
1047 			ret = kvmhv_set_partition_table(vcpu);
1048 		break;
1049 	case H_ENTER_NESTED:
1050 		ret = H_FUNCTION;
1051 		if (!nesting_enabled(vcpu->kvm))
1052 			break;
1053 		ret = kvmhv_enter_nested_guest(vcpu);
1054 		if (ret == H_INTERRUPT) {
1055 			kvmppc_set_gpr(vcpu, 3, 0);
1056 			vcpu->arch.hcall_needed = 0;
1057 			return -EINTR;
1058 		} else if (ret == H_TOO_HARD) {
1059 			kvmppc_set_gpr(vcpu, 3, 0);
1060 			vcpu->arch.hcall_needed = 0;
1061 			return RESUME_HOST;
1062 		}
1063 		break;
1064 	case H_TLB_INVALIDATE:
1065 		ret = H_FUNCTION;
1066 		if (nesting_enabled(vcpu->kvm))
1067 			ret = kvmhv_do_nested_tlbie(vcpu);
1068 		break;
1069 	case H_COPY_TOFROM_GUEST:
1070 		ret = H_FUNCTION;
1071 		if (nesting_enabled(vcpu->kvm))
1072 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1073 		break;
1074 	case H_PAGE_INIT:
1075 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1076 					 kvmppc_get_gpr(vcpu, 5),
1077 					 kvmppc_get_gpr(vcpu, 6));
1078 		break;
1079 	case H_SVM_PAGE_IN:
1080 		ret = H_UNSUPPORTED;
1081 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1082 			ret = kvmppc_h_svm_page_in(vcpu->kvm,
1083 						   kvmppc_get_gpr(vcpu, 4),
1084 						   kvmppc_get_gpr(vcpu, 5),
1085 						   kvmppc_get_gpr(vcpu, 6));
1086 		break;
1087 	case H_SVM_PAGE_OUT:
1088 		ret = H_UNSUPPORTED;
1089 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1090 			ret = kvmppc_h_svm_page_out(vcpu->kvm,
1091 						    kvmppc_get_gpr(vcpu, 4),
1092 						    kvmppc_get_gpr(vcpu, 5),
1093 						    kvmppc_get_gpr(vcpu, 6));
1094 		break;
1095 	case H_SVM_INIT_START:
1096 		ret = H_UNSUPPORTED;
1097 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1098 			ret = kvmppc_h_svm_init_start(vcpu->kvm);
1099 		break;
1100 	case H_SVM_INIT_DONE:
1101 		ret = H_UNSUPPORTED;
1102 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1103 			ret = kvmppc_h_svm_init_done(vcpu->kvm);
1104 		break;
1105 	case H_SVM_INIT_ABORT:
1106 		/*
1107 		 * Even if that call is made by the Ultravisor, the SSR1 value
1108 		 * is the guest context one, with the secure bit clear as it has
1109 		 * not yet been secured. So we can't check it here.
1110 		 * Instead the kvm->arch.secure_guest flag is checked inside
1111 		 * kvmppc_h_svm_init_abort().
1112 		 */
1113 		ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1114 		break;
1115 
1116 	default:
1117 		return RESUME_HOST;
1118 	}
1119 	kvmppc_set_gpr(vcpu, 3, ret);
1120 	vcpu->arch.hcall_needed = 0;
1121 	return RESUME_GUEST;
1122 }
1123 
1124 /*
1125  * Handle H_CEDE in the nested virtualization case where we haven't
1126  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1127  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1128  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1129  */
1130 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1131 {
1132 	vcpu->arch.shregs.msr |= MSR_EE;
1133 	vcpu->arch.ceded = 1;
1134 	smp_mb();
1135 	if (vcpu->arch.prodded) {
1136 		vcpu->arch.prodded = 0;
1137 		smp_mb();
1138 		vcpu->arch.ceded = 0;
1139 	}
1140 }
1141 
1142 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1143 {
1144 	switch (cmd) {
1145 	case H_CEDE:
1146 	case H_PROD:
1147 	case H_CONFER:
1148 	case H_REGISTER_VPA:
1149 	case H_SET_MODE:
1150 	case H_LOGICAL_CI_LOAD:
1151 	case H_LOGICAL_CI_STORE:
1152 #ifdef CONFIG_KVM_XICS
1153 	case H_XIRR:
1154 	case H_CPPR:
1155 	case H_EOI:
1156 	case H_IPI:
1157 	case H_IPOLL:
1158 	case H_XIRR_X:
1159 #endif
1160 	case H_PAGE_INIT:
1161 		return 1;
1162 	}
1163 
1164 	/* See if it's in the real-mode table */
1165 	return kvmppc_hcall_impl_hv_realmode(cmd);
1166 }
1167 
1168 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1169 {
1170 	u32 last_inst;
1171 
1172 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1173 					EMULATE_DONE) {
1174 		/*
1175 		 * Fetch failed, so return to guest and
1176 		 * try executing it again.
1177 		 */
1178 		return RESUME_GUEST;
1179 	}
1180 
1181 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1182 		vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1183 		vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1184 		return RESUME_HOST;
1185 	} else {
1186 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1187 		return RESUME_GUEST;
1188 	}
1189 }
1190 
1191 static void do_nothing(void *x)
1192 {
1193 }
1194 
1195 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1196 {
1197 	int thr, cpu, pcpu, nthreads;
1198 	struct kvm_vcpu *v;
1199 	unsigned long dpdes;
1200 
1201 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1202 	dpdes = 0;
1203 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1204 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1205 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1206 		if (!v)
1207 			continue;
1208 		/*
1209 		 * If the vcpu is currently running on a physical cpu thread,
1210 		 * interrupt it in order to pull it out of the guest briefly,
1211 		 * which will update its vcore->dpdes value.
1212 		 */
1213 		pcpu = READ_ONCE(v->cpu);
1214 		if (pcpu >= 0)
1215 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1216 		if (kvmppc_doorbell_pending(v))
1217 			dpdes |= 1 << thr;
1218 	}
1219 	return dpdes;
1220 }
1221 
1222 /*
1223  * On POWER9, emulate doorbell-related instructions in order to
1224  * give the guest the illusion of running on a multi-threaded core.
1225  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1226  * and mfspr DPDES.
1227  */
1228 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1229 {
1230 	u32 inst, rb, thr;
1231 	unsigned long arg;
1232 	struct kvm *kvm = vcpu->kvm;
1233 	struct kvm_vcpu *tvcpu;
1234 
1235 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1236 		return RESUME_GUEST;
1237 	if (get_op(inst) != 31)
1238 		return EMULATE_FAIL;
1239 	rb = get_rb(inst);
1240 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1241 	switch (get_xop(inst)) {
1242 	case OP_31_XOP_MSGSNDP:
1243 		arg = kvmppc_get_gpr(vcpu, rb);
1244 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1245 			break;
1246 		arg &= 0x7f;
1247 		if (arg >= kvm->arch.emul_smt_mode)
1248 			break;
1249 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1250 		if (!tvcpu)
1251 			break;
1252 		if (!tvcpu->arch.doorbell_request) {
1253 			tvcpu->arch.doorbell_request = 1;
1254 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1255 		}
1256 		break;
1257 	case OP_31_XOP_MSGCLRP:
1258 		arg = kvmppc_get_gpr(vcpu, rb);
1259 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1260 			break;
1261 		vcpu->arch.vcore->dpdes = 0;
1262 		vcpu->arch.doorbell_request = 0;
1263 		break;
1264 	case OP_31_XOP_MFSPR:
1265 		switch (get_sprn(inst)) {
1266 		case SPRN_TIR:
1267 			arg = thr;
1268 			break;
1269 		case SPRN_DPDES:
1270 			arg = kvmppc_read_dpdes(vcpu);
1271 			break;
1272 		default:
1273 			return EMULATE_FAIL;
1274 		}
1275 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1276 		break;
1277 	default:
1278 		return EMULATE_FAIL;
1279 	}
1280 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1281 	return RESUME_GUEST;
1282 }
1283 
1284 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1285 				 struct task_struct *tsk)
1286 {
1287 	struct kvm_run *run = vcpu->run;
1288 	int r = RESUME_HOST;
1289 
1290 	vcpu->stat.sum_exits++;
1291 
1292 	/*
1293 	 * This can happen if an interrupt occurs in the last stages
1294 	 * of guest entry or the first stages of guest exit (i.e. after
1295 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1296 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1297 	 * That can happen due to a bug, or due to a machine check
1298 	 * occurring at just the wrong time.
1299 	 */
1300 	if (vcpu->arch.shregs.msr & MSR_HV) {
1301 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1302 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1303 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1304 			vcpu->arch.shregs.msr);
1305 		kvmppc_dump_regs(vcpu);
1306 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1307 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1308 		return RESUME_HOST;
1309 	}
1310 	run->exit_reason = KVM_EXIT_UNKNOWN;
1311 	run->ready_for_interrupt_injection = 1;
1312 	switch (vcpu->arch.trap) {
1313 	/* We're good on these - the host merely wanted to get our attention */
1314 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1315 		vcpu->stat.dec_exits++;
1316 		r = RESUME_GUEST;
1317 		break;
1318 	case BOOK3S_INTERRUPT_EXTERNAL:
1319 	case BOOK3S_INTERRUPT_H_DOORBELL:
1320 	case BOOK3S_INTERRUPT_H_VIRT:
1321 		vcpu->stat.ext_intr_exits++;
1322 		r = RESUME_GUEST;
1323 		break;
1324 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1325 	case BOOK3S_INTERRUPT_HMI:
1326 	case BOOK3S_INTERRUPT_PERFMON:
1327 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1328 		r = RESUME_GUEST;
1329 		break;
1330 	case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1331 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1332 					      DEFAULT_RATELIMIT_BURST);
1333 		/*
1334 		 * Print the MCE event to host console. Ratelimit so the guest
1335 		 * can't flood the host log.
1336 		 */
1337 		if (__ratelimit(&rs))
1338 			machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1339 
1340 		/*
1341 		 * If the guest can do FWNMI, exit to userspace so it can
1342 		 * deliver a FWNMI to the guest.
1343 		 * Otherwise we synthesize a machine check for the guest
1344 		 * so that it knows that the machine check occurred.
1345 		 */
1346 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1347 			ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1348 			kvmppc_core_queue_machine_check(vcpu, flags);
1349 			r = RESUME_GUEST;
1350 			break;
1351 		}
1352 
1353 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1354 		run->exit_reason = KVM_EXIT_NMI;
1355 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1356 		/* Clear out the old NMI status from run->flags */
1357 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1358 		/* Now set the NMI status */
1359 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1360 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1361 		else
1362 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1363 
1364 		r = RESUME_HOST;
1365 		break;
1366 	}
1367 	case BOOK3S_INTERRUPT_PROGRAM:
1368 	{
1369 		ulong flags;
1370 		/*
1371 		 * Normally program interrupts are delivered directly
1372 		 * to the guest by the hardware, but we can get here
1373 		 * as a result of a hypervisor emulation interrupt
1374 		 * (e40) getting turned into a 700 by BML RTAS.
1375 		 */
1376 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1377 		kvmppc_core_queue_program(vcpu, flags);
1378 		r = RESUME_GUEST;
1379 		break;
1380 	}
1381 	case BOOK3S_INTERRUPT_SYSCALL:
1382 	{
1383 		/* hcall - punt to userspace */
1384 		int i;
1385 
1386 		/* hypercall with MSR_PR has already been handled in rmode,
1387 		 * and never reaches here.
1388 		 */
1389 
1390 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1391 		for (i = 0; i < 9; ++i)
1392 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1393 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1394 		vcpu->arch.hcall_needed = 1;
1395 		r = RESUME_HOST;
1396 		break;
1397 	}
1398 	/*
1399 	 * We get these next two if the guest accesses a page which it thinks
1400 	 * it has mapped but which is not actually present, either because
1401 	 * it is for an emulated I/O device or because the corresonding
1402 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1403 	 * have been handled already.
1404 	 */
1405 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1406 		r = RESUME_PAGE_FAULT;
1407 		break;
1408 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1409 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1410 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1411 			DSISR_SRR1_MATCH_64S;
1412 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1413 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1414 		r = RESUME_PAGE_FAULT;
1415 		break;
1416 	/*
1417 	 * This occurs if the guest executes an illegal instruction.
1418 	 * If the guest debug is disabled, generate a program interrupt
1419 	 * to the guest. If guest debug is enabled, we need to check
1420 	 * whether the instruction is a software breakpoint instruction.
1421 	 * Accordingly return to Guest or Host.
1422 	 */
1423 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1424 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1425 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1426 				swab32(vcpu->arch.emul_inst) :
1427 				vcpu->arch.emul_inst;
1428 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1429 			r = kvmppc_emulate_debug_inst(vcpu);
1430 		} else {
1431 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1432 			r = RESUME_GUEST;
1433 		}
1434 		break;
1435 	/*
1436 	 * This occurs if the guest (kernel or userspace), does something that
1437 	 * is prohibited by HFSCR.
1438 	 * On POWER9, this could be a doorbell instruction that we need
1439 	 * to emulate.
1440 	 * Otherwise, we just generate a program interrupt to the guest.
1441 	 */
1442 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1443 		r = EMULATE_FAIL;
1444 		if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1445 		    cpu_has_feature(CPU_FTR_ARCH_300))
1446 			r = kvmppc_emulate_doorbell_instr(vcpu);
1447 		if (r == EMULATE_FAIL) {
1448 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1449 			r = RESUME_GUEST;
1450 		}
1451 		break;
1452 
1453 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1454 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1455 		/*
1456 		 * This occurs for various TM-related instructions that
1457 		 * we need to emulate on POWER9 DD2.2.  We have already
1458 		 * handled the cases where the guest was in real-suspend
1459 		 * mode and was transitioning to transactional state.
1460 		 */
1461 		r = kvmhv_p9_tm_emulation(vcpu);
1462 		break;
1463 #endif
1464 
1465 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1466 		r = RESUME_PASSTHROUGH;
1467 		break;
1468 	default:
1469 		kvmppc_dump_regs(vcpu);
1470 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1471 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1472 			vcpu->arch.shregs.msr);
1473 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1474 		r = RESUME_HOST;
1475 		break;
1476 	}
1477 
1478 	return r;
1479 }
1480 
1481 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1482 {
1483 	int r;
1484 	int srcu_idx;
1485 
1486 	vcpu->stat.sum_exits++;
1487 
1488 	/*
1489 	 * This can happen if an interrupt occurs in the last stages
1490 	 * of guest entry or the first stages of guest exit (i.e. after
1491 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1492 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1493 	 * That can happen due to a bug, or due to a machine check
1494 	 * occurring at just the wrong time.
1495 	 */
1496 	if (vcpu->arch.shregs.msr & MSR_HV) {
1497 		pr_emerg("KVM trap in HV mode while nested!\n");
1498 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1499 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1500 			 vcpu->arch.shregs.msr);
1501 		kvmppc_dump_regs(vcpu);
1502 		return RESUME_HOST;
1503 	}
1504 	switch (vcpu->arch.trap) {
1505 	/* We're good on these - the host merely wanted to get our attention */
1506 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1507 		vcpu->stat.dec_exits++;
1508 		r = RESUME_GUEST;
1509 		break;
1510 	case BOOK3S_INTERRUPT_EXTERNAL:
1511 		vcpu->stat.ext_intr_exits++;
1512 		r = RESUME_HOST;
1513 		break;
1514 	case BOOK3S_INTERRUPT_H_DOORBELL:
1515 	case BOOK3S_INTERRUPT_H_VIRT:
1516 		vcpu->stat.ext_intr_exits++;
1517 		r = RESUME_GUEST;
1518 		break;
1519 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1520 	case BOOK3S_INTERRUPT_HMI:
1521 	case BOOK3S_INTERRUPT_PERFMON:
1522 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1523 		r = RESUME_GUEST;
1524 		break;
1525 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1526 	{
1527 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1528 					      DEFAULT_RATELIMIT_BURST);
1529 		/* Pass the machine check to the L1 guest */
1530 		r = RESUME_HOST;
1531 		/* Print the MCE event to host console. */
1532 		if (__ratelimit(&rs))
1533 			machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1534 		break;
1535 	}
1536 	/*
1537 	 * We get these next two if the guest accesses a page which it thinks
1538 	 * it has mapped but which is not actually present, either because
1539 	 * it is for an emulated I/O device or because the corresonding
1540 	 * host page has been paged out.
1541 	 */
1542 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1543 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1544 		r = kvmhv_nested_page_fault(vcpu);
1545 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1546 		break;
1547 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1548 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1549 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1550 					 DSISR_SRR1_MATCH_64S;
1551 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1552 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1553 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1554 		r = kvmhv_nested_page_fault(vcpu);
1555 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1556 		break;
1557 
1558 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1559 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1560 		/*
1561 		 * This occurs for various TM-related instructions that
1562 		 * we need to emulate on POWER9 DD2.2.  We have already
1563 		 * handled the cases where the guest was in real-suspend
1564 		 * mode and was transitioning to transactional state.
1565 		 */
1566 		r = kvmhv_p9_tm_emulation(vcpu);
1567 		break;
1568 #endif
1569 
1570 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1571 		vcpu->arch.trap = 0;
1572 		r = RESUME_GUEST;
1573 		if (!xics_on_xive())
1574 			kvmppc_xics_rm_complete(vcpu, 0);
1575 		break;
1576 	default:
1577 		r = RESUME_HOST;
1578 		break;
1579 	}
1580 
1581 	return r;
1582 }
1583 
1584 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1585 					    struct kvm_sregs *sregs)
1586 {
1587 	int i;
1588 
1589 	memset(sregs, 0, sizeof(struct kvm_sregs));
1590 	sregs->pvr = vcpu->arch.pvr;
1591 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1592 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1593 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1594 	}
1595 
1596 	return 0;
1597 }
1598 
1599 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1600 					    struct kvm_sregs *sregs)
1601 {
1602 	int i, j;
1603 
1604 	/* Only accept the same PVR as the host's, since we can't spoof it */
1605 	if (sregs->pvr != vcpu->arch.pvr)
1606 		return -EINVAL;
1607 
1608 	j = 0;
1609 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1610 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1611 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1612 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1613 			++j;
1614 		}
1615 	}
1616 	vcpu->arch.slb_max = j;
1617 
1618 	return 0;
1619 }
1620 
1621 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1622 		bool preserve_top32)
1623 {
1624 	struct kvm *kvm = vcpu->kvm;
1625 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1626 	u64 mask;
1627 
1628 	spin_lock(&vc->lock);
1629 	/*
1630 	 * If ILE (interrupt little-endian) has changed, update the
1631 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1632 	 */
1633 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1634 		struct kvm_vcpu *vcpu;
1635 		int i;
1636 
1637 		kvm_for_each_vcpu(i, vcpu, kvm) {
1638 			if (vcpu->arch.vcore != vc)
1639 				continue;
1640 			if (new_lpcr & LPCR_ILE)
1641 				vcpu->arch.intr_msr |= MSR_LE;
1642 			else
1643 				vcpu->arch.intr_msr &= ~MSR_LE;
1644 		}
1645 	}
1646 
1647 	/*
1648 	 * Userspace can only modify DPFD (default prefetch depth),
1649 	 * ILE (interrupt little-endian) and TC (translation control).
1650 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1651 	 */
1652 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1653 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1654 		mask |= LPCR_AIL;
1655 	/*
1656 	 * On POWER9, allow userspace to enable large decrementer for the
1657 	 * guest, whether or not the host has it enabled.
1658 	 */
1659 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1660 		mask |= LPCR_LD;
1661 
1662 	/* Broken 32-bit version of LPCR must not clear top bits */
1663 	if (preserve_top32)
1664 		mask &= 0xFFFFFFFF;
1665 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1666 	spin_unlock(&vc->lock);
1667 }
1668 
1669 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1670 				 union kvmppc_one_reg *val)
1671 {
1672 	int r = 0;
1673 	long int i;
1674 
1675 	switch (id) {
1676 	case KVM_REG_PPC_DEBUG_INST:
1677 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1678 		break;
1679 	case KVM_REG_PPC_HIOR:
1680 		*val = get_reg_val(id, 0);
1681 		break;
1682 	case KVM_REG_PPC_DABR:
1683 		*val = get_reg_val(id, vcpu->arch.dabr);
1684 		break;
1685 	case KVM_REG_PPC_DABRX:
1686 		*val = get_reg_val(id, vcpu->arch.dabrx);
1687 		break;
1688 	case KVM_REG_PPC_DSCR:
1689 		*val = get_reg_val(id, vcpu->arch.dscr);
1690 		break;
1691 	case KVM_REG_PPC_PURR:
1692 		*val = get_reg_val(id, vcpu->arch.purr);
1693 		break;
1694 	case KVM_REG_PPC_SPURR:
1695 		*val = get_reg_val(id, vcpu->arch.spurr);
1696 		break;
1697 	case KVM_REG_PPC_AMR:
1698 		*val = get_reg_val(id, vcpu->arch.amr);
1699 		break;
1700 	case KVM_REG_PPC_UAMOR:
1701 		*val = get_reg_val(id, vcpu->arch.uamor);
1702 		break;
1703 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1704 		i = id - KVM_REG_PPC_MMCR0;
1705 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1706 		break;
1707 	case KVM_REG_PPC_MMCR2:
1708 		*val = get_reg_val(id, vcpu->arch.mmcr[2]);
1709 		break;
1710 	case KVM_REG_PPC_MMCRA:
1711 		*val = get_reg_val(id, vcpu->arch.mmcra);
1712 		break;
1713 	case KVM_REG_PPC_MMCRS:
1714 		*val = get_reg_val(id, vcpu->arch.mmcrs);
1715 		break;
1716 	case KVM_REG_PPC_MMCR3:
1717 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
1718 		break;
1719 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1720 		i = id - KVM_REG_PPC_PMC1;
1721 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1722 		break;
1723 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1724 		i = id - KVM_REG_PPC_SPMC1;
1725 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1726 		break;
1727 	case KVM_REG_PPC_SIAR:
1728 		*val = get_reg_val(id, vcpu->arch.siar);
1729 		break;
1730 	case KVM_REG_PPC_SDAR:
1731 		*val = get_reg_val(id, vcpu->arch.sdar);
1732 		break;
1733 	case KVM_REG_PPC_SIER:
1734 		*val = get_reg_val(id, vcpu->arch.sier[0]);
1735 		break;
1736 	case KVM_REG_PPC_SIER2:
1737 		*val = get_reg_val(id, vcpu->arch.sier[1]);
1738 		break;
1739 	case KVM_REG_PPC_SIER3:
1740 		*val = get_reg_val(id, vcpu->arch.sier[2]);
1741 		break;
1742 	case KVM_REG_PPC_IAMR:
1743 		*val = get_reg_val(id, vcpu->arch.iamr);
1744 		break;
1745 	case KVM_REG_PPC_PSPB:
1746 		*val = get_reg_val(id, vcpu->arch.pspb);
1747 		break;
1748 	case KVM_REG_PPC_DPDES:
1749 		/*
1750 		 * On POWER9, where we are emulating msgsndp etc.,
1751 		 * we return 1 bit for each vcpu, which can come from
1752 		 * either vcore->dpdes or doorbell_request.
1753 		 * On POWER8, doorbell_request is 0.
1754 		 */
1755 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1756 				   vcpu->arch.doorbell_request);
1757 		break;
1758 	case KVM_REG_PPC_VTB:
1759 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1760 		break;
1761 	case KVM_REG_PPC_DAWR:
1762 		*val = get_reg_val(id, vcpu->arch.dawr);
1763 		break;
1764 	case KVM_REG_PPC_DAWRX:
1765 		*val = get_reg_val(id, vcpu->arch.dawrx);
1766 		break;
1767 	case KVM_REG_PPC_CIABR:
1768 		*val = get_reg_val(id, vcpu->arch.ciabr);
1769 		break;
1770 	case KVM_REG_PPC_CSIGR:
1771 		*val = get_reg_val(id, vcpu->arch.csigr);
1772 		break;
1773 	case KVM_REG_PPC_TACR:
1774 		*val = get_reg_val(id, vcpu->arch.tacr);
1775 		break;
1776 	case KVM_REG_PPC_TCSCR:
1777 		*val = get_reg_val(id, vcpu->arch.tcscr);
1778 		break;
1779 	case KVM_REG_PPC_PID:
1780 		*val = get_reg_val(id, vcpu->arch.pid);
1781 		break;
1782 	case KVM_REG_PPC_ACOP:
1783 		*val = get_reg_val(id, vcpu->arch.acop);
1784 		break;
1785 	case KVM_REG_PPC_WORT:
1786 		*val = get_reg_val(id, vcpu->arch.wort);
1787 		break;
1788 	case KVM_REG_PPC_TIDR:
1789 		*val = get_reg_val(id, vcpu->arch.tid);
1790 		break;
1791 	case KVM_REG_PPC_PSSCR:
1792 		*val = get_reg_val(id, vcpu->arch.psscr);
1793 		break;
1794 	case KVM_REG_PPC_VPA_ADDR:
1795 		spin_lock(&vcpu->arch.vpa_update_lock);
1796 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1797 		spin_unlock(&vcpu->arch.vpa_update_lock);
1798 		break;
1799 	case KVM_REG_PPC_VPA_SLB:
1800 		spin_lock(&vcpu->arch.vpa_update_lock);
1801 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1802 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1803 		spin_unlock(&vcpu->arch.vpa_update_lock);
1804 		break;
1805 	case KVM_REG_PPC_VPA_DTL:
1806 		spin_lock(&vcpu->arch.vpa_update_lock);
1807 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1808 		val->vpaval.length = vcpu->arch.dtl.len;
1809 		spin_unlock(&vcpu->arch.vpa_update_lock);
1810 		break;
1811 	case KVM_REG_PPC_TB_OFFSET:
1812 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1813 		break;
1814 	case KVM_REG_PPC_LPCR:
1815 	case KVM_REG_PPC_LPCR_64:
1816 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1817 		break;
1818 	case KVM_REG_PPC_PPR:
1819 		*val = get_reg_val(id, vcpu->arch.ppr);
1820 		break;
1821 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1822 	case KVM_REG_PPC_TFHAR:
1823 		*val = get_reg_val(id, vcpu->arch.tfhar);
1824 		break;
1825 	case KVM_REG_PPC_TFIAR:
1826 		*val = get_reg_val(id, vcpu->arch.tfiar);
1827 		break;
1828 	case KVM_REG_PPC_TEXASR:
1829 		*val = get_reg_val(id, vcpu->arch.texasr);
1830 		break;
1831 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1832 		i = id - KVM_REG_PPC_TM_GPR0;
1833 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1834 		break;
1835 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1836 	{
1837 		int j;
1838 		i = id - KVM_REG_PPC_TM_VSR0;
1839 		if (i < 32)
1840 			for (j = 0; j < TS_FPRWIDTH; j++)
1841 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1842 		else {
1843 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1844 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1845 			else
1846 				r = -ENXIO;
1847 		}
1848 		break;
1849 	}
1850 	case KVM_REG_PPC_TM_CR:
1851 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1852 		break;
1853 	case KVM_REG_PPC_TM_XER:
1854 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1855 		break;
1856 	case KVM_REG_PPC_TM_LR:
1857 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1858 		break;
1859 	case KVM_REG_PPC_TM_CTR:
1860 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1861 		break;
1862 	case KVM_REG_PPC_TM_FPSCR:
1863 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1864 		break;
1865 	case KVM_REG_PPC_TM_AMR:
1866 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1867 		break;
1868 	case KVM_REG_PPC_TM_PPR:
1869 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1870 		break;
1871 	case KVM_REG_PPC_TM_VRSAVE:
1872 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1873 		break;
1874 	case KVM_REG_PPC_TM_VSCR:
1875 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1876 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1877 		else
1878 			r = -ENXIO;
1879 		break;
1880 	case KVM_REG_PPC_TM_DSCR:
1881 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1882 		break;
1883 	case KVM_REG_PPC_TM_TAR:
1884 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1885 		break;
1886 #endif
1887 	case KVM_REG_PPC_ARCH_COMPAT:
1888 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1889 		break;
1890 	case KVM_REG_PPC_DEC_EXPIRY:
1891 		*val = get_reg_val(id, vcpu->arch.dec_expires +
1892 				   vcpu->arch.vcore->tb_offset);
1893 		break;
1894 	case KVM_REG_PPC_ONLINE:
1895 		*val = get_reg_val(id, vcpu->arch.online);
1896 		break;
1897 	case KVM_REG_PPC_PTCR:
1898 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1899 		break;
1900 	default:
1901 		r = -EINVAL;
1902 		break;
1903 	}
1904 
1905 	return r;
1906 }
1907 
1908 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1909 				 union kvmppc_one_reg *val)
1910 {
1911 	int r = 0;
1912 	long int i;
1913 	unsigned long addr, len;
1914 
1915 	switch (id) {
1916 	case KVM_REG_PPC_HIOR:
1917 		/* Only allow this to be set to zero */
1918 		if (set_reg_val(id, *val))
1919 			r = -EINVAL;
1920 		break;
1921 	case KVM_REG_PPC_DABR:
1922 		vcpu->arch.dabr = set_reg_val(id, *val);
1923 		break;
1924 	case KVM_REG_PPC_DABRX:
1925 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1926 		break;
1927 	case KVM_REG_PPC_DSCR:
1928 		vcpu->arch.dscr = set_reg_val(id, *val);
1929 		break;
1930 	case KVM_REG_PPC_PURR:
1931 		vcpu->arch.purr = set_reg_val(id, *val);
1932 		break;
1933 	case KVM_REG_PPC_SPURR:
1934 		vcpu->arch.spurr = set_reg_val(id, *val);
1935 		break;
1936 	case KVM_REG_PPC_AMR:
1937 		vcpu->arch.amr = set_reg_val(id, *val);
1938 		break;
1939 	case KVM_REG_PPC_UAMOR:
1940 		vcpu->arch.uamor = set_reg_val(id, *val);
1941 		break;
1942 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1943 		i = id - KVM_REG_PPC_MMCR0;
1944 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1945 		break;
1946 	case KVM_REG_PPC_MMCR2:
1947 		vcpu->arch.mmcr[2] = set_reg_val(id, *val);
1948 		break;
1949 	case KVM_REG_PPC_MMCRA:
1950 		vcpu->arch.mmcra = set_reg_val(id, *val);
1951 		break;
1952 	case KVM_REG_PPC_MMCRS:
1953 		vcpu->arch.mmcrs = set_reg_val(id, *val);
1954 		break;
1955 	case KVM_REG_PPC_MMCR3:
1956 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
1957 		break;
1958 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1959 		i = id - KVM_REG_PPC_PMC1;
1960 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1961 		break;
1962 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1963 		i = id - KVM_REG_PPC_SPMC1;
1964 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1965 		break;
1966 	case KVM_REG_PPC_SIAR:
1967 		vcpu->arch.siar = set_reg_val(id, *val);
1968 		break;
1969 	case KVM_REG_PPC_SDAR:
1970 		vcpu->arch.sdar = set_reg_val(id, *val);
1971 		break;
1972 	case KVM_REG_PPC_SIER:
1973 		vcpu->arch.sier[0] = set_reg_val(id, *val);
1974 		break;
1975 	case KVM_REG_PPC_SIER2:
1976 		vcpu->arch.sier[1] = set_reg_val(id, *val);
1977 		break;
1978 	case KVM_REG_PPC_SIER3:
1979 		vcpu->arch.sier[2] = set_reg_val(id, *val);
1980 		break;
1981 	case KVM_REG_PPC_IAMR:
1982 		vcpu->arch.iamr = set_reg_val(id, *val);
1983 		break;
1984 	case KVM_REG_PPC_PSPB:
1985 		vcpu->arch.pspb = set_reg_val(id, *val);
1986 		break;
1987 	case KVM_REG_PPC_DPDES:
1988 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1989 		break;
1990 	case KVM_REG_PPC_VTB:
1991 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1992 		break;
1993 	case KVM_REG_PPC_DAWR:
1994 		vcpu->arch.dawr = set_reg_val(id, *val);
1995 		break;
1996 	case KVM_REG_PPC_DAWRX:
1997 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1998 		break;
1999 	case KVM_REG_PPC_CIABR:
2000 		vcpu->arch.ciabr = set_reg_val(id, *val);
2001 		/* Don't allow setting breakpoints in hypervisor code */
2002 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2003 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
2004 		break;
2005 	case KVM_REG_PPC_CSIGR:
2006 		vcpu->arch.csigr = set_reg_val(id, *val);
2007 		break;
2008 	case KVM_REG_PPC_TACR:
2009 		vcpu->arch.tacr = set_reg_val(id, *val);
2010 		break;
2011 	case KVM_REG_PPC_TCSCR:
2012 		vcpu->arch.tcscr = set_reg_val(id, *val);
2013 		break;
2014 	case KVM_REG_PPC_PID:
2015 		vcpu->arch.pid = set_reg_val(id, *val);
2016 		break;
2017 	case KVM_REG_PPC_ACOP:
2018 		vcpu->arch.acop = set_reg_val(id, *val);
2019 		break;
2020 	case KVM_REG_PPC_WORT:
2021 		vcpu->arch.wort = set_reg_val(id, *val);
2022 		break;
2023 	case KVM_REG_PPC_TIDR:
2024 		vcpu->arch.tid = set_reg_val(id, *val);
2025 		break;
2026 	case KVM_REG_PPC_PSSCR:
2027 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2028 		break;
2029 	case KVM_REG_PPC_VPA_ADDR:
2030 		addr = set_reg_val(id, *val);
2031 		r = -EINVAL;
2032 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2033 			      vcpu->arch.dtl.next_gpa))
2034 			break;
2035 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2036 		break;
2037 	case KVM_REG_PPC_VPA_SLB:
2038 		addr = val->vpaval.addr;
2039 		len = val->vpaval.length;
2040 		r = -EINVAL;
2041 		if (addr && !vcpu->arch.vpa.next_gpa)
2042 			break;
2043 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2044 		break;
2045 	case KVM_REG_PPC_VPA_DTL:
2046 		addr = val->vpaval.addr;
2047 		len = val->vpaval.length;
2048 		r = -EINVAL;
2049 		if (addr && (len < sizeof(struct dtl_entry) ||
2050 			     !vcpu->arch.vpa.next_gpa))
2051 			break;
2052 		len -= len % sizeof(struct dtl_entry);
2053 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2054 		break;
2055 	case KVM_REG_PPC_TB_OFFSET:
2056 		/* round up to multiple of 2^24 */
2057 		vcpu->arch.vcore->tb_offset =
2058 			ALIGN(set_reg_val(id, *val), 1UL << 24);
2059 		break;
2060 	case KVM_REG_PPC_LPCR:
2061 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2062 		break;
2063 	case KVM_REG_PPC_LPCR_64:
2064 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2065 		break;
2066 	case KVM_REG_PPC_PPR:
2067 		vcpu->arch.ppr = set_reg_val(id, *val);
2068 		break;
2069 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2070 	case KVM_REG_PPC_TFHAR:
2071 		vcpu->arch.tfhar = set_reg_val(id, *val);
2072 		break;
2073 	case KVM_REG_PPC_TFIAR:
2074 		vcpu->arch.tfiar = set_reg_val(id, *val);
2075 		break;
2076 	case KVM_REG_PPC_TEXASR:
2077 		vcpu->arch.texasr = set_reg_val(id, *val);
2078 		break;
2079 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2080 		i = id - KVM_REG_PPC_TM_GPR0;
2081 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2082 		break;
2083 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2084 	{
2085 		int j;
2086 		i = id - KVM_REG_PPC_TM_VSR0;
2087 		if (i < 32)
2088 			for (j = 0; j < TS_FPRWIDTH; j++)
2089 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2090 		else
2091 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2092 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2093 			else
2094 				r = -ENXIO;
2095 		break;
2096 	}
2097 	case KVM_REG_PPC_TM_CR:
2098 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2099 		break;
2100 	case KVM_REG_PPC_TM_XER:
2101 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2102 		break;
2103 	case KVM_REG_PPC_TM_LR:
2104 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2105 		break;
2106 	case KVM_REG_PPC_TM_CTR:
2107 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2108 		break;
2109 	case KVM_REG_PPC_TM_FPSCR:
2110 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2111 		break;
2112 	case KVM_REG_PPC_TM_AMR:
2113 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2114 		break;
2115 	case KVM_REG_PPC_TM_PPR:
2116 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2117 		break;
2118 	case KVM_REG_PPC_TM_VRSAVE:
2119 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2120 		break;
2121 	case KVM_REG_PPC_TM_VSCR:
2122 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2123 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2124 		else
2125 			r = - ENXIO;
2126 		break;
2127 	case KVM_REG_PPC_TM_DSCR:
2128 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2129 		break;
2130 	case KVM_REG_PPC_TM_TAR:
2131 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2132 		break;
2133 #endif
2134 	case KVM_REG_PPC_ARCH_COMPAT:
2135 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2136 		break;
2137 	case KVM_REG_PPC_DEC_EXPIRY:
2138 		vcpu->arch.dec_expires = set_reg_val(id, *val) -
2139 			vcpu->arch.vcore->tb_offset;
2140 		break;
2141 	case KVM_REG_PPC_ONLINE:
2142 		i = set_reg_val(id, *val);
2143 		if (i && !vcpu->arch.online)
2144 			atomic_inc(&vcpu->arch.vcore->online_count);
2145 		else if (!i && vcpu->arch.online)
2146 			atomic_dec(&vcpu->arch.vcore->online_count);
2147 		vcpu->arch.online = i;
2148 		break;
2149 	case KVM_REG_PPC_PTCR:
2150 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2151 		break;
2152 	default:
2153 		r = -EINVAL;
2154 		break;
2155 	}
2156 
2157 	return r;
2158 }
2159 
2160 /*
2161  * On POWER9, threads are independent and can be in different partitions.
2162  * Therefore we consider each thread to be a subcore.
2163  * There is a restriction that all threads have to be in the same
2164  * MMU mode (radix or HPT), unfortunately, but since we only support
2165  * HPT guests on a HPT host so far, that isn't an impediment yet.
2166  */
2167 static int threads_per_vcore(struct kvm *kvm)
2168 {
2169 	if (kvm->arch.threads_indep)
2170 		return 1;
2171 	return threads_per_subcore;
2172 }
2173 
2174 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2175 {
2176 	struct kvmppc_vcore *vcore;
2177 
2178 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2179 
2180 	if (vcore == NULL)
2181 		return NULL;
2182 
2183 	spin_lock_init(&vcore->lock);
2184 	spin_lock_init(&vcore->stoltb_lock);
2185 	rcuwait_init(&vcore->wait);
2186 	vcore->preempt_tb = TB_NIL;
2187 	vcore->lpcr = kvm->arch.lpcr;
2188 	vcore->first_vcpuid = id;
2189 	vcore->kvm = kvm;
2190 	INIT_LIST_HEAD(&vcore->preempt_list);
2191 
2192 	return vcore;
2193 }
2194 
2195 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2196 static struct debugfs_timings_element {
2197 	const char *name;
2198 	size_t offset;
2199 } timings[] = {
2200 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2201 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2202 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2203 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2204 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2205 };
2206 
2207 #define N_TIMINGS	(ARRAY_SIZE(timings))
2208 
2209 struct debugfs_timings_state {
2210 	struct kvm_vcpu	*vcpu;
2211 	unsigned int	buflen;
2212 	char		buf[N_TIMINGS * 100];
2213 };
2214 
2215 static int debugfs_timings_open(struct inode *inode, struct file *file)
2216 {
2217 	struct kvm_vcpu *vcpu = inode->i_private;
2218 	struct debugfs_timings_state *p;
2219 
2220 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2221 	if (!p)
2222 		return -ENOMEM;
2223 
2224 	kvm_get_kvm(vcpu->kvm);
2225 	p->vcpu = vcpu;
2226 	file->private_data = p;
2227 
2228 	return nonseekable_open(inode, file);
2229 }
2230 
2231 static int debugfs_timings_release(struct inode *inode, struct file *file)
2232 {
2233 	struct debugfs_timings_state *p = file->private_data;
2234 
2235 	kvm_put_kvm(p->vcpu->kvm);
2236 	kfree(p);
2237 	return 0;
2238 }
2239 
2240 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2241 				    size_t len, loff_t *ppos)
2242 {
2243 	struct debugfs_timings_state *p = file->private_data;
2244 	struct kvm_vcpu *vcpu = p->vcpu;
2245 	char *s, *buf_end;
2246 	struct kvmhv_tb_accumulator tb;
2247 	u64 count;
2248 	loff_t pos;
2249 	ssize_t n;
2250 	int i, loops;
2251 	bool ok;
2252 
2253 	if (!p->buflen) {
2254 		s = p->buf;
2255 		buf_end = s + sizeof(p->buf);
2256 		for (i = 0; i < N_TIMINGS; ++i) {
2257 			struct kvmhv_tb_accumulator *acc;
2258 
2259 			acc = (struct kvmhv_tb_accumulator *)
2260 				((unsigned long)vcpu + timings[i].offset);
2261 			ok = false;
2262 			for (loops = 0; loops < 1000; ++loops) {
2263 				count = acc->seqcount;
2264 				if (!(count & 1)) {
2265 					smp_rmb();
2266 					tb = *acc;
2267 					smp_rmb();
2268 					if (count == acc->seqcount) {
2269 						ok = true;
2270 						break;
2271 					}
2272 				}
2273 				udelay(1);
2274 			}
2275 			if (!ok)
2276 				snprintf(s, buf_end - s, "%s: stuck\n",
2277 					timings[i].name);
2278 			else
2279 				snprintf(s, buf_end - s,
2280 					"%s: %llu %llu %llu %llu\n",
2281 					timings[i].name, count / 2,
2282 					tb_to_ns(tb.tb_total),
2283 					tb_to_ns(tb.tb_min),
2284 					tb_to_ns(tb.tb_max));
2285 			s += strlen(s);
2286 		}
2287 		p->buflen = s - p->buf;
2288 	}
2289 
2290 	pos = *ppos;
2291 	if (pos >= p->buflen)
2292 		return 0;
2293 	if (len > p->buflen - pos)
2294 		len = p->buflen - pos;
2295 	n = copy_to_user(buf, p->buf + pos, len);
2296 	if (n) {
2297 		if (n == len)
2298 			return -EFAULT;
2299 		len -= n;
2300 	}
2301 	*ppos = pos + len;
2302 	return len;
2303 }
2304 
2305 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2306 				     size_t len, loff_t *ppos)
2307 {
2308 	return -EACCES;
2309 }
2310 
2311 static const struct file_operations debugfs_timings_ops = {
2312 	.owner	 = THIS_MODULE,
2313 	.open	 = debugfs_timings_open,
2314 	.release = debugfs_timings_release,
2315 	.read	 = debugfs_timings_read,
2316 	.write	 = debugfs_timings_write,
2317 	.llseek	 = generic_file_llseek,
2318 };
2319 
2320 /* Create a debugfs directory for the vcpu */
2321 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2322 {
2323 	char buf[16];
2324 	struct kvm *kvm = vcpu->kvm;
2325 
2326 	snprintf(buf, sizeof(buf), "vcpu%u", id);
2327 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2328 	debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2329 			    &debugfs_timings_ops);
2330 }
2331 
2332 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2333 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2334 {
2335 }
2336 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2337 
2338 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2339 {
2340 	int err;
2341 	int core;
2342 	struct kvmppc_vcore *vcore;
2343 	struct kvm *kvm;
2344 	unsigned int id;
2345 
2346 	kvm = vcpu->kvm;
2347 	id = vcpu->vcpu_id;
2348 
2349 	vcpu->arch.shared = &vcpu->arch.shregs;
2350 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2351 	/*
2352 	 * The shared struct is never shared on HV,
2353 	 * so we can always use host endianness
2354 	 */
2355 #ifdef __BIG_ENDIAN__
2356 	vcpu->arch.shared_big_endian = true;
2357 #else
2358 	vcpu->arch.shared_big_endian = false;
2359 #endif
2360 #endif
2361 	vcpu->arch.mmcr[0] = MMCR0_FC;
2362 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2363 	/* default to host PVR, since we can't spoof it */
2364 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2365 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2366 	spin_lock_init(&vcpu->arch.tbacct_lock);
2367 	vcpu->arch.busy_preempt = TB_NIL;
2368 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2369 
2370 	/*
2371 	 * Set the default HFSCR for the guest from the host value.
2372 	 * This value is only used on POWER9.
2373 	 * On POWER9, we want to virtualize the doorbell facility, so we
2374 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2375 	 * to trap and then we emulate them.
2376 	 */
2377 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2378 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2379 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2380 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2381 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2382 			vcpu->arch.hfscr |= HFSCR_TM;
2383 	}
2384 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2385 		vcpu->arch.hfscr |= HFSCR_TM;
2386 
2387 	kvmppc_mmu_book3s_hv_init(vcpu);
2388 
2389 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2390 
2391 	init_waitqueue_head(&vcpu->arch.cpu_run);
2392 
2393 	mutex_lock(&kvm->lock);
2394 	vcore = NULL;
2395 	err = -EINVAL;
2396 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2397 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2398 			pr_devel("KVM: VCPU ID too high\n");
2399 			core = KVM_MAX_VCORES;
2400 		} else {
2401 			BUG_ON(kvm->arch.smt_mode != 1);
2402 			core = kvmppc_pack_vcpu_id(kvm, id);
2403 		}
2404 	} else {
2405 		core = id / kvm->arch.smt_mode;
2406 	}
2407 	if (core < KVM_MAX_VCORES) {
2408 		vcore = kvm->arch.vcores[core];
2409 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2410 			pr_devel("KVM: collision on id %u", id);
2411 			vcore = NULL;
2412 		} else if (!vcore) {
2413 			/*
2414 			 * Take mmu_setup_lock for mutual exclusion
2415 			 * with kvmppc_update_lpcr().
2416 			 */
2417 			err = -ENOMEM;
2418 			vcore = kvmppc_vcore_create(kvm,
2419 					id & ~(kvm->arch.smt_mode - 1));
2420 			mutex_lock(&kvm->arch.mmu_setup_lock);
2421 			kvm->arch.vcores[core] = vcore;
2422 			kvm->arch.online_vcores++;
2423 			mutex_unlock(&kvm->arch.mmu_setup_lock);
2424 		}
2425 	}
2426 	mutex_unlock(&kvm->lock);
2427 
2428 	if (!vcore)
2429 		return err;
2430 
2431 	spin_lock(&vcore->lock);
2432 	++vcore->num_threads;
2433 	spin_unlock(&vcore->lock);
2434 	vcpu->arch.vcore = vcore;
2435 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2436 	vcpu->arch.thread_cpu = -1;
2437 	vcpu->arch.prev_cpu = -1;
2438 
2439 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2440 	kvmppc_sanity_check(vcpu);
2441 
2442 	debugfs_vcpu_init(vcpu, id);
2443 
2444 	return 0;
2445 }
2446 
2447 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2448 			      unsigned long flags)
2449 {
2450 	int err;
2451 	int esmt = 0;
2452 
2453 	if (flags)
2454 		return -EINVAL;
2455 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2456 		return -EINVAL;
2457 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2458 		/*
2459 		 * On POWER8 (or POWER7), the threading mode is "strict",
2460 		 * so we pack smt_mode vcpus per vcore.
2461 		 */
2462 		if (smt_mode > threads_per_subcore)
2463 			return -EINVAL;
2464 	} else {
2465 		/*
2466 		 * On POWER9, the threading mode is "loose",
2467 		 * so each vcpu gets its own vcore.
2468 		 */
2469 		esmt = smt_mode;
2470 		smt_mode = 1;
2471 	}
2472 	mutex_lock(&kvm->lock);
2473 	err = -EBUSY;
2474 	if (!kvm->arch.online_vcores) {
2475 		kvm->arch.smt_mode = smt_mode;
2476 		kvm->arch.emul_smt_mode = esmt;
2477 		err = 0;
2478 	}
2479 	mutex_unlock(&kvm->lock);
2480 
2481 	return err;
2482 }
2483 
2484 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2485 {
2486 	if (vpa->pinned_addr)
2487 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2488 					vpa->dirty);
2489 }
2490 
2491 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2492 {
2493 	spin_lock(&vcpu->arch.vpa_update_lock);
2494 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2495 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2496 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2497 	spin_unlock(&vcpu->arch.vpa_update_lock);
2498 }
2499 
2500 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2501 {
2502 	/* Indicate we want to get back into the guest */
2503 	return 1;
2504 }
2505 
2506 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2507 {
2508 	unsigned long dec_nsec, now;
2509 
2510 	now = get_tb();
2511 	if (now > vcpu->arch.dec_expires) {
2512 		/* decrementer has already gone negative */
2513 		kvmppc_core_queue_dec(vcpu);
2514 		kvmppc_core_prepare_to_enter(vcpu);
2515 		return;
2516 	}
2517 	dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2518 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2519 	vcpu->arch.timer_running = 1;
2520 }
2521 
2522 extern int __kvmppc_vcore_entry(void);
2523 
2524 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2525 				   struct kvm_vcpu *vcpu)
2526 {
2527 	u64 now;
2528 
2529 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2530 		return;
2531 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2532 	now = mftb();
2533 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2534 		vcpu->arch.stolen_logged;
2535 	vcpu->arch.busy_preempt = now;
2536 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2537 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2538 	--vc->n_runnable;
2539 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2540 }
2541 
2542 static int kvmppc_grab_hwthread(int cpu)
2543 {
2544 	struct paca_struct *tpaca;
2545 	long timeout = 10000;
2546 
2547 	tpaca = paca_ptrs[cpu];
2548 
2549 	/* Ensure the thread won't go into the kernel if it wakes */
2550 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2551 	tpaca->kvm_hstate.kvm_vcore = NULL;
2552 	tpaca->kvm_hstate.napping = 0;
2553 	smp_wmb();
2554 	tpaca->kvm_hstate.hwthread_req = 1;
2555 
2556 	/*
2557 	 * If the thread is already executing in the kernel (e.g. handling
2558 	 * a stray interrupt), wait for it to get back to nap mode.
2559 	 * The smp_mb() is to ensure that our setting of hwthread_req
2560 	 * is visible before we look at hwthread_state, so if this
2561 	 * races with the code at system_reset_pSeries and the thread
2562 	 * misses our setting of hwthread_req, we are sure to see its
2563 	 * setting of hwthread_state, and vice versa.
2564 	 */
2565 	smp_mb();
2566 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2567 		if (--timeout <= 0) {
2568 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2569 			return -EBUSY;
2570 		}
2571 		udelay(1);
2572 	}
2573 	return 0;
2574 }
2575 
2576 static void kvmppc_release_hwthread(int cpu)
2577 {
2578 	struct paca_struct *tpaca;
2579 
2580 	tpaca = paca_ptrs[cpu];
2581 	tpaca->kvm_hstate.hwthread_req = 0;
2582 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2583 	tpaca->kvm_hstate.kvm_vcore = NULL;
2584 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2585 }
2586 
2587 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2588 {
2589 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2590 	cpumask_t *cpu_in_guest;
2591 	int i;
2592 
2593 	cpu = cpu_first_thread_sibling(cpu);
2594 	if (nested) {
2595 		cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2596 		cpu_in_guest = &nested->cpu_in_guest;
2597 	} else {
2598 		cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2599 		cpu_in_guest = &kvm->arch.cpu_in_guest;
2600 	}
2601 	/*
2602 	 * Make sure setting of bit in need_tlb_flush precedes
2603 	 * testing of cpu_in_guest bits.  The matching barrier on
2604 	 * the other side is the first smp_mb() in kvmppc_run_core().
2605 	 */
2606 	smp_mb();
2607 	for (i = 0; i < threads_per_core; ++i)
2608 		if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2609 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2610 }
2611 
2612 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2613 {
2614 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2615 	struct kvm *kvm = vcpu->kvm;
2616 	int prev_cpu;
2617 
2618 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2619 		return;
2620 
2621 	if (nested)
2622 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2623 	else
2624 		prev_cpu = vcpu->arch.prev_cpu;
2625 
2626 	/*
2627 	 * With radix, the guest can do TLB invalidations itself,
2628 	 * and it could choose to use the local form (tlbiel) if
2629 	 * it is invalidating a translation that has only ever been
2630 	 * used on one vcpu.  However, that doesn't mean it has
2631 	 * only ever been used on one physical cpu, since vcpus
2632 	 * can move around between pcpus.  To cope with this, when
2633 	 * a vcpu moves from one pcpu to another, we need to tell
2634 	 * any vcpus running on the same core as this vcpu previously
2635 	 * ran to flush the TLB.  The TLB is shared between threads,
2636 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2637 	 */
2638 	if (prev_cpu != pcpu) {
2639 		if (prev_cpu >= 0 &&
2640 		    cpu_first_thread_sibling(prev_cpu) !=
2641 		    cpu_first_thread_sibling(pcpu))
2642 			radix_flush_cpu(kvm, prev_cpu, vcpu);
2643 		if (nested)
2644 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2645 		else
2646 			vcpu->arch.prev_cpu = pcpu;
2647 	}
2648 }
2649 
2650 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2651 {
2652 	int cpu;
2653 	struct paca_struct *tpaca;
2654 	struct kvm *kvm = vc->kvm;
2655 
2656 	cpu = vc->pcpu;
2657 	if (vcpu) {
2658 		if (vcpu->arch.timer_running) {
2659 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2660 			vcpu->arch.timer_running = 0;
2661 		}
2662 		cpu += vcpu->arch.ptid;
2663 		vcpu->cpu = vc->pcpu;
2664 		vcpu->arch.thread_cpu = cpu;
2665 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2666 	}
2667 	tpaca = paca_ptrs[cpu];
2668 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2669 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2670 	tpaca->kvm_hstate.fake_suspend = 0;
2671 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2672 	smp_wmb();
2673 	tpaca->kvm_hstate.kvm_vcore = vc;
2674 	if (cpu != smp_processor_id())
2675 		kvmppc_ipi_thread(cpu);
2676 }
2677 
2678 static void kvmppc_wait_for_nap(int n_threads)
2679 {
2680 	int cpu = smp_processor_id();
2681 	int i, loops;
2682 
2683 	if (n_threads <= 1)
2684 		return;
2685 	for (loops = 0; loops < 1000000; ++loops) {
2686 		/*
2687 		 * Check if all threads are finished.
2688 		 * We set the vcore pointer when starting a thread
2689 		 * and the thread clears it when finished, so we look
2690 		 * for any threads that still have a non-NULL vcore ptr.
2691 		 */
2692 		for (i = 1; i < n_threads; ++i)
2693 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2694 				break;
2695 		if (i == n_threads) {
2696 			HMT_medium();
2697 			return;
2698 		}
2699 		HMT_low();
2700 	}
2701 	HMT_medium();
2702 	for (i = 1; i < n_threads; ++i)
2703 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2704 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2705 }
2706 
2707 /*
2708  * Check that we are on thread 0 and that any other threads in
2709  * this core are off-line.  Then grab the threads so they can't
2710  * enter the kernel.
2711  */
2712 static int on_primary_thread(void)
2713 {
2714 	int cpu = smp_processor_id();
2715 	int thr;
2716 
2717 	/* Are we on a primary subcore? */
2718 	if (cpu_thread_in_subcore(cpu))
2719 		return 0;
2720 
2721 	thr = 0;
2722 	while (++thr < threads_per_subcore)
2723 		if (cpu_online(cpu + thr))
2724 			return 0;
2725 
2726 	/* Grab all hw threads so they can't go into the kernel */
2727 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2728 		if (kvmppc_grab_hwthread(cpu + thr)) {
2729 			/* Couldn't grab one; let the others go */
2730 			do {
2731 				kvmppc_release_hwthread(cpu + thr);
2732 			} while (--thr > 0);
2733 			return 0;
2734 		}
2735 	}
2736 	return 1;
2737 }
2738 
2739 /*
2740  * A list of virtual cores for each physical CPU.
2741  * These are vcores that could run but their runner VCPU tasks are
2742  * (or may be) preempted.
2743  */
2744 struct preempted_vcore_list {
2745 	struct list_head	list;
2746 	spinlock_t		lock;
2747 };
2748 
2749 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2750 
2751 static void init_vcore_lists(void)
2752 {
2753 	int cpu;
2754 
2755 	for_each_possible_cpu(cpu) {
2756 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2757 		spin_lock_init(&lp->lock);
2758 		INIT_LIST_HEAD(&lp->list);
2759 	}
2760 }
2761 
2762 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2763 {
2764 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2765 
2766 	vc->vcore_state = VCORE_PREEMPT;
2767 	vc->pcpu = smp_processor_id();
2768 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2769 		spin_lock(&lp->lock);
2770 		list_add_tail(&vc->preempt_list, &lp->list);
2771 		spin_unlock(&lp->lock);
2772 	}
2773 
2774 	/* Start accumulating stolen time */
2775 	kvmppc_core_start_stolen(vc);
2776 }
2777 
2778 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2779 {
2780 	struct preempted_vcore_list *lp;
2781 
2782 	kvmppc_core_end_stolen(vc);
2783 	if (!list_empty(&vc->preempt_list)) {
2784 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2785 		spin_lock(&lp->lock);
2786 		list_del_init(&vc->preempt_list);
2787 		spin_unlock(&lp->lock);
2788 	}
2789 	vc->vcore_state = VCORE_INACTIVE;
2790 }
2791 
2792 /*
2793  * This stores information about the virtual cores currently
2794  * assigned to a physical core.
2795  */
2796 struct core_info {
2797 	int		n_subcores;
2798 	int		max_subcore_threads;
2799 	int		total_threads;
2800 	int		subcore_threads[MAX_SUBCORES];
2801 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2802 };
2803 
2804 /*
2805  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2806  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2807  */
2808 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2809 
2810 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2811 {
2812 	memset(cip, 0, sizeof(*cip));
2813 	cip->n_subcores = 1;
2814 	cip->max_subcore_threads = vc->num_threads;
2815 	cip->total_threads = vc->num_threads;
2816 	cip->subcore_threads[0] = vc->num_threads;
2817 	cip->vc[0] = vc;
2818 }
2819 
2820 static bool subcore_config_ok(int n_subcores, int n_threads)
2821 {
2822 	/*
2823 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2824 	 * split-core mode, with one thread per subcore.
2825 	 */
2826 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2827 		return n_subcores <= 4 && n_threads == 1;
2828 
2829 	/* On POWER8, can only dynamically split if unsplit to begin with */
2830 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2831 		return false;
2832 	if (n_subcores > MAX_SUBCORES)
2833 		return false;
2834 	if (n_subcores > 1) {
2835 		if (!(dynamic_mt_modes & 2))
2836 			n_subcores = 4;
2837 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2838 			return false;
2839 	}
2840 
2841 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2842 }
2843 
2844 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2845 {
2846 	vc->entry_exit_map = 0;
2847 	vc->in_guest = 0;
2848 	vc->napping_threads = 0;
2849 	vc->conferring_threads = 0;
2850 	vc->tb_offset_applied = 0;
2851 }
2852 
2853 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2854 {
2855 	int n_threads = vc->num_threads;
2856 	int sub;
2857 
2858 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2859 		return false;
2860 
2861 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
2862 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2863 		return false;
2864 
2865 	/* Some POWER9 chips require all threads to be in the same MMU mode */
2866 	if (no_mixing_hpt_and_radix &&
2867 	    kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2868 		return false;
2869 
2870 	if (n_threads < cip->max_subcore_threads)
2871 		n_threads = cip->max_subcore_threads;
2872 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2873 		return false;
2874 	cip->max_subcore_threads = n_threads;
2875 
2876 	sub = cip->n_subcores;
2877 	++cip->n_subcores;
2878 	cip->total_threads += vc->num_threads;
2879 	cip->subcore_threads[sub] = vc->num_threads;
2880 	cip->vc[sub] = vc;
2881 	init_vcore_to_run(vc);
2882 	list_del_init(&vc->preempt_list);
2883 
2884 	return true;
2885 }
2886 
2887 /*
2888  * Work out whether it is possible to piggyback the execution of
2889  * vcore *pvc onto the execution of the other vcores described in *cip.
2890  */
2891 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2892 			  int target_threads)
2893 {
2894 	if (cip->total_threads + pvc->num_threads > target_threads)
2895 		return false;
2896 
2897 	return can_dynamic_split(pvc, cip);
2898 }
2899 
2900 static void prepare_threads(struct kvmppc_vcore *vc)
2901 {
2902 	int i;
2903 	struct kvm_vcpu *vcpu;
2904 
2905 	for_each_runnable_thread(i, vcpu, vc) {
2906 		if (signal_pending(vcpu->arch.run_task))
2907 			vcpu->arch.ret = -EINTR;
2908 		else if (vcpu->arch.vpa.update_pending ||
2909 			 vcpu->arch.slb_shadow.update_pending ||
2910 			 vcpu->arch.dtl.update_pending)
2911 			vcpu->arch.ret = RESUME_GUEST;
2912 		else
2913 			continue;
2914 		kvmppc_remove_runnable(vc, vcpu);
2915 		wake_up(&vcpu->arch.cpu_run);
2916 	}
2917 }
2918 
2919 static void collect_piggybacks(struct core_info *cip, int target_threads)
2920 {
2921 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2922 	struct kvmppc_vcore *pvc, *vcnext;
2923 
2924 	spin_lock(&lp->lock);
2925 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2926 		if (!spin_trylock(&pvc->lock))
2927 			continue;
2928 		prepare_threads(pvc);
2929 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2930 			list_del_init(&pvc->preempt_list);
2931 			if (pvc->runner == NULL) {
2932 				pvc->vcore_state = VCORE_INACTIVE;
2933 				kvmppc_core_end_stolen(pvc);
2934 			}
2935 			spin_unlock(&pvc->lock);
2936 			continue;
2937 		}
2938 		if (!can_piggyback(pvc, cip, target_threads)) {
2939 			spin_unlock(&pvc->lock);
2940 			continue;
2941 		}
2942 		kvmppc_core_end_stolen(pvc);
2943 		pvc->vcore_state = VCORE_PIGGYBACK;
2944 		if (cip->total_threads >= target_threads)
2945 			break;
2946 	}
2947 	spin_unlock(&lp->lock);
2948 }
2949 
2950 static bool recheck_signals_and_mmu(struct core_info *cip)
2951 {
2952 	int sub, i;
2953 	struct kvm_vcpu *vcpu;
2954 	struct kvmppc_vcore *vc;
2955 
2956 	for (sub = 0; sub < cip->n_subcores; ++sub) {
2957 		vc = cip->vc[sub];
2958 		if (!vc->kvm->arch.mmu_ready)
2959 			return true;
2960 		for_each_runnable_thread(i, vcpu, vc)
2961 			if (signal_pending(vcpu->arch.run_task))
2962 				return true;
2963 	}
2964 	return false;
2965 }
2966 
2967 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2968 {
2969 	int still_running = 0, i;
2970 	u64 now;
2971 	long ret;
2972 	struct kvm_vcpu *vcpu;
2973 
2974 	spin_lock(&vc->lock);
2975 	now = get_tb();
2976 	for_each_runnable_thread(i, vcpu, vc) {
2977 		/*
2978 		 * It's safe to unlock the vcore in the loop here, because
2979 		 * for_each_runnable_thread() is safe against removal of
2980 		 * the vcpu, and the vcore state is VCORE_EXITING here,
2981 		 * so any vcpus becoming runnable will have their arch.trap
2982 		 * set to zero and can't actually run in the guest.
2983 		 */
2984 		spin_unlock(&vc->lock);
2985 		/* cancel pending dec exception if dec is positive */
2986 		if (now < vcpu->arch.dec_expires &&
2987 		    kvmppc_core_pending_dec(vcpu))
2988 			kvmppc_core_dequeue_dec(vcpu);
2989 
2990 		trace_kvm_guest_exit(vcpu);
2991 
2992 		ret = RESUME_GUEST;
2993 		if (vcpu->arch.trap)
2994 			ret = kvmppc_handle_exit_hv(vcpu,
2995 						    vcpu->arch.run_task);
2996 
2997 		vcpu->arch.ret = ret;
2998 		vcpu->arch.trap = 0;
2999 
3000 		spin_lock(&vc->lock);
3001 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3002 			if (vcpu->arch.pending_exceptions)
3003 				kvmppc_core_prepare_to_enter(vcpu);
3004 			if (vcpu->arch.ceded)
3005 				kvmppc_set_timer(vcpu);
3006 			else
3007 				++still_running;
3008 		} else {
3009 			kvmppc_remove_runnable(vc, vcpu);
3010 			wake_up(&vcpu->arch.cpu_run);
3011 		}
3012 	}
3013 	if (!is_master) {
3014 		if (still_running > 0) {
3015 			kvmppc_vcore_preempt(vc);
3016 		} else if (vc->runner) {
3017 			vc->vcore_state = VCORE_PREEMPT;
3018 			kvmppc_core_start_stolen(vc);
3019 		} else {
3020 			vc->vcore_state = VCORE_INACTIVE;
3021 		}
3022 		if (vc->n_runnable > 0 && vc->runner == NULL) {
3023 			/* make sure there's a candidate runner awake */
3024 			i = -1;
3025 			vcpu = next_runnable_thread(vc, &i);
3026 			wake_up(&vcpu->arch.cpu_run);
3027 		}
3028 	}
3029 	spin_unlock(&vc->lock);
3030 }
3031 
3032 /*
3033  * Clear core from the list of active host cores as we are about to
3034  * enter the guest. Only do this if it is the primary thread of the
3035  * core (not if a subcore) that is entering the guest.
3036  */
3037 static inline int kvmppc_clear_host_core(unsigned int cpu)
3038 {
3039 	int core;
3040 
3041 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3042 		return 0;
3043 	/*
3044 	 * Memory barrier can be omitted here as we will do a smp_wmb()
3045 	 * later in kvmppc_start_thread and we need ensure that state is
3046 	 * visible to other CPUs only after we enter guest.
3047 	 */
3048 	core = cpu >> threads_shift;
3049 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3050 	return 0;
3051 }
3052 
3053 /*
3054  * Advertise this core as an active host core since we exited the guest
3055  * Only need to do this if it is the primary thread of the core that is
3056  * exiting.
3057  */
3058 static inline int kvmppc_set_host_core(unsigned int cpu)
3059 {
3060 	int core;
3061 
3062 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3063 		return 0;
3064 
3065 	/*
3066 	 * Memory barrier can be omitted here because we do a spin_unlock
3067 	 * immediately after this which provides the memory barrier.
3068 	 */
3069 	core = cpu >> threads_shift;
3070 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3071 	return 0;
3072 }
3073 
3074 static void set_irq_happened(int trap)
3075 {
3076 	switch (trap) {
3077 	case BOOK3S_INTERRUPT_EXTERNAL:
3078 		local_paca->irq_happened |= PACA_IRQ_EE;
3079 		break;
3080 	case BOOK3S_INTERRUPT_H_DOORBELL:
3081 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3082 		break;
3083 	case BOOK3S_INTERRUPT_HMI:
3084 		local_paca->irq_happened |= PACA_IRQ_HMI;
3085 		break;
3086 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3087 		replay_system_reset();
3088 		break;
3089 	}
3090 }
3091 
3092 /*
3093  * Run a set of guest threads on a physical core.
3094  * Called with vc->lock held.
3095  */
3096 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3097 {
3098 	struct kvm_vcpu *vcpu;
3099 	int i;
3100 	int srcu_idx;
3101 	struct core_info core_info;
3102 	struct kvmppc_vcore *pvc;
3103 	struct kvm_split_mode split_info, *sip;
3104 	int split, subcore_size, active;
3105 	int sub;
3106 	bool thr0_done;
3107 	unsigned long cmd_bit, stat_bit;
3108 	int pcpu, thr;
3109 	int target_threads;
3110 	int controlled_threads;
3111 	int trap;
3112 	bool is_power8;
3113 	bool hpt_on_radix;
3114 
3115 	/*
3116 	 * Remove from the list any threads that have a signal pending
3117 	 * or need a VPA update done
3118 	 */
3119 	prepare_threads(vc);
3120 
3121 	/* if the runner is no longer runnable, let the caller pick a new one */
3122 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3123 		return;
3124 
3125 	/*
3126 	 * Initialize *vc.
3127 	 */
3128 	init_vcore_to_run(vc);
3129 	vc->preempt_tb = TB_NIL;
3130 
3131 	/*
3132 	 * Number of threads that we will be controlling: the same as
3133 	 * the number of threads per subcore, except on POWER9,
3134 	 * where it's 1 because the threads are (mostly) independent.
3135 	 */
3136 	controlled_threads = threads_per_vcore(vc->kvm);
3137 
3138 	/*
3139 	 * Make sure we are running on primary threads, and that secondary
3140 	 * threads are offline.  Also check if the number of threads in this
3141 	 * guest are greater than the current system threads per guest.
3142 	 * On POWER9, we need to be not in independent-threads mode if
3143 	 * this is a HPT guest on a radix host machine where the
3144 	 * CPU threads may not be in different MMU modes.
3145 	 */
3146 	hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3147 		!kvm_is_radix(vc->kvm);
3148 	if (((controlled_threads > 1) &&
3149 	     ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3150 	    (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3151 		for_each_runnable_thread(i, vcpu, vc) {
3152 			vcpu->arch.ret = -EBUSY;
3153 			kvmppc_remove_runnable(vc, vcpu);
3154 			wake_up(&vcpu->arch.cpu_run);
3155 		}
3156 		goto out;
3157 	}
3158 
3159 	/*
3160 	 * See if we could run any other vcores on the physical core
3161 	 * along with this one.
3162 	 */
3163 	init_core_info(&core_info, vc);
3164 	pcpu = smp_processor_id();
3165 	target_threads = controlled_threads;
3166 	if (target_smt_mode && target_smt_mode < target_threads)
3167 		target_threads = target_smt_mode;
3168 	if (vc->num_threads < target_threads)
3169 		collect_piggybacks(&core_info, target_threads);
3170 
3171 	/*
3172 	 * On radix, arrange for TLB flushing if necessary.
3173 	 * This has to be done before disabling interrupts since
3174 	 * it uses smp_call_function().
3175 	 */
3176 	pcpu = smp_processor_id();
3177 	if (kvm_is_radix(vc->kvm)) {
3178 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3179 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3180 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3181 	}
3182 
3183 	/*
3184 	 * Hard-disable interrupts, and check resched flag and signals.
3185 	 * If we need to reschedule or deliver a signal, clean up
3186 	 * and return without going into the guest(s).
3187 	 * If the mmu_ready flag has been cleared, don't go into the
3188 	 * guest because that means a HPT resize operation is in progress.
3189 	 */
3190 	local_irq_disable();
3191 	hard_irq_disable();
3192 	if (lazy_irq_pending() || need_resched() ||
3193 	    recheck_signals_and_mmu(&core_info)) {
3194 		local_irq_enable();
3195 		vc->vcore_state = VCORE_INACTIVE;
3196 		/* Unlock all except the primary vcore */
3197 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3198 			pvc = core_info.vc[sub];
3199 			/* Put back on to the preempted vcores list */
3200 			kvmppc_vcore_preempt(pvc);
3201 			spin_unlock(&pvc->lock);
3202 		}
3203 		for (i = 0; i < controlled_threads; ++i)
3204 			kvmppc_release_hwthread(pcpu + i);
3205 		return;
3206 	}
3207 
3208 	kvmppc_clear_host_core(pcpu);
3209 
3210 	/* Decide on micro-threading (split-core) mode */
3211 	subcore_size = threads_per_subcore;
3212 	cmd_bit = stat_bit = 0;
3213 	split = core_info.n_subcores;
3214 	sip = NULL;
3215 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3216 		&& !cpu_has_feature(CPU_FTR_ARCH_300);
3217 
3218 	if (split > 1 || hpt_on_radix) {
3219 		sip = &split_info;
3220 		memset(&split_info, 0, sizeof(split_info));
3221 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3222 			split_info.vc[sub] = core_info.vc[sub];
3223 
3224 		if (is_power8) {
3225 			if (split == 2 && (dynamic_mt_modes & 2)) {
3226 				cmd_bit = HID0_POWER8_1TO2LPAR;
3227 				stat_bit = HID0_POWER8_2LPARMODE;
3228 			} else {
3229 				split = 4;
3230 				cmd_bit = HID0_POWER8_1TO4LPAR;
3231 				stat_bit = HID0_POWER8_4LPARMODE;
3232 			}
3233 			subcore_size = MAX_SMT_THREADS / split;
3234 			split_info.rpr = mfspr(SPRN_RPR);
3235 			split_info.pmmar = mfspr(SPRN_PMMAR);
3236 			split_info.ldbar = mfspr(SPRN_LDBAR);
3237 			split_info.subcore_size = subcore_size;
3238 		} else {
3239 			split_info.subcore_size = 1;
3240 			if (hpt_on_radix) {
3241 				/* Use the split_info for LPCR/LPIDR changes */
3242 				split_info.lpcr_req = vc->lpcr;
3243 				split_info.lpidr_req = vc->kvm->arch.lpid;
3244 				split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3245 				split_info.do_set = 1;
3246 			}
3247 		}
3248 
3249 		/* order writes to split_info before kvm_split_mode pointer */
3250 		smp_wmb();
3251 	}
3252 
3253 	for (thr = 0; thr < controlled_threads; ++thr) {
3254 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3255 
3256 		paca->kvm_hstate.tid = thr;
3257 		paca->kvm_hstate.napping = 0;
3258 		paca->kvm_hstate.kvm_split_mode = sip;
3259 	}
3260 
3261 	/* Initiate micro-threading (split-core) on POWER8 if required */
3262 	if (cmd_bit) {
3263 		unsigned long hid0 = mfspr(SPRN_HID0);
3264 
3265 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3266 		mb();
3267 		mtspr(SPRN_HID0, hid0);
3268 		isync();
3269 		for (;;) {
3270 			hid0 = mfspr(SPRN_HID0);
3271 			if (hid0 & stat_bit)
3272 				break;
3273 			cpu_relax();
3274 		}
3275 	}
3276 
3277 	/*
3278 	 * On POWER8, set RWMR register.
3279 	 * Since it only affects PURR and SPURR, it doesn't affect
3280 	 * the host, so we don't save/restore the host value.
3281 	 */
3282 	if (is_power8) {
3283 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3284 		int n_online = atomic_read(&vc->online_count);
3285 
3286 		/*
3287 		 * Use the 8-thread value if we're doing split-core
3288 		 * or if the vcore's online count looks bogus.
3289 		 */
3290 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3291 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3292 			rwmr_val = p8_rwmr_values[n_online];
3293 		mtspr(SPRN_RWMR, rwmr_val);
3294 	}
3295 
3296 	/* Start all the threads */
3297 	active = 0;
3298 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3299 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3300 		thr0_done = false;
3301 		active |= 1 << thr;
3302 		pvc = core_info.vc[sub];
3303 		pvc->pcpu = pcpu + thr;
3304 		for_each_runnable_thread(i, vcpu, pvc) {
3305 			kvmppc_start_thread(vcpu, pvc);
3306 			kvmppc_create_dtl_entry(vcpu, pvc);
3307 			trace_kvm_guest_enter(vcpu);
3308 			if (!vcpu->arch.ptid)
3309 				thr0_done = true;
3310 			active |= 1 << (thr + vcpu->arch.ptid);
3311 		}
3312 		/*
3313 		 * We need to start the first thread of each subcore
3314 		 * even if it doesn't have a vcpu.
3315 		 */
3316 		if (!thr0_done)
3317 			kvmppc_start_thread(NULL, pvc);
3318 	}
3319 
3320 	/*
3321 	 * Ensure that split_info.do_nap is set after setting
3322 	 * the vcore pointer in the PACA of the secondaries.
3323 	 */
3324 	smp_mb();
3325 
3326 	/*
3327 	 * When doing micro-threading, poke the inactive threads as well.
3328 	 * This gets them to the nap instruction after kvm_do_nap,
3329 	 * which reduces the time taken to unsplit later.
3330 	 * For POWER9 HPT guest on radix host, we need all the secondary
3331 	 * threads woken up so they can do the LPCR/LPIDR change.
3332 	 */
3333 	if (cmd_bit || hpt_on_radix) {
3334 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3335 		for (thr = 1; thr < threads_per_subcore; ++thr)
3336 			if (!(active & (1 << thr)))
3337 				kvmppc_ipi_thread(pcpu + thr);
3338 	}
3339 
3340 	vc->vcore_state = VCORE_RUNNING;
3341 	preempt_disable();
3342 
3343 	trace_kvmppc_run_core(vc, 0);
3344 
3345 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3346 		spin_unlock(&core_info.vc[sub]->lock);
3347 
3348 	guest_enter_irqoff();
3349 
3350 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3351 
3352 	this_cpu_disable_ftrace();
3353 
3354 	/*
3355 	 * Interrupts will be enabled once we get into the guest,
3356 	 * so tell lockdep that we're about to enable interrupts.
3357 	 */
3358 	trace_hardirqs_on();
3359 
3360 	trap = __kvmppc_vcore_entry();
3361 
3362 	trace_hardirqs_off();
3363 
3364 	this_cpu_enable_ftrace();
3365 
3366 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3367 
3368 	set_irq_happened(trap);
3369 
3370 	spin_lock(&vc->lock);
3371 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3372 	vc->vcore_state = VCORE_EXITING;
3373 
3374 	/* wait for secondary threads to finish writing their state to memory */
3375 	kvmppc_wait_for_nap(controlled_threads);
3376 
3377 	/* Return to whole-core mode if we split the core earlier */
3378 	if (cmd_bit) {
3379 		unsigned long hid0 = mfspr(SPRN_HID0);
3380 		unsigned long loops = 0;
3381 
3382 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3383 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3384 		mb();
3385 		mtspr(SPRN_HID0, hid0);
3386 		isync();
3387 		for (;;) {
3388 			hid0 = mfspr(SPRN_HID0);
3389 			if (!(hid0 & stat_bit))
3390 				break;
3391 			cpu_relax();
3392 			++loops;
3393 		}
3394 	} else if (hpt_on_radix) {
3395 		/* Wait for all threads to have seen final sync */
3396 		for (thr = 1; thr < controlled_threads; ++thr) {
3397 			struct paca_struct *paca = paca_ptrs[pcpu + thr];
3398 
3399 			while (paca->kvm_hstate.kvm_split_mode) {
3400 				HMT_low();
3401 				barrier();
3402 			}
3403 			HMT_medium();
3404 		}
3405 	}
3406 	split_info.do_nap = 0;
3407 
3408 	kvmppc_set_host_core(pcpu);
3409 
3410 	local_irq_enable();
3411 	guest_exit();
3412 
3413 	/* Let secondaries go back to the offline loop */
3414 	for (i = 0; i < controlled_threads; ++i) {
3415 		kvmppc_release_hwthread(pcpu + i);
3416 		if (sip && sip->napped[i])
3417 			kvmppc_ipi_thread(pcpu + i);
3418 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3419 	}
3420 
3421 	spin_unlock(&vc->lock);
3422 
3423 	/* make sure updates to secondary vcpu structs are visible now */
3424 	smp_mb();
3425 
3426 	preempt_enable();
3427 
3428 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3429 		pvc = core_info.vc[sub];
3430 		post_guest_process(pvc, pvc == vc);
3431 	}
3432 
3433 	spin_lock(&vc->lock);
3434 
3435  out:
3436 	vc->vcore_state = VCORE_INACTIVE;
3437 	trace_kvmppc_run_core(vc, 1);
3438 }
3439 
3440 /*
3441  * Load up hypervisor-mode registers on P9.
3442  */
3443 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3444 				     unsigned long lpcr)
3445 {
3446 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3447 	s64 hdec;
3448 	u64 tb, purr, spurr;
3449 	int trap;
3450 	unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3451 	unsigned long host_ciabr = mfspr(SPRN_CIABR);
3452 	unsigned long host_dawr = mfspr(SPRN_DAWR0);
3453 	unsigned long host_dawrx = mfspr(SPRN_DAWRX0);
3454 	unsigned long host_psscr = mfspr(SPRN_PSSCR);
3455 	unsigned long host_pidr = mfspr(SPRN_PID);
3456 
3457 	/*
3458 	 * P8 and P9 suppress the HDEC exception when LPCR[HDICE] = 0,
3459 	 * so set HDICE before writing HDEC.
3460 	 */
3461 	mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr | LPCR_HDICE);
3462 	isync();
3463 
3464 	hdec = time_limit - mftb();
3465 	if (hdec < 0) {
3466 		mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3467 		isync();
3468 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3469 	}
3470 	mtspr(SPRN_HDEC, hdec);
3471 
3472 	if (vc->tb_offset) {
3473 		u64 new_tb = mftb() + vc->tb_offset;
3474 		mtspr(SPRN_TBU40, new_tb);
3475 		tb = mftb();
3476 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3477 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3478 		vc->tb_offset_applied = vc->tb_offset;
3479 	}
3480 
3481 	if (vc->pcr)
3482 		mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3483 	mtspr(SPRN_DPDES, vc->dpdes);
3484 	mtspr(SPRN_VTB, vc->vtb);
3485 
3486 	local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3487 	local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3488 	mtspr(SPRN_PURR, vcpu->arch.purr);
3489 	mtspr(SPRN_SPURR, vcpu->arch.spurr);
3490 
3491 	if (dawr_enabled()) {
3492 		mtspr(SPRN_DAWR0, vcpu->arch.dawr);
3493 		mtspr(SPRN_DAWRX0, vcpu->arch.dawrx);
3494 	}
3495 	mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3496 	mtspr(SPRN_IC, vcpu->arch.ic);
3497 	mtspr(SPRN_PID, vcpu->arch.pid);
3498 
3499 	mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3500 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3501 
3502 	mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3503 
3504 	mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3505 	mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3506 	mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3507 	mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3508 
3509 	mtspr(SPRN_AMOR, ~0UL);
3510 
3511 	mtspr(SPRN_LPCR, lpcr);
3512 	isync();
3513 
3514 	kvmppc_xive_push_vcpu(vcpu);
3515 
3516 	mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3517 	mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3518 
3519 	trap = __kvmhv_vcpu_entry_p9(vcpu);
3520 
3521 	/* Advance host PURR/SPURR by the amount used by guest */
3522 	purr = mfspr(SPRN_PURR);
3523 	spurr = mfspr(SPRN_SPURR);
3524 	mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3525 	      purr - vcpu->arch.purr);
3526 	mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3527 	      spurr - vcpu->arch.spurr);
3528 	vcpu->arch.purr = purr;
3529 	vcpu->arch.spurr = spurr;
3530 
3531 	vcpu->arch.ic = mfspr(SPRN_IC);
3532 	vcpu->arch.pid = mfspr(SPRN_PID);
3533 	vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3534 
3535 	vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3536 	vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3537 	vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3538 	vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3539 
3540 	/* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3541 	mtspr(SPRN_PSSCR, host_psscr |
3542 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3543 	mtspr(SPRN_HFSCR, host_hfscr);
3544 	mtspr(SPRN_CIABR, host_ciabr);
3545 	mtspr(SPRN_DAWR0, host_dawr);
3546 	mtspr(SPRN_DAWRX0, host_dawrx);
3547 	mtspr(SPRN_PID, host_pidr);
3548 
3549 	/*
3550 	 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3551 	 * case we interrupted the guest between a tlbie and a ptesync.
3552 	 */
3553 	asm volatile("eieio; tlbsync; ptesync");
3554 
3555 	/*
3556 	 * cp_abort is required if the processor supports local copy-paste
3557 	 * to clear the copy buffer that was under control of the guest.
3558 	 */
3559 	if (cpu_has_feature(CPU_FTR_ARCH_31))
3560 		asm volatile(PPC_CP_ABORT);
3561 
3562 	mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);	/* restore host LPID */
3563 	isync();
3564 
3565 	vc->dpdes = mfspr(SPRN_DPDES);
3566 	vc->vtb = mfspr(SPRN_VTB);
3567 	mtspr(SPRN_DPDES, 0);
3568 	if (vc->pcr)
3569 		mtspr(SPRN_PCR, PCR_MASK);
3570 
3571 	if (vc->tb_offset_applied) {
3572 		u64 new_tb = mftb() - vc->tb_offset_applied;
3573 		mtspr(SPRN_TBU40, new_tb);
3574 		tb = mftb();
3575 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3576 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3577 		vc->tb_offset_applied = 0;
3578 	}
3579 
3580 	mtspr(SPRN_HDEC, 0x7fffffff);
3581 	mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3582 
3583 	return trap;
3584 }
3585 
3586 /*
3587  * Virtual-mode guest entry for POWER9 and later when the host and
3588  * guest are both using the radix MMU.  The LPIDR has already been set.
3589  */
3590 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3591 			 unsigned long lpcr)
3592 {
3593 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3594 	unsigned long host_dscr = mfspr(SPRN_DSCR);
3595 	unsigned long host_tidr = mfspr(SPRN_TIDR);
3596 	unsigned long host_iamr = mfspr(SPRN_IAMR);
3597 	unsigned long host_amr = mfspr(SPRN_AMR);
3598 	s64 dec;
3599 	u64 tb;
3600 	int trap, save_pmu;
3601 
3602 	dec = mfspr(SPRN_DEC);
3603 	tb = mftb();
3604 	if (dec < 0)
3605 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3606 	local_paca->kvm_hstate.dec_expires = dec + tb;
3607 	if (local_paca->kvm_hstate.dec_expires < time_limit)
3608 		time_limit = local_paca->kvm_hstate.dec_expires;
3609 
3610 	vcpu->arch.ceded = 0;
3611 
3612 	kvmhv_save_host_pmu();		/* saves it to PACA kvm_hstate */
3613 
3614 	kvmppc_subcore_enter_guest();
3615 
3616 	vc->entry_exit_map = 1;
3617 	vc->in_guest = 1;
3618 
3619 	if (vcpu->arch.vpa.pinned_addr) {
3620 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3621 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3622 		lp->yield_count = cpu_to_be32(yield_count);
3623 		vcpu->arch.vpa.dirty = 1;
3624 	}
3625 
3626 	if (cpu_has_feature(CPU_FTR_TM) ||
3627 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3628 		kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3629 
3630 	kvmhv_load_guest_pmu(vcpu);
3631 
3632 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3633 	load_fp_state(&vcpu->arch.fp);
3634 #ifdef CONFIG_ALTIVEC
3635 	load_vr_state(&vcpu->arch.vr);
3636 #endif
3637 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3638 
3639 	mtspr(SPRN_DSCR, vcpu->arch.dscr);
3640 	mtspr(SPRN_IAMR, vcpu->arch.iamr);
3641 	mtspr(SPRN_PSPB, vcpu->arch.pspb);
3642 	mtspr(SPRN_FSCR, vcpu->arch.fscr);
3643 	mtspr(SPRN_TAR, vcpu->arch.tar);
3644 	mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3645 	mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3646 	mtspr(SPRN_BESCR, vcpu->arch.bescr);
3647 	mtspr(SPRN_WORT, vcpu->arch.wort);
3648 	mtspr(SPRN_TIDR, vcpu->arch.tid);
3649 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3650 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3651 	mtspr(SPRN_AMR, vcpu->arch.amr);
3652 	mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3653 
3654 	if (!(vcpu->arch.ctrl & 1))
3655 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3656 
3657 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3658 
3659 	if (kvmhv_on_pseries()) {
3660 		/*
3661 		 * We need to save and restore the guest visible part of the
3662 		 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3663 		 * doesn't do this for us. Note only required if pseries since
3664 		 * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3665 		 */
3666 		unsigned long host_psscr;
3667 		/* call our hypervisor to load up HV regs and go */
3668 		struct hv_guest_state hvregs;
3669 
3670 		host_psscr = mfspr(SPRN_PSSCR_PR);
3671 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3672 		kvmhv_save_hv_regs(vcpu, &hvregs);
3673 		hvregs.lpcr = lpcr;
3674 		vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3675 		hvregs.version = HV_GUEST_STATE_VERSION;
3676 		if (vcpu->arch.nested) {
3677 			hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3678 			hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3679 		} else {
3680 			hvregs.lpid = vcpu->kvm->arch.lpid;
3681 			hvregs.vcpu_token = vcpu->vcpu_id;
3682 		}
3683 		hvregs.hdec_expiry = time_limit;
3684 		trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3685 					  __pa(&vcpu->arch.regs));
3686 		kvmhv_restore_hv_return_state(vcpu, &hvregs);
3687 		vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3688 		vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3689 		vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3690 		vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3691 		mtspr(SPRN_PSSCR_PR, host_psscr);
3692 
3693 		/* H_CEDE has to be handled now, not later */
3694 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3695 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3696 			kvmppc_nested_cede(vcpu);
3697 			kvmppc_set_gpr(vcpu, 3, 0);
3698 			trap = 0;
3699 		}
3700 	} else {
3701 		trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3702 	}
3703 
3704 	vcpu->arch.slb_max = 0;
3705 	dec = mfspr(SPRN_DEC);
3706 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3707 		dec = (s32) dec;
3708 	tb = mftb();
3709 	vcpu->arch.dec_expires = dec + tb;
3710 	vcpu->cpu = -1;
3711 	vcpu->arch.thread_cpu = -1;
3712 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3713 
3714 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
3715 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
3716 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
3717 	vcpu->arch.tar = mfspr(SPRN_TAR);
3718 	vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3719 	vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3720 	vcpu->arch.bescr = mfspr(SPRN_BESCR);
3721 	vcpu->arch.wort = mfspr(SPRN_WORT);
3722 	vcpu->arch.tid = mfspr(SPRN_TIDR);
3723 	vcpu->arch.amr = mfspr(SPRN_AMR);
3724 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3725 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
3726 
3727 	mtspr(SPRN_PSPB, 0);
3728 	mtspr(SPRN_WORT, 0);
3729 	mtspr(SPRN_UAMOR, 0);
3730 	mtspr(SPRN_DSCR, host_dscr);
3731 	mtspr(SPRN_TIDR, host_tidr);
3732 	mtspr(SPRN_IAMR, host_iamr);
3733 	mtspr(SPRN_PSPB, 0);
3734 
3735 	if (host_amr != vcpu->arch.amr)
3736 		mtspr(SPRN_AMR, host_amr);
3737 
3738 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3739 	store_fp_state(&vcpu->arch.fp);
3740 #ifdef CONFIG_ALTIVEC
3741 	store_vr_state(&vcpu->arch.vr);
3742 #endif
3743 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3744 
3745 	if (cpu_has_feature(CPU_FTR_TM) ||
3746 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3747 		kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3748 
3749 	save_pmu = 1;
3750 	if (vcpu->arch.vpa.pinned_addr) {
3751 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3752 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3753 		lp->yield_count = cpu_to_be32(yield_count);
3754 		vcpu->arch.vpa.dirty = 1;
3755 		save_pmu = lp->pmcregs_in_use;
3756 	}
3757 	/* Must save pmu if this guest is capable of running nested guests */
3758 	save_pmu |= nesting_enabled(vcpu->kvm);
3759 
3760 	kvmhv_save_guest_pmu(vcpu, save_pmu);
3761 
3762 	vc->entry_exit_map = 0x101;
3763 	vc->in_guest = 0;
3764 
3765 	mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3766 	mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3767 
3768 	kvmhv_load_host_pmu();
3769 
3770 	kvmppc_subcore_exit_guest();
3771 
3772 	return trap;
3773 }
3774 
3775 /*
3776  * Wait for some other vcpu thread to execute us, and
3777  * wake us up when we need to handle something in the host.
3778  */
3779 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3780 				 struct kvm_vcpu *vcpu, int wait_state)
3781 {
3782 	DEFINE_WAIT(wait);
3783 
3784 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3785 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3786 		spin_unlock(&vc->lock);
3787 		schedule();
3788 		spin_lock(&vc->lock);
3789 	}
3790 	finish_wait(&vcpu->arch.cpu_run, &wait);
3791 }
3792 
3793 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3794 {
3795 	if (!halt_poll_ns_grow)
3796 		return;
3797 
3798 	vc->halt_poll_ns *= halt_poll_ns_grow;
3799 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3800 		vc->halt_poll_ns = halt_poll_ns_grow_start;
3801 }
3802 
3803 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3804 {
3805 	if (halt_poll_ns_shrink == 0)
3806 		vc->halt_poll_ns = 0;
3807 	else
3808 		vc->halt_poll_ns /= halt_poll_ns_shrink;
3809 }
3810 
3811 #ifdef CONFIG_KVM_XICS
3812 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3813 {
3814 	if (!xics_on_xive())
3815 		return false;
3816 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3817 		vcpu->arch.xive_saved_state.cppr;
3818 }
3819 #else
3820 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3821 {
3822 	return false;
3823 }
3824 #endif /* CONFIG_KVM_XICS */
3825 
3826 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3827 {
3828 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3829 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3830 		return true;
3831 
3832 	return false;
3833 }
3834 
3835 /*
3836  * Check to see if any of the runnable vcpus on the vcore have pending
3837  * exceptions or are no longer ceded
3838  */
3839 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3840 {
3841 	struct kvm_vcpu *vcpu;
3842 	int i;
3843 
3844 	for_each_runnable_thread(i, vcpu, vc) {
3845 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3846 			return 1;
3847 	}
3848 
3849 	return 0;
3850 }
3851 
3852 /*
3853  * All the vcpus in this vcore are idle, so wait for a decrementer
3854  * or external interrupt to one of the vcpus.  vc->lock is held.
3855  */
3856 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3857 {
3858 	ktime_t cur, start_poll, start_wait;
3859 	int do_sleep = 1;
3860 	u64 block_ns;
3861 
3862 	/* Poll for pending exceptions and ceded state */
3863 	cur = start_poll = ktime_get();
3864 	if (vc->halt_poll_ns) {
3865 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3866 		++vc->runner->stat.halt_attempted_poll;
3867 
3868 		vc->vcore_state = VCORE_POLLING;
3869 		spin_unlock(&vc->lock);
3870 
3871 		do {
3872 			if (kvmppc_vcore_check_block(vc)) {
3873 				do_sleep = 0;
3874 				break;
3875 			}
3876 			cur = ktime_get();
3877 		} while (single_task_running() && ktime_before(cur, stop));
3878 
3879 		spin_lock(&vc->lock);
3880 		vc->vcore_state = VCORE_INACTIVE;
3881 
3882 		if (!do_sleep) {
3883 			++vc->runner->stat.halt_successful_poll;
3884 			goto out;
3885 		}
3886 	}
3887 
3888 	prepare_to_rcuwait(&vc->wait);
3889 	set_current_state(TASK_INTERRUPTIBLE);
3890 	if (kvmppc_vcore_check_block(vc)) {
3891 		finish_rcuwait(&vc->wait);
3892 		do_sleep = 0;
3893 		/* If we polled, count this as a successful poll */
3894 		if (vc->halt_poll_ns)
3895 			++vc->runner->stat.halt_successful_poll;
3896 		goto out;
3897 	}
3898 
3899 	start_wait = ktime_get();
3900 
3901 	vc->vcore_state = VCORE_SLEEPING;
3902 	trace_kvmppc_vcore_blocked(vc, 0);
3903 	spin_unlock(&vc->lock);
3904 	schedule();
3905 	finish_rcuwait(&vc->wait);
3906 	spin_lock(&vc->lock);
3907 	vc->vcore_state = VCORE_INACTIVE;
3908 	trace_kvmppc_vcore_blocked(vc, 1);
3909 	++vc->runner->stat.halt_successful_wait;
3910 
3911 	cur = ktime_get();
3912 
3913 out:
3914 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3915 
3916 	/* Attribute wait time */
3917 	if (do_sleep) {
3918 		vc->runner->stat.halt_wait_ns +=
3919 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3920 		/* Attribute failed poll time */
3921 		if (vc->halt_poll_ns)
3922 			vc->runner->stat.halt_poll_fail_ns +=
3923 				ktime_to_ns(start_wait) -
3924 				ktime_to_ns(start_poll);
3925 	} else {
3926 		/* Attribute successful poll time */
3927 		if (vc->halt_poll_ns)
3928 			vc->runner->stat.halt_poll_success_ns +=
3929 				ktime_to_ns(cur) -
3930 				ktime_to_ns(start_poll);
3931 	}
3932 
3933 	/* Adjust poll time */
3934 	if (halt_poll_ns) {
3935 		if (block_ns <= vc->halt_poll_ns)
3936 			;
3937 		/* We slept and blocked for longer than the max halt time */
3938 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3939 			shrink_halt_poll_ns(vc);
3940 		/* We slept and our poll time is too small */
3941 		else if (vc->halt_poll_ns < halt_poll_ns &&
3942 				block_ns < halt_poll_ns)
3943 			grow_halt_poll_ns(vc);
3944 		if (vc->halt_poll_ns > halt_poll_ns)
3945 			vc->halt_poll_ns = halt_poll_ns;
3946 	} else
3947 		vc->halt_poll_ns = 0;
3948 
3949 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3950 }
3951 
3952 /*
3953  * This never fails for a radix guest, as none of the operations it does
3954  * for a radix guest can fail or have a way to report failure.
3955  * kvmhv_run_single_vcpu() relies on this fact.
3956  */
3957 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3958 {
3959 	int r = 0;
3960 	struct kvm *kvm = vcpu->kvm;
3961 
3962 	mutex_lock(&kvm->arch.mmu_setup_lock);
3963 	if (!kvm->arch.mmu_ready) {
3964 		if (!kvm_is_radix(kvm))
3965 			r = kvmppc_hv_setup_htab_rma(vcpu);
3966 		if (!r) {
3967 			if (cpu_has_feature(CPU_FTR_ARCH_300))
3968 				kvmppc_setup_partition_table(kvm);
3969 			kvm->arch.mmu_ready = 1;
3970 		}
3971 	}
3972 	mutex_unlock(&kvm->arch.mmu_setup_lock);
3973 	return r;
3974 }
3975 
3976 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
3977 {
3978 	struct kvm_run *run = vcpu->run;
3979 	int n_ceded, i, r;
3980 	struct kvmppc_vcore *vc;
3981 	struct kvm_vcpu *v;
3982 
3983 	trace_kvmppc_run_vcpu_enter(vcpu);
3984 
3985 	run->exit_reason = 0;
3986 	vcpu->arch.ret = RESUME_GUEST;
3987 	vcpu->arch.trap = 0;
3988 	kvmppc_update_vpas(vcpu);
3989 
3990 	/*
3991 	 * Synchronize with other threads in this virtual core
3992 	 */
3993 	vc = vcpu->arch.vcore;
3994 	spin_lock(&vc->lock);
3995 	vcpu->arch.ceded = 0;
3996 	vcpu->arch.run_task = current;
3997 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3998 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3999 	vcpu->arch.busy_preempt = TB_NIL;
4000 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4001 	++vc->n_runnable;
4002 
4003 	/*
4004 	 * This happens the first time this is called for a vcpu.
4005 	 * If the vcore is already running, we may be able to start
4006 	 * this thread straight away and have it join in.
4007 	 */
4008 	if (!signal_pending(current)) {
4009 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
4010 		     vc->vcore_state == VCORE_RUNNING) &&
4011 			   !VCORE_IS_EXITING(vc)) {
4012 			kvmppc_create_dtl_entry(vcpu, vc);
4013 			kvmppc_start_thread(vcpu, vc);
4014 			trace_kvm_guest_enter(vcpu);
4015 		} else if (vc->vcore_state == VCORE_SLEEPING) {
4016 		        rcuwait_wake_up(&vc->wait);
4017 		}
4018 
4019 	}
4020 
4021 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4022 	       !signal_pending(current)) {
4023 		/* See if the MMU is ready to go */
4024 		if (!vcpu->kvm->arch.mmu_ready) {
4025 			spin_unlock(&vc->lock);
4026 			r = kvmhv_setup_mmu(vcpu);
4027 			spin_lock(&vc->lock);
4028 			if (r) {
4029 				run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4030 				run->fail_entry.
4031 					hardware_entry_failure_reason = 0;
4032 				vcpu->arch.ret = r;
4033 				break;
4034 			}
4035 		}
4036 
4037 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4038 			kvmppc_vcore_end_preempt(vc);
4039 
4040 		if (vc->vcore_state != VCORE_INACTIVE) {
4041 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4042 			continue;
4043 		}
4044 		for_each_runnable_thread(i, v, vc) {
4045 			kvmppc_core_prepare_to_enter(v);
4046 			if (signal_pending(v->arch.run_task)) {
4047 				kvmppc_remove_runnable(vc, v);
4048 				v->stat.signal_exits++;
4049 				v->run->exit_reason = KVM_EXIT_INTR;
4050 				v->arch.ret = -EINTR;
4051 				wake_up(&v->arch.cpu_run);
4052 			}
4053 		}
4054 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4055 			break;
4056 		n_ceded = 0;
4057 		for_each_runnable_thread(i, v, vc) {
4058 			if (!kvmppc_vcpu_woken(v))
4059 				n_ceded += v->arch.ceded;
4060 			else
4061 				v->arch.ceded = 0;
4062 		}
4063 		vc->runner = vcpu;
4064 		if (n_ceded == vc->n_runnable) {
4065 			kvmppc_vcore_blocked(vc);
4066 		} else if (need_resched()) {
4067 			kvmppc_vcore_preempt(vc);
4068 			/* Let something else run */
4069 			cond_resched_lock(&vc->lock);
4070 			if (vc->vcore_state == VCORE_PREEMPT)
4071 				kvmppc_vcore_end_preempt(vc);
4072 		} else {
4073 			kvmppc_run_core(vc);
4074 		}
4075 		vc->runner = NULL;
4076 	}
4077 
4078 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4079 	       (vc->vcore_state == VCORE_RUNNING ||
4080 		vc->vcore_state == VCORE_EXITING ||
4081 		vc->vcore_state == VCORE_PIGGYBACK))
4082 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4083 
4084 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4085 		kvmppc_vcore_end_preempt(vc);
4086 
4087 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4088 		kvmppc_remove_runnable(vc, vcpu);
4089 		vcpu->stat.signal_exits++;
4090 		run->exit_reason = KVM_EXIT_INTR;
4091 		vcpu->arch.ret = -EINTR;
4092 	}
4093 
4094 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4095 		/* Wake up some vcpu to run the core */
4096 		i = -1;
4097 		v = next_runnable_thread(vc, &i);
4098 		wake_up(&v->arch.cpu_run);
4099 	}
4100 
4101 	trace_kvmppc_run_vcpu_exit(vcpu);
4102 	spin_unlock(&vc->lock);
4103 	return vcpu->arch.ret;
4104 }
4105 
4106 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4107 			  unsigned long lpcr)
4108 {
4109 	struct kvm_run *run = vcpu->run;
4110 	int trap, r, pcpu;
4111 	int srcu_idx, lpid;
4112 	struct kvmppc_vcore *vc;
4113 	struct kvm *kvm = vcpu->kvm;
4114 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4115 
4116 	trace_kvmppc_run_vcpu_enter(vcpu);
4117 
4118 	run->exit_reason = 0;
4119 	vcpu->arch.ret = RESUME_GUEST;
4120 	vcpu->arch.trap = 0;
4121 
4122 	vc = vcpu->arch.vcore;
4123 	vcpu->arch.ceded = 0;
4124 	vcpu->arch.run_task = current;
4125 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4126 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4127 	vcpu->arch.busy_preempt = TB_NIL;
4128 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4129 	vc->runnable_threads[0] = vcpu;
4130 	vc->n_runnable = 1;
4131 	vc->runner = vcpu;
4132 
4133 	/* See if the MMU is ready to go */
4134 	if (!kvm->arch.mmu_ready)
4135 		kvmhv_setup_mmu(vcpu);
4136 
4137 	if (need_resched())
4138 		cond_resched();
4139 
4140 	kvmppc_update_vpas(vcpu);
4141 
4142 	init_vcore_to_run(vc);
4143 	vc->preempt_tb = TB_NIL;
4144 
4145 	preempt_disable();
4146 	pcpu = smp_processor_id();
4147 	vc->pcpu = pcpu;
4148 	kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4149 
4150 	local_irq_disable();
4151 	hard_irq_disable();
4152 	if (signal_pending(current))
4153 		goto sigpend;
4154 	if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4155 		goto out;
4156 
4157 	if (!nested) {
4158 		kvmppc_core_prepare_to_enter(vcpu);
4159 		if (vcpu->arch.doorbell_request) {
4160 			vc->dpdes = 1;
4161 			smp_wmb();
4162 			vcpu->arch.doorbell_request = 0;
4163 		}
4164 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4165 			     &vcpu->arch.pending_exceptions))
4166 			lpcr |= LPCR_MER;
4167 	} else if (vcpu->arch.pending_exceptions ||
4168 		   vcpu->arch.doorbell_request ||
4169 		   xive_interrupt_pending(vcpu)) {
4170 		vcpu->arch.ret = RESUME_HOST;
4171 		goto out;
4172 	}
4173 
4174 	kvmppc_clear_host_core(pcpu);
4175 
4176 	local_paca->kvm_hstate.tid = 0;
4177 	local_paca->kvm_hstate.napping = 0;
4178 	local_paca->kvm_hstate.kvm_split_mode = NULL;
4179 	kvmppc_start_thread(vcpu, vc);
4180 	kvmppc_create_dtl_entry(vcpu, vc);
4181 	trace_kvm_guest_enter(vcpu);
4182 
4183 	vc->vcore_state = VCORE_RUNNING;
4184 	trace_kvmppc_run_core(vc, 0);
4185 
4186 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4187 		lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4188 		mtspr(SPRN_LPID, lpid);
4189 		isync();
4190 		kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4191 	}
4192 
4193 	guest_enter_irqoff();
4194 
4195 	srcu_idx = srcu_read_lock(&kvm->srcu);
4196 
4197 	this_cpu_disable_ftrace();
4198 
4199 	/* Tell lockdep that we're about to enable interrupts */
4200 	trace_hardirqs_on();
4201 
4202 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4203 	vcpu->arch.trap = trap;
4204 
4205 	trace_hardirqs_off();
4206 
4207 	this_cpu_enable_ftrace();
4208 
4209 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4210 
4211 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4212 		mtspr(SPRN_LPID, kvm->arch.host_lpid);
4213 		isync();
4214 	}
4215 
4216 	set_irq_happened(trap);
4217 
4218 	kvmppc_set_host_core(pcpu);
4219 
4220 	local_irq_enable();
4221 	guest_exit();
4222 
4223 	cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4224 
4225 	preempt_enable();
4226 
4227 	/*
4228 	 * cancel pending decrementer exception if DEC is now positive, or if
4229 	 * entering a nested guest in which case the decrementer is now owned
4230 	 * by L2 and the L1 decrementer is provided in hdec_expires
4231 	 */
4232 	if (kvmppc_core_pending_dec(vcpu) &&
4233 			((get_tb() < vcpu->arch.dec_expires) ||
4234 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4235 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4236 		kvmppc_core_dequeue_dec(vcpu);
4237 
4238 	trace_kvm_guest_exit(vcpu);
4239 	r = RESUME_GUEST;
4240 	if (trap) {
4241 		if (!nested)
4242 			r = kvmppc_handle_exit_hv(vcpu, current);
4243 		else
4244 			r = kvmppc_handle_nested_exit(vcpu);
4245 	}
4246 	vcpu->arch.ret = r;
4247 
4248 	if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4249 	    !kvmppc_vcpu_woken(vcpu)) {
4250 		kvmppc_set_timer(vcpu);
4251 		while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4252 			if (signal_pending(current)) {
4253 				vcpu->stat.signal_exits++;
4254 				run->exit_reason = KVM_EXIT_INTR;
4255 				vcpu->arch.ret = -EINTR;
4256 				break;
4257 			}
4258 			spin_lock(&vc->lock);
4259 			kvmppc_vcore_blocked(vc);
4260 			spin_unlock(&vc->lock);
4261 		}
4262 	}
4263 	vcpu->arch.ceded = 0;
4264 
4265 	vc->vcore_state = VCORE_INACTIVE;
4266 	trace_kvmppc_run_core(vc, 1);
4267 
4268  done:
4269 	kvmppc_remove_runnable(vc, vcpu);
4270 	trace_kvmppc_run_vcpu_exit(vcpu);
4271 
4272 	return vcpu->arch.ret;
4273 
4274  sigpend:
4275 	vcpu->stat.signal_exits++;
4276 	run->exit_reason = KVM_EXIT_INTR;
4277 	vcpu->arch.ret = -EINTR;
4278  out:
4279 	local_irq_enable();
4280 	preempt_enable();
4281 	goto done;
4282 }
4283 
4284 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4285 {
4286 	struct kvm_run *run = vcpu->run;
4287 	int r;
4288 	int srcu_idx;
4289 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
4290 	unsigned long user_tar = 0;
4291 	unsigned int user_vrsave;
4292 	struct kvm *kvm;
4293 
4294 	if (!vcpu->arch.sane) {
4295 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4296 		return -EINVAL;
4297 	}
4298 
4299 	/*
4300 	 * Don't allow entry with a suspended transaction, because
4301 	 * the guest entry/exit code will lose it.
4302 	 * If the guest has TM enabled, save away their TM-related SPRs
4303 	 * (they will get restored by the TM unavailable interrupt).
4304 	 */
4305 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4306 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4307 	    (current->thread.regs->msr & MSR_TM)) {
4308 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4309 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4310 			run->fail_entry.hardware_entry_failure_reason = 0;
4311 			return -EINVAL;
4312 		}
4313 		/* Enable TM so we can read the TM SPRs */
4314 		mtmsr(mfmsr() | MSR_TM);
4315 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4316 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4317 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4318 		current->thread.regs->msr &= ~MSR_TM;
4319 	}
4320 #endif
4321 
4322 	/*
4323 	 * Force online to 1 for the sake of old userspace which doesn't
4324 	 * set it.
4325 	 */
4326 	if (!vcpu->arch.online) {
4327 		atomic_inc(&vcpu->arch.vcore->online_count);
4328 		vcpu->arch.online = 1;
4329 	}
4330 
4331 	kvmppc_core_prepare_to_enter(vcpu);
4332 
4333 	/* No need to go into the guest when all we'll do is come back out */
4334 	if (signal_pending(current)) {
4335 		run->exit_reason = KVM_EXIT_INTR;
4336 		return -EINTR;
4337 	}
4338 
4339 	kvm = vcpu->kvm;
4340 	atomic_inc(&kvm->arch.vcpus_running);
4341 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4342 	smp_mb();
4343 
4344 	flush_all_to_thread(current);
4345 
4346 	/* Save userspace EBB and other register values */
4347 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4348 		ebb_regs[0] = mfspr(SPRN_EBBHR);
4349 		ebb_regs[1] = mfspr(SPRN_EBBRR);
4350 		ebb_regs[2] = mfspr(SPRN_BESCR);
4351 		user_tar = mfspr(SPRN_TAR);
4352 	}
4353 	user_vrsave = mfspr(SPRN_VRSAVE);
4354 
4355 	vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4356 	vcpu->arch.pgdir = kvm->mm->pgd;
4357 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4358 
4359 	do {
4360 		/*
4361 		 * The early POWER9 chips that can't mix radix and HPT threads
4362 		 * on the same core also need the workaround for the problem
4363 		 * where the TLB would prefetch entries in the guest exit path
4364 		 * for radix guests using the guest PIDR value and LPID 0.
4365 		 * The workaround is in the old path (kvmppc_run_vcpu())
4366 		 * but not the new path (kvmhv_run_single_vcpu()).
4367 		 */
4368 		if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4369 		    !no_mixing_hpt_and_radix)
4370 			r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4371 						  vcpu->arch.vcore->lpcr);
4372 		else
4373 			r = kvmppc_run_vcpu(vcpu);
4374 
4375 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4376 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4377 			trace_kvm_hcall_enter(vcpu);
4378 			r = kvmppc_pseries_do_hcall(vcpu);
4379 			trace_kvm_hcall_exit(vcpu, r);
4380 			kvmppc_core_prepare_to_enter(vcpu);
4381 		} else if (r == RESUME_PAGE_FAULT) {
4382 			srcu_idx = srcu_read_lock(&kvm->srcu);
4383 			r = kvmppc_book3s_hv_page_fault(vcpu,
4384 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4385 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4386 		} else if (r == RESUME_PASSTHROUGH) {
4387 			if (WARN_ON(xics_on_xive()))
4388 				r = H_SUCCESS;
4389 			else
4390 				r = kvmppc_xics_rm_complete(vcpu, 0);
4391 		}
4392 	} while (is_kvmppc_resume_guest(r));
4393 
4394 	/* Restore userspace EBB and other register values */
4395 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4396 		mtspr(SPRN_EBBHR, ebb_regs[0]);
4397 		mtspr(SPRN_EBBRR, ebb_regs[1]);
4398 		mtspr(SPRN_BESCR, ebb_regs[2]);
4399 		mtspr(SPRN_TAR, user_tar);
4400 		mtspr(SPRN_FSCR, current->thread.fscr);
4401 	}
4402 	mtspr(SPRN_VRSAVE, user_vrsave);
4403 
4404 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4405 	atomic_dec(&kvm->arch.vcpus_running);
4406 	return r;
4407 }
4408 
4409 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4410 				     int shift, int sllp)
4411 {
4412 	(*sps)->page_shift = shift;
4413 	(*sps)->slb_enc = sllp;
4414 	(*sps)->enc[0].page_shift = shift;
4415 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4416 	/*
4417 	 * Add 16MB MPSS support (may get filtered out by userspace)
4418 	 */
4419 	if (shift != 24) {
4420 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4421 		if (penc != -1) {
4422 			(*sps)->enc[1].page_shift = 24;
4423 			(*sps)->enc[1].pte_enc = penc;
4424 		}
4425 	}
4426 	(*sps)++;
4427 }
4428 
4429 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4430 					 struct kvm_ppc_smmu_info *info)
4431 {
4432 	struct kvm_ppc_one_seg_page_size *sps;
4433 
4434 	/*
4435 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4436 	 * POWER7 doesn't support keys for instruction accesses,
4437 	 * POWER8 and POWER9 do.
4438 	 */
4439 	info->data_keys = 32;
4440 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4441 
4442 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4443 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4444 	info->slb_size = 32;
4445 
4446 	/* We only support these sizes for now, and no muti-size segments */
4447 	sps = &info->sps[0];
4448 	kvmppc_add_seg_page_size(&sps, 12, 0);
4449 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4450 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4451 
4452 	/* If running as a nested hypervisor, we don't support HPT guests */
4453 	if (kvmhv_on_pseries())
4454 		info->flags |= KVM_PPC_NO_HASH;
4455 
4456 	return 0;
4457 }
4458 
4459 /*
4460  * Get (and clear) the dirty memory log for a memory slot.
4461  */
4462 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4463 					 struct kvm_dirty_log *log)
4464 {
4465 	struct kvm_memslots *slots;
4466 	struct kvm_memory_slot *memslot;
4467 	int i, r;
4468 	unsigned long n;
4469 	unsigned long *buf, *p;
4470 	struct kvm_vcpu *vcpu;
4471 
4472 	mutex_lock(&kvm->slots_lock);
4473 
4474 	r = -EINVAL;
4475 	if (log->slot >= KVM_USER_MEM_SLOTS)
4476 		goto out;
4477 
4478 	slots = kvm_memslots(kvm);
4479 	memslot = id_to_memslot(slots, log->slot);
4480 	r = -ENOENT;
4481 	if (!memslot || !memslot->dirty_bitmap)
4482 		goto out;
4483 
4484 	/*
4485 	 * Use second half of bitmap area because both HPT and radix
4486 	 * accumulate bits in the first half.
4487 	 */
4488 	n = kvm_dirty_bitmap_bytes(memslot);
4489 	buf = memslot->dirty_bitmap + n / sizeof(long);
4490 	memset(buf, 0, n);
4491 
4492 	if (kvm_is_radix(kvm))
4493 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4494 	else
4495 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4496 	if (r)
4497 		goto out;
4498 
4499 	/*
4500 	 * We accumulate dirty bits in the first half of the
4501 	 * memslot's dirty_bitmap area, for when pages are paged
4502 	 * out or modified by the host directly.  Pick up these
4503 	 * bits and add them to the map.
4504 	 */
4505 	p = memslot->dirty_bitmap;
4506 	for (i = 0; i < n / sizeof(long); ++i)
4507 		buf[i] |= xchg(&p[i], 0);
4508 
4509 	/* Harvest dirty bits from VPA and DTL updates */
4510 	/* Note: we never modify the SLB shadow buffer areas */
4511 	kvm_for_each_vcpu(i, vcpu, kvm) {
4512 		spin_lock(&vcpu->arch.vpa_update_lock);
4513 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4514 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4515 		spin_unlock(&vcpu->arch.vpa_update_lock);
4516 	}
4517 
4518 	r = -EFAULT;
4519 	if (copy_to_user(log->dirty_bitmap, buf, n))
4520 		goto out;
4521 
4522 	r = 0;
4523 out:
4524 	mutex_unlock(&kvm->slots_lock);
4525 	return r;
4526 }
4527 
4528 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4529 {
4530 	vfree(slot->arch.rmap);
4531 	slot->arch.rmap = NULL;
4532 }
4533 
4534 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4535 					struct kvm_memory_slot *slot,
4536 					const struct kvm_userspace_memory_region *mem,
4537 					enum kvm_mr_change change)
4538 {
4539 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4540 
4541 	if (change == KVM_MR_CREATE) {
4542 		slot->arch.rmap = vzalloc(array_size(npages,
4543 					  sizeof(*slot->arch.rmap)));
4544 		if (!slot->arch.rmap)
4545 			return -ENOMEM;
4546 	}
4547 
4548 	return 0;
4549 }
4550 
4551 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4552 				const struct kvm_userspace_memory_region *mem,
4553 				const struct kvm_memory_slot *old,
4554 				const struct kvm_memory_slot *new,
4555 				enum kvm_mr_change change)
4556 {
4557 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4558 
4559 	/*
4560 	 * If we are making a new memslot, it might make
4561 	 * some address that was previously cached as emulated
4562 	 * MMIO be no longer emulated MMIO, so invalidate
4563 	 * all the caches of emulated MMIO translations.
4564 	 */
4565 	if (npages)
4566 		atomic64_inc(&kvm->arch.mmio_update);
4567 
4568 	/*
4569 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4570 	 * have already called kvm_arch_flush_shadow_memslot() to
4571 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
4572 	 * previous mappings.  So the only case to handle is
4573 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4574 	 * has been changed.
4575 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4576 	 * to get rid of any THP PTEs in the partition-scoped page tables
4577 	 * so we can track dirtiness at the page level; we flush when
4578 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4579 	 * using THP PTEs.
4580 	 */
4581 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4582 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4583 		kvmppc_radix_flush_memslot(kvm, old);
4584 	/*
4585 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4586 	 */
4587 	if (!kvm->arch.secure_guest)
4588 		return;
4589 
4590 	switch (change) {
4591 	case KVM_MR_CREATE:
4592 		/*
4593 		 * @TODO kvmppc_uvmem_memslot_create() can fail and
4594 		 * return error. Fix this.
4595 		 */
4596 		kvmppc_uvmem_memslot_create(kvm, new);
4597 		break;
4598 	case KVM_MR_DELETE:
4599 		kvmppc_uvmem_memslot_delete(kvm, old);
4600 		break;
4601 	default:
4602 		/* TODO: Handle KVM_MR_MOVE */
4603 		break;
4604 	}
4605 }
4606 
4607 /*
4608  * Update LPCR values in kvm->arch and in vcores.
4609  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4610  * of kvm->arch.lpcr update).
4611  */
4612 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4613 {
4614 	long int i;
4615 	u32 cores_done = 0;
4616 
4617 	if ((kvm->arch.lpcr & mask) == lpcr)
4618 		return;
4619 
4620 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4621 
4622 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4623 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4624 		if (!vc)
4625 			continue;
4626 		spin_lock(&vc->lock);
4627 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4628 		spin_unlock(&vc->lock);
4629 		if (++cores_done >= kvm->arch.online_vcores)
4630 			break;
4631 	}
4632 }
4633 
4634 void kvmppc_setup_partition_table(struct kvm *kvm)
4635 {
4636 	unsigned long dw0, dw1;
4637 
4638 	if (!kvm_is_radix(kvm)) {
4639 		/* PS field - page size for VRMA */
4640 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4641 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4642 		/* HTABSIZE and HTABORG fields */
4643 		dw0 |= kvm->arch.sdr1;
4644 
4645 		/* Second dword as set by userspace */
4646 		dw1 = kvm->arch.process_table;
4647 	} else {
4648 		dw0 = PATB_HR | radix__get_tree_size() |
4649 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4650 		dw1 = PATB_GR | kvm->arch.process_table;
4651 	}
4652 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4653 }
4654 
4655 /*
4656  * Set up HPT (hashed page table) and RMA (real-mode area).
4657  * Must be called with kvm->arch.mmu_setup_lock held.
4658  */
4659 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4660 {
4661 	int err = 0;
4662 	struct kvm *kvm = vcpu->kvm;
4663 	unsigned long hva;
4664 	struct kvm_memory_slot *memslot;
4665 	struct vm_area_struct *vma;
4666 	unsigned long lpcr = 0, senc;
4667 	unsigned long psize, porder;
4668 	int srcu_idx;
4669 
4670 	/* Allocate hashed page table (if not done already) and reset it */
4671 	if (!kvm->arch.hpt.virt) {
4672 		int order = KVM_DEFAULT_HPT_ORDER;
4673 		struct kvm_hpt_info info;
4674 
4675 		err = kvmppc_allocate_hpt(&info, order);
4676 		/* If we get here, it means userspace didn't specify a
4677 		 * size explicitly.  So, try successively smaller
4678 		 * sizes if the default failed. */
4679 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4680 			err  = kvmppc_allocate_hpt(&info, order);
4681 
4682 		if (err < 0) {
4683 			pr_err("KVM: Couldn't alloc HPT\n");
4684 			goto out;
4685 		}
4686 
4687 		kvmppc_set_hpt(kvm, &info);
4688 	}
4689 
4690 	/* Look up the memslot for guest physical address 0 */
4691 	srcu_idx = srcu_read_lock(&kvm->srcu);
4692 	memslot = gfn_to_memslot(kvm, 0);
4693 
4694 	/* We must have some memory at 0 by now */
4695 	err = -EINVAL;
4696 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4697 		goto out_srcu;
4698 
4699 	/* Look up the VMA for the start of this memory slot */
4700 	hva = memslot->userspace_addr;
4701 	mmap_read_lock(kvm->mm);
4702 	vma = find_vma(kvm->mm, hva);
4703 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4704 		goto up_out;
4705 
4706 	psize = vma_kernel_pagesize(vma);
4707 
4708 	mmap_read_unlock(kvm->mm);
4709 
4710 	/* We can handle 4k, 64k or 16M pages in the VRMA */
4711 	if (psize >= 0x1000000)
4712 		psize = 0x1000000;
4713 	else if (psize >= 0x10000)
4714 		psize = 0x10000;
4715 	else
4716 		psize = 0x1000;
4717 	porder = __ilog2(psize);
4718 
4719 	senc = slb_pgsize_encoding(psize);
4720 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4721 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4722 	/* Create HPTEs in the hash page table for the VRMA */
4723 	kvmppc_map_vrma(vcpu, memslot, porder);
4724 
4725 	/* Update VRMASD field in the LPCR */
4726 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4727 		/* the -4 is to account for senc values starting at 0x10 */
4728 		lpcr = senc << (LPCR_VRMASD_SH - 4);
4729 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4730 	}
4731 
4732 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4733 	smp_wmb();
4734 	err = 0;
4735  out_srcu:
4736 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4737  out:
4738 	return err;
4739 
4740  up_out:
4741 	mmap_read_unlock(kvm->mm);
4742 	goto out_srcu;
4743 }
4744 
4745 /*
4746  * Must be called with kvm->arch.mmu_setup_lock held and
4747  * mmu_ready = 0 and no vcpus running.
4748  */
4749 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4750 {
4751 	if (nesting_enabled(kvm))
4752 		kvmhv_release_all_nested(kvm);
4753 	kvmppc_rmap_reset(kvm);
4754 	kvm->arch.process_table = 0;
4755 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4756 	spin_lock(&kvm->mmu_lock);
4757 	kvm->arch.radix = 0;
4758 	spin_unlock(&kvm->mmu_lock);
4759 	kvmppc_free_radix(kvm);
4760 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
4761 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4762 	return 0;
4763 }
4764 
4765 /*
4766  * Must be called with kvm->arch.mmu_setup_lock held and
4767  * mmu_ready = 0 and no vcpus running.
4768  */
4769 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4770 {
4771 	int err;
4772 
4773 	err = kvmppc_init_vm_radix(kvm);
4774 	if (err)
4775 		return err;
4776 	kvmppc_rmap_reset(kvm);
4777 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4778 	spin_lock(&kvm->mmu_lock);
4779 	kvm->arch.radix = 1;
4780 	spin_unlock(&kvm->mmu_lock);
4781 	kvmppc_free_hpt(&kvm->arch.hpt);
4782 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4783 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4784 	return 0;
4785 }
4786 
4787 #ifdef CONFIG_KVM_XICS
4788 /*
4789  * Allocate a per-core structure for managing state about which cores are
4790  * running in the host versus the guest and for exchanging data between
4791  * real mode KVM and CPU running in the host.
4792  * This is only done for the first VM.
4793  * The allocated structure stays even if all VMs have stopped.
4794  * It is only freed when the kvm-hv module is unloaded.
4795  * It's OK for this routine to fail, we just don't support host
4796  * core operations like redirecting H_IPI wakeups.
4797  */
4798 void kvmppc_alloc_host_rm_ops(void)
4799 {
4800 	struct kvmppc_host_rm_ops *ops;
4801 	unsigned long l_ops;
4802 	int cpu, core;
4803 	int size;
4804 
4805 	/* Not the first time here ? */
4806 	if (kvmppc_host_rm_ops_hv != NULL)
4807 		return;
4808 
4809 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4810 	if (!ops)
4811 		return;
4812 
4813 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4814 	ops->rm_core = kzalloc(size, GFP_KERNEL);
4815 
4816 	if (!ops->rm_core) {
4817 		kfree(ops);
4818 		return;
4819 	}
4820 
4821 	cpus_read_lock();
4822 
4823 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4824 		if (!cpu_online(cpu))
4825 			continue;
4826 
4827 		core = cpu >> threads_shift;
4828 		ops->rm_core[core].rm_state.in_host = 1;
4829 	}
4830 
4831 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4832 
4833 	/*
4834 	 * Make the contents of the kvmppc_host_rm_ops structure visible
4835 	 * to other CPUs before we assign it to the global variable.
4836 	 * Do an atomic assignment (no locks used here), but if someone
4837 	 * beats us to it, just free our copy and return.
4838 	 */
4839 	smp_wmb();
4840 	l_ops = (unsigned long) ops;
4841 
4842 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4843 		cpus_read_unlock();
4844 		kfree(ops->rm_core);
4845 		kfree(ops);
4846 		return;
4847 	}
4848 
4849 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4850 					     "ppc/kvm_book3s:prepare",
4851 					     kvmppc_set_host_core,
4852 					     kvmppc_clear_host_core);
4853 	cpus_read_unlock();
4854 }
4855 
4856 void kvmppc_free_host_rm_ops(void)
4857 {
4858 	if (kvmppc_host_rm_ops_hv) {
4859 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4860 		kfree(kvmppc_host_rm_ops_hv->rm_core);
4861 		kfree(kvmppc_host_rm_ops_hv);
4862 		kvmppc_host_rm_ops_hv = NULL;
4863 	}
4864 }
4865 #endif
4866 
4867 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4868 {
4869 	unsigned long lpcr, lpid;
4870 	char buf[32];
4871 	int ret;
4872 
4873 	mutex_init(&kvm->arch.uvmem_lock);
4874 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4875 	mutex_init(&kvm->arch.mmu_setup_lock);
4876 
4877 	/* Allocate the guest's logical partition ID */
4878 
4879 	lpid = kvmppc_alloc_lpid();
4880 	if ((long)lpid < 0)
4881 		return -ENOMEM;
4882 	kvm->arch.lpid = lpid;
4883 
4884 	kvmppc_alloc_host_rm_ops();
4885 
4886 	kvmhv_vm_nested_init(kvm);
4887 
4888 	/*
4889 	 * Since we don't flush the TLB when tearing down a VM,
4890 	 * and this lpid might have previously been used,
4891 	 * make sure we flush on each core before running the new VM.
4892 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4893 	 * does this flush for us.
4894 	 */
4895 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4896 		cpumask_setall(&kvm->arch.need_tlb_flush);
4897 
4898 	/* Start out with the default set of hcalls enabled */
4899 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4900 	       sizeof(kvm->arch.enabled_hcalls));
4901 
4902 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4903 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4904 
4905 	/* Init LPCR for virtual RMA mode */
4906 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4907 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
4908 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4909 		lpcr &= LPCR_PECE | LPCR_LPES;
4910 	} else {
4911 		lpcr = 0;
4912 	}
4913 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4914 		LPCR_VPM0 | LPCR_VPM1;
4915 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4916 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4917 	/* On POWER8 turn on online bit to enable PURR/SPURR */
4918 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
4919 		lpcr |= LPCR_ONL;
4920 	/*
4921 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4922 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
4923 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4924 	 * be unnecessary but better safe than sorry in case we re-enable
4925 	 * EE in HV mode with this LPCR still set)
4926 	 */
4927 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4928 		lpcr &= ~LPCR_VPM0;
4929 		lpcr |= LPCR_HVICE | LPCR_HEIC;
4930 
4931 		/*
4932 		 * If xive is enabled, we route 0x500 interrupts directly
4933 		 * to the guest.
4934 		 */
4935 		if (xics_on_xive())
4936 			lpcr |= LPCR_LPES;
4937 	}
4938 
4939 	/*
4940 	 * If the host uses radix, the guest starts out as radix.
4941 	 */
4942 	if (radix_enabled()) {
4943 		kvm->arch.radix = 1;
4944 		kvm->arch.mmu_ready = 1;
4945 		lpcr &= ~LPCR_VPM1;
4946 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4947 		ret = kvmppc_init_vm_radix(kvm);
4948 		if (ret) {
4949 			kvmppc_free_lpid(kvm->arch.lpid);
4950 			return ret;
4951 		}
4952 		kvmppc_setup_partition_table(kvm);
4953 	}
4954 
4955 	kvm->arch.lpcr = lpcr;
4956 
4957 	/* Initialization for future HPT resizes */
4958 	kvm->arch.resize_hpt = NULL;
4959 
4960 	/*
4961 	 * Work out how many sets the TLB has, for the use of
4962 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4963 	 */
4964 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
4965 		/*
4966 		 * P10 will flush all the congruence class with a single tlbiel
4967 		 */
4968 		kvm->arch.tlb_sets = 1;
4969 	} else if (radix_enabled())
4970 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
4971 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
4972 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
4973 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4974 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
4975 	else
4976 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
4977 
4978 	/*
4979 	 * Track that we now have a HV mode VM active. This blocks secondary
4980 	 * CPU threads from coming online.
4981 	 * On POWER9, we only need to do this if the "indep_threads_mode"
4982 	 * module parameter has been set to N.
4983 	 */
4984 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4985 		if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4986 			pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4987 			kvm->arch.threads_indep = true;
4988 		} else {
4989 			kvm->arch.threads_indep = indep_threads_mode;
4990 		}
4991 	}
4992 	if (!kvm->arch.threads_indep)
4993 		kvm_hv_vm_activated();
4994 
4995 	/*
4996 	 * Initialize smt_mode depending on processor.
4997 	 * POWER8 and earlier have to use "strict" threading, where
4998 	 * all vCPUs in a vcore have to run on the same (sub)core,
4999 	 * whereas on POWER9 the threads can each run a different
5000 	 * guest.
5001 	 */
5002 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5003 		kvm->arch.smt_mode = threads_per_subcore;
5004 	else
5005 		kvm->arch.smt_mode = 1;
5006 	kvm->arch.emul_smt_mode = 1;
5007 
5008 	/*
5009 	 * Create a debugfs directory for the VM
5010 	 */
5011 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
5012 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5013 	kvmppc_mmu_debugfs_init(kvm);
5014 	if (radix_enabled())
5015 		kvmhv_radix_debugfs_init(kvm);
5016 
5017 	return 0;
5018 }
5019 
5020 static void kvmppc_free_vcores(struct kvm *kvm)
5021 {
5022 	long int i;
5023 
5024 	for (i = 0; i < KVM_MAX_VCORES; ++i)
5025 		kfree(kvm->arch.vcores[i]);
5026 	kvm->arch.online_vcores = 0;
5027 }
5028 
5029 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5030 {
5031 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
5032 
5033 	if (!kvm->arch.threads_indep)
5034 		kvm_hv_vm_deactivated();
5035 
5036 	kvmppc_free_vcores(kvm);
5037 
5038 
5039 	if (kvm_is_radix(kvm))
5040 		kvmppc_free_radix(kvm);
5041 	else
5042 		kvmppc_free_hpt(&kvm->arch.hpt);
5043 
5044 	/* Perform global invalidation and return lpid to the pool */
5045 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5046 		if (nesting_enabled(kvm))
5047 			kvmhv_release_all_nested(kvm);
5048 		kvm->arch.process_table = 0;
5049 		if (kvm->arch.secure_guest)
5050 			uv_svm_terminate(kvm->arch.lpid);
5051 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5052 	}
5053 
5054 	kvmppc_free_lpid(kvm->arch.lpid);
5055 
5056 	kvmppc_free_pimap(kvm);
5057 }
5058 
5059 /* We don't need to emulate any privileged instructions or dcbz */
5060 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5061 				     unsigned int inst, int *advance)
5062 {
5063 	return EMULATE_FAIL;
5064 }
5065 
5066 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5067 					ulong spr_val)
5068 {
5069 	return EMULATE_FAIL;
5070 }
5071 
5072 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5073 					ulong *spr_val)
5074 {
5075 	return EMULATE_FAIL;
5076 }
5077 
5078 static int kvmppc_core_check_processor_compat_hv(void)
5079 {
5080 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5081 	    cpu_has_feature(CPU_FTR_ARCH_206))
5082 		return 0;
5083 
5084 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5085 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5086 		return 0;
5087 
5088 	return -EIO;
5089 }
5090 
5091 #ifdef CONFIG_KVM_XICS
5092 
5093 void kvmppc_free_pimap(struct kvm *kvm)
5094 {
5095 	kfree(kvm->arch.pimap);
5096 }
5097 
5098 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5099 {
5100 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5101 }
5102 
5103 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5104 {
5105 	struct irq_desc *desc;
5106 	struct kvmppc_irq_map *irq_map;
5107 	struct kvmppc_passthru_irqmap *pimap;
5108 	struct irq_chip *chip;
5109 	int i, rc = 0;
5110 
5111 	if (!kvm_irq_bypass)
5112 		return 1;
5113 
5114 	desc = irq_to_desc(host_irq);
5115 	if (!desc)
5116 		return -EIO;
5117 
5118 	mutex_lock(&kvm->lock);
5119 
5120 	pimap = kvm->arch.pimap;
5121 	if (pimap == NULL) {
5122 		/* First call, allocate structure to hold IRQ map */
5123 		pimap = kvmppc_alloc_pimap();
5124 		if (pimap == NULL) {
5125 			mutex_unlock(&kvm->lock);
5126 			return -ENOMEM;
5127 		}
5128 		kvm->arch.pimap = pimap;
5129 	}
5130 
5131 	/*
5132 	 * For now, we only support interrupts for which the EOI operation
5133 	 * is an OPAL call followed by a write to XIRR, since that's
5134 	 * what our real-mode EOI code does, or a XIVE interrupt
5135 	 */
5136 	chip = irq_data_get_irq_chip(&desc->irq_data);
5137 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5138 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5139 			host_irq, guest_gsi);
5140 		mutex_unlock(&kvm->lock);
5141 		return -ENOENT;
5142 	}
5143 
5144 	/*
5145 	 * See if we already have an entry for this guest IRQ number.
5146 	 * If it's mapped to a hardware IRQ number, that's an error,
5147 	 * otherwise re-use this entry.
5148 	 */
5149 	for (i = 0; i < pimap->n_mapped; i++) {
5150 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5151 			if (pimap->mapped[i].r_hwirq) {
5152 				mutex_unlock(&kvm->lock);
5153 				return -EINVAL;
5154 			}
5155 			break;
5156 		}
5157 	}
5158 
5159 	if (i == KVMPPC_PIRQ_MAPPED) {
5160 		mutex_unlock(&kvm->lock);
5161 		return -EAGAIN;		/* table is full */
5162 	}
5163 
5164 	irq_map = &pimap->mapped[i];
5165 
5166 	irq_map->v_hwirq = guest_gsi;
5167 	irq_map->desc = desc;
5168 
5169 	/*
5170 	 * Order the above two stores before the next to serialize with
5171 	 * the KVM real mode handler.
5172 	 */
5173 	smp_wmb();
5174 	irq_map->r_hwirq = desc->irq_data.hwirq;
5175 
5176 	if (i == pimap->n_mapped)
5177 		pimap->n_mapped++;
5178 
5179 	if (xics_on_xive())
5180 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5181 	else
5182 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5183 	if (rc)
5184 		irq_map->r_hwirq = 0;
5185 
5186 	mutex_unlock(&kvm->lock);
5187 
5188 	return 0;
5189 }
5190 
5191 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5192 {
5193 	struct irq_desc *desc;
5194 	struct kvmppc_passthru_irqmap *pimap;
5195 	int i, rc = 0;
5196 
5197 	if (!kvm_irq_bypass)
5198 		return 0;
5199 
5200 	desc = irq_to_desc(host_irq);
5201 	if (!desc)
5202 		return -EIO;
5203 
5204 	mutex_lock(&kvm->lock);
5205 	if (!kvm->arch.pimap)
5206 		goto unlock;
5207 
5208 	pimap = kvm->arch.pimap;
5209 
5210 	for (i = 0; i < pimap->n_mapped; i++) {
5211 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5212 			break;
5213 	}
5214 
5215 	if (i == pimap->n_mapped) {
5216 		mutex_unlock(&kvm->lock);
5217 		return -ENODEV;
5218 	}
5219 
5220 	if (xics_on_xive())
5221 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5222 	else
5223 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5224 
5225 	/* invalidate the entry (what do do on error from the above ?) */
5226 	pimap->mapped[i].r_hwirq = 0;
5227 
5228 	/*
5229 	 * We don't free this structure even when the count goes to
5230 	 * zero. The structure is freed when we destroy the VM.
5231 	 */
5232  unlock:
5233 	mutex_unlock(&kvm->lock);
5234 	return rc;
5235 }
5236 
5237 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5238 					     struct irq_bypass_producer *prod)
5239 {
5240 	int ret = 0;
5241 	struct kvm_kernel_irqfd *irqfd =
5242 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5243 
5244 	irqfd->producer = prod;
5245 
5246 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5247 	if (ret)
5248 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5249 			prod->irq, irqfd->gsi, ret);
5250 
5251 	return ret;
5252 }
5253 
5254 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5255 					      struct irq_bypass_producer *prod)
5256 {
5257 	int ret;
5258 	struct kvm_kernel_irqfd *irqfd =
5259 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5260 
5261 	irqfd->producer = NULL;
5262 
5263 	/*
5264 	 * When producer of consumer is unregistered, we change back to
5265 	 * default external interrupt handling mode - KVM real mode
5266 	 * will switch back to host.
5267 	 */
5268 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5269 	if (ret)
5270 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5271 			prod->irq, irqfd->gsi, ret);
5272 }
5273 #endif
5274 
5275 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5276 				 unsigned int ioctl, unsigned long arg)
5277 {
5278 	struct kvm *kvm __maybe_unused = filp->private_data;
5279 	void __user *argp = (void __user *)arg;
5280 	long r;
5281 
5282 	switch (ioctl) {
5283 
5284 	case KVM_PPC_ALLOCATE_HTAB: {
5285 		u32 htab_order;
5286 
5287 		/* If we're a nested hypervisor, we currently only support radix */
5288 		if (kvmhv_on_pseries()) {
5289 			r = -EOPNOTSUPP;
5290 			break;
5291 		}
5292 
5293 		r = -EFAULT;
5294 		if (get_user(htab_order, (u32 __user *)argp))
5295 			break;
5296 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5297 		if (r)
5298 			break;
5299 		r = 0;
5300 		break;
5301 	}
5302 
5303 	case KVM_PPC_GET_HTAB_FD: {
5304 		struct kvm_get_htab_fd ghf;
5305 
5306 		r = -EFAULT;
5307 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5308 			break;
5309 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5310 		break;
5311 	}
5312 
5313 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5314 		struct kvm_ppc_resize_hpt rhpt;
5315 
5316 		r = -EFAULT;
5317 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5318 			break;
5319 
5320 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5321 		break;
5322 	}
5323 
5324 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5325 		struct kvm_ppc_resize_hpt rhpt;
5326 
5327 		r = -EFAULT;
5328 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5329 			break;
5330 
5331 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5332 		break;
5333 	}
5334 
5335 	default:
5336 		r = -ENOTTY;
5337 	}
5338 
5339 	return r;
5340 }
5341 
5342 /*
5343  * List of hcall numbers to enable by default.
5344  * For compatibility with old userspace, we enable by default
5345  * all hcalls that were implemented before the hcall-enabling
5346  * facility was added.  Note this list should not include H_RTAS.
5347  */
5348 static unsigned int default_hcall_list[] = {
5349 	H_REMOVE,
5350 	H_ENTER,
5351 	H_READ,
5352 	H_PROTECT,
5353 	H_BULK_REMOVE,
5354 	H_GET_TCE,
5355 	H_PUT_TCE,
5356 	H_SET_DABR,
5357 	H_SET_XDABR,
5358 	H_CEDE,
5359 	H_PROD,
5360 	H_CONFER,
5361 	H_REGISTER_VPA,
5362 #ifdef CONFIG_KVM_XICS
5363 	H_EOI,
5364 	H_CPPR,
5365 	H_IPI,
5366 	H_IPOLL,
5367 	H_XIRR,
5368 	H_XIRR_X,
5369 #endif
5370 	0
5371 };
5372 
5373 static void init_default_hcalls(void)
5374 {
5375 	int i;
5376 	unsigned int hcall;
5377 
5378 	for (i = 0; default_hcall_list[i]; ++i) {
5379 		hcall = default_hcall_list[i];
5380 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5381 		__set_bit(hcall / 4, default_enabled_hcalls);
5382 	}
5383 }
5384 
5385 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5386 {
5387 	unsigned long lpcr;
5388 	int radix;
5389 	int err;
5390 
5391 	/* If not on a POWER9, reject it */
5392 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5393 		return -ENODEV;
5394 
5395 	/* If any unknown flags set, reject it */
5396 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5397 		return -EINVAL;
5398 
5399 	/* GR (guest radix) bit in process_table field must match */
5400 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5401 	if (!!(cfg->process_table & PATB_GR) != radix)
5402 		return -EINVAL;
5403 
5404 	/* Process table size field must be reasonable, i.e. <= 24 */
5405 	if ((cfg->process_table & PRTS_MASK) > 24)
5406 		return -EINVAL;
5407 
5408 	/* We can change a guest to/from radix now, if the host is radix */
5409 	if (radix && !radix_enabled())
5410 		return -EINVAL;
5411 
5412 	/* If we're a nested hypervisor, we currently only support radix */
5413 	if (kvmhv_on_pseries() && !radix)
5414 		return -EINVAL;
5415 
5416 	mutex_lock(&kvm->arch.mmu_setup_lock);
5417 	if (radix != kvm_is_radix(kvm)) {
5418 		if (kvm->arch.mmu_ready) {
5419 			kvm->arch.mmu_ready = 0;
5420 			/* order mmu_ready vs. vcpus_running */
5421 			smp_mb();
5422 			if (atomic_read(&kvm->arch.vcpus_running)) {
5423 				kvm->arch.mmu_ready = 1;
5424 				err = -EBUSY;
5425 				goto out_unlock;
5426 			}
5427 		}
5428 		if (radix)
5429 			err = kvmppc_switch_mmu_to_radix(kvm);
5430 		else
5431 			err = kvmppc_switch_mmu_to_hpt(kvm);
5432 		if (err)
5433 			goto out_unlock;
5434 	}
5435 
5436 	kvm->arch.process_table = cfg->process_table;
5437 	kvmppc_setup_partition_table(kvm);
5438 
5439 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5440 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5441 	err = 0;
5442 
5443  out_unlock:
5444 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5445 	return err;
5446 }
5447 
5448 static int kvmhv_enable_nested(struct kvm *kvm)
5449 {
5450 	if (!nested)
5451 		return -EPERM;
5452 	if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5453 		return -ENODEV;
5454 
5455 	/* kvm == NULL means the caller is testing if the capability exists */
5456 	if (kvm)
5457 		kvm->arch.nested_enable = true;
5458 	return 0;
5459 }
5460 
5461 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5462 				 int size)
5463 {
5464 	int rc = -EINVAL;
5465 
5466 	if (kvmhv_vcpu_is_radix(vcpu)) {
5467 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5468 
5469 		if (rc > 0)
5470 			rc = -EINVAL;
5471 	}
5472 
5473 	/* For now quadrants are the only way to access nested guest memory */
5474 	if (rc && vcpu->arch.nested)
5475 		rc = -EAGAIN;
5476 
5477 	return rc;
5478 }
5479 
5480 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5481 				int size)
5482 {
5483 	int rc = -EINVAL;
5484 
5485 	if (kvmhv_vcpu_is_radix(vcpu)) {
5486 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5487 
5488 		if (rc > 0)
5489 			rc = -EINVAL;
5490 	}
5491 
5492 	/* For now quadrants are the only way to access nested guest memory */
5493 	if (rc && vcpu->arch.nested)
5494 		rc = -EAGAIN;
5495 
5496 	return rc;
5497 }
5498 
5499 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5500 {
5501 	unpin_vpa(kvm, vpa);
5502 	vpa->gpa = 0;
5503 	vpa->pinned_addr = NULL;
5504 	vpa->dirty = false;
5505 	vpa->update_pending = 0;
5506 }
5507 
5508 /*
5509  * Enable a guest to become a secure VM, or test whether
5510  * that could be enabled.
5511  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5512  * tested (kvm == NULL) or enabled (kvm != NULL).
5513  */
5514 static int kvmhv_enable_svm(struct kvm *kvm)
5515 {
5516 	if (!kvmppc_uvmem_available())
5517 		return -EINVAL;
5518 	if (kvm)
5519 		kvm->arch.svm_enabled = 1;
5520 	return 0;
5521 }
5522 
5523 /*
5524  *  IOCTL handler to turn off secure mode of guest
5525  *
5526  * - Release all device pages
5527  * - Issue ucall to terminate the guest on the UV side
5528  * - Unpin the VPA pages.
5529  * - Reinit the partition scoped page tables
5530  */
5531 static int kvmhv_svm_off(struct kvm *kvm)
5532 {
5533 	struct kvm_vcpu *vcpu;
5534 	int mmu_was_ready;
5535 	int srcu_idx;
5536 	int ret = 0;
5537 	int i;
5538 
5539 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5540 		return ret;
5541 
5542 	mutex_lock(&kvm->arch.mmu_setup_lock);
5543 	mmu_was_ready = kvm->arch.mmu_ready;
5544 	if (kvm->arch.mmu_ready) {
5545 		kvm->arch.mmu_ready = 0;
5546 		/* order mmu_ready vs. vcpus_running */
5547 		smp_mb();
5548 		if (atomic_read(&kvm->arch.vcpus_running)) {
5549 			kvm->arch.mmu_ready = 1;
5550 			ret = -EBUSY;
5551 			goto out;
5552 		}
5553 	}
5554 
5555 	srcu_idx = srcu_read_lock(&kvm->srcu);
5556 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5557 		struct kvm_memory_slot *memslot;
5558 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5559 
5560 		if (!slots)
5561 			continue;
5562 
5563 		kvm_for_each_memslot(memslot, slots) {
5564 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
5565 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5566 		}
5567 	}
5568 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5569 
5570 	ret = uv_svm_terminate(kvm->arch.lpid);
5571 	if (ret != U_SUCCESS) {
5572 		ret = -EINVAL;
5573 		goto out;
5574 	}
5575 
5576 	/*
5577 	 * When secure guest is reset, all the guest pages are sent
5578 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
5579 	 * chance to run and unpin their VPA pages. Unpinning of all
5580 	 * VPA pages is done here explicitly so that VPA pages
5581 	 * can be migrated to the secure side.
5582 	 *
5583 	 * This is required to for the secure SMP guest to reboot
5584 	 * correctly.
5585 	 */
5586 	kvm_for_each_vcpu(i, vcpu, kvm) {
5587 		spin_lock(&vcpu->arch.vpa_update_lock);
5588 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5589 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5590 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5591 		spin_unlock(&vcpu->arch.vpa_update_lock);
5592 	}
5593 
5594 	kvmppc_setup_partition_table(kvm);
5595 	kvm->arch.secure_guest = 0;
5596 	kvm->arch.mmu_ready = mmu_was_ready;
5597 out:
5598 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5599 	return ret;
5600 }
5601 
5602 static struct kvmppc_ops kvm_ops_hv = {
5603 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5604 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5605 	.get_one_reg = kvmppc_get_one_reg_hv,
5606 	.set_one_reg = kvmppc_set_one_reg_hv,
5607 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
5608 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
5609 	.inject_interrupt = kvmppc_inject_interrupt_hv,
5610 	.set_msr     = kvmppc_set_msr_hv,
5611 	.vcpu_run    = kvmppc_vcpu_run_hv,
5612 	.vcpu_create = kvmppc_core_vcpu_create_hv,
5613 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
5614 	.check_requests = kvmppc_core_check_requests_hv,
5615 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5616 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
5617 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5618 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5619 	.unmap_hva_range = kvm_unmap_hva_range_hv,
5620 	.age_hva  = kvm_age_hva_hv,
5621 	.test_age_hva = kvm_test_age_hva_hv,
5622 	.set_spte_hva = kvm_set_spte_hva_hv,
5623 	.free_memslot = kvmppc_core_free_memslot_hv,
5624 	.init_vm =  kvmppc_core_init_vm_hv,
5625 	.destroy_vm = kvmppc_core_destroy_vm_hv,
5626 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5627 	.emulate_op = kvmppc_core_emulate_op_hv,
5628 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5629 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5630 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5631 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5632 	.hcall_implemented = kvmppc_hcall_impl_hv,
5633 #ifdef CONFIG_KVM_XICS
5634 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5635 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5636 #endif
5637 	.configure_mmu = kvmhv_configure_mmu,
5638 	.get_rmmu_info = kvmhv_get_rmmu_info,
5639 	.set_smt_mode = kvmhv_set_smt_mode,
5640 	.enable_nested = kvmhv_enable_nested,
5641 	.load_from_eaddr = kvmhv_load_from_eaddr,
5642 	.store_to_eaddr = kvmhv_store_to_eaddr,
5643 	.enable_svm = kvmhv_enable_svm,
5644 	.svm_off = kvmhv_svm_off,
5645 };
5646 
5647 static int kvm_init_subcore_bitmap(void)
5648 {
5649 	int i, j;
5650 	int nr_cores = cpu_nr_cores();
5651 	struct sibling_subcore_state *sibling_subcore_state;
5652 
5653 	for (i = 0; i < nr_cores; i++) {
5654 		int first_cpu = i * threads_per_core;
5655 		int node = cpu_to_node(first_cpu);
5656 
5657 		/* Ignore if it is already allocated. */
5658 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
5659 			continue;
5660 
5661 		sibling_subcore_state =
5662 			kzalloc_node(sizeof(struct sibling_subcore_state),
5663 							GFP_KERNEL, node);
5664 		if (!sibling_subcore_state)
5665 			return -ENOMEM;
5666 
5667 
5668 		for (j = 0; j < threads_per_core; j++) {
5669 			int cpu = first_cpu + j;
5670 
5671 			paca_ptrs[cpu]->sibling_subcore_state =
5672 						sibling_subcore_state;
5673 		}
5674 	}
5675 	return 0;
5676 }
5677 
5678 static int kvmppc_radix_possible(void)
5679 {
5680 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5681 }
5682 
5683 static int kvmppc_book3s_init_hv(void)
5684 {
5685 	int r;
5686 
5687 	if (!tlbie_capable) {
5688 		pr_err("KVM-HV: Host does not support TLBIE\n");
5689 		return -ENODEV;
5690 	}
5691 
5692 	/*
5693 	 * FIXME!! Do we need to check on all cpus ?
5694 	 */
5695 	r = kvmppc_core_check_processor_compat_hv();
5696 	if (r < 0)
5697 		return -ENODEV;
5698 
5699 	r = kvmhv_nested_init();
5700 	if (r)
5701 		return r;
5702 
5703 	r = kvm_init_subcore_bitmap();
5704 	if (r)
5705 		return r;
5706 
5707 	/*
5708 	 * We need a way of accessing the XICS interrupt controller,
5709 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5710 	 * indirectly, via OPAL.
5711 	 */
5712 #ifdef CONFIG_SMP
5713 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
5714 	    !local_paca->kvm_hstate.xics_phys) {
5715 		struct device_node *np;
5716 
5717 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5718 		if (!np) {
5719 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5720 			return -ENODEV;
5721 		}
5722 		/* presence of intc confirmed - node can be dropped again */
5723 		of_node_put(np);
5724 	}
5725 #endif
5726 
5727 	kvm_ops_hv.owner = THIS_MODULE;
5728 	kvmppc_hv_ops = &kvm_ops_hv;
5729 
5730 	init_default_hcalls();
5731 
5732 	init_vcore_lists();
5733 
5734 	r = kvmppc_mmu_hv_init();
5735 	if (r)
5736 		return r;
5737 
5738 	if (kvmppc_radix_possible())
5739 		r = kvmppc_radix_init();
5740 
5741 	/*
5742 	 * POWER9 chips before version 2.02 can't have some threads in
5743 	 * HPT mode and some in radix mode on the same core.
5744 	 */
5745 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5746 		unsigned int pvr = mfspr(SPRN_PVR);
5747 		if ((pvr >> 16) == PVR_POWER9 &&
5748 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5749 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5750 			no_mixing_hpt_and_radix = true;
5751 	}
5752 
5753 	r = kvmppc_uvmem_init();
5754 	if (r < 0)
5755 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5756 
5757 	return r;
5758 }
5759 
5760 static void kvmppc_book3s_exit_hv(void)
5761 {
5762 	kvmppc_uvmem_free();
5763 	kvmppc_free_host_rm_ops();
5764 	if (kvmppc_radix_possible())
5765 		kvmppc_radix_exit();
5766 	kvmppc_hv_ops = NULL;
5767 	kvmhv_nested_exit();
5768 }
5769 
5770 module_init(kvmppc_book3s_init_hv);
5771 module_exit(kvmppc_book3s_exit_hv);
5772 MODULE_LICENSE("GPL");
5773 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5774 MODULE_ALIAS("devname:kvm");
5775