xref: /linux/arch/powerpc/kvm/book3s_hv.c (revision ce615f5c1f73537c8267035d58b3d0c70e19b8da)
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 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 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
347 {
348 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
349 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
350 
351 	/* We can (emulate) our own architecture version and anything older */
352 	if (cpu_has_feature(CPU_FTR_ARCH_300))
353 		host_pcr_bit = PCR_ARCH_300;
354 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
355 		host_pcr_bit = PCR_ARCH_207;
356 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
357 		host_pcr_bit = PCR_ARCH_206;
358 	else
359 		host_pcr_bit = PCR_ARCH_205;
360 
361 	/* Determine lowest PCR bit needed to run guest in given PVR level */
362 	guest_pcr_bit = host_pcr_bit;
363 	if (arch_compat) {
364 		switch (arch_compat) {
365 		case PVR_ARCH_205:
366 			guest_pcr_bit = PCR_ARCH_205;
367 			break;
368 		case PVR_ARCH_206:
369 		case PVR_ARCH_206p:
370 			guest_pcr_bit = PCR_ARCH_206;
371 			break;
372 		case PVR_ARCH_207:
373 			guest_pcr_bit = PCR_ARCH_207;
374 			break;
375 		case PVR_ARCH_300:
376 			guest_pcr_bit = PCR_ARCH_300;
377 			break;
378 		default:
379 			return -EINVAL;
380 		}
381 	}
382 
383 	/* Check requested PCR bits don't exceed our capabilities */
384 	if (guest_pcr_bit > host_pcr_bit)
385 		return -EINVAL;
386 
387 	spin_lock(&vc->lock);
388 	vc->arch_compat = arch_compat;
389 	/*
390 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
391 	 * Also set all reserved PCR bits
392 	 */
393 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
394 	spin_unlock(&vc->lock);
395 
396 	return 0;
397 }
398 
399 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
400 {
401 	int r;
402 
403 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
404 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
405 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
406 	for (r = 0; r < 16; ++r)
407 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
408 		       r, kvmppc_get_gpr(vcpu, r),
409 		       r+16, kvmppc_get_gpr(vcpu, r+16));
410 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
411 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
412 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
413 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
414 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
415 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
416 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
417 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
418 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
419 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
420 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
421 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
422 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
423 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
424 	for (r = 0; r < vcpu->arch.slb_max; ++r)
425 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
426 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
427 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
428 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
429 	       vcpu->arch.last_inst);
430 }
431 
432 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
433 {
434 	return kvm_get_vcpu_by_id(kvm, id);
435 }
436 
437 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
438 {
439 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
440 	vpa->yield_count = cpu_to_be32(1);
441 }
442 
443 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
444 		   unsigned long addr, unsigned long len)
445 {
446 	/* check address is cacheline aligned */
447 	if (addr & (L1_CACHE_BYTES - 1))
448 		return -EINVAL;
449 	spin_lock(&vcpu->arch.vpa_update_lock);
450 	if (v->next_gpa != addr || v->len != len) {
451 		v->next_gpa = addr;
452 		v->len = addr ? len : 0;
453 		v->update_pending = 1;
454 	}
455 	spin_unlock(&vcpu->arch.vpa_update_lock);
456 	return 0;
457 }
458 
459 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
460 struct reg_vpa {
461 	u32 dummy;
462 	union {
463 		__be16 hword;
464 		__be32 word;
465 	} length;
466 };
467 
468 static int vpa_is_registered(struct kvmppc_vpa *vpap)
469 {
470 	if (vpap->update_pending)
471 		return vpap->next_gpa != 0;
472 	return vpap->pinned_addr != NULL;
473 }
474 
475 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
476 				       unsigned long flags,
477 				       unsigned long vcpuid, unsigned long vpa)
478 {
479 	struct kvm *kvm = vcpu->kvm;
480 	unsigned long len, nb;
481 	void *va;
482 	struct kvm_vcpu *tvcpu;
483 	int err;
484 	int subfunc;
485 	struct kvmppc_vpa *vpap;
486 
487 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
488 	if (!tvcpu)
489 		return H_PARAMETER;
490 
491 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
492 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
493 	    subfunc == H_VPA_REG_SLB) {
494 		/* Registering new area - address must be cache-line aligned */
495 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
496 			return H_PARAMETER;
497 
498 		/* convert logical addr to kernel addr and read length */
499 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
500 		if (va == NULL)
501 			return H_PARAMETER;
502 		if (subfunc == H_VPA_REG_VPA)
503 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
504 		else
505 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
506 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
507 
508 		/* Check length */
509 		if (len > nb || len < sizeof(struct reg_vpa))
510 			return H_PARAMETER;
511 	} else {
512 		vpa = 0;
513 		len = 0;
514 	}
515 
516 	err = H_PARAMETER;
517 	vpap = NULL;
518 	spin_lock(&tvcpu->arch.vpa_update_lock);
519 
520 	switch (subfunc) {
521 	case H_VPA_REG_VPA:		/* register VPA */
522 		/*
523 		 * The size of our lppaca is 1kB because of the way we align
524 		 * it for the guest to avoid crossing a 4kB boundary. We only
525 		 * use 640 bytes of the structure though, so we should accept
526 		 * clients that set a size of 640.
527 		 */
528 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
529 		if (len < sizeof(struct lppaca))
530 			break;
531 		vpap = &tvcpu->arch.vpa;
532 		err = 0;
533 		break;
534 
535 	case H_VPA_REG_DTL:		/* register DTL */
536 		if (len < sizeof(struct dtl_entry))
537 			break;
538 		len -= len % sizeof(struct dtl_entry);
539 
540 		/* Check that they have previously registered a VPA */
541 		err = H_RESOURCE;
542 		if (!vpa_is_registered(&tvcpu->arch.vpa))
543 			break;
544 
545 		vpap = &tvcpu->arch.dtl;
546 		err = 0;
547 		break;
548 
549 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
550 		/* Check that they have previously registered a VPA */
551 		err = H_RESOURCE;
552 		if (!vpa_is_registered(&tvcpu->arch.vpa))
553 			break;
554 
555 		vpap = &tvcpu->arch.slb_shadow;
556 		err = 0;
557 		break;
558 
559 	case H_VPA_DEREG_VPA:		/* deregister VPA */
560 		/* Check they don't still have a DTL or SLB buf registered */
561 		err = H_RESOURCE;
562 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
563 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
564 			break;
565 
566 		vpap = &tvcpu->arch.vpa;
567 		err = 0;
568 		break;
569 
570 	case H_VPA_DEREG_DTL:		/* deregister DTL */
571 		vpap = &tvcpu->arch.dtl;
572 		err = 0;
573 		break;
574 
575 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
576 		vpap = &tvcpu->arch.slb_shadow;
577 		err = 0;
578 		break;
579 	}
580 
581 	if (vpap) {
582 		vpap->next_gpa = vpa;
583 		vpap->len = len;
584 		vpap->update_pending = 1;
585 	}
586 
587 	spin_unlock(&tvcpu->arch.vpa_update_lock);
588 
589 	return err;
590 }
591 
592 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
593 {
594 	struct kvm *kvm = vcpu->kvm;
595 	void *va;
596 	unsigned long nb;
597 	unsigned long gpa;
598 
599 	/*
600 	 * We need to pin the page pointed to by vpap->next_gpa,
601 	 * but we can't call kvmppc_pin_guest_page under the lock
602 	 * as it does get_user_pages() and down_read().  So we
603 	 * have to drop the lock, pin the page, then get the lock
604 	 * again and check that a new area didn't get registered
605 	 * in the meantime.
606 	 */
607 	for (;;) {
608 		gpa = vpap->next_gpa;
609 		spin_unlock(&vcpu->arch.vpa_update_lock);
610 		va = NULL;
611 		nb = 0;
612 		if (gpa)
613 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
614 		spin_lock(&vcpu->arch.vpa_update_lock);
615 		if (gpa == vpap->next_gpa)
616 			break;
617 		/* sigh... unpin that one and try again */
618 		if (va)
619 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
620 	}
621 
622 	vpap->update_pending = 0;
623 	if (va && nb < vpap->len) {
624 		/*
625 		 * If it's now too short, it must be that userspace
626 		 * has changed the mappings underlying guest memory,
627 		 * so unregister the region.
628 		 */
629 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
630 		va = NULL;
631 	}
632 	if (vpap->pinned_addr)
633 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
634 					vpap->dirty);
635 	vpap->gpa = gpa;
636 	vpap->pinned_addr = va;
637 	vpap->dirty = false;
638 	if (va)
639 		vpap->pinned_end = va + vpap->len;
640 }
641 
642 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
643 {
644 	if (!(vcpu->arch.vpa.update_pending ||
645 	      vcpu->arch.slb_shadow.update_pending ||
646 	      vcpu->arch.dtl.update_pending))
647 		return;
648 
649 	spin_lock(&vcpu->arch.vpa_update_lock);
650 	if (vcpu->arch.vpa.update_pending) {
651 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
652 		if (vcpu->arch.vpa.pinned_addr)
653 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
654 	}
655 	if (vcpu->arch.dtl.update_pending) {
656 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
657 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
658 		vcpu->arch.dtl_index = 0;
659 	}
660 	if (vcpu->arch.slb_shadow.update_pending)
661 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
662 	spin_unlock(&vcpu->arch.vpa_update_lock);
663 }
664 
665 /*
666  * Return the accumulated stolen time for the vcore up until `now'.
667  * The caller should hold the vcore lock.
668  */
669 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
670 {
671 	u64 p;
672 	unsigned long flags;
673 
674 	spin_lock_irqsave(&vc->stoltb_lock, flags);
675 	p = vc->stolen_tb;
676 	if (vc->vcore_state != VCORE_INACTIVE &&
677 	    vc->preempt_tb != TB_NIL)
678 		p += now - vc->preempt_tb;
679 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
680 	return p;
681 }
682 
683 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
684 				    struct kvmppc_vcore *vc)
685 {
686 	struct dtl_entry *dt;
687 	struct lppaca *vpa;
688 	unsigned long stolen;
689 	unsigned long core_stolen;
690 	u64 now;
691 	unsigned long flags;
692 
693 	dt = vcpu->arch.dtl_ptr;
694 	vpa = vcpu->arch.vpa.pinned_addr;
695 	now = mftb();
696 	core_stolen = vcore_stolen_time(vc, now);
697 	stolen = core_stolen - vcpu->arch.stolen_logged;
698 	vcpu->arch.stolen_logged = core_stolen;
699 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
700 	stolen += vcpu->arch.busy_stolen;
701 	vcpu->arch.busy_stolen = 0;
702 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
703 	if (!dt || !vpa)
704 		return;
705 	memset(dt, 0, sizeof(struct dtl_entry));
706 	dt->dispatch_reason = 7;
707 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
708 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
709 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
710 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
711 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
712 	++dt;
713 	if (dt == vcpu->arch.dtl.pinned_end)
714 		dt = vcpu->arch.dtl.pinned_addr;
715 	vcpu->arch.dtl_ptr = dt;
716 	/* order writing *dt vs. writing vpa->dtl_idx */
717 	smp_wmb();
718 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
719 	vcpu->arch.dtl.dirty = true;
720 }
721 
722 /* See if there is a doorbell interrupt pending for a vcpu */
723 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
724 {
725 	int thr;
726 	struct kvmppc_vcore *vc;
727 
728 	if (vcpu->arch.doorbell_request)
729 		return true;
730 	/*
731 	 * Ensure that the read of vcore->dpdes comes after the read
732 	 * of vcpu->doorbell_request.  This barrier matches the
733 	 * smp_wmb() in kvmppc_guest_entry_inject().
734 	 */
735 	smp_rmb();
736 	vc = vcpu->arch.vcore;
737 	thr = vcpu->vcpu_id - vc->first_vcpuid;
738 	return !!(vc->dpdes & (1 << thr));
739 }
740 
741 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
742 {
743 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
744 		return true;
745 	if ((!vcpu->arch.vcore->arch_compat) &&
746 	    cpu_has_feature(CPU_FTR_ARCH_207S))
747 		return true;
748 	return false;
749 }
750 
751 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
752 			     unsigned long resource, unsigned long value1,
753 			     unsigned long value2)
754 {
755 	switch (resource) {
756 	case H_SET_MODE_RESOURCE_SET_CIABR:
757 		if (!kvmppc_power8_compatible(vcpu))
758 			return H_P2;
759 		if (value2)
760 			return H_P4;
761 		if (mflags)
762 			return H_UNSUPPORTED_FLAG_START;
763 		/* Guests can't breakpoint the hypervisor */
764 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
765 			return H_P3;
766 		vcpu->arch.ciabr  = value1;
767 		return H_SUCCESS;
768 	case H_SET_MODE_RESOURCE_SET_DAWR0:
769 		if (!kvmppc_power8_compatible(vcpu))
770 			return H_P2;
771 		if (!ppc_breakpoint_available())
772 			return H_P2;
773 		if (mflags)
774 			return H_UNSUPPORTED_FLAG_START;
775 		if (value2 & DABRX_HYP)
776 			return H_P4;
777 		vcpu->arch.dawr  = value1;
778 		vcpu->arch.dawrx = value2;
779 		return H_SUCCESS;
780 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
781 		/* KVM does not support mflags=2 (AIL=2) */
782 		if (mflags != 0 && mflags != 3)
783 			return H_UNSUPPORTED_FLAG_START;
784 		return H_TOO_HARD;
785 	default:
786 		return H_TOO_HARD;
787 	}
788 }
789 
790 /* Copy guest memory in place - must reside within a single memslot */
791 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
792 				  unsigned long len)
793 {
794 	struct kvm_memory_slot *to_memslot = NULL;
795 	struct kvm_memory_slot *from_memslot = NULL;
796 	unsigned long to_addr, from_addr;
797 	int r;
798 
799 	/* Get HPA for from address */
800 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
801 	if (!from_memslot)
802 		return -EFAULT;
803 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
804 			     << PAGE_SHIFT))
805 		return -EINVAL;
806 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
807 	if (kvm_is_error_hva(from_addr))
808 		return -EFAULT;
809 	from_addr |= (from & (PAGE_SIZE - 1));
810 
811 	/* Get HPA for to address */
812 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
813 	if (!to_memslot)
814 		return -EFAULT;
815 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
816 			   << PAGE_SHIFT))
817 		return -EINVAL;
818 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
819 	if (kvm_is_error_hva(to_addr))
820 		return -EFAULT;
821 	to_addr |= (to & (PAGE_SIZE - 1));
822 
823 	/* Perform copy */
824 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
825 			     len);
826 	if (r)
827 		return -EFAULT;
828 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
829 	return 0;
830 }
831 
832 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
833 			       unsigned long dest, unsigned long src)
834 {
835 	u64 pg_sz = SZ_4K;		/* 4K page size */
836 	u64 pg_mask = SZ_4K - 1;
837 	int ret;
838 
839 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
840 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
841 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
842 		return H_PARAMETER;
843 
844 	/* dest (and src if copy_page flag set) must be page aligned */
845 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
846 		return H_PARAMETER;
847 
848 	/* zero and/or copy the page as determined by the flags */
849 	if (flags & H_COPY_PAGE) {
850 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
851 		if (ret < 0)
852 			return H_PARAMETER;
853 	} else if (flags & H_ZERO_PAGE) {
854 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
855 		if (ret < 0)
856 			return H_PARAMETER;
857 	}
858 
859 	/* We can ignore the remaining flags */
860 
861 	return H_SUCCESS;
862 }
863 
864 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
865 {
866 	struct kvmppc_vcore *vcore = target->arch.vcore;
867 
868 	/*
869 	 * We expect to have been called by the real mode handler
870 	 * (kvmppc_rm_h_confer()) which would have directly returned
871 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
872 	 * have useful work to do and should not confer) so we don't
873 	 * recheck that here.
874 	 */
875 
876 	spin_lock(&vcore->lock);
877 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
878 	    vcore->vcore_state != VCORE_INACTIVE &&
879 	    vcore->runner)
880 		target = vcore->runner;
881 	spin_unlock(&vcore->lock);
882 
883 	return kvm_vcpu_yield_to(target);
884 }
885 
886 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
887 {
888 	int yield_count = 0;
889 	struct lppaca *lppaca;
890 
891 	spin_lock(&vcpu->arch.vpa_update_lock);
892 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
893 	if (lppaca)
894 		yield_count = be32_to_cpu(lppaca->yield_count);
895 	spin_unlock(&vcpu->arch.vpa_update_lock);
896 	return yield_count;
897 }
898 
899 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
900 {
901 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
902 	unsigned long target, ret = H_SUCCESS;
903 	int yield_count;
904 	struct kvm_vcpu *tvcpu;
905 	int idx, rc;
906 
907 	if (req <= MAX_HCALL_OPCODE &&
908 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
909 		return RESUME_HOST;
910 
911 	switch (req) {
912 	case H_CEDE:
913 		break;
914 	case H_PROD:
915 		target = kvmppc_get_gpr(vcpu, 4);
916 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
917 		if (!tvcpu) {
918 			ret = H_PARAMETER;
919 			break;
920 		}
921 		tvcpu->arch.prodded = 1;
922 		smp_mb();
923 		if (tvcpu->arch.ceded)
924 			kvmppc_fast_vcpu_kick_hv(tvcpu);
925 		break;
926 	case H_CONFER:
927 		target = kvmppc_get_gpr(vcpu, 4);
928 		if (target == -1)
929 			break;
930 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
931 		if (!tvcpu) {
932 			ret = H_PARAMETER;
933 			break;
934 		}
935 		yield_count = kvmppc_get_gpr(vcpu, 5);
936 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
937 			break;
938 		kvm_arch_vcpu_yield_to(tvcpu);
939 		break;
940 	case H_REGISTER_VPA:
941 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
942 					kvmppc_get_gpr(vcpu, 5),
943 					kvmppc_get_gpr(vcpu, 6));
944 		break;
945 	case H_RTAS:
946 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
947 			return RESUME_HOST;
948 
949 		idx = srcu_read_lock(&vcpu->kvm->srcu);
950 		rc = kvmppc_rtas_hcall(vcpu);
951 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
952 
953 		if (rc == -ENOENT)
954 			return RESUME_HOST;
955 		else if (rc == 0)
956 			break;
957 
958 		/* Send the error out to userspace via KVM_RUN */
959 		return rc;
960 	case H_LOGICAL_CI_LOAD:
961 		ret = kvmppc_h_logical_ci_load(vcpu);
962 		if (ret == H_TOO_HARD)
963 			return RESUME_HOST;
964 		break;
965 	case H_LOGICAL_CI_STORE:
966 		ret = kvmppc_h_logical_ci_store(vcpu);
967 		if (ret == H_TOO_HARD)
968 			return RESUME_HOST;
969 		break;
970 	case H_SET_MODE:
971 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
972 					kvmppc_get_gpr(vcpu, 5),
973 					kvmppc_get_gpr(vcpu, 6),
974 					kvmppc_get_gpr(vcpu, 7));
975 		if (ret == H_TOO_HARD)
976 			return RESUME_HOST;
977 		break;
978 	case H_XIRR:
979 	case H_CPPR:
980 	case H_EOI:
981 	case H_IPI:
982 	case H_IPOLL:
983 	case H_XIRR_X:
984 		if (kvmppc_xics_enabled(vcpu)) {
985 			if (xics_on_xive()) {
986 				ret = H_NOT_AVAILABLE;
987 				return RESUME_GUEST;
988 			}
989 			ret = kvmppc_xics_hcall(vcpu, req);
990 			break;
991 		}
992 		return RESUME_HOST;
993 	case H_SET_DABR:
994 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
995 		break;
996 	case H_SET_XDABR:
997 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
998 						kvmppc_get_gpr(vcpu, 5));
999 		break;
1000 #ifdef CONFIG_SPAPR_TCE_IOMMU
1001 	case H_GET_TCE:
1002 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1003 						kvmppc_get_gpr(vcpu, 5));
1004 		if (ret == H_TOO_HARD)
1005 			return RESUME_HOST;
1006 		break;
1007 	case H_PUT_TCE:
1008 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1009 						kvmppc_get_gpr(vcpu, 5),
1010 						kvmppc_get_gpr(vcpu, 6));
1011 		if (ret == H_TOO_HARD)
1012 			return RESUME_HOST;
1013 		break;
1014 	case H_PUT_TCE_INDIRECT:
1015 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1016 						kvmppc_get_gpr(vcpu, 5),
1017 						kvmppc_get_gpr(vcpu, 6),
1018 						kvmppc_get_gpr(vcpu, 7));
1019 		if (ret == H_TOO_HARD)
1020 			return RESUME_HOST;
1021 		break;
1022 	case H_STUFF_TCE:
1023 		ret = kvmppc_h_stuff_tce(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 #endif
1031 	case H_RANDOM:
1032 		if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1033 			ret = H_HARDWARE;
1034 		break;
1035 
1036 	case H_SET_PARTITION_TABLE:
1037 		ret = H_FUNCTION;
1038 		if (nesting_enabled(vcpu->kvm))
1039 			ret = kvmhv_set_partition_table(vcpu);
1040 		break;
1041 	case H_ENTER_NESTED:
1042 		ret = H_FUNCTION;
1043 		if (!nesting_enabled(vcpu->kvm))
1044 			break;
1045 		ret = kvmhv_enter_nested_guest(vcpu);
1046 		if (ret == H_INTERRUPT) {
1047 			kvmppc_set_gpr(vcpu, 3, 0);
1048 			vcpu->arch.hcall_needed = 0;
1049 			return -EINTR;
1050 		} else if (ret == H_TOO_HARD) {
1051 			kvmppc_set_gpr(vcpu, 3, 0);
1052 			vcpu->arch.hcall_needed = 0;
1053 			return RESUME_HOST;
1054 		}
1055 		break;
1056 	case H_TLB_INVALIDATE:
1057 		ret = H_FUNCTION;
1058 		if (nesting_enabled(vcpu->kvm))
1059 			ret = kvmhv_do_nested_tlbie(vcpu);
1060 		break;
1061 	case H_COPY_TOFROM_GUEST:
1062 		ret = H_FUNCTION;
1063 		if (nesting_enabled(vcpu->kvm))
1064 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1065 		break;
1066 	case H_PAGE_INIT:
1067 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1068 					 kvmppc_get_gpr(vcpu, 5),
1069 					 kvmppc_get_gpr(vcpu, 6));
1070 		break;
1071 	case H_SVM_PAGE_IN:
1072 		ret = H_UNSUPPORTED;
1073 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1074 			ret = kvmppc_h_svm_page_in(vcpu->kvm,
1075 						   kvmppc_get_gpr(vcpu, 4),
1076 						   kvmppc_get_gpr(vcpu, 5),
1077 						   kvmppc_get_gpr(vcpu, 6));
1078 		break;
1079 	case H_SVM_PAGE_OUT:
1080 		ret = H_UNSUPPORTED;
1081 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1082 			ret = kvmppc_h_svm_page_out(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_INIT_START:
1088 		ret = H_UNSUPPORTED;
1089 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1090 			ret = kvmppc_h_svm_init_start(vcpu->kvm);
1091 		break;
1092 	case H_SVM_INIT_DONE:
1093 		ret = H_UNSUPPORTED;
1094 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1095 			ret = kvmppc_h_svm_init_done(vcpu->kvm);
1096 		break;
1097 	case H_SVM_INIT_ABORT:
1098 		/*
1099 		 * Even if that call is made by the Ultravisor, the SSR1 value
1100 		 * is the guest context one, with the secure bit clear as it has
1101 		 * not yet been secured. So we can't check it here.
1102 		 * Instead the kvm->arch.secure_guest flag is checked inside
1103 		 * kvmppc_h_svm_init_abort().
1104 		 */
1105 		ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1106 		break;
1107 
1108 	default:
1109 		return RESUME_HOST;
1110 	}
1111 	kvmppc_set_gpr(vcpu, 3, ret);
1112 	vcpu->arch.hcall_needed = 0;
1113 	return RESUME_GUEST;
1114 }
1115 
1116 /*
1117  * Handle H_CEDE in the nested virtualization case where we haven't
1118  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1119  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1120  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1121  */
1122 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1123 {
1124 	vcpu->arch.shregs.msr |= MSR_EE;
1125 	vcpu->arch.ceded = 1;
1126 	smp_mb();
1127 	if (vcpu->arch.prodded) {
1128 		vcpu->arch.prodded = 0;
1129 		smp_mb();
1130 		vcpu->arch.ceded = 0;
1131 	}
1132 }
1133 
1134 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1135 {
1136 	switch (cmd) {
1137 	case H_CEDE:
1138 	case H_PROD:
1139 	case H_CONFER:
1140 	case H_REGISTER_VPA:
1141 	case H_SET_MODE:
1142 	case H_LOGICAL_CI_LOAD:
1143 	case H_LOGICAL_CI_STORE:
1144 #ifdef CONFIG_KVM_XICS
1145 	case H_XIRR:
1146 	case H_CPPR:
1147 	case H_EOI:
1148 	case H_IPI:
1149 	case H_IPOLL:
1150 	case H_XIRR_X:
1151 #endif
1152 	case H_PAGE_INIT:
1153 		return 1;
1154 	}
1155 
1156 	/* See if it's in the real-mode table */
1157 	return kvmppc_hcall_impl_hv_realmode(cmd);
1158 }
1159 
1160 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1161 {
1162 	u32 last_inst;
1163 
1164 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1165 					EMULATE_DONE) {
1166 		/*
1167 		 * Fetch failed, so return to guest and
1168 		 * try executing it again.
1169 		 */
1170 		return RESUME_GUEST;
1171 	}
1172 
1173 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1174 		vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1175 		vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1176 		return RESUME_HOST;
1177 	} else {
1178 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1179 		return RESUME_GUEST;
1180 	}
1181 }
1182 
1183 static void do_nothing(void *x)
1184 {
1185 }
1186 
1187 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1188 {
1189 	int thr, cpu, pcpu, nthreads;
1190 	struct kvm_vcpu *v;
1191 	unsigned long dpdes;
1192 
1193 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1194 	dpdes = 0;
1195 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1196 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1197 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1198 		if (!v)
1199 			continue;
1200 		/*
1201 		 * If the vcpu is currently running on a physical cpu thread,
1202 		 * interrupt it in order to pull it out of the guest briefly,
1203 		 * which will update its vcore->dpdes value.
1204 		 */
1205 		pcpu = READ_ONCE(v->cpu);
1206 		if (pcpu >= 0)
1207 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1208 		if (kvmppc_doorbell_pending(v))
1209 			dpdes |= 1 << thr;
1210 	}
1211 	return dpdes;
1212 }
1213 
1214 /*
1215  * On POWER9, emulate doorbell-related instructions in order to
1216  * give the guest the illusion of running on a multi-threaded core.
1217  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1218  * and mfspr DPDES.
1219  */
1220 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1221 {
1222 	u32 inst, rb, thr;
1223 	unsigned long arg;
1224 	struct kvm *kvm = vcpu->kvm;
1225 	struct kvm_vcpu *tvcpu;
1226 
1227 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1228 		return RESUME_GUEST;
1229 	if (get_op(inst) != 31)
1230 		return EMULATE_FAIL;
1231 	rb = get_rb(inst);
1232 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1233 	switch (get_xop(inst)) {
1234 	case OP_31_XOP_MSGSNDP:
1235 		arg = kvmppc_get_gpr(vcpu, rb);
1236 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1237 			break;
1238 		arg &= 0x3f;
1239 		if (arg >= kvm->arch.emul_smt_mode)
1240 			break;
1241 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1242 		if (!tvcpu)
1243 			break;
1244 		if (!tvcpu->arch.doorbell_request) {
1245 			tvcpu->arch.doorbell_request = 1;
1246 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1247 		}
1248 		break;
1249 	case OP_31_XOP_MSGCLRP:
1250 		arg = kvmppc_get_gpr(vcpu, rb);
1251 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1252 			break;
1253 		vcpu->arch.vcore->dpdes = 0;
1254 		vcpu->arch.doorbell_request = 0;
1255 		break;
1256 	case OP_31_XOP_MFSPR:
1257 		switch (get_sprn(inst)) {
1258 		case SPRN_TIR:
1259 			arg = thr;
1260 			break;
1261 		case SPRN_DPDES:
1262 			arg = kvmppc_read_dpdes(vcpu);
1263 			break;
1264 		default:
1265 			return EMULATE_FAIL;
1266 		}
1267 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1268 		break;
1269 	default:
1270 		return EMULATE_FAIL;
1271 	}
1272 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1273 	return RESUME_GUEST;
1274 }
1275 
1276 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1277 				 struct task_struct *tsk)
1278 {
1279 	struct kvm_run *run = vcpu->run;
1280 	int r = RESUME_HOST;
1281 
1282 	vcpu->stat.sum_exits++;
1283 
1284 	/*
1285 	 * This can happen if an interrupt occurs in the last stages
1286 	 * of guest entry or the first stages of guest exit (i.e. after
1287 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1288 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1289 	 * That can happen due to a bug, or due to a machine check
1290 	 * occurring at just the wrong time.
1291 	 */
1292 	if (vcpu->arch.shregs.msr & MSR_HV) {
1293 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1294 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1295 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1296 			vcpu->arch.shregs.msr);
1297 		kvmppc_dump_regs(vcpu);
1298 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1299 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1300 		return RESUME_HOST;
1301 	}
1302 	run->exit_reason = KVM_EXIT_UNKNOWN;
1303 	run->ready_for_interrupt_injection = 1;
1304 	switch (vcpu->arch.trap) {
1305 	/* We're good on these - the host merely wanted to get our attention */
1306 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1307 		vcpu->stat.dec_exits++;
1308 		r = RESUME_GUEST;
1309 		break;
1310 	case BOOK3S_INTERRUPT_EXTERNAL:
1311 	case BOOK3S_INTERRUPT_H_DOORBELL:
1312 	case BOOK3S_INTERRUPT_H_VIRT:
1313 		vcpu->stat.ext_intr_exits++;
1314 		r = RESUME_GUEST;
1315 		break;
1316 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1317 	case BOOK3S_INTERRUPT_HMI:
1318 	case BOOK3S_INTERRUPT_PERFMON:
1319 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1320 		r = RESUME_GUEST;
1321 		break;
1322 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1323 		/* Print the MCE event to host console. */
1324 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1325 
1326 		/*
1327 		 * If the guest can do FWNMI, exit to userspace so it can
1328 		 * deliver a FWNMI to the guest.
1329 		 * Otherwise we synthesize a machine check for the guest
1330 		 * so that it knows that the machine check occurred.
1331 		 */
1332 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1333 			ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1334 			kvmppc_core_queue_machine_check(vcpu, flags);
1335 			r = RESUME_GUEST;
1336 			break;
1337 		}
1338 
1339 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1340 		run->exit_reason = KVM_EXIT_NMI;
1341 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1342 		/* Clear out the old NMI status from run->flags */
1343 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1344 		/* Now set the NMI status */
1345 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1346 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1347 		else
1348 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1349 
1350 		r = RESUME_HOST;
1351 		break;
1352 	case BOOK3S_INTERRUPT_PROGRAM:
1353 	{
1354 		ulong flags;
1355 		/*
1356 		 * Normally program interrupts are delivered directly
1357 		 * to the guest by the hardware, but we can get here
1358 		 * as a result of a hypervisor emulation interrupt
1359 		 * (e40) getting turned into a 700 by BML RTAS.
1360 		 */
1361 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1362 		kvmppc_core_queue_program(vcpu, flags);
1363 		r = RESUME_GUEST;
1364 		break;
1365 	}
1366 	case BOOK3S_INTERRUPT_SYSCALL:
1367 	{
1368 		/* hcall - punt to userspace */
1369 		int i;
1370 
1371 		/* hypercall with MSR_PR has already been handled in rmode,
1372 		 * and never reaches here.
1373 		 */
1374 
1375 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1376 		for (i = 0; i < 9; ++i)
1377 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1378 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1379 		vcpu->arch.hcall_needed = 1;
1380 		r = RESUME_HOST;
1381 		break;
1382 	}
1383 	/*
1384 	 * We get these next two if the guest accesses a page which it thinks
1385 	 * it has mapped but which is not actually present, either because
1386 	 * it is for an emulated I/O device or because the corresonding
1387 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1388 	 * have been handled already.
1389 	 */
1390 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1391 		r = RESUME_PAGE_FAULT;
1392 		break;
1393 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1394 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1395 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1396 			DSISR_SRR1_MATCH_64S;
1397 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1398 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1399 		r = RESUME_PAGE_FAULT;
1400 		break;
1401 	/*
1402 	 * This occurs if the guest executes an illegal instruction.
1403 	 * If the guest debug is disabled, generate a program interrupt
1404 	 * to the guest. If guest debug is enabled, we need to check
1405 	 * whether the instruction is a software breakpoint instruction.
1406 	 * Accordingly return to Guest or Host.
1407 	 */
1408 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1409 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1410 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1411 				swab32(vcpu->arch.emul_inst) :
1412 				vcpu->arch.emul_inst;
1413 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1414 			r = kvmppc_emulate_debug_inst(vcpu);
1415 		} else {
1416 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1417 			r = RESUME_GUEST;
1418 		}
1419 		break;
1420 	/*
1421 	 * This occurs if the guest (kernel or userspace), does something that
1422 	 * is prohibited by HFSCR.
1423 	 * On POWER9, this could be a doorbell instruction that we need
1424 	 * to emulate.
1425 	 * Otherwise, we just generate a program interrupt to the guest.
1426 	 */
1427 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1428 		r = EMULATE_FAIL;
1429 		if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1430 		    cpu_has_feature(CPU_FTR_ARCH_300))
1431 			r = kvmppc_emulate_doorbell_instr(vcpu);
1432 		if (r == EMULATE_FAIL) {
1433 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1434 			r = RESUME_GUEST;
1435 		}
1436 		break;
1437 
1438 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1439 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1440 		/*
1441 		 * This occurs for various TM-related instructions that
1442 		 * we need to emulate on POWER9 DD2.2.  We have already
1443 		 * handled the cases where the guest was in real-suspend
1444 		 * mode and was transitioning to transactional state.
1445 		 */
1446 		r = kvmhv_p9_tm_emulation(vcpu);
1447 		break;
1448 #endif
1449 
1450 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1451 		r = RESUME_PASSTHROUGH;
1452 		break;
1453 	default:
1454 		kvmppc_dump_regs(vcpu);
1455 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1456 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1457 			vcpu->arch.shregs.msr);
1458 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1459 		r = RESUME_HOST;
1460 		break;
1461 	}
1462 
1463 	return r;
1464 }
1465 
1466 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1467 {
1468 	int r;
1469 	int srcu_idx;
1470 
1471 	vcpu->stat.sum_exits++;
1472 
1473 	/*
1474 	 * This can happen if an interrupt occurs in the last stages
1475 	 * of guest entry or the first stages of guest exit (i.e. after
1476 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1477 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1478 	 * That can happen due to a bug, or due to a machine check
1479 	 * occurring at just the wrong time.
1480 	 */
1481 	if (vcpu->arch.shregs.msr & MSR_HV) {
1482 		pr_emerg("KVM trap in HV mode while nested!\n");
1483 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1484 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1485 			 vcpu->arch.shregs.msr);
1486 		kvmppc_dump_regs(vcpu);
1487 		return RESUME_HOST;
1488 	}
1489 	switch (vcpu->arch.trap) {
1490 	/* We're good on these - the host merely wanted to get our attention */
1491 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1492 		vcpu->stat.dec_exits++;
1493 		r = RESUME_GUEST;
1494 		break;
1495 	case BOOK3S_INTERRUPT_EXTERNAL:
1496 		vcpu->stat.ext_intr_exits++;
1497 		r = RESUME_HOST;
1498 		break;
1499 	case BOOK3S_INTERRUPT_H_DOORBELL:
1500 	case BOOK3S_INTERRUPT_H_VIRT:
1501 		vcpu->stat.ext_intr_exits++;
1502 		r = RESUME_GUEST;
1503 		break;
1504 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1505 	case BOOK3S_INTERRUPT_HMI:
1506 	case BOOK3S_INTERRUPT_PERFMON:
1507 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1508 		r = RESUME_GUEST;
1509 		break;
1510 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1511 		/* Pass the machine check to the L1 guest */
1512 		r = RESUME_HOST;
1513 		/* Print the MCE event to host console. */
1514 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1515 		break;
1516 	/*
1517 	 * We get these next two if the guest accesses a page which it thinks
1518 	 * it has mapped but which is not actually present, either because
1519 	 * it is for an emulated I/O device or because the corresonding
1520 	 * host page has been paged out.
1521 	 */
1522 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1523 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1524 		r = kvmhv_nested_page_fault(vcpu);
1525 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1526 		break;
1527 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1528 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1529 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1530 					 DSISR_SRR1_MATCH_64S;
1531 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1532 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1533 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1534 		r = kvmhv_nested_page_fault(vcpu);
1535 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1536 		break;
1537 
1538 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1539 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1540 		/*
1541 		 * This occurs for various TM-related instructions that
1542 		 * we need to emulate on POWER9 DD2.2.  We have already
1543 		 * handled the cases where the guest was in real-suspend
1544 		 * mode and was transitioning to transactional state.
1545 		 */
1546 		r = kvmhv_p9_tm_emulation(vcpu);
1547 		break;
1548 #endif
1549 
1550 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1551 		vcpu->arch.trap = 0;
1552 		r = RESUME_GUEST;
1553 		if (!xics_on_xive())
1554 			kvmppc_xics_rm_complete(vcpu, 0);
1555 		break;
1556 	default:
1557 		r = RESUME_HOST;
1558 		break;
1559 	}
1560 
1561 	return r;
1562 }
1563 
1564 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1565 					    struct kvm_sregs *sregs)
1566 {
1567 	int i;
1568 
1569 	memset(sregs, 0, sizeof(struct kvm_sregs));
1570 	sregs->pvr = vcpu->arch.pvr;
1571 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1572 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1573 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1574 	}
1575 
1576 	return 0;
1577 }
1578 
1579 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1580 					    struct kvm_sregs *sregs)
1581 {
1582 	int i, j;
1583 
1584 	/* Only accept the same PVR as the host's, since we can't spoof it */
1585 	if (sregs->pvr != vcpu->arch.pvr)
1586 		return -EINVAL;
1587 
1588 	j = 0;
1589 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1590 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1591 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1592 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1593 			++j;
1594 		}
1595 	}
1596 	vcpu->arch.slb_max = j;
1597 
1598 	return 0;
1599 }
1600 
1601 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1602 		bool preserve_top32)
1603 {
1604 	struct kvm *kvm = vcpu->kvm;
1605 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1606 	u64 mask;
1607 
1608 	spin_lock(&vc->lock);
1609 	/*
1610 	 * If ILE (interrupt little-endian) has changed, update the
1611 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1612 	 */
1613 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1614 		struct kvm_vcpu *vcpu;
1615 		int i;
1616 
1617 		kvm_for_each_vcpu(i, vcpu, kvm) {
1618 			if (vcpu->arch.vcore != vc)
1619 				continue;
1620 			if (new_lpcr & LPCR_ILE)
1621 				vcpu->arch.intr_msr |= MSR_LE;
1622 			else
1623 				vcpu->arch.intr_msr &= ~MSR_LE;
1624 		}
1625 	}
1626 
1627 	/*
1628 	 * Userspace can only modify DPFD (default prefetch depth),
1629 	 * ILE (interrupt little-endian) and TC (translation control).
1630 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1631 	 */
1632 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1633 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1634 		mask |= LPCR_AIL;
1635 	/*
1636 	 * On POWER9, allow userspace to enable large decrementer for the
1637 	 * guest, whether or not the host has it enabled.
1638 	 */
1639 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1640 		mask |= LPCR_LD;
1641 
1642 	/* Broken 32-bit version of LPCR must not clear top bits */
1643 	if (preserve_top32)
1644 		mask &= 0xFFFFFFFF;
1645 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1646 	spin_unlock(&vc->lock);
1647 }
1648 
1649 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1650 				 union kvmppc_one_reg *val)
1651 {
1652 	int r = 0;
1653 	long int i;
1654 
1655 	switch (id) {
1656 	case KVM_REG_PPC_DEBUG_INST:
1657 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1658 		break;
1659 	case KVM_REG_PPC_HIOR:
1660 		*val = get_reg_val(id, 0);
1661 		break;
1662 	case KVM_REG_PPC_DABR:
1663 		*val = get_reg_val(id, vcpu->arch.dabr);
1664 		break;
1665 	case KVM_REG_PPC_DABRX:
1666 		*val = get_reg_val(id, vcpu->arch.dabrx);
1667 		break;
1668 	case KVM_REG_PPC_DSCR:
1669 		*val = get_reg_val(id, vcpu->arch.dscr);
1670 		break;
1671 	case KVM_REG_PPC_PURR:
1672 		*val = get_reg_val(id, vcpu->arch.purr);
1673 		break;
1674 	case KVM_REG_PPC_SPURR:
1675 		*val = get_reg_val(id, vcpu->arch.spurr);
1676 		break;
1677 	case KVM_REG_PPC_AMR:
1678 		*val = get_reg_val(id, vcpu->arch.amr);
1679 		break;
1680 	case KVM_REG_PPC_UAMOR:
1681 		*val = get_reg_val(id, vcpu->arch.uamor);
1682 		break;
1683 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1684 		i = id - KVM_REG_PPC_MMCR0;
1685 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1686 		break;
1687 	case KVM_REG_PPC_MMCR2:
1688 		*val = get_reg_val(id, vcpu->arch.mmcr[2]);
1689 		break;
1690 	case KVM_REG_PPC_MMCRA:
1691 		*val = get_reg_val(id, vcpu->arch.mmcra);
1692 		break;
1693 	case KVM_REG_PPC_MMCRS:
1694 		*val = get_reg_val(id, vcpu->arch.mmcrs);
1695 		break;
1696 	case KVM_REG_PPC_MMCR3:
1697 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
1698 		break;
1699 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1700 		i = id - KVM_REG_PPC_PMC1;
1701 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1702 		break;
1703 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1704 		i = id - KVM_REG_PPC_SPMC1;
1705 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1706 		break;
1707 	case KVM_REG_PPC_SIAR:
1708 		*val = get_reg_val(id, vcpu->arch.siar);
1709 		break;
1710 	case KVM_REG_PPC_SDAR:
1711 		*val = get_reg_val(id, vcpu->arch.sdar);
1712 		break;
1713 	case KVM_REG_PPC_SIER:
1714 		*val = get_reg_val(id, vcpu->arch.sier[0]);
1715 		break;
1716 	case KVM_REG_PPC_SIER2:
1717 		*val = get_reg_val(id, vcpu->arch.sier[1]);
1718 		break;
1719 	case KVM_REG_PPC_SIER3:
1720 		*val = get_reg_val(id, vcpu->arch.sier[2]);
1721 		break;
1722 	case KVM_REG_PPC_IAMR:
1723 		*val = get_reg_val(id, vcpu->arch.iamr);
1724 		break;
1725 	case KVM_REG_PPC_PSPB:
1726 		*val = get_reg_val(id, vcpu->arch.pspb);
1727 		break;
1728 	case KVM_REG_PPC_DPDES:
1729 		/*
1730 		 * On POWER9, where we are emulating msgsndp etc.,
1731 		 * we return 1 bit for each vcpu, which can come from
1732 		 * either vcore->dpdes or doorbell_request.
1733 		 * On POWER8, doorbell_request is 0.
1734 		 */
1735 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1736 				   vcpu->arch.doorbell_request);
1737 		break;
1738 	case KVM_REG_PPC_VTB:
1739 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1740 		break;
1741 	case KVM_REG_PPC_DAWR:
1742 		*val = get_reg_val(id, vcpu->arch.dawr);
1743 		break;
1744 	case KVM_REG_PPC_DAWRX:
1745 		*val = get_reg_val(id, vcpu->arch.dawrx);
1746 		break;
1747 	case KVM_REG_PPC_CIABR:
1748 		*val = get_reg_val(id, vcpu->arch.ciabr);
1749 		break;
1750 	case KVM_REG_PPC_CSIGR:
1751 		*val = get_reg_val(id, vcpu->arch.csigr);
1752 		break;
1753 	case KVM_REG_PPC_TACR:
1754 		*val = get_reg_val(id, vcpu->arch.tacr);
1755 		break;
1756 	case KVM_REG_PPC_TCSCR:
1757 		*val = get_reg_val(id, vcpu->arch.tcscr);
1758 		break;
1759 	case KVM_REG_PPC_PID:
1760 		*val = get_reg_val(id, vcpu->arch.pid);
1761 		break;
1762 	case KVM_REG_PPC_ACOP:
1763 		*val = get_reg_val(id, vcpu->arch.acop);
1764 		break;
1765 	case KVM_REG_PPC_WORT:
1766 		*val = get_reg_val(id, vcpu->arch.wort);
1767 		break;
1768 	case KVM_REG_PPC_TIDR:
1769 		*val = get_reg_val(id, vcpu->arch.tid);
1770 		break;
1771 	case KVM_REG_PPC_PSSCR:
1772 		*val = get_reg_val(id, vcpu->arch.psscr);
1773 		break;
1774 	case KVM_REG_PPC_VPA_ADDR:
1775 		spin_lock(&vcpu->arch.vpa_update_lock);
1776 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1777 		spin_unlock(&vcpu->arch.vpa_update_lock);
1778 		break;
1779 	case KVM_REG_PPC_VPA_SLB:
1780 		spin_lock(&vcpu->arch.vpa_update_lock);
1781 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1782 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1783 		spin_unlock(&vcpu->arch.vpa_update_lock);
1784 		break;
1785 	case KVM_REG_PPC_VPA_DTL:
1786 		spin_lock(&vcpu->arch.vpa_update_lock);
1787 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1788 		val->vpaval.length = vcpu->arch.dtl.len;
1789 		spin_unlock(&vcpu->arch.vpa_update_lock);
1790 		break;
1791 	case KVM_REG_PPC_TB_OFFSET:
1792 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1793 		break;
1794 	case KVM_REG_PPC_LPCR:
1795 	case KVM_REG_PPC_LPCR_64:
1796 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1797 		break;
1798 	case KVM_REG_PPC_PPR:
1799 		*val = get_reg_val(id, vcpu->arch.ppr);
1800 		break;
1801 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1802 	case KVM_REG_PPC_TFHAR:
1803 		*val = get_reg_val(id, vcpu->arch.tfhar);
1804 		break;
1805 	case KVM_REG_PPC_TFIAR:
1806 		*val = get_reg_val(id, vcpu->arch.tfiar);
1807 		break;
1808 	case KVM_REG_PPC_TEXASR:
1809 		*val = get_reg_val(id, vcpu->arch.texasr);
1810 		break;
1811 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1812 		i = id - KVM_REG_PPC_TM_GPR0;
1813 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1814 		break;
1815 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1816 	{
1817 		int j;
1818 		i = id - KVM_REG_PPC_TM_VSR0;
1819 		if (i < 32)
1820 			for (j = 0; j < TS_FPRWIDTH; j++)
1821 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1822 		else {
1823 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1824 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1825 			else
1826 				r = -ENXIO;
1827 		}
1828 		break;
1829 	}
1830 	case KVM_REG_PPC_TM_CR:
1831 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1832 		break;
1833 	case KVM_REG_PPC_TM_XER:
1834 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1835 		break;
1836 	case KVM_REG_PPC_TM_LR:
1837 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1838 		break;
1839 	case KVM_REG_PPC_TM_CTR:
1840 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1841 		break;
1842 	case KVM_REG_PPC_TM_FPSCR:
1843 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1844 		break;
1845 	case KVM_REG_PPC_TM_AMR:
1846 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1847 		break;
1848 	case KVM_REG_PPC_TM_PPR:
1849 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1850 		break;
1851 	case KVM_REG_PPC_TM_VRSAVE:
1852 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1853 		break;
1854 	case KVM_REG_PPC_TM_VSCR:
1855 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1856 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1857 		else
1858 			r = -ENXIO;
1859 		break;
1860 	case KVM_REG_PPC_TM_DSCR:
1861 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1862 		break;
1863 	case KVM_REG_PPC_TM_TAR:
1864 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1865 		break;
1866 #endif
1867 	case KVM_REG_PPC_ARCH_COMPAT:
1868 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1869 		break;
1870 	case KVM_REG_PPC_DEC_EXPIRY:
1871 		*val = get_reg_val(id, vcpu->arch.dec_expires +
1872 				   vcpu->arch.vcore->tb_offset);
1873 		break;
1874 	case KVM_REG_PPC_ONLINE:
1875 		*val = get_reg_val(id, vcpu->arch.online);
1876 		break;
1877 	case KVM_REG_PPC_PTCR:
1878 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1879 		break;
1880 	default:
1881 		r = -EINVAL;
1882 		break;
1883 	}
1884 
1885 	return r;
1886 }
1887 
1888 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1889 				 union kvmppc_one_reg *val)
1890 {
1891 	int r = 0;
1892 	long int i;
1893 	unsigned long addr, len;
1894 
1895 	switch (id) {
1896 	case KVM_REG_PPC_HIOR:
1897 		/* Only allow this to be set to zero */
1898 		if (set_reg_val(id, *val))
1899 			r = -EINVAL;
1900 		break;
1901 	case KVM_REG_PPC_DABR:
1902 		vcpu->arch.dabr = set_reg_val(id, *val);
1903 		break;
1904 	case KVM_REG_PPC_DABRX:
1905 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1906 		break;
1907 	case KVM_REG_PPC_DSCR:
1908 		vcpu->arch.dscr = set_reg_val(id, *val);
1909 		break;
1910 	case KVM_REG_PPC_PURR:
1911 		vcpu->arch.purr = set_reg_val(id, *val);
1912 		break;
1913 	case KVM_REG_PPC_SPURR:
1914 		vcpu->arch.spurr = set_reg_val(id, *val);
1915 		break;
1916 	case KVM_REG_PPC_AMR:
1917 		vcpu->arch.amr = set_reg_val(id, *val);
1918 		break;
1919 	case KVM_REG_PPC_UAMOR:
1920 		vcpu->arch.uamor = set_reg_val(id, *val);
1921 		break;
1922 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1923 		i = id - KVM_REG_PPC_MMCR0;
1924 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1925 		break;
1926 	case KVM_REG_PPC_MMCR2:
1927 		vcpu->arch.mmcr[2] = set_reg_val(id, *val);
1928 		break;
1929 	case KVM_REG_PPC_MMCRA:
1930 		vcpu->arch.mmcra = set_reg_val(id, *val);
1931 		break;
1932 	case KVM_REG_PPC_MMCRS:
1933 		vcpu->arch.mmcrs = set_reg_val(id, *val);
1934 		break;
1935 	case KVM_REG_PPC_MMCR3:
1936 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
1937 		break;
1938 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1939 		i = id - KVM_REG_PPC_PMC1;
1940 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1941 		break;
1942 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1943 		i = id - KVM_REG_PPC_SPMC1;
1944 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1945 		break;
1946 	case KVM_REG_PPC_SIAR:
1947 		vcpu->arch.siar = set_reg_val(id, *val);
1948 		break;
1949 	case KVM_REG_PPC_SDAR:
1950 		vcpu->arch.sdar = set_reg_val(id, *val);
1951 		break;
1952 	case KVM_REG_PPC_SIER:
1953 		vcpu->arch.sier[0] = set_reg_val(id, *val);
1954 		break;
1955 	case KVM_REG_PPC_SIER2:
1956 		vcpu->arch.sier[1] = set_reg_val(id, *val);
1957 		break;
1958 	case KVM_REG_PPC_SIER3:
1959 		vcpu->arch.sier[2] = set_reg_val(id, *val);
1960 		break;
1961 	case KVM_REG_PPC_IAMR:
1962 		vcpu->arch.iamr = set_reg_val(id, *val);
1963 		break;
1964 	case KVM_REG_PPC_PSPB:
1965 		vcpu->arch.pspb = set_reg_val(id, *val);
1966 		break;
1967 	case KVM_REG_PPC_DPDES:
1968 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1969 		break;
1970 	case KVM_REG_PPC_VTB:
1971 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1972 		break;
1973 	case KVM_REG_PPC_DAWR:
1974 		vcpu->arch.dawr = set_reg_val(id, *val);
1975 		break;
1976 	case KVM_REG_PPC_DAWRX:
1977 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1978 		break;
1979 	case KVM_REG_PPC_CIABR:
1980 		vcpu->arch.ciabr = set_reg_val(id, *val);
1981 		/* Don't allow setting breakpoints in hypervisor code */
1982 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1983 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1984 		break;
1985 	case KVM_REG_PPC_CSIGR:
1986 		vcpu->arch.csigr = set_reg_val(id, *val);
1987 		break;
1988 	case KVM_REG_PPC_TACR:
1989 		vcpu->arch.tacr = set_reg_val(id, *val);
1990 		break;
1991 	case KVM_REG_PPC_TCSCR:
1992 		vcpu->arch.tcscr = set_reg_val(id, *val);
1993 		break;
1994 	case KVM_REG_PPC_PID:
1995 		vcpu->arch.pid = set_reg_val(id, *val);
1996 		break;
1997 	case KVM_REG_PPC_ACOP:
1998 		vcpu->arch.acop = set_reg_val(id, *val);
1999 		break;
2000 	case KVM_REG_PPC_WORT:
2001 		vcpu->arch.wort = set_reg_val(id, *val);
2002 		break;
2003 	case KVM_REG_PPC_TIDR:
2004 		vcpu->arch.tid = set_reg_val(id, *val);
2005 		break;
2006 	case KVM_REG_PPC_PSSCR:
2007 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2008 		break;
2009 	case KVM_REG_PPC_VPA_ADDR:
2010 		addr = set_reg_val(id, *val);
2011 		r = -EINVAL;
2012 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2013 			      vcpu->arch.dtl.next_gpa))
2014 			break;
2015 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2016 		break;
2017 	case KVM_REG_PPC_VPA_SLB:
2018 		addr = val->vpaval.addr;
2019 		len = val->vpaval.length;
2020 		r = -EINVAL;
2021 		if (addr && !vcpu->arch.vpa.next_gpa)
2022 			break;
2023 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2024 		break;
2025 	case KVM_REG_PPC_VPA_DTL:
2026 		addr = val->vpaval.addr;
2027 		len = val->vpaval.length;
2028 		r = -EINVAL;
2029 		if (addr && (len < sizeof(struct dtl_entry) ||
2030 			     !vcpu->arch.vpa.next_gpa))
2031 			break;
2032 		len -= len % sizeof(struct dtl_entry);
2033 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2034 		break;
2035 	case KVM_REG_PPC_TB_OFFSET:
2036 		/* round up to multiple of 2^24 */
2037 		vcpu->arch.vcore->tb_offset =
2038 			ALIGN(set_reg_val(id, *val), 1UL << 24);
2039 		break;
2040 	case KVM_REG_PPC_LPCR:
2041 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2042 		break;
2043 	case KVM_REG_PPC_LPCR_64:
2044 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2045 		break;
2046 	case KVM_REG_PPC_PPR:
2047 		vcpu->arch.ppr = set_reg_val(id, *val);
2048 		break;
2049 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2050 	case KVM_REG_PPC_TFHAR:
2051 		vcpu->arch.tfhar = set_reg_val(id, *val);
2052 		break;
2053 	case KVM_REG_PPC_TFIAR:
2054 		vcpu->arch.tfiar = set_reg_val(id, *val);
2055 		break;
2056 	case KVM_REG_PPC_TEXASR:
2057 		vcpu->arch.texasr = set_reg_val(id, *val);
2058 		break;
2059 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2060 		i = id - KVM_REG_PPC_TM_GPR0;
2061 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2062 		break;
2063 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2064 	{
2065 		int j;
2066 		i = id - KVM_REG_PPC_TM_VSR0;
2067 		if (i < 32)
2068 			for (j = 0; j < TS_FPRWIDTH; j++)
2069 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2070 		else
2071 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2072 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2073 			else
2074 				r = -ENXIO;
2075 		break;
2076 	}
2077 	case KVM_REG_PPC_TM_CR:
2078 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2079 		break;
2080 	case KVM_REG_PPC_TM_XER:
2081 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2082 		break;
2083 	case KVM_REG_PPC_TM_LR:
2084 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2085 		break;
2086 	case KVM_REG_PPC_TM_CTR:
2087 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2088 		break;
2089 	case KVM_REG_PPC_TM_FPSCR:
2090 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2091 		break;
2092 	case KVM_REG_PPC_TM_AMR:
2093 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2094 		break;
2095 	case KVM_REG_PPC_TM_PPR:
2096 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2097 		break;
2098 	case KVM_REG_PPC_TM_VRSAVE:
2099 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2100 		break;
2101 	case KVM_REG_PPC_TM_VSCR:
2102 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2103 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2104 		else
2105 			r = - ENXIO;
2106 		break;
2107 	case KVM_REG_PPC_TM_DSCR:
2108 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2109 		break;
2110 	case KVM_REG_PPC_TM_TAR:
2111 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2112 		break;
2113 #endif
2114 	case KVM_REG_PPC_ARCH_COMPAT:
2115 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2116 		break;
2117 	case KVM_REG_PPC_DEC_EXPIRY:
2118 		vcpu->arch.dec_expires = set_reg_val(id, *val) -
2119 			vcpu->arch.vcore->tb_offset;
2120 		break;
2121 	case KVM_REG_PPC_ONLINE:
2122 		i = set_reg_val(id, *val);
2123 		if (i && !vcpu->arch.online)
2124 			atomic_inc(&vcpu->arch.vcore->online_count);
2125 		else if (!i && vcpu->arch.online)
2126 			atomic_dec(&vcpu->arch.vcore->online_count);
2127 		vcpu->arch.online = i;
2128 		break;
2129 	case KVM_REG_PPC_PTCR:
2130 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2131 		break;
2132 	default:
2133 		r = -EINVAL;
2134 		break;
2135 	}
2136 
2137 	return r;
2138 }
2139 
2140 /*
2141  * On POWER9, threads are independent and can be in different partitions.
2142  * Therefore we consider each thread to be a subcore.
2143  * There is a restriction that all threads have to be in the same
2144  * MMU mode (radix or HPT), unfortunately, but since we only support
2145  * HPT guests on a HPT host so far, that isn't an impediment yet.
2146  */
2147 static int threads_per_vcore(struct kvm *kvm)
2148 {
2149 	if (kvm->arch.threads_indep)
2150 		return 1;
2151 	return threads_per_subcore;
2152 }
2153 
2154 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2155 {
2156 	struct kvmppc_vcore *vcore;
2157 
2158 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2159 
2160 	if (vcore == NULL)
2161 		return NULL;
2162 
2163 	spin_lock_init(&vcore->lock);
2164 	spin_lock_init(&vcore->stoltb_lock);
2165 	rcuwait_init(&vcore->wait);
2166 	vcore->preempt_tb = TB_NIL;
2167 	vcore->lpcr = kvm->arch.lpcr;
2168 	vcore->first_vcpuid = id;
2169 	vcore->kvm = kvm;
2170 	INIT_LIST_HEAD(&vcore->preempt_list);
2171 
2172 	return vcore;
2173 }
2174 
2175 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2176 static struct debugfs_timings_element {
2177 	const char *name;
2178 	size_t offset;
2179 } timings[] = {
2180 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2181 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2182 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2183 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2184 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2185 };
2186 
2187 #define N_TIMINGS	(ARRAY_SIZE(timings))
2188 
2189 struct debugfs_timings_state {
2190 	struct kvm_vcpu	*vcpu;
2191 	unsigned int	buflen;
2192 	char		buf[N_TIMINGS * 100];
2193 };
2194 
2195 static int debugfs_timings_open(struct inode *inode, struct file *file)
2196 {
2197 	struct kvm_vcpu *vcpu = inode->i_private;
2198 	struct debugfs_timings_state *p;
2199 
2200 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2201 	if (!p)
2202 		return -ENOMEM;
2203 
2204 	kvm_get_kvm(vcpu->kvm);
2205 	p->vcpu = vcpu;
2206 	file->private_data = p;
2207 
2208 	return nonseekable_open(inode, file);
2209 }
2210 
2211 static int debugfs_timings_release(struct inode *inode, struct file *file)
2212 {
2213 	struct debugfs_timings_state *p = file->private_data;
2214 
2215 	kvm_put_kvm(p->vcpu->kvm);
2216 	kfree(p);
2217 	return 0;
2218 }
2219 
2220 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2221 				    size_t len, loff_t *ppos)
2222 {
2223 	struct debugfs_timings_state *p = file->private_data;
2224 	struct kvm_vcpu *vcpu = p->vcpu;
2225 	char *s, *buf_end;
2226 	struct kvmhv_tb_accumulator tb;
2227 	u64 count;
2228 	loff_t pos;
2229 	ssize_t n;
2230 	int i, loops;
2231 	bool ok;
2232 
2233 	if (!p->buflen) {
2234 		s = p->buf;
2235 		buf_end = s + sizeof(p->buf);
2236 		for (i = 0; i < N_TIMINGS; ++i) {
2237 			struct kvmhv_tb_accumulator *acc;
2238 
2239 			acc = (struct kvmhv_tb_accumulator *)
2240 				((unsigned long)vcpu + timings[i].offset);
2241 			ok = false;
2242 			for (loops = 0; loops < 1000; ++loops) {
2243 				count = acc->seqcount;
2244 				if (!(count & 1)) {
2245 					smp_rmb();
2246 					tb = *acc;
2247 					smp_rmb();
2248 					if (count == acc->seqcount) {
2249 						ok = true;
2250 						break;
2251 					}
2252 				}
2253 				udelay(1);
2254 			}
2255 			if (!ok)
2256 				snprintf(s, buf_end - s, "%s: stuck\n",
2257 					timings[i].name);
2258 			else
2259 				snprintf(s, buf_end - s,
2260 					"%s: %llu %llu %llu %llu\n",
2261 					timings[i].name, count / 2,
2262 					tb_to_ns(tb.tb_total),
2263 					tb_to_ns(tb.tb_min),
2264 					tb_to_ns(tb.tb_max));
2265 			s += strlen(s);
2266 		}
2267 		p->buflen = s - p->buf;
2268 	}
2269 
2270 	pos = *ppos;
2271 	if (pos >= p->buflen)
2272 		return 0;
2273 	if (len > p->buflen - pos)
2274 		len = p->buflen - pos;
2275 	n = copy_to_user(buf, p->buf + pos, len);
2276 	if (n) {
2277 		if (n == len)
2278 			return -EFAULT;
2279 		len -= n;
2280 	}
2281 	*ppos = pos + len;
2282 	return len;
2283 }
2284 
2285 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2286 				     size_t len, loff_t *ppos)
2287 {
2288 	return -EACCES;
2289 }
2290 
2291 static const struct file_operations debugfs_timings_ops = {
2292 	.owner	 = THIS_MODULE,
2293 	.open	 = debugfs_timings_open,
2294 	.release = debugfs_timings_release,
2295 	.read	 = debugfs_timings_read,
2296 	.write	 = debugfs_timings_write,
2297 	.llseek	 = generic_file_llseek,
2298 };
2299 
2300 /* Create a debugfs directory for the vcpu */
2301 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2302 {
2303 	char buf[16];
2304 	struct kvm *kvm = vcpu->kvm;
2305 
2306 	snprintf(buf, sizeof(buf), "vcpu%u", id);
2307 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2308 	debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2309 			    &debugfs_timings_ops);
2310 }
2311 
2312 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2313 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2314 {
2315 }
2316 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2317 
2318 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2319 {
2320 	int err;
2321 	int core;
2322 	struct kvmppc_vcore *vcore;
2323 	struct kvm *kvm;
2324 	unsigned int id;
2325 
2326 	kvm = vcpu->kvm;
2327 	id = vcpu->vcpu_id;
2328 
2329 	vcpu->arch.shared = &vcpu->arch.shregs;
2330 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2331 	/*
2332 	 * The shared struct is never shared on HV,
2333 	 * so we can always use host endianness
2334 	 */
2335 #ifdef __BIG_ENDIAN__
2336 	vcpu->arch.shared_big_endian = true;
2337 #else
2338 	vcpu->arch.shared_big_endian = false;
2339 #endif
2340 #endif
2341 	vcpu->arch.mmcr[0] = MMCR0_FC;
2342 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2343 	/* default to host PVR, since we can't spoof it */
2344 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2345 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2346 	spin_lock_init(&vcpu->arch.tbacct_lock);
2347 	vcpu->arch.busy_preempt = TB_NIL;
2348 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2349 
2350 	/*
2351 	 * Set the default HFSCR for the guest from the host value.
2352 	 * This value is only used on POWER9.
2353 	 * On POWER9, we want to virtualize the doorbell facility, so we
2354 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2355 	 * to trap and then we emulate them.
2356 	 */
2357 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2358 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2359 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2360 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2361 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2362 			vcpu->arch.hfscr |= HFSCR_TM;
2363 	}
2364 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2365 		vcpu->arch.hfscr |= HFSCR_TM;
2366 
2367 	kvmppc_mmu_book3s_hv_init(vcpu);
2368 
2369 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2370 
2371 	init_waitqueue_head(&vcpu->arch.cpu_run);
2372 
2373 	mutex_lock(&kvm->lock);
2374 	vcore = NULL;
2375 	err = -EINVAL;
2376 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2377 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2378 			pr_devel("KVM: VCPU ID too high\n");
2379 			core = KVM_MAX_VCORES;
2380 		} else {
2381 			BUG_ON(kvm->arch.smt_mode != 1);
2382 			core = kvmppc_pack_vcpu_id(kvm, id);
2383 		}
2384 	} else {
2385 		core = id / kvm->arch.smt_mode;
2386 	}
2387 	if (core < KVM_MAX_VCORES) {
2388 		vcore = kvm->arch.vcores[core];
2389 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2390 			pr_devel("KVM: collision on id %u", id);
2391 			vcore = NULL;
2392 		} else if (!vcore) {
2393 			/*
2394 			 * Take mmu_setup_lock for mutual exclusion
2395 			 * with kvmppc_update_lpcr().
2396 			 */
2397 			err = -ENOMEM;
2398 			vcore = kvmppc_vcore_create(kvm,
2399 					id & ~(kvm->arch.smt_mode - 1));
2400 			mutex_lock(&kvm->arch.mmu_setup_lock);
2401 			kvm->arch.vcores[core] = vcore;
2402 			kvm->arch.online_vcores++;
2403 			mutex_unlock(&kvm->arch.mmu_setup_lock);
2404 		}
2405 	}
2406 	mutex_unlock(&kvm->lock);
2407 
2408 	if (!vcore)
2409 		return err;
2410 
2411 	spin_lock(&vcore->lock);
2412 	++vcore->num_threads;
2413 	spin_unlock(&vcore->lock);
2414 	vcpu->arch.vcore = vcore;
2415 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2416 	vcpu->arch.thread_cpu = -1;
2417 	vcpu->arch.prev_cpu = -1;
2418 
2419 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2420 	kvmppc_sanity_check(vcpu);
2421 
2422 	debugfs_vcpu_init(vcpu, id);
2423 
2424 	return 0;
2425 }
2426 
2427 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2428 			      unsigned long flags)
2429 {
2430 	int err;
2431 	int esmt = 0;
2432 
2433 	if (flags)
2434 		return -EINVAL;
2435 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2436 		return -EINVAL;
2437 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2438 		/*
2439 		 * On POWER8 (or POWER7), the threading mode is "strict",
2440 		 * so we pack smt_mode vcpus per vcore.
2441 		 */
2442 		if (smt_mode > threads_per_subcore)
2443 			return -EINVAL;
2444 	} else {
2445 		/*
2446 		 * On POWER9, the threading mode is "loose",
2447 		 * so each vcpu gets its own vcore.
2448 		 */
2449 		esmt = smt_mode;
2450 		smt_mode = 1;
2451 	}
2452 	mutex_lock(&kvm->lock);
2453 	err = -EBUSY;
2454 	if (!kvm->arch.online_vcores) {
2455 		kvm->arch.smt_mode = smt_mode;
2456 		kvm->arch.emul_smt_mode = esmt;
2457 		err = 0;
2458 	}
2459 	mutex_unlock(&kvm->lock);
2460 
2461 	return err;
2462 }
2463 
2464 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2465 {
2466 	if (vpa->pinned_addr)
2467 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2468 					vpa->dirty);
2469 }
2470 
2471 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2472 {
2473 	spin_lock(&vcpu->arch.vpa_update_lock);
2474 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2475 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2476 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2477 	spin_unlock(&vcpu->arch.vpa_update_lock);
2478 }
2479 
2480 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2481 {
2482 	/* Indicate we want to get back into the guest */
2483 	return 1;
2484 }
2485 
2486 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2487 {
2488 	unsigned long dec_nsec, now;
2489 
2490 	now = get_tb();
2491 	if (now > vcpu->arch.dec_expires) {
2492 		/* decrementer has already gone negative */
2493 		kvmppc_core_queue_dec(vcpu);
2494 		kvmppc_core_prepare_to_enter(vcpu);
2495 		return;
2496 	}
2497 	dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2498 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2499 	vcpu->arch.timer_running = 1;
2500 }
2501 
2502 extern int __kvmppc_vcore_entry(void);
2503 
2504 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2505 				   struct kvm_vcpu *vcpu)
2506 {
2507 	u64 now;
2508 
2509 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2510 		return;
2511 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2512 	now = mftb();
2513 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2514 		vcpu->arch.stolen_logged;
2515 	vcpu->arch.busy_preempt = now;
2516 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2517 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2518 	--vc->n_runnable;
2519 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2520 }
2521 
2522 static int kvmppc_grab_hwthread(int cpu)
2523 {
2524 	struct paca_struct *tpaca;
2525 	long timeout = 10000;
2526 
2527 	tpaca = paca_ptrs[cpu];
2528 
2529 	/* Ensure the thread won't go into the kernel if it wakes */
2530 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2531 	tpaca->kvm_hstate.kvm_vcore = NULL;
2532 	tpaca->kvm_hstate.napping = 0;
2533 	smp_wmb();
2534 	tpaca->kvm_hstate.hwthread_req = 1;
2535 
2536 	/*
2537 	 * If the thread is already executing in the kernel (e.g. handling
2538 	 * a stray interrupt), wait for it to get back to nap mode.
2539 	 * The smp_mb() is to ensure that our setting of hwthread_req
2540 	 * is visible before we look at hwthread_state, so if this
2541 	 * races with the code at system_reset_pSeries and the thread
2542 	 * misses our setting of hwthread_req, we are sure to see its
2543 	 * setting of hwthread_state, and vice versa.
2544 	 */
2545 	smp_mb();
2546 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2547 		if (--timeout <= 0) {
2548 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2549 			return -EBUSY;
2550 		}
2551 		udelay(1);
2552 	}
2553 	return 0;
2554 }
2555 
2556 static void kvmppc_release_hwthread(int cpu)
2557 {
2558 	struct paca_struct *tpaca;
2559 
2560 	tpaca = paca_ptrs[cpu];
2561 	tpaca->kvm_hstate.hwthread_req = 0;
2562 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2563 	tpaca->kvm_hstate.kvm_vcore = NULL;
2564 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2565 }
2566 
2567 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2568 {
2569 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2570 	cpumask_t *cpu_in_guest;
2571 	int i;
2572 
2573 	cpu = cpu_first_thread_sibling(cpu);
2574 	if (nested) {
2575 		cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2576 		cpu_in_guest = &nested->cpu_in_guest;
2577 	} else {
2578 		cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2579 		cpu_in_guest = &kvm->arch.cpu_in_guest;
2580 	}
2581 	/*
2582 	 * Make sure setting of bit in need_tlb_flush precedes
2583 	 * testing of cpu_in_guest bits.  The matching barrier on
2584 	 * the other side is the first smp_mb() in kvmppc_run_core().
2585 	 */
2586 	smp_mb();
2587 	for (i = 0; i < threads_per_core; ++i)
2588 		if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2589 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2590 }
2591 
2592 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2593 {
2594 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2595 	struct kvm *kvm = vcpu->kvm;
2596 	int prev_cpu;
2597 
2598 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2599 		return;
2600 
2601 	if (nested)
2602 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2603 	else
2604 		prev_cpu = vcpu->arch.prev_cpu;
2605 
2606 	/*
2607 	 * With radix, the guest can do TLB invalidations itself,
2608 	 * and it could choose to use the local form (tlbiel) if
2609 	 * it is invalidating a translation that has only ever been
2610 	 * used on one vcpu.  However, that doesn't mean it has
2611 	 * only ever been used on one physical cpu, since vcpus
2612 	 * can move around between pcpus.  To cope with this, when
2613 	 * a vcpu moves from one pcpu to another, we need to tell
2614 	 * any vcpus running on the same core as this vcpu previously
2615 	 * ran to flush the TLB.  The TLB is shared between threads,
2616 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2617 	 */
2618 	if (prev_cpu != pcpu) {
2619 		if (prev_cpu >= 0 &&
2620 		    cpu_first_thread_sibling(prev_cpu) !=
2621 		    cpu_first_thread_sibling(pcpu))
2622 			radix_flush_cpu(kvm, prev_cpu, vcpu);
2623 		if (nested)
2624 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2625 		else
2626 			vcpu->arch.prev_cpu = pcpu;
2627 	}
2628 }
2629 
2630 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2631 {
2632 	int cpu;
2633 	struct paca_struct *tpaca;
2634 	struct kvm *kvm = vc->kvm;
2635 
2636 	cpu = vc->pcpu;
2637 	if (vcpu) {
2638 		if (vcpu->arch.timer_running) {
2639 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2640 			vcpu->arch.timer_running = 0;
2641 		}
2642 		cpu += vcpu->arch.ptid;
2643 		vcpu->cpu = vc->pcpu;
2644 		vcpu->arch.thread_cpu = cpu;
2645 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2646 	}
2647 	tpaca = paca_ptrs[cpu];
2648 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2649 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2650 	tpaca->kvm_hstate.fake_suspend = 0;
2651 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2652 	smp_wmb();
2653 	tpaca->kvm_hstate.kvm_vcore = vc;
2654 	if (cpu != smp_processor_id())
2655 		kvmppc_ipi_thread(cpu);
2656 }
2657 
2658 static void kvmppc_wait_for_nap(int n_threads)
2659 {
2660 	int cpu = smp_processor_id();
2661 	int i, loops;
2662 
2663 	if (n_threads <= 1)
2664 		return;
2665 	for (loops = 0; loops < 1000000; ++loops) {
2666 		/*
2667 		 * Check if all threads are finished.
2668 		 * We set the vcore pointer when starting a thread
2669 		 * and the thread clears it when finished, so we look
2670 		 * for any threads that still have a non-NULL vcore ptr.
2671 		 */
2672 		for (i = 1; i < n_threads; ++i)
2673 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2674 				break;
2675 		if (i == n_threads) {
2676 			HMT_medium();
2677 			return;
2678 		}
2679 		HMT_low();
2680 	}
2681 	HMT_medium();
2682 	for (i = 1; i < n_threads; ++i)
2683 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2684 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2685 }
2686 
2687 /*
2688  * Check that we are on thread 0 and that any other threads in
2689  * this core are off-line.  Then grab the threads so they can't
2690  * enter the kernel.
2691  */
2692 static int on_primary_thread(void)
2693 {
2694 	int cpu = smp_processor_id();
2695 	int thr;
2696 
2697 	/* Are we on a primary subcore? */
2698 	if (cpu_thread_in_subcore(cpu))
2699 		return 0;
2700 
2701 	thr = 0;
2702 	while (++thr < threads_per_subcore)
2703 		if (cpu_online(cpu + thr))
2704 			return 0;
2705 
2706 	/* Grab all hw threads so they can't go into the kernel */
2707 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2708 		if (kvmppc_grab_hwthread(cpu + thr)) {
2709 			/* Couldn't grab one; let the others go */
2710 			do {
2711 				kvmppc_release_hwthread(cpu + thr);
2712 			} while (--thr > 0);
2713 			return 0;
2714 		}
2715 	}
2716 	return 1;
2717 }
2718 
2719 /*
2720  * A list of virtual cores for each physical CPU.
2721  * These are vcores that could run but their runner VCPU tasks are
2722  * (or may be) preempted.
2723  */
2724 struct preempted_vcore_list {
2725 	struct list_head	list;
2726 	spinlock_t		lock;
2727 };
2728 
2729 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2730 
2731 static void init_vcore_lists(void)
2732 {
2733 	int cpu;
2734 
2735 	for_each_possible_cpu(cpu) {
2736 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2737 		spin_lock_init(&lp->lock);
2738 		INIT_LIST_HEAD(&lp->list);
2739 	}
2740 }
2741 
2742 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2743 {
2744 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2745 
2746 	vc->vcore_state = VCORE_PREEMPT;
2747 	vc->pcpu = smp_processor_id();
2748 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2749 		spin_lock(&lp->lock);
2750 		list_add_tail(&vc->preempt_list, &lp->list);
2751 		spin_unlock(&lp->lock);
2752 	}
2753 
2754 	/* Start accumulating stolen time */
2755 	kvmppc_core_start_stolen(vc);
2756 }
2757 
2758 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2759 {
2760 	struct preempted_vcore_list *lp;
2761 
2762 	kvmppc_core_end_stolen(vc);
2763 	if (!list_empty(&vc->preempt_list)) {
2764 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2765 		spin_lock(&lp->lock);
2766 		list_del_init(&vc->preempt_list);
2767 		spin_unlock(&lp->lock);
2768 	}
2769 	vc->vcore_state = VCORE_INACTIVE;
2770 }
2771 
2772 /*
2773  * This stores information about the virtual cores currently
2774  * assigned to a physical core.
2775  */
2776 struct core_info {
2777 	int		n_subcores;
2778 	int		max_subcore_threads;
2779 	int		total_threads;
2780 	int		subcore_threads[MAX_SUBCORES];
2781 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2782 };
2783 
2784 /*
2785  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2786  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2787  */
2788 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2789 
2790 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2791 {
2792 	memset(cip, 0, sizeof(*cip));
2793 	cip->n_subcores = 1;
2794 	cip->max_subcore_threads = vc->num_threads;
2795 	cip->total_threads = vc->num_threads;
2796 	cip->subcore_threads[0] = vc->num_threads;
2797 	cip->vc[0] = vc;
2798 }
2799 
2800 static bool subcore_config_ok(int n_subcores, int n_threads)
2801 {
2802 	/*
2803 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2804 	 * split-core mode, with one thread per subcore.
2805 	 */
2806 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2807 		return n_subcores <= 4 && n_threads == 1;
2808 
2809 	/* On POWER8, can only dynamically split if unsplit to begin with */
2810 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2811 		return false;
2812 	if (n_subcores > MAX_SUBCORES)
2813 		return false;
2814 	if (n_subcores > 1) {
2815 		if (!(dynamic_mt_modes & 2))
2816 			n_subcores = 4;
2817 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2818 			return false;
2819 	}
2820 
2821 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2822 }
2823 
2824 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2825 {
2826 	vc->entry_exit_map = 0;
2827 	vc->in_guest = 0;
2828 	vc->napping_threads = 0;
2829 	vc->conferring_threads = 0;
2830 	vc->tb_offset_applied = 0;
2831 }
2832 
2833 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2834 {
2835 	int n_threads = vc->num_threads;
2836 	int sub;
2837 
2838 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2839 		return false;
2840 
2841 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
2842 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2843 		return false;
2844 
2845 	/* Some POWER9 chips require all threads to be in the same MMU mode */
2846 	if (no_mixing_hpt_and_radix &&
2847 	    kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2848 		return false;
2849 
2850 	if (n_threads < cip->max_subcore_threads)
2851 		n_threads = cip->max_subcore_threads;
2852 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2853 		return false;
2854 	cip->max_subcore_threads = n_threads;
2855 
2856 	sub = cip->n_subcores;
2857 	++cip->n_subcores;
2858 	cip->total_threads += vc->num_threads;
2859 	cip->subcore_threads[sub] = vc->num_threads;
2860 	cip->vc[sub] = vc;
2861 	init_vcore_to_run(vc);
2862 	list_del_init(&vc->preempt_list);
2863 
2864 	return true;
2865 }
2866 
2867 /*
2868  * Work out whether it is possible to piggyback the execution of
2869  * vcore *pvc onto the execution of the other vcores described in *cip.
2870  */
2871 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2872 			  int target_threads)
2873 {
2874 	if (cip->total_threads + pvc->num_threads > target_threads)
2875 		return false;
2876 
2877 	return can_dynamic_split(pvc, cip);
2878 }
2879 
2880 static void prepare_threads(struct kvmppc_vcore *vc)
2881 {
2882 	int i;
2883 	struct kvm_vcpu *vcpu;
2884 
2885 	for_each_runnable_thread(i, vcpu, vc) {
2886 		if (signal_pending(vcpu->arch.run_task))
2887 			vcpu->arch.ret = -EINTR;
2888 		else if (vcpu->arch.vpa.update_pending ||
2889 			 vcpu->arch.slb_shadow.update_pending ||
2890 			 vcpu->arch.dtl.update_pending)
2891 			vcpu->arch.ret = RESUME_GUEST;
2892 		else
2893 			continue;
2894 		kvmppc_remove_runnable(vc, vcpu);
2895 		wake_up(&vcpu->arch.cpu_run);
2896 	}
2897 }
2898 
2899 static void collect_piggybacks(struct core_info *cip, int target_threads)
2900 {
2901 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2902 	struct kvmppc_vcore *pvc, *vcnext;
2903 
2904 	spin_lock(&lp->lock);
2905 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2906 		if (!spin_trylock(&pvc->lock))
2907 			continue;
2908 		prepare_threads(pvc);
2909 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2910 			list_del_init(&pvc->preempt_list);
2911 			if (pvc->runner == NULL) {
2912 				pvc->vcore_state = VCORE_INACTIVE;
2913 				kvmppc_core_end_stolen(pvc);
2914 			}
2915 			spin_unlock(&pvc->lock);
2916 			continue;
2917 		}
2918 		if (!can_piggyback(pvc, cip, target_threads)) {
2919 			spin_unlock(&pvc->lock);
2920 			continue;
2921 		}
2922 		kvmppc_core_end_stolen(pvc);
2923 		pvc->vcore_state = VCORE_PIGGYBACK;
2924 		if (cip->total_threads >= target_threads)
2925 			break;
2926 	}
2927 	spin_unlock(&lp->lock);
2928 }
2929 
2930 static bool recheck_signals_and_mmu(struct core_info *cip)
2931 {
2932 	int sub, i;
2933 	struct kvm_vcpu *vcpu;
2934 	struct kvmppc_vcore *vc;
2935 
2936 	for (sub = 0; sub < cip->n_subcores; ++sub) {
2937 		vc = cip->vc[sub];
2938 		if (!vc->kvm->arch.mmu_ready)
2939 			return true;
2940 		for_each_runnable_thread(i, vcpu, vc)
2941 			if (signal_pending(vcpu->arch.run_task))
2942 				return true;
2943 	}
2944 	return false;
2945 }
2946 
2947 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2948 {
2949 	int still_running = 0, i;
2950 	u64 now;
2951 	long ret;
2952 	struct kvm_vcpu *vcpu;
2953 
2954 	spin_lock(&vc->lock);
2955 	now = get_tb();
2956 	for_each_runnable_thread(i, vcpu, vc) {
2957 		/*
2958 		 * It's safe to unlock the vcore in the loop here, because
2959 		 * for_each_runnable_thread() is safe against removal of
2960 		 * the vcpu, and the vcore state is VCORE_EXITING here,
2961 		 * so any vcpus becoming runnable will have their arch.trap
2962 		 * set to zero and can't actually run in the guest.
2963 		 */
2964 		spin_unlock(&vc->lock);
2965 		/* cancel pending dec exception if dec is positive */
2966 		if (now < vcpu->arch.dec_expires &&
2967 		    kvmppc_core_pending_dec(vcpu))
2968 			kvmppc_core_dequeue_dec(vcpu);
2969 
2970 		trace_kvm_guest_exit(vcpu);
2971 
2972 		ret = RESUME_GUEST;
2973 		if (vcpu->arch.trap)
2974 			ret = kvmppc_handle_exit_hv(vcpu,
2975 						    vcpu->arch.run_task);
2976 
2977 		vcpu->arch.ret = ret;
2978 		vcpu->arch.trap = 0;
2979 
2980 		spin_lock(&vc->lock);
2981 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2982 			if (vcpu->arch.pending_exceptions)
2983 				kvmppc_core_prepare_to_enter(vcpu);
2984 			if (vcpu->arch.ceded)
2985 				kvmppc_set_timer(vcpu);
2986 			else
2987 				++still_running;
2988 		} else {
2989 			kvmppc_remove_runnable(vc, vcpu);
2990 			wake_up(&vcpu->arch.cpu_run);
2991 		}
2992 	}
2993 	if (!is_master) {
2994 		if (still_running > 0) {
2995 			kvmppc_vcore_preempt(vc);
2996 		} else if (vc->runner) {
2997 			vc->vcore_state = VCORE_PREEMPT;
2998 			kvmppc_core_start_stolen(vc);
2999 		} else {
3000 			vc->vcore_state = VCORE_INACTIVE;
3001 		}
3002 		if (vc->n_runnable > 0 && vc->runner == NULL) {
3003 			/* make sure there's a candidate runner awake */
3004 			i = -1;
3005 			vcpu = next_runnable_thread(vc, &i);
3006 			wake_up(&vcpu->arch.cpu_run);
3007 		}
3008 	}
3009 	spin_unlock(&vc->lock);
3010 }
3011 
3012 /*
3013  * Clear core from the list of active host cores as we are about to
3014  * enter the guest. Only do this if it is the primary thread of the
3015  * core (not if a subcore) that is entering the guest.
3016  */
3017 static inline int kvmppc_clear_host_core(unsigned int cpu)
3018 {
3019 	int core;
3020 
3021 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3022 		return 0;
3023 	/*
3024 	 * Memory barrier can be omitted here as we will do a smp_wmb()
3025 	 * later in kvmppc_start_thread and we need ensure that state is
3026 	 * visible to other CPUs only after we enter guest.
3027 	 */
3028 	core = cpu >> threads_shift;
3029 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3030 	return 0;
3031 }
3032 
3033 /*
3034  * Advertise this core as an active host core since we exited the guest
3035  * Only need to do this if it is the primary thread of the core that is
3036  * exiting.
3037  */
3038 static inline int kvmppc_set_host_core(unsigned int cpu)
3039 {
3040 	int core;
3041 
3042 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3043 		return 0;
3044 
3045 	/*
3046 	 * Memory barrier can be omitted here because we do a spin_unlock
3047 	 * immediately after this which provides the memory barrier.
3048 	 */
3049 	core = cpu >> threads_shift;
3050 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3051 	return 0;
3052 }
3053 
3054 static void set_irq_happened(int trap)
3055 {
3056 	switch (trap) {
3057 	case BOOK3S_INTERRUPT_EXTERNAL:
3058 		local_paca->irq_happened |= PACA_IRQ_EE;
3059 		break;
3060 	case BOOK3S_INTERRUPT_H_DOORBELL:
3061 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3062 		break;
3063 	case BOOK3S_INTERRUPT_HMI:
3064 		local_paca->irq_happened |= PACA_IRQ_HMI;
3065 		break;
3066 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3067 		replay_system_reset();
3068 		break;
3069 	}
3070 }
3071 
3072 /*
3073  * Run a set of guest threads on a physical core.
3074  * Called with vc->lock held.
3075  */
3076 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3077 {
3078 	struct kvm_vcpu *vcpu;
3079 	int i;
3080 	int srcu_idx;
3081 	struct core_info core_info;
3082 	struct kvmppc_vcore *pvc;
3083 	struct kvm_split_mode split_info, *sip;
3084 	int split, subcore_size, active;
3085 	int sub;
3086 	bool thr0_done;
3087 	unsigned long cmd_bit, stat_bit;
3088 	int pcpu, thr;
3089 	int target_threads;
3090 	int controlled_threads;
3091 	int trap;
3092 	bool is_power8;
3093 	bool hpt_on_radix;
3094 
3095 	/*
3096 	 * Remove from the list any threads that have a signal pending
3097 	 * or need a VPA update done
3098 	 */
3099 	prepare_threads(vc);
3100 
3101 	/* if the runner is no longer runnable, let the caller pick a new one */
3102 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3103 		return;
3104 
3105 	/*
3106 	 * Initialize *vc.
3107 	 */
3108 	init_vcore_to_run(vc);
3109 	vc->preempt_tb = TB_NIL;
3110 
3111 	/*
3112 	 * Number of threads that we will be controlling: the same as
3113 	 * the number of threads per subcore, except on POWER9,
3114 	 * where it's 1 because the threads are (mostly) independent.
3115 	 */
3116 	controlled_threads = threads_per_vcore(vc->kvm);
3117 
3118 	/*
3119 	 * Make sure we are running on primary threads, and that secondary
3120 	 * threads are offline.  Also check if the number of threads in this
3121 	 * guest are greater than the current system threads per guest.
3122 	 * On POWER9, we need to be not in independent-threads mode if
3123 	 * this is a HPT guest on a radix host machine where the
3124 	 * CPU threads may not be in different MMU modes.
3125 	 */
3126 	hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3127 		!kvm_is_radix(vc->kvm);
3128 	if (((controlled_threads > 1) &&
3129 	     ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3130 	    (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3131 		for_each_runnable_thread(i, vcpu, vc) {
3132 			vcpu->arch.ret = -EBUSY;
3133 			kvmppc_remove_runnable(vc, vcpu);
3134 			wake_up(&vcpu->arch.cpu_run);
3135 		}
3136 		goto out;
3137 	}
3138 
3139 	/*
3140 	 * See if we could run any other vcores on the physical core
3141 	 * along with this one.
3142 	 */
3143 	init_core_info(&core_info, vc);
3144 	pcpu = smp_processor_id();
3145 	target_threads = controlled_threads;
3146 	if (target_smt_mode && target_smt_mode < target_threads)
3147 		target_threads = target_smt_mode;
3148 	if (vc->num_threads < target_threads)
3149 		collect_piggybacks(&core_info, target_threads);
3150 
3151 	/*
3152 	 * On radix, arrange for TLB flushing if necessary.
3153 	 * This has to be done before disabling interrupts since
3154 	 * it uses smp_call_function().
3155 	 */
3156 	pcpu = smp_processor_id();
3157 	if (kvm_is_radix(vc->kvm)) {
3158 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3159 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3160 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3161 	}
3162 
3163 	/*
3164 	 * Hard-disable interrupts, and check resched flag and signals.
3165 	 * If we need to reschedule or deliver a signal, clean up
3166 	 * and return without going into the guest(s).
3167 	 * If the mmu_ready flag has been cleared, don't go into the
3168 	 * guest because that means a HPT resize operation is in progress.
3169 	 */
3170 	local_irq_disable();
3171 	hard_irq_disable();
3172 	if (lazy_irq_pending() || need_resched() ||
3173 	    recheck_signals_and_mmu(&core_info)) {
3174 		local_irq_enable();
3175 		vc->vcore_state = VCORE_INACTIVE;
3176 		/* Unlock all except the primary vcore */
3177 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3178 			pvc = core_info.vc[sub];
3179 			/* Put back on to the preempted vcores list */
3180 			kvmppc_vcore_preempt(pvc);
3181 			spin_unlock(&pvc->lock);
3182 		}
3183 		for (i = 0; i < controlled_threads; ++i)
3184 			kvmppc_release_hwthread(pcpu + i);
3185 		return;
3186 	}
3187 
3188 	kvmppc_clear_host_core(pcpu);
3189 
3190 	/* Decide on micro-threading (split-core) mode */
3191 	subcore_size = threads_per_subcore;
3192 	cmd_bit = stat_bit = 0;
3193 	split = core_info.n_subcores;
3194 	sip = NULL;
3195 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3196 		&& !cpu_has_feature(CPU_FTR_ARCH_300);
3197 
3198 	if (split > 1 || hpt_on_radix) {
3199 		sip = &split_info;
3200 		memset(&split_info, 0, sizeof(split_info));
3201 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3202 			split_info.vc[sub] = core_info.vc[sub];
3203 
3204 		if (is_power8) {
3205 			if (split == 2 && (dynamic_mt_modes & 2)) {
3206 				cmd_bit = HID0_POWER8_1TO2LPAR;
3207 				stat_bit = HID0_POWER8_2LPARMODE;
3208 			} else {
3209 				split = 4;
3210 				cmd_bit = HID0_POWER8_1TO4LPAR;
3211 				stat_bit = HID0_POWER8_4LPARMODE;
3212 			}
3213 			subcore_size = MAX_SMT_THREADS / split;
3214 			split_info.rpr = mfspr(SPRN_RPR);
3215 			split_info.pmmar = mfspr(SPRN_PMMAR);
3216 			split_info.ldbar = mfspr(SPRN_LDBAR);
3217 			split_info.subcore_size = subcore_size;
3218 		} else {
3219 			split_info.subcore_size = 1;
3220 			if (hpt_on_radix) {
3221 				/* Use the split_info for LPCR/LPIDR changes */
3222 				split_info.lpcr_req = vc->lpcr;
3223 				split_info.lpidr_req = vc->kvm->arch.lpid;
3224 				split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3225 				split_info.do_set = 1;
3226 			}
3227 		}
3228 
3229 		/* order writes to split_info before kvm_split_mode pointer */
3230 		smp_wmb();
3231 	}
3232 
3233 	for (thr = 0; thr < controlled_threads; ++thr) {
3234 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3235 
3236 		paca->kvm_hstate.tid = thr;
3237 		paca->kvm_hstate.napping = 0;
3238 		paca->kvm_hstate.kvm_split_mode = sip;
3239 	}
3240 
3241 	/* Initiate micro-threading (split-core) on POWER8 if required */
3242 	if (cmd_bit) {
3243 		unsigned long hid0 = mfspr(SPRN_HID0);
3244 
3245 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3246 		mb();
3247 		mtspr(SPRN_HID0, hid0);
3248 		isync();
3249 		for (;;) {
3250 			hid0 = mfspr(SPRN_HID0);
3251 			if (hid0 & stat_bit)
3252 				break;
3253 			cpu_relax();
3254 		}
3255 	}
3256 
3257 	/*
3258 	 * On POWER8, set RWMR register.
3259 	 * Since it only affects PURR and SPURR, it doesn't affect
3260 	 * the host, so we don't save/restore the host value.
3261 	 */
3262 	if (is_power8) {
3263 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3264 		int n_online = atomic_read(&vc->online_count);
3265 
3266 		/*
3267 		 * Use the 8-thread value if we're doing split-core
3268 		 * or if the vcore's online count looks bogus.
3269 		 */
3270 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3271 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3272 			rwmr_val = p8_rwmr_values[n_online];
3273 		mtspr(SPRN_RWMR, rwmr_val);
3274 	}
3275 
3276 	/* Start all the threads */
3277 	active = 0;
3278 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3279 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3280 		thr0_done = false;
3281 		active |= 1 << thr;
3282 		pvc = core_info.vc[sub];
3283 		pvc->pcpu = pcpu + thr;
3284 		for_each_runnable_thread(i, vcpu, pvc) {
3285 			kvmppc_start_thread(vcpu, pvc);
3286 			kvmppc_create_dtl_entry(vcpu, pvc);
3287 			trace_kvm_guest_enter(vcpu);
3288 			if (!vcpu->arch.ptid)
3289 				thr0_done = true;
3290 			active |= 1 << (thr + vcpu->arch.ptid);
3291 		}
3292 		/*
3293 		 * We need to start the first thread of each subcore
3294 		 * even if it doesn't have a vcpu.
3295 		 */
3296 		if (!thr0_done)
3297 			kvmppc_start_thread(NULL, pvc);
3298 	}
3299 
3300 	/*
3301 	 * Ensure that split_info.do_nap is set after setting
3302 	 * the vcore pointer in the PACA of the secondaries.
3303 	 */
3304 	smp_mb();
3305 
3306 	/*
3307 	 * When doing micro-threading, poke the inactive threads as well.
3308 	 * This gets them to the nap instruction after kvm_do_nap,
3309 	 * which reduces the time taken to unsplit later.
3310 	 * For POWER9 HPT guest on radix host, we need all the secondary
3311 	 * threads woken up so they can do the LPCR/LPIDR change.
3312 	 */
3313 	if (cmd_bit || hpt_on_radix) {
3314 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3315 		for (thr = 1; thr < threads_per_subcore; ++thr)
3316 			if (!(active & (1 << thr)))
3317 				kvmppc_ipi_thread(pcpu + thr);
3318 	}
3319 
3320 	vc->vcore_state = VCORE_RUNNING;
3321 	preempt_disable();
3322 
3323 	trace_kvmppc_run_core(vc, 0);
3324 
3325 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3326 		spin_unlock(&core_info.vc[sub]->lock);
3327 
3328 	guest_enter_irqoff();
3329 
3330 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3331 
3332 	this_cpu_disable_ftrace();
3333 
3334 	/*
3335 	 * Interrupts will be enabled once we get into the guest,
3336 	 * so tell lockdep that we're about to enable interrupts.
3337 	 */
3338 	trace_hardirqs_on();
3339 
3340 	trap = __kvmppc_vcore_entry();
3341 
3342 	trace_hardirqs_off();
3343 
3344 	this_cpu_enable_ftrace();
3345 
3346 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3347 
3348 	set_irq_happened(trap);
3349 
3350 	spin_lock(&vc->lock);
3351 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3352 	vc->vcore_state = VCORE_EXITING;
3353 
3354 	/* wait for secondary threads to finish writing their state to memory */
3355 	kvmppc_wait_for_nap(controlled_threads);
3356 
3357 	/* Return to whole-core mode if we split the core earlier */
3358 	if (cmd_bit) {
3359 		unsigned long hid0 = mfspr(SPRN_HID0);
3360 		unsigned long loops = 0;
3361 
3362 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3363 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3364 		mb();
3365 		mtspr(SPRN_HID0, hid0);
3366 		isync();
3367 		for (;;) {
3368 			hid0 = mfspr(SPRN_HID0);
3369 			if (!(hid0 & stat_bit))
3370 				break;
3371 			cpu_relax();
3372 			++loops;
3373 		}
3374 	} else if (hpt_on_radix) {
3375 		/* Wait for all threads to have seen final sync */
3376 		for (thr = 1; thr < controlled_threads; ++thr) {
3377 			struct paca_struct *paca = paca_ptrs[pcpu + thr];
3378 
3379 			while (paca->kvm_hstate.kvm_split_mode) {
3380 				HMT_low();
3381 				barrier();
3382 			}
3383 			HMT_medium();
3384 		}
3385 	}
3386 	split_info.do_nap = 0;
3387 
3388 	kvmppc_set_host_core(pcpu);
3389 
3390 	local_irq_enable();
3391 	guest_exit();
3392 
3393 	/* Let secondaries go back to the offline loop */
3394 	for (i = 0; i < controlled_threads; ++i) {
3395 		kvmppc_release_hwthread(pcpu + i);
3396 		if (sip && sip->napped[i])
3397 			kvmppc_ipi_thread(pcpu + i);
3398 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3399 	}
3400 
3401 	spin_unlock(&vc->lock);
3402 
3403 	/* make sure updates to secondary vcpu structs are visible now */
3404 	smp_mb();
3405 
3406 	preempt_enable();
3407 
3408 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3409 		pvc = core_info.vc[sub];
3410 		post_guest_process(pvc, pvc == vc);
3411 	}
3412 
3413 	spin_lock(&vc->lock);
3414 
3415  out:
3416 	vc->vcore_state = VCORE_INACTIVE;
3417 	trace_kvmppc_run_core(vc, 1);
3418 }
3419 
3420 /*
3421  * Load up hypervisor-mode registers on P9.
3422  */
3423 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3424 				     unsigned long lpcr)
3425 {
3426 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3427 	s64 hdec;
3428 	u64 tb, purr, spurr;
3429 	int trap;
3430 	unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3431 	unsigned long host_ciabr = mfspr(SPRN_CIABR);
3432 	unsigned long host_dawr = mfspr(SPRN_DAWR0);
3433 	unsigned long host_dawrx = mfspr(SPRN_DAWRX0);
3434 	unsigned long host_psscr = mfspr(SPRN_PSSCR);
3435 	unsigned long host_pidr = mfspr(SPRN_PID);
3436 
3437 	hdec = time_limit - mftb();
3438 	if (hdec < 0)
3439 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3440 	mtspr(SPRN_HDEC, hdec);
3441 
3442 	if (vc->tb_offset) {
3443 		u64 new_tb = mftb() + vc->tb_offset;
3444 		mtspr(SPRN_TBU40, new_tb);
3445 		tb = mftb();
3446 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3447 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3448 		vc->tb_offset_applied = vc->tb_offset;
3449 	}
3450 
3451 	if (vc->pcr)
3452 		mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3453 	mtspr(SPRN_DPDES, vc->dpdes);
3454 	mtspr(SPRN_VTB, vc->vtb);
3455 
3456 	local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3457 	local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3458 	mtspr(SPRN_PURR, vcpu->arch.purr);
3459 	mtspr(SPRN_SPURR, vcpu->arch.spurr);
3460 
3461 	if (dawr_enabled()) {
3462 		mtspr(SPRN_DAWR0, vcpu->arch.dawr);
3463 		mtspr(SPRN_DAWRX0, vcpu->arch.dawrx);
3464 	}
3465 	mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3466 	mtspr(SPRN_IC, vcpu->arch.ic);
3467 	mtspr(SPRN_PID, vcpu->arch.pid);
3468 
3469 	mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3470 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3471 
3472 	mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3473 
3474 	mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3475 	mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3476 	mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3477 	mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3478 
3479 	mtspr(SPRN_AMOR, ~0UL);
3480 
3481 	mtspr(SPRN_LPCR, lpcr);
3482 	isync();
3483 
3484 	kvmppc_xive_push_vcpu(vcpu);
3485 
3486 	mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3487 	mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3488 
3489 	trap = __kvmhv_vcpu_entry_p9(vcpu);
3490 
3491 	/* Advance host PURR/SPURR by the amount used by guest */
3492 	purr = mfspr(SPRN_PURR);
3493 	spurr = mfspr(SPRN_SPURR);
3494 	mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3495 	      purr - vcpu->arch.purr);
3496 	mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3497 	      spurr - vcpu->arch.spurr);
3498 	vcpu->arch.purr = purr;
3499 	vcpu->arch.spurr = spurr;
3500 
3501 	vcpu->arch.ic = mfspr(SPRN_IC);
3502 	vcpu->arch.pid = mfspr(SPRN_PID);
3503 	vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3504 
3505 	vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3506 	vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3507 	vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3508 	vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3509 
3510 	/* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3511 	mtspr(SPRN_PSSCR, host_psscr |
3512 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3513 	mtspr(SPRN_HFSCR, host_hfscr);
3514 	mtspr(SPRN_CIABR, host_ciabr);
3515 	mtspr(SPRN_DAWR0, host_dawr);
3516 	mtspr(SPRN_DAWRX0, host_dawrx);
3517 	mtspr(SPRN_PID, host_pidr);
3518 
3519 	/*
3520 	 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3521 	 * case we interrupted the guest between a tlbie and a ptesync.
3522 	 */
3523 	asm volatile("eieio; tlbsync; ptesync");
3524 
3525 	mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);	/* restore host LPID */
3526 	isync();
3527 
3528 	vc->dpdes = mfspr(SPRN_DPDES);
3529 	vc->vtb = mfspr(SPRN_VTB);
3530 	mtspr(SPRN_DPDES, 0);
3531 	if (vc->pcr)
3532 		mtspr(SPRN_PCR, PCR_MASK);
3533 
3534 	if (vc->tb_offset_applied) {
3535 		u64 new_tb = mftb() - vc->tb_offset_applied;
3536 		mtspr(SPRN_TBU40, new_tb);
3537 		tb = mftb();
3538 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3539 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3540 		vc->tb_offset_applied = 0;
3541 	}
3542 
3543 	mtspr(SPRN_HDEC, 0x7fffffff);
3544 	mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3545 
3546 	return trap;
3547 }
3548 
3549 /*
3550  * Virtual-mode guest entry for POWER9 and later when the host and
3551  * guest are both using the radix MMU.  The LPIDR has already been set.
3552  */
3553 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3554 			 unsigned long lpcr)
3555 {
3556 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3557 	unsigned long host_dscr = mfspr(SPRN_DSCR);
3558 	unsigned long host_tidr = mfspr(SPRN_TIDR);
3559 	unsigned long host_iamr = mfspr(SPRN_IAMR);
3560 	unsigned long host_amr = mfspr(SPRN_AMR);
3561 	s64 dec;
3562 	u64 tb;
3563 	int trap, save_pmu;
3564 
3565 	dec = mfspr(SPRN_DEC);
3566 	tb = mftb();
3567 	if (dec < 512)
3568 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3569 	local_paca->kvm_hstate.dec_expires = dec + tb;
3570 	if (local_paca->kvm_hstate.dec_expires < time_limit)
3571 		time_limit = local_paca->kvm_hstate.dec_expires;
3572 
3573 	vcpu->arch.ceded = 0;
3574 
3575 	kvmhv_save_host_pmu();		/* saves it to PACA kvm_hstate */
3576 
3577 	kvmppc_subcore_enter_guest();
3578 
3579 	vc->entry_exit_map = 1;
3580 	vc->in_guest = 1;
3581 
3582 	if (vcpu->arch.vpa.pinned_addr) {
3583 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3584 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3585 		lp->yield_count = cpu_to_be32(yield_count);
3586 		vcpu->arch.vpa.dirty = 1;
3587 	}
3588 
3589 	if (cpu_has_feature(CPU_FTR_TM) ||
3590 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3591 		kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3592 
3593 	kvmhv_load_guest_pmu(vcpu);
3594 
3595 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3596 	load_fp_state(&vcpu->arch.fp);
3597 #ifdef CONFIG_ALTIVEC
3598 	load_vr_state(&vcpu->arch.vr);
3599 #endif
3600 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3601 
3602 	mtspr(SPRN_DSCR, vcpu->arch.dscr);
3603 	mtspr(SPRN_IAMR, vcpu->arch.iamr);
3604 	mtspr(SPRN_PSPB, vcpu->arch.pspb);
3605 	mtspr(SPRN_FSCR, vcpu->arch.fscr);
3606 	mtspr(SPRN_TAR, vcpu->arch.tar);
3607 	mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3608 	mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3609 	mtspr(SPRN_BESCR, vcpu->arch.bescr);
3610 	mtspr(SPRN_WORT, vcpu->arch.wort);
3611 	mtspr(SPRN_TIDR, vcpu->arch.tid);
3612 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3613 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3614 	mtspr(SPRN_AMR, vcpu->arch.amr);
3615 	mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3616 
3617 	if (!(vcpu->arch.ctrl & 1))
3618 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3619 
3620 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3621 
3622 	if (kvmhv_on_pseries()) {
3623 		/*
3624 		 * We need to save and restore the guest visible part of the
3625 		 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3626 		 * doesn't do this for us. Note only required if pseries since
3627 		 * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3628 		 */
3629 		unsigned long host_psscr;
3630 		/* call our hypervisor to load up HV regs and go */
3631 		struct hv_guest_state hvregs;
3632 
3633 		host_psscr = mfspr(SPRN_PSSCR_PR);
3634 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3635 		kvmhv_save_hv_regs(vcpu, &hvregs);
3636 		hvregs.lpcr = lpcr;
3637 		vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3638 		hvregs.version = HV_GUEST_STATE_VERSION;
3639 		if (vcpu->arch.nested) {
3640 			hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3641 			hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3642 		} else {
3643 			hvregs.lpid = vcpu->kvm->arch.lpid;
3644 			hvregs.vcpu_token = vcpu->vcpu_id;
3645 		}
3646 		hvregs.hdec_expiry = time_limit;
3647 		trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3648 					  __pa(&vcpu->arch.regs));
3649 		kvmhv_restore_hv_return_state(vcpu, &hvregs);
3650 		vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3651 		vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3652 		vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3653 		vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3654 		mtspr(SPRN_PSSCR_PR, host_psscr);
3655 
3656 		/* H_CEDE has to be handled now, not later */
3657 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3658 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3659 			kvmppc_nested_cede(vcpu);
3660 			kvmppc_set_gpr(vcpu, 3, 0);
3661 			trap = 0;
3662 		}
3663 	} else {
3664 		trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3665 	}
3666 
3667 	vcpu->arch.slb_max = 0;
3668 	dec = mfspr(SPRN_DEC);
3669 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3670 		dec = (s32) dec;
3671 	tb = mftb();
3672 	vcpu->arch.dec_expires = dec + tb;
3673 	vcpu->cpu = -1;
3674 	vcpu->arch.thread_cpu = -1;
3675 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3676 
3677 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
3678 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
3679 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
3680 	vcpu->arch.tar = mfspr(SPRN_TAR);
3681 	vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3682 	vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3683 	vcpu->arch.bescr = mfspr(SPRN_BESCR);
3684 	vcpu->arch.wort = mfspr(SPRN_WORT);
3685 	vcpu->arch.tid = mfspr(SPRN_TIDR);
3686 	vcpu->arch.amr = mfspr(SPRN_AMR);
3687 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3688 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
3689 
3690 	mtspr(SPRN_PSPB, 0);
3691 	mtspr(SPRN_WORT, 0);
3692 	mtspr(SPRN_UAMOR, 0);
3693 	mtspr(SPRN_DSCR, host_dscr);
3694 	mtspr(SPRN_TIDR, host_tidr);
3695 	mtspr(SPRN_IAMR, host_iamr);
3696 	mtspr(SPRN_PSPB, 0);
3697 
3698 	if (host_amr != vcpu->arch.amr)
3699 		mtspr(SPRN_AMR, host_amr);
3700 
3701 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3702 	store_fp_state(&vcpu->arch.fp);
3703 #ifdef CONFIG_ALTIVEC
3704 	store_vr_state(&vcpu->arch.vr);
3705 #endif
3706 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3707 
3708 	if (cpu_has_feature(CPU_FTR_TM) ||
3709 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3710 		kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3711 
3712 	save_pmu = 1;
3713 	if (vcpu->arch.vpa.pinned_addr) {
3714 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3715 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3716 		lp->yield_count = cpu_to_be32(yield_count);
3717 		vcpu->arch.vpa.dirty = 1;
3718 		save_pmu = lp->pmcregs_in_use;
3719 	}
3720 	/* Must save pmu if this guest is capable of running nested guests */
3721 	save_pmu |= nesting_enabled(vcpu->kvm);
3722 
3723 	kvmhv_save_guest_pmu(vcpu, save_pmu);
3724 
3725 	vc->entry_exit_map = 0x101;
3726 	vc->in_guest = 0;
3727 
3728 	mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3729 	mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3730 
3731 	kvmhv_load_host_pmu();
3732 
3733 	kvmppc_subcore_exit_guest();
3734 
3735 	return trap;
3736 }
3737 
3738 /*
3739  * Wait for some other vcpu thread to execute us, and
3740  * wake us up when we need to handle something in the host.
3741  */
3742 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3743 				 struct kvm_vcpu *vcpu, int wait_state)
3744 {
3745 	DEFINE_WAIT(wait);
3746 
3747 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3748 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3749 		spin_unlock(&vc->lock);
3750 		schedule();
3751 		spin_lock(&vc->lock);
3752 	}
3753 	finish_wait(&vcpu->arch.cpu_run, &wait);
3754 }
3755 
3756 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3757 {
3758 	if (!halt_poll_ns_grow)
3759 		return;
3760 
3761 	vc->halt_poll_ns *= halt_poll_ns_grow;
3762 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3763 		vc->halt_poll_ns = halt_poll_ns_grow_start;
3764 }
3765 
3766 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3767 {
3768 	if (halt_poll_ns_shrink == 0)
3769 		vc->halt_poll_ns = 0;
3770 	else
3771 		vc->halt_poll_ns /= halt_poll_ns_shrink;
3772 }
3773 
3774 #ifdef CONFIG_KVM_XICS
3775 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3776 {
3777 	if (!xics_on_xive())
3778 		return false;
3779 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3780 		vcpu->arch.xive_saved_state.cppr;
3781 }
3782 #else
3783 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3784 {
3785 	return false;
3786 }
3787 #endif /* CONFIG_KVM_XICS */
3788 
3789 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3790 {
3791 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3792 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3793 		return true;
3794 
3795 	return false;
3796 }
3797 
3798 /*
3799  * Check to see if any of the runnable vcpus on the vcore have pending
3800  * exceptions or are no longer ceded
3801  */
3802 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3803 {
3804 	struct kvm_vcpu *vcpu;
3805 	int i;
3806 
3807 	for_each_runnable_thread(i, vcpu, vc) {
3808 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3809 			return 1;
3810 	}
3811 
3812 	return 0;
3813 }
3814 
3815 /*
3816  * All the vcpus in this vcore are idle, so wait for a decrementer
3817  * or external interrupt to one of the vcpus.  vc->lock is held.
3818  */
3819 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3820 {
3821 	ktime_t cur, start_poll, start_wait;
3822 	int do_sleep = 1;
3823 	u64 block_ns;
3824 
3825 	/* Poll for pending exceptions and ceded state */
3826 	cur = start_poll = ktime_get();
3827 	if (vc->halt_poll_ns) {
3828 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3829 		++vc->runner->stat.halt_attempted_poll;
3830 
3831 		vc->vcore_state = VCORE_POLLING;
3832 		spin_unlock(&vc->lock);
3833 
3834 		do {
3835 			if (kvmppc_vcore_check_block(vc)) {
3836 				do_sleep = 0;
3837 				break;
3838 			}
3839 			cur = ktime_get();
3840 		} while (single_task_running() && ktime_before(cur, stop));
3841 
3842 		spin_lock(&vc->lock);
3843 		vc->vcore_state = VCORE_INACTIVE;
3844 
3845 		if (!do_sleep) {
3846 			++vc->runner->stat.halt_successful_poll;
3847 			goto out;
3848 		}
3849 	}
3850 
3851 	prepare_to_rcuwait(&vc->wait);
3852 	set_current_state(TASK_INTERRUPTIBLE);
3853 	if (kvmppc_vcore_check_block(vc)) {
3854 		finish_rcuwait(&vc->wait);
3855 		do_sleep = 0;
3856 		/* If we polled, count this as a successful poll */
3857 		if (vc->halt_poll_ns)
3858 			++vc->runner->stat.halt_successful_poll;
3859 		goto out;
3860 	}
3861 
3862 	start_wait = ktime_get();
3863 
3864 	vc->vcore_state = VCORE_SLEEPING;
3865 	trace_kvmppc_vcore_blocked(vc, 0);
3866 	spin_unlock(&vc->lock);
3867 	schedule();
3868 	finish_rcuwait(&vc->wait);
3869 	spin_lock(&vc->lock);
3870 	vc->vcore_state = VCORE_INACTIVE;
3871 	trace_kvmppc_vcore_blocked(vc, 1);
3872 	++vc->runner->stat.halt_successful_wait;
3873 
3874 	cur = ktime_get();
3875 
3876 out:
3877 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3878 
3879 	/* Attribute wait time */
3880 	if (do_sleep) {
3881 		vc->runner->stat.halt_wait_ns +=
3882 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3883 		/* Attribute failed poll time */
3884 		if (vc->halt_poll_ns)
3885 			vc->runner->stat.halt_poll_fail_ns +=
3886 				ktime_to_ns(start_wait) -
3887 				ktime_to_ns(start_poll);
3888 	} else {
3889 		/* Attribute successful poll time */
3890 		if (vc->halt_poll_ns)
3891 			vc->runner->stat.halt_poll_success_ns +=
3892 				ktime_to_ns(cur) -
3893 				ktime_to_ns(start_poll);
3894 	}
3895 
3896 	/* Adjust poll time */
3897 	if (halt_poll_ns) {
3898 		if (block_ns <= vc->halt_poll_ns)
3899 			;
3900 		/* We slept and blocked for longer than the max halt time */
3901 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3902 			shrink_halt_poll_ns(vc);
3903 		/* We slept and our poll time is too small */
3904 		else if (vc->halt_poll_ns < halt_poll_ns &&
3905 				block_ns < halt_poll_ns)
3906 			grow_halt_poll_ns(vc);
3907 		if (vc->halt_poll_ns > halt_poll_ns)
3908 			vc->halt_poll_ns = halt_poll_ns;
3909 	} else
3910 		vc->halt_poll_ns = 0;
3911 
3912 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3913 }
3914 
3915 /*
3916  * This never fails for a radix guest, as none of the operations it does
3917  * for a radix guest can fail or have a way to report failure.
3918  * kvmhv_run_single_vcpu() relies on this fact.
3919  */
3920 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3921 {
3922 	int r = 0;
3923 	struct kvm *kvm = vcpu->kvm;
3924 
3925 	mutex_lock(&kvm->arch.mmu_setup_lock);
3926 	if (!kvm->arch.mmu_ready) {
3927 		if (!kvm_is_radix(kvm))
3928 			r = kvmppc_hv_setup_htab_rma(vcpu);
3929 		if (!r) {
3930 			if (cpu_has_feature(CPU_FTR_ARCH_300))
3931 				kvmppc_setup_partition_table(kvm);
3932 			kvm->arch.mmu_ready = 1;
3933 		}
3934 	}
3935 	mutex_unlock(&kvm->arch.mmu_setup_lock);
3936 	return r;
3937 }
3938 
3939 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
3940 {
3941 	struct kvm_run *run = vcpu->run;
3942 	int n_ceded, i, r;
3943 	struct kvmppc_vcore *vc;
3944 	struct kvm_vcpu *v;
3945 
3946 	trace_kvmppc_run_vcpu_enter(vcpu);
3947 
3948 	run->exit_reason = 0;
3949 	vcpu->arch.ret = RESUME_GUEST;
3950 	vcpu->arch.trap = 0;
3951 	kvmppc_update_vpas(vcpu);
3952 
3953 	/*
3954 	 * Synchronize with other threads in this virtual core
3955 	 */
3956 	vc = vcpu->arch.vcore;
3957 	spin_lock(&vc->lock);
3958 	vcpu->arch.ceded = 0;
3959 	vcpu->arch.run_task = current;
3960 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3961 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3962 	vcpu->arch.busy_preempt = TB_NIL;
3963 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3964 	++vc->n_runnable;
3965 
3966 	/*
3967 	 * This happens the first time this is called for a vcpu.
3968 	 * If the vcore is already running, we may be able to start
3969 	 * this thread straight away and have it join in.
3970 	 */
3971 	if (!signal_pending(current)) {
3972 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
3973 		     vc->vcore_state == VCORE_RUNNING) &&
3974 			   !VCORE_IS_EXITING(vc)) {
3975 			kvmppc_create_dtl_entry(vcpu, vc);
3976 			kvmppc_start_thread(vcpu, vc);
3977 			trace_kvm_guest_enter(vcpu);
3978 		} else if (vc->vcore_state == VCORE_SLEEPING) {
3979 		        rcuwait_wake_up(&vc->wait);
3980 		}
3981 
3982 	}
3983 
3984 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3985 	       !signal_pending(current)) {
3986 		/* See if the MMU is ready to go */
3987 		if (!vcpu->kvm->arch.mmu_ready) {
3988 			spin_unlock(&vc->lock);
3989 			r = kvmhv_setup_mmu(vcpu);
3990 			spin_lock(&vc->lock);
3991 			if (r) {
3992 				run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3993 				run->fail_entry.
3994 					hardware_entry_failure_reason = 0;
3995 				vcpu->arch.ret = r;
3996 				break;
3997 			}
3998 		}
3999 
4000 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4001 			kvmppc_vcore_end_preempt(vc);
4002 
4003 		if (vc->vcore_state != VCORE_INACTIVE) {
4004 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4005 			continue;
4006 		}
4007 		for_each_runnable_thread(i, v, vc) {
4008 			kvmppc_core_prepare_to_enter(v);
4009 			if (signal_pending(v->arch.run_task)) {
4010 				kvmppc_remove_runnable(vc, v);
4011 				v->stat.signal_exits++;
4012 				v->run->exit_reason = KVM_EXIT_INTR;
4013 				v->arch.ret = -EINTR;
4014 				wake_up(&v->arch.cpu_run);
4015 			}
4016 		}
4017 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4018 			break;
4019 		n_ceded = 0;
4020 		for_each_runnable_thread(i, v, vc) {
4021 			if (!kvmppc_vcpu_woken(v))
4022 				n_ceded += v->arch.ceded;
4023 			else
4024 				v->arch.ceded = 0;
4025 		}
4026 		vc->runner = vcpu;
4027 		if (n_ceded == vc->n_runnable) {
4028 			kvmppc_vcore_blocked(vc);
4029 		} else if (need_resched()) {
4030 			kvmppc_vcore_preempt(vc);
4031 			/* Let something else run */
4032 			cond_resched_lock(&vc->lock);
4033 			if (vc->vcore_state == VCORE_PREEMPT)
4034 				kvmppc_vcore_end_preempt(vc);
4035 		} else {
4036 			kvmppc_run_core(vc);
4037 		}
4038 		vc->runner = NULL;
4039 	}
4040 
4041 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4042 	       (vc->vcore_state == VCORE_RUNNING ||
4043 		vc->vcore_state == VCORE_EXITING ||
4044 		vc->vcore_state == VCORE_PIGGYBACK))
4045 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4046 
4047 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4048 		kvmppc_vcore_end_preempt(vc);
4049 
4050 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4051 		kvmppc_remove_runnable(vc, vcpu);
4052 		vcpu->stat.signal_exits++;
4053 		run->exit_reason = KVM_EXIT_INTR;
4054 		vcpu->arch.ret = -EINTR;
4055 	}
4056 
4057 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4058 		/* Wake up some vcpu to run the core */
4059 		i = -1;
4060 		v = next_runnable_thread(vc, &i);
4061 		wake_up(&v->arch.cpu_run);
4062 	}
4063 
4064 	trace_kvmppc_run_vcpu_exit(vcpu);
4065 	spin_unlock(&vc->lock);
4066 	return vcpu->arch.ret;
4067 }
4068 
4069 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4070 			  unsigned long lpcr)
4071 {
4072 	struct kvm_run *run = vcpu->run;
4073 	int trap, r, pcpu;
4074 	int srcu_idx, lpid;
4075 	struct kvmppc_vcore *vc;
4076 	struct kvm *kvm = vcpu->kvm;
4077 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4078 
4079 	trace_kvmppc_run_vcpu_enter(vcpu);
4080 
4081 	run->exit_reason = 0;
4082 	vcpu->arch.ret = RESUME_GUEST;
4083 	vcpu->arch.trap = 0;
4084 
4085 	vc = vcpu->arch.vcore;
4086 	vcpu->arch.ceded = 0;
4087 	vcpu->arch.run_task = current;
4088 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4089 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4090 	vcpu->arch.busy_preempt = TB_NIL;
4091 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4092 	vc->runnable_threads[0] = vcpu;
4093 	vc->n_runnable = 1;
4094 	vc->runner = vcpu;
4095 
4096 	/* See if the MMU is ready to go */
4097 	if (!kvm->arch.mmu_ready)
4098 		kvmhv_setup_mmu(vcpu);
4099 
4100 	if (need_resched())
4101 		cond_resched();
4102 
4103 	kvmppc_update_vpas(vcpu);
4104 
4105 	init_vcore_to_run(vc);
4106 	vc->preempt_tb = TB_NIL;
4107 
4108 	preempt_disable();
4109 	pcpu = smp_processor_id();
4110 	vc->pcpu = pcpu;
4111 	kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4112 
4113 	local_irq_disable();
4114 	hard_irq_disable();
4115 	if (signal_pending(current))
4116 		goto sigpend;
4117 	if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4118 		goto out;
4119 
4120 	if (!nested) {
4121 		kvmppc_core_prepare_to_enter(vcpu);
4122 		if (vcpu->arch.doorbell_request) {
4123 			vc->dpdes = 1;
4124 			smp_wmb();
4125 			vcpu->arch.doorbell_request = 0;
4126 		}
4127 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4128 			     &vcpu->arch.pending_exceptions))
4129 			lpcr |= LPCR_MER;
4130 	} else if (vcpu->arch.pending_exceptions ||
4131 		   vcpu->arch.doorbell_request ||
4132 		   xive_interrupt_pending(vcpu)) {
4133 		vcpu->arch.ret = RESUME_HOST;
4134 		goto out;
4135 	}
4136 
4137 	kvmppc_clear_host_core(pcpu);
4138 
4139 	local_paca->kvm_hstate.tid = 0;
4140 	local_paca->kvm_hstate.napping = 0;
4141 	local_paca->kvm_hstate.kvm_split_mode = NULL;
4142 	kvmppc_start_thread(vcpu, vc);
4143 	kvmppc_create_dtl_entry(vcpu, vc);
4144 	trace_kvm_guest_enter(vcpu);
4145 
4146 	vc->vcore_state = VCORE_RUNNING;
4147 	trace_kvmppc_run_core(vc, 0);
4148 
4149 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4150 		lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4151 		mtspr(SPRN_LPID, lpid);
4152 		isync();
4153 		kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4154 	}
4155 
4156 	guest_enter_irqoff();
4157 
4158 	srcu_idx = srcu_read_lock(&kvm->srcu);
4159 
4160 	this_cpu_disable_ftrace();
4161 
4162 	/* Tell lockdep that we're about to enable interrupts */
4163 	trace_hardirqs_on();
4164 
4165 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4166 	vcpu->arch.trap = trap;
4167 
4168 	trace_hardirqs_off();
4169 
4170 	this_cpu_enable_ftrace();
4171 
4172 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4173 
4174 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4175 		mtspr(SPRN_LPID, kvm->arch.host_lpid);
4176 		isync();
4177 	}
4178 
4179 	set_irq_happened(trap);
4180 
4181 	kvmppc_set_host_core(pcpu);
4182 
4183 	local_irq_enable();
4184 	guest_exit();
4185 
4186 	cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4187 
4188 	preempt_enable();
4189 
4190 	/*
4191 	 * cancel pending decrementer exception if DEC is now positive, or if
4192 	 * entering a nested guest in which case the decrementer is now owned
4193 	 * by L2 and the L1 decrementer is provided in hdec_expires
4194 	 */
4195 	if (kvmppc_core_pending_dec(vcpu) &&
4196 			((get_tb() < vcpu->arch.dec_expires) ||
4197 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4198 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4199 		kvmppc_core_dequeue_dec(vcpu);
4200 
4201 	trace_kvm_guest_exit(vcpu);
4202 	r = RESUME_GUEST;
4203 	if (trap) {
4204 		if (!nested)
4205 			r = kvmppc_handle_exit_hv(vcpu, current);
4206 		else
4207 			r = kvmppc_handle_nested_exit(vcpu);
4208 	}
4209 	vcpu->arch.ret = r;
4210 
4211 	if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4212 	    !kvmppc_vcpu_woken(vcpu)) {
4213 		kvmppc_set_timer(vcpu);
4214 		while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4215 			if (signal_pending(current)) {
4216 				vcpu->stat.signal_exits++;
4217 				run->exit_reason = KVM_EXIT_INTR;
4218 				vcpu->arch.ret = -EINTR;
4219 				break;
4220 			}
4221 			spin_lock(&vc->lock);
4222 			kvmppc_vcore_blocked(vc);
4223 			spin_unlock(&vc->lock);
4224 		}
4225 	}
4226 	vcpu->arch.ceded = 0;
4227 
4228 	vc->vcore_state = VCORE_INACTIVE;
4229 	trace_kvmppc_run_core(vc, 1);
4230 
4231  done:
4232 	kvmppc_remove_runnable(vc, vcpu);
4233 	trace_kvmppc_run_vcpu_exit(vcpu);
4234 
4235 	return vcpu->arch.ret;
4236 
4237  sigpend:
4238 	vcpu->stat.signal_exits++;
4239 	run->exit_reason = KVM_EXIT_INTR;
4240 	vcpu->arch.ret = -EINTR;
4241  out:
4242 	local_irq_enable();
4243 	preempt_enable();
4244 	goto done;
4245 }
4246 
4247 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4248 {
4249 	struct kvm_run *run = vcpu->run;
4250 	int r;
4251 	int srcu_idx;
4252 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
4253 	unsigned long user_tar = 0;
4254 	unsigned int user_vrsave;
4255 	struct kvm *kvm;
4256 
4257 	if (!vcpu->arch.sane) {
4258 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4259 		return -EINVAL;
4260 	}
4261 
4262 	/*
4263 	 * Don't allow entry with a suspended transaction, because
4264 	 * the guest entry/exit code will lose it.
4265 	 * If the guest has TM enabled, save away their TM-related SPRs
4266 	 * (they will get restored by the TM unavailable interrupt).
4267 	 */
4268 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4269 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4270 	    (current->thread.regs->msr & MSR_TM)) {
4271 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4272 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4273 			run->fail_entry.hardware_entry_failure_reason = 0;
4274 			return -EINVAL;
4275 		}
4276 		/* Enable TM so we can read the TM SPRs */
4277 		mtmsr(mfmsr() | MSR_TM);
4278 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4279 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4280 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4281 		current->thread.regs->msr &= ~MSR_TM;
4282 	}
4283 #endif
4284 
4285 	/*
4286 	 * Force online to 1 for the sake of old userspace which doesn't
4287 	 * set it.
4288 	 */
4289 	if (!vcpu->arch.online) {
4290 		atomic_inc(&vcpu->arch.vcore->online_count);
4291 		vcpu->arch.online = 1;
4292 	}
4293 
4294 	kvmppc_core_prepare_to_enter(vcpu);
4295 
4296 	/* No need to go into the guest when all we'll do is come back out */
4297 	if (signal_pending(current)) {
4298 		run->exit_reason = KVM_EXIT_INTR;
4299 		return -EINTR;
4300 	}
4301 
4302 	kvm = vcpu->kvm;
4303 	atomic_inc(&kvm->arch.vcpus_running);
4304 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4305 	smp_mb();
4306 
4307 	flush_all_to_thread(current);
4308 
4309 	/* Save userspace EBB and other register values */
4310 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4311 		ebb_regs[0] = mfspr(SPRN_EBBHR);
4312 		ebb_regs[1] = mfspr(SPRN_EBBRR);
4313 		ebb_regs[2] = mfspr(SPRN_BESCR);
4314 		user_tar = mfspr(SPRN_TAR);
4315 	}
4316 	user_vrsave = mfspr(SPRN_VRSAVE);
4317 
4318 	vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4319 	vcpu->arch.pgdir = kvm->mm->pgd;
4320 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4321 
4322 	do {
4323 		/*
4324 		 * The early POWER9 chips that can't mix radix and HPT threads
4325 		 * on the same core also need the workaround for the problem
4326 		 * where the TLB would prefetch entries in the guest exit path
4327 		 * for radix guests using the guest PIDR value and LPID 0.
4328 		 * The workaround is in the old path (kvmppc_run_vcpu())
4329 		 * but not the new path (kvmhv_run_single_vcpu()).
4330 		 */
4331 		if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4332 		    !no_mixing_hpt_and_radix)
4333 			r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4334 						  vcpu->arch.vcore->lpcr);
4335 		else
4336 			r = kvmppc_run_vcpu(vcpu);
4337 
4338 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4339 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4340 			trace_kvm_hcall_enter(vcpu);
4341 			r = kvmppc_pseries_do_hcall(vcpu);
4342 			trace_kvm_hcall_exit(vcpu, r);
4343 			kvmppc_core_prepare_to_enter(vcpu);
4344 		} else if (r == RESUME_PAGE_FAULT) {
4345 			srcu_idx = srcu_read_lock(&kvm->srcu);
4346 			r = kvmppc_book3s_hv_page_fault(vcpu,
4347 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4348 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4349 		} else if (r == RESUME_PASSTHROUGH) {
4350 			if (WARN_ON(xics_on_xive()))
4351 				r = H_SUCCESS;
4352 			else
4353 				r = kvmppc_xics_rm_complete(vcpu, 0);
4354 		}
4355 	} while (is_kvmppc_resume_guest(r));
4356 
4357 	/* Restore userspace EBB and other register values */
4358 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4359 		mtspr(SPRN_EBBHR, ebb_regs[0]);
4360 		mtspr(SPRN_EBBRR, ebb_regs[1]);
4361 		mtspr(SPRN_BESCR, ebb_regs[2]);
4362 		mtspr(SPRN_TAR, user_tar);
4363 		mtspr(SPRN_FSCR, current->thread.fscr);
4364 	}
4365 	mtspr(SPRN_VRSAVE, user_vrsave);
4366 
4367 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4368 	atomic_dec(&kvm->arch.vcpus_running);
4369 	return r;
4370 }
4371 
4372 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4373 				     int shift, int sllp)
4374 {
4375 	(*sps)->page_shift = shift;
4376 	(*sps)->slb_enc = sllp;
4377 	(*sps)->enc[0].page_shift = shift;
4378 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4379 	/*
4380 	 * Add 16MB MPSS support (may get filtered out by userspace)
4381 	 */
4382 	if (shift != 24) {
4383 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4384 		if (penc != -1) {
4385 			(*sps)->enc[1].page_shift = 24;
4386 			(*sps)->enc[1].pte_enc = penc;
4387 		}
4388 	}
4389 	(*sps)++;
4390 }
4391 
4392 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4393 					 struct kvm_ppc_smmu_info *info)
4394 {
4395 	struct kvm_ppc_one_seg_page_size *sps;
4396 
4397 	/*
4398 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4399 	 * POWER7 doesn't support keys for instruction accesses,
4400 	 * POWER8 and POWER9 do.
4401 	 */
4402 	info->data_keys = 32;
4403 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4404 
4405 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4406 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4407 	info->slb_size = 32;
4408 
4409 	/* We only support these sizes for now, and no muti-size segments */
4410 	sps = &info->sps[0];
4411 	kvmppc_add_seg_page_size(&sps, 12, 0);
4412 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4413 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4414 
4415 	/* If running as a nested hypervisor, we don't support HPT guests */
4416 	if (kvmhv_on_pseries())
4417 		info->flags |= KVM_PPC_NO_HASH;
4418 
4419 	return 0;
4420 }
4421 
4422 /*
4423  * Get (and clear) the dirty memory log for a memory slot.
4424  */
4425 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4426 					 struct kvm_dirty_log *log)
4427 {
4428 	struct kvm_memslots *slots;
4429 	struct kvm_memory_slot *memslot;
4430 	int i, r;
4431 	unsigned long n;
4432 	unsigned long *buf, *p;
4433 	struct kvm_vcpu *vcpu;
4434 
4435 	mutex_lock(&kvm->slots_lock);
4436 
4437 	r = -EINVAL;
4438 	if (log->slot >= KVM_USER_MEM_SLOTS)
4439 		goto out;
4440 
4441 	slots = kvm_memslots(kvm);
4442 	memslot = id_to_memslot(slots, log->slot);
4443 	r = -ENOENT;
4444 	if (!memslot || !memslot->dirty_bitmap)
4445 		goto out;
4446 
4447 	/*
4448 	 * Use second half of bitmap area because both HPT and radix
4449 	 * accumulate bits in the first half.
4450 	 */
4451 	n = kvm_dirty_bitmap_bytes(memslot);
4452 	buf = memslot->dirty_bitmap + n / sizeof(long);
4453 	memset(buf, 0, n);
4454 
4455 	if (kvm_is_radix(kvm))
4456 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4457 	else
4458 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4459 	if (r)
4460 		goto out;
4461 
4462 	/*
4463 	 * We accumulate dirty bits in the first half of the
4464 	 * memslot's dirty_bitmap area, for when pages are paged
4465 	 * out or modified by the host directly.  Pick up these
4466 	 * bits and add them to the map.
4467 	 */
4468 	p = memslot->dirty_bitmap;
4469 	for (i = 0; i < n / sizeof(long); ++i)
4470 		buf[i] |= xchg(&p[i], 0);
4471 
4472 	/* Harvest dirty bits from VPA and DTL updates */
4473 	/* Note: we never modify the SLB shadow buffer areas */
4474 	kvm_for_each_vcpu(i, vcpu, kvm) {
4475 		spin_lock(&vcpu->arch.vpa_update_lock);
4476 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4477 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4478 		spin_unlock(&vcpu->arch.vpa_update_lock);
4479 	}
4480 
4481 	r = -EFAULT;
4482 	if (copy_to_user(log->dirty_bitmap, buf, n))
4483 		goto out;
4484 
4485 	r = 0;
4486 out:
4487 	mutex_unlock(&kvm->slots_lock);
4488 	return r;
4489 }
4490 
4491 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4492 {
4493 	vfree(slot->arch.rmap);
4494 	slot->arch.rmap = NULL;
4495 }
4496 
4497 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4498 					struct kvm_memory_slot *slot,
4499 					const struct kvm_userspace_memory_region *mem,
4500 					enum kvm_mr_change change)
4501 {
4502 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4503 
4504 	if (change == KVM_MR_CREATE) {
4505 		slot->arch.rmap = vzalloc(array_size(npages,
4506 					  sizeof(*slot->arch.rmap)));
4507 		if (!slot->arch.rmap)
4508 			return -ENOMEM;
4509 	}
4510 
4511 	return 0;
4512 }
4513 
4514 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4515 				const struct kvm_userspace_memory_region *mem,
4516 				const struct kvm_memory_slot *old,
4517 				const struct kvm_memory_slot *new,
4518 				enum kvm_mr_change change)
4519 {
4520 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4521 
4522 	/*
4523 	 * If we are making a new memslot, it might make
4524 	 * some address that was previously cached as emulated
4525 	 * MMIO be no longer emulated MMIO, so invalidate
4526 	 * all the caches of emulated MMIO translations.
4527 	 */
4528 	if (npages)
4529 		atomic64_inc(&kvm->arch.mmio_update);
4530 
4531 	/*
4532 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4533 	 * have already called kvm_arch_flush_shadow_memslot() to
4534 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
4535 	 * previous mappings.  So the only case to handle is
4536 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4537 	 * has been changed.
4538 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4539 	 * to get rid of any THP PTEs in the partition-scoped page tables
4540 	 * so we can track dirtiness at the page level; we flush when
4541 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4542 	 * using THP PTEs.
4543 	 */
4544 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4545 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4546 		kvmppc_radix_flush_memslot(kvm, old);
4547 	/*
4548 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4549 	 */
4550 	if (!kvm->arch.secure_guest)
4551 		return;
4552 
4553 	switch (change) {
4554 	case KVM_MR_CREATE:
4555 		if (kvmppc_uvmem_slot_init(kvm, new))
4556 			return;
4557 		uv_register_mem_slot(kvm->arch.lpid,
4558 				     new->base_gfn << PAGE_SHIFT,
4559 				     new->npages * PAGE_SIZE,
4560 				     0, new->id);
4561 		break;
4562 	case KVM_MR_DELETE:
4563 		uv_unregister_mem_slot(kvm->arch.lpid, old->id);
4564 		kvmppc_uvmem_slot_free(kvm, old);
4565 		break;
4566 	default:
4567 		/* TODO: Handle KVM_MR_MOVE */
4568 		break;
4569 	}
4570 }
4571 
4572 /*
4573  * Update LPCR values in kvm->arch and in vcores.
4574  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4575  * of kvm->arch.lpcr update).
4576  */
4577 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4578 {
4579 	long int i;
4580 	u32 cores_done = 0;
4581 
4582 	if ((kvm->arch.lpcr & mask) == lpcr)
4583 		return;
4584 
4585 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4586 
4587 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4588 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4589 		if (!vc)
4590 			continue;
4591 		spin_lock(&vc->lock);
4592 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4593 		spin_unlock(&vc->lock);
4594 		if (++cores_done >= kvm->arch.online_vcores)
4595 			break;
4596 	}
4597 }
4598 
4599 void kvmppc_setup_partition_table(struct kvm *kvm)
4600 {
4601 	unsigned long dw0, dw1;
4602 
4603 	if (!kvm_is_radix(kvm)) {
4604 		/* PS field - page size for VRMA */
4605 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4606 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4607 		/* HTABSIZE and HTABORG fields */
4608 		dw0 |= kvm->arch.sdr1;
4609 
4610 		/* Second dword as set by userspace */
4611 		dw1 = kvm->arch.process_table;
4612 	} else {
4613 		dw0 = PATB_HR | radix__get_tree_size() |
4614 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4615 		dw1 = PATB_GR | kvm->arch.process_table;
4616 	}
4617 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4618 }
4619 
4620 /*
4621  * Set up HPT (hashed page table) and RMA (real-mode area).
4622  * Must be called with kvm->arch.mmu_setup_lock held.
4623  */
4624 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4625 {
4626 	int err = 0;
4627 	struct kvm *kvm = vcpu->kvm;
4628 	unsigned long hva;
4629 	struct kvm_memory_slot *memslot;
4630 	struct vm_area_struct *vma;
4631 	unsigned long lpcr = 0, senc;
4632 	unsigned long psize, porder;
4633 	int srcu_idx;
4634 
4635 	/* Allocate hashed page table (if not done already) and reset it */
4636 	if (!kvm->arch.hpt.virt) {
4637 		int order = KVM_DEFAULT_HPT_ORDER;
4638 		struct kvm_hpt_info info;
4639 
4640 		err = kvmppc_allocate_hpt(&info, order);
4641 		/* If we get here, it means userspace didn't specify a
4642 		 * size explicitly.  So, try successively smaller
4643 		 * sizes if the default failed. */
4644 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4645 			err  = kvmppc_allocate_hpt(&info, order);
4646 
4647 		if (err < 0) {
4648 			pr_err("KVM: Couldn't alloc HPT\n");
4649 			goto out;
4650 		}
4651 
4652 		kvmppc_set_hpt(kvm, &info);
4653 	}
4654 
4655 	/* Look up the memslot for guest physical address 0 */
4656 	srcu_idx = srcu_read_lock(&kvm->srcu);
4657 	memslot = gfn_to_memslot(kvm, 0);
4658 
4659 	/* We must have some memory at 0 by now */
4660 	err = -EINVAL;
4661 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4662 		goto out_srcu;
4663 
4664 	/* Look up the VMA for the start of this memory slot */
4665 	hva = memslot->userspace_addr;
4666 	mmap_read_lock(kvm->mm);
4667 	vma = find_vma(kvm->mm, hva);
4668 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4669 		goto up_out;
4670 
4671 	psize = vma_kernel_pagesize(vma);
4672 
4673 	mmap_read_unlock(kvm->mm);
4674 
4675 	/* We can handle 4k, 64k or 16M pages in the VRMA */
4676 	if (psize >= 0x1000000)
4677 		psize = 0x1000000;
4678 	else if (psize >= 0x10000)
4679 		psize = 0x10000;
4680 	else
4681 		psize = 0x1000;
4682 	porder = __ilog2(psize);
4683 
4684 	senc = slb_pgsize_encoding(psize);
4685 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4686 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4687 	/* Create HPTEs in the hash page table for the VRMA */
4688 	kvmppc_map_vrma(vcpu, memslot, porder);
4689 
4690 	/* Update VRMASD field in the LPCR */
4691 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4692 		/* the -4 is to account for senc values starting at 0x10 */
4693 		lpcr = senc << (LPCR_VRMASD_SH - 4);
4694 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4695 	}
4696 
4697 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4698 	smp_wmb();
4699 	err = 0;
4700  out_srcu:
4701 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4702  out:
4703 	return err;
4704 
4705  up_out:
4706 	mmap_read_unlock(kvm->mm);
4707 	goto out_srcu;
4708 }
4709 
4710 /*
4711  * Must be called with kvm->arch.mmu_setup_lock held and
4712  * mmu_ready = 0 and no vcpus running.
4713  */
4714 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4715 {
4716 	if (nesting_enabled(kvm))
4717 		kvmhv_release_all_nested(kvm);
4718 	kvmppc_rmap_reset(kvm);
4719 	kvm->arch.process_table = 0;
4720 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4721 	spin_lock(&kvm->mmu_lock);
4722 	kvm->arch.radix = 0;
4723 	spin_unlock(&kvm->mmu_lock);
4724 	kvmppc_free_radix(kvm);
4725 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
4726 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4727 	return 0;
4728 }
4729 
4730 /*
4731  * Must be called with kvm->arch.mmu_setup_lock held and
4732  * mmu_ready = 0 and no vcpus running.
4733  */
4734 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4735 {
4736 	int err;
4737 
4738 	err = kvmppc_init_vm_radix(kvm);
4739 	if (err)
4740 		return err;
4741 	kvmppc_rmap_reset(kvm);
4742 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4743 	spin_lock(&kvm->mmu_lock);
4744 	kvm->arch.radix = 1;
4745 	spin_unlock(&kvm->mmu_lock);
4746 	kvmppc_free_hpt(&kvm->arch.hpt);
4747 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4748 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4749 	return 0;
4750 }
4751 
4752 #ifdef CONFIG_KVM_XICS
4753 /*
4754  * Allocate a per-core structure for managing state about which cores are
4755  * running in the host versus the guest and for exchanging data between
4756  * real mode KVM and CPU running in the host.
4757  * This is only done for the first VM.
4758  * The allocated structure stays even if all VMs have stopped.
4759  * It is only freed when the kvm-hv module is unloaded.
4760  * It's OK for this routine to fail, we just don't support host
4761  * core operations like redirecting H_IPI wakeups.
4762  */
4763 void kvmppc_alloc_host_rm_ops(void)
4764 {
4765 	struct kvmppc_host_rm_ops *ops;
4766 	unsigned long l_ops;
4767 	int cpu, core;
4768 	int size;
4769 
4770 	/* Not the first time here ? */
4771 	if (kvmppc_host_rm_ops_hv != NULL)
4772 		return;
4773 
4774 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4775 	if (!ops)
4776 		return;
4777 
4778 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4779 	ops->rm_core = kzalloc(size, GFP_KERNEL);
4780 
4781 	if (!ops->rm_core) {
4782 		kfree(ops);
4783 		return;
4784 	}
4785 
4786 	cpus_read_lock();
4787 
4788 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4789 		if (!cpu_online(cpu))
4790 			continue;
4791 
4792 		core = cpu >> threads_shift;
4793 		ops->rm_core[core].rm_state.in_host = 1;
4794 	}
4795 
4796 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4797 
4798 	/*
4799 	 * Make the contents of the kvmppc_host_rm_ops structure visible
4800 	 * to other CPUs before we assign it to the global variable.
4801 	 * Do an atomic assignment (no locks used here), but if someone
4802 	 * beats us to it, just free our copy and return.
4803 	 */
4804 	smp_wmb();
4805 	l_ops = (unsigned long) ops;
4806 
4807 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4808 		cpus_read_unlock();
4809 		kfree(ops->rm_core);
4810 		kfree(ops);
4811 		return;
4812 	}
4813 
4814 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4815 					     "ppc/kvm_book3s:prepare",
4816 					     kvmppc_set_host_core,
4817 					     kvmppc_clear_host_core);
4818 	cpus_read_unlock();
4819 }
4820 
4821 void kvmppc_free_host_rm_ops(void)
4822 {
4823 	if (kvmppc_host_rm_ops_hv) {
4824 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4825 		kfree(kvmppc_host_rm_ops_hv->rm_core);
4826 		kfree(kvmppc_host_rm_ops_hv);
4827 		kvmppc_host_rm_ops_hv = NULL;
4828 	}
4829 }
4830 #endif
4831 
4832 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4833 {
4834 	unsigned long lpcr, lpid;
4835 	char buf[32];
4836 	int ret;
4837 
4838 	mutex_init(&kvm->arch.uvmem_lock);
4839 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4840 	mutex_init(&kvm->arch.mmu_setup_lock);
4841 
4842 	/* Allocate the guest's logical partition ID */
4843 
4844 	lpid = kvmppc_alloc_lpid();
4845 	if ((long)lpid < 0)
4846 		return -ENOMEM;
4847 	kvm->arch.lpid = lpid;
4848 
4849 	kvmppc_alloc_host_rm_ops();
4850 
4851 	kvmhv_vm_nested_init(kvm);
4852 
4853 	/*
4854 	 * Since we don't flush the TLB when tearing down a VM,
4855 	 * and this lpid might have previously been used,
4856 	 * make sure we flush on each core before running the new VM.
4857 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4858 	 * does this flush for us.
4859 	 */
4860 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4861 		cpumask_setall(&kvm->arch.need_tlb_flush);
4862 
4863 	/* Start out with the default set of hcalls enabled */
4864 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4865 	       sizeof(kvm->arch.enabled_hcalls));
4866 
4867 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4868 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4869 
4870 	/* Init LPCR for virtual RMA mode */
4871 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4872 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
4873 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4874 		lpcr &= LPCR_PECE | LPCR_LPES;
4875 	} else {
4876 		lpcr = 0;
4877 	}
4878 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4879 		LPCR_VPM0 | LPCR_VPM1;
4880 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4881 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4882 	/* On POWER8 turn on online bit to enable PURR/SPURR */
4883 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
4884 		lpcr |= LPCR_ONL;
4885 	/*
4886 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4887 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
4888 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4889 	 * be unnecessary but better safe than sorry in case we re-enable
4890 	 * EE in HV mode with this LPCR still set)
4891 	 */
4892 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4893 		lpcr &= ~LPCR_VPM0;
4894 		lpcr |= LPCR_HVICE | LPCR_HEIC;
4895 
4896 		/*
4897 		 * If xive is enabled, we route 0x500 interrupts directly
4898 		 * to the guest.
4899 		 */
4900 		if (xics_on_xive())
4901 			lpcr |= LPCR_LPES;
4902 	}
4903 
4904 	/*
4905 	 * If the host uses radix, the guest starts out as radix.
4906 	 */
4907 	if (radix_enabled()) {
4908 		kvm->arch.radix = 1;
4909 		kvm->arch.mmu_ready = 1;
4910 		lpcr &= ~LPCR_VPM1;
4911 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4912 		ret = kvmppc_init_vm_radix(kvm);
4913 		if (ret) {
4914 			kvmppc_free_lpid(kvm->arch.lpid);
4915 			return ret;
4916 		}
4917 		kvmppc_setup_partition_table(kvm);
4918 	}
4919 
4920 	kvm->arch.lpcr = lpcr;
4921 
4922 	/* Initialization for future HPT resizes */
4923 	kvm->arch.resize_hpt = NULL;
4924 
4925 	/*
4926 	 * Work out how many sets the TLB has, for the use of
4927 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4928 	 */
4929 	if (radix_enabled())
4930 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
4931 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
4932 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
4933 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4934 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
4935 	else
4936 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
4937 
4938 	/*
4939 	 * Track that we now have a HV mode VM active. This blocks secondary
4940 	 * CPU threads from coming online.
4941 	 * On POWER9, we only need to do this if the "indep_threads_mode"
4942 	 * module parameter has been set to N.
4943 	 */
4944 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4945 		if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4946 			pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4947 			kvm->arch.threads_indep = true;
4948 		} else {
4949 			kvm->arch.threads_indep = indep_threads_mode;
4950 		}
4951 	}
4952 	if (!kvm->arch.threads_indep)
4953 		kvm_hv_vm_activated();
4954 
4955 	/*
4956 	 * Initialize smt_mode depending on processor.
4957 	 * POWER8 and earlier have to use "strict" threading, where
4958 	 * all vCPUs in a vcore have to run on the same (sub)core,
4959 	 * whereas on POWER9 the threads can each run a different
4960 	 * guest.
4961 	 */
4962 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4963 		kvm->arch.smt_mode = threads_per_subcore;
4964 	else
4965 		kvm->arch.smt_mode = 1;
4966 	kvm->arch.emul_smt_mode = 1;
4967 
4968 	/*
4969 	 * Create a debugfs directory for the VM
4970 	 */
4971 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
4972 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4973 	kvmppc_mmu_debugfs_init(kvm);
4974 	if (radix_enabled())
4975 		kvmhv_radix_debugfs_init(kvm);
4976 
4977 	return 0;
4978 }
4979 
4980 static void kvmppc_free_vcores(struct kvm *kvm)
4981 {
4982 	long int i;
4983 
4984 	for (i = 0; i < KVM_MAX_VCORES; ++i)
4985 		kfree(kvm->arch.vcores[i]);
4986 	kvm->arch.online_vcores = 0;
4987 }
4988 
4989 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4990 {
4991 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
4992 
4993 	if (!kvm->arch.threads_indep)
4994 		kvm_hv_vm_deactivated();
4995 
4996 	kvmppc_free_vcores(kvm);
4997 
4998 
4999 	if (kvm_is_radix(kvm))
5000 		kvmppc_free_radix(kvm);
5001 	else
5002 		kvmppc_free_hpt(&kvm->arch.hpt);
5003 
5004 	/* Perform global invalidation and return lpid to the pool */
5005 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5006 		if (nesting_enabled(kvm))
5007 			kvmhv_release_all_nested(kvm);
5008 		kvm->arch.process_table = 0;
5009 		if (kvm->arch.secure_guest)
5010 			uv_svm_terminate(kvm->arch.lpid);
5011 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5012 	}
5013 
5014 	kvmppc_free_lpid(kvm->arch.lpid);
5015 
5016 	kvmppc_free_pimap(kvm);
5017 }
5018 
5019 /* We don't need to emulate any privileged instructions or dcbz */
5020 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5021 				     unsigned int inst, int *advance)
5022 {
5023 	return EMULATE_FAIL;
5024 }
5025 
5026 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5027 					ulong spr_val)
5028 {
5029 	return EMULATE_FAIL;
5030 }
5031 
5032 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5033 					ulong *spr_val)
5034 {
5035 	return EMULATE_FAIL;
5036 }
5037 
5038 static int kvmppc_core_check_processor_compat_hv(void)
5039 {
5040 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5041 	    cpu_has_feature(CPU_FTR_ARCH_206))
5042 		return 0;
5043 
5044 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5045 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5046 		return 0;
5047 
5048 	return -EIO;
5049 }
5050 
5051 #ifdef CONFIG_KVM_XICS
5052 
5053 void kvmppc_free_pimap(struct kvm *kvm)
5054 {
5055 	kfree(kvm->arch.pimap);
5056 }
5057 
5058 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5059 {
5060 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5061 }
5062 
5063 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5064 {
5065 	struct irq_desc *desc;
5066 	struct kvmppc_irq_map *irq_map;
5067 	struct kvmppc_passthru_irqmap *pimap;
5068 	struct irq_chip *chip;
5069 	int i, rc = 0;
5070 
5071 	if (!kvm_irq_bypass)
5072 		return 1;
5073 
5074 	desc = irq_to_desc(host_irq);
5075 	if (!desc)
5076 		return -EIO;
5077 
5078 	mutex_lock(&kvm->lock);
5079 
5080 	pimap = kvm->arch.pimap;
5081 	if (pimap == NULL) {
5082 		/* First call, allocate structure to hold IRQ map */
5083 		pimap = kvmppc_alloc_pimap();
5084 		if (pimap == NULL) {
5085 			mutex_unlock(&kvm->lock);
5086 			return -ENOMEM;
5087 		}
5088 		kvm->arch.pimap = pimap;
5089 	}
5090 
5091 	/*
5092 	 * For now, we only support interrupts for which the EOI operation
5093 	 * is an OPAL call followed by a write to XIRR, since that's
5094 	 * what our real-mode EOI code does, or a XIVE interrupt
5095 	 */
5096 	chip = irq_data_get_irq_chip(&desc->irq_data);
5097 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5098 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5099 			host_irq, guest_gsi);
5100 		mutex_unlock(&kvm->lock);
5101 		return -ENOENT;
5102 	}
5103 
5104 	/*
5105 	 * See if we already have an entry for this guest IRQ number.
5106 	 * If it's mapped to a hardware IRQ number, that's an error,
5107 	 * otherwise re-use this entry.
5108 	 */
5109 	for (i = 0; i < pimap->n_mapped; i++) {
5110 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5111 			if (pimap->mapped[i].r_hwirq) {
5112 				mutex_unlock(&kvm->lock);
5113 				return -EINVAL;
5114 			}
5115 			break;
5116 		}
5117 	}
5118 
5119 	if (i == KVMPPC_PIRQ_MAPPED) {
5120 		mutex_unlock(&kvm->lock);
5121 		return -EAGAIN;		/* table is full */
5122 	}
5123 
5124 	irq_map = &pimap->mapped[i];
5125 
5126 	irq_map->v_hwirq = guest_gsi;
5127 	irq_map->desc = desc;
5128 
5129 	/*
5130 	 * Order the above two stores before the next to serialize with
5131 	 * the KVM real mode handler.
5132 	 */
5133 	smp_wmb();
5134 	irq_map->r_hwirq = desc->irq_data.hwirq;
5135 
5136 	if (i == pimap->n_mapped)
5137 		pimap->n_mapped++;
5138 
5139 	if (xics_on_xive())
5140 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5141 	else
5142 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5143 	if (rc)
5144 		irq_map->r_hwirq = 0;
5145 
5146 	mutex_unlock(&kvm->lock);
5147 
5148 	return 0;
5149 }
5150 
5151 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5152 {
5153 	struct irq_desc *desc;
5154 	struct kvmppc_passthru_irqmap *pimap;
5155 	int i, rc = 0;
5156 
5157 	if (!kvm_irq_bypass)
5158 		return 0;
5159 
5160 	desc = irq_to_desc(host_irq);
5161 	if (!desc)
5162 		return -EIO;
5163 
5164 	mutex_lock(&kvm->lock);
5165 	if (!kvm->arch.pimap)
5166 		goto unlock;
5167 
5168 	pimap = kvm->arch.pimap;
5169 
5170 	for (i = 0; i < pimap->n_mapped; i++) {
5171 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5172 			break;
5173 	}
5174 
5175 	if (i == pimap->n_mapped) {
5176 		mutex_unlock(&kvm->lock);
5177 		return -ENODEV;
5178 	}
5179 
5180 	if (xics_on_xive())
5181 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5182 	else
5183 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5184 
5185 	/* invalidate the entry (what do do on error from the above ?) */
5186 	pimap->mapped[i].r_hwirq = 0;
5187 
5188 	/*
5189 	 * We don't free this structure even when the count goes to
5190 	 * zero. The structure is freed when we destroy the VM.
5191 	 */
5192  unlock:
5193 	mutex_unlock(&kvm->lock);
5194 	return rc;
5195 }
5196 
5197 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5198 					     struct irq_bypass_producer *prod)
5199 {
5200 	int ret = 0;
5201 	struct kvm_kernel_irqfd *irqfd =
5202 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5203 
5204 	irqfd->producer = prod;
5205 
5206 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5207 	if (ret)
5208 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5209 			prod->irq, irqfd->gsi, ret);
5210 
5211 	return ret;
5212 }
5213 
5214 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5215 					      struct irq_bypass_producer *prod)
5216 {
5217 	int ret;
5218 	struct kvm_kernel_irqfd *irqfd =
5219 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5220 
5221 	irqfd->producer = NULL;
5222 
5223 	/*
5224 	 * When producer of consumer is unregistered, we change back to
5225 	 * default external interrupt handling mode - KVM real mode
5226 	 * will switch back to host.
5227 	 */
5228 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5229 	if (ret)
5230 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5231 			prod->irq, irqfd->gsi, ret);
5232 }
5233 #endif
5234 
5235 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5236 				 unsigned int ioctl, unsigned long arg)
5237 {
5238 	struct kvm *kvm __maybe_unused = filp->private_data;
5239 	void __user *argp = (void __user *)arg;
5240 	long r;
5241 
5242 	switch (ioctl) {
5243 
5244 	case KVM_PPC_ALLOCATE_HTAB: {
5245 		u32 htab_order;
5246 
5247 		r = -EFAULT;
5248 		if (get_user(htab_order, (u32 __user *)argp))
5249 			break;
5250 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5251 		if (r)
5252 			break;
5253 		r = 0;
5254 		break;
5255 	}
5256 
5257 	case KVM_PPC_GET_HTAB_FD: {
5258 		struct kvm_get_htab_fd ghf;
5259 
5260 		r = -EFAULT;
5261 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5262 			break;
5263 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5264 		break;
5265 	}
5266 
5267 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5268 		struct kvm_ppc_resize_hpt rhpt;
5269 
5270 		r = -EFAULT;
5271 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5272 			break;
5273 
5274 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5275 		break;
5276 	}
5277 
5278 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5279 		struct kvm_ppc_resize_hpt rhpt;
5280 
5281 		r = -EFAULT;
5282 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5283 			break;
5284 
5285 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5286 		break;
5287 	}
5288 
5289 	default:
5290 		r = -ENOTTY;
5291 	}
5292 
5293 	return r;
5294 }
5295 
5296 /*
5297  * List of hcall numbers to enable by default.
5298  * For compatibility with old userspace, we enable by default
5299  * all hcalls that were implemented before the hcall-enabling
5300  * facility was added.  Note this list should not include H_RTAS.
5301  */
5302 static unsigned int default_hcall_list[] = {
5303 	H_REMOVE,
5304 	H_ENTER,
5305 	H_READ,
5306 	H_PROTECT,
5307 	H_BULK_REMOVE,
5308 	H_GET_TCE,
5309 	H_PUT_TCE,
5310 	H_SET_DABR,
5311 	H_SET_XDABR,
5312 	H_CEDE,
5313 	H_PROD,
5314 	H_CONFER,
5315 	H_REGISTER_VPA,
5316 #ifdef CONFIG_KVM_XICS
5317 	H_EOI,
5318 	H_CPPR,
5319 	H_IPI,
5320 	H_IPOLL,
5321 	H_XIRR,
5322 	H_XIRR_X,
5323 #endif
5324 	0
5325 };
5326 
5327 static void init_default_hcalls(void)
5328 {
5329 	int i;
5330 	unsigned int hcall;
5331 
5332 	for (i = 0; default_hcall_list[i]; ++i) {
5333 		hcall = default_hcall_list[i];
5334 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5335 		__set_bit(hcall / 4, default_enabled_hcalls);
5336 	}
5337 }
5338 
5339 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5340 {
5341 	unsigned long lpcr;
5342 	int radix;
5343 	int err;
5344 
5345 	/* If not on a POWER9, reject it */
5346 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5347 		return -ENODEV;
5348 
5349 	/* If any unknown flags set, reject it */
5350 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5351 		return -EINVAL;
5352 
5353 	/* GR (guest radix) bit in process_table field must match */
5354 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5355 	if (!!(cfg->process_table & PATB_GR) != radix)
5356 		return -EINVAL;
5357 
5358 	/* Process table size field must be reasonable, i.e. <= 24 */
5359 	if ((cfg->process_table & PRTS_MASK) > 24)
5360 		return -EINVAL;
5361 
5362 	/* We can change a guest to/from radix now, if the host is radix */
5363 	if (radix && !radix_enabled())
5364 		return -EINVAL;
5365 
5366 	/* If we're a nested hypervisor, we currently only support radix */
5367 	if (kvmhv_on_pseries() && !radix)
5368 		return -EINVAL;
5369 
5370 	mutex_lock(&kvm->arch.mmu_setup_lock);
5371 	if (radix != kvm_is_radix(kvm)) {
5372 		if (kvm->arch.mmu_ready) {
5373 			kvm->arch.mmu_ready = 0;
5374 			/* order mmu_ready vs. vcpus_running */
5375 			smp_mb();
5376 			if (atomic_read(&kvm->arch.vcpus_running)) {
5377 				kvm->arch.mmu_ready = 1;
5378 				err = -EBUSY;
5379 				goto out_unlock;
5380 			}
5381 		}
5382 		if (radix)
5383 			err = kvmppc_switch_mmu_to_radix(kvm);
5384 		else
5385 			err = kvmppc_switch_mmu_to_hpt(kvm);
5386 		if (err)
5387 			goto out_unlock;
5388 	}
5389 
5390 	kvm->arch.process_table = cfg->process_table;
5391 	kvmppc_setup_partition_table(kvm);
5392 
5393 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5394 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5395 	err = 0;
5396 
5397  out_unlock:
5398 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5399 	return err;
5400 }
5401 
5402 static int kvmhv_enable_nested(struct kvm *kvm)
5403 {
5404 	if (!nested)
5405 		return -EPERM;
5406 	if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5407 		return -ENODEV;
5408 
5409 	/* kvm == NULL means the caller is testing if the capability exists */
5410 	if (kvm)
5411 		kvm->arch.nested_enable = true;
5412 	return 0;
5413 }
5414 
5415 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5416 				 int size)
5417 {
5418 	int rc = -EINVAL;
5419 
5420 	if (kvmhv_vcpu_is_radix(vcpu)) {
5421 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5422 
5423 		if (rc > 0)
5424 			rc = -EINVAL;
5425 	}
5426 
5427 	/* For now quadrants are the only way to access nested guest memory */
5428 	if (rc && vcpu->arch.nested)
5429 		rc = -EAGAIN;
5430 
5431 	return rc;
5432 }
5433 
5434 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5435 				int size)
5436 {
5437 	int rc = -EINVAL;
5438 
5439 	if (kvmhv_vcpu_is_radix(vcpu)) {
5440 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5441 
5442 		if (rc > 0)
5443 			rc = -EINVAL;
5444 	}
5445 
5446 	/* For now quadrants are the only way to access nested guest memory */
5447 	if (rc && vcpu->arch.nested)
5448 		rc = -EAGAIN;
5449 
5450 	return rc;
5451 }
5452 
5453 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5454 {
5455 	unpin_vpa(kvm, vpa);
5456 	vpa->gpa = 0;
5457 	vpa->pinned_addr = NULL;
5458 	vpa->dirty = false;
5459 	vpa->update_pending = 0;
5460 }
5461 
5462 /*
5463  * Enable a guest to become a secure VM, or test whether
5464  * that could be enabled.
5465  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5466  * tested (kvm == NULL) or enabled (kvm != NULL).
5467  */
5468 static int kvmhv_enable_svm(struct kvm *kvm)
5469 {
5470 	if (!kvmppc_uvmem_available())
5471 		return -EINVAL;
5472 	if (kvm)
5473 		kvm->arch.svm_enabled = 1;
5474 	return 0;
5475 }
5476 
5477 /*
5478  *  IOCTL handler to turn off secure mode of guest
5479  *
5480  * - Release all device pages
5481  * - Issue ucall to terminate the guest on the UV side
5482  * - Unpin the VPA pages.
5483  * - Reinit the partition scoped page tables
5484  */
5485 static int kvmhv_svm_off(struct kvm *kvm)
5486 {
5487 	struct kvm_vcpu *vcpu;
5488 	int mmu_was_ready;
5489 	int srcu_idx;
5490 	int ret = 0;
5491 	int i;
5492 
5493 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5494 		return ret;
5495 
5496 	mutex_lock(&kvm->arch.mmu_setup_lock);
5497 	mmu_was_ready = kvm->arch.mmu_ready;
5498 	if (kvm->arch.mmu_ready) {
5499 		kvm->arch.mmu_ready = 0;
5500 		/* order mmu_ready vs. vcpus_running */
5501 		smp_mb();
5502 		if (atomic_read(&kvm->arch.vcpus_running)) {
5503 			kvm->arch.mmu_ready = 1;
5504 			ret = -EBUSY;
5505 			goto out;
5506 		}
5507 	}
5508 
5509 	srcu_idx = srcu_read_lock(&kvm->srcu);
5510 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5511 		struct kvm_memory_slot *memslot;
5512 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5513 
5514 		if (!slots)
5515 			continue;
5516 
5517 		kvm_for_each_memslot(memslot, slots) {
5518 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
5519 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5520 		}
5521 	}
5522 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5523 
5524 	ret = uv_svm_terminate(kvm->arch.lpid);
5525 	if (ret != U_SUCCESS) {
5526 		ret = -EINVAL;
5527 		goto out;
5528 	}
5529 
5530 	/*
5531 	 * When secure guest is reset, all the guest pages are sent
5532 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
5533 	 * chance to run and unpin their VPA pages. Unpinning of all
5534 	 * VPA pages is done here explicitly so that VPA pages
5535 	 * can be migrated to the secure side.
5536 	 *
5537 	 * This is required to for the secure SMP guest to reboot
5538 	 * correctly.
5539 	 */
5540 	kvm_for_each_vcpu(i, vcpu, kvm) {
5541 		spin_lock(&vcpu->arch.vpa_update_lock);
5542 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5543 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5544 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5545 		spin_unlock(&vcpu->arch.vpa_update_lock);
5546 	}
5547 
5548 	kvmppc_setup_partition_table(kvm);
5549 	kvm->arch.secure_guest = 0;
5550 	kvm->arch.mmu_ready = mmu_was_ready;
5551 out:
5552 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5553 	return ret;
5554 }
5555 
5556 static struct kvmppc_ops kvm_ops_hv = {
5557 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5558 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5559 	.get_one_reg = kvmppc_get_one_reg_hv,
5560 	.set_one_reg = kvmppc_set_one_reg_hv,
5561 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
5562 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
5563 	.inject_interrupt = kvmppc_inject_interrupt_hv,
5564 	.set_msr     = kvmppc_set_msr_hv,
5565 	.vcpu_run    = kvmppc_vcpu_run_hv,
5566 	.vcpu_create = kvmppc_core_vcpu_create_hv,
5567 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
5568 	.check_requests = kvmppc_core_check_requests_hv,
5569 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5570 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
5571 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5572 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5573 	.unmap_hva_range = kvm_unmap_hva_range_hv,
5574 	.age_hva  = kvm_age_hva_hv,
5575 	.test_age_hva = kvm_test_age_hva_hv,
5576 	.set_spte_hva = kvm_set_spte_hva_hv,
5577 	.free_memslot = kvmppc_core_free_memslot_hv,
5578 	.init_vm =  kvmppc_core_init_vm_hv,
5579 	.destroy_vm = kvmppc_core_destroy_vm_hv,
5580 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5581 	.emulate_op = kvmppc_core_emulate_op_hv,
5582 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5583 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5584 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5585 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5586 	.hcall_implemented = kvmppc_hcall_impl_hv,
5587 #ifdef CONFIG_KVM_XICS
5588 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5589 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5590 #endif
5591 	.configure_mmu = kvmhv_configure_mmu,
5592 	.get_rmmu_info = kvmhv_get_rmmu_info,
5593 	.set_smt_mode = kvmhv_set_smt_mode,
5594 	.enable_nested = kvmhv_enable_nested,
5595 	.load_from_eaddr = kvmhv_load_from_eaddr,
5596 	.store_to_eaddr = kvmhv_store_to_eaddr,
5597 	.enable_svm = kvmhv_enable_svm,
5598 	.svm_off = kvmhv_svm_off,
5599 };
5600 
5601 static int kvm_init_subcore_bitmap(void)
5602 {
5603 	int i, j;
5604 	int nr_cores = cpu_nr_cores();
5605 	struct sibling_subcore_state *sibling_subcore_state;
5606 
5607 	for (i = 0; i < nr_cores; i++) {
5608 		int first_cpu = i * threads_per_core;
5609 		int node = cpu_to_node(first_cpu);
5610 
5611 		/* Ignore if it is already allocated. */
5612 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
5613 			continue;
5614 
5615 		sibling_subcore_state =
5616 			kzalloc_node(sizeof(struct sibling_subcore_state),
5617 							GFP_KERNEL, node);
5618 		if (!sibling_subcore_state)
5619 			return -ENOMEM;
5620 
5621 
5622 		for (j = 0; j < threads_per_core; j++) {
5623 			int cpu = first_cpu + j;
5624 
5625 			paca_ptrs[cpu]->sibling_subcore_state =
5626 						sibling_subcore_state;
5627 		}
5628 	}
5629 	return 0;
5630 }
5631 
5632 static int kvmppc_radix_possible(void)
5633 {
5634 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5635 }
5636 
5637 static int kvmppc_book3s_init_hv(void)
5638 {
5639 	int r;
5640 
5641 	if (!tlbie_capable) {
5642 		pr_err("KVM-HV: Host does not support TLBIE\n");
5643 		return -ENODEV;
5644 	}
5645 
5646 	/*
5647 	 * FIXME!! Do we need to check on all cpus ?
5648 	 */
5649 	r = kvmppc_core_check_processor_compat_hv();
5650 	if (r < 0)
5651 		return -ENODEV;
5652 
5653 	r = kvmhv_nested_init();
5654 	if (r)
5655 		return r;
5656 
5657 	r = kvm_init_subcore_bitmap();
5658 	if (r)
5659 		return r;
5660 
5661 	/*
5662 	 * We need a way of accessing the XICS interrupt controller,
5663 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5664 	 * indirectly, via OPAL.
5665 	 */
5666 #ifdef CONFIG_SMP
5667 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
5668 	    !local_paca->kvm_hstate.xics_phys) {
5669 		struct device_node *np;
5670 
5671 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5672 		if (!np) {
5673 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5674 			return -ENODEV;
5675 		}
5676 		/* presence of intc confirmed - node can be dropped again */
5677 		of_node_put(np);
5678 	}
5679 #endif
5680 
5681 	kvm_ops_hv.owner = THIS_MODULE;
5682 	kvmppc_hv_ops = &kvm_ops_hv;
5683 
5684 	init_default_hcalls();
5685 
5686 	init_vcore_lists();
5687 
5688 	r = kvmppc_mmu_hv_init();
5689 	if (r)
5690 		return r;
5691 
5692 	if (kvmppc_radix_possible())
5693 		r = kvmppc_radix_init();
5694 
5695 	/*
5696 	 * POWER9 chips before version 2.02 can't have some threads in
5697 	 * HPT mode and some in radix mode on the same core.
5698 	 */
5699 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5700 		unsigned int pvr = mfspr(SPRN_PVR);
5701 		if ((pvr >> 16) == PVR_POWER9 &&
5702 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5703 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5704 			no_mixing_hpt_and_radix = true;
5705 	}
5706 
5707 	r = kvmppc_uvmem_init();
5708 	if (r < 0)
5709 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5710 
5711 	return r;
5712 }
5713 
5714 static void kvmppc_book3s_exit_hv(void)
5715 {
5716 	kvmppc_uvmem_free();
5717 	kvmppc_free_host_rm_ops();
5718 	if (kvmppc_radix_possible())
5719 		kvmppc_radix_exit();
5720 	kvmppc_hv_ops = NULL;
5721 	kvmhv_nested_exit();
5722 }
5723 
5724 module_init(kvmppc_book3s_init_hv);
5725 module_exit(kvmppc_book3s_exit_hv);
5726 MODULE_LICENSE("GPL");
5727 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5728 MODULE_ALIAS("devname:kvm");
5729