xref: /linux/arch/powerpc/kvm/book3s_hv.c (revision 666ed8bfd1de3b091cf32ca03b651757dd86cfff)
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 
78 #include "book3s.h"
79 
80 #define CREATE_TRACE_POINTS
81 #include "trace_hv.h"
82 
83 /* #define EXIT_DEBUG */
84 /* #define EXIT_DEBUG_SIMPLE */
85 /* #define EXIT_DEBUG_INT */
86 
87 /* Used to indicate that a guest page fault needs to be handled */
88 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
89 /* Used to indicate that a guest passthrough interrupt needs to be handled */
90 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
91 
92 /* Used as a "null" value for timebase values */
93 #define TB_NIL	(~(u64)0)
94 
95 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
96 
97 static int dynamic_mt_modes = 6;
98 module_param(dynamic_mt_modes, int, 0644);
99 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
100 static int target_smt_mode;
101 module_param(target_smt_mode, int, 0644);
102 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
103 
104 static bool indep_threads_mode = true;
105 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
106 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
107 
108 static bool one_vm_per_core;
109 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
110 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
111 
112 #ifdef CONFIG_KVM_XICS
113 static struct kernel_param_ops module_param_ops = {
114 	.set = param_set_int,
115 	.get = param_get_int,
116 };
117 
118 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
119 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
120 
121 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
122 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
123 #endif
124 
125 /* If set, guests are allowed to create and control nested guests */
126 static bool nested = true;
127 module_param(nested, bool, S_IRUGO | S_IWUSR);
128 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
129 
130 static inline bool nesting_enabled(struct kvm *kvm)
131 {
132 	return kvm->arch.nested_enable && kvm_is_radix(kvm);
133 }
134 
135 /* If set, the threads on each CPU core have to be in the same MMU mode */
136 static bool no_mixing_hpt_and_radix;
137 
138 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
139 
140 /*
141  * RWMR values for POWER8.  These control the rate at which PURR
142  * and SPURR count and should be set according to the number of
143  * online threads in the vcore being run.
144  */
145 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
146 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
147 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
148 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
149 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
150 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
151 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
152 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
153 
154 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
155 	RWMR_RPA_P8_1THREAD,
156 	RWMR_RPA_P8_1THREAD,
157 	RWMR_RPA_P8_2THREAD,
158 	RWMR_RPA_P8_3THREAD,
159 	RWMR_RPA_P8_4THREAD,
160 	RWMR_RPA_P8_5THREAD,
161 	RWMR_RPA_P8_6THREAD,
162 	RWMR_RPA_P8_7THREAD,
163 	RWMR_RPA_P8_8THREAD,
164 };
165 
166 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
167 		int *ip)
168 {
169 	int i = *ip;
170 	struct kvm_vcpu *vcpu;
171 
172 	while (++i < MAX_SMT_THREADS) {
173 		vcpu = READ_ONCE(vc->runnable_threads[i]);
174 		if (vcpu) {
175 			*ip = i;
176 			return vcpu;
177 		}
178 	}
179 	return NULL;
180 }
181 
182 /* Used to traverse the list of runnable threads for a given vcore */
183 #define for_each_runnable_thread(i, vcpu, vc) \
184 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
185 
186 static bool kvmppc_ipi_thread(int cpu)
187 {
188 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
189 
190 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
191 	if (kvmhv_on_pseries())
192 		return false;
193 
194 	/* On POWER9 we can use msgsnd to IPI any cpu */
195 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
196 		msg |= get_hard_smp_processor_id(cpu);
197 		smp_mb();
198 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
199 		return true;
200 	}
201 
202 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
203 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
204 		preempt_disable();
205 		if (cpu_first_thread_sibling(cpu) ==
206 		    cpu_first_thread_sibling(smp_processor_id())) {
207 			msg |= cpu_thread_in_core(cpu);
208 			smp_mb();
209 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
210 			preempt_enable();
211 			return true;
212 		}
213 		preempt_enable();
214 	}
215 
216 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
217 	if (cpu >= 0 && cpu < nr_cpu_ids) {
218 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
219 			xics_wake_cpu(cpu);
220 			return true;
221 		}
222 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
223 		return true;
224 	}
225 #endif
226 
227 	return false;
228 }
229 
230 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
231 {
232 	int cpu;
233 	struct swait_queue_head *wqp;
234 
235 	wqp = kvm_arch_vcpu_wq(vcpu);
236 	if (swq_has_sleeper(wqp)) {
237 		swake_up_one(wqp);
238 		++vcpu->stat.halt_wakeup;
239 	}
240 
241 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
242 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
243 		return;
244 
245 	/* CPU points to the first thread of the core */
246 	cpu = vcpu->cpu;
247 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
248 		smp_send_reschedule(cpu);
249 }
250 
251 /*
252  * We use the vcpu_load/put functions to measure stolen time.
253  * Stolen time is counted as time when either the vcpu is able to
254  * run as part of a virtual core, but the task running the vcore
255  * is preempted or sleeping, or when the vcpu needs something done
256  * in the kernel by the task running the vcpu, but that task is
257  * preempted or sleeping.  Those two things have to be counted
258  * separately, since one of the vcpu tasks will take on the job
259  * of running the core, and the other vcpu tasks in the vcore will
260  * sleep waiting for it to do that, but that sleep shouldn't count
261  * as stolen time.
262  *
263  * Hence we accumulate stolen time when the vcpu can run as part of
264  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
265  * needs its task to do other things in the kernel (for example,
266  * service a page fault) in busy_stolen.  We don't accumulate
267  * stolen time for a vcore when it is inactive, or for a vcpu
268  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
269  * a misnomer; it means that the vcpu task is not executing in
270  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
271  * the kernel.  We don't have any way of dividing up that time
272  * between time that the vcpu is genuinely stopped, time that
273  * the task is actively working on behalf of the vcpu, and time
274  * that the task is preempted, so we don't count any of it as
275  * stolen.
276  *
277  * Updates to busy_stolen are protected by arch.tbacct_lock;
278  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
279  * lock.  The stolen times are measured in units of timebase ticks.
280  * (Note that the != TB_NIL checks below are purely defensive;
281  * they should never fail.)
282  */
283 
284 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
285 {
286 	unsigned long flags;
287 
288 	spin_lock_irqsave(&vc->stoltb_lock, flags);
289 	vc->preempt_tb = mftb();
290 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
291 }
292 
293 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
294 {
295 	unsigned long flags;
296 
297 	spin_lock_irqsave(&vc->stoltb_lock, flags);
298 	if (vc->preempt_tb != TB_NIL) {
299 		vc->stolen_tb += mftb() - vc->preempt_tb;
300 		vc->preempt_tb = TB_NIL;
301 	}
302 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
303 }
304 
305 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
306 {
307 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
308 	unsigned long flags;
309 
310 	/*
311 	 * We can test vc->runner without taking the vcore lock,
312 	 * because only this task ever sets vc->runner to this
313 	 * vcpu, and once it is set to this vcpu, only this task
314 	 * ever sets it to NULL.
315 	 */
316 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
317 		kvmppc_core_end_stolen(vc);
318 
319 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
320 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
321 	    vcpu->arch.busy_preempt != TB_NIL) {
322 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
323 		vcpu->arch.busy_preempt = TB_NIL;
324 	}
325 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
326 }
327 
328 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
329 {
330 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
331 	unsigned long flags;
332 
333 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
334 		kvmppc_core_start_stolen(vc);
335 
336 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
337 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
338 		vcpu->arch.busy_preempt = mftb();
339 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
340 }
341 
342 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
343 {
344 	vcpu->arch.pvr = pvr;
345 }
346 
347 /* Dummy value used in computing PCR value below */
348 #define PCR_ARCH_300	(PCR_ARCH_207 << 1)
349 
350 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
351 {
352 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
353 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
354 
355 	/* We can (emulate) our own architecture version and anything older */
356 	if (cpu_has_feature(CPU_FTR_ARCH_300))
357 		host_pcr_bit = PCR_ARCH_300;
358 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
359 		host_pcr_bit = PCR_ARCH_207;
360 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
361 		host_pcr_bit = PCR_ARCH_206;
362 	else
363 		host_pcr_bit = PCR_ARCH_205;
364 
365 	/* Determine lowest PCR bit needed to run guest in given PVR level */
366 	guest_pcr_bit = host_pcr_bit;
367 	if (arch_compat) {
368 		switch (arch_compat) {
369 		case PVR_ARCH_205:
370 			guest_pcr_bit = PCR_ARCH_205;
371 			break;
372 		case PVR_ARCH_206:
373 		case PVR_ARCH_206p:
374 			guest_pcr_bit = PCR_ARCH_206;
375 			break;
376 		case PVR_ARCH_207:
377 			guest_pcr_bit = PCR_ARCH_207;
378 			break;
379 		case PVR_ARCH_300:
380 			guest_pcr_bit = PCR_ARCH_300;
381 			break;
382 		default:
383 			return -EINVAL;
384 		}
385 	}
386 
387 	/* Check requested PCR bits don't exceed our capabilities */
388 	if (guest_pcr_bit > host_pcr_bit)
389 		return -EINVAL;
390 
391 	spin_lock(&vc->lock);
392 	vc->arch_compat = arch_compat;
393 	/*
394 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
395 	 * Also set all reserved PCR bits
396 	 */
397 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
398 	spin_unlock(&vc->lock);
399 
400 	return 0;
401 }
402 
403 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
404 {
405 	int r;
406 
407 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
408 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
409 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
410 	for (r = 0; r < 16; ++r)
411 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
412 		       r, kvmppc_get_gpr(vcpu, r),
413 		       r+16, kvmppc_get_gpr(vcpu, r+16));
414 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
415 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
416 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
417 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
418 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
419 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
420 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
421 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
422 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
423 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
424 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
425 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
426 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
427 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
428 	for (r = 0; r < vcpu->arch.slb_max; ++r)
429 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
430 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
431 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
432 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
433 	       vcpu->arch.last_inst);
434 }
435 
436 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
437 {
438 	return kvm_get_vcpu_by_id(kvm, id);
439 }
440 
441 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
442 {
443 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
444 	vpa->yield_count = cpu_to_be32(1);
445 }
446 
447 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
448 		   unsigned long addr, unsigned long len)
449 {
450 	/* check address is cacheline aligned */
451 	if (addr & (L1_CACHE_BYTES - 1))
452 		return -EINVAL;
453 	spin_lock(&vcpu->arch.vpa_update_lock);
454 	if (v->next_gpa != addr || v->len != len) {
455 		v->next_gpa = addr;
456 		v->len = addr ? len : 0;
457 		v->update_pending = 1;
458 	}
459 	spin_unlock(&vcpu->arch.vpa_update_lock);
460 	return 0;
461 }
462 
463 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
464 struct reg_vpa {
465 	u32 dummy;
466 	union {
467 		__be16 hword;
468 		__be32 word;
469 	} length;
470 };
471 
472 static int vpa_is_registered(struct kvmppc_vpa *vpap)
473 {
474 	if (vpap->update_pending)
475 		return vpap->next_gpa != 0;
476 	return vpap->pinned_addr != NULL;
477 }
478 
479 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
480 				       unsigned long flags,
481 				       unsigned long vcpuid, unsigned long vpa)
482 {
483 	struct kvm *kvm = vcpu->kvm;
484 	unsigned long len, nb;
485 	void *va;
486 	struct kvm_vcpu *tvcpu;
487 	int err;
488 	int subfunc;
489 	struct kvmppc_vpa *vpap;
490 
491 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
492 	if (!tvcpu)
493 		return H_PARAMETER;
494 
495 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
496 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
497 	    subfunc == H_VPA_REG_SLB) {
498 		/* Registering new area - address must be cache-line aligned */
499 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
500 			return H_PARAMETER;
501 
502 		/* convert logical addr to kernel addr and read length */
503 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
504 		if (va == NULL)
505 			return H_PARAMETER;
506 		if (subfunc == H_VPA_REG_VPA)
507 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
508 		else
509 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
510 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
511 
512 		/* Check length */
513 		if (len > nb || len < sizeof(struct reg_vpa))
514 			return H_PARAMETER;
515 	} else {
516 		vpa = 0;
517 		len = 0;
518 	}
519 
520 	err = H_PARAMETER;
521 	vpap = NULL;
522 	spin_lock(&tvcpu->arch.vpa_update_lock);
523 
524 	switch (subfunc) {
525 	case H_VPA_REG_VPA:		/* register VPA */
526 		/*
527 		 * The size of our lppaca is 1kB because of the way we align
528 		 * it for the guest to avoid crossing a 4kB boundary. We only
529 		 * use 640 bytes of the structure though, so we should accept
530 		 * clients that set a size of 640.
531 		 */
532 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
533 		if (len < sizeof(struct lppaca))
534 			break;
535 		vpap = &tvcpu->arch.vpa;
536 		err = 0;
537 		break;
538 
539 	case H_VPA_REG_DTL:		/* register DTL */
540 		if (len < sizeof(struct dtl_entry))
541 			break;
542 		len -= len % sizeof(struct dtl_entry);
543 
544 		/* Check that they have previously registered a VPA */
545 		err = H_RESOURCE;
546 		if (!vpa_is_registered(&tvcpu->arch.vpa))
547 			break;
548 
549 		vpap = &tvcpu->arch.dtl;
550 		err = 0;
551 		break;
552 
553 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
554 		/* Check that they have previously registered a VPA */
555 		err = H_RESOURCE;
556 		if (!vpa_is_registered(&tvcpu->arch.vpa))
557 			break;
558 
559 		vpap = &tvcpu->arch.slb_shadow;
560 		err = 0;
561 		break;
562 
563 	case H_VPA_DEREG_VPA:		/* deregister VPA */
564 		/* Check they don't still have a DTL or SLB buf registered */
565 		err = H_RESOURCE;
566 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
567 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
568 			break;
569 
570 		vpap = &tvcpu->arch.vpa;
571 		err = 0;
572 		break;
573 
574 	case H_VPA_DEREG_DTL:		/* deregister DTL */
575 		vpap = &tvcpu->arch.dtl;
576 		err = 0;
577 		break;
578 
579 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
580 		vpap = &tvcpu->arch.slb_shadow;
581 		err = 0;
582 		break;
583 	}
584 
585 	if (vpap) {
586 		vpap->next_gpa = vpa;
587 		vpap->len = len;
588 		vpap->update_pending = 1;
589 	}
590 
591 	spin_unlock(&tvcpu->arch.vpa_update_lock);
592 
593 	return err;
594 }
595 
596 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
597 {
598 	struct kvm *kvm = vcpu->kvm;
599 	void *va;
600 	unsigned long nb;
601 	unsigned long gpa;
602 
603 	/*
604 	 * We need to pin the page pointed to by vpap->next_gpa,
605 	 * but we can't call kvmppc_pin_guest_page under the lock
606 	 * as it does get_user_pages() and down_read().  So we
607 	 * have to drop the lock, pin the page, then get the lock
608 	 * again and check that a new area didn't get registered
609 	 * in the meantime.
610 	 */
611 	for (;;) {
612 		gpa = vpap->next_gpa;
613 		spin_unlock(&vcpu->arch.vpa_update_lock);
614 		va = NULL;
615 		nb = 0;
616 		if (gpa)
617 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
618 		spin_lock(&vcpu->arch.vpa_update_lock);
619 		if (gpa == vpap->next_gpa)
620 			break;
621 		/* sigh... unpin that one and try again */
622 		if (va)
623 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
624 	}
625 
626 	vpap->update_pending = 0;
627 	if (va && nb < vpap->len) {
628 		/*
629 		 * If it's now too short, it must be that userspace
630 		 * has changed the mappings underlying guest memory,
631 		 * so unregister the region.
632 		 */
633 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
634 		va = NULL;
635 	}
636 	if (vpap->pinned_addr)
637 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
638 					vpap->dirty);
639 	vpap->gpa = gpa;
640 	vpap->pinned_addr = va;
641 	vpap->dirty = false;
642 	if (va)
643 		vpap->pinned_end = va + vpap->len;
644 }
645 
646 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
647 {
648 	if (!(vcpu->arch.vpa.update_pending ||
649 	      vcpu->arch.slb_shadow.update_pending ||
650 	      vcpu->arch.dtl.update_pending))
651 		return;
652 
653 	spin_lock(&vcpu->arch.vpa_update_lock);
654 	if (vcpu->arch.vpa.update_pending) {
655 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
656 		if (vcpu->arch.vpa.pinned_addr)
657 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
658 	}
659 	if (vcpu->arch.dtl.update_pending) {
660 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
661 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
662 		vcpu->arch.dtl_index = 0;
663 	}
664 	if (vcpu->arch.slb_shadow.update_pending)
665 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
666 	spin_unlock(&vcpu->arch.vpa_update_lock);
667 }
668 
669 /*
670  * Return the accumulated stolen time for the vcore up until `now'.
671  * The caller should hold the vcore lock.
672  */
673 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
674 {
675 	u64 p;
676 	unsigned long flags;
677 
678 	spin_lock_irqsave(&vc->stoltb_lock, flags);
679 	p = vc->stolen_tb;
680 	if (vc->vcore_state != VCORE_INACTIVE &&
681 	    vc->preempt_tb != TB_NIL)
682 		p += now - vc->preempt_tb;
683 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
684 	return p;
685 }
686 
687 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
688 				    struct kvmppc_vcore *vc)
689 {
690 	struct dtl_entry *dt;
691 	struct lppaca *vpa;
692 	unsigned long stolen;
693 	unsigned long core_stolen;
694 	u64 now;
695 	unsigned long flags;
696 
697 	dt = vcpu->arch.dtl_ptr;
698 	vpa = vcpu->arch.vpa.pinned_addr;
699 	now = mftb();
700 	core_stolen = vcore_stolen_time(vc, now);
701 	stolen = core_stolen - vcpu->arch.stolen_logged;
702 	vcpu->arch.stolen_logged = core_stolen;
703 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
704 	stolen += vcpu->arch.busy_stolen;
705 	vcpu->arch.busy_stolen = 0;
706 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
707 	if (!dt || !vpa)
708 		return;
709 	memset(dt, 0, sizeof(struct dtl_entry));
710 	dt->dispatch_reason = 7;
711 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
712 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
713 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
714 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
715 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
716 	++dt;
717 	if (dt == vcpu->arch.dtl.pinned_end)
718 		dt = vcpu->arch.dtl.pinned_addr;
719 	vcpu->arch.dtl_ptr = dt;
720 	/* order writing *dt vs. writing vpa->dtl_idx */
721 	smp_wmb();
722 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
723 	vcpu->arch.dtl.dirty = true;
724 }
725 
726 /* See if there is a doorbell interrupt pending for a vcpu */
727 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
728 {
729 	int thr;
730 	struct kvmppc_vcore *vc;
731 
732 	if (vcpu->arch.doorbell_request)
733 		return true;
734 	/*
735 	 * Ensure that the read of vcore->dpdes comes after the read
736 	 * of vcpu->doorbell_request.  This barrier matches the
737 	 * smp_wmb() in kvmppc_guest_entry_inject().
738 	 */
739 	smp_rmb();
740 	vc = vcpu->arch.vcore;
741 	thr = vcpu->vcpu_id - vc->first_vcpuid;
742 	return !!(vc->dpdes & (1 << thr));
743 }
744 
745 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
746 {
747 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
748 		return true;
749 	if ((!vcpu->arch.vcore->arch_compat) &&
750 	    cpu_has_feature(CPU_FTR_ARCH_207S))
751 		return true;
752 	return false;
753 }
754 
755 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
756 			     unsigned long resource, unsigned long value1,
757 			     unsigned long value2)
758 {
759 	switch (resource) {
760 	case H_SET_MODE_RESOURCE_SET_CIABR:
761 		if (!kvmppc_power8_compatible(vcpu))
762 			return H_P2;
763 		if (value2)
764 			return H_P4;
765 		if (mflags)
766 			return H_UNSUPPORTED_FLAG_START;
767 		/* Guests can't breakpoint the hypervisor */
768 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
769 			return H_P3;
770 		vcpu->arch.ciabr  = value1;
771 		return H_SUCCESS;
772 	case H_SET_MODE_RESOURCE_SET_DAWR:
773 		if (!kvmppc_power8_compatible(vcpu))
774 			return H_P2;
775 		if (!ppc_breakpoint_available())
776 			return H_P2;
777 		if (mflags)
778 			return H_UNSUPPORTED_FLAG_START;
779 		if (value2 & DABRX_HYP)
780 			return H_P4;
781 		vcpu->arch.dawr  = value1;
782 		vcpu->arch.dawrx = value2;
783 		return H_SUCCESS;
784 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
785 		/* KVM does not support mflags=2 (AIL=2) */
786 		if (mflags != 0 && mflags != 3)
787 			return H_UNSUPPORTED_FLAG_START;
788 		return H_TOO_HARD;
789 	default:
790 		return H_TOO_HARD;
791 	}
792 }
793 
794 /* Copy guest memory in place - must reside within a single memslot */
795 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
796 				  unsigned long len)
797 {
798 	struct kvm_memory_slot *to_memslot = NULL;
799 	struct kvm_memory_slot *from_memslot = NULL;
800 	unsigned long to_addr, from_addr;
801 	int r;
802 
803 	/* Get HPA for from address */
804 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
805 	if (!from_memslot)
806 		return -EFAULT;
807 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
808 			     << PAGE_SHIFT))
809 		return -EINVAL;
810 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
811 	if (kvm_is_error_hva(from_addr))
812 		return -EFAULT;
813 	from_addr |= (from & (PAGE_SIZE - 1));
814 
815 	/* Get HPA for to address */
816 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
817 	if (!to_memslot)
818 		return -EFAULT;
819 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
820 			   << PAGE_SHIFT))
821 		return -EINVAL;
822 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
823 	if (kvm_is_error_hva(to_addr))
824 		return -EFAULT;
825 	to_addr |= (to & (PAGE_SIZE - 1));
826 
827 	/* Perform copy */
828 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
829 			     len);
830 	if (r)
831 		return -EFAULT;
832 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
833 	return 0;
834 }
835 
836 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
837 			       unsigned long dest, unsigned long src)
838 {
839 	u64 pg_sz = SZ_4K;		/* 4K page size */
840 	u64 pg_mask = SZ_4K - 1;
841 	int ret;
842 
843 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
844 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
845 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
846 		return H_PARAMETER;
847 
848 	/* dest (and src if copy_page flag set) must be page aligned */
849 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
850 		return H_PARAMETER;
851 
852 	/* zero and/or copy the page as determined by the flags */
853 	if (flags & H_COPY_PAGE) {
854 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
855 		if (ret < 0)
856 			return H_PARAMETER;
857 	} else if (flags & H_ZERO_PAGE) {
858 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
859 		if (ret < 0)
860 			return H_PARAMETER;
861 	}
862 
863 	/* We can ignore the remaining flags */
864 
865 	return H_SUCCESS;
866 }
867 
868 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
869 {
870 	struct kvmppc_vcore *vcore = target->arch.vcore;
871 
872 	/*
873 	 * We expect to have been called by the real mode handler
874 	 * (kvmppc_rm_h_confer()) which would have directly returned
875 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
876 	 * have useful work to do and should not confer) so we don't
877 	 * recheck that here.
878 	 */
879 
880 	spin_lock(&vcore->lock);
881 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
882 	    vcore->vcore_state != VCORE_INACTIVE &&
883 	    vcore->runner)
884 		target = vcore->runner;
885 	spin_unlock(&vcore->lock);
886 
887 	return kvm_vcpu_yield_to(target);
888 }
889 
890 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
891 {
892 	int yield_count = 0;
893 	struct lppaca *lppaca;
894 
895 	spin_lock(&vcpu->arch.vpa_update_lock);
896 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
897 	if (lppaca)
898 		yield_count = be32_to_cpu(lppaca->yield_count);
899 	spin_unlock(&vcpu->arch.vpa_update_lock);
900 	return yield_count;
901 }
902 
903 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
904 {
905 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
906 	unsigned long target, ret = H_SUCCESS;
907 	int yield_count;
908 	struct kvm_vcpu *tvcpu;
909 	int idx, rc;
910 
911 	if (req <= MAX_HCALL_OPCODE &&
912 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
913 		return RESUME_HOST;
914 
915 	switch (req) {
916 	case H_CEDE:
917 		break;
918 	case H_PROD:
919 		target = kvmppc_get_gpr(vcpu, 4);
920 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
921 		if (!tvcpu) {
922 			ret = H_PARAMETER;
923 			break;
924 		}
925 		tvcpu->arch.prodded = 1;
926 		smp_mb();
927 		if (tvcpu->arch.ceded)
928 			kvmppc_fast_vcpu_kick_hv(tvcpu);
929 		break;
930 	case H_CONFER:
931 		target = kvmppc_get_gpr(vcpu, 4);
932 		if (target == -1)
933 			break;
934 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
935 		if (!tvcpu) {
936 			ret = H_PARAMETER;
937 			break;
938 		}
939 		yield_count = kvmppc_get_gpr(vcpu, 5);
940 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
941 			break;
942 		kvm_arch_vcpu_yield_to(tvcpu);
943 		break;
944 	case H_REGISTER_VPA:
945 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
946 					kvmppc_get_gpr(vcpu, 5),
947 					kvmppc_get_gpr(vcpu, 6));
948 		break;
949 	case H_RTAS:
950 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
951 			return RESUME_HOST;
952 
953 		idx = srcu_read_lock(&vcpu->kvm->srcu);
954 		rc = kvmppc_rtas_hcall(vcpu);
955 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
956 
957 		if (rc == -ENOENT)
958 			return RESUME_HOST;
959 		else if (rc == 0)
960 			break;
961 
962 		/* Send the error out to userspace via KVM_RUN */
963 		return rc;
964 	case H_LOGICAL_CI_LOAD:
965 		ret = kvmppc_h_logical_ci_load(vcpu);
966 		if (ret == H_TOO_HARD)
967 			return RESUME_HOST;
968 		break;
969 	case H_LOGICAL_CI_STORE:
970 		ret = kvmppc_h_logical_ci_store(vcpu);
971 		if (ret == H_TOO_HARD)
972 			return RESUME_HOST;
973 		break;
974 	case H_SET_MODE:
975 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
976 					kvmppc_get_gpr(vcpu, 5),
977 					kvmppc_get_gpr(vcpu, 6),
978 					kvmppc_get_gpr(vcpu, 7));
979 		if (ret == H_TOO_HARD)
980 			return RESUME_HOST;
981 		break;
982 	case H_XIRR:
983 	case H_CPPR:
984 	case H_EOI:
985 	case H_IPI:
986 	case H_IPOLL:
987 	case H_XIRR_X:
988 		if (kvmppc_xics_enabled(vcpu)) {
989 			if (xics_on_xive()) {
990 				ret = H_NOT_AVAILABLE;
991 				return RESUME_GUEST;
992 			}
993 			ret = kvmppc_xics_hcall(vcpu, req);
994 			break;
995 		}
996 		return RESUME_HOST;
997 	case H_SET_DABR:
998 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
999 		break;
1000 	case H_SET_XDABR:
1001 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1002 						kvmppc_get_gpr(vcpu, 5));
1003 		break;
1004 #ifdef CONFIG_SPAPR_TCE_IOMMU
1005 	case H_GET_TCE:
1006 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1007 						kvmppc_get_gpr(vcpu, 5));
1008 		if (ret == H_TOO_HARD)
1009 			return RESUME_HOST;
1010 		break;
1011 	case H_PUT_TCE:
1012 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1013 						kvmppc_get_gpr(vcpu, 5),
1014 						kvmppc_get_gpr(vcpu, 6));
1015 		if (ret == H_TOO_HARD)
1016 			return RESUME_HOST;
1017 		break;
1018 	case H_PUT_TCE_INDIRECT:
1019 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1020 						kvmppc_get_gpr(vcpu, 5),
1021 						kvmppc_get_gpr(vcpu, 6),
1022 						kvmppc_get_gpr(vcpu, 7));
1023 		if (ret == H_TOO_HARD)
1024 			return RESUME_HOST;
1025 		break;
1026 	case H_STUFF_TCE:
1027 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1028 						kvmppc_get_gpr(vcpu, 5),
1029 						kvmppc_get_gpr(vcpu, 6),
1030 						kvmppc_get_gpr(vcpu, 7));
1031 		if (ret == H_TOO_HARD)
1032 			return RESUME_HOST;
1033 		break;
1034 #endif
1035 	case H_RANDOM:
1036 		if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1037 			ret = H_HARDWARE;
1038 		break;
1039 
1040 	case H_SET_PARTITION_TABLE:
1041 		ret = H_FUNCTION;
1042 		if (nesting_enabled(vcpu->kvm))
1043 			ret = kvmhv_set_partition_table(vcpu);
1044 		break;
1045 	case H_ENTER_NESTED:
1046 		ret = H_FUNCTION;
1047 		if (!nesting_enabled(vcpu->kvm))
1048 			break;
1049 		ret = kvmhv_enter_nested_guest(vcpu);
1050 		if (ret == H_INTERRUPT) {
1051 			kvmppc_set_gpr(vcpu, 3, 0);
1052 			vcpu->arch.hcall_needed = 0;
1053 			return -EINTR;
1054 		} else if (ret == H_TOO_HARD) {
1055 			kvmppc_set_gpr(vcpu, 3, 0);
1056 			vcpu->arch.hcall_needed = 0;
1057 			return RESUME_HOST;
1058 		}
1059 		break;
1060 	case H_TLB_INVALIDATE:
1061 		ret = H_FUNCTION;
1062 		if (nesting_enabled(vcpu->kvm))
1063 			ret = kvmhv_do_nested_tlbie(vcpu);
1064 		break;
1065 	case H_COPY_TOFROM_GUEST:
1066 		ret = H_FUNCTION;
1067 		if (nesting_enabled(vcpu->kvm))
1068 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1069 		break;
1070 	case H_PAGE_INIT:
1071 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1072 					 kvmppc_get_gpr(vcpu, 5),
1073 					 kvmppc_get_gpr(vcpu, 6));
1074 		break;
1075 	case H_SVM_PAGE_IN:
1076 		ret = H_UNSUPPORTED;
1077 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1078 			ret = kvmppc_h_svm_page_in(vcpu->kvm,
1079 						   kvmppc_get_gpr(vcpu, 4),
1080 						   kvmppc_get_gpr(vcpu, 5),
1081 						   kvmppc_get_gpr(vcpu, 6));
1082 		break;
1083 	case H_SVM_PAGE_OUT:
1084 		ret = H_UNSUPPORTED;
1085 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1086 			ret = kvmppc_h_svm_page_out(vcpu->kvm,
1087 						    kvmppc_get_gpr(vcpu, 4),
1088 						    kvmppc_get_gpr(vcpu, 5),
1089 						    kvmppc_get_gpr(vcpu, 6));
1090 		break;
1091 	case H_SVM_INIT_START:
1092 		ret = H_UNSUPPORTED;
1093 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1094 			ret = kvmppc_h_svm_init_start(vcpu->kvm);
1095 		break;
1096 	case H_SVM_INIT_DONE:
1097 		ret = H_UNSUPPORTED;
1098 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1099 			ret = kvmppc_h_svm_init_done(vcpu->kvm);
1100 		break;
1101 	case H_SVM_INIT_ABORT:
1102 		ret = H_UNSUPPORTED;
1103 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1104 			ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1105 		break;
1106 
1107 	default:
1108 		return RESUME_HOST;
1109 	}
1110 	kvmppc_set_gpr(vcpu, 3, ret);
1111 	vcpu->arch.hcall_needed = 0;
1112 	return RESUME_GUEST;
1113 }
1114 
1115 /*
1116  * Handle H_CEDE in the nested virtualization case where we haven't
1117  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1118  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1119  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1120  */
1121 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1122 {
1123 	vcpu->arch.shregs.msr |= MSR_EE;
1124 	vcpu->arch.ceded = 1;
1125 	smp_mb();
1126 	if (vcpu->arch.prodded) {
1127 		vcpu->arch.prodded = 0;
1128 		smp_mb();
1129 		vcpu->arch.ceded = 0;
1130 	}
1131 }
1132 
1133 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1134 {
1135 	switch (cmd) {
1136 	case H_CEDE:
1137 	case H_PROD:
1138 	case H_CONFER:
1139 	case H_REGISTER_VPA:
1140 	case H_SET_MODE:
1141 	case H_LOGICAL_CI_LOAD:
1142 	case H_LOGICAL_CI_STORE:
1143 #ifdef CONFIG_KVM_XICS
1144 	case H_XIRR:
1145 	case H_CPPR:
1146 	case H_EOI:
1147 	case H_IPI:
1148 	case H_IPOLL:
1149 	case H_XIRR_X:
1150 #endif
1151 	case H_PAGE_INIT:
1152 		return 1;
1153 	}
1154 
1155 	/* See if it's in the real-mode table */
1156 	return kvmppc_hcall_impl_hv_realmode(cmd);
1157 }
1158 
1159 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1160 					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 		run->exit_reason = KVM_EXIT_DEBUG;
1175 		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_run *run, struct kvm_vcpu *vcpu,
1277 				 struct task_struct *tsk)
1278 {
1279 	int r = RESUME_HOST;
1280 
1281 	vcpu->stat.sum_exits++;
1282 
1283 	/*
1284 	 * This can happen if an interrupt occurs in the last stages
1285 	 * of guest entry or the first stages of guest exit (i.e. after
1286 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1287 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1288 	 * That can happen due to a bug, or due to a machine check
1289 	 * occurring at just the wrong time.
1290 	 */
1291 	if (vcpu->arch.shregs.msr & MSR_HV) {
1292 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1293 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1294 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1295 			vcpu->arch.shregs.msr);
1296 		kvmppc_dump_regs(vcpu);
1297 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1298 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1299 		return RESUME_HOST;
1300 	}
1301 	run->exit_reason = KVM_EXIT_UNKNOWN;
1302 	run->ready_for_interrupt_injection = 1;
1303 	switch (vcpu->arch.trap) {
1304 	/* We're good on these - the host merely wanted to get our attention */
1305 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1306 		vcpu->stat.dec_exits++;
1307 		r = RESUME_GUEST;
1308 		break;
1309 	case BOOK3S_INTERRUPT_EXTERNAL:
1310 	case BOOK3S_INTERRUPT_H_DOORBELL:
1311 	case BOOK3S_INTERRUPT_H_VIRT:
1312 		vcpu->stat.ext_intr_exits++;
1313 		r = RESUME_GUEST;
1314 		break;
1315 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1316 	case BOOK3S_INTERRUPT_HMI:
1317 	case BOOK3S_INTERRUPT_PERFMON:
1318 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1319 		r = RESUME_GUEST;
1320 		break;
1321 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1322 		/* Print the MCE event to host console. */
1323 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1324 
1325 		/*
1326 		 * If the guest can do FWNMI, exit to userspace so it can
1327 		 * deliver a FWNMI to the guest.
1328 		 * Otherwise we synthesize a machine check for the guest
1329 		 * so that it knows that the machine check occurred.
1330 		 */
1331 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1332 			ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1333 			kvmppc_core_queue_machine_check(vcpu, flags);
1334 			r = RESUME_GUEST;
1335 			break;
1336 		}
1337 
1338 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1339 		run->exit_reason = KVM_EXIT_NMI;
1340 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1341 		/* Clear out the old NMI status from run->flags */
1342 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1343 		/* Now set the NMI status */
1344 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1345 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1346 		else
1347 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1348 
1349 		r = RESUME_HOST;
1350 		break;
1351 	case BOOK3S_INTERRUPT_PROGRAM:
1352 	{
1353 		ulong flags;
1354 		/*
1355 		 * Normally program interrupts are delivered directly
1356 		 * to the guest by the hardware, but we can get here
1357 		 * as a result of a hypervisor emulation interrupt
1358 		 * (e40) getting turned into a 700 by BML RTAS.
1359 		 */
1360 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1361 		kvmppc_core_queue_program(vcpu, flags);
1362 		r = RESUME_GUEST;
1363 		break;
1364 	}
1365 	case BOOK3S_INTERRUPT_SYSCALL:
1366 	{
1367 		/* hcall - punt to userspace */
1368 		int i;
1369 
1370 		/* hypercall with MSR_PR has already been handled in rmode,
1371 		 * and never reaches here.
1372 		 */
1373 
1374 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1375 		for (i = 0; i < 9; ++i)
1376 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1377 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1378 		vcpu->arch.hcall_needed = 1;
1379 		r = RESUME_HOST;
1380 		break;
1381 	}
1382 	/*
1383 	 * We get these next two if the guest accesses a page which it thinks
1384 	 * it has mapped but which is not actually present, either because
1385 	 * it is for an emulated I/O device or because the corresonding
1386 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1387 	 * have been handled already.
1388 	 */
1389 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1390 		r = RESUME_PAGE_FAULT;
1391 		break;
1392 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1393 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1394 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1395 			DSISR_SRR1_MATCH_64S;
1396 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1397 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1398 		r = RESUME_PAGE_FAULT;
1399 		break;
1400 	/*
1401 	 * This occurs if the guest executes an illegal instruction.
1402 	 * If the guest debug is disabled, generate a program interrupt
1403 	 * to the guest. If guest debug is enabled, we need to check
1404 	 * whether the instruction is a software breakpoint instruction.
1405 	 * Accordingly return to Guest or Host.
1406 	 */
1407 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1408 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1409 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1410 				swab32(vcpu->arch.emul_inst) :
1411 				vcpu->arch.emul_inst;
1412 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1413 			r = kvmppc_emulate_debug_inst(run, vcpu);
1414 		} else {
1415 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1416 			r = RESUME_GUEST;
1417 		}
1418 		break;
1419 	/*
1420 	 * This occurs if the guest (kernel or userspace), does something that
1421 	 * is prohibited by HFSCR.
1422 	 * On POWER9, this could be a doorbell instruction that we need
1423 	 * to emulate.
1424 	 * Otherwise, we just generate a program interrupt to the guest.
1425 	 */
1426 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1427 		r = EMULATE_FAIL;
1428 		if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1429 		    cpu_has_feature(CPU_FTR_ARCH_300))
1430 			r = kvmppc_emulate_doorbell_instr(vcpu);
1431 		if (r == EMULATE_FAIL) {
1432 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1433 			r = RESUME_GUEST;
1434 		}
1435 		break;
1436 
1437 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1438 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1439 		/*
1440 		 * This occurs for various TM-related instructions that
1441 		 * we need to emulate on POWER9 DD2.2.  We have already
1442 		 * handled the cases where the guest was in real-suspend
1443 		 * mode and was transitioning to transactional state.
1444 		 */
1445 		r = kvmhv_p9_tm_emulation(vcpu);
1446 		break;
1447 #endif
1448 
1449 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1450 		r = RESUME_PASSTHROUGH;
1451 		break;
1452 	default:
1453 		kvmppc_dump_regs(vcpu);
1454 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1455 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1456 			vcpu->arch.shregs.msr);
1457 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1458 		r = RESUME_HOST;
1459 		break;
1460 	}
1461 
1462 	return r;
1463 }
1464 
1465 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1466 {
1467 	int r;
1468 	int srcu_idx;
1469 
1470 	vcpu->stat.sum_exits++;
1471 
1472 	/*
1473 	 * This can happen if an interrupt occurs in the last stages
1474 	 * of guest entry or the first stages of guest exit (i.e. after
1475 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1476 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1477 	 * That can happen due to a bug, or due to a machine check
1478 	 * occurring at just the wrong time.
1479 	 */
1480 	if (vcpu->arch.shregs.msr & MSR_HV) {
1481 		pr_emerg("KVM trap in HV mode while nested!\n");
1482 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1483 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1484 			 vcpu->arch.shregs.msr);
1485 		kvmppc_dump_regs(vcpu);
1486 		return RESUME_HOST;
1487 	}
1488 	switch (vcpu->arch.trap) {
1489 	/* We're good on these - the host merely wanted to get our attention */
1490 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1491 		vcpu->stat.dec_exits++;
1492 		r = RESUME_GUEST;
1493 		break;
1494 	case BOOK3S_INTERRUPT_EXTERNAL:
1495 		vcpu->stat.ext_intr_exits++;
1496 		r = RESUME_HOST;
1497 		break;
1498 	case BOOK3S_INTERRUPT_H_DOORBELL:
1499 	case BOOK3S_INTERRUPT_H_VIRT:
1500 		vcpu->stat.ext_intr_exits++;
1501 		r = RESUME_GUEST;
1502 		break;
1503 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1504 	case BOOK3S_INTERRUPT_HMI:
1505 	case BOOK3S_INTERRUPT_PERFMON:
1506 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1507 		r = RESUME_GUEST;
1508 		break;
1509 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1510 		/* Pass the machine check to the L1 guest */
1511 		r = RESUME_HOST;
1512 		/* Print the MCE event to host console. */
1513 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1514 		break;
1515 	/*
1516 	 * We get these next two if the guest accesses a page which it thinks
1517 	 * it has mapped but which is not actually present, either because
1518 	 * it is for an emulated I/O device or because the corresonding
1519 	 * host page has been paged out.
1520 	 */
1521 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1522 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1523 		r = kvmhv_nested_page_fault(run, vcpu);
1524 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1525 		break;
1526 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1527 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1528 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1529 					 DSISR_SRR1_MATCH_64S;
1530 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1531 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1532 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1533 		r = kvmhv_nested_page_fault(run, vcpu);
1534 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1535 		break;
1536 
1537 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1538 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1539 		/*
1540 		 * This occurs for various TM-related instructions that
1541 		 * we need to emulate on POWER9 DD2.2.  We have already
1542 		 * handled the cases where the guest was in real-suspend
1543 		 * mode and was transitioning to transactional state.
1544 		 */
1545 		r = kvmhv_p9_tm_emulation(vcpu);
1546 		break;
1547 #endif
1548 
1549 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1550 		vcpu->arch.trap = 0;
1551 		r = RESUME_GUEST;
1552 		if (!xics_on_xive())
1553 			kvmppc_xics_rm_complete(vcpu, 0);
1554 		break;
1555 	default:
1556 		r = RESUME_HOST;
1557 		break;
1558 	}
1559 
1560 	return r;
1561 }
1562 
1563 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1564 					    struct kvm_sregs *sregs)
1565 {
1566 	int i;
1567 
1568 	memset(sregs, 0, sizeof(struct kvm_sregs));
1569 	sregs->pvr = vcpu->arch.pvr;
1570 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1571 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1572 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1573 	}
1574 
1575 	return 0;
1576 }
1577 
1578 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1579 					    struct kvm_sregs *sregs)
1580 {
1581 	int i, j;
1582 
1583 	/* Only accept the same PVR as the host's, since we can't spoof it */
1584 	if (sregs->pvr != vcpu->arch.pvr)
1585 		return -EINVAL;
1586 
1587 	j = 0;
1588 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1589 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1590 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1591 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1592 			++j;
1593 		}
1594 	}
1595 	vcpu->arch.slb_max = j;
1596 
1597 	return 0;
1598 }
1599 
1600 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1601 		bool preserve_top32)
1602 {
1603 	struct kvm *kvm = vcpu->kvm;
1604 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1605 	u64 mask;
1606 
1607 	spin_lock(&vc->lock);
1608 	/*
1609 	 * If ILE (interrupt little-endian) has changed, update the
1610 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1611 	 */
1612 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1613 		struct kvm_vcpu *vcpu;
1614 		int i;
1615 
1616 		kvm_for_each_vcpu(i, vcpu, kvm) {
1617 			if (vcpu->arch.vcore != vc)
1618 				continue;
1619 			if (new_lpcr & LPCR_ILE)
1620 				vcpu->arch.intr_msr |= MSR_LE;
1621 			else
1622 				vcpu->arch.intr_msr &= ~MSR_LE;
1623 		}
1624 	}
1625 
1626 	/*
1627 	 * Userspace can only modify DPFD (default prefetch depth),
1628 	 * ILE (interrupt little-endian) and TC (translation control).
1629 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1630 	 */
1631 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1632 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1633 		mask |= LPCR_AIL;
1634 	/*
1635 	 * On POWER9, allow userspace to enable large decrementer for the
1636 	 * guest, whether or not the host has it enabled.
1637 	 */
1638 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1639 		mask |= LPCR_LD;
1640 
1641 	/* Broken 32-bit version of LPCR must not clear top bits */
1642 	if (preserve_top32)
1643 		mask &= 0xFFFFFFFF;
1644 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1645 	spin_unlock(&vc->lock);
1646 }
1647 
1648 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1649 				 union kvmppc_one_reg *val)
1650 {
1651 	int r = 0;
1652 	long int i;
1653 
1654 	switch (id) {
1655 	case KVM_REG_PPC_DEBUG_INST:
1656 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1657 		break;
1658 	case KVM_REG_PPC_HIOR:
1659 		*val = get_reg_val(id, 0);
1660 		break;
1661 	case KVM_REG_PPC_DABR:
1662 		*val = get_reg_val(id, vcpu->arch.dabr);
1663 		break;
1664 	case KVM_REG_PPC_DABRX:
1665 		*val = get_reg_val(id, vcpu->arch.dabrx);
1666 		break;
1667 	case KVM_REG_PPC_DSCR:
1668 		*val = get_reg_val(id, vcpu->arch.dscr);
1669 		break;
1670 	case KVM_REG_PPC_PURR:
1671 		*val = get_reg_val(id, vcpu->arch.purr);
1672 		break;
1673 	case KVM_REG_PPC_SPURR:
1674 		*val = get_reg_val(id, vcpu->arch.spurr);
1675 		break;
1676 	case KVM_REG_PPC_AMR:
1677 		*val = get_reg_val(id, vcpu->arch.amr);
1678 		break;
1679 	case KVM_REG_PPC_UAMOR:
1680 		*val = get_reg_val(id, vcpu->arch.uamor);
1681 		break;
1682 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1683 		i = id - KVM_REG_PPC_MMCR0;
1684 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1685 		break;
1686 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1687 		i = id - KVM_REG_PPC_PMC1;
1688 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1689 		break;
1690 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1691 		i = id - KVM_REG_PPC_SPMC1;
1692 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1693 		break;
1694 	case KVM_REG_PPC_SIAR:
1695 		*val = get_reg_val(id, vcpu->arch.siar);
1696 		break;
1697 	case KVM_REG_PPC_SDAR:
1698 		*val = get_reg_val(id, vcpu->arch.sdar);
1699 		break;
1700 	case KVM_REG_PPC_SIER:
1701 		*val = get_reg_val(id, vcpu->arch.sier);
1702 		break;
1703 	case KVM_REG_PPC_IAMR:
1704 		*val = get_reg_val(id, vcpu->arch.iamr);
1705 		break;
1706 	case KVM_REG_PPC_PSPB:
1707 		*val = get_reg_val(id, vcpu->arch.pspb);
1708 		break;
1709 	case KVM_REG_PPC_DPDES:
1710 		/*
1711 		 * On POWER9, where we are emulating msgsndp etc.,
1712 		 * we return 1 bit for each vcpu, which can come from
1713 		 * either vcore->dpdes or doorbell_request.
1714 		 * On POWER8, doorbell_request is 0.
1715 		 */
1716 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1717 				   vcpu->arch.doorbell_request);
1718 		break;
1719 	case KVM_REG_PPC_VTB:
1720 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1721 		break;
1722 	case KVM_REG_PPC_DAWR:
1723 		*val = get_reg_val(id, vcpu->arch.dawr);
1724 		break;
1725 	case KVM_REG_PPC_DAWRX:
1726 		*val = get_reg_val(id, vcpu->arch.dawrx);
1727 		break;
1728 	case KVM_REG_PPC_CIABR:
1729 		*val = get_reg_val(id, vcpu->arch.ciabr);
1730 		break;
1731 	case KVM_REG_PPC_CSIGR:
1732 		*val = get_reg_val(id, vcpu->arch.csigr);
1733 		break;
1734 	case KVM_REG_PPC_TACR:
1735 		*val = get_reg_val(id, vcpu->arch.tacr);
1736 		break;
1737 	case KVM_REG_PPC_TCSCR:
1738 		*val = get_reg_val(id, vcpu->arch.tcscr);
1739 		break;
1740 	case KVM_REG_PPC_PID:
1741 		*val = get_reg_val(id, vcpu->arch.pid);
1742 		break;
1743 	case KVM_REG_PPC_ACOP:
1744 		*val = get_reg_val(id, vcpu->arch.acop);
1745 		break;
1746 	case KVM_REG_PPC_WORT:
1747 		*val = get_reg_val(id, vcpu->arch.wort);
1748 		break;
1749 	case KVM_REG_PPC_TIDR:
1750 		*val = get_reg_val(id, vcpu->arch.tid);
1751 		break;
1752 	case KVM_REG_PPC_PSSCR:
1753 		*val = get_reg_val(id, vcpu->arch.psscr);
1754 		break;
1755 	case KVM_REG_PPC_VPA_ADDR:
1756 		spin_lock(&vcpu->arch.vpa_update_lock);
1757 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1758 		spin_unlock(&vcpu->arch.vpa_update_lock);
1759 		break;
1760 	case KVM_REG_PPC_VPA_SLB:
1761 		spin_lock(&vcpu->arch.vpa_update_lock);
1762 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1763 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1764 		spin_unlock(&vcpu->arch.vpa_update_lock);
1765 		break;
1766 	case KVM_REG_PPC_VPA_DTL:
1767 		spin_lock(&vcpu->arch.vpa_update_lock);
1768 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1769 		val->vpaval.length = vcpu->arch.dtl.len;
1770 		spin_unlock(&vcpu->arch.vpa_update_lock);
1771 		break;
1772 	case KVM_REG_PPC_TB_OFFSET:
1773 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1774 		break;
1775 	case KVM_REG_PPC_LPCR:
1776 	case KVM_REG_PPC_LPCR_64:
1777 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1778 		break;
1779 	case KVM_REG_PPC_PPR:
1780 		*val = get_reg_val(id, vcpu->arch.ppr);
1781 		break;
1782 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1783 	case KVM_REG_PPC_TFHAR:
1784 		*val = get_reg_val(id, vcpu->arch.tfhar);
1785 		break;
1786 	case KVM_REG_PPC_TFIAR:
1787 		*val = get_reg_val(id, vcpu->arch.tfiar);
1788 		break;
1789 	case KVM_REG_PPC_TEXASR:
1790 		*val = get_reg_val(id, vcpu->arch.texasr);
1791 		break;
1792 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1793 		i = id - KVM_REG_PPC_TM_GPR0;
1794 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1795 		break;
1796 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1797 	{
1798 		int j;
1799 		i = id - KVM_REG_PPC_TM_VSR0;
1800 		if (i < 32)
1801 			for (j = 0; j < TS_FPRWIDTH; j++)
1802 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1803 		else {
1804 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1805 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1806 			else
1807 				r = -ENXIO;
1808 		}
1809 		break;
1810 	}
1811 	case KVM_REG_PPC_TM_CR:
1812 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1813 		break;
1814 	case KVM_REG_PPC_TM_XER:
1815 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1816 		break;
1817 	case KVM_REG_PPC_TM_LR:
1818 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1819 		break;
1820 	case KVM_REG_PPC_TM_CTR:
1821 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1822 		break;
1823 	case KVM_REG_PPC_TM_FPSCR:
1824 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1825 		break;
1826 	case KVM_REG_PPC_TM_AMR:
1827 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1828 		break;
1829 	case KVM_REG_PPC_TM_PPR:
1830 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1831 		break;
1832 	case KVM_REG_PPC_TM_VRSAVE:
1833 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1834 		break;
1835 	case KVM_REG_PPC_TM_VSCR:
1836 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1837 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1838 		else
1839 			r = -ENXIO;
1840 		break;
1841 	case KVM_REG_PPC_TM_DSCR:
1842 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1843 		break;
1844 	case KVM_REG_PPC_TM_TAR:
1845 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1846 		break;
1847 #endif
1848 	case KVM_REG_PPC_ARCH_COMPAT:
1849 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1850 		break;
1851 	case KVM_REG_PPC_DEC_EXPIRY:
1852 		*val = get_reg_val(id, vcpu->arch.dec_expires +
1853 				   vcpu->arch.vcore->tb_offset);
1854 		break;
1855 	case KVM_REG_PPC_ONLINE:
1856 		*val = get_reg_val(id, vcpu->arch.online);
1857 		break;
1858 	case KVM_REG_PPC_PTCR:
1859 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1860 		break;
1861 	default:
1862 		r = -EINVAL;
1863 		break;
1864 	}
1865 
1866 	return r;
1867 }
1868 
1869 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1870 				 union kvmppc_one_reg *val)
1871 {
1872 	int r = 0;
1873 	long int i;
1874 	unsigned long addr, len;
1875 
1876 	switch (id) {
1877 	case KVM_REG_PPC_HIOR:
1878 		/* Only allow this to be set to zero */
1879 		if (set_reg_val(id, *val))
1880 			r = -EINVAL;
1881 		break;
1882 	case KVM_REG_PPC_DABR:
1883 		vcpu->arch.dabr = set_reg_val(id, *val);
1884 		break;
1885 	case KVM_REG_PPC_DABRX:
1886 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1887 		break;
1888 	case KVM_REG_PPC_DSCR:
1889 		vcpu->arch.dscr = set_reg_val(id, *val);
1890 		break;
1891 	case KVM_REG_PPC_PURR:
1892 		vcpu->arch.purr = set_reg_val(id, *val);
1893 		break;
1894 	case KVM_REG_PPC_SPURR:
1895 		vcpu->arch.spurr = set_reg_val(id, *val);
1896 		break;
1897 	case KVM_REG_PPC_AMR:
1898 		vcpu->arch.amr = set_reg_val(id, *val);
1899 		break;
1900 	case KVM_REG_PPC_UAMOR:
1901 		vcpu->arch.uamor = set_reg_val(id, *val);
1902 		break;
1903 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1904 		i = id - KVM_REG_PPC_MMCR0;
1905 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1906 		break;
1907 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1908 		i = id - KVM_REG_PPC_PMC1;
1909 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1910 		break;
1911 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1912 		i = id - KVM_REG_PPC_SPMC1;
1913 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1914 		break;
1915 	case KVM_REG_PPC_SIAR:
1916 		vcpu->arch.siar = set_reg_val(id, *val);
1917 		break;
1918 	case KVM_REG_PPC_SDAR:
1919 		vcpu->arch.sdar = set_reg_val(id, *val);
1920 		break;
1921 	case KVM_REG_PPC_SIER:
1922 		vcpu->arch.sier = set_reg_val(id, *val);
1923 		break;
1924 	case KVM_REG_PPC_IAMR:
1925 		vcpu->arch.iamr = set_reg_val(id, *val);
1926 		break;
1927 	case KVM_REG_PPC_PSPB:
1928 		vcpu->arch.pspb = set_reg_val(id, *val);
1929 		break;
1930 	case KVM_REG_PPC_DPDES:
1931 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1932 		break;
1933 	case KVM_REG_PPC_VTB:
1934 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1935 		break;
1936 	case KVM_REG_PPC_DAWR:
1937 		vcpu->arch.dawr = set_reg_val(id, *val);
1938 		break;
1939 	case KVM_REG_PPC_DAWRX:
1940 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1941 		break;
1942 	case KVM_REG_PPC_CIABR:
1943 		vcpu->arch.ciabr = set_reg_val(id, *val);
1944 		/* Don't allow setting breakpoints in hypervisor code */
1945 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1946 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1947 		break;
1948 	case KVM_REG_PPC_CSIGR:
1949 		vcpu->arch.csigr = set_reg_val(id, *val);
1950 		break;
1951 	case KVM_REG_PPC_TACR:
1952 		vcpu->arch.tacr = set_reg_val(id, *val);
1953 		break;
1954 	case KVM_REG_PPC_TCSCR:
1955 		vcpu->arch.tcscr = set_reg_val(id, *val);
1956 		break;
1957 	case KVM_REG_PPC_PID:
1958 		vcpu->arch.pid = set_reg_val(id, *val);
1959 		break;
1960 	case KVM_REG_PPC_ACOP:
1961 		vcpu->arch.acop = set_reg_val(id, *val);
1962 		break;
1963 	case KVM_REG_PPC_WORT:
1964 		vcpu->arch.wort = set_reg_val(id, *val);
1965 		break;
1966 	case KVM_REG_PPC_TIDR:
1967 		vcpu->arch.tid = set_reg_val(id, *val);
1968 		break;
1969 	case KVM_REG_PPC_PSSCR:
1970 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1971 		break;
1972 	case KVM_REG_PPC_VPA_ADDR:
1973 		addr = set_reg_val(id, *val);
1974 		r = -EINVAL;
1975 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1976 			      vcpu->arch.dtl.next_gpa))
1977 			break;
1978 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1979 		break;
1980 	case KVM_REG_PPC_VPA_SLB:
1981 		addr = val->vpaval.addr;
1982 		len = val->vpaval.length;
1983 		r = -EINVAL;
1984 		if (addr && !vcpu->arch.vpa.next_gpa)
1985 			break;
1986 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1987 		break;
1988 	case KVM_REG_PPC_VPA_DTL:
1989 		addr = val->vpaval.addr;
1990 		len = val->vpaval.length;
1991 		r = -EINVAL;
1992 		if (addr && (len < sizeof(struct dtl_entry) ||
1993 			     !vcpu->arch.vpa.next_gpa))
1994 			break;
1995 		len -= len % sizeof(struct dtl_entry);
1996 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1997 		break;
1998 	case KVM_REG_PPC_TB_OFFSET:
1999 		/* round up to multiple of 2^24 */
2000 		vcpu->arch.vcore->tb_offset =
2001 			ALIGN(set_reg_val(id, *val), 1UL << 24);
2002 		break;
2003 	case KVM_REG_PPC_LPCR:
2004 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2005 		break;
2006 	case KVM_REG_PPC_LPCR_64:
2007 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2008 		break;
2009 	case KVM_REG_PPC_PPR:
2010 		vcpu->arch.ppr = set_reg_val(id, *val);
2011 		break;
2012 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2013 	case KVM_REG_PPC_TFHAR:
2014 		vcpu->arch.tfhar = set_reg_val(id, *val);
2015 		break;
2016 	case KVM_REG_PPC_TFIAR:
2017 		vcpu->arch.tfiar = set_reg_val(id, *val);
2018 		break;
2019 	case KVM_REG_PPC_TEXASR:
2020 		vcpu->arch.texasr = set_reg_val(id, *val);
2021 		break;
2022 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2023 		i = id - KVM_REG_PPC_TM_GPR0;
2024 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2025 		break;
2026 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2027 	{
2028 		int j;
2029 		i = id - KVM_REG_PPC_TM_VSR0;
2030 		if (i < 32)
2031 			for (j = 0; j < TS_FPRWIDTH; j++)
2032 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2033 		else
2034 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2035 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2036 			else
2037 				r = -ENXIO;
2038 		break;
2039 	}
2040 	case KVM_REG_PPC_TM_CR:
2041 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2042 		break;
2043 	case KVM_REG_PPC_TM_XER:
2044 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2045 		break;
2046 	case KVM_REG_PPC_TM_LR:
2047 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2048 		break;
2049 	case KVM_REG_PPC_TM_CTR:
2050 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2051 		break;
2052 	case KVM_REG_PPC_TM_FPSCR:
2053 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2054 		break;
2055 	case KVM_REG_PPC_TM_AMR:
2056 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2057 		break;
2058 	case KVM_REG_PPC_TM_PPR:
2059 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2060 		break;
2061 	case KVM_REG_PPC_TM_VRSAVE:
2062 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2063 		break;
2064 	case KVM_REG_PPC_TM_VSCR:
2065 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2066 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2067 		else
2068 			r = - ENXIO;
2069 		break;
2070 	case KVM_REG_PPC_TM_DSCR:
2071 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2072 		break;
2073 	case KVM_REG_PPC_TM_TAR:
2074 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2075 		break;
2076 #endif
2077 	case KVM_REG_PPC_ARCH_COMPAT:
2078 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2079 		break;
2080 	case KVM_REG_PPC_DEC_EXPIRY:
2081 		vcpu->arch.dec_expires = set_reg_val(id, *val) -
2082 			vcpu->arch.vcore->tb_offset;
2083 		break;
2084 	case KVM_REG_PPC_ONLINE:
2085 		i = set_reg_val(id, *val);
2086 		if (i && !vcpu->arch.online)
2087 			atomic_inc(&vcpu->arch.vcore->online_count);
2088 		else if (!i && vcpu->arch.online)
2089 			atomic_dec(&vcpu->arch.vcore->online_count);
2090 		vcpu->arch.online = i;
2091 		break;
2092 	case KVM_REG_PPC_PTCR:
2093 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2094 		break;
2095 	default:
2096 		r = -EINVAL;
2097 		break;
2098 	}
2099 
2100 	return r;
2101 }
2102 
2103 /*
2104  * On POWER9, threads are independent and can be in different partitions.
2105  * Therefore we consider each thread to be a subcore.
2106  * There is a restriction that all threads have to be in the same
2107  * MMU mode (radix or HPT), unfortunately, but since we only support
2108  * HPT guests on a HPT host so far, that isn't an impediment yet.
2109  */
2110 static int threads_per_vcore(struct kvm *kvm)
2111 {
2112 	if (kvm->arch.threads_indep)
2113 		return 1;
2114 	return threads_per_subcore;
2115 }
2116 
2117 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2118 {
2119 	struct kvmppc_vcore *vcore;
2120 
2121 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2122 
2123 	if (vcore == NULL)
2124 		return NULL;
2125 
2126 	spin_lock_init(&vcore->lock);
2127 	spin_lock_init(&vcore->stoltb_lock);
2128 	init_swait_queue_head(&vcore->wq);
2129 	vcore->preempt_tb = TB_NIL;
2130 	vcore->lpcr = kvm->arch.lpcr;
2131 	vcore->first_vcpuid = id;
2132 	vcore->kvm = kvm;
2133 	INIT_LIST_HEAD(&vcore->preempt_list);
2134 
2135 	return vcore;
2136 }
2137 
2138 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2139 static struct debugfs_timings_element {
2140 	const char *name;
2141 	size_t offset;
2142 } timings[] = {
2143 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2144 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2145 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2146 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2147 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2148 };
2149 
2150 #define N_TIMINGS	(ARRAY_SIZE(timings))
2151 
2152 struct debugfs_timings_state {
2153 	struct kvm_vcpu	*vcpu;
2154 	unsigned int	buflen;
2155 	char		buf[N_TIMINGS * 100];
2156 };
2157 
2158 static int debugfs_timings_open(struct inode *inode, struct file *file)
2159 {
2160 	struct kvm_vcpu *vcpu = inode->i_private;
2161 	struct debugfs_timings_state *p;
2162 
2163 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2164 	if (!p)
2165 		return -ENOMEM;
2166 
2167 	kvm_get_kvm(vcpu->kvm);
2168 	p->vcpu = vcpu;
2169 	file->private_data = p;
2170 
2171 	return nonseekable_open(inode, file);
2172 }
2173 
2174 static int debugfs_timings_release(struct inode *inode, struct file *file)
2175 {
2176 	struct debugfs_timings_state *p = file->private_data;
2177 
2178 	kvm_put_kvm(p->vcpu->kvm);
2179 	kfree(p);
2180 	return 0;
2181 }
2182 
2183 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2184 				    size_t len, loff_t *ppos)
2185 {
2186 	struct debugfs_timings_state *p = file->private_data;
2187 	struct kvm_vcpu *vcpu = p->vcpu;
2188 	char *s, *buf_end;
2189 	struct kvmhv_tb_accumulator tb;
2190 	u64 count;
2191 	loff_t pos;
2192 	ssize_t n;
2193 	int i, loops;
2194 	bool ok;
2195 
2196 	if (!p->buflen) {
2197 		s = p->buf;
2198 		buf_end = s + sizeof(p->buf);
2199 		for (i = 0; i < N_TIMINGS; ++i) {
2200 			struct kvmhv_tb_accumulator *acc;
2201 
2202 			acc = (struct kvmhv_tb_accumulator *)
2203 				((unsigned long)vcpu + timings[i].offset);
2204 			ok = false;
2205 			for (loops = 0; loops < 1000; ++loops) {
2206 				count = acc->seqcount;
2207 				if (!(count & 1)) {
2208 					smp_rmb();
2209 					tb = *acc;
2210 					smp_rmb();
2211 					if (count == acc->seqcount) {
2212 						ok = true;
2213 						break;
2214 					}
2215 				}
2216 				udelay(1);
2217 			}
2218 			if (!ok)
2219 				snprintf(s, buf_end - s, "%s: stuck\n",
2220 					timings[i].name);
2221 			else
2222 				snprintf(s, buf_end - s,
2223 					"%s: %llu %llu %llu %llu\n",
2224 					timings[i].name, count / 2,
2225 					tb_to_ns(tb.tb_total),
2226 					tb_to_ns(tb.tb_min),
2227 					tb_to_ns(tb.tb_max));
2228 			s += strlen(s);
2229 		}
2230 		p->buflen = s - p->buf;
2231 	}
2232 
2233 	pos = *ppos;
2234 	if (pos >= p->buflen)
2235 		return 0;
2236 	if (len > p->buflen - pos)
2237 		len = p->buflen - pos;
2238 	n = copy_to_user(buf, p->buf + pos, len);
2239 	if (n) {
2240 		if (n == len)
2241 			return -EFAULT;
2242 		len -= n;
2243 	}
2244 	*ppos = pos + len;
2245 	return len;
2246 }
2247 
2248 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2249 				     size_t len, loff_t *ppos)
2250 {
2251 	return -EACCES;
2252 }
2253 
2254 static const struct file_operations debugfs_timings_ops = {
2255 	.owner	 = THIS_MODULE,
2256 	.open	 = debugfs_timings_open,
2257 	.release = debugfs_timings_release,
2258 	.read	 = debugfs_timings_read,
2259 	.write	 = debugfs_timings_write,
2260 	.llseek	 = generic_file_llseek,
2261 };
2262 
2263 /* Create a debugfs directory for the vcpu */
2264 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2265 {
2266 	char buf[16];
2267 	struct kvm *kvm = vcpu->kvm;
2268 
2269 	snprintf(buf, sizeof(buf), "vcpu%u", id);
2270 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2271 	debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2272 			    &debugfs_timings_ops);
2273 }
2274 
2275 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2276 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2277 {
2278 }
2279 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2280 
2281 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2282 {
2283 	int err;
2284 	int core;
2285 	struct kvmppc_vcore *vcore;
2286 	struct kvm *kvm;
2287 	unsigned int id;
2288 
2289 	kvm = vcpu->kvm;
2290 	id = vcpu->vcpu_id;
2291 
2292 	vcpu->arch.shared = &vcpu->arch.shregs;
2293 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2294 	/*
2295 	 * The shared struct is never shared on HV,
2296 	 * so we can always use host endianness
2297 	 */
2298 #ifdef __BIG_ENDIAN__
2299 	vcpu->arch.shared_big_endian = true;
2300 #else
2301 	vcpu->arch.shared_big_endian = false;
2302 #endif
2303 #endif
2304 	vcpu->arch.mmcr[0] = MMCR0_FC;
2305 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2306 	/* default to host PVR, since we can't spoof it */
2307 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2308 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2309 	spin_lock_init(&vcpu->arch.tbacct_lock);
2310 	vcpu->arch.busy_preempt = TB_NIL;
2311 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2312 
2313 	/*
2314 	 * Set the default HFSCR for the guest from the host value.
2315 	 * This value is only used on POWER9.
2316 	 * On POWER9, we want to virtualize the doorbell facility, so we
2317 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2318 	 * to trap and then we emulate them.
2319 	 */
2320 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2321 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2322 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2323 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2324 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2325 			vcpu->arch.hfscr |= HFSCR_TM;
2326 	}
2327 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2328 		vcpu->arch.hfscr |= HFSCR_TM;
2329 
2330 	kvmppc_mmu_book3s_hv_init(vcpu);
2331 
2332 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2333 
2334 	init_waitqueue_head(&vcpu->arch.cpu_run);
2335 
2336 	mutex_lock(&kvm->lock);
2337 	vcore = NULL;
2338 	err = -EINVAL;
2339 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2340 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2341 			pr_devel("KVM: VCPU ID too high\n");
2342 			core = KVM_MAX_VCORES;
2343 		} else {
2344 			BUG_ON(kvm->arch.smt_mode != 1);
2345 			core = kvmppc_pack_vcpu_id(kvm, id);
2346 		}
2347 	} else {
2348 		core = id / kvm->arch.smt_mode;
2349 	}
2350 	if (core < KVM_MAX_VCORES) {
2351 		vcore = kvm->arch.vcores[core];
2352 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2353 			pr_devel("KVM: collision on id %u", id);
2354 			vcore = NULL;
2355 		} else if (!vcore) {
2356 			/*
2357 			 * Take mmu_setup_lock for mutual exclusion
2358 			 * with kvmppc_update_lpcr().
2359 			 */
2360 			err = -ENOMEM;
2361 			vcore = kvmppc_vcore_create(kvm,
2362 					id & ~(kvm->arch.smt_mode - 1));
2363 			mutex_lock(&kvm->arch.mmu_setup_lock);
2364 			kvm->arch.vcores[core] = vcore;
2365 			kvm->arch.online_vcores++;
2366 			mutex_unlock(&kvm->arch.mmu_setup_lock);
2367 		}
2368 	}
2369 	mutex_unlock(&kvm->lock);
2370 
2371 	if (!vcore)
2372 		return err;
2373 
2374 	spin_lock(&vcore->lock);
2375 	++vcore->num_threads;
2376 	spin_unlock(&vcore->lock);
2377 	vcpu->arch.vcore = vcore;
2378 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2379 	vcpu->arch.thread_cpu = -1;
2380 	vcpu->arch.prev_cpu = -1;
2381 
2382 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2383 	kvmppc_sanity_check(vcpu);
2384 
2385 	debugfs_vcpu_init(vcpu, id);
2386 
2387 	return 0;
2388 }
2389 
2390 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2391 			      unsigned long flags)
2392 {
2393 	int err;
2394 	int esmt = 0;
2395 
2396 	if (flags)
2397 		return -EINVAL;
2398 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2399 		return -EINVAL;
2400 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2401 		/*
2402 		 * On POWER8 (or POWER7), the threading mode is "strict",
2403 		 * so we pack smt_mode vcpus per vcore.
2404 		 */
2405 		if (smt_mode > threads_per_subcore)
2406 			return -EINVAL;
2407 	} else {
2408 		/*
2409 		 * On POWER9, the threading mode is "loose",
2410 		 * so each vcpu gets its own vcore.
2411 		 */
2412 		esmt = smt_mode;
2413 		smt_mode = 1;
2414 	}
2415 	mutex_lock(&kvm->lock);
2416 	err = -EBUSY;
2417 	if (!kvm->arch.online_vcores) {
2418 		kvm->arch.smt_mode = smt_mode;
2419 		kvm->arch.emul_smt_mode = esmt;
2420 		err = 0;
2421 	}
2422 	mutex_unlock(&kvm->lock);
2423 
2424 	return err;
2425 }
2426 
2427 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2428 {
2429 	if (vpa->pinned_addr)
2430 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2431 					vpa->dirty);
2432 }
2433 
2434 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2435 {
2436 	spin_lock(&vcpu->arch.vpa_update_lock);
2437 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2438 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2439 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2440 	spin_unlock(&vcpu->arch.vpa_update_lock);
2441 }
2442 
2443 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2444 {
2445 	/* Indicate we want to get back into the guest */
2446 	return 1;
2447 }
2448 
2449 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2450 {
2451 	unsigned long dec_nsec, now;
2452 
2453 	now = get_tb();
2454 	if (now > vcpu->arch.dec_expires) {
2455 		/* decrementer has already gone negative */
2456 		kvmppc_core_queue_dec(vcpu);
2457 		kvmppc_core_prepare_to_enter(vcpu);
2458 		return;
2459 	}
2460 	dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2461 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2462 	vcpu->arch.timer_running = 1;
2463 }
2464 
2465 extern int __kvmppc_vcore_entry(void);
2466 
2467 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2468 				   struct kvm_vcpu *vcpu)
2469 {
2470 	u64 now;
2471 
2472 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2473 		return;
2474 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2475 	now = mftb();
2476 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2477 		vcpu->arch.stolen_logged;
2478 	vcpu->arch.busy_preempt = now;
2479 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2480 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2481 	--vc->n_runnable;
2482 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2483 }
2484 
2485 static int kvmppc_grab_hwthread(int cpu)
2486 {
2487 	struct paca_struct *tpaca;
2488 	long timeout = 10000;
2489 
2490 	tpaca = paca_ptrs[cpu];
2491 
2492 	/* Ensure the thread won't go into the kernel if it wakes */
2493 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2494 	tpaca->kvm_hstate.kvm_vcore = NULL;
2495 	tpaca->kvm_hstate.napping = 0;
2496 	smp_wmb();
2497 	tpaca->kvm_hstate.hwthread_req = 1;
2498 
2499 	/*
2500 	 * If the thread is already executing in the kernel (e.g. handling
2501 	 * a stray interrupt), wait for it to get back to nap mode.
2502 	 * The smp_mb() is to ensure that our setting of hwthread_req
2503 	 * is visible before we look at hwthread_state, so if this
2504 	 * races with the code at system_reset_pSeries and the thread
2505 	 * misses our setting of hwthread_req, we are sure to see its
2506 	 * setting of hwthread_state, and vice versa.
2507 	 */
2508 	smp_mb();
2509 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2510 		if (--timeout <= 0) {
2511 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2512 			return -EBUSY;
2513 		}
2514 		udelay(1);
2515 	}
2516 	return 0;
2517 }
2518 
2519 static void kvmppc_release_hwthread(int cpu)
2520 {
2521 	struct paca_struct *tpaca;
2522 
2523 	tpaca = paca_ptrs[cpu];
2524 	tpaca->kvm_hstate.hwthread_req = 0;
2525 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2526 	tpaca->kvm_hstate.kvm_vcore = NULL;
2527 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2528 }
2529 
2530 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2531 {
2532 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2533 	cpumask_t *cpu_in_guest;
2534 	int i;
2535 
2536 	cpu = cpu_first_thread_sibling(cpu);
2537 	if (nested) {
2538 		cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2539 		cpu_in_guest = &nested->cpu_in_guest;
2540 	} else {
2541 		cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2542 		cpu_in_guest = &kvm->arch.cpu_in_guest;
2543 	}
2544 	/*
2545 	 * Make sure setting of bit in need_tlb_flush precedes
2546 	 * testing of cpu_in_guest bits.  The matching barrier on
2547 	 * the other side is the first smp_mb() in kvmppc_run_core().
2548 	 */
2549 	smp_mb();
2550 	for (i = 0; i < threads_per_core; ++i)
2551 		if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2552 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2553 }
2554 
2555 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2556 {
2557 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2558 	struct kvm *kvm = vcpu->kvm;
2559 	int prev_cpu;
2560 
2561 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2562 		return;
2563 
2564 	if (nested)
2565 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2566 	else
2567 		prev_cpu = vcpu->arch.prev_cpu;
2568 
2569 	/*
2570 	 * With radix, the guest can do TLB invalidations itself,
2571 	 * and it could choose to use the local form (tlbiel) if
2572 	 * it is invalidating a translation that has only ever been
2573 	 * used on one vcpu.  However, that doesn't mean it has
2574 	 * only ever been used on one physical cpu, since vcpus
2575 	 * can move around between pcpus.  To cope with this, when
2576 	 * a vcpu moves from one pcpu to another, we need to tell
2577 	 * any vcpus running on the same core as this vcpu previously
2578 	 * ran to flush the TLB.  The TLB is shared between threads,
2579 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2580 	 */
2581 	if (prev_cpu != pcpu) {
2582 		if (prev_cpu >= 0 &&
2583 		    cpu_first_thread_sibling(prev_cpu) !=
2584 		    cpu_first_thread_sibling(pcpu))
2585 			radix_flush_cpu(kvm, prev_cpu, vcpu);
2586 		if (nested)
2587 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2588 		else
2589 			vcpu->arch.prev_cpu = pcpu;
2590 	}
2591 }
2592 
2593 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2594 {
2595 	int cpu;
2596 	struct paca_struct *tpaca;
2597 	struct kvm *kvm = vc->kvm;
2598 
2599 	cpu = vc->pcpu;
2600 	if (vcpu) {
2601 		if (vcpu->arch.timer_running) {
2602 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2603 			vcpu->arch.timer_running = 0;
2604 		}
2605 		cpu += vcpu->arch.ptid;
2606 		vcpu->cpu = vc->pcpu;
2607 		vcpu->arch.thread_cpu = cpu;
2608 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2609 	}
2610 	tpaca = paca_ptrs[cpu];
2611 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2612 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2613 	tpaca->kvm_hstate.fake_suspend = 0;
2614 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2615 	smp_wmb();
2616 	tpaca->kvm_hstate.kvm_vcore = vc;
2617 	if (cpu != smp_processor_id())
2618 		kvmppc_ipi_thread(cpu);
2619 }
2620 
2621 static void kvmppc_wait_for_nap(int n_threads)
2622 {
2623 	int cpu = smp_processor_id();
2624 	int i, loops;
2625 
2626 	if (n_threads <= 1)
2627 		return;
2628 	for (loops = 0; loops < 1000000; ++loops) {
2629 		/*
2630 		 * Check if all threads are finished.
2631 		 * We set the vcore pointer when starting a thread
2632 		 * and the thread clears it when finished, so we look
2633 		 * for any threads that still have a non-NULL vcore ptr.
2634 		 */
2635 		for (i = 1; i < n_threads; ++i)
2636 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2637 				break;
2638 		if (i == n_threads) {
2639 			HMT_medium();
2640 			return;
2641 		}
2642 		HMT_low();
2643 	}
2644 	HMT_medium();
2645 	for (i = 1; i < n_threads; ++i)
2646 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2647 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2648 }
2649 
2650 /*
2651  * Check that we are on thread 0 and that any other threads in
2652  * this core are off-line.  Then grab the threads so they can't
2653  * enter the kernel.
2654  */
2655 static int on_primary_thread(void)
2656 {
2657 	int cpu = smp_processor_id();
2658 	int thr;
2659 
2660 	/* Are we on a primary subcore? */
2661 	if (cpu_thread_in_subcore(cpu))
2662 		return 0;
2663 
2664 	thr = 0;
2665 	while (++thr < threads_per_subcore)
2666 		if (cpu_online(cpu + thr))
2667 			return 0;
2668 
2669 	/* Grab all hw threads so they can't go into the kernel */
2670 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2671 		if (kvmppc_grab_hwthread(cpu + thr)) {
2672 			/* Couldn't grab one; let the others go */
2673 			do {
2674 				kvmppc_release_hwthread(cpu + thr);
2675 			} while (--thr > 0);
2676 			return 0;
2677 		}
2678 	}
2679 	return 1;
2680 }
2681 
2682 /*
2683  * A list of virtual cores for each physical CPU.
2684  * These are vcores that could run but their runner VCPU tasks are
2685  * (or may be) preempted.
2686  */
2687 struct preempted_vcore_list {
2688 	struct list_head	list;
2689 	spinlock_t		lock;
2690 };
2691 
2692 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2693 
2694 static void init_vcore_lists(void)
2695 {
2696 	int cpu;
2697 
2698 	for_each_possible_cpu(cpu) {
2699 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2700 		spin_lock_init(&lp->lock);
2701 		INIT_LIST_HEAD(&lp->list);
2702 	}
2703 }
2704 
2705 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2706 {
2707 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2708 
2709 	vc->vcore_state = VCORE_PREEMPT;
2710 	vc->pcpu = smp_processor_id();
2711 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2712 		spin_lock(&lp->lock);
2713 		list_add_tail(&vc->preempt_list, &lp->list);
2714 		spin_unlock(&lp->lock);
2715 	}
2716 
2717 	/* Start accumulating stolen time */
2718 	kvmppc_core_start_stolen(vc);
2719 }
2720 
2721 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2722 {
2723 	struct preempted_vcore_list *lp;
2724 
2725 	kvmppc_core_end_stolen(vc);
2726 	if (!list_empty(&vc->preempt_list)) {
2727 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2728 		spin_lock(&lp->lock);
2729 		list_del_init(&vc->preempt_list);
2730 		spin_unlock(&lp->lock);
2731 	}
2732 	vc->vcore_state = VCORE_INACTIVE;
2733 }
2734 
2735 /*
2736  * This stores information about the virtual cores currently
2737  * assigned to a physical core.
2738  */
2739 struct core_info {
2740 	int		n_subcores;
2741 	int		max_subcore_threads;
2742 	int		total_threads;
2743 	int		subcore_threads[MAX_SUBCORES];
2744 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2745 };
2746 
2747 /*
2748  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2749  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2750  */
2751 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2752 
2753 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2754 {
2755 	memset(cip, 0, sizeof(*cip));
2756 	cip->n_subcores = 1;
2757 	cip->max_subcore_threads = vc->num_threads;
2758 	cip->total_threads = vc->num_threads;
2759 	cip->subcore_threads[0] = vc->num_threads;
2760 	cip->vc[0] = vc;
2761 }
2762 
2763 static bool subcore_config_ok(int n_subcores, int n_threads)
2764 {
2765 	/*
2766 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2767 	 * split-core mode, with one thread per subcore.
2768 	 */
2769 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2770 		return n_subcores <= 4 && n_threads == 1;
2771 
2772 	/* On POWER8, can only dynamically split if unsplit to begin with */
2773 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2774 		return false;
2775 	if (n_subcores > MAX_SUBCORES)
2776 		return false;
2777 	if (n_subcores > 1) {
2778 		if (!(dynamic_mt_modes & 2))
2779 			n_subcores = 4;
2780 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2781 			return false;
2782 	}
2783 
2784 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2785 }
2786 
2787 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2788 {
2789 	vc->entry_exit_map = 0;
2790 	vc->in_guest = 0;
2791 	vc->napping_threads = 0;
2792 	vc->conferring_threads = 0;
2793 	vc->tb_offset_applied = 0;
2794 }
2795 
2796 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2797 {
2798 	int n_threads = vc->num_threads;
2799 	int sub;
2800 
2801 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2802 		return false;
2803 
2804 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
2805 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2806 		return false;
2807 
2808 	/* Some POWER9 chips require all threads to be in the same MMU mode */
2809 	if (no_mixing_hpt_and_radix &&
2810 	    kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2811 		return false;
2812 
2813 	if (n_threads < cip->max_subcore_threads)
2814 		n_threads = cip->max_subcore_threads;
2815 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2816 		return false;
2817 	cip->max_subcore_threads = n_threads;
2818 
2819 	sub = cip->n_subcores;
2820 	++cip->n_subcores;
2821 	cip->total_threads += vc->num_threads;
2822 	cip->subcore_threads[sub] = vc->num_threads;
2823 	cip->vc[sub] = vc;
2824 	init_vcore_to_run(vc);
2825 	list_del_init(&vc->preempt_list);
2826 
2827 	return true;
2828 }
2829 
2830 /*
2831  * Work out whether it is possible to piggyback the execution of
2832  * vcore *pvc onto the execution of the other vcores described in *cip.
2833  */
2834 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2835 			  int target_threads)
2836 {
2837 	if (cip->total_threads + pvc->num_threads > target_threads)
2838 		return false;
2839 
2840 	return can_dynamic_split(pvc, cip);
2841 }
2842 
2843 static void prepare_threads(struct kvmppc_vcore *vc)
2844 {
2845 	int i;
2846 	struct kvm_vcpu *vcpu;
2847 
2848 	for_each_runnable_thread(i, vcpu, vc) {
2849 		if (signal_pending(vcpu->arch.run_task))
2850 			vcpu->arch.ret = -EINTR;
2851 		else if (vcpu->arch.vpa.update_pending ||
2852 			 vcpu->arch.slb_shadow.update_pending ||
2853 			 vcpu->arch.dtl.update_pending)
2854 			vcpu->arch.ret = RESUME_GUEST;
2855 		else
2856 			continue;
2857 		kvmppc_remove_runnable(vc, vcpu);
2858 		wake_up(&vcpu->arch.cpu_run);
2859 	}
2860 }
2861 
2862 static void collect_piggybacks(struct core_info *cip, int target_threads)
2863 {
2864 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2865 	struct kvmppc_vcore *pvc, *vcnext;
2866 
2867 	spin_lock(&lp->lock);
2868 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2869 		if (!spin_trylock(&pvc->lock))
2870 			continue;
2871 		prepare_threads(pvc);
2872 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2873 			list_del_init(&pvc->preempt_list);
2874 			if (pvc->runner == NULL) {
2875 				pvc->vcore_state = VCORE_INACTIVE;
2876 				kvmppc_core_end_stolen(pvc);
2877 			}
2878 			spin_unlock(&pvc->lock);
2879 			continue;
2880 		}
2881 		if (!can_piggyback(pvc, cip, target_threads)) {
2882 			spin_unlock(&pvc->lock);
2883 			continue;
2884 		}
2885 		kvmppc_core_end_stolen(pvc);
2886 		pvc->vcore_state = VCORE_PIGGYBACK;
2887 		if (cip->total_threads >= target_threads)
2888 			break;
2889 	}
2890 	spin_unlock(&lp->lock);
2891 }
2892 
2893 static bool recheck_signals_and_mmu(struct core_info *cip)
2894 {
2895 	int sub, i;
2896 	struct kvm_vcpu *vcpu;
2897 	struct kvmppc_vcore *vc;
2898 
2899 	for (sub = 0; sub < cip->n_subcores; ++sub) {
2900 		vc = cip->vc[sub];
2901 		if (!vc->kvm->arch.mmu_ready)
2902 			return true;
2903 		for_each_runnable_thread(i, vcpu, vc)
2904 			if (signal_pending(vcpu->arch.run_task))
2905 				return true;
2906 	}
2907 	return false;
2908 }
2909 
2910 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2911 {
2912 	int still_running = 0, i;
2913 	u64 now;
2914 	long ret;
2915 	struct kvm_vcpu *vcpu;
2916 
2917 	spin_lock(&vc->lock);
2918 	now = get_tb();
2919 	for_each_runnable_thread(i, vcpu, vc) {
2920 		/*
2921 		 * It's safe to unlock the vcore in the loop here, because
2922 		 * for_each_runnable_thread() is safe against removal of
2923 		 * the vcpu, and the vcore state is VCORE_EXITING here,
2924 		 * so any vcpus becoming runnable will have their arch.trap
2925 		 * set to zero and can't actually run in the guest.
2926 		 */
2927 		spin_unlock(&vc->lock);
2928 		/* cancel pending dec exception if dec is positive */
2929 		if (now < vcpu->arch.dec_expires &&
2930 		    kvmppc_core_pending_dec(vcpu))
2931 			kvmppc_core_dequeue_dec(vcpu);
2932 
2933 		trace_kvm_guest_exit(vcpu);
2934 
2935 		ret = RESUME_GUEST;
2936 		if (vcpu->arch.trap)
2937 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2938 						    vcpu->arch.run_task);
2939 
2940 		vcpu->arch.ret = ret;
2941 		vcpu->arch.trap = 0;
2942 
2943 		spin_lock(&vc->lock);
2944 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2945 			if (vcpu->arch.pending_exceptions)
2946 				kvmppc_core_prepare_to_enter(vcpu);
2947 			if (vcpu->arch.ceded)
2948 				kvmppc_set_timer(vcpu);
2949 			else
2950 				++still_running;
2951 		} else {
2952 			kvmppc_remove_runnable(vc, vcpu);
2953 			wake_up(&vcpu->arch.cpu_run);
2954 		}
2955 	}
2956 	if (!is_master) {
2957 		if (still_running > 0) {
2958 			kvmppc_vcore_preempt(vc);
2959 		} else if (vc->runner) {
2960 			vc->vcore_state = VCORE_PREEMPT;
2961 			kvmppc_core_start_stolen(vc);
2962 		} else {
2963 			vc->vcore_state = VCORE_INACTIVE;
2964 		}
2965 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2966 			/* make sure there's a candidate runner awake */
2967 			i = -1;
2968 			vcpu = next_runnable_thread(vc, &i);
2969 			wake_up(&vcpu->arch.cpu_run);
2970 		}
2971 	}
2972 	spin_unlock(&vc->lock);
2973 }
2974 
2975 /*
2976  * Clear core from the list of active host cores as we are about to
2977  * enter the guest. Only do this if it is the primary thread of the
2978  * core (not if a subcore) that is entering the guest.
2979  */
2980 static inline int kvmppc_clear_host_core(unsigned int cpu)
2981 {
2982 	int core;
2983 
2984 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2985 		return 0;
2986 	/*
2987 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2988 	 * later in kvmppc_start_thread and we need ensure that state is
2989 	 * visible to other CPUs only after we enter guest.
2990 	 */
2991 	core = cpu >> threads_shift;
2992 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2993 	return 0;
2994 }
2995 
2996 /*
2997  * Advertise this core as an active host core since we exited the guest
2998  * Only need to do this if it is the primary thread of the core that is
2999  * exiting.
3000  */
3001 static inline int kvmppc_set_host_core(unsigned int cpu)
3002 {
3003 	int core;
3004 
3005 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3006 		return 0;
3007 
3008 	/*
3009 	 * Memory barrier can be omitted here because we do a spin_unlock
3010 	 * immediately after this which provides the memory barrier.
3011 	 */
3012 	core = cpu >> threads_shift;
3013 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3014 	return 0;
3015 }
3016 
3017 static void set_irq_happened(int trap)
3018 {
3019 	switch (trap) {
3020 	case BOOK3S_INTERRUPT_EXTERNAL:
3021 		local_paca->irq_happened |= PACA_IRQ_EE;
3022 		break;
3023 	case BOOK3S_INTERRUPT_H_DOORBELL:
3024 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3025 		break;
3026 	case BOOK3S_INTERRUPT_HMI:
3027 		local_paca->irq_happened |= PACA_IRQ_HMI;
3028 		break;
3029 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3030 		replay_system_reset();
3031 		break;
3032 	}
3033 }
3034 
3035 /*
3036  * Run a set of guest threads on a physical core.
3037  * Called with vc->lock held.
3038  */
3039 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3040 {
3041 	struct kvm_vcpu *vcpu;
3042 	int i;
3043 	int srcu_idx;
3044 	struct core_info core_info;
3045 	struct kvmppc_vcore *pvc;
3046 	struct kvm_split_mode split_info, *sip;
3047 	int split, subcore_size, active;
3048 	int sub;
3049 	bool thr0_done;
3050 	unsigned long cmd_bit, stat_bit;
3051 	int pcpu, thr;
3052 	int target_threads;
3053 	int controlled_threads;
3054 	int trap;
3055 	bool is_power8;
3056 	bool hpt_on_radix;
3057 
3058 	/*
3059 	 * Remove from the list any threads that have a signal pending
3060 	 * or need a VPA update done
3061 	 */
3062 	prepare_threads(vc);
3063 
3064 	/* if the runner is no longer runnable, let the caller pick a new one */
3065 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3066 		return;
3067 
3068 	/*
3069 	 * Initialize *vc.
3070 	 */
3071 	init_vcore_to_run(vc);
3072 	vc->preempt_tb = TB_NIL;
3073 
3074 	/*
3075 	 * Number of threads that we will be controlling: the same as
3076 	 * the number of threads per subcore, except on POWER9,
3077 	 * where it's 1 because the threads are (mostly) independent.
3078 	 */
3079 	controlled_threads = threads_per_vcore(vc->kvm);
3080 
3081 	/*
3082 	 * Make sure we are running on primary threads, and that secondary
3083 	 * threads are offline.  Also check if the number of threads in this
3084 	 * guest are greater than the current system threads per guest.
3085 	 * On POWER9, we need to be not in independent-threads mode if
3086 	 * this is a HPT guest on a radix host machine where the
3087 	 * CPU threads may not be in different MMU modes.
3088 	 */
3089 	hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3090 		!kvm_is_radix(vc->kvm);
3091 	if (((controlled_threads > 1) &&
3092 	     ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3093 	    (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3094 		for_each_runnable_thread(i, vcpu, vc) {
3095 			vcpu->arch.ret = -EBUSY;
3096 			kvmppc_remove_runnable(vc, vcpu);
3097 			wake_up(&vcpu->arch.cpu_run);
3098 		}
3099 		goto out;
3100 	}
3101 
3102 	/*
3103 	 * See if we could run any other vcores on the physical core
3104 	 * along with this one.
3105 	 */
3106 	init_core_info(&core_info, vc);
3107 	pcpu = smp_processor_id();
3108 	target_threads = controlled_threads;
3109 	if (target_smt_mode && target_smt_mode < target_threads)
3110 		target_threads = target_smt_mode;
3111 	if (vc->num_threads < target_threads)
3112 		collect_piggybacks(&core_info, target_threads);
3113 
3114 	/*
3115 	 * On radix, arrange for TLB flushing if necessary.
3116 	 * This has to be done before disabling interrupts since
3117 	 * it uses smp_call_function().
3118 	 */
3119 	pcpu = smp_processor_id();
3120 	if (kvm_is_radix(vc->kvm)) {
3121 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3122 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3123 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3124 	}
3125 
3126 	/*
3127 	 * Hard-disable interrupts, and check resched flag and signals.
3128 	 * If we need to reschedule or deliver a signal, clean up
3129 	 * and return without going into the guest(s).
3130 	 * If the mmu_ready flag has been cleared, don't go into the
3131 	 * guest because that means a HPT resize operation is in progress.
3132 	 */
3133 	local_irq_disable();
3134 	hard_irq_disable();
3135 	if (lazy_irq_pending() || need_resched() ||
3136 	    recheck_signals_and_mmu(&core_info)) {
3137 		local_irq_enable();
3138 		vc->vcore_state = VCORE_INACTIVE;
3139 		/* Unlock all except the primary vcore */
3140 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3141 			pvc = core_info.vc[sub];
3142 			/* Put back on to the preempted vcores list */
3143 			kvmppc_vcore_preempt(pvc);
3144 			spin_unlock(&pvc->lock);
3145 		}
3146 		for (i = 0; i < controlled_threads; ++i)
3147 			kvmppc_release_hwthread(pcpu + i);
3148 		return;
3149 	}
3150 
3151 	kvmppc_clear_host_core(pcpu);
3152 
3153 	/* Decide on micro-threading (split-core) mode */
3154 	subcore_size = threads_per_subcore;
3155 	cmd_bit = stat_bit = 0;
3156 	split = core_info.n_subcores;
3157 	sip = NULL;
3158 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3159 		&& !cpu_has_feature(CPU_FTR_ARCH_300);
3160 
3161 	if (split > 1 || hpt_on_radix) {
3162 		sip = &split_info;
3163 		memset(&split_info, 0, sizeof(split_info));
3164 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3165 			split_info.vc[sub] = core_info.vc[sub];
3166 
3167 		if (is_power8) {
3168 			if (split == 2 && (dynamic_mt_modes & 2)) {
3169 				cmd_bit = HID0_POWER8_1TO2LPAR;
3170 				stat_bit = HID0_POWER8_2LPARMODE;
3171 			} else {
3172 				split = 4;
3173 				cmd_bit = HID0_POWER8_1TO4LPAR;
3174 				stat_bit = HID0_POWER8_4LPARMODE;
3175 			}
3176 			subcore_size = MAX_SMT_THREADS / split;
3177 			split_info.rpr = mfspr(SPRN_RPR);
3178 			split_info.pmmar = mfspr(SPRN_PMMAR);
3179 			split_info.ldbar = mfspr(SPRN_LDBAR);
3180 			split_info.subcore_size = subcore_size;
3181 		} else {
3182 			split_info.subcore_size = 1;
3183 			if (hpt_on_radix) {
3184 				/* Use the split_info for LPCR/LPIDR changes */
3185 				split_info.lpcr_req = vc->lpcr;
3186 				split_info.lpidr_req = vc->kvm->arch.lpid;
3187 				split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3188 				split_info.do_set = 1;
3189 			}
3190 		}
3191 
3192 		/* order writes to split_info before kvm_split_mode pointer */
3193 		smp_wmb();
3194 	}
3195 
3196 	for (thr = 0; thr < controlled_threads; ++thr) {
3197 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3198 
3199 		paca->kvm_hstate.tid = thr;
3200 		paca->kvm_hstate.napping = 0;
3201 		paca->kvm_hstate.kvm_split_mode = sip;
3202 	}
3203 
3204 	/* Initiate micro-threading (split-core) on POWER8 if required */
3205 	if (cmd_bit) {
3206 		unsigned long hid0 = mfspr(SPRN_HID0);
3207 
3208 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3209 		mb();
3210 		mtspr(SPRN_HID0, hid0);
3211 		isync();
3212 		for (;;) {
3213 			hid0 = mfspr(SPRN_HID0);
3214 			if (hid0 & stat_bit)
3215 				break;
3216 			cpu_relax();
3217 		}
3218 	}
3219 
3220 	/*
3221 	 * On POWER8, set RWMR register.
3222 	 * Since it only affects PURR and SPURR, it doesn't affect
3223 	 * the host, so we don't save/restore the host value.
3224 	 */
3225 	if (is_power8) {
3226 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3227 		int n_online = atomic_read(&vc->online_count);
3228 
3229 		/*
3230 		 * Use the 8-thread value if we're doing split-core
3231 		 * or if the vcore's online count looks bogus.
3232 		 */
3233 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3234 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3235 			rwmr_val = p8_rwmr_values[n_online];
3236 		mtspr(SPRN_RWMR, rwmr_val);
3237 	}
3238 
3239 	/* Start all the threads */
3240 	active = 0;
3241 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3242 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3243 		thr0_done = false;
3244 		active |= 1 << thr;
3245 		pvc = core_info.vc[sub];
3246 		pvc->pcpu = pcpu + thr;
3247 		for_each_runnable_thread(i, vcpu, pvc) {
3248 			kvmppc_start_thread(vcpu, pvc);
3249 			kvmppc_create_dtl_entry(vcpu, pvc);
3250 			trace_kvm_guest_enter(vcpu);
3251 			if (!vcpu->arch.ptid)
3252 				thr0_done = true;
3253 			active |= 1 << (thr + vcpu->arch.ptid);
3254 		}
3255 		/*
3256 		 * We need to start the first thread of each subcore
3257 		 * even if it doesn't have a vcpu.
3258 		 */
3259 		if (!thr0_done)
3260 			kvmppc_start_thread(NULL, pvc);
3261 	}
3262 
3263 	/*
3264 	 * Ensure that split_info.do_nap is set after setting
3265 	 * the vcore pointer in the PACA of the secondaries.
3266 	 */
3267 	smp_mb();
3268 
3269 	/*
3270 	 * When doing micro-threading, poke the inactive threads as well.
3271 	 * This gets them to the nap instruction after kvm_do_nap,
3272 	 * which reduces the time taken to unsplit later.
3273 	 * For POWER9 HPT guest on radix host, we need all the secondary
3274 	 * threads woken up so they can do the LPCR/LPIDR change.
3275 	 */
3276 	if (cmd_bit || hpt_on_radix) {
3277 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3278 		for (thr = 1; thr < threads_per_subcore; ++thr)
3279 			if (!(active & (1 << thr)))
3280 				kvmppc_ipi_thread(pcpu + thr);
3281 	}
3282 
3283 	vc->vcore_state = VCORE_RUNNING;
3284 	preempt_disable();
3285 
3286 	trace_kvmppc_run_core(vc, 0);
3287 
3288 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3289 		spin_unlock(&core_info.vc[sub]->lock);
3290 
3291 	guest_enter_irqoff();
3292 
3293 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3294 
3295 	this_cpu_disable_ftrace();
3296 
3297 	/*
3298 	 * Interrupts will be enabled once we get into the guest,
3299 	 * so tell lockdep that we're about to enable interrupts.
3300 	 */
3301 	trace_hardirqs_on();
3302 
3303 	trap = __kvmppc_vcore_entry();
3304 
3305 	trace_hardirqs_off();
3306 
3307 	this_cpu_enable_ftrace();
3308 
3309 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3310 
3311 	set_irq_happened(trap);
3312 
3313 	spin_lock(&vc->lock);
3314 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3315 	vc->vcore_state = VCORE_EXITING;
3316 
3317 	/* wait for secondary threads to finish writing their state to memory */
3318 	kvmppc_wait_for_nap(controlled_threads);
3319 
3320 	/* Return to whole-core mode if we split the core earlier */
3321 	if (cmd_bit) {
3322 		unsigned long hid0 = mfspr(SPRN_HID0);
3323 		unsigned long loops = 0;
3324 
3325 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3326 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3327 		mb();
3328 		mtspr(SPRN_HID0, hid0);
3329 		isync();
3330 		for (;;) {
3331 			hid0 = mfspr(SPRN_HID0);
3332 			if (!(hid0 & stat_bit))
3333 				break;
3334 			cpu_relax();
3335 			++loops;
3336 		}
3337 	} else if (hpt_on_radix) {
3338 		/* Wait for all threads to have seen final sync */
3339 		for (thr = 1; thr < controlled_threads; ++thr) {
3340 			struct paca_struct *paca = paca_ptrs[pcpu + thr];
3341 
3342 			while (paca->kvm_hstate.kvm_split_mode) {
3343 				HMT_low();
3344 				barrier();
3345 			}
3346 			HMT_medium();
3347 		}
3348 	}
3349 	split_info.do_nap = 0;
3350 
3351 	kvmppc_set_host_core(pcpu);
3352 
3353 	local_irq_enable();
3354 	guest_exit();
3355 
3356 	/* Let secondaries go back to the offline loop */
3357 	for (i = 0; i < controlled_threads; ++i) {
3358 		kvmppc_release_hwthread(pcpu + i);
3359 		if (sip && sip->napped[i])
3360 			kvmppc_ipi_thread(pcpu + i);
3361 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3362 	}
3363 
3364 	spin_unlock(&vc->lock);
3365 
3366 	/* make sure updates to secondary vcpu structs are visible now */
3367 	smp_mb();
3368 
3369 	preempt_enable();
3370 
3371 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3372 		pvc = core_info.vc[sub];
3373 		post_guest_process(pvc, pvc == vc);
3374 	}
3375 
3376 	spin_lock(&vc->lock);
3377 
3378  out:
3379 	vc->vcore_state = VCORE_INACTIVE;
3380 	trace_kvmppc_run_core(vc, 1);
3381 }
3382 
3383 /*
3384  * Load up hypervisor-mode registers on P9.
3385  */
3386 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3387 				     unsigned long lpcr)
3388 {
3389 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3390 	s64 hdec;
3391 	u64 tb, purr, spurr;
3392 	int trap;
3393 	unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3394 	unsigned long host_ciabr = mfspr(SPRN_CIABR);
3395 	unsigned long host_dawr = mfspr(SPRN_DAWR);
3396 	unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3397 	unsigned long host_psscr = mfspr(SPRN_PSSCR);
3398 	unsigned long host_pidr = mfspr(SPRN_PID);
3399 
3400 	hdec = time_limit - mftb();
3401 	if (hdec < 0)
3402 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3403 	mtspr(SPRN_HDEC, hdec);
3404 
3405 	if (vc->tb_offset) {
3406 		u64 new_tb = mftb() + vc->tb_offset;
3407 		mtspr(SPRN_TBU40, new_tb);
3408 		tb = mftb();
3409 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3410 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3411 		vc->tb_offset_applied = vc->tb_offset;
3412 	}
3413 
3414 	if (vc->pcr)
3415 		mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3416 	mtspr(SPRN_DPDES, vc->dpdes);
3417 	mtspr(SPRN_VTB, vc->vtb);
3418 
3419 	local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3420 	local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3421 	mtspr(SPRN_PURR, vcpu->arch.purr);
3422 	mtspr(SPRN_SPURR, vcpu->arch.spurr);
3423 
3424 	if (dawr_enabled()) {
3425 		mtspr(SPRN_DAWR, vcpu->arch.dawr);
3426 		mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3427 	}
3428 	mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3429 	mtspr(SPRN_IC, vcpu->arch.ic);
3430 	mtspr(SPRN_PID, vcpu->arch.pid);
3431 
3432 	mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3433 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3434 
3435 	mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3436 
3437 	mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3438 	mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3439 	mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3440 	mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3441 
3442 	mtspr(SPRN_AMOR, ~0UL);
3443 
3444 	mtspr(SPRN_LPCR, lpcr);
3445 	isync();
3446 
3447 	kvmppc_xive_push_vcpu(vcpu);
3448 
3449 	mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3450 	mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3451 
3452 	trap = __kvmhv_vcpu_entry_p9(vcpu);
3453 
3454 	/* Advance host PURR/SPURR by the amount used by guest */
3455 	purr = mfspr(SPRN_PURR);
3456 	spurr = mfspr(SPRN_SPURR);
3457 	mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3458 	      purr - vcpu->arch.purr);
3459 	mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3460 	      spurr - vcpu->arch.spurr);
3461 	vcpu->arch.purr = purr;
3462 	vcpu->arch.spurr = spurr;
3463 
3464 	vcpu->arch.ic = mfspr(SPRN_IC);
3465 	vcpu->arch.pid = mfspr(SPRN_PID);
3466 	vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3467 
3468 	vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3469 	vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3470 	vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3471 	vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3472 
3473 	/* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3474 	mtspr(SPRN_PSSCR, host_psscr |
3475 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3476 	mtspr(SPRN_HFSCR, host_hfscr);
3477 	mtspr(SPRN_CIABR, host_ciabr);
3478 	mtspr(SPRN_DAWR, host_dawr);
3479 	mtspr(SPRN_DAWRX, host_dawrx);
3480 	mtspr(SPRN_PID, host_pidr);
3481 
3482 	/*
3483 	 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3484 	 * case we interrupted the guest between a tlbie and a ptesync.
3485 	 */
3486 	asm volatile("eieio; tlbsync; ptesync");
3487 
3488 	mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);	/* restore host LPID */
3489 	isync();
3490 
3491 	vc->dpdes = mfspr(SPRN_DPDES);
3492 	vc->vtb = mfspr(SPRN_VTB);
3493 	mtspr(SPRN_DPDES, 0);
3494 	if (vc->pcr)
3495 		mtspr(SPRN_PCR, PCR_MASK);
3496 
3497 	if (vc->tb_offset_applied) {
3498 		u64 new_tb = mftb() - vc->tb_offset_applied;
3499 		mtspr(SPRN_TBU40, new_tb);
3500 		tb = mftb();
3501 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3502 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3503 		vc->tb_offset_applied = 0;
3504 	}
3505 
3506 	mtspr(SPRN_HDEC, 0x7fffffff);
3507 	mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3508 
3509 	return trap;
3510 }
3511 
3512 /*
3513  * Virtual-mode guest entry for POWER9 and later when the host and
3514  * guest are both using the radix MMU.  The LPIDR has already been set.
3515  */
3516 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3517 			 unsigned long lpcr)
3518 {
3519 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3520 	unsigned long host_dscr = mfspr(SPRN_DSCR);
3521 	unsigned long host_tidr = mfspr(SPRN_TIDR);
3522 	unsigned long host_iamr = mfspr(SPRN_IAMR);
3523 	unsigned long host_amr = mfspr(SPRN_AMR);
3524 	s64 dec;
3525 	u64 tb;
3526 	int trap, save_pmu;
3527 
3528 	dec = mfspr(SPRN_DEC);
3529 	tb = mftb();
3530 	if (dec < 512)
3531 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3532 	local_paca->kvm_hstate.dec_expires = dec + tb;
3533 	if (local_paca->kvm_hstate.dec_expires < time_limit)
3534 		time_limit = local_paca->kvm_hstate.dec_expires;
3535 
3536 	vcpu->arch.ceded = 0;
3537 
3538 	kvmhv_save_host_pmu();		/* saves it to PACA kvm_hstate */
3539 
3540 	kvmppc_subcore_enter_guest();
3541 
3542 	vc->entry_exit_map = 1;
3543 	vc->in_guest = 1;
3544 
3545 	if (vcpu->arch.vpa.pinned_addr) {
3546 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3547 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3548 		lp->yield_count = cpu_to_be32(yield_count);
3549 		vcpu->arch.vpa.dirty = 1;
3550 	}
3551 
3552 	if (cpu_has_feature(CPU_FTR_TM) ||
3553 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3554 		kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3555 
3556 	kvmhv_load_guest_pmu(vcpu);
3557 
3558 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3559 	load_fp_state(&vcpu->arch.fp);
3560 #ifdef CONFIG_ALTIVEC
3561 	load_vr_state(&vcpu->arch.vr);
3562 #endif
3563 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3564 
3565 	mtspr(SPRN_DSCR, vcpu->arch.dscr);
3566 	mtspr(SPRN_IAMR, vcpu->arch.iamr);
3567 	mtspr(SPRN_PSPB, vcpu->arch.pspb);
3568 	mtspr(SPRN_FSCR, vcpu->arch.fscr);
3569 	mtspr(SPRN_TAR, vcpu->arch.tar);
3570 	mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3571 	mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3572 	mtspr(SPRN_BESCR, vcpu->arch.bescr);
3573 	mtspr(SPRN_WORT, vcpu->arch.wort);
3574 	mtspr(SPRN_TIDR, vcpu->arch.tid);
3575 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3576 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3577 	mtspr(SPRN_AMR, vcpu->arch.amr);
3578 	mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3579 
3580 	if (!(vcpu->arch.ctrl & 1))
3581 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3582 
3583 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3584 
3585 	if (kvmhv_on_pseries()) {
3586 		/*
3587 		 * We need to save and restore the guest visible part of the
3588 		 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3589 		 * doesn't do this for us. Note only required if pseries since
3590 		 * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3591 		 */
3592 		unsigned long host_psscr;
3593 		/* call our hypervisor to load up HV regs and go */
3594 		struct hv_guest_state hvregs;
3595 
3596 		host_psscr = mfspr(SPRN_PSSCR_PR);
3597 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3598 		kvmhv_save_hv_regs(vcpu, &hvregs);
3599 		hvregs.lpcr = lpcr;
3600 		vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3601 		hvregs.version = HV_GUEST_STATE_VERSION;
3602 		if (vcpu->arch.nested) {
3603 			hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3604 			hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3605 		} else {
3606 			hvregs.lpid = vcpu->kvm->arch.lpid;
3607 			hvregs.vcpu_token = vcpu->vcpu_id;
3608 		}
3609 		hvregs.hdec_expiry = time_limit;
3610 		trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3611 					  __pa(&vcpu->arch.regs));
3612 		kvmhv_restore_hv_return_state(vcpu, &hvregs);
3613 		vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3614 		vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3615 		vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3616 		vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3617 		mtspr(SPRN_PSSCR_PR, host_psscr);
3618 
3619 		/* H_CEDE has to be handled now, not later */
3620 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3621 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3622 			kvmppc_nested_cede(vcpu);
3623 			kvmppc_set_gpr(vcpu, 3, 0);
3624 			trap = 0;
3625 		}
3626 	} else {
3627 		trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3628 	}
3629 
3630 	vcpu->arch.slb_max = 0;
3631 	dec = mfspr(SPRN_DEC);
3632 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3633 		dec = (s32) dec;
3634 	tb = mftb();
3635 	vcpu->arch.dec_expires = dec + tb;
3636 	vcpu->cpu = -1;
3637 	vcpu->arch.thread_cpu = -1;
3638 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3639 
3640 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
3641 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
3642 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
3643 	vcpu->arch.tar = mfspr(SPRN_TAR);
3644 	vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3645 	vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3646 	vcpu->arch.bescr = mfspr(SPRN_BESCR);
3647 	vcpu->arch.wort = mfspr(SPRN_WORT);
3648 	vcpu->arch.tid = mfspr(SPRN_TIDR);
3649 	vcpu->arch.amr = mfspr(SPRN_AMR);
3650 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3651 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
3652 
3653 	mtspr(SPRN_PSPB, 0);
3654 	mtspr(SPRN_WORT, 0);
3655 	mtspr(SPRN_UAMOR, 0);
3656 	mtspr(SPRN_DSCR, host_dscr);
3657 	mtspr(SPRN_TIDR, host_tidr);
3658 	mtspr(SPRN_IAMR, host_iamr);
3659 	mtspr(SPRN_PSPB, 0);
3660 
3661 	if (host_amr != vcpu->arch.amr)
3662 		mtspr(SPRN_AMR, host_amr);
3663 
3664 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3665 	store_fp_state(&vcpu->arch.fp);
3666 #ifdef CONFIG_ALTIVEC
3667 	store_vr_state(&vcpu->arch.vr);
3668 #endif
3669 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3670 
3671 	if (cpu_has_feature(CPU_FTR_TM) ||
3672 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3673 		kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3674 
3675 	save_pmu = 1;
3676 	if (vcpu->arch.vpa.pinned_addr) {
3677 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3678 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3679 		lp->yield_count = cpu_to_be32(yield_count);
3680 		vcpu->arch.vpa.dirty = 1;
3681 		save_pmu = lp->pmcregs_in_use;
3682 	}
3683 	/* Must save pmu if this guest is capable of running nested guests */
3684 	save_pmu |= nesting_enabled(vcpu->kvm);
3685 
3686 	kvmhv_save_guest_pmu(vcpu, save_pmu);
3687 
3688 	vc->entry_exit_map = 0x101;
3689 	vc->in_guest = 0;
3690 
3691 	mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3692 	mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3693 
3694 	kvmhv_load_host_pmu();
3695 
3696 	kvmppc_subcore_exit_guest();
3697 
3698 	return trap;
3699 }
3700 
3701 /*
3702  * Wait for some other vcpu thread to execute us, and
3703  * wake us up when we need to handle something in the host.
3704  */
3705 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3706 				 struct kvm_vcpu *vcpu, int wait_state)
3707 {
3708 	DEFINE_WAIT(wait);
3709 
3710 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3711 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3712 		spin_unlock(&vc->lock);
3713 		schedule();
3714 		spin_lock(&vc->lock);
3715 	}
3716 	finish_wait(&vcpu->arch.cpu_run, &wait);
3717 }
3718 
3719 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3720 {
3721 	if (!halt_poll_ns_grow)
3722 		return;
3723 
3724 	vc->halt_poll_ns *= halt_poll_ns_grow;
3725 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3726 		vc->halt_poll_ns = halt_poll_ns_grow_start;
3727 }
3728 
3729 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3730 {
3731 	if (halt_poll_ns_shrink == 0)
3732 		vc->halt_poll_ns = 0;
3733 	else
3734 		vc->halt_poll_ns /= halt_poll_ns_shrink;
3735 }
3736 
3737 #ifdef CONFIG_KVM_XICS
3738 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3739 {
3740 	if (!xics_on_xive())
3741 		return false;
3742 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3743 		vcpu->arch.xive_saved_state.cppr;
3744 }
3745 #else
3746 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3747 {
3748 	return false;
3749 }
3750 #endif /* CONFIG_KVM_XICS */
3751 
3752 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3753 {
3754 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3755 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3756 		return true;
3757 
3758 	return false;
3759 }
3760 
3761 /*
3762  * Check to see if any of the runnable vcpus on the vcore have pending
3763  * exceptions or are no longer ceded
3764  */
3765 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3766 {
3767 	struct kvm_vcpu *vcpu;
3768 	int i;
3769 
3770 	for_each_runnable_thread(i, vcpu, vc) {
3771 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3772 			return 1;
3773 	}
3774 
3775 	return 0;
3776 }
3777 
3778 /*
3779  * All the vcpus in this vcore are idle, so wait for a decrementer
3780  * or external interrupt to one of the vcpus.  vc->lock is held.
3781  */
3782 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3783 {
3784 	ktime_t cur, start_poll, start_wait;
3785 	int do_sleep = 1;
3786 	u64 block_ns;
3787 	DECLARE_SWAITQUEUE(wait);
3788 
3789 	/* Poll for pending exceptions and ceded state */
3790 	cur = start_poll = ktime_get();
3791 	if (vc->halt_poll_ns) {
3792 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3793 		++vc->runner->stat.halt_attempted_poll;
3794 
3795 		vc->vcore_state = VCORE_POLLING;
3796 		spin_unlock(&vc->lock);
3797 
3798 		do {
3799 			if (kvmppc_vcore_check_block(vc)) {
3800 				do_sleep = 0;
3801 				break;
3802 			}
3803 			cur = ktime_get();
3804 		} while (single_task_running() && ktime_before(cur, stop));
3805 
3806 		spin_lock(&vc->lock);
3807 		vc->vcore_state = VCORE_INACTIVE;
3808 
3809 		if (!do_sleep) {
3810 			++vc->runner->stat.halt_successful_poll;
3811 			goto out;
3812 		}
3813 	}
3814 
3815 	prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3816 
3817 	if (kvmppc_vcore_check_block(vc)) {
3818 		finish_swait(&vc->wq, &wait);
3819 		do_sleep = 0;
3820 		/* If we polled, count this as a successful poll */
3821 		if (vc->halt_poll_ns)
3822 			++vc->runner->stat.halt_successful_poll;
3823 		goto out;
3824 	}
3825 
3826 	start_wait = ktime_get();
3827 
3828 	vc->vcore_state = VCORE_SLEEPING;
3829 	trace_kvmppc_vcore_blocked(vc, 0);
3830 	spin_unlock(&vc->lock);
3831 	schedule();
3832 	finish_swait(&vc->wq, &wait);
3833 	spin_lock(&vc->lock);
3834 	vc->vcore_state = VCORE_INACTIVE;
3835 	trace_kvmppc_vcore_blocked(vc, 1);
3836 	++vc->runner->stat.halt_successful_wait;
3837 
3838 	cur = ktime_get();
3839 
3840 out:
3841 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3842 
3843 	/* Attribute wait time */
3844 	if (do_sleep) {
3845 		vc->runner->stat.halt_wait_ns +=
3846 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3847 		/* Attribute failed poll time */
3848 		if (vc->halt_poll_ns)
3849 			vc->runner->stat.halt_poll_fail_ns +=
3850 				ktime_to_ns(start_wait) -
3851 				ktime_to_ns(start_poll);
3852 	} else {
3853 		/* Attribute successful poll time */
3854 		if (vc->halt_poll_ns)
3855 			vc->runner->stat.halt_poll_success_ns +=
3856 				ktime_to_ns(cur) -
3857 				ktime_to_ns(start_poll);
3858 	}
3859 
3860 	/* Adjust poll time */
3861 	if (halt_poll_ns) {
3862 		if (block_ns <= vc->halt_poll_ns)
3863 			;
3864 		/* We slept and blocked for longer than the max halt time */
3865 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3866 			shrink_halt_poll_ns(vc);
3867 		/* We slept and our poll time is too small */
3868 		else if (vc->halt_poll_ns < halt_poll_ns &&
3869 				block_ns < halt_poll_ns)
3870 			grow_halt_poll_ns(vc);
3871 		if (vc->halt_poll_ns > halt_poll_ns)
3872 			vc->halt_poll_ns = halt_poll_ns;
3873 	} else
3874 		vc->halt_poll_ns = 0;
3875 
3876 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3877 }
3878 
3879 /*
3880  * This never fails for a radix guest, as none of the operations it does
3881  * for a radix guest can fail or have a way to report failure.
3882  * kvmhv_run_single_vcpu() relies on this fact.
3883  */
3884 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3885 {
3886 	int r = 0;
3887 	struct kvm *kvm = vcpu->kvm;
3888 
3889 	mutex_lock(&kvm->arch.mmu_setup_lock);
3890 	if (!kvm->arch.mmu_ready) {
3891 		if (!kvm_is_radix(kvm))
3892 			r = kvmppc_hv_setup_htab_rma(vcpu);
3893 		if (!r) {
3894 			if (cpu_has_feature(CPU_FTR_ARCH_300))
3895 				kvmppc_setup_partition_table(kvm);
3896 			kvm->arch.mmu_ready = 1;
3897 		}
3898 	}
3899 	mutex_unlock(&kvm->arch.mmu_setup_lock);
3900 	return r;
3901 }
3902 
3903 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3904 {
3905 	int n_ceded, i, r;
3906 	struct kvmppc_vcore *vc;
3907 	struct kvm_vcpu *v;
3908 
3909 	trace_kvmppc_run_vcpu_enter(vcpu);
3910 
3911 	kvm_run->exit_reason = 0;
3912 	vcpu->arch.ret = RESUME_GUEST;
3913 	vcpu->arch.trap = 0;
3914 	kvmppc_update_vpas(vcpu);
3915 
3916 	/*
3917 	 * Synchronize with other threads in this virtual core
3918 	 */
3919 	vc = vcpu->arch.vcore;
3920 	spin_lock(&vc->lock);
3921 	vcpu->arch.ceded = 0;
3922 	vcpu->arch.run_task = current;
3923 	vcpu->arch.kvm_run = kvm_run;
3924 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3925 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3926 	vcpu->arch.busy_preempt = TB_NIL;
3927 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3928 	++vc->n_runnable;
3929 
3930 	/*
3931 	 * This happens the first time this is called for a vcpu.
3932 	 * If the vcore is already running, we may be able to start
3933 	 * this thread straight away and have it join in.
3934 	 */
3935 	if (!signal_pending(current)) {
3936 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
3937 		     vc->vcore_state == VCORE_RUNNING) &&
3938 			   !VCORE_IS_EXITING(vc)) {
3939 			kvmppc_create_dtl_entry(vcpu, vc);
3940 			kvmppc_start_thread(vcpu, vc);
3941 			trace_kvm_guest_enter(vcpu);
3942 		} else if (vc->vcore_state == VCORE_SLEEPING) {
3943 			swake_up_one(&vc->wq);
3944 		}
3945 
3946 	}
3947 
3948 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3949 	       !signal_pending(current)) {
3950 		/* See if the MMU is ready to go */
3951 		if (!vcpu->kvm->arch.mmu_ready) {
3952 			spin_unlock(&vc->lock);
3953 			r = kvmhv_setup_mmu(vcpu);
3954 			spin_lock(&vc->lock);
3955 			if (r) {
3956 				kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3957 				kvm_run->fail_entry.
3958 					hardware_entry_failure_reason = 0;
3959 				vcpu->arch.ret = r;
3960 				break;
3961 			}
3962 		}
3963 
3964 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3965 			kvmppc_vcore_end_preempt(vc);
3966 
3967 		if (vc->vcore_state != VCORE_INACTIVE) {
3968 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3969 			continue;
3970 		}
3971 		for_each_runnable_thread(i, v, vc) {
3972 			kvmppc_core_prepare_to_enter(v);
3973 			if (signal_pending(v->arch.run_task)) {
3974 				kvmppc_remove_runnable(vc, v);
3975 				v->stat.signal_exits++;
3976 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3977 				v->arch.ret = -EINTR;
3978 				wake_up(&v->arch.cpu_run);
3979 			}
3980 		}
3981 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3982 			break;
3983 		n_ceded = 0;
3984 		for_each_runnable_thread(i, v, vc) {
3985 			if (!kvmppc_vcpu_woken(v))
3986 				n_ceded += v->arch.ceded;
3987 			else
3988 				v->arch.ceded = 0;
3989 		}
3990 		vc->runner = vcpu;
3991 		if (n_ceded == vc->n_runnable) {
3992 			kvmppc_vcore_blocked(vc);
3993 		} else if (need_resched()) {
3994 			kvmppc_vcore_preempt(vc);
3995 			/* Let something else run */
3996 			cond_resched_lock(&vc->lock);
3997 			if (vc->vcore_state == VCORE_PREEMPT)
3998 				kvmppc_vcore_end_preempt(vc);
3999 		} else {
4000 			kvmppc_run_core(vc);
4001 		}
4002 		vc->runner = NULL;
4003 	}
4004 
4005 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4006 	       (vc->vcore_state == VCORE_RUNNING ||
4007 		vc->vcore_state == VCORE_EXITING ||
4008 		vc->vcore_state == VCORE_PIGGYBACK))
4009 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4010 
4011 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4012 		kvmppc_vcore_end_preempt(vc);
4013 
4014 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4015 		kvmppc_remove_runnable(vc, vcpu);
4016 		vcpu->stat.signal_exits++;
4017 		kvm_run->exit_reason = KVM_EXIT_INTR;
4018 		vcpu->arch.ret = -EINTR;
4019 	}
4020 
4021 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4022 		/* Wake up some vcpu to run the core */
4023 		i = -1;
4024 		v = next_runnable_thread(vc, &i);
4025 		wake_up(&v->arch.cpu_run);
4026 	}
4027 
4028 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4029 	spin_unlock(&vc->lock);
4030 	return vcpu->arch.ret;
4031 }
4032 
4033 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4034 			  struct kvm_vcpu *vcpu, u64 time_limit,
4035 			  unsigned long lpcr)
4036 {
4037 	int trap, r, pcpu;
4038 	int srcu_idx, lpid;
4039 	struct kvmppc_vcore *vc;
4040 	struct kvm *kvm = vcpu->kvm;
4041 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4042 
4043 	trace_kvmppc_run_vcpu_enter(vcpu);
4044 
4045 	kvm_run->exit_reason = 0;
4046 	vcpu->arch.ret = RESUME_GUEST;
4047 	vcpu->arch.trap = 0;
4048 
4049 	vc = vcpu->arch.vcore;
4050 	vcpu->arch.ceded = 0;
4051 	vcpu->arch.run_task = current;
4052 	vcpu->arch.kvm_run = kvm_run;
4053 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4054 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4055 	vcpu->arch.busy_preempt = TB_NIL;
4056 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4057 	vc->runnable_threads[0] = vcpu;
4058 	vc->n_runnable = 1;
4059 	vc->runner = vcpu;
4060 
4061 	/* See if the MMU is ready to go */
4062 	if (!kvm->arch.mmu_ready)
4063 		kvmhv_setup_mmu(vcpu);
4064 
4065 	if (need_resched())
4066 		cond_resched();
4067 
4068 	kvmppc_update_vpas(vcpu);
4069 
4070 	init_vcore_to_run(vc);
4071 	vc->preempt_tb = TB_NIL;
4072 
4073 	preempt_disable();
4074 	pcpu = smp_processor_id();
4075 	vc->pcpu = pcpu;
4076 	kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4077 
4078 	local_irq_disable();
4079 	hard_irq_disable();
4080 	if (signal_pending(current))
4081 		goto sigpend;
4082 	if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4083 		goto out;
4084 
4085 	if (!nested) {
4086 		kvmppc_core_prepare_to_enter(vcpu);
4087 		if (vcpu->arch.doorbell_request) {
4088 			vc->dpdes = 1;
4089 			smp_wmb();
4090 			vcpu->arch.doorbell_request = 0;
4091 		}
4092 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4093 			     &vcpu->arch.pending_exceptions))
4094 			lpcr |= LPCR_MER;
4095 	} else if (vcpu->arch.pending_exceptions ||
4096 		   vcpu->arch.doorbell_request ||
4097 		   xive_interrupt_pending(vcpu)) {
4098 		vcpu->arch.ret = RESUME_HOST;
4099 		goto out;
4100 	}
4101 
4102 	kvmppc_clear_host_core(pcpu);
4103 
4104 	local_paca->kvm_hstate.tid = 0;
4105 	local_paca->kvm_hstate.napping = 0;
4106 	local_paca->kvm_hstate.kvm_split_mode = NULL;
4107 	kvmppc_start_thread(vcpu, vc);
4108 	kvmppc_create_dtl_entry(vcpu, vc);
4109 	trace_kvm_guest_enter(vcpu);
4110 
4111 	vc->vcore_state = VCORE_RUNNING;
4112 	trace_kvmppc_run_core(vc, 0);
4113 
4114 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4115 		lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4116 		mtspr(SPRN_LPID, lpid);
4117 		isync();
4118 		kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4119 	}
4120 
4121 	guest_enter_irqoff();
4122 
4123 	srcu_idx = srcu_read_lock(&kvm->srcu);
4124 
4125 	this_cpu_disable_ftrace();
4126 
4127 	/* Tell lockdep that we're about to enable interrupts */
4128 	trace_hardirqs_on();
4129 
4130 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4131 	vcpu->arch.trap = trap;
4132 
4133 	trace_hardirqs_off();
4134 
4135 	this_cpu_enable_ftrace();
4136 
4137 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4138 
4139 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4140 		mtspr(SPRN_LPID, kvm->arch.host_lpid);
4141 		isync();
4142 	}
4143 
4144 	set_irq_happened(trap);
4145 
4146 	kvmppc_set_host_core(pcpu);
4147 
4148 	local_irq_enable();
4149 	guest_exit();
4150 
4151 	cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4152 
4153 	preempt_enable();
4154 
4155 	/*
4156 	 * cancel pending decrementer exception if DEC is now positive, or if
4157 	 * entering a nested guest in which case the decrementer is now owned
4158 	 * by L2 and the L1 decrementer is provided in hdec_expires
4159 	 */
4160 	if (kvmppc_core_pending_dec(vcpu) &&
4161 			((get_tb() < vcpu->arch.dec_expires) ||
4162 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4163 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4164 		kvmppc_core_dequeue_dec(vcpu);
4165 
4166 	trace_kvm_guest_exit(vcpu);
4167 	r = RESUME_GUEST;
4168 	if (trap) {
4169 		if (!nested)
4170 			r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4171 		else
4172 			r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4173 	}
4174 	vcpu->arch.ret = r;
4175 
4176 	if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4177 	    !kvmppc_vcpu_woken(vcpu)) {
4178 		kvmppc_set_timer(vcpu);
4179 		while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4180 			if (signal_pending(current)) {
4181 				vcpu->stat.signal_exits++;
4182 				kvm_run->exit_reason = KVM_EXIT_INTR;
4183 				vcpu->arch.ret = -EINTR;
4184 				break;
4185 			}
4186 			spin_lock(&vc->lock);
4187 			kvmppc_vcore_blocked(vc);
4188 			spin_unlock(&vc->lock);
4189 		}
4190 	}
4191 	vcpu->arch.ceded = 0;
4192 
4193 	vc->vcore_state = VCORE_INACTIVE;
4194 	trace_kvmppc_run_core(vc, 1);
4195 
4196  done:
4197 	kvmppc_remove_runnable(vc, vcpu);
4198 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4199 
4200 	return vcpu->arch.ret;
4201 
4202  sigpend:
4203 	vcpu->stat.signal_exits++;
4204 	kvm_run->exit_reason = KVM_EXIT_INTR;
4205 	vcpu->arch.ret = -EINTR;
4206  out:
4207 	local_irq_enable();
4208 	preempt_enable();
4209 	goto done;
4210 }
4211 
4212 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4213 {
4214 	int r;
4215 	int srcu_idx;
4216 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
4217 	unsigned long user_tar = 0;
4218 	unsigned int user_vrsave;
4219 	struct kvm *kvm;
4220 
4221 	if (!vcpu->arch.sane) {
4222 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4223 		return -EINVAL;
4224 	}
4225 
4226 	/*
4227 	 * Don't allow entry with a suspended transaction, because
4228 	 * the guest entry/exit code will lose it.
4229 	 * If the guest has TM enabled, save away their TM-related SPRs
4230 	 * (they will get restored by the TM unavailable interrupt).
4231 	 */
4232 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4233 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4234 	    (current->thread.regs->msr & MSR_TM)) {
4235 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4236 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4237 			run->fail_entry.hardware_entry_failure_reason = 0;
4238 			return -EINVAL;
4239 		}
4240 		/* Enable TM so we can read the TM SPRs */
4241 		mtmsr(mfmsr() | MSR_TM);
4242 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4243 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4244 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4245 		current->thread.regs->msr &= ~MSR_TM;
4246 	}
4247 #endif
4248 
4249 	/*
4250 	 * Force online to 1 for the sake of old userspace which doesn't
4251 	 * set it.
4252 	 */
4253 	if (!vcpu->arch.online) {
4254 		atomic_inc(&vcpu->arch.vcore->online_count);
4255 		vcpu->arch.online = 1;
4256 	}
4257 
4258 	kvmppc_core_prepare_to_enter(vcpu);
4259 
4260 	/* No need to go into the guest when all we'll do is come back out */
4261 	if (signal_pending(current)) {
4262 		run->exit_reason = KVM_EXIT_INTR;
4263 		return -EINTR;
4264 	}
4265 
4266 	kvm = vcpu->kvm;
4267 	atomic_inc(&kvm->arch.vcpus_running);
4268 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4269 	smp_mb();
4270 
4271 	flush_all_to_thread(current);
4272 
4273 	/* Save userspace EBB and other register values */
4274 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4275 		ebb_regs[0] = mfspr(SPRN_EBBHR);
4276 		ebb_regs[1] = mfspr(SPRN_EBBRR);
4277 		ebb_regs[2] = mfspr(SPRN_BESCR);
4278 		user_tar = mfspr(SPRN_TAR);
4279 	}
4280 	user_vrsave = mfspr(SPRN_VRSAVE);
4281 
4282 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4283 	vcpu->arch.pgdir = kvm->mm->pgd;
4284 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4285 
4286 	do {
4287 		/*
4288 		 * The early POWER9 chips that can't mix radix and HPT threads
4289 		 * on the same core also need the workaround for the problem
4290 		 * where the TLB would prefetch entries in the guest exit path
4291 		 * for radix guests using the guest PIDR value and LPID 0.
4292 		 * The workaround is in the old path (kvmppc_run_vcpu())
4293 		 * but not the new path (kvmhv_run_single_vcpu()).
4294 		 */
4295 		if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4296 		    !no_mixing_hpt_and_radix)
4297 			r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4298 						  vcpu->arch.vcore->lpcr);
4299 		else
4300 			r = kvmppc_run_vcpu(run, vcpu);
4301 
4302 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4303 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4304 			trace_kvm_hcall_enter(vcpu);
4305 			r = kvmppc_pseries_do_hcall(vcpu);
4306 			trace_kvm_hcall_exit(vcpu, r);
4307 			kvmppc_core_prepare_to_enter(vcpu);
4308 		} else if (r == RESUME_PAGE_FAULT) {
4309 			srcu_idx = srcu_read_lock(&kvm->srcu);
4310 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
4311 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4312 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4313 		} else if (r == RESUME_PASSTHROUGH) {
4314 			if (WARN_ON(xics_on_xive()))
4315 				r = H_SUCCESS;
4316 			else
4317 				r = kvmppc_xics_rm_complete(vcpu, 0);
4318 		}
4319 	} while (is_kvmppc_resume_guest(r));
4320 
4321 	/* Restore userspace EBB and other register values */
4322 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4323 		mtspr(SPRN_EBBHR, ebb_regs[0]);
4324 		mtspr(SPRN_EBBRR, ebb_regs[1]);
4325 		mtspr(SPRN_BESCR, ebb_regs[2]);
4326 		mtspr(SPRN_TAR, user_tar);
4327 		mtspr(SPRN_FSCR, current->thread.fscr);
4328 	}
4329 	mtspr(SPRN_VRSAVE, user_vrsave);
4330 
4331 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4332 	atomic_dec(&kvm->arch.vcpus_running);
4333 	return r;
4334 }
4335 
4336 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4337 				     int shift, int sllp)
4338 {
4339 	(*sps)->page_shift = shift;
4340 	(*sps)->slb_enc = sllp;
4341 	(*sps)->enc[0].page_shift = shift;
4342 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4343 	/*
4344 	 * Add 16MB MPSS support (may get filtered out by userspace)
4345 	 */
4346 	if (shift != 24) {
4347 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4348 		if (penc != -1) {
4349 			(*sps)->enc[1].page_shift = 24;
4350 			(*sps)->enc[1].pte_enc = penc;
4351 		}
4352 	}
4353 	(*sps)++;
4354 }
4355 
4356 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4357 					 struct kvm_ppc_smmu_info *info)
4358 {
4359 	struct kvm_ppc_one_seg_page_size *sps;
4360 
4361 	/*
4362 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4363 	 * POWER7 doesn't support keys for instruction accesses,
4364 	 * POWER8 and POWER9 do.
4365 	 */
4366 	info->data_keys = 32;
4367 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4368 
4369 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4370 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4371 	info->slb_size = 32;
4372 
4373 	/* We only support these sizes for now, and no muti-size segments */
4374 	sps = &info->sps[0];
4375 	kvmppc_add_seg_page_size(&sps, 12, 0);
4376 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4377 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4378 
4379 	/* If running as a nested hypervisor, we don't support HPT guests */
4380 	if (kvmhv_on_pseries())
4381 		info->flags |= KVM_PPC_NO_HASH;
4382 
4383 	return 0;
4384 }
4385 
4386 /*
4387  * Get (and clear) the dirty memory log for a memory slot.
4388  */
4389 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4390 					 struct kvm_dirty_log *log)
4391 {
4392 	struct kvm_memslots *slots;
4393 	struct kvm_memory_slot *memslot;
4394 	int i, r;
4395 	unsigned long n;
4396 	unsigned long *buf, *p;
4397 	struct kvm_vcpu *vcpu;
4398 
4399 	mutex_lock(&kvm->slots_lock);
4400 
4401 	r = -EINVAL;
4402 	if (log->slot >= KVM_USER_MEM_SLOTS)
4403 		goto out;
4404 
4405 	slots = kvm_memslots(kvm);
4406 	memslot = id_to_memslot(slots, log->slot);
4407 	r = -ENOENT;
4408 	if (!memslot || !memslot->dirty_bitmap)
4409 		goto out;
4410 
4411 	/*
4412 	 * Use second half of bitmap area because both HPT and radix
4413 	 * accumulate bits in the first half.
4414 	 */
4415 	n = kvm_dirty_bitmap_bytes(memslot);
4416 	buf = memslot->dirty_bitmap + n / sizeof(long);
4417 	memset(buf, 0, n);
4418 
4419 	if (kvm_is_radix(kvm))
4420 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4421 	else
4422 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4423 	if (r)
4424 		goto out;
4425 
4426 	/*
4427 	 * We accumulate dirty bits in the first half of the
4428 	 * memslot's dirty_bitmap area, for when pages are paged
4429 	 * out or modified by the host directly.  Pick up these
4430 	 * bits and add them to the map.
4431 	 */
4432 	p = memslot->dirty_bitmap;
4433 	for (i = 0; i < n / sizeof(long); ++i)
4434 		buf[i] |= xchg(&p[i], 0);
4435 
4436 	/* Harvest dirty bits from VPA and DTL updates */
4437 	/* Note: we never modify the SLB shadow buffer areas */
4438 	kvm_for_each_vcpu(i, vcpu, kvm) {
4439 		spin_lock(&vcpu->arch.vpa_update_lock);
4440 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4441 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4442 		spin_unlock(&vcpu->arch.vpa_update_lock);
4443 	}
4444 
4445 	r = -EFAULT;
4446 	if (copy_to_user(log->dirty_bitmap, buf, n))
4447 		goto out;
4448 
4449 	r = 0;
4450 out:
4451 	mutex_unlock(&kvm->slots_lock);
4452 	return r;
4453 }
4454 
4455 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4456 {
4457 	vfree(slot->arch.rmap);
4458 	slot->arch.rmap = NULL;
4459 }
4460 
4461 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4462 					struct kvm_memory_slot *slot,
4463 					const struct kvm_userspace_memory_region *mem,
4464 					enum kvm_mr_change change)
4465 {
4466 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4467 
4468 	if (change == KVM_MR_CREATE) {
4469 		slot->arch.rmap = vzalloc(array_size(npages,
4470 					  sizeof(*slot->arch.rmap)));
4471 		if (!slot->arch.rmap)
4472 			return -ENOMEM;
4473 	}
4474 
4475 	return 0;
4476 }
4477 
4478 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4479 				const struct kvm_userspace_memory_region *mem,
4480 				const struct kvm_memory_slot *old,
4481 				const struct kvm_memory_slot *new,
4482 				enum kvm_mr_change change)
4483 {
4484 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4485 
4486 	/*
4487 	 * If we are making a new memslot, it might make
4488 	 * some address that was previously cached as emulated
4489 	 * MMIO be no longer emulated MMIO, so invalidate
4490 	 * all the caches of emulated MMIO translations.
4491 	 */
4492 	if (npages)
4493 		atomic64_inc(&kvm->arch.mmio_update);
4494 
4495 	/*
4496 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4497 	 * have already called kvm_arch_flush_shadow_memslot() to
4498 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
4499 	 * previous mappings.  So the only case to handle is
4500 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4501 	 * has been changed.
4502 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4503 	 * to get rid of any THP PTEs in the partition-scoped page tables
4504 	 * so we can track dirtiness at the page level; we flush when
4505 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4506 	 * using THP PTEs.
4507 	 */
4508 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4509 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4510 		kvmppc_radix_flush_memslot(kvm, old);
4511 	/*
4512 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4513 	 */
4514 	if (!kvm->arch.secure_guest)
4515 		return;
4516 
4517 	switch (change) {
4518 	case KVM_MR_CREATE:
4519 		if (kvmppc_uvmem_slot_init(kvm, new))
4520 			return;
4521 		uv_register_mem_slot(kvm->arch.lpid,
4522 				     new->base_gfn << PAGE_SHIFT,
4523 				     new->npages * PAGE_SIZE,
4524 				     0, new->id);
4525 		break;
4526 	case KVM_MR_DELETE:
4527 		uv_unregister_mem_slot(kvm->arch.lpid, old->id);
4528 		kvmppc_uvmem_slot_free(kvm, old);
4529 		break;
4530 	default:
4531 		/* TODO: Handle KVM_MR_MOVE */
4532 		break;
4533 	}
4534 }
4535 
4536 /*
4537  * Update LPCR values in kvm->arch and in vcores.
4538  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4539  * of kvm->arch.lpcr update).
4540  */
4541 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4542 {
4543 	long int i;
4544 	u32 cores_done = 0;
4545 
4546 	if ((kvm->arch.lpcr & mask) == lpcr)
4547 		return;
4548 
4549 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4550 
4551 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4552 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4553 		if (!vc)
4554 			continue;
4555 		spin_lock(&vc->lock);
4556 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4557 		spin_unlock(&vc->lock);
4558 		if (++cores_done >= kvm->arch.online_vcores)
4559 			break;
4560 	}
4561 }
4562 
4563 void kvmppc_setup_partition_table(struct kvm *kvm)
4564 {
4565 	unsigned long dw0, dw1;
4566 
4567 	if (!kvm_is_radix(kvm)) {
4568 		/* PS field - page size for VRMA */
4569 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4570 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4571 		/* HTABSIZE and HTABORG fields */
4572 		dw0 |= kvm->arch.sdr1;
4573 
4574 		/* Second dword as set by userspace */
4575 		dw1 = kvm->arch.process_table;
4576 	} else {
4577 		dw0 = PATB_HR | radix__get_tree_size() |
4578 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4579 		dw1 = PATB_GR | kvm->arch.process_table;
4580 	}
4581 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4582 }
4583 
4584 /*
4585  * Set up HPT (hashed page table) and RMA (real-mode area).
4586  * Must be called with kvm->arch.mmu_setup_lock held.
4587  */
4588 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4589 {
4590 	int err = 0;
4591 	struct kvm *kvm = vcpu->kvm;
4592 	unsigned long hva;
4593 	struct kvm_memory_slot *memslot;
4594 	struct vm_area_struct *vma;
4595 	unsigned long lpcr = 0, senc;
4596 	unsigned long psize, porder;
4597 	int srcu_idx;
4598 
4599 	/* Allocate hashed page table (if not done already) and reset it */
4600 	if (!kvm->arch.hpt.virt) {
4601 		int order = KVM_DEFAULT_HPT_ORDER;
4602 		struct kvm_hpt_info info;
4603 
4604 		err = kvmppc_allocate_hpt(&info, order);
4605 		/* If we get here, it means userspace didn't specify a
4606 		 * size explicitly.  So, try successively smaller
4607 		 * sizes if the default failed. */
4608 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4609 			err  = kvmppc_allocate_hpt(&info, order);
4610 
4611 		if (err < 0) {
4612 			pr_err("KVM: Couldn't alloc HPT\n");
4613 			goto out;
4614 		}
4615 
4616 		kvmppc_set_hpt(kvm, &info);
4617 	}
4618 
4619 	/* Look up the memslot for guest physical address 0 */
4620 	srcu_idx = srcu_read_lock(&kvm->srcu);
4621 	memslot = gfn_to_memslot(kvm, 0);
4622 
4623 	/* We must have some memory at 0 by now */
4624 	err = -EINVAL;
4625 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4626 		goto out_srcu;
4627 
4628 	/* Look up the VMA for the start of this memory slot */
4629 	hva = memslot->userspace_addr;
4630 	down_read(&kvm->mm->mmap_sem);
4631 	vma = find_vma(kvm->mm, hva);
4632 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4633 		goto up_out;
4634 
4635 	psize = vma_kernel_pagesize(vma);
4636 
4637 	up_read(&kvm->mm->mmap_sem);
4638 
4639 	/* We can handle 4k, 64k or 16M pages in the VRMA */
4640 	if (psize >= 0x1000000)
4641 		psize = 0x1000000;
4642 	else if (psize >= 0x10000)
4643 		psize = 0x10000;
4644 	else
4645 		psize = 0x1000;
4646 	porder = __ilog2(psize);
4647 
4648 	senc = slb_pgsize_encoding(psize);
4649 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4650 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4651 	/* Create HPTEs in the hash page table for the VRMA */
4652 	kvmppc_map_vrma(vcpu, memslot, porder);
4653 
4654 	/* Update VRMASD field in the LPCR */
4655 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4656 		/* the -4 is to account for senc values starting at 0x10 */
4657 		lpcr = senc << (LPCR_VRMASD_SH - 4);
4658 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4659 	}
4660 
4661 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4662 	smp_wmb();
4663 	err = 0;
4664  out_srcu:
4665 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4666  out:
4667 	return err;
4668 
4669  up_out:
4670 	up_read(&kvm->mm->mmap_sem);
4671 	goto out_srcu;
4672 }
4673 
4674 /*
4675  * Must be called with kvm->arch.mmu_setup_lock held and
4676  * mmu_ready = 0 and no vcpus running.
4677  */
4678 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4679 {
4680 	if (nesting_enabled(kvm))
4681 		kvmhv_release_all_nested(kvm);
4682 	kvmppc_rmap_reset(kvm);
4683 	kvm->arch.process_table = 0;
4684 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4685 	spin_lock(&kvm->mmu_lock);
4686 	kvm->arch.radix = 0;
4687 	spin_unlock(&kvm->mmu_lock);
4688 	kvmppc_free_radix(kvm);
4689 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
4690 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4691 	return 0;
4692 }
4693 
4694 /*
4695  * Must be called with kvm->arch.mmu_setup_lock held and
4696  * mmu_ready = 0 and no vcpus running.
4697  */
4698 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4699 {
4700 	int err;
4701 
4702 	err = kvmppc_init_vm_radix(kvm);
4703 	if (err)
4704 		return err;
4705 	kvmppc_rmap_reset(kvm);
4706 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4707 	spin_lock(&kvm->mmu_lock);
4708 	kvm->arch.radix = 1;
4709 	spin_unlock(&kvm->mmu_lock);
4710 	kvmppc_free_hpt(&kvm->arch.hpt);
4711 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4712 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4713 	return 0;
4714 }
4715 
4716 #ifdef CONFIG_KVM_XICS
4717 /*
4718  * Allocate a per-core structure for managing state about which cores are
4719  * running in the host versus the guest and for exchanging data between
4720  * real mode KVM and CPU running in the host.
4721  * This is only done for the first VM.
4722  * The allocated structure stays even if all VMs have stopped.
4723  * It is only freed when the kvm-hv module is unloaded.
4724  * It's OK for this routine to fail, we just don't support host
4725  * core operations like redirecting H_IPI wakeups.
4726  */
4727 void kvmppc_alloc_host_rm_ops(void)
4728 {
4729 	struct kvmppc_host_rm_ops *ops;
4730 	unsigned long l_ops;
4731 	int cpu, core;
4732 	int size;
4733 
4734 	/* Not the first time here ? */
4735 	if (kvmppc_host_rm_ops_hv != NULL)
4736 		return;
4737 
4738 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4739 	if (!ops)
4740 		return;
4741 
4742 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4743 	ops->rm_core = kzalloc(size, GFP_KERNEL);
4744 
4745 	if (!ops->rm_core) {
4746 		kfree(ops);
4747 		return;
4748 	}
4749 
4750 	cpus_read_lock();
4751 
4752 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4753 		if (!cpu_online(cpu))
4754 			continue;
4755 
4756 		core = cpu >> threads_shift;
4757 		ops->rm_core[core].rm_state.in_host = 1;
4758 	}
4759 
4760 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4761 
4762 	/*
4763 	 * Make the contents of the kvmppc_host_rm_ops structure visible
4764 	 * to other CPUs before we assign it to the global variable.
4765 	 * Do an atomic assignment (no locks used here), but if someone
4766 	 * beats us to it, just free our copy and return.
4767 	 */
4768 	smp_wmb();
4769 	l_ops = (unsigned long) ops;
4770 
4771 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4772 		cpus_read_unlock();
4773 		kfree(ops->rm_core);
4774 		kfree(ops);
4775 		return;
4776 	}
4777 
4778 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4779 					     "ppc/kvm_book3s:prepare",
4780 					     kvmppc_set_host_core,
4781 					     kvmppc_clear_host_core);
4782 	cpus_read_unlock();
4783 }
4784 
4785 void kvmppc_free_host_rm_ops(void)
4786 {
4787 	if (kvmppc_host_rm_ops_hv) {
4788 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4789 		kfree(kvmppc_host_rm_ops_hv->rm_core);
4790 		kfree(kvmppc_host_rm_ops_hv);
4791 		kvmppc_host_rm_ops_hv = NULL;
4792 	}
4793 }
4794 #endif
4795 
4796 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4797 {
4798 	unsigned long lpcr, lpid;
4799 	char buf[32];
4800 	int ret;
4801 
4802 	mutex_init(&kvm->arch.uvmem_lock);
4803 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4804 	mutex_init(&kvm->arch.mmu_setup_lock);
4805 
4806 	/* Allocate the guest's logical partition ID */
4807 
4808 	lpid = kvmppc_alloc_lpid();
4809 	if ((long)lpid < 0)
4810 		return -ENOMEM;
4811 	kvm->arch.lpid = lpid;
4812 
4813 	kvmppc_alloc_host_rm_ops();
4814 
4815 	kvmhv_vm_nested_init(kvm);
4816 
4817 	/*
4818 	 * Since we don't flush the TLB when tearing down a VM,
4819 	 * and this lpid might have previously been used,
4820 	 * make sure we flush on each core before running the new VM.
4821 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4822 	 * does this flush for us.
4823 	 */
4824 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4825 		cpumask_setall(&kvm->arch.need_tlb_flush);
4826 
4827 	/* Start out with the default set of hcalls enabled */
4828 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4829 	       sizeof(kvm->arch.enabled_hcalls));
4830 
4831 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4832 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4833 
4834 	/* Init LPCR for virtual RMA mode */
4835 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4836 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
4837 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4838 		lpcr &= LPCR_PECE | LPCR_LPES;
4839 	} else {
4840 		lpcr = 0;
4841 	}
4842 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4843 		LPCR_VPM0 | LPCR_VPM1;
4844 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4845 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4846 	/* On POWER8 turn on online bit to enable PURR/SPURR */
4847 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
4848 		lpcr |= LPCR_ONL;
4849 	/*
4850 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4851 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
4852 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4853 	 * be unnecessary but better safe than sorry in case we re-enable
4854 	 * EE in HV mode with this LPCR still set)
4855 	 */
4856 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4857 		lpcr &= ~LPCR_VPM0;
4858 		lpcr |= LPCR_HVICE | LPCR_HEIC;
4859 
4860 		/*
4861 		 * If xive is enabled, we route 0x500 interrupts directly
4862 		 * to the guest.
4863 		 */
4864 		if (xics_on_xive())
4865 			lpcr |= LPCR_LPES;
4866 	}
4867 
4868 	/*
4869 	 * If the host uses radix, the guest starts out as radix.
4870 	 */
4871 	if (radix_enabled()) {
4872 		kvm->arch.radix = 1;
4873 		kvm->arch.mmu_ready = 1;
4874 		lpcr &= ~LPCR_VPM1;
4875 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4876 		ret = kvmppc_init_vm_radix(kvm);
4877 		if (ret) {
4878 			kvmppc_free_lpid(kvm->arch.lpid);
4879 			return ret;
4880 		}
4881 		kvmppc_setup_partition_table(kvm);
4882 	}
4883 
4884 	kvm->arch.lpcr = lpcr;
4885 
4886 	/* Initialization for future HPT resizes */
4887 	kvm->arch.resize_hpt = NULL;
4888 
4889 	/*
4890 	 * Work out how many sets the TLB has, for the use of
4891 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4892 	 */
4893 	if (radix_enabled())
4894 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
4895 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
4896 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
4897 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4898 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
4899 	else
4900 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
4901 
4902 	/*
4903 	 * Track that we now have a HV mode VM active. This blocks secondary
4904 	 * CPU threads from coming online.
4905 	 * On POWER9, we only need to do this if the "indep_threads_mode"
4906 	 * module parameter has been set to N.
4907 	 */
4908 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4909 		if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4910 			pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4911 			kvm->arch.threads_indep = true;
4912 		} else {
4913 			kvm->arch.threads_indep = indep_threads_mode;
4914 		}
4915 	}
4916 	if (!kvm->arch.threads_indep)
4917 		kvm_hv_vm_activated();
4918 
4919 	/*
4920 	 * Initialize smt_mode depending on processor.
4921 	 * POWER8 and earlier have to use "strict" threading, where
4922 	 * all vCPUs in a vcore have to run on the same (sub)core,
4923 	 * whereas on POWER9 the threads can each run a different
4924 	 * guest.
4925 	 */
4926 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4927 		kvm->arch.smt_mode = threads_per_subcore;
4928 	else
4929 		kvm->arch.smt_mode = 1;
4930 	kvm->arch.emul_smt_mode = 1;
4931 
4932 	/*
4933 	 * Create a debugfs directory for the VM
4934 	 */
4935 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
4936 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4937 	kvmppc_mmu_debugfs_init(kvm);
4938 	if (radix_enabled())
4939 		kvmhv_radix_debugfs_init(kvm);
4940 
4941 	return 0;
4942 }
4943 
4944 static void kvmppc_free_vcores(struct kvm *kvm)
4945 {
4946 	long int i;
4947 
4948 	for (i = 0; i < KVM_MAX_VCORES; ++i)
4949 		kfree(kvm->arch.vcores[i]);
4950 	kvm->arch.online_vcores = 0;
4951 }
4952 
4953 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4954 {
4955 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
4956 
4957 	if (!kvm->arch.threads_indep)
4958 		kvm_hv_vm_deactivated();
4959 
4960 	kvmppc_free_vcores(kvm);
4961 
4962 
4963 	if (kvm_is_radix(kvm))
4964 		kvmppc_free_radix(kvm);
4965 	else
4966 		kvmppc_free_hpt(&kvm->arch.hpt);
4967 
4968 	/* Perform global invalidation and return lpid to the pool */
4969 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4970 		if (nesting_enabled(kvm))
4971 			kvmhv_release_all_nested(kvm);
4972 		kvm->arch.process_table = 0;
4973 		if (kvm->arch.secure_guest)
4974 			uv_svm_terminate(kvm->arch.lpid);
4975 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4976 	}
4977 
4978 	kvmppc_free_lpid(kvm->arch.lpid);
4979 
4980 	kvmppc_free_pimap(kvm);
4981 }
4982 
4983 /* We don't need to emulate any privileged instructions or dcbz */
4984 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4985 				     unsigned int inst, int *advance)
4986 {
4987 	return EMULATE_FAIL;
4988 }
4989 
4990 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4991 					ulong spr_val)
4992 {
4993 	return EMULATE_FAIL;
4994 }
4995 
4996 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4997 					ulong *spr_val)
4998 {
4999 	return EMULATE_FAIL;
5000 }
5001 
5002 static int kvmppc_core_check_processor_compat_hv(void)
5003 {
5004 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5005 	    cpu_has_feature(CPU_FTR_ARCH_206))
5006 		return 0;
5007 
5008 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5009 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5010 		return 0;
5011 
5012 	return -EIO;
5013 }
5014 
5015 #ifdef CONFIG_KVM_XICS
5016 
5017 void kvmppc_free_pimap(struct kvm *kvm)
5018 {
5019 	kfree(kvm->arch.pimap);
5020 }
5021 
5022 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5023 {
5024 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5025 }
5026 
5027 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5028 {
5029 	struct irq_desc *desc;
5030 	struct kvmppc_irq_map *irq_map;
5031 	struct kvmppc_passthru_irqmap *pimap;
5032 	struct irq_chip *chip;
5033 	int i, rc = 0;
5034 
5035 	if (!kvm_irq_bypass)
5036 		return 1;
5037 
5038 	desc = irq_to_desc(host_irq);
5039 	if (!desc)
5040 		return -EIO;
5041 
5042 	mutex_lock(&kvm->lock);
5043 
5044 	pimap = kvm->arch.pimap;
5045 	if (pimap == NULL) {
5046 		/* First call, allocate structure to hold IRQ map */
5047 		pimap = kvmppc_alloc_pimap();
5048 		if (pimap == NULL) {
5049 			mutex_unlock(&kvm->lock);
5050 			return -ENOMEM;
5051 		}
5052 		kvm->arch.pimap = pimap;
5053 	}
5054 
5055 	/*
5056 	 * For now, we only support interrupts for which the EOI operation
5057 	 * is an OPAL call followed by a write to XIRR, since that's
5058 	 * what our real-mode EOI code does, or a XIVE interrupt
5059 	 */
5060 	chip = irq_data_get_irq_chip(&desc->irq_data);
5061 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5062 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5063 			host_irq, guest_gsi);
5064 		mutex_unlock(&kvm->lock);
5065 		return -ENOENT;
5066 	}
5067 
5068 	/*
5069 	 * See if we already have an entry for this guest IRQ number.
5070 	 * If it's mapped to a hardware IRQ number, that's an error,
5071 	 * otherwise re-use this entry.
5072 	 */
5073 	for (i = 0; i < pimap->n_mapped; i++) {
5074 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5075 			if (pimap->mapped[i].r_hwirq) {
5076 				mutex_unlock(&kvm->lock);
5077 				return -EINVAL;
5078 			}
5079 			break;
5080 		}
5081 	}
5082 
5083 	if (i == KVMPPC_PIRQ_MAPPED) {
5084 		mutex_unlock(&kvm->lock);
5085 		return -EAGAIN;		/* table is full */
5086 	}
5087 
5088 	irq_map = &pimap->mapped[i];
5089 
5090 	irq_map->v_hwirq = guest_gsi;
5091 	irq_map->desc = desc;
5092 
5093 	/*
5094 	 * Order the above two stores before the next to serialize with
5095 	 * the KVM real mode handler.
5096 	 */
5097 	smp_wmb();
5098 	irq_map->r_hwirq = desc->irq_data.hwirq;
5099 
5100 	if (i == pimap->n_mapped)
5101 		pimap->n_mapped++;
5102 
5103 	if (xics_on_xive())
5104 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5105 	else
5106 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5107 	if (rc)
5108 		irq_map->r_hwirq = 0;
5109 
5110 	mutex_unlock(&kvm->lock);
5111 
5112 	return 0;
5113 }
5114 
5115 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5116 {
5117 	struct irq_desc *desc;
5118 	struct kvmppc_passthru_irqmap *pimap;
5119 	int i, rc = 0;
5120 
5121 	if (!kvm_irq_bypass)
5122 		return 0;
5123 
5124 	desc = irq_to_desc(host_irq);
5125 	if (!desc)
5126 		return -EIO;
5127 
5128 	mutex_lock(&kvm->lock);
5129 	if (!kvm->arch.pimap)
5130 		goto unlock;
5131 
5132 	pimap = kvm->arch.pimap;
5133 
5134 	for (i = 0; i < pimap->n_mapped; i++) {
5135 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5136 			break;
5137 	}
5138 
5139 	if (i == pimap->n_mapped) {
5140 		mutex_unlock(&kvm->lock);
5141 		return -ENODEV;
5142 	}
5143 
5144 	if (xics_on_xive())
5145 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5146 	else
5147 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5148 
5149 	/* invalidate the entry (what do do on error from the above ?) */
5150 	pimap->mapped[i].r_hwirq = 0;
5151 
5152 	/*
5153 	 * We don't free this structure even when the count goes to
5154 	 * zero. The structure is freed when we destroy the VM.
5155 	 */
5156  unlock:
5157 	mutex_unlock(&kvm->lock);
5158 	return rc;
5159 }
5160 
5161 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5162 					     struct irq_bypass_producer *prod)
5163 {
5164 	int ret = 0;
5165 	struct kvm_kernel_irqfd *irqfd =
5166 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5167 
5168 	irqfd->producer = prod;
5169 
5170 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5171 	if (ret)
5172 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5173 			prod->irq, irqfd->gsi, ret);
5174 
5175 	return ret;
5176 }
5177 
5178 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5179 					      struct irq_bypass_producer *prod)
5180 {
5181 	int ret;
5182 	struct kvm_kernel_irqfd *irqfd =
5183 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5184 
5185 	irqfd->producer = NULL;
5186 
5187 	/*
5188 	 * When producer of consumer is unregistered, we change back to
5189 	 * default external interrupt handling mode - KVM real mode
5190 	 * will switch back to host.
5191 	 */
5192 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5193 	if (ret)
5194 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5195 			prod->irq, irqfd->gsi, ret);
5196 }
5197 #endif
5198 
5199 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5200 				 unsigned int ioctl, unsigned long arg)
5201 {
5202 	struct kvm *kvm __maybe_unused = filp->private_data;
5203 	void __user *argp = (void __user *)arg;
5204 	long r;
5205 
5206 	switch (ioctl) {
5207 
5208 	case KVM_PPC_ALLOCATE_HTAB: {
5209 		u32 htab_order;
5210 
5211 		r = -EFAULT;
5212 		if (get_user(htab_order, (u32 __user *)argp))
5213 			break;
5214 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5215 		if (r)
5216 			break;
5217 		r = 0;
5218 		break;
5219 	}
5220 
5221 	case KVM_PPC_GET_HTAB_FD: {
5222 		struct kvm_get_htab_fd ghf;
5223 
5224 		r = -EFAULT;
5225 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5226 			break;
5227 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5228 		break;
5229 	}
5230 
5231 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5232 		struct kvm_ppc_resize_hpt rhpt;
5233 
5234 		r = -EFAULT;
5235 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5236 			break;
5237 
5238 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5239 		break;
5240 	}
5241 
5242 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5243 		struct kvm_ppc_resize_hpt rhpt;
5244 
5245 		r = -EFAULT;
5246 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5247 			break;
5248 
5249 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5250 		break;
5251 	}
5252 
5253 	default:
5254 		r = -ENOTTY;
5255 	}
5256 
5257 	return r;
5258 }
5259 
5260 /*
5261  * List of hcall numbers to enable by default.
5262  * For compatibility with old userspace, we enable by default
5263  * all hcalls that were implemented before the hcall-enabling
5264  * facility was added.  Note this list should not include H_RTAS.
5265  */
5266 static unsigned int default_hcall_list[] = {
5267 	H_REMOVE,
5268 	H_ENTER,
5269 	H_READ,
5270 	H_PROTECT,
5271 	H_BULK_REMOVE,
5272 	H_GET_TCE,
5273 	H_PUT_TCE,
5274 	H_SET_DABR,
5275 	H_SET_XDABR,
5276 	H_CEDE,
5277 	H_PROD,
5278 	H_CONFER,
5279 	H_REGISTER_VPA,
5280 #ifdef CONFIG_KVM_XICS
5281 	H_EOI,
5282 	H_CPPR,
5283 	H_IPI,
5284 	H_IPOLL,
5285 	H_XIRR,
5286 	H_XIRR_X,
5287 #endif
5288 	0
5289 };
5290 
5291 static void init_default_hcalls(void)
5292 {
5293 	int i;
5294 	unsigned int hcall;
5295 
5296 	for (i = 0; default_hcall_list[i]; ++i) {
5297 		hcall = default_hcall_list[i];
5298 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5299 		__set_bit(hcall / 4, default_enabled_hcalls);
5300 	}
5301 }
5302 
5303 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5304 {
5305 	unsigned long lpcr;
5306 	int radix;
5307 	int err;
5308 
5309 	/* If not on a POWER9, reject it */
5310 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5311 		return -ENODEV;
5312 
5313 	/* If any unknown flags set, reject it */
5314 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5315 		return -EINVAL;
5316 
5317 	/* GR (guest radix) bit in process_table field must match */
5318 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5319 	if (!!(cfg->process_table & PATB_GR) != radix)
5320 		return -EINVAL;
5321 
5322 	/* Process table size field must be reasonable, i.e. <= 24 */
5323 	if ((cfg->process_table & PRTS_MASK) > 24)
5324 		return -EINVAL;
5325 
5326 	/* We can change a guest to/from radix now, if the host is radix */
5327 	if (radix && !radix_enabled())
5328 		return -EINVAL;
5329 
5330 	/* If we're a nested hypervisor, we currently only support radix */
5331 	if (kvmhv_on_pseries() && !radix)
5332 		return -EINVAL;
5333 
5334 	mutex_lock(&kvm->arch.mmu_setup_lock);
5335 	if (radix != kvm_is_radix(kvm)) {
5336 		if (kvm->arch.mmu_ready) {
5337 			kvm->arch.mmu_ready = 0;
5338 			/* order mmu_ready vs. vcpus_running */
5339 			smp_mb();
5340 			if (atomic_read(&kvm->arch.vcpus_running)) {
5341 				kvm->arch.mmu_ready = 1;
5342 				err = -EBUSY;
5343 				goto out_unlock;
5344 			}
5345 		}
5346 		if (radix)
5347 			err = kvmppc_switch_mmu_to_radix(kvm);
5348 		else
5349 			err = kvmppc_switch_mmu_to_hpt(kvm);
5350 		if (err)
5351 			goto out_unlock;
5352 	}
5353 
5354 	kvm->arch.process_table = cfg->process_table;
5355 	kvmppc_setup_partition_table(kvm);
5356 
5357 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5358 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5359 	err = 0;
5360 
5361  out_unlock:
5362 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5363 	return err;
5364 }
5365 
5366 static int kvmhv_enable_nested(struct kvm *kvm)
5367 {
5368 	if (!nested)
5369 		return -EPERM;
5370 	if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5371 		return -ENODEV;
5372 
5373 	/* kvm == NULL means the caller is testing if the capability exists */
5374 	if (kvm)
5375 		kvm->arch.nested_enable = true;
5376 	return 0;
5377 }
5378 
5379 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5380 				 int size)
5381 {
5382 	int rc = -EINVAL;
5383 
5384 	if (kvmhv_vcpu_is_radix(vcpu)) {
5385 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5386 
5387 		if (rc > 0)
5388 			rc = -EINVAL;
5389 	}
5390 
5391 	/* For now quadrants are the only way to access nested guest memory */
5392 	if (rc && vcpu->arch.nested)
5393 		rc = -EAGAIN;
5394 
5395 	return rc;
5396 }
5397 
5398 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5399 				int size)
5400 {
5401 	int rc = -EINVAL;
5402 
5403 	if (kvmhv_vcpu_is_radix(vcpu)) {
5404 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5405 
5406 		if (rc > 0)
5407 			rc = -EINVAL;
5408 	}
5409 
5410 	/* For now quadrants are the only way to access nested guest memory */
5411 	if (rc && vcpu->arch.nested)
5412 		rc = -EAGAIN;
5413 
5414 	return rc;
5415 }
5416 
5417 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5418 {
5419 	unpin_vpa(kvm, vpa);
5420 	vpa->gpa = 0;
5421 	vpa->pinned_addr = NULL;
5422 	vpa->dirty = false;
5423 	vpa->update_pending = 0;
5424 }
5425 
5426 /*
5427  * Enable a guest to become a secure VM, or test whether
5428  * that could be enabled.
5429  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5430  * tested (kvm == NULL) or enabled (kvm != NULL).
5431  */
5432 static int kvmhv_enable_svm(struct kvm *kvm)
5433 {
5434 	if (!kvmppc_uvmem_available())
5435 		return -EINVAL;
5436 	if (kvm)
5437 		kvm->arch.svm_enabled = 1;
5438 	return 0;
5439 }
5440 
5441 /*
5442  *  IOCTL handler to turn off secure mode of guest
5443  *
5444  * - Release all device pages
5445  * - Issue ucall to terminate the guest on the UV side
5446  * - Unpin the VPA pages.
5447  * - Reinit the partition scoped page tables
5448  */
5449 static int kvmhv_svm_off(struct kvm *kvm)
5450 {
5451 	struct kvm_vcpu *vcpu;
5452 	int mmu_was_ready;
5453 	int srcu_idx;
5454 	int ret = 0;
5455 	int i;
5456 
5457 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5458 		return ret;
5459 
5460 	mutex_lock(&kvm->arch.mmu_setup_lock);
5461 	mmu_was_ready = kvm->arch.mmu_ready;
5462 	if (kvm->arch.mmu_ready) {
5463 		kvm->arch.mmu_ready = 0;
5464 		/* order mmu_ready vs. vcpus_running */
5465 		smp_mb();
5466 		if (atomic_read(&kvm->arch.vcpus_running)) {
5467 			kvm->arch.mmu_ready = 1;
5468 			ret = -EBUSY;
5469 			goto out;
5470 		}
5471 	}
5472 
5473 	srcu_idx = srcu_read_lock(&kvm->srcu);
5474 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5475 		struct kvm_memory_slot *memslot;
5476 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5477 
5478 		if (!slots)
5479 			continue;
5480 
5481 		kvm_for_each_memslot(memslot, slots) {
5482 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
5483 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5484 		}
5485 	}
5486 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5487 
5488 	ret = uv_svm_terminate(kvm->arch.lpid);
5489 	if (ret != U_SUCCESS) {
5490 		ret = -EINVAL;
5491 		goto out;
5492 	}
5493 
5494 	/*
5495 	 * When secure guest is reset, all the guest pages are sent
5496 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
5497 	 * chance to run and unpin their VPA pages. Unpinning of all
5498 	 * VPA pages is done here explicitly so that VPA pages
5499 	 * can be migrated to the secure side.
5500 	 *
5501 	 * This is required to for the secure SMP guest to reboot
5502 	 * correctly.
5503 	 */
5504 	kvm_for_each_vcpu(i, vcpu, kvm) {
5505 		spin_lock(&vcpu->arch.vpa_update_lock);
5506 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5507 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5508 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5509 		spin_unlock(&vcpu->arch.vpa_update_lock);
5510 	}
5511 
5512 	kvmppc_setup_partition_table(kvm);
5513 	kvm->arch.secure_guest = 0;
5514 	kvm->arch.mmu_ready = mmu_was_ready;
5515 out:
5516 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5517 	return ret;
5518 }
5519 
5520 static struct kvmppc_ops kvm_ops_hv = {
5521 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5522 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5523 	.get_one_reg = kvmppc_get_one_reg_hv,
5524 	.set_one_reg = kvmppc_set_one_reg_hv,
5525 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
5526 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
5527 	.inject_interrupt = kvmppc_inject_interrupt_hv,
5528 	.set_msr     = kvmppc_set_msr_hv,
5529 	.vcpu_run    = kvmppc_vcpu_run_hv,
5530 	.vcpu_create = kvmppc_core_vcpu_create_hv,
5531 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
5532 	.check_requests = kvmppc_core_check_requests_hv,
5533 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5534 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
5535 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5536 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5537 	.unmap_hva_range = kvm_unmap_hva_range_hv,
5538 	.age_hva  = kvm_age_hva_hv,
5539 	.test_age_hva = kvm_test_age_hva_hv,
5540 	.set_spte_hva = kvm_set_spte_hva_hv,
5541 	.free_memslot = kvmppc_core_free_memslot_hv,
5542 	.init_vm =  kvmppc_core_init_vm_hv,
5543 	.destroy_vm = kvmppc_core_destroy_vm_hv,
5544 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5545 	.emulate_op = kvmppc_core_emulate_op_hv,
5546 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5547 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5548 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5549 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5550 	.hcall_implemented = kvmppc_hcall_impl_hv,
5551 #ifdef CONFIG_KVM_XICS
5552 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5553 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5554 #endif
5555 	.configure_mmu = kvmhv_configure_mmu,
5556 	.get_rmmu_info = kvmhv_get_rmmu_info,
5557 	.set_smt_mode = kvmhv_set_smt_mode,
5558 	.enable_nested = kvmhv_enable_nested,
5559 	.load_from_eaddr = kvmhv_load_from_eaddr,
5560 	.store_to_eaddr = kvmhv_store_to_eaddr,
5561 	.enable_svm = kvmhv_enable_svm,
5562 	.svm_off = kvmhv_svm_off,
5563 };
5564 
5565 static int kvm_init_subcore_bitmap(void)
5566 {
5567 	int i, j;
5568 	int nr_cores = cpu_nr_cores();
5569 	struct sibling_subcore_state *sibling_subcore_state;
5570 
5571 	for (i = 0; i < nr_cores; i++) {
5572 		int first_cpu = i * threads_per_core;
5573 		int node = cpu_to_node(first_cpu);
5574 
5575 		/* Ignore if it is already allocated. */
5576 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
5577 			continue;
5578 
5579 		sibling_subcore_state =
5580 			kzalloc_node(sizeof(struct sibling_subcore_state),
5581 							GFP_KERNEL, node);
5582 		if (!sibling_subcore_state)
5583 			return -ENOMEM;
5584 
5585 
5586 		for (j = 0; j < threads_per_core; j++) {
5587 			int cpu = first_cpu + j;
5588 
5589 			paca_ptrs[cpu]->sibling_subcore_state =
5590 						sibling_subcore_state;
5591 		}
5592 	}
5593 	return 0;
5594 }
5595 
5596 static int kvmppc_radix_possible(void)
5597 {
5598 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5599 }
5600 
5601 static int kvmppc_book3s_init_hv(void)
5602 {
5603 	int r;
5604 
5605 	if (!tlbie_capable) {
5606 		pr_err("KVM-HV: Host does not support TLBIE\n");
5607 		return -ENODEV;
5608 	}
5609 
5610 	/*
5611 	 * FIXME!! Do we need to check on all cpus ?
5612 	 */
5613 	r = kvmppc_core_check_processor_compat_hv();
5614 	if (r < 0)
5615 		return -ENODEV;
5616 
5617 	r = kvmhv_nested_init();
5618 	if (r)
5619 		return r;
5620 
5621 	r = kvm_init_subcore_bitmap();
5622 	if (r)
5623 		return r;
5624 
5625 	/*
5626 	 * We need a way of accessing the XICS interrupt controller,
5627 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5628 	 * indirectly, via OPAL.
5629 	 */
5630 #ifdef CONFIG_SMP
5631 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
5632 	    !local_paca->kvm_hstate.xics_phys) {
5633 		struct device_node *np;
5634 
5635 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5636 		if (!np) {
5637 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5638 			return -ENODEV;
5639 		}
5640 		/* presence of intc confirmed - node can be dropped again */
5641 		of_node_put(np);
5642 	}
5643 #endif
5644 
5645 	kvm_ops_hv.owner = THIS_MODULE;
5646 	kvmppc_hv_ops = &kvm_ops_hv;
5647 
5648 	init_default_hcalls();
5649 
5650 	init_vcore_lists();
5651 
5652 	r = kvmppc_mmu_hv_init();
5653 	if (r)
5654 		return r;
5655 
5656 	if (kvmppc_radix_possible())
5657 		r = kvmppc_radix_init();
5658 
5659 	/*
5660 	 * POWER9 chips before version 2.02 can't have some threads in
5661 	 * HPT mode and some in radix mode on the same core.
5662 	 */
5663 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5664 		unsigned int pvr = mfspr(SPRN_PVR);
5665 		if ((pvr >> 16) == PVR_POWER9 &&
5666 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5667 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5668 			no_mixing_hpt_and_radix = true;
5669 	}
5670 
5671 	r = kvmppc_uvmem_init();
5672 	if (r < 0)
5673 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5674 
5675 	return r;
5676 }
5677 
5678 static void kvmppc_book3s_exit_hv(void)
5679 {
5680 	kvmppc_uvmem_free();
5681 	kvmppc_free_host_rm_ops();
5682 	if (kvmppc_radix_possible())
5683 		kvmppc_radix_exit();
5684 	kvmppc_hv_ops = NULL;
5685 	kvmhv_nested_exit();
5686 }
5687 
5688 module_init(kvmppc_book3s_init_hv);
5689 module_exit(kvmppc_book3s_exit_hv);
5690 MODULE_LICENSE("GPL");
5691 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5692 MODULE_ALIAS("devname:kvm");
5693