xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 2d96b44f)
1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Paul Mackerras <paulus@au1.ibm.com>
7  *    Alexander Graf <agraf@suse.de>
8  *    Kevin Wolf <mail@kevin-wolf.de>
9  *
10  * Description: KVM functions specific to running on Book 3S
11  * processors in hypervisor mode (specifically POWER7 and later).
12  *
13  * This file is derived from arch/powerpc/kvm/book3s.c,
14  * by Alexander Graf <agraf@suse.de>.
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License, version 2, as
18  * published by the Free Software Foundation.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/export.h>
28 #include <linux/fs.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpu.h>
31 #include <linux/cpumask.h>
32 #include <linux/spinlock.h>
33 #include <linux/page-flags.h>
34 #include <linux/srcu.h>
35 #include <linux/miscdevice.h>
36 #include <linux/debugfs.h>
37 
38 #include <asm/reg.h>
39 #include <asm/cputable.h>
40 #include <asm/cacheflush.h>
41 #include <asm/tlbflush.h>
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/kvm_ppc.h>
45 #include <asm/kvm_book3s.h>
46 #include <asm/mmu_context.h>
47 #include <asm/lppaca.h>
48 #include <asm/processor.h>
49 #include <asm/cputhreads.h>
50 #include <asm/page.h>
51 #include <asm/hvcall.h>
52 #include <asm/switch_to.h>
53 #include <asm/smp.h>
54 #include <asm/dbell.h>
55 #include <asm/hmi.h>
56 #include <asm/pnv-pci.h>
57 #include <linux/gfp.h>
58 #include <linux/vmalloc.h>
59 #include <linux/highmem.h>
60 #include <linux/hugetlb.h>
61 #include <linux/kvm_irqfd.h>
62 #include <linux/irqbypass.h>
63 #include <linux/module.h>
64 #include <linux/compiler.h>
65 
66 #include "book3s.h"
67 
68 #define CREATE_TRACE_POINTS
69 #include "trace_hv.h"
70 
71 /* #define EXIT_DEBUG */
72 /* #define EXIT_DEBUG_SIMPLE */
73 /* #define EXIT_DEBUG_INT */
74 
75 /* Used to indicate that a guest page fault needs to be handled */
76 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
77 /* Used to indicate that a guest passthrough interrupt needs to be handled */
78 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
79 
80 /* Used as a "null" value for timebase values */
81 #define TB_NIL	(~(u64)0)
82 
83 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
84 
85 static int dynamic_mt_modes = 6;
86 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
87 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
88 static int target_smt_mode;
89 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
90 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
91 
92 #ifdef CONFIG_KVM_XICS
93 static struct kernel_param_ops module_param_ops = {
94 	.set = param_set_int,
95 	.get = param_get_int,
96 };
97 
98 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
99 							S_IRUGO | S_IWUSR);
100 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
101 
102 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
103 							S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
105 #endif
106 
107 /* Maximum halt poll interval defaults to KVM_HALT_POLL_NS_DEFAULT */
108 static unsigned int halt_poll_max_ns = KVM_HALT_POLL_NS_DEFAULT;
109 module_param(halt_poll_max_ns, uint, S_IRUGO | S_IWUSR);
110 MODULE_PARM_DESC(halt_poll_max_ns, "Maximum halt poll time in ns");
111 
112 /* Factor by which the vcore halt poll interval is grown, default is to double
113  */
114 static unsigned int halt_poll_ns_grow = 2;
115 module_param(halt_poll_ns_grow, int, S_IRUGO);
116 MODULE_PARM_DESC(halt_poll_ns_grow, "Factor halt poll time is grown by");
117 
118 /* Factor by which the vcore halt poll interval is shrunk, default is to reset
119  */
120 static unsigned int halt_poll_ns_shrink;
121 module_param(halt_poll_ns_shrink, int, S_IRUGO);
122 MODULE_PARM_DESC(halt_poll_ns_shrink, "Factor halt poll time is shrunk by");
123 
124 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
125 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
126 
127 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
128 		int *ip)
129 {
130 	int i = *ip;
131 	struct kvm_vcpu *vcpu;
132 
133 	while (++i < MAX_SMT_THREADS) {
134 		vcpu = READ_ONCE(vc->runnable_threads[i]);
135 		if (vcpu) {
136 			*ip = i;
137 			return vcpu;
138 		}
139 	}
140 	return NULL;
141 }
142 
143 /* Used to traverse the list of runnable threads for a given vcore */
144 #define for_each_runnable_thread(i, vcpu, vc) \
145 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
146 
147 static bool kvmppc_ipi_thread(int cpu)
148 {
149 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
150 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
151 		preempt_disable();
152 		if (cpu_first_thread_sibling(cpu) ==
153 		    cpu_first_thread_sibling(smp_processor_id())) {
154 			unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
155 			msg |= cpu_thread_in_core(cpu);
156 			smp_mb();
157 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
158 			preempt_enable();
159 			return true;
160 		}
161 		preempt_enable();
162 	}
163 
164 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
165 	if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
166 		xics_wake_cpu(cpu);
167 		return true;
168 	}
169 #endif
170 
171 	return false;
172 }
173 
174 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
175 {
176 	int cpu;
177 	struct swait_queue_head *wqp;
178 
179 	wqp = kvm_arch_vcpu_wq(vcpu);
180 	if (swait_active(wqp)) {
181 		swake_up(wqp);
182 		++vcpu->stat.halt_wakeup;
183 	}
184 
185 	if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
186 		return;
187 
188 	/* CPU points to the first thread of the core */
189 	cpu = vcpu->cpu;
190 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
191 		smp_send_reschedule(cpu);
192 }
193 
194 /*
195  * We use the vcpu_load/put functions to measure stolen time.
196  * Stolen time is counted as time when either the vcpu is able to
197  * run as part of a virtual core, but the task running the vcore
198  * is preempted or sleeping, or when the vcpu needs something done
199  * in the kernel by the task running the vcpu, but that task is
200  * preempted or sleeping.  Those two things have to be counted
201  * separately, since one of the vcpu tasks will take on the job
202  * of running the core, and the other vcpu tasks in the vcore will
203  * sleep waiting for it to do that, but that sleep shouldn't count
204  * as stolen time.
205  *
206  * Hence we accumulate stolen time when the vcpu can run as part of
207  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
208  * needs its task to do other things in the kernel (for example,
209  * service a page fault) in busy_stolen.  We don't accumulate
210  * stolen time for a vcore when it is inactive, or for a vcpu
211  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
212  * a misnomer; it means that the vcpu task is not executing in
213  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
214  * the kernel.  We don't have any way of dividing up that time
215  * between time that the vcpu is genuinely stopped, time that
216  * the task is actively working on behalf of the vcpu, and time
217  * that the task is preempted, so we don't count any of it as
218  * stolen.
219  *
220  * Updates to busy_stolen are protected by arch.tbacct_lock;
221  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
222  * lock.  The stolen times are measured in units of timebase ticks.
223  * (Note that the != TB_NIL checks below are purely defensive;
224  * they should never fail.)
225  */
226 
227 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
228 {
229 	unsigned long flags;
230 
231 	spin_lock_irqsave(&vc->stoltb_lock, flags);
232 	vc->preempt_tb = mftb();
233 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
234 }
235 
236 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
237 {
238 	unsigned long flags;
239 
240 	spin_lock_irqsave(&vc->stoltb_lock, flags);
241 	if (vc->preempt_tb != TB_NIL) {
242 		vc->stolen_tb += mftb() - vc->preempt_tb;
243 		vc->preempt_tb = TB_NIL;
244 	}
245 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
246 }
247 
248 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
249 {
250 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
251 	unsigned long flags;
252 
253 	/*
254 	 * We can test vc->runner without taking the vcore lock,
255 	 * because only this task ever sets vc->runner to this
256 	 * vcpu, and once it is set to this vcpu, only this task
257 	 * ever sets it to NULL.
258 	 */
259 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
260 		kvmppc_core_end_stolen(vc);
261 
262 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
263 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
264 	    vcpu->arch.busy_preempt != TB_NIL) {
265 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
266 		vcpu->arch.busy_preempt = TB_NIL;
267 	}
268 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
269 }
270 
271 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
272 {
273 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
274 	unsigned long flags;
275 
276 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
277 		kvmppc_core_start_stolen(vc);
278 
279 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
280 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
281 		vcpu->arch.busy_preempt = mftb();
282 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
283 }
284 
285 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
286 {
287 	/*
288 	 * Check for illegal transactional state bit combination
289 	 * and if we find it, force the TS field to a safe state.
290 	 */
291 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
292 		msr &= ~MSR_TS_MASK;
293 	vcpu->arch.shregs.msr = msr;
294 	kvmppc_end_cede(vcpu);
295 }
296 
297 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
298 {
299 	vcpu->arch.pvr = pvr;
300 }
301 
302 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
303 {
304 	unsigned long pcr = 0;
305 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
306 
307 	if (arch_compat) {
308 		switch (arch_compat) {
309 		case PVR_ARCH_205:
310 			/*
311 			 * If an arch bit is set in PCR, all the defined
312 			 * higher-order arch bits also have to be set.
313 			 */
314 			pcr = PCR_ARCH_206 | PCR_ARCH_205;
315 			break;
316 		case PVR_ARCH_206:
317 		case PVR_ARCH_206p:
318 			pcr = PCR_ARCH_206;
319 			break;
320 		case PVR_ARCH_207:
321 			break;
322 		default:
323 			return -EINVAL;
324 		}
325 
326 		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
327 			/* POWER7 can't emulate POWER8 */
328 			if (!(pcr & PCR_ARCH_206))
329 				return -EINVAL;
330 			pcr &= ~PCR_ARCH_206;
331 		}
332 	}
333 
334 	spin_lock(&vc->lock);
335 	vc->arch_compat = arch_compat;
336 	vc->pcr = pcr;
337 	spin_unlock(&vc->lock);
338 
339 	return 0;
340 }
341 
342 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
343 {
344 	int r;
345 
346 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
347 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
348 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
349 	for (r = 0; r < 16; ++r)
350 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
351 		       r, kvmppc_get_gpr(vcpu, r),
352 		       r+16, kvmppc_get_gpr(vcpu, r+16));
353 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
354 	       vcpu->arch.ctr, vcpu->arch.lr);
355 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
356 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
357 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
358 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
359 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
360 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
361 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
362 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
363 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
364 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
365 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
366 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
367 	for (r = 0; r < vcpu->arch.slb_max; ++r)
368 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
369 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
370 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
371 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
372 	       vcpu->arch.last_inst);
373 }
374 
375 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
376 {
377 	struct kvm_vcpu *ret;
378 
379 	mutex_lock(&kvm->lock);
380 	ret = kvm_get_vcpu_by_id(kvm, id);
381 	mutex_unlock(&kvm->lock);
382 	return ret;
383 }
384 
385 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
386 {
387 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
388 	vpa->yield_count = cpu_to_be32(1);
389 }
390 
391 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
392 		   unsigned long addr, unsigned long len)
393 {
394 	/* check address is cacheline aligned */
395 	if (addr & (L1_CACHE_BYTES - 1))
396 		return -EINVAL;
397 	spin_lock(&vcpu->arch.vpa_update_lock);
398 	if (v->next_gpa != addr || v->len != len) {
399 		v->next_gpa = addr;
400 		v->len = addr ? len : 0;
401 		v->update_pending = 1;
402 	}
403 	spin_unlock(&vcpu->arch.vpa_update_lock);
404 	return 0;
405 }
406 
407 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
408 struct reg_vpa {
409 	u32 dummy;
410 	union {
411 		__be16 hword;
412 		__be32 word;
413 	} length;
414 };
415 
416 static int vpa_is_registered(struct kvmppc_vpa *vpap)
417 {
418 	if (vpap->update_pending)
419 		return vpap->next_gpa != 0;
420 	return vpap->pinned_addr != NULL;
421 }
422 
423 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
424 				       unsigned long flags,
425 				       unsigned long vcpuid, unsigned long vpa)
426 {
427 	struct kvm *kvm = vcpu->kvm;
428 	unsigned long len, nb;
429 	void *va;
430 	struct kvm_vcpu *tvcpu;
431 	int err;
432 	int subfunc;
433 	struct kvmppc_vpa *vpap;
434 
435 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
436 	if (!tvcpu)
437 		return H_PARAMETER;
438 
439 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
440 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
441 	    subfunc == H_VPA_REG_SLB) {
442 		/* Registering new area - address must be cache-line aligned */
443 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
444 			return H_PARAMETER;
445 
446 		/* convert logical addr to kernel addr and read length */
447 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
448 		if (va == NULL)
449 			return H_PARAMETER;
450 		if (subfunc == H_VPA_REG_VPA)
451 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
452 		else
453 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
454 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
455 
456 		/* Check length */
457 		if (len > nb || len < sizeof(struct reg_vpa))
458 			return H_PARAMETER;
459 	} else {
460 		vpa = 0;
461 		len = 0;
462 	}
463 
464 	err = H_PARAMETER;
465 	vpap = NULL;
466 	spin_lock(&tvcpu->arch.vpa_update_lock);
467 
468 	switch (subfunc) {
469 	case H_VPA_REG_VPA:		/* register VPA */
470 		if (len < sizeof(struct lppaca))
471 			break;
472 		vpap = &tvcpu->arch.vpa;
473 		err = 0;
474 		break;
475 
476 	case H_VPA_REG_DTL:		/* register DTL */
477 		if (len < sizeof(struct dtl_entry))
478 			break;
479 		len -= len % sizeof(struct dtl_entry);
480 
481 		/* Check that they have previously registered a VPA */
482 		err = H_RESOURCE;
483 		if (!vpa_is_registered(&tvcpu->arch.vpa))
484 			break;
485 
486 		vpap = &tvcpu->arch.dtl;
487 		err = 0;
488 		break;
489 
490 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
491 		/* Check that they have previously registered a VPA */
492 		err = H_RESOURCE;
493 		if (!vpa_is_registered(&tvcpu->arch.vpa))
494 			break;
495 
496 		vpap = &tvcpu->arch.slb_shadow;
497 		err = 0;
498 		break;
499 
500 	case H_VPA_DEREG_VPA:		/* deregister VPA */
501 		/* Check they don't still have a DTL or SLB buf registered */
502 		err = H_RESOURCE;
503 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
504 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
505 			break;
506 
507 		vpap = &tvcpu->arch.vpa;
508 		err = 0;
509 		break;
510 
511 	case H_VPA_DEREG_DTL:		/* deregister DTL */
512 		vpap = &tvcpu->arch.dtl;
513 		err = 0;
514 		break;
515 
516 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
517 		vpap = &tvcpu->arch.slb_shadow;
518 		err = 0;
519 		break;
520 	}
521 
522 	if (vpap) {
523 		vpap->next_gpa = vpa;
524 		vpap->len = len;
525 		vpap->update_pending = 1;
526 	}
527 
528 	spin_unlock(&tvcpu->arch.vpa_update_lock);
529 
530 	return err;
531 }
532 
533 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
534 {
535 	struct kvm *kvm = vcpu->kvm;
536 	void *va;
537 	unsigned long nb;
538 	unsigned long gpa;
539 
540 	/*
541 	 * We need to pin the page pointed to by vpap->next_gpa,
542 	 * but we can't call kvmppc_pin_guest_page under the lock
543 	 * as it does get_user_pages() and down_read().  So we
544 	 * have to drop the lock, pin the page, then get the lock
545 	 * again and check that a new area didn't get registered
546 	 * in the meantime.
547 	 */
548 	for (;;) {
549 		gpa = vpap->next_gpa;
550 		spin_unlock(&vcpu->arch.vpa_update_lock);
551 		va = NULL;
552 		nb = 0;
553 		if (gpa)
554 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
555 		spin_lock(&vcpu->arch.vpa_update_lock);
556 		if (gpa == vpap->next_gpa)
557 			break;
558 		/* sigh... unpin that one and try again */
559 		if (va)
560 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
561 	}
562 
563 	vpap->update_pending = 0;
564 	if (va && nb < vpap->len) {
565 		/*
566 		 * If it's now too short, it must be that userspace
567 		 * has changed the mappings underlying guest memory,
568 		 * so unregister the region.
569 		 */
570 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
571 		va = NULL;
572 	}
573 	if (vpap->pinned_addr)
574 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
575 					vpap->dirty);
576 	vpap->gpa = gpa;
577 	vpap->pinned_addr = va;
578 	vpap->dirty = false;
579 	if (va)
580 		vpap->pinned_end = va + vpap->len;
581 }
582 
583 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
584 {
585 	if (!(vcpu->arch.vpa.update_pending ||
586 	      vcpu->arch.slb_shadow.update_pending ||
587 	      vcpu->arch.dtl.update_pending))
588 		return;
589 
590 	spin_lock(&vcpu->arch.vpa_update_lock);
591 	if (vcpu->arch.vpa.update_pending) {
592 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
593 		if (vcpu->arch.vpa.pinned_addr)
594 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
595 	}
596 	if (vcpu->arch.dtl.update_pending) {
597 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
598 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
599 		vcpu->arch.dtl_index = 0;
600 	}
601 	if (vcpu->arch.slb_shadow.update_pending)
602 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
603 	spin_unlock(&vcpu->arch.vpa_update_lock);
604 }
605 
606 /*
607  * Return the accumulated stolen time for the vcore up until `now'.
608  * The caller should hold the vcore lock.
609  */
610 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
611 {
612 	u64 p;
613 	unsigned long flags;
614 
615 	spin_lock_irqsave(&vc->stoltb_lock, flags);
616 	p = vc->stolen_tb;
617 	if (vc->vcore_state != VCORE_INACTIVE &&
618 	    vc->preempt_tb != TB_NIL)
619 		p += now - vc->preempt_tb;
620 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
621 	return p;
622 }
623 
624 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
625 				    struct kvmppc_vcore *vc)
626 {
627 	struct dtl_entry *dt;
628 	struct lppaca *vpa;
629 	unsigned long stolen;
630 	unsigned long core_stolen;
631 	u64 now;
632 
633 	dt = vcpu->arch.dtl_ptr;
634 	vpa = vcpu->arch.vpa.pinned_addr;
635 	now = mftb();
636 	core_stolen = vcore_stolen_time(vc, now);
637 	stolen = core_stolen - vcpu->arch.stolen_logged;
638 	vcpu->arch.stolen_logged = core_stolen;
639 	spin_lock_irq(&vcpu->arch.tbacct_lock);
640 	stolen += vcpu->arch.busy_stolen;
641 	vcpu->arch.busy_stolen = 0;
642 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
643 	if (!dt || !vpa)
644 		return;
645 	memset(dt, 0, sizeof(struct dtl_entry));
646 	dt->dispatch_reason = 7;
647 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
648 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
649 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
650 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
651 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
652 	++dt;
653 	if (dt == vcpu->arch.dtl.pinned_end)
654 		dt = vcpu->arch.dtl.pinned_addr;
655 	vcpu->arch.dtl_ptr = dt;
656 	/* order writing *dt vs. writing vpa->dtl_idx */
657 	smp_wmb();
658 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
659 	vcpu->arch.dtl.dirty = true;
660 }
661 
662 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
663 {
664 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
665 		return true;
666 	if ((!vcpu->arch.vcore->arch_compat) &&
667 	    cpu_has_feature(CPU_FTR_ARCH_207S))
668 		return true;
669 	return false;
670 }
671 
672 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
673 			     unsigned long resource, unsigned long value1,
674 			     unsigned long value2)
675 {
676 	switch (resource) {
677 	case H_SET_MODE_RESOURCE_SET_CIABR:
678 		if (!kvmppc_power8_compatible(vcpu))
679 			return H_P2;
680 		if (value2)
681 			return H_P4;
682 		if (mflags)
683 			return H_UNSUPPORTED_FLAG_START;
684 		/* Guests can't breakpoint the hypervisor */
685 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
686 			return H_P3;
687 		vcpu->arch.ciabr  = value1;
688 		return H_SUCCESS;
689 	case H_SET_MODE_RESOURCE_SET_DAWR:
690 		if (!kvmppc_power8_compatible(vcpu))
691 			return H_P2;
692 		if (mflags)
693 			return H_UNSUPPORTED_FLAG_START;
694 		if (value2 & DABRX_HYP)
695 			return H_P4;
696 		vcpu->arch.dawr  = value1;
697 		vcpu->arch.dawrx = value2;
698 		return H_SUCCESS;
699 	default:
700 		return H_TOO_HARD;
701 	}
702 }
703 
704 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
705 {
706 	struct kvmppc_vcore *vcore = target->arch.vcore;
707 
708 	/*
709 	 * We expect to have been called by the real mode handler
710 	 * (kvmppc_rm_h_confer()) which would have directly returned
711 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
712 	 * have useful work to do and should not confer) so we don't
713 	 * recheck that here.
714 	 */
715 
716 	spin_lock(&vcore->lock);
717 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
718 	    vcore->vcore_state != VCORE_INACTIVE &&
719 	    vcore->runner)
720 		target = vcore->runner;
721 	spin_unlock(&vcore->lock);
722 
723 	return kvm_vcpu_yield_to(target);
724 }
725 
726 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
727 {
728 	int yield_count = 0;
729 	struct lppaca *lppaca;
730 
731 	spin_lock(&vcpu->arch.vpa_update_lock);
732 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
733 	if (lppaca)
734 		yield_count = be32_to_cpu(lppaca->yield_count);
735 	spin_unlock(&vcpu->arch.vpa_update_lock);
736 	return yield_count;
737 }
738 
739 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
740 {
741 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
742 	unsigned long target, ret = H_SUCCESS;
743 	int yield_count;
744 	struct kvm_vcpu *tvcpu;
745 	int idx, rc;
746 
747 	if (req <= MAX_HCALL_OPCODE &&
748 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
749 		return RESUME_HOST;
750 
751 	switch (req) {
752 	case H_CEDE:
753 		break;
754 	case H_PROD:
755 		target = kvmppc_get_gpr(vcpu, 4);
756 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
757 		if (!tvcpu) {
758 			ret = H_PARAMETER;
759 			break;
760 		}
761 		tvcpu->arch.prodded = 1;
762 		smp_mb();
763 		if (vcpu->arch.ceded) {
764 			if (swait_active(&vcpu->wq)) {
765 				swake_up(&vcpu->wq);
766 				vcpu->stat.halt_wakeup++;
767 			}
768 		}
769 		break;
770 	case H_CONFER:
771 		target = kvmppc_get_gpr(vcpu, 4);
772 		if (target == -1)
773 			break;
774 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
775 		if (!tvcpu) {
776 			ret = H_PARAMETER;
777 			break;
778 		}
779 		yield_count = kvmppc_get_gpr(vcpu, 5);
780 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
781 			break;
782 		kvm_arch_vcpu_yield_to(tvcpu);
783 		break;
784 	case H_REGISTER_VPA:
785 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
786 					kvmppc_get_gpr(vcpu, 5),
787 					kvmppc_get_gpr(vcpu, 6));
788 		break;
789 	case H_RTAS:
790 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
791 			return RESUME_HOST;
792 
793 		idx = srcu_read_lock(&vcpu->kvm->srcu);
794 		rc = kvmppc_rtas_hcall(vcpu);
795 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
796 
797 		if (rc == -ENOENT)
798 			return RESUME_HOST;
799 		else if (rc == 0)
800 			break;
801 
802 		/* Send the error out to userspace via KVM_RUN */
803 		return rc;
804 	case H_LOGICAL_CI_LOAD:
805 		ret = kvmppc_h_logical_ci_load(vcpu);
806 		if (ret == H_TOO_HARD)
807 			return RESUME_HOST;
808 		break;
809 	case H_LOGICAL_CI_STORE:
810 		ret = kvmppc_h_logical_ci_store(vcpu);
811 		if (ret == H_TOO_HARD)
812 			return RESUME_HOST;
813 		break;
814 	case H_SET_MODE:
815 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
816 					kvmppc_get_gpr(vcpu, 5),
817 					kvmppc_get_gpr(vcpu, 6),
818 					kvmppc_get_gpr(vcpu, 7));
819 		if (ret == H_TOO_HARD)
820 			return RESUME_HOST;
821 		break;
822 	case H_XIRR:
823 	case H_CPPR:
824 	case H_EOI:
825 	case H_IPI:
826 	case H_IPOLL:
827 	case H_XIRR_X:
828 		if (kvmppc_xics_enabled(vcpu)) {
829 			ret = kvmppc_xics_hcall(vcpu, req);
830 			break;
831 		}
832 		return RESUME_HOST;
833 	case H_PUT_TCE:
834 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
835 						kvmppc_get_gpr(vcpu, 5),
836 						kvmppc_get_gpr(vcpu, 6));
837 		if (ret == H_TOO_HARD)
838 			return RESUME_HOST;
839 		break;
840 	case H_PUT_TCE_INDIRECT:
841 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
842 						kvmppc_get_gpr(vcpu, 5),
843 						kvmppc_get_gpr(vcpu, 6),
844 						kvmppc_get_gpr(vcpu, 7));
845 		if (ret == H_TOO_HARD)
846 			return RESUME_HOST;
847 		break;
848 	case H_STUFF_TCE:
849 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
850 						kvmppc_get_gpr(vcpu, 5),
851 						kvmppc_get_gpr(vcpu, 6),
852 						kvmppc_get_gpr(vcpu, 7));
853 		if (ret == H_TOO_HARD)
854 			return RESUME_HOST;
855 		break;
856 	default:
857 		return RESUME_HOST;
858 	}
859 	kvmppc_set_gpr(vcpu, 3, ret);
860 	vcpu->arch.hcall_needed = 0;
861 	return RESUME_GUEST;
862 }
863 
864 static int kvmppc_hcall_impl_hv(unsigned long cmd)
865 {
866 	switch (cmd) {
867 	case H_CEDE:
868 	case H_PROD:
869 	case H_CONFER:
870 	case H_REGISTER_VPA:
871 	case H_SET_MODE:
872 	case H_LOGICAL_CI_LOAD:
873 	case H_LOGICAL_CI_STORE:
874 #ifdef CONFIG_KVM_XICS
875 	case H_XIRR:
876 	case H_CPPR:
877 	case H_EOI:
878 	case H_IPI:
879 	case H_IPOLL:
880 	case H_XIRR_X:
881 #endif
882 		return 1;
883 	}
884 
885 	/* See if it's in the real-mode table */
886 	return kvmppc_hcall_impl_hv_realmode(cmd);
887 }
888 
889 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
890 					struct kvm_vcpu *vcpu)
891 {
892 	u32 last_inst;
893 
894 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
895 					EMULATE_DONE) {
896 		/*
897 		 * Fetch failed, so return to guest and
898 		 * try executing it again.
899 		 */
900 		return RESUME_GUEST;
901 	}
902 
903 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
904 		run->exit_reason = KVM_EXIT_DEBUG;
905 		run->debug.arch.address = kvmppc_get_pc(vcpu);
906 		return RESUME_HOST;
907 	} else {
908 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
909 		return RESUME_GUEST;
910 	}
911 }
912 
913 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
914 				 struct task_struct *tsk)
915 {
916 	int r = RESUME_HOST;
917 
918 	vcpu->stat.sum_exits++;
919 
920 	/*
921 	 * This can happen if an interrupt occurs in the last stages
922 	 * of guest entry or the first stages of guest exit (i.e. after
923 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
924 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
925 	 * That can happen due to a bug, or due to a machine check
926 	 * occurring at just the wrong time.
927 	 */
928 	if (vcpu->arch.shregs.msr & MSR_HV) {
929 		printk(KERN_EMERG "KVM trap in HV mode!\n");
930 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
931 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
932 			vcpu->arch.shregs.msr);
933 		kvmppc_dump_regs(vcpu);
934 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
935 		run->hw.hardware_exit_reason = vcpu->arch.trap;
936 		return RESUME_HOST;
937 	}
938 	run->exit_reason = KVM_EXIT_UNKNOWN;
939 	run->ready_for_interrupt_injection = 1;
940 	switch (vcpu->arch.trap) {
941 	/* We're good on these - the host merely wanted to get our attention */
942 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
943 		vcpu->stat.dec_exits++;
944 		r = RESUME_GUEST;
945 		break;
946 	case BOOK3S_INTERRUPT_EXTERNAL:
947 	case BOOK3S_INTERRUPT_H_DOORBELL:
948 		vcpu->stat.ext_intr_exits++;
949 		r = RESUME_GUEST;
950 		break;
951 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
952 	case BOOK3S_INTERRUPT_HMI:
953 	case BOOK3S_INTERRUPT_PERFMON:
954 		r = RESUME_GUEST;
955 		break;
956 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
957 		/*
958 		 * Deliver a machine check interrupt to the guest.
959 		 * We have to do this, even if the host has handled the
960 		 * machine check, because machine checks use SRR0/1 and
961 		 * the interrupt might have trashed guest state in them.
962 		 */
963 		kvmppc_book3s_queue_irqprio(vcpu,
964 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
965 		r = RESUME_GUEST;
966 		break;
967 	case BOOK3S_INTERRUPT_PROGRAM:
968 	{
969 		ulong flags;
970 		/*
971 		 * Normally program interrupts are delivered directly
972 		 * to the guest by the hardware, but we can get here
973 		 * as a result of a hypervisor emulation interrupt
974 		 * (e40) getting turned into a 700 by BML RTAS.
975 		 */
976 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
977 		kvmppc_core_queue_program(vcpu, flags);
978 		r = RESUME_GUEST;
979 		break;
980 	}
981 	case BOOK3S_INTERRUPT_SYSCALL:
982 	{
983 		/* hcall - punt to userspace */
984 		int i;
985 
986 		/* hypercall with MSR_PR has already been handled in rmode,
987 		 * and never reaches here.
988 		 */
989 
990 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
991 		for (i = 0; i < 9; ++i)
992 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
993 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
994 		vcpu->arch.hcall_needed = 1;
995 		r = RESUME_HOST;
996 		break;
997 	}
998 	/*
999 	 * We get these next two if the guest accesses a page which it thinks
1000 	 * it has mapped but which is not actually present, either because
1001 	 * it is for an emulated I/O device or because the corresonding
1002 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1003 	 * have been handled already.
1004 	 */
1005 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1006 		r = RESUME_PAGE_FAULT;
1007 		break;
1008 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1009 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1010 		vcpu->arch.fault_dsisr = 0;
1011 		r = RESUME_PAGE_FAULT;
1012 		break;
1013 	/*
1014 	 * This occurs if the guest executes an illegal instruction.
1015 	 * If the guest debug is disabled, generate a program interrupt
1016 	 * to the guest. If guest debug is enabled, we need to check
1017 	 * whether the instruction is a software breakpoint instruction.
1018 	 * Accordingly return to Guest or Host.
1019 	 */
1020 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1021 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1022 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1023 				swab32(vcpu->arch.emul_inst) :
1024 				vcpu->arch.emul_inst;
1025 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1026 			r = kvmppc_emulate_debug_inst(run, vcpu);
1027 		} else {
1028 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1029 			r = RESUME_GUEST;
1030 		}
1031 		break;
1032 	/*
1033 	 * This occurs if the guest (kernel or userspace), does something that
1034 	 * is prohibited by HFSCR.  We just generate a program interrupt to
1035 	 * the guest.
1036 	 */
1037 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1038 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1039 		r = RESUME_GUEST;
1040 		break;
1041 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1042 		r = RESUME_PASSTHROUGH;
1043 		break;
1044 	default:
1045 		kvmppc_dump_regs(vcpu);
1046 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1047 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1048 			vcpu->arch.shregs.msr);
1049 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1050 		r = RESUME_HOST;
1051 		break;
1052 	}
1053 
1054 	return r;
1055 }
1056 
1057 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1058 					    struct kvm_sregs *sregs)
1059 {
1060 	int i;
1061 
1062 	memset(sregs, 0, sizeof(struct kvm_sregs));
1063 	sregs->pvr = vcpu->arch.pvr;
1064 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1065 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1066 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1067 	}
1068 
1069 	return 0;
1070 }
1071 
1072 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1073 					    struct kvm_sregs *sregs)
1074 {
1075 	int i, j;
1076 
1077 	/* Only accept the same PVR as the host's, since we can't spoof it */
1078 	if (sregs->pvr != vcpu->arch.pvr)
1079 		return -EINVAL;
1080 
1081 	j = 0;
1082 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1083 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1084 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1085 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1086 			++j;
1087 		}
1088 	}
1089 	vcpu->arch.slb_max = j;
1090 
1091 	return 0;
1092 }
1093 
1094 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1095 		bool preserve_top32)
1096 {
1097 	struct kvm *kvm = vcpu->kvm;
1098 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1099 	u64 mask;
1100 
1101 	mutex_lock(&kvm->lock);
1102 	spin_lock(&vc->lock);
1103 	/*
1104 	 * If ILE (interrupt little-endian) has changed, update the
1105 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1106 	 */
1107 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1108 		struct kvm_vcpu *vcpu;
1109 		int i;
1110 
1111 		kvm_for_each_vcpu(i, vcpu, kvm) {
1112 			if (vcpu->arch.vcore != vc)
1113 				continue;
1114 			if (new_lpcr & LPCR_ILE)
1115 				vcpu->arch.intr_msr |= MSR_LE;
1116 			else
1117 				vcpu->arch.intr_msr &= ~MSR_LE;
1118 		}
1119 	}
1120 
1121 	/*
1122 	 * Userspace can only modify DPFD (default prefetch depth),
1123 	 * ILE (interrupt little-endian) and TC (translation control).
1124 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1125 	 */
1126 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1127 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1128 		mask |= LPCR_AIL;
1129 
1130 	/* Broken 32-bit version of LPCR must not clear top bits */
1131 	if (preserve_top32)
1132 		mask &= 0xFFFFFFFF;
1133 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1134 	spin_unlock(&vc->lock);
1135 	mutex_unlock(&kvm->lock);
1136 }
1137 
1138 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1139 				 union kvmppc_one_reg *val)
1140 {
1141 	int r = 0;
1142 	long int i;
1143 
1144 	switch (id) {
1145 	case KVM_REG_PPC_DEBUG_INST:
1146 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1147 		break;
1148 	case KVM_REG_PPC_HIOR:
1149 		*val = get_reg_val(id, 0);
1150 		break;
1151 	case KVM_REG_PPC_DABR:
1152 		*val = get_reg_val(id, vcpu->arch.dabr);
1153 		break;
1154 	case KVM_REG_PPC_DABRX:
1155 		*val = get_reg_val(id, vcpu->arch.dabrx);
1156 		break;
1157 	case KVM_REG_PPC_DSCR:
1158 		*val = get_reg_val(id, vcpu->arch.dscr);
1159 		break;
1160 	case KVM_REG_PPC_PURR:
1161 		*val = get_reg_val(id, vcpu->arch.purr);
1162 		break;
1163 	case KVM_REG_PPC_SPURR:
1164 		*val = get_reg_val(id, vcpu->arch.spurr);
1165 		break;
1166 	case KVM_REG_PPC_AMR:
1167 		*val = get_reg_val(id, vcpu->arch.amr);
1168 		break;
1169 	case KVM_REG_PPC_UAMOR:
1170 		*val = get_reg_val(id, vcpu->arch.uamor);
1171 		break;
1172 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1173 		i = id - KVM_REG_PPC_MMCR0;
1174 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1175 		break;
1176 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1177 		i = id - KVM_REG_PPC_PMC1;
1178 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1179 		break;
1180 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1181 		i = id - KVM_REG_PPC_SPMC1;
1182 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1183 		break;
1184 	case KVM_REG_PPC_SIAR:
1185 		*val = get_reg_val(id, vcpu->arch.siar);
1186 		break;
1187 	case KVM_REG_PPC_SDAR:
1188 		*val = get_reg_val(id, vcpu->arch.sdar);
1189 		break;
1190 	case KVM_REG_PPC_SIER:
1191 		*val = get_reg_val(id, vcpu->arch.sier);
1192 		break;
1193 	case KVM_REG_PPC_IAMR:
1194 		*val = get_reg_val(id, vcpu->arch.iamr);
1195 		break;
1196 	case KVM_REG_PPC_PSPB:
1197 		*val = get_reg_val(id, vcpu->arch.pspb);
1198 		break;
1199 	case KVM_REG_PPC_DPDES:
1200 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1201 		break;
1202 	case KVM_REG_PPC_VTB:
1203 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1204 		break;
1205 	case KVM_REG_PPC_DAWR:
1206 		*val = get_reg_val(id, vcpu->arch.dawr);
1207 		break;
1208 	case KVM_REG_PPC_DAWRX:
1209 		*val = get_reg_val(id, vcpu->arch.dawrx);
1210 		break;
1211 	case KVM_REG_PPC_CIABR:
1212 		*val = get_reg_val(id, vcpu->arch.ciabr);
1213 		break;
1214 	case KVM_REG_PPC_CSIGR:
1215 		*val = get_reg_val(id, vcpu->arch.csigr);
1216 		break;
1217 	case KVM_REG_PPC_TACR:
1218 		*val = get_reg_val(id, vcpu->arch.tacr);
1219 		break;
1220 	case KVM_REG_PPC_TCSCR:
1221 		*val = get_reg_val(id, vcpu->arch.tcscr);
1222 		break;
1223 	case KVM_REG_PPC_PID:
1224 		*val = get_reg_val(id, vcpu->arch.pid);
1225 		break;
1226 	case KVM_REG_PPC_ACOP:
1227 		*val = get_reg_val(id, vcpu->arch.acop);
1228 		break;
1229 	case KVM_REG_PPC_WORT:
1230 		*val = get_reg_val(id, vcpu->arch.wort);
1231 		break;
1232 	case KVM_REG_PPC_VPA_ADDR:
1233 		spin_lock(&vcpu->arch.vpa_update_lock);
1234 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1235 		spin_unlock(&vcpu->arch.vpa_update_lock);
1236 		break;
1237 	case KVM_REG_PPC_VPA_SLB:
1238 		spin_lock(&vcpu->arch.vpa_update_lock);
1239 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1240 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1241 		spin_unlock(&vcpu->arch.vpa_update_lock);
1242 		break;
1243 	case KVM_REG_PPC_VPA_DTL:
1244 		spin_lock(&vcpu->arch.vpa_update_lock);
1245 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1246 		val->vpaval.length = vcpu->arch.dtl.len;
1247 		spin_unlock(&vcpu->arch.vpa_update_lock);
1248 		break;
1249 	case KVM_REG_PPC_TB_OFFSET:
1250 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1251 		break;
1252 	case KVM_REG_PPC_LPCR:
1253 	case KVM_REG_PPC_LPCR_64:
1254 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1255 		break;
1256 	case KVM_REG_PPC_PPR:
1257 		*val = get_reg_val(id, vcpu->arch.ppr);
1258 		break;
1259 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1260 	case KVM_REG_PPC_TFHAR:
1261 		*val = get_reg_val(id, vcpu->arch.tfhar);
1262 		break;
1263 	case KVM_REG_PPC_TFIAR:
1264 		*val = get_reg_val(id, vcpu->arch.tfiar);
1265 		break;
1266 	case KVM_REG_PPC_TEXASR:
1267 		*val = get_reg_val(id, vcpu->arch.texasr);
1268 		break;
1269 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1270 		i = id - KVM_REG_PPC_TM_GPR0;
1271 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1272 		break;
1273 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1274 	{
1275 		int j;
1276 		i = id - KVM_REG_PPC_TM_VSR0;
1277 		if (i < 32)
1278 			for (j = 0; j < TS_FPRWIDTH; j++)
1279 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1280 		else {
1281 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1282 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1283 			else
1284 				r = -ENXIO;
1285 		}
1286 		break;
1287 	}
1288 	case KVM_REG_PPC_TM_CR:
1289 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1290 		break;
1291 	case KVM_REG_PPC_TM_LR:
1292 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1293 		break;
1294 	case KVM_REG_PPC_TM_CTR:
1295 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1296 		break;
1297 	case KVM_REG_PPC_TM_FPSCR:
1298 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1299 		break;
1300 	case KVM_REG_PPC_TM_AMR:
1301 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1302 		break;
1303 	case KVM_REG_PPC_TM_PPR:
1304 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1305 		break;
1306 	case KVM_REG_PPC_TM_VRSAVE:
1307 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1308 		break;
1309 	case KVM_REG_PPC_TM_VSCR:
1310 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1311 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1312 		else
1313 			r = -ENXIO;
1314 		break;
1315 	case KVM_REG_PPC_TM_DSCR:
1316 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1317 		break;
1318 	case KVM_REG_PPC_TM_TAR:
1319 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1320 		break;
1321 #endif
1322 	case KVM_REG_PPC_ARCH_COMPAT:
1323 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1324 		break;
1325 	default:
1326 		r = -EINVAL;
1327 		break;
1328 	}
1329 
1330 	return r;
1331 }
1332 
1333 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1334 				 union kvmppc_one_reg *val)
1335 {
1336 	int r = 0;
1337 	long int i;
1338 	unsigned long addr, len;
1339 
1340 	switch (id) {
1341 	case KVM_REG_PPC_HIOR:
1342 		/* Only allow this to be set to zero */
1343 		if (set_reg_val(id, *val))
1344 			r = -EINVAL;
1345 		break;
1346 	case KVM_REG_PPC_DABR:
1347 		vcpu->arch.dabr = set_reg_val(id, *val);
1348 		break;
1349 	case KVM_REG_PPC_DABRX:
1350 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1351 		break;
1352 	case KVM_REG_PPC_DSCR:
1353 		vcpu->arch.dscr = set_reg_val(id, *val);
1354 		break;
1355 	case KVM_REG_PPC_PURR:
1356 		vcpu->arch.purr = set_reg_val(id, *val);
1357 		break;
1358 	case KVM_REG_PPC_SPURR:
1359 		vcpu->arch.spurr = set_reg_val(id, *val);
1360 		break;
1361 	case KVM_REG_PPC_AMR:
1362 		vcpu->arch.amr = set_reg_val(id, *val);
1363 		break;
1364 	case KVM_REG_PPC_UAMOR:
1365 		vcpu->arch.uamor = set_reg_val(id, *val);
1366 		break;
1367 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1368 		i = id - KVM_REG_PPC_MMCR0;
1369 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1370 		break;
1371 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1372 		i = id - KVM_REG_PPC_PMC1;
1373 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1374 		break;
1375 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1376 		i = id - KVM_REG_PPC_SPMC1;
1377 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1378 		break;
1379 	case KVM_REG_PPC_SIAR:
1380 		vcpu->arch.siar = set_reg_val(id, *val);
1381 		break;
1382 	case KVM_REG_PPC_SDAR:
1383 		vcpu->arch.sdar = set_reg_val(id, *val);
1384 		break;
1385 	case KVM_REG_PPC_SIER:
1386 		vcpu->arch.sier = set_reg_val(id, *val);
1387 		break;
1388 	case KVM_REG_PPC_IAMR:
1389 		vcpu->arch.iamr = set_reg_val(id, *val);
1390 		break;
1391 	case KVM_REG_PPC_PSPB:
1392 		vcpu->arch.pspb = set_reg_val(id, *val);
1393 		break;
1394 	case KVM_REG_PPC_DPDES:
1395 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1396 		break;
1397 	case KVM_REG_PPC_VTB:
1398 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1399 		break;
1400 	case KVM_REG_PPC_DAWR:
1401 		vcpu->arch.dawr = set_reg_val(id, *val);
1402 		break;
1403 	case KVM_REG_PPC_DAWRX:
1404 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1405 		break;
1406 	case KVM_REG_PPC_CIABR:
1407 		vcpu->arch.ciabr = set_reg_val(id, *val);
1408 		/* Don't allow setting breakpoints in hypervisor code */
1409 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1410 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1411 		break;
1412 	case KVM_REG_PPC_CSIGR:
1413 		vcpu->arch.csigr = set_reg_val(id, *val);
1414 		break;
1415 	case KVM_REG_PPC_TACR:
1416 		vcpu->arch.tacr = set_reg_val(id, *val);
1417 		break;
1418 	case KVM_REG_PPC_TCSCR:
1419 		vcpu->arch.tcscr = set_reg_val(id, *val);
1420 		break;
1421 	case KVM_REG_PPC_PID:
1422 		vcpu->arch.pid = set_reg_val(id, *val);
1423 		break;
1424 	case KVM_REG_PPC_ACOP:
1425 		vcpu->arch.acop = set_reg_val(id, *val);
1426 		break;
1427 	case KVM_REG_PPC_WORT:
1428 		vcpu->arch.wort = set_reg_val(id, *val);
1429 		break;
1430 	case KVM_REG_PPC_VPA_ADDR:
1431 		addr = set_reg_val(id, *val);
1432 		r = -EINVAL;
1433 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1434 			      vcpu->arch.dtl.next_gpa))
1435 			break;
1436 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1437 		break;
1438 	case KVM_REG_PPC_VPA_SLB:
1439 		addr = val->vpaval.addr;
1440 		len = val->vpaval.length;
1441 		r = -EINVAL;
1442 		if (addr && !vcpu->arch.vpa.next_gpa)
1443 			break;
1444 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1445 		break;
1446 	case KVM_REG_PPC_VPA_DTL:
1447 		addr = val->vpaval.addr;
1448 		len = val->vpaval.length;
1449 		r = -EINVAL;
1450 		if (addr && (len < sizeof(struct dtl_entry) ||
1451 			     !vcpu->arch.vpa.next_gpa))
1452 			break;
1453 		len -= len % sizeof(struct dtl_entry);
1454 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1455 		break;
1456 	case KVM_REG_PPC_TB_OFFSET:
1457 		/* round up to multiple of 2^24 */
1458 		vcpu->arch.vcore->tb_offset =
1459 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1460 		break;
1461 	case KVM_REG_PPC_LPCR:
1462 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1463 		break;
1464 	case KVM_REG_PPC_LPCR_64:
1465 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1466 		break;
1467 	case KVM_REG_PPC_PPR:
1468 		vcpu->arch.ppr = set_reg_val(id, *val);
1469 		break;
1470 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1471 	case KVM_REG_PPC_TFHAR:
1472 		vcpu->arch.tfhar = set_reg_val(id, *val);
1473 		break;
1474 	case KVM_REG_PPC_TFIAR:
1475 		vcpu->arch.tfiar = set_reg_val(id, *val);
1476 		break;
1477 	case KVM_REG_PPC_TEXASR:
1478 		vcpu->arch.texasr = set_reg_val(id, *val);
1479 		break;
1480 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1481 		i = id - KVM_REG_PPC_TM_GPR0;
1482 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1483 		break;
1484 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1485 	{
1486 		int j;
1487 		i = id - KVM_REG_PPC_TM_VSR0;
1488 		if (i < 32)
1489 			for (j = 0; j < TS_FPRWIDTH; j++)
1490 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1491 		else
1492 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1493 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1494 			else
1495 				r = -ENXIO;
1496 		break;
1497 	}
1498 	case KVM_REG_PPC_TM_CR:
1499 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1500 		break;
1501 	case KVM_REG_PPC_TM_LR:
1502 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1503 		break;
1504 	case KVM_REG_PPC_TM_CTR:
1505 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1506 		break;
1507 	case KVM_REG_PPC_TM_FPSCR:
1508 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1509 		break;
1510 	case KVM_REG_PPC_TM_AMR:
1511 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1512 		break;
1513 	case KVM_REG_PPC_TM_PPR:
1514 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1515 		break;
1516 	case KVM_REG_PPC_TM_VRSAVE:
1517 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1518 		break;
1519 	case KVM_REG_PPC_TM_VSCR:
1520 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1521 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1522 		else
1523 			r = - ENXIO;
1524 		break;
1525 	case KVM_REG_PPC_TM_DSCR:
1526 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1527 		break;
1528 	case KVM_REG_PPC_TM_TAR:
1529 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1530 		break;
1531 #endif
1532 	case KVM_REG_PPC_ARCH_COMPAT:
1533 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1534 		break;
1535 	default:
1536 		r = -EINVAL;
1537 		break;
1538 	}
1539 
1540 	return r;
1541 }
1542 
1543 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1544 {
1545 	struct kvmppc_vcore *vcore;
1546 
1547 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1548 
1549 	if (vcore == NULL)
1550 		return NULL;
1551 
1552 	spin_lock_init(&vcore->lock);
1553 	spin_lock_init(&vcore->stoltb_lock);
1554 	init_swait_queue_head(&vcore->wq);
1555 	vcore->preempt_tb = TB_NIL;
1556 	vcore->lpcr = kvm->arch.lpcr;
1557 	vcore->first_vcpuid = core * threads_per_subcore;
1558 	vcore->kvm = kvm;
1559 	INIT_LIST_HEAD(&vcore->preempt_list);
1560 
1561 	return vcore;
1562 }
1563 
1564 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1565 static struct debugfs_timings_element {
1566 	const char *name;
1567 	size_t offset;
1568 } timings[] = {
1569 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1570 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1571 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1572 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1573 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1574 };
1575 
1576 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1577 
1578 struct debugfs_timings_state {
1579 	struct kvm_vcpu	*vcpu;
1580 	unsigned int	buflen;
1581 	char		buf[N_TIMINGS * 100];
1582 };
1583 
1584 static int debugfs_timings_open(struct inode *inode, struct file *file)
1585 {
1586 	struct kvm_vcpu *vcpu = inode->i_private;
1587 	struct debugfs_timings_state *p;
1588 
1589 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1590 	if (!p)
1591 		return -ENOMEM;
1592 
1593 	kvm_get_kvm(vcpu->kvm);
1594 	p->vcpu = vcpu;
1595 	file->private_data = p;
1596 
1597 	return nonseekable_open(inode, file);
1598 }
1599 
1600 static int debugfs_timings_release(struct inode *inode, struct file *file)
1601 {
1602 	struct debugfs_timings_state *p = file->private_data;
1603 
1604 	kvm_put_kvm(p->vcpu->kvm);
1605 	kfree(p);
1606 	return 0;
1607 }
1608 
1609 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1610 				    size_t len, loff_t *ppos)
1611 {
1612 	struct debugfs_timings_state *p = file->private_data;
1613 	struct kvm_vcpu *vcpu = p->vcpu;
1614 	char *s, *buf_end;
1615 	struct kvmhv_tb_accumulator tb;
1616 	u64 count;
1617 	loff_t pos;
1618 	ssize_t n;
1619 	int i, loops;
1620 	bool ok;
1621 
1622 	if (!p->buflen) {
1623 		s = p->buf;
1624 		buf_end = s + sizeof(p->buf);
1625 		for (i = 0; i < N_TIMINGS; ++i) {
1626 			struct kvmhv_tb_accumulator *acc;
1627 
1628 			acc = (struct kvmhv_tb_accumulator *)
1629 				((unsigned long)vcpu + timings[i].offset);
1630 			ok = false;
1631 			for (loops = 0; loops < 1000; ++loops) {
1632 				count = acc->seqcount;
1633 				if (!(count & 1)) {
1634 					smp_rmb();
1635 					tb = *acc;
1636 					smp_rmb();
1637 					if (count == acc->seqcount) {
1638 						ok = true;
1639 						break;
1640 					}
1641 				}
1642 				udelay(1);
1643 			}
1644 			if (!ok)
1645 				snprintf(s, buf_end - s, "%s: stuck\n",
1646 					timings[i].name);
1647 			else
1648 				snprintf(s, buf_end - s,
1649 					"%s: %llu %llu %llu %llu\n",
1650 					timings[i].name, count / 2,
1651 					tb_to_ns(tb.tb_total),
1652 					tb_to_ns(tb.tb_min),
1653 					tb_to_ns(tb.tb_max));
1654 			s += strlen(s);
1655 		}
1656 		p->buflen = s - p->buf;
1657 	}
1658 
1659 	pos = *ppos;
1660 	if (pos >= p->buflen)
1661 		return 0;
1662 	if (len > p->buflen - pos)
1663 		len = p->buflen - pos;
1664 	n = copy_to_user(buf, p->buf + pos, len);
1665 	if (n) {
1666 		if (n == len)
1667 			return -EFAULT;
1668 		len -= n;
1669 	}
1670 	*ppos = pos + len;
1671 	return len;
1672 }
1673 
1674 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1675 				     size_t len, loff_t *ppos)
1676 {
1677 	return -EACCES;
1678 }
1679 
1680 static const struct file_operations debugfs_timings_ops = {
1681 	.owner	 = THIS_MODULE,
1682 	.open	 = debugfs_timings_open,
1683 	.release = debugfs_timings_release,
1684 	.read	 = debugfs_timings_read,
1685 	.write	 = debugfs_timings_write,
1686 	.llseek	 = generic_file_llseek,
1687 };
1688 
1689 /* Create a debugfs directory for the vcpu */
1690 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1691 {
1692 	char buf[16];
1693 	struct kvm *kvm = vcpu->kvm;
1694 
1695 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1696 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1697 		return;
1698 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1699 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1700 		return;
1701 	vcpu->arch.debugfs_timings =
1702 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1703 				    vcpu, &debugfs_timings_ops);
1704 }
1705 
1706 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1707 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1708 {
1709 }
1710 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1711 
1712 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1713 						   unsigned int id)
1714 {
1715 	struct kvm_vcpu *vcpu;
1716 	int err = -EINVAL;
1717 	int core;
1718 	struct kvmppc_vcore *vcore;
1719 
1720 	core = id / threads_per_subcore;
1721 	if (core >= KVM_MAX_VCORES)
1722 		goto out;
1723 
1724 	err = -ENOMEM;
1725 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1726 	if (!vcpu)
1727 		goto out;
1728 
1729 	err = kvm_vcpu_init(vcpu, kvm, id);
1730 	if (err)
1731 		goto free_vcpu;
1732 
1733 	vcpu->arch.shared = &vcpu->arch.shregs;
1734 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1735 	/*
1736 	 * The shared struct is never shared on HV,
1737 	 * so we can always use host endianness
1738 	 */
1739 #ifdef __BIG_ENDIAN__
1740 	vcpu->arch.shared_big_endian = true;
1741 #else
1742 	vcpu->arch.shared_big_endian = false;
1743 #endif
1744 #endif
1745 	vcpu->arch.mmcr[0] = MMCR0_FC;
1746 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1747 	/* default to host PVR, since we can't spoof it */
1748 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1749 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1750 	spin_lock_init(&vcpu->arch.tbacct_lock);
1751 	vcpu->arch.busy_preempt = TB_NIL;
1752 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1753 
1754 	kvmppc_mmu_book3s_hv_init(vcpu);
1755 
1756 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1757 
1758 	init_waitqueue_head(&vcpu->arch.cpu_run);
1759 
1760 	mutex_lock(&kvm->lock);
1761 	vcore = kvm->arch.vcores[core];
1762 	if (!vcore) {
1763 		vcore = kvmppc_vcore_create(kvm, core);
1764 		kvm->arch.vcores[core] = vcore;
1765 		kvm->arch.online_vcores++;
1766 	}
1767 	mutex_unlock(&kvm->lock);
1768 
1769 	if (!vcore)
1770 		goto free_vcpu;
1771 
1772 	spin_lock(&vcore->lock);
1773 	++vcore->num_threads;
1774 	spin_unlock(&vcore->lock);
1775 	vcpu->arch.vcore = vcore;
1776 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1777 	vcpu->arch.thread_cpu = -1;
1778 
1779 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1780 	kvmppc_sanity_check(vcpu);
1781 
1782 	debugfs_vcpu_init(vcpu, id);
1783 
1784 	return vcpu;
1785 
1786 free_vcpu:
1787 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1788 out:
1789 	return ERR_PTR(err);
1790 }
1791 
1792 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1793 {
1794 	if (vpa->pinned_addr)
1795 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1796 					vpa->dirty);
1797 }
1798 
1799 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1800 {
1801 	spin_lock(&vcpu->arch.vpa_update_lock);
1802 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1803 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1804 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1805 	spin_unlock(&vcpu->arch.vpa_update_lock);
1806 	kvm_vcpu_uninit(vcpu);
1807 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1808 }
1809 
1810 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1811 {
1812 	/* Indicate we want to get back into the guest */
1813 	return 1;
1814 }
1815 
1816 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1817 {
1818 	unsigned long dec_nsec, now;
1819 
1820 	now = get_tb();
1821 	if (now > vcpu->arch.dec_expires) {
1822 		/* decrementer has already gone negative */
1823 		kvmppc_core_queue_dec(vcpu);
1824 		kvmppc_core_prepare_to_enter(vcpu);
1825 		return;
1826 	}
1827 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1828 		   / tb_ticks_per_sec;
1829 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1830 		      HRTIMER_MODE_REL);
1831 	vcpu->arch.timer_running = 1;
1832 }
1833 
1834 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1835 {
1836 	vcpu->arch.ceded = 0;
1837 	if (vcpu->arch.timer_running) {
1838 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1839 		vcpu->arch.timer_running = 0;
1840 	}
1841 }
1842 
1843 extern void __kvmppc_vcore_entry(void);
1844 
1845 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1846 				   struct kvm_vcpu *vcpu)
1847 {
1848 	u64 now;
1849 
1850 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1851 		return;
1852 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1853 	now = mftb();
1854 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1855 		vcpu->arch.stolen_logged;
1856 	vcpu->arch.busy_preempt = now;
1857 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1858 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1859 	--vc->n_runnable;
1860 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
1861 }
1862 
1863 static int kvmppc_grab_hwthread(int cpu)
1864 {
1865 	struct paca_struct *tpaca;
1866 	long timeout = 10000;
1867 
1868 	tpaca = &paca[cpu];
1869 
1870 	/* Ensure the thread won't go into the kernel if it wakes */
1871 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1872 	tpaca->kvm_hstate.kvm_vcore = NULL;
1873 	tpaca->kvm_hstate.napping = 0;
1874 	smp_wmb();
1875 	tpaca->kvm_hstate.hwthread_req = 1;
1876 
1877 	/*
1878 	 * If the thread is already executing in the kernel (e.g. handling
1879 	 * a stray interrupt), wait for it to get back to nap mode.
1880 	 * The smp_mb() is to ensure that our setting of hwthread_req
1881 	 * is visible before we look at hwthread_state, so if this
1882 	 * races with the code at system_reset_pSeries and the thread
1883 	 * misses our setting of hwthread_req, we are sure to see its
1884 	 * setting of hwthread_state, and vice versa.
1885 	 */
1886 	smp_mb();
1887 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1888 		if (--timeout <= 0) {
1889 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1890 			return -EBUSY;
1891 		}
1892 		udelay(1);
1893 	}
1894 	return 0;
1895 }
1896 
1897 static void kvmppc_release_hwthread(int cpu)
1898 {
1899 	struct paca_struct *tpaca;
1900 
1901 	tpaca = &paca[cpu];
1902 	tpaca->kvm_hstate.hwthread_req = 0;
1903 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1904 	tpaca->kvm_hstate.kvm_vcore = NULL;
1905 	tpaca->kvm_hstate.kvm_split_mode = NULL;
1906 }
1907 
1908 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1909 {
1910 	int cpu;
1911 	struct paca_struct *tpaca;
1912 	struct kvmppc_vcore *mvc = vc->master_vcore;
1913 
1914 	cpu = vc->pcpu;
1915 	if (vcpu) {
1916 		if (vcpu->arch.timer_running) {
1917 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1918 			vcpu->arch.timer_running = 0;
1919 		}
1920 		cpu += vcpu->arch.ptid;
1921 		vcpu->cpu = mvc->pcpu;
1922 		vcpu->arch.thread_cpu = cpu;
1923 	}
1924 	tpaca = &paca[cpu];
1925 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1926 	tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1927 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1928 	smp_wmb();
1929 	tpaca->kvm_hstate.kvm_vcore = mvc;
1930 	if (cpu != smp_processor_id())
1931 		kvmppc_ipi_thread(cpu);
1932 }
1933 
1934 static void kvmppc_wait_for_nap(void)
1935 {
1936 	int cpu = smp_processor_id();
1937 	int i, loops;
1938 
1939 	for (loops = 0; loops < 1000000; ++loops) {
1940 		/*
1941 		 * Check if all threads are finished.
1942 		 * We set the vcore pointer when starting a thread
1943 		 * and the thread clears it when finished, so we look
1944 		 * for any threads that still have a non-NULL vcore ptr.
1945 		 */
1946 		for (i = 1; i < threads_per_subcore; ++i)
1947 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
1948 				break;
1949 		if (i == threads_per_subcore) {
1950 			HMT_medium();
1951 			return;
1952 		}
1953 		HMT_low();
1954 	}
1955 	HMT_medium();
1956 	for (i = 1; i < threads_per_subcore; ++i)
1957 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
1958 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1959 }
1960 
1961 /*
1962  * Check that we are on thread 0 and that any other threads in
1963  * this core are off-line.  Then grab the threads so they can't
1964  * enter the kernel.
1965  */
1966 static int on_primary_thread(void)
1967 {
1968 	int cpu = smp_processor_id();
1969 	int thr;
1970 
1971 	/* Are we on a primary subcore? */
1972 	if (cpu_thread_in_subcore(cpu))
1973 		return 0;
1974 
1975 	thr = 0;
1976 	while (++thr < threads_per_subcore)
1977 		if (cpu_online(cpu + thr))
1978 			return 0;
1979 
1980 	/* Grab all hw threads so they can't go into the kernel */
1981 	for (thr = 1; thr < threads_per_subcore; ++thr) {
1982 		if (kvmppc_grab_hwthread(cpu + thr)) {
1983 			/* Couldn't grab one; let the others go */
1984 			do {
1985 				kvmppc_release_hwthread(cpu + thr);
1986 			} while (--thr > 0);
1987 			return 0;
1988 		}
1989 	}
1990 	return 1;
1991 }
1992 
1993 /*
1994  * A list of virtual cores for each physical CPU.
1995  * These are vcores that could run but their runner VCPU tasks are
1996  * (or may be) preempted.
1997  */
1998 struct preempted_vcore_list {
1999 	struct list_head	list;
2000 	spinlock_t		lock;
2001 };
2002 
2003 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2004 
2005 static void init_vcore_lists(void)
2006 {
2007 	int cpu;
2008 
2009 	for_each_possible_cpu(cpu) {
2010 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2011 		spin_lock_init(&lp->lock);
2012 		INIT_LIST_HEAD(&lp->list);
2013 	}
2014 }
2015 
2016 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2017 {
2018 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2019 
2020 	vc->vcore_state = VCORE_PREEMPT;
2021 	vc->pcpu = smp_processor_id();
2022 	if (vc->num_threads < threads_per_subcore) {
2023 		spin_lock(&lp->lock);
2024 		list_add_tail(&vc->preempt_list, &lp->list);
2025 		spin_unlock(&lp->lock);
2026 	}
2027 
2028 	/* Start accumulating stolen time */
2029 	kvmppc_core_start_stolen(vc);
2030 }
2031 
2032 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2033 {
2034 	struct preempted_vcore_list *lp;
2035 
2036 	kvmppc_core_end_stolen(vc);
2037 	if (!list_empty(&vc->preempt_list)) {
2038 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2039 		spin_lock(&lp->lock);
2040 		list_del_init(&vc->preempt_list);
2041 		spin_unlock(&lp->lock);
2042 	}
2043 	vc->vcore_state = VCORE_INACTIVE;
2044 }
2045 
2046 /*
2047  * This stores information about the virtual cores currently
2048  * assigned to a physical core.
2049  */
2050 struct core_info {
2051 	int		n_subcores;
2052 	int		max_subcore_threads;
2053 	int		total_threads;
2054 	int		subcore_threads[MAX_SUBCORES];
2055 	struct kvm	*subcore_vm[MAX_SUBCORES];
2056 	struct list_head vcs[MAX_SUBCORES];
2057 };
2058 
2059 /*
2060  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2061  * respectively in 2-way micro-threading (split-core) mode.
2062  */
2063 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2064 
2065 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2066 {
2067 	int sub;
2068 
2069 	memset(cip, 0, sizeof(*cip));
2070 	cip->n_subcores = 1;
2071 	cip->max_subcore_threads = vc->num_threads;
2072 	cip->total_threads = vc->num_threads;
2073 	cip->subcore_threads[0] = vc->num_threads;
2074 	cip->subcore_vm[0] = vc->kvm;
2075 	for (sub = 0; sub < MAX_SUBCORES; ++sub)
2076 		INIT_LIST_HEAD(&cip->vcs[sub]);
2077 	list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2078 }
2079 
2080 static bool subcore_config_ok(int n_subcores, int n_threads)
2081 {
2082 	/* Can only dynamically split if unsplit to begin with */
2083 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2084 		return false;
2085 	if (n_subcores > MAX_SUBCORES)
2086 		return false;
2087 	if (n_subcores > 1) {
2088 		if (!(dynamic_mt_modes & 2))
2089 			n_subcores = 4;
2090 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2091 			return false;
2092 	}
2093 
2094 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2095 }
2096 
2097 static void init_master_vcore(struct kvmppc_vcore *vc)
2098 {
2099 	vc->master_vcore = vc;
2100 	vc->entry_exit_map = 0;
2101 	vc->in_guest = 0;
2102 	vc->napping_threads = 0;
2103 	vc->conferring_threads = 0;
2104 }
2105 
2106 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2107 {
2108 	int n_threads = vc->num_threads;
2109 	int sub;
2110 
2111 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2112 		return false;
2113 
2114 	if (n_threads < cip->max_subcore_threads)
2115 		n_threads = cip->max_subcore_threads;
2116 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2117 		return false;
2118 	cip->max_subcore_threads = n_threads;
2119 
2120 	sub = cip->n_subcores;
2121 	++cip->n_subcores;
2122 	cip->total_threads += vc->num_threads;
2123 	cip->subcore_threads[sub] = vc->num_threads;
2124 	cip->subcore_vm[sub] = vc->kvm;
2125 	init_master_vcore(vc);
2126 	list_del(&vc->preempt_list);
2127 	list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2128 
2129 	return true;
2130 }
2131 
2132 /*
2133  * Work out whether it is possible to piggyback the execution of
2134  * vcore *pvc onto the execution of the other vcores described in *cip.
2135  */
2136 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2137 			  int target_threads)
2138 {
2139 	if (cip->total_threads + pvc->num_threads > target_threads)
2140 		return false;
2141 
2142 	return can_dynamic_split(pvc, cip);
2143 }
2144 
2145 static void prepare_threads(struct kvmppc_vcore *vc)
2146 {
2147 	int i;
2148 	struct kvm_vcpu *vcpu;
2149 
2150 	for_each_runnable_thread(i, vcpu, vc) {
2151 		if (signal_pending(vcpu->arch.run_task))
2152 			vcpu->arch.ret = -EINTR;
2153 		else if (vcpu->arch.vpa.update_pending ||
2154 			 vcpu->arch.slb_shadow.update_pending ||
2155 			 vcpu->arch.dtl.update_pending)
2156 			vcpu->arch.ret = RESUME_GUEST;
2157 		else
2158 			continue;
2159 		kvmppc_remove_runnable(vc, vcpu);
2160 		wake_up(&vcpu->arch.cpu_run);
2161 	}
2162 }
2163 
2164 static void collect_piggybacks(struct core_info *cip, int target_threads)
2165 {
2166 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2167 	struct kvmppc_vcore *pvc, *vcnext;
2168 
2169 	spin_lock(&lp->lock);
2170 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2171 		if (!spin_trylock(&pvc->lock))
2172 			continue;
2173 		prepare_threads(pvc);
2174 		if (!pvc->n_runnable) {
2175 			list_del_init(&pvc->preempt_list);
2176 			if (pvc->runner == NULL) {
2177 				pvc->vcore_state = VCORE_INACTIVE;
2178 				kvmppc_core_end_stolen(pvc);
2179 			}
2180 			spin_unlock(&pvc->lock);
2181 			continue;
2182 		}
2183 		if (!can_piggyback(pvc, cip, target_threads)) {
2184 			spin_unlock(&pvc->lock);
2185 			continue;
2186 		}
2187 		kvmppc_core_end_stolen(pvc);
2188 		pvc->vcore_state = VCORE_PIGGYBACK;
2189 		if (cip->total_threads >= target_threads)
2190 			break;
2191 	}
2192 	spin_unlock(&lp->lock);
2193 }
2194 
2195 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2196 {
2197 	int still_running = 0, i;
2198 	u64 now;
2199 	long ret;
2200 	struct kvm_vcpu *vcpu;
2201 
2202 	spin_lock(&vc->lock);
2203 	now = get_tb();
2204 	for_each_runnable_thread(i, vcpu, vc) {
2205 		/* cancel pending dec exception if dec is positive */
2206 		if (now < vcpu->arch.dec_expires &&
2207 		    kvmppc_core_pending_dec(vcpu))
2208 			kvmppc_core_dequeue_dec(vcpu);
2209 
2210 		trace_kvm_guest_exit(vcpu);
2211 
2212 		ret = RESUME_GUEST;
2213 		if (vcpu->arch.trap)
2214 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2215 						    vcpu->arch.run_task);
2216 
2217 		vcpu->arch.ret = ret;
2218 		vcpu->arch.trap = 0;
2219 
2220 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2221 			if (vcpu->arch.pending_exceptions)
2222 				kvmppc_core_prepare_to_enter(vcpu);
2223 			if (vcpu->arch.ceded)
2224 				kvmppc_set_timer(vcpu);
2225 			else
2226 				++still_running;
2227 		} else {
2228 			kvmppc_remove_runnable(vc, vcpu);
2229 			wake_up(&vcpu->arch.cpu_run);
2230 		}
2231 	}
2232 	list_del_init(&vc->preempt_list);
2233 	if (!is_master) {
2234 		if (still_running > 0) {
2235 			kvmppc_vcore_preempt(vc);
2236 		} else if (vc->runner) {
2237 			vc->vcore_state = VCORE_PREEMPT;
2238 			kvmppc_core_start_stolen(vc);
2239 		} else {
2240 			vc->vcore_state = VCORE_INACTIVE;
2241 		}
2242 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2243 			/* make sure there's a candidate runner awake */
2244 			i = -1;
2245 			vcpu = next_runnable_thread(vc, &i);
2246 			wake_up(&vcpu->arch.cpu_run);
2247 		}
2248 	}
2249 	spin_unlock(&vc->lock);
2250 }
2251 
2252 /*
2253  * Clear core from the list of active host cores as we are about to
2254  * enter the guest. Only do this if it is the primary thread of the
2255  * core (not if a subcore) that is entering the guest.
2256  */
2257 static inline void kvmppc_clear_host_core(int cpu)
2258 {
2259 	int core;
2260 
2261 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2262 		return;
2263 	/*
2264 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2265 	 * later in kvmppc_start_thread and we need ensure that state is
2266 	 * visible to other CPUs only after we enter guest.
2267 	 */
2268 	core = cpu >> threads_shift;
2269 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2270 }
2271 
2272 /*
2273  * Advertise this core as an active host core since we exited the guest
2274  * Only need to do this if it is the primary thread of the core that is
2275  * exiting.
2276  */
2277 static inline void kvmppc_set_host_core(int cpu)
2278 {
2279 	int core;
2280 
2281 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2282 		return;
2283 
2284 	/*
2285 	 * Memory barrier can be omitted here because we do a spin_unlock
2286 	 * immediately after this which provides the memory barrier.
2287 	 */
2288 	core = cpu >> threads_shift;
2289 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2290 }
2291 
2292 /*
2293  * Run a set of guest threads on a physical core.
2294  * Called with vc->lock held.
2295  */
2296 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2297 {
2298 	struct kvm_vcpu *vcpu;
2299 	int i;
2300 	int srcu_idx;
2301 	struct core_info core_info;
2302 	struct kvmppc_vcore *pvc, *vcnext;
2303 	struct kvm_split_mode split_info, *sip;
2304 	int split, subcore_size, active;
2305 	int sub;
2306 	bool thr0_done;
2307 	unsigned long cmd_bit, stat_bit;
2308 	int pcpu, thr;
2309 	int target_threads;
2310 
2311 	/*
2312 	 * Remove from the list any threads that have a signal pending
2313 	 * or need a VPA update done
2314 	 */
2315 	prepare_threads(vc);
2316 
2317 	/* if the runner is no longer runnable, let the caller pick a new one */
2318 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2319 		return;
2320 
2321 	/*
2322 	 * Initialize *vc.
2323 	 */
2324 	init_master_vcore(vc);
2325 	vc->preempt_tb = TB_NIL;
2326 
2327 	/*
2328 	 * Make sure we are running on primary threads, and that secondary
2329 	 * threads are offline.  Also check if the number of threads in this
2330 	 * guest are greater than the current system threads per guest.
2331 	 */
2332 	if ((threads_per_core > 1) &&
2333 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2334 		for_each_runnable_thread(i, vcpu, vc) {
2335 			vcpu->arch.ret = -EBUSY;
2336 			kvmppc_remove_runnable(vc, vcpu);
2337 			wake_up(&vcpu->arch.cpu_run);
2338 		}
2339 		goto out;
2340 	}
2341 
2342 	/*
2343 	 * See if we could run any other vcores on the physical core
2344 	 * along with this one.
2345 	 */
2346 	init_core_info(&core_info, vc);
2347 	pcpu = smp_processor_id();
2348 	target_threads = threads_per_subcore;
2349 	if (target_smt_mode && target_smt_mode < target_threads)
2350 		target_threads = target_smt_mode;
2351 	if (vc->num_threads < target_threads)
2352 		collect_piggybacks(&core_info, target_threads);
2353 
2354 	/* Decide on micro-threading (split-core) mode */
2355 	subcore_size = threads_per_subcore;
2356 	cmd_bit = stat_bit = 0;
2357 	split = core_info.n_subcores;
2358 	sip = NULL;
2359 	if (split > 1) {
2360 		/* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2361 		if (split == 2 && (dynamic_mt_modes & 2)) {
2362 			cmd_bit = HID0_POWER8_1TO2LPAR;
2363 			stat_bit = HID0_POWER8_2LPARMODE;
2364 		} else {
2365 			split = 4;
2366 			cmd_bit = HID0_POWER8_1TO4LPAR;
2367 			stat_bit = HID0_POWER8_4LPARMODE;
2368 		}
2369 		subcore_size = MAX_SMT_THREADS / split;
2370 		sip = &split_info;
2371 		memset(&split_info, 0, sizeof(split_info));
2372 		split_info.rpr = mfspr(SPRN_RPR);
2373 		split_info.pmmar = mfspr(SPRN_PMMAR);
2374 		split_info.ldbar = mfspr(SPRN_LDBAR);
2375 		split_info.subcore_size = subcore_size;
2376 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2377 			split_info.master_vcs[sub] =
2378 				list_first_entry(&core_info.vcs[sub],
2379 					struct kvmppc_vcore, preempt_list);
2380 		/* order writes to split_info before kvm_split_mode pointer */
2381 		smp_wmb();
2382 	}
2383 	pcpu = smp_processor_id();
2384 	for (thr = 0; thr < threads_per_subcore; ++thr)
2385 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2386 
2387 	/* Initiate micro-threading (split-core) if required */
2388 	if (cmd_bit) {
2389 		unsigned long hid0 = mfspr(SPRN_HID0);
2390 
2391 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2392 		mb();
2393 		mtspr(SPRN_HID0, hid0);
2394 		isync();
2395 		for (;;) {
2396 			hid0 = mfspr(SPRN_HID0);
2397 			if (hid0 & stat_bit)
2398 				break;
2399 			cpu_relax();
2400 		}
2401 	}
2402 
2403 	kvmppc_clear_host_core(pcpu);
2404 
2405 	/* Start all the threads */
2406 	active = 0;
2407 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2408 		thr = subcore_thread_map[sub];
2409 		thr0_done = false;
2410 		active |= 1 << thr;
2411 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2412 			pvc->pcpu = pcpu + thr;
2413 			for_each_runnable_thread(i, vcpu, pvc) {
2414 				kvmppc_start_thread(vcpu, pvc);
2415 				kvmppc_create_dtl_entry(vcpu, pvc);
2416 				trace_kvm_guest_enter(vcpu);
2417 				if (!vcpu->arch.ptid)
2418 					thr0_done = true;
2419 				active |= 1 << (thr + vcpu->arch.ptid);
2420 			}
2421 			/*
2422 			 * We need to start the first thread of each subcore
2423 			 * even if it doesn't have a vcpu.
2424 			 */
2425 			if (pvc->master_vcore == pvc && !thr0_done)
2426 				kvmppc_start_thread(NULL, pvc);
2427 			thr += pvc->num_threads;
2428 		}
2429 	}
2430 
2431 	/*
2432 	 * Ensure that split_info.do_nap is set after setting
2433 	 * the vcore pointer in the PACA of the secondaries.
2434 	 */
2435 	smp_mb();
2436 	if (cmd_bit)
2437 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2438 
2439 	/*
2440 	 * When doing micro-threading, poke the inactive threads as well.
2441 	 * This gets them to the nap instruction after kvm_do_nap,
2442 	 * which reduces the time taken to unsplit later.
2443 	 */
2444 	if (split > 1)
2445 		for (thr = 1; thr < threads_per_subcore; ++thr)
2446 			if (!(active & (1 << thr)))
2447 				kvmppc_ipi_thread(pcpu + thr);
2448 
2449 	vc->vcore_state = VCORE_RUNNING;
2450 	preempt_disable();
2451 
2452 	trace_kvmppc_run_core(vc, 0);
2453 
2454 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2455 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2456 			spin_unlock(&pvc->lock);
2457 
2458 	guest_enter();
2459 
2460 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2461 
2462 	__kvmppc_vcore_entry();
2463 
2464 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2465 
2466 	spin_lock(&vc->lock);
2467 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2468 	vc->vcore_state = VCORE_EXITING;
2469 
2470 	/* wait for secondary threads to finish writing their state to memory */
2471 	kvmppc_wait_for_nap();
2472 
2473 	/* Return to whole-core mode if we split the core earlier */
2474 	if (split > 1) {
2475 		unsigned long hid0 = mfspr(SPRN_HID0);
2476 		unsigned long loops = 0;
2477 
2478 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2479 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2480 		mb();
2481 		mtspr(SPRN_HID0, hid0);
2482 		isync();
2483 		for (;;) {
2484 			hid0 = mfspr(SPRN_HID0);
2485 			if (!(hid0 & stat_bit))
2486 				break;
2487 			cpu_relax();
2488 			++loops;
2489 		}
2490 		split_info.do_nap = 0;
2491 	}
2492 
2493 	/* Let secondaries go back to the offline loop */
2494 	for (i = 0; i < threads_per_subcore; ++i) {
2495 		kvmppc_release_hwthread(pcpu + i);
2496 		if (sip && sip->napped[i])
2497 			kvmppc_ipi_thread(pcpu + i);
2498 	}
2499 
2500 	kvmppc_set_host_core(pcpu);
2501 
2502 	spin_unlock(&vc->lock);
2503 
2504 	/* make sure updates to secondary vcpu structs are visible now */
2505 	smp_mb();
2506 	guest_exit();
2507 
2508 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2509 		list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2510 					 preempt_list)
2511 			post_guest_process(pvc, pvc == vc);
2512 
2513 	spin_lock(&vc->lock);
2514 	preempt_enable();
2515 
2516  out:
2517 	vc->vcore_state = VCORE_INACTIVE;
2518 	trace_kvmppc_run_core(vc, 1);
2519 }
2520 
2521 /*
2522  * Wait for some other vcpu thread to execute us, and
2523  * wake us up when we need to handle something in the host.
2524  */
2525 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2526 				 struct kvm_vcpu *vcpu, int wait_state)
2527 {
2528 	DEFINE_WAIT(wait);
2529 
2530 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2531 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2532 		spin_unlock(&vc->lock);
2533 		schedule();
2534 		spin_lock(&vc->lock);
2535 	}
2536 	finish_wait(&vcpu->arch.cpu_run, &wait);
2537 }
2538 
2539 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2540 {
2541 	/* 10us base */
2542 	if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2543 		vc->halt_poll_ns = 10000;
2544 	else
2545 		vc->halt_poll_ns *= halt_poll_ns_grow;
2546 
2547 	if (vc->halt_poll_ns > halt_poll_max_ns)
2548 		vc->halt_poll_ns = halt_poll_max_ns;
2549 }
2550 
2551 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2552 {
2553 	if (halt_poll_ns_shrink == 0)
2554 		vc->halt_poll_ns = 0;
2555 	else
2556 		vc->halt_poll_ns /= halt_poll_ns_shrink;
2557 }
2558 
2559 /* Check to see if any of the runnable vcpus on the vcore have pending
2560  * exceptions or are no longer ceded
2561  */
2562 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2563 {
2564 	struct kvm_vcpu *vcpu;
2565 	int i;
2566 
2567 	for_each_runnable_thread(i, vcpu, vc) {
2568 		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded)
2569 			return 1;
2570 	}
2571 
2572 	return 0;
2573 }
2574 
2575 /*
2576  * All the vcpus in this vcore are idle, so wait for a decrementer
2577  * or external interrupt to one of the vcpus.  vc->lock is held.
2578  */
2579 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2580 {
2581 	ktime_t cur, start_poll, start_wait;
2582 	int do_sleep = 1;
2583 	u64 block_ns;
2584 	DECLARE_SWAITQUEUE(wait);
2585 
2586 	/* Poll for pending exceptions and ceded state */
2587 	cur = start_poll = ktime_get();
2588 	if (vc->halt_poll_ns) {
2589 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2590 		++vc->runner->stat.halt_attempted_poll;
2591 
2592 		vc->vcore_state = VCORE_POLLING;
2593 		spin_unlock(&vc->lock);
2594 
2595 		do {
2596 			if (kvmppc_vcore_check_block(vc)) {
2597 				do_sleep = 0;
2598 				break;
2599 			}
2600 			cur = ktime_get();
2601 		} while (single_task_running() && ktime_before(cur, stop));
2602 
2603 		spin_lock(&vc->lock);
2604 		vc->vcore_state = VCORE_INACTIVE;
2605 
2606 		if (!do_sleep) {
2607 			++vc->runner->stat.halt_successful_poll;
2608 			goto out;
2609 		}
2610 	}
2611 
2612 	prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2613 
2614 	if (kvmppc_vcore_check_block(vc)) {
2615 		finish_swait(&vc->wq, &wait);
2616 		do_sleep = 0;
2617 		/* If we polled, count this as a successful poll */
2618 		if (vc->halt_poll_ns)
2619 			++vc->runner->stat.halt_successful_poll;
2620 		goto out;
2621 	}
2622 
2623 	start_wait = ktime_get();
2624 
2625 	vc->vcore_state = VCORE_SLEEPING;
2626 	trace_kvmppc_vcore_blocked(vc, 0);
2627 	spin_unlock(&vc->lock);
2628 	schedule();
2629 	finish_swait(&vc->wq, &wait);
2630 	spin_lock(&vc->lock);
2631 	vc->vcore_state = VCORE_INACTIVE;
2632 	trace_kvmppc_vcore_blocked(vc, 1);
2633 	++vc->runner->stat.halt_successful_wait;
2634 
2635 	cur = ktime_get();
2636 
2637 out:
2638 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
2639 
2640 	/* Attribute wait time */
2641 	if (do_sleep) {
2642 		vc->runner->stat.halt_wait_ns +=
2643 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
2644 		/* Attribute failed poll time */
2645 		if (vc->halt_poll_ns)
2646 			vc->runner->stat.halt_poll_fail_ns +=
2647 				ktime_to_ns(start_wait) -
2648 				ktime_to_ns(start_poll);
2649 	} else {
2650 		/* Attribute successful poll time */
2651 		if (vc->halt_poll_ns)
2652 			vc->runner->stat.halt_poll_success_ns +=
2653 				ktime_to_ns(cur) -
2654 				ktime_to_ns(start_poll);
2655 	}
2656 
2657 	/* Adjust poll time */
2658 	if (halt_poll_max_ns) {
2659 		if (block_ns <= vc->halt_poll_ns)
2660 			;
2661 		/* We slept and blocked for longer than the max halt time */
2662 		else if (vc->halt_poll_ns && block_ns > halt_poll_max_ns)
2663 			shrink_halt_poll_ns(vc);
2664 		/* We slept and our poll time is too small */
2665 		else if (vc->halt_poll_ns < halt_poll_max_ns &&
2666 				block_ns < halt_poll_max_ns)
2667 			grow_halt_poll_ns(vc);
2668 	} else
2669 		vc->halt_poll_ns = 0;
2670 
2671 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
2672 }
2673 
2674 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2675 {
2676 	int n_ceded, i;
2677 	struct kvmppc_vcore *vc;
2678 	struct kvm_vcpu *v;
2679 
2680 	trace_kvmppc_run_vcpu_enter(vcpu);
2681 
2682 	kvm_run->exit_reason = 0;
2683 	vcpu->arch.ret = RESUME_GUEST;
2684 	vcpu->arch.trap = 0;
2685 	kvmppc_update_vpas(vcpu);
2686 
2687 	/*
2688 	 * Synchronize with other threads in this virtual core
2689 	 */
2690 	vc = vcpu->arch.vcore;
2691 	spin_lock(&vc->lock);
2692 	vcpu->arch.ceded = 0;
2693 	vcpu->arch.run_task = current;
2694 	vcpu->arch.kvm_run = kvm_run;
2695 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2696 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2697 	vcpu->arch.busy_preempt = TB_NIL;
2698 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
2699 	++vc->n_runnable;
2700 
2701 	/*
2702 	 * This happens the first time this is called for a vcpu.
2703 	 * If the vcore is already running, we may be able to start
2704 	 * this thread straight away and have it join in.
2705 	 */
2706 	if (!signal_pending(current)) {
2707 		if (vc->vcore_state == VCORE_PIGGYBACK) {
2708 			struct kvmppc_vcore *mvc = vc->master_vcore;
2709 			if (spin_trylock(&mvc->lock)) {
2710 				if (mvc->vcore_state == VCORE_RUNNING &&
2711 				    !VCORE_IS_EXITING(mvc)) {
2712 					kvmppc_create_dtl_entry(vcpu, vc);
2713 					kvmppc_start_thread(vcpu, vc);
2714 					trace_kvm_guest_enter(vcpu);
2715 				}
2716 				spin_unlock(&mvc->lock);
2717 			}
2718 		} else if (vc->vcore_state == VCORE_RUNNING &&
2719 			   !VCORE_IS_EXITING(vc)) {
2720 			kvmppc_create_dtl_entry(vcpu, vc);
2721 			kvmppc_start_thread(vcpu, vc);
2722 			trace_kvm_guest_enter(vcpu);
2723 		} else if (vc->vcore_state == VCORE_SLEEPING) {
2724 			swake_up(&vc->wq);
2725 		}
2726 
2727 	}
2728 
2729 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2730 	       !signal_pending(current)) {
2731 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2732 			kvmppc_vcore_end_preempt(vc);
2733 
2734 		if (vc->vcore_state != VCORE_INACTIVE) {
2735 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2736 			continue;
2737 		}
2738 		for_each_runnable_thread(i, v, vc) {
2739 			kvmppc_core_prepare_to_enter(v);
2740 			if (signal_pending(v->arch.run_task)) {
2741 				kvmppc_remove_runnable(vc, v);
2742 				v->stat.signal_exits++;
2743 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2744 				v->arch.ret = -EINTR;
2745 				wake_up(&v->arch.cpu_run);
2746 			}
2747 		}
2748 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2749 			break;
2750 		n_ceded = 0;
2751 		for_each_runnable_thread(i, v, vc) {
2752 			if (!v->arch.pending_exceptions)
2753 				n_ceded += v->arch.ceded;
2754 			else
2755 				v->arch.ceded = 0;
2756 		}
2757 		vc->runner = vcpu;
2758 		if (n_ceded == vc->n_runnable) {
2759 			kvmppc_vcore_blocked(vc);
2760 		} else if (need_resched()) {
2761 			kvmppc_vcore_preempt(vc);
2762 			/* Let something else run */
2763 			cond_resched_lock(&vc->lock);
2764 			if (vc->vcore_state == VCORE_PREEMPT)
2765 				kvmppc_vcore_end_preempt(vc);
2766 		} else {
2767 			kvmppc_run_core(vc);
2768 		}
2769 		vc->runner = NULL;
2770 	}
2771 
2772 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2773 	       (vc->vcore_state == VCORE_RUNNING ||
2774 		vc->vcore_state == VCORE_EXITING ||
2775 		vc->vcore_state == VCORE_PIGGYBACK))
2776 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2777 
2778 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2779 		kvmppc_vcore_end_preempt(vc);
2780 
2781 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2782 		kvmppc_remove_runnable(vc, vcpu);
2783 		vcpu->stat.signal_exits++;
2784 		kvm_run->exit_reason = KVM_EXIT_INTR;
2785 		vcpu->arch.ret = -EINTR;
2786 	}
2787 
2788 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2789 		/* Wake up some vcpu to run the core */
2790 		i = -1;
2791 		v = next_runnable_thread(vc, &i);
2792 		wake_up(&v->arch.cpu_run);
2793 	}
2794 
2795 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2796 	spin_unlock(&vc->lock);
2797 	return vcpu->arch.ret;
2798 }
2799 
2800 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2801 {
2802 	int r;
2803 	int srcu_idx;
2804 
2805 	if (!vcpu->arch.sane) {
2806 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2807 		return -EINVAL;
2808 	}
2809 
2810 	kvmppc_core_prepare_to_enter(vcpu);
2811 
2812 	/* No need to go into the guest when all we'll do is come back out */
2813 	if (signal_pending(current)) {
2814 		run->exit_reason = KVM_EXIT_INTR;
2815 		return -EINTR;
2816 	}
2817 
2818 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2819 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2820 	smp_mb();
2821 
2822 	/* On the first time here, set up HTAB and VRMA */
2823 	if (!vcpu->kvm->arch.hpte_setup_done) {
2824 		r = kvmppc_hv_setup_htab_rma(vcpu);
2825 		if (r)
2826 			goto out;
2827 	}
2828 
2829 	flush_all_to_thread(current);
2830 
2831 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2832 	vcpu->arch.pgdir = current->mm->pgd;
2833 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2834 
2835 	do {
2836 		r = kvmppc_run_vcpu(run, vcpu);
2837 
2838 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2839 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2840 			trace_kvm_hcall_enter(vcpu);
2841 			r = kvmppc_pseries_do_hcall(vcpu);
2842 			trace_kvm_hcall_exit(vcpu, r);
2843 			kvmppc_core_prepare_to_enter(vcpu);
2844 		} else if (r == RESUME_PAGE_FAULT) {
2845 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2846 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2847 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2848 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2849 		} else if (r == RESUME_PASSTHROUGH)
2850 			r = kvmppc_xics_rm_complete(vcpu, 0);
2851 	} while (is_kvmppc_resume_guest(r));
2852 
2853  out:
2854 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2855 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2856 	return r;
2857 }
2858 
2859 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2860 				     int linux_psize)
2861 {
2862 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2863 
2864 	if (!def->shift)
2865 		return;
2866 	(*sps)->page_shift = def->shift;
2867 	(*sps)->slb_enc = def->sllp;
2868 	(*sps)->enc[0].page_shift = def->shift;
2869 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2870 	/*
2871 	 * Add 16MB MPSS support if host supports it
2872 	 */
2873 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2874 		(*sps)->enc[1].page_shift = 24;
2875 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2876 	}
2877 	(*sps)++;
2878 }
2879 
2880 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2881 					 struct kvm_ppc_smmu_info *info)
2882 {
2883 	struct kvm_ppc_one_seg_page_size *sps;
2884 
2885 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2886 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2887 		info->flags |= KVM_PPC_1T_SEGMENTS;
2888 	info->slb_size = mmu_slb_size;
2889 
2890 	/* We only support these sizes for now, and no muti-size segments */
2891 	sps = &info->sps[0];
2892 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2893 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2894 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2895 
2896 	return 0;
2897 }
2898 
2899 /*
2900  * Get (and clear) the dirty memory log for a memory slot.
2901  */
2902 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2903 					 struct kvm_dirty_log *log)
2904 {
2905 	struct kvm_memslots *slots;
2906 	struct kvm_memory_slot *memslot;
2907 	int r;
2908 	unsigned long n;
2909 
2910 	mutex_lock(&kvm->slots_lock);
2911 
2912 	r = -EINVAL;
2913 	if (log->slot >= KVM_USER_MEM_SLOTS)
2914 		goto out;
2915 
2916 	slots = kvm_memslots(kvm);
2917 	memslot = id_to_memslot(slots, log->slot);
2918 	r = -ENOENT;
2919 	if (!memslot->dirty_bitmap)
2920 		goto out;
2921 
2922 	n = kvm_dirty_bitmap_bytes(memslot);
2923 	memset(memslot->dirty_bitmap, 0, n);
2924 
2925 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2926 	if (r)
2927 		goto out;
2928 
2929 	r = -EFAULT;
2930 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2931 		goto out;
2932 
2933 	r = 0;
2934 out:
2935 	mutex_unlock(&kvm->slots_lock);
2936 	return r;
2937 }
2938 
2939 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2940 					struct kvm_memory_slot *dont)
2941 {
2942 	if (!dont || free->arch.rmap != dont->arch.rmap) {
2943 		vfree(free->arch.rmap);
2944 		free->arch.rmap = NULL;
2945 	}
2946 }
2947 
2948 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2949 					 unsigned long npages)
2950 {
2951 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2952 	if (!slot->arch.rmap)
2953 		return -ENOMEM;
2954 
2955 	return 0;
2956 }
2957 
2958 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2959 					struct kvm_memory_slot *memslot,
2960 					const struct kvm_userspace_memory_region *mem)
2961 {
2962 	return 0;
2963 }
2964 
2965 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2966 				const struct kvm_userspace_memory_region *mem,
2967 				const struct kvm_memory_slot *old,
2968 				const struct kvm_memory_slot *new)
2969 {
2970 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2971 	struct kvm_memslots *slots;
2972 	struct kvm_memory_slot *memslot;
2973 
2974 	if (npages && old->npages) {
2975 		/*
2976 		 * If modifying a memslot, reset all the rmap dirty bits.
2977 		 * If this is a new memslot, we don't need to do anything
2978 		 * since the rmap array starts out as all zeroes,
2979 		 * i.e. no pages are dirty.
2980 		 */
2981 		slots = kvm_memslots(kvm);
2982 		memslot = id_to_memslot(slots, mem->slot);
2983 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2984 	}
2985 }
2986 
2987 /*
2988  * Update LPCR values in kvm->arch and in vcores.
2989  * Caller must hold kvm->lock.
2990  */
2991 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2992 {
2993 	long int i;
2994 	u32 cores_done = 0;
2995 
2996 	if ((kvm->arch.lpcr & mask) == lpcr)
2997 		return;
2998 
2999 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3000 
3001 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
3002 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3003 		if (!vc)
3004 			continue;
3005 		spin_lock(&vc->lock);
3006 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3007 		spin_unlock(&vc->lock);
3008 		if (++cores_done >= kvm->arch.online_vcores)
3009 			break;
3010 	}
3011 }
3012 
3013 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3014 {
3015 	return;
3016 }
3017 
3018 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3019 {
3020 	int err = 0;
3021 	struct kvm *kvm = vcpu->kvm;
3022 	unsigned long hva;
3023 	struct kvm_memory_slot *memslot;
3024 	struct vm_area_struct *vma;
3025 	unsigned long lpcr = 0, senc;
3026 	unsigned long psize, porder;
3027 	int srcu_idx;
3028 
3029 	mutex_lock(&kvm->lock);
3030 	if (kvm->arch.hpte_setup_done)
3031 		goto out;	/* another vcpu beat us to it */
3032 
3033 	/* Allocate hashed page table (if not done already) and reset it */
3034 	if (!kvm->arch.hpt_virt) {
3035 		err = kvmppc_alloc_hpt(kvm, NULL);
3036 		if (err) {
3037 			pr_err("KVM: Couldn't alloc HPT\n");
3038 			goto out;
3039 		}
3040 	}
3041 
3042 	/* Look up the memslot for guest physical address 0 */
3043 	srcu_idx = srcu_read_lock(&kvm->srcu);
3044 	memslot = gfn_to_memslot(kvm, 0);
3045 
3046 	/* We must have some memory at 0 by now */
3047 	err = -EINVAL;
3048 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3049 		goto out_srcu;
3050 
3051 	/* Look up the VMA for the start of this memory slot */
3052 	hva = memslot->userspace_addr;
3053 	down_read(&current->mm->mmap_sem);
3054 	vma = find_vma(current->mm, hva);
3055 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3056 		goto up_out;
3057 
3058 	psize = vma_kernel_pagesize(vma);
3059 	porder = __ilog2(psize);
3060 
3061 	up_read(&current->mm->mmap_sem);
3062 
3063 	/* We can handle 4k, 64k or 16M pages in the VRMA */
3064 	err = -EINVAL;
3065 	if (!(psize == 0x1000 || psize == 0x10000 ||
3066 	      psize == 0x1000000))
3067 		goto out_srcu;
3068 
3069 	/* Update VRMASD field in the LPCR */
3070 	senc = slb_pgsize_encoding(psize);
3071 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3072 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3073 	/* the -4 is to account for senc values starting at 0x10 */
3074 	lpcr = senc << (LPCR_VRMASD_SH - 4);
3075 
3076 	/* Create HPTEs in the hash page table for the VRMA */
3077 	kvmppc_map_vrma(vcpu, memslot, porder);
3078 
3079 	kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3080 
3081 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3082 	smp_wmb();
3083 	kvm->arch.hpte_setup_done = 1;
3084 	err = 0;
3085  out_srcu:
3086 	srcu_read_unlock(&kvm->srcu, srcu_idx);
3087  out:
3088 	mutex_unlock(&kvm->lock);
3089 	return err;
3090 
3091  up_out:
3092 	up_read(&current->mm->mmap_sem);
3093 	goto out_srcu;
3094 }
3095 
3096 #ifdef CONFIG_KVM_XICS
3097 static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3098 			void *hcpu)
3099 {
3100 	unsigned long cpu = (long)hcpu;
3101 
3102 	switch (action) {
3103 	case CPU_UP_PREPARE:
3104 	case CPU_UP_PREPARE_FROZEN:
3105 		kvmppc_set_host_core(cpu);
3106 		break;
3107 
3108 #ifdef CONFIG_HOTPLUG_CPU
3109 	case CPU_DEAD:
3110 	case CPU_DEAD_FROZEN:
3111 	case CPU_UP_CANCELED:
3112 	case CPU_UP_CANCELED_FROZEN:
3113 		kvmppc_clear_host_core(cpu);
3114 		break;
3115 #endif
3116 	default:
3117 		break;
3118 	}
3119 
3120 	return NOTIFY_OK;
3121 }
3122 
3123 static struct notifier_block kvmppc_cpu_notifier = {
3124 	    .notifier_call = kvmppc_cpu_notify,
3125 };
3126 
3127 /*
3128  * Allocate a per-core structure for managing state about which cores are
3129  * running in the host versus the guest and for exchanging data between
3130  * real mode KVM and CPU running in the host.
3131  * This is only done for the first VM.
3132  * The allocated structure stays even if all VMs have stopped.
3133  * It is only freed when the kvm-hv module is unloaded.
3134  * It's OK for this routine to fail, we just don't support host
3135  * core operations like redirecting H_IPI wakeups.
3136  */
3137 void kvmppc_alloc_host_rm_ops(void)
3138 {
3139 	struct kvmppc_host_rm_ops *ops;
3140 	unsigned long l_ops;
3141 	int cpu, core;
3142 	int size;
3143 
3144 	/* Not the first time here ? */
3145 	if (kvmppc_host_rm_ops_hv != NULL)
3146 		return;
3147 
3148 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3149 	if (!ops)
3150 		return;
3151 
3152 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3153 	ops->rm_core = kzalloc(size, GFP_KERNEL);
3154 
3155 	if (!ops->rm_core) {
3156 		kfree(ops);
3157 		return;
3158 	}
3159 
3160 	get_online_cpus();
3161 
3162 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3163 		if (!cpu_online(cpu))
3164 			continue;
3165 
3166 		core = cpu >> threads_shift;
3167 		ops->rm_core[core].rm_state.in_host = 1;
3168 	}
3169 
3170 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3171 
3172 	/*
3173 	 * Make the contents of the kvmppc_host_rm_ops structure visible
3174 	 * to other CPUs before we assign it to the global variable.
3175 	 * Do an atomic assignment (no locks used here), but if someone
3176 	 * beats us to it, just free our copy and return.
3177 	 */
3178 	smp_wmb();
3179 	l_ops = (unsigned long) ops;
3180 
3181 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3182 		put_online_cpus();
3183 		kfree(ops->rm_core);
3184 		kfree(ops);
3185 		return;
3186 	}
3187 
3188 	register_cpu_notifier(&kvmppc_cpu_notifier);
3189 
3190 	put_online_cpus();
3191 }
3192 
3193 void kvmppc_free_host_rm_ops(void)
3194 {
3195 	if (kvmppc_host_rm_ops_hv) {
3196 		unregister_cpu_notifier(&kvmppc_cpu_notifier);
3197 		kfree(kvmppc_host_rm_ops_hv->rm_core);
3198 		kfree(kvmppc_host_rm_ops_hv);
3199 		kvmppc_host_rm_ops_hv = NULL;
3200 	}
3201 }
3202 #endif
3203 
3204 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3205 {
3206 	unsigned long lpcr, lpid;
3207 	char buf[32];
3208 
3209 	/* Allocate the guest's logical partition ID */
3210 
3211 	lpid = kvmppc_alloc_lpid();
3212 	if ((long)lpid < 0)
3213 		return -ENOMEM;
3214 	kvm->arch.lpid = lpid;
3215 
3216 	kvmppc_alloc_host_rm_ops();
3217 
3218 	/*
3219 	 * Since we don't flush the TLB when tearing down a VM,
3220 	 * and this lpid might have previously been used,
3221 	 * make sure we flush on each core before running the new VM.
3222 	 */
3223 	cpumask_setall(&kvm->arch.need_tlb_flush);
3224 
3225 	/* Start out with the default set of hcalls enabled */
3226 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3227 	       sizeof(kvm->arch.enabled_hcalls));
3228 
3229 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3230 
3231 	/* Init LPCR for virtual RMA mode */
3232 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
3233 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3234 	lpcr &= LPCR_PECE | LPCR_LPES;
3235 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3236 		LPCR_VPM0 | LPCR_VPM1;
3237 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3238 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3239 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3240 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3241 		lpcr |= LPCR_ONL;
3242 	kvm->arch.lpcr = lpcr;
3243 
3244 	/*
3245 	 * Track that we now have a HV mode VM active. This blocks secondary
3246 	 * CPU threads from coming online.
3247 	 */
3248 	kvm_hv_vm_activated();
3249 
3250 	/*
3251 	 * Create a debugfs directory for the VM
3252 	 */
3253 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3254 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3255 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3256 		kvmppc_mmu_debugfs_init(kvm);
3257 
3258 	return 0;
3259 }
3260 
3261 static void kvmppc_free_vcores(struct kvm *kvm)
3262 {
3263 	long int i;
3264 
3265 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3266 		kfree(kvm->arch.vcores[i]);
3267 	kvm->arch.online_vcores = 0;
3268 }
3269 
3270 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3271 {
3272 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3273 
3274 	kvm_hv_vm_deactivated();
3275 
3276 	kvmppc_free_vcores(kvm);
3277 
3278 	kvmppc_free_hpt(kvm);
3279 
3280 	kvmppc_free_pimap(kvm);
3281 }
3282 
3283 /* We don't need to emulate any privileged instructions or dcbz */
3284 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3285 				     unsigned int inst, int *advance)
3286 {
3287 	return EMULATE_FAIL;
3288 }
3289 
3290 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3291 					ulong spr_val)
3292 {
3293 	return EMULATE_FAIL;
3294 }
3295 
3296 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3297 					ulong *spr_val)
3298 {
3299 	return EMULATE_FAIL;
3300 }
3301 
3302 static int kvmppc_core_check_processor_compat_hv(void)
3303 {
3304 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3305 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3306 		return -EIO;
3307 	/*
3308 	 * Disable KVM for Power9, untill the required bits merged.
3309 	 */
3310 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3311 		return -EIO;
3312 
3313 	return 0;
3314 }
3315 
3316 #ifdef CONFIG_KVM_XICS
3317 
3318 void kvmppc_free_pimap(struct kvm *kvm)
3319 {
3320 	kfree(kvm->arch.pimap);
3321 }
3322 
3323 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3324 {
3325 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3326 }
3327 
3328 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3329 {
3330 	struct irq_desc *desc;
3331 	struct kvmppc_irq_map *irq_map;
3332 	struct kvmppc_passthru_irqmap *pimap;
3333 	struct irq_chip *chip;
3334 	int i;
3335 
3336 	if (!kvm_irq_bypass)
3337 		return 1;
3338 
3339 	desc = irq_to_desc(host_irq);
3340 	if (!desc)
3341 		return -EIO;
3342 
3343 	mutex_lock(&kvm->lock);
3344 
3345 	pimap = kvm->arch.pimap;
3346 	if (pimap == NULL) {
3347 		/* First call, allocate structure to hold IRQ map */
3348 		pimap = kvmppc_alloc_pimap();
3349 		if (pimap == NULL) {
3350 			mutex_unlock(&kvm->lock);
3351 			return -ENOMEM;
3352 		}
3353 		kvm->arch.pimap = pimap;
3354 	}
3355 
3356 	/*
3357 	 * For now, we only support interrupts for which the EOI operation
3358 	 * is an OPAL call followed by a write to XIRR, since that's
3359 	 * what our real-mode EOI code does.
3360 	 */
3361 	chip = irq_data_get_irq_chip(&desc->irq_data);
3362 	if (!chip || !is_pnv_opal_msi(chip)) {
3363 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3364 			host_irq, guest_gsi);
3365 		mutex_unlock(&kvm->lock);
3366 		return -ENOENT;
3367 	}
3368 
3369 	/*
3370 	 * See if we already have an entry for this guest IRQ number.
3371 	 * If it's mapped to a hardware IRQ number, that's an error,
3372 	 * otherwise re-use this entry.
3373 	 */
3374 	for (i = 0; i < pimap->n_mapped; i++) {
3375 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
3376 			if (pimap->mapped[i].r_hwirq) {
3377 				mutex_unlock(&kvm->lock);
3378 				return -EINVAL;
3379 			}
3380 			break;
3381 		}
3382 	}
3383 
3384 	if (i == KVMPPC_PIRQ_MAPPED) {
3385 		mutex_unlock(&kvm->lock);
3386 		return -EAGAIN;		/* table is full */
3387 	}
3388 
3389 	irq_map = &pimap->mapped[i];
3390 
3391 	irq_map->v_hwirq = guest_gsi;
3392 	irq_map->desc = desc;
3393 
3394 	/*
3395 	 * Order the above two stores before the next to serialize with
3396 	 * the KVM real mode handler.
3397 	 */
3398 	smp_wmb();
3399 	irq_map->r_hwirq = desc->irq_data.hwirq;
3400 
3401 	if (i == pimap->n_mapped)
3402 		pimap->n_mapped++;
3403 
3404 	kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3405 
3406 	mutex_unlock(&kvm->lock);
3407 
3408 	return 0;
3409 }
3410 
3411 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3412 {
3413 	struct irq_desc *desc;
3414 	struct kvmppc_passthru_irqmap *pimap;
3415 	int i;
3416 
3417 	if (!kvm_irq_bypass)
3418 		return 0;
3419 
3420 	desc = irq_to_desc(host_irq);
3421 	if (!desc)
3422 		return -EIO;
3423 
3424 	mutex_lock(&kvm->lock);
3425 
3426 	if (kvm->arch.pimap == NULL) {
3427 		mutex_unlock(&kvm->lock);
3428 		return 0;
3429 	}
3430 	pimap = kvm->arch.pimap;
3431 
3432 	for (i = 0; i < pimap->n_mapped; i++) {
3433 		if (guest_gsi == pimap->mapped[i].v_hwirq)
3434 			break;
3435 	}
3436 
3437 	if (i == pimap->n_mapped) {
3438 		mutex_unlock(&kvm->lock);
3439 		return -ENODEV;
3440 	}
3441 
3442 	kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
3443 
3444 	/* invalidate the entry */
3445 	pimap->mapped[i].r_hwirq = 0;
3446 
3447 	/*
3448 	 * We don't free this structure even when the count goes to
3449 	 * zero. The structure is freed when we destroy the VM.
3450 	 */
3451 
3452 	mutex_unlock(&kvm->lock);
3453 	return 0;
3454 }
3455 
3456 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
3457 					     struct irq_bypass_producer *prod)
3458 {
3459 	int ret = 0;
3460 	struct kvm_kernel_irqfd *irqfd =
3461 		container_of(cons, struct kvm_kernel_irqfd, consumer);
3462 
3463 	irqfd->producer = prod;
3464 
3465 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3466 	if (ret)
3467 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
3468 			prod->irq, irqfd->gsi, ret);
3469 
3470 	return ret;
3471 }
3472 
3473 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
3474 					      struct irq_bypass_producer *prod)
3475 {
3476 	int ret;
3477 	struct kvm_kernel_irqfd *irqfd =
3478 		container_of(cons, struct kvm_kernel_irqfd, consumer);
3479 
3480 	irqfd->producer = NULL;
3481 
3482 	/*
3483 	 * When producer of consumer is unregistered, we change back to
3484 	 * default external interrupt handling mode - KVM real mode
3485 	 * will switch back to host.
3486 	 */
3487 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3488 	if (ret)
3489 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
3490 			prod->irq, irqfd->gsi, ret);
3491 }
3492 #endif
3493 
3494 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3495 				 unsigned int ioctl, unsigned long arg)
3496 {
3497 	struct kvm *kvm __maybe_unused = filp->private_data;
3498 	void __user *argp = (void __user *)arg;
3499 	long r;
3500 
3501 	switch (ioctl) {
3502 
3503 	case KVM_PPC_ALLOCATE_HTAB: {
3504 		u32 htab_order;
3505 
3506 		r = -EFAULT;
3507 		if (get_user(htab_order, (u32 __user *)argp))
3508 			break;
3509 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3510 		if (r)
3511 			break;
3512 		r = -EFAULT;
3513 		if (put_user(htab_order, (u32 __user *)argp))
3514 			break;
3515 		r = 0;
3516 		break;
3517 	}
3518 
3519 	case KVM_PPC_GET_HTAB_FD: {
3520 		struct kvm_get_htab_fd ghf;
3521 
3522 		r = -EFAULT;
3523 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
3524 			break;
3525 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3526 		break;
3527 	}
3528 
3529 	default:
3530 		r = -ENOTTY;
3531 	}
3532 
3533 	return r;
3534 }
3535 
3536 /*
3537  * List of hcall numbers to enable by default.
3538  * For compatibility with old userspace, we enable by default
3539  * all hcalls that were implemented before the hcall-enabling
3540  * facility was added.  Note this list should not include H_RTAS.
3541  */
3542 static unsigned int default_hcall_list[] = {
3543 	H_REMOVE,
3544 	H_ENTER,
3545 	H_READ,
3546 	H_PROTECT,
3547 	H_BULK_REMOVE,
3548 	H_GET_TCE,
3549 	H_PUT_TCE,
3550 	H_SET_DABR,
3551 	H_SET_XDABR,
3552 	H_CEDE,
3553 	H_PROD,
3554 	H_CONFER,
3555 	H_REGISTER_VPA,
3556 #ifdef CONFIG_KVM_XICS
3557 	H_EOI,
3558 	H_CPPR,
3559 	H_IPI,
3560 	H_IPOLL,
3561 	H_XIRR,
3562 	H_XIRR_X,
3563 #endif
3564 	0
3565 };
3566 
3567 static void init_default_hcalls(void)
3568 {
3569 	int i;
3570 	unsigned int hcall;
3571 
3572 	for (i = 0; default_hcall_list[i]; ++i) {
3573 		hcall = default_hcall_list[i];
3574 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3575 		__set_bit(hcall / 4, default_enabled_hcalls);
3576 	}
3577 }
3578 
3579 static struct kvmppc_ops kvm_ops_hv = {
3580 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3581 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3582 	.get_one_reg = kvmppc_get_one_reg_hv,
3583 	.set_one_reg = kvmppc_set_one_reg_hv,
3584 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
3585 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
3586 	.set_msr     = kvmppc_set_msr_hv,
3587 	.vcpu_run    = kvmppc_vcpu_run_hv,
3588 	.vcpu_create = kvmppc_core_vcpu_create_hv,
3589 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
3590 	.check_requests = kvmppc_core_check_requests_hv,
3591 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3592 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
3593 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3594 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3595 	.unmap_hva = kvm_unmap_hva_hv,
3596 	.unmap_hva_range = kvm_unmap_hva_range_hv,
3597 	.age_hva  = kvm_age_hva_hv,
3598 	.test_age_hva = kvm_test_age_hva_hv,
3599 	.set_spte_hva = kvm_set_spte_hva_hv,
3600 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
3601 	.free_memslot = kvmppc_core_free_memslot_hv,
3602 	.create_memslot = kvmppc_core_create_memslot_hv,
3603 	.init_vm =  kvmppc_core_init_vm_hv,
3604 	.destroy_vm = kvmppc_core_destroy_vm_hv,
3605 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3606 	.emulate_op = kvmppc_core_emulate_op_hv,
3607 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3608 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3609 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3610 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3611 	.hcall_implemented = kvmppc_hcall_impl_hv,
3612 #ifdef CONFIG_KVM_XICS
3613 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
3614 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
3615 #endif
3616 };
3617 
3618 static int kvm_init_subcore_bitmap(void)
3619 {
3620 	int i, j;
3621 	int nr_cores = cpu_nr_cores();
3622 	struct sibling_subcore_state *sibling_subcore_state;
3623 
3624 	for (i = 0; i < nr_cores; i++) {
3625 		int first_cpu = i * threads_per_core;
3626 		int node = cpu_to_node(first_cpu);
3627 
3628 		/* Ignore if it is already allocated. */
3629 		if (paca[first_cpu].sibling_subcore_state)
3630 			continue;
3631 
3632 		sibling_subcore_state =
3633 			kmalloc_node(sizeof(struct sibling_subcore_state),
3634 							GFP_KERNEL, node);
3635 		if (!sibling_subcore_state)
3636 			return -ENOMEM;
3637 
3638 		memset(sibling_subcore_state, 0,
3639 				sizeof(struct sibling_subcore_state));
3640 
3641 		for (j = 0; j < threads_per_core; j++) {
3642 			int cpu = first_cpu + j;
3643 
3644 			paca[cpu].sibling_subcore_state = sibling_subcore_state;
3645 		}
3646 	}
3647 	return 0;
3648 }
3649 
3650 static int kvmppc_book3s_init_hv(void)
3651 {
3652 	int r;
3653 	/*
3654 	 * FIXME!! Do we need to check on all cpus ?
3655 	 */
3656 	r = kvmppc_core_check_processor_compat_hv();
3657 	if (r < 0)
3658 		return -ENODEV;
3659 
3660 	r = kvm_init_subcore_bitmap();
3661 	if (r)
3662 		return r;
3663 
3664 	kvm_ops_hv.owner = THIS_MODULE;
3665 	kvmppc_hv_ops = &kvm_ops_hv;
3666 
3667 	init_default_hcalls();
3668 
3669 	init_vcore_lists();
3670 
3671 	r = kvmppc_mmu_hv_init();
3672 	return r;
3673 }
3674 
3675 static void kvmppc_book3s_exit_hv(void)
3676 {
3677 	kvmppc_free_host_rm_ops();
3678 	kvmppc_hv_ops = NULL;
3679 }
3680 
3681 module_init(kvmppc_book3s_init_hv);
3682 module_exit(kvmppc_book3s_exit_hv);
3683 MODULE_LICENSE("GPL");
3684 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3685 MODULE_ALIAS("devname:kvm");
3686