xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 9cfc5c90)
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/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 
37 #include <asm/reg.h>
38 #include <asm/cputable.h>
39 #include <asm/cacheflush.h>
40 #include <asm/tlbflush.h>
41 #include <asm/uaccess.h>
42 #include <asm/io.h>
43 #include <asm/kvm_ppc.h>
44 #include <asm/kvm_book3s.h>
45 #include <asm/mmu_context.h>
46 #include <asm/lppaca.h>
47 #include <asm/processor.h>
48 #include <asm/cputhreads.h>
49 #include <asm/page.h>
50 #include <asm/hvcall.h>
51 #include <asm/switch_to.h>
52 #include <asm/smp.h>
53 #include <asm/dbell.h>
54 #include <linux/gfp.h>
55 #include <linux/vmalloc.h>
56 #include <linux/highmem.h>
57 #include <linux/hugetlb.h>
58 #include <linux/module.h>
59 
60 #include "book3s.h"
61 
62 #define CREATE_TRACE_POINTS
63 #include "trace_hv.h"
64 
65 /* #define EXIT_DEBUG */
66 /* #define EXIT_DEBUG_SIMPLE */
67 /* #define EXIT_DEBUG_INT */
68 
69 /* Used to indicate that a guest page fault needs to be handled */
70 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
71 
72 /* Used as a "null" value for timebase values */
73 #define TB_NIL	(~(u64)0)
74 
75 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
76 
77 static int dynamic_mt_modes = 6;
78 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
79 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
80 static int target_smt_mode;
81 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
82 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
83 
84 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
85 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
86 
87 static bool kvmppc_ipi_thread(int cpu)
88 {
89 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
90 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
91 		preempt_disable();
92 		if (cpu_first_thread_sibling(cpu) ==
93 		    cpu_first_thread_sibling(smp_processor_id())) {
94 			unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
95 			msg |= cpu_thread_in_core(cpu);
96 			smp_mb();
97 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
98 			preempt_enable();
99 			return true;
100 		}
101 		preempt_enable();
102 	}
103 
104 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
105 	if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
106 		xics_wake_cpu(cpu);
107 		return true;
108 	}
109 #endif
110 
111 	return false;
112 }
113 
114 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
115 {
116 	int cpu;
117 	wait_queue_head_t *wqp;
118 
119 	wqp = kvm_arch_vcpu_wq(vcpu);
120 	if (waitqueue_active(wqp)) {
121 		wake_up_interruptible(wqp);
122 		++vcpu->stat.halt_wakeup;
123 	}
124 
125 	if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
126 		return;
127 
128 	/* CPU points to the first thread of the core */
129 	cpu = vcpu->cpu;
130 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
131 		smp_send_reschedule(cpu);
132 }
133 
134 /*
135  * We use the vcpu_load/put functions to measure stolen time.
136  * Stolen time is counted as time when either the vcpu is able to
137  * run as part of a virtual core, but the task running the vcore
138  * is preempted or sleeping, or when the vcpu needs something done
139  * in the kernel by the task running the vcpu, but that task is
140  * preempted or sleeping.  Those two things have to be counted
141  * separately, since one of the vcpu tasks will take on the job
142  * of running the core, and the other vcpu tasks in the vcore will
143  * sleep waiting for it to do that, but that sleep shouldn't count
144  * as stolen time.
145  *
146  * Hence we accumulate stolen time when the vcpu can run as part of
147  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
148  * needs its task to do other things in the kernel (for example,
149  * service a page fault) in busy_stolen.  We don't accumulate
150  * stolen time for a vcore when it is inactive, or for a vcpu
151  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
152  * a misnomer; it means that the vcpu task is not executing in
153  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
154  * the kernel.  We don't have any way of dividing up that time
155  * between time that the vcpu is genuinely stopped, time that
156  * the task is actively working on behalf of the vcpu, and time
157  * that the task is preempted, so we don't count any of it as
158  * stolen.
159  *
160  * Updates to busy_stolen are protected by arch.tbacct_lock;
161  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
162  * lock.  The stolen times are measured in units of timebase ticks.
163  * (Note that the != TB_NIL checks below are purely defensive;
164  * they should never fail.)
165  */
166 
167 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
168 {
169 	unsigned long flags;
170 
171 	spin_lock_irqsave(&vc->stoltb_lock, flags);
172 	vc->preempt_tb = mftb();
173 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
174 }
175 
176 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
177 {
178 	unsigned long flags;
179 
180 	spin_lock_irqsave(&vc->stoltb_lock, flags);
181 	if (vc->preempt_tb != TB_NIL) {
182 		vc->stolen_tb += mftb() - vc->preempt_tb;
183 		vc->preempt_tb = TB_NIL;
184 	}
185 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
186 }
187 
188 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
189 {
190 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
191 	unsigned long flags;
192 
193 	/*
194 	 * We can test vc->runner without taking the vcore lock,
195 	 * because only this task ever sets vc->runner to this
196 	 * vcpu, and once it is set to this vcpu, only this task
197 	 * ever sets it to NULL.
198 	 */
199 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
200 		kvmppc_core_end_stolen(vc);
201 
202 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
203 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
204 	    vcpu->arch.busy_preempt != TB_NIL) {
205 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
206 		vcpu->arch.busy_preempt = TB_NIL;
207 	}
208 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
209 }
210 
211 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
212 {
213 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
214 	unsigned long flags;
215 
216 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
217 		kvmppc_core_start_stolen(vc);
218 
219 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
220 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
221 		vcpu->arch.busy_preempt = mftb();
222 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
223 }
224 
225 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
226 {
227 	vcpu->arch.shregs.msr = msr;
228 	kvmppc_end_cede(vcpu);
229 }
230 
231 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
232 {
233 	vcpu->arch.pvr = pvr;
234 }
235 
236 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
237 {
238 	unsigned long pcr = 0;
239 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
240 
241 	if (arch_compat) {
242 		switch (arch_compat) {
243 		case PVR_ARCH_205:
244 			/*
245 			 * If an arch bit is set in PCR, all the defined
246 			 * higher-order arch bits also have to be set.
247 			 */
248 			pcr = PCR_ARCH_206 | PCR_ARCH_205;
249 			break;
250 		case PVR_ARCH_206:
251 		case PVR_ARCH_206p:
252 			pcr = PCR_ARCH_206;
253 			break;
254 		case PVR_ARCH_207:
255 			break;
256 		default:
257 			return -EINVAL;
258 		}
259 
260 		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
261 			/* POWER7 can't emulate POWER8 */
262 			if (!(pcr & PCR_ARCH_206))
263 				return -EINVAL;
264 			pcr &= ~PCR_ARCH_206;
265 		}
266 	}
267 
268 	spin_lock(&vc->lock);
269 	vc->arch_compat = arch_compat;
270 	vc->pcr = pcr;
271 	spin_unlock(&vc->lock);
272 
273 	return 0;
274 }
275 
276 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
277 {
278 	int r;
279 
280 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
281 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
282 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
283 	for (r = 0; r < 16; ++r)
284 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
285 		       r, kvmppc_get_gpr(vcpu, r),
286 		       r+16, kvmppc_get_gpr(vcpu, r+16));
287 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
288 	       vcpu->arch.ctr, vcpu->arch.lr);
289 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
290 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
291 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
292 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
293 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
294 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
295 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
296 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
297 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
298 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
299 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
300 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
301 	for (r = 0; r < vcpu->arch.slb_max; ++r)
302 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
303 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
304 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
305 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
306 	       vcpu->arch.last_inst);
307 }
308 
309 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
310 {
311 	int r;
312 	struct kvm_vcpu *v, *ret = NULL;
313 
314 	mutex_lock(&kvm->lock);
315 	kvm_for_each_vcpu(r, v, kvm) {
316 		if (v->vcpu_id == id) {
317 			ret = v;
318 			break;
319 		}
320 	}
321 	mutex_unlock(&kvm->lock);
322 	return ret;
323 }
324 
325 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
326 {
327 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
328 	vpa->yield_count = cpu_to_be32(1);
329 }
330 
331 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
332 		   unsigned long addr, unsigned long len)
333 {
334 	/* check address is cacheline aligned */
335 	if (addr & (L1_CACHE_BYTES - 1))
336 		return -EINVAL;
337 	spin_lock(&vcpu->arch.vpa_update_lock);
338 	if (v->next_gpa != addr || v->len != len) {
339 		v->next_gpa = addr;
340 		v->len = addr ? len : 0;
341 		v->update_pending = 1;
342 	}
343 	spin_unlock(&vcpu->arch.vpa_update_lock);
344 	return 0;
345 }
346 
347 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
348 struct reg_vpa {
349 	u32 dummy;
350 	union {
351 		__be16 hword;
352 		__be32 word;
353 	} length;
354 };
355 
356 static int vpa_is_registered(struct kvmppc_vpa *vpap)
357 {
358 	if (vpap->update_pending)
359 		return vpap->next_gpa != 0;
360 	return vpap->pinned_addr != NULL;
361 }
362 
363 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
364 				       unsigned long flags,
365 				       unsigned long vcpuid, unsigned long vpa)
366 {
367 	struct kvm *kvm = vcpu->kvm;
368 	unsigned long len, nb;
369 	void *va;
370 	struct kvm_vcpu *tvcpu;
371 	int err;
372 	int subfunc;
373 	struct kvmppc_vpa *vpap;
374 
375 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
376 	if (!tvcpu)
377 		return H_PARAMETER;
378 
379 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
380 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
381 	    subfunc == H_VPA_REG_SLB) {
382 		/* Registering new area - address must be cache-line aligned */
383 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
384 			return H_PARAMETER;
385 
386 		/* convert logical addr to kernel addr and read length */
387 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
388 		if (va == NULL)
389 			return H_PARAMETER;
390 		if (subfunc == H_VPA_REG_VPA)
391 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
392 		else
393 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
394 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
395 
396 		/* Check length */
397 		if (len > nb || len < sizeof(struct reg_vpa))
398 			return H_PARAMETER;
399 	} else {
400 		vpa = 0;
401 		len = 0;
402 	}
403 
404 	err = H_PARAMETER;
405 	vpap = NULL;
406 	spin_lock(&tvcpu->arch.vpa_update_lock);
407 
408 	switch (subfunc) {
409 	case H_VPA_REG_VPA:		/* register VPA */
410 		if (len < sizeof(struct lppaca))
411 			break;
412 		vpap = &tvcpu->arch.vpa;
413 		err = 0;
414 		break;
415 
416 	case H_VPA_REG_DTL:		/* register DTL */
417 		if (len < sizeof(struct dtl_entry))
418 			break;
419 		len -= len % sizeof(struct dtl_entry);
420 
421 		/* Check that they have previously registered a VPA */
422 		err = H_RESOURCE;
423 		if (!vpa_is_registered(&tvcpu->arch.vpa))
424 			break;
425 
426 		vpap = &tvcpu->arch.dtl;
427 		err = 0;
428 		break;
429 
430 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
431 		/* Check that they have previously registered a VPA */
432 		err = H_RESOURCE;
433 		if (!vpa_is_registered(&tvcpu->arch.vpa))
434 			break;
435 
436 		vpap = &tvcpu->arch.slb_shadow;
437 		err = 0;
438 		break;
439 
440 	case H_VPA_DEREG_VPA:		/* deregister VPA */
441 		/* Check they don't still have a DTL or SLB buf registered */
442 		err = H_RESOURCE;
443 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
444 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
445 			break;
446 
447 		vpap = &tvcpu->arch.vpa;
448 		err = 0;
449 		break;
450 
451 	case H_VPA_DEREG_DTL:		/* deregister DTL */
452 		vpap = &tvcpu->arch.dtl;
453 		err = 0;
454 		break;
455 
456 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
457 		vpap = &tvcpu->arch.slb_shadow;
458 		err = 0;
459 		break;
460 	}
461 
462 	if (vpap) {
463 		vpap->next_gpa = vpa;
464 		vpap->len = len;
465 		vpap->update_pending = 1;
466 	}
467 
468 	spin_unlock(&tvcpu->arch.vpa_update_lock);
469 
470 	return err;
471 }
472 
473 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
474 {
475 	struct kvm *kvm = vcpu->kvm;
476 	void *va;
477 	unsigned long nb;
478 	unsigned long gpa;
479 
480 	/*
481 	 * We need to pin the page pointed to by vpap->next_gpa,
482 	 * but we can't call kvmppc_pin_guest_page under the lock
483 	 * as it does get_user_pages() and down_read().  So we
484 	 * have to drop the lock, pin the page, then get the lock
485 	 * again and check that a new area didn't get registered
486 	 * in the meantime.
487 	 */
488 	for (;;) {
489 		gpa = vpap->next_gpa;
490 		spin_unlock(&vcpu->arch.vpa_update_lock);
491 		va = NULL;
492 		nb = 0;
493 		if (gpa)
494 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
495 		spin_lock(&vcpu->arch.vpa_update_lock);
496 		if (gpa == vpap->next_gpa)
497 			break;
498 		/* sigh... unpin that one and try again */
499 		if (va)
500 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
501 	}
502 
503 	vpap->update_pending = 0;
504 	if (va && nb < vpap->len) {
505 		/*
506 		 * If it's now too short, it must be that userspace
507 		 * has changed the mappings underlying guest memory,
508 		 * so unregister the region.
509 		 */
510 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
511 		va = NULL;
512 	}
513 	if (vpap->pinned_addr)
514 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
515 					vpap->dirty);
516 	vpap->gpa = gpa;
517 	vpap->pinned_addr = va;
518 	vpap->dirty = false;
519 	if (va)
520 		vpap->pinned_end = va + vpap->len;
521 }
522 
523 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
524 {
525 	if (!(vcpu->arch.vpa.update_pending ||
526 	      vcpu->arch.slb_shadow.update_pending ||
527 	      vcpu->arch.dtl.update_pending))
528 		return;
529 
530 	spin_lock(&vcpu->arch.vpa_update_lock);
531 	if (vcpu->arch.vpa.update_pending) {
532 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
533 		if (vcpu->arch.vpa.pinned_addr)
534 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
535 	}
536 	if (vcpu->arch.dtl.update_pending) {
537 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
538 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
539 		vcpu->arch.dtl_index = 0;
540 	}
541 	if (vcpu->arch.slb_shadow.update_pending)
542 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
543 	spin_unlock(&vcpu->arch.vpa_update_lock);
544 }
545 
546 /*
547  * Return the accumulated stolen time for the vcore up until `now'.
548  * The caller should hold the vcore lock.
549  */
550 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
551 {
552 	u64 p;
553 	unsigned long flags;
554 
555 	spin_lock_irqsave(&vc->stoltb_lock, flags);
556 	p = vc->stolen_tb;
557 	if (vc->vcore_state != VCORE_INACTIVE &&
558 	    vc->preempt_tb != TB_NIL)
559 		p += now - vc->preempt_tb;
560 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
561 	return p;
562 }
563 
564 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
565 				    struct kvmppc_vcore *vc)
566 {
567 	struct dtl_entry *dt;
568 	struct lppaca *vpa;
569 	unsigned long stolen;
570 	unsigned long core_stolen;
571 	u64 now;
572 
573 	dt = vcpu->arch.dtl_ptr;
574 	vpa = vcpu->arch.vpa.pinned_addr;
575 	now = mftb();
576 	core_stolen = vcore_stolen_time(vc, now);
577 	stolen = core_stolen - vcpu->arch.stolen_logged;
578 	vcpu->arch.stolen_logged = core_stolen;
579 	spin_lock_irq(&vcpu->arch.tbacct_lock);
580 	stolen += vcpu->arch.busy_stolen;
581 	vcpu->arch.busy_stolen = 0;
582 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
583 	if (!dt || !vpa)
584 		return;
585 	memset(dt, 0, sizeof(struct dtl_entry));
586 	dt->dispatch_reason = 7;
587 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
588 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
589 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
590 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
591 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
592 	++dt;
593 	if (dt == vcpu->arch.dtl.pinned_end)
594 		dt = vcpu->arch.dtl.pinned_addr;
595 	vcpu->arch.dtl_ptr = dt;
596 	/* order writing *dt vs. writing vpa->dtl_idx */
597 	smp_wmb();
598 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
599 	vcpu->arch.dtl.dirty = true;
600 }
601 
602 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
603 {
604 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
605 		return true;
606 	if ((!vcpu->arch.vcore->arch_compat) &&
607 	    cpu_has_feature(CPU_FTR_ARCH_207S))
608 		return true;
609 	return false;
610 }
611 
612 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
613 			     unsigned long resource, unsigned long value1,
614 			     unsigned long value2)
615 {
616 	switch (resource) {
617 	case H_SET_MODE_RESOURCE_SET_CIABR:
618 		if (!kvmppc_power8_compatible(vcpu))
619 			return H_P2;
620 		if (value2)
621 			return H_P4;
622 		if (mflags)
623 			return H_UNSUPPORTED_FLAG_START;
624 		/* Guests can't breakpoint the hypervisor */
625 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
626 			return H_P3;
627 		vcpu->arch.ciabr  = value1;
628 		return H_SUCCESS;
629 	case H_SET_MODE_RESOURCE_SET_DAWR:
630 		if (!kvmppc_power8_compatible(vcpu))
631 			return H_P2;
632 		if (mflags)
633 			return H_UNSUPPORTED_FLAG_START;
634 		if (value2 & DABRX_HYP)
635 			return H_P4;
636 		vcpu->arch.dawr  = value1;
637 		vcpu->arch.dawrx = value2;
638 		return H_SUCCESS;
639 	default:
640 		return H_TOO_HARD;
641 	}
642 }
643 
644 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
645 {
646 	struct kvmppc_vcore *vcore = target->arch.vcore;
647 
648 	/*
649 	 * We expect to have been called by the real mode handler
650 	 * (kvmppc_rm_h_confer()) which would have directly returned
651 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
652 	 * have useful work to do and should not confer) so we don't
653 	 * recheck that here.
654 	 */
655 
656 	spin_lock(&vcore->lock);
657 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
658 	    vcore->vcore_state != VCORE_INACTIVE &&
659 	    vcore->runner)
660 		target = vcore->runner;
661 	spin_unlock(&vcore->lock);
662 
663 	return kvm_vcpu_yield_to(target);
664 }
665 
666 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
667 {
668 	int yield_count = 0;
669 	struct lppaca *lppaca;
670 
671 	spin_lock(&vcpu->arch.vpa_update_lock);
672 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
673 	if (lppaca)
674 		yield_count = be32_to_cpu(lppaca->yield_count);
675 	spin_unlock(&vcpu->arch.vpa_update_lock);
676 	return yield_count;
677 }
678 
679 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
680 {
681 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
682 	unsigned long target, ret = H_SUCCESS;
683 	int yield_count;
684 	struct kvm_vcpu *tvcpu;
685 	int idx, rc;
686 
687 	if (req <= MAX_HCALL_OPCODE &&
688 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
689 		return RESUME_HOST;
690 
691 	switch (req) {
692 	case H_CEDE:
693 		break;
694 	case H_PROD:
695 		target = kvmppc_get_gpr(vcpu, 4);
696 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
697 		if (!tvcpu) {
698 			ret = H_PARAMETER;
699 			break;
700 		}
701 		tvcpu->arch.prodded = 1;
702 		smp_mb();
703 		if (vcpu->arch.ceded) {
704 			if (waitqueue_active(&vcpu->wq)) {
705 				wake_up_interruptible(&vcpu->wq);
706 				vcpu->stat.halt_wakeup++;
707 			}
708 		}
709 		break;
710 	case H_CONFER:
711 		target = kvmppc_get_gpr(vcpu, 4);
712 		if (target == -1)
713 			break;
714 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
715 		if (!tvcpu) {
716 			ret = H_PARAMETER;
717 			break;
718 		}
719 		yield_count = kvmppc_get_gpr(vcpu, 5);
720 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
721 			break;
722 		kvm_arch_vcpu_yield_to(tvcpu);
723 		break;
724 	case H_REGISTER_VPA:
725 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
726 					kvmppc_get_gpr(vcpu, 5),
727 					kvmppc_get_gpr(vcpu, 6));
728 		break;
729 	case H_RTAS:
730 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
731 			return RESUME_HOST;
732 
733 		idx = srcu_read_lock(&vcpu->kvm->srcu);
734 		rc = kvmppc_rtas_hcall(vcpu);
735 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
736 
737 		if (rc == -ENOENT)
738 			return RESUME_HOST;
739 		else if (rc == 0)
740 			break;
741 
742 		/* Send the error out to userspace via KVM_RUN */
743 		return rc;
744 	case H_LOGICAL_CI_LOAD:
745 		ret = kvmppc_h_logical_ci_load(vcpu);
746 		if (ret == H_TOO_HARD)
747 			return RESUME_HOST;
748 		break;
749 	case H_LOGICAL_CI_STORE:
750 		ret = kvmppc_h_logical_ci_store(vcpu);
751 		if (ret == H_TOO_HARD)
752 			return RESUME_HOST;
753 		break;
754 	case H_SET_MODE:
755 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
756 					kvmppc_get_gpr(vcpu, 5),
757 					kvmppc_get_gpr(vcpu, 6),
758 					kvmppc_get_gpr(vcpu, 7));
759 		if (ret == H_TOO_HARD)
760 			return RESUME_HOST;
761 		break;
762 	case H_XIRR:
763 	case H_CPPR:
764 	case H_EOI:
765 	case H_IPI:
766 	case H_IPOLL:
767 	case H_XIRR_X:
768 		if (kvmppc_xics_enabled(vcpu)) {
769 			ret = kvmppc_xics_hcall(vcpu, req);
770 			break;
771 		} /* fallthrough */
772 	default:
773 		return RESUME_HOST;
774 	}
775 	kvmppc_set_gpr(vcpu, 3, ret);
776 	vcpu->arch.hcall_needed = 0;
777 	return RESUME_GUEST;
778 }
779 
780 static int kvmppc_hcall_impl_hv(unsigned long cmd)
781 {
782 	switch (cmd) {
783 	case H_CEDE:
784 	case H_PROD:
785 	case H_CONFER:
786 	case H_REGISTER_VPA:
787 	case H_SET_MODE:
788 	case H_LOGICAL_CI_LOAD:
789 	case H_LOGICAL_CI_STORE:
790 #ifdef CONFIG_KVM_XICS
791 	case H_XIRR:
792 	case H_CPPR:
793 	case H_EOI:
794 	case H_IPI:
795 	case H_IPOLL:
796 	case H_XIRR_X:
797 #endif
798 		return 1;
799 	}
800 
801 	/* See if it's in the real-mode table */
802 	return kvmppc_hcall_impl_hv_realmode(cmd);
803 }
804 
805 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
806 					struct kvm_vcpu *vcpu)
807 {
808 	u32 last_inst;
809 
810 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
811 					EMULATE_DONE) {
812 		/*
813 		 * Fetch failed, so return to guest and
814 		 * try executing it again.
815 		 */
816 		return RESUME_GUEST;
817 	}
818 
819 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
820 		run->exit_reason = KVM_EXIT_DEBUG;
821 		run->debug.arch.address = kvmppc_get_pc(vcpu);
822 		return RESUME_HOST;
823 	} else {
824 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
825 		return RESUME_GUEST;
826 	}
827 }
828 
829 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
830 				 struct task_struct *tsk)
831 {
832 	int r = RESUME_HOST;
833 
834 	vcpu->stat.sum_exits++;
835 
836 	run->exit_reason = KVM_EXIT_UNKNOWN;
837 	run->ready_for_interrupt_injection = 1;
838 	switch (vcpu->arch.trap) {
839 	/* We're good on these - the host merely wanted to get our attention */
840 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
841 		vcpu->stat.dec_exits++;
842 		r = RESUME_GUEST;
843 		break;
844 	case BOOK3S_INTERRUPT_EXTERNAL:
845 	case BOOK3S_INTERRUPT_H_DOORBELL:
846 		vcpu->stat.ext_intr_exits++;
847 		r = RESUME_GUEST;
848 		break;
849 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
850 	case BOOK3S_INTERRUPT_HMI:
851 	case BOOK3S_INTERRUPT_PERFMON:
852 		r = RESUME_GUEST;
853 		break;
854 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
855 		/*
856 		 * Deliver a machine check interrupt to the guest.
857 		 * We have to do this, even if the host has handled the
858 		 * machine check, because machine checks use SRR0/1 and
859 		 * the interrupt might have trashed guest state in them.
860 		 */
861 		kvmppc_book3s_queue_irqprio(vcpu,
862 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
863 		r = RESUME_GUEST;
864 		break;
865 	case BOOK3S_INTERRUPT_PROGRAM:
866 	{
867 		ulong flags;
868 		/*
869 		 * Normally program interrupts are delivered directly
870 		 * to the guest by the hardware, but we can get here
871 		 * as a result of a hypervisor emulation interrupt
872 		 * (e40) getting turned into a 700 by BML RTAS.
873 		 */
874 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
875 		kvmppc_core_queue_program(vcpu, flags);
876 		r = RESUME_GUEST;
877 		break;
878 	}
879 	case BOOK3S_INTERRUPT_SYSCALL:
880 	{
881 		/* hcall - punt to userspace */
882 		int i;
883 
884 		/* hypercall with MSR_PR has already been handled in rmode,
885 		 * and never reaches here.
886 		 */
887 
888 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
889 		for (i = 0; i < 9; ++i)
890 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
891 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
892 		vcpu->arch.hcall_needed = 1;
893 		r = RESUME_HOST;
894 		break;
895 	}
896 	/*
897 	 * We get these next two if the guest accesses a page which it thinks
898 	 * it has mapped but which is not actually present, either because
899 	 * it is for an emulated I/O device or because the corresonding
900 	 * host page has been paged out.  Any other HDSI/HISI interrupts
901 	 * have been handled already.
902 	 */
903 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
904 		r = RESUME_PAGE_FAULT;
905 		break;
906 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
907 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
908 		vcpu->arch.fault_dsisr = 0;
909 		r = RESUME_PAGE_FAULT;
910 		break;
911 	/*
912 	 * This occurs if the guest executes an illegal instruction.
913 	 * If the guest debug is disabled, generate a program interrupt
914 	 * to the guest. If guest debug is enabled, we need to check
915 	 * whether the instruction is a software breakpoint instruction.
916 	 * Accordingly return to Guest or Host.
917 	 */
918 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
919 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
920 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
921 				swab32(vcpu->arch.emul_inst) :
922 				vcpu->arch.emul_inst;
923 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
924 			r = kvmppc_emulate_debug_inst(run, vcpu);
925 		} else {
926 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
927 			r = RESUME_GUEST;
928 		}
929 		break;
930 	/*
931 	 * This occurs if the guest (kernel or userspace), does something that
932 	 * is prohibited by HFSCR.  We just generate a program interrupt to
933 	 * the guest.
934 	 */
935 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
936 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
937 		r = RESUME_GUEST;
938 		break;
939 	default:
940 		kvmppc_dump_regs(vcpu);
941 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
942 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
943 			vcpu->arch.shregs.msr);
944 		run->hw.hardware_exit_reason = vcpu->arch.trap;
945 		r = RESUME_HOST;
946 		break;
947 	}
948 
949 	return r;
950 }
951 
952 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
953 					    struct kvm_sregs *sregs)
954 {
955 	int i;
956 
957 	memset(sregs, 0, sizeof(struct kvm_sregs));
958 	sregs->pvr = vcpu->arch.pvr;
959 	for (i = 0; i < vcpu->arch.slb_max; i++) {
960 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
961 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
962 	}
963 
964 	return 0;
965 }
966 
967 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
968 					    struct kvm_sregs *sregs)
969 {
970 	int i, j;
971 
972 	/* Only accept the same PVR as the host's, since we can't spoof it */
973 	if (sregs->pvr != vcpu->arch.pvr)
974 		return -EINVAL;
975 
976 	j = 0;
977 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
978 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
979 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
980 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
981 			++j;
982 		}
983 	}
984 	vcpu->arch.slb_max = j;
985 
986 	return 0;
987 }
988 
989 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
990 		bool preserve_top32)
991 {
992 	struct kvm *kvm = vcpu->kvm;
993 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
994 	u64 mask;
995 
996 	mutex_lock(&kvm->lock);
997 	spin_lock(&vc->lock);
998 	/*
999 	 * If ILE (interrupt little-endian) has changed, update the
1000 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1001 	 */
1002 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1003 		struct kvm_vcpu *vcpu;
1004 		int i;
1005 
1006 		kvm_for_each_vcpu(i, vcpu, kvm) {
1007 			if (vcpu->arch.vcore != vc)
1008 				continue;
1009 			if (new_lpcr & LPCR_ILE)
1010 				vcpu->arch.intr_msr |= MSR_LE;
1011 			else
1012 				vcpu->arch.intr_msr &= ~MSR_LE;
1013 		}
1014 	}
1015 
1016 	/*
1017 	 * Userspace can only modify DPFD (default prefetch depth),
1018 	 * ILE (interrupt little-endian) and TC (translation control).
1019 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1020 	 */
1021 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1022 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1023 		mask |= LPCR_AIL;
1024 
1025 	/* Broken 32-bit version of LPCR must not clear top bits */
1026 	if (preserve_top32)
1027 		mask &= 0xFFFFFFFF;
1028 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1029 	spin_unlock(&vc->lock);
1030 	mutex_unlock(&kvm->lock);
1031 }
1032 
1033 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1034 				 union kvmppc_one_reg *val)
1035 {
1036 	int r = 0;
1037 	long int i;
1038 
1039 	switch (id) {
1040 	case KVM_REG_PPC_DEBUG_INST:
1041 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1042 		break;
1043 	case KVM_REG_PPC_HIOR:
1044 		*val = get_reg_val(id, 0);
1045 		break;
1046 	case KVM_REG_PPC_DABR:
1047 		*val = get_reg_val(id, vcpu->arch.dabr);
1048 		break;
1049 	case KVM_REG_PPC_DABRX:
1050 		*val = get_reg_val(id, vcpu->arch.dabrx);
1051 		break;
1052 	case KVM_REG_PPC_DSCR:
1053 		*val = get_reg_val(id, vcpu->arch.dscr);
1054 		break;
1055 	case KVM_REG_PPC_PURR:
1056 		*val = get_reg_val(id, vcpu->arch.purr);
1057 		break;
1058 	case KVM_REG_PPC_SPURR:
1059 		*val = get_reg_val(id, vcpu->arch.spurr);
1060 		break;
1061 	case KVM_REG_PPC_AMR:
1062 		*val = get_reg_val(id, vcpu->arch.amr);
1063 		break;
1064 	case KVM_REG_PPC_UAMOR:
1065 		*val = get_reg_val(id, vcpu->arch.uamor);
1066 		break;
1067 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1068 		i = id - KVM_REG_PPC_MMCR0;
1069 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1070 		break;
1071 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1072 		i = id - KVM_REG_PPC_PMC1;
1073 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1074 		break;
1075 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1076 		i = id - KVM_REG_PPC_SPMC1;
1077 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1078 		break;
1079 	case KVM_REG_PPC_SIAR:
1080 		*val = get_reg_val(id, vcpu->arch.siar);
1081 		break;
1082 	case KVM_REG_PPC_SDAR:
1083 		*val = get_reg_val(id, vcpu->arch.sdar);
1084 		break;
1085 	case KVM_REG_PPC_SIER:
1086 		*val = get_reg_val(id, vcpu->arch.sier);
1087 		break;
1088 	case KVM_REG_PPC_IAMR:
1089 		*val = get_reg_val(id, vcpu->arch.iamr);
1090 		break;
1091 	case KVM_REG_PPC_PSPB:
1092 		*val = get_reg_val(id, vcpu->arch.pspb);
1093 		break;
1094 	case KVM_REG_PPC_DPDES:
1095 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1096 		break;
1097 	case KVM_REG_PPC_DAWR:
1098 		*val = get_reg_val(id, vcpu->arch.dawr);
1099 		break;
1100 	case KVM_REG_PPC_DAWRX:
1101 		*val = get_reg_val(id, vcpu->arch.dawrx);
1102 		break;
1103 	case KVM_REG_PPC_CIABR:
1104 		*val = get_reg_val(id, vcpu->arch.ciabr);
1105 		break;
1106 	case KVM_REG_PPC_CSIGR:
1107 		*val = get_reg_val(id, vcpu->arch.csigr);
1108 		break;
1109 	case KVM_REG_PPC_TACR:
1110 		*val = get_reg_val(id, vcpu->arch.tacr);
1111 		break;
1112 	case KVM_REG_PPC_TCSCR:
1113 		*val = get_reg_val(id, vcpu->arch.tcscr);
1114 		break;
1115 	case KVM_REG_PPC_PID:
1116 		*val = get_reg_val(id, vcpu->arch.pid);
1117 		break;
1118 	case KVM_REG_PPC_ACOP:
1119 		*val = get_reg_val(id, vcpu->arch.acop);
1120 		break;
1121 	case KVM_REG_PPC_WORT:
1122 		*val = get_reg_val(id, vcpu->arch.wort);
1123 		break;
1124 	case KVM_REG_PPC_VPA_ADDR:
1125 		spin_lock(&vcpu->arch.vpa_update_lock);
1126 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1127 		spin_unlock(&vcpu->arch.vpa_update_lock);
1128 		break;
1129 	case KVM_REG_PPC_VPA_SLB:
1130 		spin_lock(&vcpu->arch.vpa_update_lock);
1131 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1132 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1133 		spin_unlock(&vcpu->arch.vpa_update_lock);
1134 		break;
1135 	case KVM_REG_PPC_VPA_DTL:
1136 		spin_lock(&vcpu->arch.vpa_update_lock);
1137 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1138 		val->vpaval.length = vcpu->arch.dtl.len;
1139 		spin_unlock(&vcpu->arch.vpa_update_lock);
1140 		break;
1141 	case KVM_REG_PPC_TB_OFFSET:
1142 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1143 		break;
1144 	case KVM_REG_PPC_LPCR:
1145 	case KVM_REG_PPC_LPCR_64:
1146 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1147 		break;
1148 	case KVM_REG_PPC_PPR:
1149 		*val = get_reg_val(id, vcpu->arch.ppr);
1150 		break;
1151 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1152 	case KVM_REG_PPC_TFHAR:
1153 		*val = get_reg_val(id, vcpu->arch.tfhar);
1154 		break;
1155 	case KVM_REG_PPC_TFIAR:
1156 		*val = get_reg_val(id, vcpu->arch.tfiar);
1157 		break;
1158 	case KVM_REG_PPC_TEXASR:
1159 		*val = get_reg_val(id, vcpu->arch.texasr);
1160 		break;
1161 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1162 		i = id - KVM_REG_PPC_TM_GPR0;
1163 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1164 		break;
1165 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1166 	{
1167 		int j;
1168 		i = id - KVM_REG_PPC_TM_VSR0;
1169 		if (i < 32)
1170 			for (j = 0; j < TS_FPRWIDTH; j++)
1171 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1172 		else {
1173 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1174 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1175 			else
1176 				r = -ENXIO;
1177 		}
1178 		break;
1179 	}
1180 	case KVM_REG_PPC_TM_CR:
1181 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1182 		break;
1183 	case KVM_REG_PPC_TM_LR:
1184 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1185 		break;
1186 	case KVM_REG_PPC_TM_CTR:
1187 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1188 		break;
1189 	case KVM_REG_PPC_TM_FPSCR:
1190 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1191 		break;
1192 	case KVM_REG_PPC_TM_AMR:
1193 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1194 		break;
1195 	case KVM_REG_PPC_TM_PPR:
1196 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1197 		break;
1198 	case KVM_REG_PPC_TM_VRSAVE:
1199 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1200 		break;
1201 	case KVM_REG_PPC_TM_VSCR:
1202 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1203 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1204 		else
1205 			r = -ENXIO;
1206 		break;
1207 	case KVM_REG_PPC_TM_DSCR:
1208 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1209 		break;
1210 	case KVM_REG_PPC_TM_TAR:
1211 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1212 		break;
1213 #endif
1214 	case KVM_REG_PPC_ARCH_COMPAT:
1215 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1216 		break;
1217 	default:
1218 		r = -EINVAL;
1219 		break;
1220 	}
1221 
1222 	return r;
1223 }
1224 
1225 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1226 				 union kvmppc_one_reg *val)
1227 {
1228 	int r = 0;
1229 	long int i;
1230 	unsigned long addr, len;
1231 
1232 	switch (id) {
1233 	case KVM_REG_PPC_HIOR:
1234 		/* Only allow this to be set to zero */
1235 		if (set_reg_val(id, *val))
1236 			r = -EINVAL;
1237 		break;
1238 	case KVM_REG_PPC_DABR:
1239 		vcpu->arch.dabr = set_reg_val(id, *val);
1240 		break;
1241 	case KVM_REG_PPC_DABRX:
1242 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1243 		break;
1244 	case KVM_REG_PPC_DSCR:
1245 		vcpu->arch.dscr = set_reg_val(id, *val);
1246 		break;
1247 	case KVM_REG_PPC_PURR:
1248 		vcpu->arch.purr = set_reg_val(id, *val);
1249 		break;
1250 	case KVM_REG_PPC_SPURR:
1251 		vcpu->arch.spurr = set_reg_val(id, *val);
1252 		break;
1253 	case KVM_REG_PPC_AMR:
1254 		vcpu->arch.amr = set_reg_val(id, *val);
1255 		break;
1256 	case KVM_REG_PPC_UAMOR:
1257 		vcpu->arch.uamor = set_reg_val(id, *val);
1258 		break;
1259 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1260 		i = id - KVM_REG_PPC_MMCR0;
1261 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1262 		break;
1263 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1264 		i = id - KVM_REG_PPC_PMC1;
1265 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1266 		break;
1267 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1268 		i = id - KVM_REG_PPC_SPMC1;
1269 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1270 		break;
1271 	case KVM_REG_PPC_SIAR:
1272 		vcpu->arch.siar = set_reg_val(id, *val);
1273 		break;
1274 	case KVM_REG_PPC_SDAR:
1275 		vcpu->arch.sdar = set_reg_val(id, *val);
1276 		break;
1277 	case KVM_REG_PPC_SIER:
1278 		vcpu->arch.sier = set_reg_val(id, *val);
1279 		break;
1280 	case KVM_REG_PPC_IAMR:
1281 		vcpu->arch.iamr = set_reg_val(id, *val);
1282 		break;
1283 	case KVM_REG_PPC_PSPB:
1284 		vcpu->arch.pspb = set_reg_val(id, *val);
1285 		break;
1286 	case KVM_REG_PPC_DPDES:
1287 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1288 		break;
1289 	case KVM_REG_PPC_DAWR:
1290 		vcpu->arch.dawr = set_reg_val(id, *val);
1291 		break;
1292 	case KVM_REG_PPC_DAWRX:
1293 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1294 		break;
1295 	case KVM_REG_PPC_CIABR:
1296 		vcpu->arch.ciabr = set_reg_val(id, *val);
1297 		/* Don't allow setting breakpoints in hypervisor code */
1298 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1299 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1300 		break;
1301 	case KVM_REG_PPC_CSIGR:
1302 		vcpu->arch.csigr = set_reg_val(id, *val);
1303 		break;
1304 	case KVM_REG_PPC_TACR:
1305 		vcpu->arch.tacr = set_reg_val(id, *val);
1306 		break;
1307 	case KVM_REG_PPC_TCSCR:
1308 		vcpu->arch.tcscr = set_reg_val(id, *val);
1309 		break;
1310 	case KVM_REG_PPC_PID:
1311 		vcpu->arch.pid = set_reg_val(id, *val);
1312 		break;
1313 	case KVM_REG_PPC_ACOP:
1314 		vcpu->arch.acop = set_reg_val(id, *val);
1315 		break;
1316 	case KVM_REG_PPC_WORT:
1317 		vcpu->arch.wort = set_reg_val(id, *val);
1318 		break;
1319 	case KVM_REG_PPC_VPA_ADDR:
1320 		addr = set_reg_val(id, *val);
1321 		r = -EINVAL;
1322 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1323 			      vcpu->arch.dtl.next_gpa))
1324 			break;
1325 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1326 		break;
1327 	case KVM_REG_PPC_VPA_SLB:
1328 		addr = val->vpaval.addr;
1329 		len = val->vpaval.length;
1330 		r = -EINVAL;
1331 		if (addr && !vcpu->arch.vpa.next_gpa)
1332 			break;
1333 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1334 		break;
1335 	case KVM_REG_PPC_VPA_DTL:
1336 		addr = val->vpaval.addr;
1337 		len = val->vpaval.length;
1338 		r = -EINVAL;
1339 		if (addr && (len < sizeof(struct dtl_entry) ||
1340 			     !vcpu->arch.vpa.next_gpa))
1341 			break;
1342 		len -= len % sizeof(struct dtl_entry);
1343 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1344 		break;
1345 	case KVM_REG_PPC_TB_OFFSET:
1346 		/* round up to multiple of 2^24 */
1347 		vcpu->arch.vcore->tb_offset =
1348 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1349 		break;
1350 	case KVM_REG_PPC_LPCR:
1351 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1352 		break;
1353 	case KVM_REG_PPC_LPCR_64:
1354 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1355 		break;
1356 	case KVM_REG_PPC_PPR:
1357 		vcpu->arch.ppr = set_reg_val(id, *val);
1358 		break;
1359 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1360 	case KVM_REG_PPC_TFHAR:
1361 		vcpu->arch.tfhar = set_reg_val(id, *val);
1362 		break;
1363 	case KVM_REG_PPC_TFIAR:
1364 		vcpu->arch.tfiar = set_reg_val(id, *val);
1365 		break;
1366 	case KVM_REG_PPC_TEXASR:
1367 		vcpu->arch.texasr = set_reg_val(id, *val);
1368 		break;
1369 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1370 		i = id - KVM_REG_PPC_TM_GPR0;
1371 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1372 		break;
1373 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1374 	{
1375 		int j;
1376 		i = id - KVM_REG_PPC_TM_VSR0;
1377 		if (i < 32)
1378 			for (j = 0; j < TS_FPRWIDTH; j++)
1379 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1380 		else
1381 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1382 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1383 			else
1384 				r = -ENXIO;
1385 		break;
1386 	}
1387 	case KVM_REG_PPC_TM_CR:
1388 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1389 		break;
1390 	case KVM_REG_PPC_TM_LR:
1391 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1392 		break;
1393 	case KVM_REG_PPC_TM_CTR:
1394 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1395 		break;
1396 	case KVM_REG_PPC_TM_FPSCR:
1397 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1398 		break;
1399 	case KVM_REG_PPC_TM_AMR:
1400 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1401 		break;
1402 	case KVM_REG_PPC_TM_PPR:
1403 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1404 		break;
1405 	case KVM_REG_PPC_TM_VRSAVE:
1406 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1407 		break;
1408 	case KVM_REG_PPC_TM_VSCR:
1409 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1410 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1411 		else
1412 			r = - ENXIO;
1413 		break;
1414 	case KVM_REG_PPC_TM_DSCR:
1415 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1416 		break;
1417 	case KVM_REG_PPC_TM_TAR:
1418 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1419 		break;
1420 #endif
1421 	case KVM_REG_PPC_ARCH_COMPAT:
1422 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1423 		break;
1424 	default:
1425 		r = -EINVAL;
1426 		break;
1427 	}
1428 
1429 	return r;
1430 }
1431 
1432 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1433 {
1434 	struct kvmppc_vcore *vcore;
1435 
1436 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1437 
1438 	if (vcore == NULL)
1439 		return NULL;
1440 
1441 	INIT_LIST_HEAD(&vcore->runnable_threads);
1442 	spin_lock_init(&vcore->lock);
1443 	spin_lock_init(&vcore->stoltb_lock);
1444 	init_waitqueue_head(&vcore->wq);
1445 	vcore->preempt_tb = TB_NIL;
1446 	vcore->lpcr = kvm->arch.lpcr;
1447 	vcore->first_vcpuid = core * threads_per_subcore;
1448 	vcore->kvm = kvm;
1449 	INIT_LIST_HEAD(&vcore->preempt_list);
1450 
1451 	return vcore;
1452 }
1453 
1454 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1455 static struct debugfs_timings_element {
1456 	const char *name;
1457 	size_t offset;
1458 } timings[] = {
1459 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1460 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1461 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1462 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1463 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1464 };
1465 
1466 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1467 
1468 struct debugfs_timings_state {
1469 	struct kvm_vcpu	*vcpu;
1470 	unsigned int	buflen;
1471 	char		buf[N_TIMINGS * 100];
1472 };
1473 
1474 static int debugfs_timings_open(struct inode *inode, struct file *file)
1475 {
1476 	struct kvm_vcpu *vcpu = inode->i_private;
1477 	struct debugfs_timings_state *p;
1478 
1479 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1480 	if (!p)
1481 		return -ENOMEM;
1482 
1483 	kvm_get_kvm(vcpu->kvm);
1484 	p->vcpu = vcpu;
1485 	file->private_data = p;
1486 
1487 	return nonseekable_open(inode, file);
1488 }
1489 
1490 static int debugfs_timings_release(struct inode *inode, struct file *file)
1491 {
1492 	struct debugfs_timings_state *p = file->private_data;
1493 
1494 	kvm_put_kvm(p->vcpu->kvm);
1495 	kfree(p);
1496 	return 0;
1497 }
1498 
1499 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1500 				    size_t len, loff_t *ppos)
1501 {
1502 	struct debugfs_timings_state *p = file->private_data;
1503 	struct kvm_vcpu *vcpu = p->vcpu;
1504 	char *s, *buf_end;
1505 	struct kvmhv_tb_accumulator tb;
1506 	u64 count;
1507 	loff_t pos;
1508 	ssize_t n;
1509 	int i, loops;
1510 	bool ok;
1511 
1512 	if (!p->buflen) {
1513 		s = p->buf;
1514 		buf_end = s + sizeof(p->buf);
1515 		for (i = 0; i < N_TIMINGS; ++i) {
1516 			struct kvmhv_tb_accumulator *acc;
1517 
1518 			acc = (struct kvmhv_tb_accumulator *)
1519 				((unsigned long)vcpu + timings[i].offset);
1520 			ok = false;
1521 			for (loops = 0; loops < 1000; ++loops) {
1522 				count = acc->seqcount;
1523 				if (!(count & 1)) {
1524 					smp_rmb();
1525 					tb = *acc;
1526 					smp_rmb();
1527 					if (count == acc->seqcount) {
1528 						ok = true;
1529 						break;
1530 					}
1531 				}
1532 				udelay(1);
1533 			}
1534 			if (!ok)
1535 				snprintf(s, buf_end - s, "%s: stuck\n",
1536 					timings[i].name);
1537 			else
1538 				snprintf(s, buf_end - s,
1539 					"%s: %llu %llu %llu %llu\n",
1540 					timings[i].name, count / 2,
1541 					tb_to_ns(tb.tb_total),
1542 					tb_to_ns(tb.tb_min),
1543 					tb_to_ns(tb.tb_max));
1544 			s += strlen(s);
1545 		}
1546 		p->buflen = s - p->buf;
1547 	}
1548 
1549 	pos = *ppos;
1550 	if (pos >= p->buflen)
1551 		return 0;
1552 	if (len > p->buflen - pos)
1553 		len = p->buflen - pos;
1554 	n = copy_to_user(buf, p->buf + pos, len);
1555 	if (n) {
1556 		if (n == len)
1557 			return -EFAULT;
1558 		len -= n;
1559 	}
1560 	*ppos = pos + len;
1561 	return len;
1562 }
1563 
1564 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1565 				     size_t len, loff_t *ppos)
1566 {
1567 	return -EACCES;
1568 }
1569 
1570 static const struct file_operations debugfs_timings_ops = {
1571 	.owner	 = THIS_MODULE,
1572 	.open	 = debugfs_timings_open,
1573 	.release = debugfs_timings_release,
1574 	.read	 = debugfs_timings_read,
1575 	.write	 = debugfs_timings_write,
1576 	.llseek	 = generic_file_llseek,
1577 };
1578 
1579 /* Create a debugfs directory for the vcpu */
1580 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1581 {
1582 	char buf[16];
1583 	struct kvm *kvm = vcpu->kvm;
1584 
1585 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1586 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1587 		return;
1588 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1589 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1590 		return;
1591 	vcpu->arch.debugfs_timings =
1592 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1593 				    vcpu, &debugfs_timings_ops);
1594 }
1595 
1596 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1597 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1598 {
1599 }
1600 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1601 
1602 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1603 						   unsigned int id)
1604 {
1605 	struct kvm_vcpu *vcpu;
1606 	int err = -EINVAL;
1607 	int core;
1608 	struct kvmppc_vcore *vcore;
1609 
1610 	core = id / threads_per_subcore;
1611 	if (core >= KVM_MAX_VCORES)
1612 		goto out;
1613 
1614 	err = -ENOMEM;
1615 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1616 	if (!vcpu)
1617 		goto out;
1618 
1619 	err = kvm_vcpu_init(vcpu, kvm, id);
1620 	if (err)
1621 		goto free_vcpu;
1622 
1623 	vcpu->arch.shared = &vcpu->arch.shregs;
1624 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1625 	/*
1626 	 * The shared struct is never shared on HV,
1627 	 * so we can always use host endianness
1628 	 */
1629 #ifdef __BIG_ENDIAN__
1630 	vcpu->arch.shared_big_endian = true;
1631 #else
1632 	vcpu->arch.shared_big_endian = false;
1633 #endif
1634 #endif
1635 	vcpu->arch.mmcr[0] = MMCR0_FC;
1636 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1637 	/* default to host PVR, since we can't spoof it */
1638 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1639 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1640 	spin_lock_init(&vcpu->arch.tbacct_lock);
1641 	vcpu->arch.busy_preempt = TB_NIL;
1642 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1643 
1644 	kvmppc_mmu_book3s_hv_init(vcpu);
1645 
1646 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1647 
1648 	init_waitqueue_head(&vcpu->arch.cpu_run);
1649 
1650 	mutex_lock(&kvm->lock);
1651 	vcore = kvm->arch.vcores[core];
1652 	if (!vcore) {
1653 		vcore = kvmppc_vcore_create(kvm, core);
1654 		kvm->arch.vcores[core] = vcore;
1655 		kvm->arch.online_vcores++;
1656 	}
1657 	mutex_unlock(&kvm->lock);
1658 
1659 	if (!vcore)
1660 		goto free_vcpu;
1661 
1662 	spin_lock(&vcore->lock);
1663 	++vcore->num_threads;
1664 	spin_unlock(&vcore->lock);
1665 	vcpu->arch.vcore = vcore;
1666 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1667 	vcpu->arch.thread_cpu = -1;
1668 
1669 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1670 	kvmppc_sanity_check(vcpu);
1671 
1672 	debugfs_vcpu_init(vcpu, id);
1673 
1674 	return vcpu;
1675 
1676 free_vcpu:
1677 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1678 out:
1679 	return ERR_PTR(err);
1680 }
1681 
1682 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1683 {
1684 	if (vpa->pinned_addr)
1685 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1686 					vpa->dirty);
1687 }
1688 
1689 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1690 {
1691 	spin_lock(&vcpu->arch.vpa_update_lock);
1692 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1693 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1694 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1695 	spin_unlock(&vcpu->arch.vpa_update_lock);
1696 	kvm_vcpu_uninit(vcpu);
1697 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1698 }
1699 
1700 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1701 {
1702 	/* Indicate we want to get back into the guest */
1703 	return 1;
1704 }
1705 
1706 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1707 {
1708 	unsigned long dec_nsec, now;
1709 
1710 	now = get_tb();
1711 	if (now > vcpu->arch.dec_expires) {
1712 		/* decrementer has already gone negative */
1713 		kvmppc_core_queue_dec(vcpu);
1714 		kvmppc_core_prepare_to_enter(vcpu);
1715 		return;
1716 	}
1717 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1718 		   / tb_ticks_per_sec;
1719 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1720 		      HRTIMER_MODE_REL);
1721 	vcpu->arch.timer_running = 1;
1722 }
1723 
1724 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1725 {
1726 	vcpu->arch.ceded = 0;
1727 	if (vcpu->arch.timer_running) {
1728 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1729 		vcpu->arch.timer_running = 0;
1730 	}
1731 }
1732 
1733 extern void __kvmppc_vcore_entry(void);
1734 
1735 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1736 				   struct kvm_vcpu *vcpu)
1737 {
1738 	u64 now;
1739 
1740 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1741 		return;
1742 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1743 	now = mftb();
1744 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1745 		vcpu->arch.stolen_logged;
1746 	vcpu->arch.busy_preempt = now;
1747 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1748 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1749 	--vc->n_runnable;
1750 	list_del(&vcpu->arch.run_list);
1751 }
1752 
1753 static int kvmppc_grab_hwthread(int cpu)
1754 {
1755 	struct paca_struct *tpaca;
1756 	long timeout = 10000;
1757 
1758 	tpaca = &paca[cpu];
1759 
1760 	/* Ensure the thread won't go into the kernel if it wakes */
1761 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1762 	tpaca->kvm_hstate.kvm_vcore = NULL;
1763 	tpaca->kvm_hstate.napping = 0;
1764 	smp_wmb();
1765 	tpaca->kvm_hstate.hwthread_req = 1;
1766 
1767 	/*
1768 	 * If the thread is already executing in the kernel (e.g. handling
1769 	 * a stray interrupt), wait for it to get back to nap mode.
1770 	 * The smp_mb() is to ensure that our setting of hwthread_req
1771 	 * is visible before we look at hwthread_state, so if this
1772 	 * races with the code at system_reset_pSeries and the thread
1773 	 * misses our setting of hwthread_req, we are sure to see its
1774 	 * setting of hwthread_state, and vice versa.
1775 	 */
1776 	smp_mb();
1777 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1778 		if (--timeout <= 0) {
1779 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1780 			return -EBUSY;
1781 		}
1782 		udelay(1);
1783 	}
1784 	return 0;
1785 }
1786 
1787 static void kvmppc_release_hwthread(int cpu)
1788 {
1789 	struct paca_struct *tpaca;
1790 
1791 	tpaca = &paca[cpu];
1792 	tpaca->kvm_hstate.hwthread_req = 0;
1793 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1794 	tpaca->kvm_hstate.kvm_vcore = NULL;
1795 	tpaca->kvm_hstate.kvm_split_mode = NULL;
1796 }
1797 
1798 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1799 {
1800 	int cpu;
1801 	struct paca_struct *tpaca;
1802 	struct kvmppc_vcore *mvc = vc->master_vcore;
1803 
1804 	cpu = vc->pcpu;
1805 	if (vcpu) {
1806 		if (vcpu->arch.timer_running) {
1807 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1808 			vcpu->arch.timer_running = 0;
1809 		}
1810 		cpu += vcpu->arch.ptid;
1811 		vcpu->cpu = mvc->pcpu;
1812 		vcpu->arch.thread_cpu = cpu;
1813 	}
1814 	tpaca = &paca[cpu];
1815 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1816 	tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1817 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1818 	smp_wmb();
1819 	tpaca->kvm_hstate.kvm_vcore = mvc;
1820 	if (cpu != smp_processor_id())
1821 		kvmppc_ipi_thread(cpu);
1822 }
1823 
1824 static void kvmppc_wait_for_nap(void)
1825 {
1826 	int cpu = smp_processor_id();
1827 	int i, loops;
1828 
1829 	for (loops = 0; loops < 1000000; ++loops) {
1830 		/*
1831 		 * Check if all threads are finished.
1832 		 * We set the vcore pointer when starting a thread
1833 		 * and the thread clears it when finished, so we look
1834 		 * for any threads that still have a non-NULL vcore ptr.
1835 		 */
1836 		for (i = 1; i < threads_per_subcore; ++i)
1837 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
1838 				break;
1839 		if (i == threads_per_subcore) {
1840 			HMT_medium();
1841 			return;
1842 		}
1843 		HMT_low();
1844 	}
1845 	HMT_medium();
1846 	for (i = 1; i < threads_per_subcore; ++i)
1847 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
1848 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1849 }
1850 
1851 /*
1852  * Check that we are on thread 0 and that any other threads in
1853  * this core are off-line.  Then grab the threads so they can't
1854  * enter the kernel.
1855  */
1856 static int on_primary_thread(void)
1857 {
1858 	int cpu = smp_processor_id();
1859 	int thr;
1860 
1861 	/* Are we on a primary subcore? */
1862 	if (cpu_thread_in_subcore(cpu))
1863 		return 0;
1864 
1865 	thr = 0;
1866 	while (++thr < threads_per_subcore)
1867 		if (cpu_online(cpu + thr))
1868 			return 0;
1869 
1870 	/* Grab all hw threads so they can't go into the kernel */
1871 	for (thr = 1; thr < threads_per_subcore; ++thr) {
1872 		if (kvmppc_grab_hwthread(cpu + thr)) {
1873 			/* Couldn't grab one; let the others go */
1874 			do {
1875 				kvmppc_release_hwthread(cpu + thr);
1876 			} while (--thr > 0);
1877 			return 0;
1878 		}
1879 	}
1880 	return 1;
1881 }
1882 
1883 /*
1884  * A list of virtual cores for each physical CPU.
1885  * These are vcores that could run but their runner VCPU tasks are
1886  * (or may be) preempted.
1887  */
1888 struct preempted_vcore_list {
1889 	struct list_head	list;
1890 	spinlock_t		lock;
1891 };
1892 
1893 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1894 
1895 static void init_vcore_lists(void)
1896 {
1897 	int cpu;
1898 
1899 	for_each_possible_cpu(cpu) {
1900 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1901 		spin_lock_init(&lp->lock);
1902 		INIT_LIST_HEAD(&lp->list);
1903 	}
1904 }
1905 
1906 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1907 {
1908 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1909 
1910 	vc->vcore_state = VCORE_PREEMPT;
1911 	vc->pcpu = smp_processor_id();
1912 	if (vc->num_threads < threads_per_subcore) {
1913 		spin_lock(&lp->lock);
1914 		list_add_tail(&vc->preempt_list, &lp->list);
1915 		spin_unlock(&lp->lock);
1916 	}
1917 
1918 	/* Start accumulating stolen time */
1919 	kvmppc_core_start_stolen(vc);
1920 }
1921 
1922 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1923 {
1924 	struct preempted_vcore_list *lp;
1925 
1926 	kvmppc_core_end_stolen(vc);
1927 	if (!list_empty(&vc->preempt_list)) {
1928 		lp = &per_cpu(preempted_vcores, vc->pcpu);
1929 		spin_lock(&lp->lock);
1930 		list_del_init(&vc->preempt_list);
1931 		spin_unlock(&lp->lock);
1932 	}
1933 	vc->vcore_state = VCORE_INACTIVE;
1934 }
1935 
1936 /*
1937  * This stores information about the virtual cores currently
1938  * assigned to a physical core.
1939  */
1940 struct core_info {
1941 	int		n_subcores;
1942 	int		max_subcore_threads;
1943 	int		total_threads;
1944 	int		subcore_threads[MAX_SUBCORES];
1945 	struct kvm	*subcore_vm[MAX_SUBCORES];
1946 	struct list_head vcs[MAX_SUBCORES];
1947 };
1948 
1949 /*
1950  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
1951  * respectively in 2-way micro-threading (split-core) mode.
1952  */
1953 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
1954 
1955 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
1956 {
1957 	int sub;
1958 
1959 	memset(cip, 0, sizeof(*cip));
1960 	cip->n_subcores = 1;
1961 	cip->max_subcore_threads = vc->num_threads;
1962 	cip->total_threads = vc->num_threads;
1963 	cip->subcore_threads[0] = vc->num_threads;
1964 	cip->subcore_vm[0] = vc->kvm;
1965 	for (sub = 0; sub < MAX_SUBCORES; ++sub)
1966 		INIT_LIST_HEAD(&cip->vcs[sub]);
1967 	list_add_tail(&vc->preempt_list, &cip->vcs[0]);
1968 }
1969 
1970 static bool subcore_config_ok(int n_subcores, int n_threads)
1971 {
1972 	/* Can only dynamically split if unsplit to begin with */
1973 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
1974 		return false;
1975 	if (n_subcores > MAX_SUBCORES)
1976 		return false;
1977 	if (n_subcores > 1) {
1978 		if (!(dynamic_mt_modes & 2))
1979 			n_subcores = 4;
1980 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
1981 			return false;
1982 	}
1983 
1984 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
1985 }
1986 
1987 static void init_master_vcore(struct kvmppc_vcore *vc)
1988 {
1989 	vc->master_vcore = vc;
1990 	vc->entry_exit_map = 0;
1991 	vc->in_guest = 0;
1992 	vc->napping_threads = 0;
1993 	vc->conferring_threads = 0;
1994 }
1995 
1996 /*
1997  * See if the existing subcores can be split into 3 (or fewer) subcores
1998  * of at most two threads each, so we can fit in another vcore.  This
1999  * assumes there are at most two subcores and at most 6 threads in total.
2000  */
2001 static bool can_split_piggybacked_subcores(struct core_info *cip)
2002 {
2003 	int sub, new_sub;
2004 	int large_sub = -1;
2005 	int thr;
2006 	int n_subcores = cip->n_subcores;
2007 	struct kvmppc_vcore *vc, *vcnext;
2008 	struct kvmppc_vcore *master_vc = NULL;
2009 
2010 	for (sub = 0; sub < cip->n_subcores; ++sub) {
2011 		if (cip->subcore_threads[sub] <= 2)
2012 			continue;
2013 		if (large_sub >= 0)
2014 			return false;
2015 		large_sub = sub;
2016 		vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2017 				      preempt_list);
2018 		if (vc->num_threads > 2)
2019 			return false;
2020 		n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2021 	}
2022 	if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2023 		return false;
2024 
2025 	/*
2026 	 * Seems feasible, so go through and move vcores to new subcores.
2027 	 * Note that when we have two or more vcores in one subcore,
2028 	 * all those vcores must have only one thread each.
2029 	 */
2030 	new_sub = cip->n_subcores;
2031 	thr = 0;
2032 	sub = large_sub;
2033 	list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2034 		if (thr >= 2) {
2035 			list_del(&vc->preempt_list);
2036 			list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2037 			/* vc->num_threads must be 1 */
2038 			if (++cip->subcore_threads[new_sub] == 1) {
2039 				cip->subcore_vm[new_sub] = vc->kvm;
2040 				init_master_vcore(vc);
2041 				master_vc = vc;
2042 				++cip->n_subcores;
2043 			} else {
2044 				vc->master_vcore = master_vc;
2045 				++new_sub;
2046 			}
2047 		}
2048 		thr += vc->num_threads;
2049 	}
2050 	cip->subcore_threads[large_sub] = 2;
2051 	cip->max_subcore_threads = 2;
2052 
2053 	return true;
2054 }
2055 
2056 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2057 {
2058 	int n_threads = vc->num_threads;
2059 	int sub;
2060 
2061 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2062 		return false;
2063 
2064 	if (n_threads < cip->max_subcore_threads)
2065 		n_threads = cip->max_subcore_threads;
2066 	if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2067 		cip->max_subcore_threads = n_threads;
2068 	} else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2069 		   vc->num_threads <= 2) {
2070 		/*
2071 		 * We may be able to fit another subcore in by
2072 		 * splitting an existing subcore with 3 or 4
2073 		 * threads into two 2-thread subcores, or one
2074 		 * with 5 or 6 threads into three subcores.
2075 		 * We can only do this if those subcores have
2076 		 * piggybacked virtual cores.
2077 		 */
2078 		if (!can_split_piggybacked_subcores(cip))
2079 			return false;
2080 	} else {
2081 		return false;
2082 	}
2083 
2084 	sub = cip->n_subcores;
2085 	++cip->n_subcores;
2086 	cip->total_threads += vc->num_threads;
2087 	cip->subcore_threads[sub] = vc->num_threads;
2088 	cip->subcore_vm[sub] = vc->kvm;
2089 	init_master_vcore(vc);
2090 	list_del(&vc->preempt_list);
2091 	list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2092 
2093 	return true;
2094 }
2095 
2096 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2097 				  struct core_info *cip, int sub)
2098 {
2099 	struct kvmppc_vcore *vc;
2100 	int n_thr;
2101 
2102 	vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2103 			      preempt_list);
2104 
2105 	/* require same VM and same per-core reg values */
2106 	if (pvc->kvm != vc->kvm ||
2107 	    pvc->tb_offset != vc->tb_offset ||
2108 	    pvc->pcr != vc->pcr ||
2109 	    pvc->lpcr != vc->lpcr)
2110 		return false;
2111 
2112 	/* P8 guest with > 1 thread per core would see wrong TIR value */
2113 	if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2114 	    (vc->num_threads > 1 || pvc->num_threads > 1))
2115 		return false;
2116 
2117 	n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2118 	if (n_thr > cip->max_subcore_threads) {
2119 		if (!subcore_config_ok(cip->n_subcores, n_thr))
2120 			return false;
2121 		cip->max_subcore_threads = n_thr;
2122 	}
2123 
2124 	cip->total_threads += pvc->num_threads;
2125 	cip->subcore_threads[sub] = n_thr;
2126 	pvc->master_vcore = vc;
2127 	list_del(&pvc->preempt_list);
2128 	list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2129 
2130 	return true;
2131 }
2132 
2133 /*
2134  * Work out whether it is possible to piggyback the execution of
2135  * vcore *pvc onto the execution of the other vcores described in *cip.
2136  */
2137 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2138 			  int target_threads)
2139 {
2140 	int sub;
2141 
2142 	if (cip->total_threads + pvc->num_threads > target_threads)
2143 		return false;
2144 	for (sub = 0; sub < cip->n_subcores; ++sub)
2145 		if (cip->subcore_threads[sub] &&
2146 		    can_piggyback_subcore(pvc, cip, sub))
2147 			return true;
2148 
2149 	if (can_dynamic_split(pvc, cip))
2150 		return true;
2151 
2152 	return false;
2153 }
2154 
2155 static void prepare_threads(struct kvmppc_vcore *vc)
2156 {
2157 	struct kvm_vcpu *vcpu, *vnext;
2158 
2159 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2160 				 arch.run_list) {
2161 		if (signal_pending(vcpu->arch.run_task))
2162 			vcpu->arch.ret = -EINTR;
2163 		else if (vcpu->arch.vpa.update_pending ||
2164 			 vcpu->arch.slb_shadow.update_pending ||
2165 			 vcpu->arch.dtl.update_pending)
2166 			vcpu->arch.ret = RESUME_GUEST;
2167 		else
2168 			continue;
2169 		kvmppc_remove_runnable(vc, vcpu);
2170 		wake_up(&vcpu->arch.cpu_run);
2171 	}
2172 }
2173 
2174 static void collect_piggybacks(struct core_info *cip, int target_threads)
2175 {
2176 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2177 	struct kvmppc_vcore *pvc, *vcnext;
2178 
2179 	spin_lock(&lp->lock);
2180 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2181 		if (!spin_trylock(&pvc->lock))
2182 			continue;
2183 		prepare_threads(pvc);
2184 		if (!pvc->n_runnable) {
2185 			list_del_init(&pvc->preempt_list);
2186 			if (pvc->runner == NULL) {
2187 				pvc->vcore_state = VCORE_INACTIVE;
2188 				kvmppc_core_end_stolen(pvc);
2189 			}
2190 			spin_unlock(&pvc->lock);
2191 			continue;
2192 		}
2193 		if (!can_piggyback(pvc, cip, target_threads)) {
2194 			spin_unlock(&pvc->lock);
2195 			continue;
2196 		}
2197 		kvmppc_core_end_stolen(pvc);
2198 		pvc->vcore_state = VCORE_PIGGYBACK;
2199 		if (cip->total_threads >= target_threads)
2200 			break;
2201 	}
2202 	spin_unlock(&lp->lock);
2203 }
2204 
2205 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2206 {
2207 	int still_running = 0;
2208 	u64 now;
2209 	long ret;
2210 	struct kvm_vcpu *vcpu, *vnext;
2211 
2212 	spin_lock(&vc->lock);
2213 	now = get_tb();
2214 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2215 				 arch.run_list) {
2216 		/* cancel pending dec exception if dec is positive */
2217 		if (now < vcpu->arch.dec_expires &&
2218 		    kvmppc_core_pending_dec(vcpu))
2219 			kvmppc_core_dequeue_dec(vcpu);
2220 
2221 		trace_kvm_guest_exit(vcpu);
2222 
2223 		ret = RESUME_GUEST;
2224 		if (vcpu->arch.trap)
2225 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2226 						    vcpu->arch.run_task);
2227 
2228 		vcpu->arch.ret = ret;
2229 		vcpu->arch.trap = 0;
2230 
2231 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2232 			if (vcpu->arch.pending_exceptions)
2233 				kvmppc_core_prepare_to_enter(vcpu);
2234 			if (vcpu->arch.ceded)
2235 				kvmppc_set_timer(vcpu);
2236 			else
2237 				++still_running;
2238 		} else {
2239 			kvmppc_remove_runnable(vc, vcpu);
2240 			wake_up(&vcpu->arch.cpu_run);
2241 		}
2242 	}
2243 	list_del_init(&vc->preempt_list);
2244 	if (!is_master) {
2245 		if (still_running > 0) {
2246 			kvmppc_vcore_preempt(vc);
2247 		} else if (vc->runner) {
2248 			vc->vcore_state = VCORE_PREEMPT;
2249 			kvmppc_core_start_stolen(vc);
2250 		} else {
2251 			vc->vcore_state = VCORE_INACTIVE;
2252 		}
2253 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2254 			/* make sure there's a candidate runner awake */
2255 			vcpu = list_first_entry(&vc->runnable_threads,
2256 						struct kvm_vcpu, arch.run_list);
2257 			wake_up(&vcpu->arch.cpu_run);
2258 		}
2259 	}
2260 	spin_unlock(&vc->lock);
2261 }
2262 
2263 /*
2264  * Run a set of guest threads on a physical core.
2265  * Called with vc->lock held.
2266  */
2267 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2268 {
2269 	struct kvm_vcpu *vcpu, *vnext;
2270 	int i;
2271 	int srcu_idx;
2272 	struct core_info core_info;
2273 	struct kvmppc_vcore *pvc, *vcnext;
2274 	struct kvm_split_mode split_info, *sip;
2275 	int split, subcore_size, active;
2276 	int sub;
2277 	bool thr0_done;
2278 	unsigned long cmd_bit, stat_bit;
2279 	int pcpu, thr;
2280 	int target_threads;
2281 
2282 	/*
2283 	 * Remove from the list any threads that have a signal pending
2284 	 * or need a VPA update done
2285 	 */
2286 	prepare_threads(vc);
2287 
2288 	/* if the runner is no longer runnable, let the caller pick a new one */
2289 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2290 		return;
2291 
2292 	/*
2293 	 * Initialize *vc.
2294 	 */
2295 	init_master_vcore(vc);
2296 	vc->preempt_tb = TB_NIL;
2297 
2298 	/*
2299 	 * Make sure we are running on primary threads, and that secondary
2300 	 * threads are offline.  Also check if the number of threads in this
2301 	 * guest are greater than the current system threads per guest.
2302 	 */
2303 	if ((threads_per_core > 1) &&
2304 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2305 		list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2306 					 arch.run_list) {
2307 			vcpu->arch.ret = -EBUSY;
2308 			kvmppc_remove_runnable(vc, vcpu);
2309 			wake_up(&vcpu->arch.cpu_run);
2310 		}
2311 		goto out;
2312 	}
2313 
2314 	/*
2315 	 * See if we could run any other vcores on the physical core
2316 	 * along with this one.
2317 	 */
2318 	init_core_info(&core_info, vc);
2319 	pcpu = smp_processor_id();
2320 	target_threads = threads_per_subcore;
2321 	if (target_smt_mode && target_smt_mode < target_threads)
2322 		target_threads = target_smt_mode;
2323 	if (vc->num_threads < target_threads)
2324 		collect_piggybacks(&core_info, target_threads);
2325 
2326 	/* Decide on micro-threading (split-core) mode */
2327 	subcore_size = threads_per_subcore;
2328 	cmd_bit = stat_bit = 0;
2329 	split = core_info.n_subcores;
2330 	sip = NULL;
2331 	if (split > 1) {
2332 		/* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2333 		if (split == 2 && (dynamic_mt_modes & 2)) {
2334 			cmd_bit = HID0_POWER8_1TO2LPAR;
2335 			stat_bit = HID0_POWER8_2LPARMODE;
2336 		} else {
2337 			split = 4;
2338 			cmd_bit = HID0_POWER8_1TO4LPAR;
2339 			stat_bit = HID0_POWER8_4LPARMODE;
2340 		}
2341 		subcore_size = MAX_SMT_THREADS / split;
2342 		sip = &split_info;
2343 		memset(&split_info, 0, sizeof(split_info));
2344 		split_info.rpr = mfspr(SPRN_RPR);
2345 		split_info.pmmar = mfspr(SPRN_PMMAR);
2346 		split_info.ldbar = mfspr(SPRN_LDBAR);
2347 		split_info.subcore_size = subcore_size;
2348 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2349 			split_info.master_vcs[sub] =
2350 				list_first_entry(&core_info.vcs[sub],
2351 					struct kvmppc_vcore, preempt_list);
2352 		/* order writes to split_info before kvm_split_mode pointer */
2353 		smp_wmb();
2354 	}
2355 	pcpu = smp_processor_id();
2356 	for (thr = 0; thr < threads_per_subcore; ++thr)
2357 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2358 
2359 	/* Initiate micro-threading (split-core) if required */
2360 	if (cmd_bit) {
2361 		unsigned long hid0 = mfspr(SPRN_HID0);
2362 
2363 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2364 		mb();
2365 		mtspr(SPRN_HID0, hid0);
2366 		isync();
2367 		for (;;) {
2368 			hid0 = mfspr(SPRN_HID0);
2369 			if (hid0 & stat_bit)
2370 				break;
2371 			cpu_relax();
2372 		}
2373 	}
2374 
2375 	/* Start all the threads */
2376 	active = 0;
2377 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2378 		thr = subcore_thread_map[sub];
2379 		thr0_done = false;
2380 		active |= 1 << thr;
2381 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2382 			pvc->pcpu = pcpu + thr;
2383 			list_for_each_entry(vcpu, &pvc->runnable_threads,
2384 					    arch.run_list) {
2385 				kvmppc_start_thread(vcpu, pvc);
2386 				kvmppc_create_dtl_entry(vcpu, pvc);
2387 				trace_kvm_guest_enter(vcpu);
2388 				if (!vcpu->arch.ptid)
2389 					thr0_done = true;
2390 				active |= 1 << (thr + vcpu->arch.ptid);
2391 			}
2392 			/*
2393 			 * We need to start the first thread of each subcore
2394 			 * even if it doesn't have a vcpu.
2395 			 */
2396 			if (pvc->master_vcore == pvc && !thr0_done)
2397 				kvmppc_start_thread(NULL, pvc);
2398 			thr += pvc->num_threads;
2399 		}
2400 	}
2401 
2402 	/*
2403 	 * Ensure that split_info.do_nap is set after setting
2404 	 * the vcore pointer in the PACA of the secondaries.
2405 	 */
2406 	smp_mb();
2407 	if (cmd_bit)
2408 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2409 
2410 	/*
2411 	 * When doing micro-threading, poke the inactive threads as well.
2412 	 * This gets them to the nap instruction after kvm_do_nap,
2413 	 * which reduces the time taken to unsplit later.
2414 	 */
2415 	if (split > 1)
2416 		for (thr = 1; thr < threads_per_subcore; ++thr)
2417 			if (!(active & (1 << thr)))
2418 				kvmppc_ipi_thread(pcpu + thr);
2419 
2420 	vc->vcore_state = VCORE_RUNNING;
2421 	preempt_disable();
2422 
2423 	trace_kvmppc_run_core(vc, 0);
2424 
2425 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2426 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2427 			spin_unlock(&pvc->lock);
2428 
2429 	kvm_guest_enter();
2430 
2431 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2432 
2433 	__kvmppc_vcore_entry();
2434 
2435 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2436 
2437 	spin_lock(&vc->lock);
2438 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2439 	vc->vcore_state = VCORE_EXITING;
2440 
2441 	/* wait for secondary threads to finish writing their state to memory */
2442 	kvmppc_wait_for_nap();
2443 
2444 	/* Return to whole-core mode if we split the core earlier */
2445 	if (split > 1) {
2446 		unsigned long hid0 = mfspr(SPRN_HID0);
2447 		unsigned long loops = 0;
2448 
2449 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2450 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2451 		mb();
2452 		mtspr(SPRN_HID0, hid0);
2453 		isync();
2454 		for (;;) {
2455 			hid0 = mfspr(SPRN_HID0);
2456 			if (!(hid0 & stat_bit))
2457 				break;
2458 			cpu_relax();
2459 			++loops;
2460 		}
2461 		split_info.do_nap = 0;
2462 	}
2463 
2464 	/* Let secondaries go back to the offline loop */
2465 	for (i = 0; i < threads_per_subcore; ++i) {
2466 		kvmppc_release_hwthread(pcpu + i);
2467 		if (sip && sip->napped[i])
2468 			kvmppc_ipi_thread(pcpu + i);
2469 	}
2470 
2471 	spin_unlock(&vc->lock);
2472 
2473 	/* make sure updates to secondary vcpu structs are visible now */
2474 	smp_mb();
2475 	kvm_guest_exit();
2476 
2477 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2478 		list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2479 					 preempt_list)
2480 			post_guest_process(pvc, pvc == vc);
2481 
2482 	spin_lock(&vc->lock);
2483 	preempt_enable();
2484 
2485  out:
2486 	vc->vcore_state = VCORE_INACTIVE;
2487 	trace_kvmppc_run_core(vc, 1);
2488 }
2489 
2490 /*
2491  * Wait for some other vcpu thread to execute us, and
2492  * wake us up when we need to handle something in the host.
2493  */
2494 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2495 				 struct kvm_vcpu *vcpu, int wait_state)
2496 {
2497 	DEFINE_WAIT(wait);
2498 
2499 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2500 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2501 		spin_unlock(&vc->lock);
2502 		schedule();
2503 		spin_lock(&vc->lock);
2504 	}
2505 	finish_wait(&vcpu->arch.cpu_run, &wait);
2506 }
2507 
2508 /*
2509  * All the vcpus in this vcore are idle, so wait for a decrementer
2510  * or external interrupt to one of the vcpus.  vc->lock is held.
2511  */
2512 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2513 {
2514 	struct kvm_vcpu *vcpu;
2515 	int do_sleep = 1;
2516 
2517 	DEFINE_WAIT(wait);
2518 
2519 	prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2520 
2521 	/*
2522 	 * Check one last time for pending exceptions and ceded state after
2523 	 * we put ourselves on the wait queue
2524 	 */
2525 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2526 		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2527 			do_sleep = 0;
2528 			break;
2529 		}
2530 	}
2531 
2532 	if (!do_sleep) {
2533 		finish_wait(&vc->wq, &wait);
2534 		return;
2535 	}
2536 
2537 	vc->vcore_state = VCORE_SLEEPING;
2538 	trace_kvmppc_vcore_blocked(vc, 0);
2539 	spin_unlock(&vc->lock);
2540 	schedule();
2541 	finish_wait(&vc->wq, &wait);
2542 	spin_lock(&vc->lock);
2543 	vc->vcore_state = VCORE_INACTIVE;
2544 	trace_kvmppc_vcore_blocked(vc, 1);
2545 }
2546 
2547 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2548 {
2549 	int n_ceded;
2550 	struct kvmppc_vcore *vc;
2551 	struct kvm_vcpu *v, *vn;
2552 
2553 	trace_kvmppc_run_vcpu_enter(vcpu);
2554 
2555 	kvm_run->exit_reason = 0;
2556 	vcpu->arch.ret = RESUME_GUEST;
2557 	vcpu->arch.trap = 0;
2558 	kvmppc_update_vpas(vcpu);
2559 
2560 	/*
2561 	 * Synchronize with other threads in this virtual core
2562 	 */
2563 	vc = vcpu->arch.vcore;
2564 	spin_lock(&vc->lock);
2565 	vcpu->arch.ceded = 0;
2566 	vcpu->arch.run_task = current;
2567 	vcpu->arch.kvm_run = kvm_run;
2568 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2569 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2570 	vcpu->arch.busy_preempt = TB_NIL;
2571 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2572 	++vc->n_runnable;
2573 
2574 	/*
2575 	 * This happens the first time this is called for a vcpu.
2576 	 * If the vcore is already running, we may be able to start
2577 	 * this thread straight away and have it join in.
2578 	 */
2579 	if (!signal_pending(current)) {
2580 		if (vc->vcore_state == VCORE_PIGGYBACK) {
2581 			struct kvmppc_vcore *mvc = vc->master_vcore;
2582 			if (spin_trylock(&mvc->lock)) {
2583 				if (mvc->vcore_state == VCORE_RUNNING &&
2584 				    !VCORE_IS_EXITING(mvc)) {
2585 					kvmppc_create_dtl_entry(vcpu, vc);
2586 					kvmppc_start_thread(vcpu, vc);
2587 					trace_kvm_guest_enter(vcpu);
2588 				}
2589 				spin_unlock(&mvc->lock);
2590 			}
2591 		} else if (vc->vcore_state == VCORE_RUNNING &&
2592 			   !VCORE_IS_EXITING(vc)) {
2593 			kvmppc_create_dtl_entry(vcpu, vc);
2594 			kvmppc_start_thread(vcpu, vc);
2595 			trace_kvm_guest_enter(vcpu);
2596 		} else if (vc->vcore_state == VCORE_SLEEPING) {
2597 			wake_up(&vc->wq);
2598 		}
2599 
2600 	}
2601 
2602 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2603 	       !signal_pending(current)) {
2604 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2605 			kvmppc_vcore_end_preempt(vc);
2606 
2607 		if (vc->vcore_state != VCORE_INACTIVE) {
2608 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2609 			continue;
2610 		}
2611 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2612 					 arch.run_list) {
2613 			kvmppc_core_prepare_to_enter(v);
2614 			if (signal_pending(v->arch.run_task)) {
2615 				kvmppc_remove_runnable(vc, v);
2616 				v->stat.signal_exits++;
2617 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2618 				v->arch.ret = -EINTR;
2619 				wake_up(&v->arch.cpu_run);
2620 			}
2621 		}
2622 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2623 			break;
2624 		n_ceded = 0;
2625 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2626 			if (!v->arch.pending_exceptions)
2627 				n_ceded += v->arch.ceded;
2628 			else
2629 				v->arch.ceded = 0;
2630 		}
2631 		vc->runner = vcpu;
2632 		if (n_ceded == vc->n_runnable) {
2633 			kvmppc_vcore_blocked(vc);
2634 		} else if (need_resched()) {
2635 			kvmppc_vcore_preempt(vc);
2636 			/* Let something else run */
2637 			cond_resched_lock(&vc->lock);
2638 			if (vc->vcore_state == VCORE_PREEMPT)
2639 				kvmppc_vcore_end_preempt(vc);
2640 		} else {
2641 			kvmppc_run_core(vc);
2642 		}
2643 		vc->runner = NULL;
2644 	}
2645 
2646 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2647 	       (vc->vcore_state == VCORE_RUNNING ||
2648 		vc->vcore_state == VCORE_EXITING ||
2649 		vc->vcore_state == VCORE_PIGGYBACK))
2650 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2651 
2652 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2653 		kvmppc_vcore_end_preempt(vc);
2654 
2655 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2656 		kvmppc_remove_runnable(vc, vcpu);
2657 		vcpu->stat.signal_exits++;
2658 		kvm_run->exit_reason = KVM_EXIT_INTR;
2659 		vcpu->arch.ret = -EINTR;
2660 	}
2661 
2662 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2663 		/* Wake up some vcpu to run the core */
2664 		v = list_first_entry(&vc->runnable_threads,
2665 				     struct kvm_vcpu, arch.run_list);
2666 		wake_up(&v->arch.cpu_run);
2667 	}
2668 
2669 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2670 	spin_unlock(&vc->lock);
2671 	return vcpu->arch.ret;
2672 }
2673 
2674 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2675 {
2676 	int r;
2677 	int srcu_idx;
2678 
2679 	if (!vcpu->arch.sane) {
2680 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2681 		return -EINVAL;
2682 	}
2683 
2684 	kvmppc_core_prepare_to_enter(vcpu);
2685 
2686 	/* No need to go into the guest when all we'll do is come back out */
2687 	if (signal_pending(current)) {
2688 		run->exit_reason = KVM_EXIT_INTR;
2689 		return -EINTR;
2690 	}
2691 
2692 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2693 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2694 	smp_mb();
2695 
2696 	/* On the first time here, set up HTAB and VRMA */
2697 	if (!vcpu->kvm->arch.hpte_setup_done) {
2698 		r = kvmppc_hv_setup_htab_rma(vcpu);
2699 		if (r)
2700 			goto out;
2701 	}
2702 
2703 	flush_fp_to_thread(current);
2704 	flush_altivec_to_thread(current);
2705 	flush_vsx_to_thread(current);
2706 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2707 	vcpu->arch.pgdir = current->mm->pgd;
2708 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2709 
2710 	do {
2711 		r = kvmppc_run_vcpu(run, vcpu);
2712 
2713 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2714 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2715 			trace_kvm_hcall_enter(vcpu);
2716 			r = kvmppc_pseries_do_hcall(vcpu);
2717 			trace_kvm_hcall_exit(vcpu, r);
2718 			kvmppc_core_prepare_to_enter(vcpu);
2719 		} else if (r == RESUME_PAGE_FAULT) {
2720 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2721 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2722 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2723 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2724 		}
2725 	} while (is_kvmppc_resume_guest(r));
2726 
2727  out:
2728 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2729 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2730 	return r;
2731 }
2732 
2733 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2734 				     int linux_psize)
2735 {
2736 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2737 
2738 	if (!def->shift)
2739 		return;
2740 	(*sps)->page_shift = def->shift;
2741 	(*sps)->slb_enc = def->sllp;
2742 	(*sps)->enc[0].page_shift = def->shift;
2743 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2744 	/*
2745 	 * Add 16MB MPSS support if host supports it
2746 	 */
2747 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2748 		(*sps)->enc[1].page_shift = 24;
2749 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2750 	}
2751 	(*sps)++;
2752 }
2753 
2754 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2755 					 struct kvm_ppc_smmu_info *info)
2756 {
2757 	struct kvm_ppc_one_seg_page_size *sps;
2758 
2759 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2760 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2761 		info->flags |= KVM_PPC_1T_SEGMENTS;
2762 	info->slb_size = mmu_slb_size;
2763 
2764 	/* We only support these sizes for now, and no muti-size segments */
2765 	sps = &info->sps[0];
2766 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2767 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2768 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2769 
2770 	return 0;
2771 }
2772 
2773 /*
2774  * Get (and clear) the dirty memory log for a memory slot.
2775  */
2776 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2777 					 struct kvm_dirty_log *log)
2778 {
2779 	struct kvm_memslots *slots;
2780 	struct kvm_memory_slot *memslot;
2781 	int r;
2782 	unsigned long n;
2783 
2784 	mutex_lock(&kvm->slots_lock);
2785 
2786 	r = -EINVAL;
2787 	if (log->slot >= KVM_USER_MEM_SLOTS)
2788 		goto out;
2789 
2790 	slots = kvm_memslots(kvm);
2791 	memslot = id_to_memslot(slots, log->slot);
2792 	r = -ENOENT;
2793 	if (!memslot->dirty_bitmap)
2794 		goto out;
2795 
2796 	n = kvm_dirty_bitmap_bytes(memslot);
2797 	memset(memslot->dirty_bitmap, 0, n);
2798 
2799 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2800 	if (r)
2801 		goto out;
2802 
2803 	r = -EFAULT;
2804 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2805 		goto out;
2806 
2807 	r = 0;
2808 out:
2809 	mutex_unlock(&kvm->slots_lock);
2810 	return r;
2811 }
2812 
2813 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2814 					struct kvm_memory_slot *dont)
2815 {
2816 	if (!dont || free->arch.rmap != dont->arch.rmap) {
2817 		vfree(free->arch.rmap);
2818 		free->arch.rmap = NULL;
2819 	}
2820 }
2821 
2822 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2823 					 unsigned long npages)
2824 {
2825 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2826 	if (!slot->arch.rmap)
2827 		return -ENOMEM;
2828 
2829 	return 0;
2830 }
2831 
2832 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2833 					struct kvm_memory_slot *memslot,
2834 					const struct kvm_userspace_memory_region *mem)
2835 {
2836 	return 0;
2837 }
2838 
2839 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2840 				const struct kvm_userspace_memory_region *mem,
2841 				const struct kvm_memory_slot *old,
2842 				const struct kvm_memory_slot *new)
2843 {
2844 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2845 	struct kvm_memslots *slots;
2846 	struct kvm_memory_slot *memslot;
2847 
2848 	if (npages && old->npages) {
2849 		/*
2850 		 * If modifying a memslot, reset all the rmap dirty bits.
2851 		 * If this is a new memslot, we don't need to do anything
2852 		 * since the rmap array starts out as all zeroes,
2853 		 * i.e. no pages are dirty.
2854 		 */
2855 		slots = kvm_memslots(kvm);
2856 		memslot = id_to_memslot(slots, mem->slot);
2857 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2858 	}
2859 }
2860 
2861 /*
2862  * Update LPCR values in kvm->arch and in vcores.
2863  * Caller must hold kvm->lock.
2864  */
2865 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2866 {
2867 	long int i;
2868 	u32 cores_done = 0;
2869 
2870 	if ((kvm->arch.lpcr & mask) == lpcr)
2871 		return;
2872 
2873 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2874 
2875 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2876 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2877 		if (!vc)
2878 			continue;
2879 		spin_lock(&vc->lock);
2880 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2881 		spin_unlock(&vc->lock);
2882 		if (++cores_done >= kvm->arch.online_vcores)
2883 			break;
2884 	}
2885 }
2886 
2887 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2888 {
2889 	return;
2890 }
2891 
2892 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2893 {
2894 	int err = 0;
2895 	struct kvm *kvm = vcpu->kvm;
2896 	unsigned long hva;
2897 	struct kvm_memory_slot *memslot;
2898 	struct vm_area_struct *vma;
2899 	unsigned long lpcr = 0, senc;
2900 	unsigned long psize, porder;
2901 	int srcu_idx;
2902 
2903 	mutex_lock(&kvm->lock);
2904 	if (kvm->arch.hpte_setup_done)
2905 		goto out;	/* another vcpu beat us to it */
2906 
2907 	/* Allocate hashed page table (if not done already) and reset it */
2908 	if (!kvm->arch.hpt_virt) {
2909 		err = kvmppc_alloc_hpt(kvm, NULL);
2910 		if (err) {
2911 			pr_err("KVM: Couldn't alloc HPT\n");
2912 			goto out;
2913 		}
2914 	}
2915 
2916 	/* Look up the memslot for guest physical address 0 */
2917 	srcu_idx = srcu_read_lock(&kvm->srcu);
2918 	memslot = gfn_to_memslot(kvm, 0);
2919 
2920 	/* We must have some memory at 0 by now */
2921 	err = -EINVAL;
2922 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2923 		goto out_srcu;
2924 
2925 	/* Look up the VMA for the start of this memory slot */
2926 	hva = memslot->userspace_addr;
2927 	down_read(&current->mm->mmap_sem);
2928 	vma = find_vma(current->mm, hva);
2929 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2930 		goto up_out;
2931 
2932 	psize = vma_kernel_pagesize(vma);
2933 	porder = __ilog2(psize);
2934 
2935 	up_read(&current->mm->mmap_sem);
2936 
2937 	/* We can handle 4k, 64k or 16M pages in the VRMA */
2938 	err = -EINVAL;
2939 	if (!(psize == 0x1000 || psize == 0x10000 ||
2940 	      psize == 0x1000000))
2941 		goto out_srcu;
2942 
2943 	/* Update VRMASD field in the LPCR */
2944 	senc = slb_pgsize_encoding(psize);
2945 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2946 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
2947 	/* the -4 is to account for senc values starting at 0x10 */
2948 	lpcr = senc << (LPCR_VRMASD_SH - 4);
2949 
2950 	/* Create HPTEs in the hash page table for the VRMA */
2951 	kvmppc_map_vrma(vcpu, memslot, porder);
2952 
2953 	kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2954 
2955 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
2956 	smp_wmb();
2957 	kvm->arch.hpte_setup_done = 1;
2958 	err = 0;
2959  out_srcu:
2960 	srcu_read_unlock(&kvm->srcu, srcu_idx);
2961  out:
2962 	mutex_unlock(&kvm->lock);
2963 	return err;
2964 
2965  up_out:
2966 	up_read(&current->mm->mmap_sem);
2967 	goto out_srcu;
2968 }
2969 
2970 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2971 {
2972 	unsigned long lpcr, lpid;
2973 	char buf[32];
2974 
2975 	/* Allocate the guest's logical partition ID */
2976 
2977 	lpid = kvmppc_alloc_lpid();
2978 	if ((long)lpid < 0)
2979 		return -ENOMEM;
2980 	kvm->arch.lpid = lpid;
2981 
2982 	/*
2983 	 * Since we don't flush the TLB when tearing down a VM,
2984 	 * and this lpid might have previously been used,
2985 	 * make sure we flush on each core before running the new VM.
2986 	 */
2987 	cpumask_setall(&kvm->arch.need_tlb_flush);
2988 
2989 	/* Start out with the default set of hcalls enabled */
2990 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2991 	       sizeof(kvm->arch.enabled_hcalls));
2992 
2993 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2994 
2995 	/* Init LPCR for virtual RMA mode */
2996 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
2997 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2998 	lpcr &= LPCR_PECE | LPCR_LPES;
2999 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3000 		LPCR_VPM0 | LPCR_VPM1;
3001 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3002 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3003 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3004 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3005 		lpcr |= LPCR_ONL;
3006 	kvm->arch.lpcr = lpcr;
3007 
3008 	/*
3009 	 * Track that we now have a HV mode VM active. This blocks secondary
3010 	 * CPU threads from coming online.
3011 	 */
3012 	kvm_hv_vm_activated();
3013 
3014 	/*
3015 	 * Create a debugfs directory for the VM
3016 	 */
3017 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3018 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3019 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3020 		kvmppc_mmu_debugfs_init(kvm);
3021 
3022 	return 0;
3023 }
3024 
3025 static void kvmppc_free_vcores(struct kvm *kvm)
3026 {
3027 	long int i;
3028 
3029 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3030 		kfree(kvm->arch.vcores[i]);
3031 	kvm->arch.online_vcores = 0;
3032 }
3033 
3034 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3035 {
3036 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3037 
3038 	kvm_hv_vm_deactivated();
3039 
3040 	kvmppc_free_vcores(kvm);
3041 
3042 	kvmppc_free_hpt(kvm);
3043 }
3044 
3045 /* We don't need to emulate any privileged instructions or dcbz */
3046 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3047 				     unsigned int inst, int *advance)
3048 {
3049 	return EMULATE_FAIL;
3050 }
3051 
3052 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3053 					ulong spr_val)
3054 {
3055 	return EMULATE_FAIL;
3056 }
3057 
3058 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3059 					ulong *spr_val)
3060 {
3061 	return EMULATE_FAIL;
3062 }
3063 
3064 static int kvmppc_core_check_processor_compat_hv(void)
3065 {
3066 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3067 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3068 		return -EIO;
3069 	return 0;
3070 }
3071 
3072 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3073 				 unsigned int ioctl, unsigned long arg)
3074 {
3075 	struct kvm *kvm __maybe_unused = filp->private_data;
3076 	void __user *argp = (void __user *)arg;
3077 	long r;
3078 
3079 	switch (ioctl) {
3080 
3081 	case KVM_PPC_ALLOCATE_HTAB: {
3082 		u32 htab_order;
3083 
3084 		r = -EFAULT;
3085 		if (get_user(htab_order, (u32 __user *)argp))
3086 			break;
3087 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3088 		if (r)
3089 			break;
3090 		r = -EFAULT;
3091 		if (put_user(htab_order, (u32 __user *)argp))
3092 			break;
3093 		r = 0;
3094 		break;
3095 	}
3096 
3097 	case KVM_PPC_GET_HTAB_FD: {
3098 		struct kvm_get_htab_fd ghf;
3099 
3100 		r = -EFAULT;
3101 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
3102 			break;
3103 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3104 		break;
3105 	}
3106 
3107 	default:
3108 		r = -ENOTTY;
3109 	}
3110 
3111 	return r;
3112 }
3113 
3114 /*
3115  * List of hcall numbers to enable by default.
3116  * For compatibility with old userspace, we enable by default
3117  * all hcalls that were implemented before the hcall-enabling
3118  * facility was added.  Note this list should not include H_RTAS.
3119  */
3120 static unsigned int default_hcall_list[] = {
3121 	H_REMOVE,
3122 	H_ENTER,
3123 	H_READ,
3124 	H_PROTECT,
3125 	H_BULK_REMOVE,
3126 	H_GET_TCE,
3127 	H_PUT_TCE,
3128 	H_SET_DABR,
3129 	H_SET_XDABR,
3130 	H_CEDE,
3131 	H_PROD,
3132 	H_CONFER,
3133 	H_REGISTER_VPA,
3134 #ifdef CONFIG_KVM_XICS
3135 	H_EOI,
3136 	H_CPPR,
3137 	H_IPI,
3138 	H_IPOLL,
3139 	H_XIRR,
3140 	H_XIRR_X,
3141 #endif
3142 	0
3143 };
3144 
3145 static void init_default_hcalls(void)
3146 {
3147 	int i;
3148 	unsigned int hcall;
3149 
3150 	for (i = 0; default_hcall_list[i]; ++i) {
3151 		hcall = default_hcall_list[i];
3152 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3153 		__set_bit(hcall / 4, default_enabled_hcalls);
3154 	}
3155 }
3156 
3157 static struct kvmppc_ops kvm_ops_hv = {
3158 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3159 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3160 	.get_one_reg = kvmppc_get_one_reg_hv,
3161 	.set_one_reg = kvmppc_set_one_reg_hv,
3162 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
3163 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
3164 	.set_msr     = kvmppc_set_msr_hv,
3165 	.vcpu_run    = kvmppc_vcpu_run_hv,
3166 	.vcpu_create = kvmppc_core_vcpu_create_hv,
3167 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
3168 	.check_requests = kvmppc_core_check_requests_hv,
3169 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3170 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
3171 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3172 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3173 	.unmap_hva = kvm_unmap_hva_hv,
3174 	.unmap_hva_range = kvm_unmap_hva_range_hv,
3175 	.age_hva  = kvm_age_hva_hv,
3176 	.test_age_hva = kvm_test_age_hva_hv,
3177 	.set_spte_hva = kvm_set_spte_hva_hv,
3178 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
3179 	.free_memslot = kvmppc_core_free_memslot_hv,
3180 	.create_memslot = kvmppc_core_create_memslot_hv,
3181 	.init_vm =  kvmppc_core_init_vm_hv,
3182 	.destroy_vm = kvmppc_core_destroy_vm_hv,
3183 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3184 	.emulate_op = kvmppc_core_emulate_op_hv,
3185 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3186 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3187 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3188 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3189 	.hcall_implemented = kvmppc_hcall_impl_hv,
3190 };
3191 
3192 static int kvmppc_book3s_init_hv(void)
3193 {
3194 	int r;
3195 	/*
3196 	 * FIXME!! Do we need to check on all cpus ?
3197 	 */
3198 	r = kvmppc_core_check_processor_compat_hv();
3199 	if (r < 0)
3200 		return -ENODEV;
3201 
3202 	kvm_ops_hv.owner = THIS_MODULE;
3203 	kvmppc_hv_ops = &kvm_ops_hv;
3204 
3205 	init_default_hcalls();
3206 
3207 	init_vcore_lists();
3208 
3209 	r = kvmppc_mmu_hv_init();
3210 	return r;
3211 }
3212 
3213 static void kvmppc_book3s_exit_hv(void)
3214 {
3215 	kvmppc_hv_ops = NULL;
3216 }
3217 
3218 module_init(kvmppc_book3s_init_hv);
3219 module_exit(kvmppc_book3s_exit_hv);
3220 MODULE_LICENSE("GPL");
3221 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3222 MODULE_ALIAS("devname:kvm");
3223