• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Paul Mackerras <paulus@au1.ibm.com>
7  *    Alexander Graf <agraf@suse.de>
8  *    Kevin Wolf <mail@kevin-wolf.de>
9  *
10  * Description: KVM functions specific to running on Book 3S
11  * processors in hypervisor mode (specifically POWER7 and later).
12  *
13  * This file is derived from arch/powerpc/kvm/book3s.c,
14  * by Alexander Graf <agraf@suse.de>.
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License, version 2, as
18  * published by the Free Software Foundation.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/export.h>
28 #include <linux/fs.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpu.h>
31 #include <linux/cpumask.h>
32 #include <linux/spinlock.h>
33 #include <linux/page-flags.h>
34 #include <linux/srcu.h>
35 #include <linux/miscdevice.h>
36 #include <linux/debugfs.h>
37 
38 #include <asm/reg.h>
39 #include <asm/cputable.h>
40 #include <asm/cacheflush.h>
41 #include <asm/tlbflush.h>
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/kvm_ppc.h>
45 #include <asm/kvm_book3s.h>
46 #include <asm/mmu_context.h>
47 #include <asm/lppaca.h>
48 #include <asm/processor.h>
49 #include <asm/cputhreads.h>
50 #include <asm/page.h>
51 #include <asm/hvcall.h>
52 #include <asm/switch_to.h>
53 #include <asm/smp.h>
54 #include <asm/dbell.h>
55 #include <asm/hmi.h>
56 #include <asm/pnv-pci.h>
57 #include <linux/gfp.h>
58 #include <linux/vmalloc.h>
59 #include <linux/highmem.h>
60 #include <linux/hugetlb.h>
61 #include <linux/kvm_irqfd.h>
62 #include <linux/irqbypass.h>
63 #include <linux/module.h>
64 #include <linux/compiler.h>
65 
66 #include "book3s.h"
67 
68 #define CREATE_TRACE_POINTS
69 #include "trace_hv.h"
70 
71 /* #define EXIT_DEBUG */
72 /* #define EXIT_DEBUG_SIMPLE */
73 /* #define EXIT_DEBUG_INT */
74 
75 /* Used to indicate that a guest page fault needs to be handled */
76 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
77 /* Used to indicate that a guest passthrough interrupt needs to be handled */
78 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
79 
80 /* Used as a "null" value for timebase values */
81 #define TB_NIL	(~(u64)0)
82 
83 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
84 
85 static int dynamic_mt_modes = 6;
86 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
87 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
88 static int target_smt_mode;
89 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
90 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
91 
92 #ifdef CONFIG_KVM_XICS
93 static struct kernel_param_ops module_param_ops = {
94 	.set = param_set_int,
95 	.get = param_get_int,
96 };
97 
98 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
99 							S_IRUGO | S_IWUSR);
100 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
101 
102 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
103 							S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
105 #endif
106 
107 /* Maximum halt poll interval defaults to KVM_HALT_POLL_NS_DEFAULT */
108 static unsigned int halt_poll_max_ns = KVM_HALT_POLL_NS_DEFAULT;
109 module_param(halt_poll_max_ns, uint, S_IRUGO | S_IWUSR);
110 MODULE_PARM_DESC(halt_poll_max_ns, "Maximum halt poll time in ns");
111 
112 /* Factor by which the vcore halt poll interval is grown, default is to double
113  */
114 static unsigned int halt_poll_ns_grow = 2;
115 module_param(halt_poll_ns_grow, int, S_IRUGO);
116 MODULE_PARM_DESC(halt_poll_ns_grow, "Factor halt poll time is grown by");
117 
118 /* Factor by which the vcore halt poll interval is shrunk, default is to reset
119  */
120 static unsigned int halt_poll_ns_shrink;
121 module_param(halt_poll_ns_shrink, int, S_IRUGO);
122 MODULE_PARM_DESC(halt_poll_ns_shrink, "Factor halt poll time is shrunk by");
123 
124 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
125 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
126 
next_runnable_thread(struct kvmppc_vcore * vc,int * ip)127 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
128 		int *ip)
129 {
130 	int i = *ip;
131 	struct kvm_vcpu *vcpu;
132 
133 	while (++i < MAX_SMT_THREADS) {
134 		vcpu = READ_ONCE(vc->runnable_threads[i]);
135 		if (vcpu) {
136 			*ip = i;
137 			return vcpu;
138 		}
139 	}
140 	return NULL;
141 }
142 
143 /* Used to traverse the list of runnable threads for a given vcore */
144 #define for_each_runnable_thread(i, vcpu, vc) \
145 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
146 
kvmppc_ipi_thread(int cpu)147 static bool kvmppc_ipi_thread(int cpu)
148 {
149 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
150 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
151 		preempt_disable();
152 		if (cpu_first_thread_sibling(cpu) ==
153 		    cpu_first_thread_sibling(smp_processor_id())) {
154 			unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
155 			msg |= cpu_thread_in_core(cpu);
156 			smp_mb();
157 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
158 			preempt_enable();
159 			return true;
160 		}
161 		preempt_enable();
162 	}
163 
164 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
165 	if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
166 		xics_wake_cpu(cpu);
167 		return true;
168 	}
169 #endif
170 
171 	return false;
172 }
173 
kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu * vcpu)174 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
175 {
176 	int cpu;
177 	struct swait_queue_head *wqp;
178 
179 	wqp = kvm_arch_vcpu_wq(vcpu);
180 	if (swait_active(wqp)) {
181 		swake_up(wqp);
182 		++vcpu->stat.halt_wakeup;
183 	}
184 
185 	if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
186 		return;
187 
188 	/* CPU points to the first thread of the core */
189 	cpu = vcpu->cpu;
190 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
191 		smp_send_reschedule(cpu);
192 }
193 
194 /*
195  * We use the vcpu_load/put functions to measure stolen time.
196  * Stolen time is counted as time when either the vcpu is able to
197  * run as part of a virtual core, but the task running the vcore
198  * is preempted or sleeping, or when the vcpu needs something done
199  * in the kernel by the task running the vcpu, but that task is
200  * preempted or sleeping.  Those two things have to be counted
201  * separately, since one of the vcpu tasks will take on the job
202  * of running the core, and the other vcpu tasks in the vcore will
203  * sleep waiting for it to do that, but that sleep shouldn't count
204  * as stolen time.
205  *
206  * Hence we accumulate stolen time when the vcpu can run as part of
207  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
208  * needs its task to do other things in the kernel (for example,
209  * service a page fault) in busy_stolen.  We don't accumulate
210  * stolen time for a vcore when it is inactive, or for a vcpu
211  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
212  * a misnomer; it means that the vcpu task is not executing in
213  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
214  * the kernel.  We don't have any way of dividing up that time
215  * between time that the vcpu is genuinely stopped, time that
216  * the task is actively working on behalf of the vcpu, and time
217  * that the task is preempted, so we don't count any of it as
218  * stolen.
219  *
220  * Updates to busy_stolen are protected by arch.tbacct_lock;
221  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
222  * lock.  The stolen times are measured in units of timebase ticks.
223  * (Note that the != TB_NIL checks below are purely defensive;
224  * they should never fail.)
225  */
226 
kvmppc_core_start_stolen(struct kvmppc_vcore * vc)227 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
228 {
229 	unsigned long flags;
230 
231 	spin_lock_irqsave(&vc->stoltb_lock, flags);
232 	vc->preempt_tb = mftb();
233 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
234 }
235 
kvmppc_core_end_stolen(struct kvmppc_vcore * vc)236 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
237 {
238 	unsigned long flags;
239 
240 	spin_lock_irqsave(&vc->stoltb_lock, flags);
241 	if (vc->preempt_tb != TB_NIL) {
242 		vc->stolen_tb += mftb() - vc->preempt_tb;
243 		vc->preempt_tb = TB_NIL;
244 	}
245 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
246 }
247 
kvmppc_core_vcpu_load_hv(struct kvm_vcpu * vcpu,int cpu)248 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
249 {
250 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
251 	unsigned long flags;
252 
253 	/*
254 	 * We can test vc->runner without taking the vcore lock,
255 	 * because only this task ever sets vc->runner to this
256 	 * vcpu, and once it is set to this vcpu, only this task
257 	 * ever sets it to NULL.
258 	 */
259 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
260 		kvmppc_core_end_stolen(vc);
261 
262 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
263 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
264 	    vcpu->arch.busy_preempt != TB_NIL) {
265 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
266 		vcpu->arch.busy_preempt = TB_NIL;
267 	}
268 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
269 }
270 
kvmppc_core_vcpu_put_hv(struct kvm_vcpu * vcpu)271 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
272 {
273 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
274 	unsigned long flags;
275 
276 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
277 		kvmppc_core_start_stolen(vc);
278 
279 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
280 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
281 		vcpu->arch.busy_preempt = mftb();
282 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
283 }
284 
kvmppc_set_msr_hv(struct kvm_vcpu * vcpu,u64 msr)285 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
286 {
287 	/*
288 	 * Check for illegal transactional state bit combination
289 	 * and if we find it, force the TS field to a safe state.
290 	 */
291 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
292 		msr &= ~MSR_TS_MASK;
293 	vcpu->arch.shregs.msr = msr;
294 	kvmppc_end_cede(vcpu);
295 }
296 
kvmppc_set_pvr_hv(struct kvm_vcpu * vcpu,u32 pvr)297 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
298 {
299 	vcpu->arch.pvr = pvr;
300 }
301 
kvmppc_set_arch_compat(struct kvm_vcpu * vcpu,u32 arch_compat)302 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
303 {
304 	unsigned long pcr = 0;
305 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
306 
307 	if (arch_compat) {
308 		switch (arch_compat) {
309 		case PVR_ARCH_205:
310 			/*
311 			 * If an arch bit is set in PCR, all the defined
312 			 * higher-order arch bits also have to be set.
313 			 */
314 			pcr = PCR_ARCH_206 | PCR_ARCH_205;
315 			break;
316 		case PVR_ARCH_206:
317 		case PVR_ARCH_206p:
318 			pcr = PCR_ARCH_206;
319 			break;
320 		case PVR_ARCH_207:
321 			break;
322 		default:
323 			return -EINVAL;
324 		}
325 
326 		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
327 			/* POWER7 can't emulate POWER8 */
328 			if (!(pcr & PCR_ARCH_206))
329 				return -EINVAL;
330 			pcr &= ~PCR_ARCH_206;
331 		}
332 	}
333 
334 	spin_lock(&vc->lock);
335 	vc->arch_compat = arch_compat;
336 	vc->pcr = pcr;
337 	spin_unlock(&vc->lock);
338 
339 	return 0;
340 }
341 
kvmppc_dump_regs(struct kvm_vcpu * vcpu)342 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
343 {
344 	int r;
345 
346 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
347 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
348 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
349 	for (r = 0; r < 16; ++r)
350 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
351 		       r, kvmppc_get_gpr(vcpu, r),
352 		       r+16, kvmppc_get_gpr(vcpu, r+16));
353 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
354 	       vcpu->arch.ctr, vcpu->arch.lr);
355 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
356 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
357 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
358 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
359 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
360 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
361 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
362 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
363 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
364 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
365 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
366 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
367 	for (r = 0; r < vcpu->arch.slb_max; ++r)
368 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
369 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
370 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
371 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
372 	       vcpu->arch.last_inst);
373 }
374 
kvmppc_find_vcpu(struct kvm * kvm,int id)375 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
376 {
377 	struct kvm_vcpu *ret;
378 
379 	mutex_lock(&kvm->lock);
380 	ret = kvm_get_vcpu_by_id(kvm, id);
381 	mutex_unlock(&kvm->lock);
382 	return ret;
383 }
384 
init_vpa(struct kvm_vcpu * vcpu,struct lppaca * vpa)385 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
386 {
387 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
388 	vpa->yield_count = cpu_to_be32(1);
389 }
390 
set_vpa(struct kvm_vcpu * vcpu,struct kvmppc_vpa * v,unsigned long addr,unsigned long len)391 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
392 		   unsigned long addr, unsigned long len)
393 {
394 	/* check address is cacheline aligned */
395 	if (addr & (L1_CACHE_BYTES - 1))
396 		return -EINVAL;
397 	spin_lock(&vcpu->arch.vpa_update_lock);
398 	if (v->next_gpa != addr || v->len != len) {
399 		v->next_gpa = addr;
400 		v->len = addr ? len : 0;
401 		v->update_pending = 1;
402 	}
403 	spin_unlock(&vcpu->arch.vpa_update_lock);
404 	return 0;
405 }
406 
407 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
408 struct reg_vpa {
409 	u32 dummy;
410 	union {
411 		__be16 hword;
412 		__be32 word;
413 	} length;
414 };
415 
vpa_is_registered(struct kvmppc_vpa * vpap)416 static int vpa_is_registered(struct kvmppc_vpa *vpap)
417 {
418 	if (vpap->update_pending)
419 		return vpap->next_gpa != 0;
420 	return vpap->pinned_addr != NULL;
421 }
422 
do_h_register_vpa(struct kvm_vcpu * vcpu,unsigned long flags,unsigned long vcpuid,unsigned long vpa)423 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
424 				       unsigned long flags,
425 				       unsigned long vcpuid, unsigned long vpa)
426 {
427 	struct kvm *kvm = vcpu->kvm;
428 	unsigned long len, nb;
429 	void *va;
430 	struct kvm_vcpu *tvcpu;
431 	int err;
432 	int subfunc;
433 	struct kvmppc_vpa *vpap;
434 
435 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
436 	if (!tvcpu)
437 		return H_PARAMETER;
438 
439 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
440 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
441 	    subfunc == H_VPA_REG_SLB) {
442 		/* Registering new area - address must be cache-line aligned */
443 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
444 			return H_PARAMETER;
445 
446 		/* convert logical addr to kernel addr and read length */
447 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
448 		if (va == NULL)
449 			return H_PARAMETER;
450 		if (subfunc == H_VPA_REG_VPA)
451 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
452 		else
453 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
454 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
455 
456 		/* Check length */
457 		if (len > nb || len < sizeof(struct reg_vpa))
458 			return H_PARAMETER;
459 	} else {
460 		vpa = 0;
461 		len = 0;
462 	}
463 
464 	err = H_PARAMETER;
465 	vpap = NULL;
466 	spin_lock(&tvcpu->arch.vpa_update_lock);
467 
468 	switch (subfunc) {
469 	case H_VPA_REG_VPA:		/* register VPA */
470 		if (len < sizeof(struct lppaca))
471 			break;
472 		vpap = &tvcpu->arch.vpa;
473 		err = 0;
474 		break;
475 
476 	case H_VPA_REG_DTL:		/* register DTL */
477 		if (len < sizeof(struct dtl_entry))
478 			break;
479 		len -= len % sizeof(struct dtl_entry);
480 
481 		/* Check that they have previously registered a VPA */
482 		err = H_RESOURCE;
483 		if (!vpa_is_registered(&tvcpu->arch.vpa))
484 			break;
485 
486 		vpap = &tvcpu->arch.dtl;
487 		err = 0;
488 		break;
489 
490 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
491 		/* Check that they have previously registered a VPA */
492 		err = H_RESOURCE;
493 		if (!vpa_is_registered(&tvcpu->arch.vpa))
494 			break;
495 
496 		vpap = &tvcpu->arch.slb_shadow;
497 		err = 0;
498 		break;
499 
500 	case H_VPA_DEREG_VPA:		/* deregister VPA */
501 		/* Check they don't still have a DTL or SLB buf registered */
502 		err = H_RESOURCE;
503 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
504 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
505 			break;
506 
507 		vpap = &tvcpu->arch.vpa;
508 		err = 0;
509 		break;
510 
511 	case H_VPA_DEREG_DTL:		/* deregister DTL */
512 		vpap = &tvcpu->arch.dtl;
513 		err = 0;
514 		break;
515 
516 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
517 		vpap = &tvcpu->arch.slb_shadow;
518 		err = 0;
519 		break;
520 	}
521 
522 	if (vpap) {
523 		vpap->next_gpa = vpa;
524 		vpap->len = len;
525 		vpap->update_pending = 1;
526 	}
527 
528 	spin_unlock(&tvcpu->arch.vpa_update_lock);
529 
530 	return err;
531 }
532 
kvmppc_update_vpa(struct kvm_vcpu * vcpu,struct kvmppc_vpa * vpap)533 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
534 {
535 	struct kvm *kvm = vcpu->kvm;
536 	void *va;
537 	unsigned long nb;
538 	unsigned long gpa;
539 
540 	/*
541 	 * We need to pin the page pointed to by vpap->next_gpa,
542 	 * but we can't call kvmppc_pin_guest_page under the lock
543 	 * as it does get_user_pages() and down_read().  So we
544 	 * have to drop the lock, pin the page, then get the lock
545 	 * again and check that a new area didn't get registered
546 	 * in the meantime.
547 	 */
548 	for (;;) {
549 		gpa = vpap->next_gpa;
550 		spin_unlock(&vcpu->arch.vpa_update_lock);
551 		va = NULL;
552 		nb = 0;
553 		if (gpa)
554 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
555 		spin_lock(&vcpu->arch.vpa_update_lock);
556 		if (gpa == vpap->next_gpa)
557 			break;
558 		/* sigh... unpin that one and try again */
559 		if (va)
560 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
561 	}
562 
563 	vpap->update_pending = 0;
564 	if (va && nb < vpap->len) {
565 		/*
566 		 * If it's now too short, it must be that userspace
567 		 * has changed the mappings underlying guest memory,
568 		 * so unregister the region.
569 		 */
570 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
571 		va = NULL;
572 	}
573 	if (vpap->pinned_addr)
574 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
575 					vpap->dirty);
576 	vpap->gpa = gpa;
577 	vpap->pinned_addr = va;
578 	vpap->dirty = false;
579 	if (va)
580 		vpap->pinned_end = va + vpap->len;
581 }
582 
kvmppc_update_vpas(struct kvm_vcpu * vcpu)583 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
584 {
585 	if (!(vcpu->arch.vpa.update_pending ||
586 	      vcpu->arch.slb_shadow.update_pending ||
587 	      vcpu->arch.dtl.update_pending))
588 		return;
589 
590 	spin_lock(&vcpu->arch.vpa_update_lock);
591 	if (vcpu->arch.vpa.update_pending) {
592 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
593 		if (vcpu->arch.vpa.pinned_addr)
594 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
595 	}
596 	if (vcpu->arch.dtl.update_pending) {
597 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
598 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
599 		vcpu->arch.dtl_index = 0;
600 	}
601 	if (vcpu->arch.slb_shadow.update_pending)
602 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
603 	spin_unlock(&vcpu->arch.vpa_update_lock);
604 }
605 
606 /*
607  * Return the accumulated stolen time for the vcore up until `now'.
608  * The caller should hold the vcore lock.
609  */
vcore_stolen_time(struct kvmppc_vcore * vc,u64 now)610 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
611 {
612 	u64 p;
613 	unsigned long flags;
614 
615 	spin_lock_irqsave(&vc->stoltb_lock, flags);
616 	p = vc->stolen_tb;
617 	if (vc->vcore_state != VCORE_INACTIVE &&
618 	    vc->preempt_tb != TB_NIL)
619 		p += now - vc->preempt_tb;
620 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
621 	return p;
622 }
623 
kvmppc_create_dtl_entry(struct kvm_vcpu * vcpu,struct kvmppc_vcore * vc)624 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
625 				    struct kvmppc_vcore *vc)
626 {
627 	struct dtl_entry *dt;
628 	struct lppaca *vpa;
629 	unsigned long stolen;
630 	unsigned long core_stolen;
631 	u64 now;
632 
633 	dt = vcpu->arch.dtl_ptr;
634 	vpa = vcpu->arch.vpa.pinned_addr;
635 	now = mftb();
636 	core_stolen = vcore_stolen_time(vc, now);
637 	stolen = core_stolen - vcpu->arch.stolen_logged;
638 	vcpu->arch.stolen_logged = core_stolen;
639 	spin_lock_irq(&vcpu->arch.tbacct_lock);
640 	stolen += vcpu->arch.busy_stolen;
641 	vcpu->arch.busy_stolen = 0;
642 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
643 	if (!dt || !vpa)
644 		return;
645 	memset(dt, 0, sizeof(struct dtl_entry));
646 	dt->dispatch_reason = 7;
647 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
648 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
649 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
650 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
651 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
652 	++dt;
653 	if (dt == vcpu->arch.dtl.pinned_end)
654 		dt = vcpu->arch.dtl.pinned_addr;
655 	vcpu->arch.dtl_ptr = dt;
656 	/* order writing *dt vs. writing vpa->dtl_idx */
657 	smp_wmb();
658 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
659 	vcpu->arch.dtl.dirty = true;
660 }
661 
kvmppc_power8_compatible(struct kvm_vcpu * vcpu)662 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
663 {
664 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
665 		return true;
666 	if ((!vcpu->arch.vcore->arch_compat) &&
667 	    cpu_has_feature(CPU_FTR_ARCH_207S))
668 		return true;
669 	return false;
670 }
671 
kvmppc_h_set_mode(struct kvm_vcpu * vcpu,unsigned long mflags,unsigned long resource,unsigned long value1,unsigned long value2)672 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
673 			     unsigned long resource, unsigned long value1,
674 			     unsigned long value2)
675 {
676 	switch (resource) {
677 	case H_SET_MODE_RESOURCE_SET_CIABR:
678 		if (!kvmppc_power8_compatible(vcpu))
679 			return H_P2;
680 		if (value2)
681 			return H_P4;
682 		if (mflags)
683 			return H_UNSUPPORTED_FLAG_START;
684 		/* Guests can't breakpoint the hypervisor */
685 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
686 			return H_P3;
687 		vcpu->arch.ciabr  = value1;
688 		return H_SUCCESS;
689 	case H_SET_MODE_RESOURCE_SET_DAWR:
690 		if (!kvmppc_power8_compatible(vcpu))
691 			return H_P2;
692 		if (mflags)
693 			return H_UNSUPPORTED_FLAG_START;
694 		if (value2 & DABRX_HYP)
695 			return H_P4;
696 		vcpu->arch.dawr  = value1;
697 		vcpu->arch.dawrx = value2;
698 		return H_SUCCESS;
699 	default:
700 		return H_TOO_HARD;
701 	}
702 }
703 
kvm_arch_vcpu_yield_to(struct kvm_vcpu * target)704 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
705 {
706 	struct kvmppc_vcore *vcore = target->arch.vcore;
707 
708 	/*
709 	 * We expect to have been called by the real mode handler
710 	 * (kvmppc_rm_h_confer()) which would have directly returned
711 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
712 	 * have useful work to do and should not confer) so we don't
713 	 * recheck that here.
714 	 */
715 
716 	spin_lock(&vcore->lock);
717 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
718 	    vcore->vcore_state != VCORE_INACTIVE &&
719 	    vcore->runner)
720 		target = vcore->runner;
721 	spin_unlock(&vcore->lock);
722 
723 	return kvm_vcpu_yield_to(target);
724 }
725 
kvmppc_get_yield_count(struct kvm_vcpu * vcpu)726 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
727 {
728 	int yield_count = 0;
729 	struct lppaca *lppaca;
730 
731 	spin_lock(&vcpu->arch.vpa_update_lock);
732 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
733 	if (lppaca)
734 		yield_count = be32_to_cpu(lppaca->yield_count);
735 	spin_unlock(&vcpu->arch.vpa_update_lock);
736 	return yield_count;
737 }
738 
kvmppc_pseries_do_hcall(struct kvm_vcpu * vcpu)739 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
740 {
741 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
742 	unsigned long target, ret = H_SUCCESS;
743 	int yield_count;
744 	struct kvm_vcpu *tvcpu;
745 	int idx, rc;
746 
747 	if (req <= MAX_HCALL_OPCODE &&
748 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
749 		return RESUME_HOST;
750 
751 	switch (req) {
752 	case H_CEDE:
753 		break;
754 	case H_PROD:
755 		target = kvmppc_get_gpr(vcpu, 4);
756 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
757 		if (!tvcpu) {
758 			ret = H_PARAMETER;
759 			break;
760 		}
761 		tvcpu->arch.prodded = 1;
762 		smp_mb();
763 		if (vcpu->arch.ceded) {
764 			if (swait_active(&vcpu->wq)) {
765 				swake_up(&vcpu->wq);
766 				vcpu->stat.halt_wakeup++;
767 			}
768 		}
769 		break;
770 	case H_CONFER:
771 		target = kvmppc_get_gpr(vcpu, 4);
772 		if (target == -1)
773 			break;
774 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
775 		if (!tvcpu) {
776 			ret = H_PARAMETER;
777 			break;
778 		}
779 		yield_count = kvmppc_get_gpr(vcpu, 5);
780 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
781 			break;
782 		kvm_arch_vcpu_yield_to(tvcpu);
783 		break;
784 	case H_REGISTER_VPA:
785 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
786 					kvmppc_get_gpr(vcpu, 5),
787 					kvmppc_get_gpr(vcpu, 6));
788 		break;
789 	case H_RTAS:
790 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
791 			return RESUME_HOST;
792 
793 		idx = srcu_read_lock(&vcpu->kvm->srcu);
794 		rc = kvmppc_rtas_hcall(vcpu);
795 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
796 
797 		if (rc == -ENOENT)
798 			return RESUME_HOST;
799 		else if (rc == 0)
800 			break;
801 
802 		/* Send the error out to userspace via KVM_RUN */
803 		return rc;
804 	case H_LOGICAL_CI_LOAD:
805 		ret = kvmppc_h_logical_ci_load(vcpu);
806 		if (ret == H_TOO_HARD)
807 			return RESUME_HOST;
808 		break;
809 	case H_LOGICAL_CI_STORE:
810 		ret = kvmppc_h_logical_ci_store(vcpu);
811 		if (ret == H_TOO_HARD)
812 			return RESUME_HOST;
813 		break;
814 	case H_SET_MODE:
815 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
816 					kvmppc_get_gpr(vcpu, 5),
817 					kvmppc_get_gpr(vcpu, 6),
818 					kvmppc_get_gpr(vcpu, 7));
819 		if (ret == H_TOO_HARD)
820 			return RESUME_HOST;
821 		break;
822 	case H_XIRR:
823 	case H_CPPR:
824 	case H_EOI:
825 	case H_IPI:
826 	case H_IPOLL:
827 	case H_XIRR_X:
828 		if (kvmppc_xics_enabled(vcpu)) {
829 			ret = kvmppc_xics_hcall(vcpu, req);
830 			break;
831 		}
832 		return RESUME_HOST;
833 	case H_PUT_TCE:
834 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
835 						kvmppc_get_gpr(vcpu, 5),
836 						kvmppc_get_gpr(vcpu, 6));
837 		if (ret == H_TOO_HARD)
838 			return RESUME_HOST;
839 		break;
840 	case H_PUT_TCE_INDIRECT:
841 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
842 						kvmppc_get_gpr(vcpu, 5),
843 						kvmppc_get_gpr(vcpu, 6),
844 						kvmppc_get_gpr(vcpu, 7));
845 		if (ret == H_TOO_HARD)
846 			return RESUME_HOST;
847 		break;
848 	case H_STUFF_TCE:
849 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
850 						kvmppc_get_gpr(vcpu, 5),
851 						kvmppc_get_gpr(vcpu, 6),
852 						kvmppc_get_gpr(vcpu, 7));
853 		if (ret == H_TOO_HARD)
854 			return RESUME_HOST;
855 		break;
856 	default:
857 		return RESUME_HOST;
858 	}
859 	kvmppc_set_gpr(vcpu, 3, ret);
860 	vcpu->arch.hcall_needed = 0;
861 	return RESUME_GUEST;
862 }
863 
kvmppc_hcall_impl_hv(unsigned long cmd)864 static int kvmppc_hcall_impl_hv(unsigned long cmd)
865 {
866 	switch (cmd) {
867 	case H_CEDE:
868 	case H_PROD:
869 	case H_CONFER:
870 	case H_REGISTER_VPA:
871 	case H_SET_MODE:
872 	case H_LOGICAL_CI_LOAD:
873 	case H_LOGICAL_CI_STORE:
874 #ifdef CONFIG_KVM_XICS
875 	case H_XIRR:
876 	case H_CPPR:
877 	case H_EOI:
878 	case H_IPI:
879 	case H_IPOLL:
880 	case H_XIRR_X:
881 #endif
882 		return 1;
883 	}
884 
885 	/* See if it's in the real-mode table */
886 	return kvmppc_hcall_impl_hv_realmode(cmd);
887 }
888 
kvmppc_emulate_debug_inst(struct kvm_run * run,struct kvm_vcpu * vcpu)889 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
890 					struct kvm_vcpu *vcpu)
891 {
892 	u32 last_inst;
893 
894 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
895 					EMULATE_DONE) {
896 		/*
897 		 * Fetch failed, so return to guest and
898 		 * try executing it again.
899 		 */
900 		return RESUME_GUEST;
901 	}
902 
903 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
904 		run->exit_reason = KVM_EXIT_DEBUG;
905 		run->debug.arch.address = kvmppc_get_pc(vcpu);
906 		return RESUME_HOST;
907 	} else {
908 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
909 		return RESUME_GUEST;
910 	}
911 }
912 
kvmppc_handle_exit_hv(struct kvm_run * run,struct kvm_vcpu * vcpu,struct task_struct * tsk)913 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
914 				 struct task_struct *tsk)
915 {
916 	int r = RESUME_HOST;
917 
918 	vcpu->stat.sum_exits++;
919 
920 	/*
921 	 * This can happen if an interrupt occurs in the last stages
922 	 * of guest entry or the first stages of guest exit (i.e. after
923 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
924 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
925 	 * That can happen due to a bug, or due to a machine check
926 	 * occurring at just the wrong time.
927 	 */
928 	if (vcpu->arch.shregs.msr & MSR_HV) {
929 		printk(KERN_EMERG "KVM trap in HV mode!\n");
930 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
931 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
932 			vcpu->arch.shregs.msr);
933 		kvmppc_dump_regs(vcpu);
934 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
935 		run->hw.hardware_exit_reason = vcpu->arch.trap;
936 		return RESUME_HOST;
937 	}
938 	run->exit_reason = KVM_EXIT_UNKNOWN;
939 	run->ready_for_interrupt_injection = 1;
940 	switch (vcpu->arch.trap) {
941 	/* We're good on these - the host merely wanted to get our attention */
942 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
943 		vcpu->stat.dec_exits++;
944 		r = RESUME_GUEST;
945 		break;
946 	case BOOK3S_INTERRUPT_EXTERNAL:
947 	case BOOK3S_INTERRUPT_H_DOORBELL:
948 		vcpu->stat.ext_intr_exits++;
949 		r = RESUME_GUEST;
950 		break;
951 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
952 	case BOOK3S_INTERRUPT_HMI:
953 	case BOOK3S_INTERRUPT_PERFMON:
954 		r = RESUME_GUEST;
955 		break;
956 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
957 		/*
958 		 * Deliver a machine check interrupt to the guest.
959 		 * We have to do this, even if the host has handled the
960 		 * machine check, because machine checks use SRR0/1 and
961 		 * the interrupt might have trashed guest state in them.
962 		 */
963 		kvmppc_book3s_queue_irqprio(vcpu,
964 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
965 		r = RESUME_GUEST;
966 		break;
967 	case BOOK3S_INTERRUPT_PROGRAM:
968 	{
969 		ulong flags;
970 		/*
971 		 * Normally program interrupts are delivered directly
972 		 * to the guest by the hardware, but we can get here
973 		 * as a result of a hypervisor emulation interrupt
974 		 * (e40) getting turned into a 700 by BML RTAS.
975 		 */
976 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
977 		kvmppc_core_queue_program(vcpu, flags);
978 		r = RESUME_GUEST;
979 		break;
980 	}
981 	case BOOK3S_INTERRUPT_SYSCALL:
982 	{
983 		/* hcall - punt to userspace */
984 		int i;
985 
986 		/* hypercall with MSR_PR has already been handled in rmode,
987 		 * and never reaches here.
988 		 */
989 
990 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
991 		for (i = 0; i < 9; ++i)
992 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
993 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
994 		vcpu->arch.hcall_needed = 1;
995 		r = RESUME_HOST;
996 		break;
997 	}
998 	/*
999 	 * We get these next two if the guest accesses a page which it thinks
1000 	 * it has mapped but which is not actually present, either because
1001 	 * it is for an emulated I/O device or because the corresonding
1002 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1003 	 * have been handled already.
1004 	 */
1005 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1006 		r = RESUME_PAGE_FAULT;
1007 		break;
1008 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1009 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1010 		vcpu->arch.fault_dsisr = 0;
1011 		r = RESUME_PAGE_FAULT;
1012 		break;
1013 	/*
1014 	 * This occurs if the guest executes an illegal instruction.
1015 	 * If the guest debug is disabled, generate a program interrupt
1016 	 * to the guest. If guest debug is enabled, we need to check
1017 	 * whether the instruction is a software breakpoint instruction.
1018 	 * Accordingly return to Guest or Host.
1019 	 */
1020 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1021 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1022 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1023 				swab32(vcpu->arch.emul_inst) :
1024 				vcpu->arch.emul_inst;
1025 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1026 			r = kvmppc_emulate_debug_inst(run, vcpu);
1027 		} else {
1028 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1029 			r = RESUME_GUEST;
1030 		}
1031 		break;
1032 	/*
1033 	 * This occurs if the guest (kernel or userspace), does something that
1034 	 * is prohibited by HFSCR.  We just generate a program interrupt to
1035 	 * the guest.
1036 	 */
1037 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1038 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1039 		r = RESUME_GUEST;
1040 		break;
1041 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1042 		r = RESUME_PASSTHROUGH;
1043 		break;
1044 	default:
1045 		kvmppc_dump_regs(vcpu);
1046 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1047 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1048 			vcpu->arch.shregs.msr);
1049 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1050 		r = RESUME_HOST;
1051 		break;
1052 	}
1053 
1054 	return r;
1055 }
1056 
kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu * vcpu,struct kvm_sregs * sregs)1057 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1058 					    struct kvm_sregs *sregs)
1059 {
1060 	int i;
1061 
1062 	memset(sregs, 0, sizeof(struct kvm_sregs));
1063 	sregs->pvr = vcpu->arch.pvr;
1064 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1065 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1066 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1067 	}
1068 
1069 	return 0;
1070 }
1071 
kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu * vcpu,struct kvm_sregs * sregs)1072 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1073 					    struct kvm_sregs *sregs)
1074 {
1075 	int i, j;
1076 
1077 	/* Only accept the same PVR as the host's, since we can't spoof it */
1078 	if (sregs->pvr != vcpu->arch.pvr)
1079 		return -EINVAL;
1080 
1081 	j = 0;
1082 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1083 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1084 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1085 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1086 			++j;
1087 		}
1088 	}
1089 	vcpu->arch.slb_max = j;
1090 
1091 	return 0;
1092 }
1093 
kvmppc_set_lpcr(struct kvm_vcpu * vcpu,u64 new_lpcr,bool preserve_top32)1094 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1095 		bool preserve_top32)
1096 {
1097 	struct kvm *kvm = vcpu->kvm;
1098 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1099 	u64 mask;
1100 
1101 	mutex_lock(&kvm->lock);
1102 	spin_lock(&vc->lock);
1103 	/*
1104 	 * If ILE (interrupt little-endian) has changed, update the
1105 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1106 	 */
1107 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1108 		struct kvm_vcpu *vcpu;
1109 		int i;
1110 
1111 		kvm_for_each_vcpu(i, vcpu, kvm) {
1112 			if (vcpu->arch.vcore != vc)
1113 				continue;
1114 			if (new_lpcr & LPCR_ILE)
1115 				vcpu->arch.intr_msr |= MSR_LE;
1116 			else
1117 				vcpu->arch.intr_msr &= ~MSR_LE;
1118 		}
1119 	}
1120 
1121 	/*
1122 	 * Userspace can only modify DPFD (default prefetch depth),
1123 	 * ILE (interrupt little-endian) and TC (translation control).
1124 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1125 	 */
1126 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1127 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1128 		mask |= LPCR_AIL;
1129 
1130 	/* Broken 32-bit version of LPCR must not clear top bits */
1131 	if (preserve_top32)
1132 		mask &= 0xFFFFFFFF;
1133 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1134 	spin_unlock(&vc->lock);
1135 	mutex_unlock(&kvm->lock);
1136 }
1137 
kvmppc_get_one_reg_hv(struct kvm_vcpu * vcpu,u64 id,union kvmppc_one_reg * val)1138 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1139 				 union kvmppc_one_reg *val)
1140 {
1141 	int r = 0;
1142 	long int i;
1143 
1144 	switch (id) {
1145 	case KVM_REG_PPC_DEBUG_INST:
1146 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1147 		break;
1148 	case KVM_REG_PPC_HIOR:
1149 		*val = get_reg_val(id, 0);
1150 		break;
1151 	case KVM_REG_PPC_DABR:
1152 		*val = get_reg_val(id, vcpu->arch.dabr);
1153 		break;
1154 	case KVM_REG_PPC_DABRX:
1155 		*val = get_reg_val(id, vcpu->arch.dabrx);
1156 		break;
1157 	case KVM_REG_PPC_DSCR:
1158 		*val = get_reg_val(id, vcpu->arch.dscr);
1159 		break;
1160 	case KVM_REG_PPC_PURR:
1161 		*val = get_reg_val(id, vcpu->arch.purr);
1162 		break;
1163 	case KVM_REG_PPC_SPURR:
1164 		*val = get_reg_val(id, vcpu->arch.spurr);
1165 		break;
1166 	case KVM_REG_PPC_AMR:
1167 		*val = get_reg_val(id, vcpu->arch.amr);
1168 		break;
1169 	case KVM_REG_PPC_UAMOR:
1170 		*val = get_reg_val(id, vcpu->arch.uamor);
1171 		break;
1172 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1173 		i = id - KVM_REG_PPC_MMCR0;
1174 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1175 		break;
1176 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1177 		i = id - KVM_REG_PPC_PMC1;
1178 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1179 		break;
1180 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1181 		i = id - KVM_REG_PPC_SPMC1;
1182 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1183 		break;
1184 	case KVM_REG_PPC_SIAR:
1185 		*val = get_reg_val(id, vcpu->arch.siar);
1186 		break;
1187 	case KVM_REG_PPC_SDAR:
1188 		*val = get_reg_val(id, vcpu->arch.sdar);
1189 		break;
1190 	case KVM_REG_PPC_SIER:
1191 		*val = get_reg_val(id, vcpu->arch.sier);
1192 		break;
1193 	case KVM_REG_PPC_IAMR:
1194 		*val = get_reg_val(id, vcpu->arch.iamr);
1195 		break;
1196 	case KVM_REG_PPC_PSPB:
1197 		*val = get_reg_val(id, vcpu->arch.pspb);
1198 		break;
1199 	case KVM_REG_PPC_DPDES:
1200 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1201 		break;
1202 	case KVM_REG_PPC_VTB:
1203 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1204 		break;
1205 	case KVM_REG_PPC_DAWR:
1206 		*val = get_reg_val(id, vcpu->arch.dawr);
1207 		break;
1208 	case KVM_REG_PPC_DAWRX:
1209 		*val = get_reg_val(id, vcpu->arch.dawrx);
1210 		break;
1211 	case KVM_REG_PPC_CIABR:
1212 		*val = get_reg_val(id, vcpu->arch.ciabr);
1213 		break;
1214 	case KVM_REG_PPC_CSIGR:
1215 		*val = get_reg_val(id, vcpu->arch.csigr);
1216 		break;
1217 	case KVM_REG_PPC_TACR:
1218 		*val = get_reg_val(id, vcpu->arch.tacr);
1219 		break;
1220 	case KVM_REG_PPC_TCSCR:
1221 		*val = get_reg_val(id, vcpu->arch.tcscr);
1222 		break;
1223 	case KVM_REG_PPC_PID:
1224 		*val = get_reg_val(id, vcpu->arch.pid);
1225 		break;
1226 	case KVM_REG_PPC_ACOP:
1227 		*val = get_reg_val(id, vcpu->arch.acop);
1228 		break;
1229 	case KVM_REG_PPC_WORT:
1230 		*val = get_reg_val(id, vcpu->arch.wort);
1231 		break;
1232 	case KVM_REG_PPC_VPA_ADDR:
1233 		spin_lock(&vcpu->arch.vpa_update_lock);
1234 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1235 		spin_unlock(&vcpu->arch.vpa_update_lock);
1236 		break;
1237 	case KVM_REG_PPC_VPA_SLB:
1238 		spin_lock(&vcpu->arch.vpa_update_lock);
1239 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1240 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1241 		spin_unlock(&vcpu->arch.vpa_update_lock);
1242 		break;
1243 	case KVM_REG_PPC_VPA_DTL:
1244 		spin_lock(&vcpu->arch.vpa_update_lock);
1245 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1246 		val->vpaval.length = vcpu->arch.dtl.len;
1247 		spin_unlock(&vcpu->arch.vpa_update_lock);
1248 		break;
1249 	case KVM_REG_PPC_TB_OFFSET:
1250 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1251 		break;
1252 	case KVM_REG_PPC_LPCR:
1253 	case KVM_REG_PPC_LPCR_64:
1254 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1255 		break;
1256 	case KVM_REG_PPC_PPR:
1257 		*val = get_reg_val(id, vcpu->arch.ppr);
1258 		break;
1259 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1260 	case KVM_REG_PPC_TFHAR:
1261 		*val = get_reg_val(id, vcpu->arch.tfhar);
1262 		break;
1263 	case KVM_REG_PPC_TFIAR:
1264 		*val = get_reg_val(id, vcpu->arch.tfiar);
1265 		break;
1266 	case KVM_REG_PPC_TEXASR:
1267 		*val = get_reg_val(id, vcpu->arch.texasr);
1268 		break;
1269 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1270 		i = id - KVM_REG_PPC_TM_GPR0;
1271 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1272 		break;
1273 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1274 	{
1275 		int j;
1276 		i = id - KVM_REG_PPC_TM_VSR0;
1277 		if (i < 32)
1278 			for (j = 0; j < TS_FPRWIDTH; j++)
1279 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1280 		else {
1281 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1282 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1283 			else
1284 				r = -ENXIO;
1285 		}
1286 		break;
1287 	}
1288 	case KVM_REG_PPC_TM_CR:
1289 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1290 		break;
1291 	case KVM_REG_PPC_TM_XER:
1292 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1293 		break;
1294 	case KVM_REG_PPC_TM_LR:
1295 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1296 		break;
1297 	case KVM_REG_PPC_TM_CTR:
1298 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1299 		break;
1300 	case KVM_REG_PPC_TM_FPSCR:
1301 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1302 		break;
1303 	case KVM_REG_PPC_TM_AMR:
1304 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1305 		break;
1306 	case KVM_REG_PPC_TM_PPR:
1307 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1308 		break;
1309 	case KVM_REG_PPC_TM_VRSAVE:
1310 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1311 		break;
1312 	case KVM_REG_PPC_TM_VSCR:
1313 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1314 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1315 		else
1316 			r = -ENXIO;
1317 		break;
1318 	case KVM_REG_PPC_TM_DSCR:
1319 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1320 		break;
1321 	case KVM_REG_PPC_TM_TAR:
1322 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1323 		break;
1324 #endif
1325 	case KVM_REG_PPC_ARCH_COMPAT:
1326 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1327 		break;
1328 	default:
1329 		r = -EINVAL;
1330 		break;
1331 	}
1332 
1333 	return r;
1334 }
1335 
kvmppc_set_one_reg_hv(struct kvm_vcpu * vcpu,u64 id,union kvmppc_one_reg * val)1336 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1337 				 union kvmppc_one_reg *val)
1338 {
1339 	int r = 0;
1340 	long int i;
1341 	unsigned long addr, len;
1342 
1343 	switch (id) {
1344 	case KVM_REG_PPC_HIOR:
1345 		/* Only allow this to be set to zero */
1346 		if (set_reg_val(id, *val))
1347 			r = -EINVAL;
1348 		break;
1349 	case KVM_REG_PPC_DABR:
1350 		vcpu->arch.dabr = set_reg_val(id, *val);
1351 		break;
1352 	case KVM_REG_PPC_DABRX:
1353 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1354 		break;
1355 	case KVM_REG_PPC_DSCR:
1356 		vcpu->arch.dscr = set_reg_val(id, *val);
1357 		break;
1358 	case KVM_REG_PPC_PURR:
1359 		vcpu->arch.purr = set_reg_val(id, *val);
1360 		break;
1361 	case KVM_REG_PPC_SPURR:
1362 		vcpu->arch.spurr = set_reg_val(id, *val);
1363 		break;
1364 	case KVM_REG_PPC_AMR:
1365 		vcpu->arch.amr = set_reg_val(id, *val);
1366 		break;
1367 	case KVM_REG_PPC_UAMOR:
1368 		vcpu->arch.uamor = set_reg_val(id, *val);
1369 		break;
1370 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1371 		i = id - KVM_REG_PPC_MMCR0;
1372 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1373 		break;
1374 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1375 		i = id - KVM_REG_PPC_PMC1;
1376 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1377 		break;
1378 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1379 		i = id - KVM_REG_PPC_SPMC1;
1380 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1381 		break;
1382 	case KVM_REG_PPC_SIAR:
1383 		vcpu->arch.siar = set_reg_val(id, *val);
1384 		break;
1385 	case KVM_REG_PPC_SDAR:
1386 		vcpu->arch.sdar = set_reg_val(id, *val);
1387 		break;
1388 	case KVM_REG_PPC_SIER:
1389 		vcpu->arch.sier = set_reg_val(id, *val);
1390 		break;
1391 	case KVM_REG_PPC_IAMR:
1392 		vcpu->arch.iamr = set_reg_val(id, *val);
1393 		break;
1394 	case KVM_REG_PPC_PSPB:
1395 		vcpu->arch.pspb = set_reg_val(id, *val);
1396 		break;
1397 	case KVM_REG_PPC_DPDES:
1398 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1399 		break;
1400 	case KVM_REG_PPC_VTB:
1401 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1402 		break;
1403 	case KVM_REG_PPC_DAWR:
1404 		vcpu->arch.dawr = set_reg_val(id, *val);
1405 		break;
1406 	case KVM_REG_PPC_DAWRX:
1407 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1408 		break;
1409 	case KVM_REG_PPC_CIABR:
1410 		vcpu->arch.ciabr = set_reg_val(id, *val);
1411 		/* Don't allow setting breakpoints in hypervisor code */
1412 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1413 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1414 		break;
1415 	case KVM_REG_PPC_CSIGR:
1416 		vcpu->arch.csigr = set_reg_val(id, *val);
1417 		break;
1418 	case KVM_REG_PPC_TACR:
1419 		vcpu->arch.tacr = set_reg_val(id, *val);
1420 		break;
1421 	case KVM_REG_PPC_TCSCR:
1422 		vcpu->arch.tcscr = set_reg_val(id, *val);
1423 		break;
1424 	case KVM_REG_PPC_PID:
1425 		vcpu->arch.pid = set_reg_val(id, *val);
1426 		break;
1427 	case KVM_REG_PPC_ACOP:
1428 		vcpu->arch.acop = set_reg_val(id, *val);
1429 		break;
1430 	case KVM_REG_PPC_WORT:
1431 		vcpu->arch.wort = set_reg_val(id, *val);
1432 		break;
1433 	case KVM_REG_PPC_VPA_ADDR:
1434 		addr = set_reg_val(id, *val);
1435 		r = -EINVAL;
1436 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1437 			      vcpu->arch.dtl.next_gpa))
1438 			break;
1439 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1440 		break;
1441 	case KVM_REG_PPC_VPA_SLB:
1442 		addr = val->vpaval.addr;
1443 		len = val->vpaval.length;
1444 		r = -EINVAL;
1445 		if (addr && !vcpu->arch.vpa.next_gpa)
1446 			break;
1447 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1448 		break;
1449 	case KVM_REG_PPC_VPA_DTL:
1450 		addr = val->vpaval.addr;
1451 		len = val->vpaval.length;
1452 		r = -EINVAL;
1453 		if (addr && (len < sizeof(struct dtl_entry) ||
1454 			     !vcpu->arch.vpa.next_gpa))
1455 			break;
1456 		len -= len % sizeof(struct dtl_entry);
1457 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1458 		break;
1459 	case KVM_REG_PPC_TB_OFFSET:
1460 		/* round up to multiple of 2^24 */
1461 		vcpu->arch.vcore->tb_offset =
1462 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1463 		break;
1464 	case KVM_REG_PPC_LPCR:
1465 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1466 		break;
1467 	case KVM_REG_PPC_LPCR_64:
1468 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1469 		break;
1470 	case KVM_REG_PPC_PPR:
1471 		vcpu->arch.ppr = set_reg_val(id, *val);
1472 		break;
1473 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1474 	case KVM_REG_PPC_TFHAR:
1475 		vcpu->arch.tfhar = set_reg_val(id, *val);
1476 		break;
1477 	case KVM_REG_PPC_TFIAR:
1478 		vcpu->arch.tfiar = set_reg_val(id, *val);
1479 		break;
1480 	case KVM_REG_PPC_TEXASR:
1481 		vcpu->arch.texasr = set_reg_val(id, *val);
1482 		break;
1483 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1484 		i = id - KVM_REG_PPC_TM_GPR0;
1485 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1486 		break;
1487 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1488 	{
1489 		int j;
1490 		i = id - KVM_REG_PPC_TM_VSR0;
1491 		if (i < 32)
1492 			for (j = 0; j < TS_FPRWIDTH; j++)
1493 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1494 		else
1495 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1496 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1497 			else
1498 				r = -ENXIO;
1499 		break;
1500 	}
1501 	case KVM_REG_PPC_TM_CR:
1502 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1503 		break;
1504 	case KVM_REG_PPC_TM_XER:
1505 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1506 		break;
1507 	case KVM_REG_PPC_TM_LR:
1508 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1509 		break;
1510 	case KVM_REG_PPC_TM_CTR:
1511 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1512 		break;
1513 	case KVM_REG_PPC_TM_FPSCR:
1514 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1515 		break;
1516 	case KVM_REG_PPC_TM_AMR:
1517 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1518 		break;
1519 	case KVM_REG_PPC_TM_PPR:
1520 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1521 		break;
1522 	case KVM_REG_PPC_TM_VRSAVE:
1523 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1524 		break;
1525 	case KVM_REG_PPC_TM_VSCR:
1526 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1527 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1528 		else
1529 			r = - ENXIO;
1530 		break;
1531 	case KVM_REG_PPC_TM_DSCR:
1532 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1533 		break;
1534 	case KVM_REG_PPC_TM_TAR:
1535 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1536 		break;
1537 #endif
1538 	case KVM_REG_PPC_ARCH_COMPAT:
1539 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1540 		break;
1541 	default:
1542 		r = -EINVAL;
1543 		break;
1544 	}
1545 
1546 	return r;
1547 }
1548 
kvmppc_vcore_create(struct kvm * kvm,int core)1549 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1550 {
1551 	struct kvmppc_vcore *vcore;
1552 
1553 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1554 
1555 	if (vcore == NULL)
1556 		return NULL;
1557 
1558 	spin_lock_init(&vcore->lock);
1559 	spin_lock_init(&vcore->stoltb_lock);
1560 	init_swait_queue_head(&vcore->wq);
1561 	vcore->preempt_tb = TB_NIL;
1562 	vcore->lpcr = kvm->arch.lpcr;
1563 	vcore->first_vcpuid = core * threads_per_subcore;
1564 	vcore->kvm = kvm;
1565 	INIT_LIST_HEAD(&vcore->preempt_list);
1566 
1567 	return vcore;
1568 }
1569 
1570 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1571 static struct debugfs_timings_element {
1572 	const char *name;
1573 	size_t offset;
1574 } timings[] = {
1575 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1576 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1577 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1578 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1579 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1580 };
1581 
1582 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1583 
1584 struct debugfs_timings_state {
1585 	struct kvm_vcpu	*vcpu;
1586 	unsigned int	buflen;
1587 	char		buf[N_TIMINGS * 100];
1588 };
1589 
debugfs_timings_open(struct inode * inode,struct file * file)1590 static int debugfs_timings_open(struct inode *inode, struct file *file)
1591 {
1592 	struct kvm_vcpu *vcpu = inode->i_private;
1593 	struct debugfs_timings_state *p;
1594 
1595 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1596 	if (!p)
1597 		return -ENOMEM;
1598 
1599 	kvm_get_kvm(vcpu->kvm);
1600 	p->vcpu = vcpu;
1601 	file->private_data = p;
1602 
1603 	return nonseekable_open(inode, file);
1604 }
1605 
debugfs_timings_release(struct inode * inode,struct file * file)1606 static int debugfs_timings_release(struct inode *inode, struct file *file)
1607 {
1608 	struct debugfs_timings_state *p = file->private_data;
1609 
1610 	kvm_put_kvm(p->vcpu->kvm);
1611 	kfree(p);
1612 	return 0;
1613 }
1614 
debugfs_timings_read(struct file * file,char __user * buf,size_t len,loff_t * ppos)1615 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1616 				    size_t len, loff_t *ppos)
1617 {
1618 	struct debugfs_timings_state *p = file->private_data;
1619 	struct kvm_vcpu *vcpu = p->vcpu;
1620 	char *s, *buf_end;
1621 	struct kvmhv_tb_accumulator tb;
1622 	u64 count;
1623 	loff_t pos;
1624 	ssize_t n;
1625 	int i, loops;
1626 	bool ok;
1627 
1628 	if (!p->buflen) {
1629 		s = p->buf;
1630 		buf_end = s + sizeof(p->buf);
1631 		for (i = 0; i < N_TIMINGS; ++i) {
1632 			struct kvmhv_tb_accumulator *acc;
1633 
1634 			acc = (struct kvmhv_tb_accumulator *)
1635 				((unsigned long)vcpu + timings[i].offset);
1636 			ok = false;
1637 			for (loops = 0; loops < 1000; ++loops) {
1638 				count = acc->seqcount;
1639 				if (!(count & 1)) {
1640 					smp_rmb();
1641 					tb = *acc;
1642 					smp_rmb();
1643 					if (count == acc->seqcount) {
1644 						ok = true;
1645 						break;
1646 					}
1647 				}
1648 				udelay(1);
1649 			}
1650 			if (!ok)
1651 				snprintf(s, buf_end - s, "%s: stuck\n",
1652 					timings[i].name);
1653 			else
1654 				snprintf(s, buf_end - s,
1655 					"%s: %llu %llu %llu %llu\n",
1656 					timings[i].name, count / 2,
1657 					tb_to_ns(tb.tb_total),
1658 					tb_to_ns(tb.tb_min),
1659 					tb_to_ns(tb.tb_max));
1660 			s += strlen(s);
1661 		}
1662 		p->buflen = s - p->buf;
1663 	}
1664 
1665 	pos = *ppos;
1666 	if (pos >= p->buflen)
1667 		return 0;
1668 	if (len > p->buflen - pos)
1669 		len = p->buflen - pos;
1670 	n = copy_to_user(buf, p->buf + pos, len);
1671 	if (n) {
1672 		if (n == len)
1673 			return -EFAULT;
1674 		len -= n;
1675 	}
1676 	*ppos = pos + len;
1677 	return len;
1678 }
1679 
debugfs_timings_write(struct file * file,const char __user * buf,size_t len,loff_t * ppos)1680 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1681 				     size_t len, loff_t *ppos)
1682 {
1683 	return -EACCES;
1684 }
1685 
1686 static const struct file_operations debugfs_timings_ops = {
1687 	.owner	 = THIS_MODULE,
1688 	.open	 = debugfs_timings_open,
1689 	.release = debugfs_timings_release,
1690 	.read	 = debugfs_timings_read,
1691 	.write	 = debugfs_timings_write,
1692 	.llseek	 = generic_file_llseek,
1693 };
1694 
1695 /* Create a debugfs directory for the vcpu */
debugfs_vcpu_init(struct kvm_vcpu * vcpu,unsigned int id)1696 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1697 {
1698 	char buf[16];
1699 	struct kvm *kvm = vcpu->kvm;
1700 
1701 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1702 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1703 		return;
1704 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1705 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1706 		return;
1707 	vcpu->arch.debugfs_timings =
1708 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1709 				    vcpu, &debugfs_timings_ops);
1710 }
1711 
1712 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
debugfs_vcpu_init(struct kvm_vcpu * vcpu,unsigned int id)1713 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1714 {
1715 }
1716 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1717 
kvmppc_core_vcpu_create_hv(struct kvm * kvm,unsigned int id)1718 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1719 						   unsigned int id)
1720 {
1721 	struct kvm_vcpu *vcpu;
1722 	int err = -EINVAL;
1723 	int core;
1724 	struct kvmppc_vcore *vcore;
1725 
1726 	core = id / threads_per_subcore;
1727 	if (core >= KVM_MAX_VCORES)
1728 		goto out;
1729 
1730 	err = -ENOMEM;
1731 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1732 	if (!vcpu)
1733 		goto out;
1734 
1735 	err = kvm_vcpu_init(vcpu, kvm, id);
1736 	if (err)
1737 		goto free_vcpu;
1738 
1739 	vcpu->arch.shared = &vcpu->arch.shregs;
1740 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1741 	/*
1742 	 * The shared struct is never shared on HV,
1743 	 * so we can always use host endianness
1744 	 */
1745 #ifdef __BIG_ENDIAN__
1746 	vcpu->arch.shared_big_endian = true;
1747 #else
1748 	vcpu->arch.shared_big_endian = false;
1749 #endif
1750 #endif
1751 	vcpu->arch.mmcr[0] = MMCR0_FC;
1752 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1753 	/* default to host PVR, since we can't spoof it */
1754 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1755 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1756 	spin_lock_init(&vcpu->arch.tbacct_lock);
1757 	vcpu->arch.busy_preempt = TB_NIL;
1758 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1759 
1760 	kvmppc_mmu_book3s_hv_init(vcpu);
1761 
1762 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1763 
1764 	init_waitqueue_head(&vcpu->arch.cpu_run);
1765 
1766 	mutex_lock(&kvm->lock);
1767 	vcore = kvm->arch.vcores[core];
1768 	if (!vcore) {
1769 		vcore = kvmppc_vcore_create(kvm, core);
1770 		kvm->arch.vcores[core] = vcore;
1771 		kvm->arch.online_vcores++;
1772 	}
1773 	mutex_unlock(&kvm->lock);
1774 
1775 	if (!vcore)
1776 		goto free_vcpu;
1777 
1778 	spin_lock(&vcore->lock);
1779 	++vcore->num_threads;
1780 	spin_unlock(&vcore->lock);
1781 	vcpu->arch.vcore = vcore;
1782 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1783 	vcpu->arch.thread_cpu = -1;
1784 
1785 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1786 	kvmppc_sanity_check(vcpu);
1787 
1788 	debugfs_vcpu_init(vcpu, id);
1789 
1790 	return vcpu;
1791 
1792 free_vcpu:
1793 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1794 out:
1795 	return ERR_PTR(err);
1796 }
1797 
unpin_vpa(struct kvm * kvm,struct kvmppc_vpa * vpa)1798 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1799 {
1800 	if (vpa->pinned_addr)
1801 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1802 					vpa->dirty);
1803 }
1804 
kvmppc_core_vcpu_free_hv(struct kvm_vcpu * vcpu)1805 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1806 {
1807 	spin_lock(&vcpu->arch.vpa_update_lock);
1808 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1809 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1810 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1811 	spin_unlock(&vcpu->arch.vpa_update_lock);
1812 	kvm_vcpu_uninit(vcpu);
1813 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1814 }
1815 
kvmppc_core_check_requests_hv(struct kvm_vcpu * vcpu)1816 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1817 {
1818 	/* Indicate we want to get back into the guest */
1819 	return 1;
1820 }
1821 
kvmppc_set_timer(struct kvm_vcpu * vcpu)1822 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1823 {
1824 	unsigned long dec_nsec, now;
1825 
1826 	now = get_tb();
1827 	if (now > vcpu->arch.dec_expires) {
1828 		/* decrementer has already gone negative */
1829 		kvmppc_core_queue_dec(vcpu);
1830 		kvmppc_core_prepare_to_enter(vcpu);
1831 		return;
1832 	}
1833 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1834 		   / tb_ticks_per_sec;
1835 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1836 		      HRTIMER_MODE_REL);
1837 	vcpu->arch.timer_running = 1;
1838 }
1839 
kvmppc_end_cede(struct kvm_vcpu * vcpu)1840 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1841 {
1842 	vcpu->arch.ceded = 0;
1843 	if (vcpu->arch.timer_running) {
1844 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1845 		vcpu->arch.timer_running = 0;
1846 	}
1847 }
1848 
1849 extern void __kvmppc_vcore_entry(void);
1850 
kvmppc_remove_runnable(struct kvmppc_vcore * vc,struct kvm_vcpu * vcpu)1851 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1852 				   struct kvm_vcpu *vcpu)
1853 {
1854 	u64 now;
1855 
1856 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1857 		return;
1858 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1859 	now = mftb();
1860 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1861 		vcpu->arch.stolen_logged;
1862 	vcpu->arch.busy_preempt = now;
1863 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1864 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1865 	--vc->n_runnable;
1866 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
1867 }
1868 
kvmppc_grab_hwthread(int cpu)1869 static int kvmppc_grab_hwthread(int cpu)
1870 {
1871 	struct paca_struct *tpaca;
1872 	long timeout = 10000;
1873 
1874 	tpaca = &paca[cpu];
1875 
1876 	/* Ensure the thread won't go into the kernel if it wakes */
1877 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1878 	tpaca->kvm_hstate.kvm_vcore = NULL;
1879 	tpaca->kvm_hstate.napping = 0;
1880 	smp_wmb();
1881 	tpaca->kvm_hstate.hwthread_req = 1;
1882 
1883 	/*
1884 	 * If the thread is already executing in the kernel (e.g. handling
1885 	 * a stray interrupt), wait for it to get back to nap mode.
1886 	 * The smp_mb() is to ensure that our setting of hwthread_req
1887 	 * is visible before we look at hwthread_state, so if this
1888 	 * races with the code at system_reset_pSeries and the thread
1889 	 * misses our setting of hwthread_req, we are sure to see its
1890 	 * setting of hwthread_state, and vice versa.
1891 	 */
1892 	smp_mb();
1893 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1894 		if (--timeout <= 0) {
1895 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1896 			return -EBUSY;
1897 		}
1898 		udelay(1);
1899 	}
1900 	return 0;
1901 }
1902 
kvmppc_release_hwthread(int cpu)1903 static void kvmppc_release_hwthread(int cpu)
1904 {
1905 	struct paca_struct *tpaca;
1906 
1907 	tpaca = &paca[cpu];
1908 	tpaca->kvm_hstate.hwthread_req = 0;
1909 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1910 	tpaca->kvm_hstate.kvm_vcore = NULL;
1911 	tpaca->kvm_hstate.kvm_split_mode = NULL;
1912 }
1913 
kvmppc_start_thread(struct kvm_vcpu * vcpu,struct kvmppc_vcore * vc)1914 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1915 {
1916 	int cpu;
1917 	struct paca_struct *tpaca;
1918 	struct kvmppc_vcore *mvc = vc->master_vcore;
1919 
1920 	cpu = vc->pcpu;
1921 	if (vcpu) {
1922 		if (vcpu->arch.timer_running) {
1923 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1924 			vcpu->arch.timer_running = 0;
1925 		}
1926 		cpu += vcpu->arch.ptid;
1927 		vcpu->cpu = mvc->pcpu;
1928 		vcpu->arch.thread_cpu = cpu;
1929 	}
1930 	tpaca = &paca[cpu];
1931 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1932 	tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1933 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1934 	smp_wmb();
1935 	tpaca->kvm_hstate.kvm_vcore = mvc;
1936 	if (cpu != smp_processor_id())
1937 		kvmppc_ipi_thread(cpu);
1938 }
1939 
kvmppc_wait_for_nap(void)1940 static void kvmppc_wait_for_nap(void)
1941 {
1942 	int cpu = smp_processor_id();
1943 	int i, loops;
1944 
1945 	for (loops = 0; loops < 1000000; ++loops) {
1946 		/*
1947 		 * Check if all threads are finished.
1948 		 * We set the vcore pointer when starting a thread
1949 		 * and the thread clears it when finished, so we look
1950 		 * for any threads that still have a non-NULL vcore ptr.
1951 		 */
1952 		for (i = 1; i < threads_per_subcore; ++i)
1953 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
1954 				break;
1955 		if (i == threads_per_subcore) {
1956 			HMT_medium();
1957 			return;
1958 		}
1959 		HMT_low();
1960 	}
1961 	HMT_medium();
1962 	for (i = 1; i < threads_per_subcore; ++i)
1963 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
1964 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1965 }
1966 
1967 /*
1968  * Check that we are on thread 0 and that any other threads in
1969  * this core are off-line.  Then grab the threads so they can't
1970  * enter the kernel.
1971  */
on_primary_thread(void)1972 static int on_primary_thread(void)
1973 {
1974 	int cpu = smp_processor_id();
1975 	int thr;
1976 
1977 	/* Are we on a primary subcore? */
1978 	if (cpu_thread_in_subcore(cpu))
1979 		return 0;
1980 
1981 	thr = 0;
1982 	while (++thr < threads_per_subcore)
1983 		if (cpu_online(cpu + thr))
1984 			return 0;
1985 
1986 	/* Grab all hw threads so they can't go into the kernel */
1987 	for (thr = 1; thr < threads_per_subcore; ++thr) {
1988 		if (kvmppc_grab_hwthread(cpu + thr)) {
1989 			/* Couldn't grab one; let the others go */
1990 			do {
1991 				kvmppc_release_hwthread(cpu + thr);
1992 			} while (--thr > 0);
1993 			return 0;
1994 		}
1995 	}
1996 	return 1;
1997 }
1998 
1999 /*
2000  * A list of virtual cores for each physical CPU.
2001  * These are vcores that could run but their runner VCPU tasks are
2002  * (or may be) preempted.
2003  */
2004 struct preempted_vcore_list {
2005 	struct list_head	list;
2006 	spinlock_t		lock;
2007 };
2008 
2009 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2010 
init_vcore_lists(void)2011 static void init_vcore_lists(void)
2012 {
2013 	int cpu;
2014 
2015 	for_each_possible_cpu(cpu) {
2016 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2017 		spin_lock_init(&lp->lock);
2018 		INIT_LIST_HEAD(&lp->list);
2019 	}
2020 }
2021 
kvmppc_vcore_preempt(struct kvmppc_vcore * vc)2022 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2023 {
2024 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2025 
2026 	vc->vcore_state = VCORE_PREEMPT;
2027 	vc->pcpu = smp_processor_id();
2028 	if (vc->num_threads < threads_per_subcore) {
2029 		spin_lock(&lp->lock);
2030 		list_add_tail(&vc->preempt_list, &lp->list);
2031 		spin_unlock(&lp->lock);
2032 	}
2033 
2034 	/* Start accumulating stolen time */
2035 	kvmppc_core_start_stolen(vc);
2036 }
2037 
kvmppc_vcore_end_preempt(struct kvmppc_vcore * vc)2038 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2039 {
2040 	struct preempted_vcore_list *lp;
2041 
2042 	kvmppc_core_end_stolen(vc);
2043 	if (!list_empty(&vc->preempt_list)) {
2044 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2045 		spin_lock(&lp->lock);
2046 		list_del_init(&vc->preempt_list);
2047 		spin_unlock(&lp->lock);
2048 	}
2049 	vc->vcore_state = VCORE_INACTIVE;
2050 }
2051 
2052 /*
2053  * This stores information about the virtual cores currently
2054  * assigned to a physical core.
2055  */
2056 struct core_info {
2057 	int		n_subcores;
2058 	int		max_subcore_threads;
2059 	int		total_threads;
2060 	int		subcore_threads[MAX_SUBCORES];
2061 	struct kvm	*subcore_vm[MAX_SUBCORES];
2062 	struct list_head vcs[MAX_SUBCORES];
2063 };
2064 
2065 /*
2066  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2067  * respectively in 2-way micro-threading (split-core) mode.
2068  */
2069 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2070 
init_core_info(struct core_info * cip,struct kvmppc_vcore * vc)2071 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2072 {
2073 	int sub;
2074 
2075 	memset(cip, 0, sizeof(*cip));
2076 	cip->n_subcores = 1;
2077 	cip->max_subcore_threads = vc->num_threads;
2078 	cip->total_threads = vc->num_threads;
2079 	cip->subcore_threads[0] = vc->num_threads;
2080 	cip->subcore_vm[0] = vc->kvm;
2081 	for (sub = 0; sub < MAX_SUBCORES; ++sub)
2082 		INIT_LIST_HEAD(&cip->vcs[sub]);
2083 	list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2084 }
2085 
subcore_config_ok(int n_subcores,int n_threads)2086 static bool subcore_config_ok(int n_subcores, int n_threads)
2087 {
2088 	/* Can only dynamically split if unsplit to begin with */
2089 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2090 		return false;
2091 	if (n_subcores > MAX_SUBCORES)
2092 		return false;
2093 	if (n_subcores > 1) {
2094 		if (!(dynamic_mt_modes & 2))
2095 			n_subcores = 4;
2096 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2097 			return false;
2098 	}
2099 
2100 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2101 }
2102 
init_master_vcore(struct kvmppc_vcore * vc)2103 static void init_master_vcore(struct kvmppc_vcore *vc)
2104 {
2105 	vc->master_vcore = vc;
2106 	vc->entry_exit_map = 0;
2107 	vc->in_guest = 0;
2108 	vc->napping_threads = 0;
2109 	vc->conferring_threads = 0;
2110 }
2111 
can_dynamic_split(struct kvmppc_vcore * vc,struct core_info * cip)2112 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2113 {
2114 	int n_threads = vc->num_threads;
2115 	int sub;
2116 
2117 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2118 		return false;
2119 
2120 	if (n_threads < cip->max_subcore_threads)
2121 		n_threads = cip->max_subcore_threads;
2122 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2123 		return false;
2124 	cip->max_subcore_threads = n_threads;
2125 
2126 	sub = cip->n_subcores;
2127 	++cip->n_subcores;
2128 	cip->total_threads += vc->num_threads;
2129 	cip->subcore_threads[sub] = vc->num_threads;
2130 	cip->subcore_vm[sub] = vc->kvm;
2131 	init_master_vcore(vc);
2132 	list_del(&vc->preempt_list);
2133 	list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2134 
2135 	return true;
2136 }
2137 
2138 /*
2139  * Work out whether it is possible to piggyback the execution of
2140  * vcore *pvc onto the execution of the other vcores described in *cip.
2141  */
can_piggyback(struct kvmppc_vcore * pvc,struct core_info * cip,int target_threads)2142 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2143 			  int target_threads)
2144 {
2145 	if (cip->total_threads + pvc->num_threads > target_threads)
2146 		return false;
2147 
2148 	return can_dynamic_split(pvc, cip);
2149 }
2150 
prepare_threads(struct kvmppc_vcore * vc)2151 static void prepare_threads(struct kvmppc_vcore *vc)
2152 {
2153 	int i;
2154 	struct kvm_vcpu *vcpu;
2155 
2156 	for_each_runnable_thread(i, vcpu, vc) {
2157 		if (signal_pending(vcpu->arch.run_task))
2158 			vcpu->arch.ret = -EINTR;
2159 		else if (vcpu->arch.vpa.update_pending ||
2160 			 vcpu->arch.slb_shadow.update_pending ||
2161 			 vcpu->arch.dtl.update_pending)
2162 			vcpu->arch.ret = RESUME_GUEST;
2163 		else
2164 			continue;
2165 		kvmppc_remove_runnable(vc, vcpu);
2166 		wake_up(&vcpu->arch.cpu_run);
2167 	}
2168 }
2169 
collect_piggybacks(struct core_info * cip,int target_threads)2170 static void collect_piggybacks(struct core_info *cip, int target_threads)
2171 {
2172 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2173 	struct kvmppc_vcore *pvc, *vcnext;
2174 
2175 	spin_lock(&lp->lock);
2176 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2177 		if (!spin_trylock(&pvc->lock))
2178 			continue;
2179 		prepare_threads(pvc);
2180 		if (!pvc->n_runnable) {
2181 			list_del_init(&pvc->preempt_list);
2182 			if (pvc->runner == NULL) {
2183 				pvc->vcore_state = VCORE_INACTIVE;
2184 				kvmppc_core_end_stolen(pvc);
2185 			}
2186 			spin_unlock(&pvc->lock);
2187 			continue;
2188 		}
2189 		if (!can_piggyback(pvc, cip, target_threads)) {
2190 			spin_unlock(&pvc->lock);
2191 			continue;
2192 		}
2193 		kvmppc_core_end_stolen(pvc);
2194 		pvc->vcore_state = VCORE_PIGGYBACK;
2195 		if (cip->total_threads >= target_threads)
2196 			break;
2197 	}
2198 	spin_unlock(&lp->lock);
2199 }
2200 
post_guest_process(struct kvmppc_vcore * vc,bool is_master)2201 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2202 {
2203 	int still_running = 0, i;
2204 	u64 now;
2205 	long ret;
2206 	struct kvm_vcpu *vcpu;
2207 
2208 	spin_lock(&vc->lock);
2209 	now = get_tb();
2210 	for_each_runnable_thread(i, vcpu, vc) {
2211 		/* cancel pending dec exception if dec is positive */
2212 		if (now < vcpu->arch.dec_expires &&
2213 		    kvmppc_core_pending_dec(vcpu))
2214 			kvmppc_core_dequeue_dec(vcpu);
2215 
2216 		trace_kvm_guest_exit(vcpu);
2217 
2218 		ret = RESUME_GUEST;
2219 		if (vcpu->arch.trap)
2220 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2221 						    vcpu->arch.run_task);
2222 
2223 		vcpu->arch.ret = ret;
2224 		vcpu->arch.trap = 0;
2225 
2226 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2227 			if (vcpu->arch.pending_exceptions)
2228 				kvmppc_core_prepare_to_enter(vcpu);
2229 			if (vcpu->arch.ceded)
2230 				kvmppc_set_timer(vcpu);
2231 			else
2232 				++still_running;
2233 		} else {
2234 			kvmppc_remove_runnable(vc, vcpu);
2235 			wake_up(&vcpu->arch.cpu_run);
2236 		}
2237 	}
2238 	list_del_init(&vc->preempt_list);
2239 	if (!is_master) {
2240 		if (still_running > 0) {
2241 			kvmppc_vcore_preempt(vc);
2242 		} else if (vc->runner) {
2243 			vc->vcore_state = VCORE_PREEMPT;
2244 			kvmppc_core_start_stolen(vc);
2245 		} else {
2246 			vc->vcore_state = VCORE_INACTIVE;
2247 		}
2248 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2249 			/* make sure there's a candidate runner awake */
2250 			i = -1;
2251 			vcpu = next_runnable_thread(vc, &i);
2252 			wake_up(&vcpu->arch.cpu_run);
2253 		}
2254 	}
2255 	spin_unlock(&vc->lock);
2256 }
2257 
2258 /*
2259  * Clear core from the list of active host cores as we are about to
2260  * enter the guest. Only do this if it is the primary thread of the
2261  * core (not if a subcore) that is entering the guest.
2262  */
kvmppc_clear_host_core(int cpu)2263 static inline void kvmppc_clear_host_core(int cpu)
2264 {
2265 	int core;
2266 
2267 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2268 		return;
2269 	/*
2270 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2271 	 * later in kvmppc_start_thread and we need ensure that state is
2272 	 * visible to other CPUs only after we enter guest.
2273 	 */
2274 	core = cpu >> threads_shift;
2275 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2276 }
2277 
2278 /*
2279  * Advertise this core as an active host core since we exited the guest
2280  * Only need to do this if it is the primary thread of the core that is
2281  * exiting.
2282  */
kvmppc_set_host_core(int cpu)2283 static inline void kvmppc_set_host_core(int cpu)
2284 {
2285 	int core;
2286 
2287 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2288 		return;
2289 
2290 	/*
2291 	 * Memory barrier can be omitted here because we do a spin_unlock
2292 	 * immediately after this which provides the memory barrier.
2293 	 */
2294 	core = cpu >> threads_shift;
2295 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2296 }
2297 
2298 /*
2299  * Run a set of guest threads on a physical core.
2300  * Called with vc->lock held.
2301  */
kvmppc_run_core(struct kvmppc_vcore * vc)2302 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2303 {
2304 	struct kvm_vcpu *vcpu;
2305 	int i;
2306 	int srcu_idx;
2307 	struct core_info core_info;
2308 	struct kvmppc_vcore *pvc, *vcnext;
2309 	struct kvm_split_mode split_info, *sip;
2310 	int split, subcore_size, active;
2311 	int sub;
2312 	bool thr0_done;
2313 	unsigned long cmd_bit, stat_bit;
2314 	int pcpu, thr;
2315 	int target_threads;
2316 
2317 	/*
2318 	 * Remove from the list any threads that have a signal pending
2319 	 * or need a VPA update done
2320 	 */
2321 	prepare_threads(vc);
2322 
2323 	/* if the runner is no longer runnable, let the caller pick a new one */
2324 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2325 		return;
2326 
2327 	/*
2328 	 * Initialize *vc.
2329 	 */
2330 	init_master_vcore(vc);
2331 	vc->preempt_tb = TB_NIL;
2332 
2333 	/*
2334 	 * Make sure we are running on primary threads, and that secondary
2335 	 * threads are offline.  Also check if the number of threads in this
2336 	 * guest are greater than the current system threads per guest.
2337 	 */
2338 	if ((threads_per_core > 1) &&
2339 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2340 		for_each_runnable_thread(i, vcpu, vc) {
2341 			vcpu->arch.ret = -EBUSY;
2342 			kvmppc_remove_runnable(vc, vcpu);
2343 			wake_up(&vcpu->arch.cpu_run);
2344 		}
2345 		goto out;
2346 	}
2347 
2348 	/*
2349 	 * See if we could run any other vcores on the physical core
2350 	 * along with this one.
2351 	 */
2352 	init_core_info(&core_info, vc);
2353 	pcpu = smp_processor_id();
2354 	target_threads = threads_per_subcore;
2355 	if (target_smt_mode && target_smt_mode < target_threads)
2356 		target_threads = target_smt_mode;
2357 	if (vc->num_threads < target_threads)
2358 		collect_piggybacks(&core_info, target_threads);
2359 
2360 	/* Decide on micro-threading (split-core) mode */
2361 	subcore_size = threads_per_subcore;
2362 	cmd_bit = stat_bit = 0;
2363 	split = core_info.n_subcores;
2364 	sip = NULL;
2365 	if (split > 1) {
2366 		/* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2367 		if (split == 2 && (dynamic_mt_modes & 2)) {
2368 			cmd_bit = HID0_POWER8_1TO2LPAR;
2369 			stat_bit = HID0_POWER8_2LPARMODE;
2370 		} else {
2371 			split = 4;
2372 			cmd_bit = HID0_POWER8_1TO4LPAR;
2373 			stat_bit = HID0_POWER8_4LPARMODE;
2374 		}
2375 		subcore_size = MAX_SMT_THREADS / split;
2376 		sip = &split_info;
2377 		memset(&split_info, 0, sizeof(split_info));
2378 		split_info.rpr = mfspr(SPRN_RPR);
2379 		split_info.pmmar = mfspr(SPRN_PMMAR);
2380 		split_info.ldbar = mfspr(SPRN_LDBAR);
2381 		split_info.subcore_size = subcore_size;
2382 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2383 			split_info.master_vcs[sub] =
2384 				list_first_entry(&core_info.vcs[sub],
2385 					struct kvmppc_vcore, preempt_list);
2386 		/* order writes to split_info before kvm_split_mode pointer */
2387 		smp_wmb();
2388 	}
2389 	pcpu = smp_processor_id();
2390 	for (thr = 0; thr < threads_per_subcore; ++thr)
2391 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2392 
2393 	/* Initiate micro-threading (split-core) if required */
2394 	if (cmd_bit) {
2395 		unsigned long hid0 = mfspr(SPRN_HID0);
2396 
2397 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2398 		mb();
2399 		mtspr(SPRN_HID0, hid0);
2400 		isync();
2401 		for (;;) {
2402 			hid0 = mfspr(SPRN_HID0);
2403 			if (hid0 & stat_bit)
2404 				break;
2405 			cpu_relax();
2406 		}
2407 	}
2408 
2409 	kvmppc_clear_host_core(pcpu);
2410 
2411 	/* Start all the threads */
2412 	active = 0;
2413 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2414 		thr = subcore_thread_map[sub];
2415 		thr0_done = false;
2416 		active |= 1 << thr;
2417 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2418 			pvc->pcpu = pcpu + thr;
2419 			for_each_runnable_thread(i, vcpu, pvc) {
2420 				kvmppc_start_thread(vcpu, pvc);
2421 				kvmppc_create_dtl_entry(vcpu, pvc);
2422 				trace_kvm_guest_enter(vcpu);
2423 				if (!vcpu->arch.ptid)
2424 					thr0_done = true;
2425 				active |= 1 << (thr + vcpu->arch.ptid);
2426 			}
2427 			/*
2428 			 * We need to start the first thread of each subcore
2429 			 * even if it doesn't have a vcpu.
2430 			 */
2431 			if (pvc->master_vcore == pvc && !thr0_done)
2432 				kvmppc_start_thread(NULL, pvc);
2433 			thr += pvc->num_threads;
2434 		}
2435 	}
2436 
2437 	/*
2438 	 * Ensure that split_info.do_nap is set after setting
2439 	 * the vcore pointer in the PACA of the secondaries.
2440 	 */
2441 	smp_mb();
2442 	if (cmd_bit)
2443 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2444 
2445 	/*
2446 	 * When doing micro-threading, poke the inactive threads as well.
2447 	 * This gets them to the nap instruction after kvm_do_nap,
2448 	 * which reduces the time taken to unsplit later.
2449 	 */
2450 	if (split > 1)
2451 		for (thr = 1; thr < threads_per_subcore; ++thr)
2452 			if (!(active & (1 << thr)))
2453 				kvmppc_ipi_thread(pcpu + thr);
2454 
2455 	vc->vcore_state = VCORE_RUNNING;
2456 	preempt_disable();
2457 
2458 	trace_kvmppc_run_core(vc, 0);
2459 
2460 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2461 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2462 			spin_unlock(&pvc->lock);
2463 
2464 	guest_enter();
2465 
2466 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2467 
2468 	__kvmppc_vcore_entry();
2469 
2470 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2471 
2472 	spin_lock(&vc->lock);
2473 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2474 	vc->vcore_state = VCORE_EXITING;
2475 
2476 	/* wait for secondary threads to finish writing their state to memory */
2477 	kvmppc_wait_for_nap();
2478 
2479 	/* Return to whole-core mode if we split the core earlier */
2480 	if (split > 1) {
2481 		unsigned long hid0 = mfspr(SPRN_HID0);
2482 		unsigned long loops = 0;
2483 
2484 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2485 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2486 		mb();
2487 		mtspr(SPRN_HID0, hid0);
2488 		isync();
2489 		for (;;) {
2490 			hid0 = mfspr(SPRN_HID0);
2491 			if (!(hid0 & stat_bit))
2492 				break;
2493 			cpu_relax();
2494 			++loops;
2495 		}
2496 		split_info.do_nap = 0;
2497 	}
2498 
2499 	/* Let secondaries go back to the offline loop */
2500 	for (i = 0; i < threads_per_subcore; ++i) {
2501 		kvmppc_release_hwthread(pcpu + i);
2502 		if (sip && sip->napped[i])
2503 			kvmppc_ipi_thread(pcpu + i);
2504 	}
2505 
2506 	kvmppc_set_host_core(pcpu);
2507 
2508 	spin_unlock(&vc->lock);
2509 
2510 	/* make sure updates to secondary vcpu structs are visible now */
2511 	smp_mb();
2512 	guest_exit();
2513 
2514 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2515 		list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2516 					 preempt_list)
2517 			post_guest_process(pvc, pvc == vc);
2518 
2519 	spin_lock(&vc->lock);
2520 	preempt_enable();
2521 
2522  out:
2523 	vc->vcore_state = VCORE_INACTIVE;
2524 	trace_kvmppc_run_core(vc, 1);
2525 }
2526 
2527 /*
2528  * Wait for some other vcpu thread to execute us, and
2529  * wake us up when we need to handle something in the host.
2530  */
kvmppc_wait_for_exec(struct kvmppc_vcore * vc,struct kvm_vcpu * vcpu,int wait_state)2531 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2532 				 struct kvm_vcpu *vcpu, int wait_state)
2533 {
2534 	DEFINE_WAIT(wait);
2535 
2536 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2537 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2538 		spin_unlock(&vc->lock);
2539 		schedule();
2540 		spin_lock(&vc->lock);
2541 	}
2542 	finish_wait(&vcpu->arch.cpu_run, &wait);
2543 }
2544 
grow_halt_poll_ns(struct kvmppc_vcore * vc)2545 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2546 {
2547 	/* 10us base */
2548 	if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2549 		vc->halt_poll_ns = 10000;
2550 	else
2551 		vc->halt_poll_ns *= halt_poll_ns_grow;
2552 
2553 	if (vc->halt_poll_ns > halt_poll_max_ns)
2554 		vc->halt_poll_ns = halt_poll_max_ns;
2555 }
2556 
shrink_halt_poll_ns(struct kvmppc_vcore * vc)2557 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2558 {
2559 	if (halt_poll_ns_shrink == 0)
2560 		vc->halt_poll_ns = 0;
2561 	else
2562 		vc->halt_poll_ns /= halt_poll_ns_shrink;
2563 }
2564 
2565 /* Check to see if any of the runnable vcpus on the vcore have pending
2566  * exceptions or are no longer ceded
2567  */
kvmppc_vcore_check_block(struct kvmppc_vcore * vc)2568 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2569 {
2570 	struct kvm_vcpu *vcpu;
2571 	int i;
2572 
2573 	for_each_runnable_thread(i, vcpu, vc) {
2574 		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded)
2575 			return 1;
2576 	}
2577 
2578 	return 0;
2579 }
2580 
2581 /*
2582  * All the vcpus in this vcore are idle, so wait for a decrementer
2583  * or external interrupt to one of the vcpus.  vc->lock is held.
2584  */
kvmppc_vcore_blocked(struct kvmppc_vcore * vc)2585 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2586 {
2587 	ktime_t cur, start_poll, start_wait;
2588 	int do_sleep = 1;
2589 	u64 block_ns;
2590 	DECLARE_SWAITQUEUE(wait);
2591 
2592 	/* Poll for pending exceptions and ceded state */
2593 	cur = start_poll = ktime_get();
2594 	if (vc->halt_poll_ns) {
2595 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2596 		++vc->runner->stat.halt_attempted_poll;
2597 
2598 		vc->vcore_state = VCORE_POLLING;
2599 		spin_unlock(&vc->lock);
2600 
2601 		do {
2602 			if (kvmppc_vcore_check_block(vc)) {
2603 				do_sleep = 0;
2604 				break;
2605 			}
2606 			cur = ktime_get();
2607 		} while (single_task_running() && ktime_before(cur, stop));
2608 
2609 		spin_lock(&vc->lock);
2610 		vc->vcore_state = VCORE_INACTIVE;
2611 
2612 		if (!do_sleep) {
2613 			++vc->runner->stat.halt_successful_poll;
2614 			goto out;
2615 		}
2616 	}
2617 
2618 	prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2619 
2620 	if (kvmppc_vcore_check_block(vc)) {
2621 		finish_swait(&vc->wq, &wait);
2622 		do_sleep = 0;
2623 		/* If we polled, count this as a successful poll */
2624 		if (vc->halt_poll_ns)
2625 			++vc->runner->stat.halt_successful_poll;
2626 		goto out;
2627 	}
2628 
2629 	start_wait = ktime_get();
2630 
2631 	vc->vcore_state = VCORE_SLEEPING;
2632 	trace_kvmppc_vcore_blocked(vc, 0);
2633 	spin_unlock(&vc->lock);
2634 	schedule();
2635 	finish_swait(&vc->wq, &wait);
2636 	spin_lock(&vc->lock);
2637 	vc->vcore_state = VCORE_INACTIVE;
2638 	trace_kvmppc_vcore_blocked(vc, 1);
2639 	++vc->runner->stat.halt_successful_wait;
2640 
2641 	cur = ktime_get();
2642 
2643 out:
2644 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
2645 
2646 	/* Attribute wait time */
2647 	if (do_sleep) {
2648 		vc->runner->stat.halt_wait_ns +=
2649 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
2650 		/* Attribute failed poll time */
2651 		if (vc->halt_poll_ns)
2652 			vc->runner->stat.halt_poll_fail_ns +=
2653 				ktime_to_ns(start_wait) -
2654 				ktime_to_ns(start_poll);
2655 	} else {
2656 		/* Attribute successful poll time */
2657 		if (vc->halt_poll_ns)
2658 			vc->runner->stat.halt_poll_success_ns +=
2659 				ktime_to_ns(cur) -
2660 				ktime_to_ns(start_poll);
2661 	}
2662 
2663 	/* Adjust poll time */
2664 	if (halt_poll_max_ns) {
2665 		if (block_ns <= vc->halt_poll_ns)
2666 			;
2667 		/* We slept and blocked for longer than the max halt time */
2668 		else if (vc->halt_poll_ns && block_ns > halt_poll_max_ns)
2669 			shrink_halt_poll_ns(vc);
2670 		/* We slept and our poll time is too small */
2671 		else if (vc->halt_poll_ns < halt_poll_max_ns &&
2672 				block_ns < halt_poll_max_ns)
2673 			grow_halt_poll_ns(vc);
2674 	} else
2675 		vc->halt_poll_ns = 0;
2676 
2677 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
2678 }
2679 
kvmppc_run_vcpu(struct kvm_run * kvm_run,struct kvm_vcpu * vcpu)2680 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2681 {
2682 	int n_ceded, i;
2683 	struct kvmppc_vcore *vc;
2684 	struct kvm_vcpu *v;
2685 
2686 	trace_kvmppc_run_vcpu_enter(vcpu);
2687 
2688 	kvm_run->exit_reason = 0;
2689 	vcpu->arch.ret = RESUME_GUEST;
2690 	vcpu->arch.trap = 0;
2691 	kvmppc_update_vpas(vcpu);
2692 
2693 	/*
2694 	 * Synchronize with other threads in this virtual core
2695 	 */
2696 	vc = vcpu->arch.vcore;
2697 	spin_lock(&vc->lock);
2698 	vcpu->arch.ceded = 0;
2699 	vcpu->arch.run_task = current;
2700 	vcpu->arch.kvm_run = kvm_run;
2701 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2702 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2703 	vcpu->arch.busy_preempt = TB_NIL;
2704 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
2705 	++vc->n_runnable;
2706 
2707 	/*
2708 	 * This happens the first time this is called for a vcpu.
2709 	 * If the vcore is already running, we may be able to start
2710 	 * this thread straight away and have it join in.
2711 	 */
2712 	if (!signal_pending(current)) {
2713 		if (vc->vcore_state == VCORE_PIGGYBACK) {
2714 			struct kvmppc_vcore *mvc = vc->master_vcore;
2715 			if (spin_trylock(&mvc->lock)) {
2716 				if (mvc->vcore_state == VCORE_RUNNING &&
2717 				    !VCORE_IS_EXITING(mvc)) {
2718 					kvmppc_create_dtl_entry(vcpu, vc);
2719 					kvmppc_start_thread(vcpu, vc);
2720 					trace_kvm_guest_enter(vcpu);
2721 				}
2722 				spin_unlock(&mvc->lock);
2723 			}
2724 		} else if (vc->vcore_state == VCORE_RUNNING &&
2725 			   !VCORE_IS_EXITING(vc)) {
2726 			kvmppc_create_dtl_entry(vcpu, vc);
2727 			kvmppc_start_thread(vcpu, vc);
2728 			trace_kvm_guest_enter(vcpu);
2729 		} else if (vc->vcore_state == VCORE_SLEEPING) {
2730 			swake_up(&vc->wq);
2731 		}
2732 
2733 	}
2734 
2735 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2736 	       !signal_pending(current)) {
2737 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2738 			kvmppc_vcore_end_preempt(vc);
2739 
2740 		if (vc->vcore_state != VCORE_INACTIVE) {
2741 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2742 			continue;
2743 		}
2744 		for_each_runnable_thread(i, v, vc) {
2745 			kvmppc_core_prepare_to_enter(v);
2746 			if (signal_pending(v->arch.run_task)) {
2747 				kvmppc_remove_runnable(vc, v);
2748 				v->stat.signal_exits++;
2749 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2750 				v->arch.ret = -EINTR;
2751 				wake_up(&v->arch.cpu_run);
2752 			}
2753 		}
2754 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2755 			break;
2756 		n_ceded = 0;
2757 		for_each_runnable_thread(i, v, vc) {
2758 			if (!v->arch.pending_exceptions)
2759 				n_ceded += v->arch.ceded;
2760 			else
2761 				v->arch.ceded = 0;
2762 		}
2763 		vc->runner = vcpu;
2764 		if (n_ceded == vc->n_runnable) {
2765 			kvmppc_vcore_blocked(vc);
2766 		} else if (need_resched()) {
2767 			kvmppc_vcore_preempt(vc);
2768 			/* Let something else run */
2769 			cond_resched_lock(&vc->lock);
2770 			if (vc->vcore_state == VCORE_PREEMPT)
2771 				kvmppc_vcore_end_preempt(vc);
2772 		} else {
2773 			kvmppc_run_core(vc);
2774 		}
2775 		vc->runner = NULL;
2776 	}
2777 
2778 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2779 	       (vc->vcore_state == VCORE_RUNNING ||
2780 		vc->vcore_state == VCORE_EXITING ||
2781 		vc->vcore_state == VCORE_PIGGYBACK))
2782 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2783 
2784 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2785 		kvmppc_vcore_end_preempt(vc);
2786 
2787 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2788 		kvmppc_remove_runnable(vc, vcpu);
2789 		vcpu->stat.signal_exits++;
2790 		kvm_run->exit_reason = KVM_EXIT_INTR;
2791 		vcpu->arch.ret = -EINTR;
2792 	}
2793 
2794 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2795 		/* Wake up some vcpu to run the core */
2796 		i = -1;
2797 		v = next_runnable_thread(vc, &i);
2798 		wake_up(&v->arch.cpu_run);
2799 	}
2800 
2801 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2802 	spin_unlock(&vc->lock);
2803 	return vcpu->arch.ret;
2804 }
2805 
kvmppc_vcpu_run_hv(struct kvm_run * run,struct kvm_vcpu * vcpu)2806 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2807 {
2808 	int r;
2809 	int srcu_idx;
2810 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
2811 	unsigned long user_tar = 0;
2812 	unsigned int user_vrsave;
2813 
2814 	if (!vcpu->arch.sane) {
2815 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2816 		return -EINVAL;
2817 	}
2818 
2819 	/*
2820 	 * Don't allow entry with a suspended transaction, because
2821 	 * the guest entry/exit code will lose it.
2822 	 * If the guest has TM enabled, save away their TM-related SPRs
2823 	 * (they will get restored by the TM unavailable interrupt).
2824 	 */
2825 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2826 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
2827 	    (current->thread.regs->msr & MSR_TM)) {
2828 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
2829 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2830 			run->fail_entry.hardware_entry_failure_reason = 0;
2831 			return -EINVAL;
2832 		}
2833 		/* Enable TM so we can read the TM SPRs */
2834 		mtmsr(mfmsr() | MSR_TM);
2835 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
2836 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
2837 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
2838 		current->thread.regs->msr &= ~MSR_TM;
2839 	}
2840 #endif
2841 
2842 	kvmppc_core_prepare_to_enter(vcpu);
2843 
2844 	/* No need to go into the guest when all we'll do is come back out */
2845 	if (signal_pending(current)) {
2846 		run->exit_reason = KVM_EXIT_INTR;
2847 		return -EINTR;
2848 	}
2849 
2850 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2851 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2852 	smp_mb();
2853 
2854 	/* On the first time here, set up HTAB and VRMA */
2855 	if (!vcpu->kvm->arch.hpte_setup_done) {
2856 		r = kvmppc_hv_setup_htab_rma(vcpu);
2857 		if (r)
2858 			goto out;
2859 	}
2860 
2861 	flush_all_to_thread(current);
2862 
2863 	/* Save userspace EBB and other register values */
2864 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
2865 		ebb_regs[0] = mfspr(SPRN_EBBHR);
2866 		ebb_regs[1] = mfspr(SPRN_EBBRR);
2867 		ebb_regs[2] = mfspr(SPRN_BESCR);
2868 		user_tar = mfspr(SPRN_TAR);
2869 	}
2870 	user_vrsave = mfspr(SPRN_VRSAVE);
2871 
2872 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2873 	vcpu->arch.pgdir = current->mm->pgd;
2874 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2875 
2876 	do {
2877 		r = kvmppc_run_vcpu(run, vcpu);
2878 
2879 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2880 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2881 			trace_kvm_hcall_enter(vcpu);
2882 			r = kvmppc_pseries_do_hcall(vcpu);
2883 			trace_kvm_hcall_exit(vcpu, r);
2884 			kvmppc_core_prepare_to_enter(vcpu);
2885 		} else if (r == RESUME_PAGE_FAULT) {
2886 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2887 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2888 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2889 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2890 		} else if (r == RESUME_PASSTHROUGH)
2891 			r = kvmppc_xics_rm_complete(vcpu, 0);
2892 	} while (is_kvmppc_resume_guest(r));
2893 
2894 	/* Restore userspace EBB and other register values */
2895 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
2896 		mtspr(SPRN_EBBHR, ebb_regs[0]);
2897 		mtspr(SPRN_EBBRR, ebb_regs[1]);
2898 		mtspr(SPRN_BESCR, ebb_regs[2]);
2899 		mtspr(SPRN_TAR, user_tar);
2900 		mtspr(SPRN_FSCR, current->thread.fscr);
2901 	}
2902 	mtspr(SPRN_VRSAVE, user_vrsave);
2903 
2904  out:
2905 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2906 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2907 	return r;
2908 }
2909 
kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size ** sps,int linux_psize)2910 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2911 				     int linux_psize)
2912 {
2913 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2914 
2915 	if (!def->shift)
2916 		return;
2917 	(*sps)->page_shift = def->shift;
2918 	(*sps)->slb_enc = def->sllp;
2919 	(*sps)->enc[0].page_shift = def->shift;
2920 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2921 	/*
2922 	 * Add 16MB MPSS support if host supports it
2923 	 */
2924 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2925 		(*sps)->enc[1].page_shift = 24;
2926 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2927 	}
2928 	(*sps)++;
2929 }
2930 
kvm_vm_ioctl_get_smmu_info_hv(struct kvm * kvm,struct kvm_ppc_smmu_info * info)2931 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2932 					 struct kvm_ppc_smmu_info *info)
2933 {
2934 	struct kvm_ppc_one_seg_page_size *sps;
2935 
2936 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2937 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2938 		info->flags |= KVM_PPC_1T_SEGMENTS;
2939 	info->slb_size = mmu_slb_size;
2940 
2941 	/* We only support these sizes for now, and no muti-size segments */
2942 	sps = &info->sps[0];
2943 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2944 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2945 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2946 
2947 	return 0;
2948 }
2949 
2950 /*
2951  * Get (and clear) the dirty memory log for a memory slot.
2952  */
kvm_vm_ioctl_get_dirty_log_hv(struct kvm * kvm,struct kvm_dirty_log * log)2953 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2954 					 struct kvm_dirty_log *log)
2955 {
2956 	struct kvm_memslots *slots;
2957 	struct kvm_memory_slot *memslot;
2958 	int r;
2959 	unsigned long n;
2960 
2961 	mutex_lock(&kvm->slots_lock);
2962 
2963 	r = -EINVAL;
2964 	if (log->slot >= KVM_USER_MEM_SLOTS)
2965 		goto out;
2966 
2967 	slots = kvm_memslots(kvm);
2968 	memslot = id_to_memslot(slots, log->slot);
2969 	r = -ENOENT;
2970 	if (!memslot->dirty_bitmap)
2971 		goto out;
2972 
2973 	n = kvm_dirty_bitmap_bytes(memslot);
2974 	memset(memslot->dirty_bitmap, 0, n);
2975 
2976 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2977 	if (r)
2978 		goto out;
2979 
2980 	r = -EFAULT;
2981 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2982 		goto out;
2983 
2984 	r = 0;
2985 out:
2986 	mutex_unlock(&kvm->slots_lock);
2987 	return r;
2988 }
2989 
kvmppc_core_free_memslot_hv(struct kvm_memory_slot * free,struct kvm_memory_slot * dont)2990 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2991 					struct kvm_memory_slot *dont)
2992 {
2993 	if (!dont || free->arch.rmap != dont->arch.rmap) {
2994 		vfree(free->arch.rmap);
2995 		free->arch.rmap = NULL;
2996 	}
2997 }
2998 
kvmppc_core_create_memslot_hv(struct kvm_memory_slot * slot,unsigned long npages)2999 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3000 					 unsigned long npages)
3001 {
3002 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3003 	if (!slot->arch.rmap)
3004 		return -ENOMEM;
3005 
3006 	return 0;
3007 }
3008 
kvmppc_core_prepare_memory_region_hv(struct kvm * kvm,struct kvm_memory_slot * memslot,const struct kvm_userspace_memory_region * mem)3009 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3010 					struct kvm_memory_slot *memslot,
3011 					const struct kvm_userspace_memory_region *mem)
3012 {
3013 	return 0;
3014 }
3015 
kvmppc_core_commit_memory_region_hv(struct kvm * kvm,const struct kvm_userspace_memory_region * mem,const struct kvm_memory_slot * old,const struct kvm_memory_slot * new)3016 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3017 				const struct kvm_userspace_memory_region *mem,
3018 				const struct kvm_memory_slot *old,
3019 				const struct kvm_memory_slot *new)
3020 {
3021 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3022 	struct kvm_memslots *slots;
3023 	struct kvm_memory_slot *memslot;
3024 
3025 	if (npages && old->npages) {
3026 		/*
3027 		 * If modifying a memslot, reset all the rmap dirty bits.
3028 		 * If this is a new memslot, we don't need to do anything
3029 		 * since the rmap array starts out as all zeroes,
3030 		 * i.e. no pages are dirty.
3031 		 */
3032 		slots = kvm_memslots(kvm);
3033 		memslot = id_to_memslot(slots, mem->slot);
3034 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
3035 	}
3036 }
3037 
3038 /*
3039  * Update LPCR values in kvm->arch and in vcores.
3040  * Caller must hold kvm->lock.
3041  */
kvmppc_update_lpcr(struct kvm * kvm,unsigned long lpcr,unsigned long mask)3042 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3043 {
3044 	long int i;
3045 	u32 cores_done = 0;
3046 
3047 	if ((kvm->arch.lpcr & mask) == lpcr)
3048 		return;
3049 
3050 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3051 
3052 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
3053 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3054 		if (!vc)
3055 			continue;
3056 		spin_lock(&vc->lock);
3057 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3058 		spin_unlock(&vc->lock);
3059 		if (++cores_done >= kvm->arch.online_vcores)
3060 			break;
3061 	}
3062 }
3063 
kvmppc_mmu_destroy_hv(struct kvm_vcpu * vcpu)3064 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3065 {
3066 	return;
3067 }
3068 
kvmppc_hv_setup_htab_rma(struct kvm_vcpu * vcpu)3069 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3070 {
3071 	int err = 0;
3072 	struct kvm *kvm = vcpu->kvm;
3073 	unsigned long hva;
3074 	struct kvm_memory_slot *memslot;
3075 	struct vm_area_struct *vma;
3076 	unsigned long lpcr = 0, senc;
3077 	unsigned long psize, porder;
3078 	int srcu_idx;
3079 
3080 	mutex_lock(&kvm->lock);
3081 	if (kvm->arch.hpte_setup_done)
3082 		goto out;	/* another vcpu beat us to it */
3083 
3084 	/* Allocate hashed page table (if not done already) and reset it */
3085 	if (!kvm->arch.hpt_virt) {
3086 		err = kvmppc_alloc_hpt(kvm, NULL);
3087 		if (err) {
3088 			pr_err("KVM: Couldn't alloc HPT\n");
3089 			goto out;
3090 		}
3091 	}
3092 
3093 	/* Look up the memslot for guest physical address 0 */
3094 	srcu_idx = srcu_read_lock(&kvm->srcu);
3095 	memslot = gfn_to_memslot(kvm, 0);
3096 
3097 	/* We must have some memory at 0 by now */
3098 	err = -EINVAL;
3099 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3100 		goto out_srcu;
3101 
3102 	/* Look up the VMA for the start of this memory slot */
3103 	hva = memslot->userspace_addr;
3104 	down_read(&current->mm->mmap_sem);
3105 	vma = find_vma(current->mm, hva);
3106 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3107 		goto up_out;
3108 
3109 	psize = vma_kernel_pagesize(vma);
3110 	porder = __ilog2(psize);
3111 
3112 	up_read(&current->mm->mmap_sem);
3113 
3114 	/* We can handle 4k, 64k or 16M pages in the VRMA */
3115 	err = -EINVAL;
3116 	if (!(psize == 0x1000 || psize == 0x10000 ||
3117 	      psize == 0x1000000))
3118 		goto out_srcu;
3119 
3120 	/* Update VRMASD field in the LPCR */
3121 	senc = slb_pgsize_encoding(psize);
3122 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3123 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3124 	/* the -4 is to account for senc values starting at 0x10 */
3125 	lpcr = senc << (LPCR_VRMASD_SH - 4);
3126 
3127 	/* Create HPTEs in the hash page table for the VRMA */
3128 	kvmppc_map_vrma(vcpu, memslot, porder);
3129 
3130 	kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3131 
3132 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3133 	smp_wmb();
3134 	kvm->arch.hpte_setup_done = 1;
3135 	err = 0;
3136  out_srcu:
3137 	srcu_read_unlock(&kvm->srcu, srcu_idx);
3138  out:
3139 	mutex_unlock(&kvm->lock);
3140 	return err;
3141 
3142  up_out:
3143 	up_read(&current->mm->mmap_sem);
3144 	goto out_srcu;
3145 }
3146 
3147 #ifdef CONFIG_KVM_XICS
kvmppc_cpu_notify(struct notifier_block * self,unsigned long action,void * hcpu)3148 static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3149 			void *hcpu)
3150 {
3151 	unsigned long cpu = (long)hcpu;
3152 
3153 	switch (action) {
3154 	case CPU_UP_PREPARE:
3155 	case CPU_UP_PREPARE_FROZEN:
3156 		kvmppc_set_host_core(cpu);
3157 		break;
3158 
3159 #ifdef CONFIG_HOTPLUG_CPU
3160 	case CPU_DEAD:
3161 	case CPU_DEAD_FROZEN:
3162 	case CPU_UP_CANCELED:
3163 	case CPU_UP_CANCELED_FROZEN:
3164 		kvmppc_clear_host_core(cpu);
3165 		break;
3166 #endif
3167 	default:
3168 		break;
3169 	}
3170 
3171 	return NOTIFY_OK;
3172 }
3173 
3174 static struct notifier_block kvmppc_cpu_notifier = {
3175 	    .notifier_call = kvmppc_cpu_notify,
3176 };
3177 
3178 /*
3179  * Allocate a per-core structure for managing state about which cores are
3180  * running in the host versus the guest and for exchanging data between
3181  * real mode KVM and CPU running in the host.
3182  * This is only done for the first VM.
3183  * The allocated structure stays even if all VMs have stopped.
3184  * It is only freed when the kvm-hv module is unloaded.
3185  * It's OK for this routine to fail, we just don't support host
3186  * core operations like redirecting H_IPI wakeups.
3187  */
kvmppc_alloc_host_rm_ops(void)3188 void kvmppc_alloc_host_rm_ops(void)
3189 {
3190 	struct kvmppc_host_rm_ops *ops;
3191 	unsigned long l_ops;
3192 	int cpu, core;
3193 	int size;
3194 
3195 	/* Not the first time here ? */
3196 	if (kvmppc_host_rm_ops_hv != NULL)
3197 		return;
3198 
3199 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3200 	if (!ops)
3201 		return;
3202 
3203 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3204 	ops->rm_core = kzalloc(size, GFP_KERNEL);
3205 
3206 	if (!ops->rm_core) {
3207 		kfree(ops);
3208 		return;
3209 	}
3210 
3211 	get_online_cpus();
3212 
3213 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3214 		if (!cpu_online(cpu))
3215 			continue;
3216 
3217 		core = cpu >> threads_shift;
3218 		ops->rm_core[core].rm_state.in_host = 1;
3219 	}
3220 
3221 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3222 
3223 	/*
3224 	 * Make the contents of the kvmppc_host_rm_ops structure visible
3225 	 * to other CPUs before we assign it to the global variable.
3226 	 * Do an atomic assignment (no locks used here), but if someone
3227 	 * beats us to it, just free our copy and return.
3228 	 */
3229 	smp_wmb();
3230 	l_ops = (unsigned long) ops;
3231 
3232 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3233 		put_online_cpus();
3234 		kfree(ops->rm_core);
3235 		kfree(ops);
3236 		return;
3237 	}
3238 
3239 	register_cpu_notifier(&kvmppc_cpu_notifier);
3240 
3241 	put_online_cpus();
3242 }
3243 
kvmppc_free_host_rm_ops(void)3244 void kvmppc_free_host_rm_ops(void)
3245 {
3246 	if (kvmppc_host_rm_ops_hv) {
3247 		unregister_cpu_notifier(&kvmppc_cpu_notifier);
3248 		kfree(kvmppc_host_rm_ops_hv->rm_core);
3249 		kfree(kvmppc_host_rm_ops_hv);
3250 		kvmppc_host_rm_ops_hv = NULL;
3251 	}
3252 }
3253 #endif
3254 
kvmppc_core_init_vm_hv(struct kvm * kvm)3255 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3256 {
3257 	unsigned long lpcr, lpid;
3258 	char buf[32];
3259 
3260 	/* Allocate the guest's logical partition ID */
3261 
3262 	lpid = kvmppc_alloc_lpid();
3263 	if ((long)lpid < 0)
3264 		return -ENOMEM;
3265 	kvm->arch.lpid = lpid;
3266 
3267 	kvmppc_alloc_host_rm_ops();
3268 
3269 	/*
3270 	 * Since we don't flush the TLB when tearing down a VM,
3271 	 * and this lpid might have previously been used,
3272 	 * make sure we flush on each core before running the new VM.
3273 	 */
3274 	cpumask_setall(&kvm->arch.need_tlb_flush);
3275 
3276 	/* Start out with the default set of hcalls enabled */
3277 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3278 	       sizeof(kvm->arch.enabled_hcalls));
3279 
3280 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3281 
3282 	/* Init LPCR for virtual RMA mode */
3283 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
3284 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3285 	lpcr &= LPCR_PECE | LPCR_LPES;
3286 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3287 		LPCR_VPM0 | LPCR_VPM1;
3288 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3289 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3290 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3291 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3292 		lpcr |= LPCR_ONL;
3293 	kvm->arch.lpcr = lpcr;
3294 
3295 	/*
3296 	 * Track that we now have a HV mode VM active. This blocks secondary
3297 	 * CPU threads from coming online.
3298 	 */
3299 	kvm_hv_vm_activated();
3300 
3301 	/*
3302 	 * Create a debugfs directory for the VM
3303 	 */
3304 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3305 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3306 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3307 		kvmppc_mmu_debugfs_init(kvm);
3308 
3309 	return 0;
3310 }
3311 
kvmppc_free_vcores(struct kvm * kvm)3312 static void kvmppc_free_vcores(struct kvm *kvm)
3313 {
3314 	long int i;
3315 
3316 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3317 		kfree(kvm->arch.vcores[i]);
3318 	kvm->arch.online_vcores = 0;
3319 }
3320 
kvmppc_core_destroy_vm_hv(struct kvm * kvm)3321 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3322 {
3323 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3324 
3325 	kvm_hv_vm_deactivated();
3326 
3327 	kvmppc_free_vcores(kvm);
3328 
3329 	kvmppc_free_hpt(kvm);
3330 
3331 	kvmppc_free_pimap(kvm);
3332 }
3333 
3334 /* We don't need to emulate any privileged instructions or dcbz */
kvmppc_core_emulate_op_hv(struct kvm_run * run,struct kvm_vcpu * vcpu,unsigned int inst,int * advance)3335 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3336 				     unsigned int inst, int *advance)
3337 {
3338 	return EMULATE_FAIL;
3339 }
3340 
kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu * vcpu,int sprn,ulong spr_val)3341 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3342 					ulong spr_val)
3343 {
3344 	return EMULATE_FAIL;
3345 }
3346 
kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu * vcpu,int sprn,ulong * spr_val)3347 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3348 					ulong *spr_val)
3349 {
3350 	return EMULATE_FAIL;
3351 }
3352 
kvmppc_core_check_processor_compat_hv(void)3353 static int kvmppc_core_check_processor_compat_hv(void)
3354 {
3355 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3356 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3357 		return -EIO;
3358 	/*
3359 	 * Disable KVM for Power9, untill the required bits merged.
3360 	 */
3361 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3362 		return -EIO;
3363 
3364 	return 0;
3365 }
3366 
3367 #ifdef CONFIG_KVM_XICS
3368 
kvmppc_free_pimap(struct kvm * kvm)3369 void kvmppc_free_pimap(struct kvm *kvm)
3370 {
3371 	kfree(kvm->arch.pimap);
3372 }
3373 
kvmppc_alloc_pimap(void)3374 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3375 {
3376 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3377 }
3378 
kvmppc_set_passthru_irq(struct kvm * kvm,int host_irq,int guest_gsi)3379 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3380 {
3381 	struct irq_desc *desc;
3382 	struct kvmppc_irq_map *irq_map;
3383 	struct kvmppc_passthru_irqmap *pimap;
3384 	struct irq_chip *chip;
3385 	int i;
3386 
3387 	if (!kvm_irq_bypass)
3388 		return 1;
3389 
3390 	desc = irq_to_desc(host_irq);
3391 	if (!desc)
3392 		return -EIO;
3393 
3394 	mutex_lock(&kvm->lock);
3395 
3396 	pimap = kvm->arch.pimap;
3397 	if (pimap == NULL) {
3398 		/* First call, allocate structure to hold IRQ map */
3399 		pimap = kvmppc_alloc_pimap();
3400 		if (pimap == NULL) {
3401 			mutex_unlock(&kvm->lock);
3402 			return -ENOMEM;
3403 		}
3404 		kvm->arch.pimap = pimap;
3405 	}
3406 
3407 	/*
3408 	 * For now, we only support interrupts for which the EOI operation
3409 	 * is an OPAL call followed by a write to XIRR, since that's
3410 	 * what our real-mode EOI code does.
3411 	 */
3412 	chip = irq_data_get_irq_chip(&desc->irq_data);
3413 	if (!chip || !is_pnv_opal_msi(chip)) {
3414 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3415 			host_irq, guest_gsi);
3416 		mutex_unlock(&kvm->lock);
3417 		return -ENOENT;
3418 	}
3419 
3420 	/*
3421 	 * See if we already have an entry for this guest IRQ number.
3422 	 * If it's mapped to a hardware IRQ number, that's an error,
3423 	 * otherwise re-use this entry.
3424 	 */
3425 	for (i = 0; i < pimap->n_mapped; i++) {
3426 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
3427 			if (pimap->mapped[i].r_hwirq) {
3428 				mutex_unlock(&kvm->lock);
3429 				return -EINVAL;
3430 			}
3431 			break;
3432 		}
3433 	}
3434 
3435 	if (i == KVMPPC_PIRQ_MAPPED) {
3436 		mutex_unlock(&kvm->lock);
3437 		return -EAGAIN;		/* table is full */
3438 	}
3439 
3440 	irq_map = &pimap->mapped[i];
3441 
3442 	irq_map->v_hwirq = guest_gsi;
3443 	irq_map->desc = desc;
3444 
3445 	/*
3446 	 * Order the above two stores before the next to serialize with
3447 	 * the KVM real mode handler.
3448 	 */
3449 	smp_wmb();
3450 	irq_map->r_hwirq = desc->irq_data.hwirq;
3451 
3452 	if (i == pimap->n_mapped)
3453 		pimap->n_mapped++;
3454 
3455 	kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3456 
3457 	mutex_unlock(&kvm->lock);
3458 
3459 	return 0;
3460 }
3461 
kvmppc_clr_passthru_irq(struct kvm * kvm,int host_irq,int guest_gsi)3462 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3463 {
3464 	struct irq_desc *desc;
3465 	struct kvmppc_passthru_irqmap *pimap;
3466 	int i;
3467 
3468 	if (!kvm_irq_bypass)
3469 		return 0;
3470 
3471 	desc = irq_to_desc(host_irq);
3472 	if (!desc)
3473 		return -EIO;
3474 
3475 	mutex_lock(&kvm->lock);
3476 
3477 	if (kvm->arch.pimap == NULL) {
3478 		mutex_unlock(&kvm->lock);
3479 		return 0;
3480 	}
3481 	pimap = kvm->arch.pimap;
3482 
3483 	for (i = 0; i < pimap->n_mapped; i++) {
3484 		if (guest_gsi == pimap->mapped[i].v_hwirq)
3485 			break;
3486 	}
3487 
3488 	if (i == pimap->n_mapped) {
3489 		mutex_unlock(&kvm->lock);
3490 		return -ENODEV;
3491 	}
3492 
3493 	kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
3494 
3495 	/* invalidate the entry */
3496 	pimap->mapped[i].r_hwirq = 0;
3497 
3498 	/*
3499 	 * We don't free this structure even when the count goes to
3500 	 * zero. The structure is freed when we destroy the VM.
3501 	 */
3502 
3503 	mutex_unlock(&kvm->lock);
3504 	return 0;
3505 }
3506 
kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer * cons,struct irq_bypass_producer * prod)3507 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
3508 					     struct irq_bypass_producer *prod)
3509 {
3510 	int ret = 0;
3511 	struct kvm_kernel_irqfd *irqfd =
3512 		container_of(cons, struct kvm_kernel_irqfd, consumer);
3513 
3514 	irqfd->producer = prod;
3515 
3516 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3517 	if (ret)
3518 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
3519 			prod->irq, irqfd->gsi, ret);
3520 
3521 	return ret;
3522 }
3523 
kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer * cons,struct irq_bypass_producer * prod)3524 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
3525 					      struct irq_bypass_producer *prod)
3526 {
3527 	int ret;
3528 	struct kvm_kernel_irqfd *irqfd =
3529 		container_of(cons, struct kvm_kernel_irqfd, consumer);
3530 
3531 	irqfd->producer = NULL;
3532 
3533 	/*
3534 	 * When producer of consumer is unregistered, we change back to
3535 	 * default external interrupt handling mode - KVM real mode
3536 	 * will switch back to host.
3537 	 */
3538 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3539 	if (ret)
3540 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
3541 			prod->irq, irqfd->gsi, ret);
3542 }
3543 #endif
3544 
kvm_arch_vm_ioctl_hv(struct file * filp,unsigned int ioctl,unsigned long arg)3545 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3546 				 unsigned int ioctl, unsigned long arg)
3547 {
3548 	struct kvm *kvm __maybe_unused = filp->private_data;
3549 	void __user *argp = (void __user *)arg;
3550 	long r;
3551 
3552 	switch (ioctl) {
3553 
3554 	case KVM_PPC_ALLOCATE_HTAB: {
3555 		u32 htab_order;
3556 
3557 		r = -EFAULT;
3558 		if (get_user(htab_order, (u32 __user *)argp))
3559 			break;
3560 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3561 		if (r)
3562 			break;
3563 		r = -EFAULT;
3564 		if (put_user(htab_order, (u32 __user *)argp))
3565 			break;
3566 		r = 0;
3567 		break;
3568 	}
3569 
3570 	case KVM_PPC_GET_HTAB_FD: {
3571 		struct kvm_get_htab_fd ghf;
3572 
3573 		r = -EFAULT;
3574 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
3575 			break;
3576 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3577 		break;
3578 	}
3579 
3580 	default:
3581 		r = -ENOTTY;
3582 	}
3583 
3584 	return r;
3585 }
3586 
3587 /*
3588  * List of hcall numbers to enable by default.
3589  * For compatibility with old userspace, we enable by default
3590  * all hcalls that were implemented before the hcall-enabling
3591  * facility was added.  Note this list should not include H_RTAS.
3592  */
3593 static unsigned int default_hcall_list[] = {
3594 	H_REMOVE,
3595 	H_ENTER,
3596 	H_READ,
3597 	H_PROTECT,
3598 	H_BULK_REMOVE,
3599 	H_GET_TCE,
3600 	H_PUT_TCE,
3601 	H_SET_DABR,
3602 	H_SET_XDABR,
3603 	H_CEDE,
3604 	H_PROD,
3605 	H_CONFER,
3606 	H_REGISTER_VPA,
3607 #ifdef CONFIG_KVM_XICS
3608 	H_EOI,
3609 	H_CPPR,
3610 	H_IPI,
3611 	H_IPOLL,
3612 	H_XIRR,
3613 	H_XIRR_X,
3614 #endif
3615 	0
3616 };
3617 
init_default_hcalls(void)3618 static void init_default_hcalls(void)
3619 {
3620 	int i;
3621 	unsigned int hcall;
3622 
3623 	for (i = 0; default_hcall_list[i]; ++i) {
3624 		hcall = default_hcall_list[i];
3625 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3626 		__set_bit(hcall / 4, default_enabled_hcalls);
3627 	}
3628 }
3629 
3630 static struct kvmppc_ops kvm_ops_hv = {
3631 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3632 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3633 	.get_one_reg = kvmppc_get_one_reg_hv,
3634 	.set_one_reg = kvmppc_set_one_reg_hv,
3635 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
3636 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
3637 	.set_msr     = kvmppc_set_msr_hv,
3638 	.vcpu_run    = kvmppc_vcpu_run_hv,
3639 	.vcpu_create = kvmppc_core_vcpu_create_hv,
3640 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
3641 	.check_requests = kvmppc_core_check_requests_hv,
3642 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3643 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
3644 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3645 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3646 	.unmap_hva = kvm_unmap_hva_hv,
3647 	.unmap_hva_range = kvm_unmap_hva_range_hv,
3648 	.age_hva  = kvm_age_hva_hv,
3649 	.test_age_hva = kvm_test_age_hva_hv,
3650 	.set_spte_hva = kvm_set_spte_hva_hv,
3651 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
3652 	.free_memslot = kvmppc_core_free_memslot_hv,
3653 	.create_memslot = kvmppc_core_create_memslot_hv,
3654 	.init_vm =  kvmppc_core_init_vm_hv,
3655 	.destroy_vm = kvmppc_core_destroy_vm_hv,
3656 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3657 	.emulate_op = kvmppc_core_emulate_op_hv,
3658 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3659 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3660 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3661 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3662 	.hcall_implemented = kvmppc_hcall_impl_hv,
3663 #ifdef CONFIG_KVM_XICS
3664 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
3665 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
3666 #endif
3667 };
3668 
kvm_init_subcore_bitmap(void)3669 static int kvm_init_subcore_bitmap(void)
3670 {
3671 	int i, j;
3672 	int nr_cores = cpu_nr_cores();
3673 	struct sibling_subcore_state *sibling_subcore_state;
3674 
3675 	for (i = 0; i < nr_cores; i++) {
3676 		int first_cpu = i * threads_per_core;
3677 		int node = cpu_to_node(first_cpu);
3678 
3679 		/* Ignore if it is already allocated. */
3680 		if (paca[first_cpu].sibling_subcore_state)
3681 			continue;
3682 
3683 		sibling_subcore_state =
3684 			kmalloc_node(sizeof(struct sibling_subcore_state),
3685 							GFP_KERNEL, node);
3686 		if (!sibling_subcore_state)
3687 			return -ENOMEM;
3688 
3689 		memset(sibling_subcore_state, 0,
3690 				sizeof(struct sibling_subcore_state));
3691 
3692 		for (j = 0; j < threads_per_core; j++) {
3693 			int cpu = first_cpu + j;
3694 
3695 			paca[cpu].sibling_subcore_state = sibling_subcore_state;
3696 		}
3697 	}
3698 	return 0;
3699 }
3700 
kvmppc_book3s_init_hv(void)3701 static int kvmppc_book3s_init_hv(void)
3702 {
3703 	int r;
3704 	/*
3705 	 * FIXME!! Do we need to check on all cpus ?
3706 	 */
3707 	r = kvmppc_core_check_processor_compat_hv();
3708 	if (r < 0)
3709 		return -ENODEV;
3710 
3711 	r = kvm_init_subcore_bitmap();
3712 	if (r)
3713 		return r;
3714 
3715 	kvm_ops_hv.owner = THIS_MODULE;
3716 	kvmppc_hv_ops = &kvm_ops_hv;
3717 
3718 	init_default_hcalls();
3719 
3720 	init_vcore_lists();
3721 
3722 	r = kvmppc_mmu_hv_init();
3723 	return r;
3724 }
3725 
kvmppc_book3s_exit_hv(void)3726 static void kvmppc_book3s_exit_hv(void)
3727 {
3728 	kvmppc_free_host_rm_ops();
3729 	kvmppc_hv_ops = NULL;
3730 }
3731 
3732 module_init(kvmppc_book3s_init_hv);
3733 module_exit(kvmppc_book3s_exit_hv);
3734 MODULE_LICENSE("GPL");
3735 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3736 MODULE_ALIAS("devname:kvm");
3737