xref: /linux/arch/powerpc/kvm/powerpc.c (revision 33619f0d3ff715a2a5499520967d526ad931d70d)
1 /*
2  * This program is free software; you can redistribute it and/or modify
3  * it under the terms of the GNU General Public License, version 2, as
4  * published by the Free Software Foundation.
5  *
6  * This program is distributed in the hope that it will be useful,
7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9  * GNU General Public License for more details.
10  *
11  * You should have received a copy of the GNU General Public License
12  * along with this program; if not, write to the Free Software
13  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
14  *
15  * Copyright IBM Corp. 2007
16  *
17  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
18  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
19  */
20 
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/vmalloc.h>
26 #include <linux/hrtimer.h>
27 #include <linux/fs.h>
28 #include <linux/slab.h>
29 #include <asm/cputable.h>
30 #include <asm/uaccess.h>
31 #include <asm/kvm_ppc.h>
32 #include <asm/tlbflush.h>
33 #include "timing.h"
34 #include "../mm/mmu_decl.h"
35 
36 #define CREATE_TRACE_POINTS
37 #include "trace.h"
38 
39 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
40 {
41 	return !(v->arch.shared->msr & MSR_WE) ||
42 	       !!(v->arch.pending_exceptions);
43 }
44 
45 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
46 {
47 	int nr = kvmppc_get_gpr(vcpu, 11);
48 	int r;
49 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
50 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
51 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
52 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
53 	unsigned long r2 = 0;
54 
55 	if (!(vcpu->arch.shared->msr & MSR_SF)) {
56 		/* 32 bit mode */
57 		param1 &= 0xffffffff;
58 		param2 &= 0xffffffff;
59 		param3 &= 0xffffffff;
60 		param4 &= 0xffffffff;
61 	}
62 
63 	switch (nr) {
64 	case HC_VENDOR_KVM | KVM_HC_PPC_MAP_MAGIC_PAGE:
65 	{
66 		vcpu->arch.magic_page_pa = param1;
67 		vcpu->arch.magic_page_ea = param2;
68 
69 		r2 = KVM_MAGIC_FEAT_SR;
70 
71 		r = HC_EV_SUCCESS;
72 		break;
73 	}
74 	case HC_VENDOR_KVM | KVM_HC_FEATURES:
75 		r = HC_EV_SUCCESS;
76 #if defined(CONFIG_PPC_BOOK3S) /* XXX Missing magic page on BookE */
77 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
78 #endif
79 
80 		/* Second return value is in r4 */
81 		break;
82 	default:
83 		r = HC_EV_UNIMPLEMENTED;
84 		break;
85 	}
86 
87 	kvmppc_set_gpr(vcpu, 4, r2);
88 
89 	return r;
90 }
91 
92 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
93 {
94 	enum emulation_result er;
95 	int r;
96 
97 	er = kvmppc_emulate_instruction(run, vcpu);
98 	switch (er) {
99 	case EMULATE_DONE:
100 		/* Future optimization: only reload non-volatiles if they were
101 		 * actually modified. */
102 		r = RESUME_GUEST_NV;
103 		break;
104 	case EMULATE_DO_MMIO:
105 		run->exit_reason = KVM_EXIT_MMIO;
106 		/* We must reload nonvolatiles because "update" load/store
107 		 * instructions modify register state. */
108 		/* Future optimization: only reload non-volatiles if they were
109 		 * actually modified. */
110 		r = RESUME_HOST_NV;
111 		break;
112 	case EMULATE_FAIL:
113 		/* XXX Deliver Program interrupt to guest. */
114 		printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
115 		       kvmppc_get_last_inst(vcpu));
116 		r = RESUME_HOST;
117 		break;
118 	default:
119 		BUG();
120 	}
121 
122 	return r;
123 }
124 
125 int kvm_arch_hardware_enable(void *garbage)
126 {
127 	return 0;
128 }
129 
130 void kvm_arch_hardware_disable(void *garbage)
131 {
132 }
133 
134 int kvm_arch_hardware_setup(void)
135 {
136 	return 0;
137 }
138 
139 void kvm_arch_hardware_unsetup(void)
140 {
141 }
142 
143 void kvm_arch_check_processor_compat(void *rtn)
144 {
145 	*(int *)rtn = kvmppc_core_check_processor_compat();
146 }
147 
148 int kvm_arch_init_vm(struct kvm *kvm)
149 {
150 	return 0;
151 }
152 
153 void kvm_arch_destroy_vm(struct kvm *kvm)
154 {
155 	unsigned int i;
156 	struct kvm_vcpu *vcpu;
157 
158 	kvm_for_each_vcpu(i, vcpu, kvm)
159 		kvm_arch_vcpu_free(vcpu);
160 
161 	mutex_lock(&kvm->lock);
162 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
163 		kvm->vcpus[i] = NULL;
164 
165 	atomic_set(&kvm->online_vcpus, 0);
166 	mutex_unlock(&kvm->lock);
167 }
168 
169 void kvm_arch_sync_events(struct kvm *kvm)
170 {
171 }
172 
173 int kvm_dev_ioctl_check_extension(long ext)
174 {
175 	int r;
176 
177 	switch (ext) {
178 	case KVM_CAP_PPC_SEGSTATE:
179 	case KVM_CAP_PPC_PAIRED_SINGLES:
180 	case KVM_CAP_PPC_UNSET_IRQ:
181 	case KVM_CAP_PPC_IRQ_LEVEL:
182 	case KVM_CAP_ENABLE_CAP:
183 	case KVM_CAP_PPC_OSI:
184 	case KVM_CAP_PPC_GET_PVINFO:
185 		r = 1;
186 		break;
187 	case KVM_CAP_COALESCED_MMIO:
188 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
189 		break;
190 	default:
191 		r = 0;
192 		break;
193 	}
194 	return r;
195 
196 }
197 
198 long kvm_arch_dev_ioctl(struct file *filp,
199                         unsigned int ioctl, unsigned long arg)
200 {
201 	return -EINVAL;
202 }
203 
204 int kvm_arch_prepare_memory_region(struct kvm *kvm,
205                                    struct kvm_memory_slot *memslot,
206                                    struct kvm_memory_slot old,
207                                    struct kvm_userspace_memory_region *mem,
208                                    int user_alloc)
209 {
210 	return 0;
211 }
212 
213 void kvm_arch_commit_memory_region(struct kvm *kvm,
214                struct kvm_userspace_memory_region *mem,
215                struct kvm_memory_slot old,
216                int user_alloc)
217 {
218        return;
219 }
220 
221 
222 void kvm_arch_flush_shadow(struct kvm *kvm)
223 {
224 }
225 
226 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
227 {
228 	struct kvm_vcpu *vcpu;
229 	vcpu = kvmppc_core_vcpu_create(kvm, id);
230 	if (!IS_ERR(vcpu))
231 		kvmppc_create_vcpu_debugfs(vcpu, id);
232 	return vcpu;
233 }
234 
235 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
236 {
237 	/* Make sure we're not using the vcpu anymore */
238 	hrtimer_cancel(&vcpu->arch.dec_timer);
239 	tasklet_kill(&vcpu->arch.tasklet);
240 
241 	kvmppc_remove_vcpu_debugfs(vcpu);
242 	kvmppc_core_vcpu_free(vcpu);
243 }
244 
245 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
246 {
247 	kvm_arch_vcpu_free(vcpu);
248 }
249 
250 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
251 {
252 	return kvmppc_core_pending_dec(vcpu);
253 }
254 
255 static void kvmppc_decrementer_func(unsigned long data)
256 {
257 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
258 
259 	kvmppc_core_queue_dec(vcpu);
260 
261 	if (waitqueue_active(&vcpu->wq)) {
262 		wake_up_interruptible(&vcpu->wq);
263 		vcpu->stat.halt_wakeup++;
264 	}
265 }
266 
267 /*
268  * low level hrtimer wake routine. Because this runs in hardirq context
269  * we schedule a tasklet to do the real work.
270  */
271 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
272 {
273 	struct kvm_vcpu *vcpu;
274 
275 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
276 	tasklet_schedule(&vcpu->arch.tasklet);
277 
278 	return HRTIMER_NORESTART;
279 }
280 
281 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
282 {
283 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
284 	tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
285 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
286 
287 	return 0;
288 }
289 
290 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
291 {
292 	kvmppc_mmu_destroy(vcpu);
293 }
294 
295 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
296 {
297 	kvmppc_core_vcpu_load(vcpu, cpu);
298 }
299 
300 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
301 {
302 	kvmppc_core_vcpu_put(vcpu);
303 }
304 
305 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
306                                         struct kvm_guest_debug *dbg)
307 {
308 	return -EINVAL;
309 }
310 
311 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
312                                      struct kvm_run *run)
313 {
314 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
315 }
316 
317 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
318                                       struct kvm_run *run)
319 {
320 	u64 uninitialized_var(gpr);
321 
322 	if (run->mmio.len > sizeof(gpr)) {
323 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
324 		return;
325 	}
326 
327 	if (vcpu->arch.mmio_is_bigendian) {
328 		switch (run->mmio.len) {
329 		case 8: gpr = *(u64 *)run->mmio.data; break;
330 		case 4: gpr = *(u32 *)run->mmio.data; break;
331 		case 2: gpr = *(u16 *)run->mmio.data; break;
332 		case 1: gpr = *(u8 *)run->mmio.data; break;
333 		}
334 	} else {
335 		/* Convert BE data from userland back to LE. */
336 		switch (run->mmio.len) {
337 		case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
338 		case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
339 		case 1: gpr = *(u8 *)run->mmio.data; break;
340 		}
341 	}
342 
343 	if (vcpu->arch.mmio_sign_extend) {
344 		switch (run->mmio.len) {
345 #ifdef CONFIG_PPC64
346 		case 4:
347 			gpr = (s64)(s32)gpr;
348 			break;
349 #endif
350 		case 2:
351 			gpr = (s64)(s16)gpr;
352 			break;
353 		case 1:
354 			gpr = (s64)(s8)gpr;
355 			break;
356 		}
357 	}
358 
359 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
360 
361 	switch (vcpu->arch.io_gpr & KVM_REG_EXT_MASK) {
362 	case KVM_REG_GPR:
363 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
364 		break;
365 	case KVM_REG_FPR:
366 		vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
367 		break;
368 #ifdef CONFIG_PPC_BOOK3S
369 	case KVM_REG_QPR:
370 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
371 		break;
372 	case KVM_REG_FQPR:
373 		vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
374 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
375 		break;
376 #endif
377 	default:
378 		BUG();
379 	}
380 }
381 
382 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
383                        unsigned int rt, unsigned int bytes, int is_bigendian)
384 {
385 	if (bytes > sizeof(run->mmio.data)) {
386 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
387 		       run->mmio.len);
388 	}
389 
390 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
391 	run->mmio.len = bytes;
392 	run->mmio.is_write = 0;
393 
394 	vcpu->arch.io_gpr = rt;
395 	vcpu->arch.mmio_is_bigendian = is_bigendian;
396 	vcpu->mmio_needed = 1;
397 	vcpu->mmio_is_write = 0;
398 	vcpu->arch.mmio_sign_extend = 0;
399 
400 	return EMULATE_DO_MMIO;
401 }
402 
403 /* Same as above, but sign extends */
404 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
405                         unsigned int rt, unsigned int bytes, int is_bigendian)
406 {
407 	int r;
408 
409 	r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
410 	vcpu->arch.mmio_sign_extend = 1;
411 
412 	return r;
413 }
414 
415 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
416                         u64 val, unsigned int bytes, int is_bigendian)
417 {
418 	void *data = run->mmio.data;
419 
420 	if (bytes > sizeof(run->mmio.data)) {
421 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
422 		       run->mmio.len);
423 	}
424 
425 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
426 	run->mmio.len = bytes;
427 	run->mmio.is_write = 1;
428 	vcpu->mmio_needed = 1;
429 	vcpu->mmio_is_write = 1;
430 
431 	/* Store the value at the lowest bytes in 'data'. */
432 	if (is_bigendian) {
433 		switch (bytes) {
434 		case 8: *(u64 *)data = val; break;
435 		case 4: *(u32 *)data = val; break;
436 		case 2: *(u16 *)data = val; break;
437 		case 1: *(u8  *)data = val; break;
438 		}
439 	} else {
440 		/* Store LE value into 'data'. */
441 		switch (bytes) {
442 		case 4: st_le32(data, val); break;
443 		case 2: st_le16(data, val); break;
444 		case 1: *(u8 *)data = val; break;
445 		}
446 	}
447 
448 	return EMULATE_DO_MMIO;
449 }
450 
451 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
452 {
453 	int r;
454 	sigset_t sigsaved;
455 
456 	if (vcpu->sigset_active)
457 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
458 
459 	if (vcpu->mmio_needed) {
460 		if (!vcpu->mmio_is_write)
461 			kvmppc_complete_mmio_load(vcpu, run);
462 		vcpu->mmio_needed = 0;
463 	} else if (vcpu->arch.dcr_needed) {
464 		if (!vcpu->arch.dcr_is_write)
465 			kvmppc_complete_dcr_load(vcpu, run);
466 		vcpu->arch.dcr_needed = 0;
467 	} else if (vcpu->arch.osi_needed) {
468 		u64 *gprs = run->osi.gprs;
469 		int i;
470 
471 		for (i = 0; i < 32; i++)
472 			kvmppc_set_gpr(vcpu, i, gprs[i]);
473 		vcpu->arch.osi_needed = 0;
474 	}
475 
476 	kvmppc_core_deliver_interrupts(vcpu);
477 
478 	local_irq_disable();
479 	kvm_guest_enter();
480 	r = __kvmppc_vcpu_run(run, vcpu);
481 	kvm_guest_exit();
482 	local_irq_enable();
483 
484 	if (vcpu->sigset_active)
485 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
486 
487 	return r;
488 }
489 
490 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
491 {
492 	if (irq->irq == KVM_INTERRUPT_UNSET)
493 		kvmppc_core_dequeue_external(vcpu, irq);
494 	else
495 		kvmppc_core_queue_external(vcpu, irq);
496 
497 	if (waitqueue_active(&vcpu->wq)) {
498 		wake_up_interruptible(&vcpu->wq);
499 		vcpu->stat.halt_wakeup++;
500 	}
501 
502 	return 0;
503 }
504 
505 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
506 				     struct kvm_enable_cap *cap)
507 {
508 	int r;
509 
510 	if (cap->flags)
511 		return -EINVAL;
512 
513 	switch (cap->cap) {
514 	case KVM_CAP_PPC_OSI:
515 		r = 0;
516 		vcpu->arch.osi_enabled = true;
517 		break;
518 	default:
519 		r = -EINVAL;
520 		break;
521 	}
522 
523 	return r;
524 }
525 
526 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
527                                     struct kvm_mp_state *mp_state)
528 {
529 	return -EINVAL;
530 }
531 
532 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
533                                     struct kvm_mp_state *mp_state)
534 {
535 	return -EINVAL;
536 }
537 
538 long kvm_arch_vcpu_ioctl(struct file *filp,
539                          unsigned int ioctl, unsigned long arg)
540 {
541 	struct kvm_vcpu *vcpu = filp->private_data;
542 	void __user *argp = (void __user *)arg;
543 	long r;
544 
545 	switch (ioctl) {
546 	case KVM_INTERRUPT: {
547 		struct kvm_interrupt irq;
548 		r = -EFAULT;
549 		if (copy_from_user(&irq, argp, sizeof(irq)))
550 			goto out;
551 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
552 		goto out;
553 	}
554 
555 	case KVM_ENABLE_CAP:
556 	{
557 		struct kvm_enable_cap cap;
558 		r = -EFAULT;
559 		if (copy_from_user(&cap, argp, sizeof(cap)))
560 			goto out;
561 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
562 		break;
563 	}
564 	default:
565 		r = -EINVAL;
566 	}
567 
568 out:
569 	return r;
570 }
571 
572 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
573 {
574 	u32 inst_lis = 0x3c000000;
575 	u32 inst_ori = 0x60000000;
576 	u32 inst_nop = 0x60000000;
577 	u32 inst_sc = 0x44000002;
578 	u32 inst_imm_mask = 0xffff;
579 
580 	/*
581 	 * The hypercall to get into KVM from within guest context is as
582 	 * follows:
583 	 *
584 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
585 	 *    ori r0, KVM_SC_MAGIC_R0@l
586 	 *    sc
587 	 *    nop
588 	 */
589 	pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
590 	pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
591 	pvinfo->hcall[2] = inst_sc;
592 	pvinfo->hcall[3] = inst_nop;
593 
594 	return 0;
595 }
596 
597 long kvm_arch_vm_ioctl(struct file *filp,
598                        unsigned int ioctl, unsigned long arg)
599 {
600 	void __user *argp = (void __user *)arg;
601 	long r;
602 
603 	switch (ioctl) {
604 	case KVM_PPC_GET_PVINFO: {
605 		struct kvm_ppc_pvinfo pvinfo;
606 		memset(&pvinfo, 0, sizeof(pvinfo));
607 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
608 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
609 			r = -EFAULT;
610 			goto out;
611 		}
612 
613 		break;
614 	}
615 	default:
616 		r = -ENOTTY;
617 	}
618 
619 out:
620 	return r;
621 }
622 
623 int kvm_arch_init(void *opaque)
624 {
625 	return 0;
626 }
627 
628 void kvm_arch_exit(void)
629 {
630 }
631