xref: /openbmc/linux/arch/powerpc/kvm/booke.c (revision 81d67439)
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  * Copyright 2010-2011 Freescale Semiconductor, Inc.
17  *
18  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
19  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
20  */
21 
22 #include <linux/errno.h>
23 #include <linux/err.h>
24 #include <linux/kvm_host.h>
25 #include <linux/gfp.h>
26 #include <linux/module.h>
27 #include <linux/vmalloc.h>
28 #include <linux/fs.h>
29 
30 #include <asm/cputable.h>
31 #include <asm/uaccess.h>
32 #include <asm/kvm_ppc.h>
33 #include "timing.h"
34 #include <asm/cacheflush.h>
35 
36 #include "booke.h"
37 
38 unsigned long kvmppc_booke_handlers;
39 
40 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
41 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
42 
43 struct kvm_stats_debugfs_item debugfs_entries[] = {
44 	{ "mmio",       VCPU_STAT(mmio_exits) },
45 	{ "dcr",        VCPU_STAT(dcr_exits) },
46 	{ "sig",        VCPU_STAT(signal_exits) },
47 	{ "itlb_r",     VCPU_STAT(itlb_real_miss_exits) },
48 	{ "itlb_v",     VCPU_STAT(itlb_virt_miss_exits) },
49 	{ "dtlb_r",     VCPU_STAT(dtlb_real_miss_exits) },
50 	{ "dtlb_v",     VCPU_STAT(dtlb_virt_miss_exits) },
51 	{ "sysc",       VCPU_STAT(syscall_exits) },
52 	{ "isi",        VCPU_STAT(isi_exits) },
53 	{ "dsi",        VCPU_STAT(dsi_exits) },
54 	{ "inst_emu",   VCPU_STAT(emulated_inst_exits) },
55 	{ "dec",        VCPU_STAT(dec_exits) },
56 	{ "ext_intr",   VCPU_STAT(ext_intr_exits) },
57 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
58 	{ NULL }
59 };
60 
61 /* TODO: use vcpu_printf() */
62 void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu)
63 {
64 	int i;
65 
66 	printk("pc:   %08lx msr:  %08llx\n", vcpu->arch.pc, vcpu->arch.shared->msr);
67 	printk("lr:   %08lx ctr:  %08lx\n", vcpu->arch.lr, vcpu->arch.ctr);
68 	printk("srr0: %08llx srr1: %08llx\n", vcpu->arch.shared->srr0,
69 					    vcpu->arch.shared->srr1);
70 
71 	printk("exceptions: %08lx\n", vcpu->arch.pending_exceptions);
72 
73 	for (i = 0; i < 32; i += 4) {
74 		printk("gpr%02d: %08lx %08lx %08lx %08lx\n", i,
75 		       kvmppc_get_gpr(vcpu, i),
76 		       kvmppc_get_gpr(vcpu, i+1),
77 		       kvmppc_get_gpr(vcpu, i+2),
78 		       kvmppc_get_gpr(vcpu, i+3));
79 	}
80 }
81 
82 #ifdef CONFIG_SPE
83 void kvmppc_vcpu_disable_spe(struct kvm_vcpu *vcpu)
84 {
85 	preempt_disable();
86 	enable_kernel_spe();
87 	kvmppc_save_guest_spe(vcpu);
88 	vcpu->arch.shadow_msr &= ~MSR_SPE;
89 	preempt_enable();
90 }
91 
92 static void kvmppc_vcpu_enable_spe(struct kvm_vcpu *vcpu)
93 {
94 	preempt_disable();
95 	enable_kernel_spe();
96 	kvmppc_load_guest_spe(vcpu);
97 	vcpu->arch.shadow_msr |= MSR_SPE;
98 	preempt_enable();
99 }
100 
101 static void kvmppc_vcpu_sync_spe(struct kvm_vcpu *vcpu)
102 {
103 	if (vcpu->arch.shared->msr & MSR_SPE) {
104 		if (!(vcpu->arch.shadow_msr & MSR_SPE))
105 			kvmppc_vcpu_enable_spe(vcpu);
106 	} else if (vcpu->arch.shadow_msr & MSR_SPE) {
107 		kvmppc_vcpu_disable_spe(vcpu);
108 	}
109 }
110 #else
111 static void kvmppc_vcpu_sync_spe(struct kvm_vcpu *vcpu)
112 {
113 }
114 #endif
115 
116 /*
117  * Helper function for "full" MSR writes.  No need to call this if only
118  * EE/CE/ME/DE/RI are changing.
119  */
120 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u32 new_msr)
121 {
122 	u32 old_msr = vcpu->arch.shared->msr;
123 
124 	vcpu->arch.shared->msr = new_msr;
125 
126 	kvmppc_mmu_msr_notify(vcpu, old_msr);
127 
128 	if (vcpu->arch.shared->msr & MSR_WE) {
129 		kvm_vcpu_block(vcpu);
130 		kvmppc_set_exit_type(vcpu, EMULATED_MTMSRWE_EXITS);
131 	};
132 
133 	kvmppc_vcpu_sync_spe(vcpu);
134 }
135 
136 static void kvmppc_booke_queue_irqprio(struct kvm_vcpu *vcpu,
137                                        unsigned int priority)
138 {
139 	set_bit(priority, &vcpu->arch.pending_exceptions);
140 }
141 
142 static void kvmppc_core_queue_dtlb_miss(struct kvm_vcpu *vcpu,
143                                         ulong dear_flags, ulong esr_flags)
144 {
145 	vcpu->arch.queued_dear = dear_flags;
146 	vcpu->arch.queued_esr = esr_flags;
147 	kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DTLB_MISS);
148 }
149 
150 static void kvmppc_core_queue_data_storage(struct kvm_vcpu *vcpu,
151                                            ulong dear_flags, ulong esr_flags)
152 {
153 	vcpu->arch.queued_dear = dear_flags;
154 	vcpu->arch.queued_esr = esr_flags;
155 	kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DATA_STORAGE);
156 }
157 
158 static void kvmppc_core_queue_inst_storage(struct kvm_vcpu *vcpu,
159                                            ulong esr_flags)
160 {
161 	vcpu->arch.queued_esr = esr_flags;
162 	kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_INST_STORAGE);
163 }
164 
165 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong esr_flags)
166 {
167 	vcpu->arch.queued_esr = esr_flags;
168 	kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
169 }
170 
171 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
172 {
173 	kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DECREMENTER);
174 }
175 
176 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
177 {
178 	return test_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
179 }
180 
181 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
182 {
183 	clear_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
184 }
185 
186 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
187                                 struct kvm_interrupt *irq)
188 {
189 	unsigned int prio = BOOKE_IRQPRIO_EXTERNAL;
190 
191 	if (irq->irq == KVM_INTERRUPT_SET_LEVEL)
192 		prio = BOOKE_IRQPRIO_EXTERNAL_LEVEL;
193 
194 	kvmppc_booke_queue_irqprio(vcpu, prio);
195 }
196 
197 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
198                                   struct kvm_interrupt *irq)
199 {
200 	clear_bit(BOOKE_IRQPRIO_EXTERNAL, &vcpu->arch.pending_exceptions);
201 	clear_bit(BOOKE_IRQPRIO_EXTERNAL_LEVEL, &vcpu->arch.pending_exceptions);
202 }
203 
204 /* Deliver the interrupt of the corresponding priority, if possible. */
205 static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
206                                         unsigned int priority)
207 {
208 	int allowed = 0;
209 	ulong uninitialized_var(msr_mask);
210 	bool update_esr = false, update_dear = false;
211 	ulong crit_raw = vcpu->arch.shared->critical;
212 	ulong crit_r1 = kvmppc_get_gpr(vcpu, 1);
213 	bool crit;
214 	bool keep_irq = false;
215 
216 	/* Truncate crit indicators in 32 bit mode */
217 	if (!(vcpu->arch.shared->msr & MSR_SF)) {
218 		crit_raw &= 0xffffffff;
219 		crit_r1 &= 0xffffffff;
220 	}
221 
222 	/* Critical section when crit == r1 */
223 	crit = (crit_raw == crit_r1);
224 	/* ... and we're in supervisor mode */
225 	crit = crit && !(vcpu->arch.shared->msr & MSR_PR);
226 
227 	if (priority == BOOKE_IRQPRIO_EXTERNAL_LEVEL) {
228 		priority = BOOKE_IRQPRIO_EXTERNAL;
229 		keep_irq = true;
230 	}
231 
232 	switch (priority) {
233 	case BOOKE_IRQPRIO_DTLB_MISS:
234 	case BOOKE_IRQPRIO_DATA_STORAGE:
235 		update_dear = true;
236 		/* fall through */
237 	case BOOKE_IRQPRIO_INST_STORAGE:
238 	case BOOKE_IRQPRIO_PROGRAM:
239 		update_esr = true;
240 		/* fall through */
241 	case BOOKE_IRQPRIO_ITLB_MISS:
242 	case BOOKE_IRQPRIO_SYSCALL:
243 	case BOOKE_IRQPRIO_FP_UNAVAIL:
244 	case BOOKE_IRQPRIO_SPE_UNAVAIL:
245 	case BOOKE_IRQPRIO_SPE_FP_DATA:
246 	case BOOKE_IRQPRIO_SPE_FP_ROUND:
247 	case BOOKE_IRQPRIO_AP_UNAVAIL:
248 	case BOOKE_IRQPRIO_ALIGNMENT:
249 		allowed = 1;
250 		msr_mask = MSR_CE|MSR_ME|MSR_DE;
251 		break;
252 	case BOOKE_IRQPRIO_CRITICAL:
253 	case BOOKE_IRQPRIO_WATCHDOG:
254 		allowed = vcpu->arch.shared->msr & MSR_CE;
255 		msr_mask = MSR_ME;
256 		break;
257 	case BOOKE_IRQPRIO_MACHINE_CHECK:
258 		allowed = vcpu->arch.shared->msr & MSR_ME;
259 		msr_mask = 0;
260 		break;
261 	case BOOKE_IRQPRIO_EXTERNAL:
262 	case BOOKE_IRQPRIO_DECREMENTER:
263 	case BOOKE_IRQPRIO_FIT:
264 		allowed = vcpu->arch.shared->msr & MSR_EE;
265 		allowed = allowed && !crit;
266 		msr_mask = MSR_CE|MSR_ME|MSR_DE;
267 		break;
268 	case BOOKE_IRQPRIO_DEBUG:
269 		allowed = vcpu->arch.shared->msr & MSR_DE;
270 		msr_mask = MSR_ME;
271 		break;
272 	}
273 
274 	if (allowed) {
275 		vcpu->arch.shared->srr0 = vcpu->arch.pc;
276 		vcpu->arch.shared->srr1 = vcpu->arch.shared->msr;
277 		vcpu->arch.pc = vcpu->arch.ivpr | vcpu->arch.ivor[priority];
278 		if (update_esr == true)
279 			vcpu->arch.esr = vcpu->arch.queued_esr;
280 		if (update_dear == true)
281 			vcpu->arch.shared->dar = vcpu->arch.queued_dear;
282 		kvmppc_set_msr(vcpu, vcpu->arch.shared->msr & msr_mask);
283 
284 		if (!keep_irq)
285 			clear_bit(priority, &vcpu->arch.pending_exceptions);
286 	}
287 
288 	return allowed;
289 }
290 
291 /* Check pending exceptions and deliver one, if possible. */
292 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
293 {
294 	unsigned long *pending = &vcpu->arch.pending_exceptions;
295 	unsigned long old_pending = vcpu->arch.pending_exceptions;
296 	unsigned int priority;
297 
298 	priority = __ffs(*pending);
299 	while (priority <= BOOKE_IRQPRIO_MAX) {
300 		if (kvmppc_booke_irqprio_deliver(vcpu, priority))
301 			break;
302 
303 		priority = find_next_bit(pending,
304 		                         BITS_PER_BYTE * sizeof(*pending),
305 		                         priority + 1);
306 	}
307 
308 	/* Tell the guest about our interrupt status */
309 	if (*pending)
310 		vcpu->arch.shared->int_pending = 1;
311 	else if (old_pending)
312 		vcpu->arch.shared->int_pending = 0;
313 }
314 
315 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
316 {
317 	int ret;
318 
319 	local_irq_disable();
320 	kvm_guest_enter();
321 	ret = __kvmppc_vcpu_run(kvm_run, vcpu);
322 	kvm_guest_exit();
323 	local_irq_enable();
324 
325 	return ret;
326 }
327 
328 /**
329  * kvmppc_handle_exit
330  *
331  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
332  */
333 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
334                        unsigned int exit_nr)
335 {
336 	enum emulation_result er;
337 	int r = RESUME_HOST;
338 
339 	/* update before a new last_exit_type is rewritten */
340 	kvmppc_update_timing_stats(vcpu);
341 
342 	local_irq_enable();
343 
344 	run->exit_reason = KVM_EXIT_UNKNOWN;
345 	run->ready_for_interrupt_injection = 1;
346 
347 	switch (exit_nr) {
348 	case BOOKE_INTERRUPT_MACHINE_CHECK:
349 		printk("MACHINE CHECK: %lx\n", mfspr(SPRN_MCSR));
350 		kvmppc_dump_vcpu(vcpu);
351 		r = RESUME_HOST;
352 		break;
353 
354 	case BOOKE_INTERRUPT_EXTERNAL:
355 		kvmppc_account_exit(vcpu, EXT_INTR_EXITS);
356 		if (need_resched())
357 			cond_resched();
358 		r = RESUME_GUEST;
359 		break;
360 
361 	case BOOKE_INTERRUPT_DECREMENTER:
362 		/* Since we switched IVPR back to the host's value, the host
363 		 * handled this interrupt the moment we enabled interrupts.
364 		 * Now we just offer it a chance to reschedule the guest. */
365 		kvmppc_account_exit(vcpu, DEC_EXITS);
366 		if (need_resched())
367 			cond_resched();
368 		r = RESUME_GUEST;
369 		break;
370 
371 	case BOOKE_INTERRUPT_PROGRAM:
372 		if (vcpu->arch.shared->msr & MSR_PR) {
373 			/* Program traps generated by user-level software must be handled
374 			 * by the guest kernel. */
375 			kvmppc_core_queue_program(vcpu, vcpu->arch.fault_esr);
376 			r = RESUME_GUEST;
377 			kvmppc_account_exit(vcpu, USR_PR_INST);
378 			break;
379 		}
380 
381 		er = kvmppc_emulate_instruction(run, vcpu);
382 		switch (er) {
383 		case EMULATE_DONE:
384 			/* don't overwrite subtypes, just account kvm_stats */
385 			kvmppc_account_exit_stat(vcpu, EMULATED_INST_EXITS);
386 			/* Future optimization: only reload non-volatiles if
387 			 * they were actually modified by emulation. */
388 			r = RESUME_GUEST_NV;
389 			break;
390 		case EMULATE_DO_DCR:
391 			run->exit_reason = KVM_EXIT_DCR;
392 			r = RESUME_HOST;
393 			break;
394 		case EMULATE_FAIL:
395 			/* XXX Deliver Program interrupt to guest. */
396 			printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
397 			       __func__, vcpu->arch.pc, vcpu->arch.last_inst);
398 			/* For debugging, encode the failing instruction and
399 			 * report it to userspace. */
400 			run->hw.hardware_exit_reason = ~0ULL << 32;
401 			run->hw.hardware_exit_reason |= vcpu->arch.last_inst;
402 			r = RESUME_HOST;
403 			break;
404 		default:
405 			BUG();
406 		}
407 		break;
408 
409 	case BOOKE_INTERRUPT_FP_UNAVAIL:
410 		kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_FP_UNAVAIL);
411 		kvmppc_account_exit(vcpu, FP_UNAVAIL);
412 		r = RESUME_GUEST;
413 		break;
414 
415 #ifdef CONFIG_SPE
416 	case BOOKE_INTERRUPT_SPE_UNAVAIL: {
417 		if (vcpu->arch.shared->msr & MSR_SPE)
418 			kvmppc_vcpu_enable_spe(vcpu);
419 		else
420 			kvmppc_booke_queue_irqprio(vcpu,
421 						   BOOKE_IRQPRIO_SPE_UNAVAIL);
422 		r = RESUME_GUEST;
423 		break;
424 	}
425 
426 	case BOOKE_INTERRUPT_SPE_FP_DATA:
427 		kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_DATA);
428 		r = RESUME_GUEST;
429 		break;
430 
431 	case BOOKE_INTERRUPT_SPE_FP_ROUND:
432 		kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_ROUND);
433 		r = RESUME_GUEST;
434 		break;
435 #else
436 	case BOOKE_INTERRUPT_SPE_UNAVAIL:
437 		/*
438 		 * Guest wants SPE, but host kernel doesn't support it.  Send
439 		 * an "unimplemented operation" program check to the guest.
440 		 */
441 		kvmppc_core_queue_program(vcpu, ESR_PUO | ESR_SPV);
442 		r = RESUME_GUEST;
443 		break;
444 
445 	/*
446 	 * These really should never happen without CONFIG_SPE,
447 	 * as we should never enable the real MSR[SPE] in the guest.
448 	 */
449 	case BOOKE_INTERRUPT_SPE_FP_DATA:
450 	case BOOKE_INTERRUPT_SPE_FP_ROUND:
451 		printk(KERN_CRIT "%s: unexpected SPE interrupt %u at %08lx\n",
452 		       __func__, exit_nr, vcpu->arch.pc);
453 		run->hw.hardware_exit_reason = exit_nr;
454 		r = RESUME_HOST;
455 		break;
456 #endif
457 
458 	case BOOKE_INTERRUPT_DATA_STORAGE:
459 		kvmppc_core_queue_data_storage(vcpu, vcpu->arch.fault_dear,
460 		                               vcpu->arch.fault_esr);
461 		kvmppc_account_exit(vcpu, DSI_EXITS);
462 		r = RESUME_GUEST;
463 		break;
464 
465 	case BOOKE_INTERRUPT_INST_STORAGE:
466 		kvmppc_core_queue_inst_storage(vcpu, vcpu->arch.fault_esr);
467 		kvmppc_account_exit(vcpu, ISI_EXITS);
468 		r = RESUME_GUEST;
469 		break;
470 
471 	case BOOKE_INTERRUPT_SYSCALL:
472 		if (!(vcpu->arch.shared->msr & MSR_PR) &&
473 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
474 			/* KVM PV hypercalls */
475 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
476 			r = RESUME_GUEST;
477 		} else {
478 			/* Guest syscalls */
479 			kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SYSCALL);
480 		}
481 		kvmppc_account_exit(vcpu, SYSCALL_EXITS);
482 		r = RESUME_GUEST;
483 		break;
484 
485 	case BOOKE_INTERRUPT_DTLB_MISS: {
486 		unsigned long eaddr = vcpu->arch.fault_dear;
487 		int gtlb_index;
488 		gpa_t gpaddr;
489 		gfn_t gfn;
490 
491 #ifdef CONFIG_KVM_E500
492 		if (!(vcpu->arch.shared->msr & MSR_PR) &&
493 		    (eaddr & PAGE_MASK) == vcpu->arch.magic_page_ea) {
494 			kvmppc_map_magic(vcpu);
495 			kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
496 			r = RESUME_GUEST;
497 
498 			break;
499 		}
500 #endif
501 
502 		/* Check the guest TLB. */
503 		gtlb_index = kvmppc_mmu_dtlb_index(vcpu, eaddr);
504 		if (gtlb_index < 0) {
505 			/* The guest didn't have a mapping for it. */
506 			kvmppc_core_queue_dtlb_miss(vcpu,
507 			                            vcpu->arch.fault_dear,
508 			                            vcpu->arch.fault_esr);
509 			kvmppc_mmu_dtlb_miss(vcpu);
510 			kvmppc_account_exit(vcpu, DTLB_REAL_MISS_EXITS);
511 			r = RESUME_GUEST;
512 			break;
513 		}
514 
515 		gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
516 		gfn = gpaddr >> PAGE_SHIFT;
517 
518 		if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
519 			/* The guest TLB had a mapping, but the shadow TLB
520 			 * didn't, and it is RAM. This could be because:
521 			 * a) the entry is mapping the host kernel, or
522 			 * b) the guest used a large mapping which we're faking
523 			 * Either way, we need to satisfy the fault without
524 			 * invoking the guest. */
525 			kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
526 			kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
527 			r = RESUME_GUEST;
528 		} else {
529 			/* Guest has mapped and accessed a page which is not
530 			 * actually RAM. */
531 			vcpu->arch.paddr_accessed = gpaddr;
532 			r = kvmppc_emulate_mmio(run, vcpu);
533 			kvmppc_account_exit(vcpu, MMIO_EXITS);
534 		}
535 
536 		break;
537 	}
538 
539 	case BOOKE_INTERRUPT_ITLB_MISS: {
540 		unsigned long eaddr = vcpu->arch.pc;
541 		gpa_t gpaddr;
542 		gfn_t gfn;
543 		int gtlb_index;
544 
545 		r = RESUME_GUEST;
546 
547 		/* Check the guest TLB. */
548 		gtlb_index = kvmppc_mmu_itlb_index(vcpu, eaddr);
549 		if (gtlb_index < 0) {
550 			/* The guest didn't have a mapping for it. */
551 			kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ITLB_MISS);
552 			kvmppc_mmu_itlb_miss(vcpu);
553 			kvmppc_account_exit(vcpu, ITLB_REAL_MISS_EXITS);
554 			break;
555 		}
556 
557 		kvmppc_account_exit(vcpu, ITLB_VIRT_MISS_EXITS);
558 
559 		gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
560 		gfn = gpaddr >> PAGE_SHIFT;
561 
562 		if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
563 			/* The guest TLB had a mapping, but the shadow TLB
564 			 * didn't. This could be because:
565 			 * a) the entry is mapping the host kernel, or
566 			 * b) the guest used a large mapping which we're faking
567 			 * Either way, we need to satisfy the fault without
568 			 * invoking the guest. */
569 			kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
570 		} else {
571 			/* Guest mapped and leaped at non-RAM! */
572 			kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_MACHINE_CHECK);
573 		}
574 
575 		break;
576 	}
577 
578 	case BOOKE_INTERRUPT_DEBUG: {
579 		u32 dbsr;
580 
581 		vcpu->arch.pc = mfspr(SPRN_CSRR0);
582 
583 		/* clear IAC events in DBSR register */
584 		dbsr = mfspr(SPRN_DBSR);
585 		dbsr &= DBSR_IAC1 | DBSR_IAC2 | DBSR_IAC3 | DBSR_IAC4;
586 		mtspr(SPRN_DBSR, dbsr);
587 
588 		run->exit_reason = KVM_EXIT_DEBUG;
589 		kvmppc_account_exit(vcpu, DEBUG_EXITS);
590 		r = RESUME_HOST;
591 		break;
592 	}
593 
594 	default:
595 		printk(KERN_EMERG "exit_nr %d\n", exit_nr);
596 		BUG();
597 	}
598 
599 	local_irq_disable();
600 
601 	kvmppc_core_deliver_interrupts(vcpu);
602 
603 	if (!(r & RESUME_HOST)) {
604 		/* To avoid clobbering exit_reason, only check for signals if
605 		 * we aren't already exiting to userspace for some other
606 		 * reason. */
607 		if (signal_pending(current)) {
608 			run->exit_reason = KVM_EXIT_INTR;
609 			r = (-EINTR << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
610 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
611 		}
612 	}
613 
614 	return r;
615 }
616 
617 /* Initial guest state: 16MB mapping 0 -> 0, PC = 0, MSR = 0, R1 = 16MB */
618 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
619 {
620 	int i;
621 
622 	vcpu->arch.pc = 0;
623 	vcpu->arch.shared->msr = 0;
624 	vcpu->arch.shadow_msr = MSR_USER | MSR_DE | MSR_IS | MSR_DS;
625 	kvmppc_set_gpr(vcpu, 1, (16<<20) - 8); /* -8 for the callee-save LR slot */
626 
627 	vcpu->arch.shadow_pid = 1;
628 
629 	/* Eye-catching numbers so we know if the guest takes an interrupt
630 	 * before it's programmed its own IVPR/IVORs. */
631 	vcpu->arch.ivpr = 0x55550000;
632 	for (i = 0; i < BOOKE_IRQPRIO_MAX; i++)
633 		vcpu->arch.ivor[i] = 0x7700 | i * 4;
634 
635 	kvmppc_init_timing_stats(vcpu);
636 
637 	return kvmppc_core_vcpu_setup(vcpu);
638 }
639 
640 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
641 {
642 	int i;
643 
644 	regs->pc = vcpu->arch.pc;
645 	regs->cr = kvmppc_get_cr(vcpu);
646 	regs->ctr = vcpu->arch.ctr;
647 	regs->lr = vcpu->arch.lr;
648 	regs->xer = kvmppc_get_xer(vcpu);
649 	regs->msr = vcpu->arch.shared->msr;
650 	regs->srr0 = vcpu->arch.shared->srr0;
651 	regs->srr1 = vcpu->arch.shared->srr1;
652 	regs->pid = vcpu->arch.pid;
653 	regs->sprg0 = vcpu->arch.shared->sprg0;
654 	regs->sprg1 = vcpu->arch.shared->sprg1;
655 	regs->sprg2 = vcpu->arch.shared->sprg2;
656 	regs->sprg3 = vcpu->arch.shared->sprg3;
657 	regs->sprg4 = vcpu->arch.sprg4;
658 	regs->sprg5 = vcpu->arch.sprg5;
659 	regs->sprg6 = vcpu->arch.sprg6;
660 	regs->sprg7 = vcpu->arch.sprg7;
661 
662 	for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
663 		regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
664 
665 	return 0;
666 }
667 
668 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
669 {
670 	int i;
671 
672 	vcpu->arch.pc = regs->pc;
673 	kvmppc_set_cr(vcpu, regs->cr);
674 	vcpu->arch.ctr = regs->ctr;
675 	vcpu->arch.lr = regs->lr;
676 	kvmppc_set_xer(vcpu, regs->xer);
677 	kvmppc_set_msr(vcpu, regs->msr);
678 	vcpu->arch.shared->srr0 = regs->srr0;
679 	vcpu->arch.shared->srr1 = regs->srr1;
680 	kvmppc_set_pid(vcpu, regs->pid);
681 	vcpu->arch.shared->sprg0 = regs->sprg0;
682 	vcpu->arch.shared->sprg1 = regs->sprg1;
683 	vcpu->arch.shared->sprg2 = regs->sprg2;
684 	vcpu->arch.shared->sprg3 = regs->sprg3;
685 	vcpu->arch.sprg4 = regs->sprg4;
686 	vcpu->arch.sprg5 = regs->sprg5;
687 	vcpu->arch.sprg6 = regs->sprg6;
688 	vcpu->arch.sprg7 = regs->sprg7;
689 
690 	for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
691 		kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
692 
693 	return 0;
694 }
695 
696 static void get_sregs_base(struct kvm_vcpu *vcpu,
697                            struct kvm_sregs *sregs)
698 {
699 	u64 tb = get_tb();
700 
701 	sregs->u.e.features |= KVM_SREGS_E_BASE;
702 
703 	sregs->u.e.csrr0 = vcpu->arch.csrr0;
704 	sregs->u.e.csrr1 = vcpu->arch.csrr1;
705 	sregs->u.e.mcsr = vcpu->arch.mcsr;
706 	sregs->u.e.esr = vcpu->arch.esr;
707 	sregs->u.e.dear = vcpu->arch.shared->dar;
708 	sregs->u.e.tsr = vcpu->arch.tsr;
709 	sregs->u.e.tcr = vcpu->arch.tcr;
710 	sregs->u.e.dec = kvmppc_get_dec(vcpu, tb);
711 	sregs->u.e.tb = tb;
712 	sregs->u.e.vrsave = vcpu->arch.vrsave;
713 }
714 
715 static int set_sregs_base(struct kvm_vcpu *vcpu,
716                           struct kvm_sregs *sregs)
717 {
718 	if (!(sregs->u.e.features & KVM_SREGS_E_BASE))
719 		return 0;
720 
721 	vcpu->arch.csrr0 = sregs->u.e.csrr0;
722 	vcpu->arch.csrr1 = sregs->u.e.csrr1;
723 	vcpu->arch.mcsr = sregs->u.e.mcsr;
724 	vcpu->arch.esr = sregs->u.e.esr;
725 	vcpu->arch.shared->dar = sregs->u.e.dear;
726 	vcpu->arch.vrsave = sregs->u.e.vrsave;
727 	vcpu->arch.tcr = sregs->u.e.tcr;
728 
729 	if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_DEC)
730 		vcpu->arch.dec = sregs->u.e.dec;
731 
732 	kvmppc_emulate_dec(vcpu);
733 
734 	if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_TSR) {
735 		/*
736 		 * FIXME: existing KVM timer handling is incomplete.
737 		 * TSR cannot be read by the guest, and its value in
738 		 * vcpu->arch is always zero.  For now, just handle
739 		 * the case where the caller is trying to inject a
740 		 * decrementer interrupt.
741 		 */
742 
743 		if ((sregs->u.e.tsr & TSR_DIS) &&
744 		    (vcpu->arch.tcr & TCR_DIE))
745 			kvmppc_core_queue_dec(vcpu);
746 	}
747 
748 	return 0;
749 }
750 
751 static void get_sregs_arch206(struct kvm_vcpu *vcpu,
752                               struct kvm_sregs *sregs)
753 {
754 	sregs->u.e.features |= KVM_SREGS_E_ARCH206;
755 
756 	sregs->u.e.pir = 0;
757 	sregs->u.e.mcsrr0 = vcpu->arch.mcsrr0;
758 	sregs->u.e.mcsrr1 = vcpu->arch.mcsrr1;
759 	sregs->u.e.decar = vcpu->arch.decar;
760 	sregs->u.e.ivpr = vcpu->arch.ivpr;
761 }
762 
763 static int set_sregs_arch206(struct kvm_vcpu *vcpu,
764                              struct kvm_sregs *sregs)
765 {
766 	if (!(sregs->u.e.features & KVM_SREGS_E_ARCH206))
767 		return 0;
768 
769 	if (sregs->u.e.pir != 0)
770 		return -EINVAL;
771 
772 	vcpu->arch.mcsrr0 = sregs->u.e.mcsrr0;
773 	vcpu->arch.mcsrr1 = sregs->u.e.mcsrr1;
774 	vcpu->arch.decar = sregs->u.e.decar;
775 	vcpu->arch.ivpr = sregs->u.e.ivpr;
776 
777 	return 0;
778 }
779 
780 void kvmppc_get_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
781 {
782 	sregs->u.e.features |= KVM_SREGS_E_IVOR;
783 
784 	sregs->u.e.ivor_low[0] = vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL];
785 	sregs->u.e.ivor_low[1] = vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK];
786 	sregs->u.e.ivor_low[2] = vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE];
787 	sregs->u.e.ivor_low[3] = vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE];
788 	sregs->u.e.ivor_low[4] = vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL];
789 	sregs->u.e.ivor_low[5] = vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT];
790 	sregs->u.e.ivor_low[6] = vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM];
791 	sregs->u.e.ivor_low[7] = vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL];
792 	sregs->u.e.ivor_low[8] = vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL];
793 	sregs->u.e.ivor_low[9] = vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL];
794 	sregs->u.e.ivor_low[10] = vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER];
795 	sregs->u.e.ivor_low[11] = vcpu->arch.ivor[BOOKE_IRQPRIO_FIT];
796 	sregs->u.e.ivor_low[12] = vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG];
797 	sregs->u.e.ivor_low[13] = vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS];
798 	sregs->u.e.ivor_low[14] = vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS];
799 	sregs->u.e.ivor_low[15] = vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG];
800 }
801 
802 int kvmppc_set_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
803 {
804 	if (!(sregs->u.e.features & KVM_SREGS_E_IVOR))
805 		return 0;
806 
807 	vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL] = sregs->u.e.ivor_low[0];
808 	vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK] = sregs->u.e.ivor_low[1];
809 	vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE] = sregs->u.e.ivor_low[2];
810 	vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE] = sregs->u.e.ivor_low[3];
811 	vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL] = sregs->u.e.ivor_low[4];
812 	vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT] = sregs->u.e.ivor_low[5];
813 	vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM] = sregs->u.e.ivor_low[6];
814 	vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL] = sregs->u.e.ivor_low[7];
815 	vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL] = sregs->u.e.ivor_low[8];
816 	vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL] = sregs->u.e.ivor_low[9];
817 	vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER] = sregs->u.e.ivor_low[10];
818 	vcpu->arch.ivor[BOOKE_IRQPRIO_FIT] = sregs->u.e.ivor_low[11];
819 	vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG] = sregs->u.e.ivor_low[12];
820 	vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS] = sregs->u.e.ivor_low[13];
821 	vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS] = sregs->u.e.ivor_low[14];
822 	vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG] = sregs->u.e.ivor_low[15];
823 
824 	return 0;
825 }
826 
827 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
828                                   struct kvm_sregs *sregs)
829 {
830 	sregs->pvr = vcpu->arch.pvr;
831 
832 	get_sregs_base(vcpu, sregs);
833 	get_sregs_arch206(vcpu, sregs);
834 	kvmppc_core_get_sregs(vcpu, sregs);
835 	return 0;
836 }
837 
838 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
839                                   struct kvm_sregs *sregs)
840 {
841 	int ret;
842 
843 	if (vcpu->arch.pvr != sregs->pvr)
844 		return -EINVAL;
845 
846 	ret = set_sregs_base(vcpu, sregs);
847 	if (ret < 0)
848 		return ret;
849 
850 	ret = set_sregs_arch206(vcpu, sregs);
851 	if (ret < 0)
852 		return ret;
853 
854 	return kvmppc_core_set_sregs(vcpu, sregs);
855 }
856 
857 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
858 {
859 	return -ENOTSUPP;
860 }
861 
862 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
863 {
864 	return -ENOTSUPP;
865 }
866 
867 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
868                                   struct kvm_translation *tr)
869 {
870 	int r;
871 
872 	r = kvmppc_core_vcpu_translate(vcpu, tr);
873 	return r;
874 }
875 
876 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
877 {
878 	return -ENOTSUPP;
879 }
880 
881 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
882 				      struct kvm_userspace_memory_region *mem)
883 {
884 	return 0;
885 }
886 
887 void kvmppc_core_commit_memory_region(struct kvm *kvm,
888 				struct kvm_userspace_memory_region *mem)
889 {
890 }
891 
892 int kvmppc_core_init_vm(struct kvm *kvm)
893 {
894 	return 0;
895 }
896 
897 void kvmppc_core_destroy_vm(struct kvm *kvm)
898 {
899 }
900 
901 int __init kvmppc_booke_init(void)
902 {
903 	unsigned long ivor[16];
904 	unsigned long max_ivor = 0;
905 	int i;
906 
907 	/* We install our own exception handlers by hijacking IVPR. IVPR must
908 	 * be 16-bit aligned, so we need a 64KB allocation. */
909 	kvmppc_booke_handlers = __get_free_pages(GFP_KERNEL | __GFP_ZERO,
910 	                                         VCPU_SIZE_ORDER);
911 	if (!kvmppc_booke_handlers)
912 		return -ENOMEM;
913 
914 	/* XXX make sure our handlers are smaller than Linux's */
915 
916 	/* Copy our interrupt handlers to match host IVORs. That way we don't
917 	 * have to swap the IVORs on every guest/host transition. */
918 	ivor[0] = mfspr(SPRN_IVOR0);
919 	ivor[1] = mfspr(SPRN_IVOR1);
920 	ivor[2] = mfspr(SPRN_IVOR2);
921 	ivor[3] = mfspr(SPRN_IVOR3);
922 	ivor[4] = mfspr(SPRN_IVOR4);
923 	ivor[5] = mfspr(SPRN_IVOR5);
924 	ivor[6] = mfspr(SPRN_IVOR6);
925 	ivor[7] = mfspr(SPRN_IVOR7);
926 	ivor[8] = mfspr(SPRN_IVOR8);
927 	ivor[9] = mfspr(SPRN_IVOR9);
928 	ivor[10] = mfspr(SPRN_IVOR10);
929 	ivor[11] = mfspr(SPRN_IVOR11);
930 	ivor[12] = mfspr(SPRN_IVOR12);
931 	ivor[13] = mfspr(SPRN_IVOR13);
932 	ivor[14] = mfspr(SPRN_IVOR14);
933 	ivor[15] = mfspr(SPRN_IVOR15);
934 
935 	for (i = 0; i < 16; i++) {
936 		if (ivor[i] > max_ivor)
937 			max_ivor = ivor[i];
938 
939 		memcpy((void *)kvmppc_booke_handlers + ivor[i],
940 		       kvmppc_handlers_start + i * kvmppc_handler_len,
941 		       kvmppc_handler_len);
942 	}
943 	flush_icache_range(kvmppc_booke_handlers,
944 	                   kvmppc_booke_handlers + max_ivor + kvmppc_handler_len);
945 
946 	return 0;
947 }
948 
949 void __exit kvmppc_booke_exit(void)
950 {
951 	free_pages(kvmppc_booke_handlers, VCPU_SIZE_ORDER);
952 	kvm_exit();
953 }
954