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.
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.
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.
15 * Copyright IBM Corp. 2007
17 * Authors: Hollis Blanchard <hollisb@us.ibm.com>
18 * Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
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>
28 #include <asm/cputable.h>
29 #include <asm/uaccess.h>
30 #include <asm/kvm_ppc.h>
32 #include <asm/cacheflush.h>
33 #include <asm/kvm_44x.h>
38 unsigned long kvmppc_booke_handlers;
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
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) },
61 /* TODO: use vcpu_printf() */
62 void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu)
66 printk("pc: %08lx msr: %08lx\n", vcpu->arch.pc, vcpu->arch.msr);
67 printk("lr: %08lx ctr: %08lx\n", vcpu->arch.lr, vcpu->arch.ctr);
68 printk("srr0: %08lx srr1: %08lx\n", vcpu->arch.srr0, vcpu->arch.srr1);
70 printk("exceptions: %08lx\n", vcpu->arch.pending_exceptions);
72 for (i = 0; i < 32; i += 4) {
73 printk("gpr%02d: %08lx %08lx %08lx %08lx\n", i,
81 static void kvmppc_booke_queue_irqprio(struct kvm_vcpu *vcpu,
82 unsigned int priority)
84 set_bit(priority, &vcpu->arch.pending_exceptions);
87 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu)
89 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
92 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
94 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DECREMENTER);
97 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
99 return test_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
102 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
103 struct kvm_interrupt *irq)
105 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_EXTERNAL);
108 /* Deliver the interrupt of the corresponding priority, if possible. */
109 static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
110 unsigned int priority)
116 case BOOKE_IRQPRIO_PROGRAM:
117 case BOOKE_IRQPRIO_DTLB_MISS:
118 case BOOKE_IRQPRIO_ITLB_MISS:
119 case BOOKE_IRQPRIO_SYSCALL:
120 case BOOKE_IRQPRIO_DATA_STORAGE:
121 case BOOKE_IRQPRIO_INST_STORAGE:
122 case BOOKE_IRQPRIO_FP_UNAVAIL:
123 case BOOKE_IRQPRIO_AP_UNAVAIL:
124 case BOOKE_IRQPRIO_ALIGNMENT:
126 msr_mask = MSR_CE|MSR_ME|MSR_DE;
128 case BOOKE_IRQPRIO_CRITICAL:
129 case BOOKE_IRQPRIO_WATCHDOG:
130 allowed = vcpu->arch.msr & MSR_CE;
133 case BOOKE_IRQPRIO_MACHINE_CHECK:
134 allowed = vcpu->arch.msr & MSR_ME;
137 case BOOKE_IRQPRIO_EXTERNAL:
138 case BOOKE_IRQPRIO_DECREMENTER:
139 case BOOKE_IRQPRIO_FIT:
140 allowed = vcpu->arch.msr & MSR_EE;
141 msr_mask = MSR_CE|MSR_ME|MSR_DE;
143 case BOOKE_IRQPRIO_DEBUG:
144 allowed = vcpu->arch.msr & MSR_DE;
150 vcpu->arch.srr0 = vcpu->arch.pc;
151 vcpu->arch.srr1 = vcpu->arch.msr;
152 vcpu->arch.pc = vcpu->arch.ivpr | vcpu->arch.ivor[priority];
153 kvmppc_set_msr(vcpu, vcpu->arch.msr & msr_mask);
155 clear_bit(priority, &vcpu->arch.pending_exceptions);
161 /* Check pending exceptions and deliver one, if possible. */
162 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
164 unsigned long *pending = &vcpu->arch.pending_exceptions;
165 unsigned int priority;
167 priority = __ffs(*pending);
168 while (priority <= BOOKE_MAX_INTERRUPT) {
169 if (kvmppc_booke_irqprio_deliver(vcpu, priority))
172 priority = find_next_bit(pending,
173 BITS_PER_BYTE * sizeof(*pending),
181 * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
183 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
184 unsigned int exit_nr)
186 enum emulation_result er;
189 /* update before a new last_exit_type is rewritten */
190 kvmppc_update_timing_stats(vcpu);
194 run->exit_reason = KVM_EXIT_UNKNOWN;
195 run->ready_for_interrupt_injection = 1;
198 case BOOKE_INTERRUPT_MACHINE_CHECK:
199 printk("MACHINE CHECK: %lx\n", mfspr(SPRN_MCSR));
200 kvmppc_dump_vcpu(vcpu);
204 case BOOKE_INTERRUPT_EXTERNAL:
205 kvmppc_account_exit(vcpu, EXT_INTR_EXITS);
211 case BOOKE_INTERRUPT_DECREMENTER:
212 /* Since we switched IVPR back to the host's value, the host
213 * handled this interrupt the moment we enabled interrupts.
214 * Now we just offer it a chance to reschedule the guest. */
215 kvmppc_account_exit(vcpu, DEC_EXITS);
221 case BOOKE_INTERRUPT_PROGRAM:
222 if (vcpu->arch.msr & MSR_PR) {
223 /* Program traps generated by user-level software must be handled
224 * by the guest kernel. */
225 vcpu->arch.esr = vcpu->arch.fault_esr;
226 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
228 kvmppc_account_exit(vcpu, USR_PR_INST);
232 er = kvmppc_emulate_instruction(run, vcpu);
235 /* don't overwrite subtypes, just account kvm_stats */
236 kvmppc_account_exit_stat(vcpu, EMULATED_INST_EXITS);
237 /* Future optimization: only reload non-volatiles if
238 * they were actually modified by emulation. */
242 run->exit_reason = KVM_EXIT_DCR;
246 /* XXX Deliver Program interrupt to guest. */
247 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
248 __func__, vcpu->arch.pc, vcpu->arch.last_inst);
249 /* For debugging, encode the failing instruction and
250 * report it to userspace. */
251 run->hw.hardware_exit_reason = ~0ULL << 32;
252 run->hw.hardware_exit_reason |= vcpu->arch.last_inst;
260 case BOOKE_INTERRUPT_FP_UNAVAIL:
261 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_FP_UNAVAIL);
262 kvmppc_account_exit(vcpu, FP_UNAVAIL);
266 case BOOKE_INTERRUPT_DATA_STORAGE:
267 vcpu->arch.dear = vcpu->arch.fault_dear;
268 vcpu->arch.esr = vcpu->arch.fault_esr;
269 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DATA_STORAGE);
270 kvmppc_account_exit(vcpu, DSI_EXITS);
274 case BOOKE_INTERRUPT_INST_STORAGE:
275 vcpu->arch.esr = vcpu->arch.fault_esr;
276 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_INST_STORAGE);
277 kvmppc_account_exit(vcpu, ISI_EXITS);
281 case BOOKE_INTERRUPT_SYSCALL:
282 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SYSCALL);
283 kvmppc_account_exit(vcpu, SYSCALL_EXITS);
287 /* XXX move to a 440-specific file. */
288 case BOOKE_INTERRUPT_DTLB_MISS: {
289 struct kvmppc_vcpu_44x *vcpu_44x = to_44x(vcpu);
290 struct kvmppc_44x_tlbe *gtlbe;
291 unsigned long eaddr = vcpu->arch.fault_dear;
295 /* Check the guest TLB. */
296 gtlb_index = kvmppc_44x_dtlb_index(vcpu, eaddr);
297 if (gtlb_index < 0) {
298 /* The guest didn't have a mapping for it. */
299 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DTLB_MISS);
300 vcpu->arch.dear = vcpu->arch.fault_dear;
301 vcpu->arch.esr = vcpu->arch.fault_esr;
302 kvmppc_account_exit(vcpu, DTLB_REAL_MISS_EXITS);
307 gtlbe = &vcpu_44x->guest_tlb[gtlb_index];
308 vcpu->arch.paddr_accessed = tlb_xlate(gtlbe, eaddr);
309 gfn = vcpu->arch.paddr_accessed >> PAGE_SHIFT;
311 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
312 /* The guest TLB had a mapping, but the shadow TLB
313 * didn't, and it is RAM. This could be because:
314 * a) the entry is mapping the host kernel, or
315 * b) the guest used a large mapping which we're faking
316 * Either way, we need to satisfy the fault without
317 * invoking the guest. */
318 kvmppc_mmu_map(vcpu, eaddr, vcpu->arch.paddr_accessed, gtlbe->tid,
319 gtlbe->word2, get_tlb_bytes(gtlbe), gtlb_index);
320 kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
323 /* Guest has mapped and accessed a page which is not
325 r = kvmppc_emulate_mmio(run, vcpu);
326 kvmppc_account_exit(vcpu, MMIO_EXITS);
332 /* XXX move to a 440-specific file. */
333 case BOOKE_INTERRUPT_ITLB_MISS: {
334 struct kvmppc_vcpu_44x *vcpu_44x = to_44x(vcpu);
335 struct kvmppc_44x_tlbe *gtlbe;
336 unsigned long eaddr = vcpu->arch.pc;
343 /* Check the guest TLB. */
344 gtlb_index = kvmppc_44x_itlb_index(vcpu, eaddr);
345 if (gtlb_index < 0) {
346 /* The guest didn't have a mapping for it. */
347 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ITLB_MISS);
348 kvmppc_account_exit(vcpu, ITLB_REAL_MISS_EXITS);
352 kvmppc_account_exit(vcpu, ITLB_VIRT_MISS_EXITS);
354 gtlbe = &vcpu_44x->guest_tlb[gtlb_index];
355 gpaddr = tlb_xlate(gtlbe, eaddr);
356 gfn = gpaddr >> PAGE_SHIFT;
358 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
359 /* The guest TLB had a mapping, but the shadow TLB
360 * didn't. This could be because:
361 * a) the entry is mapping the host kernel, or
362 * b) the guest used a large mapping which we're faking
363 * Either way, we need to satisfy the fault without
364 * invoking the guest. */
365 kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlbe->tid,
366 gtlbe->word2, get_tlb_bytes(gtlbe), gtlb_index);
368 /* Guest mapped and leaped at non-RAM! */
369 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_MACHINE_CHECK);
375 case BOOKE_INTERRUPT_DEBUG: {
378 vcpu->arch.pc = mfspr(SPRN_CSRR0);
380 /* clear IAC events in DBSR register */
381 dbsr = mfspr(SPRN_DBSR);
382 dbsr &= DBSR_IAC1 | DBSR_IAC2 | DBSR_IAC3 | DBSR_IAC4;
383 mtspr(SPRN_DBSR, dbsr);
385 run->exit_reason = KVM_EXIT_DEBUG;
386 kvmppc_account_exit(vcpu, DEBUG_EXITS);
392 printk(KERN_EMERG "exit_nr %d\n", exit_nr);
398 kvmppc_core_deliver_interrupts(vcpu);
400 if (!(r & RESUME_HOST)) {
401 /* To avoid clobbering exit_reason, only check for signals if
402 * we aren't already exiting to userspace for some other
404 if (signal_pending(current)) {
405 run->exit_reason = KVM_EXIT_INTR;
406 r = (-EINTR << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
407 kvmppc_account_exit(vcpu, SIGNAL_EXITS);
414 /* Initial guest state: 16MB mapping 0 -> 0, PC = 0, MSR = 0, R1 = 16MB */
415 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
419 vcpu->arch.gpr[1] = (16<<20) - 8; /* -8 for the callee-save LR slot */
421 vcpu->arch.shadow_pid = 1;
423 /* Eye-catching number so we know if the guest takes an interrupt
424 * before it's programmed its own IVPR. */
425 vcpu->arch.ivpr = 0x55550000;
427 kvmppc_init_timing_stats(vcpu);
429 return kvmppc_core_vcpu_setup(vcpu);
432 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
436 regs->pc = vcpu->arch.pc;
437 regs->cr = vcpu->arch.cr;
438 regs->ctr = vcpu->arch.ctr;
439 regs->lr = vcpu->arch.lr;
440 regs->xer = vcpu->arch.xer;
441 regs->msr = vcpu->arch.msr;
442 regs->srr0 = vcpu->arch.srr0;
443 regs->srr1 = vcpu->arch.srr1;
444 regs->pid = vcpu->arch.pid;
445 regs->sprg0 = vcpu->arch.sprg0;
446 regs->sprg1 = vcpu->arch.sprg1;
447 regs->sprg2 = vcpu->arch.sprg2;
448 regs->sprg3 = vcpu->arch.sprg3;
449 regs->sprg5 = vcpu->arch.sprg4;
450 regs->sprg6 = vcpu->arch.sprg5;
451 regs->sprg7 = vcpu->arch.sprg6;
453 for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
454 regs->gpr[i] = vcpu->arch.gpr[i];
459 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
463 vcpu->arch.pc = regs->pc;
464 vcpu->arch.cr = regs->cr;
465 vcpu->arch.ctr = regs->ctr;
466 vcpu->arch.lr = regs->lr;
467 vcpu->arch.xer = regs->xer;
468 kvmppc_set_msr(vcpu, regs->msr);
469 vcpu->arch.srr0 = regs->srr0;
470 vcpu->arch.srr1 = regs->srr1;
471 vcpu->arch.sprg0 = regs->sprg0;
472 vcpu->arch.sprg1 = regs->sprg1;
473 vcpu->arch.sprg2 = regs->sprg2;
474 vcpu->arch.sprg3 = regs->sprg3;
475 vcpu->arch.sprg5 = regs->sprg4;
476 vcpu->arch.sprg6 = regs->sprg5;
477 vcpu->arch.sprg7 = regs->sprg6;
479 for (i = 0; i < ARRAY_SIZE(vcpu->arch.gpr); i++)
480 vcpu->arch.gpr[i] = regs->gpr[i];
485 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
486 struct kvm_sregs *sregs)
491 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
492 struct kvm_sregs *sregs)
497 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
502 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
507 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
508 struct kvm_translation *tr)
510 return kvmppc_core_vcpu_translate(vcpu, tr);
513 int kvmppc_booke_init(void)
515 unsigned long ivor[16];
516 unsigned long max_ivor = 0;
519 /* We install our own exception handlers by hijacking IVPR. IVPR must
520 * be 16-bit aligned, so we need a 64KB allocation. */
521 kvmppc_booke_handlers = __get_free_pages(GFP_KERNEL | __GFP_ZERO,
523 if (!kvmppc_booke_handlers)
526 /* XXX make sure our handlers are smaller than Linux's */
528 /* Copy our interrupt handlers to match host IVORs. That way we don't
529 * have to swap the IVORs on every guest/host transition. */
530 ivor[0] = mfspr(SPRN_IVOR0);
531 ivor[1] = mfspr(SPRN_IVOR1);
532 ivor[2] = mfspr(SPRN_IVOR2);
533 ivor[3] = mfspr(SPRN_IVOR3);
534 ivor[4] = mfspr(SPRN_IVOR4);
535 ivor[5] = mfspr(SPRN_IVOR5);
536 ivor[6] = mfspr(SPRN_IVOR6);
537 ivor[7] = mfspr(SPRN_IVOR7);
538 ivor[8] = mfspr(SPRN_IVOR8);
539 ivor[9] = mfspr(SPRN_IVOR9);
540 ivor[10] = mfspr(SPRN_IVOR10);
541 ivor[11] = mfspr(SPRN_IVOR11);
542 ivor[12] = mfspr(SPRN_IVOR12);
543 ivor[13] = mfspr(SPRN_IVOR13);
544 ivor[14] = mfspr(SPRN_IVOR14);
545 ivor[15] = mfspr(SPRN_IVOR15);
547 for (i = 0; i < 16; i++) {
548 if (ivor[i] > max_ivor)
551 memcpy((void *)kvmppc_booke_handlers + ivor[i],
552 kvmppc_handlers_start + i * kvmppc_handler_len,
555 flush_icache_range(kvmppc_booke_handlers,
556 kvmppc_booke_handlers + max_ivor + kvmppc_handler_len);
561 void __exit kvmppc_booke_exit(void)
563 free_pages(kvmppc_booke_handlers, VCPU_SIZE_ORDER);