KVM: ppc: remove last 44x-specific bits from booke.c
[linux-2.6] / arch / powerpc / kvm / booke.c
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/fs.h>
27
28 #include <asm/cputable.h>
29 #include <asm/uaccess.h>
30 #include <asm/kvm_ppc.h>
31 #include "timing.h"
32 #include <asm/cacheflush.h>
33
34 #include "booke.h"
35
36 unsigned long kvmppc_booke_handlers;
37
38 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
39 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
40
41 struct kvm_stats_debugfs_item debugfs_entries[] = {
42         { "mmio",       VCPU_STAT(mmio_exits) },
43         { "dcr",        VCPU_STAT(dcr_exits) },
44         { "sig",        VCPU_STAT(signal_exits) },
45         { "itlb_r",     VCPU_STAT(itlb_real_miss_exits) },
46         { "itlb_v",     VCPU_STAT(itlb_virt_miss_exits) },
47         { "dtlb_r",     VCPU_STAT(dtlb_real_miss_exits) },
48         { "dtlb_v",     VCPU_STAT(dtlb_virt_miss_exits) },
49         { "sysc",       VCPU_STAT(syscall_exits) },
50         { "isi",        VCPU_STAT(isi_exits) },
51         { "dsi",        VCPU_STAT(dsi_exits) },
52         { "inst_emu",   VCPU_STAT(emulated_inst_exits) },
53         { "dec",        VCPU_STAT(dec_exits) },
54         { "ext_intr",   VCPU_STAT(ext_intr_exits) },
55         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
56         { NULL }
57 };
58
59 /* TODO: use vcpu_printf() */
60 void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu)
61 {
62         int i;
63
64         printk("pc:   %08lx msr:  %08lx\n", vcpu->arch.pc, vcpu->arch.msr);
65         printk("lr:   %08lx ctr:  %08lx\n", vcpu->arch.lr, vcpu->arch.ctr);
66         printk("srr0: %08lx srr1: %08lx\n", vcpu->arch.srr0, vcpu->arch.srr1);
67
68         printk("exceptions: %08lx\n", vcpu->arch.pending_exceptions);
69
70         for (i = 0; i < 32; i += 4) {
71                 printk("gpr%02d: %08lx %08lx %08lx %08lx\n", i,
72                        vcpu->arch.gpr[i],
73                        vcpu->arch.gpr[i+1],
74                        vcpu->arch.gpr[i+2],
75                        vcpu->arch.gpr[i+3]);
76         }
77 }
78
79 static void kvmppc_booke_queue_irqprio(struct kvm_vcpu *vcpu,
80                                        unsigned int priority)
81 {
82         set_bit(priority, &vcpu->arch.pending_exceptions);
83 }
84
85 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu)
86 {
87         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
88 }
89
90 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
91 {
92         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DECREMENTER);
93 }
94
95 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
96 {
97         return test_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
98 }
99
100 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
101                                 struct kvm_interrupt *irq)
102 {
103         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_EXTERNAL);
104 }
105
106 /* Deliver the interrupt of the corresponding priority, if possible. */
107 static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
108                                         unsigned int priority)
109 {
110         int allowed = 0;
111         ulong msr_mask;
112
113         switch (priority) {
114         case BOOKE_IRQPRIO_PROGRAM:
115         case BOOKE_IRQPRIO_DTLB_MISS:
116         case BOOKE_IRQPRIO_ITLB_MISS:
117         case BOOKE_IRQPRIO_SYSCALL:
118         case BOOKE_IRQPRIO_DATA_STORAGE:
119         case BOOKE_IRQPRIO_INST_STORAGE:
120         case BOOKE_IRQPRIO_FP_UNAVAIL:
121         case BOOKE_IRQPRIO_AP_UNAVAIL:
122         case BOOKE_IRQPRIO_ALIGNMENT:
123                 allowed = 1;
124                 msr_mask = MSR_CE|MSR_ME|MSR_DE;
125                 break;
126         case BOOKE_IRQPRIO_CRITICAL:
127         case BOOKE_IRQPRIO_WATCHDOG:
128                 allowed = vcpu->arch.msr & MSR_CE;
129                 msr_mask = MSR_ME;
130                 break;
131         case BOOKE_IRQPRIO_MACHINE_CHECK:
132                 allowed = vcpu->arch.msr & MSR_ME;
133                 msr_mask = 0;
134                 break;
135         case BOOKE_IRQPRIO_EXTERNAL:
136         case BOOKE_IRQPRIO_DECREMENTER:
137         case BOOKE_IRQPRIO_FIT:
138                 allowed = vcpu->arch.msr & MSR_EE;
139                 msr_mask = MSR_CE|MSR_ME|MSR_DE;
140                 break;
141         case BOOKE_IRQPRIO_DEBUG:
142                 allowed = vcpu->arch.msr & MSR_DE;
143                 msr_mask = MSR_ME;
144                 break;
145         }
146
147         if (allowed) {
148                 vcpu->arch.srr0 = vcpu->arch.pc;
149                 vcpu->arch.srr1 = vcpu->arch.msr;
150                 vcpu->arch.pc = vcpu->arch.ivpr | vcpu->arch.ivor[priority];
151                 kvmppc_set_msr(vcpu, vcpu->arch.msr & msr_mask);
152
153                 clear_bit(priority, &vcpu->arch.pending_exceptions);
154         }
155
156         return allowed;
157 }
158
159 /* Check pending exceptions and deliver one, if possible. */
160 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
161 {
162         unsigned long *pending = &vcpu->arch.pending_exceptions;
163         unsigned int priority;
164
165         priority = __ffs(*pending);
166         while (priority <= BOOKE_MAX_INTERRUPT) {
167                 if (kvmppc_booke_irqprio_deliver(vcpu, priority))
168                         break;
169
170                 priority = find_next_bit(pending,
171                                          BITS_PER_BYTE * sizeof(*pending),
172                                          priority + 1);
173         }
174 }
175
176 /**
177  * kvmppc_handle_exit
178  *
179  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
180  */
181 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
182                        unsigned int exit_nr)
183 {
184         enum emulation_result er;
185         int r = RESUME_HOST;
186
187         /* update before a new last_exit_type is rewritten */
188         kvmppc_update_timing_stats(vcpu);
189
190         local_irq_enable();
191
192         run->exit_reason = KVM_EXIT_UNKNOWN;
193         run->ready_for_interrupt_injection = 1;
194
195         switch (exit_nr) {
196         case BOOKE_INTERRUPT_MACHINE_CHECK:
197                 printk("MACHINE CHECK: %lx\n", mfspr(SPRN_MCSR));
198                 kvmppc_dump_vcpu(vcpu);
199                 r = RESUME_HOST;
200                 break;
201
202         case BOOKE_INTERRUPT_EXTERNAL:
203                 kvmppc_account_exit(vcpu, EXT_INTR_EXITS);
204                 if (need_resched())
205                         cond_resched();
206                 r = RESUME_GUEST;
207                 break;
208
209         case BOOKE_INTERRUPT_DECREMENTER:
210                 /* Since we switched IVPR back to the host's value, the host
211                  * handled this interrupt the moment we enabled interrupts.
212                  * Now we just offer it a chance to reschedule the guest. */
213                 kvmppc_account_exit(vcpu, DEC_EXITS);
214                 if (need_resched())
215                         cond_resched();
216                 r = RESUME_GUEST;
217                 break;
218
219         case BOOKE_INTERRUPT_PROGRAM:
220                 if (vcpu->arch.msr & MSR_PR) {
221                         /* Program traps generated by user-level software must be handled
222                          * by the guest kernel. */
223                         vcpu->arch.esr = vcpu->arch.fault_esr;
224                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
225                         r = RESUME_GUEST;
226                         kvmppc_account_exit(vcpu, USR_PR_INST);
227                         break;
228                 }
229
230                 er = kvmppc_emulate_instruction(run, vcpu);
231                 switch (er) {
232                 case EMULATE_DONE:
233                         /* don't overwrite subtypes, just account kvm_stats */
234                         kvmppc_account_exit_stat(vcpu, EMULATED_INST_EXITS);
235                         /* Future optimization: only reload non-volatiles if
236                          * they were actually modified by emulation. */
237                         r = RESUME_GUEST_NV;
238                         break;
239                 case EMULATE_DO_DCR:
240                         run->exit_reason = KVM_EXIT_DCR;
241                         r = RESUME_HOST;
242                         break;
243                 case EMULATE_FAIL:
244                         /* XXX Deliver Program interrupt to guest. */
245                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
246                                __func__, vcpu->arch.pc, vcpu->arch.last_inst);
247                         /* For debugging, encode the failing instruction and
248                          * report it to userspace. */
249                         run->hw.hardware_exit_reason = ~0ULL << 32;
250                         run->hw.hardware_exit_reason |= vcpu->arch.last_inst;
251                         r = RESUME_HOST;
252                         break;
253                 default:
254                         BUG();
255                 }
256                 break;
257
258         case BOOKE_INTERRUPT_FP_UNAVAIL:
259                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_FP_UNAVAIL);
260                 kvmppc_account_exit(vcpu, FP_UNAVAIL);
261                 r = RESUME_GUEST;
262                 break;
263
264         case BOOKE_INTERRUPT_DATA_STORAGE:
265                 vcpu->arch.dear = vcpu->arch.fault_dear;
266                 vcpu->arch.esr = vcpu->arch.fault_esr;
267                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DATA_STORAGE);
268                 kvmppc_account_exit(vcpu, DSI_EXITS);
269                 r = RESUME_GUEST;
270                 break;
271
272         case BOOKE_INTERRUPT_INST_STORAGE:
273                 vcpu->arch.esr = vcpu->arch.fault_esr;
274                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_INST_STORAGE);
275                 kvmppc_account_exit(vcpu, ISI_EXITS);
276                 r = RESUME_GUEST;
277                 break;
278
279         case BOOKE_INTERRUPT_SYSCALL:
280                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SYSCALL);
281                 kvmppc_account_exit(vcpu, SYSCALL_EXITS);
282                 r = RESUME_GUEST;
283                 break;
284
285         case BOOKE_INTERRUPT_DTLB_MISS: {
286                 unsigned long eaddr = vcpu->arch.fault_dear;
287                 int gtlb_index;
288                 gpa_t gpaddr;
289                 gfn_t gfn;
290
291                 /* Check the guest TLB. */
292                 gtlb_index = kvmppc_mmu_dtlb_index(vcpu, eaddr);
293                 if (gtlb_index < 0) {
294                         /* The guest didn't have a mapping for it. */
295                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DTLB_MISS);
296                         vcpu->arch.dear = vcpu->arch.fault_dear;
297                         vcpu->arch.esr = vcpu->arch.fault_esr;
298                         kvmppc_account_exit(vcpu, DTLB_REAL_MISS_EXITS);
299                         r = RESUME_GUEST;
300                         break;
301                 }
302
303                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
304                 gfn = gpaddr >> PAGE_SHIFT;
305
306                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
307                         /* The guest TLB had a mapping, but the shadow TLB
308                          * didn't, and it is RAM. This could be because:
309                          * a) the entry is mapping the host kernel, or
310                          * b) the guest used a large mapping which we're faking
311                          * Either way, we need to satisfy the fault without
312                          * invoking the guest. */
313                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
314                         kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
315                         r = RESUME_GUEST;
316                 } else {
317                         /* Guest has mapped and accessed a page which is not
318                          * actually RAM. */
319                         vcpu->arch.paddr_accessed = gpaddr;
320                         r = kvmppc_emulate_mmio(run, vcpu);
321                         kvmppc_account_exit(vcpu, MMIO_EXITS);
322                 }
323
324                 break;
325         }
326
327         case BOOKE_INTERRUPT_ITLB_MISS: {
328                 unsigned long eaddr = vcpu->arch.pc;
329                 gpa_t gpaddr;
330                 gfn_t gfn;
331                 int gtlb_index;
332
333                 r = RESUME_GUEST;
334
335                 /* Check the guest TLB. */
336                 gtlb_index = kvmppc_mmu_itlb_index(vcpu, eaddr);
337                 if (gtlb_index < 0) {
338                         /* The guest didn't have a mapping for it. */
339                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ITLB_MISS);
340                         kvmppc_account_exit(vcpu, ITLB_REAL_MISS_EXITS);
341                         break;
342                 }
343
344                 kvmppc_account_exit(vcpu, ITLB_VIRT_MISS_EXITS);
345
346                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
347                 gfn = gpaddr >> PAGE_SHIFT;
348
349                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
350                         /* The guest TLB had a mapping, but the shadow TLB
351                          * didn't. This could be because:
352                          * a) the entry is mapping the host kernel, or
353                          * b) the guest used a large mapping which we're faking
354                          * Either way, we need to satisfy the fault without
355                          * invoking the guest. */
356                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
357                 } else {
358                         /* Guest mapped and leaped at non-RAM! */
359                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_MACHINE_CHECK);
360                 }
361
362                 break;
363         }
364
365         case BOOKE_INTERRUPT_DEBUG: {
366                 u32 dbsr;
367
368                 vcpu->arch.pc = mfspr(SPRN_CSRR0);
369
370                 /* clear IAC events in DBSR register */
371                 dbsr = mfspr(SPRN_DBSR);
372                 dbsr &= DBSR_IAC1 | DBSR_IAC2 | DBSR_IAC3 | DBSR_IAC4;
373                 mtspr(SPRN_DBSR, dbsr);
374
375                 run->exit_reason = KVM_EXIT_DEBUG;
376                 kvmppc_account_exit(vcpu, DEBUG_EXITS);
377                 r = RESUME_HOST;
378                 break;
379         }
380
381         default:
382                 printk(KERN_EMERG "exit_nr %d\n", exit_nr);
383                 BUG();
384         }
385
386         local_irq_disable();
387
388         kvmppc_core_deliver_interrupts(vcpu);
389
390         if (!(r & RESUME_HOST)) {
391                 /* To avoid clobbering exit_reason, only check for signals if
392                  * we aren't already exiting to userspace for some other
393                  * reason. */
394                 if (signal_pending(current)) {
395                         run->exit_reason = KVM_EXIT_INTR;
396                         r = (-EINTR << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
397                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
398                 }
399         }
400
401         return r;
402 }
403
404 /* Initial guest state: 16MB mapping 0 -> 0, PC = 0, MSR = 0, R1 = 16MB */
405 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
406 {
407         vcpu->arch.pc = 0;
408         vcpu->arch.msr = 0;
409         vcpu->arch.gpr[1] = (16<<20) - 8; /* -8 for the callee-save LR slot */
410
411         vcpu->arch.shadow_pid = 1;
412
413         /* Eye-catching number so we know if the guest takes an interrupt
414          * before it's programmed its own IVPR. */
415         vcpu->arch.ivpr = 0x55550000;
416
417         kvmppc_init_timing_stats(vcpu);
418
419         return kvmppc_core_vcpu_setup(vcpu);
420 }
421
422 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
423 {
424         int i;
425
426         regs->pc = vcpu->arch.pc;
427         regs->cr = vcpu->arch.cr;
428         regs->ctr = vcpu->arch.ctr;
429         regs->lr = vcpu->arch.lr;
430         regs->xer = vcpu->arch.xer;
431         regs->msr = vcpu->arch.msr;
432         regs->srr0 = vcpu->arch.srr0;
433         regs->srr1 = vcpu->arch.srr1;
434         regs->pid = vcpu->arch.pid;
435         regs->sprg0 = vcpu->arch.sprg0;
436         regs->sprg1 = vcpu->arch.sprg1;
437         regs->sprg2 = vcpu->arch.sprg2;
438         regs->sprg3 = vcpu->arch.sprg3;
439         regs->sprg5 = vcpu->arch.sprg4;
440         regs->sprg6 = vcpu->arch.sprg5;
441         regs->sprg7 = vcpu->arch.sprg6;
442
443         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
444                 regs->gpr[i] = vcpu->arch.gpr[i];
445
446         return 0;
447 }
448
449 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
450 {
451         int i;
452
453         vcpu->arch.pc = regs->pc;
454         vcpu->arch.cr = regs->cr;
455         vcpu->arch.ctr = regs->ctr;
456         vcpu->arch.lr = regs->lr;
457         vcpu->arch.xer = regs->xer;
458         kvmppc_set_msr(vcpu, regs->msr);
459         vcpu->arch.srr0 = regs->srr0;
460         vcpu->arch.srr1 = regs->srr1;
461         vcpu->arch.sprg0 = regs->sprg0;
462         vcpu->arch.sprg1 = regs->sprg1;
463         vcpu->arch.sprg2 = regs->sprg2;
464         vcpu->arch.sprg3 = regs->sprg3;
465         vcpu->arch.sprg5 = regs->sprg4;
466         vcpu->arch.sprg6 = regs->sprg5;
467         vcpu->arch.sprg7 = regs->sprg6;
468
469         for (i = 0; i < ARRAY_SIZE(vcpu->arch.gpr); i++)
470                 vcpu->arch.gpr[i] = regs->gpr[i];
471
472         return 0;
473 }
474
475 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
476                                   struct kvm_sregs *sregs)
477 {
478         return -ENOTSUPP;
479 }
480
481 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
482                                   struct kvm_sregs *sregs)
483 {
484         return -ENOTSUPP;
485 }
486
487 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
488 {
489         return -ENOTSUPP;
490 }
491
492 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
493 {
494         return -ENOTSUPP;
495 }
496
497 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
498                                   struct kvm_translation *tr)
499 {
500         return kvmppc_core_vcpu_translate(vcpu, tr);
501 }
502
503 int kvmppc_booke_init(void)
504 {
505         unsigned long ivor[16];
506         unsigned long max_ivor = 0;
507         int i;
508
509         /* We install our own exception handlers by hijacking IVPR. IVPR must
510          * be 16-bit aligned, so we need a 64KB allocation. */
511         kvmppc_booke_handlers = __get_free_pages(GFP_KERNEL | __GFP_ZERO,
512                                                  VCPU_SIZE_ORDER);
513         if (!kvmppc_booke_handlers)
514                 return -ENOMEM;
515
516         /* XXX make sure our handlers are smaller than Linux's */
517
518         /* Copy our interrupt handlers to match host IVORs. That way we don't
519          * have to swap the IVORs on every guest/host transition. */
520         ivor[0] = mfspr(SPRN_IVOR0);
521         ivor[1] = mfspr(SPRN_IVOR1);
522         ivor[2] = mfspr(SPRN_IVOR2);
523         ivor[3] = mfspr(SPRN_IVOR3);
524         ivor[4] = mfspr(SPRN_IVOR4);
525         ivor[5] = mfspr(SPRN_IVOR5);
526         ivor[6] = mfspr(SPRN_IVOR6);
527         ivor[7] = mfspr(SPRN_IVOR7);
528         ivor[8] = mfspr(SPRN_IVOR8);
529         ivor[9] = mfspr(SPRN_IVOR9);
530         ivor[10] = mfspr(SPRN_IVOR10);
531         ivor[11] = mfspr(SPRN_IVOR11);
532         ivor[12] = mfspr(SPRN_IVOR12);
533         ivor[13] = mfspr(SPRN_IVOR13);
534         ivor[14] = mfspr(SPRN_IVOR14);
535         ivor[15] = mfspr(SPRN_IVOR15);
536
537         for (i = 0; i < 16; i++) {
538                 if (ivor[i] > max_ivor)
539                         max_ivor = ivor[i];
540
541                 memcpy((void *)kvmppc_booke_handlers + ivor[i],
542                        kvmppc_handlers_start + i * kvmppc_handler_len,
543                        kvmppc_handler_len);
544         }
545         flush_icache_range(kvmppc_booke_handlers,
546                            kvmppc_booke_handlers + max_ivor + kvmppc_handler_len);
547
548         return 0;
549 }
550
551 void __exit kvmppc_booke_exit(void)
552 {
553         free_pages(kvmppc_booke_handlers, VCPU_SIZE_ORDER);
554         kvm_exit();
555 }