]> git.karo-electronics.de Git - karo-tx-linux.git/blob - arch/powerpc/kvm/booke.c
KVM: ppc: optimize irq delivery path
[karo-tx-linux.git] / 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 #include <asm/cputable.h>
28 #include <asm/uaccess.h>
29 #include <asm/kvm_ppc.h>
30 #include <asm/cacheflush.h>
31
32 #include "booke.h"
33 #include "44x_tlb.h"
34
35 unsigned long kvmppc_booke_handlers;
36
37 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
38 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
39
40 struct kvm_stats_debugfs_item debugfs_entries[] = {
41         { "mmio",       VCPU_STAT(mmio_exits) },
42         { "dcr",        VCPU_STAT(dcr_exits) },
43         { "sig",        VCPU_STAT(signal_exits) },
44         { "itlb_r",     VCPU_STAT(itlb_real_miss_exits) },
45         { "itlb_v",     VCPU_STAT(itlb_virt_miss_exits) },
46         { "dtlb_r",     VCPU_STAT(dtlb_real_miss_exits) },
47         { "dtlb_v",     VCPU_STAT(dtlb_virt_miss_exits) },
48         { "sysc",       VCPU_STAT(syscall_exits) },
49         { "isi",        VCPU_STAT(isi_exits) },
50         { "dsi",        VCPU_STAT(dsi_exits) },
51         { "inst_emu",   VCPU_STAT(emulated_inst_exits) },
52         { "dec",        VCPU_STAT(dec_exits) },
53         { "ext_intr",   VCPU_STAT(ext_intr_exits) },
54         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
55         { NULL }
56 };
57
58 /* TODO: use vcpu_printf() */
59 void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu)
60 {
61         int i;
62
63         printk("pc:   %08lx msr:  %08lx\n", vcpu->arch.pc, vcpu->arch.msr);
64         printk("lr:   %08lx ctr:  %08lx\n", vcpu->arch.lr, vcpu->arch.ctr);
65         printk("srr0: %08lx srr1: %08lx\n", vcpu->arch.srr0, vcpu->arch.srr1);
66
67         printk("exceptions: %08lx\n", vcpu->arch.pending_exceptions);
68
69         for (i = 0; i < 32; i += 4) {
70                 printk("gpr%02d: %08lx %08lx %08lx %08lx\n", i,
71                        vcpu->arch.gpr[i],
72                        vcpu->arch.gpr[i+1],
73                        vcpu->arch.gpr[i+2],
74                        vcpu->arch.gpr[i+3]);
75         }
76 }
77
78 static void kvmppc_booke_queue_irqprio(struct kvm_vcpu *vcpu,
79                                        unsigned int priority)
80 {
81         set_bit(priority, &vcpu->arch.pending_exceptions);
82 }
83
84 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu)
85 {
86         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
87 }
88
89 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
90 {
91         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DECREMENTER);
92 }
93
94 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
95 {
96         return test_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
97 }
98
99 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
100                                 struct kvm_interrupt *irq)
101 {
102         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_EXTERNAL);
103 }
104
105 /* Deliver the interrupt of the corresponding priority, if possible. */
106 static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
107                                         unsigned int priority)
108 {
109         int allowed = 0;
110         ulong msr_mask;
111
112         switch (priority) {
113         case BOOKE_IRQPRIO_PROGRAM:
114         case BOOKE_IRQPRIO_DTLB_MISS:
115         case BOOKE_IRQPRIO_ITLB_MISS:
116         case BOOKE_IRQPRIO_SYSCALL:
117         case BOOKE_IRQPRIO_DATA_STORAGE:
118         case BOOKE_IRQPRIO_INST_STORAGE:
119         case BOOKE_IRQPRIO_FP_UNAVAIL:
120         case BOOKE_IRQPRIO_AP_UNAVAIL:
121         case BOOKE_IRQPRIO_ALIGNMENT:
122                 allowed = 1;
123                 msr_mask = MSR_CE|MSR_ME|MSR_DE;
124                 break;
125         case BOOKE_IRQPRIO_CRITICAL:
126         case BOOKE_IRQPRIO_WATCHDOG:
127                 allowed = vcpu->arch.msr & MSR_CE;
128                 msr_mask = MSR_ME;
129                 break;
130         case BOOKE_IRQPRIO_MACHINE_CHECK:
131                 allowed = vcpu->arch.msr & MSR_ME;
132                 msr_mask = 0;
133                 break;
134         case BOOKE_IRQPRIO_EXTERNAL:
135         case BOOKE_IRQPRIO_DECREMENTER:
136         case BOOKE_IRQPRIO_FIT:
137                 allowed = vcpu->arch.msr & MSR_EE;
138                 msr_mask = MSR_CE|MSR_ME|MSR_DE;
139                 break;
140         case BOOKE_IRQPRIO_DEBUG:
141                 allowed = vcpu->arch.msr & MSR_DE;
142                 msr_mask = MSR_ME;
143                 break;
144         }
145
146         if (allowed) {
147                 vcpu->arch.srr0 = vcpu->arch.pc;
148                 vcpu->arch.srr1 = vcpu->arch.msr;
149                 vcpu->arch.pc = vcpu->arch.ivpr | vcpu->arch.ivor[priority];
150                 kvmppc_set_msr(vcpu, vcpu->arch.msr & msr_mask);
151
152                 clear_bit(priority, &vcpu->arch.pending_exceptions);
153         }
154
155         return allowed;
156 }
157
158 /* Check pending exceptions and deliver one, if possible. */
159 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
160 {
161         unsigned long *pending = &vcpu->arch.pending_exceptions;
162         unsigned int priority;
163
164         priority = __ffs(*pending);
165         while (priority <= BOOKE_MAX_INTERRUPT) {
166                 if (kvmppc_booke_irqprio_deliver(vcpu, priority))
167                         break;
168
169                 priority = find_next_bit(pending,
170                                          BITS_PER_BYTE * sizeof(*pending),
171                                          priority + 1);
172         }
173 }
174
175 /**
176  * kvmppc_handle_exit
177  *
178  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
179  */
180 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
181                        unsigned int exit_nr)
182 {
183         enum emulation_result er;
184         int r = RESUME_HOST;
185
186         local_irq_enable();
187
188         run->exit_reason = KVM_EXIT_UNKNOWN;
189         run->ready_for_interrupt_injection = 1;
190
191         switch (exit_nr) {
192         case BOOKE_INTERRUPT_MACHINE_CHECK:
193                 printk("MACHINE CHECK: %lx\n", mfspr(SPRN_MCSR));
194                 kvmppc_dump_vcpu(vcpu);
195                 r = RESUME_HOST;
196                 break;
197
198         case BOOKE_INTERRUPT_EXTERNAL:
199                 vcpu->stat.ext_intr_exits++;
200                 if (need_resched())
201                         cond_resched();
202                 r = RESUME_GUEST;
203                 break;
204
205         case BOOKE_INTERRUPT_DECREMENTER:
206                 /* Since we switched IVPR back to the host's value, the host
207                  * handled this interrupt the moment we enabled interrupts.
208                  * Now we just offer it a chance to reschedule the guest. */
209
210                 /* XXX At this point the TLB still holds our shadow TLB, so if
211                  * we do reschedule the host will fault over it. Perhaps we
212                  * should politely restore the host's entries to minimize
213                  * misses before ceding control. */
214                 vcpu->stat.dec_exits++;
215                 if (need_resched())
216                         cond_resched();
217                 r = RESUME_GUEST;
218                 break;
219
220         case BOOKE_INTERRUPT_PROGRAM:
221                 if (vcpu->arch.msr & MSR_PR) {
222                         /* Program traps generated by user-level software must be handled
223                          * by the guest kernel. */
224                         vcpu->arch.esr = vcpu->arch.fault_esr;
225                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
226                         r = RESUME_GUEST;
227                         break;
228                 }
229
230                 er = kvmppc_emulate_instruction(run, vcpu);
231                 switch (er) {
232                 case EMULATE_DONE:
233                         /* Future optimization: only reload non-volatiles if
234                          * they were actually modified by emulation. */
235                         vcpu->stat.emulated_inst_exits++;
236                         r = RESUME_GUEST_NV;
237                         break;
238                 case EMULATE_DO_DCR:
239                         run->exit_reason = KVM_EXIT_DCR;
240                         vcpu->stat.dcr_exits++;
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                 r = RESUME_GUEST;
261                 break;
262
263         case BOOKE_INTERRUPT_DATA_STORAGE:
264                 vcpu->arch.dear = vcpu->arch.fault_dear;
265                 vcpu->arch.esr = vcpu->arch.fault_esr;
266                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DATA_STORAGE);
267                 vcpu->stat.dsi_exits++;
268                 r = RESUME_GUEST;
269                 break;
270
271         case BOOKE_INTERRUPT_INST_STORAGE:
272                 vcpu->arch.esr = vcpu->arch.fault_esr;
273                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_INST_STORAGE);
274                 vcpu->stat.isi_exits++;
275                 r = RESUME_GUEST;
276                 break;
277
278         case BOOKE_INTERRUPT_SYSCALL:
279                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SYSCALL);
280                 vcpu->stat.syscall_exits++;
281                 r = RESUME_GUEST;
282                 break;
283
284         case BOOKE_INTERRUPT_DTLB_MISS: {
285                 struct kvmppc_44x_tlbe *gtlbe;
286                 unsigned long eaddr = vcpu->arch.fault_dear;
287                 gfn_t gfn;
288
289                 /* Check the guest TLB. */
290                 gtlbe = kvmppc_44x_dtlb_search(vcpu, eaddr);
291                 if (!gtlbe) {
292                         /* The guest didn't have a mapping for it. */
293                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DTLB_MISS);
294                         vcpu->arch.dear = vcpu->arch.fault_dear;
295                         vcpu->arch.esr = vcpu->arch.fault_esr;
296                         vcpu->stat.dtlb_real_miss_exits++;
297                         r = RESUME_GUEST;
298                         break;
299                 }
300
301                 vcpu->arch.paddr_accessed = tlb_xlate(gtlbe, eaddr);
302                 gfn = vcpu->arch.paddr_accessed >> PAGE_SHIFT;
303
304                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
305                         /* The guest TLB had a mapping, but the shadow TLB
306                          * didn't, and it is RAM. This could be because:
307                          * a) the entry is mapping the host kernel, or
308                          * b) the guest used a large mapping which we're faking
309                          * Either way, we need to satisfy the fault without
310                          * invoking the guest. */
311                         kvmppc_mmu_map(vcpu, eaddr, gfn, gtlbe->tid,
312                                        gtlbe->word2);
313                         vcpu->stat.dtlb_virt_miss_exits++;
314                         r = RESUME_GUEST;
315                 } else {
316                         /* Guest has mapped and accessed a page which is not
317                          * actually RAM. */
318                         r = kvmppc_emulate_mmio(run, vcpu);
319                         vcpu->stat.mmio_exits++;
320                 }
321
322                 break;
323         }
324
325         case BOOKE_INTERRUPT_ITLB_MISS: {
326                 struct kvmppc_44x_tlbe *gtlbe;
327                 unsigned long eaddr = vcpu->arch.pc;
328                 gfn_t gfn;
329
330                 r = RESUME_GUEST;
331
332                 /* Check the guest TLB. */
333                 gtlbe = kvmppc_44x_itlb_search(vcpu, eaddr);
334                 if (!gtlbe) {
335                         /* The guest didn't have a mapping for it. */
336                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ITLB_MISS);
337                         vcpu->stat.itlb_real_miss_exits++;
338                         break;
339                 }
340
341                 vcpu->stat.itlb_virt_miss_exits++;
342
343                 gfn = tlb_xlate(gtlbe, eaddr) >> PAGE_SHIFT;
344
345                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
346                         /* The guest TLB had a mapping, but the shadow TLB
347                          * didn't. This could be because:
348                          * a) the entry is mapping the host kernel, or
349                          * b) the guest used a large mapping which we're faking
350                          * Either way, we need to satisfy the fault without
351                          * invoking the guest. */
352                         kvmppc_mmu_map(vcpu, eaddr, gfn, gtlbe->tid,
353                                        gtlbe->word2);
354                 } else {
355                         /* Guest mapped and leaped at non-RAM! */
356                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_MACHINE_CHECK);
357                 }
358
359                 break;
360         }
361
362         case BOOKE_INTERRUPT_DEBUG: {
363                 u32 dbsr;
364
365                 vcpu->arch.pc = mfspr(SPRN_CSRR0);
366
367                 /* clear IAC events in DBSR register */
368                 dbsr = mfspr(SPRN_DBSR);
369                 dbsr &= DBSR_IAC1 | DBSR_IAC2 | DBSR_IAC3 | DBSR_IAC4;
370                 mtspr(SPRN_DBSR, dbsr);
371
372                 run->exit_reason = KVM_EXIT_DEBUG;
373                 r = RESUME_HOST;
374                 break;
375         }
376
377         default:
378                 printk(KERN_EMERG "exit_nr %d\n", exit_nr);
379                 BUG();
380         }
381
382         local_irq_disable();
383
384         kvmppc_core_deliver_interrupts(vcpu);
385
386         if (!(r & RESUME_HOST)) {
387                 /* To avoid clobbering exit_reason, only check for signals if
388                  * we aren't already exiting to userspace for some other
389                  * reason. */
390                 if (signal_pending(current)) {
391                         run->exit_reason = KVM_EXIT_INTR;
392                         r = (-EINTR << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
393                         vcpu->stat.signal_exits++;
394                 }
395         }
396
397         return r;
398 }
399
400 /* Initial guest state: 16MB mapping 0 -> 0, PC = 0, MSR = 0, R1 = 16MB */
401 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
402 {
403         vcpu->arch.pc = 0;
404         vcpu->arch.msr = 0;
405         vcpu->arch.gpr[1] = (16<<20) - 8; /* -8 for the callee-save LR slot */
406
407         vcpu->arch.shadow_pid = 1;
408
409         /* Eye-catching number so we know if the guest takes an interrupt
410          * before it's programmed its own IVPR. */
411         vcpu->arch.ivpr = 0x55550000;
412
413         return kvmppc_core_vcpu_setup(vcpu);
414 }
415
416 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
417 {
418         int i;
419
420         regs->pc = vcpu->arch.pc;
421         regs->cr = vcpu->arch.cr;
422         regs->ctr = vcpu->arch.ctr;
423         regs->lr = vcpu->arch.lr;
424         regs->xer = vcpu->arch.xer;
425         regs->msr = vcpu->arch.msr;
426         regs->srr0 = vcpu->arch.srr0;
427         regs->srr1 = vcpu->arch.srr1;
428         regs->pid = vcpu->arch.pid;
429         regs->sprg0 = vcpu->arch.sprg0;
430         regs->sprg1 = vcpu->arch.sprg1;
431         regs->sprg2 = vcpu->arch.sprg2;
432         regs->sprg3 = vcpu->arch.sprg3;
433         regs->sprg5 = vcpu->arch.sprg4;
434         regs->sprg6 = vcpu->arch.sprg5;
435         regs->sprg7 = vcpu->arch.sprg6;
436
437         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
438                 regs->gpr[i] = vcpu->arch.gpr[i];
439
440         return 0;
441 }
442
443 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
444 {
445         int i;
446
447         vcpu->arch.pc = regs->pc;
448         vcpu->arch.cr = regs->cr;
449         vcpu->arch.ctr = regs->ctr;
450         vcpu->arch.lr = regs->lr;
451         vcpu->arch.xer = regs->xer;
452         kvmppc_set_msr(vcpu, regs->msr);
453         vcpu->arch.srr0 = regs->srr0;
454         vcpu->arch.srr1 = regs->srr1;
455         vcpu->arch.sprg0 = regs->sprg0;
456         vcpu->arch.sprg1 = regs->sprg1;
457         vcpu->arch.sprg2 = regs->sprg2;
458         vcpu->arch.sprg3 = regs->sprg3;
459         vcpu->arch.sprg5 = regs->sprg4;
460         vcpu->arch.sprg6 = regs->sprg5;
461         vcpu->arch.sprg7 = regs->sprg6;
462
463         for (i = 0; i < ARRAY_SIZE(vcpu->arch.gpr); i++)
464                 vcpu->arch.gpr[i] = regs->gpr[i];
465
466         return 0;
467 }
468
469 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
470                                   struct kvm_sregs *sregs)
471 {
472         return -ENOTSUPP;
473 }
474
475 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
476                                   struct kvm_sregs *sregs)
477 {
478         return -ENOTSUPP;
479 }
480
481 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
482 {
483         return -ENOTSUPP;
484 }
485
486 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
487 {
488         return -ENOTSUPP;
489 }
490
491 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
492                                   struct kvm_translation *tr)
493 {
494         return kvmppc_core_vcpu_translate(vcpu, tr);
495 }
496
497 int kvmppc_booke_init(void)
498 {
499         unsigned long ivor[16];
500         unsigned long max_ivor = 0;
501         int i;
502
503         /* We install our own exception handlers by hijacking IVPR. IVPR must
504          * be 16-bit aligned, so we need a 64KB allocation. */
505         kvmppc_booke_handlers = __get_free_pages(GFP_KERNEL | __GFP_ZERO,
506                                                  VCPU_SIZE_ORDER);
507         if (!kvmppc_booke_handlers)
508                 return -ENOMEM;
509
510         /* XXX make sure our handlers are smaller than Linux's */
511
512         /* Copy our interrupt handlers to match host IVORs. That way we don't
513          * have to swap the IVORs on every guest/host transition. */
514         ivor[0] = mfspr(SPRN_IVOR0);
515         ivor[1] = mfspr(SPRN_IVOR1);
516         ivor[2] = mfspr(SPRN_IVOR2);
517         ivor[3] = mfspr(SPRN_IVOR3);
518         ivor[4] = mfspr(SPRN_IVOR4);
519         ivor[5] = mfspr(SPRN_IVOR5);
520         ivor[6] = mfspr(SPRN_IVOR6);
521         ivor[7] = mfspr(SPRN_IVOR7);
522         ivor[8] = mfspr(SPRN_IVOR8);
523         ivor[9] = mfspr(SPRN_IVOR9);
524         ivor[10] = mfspr(SPRN_IVOR10);
525         ivor[11] = mfspr(SPRN_IVOR11);
526         ivor[12] = mfspr(SPRN_IVOR12);
527         ivor[13] = mfspr(SPRN_IVOR13);
528         ivor[14] = mfspr(SPRN_IVOR14);
529         ivor[15] = mfspr(SPRN_IVOR15);
530
531         for (i = 0; i < 16; i++) {
532                 if (ivor[i] > max_ivor)
533                         max_ivor = ivor[i];
534
535                 memcpy((void *)kvmppc_booke_handlers + ivor[i],
536                        kvmppc_handlers_start + i * kvmppc_handler_len,
537                        kvmppc_handler_len);
538         }
539         flush_icache_range(kvmppc_booke_handlers,
540                            kvmppc_booke_handlers + max_ivor + kvmppc_handler_len);
541
542         return 0;
543 }
544
545 void __exit kvmppc_booke_exit(void)
546 {
547         free_pages(kvmppc_booke_handlers, VCPU_SIZE_ORDER);
548         kvm_exit();
549 }