[S390] pgtables: Fix race in enable_sie vs. page table ops
[linux-2.6.git] / arch / powerpc / kvm / powerpc.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/tlbflush.h>
31
32
33 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
34 {
35         return gfn;
36 }
37
38 int kvm_cpu_has_interrupt(struct kvm_vcpu *v)
39 {
40         return !!(v->arch.pending_exceptions);
41 }
42
43 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
44 {
45         return !(v->arch.msr & MSR_WE);
46 }
47
48
49 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
50 {
51         enum emulation_result er;
52         int r;
53
54         er = kvmppc_emulate_instruction(run, vcpu);
55         switch (er) {
56         case EMULATE_DONE:
57                 /* Future optimization: only reload non-volatiles if they were
58                  * actually modified. */
59                 r = RESUME_GUEST_NV;
60                 break;
61         case EMULATE_DO_MMIO:
62                 run->exit_reason = KVM_EXIT_MMIO;
63                 /* We must reload nonvolatiles because "update" load/store
64                  * instructions modify register state. */
65                 /* Future optimization: only reload non-volatiles if they were
66                  * actually modified. */
67                 r = RESUME_HOST_NV;
68                 break;
69         case EMULATE_FAIL:
70                 /* XXX Deliver Program interrupt to guest. */
71                 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
72                        vcpu->arch.last_inst);
73                 r = RESUME_HOST;
74                 break;
75         default:
76                 BUG();
77         }
78
79         return r;
80 }
81
82 void kvm_arch_hardware_enable(void *garbage)
83 {
84 }
85
86 void kvm_arch_hardware_disable(void *garbage)
87 {
88 }
89
90 int kvm_arch_hardware_setup(void)
91 {
92         return 0;
93 }
94
95 void kvm_arch_hardware_unsetup(void)
96 {
97 }
98
99 void kvm_arch_check_processor_compat(void *rtn)
100 {
101         int r;
102
103         if (strcmp(cur_cpu_spec->platform, "ppc440") == 0)
104                 r = 0;
105         else
106                 r = -ENOTSUPP;
107
108         *(int *)rtn = r;
109 }
110
111 struct kvm *kvm_arch_create_vm(void)
112 {
113         struct kvm *kvm;
114
115         kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
116         if (!kvm)
117                 return ERR_PTR(-ENOMEM);
118
119         return kvm;
120 }
121
122 static void kvmppc_free_vcpus(struct kvm *kvm)
123 {
124         unsigned int i;
125
126         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
127                 if (kvm->vcpus[i]) {
128                         kvm_arch_vcpu_free(kvm->vcpus[i]);
129                         kvm->vcpus[i] = NULL;
130                 }
131         }
132 }
133
134 void kvm_arch_destroy_vm(struct kvm *kvm)
135 {
136         kvmppc_free_vcpus(kvm);
137         kvm_free_physmem(kvm);
138         kfree(kvm);
139 }
140
141 int kvm_dev_ioctl_check_extension(long ext)
142 {
143         int r;
144
145         switch (ext) {
146         case KVM_CAP_USER_MEMORY:
147                 r = 1;
148                 break;
149         case KVM_CAP_COALESCED_MMIO:
150                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
151                 break;
152         default:
153                 r = 0;
154                 break;
155         }
156         return r;
157
158 }
159
160 long kvm_arch_dev_ioctl(struct file *filp,
161                         unsigned int ioctl, unsigned long arg)
162 {
163         return -EINVAL;
164 }
165
166 int kvm_arch_set_memory_region(struct kvm *kvm,
167                                struct kvm_userspace_memory_region *mem,
168                                struct kvm_memory_slot old,
169                                int user_alloc)
170 {
171         return 0;
172 }
173
174 void kvm_arch_flush_shadow(struct kvm *kvm)
175 {
176 }
177
178 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
179 {
180         struct kvm_vcpu *vcpu;
181         int err;
182
183         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
184         if (!vcpu) {
185                 err = -ENOMEM;
186                 goto out;
187         }
188
189         err = kvm_vcpu_init(vcpu, kvm, id);
190         if (err)
191                 goto free_vcpu;
192
193         return vcpu;
194
195 free_vcpu:
196         kmem_cache_free(kvm_vcpu_cache, vcpu);
197 out:
198         return ERR_PTR(err);
199 }
200
201 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
202 {
203         kvm_vcpu_uninit(vcpu);
204         kmem_cache_free(kvm_vcpu_cache, vcpu);
205 }
206
207 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
208 {
209         kvm_arch_vcpu_free(vcpu);
210 }
211
212 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
213 {
214         unsigned int priority = exception_priority[BOOKE_INTERRUPT_DECREMENTER];
215
216         return test_bit(priority, &vcpu->arch.pending_exceptions);
217 }
218
219 static void kvmppc_decrementer_func(unsigned long data)
220 {
221         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
222
223         kvmppc_queue_exception(vcpu, BOOKE_INTERRUPT_DECREMENTER);
224
225         if (waitqueue_active(&vcpu->wq)) {
226                 wake_up_interruptible(&vcpu->wq);
227                 vcpu->stat.halt_wakeup++;
228         }
229 }
230
231 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
232 {
233         setup_timer(&vcpu->arch.dec_timer, kvmppc_decrementer_func,
234                     (unsigned long)vcpu);
235
236         return 0;
237 }
238
239 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
240 {
241 }
242
243 /* Note: clearing MSR[DE] just means that the debug interrupt will not be
244  * delivered *immediately*. Instead, it simply sets the appropriate DBSR bits.
245  * If those DBSR bits are still set when MSR[DE] is re-enabled, the interrupt
246  * will be delivered as an "imprecise debug event" (which is indicated by
247  * DBSR[IDE].
248  */
249 static void kvmppc_disable_debug_interrupts(void)
250 {
251         mtmsr(mfmsr() & ~MSR_DE);
252 }
253
254 static void kvmppc_restore_host_debug_state(struct kvm_vcpu *vcpu)
255 {
256         kvmppc_disable_debug_interrupts();
257
258         mtspr(SPRN_IAC1, vcpu->arch.host_iac[0]);
259         mtspr(SPRN_IAC2, vcpu->arch.host_iac[1]);
260         mtspr(SPRN_IAC3, vcpu->arch.host_iac[2]);
261         mtspr(SPRN_IAC4, vcpu->arch.host_iac[3]);
262         mtspr(SPRN_DBCR1, vcpu->arch.host_dbcr1);
263         mtspr(SPRN_DBCR2, vcpu->arch.host_dbcr2);
264         mtspr(SPRN_DBCR0, vcpu->arch.host_dbcr0);
265         mtmsr(vcpu->arch.host_msr);
266 }
267
268 static void kvmppc_load_guest_debug_registers(struct kvm_vcpu *vcpu)
269 {
270         struct kvm_guest_debug *dbg = &vcpu->guest_debug;
271         u32 dbcr0 = 0;
272
273         vcpu->arch.host_msr = mfmsr();
274         kvmppc_disable_debug_interrupts();
275
276         /* Save host debug register state. */
277         vcpu->arch.host_iac[0] = mfspr(SPRN_IAC1);
278         vcpu->arch.host_iac[1] = mfspr(SPRN_IAC2);
279         vcpu->arch.host_iac[2] = mfspr(SPRN_IAC3);
280         vcpu->arch.host_iac[3] = mfspr(SPRN_IAC4);
281         vcpu->arch.host_dbcr0 = mfspr(SPRN_DBCR0);
282         vcpu->arch.host_dbcr1 = mfspr(SPRN_DBCR1);
283         vcpu->arch.host_dbcr2 = mfspr(SPRN_DBCR2);
284
285         /* set registers up for guest */
286
287         if (dbg->bp[0]) {
288                 mtspr(SPRN_IAC1, dbg->bp[0]);
289                 dbcr0 |= DBCR0_IAC1 | DBCR0_IDM;
290         }
291         if (dbg->bp[1]) {
292                 mtspr(SPRN_IAC2, dbg->bp[1]);
293                 dbcr0 |= DBCR0_IAC2 | DBCR0_IDM;
294         }
295         if (dbg->bp[2]) {
296                 mtspr(SPRN_IAC3, dbg->bp[2]);
297                 dbcr0 |= DBCR0_IAC3 | DBCR0_IDM;
298         }
299         if (dbg->bp[3]) {
300                 mtspr(SPRN_IAC4, dbg->bp[3]);
301                 dbcr0 |= DBCR0_IAC4 | DBCR0_IDM;
302         }
303
304         mtspr(SPRN_DBCR0, dbcr0);
305         mtspr(SPRN_DBCR1, 0);
306         mtspr(SPRN_DBCR2, 0);
307 }
308
309 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
310 {
311         int i;
312
313         if (vcpu->guest_debug.enabled)
314                 kvmppc_load_guest_debug_registers(vcpu);
315
316         /* Mark every guest entry in the shadow TLB entry modified, so that they
317          * will all be reloaded on the next vcpu run (instead of being
318          * demand-faulted). */
319         for (i = 0; i <= tlb_44x_hwater; i++)
320                 kvmppc_tlbe_set_modified(vcpu, i);
321 }
322
323 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
324 {
325         if (vcpu->guest_debug.enabled)
326                 kvmppc_restore_host_debug_state(vcpu);
327
328         /* Don't leave guest TLB entries resident when being de-scheduled. */
329         /* XXX It would be nice to differentiate between heavyweight exit and
330          * sched_out here, since we could avoid the TLB flush for heavyweight
331          * exits. */
332         _tlbia();
333 }
334
335 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
336                                     struct kvm_debug_guest *dbg)
337 {
338         int i;
339
340         vcpu->guest_debug.enabled = dbg->enabled;
341         if (vcpu->guest_debug.enabled) {
342                 for (i=0; i < ARRAY_SIZE(vcpu->guest_debug.bp); i++) {
343                         if (dbg->breakpoints[i].enabled)
344                                 vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
345                         else
346                                 vcpu->guest_debug.bp[i] = 0;
347                 }
348         }
349
350         return 0;
351 }
352
353 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
354                                      struct kvm_run *run)
355 {
356         u32 *gpr = &vcpu->arch.gpr[vcpu->arch.io_gpr];
357         *gpr = run->dcr.data;
358 }
359
360 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
361                                       struct kvm_run *run)
362 {
363         u32 *gpr = &vcpu->arch.gpr[vcpu->arch.io_gpr];
364
365         if (run->mmio.len > sizeof(*gpr)) {
366                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
367                 return;
368         }
369
370         if (vcpu->arch.mmio_is_bigendian) {
371                 switch (run->mmio.len) {
372                 case 4: *gpr = *(u32 *)run->mmio.data; break;
373                 case 2: *gpr = *(u16 *)run->mmio.data; break;
374                 case 1: *gpr = *(u8 *)run->mmio.data; break;
375                 }
376         } else {
377                 /* Convert BE data from userland back to LE. */
378                 switch (run->mmio.len) {
379                 case 4: *gpr = ld_le32((u32 *)run->mmio.data); break;
380                 case 2: *gpr = ld_le16((u16 *)run->mmio.data); break;
381                 case 1: *gpr = *(u8 *)run->mmio.data; break;
382                 }
383         }
384 }
385
386 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
387                        unsigned int rt, unsigned int bytes, int is_bigendian)
388 {
389         if (bytes > sizeof(run->mmio.data)) {
390                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
391                        run->mmio.len);
392         }
393
394         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
395         run->mmio.len = bytes;
396         run->mmio.is_write = 0;
397
398         vcpu->arch.io_gpr = rt;
399         vcpu->arch.mmio_is_bigendian = is_bigendian;
400         vcpu->mmio_needed = 1;
401         vcpu->mmio_is_write = 0;
402
403         return EMULATE_DO_MMIO;
404 }
405
406 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
407                         u32 val, unsigned int bytes, int is_bigendian)
408 {
409         void *data = run->mmio.data;
410
411         if (bytes > sizeof(run->mmio.data)) {
412                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
413                        run->mmio.len);
414         }
415
416         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
417         run->mmio.len = bytes;
418         run->mmio.is_write = 1;
419         vcpu->mmio_needed = 1;
420         vcpu->mmio_is_write = 1;
421
422         /* Store the value at the lowest bytes in 'data'. */
423         if (is_bigendian) {
424                 switch (bytes) {
425                 case 4: *(u32 *)data = val; break;
426                 case 2: *(u16 *)data = val; break;
427                 case 1: *(u8  *)data = val; break;
428                 }
429         } else {
430                 /* Store LE value into 'data'. */
431                 switch (bytes) {
432                 case 4: st_le32(data, val); break;
433                 case 2: st_le16(data, val); break;
434                 case 1: *(u8 *)data = val; break;
435                 }
436         }
437
438         return EMULATE_DO_MMIO;
439 }
440
441 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
442 {
443         int r;
444         sigset_t sigsaved;
445
446         vcpu_load(vcpu);
447
448         if (vcpu->sigset_active)
449                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
450
451         if (vcpu->mmio_needed) {
452                 if (!vcpu->mmio_is_write)
453                         kvmppc_complete_mmio_load(vcpu, run);
454                 vcpu->mmio_needed = 0;
455         } else if (vcpu->arch.dcr_needed) {
456                 if (!vcpu->arch.dcr_is_write)
457                         kvmppc_complete_dcr_load(vcpu, run);
458                 vcpu->arch.dcr_needed = 0;
459         }
460
461         kvmppc_check_and_deliver_interrupts(vcpu);
462
463         local_irq_disable();
464         kvm_guest_enter();
465         r = __kvmppc_vcpu_run(run, vcpu);
466         kvm_guest_exit();
467         local_irq_enable();
468
469         if (vcpu->sigset_active)
470                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
471
472         vcpu_put(vcpu);
473
474         return r;
475 }
476
477 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
478 {
479         kvmppc_queue_exception(vcpu, BOOKE_INTERRUPT_EXTERNAL);
480
481         if (waitqueue_active(&vcpu->wq)) {
482                 wake_up_interruptible(&vcpu->wq);
483                 vcpu->stat.halt_wakeup++;
484         }
485
486         return 0;
487 }
488
489 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
490                                     struct kvm_mp_state *mp_state)
491 {
492         return -EINVAL;
493 }
494
495 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
496                                     struct kvm_mp_state *mp_state)
497 {
498         return -EINVAL;
499 }
500
501 long kvm_arch_vcpu_ioctl(struct file *filp,
502                          unsigned int ioctl, unsigned long arg)
503 {
504         struct kvm_vcpu *vcpu = filp->private_data;
505         void __user *argp = (void __user *)arg;
506         long r;
507
508         switch (ioctl) {
509         case KVM_INTERRUPT: {
510                 struct kvm_interrupt irq;
511                 r = -EFAULT;
512                 if (copy_from_user(&irq, argp, sizeof(irq)))
513                         goto out;
514                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
515                 break;
516         }
517         default:
518                 r = -EINVAL;
519         }
520
521 out:
522         return r;
523 }
524
525 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
526 {
527         return -ENOTSUPP;
528 }
529
530 long kvm_arch_vm_ioctl(struct file *filp,
531                        unsigned int ioctl, unsigned long arg)
532 {
533         long r;
534
535         switch (ioctl) {
536         default:
537                 r = -EINVAL;
538         }
539
540         return r;
541 }
542
543 int kvm_arch_init(void *opaque)
544 {
545         return 0;
546 }
547
548 void kvm_arch_exit(void)
549 {
550 }