]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * hosting zSeries kernel virtual machines | |
3 | * | |
4 | * Copyright IBM Corp. 2008, 2009 | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License (version 2 only) | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * Author(s): Carsten Otte <cotte@de.ibm.com> | |
11 | * Christian Borntraeger <borntraeger@de.ibm.com> | |
12 | * Heiko Carstens <heiko.carstens@de.ibm.com> | |
13 | * Christian Ehrhardt <ehrhardt@de.ibm.com> | |
14 | * Jason J. Herne <jjherne@us.ibm.com> | |
15 | */ | |
16 | ||
17 | #include <linux/compiler.h> | |
18 | #include <linux/err.h> | |
19 | #include <linux/fs.h> | |
20 | #include <linux/hrtimer.h> | |
21 | #include <linux/init.h> | |
22 | #include <linux/kvm.h> | |
23 | #include <linux/kvm_host.h> | |
24 | #include <linux/module.h> | |
25 | #include <linux/slab.h> | |
26 | #include <linux/timer.h> | |
27 | #include <asm/asm-offsets.h> | |
28 | #include <asm/lowcore.h> | |
29 | #include <asm/pgtable.h> | |
30 | #include <asm/nmi.h> | |
31 | #include <asm/switch_to.h> | |
32 | #include <asm/facility.h> | |
33 | #include <asm/sclp.h> | |
34 | #include "kvm-s390.h" | |
35 | #include "gaccess.h" | |
36 | ||
37 | #define CREATE_TRACE_POINTS | |
38 | #include "trace.h" | |
39 | #include "trace-s390.h" | |
40 | ||
41 | #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU | |
42 | ||
43 | struct kvm_stats_debugfs_item debugfs_entries[] = { | |
44 | { "userspace_handled", VCPU_STAT(exit_userspace) }, | |
45 | { "exit_null", VCPU_STAT(exit_null) }, | |
46 | { "exit_validity", VCPU_STAT(exit_validity) }, | |
47 | { "exit_stop_request", VCPU_STAT(exit_stop_request) }, | |
48 | { "exit_external_request", VCPU_STAT(exit_external_request) }, | |
49 | { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) }, | |
50 | { "exit_instruction", VCPU_STAT(exit_instruction) }, | |
51 | { "exit_program_interruption", VCPU_STAT(exit_program_interruption) }, | |
52 | { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) }, | |
53 | { "instruction_lctlg", VCPU_STAT(instruction_lctlg) }, | |
54 | { "instruction_lctl", VCPU_STAT(instruction_lctl) }, | |
55 | { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) }, | |
56 | { "deliver_external_call", VCPU_STAT(deliver_external_call) }, | |
57 | { "deliver_service_signal", VCPU_STAT(deliver_service_signal) }, | |
58 | { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) }, | |
59 | { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) }, | |
60 | { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) }, | |
61 | { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) }, | |
62 | { "deliver_program_interruption", VCPU_STAT(deliver_program_int) }, | |
63 | { "exit_wait_state", VCPU_STAT(exit_wait_state) }, | |
64 | { "instruction_pfmf", VCPU_STAT(instruction_pfmf) }, | |
65 | { "instruction_stidp", VCPU_STAT(instruction_stidp) }, | |
66 | { "instruction_spx", VCPU_STAT(instruction_spx) }, | |
67 | { "instruction_stpx", VCPU_STAT(instruction_stpx) }, | |
68 | { "instruction_stap", VCPU_STAT(instruction_stap) }, | |
69 | { "instruction_storage_key", VCPU_STAT(instruction_storage_key) }, | |
70 | { "instruction_stsch", VCPU_STAT(instruction_stsch) }, | |
71 | { "instruction_chsc", VCPU_STAT(instruction_chsc) }, | |
72 | { "instruction_essa", VCPU_STAT(instruction_essa) }, | |
73 | { "instruction_stsi", VCPU_STAT(instruction_stsi) }, | |
74 | { "instruction_stfl", VCPU_STAT(instruction_stfl) }, | |
75 | { "instruction_tprot", VCPU_STAT(instruction_tprot) }, | |
76 | { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) }, | |
77 | { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) }, | |
78 | { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) }, | |
79 | { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) }, | |
80 | { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) }, | |
81 | { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) }, | |
82 | { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) }, | |
83 | { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) }, | |
84 | { "diagnose_10", VCPU_STAT(diagnose_10) }, | |
85 | { "diagnose_44", VCPU_STAT(diagnose_44) }, | |
86 | { "diagnose_9c", VCPU_STAT(diagnose_9c) }, | |
87 | { NULL } | |
88 | }; | |
89 | ||
90 | unsigned long *vfacilities; | |
91 | static struct gmap_notifier gmap_notifier; | |
92 | ||
93 | /* test availability of vfacility */ | |
94 | static inline int test_vfacility(unsigned long nr) | |
95 | { | |
96 | return __test_facility(nr, (void *) vfacilities); | |
97 | } | |
98 | ||
99 | /* Section: not file related */ | |
100 | int kvm_arch_hardware_enable(void *garbage) | |
101 | { | |
102 | /* every s390 is virtualization enabled ;-) */ | |
103 | return 0; | |
104 | } | |
105 | ||
106 | void kvm_arch_hardware_disable(void *garbage) | |
107 | { | |
108 | } | |
109 | ||
110 | static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address); | |
111 | ||
112 | int kvm_arch_hardware_setup(void) | |
113 | { | |
114 | gmap_notifier.notifier_call = kvm_gmap_notifier; | |
115 | gmap_register_ipte_notifier(&gmap_notifier); | |
116 | return 0; | |
117 | } | |
118 | ||
119 | void kvm_arch_hardware_unsetup(void) | |
120 | { | |
121 | gmap_unregister_ipte_notifier(&gmap_notifier); | |
122 | } | |
123 | ||
124 | void kvm_arch_check_processor_compat(void *rtn) | |
125 | { | |
126 | } | |
127 | ||
128 | int kvm_arch_init(void *opaque) | |
129 | { | |
130 | return 0; | |
131 | } | |
132 | ||
133 | void kvm_arch_exit(void) | |
134 | { | |
135 | } | |
136 | ||
137 | /* Section: device related */ | |
138 | long kvm_arch_dev_ioctl(struct file *filp, | |
139 | unsigned int ioctl, unsigned long arg) | |
140 | { | |
141 | if (ioctl == KVM_S390_ENABLE_SIE) | |
142 | return s390_enable_sie(); | |
143 | return -EINVAL; | |
144 | } | |
145 | ||
146 | int kvm_dev_ioctl_check_extension(long ext) | |
147 | { | |
148 | int r; | |
149 | ||
150 | switch (ext) { | |
151 | case KVM_CAP_S390_PSW: | |
152 | case KVM_CAP_S390_GMAP: | |
153 | case KVM_CAP_SYNC_MMU: | |
154 | #ifdef CONFIG_KVM_S390_UCONTROL | |
155 | case KVM_CAP_S390_UCONTROL: | |
156 | #endif | |
157 | case KVM_CAP_ASYNC_PF: | |
158 | case KVM_CAP_SYNC_REGS: | |
159 | case KVM_CAP_ONE_REG: | |
160 | case KVM_CAP_ENABLE_CAP: | |
161 | case KVM_CAP_S390_CSS_SUPPORT: | |
162 | case KVM_CAP_IOEVENTFD: | |
163 | case KVM_CAP_DEVICE_CTRL: | |
164 | case KVM_CAP_ENABLE_CAP_VM: | |
165 | r = 1; | |
166 | break; | |
167 | case KVM_CAP_NR_VCPUS: | |
168 | case KVM_CAP_MAX_VCPUS: | |
169 | r = KVM_MAX_VCPUS; | |
170 | break; | |
171 | case KVM_CAP_NR_MEMSLOTS: | |
172 | r = KVM_USER_MEM_SLOTS; | |
173 | break; | |
174 | case KVM_CAP_S390_COW: | |
175 | r = MACHINE_HAS_ESOP; | |
176 | break; | |
177 | default: | |
178 | r = 0; | |
179 | } | |
180 | return r; | |
181 | } | |
182 | ||
183 | static void kvm_s390_sync_dirty_log(struct kvm *kvm, | |
184 | struct kvm_memory_slot *memslot) | |
185 | { | |
186 | gfn_t cur_gfn, last_gfn; | |
187 | unsigned long address; | |
188 | struct gmap *gmap = kvm->arch.gmap; | |
189 | ||
190 | down_read(&gmap->mm->mmap_sem); | |
191 | /* Loop over all guest pages */ | |
192 | last_gfn = memslot->base_gfn + memslot->npages; | |
193 | for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) { | |
194 | address = gfn_to_hva_memslot(memslot, cur_gfn); | |
195 | ||
196 | if (gmap_test_and_clear_dirty(address, gmap)) | |
197 | mark_page_dirty(kvm, cur_gfn); | |
198 | } | |
199 | up_read(&gmap->mm->mmap_sem); | |
200 | } | |
201 | ||
202 | /* Section: vm related */ | |
203 | /* | |
204 | * Get (and clear) the dirty memory log for a memory slot. | |
205 | */ | |
206 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
207 | struct kvm_dirty_log *log) | |
208 | { | |
209 | int r; | |
210 | unsigned long n; | |
211 | struct kvm_memory_slot *memslot; | |
212 | int is_dirty = 0; | |
213 | ||
214 | mutex_lock(&kvm->slots_lock); | |
215 | ||
216 | r = -EINVAL; | |
217 | if (log->slot >= KVM_USER_MEM_SLOTS) | |
218 | goto out; | |
219 | ||
220 | memslot = id_to_memslot(kvm->memslots, log->slot); | |
221 | r = -ENOENT; | |
222 | if (!memslot->dirty_bitmap) | |
223 | goto out; | |
224 | ||
225 | kvm_s390_sync_dirty_log(kvm, memslot); | |
226 | r = kvm_get_dirty_log(kvm, log, &is_dirty); | |
227 | if (r) | |
228 | goto out; | |
229 | ||
230 | /* Clear the dirty log */ | |
231 | if (is_dirty) { | |
232 | n = kvm_dirty_bitmap_bytes(memslot); | |
233 | memset(memslot->dirty_bitmap, 0, n); | |
234 | } | |
235 | r = 0; | |
236 | out: | |
237 | mutex_unlock(&kvm->slots_lock); | |
238 | return r; | |
239 | } | |
240 | ||
241 | static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap) | |
242 | { | |
243 | int r; | |
244 | ||
245 | if (cap->flags) | |
246 | return -EINVAL; | |
247 | ||
248 | switch (cap->cap) { | |
249 | case KVM_CAP_S390_IRQCHIP: | |
250 | kvm->arch.use_irqchip = 1; | |
251 | r = 0; | |
252 | break; | |
253 | default: | |
254 | r = -EINVAL; | |
255 | break; | |
256 | } | |
257 | return r; | |
258 | } | |
259 | ||
260 | long kvm_arch_vm_ioctl(struct file *filp, | |
261 | unsigned int ioctl, unsigned long arg) | |
262 | { | |
263 | struct kvm *kvm = filp->private_data; | |
264 | void __user *argp = (void __user *)arg; | |
265 | int r; | |
266 | ||
267 | switch (ioctl) { | |
268 | case KVM_S390_INTERRUPT: { | |
269 | struct kvm_s390_interrupt s390int; | |
270 | ||
271 | r = -EFAULT; | |
272 | if (copy_from_user(&s390int, argp, sizeof(s390int))) | |
273 | break; | |
274 | r = kvm_s390_inject_vm(kvm, &s390int); | |
275 | break; | |
276 | } | |
277 | case KVM_ENABLE_CAP: { | |
278 | struct kvm_enable_cap cap; | |
279 | r = -EFAULT; | |
280 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
281 | break; | |
282 | r = kvm_vm_ioctl_enable_cap(kvm, &cap); | |
283 | break; | |
284 | } | |
285 | case KVM_CREATE_IRQCHIP: { | |
286 | struct kvm_irq_routing_entry routing; | |
287 | ||
288 | r = -EINVAL; | |
289 | if (kvm->arch.use_irqchip) { | |
290 | /* Set up dummy routing. */ | |
291 | memset(&routing, 0, sizeof(routing)); | |
292 | kvm_set_irq_routing(kvm, &routing, 0, 0); | |
293 | r = 0; | |
294 | } | |
295 | break; | |
296 | } | |
297 | default: | |
298 | r = -ENOTTY; | |
299 | } | |
300 | ||
301 | return r; | |
302 | } | |
303 | ||
304 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) | |
305 | { | |
306 | int rc; | |
307 | char debug_name[16]; | |
308 | static unsigned long sca_offset; | |
309 | ||
310 | rc = -EINVAL; | |
311 | #ifdef CONFIG_KVM_S390_UCONTROL | |
312 | if (type & ~KVM_VM_S390_UCONTROL) | |
313 | goto out_err; | |
314 | if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN))) | |
315 | goto out_err; | |
316 | #else | |
317 | if (type) | |
318 | goto out_err; | |
319 | #endif | |
320 | ||
321 | rc = s390_enable_sie(); | |
322 | if (rc) | |
323 | goto out_err; | |
324 | ||
325 | rc = -ENOMEM; | |
326 | ||
327 | kvm->arch.sca = (struct sca_block *) get_zeroed_page(GFP_KERNEL); | |
328 | if (!kvm->arch.sca) | |
329 | goto out_err; | |
330 | spin_lock(&kvm_lock); | |
331 | sca_offset = (sca_offset + 16) & 0x7f0; | |
332 | kvm->arch.sca = (struct sca_block *) ((char *) kvm->arch.sca + sca_offset); | |
333 | spin_unlock(&kvm_lock); | |
334 | ||
335 | sprintf(debug_name, "kvm-%u", current->pid); | |
336 | ||
337 | kvm->arch.dbf = debug_register(debug_name, 8, 2, 8 * sizeof(long)); | |
338 | if (!kvm->arch.dbf) | |
339 | goto out_nodbf; | |
340 | ||
341 | spin_lock_init(&kvm->arch.float_int.lock); | |
342 | INIT_LIST_HEAD(&kvm->arch.float_int.list); | |
343 | ||
344 | debug_register_view(kvm->arch.dbf, &debug_sprintf_view); | |
345 | VM_EVENT(kvm, 3, "%s", "vm created"); | |
346 | ||
347 | if (type & KVM_VM_S390_UCONTROL) { | |
348 | kvm->arch.gmap = NULL; | |
349 | } else { | |
350 | kvm->arch.gmap = gmap_alloc(current->mm); | |
351 | if (!kvm->arch.gmap) | |
352 | goto out_nogmap; | |
353 | kvm->arch.gmap->private = kvm; | |
354 | kvm->arch.gmap->pfault_enabled = 0; | |
355 | } | |
356 | ||
357 | kvm->arch.css_support = 0; | |
358 | kvm->arch.use_irqchip = 0; | |
359 | ||
360 | return 0; | |
361 | out_nogmap: | |
362 | debug_unregister(kvm->arch.dbf); | |
363 | out_nodbf: | |
364 | free_page((unsigned long)(kvm->arch.sca)); | |
365 | out_err: | |
366 | return rc; | |
367 | } | |
368 | ||
369 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
370 | { | |
371 | VCPU_EVENT(vcpu, 3, "%s", "free cpu"); | |
372 | trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id); | |
373 | kvm_clear_async_pf_completion_queue(vcpu); | |
374 | if (!kvm_is_ucontrol(vcpu->kvm)) { | |
375 | clear_bit(63 - vcpu->vcpu_id, | |
376 | (unsigned long *) &vcpu->kvm->arch.sca->mcn); | |
377 | if (vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda == | |
378 | (__u64) vcpu->arch.sie_block) | |
379 | vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda = 0; | |
380 | } | |
381 | smp_mb(); | |
382 | ||
383 | if (kvm_is_ucontrol(vcpu->kvm)) | |
384 | gmap_free(vcpu->arch.gmap); | |
385 | ||
386 | if (vcpu->arch.sie_block->cbrlo) | |
387 | __free_page(__pfn_to_page( | |
388 | vcpu->arch.sie_block->cbrlo >> PAGE_SHIFT)); | |
389 | free_page((unsigned long)(vcpu->arch.sie_block)); | |
390 | ||
391 | kvm_vcpu_uninit(vcpu); | |
392 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
393 | } | |
394 | ||
395 | static void kvm_free_vcpus(struct kvm *kvm) | |
396 | { | |
397 | unsigned int i; | |
398 | struct kvm_vcpu *vcpu; | |
399 | ||
400 | kvm_for_each_vcpu(i, vcpu, kvm) | |
401 | kvm_arch_vcpu_destroy(vcpu); | |
402 | ||
403 | mutex_lock(&kvm->lock); | |
404 | for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) | |
405 | kvm->vcpus[i] = NULL; | |
406 | ||
407 | atomic_set(&kvm->online_vcpus, 0); | |
408 | mutex_unlock(&kvm->lock); | |
409 | } | |
410 | ||
411 | void kvm_arch_sync_events(struct kvm *kvm) | |
412 | { | |
413 | } | |
414 | ||
415 | void kvm_arch_destroy_vm(struct kvm *kvm) | |
416 | { | |
417 | kvm_free_vcpus(kvm); | |
418 | free_page((unsigned long)(kvm->arch.sca)); | |
419 | debug_unregister(kvm->arch.dbf); | |
420 | if (!kvm_is_ucontrol(kvm)) | |
421 | gmap_free(kvm->arch.gmap); | |
422 | kvm_s390_destroy_adapters(kvm); | |
423 | } | |
424 | ||
425 | /* Section: vcpu related */ | |
426 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) | |
427 | { | |
428 | vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID; | |
429 | kvm_clear_async_pf_completion_queue(vcpu); | |
430 | if (kvm_is_ucontrol(vcpu->kvm)) { | |
431 | vcpu->arch.gmap = gmap_alloc(current->mm); | |
432 | if (!vcpu->arch.gmap) | |
433 | return -ENOMEM; | |
434 | vcpu->arch.gmap->private = vcpu->kvm; | |
435 | return 0; | |
436 | } | |
437 | ||
438 | vcpu->arch.gmap = vcpu->kvm->arch.gmap; | |
439 | vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX | | |
440 | KVM_SYNC_GPRS | | |
441 | KVM_SYNC_ACRS | | |
442 | KVM_SYNC_CRS; | |
443 | return 0; | |
444 | } | |
445 | ||
446 | void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) | |
447 | { | |
448 | /* Nothing todo */ | |
449 | } | |
450 | ||
451 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) | |
452 | { | |
453 | save_fp_ctl(&vcpu->arch.host_fpregs.fpc); | |
454 | save_fp_regs(vcpu->arch.host_fpregs.fprs); | |
455 | save_access_regs(vcpu->arch.host_acrs); | |
456 | restore_fp_ctl(&vcpu->arch.guest_fpregs.fpc); | |
457 | restore_fp_regs(vcpu->arch.guest_fpregs.fprs); | |
458 | restore_access_regs(vcpu->run->s.regs.acrs); | |
459 | gmap_enable(vcpu->arch.gmap); | |
460 | atomic_set_mask(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags); | |
461 | } | |
462 | ||
463 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
464 | { | |
465 | atomic_clear_mask(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags); | |
466 | gmap_disable(vcpu->arch.gmap); | |
467 | save_fp_ctl(&vcpu->arch.guest_fpregs.fpc); | |
468 | save_fp_regs(vcpu->arch.guest_fpregs.fprs); | |
469 | save_access_regs(vcpu->run->s.regs.acrs); | |
470 | restore_fp_ctl(&vcpu->arch.host_fpregs.fpc); | |
471 | restore_fp_regs(vcpu->arch.host_fpregs.fprs); | |
472 | restore_access_regs(vcpu->arch.host_acrs); | |
473 | } | |
474 | ||
475 | static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu) | |
476 | { | |
477 | /* this equals initial cpu reset in pop, but we don't switch to ESA */ | |
478 | vcpu->arch.sie_block->gpsw.mask = 0UL; | |
479 | vcpu->arch.sie_block->gpsw.addr = 0UL; | |
480 | kvm_s390_set_prefix(vcpu, 0); | |
481 | vcpu->arch.sie_block->cputm = 0UL; | |
482 | vcpu->arch.sie_block->ckc = 0UL; | |
483 | vcpu->arch.sie_block->todpr = 0; | |
484 | memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64)); | |
485 | vcpu->arch.sie_block->gcr[0] = 0xE0UL; | |
486 | vcpu->arch.sie_block->gcr[14] = 0xC2000000UL; | |
487 | vcpu->arch.guest_fpregs.fpc = 0; | |
488 | asm volatile("lfpc %0" : : "Q" (vcpu->arch.guest_fpregs.fpc)); | |
489 | vcpu->arch.sie_block->gbea = 1; | |
490 | vcpu->arch.sie_block->pp = 0; | |
491 | vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID; | |
492 | kvm_clear_async_pf_completion_queue(vcpu); | |
493 | atomic_set_mask(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags); | |
494 | kvm_s390_clear_local_irqs(vcpu); | |
495 | } | |
496 | ||
497 | int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) | |
498 | { | |
499 | return 0; | |
500 | } | |
501 | ||
502 | int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu) | |
503 | { | |
504 | struct page *cbrl; | |
505 | ||
506 | atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH | | |
507 | CPUSTAT_SM | | |
508 | CPUSTAT_STOPPED | | |
509 | CPUSTAT_GED); | |
510 | vcpu->arch.sie_block->ecb = 6; | |
511 | if (test_vfacility(50) && test_vfacility(73)) | |
512 | vcpu->arch.sie_block->ecb |= 0x10; | |
513 | ||
514 | vcpu->arch.sie_block->ecb2 = 8; | |
515 | vcpu->arch.sie_block->eca = 0xC1002001U; | |
516 | vcpu->arch.sie_block->fac = (int) (long) vfacilities; | |
517 | vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE; | |
518 | if (kvm_enabled_cmma()) { | |
519 | cbrl = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
520 | if (cbrl) { | |
521 | vcpu->arch.sie_block->ecb2 |= 0x80; | |
522 | vcpu->arch.sie_block->ecb2 &= ~0x08; | |
523 | vcpu->arch.sie_block->cbrlo = page_to_phys(cbrl); | |
524 | } | |
525 | } | |
526 | hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS); | |
527 | tasklet_init(&vcpu->arch.tasklet, kvm_s390_tasklet, | |
528 | (unsigned long) vcpu); | |
529 | vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup; | |
530 | get_cpu_id(&vcpu->arch.cpu_id); | |
531 | vcpu->arch.cpu_id.version = 0xff; | |
532 | return 0; | |
533 | } | |
534 | ||
535 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, | |
536 | unsigned int id) | |
537 | { | |
538 | struct kvm_vcpu *vcpu; | |
539 | struct sie_page *sie_page; | |
540 | int rc = -EINVAL; | |
541 | ||
542 | if (id >= KVM_MAX_VCPUS) | |
543 | goto out; | |
544 | ||
545 | rc = -ENOMEM; | |
546 | ||
547 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); | |
548 | if (!vcpu) | |
549 | goto out; | |
550 | ||
551 | sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL); | |
552 | if (!sie_page) | |
553 | goto out_free_cpu; | |
554 | ||
555 | vcpu->arch.sie_block = &sie_page->sie_block; | |
556 | vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb; | |
557 | ||
558 | vcpu->arch.sie_block->icpua = id; | |
559 | if (!kvm_is_ucontrol(kvm)) { | |
560 | if (!kvm->arch.sca) { | |
561 | WARN_ON_ONCE(1); | |
562 | goto out_free_cpu; | |
563 | } | |
564 | if (!kvm->arch.sca->cpu[id].sda) | |
565 | kvm->arch.sca->cpu[id].sda = | |
566 | (__u64) vcpu->arch.sie_block; | |
567 | vcpu->arch.sie_block->scaoh = | |
568 | (__u32)(((__u64)kvm->arch.sca) >> 32); | |
569 | vcpu->arch.sie_block->scaol = (__u32)(__u64)kvm->arch.sca; | |
570 | set_bit(63 - id, (unsigned long *) &kvm->arch.sca->mcn); | |
571 | } | |
572 | ||
573 | spin_lock_init(&vcpu->arch.local_int.lock); | |
574 | INIT_LIST_HEAD(&vcpu->arch.local_int.list); | |
575 | vcpu->arch.local_int.float_int = &kvm->arch.float_int; | |
576 | vcpu->arch.local_int.wq = &vcpu->wq; | |
577 | vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags; | |
578 | ||
579 | rc = kvm_vcpu_init(vcpu, kvm, id); | |
580 | if (rc) | |
581 | goto out_free_sie_block; | |
582 | VM_EVENT(kvm, 3, "create cpu %d at %p, sie block at %p", id, vcpu, | |
583 | vcpu->arch.sie_block); | |
584 | trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block); | |
585 | ||
586 | return vcpu; | |
587 | out_free_sie_block: | |
588 | free_page((unsigned long)(vcpu->arch.sie_block)); | |
589 | out_free_cpu: | |
590 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
591 | out: | |
592 | return ERR_PTR(rc); | |
593 | } | |
594 | ||
595 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) | |
596 | { | |
597 | return kvm_cpu_has_interrupt(vcpu); | |
598 | } | |
599 | ||
600 | void s390_vcpu_block(struct kvm_vcpu *vcpu) | |
601 | { | |
602 | atomic_set_mask(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20); | |
603 | } | |
604 | ||
605 | void s390_vcpu_unblock(struct kvm_vcpu *vcpu) | |
606 | { | |
607 | atomic_clear_mask(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20); | |
608 | } | |
609 | ||
610 | /* | |
611 | * Kick a guest cpu out of SIE and wait until SIE is not running. | |
612 | * If the CPU is not running (e.g. waiting as idle) the function will | |
613 | * return immediately. */ | |
614 | void exit_sie(struct kvm_vcpu *vcpu) | |
615 | { | |
616 | atomic_set_mask(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags); | |
617 | while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE) | |
618 | cpu_relax(); | |
619 | } | |
620 | ||
621 | /* Kick a guest cpu out of SIE and prevent SIE-reentry */ | |
622 | void exit_sie_sync(struct kvm_vcpu *vcpu) | |
623 | { | |
624 | s390_vcpu_block(vcpu); | |
625 | exit_sie(vcpu); | |
626 | } | |
627 | ||
628 | static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address) | |
629 | { | |
630 | int i; | |
631 | struct kvm *kvm = gmap->private; | |
632 | struct kvm_vcpu *vcpu; | |
633 | ||
634 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
635 | /* match against both prefix pages */ | |
636 | if (vcpu->arch.sie_block->prefix == (address & ~0x1000UL)) { | |
637 | VCPU_EVENT(vcpu, 2, "gmap notifier for %lx", address); | |
638 | kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu); | |
639 | exit_sie_sync(vcpu); | |
640 | } | |
641 | } | |
642 | } | |
643 | ||
644 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) | |
645 | { | |
646 | /* kvm common code refers to this, but never calls it */ | |
647 | BUG(); | |
648 | return 0; | |
649 | } | |
650 | ||
651 | static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, | |
652 | struct kvm_one_reg *reg) | |
653 | { | |
654 | int r = -EINVAL; | |
655 | ||
656 | switch (reg->id) { | |
657 | case KVM_REG_S390_TODPR: | |
658 | r = put_user(vcpu->arch.sie_block->todpr, | |
659 | (u32 __user *)reg->addr); | |
660 | break; | |
661 | case KVM_REG_S390_EPOCHDIFF: | |
662 | r = put_user(vcpu->arch.sie_block->epoch, | |
663 | (u64 __user *)reg->addr); | |
664 | break; | |
665 | case KVM_REG_S390_CPU_TIMER: | |
666 | r = put_user(vcpu->arch.sie_block->cputm, | |
667 | (u64 __user *)reg->addr); | |
668 | break; | |
669 | case KVM_REG_S390_CLOCK_COMP: | |
670 | r = put_user(vcpu->arch.sie_block->ckc, | |
671 | (u64 __user *)reg->addr); | |
672 | break; | |
673 | case KVM_REG_S390_PFTOKEN: | |
674 | r = put_user(vcpu->arch.pfault_token, | |
675 | (u64 __user *)reg->addr); | |
676 | break; | |
677 | case KVM_REG_S390_PFCOMPARE: | |
678 | r = put_user(vcpu->arch.pfault_compare, | |
679 | (u64 __user *)reg->addr); | |
680 | break; | |
681 | case KVM_REG_S390_PFSELECT: | |
682 | r = put_user(vcpu->arch.pfault_select, | |
683 | (u64 __user *)reg->addr); | |
684 | break; | |
685 | case KVM_REG_S390_PP: | |
686 | r = put_user(vcpu->arch.sie_block->pp, | |
687 | (u64 __user *)reg->addr); | |
688 | break; | |
689 | case KVM_REG_S390_GBEA: | |
690 | r = put_user(vcpu->arch.sie_block->gbea, | |
691 | (u64 __user *)reg->addr); | |
692 | break; | |
693 | default: | |
694 | break; | |
695 | } | |
696 | ||
697 | return r; | |
698 | } | |
699 | ||
700 | static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, | |
701 | struct kvm_one_reg *reg) | |
702 | { | |
703 | int r = -EINVAL; | |
704 | ||
705 | switch (reg->id) { | |
706 | case KVM_REG_S390_TODPR: | |
707 | r = get_user(vcpu->arch.sie_block->todpr, | |
708 | (u32 __user *)reg->addr); | |
709 | break; | |
710 | case KVM_REG_S390_EPOCHDIFF: | |
711 | r = get_user(vcpu->arch.sie_block->epoch, | |
712 | (u64 __user *)reg->addr); | |
713 | break; | |
714 | case KVM_REG_S390_CPU_TIMER: | |
715 | r = get_user(vcpu->arch.sie_block->cputm, | |
716 | (u64 __user *)reg->addr); | |
717 | break; | |
718 | case KVM_REG_S390_CLOCK_COMP: | |
719 | r = get_user(vcpu->arch.sie_block->ckc, | |
720 | (u64 __user *)reg->addr); | |
721 | break; | |
722 | case KVM_REG_S390_PFTOKEN: | |
723 | r = get_user(vcpu->arch.pfault_token, | |
724 | (u64 __user *)reg->addr); | |
725 | break; | |
726 | case KVM_REG_S390_PFCOMPARE: | |
727 | r = get_user(vcpu->arch.pfault_compare, | |
728 | (u64 __user *)reg->addr); | |
729 | break; | |
730 | case KVM_REG_S390_PFSELECT: | |
731 | r = get_user(vcpu->arch.pfault_select, | |
732 | (u64 __user *)reg->addr); | |
733 | break; | |
734 | case KVM_REG_S390_PP: | |
735 | r = get_user(vcpu->arch.sie_block->pp, | |
736 | (u64 __user *)reg->addr); | |
737 | break; | |
738 | case KVM_REG_S390_GBEA: | |
739 | r = get_user(vcpu->arch.sie_block->gbea, | |
740 | (u64 __user *)reg->addr); | |
741 | break; | |
742 | default: | |
743 | break; | |
744 | } | |
745 | ||
746 | return r; | |
747 | } | |
748 | ||
749 | static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu) | |
750 | { | |
751 | kvm_s390_vcpu_initial_reset(vcpu); | |
752 | return 0; | |
753 | } | |
754 | ||
755 | int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) | |
756 | { | |
757 | memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs)); | |
758 | return 0; | |
759 | } | |
760 | ||
761 | int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) | |
762 | { | |
763 | memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs)); | |
764 | return 0; | |
765 | } | |
766 | ||
767 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, | |
768 | struct kvm_sregs *sregs) | |
769 | { | |
770 | memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs)); | |
771 | memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs)); | |
772 | restore_access_regs(vcpu->run->s.regs.acrs); | |
773 | return 0; | |
774 | } | |
775 | ||
776 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, | |
777 | struct kvm_sregs *sregs) | |
778 | { | |
779 | memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs)); | |
780 | memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs)); | |
781 | return 0; | |
782 | } | |
783 | ||
784 | int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) | |
785 | { | |
786 | if (test_fp_ctl(fpu->fpc)) | |
787 | return -EINVAL; | |
788 | memcpy(&vcpu->arch.guest_fpregs.fprs, &fpu->fprs, sizeof(fpu->fprs)); | |
789 | vcpu->arch.guest_fpregs.fpc = fpu->fpc; | |
790 | restore_fp_ctl(&vcpu->arch.guest_fpregs.fpc); | |
791 | restore_fp_regs(vcpu->arch.guest_fpregs.fprs); | |
792 | return 0; | |
793 | } | |
794 | ||
795 | int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) | |
796 | { | |
797 | memcpy(&fpu->fprs, &vcpu->arch.guest_fpregs.fprs, sizeof(fpu->fprs)); | |
798 | fpu->fpc = vcpu->arch.guest_fpregs.fpc; | |
799 | return 0; | |
800 | } | |
801 | ||
802 | static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw) | |
803 | { | |
804 | int rc = 0; | |
805 | ||
806 | if (!(atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_STOPPED)) | |
807 | rc = -EBUSY; | |
808 | else { | |
809 | vcpu->run->psw_mask = psw.mask; | |
810 | vcpu->run->psw_addr = psw.addr; | |
811 | } | |
812 | return rc; | |
813 | } | |
814 | ||
815 | int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, | |
816 | struct kvm_translation *tr) | |
817 | { | |
818 | return -EINVAL; /* not implemented yet */ | |
819 | } | |
820 | ||
821 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, | |
822 | struct kvm_guest_debug *dbg) | |
823 | { | |
824 | return -EINVAL; /* not implemented yet */ | |
825 | } | |
826 | ||
827 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, | |
828 | struct kvm_mp_state *mp_state) | |
829 | { | |
830 | return -EINVAL; /* not implemented yet */ | |
831 | } | |
832 | ||
833 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
834 | struct kvm_mp_state *mp_state) | |
835 | { | |
836 | return -EINVAL; /* not implemented yet */ | |
837 | } | |
838 | ||
839 | static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu) | |
840 | { | |
841 | /* | |
842 | * We use MMU_RELOAD just to re-arm the ipte notifier for the | |
843 | * guest prefix page. gmap_ipte_notify will wait on the ptl lock. | |
844 | * This ensures that the ipte instruction for this request has | |
845 | * already finished. We might race against a second unmapper that | |
846 | * wants to set the blocking bit. Lets just retry the request loop. | |
847 | */ | |
848 | while (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) { | |
849 | int rc; | |
850 | rc = gmap_ipte_notify(vcpu->arch.gmap, | |
851 | vcpu->arch.sie_block->prefix, | |
852 | PAGE_SIZE * 2); | |
853 | if (rc) | |
854 | return rc; | |
855 | s390_vcpu_unblock(vcpu); | |
856 | } | |
857 | return 0; | |
858 | } | |
859 | ||
860 | static long kvm_arch_fault_in_sync(struct kvm_vcpu *vcpu) | |
861 | { | |
862 | long rc; | |
863 | hva_t fault = gmap_fault(current->thread.gmap_addr, vcpu->arch.gmap); | |
864 | struct mm_struct *mm = current->mm; | |
865 | down_read(&mm->mmap_sem); | |
866 | rc = get_user_pages(current, mm, fault, 1, 1, 0, NULL, NULL); | |
867 | up_read(&mm->mmap_sem); | |
868 | return rc; | |
869 | } | |
870 | ||
871 | static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token, | |
872 | unsigned long token) | |
873 | { | |
874 | struct kvm_s390_interrupt inti; | |
875 | inti.parm64 = token; | |
876 | ||
877 | if (start_token) { | |
878 | inti.type = KVM_S390_INT_PFAULT_INIT; | |
879 | WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &inti)); | |
880 | } else { | |
881 | inti.type = KVM_S390_INT_PFAULT_DONE; | |
882 | WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti)); | |
883 | } | |
884 | } | |
885 | ||
886 | void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, | |
887 | struct kvm_async_pf *work) | |
888 | { | |
889 | trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token); | |
890 | __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token); | |
891 | } | |
892 | ||
893 | void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, | |
894 | struct kvm_async_pf *work) | |
895 | { | |
896 | trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token); | |
897 | __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token); | |
898 | } | |
899 | ||
900 | void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, | |
901 | struct kvm_async_pf *work) | |
902 | { | |
903 | /* s390 will always inject the page directly */ | |
904 | } | |
905 | ||
906 | bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu) | |
907 | { | |
908 | /* | |
909 | * s390 will always inject the page directly, | |
910 | * but we still want check_async_completion to cleanup | |
911 | */ | |
912 | return true; | |
913 | } | |
914 | ||
915 | static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu) | |
916 | { | |
917 | hva_t hva; | |
918 | struct kvm_arch_async_pf arch; | |
919 | int rc; | |
920 | ||
921 | if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) | |
922 | return 0; | |
923 | if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) != | |
924 | vcpu->arch.pfault_compare) | |
925 | return 0; | |
926 | if (psw_extint_disabled(vcpu)) | |
927 | return 0; | |
928 | if (kvm_cpu_has_interrupt(vcpu)) | |
929 | return 0; | |
930 | if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul)) | |
931 | return 0; | |
932 | if (!vcpu->arch.gmap->pfault_enabled) | |
933 | return 0; | |
934 | ||
935 | hva = gmap_fault(current->thread.gmap_addr, vcpu->arch.gmap); | |
936 | if (copy_from_guest(vcpu, &arch.pfault_token, vcpu->arch.pfault_token, 8)) | |
937 | return 0; | |
938 | ||
939 | rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch); | |
940 | return rc; | |
941 | } | |
942 | ||
943 | static int vcpu_pre_run(struct kvm_vcpu *vcpu) | |
944 | { | |
945 | int rc, cpuflags; | |
946 | ||
947 | /* | |
948 | * On s390 notifications for arriving pages will be delivered directly | |
949 | * to the guest but the house keeping for completed pfaults is | |
950 | * handled outside the worker. | |
951 | */ | |
952 | kvm_check_async_pf_completion(vcpu); | |
953 | ||
954 | memcpy(&vcpu->arch.sie_block->gg14, &vcpu->run->s.regs.gprs[14], 16); | |
955 | ||
956 | if (need_resched()) | |
957 | schedule(); | |
958 | ||
959 | if (test_thread_flag(TIF_MCCK_PENDING)) | |
960 | s390_handle_mcck(); | |
961 | ||
962 | if (!kvm_is_ucontrol(vcpu->kvm)) | |
963 | kvm_s390_deliver_pending_interrupts(vcpu); | |
964 | ||
965 | rc = kvm_s390_handle_requests(vcpu); | |
966 | if (rc) | |
967 | return rc; | |
968 | ||
969 | vcpu->arch.sie_block->icptcode = 0; | |
970 | cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags); | |
971 | VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags); | |
972 | trace_kvm_s390_sie_enter(vcpu, cpuflags); | |
973 | ||
974 | return 0; | |
975 | } | |
976 | ||
977 | static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason) | |
978 | { | |
979 | int rc = -1; | |
980 | ||
981 | VCPU_EVENT(vcpu, 6, "exit sie icptcode %d", | |
982 | vcpu->arch.sie_block->icptcode); | |
983 | trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode); | |
984 | ||
985 | if (exit_reason >= 0) { | |
986 | rc = 0; | |
987 | } else if (kvm_is_ucontrol(vcpu->kvm)) { | |
988 | vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL; | |
989 | vcpu->run->s390_ucontrol.trans_exc_code = | |
990 | current->thread.gmap_addr; | |
991 | vcpu->run->s390_ucontrol.pgm_code = 0x10; | |
992 | rc = -EREMOTE; | |
993 | ||
994 | } else if (current->thread.gmap_pfault) { | |
995 | trace_kvm_s390_major_guest_pfault(vcpu); | |
996 | current->thread.gmap_pfault = 0; | |
997 | if (kvm_arch_setup_async_pf(vcpu) || | |
998 | (kvm_arch_fault_in_sync(vcpu) >= 0)) | |
999 | rc = 0; | |
1000 | } | |
1001 | ||
1002 | if (rc == -1) { | |
1003 | VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction"); | |
1004 | trace_kvm_s390_sie_fault(vcpu); | |
1005 | rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); | |
1006 | } | |
1007 | ||
1008 | memcpy(&vcpu->run->s.regs.gprs[14], &vcpu->arch.sie_block->gg14, 16); | |
1009 | ||
1010 | if (rc == 0) { | |
1011 | if (kvm_is_ucontrol(vcpu->kvm)) | |
1012 | /* Don't exit for host interrupts. */ | |
1013 | rc = vcpu->arch.sie_block->icptcode ? -EOPNOTSUPP : 0; | |
1014 | else | |
1015 | rc = kvm_handle_sie_intercept(vcpu); | |
1016 | } | |
1017 | ||
1018 | return rc; | |
1019 | } | |
1020 | ||
1021 | bool kvm_enabled_cmma(void) | |
1022 | { | |
1023 | if (!MACHINE_IS_LPAR) | |
1024 | return false; | |
1025 | /* only enable for z10 and later */ | |
1026 | if (!MACHINE_HAS_EDAT1) | |
1027 | return false; | |
1028 | return true; | |
1029 | } | |
1030 | ||
1031 | static int __vcpu_run(struct kvm_vcpu *vcpu) | |
1032 | { | |
1033 | int rc, exit_reason; | |
1034 | ||
1035 | /* | |
1036 | * We try to hold kvm->srcu during most of vcpu_run (except when run- | |
1037 | * ning the guest), so that memslots (and other stuff) are protected | |
1038 | */ | |
1039 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); | |
1040 | ||
1041 | do { | |
1042 | rc = vcpu_pre_run(vcpu); | |
1043 | if (rc) | |
1044 | break; | |
1045 | ||
1046 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); | |
1047 | /* | |
1048 | * As PF_VCPU will be used in fault handler, between | |
1049 | * guest_enter and guest_exit should be no uaccess. | |
1050 | */ | |
1051 | preempt_disable(); | |
1052 | kvm_guest_enter(); | |
1053 | preempt_enable(); | |
1054 | exit_reason = sie64a(vcpu->arch.sie_block, | |
1055 | vcpu->run->s.regs.gprs); | |
1056 | kvm_guest_exit(); | |
1057 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); | |
1058 | ||
1059 | rc = vcpu_post_run(vcpu, exit_reason); | |
1060 | } while (!signal_pending(current) && !rc); | |
1061 | ||
1062 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); | |
1063 | return rc; | |
1064 | } | |
1065 | ||
1066 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) | |
1067 | { | |
1068 | int rc; | |
1069 | sigset_t sigsaved; | |
1070 | ||
1071 | if (vcpu->sigset_active) | |
1072 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
1073 | ||
1074 | atomic_clear_mask(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags); | |
1075 | ||
1076 | switch (kvm_run->exit_reason) { | |
1077 | case KVM_EXIT_S390_SIEIC: | |
1078 | case KVM_EXIT_UNKNOWN: | |
1079 | case KVM_EXIT_INTR: | |
1080 | case KVM_EXIT_S390_RESET: | |
1081 | case KVM_EXIT_S390_UCONTROL: | |
1082 | case KVM_EXIT_S390_TSCH: | |
1083 | break; | |
1084 | default: | |
1085 | BUG(); | |
1086 | } | |
1087 | ||
1088 | vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask; | |
1089 | vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr; | |
1090 | if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX) { | |
1091 | kvm_run->kvm_dirty_regs &= ~KVM_SYNC_PREFIX; | |
1092 | kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix); | |
1093 | } | |
1094 | if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) { | |
1095 | kvm_run->kvm_dirty_regs &= ~KVM_SYNC_CRS; | |
1096 | memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128); | |
1097 | kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix); | |
1098 | } | |
1099 | ||
1100 | might_fault(); | |
1101 | rc = __vcpu_run(vcpu); | |
1102 | ||
1103 | if (signal_pending(current) && !rc) { | |
1104 | kvm_run->exit_reason = KVM_EXIT_INTR; | |
1105 | rc = -EINTR; | |
1106 | } | |
1107 | ||
1108 | if (rc == -EOPNOTSUPP) { | |
1109 | /* intercept cannot be handled in-kernel, prepare kvm-run */ | |
1110 | kvm_run->exit_reason = KVM_EXIT_S390_SIEIC; | |
1111 | kvm_run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode; | |
1112 | kvm_run->s390_sieic.ipa = vcpu->arch.sie_block->ipa; | |
1113 | kvm_run->s390_sieic.ipb = vcpu->arch.sie_block->ipb; | |
1114 | rc = 0; | |
1115 | } | |
1116 | ||
1117 | if (rc == -EREMOTE) { | |
1118 | /* intercept was handled, but userspace support is needed | |
1119 | * kvm_run has been prepared by the handler */ | |
1120 | rc = 0; | |
1121 | } | |
1122 | ||
1123 | kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask; | |
1124 | kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr; | |
1125 | kvm_run->s.regs.prefix = vcpu->arch.sie_block->prefix; | |
1126 | memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128); | |
1127 | ||
1128 | if (vcpu->sigset_active) | |
1129 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
1130 | ||
1131 | vcpu->stat.exit_userspace++; | |
1132 | return rc; | |
1133 | } | |
1134 | ||
1135 | static int __guestcopy(struct kvm_vcpu *vcpu, u64 guestdest, void *from, | |
1136 | unsigned long n, int prefix) | |
1137 | { | |
1138 | if (prefix) | |
1139 | return copy_to_guest(vcpu, guestdest, from, n); | |
1140 | else | |
1141 | return copy_to_guest_absolute(vcpu, guestdest, from, n); | |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * store status at address | |
1146 | * we use have two special cases: | |
1147 | * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit | |
1148 | * KVM_S390_STORE_STATUS_PREFIXED: -> prefix | |
1149 | */ | |
1150 | int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long addr) | |
1151 | { | |
1152 | unsigned char archmode = 1; | |
1153 | int prefix; | |
1154 | u64 clkcomp; | |
1155 | ||
1156 | if (addr == KVM_S390_STORE_STATUS_NOADDR) { | |
1157 | if (copy_to_guest_absolute(vcpu, 163ul, &archmode, 1)) | |
1158 | return -EFAULT; | |
1159 | addr = SAVE_AREA_BASE; | |
1160 | prefix = 0; | |
1161 | } else if (addr == KVM_S390_STORE_STATUS_PREFIXED) { | |
1162 | if (copy_to_guest(vcpu, 163ul, &archmode, 1)) | |
1163 | return -EFAULT; | |
1164 | addr = SAVE_AREA_BASE; | |
1165 | prefix = 1; | |
1166 | } else | |
1167 | prefix = 0; | |
1168 | ||
1169 | if (__guestcopy(vcpu, addr + offsetof(struct save_area, fp_regs), | |
1170 | vcpu->arch.guest_fpregs.fprs, 128, prefix)) | |
1171 | return -EFAULT; | |
1172 | ||
1173 | if (__guestcopy(vcpu, addr + offsetof(struct save_area, gp_regs), | |
1174 | vcpu->run->s.regs.gprs, 128, prefix)) | |
1175 | return -EFAULT; | |
1176 | ||
1177 | if (__guestcopy(vcpu, addr + offsetof(struct save_area, psw), | |
1178 | &vcpu->arch.sie_block->gpsw, 16, prefix)) | |
1179 | return -EFAULT; | |
1180 | ||
1181 | if (__guestcopy(vcpu, addr + offsetof(struct save_area, pref_reg), | |
1182 | &vcpu->arch.sie_block->prefix, 4, prefix)) | |
1183 | return -EFAULT; | |
1184 | ||
1185 | if (__guestcopy(vcpu, | |
1186 | addr + offsetof(struct save_area, fp_ctrl_reg), | |
1187 | &vcpu->arch.guest_fpregs.fpc, 4, prefix)) | |
1188 | return -EFAULT; | |
1189 | ||
1190 | if (__guestcopy(vcpu, addr + offsetof(struct save_area, tod_reg), | |
1191 | &vcpu->arch.sie_block->todpr, 4, prefix)) | |
1192 | return -EFAULT; | |
1193 | ||
1194 | if (__guestcopy(vcpu, addr + offsetof(struct save_area, timer), | |
1195 | &vcpu->arch.sie_block->cputm, 8, prefix)) | |
1196 | return -EFAULT; | |
1197 | ||
1198 | clkcomp = vcpu->arch.sie_block->ckc >> 8; | |
1199 | if (__guestcopy(vcpu, addr + offsetof(struct save_area, clk_cmp), | |
1200 | &clkcomp, 8, prefix)) | |
1201 | return -EFAULT; | |
1202 | ||
1203 | if (__guestcopy(vcpu, addr + offsetof(struct save_area, acc_regs), | |
1204 | &vcpu->run->s.regs.acrs, 64, prefix)) | |
1205 | return -EFAULT; | |
1206 | ||
1207 | if (__guestcopy(vcpu, | |
1208 | addr + offsetof(struct save_area, ctrl_regs), | |
1209 | &vcpu->arch.sie_block->gcr, 128, prefix)) | |
1210 | return -EFAULT; | |
1211 | return 0; | |
1212 | } | |
1213 | ||
1214 | int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr) | |
1215 | { | |
1216 | /* | |
1217 | * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy | |
1218 | * copying in vcpu load/put. Lets update our copies before we save | |
1219 | * it into the save area | |
1220 | */ | |
1221 | save_fp_ctl(&vcpu->arch.guest_fpregs.fpc); | |
1222 | save_fp_regs(vcpu->arch.guest_fpregs.fprs); | |
1223 | save_access_regs(vcpu->run->s.regs.acrs); | |
1224 | ||
1225 | return kvm_s390_store_status_unloaded(vcpu, addr); | |
1226 | } | |
1227 | ||
1228 | static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, | |
1229 | struct kvm_enable_cap *cap) | |
1230 | { | |
1231 | int r; | |
1232 | ||
1233 | if (cap->flags) | |
1234 | return -EINVAL; | |
1235 | ||
1236 | switch (cap->cap) { | |
1237 | case KVM_CAP_S390_CSS_SUPPORT: | |
1238 | if (!vcpu->kvm->arch.css_support) { | |
1239 | vcpu->kvm->arch.css_support = 1; | |
1240 | trace_kvm_s390_enable_css(vcpu->kvm); | |
1241 | } | |
1242 | r = 0; | |
1243 | break; | |
1244 | default: | |
1245 | r = -EINVAL; | |
1246 | break; | |
1247 | } | |
1248 | return r; | |
1249 | } | |
1250 | ||
1251 | long kvm_arch_vcpu_ioctl(struct file *filp, | |
1252 | unsigned int ioctl, unsigned long arg) | |
1253 | { | |
1254 | struct kvm_vcpu *vcpu = filp->private_data; | |
1255 | void __user *argp = (void __user *)arg; | |
1256 | int idx; | |
1257 | long r; | |
1258 | ||
1259 | switch (ioctl) { | |
1260 | case KVM_S390_INTERRUPT: { | |
1261 | struct kvm_s390_interrupt s390int; | |
1262 | ||
1263 | r = -EFAULT; | |
1264 | if (copy_from_user(&s390int, argp, sizeof(s390int))) | |
1265 | break; | |
1266 | r = kvm_s390_inject_vcpu(vcpu, &s390int); | |
1267 | break; | |
1268 | } | |
1269 | case KVM_S390_STORE_STATUS: | |
1270 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
1271 | r = kvm_s390_vcpu_store_status(vcpu, arg); | |
1272 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
1273 | break; | |
1274 | case KVM_S390_SET_INITIAL_PSW: { | |
1275 | psw_t psw; | |
1276 | ||
1277 | r = -EFAULT; | |
1278 | if (copy_from_user(&psw, argp, sizeof(psw))) | |
1279 | break; | |
1280 | r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw); | |
1281 | break; | |
1282 | } | |
1283 | case KVM_S390_INITIAL_RESET: | |
1284 | r = kvm_arch_vcpu_ioctl_initial_reset(vcpu); | |
1285 | break; | |
1286 | case KVM_SET_ONE_REG: | |
1287 | case KVM_GET_ONE_REG: { | |
1288 | struct kvm_one_reg reg; | |
1289 | r = -EFAULT; | |
1290 | if (copy_from_user(®, argp, sizeof(reg))) | |
1291 | break; | |
1292 | if (ioctl == KVM_SET_ONE_REG) | |
1293 | r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®); | |
1294 | else | |
1295 | r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®); | |
1296 | break; | |
1297 | } | |
1298 | #ifdef CONFIG_KVM_S390_UCONTROL | |
1299 | case KVM_S390_UCAS_MAP: { | |
1300 | struct kvm_s390_ucas_mapping ucasmap; | |
1301 | ||
1302 | if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) { | |
1303 | r = -EFAULT; | |
1304 | break; | |
1305 | } | |
1306 | ||
1307 | if (!kvm_is_ucontrol(vcpu->kvm)) { | |
1308 | r = -EINVAL; | |
1309 | break; | |
1310 | } | |
1311 | ||
1312 | r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr, | |
1313 | ucasmap.vcpu_addr, ucasmap.length); | |
1314 | break; | |
1315 | } | |
1316 | case KVM_S390_UCAS_UNMAP: { | |
1317 | struct kvm_s390_ucas_mapping ucasmap; | |
1318 | ||
1319 | if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) { | |
1320 | r = -EFAULT; | |
1321 | break; | |
1322 | } | |
1323 | ||
1324 | if (!kvm_is_ucontrol(vcpu->kvm)) { | |
1325 | r = -EINVAL; | |
1326 | break; | |
1327 | } | |
1328 | ||
1329 | r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr, | |
1330 | ucasmap.length); | |
1331 | break; | |
1332 | } | |
1333 | #endif | |
1334 | case KVM_S390_VCPU_FAULT: { | |
1335 | r = gmap_fault(arg, vcpu->arch.gmap); | |
1336 | if (!IS_ERR_VALUE(r)) | |
1337 | r = 0; | |
1338 | break; | |
1339 | } | |
1340 | case KVM_ENABLE_CAP: | |
1341 | { | |
1342 | struct kvm_enable_cap cap; | |
1343 | r = -EFAULT; | |
1344 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
1345 | break; | |
1346 | r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); | |
1347 | break; | |
1348 | } | |
1349 | default: | |
1350 | r = -ENOTTY; | |
1351 | } | |
1352 | return r; | |
1353 | } | |
1354 | ||
1355 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) | |
1356 | { | |
1357 | #ifdef CONFIG_KVM_S390_UCONTROL | |
1358 | if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET) | |
1359 | && (kvm_is_ucontrol(vcpu->kvm))) { | |
1360 | vmf->page = virt_to_page(vcpu->arch.sie_block); | |
1361 | get_page(vmf->page); | |
1362 | return 0; | |
1363 | } | |
1364 | #endif | |
1365 | return VM_FAULT_SIGBUS; | |
1366 | } | |
1367 | ||
1368 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
1369 | struct kvm_memory_slot *dont) | |
1370 | { | |
1371 | } | |
1372 | ||
1373 | int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, | |
1374 | unsigned long npages) | |
1375 | { | |
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | void kvm_arch_memslots_updated(struct kvm *kvm) | |
1380 | { | |
1381 | } | |
1382 | ||
1383 | /* Section: memory related */ | |
1384 | int kvm_arch_prepare_memory_region(struct kvm *kvm, | |
1385 | struct kvm_memory_slot *memslot, | |
1386 | struct kvm_userspace_memory_region *mem, | |
1387 | enum kvm_mr_change change) | |
1388 | { | |
1389 | /* A few sanity checks. We can have memory slots which have to be | |
1390 | located/ended at a segment boundary (1MB). The memory in userland is | |
1391 | ok to be fragmented into various different vmas. It is okay to mmap() | |
1392 | and munmap() stuff in this slot after doing this call at any time */ | |
1393 | ||
1394 | if (mem->userspace_addr & 0xffffful) | |
1395 | return -EINVAL; | |
1396 | ||
1397 | if (mem->memory_size & 0xffffful) | |
1398 | return -EINVAL; | |
1399 | ||
1400 | return 0; | |
1401 | } | |
1402 | ||
1403 | void kvm_arch_commit_memory_region(struct kvm *kvm, | |
1404 | struct kvm_userspace_memory_region *mem, | |
1405 | const struct kvm_memory_slot *old, | |
1406 | enum kvm_mr_change change) | |
1407 | { | |
1408 | int rc; | |
1409 | ||
1410 | /* If the basics of the memslot do not change, we do not want | |
1411 | * to update the gmap. Every update causes several unnecessary | |
1412 | * segment translation exceptions. This is usually handled just | |
1413 | * fine by the normal fault handler + gmap, but it will also | |
1414 | * cause faults on the prefix page of running guest CPUs. | |
1415 | */ | |
1416 | if (old->userspace_addr == mem->userspace_addr && | |
1417 | old->base_gfn * PAGE_SIZE == mem->guest_phys_addr && | |
1418 | old->npages * PAGE_SIZE == mem->memory_size) | |
1419 | return; | |
1420 | ||
1421 | rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr, | |
1422 | mem->guest_phys_addr, mem->memory_size); | |
1423 | if (rc) | |
1424 | printk(KERN_WARNING "kvm-s390: failed to commit memory region\n"); | |
1425 | return; | |
1426 | } | |
1427 | ||
1428 | void kvm_arch_flush_shadow_all(struct kvm *kvm) | |
1429 | { | |
1430 | } | |
1431 | ||
1432 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, | |
1433 | struct kvm_memory_slot *slot) | |
1434 | { | |
1435 | } | |
1436 | ||
1437 | static int __init kvm_s390_init(void) | |
1438 | { | |
1439 | int ret; | |
1440 | ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1441 | if (ret) | |
1442 | return ret; | |
1443 | ||
1444 | /* | |
1445 | * guests can ask for up to 255+1 double words, we need a full page | |
1446 | * to hold the maximum amount of facilities. On the other hand, we | |
1447 | * only set facilities that are known to work in KVM. | |
1448 | */ | |
1449 | vfacilities = (unsigned long *) get_zeroed_page(GFP_KERNEL|GFP_DMA); | |
1450 | if (!vfacilities) { | |
1451 | kvm_exit(); | |
1452 | return -ENOMEM; | |
1453 | } | |
1454 | memcpy(vfacilities, S390_lowcore.stfle_fac_list, 16); | |
1455 | vfacilities[0] &= 0xff82fff3f4fc2000UL; | |
1456 | vfacilities[1] &= 0x005c000000000000UL; | |
1457 | return 0; | |
1458 | } | |
1459 | ||
1460 | static void __exit kvm_s390_exit(void) | |
1461 | { | |
1462 | free_page((unsigned long) vfacilities); | |
1463 | kvm_exit(); | |
1464 | } | |
1465 | ||
1466 | module_init(kvm_s390_init); | |
1467 | module_exit(kvm_s390_exit); | |
1468 | ||
1469 | /* | |
1470 | * Enable autoloading of the kvm module. | |
1471 | * Note that we add the module alias here instead of virt/kvm/kvm_main.c | |
1472 | * since x86 takes a different approach. | |
1473 | */ | |
1474 | #include <linux/miscdevice.h> | |
1475 | MODULE_ALIAS_MISCDEV(KVM_MINOR); | |
1476 | MODULE_ALIAS("devname:kvm"); |