23a8370b104506e6a21971b522ff284b38c0be92
[pandora-kernel.git] / arch / s390 / kvm / priv.c
1 /*
2  * handling privileged instructions
3  *
4  * Copyright IBM Corp. 2008
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  */
13
14 #include <linux/kvm.h>
15 #include <linux/gfp.h>
16 #include <linux/errno.h>
17 #include <asm/asm-offsets.h>
18 #include <asm/current.h>
19 #include <asm/debug.h>
20 #include <asm/ebcdic.h>
21 #include <asm/sysinfo.h>
22 #include <asm/ptrace.h>
23 #include <asm/compat.h>
24 #include "gaccess.h"
25 #include "kvm-s390.h"
26 #include "trace.h"
27
28 static int handle_set_prefix(struct kvm_vcpu *vcpu)
29 {
30         u64 operand2;
31         u32 address = 0;
32         u8 tmp;
33
34         vcpu->stat.instruction_spx++;
35
36         operand2 = kvm_s390_get_base_disp_s(vcpu);
37
38         /* must be word boundary */
39         if (operand2 & 3)
40                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
41
42         /* get the value */
43         if (get_guest(vcpu, address, (u32 __user *) operand2))
44                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
45
46         address = address & 0x7fffe000u;
47
48         /* make sure that the new value is valid memory */
49         if (copy_from_guest_absolute(vcpu, &tmp, address, 1) ||
50            (copy_from_guest_absolute(vcpu, &tmp, address + PAGE_SIZE, 1)))
51                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
52
53         kvm_s390_set_prefix(vcpu, address);
54
55         VCPU_EVENT(vcpu, 5, "setting prefix to %x", address);
56         trace_kvm_s390_handle_prefix(vcpu, 1, address);
57         return 0;
58 }
59
60 static int handle_store_prefix(struct kvm_vcpu *vcpu)
61 {
62         u64 operand2;
63         u32 address;
64
65         vcpu->stat.instruction_stpx++;
66
67         operand2 = kvm_s390_get_base_disp_s(vcpu);
68
69         /* must be word boundary */
70         if (operand2 & 3)
71                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
72
73         address = vcpu->arch.sie_block->prefix;
74         address = address & 0x7fffe000u;
75
76         /* get the value */
77         if (put_guest(vcpu, address, (u32 __user *)operand2))
78                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
79
80         VCPU_EVENT(vcpu, 5, "storing prefix to %x", address);
81         trace_kvm_s390_handle_prefix(vcpu, 0, address);
82         return 0;
83 }
84
85 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
86 {
87         u64 useraddr;
88
89         vcpu->stat.instruction_stap++;
90
91         useraddr = kvm_s390_get_base_disp_s(vcpu);
92
93         if (useraddr & 1)
94                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
95
96         if (put_guest(vcpu, vcpu->vcpu_id, (u16 __user *)useraddr))
97                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
98
99         VCPU_EVENT(vcpu, 5, "storing cpu address to %llx", useraddr);
100         trace_kvm_s390_handle_stap(vcpu, useraddr);
101         return 0;
102 }
103
104 static int handle_skey(struct kvm_vcpu *vcpu)
105 {
106         vcpu->stat.instruction_storage_key++;
107         vcpu->arch.sie_block->gpsw.addr -= 4;
108         VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
109         return 0;
110 }
111
112 static int handle_tpi(struct kvm_vcpu *vcpu)
113 {
114         struct kvm_s390_interrupt_info *inti;
115         u64 addr;
116         int cc;
117
118         addr = kvm_s390_get_base_disp_s(vcpu);
119         if (addr & 3)
120                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
121         cc = 0;
122         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->run->s.regs.crs[6], 0);
123         if (!inti)
124                 goto no_interrupt;
125         cc = 1;
126         if (addr) {
127                 /*
128                  * Store the two-word I/O interruption code into the
129                  * provided area.
130                  */
131                 put_guest(vcpu, inti->io.subchannel_id, (u16 __user *) addr);
132                 put_guest(vcpu, inti->io.subchannel_nr, (u16 __user *) (addr + 2));
133                 put_guest(vcpu, inti->io.io_int_parm, (u32 __user *) (addr + 4));
134         } else {
135                 /*
136                  * Store the three-word I/O interruption code into
137                  * the appropriate lowcore area.
138                  */
139                 put_guest(vcpu, inti->io.subchannel_id, (u16 __user *) __LC_SUBCHANNEL_ID);
140                 put_guest(vcpu, inti->io.subchannel_nr, (u16 __user *) __LC_SUBCHANNEL_NR);
141                 put_guest(vcpu, inti->io.io_int_parm, (u32 __user *) __LC_IO_INT_PARM);
142                 put_guest(vcpu, inti->io.io_int_word, (u32 __user *) __LC_IO_INT_WORD);
143         }
144         kfree(inti);
145 no_interrupt:
146         /* Set condition code and we're done. */
147         vcpu->arch.sie_block->gpsw.mask &= ~(3ul << 44);
148         vcpu->arch.sie_block->gpsw.mask |= (cc & 3ul) << 44;
149         return 0;
150 }
151
152 static int handle_tsch(struct kvm_vcpu *vcpu)
153 {
154         struct kvm_s390_interrupt_info *inti;
155
156         inti = kvm_s390_get_io_int(vcpu->kvm, 0,
157                                    vcpu->run->s.regs.gprs[1]);
158
159         /*
160          * Prepare exit to userspace.
161          * We indicate whether we dequeued a pending I/O interrupt
162          * so that userspace can re-inject it if the instruction gets
163          * a program check. While this may re-order the pending I/O
164          * interrupts, this is no problem since the priority is kept
165          * intact.
166          */
167         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
168         vcpu->run->s390_tsch.dequeued = !!inti;
169         if (inti) {
170                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
171                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
172                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
173                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
174         }
175         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
176         kfree(inti);
177         return -EREMOTE;
178 }
179
180 static int handle_io_inst(struct kvm_vcpu *vcpu)
181 {
182         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
183
184         if (vcpu->kvm->arch.css_support) {
185                 /*
186                  * Most I/O instructions will be handled by userspace.
187                  * Exceptions are tpi and the interrupt portion of tsch.
188                  */
189                 if (vcpu->arch.sie_block->ipa == 0xb236)
190                         return handle_tpi(vcpu);
191                 if (vcpu->arch.sie_block->ipa == 0xb235)
192                         return handle_tsch(vcpu);
193                 /* Handle in userspace. */
194                 return -EOPNOTSUPP;
195         } else {
196                 /*
197                  * Set condition code 3 to stop the guest from issueing channel
198                  * I/O instructions.
199                  */
200                 vcpu->arch.sie_block->gpsw.mask &= ~(3ul << 44);
201                 vcpu->arch.sie_block->gpsw.mask |= (3 & 3ul) << 44;
202                 return 0;
203         }
204 }
205
206 static int handle_stfl(struct kvm_vcpu *vcpu)
207 {
208         unsigned int facility_list;
209         int rc;
210
211         vcpu->stat.instruction_stfl++;
212         /* only pass the facility bits, which we can handle */
213         facility_list = S390_lowcore.stfl_fac_list & 0xff00fff3;
214
215         rc = copy_to_guest(vcpu, offsetof(struct _lowcore, stfl_fac_list),
216                            &facility_list, sizeof(facility_list));
217         if (rc)
218                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
219         VCPU_EVENT(vcpu, 5, "store facility list value %x", facility_list);
220         trace_kvm_s390_handle_stfl(vcpu, facility_list);
221         return 0;
222 }
223
224 static void handle_new_psw(struct kvm_vcpu *vcpu)
225 {
226         /* Check whether the new psw is enabled for machine checks. */
227         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK)
228                 kvm_s390_deliver_pending_machine_checks(vcpu);
229 }
230
231 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
232 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
233 #define PSW_ADDR_24 0x0000000000ffffffUL
234 #define PSW_ADDR_31 0x000000007fffffffUL
235
236 static int is_valid_psw(psw_t *psw) {
237         if (psw->mask & PSW_MASK_UNASSIGNED)
238                 return 0;
239         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
240                 if (psw->addr & ~PSW_ADDR_31)
241                         return 0;
242         }
243         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
244                 return 0;
245         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
246                 return 0;
247         return 1;
248 }
249
250 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
251 {
252         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
253         psw_compat_t new_psw;
254         u64 addr;
255
256         if (gpsw->mask & PSW_MASK_PSTATE)
257                 return kvm_s390_inject_program_int(vcpu,
258                                                    PGM_PRIVILEGED_OPERATION);
259         addr = kvm_s390_get_base_disp_s(vcpu);
260         if (addr & 7)
261                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
262         if (copy_from_guest(vcpu, &new_psw, addr, sizeof(new_psw)))
263                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
264         if (!(new_psw.mask & PSW32_MASK_BASE))
265                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
266         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
267         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
268         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
269         if (!is_valid_psw(gpsw))
270                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
271         handle_new_psw(vcpu);
272         return 0;
273 }
274
275 static int handle_lpswe(struct kvm_vcpu *vcpu)
276 {
277         psw_t new_psw;
278         u64 addr;
279
280         addr = kvm_s390_get_base_disp_s(vcpu);
281         if (addr & 7)
282                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
283         if (copy_from_guest(vcpu, &new_psw, addr, sizeof(new_psw)))
284                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
285         vcpu->arch.sie_block->gpsw = new_psw;
286         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
287                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
288         handle_new_psw(vcpu);
289         return 0;
290 }
291
292 static int handle_stidp(struct kvm_vcpu *vcpu)
293 {
294         u64 operand2;
295
296         vcpu->stat.instruction_stidp++;
297
298         operand2 = kvm_s390_get_base_disp_s(vcpu);
299
300         if (operand2 & 7)
301                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
302
303         if (put_guest(vcpu, vcpu->arch.stidp_data, (u64 __user *)operand2))
304                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
305
306         VCPU_EVENT(vcpu, 5, "%s", "store cpu id");
307         return 0;
308 }
309
310 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
311 {
312         struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
313         int cpus = 0;
314         int n;
315
316         spin_lock(&fi->lock);
317         for (n = 0; n < KVM_MAX_VCPUS; n++)
318                 if (fi->local_int[n])
319                         cpus++;
320         spin_unlock(&fi->lock);
321
322         /* deal with other level 3 hypervisors */
323         if (stsi(mem, 3, 2, 2))
324                 mem->count = 0;
325         if (mem->count < 8)
326                 mem->count++;
327         for (n = mem->count - 1; n > 0 ; n--)
328                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
329
330         mem->vm[0].cpus_total = cpus;
331         mem->vm[0].cpus_configured = cpus;
332         mem->vm[0].cpus_standby = 0;
333         mem->vm[0].cpus_reserved = 0;
334         mem->vm[0].caf = 1000;
335         memcpy(mem->vm[0].name, "KVMguest", 8);
336         ASCEBC(mem->vm[0].name, 8);
337         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
338         ASCEBC(mem->vm[0].cpi, 16);
339 }
340
341 static int handle_stsi(struct kvm_vcpu *vcpu)
342 {
343         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
344         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
345         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
346         u64 operand2;
347         unsigned long mem;
348         int rc = 0;
349
350         vcpu->stat.instruction_stsi++;
351         VCPU_EVENT(vcpu, 4, "stsi: fc: %x sel1: %x sel2: %x", fc, sel1, sel2);
352
353         operand2 = kvm_s390_get_base_disp_s(vcpu);
354
355         if (operand2 & 0xfff && fc > 0)
356                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
357
358         switch (fc) {
359         case 0:
360                 vcpu->run->s.regs.gprs[0] = 3 << 28;
361                 vcpu->arch.sie_block->gpsw.mask &= ~(3ul << 44);
362                 return 0;
363         case 1: /* same handling for 1 and 2 */
364         case 2:
365                 mem = get_zeroed_page(GFP_KERNEL);
366                 if (!mem)
367                         goto out_fail;
368                 if (stsi((void *) mem, fc, sel1, sel2))
369                         goto out_mem;
370                 break;
371         case 3:
372                 if (sel1 != 2 || sel2 != 2)
373                         goto out_fail;
374                 mem = get_zeroed_page(GFP_KERNEL);
375                 if (!mem)
376                         goto out_fail;
377                 handle_stsi_3_2_2(vcpu, (void *) mem);
378                 break;
379         default:
380                 goto out_fail;
381         }
382
383         if (copy_to_guest_absolute(vcpu, operand2, (void *) mem, PAGE_SIZE)) {
384                 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
385                 goto out_mem;
386         }
387         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
388         free_page(mem);
389         vcpu->arch.sie_block->gpsw.mask &= ~(3ul << 44);
390         vcpu->run->s.regs.gprs[0] = 0;
391         return 0;
392 out_mem:
393         free_page(mem);
394 out_fail:
395         /* condition code 3 */
396         vcpu->arch.sie_block->gpsw.mask |= 3ul << 44;
397         return rc;
398 }
399
400 static const intercept_handler_t b2_handlers[256] = {
401         [0x02] = handle_stidp,
402         [0x10] = handle_set_prefix,
403         [0x11] = handle_store_prefix,
404         [0x12] = handle_store_cpu_address,
405         [0x29] = handle_skey,
406         [0x2a] = handle_skey,
407         [0x2b] = handle_skey,
408         [0x30] = handle_io_inst,
409         [0x31] = handle_io_inst,
410         [0x32] = handle_io_inst,
411         [0x33] = handle_io_inst,
412         [0x34] = handle_io_inst,
413         [0x35] = handle_io_inst,
414         [0x36] = handle_io_inst,
415         [0x37] = handle_io_inst,
416         [0x38] = handle_io_inst,
417         [0x39] = handle_io_inst,
418         [0x3a] = handle_io_inst,
419         [0x3b] = handle_io_inst,
420         [0x3c] = handle_io_inst,
421         [0x5f] = handle_io_inst,
422         [0x74] = handle_io_inst,
423         [0x76] = handle_io_inst,
424         [0x7d] = handle_stsi,
425         [0xb1] = handle_stfl,
426         [0xb2] = handle_lpswe,
427 };
428
429 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
430 {
431         intercept_handler_t handler;
432
433         /*
434          * a lot of B2 instructions are priviledged. We first check for
435          * the privileged ones, that we can handle in the kernel. If the
436          * kernel can handle this instruction, we check for the problem
437          * state bit and (a) handle the instruction or (b) send a code 2
438          * program check.
439          * Anything else goes to userspace.*/
440         handler = b2_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
441         if (handler) {
442                 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
443                         return kvm_s390_inject_program_int(vcpu,
444                                                    PGM_PRIVILEGED_OPERATION);
445                 else
446                         return handler(vcpu);
447         }
448         return -EOPNOTSUPP;
449 }
450
451 static int handle_epsw(struct kvm_vcpu *vcpu)
452 {
453         int reg1, reg2;
454
455         reg1 = (vcpu->arch.sie_block->ipb & 0x00f00000) >> 24;
456         reg2 = (vcpu->arch.sie_block->ipb & 0x000f0000) >> 16;
457
458         /* This basically extracts the mask half of the psw. */
459         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000;
460         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
461         if (reg2) {
462                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000;
463                 vcpu->run->s.regs.gprs[reg2] |=
464                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffff;
465         }
466         return 0;
467 }
468
469 static const intercept_handler_t b9_handlers[256] = {
470         [0x8d] = handle_epsw,
471         [0x9c] = handle_io_inst,
472 };
473
474 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
475 {
476         intercept_handler_t handler;
477
478         /* This is handled just as for the B2 instructions. */
479         handler = b9_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
480         if (handler) {
481                 if ((handler != handle_epsw) &&
482                     (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE))
483                         return kvm_s390_inject_program_int(vcpu,
484                                                    PGM_PRIVILEGED_OPERATION);
485                 else
486                         return handler(vcpu);
487         }
488         return -EOPNOTSUPP;
489 }
490
491 static const intercept_handler_t eb_handlers[256] = {
492         [0x8a] = handle_io_inst,
493 };
494
495 int kvm_s390_handle_priv_eb(struct kvm_vcpu *vcpu)
496 {
497         intercept_handler_t handler;
498
499         /* All eb instructions that end up here are privileged. */
500         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
501                 return kvm_s390_inject_program_int(vcpu,
502                                                    PGM_PRIVILEGED_OPERATION);
503         handler = eb_handlers[vcpu->arch.sie_block->ipb & 0xff];
504         if (handler)
505                 return handler(vcpu);
506         return -EOPNOTSUPP;
507 }
508
509 static int handle_tprot(struct kvm_vcpu *vcpu)
510 {
511         u64 address1, address2;
512         struct vm_area_struct *vma;
513         unsigned long user_address;
514
515         vcpu->stat.instruction_tprot++;
516
517         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2);
518
519         /* we only handle the Linux memory detection case:
520          * access key == 0
521          * guest DAT == off
522          * everything else goes to userspace. */
523         if (address2 & 0xf0)
524                 return -EOPNOTSUPP;
525         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
526                 return -EOPNOTSUPP;
527
528         down_read(&current->mm->mmap_sem);
529         user_address = __gmap_translate(address1, vcpu->arch.gmap);
530         if (IS_ERR_VALUE(user_address))
531                 goto out_inject;
532         vma = find_vma(current->mm, user_address);
533         if (!vma)
534                 goto out_inject;
535         vcpu->arch.sie_block->gpsw.mask &= ~(3ul << 44);
536         if (!(vma->vm_flags & VM_WRITE) && (vma->vm_flags & VM_READ))
537                 vcpu->arch.sie_block->gpsw.mask |= (1ul << 44);
538         if (!(vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_READ))
539                 vcpu->arch.sie_block->gpsw.mask |= (2ul << 44);
540
541         up_read(&current->mm->mmap_sem);
542         return 0;
543
544 out_inject:
545         up_read(&current->mm->mmap_sem);
546         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
547 }
548
549 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
550 {
551         /* For e5xx... instructions we only handle TPROT */
552         if ((vcpu->arch.sie_block->ipa & 0x00ff) == 0x01)
553                 return handle_tprot(vcpu);
554         return -EOPNOTSUPP;
555 }
556
557 static int handle_sckpf(struct kvm_vcpu *vcpu)
558 {
559         u32 value;
560
561         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
562                 return kvm_s390_inject_program_int(vcpu,
563                                                    PGM_PRIVILEGED_OPERATION);
564
565         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
566                 return kvm_s390_inject_program_int(vcpu,
567                                                    PGM_SPECIFICATION);
568
569         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
570         vcpu->arch.sie_block->todpr = value;
571
572         return 0;
573 }
574
575 static const intercept_handler_t x01_handlers[256] = {
576         [0x07] = handle_sckpf,
577 };
578
579 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
580 {
581         intercept_handler_t handler;
582
583         handler = x01_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
584         if (handler)
585                 return handler(vcpu);
586         return -EOPNOTSUPP;
587 }