KVM: s390: prevent buffer overrun on memory hotplug during migration
[sfrench/cifs-2.6.git] / arch / s390 / kvm / priv.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * handling privileged instructions
4  *
5  * Copyright IBM Corp. 2008, 2013
6  *
7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
8  *               Christian Borntraeger <borntraeger@de.ibm.com>
9  */
10
11 #include <linux/kvm.h>
12 #include <linux/gfp.h>
13 #include <linux/errno.h>
14 #include <linux/compat.h>
15 #include <linux/mm_types.h>
16
17 #include <asm/asm-offsets.h>
18 #include <asm/facility.h>
19 #include <asm/current.h>
20 #include <asm/debug.h>
21 #include <asm/ebcdic.h>
22 #include <asm/sysinfo.h>
23 #include <asm/pgtable.h>
24 #include <asm/page-states.h>
25 #include <asm/pgalloc.h>
26 #include <asm/gmap.h>
27 #include <asm/io.h>
28 #include <asm/ptrace.h>
29 #include <asm/compat.h>
30 #include <asm/sclp.h>
31 #include "gaccess.h"
32 #include "kvm-s390.h"
33 #include "trace.h"
34
35 static int handle_ri(struct kvm_vcpu *vcpu)
36 {
37         if (test_kvm_facility(vcpu->kvm, 64)) {
38                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
39                 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
40                 kvm_s390_retry_instr(vcpu);
41                 return 0;
42         } else
43                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
44 }
45
46 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
47 {
48         if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
49                 return handle_ri(vcpu);
50         else
51                 return -EOPNOTSUPP;
52 }
53
54 static int handle_gs(struct kvm_vcpu *vcpu)
55 {
56         if (test_kvm_facility(vcpu->kvm, 133)) {
57                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
58                 preempt_disable();
59                 __ctl_set_bit(2, 4);
60                 current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
61                 restore_gs_cb(current->thread.gs_cb);
62                 preempt_enable();
63                 vcpu->arch.sie_block->ecb |= ECB_GS;
64                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
65                 vcpu->arch.gs_enabled = 1;
66                 kvm_s390_retry_instr(vcpu);
67                 return 0;
68         } else
69                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
70 }
71
72 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
73 {
74         int code = vcpu->arch.sie_block->ipb & 0xff;
75
76         if (code == 0x49 || code == 0x4d)
77                 return handle_gs(vcpu);
78         else
79                 return -EOPNOTSUPP;
80 }
81 /* Handle SCK (SET CLOCK) interception */
82 static int handle_set_clock(struct kvm_vcpu *vcpu)
83 {
84         int rc;
85         u8 ar;
86         u64 op2, val;
87
88         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
89                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
90
91         op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
92         if (op2 & 7)    /* Operand must be on a doubleword boundary */
93                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
94         rc = read_guest(vcpu, op2, ar, &val, sizeof(val));
95         if (rc)
96                 return kvm_s390_inject_prog_cond(vcpu, rc);
97
98         VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", val);
99         kvm_s390_set_tod_clock(vcpu->kvm, val);
100
101         kvm_s390_set_psw_cc(vcpu, 0);
102         return 0;
103 }
104
105 static int handle_set_prefix(struct kvm_vcpu *vcpu)
106 {
107         u64 operand2;
108         u32 address;
109         int rc;
110         u8 ar;
111
112         vcpu->stat.instruction_spx++;
113
114         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
115                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
116
117         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
118
119         /* must be word boundary */
120         if (operand2 & 3)
121                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
122
123         /* get the value */
124         rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
125         if (rc)
126                 return kvm_s390_inject_prog_cond(vcpu, rc);
127
128         address &= 0x7fffe000u;
129
130         /*
131          * Make sure the new value is valid memory. We only need to check the
132          * first page, since address is 8k aligned and memory pieces are always
133          * at least 1MB aligned and have at least a size of 1MB.
134          */
135         if (kvm_is_error_gpa(vcpu->kvm, address))
136                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
137
138         kvm_s390_set_prefix(vcpu, address);
139         trace_kvm_s390_handle_prefix(vcpu, 1, address);
140         return 0;
141 }
142
143 static int handle_store_prefix(struct kvm_vcpu *vcpu)
144 {
145         u64 operand2;
146         u32 address;
147         int rc;
148         u8 ar;
149
150         vcpu->stat.instruction_stpx++;
151
152         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
153                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
154
155         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
156
157         /* must be word boundary */
158         if (operand2 & 3)
159                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
160
161         address = kvm_s390_get_prefix(vcpu);
162
163         /* get the value */
164         rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
165         if (rc)
166                 return kvm_s390_inject_prog_cond(vcpu, rc);
167
168         VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
169         trace_kvm_s390_handle_prefix(vcpu, 0, address);
170         return 0;
171 }
172
173 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
174 {
175         u16 vcpu_id = vcpu->vcpu_id;
176         u64 ga;
177         int rc;
178         u8 ar;
179
180         vcpu->stat.instruction_stap++;
181
182         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
183                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
184
185         ga = kvm_s390_get_base_disp_s(vcpu, &ar);
186
187         if (ga & 1)
188                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
189
190         rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
191         if (rc)
192                 return kvm_s390_inject_prog_cond(vcpu, rc);
193
194         VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
195         trace_kvm_s390_handle_stap(vcpu, ga);
196         return 0;
197 }
198
199 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
200 {
201         int rc = 0;
202         struct kvm_s390_sie_block *sie_block = vcpu->arch.sie_block;
203
204         trace_kvm_s390_skey_related_inst(vcpu);
205         if (!(sie_block->ictl & (ICTL_ISKE | ICTL_SSKE | ICTL_RRBE)) &&
206             !(atomic_read(&sie_block->cpuflags) & CPUSTAT_KSS))
207                 return rc;
208
209         rc = s390_enable_skey();
210         VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
211         if (!rc) {
212                 if (atomic_read(&sie_block->cpuflags) & CPUSTAT_KSS)
213                         atomic_andnot(CPUSTAT_KSS, &sie_block->cpuflags);
214                 else
215                         sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE |
216                                              ICTL_RRBE);
217         }
218         return rc;
219 }
220
221 static int try_handle_skey(struct kvm_vcpu *vcpu)
222 {
223         int rc;
224
225         vcpu->stat.instruction_storage_key++;
226         rc = kvm_s390_skey_check_enable(vcpu);
227         if (rc)
228                 return rc;
229         if (sclp.has_skey) {
230                 /* with storage-key facility, SIE interprets it for us */
231                 kvm_s390_retry_instr(vcpu);
232                 VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
233                 return -EAGAIN;
234         }
235         return 0;
236 }
237
238 static int handle_iske(struct kvm_vcpu *vcpu)
239 {
240         unsigned long addr;
241         unsigned char key;
242         int reg1, reg2;
243         int rc;
244
245         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
246                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
247
248         rc = try_handle_skey(vcpu);
249         if (rc)
250                 return rc != -EAGAIN ? rc : 0;
251
252         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
253
254         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
255         addr = kvm_s390_logical_to_effective(vcpu, addr);
256         addr = kvm_s390_real_to_abs(vcpu, addr);
257         addr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(addr));
258         if (kvm_is_error_hva(addr))
259                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
260
261         down_read(&current->mm->mmap_sem);
262         rc = get_guest_storage_key(current->mm, addr, &key);
263         up_read(&current->mm->mmap_sem);
264         if (rc)
265                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
266         vcpu->run->s.regs.gprs[reg1] &= ~0xff;
267         vcpu->run->s.regs.gprs[reg1] |= key;
268         return 0;
269 }
270
271 static int handle_rrbe(struct kvm_vcpu *vcpu)
272 {
273         unsigned long addr;
274         int reg1, reg2;
275         int rc;
276
277         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
278                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
279
280         rc = try_handle_skey(vcpu);
281         if (rc)
282                 return rc != -EAGAIN ? rc : 0;
283
284         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
285
286         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
287         addr = kvm_s390_logical_to_effective(vcpu, addr);
288         addr = kvm_s390_real_to_abs(vcpu, addr);
289         addr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(addr));
290         if (kvm_is_error_hva(addr))
291                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
292
293         down_read(&current->mm->mmap_sem);
294         rc = reset_guest_reference_bit(current->mm, addr);
295         up_read(&current->mm->mmap_sem);
296         if (rc < 0)
297                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
298
299         kvm_s390_set_psw_cc(vcpu, rc);
300         return 0;
301 }
302
303 #define SSKE_NQ 0x8
304 #define SSKE_MR 0x4
305 #define SSKE_MC 0x2
306 #define SSKE_MB 0x1
307 static int handle_sske(struct kvm_vcpu *vcpu)
308 {
309         unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
310         unsigned long start, end;
311         unsigned char key, oldkey;
312         int reg1, reg2;
313         int rc;
314
315         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
316                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
317
318         rc = try_handle_skey(vcpu);
319         if (rc)
320                 return rc != -EAGAIN ? rc : 0;
321
322         if (!test_kvm_facility(vcpu->kvm, 8))
323                 m3 &= ~SSKE_MB;
324         if (!test_kvm_facility(vcpu->kvm, 10))
325                 m3 &= ~(SSKE_MC | SSKE_MR);
326         if (!test_kvm_facility(vcpu->kvm, 14))
327                 m3 &= ~SSKE_NQ;
328
329         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
330
331         key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
332         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
333         start = kvm_s390_logical_to_effective(vcpu, start);
334         if (m3 & SSKE_MB) {
335                 /* start already designates an absolute address */
336                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
337         } else {
338                 start = kvm_s390_real_to_abs(vcpu, start);
339                 end = start + PAGE_SIZE;
340         }
341
342         while (start != end) {
343                 unsigned long addr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
344
345                 if (kvm_is_error_hva(addr))
346                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
347
348                 down_read(&current->mm->mmap_sem);
349                 rc = cond_set_guest_storage_key(current->mm, addr, key, &oldkey,
350                                                 m3 & SSKE_NQ, m3 & SSKE_MR,
351                                                 m3 & SSKE_MC);
352                 up_read(&current->mm->mmap_sem);
353                 if (rc < 0)
354                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
355                 start += PAGE_SIZE;
356         }
357
358         if (m3 & (SSKE_MC | SSKE_MR)) {
359                 if (m3 & SSKE_MB) {
360                         /* skey in reg1 is unpredictable */
361                         kvm_s390_set_psw_cc(vcpu, 3);
362                 } else {
363                         kvm_s390_set_psw_cc(vcpu, rc);
364                         vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
365                         vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
366                 }
367         }
368         if (m3 & SSKE_MB) {
369                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
370                         vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
371                 else
372                         vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
373                 end = kvm_s390_logical_to_effective(vcpu, end);
374                 vcpu->run->s.regs.gprs[reg2] |= end;
375         }
376         return 0;
377 }
378
379 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
380 {
381         vcpu->stat.instruction_ipte_interlock++;
382         if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
383                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
384         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
385         kvm_s390_retry_instr(vcpu);
386         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
387         return 0;
388 }
389
390 static int handle_test_block(struct kvm_vcpu *vcpu)
391 {
392         gpa_t addr;
393         int reg2;
394
395         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
396                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
397
398         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
399         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
400         addr = kvm_s390_logical_to_effective(vcpu, addr);
401         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
402                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
403         addr = kvm_s390_real_to_abs(vcpu, addr);
404
405         if (kvm_is_error_gpa(vcpu->kvm, addr))
406                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
407         /*
408          * We don't expect errors on modern systems, and do not care
409          * about storage keys (yet), so let's just clear the page.
410          */
411         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
412                 return -EFAULT;
413         kvm_s390_set_psw_cc(vcpu, 0);
414         vcpu->run->s.regs.gprs[0] = 0;
415         return 0;
416 }
417
418 static int handle_tpi(struct kvm_vcpu *vcpu)
419 {
420         struct kvm_s390_interrupt_info *inti;
421         unsigned long len;
422         u32 tpi_data[3];
423         int rc;
424         u64 addr;
425         u8 ar;
426
427         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
428         if (addr & 3)
429                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
430
431         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
432         if (!inti) {
433                 kvm_s390_set_psw_cc(vcpu, 0);
434                 return 0;
435         }
436
437         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
438         tpi_data[1] = inti->io.io_int_parm;
439         tpi_data[2] = inti->io.io_int_word;
440         if (addr) {
441                 /*
442                  * Store the two-word I/O interruption code into the
443                  * provided area.
444                  */
445                 len = sizeof(tpi_data) - 4;
446                 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
447                 if (rc) {
448                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
449                         goto reinject_interrupt;
450                 }
451         } else {
452                 /*
453                  * Store the three-word I/O interruption code into
454                  * the appropriate lowcore area.
455                  */
456                 len = sizeof(tpi_data);
457                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
458                         /* failed writes to the low core are not recoverable */
459                         rc = -EFAULT;
460                         goto reinject_interrupt;
461                 }
462         }
463
464         /* irq was successfully handed to the guest */
465         kfree(inti);
466         kvm_s390_set_psw_cc(vcpu, 1);
467         return 0;
468 reinject_interrupt:
469         /*
470          * If we encounter a problem storing the interruption code, the
471          * instruction is suppressed from the guest's view: reinject the
472          * interrupt.
473          */
474         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
475                 kfree(inti);
476                 rc = -EFAULT;
477         }
478         /* don't set the cc, a pgm irq was injected or we drop to user space */
479         return rc ? -EFAULT : 0;
480 }
481
482 static int handle_tsch(struct kvm_vcpu *vcpu)
483 {
484         struct kvm_s390_interrupt_info *inti = NULL;
485         const u64 isc_mask = 0xffUL << 24; /* all iscs set */
486
487         /* a valid schid has at least one bit set */
488         if (vcpu->run->s.regs.gprs[1])
489                 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
490                                            vcpu->run->s.regs.gprs[1]);
491
492         /*
493          * Prepare exit to userspace.
494          * We indicate whether we dequeued a pending I/O interrupt
495          * so that userspace can re-inject it if the instruction gets
496          * a program check. While this may re-order the pending I/O
497          * interrupts, this is no problem since the priority is kept
498          * intact.
499          */
500         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
501         vcpu->run->s390_tsch.dequeued = !!inti;
502         if (inti) {
503                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
504                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
505                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
506                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
507         }
508         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
509         kfree(inti);
510         return -EREMOTE;
511 }
512
513 static int handle_io_inst(struct kvm_vcpu *vcpu)
514 {
515         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
516
517         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
518                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
519
520         if (vcpu->kvm->arch.css_support) {
521                 /*
522                  * Most I/O instructions will be handled by userspace.
523                  * Exceptions are tpi and the interrupt portion of tsch.
524                  */
525                 if (vcpu->arch.sie_block->ipa == 0xb236)
526                         return handle_tpi(vcpu);
527                 if (vcpu->arch.sie_block->ipa == 0xb235)
528                         return handle_tsch(vcpu);
529                 /* Handle in userspace. */
530                 return -EOPNOTSUPP;
531         } else {
532                 /*
533                  * Set condition code 3 to stop the guest from issuing channel
534                  * I/O instructions.
535                  */
536                 kvm_s390_set_psw_cc(vcpu, 3);
537                 return 0;
538         }
539 }
540
541 static int handle_stfl(struct kvm_vcpu *vcpu)
542 {
543         int rc;
544         unsigned int fac;
545
546         vcpu->stat.instruction_stfl++;
547
548         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
549                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
550
551         /*
552          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
553          * into a u32 memory representation. They will remain bits 0-31.
554          */
555         fac = *vcpu->kvm->arch.model.fac_list >> 32;
556         rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
557                             &fac, sizeof(fac));
558         if (rc)
559                 return rc;
560         VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
561         trace_kvm_s390_handle_stfl(vcpu, fac);
562         return 0;
563 }
564
565 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
566 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
567 #define PSW_ADDR_24 0x0000000000ffffffUL
568 #define PSW_ADDR_31 0x000000007fffffffUL
569
570 int is_valid_psw(psw_t *psw)
571 {
572         if (psw->mask & PSW_MASK_UNASSIGNED)
573                 return 0;
574         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
575                 if (psw->addr & ~PSW_ADDR_31)
576                         return 0;
577         }
578         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
579                 return 0;
580         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
581                 return 0;
582         if (psw->addr & 1)
583                 return 0;
584         return 1;
585 }
586
587 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
588 {
589         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
590         psw_compat_t new_psw;
591         u64 addr;
592         int rc;
593         u8 ar;
594
595         if (gpsw->mask & PSW_MASK_PSTATE)
596                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
597
598         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
599         if (addr & 7)
600                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
601
602         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
603         if (rc)
604                 return kvm_s390_inject_prog_cond(vcpu, rc);
605         if (!(new_psw.mask & PSW32_MASK_BASE))
606                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
607         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
608         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
609         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
610         if (!is_valid_psw(gpsw))
611                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
612         return 0;
613 }
614
615 static int handle_lpswe(struct kvm_vcpu *vcpu)
616 {
617         psw_t new_psw;
618         u64 addr;
619         int rc;
620         u8 ar;
621
622         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
623                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
624
625         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
626         if (addr & 7)
627                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
628         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
629         if (rc)
630                 return kvm_s390_inject_prog_cond(vcpu, rc);
631         vcpu->arch.sie_block->gpsw = new_psw;
632         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
633                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
634         return 0;
635 }
636
637 static int handle_stidp(struct kvm_vcpu *vcpu)
638 {
639         u64 stidp_data = vcpu->kvm->arch.model.cpuid;
640         u64 operand2;
641         int rc;
642         u8 ar;
643
644         vcpu->stat.instruction_stidp++;
645
646         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
647                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
648
649         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
650
651         if (operand2 & 7)
652                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
653
654         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
655         if (rc)
656                 return kvm_s390_inject_prog_cond(vcpu, rc);
657
658         VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
659         return 0;
660 }
661
662 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
663 {
664         int cpus = 0;
665         int n;
666
667         cpus = atomic_read(&vcpu->kvm->online_vcpus);
668
669         /* deal with other level 3 hypervisors */
670         if (stsi(mem, 3, 2, 2))
671                 mem->count = 0;
672         if (mem->count < 8)
673                 mem->count++;
674         for (n = mem->count - 1; n > 0 ; n--)
675                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
676
677         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
678         mem->vm[0].cpus_total = cpus;
679         mem->vm[0].cpus_configured = cpus;
680         mem->vm[0].cpus_standby = 0;
681         mem->vm[0].cpus_reserved = 0;
682         mem->vm[0].caf = 1000;
683         memcpy(mem->vm[0].name, "KVMguest", 8);
684         ASCEBC(mem->vm[0].name, 8);
685         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
686         ASCEBC(mem->vm[0].cpi, 16);
687 }
688
689 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
690                                  u8 fc, u8 sel1, u16 sel2)
691 {
692         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
693         vcpu->run->s390_stsi.addr = addr;
694         vcpu->run->s390_stsi.ar = ar;
695         vcpu->run->s390_stsi.fc = fc;
696         vcpu->run->s390_stsi.sel1 = sel1;
697         vcpu->run->s390_stsi.sel2 = sel2;
698 }
699
700 static int handle_stsi(struct kvm_vcpu *vcpu)
701 {
702         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
703         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
704         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
705         unsigned long mem = 0;
706         u64 operand2;
707         int rc = 0;
708         u8 ar;
709
710         vcpu->stat.instruction_stsi++;
711         VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
712
713         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
714                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
715
716         if (fc > 3) {
717                 kvm_s390_set_psw_cc(vcpu, 3);
718                 return 0;
719         }
720
721         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
722             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
723                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
724
725         if (fc == 0) {
726                 vcpu->run->s.regs.gprs[0] = 3 << 28;
727                 kvm_s390_set_psw_cc(vcpu, 0);
728                 return 0;
729         }
730
731         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
732
733         if (operand2 & 0xfff)
734                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
735
736         switch (fc) {
737         case 1: /* same handling for 1 and 2 */
738         case 2:
739                 mem = get_zeroed_page(GFP_KERNEL);
740                 if (!mem)
741                         goto out_no_data;
742                 if (stsi((void *) mem, fc, sel1, sel2))
743                         goto out_no_data;
744                 break;
745         case 3:
746                 if (sel1 != 2 || sel2 != 2)
747                         goto out_no_data;
748                 mem = get_zeroed_page(GFP_KERNEL);
749                 if (!mem)
750                         goto out_no_data;
751                 handle_stsi_3_2_2(vcpu, (void *) mem);
752                 break;
753         }
754
755         rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
756         if (rc) {
757                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
758                 goto out;
759         }
760         if (vcpu->kvm->arch.user_stsi) {
761                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
762                 rc = -EREMOTE;
763         }
764         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
765         free_page(mem);
766         kvm_s390_set_psw_cc(vcpu, 0);
767         vcpu->run->s.regs.gprs[0] = 0;
768         return rc;
769 out_no_data:
770         kvm_s390_set_psw_cc(vcpu, 3);
771 out:
772         free_page(mem);
773         return rc;
774 }
775
776 static const intercept_handler_t b2_handlers[256] = {
777         [0x02] = handle_stidp,
778         [0x04] = handle_set_clock,
779         [0x10] = handle_set_prefix,
780         [0x11] = handle_store_prefix,
781         [0x12] = handle_store_cpu_address,
782         [0x14] = kvm_s390_handle_vsie,
783         [0x21] = handle_ipte_interlock,
784         [0x29] = handle_iske,
785         [0x2a] = handle_rrbe,
786         [0x2b] = handle_sske,
787         [0x2c] = handle_test_block,
788         [0x30] = handle_io_inst,
789         [0x31] = handle_io_inst,
790         [0x32] = handle_io_inst,
791         [0x33] = handle_io_inst,
792         [0x34] = handle_io_inst,
793         [0x35] = handle_io_inst,
794         [0x36] = handle_io_inst,
795         [0x37] = handle_io_inst,
796         [0x38] = handle_io_inst,
797         [0x39] = handle_io_inst,
798         [0x3a] = handle_io_inst,
799         [0x3b] = handle_io_inst,
800         [0x3c] = handle_io_inst,
801         [0x50] = handle_ipte_interlock,
802         [0x56] = handle_sthyi,
803         [0x5f] = handle_io_inst,
804         [0x74] = handle_io_inst,
805         [0x76] = handle_io_inst,
806         [0x7d] = handle_stsi,
807         [0xb1] = handle_stfl,
808         [0xb2] = handle_lpswe,
809 };
810
811 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
812 {
813         intercept_handler_t handler;
814
815         /*
816          * A lot of B2 instructions are priviledged. Here we check for
817          * the privileged ones, that we can handle in the kernel.
818          * Anything else goes to userspace.
819          */
820         handler = b2_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
821         if (handler)
822                 return handler(vcpu);
823
824         return -EOPNOTSUPP;
825 }
826
827 static int handle_epsw(struct kvm_vcpu *vcpu)
828 {
829         int reg1, reg2;
830
831         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
832
833         /* This basically extracts the mask half of the psw. */
834         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
835         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
836         if (reg2) {
837                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
838                 vcpu->run->s.regs.gprs[reg2] |=
839                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
840         }
841         return 0;
842 }
843
844 #define PFMF_RESERVED   0xfffc0101UL
845 #define PFMF_SK         0x00020000UL
846 #define PFMF_CF         0x00010000UL
847 #define PFMF_UI         0x00008000UL
848 #define PFMF_FSC        0x00007000UL
849 #define PFMF_NQ         0x00000800UL
850 #define PFMF_MR         0x00000400UL
851 #define PFMF_MC         0x00000200UL
852 #define PFMF_KEY        0x000000feUL
853
854 static int handle_pfmf(struct kvm_vcpu *vcpu)
855 {
856         bool mr = false, mc = false, nq;
857         int reg1, reg2;
858         unsigned long start, end;
859         unsigned char key;
860
861         vcpu->stat.instruction_pfmf++;
862
863         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
864
865         if (!test_kvm_facility(vcpu->kvm, 8))
866                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
867
868         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
869                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
870
871         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
872                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
873
874         /* Only provide non-quiescing support if enabled for the guest */
875         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
876             !test_kvm_facility(vcpu->kvm, 14))
877                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
878
879         /* Only provide conditional-SSKE support if enabled for the guest */
880         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
881             test_kvm_facility(vcpu->kvm, 10)) {
882                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
883                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
884         }
885
886         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
887         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
888         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
889         start = kvm_s390_logical_to_effective(vcpu, start);
890
891         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
892                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
893                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
894         }
895
896         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
897         case 0x00000000:
898                 /* only 4k frames specify a real address */
899                 start = kvm_s390_real_to_abs(vcpu, start);
900                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
901                 break;
902         case 0x00001000:
903                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
904                 break;
905         case 0x00002000:
906                 /* only support 2G frame size if EDAT2 is available and we are
907                    not in 24-bit addressing mode */
908                 if (!test_kvm_facility(vcpu->kvm, 78) ||
909                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
910                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
911                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
912                 break;
913         default:
914                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
915         }
916
917         while (start != end) {
918                 unsigned long useraddr;
919
920                 /* Translate guest address to host address */
921                 useraddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
922                 if (kvm_is_error_hva(useraddr))
923                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
924
925                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
926                         if (clear_user((void __user *)useraddr, PAGE_SIZE))
927                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
928                 }
929
930                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
931                         int rc = kvm_s390_skey_check_enable(vcpu);
932
933                         if (rc)
934                                 return rc;
935                         down_read(&current->mm->mmap_sem);
936                         rc = cond_set_guest_storage_key(current->mm, useraddr,
937                                                         key, NULL, nq, mr, mc);
938                         up_read(&current->mm->mmap_sem);
939                         if (rc < 0)
940                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
941                 }
942
943                 start += PAGE_SIZE;
944         }
945         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
946                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
947                         vcpu->run->s.regs.gprs[reg2] = end;
948                 } else {
949                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
950                         end = kvm_s390_logical_to_effective(vcpu, end);
951                         vcpu->run->s.regs.gprs[reg2] |= end;
952                 }
953         }
954         return 0;
955 }
956
957 static inline int do_essa(struct kvm_vcpu *vcpu, const int orc)
958 {
959         struct kvm_s390_migration_state *ms = vcpu->kvm->arch.migration_state;
960         int r1, r2, nappended, entries;
961         unsigned long gfn, hva, res, pgstev, ptev;
962         unsigned long *cbrlo;
963
964         /*
965          * We don't need to set SD.FPF.SK to 1 here, because if we have a
966          * machine check here we either handle it or crash
967          */
968
969         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
970         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
971         hva = gfn_to_hva(vcpu->kvm, gfn);
972         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
973
974         if (kvm_is_error_hva(hva))
975                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
976
977         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
978         if (nappended < 0) {
979                 res = orc ? 0x10 : 0;
980                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
981                 return 0;
982         }
983         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
984         /*
985          * Set the block-content state part of the result. 0 means resident, so
986          * nothing to do if the page is valid. 2 is for preserved pages
987          * (non-present and non-zero), and 3 for zero pages (non-present and
988          * zero).
989          */
990         if (ptev & _PAGE_INVALID) {
991                 res |= 2;
992                 if (pgstev & _PGSTE_GPS_ZERO)
993                         res |= 1;
994         }
995         if (pgstev & _PGSTE_GPS_NODAT)
996                 res |= 0x20;
997         vcpu->run->s.regs.gprs[r1] = res;
998         /*
999          * It is possible that all the normal 511 slots were full, in which case
1000          * we will now write in the 512th slot, which is reserved for host use.
1001          * In both cases we let the normal essa handling code process all the
1002          * slots, including the reserved one, if needed.
1003          */
1004         if (nappended > 0) {
1005                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1006                 cbrlo[entries] = gfn << PAGE_SHIFT;
1007         }
1008
1009         if (orc && gfn < ms->bitmap_size) {
1010                 /* increment only if we are really flipping the bit to 1 */
1011                 if (!test_and_set_bit(gfn, ms->pgste_bitmap))
1012                         atomic64_inc(&ms->dirty_pages);
1013         }
1014
1015         return nappended;
1016 }
1017
1018 static int handle_essa(struct kvm_vcpu *vcpu)
1019 {
1020         /* entries expected to be 1FF */
1021         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1022         unsigned long *cbrlo;
1023         struct gmap *gmap;
1024         int i, orc;
1025
1026         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1027         gmap = vcpu->arch.gmap;
1028         vcpu->stat.instruction_essa++;
1029         if (!vcpu->kvm->arch.use_cmma)
1030                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1031
1032         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1033                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1034         /* Check for invalid operation request code */
1035         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1036         /* ORCs 0-6 are always valid */
1037         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1038                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1039                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1040
1041         if (likely(!vcpu->kvm->arch.migration_state)) {
1042                 /*
1043                  * CMMA is enabled in the KVM settings, but is disabled in
1044                  * the SIE block and in the mm_context, and we are not doing
1045                  * a migration. Enable CMMA in the mm_context.
1046                  * Since we need to take a write lock to write to the context
1047                  * to avoid races with storage keys handling, we check if the
1048                  * value really needs to be written to; if the value is
1049                  * already correct, we do nothing and avoid the lock.
1050                  */
1051                 if (vcpu->kvm->mm->context.use_cmma == 0) {
1052                         down_write(&vcpu->kvm->mm->mmap_sem);
1053                         vcpu->kvm->mm->context.use_cmma = 1;
1054                         up_write(&vcpu->kvm->mm->mmap_sem);
1055                 }
1056                 /*
1057                  * If we are here, we are supposed to have CMMA enabled in
1058                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1059                  * while the context use_cmma flag is per process.
1060                  * It's possible that the context flag is enabled and the
1061                  * SIE flag is not, so we set the flag always; if it was
1062                  * already set, nothing changes, otherwise we enable it
1063                  * on this CPU too.
1064                  */
1065                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1066                 /* Retry the ESSA instruction */
1067                 kvm_s390_retry_instr(vcpu);
1068         } else {
1069                 /* Account for the possible extra cbrl entry */
1070                 i = do_essa(vcpu, orc);
1071                 if (i < 0)
1072                         return i;
1073                 entries += i;
1074         }
1075         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1076         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1077         down_read(&gmap->mm->mmap_sem);
1078         for (i = 0; i < entries; ++i)
1079                 __gmap_zap(gmap, cbrlo[i]);
1080         up_read(&gmap->mm->mmap_sem);
1081         return 0;
1082 }
1083
1084 static const intercept_handler_t b9_handlers[256] = {
1085         [0x8a] = handle_ipte_interlock,
1086         [0x8d] = handle_epsw,
1087         [0x8e] = handle_ipte_interlock,
1088         [0x8f] = handle_ipte_interlock,
1089         [0xab] = handle_essa,
1090         [0xaf] = handle_pfmf,
1091 };
1092
1093 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1094 {
1095         intercept_handler_t handler;
1096
1097         /* This is handled just as for the B2 instructions. */
1098         handler = b9_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
1099         if (handler)
1100                 return handler(vcpu);
1101
1102         return -EOPNOTSUPP;
1103 }
1104
1105 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1106 {
1107         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1108         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1109         int reg, rc, nr_regs;
1110         u32 ctl_array[16];
1111         u64 ga;
1112         u8 ar;
1113
1114         vcpu->stat.instruction_lctl++;
1115
1116         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1117                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1118
1119         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1120
1121         if (ga & 3)
1122                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1123
1124         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1125         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1126
1127         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1128         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1129         if (rc)
1130                 return kvm_s390_inject_prog_cond(vcpu, rc);
1131         reg = reg1;
1132         nr_regs = 0;
1133         do {
1134                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1135                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1136                 if (reg == reg3)
1137                         break;
1138                 reg = (reg + 1) % 16;
1139         } while (1);
1140         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1141         return 0;
1142 }
1143
1144 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1145 {
1146         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1147         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1148         int reg, rc, nr_regs;
1149         u32 ctl_array[16];
1150         u64 ga;
1151         u8 ar;
1152
1153         vcpu->stat.instruction_stctl++;
1154
1155         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1156                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1157
1158         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1159
1160         if (ga & 3)
1161                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1162
1163         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1164         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1165
1166         reg = reg1;
1167         nr_regs = 0;
1168         do {
1169                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1170                 if (reg == reg3)
1171                         break;
1172                 reg = (reg + 1) % 16;
1173         } while (1);
1174         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1175         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1176 }
1177
1178 static int handle_lctlg(struct kvm_vcpu *vcpu)
1179 {
1180         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1181         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1182         int reg, rc, nr_regs;
1183         u64 ctl_array[16];
1184         u64 ga;
1185         u8 ar;
1186
1187         vcpu->stat.instruction_lctlg++;
1188
1189         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1190                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1191
1192         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1193
1194         if (ga & 7)
1195                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1196
1197         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1198         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1199
1200         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1201         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1202         if (rc)
1203                 return kvm_s390_inject_prog_cond(vcpu, rc);
1204         reg = reg1;
1205         nr_regs = 0;
1206         do {
1207                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1208                 if (reg == reg3)
1209                         break;
1210                 reg = (reg + 1) % 16;
1211         } while (1);
1212         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1213         return 0;
1214 }
1215
1216 static int handle_stctg(struct kvm_vcpu *vcpu)
1217 {
1218         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1219         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1220         int reg, rc, nr_regs;
1221         u64 ctl_array[16];
1222         u64 ga;
1223         u8 ar;
1224
1225         vcpu->stat.instruction_stctg++;
1226
1227         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1228                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1229
1230         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1231
1232         if (ga & 7)
1233                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1234
1235         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1236         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1237
1238         reg = reg1;
1239         nr_regs = 0;
1240         do {
1241                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1242                 if (reg == reg3)
1243                         break;
1244                 reg = (reg + 1) % 16;
1245         } while (1);
1246         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1247         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1248 }
1249
1250 static const intercept_handler_t eb_handlers[256] = {
1251         [0x2f] = handle_lctlg,
1252         [0x25] = handle_stctg,
1253         [0x60] = handle_ri,
1254         [0x61] = handle_ri,
1255         [0x62] = handle_ri,
1256 };
1257
1258 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1259 {
1260         intercept_handler_t handler;
1261
1262         handler = eb_handlers[vcpu->arch.sie_block->ipb & 0xff];
1263         if (handler)
1264                 return handler(vcpu);
1265         return -EOPNOTSUPP;
1266 }
1267
1268 static int handle_tprot(struct kvm_vcpu *vcpu)
1269 {
1270         u64 address1, address2;
1271         unsigned long hva, gpa;
1272         int ret = 0, cc = 0;
1273         bool writable;
1274         u8 ar;
1275
1276         vcpu->stat.instruction_tprot++;
1277
1278         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1279                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1280
1281         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1282
1283         /* we only handle the Linux memory detection case:
1284          * access key == 0
1285          * everything else goes to userspace. */
1286         if (address2 & 0xf0)
1287                 return -EOPNOTSUPP;
1288         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1289                 ipte_lock(vcpu);
1290         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1291         if (ret == PGM_PROTECTION) {
1292                 /* Write protected? Try again with read-only... */
1293                 cc = 1;
1294                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1295                                               GACC_FETCH);
1296         }
1297         if (ret) {
1298                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1299                         ret = kvm_s390_inject_program_int(vcpu, ret);
1300                 } else if (ret > 0) {
1301                         /* Translation not available */
1302                         kvm_s390_set_psw_cc(vcpu, 3);
1303                         ret = 0;
1304                 }
1305                 goto out_unlock;
1306         }
1307
1308         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1309         if (kvm_is_error_hva(hva)) {
1310                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1311         } else {
1312                 if (!writable)
1313                         cc = 1;         /* Write not permitted ==> read-only */
1314                 kvm_s390_set_psw_cc(vcpu, cc);
1315                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1316         }
1317 out_unlock:
1318         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1319                 ipte_unlock(vcpu);
1320         return ret;
1321 }
1322
1323 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1324 {
1325         /* For e5xx... instructions we only handle TPROT */
1326         if ((vcpu->arch.sie_block->ipa & 0x00ff) == 0x01)
1327                 return handle_tprot(vcpu);
1328         return -EOPNOTSUPP;
1329 }
1330
1331 static int handle_sckpf(struct kvm_vcpu *vcpu)
1332 {
1333         u32 value;
1334
1335         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1336                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1337
1338         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1339                 return kvm_s390_inject_program_int(vcpu,
1340                                                    PGM_SPECIFICATION);
1341
1342         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1343         vcpu->arch.sie_block->todpr = value;
1344
1345         return 0;
1346 }
1347
1348 static int handle_ptff(struct kvm_vcpu *vcpu)
1349 {
1350         /* we don't emulate any control instructions yet */
1351         kvm_s390_set_psw_cc(vcpu, 3);
1352         return 0;
1353 }
1354
1355 static const intercept_handler_t x01_handlers[256] = {
1356         [0x04] = handle_ptff,
1357         [0x07] = handle_sckpf,
1358 };
1359
1360 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1361 {
1362         intercept_handler_t handler;
1363
1364         handler = x01_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
1365         if (handler)
1366                 return handler(vcpu);
1367         return -EOPNOTSUPP;
1368 }