xref: /linux/arch/s390/kvm/priv.c (revision 44f57d78)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * handling privileged instructions
4  *
5  * Copyright IBM Corp. 2008, 2018
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/sclp.h>
30 #include "gaccess.h"
31 #include "kvm-s390.h"
32 #include "trace.h"
33 
34 static int handle_ri(struct kvm_vcpu *vcpu)
35 {
36 	vcpu->stat.instruction_ri++;
37 
38 	if (test_kvm_facility(vcpu->kvm, 64)) {
39 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
40 		vcpu->arch.sie_block->ecb3 |= ECB3_RI;
41 		kvm_s390_retry_instr(vcpu);
42 		return 0;
43 	} else
44 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
45 }
46 
47 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
48 {
49 	if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
50 		return handle_ri(vcpu);
51 	else
52 		return -EOPNOTSUPP;
53 }
54 
55 static int handle_gs(struct kvm_vcpu *vcpu)
56 {
57 	vcpu->stat.instruction_gs++;
58 
59 	if (test_kvm_facility(vcpu->kvm, 133)) {
60 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
61 		preempt_disable();
62 		__ctl_set_bit(2, 4);
63 		current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
64 		restore_gs_cb(current->thread.gs_cb);
65 		preempt_enable();
66 		vcpu->arch.sie_block->ecb |= ECB_GS;
67 		vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
68 		vcpu->arch.gs_enabled = 1;
69 		kvm_s390_retry_instr(vcpu);
70 		return 0;
71 	} else
72 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
73 }
74 
75 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
76 {
77 	int code = vcpu->arch.sie_block->ipb & 0xff;
78 
79 	if (code == 0x49 || code == 0x4d)
80 		return handle_gs(vcpu);
81 	else
82 		return -EOPNOTSUPP;
83 }
84 /* Handle SCK (SET CLOCK) interception */
85 static int handle_set_clock(struct kvm_vcpu *vcpu)
86 {
87 	struct kvm_s390_vm_tod_clock gtod = { 0 };
88 	int rc;
89 	u8 ar;
90 	u64 op2;
91 
92 	vcpu->stat.instruction_sck++;
93 
94 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
95 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
96 
97 	op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
98 	if (op2 & 7)	/* Operand must be on a doubleword boundary */
99 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
100 	rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
101 	if (rc)
102 		return kvm_s390_inject_prog_cond(vcpu, rc);
103 
104 	VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
105 	kvm_s390_set_tod_clock(vcpu->kvm, &gtod);
106 
107 	kvm_s390_set_psw_cc(vcpu, 0);
108 	return 0;
109 }
110 
111 static int handle_set_prefix(struct kvm_vcpu *vcpu)
112 {
113 	u64 operand2;
114 	u32 address;
115 	int rc;
116 	u8 ar;
117 
118 	vcpu->stat.instruction_spx++;
119 
120 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
121 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
122 
123 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
124 
125 	/* must be word boundary */
126 	if (operand2 & 3)
127 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
128 
129 	/* get the value */
130 	rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
131 	if (rc)
132 		return kvm_s390_inject_prog_cond(vcpu, rc);
133 
134 	address &= 0x7fffe000u;
135 
136 	/*
137 	 * Make sure the new value is valid memory. We only need to check the
138 	 * first page, since address is 8k aligned and memory pieces are always
139 	 * at least 1MB aligned and have at least a size of 1MB.
140 	 */
141 	if (kvm_is_error_gpa(vcpu->kvm, address))
142 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
143 
144 	kvm_s390_set_prefix(vcpu, address);
145 	trace_kvm_s390_handle_prefix(vcpu, 1, address);
146 	return 0;
147 }
148 
149 static int handle_store_prefix(struct kvm_vcpu *vcpu)
150 {
151 	u64 operand2;
152 	u32 address;
153 	int rc;
154 	u8 ar;
155 
156 	vcpu->stat.instruction_stpx++;
157 
158 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
159 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
160 
161 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
162 
163 	/* must be word boundary */
164 	if (operand2 & 3)
165 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
166 
167 	address = kvm_s390_get_prefix(vcpu);
168 
169 	/* get the value */
170 	rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
171 	if (rc)
172 		return kvm_s390_inject_prog_cond(vcpu, rc);
173 
174 	VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
175 	trace_kvm_s390_handle_prefix(vcpu, 0, address);
176 	return 0;
177 }
178 
179 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
180 {
181 	u16 vcpu_id = vcpu->vcpu_id;
182 	u64 ga;
183 	int rc;
184 	u8 ar;
185 
186 	vcpu->stat.instruction_stap++;
187 
188 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
189 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
190 
191 	ga = kvm_s390_get_base_disp_s(vcpu, &ar);
192 
193 	if (ga & 1)
194 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
195 
196 	rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
197 	if (rc)
198 		return kvm_s390_inject_prog_cond(vcpu, rc);
199 
200 	VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
201 	trace_kvm_s390_handle_stap(vcpu, ga);
202 	return 0;
203 }
204 
205 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
206 {
207 	int rc;
208 
209 	trace_kvm_s390_skey_related_inst(vcpu);
210 	/* Already enabled? */
211 	if (vcpu->arch.skey_enabled)
212 		return 0;
213 
214 	rc = s390_enable_skey();
215 	VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
216 	if (rc)
217 		return rc;
218 
219 	if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
220 		kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
221 	if (!vcpu->kvm->arch.use_skf)
222 		vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
223 	else
224 		vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
225 	vcpu->arch.skey_enabled = true;
226 	return 0;
227 }
228 
229 static int try_handle_skey(struct kvm_vcpu *vcpu)
230 {
231 	int rc;
232 
233 	rc = kvm_s390_skey_check_enable(vcpu);
234 	if (rc)
235 		return rc;
236 	if (vcpu->kvm->arch.use_skf) {
237 		/* with storage-key facility, SIE interprets it for us */
238 		kvm_s390_retry_instr(vcpu);
239 		VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
240 		return -EAGAIN;
241 	}
242 	return 0;
243 }
244 
245 static int handle_iske(struct kvm_vcpu *vcpu)
246 {
247 	unsigned long gaddr, vmaddr;
248 	unsigned char key;
249 	int reg1, reg2;
250 	bool unlocked;
251 	int rc;
252 
253 	vcpu->stat.instruction_iske++;
254 
255 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
256 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
257 
258 	rc = try_handle_skey(vcpu);
259 	if (rc)
260 		return rc != -EAGAIN ? rc : 0;
261 
262 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
263 
264 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
265 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
266 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
267 	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
268 	if (kvm_is_error_hva(vmaddr))
269 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
270 retry:
271 	unlocked = false;
272 	down_read(&current->mm->mmap_sem);
273 	rc = get_guest_storage_key(current->mm, vmaddr, &key);
274 
275 	if (rc) {
276 		rc = fixup_user_fault(current, current->mm, vmaddr,
277 				      FAULT_FLAG_WRITE, &unlocked);
278 		if (!rc) {
279 			up_read(&current->mm->mmap_sem);
280 			goto retry;
281 		}
282 	}
283 	up_read(&current->mm->mmap_sem);
284 	if (rc == -EFAULT)
285 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
286 	if (rc < 0)
287 		return rc;
288 	vcpu->run->s.regs.gprs[reg1] &= ~0xff;
289 	vcpu->run->s.regs.gprs[reg1] |= key;
290 	return 0;
291 }
292 
293 static int handle_rrbe(struct kvm_vcpu *vcpu)
294 {
295 	unsigned long vmaddr, gaddr;
296 	int reg1, reg2;
297 	bool unlocked;
298 	int rc;
299 
300 	vcpu->stat.instruction_rrbe++;
301 
302 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
303 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
304 
305 	rc = try_handle_skey(vcpu);
306 	if (rc)
307 		return rc != -EAGAIN ? rc : 0;
308 
309 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
310 
311 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
312 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
313 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
314 	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
315 	if (kvm_is_error_hva(vmaddr))
316 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
317 retry:
318 	unlocked = false;
319 	down_read(&current->mm->mmap_sem);
320 	rc = reset_guest_reference_bit(current->mm, vmaddr);
321 	if (rc < 0) {
322 		rc = fixup_user_fault(current, current->mm, vmaddr,
323 				      FAULT_FLAG_WRITE, &unlocked);
324 		if (!rc) {
325 			up_read(&current->mm->mmap_sem);
326 			goto retry;
327 		}
328 	}
329 	up_read(&current->mm->mmap_sem);
330 	if (rc == -EFAULT)
331 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
332 	if (rc < 0)
333 		return rc;
334 	kvm_s390_set_psw_cc(vcpu, rc);
335 	return 0;
336 }
337 
338 #define SSKE_NQ 0x8
339 #define SSKE_MR 0x4
340 #define SSKE_MC 0x2
341 #define SSKE_MB 0x1
342 static int handle_sske(struct kvm_vcpu *vcpu)
343 {
344 	unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
345 	unsigned long start, end;
346 	unsigned char key, oldkey;
347 	int reg1, reg2;
348 	bool unlocked;
349 	int rc;
350 
351 	vcpu->stat.instruction_sske++;
352 
353 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
354 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
355 
356 	rc = try_handle_skey(vcpu);
357 	if (rc)
358 		return rc != -EAGAIN ? rc : 0;
359 
360 	if (!test_kvm_facility(vcpu->kvm, 8))
361 		m3 &= ~SSKE_MB;
362 	if (!test_kvm_facility(vcpu->kvm, 10))
363 		m3 &= ~(SSKE_MC | SSKE_MR);
364 	if (!test_kvm_facility(vcpu->kvm, 14))
365 		m3 &= ~SSKE_NQ;
366 
367 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
368 
369 	key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
370 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
371 	start = kvm_s390_logical_to_effective(vcpu, start);
372 	if (m3 & SSKE_MB) {
373 		/* start already designates an absolute address */
374 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
375 	} else {
376 		start = kvm_s390_real_to_abs(vcpu, start);
377 		end = start + PAGE_SIZE;
378 	}
379 
380 	while (start != end) {
381 		unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
382 		unlocked = false;
383 
384 		if (kvm_is_error_hva(vmaddr))
385 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
386 
387 		down_read(&current->mm->mmap_sem);
388 		rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
389 						m3 & SSKE_NQ, m3 & SSKE_MR,
390 						m3 & SSKE_MC);
391 
392 		if (rc < 0) {
393 			rc = fixup_user_fault(current, current->mm, vmaddr,
394 					      FAULT_FLAG_WRITE, &unlocked);
395 			rc = !rc ? -EAGAIN : rc;
396 		}
397 		up_read(&current->mm->mmap_sem);
398 		if (rc == -EFAULT)
399 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
400 		if (rc < 0)
401 			return rc;
402 		start += PAGE_SIZE;
403 	}
404 
405 	if (m3 & (SSKE_MC | SSKE_MR)) {
406 		if (m3 & SSKE_MB) {
407 			/* skey in reg1 is unpredictable */
408 			kvm_s390_set_psw_cc(vcpu, 3);
409 		} else {
410 			kvm_s390_set_psw_cc(vcpu, rc);
411 			vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
412 			vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
413 		}
414 	}
415 	if (m3 & SSKE_MB) {
416 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
417 			vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
418 		else
419 			vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
420 		end = kvm_s390_logical_to_effective(vcpu, end);
421 		vcpu->run->s.regs.gprs[reg2] |= end;
422 	}
423 	return 0;
424 }
425 
426 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
427 {
428 	vcpu->stat.instruction_ipte_interlock++;
429 	if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
430 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
431 	wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
432 	kvm_s390_retry_instr(vcpu);
433 	VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
434 	return 0;
435 }
436 
437 static int handle_test_block(struct kvm_vcpu *vcpu)
438 {
439 	gpa_t addr;
440 	int reg2;
441 
442 	vcpu->stat.instruction_tb++;
443 
444 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
445 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
446 
447 	kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
448 	addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
449 	addr = kvm_s390_logical_to_effective(vcpu, addr);
450 	if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
451 		return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
452 	addr = kvm_s390_real_to_abs(vcpu, addr);
453 
454 	if (kvm_is_error_gpa(vcpu->kvm, addr))
455 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
456 	/*
457 	 * We don't expect errors on modern systems, and do not care
458 	 * about storage keys (yet), so let's just clear the page.
459 	 */
460 	if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
461 		return -EFAULT;
462 	kvm_s390_set_psw_cc(vcpu, 0);
463 	vcpu->run->s.regs.gprs[0] = 0;
464 	return 0;
465 }
466 
467 static int handle_tpi(struct kvm_vcpu *vcpu)
468 {
469 	struct kvm_s390_interrupt_info *inti;
470 	unsigned long len;
471 	u32 tpi_data[3];
472 	int rc;
473 	u64 addr;
474 	u8 ar;
475 
476 	vcpu->stat.instruction_tpi++;
477 
478 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
479 	if (addr & 3)
480 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
481 
482 	inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
483 	if (!inti) {
484 		kvm_s390_set_psw_cc(vcpu, 0);
485 		return 0;
486 	}
487 
488 	tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
489 	tpi_data[1] = inti->io.io_int_parm;
490 	tpi_data[2] = inti->io.io_int_word;
491 	if (addr) {
492 		/*
493 		 * Store the two-word I/O interruption code into the
494 		 * provided area.
495 		 */
496 		len = sizeof(tpi_data) - 4;
497 		rc = write_guest(vcpu, addr, ar, &tpi_data, len);
498 		if (rc) {
499 			rc = kvm_s390_inject_prog_cond(vcpu, rc);
500 			goto reinject_interrupt;
501 		}
502 	} else {
503 		/*
504 		 * Store the three-word I/O interruption code into
505 		 * the appropriate lowcore area.
506 		 */
507 		len = sizeof(tpi_data);
508 		if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
509 			/* failed writes to the low core are not recoverable */
510 			rc = -EFAULT;
511 			goto reinject_interrupt;
512 		}
513 	}
514 
515 	/* irq was successfully handed to the guest */
516 	kfree(inti);
517 	kvm_s390_set_psw_cc(vcpu, 1);
518 	return 0;
519 reinject_interrupt:
520 	/*
521 	 * If we encounter a problem storing the interruption code, the
522 	 * instruction is suppressed from the guest's view: reinject the
523 	 * interrupt.
524 	 */
525 	if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
526 		kfree(inti);
527 		rc = -EFAULT;
528 	}
529 	/* don't set the cc, a pgm irq was injected or we drop to user space */
530 	return rc ? -EFAULT : 0;
531 }
532 
533 static int handle_tsch(struct kvm_vcpu *vcpu)
534 {
535 	struct kvm_s390_interrupt_info *inti = NULL;
536 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
537 
538 	vcpu->stat.instruction_tsch++;
539 
540 	/* a valid schid has at least one bit set */
541 	if (vcpu->run->s.regs.gprs[1])
542 		inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
543 					   vcpu->run->s.regs.gprs[1]);
544 
545 	/*
546 	 * Prepare exit to userspace.
547 	 * We indicate whether we dequeued a pending I/O interrupt
548 	 * so that userspace can re-inject it if the instruction gets
549 	 * a program check. While this may re-order the pending I/O
550 	 * interrupts, this is no problem since the priority is kept
551 	 * intact.
552 	 */
553 	vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
554 	vcpu->run->s390_tsch.dequeued = !!inti;
555 	if (inti) {
556 		vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
557 		vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
558 		vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
559 		vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
560 	}
561 	vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
562 	kfree(inti);
563 	return -EREMOTE;
564 }
565 
566 static int handle_io_inst(struct kvm_vcpu *vcpu)
567 {
568 	VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
569 
570 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
571 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
572 
573 	if (vcpu->kvm->arch.css_support) {
574 		/*
575 		 * Most I/O instructions will be handled by userspace.
576 		 * Exceptions are tpi and the interrupt portion of tsch.
577 		 */
578 		if (vcpu->arch.sie_block->ipa == 0xb236)
579 			return handle_tpi(vcpu);
580 		if (vcpu->arch.sie_block->ipa == 0xb235)
581 			return handle_tsch(vcpu);
582 		/* Handle in userspace. */
583 		vcpu->stat.instruction_io_other++;
584 		return -EOPNOTSUPP;
585 	} else {
586 		/*
587 		 * Set condition code 3 to stop the guest from issuing channel
588 		 * I/O instructions.
589 		 */
590 		kvm_s390_set_psw_cc(vcpu, 3);
591 		return 0;
592 	}
593 }
594 
595 static int handle_stfl(struct kvm_vcpu *vcpu)
596 {
597 	int rc;
598 	unsigned int fac;
599 
600 	vcpu->stat.instruction_stfl++;
601 
602 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
603 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
604 
605 	/*
606 	 * We need to shift the lower 32 facility bits (bit 0-31) from a u64
607 	 * into a u32 memory representation. They will remain bits 0-31.
608 	 */
609 	fac = *vcpu->kvm->arch.model.fac_list >> 32;
610 	rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
611 			    &fac, sizeof(fac));
612 	if (rc)
613 		return rc;
614 	VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
615 	trace_kvm_s390_handle_stfl(vcpu, fac);
616 	return 0;
617 }
618 
619 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
620 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
621 #define PSW_ADDR_24 0x0000000000ffffffUL
622 #define PSW_ADDR_31 0x000000007fffffffUL
623 
624 int is_valid_psw(psw_t *psw)
625 {
626 	if (psw->mask & PSW_MASK_UNASSIGNED)
627 		return 0;
628 	if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
629 		if (psw->addr & ~PSW_ADDR_31)
630 			return 0;
631 	}
632 	if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
633 		return 0;
634 	if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
635 		return 0;
636 	if (psw->addr & 1)
637 		return 0;
638 	return 1;
639 }
640 
641 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
642 {
643 	psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
644 	psw_compat_t new_psw;
645 	u64 addr;
646 	int rc;
647 	u8 ar;
648 
649 	vcpu->stat.instruction_lpsw++;
650 
651 	if (gpsw->mask & PSW_MASK_PSTATE)
652 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
653 
654 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
655 	if (addr & 7)
656 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
657 
658 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
659 	if (rc)
660 		return kvm_s390_inject_prog_cond(vcpu, rc);
661 	if (!(new_psw.mask & PSW32_MASK_BASE))
662 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
663 	gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
664 	gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
665 	gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
666 	if (!is_valid_psw(gpsw))
667 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
668 	return 0;
669 }
670 
671 static int handle_lpswe(struct kvm_vcpu *vcpu)
672 {
673 	psw_t new_psw;
674 	u64 addr;
675 	int rc;
676 	u8 ar;
677 
678 	vcpu->stat.instruction_lpswe++;
679 
680 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
681 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
682 
683 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
684 	if (addr & 7)
685 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
686 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
687 	if (rc)
688 		return kvm_s390_inject_prog_cond(vcpu, rc);
689 	vcpu->arch.sie_block->gpsw = new_psw;
690 	if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
691 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
692 	return 0;
693 }
694 
695 static int handle_stidp(struct kvm_vcpu *vcpu)
696 {
697 	u64 stidp_data = vcpu->kvm->arch.model.cpuid;
698 	u64 operand2;
699 	int rc;
700 	u8 ar;
701 
702 	vcpu->stat.instruction_stidp++;
703 
704 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
705 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
706 
707 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
708 
709 	if (operand2 & 7)
710 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
711 
712 	rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
713 	if (rc)
714 		return kvm_s390_inject_prog_cond(vcpu, rc);
715 
716 	VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
717 	return 0;
718 }
719 
720 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
721 {
722 	int cpus = 0;
723 	int n;
724 
725 	cpus = atomic_read(&vcpu->kvm->online_vcpus);
726 
727 	/* deal with other level 3 hypervisors */
728 	if (stsi(mem, 3, 2, 2))
729 		mem->count = 0;
730 	if (mem->count < 8)
731 		mem->count++;
732 	for (n = mem->count - 1; n > 0 ; n--)
733 		memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
734 
735 	memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
736 	mem->vm[0].cpus_total = cpus;
737 	mem->vm[0].cpus_configured = cpus;
738 	mem->vm[0].cpus_standby = 0;
739 	mem->vm[0].cpus_reserved = 0;
740 	mem->vm[0].caf = 1000;
741 	memcpy(mem->vm[0].name, "KVMguest", 8);
742 	ASCEBC(mem->vm[0].name, 8);
743 	memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
744 	ASCEBC(mem->vm[0].cpi, 16);
745 }
746 
747 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
748 				 u8 fc, u8 sel1, u16 sel2)
749 {
750 	vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
751 	vcpu->run->s390_stsi.addr = addr;
752 	vcpu->run->s390_stsi.ar = ar;
753 	vcpu->run->s390_stsi.fc = fc;
754 	vcpu->run->s390_stsi.sel1 = sel1;
755 	vcpu->run->s390_stsi.sel2 = sel2;
756 }
757 
758 static int handle_stsi(struct kvm_vcpu *vcpu)
759 {
760 	int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
761 	int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
762 	int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
763 	unsigned long mem = 0;
764 	u64 operand2;
765 	int rc = 0;
766 	u8 ar;
767 
768 	vcpu->stat.instruction_stsi++;
769 	VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
770 
771 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
772 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
773 
774 	if (fc > 3) {
775 		kvm_s390_set_psw_cc(vcpu, 3);
776 		return 0;
777 	}
778 
779 	if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
780 	    || vcpu->run->s.regs.gprs[1] & 0xffff0000)
781 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
782 
783 	if (fc == 0) {
784 		vcpu->run->s.regs.gprs[0] = 3 << 28;
785 		kvm_s390_set_psw_cc(vcpu, 0);
786 		return 0;
787 	}
788 
789 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
790 
791 	if (operand2 & 0xfff)
792 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
793 
794 	switch (fc) {
795 	case 1: /* same handling for 1 and 2 */
796 	case 2:
797 		mem = get_zeroed_page(GFP_KERNEL);
798 		if (!mem)
799 			goto out_no_data;
800 		if (stsi((void *) mem, fc, sel1, sel2))
801 			goto out_no_data;
802 		break;
803 	case 3:
804 		if (sel1 != 2 || sel2 != 2)
805 			goto out_no_data;
806 		mem = get_zeroed_page(GFP_KERNEL);
807 		if (!mem)
808 			goto out_no_data;
809 		handle_stsi_3_2_2(vcpu, (void *) mem);
810 		break;
811 	}
812 
813 	rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
814 	if (rc) {
815 		rc = kvm_s390_inject_prog_cond(vcpu, rc);
816 		goto out;
817 	}
818 	if (vcpu->kvm->arch.user_stsi) {
819 		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
820 		rc = -EREMOTE;
821 	}
822 	trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
823 	free_page(mem);
824 	kvm_s390_set_psw_cc(vcpu, 0);
825 	vcpu->run->s.regs.gprs[0] = 0;
826 	return rc;
827 out_no_data:
828 	kvm_s390_set_psw_cc(vcpu, 3);
829 out:
830 	free_page(mem);
831 	return rc;
832 }
833 
834 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
835 {
836 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
837 	case 0x02:
838 		return handle_stidp(vcpu);
839 	case 0x04:
840 		return handle_set_clock(vcpu);
841 	case 0x10:
842 		return handle_set_prefix(vcpu);
843 	case 0x11:
844 		return handle_store_prefix(vcpu);
845 	case 0x12:
846 		return handle_store_cpu_address(vcpu);
847 	case 0x14:
848 		return kvm_s390_handle_vsie(vcpu);
849 	case 0x21:
850 	case 0x50:
851 		return handle_ipte_interlock(vcpu);
852 	case 0x29:
853 		return handle_iske(vcpu);
854 	case 0x2a:
855 		return handle_rrbe(vcpu);
856 	case 0x2b:
857 		return handle_sske(vcpu);
858 	case 0x2c:
859 		return handle_test_block(vcpu);
860 	case 0x30:
861 	case 0x31:
862 	case 0x32:
863 	case 0x33:
864 	case 0x34:
865 	case 0x35:
866 	case 0x36:
867 	case 0x37:
868 	case 0x38:
869 	case 0x39:
870 	case 0x3a:
871 	case 0x3b:
872 	case 0x3c:
873 	case 0x5f:
874 	case 0x74:
875 	case 0x76:
876 		return handle_io_inst(vcpu);
877 	case 0x56:
878 		return handle_sthyi(vcpu);
879 	case 0x7d:
880 		return handle_stsi(vcpu);
881 	case 0xb1:
882 		return handle_stfl(vcpu);
883 	case 0xb2:
884 		return handle_lpswe(vcpu);
885 	default:
886 		return -EOPNOTSUPP;
887 	}
888 }
889 
890 static int handle_epsw(struct kvm_vcpu *vcpu)
891 {
892 	int reg1, reg2;
893 
894 	vcpu->stat.instruction_epsw++;
895 
896 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
897 
898 	/* This basically extracts the mask half of the psw. */
899 	vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
900 	vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
901 	if (reg2) {
902 		vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
903 		vcpu->run->s.regs.gprs[reg2] |=
904 			vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
905 	}
906 	return 0;
907 }
908 
909 #define PFMF_RESERVED   0xfffc0101UL
910 #define PFMF_SK         0x00020000UL
911 #define PFMF_CF         0x00010000UL
912 #define PFMF_UI         0x00008000UL
913 #define PFMF_FSC        0x00007000UL
914 #define PFMF_NQ         0x00000800UL
915 #define PFMF_MR         0x00000400UL
916 #define PFMF_MC         0x00000200UL
917 #define PFMF_KEY        0x000000feUL
918 
919 static int handle_pfmf(struct kvm_vcpu *vcpu)
920 {
921 	bool mr = false, mc = false, nq;
922 	int reg1, reg2;
923 	unsigned long start, end;
924 	unsigned char key;
925 
926 	vcpu->stat.instruction_pfmf++;
927 
928 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
929 
930 	if (!test_kvm_facility(vcpu->kvm, 8))
931 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
932 
933 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
934 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
935 
936 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
937 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
938 
939 	/* Only provide non-quiescing support if enabled for the guest */
940 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
941 	    !test_kvm_facility(vcpu->kvm, 14))
942 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
943 
944 	/* Only provide conditional-SSKE support if enabled for the guest */
945 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
946 	    test_kvm_facility(vcpu->kvm, 10)) {
947 		mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
948 		mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
949 	}
950 
951 	nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
952 	key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
953 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
954 	start = kvm_s390_logical_to_effective(vcpu, start);
955 
956 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
957 		if (kvm_s390_check_low_addr_prot_real(vcpu, start))
958 			return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
959 	}
960 
961 	switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
962 	case 0x00000000:
963 		/* only 4k frames specify a real address */
964 		start = kvm_s390_real_to_abs(vcpu, start);
965 		end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
966 		break;
967 	case 0x00001000:
968 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
969 		break;
970 	case 0x00002000:
971 		/* only support 2G frame size if EDAT2 is available and we are
972 		   not in 24-bit addressing mode */
973 		if (!test_kvm_facility(vcpu->kvm, 78) ||
974 		    psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
975 			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
976 		end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
977 		break;
978 	default:
979 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
980 	}
981 
982 	while (start != end) {
983 		unsigned long vmaddr;
984 		bool unlocked = false;
985 
986 		/* Translate guest address to host address */
987 		vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
988 		if (kvm_is_error_hva(vmaddr))
989 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
990 
991 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
992 			if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
993 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
994 		}
995 
996 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
997 			int rc = kvm_s390_skey_check_enable(vcpu);
998 
999 			if (rc)
1000 				return rc;
1001 			down_read(&current->mm->mmap_sem);
1002 			rc = cond_set_guest_storage_key(current->mm, vmaddr,
1003 							key, NULL, nq, mr, mc);
1004 			if (rc < 0) {
1005 				rc = fixup_user_fault(current, current->mm, vmaddr,
1006 						      FAULT_FLAG_WRITE, &unlocked);
1007 				rc = !rc ? -EAGAIN : rc;
1008 			}
1009 			up_read(&current->mm->mmap_sem);
1010 			if (rc == -EFAULT)
1011 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1012 			if (rc == -EAGAIN)
1013 				continue;
1014 			if (rc < 0)
1015 				return rc;
1016 		}
1017 		start += PAGE_SIZE;
1018 	}
1019 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1020 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1021 			vcpu->run->s.regs.gprs[reg2] = end;
1022 		} else {
1023 			vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1024 			end = kvm_s390_logical_to_effective(vcpu, end);
1025 			vcpu->run->s.regs.gprs[reg2] |= end;
1026 		}
1027 	}
1028 	return 0;
1029 }
1030 
1031 /*
1032  * Must be called with relevant read locks held (kvm->mm->mmap_sem, kvm->srcu)
1033  */
1034 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1035 {
1036 	int r1, r2, nappended, entries;
1037 	unsigned long gfn, hva, res, pgstev, ptev;
1038 	unsigned long *cbrlo;
1039 
1040 	/*
1041 	 * We don't need to set SD.FPF.SK to 1 here, because if we have a
1042 	 * machine check here we either handle it or crash
1043 	 */
1044 
1045 	kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1046 	gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1047 	hva = gfn_to_hva(vcpu->kvm, gfn);
1048 	entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1049 
1050 	if (kvm_is_error_hva(hva))
1051 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1052 
1053 	nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1054 	if (nappended < 0) {
1055 		res = orc ? 0x10 : 0;
1056 		vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1057 		return 0;
1058 	}
1059 	res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1060 	/*
1061 	 * Set the block-content state part of the result. 0 means resident, so
1062 	 * nothing to do if the page is valid. 2 is for preserved pages
1063 	 * (non-present and non-zero), and 3 for zero pages (non-present and
1064 	 * zero).
1065 	 */
1066 	if (ptev & _PAGE_INVALID) {
1067 		res |= 2;
1068 		if (pgstev & _PGSTE_GPS_ZERO)
1069 			res |= 1;
1070 	}
1071 	if (pgstev & _PGSTE_GPS_NODAT)
1072 		res |= 0x20;
1073 	vcpu->run->s.regs.gprs[r1] = res;
1074 	/*
1075 	 * It is possible that all the normal 511 slots were full, in which case
1076 	 * we will now write in the 512th slot, which is reserved for host use.
1077 	 * In both cases we let the normal essa handling code process all the
1078 	 * slots, including the reserved one, if needed.
1079 	 */
1080 	if (nappended > 0) {
1081 		cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1082 		cbrlo[entries] = gfn << PAGE_SHIFT;
1083 	}
1084 
1085 	if (orc) {
1086 		struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1087 
1088 		/* Increment only if we are really flipping the bit */
1089 		if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1090 			atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1091 	}
1092 
1093 	return nappended;
1094 }
1095 
1096 static int handle_essa(struct kvm_vcpu *vcpu)
1097 {
1098 	/* entries expected to be 1FF */
1099 	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1100 	unsigned long *cbrlo;
1101 	struct gmap *gmap;
1102 	int i, orc;
1103 
1104 	VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1105 	gmap = vcpu->arch.gmap;
1106 	vcpu->stat.instruction_essa++;
1107 	if (!vcpu->kvm->arch.use_cmma)
1108 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1109 
1110 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1111 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1112 	/* Check for invalid operation request code */
1113 	orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1114 	/* ORCs 0-6 are always valid */
1115 	if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1116 						: ESSA_SET_STABLE_IF_RESIDENT))
1117 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1118 
1119 	if (!vcpu->kvm->arch.migration_mode) {
1120 		/*
1121 		 * CMMA is enabled in the KVM settings, but is disabled in
1122 		 * the SIE block and in the mm_context, and we are not doing
1123 		 * a migration. Enable CMMA in the mm_context.
1124 		 * Since we need to take a write lock to write to the context
1125 		 * to avoid races with storage keys handling, we check if the
1126 		 * value really needs to be written to; if the value is
1127 		 * already correct, we do nothing and avoid the lock.
1128 		 */
1129 		if (vcpu->kvm->mm->context.uses_cmm == 0) {
1130 			down_write(&vcpu->kvm->mm->mmap_sem);
1131 			vcpu->kvm->mm->context.uses_cmm = 1;
1132 			up_write(&vcpu->kvm->mm->mmap_sem);
1133 		}
1134 		/*
1135 		 * If we are here, we are supposed to have CMMA enabled in
1136 		 * the SIE block. Enabling CMMA works on a per-CPU basis,
1137 		 * while the context use_cmma flag is per process.
1138 		 * It's possible that the context flag is enabled and the
1139 		 * SIE flag is not, so we set the flag always; if it was
1140 		 * already set, nothing changes, otherwise we enable it
1141 		 * on this CPU too.
1142 		 */
1143 		vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1144 		/* Retry the ESSA instruction */
1145 		kvm_s390_retry_instr(vcpu);
1146 	} else {
1147 		int srcu_idx;
1148 
1149 		down_read(&vcpu->kvm->mm->mmap_sem);
1150 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1151 		i = __do_essa(vcpu, orc);
1152 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1153 		up_read(&vcpu->kvm->mm->mmap_sem);
1154 		if (i < 0)
1155 			return i;
1156 		/* Account for the possible extra cbrl entry */
1157 		entries += i;
1158 	}
1159 	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;	/* reset nceo */
1160 	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1161 	down_read(&gmap->mm->mmap_sem);
1162 	for (i = 0; i < entries; ++i)
1163 		__gmap_zap(gmap, cbrlo[i]);
1164 	up_read(&gmap->mm->mmap_sem);
1165 	return 0;
1166 }
1167 
1168 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1169 {
1170 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1171 	case 0x8a:
1172 	case 0x8e:
1173 	case 0x8f:
1174 		return handle_ipte_interlock(vcpu);
1175 	case 0x8d:
1176 		return handle_epsw(vcpu);
1177 	case 0xab:
1178 		return handle_essa(vcpu);
1179 	case 0xaf:
1180 		return handle_pfmf(vcpu);
1181 	default:
1182 		return -EOPNOTSUPP;
1183 	}
1184 }
1185 
1186 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1187 {
1188 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1189 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1190 	int reg, rc, nr_regs;
1191 	u32 ctl_array[16];
1192 	u64 ga;
1193 	u8 ar;
1194 
1195 	vcpu->stat.instruction_lctl++;
1196 
1197 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1198 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1199 
1200 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1201 
1202 	if (ga & 3)
1203 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1204 
1205 	VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1206 	trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1207 
1208 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1209 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1210 	if (rc)
1211 		return kvm_s390_inject_prog_cond(vcpu, rc);
1212 	reg = reg1;
1213 	nr_regs = 0;
1214 	do {
1215 		vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1216 		vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1217 		if (reg == reg3)
1218 			break;
1219 		reg = (reg + 1) % 16;
1220 	} while (1);
1221 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1222 	return 0;
1223 }
1224 
1225 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1226 {
1227 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1228 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1229 	int reg, rc, nr_regs;
1230 	u32 ctl_array[16];
1231 	u64 ga;
1232 	u8 ar;
1233 
1234 	vcpu->stat.instruction_stctl++;
1235 
1236 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1237 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1238 
1239 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1240 
1241 	if (ga & 3)
1242 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1243 
1244 	VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1245 	trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1246 
1247 	reg = reg1;
1248 	nr_regs = 0;
1249 	do {
1250 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1251 		if (reg == reg3)
1252 			break;
1253 		reg = (reg + 1) % 16;
1254 	} while (1);
1255 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1256 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1257 }
1258 
1259 static int handle_lctlg(struct kvm_vcpu *vcpu)
1260 {
1261 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1262 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1263 	int reg, rc, nr_regs;
1264 	u64 ctl_array[16];
1265 	u64 ga;
1266 	u8 ar;
1267 
1268 	vcpu->stat.instruction_lctlg++;
1269 
1270 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1271 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1272 
1273 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1274 
1275 	if (ga & 7)
1276 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1277 
1278 	VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1279 	trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1280 
1281 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1282 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1283 	if (rc)
1284 		return kvm_s390_inject_prog_cond(vcpu, rc);
1285 	reg = reg1;
1286 	nr_regs = 0;
1287 	do {
1288 		vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1289 		if (reg == reg3)
1290 			break;
1291 		reg = (reg + 1) % 16;
1292 	} while (1);
1293 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1294 	return 0;
1295 }
1296 
1297 static int handle_stctg(struct kvm_vcpu *vcpu)
1298 {
1299 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1300 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1301 	int reg, rc, nr_regs;
1302 	u64 ctl_array[16];
1303 	u64 ga;
1304 	u8 ar;
1305 
1306 	vcpu->stat.instruction_stctg++;
1307 
1308 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1309 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1310 
1311 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1312 
1313 	if (ga & 7)
1314 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1315 
1316 	VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1317 	trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1318 
1319 	reg = reg1;
1320 	nr_regs = 0;
1321 	do {
1322 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1323 		if (reg == reg3)
1324 			break;
1325 		reg = (reg + 1) % 16;
1326 	} while (1);
1327 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1328 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1329 }
1330 
1331 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1332 {
1333 	switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1334 	case 0x25:
1335 		return handle_stctg(vcpu);
1336 	case 0x2f:
1337 		return handle_lctlg(vcpu);
1338 	case 0x60:
1339 	case 0x61:
1340 	case 0x62:
1341 		return handle_ri(vcpu);
1342 	default:
1343 		return -EOPNOTSUPP;
1344 	}
1345 }
1346 
1347 static int handle_tprot(struct kvm_vcpu *vcpu)
1348 {
1349 	u64 address1, address2;
1350 	unsigned long hva, gpa;
1351 	int ret = 0, cc = 0;
1352 	bool writable;
1353 	u8 ar;
1354 
1355 	vcpu->stat.instruction_tprot++;
1356 
1357 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1358 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1359 
1360 	kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1361 
1362 	/* we only handle the Linux memory detection case:
1363 	 * access key == 0
1364 	 * everything else goes to userspace. */
1365 	if (address2 & 0xf0)
1366 		return -EOPNOTSUPP;
1367 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1368 		ipte_lock(vcpu);
1369 	ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1370 	if (ret == PGM_PROTECTION) {
1371 		/* Write protected? Try again with read-only... */
1372 		cc = 1;
1373 		ret = guest_translate_address(vcpu, address1, ar, &gpa,
1374 					      GACC_FETCH);
1375 	}
1376 	if (ret) {
1377 		if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1378 			ret = kvm_s390_inject_program_int(vcpu, ret);
1379 		} else if (ret > 0) {
1380 			/* Translation not available */
1381 			kvm_s390_set_psw_cc(vcpu, 3);
1382 			ret = 0;
1383 		}
1384 		goto out_unlock;
1385 	}
1386 
1387 	hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1388 	if (kvm_is_error_hva(hva)) {
1389 		ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1390 	} else {
1391 		if (!writable)
1392 			cc = 1;		/* Write not permitted ==> read-only */
1393 		kvm_s390_set_psw_cc(vcpu, cc);
1394 		/* Note: CC2 only occurs for storage keys (not supported yet) */
1395 	}
1396 out_unlock:
1397 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1398 		ipte_unlock(vcpu);
1399 	return ret;
1400 }
1401 
1402 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1403 {
1404 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1405 	case 0x01:
1406 		return handle_tprot(vcpu);
1407 	default:
1408 		return -EOPNOTSUPP;
1409 	}
1410 }
1411 
1412 static int handle_sckpf(struct kvm_vcpu *vcpu)
1413 {
1414 	u32 value;
1415 
1416 	vcpu->stat.instruction_sckpf++;
1417 
1418 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1419 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1420 
1421 	if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1422 		return kvm_s390_inject_program_int(vcpu,
1423 						   PGM_SPECIFICATION);
1424 
1425 	value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1426 	vcpu->arch.sie_block->todpr = value;
1427 
1428 	return 0;
1429 }
1430 
1431 static int handle_ptff(struct kvm_vcpu *vcpu)
1432 {
1433 	vcpu->stat.instruction_ptff++;
1434 
1435 	/* we don't emulate any control instructions yet */
1436 	kvm_s390_set_psw_cc(vcpu, 3);
1437 	return 0;
1438 }
1439 
1440 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1441 {
1442 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1443 	case 0x04:
1444 		return handle_ptff(vcpu);
1445 	case 0x07:
1446 		return handle_sckpf(vcpu);
1447 	default:
1448 		return -EOPNOTSUPP;
1449 	}
1450 }
1451