xref: /linux/arch/s390/kvm/priv.c (revision 03c305861c70d6db898dd2379b882e7772a5c5d0)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * handling privileged instructions
4  *
5  * Copyright IBM Corp. 2008, 2020
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/mm_types.h>
15 #include <linux/pgtable.h>
16 #include <linux/io.h>
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/page-states.h>
24 #include <asm/gmap.h>
25 #include <asm/ptrace.h>
26 #include <asm/sclp.h>
27 #include <asm/ap.h>
28 #include "gaccess.h"
29 #include "kvm-s390.h"
30 #include "trace.h"
31 
32 static int handle_ri(struct kvm_vcpu *vcpu)
33 {
34 	vcpu->stat.instruction_ri++;
35 
36 	if (test_kvm_facility(vcpu->kvm, 64)) {
37 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
38 		vcpu->arch.sie_block->ecb3 |= ECB3_RI;
39 		kvm_s390_retry_instr(vcpu);
40 		return 0;
41 	} else
42 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
43 }
44 
45 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
46 {
47 	if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
48 		return handle_ri(vcpu);
49 	else
50 		return -EOPNOTSUPP;
51 }
52 
53 static int handle_gs(struct kvm_vcpu *vcpu)
54 {
55 	vcpu->stat.instruction_gs++;
56 
57 	if (test_kvm_facility(vcpu->kvm, 133)) {
58 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
59 		preempt_disable();
60 		local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
61 		current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
62 		restore_gs_cb(current->thread.gs_cb);
63 		preempt_enable();
64 		vcpu->arch.sie_block->ecb |= ECB_GS;
65 		vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
66 		vcpu->arch.gs_enabled = 1;
67 		kvm_s390_retry_instr(vcpu);
68 		return 0;
69 	} else
70 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
71 }
72 
73 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
74 {
75 	int code = vcpu->arch.sie_block->ipb & 0xff;
76 
77 	if (code == 0x49 || code == 0x4d)
78 		return handle_gs(vcpu);
79 	else
80 		return -EOPNOTSUPP;
81 }
82 /* Handle SCK (SET CLOCK) interception */
83 static int handle_set_clock(struct kvm_vcpu *vcpu)
84 {
85 	struct kvm_s390_vm_tod_clock gtod = { 0 };
86 	int rc;
87 	u8 ar;
88 	u64 op2;
89 
90 	vcpu->stat.instruction_sck++;
91 
92 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
93 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
94 
95 	op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
96 	if (op2 & 7)	/* Operand must be on a doubleword boundary */
97 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
98 	rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
99 	if (rc)
100 		return kvm_s390_inject_prog_cond(vcpu, rc);
101 
102 	VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
103 	/*
104 	 * To set the TOD clock the kvm lock must be taken, but the vcpu lock
105 	 * is already held in handle_set_clock. The usual lock order is the
106 	 * opposite.  As SCK is deprecated and should not be used in several
107 	 * cases, for example when the multiple epoch facility or TOD clock
108 	 * steering facility is installed (see Principles of Operation),  a
109 	 * slow path can be used.  If the lock can not be taken via try_lock,
110 	 * the instruction will be retried via -EAGAIN at a later point in
111 	 * time.
112 	 */
113 	if (!kvm_s390_try_set_tod_clock(vcpu->kvm, &gtod)) {
114 		kvm_s390_retry_instr(vcpu);
115 		return -EAGAIN;
116 	}
117 
118 	kvm_s390_set_psw_cc(vcpu, 0);
119 	return 0;
120 }
121 
122 static int handle_set_prefix(struct kvm_vcpu *vcpu)
123 {
124 	u64 operand2;
125 	u32 address;
126 	int rc;
127 	u8 ar;
128 
129 	vcpu->stat.instruction_spx++;
130 
131 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
132 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
133 
134 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
135 
136 	/* must be word boundary */
137 	if (operand2 & 3)
138 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
139 
140 	/* get the value */
141 	rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
142 	if (rc)
143 		return kvm_s390_inject_prog_cond(vcpu, rc);
144 
145 	address &= 0x7fffe000u;
146 
147 	/*
148 	 * Make sure the new value is valid memory. We only need to check the
149 	 * first page, since address is 8k aligned and memory pieces are always
150 	 * at least 1MB aligned and have at least a size of 1MB.
151 	 */
152 	if (kvm_is_error_gpa(vcpu->kvm, address))
153 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
154 
155 	kvm_s390_set_prefix(vcpu, address);
156 	trace_kvm_s390_handle_prefix(vcpu, 1, address);
157 	return 0;
158 }
159 
160 static int handle_store_prefix(struct kvm_vcpu *vcpu)
161 {
162 	u64 operand2;
163 	u32 address;
164 	int rc;
165 	u8 ar;
166 
167 	vcpu->stat.instruction_stpx++;
168 
169 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
170 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
171 
172 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
173 
174 	/* must be word boundary */
175 	if (operand2 & 3)
176 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
177 
178 	address = kvm_s390_get_prefix(vcpu);
179 
180 	/* get the value */
181 	rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
182 	if (rc)
183 		return kvm_s390_inject_prog_cond(vcpu, rc);
184 
185 	VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
186 	trace_kvm_s390_handle_prefix(vcpu, 0, address);
187 	return 0;
188 }
189 
190 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
191 {
192 	u16 vcpu_id = vcpu->vcpu_id;
193 	u64 ga;
194 	int rc;
195 	u8 ar;
196 
197 	vcpu->stat.instruction_stap++;
198 
199 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
200 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
201 
202 	ga = kvm_s390_get_base_disp_s(vcpu, &ar);
203 
204 	if (ga & 1)
205 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
206 
207 	rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
208 	if (rc)
209 		return kvm_s390_inject_prog_cond(vcpu, rc);
210 
211 	VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
212 	trace_kvm_s390_handle_stap(vcpu, ga);
213 	return 0;
214 }
215 
216 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
217 {
218 	int rc;
219 
220 	trace_kvm_s390_skey_related_inst(vcpu);
221 	/* Already enabled? */
222 	if (vcpu->arch.skey_enabled)
223 		return 0;
224 
225 	rc = s390_enable_skey();
226 	VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
227 	if (rc)
228 		return rc;
229 
230 	if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
231 		kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
232 	if (!vcpu->kvm->arch.use_skf)
233 		vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
234 	else
235 		vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
236 	vcpu->arch.skey_enabled = true;
237 	return 0;
238 }
239 
240 static int try_handle_skey(struct kvm_vcpu *vcpu)
241 {
242 	int rc;
243 
244 	rc = kvm_s390_skey_check_enable(vcpu);
245 	if (rc)
246 		return rc;
247 	if (vcpu->kvm->arch.use_skf) {
248 		/* with storage-key facility, SIE interprets it for us */
249 		kvm_s390_retry_instr(vcpu);
250 		VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
251 		return -EAGAIN;
252 	}
253 	return 0;
254 }
255 
256 static int handle_iske(struct kvm_vcpu *vcpu)
257 {
258 	unsigned long gaddr, vmaddr;
259 	unsigned char key;
260 	int reg1, reg2;
261 	bool unlocked;
262 	int rc;
263 
264 	vcpu->stat.instruction_iske++;
265 
266 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
267 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
268 
269 	rc = try_handle_skey(vcpu);
270 	if (rc)
271 		return rc != -EAGAIN ? rc : 0;
272 
273 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
274 
275 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
276 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
277 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
278 	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
279 	if (kvm_is_error_hva(vmaddr))
280 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
281 retry:
282 	unlocked = false;
283 	mmap_read_lock(current->mm);
284 	rc = get_guest_storage_key(current->mm, vmaddr, &key);
285 
286 	if (rc) {
287 		rc = fixup_user_fault(current->mm, vmaddr,
288 				      FAULT_FLAG_WRITE, &unlocked);
289 		if (!rc) {
290 			mmap_read_unlock(current->mm);
291 			goto retry;
292 		}
293 	}
294 	mmap_read_unlock(current->mm);
295 	if (rc == -EFAULT)
296 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
297 	if (rc < 0)
298 		return rc;
299 	vcpu->run->s.regs.gprs[reg1] &= ~0xff;
300 	vcpu->run->s.regs.gprs[reg1] |= key;
301 	return 0;
302 }
303 
304 static int handle_rrbe(struct kvm_vcpu *vcpu)
305 {
306 	unsigned long vmaddr, gaddr;
307 	int reg1, reg2;
308 	bool unlocked;
309 	int rc;
310 
311 	vcpu->stat.instruction_rrbe++;
312 
313 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
314 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
315 
316 	rc = try_handle_skey(vcpu);
317 	if (rc)
318 		return rc != -EAGAIN ? rc : 0;
319 
320 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
321 
322 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
323 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
324 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
325 	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
326 	if (kvm_is_error_hva(vmaddr))
327 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
328 retry:
329 	unlocked = false;
330 	mmap_read_lock(current->mm);
331 	rc = reset_guest_reference_bit(current->mm, vmaddr);
332 	if (rc < 0) {
333 		rc = fixup_user_fault(current->mm, vmaddr,
334 				      FAULT_FLAG_WRITE, &unlocked);
335 		if (!rc) {
336 			mmap_read_unlock(current->mm);
337 			goto retry;
338 		}
339 	}
340 	mmap_read_unlock(current->mm);
341 	if (rc == -EFAULT)
342 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
343 	if (rc < 0)
344 		return rc;
345 	kvm_s390_set_psw_cc(vcpu, rc);
346 	return 0;
347 }
348 
349 #define SSKE_NQ 0x8
350 #define SSKE_MR 0x4
351 #define SSKE_MC 0x2
352 #define SSKE_MB 0x1
353 static int handle_sske(struct kvm_vcpu *vcpu)
354 {
355 	unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
356 	unsigned long start, end;
357 	unsigned char key, oldkey;
358 	int reg1, reg2;
359 	bool unlocked;
360 	int rc;
361 
362 	vcpu->stat.instruction_sske++;
363 
364 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
365 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
366 
367 	rc = try_handle_skey(vcpu);
368 	if (rc)
369 		return rc != -EAGAIN ? rc : 0;
370 
371 	if (!test_kvm_facility(vcpu->kvm, 8))
372 		m3 &= ~SSKE_MB;
373 	if (!test_kvm_facility(vcpu->kvm, 10))
374 		m3 &= ~(SSKE_MC | SSKE_MR);
375 	if (!test_kvm_facility(vcpu->kvm, 14))
376 		m3 &= ~SSKE_NQ;
377 
378 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
379 
380 	key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
381 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
382 	start = kvm_s390_logical_to_effective(vcpu, start);
383 	if (m3 & SSKE_MB) {
384 		/* start already designates an absolute address */
385 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
386 	} else {
387 		start = kvm_s390_real_to_abs(vcpu, start);
388 		end = start + PAGE_SIZE;
389 	}
390 
391 	while (start != end) {
392 		unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
393 		unlocked = false;
394 
395 		if (kvm_is_error_hva(vmaddr))
396 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
397 
398 		mmap_read_lock(current->mm);
399 		rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
400 						m3 & SSKE_NQ, m3 & SSKE_MR,
401 						m3 & SSKE_MC);
402 
403 		if (rc < 0) {
404 			rc = fixup_user_fault(current->mm, vmaddr,
405 					      FAULT_FLAG_WRITE, &unlocked);
406 			rc = !rc ? -EAGAIN : rc;
407 		}
408 		mmap_read_unlock(current->mm);
409 		if (rc == -EFAULT)
410 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
411 		if (rc == -EAGAIN)
412 			continue;
413 		if (rc < 0)
414 			return rc;
415 		start += PAGE_SIZE;
416 	}
417 
418 	if (m3 & (SSKE_MC | SSKE_MR)) {
419 		if (m3 & SSKE_MB) {
420 			/* skey in reg1 is unpredictable */
421 			kvm_s390_set_psw_cc(vcpu, 3);
422 		} else {
423 			kvm_s390_set_psw_cc(vcpu, rc);
424 			vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
425 			vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
426 		}
427 	}
428 	if (m3 & SSKE_MB) {
429 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
430 			vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
431 		else
432 			vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
433 		end = kvm_s390_logical_to_effective(vcpu, end);
434 		vcpu->run->s.regs.gprs[reg2] |= end;
435 	}
436 	return 0;
437 }
438 
439 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
440 {
441 	vcpu->stat.instruction_ipte_interlock++;
442 	if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
443 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
444 	wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu->kvm));
445 	kvm_s390_retry_instr(vcpu);
446 	VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
447 	return 0;
448 }
449 
450 static int handle_test_block(struct kvm_vcpu *vcpu)
451 {
452 	gpa_t addr;
453 	int reg2;
454 
455 	vcpu->stat.instruction_tb++;
456 
457 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
458 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
459 
460 	kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
461 	addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
462 	addr = kvm_s390_logical_to_effective(vcpu, addr);
463 	if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
464 		return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
465 	addr = kvm_s390_real_to_abs(vcpu, addr);
466 
467 	if (kvm_is_error_gpa(vcpu->kvm, addr))
468 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
469 	/*
470 	 * We don't expect errors on modern systems, and do not care
471 	 * about storage keys (yet), so let's just clear the page.
472 	 */
473 	if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
474 		return -EFAULT;
475 	kvm_s390_set_psw_cc(vcpu, 0);
476 	vcpu->run->s.regs.gprs[0] = 0;
477 	return 0;
478 }
479 
480 static int handle_tpi(struct kvm_vcpu *vcpu)
481 {
482 	struct kvm_s390_interrupt_info *inti;
483 	unsigned long len;
484 	u32 tpi_data[3];
485 	int rc;
486 	u64 addr;
487 	u8 ar;
488 
489 	vcpu->stat.instruction_tpi++;
490 
491 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
492 	if (addr & 3)
493 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
494 
495 	inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
496 	if (!inti) {
497 		kvm_s390_set_psw_cc(vcpu, 0);
498 		return 0;
499 	}
500 
501 	tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
502 	tpi_data[1] = inti->io.io_int_parm;
503 	tpi_data[2] = inti->io.io_int_word;
504 	if (addr) {
505 		/*
506 		 * Store the two-word I/O interruption code into the
507 		 * provided area.
508 		 */
509 		len = sizeof(tpi_data) - 4;
510 		rc = write_guest(vcpu, addr, ar, &tpi_data, len);
511 		if (rc) {
512 			rc = kvm_s390_inject_prog_cond(vcpu, rc);
513 			goto reinject_interrupt;
514 		}
515 	} else {
516 		/*
517 		 * Store the three-word I/O interruption code into
518 		 * the appropriate lowcore area.
519 		 */
520 		len = sizeof(tpi_data);
521 		if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
522 			/* failed writes to the low core are not recoverable */
523 			rc = -EFAULT;
524 			goto reinject_interrupt;
525 		}
526 	}
527 
528 	/* irq was successfully handed to the guest */
529 	kfree(inti);
530 	kvm_s390_set_psw_cc(vcpu, 1);
531 	return 0;
532 reinject_interrupt:
533 	/*
534 	 * If we encounter a problem storing the interruption code, the
535 	 * instruction is suppressed from the guest's view: reinject the
536 	 * interrupt.
537 	 */
538 	if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
539 		kfree(inti);
540 		rc = -EFAULT;
541 	}
542 	/* don't set the cc, a pgm irq was injected or we drop to user space */
543 	return rc ? -EFAULT : 0;
544 }
545 
546 static int handle_tsch(struct kvm_vcpu *vcpu)
547 {
548 	struct kvm_s390_interrupt_info *inti = NULL;
549 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
550 
551 	vcpu->stat.instruction_tsch++;
552 
553 	/* a valid schid has at least one bit set */
554 	if (vcpu->run->s.regs.gprs[1])
555 		inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
556 					   vcpu->run->s.regs.gprs[1]);
557 
558 	/*
559 	 * Prepare exit to userspace.
560 	 * We indicate whether we dequeued a pending I/O interrupt
561 	 * so that userspace can re-inject it if the instruction gets
562 	 * a program check. While this may re-order the pending I/O
563 	 * interrupts, this is no problem since the priority is kept
564 	 * intact.
565 	 */
566 	vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
567 	vcpu->run->s390_tsch.dequeued = !!inti;
568 	if (inti) {
569 		vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
570 		vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
571 		vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
572 		vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
573 	}
574 	vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
575 	kfree(inti);
576 	return -EREMOTE;
577 }
578 
579 static int handle_io_inst(struct kvm_vcpu *vcpu)
580 {
581 	VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
582 
583 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
584 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
585 
586 	if (vcpu->kvm->arch.css_support) {
587 		/*
588 		 * Most I/O instructions will be handled by userspace.
589 		 * Exceptions are tpi and the interrupt portion of tsch.
590 		 */
591 		if (vcpu->arch.sie_block->ipa == 0xb236)
592 			return handle_tpi(vcpu);
593 		if (vcpu->arch.sie_block->ipa == 0xb235)
594 			return handle_tsch(vcpu);
595 		/* Handle in userspace. */
596 		vcpu->stat.instruction_io_other++;
597 		return -EOPNOTSUPP;
598 	} else {
599 		/*
600 		 * Set condition code 3 to stop the guest from issuing channel
601 		 * I/O instructions.
602 		 */
603 		kvm_s390_set_psw_cc(vcpu, 3);
604 		return 0;
605 	}
606 }
607 
608 /*
609  * handle_pqap: Handling pqap interception
610  * @vcpu: the vcpu having issue the pqap instruction
611  *
612  * We now support PQAP/AQIC instructions and we need to correctly
613  * answer the guest even if no dedicated driver's hook is available.
614  *
615  * The intercepting code calls a dedicated callback for this instruction
616  * if a driver did register one in the CRYPTO satellite of the
617  * SIE block.
618  *
619  * If no callback is available, the queues are not available, return this
620  * response code to the caller and set CC to 3.
621  * Else return the response code returned by the callback.
622  */
623 static int handle_pqap(struct kvm_vcpu *vcpu)
624 {
625 	struct ap_queue_status status = {};
626 	crypto_hook pqap_hook;
627 	unsigned long reg0;
628 	int ret;
629 	uint8_t fc;
630 
631 	/* Verify that the AP instruction are available */
632 	if (!ap_instructions_available())
633 		return -EOPNOTSUPP;
634 	/* Verify that the guest is allowed to use AP instructions */
635 	if (!(vcpu->arch.sie_block->eca & ECA_APIE))
636 		return -EOPNOTSUPP;
637 	/*
638 	 * The only possibly intercepted functions when AP instructions are
639 	 * available for the guest are AQIC and TAPQ with the t bit set
640 	 * since we do not set IC.3 (FIII) we currently will only intercept
641 	 * the AQIC function code.
642 	 * Note: running nested under z/VM can result in intercepts for other
643 	 * function codes, e.g. PQAP(QCI). We do not support this and bail out.
644 	 */
645 	reg0 = vcpu->run->s.regs.gprs[0];
646 	fc = (reg0 >> 24) & 0xff;
647 	if (fc != 0x03)
648 		return -EOPNOTSUPP;
649 
650 	/* PQAP instruction is allowed for guest kernel only */
651 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
652 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
653 
654 	/* Common PQAP instruction specification exceptions */
655 	/* bits 41-47 must all be zeros */
656 	if (reg0 & 0x007f0000UL)
657 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
658 	/* APFT not install and T bit set */
659 	if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
660 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
661 	/* APXA not installed and APID greater 64 or APQI greater 16 */
662 	if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
663 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
664 
665 	/* AQIC function code specific exception */
666 	/* facility 65 not present for AQIC function code */
667 	if (!test_kvm_facility(vcpu->kvm, 65))
668 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
669 
670 	/*
671 	 * If the hook callback is registered, there will be a pointer to the
672 	 * hook function pointer in the kvm_s390_crypto structure. Lock the
673 	 * owner, retrieve the hook function pointer and call the hook.
674 	 */
675 	down_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
676 	if (vcpu->kvm->arch.crypto.pqap_hook) {
677 		pqap_hook = *vcpu->kvm->arch.crypto.pqap_hook;
678 		ret = pqap_hook(vcpu);
679 		if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
680 			kvm_s390_set_psw_cc(vcpu, 3);
681 		up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
682 		return ret;
683 	}
684 	up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
685 	/*
686 	 * A vfio_driver must register a hook.
687 	 * No hook means no driver to enable the SIE CRYCB and no queues.
688 	 * We send this response to the guest.
689 	 */
690 	status.response_code = 0x01;
691 	memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
692 	kvm_s390_set_psw_cc(vcpu, 3);
693 	return 0;
694 }
695 
696 static int handle_stfl(struct kvm_vcpu *vcpu)
697 {
698 	int rc;
699 	unsigned int fac;
700 
701 	vcpu->stat.instruction_stfl++;
702 
703 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
704 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
705 
706 	/*
707 	 * We need to shift the lower 32 facility bits (bit 0-31) from a u64
708 	 * into a u32 memory representation. They will remain bits 0-31.
709 	 */
710 	fac = *vcpu->kvm->arch.model.fac_list >> 32;
711 	rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
712 			    &fac, sizeof(fac));
713 	if (rc)
714 		return rc;
715 	VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
716 	trace_kvm_s390_handle_stfl(vcpu, fac);
717 	return 0;
718 }
719 
720 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
721 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
722 #define PSW_ADDR_24 0x0000000000ffffffUL
723 #define PSW_ADDR_31 0x000000007fffffffUL
724 
725 int is_valid_psw(psw_t *psw)
726 {
727 	if (psw->mask & PSW_MASK_UNASSIGNED)
728 		return 0;
729 	if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
730 		if (psw->addr & ~PSW_ADDR_31)
731 			return 0;
732 	}
733 	if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
734 		return 0;
735 	if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
736 		return 0;
737 	if (psw->addr & 1)
738 		return 0;
739 	return 1;
740 }
741 
742 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
743 {
744 	psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
745 	psw_compat_t new_psw;
746 	u64 addr;
747 	int rc;
748 	u8 ar;
749 
750 	vcpu->stat.instruction_lpsw++;
751 
752 	if (gpsw->mask & PSW_MASK_PSTATE)
753 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
754 
755 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
756 	if (addr & 7)
757 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
758 
759 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
760 	if (rc)
761 		return kvm_s390_inject_prog_cond(vcpu, rc);
762 	if (!(new_psw.mask & PSW32_MASK_BASE))
763 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
764 	gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
765 	gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
766 	gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
767 	if (!is_valid_psw(gpsw))
768 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
769 	return 0;
770 }
771 
772 static int handle_lpswe(struct kvm_vcpu *vcpu)
773 {
774 	psw_t new_psw;
775 	u64 addr;
776 	int rc;
777 	u8 ar;
778 
779 	vcpu->stat.instruction_lpswe++;
780 
781 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
782 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
783 
784 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
785 	if (addr & 7)
786 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
787 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
788 	if (rc)
789 		return kvm_s390_inject_prog_cond(vcpu, rc);
790 	vcpu->arch.sie_block->gpsw = new_psw;
791 	if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
792 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
793 	return 0;
794 }
795 
796 static int handle_stidp(struct kvm_vcpu *vcpu)
797 {
798 	u64 stidp_data = vcpu->kvm->arch.model.cpuid;
799 	u64 operand2;
800 	int rc;
801 	u8 ar;
802 
803 	vcpu->stat.instruction_stidp++;
804 
805 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
806 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
807 
808 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
809 
810 	if (operand2 & 7)
811 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
812 
813 	rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
814 	if (rc)
815 		return kvm_s390_inject_prog_cond(vcpu, rc);
816 
817 	VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
818 	return 0;
819 }
820 
821 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
822 {
823 	int cpus = 0;
824 	int n;
825 
826 	cpus = atomic_read(&vcpu->kvm->online_vcpus);
827 
828 	/* deal with other level 3 hypervisors */
829 	if (stsi(mem, 3, 2, 2))
830 		mem->count = 0;
831 	if (mem->count < 8)
832 		mem->count++;
833 	for (n = mem->count - 1; n > 0 ; n--)
834 		memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
835 
836 	memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
837 	mem->vm[0].cpus_total = cpus;
838 	mem->vm[0].cpus_configured = cpus;
839 	mem->vm[0].cpus_standby = 0;
840 	mem->vm[0].cpus_reserved = 0;
841 	mem->vm[0].caf = 1000;
842 	memcpy(mem->vm[0].name, "KVMguest", 8);
843 	ASCEBC(mem->vm[0].name, 8);
844 	memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
845 	ASCEBC(mem->vm[0].cpi, 16);
846 }
847 
848 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
849 				 u8 fc, u8 sel1, u16 sel2)
850 {
851 	vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
852 	vcpu->run->s390_stsi.addr = addr;
853 	vcpu->run->s390_stsi.ar = ar;
854 	vcpu->run->s390_stsi.fc = fc;
855 	vcpu->run->s390_stsi.sel1 = sel1;
856 	vcpu->run->s390_stsi.sel2 = sel2;
857 }
858 
859 static int handle_stsi(struct kvm_vcpu *vcpu)
860 {
861 	int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
862 	int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
863 	int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
864 	unsigned long mem = 0;
865 	u64 operand2;
866 	int rc = 0;
867 	u8 ar;
868 
869 	vcpu->stat.instruction_stsi++;
870 	VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
871 
872 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
873 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
874 
875 	/* Bailout forbidden function codes */
876 	if (fc > 3 && fc != 15)
877 		goto out_no_data;
878 
879 	/*
880 	 * fc 15 is provided only with
881 	 *   - PTF/CPU topology support through facility 15
882 	 *   - KVM_CAP_S390_USER_STSI
883 	 */
884 	if (fc == 15 && (!test_kvm_facility(vcpu->kvm, 11) ||
885 			 !vcpu->kvm->arch.user_stsi))
886 		goto out_no_data;
887 
888 	if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
889 	    || vcpu->run->s.regs.gprs[1] & 0xffff0000)
890 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
891 
892 	if (fc == 0) {
893 		vcpu->run->s.regs.gprs[0] = 3 << 28;
894 		kvm_s390_set_psw_cc(vcpu, 0);
895 		return 0;
896 	}
897 
898 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
899 
900 	if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff))
901 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
902 
903 	switch (fc) {
904 	case 1: /* same handling for 1 and 2 */
905 	case 2:
906 		mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
907 		if (!mem)
908 			goto out_no_data;
909 		if (stsi((void *) mem, fc, sel1, sel2))
910 			goto out_no_data;
911 		break;
912 	case 3:
913 		if (sel1 != 2 || sel2 != 2)
914 			goto out_no_data;
915 		mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
916 		if (!mem)
917 			goto out_no_data;
918 		handle_stsi_3_2_2(vcpu, (void *) mem);
919 		break;
920 	case 15: /* fc 15 is fully handled in userspace */
921 		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
922 		trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
923 		return -EREMOTE;
924 	}
925 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
926 		memcpy(sida_addr(vcpu->arch.sie_block), (void *)mem, PAGE_SIZE);
927 		rc = 0;
928 	} else {
929 		rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
930 	}
931 	if (rc) {
932 		rc = kvm_s390_inject_prog_cond(vcpu, rc);
933 		goto out;
934 	}
935 	if (vcpu->kvm->arch.user_stsi) {
936 		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
937 		rc = -EREMOTE;
938 	}
939 	trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
940 	free_page(mem);
941 	kvm_s390_set_psw_cc(vcpu, 0);
942 	vcpu->run->s.regs.gprs[0] = 0;
943 	return rc;
944 out_no_data:
945 	kvm_s390_set_psw_cc(vcpu, 3);
946 out:
947 	free_page(mem);
948 	return rc;
949 }
950 
951 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
952 {
953 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
954 	case 0x02:
955 		return handle_stidp(vcpu);
956 	case 0x04:
957 		return handle_set_clock(vcpu);
958 	case 0x10:
959 		return handle_set_prefix(vcpu);
960 	case 0x11:
961 		return handle_store_prefix(vcpu);
962 	case 0x12:
963 		return handle_store_cpu_address(vcpu);
964 	case 0x14:
965 		return kvm_s390_handle_vsie(vcpu);
966 	case 0x21:
967 	case 0x50:
968 		return handle_ipte_interlock(vcpu);
969 	case 0x29:
970 		return handle_iske(vcpu);
971 	case 0x2a:
972 		return handle_rrbe(vcpu);
973 	case 0x2b:
974 		return handle_sske(vcpu);
975 	case 0x2c:
976 		return handle_test_block(vcpu);
977 	case 0x30:
978 	case 0x31:
979 	case 0x32:
980 	case 0x33:
981 	case 0x34:
982 	case 0x35:
983 	case 0x36:
984 	case 0x37:
985 	case 0x38:
986 	case 0x39:
987 	case 0x3a:
988 	case 0x3b:
989 	case 0x3c:
990 	case 0x5f:
991 	case 0x74:
992 	case 0x76:
993 		return handle_io_inst(vcpu);
994 	case 0x56:
995 		return handle_sthyi(vcpu);
996 	case 0x7d:
997 		return handle_stsi(vcpu);
998 	case 0xaf:
999 		return handle_pqap(vcpu);
1000 	case 0xb1:
1001 		return handle_stfl(vcpu);
1002 	case 0xb2:
1003 		return handle_lpswe(vcpu);
1004 	default:
1005 		return -EOPNOTSUPP;
1006 	}
1007 }
1008 
1009 static int handle_epsw(struct kvm_vcpu *vcpu)
1010 {
1011 	int reg1, reg2;
1012 
1013 	vcpu->stat.instruction_epsw++;
1014 
1015 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1016 
1017 	/* This basically extracts the mask half of the psw. */
1018 	vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
1019 	vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
1020 	if (reg2) {
1021 		vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
1022 		vcpu->run->s.regs.gprs[reg2] |=
1023 			vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
1024 	}
1025 	return 0;
1026 }
1027 
1028 #define PFMF_RESERVED   0xfffc0101UL
1029 #define PFMF_SK         0x00020000UL
1030 #define PFMF_CF         0x00010000UL
1031 #define PFMF_UI         0x00008000UL
1032 #define PFMF_FSC        0x00007000UL
1033 #define PFMF_NQ         0x00000800UL
1034 #define PFMF_MR         0x00000400UL
1035 #define PFMF_MC         0x00000200UL
1036 #define PFMF_KEY        0x000000feUL
1037 
1038 static int handle_pfmf(struct kvm_vcpu *vcpu)
1039 {
1040 	bool mr = false, mc = false, nq;
1041 	int reg1, reg2;
1042 	unsigned long start, end;
1043 	unsigned char key;
1044 
1045 	vcpu->stat.instruction_pfmf++;
1046 
1047 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1048 
1049 	if (!test_kvm_facility(vcpu->kvm, 8))
1050 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1051 
1052 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1053 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1054 
1055 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1056 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1057 
1058 	/* Only provide non-quiescing support if enabled for the guest */
1059 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1060 	    !test_kvm_facility(vcpu->kvm, 14))
1061 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1062 
1063 	/* Only provide conditional-SSKE support if enabled for the guest */
1064 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1065 	    test_kvm_facility(vcpu->kvm, 10)) {
1066 		mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1067 		mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1068 	}
1069 
1070 	nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1071 	key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1072 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1073 	start = kvm_s390_logical_to_effective(vcpu, start);
1074 
1075 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1076 		if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1077 			return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1078 	}
1079 
1080 	switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1081 	case 0x00000000:
1082 		/* only 4k frames specify a real address */
1083 		start = kvm_s390_real_to_abs(vcpu, start);
1084 		end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1085 		break;
1086 	case 0x00001000:
1087 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1088 		break;
1089 	case 0x00002000:
1090 		/* only support 2G frame size if EDAT2 is available and we are
1091 		   not in 24-bit addressing mode */
1092 		if (!test_kvm_facility(vcpu->kvm, 78) ||
1093 		    psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1094 			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1095 		end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1096 		break;
1097 	default:
1098 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1099 	}
1100 
1101 	while (start != end) {
1102 		unsigned long vmaddr;
1103 		bool unlocked = false;
1104 
1105 		/* Translate guest address to host address */
1106 		vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
1107 		if (kvm_is_error_hva(vmaddr))
1108 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1109 
1110 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1111 			if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1112 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1113 		}
1114 
1115 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1116 			int rc = kvm_s390_skey_check_enable(vcpu);
1117 
1118 			if (rc)
1119 				return rc;
1120 			mmap_read_lock(current->mm);
1121 			rc = cond_set_guest_storage_key(current->mm, vmaddr,
1122 							key, NULL, nq, mr, mc);
1123 			if (rc < 0) {
1124 				rc = fixup_user_fault(current->mm, vmaddr,
1125 						      FAULT_FLAG_WRITE, &unlocked);
1126 				rc = !rc ? -EAGAIN : rc;
1127 			}
1128 			mmap_read_unlock(current->mm);
1129 			if (rc == -EFAULT)
1130 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1131 			if (rc == -EAGAIN)
1132 				continue;
1133 			if (rc < 0)
1134 				return rc;
1135 		}
1136 		start += PAGE_SIZE;
1137 	}
1138 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1139 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1140 			vcpu->run->s.regs.gprs[reg2] = end;
1141 		} else {
1142 			vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1143 			end = kvm_s390_logical_to_effective(vcpu, end);
1144 			vcpu->run->s.regs.gprs[reg2] |= end;
1145 		}
1146 	}
1147 	return 0;
1148 }
1149 
1150 /*
1151  * Must be called with relevant read locks held (kvm->mm->mmap_lock, kvm->srcu)
1152  */
1153 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1154 {
1155 	int r1, r2, nappended, entries;
1156 	unsigned long gfn, hva, res, pgstev, ptev;
1157 	unsigned long *cbrlo;
1158 
1159 	/*
1160 	 * We don't need to set SD.FPF.SK to 1 here, because if we have a
1161 	 * machine check here we either handle it or crash
1162 	 */
1163 
1164 	kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1165 	gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1166 	hva = gfn_to_hva(vcpu->kvm, gfn);
1167 	entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1168 
1169 	if (kvm_is_error_hva(hva))
1170 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1171 
1172 	nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1173 	if (nappended < 0) {
1174 		res = orc ? 0x10 : 0;
1175 		vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1176 		return 0;
1177 	}
1178 	res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1179 	/*
1180 	 * Set the block-content state part of the result. 0 means resident, so
1181 	 * nothing to do if the page is valid. 2 is for preserved pages
1182 	 * (non-present and non-zero), and 3 for zero pages (non-present and
1183 	 * zero).
1184 	 */
1185 	if (ptev & _PAGE_INVALID) {
1186 		res |= 2;
1187 		if (pgstev & _PGSTE_GPS_ZERO)
1188 			res |= 1;
1189 	}
1190 	if (pgstev & _PGSTE_GPS_NODAT)
1191 		res |= 0x20;
1192 	vcpu->run->s.regs.gprs[r1] = res;
1193 	/*
1194 	 * It is possible that all the normal 511 slots were full, in which case
1195 	 * we will now write in the 512th slot, which is reserved for host use.
1196 	 * In both cases we let the normal essa handling code process all the
1197 	 * slots, including the reserved one, if needed.
1198 	 */
1199 	if (nappended > 0) {
1200 		cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1201 		cbrlo[entries] = gfn << PAGE_SHIFT;
1202 	}
1203 
1204 	if (orc) {
1205 		struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1206 
1207 		/* Increment only if we are really flipping the bit */
1208 		if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1209 			atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1210 	}
1211 
1212 	return nappended;
1213 }
1214 
1215 static int handle_essa(struct kvm_vcpu *vcpu)
1216 {
1217 	/* entries expected to be 1FF */
1218 	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1219 	unsigned long *cbrlo;
1220 	struct gmap *gmap;
1221 	int i, orc;
1222 
1223 	VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1224 	gmap = vcpu->arch.gmap;
1225 	vcpu->stat.instruction_essa++;
1226 	if (!vcpu->kvm->arch.use_cmma)
1227 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1228 
1229 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1230 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1231 	/* Check for invalid operation request code */
1232 	orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1233 	/* ORCs 0-6 are always valid */
1234 	if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1235 						: ESSA_SET_STABLE_IF_RESIDENT))
1236 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1237 
1238 	if (!vcpu->kvm->arch.migration_mode) {
1239 		/*
1240 		 * CMMA is enabled in the KVM settings, but is disabled in
1241 		 * the SIE block and in the mm_context, and we are not doing
1242 		 * a migration. Enable CMMA in the mm_context.
1243 		 * Since we need to take a write lock to write to the context
1244 		 * to avoid races with storage keys handling, we check if the
1245 		 * value really needs to be written to; if the value is
1246 		 * already correct, we do nothing and avoid the lock.
1247 		 */
1248 		if (vcpu->kvm->mm->context.uses_cmm == 0) {
1249 			mmap_write_lock(vcpu->kvm->mm);
1250 			vcpu->kvm->mm->context.uses_cmm = 1;
1251 			mmap_write_unlock(vcpu->kvm->mm);
1252 		}
1253 		/*
1254 		 * If we are here, we are supposed to have CMMA enabled in
1255 		 * the SIE block. Enabling CMMA works on a per-CPU basis,
1256 		 * while the context use_cmma flag is per process.
1257 		 * It's possible that the context flag is enabled and the
1258 		 * SIE flag is not, so we set the flag always; if it was
1259 		 * already set, nothing changes, otherwise we enable it
1260 		 * on this CPU too.
1261 		 */
1262 		vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1263 		/* Retry the ESSA instruction */
1264 		kvm_s390_retry_instr(vcpu);
1265 	} else {
1266 		int srcu_idx;
1267 
1268 		mmap_read_lock(vcpu->kvm->mm);
1269 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1270 		i = __do_essa(vcpu, orc);
1271 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1272 		mmap_read_unlock(vcpu->kvm->mm);
1273 		if (i < 0)
1274 			return i;
1275 		/* Account for the possible extra cbrl entry */
1276 		entries += i;
1277 	}
1278 	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;	/* reset nceo */
1279 	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1280 	mmap_read_lock(gmap->mm);
1281 	for (i = 0; i < entries; ++i)
1282 		__gmap_zap(gmap, cbrlo[i]);
1283 	mmap_read_unlock(gmap->mm);
1284 	return 0;
1285 }
1286 
1287 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1288 {
1289 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1290 	case 0x8a:
1291 	case 0x8e:
1292 	case 0x8f:
1293 		return handle_ipte_interlock(vcpu);
1294 	case 0x8d:
1295 		return handle_epsw(vcpu);
1296 	case 0xab:
1297 		return handle_essa(vcpu);
1298 	case 0xaf:
1299 		return handle_pfmf(vcpu);
1300 	default:
1301 		return -EOPNOTSUPP;
1302 	}
1303 }
1304 
1305 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1306 {
1307 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1308 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1309 	int reg, rc, nr_regs;
1310 	u32 ctl_array[16];
1311 	u64 ga;
1312 	u8 ar;
1313 
1314 	vcpu->stat.instruction_lctl++;
1315 
1316 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1317 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1318 
1319 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1320 
1321 	if (ga & 3)
1322 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1323 
1324 	VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1325 	trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1326 
1327 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1328 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1329 	if (rc)
1330 		return kvm_s390_inject_prog_cond(vcpu, rc);
1331 	reg = reg1;
1332 	nr_regs = 0;
1333 	do {
1334 		vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1335 		vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1336 		if (reg == reg3)
1337 			break;
1338 		reg = (reg + 1) % 16;
1339 	} while (1);
1340 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1341 	return 0;
1342 }
1343 
1344 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1345 {
1346 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1347 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1348 	int reg, rc, nr_regs;
1349 	u32 ctl_array[16];
1350 	u64 ga;
1351 	u8 ar;
1352 
1353 	vcpu->stat.instruction_stctl++;
1354 
1355 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1356 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1357 
1358 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1359 
1360 	if (ga & 3)
1361 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1362 
1363 	VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1364 	trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1365 
1366 	reg = reg1;
1367 	nr_regs = 0;
1368 	do {
1369 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1370 		if (reg == reg3)
1371 			break;
1372 		reg = (reg + 1) % 16;
1373 	} while (1);
1374 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1375 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1376 }
1377 
1378 static int handle_lctlg(struct kvm_vcpu *vcpu)
1379 {
1380 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1381 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1382 	int reg, rc, nr_regs;
1383 	u64 ctl_array[16];
1384 	u64 ga;
1385 	u8 ar;
1386 
1387 	vcpu->stat.instruction_lctlg++;
1388 
1389 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1390 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1391 
1392 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1393 
1394 	if (ga & 7)
1395 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1396 
1397 	VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1398 	trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1399 
1400 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1401 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1402 	if (rc)
1403 		return kvm_s390_inject_prog_cond(vcpu, rc);
1404 	reg = reg1;
1405 	nr_regs = 0;
1406 	do {
1407 		vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1408 		if (reg == reg3)
1409 			break;
1410 		reg = (reg + 1) % 16;
1411 	} while (1);
1412 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1413 	return 0;
1414 }
1415 
1416 static int handle_stctg(struct kvm_vcpu *vcpu)
1417 {
1418 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1419 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1420 	int reg, rc, nr_regs;
1421 	u64 ctl_array[16];
1422 	u64 ga;
1423 	u8 ar;
1424 
1425 	vcpu->stat.instruction_stctg++;
1426 
1427 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1428 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1429 
1430 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1431 
1432 	if (ga & 7)
1433 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1434 
1435 	VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1436 	trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1437 
1438 	reg = reg1;
1439 	nr_regs = 0;
1440 	do {
1441 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1442 		if (reg == reg3)
1443 			break;
1444 		reg = (reg + 1) % 16;
1445 	} while (1);
1446 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1447 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1448 }
1449 
1450 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1451 {
1452 	switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1453 	case 0x25:
1454 		return handle_stctg(vcpu);
1455 	case 0x2f:
1456 		return handle_lctlg(vcpu);
1457 	case 0x60:
1458 	case 0x61:
1459 	case 0x62:
1460 		return handle_ri(vcpu);
1461 	default:
1462 		return -EOPNOTSUPP;
1463 	}
1464 }
1465 
1466 static int handle_tprot(struct kvm_vcpu *vcpu)
1467 {
1468 	u64 address, operand2;
1469 	unsigned long gpa;
1470 	u8 access_key;
1471 	bool writable;
1472 	int ret, cc;
1473 	u8 ar;
1474 
1475 	vcpu->stat.instruction_tprot++;
1476 
1477 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1478 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1479 
1480 	kvm_s390_get_base_disp_sse(vcpu, &address, &operand2, &ar, NULL);
1481 	access_key = (operand2 & 0xf0) >> 4;
1482 
1483 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1484 		ipte_lock(vcpu->kvm);
1485 
1486 	ret = guest_translate_address_with_key(vcpu, address, ar, &gpa,
1487 					       GACC_STORE, access_key);
1488 	if (ret == 0) {
1489 		gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1490 	} else if (ret == PGM_PROTECTION) {
1491 		writable = false;
1492 		/* Write protected? Try again with read-only... */
1493 		ret = guest_translate_address_with_key(vcpu, address, ar, &gpa,
1494 						       GACC_FETCH, access_key);
1495 	}
1496 	if (ret >= 0) {
1497 		cc = -1;
1498 
1499 		/* Fetching permitted; storing permitted */
1500 		if (ret == 0 && writable)
1501 			cc = 0;
1502 		/* Fetching permitted; storing not permitted */
1503 		else if (ret == 0 && !writable)
1504 			cc = 1;
1505 		/* Fetching not permitted; storing not permitted */
1506 		else if (ret == PGM_PROTECTION)
1507 			cc = 2;
1508 		/* Translation not available */
1509 		else if (ret != PGM_ADDRESSING && ret != PGM_TRANSLATION_SPEC)
1510 			cc = 3;
1511 
1512 		if (cc != -1) {
1513 			kvm_s390_set_psw_cc(vcpu, cc);
1514 			ret = 0;
1515 		} else {
1516 			ret = kvm_s390_inject_program_int(vcpu, ret);
1517 		}
1518 	}
1519 
1520 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1521 		ipte_unlock(vcpu->kvm);
1522 	return ret;
1523 }
1524 
1525 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1526 {
1527 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1528 	case 0x01:
1529 		return handle_tprot(vcpu);
1530 	default:
1531 		return -EOPNOTSUPP;
1532 	}
1533 }
1534 
1535 static int handle_sckpf(struct kvm_vcpu *vcpu)
1536 {
1537 	u32 value;
1538 
1539 	vcpu->stat.instruction_sckpf++;
1540 
1541 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1542 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1543 
1544 	if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1545 		return kvm_s390_inject_program_int(vcpu,
1546 						   PGM_SPECIFICATION);
1547 
1548 	value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1549 	vcpu->arch.sie_block->todpr = value;
1550 
1551 	return 0;
1552 }
1553 
1554 static int handle_ptff(struct kvm_vcpu *vcpu)
1555 {
1556 	vcpu->stat.instruction_ptff++;
1557 
1558 	/* we don't emulate any control instructions yet */
1559 	kvm_s390_set_psw_cc(vcpu, 3);
1560 	return 0;
1561 }
1562 
1563 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1564 {
1565 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1566 	case 0x04:
1567 		return handle_ptff(vcpu);
1568 	case 0x07:
1569 		return handle_sckpf(vcpu);
1570 	default:
1571 		return -EOPNOTSUPP;
1572 	}
1573 }
1574