xref: /openbmc/linux/arch/s390/kvm/priv.c (revision 022dacdd)
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/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 <asm/ap.h>
31 #include "gaccess.h"
32 #include "kvm-s390.h"
33 #include "trace.h"
34 
35 static int handle_ri(struct kvm_vcpu *vcpu)
36 {
37 	vcpu->stat.instruction_ri++;
38 
39 	if (test_kvm_facility(vcpu->kvm, 64)) {
40 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
41 		vcpu->arch.sie_block->ecb3 |= ECB3_RI;
42 		kvm_s390_retry_instr(vcpu);
43 		return 0;
44 	} else
45 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
46 }
47 
48 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
49 {
50 	if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
51 		return handle_ri(vcpu);
52 	else
53 		return -EOPNOTSUPP;
54 }
55 
56 static int handle_gs(struct kvm_vcpu *vcpu)
57 {
58 	vcpu->stat.instruction_gs++;
59 
60 	if (test_kvm_facility(vcpu->kvm, 133)) {
61 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
62 		preempt_disable();
63 		__ctl_set_bit(2, 4);
64 		current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
65 		restore_gs_cb(current->thread.gs_cb);
66 		preempt_enable();
67 		vcpu->arch.sie_block->ecb |= ECB_GS;
68 		vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
69 		vcpu->arch.gs_enabled = 1;
70 		kvm_s390_retry_instr(vcpu);
71 		return 0;
72 	} else
73 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
74 }
75 
76 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
77 {
78 	int code = vcpu->arch.sie_block->ipb & 0xff;
79 
80 	if (code == 0x49 || code == 0x4d)
81 		return handle_gs(vcpu);
82 	else
83 		return -EOPNOTSUPP;
84 }
85 /* Handle SCK (SET CLOCK) interception */
86 static int handle_set_clock(struct kvm_vcpu *vcpu)
87 {
88 	struct kvm_s390_vm_tod_clock gtod = { 0 };
89 	int rc;
90 	u8 ar;
91 	u64 op2;
92 
93 	vcpu->stat.instruction_sck++;
94 
95 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
96 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
97 
98 	op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
99 	if (op2 & 7)	/* Operand must be on a doubleword boundary */
100 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
101 	rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
102 	if (rc)
103 		return kvm_s390_inject_prog_cond(vcpu, rc);
104 
105 	VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
106 	kvm_s390_set_tod_clock(vcpu->kvm, &gtod);
107 
108 	kvm_s390_set_psw_cc(vcpu, 0);
109 	return 0;
110 }
111 
112 static int handle_set_prefix(struct kvm_vcpu *vcpu)
113 {
114 	u64 operand2;
115 	u32 address;
116 	int rc;
117 	u8 ar;
118 
119 	vcpu->stat.instruction_spx++;
120 
121 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
122 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
123 
124 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
125 
126 	/* must be word boundary */
127 	if (operand2 & 3)
128 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
129 
130 	/* get the value */
131 	rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
132 	if (rc)
133 		return kvm_s390_inject_prog_cond(vcpu, rc);
134 
135 	address &= 0x7fffe000u;
136 
137 	/*
138 	 * Make sure the new value is valid memory. We only need to check the
139 	 * first page, since address is 8k aligned and memory pieces are always
140 	 * at least 1MB aligned and have at least a size of 1MB.
141 	 */
142 	if (kvm_is_error_gpa(vcpu->kvm, address))
143 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
144 
145 	kvm_s390_set_prefix(vcpu, address);
146 	trace_kvm_s390_handle_prefix(vcpu, 1, address);
147 	return 0;
148 }
149 
150 static int handle_store_prefix(struct kvm_vcpu *vcpu)
151 {
152 	u64 operand2;
153 	u32 address;
154 	int rc;
155 	u8 ar;
156 
157 	vcpu->stat.instruction_stpx++;
158 
159 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
160 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
161 
162 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
163 
164 	/* must be word boundary */
165 	if (operand2 & 3)
166 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
167 
168 	address = kvm_s390_get_prefix(vcpu);
169 
170 	/* get the value */
171 	rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
172 	if (rc)
173 		return kvm_s390_inject_prog_cond(vcpu, rc);
174 
175 	VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
176 	trace_kvm_s390_handle_prefix(vcpu, 0, address);
177 	return 0;
178 }
179 
180 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
181 {
182 	u16 vcpu_id = vcpu->vcpu_id;
183 	u64 ga;
184 	int rc;
185 	u8 ar;
186 
187 	vcpu->stat.instruction_stap++;
188 
189 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
190 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
191 
192 	ga = kvm_s390_get_base_disp_s(vcpu, &ar);
193 
194 	if (ga & 1)
195 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
196 
197 	rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
198 	if (rc)
199 		return kvm_s390_inject_prog_cond(vcpu, rc);
200 
201 	VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
202 	trace_kvm_s390_handle_stap(vcpu, ga);
203 	return 0;
204 }
205 
206 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
207 {
208 	int rc;
209 
210 	trace_kvm_s390_skey_related_inst(vcpu);
211 	/* Already enabled? */
212 	if (vcpu->arch.skey_enabled)
213 		return 0;
214 
215 	rc = s390_enable_skey();
216 	VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
217 	if (rc)
218 		return rc;
219 
220 	if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
221 		kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
222 	if (!vcpu->kvm->arch.use_skf)
223 		vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
224 	else
225 		vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
226 	vcpu->arch.skey_enabled = true;
227 	return 0;
228 }
229 
230 static int try_handle_skey(struct kvm_vcpu *vcpu)
231 {
232 	int rc;
233 
234 	rc = kvm_s390_skey_check_enable(vcpu);
235 	if (rc)
236 		return rc;
237 	if (vcpu->kvm->arch.use_skf) {
238 		/* with storage-key facility, SIE interprets it for us */
239 		kvm_s390_retry_instr(vcpu);
240 		VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
241 		return -EAGAIN;
242 	}
243 	return 0;
244 }
245 
246 static int handle_iske(struct kvm_vcpu *vcpu)
247 {
248 	unsigned long gaddr, vmaddr;
249 	unsigned char key;
250 	int reg1, reg2;
251 	bool unlocked;
252 	int rc;
253 
254 	vcpu->stat.instruction_iske++;
255 
256 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
257 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
258 
259 	rc = try_handle_skey(vcpu);
260 	if (rc)
261 		return rc != -EAGAIN ? rc : 0;
262 
263 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
264 
265 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
266 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
267 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
268 	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
269 	if (kvm_is_error_hva(vmaddr))
270 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
271 retry:
272 	unlocked = false;
273 	down_read(&current->mm->mmap_sem);
274 	rc = get_guest_storage_key(current->mm, vmaddr, &key);
275 
276 	if (rc) {
277 		rc = fixup_user_fault(current, current->mm, vmaddr,
278 				      FAULT_FLAG_WRITE, &unlocked);
279 		if (!rc) {
280 			up_read(&current->mm->mmap_sem);
281 			goto retry;
282 		}
283 	}
284 	up_read(&current->mm->mmap_sem);
285 	if (rc == -EFAULT)
286 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
287 	if (rc < 0)
288 		return rc;
289 	vcpu->run->s.regs.gprs[reg1] &= ~0xff;
290 	vcpu->run->s.regs.gprs[reg1] |= key;
291 	return 0;
292 }
293 
294 static int handle_rrbe(struct kvm_vcpu *vcpu)
295 {
296 	unsigned long vmaddr, gaddr;
297 	int reg1, reg2;
298 	bool unlocked;
299 	int rc;
300 
301 	vcpu->stat.instruction_rrbe++;
302 
303 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
304 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
305 
306 	rc = try_handle_skey(vcpu);
307 	if (rc)
308 		return rc != -EAGAIN ? rc : 0;
309 
310 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
311 
312 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
313 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
314 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
315 	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
316 	if (kvm_is_error_hva(vmaddr))
317 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
318 retry:
319 	unlocked = false;
320 	down_read(&current->mm->mmap_sem);
321 	rc = reset_guest_reference_bit(current->mm, vmaddr);
322 	if (rc < 0) {
323 		rc = fixup_user_fault(current, current->mm, vmaddr,
324 				      FAULT_FLAG_WRITE, &unlocked);
325 		if (!rc) {
326 			up_read(&current->mm->mmap_sem);
327 			goto retry;
328 		}
329 	}
330 	up_read(&current->mm->mmap_sem);
331 	if (rc == -EFAULT)
332 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
333 	if (rc < 0)
334 		return rc;
335 	kvm_s390_set_psw_cc(vcpu, rc);
336 	return 0;
337 }
338 
339 #define SSKE_NQ 0x8
340 #define SSKE_MR 0x4
341 #define SSKE_MC 0x2
342 #define SSKE_MB 0x1
343 static int handle_sske(struct kvm_vcpu *vcpu)
344 {
345 	unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
346 	unsigned long start, end;
347 	unsigned char key, oldkey;
348 	int reg1, reg2;
349 	bool unlocked;
350 	int rc;
351 
352 	vcpu->stat.instruction_sske++;
353 
354 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
355 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
356 
357 	rc = try_handle_skey(vcpu);
358 	if (rc)
359 		return rc != -EAGAIN ? rc : 0;
360 
361 	if (!test_kvm_facility(vcpu->kvm, 8))
362 		m3 &= ~SSKE_MB;
363 	if (!test_kvm_facility(vcpu->kvm, 10))
364 		m3 &= ~(SSKE_MC | SSKE_MR);
365 	if (!test_kvm_facility(vcpu->kvm, 14))
366 		m3 &= ~SSKE_NQ;
367 
368 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
369 
370 	key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
371 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
372 	start = kvm_s390_logical_to_effective(vcpu, start);
373 	if (m3 & SSKE_MB) {
374 		/* start already designates an absolute address */
375 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
376 	} else {
377 		start = kvm_s390_real_to_abs(vcpu, start);
378 		end = start + PAGE_SIZE;
379 	}
380 
381 	while (start != end) {
382 		unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
383 		unlocked = false;
384 
385 		if (kvm_is_error_hva(vmaddr))
386 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
387 
388 		down_read(&current->mm->mmap_sem);
389 		rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
390 						m3 & SSKE_NQ, m3 & SSKE_MR,
391 						m3 & SSKE_MC);
392 
393 		if (rc < 0) {
394 			rc = fixup_user_fault(current, current->mm, vmaddr,
395 					      FAULT_FLAG_WRITE, &unlocked);
396 			rc = !rc ? -EAGAIN : rc;
397 		}
398 		up_read(&current->mm->mmap_sem);
399 		if (rc == -EFAULT)
400 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
401 		if (rc < 0)
402 			return rc;
403 		start += PAGE_SIZE;
404 	}
405 
406 	if (m3 & (SSKE_MC | SSKE_MR)) {
407 		if (m3 & SSKE_MB) {
408 			/* skey in reg1 is unpredictable */
409 			kvm_s390_set_psw_cc(vcpu, 3);
410 		} else {
411 			kvm_s390_set_psw_cc(vcpu, rc);
412 			vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
413 			vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
414 		}
415 	}
416 	if (m3 & SSKE_MB) {
417 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
418 			vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
419 		else
420 			vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
421 		end = kvm_s390_logical_to_effective(vcpu, end);
422 		vcpu->run->s.regs.gprs[reg2] |= end;
423 	}
424 	return 0;
425 }
426 
427 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
428 {
429 	vcpu->stat.instruction_ipte_interlock++;
430 	if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
431 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
432 	wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
433 	kvm_s390_retry_instr(vcpu);
434 	VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
435 	return 0;
436 }
437 
438 static int handle_test_block(struct kvm_vcpu *vcpu)
439 {
440 	gpa_t addr;
441 	int reg2;
442 
443 	vcpu->stat.instruction_tb++;
444 
445 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
446 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
447 
448 	kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
449 	addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
450 	addr = kvm_s390_logical_to_effective(vcpu, addr);
451 	if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
452 		return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
453 	addr = kvm_s390_real_to_abs(vcpu, addr);
454 
455 	if (kvm_is_error_gpa(vcpu->kvm, addr))
456 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
457 	/*
458 	 * We don't expect errors on modern systems, and do not care
459 	 * about storage keys (yet), so let's just clear the page.
460 	 */
461 	if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
462 		return -EFAULT;
463 	kvm_s390_set_psw_cc(vcpu, 0);
464 	vcpu->run->s.regs.gprs[0] = 0;
465 	return 0;
466 }
467 
468 static int handle_tpi(struct kvm_vcpu *vcpu)
469 {
470 	struct kvm_s390_interrupt_info *inti;
471 	unsigned long len;
472 	u32 tpi_data[3];
473 	int rc;
474 	u64 addr;
475 	u8 ar;
476 
477 	vcpu->stat.instruction_tpi++;
478 
479 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
480 	if (addr & 3)
481 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
482 
483 	inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
484 	if (!inti) {
485 		kvm_s390_set_psw_cc(vcpu, 0);
486 		return 0;
487 	}
488 
489 	tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
490 	tpi_data[1] = inti->io.io_int_parm;
491 	tpi_data[2] = inti->io.io_int_word;
492 	if (addr) {
493 		/*
494 		 * Store the two-word I/O interruption code into the
495 		 * provided area.
496 		 */
497 		len = sizeof(tpi_data) - 4;
498 		rc = write_guest(vcpu, addr, ar, &tpi_data, len);
499 		if (rc) {
500 			rc = kvm_s390_inject_prog_cond(vcpu, rc);
501 			goto reinject_interrupt;
502 		}
503 	} else {
504 		/*
505 		 * Store the three-word I/O interruption code into
506 		 * the appropriate lowcore area.
507 		 */
508 		len = sizeof(tpi_data);
509 		if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
510 			/* failed writes to the low core are not recoverable */
511 			rc = -EFAULT;
512 			goto reinject_interrupt;
513 		}
514 	}
515 
516 	/* irq was successfully handed to the guest */
517 	kfree(inti);
518 	kvm_s390_set_psw_cc(vcpu, 1);
519 	return 0;
520 reinject_interrupt:
521 	/*
522 	 * If we encounter a problem storing the interruption code, the
523 	 * instruction is suppressed from the guest's view: reinject the
524 	 * interrupt.
525 	 */
526 	if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
527 		kfree(inti);
528 		rc = -EFAULT;
529 	}
530 	/* don't set the cc, a pgm irq was injected or we drop to user space */
531 	return rc ? -EFAULT : 0;
532 }
533 
534 static int handle_tsch(struct kvm_vcpu *vcpu)
535 {
536 	struct kvm_s390_interrupt_info *inti = NULL;
537 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
538 
539 	vcpu->stat.instruction_tsch++;
540 
541 	/* a valid schid has at least one bit set */
542 	if (vcpu->run->s.regs.gprs[1])
543 		inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
544 					   vcpu->run->s.regs.gprs[1]);
545 
546 	/*
547 	 * Prepare exit to userspace.
548 	 * We indicate whether we dequeued a pending I/O interrupt
549 	 * so that userspace can re-inject it if the instruction gets
550 	 * a program check. While this may re-order the pending I/O
551 	 * interrupts, this is no problem since the priority is kept
552 	 * intact.
553 	 */
554 	vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
555 	vcpu->run->s390_tsch.dequeued = !!inti;
556 	if (inti) {
557 		vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
558 		vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
559 		vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
560 		vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
561 	}
562 	vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
563 	kfree(inti);
564 	return -EREMOTE;
565 }
566 
567 static int handle_io_inst(struct kvm_vcpu *vcpu)
568 {
569 	VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
570 
571 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
572 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
573 
574 	if (vcpu->kvm->arch.css_support) {
575 		/*
576 		 * Most I/O instructions will be handled by userspace.
577 		 * Exceptions are tpi and the interrupt portion of tsch.
578 		 */
579 		if (vcpu->arch.sie_block->ipa == 0xb236)
580 			return handle_tpi(vcpu);
581 		if (vcpu->arch.sie_block->ipa == 0xb235)
582 			return handle_tsch(vcpu);
583 		/* Handle in userspace. */
584 		vcpu->stat.instruction_io_other++;
585 		return -EOPNOTSUPP;
586 	} else {
587 		/*
588 		 * Set condition code 3 to stop the guest from issuing channel
589 		 * I/O instructions.
590 		 */
591 		kvm_s390_set_psw_cc(vcpu, 3);
592 		return 0;
593 	}
594 }
595 
596 /*
597  * handle_pqap: Handling pqap interception
598  * @vcpu: the vcpu having issue the pqap instruction
599  *
600  * We now support PQAP/AQIC instructions and we need to correctly
601  * answer the guest even if no dedicated driver's hook is available.
602  *
603  * The intercepting code calls a dedicated callback for this instruction
604  * if a driver did register one in the CRYPTO satellite of the
605  * SIE block.
606  *
607  * If no callback is available, the queues are not available, return this
608  * response code to the caller and set CC to 3.
609  * Else return the response code returned by the callback.
610  */
611 static int handle_pqap(struct kvm_vcpu *vcpu)
612 {
613 	struct ap_queue_status status = {};
614 	unsigned long reg0;
615 	int ret;
616 	uint8_t fc;
617 
618 	/* Verify that the AP instruction are available */
619 	if (!ap_instructions_available())
620 		return -EOPNOTSUPP;
621 	/* Verify that the guest is allowed to use AP instructions */
622 	if (!(vcpu->arch.sie_block->eca & ECA_APIE))
623 		return -EOPNOTSUPP;
624 	/*
625 	 * The only possibly intercepted functions when AP instructions are
626 	 * available for the guest are AQIC and TAPQ with the t bit set
627 	 * since we do not set IC.3 (FIII) we currently will only intercept
628 	 * the AQIC function code.
629 	 */
630 	reg0 = vcpu->run->s.regs.gprs[0];
631 	fc = (reg0 >> 24) & 0xff;
632 	if (WARN_ON_ONCE(fc != 0x03))
633 		return -EOPNOTSUPP;
634 
635 	/* PQAP instruction is allowed for guest kernel only */
636 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
637 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
638 
639 	/* Common PQAP instruction specification exceptions */
640 	/* bits 41-47 must all be zeros */
641 	if (reg0 & 0x007f0000UL)
642 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
643 	/* APFT not install and T bit set */
644 	if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
645 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
646 	/* APXA not installed and APID greater 64 or APQI greater 16 */
647 	if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
648 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
649 
650 	/* AQIC function code specific exception */
651 	/* facility 65 not present for AQIC function code */
652 	if (!test_kvm_facility(vcpu->kvm, 65))
653 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
654 
655 	/*
656 	 * Verify that the hook callback is registered, lock the owner
657 	 * and call the hook.
658 	 */
659 	if (vcpu->kvm->arch.crypto.pqap_hook) {
660 		if (!try_module_get(vcpu->kvm->arch.crypto.pqap_hook->owner))
661 			return -EOPNOTSUPP;
662 		ret = vcpu->kvm->arch.crypto.pqap_hook->hook(vcpu);
663 		module_put(vcpu->kvm->arch.crypto.pqap_hook->owner);
664 		if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
665 			kvm_s390_set_psw_cc(vcpu, 3);
666 		return ret;
667 	}
668 	/*
669 	 * A vfio_driver must register a hook.
670 	 * No hook means no driver to enable the SIE CRYCB and no queues.
671 	 * We send this response to the guest.
672 	 */
673 	status.response_code = 0x01;
674 	memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
675 	kvm_s390_set_psw_cc(vcpu, 3);
676 	return 0;
677 }
678 
679 static int handle_stfl(struct kvm_vcpu *vcpu)
680 {
681 	int rc;
682 	unsigned int fac;
683 
684 	vcpu->stat.instruction_stfl++;
685 
686 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
687 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
688 
689 	/*
690 	 * We need to shift the lower 32 facility bits (bit 0-31) from a u64
691 	 * into a u32 memory representation. They will remain bits 0-31.
692 	 */
693 	fac = *vcpu->kvm->arch.model.fac_list >> 32;
694 	rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
695 			    &fac, sizeof(fac));
696 	if (rc)
697 		return rc;
698 	VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
699 	trace_kvm_s390_handle_stfl(vcpu, fac);
700 	return 0;
701 }
702 
703 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
704 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
705 #define PSW_ADDR_24 0x0000000000ffffffUL
706 #define PSW_ADDR_31 0x000000007fffffffUL
707 
708 int is_valid_psw(psw_t *psw)
709 {
710 	if (psw->mask & PSW_MASK_UNASSIGNED)
711 		return 0;
712 	if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
713 		if (psw->addr & ~PSW_ADDR_31)
714 			return 0;
715 	}
716 	if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
717 		return 0;
718 	if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
719 		return 0;
720 	if (psw->addr & 1)
721 		return 0;
722 	return 1;
723 }
724 
725 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
726 {
727 	psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
728 	psw_compat_t new_psw;
729 	u64 addr;
730 	int rc;
731 	u8 ar;
732 
733 	vcpu->stat.instruction_lpsw++;
734 
735 	if (gpsw->mask & PSW_MASK_PSTATE)
736 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
737 
738 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
739 	if (addr & 7)
740 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
741 
742 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
743 	if (rc)
744 		return kvm_s390_inject_prog_cond(vcpu, rc);
745 	if (!(new_psw.mask & PSW32_MASK_BASE))
746 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
747 	gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
748 	gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
749 	gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
750 	if (!is_valid_psw(gpsw))
751 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
752 	return 0;
753 }
754 
755 static int handle_lpswe(struct kvm_vcpu *vcpu)
756 {
757 	psw_t new_psw;
758 	u64 addr;
759 	int rc;
760 	u8 ar;
761 
762 	vcpu->stat.instruction_lpswe++;
763 
764 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
765 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
766 
767 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
768 	if (addr & 7)
769 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
770 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
771 	if (rc)
772 		return kvm_s390_inject_prog_cond(vcpu, rc);
773 	vcpu->arch.sie_block->gpsw = new_psw;
774 	if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
775 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
776 	return 0;
777 }
778 
779 static int handle_stidp(struct kvm_vcpu *vcpu)
780 {
781 	u64 stidp_data = vcpu->kvm->arch.model.cpuid;
782 	u64 operand2;
783 	int rc;
784 	u8 ar;
785 
786 	vcpu->stat.instruction_stidp++;
787 
788 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
789 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
790 
791 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
792 
793 	if (operand2 & 7)
794 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
795 
796 	rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
797 	if (rc)
798 		return kvm_s390_inject_prog_cond(vcpu, rc);
799 
800 	VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
801 	return 0;
802 }
803 
804 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
805 {
806 	int cpus = 0;
807 	int n;
808 
809 	cpus = atomic_read(&vcpu->kvm->online_vcpus);
810 
811 	/* deal with other level 3 hypervisors */
812 	if (stsi(mem, 3, 2, 2))
813 		mem->count = 0;
814 	if (mem->count < 8)
815 		mem->count++;
816 	for (n = mem->count - 1; n > 0 ; n--)
817 		memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
818 
819 	memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
820 	mem->vm[0].cpus_total = cpus;
821 	mem->vm[0].cpus_configured = cpus;
822 	mem->vm[0].cpus_standby = 0;
823 	mem->vm[0].cpus_reserved = 0;
824 	mem->vm[0].caf = 1000;
825 	memcpy(mem->vm[0].name, "KVMguest", 8);
826 	ASCEBC(mem->vm[0].name, 8);
827 	memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
828 	ASCEBC(mem->vm[0].cpi, 16);
829 }
830 
831 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
832 				 u8 fc, u8 sel1, u16 sel2)
833 {
834 	vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
835 	vcpu->run->s390_stsi.addr = addr;
836 	vcpu->run->s390_stsi.ar = ar;
837 	vcpu->run->s390_stsi.fc = fc;
838 	vcpu->run->s390_stsi.sel1 = sel1;
839 	vcpu->run->s390_stsi.sel2 = sel2;
840 }
841 
842 static int handle_stsi(struct kvm_vcpu *vcpu)
843 {
844 	int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
845 	int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
846 	int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
847 	unsigned long mem = 0;
848 	u64 operand2;
849 	int rc = 0;
850 	u8 ar;
851 
852 	vcpu->stat.instruction_stsi++;
853 	VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
854 
855 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
856 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
857 
858 	if (fc > 3) {
859 		kvm_s390_set_psw_cc(vcpu, 3);
860 		return 0;
861 	}
862 
863 	if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
864 	    || vcpu->run->s.regs.gprs[1] & 0xffff0000)
865 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
866 
867 	if (fc == 0) {
868 		vcpu->run->s.regs.gprs[0] = 3 << 28;
869 		kvm_s390_set_psw_cc(vcpu, 0);
870 		return 0;
871 	}
872 
873 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
874 
875 	if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff))
876 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
877 
878 	switch (fc) {
879 	case 1: /* same handling for 1 and 2 */
880 	case 2:
881 		mem = get_zeroed_page(GFP_KERNEL);
882 		if (!mem)
883 			goto out_no_data;
884 		if (stsi((void *) mem, fc, sel1, sel2))
885 			goto out_no_data;
886 		break;
887 	case 3:
888 		if (sel1 != 2 || sel2 != 2)
889 			goto out_no_data;
890 		mem = get_zeroed_page(GFP_KERNEL);
891 		if (!mem)
892 			goto out_no_data;
893 		handle_stsi_3_2_2(vcpu, (void *) mem);
894 		break;
895 	}
896 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
897 		memcpy((void *)sida_origin(vcpu->arch.sie_block), (void *)mem,
898 		       PAGE_SIZE);
899 		rc = 0;
900 	} else {
901 		rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
902 	}
903 	if (rc) {
904 		rc = kvm_s390_inject_prog_cond(vcpu, rc);
905 		goto out;
906 	}
907 	if (vcpu->kvm->arch.user_stsi) {
908 		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
909 		rc = -EREMOTE;
910 	}
911 	trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
912 	free_page(mem);
913 	kvm_s390_set_psw_cc(vcpu, 0);
914 	vcpu->run->s.regs.gprs[0] = 0;
915 	return rc;
916 out_no_data:
917 	kvm_s390_set_psw_cc(vcpu, 3);
918 out:
919 	free_page(mem);
920 	return rc;
921 }
922 
923 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
924 {
925 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
926 	case 0x02:
927 		return handle_stidp(vcpu);
928 	case 0x04:
929 		return handle_set_clock(vcpu);
930 	case 0x10:
931 		return handle_set_prefix(vcpu);
932 	case 0x11:
933 		return handle_store_prefix(vcpu);
934 	case 0x12:
935 		return handle_store_cpu_address(vcpu);
936 	case 0x14:
937 		return kvm_s390_handle_vsie(vcpu);
938 	case 0x21:
939 	case 0x50:
940 		return handle_ipte_interlock(vcpu);
941 	case 0x29:
942 		return handle_iske(vcpu);
943 	case 0x2a:
944 		return handle_rrbe(vcpu);
945 	case 0x2b:
946 		return handle_sske(vcpu);
947 	case 0x2c:
948 		return handle_test_block(vcpu);
949 	case 0x30:
950 	case 0x31:
951 	case 0x32:
952 	case 0x33:
953 	case 0x34:
954 	case 0x35:
955 	case 0x36:
956 	case 0x37:
957 	case 0x38:
958 	case 0x39:
959 	case 0x3a:
960 	case 0x3b:
961 	case 0x3c:
962 	case 0x5f:
963 	case 0x74:
964 	case 0x76:
965 		return handle_io_inst(vcpu);
966 	case 0x56:
967 		return handle_sthyi(vcpu);
968 	case 0x7d:
969 		return handle_stsi(vcpu);
970 	case 0xaf:
971 		return handle_pqap(vcpu);
972 	case 0xb1:
973 		return handle_stfl(vcpu);
974 	case 0xb2:
975 		return handle_lpswe(vcpu);
976 	default:
977 		return -EOPNOTSUPP;
978 	}
979 }
980 
981 static int handle_epsw(struct kvm_vcpu *vcpu)
982 {
983 	int reg1, reg2;
984 
985 	vcpu->stat.instruction_epsw++;
986 
987 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
988 
989 	/* This basically extracts the mask half of the psw. */
990 	vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
991 	vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
992 	if (reg2) {
993 		vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
994 		vcpu->run->s.regs.gprs[reg2] |=
995 			vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
996 	}
997 	return 0;
998 }
999 
1000 #define PFMF_RESERVED   0xfffc0101UL
1001 #define PFMF_SK         0x00020000UL
1002 #define PFMF_CF         0x00010000UL
1003 #define PFMF_UI         0x00008000UL
1004 #define PFMF_FSC        0x00007000UL
1005 #define PFMF_NQ         0x00000800UL
1006 #define PFMF_MR         0x00000400UL
1007 #define PFMF_MC         0x00000200UL
1008 #define PFMF_KEY        0x000000feUL
1009 
1010 static int handle_pfmf(struct kvm_vcpu *vcpu)
1011 {
1012 	bool mr = false, mc = false, nq;
1013 	int reg1, reg2;
1014 	unsigned long start, end;
1015 	unsigned char key;
1016 
1017 	vcpu->stat.instruction_pfmf++;
1018 
1019 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1020 
1021 	if (!test_kvm_facility(vcpu->kvm, 8))
1022 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1023 
1024 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1025 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1026 
1027 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1028 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1029 
1030 	/* Only provide non-quiescing support if enabled for the guest */
1031 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1032 	    !test_kvm_facility(vcpu->kvm, 14))
1033 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1034 
1035 	/* Only provide conditional-SSKE support if enabled for the guest */
1036 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1037 	    test_kvm_facility(vcpu->kvm, 10)) {
1038 		mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1039 		mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1040 	}
1041 
1042 	nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1043 	key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1044 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1045 	start = kvm_s390_logical_to_effective(vcpu, start);
1046 
1047 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1048 		if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1049 			return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1050 	}
1051 
1052 	switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1053 	case 0x00000000:
1054 		/* only 4k frames specify a real address */
1055 		start = kvm_s390_real_to_abs(vcpu, start);
1056 		end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1057 		break;
1058 	case 0x00001000:
1059 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1060 		break;
1061 	case 0x00002000:
1062 		/* only support 2G frame size if EDAT2 is available and we are
1063 		   not in 24-bit addressing mode */
1064 		if (!test_kvm_facility(vcpu->kvm, 78) ||
1065 		    psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1066 			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1067 		end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1068 		break;
1069 	default:
1070 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1071 	}
1072 
1073 	while (start != end) {
1074 		unsigned long vmaddr;
1075 		bool unlocked = false;
1076 
1077 		/* Translate guest address to host address */
1078 		vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
1079 		if (kvm_is_error_hva(vmaddr))
1080 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1081 
1082 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1083 			if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1084 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1085 		}
1086 
1087 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1088 			int rc = kvm_s390_skey_check_enable(vcpu);
1089 
1090 			if (rc)
1091 				return rc;
1092 			down_read(&current->mm->mmap_sem);
1093 			rc = cond_set_guest_storage_key(current->mm, vmaddr,
1094 							key, NULL, nq, mr, mc);
1095 			if (rc < 0) {
1096 				rc = fixup_user_fault(current, current->mm, vmaddr,
1097 						      FAULT_FLAG_WRITE, &unlocked);
1098 				rc = !rc ? -EAGAIN : rc;
1099 			}
1100 			up_read(&current->mm->mmap_sem);
1101 			if (rc == -EFAULT)
1102 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1103 			if (rc == -EAGAIN)
1104 				continue;
1105 			if (rc < 0)
1106 				return rc;
1107 		}
1108 		start += PAGE_SIZE;
1109 	}
1110 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1111 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1112 			vcpu->run->s.regs.gprs[reg2] = end;
1113 		} else {
1114 			vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1115 			end = kvm_s390_logical_to_effective(vcpu, end);
1116 			vcpu->run->s.regs.gprs[reg2] |= end;
1117 		}
1118 	}
1119 	return 0;
1120 }
1121 
1122 /*
1123  * Must be called with relevant read locks held (kvm->mm->mmap_sem, kvm->srcu)
1124  */
1125 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1126 {
1127 	int r1, r2, nappended, entries;
1128 	unsigned long gfn, hva, res, pgstev, ptev;
1129 	unsigned long *cbrlo;
1130 
1131 	/*
1132 	 * We don't need to set SD.FPF.SK to 1 here, because if we have a
1133 	 * machine check here we either handle it or crash
1134 	 */
1135 
1136 	kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1137 	gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1138 	hva = gfn_to_hva(vcpu->kvm, gfn);
1139 	entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1140 
1141 	if (kvm_is_error_hva(hva))
1142 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1143 
1144 	nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1145 	if (nappended < 0) {
1146 		res = orc ? 0x10 : 0;
1147 		vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1148 		return 0;
1149 	}
1150 	res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1151 	/*
1152 	 * Set the block-content state part of the result. 0 means resident, so
1153 	 * nothing to do if the page is valid. 2 is for preserved pages
1154 	 * (non-present and non-zero), and 3 for zero pages (non-present and
1155 	 * zero).
1156 	 */
1157 	if (ptev & _PAGE_INVALID) {
1158 		res |= 2;
1159 		if (pgstev & _PGSTE_GPS_ZERO)
1160 			res |= 1;
1161 	}
1162 	if (pgstev & _PGSTE_GPS_NODAT)
1163 		res |= 0x20;
1164 	vcpu->run->s.regs.gprs[r1] = res;
1165 	/*
1166 	 * It is possible that all the normal 511 slots were full, in which case
1167 	 * we will now write in the 512th slot, which is reserved for host use.
1168 	 * In both cases we let the normal essa handling code process all the
1169 	 * slots, including the reserved one, if needed.
1170 	 */
1171 	if (nappended > 0) {
1172 		cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1173 		cbrlo[entries] = gfn << PAGE_SHIFT;
1174 	}
1175 
1176 	if (orc) {
1177 		struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1178 
1179 		/* Increment only if we are really flipping the bit */
1180 		if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1181 			atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1182 	}
1183 
1184 	return nappended;
1185 }
1186 
1187 static int handle_essa(struct kvm_vcpu *vcpu)
1188 {
1189 	/* entries expected to be 1FF */
1190 	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1191 	unsigned long *cbrlo;
1192 	struct gmap *gmap;
1193 	int i, orc;
1194 
1195 	VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1196 	gmap = vcpu->arch.gmap;
1197 	vcpu->stat.instruction_essa++;
1198 	if (!vcpu->kvm->arch.use_cmma)
1199 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1200 
1201 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1202 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1203 	/* Check for invalid operation request code */
1204 	orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1205 	/* ORCs 0-6 are always valid */
1206 	if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1207 						: ESSA_SET_STABLE_IF_RESIDENT))
1208 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1209 
1210 	if (!vcpu->kvm->arch.migration_mode) {
1211 		/*
1212 		 * CMMA is enabled in the KVM settings, but is disabled in
1213 		 * the SIE block and in the mm_context, and we are not doing
1214 		 * a migration. Enable CMMA in the mm_context.
1215 		 * Since we need to take a write lock to write to the context
1216 		 * to avoid races with storage keys handling, we check if the
1217 		 * value really needs to be written to; if the value is
1218 		 * already correct, we do nothing and avoid the lock.
1219 		 */
1220 		if (vcpu->kvm->mm->context.uses_cmm == 0) {
1221 			down_write(&vcpu->kvm->mm->mmap_sem);
1222 			vcpu->kvm->mm->context.uses_cmm = 1;
1223 			up_write(&vcpu->kvm->mm->mmap_sem);
1224 		}
1225 		/*
1226 		 * If we are here, we are supposed to have CMMA enabled in
1227 		 * the SIE block. Enabling CMMA works on a per-CPU basis,
1228 		 * while the context use_cmma flag is per process.
1229 		 * It's possible that the context flag is enabled and the
1230 		 * SIE flag is not, so we set the flag always; if it was
1231 		 * already set, nothing changes, otherwise we enable it
1232 		 * on this CPU too.
1233 		 */
1234 		vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1235 		/* Retry the ESSA instruction */
1236 		kvm_s390_retry_instr(vcpu);
1237 	} else {
1238 		int srcu_idx;
1239 
1240 		down_read(&vcpu->kvm->mm->mmap_sem);
1241 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1242 		i = __do_essa(vcpu, orc);
1243 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1244 		up_read(&vcpu->kvm->mm->mmap_sem);
1245 		if (i < 0)
1246 			return i;
1247 		/* Account for the possible extra cbrl entry */
1248 		entries += i;
1249 	}
1250 	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;	/* reset nceo */
1251 	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1252 	down_read(&gmap->mm->mmap_sem);
1253 	for (i = 0; i < entries; ++i)
1254 		__gmap_zap(gmap, cbrlo[i]);
1255 	up_read(&gmap->mm->mmap_sem);
1256 	return 0;
1257 }
1258 
1259 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1260 {
1261 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1262 	case 0x8a:
1263 	case 0x8e:
1264 	case 0x8f:
1265 		return handle_ipte_interlock(vcpu);
1266 	case 0x8d:
1267 		return handle_epsw(vcpu);
1268 	case 0xab:
1269 		return handle_essa(vcpu);
1270 	case 0xaf:
1271 		return handle_pfmf(vcpu);
1272 	default:
1273 		return -EOPNOTSUPP;
1274 	}
1275 }
1276 
1277 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1278 {
1279 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1280 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1281 	int reg, rc, nr_regs;
1282 	u32 ctl_array[16];
1283 	u64 ga;
1284 	u8 ar;
1285 
1286 	vcpu->stat.instruction_lctl++;
1287 
1288 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1289 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1290 
1291 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1292 
1293 	if (ga & 3)
1294 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1295 
1296 	VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1297 	trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1298 
1299 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1300 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1301 	if (rc)
1302 		return kvm_s390_inject_prog_cond(vcpu, rc);
1303 	reg = reg1;
1304 	nr_regs = 0;
1305 	do {
1306 		vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1307 		vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1308 		if (reg == reg3)
1309 			break;
1310 		reg = (reg + 1) % 16;
1311 	} while (1);
1312 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1313 	return 0;
1314 }
1315 
1316 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1317 {
1318 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1319 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1320 	int reg, rc, nr_regs;
1321 	u32 ctl_array[16];
1322 	u64 ga;
1323 	u8 ar;
1324 
1325 	vcpu->stat.instruction_stctl++;
1326 
1327 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1328 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1329 
1330 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1331 
1332 	if (ga & 3)
1333 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1334 
1335 	VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1336 	trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1337 
1338 	reg = reg1;
1339 	nr_regs = 0;
1340 	do {
1341 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1342 		if (reg == reg3)
1343 			break;
1344 		reg = (reg + 1) % 16;
1345 	} while (1);
1346 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1347 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1348 }
1349 
1350 static int handle_lctlg(struct kvm_vcpu *vcpu)
1351 {
1352 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1353 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1354 	int reg, rc, nr_regs;
1355 	u64 ctl_array[16];
1356 	u64 ga;
1357 	u8 ar;
1358 
1359 	vcpu->stat.instruction_lctlg++;
1360 
1361 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1362 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1363 
1364 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1365 
1366 	if (ga & 7)
1367 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1368 
1369 	VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1370 	trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1371 
1372 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1373 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1374 	if (rc)
1375 		return kvm_s390_inject_prog_cond(vcpu, rc);
1376 	reg = reg1;
1377 	nr_regs = 0;
1378 	do {
1379 		vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1380 		if (reg == reg3)
1381 			break;
1382 		reg = (reg + 1) % 16;
1383 	} while (1);
1384 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1385 	return 0;
1386 }
1387 
1388 static int handle_stctg(struct kvm_vcpu *vcpu)
1389 {
1390 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1391 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1392 	int reg, rc, nr_regs;
1393 	u64 ctl_array[16];
1394 	u64 ga;
1395 	u8 ar;
1396 
1397 	vcpu->stat.instruction_stctg++;
1398 
1399 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1400 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1401 
1402 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1403 
1404 	if (ga & 7)
1405 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1406 
1407 	VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1408 	trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1409 
1410 	reg = reg1;
1411 	nr_regs = 0;
1412 	do {
1413 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1414 		if (reg == reg3)
1415 			break;
1416 		reg = (reg + 1) % 16;
1417 	} while (1);
1418 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1419 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1420 }
1421 
1422 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1423 {
1424 	switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1425 	case 0x25:
1426 		return handle_stctg(vcpu);
1427 	case 0x2f:
1428 		return handle_lctlg(vcpu);
1429 	case 0x60:
1430 	case 0x61:
1431 	case 0x62:
1432 		return handle_ri(vcpu);
1433 	default:
1434 		return -EOPNOTSUPP;
1435 	}
1436 }
1437 
1438 static int handle_tprot(struct kvm_vcpu *vcpu)
1439 {
1440 	u64 address1, address2;
1441 	unsigned long hva, gpa;
1442 	int ret = 0, cc = 0;
1443 	bool writable;
1444 	u8 ar;
1445 
1446 	vcpu->stat.instruction_tprot++;
1447 
1448 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1449 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1450 
1451 	kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1452 
1453 	/* we only handle the Linux memory detection case:
1454 	 * access key == 0
1455 	 * everything else goes to userspace. */
1456 	if (address2 & 0xf0)
1457 		return -EOPNOTSUPP;
1458 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1459 		ipte_lock(vcpu);
1460 	ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1461 	if (ret == PGM_PROTECTION) {
1462 		/* Write protected? Try again with read-only... */
1463 		cc = 1;
1464 		ret = guest_translate_address(vcpu, address1, ar, &gpa,
1465 					      GACC_FETCH);
1466 	}
1467 	if (ret) {
1468 		if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1469 			ret = kvm_s390_inject_program_int(vcpu, ret);
1470 		} else if (ret > 0) {
1471 			/* Translation not available */
1472 			kvm_s390_set_psw_cc(vcpu, 3);
1473 			ret = 0;
1474 		}
1475 		goto out_unlock;
1476 	}
1477 
1478 	hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1479 	if (kvm_is_error_hva(hva)) {
1480 		ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1481 	} else {
1482 		if (!writable)
1483 			cc = 1;		/* Write not permitted ==> read-only */
1484 		kvm_s390_set_psw_cc(vcpu, cc);
1485 		/* Note: CC2 only occurs for storage keys (not supported yet) */
1486 	}
1487 out_unlock:
1488 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1489 		ipte_unlock(vcpu);
1490 	return ret;
1491 }
1492 
1493 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1494 {
1495 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1496 	case 0x01:
1497 		return handle_tprot(vcpu);
1498 	default:
1499 		return -EOPNOTSUPP;
1500 	}
1501 }
1502 
1503 static int handle_sckpf(struct kvm_vcpu *vcpu)
1504 {
1505 	u32 value;
1506 
1507 	vcpu->stat.instruction_sckpf++;
1508 
1509 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1510 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1511 
1512 	if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1513 		return kvm_s390_inject_program_int(vcpu,
1514 						   PGM_SPECIFICATION);
1515 
1516 	value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1517 	vcpu->arch.sie_block->todpr = value;
1518 
1519 	return 0;
1520 }
1521 
1522 static int handle_ptff(struct kvm_vcpu *vcpu)
1523 {
1524 	vcpu->stat.instruction_ptff++;
1525 
1526 	/* we don't emulate any control instructions yet */
1527 	kvm_s390_set_psw_cc(vcpu, 3);
1528 	return 0;
1529 }
1530 
1531 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1532 {
1533 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1534 	case 0x04:
1535 		return handle_ptff(vcpu);
1536 	case 0x07:
1537 		return handle_sckpf(vcpu);
1538 	default:
1539 		return -EOPNOTSUPP;
1540 	}
1541 }
1542