xref: /openbmc/linux/arch/s390/kvm/priv.c (revision e1e38ea1)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * handling privileged instructions
4  *
5  * Copyright IBM Corp. 2008, 2018
6  *
7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
8  *               Christian Borntraeger <borntraeger@de.ibm.com>
9  */
10 
11 #include <linux/kvm.h>
12 #include <linux/gfp.h>
13 #include <linux/errno.h>
14 #include <linux/compat.h>
15 #include <linux/mm_types.h>
16 
17 #include <asm/asm-offsets.h>
18 #include <asm/facility.h>
19 #include <asm/current.h>
20 #include <asm/debug.h>
21 #include <asm/ebcdic.h>
22 #include <asm/sysinfo.h>
23 #include <asm/pgtable.h>
24 #include <asm/page-states.h>
25 #include <asm/pgalloc.h>
26 #include <asm/gmap.h>
27 #include <asm/io.h>
28 #include <asm/ptrace.h>
29 #include <asm/sclp.h>
30 #include "gaccess.h"
31 #include "kvm-s390.h"
32 #include "trace.h"
33 
34 static int handle_ri(struct kvm_vcpu *vcpu)
35 {
36 	vcpu->stat.instruction_ri++;
37 
38 	if (test_kvm_facility(vcpu->kvm, 64)) {
39 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
40 		vcpu->arch.sie_block->ecb3 |= ECB3_RI;
41 		kvm_s390_retry_instr(vcpu);
42 		return 0;
43 	} else
44 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
45 }
46 
47 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
48 {
49 	if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
50 		return handle_ri(vcpu);
51 	else
52 		return -EOPNOTSUPP;
53 }
54 
55 static int handle_gs(struct kvm_vcpu *vcpu)
56 {
57 	vcpu->stat.instruction_gs++;
58 
59 	if (test_kvm_facility(vcpu->kvm, 133)) {
60 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
61 		preempt_disable();
62 		__ctl_set_bit(2, 4);
63 		current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
64 		restore_gs_cb(current->thread.gs_cb);
65 		preempt_enable();
66 		vcpu->arch.sie_block->ecb |= ECB_GS;
67 		vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
68 		vcpu->arch.gs_enabled = 1;
69 		kvm_s390_retry_instr(vcpu);
70 		return 0;
71 	} else
72 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
73 }
74 
75 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
76 {
77 	int code = vcpu->arch.sie_block->ipb & 0xff;
78 
79 	if (code == 0x49 || code == 0x4d)
80 		return handle_gs(vcpu);
81 	else
82 		return -EOPNOTSUPP;
83 }
84 /* Handle SCK (SET CLOCK) interception */
85 static int handle_set_clock(struct kvm_vcpu *vcpu)
86 {
87 	struct kvm_s390_vm_tod_clock gtod = { 0 };
88 	int rc;
89 	u8 ar;
90 	u64 op2;
91 
92 	vcpu->stat.instruction_sck++;
93 
94 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
95 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
96 
97 	op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
98 	if (op2 & 7)	/* Operand must be on a doubleword boundary */
99 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
100 	rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
101 	if (rc)
102 		return kvm_s390_inject_prog_cond(vcpu, rc);
103 
104 	VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
105 	kvm_s390_set_tod_clock(vcpu->kvm, &gtod);
106 
107 	kvm_s390_set_psw_cc(vcpu, 0);
108 	return 0;
109 }
110 
111 static int handle_set_prefix(struct kvm_vcpu *vcpu)
112 {
113 	u64 operand2;
114 	u32 address;
115 	int rc;
116 	u8 ar;
117 
118 	vcpu->stat.instruction_spx++;
119 
120 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
121 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
122 
123 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
124 
125 	/* must be word boundary */
126 	if (operand2 & 3)
127 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
128 
129 	/* get the value */
130 	rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
131 	if (rc)
132 		return kvm_s390_inject_prog_cond(vcpu, rc);
133 
134 	address &= 0x7fffe000u;
135 
136 	/*
137 	 * Make sure the new value is valid memory. We only need to check the
138 	 * first page, since address is 8k aligned and memory pieces are always
139 	 * at least 1MB aligned and have at least a size of 1MB.
140 	 */
141 	if (kvm_is_error_gpa(vcpu->kvm, address))
142 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
143 
144 	kvm_s390_set_prefix(vcpu, address);
145 	trace_kvm_s390_handle_prefix(vcpu, 1, address);
146 	return 0;
147 }
148 
149 static int handle_store_prefix(struct kvm_vcpu *vcpu)
150 {
151 	u64 operand2;
152 	u32 address;
153 	int rc;
154 	u8 ar;
155 
156 	vcpu->stat.instruction_stpx++;
157 
158 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
159 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
160 
161 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
162 
163 	/* must be word boundary */
164 	if (operand2 & 3)
165 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
166 
167 	address = kvm_s390_get_prefix(vcpu);
168 
169 	/* get the value */
170 	rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
171 	if (rc)
172 		return kvm_s390_inject_prog_cond(vcpu, rc);
173 
174 	VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
175 	trace_kvm_s390_handle_prefix(vcpu, 0, address);
176 	return 0;
177 }
178 
179 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
180 {
181 	u16 vcpu_id = vcpu->vcpu_id;
182 	u64 ga;
183 	int rc;
184 	u8 ar;
185 
186 	vcpu->stat.instruction_stap++;
187 
188 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
189 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
190 
191 	ga = kvm_s390_get_base_disp_s(vcpu, &ar);
192 
193 	if (ga & 1)
194 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
195 
196 	rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
197 	if (rc)
198 		return kvm_s390_inject_prog_cond(vcpu, rc);
199 
200 	VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
201 	trace_kvm_s390_handle_stap(vcpu, ga);
202 	return 0;
203 }
204 
205 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
206 {
207 	int rc;
208 
209 	trace_kvm_s390_skey_related_inst(vcpu);
210 	/* Already enabled? */
211 	if (vcpu->arch.skey_enabled)
212 		return 0;
213 
214 	rc = s390_enable_skey();
215 	VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
216 	if (rc)
217 		return rc;
218 
219 	if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
220 		kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
221 	if (!vcpu->kvm->arch.use_skf)
222 		vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
223 	else
224 		vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
225 	vcpu->arch.skey_enabled = true;
226 	return 0;
227 }
228 
229 static int try_handle_skey(struct kvm_vcpu *vcpu)
230 {
231 	int rc;
232 
233 	rc = kvm_s390_skey_check_enable(vcpu);
234 	if (rc)
235 		return rc;
236 	if (vcpu->kvm->arch.use_skf) {
237 		/* with storage-key facility, SIE interprets it for us */
238 		kvm_s390_retry_instr(vcpu);
239 		VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
240 		return -EAGAIN;
241 	}
242 	return 0;
243 }
244 
245 static int handle_iske(struct kvm_vcpu *vcpu)
246 {
247 	unsigned long gaddr, vmaddr;
248 	unsigned char key;
249 	int reg1, reg2;
250 	bool unlocked;
251 	int rc;
252 
253 	vcpu->stat.instruction_iske++;
254 
255 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
256 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
257 
258 	rc = try_handle_skey(vcpu);
259 	if (rc)
260 		return rc != -EAGAIN ? rc : 0;
261 
262 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
263 
264 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
265 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
266 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
267 	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
268 	if (kvm_is_error_hva(vmaddr))
269 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
270 retry:
271 	unlocked = false;
272 	down_read(&current->mm->mmap_sem);
273 	rc = get_guest_storage_key(current->mm, vmaddr, &key);
274 
275 	if (rc) {
276 		rc = fixup_user_fault(current, current->mm, vmaddr,
277 				      FAULT_FLAG_WRITE, &unlocked);
278 		if (!rc) {
279 			up_read(&current->mm->mmap_sem);
280 			goto retry;
281 		}
282 	}
283 	if (rc)
284 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
285 	up_read(&current->mm->mmap_sem);
286 	vcpu->run->s.regs.gprs[reg1] &= ~0xff;
287 	vcpu->run->s.regs.gprs[reg1] |= key;
288 	return 0;
289 }
290 
291 static int handle_rrbe(struct kvm_vcpu *vcpu)
292 {
293 	unsigned long vmaddr, gaddr;
294 	int reg1, reg2;
295 	bool unlocked;
296 	int rc;
297 
298 	vcpu->stat.instruction_rrbe++;
299 
300 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
301 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
302 
303 	rc = try_handle_skey(vcpu);
304 	if (rc)
305 		return rc != -EAGAIN ? rc : 0;
306 
307 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
308 
309 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
310 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
311 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
312 	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
313 	if (kvm_is_error_hva(vmaddr))
314 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
315 retry:
316 	unlocked = false;
317 	down_read(&current->mm->mmap_sem);
318 	rc = reset_guest_reference_bit(current->mm, vmaddr);
319 	if (rc < 0) {
320 		rc = fixup_user_fault(current, current->mm, vmaddr,
321 				      FAULT_FLAG_WRITE, &unlocked);
322 		if (!rc) {
323 			up_read(&current->mm->mmap_sem);
324 			goto retry;
325 		}
326 	}
327 	if (rc < 0)
328 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
329 	up_read(&current->mm->mmap_sem);
330 	kvm_s390_set_psw_cc(vcpu, rc);
331 	return 0;
332 }
333 
334 #define SSKE_NQ 0x8
335 #define SSKE_MR 0x4
336 #define SSKE_MC 0x2
337 #define SSKE_MB 0x1
338 static int handle_sske(struct kvm_vcpu *vcpu)
339 {
340 	unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
341 	unsigned long start, end;
342 	unsigned char key, oldkey;
343 	int reg1, reg2;
344 	bool unlocked;
345 	int rc;
346 
347 	vcpu->stat.instruction_sske++;
348 
349 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
350 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
351 
352 	rc = try_handle_skey(vcpu);
353 	if (rc)
354 		return rc != -EAGAIN ? rc : 0;
355 
356 	if (!test_kvm_facility(vcpu->kvm, 8))
357 		m3 &= ~SSKE_MB;
358 	if (!test_kvm_facility(vcpu->kvm, 10))
359 		m3 &= ~(SSKE_MC | SSKE_MR);
360 	if (!test_kvm_facility(vcpu->kvm, 14))
361 		m3 &= ~SSKE_NQ;
362 
363 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
364 
365 	key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
366 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
367 	start = kvm_s390_logical_to_effective(vcpu, start);
368 	if (m3 & SSKE_MB) {
369 		/* start already designates an absolute address */
370 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
371 	} else {
372 		start = kvm_s390_real_to_abs(vcpu, start);
373 		end = start + PAGE_SIZE;
374 	}
375 
376 	while (start != end) {
377 		unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
378 		unlocked = false;
379 
380 		if (kvm_is_error_hva(vmaddr))
381 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
382 
383 		down_read(&current->mm->mmap_sem);
384 		rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
385 						m3 & SSKE_NQ, m3 & SSKE_MR,
386 						m3 & SSKE_MC);
387 
388 		if (rc < 0) {
389 			rc = fixup_user_fault(current, current->mm, vmaddr,
390 					      FAULT_FLAG_WRITE, &unlocked);
391 			rc = !rc ? -EAGAIN : rc;
392 		}
393 		if (rc == -EFAULT)
394 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
395 
396 		up_read(&current->mm->mmap_sem);
397 		if (rc >= 0)
398 			start += PAGE_SIZE;
399 	}
400 
401 	if (m3 & (SSKE_MC | SSKE_MR)) {
402 		if (m3 & SSKE_MB) {
403 			/* skey in reg1 is unpredictable */
404 			kvm_s390_set_psw_cc(vcpu, 3);
405 		} else {
406 			kvm_s390_set_psw_cc(vcpu, rc);
407 			vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
408 			vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
409 		}
410 	}
411 	if (m3 & SSKE_MB) {
412 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
413 			vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
414 		else
415 			vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
416 		end = kvm_s390_logical_to_effective(vcpu, end);
417 		vcpu->run->s.regs.gprs[reg2] |= end;
418 	}
419 	return 0;
420 }
421 
422 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
423 {
424 	vcpu->stat.instruction_ipte_interlock++;
425 	if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
426 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
427 	wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
428 	kvm_s390_retry_instr(vcpu);
429 	VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
430 	return 0;
431 }
432 
433 static int handle_test_block(struct kvm_vcpu *vcpu)
434 {
435 	gpa_t addr;
436 	int reg2;
437 
438 	vcpu->stat.instruction_tb++;
439 
440 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
441 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
442 
443 	kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
444 	addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
445 	addr = kvm_s390_logical_to_effective(vcpu, addr);
446 	if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
447 		return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
448 	addr = kvm_s390_real_to_abs(vcpu, addr);
449 
450 	if (kvm_is_error_gpa(vcpu->kvm, addr))
451 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
452 	/*
453 	 * We don't expect errors on modern systems, and do not care
454 	 * about storage keys (yet), so let's just clear the page.
455 	 */
456 	if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
457 		return -EFAULT;
458 	kvm_s390_set_psw_cc(vcpu, 0);
459 	vcpu->run->s.regs.gprs[0] = 0;
460 	return 0;
461 }
462 
463 static int handle_tpi(struct kvm_vcpu *vcpu)
464 {
465 	struct kvm_s390_interrupt_info *inti;
466 	unsigned long len;
467 	u32 tpi_data[3];
468 	int rc;
469 	u64 addr;
470 	u8 ar;
471 
472 	vcpu->stat.instruction_tpi++;
473 
474 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
475 	if (addr & 3)
476 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
477 
478 	inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
479 	if (!inti) {
480 		kvm_s390_set_psw_cc(vcpu, 0);
481 		return 0;
482 	}
483 
484 	tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
485 	tpi_data[1] = inti->io.io_int_parm;
486 	tpi_data[2] = inti->io.io_int_word;
487 	if (addr) {
488 		/*
489 		 * Store the two-word I/O interruption code into the
490 		 * provided area.
491 		 */
492 		len = sizeof(tpi_data) - 4;
493 		rc = write_guest(vcpu, addr, ar, &tpi_data, len);
494 		if (rc) {
495 			rc = kvm_s390_inject_prog_cond(vcpu, rc);
496 			goto reinject_interrupt;
497 		}
498 	} else {
499 		/*
500 		 * Store the three-word I/O interruption code into
501 		 * the appropriate lowcore area.
502 		 */
503 		len = sizeof(tpi_data);
504 		if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
505 			/* failed writes to the low core are not recoverable */
506 			rc = -EFAULT;
507 			goto reinject_interrupt;
508 		}
509 	}
510 
511 	/* irq was successfully handed to the guest */
512 	kfree(inti);
513 	kvm_s390_set_psw_cc(vcpu, 1);
514 	return 0;
515 reinject_interrupt:
516 	/*
517 	 * If we encounter a problem storing the interruption code, the
518 	 * instruction is suppressed from the guest's view: reinject the
519 	 * interrupt.
520 	 */
521 	if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
522 		kfree(inti);
523 		rc = -EFAULT;
524 	}
525 	/* don't set the cc, a pgm irq was injected or we drop to user space */
526 	return rc ? -EFAULT : 0;
527 }
528 
529 static int handle_tsch(struct kvm_vcpu *vcpu)
530 {
531 	struct kvm_s390_interrupt_info *inti = NULL;
532 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
533 
534 	vcpu->stat.instruction_tsch++;
535 
536 	/* a valid schid has at least one bit set */
537 	if (vcpu->run->s.regs.gprs[1])
538 		inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
539 					   vcpu->run->s.regs.gprs[1]);
540 
541 	/*
542 	 * Prepare exit to userspace.
543 	 * We indicate whether we dequeued a pending I/O interrupt
544 	 * so that userspace can re-inject it if the instruction gets
545 	 * a program check. While this may re-order the pending I/O
546 	 * interrupts, this is no problem since the priority is kept
547 	 * intact.
548 	 */
549 	vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
550 	vcpu->run->s390_tsch.dequeued = !!inti;
551 	if (inti) {
552 		vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
553 		vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
554 		vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
555 		vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
556 	}
557 	vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
558 	kfree(inti);
559 	return -EREMOTE;
560 }
561 
562 static int handle_io_inst(struct kvm_vcpu *vcpu)
563 {
564 	VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
565 
566 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
567 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
568 
569 	if (vcpu->kvm->arch.css_support) {
570 		/*
571 		 * Most I/O instructions will be handled by userspace.
572 		 * Exceptions are tpi and the interrupt portion of tsch.
573 		 */
574 		if (vcpu->arch.sie_block->ipa == 0xb236)
575 			return handle_tpi(vcpu);
576 		if (vcpu->arch.sie_block->ipa == 0xb235)
577 			return handle_tsch(vcpu);
578 		/* Handle in userspace. */
579 		vcpu->stat.instruction_io_other++;
580 		return -EOPNOTSUPP;
581 	} else {
582 		/*
583 		 * Set condition code 3 to stop the guest from issuing channel
584 		 * I/O instructions.
585 		 */
586 		kvm_s390_set_psw_cc(vcpu, 3);
587 		return 0;
588 	}
589 }
590 
591 static int handle_stfl(struct kvm_vcpu *vcpu)
592 {
593 	int rc;
594 	unsigned int fac;
595 
596 	vcpu->stat.instruction_stfl++;
597 
598 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
599 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
600 
601 	/*
602 	 * We need to shift the lower 32 facility bits (bit 0-31) from a u64
603 	 * into a u32 memory representation. They will remain bits 0-31.
604 	 */
605 	fac = *vcpu->kvm->arch.model.fac_list >> 32;
606 	rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
607 			    &fac, sizeof(fac));
608 	if (rc)
609 		return rc;
610 	VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
611 	trace_kvm_s390_handle_stfl(vcpu, fac);
612 	return 0;
613 }
614 
615 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
616 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
617 #define PSW_ADDR_24 0x0000000000ffffffUL
618 #define PSW_ADDR_31 0x000000007fffffffUL
619 
620 int is_valid_psw(psw_t *psw)
621 {
622 	if (psw->mask & PSW_MASK_UNASSIGNED)
623 		return 0;
624 	if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
625 		if (psw->addr & ~PSW_ADDR_31)
626 			return 0;
627 	}
628 	if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
629 		return 0;
630 	if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
631 		return 0;
632 	if (psw->addr & 1)
633 		return 0;
634 	return 1;
635 }
636 
637 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
638 {
639 	psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
640 	psw_compat_t new_psw;
641 	u64 addr;
642 	int rc;
643 	u8 ar;
644 
645 	vcpu->stat.instruction_lpsw++;
646 
647 	if (gpsw->mask & PSW_MASK_PSTATE)
648 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
649 
650 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
651 	if (addr & 7)
652 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
653 
654 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
655 	if (rc)
656 		return kvm_s390_inject_prog_cond(vcpu, rc);
657 	if (!(new_psw.mask & PSW32_MASK_BASE))
658 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
659 	gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
660 	gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
661 	gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
662 	if (!is_valid_psw(gpsw))
663 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
664 	return 0;
665 }
666 
667 static int handle_lpswe(struct kvm_vcpu *vcpu)
668 {
669 	psw_t new_psw;
670 	u64 addr;
671 	int rc;
672 	u8 ar;
673 
674 	vcpu->stat.instruction_lpswe++;
675 
676 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
677 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
678 
679 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
680 	if (addr & 7)
681 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
682 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
683 	if (rc)
684 		return kvm_s390_inject_prog_cond(vcpu, rc);
685 	vcpu->arch.sie_block->gpsw = new_psw;
686 	if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
687 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
688 	return 0;
689 }
690 
691 static int handle_stidp(struct kvm_vcpu *vcpu)
692 {
693 	u64 stidp_data = vcpu->kvm->arch.model.cpuid;
694 	u64 operand2;
695 	int rc;
696 	u8 ar;
697 
698 	vcpu->stat.instruction_stidp++;
699 
700 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
701 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
702 
703 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
704 
705 	if (operand2 & 7)
706 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
707 
708 	rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
709 	if (rc)
710 		return kvm_s390_inject_prog_cond(vcpu, rc);
711 
712 	VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
713 	return 0;
714 }
715 
716 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
717 {
718 	int cpus = 0;
719 	int n;
720 
721 	cpus = atomic_read(&vcpu->kvm->online_vcpus);
722 
723 	/* deal with other level 3 hypervisors */
724 	if (stsi(mem, 3, 2, 2))
725 		mem->count = 0;
726 	if (mem->count < 8)
727 		mem->count++;
728 	for (n = mem->count - 1; n > 0 ; n--)
729 		memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
730 
731 	memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
732 	mem->vm[0].cpus_total = cpus;
733 	mem->vm[0].cpus_configured = cpus;
734 	mem->vm[0].cpus_standby = 0;
735 	mem->vm[0].cpus_reserved = 0;
736 	mem->vm[0].caf = 1000;
737 	memcpy(mem->vm[0].name, "KVMguest", 8);
738 	ASCEBC(mem->vm[0].name, 8);
739 	memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
740 	ASCEBC(mem->vm[0].cpi, 16);
741 }
742 
743 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
744 				 u8 fc, u8 sel1, u16 sel2)
745 {
746 	vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
747 	vcpu->run->s390_stsi.addr = addr;
748 	vcpu->run->s390_stsi.ar = ar;
749 	vcpu->run->s390_stsi.fc = fc;
750 	vcpu->run->s390_stsi.sel1 = sel1;
751 	vcpu->run->s390_stsi.sel2 = sel2;
752 }
753 
754 static int handle_stsi(struct kvm_vcpu *vcpu)
755 {
756 	int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
757 	int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
758 	int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
759 	unsigned long mem = 0;
760 	u64 operand2;
761 	int rc = 0;
762 	u8 ar;
763 
764 	vcpu->stat.instruction_stsi++;
765 	VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
766 
767 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
768 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
769 
770 	if (fc > 3) {
771 		kvm_s390_set_psw_cc(vcpu, 3);
772 		return 0;
773 	}
774 
775 	if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
776 	    || vcpu->run->s.regs.gprs[1] & 0xffff0000)
777 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
778 
779 	if (fc == 0) {
780 		vcpu->run->s.regs.gprs[0] = 3 << 28;
781 		kvm_s390_set_psw_cc(vcpu, 0);
782 		return 0;
783 	}
784 
785 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
786 
787 	if (operand2 & 0xfff)
788 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
789 
790 	switch (fc) {
791 	case 1: /* same handling for 1 and 2 */
792 	case 2:
793 		mem = get_zeroed_page(GFP_KERNEL);
794 		if (!mem)
795 			goto out_no_data;
796 		if (stsi((void *) mem, fc, sel1, sel2))
797 			goto out_no_data;
798 		break;
799 	case 3:
800 		if (sel1 != 2 || sel2 != 2)
801 			goto out_no_data;
802 		mem = get_zeroed_page(GFP_KERNEL);
803 		if (!mem)
804 			goto out_no_data;
805 		handle_stsi_3_2_2(vcpu, (void *) mem);
806 		break;
807 	}
808 
809 	rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
810 	if (rc) {
811 		rc = kvm_s390_inject_prog_cond(vcpu, rc);
812 		goto out;
813 	}
814 	if (vcpu->kvm->arch.user_stsi) {
815 		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
816 		rc = -EREMOTE;
817 	}
818 	trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
819 	free_page(mem);
820 	kvm_s390_set_psw_cc(vcpu, 0);
821 	vcpu->run->s.regs.gprs[0] = 0;
822 	return rc;
823 out_no_data:
824 	kvm_s390_set_psw_cc(vcpu, 3);
825 out:
826 	free_page(mem);
827 	return rc;
828 }
829 
830 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
831 {
832 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
833 	case 0x02:
834 		return handle_stidp(vcpu);
835 	case 0x04:
836 		return handle_set_clock(vcpu);
837 	case 0x10:
838 		return handle_set_prefix(vcpu);
839 	case 0x11:
840 		return handle_store_prefix(vcpu);
841 	case 0x12:
842 		return handle_store_cpu_address(vcpu);
843 	case 0x14:
844 		return kvm_s390_handle_vsie(vcpu);
845 	case 0x21:
846 	case 0x50:
847 		return handle_ipte_interlock(vcpu);
848 	case 0x29:
849 		return handle_iske(vcpu);
850 	case 0x2a:
851 		return handle_rrbe(vcpu);
852 	case 0x2b:
853 		return handle_sske(vcpu);
854 	case 0x2c:
855 		return handle_test_block(vcpu);
856 	case 0x30:
857 	case 0x31:
858 	case 0x32:
859 	case 0x33:
860 	case 0x34:
861 	case 0x35:
862 	case 0x36:
863 	case 0x37:
864 	case 0x38:
865 	case 0x39:
866 	case 0x3a:
867 	case 0x3b:
868 	case 0x3c:
869 	case 0x5f:
870 	case 0x74:
871 	case 0x76:
872 		return handle_io_inst(vcpu);
873 	case 0x56:
874 		return handle_sthyi(vcpu);
875 	case 0x7d:
876 		return handle_stsi(vcpu);
877 	case 0xb1:
878 		return handle_stfl(vcpu);
879 	case 0xb2:
880 		return handle_lpswe(vcpu);
881 	default:
882 		return -EOPNOTSUPP;
883 	}
884 }
885 
886 static int handle_epsw(struct kvm_vcpu *vcpu)
887 {
888 	int reg1, reg2;
889 
890 	vcpu->stat.instruction_epsw++;
891 
892 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
893 
894 	/* This basically extracts the mask half of the psw. */
895 	vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
896 	vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
897 	if (reg2) {
898 		vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
899 		vcpu->run->s.regs.gprs[reg2] |=
900 			vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
901 	}
902 	return 0;
903 }
904 
905 #define PFMF_RESERVED   0xfffc0101UL
906 #define PFMF_SK         0x00020000UL
907 #define PFMF_CF         0x00010000UL
908 #define PFMF_UI         0x00008000UL
909 #define PFMF_FSC        0x00007000UL
910 #define PFMF_NQ         0x00000800UL
911 #define PFMF_MR         0x00000400UL
912 #define PFMF_MC         0x00000200UL
913 #define PFMF_KEY        0x000000feUL
914 
915 static int handle_pfmf(struct kvm_vcpu *vcpu)
916 {
917 	bool mr = false, mc = false, nq;
918 	int reg1, reg2;
919 	unsigned long start, end;
920 	unsigned char key;
921 
922 	vcpu->stat.instruction_pfmf++;
923 
924 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
925 
926 	if (!test_kvm_facility(vcpu->kvm, 8))
927 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
928 
929 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
930 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
931 
932 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
933 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
934 
935 	/* Only provide non-quiescing support if enabled for the guest */
936 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
937 	    !test_kvm_facility(vcpu->kvm, 14))
938 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
939 
940 	/* Only provide conditional-SSKE support if enabled for the guest */
941 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
942 	    test_kvm_facility(vcpu->kvm, 10)) {
943 		mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
944 		mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
945 	}
946 
947 	nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
948 	key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
949 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
950 	start = kvm_s390_logical_to_effective(vcpu, start);
951 
952 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
953 		if (kvm_s390_check_low_addr_prot_real(vcpu, start))
954 			return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
955 	}
956 
957 	switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
958 	case 0x00000000:
959 		/* only 4k frames specify a real address */
960 		start = kvm_s390_real_to_abs(vcpu, start);
961 		end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
962 		break;
963 	case 0x00001000:
964 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
965 		break;
966 	case 0x00002000:
967 		/* only support 2G frame size if EDAT2 is available and we are
968 		   not in 24-bit addressing mode */
969 		if (!test_kvm_facility(vcpu->kvm, 78) ||
970 		    psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
971 			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
972 		end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
973 		break;
974 	default:
975 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
976 	}
977 
978 	while (start != end) {
979 		unsigned long vmaddr;
980 		bool unlocked = false;
981 
982 		/* Translate guest address to host address */
983 		vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
984 		if (kvm_is_error_hva(vmaddr))
985 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
986 
987 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
988 			if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
989 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
990 		}
991 
992 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
993 			int rc = kvm_s390_skey_check_enable(vcpu);
994 
995 			if (rc)
996 				return rc;
997 			down_read(&current->mm->mmap_sem);
998 			rc = cond_set_guest_storage_key(current->mm, vmaddr,
999 							key, NULL, nq, mr, mc);
1000 			if (rc < 0) {
1001 				rc = fixup_user_fault(current, current->mm, vmaddr,
1002 						      FAULT_FLAG_WRITE, &unlocked);
1003 				rc = !rc ? -EAGAIN : rc;
1004 			}
1005 			if (rc == -EFAULT)
1006 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1007 
1008 			up_read(&current->mm->mmap_sem);
1009 			if (rc >= 0)
1010 				start += PAGE_SIZE;
1011 		}
1012 	}
1013 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1014 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1015 			vcpu->run->s.regs.gprs[reg2] = end;
1016 		} else {
1017 			vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1018 			end = kvm_s390_logical_to_effective(vcpu, end);
1019 			vcpu->run->s.regs.gprs[reg2] |= end;
1020 		}
1021 	}
1022 	return 0;
1023 }
1024 
1025 /*
1026  * Must be called with relevant read locks held (kvm->mm->mmap_sem, kvm->srcu)
1027  */
1028 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1029 {
1030 	int r1, r2, nappended, entries;
1031 	unsigned long gfn, hva, res, pgstev, ptev;
1032 	unsigned long *cbrlo;
1033 
1034 	/*
1035 	 * We don't need to set SD.FPF.SK to 1 here, because if we have a
1036 	 * machine check here we either handle it or crash
1037 	 */
1038 
1039 	kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1040 	gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1041 	hva = gfn_to_hva(vcpu->kvm, gfn);
1042 	entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1043 
1044 	if (kvm_is_error_hva(hva))
1045 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1046 
1047 	nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1048 	if (nappended < 0) {
1049 		res = orc ? 0x10 : 0;
1050 		vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1051 		return 0;
1052 	}
1053 	res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1054 	/*
1055 	 * Set the block-content state part of the result. 0 means resident, so
1056 	 * nothing to do if the page is valid. 2 is for preserved pages
1057 	 * (non-present and non-zero), and 3 for zero pages (non-present and
1058 	 * zero).
1059 	 */
1060 	if (ptev & _PAGE_INVALID) {
1061 		res |= 2;
1062 		if (pgstev & _PGSTE_GPS_ZERO)
1063 			res |= 1;
1064 	}
1065 	if (pgstev & _PGSTE_GPS_NODAT)
1066 		res |= 0x20;
1067 	vcpu->run->s.regs.gprs[r1] = res;
1068 	/*
1069 	 * It is possible that all the normal 511 slots were full, in which case
1070 	 * we will now write in the 512th slot, which is reserved for host use.
1071 	 * In both cases we let the normal essa handling code process all the
1072 	 * slots, including the reserved one, if needed.
1073 	 */
1074 	if (nappended > 0) {
1075 		cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1076 		cbrlo[entries] = gfn << PAGE_SHIFT;
1077 	}
1078 
1079 	if (orc) {
1080 		struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1081 
1082 		/* Increment only if we are really flipping the bit */
1083 		if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1084 			atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1085 	}
1086 
1087 	return nappended;
1088 }
1089 
1090 static int handle_essa(struct kvm_vcpu *vcpu)
1091 {
1092 	/* entries expected to be 1FF */
1093 	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1094 	unsigned long *cbrlo;
1095 	struct gmap *gmap;
1096 	int i, orc;
1097 
1098 	VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1099 	gmap = vcpu->arch.gmap;
1100 	vcpu->stat.instruction_essa++;
1101 	if (!vcpu->kvm->arch.use_cmma)
1102 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1103 
1104 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1105 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1106 	/* Check for invalid operation request code */
1107 	orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1108 	/* ORCs 0-6 are always valid */
1109 	if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1110 						: ESSA_SET_STABLE_IF_RESIDENT))
1111 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1112 
1113 	if (!vcpu->kvm->arch.migration_mode) {
1114 		/*
1115 		 * CMMA is enabled in the KVM settings, but is disabled in
1116 		 * the SIE block and in the mm_context, and we are not doing
1117 		 * a migration. Enable CMMA in the mm_context.
1118 		 * Since we need to take a write lock to write to the context
1119 		 * to avoid races with storage keys handling, we check if the
1120 		 * value really needs to be written to; if the value is
1121 		 * already correct, we do nothing and avoid the lock.
1122 		 */
1123 		if (vcpu->kvm->mm->context.uses_cmm == 0) {
1124 			down_write(&vcpu->kvm->mm->mmap_sem);
1125 			vcpu->kvm->mm->context.uses_cmm = 1;
1126 			up_write(&vcpu->kvm->mm->mmap_sem);
1127 		}
1128 		/*
1129 		 * If we are here, we are supposed to have CMMA enabled in
1130 		 * the SIE block. Enabling CMMA works on a per-CPU basis,
1131 		 * while the context use_cmma flag is per process.
1132 		 * It's possible that the context flag is enabled and the
1133 		 * SIE flag is not, so we set the flag always; if it was
1134 		 * already set, nothing changes, otherwise we enable it
1135 		 * on this CPU too.
1136 		 */
1137 		vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1138 		/* Retry the ESSA instruction */
1139 		kvm_s390_retry_instr(vcpu);
1140 	} else {
1141 		int srcu_idx;
1142 
1143 		down_read(&vcpu->kvm->mm->mmap_sem);
1144 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1145 		i = __do_essa(vcpu, orc);
1146 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1147 		up_read(&vcpu->kvm->mm->mmap_sem);
1148 		if (i < 0)
1149 			return i;
1150 		/* Account for the possible extra cbrl entry */
1151 		entries += i;
1152 	}
1153 	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;	/* reset nceo */
1154 	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1155 	down_read(&gmap->mm->mmap_sem);
1156 	for (i = 0; i < entries; ++i)
1157 		__gmap_zap(gmap, cbrlo[i]);
1158 	up_read(&gmap->mm->mmap_sem);
1159 	return 0;
1160 }
1161 
1162 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1163 {
1164 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1165 	case 0x8a:
1166 	case 0x8e:
1167 	case 0x8f:
1168 		return handle_ipte_interlock(vcpu);
1169 	case 0x8d:
1170 		return handle_epsw(vcpu);
1171 	case 0xab:
1172 		return handle_essa(vcpu);
1173 	case 0xaf:
1174 		return handle_pfmf(vcpu);
1175 	default:
1176 		return -EOPNOTSUPP;
1177 	}
1178 }
1179 
1180 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1181 {
1182 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1183 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1184 	int reg, rc, nr_regs;
1185 	u32 ctl_array[16];
1186 	u64 ga;
1187 	u8 ar;
1188 
1189 	vcpu->stat.instruction_lctl++;
1190 
1191 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1192 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1193 
1194 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1195 
1196 	if (ga & 3)
1197 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1198 
1199 	VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1200 	trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1201 
1202 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1203 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1204 	if (rc)
1205 		return kvm_s390_inject_prog_cond(vcpu, rc);
1206 	reg = reg1;
1207 	nr_regs = 0;
1208 	do {
1209 		vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1210 		vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1211 		if (reg == reg3)
1212 			break;
1213 		reg = (reg + 1) % 16;
1214 	} while (1);
1215 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1216 	return 0;
1217 }
1218 
1219 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1220 {
1221 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1222 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1223 	int reg, rc, nr_regs;
1224 	u32 ctl_array[16];
1225 	u64 ga;
1226 	u8 ar;
1227 
1228 	vcpu->stat.instruction_stctl++;
1229 
1230 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1231 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1232 
1233 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1234 
1235 	if (ga & 3)
1236 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1237 
1238 	VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1239 	trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1240 
1241 	reg = reg1;
1242 	nr_regs = 0;
1243 	do {
1244 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1245 		if (reg == reg3)
1246 			break;
1247 		reg = (reg + 1) % 16;
1248 	} while (1);
1249 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1250 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1251 }
1252 
1253 static int handle_lctlg(struct kvm_vcpu *vcpu)
1254 {
1255 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1256 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1257 	int reg, rc, nr_regs;
1258 	u64 ctl_array[16];
1259 	u64 ga;
1260 	u8 ar;
1261 
1262 	vcpu->stat.instruction_lctlg++;
1263 
1264 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1265 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1266 
1267 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1268 
1269 	if (ga & 7)
1270 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1271 
1272 	VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1273 	trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1274 
1275 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1276 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1277 	if (rc)
1278 		return kvm_s390_inject_prog_cond(vcpu, rc);
1279 	reg = reg1;
1280 	nr_regs = 0;
1281 	do {
1282 		vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1283 		if (reg == reg3)
1284 			break;
1285 		reg = (reg + 1) % 16;
1286 	} while (1);
1287 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1288 	return 0;
1289 }
1290 
1291 static int handle_stctg(struct kvm_vcpu *vcpu)
1292 {
1293 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1294 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1295 	int reg, rc, nr_regs;
1296 	u64 ctl_array[16];
1297 	u64 ga;
1298 	u8 ar;
1299 
1300 	vcpu->stat.instruction_stctg++;
1301 
1302 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1303 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1304 
1305 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1306 
1307 	if (ga & 7)
1308 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1309 
1310 	VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1311 	trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1312 
1313 	reg = reg1;
1314 	nr_regs = 0;
1315 	do {
1316 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1317 		if (reg == reg3)
1318 			break;
1319 		reg = (reg + 1) % 16;
1320 	} while (1);
1321 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1322 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1323 }
1324 
1325 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1326 {
1327 	switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1328 	case 0x25:
1329 		return handle_stctg(vcpu);
1330 	case 0x2f:
1331 		return handle_lctlg(vcpu);
1332 	case 0x60:
1333 	case 0x61:
1334 	case 0x62:
1335 		return handle_ri(vcpu);
1336 	default:
1337 		return -EOPNOTSUPP;
1338 	}
1339 }
1340 
1341 static int handle_tprot(struct kvm_vcpu *vcpu)
1342 {
1343 	u64 address1, address2;
1344 	unsigned long hva, gpa;
1345 	int ret = 0, cc = 0;
1346 	bool writable;
1347 	u8 ar;
1348 
1349 	vcpu->stat.instruction_tprot++;
1350 
1351 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1352 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1353 
1354 	kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1355 
1356 	/* we only handle the Linux memory detection case:
1357 	 * access key == 0
1358 	 * everything else goes to userspace. */
1359 	if (address2 & 0xf0)
1360 		return -EOPNOTSUPP;
1361 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1362 		ipte_lock(vcpu);
1363 	ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1364 	if (ret == PGM_PROTECTION) {
1365 		/* Write protected? Try again with read-only... */
1366 		cc = 1;
1367 		ret = guest_translate_address(vcpu, address1, ar, &gpa,
1368 					      GACC_FETCH);
1369 	}
1370 	if (ret) {
1371 		if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1372 			ret = kvm_s390_inject_program_int(vcpu, ret);
1373 		} else if (ret > 0) {
1374 			/* Translation not available */
1375 			kvm_s390_set_psw_cc(vcpu, 3);
1376 			ret = 0;
1377 		}
1378 		goto out_unlock;
1379 	}
1380 
1381 	hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1382 	if (kvm_is_error_hva(hva)) {
1383 		ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1384 	} else {
1385 		if (!writable)
1386 			cc = 1;		/* Write not permitted ==> read-only */
1387 		kvm_s390_set_psw_cc(vcpu, cc);
1388 		/* Note: CC2 only occurs for storage keys (not supported yet) */
1389 	}
1390 out_unlock:
1391 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1392 		ipte_unlock(vcpu);
1393 	return ret;
1394 }
1395 
1396 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1397 {
1398 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1399 	case 0x01:
1400 		return handle_tprot(vcpu);
1401 	default:
1402 		return -EOPNOTSUPP;
1403 	}
1404 }
1405 
1406 static int handle_sckpf(struct kvm_vcpu *vcpu)
1407 {
1408 	u32 value;
1409 
1410 	vcpu->stat.instruction_sckpf++;
1411 
1412 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1413 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1414 
1415 	if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1416 		return kvm_s390_inject_program_int(vcpu,
1417 						   PGM_SPECIFICATION);
1418 
1419 	value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1420 	vcpu->arch.sie_block->todpr = value;
1421 
1422 	return 0;
1423 }
1424 
1425 static int handle_ptff(struct kvm_vcpu *vcpu)
1426 {
1427 	vcpu->stat.instruction_ptff++;
1428 
1429 	/* we don't emulate any control instructions yet */
1430 	kvm_s390_set_psw_cc(vcpu, 3);
1431 	return 0;
1432 }
1433 
1434 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1435 {
1436 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1437 	case 0x04:
1438 		return handle_ptff(vcpu);
1439 	case 0x07:
1440 		return handle_sckpf(vcpu);
1441 	default:
1442 		return -EOPNOTSUPP;
1443 	}
1444 }
1445