xref: /linux/arch/s390/kvm/priv.c (revision 53597deca0e38c30e6cd4ba2114fa42d2bcd85bb)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * handling privileged instructions
4  *
5  * Copyright IBM Corp. 2008, 2020
6  *
7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
8  *               Christian Borntraeger <borntraeger@de.ibm.com>
9  */
10 
11 #include <linux/kvm.h>
12 #include <linux/gfp.h>
13 #include <linux/errno.h>
14 #include <linux/mm_types.h>
15 #include <linux/pgtable.h>
16 #include <linux/io.h>
17 #include <asm/asm-offsets.h>
18 #include <asm/facility.h>
19 #include <asm/current.h>
20 #include <asm/debug.h>
21 #include <asm/ebcdic.h>
22 #include <asm/sysinfo.h>
23 #include <asm/page-states.h>
24 #include <asm/ptrace.h>
25 #include <asm/sclp.h>
26 #include <asm/ap.h>
27 #include <asm/gmap_helpers.h>
28 #include "gaccess.h"
29 #include "kvm-s390.h"
30 #include "trace.h"
31 #include "gmap.h"
32 
33 static int handle_ri(struct kvm_vcpu *vcpu)
34 {
35 	vcpu->stat.instruction_ri++;
36 
37 	if (test_kvm_facility(vcpu->kvm, 64)) {
38 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
39 		vcpu->arch.sie_block->ecb3 |= ECB3_RI;
40 		kvm_s390_retry_instr(vcpu);
41 		return 0;
42 	} else
43 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
44 }
45 
46 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
47 {
48 	if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
49 		return handle_ri(vcpu);
50 	else
51 		return -EOPNOTSUPP;
52 }
53 
54 static int handle_gs(struct kvm_vcpu *vcpu)
55 {
56 	vcpu->stat.instruction_gs++;
57 
58 	if (test_kvm_facility(vcpu->kvm, 133)) {
59 		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
60 		preempt_disable();
61 		local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
62 		current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
63 		restore_gs_cb(current->thread.gs_cb);
64 		preempt_enable();
65 		vcpu->arch.sie_block->ecb |= ECB_GS;
66 		vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
67 		vcpu->arch.gs_enabled = 1;
68 		kvm_s390_retry_instr(vcpu);
69 		return 0;
70 	} else
71 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
72 }
73 
74 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
75 {
76 	int code = vcpu->arch.sie_block->ipb & 0xff;
77 
78 	if (code == 0x49 || code == 0x4d)
79 		return handle_gs(vcpu);
80 	else
81 		return -EOPNOTSUPP;
82 }
83 /* Handle SCK (SET CLOCK) interception */
84 static int handle_set_clock(struct kvm_vcpu *vcpu)
85 {
86 	struct kvm_s390_vm_tod_clock gtod = { 0 };
87 	int rc;
88 	u8 ar;
89 	u64 op2;
90 
91 	vcpu->stat.instruction_sck++;
92 
93 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
94 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
95 
96 	op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
97 	if (op2 & 7)	/* Operand must be on a doubleword boundary */
98 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
99 	rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
100 	if (rc)
101 		return kvm_s390_inject_prog_cond(vcpu, rc);
102 
103 	VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
104 	/*
105 	 * To set the TOD clock the kvm lock must be taken, but the vcpu lock
106 	 * is already held in handle_set_clock. The usual lock order is the
107 	 * opposite.  As SCK is deprecated and should not be used in several
108 	 * cases, for example when the multiple epoch facility or TOD clock
109 	 * steering facility is installed (see Principles of Operation),  a
110 	 * slow path can be used.  If the lock can not be taken via try_lock,
111 	 * the instruction will be retried via -EAGAIN at a later point in
112 	 * time.
113 	 */
114 	if (!kvm_s390_try_set_tod_clock(vcpu->kvm, &gtod)) {
115 		kvm_s390_retry_instr(vcpu);
116 		return -EAGAIN;
117 	}
118 
119 	kvm_s390_set_psw_cc(vcpu, 0);
120 	return 0;
121 }
122 
123 static int handle_set_prefix(struct kvm_vcpu *vcpu)
124 {
125 	u64 operand2;
126 	u32 address;
127 	int rc;
128 	u8 ar;
129 
130 	vcpu->stat.instruction_spx++;
131 
132 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
133 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
134 
135 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
136 
137 	/* must be word boundary */
138 	if (operand2 & 3)
139 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
140 
141 	/* get the value */
142 	rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
143 	if (rc)
144 		return kvm_s390_inject_prog_cond(vcpu, rc);
145 
146 	address &= 0x7fffe000u;
147 
148 	/*
149 	 * Make sure the new value is valid memory. We only need to check the
150 	 * first page, since address is 8k aligned and memory pieces are always
151 	 * at least 1MB aligned and have at least a size of 1MB.
152 	 */
153 	if (!kvm_is_gpa_in_memslot(vcpu->kvm, address))
154 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
155 
156 	kvm_s390_set_prefix(vcpu, address);
157 	trace_kvm_s390_handle_prefix(vcpu, 1, address);
158 	return 0;
159 }
160 
161 static int handle_store_prefix(struct kvm_vcpu *vcpu)
162 {
163 	u64 operand2;
164 	u32 address;
165 	int rc;
166 	u8 ar;
167 
168 	vcpu->stat.instruction_stpx++;
169 
170 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
171 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
172 
173 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
174 
175 	/* must be word boundary */
176 	if (operand2 & 3)
177 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
178 
179 	address = kvm_s390_get_prefix(vcpu);
180 
181 	/* get the value */
182 	rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
183 	if (rc)
184 		return kvm_s390_inject_prog_cond(vcpu, rc);
185 
186 	VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
187 	trace_kvm_s390_handle_prefix(vcpu, 0, address);
188 	return 0;
189 }
190 
191 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
192 {
193 	u16 vcpu_id = vcpu->vcpu_id;
194 	u64 ga;
195 	int rc;
196 	u8 ar;
197 
198 	vcpu->stat.instruction_stap++;
199 
200 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
201 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
202 
203 	ga = kvm_s390_get_base_disp_s(vcpu, &ar);
204 
205 	if (ga & 1)
206 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
207 
208 	rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
209 	if (rc)
210 		return kvm_s390_inject_prog_cond(vcpu, rc);
211 
212 	VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
213 	trace_kvm_s390_handle_stap(vcpu, ga);
214 	return 0;
215 }
216 
217 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
218 {
219 	int rc;
220 
221 	trace_kvm_s390_skey_related_inst(vcpu);
222 	/* Already enabled? */
223 	if (vcpu->arch.skey_enabled)
224 		return 0;
225 
226 	rc = gmap_enable_skeys(vcpu->arch.gmap);
227 	VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
228 	if (rc)
229 		return rc;
230 
231 	if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
232 		kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
233 	if (!vcpu->kvm->arch.use_skf)
234 		vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
235 	else
236 		vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
237 	vcpu->arch.skey_enabled = true;
238 	return 0;
239 }
240 
241 static int try_handle_skey(struct kvm_vcpu *vcpu)
242 {
243 	int rc;
244 
245 	rc = kvm_s390_skey_check_enable(vcpu);
246 	if (rc)
247 		return rc;
248 	if (vcpu->kvm->arch.use_skf) {
249 		/* with storage-key facility, SIE interprets it for us */
250 		kvm_s390_retry_instr(vcpu);
251 		VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
252 		return -EAGAIN;
253 	}
254 	return 0;
255 }
256 
257 static int handle_iske(struct kvm_vcpu *vcpu)
258 {
259 	unsigned long gaddr;
260 	int reg1, reg2;
261 	union skey key;
262 	int rc;
263 
264 	vcpu->stat.instruction_iske++;
265 
266 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
267 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
268 
269 	rc = try_handle_skey(vcpu);
270 	if (rc)
271 		return rc != -EAGAIN ? rc : 0;
272 
273 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
274 
275 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
276 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
277 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
278 	scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
279 		rc = dat_get_storage_key(vcpu->arch.gmap->asce, gpa_to_gfn(gaddr), &key);
280 	if (rc > 0)
281 		return kvm_s390_inject_program_int(vcpu, rc);
282 	if (rc < 0)
283 		return rc;
284 	vcpu->run->s.regs.gprs[reg1] &= ~0xff;
285 	vcpu->run->s.regs.gprs[reg1] |= key.skey;
286 	return 0;
287 }
288 
289 static int handle_rrbe(struct kvm_vcpu *vcpu)
290 {
291 	unsigned long gaddr;
292 	int reg1, reg2;
293 	int rc;
294 
295 	vcpu->stat.instruction_rrbe++;
296 
297 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
298 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
299 
300 	rc = try_handle_skey(vcpu);
301 	if (rc)
302 		return rc != -EAGAIN ? rc : 0;
303 
304 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
305 
306 	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
307 	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
308 	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
309 	scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
310 		rc = dat_reset_reference_bit(vcpu->arch.gmap->asce, gpa_to_gfn(gaddr));
311 	if (rc > 0)
312 		return kvm_s390_inject_program_int(vcpu, rc);
313 	if (rc < 0)
314 		return rc;
315 	kvm_s390_set_psw_cc(vcpu, rc);
316 	return 0;
317 }
318 
319 #define SSKE_NQ 0x8
320 #define SSKE_MR 0x4
321 #define SSKE_MC 0x2
322 #define SSKE_MB 0x1
323 static int handle_sske(struct kvm_vcpu *vcpu)
324 {
325 	unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
326 	unsigned long start, end;
327 	union skey key, oldkey;
328 	int reg1, reg2;
329 	int rc;
330 
331 	vcpu->stat.instruction_sske++;
332 
333 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
334 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
335 
336 	rc = try_handle_skey(vcpu);
337 	if (rc)
338 		return rc != -EAGAIN ? rc : 0;
339 
340 	if (!test_kvm_facility(vcpu->kvm, 8))
341 		m3 &= ~SSKE_MB;
342 	if (!test_kvm_facility(vcpu->kvm, 10))
343 		m3 &= ~(SSKE_MC | SSKE_MR);
344 	if (!test_kvm_facility(vcpu->kvm, 14))
345 		m3 &= ~SSKE_NQ;
346 
347 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
348 
349 	key.skey = vcpu->run->s.regs.gprs[reg1] & 0xfe;
350 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
351 	start = kvm_s390_logical_to_effective(vcpu, start);
352 	if (m3 & SSKE_MB) {
353 		/* start already designates an absolute address */
354 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
355 	} else {
356 		start = kvm_s390_real_to_abs(vcpu, start);
357 		end = start + PAGE_SIZE;
358 	}
359 
360 	while (start != end) {
361 		scoped_guard(read_lock, &vcpu->kvm->mmu_lock) {
362 			rc = dat_cond_set_storage_key(vcpu->arch.mc, vcpu->arch.gmap->asce,
363 						      gpa_to_gfn(start), key, &oldkey,
364 						      m3 & SSKE_NQ, m3 & SSKE_MR, m3 & SSKE_MC);
365 		}
366 		if (rc > 1)
367 			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
368 		if (rc == -ENOMEM) {
369 			kvm_s390_mmu_cache_topup(vcpu->arch.mc);
370 			continue;
371 		}
372 		if (rc < 0)
373 			return rc;
374 		start += PAGE_SIZE;
375 	}
376 
377 	if (m3 & (SSKE_MC | SSKE_MR)) {
378 		if (m3 & SSKE_MB) {
379 			/* skey in reg1 is unpredictable */
380 			kvm_s390_set_psw_cc(vcpu, 3);
381 		} else {
382 			kvm_s390_set_psw_cc(vcpu, rc);
383 			vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
384 			vcpu->run->s.regs.gprs[reg1] |= (u64)oldkey.skey << 8;
385 		}
386 	}
387 	if (m3 & SSKE_MB) {
388 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
389 			vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
390 		else
391 			vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
392 		end = kvm_s390_logical_to_effective(vcpu, end);
393 		vcpu->run->s.regs.gprs[reg2] |= end;
394 	}
395 	return 0;
396 }
397 
398 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
399 {
400 	vcpu->stat.instruction_ipte_interlock++;
401 	if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
402 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
403 	wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu->kvm));
404 	kvm_s390_retry_instr(vcpu);
405 	VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
406 	return 0;
407 }
408 
409 static int handle_test_block(struct kvm_vcpu *vcpu)
410 {
411 	gpa_t addr;
412 	int reg2;
413 
414 	vcpu->stat.instruction_tb++;
415 
416 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
417 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
418 
419 	kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
420 	addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
421 	addr = kvm_s390_logical_to_effective(vcpu, addr);
422 	if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
423 		return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
424 	addr = kvm_s390_real_to_abs(vcpu, addr);
425 
426 	if (!kvm_is_gpa_in_memslot(vcpu->kvm, addr))
427 		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
428 	/*
429 	 * We don't expect errors on modern systems, and do not care
430 	 * about storage keys (yet), so let's just clear the page.
431 	 */
432 	if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
433 		return -EFAULT;
434 	kvm_s390_set_psw_cc(vcpu, 0);
435 	vcpu->run->s.regs.gprs[0] = 0;
436 	return 0;
437 }
438 
439 static int handle_tpi(struct kvm_vcpu *vcpu)
440 {
441 	struct kvm_s390_interrupt_info *inti;
442 	unsigned long len;
443 	u32 tpi_data[3];
444 	int rc;
445 	u64 addr;
446 	u8 ar;
447 
448 	vcpu->stat.instruction_tpi++;
449 
450 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
451 	if (addr & 3)
452 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
453 
454 	inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
455 	if (!inti) {
456 		kvm_s390_set_psw_cc(vcpu, 0);
457 		return 0;
458 	}
459 
460 	tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
461 	tpi_data[1] = inti->io.io_int_parm;
462 	tpi_data[2] = inti->io.io_int_word;
463 	if (addr) {
464 		/*
465 		 * Store the two-word I/O interruption code into the
466 		 * provided area.
467 		 */
468 		len = sizeof(tpi_data) - 4;
469 		rc = write_guest(vcpu, addr, ar, &tpi_data, len);
470 		if (rc) {
471 			rc = kvm_s390_inject_prog_cond(vcpu, rc);
472 			goto reinject_interrupt;
473 		}
474 	} else {
475 		/*
476 		 * Store the three-word I/O interruption code into
477 		 * the appropriate lowcore area.
478 		 */
479 		len = sizeof(tpi_data);
480 		if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
481 			/* failed writes to the low core are not recoverable */
482 			rc = -EFAULT;
483 			goto reinject_interrupt;
484 		}
485 	}
486 
487 	/* irq was successfully handed to the guest */
488 	kfree(inti);
489 	kvm_s390_set_psw_cc(vcpu, 1);
490 	return 0;
491 reinject_interrupt:
492 	/*
493 	 * If we encounter a problem storing the interruption code, the
494 	 * instruction is suppressed from the guest's view: reinject the
495 	 * interrupt.
496 	 */
497 	if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
498 		kfree(inti);
499 		rc = -EFAULT;
500 	}
501 	/* don't set the cc, a pgm irq was injected or we drop to user space */
502 	return rc ? -EFAULT : 0;
503 }
504 
505 static int handle_tsch(struct kvm_vcpu *vcpu)
506 {
507 	struct kvm_s390_interrupt_info *inti = NULL;
508 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
509 
510 	vcpu->stat.instruction_tsch++;
511 
512 	/* a valid schid has at least one bit set */
513 	if (vcpu->run->s.regs.gprs[1])
514 		inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
515 					   vcpu->run->s.regs.gprs[1]);
516 
517 	/*
518 	 * Prepare exit to userspace.
519 	 * We indicate whether we dequeued a pending I/O interrupt
520 	 * so that userspace can re-inject it if the instruction gets
521 	 * a program check. While this may re-order the pending I/O
522 	 * interrupts, this is no problem since the priority is kept
523 	 * intact.
524 	 */
525 	vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
526 	vcpu->run->s390_tsch.dequeued = !!inti;
527 	if (inti) {
528 		vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
529 		vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
530 		vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
531 		vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
532 	}
533 	vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
534 	kfree(inti);
535 	return -EREMOTE;
536 }
537 
538 static int handle_io_inst(struct kvm_vcpu *vcpu)
539 {
540 	VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
541 
542 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
543 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
544 
545 	if (vcpu->kvm->arch.css_support) {
546 		/*
547 		 * Most I/O instructions will be handled by userspace.
548 		 * Exceptions are tpi and the interrupt portion of tsch.
549 		 */
550 		if (vcpu->arch.sie_block->ipa == 0xb236)
551 			return handle_tpi(vcpu);
552 		if (vcpu->arch.sie_block->ipa == 0xb235)
553 			return handle_tsch(vcpu);
554 		/* Handle in userspace. */
555 		vcpu->stat.instruction_io_other++;
556 		return -EOPNOTSUPP;
557 	} else {
558 		/*
559 		 * Set condition code 3 to stop the guest from issuing channel
560 		 * I/O instructions.
561 		 */
562 		kvm_s390_set_psw_cc(vcpu, 3);
563 		return 0;
564 	}
565 }
566 
567 #if IS_ENABLED(CONFIG_VFIO_AP)
568 bool kvm_s390_is_gpa_in_memslot(struct kvm *kvm, gpa_t gpa)
569 {
570 	return kvm_is_gpa_in_memslot(kvm, gpa);
571 }
572 EXPORT_SYMBOL_FOR_MODULES(kvm_s390_is_gpa_in_memslot, "vfio_ap");
573 #endif
574 
575 /*
576  * handle_pqap: Handling pqap interception
577  * @vcpu: the vcpu having issue the pqap instruction
578  *
579  * We now support PQAP/AQIC instructions and we need to correctly
580  * answer the guest even if no dedicated driver's hook is available.
581  *
582  * The intercepting code calls a dedicated callback for this instruction
583  * if a driver did register one in the CRYPTO satellite of the
584  * SIE block.
585  *
586  * If no callback is available, the queues are not available, return this
587  * response code to the caller and set CC to 3.
588  * Else return the response code returned by the callback.
589  */
590 static int handle_pqap(struct kvm_vcpu *vcpu)
591 {
592 	struct ap_queue_status status = {};
593 	crypto_hook pqap_hook;
594 	unsigned long reg0;
595 	int ret;
596 	uint8_t fc;
597 
598 	/* Verify that the AP instruction are available */
599 	if (!ap_instructions_available())
600 		return -EOPNOTSUPP;
601 	/* Verify that the guest is allowed to use AP instructions */
602 	if (!(vcpu->arch.sie_block->eca & ECA_APIE))
603 		return -EOPNOTSUPP;
604 	/*
605 	 * The only possibly intercepted functions when AP instructions are
606 	 * available for the guest are AQIC and TAPQ with the t bit set
607 	 * since we do not set IC.3 (FIII) we currently will only intercept
608 	 * the AQIC function code.
609 	 * Note: running nested under z/VM can result in intercepts for other
610 	 * function codes, e.g. PQAP(QCI). We do not support this and bail out.
611 	 */
612 	reg0 = vcpu->run->s.regs.gprs[0];
613 	fc = (reg0 >> 24) & 0xff;
614 	if (fc != 0x03)
615 		return -EOPNOTSUPP;
616 
617 	/* PQAP instruction is allowed for guest kernel only */
618 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
619 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
620 
621 	/* Common PQAP instruction specification exceptions */
622 	/* bits 41-47 must all be zeros */
623 	if (reg0 & 0x007f0000UL)
624 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
625 	/* APFT not install and T bit set */
626 	if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
627 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
628 	/* APXA not installed and APID greater 64 or APQI greater 16 */
629 	if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
630 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
631 
632 	/* AQIC function code specific exception */
633 	/* facility 65 not present for AQIC function code */
634 	if (!test_kvm_facility(vcpu->kvm, 65))
635 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
636 
637 	/*
638 	 * If the hook callback is registered, there will be a pointer to the
639 	 * hook function pointer in the kvm_s390_crypto structure. Lock the
640 	 * owner, retrieve the hook function pointer and call the hook.
641 	 */
642 	down_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
643 	if (vcpu->kvm->arch.crypto.pqap_hook) {
644 		pqap_hook = *vcpu->kvm->arch.crypto.pqap_hook;
645 		ret = pqap_hook(vcpu);
646 		if (!ret) {
647 			if (vcpu->run->s.regs.gprs[1] & 0x00ff0000)
648 				kvm_s390_set_psw_cc(vcpu, 3);
649 			else
650 				kvm_s390_set_psw_cc(vcpu, 0);
651 		}
652 		up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
653 		return ret;
654 	}
655 	up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
656 	/*
657 	 * A vfio_driver must register a hook.
658 	 * No hook means no driver to enable the SIE CRYCB and no queues.
659 	 * We send this response to the guest.
660 	 */
661 	status.response_code = 0x01;
662 	memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
663 	kvm_s390_set_psw_cc(vcpu, 3);
664 	return 0;
665 }
666 
667 static int handle_stfl(struct kvm_vcpu *vcpu)
668 {
669 	int rc;
670 	unsigned int fac;
671 
672 	vcpu->stat.instruction_stfl++;
673 
674 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
675 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
676 
677 	/*
678 	 * We need to shift the lower 32 facility bits (bit 0-31) from a u64
679 	 * into a u32 memory representation. They will remain bits 0-31.
680 	 */
681 	fac = *vcpu->kvm->arch.model.fac_list >> 32;
682 	rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
683 			    &fac, sizeof(fac));
684 	if (rc)
685 		return rc;
686 	VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
687 	trace_kvm_s390_handle_stfl(vcpu, fac);
688 	return 0;
689 }
690 
691 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
692 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
693 #define PSW_ADDR_24 0x0000000000ffffffUL
694 #define PSW_ADDR_31 0x000000007fffffffUL
695 
696 int is_valid_psw(psw_t *psw)
697 {
698 	if (psw->mask & PSW_MASK_UNASSIGNED)
699 		return 0;
700 	if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
701 		if (psw->addr & ~PSW_ADDR_31)
702 			return 0;
703 	}
704 	if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
705 		return 0;
706 	if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
707 		return 0;
708 	if (psw->addr & 1)
709 		return 0;
710 	return 1;
711 }
712 
713 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
714 {
715 	psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
716 	psw32_t new_psw;
717 	u64 addr, iaddr;
718 	int rc;
719 	u8 ar;
720 
721 	vcpu->stat.instruction_lpsw++;
722 
723 	iaddr = gpsw->addr - kvm_s390_get_ilen(vcpu);
724 	if (gpsw->mask & PSW_MASK_PSTATE)
725 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
726 
727 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
728 	if (addr & 7)
729 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
730 
731 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
732 	if (rc)
733 		return kvm_s390_inject_prog_cond(vcpu, rc);
734 	if (!(new_psw.mask & PSW32_MASK_BASE))
735 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
736 	gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
737 	gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
738 	gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
739 	if (!is_valid_psw(gpsw))
740 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
741 	vcpu->arch.sie_block->gbea = iaddr;
742 	return 0;
743 }
744 
745 static int handle_lpswe(struct kvm_vcpu *vcpu)
746 {
747 	psw_t new_psw;
748 	u64 addr, iaddr;
749 	int rc;
750 	u8 ar;
751 
752 	vcpu->stat.instruction_lpswe++;
753 
754 	iaddr = vcpu->arch.sie_block->gpsw.addr - kvm_s390_get_ilen(vcpu);
755 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
756 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
757 
758 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
759 	if (addr & 7)
760 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
761 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
762 	if (rc)
763 		return kvm_s390_inject_prog_cond(vcpu, rc);
764 	vcpu->arch.sie_block->gpsw = new_psw;
765 	if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
766 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
767 	vcpu->arch.sie_block->gbea = iaddr;
768 	return 0;
769 }
770 
771 static int handle_lpswey(struct kvm_vcpu *vcpu)
772 {
773 	psw_t new_psw;
774 	u64 addr;
775 	int rc;
776 	u8 ar;
777 
778 	vcpu->stat.instruction_lpswey++;
779 
780 	if (!test_kvm_facility(vcpu->kvm, 193))
781 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
782 
783 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
784 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
785 
786 	addr = kvm_s390_get_base_disp_siy(vcpu, &ar);
787 	if (addr & 7)
788 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
789 
790 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
791 	if (rc)
792 		return kvm_s390_inject_prog_cond(vcpu, rc);
793 
794 	vcpu->arch.sie_block->gpsw = new_psw;
795 	if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
796 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
797 
798 	return 0;
799 }
800 
801 static int handle_stidp(struct kvm_vcpu *vcpu)
802 {
803 	u64 stidp_data = vcpu->kvm->arch.model.cpuid;
804 	u64 operand2;
805 	int rc;
806 	u8 ar;
807 
808 	vcpu->stat.instruction_stidp++;
809 
810 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
811 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
812 
813 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
814 
815 	if (operand2 & 7)
816 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
817 
818 	rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
819 	if (rc)
820 		return kvm_s390_inject_prog_cond(vcpu, rc);
821 
822 	VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
823 	return 0;
824 }
825 
826 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
827 {
828 	int cpus = 0;
829 	int n;
830 
831 	cpus = atomic_read(&vcpu->kvm->online_vcpus);
832 
833 	/* deal with other level 3 hypervisors */
834 	if (stsi(mem, 3, 2, 2))
835 		mem->count = 0;
836 	if (mem->count < 8)
837 		mem->count++;
838 	for (n = mem->count - 1; n > 0 ; n--)
839 		memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
840 
841 	memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
842 	mem->vm[0].cpus_total = cpus;
843 	mem->vm[0].cpus_configured = cpus;
844 	mem->vm[0].cpus_standby = 0;
845 	mem->vm[0].cpus_reserved = 0;
846 	mem->vm[0].caf = 1000;
847 	memcpy(mem->vm[0].name, "KVMguest", 8);
848 	ASCEBC(mem->vm[0].name, 8);
849 	memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
850 	ASCEBC(mem->vm[0].cpi, 16);
851 }
852 
853 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
854 				 u8 fc, u8 sel1, u16 sel2)
855 {
856 	vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
857 	vcpu->run->s390_stsi.addr = addr;
858 	vcpu->run->s390_stsi.ar = ar;
859 	vcpu->run->s390_stsi.fc = fc;
860 	vcpu->run->s390_stsi.sel1 = sel1;
861 	vcpu->run->s390_stsi.sel2 = sel2;
862 }
863 
864 static int handle_stsi(struct kvm_vcpu *vcpu)
865 {
866 	int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
867 	int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
868 	int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
869 	unsigned long mem = 0;
870 	u64 operand2;
871 	int rc = 0;
872 	u8 ar;
873 
874 	vcpu->stat.instruction_stsi++;
875 	VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
876 
877 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
878 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
879 
880 	/* Bailout forbidden function codes */
881 	if (fc > 3 && fc != 15)
882 		goto out_no_data;
883 
884 	/*
885 	 * fc 15 is provided only with
886 	 *   - PTF/CPU topology support through facility 15
887 	 *   - KVM_CAP_S390_USER_STSI
888 	 */
889 	if (fc == 15 && (!test_kvm_facility(vcpu->kvm, 11) ||
890 			 !vcpu->kvm->arch.user_stsi))
891 		goto out_no_data;
892 
893 	if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
894 	    || vcpu->run->s.regs.gprs[1] & 0xffff0000)
895 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
896 
897 	if (fc == 0) {
898 		vcpu->run->s.regs.gprs[0] = 3 << 28;
899 		kvm_s390_set_psw_cc(vcpu, 0);
900 		return 0;
901 	}
902 
903 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
904 
905 	if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff))
906 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
907 
908 	switch (fc) {
909 	case 1: /* same handling for 1 and 2 */
910 	case 2:
911 		mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
912 		if (!mem)
913 			goto out_no_data;
914 		if (stsi((void *) mem, fc, sel1, sel2))
915 			goto out_no_data;
916 		break;
917 	case 3:
918 		if (sel1 != 2 || sel2 != 2)
919 			goto out_no_data;
920 		mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
921 		if (!mem)
922 			goto out_no_data;
923 		handle_stsi_3_2_2(vcpu, (void *) mem);
924 		break;
925 	case 15: /* fc 15 is fully handled in userspace */
926 		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
927 		trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
928 		return -EREMOTE;
929 	}
930 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
931 		memcpy(sida_addr(vcpu->arch.sie_block), (void *)mem, PAGE_SIZE);
932 		rc = 0;
933 	} else {
934 		rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
935 	}
936 	if (rc) {
937 		rc = kvm_s390_inject_prog_cond(vcpu, rc);
938 		goto out;
939 	}
940 	if (vcpu->kvm->arch.user_stsi) {
941 		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
942 		rc = -EREMOTE;
943 	}
944 	trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
945 	free_page(mem);
946 	kvm_s390_set_psw_cc(vcpu, 0);
947 	vcpu->run->s.regs.gprs[0] = 0;
948 	return rc;
949 out_no_data:
950 	kvm_s390_set_psw_cc(vcpu, 3);
951 out:
952 	free_page(mem);
953 	return rc;
954 }
955 
956 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
957 {
958 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
959 	case 0x02:
960 		return handle_stidp(vcpu);
961 	case 0x04:
962 		return handle_set_clock(vcpu);
963 	case 0x10:
964 		return handle_set_prefix(vcpu);
965 	case 0x11:
966 		return handle_store_prefix(vcpu);
967 	case 0x12:
968 		return handle_store_cpu_address(vcpu);
969 	case 0x14:
970 		return kvm_s390_handle_vsie(vcpu);
971 	case 0x21:
972 	case 0x50:
973 		return handle_ipte_interlock(vcpu);
974 	case 0x29:
975 		return handle_iske(vcpu);
976 	case 0x2a:
977 		return handle_rrbe(vcpu);
978 	case 0x2b:
979 		return handle_sske(vcpu);
980 	case 0x2c:
981 		return handle_test_block(vcpu);
982 	case 0x30:
983 	case 0x31:
984 	case 0x32:
985 	case 0x33:
986 	case 0x34:
987 	case 0x35:
988 	case 0x36:
989 	case 0x37:
990 	case 0x38:
991 	case 0x39:
992 	case 0x3a:
993 	case 0x3b:
994 	case 0x3c:
995 	case 0x5f:
996 	case 0x74:
997 	case 0x76:
998 		return handle_io_inst(vcpu);
999 	case 0x56:
1000 		return handle_sthyi(vcpu);
1001 	case 0x7d:
1002 		return handle_stsi(vcpu);
1003 	case 0xaf:
1004 		return handle_pqap(vcpu);
1005 	case 0xb1:
1006 		return handle_stfl(vcpu);
1007 	case 0xb2:
1008 		return handle_lpswe(vcpu);
1009 	default:
1010 		return -EOPNOTSUPP;
1011 	}
1012 }
1013 
1014 static int handle_epsw(struct kvm_vcpu *vcpu)
1015 {
1016 	int reg1, reg2;
1017 
1018 	vcpu->stat.instruction_epsw++;
1019 
1020 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1021 
1022 	/* This basically extracts the mask half of the psw. */
1023 	vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
1024 	vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
1025 	if (reg2) {
1026 		vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
1027 		vcpu->run->s.regs.gprs[reg2] |=
1028 			vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
1029 	}
1030 	return 0;
1031 }
1032 
1033 #define PFMF_RESERVED   0xfffc0101UL
1034 #define PFMF_SK         0x00020000UL
1035 #define PFMF_CF         0x00010000UL
1036 #define PFMF_UI         0x00008000UL
1037 #define PFMF_FSC        0x00007000UL
1038 #define PFMF_NQ         0x00000800UL
1039 #define PFMF_MR         0x00000400UL
1040 #define PFMF_MC         0x00000200UL
1041 #define PFMF_KEY        0x000000feUL
1042 
1043 static int handle_pfmf(struct kvm_vcpu *vcpu)
1044 {
1045 	bool mr = false, mc = false, nq;
1046 	int reg1, reg2;
1047 	unsigned long start, end;
1048 	union skey key;
1049 
1050 	vcpu->stat.instruction_pfmf++;
1051 
1052 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1053 
1054 	if (!test_kvm_facility(vcpu->kvm, 8))
1055 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1056 
1057 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1058 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1059 
1060 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1061 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1062 
1063 	/* Only provide non-quiescing support if enabled for the guest */
1064 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1065 	    !test_kvm_facility(vcpu->kvm, 14))
1066 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1067 
1068 	/* Only provide conditional-SSKE support if enabled for the guest */
1069 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1070 	    test_kvm_facility(vcpu->kvm, 10)) {
1071 		mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1072 		mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1073 	}
1074 
1075 	nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1076 	key.skey = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1077 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1078 	start = kvm_s390_logical_to_effective(vcpu, start);
1079 
1080 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1081 		if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1082 			return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1083 	}
1084 
1085 	switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1086 	case 0x00000000:
1087 		/* only 4k frames specify a real address */
1088 		start = kvm_s390_real_to_abs(vcpu, start);
1089 		end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1090 		break;
1091 	case 0x00001000:
1092 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1093 		break;
1094 	case 0x00002000:
1095 		/* only support 2G frame size if EDAT2 is available and we are
1096 		   not in 24-bit addressing mode */
1097 		if (!test_kvm_facility(vcpu->kvm, 78) ||
1098 		    psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1099 			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1100 		end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1101 		break;
1102 	default:
1103 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1104 	}
1105 
1106 	while (start != end) {
1107 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1108 			if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1109 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1110 		}
1111 
1112 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1113 			int rc = kvm_s390_skey_check_enable(vcpu);
1114 
1115 			if (rc)
1116 				return rc;
1117 			scoped_guard(read_lock, &vcpu->kvm->mmu_lock) {
1118 				rc = dat_cond_set_storage_key(vcpu->arch.mc, vcpu->arch.gmap->asce,
1119 							      gpa_to_gfn(start), key,
1120 							      NULL, nq, mr, mc);
1121 			}
1122 			if (rc > 1)
1123 				return kvm_s390_inject_program_int(vcpu, rc);
1124 			if (rc == -ENOMEM) {
1125 				kvm_s390_mmu_cache_topup(vcpu->arch.mc);
1126 				continue;
1127 			}
1128 			if (rc < 0)
1129 				return rc;
1130 		}
1131 		start += PAGE_SIZE;
1132 	}
1133 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1134 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1135 			vcpu->run->s.regs.gprs[reg2] = end;
1136 		} else {
1137 			vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1138 			end = kvm_s390_logical_to_effective(vcpu, end);
1139 			vcpu->run->s.regs.gprs[reg2] |= end;
1140 		}
1141 	}
1142 	return 0;
1143 }
1144 
1145 /*
1146  * Must be called with relevant read locks held (kvm->mm->mmap_lock, kvm->srcu)
1147  */
1148 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1149 {
1150 	int r1, r2, nappended, entries;
1151 	union essa_state state;
1152 	unsigned long *cbrlo;
1153 	unsigned long gfn;
1154 	bool dirtied;
1155 
1156 	/*
1157 	 * We don't need to set SD.FPF.SK to 1 here, because if we have a
1158 	 * machine check here we either handle it or crash
1159 	 */
1160 
1161 	kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1162 	gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1163 	entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1164 
1165 	nappended = dat_perform_essa(vcpu->arch.gmap->asce, gfn, orc, &state, &dirtied);
1166 	vcpu->run->s.regs.gprs[r1] = state.val;
1167 	if (nappended < 0)
1168 		return 0;
1169 	/*
1170 	 * It is possible that all the normal 511 slots were full, in which case
1171 	 * we will now write in the 512th slot, which is reserved for host use.
1172 	 * In both cases we let the normal essa handling code process all the
1173 	 * slots, including the reserved one, if needed.
1174 	 */
1175 	if (nappended > 0) {
1176 		cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1177 		cbrlo[entries] = gfn << PAGE_SHIFT;
1178 	}
1179 
1180 	if (dirtied)
1181 		atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1182 
1183 	return nappended;
1184 }
1185 
1186 static void _essa_clear_cbrl(struct kvm_vcpu *vcpu, unsigned long *cbrl, int len)
1187 {
1188 	union crste *crstep;
1189 	union pgste pgste;
1190 	union pte *ptep;
1191 	int i;
1192 
1193 	lockdep_assert_held(&vcpu->kvm->mmu_lock);
1194 
1195 	for (i = 0; i < len; i++) {
1196 		if (dat_entry_walk(NULL, gpa_to_gfn(cbrl[i]), vcpu->arch.gmap->asce,
1197 				   0, TABLE_TYPE_PAGE_TABLE, &crstep, &ptep))
1198 			continue;
1199 		if (!ptep || ptep->s.pr)
1200 			continue;
1201 		pgste = pgste_get_lock(ptep);
1202 		if (pgste.usage == PGSTE_GPS_USAGE_UNUSED || pgste.zero)
1203 			gmap_helper_zap_one_page(vcpu->kvm->mm, cbrl[i]);
1204 		pgste_set_unlock(ptep, pgste);
1205 	}
1206 }
1207 
1208 static int handle_essa(struct kvm_vcpu *vcpu)
1209 {
1210 	lockdep_assert_held(&vcpu->kvm->srcu);
1211 
1212 	/* entries expected to be 1FF */
1213 	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1214 	unsigned long *cbrlo;
1215 	int i, orc;
1216 
1217 	VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1218 	vcpu->stat.instruction_essa++;
1219 	if (!vcpu->kvm->arch.use_cmma)
1220 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1221 
1222 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1223 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1224 	/* Check for invalid operation request code */
1225 	orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1226 	/* ORCs 0-6 are always valid */
1227 	if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1228 						: ESSA_SET_STABLE_IF_RESIDENT))
1229 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1230 
1231 	if (!vcpu->kvm->arch.migration_mode) {
1232 		/*
1233 		 * CMMA is enabled in the KVM settings, but is disabled in
1234 		 * the SIE block and in the mm_context, and we are not doing
1235 		 * a migration. Enable CMMA in the mm_context.
1236 		 * Since we need to take a write lock to write to the context
1237 		 * to avoid races with storage keys handling, we check if the
1238 		 * value really needs to be written to; if the value is
1239 		 * already correct, we do nothing and avoid the lock.
1240 		 */
1241 		set_bit(GMAP_FLAG_USES_CMM, &vcpu->arch.gmap->flags);
1242 		/*
1243 		 * If we are here, we are supposed to have CMMA enabled in
1244 		 * the SIE block. Enabling CMMA works on a per-CPU basis,
1245 		 * while the context use_cmma flag is per process.
1246 		 * It's possible that the context flag is enabled and the
1247 		 * SIE flag is not, so we set the flag always; if it was
1248 		 * already set, nothing changes, otherwise we enable it
1249 		 * on this CPU too.
1250 		 */
1251 		vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1252 		/* Retry the ESSA instruction */
1253 		kvm_s390_retry_instr(vcpu);
1254 	} else {
1255 		scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
1256 			i = __do_essa(vcpu, orc);
1257 		if (i < 0)
1258 			return i;
1259 		/* Account for the possible extra cbrl entry */
1260 		entries += i;
1261 	}
1262 	/* reset nceo */
1263 	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;
1264 	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1265 
1266 	mmap_read_lock(vcpu->kvm->mm);
1267 	scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
1268 		_essa_clear_cbrl(vcpu, cbrlo, entries);
1269 	mmap_read_unlock(vcpu->kvm->mm);
1270 
1271 	return 0;
1272 }
1273 
1274 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1275 {
1276 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1277 	case 0x8a:
1278 	case 0x8e:
1279 	case 0x8f:
1280 		return handle_ipte_interlock(vcpu);
1281 	case 0x8d:
1282 		return handle_epsw(vcpu);
1283 	case 0xab:
1284 		return handle_essa(vcpu);
1285 	case 0xaf:
1286 		return handle_pfmf(vcpu);
1287 	default:
1288 		return -EOPNOTSUPP;
1289 	}
1290 }
1291 
1292 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1293 {
1294 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1295 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1296 	int reg, rc, nr_regs;
1297 	u32 ctl_array[16];
1298 	u64 ga;
1299 	u8 ar;
1300 
1301 	vcpu->stat.instruction_lctl++;
1302 
1303 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1304 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1305 
1306 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1307 
1308 	if (ga & 3)
1309 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1310 
1311 	VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1312 	trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1313 
1314 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1315 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1316 	if (rc)
1317 		return kvm_s390_inject_prog_cond(vcpu, rc);
1318 	reg = reg1;
1319 	nr_regs = 0;
1320 	do {
1321 		vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1322 		vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1323 		if (reg == reg3)
1324 			break;
1325 		reg = (reg + 1) % 16;
1326 	} while (1);
1327 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1328 	return 0;
1329 }
1330 
1331 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1332 {
1333 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1334 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1335 	int reg, rc, nr_regs;
1336 	u32 ctl_array[16];
1337 	u64 ga;
1338 	u8 ar;
1339 
1340 	vcpu->stat.instruction_stctl++;
1341 
1342 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1343 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1344 
1345 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1346 
1347 	if (ga & 3)
1348 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1349 
1350 	VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1351 	trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1352 
1353 	reg = reg1;
1354 	nr_regs = 0;
1355 	do {
1356 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1357 		if (reg == reg3)
1358 			break;
1359 		reg = (reg + 1) % 16;
1360 	} while (1);
1361 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1362 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1363 }
1364 
1365 static int handle_lctlg(struct kvm_vcpu *vcpu)
1366 {
1367 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1368 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1369 	int reg, rc, nr_regs;
1370 	u64 ctl_array[16];
1371 	u64 ga;
1372 	u8 ar;
1373 
1374 	vcpu->stat.instruction_lctlg++;
1375 
1376 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1377 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1378 
1379 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1380 
1381 	if (ga & 7)
1382 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1383 
1384 	VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1385 	trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1386 
1387 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1388 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1389 	if (rc)
1390 		return kvm_s390_inject_prog_cond(vcpu, rc);
1391 	reg = reg1;
1392 	nr_regs = 0;
1393 	do {
1394 		vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1395 		if (reg == reg3)
1396 			break;
1397 		reg = (reg + 1) % 16;
1398 	} while (1);
1399 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1400 	return 0;
1401 }
1402 
1403 static int handle_stctg(struct kvm_vcpu *vcpu)
1404 {
1405 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1406 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1407 	int reg, rc, nr_regs;
1408 	u64 ctl_array[16];
1409 	u64 ga;
1410 	u8 ar;
1411 
1412 	vcpu->stat.instruction_stctg++;
1413 
1414 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1415 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1416 
1417 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1418 
1419 	if (ga & 7)
1420 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1421 
1422 	VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1423 	trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1424 
1425 	reg = reg1;
1426 	nr_regs = 0;
1427 	do {
1428 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1429 		if (reg == reg3)
1430 			break;
1431 		reg = (reg + 1) % 16;
1432 	} while (1);
1433 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1434 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1435 }
1436 
1437 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1438 {
1439 	switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1440 	case 0x25:
1441 		return handle_stctg(vcpu);
1442 	case 0x2f:
1443 		return handle_lctlg(vcpu);
1444 	case 0x60:
1445 	case 0x61:
1446 	case 0x62:
1447 		return handle_ri(vcpu);
1448 	case 0x71:
1449 		return handle_lpswey(vcpu);
1450 	default:
1451 		return -EOPNOTSUPP;
1452 	}
1453 }
1454 
1455 static int handle_tprot(struct kvm_vcpu *vcpu)
1456 {
1457 	u64 address, operand2;
1458 	unsigned long gpa;
1459 	u8 access_key;
1460 	bool writable;
1461 	int ret, cc;
1462 	u8 ar;
1463 
1464 	vcpu->stat.instruction_tprot++;
1465 
1466 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1467 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1468 
1469 	kvm_s390_get_base_disp_sse(vcpu, &address, &operand2, &ar, NULL);
1470 	access_key = (operand2 & 0xf0) >> 4;
1471 
1472 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1473 		ipte_lock(vcpu->kvm);
1474 
1475 	ret = guest_translate_address_with_key(vcpu, address, ar, &gpa,
1476 					       GACC_STORE, access_key);
1477 	if (ret == 0) {
1478 		gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1479 	} else if (ret == PGM_PROTECTION) {
1480 		writable = false;
1481 		/* Write protected? Try again with read-only... */
1482 		ret = guest_translate_address_with_key(vcpu, address, ar, &gpa,
1483 						       GACC_FETCH, access_key);
1484 	}
1485 	if (ret >= 0) {
1486 		cc = -1;
1487 
1488 		/* Fetching permitted; storing permitted */
1489 		if (ret == 0 && writable)
1490 			cc = 0;
1491 		/* Fetching permitted; storing not permitted */
1492 		else if (ret == 0 && !writable)
1493 			cc = 1;
1494 		/* Fetching not permitted; storing not permitted */
1495 		else if (ret == PGM_PROTECTION)
1496 			cc = 2;
1497 		/* Translation not available */
1498 		else if (ret != PGM_ADDRESSING && ret != PGM_TRANSLATION_SPEC)
1499 			cc = 3;
1500 
1501 		if (cc != -1) {
1502 			kvm_s390_set_psw_cc(vcpu, cc);
1503 			ret = 0;
1504 		} else {
1505 			ret = kvm_s390_inject_program_int(vcpu, ret);
1506 		}
1507 	}
1508 
1509 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1510 		ipte_unlock(vcpu->kvm);
1511 	return ret;
1512 }
1513 
1514 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1515 {
1516 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1517 	case 0x01:
1518 		return handle_tprot(vcpu);
1519 	default:
1520 		return -EOPNOTSUPP;
1521 	}
1522 }
1523 
1524 static int handle_sckpf(struct kvm_vcpu *vcpu)
1525 {
1526 	u32 value;
1527 
1528 	vcpu->stat.instruction_sckpf++;
1529 
1530 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1531 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1532 
1533 	if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1534 		return kvm_s390_inject_program_int(vcpu,
1535 						   PGM_SPECIFICATION);
1536 
1537 	value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1538 	vcpu->arch.sie_block->todpr = value;
1539 
1540 	return 0;
1541 }
1542 
1543 static int handle_ptff(struct kvm_vcpu *vcpu)
1544 {
1545 	vcpu->stat.instruction_ptff++;
1546 
1547 	/* we don't emulate any control instructions yet */
1548 	kvm_s390_set_psw_cc(vcpu, 3);
1549 	return 0;
1550 }
1551 
1552 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1553 {
1554 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1555 	case 0x04:
1556 		return handle_ptff(vcpu);
1557 	case 0x07:
1558 		return handle_sckpf(vcpu);
1559 	default:
1560 		return -EOPNOTSUPP;
1561 	}
1562 }
1563