xref: /linux/arch/s390/kvm/priv.c (revision 1fd1dc41724319406b0aff221a352a400b0ddfc5)
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;
718 	int rc;
719 	u8 ar;
720 
721 	vcpu->stat.instruction_lpsw++;
722 
723 	if (gpsw->mask & PSW_MASK_PSTATE)
724 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
725 
726 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
727 	if (addr & 7)
728 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
729 
730 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
731 	if (rc)
732 		return kvm_s390_inject_prog_cond(vcpu, rc);
733 	if (!(new_psw.mask & PSW32_MASK_BASE))
734 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
735 	gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
736 	gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
737 	gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
738 	if (!is_valid_psw(gpsw))
739 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
740 	return 0;
741 }
742 
743 static int handle_lpswe(struct kvm_vcpu *vcpu)
744 {
745 	psw_t new_psw;
746 	u64 addr;
747 	int rc;
748 	u8 ar;
749 
750 	vcpu->stat.instruction_lpswe++;
751 
752 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
753 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
754 
755 	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
756 	if (addr & 7)
757 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
758 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
759 	if (rc)
760 		return kvm_s390_inject_prog_cond(vcpu, rc);
761 	vcpu->arch.sie_block->gpsw = new_psw;
762 	if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
763 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
764 	return 0;
765 }
766 
767 static int handle_lpswey(struct kvm_vcpu *vcpu)
768 {
769 	psw_t new_psw;
770 	u64 addr;
771 	int rc;
772 	u8 ar;
773 
774 	vcpu->stat.instruction_lpswey++;
775 
776 	if (!test_kvm_facility(vcpu->kvm, 193))
777 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
778 
779 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
780 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
781 
782 	addr = kvm_s390_get_base_disp_siy(vcpu, &ar);
783 	if (addr & 7)
784 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
785 
786 	rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
787 	if (rc)
788 		return kvm_s390_inject_prog_cond(vcpu, rc);
789 
790 	vcpu->arch.sie_block->gpsw = new_psw;
791 	if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
792 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
793 
794 	return 0;
795 }
796 
797 static int handle_stidp(struct kvm_vcpu *vcpu)
798 {
799 	u64 stidp_data = vcpu->kvm->arch.model.cpuid;
800 	u64 operand2;
801 	int rc;
802 	u8 ar;
803 
804 	vcpu->stat.instruction_stidp++;
805 
806 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
807 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
808 
809 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
810 
811 	if (operand2 & 7)
812 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
813 
814 	rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
815 	if (rc)
816 		return kvm_s390_inject_prog_cond(vcpu, rc);
817 
818 	VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
819 	return 0;
820 }
821 
822 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
823 {
824 	int cpus = 0;
825 	int n;
826 
827 	cpus = atomic_read(&vcpu->kvm->online_vcpus);
828 
829 	/* deal with other level 3 hypervisors */
830 	if (stsi(mem, 3, 2, 2))
831 		mem->count = 0;
832 	if (mem->count < 8)
833 		mem->count++;
834 	for (n = mem->count - 1; n > 0 ; n--)
835 		memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
836 
837 	memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
838 	mem->vm[0].cpus_total = cpus;
839 	mem->vm[0].cpus_configured = cpus;
840 	mem->vm[0].cpus_standby = 0;
841 	mem->vm[0].cpus_reserved = 0;
842 	mem->vm[0].caf = 1000;
843 	memcpy(mem->vm[0].name, "KVMguest", 8);
844 	ASCEBC(mem->vm[0].name, 8);
845 	memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
846 	ASCEBC(mem->vm[0].cpi, 16);
847 }
848 
849 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
850 				 u8 fc, u8 sel1, u16 sel2)
851 {
852 	vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
853 	vcpu->run->s390_stsi.addr = addr;
854 	vcpu->run->s390_stsi.ar = ar;
855 	vcpu->run->s390_stsi.fc = fc;
856 	vcpu->run->s390_stsi.sel1 = sel1;
857 	vcpu->run->s390_stsi.sel2 = sel2;
858 }
859 
860 static int handle_stsi(struct kvm_vcpu *vcpu)
861 {
862 	int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
863 	int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
864 	int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
865 	unsigned long mem = 0;
866 	u64 operand2;
867 	int rc = 0;
868 	u8 ar;
869 
870 	vcpu->stat.instruction_stsi++;
871 	VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
872 
873 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
874 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
875 
876 	/* Bailout forbidden function codes */
877 	if (fc > 3 && fc != 15)
878 		goto out_no_data;
879 
880 	/*
881 	 * fc 15 is provided only with
882 	 *   - PTF/CPU topology support through facility 15
883 	 *   - KVM_CAP_S390_USER_STSI
884 	 */
885 	if (fc == 15 && (!test_kvm_facility(vcpu->kvm, 11) ||
886 			 !vcpu->kvm->arch.user_stsi))
887 		goto out_no_data;
888 
889 	if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
890 	    || vcpu->run->s.regs.gprs[1] & 0xffff0000)
891 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
892 
893 	if (fc == 0) {
894 		vcpu->run->s.regs.gprs[0] = 3 << 28;
895 		kvm_s390_set_psw_cc(vcpu, 0);
896 		return 0;
897 	}
898 
899 	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
900 
901 	if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff))
902 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
903 
904 	switch (fc) {
905 	case 1: /* same handling for 1 and 2 */
906 	case 2:
907 		mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
908 		if (!mem)
909 			goto out_no_data;
910 		if (stsi((void *) mem, fc, sel1, sel2))
911 			goto out_no_data;
912 		break;
913 	case 3:
914 		if (sel1 != 2 || sel2 != 2)
915 			goto out_no_data;
916 		mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
917 		if (!mem)
918 			goto out_no_data;
919 		handle_stsi_3_2_2(vcpu, (void *) mem);
920 		break;
921 	case 15: /* fc 15 is fully handled in userspace */
922 		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
923 		trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
924 		return -EREMOTE;
925 	}
926 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
927 		memcpy(sida_addr(vcpu->arch.sie_block), (void *)mem, PAGE_SIZE);
928 		rc = 0;
929 	} else {
930 		rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
931 	}
932 	if (rc) {
933 		rc = kvm_s390_inject_prog_cond(vcpu, rc);
934 		goto out;
935 	}
936 	if (vcpu->kvm->arch.user_stsi) {
937 		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
938 		rc = -EREMOTE;
939 	}
940 	trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
941 	free_page(mem);
942 	kvm_s390_set_psw_cc(vcpu, 0);
943 	vcpu->run->s.regs.gprs[0] = 0;
944 	return rc;
945 out_no_data:
946 	kvm_s390_set_psw_cc(vcpu, 3);
947 out:
948 	free_page(mem);
949 	return rc;
950 }
951 
952 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
953 {
954 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
955 	case 0x02:
956 		return handle_stidp(vcpu);
957 	case 0x04:
958 		return handle_set_clock(vcpu);
959 	case 0x10:
960 		return handle_set_prefix(vcpu);
961 	case 0x11:
962 		return handle_store_prefix(vcpu);
963 	case 0x12:
964 		return handle_store_cpu_address(vcpu);
965 	case 0x14:
966 		return kvm_s390_handle_vsie(vcpu);
967 	case 0x21:
968 	case 0x50:
969 		return handle_ipte_interlock(vcpu);
970 	case 0x29:
971 		return handle_iske(vcpu);
972 	case 0x2a:
973 		return handle_rrbe(vcpu);
974 	case 0x2b:
975 		return handle_sske(vcpu);
976 	case 0x2c:
977 		return handle_test_block(vcpu);
978 	case 0x30:
979 	case 0x31:
980 	case 0x32:
981 	case 0x33:
982 	case 0x34:
983 	case 0x35:
984 	case 0x36:
985 	case 0x37:
986 	case 0x38:
987 	case 0x39:
988 	case 0x3a:
989 	case 0x3b:
990 	case 0x3c:
991 	case 0x5f:
992 	case 0x74:
993 	case 0x76:
994 		return handle_io_inst(vcpu);
995 	case 0x56:
996 		return handle_sthyi(vcpu);
997 	case 0x7d:
998 		return handle_stsi(vcpu);
999 	case 0xaf:
1000 		return handle_pqap(vcpu);
1001 	case 0xb1:
1002 		return handle_stfl(vcpu);
1003 	case 0xb2:
1004 		return handle_lpswe(vcpu);
1005 	default:
1006 		return -EOPNOTSUPP;
1007 	}
1008 }
1009 
1010 static int handle_epsw(struct kvm_vcpu *vcpu)
1011 {
1012 	int reg1, reg2;
1013 
1014 	vcpu->stat.instruction_epsw++;
1015 
1016 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1017 
1018 	/* This basically extracts the mask half of the psw. */
1019 	vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
1020 	vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
1021 	if (reg2) {
1022 		vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
1023 		vcpu->run->s.regs.gprs[reg2] |=
1024 			vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
1025 	}
1026 	return 0;
1027 }
1028 
1029 #define PFMF_RESERVED   0xfffc0101UL
1030 #define PFMF_SK         0x00020000UL
1031 #define PFMF_CF         0x00010000UL
1032 #define PFMF_UI         0x00008000UL
1033 #define PFMF_FSC        0x00007000UL
1034 #define PFMF_NQ         0x00000800UL
1035 #define PFMF_MR         0x00000400UL
1036 #define PFMF_MC         0x00000200UL
1037 #define PFMF_KEY        0x000000feUL
1038 
1039 static int handle_pfmf(struct kvm_vcpu *vcpu)
1040 {
1041 	bool mr = false, mc = false, nq;
1042 	int reg1, reg2;
1043 	unsigned long start, end;
1044 	union skey key;
1045 
1046 	vcpu->stat.instruction_pfmf++;
1047 
1048 	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1049 
1050 	if (!test_kvm_facility(vcpu->kvm, 8))
1051 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1052 
1053 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1054 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1055 
1056 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1057 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1058 
1059 	/* Only provide non-quiescing support if enabled for the guest */
1060 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1061 	    !test_kvm_facility(vcpu->kvm, 14))
1062 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1063 
1064 	/* Only provide conditional-SSKE support if enabled for the guest */
1065 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1066 	    test_kvm_facility(vcpu->kvm, 10)) {
1067 		mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1068 		mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1069 	}
1070 
1071 	nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1072 	key.skey = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1073 	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1074 	start = kvm_s390_logical_to_effective(vcpu, start);
1075 
1076 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1077 		if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1078 			return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1079 	}
1080 
1081 	switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1082 	case 0x00000000:
1083 		/* only 4k frames specify a real address */
1084 		start = kvm_s390_real_to_abs(vcpu, start);
1085 		end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1086 		break;
1087 	case 0x00001000:
1088 		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1089 		break;
1090 	case 0x00002000:
1091 		/* only support 2G frame size if EDAT2 is available and we are
1092 		   not in 24-bit addressing mode */
1093 		if (!test_kvm_facility(vcpu->kvm, 78) ||
1094 		    psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1095 			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1096 		end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1097 		break;
1098 	default:
1099 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1100 	}
1101 
1102 	while (start != end) {
1103 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1104 			if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1105 				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1106 		}
1107 
1108 		if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1109 			int rc = kvm_s390_skey_check_enable(vcpu);
1110 
1111 			if (rc)
1112 				return rc;
1113 			scoped_guard(read_lock, &vcpu->kvm->mmu_lock) {
1114 				rc = dat_cond_set_storage_key(vcpu->arch.mc, vcpu->arch.gmap->asce,
1115 							      gpa_to_gfn(start), key,
1116 							      NULL, nq, mr, mc);
1117 			}
1118 			if (rc > 1)
1119 				return kvm_s390_inject_program_int(vcpu, rc);
1120 			if (rc == -ENOMEM) {
1121 				kvm_s390_mmu_cache_topup(vcpu->arch.mc);
1122 				continue;
1123 			}
1124 			if (rc < 0)
1125 				return rc;
1126 		}
1127 		start += PAGE_SIZE;
1128 	}
1129 	if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1130 		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1131 			vcpu->run->s.regs.gprs[reg2] = end;
1132 		} else {
1133 			vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1134 			end = kvm_s390_logical_to_effective(vcpu, end);
1135 			vcpu->run->s.regs.gprs[reg2] |= end;
1136 		}
1137 	}
1138 	return 0;
1139 }
1140 
1141 /*
1142  * Must be called with relevant read locks held (kvm->mm->mmap_lock, kvm->srcu)
1143  */
1144 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1145 {
1146 	int r1, r2, nappended, entries;
1147 	union essa_state state;
1148 	unsigned long *cbrlo;
1149 	unsigned long gfn;
1150 	bool dirtied;
1151 
1152 	/*
1153 	 * We don't need to set SD.FPF.SK to 1 here, because if we have a
1154 	 * machine check here we either handle it or crash
1155 	 */
1156 
1157 	kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1158 	gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1159 	entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1160 
1161 	nappended = dat_perform_essa(vcpu->arch.gmap->asce, gfn, orc, &state, &dirtied);
1162 	vcpu->run->s.regs.gprs[r1] = state.val;
1163 	if (nappended < 0)
1164 		return 0;
1165 	/*
1166 	 * It is possible that all the normal 511 slots were full, in which case
1167 	 * we will now write in the 512th slot, which is reserved for host use.
1168 	 * In both cases we let the normal essa handling code process all the
1169 	 * slots, including the reserved one, if needed.
1170 	 */
1171 	if (nappended > 0) {
1172 		cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1173 		cbrlo[entries] = gfn << PAGE_SHIFT;
1174 	}
1175 
1176 	if (dirtied)
1177 		atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1178 
1179 	return nappended;
1180 }
1181 
1182 static void _essa_clear_cbrl(struct kvm_vcpu *vcpu, unsigned long *cbrl, int len)
1183 {
1184 	union crste *crstep;
1185 	union pgste pgste;
1186 	union pte *ptep;
1187 	int i;
1188 
1189 	lockdep_assert_held(&vcpu->kvm->mmu_lock);
1190 
1191 	for (i = 0; i < len; i++) {
1192 		if (dat_entry_walk(NULL, gpa_to_gfn(cbrl[i]), vcpu->arch.gmap->asce,
1193 				   0, TABLE_TYPE_PAGE_TABLE, &crstep, &ptep))
1194 			continue;
1195 		if (!ptep || ptep->s.pr)
1196 			continue;
1197 		pgste = pgste_get_lock(ptep);
1198 		if (pgste.usage == PGSTE_GPS_USAGE_UNUSED || pgste.zero)
1199 			gmap_helper_zap_one_page(vcpu->kvm->mm, cbrl[i]);
1200 		pgste_set_unlock(ptep, pgste);
1201 	}
1202 }
1203 
1204 static int handle_essa(struct kvm_vcpu *vcpu)
1205 {
1206 	lockdep_assert_held(&vcpu->kvm->srcu);
1207 
1208 	/* entries expected to be 1FF */
1209 	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1210 	unsigned long *cbrlo;
1211 	int i, orc;
1212 
1213 	VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1214 	vcpu->stat.instruction_essa++;
1215 	if (!vcpu->kvm->arch.use_cmma)
1216 		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1217 
1218 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1219 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1220 	/* Check for invalid operation request code */
1221 	orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1222 	/* ORCs 0-6 are always valid */
1223 	if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1224 						: ESSA_SET_STABLE_IF_RESIDENT))
1225 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1226 
1227 	if (!vcpu->kvm->arch.migration_mode) {
1228 		/*
1229 		 * CMMA is enabled in the KVM settings, but is disabled in
1230 		 * the SIE block and in the mm_context, and we are not doing
1231 		 * a migration. Enable CMMA in the mm_context.
1232 		 * Since we need to take a write lock to write to the context
1233 		 * to avoid races with storage keys handling, we check if the
1234 		 * value really needs to be written to; if the value is
1235 		 * already correct, we do nothing and avoid the lock.
1236 		 */
1237 		set_bit(GMAP_FLAG_USES_CMM, &vcpu->arch.gmap->flags);
1238 		/*
1239 		 * If we are here, we are supposed to have CMMA enabled in
1240 		 * the SIE block. Enabling CMMA works on a per-CPU basis,
1241 		 * while the context use_cmma flag is per process.
1242 		 * It's possible that the context flag is enabled and the
1243 		 * SIE flag is not, so we set the flag always; if it was
1244 		 * already set, nothing changes, otherwise we enable it
1245 		 * on this CPU too.
1246 		 */
1247 		vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1248 		/* Retry the ESSA instruction */
1249 		kvm_s390_retry_instr(vcpu);
1250 	} else {
1251 		scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
1252 			i = __do_essa(vcpu, orc);
1253 		if (i < 0)
1254 			return i;
1255 		/* Account for the possible extra cbrl entry */
1256 		entries += i;
1257 	}
1258 	/* reset nceo */
1259 	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;
1260 	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1261 
1262 	mmap_read_lock(vcpu->kvm->mm);
1263 	scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
1264 		_essa_clear_cbrl(vcpu, cbrlo, entries);
1265 	mmap_read_unlock(vcpu->kvm->mm);
1266 
1267 	return 0;
1268 }
1269 
1270 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1271 {
1272 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1273 	case 0x8a:
1274 	case 0x8e:
1275 	case 0x8f:
1276 		return handle_ipte_interlock(vcpu);
1277 	case 0x8d:
1278 		return handle_epsw(vcpu);
1279 	case 0xab:
1280 		return handle_essa(vcpu);
1281 	case 0xaf:
1282 		return handle_pfmf(vcpu);
1283 	default:
1284 		return -EOPNOTSUPP;
1285 	}
1286 }
1287 
1288 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1289 {
1290 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1291 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1292 	int reg, rc, nr_regs;
1293 	u32 ctl_array[16];
1294 	u64 ga;
1295 	u8 ar;
1296 
1297 	vcpu->stat.instruction_lctl++;
1298 
1299 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1300 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1301 
1302 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1303 
1304 	if (ga & 3)
1305 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1306 
1307 	VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1308 	trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1309 
1310 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1311 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1312 	if (rc)
1313 		return kvm_s390_inject_prog_cond(vcpu, rc);
1314 	reg = reg1;
1315 	nr_regs = 0;
1316 	do {
1317 		vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1318 		vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1319 		if (reg == reg3)
1320 			break;
1321 		reg = (reg + 1) % 16;
1322 	} while (1);
1323 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1324 	return 0;
1325 }
1326 
1327 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1328 {
1329 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1330 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1331 	int reg, rc, nr_regs;
1332 	u32 ctl_array[16];
1333 	u64 ga;
1334 	u8 ar;
1335 
1336 	vcpu->stat.instruction_stctl++;
1337 
1338 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1339 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1340 
1341 	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1342 
1343 	if (ga & 3)
1344 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1345 
1346 	VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1347 	trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1348 
1349 	reg = reg1;
1350 	nr_regs = 0;
1351 	do {
1352 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1353 		if (reg == reg3)
1354 			break;
1355 		reg = (reg + 1) % 16;
1356 	} while (1);
1357 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1358 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1359 }
1360 
1361 static int handle_lctlg(struct kvm_vcpu *vcpu)
1362 {
1363 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1364 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1365 	int reg, rc, nr_regs;
1366 	u64 ctl_array[16];
1367 	u64 ga;
1368 	u8 ar;
1369 
1370 	vcpu->stat.instruction_lctlg++;
1371 
1372 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1373 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1374 
1375 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1376 
1377 	if (ga & 7)
1378 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1379 
1380 	VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1381 	trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1382 
1383 	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1384 	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1385 	if (rc)
1386 		return kvm_s390_inject_prog_cond(vcpu, rc);
1387 	reg = reg1;
1388 	nr_regs = 0;
1389 	do {
1390 		vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1391 		if (reg == reg3)
1392 			break;
1393 		reg = (reg + 1) % 16;
1394 	} while (1);
1395 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1396 	return 0;
1397 }
1398 
1399 static int handle_stctg(struct kvm_vcpu *vcpu)
1400 {
1401 	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1402 	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1403 	int reg, rc, nr_regs;
1404 	u64 ctl_array[16];
1405 	u64 ga;
1406 	u8 ar;
1407 
1408 	vcpu->stat.instruction_stctg++;
1409 
1410 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1411 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1412 
1413 	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1414 
1415 	if (ga & 7)
1416 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1417 
1418 	VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1419 	trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1420 
1421 	reg = reg1;
1422 	nr_regs = 0;
1423 	do {
1424 		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1425 		if (reg == reg3)
1426 			break;
1427 		reg = (reg + 1) % 16;
1428 	} while (1);
1429 	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1430 	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1431 }
1432 
1433 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1434 {
1435 	switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1436 	case 0x25:
1437 		return handle_stctg(vcpu);
1438 	case 0x2f:
1439 		return handle_lctlg(vcpu);
1440 	case 0x60:
1441 	case 0x61:
1442 	case 0x62:
1443 		return handle_ri(vcpu);
1444 	case 0x71:
1445 		return handle_lpswey(vcpu);
1446 	default:
1447 		return -EOPNOTSUPP;
1448 	}
1449 }
1450 
1451 static int handle_tprot(struct kvm_vcpu *vcpu)
1452 {
1453 	u64 address, operand2;
1454 	unsigned long gpa;
1455 	u8 access_key;
1456 	bool writable;
1457 	int ret, cc;
1458 	u8 ar;
1459 
1460 	vcpu->stat.instruction_tprot++;
1461 
1462 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1463 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1464 
1465 	kvm_s390_get_base_disp_sse(vcpu, &address, &operand2, &ar, NULL);
1466 	access_key = (operand2 & 0xf0) >> 4;
1467 
1468 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1469 		ipte_lock(vcpu->kvm);
1470 
1471 	ret = guest_translate_address_with_key(vcpu, address, ar, &gpa,
1472 					       GACC_STORE, access_key);
1473 	if (ret == 0) {
1474 		gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1475 	} else if (ret == PGM_PROTECTION) {
1476 		writable = false;
1477 		/* Write protected? Try again with read-only... */
1478 		ret = guest_translate_address_with_key(vcpu, address, ar, &gpa,
1479 						       GACC_FETCH, access_key);
1480 	}
1481 	if (ret >= 0) {
1482 		cc = -1;
1483 
1484 		/* Fetching permitted; storing permitted */
1485 		if (ret == 0 && writable)
1486 			cc = 0;
1487 		/* Fetching permitted; storing not permitted */
1488 		else if (ret == 0 && !writable)
1489 			cc = 1;
1490 		/* Fetching not permitted; storing not permitted */
1491 		else if (ret == PGM_PROTECTION)
1492 			cc = 2;
1493 		/* Translation not available */
1494 		else if (ret != PGM_ADDRESSING && ret != PGM_TRANSLATION_SPEC)
1495 			cc = 3;
1496 
1497 		if (cc != -1) {
1498 			kvm_s390_set_psw_cc(vcpu, cc);
1499 			ret = 0;
1500 		} else {
1501 			ret = kvm_s390_inject_program_int(vcpu, ret);
1502 		}
1503 	}
1504 
1505 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1506 		ipte_unlock(vcpu->kvm);
1507 	return ret;
1508 }
1509 
1510 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1511 {
1512 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1513 	case 0x01:
1514 		return handle_tprot(vcpu);
1515 	default:
1516 		return -EOPNOTSUPP;
1517 	}
1518 }
1519 
1520 static int handle_sckpf(struct kvm_vcpu *vcpu)
1521 {
1522 	u32 value;
1523 
1524 	vcpu->stat.instruction_sckpf++;
1525 
1526 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1527 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1528 
1529 	if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1530 		return kvm_s390_inject_program_int(vcpu,
1531 						   PGM_SPECIFICATION);
1532 
1533 	value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1534 	vcpu->arch.sie_block->todpr = value;
1535 
1536 	return 0;
1537 }
1538 
1539 static int handle_ptff(struct kvm_vcpu *vcpu)
1540 {
1541 	vcpu->stat.instruction_ptff++;
1542 
1543 	/* we don't emulate any control instructions yet */
1544 	kvm_s390_set_psw_cc(vcpu, 3);
1545 	return 0;
1546 }
1547 
1548 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1549 {
1550 	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1551 	case 0x04:
1552 		return handle_ptff(vcpu);
1553 	case 0x07:
1554 		return handle_sckpf(vcpu);
1555 	default:
1556 		return -EOPNOTSUPP;
1557 	}
1558 }
1559