xref: /kernel/linux/linux-5.10/arch/s390/kvm/priv.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * handling privileged instructions
4 *
5 * Copyright IBM Corp. 2008, 2020
6 *
7 *    Author(s): Carsten Otte <cotte@de.ibm.com>
8 *               Christian Borntraeger <borntraeger@de.ibm.com>
9 */
10
11#include <linux/kvm.h>
12#include <linux/gfp.h>
13#include <linux/errno.h>
14#include <linux/compat.h>
15#include <linux/mm_types.h>
16#include <linux/pgtable.h>
17
18#include <asm/asm-offsets.h>
19#include <asm/facility.h>
20#include <asm/current.h>
21#include <asm/debug.h>
22#include <asm/ebcdic.h>
23#include <asm/sysinfo.h>
24#include <asm/page-states.h>
25#include <asm/gmap.h>
26#include <asm/io.h>
27#include <asm/ptrace.h>
28#include <asm/sclp.h>
29#include <asm/ap.h>
30#include "gaccess.h"
31#include "kvm-s390.h"
32#include "trace.h"
33
34static int handle_ri(struct kvm_vcpu *vcpu)
35{
36	vcpu->stat.instruction_ri++;
37
38	if (test_kvm_facility(vcpu->kvm, 64)) {
39		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
40		vcpu->arch.sie_block->ecb3 |= ECB3_RI;
41		kvm_s390_retry_instr(vcpu);
42		return 0;
43	} else
44		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
45}
46
47int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
48{
49	if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
50		return handle_ri(vcpu);
51	else
52		return -EOPNOTSUPP;
53}
54
55static int handle_gs(struct kvm_vcpu *vcpu)
56{
57	vcpu->stat.instruction_gs++;
58
59	if (test_kvm_facility(vcpu->kvm, 133)) {
60		VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
61		preempt_disable();
62		__ctl_set_bit(2, 4);
63		current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
64		restore_gs_cb(current->thread.gs_cb);
65		preempt_enable();
66		vcpu->arch.sie_block->ecb |= ECB_GS;
67		vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
68		vcpu->arch.gs_enabled = 1;
69		kvm_s390_retry_instr(vcpu);
70		return 0;
71	} else
72		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
73}
74
75int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
76{
77	int code = vcpu->arch.sie_block->ipb & 0xff;
78
79	if (code == 0x49 || code == 0x4d)
80		return handle_gs(vcpu);
81	else
82		return -EOPNOTSUPP;
83}
84/* Handle SCK (SET CLOCK) interception */
85static int handle_set_clock(struct kvm_vcpu *vcpu)
86{
87	struct kvm_s390_vm_tod_clock gtod = { 0 };
88	int rc;
89	u8 ar;
90	u64 op2;
91
92	vcpu->stat.instruction_sck++;
93
94	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
95		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
96
97	op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
98	if (op2 & 7)	/* Operand must be on a doubleword boundary */
99		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
100	rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
101	if (rc)
102		return kvm_s390_inject_prog_cond(vcpu, rc);
103
104	VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
105	/*
106	 * To set the TOD clock the kvm lock must be taken, but the vcpu lock
107	 * is already held in handle_set_clock. The usual lock order is the
108	 * opposite.  As SCK is deprecated and should not be used in several
109	 * cases, for example when the multiple epoch facility or TOD clock
110	 * steering facility is installed (see Principles of Operation),  a
111	 * slow path can be used.  If the lock can not be taken via try_lock,
112	 * the instruction will be retried via -EAGAIN at a later point in
113	 * time.
114	 */
115	if (!kvm_s390_try_set_tod_clock(vcpu->kvm, &gtod)) {
116		kvm_s390_retry_instr(vcpu);
117		return -EAGAIN;
118	}
119
120	kvm_s390_set_psw_cc(vcpu, 0);
121	return 0;
122}
123
124static int handle_set_prefix(struct kvm_vcpu *vcpu)
125{
126	u64 operand2;
127	u32 address;
128	int rc;
129	u8 ar;
130
131	vcpu->stat.instruction_spx++;
132
133	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
134		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
135
136	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
137
138	/* must be word boundary */
139	if (operand2 & 3)
140		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
141
142	/* get the value */
143	rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
144	if (rc)
145		return kvm_s390_inject_prog_cond(vcpu, rc);
146
147	address &= 0x7fffe000u;
148
149	/*
150	 * Make sure the new value is valid memory. We only need to check the
151	 * first page, since address is 8k aligned and memory pieces are always
152	 * at least 1MB aligned and have at least a size of 1MB.
153	 */
154	if (kvm_is_error_gpa(vcpu->kvm, address))
155		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
156
157	kvm_s390_set_prefix(vcpu, address);
158	trace_kvm_s390_handle_prefix(vcpu, 1, address);
159	return 0;
160}
161
162static int handle_store_prefix(struct kvm_vcpu *vcpu)
163{
164	u64 operand2;
165	u32 address;
166	int rc;
167	u8 ar;
168
169	vcpu->stat.instruction_stpx++;
170
171	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
172		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
173
174	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
175
176	/* must be word boundary */
177	if (operand2 & 3)
178		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
179
180	address = kvm_s390_get_prefix(vcpu);
181
182	/* get the value */
183	rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
184	if (rc)
185		return kvm_s390_inject_prog_cond(vcpu, rc);
186
187	VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
188	trace_kvm_s390_handle_prefix(vcpu, 0, address);
189	return 0;
190}
191
192static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
193{
194	u16 vcpu_id = vcpu->vcpu_id;
195	u64 ga;
196	int rc;
197	u8 ar;
198
199	vcpu->stat.instruction_stap++;
200
201	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
202		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
203
204	ga = kvm_s390_get_base_disp_s(vcpu, &ar);
205
206	if (ga & 1)
207		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
208
209	rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
210	if (rc)
211		return kvm_s390_inject_prog_cond(vcpu, rc);
212
213	VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
214	trace_kvm_s390_handle_stap(vcpu, ga);
215	return 0;
216}
217
218int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
219{
220	int rc;
221
222	trace_kvm_s390_skey_related_inst(vcpu);
223	/* Already enabled? */
224	if (vcpu->arch.skey_enabled)
225		return 0;
226
227	rc = s390_enable_skey();
228	VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
229	if (rc)
230		return rc;
231
232	if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
233		kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
234	if (!vcpu->kvm->arch.use_skf)
235		vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
236	else
237		vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
238	vcpu->arch.skey_enabled = true;
239	return 0;
240}
241
242static int try_handle_skey(struct kvm_vcpu *vcpu)
243{
244	int rc;
245
246	rc = kvm_s390_skey_check_enable(vcpu);
247	if (rc)
248		return rc;
249	if (vcpu->kvm->arch.use_skf) {
250		/* with storage-key facility, SIE interprets it for us */
251		kvm_s390_retry_instr(vcpu);
252		VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
253		return -EAGAIN;
254	}
255	return 0;
256}
257
258static int handle_iske(struct kvm_vcpu *vcpu)
259{
260	unsigned long gaddr, vmaddr;
261	unsigned char key;
262	int reg1, reg2;
263	bool unlocked;
264	int rc;
265
266	vcpu->stat.instruction_iske++;
267
268	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
269		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
270
271	rc = try_handle_skey(vcpu);
272	if (rc)
273		return rc != -EAGAIN ? rc : 0;
274
275	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
276
277	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
278	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
279	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
280	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
281	if (kvm_is_error_hva(vmaddr))
282		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
283retry:
284	unlocked = false;
285	mmap_read_lock(current->mm);
286	rc = get_guest_storage_key(current->mm, vmaddr, &key);
287
288	if (rc) {
289		rc = fixup_user_fault(current->mm, vmaddr,
290				      FAULT_FLAG_WRITE, &unlocked);
291		if (!rc) {
292			mmap_read_unlock(current->mm);
293			goto retry;
294		}
295	}
296	mmap_read_unlock(current->mm);
297	if (rc == -EFAULT)
298		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
299	if (rc < 0)
300		return rc;
301	vcpu->run->s.regs.gprs[reg1] &= ~0xff;
302	vcpu->run->s.regs.gprs[reg1] |= key;
303	return 0;
304}
305
306static int handle_rrbe(struct kvm_vcpu *vcpu)
307{
308	unsigned long vmaddr, gaddr;
309	int reg1, reg2;
310	bool unlocked;
311	int rc;
312
313	vcpu->stat.instruction_rrbe++;
314
315	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
316		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
317
318	rc = try_handle_skey(vcpu);
319	if (rc)
320		return rc != -EAGAIN ? rc : 0;
321
322	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
323
324	gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
325	gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
326	gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
327	vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
328	if (kvm_is_error_hva(vmaddr))
329		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
330retry:
331	unlocked = false;
332	mmap_read_lock(current->mm);
333	rc = reset_guest_reference_bit(current->mm, vmaddr);
334	if (rc < 0) {
335		rc = fixup_user_fault(current->mm, vmaddr,
336				      FAULT_FLAG_WRITE, &unlocked);
337		if (!rc) {
338			mmap_read_unlock(current->mm);
339			goto retry;
340		}
341	}
342	mmap_read_unlock(current->mm);
343	if (rc == -EFAULT)
344		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
345	if (rc < 0)
346		return rc;
347	kvm_s390_set_psw_cc(vcpu, rc);
348	return 0;
349}
350
351#define SSKE_NQ 0x8
352#define SSKE_MR 0x4
353#define SSKE_MC 0x2
354#define SSKE_MB 0x1
355static int handle_sske(struct kvm_vcpu *vcpu)
356{
357	unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
358	unsigned long start, end;
359	unsigned char key, oldkey;
360	int reg1, reg2;
361	bool unlocked;
362	int rc;
363
364	vcpu->stat.instruction_sske++;
365
366	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
367		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
368
369	rc = try_handle_skey(vcpu);
370	if (rc)
371		return rc != -EAGAIN ? rc : 0;
372
373	if (!test_kvm_facility(vcpu->kvm, 8))
374		m3 &= ~SSKE_MB;
375	if (!test_kvm_facility(vcpu->kvm, 10))
376		m3 &= ~(SSKE_MC | SSKE_MR);
377	if (!test_kvm_facility(vcpu->kvm, 14))
378		m3 &= ~SSKE_NQ;
379
380	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
381
382	key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
383	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
384	start = kvm_s390_logical_to_effective(vcpu, start);
385	if (m3 & SSKE_MB) {
386		/* start already designates an absolute address */
387		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
388	} else {
389		start = kvm_s390_real_to_abs(vcpu, start);
390		end = start + PAGE_SIZE;
391	}
392
393	while (start != end) {
394		unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
395		unlocked = false;
396
397		if (kvm_is_error_hva(vmaddr))
398			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
399
400		mmap_read_lock(current->mm);
401		rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
402						m3 & SSKE_NQ, m3 & SSKE_MR,
403						m3 & SSKE_MC);
404
405		if (rc < 0) {
406			rc = fixup_user_fault(current->mm, vmaddr,
407					      FAULT_FLAG_WRITE, &unlocked);
408			rc = !rc ? -EAGAIN : rc;
409		}
410		mmap_read_unlock(current->mm);
411		if (rc == -EFAULT)
412			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
413		if (rc == -EAGAIN)
414			continue;
415		if (rc < 0)
416			return rc;
417		start += PAGE_SIZE;
418	}
419
420	if (m3 & (SSKE_MC | SSKE_MR)) {
421		if (m3 & SSKE_MB) {
422			/* skey in reg1 is unpredictable */
423			kvm_s390_set_psw_cc(vcpu, 3);
424		} else {
425			kvm_s390_set_psw_cc(vcpu, rc);
426			vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
427			vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
428		}
429	}
430	if (m3 & SSKE_MB) {
431		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
432			vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
433		else
434			vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
435		end = kvm_s390_logical_to_effective(vcpu, end);
436		vcpu->run->s.regs.gprs[reg2] |= end;
437	}
438	return 0;
439}
440
441static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
442{
443	vcpu->stat.instruction_ipte_interlock++;
444	if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
445		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
446	wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
447	kvm_s390_retry_instr(vcpu);
448	VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
449	return 0;
450}
451
452static int handle_test_block(struct kvm_vcpu *vcpu)
453{
454	gpa_t addr;
455	int reg2;
456
457	vcpu->stat.instruction_tb++;
458
459	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
460		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
461
462	kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
463	addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
464	addr = kvm_s390_logical_to_effective(vcpu, addr);
465	if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
466		return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
467	addr = kvm_s390_real_to_abs(vcpu, addr);
468
469	if (kvm_is_error_gpa(vcpu->kvm, addr))
470		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
471	/*
472	 * We don't expect errors on modern systems, and do not care
473	 * about storage keys (yet), so let's just clear the page.
474	 */
475	if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
476		return -EFAULT;
477	kvm_s390_set_psw_cc(vcpu, 0);
478	vcpu->run->s.regs.gprs[0] = 0;
479	return 0;
480}
481
482static int handle_tpi(struct kvm_vcpu *vcpu)
483{
484	struct kvm_s390_interrupt_info *inti;
485	unsigned long len;
486	u32 tpi_data[3];
487	int rc;
488	u64 addr;
489	u8 ar;
490
491	vcpu->stat.instruction_tpi++;
492
493	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
494	if (addr & 3)
495		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
496
497	inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
498	if (!inti) {
499		kvm_s390_set_psw_cc(vcpu, 0);
500		return 0;
501	}
502
503	tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
504	tpi_data[1] = inti->io.io_int_parm;
505	tpi_data[2] = inti->io.io_int_word;
506	if (addr) {
507		/*
508		 * Store the two-word I/O interruption code into the
509		 * provided area.
510		 */
511		len = sizeof(tpi_data) - 4;
512		rc = write_guest(vcpu, addr, ar, &tpi_data, len);
513		if (rc) {
514			rc = kvm_s390_inject_prog_cond(vcpu, rc);
515			goto reinject_interrupt;
516		}
517	} else {
518		/*
519		 * Store the three-word I/O interruption code into
520		 * the appropriate lowcore area.
521		 */
522		len = sizeof(tpi_data);
523		if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
524			/* failed writes to the low core are not recoverable */
525			rc = -EFAULT;
526			goto reinject_interrupt;
527		}
528	}
529
530	/* irq was successfully handed to the guest */
531	kfree(inti);
532	kvm_s390_set_psw_cc(vcpu, 1);
533	return 0;
534reinject_interrupt:
535	/*
536	 * If we encounter a problem storing the interruption code, the
537	 * instruction is suppressed from the guest's view: reinject the
538	 * interrupt.
539	 */
540	if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
541		kfree(inti);
542		rc = -EFAULT;
543	}
544	/* don't set the cc, a pgm irq was injected or we drop to user space */
545	return rc ? -EFAULT : 0;
546}
547
548static int handle_tsch(struct kvm_vcpu *vcpu)
549{
550	struct kvm_s390_interrupt_info *inti = NULL;
551	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
552
553	vcpu->stat.instruction_tsch++;
554
555	/* a valid schid has at least one bit set */
556	if (vcpu->run->s.regs.gprs[1])
557		inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
558					   vcpu->run->s.regs.gprs[1]);
559
560	/*
561	 * Prepare exit to userspace.
562	 * We indicate whether we dequeued a pending I/O interrupt
563	 * so that userspace can re-inject it if the instruction gets
564	 * a program check. While this may re-order the pending I/O
565	 * interrupts, this is no problem since the priority is kept
566	 * intact.
567	 */
568	vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
569	vcpu->run->s390_tsch.dequeued = !!inti;
570	if (inti) {
571		vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
572		vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
573		vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
574		vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
575	}
576	vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
577	kfree(inti);
578	return -EREMOTE;
579}
580
581static int handle_io_inst(struct kvm_vcpu *vcpu)
582{
583	VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
584
585	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
586		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
587
588	if (vcpu->kvm->arch.css_support) {
589		/*
590		 * Most I/O instructions will be handled by userspace.
591		 * Exceptions are tpi and the interrupt portion of tsch.
592		 */
593		if (vcpu->arch.sie_block->ipa == 0xb236)
594			return handle_tpi(vcpu);
595		if (vcpu->arch.sie_block->ipa == 0xb235)
596			return handle_tsch(vcpu);
597		/* Handle in userspace. */
598		vcpu->stat.instruction_io_other++;
599		return -EOPNOTSUPP;
600	} else {
601		/*
602		 * Set condition code 3 to stop the guest from issuing channel
603		 * I/O instructions.
604		 */
605		kvm_s390_set_psw_cc(vcpu, 3);
606		return 0;
607	}
608}
609
610/*
611 * handle_pqap: Handling pqap interception
612 * @vcpu: the vcpu having issue the pqap instruction
613 *
614 * We now support PQAP/AQIC instructions and we need to correctly
615 * answer the guest even if no dedicated driver's hook is available.
616 *
617 * The intercepting code calls a dedicated callback for this instruction
618 * if a driver did register one in the CRYPTO satellite of the
619 * SIE block.
620 *
621 * If no callback is available, the queues are not available, return this
622 * response code to the caller and set CC to 3.
623 * Else return the response code returned by the callback.
624 */
625static int handle_pqap(struct kvm_vcpu *vcpu)
626{
627	struct ap_queue_status status = {};
628	unsigned long reg0;
629	int ret;
630	uint8_t fc;
631
632	/* Verify that the AP instruction are available */
633	if (!ap_instructions_available())
634		return -EOPNOTSUPP;
635	/* Verify that the guest is allowed to use AP instructions */
636	if (!(vcpu->arch.sie_block->eca & ECA_APIE))
637		return -EOPNOTSUPP;
638	/*
639	 * The only possibly intercepted functions when AP instructions are
640	 * available for the guest are AQIC and TAPQ with the t bit set
641	 * since we do not set IC.3 (FIII) we currently will only intercept
642	 * the AQIC function code.
643	 * Note: running nested under z/VM can result in intercepts for other
644	 * function codes, e.g. PQAP(QCI). We do not support this and bail out.
645	 */
646	reg0 = vcpu->run->s.regs.gprs[0];
647	fc = (reg0 >> 24) & 0xff;
648	if (fc != 0x03)
649		return -EOPNOTSUPP;
650
651	/* PQAP instruction is allowed for guest kernel only */
652	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
653		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
654
655	/* Common PQAP instruction specification exceptions */
656	/* bits 41-47 must all be zeros */
657	if (reg0 & 0x007f0000UL)
658		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
659	/* APFT not install and T bit set */
660	if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
661		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
662	/* APXA not installed and APID greater 64 or APQI greater 16 */
663	if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
664		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
665
666	/* AQIC function code specific exception */
667	/* facility 65 not present for AQIC function code */
668	if (!test_kvm_facility(vcpu->kvm, 65))
669		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
670
671	/*
672	 * Verify that the hook callback is registered, lock the owner
673	 * and call the hook.
674	 */
675	if (vcpu->kvm->arch.crypto.pqap_hook) {
676		if (!try_module_get(vcpu->kvm->arch.crypto.pqap_hook->owner))
677			return -EOPNOTSUPP;
678		ret = vcpu->kvm->arch.crypto.pqap_hook->hook(vcpu);
679		module_put(vcpu->kvm->arch.crypto.pqap_hook->owner);
680		if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
681			kvm_s390_set_psw_cc(vcpu, 3);
682		return ret;
683	}
684	/*
685	 * A vfio_driver must register a hook.
686	 * No hook means no driver to enable the SIE CRYCB and no queues.
687	 * We send this response to the guest.
688	 */
689	status.response_code = 0x01;
690	memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
691	kvm_s390_set_psw_cc(vcpu, 3);
692	return 0;
693}
694
695static int handle_stfl(struct kvm_vcpu *vcpu)
696{
697	int rc;
698	unsigned int fac;
699
700	vcpu->stat.instruction_stfl++;
701
702	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
703		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
704
705	/*
706	 * We need to shift the lower 32 facility bits (bit 0-31) from a u64
707	 * into a u32 memory representation. They will remain bits 0-31.
708	 */
709	fac = *vcpu->kvm->arch.model.fac_list >> 32;
710	rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
711			    &fac, sizeof(fac));
712	if (rc)
713		return rc;
714	VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
715	trace_kvm_s390_handle_stfl(vcpu, fac);
716	return 0;
717}
718
719#define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
720#define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
721#define PSW_ADDR_24 0x0000000000ffffffUL
722#define PSW_ADDR_31 0x000000007fffffffUL
723
724int is_valid_psw(psw_t *psw)
725{
726	if (psw->mask & PSW_MASK_UNASSIGNED)
727		return 0;
728	if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
729		if (psw->addr & ~PSW_ADDR_31)
730			return 0;
731	}
732	if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
733		return 0;
734	if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
735		return 0;
736	if (psw->addr & 1)
737		return 0;
738	return 1;
739}
740
741int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
742{
743	psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
744	psw_compat_t new_psw;
745	u64 addr;
746	int rc;
747	u8 ar;
748
749	vcpu->stat.instruction_lpsw++;
750
751	if (gpsw->mask & PSW_MASK_PSTATE)
752		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
753
754	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
755	if (addr & 7)
756		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
757
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	if (!(new_psw.mask & PSW32_MASK_BASE))
762		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
763	gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
764	gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
765	gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
766	if (!is_valid_psw(gpsw))
767		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
768	return 0;
769}
770
771static int handle_lpswe(struct kvm_vcpu *vcpu)
772{
773	psw_t new_psw;
774	u64 addr;
775	int rc;
776	u8 ar;
777
778	vcpu->stat.instruction_lpswe++;
779
780	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
781		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
782
783	addr = kvm_s390_get_base_disp_s(vcpu, &ar);
784	if (addr & 7)
785		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
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	vcpu->arch.sie_block->gpsw = new_psw;
790	if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
791		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
792	return 0;
793}
794
795static int handle_stidp(struct kvm_vcpu *vcpu)
796{
797	u64 stidp_data = vcpu->kvm->arch.model.cpuid;
798	u64 operand2;
799	int rc;
800	u8 ar;
801
802	vcpu->stat.instruction_stidp++;
803
804	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
805		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
806
807	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
808
809	if (operand2 & 7)
810		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
811
812	rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
813	if (rc)
814		return kvm_s390_inject_prog_cond(vcpu, rc);
815
816	VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
817	return 0;
818}
819
820static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
821{
822	int cpus = 0;
823	int n;
824
825	cpus = atomic_read(&vcpu->kvm->online_vcpus);
826
827	/* deal with other level 3 hypervisors */
828	if (stsi(mem, 3, 2, 2))
829		mem->count = 0;
830	if (mem->count < 8)
831		mem->count++;
832	for (n = mem->count - 1; n > 0 ; n--)
833		memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
834
835	memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
836	mem->vm[0].cpus_total = cpus;
837	mem->vm[0].cpus_configured = cpus;
838	mem->vm[0].cpus_standby = 0;
839	mem->vm[0].cpus_reserved = 0;
840	mem->vm[0].caf = 1000;
841	memcpy(mem->vm[0].name, "KVMguest", 8);
842	ASCEBC(mem->vm[0].name, 8);
843	memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
844	ASCEBC(mem->vm[0].cpi, 16);
845}
846
847static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
848				 u8 fc, u8 sel1, u16 sel2)
849{
850	vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
851	vcpu->run->s390_stsi.addr = addr;
852	vcpu->run->s390_stsi.ar = ar;
853	vcpu->run->s390_stsi.fc = fc;
854	vcpu->run->s390_stsi.sel1 = sel1;
855	vcpu->run->s390_stsi.sel2 = sel2;
856}
857
858static int handle_stsi(struct kvm_vcpu *vcpu)
859{
860	int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
861	int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
862	int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
863	unsigned long mem = 0;
864	u64 operand2;
865	int rc = 0;
866	u8 ar;
867
868	vcpu->stat.instruction_stsi++;
869	VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
870
871	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
872		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
873
874	if (fc > 3) {
875		kvm_s390_set_psw_cc(vcpu, 3);
876		return 0;
877	}
878
879	if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
880	    || vcpu->run->s.regs.gprs[1] & 0xffff0000)
881		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
882
883	if (fc == 0) {
884		vcpu->run->s.regs.gprs[0] = 3 << 28;
885		kvm_s390_set_psw_cc(vcpu, 0);
886		return 0;
887	}
888
889	operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
890
891	if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff))
892		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
893
894	switch (fc) {
895	case 1: /* same handling for 1 and 2 */
896	case 2:
897		mem = get_zeroed_page(GFP_KERNEL);
898		if (!mem)
899			goto out_no_data;
900		if (stsi((void *) mem, fc, sel1, sel2))
901			goto out_no_data;
902		break;
903	case 3:
904		if (sel1 != 2 || sel2 != 2)
905			goto out_no_data;
906		mem = get_zeroed_page(GFP_KERNEL);
907		if (!mem)
908			goto out_no_data;
909		handle_stsi_3_2_2(vcpu, (void *) mem);
910		break;
911	}
912	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
913		memcpy((void *)sida_origin(vcpu->arch.sie_block), (void *)mem,
914		       PAGE_SIZE);
915		rc = 0;
916	} else {
917		rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
918	}
919	if (rc) {
920		rc = kvm_s390_inject_prog_cond(vcpu, rc);
921		goto out;
922	}
923	if (vcpu->kvm->arch.user_stsi) {
924		insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
925		rc = -EREMOTE;
926	}
927	trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
928	free_page(mem);
929	kvm_s390_set_psw_cc(vcpu, 0);
930	vcpu->run->s.regs.gprs[0] = 0;
931	return rc;
932out_no_data:
933	kvm_s390_set_psw_cc(vcpu, 3);
934out:
935	free_page(mem);
936	return rc;
937}
938
939int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
940{
941	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
942	case 0x02:
943		return handle_stidp(vcpu);
944	case 0x04:
945		return handle_set_clock(vcpu);
946	case 0x10:
947		return handle_set_prefix(vcpu);
948	case 0x11:
949		return handle_store_prefix(vcpu);
950	case 0x12:
951		return handle_store_cpu_address(vcpu);
952	case 0x14:
953		return kvm_s390_handle_vsie(vcpu);
954	case 0x21:
955	case 0x50:
956		return handle_ipte_interlock(vcpu);
957	case 0x29:
958		return handle_iske(vcpu);
959	case 0x2a:
960		return handle_rrbe(vcpu);
961	case 0x2b:
962		return handle_sske(vcpu);
963	case 0x2c:
964		return handle_test_block(vcpu);
965	case 0x30:
966	case 0x31:
967	case 0x32:
968	case 0x33:
969	case 0x34:
970	case 0x35:
971	case 0x36:
972	case 0x37:
973	case 0x38:
974	case 0x39:
975	case 0x3a:
976	case 0x3b:
977	case 0x3c:
978	case 0x5f:
979	case 0x74:
980	case 0x76:
981		return handle_io_inst(vcpu);
982	case 0x56:
983		return handle_sthyi(vcpu);
984	case 0x7d:
985		return handle_stsi(vcpu);
986	case 0xaf:
987		return handle_pqap(vcpu);
988	case 0xb1:
989		return handle_stfl(vcpu);
990	case 0xb2:
991		return handle_lpswe(vcpu);
992	default:
993		return -EOPNOTSUPP;
994	}
995}
996
997static int handle_epsw(struct kvm_vcpu *vcpu)
998{
999	int reg1, reg2;
1000
1001	vcpu->stat.instruction_epsw++;
1002
1003	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1004
1005	/* This basically extracts the mask half of the psw. */
1006	vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
1007	vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
1008	if (reg2) {
1009		vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
1010		vcpu->run->s.regs.gprs[reg2] |=
1011			vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
1012	}
1013	return 0;
1014}
1015
1016#define PFMF_RESERVED   0xfffc0101UL
1017#define PFMF_SK         0x00020000UL
1018#define PFMF_CF         0x00010000UL
1019#define PFMF_UI         0x00008000UL
1020#define PFMF_FSC        0x00007000UL
1021#define PFMF_NQ         0x00000800UL
1022#define PFMF_MR         0x00000400UL
1023#define PFMF_MC         0x00000200UL
1024#define PFMF_KEY        0x000000feUL
1025
1026static int handle_pfmf(struct kvm_vcpu *vcpu)
1027{
1028	bool mr = false, mc = false, nq;
1029	int reg1, reg2;
1030	unsigned long start, end;
1031	unsigned char key;
1032
1033	vcpu->stat.instruction_pfmf++;
1034
1035	kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1036
1037	if (!test_kvm_facility(vcpu->kvm, 8))
1038		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1039
1040	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1041		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1042
1043	if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1044		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1045
1046	/* Only provide non-quiescing support if enabled for the guest */
1047	if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1048	    !test_kvm_facility(vcpu->kvm, 14))
1049		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1050
1051	/* Only provide conditional-SSKE support if enabled for the guest */
1052	if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1053	    test_kvm_facility(vcpu->kvm, 10)) {
1054		mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1055		mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1056	}
1057
1058	nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1059	key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1060	start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1061	start = kvm_s390_logical_to_effective(vcpu, start);
1062
1063	if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1064		if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1065			return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1066	}
1067
1068	switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1069	case 0x00000000:
1070		/* only 4k frames specify a real address */
1071		start = kvm_s390_real_to_abs(vcpu, start);
1072		end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1073		break;
1074	case 0x00001000:
1075		end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1076		break;
1077	case 0x00002000:
1078		/* only support 2G frame size if EDAT2 is available and we are
1079		   not in 24-bit addressing mode */
1080		if (!test_kvm_facility(vcpu->kvm, 78) ||
1081		    psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1082			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1083		end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1084		break;
1085	default:
1086		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1087	}
1088
1089	while (start != end) {
1090		unsigned long vmaddr;
1091		bool unlocked = false;
1092
1093		/* Translate guest address to host address */
1094		vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
1095		if (kvm_is_error_hva(vmaddr))
1096			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1097
1098		if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1099			if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1100				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1101		}
1102
1103		if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1104			int rc = kvm_s390_skey_check_enable(vcpu);
1105
1106			if (rc)
1107				return rc;
1108			mmap_read_lock(current->mm);
1109			rc = cond_set_guest_storage_key(current->mm, vmaddr,
1110							key, NULL, nq, mr, mc);
1111			if (rc < 0) {
1112				rc = fixup_user_fault(current->mm, vmaddr,
1113						      FAULT_FLAG_WRITE, &unlocked);
1114				rc = !rc ? -EAGAIN : rc;
1115			}
1116			mmap_read_unlock(current->mm);
1117			if (rc == -EFAULT)
1118				return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1119			if (rc == -EAGAIN)
1120				continue;
1121			if (rc < 0)
1122				return rc;
1123		}
1124		start += PAGE_SIZE;
1125	}
1126	if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1127		if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1128			vcpu->run->s.regs.gprs[reg2] = end;
1129		} else {
1130			vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1131			end = kvm_s390_logical_to_effective(vcpu, end);
1132			vcpu->run->s.regs.gprs[reg2] |= end;
1133		}
1134	}
1135	return 0;
1136}
1137
1138/*
1139 * Must be called with relevant read locks held (kvm->mm->mmap_lock, kvm->srcu)
1140 */
1141static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1142{
1143	int r1, r2, nappended, entries;
1144	unsigned long gfn, hva, res, pgstev, ptev;
1145	unsigned long *cbrlo;
1146
1147	/*
1148	 * We don't need to set SD.FPF.SK to 1 here, because if we have a
1149	 * machine check here we either handle it or crash
1150	 */
1151
1152	kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1153	gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1154	hva = gfn_to_hva(vcpu->kvm, gfn);
1155	entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1156
1157	if (kvm_is_error_hva(hva))
1158		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1159
1160	nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1161	if (nappended < 0) {
1162		res = orc ? 0x10 : 0;
1163		vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1164		return 0;
1165	}
1166	res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1167	/*
1168	 * Set the block-content state part of the result. 0 means resident, so
1169	 * nothing to do if the page is valid. 2 is for preserved pages
1170	 * (non-present and non-zero), and 3 for zero pages (non-present and
1171	 * zero).
1172	 */
1173	if (ptev & _PAGE_INVALID) {
1174		res |= 2;
1175		if (pgstev & _PGSTE_GPS_ZERO)
1176			res |= 1;
1177	}
1178	if (pgstev & _PGSTE_GPS_NODAT)
1179		res |= 0x20;
1180	vcpu->run->s.regs.gprs[r1] = res;
1181	/*
1182	 * It is possible that all the normal 511 slots were full, in which case
1183	 * we will now write in the 512th slot, which is reserved for host use.
1184	 * In both cases we let the normal essa handling code process all the
1185	 * slots, including the reserved one, if needed.
1186	 */
1187	if (nappended > 0) {
1188		cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1189		cbrlo[entries] = gfn << PAGE_SHIFT;
1190	}
1191
1192	if (orc) {
1193		struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1194
1195		/* Increment only if we are really flipping the bit */
1196		if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1197			atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1198	}
1199
1200	return nappended;
1201}
1202
1203static int handle_essa(struct kvm_vcpu *vcpu)
1204{
1205	/* entries expected to be 1FF */
1206	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1207	unsigned long *cbrlo;
1208	struct gmap *gmap;
1209	int i, orc;
1210
1211	VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1212	gmap = vcpu->arch.gmap;
1213	vcpu->stat.instruction_essa++;
1214	if (!vcpu->kvm->arch.use_cmma)
1215		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1216
1217	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1218		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1219	/* Check for invalid operation request code */
1220	orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1221	/* ORCs 0-6 are always valid */
1222	if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1223						: ESSA_SET_STABLE_IF_RESIDENT))
1224		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1225
1226	if (!vcpu->kvm->arch.migration_mode) {
1227		/*
1228		 * CMMA is enabled in the KVM settings, but is disabled in
1229		 * the SIE block and in the mm_context, and we are not doing
1230		 * a migration. Enable CMMA in the mm_context.
1231		 * Since we need to take a write lock to write to the context
1232		 * to avoid races with storage keys handling, we check if the
1233		 * value really needs to be written to; if the value is
1234		 * already correct, we do nothing and avoid the lock.
1235		 */
1236		if (vcpu->kvm->mm->context.uses_cmm == 0) {
1237			mmap_write_lock(vcpu->kvm->mm);
1238			vcpu->kvm->mm->context.uses_cmm = 1;
1239			mmap_write_unlock(vcpu->kvm->mm);
1240		}
1241		/*
1242		 * If we are here, we are supposed to have CMMA enabled in
1243		 * the SIE block. Enabling CMMA works on a per-CPU basis,
1244		 * while the context use_cmma flag is per process.
1245		 * It's possible that the context flag is enabled and the
1246		 * SIE flag is not, so we set the flag always; if it was
1247		 * already set, nothing changes, otherwise we enable it
1248		 * on this CPU too.
1249		 */
1250		vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1251		/* Retry the ESSA instruction */
1252		kvm_s390_retry_instr(vcpu);
1253	} else {
1254		int srcu_idx;
1255
1256		mmap_read_lock(vcpu->kvm->mm);
1257		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1258		i = __do_essa(vcpu, orc);
1259		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1260		mmap_read_unlock(vcpu->kvm->mm);
1261		if (i < 0)
1262			return i;
1263		/* Account for the possible extra cbrl entry */
1264		entries += i;
1265	}
1266	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;	/* reset nceo */
1267	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1268	mmap_read_lock(gmap->mm);
1269	for (i = 0; i < entries; ++i)
1270		__gmap_zap(gmap, cbrlo[i]);
1271	mmap_read_unlock(gmap->mm);
1272	return 0;
1273}
1274
1275int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1276{
1277	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1278	case 0x8a:
1279	case 0x8e:
1280	case 0x8f:
1281		return handle_ipte_interlock(vcpu);
1282	case 0x8d:
1283		return handle_epsw(vcpu);
1284	case 0xab:
1285		return handle_essa(vcpu);
1286	case 0xaf:
1287		return handle_pfmf(vcpu);
1288	default:
1289		return -EOPNOTSUPP;
1290	}
1291}
1292
1293int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1294{
1295	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1296	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1297	int reg, rc, nr_regs;
1298	u32 ctl_array[16];
1299	u64 ga;
1300	u8 ar;
1301
1302	vcpu->stat.instruction_lctl++;
1303
1304	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1305		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1306
1307	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1308
1309	if (ga & 3)
1310		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1311
1312	VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1313	trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1314
1315	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1316	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1317	if (rc)
1318		return kvm_s390_inject_prog_cond(vcpu, rc);
1319	reg = reg1;
1320	nr_regs = 0;
1321	do {
1322		vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1323		vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1324		if (reg == reg3)
1325			break;
1326		reg = (reg + 1) % 16;
1327	} while (1);
1328	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1329	return 0;
1330}
1331
1332int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1333{
1334	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1335	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1336	int reg, rc, nr_regs;
1337	u32 ctl_array[16];
1338	u64 ga;
1339	u8 ar;
1340
1341	vcpu->stat.instruction_stctl++;
1342
1343	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1344		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1345
1346	ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1347
1348	if (ga & 3)
1349		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1350
1351	VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1352	trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1353
1354	reg = reg1;
1355	nr_regs = 0;
1356	do {
1357		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1358		if (reg == reg3)
1359			break;
1360		reg = (reg + 1) % 16;
1361	} while (1);
1362	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1363	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1364}
1365
1366static int handle_lctlg(struct kvm_vcpu *vcpu)
1367{
1368	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1369	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1370	int reg, rc, nr_regs;
1371	u64 ctl_array[16];
1372	u64 ga;
1373	u8 ar;
1374
1375	vcpu->stat.instruction_lctlg++;
1376
1377	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1378		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1379
1380	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1381
1382	if (ga & 7)
1383		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1384
1385	VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1386	trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1387
1388	nr_regs = ((reg3 - reg1) & 0xf) + 1;
1389	rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1390	if (rc)
1391		return kvm_s390_inject_prog_cond(vcpu, rc);
1392	reg = reg1;
1393	nr_regs = 0;
1394	do {
1395		vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1396		if (reg == reg3)
1397			break;
1398		reg = (reg + 1) % 16;
1399	} while (1);
1400	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1401	return 0;
1402}
1403
1404static int handle_stctg(struct kvm_vcpu *vcpu)
1405{
1406	int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1407	int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1408	int reg, rc, nr_regs;
1409	u64 ctl_array[16];
1410	u64 ga;
1411	u8 ar;
1412
1413	vcpu->stat.instruction_stctg++;
1414
1415	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1416		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1417
1418	ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1419
1420	if (ga & 7)
1421		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1422
1423	VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1424	trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1425
1426	reg = reg1;
1427	nr_regs = 0;
1428	do {
1429		ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1430		if (reg == reg3)
1431			break;
1432		reg = (reg + 1) % 16;
1433	} while (1);
1434	rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1435	return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1436}
1437
1438int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1439{
1440	switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1441	case 0x25:
1442		return handle_stctg(vcpu);
1443	case 0x2f:
1444		return handle_lctlg(vcpu);
1445	case 0x60:
1446	case 0x61:
1447	case 0x62:
1448		return handle_ri(vcpu);
1449	default:
1450		return -EOPNOTSUPP;
1451	}
1452}
1453
1454static int handle_tprot(struct kvm_vcpu *vcpu)
1455{
1456	u64 address1, address2;
1457	unsigned long hva, gpa;
1458	int ret = 0, cc = 0;
1459	bool writable;
1460	u8 ar;
1461
1462	vcpu->stat.instruction_tprot++;
1463
1464	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1465		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1466
1467	kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1468
1469	/* we only handle the Linux memory detection case:
1470	 * access key == 0
1471	 * everything else goes to userspace. */
1472	if (address2 & 0xf0)
1473		return -EOPNOTSUPP;
1474	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1475		ipte_lock(vcpu);
1476	ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1477	if (ret == PGM_PROTECTION) {
1478		/* Write protected? Try again with read-only... */
1479		cc = 1;
1480		ret = guest_translate_address(vcpu, address1, ar, &gpa,
1481					      GACC_FETCH);
1482	}
1483	if (ret) {
1484		if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1485			ret = kvm_s390_inject_program_int(vcpu, ret);
1486		} else if (ret > 0) {
1487			/* Translation not available */
1488			kvm_s390_set_psw_cc(vcpu, 3);
1489			ret = 0;
1490		}
1491		goto out_unlock;
1492	}
1493
1494	hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1495	if (kvm_is_error_hva(hva)) {
1496		ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1497	} else {
1498		if (!writable)
1499			cc = 1;		/* Write not permitted ==> read-only */
1500		kvm_s390_set_psw_cc(vcpu, cc);
1501		/* Note: CC2 only occurs for storage keys (not supported yet) */
1502	}
1503out_unlock:
1504	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1505		ipte_unlock(vcpu);
1506	return ret;
1507}
1508
1509int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1510{
1511	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1512	case 0x01:
1513		return handle_tprot(vcpu);
1514	default:
1515		return -EOPNOTSUPP;
1516	}
1517}
1518
1519static int handle_sckpf(struct kvm_vcpu *vcpu)
1520{
1521	u32 value;
1522
1523	vcpu->stat.instruction_sckpf++;
1524
1525	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1526		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1527
1528	if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1529		return kvm_s390_inject_program_int(vcpu,
1530						   PGM_SPECIFICATION);
1531
1532	value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1533	vcpu->arch.sie_block->todpr = value;
1534
1535	return 0;
1536}
1537
1538static int handle_ptff(struct kvm_vcpu *vcpu)
1539{
1540	vcpu->stat.instruction_ptff++;
1541
1542	/* we don't emulate any control instructions yet */
1543	kvm_s390_set_psw_cc(vcpu, 3);
1544	return 0;
1545}
1546
1547int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1548{
1549	switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1550	case 0x04:
1551		return handle_ptff(vcpu);
1552	case 0x07:
1553		return handle_sckpf(vcpu);
1554	default:
1555		return -EOPNOTSUPP;
1556	}
1557}
1558