2 * kvm guest debug support
4 * Copyright IBM Corp. 2014
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License (version 2 only)
8 * as published by the Free Software Foundation.
12 #include <linux/kvm_host.h>
13 #include <linux/errno.h>
18 * Extends the address range given by *start and *stop to include the address
19 * range starting with estart and the length len. Takes care of overflowing
20 * intervals and tries to minimize the overall interval size.
22 static void extend_address_range(u64 *start, u64 *stop, u64 estart, int len)
33 /* 0-0 range represents "not set" */
34 if ((*start == 0) && (*stop == 0)) {
37 } else if (*start <= *stop) {
38 /* increase the existing range */
44 /* "overflowing" interval, whereby *stop > *start */
45 if (estart <= *stop) {
48 } else if (estop > *start) {
52 /* minimize the range */
53 else if ((estop - *stop) < (*start - estart))
60 #define MAX_INST_SIZE 6
62 static void enable_all_hw_bp(struct kvm_vcpu *vcpu)
64 unsigned long start, len;
65 u64 *cr9 = &vcpu->arch.sie_block->gcr[9];
66 u64 *cr10 = &vcpu->arch.sie_block->gcr[10];
67 u64 *cr11 = &vcpu->arch.sie_block->gcr[11];
70 if (vcpu->arch.guestdbg.nr_hw_bp <= 0 ||
71 vcpu->arch.guestdbg.hw_bp_info == NULL)
75 * If the guest is not interested in branching events, we can safely
76 * limit them to the PER address range.
78 if (!(*cr9 & PER_EVENT_BRANCH))
79 *cr9 |= PER_CONTROL_BRANCH_ADDRESS;
80 *cr9 |= PER_EVENT_IFETCH | PER_EVENT_BRANCH;
82 for (i = 0; i < vcpu->arch.guestdbg.nr_hw_bp; i++) {
83 start = vcpu->arch.guestdbg.hw_bp_info[i].addr;
84 len = vcpu->arch.guestdbg.hw_bp_info[i].len;
87 * The instruction in front of the desired bp has to
88 * report instruction-fetching events
90 if (start < MAX_INST_SIZE) {
94 start -= MAX_INST_SIZE;
98 extend_address_range(cr10, cr11, start, len);
102 static void enable_all_hw_wp(struct kvm_vcpu *vcpu)
104 unsigned long start, len;
105 u64 *cr9 = &vcpu->arch.sie_block->gcr[9];
106 u64 *cr10 = &vcpu->arch.sie_block->gcr[10];
107 u64 *cr11 = &vcpu->arch.sie_block->gcr[11];
110 if (vcpu->arch.guestdbg.nr_hw_wp <= 0 ||
111 vcpu->arch.guestdbg.hw_wp_info == NULL)
114 /* if host uses storage alternation for special address
115 * spaces, enable all events and give all to the guest */
116 if (*cr9 & PER_EVENT_STORE && *cr9 & PER_CONTROL_ALTERATION) {
117 *cr9 &= ~PER_CONTROL_ALTERATION;
121 *cr9 &= ~PER_CONTROL_ALTERATION;
122 *cr9 |= PER_EVENT_STORE;
124 for (i = 0; i < vcpu->arch.guestdbg.nr_hw_wp; i++) {
125 start = vcpu->arch.guestdbg.hw_wp_info[i].addr;
126 len = vcpu->arch.guestdbg.hw_wp_info[i].len;
128 extend_address_range(cr10, cr11, start, len);
133 void kvm_s390_backup_guest_per_regs(struct kvm_vcpu *vcpu)
135 vcpu->arch.guestdbg.cr0 = vcpu->arch.sie_block->gcr[0];
136 vcpu->arch.guestdbg.cr9 = vcpu->arch.sie_block->gcr[9];
137 vcpu->arch.guestdbg.cr10 = vcpu->arch.sie_block->gcr[10];
138 vcpu->arch.guestdbg.cr11 = vcpu->arch.sie_block->gcr[11];
141 void kvm_s390_restore_guest_per_regs(struct kvm_vcpu *vcpu)
143 vcpu->arch.sie_block->gcr[0] = vcpu->arch.guestdbg.cr0;
144 vcpu->arch.sie_block->gcr[9] = vcpu->arch.guestdbg.cr9;
145 vcpu->arch.sie_block->gcr[10] = vcpu->arch.guestdbg.cr10;
146 vcpu->arch.sie_block->gcr[11] = vcpu->arch.guestdbg.cr11;
149 void kvm_s390_patch_guest_per_regs(struct kvm_vcpu *vcpu)
152 * TODO: if guest psw has per enabled, otherwise 0s!
153 * This reduces the amount of reported events.
154 * Need to intercept all psw changes!
157 if (guestdbg_sstep_enabled(vcpu)) {
158 /* disable timer (clock-comparator) interrupts */
159 vcpu->arch.sie_block->gcr[0] &= ~0x800ul;
160 vcpu->arch.sie_block->gcr[9] |= PER_EVENT_IFETCH;
161 vcpu->arch.sie_block->gcr[10] = 0;
162 vcpu->arch.sie_block->gcr[11] = -1UL;
165 if (guestdbg_hw_bp_enabled(vcpu)) {
166 enable_all_hw_bp(vcpu);
167 enable_all_hw_wp(vcpu);
170 /* TODO: Instruction-fetching-nullification not allowed for now */
171 if (vcpu->arch.sie_block->gcr[9] & PER_EVENT_NULLIFICATION)
172 vcpu->arch.sie_block->gcr[9] &= ~PER_EVENT_NULLIFICATION;
175 #define MAX_WP_SIZE 100
177 static int __import_wp_info(struct kvm_vcpu *vcpu,
178 struct kvm_hw_breakpoint *bp_data,
179 struct kvm_hw_wp_info_arch *wp_info)
182 wp_info->len = bp_data->len;
183 wp_info->addr = bp_data->addr;
184 wp_info->phys_addr = bp_data->phys_addr;
185 wp_info->old_data = NULL;
187 if (wp_info->len < 0 || wp_info->len > MAX_WP_SIZE)
190 wp_info->old_data = kmalloc(bp_data->len, GFP_KERNEL);
191 if (!wp_info->old_data)
193 /* try to backup the original value */
194 ret = read_guest_abs(vcpu, wp_info->phys_addr, wp_info->old_data,
197 kfree(wp_info->old_data);
198 wp_info->old_data = NULL;
204 #define MAX_BP_COUNT 50
206 int kvm_s390_import_bp_data(struct kvm_vcpu *vcpu,
207 struct kvm_guest_debug *dbg)
209 int ret = 0, nr_wp = 0, nr_bp = 0, i;
210 struct kvm_hw_breakpoint *bp_data = NULL;
211 struct kvm_hw_wp_info_arch *wp_info = NULL;
212 struct kvm_hw_bp_info_arch *bp_info = NULL;
214 if (dbg->arch.nr_hw_bp <= 0 || !dbg->arch.hw_bp)
216 else if (dbg->arch.nr_hw_bp > MAX_BP_COUNT)
219 bp_data = memdup_user(dbg->arch.hw_bp,
220 sizeof(*bp_data) * dbg->arch.nr_hw_bp);
222 return PTR_ERR(bp_data);
224 for (i = 0; i < dbg->arch.nr_hw_bp; i++) {
225 switch (bp_data[i].type) {
226 case KVM_HW_WP_WRITE:
238 wp_info = kmalloc_array(nr_wp,
247 bp_info = kmalloc_array(nr_bp,
256 for (nr_wp = 0, nr_bp = 0, i = 0; i < dbg->arch.nr_hw_bp; i++) {
257 switch (bp_data[i].type) {
258 case KVM_HW_WP_WRITE:
259 ret = __import_wp_info(vcpu, &bp_data[i],
266 bp_info[nr_bp].len = bp_data[i].len;
267 bp_info[nr_bp].addr = bp_data[i].addr;
273 vcpu->arch.guestdbg.nr_hw_bp = nr_bp;
274 vcpu->arch.guestdbg.hw_bp_info = bp_info;
275 vcpu->arch.guestdbg.nr_hw_wp = nr_wp;
276 vcpu->arch.guestdbg.hw_wp_info = wp_info;
285 void kvm_s390_clear_bp_data(struct kvm_vcpu *vcpu)
288 struct kvm_hw_wp_info_arch *hw_wp_info = NULL;
290 for (i = 0; i < vcpu->arch.guestdbg.nr_hw_wp; i++) {
291 hw_wp_info = &vcpu->arch.guestdbg.hw_wp_info[i];
292 kfree(hw_wp_info->old_data);
293 hw_wp_info->old_data = NULL;
295 kfree(vcpu->arch.guestdbg.hw_wp_info);
296 vcpu->arch.guestdbg.hw_wp_info = NULL;
298 kfree(vcpu->arch.guestdbg.hw_bp_info);
299 vcpu->arch.guestdbg.hw_bp_info = NULL;
301 vcpu->arch.guestdbg.nr_hw_wp = 0;
302 vcpu->arch.guestdbg.nr_hw_bp = 0;
305 static inline int in_addr_range(u64 addr, u64 a, u64 b)
308 return (addr >= a) && (addr <= b);
310 /* "overflowing" interval */
311 return (addr <= a) && (addr >= b);
314 #define end_of_range(bp_info) (bp_info->addr + bp_info->len - 1)
316 static struct kvm_hw_bp_info_arch *find_hw_bp(struct kvm_vcpu *vcpu,
319 struct kvm_hw_bp_info_arch *bp_info = vcpu->arch.guestdbg.hw_bp_info;
322 if (vcpu->arch.guestdbg.nr_hw_bp == 0)
325 for (i = 0; i < vcpu->arch.guestdbg.nr_hw_bp; i++) {
326 /* addr is directly the start or in the range of a bp */
327 if (addr == bp_info->addr)
329 if (bp_info->len > 0 &&
330 in_addr_range(addr, bp_info->addr, end_of_range(bp_info)))
341 static struct kvm_hw_wp_info_arch *any_wp_changed(struct kvm_vcpu *vcpu)
344 struct kvm_hw_wp_info_arch *wp_info = NULL;
347 if (vcpu->arch.guestdbg.nr_hw_wp == 0)
350 for (i = 0; i < vcpu->arch.guestdbg.nr_hw_wp; i++) {
351 wp_info = &vcpu->arch.guestdbg.hw_wp_info[i];
352 if (!wp_info || !wp_info->old_data || wp_info->len <= 0)
355 temp = kmalloc(wp_info->len, GFP_KERNEL);
359 /* refetch the wp data and compare it to the old value */
360 if (!read_guest_abs(vcpu, wp_info->phys_addr, temp,
362 if (memcmp(temp, wp_info->old_data, wp_info->len)) {
374 void kvm_s390_prepare_debug_exit(struct kvm_vcpu *vcpu)
376 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
377 vcpu->guest_debug &= ~KVM_GUESTDBG_EXIT_PENDING;
380 #define PER_CODE_MASK (PER_EVENT_MASK >> 24)
381 #define PER_CODE_BRANCH (PER_EVENT_BRANCH >> 24)
382 #define PER_CODE_IFETCH (PER_EVENT_IFETCH >> 24)
383 #define PER_CODE_STORE (PER_EVENT_STORE >> 24)
384 #define PER_CODE_STORE_REAL (PER_EVENT_STORE_REAL >> 24)
386 #define per_bp_event(code) \
387 (code & (PER_CODE_IFETCH | PER_CODE_BRANCH))
388 #define per_write_wp_event(code) \
389 (code & (PER_CODE_STORE | PER_CODE_STORE_REAL))
391 static int debug_exit_required(struct kvm_vcpu *vcpu, u8 perc,
392 unsigned long peraddr)
394 struct kvm_debug_exit_arch *debug_exit = &vcpu->run->debug.arch;
395 struct kvm_hw_wp_info_arch *wp_info = NULL;
396 struct kvm_hw_bp_info_arch *bp_info = NULL;
397 unsigned long addr = vcpu->arch.sie_block->gpsw.addr;
399 if (guestdbg_hw_bp_enabled(vcpu)) {
400 if (per_write_wp_event(perc) &&
401 vcpu->arch.guestdbg.nr_hw_wp > 0) {
402 wp_info = any_wp_changed(vcpu);
404 debug_exit->addr = wp_info->addr;
405 debug_exit->type = KVM_HW_WP_WRITE;
409 if (per_bp_event(perc) &&
410 vcpu->arch.guestdbg.nr_hw_bp > 0) {
411 bp_info = find_hw_bp(vcpu, addr);
412 /* remove duplicate events if PC==PER address */
413 if (bp_info && (addr != peraddr)) {
414 debug_exit->addr = addr;
415 debug_exit->type = KVM_HW_BP;
416 vcpu->arch.guestdbg.last_bp = addr;
419 /* breakpoint missed */
420 bp_info = find_hw_bp(vcpu, peraddr);
421 if (bp_info && vcpu->arch.guestdbg.last_bp != peraddr) {
422 debug_exit->addr = peraddr;
423 debug_exit->type = KVM_HW_BP;
428 if (guestdbg_sstep_enabled(vcpu) && per_bp_event(perc)) {
429 debug_exit->addr = addr;
430 debug_exit->type = KVM_SINGLESTEP;
439 static int per_fetched_addr(struct kvm_vcpu *vcpu, unsigned long *addr)
445 if (vcpu->arch.sie_block->icptcode == ICPT_PROGI) {
446 /* PER address references the fetched or the execute instr */
447 *addr = vcpu->arch.sie_block->peraddr;
449 * Manually detect if we have an EXECUTE instruction. As
450 * instructions are always 2 byte aligned we can read the
451 * first two bytes unconditionally
453 rc = read_guest_instr(vcpu, *addr, &opcode, 2);
456 if (opcode[0] >> 8 == 0x44)
458 if ((opcode[0] & 0xff0f) == 0xc600)
461 /* instr was suppressed, calculate the responsible instr */
462 *addr = __rewind_psw(vcpu->arch.sie_block->gpsw,
463 kvm_s390_get_ilen(vcpu));
464 if (vcpu->arch.sie_block->icptstatus & 0x01) {
465 exec_ilen = (vcpu->arch.sie_block->icptstatus & 0x60) >> 4;
472 /* read the complete EXECUTE instr to detect the fetched addr */
473 rc = read_guest_instr(vcpu, *addr, &opcode, exec_ilen);
476 if (exec_ilen == 6) {
477 /* EXECUTE RELATIVE LONG - RIL-b format */
478 s32 rl = *((s32 *) (opcode + 1));
480 /* rl is a _signed_ 32 bit value specifying halfwords */
481 *addr += (u64)(s64) rl * 2;
483 /* EXECUTE - RX-a format */
484 u32 base = (opcode[1] & 0xf000) >> 12;
485 u32 disp = opcode[1] & 0x0fff;
486 u32 index = opcode[0] & 0x000f;
488 *addr = base ? vcpu->run->s.regs.gprs[base] : 0;
489 *addr += index ? vcpu->run->s.regs.gprs[index] : 0;
492 *addr = kvm_s390_logical_to_effective(vcpu, *addr);
497 #define guest_per_enabled(vcpu) \
498 (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PER)
500 int kvm_s390_handle_per_ifetch_icpt(struct kvm_vcpu *vcpu)
502 const u64 cr10 = vcpu->arch.sie_block->gcr[10];
503 const u64 cr11 = vcpu->arch.sie_block->gcr[11];
504 const u8 ilen = kvm_s390_get_ilen(vcpu);
505 struct kvm_s390_pgm_info pgm_info = {
507 .per_code = PER_CODE_IFETCH,
508 .per_address = __rewind_psw(vcpu->arch.sie_block->gpsw, ilen),
510 unsigned long fetched_addr;
514 * The PSW points to the next instruction, therefore the intercepted
515 * instruction generated a PER i-fetch event. PER address therefore
516 * points at the previous PSW address (could be an EXECUTE function).
518 if (!guestdbg_enabled(vcpu))
519 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
521 if (debug_exit_required(vcpu, pgm_info.per_code, pgm_info.per_address))
522 vcpu->guest_debug |= KVM_GUESTDBG_EXIT_PENDING;
524 if (!guest_per_enabled(vcpu) ||
525 !(vcpu->arch.sie_block->gcr[9] & PER_EVENT_IFETCH))
528 rc = per_fetched_addr(vcpu, &fetched_addr);
532 /* instruction-fetching exceptions */
533 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
535 if (in_addr_range(fetched_addr, cr10, cr11))
536 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
540 static int filter_guest_per_event(struct kvm_vcpu *vcpu)
542 const u8 perc = vcpu->arch.sie_block->perc;
543 u64 addr = vcpu->arch.sie_block->gpsw.addr;
544 u64 cr9 = vcpu->arch.sie_block->gcr[9];
545 u64 cr10 = vcpu->arch.sie_block->gcr[10];
546 u64 cr11 = vcpu->arch.sie_block->gcr[11];
547 /* filter all events, demanded by the guest */
548 u8 guest_perc = perc & (cr9 >> 24) & PER_CODE_MASK;
549 unsigned long fetched_addr;
552 if (!guest_per_enabled(vcpu))
555 /* filter "successful-branching" events */
556 if (guest_perc & PER_CODE_BRANCH &&
557 cr9 & PER_CONTROL_BRANCH_ADDRESS &&
558 !in_addr_range(addr, cr10, cr11))
559 guest_perc &= ~PER_CODE_BRANCH;
561 /* filter "instruction-fetching" events */
562 if (guest_perc & PER_CODE_IFETCH) {
563 rc = per_fetched_addr(vcpu, &fetched_addr);
567 * Don't inject an irq on exceptions. This would make handling
568 * on icpt code 8 very complex (as PSW was already rewound).
570 if (rc || !in_addr_range(fetched_addr, cr10, cr11))
571 guest_perc &= ~PER_CODE_IFETCH;
574 /* All other PER events will be given to the guest */
575 /* TODO: Check altered address/address space */
577 vcpu->arch.sie_block->perc = guest_perc;
580 vcpu->arch.sie_block->iprcc &= ~PGM_PER;
584 #define pssec(vcpu) (vcpu->arch.sie_block->gcr[1] & _ASCE_SPACE_SWITCH)
585 #define hssec(vcpu) (vcpu->arch.sie_block->gcr[13] & _ASCE_SPACE_SWITCH)
586 #define old_ssec(vcpu) ((vcpu->arch.sie_block->tecmc >> 31) & 0x1)
587 #define old_as_is_home(vcpu) !(vcpu->arch.sie_block->tecmc & 0xffff)
589 int kvm_s390_handle_per_event(struct kvm_vcpu *vcpu)
593 if (debug_exit_required(vcpu, vcpu->arch.sie_block->perc,
594 vcpu->arch.sie_block->peraddr))
595 vcpu->guest_debug |= KVM_GUESTDBG_EXIT_PENDING;
597 rc = filter_guest_per_event(vcpu);
602 * Only RP, SAC, SACF, PT, PTI, PR, PC instructions can trigger
603 * a space-switch event. PER events enforce space-switch events
604 * for these instructions. So if no PER event for the guest is left,
605 * we might have to filter the space-switch element out, too.
607 if (vcpu->arch.sie_block->iprcc == PGM_SPACE_SWITCH) {
608 vcpu->arch.sie_block->iprcc = 0;
609 new_as = psw_bits(vcpu->arch.sie_block->gpsw).as;
612 * If the AS changed from / to home, we had RP, SAC or SACF
613 * instruction. Check primary and home space-switch-event
614 * controls. (theoretically home -> home produced no event)
616 if (((new_as == PSW_BITS_AS_HOME) ^ old_as_is_home(vcpu)) &&
617 (pssec(vcpu) || hssec(vcpu)))
618 vcpu->arch.sie_block->iprcc = PGM_SPACE_SWITCH;
621 * PT, PTI, PR, PC instruction operate on primary AS only. Check
622 * if the primary-space-switch-event control was or got set.
624 if (new_as == PSW_BITS_AS_PRIMARY && !old_as_is_home(vcpu) &&
625 (pssec(vcpu) || old_ssec(vcpu)))
626 vcpu->arch.sie_block->iprcc = PGM_SPACE_SWITCH;