2 * ARM implementation of KVM hooks, 64 bit specific code
4 * Copyright Mian-M. Hamayun 2013, Virtual Open Systems
5 * Copyright Alex Bennée 2014, Linaro
7 * This work is licensed under the terms of the GNU GPL, version 2 or later.
8 * See the COPYING file in the top-level directory.
12 #include "qemu/osdep.h"
13 #include <sys/ioctl.h>
14 #include <sys/ptrace.h>
16 #include <linux/elf.h>
17 #include <linux/kvm.h>
19 #include "qemu-common.h"
21 #include "qemu/timer.h"
22 #include "qemu/error-report.h"
23 #include "qemu/host-utils.h"
24 #include "qemu/main-loop.h"
25 #include "exec/gdbstub.h"
26 #include "sysemu/sysemu.h"
27 #include "sysemu/kvm.h"
29 #include "internals.h"
31 static bool have_guest_debug;
34 * Although the ARM implementation of hardware assisted debugging
35 * allows for different breakpoints per-core, the current GDB
36 * interface treats them as a global pool of registers (which seems to
37 * be the case for x86, ppc and s390). As a result we store one copy
38 * of registers which is used for all active cores.
40 * Write access is serialised by virtue of the GDB protocol which
41 * updates things. Read access (i.e. when the values are copied to the
42 * vCPU) is also gated by GDB's run control.
44 * This is not unreasonable as most of the time debugging kernels you
45 * never know which core will eventually execute your function.
53 /* The watchpoint registers can cover more area than the requested
54 * watchpoint so we need to store the additional information
55 * somewhere. We also need to supply a CPUWatchpoint to the GDB stub
56 * when the watchpoint is hit.
61 CPUWatchpoint details;
64 /* Maximum and current break/watch point counts */
65 int max_hw_bps, max_hw_wps;
66 GArray *hw_breakpoints, *hw_watchpoints;
68 #define cur_hw_wps (hw_watchpoints->len)
69 #define cur_hw_bps (hw_breakpoints->len)
70 #define get_hw_bp(i) (&g_array_index(hw_breakpoints, HWBreakpoint, i))
71 #define get_hw_wp(i) (&g_array_index(hw_watchpoints, HWWatchpoint, i))
74 * kvm_arm_init_debug() - check for guest debug capabilities
77 * kvm_check_extension returns the number of debug registers we have
78 * or 0 if we have none.
81 static void kvm_arm_init_debug(CPUState *cs)
83 have_guest_debug = kvm_check_extension(cs->kvm_state,
84 KVM_CAP_SET_GUEST_DEBUG);
86 max_hw_wps = kvm_check_extension(cs->kvm_state, KVM_CAP_GUEST_DEBUG_HW_WPS);
87 hw_watchpoints = g_array_sized_new(true, true,
88 sizeof(HWWatchpoint), max_hw_wps);
90 max_hw_bps = kvm_check_extension(cs->kvm_state, KVM_CAP_GUEST_DEBUG_HW_BPS);
91 hw_breakpoints = g_array_sized_new(true, true,
92 sizeof(HWBreakpoint), max_hw_bps);
97 * insert_hw_breakpoint()
98 * @addr: address of breakpoint
100 * See ARM ARM D2.9.1 for details but here we are only going to create
101 * simple un-linked breakpoints (i.e. we don't chain breakpoints
102 * together to match address and context or vmid). The hardware is
103 * capable of fancier matching but that will require exposing that
104 * fanciness to GDB's interface
106 * DBGBCR<n>_EL1, Debug Breakpoint Control Registers
108 * 31 24 23 20 19 16 15 14 13 12 9 8 5 4 3 2 1 0
109 * +------+------+-------+-----+----+------+-----+------+-----+---+
110 * | RES0 | BT | LBN | SSC | HMC| RES0 | BAS | RES0 | PMC | E |
111 * +------+------+-------+-----+----+------+-----+------+-----+---+
113 * BT: Breakpoint type (0 = unlinked address match)
114 * LBN: Linked BP number (0 = unused)
115 * SSC/HMC/PMC: Security, Higher and Priv access control (Table D-12)
116 * BAS: Byte Address Select (RES1 for AArch64)
119 * DBGBVR<n>_EL1, Debug Breakpoint Value Registers
121 * 63 53 52 49 48 2 1 0
122 * +------+-----------+----------+-----+
123 * | RESS | VA[52:49] | VA[48:2] | 0 0 |
124 * +------+-----------+----------+-----+
126 * Depending on the addressing mode bits the top bits of the register
127 * are a sign extension of the highest applicable VA bit. Some
128 * versions of GDB don't do it correctly so we ensure they are correct
129 * here so future PC comparisons will work properly.
132 static int insert_hw_breakpoint(target_ulong addr)
135 .bcr = 0x1, /* BCR E=1, enable */
136 .bvr = sextract64(addr, 0, 53)
139 if (cur_hw_bps >= max_hw_bps) {
143 brk.bcr = deposit32(brk.bcr, 1, 2, 0x3); /* PMC = 11 */
144 brk.bcr = deposit32(brk.bcr, 5, 4, 0xf); /* BAS = RES1 */
146 g_array_append_val(hw_breakpoints, brk);
152 * delete_hw_breakpoint()
153 * @pc: address of breakpoint
155 * Delete a breakpoint and shuffle any above down
158 static int delete_hw_breakpoint(target_ulong pc)
161 for (i = 0; i < hw_breakpoints->len; i++) {
162 HWBreakpoint *brk = get_hw_bp(i);
163 if (brk->bvr == pc) {
164 g_array_remove_index(hw_breakpoints, i);
172 * insert_hw_watchpoint()
173 * @addr: address of watch point
175 * @type: type of watch point
177 * See ARM ARM D2.10. As with the breakpoints we can do some advanced
178 * stuff if we want to. The watch points can be linked with the break
179 * points above to make them context aware. However for simplicity
180 * currently we only deal with simple read/write watch points.
182 * D7.3.11 DBGWCR<n>_EL1, Debug Watchpoint Control Registers
184 * 31 29 28 24 23 21 20 19 16 15 14 13 12 5 4 3 2 1 0
185 * +------+-------+------+----+-----+-----+-----+-----+-----+-----+---+
186 * | RES0 | MASK | RES0 | WT | LBN | SSC | HMC | BAS | LSC | PAC | E |
187 * +------+-------+------+----+-----+-----+-----+-----+-----+-----+---+
189 * MASK: num bits addr mask (0=none,01/10=res,11=3 bits (8 bytes))
190 * WT: 0 - unlinked, 1 - linked (not currently used)
191 * LBN: Linked BP number (not currently used)
192 * SSC/HMC/PAC: Security, Higher and Priv access control (Table D2-11)
193 * BAS: Byte Address Select
194 * LSC: Load/Store control (01: load, 10: store, 11: both)
197 * The bottom 2 bits of the value register are masked. Therefore to
198 * break on any sizes smaller than an unaligned word you need to set
199 * MASK=0, BAS=bit per byte in question. For larger regions (^2) you
200 * need to ensure you mask the address as required and set BAS=0xff
203 static int insert_hw_watchpoint(target_ulong addr,
204 target_ulong len, int type)
207 .wcr = 1, /* E=1, enable */
208 .wvr = addr & (~0x7ULL),
209 .details = { .vaddr = addr, .len = len }
212 if (cur_hw_wps >= max_hw_wps) {
217 * HMC=0 SSC=0 PAC=3 will hit EL0 or EL1, any security state,
218 * valid whether EL3 is implemented or not
220 wp.wcr = deposit32(wp.wcr, 1, 2, 3);
223 case GDB_WATCHPOINT_READ:
224 wp.wcr = deposit32(wp.wcr, 3, 2, 1);
225 wp.details.flags = BP_MEM_READ;
227 case GDB_WATCHPOINT_WRITE:
228 wp.wcr = deposit32(wp.wcr, 3, 2, 2);
229 wp.details.flags = BP_MEM_WRITE;
231 case GDB_WATCHPOINT_ACCESS:
232 wp.wcr = deposit32(wp.wcr, 3, 2, 3);
233 wp.details.flags = BP_MEM_ACCESS;
236 g_assert_not_reached();
240 /* we align the address and set the bits in BAS */
241 int off = addr & 0x7;
242 int bas = (1 << len) - 1;
244 wp.wcr = deposit32(wp.wcr, 5 + off, 8 - off, bas);
246 /* For ranges above 8 bytes we need to be a power of 2 */
247 if (is_power_of_2(len)) {
248 int bits = ctz64(len);
250 wp.wvr &= ~((1 << bits) - 1);
251 wp.wcr = deposit32(wp.wcr, 24, 4, bits);
252 wp.wcr = deposit32(wp.wcr, 5, 8, 0xff);
258 g_array_append_val(hw_watchpoints, wp);
263 static bool check_watchpoint_in_range(int i, target_ulong addr)
265 HWWatchpoint *wp = get_hw_wp(i);
266 uint64_t addr_top, addr_bottom = wp->wvr;
267 int bas = extract32(wp->wcr, 5, 8);
268 int mask = extract32(wp->wcr, 24, 4);
271 addr_top = addr_bottom + (1 << mask);
273 /* BAS must be contiguous but can offset against the base
274 * address in DBGWVR */
275 addr_bottom = addr_bottom + ctz32(bas);
276 addr_top = addr_bottom + clo32(bas);
279 if (addr >= addr_bottom && addr <= addr_top) {
287 * delete_hw_watchpoint()
288 * @addr: address of breakpoint
290 * Delete a breakpoint and shuffle any above down
293 static int delete_hw_watchpoint(target_ulong addr,
294 target_ulong len, int type)
297 for (i = 0; i < cur_hw_wps; i++) {
298 if (check_watchpoint_in_range(i, addr)) {
299 g_array_remove_index(hw_watchpoints, i);
307 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
308 target_ulong len, int type)
311 case GDB_BREAKPOINT_HW:
312 return insert_hw_breakpoint(addr);
314 case GDB_WATCHPOINT_READ:
315 case GDB_WATCHPOINT_WRITE:
316 case GDB_WATCHPOINT_ACCESS:
317 return insert_hw_watchpoint(addr, len, type);
323 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
324 target_ulong len, int type)
327 case GDB_BREAKPOINT_HW:
328 return delete_hw_breakpoint(addr);
330 case GDB_WATCHPOINT_READ:
331 case GDB_WATCHPOINT_WRITE:
332 case GDB_WATCHPOINT_ACCESS:
333 return delete_hw_watchpoint(addr, len, type);
340 void kvm_arch_remove_all_hw_breakpoints(void)
342 if (cur_hw_wps > 0) {
343 g_array_remove_range(hw_watchpoints, 0, cur_hw_wps);
345 if (cur_hw_bps > 0) {
346 g_array_remove_range(hw_breakpoints, 0, cur_hw_bps);
350 void kvm_arm_copy_hw_debug_data(struct kvm_guest_debug_arch *ptr)
353 memset(ptr, 0, sizeof(struct kvm_guest_debug_arch));
355 for (i = 0; i < max_hw_wps; i++) {
356 HWWatchpoint *wp = get_hw_wp(i);
357 ptr->dbg_wcr[i] = wp->wcr;
358 ptr->dbg_wvr[i] = wp->wvr;
360 for (i = 0; i < max_hw_bps; i++) {
361 HWBreakpoint *bp = get_hw_bp(i);
362 ptr->dbg_bcr[i] = bp->bcr;
363 ptr->dbg_bvr[i] = bp->bvr;
367 bool kvm_arm_hw_debug_active(CPUState *cs)
369 return ((cur_hw_wps > 0) || (cur_hw_bps > 0));
372 static bool find_hw_breakpoint(CPUState *cpu, target_ulong pc)
376 for (i = 0; i < cur_hw_bps; i++) {
377 HWBreakpoint *bp = get_hw_bp(i);
385 static CPUWatchpoint *find_hw_watchpoint(CPUState *cpu, target_ulong addr)
389 for (i = 0; i < cur_hw_wps; i++) {
390 if (check_watchpoint_in_range(i, addr)) {
391 return &get_hw_wp(i)->details;
397 static bool kvm_arm_pmu_set_attr(CPUState *cs, struct kvm_device_attr *attr)
401 err = kvm_vcpu_ioctl(cs, KVM_HAS_DEVICE_ATTR, attr);
403 error_report("PMU: KVM_HAS_DEVICE_ATTR: %s", strerror(-err));
407 err = kvm_vcpu_ioctl(cs, KVM_SET_DEVICE_ATTR, attr);
409 error_report("PMU: KVM_SET_DEVICE_ATTR: %s", strerror(-err));
416 void kvm_arm_pmu_init(CPUState *cs)
418 struct kvm_device_attr attr = {
419 .group = KVM_ARM_VCPU_PMU_V3_CTRL,
420 .attr = KVM_ARM_VCPU_PMU_V3_INIT,
423 if (!ARM_CPU(cs)->has_pmu) {
426 if (!kvm_arm_pmu_set_attr(cs, &attr)) {
427 error_report("failed to init PMU");
432 void kvm_arm_pmu_set_irq(CPUState *cs, int irq)
434 struct kvm_device_attr attr = {
435 .group = KVM_ARM_VCPU_PMU_V3_CTRL,
436 .addr = (intptr_t)&irq,
437 .attr = KVM_ARM_VCPU_PMU_V3_IRQ,
440 if (!ARM_CPU(cs)->has_pmu) {
443 if (!kvm_arm_pmu_set_attr(cs, &attr)) {
444 error_report("failed to set irq for PMU");
449 static inline void set_feature(uint64_t *features, int feature)
451 *features |= 1ULL << feature;
454 static inline void unset_feature(uint64_t *features, int feature)
456 *features &= ~(1ULL << feature);
459 static int read_sys_reg32(int fd, uint32_t *pret, uint64_t id)
462 struct kvm_one_reg idreg = { .id = id, .addr = (uintptr_t)&ret };
465 assert((id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64);
466 err = ioctl(fd, KVM_GET_ONE_REG, &idreg);
474 static int read_sys_reg64(int fd, uint64_t *pret, uint64_t id)
476 struct kvm_one_reg idreg = { .id = id, .addr = (uintptr_t)pret };
478 assert((id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64);
479 return ioctl(fd, KVM_GET_ONE_REG, &idreg);
482 bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
484 /* Identify the feature bits corresponding to the host CPU, and
485 * fill out the ARMHostCPUClass fields accordingly. To do this
486 * we have to create a scratch VM, create a single CPU inside it,
487 * and then query that CPU for the relevant ID registers.
490 uint64_t features = 0;
493 /* Old kernels may not know about the PREFERRED_TARGET ioctl: however
494 * we know these will only support creating one kind of guest CPU,
495 * which is its preferred CPU type. Fortunately these old kernels
496 * support only a very limited number of CPUs.
498 static const uint32_t cpus_to_try[] = {
499 KVM_ARM_TARGET_AEM_V8,
500 KVM_ARM_TARGET_FOUNDATION_V8,
501 KVM_ARM_TARGET_CORTEX_A57,
502 QEMU_KVM_ARM_TARGET_NONE
504 struct kvm_vcpu_init init;
506 if (!kvm_arm_create_scratch_host_vcpu(cpus_to_try, fdarray, &init)) {
510 ahcf->target = init.target;
511 ahcf->dtb_compatible = "arm,arm-v8";
513 err = read_sys_reg64(fdarray[2], &ahcf->isar.id_aa64pfr0,
514 ARM64_SYS_REG(3, 0, 0, 4, 0));
515 if (unlikely(err < 0)) {
517 * Before v4.15, the kernel only exposed a limited number of system
518 * registers, not including any of the interesting AArch64 ID regs.
519 * For the most part we could leave these fields as zero with minimal
520 * effect, since this does not affect the values seen by the guest.
522 * However, it could cause problems down the line for QEMU,
523 * so provide a minimal v8.0 default.
525 * ??? Could read MIDR and use knowledge from cpu64.c.
526 * ??? Could map a page of memory into our temp guest and
527 * run the tiniest of hand-crafted kernels to extract
528 * the values seen by the guest.
529 * ??? Either of these sounds like too much effort just
530 * to work around running a modern host kernel.
532 ahcf->isar.id_aa64pfr0 = 0x00000011; /* EL1&0, AArch64 only */
535 err |= read_sys_reg64(fdarray[2], &ahcf->isar.id_aa64pfr1,
536 ARM64_SYS_REG(3, 0, 0, 4, 1));
537 err |= read_sys_reg64(fdarray[2], &ahcf->isar.id_aa64isar0,
538 ARM64_SYS_REG(3, 0, 0, 6, 0));
539 err |= read_sys_reg64(fdarray[2], &ahcf->isar.id_aa64isar1,
540 ARM64_SYS_REG(3, 0, 0, 6, 1));
541 err |= read_sys_reg64(fdarray[2], &ahcf->isar.id_aa64mmfr0,
542 ARM64_SYS_REG(3, 0, 0, 7, 0));
543 err |= read_sys_reg64(fdarray[2], &ahcf->isar.id_aa64mmfr1,
544 ARM64_SYS_REG(3, 0, 0, 7, 1));
547 * Note that if AArch32 support is not present in the host,
548 * the AArch32 sysregs are present to be read, but will
549 * return UNKNOWN values. This is neither better nor worse
550 * than skipping the reads and leaving 0, as we must avoid
551 * considering the values in every case.
553 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar0,
554 ARM64_SYS_REG(3, 0, 0, 2, 0));
555 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar1,
556 ARM64_SYS_REG(3, 0, 0, 2, 1));
557 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar2,
558 ARM64_SYS_REG(3, 0, 0, 2, 2));
559 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar3,
560 ARM64_SYS_REG(3, 0, 0, 2, 3));
561 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar4,
562 ARM64_SYS_REG(3, 0, 0, 2, 4));
563 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar5,
564 ARM64_SYS_REG(3, 0, 0, 2, 5));
565 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar6,
566 ARM64_SYS_REG(3, 0, 0, 2, 7));
568 err |= read_sys_reg32(fdarray[2], &ahcf->isar.mvfr0,
569 ARM64_SYS_REG(3, 0, 0, 3, 0));
570 err |= read_sys_reg32(fdarray[2], &ahcf->isar.mvfr1,
571 ARM64_SYS_REG(3, 0, 0, 3, 1));
572 err |= read_sys_reg32(fdarray[2], &ahcf->isar.mvfr2,
573 ARM64_SYS_REG(3, 0, 0, 3, 2));
576 kvm_arm_destroy_scratch_host_vcpu(fdarray);
582 /* We can assume any KVM supporting CPU is at least a v8
583 * with VFPv4+Neon; this in turn implies most of the other
586 set_feature(&features, ARM_FEATURE_V8);
587 set_feature(&features, ARM_FEATURE_VFP4);
588 set_feature(&features, ARM_FEATURE_NEON);
589 set_feature(&features, ARM_FEATURE_AARCH64);
590 set_feature(&features, ARM_FEATURE_PMU);
592 ahcf->features = features;
597 #define ARM_CPU_ID_MPIDR 3, 0, 0, 0, 5
599 int kvm_arch_init_vcpu(CPUState *cs)
603 ARMCPU *cpu = ARM_CPU(cs);
604 CPUARMState *env = &cpu->env;
606 if (cpu->kvm_target == QEMU_KVM_ARM_TARGET_NONE ||
607 !object_dynamic_cast(OBJECT(cpu), TYPE_AARCH64_CPU)) {
608 fprintf(stderr, "KVM is not supported for this guest CPU type\n");
612 /* Determine init features for this CPU */
613 memset(cpu->kvm_init_features, 0, sizeof(cpu->kvm_init_features));
614 if (cpu->start_powered_off) {
615 cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_POWER_OFF;
617 if (kvm_check_extension(cs->kvm_state, KVM_CAP_ARM_PSCI_0_2)) {
618 cpu->psci_version = 2;
619 cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_PSCI_0_2;
621 if (!arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
622 cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_EL1_32BIT;
624 if (!kvm_check_extension(cs->kvm_state, KVM_CAP_ARM_PMU_V3)) {
625 cpu->has_pmu = false;
628 cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_PMU_V3;
630 unset_feature(&env->features, ARM_FEATURE_PMU);
633 /* Do KVM_ARM_VCPU_INIT ioctl */
634 ret = kvm_arm_vcpu_init(cs);
640 * When KVM is in use, PSCI is emulated in-kernel and not by qemu.
641 * Currently KVM has its own idea about MPIDR assignment, so we
642 * override our defaults with what we get from KVM.
644 ret = kvm_get_one_reg(cs, ARM64_SYS_REG(ARM_CPU_ID_MPIDR), &mpidr);
648 cpu->mp_affinity = mpidr & ARM64_AFFINITY_MASK;
650 kvm_arm_init_debug(cs);
652 /* Check whether user space can specify guest syndrome value */
653 kvm_arm_init_serror_injection(cs);
655 return kvm_arm_init_cpreg_list(cpu);
658 int kvm_arch_destroy_vcpu(CPUState *cs)
663 bool kvm_arm_reg_syncs_via_cpreg_list(uint64_t regidx)
665 /* Return true if the regidx is a register we should synchronize
666 * via the cpreg_tuples array (ie is not a core reg we sync by
667 * hand in kvm_arch_get/put_registers())
669 switch (regidx & KVM_REG_ARM_COPROC_MASK) {
670 case KVM_REG_ARM_CORE:
677 typedef struct CPRegStateLevel {
682 /* All system registers not listed in the following table are assumed to be
683 * of the level KVM_PUT_RUNTIME_STATE. If a register should be written less
684 * often, you must add it to this table with a state of either
685 * KVM_PUT_RESET_STATE or KVM_PUT_FULL_STATE.
687 static const CPRegStateLevel non_runtime_cpregs[] = {
688 { KVM_REG_ARM_TIMER_CNT, KVM_PUT_FULL_STATE },
691 int kvm_arm_cpreg_level(uint64_t regidx)
695 for (i = 0; i < ARRAY_SIZE(non_runtime_cpregs); i++) {
696 const CPRegStateLevel *l = &non_runtime_cpregs[i];
697 if (l->regidx == regidx) {
702 return KVM_PUT_RUNTIME_STATE;
705 #define AARCH64_CORE_REG(x) (KVM_REG_ARM64 | KVM_REG_SIZE_U64 | \
706 KVM_REG_ARM_CORE | KVM_REG_ARM_CORE_REG(x))
708 #define AARCH64_SIMD_CORE_REG(x) (KVM_REG_ARM64 | KVM_REG_SIZE_U128 | \
709 KVM_REG_ARM_CORE | KVM_REG_ARM_CORE_REG(x))
711 #define AARCH64_SIMD_CTRL_REG(x) (KVM_REG_ARM64 | KVM_REG_SIZE_U32 | \
712 KVM_REG_ARM_CORE | KVM_REG_ARM_CORE_REG(x))
714 int kvm_arch_put_registers(CPUState *cs, int level)
716 struct kvm_one_reg reg;
723 ARMCPU *cpu = ARM_CPU(cs);
724 CPUARMState *env = &cpu->env;
726 /* If we are in AArch32 mode then we need to copy the AArch32 regs to the
727 * AArch64 registers before pushing them out to 64-bit KVM.
730 aarch64_sync_32_to_64(env);
733 for (i = 0; i < 31; i++) {
734 reg.id = AARCH64_CORE_REG(regs.regs[i]);
735 reg.addr = (uintptr_t) &env->xregs[i];
736 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
742 /* KVM puts SP_EL0 in regs.sp and SP_EL1 in regs.sp_el1. On the
743 * QEMU side we keep the current SP in xregs[31] as well.
745 aarch64_save_sp(env, 1);
747 reg.id = AARCH64_CORE_REG(regs.sp);
748 reg.addr = (uintptr_t) &env->sp_el[0];
749 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
754 reg.id = AARCH64_CORE_REG(sp_el1);
755 reg.addr = (uintptr_t) &env->sp_el[1];
756 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
761 /* Note that KVM thinks pstate is 64 bit but we use a uint32_t */
763 val = pstate_read(env);
765 val = cpsr_read(env);
767 reg.id = AARCH64_CORE_REG(regs.pstate);
768 reg.addr = (uintptr_t) &val;
769 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
774 reg.id = AARCH64_CORE_REG(regs.pc);
775 reg.addr = (uintptr_t) &env->pc;
776 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
781 reg.id = AARCH64_CORE_REG(elr_el1);
782 reg.addr = (uintptr_t) &env->elr_el[1];
783 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
788 /* Saved Program State Registers
790 * Before we restore from the banked_spsr[] array we need to
791 * ensure that any modifications to env->spsr are correctly
792 * reflected in the banks.
794 el = arm_current_el(env);
795 if (el > 0 && !is_a64(env)) {
796 i = bank_number(env->uncached_cpsr & CPSR_M);
797 env->banked_spsr[i] = env->spsr;
800 /* KVM 0-4 map to QEMU banks 1-5 */
801 for (i = 0; i < KVM_NR_SPSR; i++) {
802 reg.id = AARCH64_CORE_REG(spsr[i]);
803 reg.addr = (uintptr_t) &env->banked_spsr[i + 1];
804 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
810 /* Advanced SIMD and FP registers. */
811 for (i = 0; i < 32; i++) {
812 uint64_t *q = aa64_vfp_qreg(env, i);
813 #ifdef HOST_WORDS_BIGENDIAN
814 uint64_t fp_val[2] = { q[1], q[0] };
815 reg.addr = (uintptr_t)fp_val;
817 reg.addr = (uintptr_t)q;
819 reg.id = AARCH64_SIMD_CORE_REG(fp_regs.vregs[i]);
820 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
826 reg.addr = (uintptr_t)(&fpr);
827 fpr = vfp_get_fpsr(env);
828 reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpsr);
829 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
834 fpr = vfp_get_fpcr(env);
835 reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpcr);
836 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
841 ret = kvm_put_vcpu_events(cpu);
846 write_cpustate_to_list(cpu, true);
848 if (!write_list_to_kvmstate(cpu, level)) {
852 kvm_arm_sync_mpstate_to_kvm(cpu);
857 int kvm_arch_get_registers(CPUState *cs)
859 struct kvm_one_reg reg;
866 ARMCPU *cpu = ARM_CPU(cs);
867 CPUARMState *env = &cpu->env;
869 for (i = 0; i < 31; i++) {
870 reg.id = AARCH64_CORE_REG(regs.regs[i]);
871 reg.addr = (uintptr_t) &env->xregs[i];
872 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
878 reg.id = AARCH64_CORE_REG(regs.sp);
879 reg.addr = (uintptr_t) &env->sp_el[0];
880 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
885 reg.id = AARCH64_CORE_REG(sp_el1);
886 reg.addr = (uintptr_t) &env->sp_el[1];
887 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
892 reg.id = AARCH64_CORE_REG(regs.pstate);
893 reg.addr = (uintptr_t) &val;
894 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
899 env->aarch64 = ((val & PSTATE_nRW) == 0);
901 pstate_write(env, val);
903 cpsr_write(env, val, 0xffffffff, CPSRWriteRaw);
906 /* KVM puts SP_EL0 in regs.sp and SP_EL1 in regs.sp_el1. On the
907 * QEMU side we keep the current SP in xregs[31] as well.
909 aarch64_restore_sp(env, 1);
911 reg.id = AARCH64_CORE_REG(regs.pc);
912 reg.addr = (uintptr_t) &env->pc;
913 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
918 /* If we are in AArch32 mode then we need to sync the AArch32 regs with the
919 * incoming AArch64 regs received from 64-bit KVM.
920 * We must perform this after all of the registers have been acquired from
924 aarch64_sync_64_to_32(env);
927 reg.id = AARCH64_CORE_REG(elr_el1);
928 reg.addr = (uintptr_t) &env->elr_el[1];
929 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
934 /* Fetch the SPSR registers
936 * KVM SPSRs 0-4 map to QEMU banks 1-5
938 for (i = 0; i < KVM_NR_SPSR; i++) {
939 reg.id = AARCH64_CORE_REG(spsr[i]);
940 reg.addr = (uintptr_t) &env->banked_spsr[i + 1];
941 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
947 el = arm_current_el(env);
948 if (el > 0 && !is_a64(env)) {
949 i = bank_number(env->uncached_cpsr & CPSR_M);
950 env->spsr = env->banked_spsr[i];
953 /* Advanced SIMD and FP registers */
954 for (i = 0; i < 32; i++) {
955 uint64_t *q = aa64_vfp_qreg(env, i);
956 reg.id = AARCH64_SIMD_CORE_REG(fp_regs.vregs[i]);
957 reg.addr = (uintptr_t)q;
958 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
962 #ifdef HOST_WORDS_BIGENDIAN
964 t = q[0], q[0] = q[1], q[1] = t;
969 reg.addr = (uintptr_t)(&fpr);
970 reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpsr);
971 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
975 vfp_set_fpsr(env, fpr);
977 reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpcr);
978 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
982 vfp_set_fpcr(env, fpr);
984 ret = kvm_get_vcpu_events(cpu);
989 if (!write_kvmstate_to_list(cpu)) {
992 /* Note that it's OK to have registers which aren't in CPUState,
993 * so we can ignore a failure return here.
995 write_list_to_cpustate(cpu);
997 kvm_arm_sync_mpstate_to_qemu(cpu);
999 /* TODO: other registers */
1003 /* C6.6.29 BRK instruction */
1004 static const uint32_t brk_insn = 0xd4200000;
1006 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1008 if (have_guest_debug) {
1009 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 0) ||
1010 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&brk_insn, 4, 1)) {
1015 error_report("guest debug not supported on this kernel");
1020 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1022 static uint32_t brk;
1024 if (have_guest_debug) {
1025 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&brk, 4, 0) ||
1027 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 1)) {
1032 error_report("guest debug not supported on this kernel");
1037 /* See v8 ARM ARM D7.2.27 ESR_ELx, Exception Syndrome Register
1039 * To minimise translating between kernel and user-space the kernel
1040 * ABI just provides user-space with the full exception syndrome
1041 * register value to be decoded in QEMU.
1044 bool kvm_arm_handle_debug(CPUState *cs, struct kvm_debug_exit_arch *debug_exit)
1046 int hsr_ec = syn_get_ec(debug_exit->hsr);
1047 ARMCPU *cpu = ARM_CPU(cs);
1048 CPUClass *cc = CPU_GET_CLASS(cs);
1049 CPUARMState *env = &cpu->env;
1051 /* Ensure PC is synchronised */
1052 kvm_cpu_synchronize_state(cs);
1055 case EC_SOFTWARESTEP:
1056 if (cs->singlestep_enabled) {
1060 * The kernel should have suppressed the guest's ability to
1061 * single step at this point so something has gone wrong.
1063 error_report("%s: guest single-step while debugging unsupported"
1064 " (%"PRIx64", %"PRIx32")",
1065 __func__, env->pc, debug_exit->hsr);
1070 if (kvm_find_sw_breakpoint(cs, env->pc)) {
1075 if (find_hw_breakpoint(cs, env->pc)) {
1081 CPUWatchpoint *wp = find_hw_watchpoint(cs, debug_exit->far);
1083 cs->watchpoint_hit = wp;
1089 error_report("%s: unhandled debug exit (%"PRIx32", %"PRIx64")",
1090 __func__, debug_exit->hsr, env->pc);
1093 /* If we are not handling the debug exception it must belong to
1094 * the guest. Let's re-use the existing TCG interrupt code to set
1095 * everything up properly.
1097 cs->exception_index = EXCP_BKPT;
1098 env->exception.syndrome = debug_exit->hsr;
1099 env->exception.vaddress = debug_exit->far;
1100 env->exception.target_el = 1;
1101 qemu_mutex_lock_iothread();
1102 cc->do_interrupt(cs);
1103 qemu_mutex_unlock_iothread();