Currently QEMU marks a vcpu state as dirty also on merely inspectional reads, e.g. via "info registers", which triggers a writeback to the hypervisor. For MSRs that cause a side-effect when written to this is undesired.
We introduce an MshvPutState flag into the writeback logic to differentiate "runtime" from "full" and "reset" writebacks that should occur e.g. after a migration or startup. Discrete side-effectful MSRs are excluded in runtime writebacks. So far we only mark one MSR as not runtime-safe, but probably more will be added eventually. Signed-off-by: Magnus Kulke <[email protected]> --- accel/mshv/mshv-all.c | 10 +++++----- include/system/mshv_int.h | 13 +++++++++++-- target/i386/mshv/mshv-cpu.c | 10 ++++++++-- target/i386/mshv/msr.c | 21 ++++++++++++++++++++- 4 files changed, 44 insertions(+), 10 deletions(-) diff --git a/accel/mshv/mshv-all.c b/accel/mshv/mshv-all.c index d820a67b95..d64599e2e2 100644 --- a/accel/mshv/mshv-all.c +++ b/accel/mshv/mshv-all.c @@ -759,10 +759,10 @@ static int mshv_cpu_exec(CPUState *cpu) do { if (cpu->vcpu_dirty) { - ret = mshv_arch_store_vcpu_state(cpu); + ret = mshv_arch_store_vcpu_state(cpu, MSHV_PUT_RUNTIME_STATE); if (ret) { - error_report("Failed to put registers after init: %s", - strerror(-ret)); + error_report("Failed to put registers at runtime: %s", + strerror(-ret)); ret = -1; break; } @@ -890,7 +890,7 @@ static void mshv_start_vcpu_thread(CPUState *cpu) static void do_mshv_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg) { - int ret = mshv_arch_store_vcpu_state(cpu); + int ret = mshv_arch_store_vcpu_state(cpu, MSHV_PUT_FULL_STATE); if (ret < 0) { error_report("Failed to put registers after init: %s", strerror(-ret)); abort(); @@ -906,7 +906,7 @@ static void mshv_cpu_synchronize_post_init(CPUState *cpu) static void mshv_cpu_synchronize_post_reset(CPUState *cpu) { - int ret = mshv_arch_store_vcpu_state(cpu); + int ret = mshv_arch_store_vcpu_state(cpu, MSHV_PUT_RESET_STATE); if (ret) { error_report("Failed to put registers after reset: %s", strerror(-ret)); diff --git a/include/system/mshv_int.h b/include/system/mshv_int.h index 1f77ffd152..b7b7c56443 100644 --- a/include/system/mshv_int.h +++ b/include/system/mshv_int.h @@ -98,6 +98,15 @@ typedef enum MshvVmExit { MshvVmExitDebug = 3, } MshvVmExit; +typedef enum MshvPutState { + /* state subset only touched by the VCPU itself during runtime */ + MSHV_PUT_RUNTIME_STATE = 1, + /* state subset modified during VCPU reset */ + MSHV_PUT_RESET_STATE = 2, + /* full state set, modified during initialization or on vmload */ + MSHV_PUT_FULL_STATE = 3, +} MshvPutState; + void mshv_init_mmio_emu(void); int mshv_create_vcpu(int vm_fd, uint8_t vp_index, int *cpu_fd); void mshv_remove_vcpu(int vm_fd, int cpu_fd); @@ -107,7 +116,7 @@ int mshv_set_generic_regs(const CPUState *cpu, const hv_register_assoc *assocs, size_t n_regs); int mshv_get_generic_regs(CPUState *cpu, hv_register_assoc *assocs, size_t n_regs); -int mshv_arch_store_vcpu_state(const CPUState *cpu); +int mshv_arch_store_vcpu_state(const CPUState *cpu, MshvPutState state); int mshv_arch_load_vcpu_state(CPUState *cpu); int mshv_arch_set_partition_msrs(const CPUState *cpu); int mshv_arch_set_mp_state(const CPUState *cpu); @@ -152,7 +161,7 @@ void mshv_log_global_stop(MemoryListener *listener); /* msr */ int mshv_init_msrs(const CPUState *cpu); int mshv_get_msrs(CPUState *cpu); -int mshv_set_msrs(const CPUState *cpu); +int mshv_set_msrs(const CPUState *cpu, MshvPutState put_state); /* synic */ int mshv_get_simp(int cpu_fd, uint8_t *page); diff --git a/target/i386/mshv/mshv-cpu.c b/target/i386/mshv/mshv-cpu.c index 4df73bb497..72ef01a597 100644 --- a/target/i386/mshv/mshv-cpu.c +++ b/target/i386/mshv/mshv-cpu.c @@ -1511,7 +1511,13 @@ static int set_synic_state(const CPUState *cpu) return 0; } -int mshv_arch_store_vcpu_state(const CPUState *cpu) +/* + * The put_state param is passed to decide whether we need a full restoration + * of all registers (e.g. after a migration ) or just a subset during runtime + * (e.g after "info registers"). The latter should avoid writing registers that + * cause a side-effect. + */ +int mshv_arch_store_vcpu_state(const CPUState *cpu, MshvPutState put_state) { int ret; @@ -1541,7 +1547,7 @@ int mshv_arch_store_vcpu_state(const CPUState *cpu) return ret; } - ret = mshv_set_msrs(cpu); + ret = mshv_set_msrs(cpu, put_state); if (ret < 0) { return ret; } diff --git a/target/i386/mshv/msr.c b/target/i386/mshv/msr.c index 96bd552916..a06844cc99 100644 --- a/target/i386/mshv/msr.c +++ b/target/i386/mshv/msr.c @@ -324,6 +324,20 @@ static bool msr_supported(uint32_t name) return true; } +static bool msr_is_runtime_safe(uint32_t name) +{ + switch (name) { + /* + * HV_X64_REGISTER_TSC will invalidate the guest's tsc_page if written to + * while the guest's timer is not frozen. + */ + case HV_X64_REGISTER_TSC: + return false; + default: + return true; + } +} + int mshv_get_msrs(CPUState *cpu) { int ret = 0; @@ -406,13 +420,14 @@ static void load_from_env(const CPUState *cpu, struct hv_register_assoc *assocs, load_from_env_mtrr_phys(cpu, &assocs[mtrr_offset], MSHV_MTRR_MSR_COUNT); } -int mshv_set_msrs(const CPUState *cpu) +int mshv_set_msrs(const CPUState *cpu, MshvPutState put_state) { size_t n_assocs = MSHV_MSR_TOTAL_COUNT; struct hv_register_assoc assocs[MSHV_MSR_TOTAL_COUNT] = { 0 }; int ret; size_t i, j; X86CPU *x86cpu = X86_CPU(cpu); + bool runtime = (put_state <= MSHV_PUT_RUNTIME_STATE); load_from_env(cpu, assocs, n_assocs); @@ -432,6 +447,10 @@ int mshv_set_msrs(const CPUState *cpu) continue; } + if (runtime && !msr_is_runtime_safe(name)) { + continue; + } + if (j != i) { assocs[j] = assocs[i]; } -- 2.34.1
