i386: KVM: expand Hyper-V features early and provide simple 'hv-default=on' option
v4 cover i386: KVM: expand Hyper-V features early and provide simple 'hv-default=on' option
Changes since v3:
- Make 'hv-default' override 'hv-*' options which were already set (e.g. 'hv-feature=on,hv-default' case) [Igor]. Make 'hv-passthrough' behave the same way.
- Add "i386: be more picky about implicit 'hv-evmcs' enablement" patch to avoid enabling 'hv-evmcs' with hv-default/hv-passthrough when guest CPU lacks VMX.
- Add "i386: support 'hv-passthrough,hv-feature=off' on the command line" patch to make 'hv-passthrough' semantics match the newly introduced 'hv-default'.
- Add "i386: track explicit 'hv-*' features enablement/disablement" patch to support the above mentioned changes.
- Expand qtest to check the above mentioned improvements.
Original description:
Upper layer tools like libvirt want to figure out which Hyper-V features are supported by the underlying stack (QEMU/KVM) but currently they are unable to do so. We have a nice 'hv_passthrough' CPU flag supported by QEMU but it has no effect on e.g. QMP's
query-cpu-model-expansion type=full model={"name":"host","props":{"hv-passthrough":true}}
command as we parse Hyper-V features after creating KVM vCPUs and not at feature expansion time. To support the use-case we first need to make KVM_GET_SUPPORTED_HV_CPUID ioctl a system-wide ioctl as the existing vCPU version can't be used that early. This is what KVM part does. With that done, we can make early Hyper-V feature expansion (this series).
In addition, provide a simple 'hv-default' option which enables (and requires from KVM) all currently supported Hyper-V enlightenments. Unlike 'hv-passthrough' mode, this is going to be migratable.
Vitaly Kuznetsov (21):
i386: keep hyperv_vendor string up-to-date
i386: invert hyperv_spinlock_attempts setting logic with
hv_passthrough
i386: always fill Hyper-V CPUID feature leaves from X86CPU data
i386: stop using env->features[] for filling Hyper-V CPUIDs
i386: introduce hyperv_feature_supported()
i386: introduce hv_cpuid_get_host()
i386: drop FEAT_HYPERV feature leaves
i386: introduce hv_cpuid_cache
i386: split hyperv_handle_properties() into
hyperv_expand_features()/hyperv_fill_cpuids()
i386: move eVMCS enablement to hyperv_init_vcpu()
i386: switch hyperv_expand_features() to using error_setg()
i386: adjust the expected KVM_GET_SUPPORTED_HV_CPUID array size
i386: prefer system KVM_GET_SUPPORTED_HV_CPUID ioctl over vCPU's one
i386: use global kvm_state in hyperv_enabled() check
i386: expand Hyper-V features during CPU feature expansion time
i386: track explicit 'hv-*' features enablement/disablement
i386: support 'hv-passthrough,hv-feature=off' on the command line
i386: be more picky about implicit 'hv-evmcs' enablement
i386: introduce kvm_hv_evmcs_available()
i386: provide simple 'hv-default=on' option
qtest/hyperv: Introduce a simple hyper-v test
MAINTAINERS | 1 + docs/hyperv.txt | 16 +- target/i386/cpu.c | 430 ++++++++++++++++++++--------- target/i386/cpu.h | 11 +- target/i386/kvm/kvm-stub.c | 10 + target/i386/kvm/kvm.c | 535 +++++++++++++++++++++---------------- target/i386/kvm/kvm_i386.h | 2 + tests/qtest/hyperv-test.c | 312 +++++++++++++++++++++ tests/qtest/meson.build | 3 +- 9 files changed, 950 insertions(+), 370 deletions(-) create mode 100644 tests/qtest/hyperv-test.c
Signature · 1 line
2.29.2
PATCH v4 1/21 i386: keep hyperv_vendor string up-to-date
When cpu->hyperv_vendor is not set manually we default to "Microsoft Hv" and in 'hv_passthrough' mode we get the information from the host. This information is stored in cpu->hyperv_vendor_id[] array but we don't update cpu->hyperv_vendor string so e.g. QMP's query-cpu-model-expansion output is incorrect.
--- target/i386/cpu.c | 19 +++++++++---------- target/i386/kvm/kvm.c | 4 ++++ 2 files changed, 13 insertions(+), 10 deletions(-)
target/i386/cpu.c +9 -10
@@ -6547,17 +6547,16 @@ static void x86_cpu_hyperv_realize(X86CPU *cpu)/* Hyper-V vendor id */ if (!cpu->hyperv_vendor) { -memcpy (cpu->hyperv_vendor_id,"Microsoft Hv" ,12 ); - }else { - len =strlen (cpu->hyperv_vendor); - -if (len >12 ) { -warn_report ("hv-vendor-id truncated to 12 characters" ); - len =12 ; - } -memset (cpu->hyperv_vendor_id,0 ,12 ); -memcpy (cpu->hyperv_vendor_id, cpu->hyperv_vendor, len); +object_property_set_str (OBJECT (cpu),"hv-vendor-id" ,"Microsoft Hv" , + &error_abort); + } + len =strlen (cpu->hyperv_vendor); +if (len >12 ) { +warn_report ("hv-vendor-id truncated to 12 characters" ); + len =12 ; } +memset (cpu->hyperv_vendor_id,0 ,12 ); +memcpy (cpu->hyperv_vendor_id, cpu->hyperv_vendor, len);/* 'Hv#1' interface identification*/ cpu->hyperv_interface_id[0 ] =0x31237648 ;
target/i386/kvm/kvm.c +4 -0
@@ -1214,6 +1214,10 @@ static int hyperv_handle_properties(CPUState *cs, cpu->hyperv_vendor_id[0 ] = c->ebx; cpu->hyperv_vendor_id[1 ] = c->ecx; cpu->hyperv_vendor_id[2 ] = c->edx; + cpu->hyperv_vendor =g_realloc (cpu->hyperv_vendor, +sizeof (cpu->hyperv_vendor_id) +1 ); +memcpy (cpu->hyperv_vendor, cpu->hyperv_vendor_id, +sizeof (cpu->hyperv_vendor_id)); } c =cpuid_find_entry (cpuid,HV_CPUID_INTERFACE ,0 );
Signature · 1 line
2.29.2
PATCH v4 2/21 i386: invert hyperv_spinlock_attempts setting logic with hv_passthrough
In reply to Vitaly KuznetsovThere is no need to have this special case: like all other Hyper-V enlightenments we can just use kernel's supplied value in hv_passthrough mode.
--- target/i386/kvm/kvm.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)
target/i386/kvm/kvm.c +1 -5
@@ -1254,11 +1254,7 @@ static int hyperv_handle_properties(CPUState *cs, c =cpuid_find_entry (cpuid,HV_CPUID_ENLIGHTMENT_INFO ,0 );if (c) { env->features[FEAT_HV_RECOMM_EAX ] = c->eax; - -/* hv-spinlocks may have been overriden */ -if (cpu->hyperv_spinlock_attempts !=HYPERV_SPINLOCK_NEVER_NOTIFY ) { - c->ebx = cpu->hyperv_spinlock_attempts; - } + cpu->hyperv_spinlock_attempts = c->ebx; } c =cpuid_find_entry (cpuid,HV_CPUID_NESTED_FEATURES ,0 );if (c) {
Signature · 1 line
2.29.2
PATCH v4 3/21 i386: always fill Hyper-V CPUID feature leaves from X86CPU data
In reply to Vitaly KuznetsovWe have all the required data in X86CPU already and as we are about to split hyperv_handle_properties() into hyperv_expand_features()/ hyperv_fill_cpuids() we can remove the blind copy. The functional change is that QEMU won't pass CPUID leaves it doesn't currently know about to the guest but arguably this is a good change.
--- target/i386/kvm/kvm.c | 9 --------- 1 file changed, 9 deletions(-)
target/i386/kvm/kvm.c +0 -9
@@ -1206,9 +1206,6 @@ static int hyperv_handle_properties(CPUState *cs, }if (cpu->hyperv_passthrough) { -memcpy (cpuid_ent, &cpuid->entries[0 ], - cpuid->nent *sizeof (cpuid->entries[0 ])); - c =cpuid_find_entry (cpuid,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,0 );if (c) { cpu->hyperv_vendor_id[0 ] = c->ebx;
@@ -1307,12 +1304,6 @@ static int hyperv_handle_properties(CPUState *cs, goto free; } -if (cpu->hyperv_passthrough) { -/* We already copied all feature words from KVM as is */ - r = cpuid->nent; - goto free; - } - c = &cpuid_ent[cpuid_i++]; c->function =HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ; c->eax = hyperv_feat_enabled(cpu, HYPERV_FEAT_EVMCS) ?
Signature · 1 line
2.29.2
PATCH v4 4/21 i386: stop using env->features[] for filling Hyper-V CPUIDs
In reply to Vitaly KuznetsovAs a preparatory patch to dropping Hyper-V CPUID leaves from feature_word_info[] stop using env->features[] as a temporary storage of Hyper-V CPUIDs, just build Hyper-V CPUID leaves directly from kvm_hyperv_properties[] data.
--- target/i386/cpu.h | 1 + target/i386/kvm/kvm.c | 80 +++++++++++++++++++++++-------------------- 2 files changed, 43 insertions(+), 38 deletions(-)
target/i386/cpu.h +1 -0
@@ -1678,6 +1678,7 @@ struct X86CPU {uint32_t hyperv_interface_id[4 ];uint32_t hyperv_version_id[4 ];uint32_t hyperv_limits[3 ]; +uint32_t hyperv_nested[4 ];bool check_cpuid;bool enforce_cpuid;
target/i386/kvm/kvm.c +42 -38
148 lines not on this page · raw message
Signature · 1 line
2.29.2
PATCH v4 5/21 i386: introduce hyperv_feature_supported()
In reply to Vitaly KuznetsovClean up hv_cpuid_check_and_set() by separating hyperv_feature_supported() off it. No functional change intended.
--- target/i386/kvm/kvm.c | 49 ++++++++++++++++++++++++++----------------- 1 file changed, 30 insertions(+), 19 deletions(-)
target/i386/kvm/kvm.c +30 -19
@@ -1106,13 +1106,33 @@ static int hv_cpuid_get_fw(struct kvm_cpuid2 *cpuid, int fw, uint32_t *r)return 0 ; } +static bool hyperv_feature_supported (struct kvm_cpuid2 *cpuid,int feature) +{ +uint32_t r, fw, bits; +int i; + +for (i =0 ; i <ARRAY_SIZE (kvm_hyperv_properties[feature].flags); i++) { + fw = kvm_hyperv_properties[feature].flags[i].fw; + bits = kvm_hyperv_properties[feature].flags[i].bits; + +if (!fw) { +continue ; + } + +if (hv_cpuid_get_fw (cpuid, fw, &r) || (r & bits) != bits) { +return false; + } + } + +return true; +} +static int hv_cpuid_check_and_set (CPUState *cs,struct kvm_cpuid2 *cpuid,int feature) {X86CPU *cpu =X86_CPU (cs); -uint32_t r, fw, bits;uint64_t deps; -int i, dep_feat; +int dep_feat;if (!hyperv_feat_enabled (cpu, feature) && !cpu->hyperv_passthrough) {return 0 ;
@@ -1131,23 +1151,14 @@ static int hv_cpuid_check_and_set(CPUState *cs, struct kvm_cpuid2 *cpuid, deps &= ~(1ull << dep_feat); } -for (i =0 ; i <ARRAY_SIZE (kvm_hyperv_properties[feature].flags); i++) { - fw = kvm_hyperv_properties[feature].flags[i].fw; - bits = kvm_hyperv_properties[feature].flags[i].bits; - -if (!fw) { -continue ; - } - -if (hv_cpuid_get_fw (cpuid, fw, &r) || (r & bits) != bits) { -if (hyperv_feat_enabled (cpu, feature)) { -fprintf (stderr, -"Hyper-V %s is not supported by kernel\n" , - kvm_hyperv_properties[feature].desc); -return 1 ; - }else { -return 0 ; - } +if (!hyperv_feature_supported (cpuid, feature)) { +if (hyperv_feat_enabled (cpu, feature)) { +fprintf (stderr, +"Hyper-V %s is not supported by kernel\n" , + kvm_hyperv_properties[feature].desc); +return 1 ; + }else { +return 0 ; } }
Signature · 1 line
2.29.2
PATCH v4 6/21 i386: introduce hv_cpuid_get_host()
In reply to Vitaly KuznetsovAs a preparation to implementing hv_cpuid_cache intro introduce hv_cpuid_get_host(). No functional change intended.
--- target/i386/kvm/kvm.c | 100 +++++++++++++++++++++++------------------- 1 file changed, 56 insertions(+), 44 deletions(-)
target/i386/kvm/kvm.c +56 -44
@@ -1106,6 +1106,19 @@ static int hv_cpuid_get_fw(struct kvm_cpuid2 *cpuid, int fw, uint32_t *r)return 0 ; } +static uint32_t hv_cpuid_get_host (struct kvm_cpuid2 *cpuid,uint32_t func, +int reg) +{ +struct kvm_cpuid_entry2 *entry; + + entry =cpuid_find_entry (cpuid, func,0 ); +if (!entry) { +return 0 ; + } + +return cpuid_entry_get_reg (entry, reg); +} +static bool hyperv_feature_supported (struct kvm_cpuid2 *cpuid,int feature) {uint32_t r, fw, bits;
@@ -1202,7 +1215,7 @@ static int hyperv_handle_properties(CPUState *cs, {X86CPU *cpu =X86_CPU (cs);struct kvm_cpuid2 *cpuid; -struct kvm_cpuid_entry2 *c, *c2; +struct kvm_cpuid_entry2 *c;uint32_t cpuid_i =0 ;int r;
@@ -1234,45 +1247,46 @@ static int hyperv_handle_properties(CPUState *cs, }if (cpu->hyperv_passthrough) { - c =cpuid_find_entry (cpuid,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,0 ); -if (c) { - cpu->hyperv_vendor_id[0 ] = c->ebx; - cpu->hyperv_vendor_id[1 ] = c->ecx; - cpu->hyperv_vendor_id[2 ] = c->edx; - cpu->hyperv_vendor =g_realloc (cpu->hyperv_vendor, -sizeof (cpu->hyperv_vendor_id) +1 ); -memcpy (cpu->hyperv_vendor, cpu->hyperv_vendor_id, -sizeof (cpu->hyperv_vendor_id)); - } - - c =cpuid_find_entry (cpuid,HV_CPUID_INTERFACE ,0 ); -if (c) { - cpu->hyperv_interface_id[0 ] = c->eax; - cpu->hyperv_interface_id[1 ] = c->ebx; - cpu->hyperv_interface_id[2 ] = c->ecx; - cpu->hyperv_interface_id[3 ] = c->edx; - } - - c =cpuid_find_entry (cpuid,HV_CPUID_VERSION ,0 ); -if (c) { - cpu->hyperv_version_id[0 ] = c->eax; - cpu->hyperv_version_id[1 ] = c->ebx; - cpu->hyperv_version_id[2 ] = c->ecx; - cpu->hyperv_version_id[3 ] = c->edx; - } - - c =cpuid_find_entry (cpuid,HV_CPUID_IMPLEMENT_LIMITS ,0 ); -if (c) { - cpu->hv_max_vps = c->eax; - cpu->hyperv_limits[0 ] = c->ebx; - cpu->hyperv_limits[1 ] = c->ecx; - cpu->hyperv_limits[2 ] = c->edx; - } - - c =cpuid_find_entry (cpuid,HV_CPUID_ENLIGHTMENT_INFO ,0 ); -if (c) { - cpu->hyperv_spinlock_attempts = c->ebx; - } + cpu->hyperv_vendor_id[0 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_EBX ); + cpu->hyperv_vendor_id[1 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_ECX ); + cpu->hyperv_vendor_id[2 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_EDX ); + cpu->hyperv_vendor =g_realloc (cpu->hyperv_vendor, +sizeof (cpu->hyperv_vendor_id) +1 ); +memcpy (cpu->hyperv_vendor, cpu->hyperv_vendor_id, +sizeof (cpu->hyperv_vendor_id)); + + cpu->hyperv_interface_id[0 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_INTERFACE ,R_EAX ); + cpu->hyperv_interface_id[1 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_INTERFACE ,R_EBX ); + cpu->hyperv_interface_id[2 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_INTERFACE ,R_ECX ); + cpu->hyperv_interface_id[3 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_INTERFACE ,R_EDX ); + + cpu->hyperv_version_id[0 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_VERSION ,R_EAX ); + cpu->hyperv_version_id[1 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_VERSION ,R_EBX ); + cpu->hyperv_version_id[2 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_VERSION ,R_ECX ); + cpu->hyperv_version_id[3 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_VERSION ,R_EDX ); + + cpu->hv_max_vps =hv_cpuid_get_host (cpuid,HV_CPUID_IMPLEMENT_LIMITS , +R_EAX ); + cpu->hyperv_limits[0 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_IMPLEMENT_LIMITS ,R_EBX ); + cpu->hyperv_limits[1 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_IMPLEMENT_LIMITS ,R_ECX ); + cpu->hyperv_limits[2 ] = +hv_cpuid_get_host (cpuid,HV_CPUID_IMPLEMENT_LIMITS ,R_EDX ); + + cpu->hyperv_spinlock_attempts = +hv_cpuid_get_host (cpuid,HV_CPUID_ENLIGHTMENT_INFO ,R_EBX ); }/* Features */
@@ -1346,10 +1360,8 @@ static int hyperv_handle_properties(CPUState *cs,if (cpu->hyperv_no_nonarch_cs ==ON_OFF_AUTO_ON ) { c->eax |=HV_NO_NONARCH_CORESHARING ; }else if (cpu->hyperv_no_nonarch_cs ==ON_OFF_AUTO_AUTO ) { - c2 =cpuid_find_entry (cpuid,HV_CPUID_ENLIGHTMENT_INFO ,0 ); -if (c2) { - c->eax |= c2->eax &HV_NO_NONARCH_CORESHARING ; - } + c->eax |=hv_cpuid_get_host (cpuid,HV_CPUID_ENLIGHTMENT_INFO ,R_EAX ) & +HV_NO_NONARCH_CORESHARING ; } c = &cpuid_ent[cpuid_i++];
Signature · 1 line
2.29.2
PATCH v4 7/21 i386: drop FEAT_HYPERV feature leaves
In reply to Vitaly KuznetsovHyper-V feature leaves are weird. We have some of them in feature_word_info[] array but we don't use feature_word_info magic to enable them. Neither do we use feature_dependencies[] mechanism to validate the configuration as it doesn't allign well with Hyper-V's many-to-many dependency chains. Some of the feature leaves hold not only feature bits, but also values. E.g. FEAT_HV_NESTED_EAX contains both features and the supported Enlightened VMCS range.
Hyper-V features are already represented in 'struct X86CPU' with uint64_t hyperv_features so duplicating them in env->features adds little (or zero) benefits. THe other half of Hyper-V emulation features is also stored with values in hyperv_vendor_id[], hyperv_limits[],... so env->features[] is already incomplete.
Remove Hyper-V feature leaves from env->features[] completely. kvm_hyperv_properties[] is converted to using raw CPUID func/reg pairs for features, this allows us to get rid of hv_cpuid_get_fw() conversion.
--- target/i386/cpu.c | 90 +---------------------------------- target/i386/cpu.h | 5 -- target/i386/kvm/kvm.c | 108 ++++++++++++++---------------------------- 3 files changed, 37 insertions(+), 166 deletions(-)
target/i386/cpu.c +1 -89
@@ -832,94 +832,6 @@ static FeatureWordInfo feature_word_info[FEATURE_WORDS] = {*/ .no_autoenable_flags = ~0U, }, -/* -* .feat_names are commented out for Hyper-V enlightenments because we -* don't want to have two different ways for enabling them on QEMU command -* line. Some features (e.g. "hyperv_time", "hyperv_vapic", ...) require -* enabling several feature bits simultaneously, exposing these bits -* individually may just confuse guests. -*/ - [FEAT_HYPERV_EAX] = { - .type = CPUID_FEATURE_WORD, - .feat_names = { -NULL /* hv_msr_vp_runtime_access */ ,NULL /* hv_msr_time_refcount_access */ , -NULL /* hv_msr_synic_access */ ,NULL /* hv_msr_stimer_access */ , -NULL /* hv_msr_apic_access */ ,NULL /* hv_msr_hypercall_access */ , -NULL /* hv_vpindex_access */ ,NULL /* hv_msr_reset_access */ , -NULL /* hv_msr_stats_access */ ,NULL /* hv_reftsc_access */ , -NULL /* hv_msr_idle_access */ ,NULL /* hv_msr_frequency_access */ , -NULL /* hv_msr_debug_access */ ,NULL /* hv_msr_reenlightenment_access */ , -NULL ,NULL , -NULL ,NULL ,NULL ,NULL , -NULL ,NULL ,NULL ,NULL , -NULL ,NULL ,NULL ,NULL , - NULL, NULL, NULL, NULL, - }, - .cpuid = { .eax = 0x40000003, .reg = R_EAX, }, - }, - [FEAT_HYPERV_EBX] = { - .type =CPUID_FEATURE_WORD , - .feat_names = { -NULL /* hv_create_partitions */ ,NULL /* hv_access_partition_id */ , -NULL /* hv_access_memory_pool */ ,NULL /* hv_adjust_message_buffers */ , -NULL /* hv_post_messages */ ,NULL /* hv_signal_events */ , -NULL /* hv_create_port */ ,NULL /* hv_connect_port */ , -NULL /* hv_access_stats */ ,NULL ,NULL ,NULL /* hv_debugging */ , -NULL /* hv_cpu_power_management */ ,NULL /* hv_configure_profiler */ , -NULL ,NULL , -NULL ,NULL ,NULL ,NULL , -NULL ,NULL ,NULL ,NULL , -NULL ,NULL ,NULL ,NULL , - NULL, NULL, NULL, NULL, - }, - .cpuid = { .eax = 0x40000003, .reg = R_EBX, }, - }, - [FEAT_HYPERV_EDX] = { - .type =CPUID_FEATURE_WORD , - .feat_names = { -NULL /* hv_mwait */ ,NULL /* hv_guest_debugging */ , -NULL /* hv_perf_monitor */ ,NULL /* hv_cpu_dynamic_part */ , -NULL /* hv_hypercall_params_xmm */ ,NULL /* hv_guest_idle_state */ , -NULL ,NULL , -NULL ,NULL ,NULL /* hv_guest_crash_msr */ ,NULL , -NULL ,NULL ,NULL ,NULL , -NULL ,NULL ,NULL ,NULL , -NULL ,NULL ,NULL ,NULL , -NULL ,NULL ,NULL ,NULL , - NULL, NULL, NULL, NULL, - }, - .cpuid = { .eax = 0x40000003, .reg = R_EDX, }, - }, - [FEAT_HV_RECOMM_EAX] = { - .type =CPUID_FEATURE_WORD , - .feat_names = { -NULL /* hv_recommend_pv_as_switch */ , -NULL /* hv_recommend_pv_tlbflush_local */ , -NULL /* hv_recommend_pv_tlbflush_remote */ , -NULL /* hv_recommend_msr_apic_access */ , -NULL /* hv_recommend_msr_reset */ , -NULL /* hv_recommend_relaxed_timing */ , -NULL /* hv_recommend_dma_remapping */ , -NULL /* hv_recommend_int_remapping */ , -NULL /* hv_recommend_x2apic_msrs */ , -NULL /* hv_recommend_autoeoi_deprecation */ , -NULL /* hv_recommend_pv_ipi */ , -NULL /* hv_recommend_ex_hypercalls */ , -NULL /* hv_hypervisor_is_nested */ , -NULL /* hv_recommend_int_mbec */ , -NULL /* hv_recommend_evmcs */ , -NULL , -NULL ,NULL ,NULL ,NULL , -NULL ,NULL ,NULL ,NULL , -NULL ,NULL ,NULL ,NULL , - NULL, NULL, NULL, NULL, - }, - .cpuid = { .eax = 0x40000004, .reg = R_EAX, }, - }, - [FEAT_HV_NESTED_EAX] = { - .type =CPUID_FEATURE_WORD , - .cpuid = { .eax = 0x4000000A, .reg = R_EAX, }, - }, [FEAT_SVM] = { .type = CPUID_FEATURE_WORD, .feat_names = {
@@ -6951,7 +6863,7 @@ static GuestPanicInformation *x86_cpu_get_crash_info(CPUState *cs)CPUX86State *env = &cpu->env;GuestPanicInformation *panic_info =NULL ; -if (env->features[FEAT_HYPERV_EDX ] &HV_GUEST_CRASH_MSR_AVAILABLE ) { +if (hyperv_feat_enabled (cpu,HYPERV_FEAT_CRASH )) { panic_info =g_malloc0 (sizeof (GuestPanicInformation)); panic_info->type =GUEST_PANIC_INFORMATION_TYPE_HYPER_V ;
target/i386/cpu.h +0 -5
@@ -518,11 +518,6 @@ typedef enum FeatureWord {FEAT_C000_0001_EDX ,/* CPUID[C000_0001].EDX */ FEAT_KVM ,/* CPUID[4000_0001].EAX (KVM_CPUID_FEATURES) */ FEAT_KVM_HINTS ,/* CPUID[4000_0001].EDX */ -FEAT_HYPERV_EAX ,/* CPUID[4000_0003].EAX */ -FEAT_HYPERV_EBX ,/* CPUID[4000_0003].EBX */ -FEAT_HYPERV_EDX ,/* CPUID[4000_0003].EDX */ -FEAT_HV_RECOMM_EAX ,/* CPUID[4000_0004].EAX */ -FEAT_HV_NESTED_EAX ,/* CPUID[4000_000A].EAX */ FEAT_SVM ,/* CPUID[8000_000A].EDX */ FEAT_XSAVE ,/* CPUID[EAX=0xd,ECX=1].EAX */ FEAT_6_EAX, /* CPUID[6].EAX */
target/i386/kvm/kvm.c +36 -72
250 lines not on this page · raw message
Signature · 1 line
2.29.2
PATCH v4 8/21 i386: introduce hv_cpuid_cache
In reply to Vitaly KuznetsovJust like with cpuid_cache, it makes no sense to call KVM_GET_SUPPORTED_HV_CPUID more than once and instead of (ab)using env->features[] and/or trying to keep all the code in one place, it is better to introduce persistent hv_cpuid_cache and hv_cpuid_get_host() accessor to it.
Note, hv_cpuid_get_fw() is converted to using hv_cpuid_get_host() just to be removed later with Hyper-V specific feature words.
--- target/i386/kvm/kvm.c | 109 ++++++++++++++++++++++-------------------- 1 file changed, 56 insertions(+), 53 deletions(-)
target/i386/kvm/kvm.c +56 -53
@@ -127,6 +127,7 @@ static int has_exception_payload;static bool has_msr_mcg_ext_ctl;static struct kvm_cpuid2 *cpuid_cache; +static struct kvm_cpuid2 *hv_cpuid_cache;static struct kvm_msr_list *kvm_feature_msrs;int kvm_has_pit_state2 (void )
@@ -1065,10 +1066,25 @@ static struct kvm_cpuid2 *get_supported_hv_cpuid_legacy(CPUState *cs)return cpuid; } -static uint32_t hv_cpuid_get_host (struct kvm_cpuid2 *cpuid,uint32_t func, -int reg) +static uint32_t hv_cpuid_get_host (CPUState *cs,uint32_t func,int reg) {struct kvm_cpuid_entry2 *entry; +struct kvm_cpuid2 *cpuid; + +if (hv_cpuid_cache) { + cpuid = hv_cpuid_cache; + }else { +if (kvm_check_extension (kvm_state,KVM_CAP_HYPERV_CPUID ) >0 ) { + cpuid =get_supported_hv_cpuid (cs); + }else { + cpuid =get_supported_hv_cpuid_legacy (cs); + } + hv_cpuid_cache = cpuid; + } + +if (!cpuid) { +return 0 ; + } entry =cpuid_find_entry (cpuid, func,0 );if (!entry) {
@@ -1078,7 +1094,7 @@ static uint32_t hv_cpuid_get_host(struct kvm_cpuid2 *cpuid, uint32_t func,return cpuid_entry_get_reg (entry, reg); } -static bool hyperv_feature_supported (struct kvm_cpuid2 *cpuid,int feature) +static bool hyperv_feature_supported (CPUState *cs,int feature) {uint32_t func, bits;int i, reg;
@@ -1093,7 +1109,7 @@ static bool hyperv_feature_supported(struct kvm_cpuid2 *cpuid, int feature)continue ; } -if ((hv_cpuid_get_host (cpuid, func, reg) & bits) != bits) { +if ((hv_cpuid_get_host (cs, func, reg) & bits) != bits) {return false; } }
@@ -1101,8 +1117,7 @@ static bool hyperv_feature_supported(struct kvm_cpuid2 *cpuid, int feature)return true; } -static int hv_cpuid_check_and_set (CPUState *cs,struct kvm_cpuid2 *cpuid, -int feature) +static int hv_cpuid_check_and_set (CPUState *cs,int feature) {X86CPU *cpu =X86_CPU (cs);uint64_t deps;
@@ -1125,7 +1140,7 @@ static int hv_cpuid_check_and_set(CPUState *cs, struct kvm_cpuid2 *cpuid, deps &= ~(1ull << dep_feat); } -if (!hyperv_feature_supported (cpuid, feature)) { +if (!hyperv_feature_supported (cs, feature)) {if (hyperv_feat_enabled (cpu, feature)) { fprintf(stderr, "Hyper-V %s is not supported by kernel\n",
@@ -1178,7 +1193,6 @@ static int hyperv_handle_properties(CPUState *cs, struct kvm_cpuid_entry2 *cpuid_ent) {X86CPU *cpu =X86_CPU (cs); -struct kvm_cpuid2 *cpuid;struct kvm_cpuid_entry2 *c;uint32_t cpuid_i =0 ;int r;
@@ -1204,71 +1218,65 @@ static int hyperv_handle_properties(CPUState *cs, } } -if (kvm_check_extension (cs->kvm_state,KVM_CAP_HYPERV_CPUID ) >0 ) { - cpuid =get_supported_hv_cpuid (cs); - }else { - cpuid =get_supported_hv_cpuid_legacy (cs); - } -if (cpu->hyperv_passthrough) { cpu->hyperv_vendor_id[0 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_EBX ); +hv_cpuid_get_host (cs,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_EBX ); cpu->hyperv_vendor_id[1 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_ECX ); +hv_cpuid_get_host (cs,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_ECX ); cpu->hyperv_vendor_id[2 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_EDX ); +hv_cpuid_get_host (cs,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_EDX ); cpu->hyperv_vendor =g_realloc (cpu->hyperv_vendor,sizeof (cpu->hyperv_vendor_id) +1 );memcpy (cpu->hyperv_vendor, cpu->hyperv_vendor_id,sizeof (cpu->hyperv_vendor_id)); cpu->hyperv_interface_id[0 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_INTERFACE ,R_EAX ); +hv_cpuid_get_host (cs,HV_CPUID_INTERFACE ,R_EAX ); cpu->hyperv_interface_id[1 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_INTERFACE ,R_EBX ); +hv_cpuid_get_host (cs,HV_CPUID_INTERFACE ,R_EBX ); cpu->hyperv_interface_id[2 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_INTERFACE ,R_ECX ); +hv_cpuid_get_host (cs,HV_CPUID_INTERFACE ,R_ECX ); cpu->hyperv_interface_id[3 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_INTERFACE ,R_EDX ); +hv_cpuid_get_host (cs,HV_CPUID_INTERFACE ,R_EDX ); cpu->hyperv_version_id[0 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_VERSION ,R_EAX ); +hv_cpuid_get_host (cs,HV_CPUID_VERSION ,R_EAX ); cpu->hyperv_version_id[1 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_VERSION ,R_EBX ); +hv_cpuid_get_host (cs,HV_CPUID_VERSION ,R_EBX ); cpu->hyperv_version_id[2 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_VERSION ,R_ECX ); +hv_cpuid_get_host (cs,HV_CPUID_VERSION ,R_ECX ); cpu->hyperv_version_id[3 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_VERSION ,R_EDX ); +hv_cpuid_get_host (cs,HV_CPUID_VERSION ,R_EDX ); - cpu->hv_max_vps =hv_cpuid_get_host (cpuid,HV_CPUID_IMPLEMENT_LIMITS , + cpu->hv_max_vps =hv_cpuid_get_host (cs,HV_CPUID_IMPLEMENT_LIMITS ,R_EAX ); cpu->hyperv_limits[0 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_IMPLEMENT_LIMITS ,R_EBX ); +hv_cpuid_get_host (cs,HV_CPUID_IMPLEMENT_LIMITS ,R_EBX ); cpu->hyperv_limits[1 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_IMPLEMENT_LIMITS ,R_ECX ); +hv_cpuid_get_host (cs,HV_CPUID_IMPLEMENT_LIMITS ,R_ECX ); cpu->hyperv_limits[2 ] = -hv_cpuid_get_host (cpuid,HV_CPUID_IMPLEMENT_LIMITS ,R_EDX ); +hv_cpuid_get_host (cs,HV_CPUID_IMPLEMENT_LIMITS ,R_EDX ); cpu->hyperv_spinlock_attempts = -hv_cpuid_get_host (cpuid,HV_CPUID_ENLIGHTMENT_INFO ,R_EBX ); +hv_cpuid_get_host (cs,HV_CPUID_ENLIGHTMENT_INFO ,R_EBX ); }/* Features */ - r =hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_RELAXED ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_VAPIC ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_TIME ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_CRASH ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_RESET ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_VPINDEX ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_RUNTIME ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_SYNIC ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_STIMER ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_FREQUENCIES ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_REENLIGHTENMENT ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_TLBFLUSH ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_EVMCS ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_IPI ); - r |=hv_cpuid_check_and_set (cs, cpuid,HYPERV_FEAT_STIMER_DIRECT ); + r =hv_cpuid_check_and_set (cs,HYPERV_FEAT_RELAXED ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_VAPIC ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_TIME ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_CRASH ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_RESET ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_VPINDEX ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_RUNTIME ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_SYNIC ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_STIMER ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_FREQUENCIES ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_REENLIGHTENMENT ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_TLBFLUSH ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_EVMCS ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_IPI ); + r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_STIMER_DIRECT );/* Additional dependencies not covered by kvm_hyperv_properties[] */ if (hyperv_feat_enabled(cpu, HYPERV_FEAT_SYNIC) &&
@@ -1281,8 +1289,7 @@ static int hyperv_handle_properties(CPUState *cs, }if (r) { - r = -ENOSYS ; - goto free; +return -ENOSYS ; } c = &cpuid_ent[cpuid_i++];
@@ -1324,7 +1331,7 @@ static int hyperv_handle_properties(CPUState *cs,if (cpu->hyperv_no_nonarch_cs ==ON_OFF_AUTO_ON ) { c->eax |=HV_NO_NONARCH_CORESHARING ; }else if (cpu->hyperv_no_nonarch_cs ==ON_OFF_AUTO_AUTO ) { - c->eax |=hv_cpuid_get_host (cpuid,HV_CPUID_ENLIGHTMENT_INFO ,R_EAX ) & + c->eax |=hv_cpuid_get_host (cs,HV_CPUID_ENLIGHTMENT_INFO ,R_EAX ) &HV_NO_NONARCH_CORESHARING ; }
@@ -1349,12 +1356,8 @@ static int hyperv_handle_properties(CPUState *cs, c->function =HV_CPUID_NESTED_FEATURES ; c->eax = cpu->hyperv_nested[0 ]; } - r = cpuid_i; -free: -g_free (cpuid); - -return r; +return cpuid_i; }static Error *hv_passthrough_mig_blocker;
Signature · 1 line
2.29.2
PATCH v4 9/21 i386: split hyperv_handle_properties() into hyperv_expand_features()/hyperv_fill_cpuids()
In reply to Vitaly KuznetsovThe intention is to call hyperv_expand_features() early, before vCPUs are created and use the acquired data later when we set guest visible CPUID data.
--- target/i386/kvm/kvm.c | 34 ++++++++++++++++++++++++---------- 1 file changed, 24 insertions(+), 10 deletions(-)
target/i386/kvm/kvm.c +24 -10
@@ -1185,16 +1185,15 @@ static uint32_t hv_build_cpuid_leaf(CPUState *cs, uint32_t func, int reg) }/* -* Fill in Hyper-V CPUIDs. Returns the number of entries filled in cpuid_ent in -* case of success, errno < 0 in case of failure and 0 when no Hyper-V -* extentions are enabled. +* Expand Hyper-V CPU features. In partucular, check that all the requested +* features are supported by the host and the sanity of the configuration +* (that all the required dependencies are included). Also, this takes care +* of 'hv_passthrough' mode and fills the environment with all supported +* Hyper-V features. */ -static int hyperv_handle_properties (CPUState *cs, -struct kvm_cpuid_entry2 *cpuid_ent) +static int hyperv_expand_features (CPUState *cs) {X86CPU *cpu =X86_CPU (cs); -struct kvm_cpuid_entry2 *c; -uint32_t cpuid_i =0 ;int r;if (!hyperv_enabled (cpu))
@@ -1292,6 +1291,19 @@ static int hyperv_handle_properties(CPUState *cs,return -ENOSYS ; } +return 0 ; +} + +/* +* Fill in Hyper-V CPUIDs. Returns the number of entries filled in cpuid_ent. +*/ +static int hyperv_fill_cpuids (CPUState *cs, +struct kvm_cpuid_entry2 *cpuid_ent) +{ +X86CPU *cpu =X86_CPU (cs); +struct kvm_cpuid_entry2 *c; +uint32_t cpuid_i =0 ; + c = &cpuid_ent[cpuid_i++]; c->function =HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ; c->eax = hyperv_feat_enabled(cpu, HYPERV_FEAT_EVMCS) ?
@@ -1499,11 +1511,13 @@ int kvm_arch_init_vcpu(CPUState *cs) env->apic_bus_freq =KVM_APIC_BUS_FREQUENCY ;/* Paravirtualization CPUIDs */ - r =hyperv_handle_properties (cs, cpuid_data.entries); + r =hyperv_expand_features (cs);if (r <0 ) {return r; - }else if (r >0 ) { - cpuid_i = r; + } + +if (hyperv_enabled (cpu)) { + cpuid_i =hyperv_fill_cpuids (cs, cpuid_data.entries); kvm_base =KVM_CPUID_SIGNATURE_NEXT ; has_msr_hv_hypercall = true; }
Signature · 1 line
2.29.2
PATCH v4 10/21 i386: move eVMCS enablement to hyperv_init_vcpu()
In reply to Vitaly Kuznetsovhyperv_expand_features() will be called before we create vCPU so evmcs enablement should go away. hyperv_init_vcpu() looks like the right place.
--- target/i386/kvm/kvm.c | 60 ++++++++++++++++++++++++++----------------- 1 file changed, 37 insertions(+), 23 deletions(-)
target/i386/kvm/kvm.c +37 -23
@@ -961,6 +961,7 @@ static struct kvm_cpuid2 *get_supported_hv_cpuid(CPUState *cs) {struct kvm_cpuid2 *cpuid;int max =7 ;/* 0x40000000..0x40000005, 0x4000000A */ +int i;/* * When the buffer is too small, KVM_GET_SUPPORTED_HV_CPUID fails with
@@ -970,6 +971,22 @@ static struct kvm_cpuid2 *get_supported_hv_cpuid(CPUState *cs)while ((cpuid =try_get_hv_cpuid (cs, max)) ==NULL ) { max++; } + +/* +* KVM_GET_SUPPORTED_HV_CPUID does not set EVMCS CPUID bit before +* KVM_CAP_HYPERV_ENLIGHTENED_VMCS is enabled but we want to get the +* information early, just check for the capability and set the bit +* manually. +*/ +if (kvm_check_extension (cs->kvm_state, +KVM_CAP_HYPERV_ENLIGHTENED_VMCS ) >0 ) { +for (i =0 ; i < cpuid->nent; i++) { +if (cpuid->entries[i].function ==HV_CPUID_ENLIGHTMENT_INFO ) { + cpuid->entries[i].eax |=HV_ENLIGHTENED_VMCS_RECOMMENDED ; + } + } + } +return cpuid; }
@@ -1199,24 +1216,6 @@ static int hyperv_expand_features(CPUState *cs)if (!hyperv_enabled (cpu))return 0 ; -if (hyperv_feat_enabled (cpu,HYPERV_FEAT_EVMCS ) || - cpu->hyperv_passthrough) { -uint16_t evmcs_version; - - r =kvm_vcpu_enable_cap (cs,KVM_CAP_HYPERV_ENLIGHTENED_VMCS ,0 , - (uintptr_t )&evmcs_version); - -if (hyperv_feat_enabled (cpu,HYPERV_FEAT_EVMCS ) && r) { -fprintf (stderr,"Hyper-V %s is not supported by kernel\n" , - kvm_hyperv_properties[HYPERV_FEAT_EVMCS ].desc); -return -ENOSYS ; - } - -if (!r) { - cpu->hyperv_nested[0 ] = evmcs_version; - } - } -if (cpu->hyperv_passthrough) { cpu->hyperv_vendor_id[0 ] =hv_cpuid_get_host (cs,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_EBX );
@@ -1453,6 +1452,21 @@ static int hyperv_init_vcpu(X86CPU *cpu) } } +if (hyperv_feat_enabled (cpu,HYPERV_FEAT_EVMCS )) { +uint16_t evmcs_version; + + ret =kvm_vcpu_enable_cap (cs,KVM_CAP_HYPERV_ENLIGHTENED_VMCS ,0 , + (uintptr_t )&evmcs_version); + +if (ret <0 ) { +fprintf (stderr,"Hyper-V %s is not supported by kernel\n" , + kvm_hyperv_properties[HYPERV_FEAT_EVMCS ].desc); +return ret; + } + + cpu->hyperv_nested[0 ] = evmcs_version; + } +return 0 ; }
@@ -1517,6 +1531,11 @@ int kvm_arch_init_vcpu(CPUState *cs) }if (hyperv_enabled (cpu)) { + r =hyperv_init_vcpu (cpu); +if (r) { +return r; + } + cpuid_i =hyperv_fill_cpuids (cs, cpuid_data.entries); kvm_base =KVM_CPUID_SIGNATURE_NEXT ; has_msr_hv_hypercall = true;
@@ -1866,11 +1885,6 @@ int kvm_arch_init_vcpu(CPUState *cs)kvm_init_msrs (cpu); - r =hyperv_init_vcpu (cpu); -if (r) { - goto fail; - } -return 0 ; fail:
Signature · 1 line
2.29.2
PATCH v4 11/21 i386: switch hyperv_expand_features() to using error_setg()
In reply to Vitaly KuznetsovUse standard error_setg() mechanism in hyperv_expand_features().
--- target/i386/kvm/kvm.c | 101 +++++++++++++++++++++++++----------------- 1 file changed, 61 insertions(+), 40 deletions(-)
target/i386/kvm/kvm.c +61 -40
@@ -1134,7 +1134,7 @@ static bool hyperv_feature_supported(CPUState *cs, int feature)return true; } -static int hv_cpuid_check_and_set (CPUState *cs,int feature) +static int hv_cpuid_check_and_set (CPUState *cs,int feature,Error **errp) {X86CPU *cpu =X86_CPU (cs);uint64_t deps;
@@ -1148,20 +1148,18 @@ static int hv_cpuid_check_and_set(CPUState *cs, int feature)while (deps) { dep_feat =ctz64 (deps);if (!(hyperv_feat_enabled (cpu, dep_feat))) { -fprintf (stderr, -"Hyper-V %s requires Hyper-V %s\n" , - kvm_hyperv_properties[feature].desc, - kvm_hyperv_properties[dep_feat].desc); -return 1 ; +error_setg (errp,"Hyper-V %s requires Hyper-V %s" , + kvm_hyperv_properties[feature].desc, + kvm_hyperv_properties[dep_feat].desc); +return 1 ; } deps &= ~(1ull << dep_feat); }if (!hyperv_feature_supported (cs, feature)) {if (hyperv_feat_enabled (cpu, feature)) { -fprintf (stderr, -"Hyper-V %s is not supported by kernel\n" , - kvm_hyperv_properties[feature].desc); +error_setg (errp,"Hyper-V %s is not supported by kernel" , + kvm_hyperv_properties[feature].desc);return 1 ; }else {return 0 ;
@@ -1208,13 +1206,12 @@ static uint32_t hv_build_cpuid_leaf(CPUState *cs, uint32_t func, int reg)* of 'hv_passthrough' mode and fills the environment with all supported * Hyper-V features. */ -static int hyperv_expand_features (CPUState *cs) +static void hyperv_expand_features (CPUState *cs,Error **errp) {X86CPU *cpu =X86_CPU (cs); -int r;if (!hyperv_enabled (cpu)) -return 0 ; +return ;if (cpu->hyperv_passthrough) { cpu->hyperv_vendor_id[0] =
@@ -1260,37 +1257,60 @@ static int hyperv_expand_features(CPUState *cs) }/* Features */ - r =hv_cpuid_check_and_set (cs,HYPERV_FEAT_RELAXED ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_VAPIC ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_TIME ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_CRASH ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_RESET ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_VPINDEX ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_RUNTIME ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_SYNIC ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_STIMER ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_FREQUENCIES ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_REENLIGHTENMENT ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_TLBFLUSH ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_EVMCS ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_IPI ); - r |=hv_cpuid_check_and_set (cs,HYPERV_FEAT_STIMER_DIRECT ); +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_RELAXED , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_VAPIC , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_TIME , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_CRASH , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_RESET , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_VPINDEX , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_RUNTIME , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_SYNIC , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_STIMER , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_FREQUENCIES , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_REENLIGHTENMENT , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_TLBFLUSH , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_EVMCS , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_IPI , errp)) { +return ; + } +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_STIMER_DIRECT , errp)) { +return ; + }/* Additional dependencies not covered by kvm_hyperv_properties[] */ if (hyperv_feat_enabled (cpu,HYPERV_FEAT_SYNIC ) && !cpu->hyperv_synic_kvm_only && !hyperv_feat_enabled (cpu,HYPERV_FEAT_VPINDEX )) { -fprintf (stderr,"Hyper-V %s requires Hyper-V %s\n" , - kvm_hyperv_properties[HYPERV_FEAT_SYNIC ].desc, - kvm_hyperv_properties[HYPERV_FEAT_VPINDEX ].desc); - r |=1 ; - } - -if (r) { -return -ENOSYS ; +error_setg (errp,"Hyper-V %s requires Hyper-V %s" , + kvm_hyperv_properties[HYPERV_FEAT_SYNIC ].desc, + kvm_hyperv_properties[HYPERV_FEAT_VPINDEX ].desc); } - -return 0 ; }/*
@@ -1525,9 +1545,10 @@ int kvm_arch_init_vcpu(CPUState *cs) env->apic_bus_freq =KVM_APIC_BUS_FREQUENCY ;/* Paravirtualization CPUIDs */ - r =hyperv_expand_features (cs); -if (r <0 ) { -return r; +hyperv_expand_features (cs, &local_err); +if (local_err) { +error_report_err (local_err); +return -ENOSYS ; }if (hyperv_enabled (cpu)) {
Signature · 1 line
2.29.2
PATCH v4 12/21 i386: adjust the expected KVM_GET_SUPPORTED_HV_CPUID array size
In reply to Vitaly KuznetsovSYNDBG leaves were recently (Linux-5.8) added to KVM but we haven't updated the expected size of KVM_GET_SUPPORTED_HV_CPUID output in KVM so we now make serveral tries before succeeding. Update the default.
--- target/i386/kvm/kvm.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
target/i386/kvm/kvm.c +2 -1
@@ -960,7 +960,8 @@ static struct kvm_cpuid2 *try_get_hv_cpuid(CPUState *cs, int max)static struct kvm_cpuid2 *get_supported_hv_cpuid (CPUState *cs) {struct kvm_cpuid2 *cpuid; -int max =7 ;/* 0x40000000..0x40000005, 0x4000000A */ +/* 0x40000000..0x40000005, 0x4000000A, 0x40000080..0x40000080 leaves */ +int max =10 ;int i;/*
Signature · 1 line
2.29.2
PATCH v4 13/21 i386: prefer system KVM_GET_SUPPORTED_HV_CPUID ioctl over vCPU's one
In reply to Vitaly KuznetsovKVM_GET_SUPPORTED_HV_CPUID was made a system wide ioctl which can be called prior to creating vCPUs and we are going to use that to expand Hyper-V cpu features early. Use it when it is supported by KVM.
--- target/i386/kvm/kvm.c | 17 +++++++++++++---- 1 file changed, 13 insertions(+), 4 deletions(-)
target/i386/kvm/kvm.c +13 -4
@@ -927,7 +927,8 @@ static struct { }, }; -static struct kvm_cpuid2 *try_get_hv_cpuid (CPUState *cs,int max) +static struct kvm_cpuid2 *try_get_hv_cpuid (CPUState *cs,int max, +bool do_sys_ioctl) {struct kvm_cpuid2 *cpuid;int r, size;
@@ -936,7 +937,11 @@ static struct kvm_cpuid2 *try_get_hv_cpuid(CPUState *cs, int max) cpuid =g_malloc0 (size); cpuid->nent = max; - r =kvm_vcpu_ioctl (cs,KVM_GET_SUPPORTED_HV_CPUID , cpuid); +if (do_sys_ioctl) { + r =kvm_ioctl (kvm_state,KVM_GET_SUPPORTED_HV_CPUID , cpuid); + }else { + r =kvm_vcpu_ioctl (cs,KVM_GET_SUPPORTED_HV_CPUID , cpuid); + }if (r ==0 && cpuid->nent >= max) { r = -E2BIG ; }
@@ -963,13 +968,17 @@ static struct kvm_cpuid2 *get_supported_hv_cpuid(CPUState *cs)/* 0x40000000..0x40000005, 0x4000000A, 0x40000080..0x40000080 leaves */ int max =10 ;int i; +bool do_sys_ioctl; + + do_sys_ioctl = +kvm_check_extension (kvm_state,KVM_CAP_SYS_HYPERV_CPUID ) >0 ;/* * When the buffer is too small, KVM_GET_SUPPORTED_HV_CPUID fails with * -E2BIG, however, it doesn't report back the right size. Keep increasing * it and re-trying until we succeed. */ -while ((cpuid =try_get_hv_cpuid (cs, max)) ==NULL ) { +while ((cpuid =try_get_hv_cpuid (cs, max, do_sys_ioctl)) ==NULL ) { max++; }
@@ -979,7 +988,7 @@ static struct kvm_cpuid2 *get_supported_hv_cpuid(CPUState *cs)* information early, just check for the capability and set the bit * manually. */ -if (kvm_check_extension (cs->kvm_state, +if (!do_sys_ioctl &&kvm_check_extension (cs->kvm_state,KVM_CAP_HYPERV_ENLIGHTENED_VMCS ) >0 ) {for (i =0 ; i < cpuid->nent; i++) {if (cpuid->entries[i].function ==HV_CPUID_ENLIGHTMENT_INFO ) {
Signature · 1 line
2.29.2
PATCH v4 14/21 i386: use global kvm_state in hyperv_enabled() check
In reply to Vitaly KuznetsovThere is no need to use vCPU-specific kvm state in hyperv_enabled() check and we need to do that when feature expansion happens early, before vCPU specific KVM state is created.
--- target/i386/kvm/kvm.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
target/i386/kvm/kvm.c +1 -2
@@ -714,8 +714,7 @@ unsigned long kvm_arch_vcpu_id(CPUState *cs)static bool hyperv_enabled (X86CPU *cpu) { -CPUState *cs =CPU (cpu); -return kvm_check_extension (cs->kvm_state,KVM_CAP_HYPERV ) >0 && +return kvm_check_extension (kvm_state,KVM_CAP_HYPERV ) >0 && ((cpu->hyperv_spinlock_attempts !=HYPERV_SPINLOCK_NEVER_NOTIFY ) || cpu->hyperv_features || cpu->hyperv_passthrough); }
Signature · 1 line
2.29.2
PATCH v4 15/21 i386: expand Hyper-V features during CPU feature expansion time
In reply to Vitaly KuznetsovTo make Hyper-V features appear in e.g. QMP query-cpu-model-expansion we need to expand and set the corresponding CPUID leaves early. Modify x86_cpu_get_supported_feature_word() to call newly intoduced Hyper-V specific kvm_hv_get_supported_cpuid() instead of kvm_arch_get_supported_cpuid(). We can't use kvm_arch_get_supported_cpuid() as Hyper-V specific CPUID leaves intersect with KVM's.
Note, early expansion will only happen when KVM supports system wide KVM_GET_SUPPORTED_HV_CPUID ioctl (KVM_CAP_SYS_HYPERV_CPUID).
--- target/i386/cpu.c | 4 ++++ target/i386/kvm/kvm-stub.c | 5 +++++ target/i386/kvm/kvm.c | 15 ++++++++++++--- target/i386/kvm/kvm_i386.h | 1 + 4 files changed, 22 insertions(+), 3 deletions(-)
target/i386/cpu.c +4 -0
@@ -6396,6 +6396,10 @@ static void x86_cpu_expand_features(X86CPU *cpu, Error **errp)if (env->cpuid_xlevel2 ==UINT32_MAX ) { env->cpuid_xlevel2 = env->cpuid_min_xlevel2; } + +if (kvm_enabled ()) { +kvm_hyperv_expand_features (cpu, errp); + } }/*
target/i386/kvm/kvm-stub.c +5 -0
@@ -39,3 +39,8 @@ bool kvm_hv_vpindex_settable(void) {return false; } + +void kvm_hyperv_expand_features (X86CPU *cpu,Error **errp) +{ +return ; +}
target/i386/kvm/kvm.c +12 -3
@@ -1215,13 +1215,22 @@ static uint32_t hv_build_cpuid_leaf(CPUState *cs, uint32_t func, int reg)* of 'hv_passthrough' mode and fills the environment with all supported * Hyper-V features. */ -static void hyperv_expand_features (CPUState *cs,Error **errp) +void kvm_hyperv_expand_features (X86CPU *cpu,Error **errp) { -X86CPU *cpu =X86_CPU (cs); +CPUState *cs =CPU (cpu);if (!hyperv_enabled (cpu))return ; +/* +* When kvm_hyperv_expand_features is called at CPU feature expansion +* time per-CPU kvm_state is not available yet so we can only proceed +* when KVM_CAP_SYS_HYPERV_CPUID is supported. +*/ +if (!cs->kvm_state && + !kvm_check_extension (kvm_state,KVM_CAP_SYS_HYPERV_CPUID )) +return ; +if (cpu->hyperv_passthrough) { cpu->hyperv_vendor_id[0 ] =hv_cpuid_get_host (cs,HV_CPUID_VENDOR_AND_MAX_FUNCTIONS ,R_EBX );
@@ -1554,7 +1563,7 @@ int kvm_arch_init_vcpu(CPUState *cs) env->apic_bus_freq =KVM_APIC_BUS_FREQUENCY ;/* Paravirtualization CPUIDs */ -hyperv_expand_features (cs, &local_err); +kvm_hyperv_expand_features (cpu, &local_err);if (local_err) {error_report_err (local_err);return -ENOSYS ;
target/i386/kvm/kvm_i386.h +1 -0
@@ -47,6 +47,7 @@ bool kvm_has_x2apic_api(void);bool kvm_has_waitpkg (void );bool kvm_hv_vpindex_settable (void ); +void kvm_hyperv_expand_features (X86CPU *cpu,Error **errp);uint64_t kvm_swizzle_msi_ext_dest_id (uint64_t address);
Signature · 1 line
2.29.2
PATCH v4 16/21 i386: track explicit 'hv-*' features enablement/disablement
In reply to Vitaly KuznetsovSometimes we'd like to know which features were explicitly enabled and which were explicitly disabled on the command line. E.g. it seems logical to handle 'hv_passthrough,hv_feature=off' as "enable everything supported by the host except for hv_feature" but this doesn't seem to be possible with the current 'hyperv_features' bit array. Introduce 'hv_features_on'/'hv_features_off' add-ons and track explicit enablement/disablement there.
Note, it doesn't seem to be possible to fill 'hyperv_features' array during CPU creation time when 'hv-passthrough' is specified and we're running on an older kernel without KVM_CAP_SYS_HYPERV_CPUID support. To get the list of the supported Hyper-V features we need to actually create KVM VCPU and this happens much later.
No functional change intended.
--- target/i386/cpu.c | 237 ++++++++++++++++++++++++++++++++++++++++------ target/i386/cpu.h | 2 + 2 files changed, 209 insertions(+), 30 deletions(-)
target/i386/cpu.c +207 -30
@@ -4553,6 +4553,178 @@ static void x86_cpuid_set_tsc_freq(Object *obj, Visitor *v, const char *name, cpu->env.tsc_khz = cpu->env.user_tsc_khz = value /1000 ; } +static bool x86_hv_feature_get (Object *obj,int feature) +{ +X86CPU *cpu =X86_CPU (obj); + +return cpu->hyperv_features &BIT (feature); +} + +static void x86_hv_feature_set (Object *obj,bool value,int feature) +{ +X86CPU *cpu =X86_CPU (obj); + +if (value) { + cpu->hyperv_features |=BIT (feature); + cpu->hyperv_features_on |=BIT (feature); + cpu->hyperv_features_off &= ~BIT (feature); + }else { + cpu->hyperv_features &= ~BIT (feature); + cpu->hyperv_features_on &= ~BIT (feature); + cpu->hyperv_features_off |=BIT (feature); + } +} + +static bool x86_hv_relaxed_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_RELAXED ); +} + +static void x86_hv_relaxed_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_RELAXED ); +} + +static bool x86_hv_vapic_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_VAPIC ); +} + +static void x86_hv_vapic_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_VAPIC ); +} + +static bool x86_hv_time_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_TIME ); +} + +static void x86_hv_time_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_TIME ); +} + +static bool x86_hv_crash_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_CRASH ); +} + +static void x86_hv_crash_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_CRASH ); +} + +static bool x86_hv_reset_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_RESET ); +} + +static void x86_hv_reset_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_RESET ); +} + +static bool x86_hv_vpindex_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_VPINDEX ); +} + +static void x86_hv_vpindex_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_VPINDEX ); +} + +static bool x86_hv_runtime_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_RUNTIME ); +} + +static void x86_hv_runtime_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_RUNTIME ); +} + +static bool x86_hv_synic_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_SYNIC ); +} + +static void x86_hv_synic_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_SYNIC ); +} + +static bool x86_hv_stimer_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_STIMER ); +} + +static void x86_hv_stimer_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_STIMER ); +} + +static bool x86_hv_frequencies_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_FREQUENCIES ); +} + +static void x86_hv_frequencies_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_FREQUENCIES ); +} + +static bool x86_hv_reenlightenment_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_REENLIGHTENMENT ); +} + +static void x86_hv_reenlightenment_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_REENLIGHTENMENT ); +} + +static bool x86_hv_tlbflush_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_TLBFLUSH ); +} + +static void x86_hv_tlbflush_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_TLBFLUSH ); +} + +static bool x86_hv_evmcs_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_EVMCS ); +} + +static void x86_hv_evmcs_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_EVMCS ); +} + +static bool x86_hv_ipi_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_IPI ); +} + +static void x86_hv_ipi_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_IPI ); +} + +static bool x86_hv_stimer_direct_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_STIMER_DIRECT ); +} + +static void x86_hv_stimer_direct_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_STIMER_DIRECT ); +} +/* Generic getter for "feature-words" and "filtered-features" properties */ static void x86_cpu_get_feature_words(Object *obj, Visitor *v, const char *name, void *opaque,
@@ -7107,36 +7279,6 @@ static Property x86_cpu_properties[] = {DEFINE_PROP_UINT32 ("hv-spinlocks" ,X86CPU , hyperv_spinlock_attempts,HYPERV_SPINLOCK_NEVER_NOTIFY ), -DEFINE_PROP_BIT64 ("hv-relaxed" ,X86CPU , hyperv_features, -HYPERV_FEAT_RELAXED ,0 ), -DEFINE_PROP_BIT64 ("hv-vapic" ,X86CPU , hyperv_features, -HYPERV_FEAT_VAPIC ,0 ), -DEFINE_PROP_BIT64 ("hv-time" ,X86CPU , hyperv_features, -HYPERV_FEAT_TIME ,0 ), -DEFINE_PROP_BIT64 ("hv-crash" ,X86CPU , hyperv_features, -HYPERV_FEAT_CRASH ,0 ), -DEFINE_PROP_BIT64 ("hv-reset" ,X86CPU , hyperv_features, -HYPERV_FEAT_RESET ,0 ), -DEFINE_PROP_BIT64 ("hv-vpindex" ,X86CPU , hyperv_features, -HYPERV_FEAT_VPINDEX ,0 ), -DEFINE_PROP_BIT64 ("hv-runtime" ,X86CPU , hyperv_features, -HYPERV_FEAT_RUNTIME ,0 ), -DEFINE_PROP_BIT64 ("hv-synic" ,X86CPU , hyperv_features, -HYPERV_FEAT_SYNIC ,0 ), -DEFINE_PROP_BIT64 ("hv-stimer" ,X86CPU , hyperv_features, -HYPERV_FEAT_STIMER ,0 ), -DEFINE_PROP_BIT64 ("hv-frequencies" ,X86CPU , hyperv_features, -HYPERV_FEAT_FREQUENCIES ,0 ), -DEFINE_PROP_BIT64 ("hv-reenlightenment" ,X86CPU , hyperv_features, -HYPERV_FEAT_REENLIGHTENMENT ,0 ), -DEFINE_PROP_BIT64 ("hv-tlbflush" ,X86CPU , hyperv_features, -HYPERV_FEAT_TLBFLUSH ,0 ), -DEFINE_PROP_BIT64 ("hv-evmcs" ,X86CPU , hyperv_features, -HYPERV_FEAT_EVMCS ,0 ), -DEFINE_PROP_BIT64 ("hv-ipi" ,X86CPU , hyperv_features, -HYPERV_FEAT_IPI ,0 ), -DEFINE_PROP_BIT64 ("hv-stimer-direct" ,X86CPU , hyperv_features, -HYPERV_FEAT_STIMER_DIRECT ,0 ),DEFINE_PROP_ON_OFF_AUTO ("hv-no-nonarch-coresharing" ,X86CPU , hyperv_no_nonarch_cs,ON_OFF_AUTO_OFF ), DEFINE_PROP_BOOL("hv-passthrough", X86CPU, hyperv_passthrough, false),
@@ -7283,6 +7425,41 @@ static void x86_cpu_common_class_init(ObjectClass *oc, void *data) x86_cpu_get_crash_info_qom, NULL, NULL, NULL);#endif +object_class_property_add_bool (oc,"hv-relaxed" , + x86_hv_relaxed_get, x86_hv_relaxed_set); +object_class_property_add_bool (oc,"hv-vapic" , + x86_hv_vapic_get, x86_hv_vapic_set); +object_class_property_add_bool (oc,"hv-time" , + x86_hv_time_get, x86_hv_time_set); +object_class_property_add_bool (oc,"hv-crash" , + x86_hv_crash_get, x86_hv_crash_set); +object_class_property_add_bool (oc,"hv-reset" , + x86_hv_reset_get, x86_hv_reset_set); +object_class_property_add_bool (oc,"hv-vpindex" , + x86_hv_vpindex_get, x86_hv_vpindex_set); +object_class_property_add_bool (oc,"hv-runtime" , + x86_hv_runtime_get, x86_hv_runtime_set); +object_class_property_add_bool (oc,"hv-synic" , + x86_hv_synic_get, x86_hv_synic_set); +object_class_property_add_bool (oc,"hv-stimer" , + x86_hv_stimer_get, x86_hv_stimer_set); +object_class_property_add_bool (oc,"hv-frequencies" , + x86_hv_frequencies_get, + x86_hv_frequencies_set); +object_class_property_add_bool (oc,"hv-reenlightenment" , + x86_hv_reenlightenment_get, + x86_hv_reenlightenment_set); +object_class_property_add_bool (oc,"hv-tlbflush" , + x86_hv_tlbflush_get, x86_hv_tlbflush_set); +object_class_property_add_bool (oc,"hv-evmcs" , + x86_hv_evmcs_get, + x86_hv_evmcs_set); +object_class_property_add_bool (oc,"hv-ipi" , + x86_hv_ipi_get, x86_hv_ipi_set); +object_class_property_add_bool (oc,"hv-stimer-direct" , + x86_hv_stimer_direct_get, + x86_hv_stimer_direct_set); +for (w =0 ; w <FEATURE_WORDS ; w++) {int bitnr;for (bitnr =0 ; bitnr <64 ; bitnr++) {
target/i386/cpu.h +2 -0
@@ -1667,6 +1667,8 @@ struct X86CPU {char *hyperv_vendor;bool hyperv_synic_kvm_only;uint64_t hyperv_features; +uint64_t hyperv_features_on; +uint64_t hyperv_features_off;bool hyperv_passthrough;OnOffAuto hyperv_no_nonarch_cs;uint32_t hyperv_vendor_id[3 ];
Signature · 1 line
2.29.2
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 7 lines · On Wed, 10 Feb 2021 17:40:28 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Sometimes we'd like to know which features were explicitly enabled and which were explicitly disabled on the command line. E.g. it seems logical to handle 'hv_passthrough,hv_feature=off' as "enable everything supported by the host except for
hv_feature" but this doesn't seem to be possible with the current 'hyperv_features' bit array. Introduce 'hv_features_on'/'hv_features_off' add-ons and track explicit enablement/disablement there.
Quoted 5 lines
Note, it doesn't seem to be possible to fill '
hyperv_features' array during CPU creation time when 'hv-passthrough' is specified and we're running on an older kernel withoutKVM_CAP_SYS_HYPERV_CPUIDsupport. To get the list of the supported Hyper-V features we need to actually create KVM VCPU and this happens much later.
seems to me that we are returning back to +-feat parsing, this time only for hyperv.
I'm not sure I like it back, especially considering we are going to drop "-feat" priority for x86.
now about impossible, see arm/kvm/virt, they create a 'sample' VCPU at KVM init time to probe for some CPU features in advance. You can use similar approach to prepare value for hyperv_features.
Quoted 284 lines
No functional change intended.
Signed-off-by Vitaly Kuznetsov <[email protected]>--- target/i386/cpu.c | 237 ++++++++++++++++++++++++++++++++++++++++------ target/i386/cpu.h | 2 + 2 files changed, 209 insertions(+), 30 deletions(-)target/i386/cpu.c +207
-30@@ -4553,6 +4553,178 @@ static void x86_cpuid_set_tsc_freq(Object *obj, Visitor *v, const char *name, cpu->env.tsc_khz = cpu->env.user_tsc_khz = value /1000 ; } +static bool x86_hv_feature_get (Object *obj,int feature) +{ +X86CPU *cpu =X86_CPU (obj); + +return cpu->hyperv_features &BIT (feature); +} + +static void x86_hv_feature_set (Object *obj,bool value,int feature) +{ +X86CPU *cpu =X86_CPU (obj); + +if (value) { + cpu->hyperv_features |=BIT (feature); + cpu->hyperv_features_on |=BIT (feature); + cpu->hyperv_features_off &= ~BIT (feature); + }else { + cpu->hyperv_features &= ~BIT (feature); + cpu->hyperv_features_on &= ~BIT (feature); + cpu->hyperv_features_off |=BIT (feature); + } +} + +static bool x86_hv_relaxed_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_RELAXED ); +} + +static void x86_hv_relaxed_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_RELAXED ); +} + +static bool x86_hv_vapic_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_VAPIC ); +} + +static void x86_hv_vapic_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_VAPIC ); +} + +static bool x86_hv_time_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_TIME ); +} + +static void x86_hv_time_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_TIME ); +} + +static bool x86_hv_crash_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_CRASH ); +} + +static void x86_hv_crash_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_CRASH ); +} + +static bool x86_hv_reset_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_RESET ); +} + +static void x86_hv_reset_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_RESET ); +} + +static bool x86_hv_vpindex_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_VPINDEX ); +} + +static void x86_hv_vpindex_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_VPINDEX ); +} + +static bool x86_hv_runtime_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_RUNTIME ); +} + +static void x86_hv_runtime_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_RUNTIME ); +} + +static bool x86_hv_synic_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_SYNIC ); +} + +static void x86_hv_synic_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_SYNIC ); +} + +static bool x86_hv_stimer_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_STIMER ); +} + +static void x86_hv_stimer_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_STIMER ); +} + +static bool x86_hv_frequencies_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_FREQUENCIES ); +} + +static void x86_hv_frequencies_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_FREQUENCIES ); +} + +static bool x86_hv_reenlightenment_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_REENLIGHTENMENT ); +} + +static void x86_hv_reenlightenment_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_REENLIGHTENMENT ); +} + +static bool x86_hv_tlbflush_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_TLBFLUSH ); +} + +static void x86_hv_tlbflush_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_TLBFLUSH ); +} + +static bool x86_hv_evmcs_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_EVMCS ); +} + +static void x86_hv_evmcs_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_EVMCS ); +} + +static bool x86_hv_ipi_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_IPI ); +} + +static void x86_hv_ipi_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_IPI ); +} + +static bool x86_hv_stimer_direct_get (Object *obj,Error **errp) +{ +return x86_hv_feature_get (obj,HYPERV_FEAT_STIMER_DIRECT ); +} + +static void x86_hv_stimer_direct_set (Object *obj,bool value,Error **errp) +{ +x86_hv_feature_set (obj, value,HYPERV_FEAT_STIMER_DIRECT ); +} +/* Generic getter for "feature-words" and "filtered-features" properties */ static void x86_cpu_get_feature_words(Object *obj, Visitor *v, const char *name, void *opaque,@@ -7107,36 +7279,6 @@ static Property x86_cpu_properties[] = {DEFINE_PROP_UINT32 ("hv-spinlocks" ,X86CPU , hyperv_spinlock_attempts,HYPERV_SPINLOCK_NEVER_NOTIFY ), -DEFINE_PROP_BIT64 ("hv-relaxed" ,X86CPU , hyperv_features, -HYPERV_FEAT_RELAXED ,0 ), -DEFINE_PROP_BIT64 ("hv-vapic" ,X86CPU , hyperv_features, -HYPERV_FEAT_VAPIC ,0 ), -DEFINE_PROP_BIT64 ("hv-time" ,X86CPU , hyperv_features, -HYPERV_FEAT_TIME ,0 ), -DEFINE_PROP_BIT64 ("hv-crash" ,X86CPU , hyperv_features, -HYPERV_FEAT_CRASH ,0 ), -DEFINE_PROP_BIT64 ("hv-reset" ,X86CPU , hyperv_features, -HYPERV_FEAT_RESET ,0 ), -DEFINE_PROP_BIT64 ("hv-vpindex" ,X86CPU , hyperv_features, -HYPERV_FEAT_VPINDEX ,0 ), -DEFINE_PROP_BIT64 ("hv-runtime" ,X86CPU , hyperv_features, -HYPERV_FEAT_RUNTIME ,0 ), -DEFINE_PROP_BIT64 ("hv-synic" ,X86CPU , hyperv_features, -HYPERV_FEAT_SYNIC ,0 ), -DEFINE_PROP_BIT64 ("hv-stimer" ,X86CPU , hyperv_features, -HYPERV_FEAT_STIMER ,0 ), -DEFINE_PROP_BIT64 ("hv-frequencies" ,X86CPU , hyperv_features, -HYPERV_FEAT_FREQUENCIES ,0 ), -DEFINE_PROP_BIT64 ("hv-reenlightenment" ,X86CPU , hyperv_features, -HYPERV_FEAT_REENLIGHTENMENT ,0 ), -DEFINE_PROP_BIT64 ("hv-tlbflush" ,X86CPU , hyperv_features, -HYPERV_FEAT_TLBFLUSH ,0 ), -DEFINE_PROP_BIT64 ("hv-evmcs" ,X86CPU , hyperv_features, -HYPERV_FEAT_EVMCS ,0 ), -DEFINE_PROP_BIT64 ("hv-ipi" ,X86CPU , hyperv_features, -HYPERV_FEAT_IPI ,0 ), -DEFINE_PROP_BIT64 ("hv-stimer-direct" ,X86CPU , hyperv_features, -HYPERV_FEAT_STIMER_DIRECT ,0 ),DEFINE_PROP_ON_OFF_AUTO ("hv-no-nonarch-coresharing" ,X86CPU , hyperv_no_nonarch_cs,ON_OFF_AUTO_OFF ), DEFINE_PROP_BOOL("hv-passthrough", X86CPU, hyperv_passthrough, false),@@ -7283,6 +7425,41 @@ static void x86_cpu_common_class_init(ObjectClass *oc, void *data) x86_cpu_get_crash_info_qom, NULL, NULL, NULL);#endif +object_class_property_add_bool (oc,"hv-relaxed" , + x86_hv_relaxed_get, x86_hv_relaxed_set); +object_class_property_add_bool (oc,"hv-vapic" , + x86_hv_vapic_get, x86_hv_vapic_set); +object_class_property_add_bool (oc,"hv-time" , + x86_hv_time_get, x86_hv_time_set); +object_class_property_add_bool (oc,"hv-crash" , + x86_hv_crash_get, x86_hv_crash_set); +object_class_property_add_bool (oc,"hv-reset" , + x86_hv_reset_get, x86_hv_reset_set); +object_class_property_add_bool (oc,"hv-vpindex" , + x86_hv_vpindex_get, x86_hv_vpindex_set); +object_class_property_add_bool (oc,"hv-runtime" , + x86_hv_runtime_get, x86_hv_runtime_set); +object_class_property_add_bool (oc,"hv-synic" , + x86_hv_synic_get, x86_hv_synic_set); +object_class_property_add_bool (oc,"hv-stimer" , + x86_hv_stimer_get, x86_hv_stimer_set); +object_class_property_add_bool (oc,"hv-frequencies" , + x86_hv_frequencies_get, + x86_hv_frequencies_set); +object_class_property_add_bool (oc,"hv-reenlightenment" , + x86_hv_reenlightenment_get, + x86_hv_reenlightenment_set); +object_class_property_add_bool (oc,"hv-tlbflush" , + x86_hv_tlbflush_get, x86_hv_tlbflush_set); +object_class_property_add_bool (oc,"hv-evmcs" , + x86_hv_evmcs_get, + x86_hv_evmcs_set); +object_class_property_add_bool (oc,"hv-ipi" , + x86_hv_ipi_get, x86_hv_ipi_set); +object_class_property_add_bool (oc,"hv-stimer-direct" , + x86_hv_stimer_direct_get, + x86_hv_stimer_direct_set); +for (w =0 ; w <FEATURE_WORDS ; w++) {int bitnr;for (bitnr =0 ; bitnr <64 ; bitnr++) {target/i386/cpu.h +2
-0@@ -1667,6 +1667,8 @@ struct X86CPU {char *hyperv_vendor;bool hyperv_synic_kvm_only;uint64_t hyperv_features; +uint64_t hyperv_features_on; +uint64_t hyperv_features_off;bool hyperv_passthrough;OnOffAuto hyperv_no_nonarch_cs;uint32_t hyperv_vendor_id[3 ];
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 16 lines · Igor Mammedov
Quoted 7 lines · On Wed, 10 Feb 2021 17:40:28 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Sometimes we'd like to know which features were explicitly enabled and which were explicitly disabled on the command line. E.g. it seems logical to handle 'hv_passthrough,hv_feature=off' as "enable everything supported by the host except for
hv_feature" but this doesn't seem to be possible with the current 'hyperv_features' bit array. Introduce 'hv_features_on'/'hv_features_off' add-ons and track explicit enablement/disablement there.Quoted 5 lines
Note, it doesn't seem to be possible to fill '
hyperv_features' array during CPU creation time when 'hv-passthrough' is specified and we're running on an older kernel withoutKVM_CAP_SYS_HYPERV_CPUIDsupport. To get the list of the supported Hyper-V features we need to actually create KVM VCPU and this happens much later.
Quoted 9 lines
seems to me that we are returning back to +-feat parsing, this time only for hyperv.
I'm not sure I like it back, especially considering we are going to drop "-feat" priority for x86.now about impossible, see arm/kvm/virt, they create a 'sample' VCPU at KVM init time to probe for some CPU features in advance. You can use similar approach to prepare value for
hyperv_features.
KVM_CAP_SYS_HYPERV_CPUID is supported since 5.11 and eventually it'll make it to all kernels we care about so I'd really like to avoid any 'sample' CPUs for the time being. On/off parsing looks like a much lesser evil.
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 28 lines · On Fri, 12 Feb 2021 09:45:52 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 16 lines · Igor Mammedov
Quoted 7 lines · On Wed, 10 Feb 2021 17:40:28 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Sometimes we'd like to know which features were explicitly enabled and which were explicitly disabled on the command line. E.g. it seems logical to handle 'hv_passthrough,hv_feature=off' as "enable everything supported by the host except for
hv_feature" but this doesn't seem to be possible with the current 'hyperv_features' bit array. Introduce 'hv_features_on'/'hv_features_off' add-ons and track explicit enablement/disablement there.Quoted 5 lines
Note, it doesn't seem to be possible to fill '
hyperv_features' array during CPU creation time when 'hv-passthrough' is specified and we're running on an older kernel withoutKVM_CAP_SYS_HYPERV_CPUIDsupport. To get the list of the supported Hyper-V features we need to actually create KVM VCPU and this happens much later.Quoted 9 lines
seems to me that we are returning back to +-feat parsing, this time only for hyperv.
I'm not sure I like it back, especially considering we are going to drop "-feat" priority for x86.now about impossible, see arm/kvm/virt, they create a 'sample' VCPU at KVM init time to probe for some CPU features in advance. You can use similar approach to prepare value for
hyperv_features.
Quoted 4 lines
KVM_CAP_SYS_HYPERV_CPUIDis supported since 5.11 and eventually it'll make it to all kernels we care about so I'd really like to avoid any 'sample' CPUs for the time being. On/off parsing looks like a much lesser evil.
When minimum supported by QEMU kernel version gets there, you can remove scratch CPU in QEMU (if hyperv will remain its sole user).
writing your own property parser like in this series, is possible too but it adds extra fields to track state and hard to follow logic. On top it adds a lot of churn by switching hv_ features to dynamic properties, which is not necessary if scratch CPU approach is used.
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.
in proto would look like:
* kvm_init:
hv_passthrough_cached = scratch_cpu->hyperv_features
* property parsing time:
x86_hv_passthrough_set()
cpu->hyperv_features = hv_passthrough_cached
all other features handled by generic property parsing, you don't have to do any special handling for them.
* cpu_relalize()
hv_expand() to check for dependencies, conflicts availability of features.
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 43 lines · Igor Mammedov
Quoted 28 lines · On Fri, 12 Feb 2021 09:45:52 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 16 lines · Igor Mammedov
Quoted 7 lines · On Wed, 10 Feb 2021 17:40:28 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Sometimes we'd like to know which features were explicitly enabled and which were explicitly disabled on the command line. E.g. it seems logical to handle 'hv_passthrough,hv_feature=off' as "enable everything supported by the host except for
hv_feature" but this doesn't seem to be possible with the current 'hyperv_features' bit array. Introduce 'hv_features_on'/'hv_features_off' add-ons and track explicit enablement/disablement there.Quoted 5 lines
Note, it doesn't seem to be possible to fill '
hyperv_features' array during CPU creation time when 'hv-passthrough' is specified and we're running on an older kernel withoutKVM_CAP_SYS_HYPERV_CPUIDsupport. To get the list of the supported Hyper-V features we need to actually create KVM VCPU and this happens much later.Quoted 9 lines
seems to me that we are returning back to +-feat parsing, this time only for hyperv.
I'm not sure I like it back, especially considering we are going to drop "-feat" priority for x86.now about impossible, see arm/kvm/virt, they create a 'sample' VCPU at KVM init time to probe for some CPU features in advance. You can use similar approach to prepare value for
hyperv_features.Quoted 4 lines
KVM_CAP_SYS_HYPERV_CPUIDis supported since 5.11 and eventually it'll make it to all kernels we care about so I'd really like to avoid any 'sample' CPUs for the time being. On/off parsing looks like a much lesser evil.When minimum supported by QEMU kernel version gets there, you can remove scratch CPU in QEMU (if hyperv will remain its sole user).
writing your own property parser like in this series, is possible too but it adds extra fields to track state and hard to follow logic. On top it adds a lot of churn by switching
hv_features to dynamic properties, which is not necessary if scratch CPU approach is used.
Quoted 4 lines
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.
Even if I do that (and I serioulsy doubt it's going to be easier than just adding two 'u64's, kvm_arm_get_host_cpu_features() alone is 200 lines long) this is not going to give us what we need to distinguish between
'hv-passthrough,hv-evmcs'
and
'hv-passthrough'
when 'hv-evmcs' *is* supported by the host. When guest CPU lacks VMX we don't want to enable it unless it was requested explicitly (former but not the later).
Moreover, instead of just adding two 'u64's we're now doing an ioctl which can fail, be subject to limits,... Creating and destroying a CPU is also slow. Sorry, I hardly see how this is better, maybe just from 'code purity' point of view.
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 15 lines · Vitaly Kuznetsov
Quoted 5 lines · Igor Mammedov
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.Even if I do that (and I serioulsy doubt it's going to be easier than just adding two 'u64's,
kvm_arm_get_host_cpu_features()alone is 200 lines long) this is not going to give us what we need to distinguish between'hv-passthrough,hv-evmcs'
Quoted 7 lines
and
'hv-passthrough'
when 'hv-evmcs' *is* supported by the host. When guest CPU lacks VMX we don't want to enable it unless it was requested explicitly (former but not the later).
... and if for whatever reason we decide that this is also bad/not needed, I can just drop patches 16-18 from the series (leaving 'hv-passthrough,hv-feature=off' problem to better times).
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 21 lines · On Fri, 12 Feb 2021 16:26:03 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 19 lines · Vitaly Kuznetsov
Quoted 5 lines · Igor Mammedov
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.Even if I do that (and I serioulsy doubt it's going to be easier than just adding two 'u64's,
kvm_arm_get_host_cpu_features()alone is 200 lines long) this is not going to give us what we need to distinguish between'hv-passthrough,hv-evmcs'
and
'hv-passthrough'
Quoted 7 lines
Quoted 3 lines
when 'hv-evmcs' *is* supported by the host. When guest CPU lacks VMX we don't want to enable it unless it was requested explicitly (former but not the later).
... and if for whatever reason we decide that this is also bad/not needed, I can just drop patches 16-18 from the series (leaving 'hv-passthrough,hv-feature=off' problem to better times).
that's also an option,
we would need to make sure that hv-passthrough is mutually exclusive with ''all'' other hv- properties to avoid above combination being ever (mis)used.
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 28 lines · Igor Mammedov
Quoted 25 lines · On Fri, 12 Feb 2021 16:26:03 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 22 lines · Vitaly Kuznetsov
Quoted 5 lines · Igor Mammedov
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.Even if I do that (and I serioulsy doubt it's going to be easier than just adding two 'u64's,
kvm_arm_get_host_cpu_features()alone is 200 lines long) this is not going to give us what we need to distinguish between'hv-passthrough,hv-evmcs'
and
'hv-passthrough'
when 'hv-evmcs' *is* supported by the host. When guest CPU lacks VMX we don't want to enable it unless it was requested explicitly (former but not the later).
Quoted 7 lines
Quoted 3 lines
... and if for whatever reason we decide that this is also bad/not needed, I can just drop patches 16-18 from the series (leaving 'hv-passthrough,hv-feature=off' problem to better times).
that's also an option,
we would need to make sure that hv-passthrough is mutually exclusive with ''all'' other hv- properties to avoid above combination being ever (mis)used.
That's an option to finally get these patches merged, not a good option for end users.
'hv-passthrough,hv-feature' works today and it's useful. Should we drop it?
'hv-passthrough/hv-default' and 'hv-passthrough/hv-default,hv-evmcs' should give us sane results.
'hv-passthrough,hv-feature=off' is convenient.
Why droppping this all? To save 9 (nine) lines of code in the parser?
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 38 lines · On Mon, 15 Feb 2021 09:56:19 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 28 lines · Igor Mammedov
Quoted 25 lines · On Fri, 12 Feb 2021 16:26:03 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 22 lines · Vitaly Kuznetsov
Quoted 5 lines · Igor Mammedov
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.Even if I do that (and I serioulsy doubt it's going to be easier than just adding two 'u64's,
kvm_arm_get_host_cpu_features()alone is 200 lines long) this is not going to give us what we need to distinguish between'hv-passthrough,hv-evmcs'
and
'hv-passthrough'
when 'hv-evmcs' *is* supported by the host. When guest CPU lacks VMX we don't want to enable it unless it was requested explicitly (former but not the later).
Quoted 7 lines
Quoted 3 lines
... and if for whatever reason we decide that this is also bad/not needed, I can just drop patches 16-18 from the series (leaving 'hv-passthrough,hv-feature=off' problem to better times).
that's also an option,
we would need to make sure that hv-passthrough is mutually exclusive with ''all'' other hv- properties to avoid above combination being ever (mis)used.
Quoted 5 lines
That's an option to finally get these patches merged, not a good option for end users.
'hv-passthrough,hv-feature' works today and it's useful. Should we drop it?
well,
try suggested idea about using scratch CPU and it might get merged sooner. (it's not like I'm suggesting you to rewrite half of QEMU, just some of patches, which most likely would simplify series from my point of view and would be easier to maintain)
Quoted 7 lines
'hv-passthrough/hv-default' and 'hv-passthrough/hv-default,hv-evmcs' should give us sane results.
'hv-passthrough,hv-feature=off' is convenient.
Why droppping this all? To save 9 (nine) lines of code in the parser?
it's doing what generic property parsing is capable off, provided you fish out hv-passthrough value in advance like arm/virt does (I think ppc does similar hing also), so I consider it as unnecessary code duplication/ complication and maintenance burden.
If it were a hotfix during hard-freeze may be I'd agree (with promise to rework it later to something more palatable), but it's not, for patches in state they are now I'm not confident enough to ACK them.
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 12 lines · On Mon, 15 Feb 2021 16:55:02 +0100 Igor Mammedov <[email protected]> wrote:
Quoted 9 lines · On Mon, 15 Feb 2021 09:56:19 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 7 lines · Igor Mammedov
Quoted 4 lines · On Fri, 12 Feb 2021 16:26:03 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 2 lines · Vitaly Kuznetsov
Igor Mammedov <[email protected]> writes:
[...]
Quoted 1 line
Quoted 1 line
(I think ppc does similar hing also)
well scratch that off, I can't find PPC part anymore. Maybe I've confused that with something else.
[...]
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
In reply to Igor MammedovQuoted 44 lines · Igor Mammedov
Quoted 41 lines · On Mon, 15 Feb 2021 09:56:19 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 28 lines · Igor Mammedov
Quoted 25 lines · On Fri, 12 Feb 2021 16:26:03 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 22 lines · Vitaly Kuznetsov
Quoted 5 lines · Igor Mammedov
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.Even if I do that (and I serioulsy doubt it's going to be easier than just adding two 'u64's,
kvm_arm_get_host_cpu_features()alone is 200 lines long) this is not going to give us what we need to distinguish between'hv-passthrough,hv-evmcs'
and
'hv-passthrough'
when 'hv-evmcs' *is* supported by the host. When guest CPU lacks VMX we don't want to enable it unless it was requested explicitly (former but not the later).
Quoted 7 lines
Quoted 3 lines
... and if for whatever reason we decide that this is also bad/not needed, I can just drop patches 16-18 from the series (leaving 'hv-passthrough,hv-feature=off' problem to better times).
that's also an option,
we would need to make sure that hv-passthrough is mutually exclusive with ''all'' other hv- properties to avoid above combination being ever (mis)used.That's an option to finally get these patches merged, not a good option for end users.
Quoted 8 lines
Quoted 2 lines
'hv-passthrough,hv-feature' works today and it's useful. Should we drop it?
well,
try suggested idea about using scratch CPU and it might get merged sooner. (it's not like I'm suggesting you to rewrite half of QEMU, just some of patches, which most likely would simplify series from my point of view and would be easier to maintain)
I don't see anything in the series which will go away if I implement this idea but as I hate it deerly I'm likely not going to.
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
In reply to Vitaly KuznetsovQuoted 45 lines · On Fri, 12 Feb 2021 16:19:24 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 43 lines · Igor Mammedov
Quoted 28 lines · On Fri, 12 Feb 2021 09:45:52 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 16 lines · Igor Mammedov
Quoted 7 lines · On Wed, 10 Feb 2021 17:40:28 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Sometimes we'd like to know which features were explicitly enabled and which were explicitly disabled on the command line. E.g. it seems logical to handle 'hv_passthrough,hv_feature=off' as "enable everything supported by the host except for
hv_feature" but this doesn't seem to be possible with the current 'hyperv_features' bit array. Introduce 'hv_features_on'/'hv_features_off' add-ons and track explicit enablement/disablement there.Quoted 5 lines
Note, it doesn't seem to be possible to fill '
hyperv_features' array during CPU creation time when 'hv-passthrough' is specified and we're running on an older kernel withoutKVM_CAP_SYS_HYPERV_CPUIDsupport. To get the list of the supported Hyper-V features we need to actually create KVM VCPU and this happens much later.Quoted 9 lines
seems to me that we are returning back to +-feat parsing, this time only for hyperv.
I'm not sure I like it back, especially considering we are going to drop "-feat" priority for x86.now about impossible, see arm/kvm/virt, they create a 'sample' VCPU at KVM init time to probe for some CPU features in advance. You can use similar approach to prepare value for
hyperv_features.Quoted 4 lines
KVM_CAP_SYS_HYPERV_CPUIDis supported since 5.11 and eventually it'll make it to all kernels we care about so I'd really like to avoid any 'sample' CPUs for the time being. On/off parsing looks like a much lesser evil.When minimum supported by QEMU kernel version gets there, you can remove scratch CPU in QEMU (if hyperv will remain its sole user).
writing your own property parser like in this series, is possible too but it adds extra fields to track state and hard to follow logic. On top it adds a lot of churn by switching
hv_features to dynamic properties, which is not necessary if scratch CPU approach is used.
Quoted 7 lines
Quoted 4 lines
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.Even if I do that (and I serioulsy doubt it's going to be easier than just adding two 'u64's,
kvm_arm_get_host_cpu_features()alone is 200
it does a lot more then what you need, kvm_arm_create_scratch_host_vcpu() which it uses will do the job and even that could be made smaller for hv usecase.
Quoted 5 lines
lines long) this is not going to give us what we need to distinguish between
'hv-passthrough,hv-evmcs'
Quoted 7 lines
and
'hv-passthrough'
when 'hv-evmcs' *is* supported by the host. When guest CPU lacks VMX we don't want to enable it unless it was requested explicitly (former but not the later).
could you elaborate more on it, i.e. why do we need to distinguish and why do we need evmcs without VMX if user asked for it (will it be usable)
Quoted 4 lines
Moreover, instead of just adding two 'u64's we're now doing an ioctl which can fail, be subject to limits,... Creating and destroying a CPU is also slow. Sorry, I hardly see how this is better, maybe just from 'code purity' point of view.
readable and easy to maintain code is not a thing to neglect.
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 19 lines · Igor Mammedov
Quoted 8 lines
Quoted 5 lines
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.Even if I do that (and I serioulsy doubt it's going to be easier than just adding two 'u64's,
kvm_arm_get_host_cpu_features()alone is 200it does a lot more then what you need,
kvm_arm_create_scratch_host_vcpu()which it uses will do the job and even that could be made smaller for hv usecase.Quoted 7 lines
lines long) this is not going to give us what we need to distinguish between
'hv-passthrough,hv-evmcs'
and
Quoted 8 lines
Quoted 5 lines
'hv-passthrough'
when 'hv-evmcs' *is* supported by the host. When guest CPU lacks VMX we don't want to enable it unless it was requested explicitly (former but not the later).
could you elaborate more on it, i.e. why do we need to distinguish and why do we need evmcs without VMX if user asked for it (will it be usable)
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it, it comes with nesting (-ExposeVirtualizationExtensions $true). When we create a default set of Hyper-V enlightenments (either 'hv-default' or 'hv-passthrough') we should be as close as possible to genuine Hyper-V to not create unsupported Frankenstiens which can break with any Windows update (because nobody tested these configurations). That bein said, if guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should
'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
Quoted 5 lines
Quoted 4 lines
Moreover, instead of just adding two 'u64's we're now doing an ioctl which can fail, be subject to limits,... Creating and destroying a CPU is also slow. Sorry, I hardly see how this is better, maybe just from 'code purity' point of view.
readable and easy to maintain code is not a thing to neglect.
Of couse, but 'scratch CPU' idea is not a good design decision, it is an ugly hack we should get rid of in ARM land, not try bringing it to other architectures. Generally, KVM should allow to query all its capabilities without the need to create a vCPU or, if not possible, we should create 'real' QEMU VCPUs and use one/all of the to query capabilities, avoiding 'scratch' because:
- Creating and destroying a vCPU makes VM startup slower, much slower. E.g. for a single-CPU VM you're doubling the time required to create vCPUs!
- vCPUs in KVM are quite memory consuming. Just 'struct kvm_vcpu_arch' was something like 12kb last time I looked at it.
I have no clue why scratch vCPUs were implemented on ARM, however, I'd very much want us to avoid doing the same on x86. We do have use-cases where startup time and consumed memory is important. There is a point in limiting ioctls for security reasons (e.g. if I'm creating a single vCPU VM I may want to limit userspace process to one and only one KVM_CREATE_VCPU call).
Now to the code you complain about. The 'hard to read and maintain' code is literaly this:
+static void x86_hv_feature_set(Object *obj, bool value, int feature) +{
+ X86CPU *cpu = X86_CPU(obj);
+
+ if (value) {
+ cpu->hyperv_features |= BIT(feature);
+ cpu->hyperv_features_on |= BIT(feature);
+ cpu->hyperv_features_off &= ~BIT(feature);
+ } else {
+ cpu->hyperv_features &= ~BIT(feature);
+ cpu->hyperv_features_on &= ~BIT(feature);
+ cpu->hyperv_features_off |= BIT(feature);+ }
+}
I can add as many comments here as needed, however, I don't see what requires additional explanaition. We just want to know two things:
- What's the 'effective' setting of the control
- Was it explicitly enabled or disabled on the command line.
Custom parsers are not new in QEMU and they're not going anywhere I believe. There are options with simple enablent and there are some with additional considerations. Trying to make CPU objects somewhat 'special' by forcing all options to be of type-1 (and thus crippling user experience) is not the way to go IMHO. I'd very much like us to go in another direction, make our option parser better so my very simple use-case is covered 'out-of-the-box'.
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 1 line · Vitaly Kuznetsov
I have no clue why scratch vCPUs were implemented on ARM, however, I'd
We don't have an ioctl like KVM_GET_SUPPORTED_CPUID, which operates on the KVM fd. Perhaps we should.
Thanks,
drew
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
In reply to Vitaly KuznetsovQuoted 34 lines · On Mon, 15 Feb 2021 09:53:50 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 19 lines · Igor Mammedov
Quoted 8 lines
Quoted 5 lines
Please try reusing scratch CPU approach, see
kvm_arm_get_host_cpu_features()
for an example. You will very likely end up with simpler series, compared to reinventing wheel.Even if I do that (and I serioulsy doubt it's going to be easier than just adding two 'u64's,
kvm_arm_get_host_cpu_features()alone is 200it does a lot more then what you need,
kvm_arm_create_scratch_host_vcpu()which it uses will do the job and even that could be made smaller for hv usecase.Quoted 7 lines
lines long) this is not going to give us what we need to distinguish between
'hv-passthrough,hv-evmcs'
and
Quoted 8 lines
Quoted 5 lines
'hv-passthrough'
when 'hv-evmcs' *is* supported by the host. When guest CPU lacks VMX we don't want to enable it unless it was requested explicitly (former but not the later).
could you elaborate more on it, i.e. why do we need to distinguish and why do we need evmcs without VMX if user asked for it (will it be usable)
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
if later on we find usecase for VMX=off + hv-evmcs=on, we will be able to drop error without affecting existing users, but not other way around.
Quoted 18 lines
Quoted 5 lines
Quoted 4 lines
Moreover, instead of just adding two 'u64's we're now doing an ioctl which can fail, be subject to limits,... Creating and destroying a CPU is also slow. Sorry, I hardly see how this is better, maybe just from 'code purity' point of view.
readable and easy to maintain code is not a thing to neglect.
Of couse, but 'scratch CPU' idea is not a good design decision, it is an ugly hack we should get rid of in ARM land, not try bringing it to other architectures. Generally, KVM should allow to query all its capabilities without the need to create a vCPU or, if not possible, we should create 'real' QEMU VCPUs and use one/all of the to query capabilities, avoiding 'scratch' because:
- Creating and destroying a vCPU makes VM startup slower, much slower. E.g. for a single-CPU VM you're doubling the time required to create vCPUs!
- vCPUs in KVM are quite memory consuming. Just 'structkvm_vcpu_arch' was something like 12kb last time I looked at it.
Quoted 6 lines
I have no clue why scratch vCPUs were implemented on ARM, however, I'd very much want us to avoid doing the same on x86. We do have use-cases where startup time and consumed memory is important. There is a point in limiting ioctls for security reasons (e.g. if I'm creating a single vCPU VM I may want to limit userspace process to one and only one
KVM_CREATE_VCPUcall).
it should be possible to reuse scratch VCPU (kvm file descriptor) as the first CPU of VM, if there is a will/need, without creating unnecessary overhead. I don't like scratch CPU either but from my pov it's a lesser evil to spawning custom parser every time someone fills like it.
Quoted 7 lines
Now to the code you complain about. The 'hard to read and maintain' code is literaly this:
(quoted diff) +4
-0+static void x86_hv_feature_set(Object *obj, bool value, int feature) +{ + X86CPU *cpu = X86_CPU(obj); +
Review
Quoted 10 lines
(quoted diff) +10
-0+ if (value) { + cpu->hyperv_features |= BIT(feature); + cpu->hyperv_features_on |= BIT(feature); + cpu->hyperv_features_off &= ~BIT(feature); + } else { + cpu->hyperv_features &= ~BIT(feature); + cpu->hyperv_features_on &= ~BIT(feature); + cpu->hyperv_features_off |= BIT(feature); + } +}
It's not just that code but the rest that uses above variables to get final hyperv_features feature set. There is a lot of invariants that are hidden in hv specific code that you put in hyperv kvm specific part.
btw why can't we get supported hyperv_features in passthrough mode during time we initialize KVM (without a vCPU)?
Quoted 5 lines
I can add as many comments here as needed, however, I don't see what requires additional explanaition. We just want to know two things:
- What's the 'effective' setting of the control
- Was it explicitly enabled or disabled on the command line.
Quoted 7 lines
Custom parsers are not new in QEMU and they're not going anywhere I believe. There are options with simple enablent and there are some with additional considerations. Trying to make CPU objects somewhat 'special' by forcing all options to be of type-1 (and thus crippling user experience) is not the way to go IMHO. I'd very much like us to go in another direction, make our option parser better so my very simple use-case is covered 'out-of-the-box'.
there is a lot of effort spent on getting rid of custom parsers that QEMU accumulated over years. Probably there were good reasons to add them back then, and now someone else has to spend time to clean them up.
hyperv case is not any special in that regard (at least I'm not convinced at this point). Try alternative(s) first, if that doesn't work out, then custom parser might be necessary.
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user. Ok, you can replace 'hyperv_features_on' with 'evmcs_was_explicitly_requested' but how is it better?
Quoted 29 lines
if later on we find usecase for VMX=off + hv-evmcs=on, we will be able to drop error without affecting existing users, but not other way around.
Quoted 18 lines
Quoted 5 lines
Quoted 4 lines
Moreover, instead of just adding two 'u64's we're now doing an ioctl which can fail, be subject to limits,... Creating and destroying a CPU is also slow. Sorry, I hardly see how this is better, maybe just from 'code purity' point of view.
readable and easy to maintain code is not a thing to neglect.
Of couse, but 'scratch CPU' idea is not a good design decision, it is an ugly hack we should get rid of in ARM land, not try bringing it to other architectures. Generally, KVM should allow to query all its capabilities without the need to create a vCPU or, if not possible, we should create 'real' QEMU VCPUs and use one/all of the to query capabilities, avoiding 'scratch' because:
- Creating and destroying a vCPU makes VM startup slower, much slower. E.g. for a single-CPU VM you're doubling the time required to create vCPUs!
- vCPUs in KVM are quite memory consuming. Just 'structkvm_vcpu_arch' was something like 12kb last time I looked at it.Quoted 6 lines
I have no clue why scratch vCPUs were implemented on ARM, however, I'd very much want us to avoid doing the same on x86. We do have use-cases where startup time and consumed memory is important. There is a point in limiting ioctls for security reasons (e.g. if I'm creating a single vCPU VM I may want to limit userspace process to one and only one
KVM_CREATE_VCPUcall).
Quoted 4 lines
it should be possible to reuse scratch VCPU (kvm file descriptor) as the first CPU of VM, if there is a will/need, without creating unnecessary overhead. I don't like scratch CPU either but from my pov it's a lesser evil to spawning custom parser every time someone fills like it.
I respectfully disagree.
Quoted 15 lines
Quoted 15 lines
Now to the code you complain about. The 'hard to read and maintain' code is literaly this:
(quoted diff) +10
-0+static void x86_hv_feature_set(Object *obj, bool value, int feature) +{ + X86CPU *cpu = X86_CPU(obj); + + if (value) { + cpu->hyperv_features |= BIT(feature); + cpu->hyperv_features_on |= BIT(feature); + cpu->hyperv_features_off &= ~BIT(feature); + } else { + cpu->hyperv_features &= ~BIT(feature);
Quoted 8 lines
Review
Quoted 4 lines
(quoted diff) +4
-0+ cpu->hyperv_features_on &= ~BIT(feature); + cpu->hyperv_features_off |= BIT(feature); + } +}It's not just that code but the rest that uses above variables to get final
hyperv_featuresfeature set. There is a lot of invariants that are hidden in hv specific code that you put in hyperv kvm specific part.
Could you give an example please?
Quoted 3 lines
btw why can't we get supported
hyperv_featuresin passthrough mode during time we initialize KVM (without a vCPU)?
I think I already explained that: KVM_GET_SUPPORTED_HV_CPUID works on KVM fd from 5.11, it requires a vCPU prior to that.
Quoted 13 lines
Quoted 6 lines
I can add as many comments here as needed, however, I don't see what requires additional explanaition. We just want to know two things:
- What's the 'effective' setting of the control
- Was it explicitly enabled or disabled on the command line.Quoted 7 lines
Custom parsers are not new in QEMU and they're not going anywhere I believe. There are options with simple enablent and there are some with additional considerations. Trying to make CPU objects somewhat 'special' by forcing all options to be of type-1 (and thus crippling user experience) is not the way to go IMHO. I'd very much like us to go in another direction, make our option parser better so my very simple use-case is covered 'out-of-the-box'.
Quoted 7 lines
there is a lot of effort spent on getting rid of custom parsers that QEMU accumulated over years. Probably there were good reasons to add them back then, and now someone else has to spend time to clean them up.
hyperv case is not any special in that regard (at least I'm not convinced at this point). Try alternative(s) first, if that doesn't work out, then custom parser might be necessary.
Please explain to me how
'hv-passthrough,hv-evmcs,-vmx' is going to throw an error and
'hv-passthrough,-vmx' is not.
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 25 lines · Vitaly Kuznetsov
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
Quoted 6 lines
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user.
Igor,
could you please comment on the above? In case my line of thought is correct, and it is impossible to distinguish between e.g.
'hv-passthrough,hv-evmcs,-vmx'
and
'hv-passthrough,-vmx'
without a custom parser (written just exactly the way I did in this version, for example) regardless of when 'hv-passthrough' is expanded. E.g. we have the exact same problem with 'hv-default,hv-evmcs,-vmx'. I that case I see no point in discussing 'scratch CPUs' idea at this point because it is not going to change anything at all ('hv_features_on' will stay, custom parsers will stay).
Am I missing something?
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 39 lines · On Mon, 22 Feb 2021 11:20:34 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 25 lines · Vitaly Kuznetsov
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
Quoted 6 lines
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user.
Igor,
could you please comment on the above? In case my line of thought is correct, and it is impossible to distinguish between e.g.
Quoted 8 lines
'hv-passthrough,hv-evmcs,-vmx'
and
'hv-passthrough,-vmx'without a custom parser (written just exactly the way I did in this version, for example) regardless of when 'hv-passthrough' is expanded. E.g. we have the exact same problem with 'hv-default,hv-evmcs,-vmx'. I that case I see no point in discussing
right, if we need to distinguish between explicit and implicit hv-evmcs set by hv-passthrough custom parser probably the way to go.
However do we need actually need to do it?
I'd treat 'hv-passthrough,-vmx' the same way as 'hv-passthrough,hv-evmcs,-vmx' and it applies not only hv-evmcs but other features hv-passthrough might set (i.e. if whatever was [un]set by hv-passthrough in combination with other features results in invalid config, QEMU shall error out instead of magically altering host provided hv-passthrough value).
something like:
'hv-passthrough,-vmx' when hv-passthrough makes hv-evmcs bit set
should result in
error_setg(errp,"'vmx' feature can't be disabled when hv-evmcs is enabled,"
" either enable 'vmx' or disable 'hv-evmcs' along with disabling 'vmx'"
making host's features set, *magically* mutable, depending on other user provided features is a bit confusing. One would never know what hv-passthrough actually means, and if enabling/disabling 'random' feature changes it.
It's cleaner to do just what user asked (whether implicitly or explicitly) and error out in case it ends up in nonsense configuration.
Quoted 5 lines
'scratch CPUs' idea at this point because it is not going to change anything at all ('
hv_features_on' will stay, custom parsers will stay).gAm I missing something?
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 51 lines · Igor Mammedov
Quoted 39 lines · On Mon, 22 Feb 2021 11:20:34 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 25 lines · Vitaly Kuznetsov
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
Quoted 6 lines
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user.
Igor,
could you please comment on the above? In case my line of thought is correct, and it is impossible to distinguish between e.g.
Quoted 8 lines
'hv-passthrough,hv-evmcs,-vmx'
and
'hv-passthrough,-vmx'without a custom parser (written just exactly the way I did in this version, for example) regardless of when 'hv-passthrough' is expanded. E.g. we have the exact same problem with 'hv-default,hv-evmcs,-vmx'. I that case I see no point in discussing
Quoted 4 lines
right, if we need to distinguish between explicit and implicit hv-evmcs set by hv-passthrough custom parser probably the way to go.
However do we need actually need to do it?
I think we really need that. See below ...
Quoted 12 lines
I'd treat 'hv-passthrough,-vmx' the same way as 'hv-passthrough,hv-evmcs,-vmx' and it applies not only hv-evmcs but other features hv-passthrough might set (i.e. if whatever was [un]set by hv-passthrough in combination with other features results in invalid config, QEMU shall error out instead of magically altering host provided hv-passthrough value).
something like:
'hv-passthrough,-vmx' when hv-passthrough makes hv-evmcs bit set
should result in
error_setg(errp,"'vmx' feature can't be disabled when hv-evmcs is enabled,"
" either enable 'vmx' or disable 'hv-evmcs' along with disabling 'vmx'"
Quoted 7 lines
making host's features set, *magically* mutable, depending on other user provided features is a bit confusing. One would never know what hv-passthrough actually means, and if enabling/disabling 'random' feature changes it.
It's cleaner to do just what user asked (whether implicitly or explicitly) and error out in case it ends up in nonsense configuration.
I don't seem to agree this is a sane behavior, especially if you replace 'hv-passthrough' with 'hv-default' above. Removing 'vmx' from CPU for Windows guests is common if you'd want to avoid nested configuration:
even without any Hyper-V guests created, Windows itself is a Hyper-V partition.
So a sane user will do:
'-cpu host,hv-default,vmx=off'
and on Intel he will get an error, and on AMD he won't.
So what you're suggesting actually defeats the whole purpose of 'hv-default' as upper-layer tools (think libvirt) will need to know that Intel configurations for Windows guests are somewhat different. They'll need to know what 'hv-evmcs' is. We're back to where we've started.
If we are to follow this approach let's just throw away 'hv-evmcs' from 'hv-default' set, it's going to be much cleaner. But again, I don't really believe it's the right way to go.
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 86 lines · On Tue, 23 Feb 2021 16:46:50 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 51 lines · Igor Mammedov
Quoted 39 lines · On Mon, 22 Feb 2021 11:20:34 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 25 lines · Vitaly Kuznetsov
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
Quoted 6 lines
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user.
Igor,
could you please comment on the above? In case my line of thought is correct, and it is impossible to distinguish between e.g.
Quoted 8 lines
'hv-passthrough,hv-evmcs,-vmx'
and
'hv-passthrough,-vmx'without a custom parser (written just exactly the way I did in this version, for example) regardless of when 'hv-passthrough' is expanded. E.g. we have the exact same problem with 'hv-default,hv-evmcs,-vmx'. I that case I see no point in discussing
Quoted 4 lines
right, if we need to distinguish between explicit and implicit hv-evmcs set by hv-passthrough custom parser probably the way to go.
However do we need actually need to do it?
I think we really need that. See below ...
Quoted 12 lines
I'd treat 'hv-passthrough,-vmx' the same way as 'hv-passthrough,hv-evmcs,-vmx' and it applies not only hv-evmcs but other features hv-passthrough might set (i.e. if whatever was [un]set by hv-passthrough in combination with other features results in invalid config, QEMU shall error out instead of magically altering host provided hv-passthrough value).
something like:
'hv-passthrough,-vmx' when hv-passthrough makes hv-evmcs bit set
should result in
error_setg(errp,"'vmx' feature can't be disabled when hv-evmcs is enabled,"
" either enable 'vmx' or disable 'hv-evmcs' along with disabling 'vmx'"Quoted 7 lines
making host's features set, *magically* mutable, depending on other user provided features is a bit confusing. One would never know what hv-passthrough actually means, and if enabling/disabling 'random' feature changes it.
It's cleaner to do just what user asked (whether implicitly or explicitly) and error out in case it ends up in nonsense configuration.
I don't seem to agree this is a sane behavior, especially if you replace 'hv-passthrough' with 'hv-default' above. Removing 'vmx' from CPU for Windows guests is common if you'd want to avoid nested configuration:
even without any Hyper-V guests created, Windows itself is a Hyper-V partition.
Quoted 8 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
and on Intel he will get an error, and on AMD he won't.
So what you're suggesting actually defeats the whole purpose of 'hv-default' as upper-layer tools (think libvirt) will need to know that
I'd assume it would be hard for libvirt to use 'hv-default' from migration point of view. It's semi opaque (one can find out what features it sets indirectly inspecting individual hv_foo features, and mgmt will need to know about them). If it will mutate when other features [un]set, upper layers might need to enumerate all these permutations to know which hosts are compatible or compare host feature sets every time before attempting migration.
Quoted 2 lines
Intel configurations for Windows guests are somewhat different. They'll need to know what 'hv-evmcs' is. We're back to where we've started.
we were talking about hv-passthrough, and if host advertises hv-evmcs QEMU should complain if user disabled features it depends on ( not silently fixing up configuration error).
But the same applies to hv-default.
Quoted 3 lines
If we are to follow this approach let's just throw away 'hv-evmcs' from 'hv-default' set, it's going to be much cleaner. But again, I don't really believe it's the right way to go.
if desired behavior, on Intel host for above config, to start without error then indeed defaults should not set 'hv-evmcs' if it results in invalid feature set.
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 97 lines · Igor Mammedov
Quoted 86 lines · On Tue, 23 Feb 2021 16:46:50 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 51 lines · Igor Mammedov
Quoted 39 lines · On Mon, 22 Feb 2021 11:20:34 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 25 lines · Vitaly Kuznetsov
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
Quoted 6 lines
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user.
Igor,
could you please comment on the above? In case my line of thought is correct, and it is impossible to distinguish between e.g.
Quoted 8 lines
'hv-passthrough,hv-evmcs,-vmx'
and
'hv-passthrough,-vmx'without a custom parser (written just exactly the way I did in this version, for example) regardless of when 'hv-passthrough' is expanded. E.g. we have the exact same problem with 'hv-default,hv-evmcs,-vmx'. I that case I see no point in discussing
Quoted 4 lines
right, if we need to distinguish between explicit and implicit hv-evmcs set by hv-passthrough custom parser probably the way to go.
However do we need actually need to do it?
I think we really need that. See below ...
Quoted 12 lines
I'd treat 'hv-passthrough,-vmx' the same way as 'hv-passthrough,hv-evmcs,-vmx' and it applies not only hv-evmcs but other features hv-passthrough might set (i.e. if whatever was [un]set by hv-passthrough in combination with other features results in invalid config, QEMU shall error out instead of magically altering host provided hv-passthrough value).
something like:
'hv-passthrough,-vmx' when hv-passthrough makes hv-evmcs bit set
should result in
error_setg(errp,"'vmx' feature can't be disabled when hv-evmcs is enabled,"
" either enable 'vmx' or disable 'hv-evmcs' along with disabling 'vmx'"Quoted 7 lines
making host's features set, *magically* mutable, depending on other user provided features is a bit confusing. One would never know what hv-passthrough actually means, and if enabling/disabling 'random' feature changes it.
It's cleaner to do just what user asked (whether implicitly or explicitly) and error out in case it ends up in nonsense configuration.
I don't seem to agree this is a sane behavior, especially if you replace 'hv-passthrough' with 'hv-default' above. Removing 'vmx' from CPU for Windows guests is common if you'd want to avoid nested configuration:
even without any Hyper-V guests created, Windows itself is a Hyper-V partition.Quoted 8 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
and on Intel he will get an error, and on AMD he won't.
So what you're suggesting actually defeats the whole purpose of 'hv-default' as upper-layer tools (think libvirt) will need to know that
Quoted 8 lines
I'd assume it would be hard for libvirt to use 'hv-default' from migration point of view. It's semi opaque (one can find out what features it sets indirectly inspecting individual
hv_foofeatures, and mgmt will need to know about them). If it will mutate when other features [un]set, upper layers might need to enumerate all these permutations to know which hosts are compatible or compare host feature sets every time before attempting migration.
That's exactly the opposite of what's the goal here which is: make it possible for upper layers to not know anything about Hyper-V enlightenments besides 'hv-default'. Migration should work just fine, if the rest of guest configuration matches -- then 'hv-default' will create the exact same things (e.g. if 'vmx' was disabled on the source it has to be enabled on the destination, it can't be different)
Quoted 7 lines
Quoted 2 lines
Intel configurations for Windows guests are somewhat different. They'll need to know what 'hv-evmcs' is. We're back to where we've started.
we were talking about hv-passthrough, and if host advertises hv-evmcs QEMU should complain if user disabled features it depends on ( not silently fixing up configuration error).
But the same applies to hv-default.
Let's forget about hv-passthrough completely for a while as this series is kind of unrelated to it.
In the previous submission I was setting 'hv-default' based on host availability of the feature only. That is: set on Intel, unset on AMD. We have to at least preserve that because it would be insane to crash on
-cpu host,hv-default
on AMD because AMD doesn't (and never will!) support hv-evmcs, right?
Quoted 8 lines
Quoted 4 lines
If we are to follow this approach let's just throw away 'hv-evmcs' from 'hv-default' set, it's going to be much cleaner. But again, I don't really believe it's the right way to go.
if desired behavior, on Intel host for above config, to start without error then indeed defaults should not set 'hv-evmcs' if it results in invalid feature set.
This is problematic as it is still sane for everyone to enable it as it gives performance advantage. If we just for a second forget about custom parsers and all that -- which is just an implementation detail, why can't we tie 'hv-evmcs' bit in 'hv-default' to 'vxm' 1:1?
Again, the end goal is: make it possible for upper layers to now know anything about Hyper-V enlightenments other than 'hv-default'. Technically, it is possible to make it work.
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 107 lines · On Tue, 23 Feb 2021 19:08:42 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 97 lines · Igor Mammedov
Quoted 86 lines · On Tue, 23 Feb 2021 16:46:50 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 51 lines · Igor Mammedov
Quoted 39 lines · On Mon, 22 Feb 2021 11:20:34 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 25 lines · Vitaly Kuznetsov
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
Quoted 6 lines
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user.
Igor,
could you please comment on the above? In case my line of thought is correct, and it is impossible to distinguish between e.g.
Quoted 8 lines
'hv-passthrough,hv-evmcs,-vmx'
and
'hv-passthrough,-vmx'without a custom parser (written just exactly the way I did in this version, for example) regardless of when 'hv-passthrough' is expanded. E.g. we have the exact same problem with 'hv-default,hv-evmcs,-vmx'. I that case I see no point in discussing
Quoted 4 lines
right, if we need to distinguish between explicit and implicit hv-evmcs set by hv-passthrough custom parser probably the way to go.
However do we need actually need to do it?
I think we really need that. See below ...
Quoted 12 lines
I'd treat 'hv-passthrough,-vmx' the same way as 'hv-passthrough,hv-evmcs,-vmx' and it applies not only hv-evmcs but other features hv-passthrough might set (i.e. if whatever was [un]set by hv-passthrough in combination with other features results in invalid config, QEMU shall error out instead of magically altering host provided hv-passthrough value).
something like:
'hv-passthrough,-vmx' when hv-passthrough makes hv-evmcs bit set
should result in
error_setg(errp,"'vmx' feature can't be disabled when hv-evmcs is enabled,"
" either enable 'vmx' or disable 'hv-evmcs' along with disabling 'vmx'"Quoted 7 lines
making host's features set, *magically* mutable, depending on other user provided features is a bit confusing. One would never know what hv-passthrough actually means, and if enabling/disabling 'random' feature changes it.
It's cleaner to do just what user asked (whether implicitly or explicitly) and error out in case it ends up in nonsense configuration.
I don't seem to agree this is a sane behavior, especially if you replace 'hv-passthrough' with 'hv-default' above. Removing 'vmx' from CPU for Windows guests is common if you'd want to avoid nested configuration:
even without any Hyper-V guests created, Windows itself is a Hyper-V partition.Quoted 8 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
and on Intel he will get an error, and on AMD he won't.
So what you're suggesting actually defeats the whole purpose of 'hv-default' as upper-layer tools (think libvirt) will need to know that
Quoted 7 lines
I'd assume it would be hard for libvirt to use 'hv-default' from migration point of view. It's semi opaque (one can find out what features it sets indirectly inspecting individual
hv_foofeatures, and mgmt will need to know about them). If it will mutate when other features [un]set, upper layers might need to enumerate all these permutations to know which hosts are compatible or compare host feature sets every time before attempting migration.
Quoted 5 lines
That's exactly the opposite of what's the goal here which is: make it possible for upper layers to not know anything about Hyper-V enlightenments besides 'hv-default'. Migration should work just fine, if the rest of guest configuration matches -- then 'hv-default' will create the exact same things (e.g. if 'vmx' was disabled on the source it has
^^^^^
I'm not convinced in that yet (not with current impl. more on that at the end of reply)
Quoted 6 lines
to be enabled on the destination, it can't be different)
Quoted 3 lines
Quoted 2 lines
Intel configurations for Windows guests are somewhat different. They'll need to know what 'hv-evmcs' is. We're back to where we've started.
Quoted 7 lines
Quoted 4 lines
we were talking about hv-passthrough, and if host advertises hv-evmcs QEMU should complain if user disabled features it depends on ( not silently fixing up configuration error).
But the same applies to hv-default.Let's forget about hv-passthrough completely for a while as this series is kind of unrelated to it.
It adds a lot for unrelated code (not just couple of lines),
I've played with scratch CPU idea, here is demo of it
https://github.com/imammedo/qemu/commit/a4b107d5368ebf72d45082bc8310a6b88a4ba6fb
I didn't rework caps/cpuid querying parts (just hacked around it),
and even without that it saves us ~200LOC (not a small part of which comes with this series).
I also split horrible hv_cpuid_check_and_set into separate 'set' and 'check' stages. Granted it was sort-of pre-existing ugly code, some of your re-factoring made it a bit better but it's still far from readable.
Quoted 8 lines
In the previous submission I was setting 'hv-default' based on host availability of the feature only. That is: set on Intel, unset on AMD. We have to at least preserve that because it would be insane to crash on
-cpu host,hv-default
on AMD because AMD doesn't (and never will!) support hv-evmcs, right?
If QEMU prevents cross arch migration i.e. it's not supported,
then I guess we can make hv-default different depending on AMD or Intel host. If not then we might need to be conservative i.e. exclude hv-evmcs from defaults.
Quoted 10 lines
Quoted 7 lines
Quoted 3 lines
If we are to follow this approach let's just throw away 'hv-evmcs' from 'hv-default' set, it's going to be much cleaner. But again, I don't really believe it's the right way to go.
if desired behavior, on Intel host for above config, to start without error then indeed defaults should not set 'hv-evmcs' if it results in invalid feature set.
This is problematic as it is still sane for everyone to enable it as it gives performance advantage. If we just for a second forget about custom
"
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
"
it's not easy picking defaults.
Quoted 2 lines
parsers and all that -- which is just an implementation detail, why can't we tie 'hv-evmcs' bit in 'hv-default' to 'vxm' 1:1?
migration wise I don't see issues wrt vmx=off turning of hv-evmcs, however ...
we were replacing user input fixups with hard errors asking user to fix CLI and removing custom parsers in favor of generic ones.
In vmx=off case we would be fixing up what 'hv-default' explicitly set. Same applies to other hv-foo set by hv-default.
ex: 'hv-default,hv-dep1=off', will turn off some dependent feature for other hv feature in hv-default set and it will error out, same goes on for enabling feature that has dependencies. Why should we treat hv-evmcs/vmx pair any different?
Granted exiting with error is not the best UX, but at least it says to user what's wrong with CLI and how to fix it. Also it lets to keep QEMU code manageable and with consistent behavior.
Quoted 2 lines
Again, the end goal is: make it possible for upper layers to now know anything about Hyper-V enlightenments other than 'hv-default'.
I'm still doubtful about feasibility of this goal when migration is considered. It sure would work if hosts are identical hw/sw wise.
In mixed setup all features, except of hv-evmcs, that included in 'hv-default', will error out in case host doesn't support it, which should prevent incompatible migration so that's also fine.
But hv-evmcs will silently go away if host doesn't support it, which is issue when migration happens to/from host that supports it.
Maybe to help mgmt to figure out hosts compatibility
1. it should know about hv-evmcs to query it's status
2. or default value set by 'hv-default' should be exposed to mgmt so it could compare whole feature-set in one go without being aware of individual features.
Additionally on QEMU side for such conditional features we can theoretically add a subsection to migration stream when feature is enabled, that way we at least can prevent 'successful' migration, when destination value doesn't match. But this might already be over-engineering on my part.
Quoted 1 line
Technically, it is possible to make it work.
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 132 lines · Igor Mammedov
Quoted 107 lines · On Tue, 23 Feb 2021 19:08:42 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 97 lines · Igor Mammedov
Quoted 86 lines · On Tue, 23 Feb 2021 16:46:50 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 51 lines · Igor Mammedov
Quoted 39 lines · On Mon, 22 Feb 2021 11:20:34 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 25 lines · Vitaly Kuznetsov
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
Quoted 6 lines
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user.
Igor,
could you please comment on the above? In case my line of thought is correct, and it is impossible to distinguish between e.g.
Quoted 8 lines
'hv-passthrough,hv-evmcs,-vmx'
and
'hv-passthrough,-vmx'without a custom parser (written just exactly the way I did in this version, for example) regardless of when 'hv-passthrough' is expanded. E.g. we have the exact same problem with 'hv-default,hv-evmcs,-vmx'. I that case I see no point in discussing
Quoted 4 lines
right, if we need to distinguish between explicit and implicit hv-evmcs set by hv-passthrough custom parser probably the way to go.
However do we need actually need to do it?
I think we really need that. See below ...
Quoted 12 lines
I'd treat 'hv-passthrough,-vmx' the same way as 'hv-passthrough,hv-evmcs,-vmx' and it applies not only hv-evmcs but other features hv-passthrough might set (i.e. if whatever was [un]set by hv-passthrough in combination with other features results in invalid config, QEMU shall error out instead of magically altering host provided hv-passthrough value).
something like:
'hv-passthrough,-vmx' when hv-passthrough makes hv-evmcs bit set
should result in
error_setg(errp,"'vmx' feature can't be disabled when hv-evmcs is enabled,"
" either enable 'vmx' or disable 'hv-evmcs' along with disabling 'vmx'"Quoted 7 lines
making host's features set, *magically* mutable, depending on other user provided features is a bit confusing. One would never know what hv-passthrough actually means, and if enabling/disabling 'random' feature changes it.
It's cleaner to do just what user asked (whether implicitly or explicitly) and error out in case it ends up in nonsense configuration.
I don't seem to agree this is a sane behavior, especially if you replace 'hv-passthrough' with 'hv-default' above. Removing 'vmx' from CPU for Windows guests is common if you'd want to avoid nested configuration:
even without any Hyper-V guests created, Windows itself is a Hyper-V partition.Quoted 8 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
and on Intel he will get an error, and on AMD he won't.
So what you're suggesting actually defeats the whole purpose of 'hv-default' as upper-layer tools (think libvirt) will need to know that
Quoted 7 lines
I'd assume it would be hard for libvirt to use 'hv-default' from migration point of view. It's semi opaque (one can find out what features it sets indirectly inspecting individual
hv_foofeatures, and mgmt will need to know about them). If it will mutate when other features [un]set, upper layers might need to enumerate all these permutations to know which hosts are compatible or compare host feature sets every time before attempting migration.Quoted 5 lines
That's exactly the opposite of what's the goal here which is: make it possible for upper layers to not know anything about Hyper-V enlightenments besides 'hv-default'. Migration should work just fine, if the rest of guest configuration matches -- then 'hv-default' will create the exact same things (e.g. if 'vmx' was disabled on the source it has
^^^^^I'm not convinced in that yet (not with current impl. more on that at the end of reply)
Quoted 6 lines
to be enabled on the destination, it can't be different)
Quoted 3 lines
Quoted 2 lines
Intel configurations for Windows guests are somewhat different. They'll need to know what 'hv-evmcs' is. We're back to where we've started.
Quoted 7 lines
Quoted 4 lines
we were talking about hv-passthrough, and if host advertises hv-evmcs QEMU should complain if user disabled features it depends on ( not silently fixing up configuration error).
But the same applies to hv-default.Let's forget about hv-passthrough completely for a while as this series is kind of unrelated to it.
Quoted 6 lines
It adds a lot for unrelated code (not just couple of lines), I've played with scratch CPU idea, here is demo of it
https://github.com/imammedo/qemu/commit/a4b107d5368ebf72d45082bc8310a6b88a4ba6fb
I didn't rework caps/cpuid querying parts (just hacked around it), and even without that it saves us ~200LOC (not a small part of which comes with this series).
All your savings come from throwing away custom parsers -- which are not needed at all if we don't distinguish between
'hv-default,hv-evmcs' and 'hv-default'
it's just not needed, don't count these patches in. Or, if it is needed, please explain how your scratch CPU is making things different. I guess it is not so we can discuss this outside of this series.
Quoted 4 lines
I also split horrible
hv_cpuid_check_and_setinto separate 'set' and 'check' stages. Granted it was sort-of pre-existing ugly code, some of your re-factoring made it a bit better but it's still far from readable.
hv_cpuid_check_and_set() is already there, I'm not at all opposed to making this code even better but I don't see it as a must for this particular feature (hv-default).
Quoted 5 lines
Quoted 5 lines
In the previous submission I was setting 'hv-default' based on host availability of the feature only. That is: set on Intel, unset on AMD. We have to at least preserve that because it would be insane to crash on
Quoted 8 lines
Quoted 3 lines
-cpu host,hv-default
on AMD because AMD doesn't (and never will!) support hv-evmcs, right?
If QEMU prevents cross arch migration i.e. it's not supported,
then I guess we can make hv-default different depending on AMD or Intel host. If not then we might need to be conservative i.e. exclude hv-evmcs from defaults.
Forget about cross vendor. I want to tie it 1:1 to VMX feature on the guest CPU -- which happens to be only available on Intel. It is absolutely impossible to migrate VMX enabled guest to VMX-disabled destination, with or without evmcs.
Quoted 25 lines
Quoted 10 lines
Quoted 7 lines
Quoted 3 lines
If we are to follow this approach let's just throw away 'hv-evmcs' from 'hv-default' set, it's going to be much cleaner. But again, I don't really believe it's the right way to go.
if desired behavior, on Intel host for above config, to start without error then indeed defaults should not set 'hv-evmcs' if it results in invalid feature set.
This is problematic as it is still sane for everyone to enable it as it gives performance advantage. If we just for a second forget about custom
"
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
"
it's not easy picking defaults.Quoted 2 lines
parsers and all that -- which is just an implementation detail, why can't we tie 'hv-evmcs' bit in 'hv-default' to 'vxm' 1:1?
migration wise I don't see issues wrt vmx=off turning of hv-evmcs, however ...
we were replacing user input fixups with hard errors asking user to fix CLI and removing custom parsers in favor of generic ones.
Quoted 8 lines
In vmx=off case we would be fixing up what 'hv-default' explicitly set. Same applies to other hv-foo set by hv-default.
ex: 'hv-default,hv-dep1=off', will turn off some dependent feature for other hv feature in hv-default set and it will error out, same goes on for enabling feature that has dependencies. Why should we treat hv-evmcs/vmx pair any different?
VMX is not part of Hyper-V enlightenments, is it? It can also be coming from a CPU model:
"-cpu MyLovelyModelWithVmx,hv-default"
should not throw an error!
Again, the goal is for userspace to not know anything besides 'hv-default' for Hyper-V enlightenments.
Quoted 3 lines
Granted exiting with error is not the best UX, but at least it says to user what's wrong with CLI and how to fix it. Also it lets to keep QEMU code manageable and with consistent behavior.
Enabling EVMCS only when guest CPU has VMX is a smart behavior, all users want that. It is very consistent with how genuine Hyper-V behaves.
Quoted 11 lines
Quoted 3 lines
Again, the end goal is: make it possible for upper layers to now know anything about Hyper-V enlightenments other than 'hv-default'.
I'm still doubtful about feasibility of this goal when migration is considered. It sure would work if hosts are identical hw/sw wise.
In mixed setup all features, except of hv-evmcs, that included in 'hv-default', will error out in case host doesn't support it, which should prevent incompatible migration so that's also fine.But hv-evmcs will silently go away if host doesn't support it, which is issue when migration happens to/from host that supports it.
But how can it go away? In KVM, hv-evmcs support is not conditional, basically, all KVMs which support netsting state migration also support EVMCS.
Quoted 7 lines
Maybe to help mgmt to figure out hosts compatibility
1. it should know about hv-evmcs to query it's status
2. or default value set by 'hv-default' should be exposed to mgmt so it could compare whole feature-set in one go without being aware of individual features.
Quoted 5 lines
Additionally on QEMU side for such conditional features we can theoretically add a subsection to migration stream when feature is enabled, that way we at least can prevent 'successful' migration, when destination value doesn't match. But this might already be over-engineering on my part.
I think you're trying to solve an issue which doesn't exist. In case we're successfully migrating a nested enabled guest, our KVM is modern enough and supports EVMCS (on Intel, of course). Also, nested state (which is part of the migration stream) has EVMCS flag, we can't migrate somewhere where the flag is unsupported.
Anyway, I feel we're walking in circles. I'm ready to just drop all EVMCS related bits from this seies to get it merged. This will make the part which you hate the most ("custom parsers") go away too. We can discuss EVMCS to death after that and, when we finally decide that user convenience is actually worth something, we can add 'hv-evmcs' to the new
Eduardo, do you see any way forward here?
Signature · 1 line
Vitaly
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 141 lines · On Wed, 24 Feb 2021 18:00:43 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 132 lines · Igor Mammedov
Quoted 107 lines · On Tue, 23 Feb 2021 19:08:42 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 97 lines · Igor Mammedov
Quoted 86 lines · On Tue, 23 Feb 2021 16:46:50 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 51 lines · Igor Mammedov
Quoted 39 lines · On Mon, 22 Feb 2021 11:20:34 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 25 lines · Vitaly Kuznetsov
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
Quoted 6 lines
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user.
Igor,
could you please comment on the above? In case my line of thought is correct, and it is impossible to distinguish between e.g.
Quoted 8 lines
'hv-passthrough,hv-evmcs,-vmx'
and
'hv-passthrough,-vmx'without a custom parser (written just exactly the way I did in this version, for example) regardless of when 'hv-passthrough' is expanded. E.g. we have the exact same problem with 'hv-default,hv-evmcs,-vmx'. I that case I see no point in discussing
Quoted 4 lines
right, if we need to distinguish between explicit and implicit hv-evmcs set by hv-passthrough custom parser probably the way to go.
However do we need actually need to do it?
I think we really need that. See below ...
Quoted 12 lines
I'd treat 'hv-passthrough,-vmx' the same way as 'hv-passthrough,hv-evmcs,-vmx' and it applies not only hv-evmcs but other features hv-passthrough might set (i.e. if whatever was [un]set by hv-passthrough in combination with other features results in invalid config, QEMU shall error out instead of magically altering host provided hv-passthrough value).
something like:
'hv-passthrough,-vmx' when hv-passthrough makes hv-evmcs bit set
should result in
error_setg(errp,"'vmx' feature can't be disabled when hv-evmcs is enabled,"
" either enable 'vmx' or disable 'hv-evmcs' along with disabling 'vmx'"Quoted 7 lines
making host's features set, *magically* mutable, depending on other user provided features is a bit confusing. One would never know what hv-passthrough actually means, and if enabling/disabling 'random' feature changes it.
It's cleaner to do just what user asked (whether implicitly or explicitly) and error out in case it ends up in nonsense configuration.
I don't seem to agree this is a sane behavior, especially if you replace 'hv-passthrough' with 'hv-default' above. Removing 'vmx' from CPU for Windows guests is common if you'd want to avoid nested configuration:
even without any Hyper-V guests created, Windows itself is a Hyper-V partition.Quoted 8 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
and on Intel he will get an error, and on AMD he won't.
So what you're suggesting actually defeats the whole purpose of 'hv-default' as upper-layer tools (think libvirt) will need to know that
Quoted 7 lines
I'd assume it would be hard for libvirt to use 'hv-default' from migration point of view. It's semi opaque (one can find out what features it sets indirectly inspecting individual
hv_foofeatures, and mgmt will need to know about them). If it will mutate when other features [un]set, upper layers might need to enumerate all these permutations to know which hosts are compatible or compare host feature sets every time before attempting migration.Quoted 5 lines
That's exactly the opposite of what's the goal here which is: make it possible for upper layers to not know anything about Hyper-V enlightenments besides 'hv-default'. Migration should work just fine, if the rest of guest configuration matches -- then 'hv-default' will create the exact same things (e.g. if 'vmx' was disabled on the source it has
^^^^^I'm not convinced in that yet (not with current impl. more on that at the end of reply)
Quoted 6 lines
to be enabled on the destination, it can't be different)
Quoted 3 lines
Quoted 2 lines
Intel configurations for Windows guests are somewhat different. They'll need to know what 'hv-evmcs' is. We're back to where we've started.
Quoted 7 lines
Quoted 4 lines
we were talking about hv-passthrough, and if host advertises hv-evmcs QEMU should complain if user disabled features it depends on ( not silently fixing up configuration error).
But the same applies to hv-default.Let's forget about hv-passthrough completely for a while as this series is kind of unrelated to it.
Quoted 6 lines
It adds a lot for unrelated code (not just couple of lines), I've played with scratch CPU idea, here is demo of it
https://github.com/imammedo/qemu/commit/a4b107d5368ebf72d45082bc8310a6b88a4ba6fb
I didn't rework caps/cpuid querying parts (just hacked around it), and even without that it saves us ~200LOC (not a small part of which comes with this series).
Quoted 8 lines
All your savings come from throwing away custom parsers -- which are not needed at all if we don't distinguish between
'hv-default,hv-evmcs' and 'hv-default'
it's just not needed, don't count these patches in. Or, if it is needed, please explain how your scratch CPU is making things different. I guess it is not so we can discuss this outside of this series.
scratch CPU helps with hostpassthrough refactoring which also brings in dependency on custom parser.
Quoted 8 lines
Quoted 4 lines
I also split horrible
hv_cpuid_check_and_setinto separate 'set' and 'check' stages. Granted it was sort-of pre-existing ugly code, some of your re-factoring made it a bit better but it's still far from readable.
hv_cpuid_check_and_set()is already there, I'm not at all opposed to making this code even better but I don't see it as a must for this particular feature (hv-default).
it's not a must have for hv-default especially if you don't touch it. (however if you touch it & co, I'd ask to clean it up first)
Quoted 14 lines
Quoted 5 lines
Quoted 5 lines
In the previous submission I was setting 'hv-default' based on host availability of the feature only. That is: set on Intel, unset on AMD. We have to at least preserve that because it would be insane to crash on
Quoted 8 lines
Quoted 3 lines
-cpu host,hv-default
on AMD because AMD doesn't (and never will!) support hv-evmcs, right?
If QEMU prevents cross arch migration i.e. it's not supported,
then I guess we can make hv-default different depending on AMD or Intel host. If not then we might need to be conservative i.e. exclude hv-evmcs from defaults.
Quoted 4 lines
Forget about cross vendor. I want to tie it 1:1 to VMX feature on the guest CPU -- which happens to be only available on Intel. It is absolutely impossible to migrate VMX enabled guest to VMX-disabled destination, with or without evmcs.
Ok
Quoted 45 lines
Quoted 25 lines
Quoted 10 lines
Quoted 7 lines
Quoted 3 lines
If we are to follow this approach let's just throw away 'hv-evmcs' from 'hv-default' set, it's going to be much cleaner. But again, I don't really believe it's the right way to go.
if desired behavior, on Intel host for above config, to start without error then indeed defaults should not set 'hv-evmcs' if it results in invalid feature set.
This is problematic as it is still sane for everyone to enable it as it gives performance advantage. If we just for a second forget about custom
"
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
"
it's not easy picking defaults.Quoted 2 lines
parsers and all that -- which is just an implementation detail, why can't we tie 'hv-evmcs' bit in 'hv-default' to 'vxm' 1:1?
migration wise I don't see issues wrt vmx=off turning of hv-evmcs, however ...
we were replacing user input fixups with hard errors asking user to fix CLI and removing custom parsers in favor of generic ones.
Quoted 8 lines
In vmx=off case we would be fixing up what 'hv-default' explicitly set. Same applies to other hv-foo set by hv-default.
ex: 'hv-default,hv-dep1=off', will turn off some dependent feature for other hv feature in hv-default set and it will error out, same goes on for enabling feature that has dependencies. Why should we treat hv-evmcs/vmx pair any different?
VMX is not part of Hyper-V enlightenments, is it? It can also be coming from a CPU model:
"-cpu MyLovelyModelWithVmx,hv-default"
should not throw an error!
Again, the goal is for userspace to not know anything besides 'hv-default' for Hyper-V enlightenments.
Quoted 6 lines
Quoted 3 lines
Granted exiting with error is not the best UX, but at least it says to user what's wrong with CLI and how to fix it. Also it lets to keep QEMU code manageable and with consistent behavior.
Enabling EVMCS only when guest CPU has VMX is a smart behavior, all users want that. It is very consistent with how genuine Hyper-V behaves.
it's all good,
unless VMX is absent for whatever reasons (cpumodel or vmx=off on CLI), in this case just error out and say what's wrong instead of trying to fix CLI up.
Quoted 8 lines
Quoted 8 lines
Quoted 2 lines
Again, the end goal is: make it possible for upper layers to now know anything about Hyper-V enlightenments other than 'hv-default'.
I'm still doubtful about feasibility of this goal when migration is considered. It sure would work if hosts are identical hw/sw wise.
In mixed setup all features, except of hv-evmcs, that included in 'hv-default', will error out in case host doesn't support it, which should prevent incompatible migration so that's also fine.
Quoted 6 lines
Quoted 2 lines
But hv-evmcs will silently go away if host doesn't support it, which is issue when migration happens to/from host that supports it.
But how can it go away? In KVM, hv-evmcs support is not conditional, basically, all KVMs which support netsting state migration also support EVMCS.
you added that in kernel between 4.19-4.20 (8cab6507f64),
so on Intel host flag can change depending on not so old kernel version.
Unless QEMU minimum supported kernel version is 4.20, we can't ignore that.
(downstreams will have to take care of it on its own)
Quoted 13 lines
Quoted 7 lines
Maybe to help mgmt to figure out hosts compatibility
1. it should know about hv-evmcs to query it's status
2. or default value set by 'hv-default' should be exposed to mgmt so it could compare whole feature-set in one go without being aware of individual features.Quoted 5 lines
Additionally on QEMU side for such conditional features we can theoretically add a subsection to migration stream when feature is enabled, that way we at least can prevent 'successful' migration, when destination value doesn't match. But this might already be over-engineering on my part.
Quoted 5 lines
I think you're trying to solve an issue which doesn't exist. In case we're successfully migrating a nested enabled guest, our KVM is modern enough and supports EVMCS (on Intel, of course). Also, nested state (which is part of the migration stream) has EVMCS flag, we can't migrate somewhere where the flag is unsupported.
that's what I was missing.
Can you point out to the code that makes sure that migration fails? (a comment where hv-evcms is added to default set explaining why it's safe, pls)
Quoted 6 lines
Anyway, I feel we're walking in circles. I'm ready to just drop all EVMCS related bits from this seies to get it merged. This will make the part which you hate the most ("custom parsers") go away too. We can discuss EVMCS to death after that and, when we finally decide that user convenience is actually worth something, we can add 'hv-evmcs' to the new
fine with me, it would be even easier to review if it were just a patch that would add 'hv-defaults', without non must have refactoring. (cleanup/refactoring could be another series)
If we can guarantee that hv-evcms won't flip on/off on all supported kernels, I'm also fine with keeping it in default set and error-ing out if vmx ends up in off state.
However if it changes, we need to expose 'default set' to mgmt somehow, so it will know that hosts aren't compatible (instead of finding out it hard way in form of failed migration (assuming it fails))
Quoted 2 lines
Eduardo, do you see any way forward here?
Re: [PATCH v4 16/21] i386: track explicit 'hv-*' features enablement/disablement
Quoted 147 lines · Igor Mammedov
Quoted 144 lines · On Wed, 24 Feb 2021 18:00:43 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 132 lines · Igor Mammedov
Quoted 107 lines · On Tue, 23 Feb 2021 19:08:42 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 97 lines · Igor Mammedov
Quoted 86 lines · On Tue, 23 Feb 2021 16:46:50 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 51 lines · Igor Mammedov
Quoted 39 lines · On Mon, 22 Feb 2021 11:20:34 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 25 lines · Vitaly Kuznetsov
Quoted 6 lines · Igor Mammedov
Quoted 5 lines
We need to distinguish because that would be sane.
Enlightened VMCS is an extension to VMX, it can't be used without it. Genuine Hyper-V doesn't have a knob for enabling and disabling it,
...
Quoted 8 lines
Quoted 6 lines
That bein said, if
guest CPU lacks VMX it is counter-productive to expose EVMCS. However, there is a problem with explicit enablement: what should'hv-passthrough,hv-evmcs' option do? Just silently drop EVMCS? Doesn't sound sane to me.
based on above I'd error out is user asks for unsupported option i.e. no VMX -> no hv-evmcs - if explicitly asked -> error out
That's what I keep telling you but you don't seem to listen. 'Scratch CPU' can't possibly help with this use-case because when you parse
'hv-passthrough,hv-evmcs,vmx=off' you
1) "hv-passthrough" -> set EVMCS bit to '1' as it is supported by the host.
Quoted 6 lines
2) 'hv-evmcs' -> keep EVMCS bit '1'
3) 'vmx=off' -> you have no idea where EVMCS bit came from.
We have to remember which options were aquired from the host and which were set explicitly by the user.
Igor,
could you please comment on the above? In case my line of thought is correct, and it is impossible to distinguish between e.g.
Quoted 8 lines
'hv-passthrough,hv-evmcs,-vmx'
and
'hv-passthrough,-vmx'without a custom parser (written just exactly the way I did in this version, for example) regardless of when 'hv-passthrough' is expanded. E.g. we have the exact same problem with 'hv-default,hv-evmcs,-vmx'. I that case I see no point in discussing
Quoted 4 lines
right, if we need to distinguish between explicit and implicit hv-evmcs set by hv-passthrough custom parser probably the way to go.
However do we need actually need to do it?
I think we really need that. See below ...
Quoted 12 lines
I'd treat 'hv-passthrough,-vmx' the same way as 'hv-passthrough,hv-evmcs,-vmx' and it applies not only hv-evmcs but other features hv-passthrough might set (i.e. if whatever was [un]set by hv-passthrough in combination with other features results in invalid config, QEMU shall error out instead of magically altering host provided hv-passthrough value).
something like:
'hv-passthrough,-vmx' when hv-passthrough makes hv-evmcs bit set
should result in
error_setg(errp,"'vmx' feature can't be disabled when hv-evmcs is enabled,"
" either enable 'vmx' or disable 'hv-evmcs' along with disabling 'vmx'"Quoted 7 lines
making host's features set, *magically* mutable, depending on other user provided features is a bit confusing. One would never know what hv-passthrough actually means, and if enabling/disabling 'random' feature changes it.
It's cleaner to do just what user asked (whether implicitly or explicitly) and error out in case it ends up in nonsense configuration.
I don't seem to agree this is a sane behavior, especially if you replace 'hv-passthrough' with 'hv-default' above. Removing 'vmx' from CPU for Windows guests is common if you'd want to avoid nested configuration:
even without any Hyper-V guests created, Windows itself is a Hyper-V partition.Quoted 8 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
and on Intel he will get an error, and on AMD he won't.
So what you're suggesting actually defeats the whole purpose of 'hv-default' as upper-layer tools (think libvirt) will need to know that
Quoted 7 lines
I'd assume it would be hard for libvirt to use 'hv-default' from migration point of view. It's semi opaque (one can find out what features it sets indirectly inspecting individual
hv_foofeatures, and mgmt will need to know about them). If it will mutate when other features [un]set, upper layers might need to enumerate all these permutations to know which hosts are compatible or compare host feature sets every time before attempting migration.Quoted 5 lines
That's exactly the opposite of what's the goal here which is: make it possible for upper layers to not know anything about Hyper-V enlightenments besides 'hv-default'. Migration should work just fine, if the rest of guest configuration matches -- then 'hv-default' will create the exact same things (e.g. if 'vmx' was disabled on the source it has
^^^^^I'm not convinced in that yet (not with current impl. more on that at the end of reply)
Quoted 6 lines
to be enabled on the destination, it can't be different)
Quoted 3 lines
Quoted 2 lines
Intel configurations for Windows guests are somewhat different. They'll need to know what 'hv-evmcs' is. We're back to where we've started.
Quoted 7 lines
Quoted 4 lines
we were talking about hv-passthrough, and if host advertises hv-evmcs QEMU should complain if user disabled features it depends on ( not silently fixing up configuration error).
But the same applies to hv-default.Let's forget about hv-passthrough completely for a while as this series is kind of unrelated to it.
Quoted 6 lines
It adds a lot for unrelated code (not just couple of lines), I've played with scratch CPU idea, here is demo of it
https://github.com/imammedo/qemu/commit/a4b107d5368ebf72d45082bc8310a6b88a4ba6fb
I didn't rework caps/cpuid querying parts (just hacked around it), and even without that it saves us ~200LOC (not a small part of which comes with this series).All your savings come from throwing away custom parsers -- which are not needed at all if we don't distinguish between
Quoted 8 lines
Quoted 5 lines
'hv-default,hv-evmcs' and 'hv-default'
it's just not needed, don't count these patches in. Or, if it is needed, please explain how your scratch CPU is making things different. I guess it is not so we can discuss this outside of this series.
scratch CPU helps with hostpassthrough refactoring which also brings in dependency on custom parser.
I think I already said that but let me repeat myself: scratch CPUs don't help us in any way to distinguish between
'hv-passthrough,hv-evmcs,vmx=off'
and
'hv-passthrough,vmx=off'
and thus are utterly useless (for this particular feature). It's not 'refactoring', it's throwing a feature away for the sake of code purity.
Quoted 83 lines
Quoted 10 lines
Quoted 6 lines
I also split horrible
hv_cpuid_check_and_setinto separate 'set' and 'check' stages. Granted it was sort-of pre-existing ugly code, some of your re-factoring made it a bit better but it's still far from readable.
hv_cpuid_check_and_set()is already there, I'm not at all opposed to making this code even better but I don't see it as a must for this particular feature (hv-default).it's not a must have for hv-default especially if you don't touch it. (however if you touch it & co, I'd ask to clean it up first)
Quoted 14 lines
Quoted 5 lines
Quoted 5 lines
In the previous submission I was setting 'hv-default' based on host availability of the feature only. That is: set on Intel, unset on AMD. We have to at least preserve that because it would be insane to crash on
Quoted 8 lines
Quoted 3 lines
-cpu host,hv-default
on AMD because AMD doesn't (and never will!) support hv-evmcs, right?
If QEMU prevents cross arch migration i.e. it's not supported,
then I guess we can make hv-default different depending on AMD or Intel host. If not then we might need to be conservative i.e. exclude hv-evmcs from defaults.Quoted 4 lines
Forget about cross vendor. I want to tie it 1:1 to VMX feature on the guest CPU -- which happens to be only available on Intel. It is absolutely impossible to migrate VMX enabled guest to VMX-disabled destination, with or without evmcs.
Ok
Quoted 49 lines
Quoted 25 lines
Quoted 10 lines
Quoted 7 lines
Quoted 3 lines
If we are to follow this approach let's just throw away 'hv-evmcs' from 'hv-default' set, it's going to be much cleaner. But again, I don't really believe it's the right way to go.
if desired behavior, on Intel host for above config, to start without error then indeed defaults should not set 'hv-evmcs' if it results in invalid feature set.
This is problematic as it is still sane for everyone to enable it as it gives performance advantage. If we just for a second forget about custom
"
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
So a sane user will do:
'-cpu host,hv-default,vmx=off'
"
it's not easy picking defaults.Quoted 2 lines
parsers and all that -- which is just an implementation detail, why can't we tie 'hv-evmcs' bit in 'hv-default' to 'vxm' 1:1?
migration wise I don't see issues wrt vmx=off turning of hv-evmcs, however ...
we were replacing user input fixups with hard errors asking user to fix CLI and removing custom parsers in favor of generic ones.
Quoted 8 lines
In vmx=off case we would be fixing up what 'hv-default' explicitly set. Same applies to other hv-foo set by hv-default.
ex: 'hv-default,hv-dep1=off', will turn off some dependent feature for other hv feature in hv-default set and it will error out, same goes on for enabling feature that has dependencies. Why should we treat hv-evmcs/vmx pair any different?
VMX is not part of Hyper-V enlightenments, is it? It can also be coming from a CPU model:
"-cpu MyLovelyModelWithVmx,hv-default"
should not throw an error!
Again, the goal is for userspace to not know anything besides 'hv-default' for Hyper-V enlightenments.
Quoted 3 lines
Granted exiting with error is not the best UX, but at least it says to user what's wrong with CLI and how to fix it. Also it lets to keep QEMU code manageable and with consistent behavior.
Quoted 6 lines
Quoted 2 lines
Enabling EVMCS only when guest CPU has VMX is a smart behavior, all users want that. It is very consistent with how genuine Hyper-V behaves.
it's all good,
unless VMX is absent for whatever reasons (cpumodel or vmx=off on CLI), in this case just error out and say what's wrong instead of trying to fix CLI up.
It's easier, of course, but I don't think we should do that. At least
query-cpu-model-expansion type=full model={"name":"host","props":{"hv-passthrough":true}}
should give upper layer the full list of supported features, including evmcs, and not just throw an error. Same with 'hv-default'
Quoted 16 lines
Quoted 9 lines
Quoted 9 lines
Quoted 3 lines
Again, the end goal is: make it possible for upper layers to now know anything about Hyper-V enlightenments other than 'hv-default'.
I'm still doubtful about feasibility of this goal when migration is considered. It sure would work if hosts are identical hw/sw wise.
In mixed setup all features, except of hv-evmcs, that included in 'hv-default', will error out in case host doesn't support it, which should prevent incompatible migration so that's also fine.Quoted 6 lines
Quoted 2 lines
But hv-evmcs will silently go away if host doesn't support it, which is issue when migration happens to/from host that supports it.
But how can it go away? In KVM, hv-evmcs support is not conditional, basically, all KVMs which support netsting state migration also support EVMCS.
Quoted 6 lines
you added that in kernel between 4.19-4.20 (
8cab6507f64),
so on Intel host flag can change depending on not so old kernel version.Unless QEMU minimum supported kernel version is 4.20, we can't ignore that.
(downstreams will have to take care of it on its own)
You forget that EVMCS is not a feature in vacuum, it is an extension to VMX. To migrate a guest which has it enabled, nested state migration has to support it. Assuming the feature was enabled on the source, you need at least the following:
commit 8cab6507f64eff0ccfea01fccbc7e3e05e2aaf7e
Author: Vitaly Kuznetsov <[email protected]> Date: Tue Oct 16 18:50:09 2018 +0200
x86/kvm/nVMX: nested state migration for Enlightened VMCS
which is also in 4.20, I don't see how migration without it can even succeed.
Quoted 14 lines
Quoted 14 lines
Quoted 13 lines
Maybe to help mgmt to figure out hosts compatibility
1. it should know about hv-evmcs to query it's status
2. or default value set by 'hv-default' should be exposed to mgmt so it could compare whole feature-set in one go without being aware of individual features.Additionally on QEMU side for such conditional features we can theoretically add a subsection to migration stream when feature is enabled, that way we at least can prevent 'successful' migration, when destination value doesn't match. But this might already be over-engineering on my part.
Quoted 9 lines
Quoted 5 lines
I think you're trying to solve an issue which doesn't exist. In case we're successfully migrating a nested enabled guest, our KVM is modern enough and supports EVMCS (on Intel, of course). Also, nested state (which is part of the migration stream) has EVMCS flag, we can't migrate somewhere where the flag is unsupported.
that's what I was missing.
Can you point out to the code that makes sure that migration fails? (a comment where hv-evcms is added to default set explaining why it's safe, pls)
If you look at KVM_SET_NESTED_STATE implementation in KVM (see kvm_arch_vcpu_ioctl()), it checks for supported flags
arch/x86/kvm/x86.c- if (kvm_state.flags &arch/x86/kvm/x86.c- ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODEarch/x86/kvm/x86.c- | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDINGarch/x86/kvm/x86.c- | KVM_STATE_NESTED_GIF_SET))arch/x86/kvm/x86.c- break;
if KVM_STATE_NESTED_EVMCS is unsupported (KVM before 8cab6507f64) this condition will fail.
In reality, regardless of evmcs, there are so many bugs in nested migration prior to 5.11/5.10(maybe) that chances for successful migration on these kernels are pretty slim. Worst is, the migration won't fail, you guest (with its nested guests) will just crash. I don't recommend anyone doing nested migration in production with some older kernel, it's just not safe.
Quoted 8 lines
Quoted 7 lines
Anyway, I feel we're walking in circles. I'm ready to just drop all EVMCS related bits from this seies to get it merged. This will make the part which you hate the most ("custom parsers") go away too. We can discuss EVMCS to death after that and, when we finally decide that user convenience is actually worth something, we can add 'hv-evmcs' to the new
Quoted 4 lines
fine with me, it would be even easier to review if it were just a patch that would add 'hv-defaults', without non must have refactoring. (cleanup/refactoring could be another series)
Ok, v5 is on the list.
Quoted 3 lines
If we can guarantee that hv-evcms won't flip on/off on all supported kernels, I'm also fine with keeping it in default set and error-ing out if vmx ends up in off state.
I still don't see why and how it should (or could) flip.
Quoted 3 lines
However if it changes, we need to expose 'default set' to mgmt somehow, so it will know that hosts aren't compatible (instead of finding out it hard way in form of failed migration (assuming it fails))
The first part of this series does exactly that: exposes Hyper-V enlightenments to upper layers in QMP. Upper layer can do e.g.
query-cpu-model-expansion type=full model={"name":"host","props":{"hv-passthrough":true}}
and it will get all the supported features, including evmcs. This is one of the reasons I insist this should *always* work instead of throwing an error (with 'hv-default too).
Signature · 1 line
Vitaly
PATCH v4 17/21 i386: support 'hv-passthrough, hv-feature=off' on the command line
In reply to Vitaly KuznetsovCurrently, we support 'hv-passthrough,hv-feature=on' enablement, this is supposed to mean "hv-feature is mandatory, don't start without it". Add support for 'hv-passthrough,hv-feature=off' meaning "enable everything supported by the host except for hv-feature".
While on it, make 'hv-passthrough' parse semantics in-line with other options in qemu: when specified, it overrides what was previously set with what's supported by the host. This can later be modified with 'hv-feature=on'/ 'hv-feature=off'.
--- target/i386/cpu.c | 28 +++++++++++++++++++++++++++- target/i386/kvm/kvm.c | 4 ++++ 2 files changed, 31 insertions(+), 1 deletion(-)
target/i386/cpu.c +27 -1
@@ -4725,6 +4725,29 @@ static void x86_hv_stimer_direct_set(Object *obj, bool value, Error **errp)x86_hv_feature_set (obj, value,HYPERV_FEAT_STIMER_DIRECT ); } +static bool x86_hv_passthrough_get (Object *obj,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + +return cpu->hyperv_passthrough; +} + +static void x86_hv_passthrough_set (Object *obj,bool value,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + + cpu->hyperv_passthrough = value; + +/* hv-passthrough overrides everything with what's supported by the host */ +if (value) { + cpu->hyperv_features =0 ; + cpu->hyperv_features_on =0 ; + cpu->hyperv_features_off =0 ; + } + +return ; +} +/* Generic getter for "feature-words" and "filtered-features" properties */ static void x86_cpu_get_feature_words(Object *obj, Visitor *v, const char *name, void *opaque,
@@ -7281,7 +7304,6 @@ static Property x86_cpu_properties[] = { HYPERV_SPINLOCK_NEVER_NOTIFY),DEFINE_PROP_ON_OFF_AUTO ("hv-no-nonarch-coresharing" ,X86CPU , hyperv_no_nonarch_cs,ON_OFF_AUTO_OFF ), -DEFINE_PROP_BOOL ("hv-passthrough" ,X86CPU , hyperv_passthrough, false),DEFINE_PROP_BOOL ("check" ,X86CPU , check_cpuid, true), DEFINE_PROP_BOOL("enforce", X86CPU, enforce_cpuid, false),
@@ -7460,6 +7482,10 @@ static void x86_cpu_common_class_init(ObjectClass *oc, void *data) x86_hv_stimer_direct_get, x86_hv_stimer_direct_set); +object_class_property_add_bool (oc,"hv-passthrough" , + x86_hv_passthrough_get, + x86_hv_passthrough_set); +for (w =0 ; w <FEATURE_WORDS ; w++) {int bitnr;for (bitnr =0 ; bitnr <64 ; bitnr++) {
target/i386/kvm/kvm.c +4 -0
@@ -1153,6 +1153,10 @@ static int hv_cpuid_check_and_set(CPUState *cs, int feature, Error **errp)return 0 ; } +if (cpu->hyperv_passthrough && (cpu->hyperv_features_off &BIT (feature))) { +return 0 ; + } + deps = kvm_hyperv_properties[feature].dependencies;while (deps) { dep_feat =ctz64 (deps);
Signature · 1 line
2.29.2
Re: [PATCH v4 17/21] i386: support 'hv-passthrough, hv-feature=off' on the command line
Quoted 32 lines · On Wed, 10 Feb 2021 17:40:29 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Currently, we support 'hv-passthrough,hv-feature=on' enablement, this is supposed to mean "hv-feature is mandatory, don't start without it". Add support for 'hv-passthrough,hv-feature=off' meaning "enable everything supported by the host except for hv-feature".
While on it, make 'hv-passthrough' parse semantics in-line with other options in qemu: when specified, it overrides what was previously set with what's supported by the host. This can later be modified with 'hv-feature=on'/ 'hv-feature=off'.
Signed-off-by Vitaly Kuznetsov <[email protected]>--- target/i386/cpu.c | 28 +++++++++++++++++++++++++++- target/i386/kvm/kvm.c | 4 ++++ 2 files changed, 31 insertions(+), 1 deletion(-)target/i386/cpu.c +8
-0@@ -4725,6 +4725,29 @@ static void x86_hv_stimer_direct_set(Object *obj, bool value, Error **errp)x86_hv_feature_set (obj, value,HYPERV_FEAT_STIMER_DIRECT ); } +static bool x86_hv_passthrough_get (Object *obj,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + +return cpu->hyperv_passthrough; +} + +static void x86_hv_passthrough_set (Object *obj,bool value,Error **errp)
target/i386/cpu.c static void x86_hv_stimer_direct_set(Object *obj, bool value, Error **errp) · line 4745
Quoted 10 lines
target/i386/cpu.c +10
-0+{ +X86CPU *cpu =X86_CPU (obj); + + cpu->hyperv_passthrough = value; + +/* hv-passthrough overrides everything with what's supported by the host */ +if (value) { + cpu->hyperv_features =0 ; + cpu->hyperv_features_on =0 ; + cpu->hyperv_features_off =0 ;
why do we have _on|_off fields?
Quoted 42 lines
target/i386/cpu.c +9
-1+ } + +return ; +} +/* Generic getter for "feature-words" and "filtered-features" properties */ static void x86_cpu_get_feature_words(Object *obj, Visitor *v, const char *name, void *opaque,@@ -7281,7 +7304,6 @@ static Property x86_cpu_properties[] = { HYPERV_SPINLOCK_NEVER_NOTIFY),DEFINE_PROP_ON_OFF_AUTO ("hv-no-nonarch-coresharing" ,X86CPU , hyperv_no_nonarch_cs,ON_OFF_AUTO_OFF ), -DEFINE_PROP_BOOL ("hv-passthrough" ,X86CPU , hyperv_passthrough, false),DEFINE_PROP_BOOL ("check" ,X86CPU , check_cpuid, true), DEFINE_PROP_BOOL("enforce", X86CPU, enforce_cpuid, false),@@ -7460,6 +7482,10 @@ static void x86_cpu_common_class_init(ObjectClass *oc, void *data) x86_hv_stimer_direct_get, x86_hv_stimer_direct_set); +object_class_property_add_bool (oc,"hv-passthrough" , + x86_hv_passthrough_get, + x86_hv_passthrough_set); +for (w =0 ; w <FEATURE_WORDS ; w++) {int bitnr;for (bitnr =0 ; bitnr <64 ; bitnr++) {target/i386/kvm/kvm.c +4
-0@@ -1153,6 +1153,10 @@ static int hv_cpuid_check_and_set(CPUState *cs, int feature, Error **errp)return 0 ; } +if (cpu->hyperv_passthrough && (cpu->hyperv_features_off &BIT (feature))) { +return 0 ; + } + deps = kvm_hyperv_properties[feature].dependencies;while (deps) { dep_feat =ctz64 (deps);
Re: [PATCH v4 17/21] i386: support 'hv-passthrough, hv-feature=off' on the command line
Quoted 39 lines · Igor Mammedov
Quoted 36 lines · On Wed, 10 Feb 2021 17:40:29 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Currently, we support 'hv-passthrough,hv-feature=on' enablement, this is supposed to mean "hv-feature is mandatory, don't start without it". Add support for 'hv-passthrough,hv-feature=off' meaning "enable everything supported by the host except for hv-feature".
While on it, make 'hv-passthrough' parse semantics in-line with other options in qemu: when specified, it overrides what was previously set with what's supported by the host. This can later be modified with 'hv-feature=on'/ 'hv-feature=off'.
Signed-off-by Vitaly Kuznetsov <[email protected]>--- target/i386/cpu.c | 28 +++++++++++++++++++++++++++- target/i386/kvm/kvm.c | 4 ++++ 2 files changed, 31 insertions(+), 1 deletion(-)target/i386/cpu.c +12
-0@@ -4725,6 +4725,29 @@ static void x86_hv_stimer_direct_set(Object *obj, bool value, Error **errp)x86_hv_feature_set (obj, value,HYPERV_FEAT_STIMER_DIRECT ); } +static bool x86_hv_passthrough_get (Object *obj,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + +return cpu->hyperv_passthrough; +} + +static void x86_hv_passthrough_set (Object *obj,bool value,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + + cpu->hyperv_passthrough = value;
Quoted 9 lines
target/i386/cpu.c static void x86_hv_stimer_direct_set(Object *obj, bool value, Error **errp) · line 4745
Quoted 6 lines
target/i386/cpu.c +6
-0+ +/* hv-passthrough overrides everything with what's supported by the host */ +if (value) { + cpu->hyperv_features =0 ; + cpu->hyperv_features_on =0 ; + cpu->hyperv_features_off =0 ;why do we have _on|_off fields?
You mean 'why do we have them at all' or 'why do we reset them here'? For the former, we need to distinguish between 'hv-passthroug,hv-feature=off' and just 'hv-passthrough'; 'hv-passthrough,hv-evmcs=on' and just 'hv-passthrough'. For the later, I'd like to make the samentics you've asked for:
'hv-feature,hv-passthrough' == 'hv-passthrough'
(though I still see it as a gotcha for an unprepared user)
Quoted 43 lines
Quoted 42 lines
target/i386/cpu.c +9
-1+ } + +return ; +} +/* Generic getter for "feature-words" and "filtered-features" properties */ static void x86_cpu_get_feature_words(Object *obj, Visitor *v, const char *name, void *opaque,@@ -7281,7 +7304,6 @@ static Property x86_cpu_properties[] = { HYPERV_SPINLOCK_NEVER_NOTIFY),DEFINE_PROP_ON_OFF_AUTO ("hv-no-nonarch-coresharing" ,X86CPU , hyperv_no_nonarch_cs,ON_OFF_AUTO_OFF ), -DEFINE_PROP_BOOL ("hv-passthrough" ,X86CPU , hyperv_passthrough, false),DEFINE_PROP_BOOL ("check" ,X86CPU , check_cpuid, true), DEFINE_PROP_BOOL("enforce", X86CPU, enforce_cpuid, false),@@ -7460,6 +7482,10 @@ static void x86_cpu_common_class_init(ObjectClass *oc, void *data) x86_hv_stimer_direct_get, x86_hv_stimer_direct_set); +object_class_property_add_bool (oc,"hv-passthrough" , + x86_hv_passthrough_get, + x86_hv_passthrough_set); +for (w =0 ; w <FEATURE_WORDS ; w++) {int bitnr;for (bitnr =0 ; bitnr <64 ; bitnr++) {target/i386/kvm/kvm.c +4
-0@@ -1153,6 +1153,10 @@ static int hv_cpuid_check_and_set(CPUState *cs, int feature, Error **errp)return 0 ; } +if (cpu->hyperv_passthrough && (cpu->hyperv_features_off &BIT (feature))) { +return 0 ; + } + deps = kvm_hyperv_properties[feature].dependencies;while (deps) { dep_feat =ctz64 (deps);
Signature · 1 line
Vitaly
Re: [PATCH v4 17/21] i386: support 'hv-passthrough, hv-feature=off' on the command line
Quoted 51 lines · Vitaly Kuznetsov
Quoted 39 lines · Igor Mammedov
Quoted 36 lines · On Wed, 10 Feb 2021 17:40:29 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Currently, we support 'hv-passthrough,hv-feature=on' enablement, this is supposed to mean "hv-feature is mandatory, don't start without it". Add support for 'hv-passthrough,hv-feature=off' meaning "enable everything supported by the host except for hv-feature".
While on it, make 'hv-passthrough' parse semantics in-line with other options in qemu: when specified, it overrides what was previously set with what's supported by the host. This can later be modified with 'hv-feature=on'/ 'hv-feature=off'.
Signed-off-by Vitaly Kuznetsov <[email protected]>--- target/i386/cpu.c | 28 +++++++++++++++++++++++++++- target/i386/kvm/kvm.c | 4 ++++ 2 files changed, 31 insertions(+), 1 deletion(-)target/i386/cpu.c +12
-0@@ -4725,6 +4725,29 @@ static void x86_hv_stimer_direct_set(Object *obj, bool value, Error **errp)x86_hv_feature_set (obj, value,HYPERV_FEAT_STIMER_DIRECT ); } +static bool x86_hv_passthrough_get (Object *obj,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + +return cpu->hyperv_passthrough; +} + +static void x86_hv_passthrough_set (Object *obj,bool value,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + + cpu->hyperv_passthrough = value;Quoted 9 lines
target/i386/cpu.c static void x86_hv_stimer_direct_set(Object *obj, bool value, Error **errp) · line 4745
Quoted 6 lines
target/i386/cpu.c +6
-0+ +/* hv-passthrough overrides everything with what's supported by the host */ +if (value) { + cpu->hyperv_features =0 ; + cpu->hyperv_features_on =0 ; + cpu->hyperv_features_off =0 ;why do we have _on|_off fields?
Quoted 7 lines
You mean 'why do we have them at all' or 'why do we reset them here'? For the former, we need to distinguish between 'hv-passthroug,hv-feature=off' and just 'hv-passthrough'; 'hv-passthrough,hv-evmcs=on' and just 'hv-passthrough'. For the later, I'd like to make the samentics you've asked for:
'hv-feature,hv-passthrough' == 'hv-passthrough'
(though I still see it as a gotcha for an unprepared user)
Either approach will confuse *someone*, I think.
This way at least behaves better if someone/something is composing the feature strings via concatenation.
Quoted 35 lines
Quoted 35 lines
Quoted 35 lines
target/i386/cpu.c +9
-1+ } + +return ; +} +/* Generic getter for "feature-words" and "filtered-features" properties */ static void x86_cpu_get_feature_words(Object *obj, Visitor *v, const char *name, void *opaque,@@ -7281,7 +7304,6 @@ static Property x86_cpu_properties[] = { HYPERV_SPINLOCK_NEVER_NOTIFY),DEFINE_PROP_ON_OFF_AUTO ("hv-no-nonarch-coresharing" ,X86CPU , hyperv_no_nonarch_cs,ON_OFF_AUTO_OFF ), -DEFINE_PROP_BOOL ("hv-passthrough" ,X86CPU , hyperv_passthrough, false),DEFINE_PROP_BOOL ("check" ,X86CPU , check_cpuid, true), DEFINE_PROP_BOOL("enforce", X86CPU, enforce_cpuid, false),@@ -7460,6 +7482,10 @@ static void x86_cpu_common_class_init(ObjectClass *oc, void *data) x86_hv_stimer_direct_get, x86_hv_stimer_direct_set); +object_class_property_add_bool (oc,"hv-passthrough" , + x86_hv_passthrough_get, + x86_hv_passthrough_set); +for (w =0 ; w <FEATURE_WORDS ; w++) {int bitnr;for (bitnr =0 ; bitnr <64 ; bitnr++) {target/i386/kvm/kvm.c +0
-0@@ -1153,6 +1153,10 @@ static int hv_cpuid_check_and_set(CPUState *cs, int feature, Error **errp)return 0 ; }
Quoted 11 lines
Quoted 8 lines
Quoted 7 lines
target/i386/kvm/kvm.c +4
-0+if (cpu->hyperv_passthrough && (cpu->hyperv_features_off &BIT (feature))) { +return 0 ; + } + deps = kvm_hyperv_properties[feature].dependencies;while (deps) { dep_feat =ctz64 (deps);Signature · 1 line
Vitaly
dme.
Signature · 1 line
J'aurais toujours faim de toi.
Re: [PATCH v4 17/21] i386: support 'hv-passthrough, hv-feature=off' on the command line
In reply to Vitaly KuznetsovQuoted 51 lines · On Fri, 12 Feb 2021 09:49:46 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Quoted 39 lines · Igor Mammedov
Quoted 36 lines · On Wed, 10 Feb 2021 17:40:29 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Currently, we support 'hv-passthrough,hv-feature=on' enablement, this is supposed to mean "hv-feature is mandatory, don't start without it". Add support for 'hv-passthrough,hv-feature=off' meaning "enable everything supported by the host except for hv-feature".
While on it, make 'hv-passthrough' parse semantics in-line with other options in qemu: when specified, it overrides what was previously set with what's supported by the host. This can later be modified with 'hv-feature=on'/ 'hv-feature=off'.
Signed-off-by Vitaly Kuznetsov <[email protected]>--- target/i386/cpu.c | 28 +++++++++++++++++++++++++++- target/i386/kvm/kvm.c | 4 ++++ 2 files changed, 31 insertions(+), 1 deletion(-)target/i386/cpu.c +12
-0@@ -4725,6 +4725,29 @@ static void x86_hv_stimer_direct_set(Object *obj, bool value, Error **errp)x86_hv_feature_set (obj, value,HYPERV_FEAT_STIMER_DIRECT ); } +static bool x86_hv_passthrough_get (Object *obj,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + +return cpu->hyperv_passthrough; +} + +static void x86_hv_passthrough_set (Object *obj,bool value,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + + cpu->hyperv_passthrough = value;Quoted 9 lines
target/i386/cpu.c static void x86_hv_stimer_direct_set(Object *obj, bool value, Error **errp) · line 4745
Quoted 6 lines
target/i386/cpu.c +6
-0+ +/* hv-passthrough overrides everything with what's supported by the host */ +if (value) { + cpu->hyperv_features =0 ; + cpu->hyperv_features_on =0 ; + cpu->hyperv_features_off =0 ;why do we have _on|_off fields?
Quoted 4 lines
You mean 'why do we have them at all' or 'why do we reset them here'? For the former, we need to distinguish between 'hv-passthroug,hv-feature=off' and just 'hv-passthrough'; 'hv-passthrough,hv-evmcs=on' and just 'hv-passthrough'. For the later,
that's what I was asking for, eventually I found it being buried in kvm code
Quoted 2 lines
I'd like to make the samentics you've asked for:
'hv-feature,hv-passthrough' == 'hv-passthrough'
you essentially you wrote your own hv-foo parser in kvm_hyperv_expand_features(), which is a bit complicated for my taste.
With scratch CPU you can simplify and make it easier to read.
Quoted 1 line
(though I still see it as a gotcha for an unprepared user)
this way it at least works the same way like any other property.
Quoted 44 lines
Quoted 43 lines
Quoted 42 lines
target/i386/cpu.c +9
-1+ } + +return ; +} +/* Generic getter for "feature-words" and "filtered-features" properties */ static void x86_cpu_get_feature_words(Object *obj, Visitor *v, const char *name, void *opaque,@@ -7281,7 +7304,6 @@ static Property x86_cpu_properties[] = { HYPERV_SPINLOCK_NEVER_NOTIFY),DEFINE_PROP_ON_OFF_AUTO ("hv-no-nonarch-coresharing" ,X86CPU , hyperv_no_nonarch_cs,ON_OFF_AUTO_OFF ), -DEFINE_PROP_BOOL ("hv-passthrough" ,X86CPU , hyperv_passthrough, false),DEFINE_PROP_BOOL ("check" ,X86CPU , check_cpuid, true), DEFINE_PROP_BOOL("enforce", X86CPU, enforce_cpuid, false),@@ -7460,6 +7482,10 @@ static void x86_cpu_common_class_init(ObjectClass *oc, void *data) x86_hv_stimer_direct_get, x86_hv_stimer_direct_set); +object_class_property_add_bool (oc,"hv-passthrough" , + x86_hv_passthrough_get, + x86_hv_passthrough_set); +for (w =0 ; w <FEATURE_WORDS ; w++) {int bitnr;for (bitnr =0 ; bitnr <64 ; bitnr++) {target/i386/kvm/kvm.c +4
-0@@ -1153,6 +1153,10 @@ static int hv_cpuid_check_and_set(CPUState *cs, int feature, Error **errp)return 0 ; } +if (cpu->hyperv_passthrough && (cpu->hyperv_features_off &BIT (feature))) { +return 0 ; + } + deps = kvm_hyperv_properties[feature].dependencies;while (deps) { dep_feat =ctz64 (deps);
PATCH v4 18/21 i386: be more picky about implicit 'hv-evmcs' enablement
In reply to Vitaly KuznetsovEnlightened VMCS is the only (currently implemented in QEMU) Hyper-V feature with hardware dependencies, it pairs with Intel VMX. It doesn't seem right to enable this feature when VMX wasn't enabled in the guest and when it wasn't explicitly requested on the command line. Currently, the only possible scenario is 'hv-passthrough' which will enable 'hv-evmcs' when the host supports it, regardless of guest VMX exposure. The upcoming 'hv-default' should also avoid enabling 'hv-evmcs' without VMX.
--- target/i386/kvm/kvm.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-)
target/i386/kvm/kvm.c +11 -2
@@ -1315,8 +1315,17 @@ void kvm_hyperv_expand_features(X86CPU *cpu, Error **errp)if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_TLBFLUSH , errp)) {return ; } -if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_EVMCS , errp)) { -return ; +/* +* 'hv-evmcs' is not enabled when it wasn't explicitly requested and guest +* CPU lacks VMX. +*/ +if (cpu_has_vmx (&cpu->env) || + (cpu->hyperv_features_on &BIT (HYPERV_FEAT_EVMCS ))) { +if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_EVMCS , errp)) { +return ; + } + }else { + cpu->hyperv_features &= ~BIT (HYPERV_FEAT_EVMCS ); }if (hv_cpuid_check_and_set (cs,HYPERV_FEAT_IPI , errp)) {return ;
Signature · 1 line
2.29.2
PATCH v4 19/21 i386: introduce kvm_hv_evmcs_available()
In reply to Vitaly KuznetsovEnlightened VMCS feature is hardware specific, it is only supported on Intel CPUs. Introduce a simple kvm_hv_evmcs_available() helper, it will be used to filter out 'hv_evmcs' when 'hyperv=on' option is added to X86MachineClass.
--- target/i386/kvm/kvm-stub.c | 5 +++++ target/i386/kvm/kvm.c | 8 ++++++++ target/i386/kvm/kvm_i386.h | 1 + 3 files changed, 14 insertions(+)
target/i386/kvm/kvm-stub.c +5 -0
@@ -44,3 +44,8 @@ void kvm_hyperv_expand_features(X86CPU *cpu, Error **errp) {return ; } + +bool kvm_hv_evmcs_available (void ) +{ +return false; +}
target/i386/kvm/kvm.c +8 -0
@@ -96,6 +96,7 @@ static bool has_msr_hv_crash;static bool has_msr_hv_reset;static bool has_msr_hv_vpindex;static bool hv_vpindex_settable; +static bool hv_evmcs_available;static bool has_msr_hv_runtime;static bool has_msr_hv_synic;static bool has_msr_hv_stimer;
@@ -195,6 +196,11 @@ bool kvm_hv_vpindex_settable(void)return hv_vpindex_settable; } +bool kvm_hv_evmcs_available (void ) +{ +return hv_evmcs_available; +} +static int kvm_get_tsc (CPUState *cs) {X86CPU *cpu =X86_CPU (cs);
@@ -2235,6 +2241,8 @@ int kvm_arch_init(MachineState *ms, KVMState *s) has_pit_state2 =kvm_check_extension (s,KVM_CAP_PIT_STATE2 ); hv_vpindex_settable =kvm_check_extension (s,KVM_CAP_HYPERV_VP_INDEX ); + hv_evmcs_available = +kvm_check_extension (s,KVM_CAP_HYPERV_ENLIGHTENED_VMCS ); has_exception_payload =kvm_check_extension (s,KVM_CAP_EXCEPTION_PAYLOAD );if (has_exception_payload) {
target/i386/kvm/kvm_i386.h +1 -0
@@ -48,6 +48,7 @@ bool kvm_has_waitpkg(void);bool kvm_hv_vpindex_settable (void );void kvm_hyperv_expand_features (X86CPU *cpu,Error **errp); +bool kvm_hv_evmcs_available (void );uint64_t kvm_swizzle_msi_ext_dest_id (uint64_t address);
Signature · 1 line
2.29.2
Quoted 27 lines · Vitaly Kuznetsov
Changes since v3:
- Make 'hv-default' override 'hv-*' options which were already set (e.g. 'hv-feature=on,hv-default' case) [Igor]. Make 'hv-passthrough' behave the same way.
- Add "i386: be more picky about implicit 'hv-evmcs' enablement" patch to avoid enabling 'hv-evmcs' with hv-default/hv-passthrough when guest CPU lacks VMX.
- Add "i386: support 'hv-passthrough,hv-feature=off' on the command line" patch to make 'hv-passthrough' semantics match the newly introduced 'hv-default'.
- Add "i386: track explicit 'hv-*' features enablement/disablement" patch to support the above mentioned changes.
- Expand qtest to check the above mentioned improvements.Original description:
Upper layer tools like libvirt want to figure out which Hyper-V features are supported by the underlying stack (QEMU/KVM) but currently they are unable to do so. We have a nice '
hv_passthrough' CPU flag supported by QEMU but it has no effect on e.g. QMP'squery-cpu-model-expansion type=full model={"name":"host","props":{"hv-passthrough":true}}
command as we parse Hyper-V features after creating KVM vCPUs and not at feature expansion time. To support the use-case we first need to make
KVM_GET_SUPPORTED_HV_CPUIDioctl a system-wide ioctl as the existing vCPU version can't be used that early. This is what KVM part does. With that done, we can make early Hyper-V feature expansion (this series).
Quoted 3 lines
In addition, provide a simple 'hv-default' option which enables (and requires from KVM) all currently supported Hyper-V enlightenments. Unlike 'hv-passthrough' mode, this is going to be migratable.
How is it going to be migratable if the semantics vary depending on the host kernel KVM reporting features, because different kernels will expose different features ?
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Quoted 28 lines · Daniel P. Berrangé
Quoted 27 lines · Vitaly Kuznetsov
Changes since v3:
- Make 'hv-default' override 'hv-*' options which were already set (e.g. 'hv-feature=on,hv-default' case) [Igor]. Make 'hv-passthrough' behave the same way.
- Add "i386: be more picky about implicit 'hv-evmcs' enablement" patch to avoid enabling 'hv-evmcs' with hv-default/hv-passthrough when guest CPU lacks VMX.
- Add "i386: support 'hv-passthrough,hv-feature=off' on the command line" patch to make 'hv-passthrough' semantics match the newly introduced 'hv-default'.
- Add "i386: track explicit 'hv-*' features enablement/disablement" patch to support the above mentioned changes.
- Expand qtest to check the above mentioned improvements.Original description:
Upper layer tools like libvirt want to figure out which Hyper-V features are supported by the underlying stack (QEMU/KVM) but currently they are unable to do so. We have a nice '
hv_passthrough' CPU flag supported by QEMU but it has no effect on e.g. QMP'squery-cpu-model-expansion type=full model={"name":"host","props":{"hv-passthrough":true}}
command as we parse Hyper-V features after creating KVM vCPUs and not at feature expansion time. To support the use-case we first need to make
KVM_GET_SUPPORTED_HV_CPUIDioctl a system-wide ioctl as the existing vCPU version can't be used that early. This is what KVM part does. With that done, we can make early Hyper-V feature expansion (this series).
Quoted 7 lines
Quoted 3 lines
In addition, provide a simple 'hv-default' option which enables (and requires from KVM) all currently supported Hyper-V enlightenments. Unlike 'hv-passthrough' mode, this is going to be migratable.
How is it going to be migratable if the semantics vary depending on the host kernel KVM reporting features, because different kernels will expose different features ?
"all currently supported" in this context means "all features supported when the machine type was added", not "all features supported by the host kernel".
Signature · 1 line
Eduardo
Quoted 33 lines · Eduardo Habkost
Quoted 32 lines · Daniel P. Berrangé
Quoted 30 lines · Vitaly Kuznetsov
Changes since v3:
- Make 'hv-default' override 'hv-*' options which were already set (e.g. 'hv-feature=on,hv-default' case) [Igor]. Make 'hv-passthrough' behave the same way.
- Add "i386: be more picky about implicit 'hv-evmcs' enablement" patch to avoid enabling 'hv-evmcs' with hv-default/hv-passthrough when guest CPU lacks VMX.
- Add "i386: support 'hv-passthrough,hv-feature=off' on the command line" patch to make 'hv-passthrough' semantics match the newly introduced 'hv-default'.
- Add "i386: track explicit 'hv-*' features enablement/disablement" patch to support the above mentioned changes.
- Expand qtest to check the above mentioned improvements.Original description:
Upper layer tools like libvirt want to figure out which Hyper-V features are supported by the underlying stack (QEMU/KVM) but currently they are unable to do so. We have a nice '
hv_passthrough' CPU flag supported by QEMU but it has no effect on e.g. QMP'squery-cpu-model-expansion type=full model={"name":"host","props":{"hv-passthrough":true}}
command as we parse Hyper-V features after creating KVM vCPUs and not at feature expansion time. To support the use-case we first need to make
KVM_GET_SUPPORTED_HV_CPUIDioctl a system-wide ioctl as the existing vCPU version can't be used that early. This is what KVM part does. With that done, we can make early Hyper-V feature expansion (this series).In addition, provide a simple 'hv-default' option which enables (and requires from KVM) all currently supported Hyper-V enlightenments. Unlike 'hv-passthrough' mode, this is going to be migratable.
Quoted 7 lines
Quoted 3 lines
How is it going to be migratable if the semantics vary depending on the host kernel KVM reporting features, because different kernels will expose different features ?
"all currently supported" in this context means "all features supported when the machine type was added", not "all features supported by the host kernel".
Yes, exactly.
'hv-passthrough' enables 'everything supported by the host' and this is not migratable.
'hv-default' requires a certain set of features (depending on the machine type) so the VM won't start if the host lacks something.
Signature · 1 line
Vitaly
Quoted 43 lines · Vitaly Kuznetsov
Quoted 33 lines · Eduardo Habkost
Quoted 32 lines · Daniel P. Berrangé
Quoted 30 lines · Vitaly Kuznetsov
Changes since v3:
- Make 'hv-default' override 'hv-*' options which were already set (e.g. 'hv-feature=on,hv-default' case) [Igor]. Make 'hv-passthrough' behave the same way.
- Add "i386: be more picky about implicit 'hv-evmcs' enablement" patch to avoid enabling 'hv-evmcs' with hv-default/hv-passthrough when guest CPU lacks VMX.
- Add "i386: support 'hv-passthrough,hv-feature=off' on the command line" patch to make 'hv-passthrough' semantics match the newly introduced 'hv-default'.
- Add "i386: track explicit 'hv-*' features enablement/disablement" patch to support the above mentioned changes.
- Expand qtest to check the above mentioned improvements.Original description:
Upper layer tools like libvirt want to figure out which Hyper-V features are supported by the underlying stack (QEMU/KVM) but currently they are unable to do so. We have a nice '
hv_passthrough' CPU flag supported by QEMU but it has no effect on e.g. QMP'squery-cpu-model-expansion type=full model={"name":"host","props":{"hv-passthrough":true}}
command as we parse Hyper-V features after creating KVM vCPUs and not at feature expansion time. To support the use-case we first need to make
KVM_GET_SUPPORTED_HV_CPUIDioctl a system-wide ioctl as the existing vCPU version can't be used that early. This is what KVM part does. With that done, we can make early Hyper-V feature expansion (this series).In addition, provide a simple 'hv-default' option which enables (and requires from KVM) all currently supported Hyper-V enlightenments. Unlike 'hv-passthrough' mode, this is going to be migratable.
Quoted 7 lines
Quoted 3 lines
How is it going to be migratable if the semantics vary depending on the host kernel KVM reporting features, because different kernels will expose different features ?
"all currently supported" in this context means "all features supported when the machine type was added", not "all features supported by the host kernel".
Quoted 7 lines
Yes, exactly.
'hv-passthrough' enables 'everything supported by the host' and this is not migratable.
'hv-default' requires a certain set of features (depending on the machine type) so the VM won't start if the host lacks something.
Ok, so I presume HV features will only be added to hv-default when we know they are available in the oldest kernel we are targetting ? Upsteam is more conservative in this respect than downstreams, the latter can guarantee much more modern kernels.
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Quoted 49 lines · Daniel P. Berrangé
Quoted 48 lines · Vitaly Kuznetsov
Quoted 33 lines · Eduardo Habkost
Quoted 32 lines · Daniel P. Berrangé
Quoted 30 lines · Vitaly Kuznetsov
Changes since v3:
- Make 'hv-default' override 'hv-*' options which were already set (e.g. 'hv-feature=on,hv-default' case) [Igor]. Make 'hv-passthrough' behave the same way.
- Add "i386: be more picky about implicit 'hv-evmcs' enablement" patch to avoid enabling 'hv-evmcs' with hv-default/hv-passthrough when guest CPU lacks VMX.
- Add "i386: support 'hv-passthrough,hv-feature=off' on the command line" patch to make 'hv-passthrough' semantics match the newly introduced 'hv-default'.
- Add "i386: track explicit 'hv-*' features enablement/disablement" patch to support the above mentioned changes.
- Expand qtest to check the above mentioned improvements.Original description:
Upper layer tools like libvirt want to figure out which Hyper-V features are supported by the underlying stack (QEMU/KVM) but currently they are unable to do so. We have a nice '
hv_passthrough' CPU flag supported by QEMU but it has no effect on e.g. QMP'squery-cpu-model-expansion type=full model={"name":"host","props":{"hv-passthrough":true}}
command as we parse Hyper-V features after creating KVM vCPUs and not at feature expansion time. To support the use-case we first need to make
KVM_GET_SUPPORTED_HV_CPUIDioctl a system-wide ioctl as the existing vCPU version can't be used that early. This is what KVM part does. With that done, we can make early Hyper-V feature expansion (this series).In addition, provide a simple 'hv-default' option which enables (and requires from KVM) all currently supported Hyper-V enlightenments. Unlike 'hv-passthrough' mode, this is going to be migratable.
Quoted 7 lines
Quoted 3 lines
How is it going to be migratable if the semantics vary depending on the host kernel KVM reporting features, because different kernels will expose different features ?
"all currently supported" in this context means "all features supported when the machine type was added", not "all features supported by the host kernel".
Yes, exactly.
'hv-passthrough' enables 'everything supported by the host' and this is not migratable.
Quoted 8 lines
Quoted 2 lines
'hv-default' requires a certain set of features (depending on the machine type) so the VM won't start if the host lacks something.
Ok, so I presume HV features will only be added to hv-default when we know they are available in the oldest kernel we are targetting ? Upsteam is more conservative in this respect than downstreams, the latter can guarantee much more modern kernels.
Yes, it is kind of an open question when a feature gets 'promoted' to 'hv-default'. Currently, the latest feature we include is 'HYPERV_FEAT_STIMER_DIRECT' which dates back to Linux 5.0. It is also possible to use something like
'hv-default,hv-stimer-direct=off,...'
when running on an older kernel (and this is still migratable).
Signature · 1 line
Vitaly
Quoted 60 lines · Vitaly Kuznetsov
Quoted 49 lines · Daniel P. Berrangé
Quoted 48 lines · Vitaly Kuznetsov
Quoted 33 lines · Eduardo Habkost
Quoted 32 lines · Daniel P. Berrangé
Quoted 30 lines · Vitaly Kuznetsov
Changes since v3:
- Make 'hv-default' override 'hv-*' options which were already set (e.g. 'hv-feature=on,hv-default' case) [Igor]. Make 'hv-passthrough' behave the same way.
- Add "i386: be more picky about implicit 'hv-evmcs' enablement" patch to avoid enabling 'hv-evmcs' with hv-default/hv-passthrough when guest CPU lacks VMX.
- Add "i386: support 'hv-passthrough,hv-feature=off' on the command line" patch to make 'hv-passthrough' semantics match the newly introduced 'hv-default'.
- Add "i386: track explicit 'hv-*' features enablement/disablement" patch to support the above mentioned changes.
- Expand qtest to check the above mentioned improvements.Original description:
Upper layer tools like libvirt want to figure out which Hyper-V features are supported by the underlying stack (QEMU/KVM) but currently they are unable to do so. We have a nice '
hv_passthrough' CPU flag supported by QEMU but it has no effect on e.g. QMP'squery-cpu-model-expansion type=full model={"name":"host","props":{"hv-passthrough":true}}
command as we parse Hyper-V features after creating KVM vCPUs and not at feature expansion time. To support the use-case we first need to make
KVM_GET_SUPPORTED_HV_CPUIDioctl a system-wide ioctl as the existing vCPU version can't be used that early. This is what KVM part does. With that done, we can make early Hyper-V feature expansion (this series).In addition, provide a simple 'hv-default' option which enables (and requires from KVM) all currently supported Hyper-V enlightenments. Unlike 'hv-passthrough' mode, this is going to be migratable.
Quoted 7 lines
Quoted 3 lines
How is it going to be migratable if the semantics vary depending on the host kernel KVM reporting features, because different kernels will expose different features ?
"all currently supported" in this context means "all features supported when the machine type was added", not "all features supported by the host kernel".
Yes, exactly.
'hv-passthrough' enables 'everything supported by the host' and this is not migratable.
Quoted 8 lines
Quoted 2 lines
'hv-default' requires a certain set of features (depending on the machine type) so the VM won't start if the host lacks something.
Ok, so I presume HV features will only be added to hv-default when we know they are available in the oldest kernel we are targetting ? Upsteam is more conservative in this respect than downstreams, the latter can guarantee much more modern kernels.
Quoted 4 lines
Yes, it is kind of an open question when a feature gets 'promoted' to 'hv-default'. Currently, the latest feature we include is '
HYPERV_FEAT_STIMER_DIRECT' which dates back to Linux 5.0. It is also possible to use something like
Upstream we have a clear set of targetted OS platforms:
https://qemu.readthedocs.io/en/latest/system/build-platforms.html
This will inform what our minimum possible kernel version will be, which should influence when we can promote something.
Downstream's of course have their own min kernel model, so they can do things differently if they have their own machine types.
Quoted 3 lines
'hv-default,hv-stimer-direct=off,...'
when running on an older kernel (and this is still migratable).
At that point the mgmt app needs to know exactly what features the host supports, and if they know that, there's no real need to use hv-default in the first place. IOW, I think we need to strive to ensure "hv-default" is always usable on supported platforms without needing to know about turning off things.
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
PATCH v4 20/21 i386: provide simple 'hv-default=on' option
In reply to Vitaly KuznetsovEnabling Hyper-V emulation for a Windows VM is a tiring experience as it requires listing all currently supported enlightenments ("hv-*" CPU features) explicitly. We do have 'hv-passthrough' mode enabling everything but it can't be used in production as it prevents migration.
Introduce a simple 'hv-default=on' CPU flag enabling all currently supported Hyper-V enlightenments. Later, when new enlightenments get implemented, compat_props mechanism will be used to disable them for legacy machine types, this will keep 'hv-default=on' configurations migratable.
--- docs/hyperv.txt | 16 ++++++++++++--- target/i386/cpu.c | 52 +++++++++++++++++++++++++++++++++++++++++++++++ target/i386/cpu.h | 3 +++ 3 files changed, 68 insertions(+), 3 deletions(-)
docs/hyperv.txt +13 -3
@@ -17,10 +17,20 @@ compatible hypervisor and use Hyper-V specific features. 2. Setup ========= -No Hyper-V enlightenments are enabled by default by either KVM or QEMU. In -QEMU, individual enlightenments can be enabled through CPU flags, e.g: +All currently supported Hyper-V enlightenments can be enabled by specifying +'hv-default=on' CPU flag: - qemu-system-x86_64 --enable-kvm --cpu host,hv_relaxed,hv_vpindex,hv_time, ... + qemu-system-x86_64 --enable-kvm --cpu host,hv-default ... + +Alternatively, it is possible to do fine-grained enablement through CPU flags, +e.g: + + qemu-system-x86_64 --enable-kvm --cpu host,hv-relaxed,hv-vpindex,hv-time ... + +It is also possible to disable individual enlightenments from the default list, +this can be used for debugging purposes: + + qemu-system-x86_64 --enable-kvm --cpu host,hv-default=on,hv-evmcs=off ... Sometimes there are dependencies between enlightenments, QEMU is supposed to check that the supplied configuration is sane.
target/i386/cpu.c +52 -0
76 lines not on this page · raw message
target/i386/cpu.h +3 -0
@@ -1670,6 +1670,9 @@ struct X86CPU {uint64_t hyperv_features_on;uint64_t hyperv_features_off;bool hyperv_passthrough; +/* 'hv-default' enablement */ +uint64_t hyperv_default_features; +bool hyperv_default;OnOffAuto hyperv_no_nonarch_cs;uint32_t hyperv_vendor_id[3 ];uint32_t hyperv_interface_id[4 ];
Signature · 1 line
2.29.2
Re: [PATCH v4 20/21] i386: provide simple 'hv-default=on' option
Quoted 70 lines · On Wed, 10 Feb 2021 17:40:32 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Enabling Hyper-V emulation for a Windows VM is a tiring experience as it requires listing all currently supported enlightenments ("hv-*" CPU features) explicitly. We do have 'hv-passthrough' mode enabling everything but it can't be used in production as it prevents migration.
Introduce a simple 'hv-default=on' CPU flag enabling all currently supported Hyper-V enlightenments. Later, when new enlightenments get implemented,
compat_propsmechanism will be used to disable them for legacy machine types, this will keep 'hv-default=on' configurations migratable.Signed-off-by Vitaly Kuznetsov <[email protected]>--- docs/hyperv.txt | 16 ++++++++++++--- target/i386/cpu.c | 52 +++++++++++++++++++++++++++++++++++++++++++++++ target/i386/cpu.h | 3 +++ 3 files changed, 68 insertions(+), 3 deletions(-)docs/hyperv.txt +13
-3@@ -17,10 +17,20 @@ compatible hypervisor and use Hyper-V specific features. 2. Setup ========= -No Hyper-V enlightenments are enabled by default by either KVM or QEMU. In -QEMU, individual enlightenments can be enabled through CPU flags, e.g: +All currently supported Hyper-V enlightenments can be enabled by specifying +'hv-default=on' CPU flag: - qemu-system-x86_64 --enable-kvm --cpu host,hv_relaxed,hv_vpindex,hv_time, ... + qemu-system-x86_64 --enable-kvm --cpu host,hv-default ... + +Alternatively, it is possible to do fine-grained enablement through CPU flags, +e.g: + + qemu-system-x86_64 --enable-kvm --cpu host,hv-relaxed,hv-vpindex,hv-time ... + +It is also possible to disable individual enlightenments from the default list, +this can be used for debugging purposes: + + qemu-system-x86_64 --enable-kvm --cpu host,hv-default=on,hv-evmcs=off ... Sometimes there are dependencies between enlightenments, QEMU is supposed to check that the supplied configuration is sane.target/i386/cpu.c +10
-0@@ -4736,6 +4736,12 @@ static void x86_hv_passthrough_set(Object *obj, bool value, Error **errp) {X86CPU *cpu =X86_CPU (obj); +if (cpu->hyperv_default) { +error_setg (errp, +"'hv-default' and 'hv-paththrough' are mutually exclusive" ); +return ; + } + cpu->hyperv_passthrough = value;/* hv-passthrough overrides everything with what's supported by the host */ @@ -4748,6 +4754,33 @@ static void x86_hv_passthrough_set(Object *obj, bool value, Error **errp)return ; } +static bool x86_hv_default_get (Object *obj,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); +
target/i386/cpu.c static void x86_hv_passthrough_set(Object *obj, bool value, Error **errp) · line 4770
Quoted 10 lines
target/i386/cpu.c +10
-0+return cpu->hyperv_default; +} + +static void x86_hv_default_set (Object *obj,bool value,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + +if (cpu->hyperv_passthrough) { +error_setg (errp, +"'hv-default' and 'hv-paththrough' are mutually exclusive" );
this check will work only half way, i.e.: hv-paththrough=on,hv-default=on|off (where off value looks a bit wierd)
but not other way around: hv-default=on,hv-paththrough=on
were you thinking about following error:
"hv-default can't be used after hv-paththrough were enabled"
or if it symmetric, then putting this check in realizefn() will do the job as both properties are processed by that time.
Quoted 63 lines
target/i386/cpu.c +32
-0+return ; + } + + cpu->hyperv_default = value; + +/* hv-default overrides everything with the default set */ +if (value) { + cpu->hyperv_features = cpu->hyperv_default_features; + cpu->hyperv_features_on =0 ; + cpu->hyperv_features_off =0 ; + } +} +/* Generic getter for "feature-words" and "filtered-features" properties */ static void x86_cpu_get_feature_words(Object *obj, Visitor *v, const char *name, void *opaque,@@ -7152,6 +7185,21 @@ static void x86_cpu_initfn(Object *obj)if (xcc->model) {x86_cpu_load_model (cpu, xcc->model); } + +/* Hyper-V features enabled with 'hv-default=on' */ + cpu->hyperv_default_features =BIT (HYPERV_FEAT_RELAXED ) | +BIT (HYPERV_FEAT_VAPIC ) |BIT (HYPERV_FEAT_TIME ) | +BIT (HYPERV_FEAT_CRASH ) |BIT (HYPERV_FEAT_RESET ) | +BIT (HYPERV_FEAT_VPINDEX ) |BIT (HYPERV_FEAT_RUNTIME ) | +BIT (HYPERV_FEAT_SYNIC ) |BIT (HYPERV_FEAT_STIMER ) | +BIT (HYPERV_FEAT_FREQUENCIES ) |BIT (HYPERV_FEAT_REENLIGHTENMENT ) | +BIT (HYPERV_FEAT_TLBFLUSH ) |BIT (HYPERV_FEAT_IPI ) | +BIT (HYPERV_FEAT_STIMER_DIRECT ); + +/* Enlightened VMCS is only available on Intel/VMX */ +if (kvm_hv_evmcs_available ()) { + cpu->hyperv_default_features |=BIT (HYPERV_FEAT_EVMCS ); + } }static int64_t x86_cpu_get_arch_id (CPUState *cs)@@ -7486,6 +7534,10 @@ static void x86_cpu_common_class_init(ObjectClass *oc, void *data) x86_hv_passthrough_get, x86_hv_passthrough_set); +object_class_property_add_bool (oc,"hv-default" , + x86_hv_default_get, + x86_hv_default_set); +for (w =0 ; w <FEATURE_WORDS ; w++) {int bitnr;for (bitnr =0 ; bitnr <64 ; bitnr++) {target/i386/cpu.h +3
-0@@ -1670,6 +1670,9 @@ struct X86CPU {uint64_t hyperv_features_on;uint64_t hyperv_features_off;bool hyperv_passthrough; +/* 'hv-default' enablement */ +uint64_t hyperv_default_features; +bool hyperv_default;OnOffAuto hyperv_no_nonarch_cs;uint32_t hyperv_vendor_id[3 ];uint32_t hyperv_interface_id[4 ];
Re: [PATCH v4 20/21] i386: provide simple 'hv-default=on' option
Quoted 56 lines · Igor Mammedov
Quoted 53 lines · On Wed, 10 Feb 2021 17:40:32 +0100 Vitaly Kuznetsov <[email protected]> wrote:
Enabling Hyper-V emulation for a Windows VM is a tiring experience as it requires listing all currently supported enlightenments ("hv-*" CPU features) explicitly. We do have 'hv-passthrough' mode enabling everything but it can't be used in production as it prevents migration.
Introduce a simple 'hv-default=on' CPU flag enabling all currently supported Hyper-V enlightenments. Later, when new enlightenments get implemented,
compat_propsmechanism will be used to disable them for legacy machine types, this will keep 'hv-default=on' configurations migratable.Signed-off-by Vitaly Kuznetsov <[email protected]>--- docs/hyperv.txt | 16 ++++++++++++--- target/i386/cpu.c | 52 +++++++++++++++++++++++++++++++++++++++++++++++ target/i386/cpu.h | 3 +++ 3 files changed, 68 insertions(+), 3 deletions(-)docs/hyperv.txt +13
-3@@ -17,10 +17,20 @@ compatible hypervisor and use Hyper-V specific features. 2. Setup ========= -No Hyper-V enlightenments are enabled by default by either KVM or QEMU. In -QEMU, individual enlightenments can be enabled through CPU flags, e.g: +All currently supported Hyper-V enlightenments can be enabled by specifying +'hv-default=on' CPU flag: - qemu-system-x86_64 --enable-kvm --cpu host,hv_relaxed,hv_vpindex,hv_time, ... + qemu-system-x86_64 --enable-kvm --cpu host,hv-default ... + +Alternatively, it is possible to do fine-grained enablement through CPU flags, +e.g: + + qemu-system-x86_64 --enable-kvm --cpu host,hv-relaxed,hv-vpindex,hv-time ... + +It is also possible to disable individual enlightenments from the default list, +this can be used for debugging purposes: + + qemu-system-x86_64 --enable-kvm --cpu host,hv-default=on,hv-evmcs=off ... Sometimes there are dependencies between enlightenments, QEMU is supposed to check that the supplied configuration is sane.target/i386/cpu.c +0
-0@@ -4736,6 +4736,12 @@ static void x86_hv_passthrough_set(Object *obj, bool value, Error **errp) {X86CPU *cpu =X86_CPU (obj);
Quoted 1 line
target/i386/cpu.c static void x86_hv_passthrough_set(Object *obj, bool value, Error **errp) · line 4739
Quoted 1 line
target/i386/cpu.c +1
-0+if (cpu->hyperv_default) {
^^^ here ^^^
Quoted 21 lines
Quoted 21 lines
target/i386/cpu.c +14
-0+error_setg (errp, +"'hv-default' and 'hv-paththrough' are mutually exclusive" ); +return ; + } + cpu->hyperv_passthrough = value;/* hv-passthrough overrides everything with what's supported by the host */ @@ -4748,6 +4754,33 @@ static void x86_hv_passthrough_set(Object *obj, bool value, Error **errp)return ; } +static bool x86_hv_default_get (Object *obj,Error **errp) +{ +X86CPU *cpu =X86_CPU (obj); + +return cpu->hyperv_default; +} + +static void x86_hv_default_set (Object *obj,bool value,Error **errp) +{
Quoted 8 lines
target/i386/cpu.c static void x86_hv_passthrough_set(Object *obj, bool value, Error **errp) · line 4770
Quoted 5 lines
target/i386/cpu.c +5
-0+X86CPU *cpu =X86_CPU (obj); + +if (cpu->hyperv_passthrough) { +error_setg (errp, +"'hv-default' and 'hv-paththrough' are mutually exclusive" );this check will work only half way, i.e.: hv-paththrough=on,hv-default=on|off (where off value looks a bit wierd)
but not other way around: hv-default=on,hv-paththrough=on
The check in x86_hv_passthrough_set() chechs the opposite scenario.
Quoted 6 lines
were you thinking about following error:
"hv-default can't be used after hv-paththrough were enabled"or if it symmetric, then putting this check in
realizefn()will do the job as both properties are processed by that time.
I can move the check there but I think that two checks I add here cover what we need (and we don't need to care what to set 'hyperv_features' to in the interim).
Quoted 65 lines
Quoted 64 lines
target/i386/cpu.c +32
-0+return ; + } + + cpu->hyperv_default = value; + +/* hv-default overrides everything with the default set */ +if (value) { + cpu->hyperv_features = cpu->hyperv_default_features; + cpu->hyperv_features_on =0 ; + cpu->hyperv_features_off =0 ; + } +} +/* Generic getter for "feature-words" and "filtered-features" properties */ static void x86_cpu_get_feature_words(Object *obj, Visitor *v, const char *name, void *opaque,@@ -7152,6 +7185,21 @@ static void x86_cpu_initfn(Object *obj)if (xcc->model) {x86_cpu_load_model (cpu, xcc->model); } + +/* Hyper-V features enabled with 'hv-default=on' */ + cpu->hyperv_default_features =BIT (HYPERV_FEAT_RELAXED ) | +BIT (HYPERV_FEAT_VAPIC ) |BIT (HYPERV_FEAT_TIME ) | +BIT (HYPERV_FEAT_CRASH ) |BIT (HYPERV_FEAT_RESET ) | +BIT (HYPERV_FEAT_VPINDEX ) |BIT (HYPERV_FEAT_RUNTIME ) | +BIT (HYPERV_FEAT_SYNIC ) |BIT (HYPERV_FEAT_STIMER ) | +BIT (HYPERV_FEAT_FREQUENCIES ) |BIT (HYPERV_FEAT_REENLIGHTENMENT ) | +BIT (HYPERV_FEAT_TLBFLUSH ) |BIT (HYPERV_FEAT_IPI ) | +BIT (HYPERV_FEAT_STIMER_DIRECT ); + +/* Enlightened VMCS is only available on Intel/VMX */ +if (kvm_hv_evmcs_available ()) { + cpu->hyperv_default_features |=BIT (HYPERV_FEAT_EVMCS ); + } }static int64_t x86_cpu_get_arch_id (CPUState *cs)@@ -7486,6 +7534,10 @@ static void x86_cpu_common_class_init(ObjectClass *oc, void *data) x86_hv_passthrough_get, x86_hv_passthrough_set); +object_class_property_add_bool (oc,"hv-default" , + x86_hv_default_get, + x86_hv_default_set); +for (w =0 ; w <FEATURE_WORDS ; w++) {int bitnr;for (bitnr =0 ; bitnr <64 ; bitnr++) {target/i386/cpu.h +3
-0@@ -1670,6 +1670,9 @@ struct X86CPU {uint64_t hyperv_features_on;uint64_t hyperv_features_off;bool hyperv_passthrough; +/* 'hv-default' enablement */ +uint64_t hyperv_default_features; +bool hyperv_default;OnOffAuto hyperv_no_nonarch_cs;uint32_t hyperv_vendor_id[3 ];uint32_t hyperv_interface_id[4 ];
Signature · 1 line
Vitaly
PATCH v4 21/21 qtest/hyperv: Introduce a simple hyper-v test
In reply to Vitaly KuznetsovFor the beginning, just test 'hv-default', 'hv-passthrough' and a couple of custom Hyper-V enlightenments configurations through QMP. Later, it would be great to complement this by checking CPUID values from within the guest.
--- MAINTAINERS | 1 + tests/qtest/hyperv-test.c | 312 ++++++++++++++++++++++++++++++++++++++ tests/qtest/meson.build | 3 +- 3 files changed, 315 insertions(+), 1 deletion(-) create mode 100644 tests/qtest/hyperv-test.c
MAINTAINERS +1 -0
@@ -1505,6 +1505,7 @@ F: hw/isa/apm.c F: include/hw/isa/apm.h F: tests/test-x86-cpuid.c F: tests/qtest/test-x86-cpuid-compat.c +F: tests/qtest/hyperv-test.c PC Chipset M: Michael S. Tsirkin <[email protected]>
tests/qtest/hyperv-test.c new file +312 -0
312 lines not on this page · raw message
tests/qtest/meson.build +2 -1
@@ -65,7 +65,8 @@ qtests_i386 = \ 'vmgenid-test', 'migration-test', 'test-x86-cpuid-compat', - 'numa-test'] + 'numa-test', + 'hyperv-test'] dbus_daemon = find_program('dbus-daemon', required: false) if dbus_daemon.found() and config_host.has_key('GDBUS_CODEGEN')
Signature · 1 line
2.29.2