| Message ID | 20180921082217.29481-4-rkagan@virtuozzo.com |
|---|---|
| State | New |
| Headers | show |
| Series | hyperv: add connection infrastructure | expand |
On Fri, Sep 21, 2018 at 11:22:10AM +0300, Roman Kagan wrote: > Certain configurations do not allow SynIC to be used in QEMU. In > particular, > > - when hyperv_vpindex is off, SINT routes can't be used as they refer to > the destination vCPU by vp_index > > - older KVM (which doesn't expose KVM_CAP_HYPERV_SYNIC2) zeroes out > SynIC message and event pages on every msr load, breaking migration > > OTOH in-KVM users of SynIC -- SynIC timers -- do work in those > configurations, and we shouldn't stop the guest from using them. > > To cover both scenarios, introduce an X86CPU property that makes CPU > init code to skip creation of the SynIC object (and thus disables any > SynIC use in QEMU) but keeps the KVM part of the SynIC working. > The property is clear by default but is set via compat logic for older > machine types. > > As a result, when hv_synic and a modern machine type are specified, QEMU > will refuse to run unless vp_index is on and the kernel is recent > enough. OTOH with an older machine type QEMU will run fine with > hv_synic=on against an older kernel and/or without vp_index enabled but > will disallow the in-QEMU uses of SynIC (in e.g. VMBus). > > Signed-off-by: Roman Kagan <rkagan@virtuozzo.com> > --- > include/hw/i386/pc.h | 8 ++++++++ > target/i386/cpu.h | 1 + > target/i386/cpu.c | 2 ++ > target/i386/kvm.c | 30 ++++++++++++++++++++++-------- > 4 files changed, 33 insertions(+), 8 deletions(-) > > diff --git a/include/hw/i386/pc.h b/include/hw/i386/pc.h > index 6894f37df1..dfe6746692 100644 > --- a/include/hw/i386/pc.h > +++ b/include/hw/i386/pc.h > @@ -294,6 +294,14 @@ int e820_add_entry(uint64_t, uint64_t, uint32_t); > int e820_get_num_entries(void); > bool e820_get_entry(int, uint32_t, uint64_t *, uint64_t *); > > +#define PC_COMPAT_3_0 \ > + HW_COMPAT_3_0 \ > + {\ > + .driver = TYPE_X86_CPU,\ > + .property = "x-hv-synic-kvm-only",\ > + .value = "on",\ > + } Oops. This macro does nothing if the 3.1 machine-types aren't updated to use it. This patch breaks compatibility on pc-*-3.1. > [...]
On Fri, 21 Sep 2018 11:22:10 +0300 Roman Kagan <rkagan@virtuozzo.com> wrote: > Certain configurations do not allow SynIC to be used in QEMU. In > particular, > > - when hyperv_vpindex is off, SINT routes can't be used as they refer to > the destination vCPU by vp_index > > - older KVM (which doesn't expose KVM_CAP_HYPERV_SYNIC2) zeroes out > SynIC message and event pages on every msr load, breaking migration > > OTOH in-KVM users of SynIC -- SynIC timers -- do work in those > configurations, and we shouldn't stop the guest from using them. > > To cover both scenarios, introduce an X86CPU property that makes CPU > init code to skip creation of the SynIC object (and thus disables any > SynIC use in QEMU) but keeps the KVM part of the SynIC working. > The property is clear by default but is set via compat logic for older > machine types. > > As a result, when hv_synic and a modern machine type are specified, QEMU > will refuse to run unless vp_index is on and the kernel is recent > enough. OTOH with an older machine type QEMU will run fine with > hv_synic=on against an older kernel and/or without vp_index enabled but > will disallow the in-QEMU uses of SynIC (in e.g. VMBus). > > Signed-off-by: Roman Kagan <rkagan@virtuozzo.com> With current upstream and x-hv-synic-kvm-only=on QEMU will SIGSEGV. Problem was unnoticed since added compat property wasn't actually used until much later commit 4a93722f9c hw/i386: add pc-i440fx-3.1 & pc-q35-3.1 which put compat property in use. qemu-system-x86_64 -machine pc-i440fx-2.10,accel=kvm \ -cpu host,-vmx,hv-relaxed,hv_spinlocks=0x1fff,hv-vpindex,hv-synic simpler reproducer: qemu-system-x86_64 -enable-kvm -cpu host,hv-synic,x-hv-synic-kvm-only=on Reported-by: Vitaly Kuznetsov <vkuznets@redhat.com> > --- > include/hw/i386/pc.h | 8 ++++++++ > target/i386/cpu.h | 1 + > target/i386/cpu.c | 2 ++ > target/i386/kvm.c | 30 ++++++++++++++++++++++-------- > 4 files changed, 33 insertions(+), 8 deletions(-) > > diff --git a/include/hw/i386/pc.h b/include/hw/i386/pc.h > index 6894f37df1..dfe6746692 100644 > --- a/include/hw/i386/pc.h > +++ b/include/hw/i386/pc.h > @@ -294,6 +294,14 @@ int e820_add_entry(uint64_t, uint64_t, uint32_t); > int e820_get_num_entries(void); > bool e820_get_entry(int, uint32_t, uint64_t *, uint64_t *); > > +#define PC_COMPAT_3_0 \ > + HW_COMPAT_3_0 \ > + {\ > + .driver = TYPE_X86_CPU,\ > + .property = "x-hv-synic-kvm-only",\ > + .value = "on",\ > + } > + > #define PC_COMPAT_2_12 \ > HW_COMPAT_2_12 \ > {\ > diff --git a/target/i386/cpu.h b/target/i386/cpu.h > index b572a8e4aa..e2e87dc13f 100644 > --- a/target/i386/cpu.h > +++ b/target/i386/cpu.h > @@ -1376,6 +1376,7 @@ struct X86CPU { > bool hyperv_vpindex; > bool hyperv_runtime; > bool hyperv_synic; > + bool hyperv_synic_kvm_only; > bool hyperv_stimer; > bool hyperv_frequencies; > bool hyperv_reenlightenment; > diff --git a/target/i386/cpu.c b/target/i386/cpu.c > index f24295e6e4..9c29c5db81 100644 > --- a/target/i386/cpu.c > +++ b/target/i386/cpu.c > @@ -5575,6 +5575,8 @@ static Property x86_cpu_properties[] = { > * to the specific Windows version being used." > */ > DEFINE_PROP_INT32("x-hv-max-vps", X86CPU, hv_max_vps, -1), > + DEFINE_PROP_BOOL("x-hv-synic-kvm-only", X86CPU, hyperv_synic_kvm_only, > + false), > DEFINE_PROP_END_OF_LIST() > }; > > diff --git a/target/i386/kvm.c b/target/i386/kvm.c > index 47427d98f8..056a482d3a 100644 > --- a/target/i386/kvm.c > +++ b/target/i386/kvm.c > @@ -733,8 +733,18 @@ static int hyperv_handle_properties(CPUState *cs) > env->features[FEAT_HYPERV_EAX] |= HV_VP_RUNTIME_AVAILABLE; > } > if (cpu->hyperv_synic) { > - if (!has_msr_hv_synic || > - !kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV_SYNIC)) { > + unsigned int cap = KVM_CAP_HYPERV_SYNIC; > + if (!cpu->hyperv_synic_kvm_only) { > + if (!cpu->hyperv_vpindex) { > + fprintf(stderr, "Hyper-V SynIC " > + "(requested by 'hv-synic' cpu flag) " > + "requires Hyper-V VP_INDEX ('hv-vpindex')\n"); > + return -ENOSYS; > + } > + cap = KVM_CAP_HYPERV_SYNIC2; > + } > + > + if (!has_msr_hv_synic || !kvm_check_extension(cs->kvm_state, cap)) { > fprintf(stderr, "Hyper-V SynIC (requested by 'hv-synic' cpu flag) " > "is not supported by kernel\n"); > return -ENOSYS; > @@ -783,18 +793,22 @@ static int hyperv_init_vcpu(X86CPU *cpu) > } > > if (cpu->hyperv_synic) { > - ret = kvm_vcpu_enable_cap(cs, KVM_CAP_HYPERV_SYNIC, 0); > + uint32_t synic_cap = cpu->hyperv_synic_kvm_only ? > + KVM_CAP_HYPERV_SYNIC : KVM_CAP_HYPERV_SYNIC2; > + ret = kvm_vcpu_enable_cap(cs, synic_cap, 0); > if (ret < 0) { > error_report("failed to turn on HyperV SynIC in KVM: %s", > strerror(-ret)); > return ret; > } > > - ret = hyperv_x86_synic_add(cpu); > - if (ret < 0) { > - error_report("failed to create HyperV SynIC: %s", > - strerror(-ret)); > - return ret; > + if (!cpu->hyperv_synic_kvm_only) { > + ret = hyperv_x86_synic_add(cpu); > + if (ret < 0) { > + error_report("failed to create HyperV SynIC: %s", > + strerror(-ret)); > + return ret; > + } > } > } >
On Mon, Nov 26, 2018 at 03:45:03PM +0100, Igor Mammedov wrote: > On Fri, 21 Sep 2018 11:22:10 +0300 > Roman Kagan <rkagan@virtuozzo.com> wrote: > > > Certain configurations do not allow SynIC to be used in QEMU. In > > particular, > > > > - when hyperv_vpindex is off, SINT routes can't be used as they refer to > > the destination vCPU by vp_index > > > > - older KVM (which doesn't expose KVM_CAP_HYPERV_SYNIC2) zeroes out > > SynIC message and event pages on every msr load, breaking migration > > > > OTOH in-KVM users of SynIC -- SynIC timers -- do work in those > > configurations, and we shouldn't stop the guest from using them. > > > > To cover both scenarios, introduce an X86CPU property that makes CPU > > init code to skip creation of the SynIC object (and thus disables any > > SynIC use in QEMU) but keeps the KVM part of the SynIC working. > > The property is clear by default but is set via compat logic for older > > machine types. > > > > As a result, when hv_synic and a modern machine type are specified, QEMU > > will refuse to run unless vp_index is on and the kernel is recent > > enough. OTOH with an older machine type QEMU will run fine with > > hv_synic=on against an older kernel and/or without vp_index enabled but > > will disallow the in-QEMU uses of SynIC (in e.g. VMBus). > > > > Signed-off-by: Roman Kagan <rkagan@virtuozzo.com> > > With current upstream and x-hv-synic-kvm-only=on QEMU will SIGSEGV. > Problem was unnoticed since added compat property wasn't actually used > until much later commit > 4a93722f9c hw/i386: add pc-i440fx-3.1 & pc-q35-3.1 > which put compat property in use. > > qemu-system-x86_64 -machine pc-i440fx-2.10,accel=kvm \ > -cpu host,-vmx,hv-relaxed,hv_spinlocks=0x1fff,hv-vpindex,hv-synic > > simpler reproducer: > qemu-system-x86_64 -enable-kvm -cpu host,hv-synic,x-hv-synic-kvm-only=on > > Reported-by: Vitaly Kuznetsov <vkuznets@redhat.com> Thanks for the report, fix is on the way to ML. Roman.
diff --git a/include/hw/i386/pc.h b/include/hw/i386/pc.h index 6894f37df1..dfe6746692 100644 --- a/include/hw/i386/pc.h +++ b/include/hw/i386/pc.h @@ -294,6 +294,14 @@ int e820_add_entry(uint64_t, uint64_t, uint32_t); int e820_get_num_entries(void); bool e820_get_entry(int, uint32_t, uint64_t *, uint64_t *); +#define PC_COMPAT_3_0 \ + HW_COMPAT_3_0 \ + {\ + .driver = TYPE_X86_CPU,\ + .property = "x-hv-synic-kvm-only",\ + .value = "on",\ + } + #define PC_COMPAT_2_12 \ HW_COMPAT_2_12 \ {\ diff --git a/target/i386/cpu.h b/target/i386/cpu.h index b572a8e4aa..e2e87dc13f 100644 --- a/target/i386/cpu.h +++ b/target/i386/cpu.h @@ -1376,6 +1376,7 @@ struct X86CPU { bool hyperv_vpindex; bool hyperv_runtime; bool hyperv_synic; + bool hyperv_synic_kvm_only; bool hyperv_stimer; bool hyperv_frequencies; bool hyperv_reenlightenment; diff --git a/target/i386/cpu.c b/target/i386/cpu.c index f24295e6e4..9c29c5db81 100644 --- a/target/i386/cpu.c +++ b/target/i386/cpu.c @@ -5575,6 +5575,8 @@ static Property x86_cpu_properties[] = { * to the specific Windows version being used." */ DEFINE_PROP_INT32("x-hv-max-vps", X86CPU, hv_max_vps, -1), + DEFINE_PROP_BOOL("x-hv-synic-kvm-only", X86CPU, hyperv_synic_kvm_only, + false), DEFINE_PROP_END_OF_LIST() }; diff --git a/target/i386/kvm.c b/target/i386/kvm.c index 47427d98f8..056a482d3a 100644 --- a/target/i386/kvm.c +++ b/target/i386/kvm.c @@ -733,8 +733,18 @@ static int hyperv_handle_properties(CPUState *cs) env->features[FEAT_HYPERV_EAX] |= HV_VP_RUNTIME_AVAILABLE; } if (cpu->hyperv_synic) { - if (!has_msr_hv_synic || - !kvm_check_extension(cs->kvm_state, KVM_CAP_HYPERV_SYNIC)) { + unsigned int cap = KVM_CAP_HYPERV_SYNIC; + if (!cpu->hyperv_synic_kvm_only) { + if (!cpu->hyperv_vpindex) { + fprintf(stderr, "Hyper-V SynIC " + "(requested by 'hv-synic' cpu flag) " + "requires Hyper-V VP_INDEX ('hv-vpindex')\n"); + return -ENOSYS; + } + cap = KVM_CAP_HYPERV_SYNIC2; + } + + if (!has_msr_hv_synic || !kvm_check_extension(cs->kvm_state, cap)) { fprintf(stderr, "Hyper-V SynIC (requested by 'hv-synic' cpu flag) " "is not supported by kernel\n"); return -ENOSYS; @@ -783,18 +793,22 @@ static int hyperv_init_vcpu(X86CPU *cpu) } if (cpu->hyperv_synic) { - ret = kvm_vcpu_enable_cap(cs, KVM_CAP_HYPERV_SYNIC, 0); + uint32_t synic_cap = cpu->hyperv_synic_kvm_only ? + KVM_CAP_HYPERV_SYNIC : KVM_CAP_HYPERV_SYNIC2; + ret = kvm_vcpu_enable_cap(cs, synic_cap, 0); if (ret < 0) { error_report("failed to turn on HyperV SynIC in KVM: %s", strerror(-ret)); return ret; } - ret = hyperv_x86_synic_add(cpu); - if (ret < 0) { - error_report("failed to create HyperV SynIC: %s", - strerror(-ret)); - return ret; + if (!cpu->hyperv_synic_kvm_only) { + ret = hyperv_x86_synic_add(cpu); + if (ret < 0) { + error_report("failed to create HyperV SynIC: %s", + strerror(-ret)); + return ret; + } } }
Certain configurations do not allow SynIC to be used in QEMU. In particular, - when hyperv_vpindex is off, SINT routes can't be used as they refer to the destination vCPU by vp_index - older KVM (which doesn't expose KVM_CAP_HYPERV_SYNIC2) zeroes out SynIC message and event pages on every msr load, breaking migration OTOH in-KVM users of SynIC -- SynIC timers -- do work in those configurations, and we shouldn't stop the guest from using them. To cover both scenarios, introduce an X86CPU property that makes CPU init code to skip creation of the SynIC object (and thus disables any SynIC use in QEMU) but keeps the KVM part of the SynIC working. The property is clear by default but is set via compat logic for older machine types. As a result, when hv_synic and a modern machine type are specified, QEMU will refuse to run unless vp_index is on and the kernel is recent enough. OTOH with an older machine type QEMU will run fine with hv_synic=on against an older kernel and/or without vp_index enabled but will disallow the in-QEMU uses of SynIC (in e.g. VMBus). Signed-off-by: Roman Kagan <rkagan@virtuozzo.com> --- include/hw/i386/pc.h | 8 ++++++++ target/i386/cpu.h | 1 + target/i386/cpu.c | 2 ++ target/i386/kvm.c | 30 ++++++++++++++++++++++-------- 4 files changed, 33 insertions(+), 8 deletions(-)