| Message ID | 20231115071519.2864957-27-xiaoyao.li@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | QEMU Guest memfd + QEMU TDX support | expand |
On Wed, Nov 15, 2023 at 02:14:35AM -0500, Xiaoyao Li wrote: > Invoke KVM_TDX_INIT in kvm_arch_pre_create_vcpu() that KVM_TDX_INIT > configures global TD configurations, e.g. the canonical CPUID config, > and must be executed prior to creating vCPUs. > > Use kvm_x86_arch_cpuid() to setup the CPUID settings for TDX VM. > > Note, this doesn't address the fact that QEMU may change the CPUID > configuration when creating vCPUs, i.e. punts on refactoring QEMU to > provide a stable CPUID config prior to kvm_arch_init(). > > Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> > Acked-by: Gerd Hoffmann <kraxel@redhat.com> > --- > Changes in v3: > - Pass @errp in tdx_pre_create_vcpu() and pass error info to it. (Daniel) > --- > accel/kvm/kvm-all.c | 9 +++++++- > target/i386/kvm/kvm.c | 9 ++++++++ > target/i386/kvm/tdx-stub.c | 5 +++++ > target/i386/kvm/tdx.c | 45 ++++++++++++++++++++++++++++++++++++++ > target/i386/kvm/tdx.h | 4 ++++ > 5 files changed, 71 insertions(+), 1 deletion(-) > > diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c > index 6b5f4d62f961..a92fff471b58 100644 > --- a/accel/kvm/kvm-all.c > +++ b/accel/kvm/kvm-all.c > @@ -441,8 +441,15 @@ int kvm_init_vcpu(CPUState *cpu, Error **errp) > > trace_kvm_init_vcpu(cpu->cpu_index, kvm_arch_vcpu_id(cpu)); > > + /* > + * tdx_pre_create_vcpu() may call cpu_x86_cpuid(). It in turn may call > + * kvm_vm_ioctl(). Set cpu->kvm_state in advance to avoid NULL pointer > + * dereference. > + */ > + cpu->kvm_state = s; > ret = kvm_arch_pre_create_vcpu(cpu, errp); > if (ret < 0) { > + cpu->kvm_state = NULL; > goto err; > } > > @@ -450,11 +457,11 @@ int kvm_init_vcpu(CPUState *cpu, Error **errp) > if (ret < 0) { > error_setg_errno(errp, -ret, "kvm_init_vcpu: kvm_get_vcpu failed (%lu)", > kvm_arch_vcpu_id(cpu)); > + cpu->kvm_state = NULL; > goto err; > } > > cpu->kvm_fd = ret; > - cpu->kvm_state = s; > cpu->vcpu_dirty = true; > cpu->dirty_pages = 0; > cpu->throttle_us_per_full = 0; > diff --git a/target/i386/kvm/kvm.c b/target/i386/kvm/kvm.c > index dafe4d262977..fc840653ceb6 100644 > --- a/target/i386/kvm/kvm.c > +++ b/target/i386/kvm/kvm.c > @@ -2268,6 +2268,15 @@ int kvm_arch_init_vcpu(CPUState *cs) > return r; > } > > +int kvm_arch_pre_create_vcpu(CPUState *cpu, Error **errp) > +{ > + if (is_tdx_vm()) { > + return tdx_pre_create_vcpu(cpu, errp); > + } > + > + return 0; > +} > + > int kvm_arch_destroy_vcpu(CPUState *cs) > { > X86CPU *cpu = X86_CPU(cs); > diff --git a/target/i386/kvm/tdx-stub.c b/target/i386/kvm/tdx-stub.c > index 1d866d5496bf..3877d432a397 100644 > --- a/target/i386/kvm/tdx-stub.c > +++ b/target/i386/kvm/tdx-stub.c > @@ -6,3 +6,8 @@ int tdx_kvm_init(MachineState *ms, Error **errp) > { > return -EINVAL; > } > + > +int tdx_pre_create_vcpu(CPUState *cpu, Error **errp) > +{ > + return -EINVAL; > +} > diff --git a/target/i386/kvm/tdx.c b/target/i386/kvm/tdx.c > index 1f5d8117d1a9..122a37c93de3 100644 > --- a/target/i386/kvm/tdx.c > +++ b/target/i386/kvm/tdx.c > @@ -467,6 +467,49 @@ int tdx_kvm_init(MachineState *ms, Error **errp) > return 0; > } > > +int tdx_pre_create_vcpu(CPUState *cpu, Error **errp) > +{ > + MachineState *ms = MACHINE(qdev_get_machine()); > + X86CPU *x86cpu = X86_CPU(cpu); > + CPUX86State *env = &x86cpu->env; > + struct kvm_tdx_init_vm *init_vm; Mark this as auto-free to avoid the g_free() requirement g_autofree struct kvm_tdx_init_vm *init_vm = NULL; > + int r = 0; > + > + qemu_mutex_lock(&tdx_guest->lock); QEMU_LOCK_GUARD(&tdx_guest->lock); to eliminate the mutex_unlock requirement, thus eliminating all 'goto' jumps and label targets, in favour of a plain 'return -1' everywhere. > + if (tdx_guest->initialized) { > + goto out; > + } > + > + init_vm = g_malloc0(sizeof(struct kvm_tdx_init_vm) + > + sizeof(struct kvm_cpuid_entry2) * KVM_MAX_CPUID_ENTRIES); > + > + r = kvm_vm_enable_cap(kvm_state, KVM_CAP_MAX_VCPUS, 0, ms->smp.cpus); > + if (r < 0) { > + error_setg(errp, "Unable to set MAX VCPUS to %d", ms->smp.cpus); > + goto out_free; > + } > + > + init_vm->cpuid.nent = kvm_x86_arch_cpuid(env, init_vm->cpuid.entries, 0); > + > + init_vm->attributes = tdx_guest->attributes; > + > + do { > + r = tdx_vm_ioctl(KVM_TDX_INIT_VM, 0, init_vm); > + } while (r == -EAGAIN); > + if (r < 0) { > + error_setg_errno(errp, -r, "KVM_TDX_INIT_VM failed"); > + goto out_free; > + } > + > + tdx_guest->initialized = true; > + > +out_free: > + g_free(init_vm); > +out: > + qemu_mutex_unlock(&tdx_guest->lock); > + return r; > +} > + > /* tdx guest */ > OBJECT_DEFINE_TYPE_WITH_INTERFACES(TdxGuest, > tdx_guest, > @@ -479,6 +522,8 @@ static void tdx_guest_init(Object *obj) > { > TdxGuest *tdx = TDX_GUEST(obj); > > + qemu_mutex_init(&tdx->lock); > + > tdx->attributes = 0; > } > > diff --git a/target/i386/kvm/tdx.h b/target/i386/kvm/tdx.h > index 06599b65b827..432077723ac5 100644 > --- a/target/i386/kvm/tdx.h > +++ b/target/i386/kvm/tdx.h > @@ -17,6 +17,9 @@ typedef struct TdxGuestClass { > typedef struct TdxGuest { > ConfidentialGuestSupport parent_obj; > > + QemuMutex lock; > + > + bool initialized; > uint64_t attributes; /* TD attributes */ > } TdxGuest; > > @@ -29,5 +32,6 @@ bool is_tdx_vm(void); > int tdx_kvm_init(MachineState *ms, Error **errp); > void tdx_get_supported_cpuid(uint32_t function, uint32_t index, int reg, > uint32_t *ret); > +int tdx_pre_create_vcpu(CPUState *cpu, Error **errp); > > #endif /* QEMU_I386_TDX_H */ > -- > 2.34.1 > With regards, Daniel
On 11/15/2023 7:01 PM, Daniel P. Berrangé wrote: > On Wed, Nov 15, 2023 at 02:14:35AM -0500, Xiaoyao Li wrote: >> Invoke KVM_TDX_INIT in kvm_arch_pre_create_vcpu() that KVM_TDX_INIT >> configures global TD configurations, e.g. the canonical CPUID config, >> and must be executed prior to creating vCPUs. >> >> Use kvm_x86_arch_cpuid() to setup the CPUID settings for TDX VM. >> >> Note, this doesn't address the fact that QEMU may change the CPUID >> configuration when creating vCPUs, i.e. punts on refactoring QEMU to >> provide a stable CPUID config prior to kvm_arch_init(). >> >> Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> >> Acked-by: Gerd Hoffmann <kraxel@redhat.com> >> --- >> Changes in v3: >> - Pass @errp in tdx_pre_create_vcpu() and pass error info to it. (Daniel) >> --- >> accel/kvm/kvm-all.c | 9 +++++++- >> target/i386/kvm/kvm.c | 9 ++++++++ >> target/i386/kvm/tdx-stub.c | 5 +++++ >> target/i386/kvm/tdx.c | 45 ++++++++++++++++++++++++++++++++++++++ >> target/i386/kvm/tdx.h | 4 ++++ >> 5 files changed, 71 insertions(+), 1 deletion(-) >> >> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c >> index 6b5f4d62f961..a92fff471b58 100644 >> --- a/accel/kvm/kvm-all.c >> +++ b/accel/kvm/kvm-all.c >> @@ -441,8 +441,15 @@ int kvm_init_vcpu(CPUState *cpu, Error **errp) >> >> trace_kvm_init_vcpu(cpu->cpu_index, kvm_arch_vcpu_id(cpu)); >> >> + /* >> + * tdx_pre_create_vcpu() may call cpu_x86_cpuid(). It in turn may call >> + * kvm_vm_ioctl(). Set cpu->kvm_state in advance to avoid NULL pointer >> + * dereference. >> + */ >> + cpu->kvm_state = s; >> ret = kvm_arch_pre_create_vcpu(cpu, errp); >> if (ret < 0) { >> + cpu->kvm_state = NULL; >> goto err; >> } >> >> @@ -450,11 +457,11 @@ int kvm_init_vcpu(CPUState *cpu, Error **errp) >> if (ret < 0) { >> error_setg_errno(errp, -ret, "kvm_init_vcpu: kvm_get_vcpu failed (%lu)", >> kvm_arch_vcpu_id(cpu)); >> + cpu->kvm_state = NULL; >> goto err; >> } >> >> cpu->kvm_fd = ret; >> - cpu->kvm_state = s; >> cpu->vcpu_dirty = true; >> cpu->dirty_pages = 0; >> cpu->throttle_us_per_full = 0; >> diff --git a/target/i386/kvm/kvm.c b/target/i386/kvm/kvm.c >> index dafe4d262977..fc840653ceb6 100644 >> --- a/target/i386/kvm/kvm.c >> +++ b/target/i386/kvm/kvm.c >> @@ -2268,6 +2268,15 @@ int kvm_arch_init_vcpu(CPUState *cs) >> return r; >> } >> >> +int kvm_arch_pre_create_vcpu(CPUState *cpu, Error **errp) >> +{ >> + if (is_tdx_vm()) { >> + return tdx_pre_create_vcpu(cpu, errp); >> + } >> + >> + return 0; >> +} >> + >> int kvm_arch_destroy_vcpu(CPUState *cs) >> { >> X86CPU *cpu = X86_CPU(cs); >> diff --git a/target/i386/kvm/tdx-stub.c b/target/i386/kvm/tdx-stub.c >> index 1d866d5496bf..3877d432a397 100644 >> --- a/target/i386/kvm/tdx-stub.c >> +++ b/target/i386/kvm/tdx-stub.c >> @@ -6,3 +6,8 @@ int tdx_kvm_init(MachineState *ms, Error **errp) >> { >> return -EINVAL; >> } >> + >> +int tdx_pre_create_vcpu(CPUState *cpu, Error **errp) >> +{ >> + return -EINVAL; >> +} >> diff --git a/target/i386/kvm/tdx.c b/target/i386/kvm/tdx.c >> index 1f5d8117d1a9..122a37c93de3 100644 >> --- a/target/i386/kvm/tdx.c >> +++ b/target/i386/kvm/tdx.c >> @@ -467,6 +467,49 @@ int tdx_kvm_init(MachineState *ms, Error **errp) >> return 0; >> } >> >> +int tdx_pre_create_vcpu(CPUState *cpu, Error **errp) >> +{ >> + MachineState *ms = MACHINE(qdev_get_machine()); >> + X86CPU *x86cpu = X86_CPU(cpu); >> + CPUX86State *env = &x86cpu->env; >> + struct kvm_tdx_init_vm *init_vm; > > Mark this as auto-free to avoid the g_free() requirement > > g_autofree struct kvm_tdx_init_vm *init_vm = NULL; > >> + int r = 0; >> + >> + qemu_mutex_lock(&tdx_guest->lock); > > QEMU_LOCK_GUARD(&tdx_guest->lock); > > to eliminate the mutex_unlock requirement, thus eliminating all > 'goto' jumps and label targets, in favour of a plain 'return -1' > everywhere. > Learned! thanks!
diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c index 6b5f4d62f961..a92fff471b58 100644 --- a/accel/kvm/kvm-all.c +++ b/accel/kvm/kvm-all.c @@ -441,8 +441,15 @@ int kvm_init_vcpu(CPUState *cpu, Error **errp) trace_kvm_init_vcpu(cpu->cpu_index, kvm_arch_vcpu_id(cpu)); + /* + * tdx_pre_create_vcpu() may call cpu_x86_cpuid(). It in turn may call + * kvm_vm_ioctl(). Set cpu->kvm_state in advance to avoid NULL pointer + * dereference. + */ + cpu->kvm_state = s; ret = kvm_arch_pre_create_vcpu(cpu, errp); if (ret < 0) { + cpu->kvm_state = NULL; goto err; } @@ -450,11 +457,11 @@ int kvm_init_vcpu(CPUState *cpu, Error **errp) if (ret < 0) { error_setg_errno(errp, -ret, "kvm_init_vcpu: kvm_get_vcpu failed (%lu)", kvm_arch_vcpu_id(cpu)); + cpu->kvm_state = NULL; goto err; } cpu->kvm_fd = ret; - cpu->kvm_state = s; cpu->vcpu_dirty = true; cpu->dirty_pages = 0; cpu->throttle_us_per_full = 0; diff --git a/target/i386/kvm/kvm.c b/target/i386/kvm/kvm.c index dafe4d262977..fc840653ceb6 100644 --- a/target/i386/kvm/kvm.c +++ b/target/i386/kvm/kvm.c @@ -2268,6 +2268,15 @@ int kvm_arch_init_vcpu(CPUState *cs) return r; } +int kvm_arch_pre_create_vcpu(CPUState *cpu, Error **errp) +{ + if (is_tdx_vm()) { + return tdx_pre_create_vcpu(cpu, errp); + } + + return 0; +} + int kvm_arch_destroy_vcpu(CPUState *cs) { X86CPU *cpu = X86_CPU(cs); diff --git a/target/i386/kvm/tdx-stub.c b/target/i386/kvm/tdx-stub.c index 1d866d5496bf..3877d432a397 100644 --- a/target/i386/kvm/tdx-stub.c +++ b/target/i386/kvm/tdx-stub.c @@ -6,3 +6,8 @@ int tdx_kvm_init(MachineState *ms, Error **errp) { return -EINVAL; } + +int tdx_pre_create_vcpu(CPUState *cpu, Error **errp) +{ + return -EINVAL; +} diff --git a/target/i386/kvm/tdx.c b/target/i386/kvm/tdx.c index 1f5d8117d1a9..122a37c93de3 100644 --- a/target/i386/kvm/tdx.c +++ b/target/i386/kvm/tdx.c @@ -467,6 +467,49 @@ int tdx_kvm_init(MachineState *ms, Error **errp) return 0; } +int tdx_pre_create_vcpu(CPUState *cpu, Error **errp) +{ + MachineState *ms = MACHINE(qdev_get_machine()); + X86CPU *x86cpu = X86_CPU(cpu); + CPUX86State *env = &x86cpu->env; + struct kvm_tdx_init_vm *init_vm; + int r = 0; + + qemu_mutex_lock(&tdx_guest->lock); + if (tdx_guest->initialized) { + goto out; + } + + init_vm = g_malloc0(sizeof(struct kvm_tdx_init_vm) + + sizeof(struct kvm_cpuid_entry2) * KVM_MAX_CPUID_ENTRIES); + + r = kvm_vm_enable_cap(kvm_state, KVM_CAP_MAX_VCPUS, 0, ms->smp.cpus); + if (r < 0) { + error_setg(errp, "Unable to set MAX VCPUS to %d", ms->smp.cpus); + goto out_free; + } + + init_vm->cpuid.nent = kvm_x86_arch_cpuid(env, init_vm->cpuid.entries, 0); + + init_vm->attributes = tdx_guest->attributes; + + do { + r = tdx_vm_ioctl(KVM_TDX_INIT_VM, 0, init_vm); + } while (r == -EAGAIN); + if (r < 0) { + error_setg_errno(errp, -r, "KVM_TDX_INIT_VM failed"); + goto out_free; + } + + tdx_guest->initialized = true; + +out_free: + g_free(init_vm); +out: + qemu_mutex_unlock(&tdx_guest->lock); + return r; +} + /* tdx guest */ OBJECT_DEFINE_TYPE_WITH_INTERFACES(TdxGuest, tdx_guest, @@ -479,6 +522,8 @@ static void tdx_guest_init(Object *obj) { TdxGuest *tdx = TDX_GUEST(obj); + qemu_mutex_init(&tdx->lock); + tdx->attributes = 0; } diff --git a/target/i386/kvm/tdx.h b/target/i386/kvm/tdx.h index 06599b65b827..432077723ac5 100644 --- a/target/i386/kvm/tdx.h +++ b/target/i386/kvm/tdx.h @@ -17,6 +17,9 @@ typedef struct TdxGuestClass { typedef struct TdxGuest { ConfidentialGuestSupport parent_obj; + QemuMutex lock; + + bool initialized; uint64_t attributes; /* TD attributes */ } TdxGuest; @@ -29,5 +32,6 @@ bool is_tdx_vm(void); int tdx_kvm_init(MachineState *ms, Error **errp); void tdx_get_supported_cpuid(uint32_t function, uint32_t index, int reg, uint32_t *ret); +int tdx_pre_create_vcpu(CPUState *cpu, Error **errp); #endif /* QEMU_I386_TDX_H */