Message ID | 1538765165-19177-1-git-send-email-linux@roeck-us.net (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | KVM: X86: Add missing KVM_AMD dependency | expand |
On 05/10/2018 20:46, Guenter Roeck wrote: > Analysis shows that commit 59414c9892208 ("KVM: SVM: Add support for > KVM_SEV_LAUNCH_START command") added a dependency of KVM_AMD on > CRYPTO_DEV_CCP_DD if CRYPTO_DEV_SP_PSP is enabled: If CRYPTO_DEV_CCP_DD > is built as module, KVM_AMD must be built as module as well. > > Fixes: 59414c9892208 ("KVM: SVM: Add support for KVM_SEV_LAUNCH_START command") > Cc: Brijesh Singh <brijesh.singh@amd.com> > Cc: Borislav Petkov <bp@suse.de> > Signed-off-by: Guenter Roeck <linux@roeck-us.net> This should be handled by config KVM_AMD_SEV def_bool y bool "AMD Secure Encrypted Virtualization (SEV) support" depends on KVM_AMD && X86_64 depends on CRYPTO_DEV_SP_PSP && !(KVM_AMD=y && CRYPTO_DEV_CCP_DD=m) ---help--- Provides support for launching Encrypted VMs on AMD processors. Maybe this works as well? I haven't tested it yet: diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c index 89c4c5aa15f1..55f10b17d044 100644 --- a/arch/x86/kvm/svm.c +++ b/arch/x86/kvm/svm.c @@ -441,9 +441,13 @@ static inline bool svm_sev_enabled(void) static inline bool sev_guest(struct kvm *kvm) { +#ifdef CONFIG_KVM_AMD_SEV struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; return sev->active; +#else + return false; +#endif } static inline int sev_get_asid(struct kvm *kvm) Thanks, Paolo
On Fri, Oct 05, 2018 at 10:41:55PM +0200, Paolo Bonzini wrote: > On 05/10/2018 20:46, Guenter Roeck wrote: > > Analysis shows that commit 59414c9892208 ("KVM: SVM: Add support for > > KVM_SEV_LAUNCH_START command") added a dependency of KVM_AMD on > > CRYPTO_DEV_CCP_DD if CRYPTO_DEV_SP_PSP is enabled: If CRYPTO_DEV_CCP_DD > > is built as module, KVM_AMD must be built as module as well. > > > > Fixes: 59414c9892208 ("KVM: SVM: Add support for KVM_SEV_LAUNCH_START command") > > Cc: Brijesh Singh <brijesh.singh@amd.com> > > Cc: Borislav Petkov <bp@suse.de> > > Signed-off-by: Guenter Roeck <linux@roeck-us.net> > > This should be handled by > > config KVM_AMD_SEV > def_bool y > bool "AMD Secure Encrypted Virtualization (SEV) support" > depends on KVM_AMD && X86_64 > depends on CRYPTO_DEV_SP_PSP && !(KVM_AMD=y && CRYPTO_DEV_CCP_DD=m) > ---help--- > Provides support for launching Encrypted VMs on AMD processors. > Unfortunately it doesn't. It disables KVM_AMD_SEV, but that doesn't prevent the calls. > Maybe this works as well? I haven't tested it yet: > I am sure there are many possible solutions. I would personally prefer one that enforces KVM_AMD=m with CRYPTO_DEV_CCP_DD=m, but that is just me. Thanks, Guenter > diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c > index 89c4c5aa15f1..55f10b17d044 100644 > --- a/arch/x86/kvm/svm.c > +++ b/arch/x86/kvm/svm.c > @@ -441,9 +441,13 @@ static inline bool svm_sev_enabled(void) > > static inline bool sev_guest(struct kvm *kvm) > { > +#ifdef CONFIG_KVM_AMD_SEV > struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info; > > return sev->active; > +#else > + return false; > +#endif > } > > static inline int sev_get_asid(struct kvm *kvm) > > Thanks, > > Paolo
On 06/10/2018 00:03, Guenter Roeck wrote: >> This should be handled by >> >> config KVM_AMD_SEV >> def_bool y >> bool "AMD Secure Encrypted Virtualization (SEV) support" >> depends on KVM_AMD && X86_64 >> depends on CRYPTO_DEV_SP_PSP && !(KVM_AMD=y && CRYPTO_DEV_CCP_DD=m) >> ---help--- >> Provides support for launching Encrypted VMs on AMD processors. >> > Unfortunately it doesn't. It disables KVM_AMD_SEV, but that doesn't prevent > the calls. Yes, exactly - that's why I mentioned the sev_guest patch that should cull all the SEV code from a !KVM_AMD_SEV build. >> Maybe this works as well? I haven't tested it yet: >> > I am sure there are many possible solutions. I would personally prefer one > that enforces KVM_AMD=m with CRYPTO_DEV_CCP_DD=m, but that is just me. Well, KVM_AMD=y is a relatively unusual choice to begin with. The question is whether then you want to disable this choice completely when CRYPTO_DEV_CCP_DD=m, or just disable SEV. My patch is a good idea anyway, if I may say so :), because it culls a lot of code from a !KVM_AMD_SEV build. But if it is not enough, we certainly have to do something else about the failure you're reporting. Paolo
Hi Paolo, On 10/05/2018 03:18 PM, Paolo Bonzini wrote: > On 06/10/2018 00:03, Guenter Roeck wrote: >>> This should be handled by >>> >>> config KVM_AMD_SEV >>> def_bool y >>> bool "AMD Secure Encrypted Virtualization (SEV) support" >>> depends on KVM_AMD && X86_64 >>> depends on CRYPTO_DEV_SP_PSP && !(KVM_AMD=y && CRYPTO_DEV_CCP_DD=m) >>> ---help--- >>> Provides support for launching Encrypted VMs on AMD processors. >>> >> Unfortunately it doesn't. It disables KVM_AMD_SEV, but that doesn't prevent >> the calls. > > Yes, exactly - that's why I mentioned the sev_guest patch that should > cull all the SEV code from a !KVM_AMD_SEV build. > >>> Maybe this works as well? I haven't tested it yet: >>> >> I am sure there are many possible solutions. I would personally prefer one >> that enforces KVM_AMD=m with CRYPTO_DEV_CCP_DD=m, but that is just me. > > Well, KVM_AMD=y is a relatively unusual choice to begin with. The It is common enough that we are not the only ones affected. Also, even a "relatively unusual choice" should, in my opinion, not result in a build error. Never mind, I'll just apply the suggested workaround and configure CRYPTO_DEV_CCP_DD=y. I need to do that anyway, after all, if I want to keep KVM_AMD=y. Thanks, Guenter
On 06/10/2018 22:43, Guenter Roeck wrote: >> >>>> Maybe this works as well? I haven't tested it yet: >>>> >>> I am sure there are many possible solutions. I would personally >>> prefer one >>> that enforces KVM_AMD=m with CRYPTO_DEV_CCP_DD=m, but that is just me. >> >> Well, KVM_AMD=y is a relatively unusual choice to begin with. The > > It is common enough that we are not the only ones affected. Also, even a > "relatively unusual choice" should, in my opinion, not result in a build > error. Of course not! The question is whether to solve it by disabling KVM_AMD_SEV (which is what the current code attempts to do, and my patch should fix that) or KVM_AMD (your patch). Paolo > Never mind, I'll just apply the suggested workaround and configure > CRYPTO_DEV_CCP_DD=y. I need to do that anyway, after all, if I want to keep > KVM_AMD=y.
On 10/08/2018 06:27 AM, Paolo Bonzini wrote: > On 06/10/2018 22:43, Guenter Roeck wrote: >>> >>>>> Maybe this works as well? I haven't tested it yet: >>>>> >>>> I am sure there are many possible solutions. I would personally >>>> prefer one >>>> that enforces KVM_AMD=m with CRYPTO_DEV_CCP_DD=m, but that is just me. >>> >>> Well, KVM_AMD=y is a relatively unusual choice to begin with. The >> >> It is common enough that we are not the only ones affected. Also, even a >> "relatively unusual choice" should, in my opinion, not result in a build >> error. > > Of course not! The question is whether to solve it by disabling > KVM_AMD_SEV (which is what the current code attempts to do, and my patch > should fix that) or KVM_AMD (your patch). IMHO, Paolo's patch make sense; it removes all the SEV specific code when KVM_AMD_SEV=n. It saves ~4K in text section. Paolo, Does it make sense to move all the SEV specific code in svm-sev.c ? I am looking to add SEV migration support very soon, and can see myself adding more SEV command handling which will grow svm.c further. -Brijesh
On Mon, Oct 08, 2018 at 02:52:46PM +0000, Singh, Brijesh wrote: > Does it make sense to move all the SEV specific code in svm-sev.c ? > I am looking to add SEV migration support very soon, and can see > myself adding more SEV command handling which will grow svm.c further. Amen to that - svm.c is pretty huge already. And if you end up moving facilities, you can simply call it sev.c Just my 2¢. Thx.
On 08/10/2018 19:32, Borislav Petkov wrote: > On Mon, Oct 08, 2018 at 02:52:46PM +0000, Singh, Brijesh wrote: >> Does it make sense to move all the SEV specific code in svm-sev.c ? >> I am looking to add SEV migration support very soon, and can see >> myself adding more SEV command handling which will grow svm.c further. > > Amen to that - svm.c is pretty huge already. And if you end up moving > facilities, you can simply call it sev.c Creating arch/x86/kvm/{mmu,vmx,svm}/ has been on my todo list for a while. If you would like to split the SEV bits, that would be a good start. Paolo
diff --git a/arch/x86/kvm/Kconfig b/arch/x86/kvm/Kconfig index 1bbec387d289..a8eff6910ea3 100644 --- a/arch/x86/kvm/Kconfig +++ b/arch/x86/kvm/Kconfig @@ -74,6 +74,7 @@ config KVM_INTEL config KVM_AMD tristate "KVM for AMD processors support" depends on KVM + depends on !CRYPTO_DEV_SP_PSP || (CRYPTO_DEV_SP_PSP && CRYPTO_DEV_CCP_DD) ---help--- Provides support for KVM on AMD processors equipped with the AMD-V (SVM) extensions.
Building an image with KVM_AMD=y, CRYPTO_DEV_SP_PSP=y, and CRYPTO_DEV_CCP_DD=m fails with the following error messages. arch/x86/kvm/svm.c:6287: undefined reference to `sev_issue_cmd_external_user' arch/x86/kvm/svm.o: In function `sev_unbind_asid': arch/x86/kvm/svm.c:1747: undefined reference to `sev_guest_deactivate' arch/x86/kvm/svm.c:1750: undefined reference to `sev_guest_df_flush' arch/x86/kvm/svm.c:1759: undefined reference to `sev_guest_decommission' Analysis shows that commit 59414c9892208 ("KVM: SVM: Add support for KVM_SEV_LAUNCH_START command") added a dependency of KVM_AMD on CRYPTO_DEV_CCP_DD if CRYPTO_DEV_SP_PSP is enabled: If CRYPTO_DEV_CCP_DD is built as module, KVM_AMD must be built as module as well. Fixes: 59414c9892208 ("KVM: SVM: Add support for KVM_SEV_LAUNCH_START command") Cc: Brijesh Singh <brijesh.singh@amd.com> Cc: Borislav Petkov <bp@suse.de> Signed-off-by: Guenter Roeck <linux@roeck-us.net> --- arch/x86/kvm/Kconfig | 1 + 1 file changed, 1 insertion(+)