| Message ID | 20240416063534.3469482-1-Sergiy_Kibrik@epam.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | x86: make cpu virtualization support configurable | expand |
On 16/04/2024 7:35 am, Sergiy Kibrik wrote: > diff --git a/xen/arch/x86/cpu/Makefile b/xen/arch/x86/cpu/Makefile > index 35561fe51d..d3d7b8fb2e 100644 > --- a/xen/arch/x86/cpu/Makefile > +++ b/xen/arch/x86/cpu/Makefile > @@ -10,4 +10,6 @@ obj-y += intel.o > obj-y += intel_cacheinfo.o > obj-y += mwait-idle.o > obj-y += shanghai.o > -obj-y += vpmu.o vpmu_amd.o vpmu_intel.o > +obj-y += vpmu.o > +obj-$(CONFIG_SVM) += vpmu_amd.o > +obj-$(CONFIG_VMX) += vpmu_intel.o I'm afraid this breaks perf counters on PV guests. These files are joint guest-type implementations. Seeing as you leave vpmu.o alone, I guess that all you're actually wanting to do is compile out vpmu_intel.o? In which case, use CONFIG_{AMD,INTEL} rather than CONFIG_{SVM,VMX} please. ~Andrew
16.04.24 14:05, Andrew Cooper: > On 16/04/2024 7:35 am, Sergiy Kibrik wrote: >> diff --git a/xen/arch/x86/cpu/Makefile b/xen/arch/x86/cpu/Makefile >> index 35561fe51d..d3d7b8fb2e 100644 >> --- a/xen/arch/x86/cpu/Makefile >> +++ b/xen/arch/x86/cpu/Makefile >> @@ -10,4 +10,6 @@ obj-y += intel.o >> obj-y += intel_cacheinfo.o >> obj-y += mwait-idle.o >> obj-y += shanghai.o >> -obj-y += vpmu.o vpmu_amd.o vpmu_intel.o >> +obj-y += vpmu.o >> +obj-$(CONFIG_SVM) += vpmu_amd.o >> +obj-$(CONFIG_VMX) += vpmu_intel.o > > I'm afraid this breaks perf counters on PV guests. These files are > joint guest-type implementations. > > Seeing as you leave vpmu.o alone, I guess that all you're actually > wanting to do is compile out vpmu_intel.o? In which case, use > CONFIG_{AMD,INTEL} rather than CONFIG_{SVM,VMX} please. > Thanks for pointing that out. I think I'll just exclude this patch from the series, and make a separate series with CONFIG_{AMD,INTEL} option and code separation that unrelated to VMX/SVM & HVM/PV, only to CPUs themselves. BTW, how would you suggest CONFIG_{AMD,INTEL} shall relate to CONFIG_{SVM,VMX}? Should CONFIG_VMX just plainly depend on CONFIG_AMD, or more complex relations needed? -Sergiy
On 18.04.2024 15:25, Sergiy Kibrik wrote: > 16.04.24 14:05, Andrew Cooper: >> On 16/04/2024 7:35 am, Sergiy Kibrik wrote: >>> diff --git a/xen/arch/x86/cpu/Makefile b/xen/arch/x86/cpu/Makefile >>> index 35561fe51d..d3d7b8fb2e 100644 >>> --- a/xen/arch/x86/cpu/Makefile >>> +++ b/xen/arch/x86/cpu/Makefile >>> @@ -10,4 +10,6 @@ obj-y += intel.o >>> obj-y += intel_cacheinfo.o >>> obj-y += mwait-idle.o >>> obj-y += shanghai.o >>> -obj-y += vpmu.o vpmu_amd.o vpmu_intel.o >>> +obj-y += vpmu.o >>> +obj-$(CONFIG_SVM) += vpmu_amd.o >>> +obj-$(CONFIG_VMX) += vpmu_intel.o >> >> I'm afraid this breaks perf counters on PV guests. These files are >> joint guest-type implementations. >> >> Seeing as you leave vpmu.o alone, I guess that all you're actually >> wanting to do is compile out vpmu_intel.o? In which case, use >> CONFIG_{AMD,INTEL} rather than CONFIG_{SVM,VMX} please. >> > > Thanks for pointing that out. > I think I'll just exclude this patch from the series, and make a > separate series with CONFIG_{AMD,INTEL} option and code separation that > unrelated to VMX/SVM & HVM/PV, only to CPUs themselves. > > BTW, how would you suggest CONFIG_{AMD,INTEL} shall relate to > CONFIG_{SVM,VMX}? Should CONFIG_VMX just plainly depend on CONFIG_AMD, > or more complex relations needed? The answer to this, from my perspective, is actually one of the reasons why I would have wished we'd have separate CONFIG_INTEL and CONFIG_INTEL_CPU. I'd then suggest to have VMX simply default to INTEL, permitting at least experts to select it also when INTEL=n. Perhaps that would still be the best approach even with this INTEL / INTEL_CPU distinction. Jan
On 18/04/2024 2:25 pm, Sergiy Kibrik wrote: > 16.04.24 14:05, Andrew Cooper: >> On 16/04/2024 7:35 am, Sergiy Kibrik wrote: >>> diff --git a/xen/arch/x86/cpu/Makefile b/xen/arch/x86/cpu/Makefile >>> index 35561fe51d..d3d7b8fb2e 100644 >>> --- a/xen/arch/x86/cpu/Makefile >>> +++ b/xen/arch/x86/cpu/Makefile >>> @@ -10,4 +10,6 @@ obj-y += intel.o >>> obj-y += intel_cacheinfo.o >>> obj-y += mwait-idle.o >>> obj-y += shanghai.o >>> -obj-y += vpmu.o vpmu_amd.o vpmu_intel.o >>> +obj-y += vpmu.o >>> +obj-$(CONFIG_SVM) += vpmu_amd.o >>> +obj-$(CONFIG_VMX) += vpmu_intel.o >> >> I'm afraid this breaks perf counters on PV guests. These files are >> joint guest-type implementations. >> >> Seeing as you leave vpmu.o alone, I guess that all you're actually >> wanting to do is compile out vpmu_intel.o? In which case, use >> CONFIG_{AMD,INTEL} rather than CONFIG_{SVM,VMX} please. >> > > Thanks for pointing that out. > I think I'll just exclude this patch from the series, and make a > separate series with CONFIG_{AMD,INTEL} option and code separation > that unrelated to VMX/SVM & HVM/PV, only to CPUs themselves. > > BTW, how would you suggest CONFIG_{AMD,INTEL} shall relate to > CONFIG_{SVM,VMX}? Should CONFIG_VMX just plainly depend on CONFIG_AMD, > or more complex relations needed? To a first approximation, no linkage. Centaur have an implementation of VMX on the market, and Hygon have an implementation of SVM. ~Andrew
diff --git a/xen/arch/x86/cpu/Makefile b/xen/arch/x86/cpu/Makefile index 35561fe51d..d3d7b8fb2e 100644 --- a/xen/arch/x86/cpu/Makefile +++ b/xen/arch/x86/cpu/Makefile @@ -10,4 +10,6 @@ obj-y += intel.o obj-y += intel_cacheinfo.o obj-y += mwait-idle.o obj-y += shanghai.o -obj-y += vpmu.o vpmu_amd.o vpmu_intel.o +obj-y += vpmu.o +obj-$(CONFIG_SVM) += vpmu_amd.o +obj-$(CONFIG_VMX) += vpmu_intel.o diff --git a/xen/arch/x86/include/asm/vpmu.h b/xen/arch/x86/include/asm/vpmu.h index dae9b43dac..da86f2e420 100644 --- a/xen/arch/x86/include/asm/vpmu.h +++ b/xen/arch/x86/include/asm/vpmu.h @@ -11,6 +11,7 @@ #define __ASM_X86_HVM_VPMU_H_ #include <public/pmu.h> +#include <xen/err.h> #define vcpu_vpmu(vcpu) (&(vcpu)->arch.vpmu) #define vpmu_vcpu(vpmu) container_of((vpmu), struct vcpu, arch.vpmu) @@ -42,9 +43,27 @@ struct arch_vpmu_ops { #endif }; +#ifdef CONFIG_VMX const struct arch_vpmu_ops *core2_vpmu_init(void); +#else +static inline const struct arch_vpmu_ops* core2_vpmu_init(void) +{ + return ERR_PTR(-ENODEV); +} +#endif +#ifdef CONFIG_SVM const struct arch_vpmu_ops *amd_vpmu_init(void); const struct arch_vpmu_ops *hygon_vpmu_init(void); +#else +static inline const struct arch_vpmu_ops* amd_vpmu_init(void) +{ + return ERR_PTR(-ENODEV); +} +static inline const struct arch_vpmu_ops* hygon_vpmu_init(void) +{ + return ERR_PTR(-ENODEV); +} +#endif struct vpmu_struct { u32 flags;
Build AMD vPMU when CONFIG_SVM is on, and Intel vPMU when CONFIG_VMX is on respectively, allowing for a plaftorm-specific build. Also separate arch_vpmu_ops initializers using these options and static inline stubs. No functional change intended. Signed-off-by: Sergiy Kibrik <Sergiy_Kibrik@epam.com> --- xen/arch/x86/cpu/Makefile | 4 +++- xen/arch/x86/include/asm/vpmu.h | 19 +++++++++++++++++++ 2 files changed, 22 insertions(+), 1 deletion(-)