[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] [PATCH] Fix PV CPUID virtualization of XSave
Jan, Keir, Updated patch is attached. The patch will fix XSave CPUID virtualization for PV guests. The XSave area size returned by CPUID leaf D is changed dynamically depending on the XCR0. Tools/libxc only assigns a static value. The fix will adjust xsave area size during runtime. Note: This fix is already in HVM cpuid virtualization. And Dom0 is not affected, either. Signed-off-by: Shan Haitao <haitao.shan@xxxxxxxxx> diff -r e90438f6e6d1 xen/arch/x86/hvm/hvm.c --- a/xen/arch/x86/hvm/hvm.c Wed Sep 14 11:38:13 2011 +0100 +++ b/xen/arch/x86/hvm/hvm.c Mon Sep 19 14:34:15 2011 +0800 @@ -2426,7 +2426,7 @@ void hvm_cpuid(unsigned int input, unsig { /* reset EBX to default value first */ *ebx = XSTATE_AREA_MIN_SIZE; - for ( sub_leaf = 2; sub_leaf < 64; sub_leaf++ ) + for ( sub_leaf = 2; sub_leaf < 63; sub_leaf++ ) { if ( !(v->arch.xcr0 & (1ULL << sub_leaf)) ) continue; diff -r e90438f6e6d1 xen/arch/x86/traps.c --- a/xen/arch/x86/traps.c Wed Sep 14 11:38:13 2011 +0100 +++ b/xen/arch/x86/traps.c Mon Sep 19 14:34:15 2011 +0800 @@ -770,6 +770,30 @@ static void pv_cpuid(struct cpu_user_reg { if ( !cpuid_hypervisor_leaves(a, c, &a, &b, &c, &d) ) domain_cpuid(current->domain, a, c, &a, &b, &c, &d); + + switch ( a ) + { + case 0xd: + { + unsigned int sub_leaf, _eax, _ebx, _ecx, _edx; + /* EBX value of main leaf 0 depends on enabled xsave features */ + if ( c == 0 && current->arch.xcr0 ) + { + /* reset EBX to default value first */ + b = XSTATE_AREA_MIN_SIZE; + for ( sub_leaf = 2; sub_leaf < 63; sub_leaf++ ) + { + if ( !(current->arch.xcr0 & (1ULL << sub_leaf)) ) + continue; + domain_cpuid(current->domain, a, c, &_eax, &_ebx, &_ecx, + &_edx); + if ( (_eax + _ebx) > b ) + b = _eax + _ebx; + } + } + break; + } + } goto out; } 2011/9/18 Haitao Shan <maillists.shan@xxxxxxxxx>: > 2011/9/16 Jan Beulich <JBeulich@xxxxxxxx>: >>>>> On 16.09.11 at 02:46, Haitao Shan <maillists.shan@xxxxxxxxx> wrote: >>> Hi, Keir, >>> >>> The patch will fix XSave CPUID virtualization for PV guests. The XSave >>> area size returned by CPUID leaf D is changed dynamically depending on >>> the XCR0. Tools/libxc only assigns a static value. The fix will adjust >>> xsave area size during runtime. >>> >>> Note: This fix is already in HVM cpuid virtualization. And Dom0 is not >>> affected, either. >>> >>> Signed-off-by: Shan Haitao <haitao.shan@xxxxxxxxx> >>> >>> Shan Haitao >>> >>> diff -r 5fe770c8a8a3 xen/arch/x86/traps.c >>> --- a/xen/arch/x86/traps.c Tue Sep 06 15:49:40 2011 +0100 >>> +++ b/xen/arch/x86/traps.c Wed Sep 07 02:09:12 2011 +0800 >>> @@ -770,6 +770,30 @@ static void pv_cpuid(struct cpu_user_reg >>> { >>> if ( !cpuid_hypervisor_leaves(a, c, &a, &b, &c, &d) ) >>> domain_cpuid(current->domain, a, c, &a, &b, &c, &d); >>> + >>> + switch ( a ) >>> + { >>> + case 0xd: >>> + { >>> + unsigned int sub_leaf, _eax, _ebx, _ecx, _edx; >>> + /* EBX value of main leaf 0 depends on enabled xsave features >>> */ >>> + if ( c == 0 && current->arch.xcr0 ) >>> + { >>> + /* reset EBX to default value first */ >>> + b = XSTATE_AREA_MIN_SIZE; >>> + for ( sub_leaf = 2; sub_leaf < 64; sub_leaf++ ) >> >> Shouldn't the upper bound be 63 here (as bit 63 serves a different >> purpose, and if that bit was set code changes would be required in >> various other places)? >> >> Jan > Nice catch! I will update the patch. The same piece of code is > borrowed from hvm_cpuid(), where I can change the value from 64 to 63, > too. > > Shan Haitao > >> >>> + { >>> + if ( !(current->arch.xcr0 & (1ULL << sub_leaf)) ) >>> + continue; >>> + domain_cpuid(current->domain, a, c, &_eax, &_ebx, >>> &_ecx, >>> + &_edx); >>> + if ( (_eax + _ebx) > b ) >>> + b = _eax + _ebx; >>> + } >>> + } >>> + break; >>> + } >>> + } >>> goto out; >>> } >> >> >> >> > Attachment:
pv_xsave_cpuid_fix.patch _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |