KVM: CPUID: Enable supervisor XSAVE states in CPUID enumeration and XSS
RFC PATCH 1/2 KVM: CPUID: Enable supervisor XSAVE states in CPUID enumeration and XSS
CPUID.(EAX=DH, ECX={i}H i>=0) enumerates XSAVE related leaves/sub-leaves, but supervisor states are not taken into account. Meanwhile,more and more new features, e.g., CET, PT, LBR, rely on supervisor states to enhance performance, so updating related KVM code becomes necessary.
With Aaron Lewis's <[email protected]> patches in place, i.e., {c90992bfb080, 52297436199d, 864e2ab2b46d, 139a12cfe1a0, 9753d68865c5, 312a1c87798e, 78958563d802, c034f2aa8622, 7204160eb780}, this patch is to enable suppervisor XSAVE states support in CPUID enumeration and MSR IA32_XSS. KVM_SUPPORTED_XSS is a static mask for KVM/Guest supervisor states and guest_supported_xss is a dynamic mask which consolidates current host IA32_XSS and QEMU configuration together with static mask.
Right now, supervisor states in IA32_XSS haven't been used in upstreamed KVM code, so set KVM_SUPPORTED_XSS to 0 in the patch, new XSAVES related features can expand the macro to enable save/restore with XSAVES/XSTORS instruction.
To test the patch, I first set the KVM_SUPPORTED_XSS to 0x3900 and inject value to IA32_XSS too, 0x3900 corresponds to the most recent possible candidate supervisor states on Intel platforms, tested on TGL platform as results below:
cpuid.[d.0]: eax = 0x000002e7, ebx = 0x00000a88, ecx = 0x00000a88, edx = 0x00000000 cpuid.[d.1]: eax = 0x0000000f, ebx = 0x00000a38, ecx = 0x00003900, edx = 0x00000000 cpuid.[d.2]: eax = 0x00000100, ebx = 0x00000240, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.5]: eax = 0x00000040, ebx = 0x00000440, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.6]: eax = 0x00000200, ebx = 0x00000480, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.7]: eax = 0x00000400, ebx = 0x00000680, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.8]: eax = 0x00000080, ebx = 0x00000000, ecx = 0x00000001, edx = 0x00000000 cpuid.[d.9]: eax = 0x00000008, ebx = 0x00000a80, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.11]: eax = 0x00000010, ebx = 0x00000000, ecx = 0x00000001, edx = 0x00000000 cpuid.[d.12]: eax = 0x00000018, ebx = 0x00000000, ecx = 0x00000001, edx = 0x00000000 cpuid.[d.13]: eax = 0x00000008, ebx = 0x00000000, ecx = 0x00000001, edx = 0x00000000 bit[8] in MSR_IA32_XSS is supported
bit[11] in MSR_IA32_XSS is supported
bit[12] in MSR_IA32_XSS is supported
bit[13] in MSR_IA32_XSS is supported
Supported bit mask in MSR_IA32_XSS is : 0x3900
When IA32_XSS and KVM_SUPPORTED_XSS are 0, got below output:
cpuid.[d.0]: eax = 0x000002e7, ebx = 0x00000a88, ecx = 0x00000a88, edx = 0x00000000 cpuid.[d.1]: eax = 0x0000000f, ebx = 0x00000988, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.2]: eax = 0x00000100, ebx = 0x00000240, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.5]: eax = 0x00000040, ebx = 0x00000440, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.6]: eax = 0x00000200, ebx = 0x00000480, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.7]: eax = 0x00000400, ebx = 0x00000680, ecx = 0x00000000, edx = 0x00000000 cpuid.[d.9]: eax = 0x00000008, ebx = 0x00000a80, ecx = 0x00000000, edx = 0x00000000 Supported bit mask in MSR_IA32_XSS is : 0x0
--- arch/x86/include/asm/kvm_host.h | 1 + arch/x86/kvm/cpuid.c | 111 ++++++++++++++++++++++---------- arch/x86/kvm/x86.c | 4 +- arch/x86/kvm/x86.h | 8 +++ 4 files changed, 87 insertions(+), 37 deletions(-)
arch/x86/include/asm/kvm_host.h +1 -0
@@ -638,6 +638,7 @@ struct kvm_vcpu_arch {u64 xcr0;u64 guest_supported_xcr0; +u64 guest_supported_xss;u32 guest_xstate_size;struct kvm_pio_request pio;
arch/x86/kvm/cpuid.c +76 -35
153 lines not on this page · raw message
arch/x86/kvm/x86.c +2 -2
@@ -177,7 +177,7 @@ struct kvm_shared_msrs {static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;static struct kvm_shared_msrs __percpu *shared_msrs; -static u64 __read_mostly host_xss; +u64 __read_mostly host_xss;struct kvm_stats_debugfs_item debugfs_entries[] = { {"pf_fixed" ,VCPU_STAT (pf_fixed) },
@@ -2732,7 +2732,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)* RDMSR/WRMSR rather than XSAVES/XRSTORS to save/restore PT * MSRs. */ -if (data !=0 ) +if (data & ~vcpu->arch.guest_supported_xss)return 1 ; vcpu->arch.ia32_xss = data;break ;
arch/x86/kvm/x86.h +8 -0
@@ -296,6 +296,14 @@ int x86_emulate_instruction(struct kvm_vcpu *vcpu, unsigned long cr2, | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \ | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \ | XFEATURE_MASK_PKRU) + +/* +* In future, new XSS bits can be ORed here to make them available +* to KVM and guest, right now, it's 0, meaning no XSS bits are +* supported. +*/ +#define KVM_SUPPORTED_XSS 0 +extern u64 host_xcr0;extern u64 kvm_supported_xcr0 (void );
Signature · 1 line
2.17.2
RFC PATCH 2/2 KVM: tests: Selftest for xsave CPUID enumeration
Test CPUID(EAX=DH, ECX=i, i>=0) enumeration and supervisor XSAVE bits support in MSR IA32_XSS.
--- tools/testing/selftests/kvm/Makefile | 1 + .../selftests/kvm/x86_64/xsave_cpuid_test.c | 81 +++++++++++++++++++ 2 files changed, 82 insertions(+) create mode 100644 tools/testing/selftests/kvm/x86_64/xsave_cpuid_test.c
tools/testing/selftests/kvm/Makefile +1 -0
@@ -26,6 +26,7 @@ TEST_GEN_PROGS_x86_64 += x86_64/vmx_dirty_log_test TEST_GEN_PROGS_x86_64 += x86_64/vmx_set_nested_state_test TEST_GEN_PROGS_x86_64 += x86_64/vmx_tsc_adjust_test TEST_GEN_PROGS_x86_64 += x86_64/xss_msr_test +TEST_GEN_PROGS_x86_64 += x86_64/xsave_cpuid_test TEST_GEN_PROGS_x86_64 += clear_dirty_log_test TEST_GEN_PROGS_x86_64 += dirty_log_test TEST_GEN_PROGS_x86_64 += kvm_create_max_vcpus
tools/testing/selftests/kvm/x86_64/xsave_cpuid_test.c new file +81 -0
81 lines not on this page · raw message
Signature · 1 line
2.17.2
kbuild test robot
build bot[if your patch is applied to the wrong git tree, please drop us a note to help improve the system. BTW, we also suggest to use '--base' option to specify the base tree in git format-patch, please…
Hi Yang,
[FYI, it's a private test report for your RFC patch.]
[auto build test WARNING on vhost/linux-next]
[cannot apply to kvm/linux-next v5.6-rc1 next-20200213]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system. BTW, we also suggest to use '--base' option to specify the base tree in git format-patch, please see https://stackoverflow.com/a/37406982]
url: https://github.com/0day-ci/linux/commits/Yang-Weijiang/KVM-CPUID-Enable-supervisor-XSAVE-states-in-CPUID-enumeration-and-XSS/20200213-113112 base: https://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost.git linux-next reproduce: # apt-get install sparse # sparse version: v0.6.1-166-g7e4a5b6f-dirty make ARCH=x86_64 allmodconfig make C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__'
If you fix the issue, kindly add following tag
sparse warnings: (new ones prefixed by >>)
>> arch/x86/kvm/cpuid.c:66:5: sparse: sparse: symbol 'kvm_supported_xss' was not declared. Should it be static?
Please review and possibly fold the followup patch.
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org
kbuild test robot
build bot[RFC PATCH] KVM: CPUID: kvm_supported_xss() can be static
In reply to Yang WeijiangFixes: c0a25f0ace60 ("KVM: CPUID: Enable supervisor XSAVE states in CPUID enumeration and XSS")
--- cpuid.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/cpuid.c +1 -1
@@ -63,7 +63,7 @@ u64 kvm_supported_xcr0(void) }extern int host_xss; -u64 kvm_supported_xss (void ) +static u64 kvm_supported_xss (void ) {return KVM_SUPPORTED_XSS & host_xss; }
Quoted 75 lines · Yang Weijiang
CPUID.(EAX=DH, ECX={i}H i>=0) enumerates XSAVE related leaves/sub-leaves, but supervisor states are not taken into account. Meanwhile,more and more new features, e.g., CET, PT, LBR, rely on supervisor states to enhance performance, so updating related KVM code becomes necessary.
With Aaron Lewis's <[email protected]> patches in place, i.e., {
c90992bfb080,52297436199d,864e2ab2b46d,139a12cfe1a0,9753d68865c5,312a1c87798e,78958563d802,c034f2aa8622,7204160eb780}, this patch is to enable suppervisor XSAVE states support in CPUID enumeration and MSRIA32_XSS.KVM_SUPPORTED_XSSis a static mask for KVM/Guest supervisor states andguest_supported_xssis a dynamic mask which consolidates current hostIA32_XSSand QEMU configuration together with static mask.Right now, supervisor states in
IA32_XSShaven't been used in upstreamed KVM code, so setKVM_SUPPORTED_XSSto 0 in the patch, new XSAVES related features can expand the macro to enable save/restore with XSAVES/XSTORS instruction.To test the patch, I first set the
KVM_SUPPORTED_XSSto0x3900and inject value toIA32_XSStoo,0x3900corresponds to the most recent possible candidate supervisor states on Intel platforms, tested on TGL platform as results below:cpuid.[d.0]: eax =
0x000002e7, ebx =0x00000a88, ecx =0x00000a88, edx =0x00000000cpuid.[d.1]: eax =0x0000000f, ebx =0x00000a38, ecx =0x00003900, edx =0x00000000cpuid.[d.2]: eax =0x00000100, ebx =0x00000240, ecx =0x00000000, edx =0x00000000cpuid.[d.5]: eax =0x00000040, ebx =0x00000440, ecx =0x00000000, edx =0x00000000cpuid.[d.6]: eax =0x00000200, ebx =0x00000480, ecx =0x00000000, edx =0x00000000cpuid.[d.7]: eax =0x00000400, ebx =0x00000680, ecx =0x00000000, edx =0x00000000cpuid.[d.8]: eax =0x00000080, ebx =0x00000000, ecx =0x00000001, edx =0x00000000cpuid.[d.9]: eax =0x00000008, ebx =0x00000a80, ecx =0x00000000, edx =0x00000000cpuid.[d.11]: eax =0x00000010, ebx =0x00000000, ecx =0x00000001, edx =0x00000000cpuid.[d.12]: eax =0x00000018, ebx =0x00000000, ecx =0x00000001, edx =0x00000000cpuid.[d.13]: eax =0x00000008, ebx =0x00000000, ecx =0x00000001, edx =0x00000000bit[8] inMSR_IA32_XSSis supported
bit[11] inMSR_IA32_XSSis supported
bit[12] inMSR_IA32_XSSis supported
bit[13] inMSR_IA32_XSSis supported
Supported bit mask inMSR_IA32_XSSis :0x3900When
IA32_XSSandKVM_SUPPORTED_XSSare 0, got below output:
cpuid.[d.0]: eax =0x000002e7, ebx =0x00000a88, ecx =0x00000a88, edx =0x00000000cpuid.[d.1]: eax =0x0000000f, ebx =0x00000988, ecx =0x00000000, edx =0x00000000cpuid.[d.2]: eax =0x00000100, ebx =0x00000240, ecx =0x00000000, edx =0x00000000cpuid.[d.5]: eax =0x00000040, ebx =0x00000440, ecx =0x00000000, edx =0x00000000cpuid.[d.6]: eax =0x00000200, ebx =0x00000480, ecx =0x00000000, edx =0x00000000cpuid.[d.7]: eax =0x00000400, ebx =0x00000680, ecx =0x00000000, edx =0x00000000cpuid.[d.9]: eax =0x00000008, ebx =0x00000a80, ecx =0x00000000, edx =0x00000000Supported bit mask inMSR_IA32_XSSis :0x0Signed-off-by Yang Weijiang <[email protected]>--- arch/x86/include/asm/kvm_host.h | 1 + arch/x86/kvm/cpuid.c | 111 ++++++++++++++++++++++---------- arch/x86/kvm/x86.c | 4 +- arch/x86/kvm/x86.h | 8 +++ 4 files changed, 87 insertions(+), 37 deletions(-)arch/x86/include/asm/kvm_host.h +1
-0@@ -638,6 +638,7 @@ struct kvm_vcpu_arch {u64 xcr0;u64 guest_supported_xcr0; +u64 guest_supported_xss;u32 guest_xstate_size;struct kvm_pio_request pio;arch/x86/kvm/cpuid.c +0
-0
arch/x86/kvm/cpuid.c u64 kvm_supported_xcr0(void) · line 70
Quoted 9 lines
arch/x86/kvm/cpuid.c +6
-0@@ -62,6 +62,12 @@ u64 kvm_supported_xcr0(void)return xcr0; } +extern int host_xss; +u64 kvm_supported_xss (void ) +{ +return KVM_SUPPORTED_XSS & host_xss; +} +
How about using a global variable, supported_xss, instead of calculating the mask on every call. Just like what Sean posted on
https://lore.kernel.org/kvm/[email protected]/
Quoted 49 lines
arch/x86/kvm/cpuid.c +31
-3#define F (x) bit(X86_FEATURE_##x)int kvm_update_cpuid (struct kvm_vcpu *vcpu)@@ -112,10 +118,17 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu) vcpu->arch.guest_xstate_size = best->ebx =xstate_required_size (vcpu->arch.xcr0, false); } - best =kvm_find_cpuid_entry (vcpu,0xD ,1 ); -if (best && (best->eax & (F (XSAVES ) |F (XSAVEC )))) - best->ebx =xstate_required_size (vcpu->arch.xcr0, true); +if (best && (best->eax & (F (XSAVES ) |F (XSAVEC )))) { +u64 xstate = vcpu->arch.xcr0 | vcpu->arch.ia32_xss; + + best->ebx =xstate_required_size (xstate, true); + vcpu->arch.guest_supported_xss = + (best->ecx | ((u64 )best->edx <<32 )) & +kvm_supported_xss (); + }else { + vcpu->arch.guest_supported_xss =0 ; + }/* * The existing code assumes virtual address is 48-bit or 57-bit in the @@ -426,6 +439,56 @@ static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) } } +static inline bool do_cpuid_0xd_mask (struct kvm_cpuid_entry2 *entry,int index) +{ +unsigned int f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ; +/* cpuid 0xD.1.eax */ +const u32 kvm_cpuid_D_1_eax_x86_features = +F (XSAVEOPT ) |F (XSAVEC ) |F (XGETBV1 ) | f_xsaves; +u64 u_supported =kvm_supported_xcr0 (); +u64 s_supported =kvm_supported_xss (); +u64 supported; + +switch (index) { +case 0 : +if (!u_supported) { + entry->eax =0 ; + entry->ebx =0 ; + entry->ecx =0 ; + entry->edx =0 ; +return false; + } + entry->eax &= u_supported; + entry->ebx =xstate_required_size (u_supported, false);
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 472
Quoted 10 lines
arch/x86/kvm/cpuid.c +10
-0+ entry->ecx = entry->ebx; + entry->edx &= u_supported >>32 ; +break ; +case 1 : + supported = u_supported | s_supported; + entry->eax &= kvm_cpuid_D_1_eax_x86_features; +cpuid_mask (&entry->eax,CPUID_D_1_EAX ); + entry->ebx =0 ; + entry->edx &= s_supported >>32 ; + entry->ecx &= s_supported;
We'd better initialize msr_ia32_xss bitmap (entry->ecx & entry-edx) as zeros here.
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 474
Quoted 2 lines
arch/x86/kvm/cpuid.c +2
-0+if (entry->eax & (F (XSAVES ) |F (XSAVEC ))) + entry->ebx =xstate_required_size (supported, true);
And setup msr_ia32_xss bitmap based on the s_supported within this condition when F(XSAVES) is supported.
Quoted 130 lines
arch/x86/kvm/cpuid.c +27
-32+break ; +default : + supported = (entry->ecx &0x1 ) ? s_supported : u_supported; +if (!(supported & (BIT_ULL (index)))) { + entry->eax =0 ; + entry->ebx =0 ; + entry->ecx =0 ; + entry->edx =0 ; +return false; + } +if (entry->ecx &0x1 ) + entry->ebx =0 ; +break ; + } +return true; +} +static inline int __do_cpuid_func (struct kvm_cpuid_entry2 *entry,u32 function,int *nent,int maxnent) {@@ -440,7 +503,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,unsigned f_lm =0 ;#endif unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported () ?F (RDTSCP ) :0 ; -unsigned f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ;unsigned f_intel_pt = kvm_x86_ops->pt_supported () ?F (INTEL_PT ) :0 ;/* cpuid 1.edx */ @@ -495,10 +557,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,F (ACE2 ) |F (ACE2_EN ) |F (PHE ) |F (PHE_EN ) |F (PMM ) |F (PMM_EN ); -/* cpuid 0xD.1.eax */ -const u32 kvm_cpuid_D_1_eax_x86_features = -F (XSAVEOPT ) |F (XSAVEC ) |F (XGETBV1 ) | f_xsaves; -/* all calls to cpuid_count() should be made on the same cpu */ get_cpu ();@@ -639,38 +697,21 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,break ; }case 0xd : { -int idx, i; -u64 supported =kvm_supported_xcr0 (); +int i, idx; - entry->eax &= supported; - entry->ebx =xstate_required_size (supported, false); - entry->ecx = entry->ebx; - entry->edx &= supported >>32 ; -if (!supported) +if (!do_cpuid_0xd_mask (&entry[0 ],0 ))break ; - -for (idx =1 , i =1 ; idx <64 ; ++idx) { -u64 mask = ((u64 )1 << idx); +for (i =1 , idx =1 ; idx <64 ; ++idx) {if (*nent >= maxnent) goto out; -do_host_cpuid (&entry[i], function, idx); -if (idx ==1 ) { - entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; -cpuid_mask (&entry[i].eax,CPUID_D_1_EAX ); - entry[i].ebx =0 ; -if (entry[i].eax & (F (XSAVES )|F (XSAVEC ))) - entry[i].ebx = -xstate_required_size (supported, - true); - }else { -if (entry[i].eax ==0 || !(supported & mask)) -continue ; -if (WARN_ON_ONCE (entry[i].ecx &1 )) -continue ; - } - entry[i].ecx =0 ; - entry[i].edx =0 ; + +if (entry[i].eax ==0 && entry[i].ebx ==0 && + entry[i].ecx ==0 && entry[i].edx ==0 ) +continue ; + +if (!do_cpuid_0xd_mask (&entry[i], idx)) +continue ; ++*nent; ++i; }arch/x86/kvm/x86.c +2
-2@@ -177,7 +177,7 @@ struct kvm_shared_msrs {static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;static struct kvm_shared_msrs __percpu *shared_msrs; -static u64 __read_mostly host_xss; +u64 __read_mostly host_xss;struct kvm_stats_debugfs_item debugfs_entries[] = { {"pf_fixed" ,VCPU_STAT (pf_fixed) },@@ -2732,7 +2732,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)* RDMSR/WRMSR rather than XSAVES/XRSTORS to save/restore PT * MSRs. */ -if (data !=0 ) +if (data & ~vcpu->arch.guest_supported_xss)return 1 ; vcpu->arch.ia32_xss = data;break ;arch/x86/kvm/x86.h +8
-0@@ -296,6 +296,14 @@ int x86_emulate_instruction(struct kvm_vcpu *vcpu, unsigned long cr2, | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \ | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \ | XFEATURE_MASK_PKRU) + +/* +* In future, new XSS bits can be ORed here to make them available +* to KVM and guest, right now, it's 0, meaning no XSS bits are +* supported. +*/ +#define KVM_SUPPORTED_XSS 0 +extern u64 host_xcr0;extern u64 kvm_supported_xcr0 (void );
Quoted 13 lines · Xiaoyao Li
arch/x86/kvm/cpuid.c u64 kvm_supported_xcr0(void) · line 70
Quoted 7 lines · Yang Weijiang
CPUID.(EAX=DH, ECX={i}H i>=0) enumerates XSAVE related leaves/sub-leaves,
arch/x86/kvm/cpuid.c +6
-0+extern int host_xss; +u64 kvm_supported_xss (void ) +{ +return KVM_SUPPORTED_XSS & host_xss; +} +How about using a global variable,
supported_xss, instead of calculating the mask on every call. Just like what Sean posted on
https://lore.kernel.org/kvm/[email protected]/
Thanks Xiaoyao for the comments!
Good suggestion, I'll change it in next version.
Quoted 51 lines
Quoted 51 lines
#define F(x) bit(X86_FEATURE_##x) int kvm_update_cpuid(struct kvm_vcpu *vcpu)arch/x86/kvm/cpuid.c +36
-3@@ -112,10 +118,17 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu) vcpu->arch.guest_xstate_size = best->ebx =xstate_required_size (vcpu->arch.xcr0, false); } - best =kvm_find_cpuid_entry (vcpu,0xD ,1 ); -if (best && (best->eax & (F (XSAVES ) |F (XSAVEC )))) - best->ebx =xstate_required_size (vcpu->arch.xcr0, true); +if (best && (best->eax & (F (XSAVES ) |F (XSAVEC )))) { +u64 xstate = vcpu->arch.xcr0 | vcpu->arch.ia32_xss; + + best->ebx =xstate_required_size (xstate, true); + vcpu->arch.guest_supported_xss = + (best->ecx | ((u64 )best->edx <<32 )) & +kvm_supported_xss (); + }else { + vcpu->arch.guest_supported_xss =0 ; + }/* * The existing code assumes virtual address is 48-bit or 57-bit in the @@ -426,6 +439,56 @@ static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) } } +static inline bool do_cpuid_0xd_mask (struct kvm_cpuid_entry2 *entry,int index) +{ +unsigned int f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ; +/* cpuid 0xD.1.eax */ +const u32 kvm_cpuid_D_1_eax_x86_features = +F (XSAVEOPT ) |F (XSAVEC ) |F (XGETBV1 ) | f_xsaves; +u64 u_supported =kvm_supported_xcr0 (); +u64 s_supported =kvm_supported_xss (); +u64 supported; + +switch (index) { +case 0 : +if (!u_supported) { + entry->eax =0 ; + entry->ebx =0 ; + entry->ecx =0 ; + entry->edx =0 ; +return false; + } + entry->eax &= u_supported; + entry->ebx =xstate_required_size (u_supported, false); + entry->ecx = entry->ebx; + entry->edx &= u_supported >>32 ; +break ; +case 1 : + supported = u_supported | s_supported;
Quoted 8 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 472
Quoted 5 lines
arch/x86/kvm/cpuid.c +5
-0+ entry->eax &= kvm_cpuid_D_1_eax_x86_features; +cpuid_mask (&entry->eax,CPUID_D_1_EAX ); + entry->ebx =0 ; + entry->edx &= s_supported >>32 ; + entry->ecx &= s_supported;We'd better initialize
msr_ia32_xssbitmap (entry->ecx& entry-edx) as zeros here.
Hmm, explicit setting the MSR to 0 is good in this case, but there's implied flow to ensure guest MSR_IA32_XSS will be 0 if entry->ecx and entry->edx are 0s. In above kvm_update_cpuid(), vcpu->arch.guest_supported_xss is set to 0 when they're 0s. this masks guest cannot set non-zero value to this MSR. And in kvm_vcpu_reset(), vcpu->arch.ia32_xss is initialized to 0, in kvm_load_guest_xsave_state() MSR_IA32_XSS is set to ia32_xss, therefore the MSR is kept to 0.
Quoted 6 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 474
Quoted 3 lines
arch/x86/kvm/cpuid.c +2
-0+if (entry->eax & (F (XSAVES ) |F (XSAVEC ))) + entry->ebx =xstate_required_size (supported, true);And setup
msr_ia32_xssbitmap based on thes_supportedwithin this condition when F(XSAVES) is supported.
IIUC, both XSAVEC and XSAVES use compacted format of the extended region, so if XSAVEC is supported while XSAVES is not, guest still can get correct size, so in existing code the two bits are ORed.
Quoted 86 lines
Quoted 85 lines
arch/x86/kvm/cpuid.c +27
-32+break ; +default : + supported = (entry->ecx &0x1 ) ? s_supported : u_supported; +if (!(supported & (BIT_ULL (index)))) { + entry->eax =0 ; + entry->ebx =0 ; + entry->ecx =0 ; + entry->edx =0 ; +return false; + } +if (entry->ecx &0x1 ) + entry->ebx =0 ; +break ; + } +return true; +} +static inline int __do_cpuid_func (struct kvm_cpuid_entry2 *entry,u32 function,int *nent,int maxnent) {@@ -440,7 +503,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,unsigned f_lm =0 ;#endif unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported () ?F (RDTSCP ) :0 ; -unsigned f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ;unsigned f_intel_pt = kvm_x86_ops->pt_supported () ?F (INTEL_PT ) :0 ;/* cpuid 1.edx */ @@ -495,10 +557,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,F (ACE2 ) |F (ACE2_EN ) |F (PHE ) |F (PHE_EN ) |F (PMM ) |F (PMM_EN ); -/* cpuid 0xD.1.eax */ -const u32 kvm_cpuid_D_1_eax_x86_features = -F (XSAVEOPT ) |F (XSAVEC ) |F (XGETBV1 ) | f_xsaves; -/* all calls to cpuid_count() should be made on the same cpu */ get_cpu ();@@ -639,38 +697,21 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,break ; }case 0xd : { -int idx, i; -u64 supported =kvm_supported_xcr0 (); +int i, idx; - entry->eax &= supported; - entry->ebx =xstate_required_size (supported, false); - entry->ecx = entry->ebx; - entry->edx &= supported >>32 ; -if (!supported) +if (!do_cpuid_0xd_mask (&entry[0 ],0 ))break ; - -for (idx =1 , i =1 ; idx <64 ; ++idx) { -u64 mask = ((u64 )1 << idx); +for (i =1 , idx =1 ; idx <64 ; ++idx) {if (*nent >= maxnent) goto out; -do_host_cpuid (&entry[i], function, idx); -if (idx ==1 ) { - entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; -cpuid_mask (&entry[i].eax,CPUID_D_1_EAX ); - entry[i].ebx =0 ; -if (entry[i].eax & (F (XSAVES )|F (XSAVEC ))) - entry[i].ebx = -xstate_required_size (supported, - true); - }else { -if (entry[i].eax ==0 || !(supported & mask)) -continue ; -if (WARN_ON_ONCE (entry[i].ecx &1 )) -continue ; - } - entry[i].ecx =0 ; - entry[i].edx =0 ; + +if (entry[i].eax ==0 && entry[i].ebx ==0 && + entry[i].ecx ==0 && entry[i].edx ==0 ) +continue ; + +if (!do_cpuid_0xd_mask (&entry[i], idx)) +continue ; ++*nent; ++i; }
Quoted 76 lines · Yang Weijiang
Quoted 13 lines · Xiaoyao Li
arch/x86/kvm/cpuid.c u64 kvm_supported_xcr0(void) · line 70
Quoted 7 lines · Yang Weijiang
CPUID.(EAX=DH, ECX={i}H i>=0) enumerates XSAVE related leaves/sub-leaves,
arch/x86/kvm/cpuid.c +6
-0+extern int host_xss; +u64 kvm_supported_xss (void ) +{ +return KVM_SUPPORTED_XSS & host_xss; +} +How about using a global variable,
supported_xss, instead of calculating the mask on every call. Just like what Sean posted on
https://lore.kernel.org/kvm/[email protected]/Thanks Xiaoyao for the comments!
Good suggestion, I'll change it in next version.Quoted 51 lines
Quoted 51 lines
#define F(x) bit(X86_FEATURE_##x) int kvm_update_cpuid(struct kvm_vcpu *vcpu)arch/x86/kvm/cpuid.c +36
-3@@ -112,10 +118,17 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu) vcpu->arch.guest_xstate_size = best->ebx =xstate_required_size (vcpu->arch.xcr0, false); } - best =kvm_find_cpuid_entry (vcpu,0xD ,1 ); -if (best && (best->eax & (F (XSAVES ) |F (XSAVEC )))) - best->ebx =xstate_required_size (vcpu->arch.xcr0, true); +if (best && (best->eax & (F (XSAVES ) |F (XSAVEC )))) { +u64 xstate = vcpu->arch.xcr0 | vcpu->arch.ia32_xss; + + best->ebx =xstate_required_size (xstate, true); + vcpu->arch.guest_supported_xss = + (best->ecx | ((u64 )best->edx <<32 )) & +kvm_supported_xss (); + }else { + vcpu->arch.guest_supported_xss =0 ; + }/* * The existing code assumes virtual address is 48-bit or 57-bit in the @@ -426,6 +439,56 @@ static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) } } +static inline bool do_cpuid_0xd_mask (struct kvm_cpuid_entry2 *entry,int index) +{ +unsigned int f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ; +/* cpuid 0xD.1.eax */ +const u32 kvm_cpuid_D_1_eax_x86_features = +F (XSAVEOPT ) |F (XSAVEC ) |F (XGETBV1 ) | f_xsaves; +u64 u_supported =kvm_supported_xcr0 (); +u64 s_supported =kvm_supported_xss (); +u64 supported; + +switch (index) { +case 0 : +if (!u_supported) { + entry->eax =0 ; + entry->ebx =0 ; + entry->ecx =0 ; + entry->edx =0 ; +return false; + } + entry->eax &= u_supported; + entry->ebx =xstate_required_size (u_supported, false); + entry->ecx = entry->ebx; + entry->edx &= u_supported >>32 ; +break ; +case 1 : + supported = u_supported | s_supported;Quoted 8 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 472
Quoted 5 lines
arch/x86/kvm/cpuid.c +5
-0+ entry->eax &= kvm_cpuid_D_1_eax_x86_features; +cpuid_mask (&entry->eax,CPUID_D_1_EAX ); + entry->ebx =0 ; + entry->edx &= s_supported >>32 ; + entry->ecx &= s_supported;We'd better initialize
msr_ia32_xssbitmap (entry->ecx& entry-edx) as zeros here.
Quoted 7 lines
Hmm, explicit setting the MSR to 0 is good in this case, but there's implied flow to ensure guest
MSR_IA32_XSSwill be 0 ifentry->ecxandentry->edxare 0s. In abovekvm_update_cpuid(),vcpu->arch.guest_supported_xssis set to 0 when they're 0s. this masks guest cannot set non-zero value to this MSR. And inkvm_vcpu_reset(),vcpu->arch.ia32_xssis initialized to 0, inkvm_load_guest_xsave_state()MSR_IA32_XSSis set toia32_xss, therefore the MSR is kept to 0.
Sorry, I think what I said "msr_ia32_xss bitmap" misled you.
"msr_ia32_xss bitmap" is not MSR_IA32_XSS, but the (entry->ecx | entry->edx >> 32) of cpuid.D_1
I meant we'd better set entry->ecx and entry->edx to 0 here.
Quoted 7 lines
Quoted 6 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 474
Quoted 3 lines
arch/x86/kvm/cpuid.c +2
-0+if (entry->eax & (F (XSAVES ) |F (XSAVEC ))) + entry->ebx =xstate_required_size (supported, true);And setup
msr_ia32_xssbitmap based on thes_supportedwithin this condition when F(XSAVES) is supported.
And set entry->ecx & entry->edx only when F(XSAVES) is supported.
Quoted 3 lines
IIUC, both XSAVEC and XSAVES use compacted format of the extended region, so if XSAVEC is supported while XSAVES is not, guest still can get correct size, so in existing code the two bits are ORed.
Yeah, but entry->ecx and entry->edx should be non-zero only when F(XSAVES) is set.
Quoted 87 lines
Quoted 86 lines
Quoted 85 lines
arch/x86/kvm/cpuid.c +27
-32+break ; +default : + supported = (entry->ecx &0x1 ) ? s_supported : u_supported; +if (!(supported & (BIT_ULL (index)))) { + entry->eax =0 ; + entry->ebx =0 ; + entry->ecx =0 ; + entry->edx =0 ; +return false; + } +if (entry->ecx &0x1 ) + entry->ebx =0 ; +break ; + } +return true; +} +static inline int __do_cpuid_func (struct kvm_cpuid_entry2 *entry,u32 function,int *nent,int maxnent) {@@ -440,7 +503,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,unsigned f_lm =0 ;#endif unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported () ?F (RDTSCP ) :0 ; -unsigned f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ;unsigned f_intel_pt = kvm_x86_ops->pt_supported () ?F (INTEL_PT ) :0 ;/* cpuid 1.edx */ @@ -495,10 +557,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,F (ACE2 ) |F (ACE2_EN ) |F (PHE ) |F (PHE_EN ) |F (PMM ) |F (PMM_EN ); -/* cpuid 0xD.1.eax */ -const u32 kvm_cpuid_D_1_eax_x86_features = -F (XSAVEOPT ) |F (XSAVEC ) |F (XGETBV1 ) | f_xsaves; -/* all calls to cpuid_count() should be made on the same cpu */ get_cpu ();@@ -639,38 +697,21 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,break ; }case 0xd : { -int idx, i; -u64 supported =kvm_supported_xcr0 (); +int i, idx; - entry->eax &= supported; - entry->ebx =xstate_required_size (supported, false); - entry->ecx = entry->ebx; - entry->edx &= supported >>32 ; -if (!supported) +if (!do_cpuid_0xd_mask (&entry[0 ],0 ))break ; - -for (idx =1 , i =1 ; idx <64 ; ++idx) { -u64 mask = ((u64 )1 << idx); +for (i =1 , idx =1 ; idx <64 ; ++idx) {if (*nent >= maxnent) goto out; -do_host_cpuid (&entry[i], function, idx); -if (idx ==1 ) { - entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; -cpuid_mask (&entry[i].eax,CPUID_D_1_EAX ); - entry[i].ebx =0 ; -if (entry[i].eax & (F (XSAVES )|F (XSAVEC ))) - entry[i].ebx = -xstate_required_size (supported, - true); - }else { -if (entry[i].eax ==0 || !(supported & mask)) -continue ; -if (WARN_ON_ONCE (entry[i].ecx &1 )) -continue ; - } - entry[i].ecx =0 ; - entry[i].edx =0 ; + +if (entry[i].eax ==0 && entry[i].ebx ==0 && + entry[i].ecx ==0 && entry[i].edx ==0 ) +continue ; + +if (!do_cpuid_0xd_mask (&entry[i], idx)) +continue ; ++*nent; ++i; }
Quoted 31 lines · Xiaoyao Li
Quoted 30 lines · Yang Weijiang
Quoted 14 lines · Xiaoyao Li
arch/x86/kvm/cpuid.c u64 kvm_supported_xcr0(void) · line 70
Quoted 7 lines · Yang Weijiang
CPUID.(EAX=DH, ECX={i}H i>=0) enumerates XSAVE related leaves/sub-leaves,
arch/x86/kvm/cpuid.c +6
-0+extern int host_xss; +u64 kvm_supported_xss (void ) +{ + return KVM_SUPPORTED_XSS & host_xss; +} +How about using a global variable,
supported_xss, instead of calculating the mask on every call. Just like what Sean posted on
https://lore.kernel.org/kvm/[email protected]/Thanks Xiaoyao for the comments!
Good suggestion, I'll change it in next version.Quoted 12 lines
Quoted 12 lines
#define F(x)
bit(X86_FEATURE_##x)
intkvm_update_cpuid(struct kvm_vcpu *vcpu)(quoted diff) +0
-0@@ -112,10 +118,17 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu)
vcpu->arch.guest_xstate_size=best->ebx=xstate_required_size(vcpu->arch.xcr0, false); }
-
best =kvm_find_cpuid_entry(vcpu, 0xD, 1);
- if (best && (best->eax& (F(XSAVES) | F(XSAVEC)))) -best->ebx=xstate_required_size(vcpu->arch.xcr0, true); + if (best && (best->eax& (F(XSAVES) | F(XSAVEC)))) { + u64 xstate =vcpu->arch.xcr0|vcpu->arch.ia32_xss;
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
+
+best->ebx=xstate_required_size(xstate, true); +vcpu->arch.guest_supported_xss=
+ (best->ecx| ((u64)best->edx << 32)) & +kvm_supported_xss();
+ } else {
+vcpu->arch.guest_supported_xss= 0;
+ }
also here should be something like below:
if (best) {
if (best->eax & (F(XSAVES) | F(XSAVEC)))) {
u64 xstate = vcpu->arch.xcr0 | vcpu->arch.ia32_xss;
best->ebx = xstate_required_size(xstate, true);
}
if (best->eax & F(XSAVES) {
vcpu->arch.guest_supported_xss =
(best->ecx | ((u64)best->edx << 32)) &
kvm_supported_xss();
} else {
best->ecx = 0;
best->edx = 0;
vcpu->arch.guest_supported_xss = 0;
}
}Quoted 165 lines
Quoted 42 lines
Quoted 34 lines
Quoted 34 lines
/*
* The existing code assumes virtual address is 48-bit or 57-bit in thearch/x86/kvm/cpuid.c +26
-0@@ -426,6 +439,56 @@ static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) } } +static inline bool do_cpuid_0xd_mask (struct kvm_cpuid_entry2 *entry,int index) +{ + unsigned int f_xsaves = kvm_x86_ops->xsaves_supported() ?F (XSAVES ) :0 ; + /* cpuid 0xD.1.eax */ + const u32 kvm_cpuid_D_1_eax_x86_features = + F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves; + u64 u_supported = kvm_supported_xcr0(); + u64 s_supported = kvm_supported_xss(); + u64 supported; + + switch (index) { + case 0: + if (!u_supported) { + entry->eax = 0; + entry->ebx = 0; + entry->ecx = 0; + entry->edx = 0; + return false; + } + entry->eax &= u_supported; + entry->ebx = xstate_required_size(u_supported, false); + entry->ecx = entry->ebx; + entry->edx &= u_supported >> 32; + break; + case 1: + supported = u_supported | s_supported;Quoted 8 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 472
Quoted 5 lines
arch/x86/kvm/cpuid.c +5
-0+ entry->eax &= kvm_cpuid_D_1_eax_x86_features; + cpuid_mask(&entry->eax, CPUID_D_1_EAX); + entry->ebx = 0; + entry->edx &= s_supported >> 32; + entry->ecx &= s_supported;We'd better initialize
msr_ia32_xssbitmap (entry->ecx& entry-edx) as zeros here.Quoted 7 lines
Hmm, explicit setting the MSR to 0 is good in this case, but there's implied flow to ensure guest
MSR_IA32_XSSwill be 0 ifentry->ecxandentry->edxare 0s. In abovekvm_update_cpuid(),vcpu->arch.guest_supported_xssis set to 0 when they're 0s. this masks guest cannot set non-zero value to this MSR. And inkvm_vcpu_reset(),vcpu->arch.ia32_xssis initialized to 0, inkvm_load_guest_xsave_state()MSR_IA32_XSSis set toia32_xss, therefore the MSR is kept to 0.Sorry, I think what I said "
msr_ia32_xssbitmap" misled you."
msr_ia32_xssbitmap" is notMSR_IA32_XSS, but the (entry->ecx|entry->edx>> 32) of cpuid.D_1I meant we'd better set
entry->ecxandentry->edxto 0 here.Quoted 7 lines
Quoted 6 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 474
Quoted 3 lines
arch/x86/kvm/cpuid.c +2
-0+ if (entry->eax & (F(XSAVES) | F(XSAVEC))) + entry->ebx = xstate_required_size(supported, true);And setup
msr_ia32_xssbitmap based on thes_supportedwithin this condition when F(XSAVES) is supported.And set
entry->ecx&entry->edxonly when F(XSAVES) is supported.Quoted 3 lines
IIUC, both XSAVEC and XSAVES use compacted format of the extended region, so if XSAVEC is supported while XSAVES is not, guest still can get correct size, so in existing code the two bits are ORed.
Yeah, but
entry->ecxandentry->edxshould be non-zero only when F(XSAVES) is set.Quoted 91 lines
Quoted 81 lines
Quoted 81 lines
arch/x86/kvm/cpuid.c +17
-0+ break; + default: + supported = (entry->ecx & 0x1) ? s_supported : u_supported; + if (!(supported & (BIT_ULL(index)))) { + entry->eax = 0; + entry->ebx = 0; + entry->ecx = 0; + entry->edx = 0; + return false; + } + if (entry->ecx & 0x1) + entry->ebx = 0; + break; + } + return true; +} + static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,int *nent, int maxnent)
{arch/x86/kvm/cpuid.c +0
-1@@ -440,7 +503,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function, unsigned f_lm = 0; #endif unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0; - unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0; unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;/* cpuid 1.edx */
(quoted diff) +0
-0@@ -495,10 +557,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,F(ACE2) |
F(ACE2_EN)| F(PHE) |F(PHE_EN)| F(PMM) |F(PMM_EN);
- /* cpuid 0xD.1.eax */
- const u32kvm_cpuid_D_1_eax_x86_features=
- F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) |f_xsaves;
-
/* all calls tocpuid_count()should be made on the same cpu */get_cpu();arch/x86/kvm/cpuid.c +5
-27@@ -639,38 +697,21 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function, break; } case 0xd: { - int idx, i; - u64 supported = kvm_supported_xcr0(); + int i, idx; - entry->eax &= supported; - entry->ebx = xstate_required_size(supported, false); - entry->ecx = entry->ebx; - entry->edx &= supported >> 32; - if (!supported) + if (!do_cpuid_0xd_mask(&entry[0], 0)) break; - - for (idx = 1, i = 1; idx < 64; ++idx) { - u64 mask = ((u64)1 << idx); + for (i = 1, idx = 1; idx < 64; ++idx) { if (*nent >= maxnent) goto out; - do_host_cpuid(&entry[i], function, idx); - if (idx == 1) { - entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; - cpuid_mask(&entry[i].eax, CPUID_D_1_EAX); - entry[i].ebx = 0; - if (entry[i].eax & (F(XSAVES)|F(XSAVEC))) - entry[i].ebx = - xstate_required_size(supported, - true); - } else { - if (entry[i].eax == 0 || !(supported & mask)) - continue; - if (WARN_ON_ONCE(entry[i].ecx & 1)) - continue; - } - entry[i].ecx = 0; - entry[i].edx = 0; + + if (entry[i].eax == 0 && entry[i].ebx == 0 &&Quoted 8 lines
arch/x86/kvm/cpuid.c static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function, · line 717
Quoted 8 lines
arch/x86/kvm/cpuid.c +5
-0+ entry[i].ecx == 0 && entry[i].edx == 0) + continue; + + if (!do_cpuid_0xd_mask(&entry[i], idx)) + continue; ++*nent; ++i; }>
Quoted 45 lines · Xiaoyao Li
Quoted 33 lines · Xiaoyao Li
Quoted 32 lines · Yang Weijiang
Quoted 16 lines · Xiaoyao Li
arch/x86/kvm/cpuid.c u64 kvm_supported_xcr0(void) · line 70
Quoted 8 lines · Yang Weijiang
CPUID.(EAX=DH, ECX={i}H i>=0) enumerates XSAVE related leaves/sub-leaves,
arch/x86/kvm/cpuid.c +6
-0+extern int host_xss; +u64 kvm_supported_xss (void ) +{ + return KVM_SUPPORTED_XSS & host_xss; +} +How about using a global variable,
supported_xss, instead of calculating the
mask on every call. Just like what Sean posted on
https://lore.kernel.org/kvm/[email protected]/Thanks Xiaoyao for the comments!
Good suggestion, I'll change it in next version.Quoted 12 lines
Quoted 12 lines
#define F(x)
bit(X86_FEATURE_##x)
intkvm_update_cpuid(struct kvm_vcpu *vcpu)(quoted diff) +0
-0@@ -112,10 +118,17 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu)
vcpu->arch.guest_xstate_size=best->ebx=xstate_required_size(vcpu->arch.xcr0, false); }
-
best =kvm_find_cpuid_entry(vcpu, 0xD, 1);
- if (best && (best->eax& (F(XSAVES) | F(XSAVEC)))) -best->ebx=xstate_required_size(vcpu->arch.xcr0, true); + if (best && (best->eax& (F(XSAVES) | F(XSAVEC)))) { + u64 xstate =vcpu->arch.xcr0|vcpu->arch.ia32_xss;Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
+
+best->ebx=xstate_required_size(xstate, true); +vcpu->arch.guest_supported_xss=
+ (best->ecx| ((u64)best->edx << 32)) & +kvm_supported_xss();
+ } else {
+vcpu->arch.guest_supported_xss= 0;
+ }also here should be something like below:
Quoted 16 lines
if (best) {
if (best->eax & (F(XSAVES) | F(XSAVEC)))) {
u64 xstate = vcpu->arch.xcr0 | vcpu->arch.ia32_xss;
best->ebx = xstate_required_size(xstate, true);
}
if (best->eax & F(XSAVES) {
vcpu->arch.guest_supported_xss =
(best->ecx | ((u64)best->edx << 32)) &
kvm_supported_xss();
} else {
best->ecx = 0;
best->edx = 0;
vcpu->arch.guest_supported_xss = 0;
}
}Right, to keep it consistent with CPUID enumeration, thanks for review!
Quoted 177 lines
Quoted 177 lines
Quoted 45 lines
Quoted 37 lines
Quoted 37 lines
/*
* The existing code assumes virtual address is 48-bit or 57-bit in thearch/x86/kvm/cpuid.c +27
-0@@ -426,6 +439,56 @@ static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) } } +static inline bool do_cpuid_0xd_mask (struct kvm_cpuid_entry2 *entry,int index) +{ + unsigned int f_xsaves = kvm_x86_ops->xsaves_supported() ?F (XSAVES ) :0 ; + /* cpuid 0xD.1.eax */ + const u32 kvm_cpuid_D_1_eax_x86_features = + F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves; + u64 u_supported = kvm_supported_xcr0(); + u64 s_supported = kvm_supported_xss(); + u64 supported; + + switch (index) { + case 0: + if (!u_supported) { + entry->eax = 0; + entry->ebx = 0; + entry->ecx = 0; + entry->edx = 0; + return false; + } + entry->eax &= u_supported; + entry->ebx = xstate_required_size(u_supported, false); + entry->ecx = entry->ebx; + entry->edx &= u_supported >> 32; + break; + case 1: + supported = u_supported | s_supported; + entry->eax &= kvm_cpuid_D_1_eax_x86_features;Quoted 8 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 472
Quoted 4 lines
arch/x86/kvm/cpuid.c +4
-0+ cpuid_mask(&entry->eax, CPUID_D_1_EAX); + entry->ebx = 0; + entry->edx &= s_supported >> 32; + entry->ecx &= s_supported;We'd better initialize
msr_ia32_xssbitmap (entry->ecx& entry-edx) as zeros
here.Quoted 9 lines
Hmm, explicit setting the MSR to 0 is good in this case, but there's implied
flow to ensure guestMSR_IA32_XSSwill be 0 ifentry->ecxandentry->edxare 0s.
In abovekvm_update_cpuid(),vcpu->arch.guest_supported_xssis set to 0 when they're 0s. this masks guest cannot set non-zero value to this MSR. And inkvm_vcpu_reset(),vcpu->arch.ia32_xssis initialized to 0, inkvm_load_guest_xsave_state()MSR_IA32_XSSis set toia32_xss, therefore the MSR is kept to 0.Sorry, I think what I said "
msr_ia32_xssbitmap" misled you."
msr_ia32_xssbitmap" is notMSR_IA32_XSS, but the (entry->ecx|entry->edx>> 32) of cpuid.D_1I meant we'd better set
entry->ecxandentry->edxto 0 here.Quoted 8 lines
Quoted 7 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 474
Quoted 3 lines
arch/x86/kvm/cpuid.c +2
-0+ if (entry->eax & (F(XSAVES) | F(XSAVEC))) + entry->ebx = xstate_required_size(supported, true);And setup
msr_ia32_xssbitmap based on thes_supportedwithin this condition
when F(XSAVES) is supported.And set
entry->ecx&entry->edxonly when F(XSAVES) is supported.Quoted 3 lines
IIUC, both XSAVEC and XSAVES use compacted format of the extended region, so if XSAVEC is supported while XSAVES is not, guest still can get correct size, so in existing code the two bits are ORed.
Yeah, but
entry->ecxandentry->edxshould be non-zero only when F(XSAVES) is set.Quoted 96 lines
Quoted 86 lines
Quoted 86 lines
arch/x86/kvm/cpuid.c +17
-0+ break; + default: + supported = (entry->ecx & 0x1) ? s_supported : u_supported; + if (!(supported & (BIT_ULL(index)))) { + entry->eax = 0; + entry->ebx = 0; + entry->ecx = 0; + entry->edx = 0; + return false; + } + if (entry->ecx & 0x1) + entry->ebx = 0; + break; + } + return true; +} +static inline int __do_cpuid_func(struct
kvm_cpuid_entry2*entry, u32 function,
int *nent, int maxnent)
{arch/x86/kvm/cpuid.c +0
-0@@ -440,7 +503,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function, unsigned f_lm = 0; #endifunsigned
f_rdtscp=kvm_x86_ops->rdtscp_supported()? F(RDTSCP) : 0;
- unsignedf_xsaves=kvm_x86_ops->xsaves_supported()? F(XSAVES) : 0;
unsignedf_intel_pt=kvm_x86_ops->pt_supported()?F(INTEL_PT): 0;
/* cpuid 1.edx */(quoted diff) +0
-0@@ -495,10 +557,6 @@ static inline int__do_cpuid_func(struct
kvm_cpuid_entry2*entry, u32 function,
F(ACE2) |F(ACE2_EN)| F(PHE) |F(PHE_EN)| F(PMM) |F(PMM_EN);
- /* cpuid 0xD.1.eax */
- const u32kvm_cpuid_D_1_eax_x86_features=
- F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) |f_xsaves;
-
/* all calls tocpuid_count()should be made on the same cpu */get_cpu();(quoted diff) +0
-0@@ -639,38 +697,21 @@ static inline int__do_cpuid_func(struct
kvm_cpuid_entry2*entry, u32 function,
break;
}
case0xd: {
- int idx, i;
- u64 supported =kvm_supported_xcr0();
+ int i, idx;
-entry->eax&= supported;
-entry->ebx=xstate_required_size(supported, false); -entry->ecx=entry->ebx;
-entry->edx&= supported >> 32;
- if (!supported)
+ if (!do_cpuid_0xd_mask(&entry[0], 0))
break;
-
- for (idx = 1, i = 1; idx < 64; ++idx) {
- u64 mask = ((u64)1 << idx);
+ for (i = 1, idx = 1; idx < 64; ++idx) {
if (*nent >= maxnent)
goto out;
-
do_host_cpuid(&entry[i], function, idx); - if (idx == 1) {
- entry[i].eax &=kvm_cpuid_D_1_eax_x86_features; -cpuid_mask(&entry[i].eax, CPUID_D_1_EAX); - entry[i].ebx = 0;
- if (entry[i].eax & (F(XSAVES)|F(XSAVEC))) - entry[i].ebx =
-xstate_required_size(supported, - true);
- } else {
- if (entry[i].eax == 0 || !(supported & mask)) - continue;
- if (WARN_ON_ONCE(entry[i].ecx & 1))
- continue;
- }
- entry[i].ecx = 0;
- entry[i].edx = 0;
+
+ if (entry[i].eax == 0 && entry[i].ebx == 0 &&Quoted 8 lines
Quoted 8 lines
+ entry[i].ecx == 0 && entry[i].edx == 0) + continue;
+
+ if (!do_cpuid_0xd_mask(&entry[i], idx)) + continue;
++*nent;
++i;
}>
Quoted 93 lines · Xiaoyao Li
Quoted 76 lines · Yang Weijiang
Quoted 13 lines · Xiaoyao Li
arch/x86/kvm/cpuid.c u64 kvm_supported_xcr0(void) · line 70
Quoted 7 lines · Yang Weijiang
CPUID.(EAX=DH, ECX={i}H i>=0) enumerates XSAVE related leaves/sub-leaves,
arch/x86/kvm/cpuid.c +6
-0+extern int host_xss; +u64 kvm_supported_xss (void ) +{ +return KVM_SUPPORTED_XSS & host_xss; +} +How about using a global variable,
supported_xss, instead of calculating the mask on every call. Just like what Sean posted on
https://lore.kernel.org/kvm/[email protected]/Thanks Xiaoyao for the comments!
Good suggestion, I'll change it in next version.Quoted 51 lines
Quoted 51 lines
#define F(x) bit(X86_FEATURE_##x) int kvm_update_cpuid(struct kvm_vcpu *vcpu)arch/x86/kvm/cpuid.c +36
-3@@ -112,10 +118,17 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu) vcpu->arch.guest_xstate_size = best->ebx =xstate_required_size (vcpu->arch.xcr0, false); } - best =kvm_find_cpuid_entry (vcpu,0xD ,1 ); -if (best && (best->eax & (F (XSAVES ) |F (XSAVEC )))) - best->ebx =xstate_required_size (vcpu->arch.xcr0, true); +if (best && (best->eax & (F (XSAVES ) |F (XSAVEC )))) { +u64 xstate = vcpu->arch.xcr0 | vcpu->arch.ia32_xss; + + best->ebx =xstate_required_size (xstate, true); + vcpu->arch.guest_supported_xss = + (best->ecx | ((u64 )best->edx <<32 )) & +kvm_supported_xss (); + }else { + vcpu->arch.guest_supported_xss =0 ; + }/* * The existing code assumes virtual address is 48-bit or 57-bit in the @@ -426,6 +439,56 @@ static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) } } +static inline bool do_cpuid_0xd_mask (struct kvm_cpuid_entry2 *entry,int index) +{ +unsigned int f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ; +/* cpuid 0xD.1.eax */ +const u32 kvm_cpuid_D_1_eax_x86_features = +F (XSAVEOPT ) |F (XSAVEC ) |F (XGETBV1 ) | f_xsaves; +u64 u_supported =kvm_supported_xcr0 (); +u64 s_supported =kvm_supported_xss (); +u64 supported; + +switch (index) { +case 0 : +if (!u_supported) { + entry->eax =0 ; + entry->ebx =0 ; + entry->ecx =0 ; + entry->edx =0 ; +return false; + } + entry->eax &= u_supported; + entry->ebx =xstate_required_size (u_supported, false); + entry->ecx = entry->ebx; + entry->edx &= u_supported >>32 ; +break ; +case 1 : + supported = u_supported | s_supported;Quoted 8 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 472
Quoted 5 lines
arch/x86/kvm/cpuid.c +5
-0+ entry->eax &= kvm_cpuid_D_1_eax_x86_features; +cpuid_mask (&entry->eax,CPUID_D_1_EAX ); + entry->ebx =0 ; + entry->edx &= s_supported >>32 ; + entry->ecx &= s_supported;We'd better initialize
msr_ia32_xssbitmap (entry->ecx& entry-edx) as zeros here.Quoted 7 lines
Hmm, explicit setting the MSR to 0 is good in this case, but there's implied flow to ensure guest
MSR_IA32_XSSwill be 0 ifentry->ecxandentry->edxare 0s. In abovekvm_update_cpuid(),vcpu->arch.guest_supported_xssis set to 0 when they're 0s. this masks guest cannot set non-zero value to this MSR. And inkvm_vcpu_reset(),vcpu->arch.ia32_xssis initialized to 0, inkvm_load_guest_xsave_state()MSR_IA32_XSSis set toia32_xss, therefore the MSR is kept to 0.Sorry, I think what I said "
msr_ia32_xssbitmap" misled you."
msr_ia32_xssbitmap" is notMSR_IA32_XSS, but the (entry->ecx|entry->edx>> 32) of cpuid.D_1I meant we'd better set
entry->ecxandentry->edxto 0 here.Quoted 1 line
Quoted 1 line
Quoted 9 lines
Quoted 6 lines
Quoted 5 lines
arch/x86/kvm/cpuid.c static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2 *entry, int index) · line 474
Quoted 2 lines
arch/x86/kvm/cpuid.c +2
-0+if (entry->eax & (F (XSAVES ) |F (XSAVEC ))) + entry->ebx =xstate_required_size (supported, true);And setup
msr_ia32_xssbitmap based on thes_supportedwithin this condition when F(XSAVES) is supported.And set
entry->ecx&entry->edxonly when F(XSAVES) is supported.
Yep, make sense, will change it, thanks!
Quoted 7 lines
Quoted 3 lines
IIUC, both XSAVEC and XSAVES use compacted format of the extended region, so if XSAVEC is supported while XSAVES is not, guest still can get correct size, so in existing code the two bits are ORed.
Yeah, but
entry->ecxandentry->edxshould be non-zero only when F(XSAVES) is set.
Yes, when F(XSAVES) is not supported, we just init them to 0s.
Quoted 87 lines
Quoted 87 lines
Quoted 86 lines
Quoted 85 lines
arch/x86/kvm/cpuid.c +27
-32+break ; +default : + supported = (entry->ecx &0x1 ) ? s_supported : u_supported; +if (!(supported & (BIT_ULL (index)))) { + entry->eax =0 ; + entry->ebx =0 ; + entry->ecx =0 ; + entry->edx =0 ; +return false; + } +if (entry->ecx &0x1 ) + entry->ebx =0 ; +break ; + } +return true; +} +static inline int __do_cpuid_func (struct kvm_cpuid_entry2 *entry,u32 function,int *nent,int maxnent) {@@ -440,7 +503,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,unsigned f_lm =0 ;#endif unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported () ?F (RDTSCP ) :0 ; -unsigned f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ;unsigned f_intel_pt = kvm_x86_ops->pt_supported () ?F (INTEL_PT ) :0 ;/* cpuid 1.edx */ @@ -495,10 +557,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,F (ACE2 ) |F (ACE2_EN ) |F (PHE ) |F (PHE_EN ) |F (PMM ) |F (PMM_EN ); -/* cpuid 0xD.1.eax */ -const u32 kvm_cpuid_D_1_eax_x86_features = -F (XSAVEOPT ) |F (XSAVEC ) |F (XGETBV1 ) | f_xsaves; -/* all calls to cpuid_count() should be made on the same cpu */ get_cpu ();@@ -639,38 +697,21 @@ static inline int __do_cpuid_func(struct kvm_cpuid_entry2 *entry, u32 function,break ; }case 0xd : { -int idx, i; -u64 supported =kvm_supported_xcr0 (); +int i, idx; - entry->eax &= supported; - entry->ebx =xstate_required_size (supported, false); - entry->ecx = entry->ebx; - entry->edx &= supported >>32 ; -if (!supported) +if (!do_cpuid_0xd_mask (&entry[0 ],0 ))break ; - -for (idx =1 , i =1 ; idx <64 ; ++idx) { -u64 mask = ((u64 )1 << idx); +for (i =1 , idx =1 ; idx <64 ; ++idx) {if (*nent >= maxnent) goto out; -do_host_cpuid (&entry[i], function, idx); -if (idx ==1 ) { - entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; -cpuid_mask (&entry[i].eax,CPUID_D_1_EAX ); - entry[i].ebx =0 ; -if (entry[i].eax & (F (XSAVES )|F (XSAVEC ))) - entry[i].ebx = -xstate_required_size (supported, - true); - }else { -if (entry[i].eax ==0 || !(supported & mask)) -continue ; -if (WARN_ON_ONCE (entry[i].ecx &1 )) -continue ; - } - entry[i].ecx =0 ; - entry[i].edx =0 ; + +if (entry[i].eax ==0 && entry[i].ebx ==0 && + entry[i].ecx ==0 && entry[i].edx ==0 ) +continue ; + +if (!do_cpuid_0xd_mask (&entry[i], idx)) +continue ; ++*nent; ++i; }