Message ID | 20240415172542.1830566-1-mizhang@google.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [kvm-unit-tests] x86: msr: Remove the loop for testing reserved bits in MSR_IA32_FLUSH_CMD | expand |
On Mon, Apr 15, 2024, Mingwei Zhang wrote: > Avoid testing reserved bits in MSR_IA32_FLUSH_CMD. Since KVM passes through > the MSR at runtime, testing reserved bits directly touches the HW and > should generate #GP. However, some older CPU models like skylake with > certain FMS do not generate #GP. > > Ideally, it could be fixed by enumerating all such CPU models. The value > added is would be low. So just remove the testing loop and allow the test > pass. > > Signed-off-by: Mingwei Zhang <mizhang@google.com> > --- > x86/msr.c | 2 -- > 1 file changed, 2 deletions(-) > > diff --git a/x86/msr.c b/x86/msr.c > index 3a041fab..76c80d29 100644 > --- a/x86/msr.c > +++ b/x86/msr.c > @@ -302,8 +302,6 @@ static void test_cmd_msrs(void) > test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", 0); > test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", L1D_FLUSH); > } > - for (i = 1; i < 64; i++) > - test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", BIT_ULL(i)); Rather than remove this entirely, what forcing emulation? E.g. (compile tested only, and haven't verified all macros) --- lib/x86/desc.h | 30 ++++++++++++++++++++++++------ lib/x86/processor.h | 18 ++++++++++++++---- x86/msr.c | 16 +++++++++++++++- 3 files changed, 53 insertions(+), 11 deletions(-) diff --git a/lib/x86/desc.h b/lib/x86/desc.h index 7778a0f8..92c45a48 100644 --- a/lib/x86/desc.h +++ b/lib/x86/desc.h @@ -272,9 +272,9 @@ extern gdt_entry_t *get_tss_descr(void); extern unsigned long get_gdt_entry_base(gdt_entry_t *entry); extern unsigned long get_gdt_entry_limit(gdt_entry_t *entry); -#define asm_safe(insn, inputs...) \ +#define __asm_safe(fep, insn, inputs...) \ ({ \ - asm volatile(ASM_TRY("1f") \ + asm volatile(__ASM_TRY(fep, "1f") \ insn "\n\t" \ "1:\n\t" \ : \ @@ -283,9 +283,15 @@ extern unsigned long get_gdt_entry_limit(gdt_entry_t *entry); exception_vector(); \ }) -#define asm_safe_out1(insn, output, inputs...) \ +#define asm_safe(insn, inputs...) \ + __asm_safe("", insn, inputs) + +#define asm_fep_safe(insn, output, inputs...) \ + __asm_safe_out1(KVM_FEP, insn, output, inputs) + +#define __asm_safe_out1(fep, insn, output, inputs...) \ ({ \ - asm volatile(ASM_TRY("1f") \ + asm volatile(__ASM_TRY(fep, "1f") \ insn "\n\t" \ "1:\n\t" \ : output \ @@ -294,9 +300,15 @@ extern unsigned long get_gdt_entry_limit(gdt_entry_t *entry); exception_vector(); \ }) -#define asm_safe_out2(insn, output1, output2, inputs...) \ +#define asm_safe_out1(insn, output, inputs...) \ + __asm_safe_out1("", insn, output, inputs) + +#define asm_fep_safe_out1(insn, output, inputs...) \ + __asm_safe_out1(KVM_FEP, insn, output, inputs) + +#define __asm_safe_out2(fep, insn, output1, output2, inputs...) \ ({ \ - asm volatile(ASM_TRY("1f") \ + asm volatile(__ASM_TRY(fep, "1f") \ insn "\n\t" \ "1:\n\t" \ : output1, output2 \ @@ -305,6 +317,12 @@ extern unsigned long get_gdt_entry_limit(gdt_entry_t *entry); exception_vector(); \ }) +#define asm_safe_out2(fep, insn, output1, output2, inputs...) \ + __asm_safe_out2("", insn, output1, output2, inputs) + +#define asm_fep_safe_out2(insn, output1, output2, inputs...) \ + __asm_safe_out2(KVM_FEP, insn, output1, output2, inputs) + #define __asm_safe_report(want, insn, inputs...) \ do { \ int vector = asm_safe(insn, inputs); \ diff --git a/lib/x86/processor.h b/lib/x86/processor.h index 44f4fd1e..d20496c0 100644 --- a/lib/x86/processor.h +++ b/lib/x86/processor.h @@ -430,12 +430,12 @@ static inline void wrmsr(u32 index, u64 val) asm volatile ("wrmsr" : : "a"(a), "d"(d), "c"(index) : "memory"); } -#define rdreg64_safe(insn, index, val) \ +#define __rdreg64_safe(fep, insn, index, val) \ ({ \ uint32_t a, d; \ int vector; \ \ - vector = asm_safe_out2(insn, "=a"(a), "=d"(d), "c"(index)); \ + vector = __asm_safe_out2(fep, insn, "=a"(a), "=d"(d), "c"(index));\ \ if (vector) \ *(val) = 0; \ @@ -444,13 +444,18 @@ static inline void wrmsr(u32 index, u64 val) vector; \ }) -#define wrreg64_safe(insn, index, val) \ +#define rdreg64_safe(insn, index, val) \ + __rdreg64_safe("", insn, index, val) + +#define __wrreg64_safe(fep, insn, index, val) \ ({ \ uint32_t eax = (val), edx = (val) >> 32; \ \ - asm_safe(insn, "a" (eax), "d" (edx), "c" (index)); \ + __asm_safe(fep, insn, "a" (eax), "d" (edx), "c" (index)); \ }) +#define wrreg64_safe(insn, index, val) \ + __wrreg64_safe("", insn, index, val) static inline int rdmsr_safe(u32 index, uint64_t *val) { @@ -462,6 +467,11 @@ static inline int wrmsr_safe(u32 index, u64 val) return wrreg64_safe("wrmsr", index, val); } +static inline int wrmsr_fep_safe(u32 index, u64 val) +{ + return __wrreg64_safe(KVM_FEP, "wrmsr", index, val); +} + static inline int rdpmc_safe(u32 index, uint64_t *val) { return rdreg64_safe("rdpmc", index, val); diff --git a/x86/msr.c b/x86/msr.c index 3a041fab..2830530b 100644 --- a/x86/msr.c +++ b/x86/msr.c @@ -112,6 +112,16 @@ static void test_rdmsr_fault(u32 msr, const char *name) "Expected #GP on RDSMR(%s), got vector %d", name, vector); } +static void test_wrmsr_fep_fault(u32 msr, const char *name, + unsigned long long val) +{ + unsigned char vector = wrmsr_fep_safe(msr, val); + + report(vector == GP_VECTOR, + "Expected #GP on emulated WRSMR(%s, 0x%llx), got vector %d", + name, val, vector); +} + static void test_msr(struct msr_info *msr, bool is_64bit_host) { if (is_64bit_host || !msr->is_64bit_only) { @@ -302,8 +312,12 @@ static void test_cmd_msrs(void) test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", 0); test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", L1D_FLUSH); } + + if (!is_fep_available()) + return; + for (i = 1; i < 64; i++) - test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", BIT_ULL(i)); + test_wrmsr_fep_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", BIT_ULL(i)); } int main(int ac, char **av) base-commit: 38135e08a580b9f3696f9b4ae5ca228dc71a1a56 --
On Tue, Apr 16, 2024, Sean Christopherson wrote: > On Mon, Apr 15, 2024, Mingwei Zhang wrote: > > Avoid testing reserved bits in MSR_IA32_FLUSH_CMD. Since KVM passes through > > the MSR at runtime, testing reserved bits directly touches the HW and > > should generate #GP. However, some older CPU models like skylake with > > certain FMS do not generate #GP. > > > > Ideally, it could be fixed by enumerating all such CPU models. The value > > added is would be low. So just remove the testing loop and allow the test > > pass. > > > > Signed-off-by: Mingwei Zhang <mizhang@google.com> > > --- > > x86/msr.c | 2 -- > > 1 file changed, 2 deletions(-) > > > > diff --git a/x86/msr.c b/x86/msr.c > > index 3a041fab..76c80d29 100644 > > --- a/x86/msr.c > > +++ b/x86/msr.c > > @@ -302,8 +302,6 @@ static void test_cmd_msrs(void) > > test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", 0); > > test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", L1D_FLUSH); > > } > > - for (i = 1; i < 64; i++) > > - test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", BIT_ULL(i)); > > Rather than remove this entirely, what forcing emulation? E.g. (compile tested > only, and haven't verified all macros) For sure. Nice, thank you! I think this FEP thing also benefits other tests like PMU. Thanks. -Mingwei > > --- > lib/x86/desc.h | 30 ++++++++++++++++++++++++------ > lib/x86/processor.h | 18 ++++++++++++++---- > x86/msr.c | 16 +++++++++++++++- > 3 files changed, 53 insertions(+), 11 deletions(-) > > diff --git a/lib/x86/desc.h b/lib/x86/desc.h > index 7778a0f8..92c45a48 100644 > --- a/lib/x86/desc.h > +++ b/lib/x86/desc.h > @@ -272,9 +272,9 @@ extern gdt_entry_t *get_tss_descr(void); > extern unsigned long get_gdt_entry_base(gdt_entry_t *entry); > extern unsigned long get_gdt_entry_limit(gdt_entry_t *entry); > > -#define asm_safe(insn, inputs...) \ > +#define __asm_safe(fep, insn, inputs...) \ > ({ \ > - asm volatile(ASM_TRY("1f") \ > + asm volatile(__ASM_TRY(fep, "1f") \ > insn "\n\t" \ > "1:\n\t" \ > : \ > @@ -283,9 +283,15 @@ extern unsigned long get_gdt_entry_limit(gdt_entry_t *entry); > exception_vector(); \ > }) > > -#define asm_safe_out1(insn, output, inputs...) \ > +#define asm_safe(insn, inputs...) \ > + __asm_safe("", insn, inputs) > + > +#define asm_fep_safe(insn, output, inputs...) \ > + __asm_safe_out1(KVM_FEP, insn, output, inputs) > + > +#define __asm_safe_out1(fep, insn, output, inputs...) \ > ({ \ > - asm volatile(ASM_TRY("1f") \ > + asm volatile(__ASM_TRY(fep, "1f") \ > insn "\n\t" \ > "1:\n\t" \ > : output \ > @@ -294,9 +300,15 @@ extern unsigned long get_gdt_entry_limit(gdt_entry_t *entry); > exception_vector(); \ > }) > > -#define asm_safe_out2(insn, output1, output2, inputs...) \ > +#define asm_safe_out1(insn, output, inputs...) \ > + __asm_safe_out1("", insn, output, inputs) > + > +#define asm_fep_safe_out1(insn, output, inputs...) \ > + __asm_safe_out1(KVM_FEP, insn, output, inputs) > + > +#define __asm_safe_out2(fep, insn, output1, output2, inputs...) \ > ({ \ > - asm volatile(ASM_TRY("1f") \ > + asm volatile(__ASM_TRY(fep, "1f") \ > insn "\n\t" \ > "1:\n\t" \ > : output1, output2 \ > @@ -305,6 +317,12 @@ extern unsigned long get_gdt_entry_limit(gdt_entry_t *entry); > exception_vector(); \ > }) > > +#define asm_safe_out2(fep, insn, output1, output2, inputs...) \ > + __asm_safe_out2("", insn, output1, output2, inputs) > + > +#define asm_fep_safe_out2(insn, output1, output2, inputs...) \ > + __asm_safe_out2(KVM_FEP, insn, output1, output2, inputs) > + > #define __asm_safe_report(want, insn, inputs...) \ > do { \ > int vector = asm_safe(insn, inputs); \ > diff --git a/lib/x86/processor.h b/lib/x86/processor.h > index 44f4fd1e..d20496c0 100644 > --- a/lib/x86/processor.h > +++ b/lib/x86/processor.h > @@ -430,12 +430,12 @@ static inline void wrmsr(u32 index, u64 val) > asm volatile ("wrmsr" : : "a"(a), "d"(d), "c"(index) : "memory"); > } > > -#define rdreg64_safe(insn, index, val) \ > +#define __rdreg64_safe(fep, insn, index, val) \ > ({ \ > uint32_t a, d; \ > int vector; \ > \ > - vector = asm_safe_out2(insn, "=a"(a), "=d"(d), "c"(index)); \ > + vector = __asm_safe_out2(fep, insn, "=a"(a), "=d"(d), "c"(index));\ > \ > if (vector) \ > *(val) = 0; \ > @@ -444,13 +444,18 @@ static inline void wrmsr(u32 index, u64 val) > vector; \ > }) > > -#define wrreg64_safe(insn, index, val) \ > +#define rdreg64_safe(insn, index, val) \ > + __rdreg64_safe("", insn, index, val) > + > +#define __wrreg64_safe(fep, insn, index, val) \ > ({ \ > uint32_t eax = (val), edx = (val) >> 32; \ > \ > - asm_safe(insn, "a" (eax), "d" (edx), "c" (index)); \ > + __asm_safe(fep, insn, "a" (eax), "d" (edx), "c" (index)); \ > }) > > +#define wrreg64_safe(insn, index, val) \ > + __wrreg64_safe("", insn, index, val) > > static inline int rdmsr_safe(u32 index, uint64_t *val) > { > @@ -462,6 +467,11 @@ static inline int wrmsr_safe(u32 index, u64 val) > return wrreg64_safe("wrmsr", index, val); > } > > +static inline int wrmsr_fep_safe(u32 index, u64 val) > +{ > + return __wrreg64_safe(KVM_FEP, "wrmsr", index, val); > +} > + > static inline int rdpmc_safe(u32 index, uint64_t *val) > { > return rdreg64_safe("rdpmc", index, val); > diff --git a/x86/msr.c b/x86/msr.c > index 3a041fab..2830530b 100644 > --- a/x86/msr.c > +++ b/x86/msr.c > @@ -112,6 +112,16 @@ static void test_rdmsr_fault(u32 msr, const char *name) > "Expected #GP on RDSMR(%s), got vector %d", name, vector); > } > > +static void test_wrmsr_fep_fault(u32 msr, const char *name, > + unsigned long long val) > +{ > + unsigned char vector = wrmsr_fep_safe(msr, val); > + > + report(vector == GP_VECTOR, > + "Expected #GP on emulated WRSMR(%s, 0x%llx), got vector %d", > + name, val, vector); > +} > + > static void test_msr(struct msr_info *msr, bool is_64bit_host) > { > if (is_64bit_host || !msr->is_64bit_only) { > @@ -302,8 +312,12 @@ static void test_cmd_msrs(void) > test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", 0); > test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", L1D_FLUSH); > } > + > + if (!is_fep_available()) > + return; > + > for (i = 1; i < 64; i++) > - test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", BIT_ULL(i)); > + test_wrmsr_fep_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", BIT_ULL(i)); > } > > int main(int ac, char **av) > > base-commit: 38135e08a580b9f3696f9b4ae5ca228dc71a1a56 > -- >
diff --git a/x86/msr.c b/x86/msr.c index 3a041fab..76c80d29 100644 --- a/x86/msr.c +++ b/x86/msr.c @@ -302,8 +302,6 @@ static void test_cmd_msrs(void) test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", 0); test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", L1D_FLUSH); } - for (i = 1; i < 64; i++) - test_wrmsr_fault(MSR_IA32_FLUSH_CMD, "FLUSH_CMD", BIT_ULL(i)); } int main(int ac, char **av)
Avoid testing reserved bits in MSR_IA32_FLUSH_CMD. Since KVM passes through the MSR at runtime, testing reserved bits directly touches the HW and should generate #GP. However, some older CPU models like skylake with certain FMS do not generate #GP. Ideally, it could be fixed by enumerating all such CPU models. The value added is would be low. So just remove the testing loop and allow the test pass. Signed-off-by: Mingwei Zhang <mizhang@google.com> --- x86/msr.c | 2 -- 1 file changed, 2 deletions(-) base-commit: 7b0147ea57dc29ba844f5b60393a0639e55e88af