x86: xsaveopt kernel enabling patches - v2
cover x86: xsaveopt kernel enabling patches - v2
AVX Programming Reference (June, 2010 version) located at
http://software.intel.com/en-us/avx/ refers to the optimized extended state save instruction "xsaveopt".
xsaveopt is similar to xsave and incorporates optimizations specifically targeted for speeding up context switch save, like:
a. if a state is in an INIT state, the instruction may clear the corresponding
bit in the xheader.bv with out writing the actual state in to the memory.
b. if the state hasn't been modified since the last xrstor, the instruction
may not write out the state to the save area.
Following patches enable the use of xsaveopt for context switch save/restore of the extended state.
changes in v2:
Split the cpu feature detection bits in to multiple patches along with the misc changes in the init_scattered_cpuid_features() -- Requested by hpa.
PATCH 1/5 x86: track the offset, size of features state in the xsave layout
Subleaves of the cpuid vector 0xd provides the offset and size of different feature state that are managed by the xsave/xrstor. Track this for the upcoming usage during signal handling.
--- arch/x86/kernel/xsave.c | 29 +++++++++++++++++++++++++++++ 1 file changed, 29 insertions(+)
arch/x86/kernel/xsave.c +29 -0
@@ -21,6 +21,8 @@ struct _fpx_sw_bytes fx_sw_reserved;struct _fpx_sw_bytes fx_sw_reserved_ia32;#endif +static unsigned int *xstate_offsets, *xstate_sizes, xstate_features; +/* * Check for the presence of extended state information in the * user fpstate pointer in the sigcontext.
@@ -294,12 +296,39 @@ void __cpuinit xsave_init(void) }/* +* Record the offsets and sizes of different state managed by the xsave +* memory layout. +*/ +static void setup_xstate_features (void ) +{ +int eax, ebx, ecx, edx, leaf =0x2 ; + + xstate_features =fls64 (pcntxt_mask); + xstate_offsets =alloc_bootmem (xstate_features *sizeof (int )); + xstate_sizes =alloc_bootmem (xstate_features *sizeof (int )); + +do { +cpuid_count (0xd , leaf, &eax, &ebx, &ecx, &edx); + +if (eax ==0 ) +break ; + + xstate_offsets[leaf] = ebx; + xstate_sizes[leaf] = eax; + + leaf++; + }while (1 ); +} + +/* * setup the xstate image representing the init state */ static void __init setup_xstate_init (void ) { init_xstate_buf =alloc_bootmem (xstate_size); init_xstate_buf->i387.mxcsr =MXCSR_DEFAULT ; + +setup_xstate_features (); }/*
[tip:x86/xsave] x86, xsave: Track the offset, size of state in the xsave layout
Commit-ID: a1488f8bf4d72ad724700f6e982469a1240e4264 Gitweb: http://git.kernel.org/tip/a1488f8bf4d72ad724700f6e982469a1240e4264 Author: Suresh Siddha <[email protected]> AuthorDate: Mon, 19 Jul 2010 16:05:48 -0700 Committer: H. Peter Anvin <[email protected]> CommitDate: Mon, 19 Jul 2010 17:51:17 -0700
x86, xsave: Track the offset, size of state in the xsave layout
Subleaves of the cpuid vector 0xd provides the offset and size of different feature state that are managed by the xsave/xrstor. Track this for the upcoming usage during signal handling.
LKML-Reference: <[email protected]>
--- arch/x86/kernel/xsave.c | 29 +++++++++++++++++++++++++++++ 1 files changed, 29 insertions(+), 0 deletions(-)
arch/x86/kernel/xsave.c +29 -0
@@ -21,6 +21,8 @@ struct _fpx_sw_bytes fx_sw_reserved;struct _fpx_sw_bytes fx_sw_reserved_ia32;#endif +static unsigned int *xstate_offsets, *xstate_sizes, xstate_features; +/* * Check for the presence of extended state information in the * user fpstate pointer in the sigcontext.
@@ -302,12 +304,39 @@ void __cpuinit xsave_init(void) }/* +* Record the offsets and sizes of different state managed by the xsave +* memory layout. +*/ +static void setup_xstate_features (void ) +{ +int eax, ebx, ecx, edx, leaf =0x2 ; + + xstate_features =fls64 (pcntxt_mask); + xstate_offsets =alloc_bootmem (xstate_features *sizeof (int )); + xstate_sizes =alloc_bootmem (xstate_features *sizeof (int )); + +do { +cpuid_count (0xd , leaf, &eax, &ebx, &ecx, &edx); + +if (eax ==0 ) +break ; + + xstate_offsets[leaf] = ebx; + xstate_sizes[leaf] = eax; + + leaf++; + }while (1 ); +} + +/* * setup the xstate image representing the init state */ static void __init setup_xstate_init (void ) { init_xstate_buf =alloc_bootmem (xstate_size); init_xstate_buf->i387.mxcsr =MXCSR_DEFAULT ; + +setup_xstate_features (); }/*
PATCH 2/5 x86: sync xsave memory layout with its header for user handling
In reply to Suresh SiddhaWith xsaveopt, if a processor implementation discern that a processor state component is in its initialized state it may modify the corresponding bit in the xsave_hdr.xstate_bv as '0', with out modifying the corresponding memory layout. Hence wHile presenting the xstate information to the user, we always ensure that the memory layout of a feature will be in the init state if the corresponding header bit is zero. This ensures the consistency and avoids the condition of the user seeing some some stale state in the memory layout during signal handling, debugging etc.
--- arch/x86/include/asm/i387.h | 14 ++++++ arch/x86/include/asm/xsave.h | 10 ++++ arch/x86/kernel/i387.c | 11 +++++ arch/x86/kernel/xsave.c | 89 ++++++++++++++++++++++++++++++++++++++++++- 4 files changed, 123 insertions(+), 1 deletion(-)
arch/x86/kernel/xsave.c +88 -1
@@ -24,6 +24,76 @@ struct _fpx_sw_bytes fx_sw_reserved_ia32static unsigned int *xstate_offsets, *xstate_sizes, xstate_features;/* +* If a processor implementation discern that a processor state component is +* in its initialized state it may modify the corresponding bit in the +* xsave_hdr.xstate_bv as '0', with out modifying the corresponding memory +* layout in the case of xsaveopt. While presenting the xstate information to +* the user, we always ensure that the memory layout of a feature will be in +* the init state if the corresponding header bit is zero. This is to ensure +* that the user doesn't see some stale state in the memory layout during +* signal handling, debugging etc. +*/ +void __sanitize_i387_state (struct task_struct *tsk) +{ +u64 xstate_bv; +int feature_bit =0x2 ; +struct i387_fxsave_struct *fx = &tsk->thread.fpu.state->fxsave; + +if (!fx) +return ; + +BUG_ON (task_thread_info (tsk)->status &TS_USEDFPU ); + + xstate_bv = tsk->thread.fpu.state->xsave.xsave_hdr.xstate_bv; + +/* +* None of the feature bits are in init state. So nothing else +* to do for us, as the memory layout is upto date. +*/ +if ((xstate_bv & pcntxt_mask) == pcntxt_mask) +return ; + +/* +* FP is in init state +*/ +if (!(xstate_bv &XSTATE_FP )) { + fx->cwd =0x37f ; + fx->swd =0 ; + fx->twd =0 ; + fx->fop =0 ; + fx->rip =0 ; + fx->rdp =0 ; +memset (&fx->st_space[0 ],0 ,128 ); + } + +/* +* SSE is in init state +*/ +if (!(xstate_bv &XSTATE_SSE )) +memset (&fx->xmm_space[0 ],0 ,256 ); + + xstate_bv = (pcntxt_mask & ~xstate_bv) >>2 ; + +/* +* Update all the other memory layouts for which the corresponding +* header bit is in the init state. +*/ +while (xstate_bv) { +if (xstate_bv &0x1 ) { +int offset = xstate_offsets[feature_bit]; +int size = xstate_sizes[feature_bit]; + +memcpy (((void *) fx) + offset, + ((void *) init_xstate_buf) + offset, + size); + } + + xstate_bv >>=1 ; + feature_bit++; + } +} + +/* * Check for the presence of extended state information in the * user fpstate pointer in the sigcontext. */
@@ -104,6 +174,7 @@ int save_i387_xstate(void __user *buf)task_thread_info (tsk)->status &= ~TS_USEDFPU ;stts (); }else { +sanitize_i387_state (tsk);if (__copy_to_user (buf, &tsk->thread.fpu.state->fxsave, xstate_size))return -1 ;
@@ -325,10 +396,26 @@ static void setup_xstate_features(void)*/ static void __init setup_xstate_init (void ) { +setup_xstate_features (); + +/* +* Setup init_xstate_buf to represent the init state of +* all the features managed by the xsave +*/ init_xstate_buf =alloc_bootmem (xstate_size); init_xstate_buf->i387.mxcsr =MXCSR_DEFAULT ; -setup_xstate_features (); +clts (); +/* +* Init all the features state with header_bv being 0x0 +*/ +xrstor_state (init_xstate_buf,-1 ); +/* +* Dump the init state again. This is to identify the init state +* of any feature which is not represented by all zero's. +*/ +xsave_state (init_xstate_buf,-1 ); +stts (); }/*
arch/x86/kernel/i387.c +11 -0
@@ -190,6 +190,8 @@ int xfpregs_get(struct task_struct *targif (ret)return ret; +sanitize_i387_state (target); +return user_regset_copyout (&pos, &count, &kbuf, &ubuf, &target->thread.fpu.state->fxsave,0 ,-1 ); }
@@ -207,6 +209,8 @@ int xfpregs_set(struct task_struct *targif (ret)return ret; +sanitize_i387_state (target); + ret =user_regset_copyin (&pos, &count, &kbuf, &ubuf, &target->thread.fpu.state->fxsave,0 ,-1 );
@@ -446,6 +450,8 @@ int fpregs_get(struct task_struct *targe -1); } +sanitize_i387_state (target); +if (kbuf && pos ==0 && count ==sizeof (env)) {convert_from_fxsr (kbuf, target);return 0 ;
@@ -467,6 +473,8 @@ int fpregs_set(struct task_struct *targeif (ret)return ret; +sanitize_i387_state (target); +if (!HAVE_HWFP )return fpregs_soft_set (target, regset, pos, count, kbuf, ubuf);
@@ -533,6 +541,9 @@ static int save_i387_xsave(void __user *struct _fpstate_ia32 __user *fx = buf;int err =0 ; + +sanitize_i387_state (tsk); +/* * For legacy compatible, we always set FP/SSE bits in the bit * vector while saving the state to the user context.
arch/x86/include/asm/i387.h +14 -0
@@ -58,11 +58,25 @@ extern int restore_i387_xstate_ia32(void#define X87_FSW_ES (1 << 7)/* Exception Summary */ +static __always_inline __pure bool use_xsaveopt(void) +{ +return 0 ; +} + static __always_inline __pure bool use_xsave(void) {return static_cpu_has (X86_FEATURE_XSAVE ); } +extern void __sanitize_i387_state (struct task_struct *); + +static inline void sanitize_i387_state (struct task_struct *tsk) +{ +if (!use_xsaveopt ()) +return ; +__sanitize_i387_state (tsk); +} +#ifdef CONFIG_X86_64 /* Ignore delayed exceptions from user space */
arch/x86/include/asm/xsave.h +10 -0
@@ -121,6 +121,16 @@ static inline void xrstor_state(struct x : "memory"); } +static inline void xsave_state (struct xsave_struct *fx,u64 mask) +{ +u32 lmask = mask; +u32 hmask = mask >>32 ; + + asmvolatile (".byte " REX_PREFIX "0x0f,0xae,0x27\n\t" + : :"D" (fx),"m" (*fx),"a" (lmask),"d" (hmask) + :"memory" ); +} +static inline void fpu_xsave (struct fpu *fpu) {/* This, however, we can work around by forcing the compiler to select
[tip:x86/xsave] x86, xsave: Sync xsave memory layout with its header for user handling
Commit-ID: 29104e101d710dd152f807978884643a52eca8b7 Gitweb: http://git.kernel.org/tip/29104e101d710dd152f807978884643a52eca8b7 Author: Suresh Siddha <[email protected]> AuthorDate: Mon, 19 Jul 2010 16:05:49 -0700 Committer: H. Peter Anvin <[email protected]> CommitDate: Mon, 19 Jul 2010 17:51:30 -0700
x86, xsave: Sync xsave memory layout with its header for user handling
With xsaveopt, if a processor implementation discern that a processor state component is in its initialized state it may modify the corresponding bit in the xsave_hdr.xstate_bv as '0', with out modifying the corresponding memory layout. Hence wHile presenting the xstate information to the user, we always ensure that the memory layout of a feature will be in the init state if the corresponding header bit is zero. This ensures the consistency and avoids the condition of the user seeing some some stale state in the memory layout during signal handling, debugging etc.
LKML-Reference: <[email protected]>
--- arch/x86/include/asm/i387.h | 14 +++++++ arch/x86/include/asm/xsave.h | 10 +++++ arch/x86/kernel/i387.c | 11 +++++ arch/x86/kernel/xsave.c | 89 +++++++++++++++++++++++++++++++++++++++++- 4 files changed, 123 insertions(+), 1 deletions(-)
arch/x86/include/asm/i387.h +14 -0
@@ -58,11 +58,25 @@ extern int restore_i387_xstate_ia32(void __user *buf);#define X87_FSW_ES (1 << 7)/* Exception Summary */ +static __always_inline __pure bool use_xsaveopt(void) +{ +return 0 ; +} + static __always_inline __pure bool use_xsave(void) {return static_cpu_has (X86_FEATURE_XSAVE ); } +extern void __sanitize_i387_state (struct task_struct *); + +static inline void sanitize_i387_state (struct task_struct *tsk) +{ +if (!use_xsaveopt ()) +return ; +__sanitize_i387_state (tsk); +} +#ifdef CONFIG_X86_64 /* Ignore delayed exceptions from user space */
arch/x86/include/asm/xsave.h +10 -0
@@ -111,6 +111,16 @@ static inline void xrstor_state(struct xsave_struct *fx, u64 mask) : "memory"); } +static inline void xsave_state (struct xsave_struct *fx,u64 mask) +{ +u32 lmask = mask; +u32 hmask = mask >>32 ; + + asmvolatile (".byte " REX_PREFIX "0x0f,0xae,0x27\n\t" + : :"D" (fx),"m" (*fx),"a" (lmask),"d" (hmask) + :"memory" ); +} +static inline void fpu_xsave (struct fpu *fpu) {/* This, however, we can work around by forcing the compiler to select
arch/x86/kernel/i387.c +11 -0
@@ -190,6 +190,8 @@ int xfpregs_get(struct task_struct *target, const struct user_regset *regset,if (ret)return ret; +sanitize_i387_state (target); +return user_regset_copyout (&pos, &count, &kbuf, &ubuf, &target->thread.fpu.state->fxsave,0 ,-1 ); }
@@ -207,6 +209,8 @@ int xfpregs_set(struct task_struct *target, const struct user_regset *regset,if (ret)return ret; +sanitize_i387_state (target); + ret =user_regset_copyin (&pos, &count, &kbuf, &ubuf, &target->thread.fpu.state->fxsave,0 ,-1 );
@@ -446,6 +450,8 @@ int fpregs_get(struct task_struct *target, const struct user_regset *regset, -1); } +sanitize_i387_state (target); +if (kbuf && pos ==0 && count ==sizeof (env)) {convert_from_fxsr (kbuf, target);return 0 ;
@@ -467,6 +473,8 @@ int fpregs_set(struct task_struct *target, const struct user_regset *regset,if (ret)return ret; +sanitize_i387_state (target); +if (!HAVE_HWFP )return fpregs_soft_set (target, regset, pos, count, kbuf, ubuf);
@@ -533,6 +541,9 @@ static int save_i387_xsave(void __user *buf)struct _fpstate_ia32 __user *fx = buf;int err =0 ; + +sanitize_i387_state (tsk); +/* * For legacy compatible, we always set FP/SSE bits in the bit * vector while saving the state to the user context.
arch/x86/kernel/xsave.c +88 -1
110 lines not on this page · raw message
PATCH 3/5 x86: extend init_scattered_cpuid_features() to consider cpuid subleaves
In reply to Suresh Siddhacpuid features (like xsaveopt) are enumerated using cpuid subleaves.
Extend init_scattered_cpuid_features() to take subleaf into account.
--- arch/x86/kernel/cpu/addon_cpuid_features.c | 25 +++++++++++++------------ 1 file changed, 13 insertions(+), 12 deletions(-)
arch/x86/kernel/cpu/addon_cpuid_features.c +13 -12
@@ -14,6 +14,7 @@ struct cpuid_bit {u8 reg;u8 bit;u32 level; +u32 sub_leaf; };enum cpuid_regs {
@@ -30,16 +31,16 @@ void __cpuinit init_scattered_cpuid_featconst struct cpuid_bit *cb;static const struct cpuid_bit __cpuinitconst cpuid_bits[] = { - {X86_FEATURE_IDA ,CR_EAX ,1 ,0x00000006 }, - {X86_FEATURE_ARAT ,CR_EAX ,2 ,0x00000006 }, - {X86_FEATURE_APERFMPERF ,CR_ECX ,0 ,0x00000006 }, - {X86_FEATURE_EPB ,CR_ECX ,3 ,0x00000006 }, - {X86_FEATURE_CPB ,CR_EDX ,9 ,0x80000007 }, - {X86_FEATURE_NPT ,CR_EDX ,0 ,0x8000000a }, - {X86_FEATURE_LBRV ,CR_EDX ,1 ,0x8000000a }, - {X86_FEATURE_SVML ,CR_EDX ,2 ,0x8000000a }, - {X86_FEATURE_NRIPS ,CR_EDX ,3 ,0x8000000a }, - {0 ,0 ,0 ,0 } + {X86_FEATURE_IDA ,CR_EAX ,1 ,0x00000006 ,0 }, + {X86_FEATURE_ARAT ,CR_EAX ,2 ,0x00000006 ,0 }, + {X86_FEATURE_APERFMPERF ,CR_ECX ,0 ,0x00000006 ,0 }, + {X86_FEATURE_EPB ,CR_ECX ,3 ,0x00000006 ,0 }, + {X86_FEATURE_CPB ,CR_EDX ,9 ,0x80000007 ,0 }, + {X86_FEATURE_NPT ,CR_EDX ,0 ,0x8000000a ,0 }, + {X86_FEATURE_LBRV ,CR_EDX ,1 ,0x8000000a ,0 }, + {X86_FEATURE_SVML ,CR_EDX ,2 ,0x8000000a ,0 }, + {X86_FEATURE_NRIPS ,CR_EDX ,3 ,0x8000000a ,0 }, + {0 ,0 ,0 ,0 ,0 } };for (cb = cpuid_bits; cb->feature; cb++) {
@@ -50,8 +51,8 @@ void __cpuinit init_scattered_cpuid_feat max_level > (cb->level | 0xffff))continue ; -cpuid (cb->level, ®s[CR_EAX ], ®s[CR_EBX ], - ®s[CR_ECX ], ®s[CR_EDX ]); +cpuid_count (cb->level, cb->sub_leaf, ®s[CR_EAX ], + ®s[CR_EBX ], ®s[CR_ECX ], ®s[CR_EDX ]);if (regs[cb->reg] & (1 << cb->bit))set_cpu_cap (c, cb->feature);
[tip:x86/cpu] x86, cpu: Make init_scattered_cpuid_features() consider cpuid subleaves
Commit-ID: edb18f8ab02843453306601c4aa697f9691129cd Gitweb: http://git.kernel.org/tip/edb18f8ab02843453306601c4aa697f9691129cd Author: Suresh Siddha <[email protected]> AuthorDate: Mon, 19 Jul 2010 16:05:50 -0700 Committer: H. Peter Anvin <[email protected]> CommitDate: Mon, 19 Jul 2010 16:47:59 -0700
x86, cpu: Make init_scattered_cpuid_features() consider cpuid subleaves
Some cpuid features (like xsaveopt) are enumerated using cpuid subleaves.
Extend init_scattered_cpuid_features() to take subleaf into account.
LKML-Reference: <[email protected]>
--- arch/x86/kernel/cpu/addon_cpuid_features.c | 25 +++++++++++++------------ 1 files changed, 13 insertions(+), 12 deletions(-)
arch/x86/kernel/cpu/addon_cpuid_features.c +13 -12
@@ -14,6 +14,7 @@ struct cpuid_bit {u8 reg;u8 bit;u32 level; +u32 sub_leaf; };enum cpuid_regs {
@@ -30,16 +31,16 @@ void __cpuinit init_scattered_cpuid_features(struct cpuinfo_x86 *c)const struct cpuid_bit *cb;static const struct cpuid_bit __cpuinitconst cpuid_bits[] = { - {X86_FEATURE_IDA ,CR_EAX ,1 ,0x00000006 }, - {X86_FEATURE_ARAT ,CR_EAX ,2 ,0x00000006 }, - {X86_FEATURE_APERFMPERF ,CR_ECX ,0 ,0x00000006 }, - {X86_FEATURE_EPB ,CR_ECX ,3 ,0x00000006 }, - {X86_FEATURE_CPB ,CR_EDX ,9 ,0x80000007 }, - {X86_FEATURE_NPT ,CR_EDX ,0 ,0x8000000a }, - {X86_FEATURE_LBRV ,CR_EDX ,1 ,0x8000000a }, - {X86_FEATURE_SVML ,CR_EDX ,2 ,0x8000000a }, - {X86_FEATURE_NRIPS ,CR_EDX ,3 ,0x8000000a }, - {0 ,0 ,0 ,0 } + {X86_FEATURE_IDA ,CR_EAX ,1 ,0x00000006 ,0 }, + {X86_FEATURE_ARAT ,CR_EAX ,2 ,0x00000006 ,0 }, + {X86_FEATURE_APERFMPERF ,CR_ECX ,0 ,0x00000006 ,0 }, + {X86_FEATURE_EPB ,CR_ECX ,3 ,0x00000006 ,0 }, + {X86_FEATURE_CPB ,CR_EDX ,9 ,0x80000007 ,0 }, + {X86_FEATURE_NPT ,CR_EDX ,0 ,0x8000000a ,0 }, + {X86_FEATURE_LBRV ,CR_EDX ,1 ,0x8000000a ,0 }, + {X86_FEATURE_SVML ,CR_EDX ,2 ,0x8000000a ,0 }, + {X86_FEATURE_NRIPS ,CR_EDX ,3 ,0x8000000a ,0 }, + {0 ,0 ,0 ,0 ,0 } };for (cb = cpuid_bits; cb->feature; cb++) {
@@ -50,8 +51,8 @@ void __cpuinit init_scattered_cpuid_features(struct cpuinfo_x86 *c) max_level > (cb->level | 0xffff))continue ; -cpuid (cb->level, ®s[CR_EAX ], ®s[CR_EBX ], - ®s[CR_ECX ], ®s[CR_EDX ]); +cpuid_count (cb->level, cb->sub_leaf, ®s[CR_EAX ], + ®s[CR_EBX ], ®s[CR_ECX ], ®s[CR_EDX ]);if (regs[cb->reg] & (1 << cb->bit))set_cpu_cap (c, cb->feature);
PATCH 4/5 x86: Add xsaveopt cpufeature
In reply to Suresh SiddhaAdd cpu feature bit support for the XSAVEOPT instruction.
--- arch/x86/include/asm/cpufeature.h | 1 + 1 file changed, 1 insertion(+)
arch/x86/include/asm/cpufeature.h +1 -0
@@ -165,6 +165,7 @@#define X86_FEATURE_ARAT (7*32+ 1)/* Always Running APIC Timer */ #define X86_FEATURE_CPB (7*32+ 2)/* AMD Core Performance Boost */ #define X86_FEATURE_EPB (7*32+ 3)/* IA32_ENERGY_PERF_BIAS support */ +#define X86_FEATURE_XSAVEOPT (7*32+4)/* "xsaveopt" Optimized Xsave */ /* Virtualization flags: Linux defined, word 8 */ #define X86_FEATURE_TPR_SHADOW (8*32+ 0)/* Intel TPR Shadow */
[tip:x86/cpu] x86, cpu: Add xsaveopt cpufeature
Commit-ID: 40e1d7a4ffee5cb17f5c36f4c3c4a011ab103ebe Gitweb: http://git.kernel.org/tip/40e1d7a4ffee5cb17f5c36f4c3c4a011ab103ebe Author: Suresh Siddha <[email protected]> AuthorDate: Mon, 19 Jul 2010 16:05:51 -0700 Committer: H. Peter Anvin <[email protected]> CommitDate: Mon, 19 Jul 2010 16:48:06 -0700
x86, cpu: Add xsaveopt cpufeature
Add cpu feature bit support for the XSAVEOPT instruction.
LKML-Reference: <[email protected]>
--- arch/x86/include/asm/cpufeature.h | 1 + 1 files changed, 1 insertions(+), 0 deletions(-)
arch/x86/include/asm/cpufeature.h +1 -0
@@ -165,6 +165,7 @@#define X86_FEATURE_ARAT (7*32+ 1)/* Always Running APIC Timer */ #define X86_FEATURE_CPB (7*32+ 2)/* AMD Core Performance Boost */ #define X86_FEATURE_EPB (7*32+ 3)/* IA32_ENERGY_PERF_BIAS support */ +#define X86_FEATURE_XSAVEOPT (7*32+4)/* "xsaveopt" Optimized Xsave */ /* Virtualization flags: Linux defined, word 8 */ #define X86_FEATURE_TPR_SHADOW (8*32+ 0)/* Intel TPR Shadow */
PATCH 5/5 x86: use xsaveopt in context-switch path when supported
In reply to Suresh Siddhaxsaveopt is a more optimized form of xsave specifically designed for the context switch usage. xsaveopt doesn't save the state that's not modified from the prior xrstor. And if a specific feature state gets modified to the init state, then xsaveopt just updates the header bit in the xsave memory layout without updating the corresponding memory layout.
--- arch/x86/include/asm/i387.h | 2 +- arch/x86/include/asm/xsave.h | 9 ++++++--- arch/x86/kernel/cpu/addon_cpuid_features.c | 1 + arch/x86/kernel/cpu/common.c | 8 ++++++++ 4 files changed, 16 insertions(+), 4 deletions(-)
arch/x86/include/asm/xsave.h +6 -3
@@ -135,8 +135,11 @@ static inline void fpu_xsave(struct fpu {/* This, however, we can work around by forcing the compiler to select an addressing mode that doesn't require extended registers. */ - __asm__ __volatile__(".byte " REX_PREFIX "0x0f,0xae,0x27" - : :"D" (&(fpu->state->xsave)), -"a" (-1 ),"d" (-1 ) :"memory" ); +alternative_input ( +".byte " REX_PREFIX "0x0f,0xae,0x27" , +".byte " REX_PREFIX "0x0f,0xae,0x37" , + X86_FEATURE_XSAVEOPT, + [fx] "D" (&fpu->state->xsave), "a" (-1), "d" (-1) : +"memory" ); }#endif
arch/x86/kernel/cpu/common.c +8 -0
@@ -140,10 +140,18 @@ EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);static int __init x86_xsave_setup (char *s) {setup_clear_cpu_cap (X86_FEATURE_XSAVE ); +setup_clear_cpu_cap (X86_FEATURE_XSAVEOPT );return 1 ; }__setup ("noxsave" , x86_xsave_setup); +static int __init x86_xsaveopt_setup (char *s) +{ +setup_clear_cpu_cap (X86_FEATURE_XSAVEOPT ); +return 1 ; +} +__setup ("noxsaveopt" , x86_xsaveopt_setup); +#ifdef CONFIG_X86_32 static int cachesize_override__cpuinitdata =-1 ;static int disable_x86_serial_nr__cpuinitdata =1 ;
arch/x86/include/asm/i387.h +1 -1
@@ -60,7 +60,7 @@ extern int restore_i387_xstate_ia32(void static __always_inline __pure bool use_xsaveopt(void) { -return 0 ; +return static_cpu_has (X86_FEATURE_XSAVEOPT ); } static __always_inline __pure bool use_xsave(void)
arch/x86/kernel/cpu/addon_cpuid_features.c +1 -0
@@ -35,6 +35,7 @@ void __cpuinit init_scattered_cpuid_feat { X86_FEATURE_ARAT, CR_EAX, 2, 0x00000006, 0 }, { X86_FEATURE_APERFMPERF, CR_ECX, 0, 0x00000006, 0 }, { X86_FEATURE_EPB, CR_ECX, 3, 0x00000006, 0 }, + { X86_FEATURE_XSAVEOPT, CR_EAX, 0, 0x0000000d, 1 }, { X86_FEATURE_CPB, CR_EDX, 9, 0x80000007, 0 }, { X86_FEATURE_NPT, CR_EDX, 0, 0x8000000a, 0 }, { X86_FEATURE_LBRV, CR_EDX, 1, 0x8000000a, 0 },
[tip:x86/cpu] x86, cpu: Enumerate xsaveopt
Commit-ID: 5734f62b6601d88fd8ec720cb56b93fd3a030557 Gitweb: http://git.kernel.org/tip/5734f62b6601d88fd8ec720cb56b93fd3a030557 Author: Suresh Siddha <[email protected]> AuthorDate: Mon, 19 Jul 2010 16:05:52 -0700 Committer: H. Peter Anvin <[email protected]> CommitDate: Mon, 19 Jul 2010 16:50:40 -0700
x86, cpu: Enumerate xsaveopt
Enumerate the xsaveopt feature.
LKML-Reference: <[email protected]>
--- arch/x86/kernel/cpu/addon_cpuid_features.c | 1 + 1 files changed, 1 insertions(+), 0 deletions(-)
arch/x86/kernel/cpu/addon_cpuid_features.c +1 -0
@@ -35,6 +35,7 @@ void __cpuinit init_scattered_cpuid_features(struct cpuinfo_x86 *c) { X86_FEATURE_ARAT, CR_EAX, 2, 0x00000006, 0 }, { X86_FEATURE_APERFMPERF, CR_ECX, 0, 0x00000006, 0 }, { X86_FEATURE_EPB, CR_ECX, 3, 0x00000006, 0 }, + { X86_FEATURE_XSAVEOPT, CR_EAX, 0, 0x0000000d, 1 }, { X86_FEATURE_CPB, CR_EDX, 9, 0x80000007, 0 }, { X86_FEATURE_NPT, CR_EDX, 0, 0x8000000a, 0 }, { X86_FEATURE_LBRV, CR_EDX, 1, 0x8000000a, 0 },
[tip:x86/xsave] x86, xsave: Use xsaveopt in context-switch path when supported
In reply to Suresh SiddhaCommit-ID: 6bad06b768920e278c7cedfdda56a0b4c6a35ee9 Gitweb: http://git.kernel.org/tip/6bad06b768920e278c7cedfdda56a0b4c6a35ee9 Author: Suresh Siddha <[email protected]> AuthorDate: Mon, 19 Jul 2010 16:05:52 -0700 Committer: H. Peter Anvin <[email protected]> CommitDate: Mon, 19 Jul 2010 17:52:24 -0700
x86, xsave: Use xsaveopt in context-switch path when supported
xsaveopt is a more optimized form of xsave specifically designed for the context switch usage. xsaveopt doesn't save the state that's not modified from the prior xrstor. And if a specific feature state gets modified to the init state, then xsaveopt just updates the header bit in the xsave memory layout without updating the corresponding memory layout.
LKML-Reference: <[email protected]>
--- arch/x86/include/asm/i387.h | 2 +- arch/x86/include/asm/xsave.h | 9 ++++++--- arch/x86/kernel/cpu/common.c | 8 ++++++++ 3 files changed, 15 insertions(+), 4 deletions(-)
arch/x86/include/asm/i387.h +1 -1
@@ -60,7 +60,7 @@ extern int restore_i387_xstate_ia32(void __user *buf); static __always_inline __pure bool use_xsaveopt(void) { -return 0 ; +return static_cpu_has (X86_FEATURE_XSAVEOPT ); } static __always_inline __pure bool use_xsave(void)
arch/x86/include/asm/xsave.h +6 -3
@@ -125,8 +125,11 @@ static inline void fpu_xsave(struct fpu *fpu) {/* This, however, we can work around by forcing the compiler to select an addressing mode that doesn't require extended registers. */ - __asm__ __volatile__(".byte " REX_PREFIX "0x0f,0xae,0x27" - : :"D" (&(fpu->state->xsave)), -"a" (-1 ),"d" (-1 ) :"memory" ); +alternative_input ( +".byte " REX_PREFIX "0x0f,0xae,0x27" , +".byte " REX_PREFIX "0x0f,0xae,0x37" , + X86_FEATURE_XSAVEOPT, + [fx] "D" (&fpu->state->xsave), "a" (-1), "d" (-1) : +"memory" ); }#endif
arch/x86/kernel/cpu/common.c +8 -0
@@ -140,10 +140,18 @@ EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);static int __init x86_xsave_setup (char *s) {setup_clear_cpu_cap (X86_FEATURE_XSAVE ); +setup_clear_cpu_cap (X86_FEATURE_XSAVEOPT );return 1 ; }__setup ("noxsave" , x86_xsave_setup); +static int __init x86_xsaveopt_setup (char *s) +{ +setup_clear_cpu_cap (X86_FEATURE_XSAVEOPT ); +return 1 ; +} +__setup ("noxsaveopt" , x86_xsaveopt_setup); +#ifdef CONFIG_X86_32 static int cachesize_override__cpuinitdata =-1 ;static int disable_x86_serial_nr__cpuinitdata =1 ;