x86/cpufeatures: Re-arrange a few features and enumerate AVX512 BFLOAT16 intructions
v2 cover x86/cpufeatures: Re-arrange a few features and enumerate AVX512 BFLOAT16 intructions
To enumerate AVX512 BFLOAT16 feature CPUID.7.1:EAX[5] and other future features in CPUID.7.1:EAX, Boris suggests to create a new pure feature bits word.
Boris further suggests to re-define word 11 as a scattered features word and move the four X86_FEATURE_CQM_* features in existing word 11 and word 12 into the new word 11. Then use word 12 to hold features in CPUID.7.1:EAX including AVX512 BFLOAT16 instructions.
Change Log:
v2:
- Remove patch 0001 and keep x86_cache_max_rmid and x86_cache_occ_scale in cpuinfo_x86 per Thomas Gleixner's comment.
v1:
Address all comments from Borislav Petkov:
- Re-define feature word 11 as a scattered features word and move the four X86_FEATURE_CQM_* features into the word
- Re-define feature word 12 to hold CPUID.7.1:EAX including BFLOAT16
- Name a dummy leaf 12 in cpuid_leafs in patch 0002 to avoid bisect error
- Simplify code to get number of rmid and monitoring scale in rdt_get_mon_l3_config()
Fenghua Yu (2):
x86/cpufeatures: Combine word 11 and 12 into new scattered features
word 11
x86/cpufeatures: Enumerate new AVX512 BFLOAT16 instructions
arch/x86/include/asm/cpufeature.h | 4 +-- arch/x86/include/asm/cpufeatures.h | 18 ++++++---- arch/x86/kernel/cpu/common.c | 56 ++++++++++++++++-------------- arch/x86/kernel/cpu/cpuid-deps.c | 4 +++ arch/x86/kernel/cpu/scattered.c | 4 +++ arch/x86/kvm/cpuid.h | 2 -- 6 files changed, 51 insertions(+), 37 deletions(-)
Signature · 1 line
2.19.1
PATCH v2 1/2 x86/cpufeatures: Combine word 11 and 12 into new scattered features word 11
It's a waste for the four X86_FEATURE_CQM_* features to occupy two pure feature bits words. To better utilize feature words, re-define word 11 to host scattered features and move the four X86_FEATURE_CQM_* features into Linux defined word 11. More scattered features can be added in word 11 in the future.
Leaf 11 in cpuid_leafs is renamed as CPUID_LNX_4 to reflect it's a Linux defined leaf.
Although word 12 doesn't have any feature now, leaf 12 in cpuid_leafs still needs to be kept because cpuid_leafs must have NCAPINTS leafs. Rename leaf 12 as CPUID_DUMMY which will be replaced by a meaningful name in the next patch when CPUID.7.1:EAX occupies world 12.
Maximum number of RMID and cache occupancy scale are retrieved from CPUID.0xf.1 after scattered CQM features are enumerated. Carve out the code into a separate function.
KVM doesn't support resctrl now. So it's safe to move the X86_FEATURE_CQM_* features to scattered features word 11 for KVM.
--- arch/x86/include/asm/cpufeature.h | 4 +-- arch/x86/include/asm/cpufeatures.h | 17 ++++++---- arch/x86/kernel/cpu/common.c | 53 +++++++++++++++--------------- arch/x86/kernel/cpu/cpuid-deps.c | 3 ++ arch/x86/kernel/cpu/scattered.c | 4 +++ arch/x86/kvm/cpuid.h | 2 -- 6 files changed, 45 insertions(+), 38 deletions(-)
arch/x86/include/asm/cpufeature.h +2 -2
@@ -22,8 +22,8 @@ enum cpuid_leafsCPUID_LNX_3 ,CPUID_7_0_EBX ,CPUID_D_1_EAX , -CPUID_F_0_EDX , -CPUID_F_1_EDX , +CPUID_LNX_4 , +CPUID_DUMMY ,CPUID_8000_0008_EBX ,CPUID_6_EAX , CPUID_8000_000A_EDX,
arch/x86/include/asm/cpufeatures.h +10 -7
@@ -269,13 +269,16 @@#define X86_FEATURE_XGETBV1 (10*32+ 2)/* XGETBV with ECX = 1 instruction */ #define X86_FEATURE_XSAVES (10*32+ 3)/* XSAVES/XRSTORS instructions */ -/* Intel-defined CPU QoS Sub-leaf, CPUID level 0x0000000F:0 (EDX), word 11 */ -#define X86_FEATURE_CQM_LLC (11*32+ 1)/* LLC QoS if 1 */ - -/* Intel-defined CPU QoS Sub-leaf, CPUID level 0x0000000F:1 (EDX), word 12 */ -#define X86_FEATURE_CQM_OCCUP_LLC (12*32+ 0)/* LLC occupancy monitoring */ -#define X86_FEATURE_CQM_MBM_TOTAL (12*32+ 1)/* LLC Total MBM monitoring */ -#define X86_FEATURE_CQM_MBM_LOCAL (12*32+ 2)/* LLC Local MBM monitoring */ +/* +* Extended auxiliary flags: Linux defined - For features scattered in various +* CPUID levels like 0xf, word 11. +* +* Reuse free bits when adding new feature flags! +*/ +#define X86_FEATURE_CQM_LLC (11*32+ 0)/* LLC QoS if 1 */ +#define X86_FEATURE_CQM_OCCUP_LLC (11*32+ 1)/* LLC occupancy monitoring */ +#define X86_FEATURE_CQM_MBM_TOTAL (11*32+ 2)/* LLC Total MBM monitoring */ +#define X86_FEATURE_CQM_MBM_LOCAL (11*32+ 3)/* LLC Local MBM monitoring */ /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */ #define X86_FEATURE_CLZERO (13*32+ 0)/* CLZERO instruction */
arch/x86/kernel/cpu/common.c +26 -27
71 lines not on this page · raw message
arch/x86/kernel/cpu/cpuid-deps.c +3 -0
@@ -59,6 +59,9 @@ static const struct cpuid_dep cpuid_deps[] = { { X86_FEATURE_AVX512_4VNNIW, X86_FEATURE_AVX512F }, { X86_FEATURE_AVX512_4FMAPS, X86_FEATURE_AVX512F }, { X86_FEATURE_AVX512_VPOPCNTDQ, X86_FEATURE_AVX512F }, + { X86_FEATURE_CQM_OCCUP_LLC, X86_FEATURE_CQM_LLC }, + { X86_FEATURE_CQM_MBM_TOTAL, X86_FEATURE_CQM_LLC }, + { X86_FEATURE_CQM_MBM_LOCAL, X86_FEATURE_CQM_LLC }, {} };
arch/x86/kernel/cpu/scattered.c +4 -0
@@ -26,6 +26,10 @@ struct cpuid_bit {static const struct cpuid_bit cpuid_bits[] = { {X86_FEATURE_APERFMPERF ,CPUID_ECX ,0 ,0x00000006 ,0 }, {X86_FEATURE_EPB ,CPUID_ECX ,3 ,0x00000006 ,0 }, + {X86_FEATURE_CQM_LLC ,CPUID_EDX ,1 ,0x0000000f ,0 }, + {X86_FEATURE_CQM_OCCUP_LLC ,CPUID_EDX ,0 ,0x0000000f ,1 }, + {X86_FEATURE_CQM_MBM_TOTAL ,CPUID_EDX ,1 ,0x0000000f ,1 }, + {X86_FEATURE_CQM_MBM_LOCAL ,CPUID_EDX ,2 ,0x0000000f ,1 }, {X86_FEATURE_CAT_L3 ,CPUID_EBX ,1 ,0x00000010 ,0 }, {X86_FEATURE_CAT_L2 ,CPUID_EBX ,2 ,0x00000010 ,0 }, {X86_FEATURE_CDP_L3 ,CPUID_ECX ,2 ,0x00000010 ,1 },
arch/x86/kvm/cpuid.h +0 -2
@@ -47,8 +47,6 @@ static const struct cpuid_reg reverse_cpuid[] = { [CPUID_8000_0001_ECX] = {0x80000001, 0, CPUID_ECX}, [CPUID_7_0_EBX] = { 7, 0, CPUID_EBX}, [CPUID_D_1_EAX] = { 0xd, 1, CPUID_EAX}, - [CPUID_F_0_EDX] = { 0xf, 0, CPUID_EDX}, - [CPUID_F_1_EDX] = { 0xf, 1, CPUID_EDX}, [CPUID_8000_0008_EBX] = {0x80000008, 0, CPUID_EBX}, [CPUID_6_EAX] = { 6, 0, CPUID_EAX}, [CPUID_8000_000A_EDX] = {0x8000000a, 0, CPUID_EDX},
Signature · 1 line
2.19.1
Borislav Petkov
maintainerRe: [PATCH v2 1/2] x86/cpufeatures: Combine word 11 and 12 into new scattered features word 11
Quoted 24 lines · Fenghua Yu
(quoted diff) +0
-20@@ -832,33 +857,6 @@ void get_cpu_cap(struct cpuinfo_x86 *c) c->x86_capability[CPUID_D_1_EAX] = eax; } - /* Additional Intel-defined flags: level 0x0000000F */ - if (c->cpuid_level >= 0x0000000F) { - - /* QoS sub-leaf, EAX=0Fh, ECX=0 */ - cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_F_0_EDX] = edx; - - if (cpu_has(c, X86_FEATURE_CQM_LLC)) { - /* will be overridden if occupancy monitoring exists */ - c->x86_cache_max_rmid = ebx; - - /* QoS sub-leaf, EAX=0Fh, ECX=1 */ - cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_F_1_EDX] = edx; - - if ((cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) || - ((cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL)) || - (cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)))) { - c->x86_cache_max_rmid = ecx; - c->x86_cache_occ_scale = ebx;
Review void get_cpu_cap(struct cpuinfo_x86 *c) · line 862
Quoted 10 lines
(quoted diff) +0
-7- } - } else { - c->x86_cache_max_rmid = -1; - c->x86_cache_occ_scale = -1; - } - } - /* AMD-defined flags: level 0x80000001 */ eax = cpuid_eax(0x80000000); c->extended_cpuid_level = eax;
What I meant with having a separate patch doing the carve out is to have a single patch doing *only* code movement - no changes, no nothing. So that it is clear what happens. Intermixing code movement and changes is a bad idea and hard to review.
IOW, I did this:
---
From cef4f58a3da0465bbff33b2d669cc600b775f3ba Mon Sep 17 00:00:00 2001
From: Borislav Petkov <[email protected]>
Date: Wed, 19 Jun 2019 17:24:34 +0200
Subject: [PATCH] x86/cpufeatures: Carve out CQM features retrieval
... into a separate function for better readability. Split out from a patch from Fenghua Yu <[email protected]> to keep the mechanical, sole code movement separate for easy review.
No functional changes.
--- arch/x86/kernel/cpu/common.c | 60 ++++++++++++++++++++---------------- 1 file changed, 33 insertions(+), 27 deletions(-)
arch/x86/kernel/cpu/common.c +33 -27
@@ -801,6 +801,38 @@ static void init_speculation_control(struct cpuinfo_x86 *c) } } +static void init_cqm (struct cpuinfo_x86 *c) +{ +u32 eax, ebx, ecx, edx; + +/* Additional Intel-defined flags: level 0x0000000F */ +if (c->cpuid_level >=0x0000000F ) { + +/* QoS sub-leaf, EAX=0Fh, ECX=0 */ +cpuid_count (0x0000000F ,0 , &eax, &ebx, &ecx, &edx); + c->x86_capability[CPUID_F_0_EDX ] = edx; + +if (cpu_has (c,X86_FEATURE_CQM_LLC )) { +/* will be overridden if occupancy monitoring exists */ + c->x86_cache_max_rmid = ebx; + +/* QoS sub-leaf, EAX=0Fh, ECX=1 */ +cpuid_count (0x0000000F ,1 , &eax, &ebx, &ecx, &edx); + c->x86_capability[CPUID_F_1_EDX ] = edx; + +if ((cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC )) || + ((cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL )) || + (cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )))) { + c->x86_cache_max_rmid = ecx; + c->x86_cache_occ_scale = ebx; + } + }else { + c->x86_cache_max_rmid =-1 ; + c->x86_cache_occ_scale =-1 ; + } + } +} +void get_cpu_cap (struct cpuinfo_x86 *c) {u32 eax, ebx, ecx, edx;
@@ -832,33 +864,6 @@ void get_cpu_cap(struct cpuinfo_x86 *c) c->x86_capability[CPUID_D_1_EAX ] = eax; } -/* Additional Intel-defined flags: level 0x0000000F */ -if (c->cpuid_level >=0x0000000F ) { - -/* QoS sub-leaf, EAX=0Fh, ECX=0 */ -cpuid_count (0x0000000F ,0 , &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_F_0_EDX ] = edx; - -if (cpu_has (c,X86_FEATURE_CQM_LLC )) { -/* will be overridden if occupancy monitoring exists */ - c->x86_cache_max_rmid = ebx; - -/* QoS sub-leaf, EAX=0Fh, ECX=1 */ -cpuid_count (0x0000000F ,1 , &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_F_1_EDX ] = edx; - -if ((cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC )) || - ((cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL )) || - (cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )))) { - c->x86_cache_max_rmid = ecx; - c->x86_cache_occ_scale = ebx; - } - }else { - c->x86_cache_max_rmid =-1 ; - c->x86_cache_occ_scale =-1 ; - } - } -/* AMD-defined flags: level 0x80000001 */ eax =cpuid_eax (0x80000000 ); c->extended_cpuid_level = eax;
@@ -889,6 +894,7 @@ void get_cpu_cap(struct cpuinfo_x86 *c)init_scattered_cpuid_features (c);init_speculation_control (c); +init_cqm (c);/* * Clear/Set all flags overridden by options, after probe.
Signature · 80 lines
2.21.0 --- This way you have *pure* code movement only. And then your second patch turns into this, which shows *exactly* what has been changed in init_cqm(). Please have a look and send me only the now third patch with corrected commit message. Thx. --- From e33527b8cde8bef84cdc90651d1a1c7a9a5234d7 Mon Sep 17 00:00:00 2001 From: Fenghua Yu <[email protected]> Date: Wed, 19 Jun 2019 18:51:09 +0200 Subject: [PATCH] x86/cpufeatures: Combine word 11 and 12 into a new scattered features word MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit It's a waste for the four X86_FEATURE_CQM_* feature bits to occupy two whole feature bits words. To better utilize feature words, re-define word 11 to host scattered features and move the four X86_FEATURE_CQM_* features into Linux defined word 11. More scattered features can be added in word 11 in the future. Rename leaf 11 in cpuid_leafs to CPUID_LNX_4 to reflect it's a Linux-defined leaf. Rename leaf 12 as CPUID_DUMMY which will be replaced by a meaningful name in the next patch when CPUID.7.1:EAX occupies world 12. Maximum number of RMID and cache occupancy scale are retrieved from CPUID.0xf.1 after scattered CQM features are enumerated. Carve out the code into a separate function. KVM doesn't support resctrl now. So it's safe to move the X86_FEATURE_CQM_* features to scattered features word 11 for KVM. Signed-off-by: Fenghua Yu <[email protected]> Signed-off-by: Borislav Petkov <[email protected]> Cc: Aaron Lewis <[email protected]> Cc: Andy Lutomirski <[email protected]> Cc: Babu Moger <[email protected]> Cc: "Chang S. Bae" <[email protected]> Cc: "Sean J Christopherson" <[email protected]> Cc: Frederic Weisbecker <[email protected]> Cc: "H. Peter Anvin" <[email protected]> Cc: Ingo Molnar <[email protected]> Cc: Jann Horn <[email protected]> Cc: Juergen Gross <[email protected]> Cc: Konrad Rzeszutek Wilk <[email protected]> Cc: kvm ML <[email protected]> Cc: Masahiro Yamada <[email protected]> Cc: Masami Hiramatsu <[email protected]> Cc: Nadav Amit <[email protected]> Cc: Paolo Bonzini <[email protected]> Cc: Pavel Tatashin <[email protected]> Cc: Peter Feiner <[email protected]> Cc: "Peter Zijlstra (Intel)" <[email protected]> Cc: "Radim Krčmář" <[email protected]> Cc: "Rafael J. Wysocki" <[email protected]> Cc: Ravi V Shankar <[email protected]> Cc: Sherry Hurwitz <[email protected]> Cc: Thomas Gleixner <[email protected]> Cc: Thomas Lendacky <[email protected]> Cc: x86 <[email protected]> Link: https://lkml.kernel.org/r/[email protected] --- arch/x86/include/asm/cpufeature.h | 4 ++-- arch/x86/include/asm/cpufeatures.h | 17 +++++++------ arch/x86/kernel/cpu/common.c | 38 ++++++++++++------------------ arch/x86/kernel/cpu/cpuid-deps.c | 3 +++ arch/x86/kernel/cpu/scattered.c | 4 ++++ arch/x86/kvm/cpuid.h | 2 -- 6 files changed, 34 insertions(+), 34 deletions(-)
arch/x86/include/asm/cpufeature.h +2 -2
@@ -22,8 +22,8 @@ enum cpuid_leafsCPUID_LNX_3 ,CPUID_7_0_EBX ,CPUID_D_1_EAX , -CPUID_F_0_EDX , -CPUID_F_1_EDX , +CPUID_LNX_4 , +CPUID_DUMMY ,CPUID_8000_0008_EBX ,CPUID_6_EAX , CPUID_8000_000A_EDX,
arch/x86/include/asm/cpufeatures.h +10 -7
@@ -271,13 +271,16 @@#define X86_FEATURE_XGETBV1 (10*32+ 2)/* XGETBV with ECX = 1 instruction */ #define X86_FEATURE_XSAVES (10*32+ 3)/* XSAVES/XRSTORS instructions */ -/* Intel-defined CPU QoS Sub-leaf, CPUID level 0x0000000F:0 (EDX), word 11 */ -#define X86_FEATURE_CQM_LLC (11*32+ 1)/* LLC QoS if 1 */ - -/* Intel-defined CPU QoS Sub-leaf, CPUID level 0x0000000F:1 (EDX), word 12 */ -#define X86_FEATURE_CQM_OCCUP_LLC (12*32+ 0)/* LLC occupancy monitoring */ -#define X86_FEATURE_CQM_MBM_TOTAL (12*32+ 1)/* LLC Total MBM monitoring */ -#define X86_FEATURE_CQM_MBM_LOCAL (12*32+ 2)/* LLC Local MBM monitoring */ +/* +* Extended auxiliary flags: Linux defined - for features scattered in various +* CPUID levels like 0xf, etc. +* +* Reuse free bits when adding new feature flags! +*/ +#define X86_FEATURE_CQM_LLC (11*32+ 0)/* LLC QoS if 1 */ +#define X86_FEATURE_CQM_OCCUP_LLC (11*32+ 1)/* LLC occupancy monitoring */ +#define X86_FEATURE_CQM_MBM_TOTAL (11*32+ 2)/* LLC Total MBM monitoring */ +#define X86_FEATURE_CQM_MBM_LOCAL (11*32+ 3)/* LLC Local MBM monitoring */ /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */ #define X86_FEATURE_CLZERO (13*32+ 0)/* CLZERO instruction */
arch/x86/kernel/cpu/common.c +15 -23
@@ -803,33 +803,25 @@ static void init_speculation_control(struct cpuinfo_x86 *c)static void init_cqm (struct cpuinfo_x86 *c) { -u32 eax, ebx, ecx, edx; - -/* Additional Intel-defined flags: level 0x0000000F */ -if (c->cpuid_level >=0x0000000F ) { +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { + c->x86_cache_max_rmid =-1 ; + c->x86_cache_occ_scale =-1 ; +return ; + } -/* QoS sub-leaf, EAX=0Fh, ECX=0 */ -cpuid_count (0x0000000F ,0 , &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_F_0_EDX ] = edx; +/* will be overridden if occupancy monitoring exists */ + c->x86_cache_max_rmid =cpuid_ebx (0xf ); -if (cpu_has (c,X86_FEATURE_CQM_LLC )) { -/* will be overridden if occupancy monitoring exists */ - c->x86_cache_max_rmid = ebx; +if (cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC ) || +cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL ) || +cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )) { +u32 eax, ebx, ecx, edx; -/* QoS sub-leaf, EAX=0Fh, ECX=1 */ -cpuid_count (0x0000000F ,1 , &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_F_1_EDX ] = edx; +/* QoS sub-leaf, EAX=0Fh, ECX=1 */ +cpuid_count (0xf ,1 , &eax, &ebx, &ecx, &edx); -if ((cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC )) || - ((cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL )) || - (cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )))) { - c->x86_cache_max_rmid = ecx; - c->x86_cache_occ_scale = ebx; - } - }else { - c->x86_cache_max_rmid =-1 ; - c->x86_cache_occ_scale =-1 ; - } + c->x86_cache_max_rmid = ecx; + c->x86_cache_occ_scale = ebx; } }
arch/x86/kernel/cpu/cpuid-deps.c +3 -0
@@ -59,6 +59,9 @@ static const struct cpuid_dep cpuid_deps[] = { { X86_FEATURE_AVX512_4VNNIW, X86_FEATURE_AVX512F }, { X86_FEATURE_AVX512_4FMAPS, X86_FEATURE_AVX512F }, { X86_FEATURE_AVX512_VPOPCNTDQ, X86_FEATURE_AVX512F }, + { X86_FEATURE_CQM_OCCUP_LLC, X86_FEATURE_CQM_LLC }, + { X86_FEATURE_CQM_MBM_TOTAL, X86_FEATURE_CQM_LLC }, + { X86_FEATURE_CQM_MBM_LOCAL, X86_FEATURE_CQM_LLC }, {} };
arch/x86/kernel/cpu/scattered.c +4 -0
@@ -26,6 +26,10 @@ struct cpuid_bit {static const struct cpuid_bit cpuid_bits[] = { {X86_FEATURE_APERFMPERF ,CPUID_ECX ,0 ,0x00000006 ,0 }, {X86_FEATURE_EPB ,CPUID_ECX ,3 ,0x00000006 ,0 }, + {X86_FEATURE_CQM_LLC ,CPUID_EDX ,1 ,0x0000000f ,0 }, + {X86_FEATURE_CQM_OCCUP_LLC ,CPUID_EDX ,0 ,0x0000000f ,1 }, + {X86_FEATURE_CQM_MBM_TOTAL ,CPUID_EDX ,1 ,0x0000000f ,1 }, + {X86_FEATURE_CQM_MBM_LOCAL ,CPUID_EDX ,2 ,0x0000000f ,1 }, {X86_FEATURE_CAT_L3 ,CPUID_EBX ,1 ,0x00000010 ,0 }, {X86_FEATURE_CAT_L2 ,CPUID_EBX ,2 ,0x00000010 ,0 }, {X86_FEATURE_CDP_L3 ,CPUID_ECX ,2 ,0x00000010 ,1 },
arch/x86/kvm/cpuid.h +0 -2
@@ -47,8 +47,6 @@ static const struct cpuid_reg reverse_cpuid[] = { [CPUID_8000_0001_ECX] = {0x80000001, 0, CPUID_ECX}, [CPUID_7_0_EBX] = { 7, 0, CPUID_EBX}, [CPUID_D_1_EAX] = { 0xd, 1, CPUID_EAX}, - [CPUID_F_0_EDX] = { 0xf, 0, CPUID_EDX}, - [CPUID_F_1_EDX] = { 0xf, 1, CPUID_EDX}, [CPUID_8000_0008_EBX] = {0x80000008, 0, CPUID_EBX}, [CPUID_6_EAX] = { 6, 0, CPUID_EAX}, [CPUID_8000_000A_EDX] = {0x8000000a, 0, CPUID_EDX},
Signature · 8 lines
2.21.0
--
Regards/Gruss,
Boris.
Good mailing practices for 400: avoid top-posting and trim the reply.Re: [PATCH v2 1/2] x86/cpufeatures: Combine word 11 and 12 into new scattered features word 11
Quoted 45 lines · Borislav Petkov
arch/x86/kernel/cpu/common.c void get_cpu_cap(struct cpuinfo_x86 *c) · line 834
Quoted 4 lines · Fenghua Yu
arch/x86/kernel/cpu/common.c +0
-2@@ -832,33 +857,6 @@ void get_cpu_cap(struct cpuinfo_x86 *c) c->x86_capability[CPUID_D_1_EAX ] = eax; -/* Additional Intel-defined flags: level 0x0000000F */ -if (c->cpuid_level >=0x0000000F ) {What I meant with having a separate patch doing the carve out is to have a single patch doing *only* code movement - no changes, no nothing. So that it is clear what happens. Intermixing code movement and changes is a bad idea and hard to review.
IOW, I did this:
---
Fromcef4f58a3da0465bbff33b2d669cc600b775f3baMon Sep 17 00:00:00 2001
From: Borislav Petkov <[email protected]>
Date: Wed, 19 Jun 2019 17:24:34 +0200
Subject: [PATCH] x86/cpufeatures: Carve out CQM features retrieval... into a separate function for better readability. Split out from a patch from Fenghua Yu <[email protected]> to keep the mechanical, sole code movement separate for easy review.
No functional changes.
Signed-off-by Borislav Petkov <[email protected]>Cc Fenghua Yu <[email protected]>--- arch/x86/kernel/cpu/common.c | 60 ++++++++++++++++++++---------------- 1 file changed, 33 insertions(+), 27 deletions(-)arch/x86/kernel/cpu/common.c +3
-0@@ -801,6 +801,38 @@ static void init_speculation_control(struct cpuinfo_x86 *c) } } +static void init_cqm (struct cpuinfo_x86 *c) +{ +u32 eax, ebx, ecx, edx;This way you have *pure* code movement only.
Quoted 2 lines
And then your second patch turns into this, which shows *exactly* what has been changed in
init_cqm().
Yes, the added patch makes this patch set more clear and readable.
Quoted 13 lines
Please have a look and send me only the now third patch with corrected commit message.
From
e33527b8cde8bef84cdc90651d1a1c7a9a5234d7Mon Sep 17 00:00:00 2001
From: Fenghua Yu <[email protected]>
Date: Wed, 19 Jun 2019 18:51:09 +0200
Subject: [PATCH] x86/cpufeatures: Combine word 11 and 12 into a new
scattered features word
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Quoted 5 lines
It's a waste for the four X86_FEATURE_CQM_* feature bits to occupy two whole feature bits words. To better utilize feature words, re-define word 11 to host scattered features and move the four X86_FEATURE_CQM_* features into Linux defined word 11. More scattered features can be added in word 11 in the future.
I checked and tested the updated patch set (now three patches). They look much better than v2.
I will send you the now third patch with corrected commit message in the other email thread.
Thanks.
-Fenghua
tip-bot for Fenghua Yu
merge bot[tip:x86/cpu] x86/cpufeatures: Combine word 11 and 12 into a new scattered features word
In reply to Fenghua Yux86/cpufeatures: Combine word 11 and 12 into a new scattered features word
Commit-ID: acec0ce081de0c36459eea91647faf99296445a3 Gitweb: https://git.kernel.org/tip/acec0ce081de0c36459eea91647faf99296445a3 Author: Fenghua Yu <[email protected]> AuthorDate: Wed, 19 Jun 2019 18:51:09 +0200 Committer: Borislav Petkov <[email protected]> CommitDate: Thu, 20 Jun 2019 12:38:44 +0200
x86/cpufeatures: Combine word 11 and 12 into a new scattered features word
It's a waste for the four X86_FEATURE_CQM_* feature bits to occupy two whole feature bits words. To better utilize feature words, re-define word 11 to host scattered features and move the four X86_FEATURE_CQM_* features into Linux defined word 11. More scattered features can be added in word 11 in the future.
Rename leaf 11 in cpuid_leafs to CPUID_LNX_4 to reflect it's a Linux-defined leaf.
Rename leaf 12 as CPUID_DUMMY which will be replaced by a meaningful name in the next patch when CPUID.7.1:EAX occupies world 12.
Maximum number of RMID and cache occupancy scale are retrieved from CPUID.0xf.1 after scattered CQM features are enumerated. Carve out the code into a separate function.
KVM doesn't support resctrl now. So it's safe to move the X86_FEATURE_CQM_* features to scattered features word 11 for KVM.
--- arch/x86/include/asm/cpufeature.h | 4 ++-- arch/x86/include/asm/cpufeatures.h | 17 ++++++++++------- arch/x86/kernel/cpu/common.c | 38 +++++++++++++++----------------------- arch/x86/kernel/cpu/cpuid-deps.c | 3 +++ arch/x86/kernel/cpu/scattered.c | 4 ++++ arch/x86/kvm/cpuid.h | 2 -- 6 files changed, 34 insertions(+), 34 deletions(-)
arch/x86/include/asm/cpufeature.h +2 -2
@@ -22,8 +22,8 @@ enum cpuid_leafsCPUID_LNX_3 ,CPUID_7_0_EBX ,CPUID_D_1_EAX , -CPUID_F_0_EDX , -CPUID_F_1_EDX , +CPUID_LNX_4 , +CPUID_DUMMY ,CPUID_8000_0008_EBX ,CPUID_6_EAX , CPUID_8000_000A_EDX,
arch/x86/include/asm/cpufeatures.h +10 -7
@@ -271,13 +271,16 @@#define X86_FEATURE_XGETBV1 (10*32+ 2)/* XGETBV with ECX = 1 instruction */ #define X86_FEATURE_XSAVES (10*32+ 3)/* XSAVES/XRSTORS instructions */ -/* Intel-defined CPU QoS Sub-leaf, CPUID level 0x0000000F:0 (EDX), word 11 */ -#define X86_FEATURE_CQM_LLC (11*32+ 1)/* LLC QoS if 1 */ - -/* Intel-defined CPU QoS Sub-leaf, CPUID level 0x0000000F:1 (EDX), word 12 */ -#define X86_FEATURE_CQM_OCCUP_LLC (12*32+ 0)/* LLC occupancy monitoring */ -#define X86_FEATURE_CQM_MBM_TOTAL (12*32+ 1)/* LLC Total MBM monitoring */ -#define X86_FEATURE_CQM_MBM_LOCAL (12*32+ 2)/* LLC Local MBM monitoring */ +/* +* Extended auxiliary flags: Linux defined - for features scattered in various +* CPUID levels like 0xf, etc. +* +* Reuse free bits when adding new feature flags! +*/ +#define X86_FEATURE_CQM_LLC (11*32+ 0)/* LLC QoS if 1 */ +#define X86_FEATURE_CQM_OCCUP_LLC (11*32+ 1)/* LLC occupancy monitoring */ +#define X86_FEATURE_CQM_MBM_TOTAL (11*32+ 2)/* LLC Total MBM monitoring */ +#define X86_FEATURE_CQM_MBM_LOCAL (11*32+ 3)/* LLC Local MBM monitoring */ /* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */ #define X86_FEATURE_CLZERO (13*32+ 0)/* CLZERO instruction */
arch/x86/kernel/cpu/common.c +15 -23
@@ -803,33 +803,25 @@ static void init_speculation_control(struct cpuinfo_x86 *c)static void init_cqm (struct cpuinfo_x86 *c) { -u32 eax, ebx, ecx, edx; - -/* Additional Intel-defined flags: level 0x0000000F */ -if (c->cpuid_level >=0x0000000F ) { +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { + c->x86_cache_max_rmid =-1 ; + c->x86_cache_occ_scale =-1 ; +return ; + } -/* QoS sub-leaf, EAX=0Fh, ECX=0 */ -cpuid_count (0x0000000F ,0 , &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_F_0_EDX ] = edx; +/* will be overridden if occupancy monitoring exists */ + c->x86_cache_max_rmid =cpuid_ebx (0xf ); -if (cpu_has (c,X86_FEATURE_CQM_LLC )) { -/* will be overridden if occupancy monitoring exists */ - c->x86_cache_max_rmid = ebx; +if (cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC ) || +cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL ) || +cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )) { +u32 eax, ebx, ecx, edx; -/* QoS sub-leaf, EAX=0Fh, ECX=1 */ -cpuid_count (0x0000000F ,1 , &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_F_1_EDX ] = edx; +/* QoS sub-leaf, EAX=0Fh, ECX=1 */ +cpuid_count (0xf ,1 , &eax, &ebx, &ecx, &edx); -if ((cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC )) || - ((cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL )) || - (cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )))) { - c->x86_cache_max_rmid = ecx; - c->x86_cache_occ_scale = ebx; - } - }else { - c->x86_cache_max_rmid =-1 ; - c->x86_cache_occ_scale =-1 ; - } + c->x86_cache_max_rmid = ecx; + c->x86_cache_occ_scale = ebx; } }
arch/x86/kernel/cpu/cpuid-deps.c +3 -0
@@ -59,6 +59,9 @@ static const struct cpuid_dep cpuid_deps[] = { { X86_FEATURE_AVX512_4VNNIW, X86_FEATURE_AVX512F }, { X86_FEATURE_AVX512_4FMAPS, X86_FEATURE_AVX512F }, { X86_FEATURE_AVX512_VPOPCNTDQ, X86_FEATURE_AVX512F }, + { X86_FEATURE_CQM_OCCUP_LLC, X86_FEATURE_CQM_LLC }, + { X86_FEATURE_CQM_MBM_TOTAL, X86_FEATURE_CQM_LLC }, + { X86_FEATURE_CQM_MBM_LOCAL, X86_FEATURE_CQM_LLC }, {} };
arch/x86/kernel/cpu/scattered.c +4 -0
@@ -26,6 +26,10 @@ struct cpuid_bit {static const struct cpuid_bit cpuid_bits[] = { {X86_FEATURE_APERFMPERF ,CPUID_ECX ,0 ,0x00000006 ,0 }, {X86_FEATURE_EPB ,CPUID_ECX ,3 ,0x00000006 ,0 }, + {X86_FEATURE_CQM_LLC ,CPUID_EDX ,1 ,0x0000000f ,0 }, + {X86_FEATURE_CQM_OCCUP_LLC ,CPUID_EDX ,0 ,0x0000000f ,1 }, + {X86_FEATURE_CQM_MBM_TOTAL ,CPUID_EDX ,1 ,0x0000000f ,1 }, + {X86_FEATURE_CQM_MBM_LOCAL ,CPUID_EDX ,2 ,0x0000000f ,1 }, {X86_FEATURE_CAT_L3 ,CPUID_EBX ,1 ,0x00000010 ,0 }, {X86_FEATURE_CAT_L2 ,CPUID_EBX ,2 ,0x00000010 ,0 }, {X86_FEATURE_CDP_L3 ,CPUID_ECX ,2 ,0x00000010 ,1 },
arch/x86/kvm/cpuid.h +0 -2
@@ -47,8 +47,6 @@ static const struct cpuid_reg reverse_cpuid[] = { [CPUID_8000_0001_ECX] = {0x80000001, 0, CPUID_ECX}, [CPUID_7_0_EBX] = { 7, 0, CPUID_EBX}, [CPUID_D_1_EAX] = { 0xd, 1, CPUID_EAX}, - [CPUID_F_0_EDX] = { 0xf, 0, CPUID_EDX}, - [CPUID_F_1_EDX] = { 0xf, 1, CPUID_EDX}, [CPUID_8000_0008_EBX] = {0x80000008, 0, CPUID_EBX}, [CPUID_6_EAX] = { 6, 0, CPUID_EAX}, [CPUID_8000_000A_EDX] = {0x8000000a, 0, CPUID_EDX},
PATCH v2 2/2 x86/cpufeatures: Enumerate new AVX512 BFLOAT16 instructions
In reply to Fenghua YuAVX512 Vector Neural Network Instructions (VNNI) in Intel Deep Learning Boost support BFLOAT16 format (BF16). BF16 is a short version of FP32 and has several advantages over FP16. BF16 offers more than enough range for deep learning training tasks and doesn't need to handle hardware exception as this is a performance optimization. FP32 accumulation after the multiply is essential to achieve sufficient numerical behavior on an application level.
AVX512 BFLOAT16 instructions can be enumerated by:
CPUID.7.1:EAX[bit 5] AVX512_BF16
Use word 12, which is empty now, to hold features in CPUID.7.1:EAX including AVX512_BF16. Leaf CPUID_DUMMY is renamed as CPUID_7_1_EAX.
Detailed information of the CPUID bit and AVX512 BFLOAT16 instructions can be found in the latest Intel Architecture Instruction Set Extensions and Future Features Programming Reference.
--- arch/x86/include/asm/cpufeature.h | 2 +- arch/x86/include/asm/cpufeatures.h | 3 +++ arch/x86/kernel/cpu/common.c | 3 +++ arch/x86/kernel/cpu/cpuid-deps.c | 1 + 4 files changed, 8 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeature.h +1 -1
@@ -23,7 +23,7 @@ enum cpuid_leafsCPUID_7_0_EBX ,CPUID_D_1_EAX ,CPUID_LNX_4 , -CPUID_DUMMY , +CPUID_7_1_EAX ,CPUID_8000_0008_EBX ,CPUID_6_EAX , CPUID_8000_000A_EDX,
arch/x86/include/asm/cpufeatures.h +3 -0
@@ -280,6 +280,9 @@#define X86_FEATURE_CQM_MBM_TOTAL (11*32+ 2)/* LLC Total MBM monitoring */ #define X86_FEATURE_CQM_MBM_LOCAL (11*32+ 3)/* LLC Local MBM monitoring */ +/* Intel-defined CPU features, CPUID level 0x00000007:1 (EAX), word 12 */ +#define X86_FEATURE_AVX512_BF16 (12*32+ 5)/* AVX512 BFLOAT16 instructions */ +/* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */ #define X86_FEATURE_CLZERO (13*32+ 0)/* CLZERO instruction */ #define X86_FEATURE_IRPERF (13*32+ 1)/* Instructions Retired Count */
arch/x86/kernel/cpu/common.c +3 -0
@@ -848,6 +848,9 @@ void get_cpu_cap(struct cpuinfo_x86 *c) c->x86_capability[CPUID_7_0_EBX ] = ebx; c->x86_capability[CPUID_7_ECX ] = ecx; c->x86_capability[CPUID_7_EDX ] = edx; + +cpuid_count (0x00000007 ,1 , &eax, &ebx, &ecx, &edx); + c->x86_capability[CPUID_7_1_EAX ] = eax; }/* Extended state features: level 0x0000000d */
arch/x86/kernel/cpu/cpuid-deps.c +1 -0
@@ -62,6 +62,7 @@ static const struct cpuid_dep cpuid_deps[] = { { X86_FEATURE_CQM_OCCUP_LLC, X86_FEATURE_CQM_LLC }, { X86_FEATURE_CQM_MBM_TOTAL, X86_FEATURE_CQM_LLC }, { X86_FEATURE_CQM_MBM_LOCAL, X86_FEATURE_CQM_LLC }, + { X86_FEATURE_AVX512_BF16, X86_FEATURE_AVX512VL }, {} };
Signature · 1 line
2.19.1
Borislav Petkov
maintainerRe: [PATCH v2 2/2] x86/cpufeatures: Enumerate new AVX512 BFLOAT16 instructions
Quoted 2 lines · Fenghua Yu
AVX512 Vector Neural Network Instructions (VNNI) in Intel Deep Learning Boost support BFLOAT16 format (BF16).
That sentence is a mouthful and I have no clue what it means. Marketing junk? If so, either rewrite it for mere mortals or kill it.
Quoted 1 line
BF16 is a short version of FP32
FP32?
Please write out.
Quoted 1 line
and has several advantages over FP16.
Ditto.
Quoted 11 lines
BF16 offers more than enough range for
deep learning training tasks and doesn't need to handle hardware exception as this is a performance optimization. FP32 accumulation after the multiply is essential to achieve sufficient numerical behavior on an application level.AVX512 BFLOAT16 instructions can be enumerated by:
CPUID.7.1:EAX[bit 5] AVX512_BF16Use word 12, which is empty now, to hold features in CPUID.7.1:EAX including
AVX512_BF16.
... because that leaf is features only, right?
Quoted 1 line
Leaf
CPUID_DUMMYis renamed asCPUID_7_1_EAX.
That's obvious from the patch, ain't it?
Quoted 3 lines
Detailed information of the CPUID bit and AVX512 BFLOAT16 instructions can be found in the latest Intel Architecture Instruction Set Extensions and Future Features Programming Reference.
Signature · 4 lines
Regards/Gruss,
Boris.
Good mailing practices for 400: avoid top-posting and trim the reply.Re: [PATCH v2 2/2] x86/cpufeatures: Enumerate new AVX512 BFLOAT16 instructions
Quoted 26 lines · Borislav Petkov
Quoted 2 lines · Fenghua Yu
AVX512 Vector Neural Network Instructions (VNNI) in Intel Deep Learning Boost support BFLOAT16 format (BF16).
That sentence is a mouthful and I have no clue what it means. Marketing junk? If so, either rewrite it for mere mortals or kill it.
Quoted 1 line
BF16 is a short version of FP32
FP32?
Please write out.
Quoted 1 line
and has several advantages over FP16.
Ditto.
Quoted 9 lines
BF16 offers more than enough range for
deep learning training tasks and doesn't need to handle hardware exception as this is a performance optimization. FP32 accumulation after the multiply is essential to achieve sufficient numerical behavior on an application level.AVX512 BFLOAT16 instructions can be enumerated by:
CPUID.7.1:EAX[bit 5] AVX512_BF16
Quoted 9 lines
Quoted 2 lines
Use word 12, which is empty now, to hold features in CPUID.7.1:EAX including
AVX512_BF16.... because that leaf is features only, right?
Quoted 1 line
Leaf
CPUID_DUMMYis renamed asCPUID_7_1_EAX.That's obvious from the patch, ain't it?
Hi, Boris,
I corrected the commit message per your comment. Except the commit message, I didn't change anything else.
Now I send the updated patch here. Is the patch right now? Should I send the patch to you in another thread?
Thanks.
-Fenghua
From 6265ae0a0fac5ecf7e67b8bfc4b6791afb88e56c Mon Sep 17 00:00:00 2001
From: Fenghua Yu <[email protected]>
Date: Tue, 11 Jun 2019 23:06:05 +0000
Subject: [PATCH v3 3/3] x86/cpufeatures: Enumerate new AVX512 BFLOAT16
instructions
AVX512 BFLOAT16 instructions support 16-bit BFLOAT16 floating-point format (BF16) for deep learning optimization.
BF16 is a short version of 32-bit single-precision floating-point format (FP32) and has several advantages over 16-bit half-precision floating-point format (FP16). BF16 keeps FP32 accumulation after multiplication without loss of precision, offers more than enough range for deep learning training tasks, and doesn't need to handle hardware exception.
AVX512 BFLOAT16 instructions can be enumerated by:
CPUID.7.1:EAX[bit 5] AVX512_BF16
CPUID.7.1:EAX contains only feature bits. Re-define currently empty word 12 as a pure features word to hold the feature bits including AVV512_BF16.
Detailed information of the CPUID bit and AVX512 BFLOAT16 instructions can be found in the latest Intel Architecture Instruction Set Extensions and Future Features Programming Reference.
--- arch/x86/include/asm/cpufeature.h | 2 +- arch/x86/include/asm/cpufeatures.h | 3 +++ arch/x86/kernel/cpu/common.c | 3 +++ arch/x86/kernel/cpu/cpuid-deps.c | 1 + 4 files changed, 8 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeature.h +1 -1
@@ -23,7 +23,7 @@ enum cpuid_leafsCPUID_7_0_EBX ,CPUID_D_1_EAX ,CPUID_LNX_4 , -CPUID_DUMMY , +CPUID_7_1_EAX ,CPUID_8000_0008_EBX ,CPUID_6_EAX , CPUID_8000_000A_EDX,
arch/x86/include/asm/cpufeatures.h +3 -0
@@ -280,6 +280,9 @@#define X86_FEATURE_CQM_MBM_TOTAL (11*32+ 2)/* LLC Total MBM monitoring */ #define X86_FEATURE_CQM_MBM_LOCAL (11*32+ 3)/* LLC Local MBM monitoring */ +/* Intel-defined CPU features, CPUID level 0x00000007:1 (EAX), word 12 */ +#define X86_FEATURE_AVX512_BF16 (12*32+ 5)/* AVX512 BFLOAT16 instructions */ +/* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */ #define X86_FEATURE_CLZERO (13*32+ 0)/* CLZERO instruction */ #define X86_FEATURE_IRPERF (13*32+ 1)/* Instructions Retired Count */
arch/x86/kernel/cpu/common.c +3 -0
@@ -847,6 +847,9 @@ void get_cpu_cap(struct cpuinfo_x86 *c) c->x86_capability[CPUID_7_0_EBX ] = ebx; c->x86_capability[CPUID_7_ECX ] = ecx; c->x86_capability[CPUID_7_EDX ] = edx; + +cpuid_count (0x00000007 ,1 , &eax, &ebx, &ecx, &edx); + c->x86_capability[CPUID_7_1_EAX ] = eax; }/* Extended state features: level 0x0000000d */
arch/x86/kernel/cpu/cpuid-deps.c +1 -0
@@ -62,6 +62,7 @@ static const struct cpuid_dep cpuid_deps[] = { { X86_FEATURE_CQM_OCCUP_LLC, X86_FEATURE_CQM_LLC }, { X86_FEATURE_CQM_MBM_TOTAL, X86_FEATURE_CQM_LLC }, { X86_FEATURE_CQM_MBM_LOCAL, X86_FEATURE_CQM_LLC }, + { X86_FEATURE_AVX512_BF16, X86_FEATURE_AVX512VL }, {} };
Signature · 1 line
2.19.1
Borislav Petkov
maintainerRe: [PATCH v2 2/2] x86/cpufeatures: Enumerate new AVX512 BFLOAT16 instructions
arch/x86/kernel/cpu/common.c void get_cpu_cap(struct cpuinfo_x86 *c) · line 853
Quoted 12 lines · Fenghua Yu
arch/x86/kernel/cpu/common.c +3
-0@@ -847,6 +847,9 @@ void get_cpu_cap(struct cpuinfo_x86 *c) c->x86_capability[CPUID_7_0_EBX ] = ebx; c->x86_capability[CPUID_7_ECX ] = ecx; c->x86_capability[CPUID_7_EDX ] = edx; + +cpuid_count (0x00000007 ,1 , &eax, &ebx, &ecx, &edx); + c->x86_capability[CPUID_7_1_EAX ] = eax; }
You need to test the sub-leaf index validity here before accessing subleaf 1:
arch/x86/kernel/cpu/common.c +5 -2
@@ -848,8 +848,11 @@ void get_cpu_cap(struct cpuinfo_x86 *c) c->x86_capability[CPUID_7_ECX ] = ecx; c->x86_capability[CPUID_7_EDX ] = edx; -cpuid_count (0x00000007 ,1 , &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_7_1_EAX ] = eax; +/* Check valid sub-leaf index before accessing it */ +if (eax >=1 ) { +cpuid_count (0x00000007 ,1 , &eax, &ebx, &ecx, &edx); + c->x86_capability[CPUID_7_1_EAX ] = eax; + } }/* Extended state features: level 0x0000000d */
Signature · 4 lines
Regards/Gruss,
Boris.
Good mailing practices for 400: avoid top-posting and trim the reply.Re: [PATCH v2 2/2] x86/cpufeatures: Enumerate new AVX512 BFLOAT16 instructions
Quoted 9 lines · Borislav Petkov
On Wed, Jun 19, 2019 at 02:34:04PM -0700, Fenghua Yu wrote:
You need to test the sub-leaf index validity here before accessing subleaf 1:arch/x86/kernel/cpu/common.c +0
-0
arch/x86/kernel/cpu/common.c void get_cpu_cap(struct cpuinfo_x86 *c) · line 855
Quoted 10 lines
arch/x86/kernel/cpu/common.c +5
-2@@ -848,8 +848,11 @@ void get_cpu_cap(struct cpuinfo_x86 *c) c->x86_capability[CPUID_7_ECX ] = ecx; c->x86_capability[CPUID_7_EDX ] = edx; -cpuid_count (0x00000007 ,1 , &eax, &ebx, &ecx, &edx); - c->x86_capability[CPUID_7_1_EAX ] = eax; +/* Check valid sub-leaf index before accessing it */ +if (eax >=1 ) { +cpuid_count (0x00000007 ,1 , &eax, &ebx, &ecx, &edx); + c->x86_capability[CPUID_7_1_EAX ] = eax; + }
You are right.
I tested the three patches in tip tree and they work as expected.
Thank you very much!
-Fenghua
tip-bot for Fenghua Yu
merge bot[tip:x86/cpu] x86/cpufeatures: Enumerate the new AVX512 BFLOAT16 instructions
In reply to Fenghua Yux86/cpufeatures: Enumerate the new AVX512 BFLOAT16 instructions
Commit-ID: b302e4b176d00e1cbc80148c5d0aee36751f7480 Gitweb: https://git.kernel.org/tip/b302e4b176d00e1cbc80148c5d0aee36751f7480 Author: Fenghua Yu <[email protected]> AuthorDate: Mon, 17 Jun 2019 11:00:16 -0700 Committer: Borislav Petkov <[email protected]> CommitDate: Thu, 20 Jun 2019 12:38:49 +0200
x86/cpufeatures: Enumerate the new AVX512 BFLOAT16 instructions
AVX512 BFLOAT16 instructions support 16-bit BFLOAT16 floating-point format (BF16) for deep learning optimization.
BF16 is a short version of 32-bit single-precision floating-point format (FP32) and has several advantages over 16-bit half-precision floating-point format (FP16). BF16 keeps FP32 accumulation after multiplication without loss of precision, offers more than enough range for deep learning training tasks, and doesn't need to handle hardware exception.
AVX512 BFLOAT16 instructions are enumerated in CPUID.7.1:EAX[bit 5] AVX512_BF16.
CPUID.7.1:EAX contains only feature bits. Reuse the currently empty word 12 as a pure features word to hold the feature bits including AVX512_BF16.
Detailed information of the CPUID bit and AVX512 BFLOAT16 instructions can be found in the latest Intel Architecture Instruction Set Extensions and Future Features Programming Reference.
[ bp: Check CPUID(7) subleaf validity before accessing subleaf 1. ]
--- arch/x86/include/asm/cpufeature.h | 2 +- arch/x86/include/asm/cpufeatures.h | 3 +++ arch/x86/kernel/cpu/common.c | 6 ++++++ arch/x86/kernel/cpu/cpuid-deps.c | 1 + 4 files changed, 11 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeature.h +1 -1
@@ -23,7 +23,7 @@ enum cpuid_leafsCPUID_7_0_EBX ,CPUID_D_1_EAX ,CPUID_LNX_4 , -CPUID_DUMMY , +CPUID_7_1_EAX ,CPUID_8000_0008_EBX ,CPUID_6_EAX , CPUID_8000_000A_EDX,
arch/x86/include/asm/cpufeatures.h +3 -0
@@ -282,6 +282,9 @@#define X86_FEATURE_CQM_MBM_TOTAL (11*32+ 2)/* LLC Total MBM monitoring */ #define X86_FEATURE_CQM_MBM_LOCAL (11*32+ 3)/* LLC Local MBM monitoring */ +/* Intel-defined CPU features, CPUID level 0x00000007:1 (EAX), word 12 */ +#define X86_FEATURE_AVX512_BF16 (12*32+ 5)/* AVX512 BFLOAT16 instructions */ +/* AMD-defined CPU features, CPUID level 0x80000008 (EBX), word 13 */ #define X86_FEATURE_CLZERO (13*32+ 0)/* CLZERO instruction */ #define X86_FEATURE_IRPERF (13*32+ 1)/* Instructions Retired Count */
arch/x86/kernel/cpu/common.c +6 -0
@@ -847,6 +847,12 @@ void get_cpu_cap(struct cpuinfo_x86 *c) c->x86_capability[CPUID_7_0_EBX ] = ebx; c->x86_capability[CPUID_7_ECX ] = ecx; c->x86_capability[CPUID_7_EDX ] = edx; + +/* Check valid sub-leaf index before accessing it */ +if (eax >=1 ) { +cpuid_count (0x00000007 ,1 , &eax, &ebx, &ecx, &edx); + c->x86_capability[CPUID_7_1_EAX ] = eax; + } }/* Extended state features: level 0x0000000d */
arch/x86/kernel/cpu/cpuid-deps.c +1 -0
@@ -62,6 +62,7 @@ static const struct cpuid_dep cpuid_deps[] = { { X86_FEATURE_CQM_OCCUP_LLC, X86_FEATURE_CQM_LLC }, { X86_FEATURE_CQM_MBM_TOTAL, X86_FEATURE_CQM_LLC }, { X86_FEATURE_CQM_MBM_LOCAL, X86_FEATURE_CQM_LLC }, + { X86_FEATURE_AVX512_BF16, X86_FEATURE_AVX512VL }, {} };