x86/cpu: Use SERIALIZE in sync_core() when available
Subsystems (3)
PATCH v4 1/1 x86/cpu: Use SERIALIZE in sync_core() when available
The SERIALIZE instruction gives software a way to force the processor to complete all modifications to flags, registers and memory from previous instructions and drain all buffered writes to memory before the next instruction is fetched and executed. Thus, it serves the purpose of sync_core(). Use it when available.
---
This is v4 from my three previous submission [1], [2], and [3]. The first three patches of the series have been merged in Linus' tree. Hence, I am submitting only this patch for review.
[1]. https://lkml.org/lkml/2020/7/27/8
[2]. https://lkml.org/lkml/2020/8/4/1090
[3]. https://lkml.org/lkml/2020/8/6/808
Changes since v3:
* Reworked comments in sync_core() for better readability. (Dave Hansen)
* Reworked the implementation to align with the style in special_insns.h. No functional changes were introduced. (Tony Luck)
Changes since v2:
* Support serialize with static_cpu_has() instead of using alternative runtime patching directly. (Borislav Petkov)
Changes since v1:
* Support SERIALIZE using alternative runtime patching. (Peter Zijlstra, H. Peter Anvin)
* Added a note to specify which version of binutils supports SERIALIZE. (Peter Zijlstra)
* Verified that (::: "memory") is used. (H. Peter Anvin)
--- arch/x86/include/asm/special_insns.h | 6 ++++++ arch/x86/include/asm/sync_core.h | 26 ++++++++++++++++++-------- 2 files changed, 24 insertions(+), 8 deletions(-)
arch/x86/include/asm/special_insns.h +6 -0
@@ -234,6 +234,12 @@ static inline void clwb(volatile void *__p)#define nop () asm volatile ("nop") +static inline void serialize (void ) +{ +/* Instruction opcode for SERIALIZE; supported in binutils >= 2.35. */ + asmvolatile (".byte 0xf, 0x1, 0xe8" :::"memory" ); +} +#endif /* __KERNEL__ */ #endif /* _ASM_X86_SPECIAL_INSNS_H */
arch/x86/include/asm/sync_core.h +18 -8
@@ -5,6 +5,7 @@#include <linux/preempt.h> #include <asm/processor.h> #include <asm/cpufeature.h> +#include <asm/special_insns.h> #ifdef CONFIG_X86_32 static inline void iret_to_self (void )
@@ -54,14 +55,23 @@ static inline void iret_to_self(void)static inline void sync_core (void ) {/* -* There are quite a few ways to do this. IRET-to-self is nice -* because it works on every CPU, at any CPL (so it's compatible -* with paravirtualization), and it never exits to a hypervisor. -* The only down sides are that it's a bit slow (it seems to be -* a bit more than 2x slower than the fastest options) and that -* it unmasks NMIs. The "push %cs" is needed because, in -* paravirtual environments, __KERNEL_CS may not be a valid CS -* value when we do IRET directly. +* The SERIALIZE instruction is the most straightforward way to +* do this but it not universally available. +*/ +if (static_cpu_has (X86_FEATURE_SERIALIZE )) { +serialize (); +return ; + } + +/* +* For all other processors, there are quite a few ways to do this +* IRET-to-self is nice because it works on every CPU, at any CPL +* (so it's compatible with paravirtualization), and it never exits +* to a hypervisor. The only down sides are that it's a bit slow +* (it seems to be a bit more than 2x slower than the fastest +* options) and that it unmasks NMIs. The "push %cs" is needed +* because, in paravirtual environments, __KERNEL_CS may not be a +* valid CS value when we do IRET directly. * * In case NMI unmasking or performance ever becomes a problem, * the next best option appears to be MOV-to-CR2 and an
Signature · 1 line
2.17.1
The following commit has been merged into the x86/cpu branch of tip:
The following commit has been merged into the x86/cpu branch of tip:
Commit-ID: bf9c912f9a649776c2d741310486a6984edaac72 Gitweb: https://git.kernel.org/tip/bf9c912f9a649776c2d741310486a6984edaac72 Author: Ricardo Neri <[email protected]> AuthorDate: Thu, 06 Aug 2020 20:28:33 -07:00 Committer: Borislav Petkov <[email protected]> CommitterDate: Mon, 17 Aug 2020 17:23:04 +02:00
x86/cpu: Use SERIALIZE in sync_core() when available
The SERIALIZE instruction gives software a way to force the processor to complete all modifications to flags, registers and memory from previous instructions and drain all buffered writes to memory before the next instruction is fetched and executed. Thus, it serves the purpose of sync_core(). Use it when available.
--- arch/x86/include/asm/special_insns.h | 6 ++++++ arch/x86/include/asm/sync_core.h | 26 ++++++++++++++++++-------- 2 files changed, 24 insertions(+), 8 deletions(-)
arch/x86/include/asm/special_insns.h +6 -0
@@ -234,6 +234,12 @@ static inline void clwb(volatile void *__p)#define nop () asm volatile ("nop") +static inline void serialize (void ) +{ +/* Instruction opcode for SERIALIZE; supported in binutils >= 2.35. */ + asmvolatile (".byte 0xf, 0x1, 0xe8" :::"memory" ); +} +#endif /* __KERNEL__ */ #endif /* _ASM_X86_SPECIAL_INSNS_H */
arch/x86/include/asm/sync_core.h +18 -8
@@ -5,6 +5,7 @@#include <linux/preempt.h> #include <asm/processor.h> #include <asm/cpufeature.h> +#include <asm/special_insns.h> #ifdef CONFIG_X86_32 static inline void iret_to_self (void )
@@ -54,14 +55,23 @@ static inline void iret_to_self(void)static inline void sync_core (void ) {/* -* There are quite a few ways to do this. IRET-to-self is nice -* because it works on every CPU, at any CPL (so it's compatible -* with paravirtualization), and it never exits to a hypervisor. -* The only down sides are that it's a bit slow (it seems to be -* a bit more than 2x slower than the fastest options) and that -* it unmasks NMIs. The "push %cs" is needed because, in -* paravirtual environments, __KERNEL_CS may not be a valid CS -* value when we do IRET directly. +* The SERIALIZE instruction is the most straightforward way to +* do this but it not universally available. +*/ +if (static_cpu_has (X86_FEATURE_SERIALIZE )) { +serialize (); +return ; + } + +/* +* For all other processors, there are quite a few ways to do this. +* IRET-to-self is nice because it works on every CPU, at any CPL +* (so it's compatible with paravirtualization), and it never exits +* to a hypervisor. The only down sides are that it's a bit slow +* (it seems to be a bit more than 2x slower than the fastest +* options) and that it unmasks NMIs. The "push %cs" is needed +* because, in paravirtual environments, __KERNEL_CS may not be a +* valid CS value when we do IRET directly. * * In case NMI unmasking or performance ever becomes a problem, * the next best option appears to be MOV-to-CR2 and an
PATCH 1/1 x86/cpu: Fix typos and improve the comments in sync_core()
Quoted 21 lines · * tip-bot2 for Ricardo Neri <[email protected]> wrote:
arch/x86/include/asm/sync_core.h +8
-0@@ -5,6 +5,7 @@#include <linux/preempt.h> #include <asm/processor.h> #include <asm/cpufeature.h> +#include <asm/special_insns.h> #ifdef CONFIG_X86_32 static inline void iret_to_self (void )@@ -54,14 +55,23 @@ static inline void iret_to_self(void)static inline void sync_core (void ) {/* +* The SERIALIZE instruction is the most straightforward way to +* do this but it not universally available. +*/ +if (static_cpu_has (X86_FEATURE_SERIALIZE )) { +serialize (); +return ; + }
arch/x86/include/asm/sync_core.h static inline void iret_to_self(void) · line 74
Quoted 10 lines
arch/x86/include/asm/sync_core.h +10
-0+ +/* +* For all other processors, there are quite a few ways to do this. +* IRET-to-self is nice because it works on every CPU, at any CPL +* (so it's compatible with paravirtualization), and it never exits +* to a hypervisor. The only down sides are that it's a bit slow +* (it seems to be a bit more than 2x slower than the fastest +* options) and that it unmasks NMIs. The "push %cs" is needed +* because, in paravirtual environments, __KERNEL_CS may not be a +* valid CS value when we do IRET directly.
So there's two typos in the new comments, there are at least two misapplied commas, it departs from existing style, and there's a typo in the existing comments as well.
Also, before this patch the 'compiler barrier' comment was valid for the whole function (there was no branching), but after this patch it reads of it was only valid for the legacy IRET-to-self branch.
Which together broke my detector and triggered a bit of compulsive bike-shed painting. ;-) See the resulting patch below.
Thanks,
Ingo
================>
From: Ingo Molnar <[email protected]>
Date: Tue, 18 Aug 2020 07:24:05 +0200
Subject: [PATCH] x86/cpu: Fix typos and improve the comments in sync_core()
- Fix typos.
- Move the compiler barrier comment to the top, because it's valid for the whole function, not just the legacy branch.
--- arch/x86/include/asm/sync_core.h | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-)
arch/x86/include/asm/sync_core.h +8 -8
@@ -47,16 +47,19 @@ static inline void iret_to_self(void)* * b) Text was modified on a different CPU, may subsequently be * executed on this CPU, and you want to make sure the new version -* gets executed. This generally means you're calling this in a IPI. +* gets executed. This generally means you're calling this in an IPI. * * If you're calling this for a different reason, you're probably doing * it wrong. +* +* Like all of Linux's memory ordering operations, this is a +* compiler barrier as well. */ static inline void sync_core (void ) {/* * The SERIALIZE instruction is the most straightforward way to -* do this but it not universally available. +* do this, but it is not universally available. */ if (static_cpu_has (X86_FEATURE_SERIALIZE )) {serialize ();
@@ -67,10 +70,10 @@ static inline void sync_core(void)* For all other processors, there are quite a few ways to do this. * IRET-to-self is nice because it works on every CPU, at any CPL * (so it's compatible with paravirtualization), and it never exits -* to a hypervisor. The only down sides are that it's a bit slow +* to a hypervisor. The only downsides are that it's a bit slow * (it seems to be a bit more than 2x slower than the fastest -* options) and that it unmasks NMIs. The "push %cs" is needed -* because, in paravirtual environments, __KERNEL_CS may not be a +* options) and that it unmasks NMIs. The "push %cs" is needed, +* because in paravirtual environments __KERNEL_CS may not be a * valid CS value when we do IRET directly. * * In case NMI unmasking or performance ever becomes a problem,
@@ -81,9 +84,6 @@ static inline void sync_core(void)* CPUID is the conventional way, but it's nasty: it doesn't * exist on some 486-like CPUs, and it usually exits to a * hypervisor. -* -* Like all of Linux's memory ordering operations, this is a -* compiler barrier as well. */ iret_to_self (); }
Re: [PATCH] x86/cpu: Fix typos and improve the comments in sync_core()
Quoted 78 lines · Ingo Molnar
Quoted 21 lines · * tip-bot2 for Ricardo Neri <[email protected]> wrote:
arch/x86/include/asm/sync_core.h +8
-0@@ -5,6 +5,7 @@#include <linux/preempt.h> #include <asm/processor.h> #include <asm/cpufeature.h> +#include <asm/special_insns.h> #ifdef CONFIG_X86_32 static inline void iret_to_self (void )@@ -54,14 +55,23 @@ static inline void iret_to_self(void)static inline void sync_core (void ) {/* +* The SERIALIZE instruction is the most straightforward way to +* do this but it not universally available. +*/ +if (static_cpu_has (X86_FEATURE_SERIALIZE )) { +serialize (); +return ; + }arch/x86/include/asm/sync_core.h static inline void iret_to_self(void) · line 74
Quoted 10 lines
arch/x86/include/asm/sync_core.h +10
-0+ +/* +* For all other processors, there are quite a few ways to do this. +* IRET-to-self is nice because it works on every CPU, at any CPL +* (so it's compatible with paravirtualization), and it never exits +* to a hypervisor. The only down sides are that it's a bit slow +* (it seems to be a bit more than 2x slower than the fastest +* options) and that it unmasks NMIs. The "push %cs" is needed +* because, in paravirtual environments, __KERNEL_CS may not be a +* valid CS value when we do IRET directly. So there's two typos in the new comments, there are at least two misapplied commas, it departs from existing style, and there's a typo in the existing comments as well.
Also, before this patch the 'compiler barrier' comment was valid for the whole function (there was no branching), but after this patch it reads of it was only valid for the legacy IRET-to-self branch.
Which together broke my detector and triggered a bit of compulsive bike-shed painting. ;-) See the resulting patch below.
Thanks,
Ingo
================>
From: Ingo Molnar <[email protected]>
Date: Tue, 18 Aug 2020 07:24:05 +0200
Subject: [PATCH] x86/cpu: Fix typos and improve the comments insync_core()- Fix typos.
- Move the compiler barrier comment to the top, because it's valid for the whole function, not just the legacy branch.
Signed-off-by Ingo Molnar <[email protected]>--- arch/x86/include/asm/sync_core.h | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-)arch/x86/include/asm/sync_core.h +1
-1@@ -47,16 +47,19 @@ static inline void iret_to_self(void)* * b) Text was modified on a different CPU, may subsequently be * executed on this CPU, and you want to make sure the new version -* gets executed. This generally means you're calling this in a IPI. +* gets executed. This generally means you're calling this in an IPI. * * If you're calling this for a different reason, you're probably doing * it wrong.
arch/x86/include/asm/sync_core.h static inline void iret_to_self(void) · line 62
Quoted 10 lines
arch/x86/include/asm/sync_core.h +4
-1+* +* Like all of Linux's memory ordering operations, this is a +* compiler barrier as well. */ static inline void sync_core (void ) {/* * The SERIALIZE instruction is the most straightforward way to -* do this but it not universally available. +* do this, but it is not universally available.
Indeed, I missed this grammar error.
Quoted 4 lines
arch/x86/include/asm/sync_core.h +0
-0*/ if (static_cpu_has (X86_FEATURE_SERIALIZE )) {serialize ();
arch/x86/include/asm/sync_core.h static inline void sync_core(void) · line 76
Quoted 10 lines
arch/x86/include/asm/sync_core.h +3
-3@@ -67,10 +70,10 @@ static inline void sync_core(void)* For all other processors, there are quite a few ways to do this. * IRET-to-self is nice because it works on every CPU, at any CPL * (so it's compatible with paravirtualization), and it never exits -* to a hypervisor. The only down sides are that it's a bit slow +* to a hypervisor. The only downsides are that it's a bit slow * (it seems to be a bit more than 2x slower than the fastest -* options) and that it unmasks NMIs. The "push %cs" is needed -* because, in paravirtual environments, __KERNEL_CS may not be a +* options) and that it unmasks NMIs. The "push %cs" is needed, +* because in paravirtual environments __KERNEL_CS may not be a
I didn't realize that the double spaces after the period were part of the style.
FWIW,
Re: [PATCH] x86/cpu: Fix typos and improve the comments in sync_core()
Quoted 22 lines · Ricardo Neri
Quoted 9 lines
arch/x86/include/asm/sync_core.h +1
-1@@ -47,16 +47,19 @@ static inline void iret_to_self(void)* * b) Text was modified on a different CPU, may subsequently be * executed on this CPU, and you want to make sure the new version -* gets executed. This generally means you're calling this in a IPI. +* gets executed. This generally means you're calling this in an IPI. * * If you're calling this for a different reason, you're probably doing * it wrong. arch/x86/include/asm/sync_core.h static inline void iret_to_self(void) · line 62
Quoted 10 lines
arch/x86/include/asm/sync_core.h +4
-1+* +* Like all of Linux's memory ordering operations, this is a +* compiler barrier as well. */ static inline void sync_core (void ) {/* * The SERIALIZE instruction is the most straightforward way to -* do this but it not universally available. +* do this, but it is not universally available. Indeed, I missed this grammar error.
Quoted 9 lines
arch/x86/include/asm/sync_core.h static inline void sync_core(void) · line 73
Quoted 9 lines
arch/x86/include/asm/sync_core.h +1
-1*/ if (static_cpu_has (X86_FEATURE_SERIALIZE )) {serialize ();@@ -67,10 +70,10 @@ static inline void sync_core(void)* For all other processors, there are quite a few ways to do this. * IRET-to-self is nice because it works on every CPU, at any CPL * (so it's compatible with paravirtualization), and it never exits -* to a hypervisor. The only down sides are that it's a bit slow +* to a hypervisor. The only downsides are that it's a bit slow
And this one - it's "downsides" not "down sides".
Quoted 8 lines
arch/x86/include/asm/sync_core.h static inline void sync_core(void) · line 76
Quoted 5 lines
arch/x86/include/asm/sync_core.h +2
-2* (it seems to be a bit more than 2x slower than the fastest -* options) and that it unmasks NMIs. The "push %cs" is needed -* because, in paravirtual environments, __KERNEL_CS may not be a +* options) and that it unmasks NMIs. The "push %cs" is needed, +* because in paravirtual environments __KERNEL_CS may not be a I didn't realize that the double spaces after the period were part of the style.
They are not, but *consistent* use of typographic details is part of the style, and here we were mixing two styles within the same comment block.
Quoted 3 lines
FWIW,
Reviewed-by Ricardo Neri <[email protected]>
Thanks,
Ingo
[tip: x86/cpu] x86/cpu: Fix typos and improve the comments in sync_core()
In reply to Ingo MolnarThe following commit has been merged into the x86/cpu branch of tip:
The following commit has been merged into the x86/cpu branch of tip:
Commit-ID: 40eb0cb4939e462acfedea8c8064571e886b9773 Gitweb: https://git.kernel.org/tip/40eb0cb4939e462acfedea8c8064571e886b9773 Author: Ingo Molnar <[email protected]> AuthorDate: Tue, 18 Aug 2020 07:31:30 +02:00 Committer: Ingo Molnar <[email protected]> CommitterDate: Wed, 19 Aug 2020 09:56:36 +02:00
x86/cpu: Fix typos and improve the comments in sync_core()
- Fix typos.
- Move the compiler barrier comment to the top, because it's valid for the whole function, not just the legacy branch.
--- arch/x86/include/asm/sync_core.h | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-)
arch/x86/include/asm/sync_core.h +8 -8
@@ -47,16 +47,19 @@ static inline void iret_to_self(void)* * b) Text was modified on a different CPU, may subsequently be * executed on this CPU, and you want to make sure the new version -* gets executed. This generally means you're calling this in a IPI. +* gets executed. This generally means you're calling this in an IPI. * * If you're calling this for a different reason, you're probably doing * it wrong. +* +* Like all of Linux's memory ordering operations, this is a +* compiler barrier as well. */ static inline void sync_core (void ) {/* * The SERIALIZE instruction is the most straightforward way to -* do this but it not universally available. +* do this, but it is not universally available. */ if (static_cpu_has (X86_FEATURE_SERIALIZE )) {serialize ();
@@ -67,10 +70,10 @@ static inline void sync_core(void)* For all other processors, there are quite a few ways to do this. * IRET-to-self is nice because it works on every CPU, at any CPL * (so it's compatible with paravirtualization), and it never exits -* to a hypervisor. The only down sides are that it's a bit slow +* to a hypervisor. The only downsides are that it's a bit slow * (it seems to be a bit more than 2x slower than the fastest -* options) and that it unmasks NMIs. The "push %cs" is needed -* because, in paravirtual environments, __KERNEL_CS may not be a +* options) and that it unmasks NMIs. The "push %cs" is needed, +* because in paravirtual environments __KERNEL_CS may not be a * valid CS value when we do IRET directly. * * In case NMI unmasking or performance ever becomes a problem,
@@ -81,9 +84,6 @@ static inline void sync_core(void)* CPUID is the conventional way, but it's nasty: it doesn't * exist on some 486-like CPUs, and it usually exits to a * hypervisor. -* -* Like all of Linux's memory ordering operations, this is a -* compiler barrier as well. */ iret_to_self (); }