From: Peter Zijlstra <peterz@infradead.org>
To: torvalds@linux-foundation.org
Cc: corbet@lwn.net, will@kernel.org, boqun.feng@gmail.com,
mark.rutland@arm.com, catalin.marinas@arm.com, dennis@kernel.org,
tj@kernel.org, cl@linux.com, hca@linux.ibm.com,
gor@linux.ibm.com, agordeev@linux.ibm.com,
borntraeger@linux.ibm.com, svens@linux.ibm.com,
tglx@linutronix.de, mingo@redhat.com, bp@alien8.de,
dave.hansen@linux.intel.com, x86@kernel.org, hpa@zytor.com,
joro@8bytes.org, suravee.suthikulpanit@amd.com,
robin.murphy@arm.com, dwmw2@infradead.org,
baolu.lu@linux.intel.com, Arnd Bergmann <arnd@arndb.de>,
Herbert Xu <herbert@gondor.apana.org.au>,
davem@davemloft.net, penberg@kernel.org, rientjes@google.com,
iamjoonsoo.kim@lge.com, Andrew Morton <akpm@linux-foundation.org>,
vbabka@suse.cz, roman.gushchin@linux.dev, 42.hyeyoo@gmail.com,
linux-doc@vger.kernel.org, linux-kernel@vger.kernel.org,
linux-mm@kvack.org, linux-s390@vger.kernel.org,
iommu@lists.linux.dev, linux-arch@vger.kernel.org,
linux-crypto@vger.kernel.org
Subject: Re: [PATCH v3 05/11] percpu: Wire up cmpxchg128
Date: Thu, 25 May 2023 14:49:55 +0200 [thread overview]
Message-ID: <20230525124955.GS83892@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <20230515080554.248739380@infradead.org>
On Mon, May 15, 2023 at 09:57:04AM +0200, Peter Zijlstra wrote:
> In order to replace cmpxchg_double() with the newly minted
> cmpxchg128() family of functions, wire it up in this_cpu_cmpxchg().
>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
> --- a/arch/x86/include/asm/percpu.h
> +++ b/arch/x86/include/asm/percpu.h
> @@ -210,6 +210,65 @@ do { \
> (typeof(_var))(unsigned long) pco_old__; \
> })
>
> +#if defined(CONFIG_X86_32) && defined(CONFIG_X86_CMPXCHG64)
> +#define percpu_cmpxchg64_op(size, qual, _var, _oval, _nval) \
> +({ \
> + union { \
> + u64 var; \
> + struct { \
> + u32 low, high; \
> + }; \
> + } old__, new__; \
> + \
> + old__.var = _oval; \
> + new__.var = _nval; \
> + \
> + asm qual ("cmpxchg8b " __percpu_arg([var]) \
> + : [var] "+m" (_var), \
> + "+a" (old__.low), \
> + "+d" (old__.high) \
> + : "b" (new__.low), \
> + "c" (new__.high) \
> + : "memory"); \
> + \
> + old__.var; \
> +})
> +
> +#define raw_cpu_cmpxchg64(pcp, oval, nval) percpu_cmpxchg64_op(8, , pcp, oval, nval)
> +#define this_cpu_cmpxchg64(pcp, oval, nval) percpu_cmpxchg64_op(8, volatile, pcp, oval, nval)
> +#endif
> +
> +#ifdef CONFIG_X86_64
> +#define raw_cpu_cmpxchg64(pcp, oval, nval) percpu_cmpxchg_op(8, , pcp, oval, nval);
> +#define this_cpu_cmpxchg64(pcp, oval, nval) percpu_cmpxchg_op(8, volatile, pcp, oval, nval);
> +
> +#define percpu_cmpxchg128_op(size, qual, _var, _oval, _nval) \
> +({ \
> + union { \
> + u128 var; \
> + struct { \
> + u64 low, high; \
> + }; \
> + } old__, new__; \
> + \
> + old__.var = _oval; \
> + new__.var = _nval; \
> + \
> + asm qual ("cmpxchg16b " __percpu_arg([var]) \
> + : [var] "+m" (_var), \
> + "+a" (old__.low), \
> + "+d" (old__.high) \
> + : "b" (new__.low), \
> + "c" (new__.high) \
> + : "memory"); \
> + \
> + old__.var; \
> +})
> +
> +#define raw_cpu_cmpxchg128(pcp, oval, nval) percpu_cmpxchg128_op(16, , pcp, oval, nval)
> +#define this_cpu_cmpxchg128(pcp, oval, nval) percpu_cmpxchg128_op(16, volatile, pcp, oval, nval)
> +#endif
> +
> /*
> * this_cpu_read() makes gcc load the percpu variable every time it is
> * accessed while this_cpu_read_stable() allows the value to be cached.
Since this_cpu_cmpxchg*() is assumed to always be present (their usage
is not guarded with system_has_cmpxchg*() it needs a fallback for then
the instruction is not present.
The below has been tested with clearcpuid=cx16; adding an obvious defect
in the fallback implemention crashes the kernel, with the code as
presented it boots.
Build tested i386-defconfig.
(the things we do for museum pieces :/)
---
include/asm/percpu.h | 25 +++++++++++---------
lib/Makefile | 3 +-
lib/cmpxchg16b_emu.S | 43 ++++++++++++++++++++--------------
lib/cmpxchg8b_emu.S | 63 +++++++++++++++++++++++++++++++++++++++------------
4 files changed, 90 insertions(+), 44 deletions(-)
--- a/arch/x86/include/asm/percpu.h
+++ b/arch/x86/include/asm/percpu.h
@@ -210,7 +210,7 @@ do { \
(typeof(_var))(unsigned long) pco_old__; \
})
-#if defined(CONFIG_X86_32) && defined(CONFIG_X86_CMPXCHG64)
+#ifdef CONFIG_X86_32
#define percpu_cmpxchg64_op(size, qual, _var, _oval, _nval) \
({ \
union { \
@@ -223,13 +223,14 @@ do { \
old__.var = _oval; \
new__.var = _nval; \
\
- asm qual ("cmpxchg8b " __percpu_arg([var]) \
+ asm qual (ALTERNATIVE("leal %P[var], %%esi; call this_cpu_cmpxchg8b_emu", \
+ "cmpxchg8b " __percpu_arg([var]), X86_FEATURE_CX8) \
: [var] "+m" (_var), \
"+a" (old__.low), \
"+d" (old__.high) \
: "b" (new__.low), \
"c" (new__.high) \
- : "memory"); \
+ : "memory", "esi"); \
\
old__.var; \
})
@@ -254,13 +255,14 @@ do { \
old__.var = _oval; \
new__.var = _nval; \
\
- asm qual ("cmpxchg16b " __percpu_arg([var]) \
+ asm qual (ALTERNATIVE("leaq %P[var], %%rsi; call this_cpu_cmpxchg16b_emu", \
+ "cmpxchg16b " __percpu_arg([var]), X86_FEATURE_CX16) \
: [var] "+m" (_var), \
"+a" (old__.low), \
"+d" (old__.high) \
: "b" (new__.low), \
"c" (new__.high) \
- : "memory"); \
+ : "memory", "rsi"); \
\
old__.var; \
})
@@ -400,12 +402,13 @@ do { \
bool __ret; \
typeof(pcp1) __o1 = (o1), __n1 = (n1); \
typeof(pcp2) __o2 = (o2), __n2 = (n2); \
- alternative_io("leaq %P1,%%rsi\n\tcall this_cpu_cmpxchg16b_emu\n\t", \
- "cmpxchg16b " __percpu_arg(1) "\n\tsetz %0\n\t", \
- X86_FEATURE_CX16, \
- ASM_OUTPUT2("=a" (__ret), "+m" (pcp1), \
- "+m" (pcp2), "+d" (__o2)), \
- "b" (__n1), "c" (__n2), "a" (__o1) : "rsi"); \
+ asm volatile (ALTERNATIVE("leaq %P1, %%rsi; call this_cpu_cmpxchg16b_emu", \
+ "cmpxchg16b " __percpu_arg(1), X86_FEATURE_CX16) \
+ "setz %0" \
+ : "=a" (__ret), "+m" (pcp1) \
+ : "b" (__n1), "c" (__n2), \
+ "a" (__o1), "d" (__o2) \
+ : "memory", "rsi"); \
__ret; \
})
--- a/arch/x86/lib/Makefile
+++ b/arch/x86/lib/Makefile
@@ -61,8 +61,9 @@ ifeq ($(CONFIG_X86_32),y)
lib-y += strstr_32.o
lib-y += string_32.o
lib-y += memmove_32.o
+ lib-y += cmpxchg8b_emu.o
ifneq ($(CONFIG_X86_CMPXCHG64),y)
- lib-y += cmpxchg8b_emu.o atomic64_386_32.o
+ lib-y += atomic64_386_32.o
endif
else
obj-y += iomap_copy_64.o
--- a/arch/x86/lib/cmpxchg16b_emu.S
+++ b/arch/x86/lib/cmpxchg16b_emu.S
@@ -1,47 +1,54 @@
/* SPDX-License-Identifier: GPL-2.0-only */
#include <linux/linkage.h>
#include <asm/percpu.h>
+#include <asm/processor-flags.h>
.text
/*
+ * Emulate 'cmpxchg16b %gs:(%rsi)'
+ *
* Inputs:
* %rsi : memory location to compare
* %rax : low 64 bits of old value
* %rdx : high 64 bits of old value
* %rbx : low 64 bits of new value
* %rcx : high 64 bits of new value
- * %al : Operation successful
+ *
+ * Notably this is not LOCK prefixed and is not safe against NMIs
*/
SYM_FUNC_START(this_cpu_cmpxchg16b_emu)
-#
-# Emulate 'cmpxchg16b %gs:(%rsi)' except we return the result in %al not
-# via the ZF. Caller will access %al to get result.
-#
-# Note that this is only useful for a cpuops operation. Meaning that we
-# do *not* have a fully atomic operation but just an operation that is
-# *atomic* on a single cpu (as provided by the this_cpu_xx class of
-# macros).
-#
pushfq
cli
- cmpq PER_CPU_VAR((%rsi)), %rax
- jne .Lnot_same
- cmpq PER_CPU_VAR(8(%rsi)), %rdx
- jne .Lnot_same
+ /* if (*ptr == old) */
+ cmpq PER_CPU_VAR(0(%rsi)), %rax
+ jne .Lnot_same
+ cmpq PER_CPU_VAR(8(%rsi)), %rdx
+ jne .Lnot_same
+
+ /* *ptr = new */
+ movq %rbx, PER_CPU_VAR(0(%rsi))
+ movq %rcx, PER_CPU_VAR(8(%rsi))
- movq %rbx, PER_CPU_VAR((%rsi))
- movq %rcx, PER_CPU_VAR(8(%rsi))
+ /* set ZF in EFLAGS to indicate success */
+ orl $X86_EFLAGS_ZF, (%rsp)
popfq
- mov $1, %al
RET
.Lnot_same:
+ /* *ptr != old */
+
+ /* old = *ptr */
+ movq PER_CPU_VAR(0(%rsi)), %rax
+ movq PER_CPU_VAR(8(%rsi)), %rdx
+
+ /* clear ZF in EFLAGS to indicate failure */
+ andl $(~X86_EFLAGS_ZF), (%rsp)
+
popfq
- xor %al,%al
RET
SYM_FUNC_END(this_cpu_cmpxchg16b_emu)
--- a/arch/x86/lib/cmpxchg8b_emu.S
+++ b/arch/x86/lib/cmpxchg8b_emu.S
@@ -2,10 +2,16 @@
#include <linux/linkage.h>
#include <asm/export.h>
+#include <asm/percpu.h>
+#include <asm/processor-flags.h>
.text
+#ifndef CONFIG_X86_CMPXCHG64
+
/*
+ * Emulate 'cmpxchg8b (%esi)' on UP
+ *
* Inputs:
* %esi : memory location to compare
* %eax : low 32 bits of old value
@@ -15,32 +21,61 @@
*/
SYM_FUNC_START(cmpxchg8b_emu)
-#
-# Emulate 'cmpxchg8b (%esi)' on UP except we don't
-# set the whole ZF thing (caller will just compare
-# eax:edx with the expected value)
-#
pushfl
cli
- cmpl (%esi), %eax
- jne .Lnot_same
- cmpl 4(%esi), %edx
- jne .Lhalf_same
+ cmpl 0(%esi), %eax
+ jne .Lnot_same
+ cmpl 4(%esi), %edx
+ jne .Lnot_same
+
+ movl %ebx, 0(%esi)
+ movl %ecx, 4(%esi)
- movl %ebx, (%esi)
- movl %ecx, 4(%esi)
+ orl $X86_EFLAGS_ZF, (%esp)
popfl
RET
.Lnot_same:
- movl (%esi), %eax
-.Lhalf_same:
- movl 4(%esi), %edx
+ movl 0(%esi), %eax
+ movl 4(%esi), %edx
+
+ andl $(~X86_EFLAGS_ZF), (%esp)
popfl
RET
SYM_FUNC_END(cmpxchg8b_emu)
EXPORT_SYMBOL(cmpxchg8b_emu)
+
+#endif
+
+SYM_FUNC_START(this_cpu_cmpxchg8b_emu)
+
+ pushfl
+ cli
+
+ cmpl PER_CPU_VAR(0(%esi)), %eax
+ jne .Lnot_same2
+ cmpl PER_CPU_VAR(4(%esi)), %edx
+ jne .Lnot_same2
+
+ movl %ebx, PER_CPU_VAR(0(%esi))
+ movl %ecx, PER_CPU_VAR(4(%esi))
+
+ orl $X86_EFLAGS_ZF, (%esp)
+
+ popfl
+ RET
+
+.Lnot_same2:
+ movl PER_CPU_VAR(0(%esi)), %eax
+ movl PER_CPU_VAR(4(%esi)), %edx
+
+ andl $(~X86_EFLAGS_ZF), (%esp)
+
+ popfl
+ RET
+
+SYM_FUNC_END(this_cpu_cmpxchg8b_emu)
next prev parent reply other threads:[~2023-05-25 12:51 UTC|newest]
Thread overview: 45+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-05-15 7:56 [PATCH v3 00/11] Introduce cmpxchg128() -- aka. the demise of cmpxchg_double() Peter Zijlstra
2023-05-15 7:57 ` [PATCH v3 01/11] cyrpto/b128ops: Remove struct u128 Peter Zijlstra
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-15 7:57 ` [PATCH v3 02/11] types: Introduce [us]128 Peter Zijlstra
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-15 7:57 ` [PATCH v3 03/11] arch: Introduce arch_{,try_}_cmpxchg128{,_local}() Peter Zijlstra
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-15 7:57 ` [PATCH v3 04/11] instrumentation: Wire up cmpxchg128() Peter Zijlstra
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-15 7:57 ` [PATCH v3 05/11] percpu: Wire up cmpxchg128 Peter Zijlstra
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-25 12:49 ` Peter Zijlstra [this message]
2023-05-25 22:59 ` [PATCH v3 05/11] " Petr Tesařík
2023-05-15 7:57 ` [PATCH v3 06/11] x86,amd_iommu: Replace cmpxchg_double() Peter Zijlstra
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-15 7:57 ` [PATCH v3 07/11] x86,intel_iommu: " Peter Zijlstra
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-15 7:57 ` [PATCH v3 08/11] slub: " Peter Zijlstra
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-24 9:32 ` [PATCH v3 08/11] " Peter Zijlstra
2023-05-24 10:13 ` Vlastimil Babka
2023-05-25 10:29 ` Peter Zijlstra
2023-05-25 10:52 ` Arnd Bergmann
2023-05-25 13:10 ` Peter Zijlstra
2023-05-30 14:22 ` Peter Zijlstra
2023-05-30 19:32 ` Peter Zijlstra
2023-05-15 7:57 ` [PATCH v3 09/11] mm/slub: Fold slab_update_freelist() Peter Zijlstra
2023-05-24 11:58 ` Vlastimil Babka
2023-05-15 7:57 ` [PATCH v3 10/11] arch: Remove cmpxchg_double Peter Zijlstra
2023-05-15 8:52 ` Heiko Carstens
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-15 7:57 ` [PATCH v3 11/11] s390/cpum_sf: Convert to cmpxchg128() Peter Zijlstra
2023-05-20 10:49 ` [tip: locking/core] " tip-bot2 for Peter Zijlstra
2023-05-22 10:27 ` tip-bot2 for Peter Zijlstra
2023-05-15 9:42 ` [PATCH v3 00/11] Introduce cmpxchg128() -- aka. the demise of cmpxchg_double() Arnd Bergmann
2023-05-24 9:39 ` Peter Zijlstra
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230525124955.GS83892@hirez.programming.kicks-ass.net \
--to=peterz@infradead.org \
--cc=42.hyeyoo@gmail.com \
--cc=agordeev@linux.ibm.com \
--cc=akpm@linux-foundation.org \
--cc=arnd@arndb.de \
--cc=baolu.lu@linux.intel.com \
--cc=boqun.feng@gmail.com \
--cc=borntraeger@linux.ibm.com \
--cc=bp@alien8.de \
--cc=catalin.marinas@arm.com \
--cc=cl@linux.com \
--cc=corbet@lwn.net \
--cc=dave.hansen@linux.intel.com \
--cc=davem@davemloft.net \
--cc=dennis@kernel.org \
--cc=dwmw2@infradead.org \
--cc=gor@linux.ibm.com \
--cc=hca@linux.ibm.com \
--cc=herbert@gondor.apana.org.au \
--cc=hpa@zytor.com \
--cc=iamjoonsoo.kim@lge.com \
--cc=iommu@lists.linux.dev \
--cc=joro@8bytes.org \
--cc=linux-arch@vger.kernel.org \
--cc=linux-crypto@vger.kernel.org \
--cc=linux-doc@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=linux-s390@vger.kernel.org \
--cc=mark.rutland@arm.com \
--cc=mingo@redhat.com \
--cc=penberg@kernel.org \
--cc=rientjes@google.com \
--cc=robin.murphy@arm.com \
--cc=roman.gushchin@linux.dev \
--cc=suravee.suthikulpanit@amd.com \
--cc=svens@linux.ibm.com \
--cc=tglx@linutronix.de \
--cc=tj@kernel.org \
--cc=torvalds@linux-foundation.org \
--cc=vbabka@suse.cz \
--cc=will@kernel.org \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).