| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Tue, 19 Aug 2008 12:02:00 -0700
From: Harvey Harrison <harvey.harrison@...il.com>
To: David Miller <davem@...emloft.net>
Cc: akpm@...ux-foundation.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 18/23] sparc: use the new byteorder headers
On Tue, 2008-08-19 at 01:43 -0700, David Miller wrote:
> From: Harvey Harrison <harvey.harrison@...il.com>
> Date: Mon, 18 Aug 2008 17:48:17 -0700
>
> > Signed-off-by: Harvey Harrison <harvey.harrison@...il.com>
>
> I'm not so sure about this.
>
> If I understand the ___swab*() inlines in linux/swab.h,
> it has the following priority of swapping methods:
>
> 1) If arch defines __arch_swab*(), this is used.
>
> 2) If arch defines __arch_swab*p(), variable is popped onto
> the stack and we do the pointer based operation.
2a) If defined(__SWAB_64_THRU_32__) swab64 uses swab32 internally.
>
> 3) Else normal C version is used.
>
Your above understanding is correct.
> Case #2 is totally disagree with.
>
> Especially for small swaps such as 16-bit the inline expansion
> of the portable C code is going to be much better than popping
> the variable onto and then back off the stack.
>
I thought gcc wasn't too bad for this case these days for attribute_const
inlines? But without evidence to back it up, you're right that the
generic C version should just be used if an arch hasn't provided an
override.
> Sparc 64-bit only provides the __arch_swab*p() routines so
> #2 is what will in fact be used here.
>
> So NACK based upon that analysis.
From: Harvey Harrison <harvey.harrison@...il.com>
Subject: [PATCH] byteorder: use generic C version for value byteswapping
David Miller noted that popping the variable on and back off the stack
will probably be more expensive than just using the generic C byteswapping
code. Remove the fallback to the swap-from-pointer helper when no
arch override for the value byteswap has been defined.
Signed-off-by: Harvey Harrison <harvey.harrison@...il.com>
---
include/linux/swab.h | 10 ----------
1 files changed, 0 insertions(+), 10 deletions(-)
diff --git a/include/linux/swab.h b/include/linux/swab.h
index 270d5c2..bbed279 100644
--- a/include/linux/swab.h
+++ b/include/linux/swab.h
@@ -47,8 +47,6 @@ static inline __attribute_const__ __u16 ___swab16(__u16 val)
{
#ifdef __arch_swab16
return __arch_swab16(val);
-#elif defined(__arch_swab16p)
- return __arch_swab16p(&val);
#else
return __const_swab16(val);
#endif
@@ -58,8 +56,6 @@ static inline __attribute_const__ __u32 ___swab32(__u32 val)
{
#ifdef __arch_swab32
return __arch_swab32(val);
-#elif defined(__arch_swab32p)
- return __arch_swab32p(&val);
#else
return __const_swab32(val);
#endif
@@ -69,8 +65,6 @@ static inline __attribute_const__ __u64 ___swab64(__u64 val)
{
#ifdef __arch_swab64
return __arch_swab64(val);
-#elif defined(__arch_swab64p)
- return __arch_swab64p(&val);
#elif defined(__SWAB_64_THRU_32__)
__u32 h = val >> 32;
__u32 l = val & ((1ULL << 32) - 1);
@@ -84,8 +78,6 @@ static inline __attribute_const__ __u32 ___swahw32(__u32 val)
{
#ifdef __arch_swahw32
return __arch_swahw32(val);
-#elif defined(__arch_swahw32p)
- return __arch_swahw32p(&val);
#else
return __const_swahw32(val);
#endif
@@ -95,8 +87,6 @@ static inline __attribute_const__ __u32 ___swahb32(__u32 val)
{
#ifdef __arch_swahb32
return __arch_swahb32(val);
-#elif defined(__arch_swahb32p)
- return __arch_swahb32p(&val);
#else
return __const_swahb32(val);
#endif
--
1.6.0.274.g8aacc
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists