s390: missing ifdef in bitops.h
PATCH 1/1 s390: missing ifdef in bitops.h
From: Cedric Le Goater <[email protected]>
Add missing #ifdef __KERNEL__ to asm-s390/bitops.h
---
Patch is for -mm tree only and resolves a merge conflict.
include/asm-s390/bitops.h | 3 +++ 1 file changed, 3 insertions(+)
include/asm-s390/bitops.h +3 -0
@@ -12,6 +12,9 @@* Copyright (C) 1992, Linus Torvalds * */ + +#ifdef __KERNEL__ +#include <linux/compiler.h> /*
Quoted 1 line · Heiko Carstens
Add missing #ifdef
__KERNEL__toasm-s390/bitops.h
But asm/bitops.h isn't suitable for userspace _anyway_, is it? We should be able to drop all instances of __KERNEL__ from it.
Which means that asm-s390/posix_types.h probably ought never to be trying to include asm/bitops.h in the !__KERNEL__ case... we never had libc5 on S390 anyway, did we?
Martin, is it OK for me to add this to the hdrcleanup-2.6.git tree?
include/asm-s390/posix_types.h +2 -2
@@ -76,7 +76,7 @@ #endif /* !defined } __kernel_fsid_t; -#if defined(__KERNEL__) || !defined(__GLIBC__) || (__GLIBC__ <2 ) +#ifdef __KERNEL__#ifndef _S390_BITOPS_H#include <asm/bitops.h>
@@ -94,6 +94,6 @@ #define __FD_ISSET(fd,fdsetp) test_bit(#undef __FD_ZERO#define __FD_ZERO (fdsetp) (memset ((fdsetp), 0, sizeof(*(fd_set *)(fdsetp)))) -#endif /* defined(__KERNEL__) || !defined(__GLIBC__) || (__GLIBC__ < 2)*/ +#endif /* __KERNEL__ */ #endif
Signature · 1 line
dwmw2
Am Tuesday 13 June 2006 17:17 schrieb David Woodhouse:
Quoted 10 lines
include/asm-s390/posix_types.h +1
-1-#if defined(__KERNEL__) || !defined(__GLIBC__) || (__GLIBC__ <2 ) +#ifdef __KERNEL__
#ifndef_S390_BITOPS_H
#include <asm/bitops.h>(quoted diff) +0
-1@@ -94,6 +94,6 @@ #define __FD_ISSET(fd,fdsetp) test_bit( #undef __FD_ZERO #define __FD_ZERO(fdsetp) (memset ((fdsetp), 0, sizeof(*(fd_set *)(fdsetp)))) -#endif /* defined(__KERNEL__) || !defined(__GLIBC__) || (__GLIBC__ <
Quoted 2 lines
2)*/ +#endif /*
__KERNEL__*/
Erm, I don't think the kernel even uses those definitions, the only reason to keep them is old user space.
Now that is true for all architectures, although they all handle things slightly differently:
cris, v850, s390: use bitops, which can be considered broken arm, arm26, frv, h8300, m68k: use a trivial C macro
i386: has an inline assembly
sparc64, x86_64, alpha, xtensa, sparc, ia64, powerpc, sh, parisc, mips:
use variations of the same unrolled C inline functions
I'd suggest using only one version. In doubt, I would move the parisc version to asm-generic/fd_set.h or similar and include that from everywhere.
Arnd <><
Quoted 3 lines · Arnd Bergmann
I'd suggest using only one version. In doubt, I would move the parisc version to
asm-generic/fd_set.hor similar and include that from everywhere.
That makes sense. It's not _entirely_ trivial though, so I don't want to do it in my hdrcleanup-2.6.git tree. I'll leave it for later.
Signature · 1 line
dwmw2
Quoted 7 lines · David Woodhouse
Quoted 3 lines · Arnd Bergmann
I'd suggest using only one version. In doubt, I would move the parisc version to
asm-generic/fd_set.hor similar and include that from everywhere.That makes sense. It's not _entirely_ trivial though, so I don't want to do it in my hdrcleanup-2.6.git tree. I'll leave it for later.
In that case it would be good to get Cedric's bitops patch merged, since it fixes compilation.
Am Tuesday 13 June 2006 22:33 schrieb Arnd Bergmann:
Quoted 2 lines
Erm, I don't think the kernel even uses those definitions, the only reason to keep them is old user space.
Sorry, misinformation on my side. The kernel uses the FD_foo versions, not the __FD_foo ones, so I did not find them at first.
It's only used in fs/open.c though, and I wonder if that could just use the bitops versions directly, as these are more likely to be optimized well for a given architecture.
Arnd <><
Quoted 10 lines · Arnd Bergmann
Am Tuesday 13 June 2006 22:33 schrieb Arnd Bergmann:
Quoted 2 lines
Erm, I don't think the kernel even uses those definitions, the only reason to keep them is old user space.
Sorry, misinformation on my side. The kernel uses the
FD_fooversions, not the__FD_fooones, so I did not find them at first.It's only used in
fs/open.cthough, and I wonder if that could just use the bitops versions directly, as these are more likely to be optimized well for a given architecture.
Using set_bit and clear_bit is actually overkill since they use compare and swap to do an atomic update. That is not needed for __FD_foo. How about the attached patch ?
Signature · 82 lines
blue skies, Martin. Martin Schwidefsky Linux for zSeries Development & Services IBM Deutschland Entwicklung GmbH "Reality continues to ruin my life." - Calvin. -- From: Martin Schwidefsky <[email protected]> [S390] __FD_foo definitions. Make the definitions of __FD_SET, __FD_CLR and __FD_ISSET independent from asm/bitops.h and remove the macro magic that tests for __GLIBC__. Use simple C inline functions instead of set_bit, clear_bit and test_bit. Spotted by David Woodhouse <[email protected]> Signed-off-by: Martin Schwidefsky <[email protected]> --- include/asm-s390/posix_types.h | 42 ++++++++++++++++++++++++++--------------- 1 files changed, 27 insertions(+), 15 deletions(-) diff -urpN linux-2.6/include/asm-s390/posix_types.h linux-2.6-patched/include/asm-s390/posix_types.h --- linux-2.6/include/asm-s390/posix_types.h 2006-03-20 06:53:29.000000000 +0100 +++ linux-2.6-patched/include/asm-s390/posix_types.h 2006-06-14 12:37:54.000000000 +0200 @@ -76,24 +76,36 @@ typedef struct { } __kernel_fsid_t; -#if defined(__KERNEL__) || !defined(__GLIBC__) || (__GLIBC__ < 2) +#ifdef __KERNEL__ -#ifndef _S390_BITOPS_H -#include <asm/bitops.h> -#endif - -#undef __FD_SET -#define __FD_SET(fd,fdsetp) set_bit((fd),(fdsetp)->fds_bits) - -#undef __FD_CLR -#define __FD_CLR(fd,fdsetp) clear_bit((fd),(fdsetp)->fds_bits) - -#undef __FD_ISSET -#define __FD_ISSET(fd,fdsetp) test_bit((fd),(fdsetp)->fds_bits) +#undef __FD_SET +static inline void __FD_SET(unsigned long fd, __kernel_fd_set *fdsetp) +{ + unsigned long _tmp = fd / __NFDBITS; + unsigned long _rem = fd % __NFDBITS; + fdsetp->fds_bits[_tmp] |= (1UL<<_rem); +} + +#undef __FD_CLR +static inline void __FD_CLR(unsigned long fd, __kernel_fd_set *fdsetp) +{ + unsigned long _tmp = fd / __NFDBITS; + unsigned long _rem = fd % __NFDBITS; + fdsetp->fds_bits[_tmp] &= ~(1UL<<_rem); +} + +#undef __FD_ISSET +static inline int __FD_ISSET(unsigned long fd, const __kernel_fd_set *fdsetp) +{ + unsigned long _tmp = fd / __NFDBITS; + unsigned long _rem = fd % __NFDBITS; + return (fdsetp->fds_bits[_tmp] & (1UL<<_rem)) != 0; +} #undef __FD_ZERO -#define __FD_ZERO(fdsetp) (memset ((fdsetp), 0, sizeof(*(fd_set *)(fdsetp)))) +#define __FD_ZERO(fdsetp) \ + ((void) memset ((__ptr_t) (fdsetp), 0, sizeof (__kernel_fd_set))) -#endif /* defined(__KERNEL__) || !defined(__GLIBC__) || (__GLIBC__ < 2)*/ +#endif /* __KERNEL__ */ #endif
Quoted 3 lines · Martin Schwidefsky
Using
set_bitandclear_bitis actually overkill since they use compare and swap to do an atomic update. That is not needed for__FD_foo. How about the attached patch ?
Looks good. I'll stick that in the hdrcleanup-2.6.git tree right now, if that's OK?
Signature · 1 line
dwmw2
Quoted 7 lines · David Woodhouse
Quoted 3 lines · Martin Schwidefsky
Using
set_bitandclear_bitis actually overkill since they use compare and swap to do an atomic update. That is not needed for__FD_foo. How about the attached patch ?Looks good. I'll stick that in the hdrcleanup-2.6.git tree right now, if that's OK?
Fine with me.
Signature · 8 lines
blue skies, Martin. Martin Schwidefsky Linux for zSeries Development & Services IBM Deutschland Entwicklung GmbH "Reality continues to ruin my life." - Calvin.
Quoted 5 lines · David Woodhouse
Quoted 1 line · Heiko Carstens
Add missing #ifdef
__KERNEL__toasm-s390/bitops.hBut
asm/bitops.hisn't suitable for userspace _anyway_, is it? We should be able to drop all instances of__KERNEL__from it.
Nothing from asm/bitops.h should ever be used in userspace.
Quoted 3 lines
Which means that
asm-s390/posix_types.hprobably ought never to be trying to includeasm/bitops.hin the !__KERNEL__ case... we never had libc5 on S390 anyway, did we?
Correct. And the depedency to bitops.h in case of __KERNEL__ = 1 should not be there either, in particular since __FD_SET/__FD_CLR do not have to be atomic. A simple piece of C code should be used instead of set_bit/clear_bit.
Quoted 16 lines
Martin, is it OK for me to add this to the hdrcleanup-2.6.git tree?
include/asm-s390/posix_types.h +1
-1@@ -76,7 +76,7 @@ #endif /* !defined } __kernel_fsid_t; -#if defined(__KERNEL__) || !defined(__GLIBC__) || (__GLIBC__ <2 ) +#ifdef __KERNEL__#ifndef _S390_BITOPS_H#include <asm/bitops.h>
include/asm-s390/posix_types.h #define __FD_ISSET(fd,fdsetp) test_bit( · line 99
Quoted 7 lines
include/asm-s390/posix_types.h +1
-1@@ -94,6 +94,6 @@ #define __FD_ISSET(fd,fdsetp) test_bit(#undef __FD_ZERO#define __FD_ZERO (fdsetp) (memset ((fdsetp), 0, sizeof(*(fd_set *)(fdsetp)))) -#endif /* defined(__KERNEL__) || !defined(__GLIBC__) || (__GLIBC__ < 2)*/ +#endif /* __KERNEL__ */ #endif
It should not hurt. s390 always used a glibc with a major >= 2.
Signature · 8 lines
blue skies, Martin. Martin Schwidefsky Linux for zSeries Development & Services IBM Deutschland Entwicklung GmbH "Reality continues to ruin my life." - Calvin.