| Message ID | 1519657500-15094-1-git-send-email-will.deacon@arm.com |
|---|---|
| Headers | show |
| Series | Rewrite asm-generic/bitops/{atomic,lock}.h and use on arm64 | expand |
2018-02-27 0:04 GMT+09:00 Will Deacon <will.deacon@arm.com>: > Hi everyone, > > This is version two of the RFC I previously posted here: > > https://www.spinics.net/lists/arm-kernel/msg634719.html > > Changes since v1 include: > > * Fixed __clear_bit_unlock to work on archs with lock-based atomics > * Moved lock ops into bitops/lock.h > * Fixed build breakage on lesser-spotted architectures > > Trying to fix the circular #includes introduced by pulling atomic.h > into btops/lock.h has been driving me insane. I've ended up moving some > basic BIT definitions into bits.h, but this might all be better in > const.h which is being proposed by Masahiro. Feedback is especially > welcome on this part. Info for reviewers: You can see my patches at the following: 1/5: https://patchwork.kernel.org/patch/10235457/ 2/5: https://patchwork.kernel.org/patch/10235461/ 3/5: https://patchwork.kernel.org/patch/10235463/ 4/5: https://patchwork.kernel.org/patch/10235469/ 5/5: https://patchwork.kernel.org/patch/10235471/ 5/5 has conflict with Will's 2/12. Fortunately, it is at the tail of the series. It is easy to pick/drop/change when we decide how to organize it. > I've not bothered optimising for the case of a 64-bit, big-endian > architecture that uses the generic implementation of atomic64_t because > it's both messy and hypothetical. The code here should still work > correctly for that case, it just sucks (as does the implementation > currently in mainline). > > Cheers, > > Will > > --->8 > > Will Deacon (12): > h8300: Don't include linux/kernel.h in asm/atomic.h > m68k: Don't use asm-generic/bitops/lock.h > asm-generic: Move some macros from linux/bitops.h to a new bits.h file > openrisc: Don't pull in all of linux/bitops.h in asm/cmpxchg.h > sh: Don't pull in all of linux/bitops.h in asm/cmpxchg-xchg.h > arm64: fpsimd: include <linux/init.h> in fpsimd.h > arm64: lse: Include compiler_types.h and export.h for out-of-line > LL/SC > arm64: cmpxchg: Include build_bug.h instead of bug.h for BUILD_BUG > asm-generic/bitops/atomic.h: Rewrite using atomic_fetch_* > asm-generic/bitops/lock.h: Rewrite using atomic_fetch_* > arm64: Replace our atomic/lock bitop implementations with asm-generic > arm64: bitops: Include <asm-generic/bitops/ext2-atomic-setbit.h> > > arch/arm64/include/asm/bitops.h | 21 +--- > arch/arm64/include/asm/cmpxchg.h | 2 +- > arch/arm64/include/asm/fpsimd.h | 1 + > arch/arm64/include/asm/lse.h | 2 + > arch/arm64/lib/Makefile | 2 +- > arch/arm64/lib/bitops.S | 76 --------------- > arch/h8300/include/asm/atomic.h | 4 +- > arch/m68k/include/asm/bitops.h | 6 +- > arch/openrisc/include/asm/cmpxchg.h | 3 +- > arch/sh/include/asm/cmpxchg-xchg.h | 3 +- > include/asm-generic/bitops/atomic.h | 188 +++++++----------------------------- > include/asm-generic/bitops/lock.h | 68 ++++++++++--- > include/asm-generic/bits.h | 26 +++++ > include/linux/bitops.h | 22 +---- > 14 files changed, 135 insertions(+), 289 deletions(-) > delete mode 100644 arch/arm64/lib/bitops.S > create mode 100644 include/asm-generic/bits.h > > -- > 2.1.4 > -- Best Regards Masahiro Yamada
Hi Will, 2018-03-01 16:16 GMT+09:00 Masahiro Yamada <yamada.masahiro@socionext.com>: > 2018-02-27 0:04 GMT+09:00 Will Deacon <will.deacon@arm.com>: >> Hi everyone, >> >> This is version two of the RFC I previously posted here: >> >> https://www.spinics.net/lists/arm-kernel/msg634719.html >> >> Changes since v1 include: >> >> * Fixed __clear_bit_unlock to work on archs with lock-based atomics >> * Moved lock ops into bitops/lock.h >> * Fixed build breakage on lesser-spotted architectures >> >> Trying to fix the circular #includes introduced by pulling atomic.h >> into btops/lock.h has been driving me insane. I've ended up moving some >> basic BIT definitions into bits.h, but this might all be better in >> const.h which is being proposed by Masahiro. Feedback is especially >> welcome on this part. > > > Info for reviewers: > > You can see my patches at the following: > > 1/5: https://patchwork.kernel.org/patch/10235457/ > 2/5: https://patchwork.kernel.org/patch/10235461/ > 3/5: https://patchwork.kernel.org/patch/10235463/ > 4/5: https://patchwork.kernel.org/patch/10235469/ > 5/5: https://patchwork.kernel.org/patch/10235471/ > > > 5/5 has conflict with Will's 2/12. > > Fortunately, it is at the tail of the series. > It is easy to pick/drop/change > when we decide how to organize it. No comments so far about this part. I think your approach is better since putting BIT* macros into a single header is more consistent. So, I will ask Andrew to drop mine. However, I think <linux/bits.h> will make more sense than <asm-generic/bits.h> These macros are really arch-agnostic. So, we would not expect to have <asm/bits.h> that could fall back to <asm-generic/bits.h>, right? -- Best Regards Masahiro Yamada
Hi Masahiro, On Mon, Mar 12, 2018 at 12:56:28PM +0900, Masahiro Yamada wrote: > 2018-03-01 16:16 GMT+09:00 Masahiro Yamada <yamada.masahiro@socionext.com>: > > 2018-02-27 0:04 GMT+09:00 Will Deacon <will.deacon@arm.com>: > >> Hi everyone, > >> > >> This is version two of the RFC I previously posted here: > >> > >> https://www.spinics.net/lists/arm-kernel/msg634719.html > >> > >> Changes since v1 include: > >> > >> * Fixed __clear_bit_unlock to work on archs with lock-based atomics > >> * Moved lock ops into bitops/lock.h > >> * Fixed build breakage on lesser-spotted architectures > >> > >> Trying to fix the circular #includes introduced by pulling atomic.h > >> into btops/lock.h has been driving me insane. I've ended up moving some > >> basic BIT definitions into bits.h, but this might all be better in > >> const.h which is being proposed by Masahiro. Feedback is especially > >> welcome on this part. > > > > > > Info for reviewers: > > > > You can see my patches at the following: > > > > 1/5: https://patchwork.kernel.org/patch/10235457/ > > 2/5: https://patchwork.kernel.org/patch/10235461/ > > 3/5: https://patchwork.kernel.org/patch/10235463/ > > 4/5: https://patchwork.kernel.org/patch/10235469/ > > 5/5: https://patchwork.kernel.org/patch/10235471/ > > > > > > 5/5 has conflict with Will's 2/12. > > > > Fortunately, it is at the tail of the series. > > It is easy to pick/drop/change > > when we decide how to organize it. > > > No comments so far about this part. > > I think your approach is better > since putting BIT* macros into a single header > is more consistent. > > So, I will ask Andrew to drop mine. Thanks. > However, I think <linux/bits.h> will make more sense > than <asm-generic/bits.h> > > These macros are really arch-agnostic. > So, we would not expect to have <asm/bits.h> > that could fall back to <asm-generic/bits.h>, right? That's fair. I'll do a respin using linux/*. Cheers, Will
Hi everyone, This is version two of the RFC I previously posted here: https://www.spinics.net/lists/arm-kernel/msg634719.html Changes since v1 include: * Fixed __clear_bit_unlock to work on archs with lock-based atomics * Moved lock ops into bitops/lock.h * Fixed build breakage on lesser-spotted architectures Trying to fix the circular #includes introduced by pulling atomic.h into btops/lock.h has been driving me insane. I've ended up moving some basic BIT definitions into bits.h, but this might all be better in const.h which is being proposed by Masahiro. Feedback is especially welcome on this part. I've not bothered optimising for the case of a 64-bit, big-endian architecture that uses the generic implementation of atomic64_t because it's both messy and hypothetical. The code here should still work correctly for that case, it just sucks (as does the implementation currently in mainline). Cheers, Will --->8 Will Deacon (12): h8300: Don't include linux/kernel.h in asm/atomic.h m68k: Don't use asm-generic/bitops/lock.h asm-generic: Move some macros from linux/bitops.h to a new bits.h file openrisc: Don't pull in all of linux/bitops.h in asm/cmpxchg.h sh: Don't pull in all of linux/bitops.h in asm/cmpxchg-xchg.h arm64: fpsimd: include <linux/init.h> in fpsimd.h arm64: lse: Include compiler_types.h and export.h for out-of-line LL/SC arm64: cmpxchg: Include build_bug.h instead of bug.h for BUILD_BUG asm-generic/bitops/atomic.h: Rewrite using atomic_fetch_* asm-generic/bitops/lock.h: Rewrite using atomic_fetch_* arm64: Replace our atomic/lock bitop implementations with asm-generic arm64: bitops: Include <asm-generic/bitops/ext2-atomic-setbit.h> arch/arm64/include/asm/bitops.h | 21 +--- arch/arm64/include/asm/cmpxchg.h | 2 +- arch/arm64/include/asm/fpsimd.h | 1 + arch/arm64/include/asm/lse.h | 2 + arch/arm64/lib/Makefile | 2 +- arch/arm64/lib/bitops.S | 76 --------------- arch/h8300/include/asm/atomic.h | 4 +- arch/m68k/include/asm/bitops.h | 6 +- arch/openrisc/include/asm/cmpxchg.h | 3 +- arch/sh/include/asm/cmpxchg-xchg.h | 3 +- include/asm-generic/bitops/atomic.h | 188 +++++++----------------------------- include/asm-generic/bitops/lock.h | 68 ++++++++++--- include/asm-generic/bits.h | 26 +++++ include/linux/bitops.h | 22 +---- 14 files changed, 135 insertions(+), 289 deletions(-) delete mode 100644 arch/arm64/lib/bitops.S create mode 100644 include/asm-generic/bits.h -- 2.1.4