Buildresult: linus/um-defconfig/um-x86_64 built on Dec 30 2016, 10:47

kisskb

Revisions | Branches | Compilers | Configs | Build Results | Build Failures |

Status:	Failed
Date/Time:	Dec 30 2016, 10:47
Duration:	0:00:07.598920
Builder:	ka3
Revision:	mm: optimize PageWaiters bit use for unlock_page() (`b91e1302ad9b80c174a4855533f7e3aa2873355e)`
Target:	linus/um-defconfig/um-x86_64
Branch:	linus
Compiler:	um-x86_64 (x86_64-linux-gcc (GCC) 4.6.3)
Config:	defconfig (download)
Log:	Download original Possible errors /opt/cross/gcc-4.6.3-nolibc/x86_64-linux/lib/gcc/x86_64-linux/4.6.3/../../../../x86_64-linux/include/sys/cdefs.h:385:27: fatal error: bits/wordsize.h: No such file or directory make[2]: * [arch/x86/um/user-offsets.s] Error 1 make[1]: * [arch/x86/um/user-offsets.s] Error 2 make: * [sub-make] Error 2 Possible warnings (2) kernel/time/Kconfig:155:warning: range is invalid kernel/time/Kconfig:155:warning: range is invalid Full Log # git rev-parse -q --verify b91e1302ad9b80c174a4855533f7e3aa2873355e^{commit} b91e1302ad9b80c174a4855533f7e3aa2873355e already have revision, skipping fetch # git checkout -q -f -B kisskb b91e1302ad9b80c174a4855533f7e3aa2873355e # git clean -qxdf # git log -1 commit b91e1302ad9b80c174a4855533f7e3aa2873355e Author: Linus Torvalds <torvalds@linux-foundation.org> Date: Tue Dec 27 11:40:38 2016 -0800 mm: optimize PageWaiters bit use for unlock_page() In commit 62906027091f ("mm: add PageWaiters indicating tasks are waiting for a page bit") Nick Piggin made our page locking no longer unconditionally touch the hashed page waitqueue, which not only helps performance in general, but is particularly helpful on NUMA machines where the hashed wait queues can bounce around a lot. However, the "clear lock bit atomically and then test the waiters bit" sequence turns out to be much more expensive than it needs to be, because you get a nasty stall when trying to access the same word that just got updated atomically. On architectures where locking is done with LL/SC, this would be trivial to fix with a new primitive that clears one bit and tests another atomically, but that ends up not working on x86, where the only atomic operations that return the result end up being cmpxchg and xadd. The atomic bit operations return the old value of the same bit we changed, not the value of an unrelated bit. On x86, we could put the lock bit in the high bit of the byte, and use "xadd" with that bit (where the overflow ends up not touching other bits), and look at the other bits of the result. However, an even simpler model is to just use a regular atomic "and" to clear the lock bit, and then the sign bit in eflags will indicate the resulting state of the unrelated bit #7. So by moving the PageWaiters bit up to bit #7, we can atomically clear the lock bit and test the waiters bit on x86 too. And architectures with LL/SC (which is all the usual RISC suspects), the particular bit doesn't matter, so they are fine with this approach too. This avoids the extra access to the same atomic word, and thus avoids the costly stall at page unlock time. The only downside is that the interface ends up being a bit odd and specialized: clear a bit in a byte, and test the sign bit. Nick doesn't love the resulting name of the new primitive, but I'd rather make the name be descriptive and very clear about the limitation imposed by trying to work across all relevant architectures than make it be some generic thing that doesn't make the odd semantics explicit. So this introduces the new architecture primitive clear_bit_unlock_is_negative_byte(); and adds the trivial implementation for x86. We have a generic non-optimized fallback (that just does a "clear_bit()"+"test_bit(7)" combination) which can be overridden by any architecture that can do better. According to Nick, Power has the same hickup x86 has, for example, but some other architectures may not even care. All these optimizations mean that my page locking stress-test (which is just executing a lot of small short-lived shell scripts: "make test" in the git source tree) no longer makes our page locking look horribly bad. Before all these optimizations, just the unlock_page() costs were just over 3% of all CPU overhead on "make test". After this, it's down to 0.66%, so just a quarter of the cost it used to be. (The difference on NUMA is bigger, but there this micro-optimization is likely less noticeable, since the big issue on NUMA was not the accesses to 'struct page', but the waitqueue accesses that were already removed by Nick's earlier commit). Acked-by: Nick Piggin <npiggin@gmail.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Bob Peterson <rpeterso@redhat.com> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Andrew Lutomirski <luto@kernel.org> Cc: Andreas Gruenbacher <agruenba@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> # < /opt/cross/kisskb/gcc-4.6.3-nolibc/um-x86_64-linux/bin/x86_64-linux-gcc --version # < git log --format=%s --max-count=1 b91e1302ad9b80c174a4855533f7e3aa2873355e # < make -j 10 ARCH=um O=/home/kisskb/slave/build/linus_um-defconfig_um-x86_64 CROSS_COMPILE=/opt/cross/kisskb/gcc-4.6.3-nolibc/um-x86_64-linux/bin/x86_64-linux- SUBARCH=x86_64 defconfig kernel/time/Kconfig:155:warning: range is invalid # make -j 10 ARCH=um O=/home/kisskb/slave/build/linus_um-defconfig_um-x86_64 CROSS_COMPILE=/opt/cross/kisskb/gcc-4.6.3-nolibc/um-x86_64-linux/bin/x86_64-linux- SUBARCH=x86_64 make[1]: Entering directory '/home/kisskb/slave/build/linus_um-defconfig_um-x86_64' GEN ./Makefile scripts/kconfig/conf --silentoldconfig arch/x86/um/Kconfig kernel/time/Kconfig:155:warning: range is invalid CHK include/generated/uapi/linux/version.h UPD include/generated/uapi/linux/version.h HOSTCC scripts/basic/bin2c WRAP arch/um/include/generated/asm/barrier.h WRAP arch/um/include/generated/asm/bug.h WRAP arch/um/include/generated/asm/clkdev.h WRAP arch/um/include/generated/asm/cputime.h WRAP arch/um/include/generated/asm/delay.h WRAP arch/um/include/generated/asm/device.h WRAP arch/um/include/generated/asm/current.h WRAP arch/um/include/generated/asm/emergency-restart.h WRAP arch/um/include/generated/asm/exec.h WRAP arch/um/include/generated/asm/ftrace.h WRAP arch/um/include/generated/asm/futex.h WRAP arch/um/include/generated/asm/hardirq.h WRAP arch/um/include/generated/asm/hw_irq.h WRAP arch/um/include/generated/asm/io.h WRAP arch/um/include/generated/asm/irq_regs.h WRAP arch/um/include/generated/asm/irq_work.h WRAP arch/um/include/generated/asm/kdebug.h WRAP arch/um/include/generated/asm/mcs_spinlock.h WRAP arch/um/include/generated/asm/mm-arch-hooks.h WRAP arch/um/include/generated/asm/param.h WRAP arch/um/include/generated/asm/pci.h WRAP arch/um/include/generated/asm/percpu.h WRAP arch/um/include/generated/asm/preempt.h WRAP arch/um/include/generated/asm/switch_to.h WRAP arch/um/include/generated/asm/topology.h WRAP arch/um/include/generated/asm/trace_clock.h WRAP arch/um/include/generated/asm/word-at-a-time.h WRAP arch/um/include/generated/asm/xor.h CHK include/config/kernel.release GEN ./Makefile HOSTCC scripts/kallsyms CC scripts/mod/empty.o HOSTCC scripts/mod/mk_elfconfig CC scripts/mod/devicetable-offsets.s GEN scripts/mod/devicetable-offsets.h MKELF scripts/mod/elfconfig.h HOSTCC scripts/mod/modpost.o HOSTCC scripts/mod/file2alias.o HOSTCC scripts/mod/sumversion.o UPD include/config/kernel.release SYSTBL arch/x86/entry/syscalls/../../include/generated/asm/syscalls_32.h SYSHDR arch/x86/entry/syscalls/../../include/generated/asm/unistd_32_ia32.h SYSHDR arch/x86/entry/syscalls/../../include/generated/asm/unistd_64_x32.h SYSTBL arch/x86/entry/syscalls/../../include/generated/asm/syscalls_64.h SYSHDR arch/x86/entry/syscalls/../../include/generated/uapi/asm/unistd_32.h SYSHDR arch/x86/entry/syscalls/../../include/generated/uapi/asm/unistd_64.h SYSHDR arch/x86/entry/syscalls/../../include/generated/uapi/asm/unistd_x32.h Using /home/kisskb/slave/src as source for kernel CHK include/generated/utsrelease.h UPD include/generated/utsrelease.h HOSTLD scripts/mod/modpost HOSTCC scripts/unifdef CC arch/x86/um/user-offsets.s In file included from /usr/include/features.h:374:0, from /usr/include/stdio.h:27, from /home/kisskb/slave/src/arch/x86/um/user-offsets.c:1: /opt/cross/gcc-4.6.3-nolibc/x86_64-linux/lib/gcc/x86_64-linux/4.6.3/../../../../x86_64-linux/include/sys/cdefs.h:385:27: fatal error: bits/wordsize.h: No such file or directory compilation terminated. /home/kisskb/slave/src/scripts/Makefile.build:154: recipe for target 'arch/x86/um/user-offsets.s' failed make[2]: * [arch/x86/um/user-offsets.s] Error 1 arch/um/Makefile:150: recipe for target 'arch/x86/um/user-offsets.s' failed make[1]: * [arch/x86/um/user-offsets.s] Error 2 make[1]: Leaving directory '/home/kisskb/slave/build/linus_um-defconfig_um-x86_64' Makefile:150: recipe for target 'sub-make' failed make: * [sub-make] Error 2 Command 'make -j 10 ARCH=um O=/home/kisskb/slave/build/linus_um-defconfig_um-x86_64 CROSS_COMPILE=/opt/cross/kisskb/gcc-4.6.3-nolibc/um-x86_64-linux/bin/x86_64-linux- SUBARCH=x86_64 ' returned non-zero exit status 2 # rm -rf /home/kisskb/slave/build/linus_um-defconfig_um-x86_64 # Build took: 0:00:07.598920

© Michael Ellerman 2006-2018.