On 02/12/2019 01:36 PM, Waiman Long wrote: > On 02/12/2019 08:25 AM, Peter Zijlstra wrote: >> On Tue, Feb 12, 2019 at 02:24:04PM +0100, Peter Zijlstra wrote: >>> On Mon, Feb 11, 2019 at 02:31:26PM -0500, Waiman Long wrote: >>>> Modify __down_read_trylock() to make it generate slightly better code >>>> (smaller and maybe a tiny bit faster). >>>> >>>> Before this patch, down_read_trylock: >>>> >>>> 0x0000000000000000 <+0>: callq 0x5 <down_read_trylock+5> >>>> 0x0000000000000005 <+5>: jmp 0x18 <down_read_trylock+24> >>>> 0x0000000000000007 <+7>: lea 0x1(%rdx),%rcx >>>> 0x000000000000000b <+11>: mov %rdx,%rax >>>> 0x000000000000000e <+14>: lock cmpxchg %rcx,(%rdi) >>>> 0x0000000000000013 <+19>: cmp %rax,%rdx >>>> 0x0000000000000016 <+22>: je 0x23 <down_read_trylock+35> >>>> 0x0000000000000018 <+24>: mov (%rdi),%rdx >>>> 0x000000000000001b <+27>: test %rdx,%rdx >>>> 0x000000000000001e <+30>: jns 0x7 <down_read_trylock+7> >>>> 0x0000000000000020 <+32>: xor %eax,%eax >>>> 0x0000000000000022 <+34>: retq >>>> 0x0000000000000023 <+35>: mov %gs:0x0,%rax >>>> 0x000000000000002c <+44>: or $0x3,%rax >>>> 0x0000000000000030 <+48>: mov %rax,0x20(%rdi) >>>> 0x0000000000000034 <+52>: mov $0x1,%eax >>>> 0x0000000000000039 <+57>: retq >>>> >>>> After patch, down_read_trylock: >>>> >>>> 0x0000000000000000 <+0>: callq 0x5 <down_read_trylock+5> >>>> 0x0000000000000005 <+5>: mov (%rdi),%rax >>>> 0x0000000000000008 <+8>: test %rax,%rax >>>> 0x000000000000000b <+11>: js 0x2f <down_read_trylock+47> >>>> 0x000000000000000d <+13>: lea 0x1(%rax),%rdx >>>> 0x0000000000000011 <+17>: lock cmpxchg %rdx,(%rdi) >>>> 0x0000000000000016 <+22>: jne 0x8 <down_read_trylock+8> >>>> 0x0000000000000018 <+24>: mov %gs:0x0,%rax >>>> 0x0000000000000021 <+33>: or $0x3,%rax >>>> 0x0000000000000025 <+37>: mov %rax,0x20(%rdi) >>>> 0x0000000000000029 <+41>: mov $0x1,%eax >>>> 0x000000000000002e <+46>: retq >>>> 0x000000000000002f <+47>: xor %eax,%eax >>>> 0x0000000000000031 <+49>: retq >>>> >>>> By using a rwsem microbenchmark, the down_read_trylock() rate on a >>>> x86-64 system before and after the patch were: >>>> >>>> Before Patch After Patch >>>> # of Threads rlock rlock >>>> ------------ ----- ----- >>>> 1 27,787 28,259 >>>> 2 8,359 9,234 >>> From 1/2: >>> >>> 1 29,201 30,143 29,458 28,615 30,172 29,201 >>> 2 6,807 13,299 1,171 7,725 15,025 1,804 >> Argh, fat fingered and send before I was done typing. >> >> What I wanted to say was; those rlock numbers don't match up. What >> gives? >> >> The before _this_ patch number of 27k787 should be the same as the after >> first patch number of 30k172. > The rlock number in patch 1 refers to down_read() which uses xadd. The > number here in patch 2 refers specifically to down_read_trylock() which > uses cmpxchg() as this patch changes only __down_read_tryulock(). So the > performance data differ. You can see that the performance is worse if we use cmpxchg for down_read instead of using xadd. Cheers, Longman
