[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <f901e346-6203-371d-a39a-ed9231187570@suse.cz>
Date: Mon, 5 Nov 2018 21:09:24 +0100
From: Vlastimil Babka <vbabka@...e.cz>
To: Linus Torvalds <torvalds@...ux-foundation.org>,
rong.a.chen@...el.com
Cc: yang.shi@...ux.alibaba.com, kirill.shutemov@...ux.intel.com,
mhocko@...nel.org, willy@...radead.org, ldufour@...ux.vnet.ibm.com,
Colin King <colin.king@...onical.com>,
Andrew Morton <akpm@...ux-foundation.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
lkp@...org
Subject: Re: [LKP] [mm] 9bc8039e71: will-it-scale.per_thread_ops -64.1%
regression
On 11/5/18 6:50 PM, Linus Torvalds wrote:
> On Sun, Nov 4, 2018 at 9:08 PM kernel test robot <rong.a.chen@...el.com> wrote:
>>
>> FYI, we noticed a -64.1% regression of will-it-scale.per_thread_ops
>> due to commit 9bc8039e715d ("mm: brk: downgrade mmap_sem to read when
>> shrinking")
>
> Ugh. That looks pretty bad.
>
>> in testcase: will-it-scale
>> on test machine: 8 threads Ivy Bridge with 16G memory
>> with following parameters:
>>
>> nr_task: 100%
>> mode: thread
>> test: brk1
>> ucode: 0x20
>> cpufreq_governor: performance
>
> The reason seems to be way more scheduler time due to lots more
> context switches:
>
>> 34925294 ± 18% +270.3% 1.293e+08 ± 4% will-it-scale.time.voluntary_context_switches
And what about this:
0.83 ± 27% +25.9 26.75 ± 11% perf-profile.calltrace.cycles-pp.intel_idle.cpuidle_enter_state.do_idle.cpu_startup_entry.start_secondary
1.09 ± 32% +30.9 31.97 ± 10% perf-profile.calltrace.cycles-pp.cpuidle_enter_state.do_idle.cpu_startup_entry.start_secondary.secondary_startup_64
1.62 ± 36% +44.4 46.01 ± 9% perf-profile.calltrace.cycles-pp.do_idle.cpu_startup_entry.start_secondary.secondary_startup_64
1.63 ± 36% +44.5 46.18 ± 9% perf-profile.calltrace.cycles-pp.cpu_startup_entry.start_secondary.secondary_startup_64
1.63 ± 36% +44.6 46.21 ± 9% perf-profile.calltrace.cycles-pp.start_secondary.secondary_startup_64
1.73 ± 29% +51.1 52.86 ± 2% perf-profile.calltrace.cycles-pp.secondary_startup_64
And the graphs showing less user/kernel time and less "percent_of_cpu_this_job_got"...
I didn't spot an obvious mistake in the patch itself, so it looks
like some bad interaction between scheduler and the mmap downgrade?
> Yang Shi, would you mind taking a look at what's going on?
>
> Linus
>
Powered by blists - more mailing lists