Thread (27 messages) 27 messages, 7 authors, 2021-06-17

RE: [LKP] Re: [mm/mremap] ecf8443e51: vm-scalability.throughput -29.4% regression

From: Liu, Yujie <hidden>
Date: 2021-06-17 02:38:45
Also in: linuxppc-dev, oe-lkp

-----Original Message-----
From: Aneesh Kumar K.V <redacted>
Sent: Tuesday, June 15, 2021 12:09 AM
To: Sang, Oliver <redacted>
Cc: lkp <redacted>; LKML <redacted>; lkp@lists.01.org; linux-mm@kvack.org; akpm@linux-foundation.org;
mpe@ellerman.id.au; linuxppc-dev@lists.ozlabs.org; kaleshsingh@google.com; npiggin@gmail.com; joel@joelfernandes.org; Christophe
Leroy [off-list ref]; Linus Torvalds [off-list ref]; Kirill A . Shutemov [off-list ref]
Subject: [LKP] Re: [mm/mremap] ecf8443e51: vm-scalability.throughput -29.4% regression

On 6/14/21 8:25 PM, kernel test robot wrote:
quoted

Greeting,

FYI, we noticed a -29.4% regression of vm-scalability.throughput due to commit:


commit: ecf8443e51a862b261313c2319ab4e4aed9e6b7e ("[PATCH v7 02/11]
mm/mremap: Fix race between MOVE_PUD mremap and pageout")
url:
https://github.com/0day-ci/linux/commits/Aneesh-Kumar-K-V/Speedup-mrem
ap-on-ppc64/20210607-135424
base: https://git.kernel.org/cgit/linux/kernel/git/powerpc/linux.git
next
We dropped that approach and is now using https://lore.kernel.org/linux-mm/20210610083549.386085-1-aneesh.kumar@linux.ibm.com (local)


Instead of pud lock we are now using rmap lock with mremap.

Can you check with that series?
Hi Aneesh,

Could you please specify the base commit of the patch series? 

We have applied new patch series(rmap lock with mremap) on commit 027f55e87c30 (tty: hvc: udbg_hvc: retry putc on -EAGAIN), and it shows no regression after patch.
=========================================================================================                          
compiler/cpufreq_governor/kconfig/nr_ssd/nr_task/rootfs/runtime/tbox_group/test/testcase/thp_defrag/thp_enabled/uco
de:                                                                                                                
  gcc-9/performance/x86_64-rhel-8.3/1/8/debian-10.4-x86_64-20200603.cgz/300/lkp-csl-2ap1/swap-w-seq-mt/vm-scalabili
ty/always/never/0x4003006                                                                                          
                                                                                                                   
commit:                                                                                                            
  027f55e87c30 (tty: hvc: udbg_hvc: retry putc on -EAGAIN)        <---  patch's base commit      
  ecf8443e51a8 (mm/mremap: Fix race between MOVE_PUD mremap and pageout)      <---  first bad commit  
  57da7477067d  (fixup)    <---  apply patch on 027f55e87c30
                                                                                                                   
027f55e87c309427 ecf8443e51a862b261313c2319a 57da7477067dbe29247484eda0e                                           
---------------- --------------------------- ---------------------------                                           
         %stddev     %change         %stddev     %change         %stddev                              
             \          |                \          |                \                                
    371814 ±  3%     -29.1%     263582 ±  3%      -0.1%     371286 ±  2%  vm-scalability.median       


We also tried to apply patches on commit ecf8443e51a8 (mm/mremap: Fix race between MOVE_PUD mremap and pageout), and the regression increased to  -33.9%
=========================================================================================                          
compiler/cpufreq_governor/kconfig/nr_ssd/nr_task/rootfs/runtime/tbox_group/test/testcase/thp_defrag/thp_enabled/uco
de:                                                                                                                
  gcc-9/performance/x86_64-rhel-8.3/1/8/debian-10.4-x86_64-20200603.cgz/300/lkp-csl-2ap1/swap-w-seq-mt/vm-scalabili
ty/always/never/0x4003006                                                                                          
                                                                                                                   
commit:                                                                                                            
  5f80ee2fc08b (mm/mremap: Fix race between MOVE_PUD mremap and pageout)      <--- first bad commit's parent                                                             
  ecf8443e51a8 (mm/mremap: Fix race between MOVE_PMD mremap and pageout)    <--- first bad commit                                                            
  8ae369d45894 (fixup)      <--- apply patch on ecf8443e51a8
                                                                                                                   
5f80ee2fc08b3613 ecf8443e51a862b261313c2319a 8ae369d4589492c7f7198cd742d                                           
---------------- --------------------------- ---------------------------                                           
          %stddev     %change         %stddev     %change         %stddev                                           
              \          |                \          |                \                                             
     373210           -29.4%     263582 ±  3%     -33.9%     246626 ±  2%  vm-scalability.median                    

Thanks 
-aneesh
_______________________________________________
LKP mailing list -- lkp@lists.01.org
To unsubscribe send an email to lkp-leave@lists.01.org
  
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help