Re: [PATCH 0/2] mm: tlb swap entries batch async release
From: zhiguojiang
Date: Wed Jul 31 2024 - 00:57:08 EST
在 2024/7/31 10:18, Barry Song 写道:
[Some people who received this message don't often get email from 21cnbao@xxxxxxxxx. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
On Tue, Jul 30, 2024 at 7:44 PM Zhiguo Jiang <justinjiang@xxxxxxxx> wrote:
The main reasons for the prolonged exit of a background process is the
time-consuming release of its swap entries. The proportion of swap memory
occupied by the background process increases with its duration in the
background, and after a period of time, this value can reach 60% or more.
Do you know the reason? Could they be contending for a cluster lock or
something?
Is there any perf data or flamegraph available here?
Hi,
Testing datas of application occuping different physical memory sizes at
different time
points in the background:
Testing Platform: 8GB RAM
Testing procedure:
After booting up, start 15 applications first, and then observe the
physical memory size
occupied by the last launched application at different time points in
the background.
foreground - abbreviation FG
background - abbreviation BG
The app launched last: com.qiyi.video app
| memory type | FG 5s | BG 5s | BG 1min | BG 3min | BG 5min | BG
10min | BG 15min |
---------------------------------------------------------------------------------------
| VmRSS(KB) | 453832 | 252300 | 207724 | 206776 | 204364 |
199944 | 199748 |
| RssAnon(KB) | 247348 | 99296 | 71816 | 71484 | 71268 |
67808 | 67660 |
| RssFile(KB) | 205536 | 152020 | 134956 | 134340 | 132144 |
131184 | 131136 |
| RssShmem(KB) | 1048 | 984 | 952 | 952 | 952 |
952 | 952 |
| VmSwap(KB) | 202692 | 334852 | 362332 | 362664 | 362880 |
366340 | 366488 |
| Swap ratio(%) | 30.87% | 57.03% | 63.56% | 63.69% | 63.97% |
64.69% | 64.72% |
The app launched last: com.netease.sky.vivo
| memory type | FG 5s | BG 5s | BG 1min | BG 3min | BG 5min | BG
10min | BG 15min |
---------------------------------------------------------------------------------------
| VmRSS(KB) | 435424 | 403564 | 403200 | 401688 | 402996 |
396372 | 396268 |
| RssAnon(KB) | 151616 | 117252 | 117244 | 115888 | 117088 |
110780 | 110684 |
| RssFile(KB) | 281672 | 284192 | 283836 | 283680 | 283788 |
283472 | 283464 |
| RssShmem(KB) | 2136 | 2120 | 2120 | 2120 | 2120 |
2120 | 2120 |
| VmSwap(KB) | 546584 | 559920 | 559928 | 561284 | 560084 |
566392 | 566488 |
| Swap ratio(%) | 55.66% | 58.11% | 58.14% | 58.29% | 58.16% |
58.83% | 58.84% |
A background exiting process's perfedata:
| interfaces | cost(ms) | exe(ms) | average(ms) | run
counts |
--------------------------------------------------------------------------------
| do_signal | 791.813 | 0 | 791.813 |
1 |
| get_signal | 791.813 | 0 | 791.813 |
1 |
| do_group_exit | 791.813 | 0 | 791.813 |
1 |
| do_exit | 791.813 | 0.148 | 791.813 |
1 |
| exit_mm | 577.859 | 0 | 577.859 |
1 |
| __mmput | 577.859 | 0.202 | 577.859 |
1 |
| exit_mmap | 577.497 | 1.806 | 192.499 |
3 |
| __oom_reap_task_mm | 562.869 | 2.695 | 562.869 |
1 |
| unmap_page_range | 562.07 | 3.185 | 20.817 |
27 |
| zap_pte_range | 558.645 | 123.958 | 15.518 |
36 |
| free_swap_and_cache | 433.381 | 28.831 | 6.879 |
63 |
| free_swap_slot | 403.568 | 4.876 | 4.248 |
95 |
| swapcache_free_entries | 398.292 | 3.578 | 3.588 |
111 |
| swap_entry_free | 393.863 | 13.953 | 3.176 |
124 |
| swap_range_free | 372.602 | 202.478 | 1.791 |
208 |
| $x.204 [zram] | 132.389 | 0.341 | 0.33 |
401 |
| zram_reset_device | 131.888 | 22.376 | 0.326 |
405 |
| obj_free | 80.101 | 29.517 | 0.21 |
381 |
| zs_create_pool | 29.381 | 2.772 | 0.124 |
237 |
| clear_shadow_from_swap_cache | 22.846 | 22.686 | 0.11 |
208 |
| __put_page | 19.317 | 10.088 | 0.105 |
184 |
| pr_memcg_info | 13.038 | 1.181 | 0.11 |
118 |
| free_pcp_prepare | 9.229 | 0.812 | 0.094 |
98 |
| xxx_memcg_out | 9.223 | 4.746 | 0.098 |
94 |
| free_pgtables | 8.813 | 3.302 | 8.813 |
1 |
| zs_compact | 8.617 | 8.43 | 0.097 |
89 |
| kmem_cache_free | 7.483 | 4.595 | 0.084 |
89 |
| __mem_cgroup_uncharge_swap | 6.348 | 3.03 | 0.086 |
74 |
| $x.178 [zsmalloc] | 6.182 | 0.32 | 0.09 |
69 |
| $x.182 [zsmalloc] | 5.019 | 0.08 | 0.088 |
57 |
cost - total time consumption.
exe - total actual execution time.
According to perfdata, we can observe that the main reason for the
prolonged exit
of a background process is the time-consuming release of its swap entries.
The reason for the time-consuming release of swap entries is not only
due to cluster
locks, but also swp_slots lock and swap_info lock, additionally zram and
swapdisk free
path time-consuming .
Additionally, the relatively lengthy path for releasing swap entries
further contributes to the longer time required for the background process
to release its swap entries.
In the multiple background applications scenario, when launching a large
memory application such as a camera, system may enter a low memory state,
which will triggers the killing of multiple background processes at the
same time. Due to multiple exiting processes occupying multiple CPUs for
concurrent execution, the current foreground application's CPU resources
are tight and may cause issues such as lagging.
To solve this problem, we have introduced the multiple exiting process
asynchronous swap memory release mechanism, which isolates and caches
swap entries occupied by multiple exit processes, and hands them over
to an asynchronous kworker to complete the release. This allows the
exiting processes to complete quickly and release CPU resources. We have
validated this modification on the products and achieved the expected
benefits.
It offers several benefits:
1. Alleviate the high system cpu load caused by multiple exiting
processes running simultaneously.
2. Reduce lock competition in swap entry free path by an asynchronous
Do you have data on which lock is affected? Could it be a cluster lock?
The reason for the time-consuming release of swap entries is not only
due to cluster
locks, but also swp_slots lock and swap_info lock, additionally zram and
swapdisk free
path time-consuming . In short, swap entry release path is relatively
long compared to
file and anonymous folio release path.
kworker instead of multiple exiting processes parallel execution.
3. Release memory occupied by exiting processes more efficiently.
Zhiguo Jiang (2):
mm: move task_is_dying to h headfile
mm: tlb: multiple exiting processes's swap entries async release
include/asm-generic/tlb.h | 50 +++++++
include/linux/mm_types.h | 58 ++++++++
include/linux/oom.h | 6 +
mm/memcontrol.c | 6 -
mm/memory.c | 3 +-
mm/mmu_gather.c | 297 ++++++++++++++++++++++++++++++++++++++
6 files changed, 413 insertions(+), 7 deletions(-)
mode change 100644 => 100755 include/asm-generic/tlb.h
mode change 100644 => 100755 include/linux/mm_types.h
mode change 100644 => 100755 include/linux/oom.h
mode change 100644 => 100755 mm/memcontrol.c
mode change 100644 => 100755 mm/memory.c
mode change 100644 => 100755 mm/mmu_gather.c
Can you check your local filesystem to determine why you're running
the chmod command?
Ok, I will check it carefully.
Thanks
Zhiguo
--
2.39.0
Thanks
Barry