[Deepin-Kernel-SIG] [linux 6.6-y] [Openeuler] cgroup: disable kernel memory accounting for all memory cgroups by de… #967
Conversation
…fault hulk inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I8QLND CVE: NA ---------------------------------------- The kernel memory accounting for all memory cgroups is not stable, and it will cause a 100% regression in hackbench compared with kernel-4.19, so disable it by default. We can use the following command line to enable or disable it: cgroup.memory=kmem or cgroup.memory=nokmem. Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Signed-off-by: Zheng Zengkai <zhengzengkai@huawei.com> Signed-off-by: chenridong <chenridong@huawei.com> Signed-off-by: Wentao Guan <guanwentao@uniontech.com>
Reviewer's GuideDisable kernel memory accounting for all memory cgroups by default and introduce a "kmem" kernel parameter to re-enable it, with corresponding documentation updates. Class diagram for cgroup memory kernel parameter handlingclassDiagram
class memcontrol.c {
+bool cgroup_memory_nokmem
+int __init cgroup_memory(char *s)
}
memcontrol.c : cgroup_memory_nokmem = true (default)
memcontrol.c : cgroup_memory(char *s) parses "kmem" and "nokmem" tokens
Flow diagram for cgroup.memory kernel parameter parsingflowchart TD
A[Boot with kernel parameters] --> B{cgroup.memory token}
B -->|kmem| C[Set cgroup_memory_nokmem = false]
B -->|nokmem| D[Set cgroup_memory_nokmem = true]
B -->|other| E[No change to cgroup_memory_nokmem]
File-Level Changes
Tips and commandsInteracting with Sourcery
Customizing Your ExperienceAccess your dashboard to:
Getting Help
|
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here.
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
deepin pr auto review代码审查意见：
综上所述，代码的更改看起来是为了调整内核内存计费的行为，但是需要确保这些更改与文档和代码的其他部分保持一致，并且进行了充分的测试以确保没有引入新的问题。 |
|
Test result: After patch:Benchmark Run: 二 7月 22 2025 18:35:19 - 18:42:02 Dhrystone 2 using register variables 50758627.6 lps (10.0 s, 1 samples) System Benchmarks Index Values BASELINE RESULT INDEX Benchmark Run: 二 7月 22 2025 18:42:02 - 18:48:45 Dhrystone 2 using register variables 207921797.3 lps (10.0 s, 1 samples) System Benchmarks Index Values BASELINE RESULT INDEX before: Dhrystone 2 using register variables 49626251.2 lps (10.0 s, 1 samples) System Benchmarks Index Values BASELINE RESULT INDEX Benchmark Run: 日 7月 20 2025 15:44:59 - 15:51:42 Dhrystone 2 using register variables 207921608.7 lps (10.0 s, 1 samples) System Benchmarks Index Values BASELINE RESULT INDEX |
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
There was a problem hiding this comment.
Hey @opsiff - I've reviewed your changes - here's some feedback:
- Consider using the kernel’s match_token helper instead of manual strcmp loops for parsing the cgroup.memory tokens to make the code more maintainable and extensible.
- Add explicit conflict detection or precedence handling when both “kmem” and “nokmem” are passed so users aren’t left with ambiguous behavior.
- Double-check that cgroup v2’s memory controller either matches this default-off kmem behavior or provides its own symmetric option to avoid inconsistent defaults across cgroup versions.
Prompt for AI Agents
Please address the comments from this code review:
## Overall Comments
- Consider using the kernel’s match_token helper instead of manual strcmp loops for parsing the cgroup.memory tokens to make the code more maintainable and extensible.
- Add explicit conflict detection or precedence handling when both “kmem” and “nokmem” are passed so users aren’t left with ambiguous behavior.
- Double-check that cgroup v2’s memory controller either matches this default-off kmem behavior or provides its own symmetric option to avoid inconsistent defaults across cgroup versions.Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.
|
Tldr: execl from 5356 to 5573(+4%) |
There was a problem hiding this comment.
Pull Request Overview
This PR disables kernel memory accounting for all memory cgroups by default to address performance regression issues, specifically a 100% regression in hackbench compared to kernel-4.19. The change introduces a command-line option to re-enable the feature when needed.
- Changes the default state of kernel memory accounting from enabled to disabled
- Adds a new "kmem" command-line parameter to explicitly enable kernel memory accounting
- Updates documentation to reflect the new default behavior and available options
Reviewed Changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| mm/memcontrol.c | Changes default value of cgroup_memory_nokmem to true and adds kmem parameter handling |
| Documentation/admin-guide/kernel-parameters.txt | Documents the new kmem command-line option |
| Documentation/admin-guide/cgroup-v1/memory.rst | Updates memory cgroup documentation to reflect disabled-by-default behavior |
| @@ -7618,6 +7618,8 @@ static int __init cgroup_memory(char *s) | |||
| cgroup_memory_nosocket = true; | |||
| if (!strcmp(token, "nokmem")) | |||
| cgroup_memory_nokmem = true; | |||
| else if (!strcmp(token, "kmem")) | |||
There was a problem hiding this comment.
The else-if condition should be changed to a separate if statement for consistency with the existing pattern. The current code has separate if statements for 'nosocket', 'nokmem', and 'nobpf' tokens, but uses else-if for 'kmem', which creates inconsistent structure.
| else if (!strcmp(token, "kmem")) | |
| if (!strcmp(token, "kmem")) |
Copilot uses AI. Check for mistakes.
| at boot time. In this case, kernel memory will not be accounted at all. | ||
| Kernel memory accounting is disabled for all memory cgroups by default. But | ||
| it can be enabled system-wide by passing cgroup.memory=kmem to the kernel | ||
| at boot time. In this case, kernel memory will all be accounted. |
There was a problem hiding this comment.
The phrase 'kernel memory will all be accounted' is grammatically awkward. It should be 'all kernel memory will be accounted' or 'kernel memory will be accounted'.
| at boot time. In this case, kernel memory will all be accounted. | |
| at boot time. In this case, all kernel memory will be accounted. |
Copilot uses AI. Check for mistakes.
|
@winnscode 你怎么看 |
| @@ -87,7 +87,7 @@ EXPORT_PER_CPU_SYMBOL_GPL(int_active_memcg); | |||
| static bool cgroup_memory_nosocket __ro_after_init; | |||
|
|
|||
| /* Kernel memory accounting disabled? */ | |||
| static bool cgroup_memory_nokmem __ro_after_init; | |||
| static bool cgroup_memory_nokmem __ro_after_init = true; | |||
…fault
hulk inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I8QLND CVE: NA
The kernel memory accounting for all memory cgroups is not stable, and it will cause a 100% regression in hackbench compared with kernel-4.19, so disable it by default. We can use the following command line to enable or disable it:
cgroup.memory=kmem or cgroup.memory=nokmem.
Summary by Sourcery
Disable kernel memory accounting for all memory cgroups by default to avoid performance regressions and introduce a command-line option to re-enable it.
New Features:
Bug Fixes:
Documentation: