linux-kernel - [PATCH] interconnect:Replace mutex with rt_mutex for icc_bw

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date: Tue, 20 Feb 2024 20:24:47 +0800
From: Rumeng Wang <wangrumeng@...omi.corp-partner.google.com>
To: djakov@...nel.org
Cc: fengqi@...omi.com,
	linux-kernel@...r.kernel.org,
	linux-pm@...r.kernel.org,
	tkjos@...gle.com,
	wangrumeng@...omi.corp-partner.google.com,
	xuyingfeng@...omi.com
Subject: [PATCH] interconnect:Replace mutex with rt_mutex for icc_bw_lock

Thank you for your reply.
The modification you mentioned is a modification made by Qualcomm in 2022 to address the issue of icc_set_bw, which involves changing the icc_lock from mutex to rt_mutex.
However, this modification has not been introduced into the new version as it has been upgraded over the years. Therefore, there is still an issue with icc_set_bw on Android U.
My modification this time is to change the icc_bw_lock from mutex to rt_mutex, only modifying the icc_bw_lock without modifying the icc_lock, in order to reduce the scope of impact.
Here are the reasons for the modification:
We execute test scripts on Xiaomi phones. Each test will have 5 rounds, and each round of testing will involve continuously entering and exiting 30 applications. Every time an application enters or exits, it will grab a trace to count the frame loss situation.
We tested it four times using the Google solution, which is icc_set_bw use mutex.
Every time the test is conducted, there will be frame loss caused by the icc_set_bw D state issue.
The recurrence probability of the iccset_bw D state problem is 5/600
We tested it three times using the Xiaomi solution, which is icc_set_bw use rt_mutex.
Every time there is a test, even if there is a frame loss, it is not caused by the icc_set_bw D state issue.
The recurrence probability of the iccset_bw D state problem is 0/450
Of course, the above tests still have their limitations, as they only tested the application in and out scenarios, without testing any other scenarios. However, based on the current results, modifying mutex to rt_mutex has a significant optimization effect on frame loss.