summaryrefslogtreecommitdiff
path: root/runtime/class_linker_test.cc
diff options
context:
space:
mode:
author Hans Boehm <hboehm@google.com> 2024-09-19 10:04:31 -0700
committer Hans Boehm <hboehm@google.com> 2024-09-21 00:06:03 +0000
commite23d7335d327e3ad2d1c6c88261ea22835ecb342 (patch)
tree822b10aca4a6ec5a863d15c08d639fe9eb1967c7 /runtime/class_linker_test.cc
parented67125787c5ec24285e694e56f6e4f1f7fef1be (diff)
More suspend timeout improvements
1) Increase suspend timeouts when the request comes from a low priority thread. We used to double the timeout, we now triple it. 2) Rather than including /proc/<tid>/stat prefixes in timeout messages. use /proc/<pid>/task/<tid>/stat. The former reports running times for the process, while the latter reports thread-specific numbers. (In spite of what the man page states.) 3) Improve the timeout message format to eliminate some redundant information and instead include 10 more characters from the stat files, better exposing priority-related fields. 4) We print a calculated approximation of the total wait time before the timeout. Just printing the final wait time turned out to be confusing. 5) Rather than lengthening the third out of 4 wait attempts more than the others, we now lengthen the 4th and final one instead. This means the two stat file snapshots may be further apart. This should somewhat reduce timeouts, but mostly it further improves the message we get when we do time out. Bug: 366061900 Test: Treehugger Test: Check messages with kShortSuspendTimeouts set. Change-Id: Ife42f0b9c37c4fabe82c076c66923d9592df53f7
Diffstat (limited to 'runtime/class_linker_test.cc')
0 files changed, 0 insertions, 0 deletions