Conversation

great architecting today, much success, I fixed a release blocking perf regression by

... reading the docs 🤯

team already did all the debugging and triaging and root-causing, but apparently reading the docs to figure out what prctl we need to set needs the big guns 😱

2
0
1
@sima please tell me it wasn't PR_SET_THP_DISABLE
2
1
4
@vbabka @sima good to be aware of this +1
0
0
1

@vbabka perf regression was with a very clever hack for buffer sharing in drm across processes. but they already had a new one using pidfd_getfd as an actually cleaner design, which needs PR_SET_TRACER to work with ptrace_scope = 1

apparently no one knew this prctl exists, and they tried to ship with a "pls add CAP_SYS_PTRACE" in the release notes

0
0
1