Support kernel syscall trace injection in the drmemtrace scheduler #7157
Assignees
Labels
Comments
abhinav92003
added a commit
that referenced
this issue
Dec 18, 2024
Adds handling for statically injected kernel syscall traces in the scheduler. Ensures that quantum context switches are not done in the middle of a statically-injected syscall trace. Also ensures that voluntary context switches are also delayed until after the syscall trace. This involved fixing the bookkeeping logic which is done on the next user-space instr. We keep status quo on the scheduler behavior of showing the post-syscall markers before the switch. Adds a unit test for statically injected kernel syscall trace handling by the scheduler. Issue: #7157
abhinav92003
added a commit
that referenced
this issue
Dec 19, 2024
…ler (#7158) Adds handling for statically-injected kernel syscall traces in the scheduler. Ensures that quantum and voluntary context switches are delayed until after the statically-injected syscall trace. This involved fixing the bookkeeping logic which is done on the next user-space instr now. Note that the injected kernel syscall traces do not include any scheduling-related markers such as TRACE_MARKER_TYPE_SYSCALL_UNSCHEDULE and TRACE_MARKER_TYPE_TIMESTAMP. For now we keep status quo on the scheduler behavior of showing the post-syscall markers before the switch. Adds a unit test for static-injected kernel syscall trace handling by the scheduler for various scenarios involving syscall sequences shorter and longer than the quantum, for system calls that do and do not cause context switches, and occurring at different offsets into the quantum. Issue: #7157
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
There are various cases where the injected syscall traces (both statically and dynamically) need to be accounted for by the drmemtrace scheduler. E.g., there shouldn't be a context switch in the middle of the syscall trace, syscall traces shouldn't break book-keeping done by the scheduler.
This issue is to explore and handle those cases.
The text was updated successfully, but these errors were encountered: