These are changes to improve affinity on cyclictest
There are some fixes to the parsing of -a CPUSET that occurred in a few
exceptional instances.
Also, cyclictest was written as if it owned the whole machine. So before
these changes we would run on all cpus or on the CPUSET that the user
passed with the command line option even if those cpus were not in the
CPUSET from the affinity in the runtime environment. These changes here
fix that. For example,
Before the changes
taskset 4-7 cyclictest -t -a 1-6
the taskset 4-7 would only apply to the main cyclictest thread, and then
the 1-6 would only apply to every newly created thread.
After the changes
taskset 4-7 cyclictest -t -a 1-6
Now, 4-7 still applies to the main thread, but all newly created threads
run on the interesection of the cpus 4-7 and 1-6, in other words on 4,5
and 6
In addition if the user doesn't pass and affinity with the command line
option, the runtime environment is respected.
For example
Before
taskset 0-4 cyclictest -t
Only the main cyclictest thread would run on cpus 0-4 and the other
threads could run on any cpu
After
taskset 0-4 cyclictest -t
All cyclictest threads run on 0-4
John Kacur (6):
rt-tests: cyclictest: Set errno to 0 before call to mlock
rt-tests: cyclictest: Report all errors from pthread_setaffinity_np
rt-tests: cyclictest: Fix parsing affinity with a space and a leading
zero
rt-tests: cyclictest: Fix parsing affinity with leading exclamation
mark
rt-tests: cyclictest: Only run on available cpus according to the
affinity
rt-tests: cyclictest: Only run on runtime affinity and user supplied
affinity
src/cyclictest/cyclictest.c | 78 +++++++++++++++++++++++++++++++++----
1 file changed, 71 insertions(+), 7 deletions(-)
--
2.20.1
