* priority inversion regarding workqueues?
@ 2014-06-18 11:51 Alexander Stein
0 siblings, 0 replies; only message in thread
From: Alexander Stein @ 2014-06-18 11:51 UTC (permalink / raw)
To: linux-rt-users
Hello,
i think I noticed a priority inversion regarding workqueues. I have a high-
prio task (SCHED_FIFO prio 69) that reads a file in sysfs which actually
starts a SPI transfer (4 Bytes). I already boosted (manually) the IRQ thread
to SCHED_FIFO prio 96 and also the workqueue (imx35-cspi.0) attached to the
SPI driver (spi-imx). The execution time of my task is usually about 600us,
but with that SPI it sometimes (well rather often though) it is of about 2ms
or even mugh higher.
When I remove the SPI transfer from the driver behing the sysfs file the
execution time is rather stable, so the problem lies in that SPI transfer.
After trying some debug output I came to the conclusion i have change the
priority of [kworker/u:1] to SCHED_FIFO prio 97 in my case. I think this
should be done automatically. To be held: I do not know what [kworker/u:1]
actually is. There seem to be a task for that workqueue already, so what is
this kworker? BTW: This is a v3.4.33-rt47 kernel.
As the some low priority tasks can delay my high-prio task waiting for SPI
transfer completion this seems like priority inversion to me.
Any opinions?
Best regards,
Alexander
PS: I tried a 3.10.39-rt40 kernel and the problem is still the same, just the
kworker task names have changed.
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2014-06-18 11:53 UTC | newest]
Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2014-06-18 11:51 priority inversion regarding workqueues? Alexander Stein
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).