On Mon, Jan 31, 2022 at 12:16:06PM +0200, Tony Lindgren wrote:
> Hi,
>
> * Drew Fustini <dfustini@xxxxxxxxxxxx> [220129 21:05]:
> > On Tue, Mar 23, 2021 at 09:43:26AM +0200, Tony Lindgren wrote:
> > > There is a timer wrap issue on dra7 for the ARM architected timer.
> > > In a typical clock configuration the timer fails to wrap after 388 days.
> > >
> > > To work around the issue, we need to use timer-ti-dm percpu timers instead.
> > >
> > > Let's configure dmtimer3 and 4 as percpu timers by default, and warn about
> > > the issue if the dtb is not configured properly.
> >
> > Hi Tony,
> >
> > This causes a conflict for IPU2 which is using timer 3 and 4.
> >
> > From arch/arm/boot/dts/dra7-ipu-dsp-common.dtsi:
> >
> > &ipu2 {
> > mboxes = <&mailbox6 &mbox_ipu2_ipc3x>;
> > ti,timers = <&timer3>;
> > ti,watchdog-timers = <&timer4>, <&timer9>;
> > };
>
> OK, sorry I missed that part.
>
> > I noticed an error ("could not get timer platform device") when booting
> > mainline on a BeagleBoard X15 (AM578):
> >
> > omap-rproc 55020000.ipu: assigned reserved memory node ipu2-memory@95800000
> > remoteproc remoteproc1: 55020000.ipu is available
> > remoteproc remoteproc1: powering up 55020000.ipu
> > remoteproc remoteproc1: Booting fw image dra7-ipu2-fw.xem4, size 3747220
> > omap-rproc 55020000.ipu: could not get timer platform device
> > omap-rproc 55020000.ipu: omap_rproc_enable_timers failed: -19
> > remoteproc remoteproc1: can't start rproc 55020000.ipu: -19
> >
> > I switched this errata fix to use timer 15 and 16 instead which resolves
> > the error. Do you think that is an acceptable solution?
>
> I think the only difference is that timers 15 and 16 are in l4_per3 instead
> of l4_per1. I doubt that matters as they are pretty much always clocked in
> this case. If you want to check you can run cyclictest :)
I ran this with existing errata fix with dmtimer 3 and 4:
root@am57xx-evm:~# cyclictest --mlockall --smp --priority=80 --interval=200 --distance=0
# /dev/cpu_dma_latency set to 0us
policy: fifo: loadavg: 0.02 0.03 0.05
T: 0 ( 1449) P:80 I:200 C: 800368 Min: 0 Act: 32 Avg: 22 Max: 128
T: 1 ( 1450) P:80 I:200 C: 800301 Min: 0 Act: 12 Avg: 23 Max: 70
And the results after my switch to dmtimer 15 and 16:
root@am57xx-evm:~# cyclictest --mlockall --smp --priority=80 --interval=200 --distance=0
# /dev/cpu_dma_latency set to 0us
policy: fifo: loadavg: 0.36 0.19 0.07
T: 0 ( 1711) P:80 I:200 C: 759599 Min: 0 Act: 6 Avg: 22 Max: 108
T: 1 ( 1712) P:80 I:200 C: 759539 Min: 0 Act: 19 Avg: 23 Max: 79
This doesn't appear to show any latency regression.
Any other options for cyclictest that that you'd recommend trying?
Thank you,
Drew
