When connecting Streamline to Juno, all I see are A53 counters. Any chance of getting the A57 (or better yet both) sets of counters or is HMP Streamline fundamentally broken?
-eric
Streamline/gator should be able to pick up both sets of counters. However having just got access to a board I think I have observed similar behaviour. I'll let the Streamline team know and see if we can make some progress.
Standby for more...
Hi Eric,
The issue is that the Linux kernel does not currently support collecting data from multiple CPU PMUs. As such it only displays one of the PMUs - in the case of the Juno development platform it chooses the Cortex-A53's PMU.
There is a fix to the kernel currently being prepared to address this problem.
In the mean time while you are not able to see the Cortex-A57's counters explicitly, there is sufficient similarity between the cores that collecting the Cortex-A53's counters will allow you to read most values from the Cortex-A57 cluster as well - i.e. if you collect data from a Cortex-A53 counter it will, providing that counter exists on the Cortex-A57 as well, collect from both clusters.
Once the Linux kernel supports having multiple types of PMU counters it should work as expected.
Hope that helps,
Jon
Yes, it seems like the common counters work, but we are observing that certain counters only report values when the A53s are active (things like branch: mispredicted, l2 data access, etc. So seems like things are really bad off unless you are only measuring on the A53s.
Hi,
Has the latest linaro kernel fixed this issue?
Please let me know.
Thanks,
Biswa
Hi Biswa,
I'm afraid this has not been resolved yet. I don't have any estimates about how long it will take but it is likely still several months away.
Sorry,