Oliver Freyermuth

Trying the commands: ``` # vcgencmd measure_clock pllb frequency(100)=3000000000 # vcgencmd measure_clock arm frequency(48)=750199232 ``` seems to work fine in that situation.

I grepped through the logs and found the following, potentially related messages: ``` [567278.869559] xhci_hcd 0000:01:00.0: Ring expansion failed [567278.869581] hub 1-1.2.4:1.0: activate --> -12 [567294.010078] xhci_hcd 0000:01:00.0: Ring expansion...

After reading through #3040 which might be related(?), I have now added: ``` coherent_pool=4M ``` to my ``/boot/cmdline`` and rebooted. Other suggestions welcome. Will report back latest in ~2 weeks...

HI @lategoodbye , thanks! I wondered whether this is a leak or "just" fragmentation within the coherent memory region. I'm a bit astonished this would only hit me, though, since...

@popcornmix Thanks, compiling now. Hope it'll be straightforward to go back to the upstream built kernel (i.e."just upgrade"). Will report back once I found something. I wonder if: ``` CONFIG_DEBUG_KMEMLEAK=Y...

Modified kernel is running, at least there is no immediate leak visible (even when triggering an active `scan`). Guess I'll have to wait a few days. Or maybe it is...

I have by now executed all the workloads I execute on the device, including SMART tests of the attached HDD, running `hd-idle`, running `fstrim` on the SD card, running a...

@lategoodbye , it's 5 days now and `kmemleak` still shows up empty. So I think we can already exclude a leak, right? Since I increased to `coherent_pool=4M`, if it is...

@lategoodbye , the issue is now back after 23 days of uptime, so the increase to `coherent_pool=4M` seems to have delayed it. However: ``` sudo cat /sys/kernel/debug/kmemleak ``` gives me...

Since there were still no leaks and no further ideas, I've now: - Reverted to the stock kernel. - Switched from the USB 2 hub to a USB 3 hub...