On 16.9.2019 12.41, Peter Chen wrote:
Hi Mathias, I met "event ring full error" like below, this error is met when I do iozone test on UAS storage at v4.19.35 kernel, but not meet this error at linux-next tree (08/24). The same host and test UAS storage device are used. This issue is due to xhci_handle_event does not return 0 long time, maybe the xHC speed is fast enough at that time. If I force the xhci_handle_event only run 100 times before update ERST dequene pointer, it will not occur this error. I did not see any changes for xhci_handle_event at the latest code, so in theory, it should have this issue too. Do you think if we need to improve xhci_handle_event to avoid event ring?
Possibly. We need to check the details of what types of events the ring is filled with, and why handling them takes so long. does irqsoff tracing show anything blocking interrupts for long? It's also possible that we don't get interrupts early enough. Either if interupts are moderated, or event ring is filled with events that don't generate interrupts (BEI flag set). -Mathias
