Le dim. 14 févr. 2021 à 13:28, Mathew, Cherry G. <c%bow.st@localhost> a écrit :
I'm not sure if the interrupt sharing is a problem either because all interrupts are despatched through a central stub anyway - however the numbers you show below, if for similar time periods show a high number which is unusual. if, as I earlier requested, you could get fixed interval numbers and average them across several time slices across the iperf test you are running, it would be good.
Avoiding the interrupt sharing is still useful, since you avoid the
I/O read from the interrupt routine. So that can have an effect.
That was the primary driver for the MSI support I've added.
Unfortunately MSI-X didn't work out, so it's still disabled - so for
the re0 it still falls back to INTx. I'd like to get back to getting
MSI-X working.
Matthias, can you check if disabling sdhc* and ichsmb* does anything?