On Fri, Sep 16, 2022 at 09:39:56AM +0200, Stefan Sperling wrote: > On Fri, Sep 16, 2022 at 08:27:59AM +0200, Adam Szewczyk wrote: > > Checked on actual hardware it seems to work fine. So I report it also to > > Qubes and Xen Project, but marmarek (head of qubes development team) > > suggest that it can be a problem with driver itself: > > "I'm not sure if that's relevant, but a common cause is incorrect MSI / > > MSI-X detection/support by the driver. HVM in Qubes do not support MSI-X, > > but do support MSI. MSI-X, even if the device really supports it, is masked > > from PCI capabilities. Driver should fallback to MSI (or INTx) if MSI-X > > fails to setup (or isn't there at all), but many drivers have this path > > buggy, since it isn't exercised in most common (native) scenario." > > re(4) only uses MSI, it does not ever even try MSI-X. As can be seen here: > https://github.com/openbsd/src/blob/master/sys/dev/pci/if_re_pci.c#L161
But there has been at least one report of MSI, (not MSI-X), causing the error that the OP is seeing with an 're' device on bare hardware, (no hypervisor): https://marc.info/?l=openbsd-tech&m=165987603706259&w=2 So maybe test the patch in that mail to see if using a regular non-MSI interrupt mitigates the problem. Then look for the real bug if it does.
