| From: Alex Williamson [alex.william...@redhat.com] | Sent: Thursday, June 25, 2015 1:57 PM | | We expose [the PBA Offset] to the guest VM, but the guest probably | doesn't consume it, I don't know of any drivers that do.
Ah, okay. You're doing this for the Hypervisor traps of accesses to the device's Configuration Space. Makes sense. | It would be really helpful if these kinds of details were in the commit | log, specifically the value we expect to see for pba_offset when it's | wrong, and in this case that it's obviously wrong because it's beyond | the end of the BAR. Knowing that it's limited to VFs is also helpful if | someone needs to go back and rework the quirk later. Yeah, sorry about that. Gabriel sent it around for internal review but I was overwhelmed with other things and didn't get a chance to make revision comments. Gabriel's new to the process and assumed when someone else said that "it's sort of looks like what you'd need to do ..." that that was ACK enough ... :-) I certainly would have noted that he was applying the quirk to all the T5 PCI-E Functions instead of only the SR-IOV Virtual Functions. | And, because it's obviously wrong, we can further limit the scope of the | quirk by testing those parameters. That would have automatically | skipped the quirk if we had used the wrong device ID match or when | Chelsio runs out of device IDs and starts using gaps left in the device | ID address space (I only see ~34 device IDs in the 58xx space used so | far according to pci-ids). Thanks, Sure, if you like. At our current rate we would end up reusing these PCI Device IDs in about 50 years though. Casey
