On Mon, Apr 19, 2021 at 11:08 PM Jason Gunthorpe wrote: > > On Sun, Apr 18, 2021 at 12:09:16AM +0530, Devesh Sharma wrote: > > > The host crash I indicated earlier is actually caused by patch 4 and > > not by patch 3 from this series. I spent time to root cause the > > This makes a lot more sense. > > The ulp_id stuff does need to go away as well though. We shall address this concern with Aux implementation. > > > problem and realized that patch-4 is touching quite many areas which > > would require much intrusive testing and validation. > > As I indicated earlier, we are implementing the PCI Aux driver > > interface at a faster pace. > > Doing an aux driver doesn't mean you get to keep all these single > implementation function pointers - see the discussion around Intel's > patches. Sure, We will try addressing this concern as well. Could you please point me to the exact patches please... > > > The problem of module referencing would be rectified with PCI aux > > change by inheritance. > > The first three patches are clearly an improvement, and quite trivial, > so I'm going to take them. To my mind the first 3 patches in this series are fine and I agree to Ack those. Of-course I want to spend some more time establishing all of our internal test harness passing. I can Ack those in a couple of days at the earliest. Further, I do not want to proceed with patch 4 and 5 as those patches are too intrusive and cause host hang/crash during rmmod bnxt_re followed by rmmod bnxt_en. Probably this is some kind of race. If I try to add a few prints to the debug, the problem does not appear. Since, we want to remain focused on PCI Aux implementation instead of resolving instabilities at this point, we would like to pick up the idea from 4 and 5 in our PCI Aux implementation. > > Jason -- -Regards Devesh