On 10/20/2026 2:24 AM, Przemek Kitszel wrote: > On 10/19/26 12:51, Aaron Ma wrote: >> Adapter-scoped synchronization must be available to Admin Queue users >> from initial queue creation through the last teardown command. Probe >> currently obtains its adapter reference only after hardware and package >> initialization, and error cleanup and removal drop it before hardware >> teardown. >> >> Acquire the reference before either normal hardware initialization or >> firmware recovery-mode initialization. Release it after hardware and >> device cleanup, balancing recovery-mode probe failures and removal too. >> >> This establishes the lifetime prerequisite for moving package ownership >> exclusion from a module-global mutex to the shared adapter. >> >> Signed-off-by: Aaron Ma <[email protected]> > > it was me to move hw init before adapter init, to make hw details > readily accessible to the adapter > > but I agree that SW lock protection covering HW reinit will be great, > and my needs could be achieved by other means, even if less elegant > > so, I'm happy with your patch > OTOH I have a big series that git-conflicts, would be easiest if your > series will be merged first (so I will adjust), but I don't want to be > delayed (will be CCing you on it going forward) > > the real thing is of course the 3rd patch not this one >
I got a Sashiko report about adapter not being present during hw deinit, which was problematic for PTP. I think the lifetime rules make more sense if the adapter outlives the PFs. Sounds good. Thanks, Jake
