On Sat, Jan 30, 2021 at 01:19:36AM +0000, Saleem, Shiraz wrote:

quoted

Subject: Re: [PATCH 07/22] RDMA/irdma: Register an auxiliary driver and
implement private channel OPs

On Wed, Jan 27, 2021 at 07:16:41PM -0400, Jason Gunthorpe wrote:

quoted

On Wed, Jan 27, 2021 at 10:17:56PM +0000, Saleem, Shiraz wrote:

quoted

Even with another core PCI driver, there still needs to be private
communication channel between the aux rdma driver and this PCI
driver to pass things like QoS updates.

Data pushed from the core driver to its aux drivers should either be
done through new callbacks in a struct device_driver or by having a
notifier chain scheme from the core driver.

Right, and internal to driver/core device_lock will protect from parallel
probe/remove and PCI flows.

OK. We will hold the device_lock while issuing the .ops callbacks from core driver.
This should solve our synchronization issue.

There have been a few discussions in this thread. And I would like to be clear on what
to do.

So we will,

1. Remove .open/.close, .peer_register/.peer_unregister
2. Protect ops callbacks issued from core driver to the aux driver
with device_lock

A notifier chain is probably better, honestly.

Especially since you don't want to split the netdev side, a notifier
chain can be used by both cases equally.

Jason