Re: [RFC] rdma/uverbs: Sketch for an ioctl framework

[Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>]

On Fri, May 27, 2016 at 12:22:56AM +0000, Hefty, Sean wrote:

> A lock associated with the idr is held while objects are acquired.
> A simple reference count is bumped for each object when used.  If
> exclusive access is required, a flag is set.  If the exclusive flag
> is set, then any attempt to acquire the object is rejected and the
> ioctl fails.

How is that different from your basic idr+kref locking scheme common
in the kernel?

What is a non-blocking exclusive useful for? Removal?

> Coordination with the removal of the kernel client (i.e. device) is
> done using a rw-lock.  Ioctls acquire a read lock, while the removal
> processing acquires the write lock.

uverbs ended up using a more complex srcu scheme for this citing
performance reasons.. I argued for the simpler rwlock at the time..

> synchronization seems sufficient.  Any lower-level locking is the
> responsibility of the kernel client.

So there are still object locks, but the core doesn't help with them
at all.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html