Re: [PATCH rdma-next 01/10] RDMA/umem: Add reference counting to ib_umem

[Leon Romanovsky <leon@kernel.org>]

On Tue, Feb 03, 2026 at 10:51:38AM -0400, Jason Gunthorpe wrote:
> On Tue, Feb 03, 2026 at 09:49:53AM +0100, Jiri Pirko wrote:
> > From: Jiri Pirko <jiri@nvidia.com>
> > 
> > Introduce reference counting for ib_umem objects to simplify memory
> > lifecycle management when umem buffers are shared between the core
> > layer and device drivers.
> 
> I admit I have reservations about this too.. The flow should not be so
> convoluted that a refcount is necessary. The lifecycle of a umem is
> not uncertain at all.
> 
> I imagine'd it would be like:
> 
> core code:
>   if (ops->create_cq_umem) {
>      umem = umem_get
>      rc = ops->create_cq_umem(umem)
>      if (rc)
>       umem_free(umem)
>   } else {
>      rc = ops->create_cq()
>   }
> 
> Driver:
>   create_cq():
>     copy_from_user(drvdata)
>     umem = umem_get()
>     rc = driver_create_cq_umem(umem, &drvdata))
>     if (rc)
>       umem_free(umem)
> 
>    create_cq_umem()
>      copy_from_user(drvdata)
>      return driver_create_cq_umem(umem, &drvdata)
> 
>    destroy_cq()
>      destry_hw
>      umem_free()
> 
> This basically moves all the working code in the driver to the
> driver_create_cq_umem() which *always* gets a umem as a parameter.
> 
> If the user uses the drvdata path to specify the umem then the driver
> helper create_cq() creates the umem from the drvdata parameters,
> otherwise the core creates it from the common UATTRs.
> 
> We can keep things so the umem is always freed by the driver on
> success, which doesn't require any driver changes.
> 
> It should never be "shared", this is just a very simple unwind on
> error kind of pattern.
> 
> I think the challenge here is to unwind the drivers into the above
> three functions so they don't have a mess of convoluted error handling
> around the umem.

I opted for an even simpler approach: embed the umem directly in ib_cq, set  
cq->umem in ib_core’s create_cq, and clean it up in ib_core’s destroy_cq.

The beginning of the series is available here:  
https://git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git/log/?h=refactor-umem-v1

Thanks

> 
> Jason