Re: [PATCH for-next resend 07/24] RDMA/hfi2: Add system core header files

[Leon Romanovsky <leon@kernel.org>]

On Mon, Mar 16, 2026 at 05:37:30PM -0400, Dennis Dalessandro wrote:
> On 3/16/26 11:58 AM, Leon Romanovsky wrote:
> > On Wed, Mar 11, 2026 at 01:54:20PM -0400, Dennis Dalessandro wrote:
> > > Add header files for hooking into the core system, things like IRQs and
> > > affinity.
> > > 
> > > Co-developed-by: Dean Luick <dean.luick@cornelisnetworks.com>
> > > Signed-off-by: Dean Luick <dean.luick@cornelisnetworks.com>
> > > Co-developed-by: Bendan Cunningham <brendan.cunningham@cornelisnetworks.com>
> > > Signed-off-by: Breandan Cunningham <brendan.cunningham@cornelisnetworks.com>
> > > Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
> > > ---
> > >   drivers/infiniband/hw/hfi2/affinity.h |   86 +++++++++++++++++++++++++++++++++
> > >   drivers/infiniband/hw/hfi2/aspm.h     |   36 ++++++++++++++
> > >   drivers/infiniband/hw/hfi2/efivar.h   |   17 +++++++
> > >   drivers/infiniband/hw/hfi2/eprom.h    |   11 ++++
> > >   drivers/infiniband/hw/hfi2/mmu_rb.h   |   78 ++++++++++++++++++++++++++++++
> > >   drivers/infiniband/hw/hfi2/msix.h     |   31 ++++++++++++
> > >   6 files changed, 259 insertions(+)
> > >   create mode 100644 drivers/infiniband/hw/hfi2/affinity.h
> > >   create mode 100644 drivers/infiniband/hw/hfi2/aspm.h
> > >   create mode 100644 drivers/infiniband/hw/hfi2/efivar.h
> > >   create mode 100644 drivers/infiniband/hw/hfi2/eprom.h
> > >   create mode 100644 drivers/infiniband/hw/hfi2/mmu_rb.h
> > >   create mode 100644 drivers/infiniband/hw/hfi2/msix.h
> > 
> > <...>
> > 
> > > +/* Can be used for both memory and cpu */
> > > +enum affinity_flags {
> > > +	AFF_AUTO,
> > > +	AFF_NUMA_LOCAL,
> > > +	AFF_DEV_LOCAL,
> > > +	AFF_IRQ_LOCAL
> > > +};
> > 
> > Maybe I'm misremembering, but I recall we already discussed affinity
> > management in the driver in the context of hfi1, and our position at the
> > time was that handling affinity belongs in user space.
> 
> I don't recall specifically, but I do agree it should be handled in user
> space as the ultimate decider. The driver just provides hints as to where
> best to run things for optimal performance. User space is free to manage if
> it chooses.

So there is no need in all this affinity code, which looks like policy,
just use right attributes from the beginning.

Thanks

> 
> -Denny
>