Re: [PATCH rdma-next v2 1/1] RDMA/mana_ib: memory windows

[Konstantin Taranov]

> On Wed, Mar 18, 2026 at 10:23:23AM -0700, Konstantin Taranov wrote:
> > From: Konstantin Taranov <kotaranov@xxxxxxxxxxxxx>
> >
> > Implement .alloc_mw() and .dealloc_mw() for mana device.
> >
> > Signed-off-by: Konstantin Taranov <kotaranov@xxxxxxxxxxxxx>
> > ---
> > v2: fixed comments. Cleaned up the use of mana_gd_send_request()
> > drivers/infiniband/hw/mana/device.c  |  3 ++
> > drivers/infiniband/hw/mana/mana_ib.h |  8 +++++
> >  drivers/infiniband/hw/mana/mr.c      | 53
> +++++++++++++++++++++++++++-
> >  include/net/mana/gdma.h              |  5 +++
> >  4 files changed, 68 insertions(+), 1 deletion(-)
> 
> <...>
> 
> > +static int mana_ib_gd_create_mw(struct mana_ib_dev *dev, struct
> > +mana_ib_pd *pd, struct ib_mw *ibmw) {
> > +	struct mana_ib_mw *mw = container_of(ibmw, struct mana_ib_mw,
> ibmw);
> > +	struct gdma_context *gc = mdev_to_gc(dev);
> > +	struct gdma_create_mr_response resp = {};
> > +	struct gdma_create_mr_request req = {};
> > +	int err;
> > +
> > +	mana_gd_init_req_hdr(&req.hdr, GDMA_CREATE_MR, sizeof(req),
> sizeof(resp));
> > +	req.pd_handle = pd->pd_handle;
> 
> Both sashiko, which runs on Gemnini
> https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fsashi
> ko.dev%2F%23%2Fpatchset%2F20260318172323.1416803-1-
> kotaranov%40linux.microsoft.com&data=05%7C02%7Ckotaranov%40microso
> ft.com%7Ce8888ce68913479d8db508de8e53a186%7C72f988bf86f141af91ab
> 2d7cd011db47%7C1%7C0%7C639104687038043303%7CUnknown%7CTWFp
> bGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4z
> MiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=r9n1Jmd
> Ke5smdwDQto80%2BHWosk3DrRa1z4jdNmzjLjk%3D&reserved=0
> and my local claude checks pointed to the same area of the code:
> 
>    25 > +     mana_gd_init_req_hdr(&req.hdr, GDMA_CREATE_MR, sizeof(req),
> sizeof(resp));
>    26 > +     req.pd_handle = pd->pd_handle;
>    27
>    28 mana_gd_init_req_hdr() sets msg_version to GDMA_MESSAGE_V1.
>    29 mana_ib_gd_create_mr() explicitly upgrades this to
> GDMA_MESSAGE_V2
>    30 for all GDMA_CREATE_MR requests, regardless of mr_type:
>    31
>    32 mana_ib_gd_create_mr() {
>    33         mana_gd_init_req_hdr(&req.hdr, GDMA_CREATE_MR, ...);
>    34         req.hdr.req.msg_version = GDMA_MESSAGE_V2;
>    35         ...
>    36 }
>    37
>    38 Is it intentional that mana_ib_gd_create_mw() sends
> GDMA_CREATE_MR
>    39 at GDMA_MESSAGE_V1 rather than GDMA_MESSAGE_V2?
> 
Thanks. I sent v3. It does not matter v1 vs v2 as HW supports both, I just wrote this patch long time ago.
I changed to v2 to make code more unified.

> Thanks