> -----Original Message-----
> From: Chuck Lever III <chuck.lever@xxxxxxxxxx>
> Sent: Thursday, 27 April 2023 19:48
> To: Bernard Metzler <BMT@xxxxxxxxxxxxxx>
> Cc: Chuck Lever <cel@xxxxxxxxxx>; linux-rdma@xxxxxxxxxxxxxxx; Linux NFS
> Mailing List <linux-nfs@xxxxxxxxxxxxxxx>
> Subject: [EXTERNAL] Re: [PATCH RFC] RDMA/core: Store zero GIDs in some
> cases
>
>
>
> > On Apr 27, 2023, at 1:46 PM, Bernard Metzler <BMT@xxxxxxxxxxxxxx> wrote:
> >
> >
> >
> >> -----Original Message-----
> >> From: Chuck Lever <cel@xxxxxxxxxx>
> >> Sent: Thursday, 27 April 2023 19:15
> >> To: Bernard Metzler <BMT@xxxxxxxxxxxxxx>
> >> Cc: linux-rdma@xxxxxxxxxxxxxxx; linux-nfs@xxxxxxxxxxxxxxx
> >> Subject: [EXTERNAL] [PATCH RFC] RDMA/core: Store zero GIDs in some cases
> >>
> >> From: Bernard Metzler <bmt@xxxxxxxxxxxxxx>
> >>
> >> Tunnel devices have zero GIDs, so skip the zero GID check when
> >> setting up soft iWARP over a tunnel device.
> >>
> >> Suggested-by: Bernard Metzler <bmt@xxxxxxxxxxxxxx>
> >> Signed-off-by: Chuck Lever <chuck.lever@xxxxxxxxxx>
> >> ---
> >> drivers/infiniband/core/cache.c | 4 +++-
> >> drivers/infiniband/sw/siw/siw_main.c | 1 +
> >> include/rdma/iw_cm.h | 9 ++++++++-
> >> 3 files changed, 12 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/infiniband/core/cache.c
> >> b/drivers/infiniband/core/cache.c
> >> index 2e91d8879326..2493ca4f2739 100644
> >> --- a/drivers/infiniband/core/cache.c
> >> +++ b/drivers/infiniband/core/cache.c
> >> @@ -41,6 +41,7 @@
> >> #include <net/addrconf.h>
> >>
> >> #include <rdma/ib_cache.h>
> >> +#include <rdma/iw_cm.h>
> >>
> >> #include "core_priv.h"
> >>
> >> @@ -441,7 +442,8 @@ static int add_modify_gid(struct ib_gid_table
> *table,
> >> * leave other unused entries as the zero GID. Convert zero GIDs to
> >> * empty table entries instead of storing them.
> >> */
> >> - if (rdma_is_zero_gid(&attr->gid))
> >> + if (rdma_is_zero_gid(&attr->gid) &&
> >> + !(attr->device->iw_driver_flags & IW_F_STORE_0GID))
> >> return 0;
> >>
> >> entry = alloc_gid_entry(attr);
> >> diff --git a/drivers/infiniband/sw/siw/siw_main.c
> >> b/drivers/infiniband/sw/siw/siw_main.c
> >> index dacc174604bf..842a039fa457 100644
> >> --- a/drivers/infiniband/sw/siw/siw_main.c
> >> +++ b/drivers/infiniband/sw/siw/siw_main.c
> >> @@ -359,6 +359,7 @@ static struct siw_device *siw_device_create(struct
> >> net_device *netdev)
> >>
> >> /* Disable TCP port mapping */
> >> base_dev->iw_driver_flags = IW_F_NO_PORT_MAP;
> >> + base_dev->iw_driver_flags = IW_F_STORE_0GID;
> >>
> > That overwrites the first assignment. Probably better
> > '|= IW_F_STORE_0GID;' ? Or put them on one line...
>
> D'oh! Will fix.
Otherwise looks good of course. Could you
please check if that does not break 'normal' Ethernet
behavior? I am off from any kernel test infrastructure
for the next four (!!) weeks unfortunately. Sorry about
that.
Thanks for the patch!
Bernard.
>
> >> sdev->attrs.max_qp = SIW_MAX_QP;
> >> sdev->attrs.max_qp_wr = SIW_MAX_QP_WR;
> >> diff --git a/include/rdma/iw_cm.h b/include/rdma/iw_cm.h
> >> index 03abd30e6c8c..c48f2cbe37b5 100644
> >> --- a/include/rdma/iw_cm.h
> >> +++ b/include/rdma/iw_cm.h
> >> @@ -90,7 +90,14 @@ enum iw_flags {
> >> * reserve the port. This is required for soft iwarp
> >> * to play in the port mapped iwarp space.
> >> */
> >> - IW_F_NO_PORT_MAP = (1 << 0),
> >> + IW_F_NO_PORT_MAP = BIT(0),
> >> +
> >> + /*
> >> + * This flag allows the insertion of zero GIDs into the
> >> + * stored GID table. That is needed to enable soft iWARP
> >> + * on tunnel devices.
> >> + */
> >> + IW_F_STORE_0GID = BIT(1),
> >> };
> >>
> >> /**
>
>
> --
> Chuck Lever
>
