On Wed, Feb 27, 2019 at 09:45:06AM +0200, Gal Pressman wrote:
> On 26-Feb-19 21:20, Leon Romanovsky wrote:
> > On Thu, Feb 21, 2019 at 05:33:03PM +0200, Gal Pressman wrote:
> >> Add EFA node, transport and protocol types to core code.
> >> In addition, EFA driver ID is added to the IOCTL interface uapi.
> >>
> >> Signed-off-by: Gal Pressman <galpress@xxxxxxxxxx>
> >> ---
> >> drivers/infiniband/core/sysfs.c | 1 +
> >> drivers/infiniband/core/verbs.c | 2 ++
> >> include/rdma/ib_verbs.h | 8 ++++++--
> >> include/uapi/rdma/rdma_user_ioctl_cmds.h | 1 +
> >> 4 files changed, 10 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/infiniband/core/sysfs.c b/drivers/infiniband/core/sysfs.c
> >> index 9b6a065bdfa5..f7fac4c81559 100644
> >> --- a/drivers/infiniband/core/sysfs.c
> >> +++ b/drivers/infiniband/core/sysfs.c
> >> @@ -1194,6 +1194,7 @@ static ssize_t node_type_show(struct device *device,
> >> case RDMA_NODE_RNIC: return sprintf(buf, "%d: RNIC\n", dev->node_type);
> >> case RDMA_NODE_USNIC: return sprintf(buf, "%d: usNIC\n", dev->node_type);
> >> case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
> >> + case RDMA_NODE_EFA: return sprintf(buf, "%d: EFA\n", dev->node_type);
> >> case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
> >> case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
> >> default: return sprintf(buf, "%d: <unknown>\n", dev->node_type);
> >> diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
> >> index 5a5e83f5f0fc..fc6f748a1933 100644
> >> --- a/drivers/infiniband/core/verbs.c
> >> +++ b/drivers/infiniband/core/verbs.c
> >> @@ -218,6 +218,8 @@ rdma_node_get_transport(enum rdma_node_type node_type)
> >> return RDMA_TRANSPORT_USNIC_UDP;
> >> if (node_type == RDMA_NODE_RNIC)
> >> return RDMA_TRANSPORT_IWARP;
> >> + if (node_type == RDMA_NODE_EFA)
> >> + return RDMA_TRANSPORT_EFA;
> >>
> >> return RDMA_TRANSPORT_IB;
> >> }
> >> diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
> >> index 225cb76d469f..55534aba22bf 100644
> >> --- a/include/rdma/ib_verbs.h
> >> +++ b/include/rdma/ib_verbs.h
> >> @@ -108,6 +108,7 @@ enum rdma_node_type {
> >> RDMA_NODE_RNIC,
> >> RDMA_NODE_USNIC,
> >> RDMA_NODE_USNIC_UDP,
> >> + RDMA_NODE_EFA,
> >> };
> >>
> >> enum {
> >> @@ -119,14 +120,16 @@ enum rdma_transport_type {
> >> RDMA_TRANSPORT_IB,
> >> RDMA_TRANSPORT_IWARP,
> >> RDMA_TRANSPORT_USNIC,
> >> - RDMA_TRANSPORT_USNIC_UDP
> >> + RDMA_TRANSPORT_USNIC_UDP,
> >> + RDMA_TRANSPORT_EFA,
> >> };
> >>
> >> enum rdma_protocol_type {
> >> RDMA_PROTOCOL_IB,
> >> RDMA_PROTOCOL_IBOE,
> >> RDMA_PROTOCOL_IWARP,
> >> - RDMA_PROTOCOL_USNIC_UDP
> >> + RDMA_PROTOCOL_USNIC_UDP,
> >> + RDMA_PROTOCOL_EFA,
> >> };
> >
> > Everything above is backed by specification. I disagree that we need
> > to declare new transport, protocol and node every time some large company
> > decides to reinvent the wheel. Especially if this company refuses
> > provide specification grade documentation.
>
> EFA is not an IB_CA node/IB transport/IB protocol.
> We have two options:
> 1. Add EFA definitions
> 2. Add 'unspecified' definitions
>
> Both will not be backed up by the spec, at least with option #1 we'll have more
> information than an ambiguous unspecified.
Excellent options for out-of-tree driver.
Those fields are important for interopability, both EFA and "unspecified"
doesn't qualify for that. So there is third option - do standardization.
Thanks
Attachment:
signature.asc
Description: PGP signature
