On Thu, Feb 6, 2020 at 11:38 AM Mark Bloch <markb@xxxxxxxxxxxx> wrote:
>
>
>
> On 2/5/2020 21:44, Devesh Sharma wrote:
> > It becomes difficult to make out from the output of ibv_devinfo
> > if a particular gid index is RoCE v2 or not.
> >
> > Adding a string to the output of ibv_devinfo -v to display the
> > gid type at the end of gid.
> >
> > The output would look something like below:
> > $ ibv_devinfo -v -d bnxt_re2
> > hca_id: bnxt_re2
> > transport: InfiniBand (0)
> > fw_ver: 216.0.220.0
> > node_guid: b226:28ff:fed3:b0f0
> > sys_image_guid: b226:28ff:fed3:b0f0
> > .
> > .
> > .
> > .
> > phys_state: LINK_UP (5)
> > GID[ 0]: fe80:0000:0000:0000:b226:28ff:fed3:b0f0, IB/RoCE v1
> > GID[ 1]: fe80::b226:28ff:fed3:b0f0, RoCE v2
> > GID[ 2]: 0000:0000:0000:0000:0000:ffff:c0aa:0165, IB/RoCE v1
> > GID[ 3]: ::ffff:192.170.1.101, RoCE v2
> > $
> > $
> > $
> >
> > Reviewed-by: Jason Gunthrope <jgg@xxxxxxxx>
> > Reviewed-by: Leon Romanovsky <leon@xxxxxxxxxx>
> > Reviewed-by: Gal Pressman <galpress@xxxxxxxxxx>
> > Reviewed-by: Parav Pandit <parav@xxxxxxxxxxxx>
> > Signed-off-by: Devesh Sharma <devesh.sharma@xxxxxxxxxxxx>
> > ---
> > libibverbs/examples/devinfo.c | 61 ++++++++++++++++++++++++++++++++++---------
> > 1 file changed, 49 insertions(+), 12 deletions(-)
> >
> > diff --git a/libibverbs/examples/devinfo.c b/libibverbs/examples/devinfo.c
> > index bf53eac..7e2015b 100644
> > --- a/libibverbs/examples/devinfo.c
> > +++ b/libibverbs/examples/devinfo.c
> > @@ -40,6 +40,7 @@
> > #include <getopt.h>
> > #include <endian.h>
> > #include <inttypes.h>
> > +#include <arpa/inet.h>
> >
> > #include <infiniband/verbs.h>
> > #include <infiniband/driver.h>
> > @@ -162,12 +163,50 @@ static const char *vl_str(uint8_t vl_num)
> > }
> > }
> >
> > -static int print_all_port_gids(struct ibv_context *ctx, uint8_t port_num, int tbl_len)
> > +#define DEVINFO_INVALID_GID_TYPE 2
> > +static const char *gid_type_str(enum ibv_gid_type type)
> > {
> > + switch (type) {
> > + case IBV_GID_TYPE_IB_ROCE_V1: return "IB/RoCE v1";
>
> You call this function only if link layer is ethernet, why do you need the "IB/" part?
Jason do you have any suggestion here, "IB/ROCE v1" is the standard
string rdma-cm and
/sys/class/infiniband/bnxt_re0/ports/1/gid_attrs/types/* use to
distinguish v2 gid from v1/IB gids.
>
>
> > + case IBV_GID_TYPE_ROCE_V2: return "RoCE v2";
> > + default: return "Invalid gid type";
> > + }
> > +}
> > +
> > +static void print_formated_gid(union ibv_gid *gid, int i,
> > + enum ibv_gid_type type, int ll)
> > +{
> > + char gid_str[INET6_ADDRSTRLEN] = {};
> > + char str[20] = {};
> > +
> > + if (ll == IBV_LINK_LAYER_ETHERNET)
> > + sprintf(str, ", %s", gid_type_str(type));
> > +
> > + if (type == IBV_GID_TYPE_IB_ROCE_V1)
> > + printf("\t\t\tGID[%3d]:\t\t%02x%02x:%02x%02x:%02x%02x:%02x%02x:%02x%02x:%02x%02x:%02x%02x:%02x%02x%s\n",
> > + i, gid->raw[0], gid->raw[1], gid->raw[2],
> > + gid->raw[3], gid->raw[4], gid->raw[5], gid->raw[6],
> > + gid->raw[7], gid->raw[8], gid->raw[9], gid->raw[10],
> > + gid->raw[11], gid->raw[12], gid->raw[13], gid->raw[14],
> > + gid->raw[15], str);
> > +
> > + if (type == IBV_GID_TYPE_ROCE_V2) {
> > + inet_ntop(AF_INET6, gid->raw, gid_str, sizeof(gid_str));
> > + printf("\t\t\tGID[%3d]:\t\t%s%s\n", i, gid_str, str);
> > + }
> > +}
> > +
> > +static int print_all_port_gids(struct ibv_context *ctx,
> > + struct ibv_port_attr *port_attr,
> > + uint8_t port_num)
> > +{
> > + enum ibv_gid_type type;
> > union ibv_gid gid;
> > + int tbl_len;
> > int rc = 0;
> > int i;
> >
> > + tbl_len = port_attr->gid_tbl_len;
> > for (i = 0; i < tbl_len; i++) {
> > rc = ibv_query_gid(ctx, port_num, i, &gid);
> > if (rc) {
> > @@ -175,17 +214,15 @@ static int print_all_port_gids(struct ibv_context *ctx, uint8_t port_num, int tb
> > port_num, i);
> > return rc;
> > }
> > +
> > + rc = ibv_query_gid_type(ctx, port_num, i, &type);
>
> I really dislike this, what if in the future a new gid type will be added (I know, I know, what are
> the chances :) )
>
> old rdma-core won't display anything in that index (while today it displays anything as long
> as the gid isn't a null gid).
Parav, do you have any suggestions here?
>
> Mark
>
> > + if (rc) {
> > + rc = 0;
> > + type = DEVINFO_INVALID_GID_TYPE;
> > + }
> > if (!null_gid(&gid))
> > - printf("\t\t\tGID[%3d]:\t\t%02x%02x:%02x%02x:%02x%02x:%02x%02x:%02x%02x:%02x%02x:%02x%02x:%02x%02x\n",
> > - i,
> > - gid.raw[ 0], gid.raw[ 1],
> > - gid.raw[ 2], gid.raw[ 3],
> > - gid.raw[ 4], gid.raw[ 5],
> > - gid.raw[ 6], gid.raw[ 7],
> > - gid.raw[ 8], gid.raw[ 9],
> > - gid.raw[10], gid.raw[11],
> > - gid.raw[12], gid.raw[13],
> > - gid.raw[14], gid.raw[15]);
> > + print_formated_gid(&gid, i, type,
> > + port_attr->link_layer);
> > }
> > return rc;
> > }
> > @@ -614,7 +651,7 @@ static int print_hca_cap(struct ibv_device *ib_dev, uint8_t ib_port)
> > printf("\t\t\tphys_state:\t\t%s (%d)\n",
> > port_phy_state_str(port_attr.phys_state), port_attr.phys_state);
> >
> > - if (print_all_port_gids(ctx, port, port_attr.gid_tbl_len))
> > + if (print_all_port_gids(ctx, &port_attr, port))
> > goto cleanup;
> > }
> > printf("\n");
> >
