On 9/21/2020 7:49 PM, Jason Gunthorpe wrote:
On Mon, Sep 14, 2020 at 09:34:58AM +0300, Yishai Hadas wrote:
+static int query_sysfs_gid_ndev_ifindex(struct ibv_context *context,
+ uint8_t port_num, uint32_t gid_index,
+ uint32_t *ndev_ifindex)
+{
+ struct verbs_device *verbs_device = verbs_get_device(context->device);
+ char buff[IF_NAMESIZE] = {};
This init is not necessary
OK
diff --git a/libibverbs/verbs.c b/libibverbs/verbs.c
index 9507ffd..9427aba 100644
+++ b/libibverbs/verbs.c
@@ -240,6 +240,14 @@ LATEST_SYMVER_FUNC(ibv_query_gid, 1_1, "IBVERBS_1.1",
return 0;
}
+int _ibv_query_gid_ex(struct ibv_context *context, uint32_t port_num,
+ uint32_t gid_index, struct ibv_gid_entry *entry,
+ uint32_t flags, size_t entry_size)
+{
+ return ibv_cmd_query_gid_entry(context, port_num, gid_index, entry,
+ flags, entry_size);
+}
This extra function seems unncessary.
We've been creating C files for each object type, the gid stuff could
go in device.c
Jason
In two patches ahead we introduce some mask to optimize ibv_query_gid()
and ibv_query_gid_type() once they fallback over sysfs, so we may need
to differentiate
between this external verb API to the internal command, but YES, all of
this can be done under cmd_device.c, will be part of V1.
Thanks,
Yishai
