On Wed, Nov 18, 2020 at 02:43:57PM +0200, Gal Pressman wrote:
> On 16/11/2020 22:23, Jason Gunthorpe wrote:
> > +int ibv_cmd_query_device_any(struct ibv_context *context,
> > + const struct ibv_query_device_ex_input *input,
> > + struct ibv_device_attr_ex *attr, size_t attr_size,
> > + struct ib_uverbs_ex_query_device_resp *resp,
> > + size_t *resp_size)
> > +{
> > + struct ib_uverbs_ex_query_device_resp internal_resp;
> > + size_t internal_resp_size;
> > + int err;
> > +
> > + if (input && input->comp_mask)
> > + return EINVAL;
> > + if (attr_size < sizeof(attr->orig_attr))
> > + return EINVAL;
> > +
> > + if (!resp) {
> > + resp = &internal_resp;
> > + internal_resp_size = sizeof(internal_resp);
> > + resp_size = &internal_resp_size;
> > + }
> > + memset(attr, 0, attr_size);
> > + memset(resp, 0, *resp_size);
> > +
> > + if (attr_size > sizeof(attr->orig_attr)) {
> > + struct ibv_query_device_ex cmd = {};
> > +
> > + err = execute_cmd_write_ex(context,
> > + IB_USER_VERBS_EX_CMD_QUERY_DEVICE,
> > + &cmd, sizeof(cmd), resp, *resp_size);
> > + if (err) {
> > + if (err != EOPNOTSUPP)
>
> Are you sure about that?
> I think older kernels return ENOSYS.
Oh, that is possibly true, we changed those at one point, I didn't
test that far back.
Jason
