On 6/19/2020 3:33 PM, Jason Gunthorpe wrote:
On Wed, Jun 17, 2020 at 10:45:47AM +0300, Yishai Hadas wrote:
@@ -418,7 +427,13 @@ static struct ibv_context *verbs_import_device(int cmd_fd)
set_lib_ops(context_ex);
+ context_ex->priv->imported = true;
ctx = &context_ex->context;
+ ret = ibv_cmd_alloc_async_fd(ctx);
+ if (ret) {
+ ibv_close_device(ctx);
+ ctx = NULL;
+ }
out:
ibv_free_device_list(dev_list);
return ctx;
This hunk should probably be in the prior patch, or ideally the order
of these two patches should be swapped.
I can swap the order of those two patches as you suggested.
However, in this case, the first one will be some preparation to force
the ioctl mode upon an 'imported' flow and only then the second one will
introduce the 'imported' flow as part of adding ibv_import_device().
If you are fine with that let's go by that approach.
Yishai
