On Mon, Nov 14, 2022 at 9:00 AM Guoqing Jiang <guoqing.jiang@xxxxxxxxx> wrote:
>
> Hi Jinpu,
>
> On 11/14/22 3:39 PM, Jinpu Wang wrote:
> > Hi Guoqing,
> >
> > Thx for the patch, see comments below.
> > On Sun, Nov 13, 2022 at 2:08 AM Guoqing Jiang <guoqing.jiang@xxxxxxxxx> wrote:
> >> The ib_dev_count is supposed to track the number of added ib devices
> >> which is only used in rtrs_srv_{add,remove}_one.
> >>
> >> However we only trigger rtrs_srv_add_one from rnbd_srv_init_module
> >> -> rtrs_srv_open -> ib_register_client -> client->add which should
> >> happen only once.
> > client->add is call per ib_device, eg:
> > jwang@xxxxxxxxxxxx:~$ ls -l /sys/class/infiniband/mlx5_*
> > lrwxrwxrwx 1 root root 0 Nov 8 13:49 /sys/class/infiniband/mlx5_0 ->
> > ../../devices/pci0000:ae/0000:ae:00.0/0000:af:00.0/infiniband/mlx5_0
> > lrwxrwxrwx 1 root root 0 Nov 8 13:49 /sys/class/infiniband/mlx5_1 ->
> > ../../devices/pci0000:ae/0000:ae:00.0/0000:af:00.1/infiniband/mlx5_1
> > rtrs will be call twice for mlx5_0 and mlx5_1 devices
>
> Ah, yes.
>
> But still we can only load/unload module once, I guess it was used to avoid
> racy condition (concurrent loading/unloading module?), could you elaborate
> why it is needed?
The change was introduced due to a bug report, you can follow the
discussion here for the history and reason:
https://lore.kernel.org/linux-rdma/20200617103732.10356-1-haris.iqbal@xxxxxxxxxxxxxxx/
>
> Thanks,
> Guoqing
Thx!
