Hi Ophir, Adrien,
On Fri, Mar 02, 2018 at 08:12:58PM +0100, Adrien Mazarguil wrote:
> On Wed, Feb 28, 2018 at 04:47:30PM +0000, Ophir Munk wrote:
> > The following scenario causes a crash in function mlx4_get_ifname
> > 1. On testpmd startup mlx4 device is probed and started
> > 2. mlx4 sriov is disabled. As a result an RMV event is sent to
> > testpmd which closes the device and nullify the priv struct
> > members.
> > 3. Running 'show port info all' in testpmd results in segmentation
> > fault because of accessing NULL pointer priv->ctx
> >
> > The fix is to return with an error from mlx4_get_ifname() if priv->ctx
> > member is NULL.
>
> So priv->ctx is NULL after mlx4_dev_close() was called. In my opinion the
> fact testpmd is allowed to call rte_eth_dev_info_get() on a closed port is
> fishy, there are quite a few other ethdev callbacks that may end up
> crashing in such a scenario.
>
> Since commit 284c908cc588 ("app/testpmd: request device removal interrupt"),
> testpmd automatically calls rte_eth_dev_close() and rte_eal_dev_detach()
> after receiving a removal event on a port.
>
> rte_eth_dev_close() documentation says:
>
> "Close a stopped Ethernet device. The device cannot be restarted!
> The function frees all resources except for needed by the
> closed state. To free these resources, call rte_eth_dev_detach()."
>
> Unfortunately testpmd calls rte_*eal*_dev_detach() not
> rte_*eth*_dev_detach(), the former doesn't release ethdev ports while the
> latter does. I think it's a mistake, there's no point in keeping a closed
> device around after it's been physically unplugged.
>
> In short, rmv_event_callback() should call detach_port() instead of
> rte_eal_dev_detach().
>
This is correct, the issue is actually testpmd making a mistake when
calling directly rte_eal_dev_detach().
The fix should thus be simply to change this call.
> > Fixes: 61cbdd419478 ("net/mlx4: separate device control functions")
> > Cc: sta...@dpdk.org
> >
> > Signed-off-by: Ophir Munk <ophi...@mellanox.com>
>
> The above suggests the problem is actually in testpmd and was introduced in
> v17.05 by commit 284c908cc588. The proposed patch is a workaround that
> doesn't address the underlying issue, thus NACK unless proven otherwise :)
>
> > ---
> > drivers/net/mlx4/mlx4_ethdev.c | 3 +++
> > 1 file changed, 3 insertions(+)
> >
> > diff --git a/drivers/net/mlx4/mlx4_ethdev.c b/drivers/net/mlx4/mlx4_ethdev.c
> > index 3bc6927..cca5223 100644
> > --- a/drivers/net/mlx4/mlx4_ethdev.c
> > +++ b/drivers/net/mlx4/mlx4_ethdev.c
> > @@ -67,6 +67,9 @@ mlx4_get_ifname(const struct priv *priv, char
> > (*ifname)[IF_NAMESIZE])
> > char match[IF_NAMESIZE] = "";
> >
> > {
> > + if (priv->ctx == NULL)
> > + return -ENOENT;
> > +
> > MKSTR(path, "%s/device/net", priv->ctx->device->ibdev_path);
> >
> > dir = opendir(path);
> > --
> > 2.7.4
> >
>
> --
> Adrien Mazarguil
> 6WIND
--
Gaëtan Rivet
6WIND
