Re: [dpdk-stable] [PATCH] net/vdev_netvsc: Prevent alarm lost on failed device probe

[Long Li <longli@microsoft.com>]

> Subject: RE: [PATCH] net/vdev_netvsc: Prevent alarm lost on failed device
> probe
> 
> 
> 
> From: Long Li
> > > Subject: RE: [PATCH] net/vdev_netvsc: Prevent alarm lost on failed
> > > device probe
> > >
> > >
> > >
> > > From: Long Li
> > > > If a device probe fails, the alarm is canceled and will no longer
> > > > work for previously probed devices.
> > > >
> > > > Fix this by introducing a flag to track if alarm has been set.
> > > > Because it's possible that an alarm is triggered while probing is
> > > > in progress that may modify vdev_netvsc_ctx_list, introduce a lock
> > > > to
> > protect it.
> > > >
> > > > Cc: stable@dpdk.org
> > > > Signed-off-by: Long Li <longli@microsoft.com>
> > > > ---
> > > >  drivers/net/vdev_netvsc/vdev_netvsc.c | 41
> > > > +++++++++++++++++++++++++----------
> > > >  1 file changed, 30 insertions(+), 11 deletions(-)
> > > >
> > > > diff --git a/drivers/net/vdev_netvsc/vdev_netvsc.c
> > > > b/drivers/net/vdev_netvsc/vdev_netvsc.c
> > > > index be8f19c..bd7e308 100644
> > > > --- a/drivers/net/vdev_netvsc/vdev_netvsc.c
> > > > +++ b/drivers/net/vdev_netvsc/vdev_netvsc.c
> > > > @@ -76,6 +76,11 @@ struct vdev_netvsc_ctx {
> > > >  /** Context list is common to all driver instances. */  static
> > > > LIST_HEAD(,
> > > > vdev_netvsc_ctx) vdev_netvsc_ctx_list =
> > > >         LIST_HEAD_INITIALIZER(vdev_netvsc_ctx_list);
> > > > +/* Lock to protect concurrent accesses to vdev_netvsc_ctx_list */
> > > > +static rte_rwlock_t vdev_netvsc_ctx_list_lock;
> > > > +
> > > > +/* Flag to track if alarm has been set */ static int
> > > > +vdev_netvsc_alarm_set;
> > > >
> > > >  /** Number of entries in context list. */  static unsigned int
> > > > vdev_netvsc_ctx_count; @@ -454,19 +459,26 @@ static LIST_HEAD(,
> > > > vdev_netvsc_ctx) vdev_netvsc_ctx_list =
> > > >         struct vdev_netvsc_ctx *ctx;
> > > >         int ret;
> > > >
> > > > +       rte_rwlock_write_lock(&vdev_netvsc_ctx_list_lock);
> > > >         LIST_FOREACH(ctx, &vdev_netvsc_ctx_list, entry) {
> > > >                 ret = vdev_netvsc_foreach_iface(vdev_netvsc_device_probe,
> 0,
> > > >                       ctx);
> > > >                 if (ret < 0)
> > > >                         break;
> > > >         }
> > > > -       if (!vdev_netvsc_ctx_count)
> > > > +       rte_rwlock_write_unlock(&vdev_netvsc_ctx_list_lock);
> > > > +
> > > > +       if (!vdev_netvsc_ctx_count) {
> > > > +               vdev_netvsc_alarm_set = 0;
> > > >                 return;
> > > > +       }
> > > > +
> > > >         ret = rte_eal_alarm_set(VDEV_NETVSC_PROBE_MS * 1000,
> > > >                                 vdev_netvsc_alarm, NULL);
> > > >         if (ret < 0) {
> > > >                 DRV_LOG(ERR, "unable to reschedule alarm callback: %s",
> > > >                         rte_strerror(-ret));
> > > > +               vdev_netvsc_alarm_set = 0;
> > > >         }
> > > >  }
> > > >
> > > > @@ -698,34 +710,41 @@ static LIST_HEAD(, vdev_netvsc_ctx)
> > > > vdev_netvsc_ctx_list =
> > > >                         " device.");
> > > >                 goto error;
> > > >         }
> > > > -       rte_eal_alarm_cancel(vdev_netvsc_alarm, NULL);
> > > > +
> > > > +       rte_rwlock_write_lock(&vdev_netvsc_ctx_list_lock);
> > > >         /* Gather interfaces. */
> > > >         ret = vdev_netvsc_foreach_iface(vdev_netvsc_netvsc_probe,
> > > > 1,
> > > name,
> > > >                                         kvargs, specified, &matched);
> > > >         if (ret < 0)
> > > > -               goto error;
> > > > +               goto error_unlock;
> > > >         if (specified && matched < specified) {
> > > >                 if (!force) {
> > > >                         DRV_LOG(ERR, "Cannot find the specified netvsc device");
> > > > -                       goto error;
> > > > +                       goto error_unlock;
> > > >                 }
> > > >                 /* Try to force probing on non-netvsc specified device. */
> > > >                 if
> > > > (vdev_netvsc_foreach_iface(vdev_netvsc_netvsc_probe, 0,
> > > name,
> > > >                                               kvargs, specified, &matched) < 0)
> > > > -                       goto error;
> > > > +                       goto error_unlock;
> > > >                 if (matched < specified) {
> > > >                         DRV_LOG(ERR, "Cannot find the specified device");
> > > > -                       goto error;
> > > > +                       goto error_unlock;
> > > >                 }
> > > >                 DRV_LOG(WARNING, "non-netvsc device was probed as
> > netvsc");
> > > >         }
> > > > -       ret = rte_eal_alarm_set(VDEV_NETVSC_PROBE_MS * 1000,
> > > > +       if (!vdev_netvsc_alarm_set) {
> > > > +               ret = rte_eal_alarm_set(VDEV_NETVSC_PROBE_MS *
> > > > + 1000,
> > > >                                 vdev_netvsc_alarm, NULL);
> > > > -       if (ret < 0) {
> > > > -               DRV_LOG(ERR, "unable to schedule alarm callback: %s",
> > > > -                       rte_strerror(-ret));
> > > > -               goto error;
> > > > +               if (ret < 0)
> > > > +                       DRV_LOG(ERR, "unable to schedule alarm callback: %s",
> > > > +                               rte_strerror(-ret));
> > > > +               else
> > > > +                       vdev_netvsc_alarm_set = 1;
> > > >         }
> > > > +
> > > > +error_unlock:
> > > > +       rte_rwlock_write_unlock(&vdev_netvsc_ctx_list_lock);
> > > > +
> > > >  error:
> > > >         if (kvargs)
> > > >                 rte_kvargs_free(kvargs);
> > > > --
> > >
> > > Hi
> > >
> > > Nice direction.
> > >
> > > As I understand from your patch it looks like you want to add
> > > support for the vdev_netvsc driver itself to be hot-plug aware - so
> > > it can be probed\removed in run-time (even after EAL initialization).
> > > So I think the title and commit log should be converted to this:
> > > net/vdev_netvsc: support driver instance hot-plug ...
> >
> > Hi Matan,
> >
> > Actually I was not trying to add support for hot-plug (although it
> > made half way there).
> >
> > The problem I'm trying to solve is on a system with multiple
> > vdev_netvsc interfaces. For example, if the system has eth1, eth2 ..
> > eth10. What if the probing on eth1 to eth9 succeed, but on eth10
> > fails? Current behavior is that we can still have DPDK run on eth1 to
> > eth9, but without any alarms. Because the alarm is canceled while
> > trying to probing eth10. Then once eth1 to eth9 lost VF, they won't get VF
> back because we don't have any alarms.
> >
> > Long
> 
> Ok, got you.
> 
> So, why to manage critical sections in the code, better, to stop alarm before
> adding a new device, and apply it back if the list of devices is not empty.
> It will make it simpler, no?

Okay, I will send v2 based on your suggestion. Thanks.

> 
> 
> 
> > >
> > > More things:
> > > Look on convention of the file - blank line only after variables
> > > declarations inside the blocks.
> > >
> > > Need to adjust also vdev_netvsc_vdev_remove to the feature.
> > >
> > > Matan