On di, 16 jun 2020 09:51:00 +0000, Felix Riemann wrote:
> Hi Kurt,
>
> > -----Original Message-----
> > From: Kurt Van Dijck <dev.kurt@xxxxxxxxxxxxxxxxxxxxxx>
> > Sent: Monday, June 15, 2020 5:31 PM
> > Subject: Re: [PATCH] can: c_can: Handle lost bus-off interrupt while IRQs are
> > disabled
> >
> > On ma, 15 jun 2020 11:11:17 +0000, Felix Riemann wrote:
> > >
> > > Hi Kurt,
> > >
> > > see below.
> > >
> > > > -----Original Message-----
> > > >
> > > > Hi Felix,
> > > >
> > > > Next question comes up.
> > > > See inline.
> > > >
> > > > On vr, 12 jun 2020 09:10:44 +0200, Kurt Van Dijck wrote:
> > > > > Thanks for verifying,
> > > > >
> > > > > I'll take a look how that can happen.
> > > > >
> > > > > I think packet loss is better (less packets lost) than before.
> > > > >
> > > > > Kurt
> > > > >
> > > > > On wo, 10 jun 2020 08:08:28 +0000, Felix Riemann wrote:
> > > > > > Hi Kurt,
> > > > > >
> > > > > > > -----Original Message-----
> > > > > > >
> > > > > > > I have the impression that the recv path is covered, at least
> > > > > > > the null ptr would have caused issued earlier already.
> > > > > > > Would the problem arise in c_can_do_tx fetching a NULL echo_skb?
> > > > > >
> > > > > > It really looks like it.
> > > > > >
> > > > > > Adding a BUG_ON to the loop in c_can_do_tx triggers with a NULL
> > > > > > skb
> > > > returned by __can_get_echo_skb().
> > > > > >
> > > > > > For testing I replaced the BUG_ON with a check that skips the
> > > > > > skb if that
> > > > happens (not sure if that would be a valid fix in that place):
> > > > > >
> > > > > > while ((obj = c_can_ffs64(pend))) { pend &= ~((u64)1 << (obj -
> > > > > > 1)); c_can_inval_tx_object(dev, IF_RX, obj); idx = obj -
> > > > > > priv->obj.send_frst; skb = __can_get_echo_skb(dev, idx, &len);
> > > > > > if(!skb)
> > > >
> > > > It would be interesting to know idx and priv->tx_active here.
> > >
> > > That's the statistic over about 1600 lines I captured. First number is number
> > of occurrences:
> >
> > just double-check:
> > These are all when __can_get_echo_skb returned NULL?
>
> Yes, that's correct. The printout occurs directly before it skips to the next item in the loop.
thanks.
