On Fri, Feb 03, 2023 at 05:01:21PM +0100, Luca Ceresoli wrote:
> Hello Johan,
>
> On Wed, 1 Feb 2023 11:15:40 +0100
> Johan Hovold <johan+linaro@xxxxxxxxxx> wrote:
>
> > The current interconnect provider registration interface is inherently
> > racy as nodes are not added until the after adding the provider. This
> > can specifically cause racing DT lookups to fail.
> >
> > Switch to using the new API where the provider is not registered until
> > after it has been fully initialised.
> >
> > Fixes: f0d8048525d7 ("interconnect: Add imx core driver")
> > Cc: stable@xxxxxxxxxxxxxxx # 5.8
> > Cc: Leonard Crestez <leonard.crestez@xxxxxxx>
> > Cc: Alexandre Bailon <abailon@xxxxxxxxxxxx>
> > Signed-off-by: Johan Hovold <johan+linaro@xxxxxxxxxx>
>
> Georgi pointed me to this series after I reported a bug yesterday [0],
> that I found on iMX8MP. So I ran some tests with my original, failing
> tree, minus one patch with my debugging code to hunt for the bug, plus
> patches 1-4 of this series.
>
> The original code was failing approx 5~10% of the times. With your 4
> patches applied it ran 139 times with zero errors, which looks great! I
> won't be able to do more testing until next Monday to be extra sure.
Thanks for testing.
It indeed looks like you're hitting the same race, and as the imx
interconnect driver also initialises the provider data num_nodes count
before adding the nodes it results in that NULL-deref (where the qcom
driver failed a bit more gracefully).
Johan
> [0]
> https://lore.kernel.org/linux-arm-kernel/20230202175525.3dba79a7@booty/T/#u
