Hello Robin,
On Sun, Jun 16, 2019 at 10:02 PM Robin Gong <yibin.gong@xxxxxxx> wrote:
>
> The default imx defconfig and dts should be ok, because firmware load
> is delayed after rootfs mounted where firmware located in and before
> that, some driver which use sdma such as spi/uart/audio may have
> already enable sdma clock which means channel0 interrupt could be
> cleared immediately without interrupt storm. That's why I can't
> reproduce your issue at first, but catch it once I sync with your
> directly firmware load defconfig. So seems not very must to CC stable
> tree?
As far as I know, the bug/crash does not happen if you're loading the
sdma firmware from a filesystem. So the vast majority of users would
never see the crash.
I agree that this is not a high-priority bugfix. But it's worthwhile for the
stable trees to have it.
> Yes, but Michael's patch is the right direction, at least it fix RT
> case and meaningless channel0 interrupt storm coming before clearing
> channel0 interrupt status in sdma_run_channel0(). Now, this patch could
> fix its minor side-effect.
I'm not suggesting that we should revert or change Michael's patch. Just
that it would be good for the v2 patch to contain:
Fixes: 1d069bfa3c78 ("dmaengine: imx-sdma: ack channel 0 IRQ in the
interrupt handler")
This should allow stable maintainers to pull in your patch if and only if
their release already contains 1d069bfa3c78.
