On Fri, 2024-02-16 at 17:52 +0530, Vinod Koul wrote:
> On 14-02-24, 13:29, Nuno Sa wrote:
> > In axi_dmac_probe(), there's a mix in using device managed APIs and
> > explicitly cleaning things in the driver .remove() hook. Move to use
> > device managed APIs and thus drop the .remove() hook.
> >
> > While doing this move request_irq() before of_dma_controller_register()
> > so the previous cleanup order in the .remove() hook is preserved.
>
> It is good that we are doing this, but moving irq to devm doesnt help a
> lot, there exists a race
>
> I would suggest you use the axi_dmac_free_dma_controller() to all free
> up irq explicitly, that will ensure things no chance of irq firing and
> scheduling tasklet when we are removing
>
Hi Vinod,
Hmm wait... Before the patch we had:
static void axi_dmac_remove(struct platform_device *pdev)
{
struct axi_dmac *dmac = platform_get_drvdata(pdev);
of_dma_controller_free(pdev->dev.of_node);
free_irq(dmac->irq, dmac);
So, you mean that there's actually a possible bug/race in the above? The irq is only
being released after of_dma_controller_free(). So, I guess I should not move
request_irq() at all and then I can still use devm_request_irq(). If I do it after
registering the device then it will be released before of_dma_controller_free(). Am I
missing something?
Thanks for the feedback!
- Nuno Sá
>
