Re: [PATCH] media: venus: hfi: explicitly release IRQ during teardown

[Jorge Ramirez <jorge.ramirez@oss.qualcomm.com>]

On 18/06/25 12:08:28, Bryan O'Donoghue wrote:
> On 18/06/2025 08:07, Jorge Ramirez wrote:
> > On 16/06/25 17:32:38, Jorge Ramirez wrote:
> > > On 16/06/25 20:14:36, Vikash Garodia wrote:
> > > > Hi Jorge,
> > > > 
> > > > On 6/16/2025 6:29 PM, Jorge Ramirez wrote:
> > > > > On 16/06/25 17:26:24, Vikash Garodia wrote:
> > > > > > 
> > > > > > On 6/12/2025 1:59 PM, Jorge Ramirez-Ortiz wrote:
> > > > > > > Ensure the IRQ is released before dismantling the ISR handler and
> > > > > > > clearing related pointers.
> > > > > > > 
> > > > > > > This prevents any possibility of the interrupt triggering after the
> > > > > > > handler context has been invalidated.
> > > > > > > 
> > > > > > > Fixes: d96d3f30c0f2 ("[media] media: venus: hfi: add Venus HFI files")
> > > > > > > Signed-off-by: Jorge Ramirez-Ortiz <jorge.ramirez@oss.qualcomm.com>
> > > > > > > ---
> > > > > > >   drivers/media/platform/qcom/venus/hfi_venus.c | 1 +
> > > > > > >   1 file changed, 1 insertion(+)
> > > > > > > 
> > > > > > > diff --git a/drivers/media/platform/qcom/venus/hfi_venus.c b/drivers/media/platform/qcom/venus/hfi_venus.c
> > > > > > > index b5f2ea879950..d9d62d965bcf 100644
> > > > > > > --- a/drivers/media/platform/qcom/venus/hfi_venus.c
> > > > > > > +++ b/drivers/media/platform/qcom/venus/hfi_venus.c
> > > > > > > @@ -1678,6 +1678,7 @@ void venus_hfi_destroy(struct venus_core *core)
> > > > > > >   	venus_interface_queues_release(hdev);
> > > > > > >   	mutex_destroy(&hdev->lock);
> > > > > > >   	kfree(hdev);
> > > > > > > +	devm_free_irq(core->dev, core->irq, core);
> > > > > > Could you please check and add the handling here [1] as well ?
> > > > > > 
> > > > > > [1]
> > > > > > https://elixir.bootlin.com/linux/v6.16-rc1/source/drivers/media/platform/qcom/venus/core.c#L427
> > > > > 
> > > > > hi Vikash, sorry I dont get your point - what do you mean?
> > > > IRQ need to be freed even for error cases during venus_probe().
> > > > 
> > > 
> > > but  this is what the current patch does (venus_hfi_destroy is called at
> > > the end of probe error handling as well).
> > > 
> > 
> > for background, this fixes a null derreference in the Venus driver -
> > reproduceable in RB3Gen2 on a particular error condition during probe.
> Shouldn't it be the case that devm removes the handler for us anyway ?
> 
> Why not ->         disable_irq_nosync(core->irq);

I agree, this seems better to me too.

I guess disable_irq() is the safer/more meaningfull choice since we are
calling from non irq context.

will fix - thanks for the suggestion!

> 
> i.e. disable the IRQ until the normal/expected exit path removes it.
> 
> ---
> bod