Re: [PATCH 6/6] dmaengine: tegra-apb: Disable interrupts on removal

From: Jon Hunter <jonathanh@nvidia.com>
To: Vinod Koul <vinod.koul@intel.com>
Cc: Lars-Peter Clausen <lars@metafoo.de>,
	Laxman Dewangan <ldewangan@nvidia.com>,
	Stephen Warren <swarren@wwwdotorg.org>,
	"Thierry Reding" <thierry.reding@gmail.com>,
	Alexandre Courbot <gnurou@gmail.com>, <dmaengine@vger.kernel.org>,
	<linux-tegra@vger.kernel.org>, <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 6/6] dmaengine: tegra-apb: Disable interrupts on removal
Date: Wed, 28 Oct 2015 13:34:06 +0000	[thread overview]
Message-ID: <5630CECE.2000600@nvidia.com> (raw)
In-Reply-To: <20151028065339.GE3041@vkoul-mobl.iind.intel.com>

On 28/10/15 06:53, Vinod Koul wrote:
> On Fri, Oct 16, 2015 at 11:57:02AM +0100, Jon Hunter wrote:
>>>>> How about just calling free_irq()? That's how you'd typically handle this.
>>>>
>>>> Yes, however, the interrupt is requested by devm_request_irq(). I guess
>>>> I could call devm_free_irq() here?
>>>
>>> Just use request_irq() instead of devm_request_irq(). You have the same
>>> issue on the error path in the probe function anyway and also need to add
>>> the free_irq() before the tasklet_kill() there as well.
>>
>> I was wondering about that but the tasklets should never be scheduled if
>> the probe does not succeed, so I think it is ok.
> 
> This is actually very racy, if probe fails but due to devm_ calls your irq
> is alive till it freed by core
> 
> And a faulty device triggering irq can complicate matters, so for irq IMHO
> we don't get much benefit with devm_ variant

That's fine, I will drop the devm_ usage here then.

Jon