Re: [PATCH] video: fbdev: pxa3xx_gcu: Fix some resource leak in an error handling path in 'pxa3xx_gcu_probe()'

[Christophe JAILLET <christophe.jaillet@wanadoo.fr>]

Le 29/04/2020 à 14:25, Dan Carpenter a écrit :
> On Wed, Apr 29, 2020 at 06:34:38AM +0200, Christophe JAILLET wrote:
>> If an error occurs in the loop where we call 'pxa3xx_gcu_add_buffer()',
>> any resource already allocated should be freed.
>>
>> In order to fix it, add a call to 'pxa3xx_gcu_free_buffers()' in the error
>> handling path, as already done in the remove function.
>>
>> Fixes: 364dbdf3b6c3 ("video: add driver for PXA3xx 2D graphics accelerator")
>> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
>> ---
>>   drivers/video/fbdev/pxa3xx-gcu.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/video/fbdev/pxa3xx-gcu.c b/drivers/video/fbdev/pxa3xx-gcu.c
>> index 4279e13a3b58..68d9c7a681d4 100644
>> --- a/drivers/video/fbdev/pxa3xx-gcu.c
>> +++ b/drivers/video/fbdev/pxa3xx-gcu.c
>> @@ -675,6 +675,7 @@ static int pxa3xx_gcu_probe(struct platform_device *pdev)
>>   
>>   err_disable_clk:
>>   	clk_disable_unprepare(priv->clk);
>> +	pxa3xx_gcu_free_buffers(dev, priv);
> The error handling in this function makes no sense and is buggy.  It
> should be that it unwinds in the reverse order from the allocation.  The
> goto should be "goto free_most_recently_allocated_resource;".  Since the
> unwind is done in the wrong order it causes a couple bugs.
>
> These buffers are the last thing which we allocated so they should be
> the first thing which we free.  In this case, calling
> pxa3xx_gcu_free_buffers() before the buffers are allocated is confusing
> but harmless.  The clk_disable_unprepare() is done on some paths where
> the clock was not enabled and that will trigger a WARN() so that's a
> bug.  Syzcaller will complain and if you have reboot on WARN then it's
> annoying.
>
> The second bug is that we don't deregister the misc device or release
> the DMA memory on failure when we allocate the buffers in the loop.
>
> regards,
> dan carpenter
>
Agreed. I've been a little too fast on this one.
I'll update it.

Thx for the review.

CJ