[PATCH] drm/mediatek/dp: fix spurious kfree()

[Michael Walle]

On Fri May 17, 2024 at 1:09 PM CEST, AngeloGioacchino Del Regno wrote:
> Il 17/05/24 13:07, Michael Walle ha scritto:
> > On Fri May 17, 2024 at 12:35 PM CEST, AngeloGioacchino Del Regno wrote:
> >> Il 17/05/24 11:30, Michael Walle ha scritto:
> >>> drm_edid_to_sad() might return an error or just zero. If that is the
> >>> case, we must not free the SADs because there was no allocation in
> >>> the first place.
> >>>
> >>> Fixes: dab12fa8d2bd ("drm/mediatek/dp: fix memory leak on ->get_edid callback audio detection")
> >>> Signed-off-by: Michael Walle <mwalle at kernel.org>
> >>> ---
> >>>    drivers/gpu/drm/mediatek/mtk_dp.c | 10 ++++++++--
> >>>    1 file changed, 8 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/drivers/gpu/drm/mediatek/mtk_dp.c b/drivers/gpu/drm/mediatek/mtk_dp.c
> >>> index 536366956447..ada12927bbac 100644
> >>> --- a/drivers/gpu/drm/mediatek/mtk_dp.c
> >>> +++ b/drivers/gpu/drm/mediatek/mtk_dp.c
> >>> @@ -2073,9 +2073,15 @@ static const struct drm_edid *mtk_dp_edid_read(struct drm_bridge *bridge,
> >>>    		 */
> >>>    		const struct edid *edid = drm_edid_raw(drm_edid);
> >>>    		struct cea_sad *sads;
> >>> +		int ret;
> >>>    
> >>> -		audio_caps->sad_count = drm_edid_to_sad(edid, &sads);
> >>> -		kfree(sads);
> >>> +		ret = drm_edid_to_sad(edid, &sads);
> >>> +		/* Ignore any errors */
> >>> +		if (ret < 0)
> >>> +			ret = 0;
> >>> +		if (ret)
> >>
> >> Eh, this will never work, because you're clearing the error before checking
> >> if there's any error here?!?! :-P
> > 
> > Don't get what you mean? Yes, I'm ignoring the error. Thus, in case
> > of an error ret will be zero and there will be no free. If ret was
> > zero, there won't be a free either. So you're left with the "normal"
> > case, where you have to free the sads. Just like before.
> > 
> >> Anyway in reality, it returns -ENOMEM if the allocation was not successful...
> >> in the event that any future update adds any other error we'd be back with the same
> >> issue, but I'm not sure how much should we worry about that.
> >>
> >> To be extremely safe, we could do...
> >>
> >> if (ret != -ENOMEM)
> >> 	kfree(sads)
> >>
> >> audio_caps->sad_count = ret < 0 ? 0 : ret;
> > 
> > Which is the same as above, but you only check for ENOMEM?
> > 
>
> Yes, the point is to avoid kfree(sads) for -ENOMEM only, as other errors that may
> be introduced later might still allocate it and leave it allocated.

Honestly, I doubt that any sane function will allocate memory, then
return an error and expect the caller to free it.

-michael
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 297 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20240517/6391b1a7/attachment-0001.sig>