1064875 - Consider replacing the FilterNodeCompositeSoftware code with Skia compositing through DrawTargetSkia

Status: ASSIGNED

(Core :: Graphics, defect)

Description

[Markus Stange [:mstange]]

Attachment: v1

I think the only argument against this is that the pixman code might not be safe against timing attacks. But if that's the case then we're already vulnerable through more common uses of the code, e.g. through normal web page compositing.

Comment 1

[Markus Stange [:mstange]]

Attachment: v2

Comment 2

[Bas Schouten (:bas.schouten)]

How about timing attacks with Skia?

Comment 3

[Markus Stange [:mstange]]

Comment on attachment 8498664 [details] [diff] [review]
v2

Right, I still need to check that.

Comment 4

[Jonathan Watt [:jwatt]]

Markus, any update on this? Could you unbitrot your patch?

Comment 5

[Markus Stange [:mstange]]

No update, but sure, I can update it.
As for the timing attacks, it's unlikely that Skia is vulnerable to them. Since we're using Skia composition for content drawing on some platforms now (and so is Chrome), we'd be vulnerable already.

Comment 6

[Markus Stange [:mstange]]

For the record: This patch will probably not improve the performance of the feComposite filter on desktop. Last I checked, the SSE2 code I wrote was roughly competitive with Skia.
I did not write a NEON version of the composite filter, but Skia has NEON versions for this, so this patch will definitely improve performance on Android.