[PATCH v4 0/8] arm64/sve: First steps towards optimizing syscalls

[Will Deacon]

On Mon, Sep 21, 2020 at 01:56:28PM +0100, Mark Brown wrote:
> On Mon, Sep 21, 2020 at 01:42:13PM +0100, Will Deacon wrote:
> 
> > Having three flags to track the fp state and then a bunch of WARN()s
> > checking for invalid combinations is quite brittle, so any documentation
> > that can help to justify this would certainly be useful!
> 
> > I've left a couple of comments on some of the patches, but it looks like
> > Dave was reviewing them but stopped short of the meat and potatoes in the
> > last two patches. I'd like to see his Ack on those before picking them up,
> > as well as testing from somebody with hardware because this is _very_
> > subtle stuff.
> 
> Right.  The previous version was tested on hardware but I dropped the
> Tested-by since I felt there were more changes than I was comfortable
> with.  I have to say that a bunch of the things you've flagged up were
> things that were requested on previous rounds of review.

If you have links to specifics, I'm happy to take a look. I like what this
patch series is trying to do, but the implementation is piling complexity
on top of something that is already horribly complicated and I don't
immediately see the justification for why that is necessary.

> > Is it worth me picking some of the preparatory patches up on their own?
> 
> I think so, yes - it'd make the series easier to manage and mean there's
> less to redo per-patch validation on each time if nothing else.  They
> don't do any harm and seem like they'd be useful even if a completely
> different approach is adopted.

Ok, I'll see if I can reduce this a bit then.

Will