Re: [PATCH v2 2/2] mm, sl[aou]b: guarantee natural alignment for kmalloc(power-of-two)

[Darrick J. Wong]

On Wed, Sep 25, 2019 at 07:53:53AM +1000, Dave Chinner wrote:
> On Tue, Sep 24, 2019 at 11:19:29PM +0200, Vlastimil Babka wrote:
> > On 9/23/19 7:51 PM, Darrick J. Wong wrote:
> > > On Mon, Sep 23, 2019 at 07:17:10PM +0200, David Sterba wrote:
> > >> On Mon, Sep 23, 2019 at 06:36:32PM +0200, Vlastimil Babka wrote:
> > >>> So if anyone thinks this is a good idea, please express it (preferably
> > >>> in a formal way such as Acked-by), otherwise it seems the patch will be
> > >>> dropped (due to a private NACK, apparently).
> > > 
> > > Oh, I didn't realize  ^^^^^^^^^^^^ that *some* of us are allowed the
> > > privilege of gutting a patch via private NAK without any of that open
> > > development discussion incovenience. <grumble>
> > > 
> > > As far as XFS is concerned I merged Dave's series that checks the
> > > alignment of io memory allocations and falls back to vmalloc if the
> > > alignment won't work, because I got tired of scrolling past the endless
> > > discussion and bug reports and inaction spanning months.
> > 
> > I think it's a big fail of kmalloc API that you have to do that, and
> > especially with vmalloc, which has the overhead of setting up page
> > tables, and it's a waste for allocation requests smaller than page size.
> > I wish we could have nice things.
> 
> I don't think the problem here is the code. The problem here is that
> we have a dysfunctional development community and there are no
> processes we can follow to ensure architectural problems in core
> subsystems are addressed in a timely manner...
> 
> And this criticism isn't just of the mm/ here - this alignment
> problem is exacerbated by exactly the same issue on the block layer
> side. i.e. the block layer and drivers have -zero- bounds checking
> to catch these sorts of things and the block layer maintainer will
> not accept patches for runtime checks that would catch these issues
> and make them instantly visible to us.
> 
> These are not code problems: we can fix the problems with code (and
> I have done so to demonstrate "this is how we do what you say is
> impossible").  The problem here is people in positions of
> control/power are repeatedly demonstrating an inability to
> compromise to reach a solution that works for everyone.
> 
> It's far better for us just to work around bullshit like this in XFS
> now, then when the core subsystems get they act together years down
> the track we can remove the workaround from XFS. Users don't care
> how we fix the problem, they just want it fixed. If that means we
> have to route around dysfunctional developer groups, then we'll just
> have to do that....

Seconded.

--D

> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> david@xxxxxxxxxxxxx