Re: [PATCH] z3fold: encode object length in the handle

[Andrew Morton]

On Mon, 29 Oct 2018 13:27:36 +0100 Vitaly Wool <vitalywool@xxxxxxxxx> wrote:

> Hi Andrew,
> 
> Den tors 25 okt. 2018 kl 21:42 skrev Andrew Morton <
> akpm@xxxxxxxxxxxxxxxxxxxx>:
> 
> > On Thu, 25 Oct 2018 11:28:21 +0200 Vitaly Wool <vitalywool@xxxxxxxxx>
> > wrote:
> >
> > > Reclaim and free can race on an object (which is basically ok) but
> > > in order for reclaim to be able to  map "freed" object we need to
> > > encode object length in the handle. handle_to_chunks() is thus
> > > introduced to extract object length from a handle and use it during
> > > mapping of the last object we couldn't correctly map before.
> >
> > What are the runtime effects of this change?
> >
> 
> I haven't observed any adverse impact with this change used in zswap (and
> in fact, this is a bugfix for zswap operation). There is a slight under 1%
> impact when z3fold is used with ZRAM but since the support for ZRAM over
> zpool is still out of tree, I take it doesn't matter at this point, right?
> 

I mean "runtime effects", not "run time effects" ;)

Apart from wishing to document this change fully, I'm trying to
understand which kernel version(s) need the fix.  To understand that, I
need to know the effect upon end-user-visible behaviour.  You say it
fixes a bug - please describe that bug: how it is triggered, what
effect is has, etc.

Also, any suggestions as to which kernel versions we should fix is
always welcome.