Re: [PATCH 0/4] mtd: add support for pairing scheme description

[Brian Norris]

On Sat, Jun 11, 2016 at 08:45:18AM +0200, Boris Brezillon wrote:
> On Fri, 10 Jun 2016 19:16:25 -0700
> Brian Norris <computersforpeace@xxxxxxxxx> wrote:
> > On Mon, Apr 25, 2016 at 12:01:17PM +0200, Boris Brezillon wrote:
> > > Hi,
> > > 
> > > This series is the first step towards reliable MLC/TLC NAND support.
> > > Those patches allows the NAND layer to expose page pairing information
> > > to MTD users.  
> > 
> > Have you surveyed many types of NAND to get a representative sampling of
> > what kind of pairing schemes are out there? Do you think you've covered
> > the possibilities well enough in your API? I have a few comments on the
> > patches to this effect. I honestly don't know the answer to these
> > questions, because AFAIR, this is rarely well documented in datasheets.
> 
> I only tested on 3 different NANDs from Micron, Toshiba and Hynix, but

I'm curious, do you have an example part number for Micron? When I
looked briefly last week, I only found either MLC that don't mention it
at all (they fundamentally *have* to have write pairing, don't they?) or
TLC that required too much work for me to get past their login screens.

> I had a look at several datasheets. Unlike read-retry this part is
> usually documented in public datasheets, and on a panel of approximately
> 20 NANDs (mainly from Toshiba, Samsung, Hynix and Micron), all of them
> where using the 'distance 3' or 'distance 6' pairing scheme.
> The only exception I've seen so far is the one pointed by Bean here [1],
> and it can be described using the mtd_pairing_scheme approach.

Yeah, I suppose the API is rather generic. It doesn't really assume
anything about patterns/distances -- just that the pairings are formed
in groups of the same size.

> > > The plan is to teach UBI about those constraints and let UBI code take
> > > the appropriate precautions when dealing with those multi-level cells
> > > NANDs. The way we'll handle this "paired pages" constraint will be
> > > described soon in a series adapting the UBI layer, so stay tune ;).
> > > 
> > > Note that this implementation only allows page pairing scheme description
> > > when the NAND has a full-id entry in the nand_ids table.
> > > This should be addressed in some way for ONFI and JEDEC NANDs, though
> > > I'm not sure how to handle this yet.  
> > 
> > Do ONFI or JEDEC parameter pages even provide this kind of info? The
> > ONFI spec doesn't mention paired pages.
> 
> Nope that's the problem. The only way you can deduce that is to extract
> it from other information, but I think my series reworking the NAND
> initialization will help us [2].

Sure, I suppose.

Brian