Re: [EXT] [PATCH v2 3/3] mtd: rawnand: micron: Address the shallow erase issue

From: Boris Brezillon <boris.brezillon@collabora.com>
To: Richard Weinberger <richard@nod.at>
Cc: Vignesh Raghavendra <vigneshr@ti.com>,
	Tudor Ambarus <Tudor.Ambarus@microchip.com>,
	Steve deRosier <derosier@gmail.com>,
	"Zoltan Szubbocsev, zszubbocsev" <zszubbocsev@micron.com>,
	linux-mtd <linux-mtd@lists.infradead.org>,
	Thomas Petazzoni <thomas.petazzoni@bootlin.com>,
	Miquel Raynal <miquel.raynal@bootlin.com>,
	tglx <tglx@linutronix.de>,
	Piotr Wojtaszczyk <WojtaszczykP@cumminsallison.com>,
	"Bean Huo, beanhuo" <beanhuo@micron.com>
Subject: Re: [EXT] [PATCH v2 3/3] mtd: rawnand: micron: Address the shallow erase issue
Date: Wed, 6 May 2020 23:11:16 +0200	[thread overview]
Message-ID: <20200506231116.2d16b747@collabora.com> (raw)
In-Reply-To: <130342272.192940.1588798779898.JavaMail.zimbra@nod.at>

On Wed, 6 May 2020 22:59:39 +0200 (CEST)
Richard Weinberger <richard@nod.at> wrote:

> ----- Ursprüngliche Mail -----
> >> UBI scans only headers during attach. If you don't touch these, no way.  
> > 
> > Sorry, I misunderstood what you meant by payload. UBI should schedule
> > the PEB for erase if the EC/VID header is corrupted.  
> 
> UBI even tries to recover from such a situation. If only the EC header is bad,
> it will create a new one. Only of the VID header is bad/missing and the payload
> is corrupted (ECC errors or bit-flips) it will erase it.

Yep, I remember that, though in the case of a corrupted EC, UBI will
schedule a PEB recovery, and if the payload is corrupted, the read
should fail when copying the LEB content to another block.

> 
> A missing VID header plus good payload will cause UBI to stop attaching since it
> violates the IO model.

Sure, and that's not what we want to do anyway. We basically have 2
choices here:

1/ overwrite all pages starting from page 0 and ending at page 15. This
will lead to ECC errors on already written pages, and 0-filled pages
for others (unless we go for a raw write, in which case it might
actually lead to ECC errors depending on the engine).
2/ track already written pages (by reading them first), and only writes
those that have not been written yet. That means no ECC error in that
case, and no corrupted EC/VID header as well.

Clearly, #2 would help mitigate the perf penalty incurred by the Micron
workaround, but if it's not an option, maybe we can just start with
#1 (which is basically what Miquel implemented).

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/