[RFC 0/5] fix data+OOB writes, add ioctl

[Ricard Wanderlof]

On Tue, 23 Aug 2011, Brian Norris wrote:

> On Mon, Aug 22, 2011 at 11:48 PM, Ricard Wanderlof
> <ricard.wanderlof at axis.com> wrote:
>>>>> On Wed, 2011-08-17 at 16:50 -0700, Brian Norris wrote:
>>>>>> 1) Broken "noecc" writes; `nandwrite -n -o' does not work on my
>>>>>>   hardware, at least, without these patches
>>
>> This particular issue was fixed in a patch by Peter Wippich on June 6th:
>>
>> http://lists.infradead.org/pipermail/linux-mtd/2011-June/036016.html
>
> Right, I actually found this patch already and have it lined up in my
> test kernel, but apparently my issue is separate.
>
>> I could test patch 1 of this set to see if it accomplishes the same thing,
>> but initially to me it seems that the issue has already been fixed, albeit
>> perhaps not pushed yet?
>
> I am now doubting that my fix will solve your problem like Peter
> Wippich's patch did. I believe the problem I am having is specifically
> related to my NAND controller's characteristics (how it handles ECC).
>
> So for hardware on which Peter's fix solves your problems, I suppose I
> would be interested mostly in seeing if my patch *breaks* your build.
> FYI, I already see an issue for those who use both the builtin
> `nand_write_page_raw()' and `nand_write_buf()' functions. It can be
> fixed if you amend my patch with the following. I can resend later,
> but test these for breakage if you can!

Ok. I tried applying patch #1 of your patch set, to a tree in which 
Peter's patch has already been applying, to see what would happen. I had a 
problem in that in the mtdchar.c I have it looks like this:

      memset(chip->oob_poi, 0xff, mtd->oobsize);
      nand_fill_oob(chip, ops->oobbuf, ops->ooblen, ops);
      status = chip->ecc.write_oob(mtd, chip, page & chip->pagemask);
      memset(chip->oob_poi, 0xff, mtd->oobsize);

whereas your patch looks like it was made against a version which lacks 
the memsets. First I thought it was because I was running an older kernel 
(2.6.35), but I looked at HEAD of the linux-2.6 and mtd-2.6 trees at 
git.infradead.org, and it's the same there. So I'm not sure exactly which 
version your patch was made against. Perhaps it's obvious to someone but 
not me right now.

Nevertheless, after applying the patch, as you suspected, using 'nandwrite 
-o -n' fails, in this case the application hangs after outputting

Writing data to block 0

After applying the following patch from your email, 'nandwrite -o -n' 
initially appears to work, however upon closer inspection the oob's of all 
the written pages are all-ff, i.e. the oob's are never written.

> --- a/drivers/mtd/nand/nand_base.c
> +++ b/drivers/mtd/nand/nand_base.c
> @@ -1898,7 +1898,8 @@ out:
> static void nand_write_page_raw(struct mtd_info *mtd, struct nand_chip *chip,
>                                const uint8_t *buf)
> {
> -       chip->write_buf(mtd, buf, mtd->writesize);
> +       if (buf)
> +               chip->write_buf(mtd, buf, mtd->writesize);
>        chip->write_buf(mtd, chip->oob_poi, mtd->oobsize);
> }
>

Now, this may all be nonsense as as I mentioned I'm not sure that I've had 
the right version to start with. The memsets bug me, as they would explain 
the all-ff's stuff, but I don't really feel like just experimenting.

/Ricard
-- 
Ricard Wolf Wanderlöf                           ricardw(at)axis.com
Axis Communications AB, Lund, Sweden            www.axis.com
Phone +46 46 272 2016                           Fax +46 46 13 61 30