Re: [PATCH] mtd: rawnand: stop BBT creation when the marker scan fails

From: James Hilliard

Date: Thu Nov 19 2026 - 04:35:57 EST


On Thu, Nov 19, 2026 at 1:38 AM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
>
> On 18/11/2026 at 12:11:19 -07, James Hilliard <james.hilliard1@xxxxxxxxx> wrote:
>
> > On Wed, Nov 18, 2026 at 2:45 AM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
> >>
> >> On 03/11/2026 at 22:52:39 -07, James Hilliard <james.hilliard1@xxxxxxxxx> wrote:
> >>
> >> > create_bbt() returns an error if reading a bad-block marker fails,
> >>
> >> This sentence does not mean anything.
> >>
> >> > but
> >> > check_create() ignores that result. It can assign version 1 to the
> >>
> >> That's also unclear.
> >
> > It's just stating where an unchecked return exists along with the
> > potential consequences of continuing without checking the return.
> > At least that was how I read it.
> >
> > Is this commit message phrasing/style weird or something?
>
> See just below one example:
>
> >> > incomplete RAM table and write it to the primary and mirror
> >>locations.
>
> "the primary and mirror locations" -> this is a detail, that's not what
> matters.
>
> >> > Unscanned factory-bad blocks then appear good in the persisted
> >> > tables.
> >>
> >> Here is a proposal which I believe makes more sense:
> >>
> >> mtd: rawnand: abort BBT creation on scan failure
> >>
> >> create_bbt() can fail on a non-ECC OOB read error, leaving the RAM BBT
> >> partially populated. It is preferable to bail out early rather than bump
> >> the table versions and write incomplete primary and mirror tables, which
> >> would mark the unscanned factory-bad blocks as good.
> >
> > Personally I try to avoid terminology like "preferable" when
> > describing
>
> Yes, fine.
>
> > code behavior that's clearly wrong, as "preferable" to me at least weakly
> > implies that the code issue being fixed could be more of a style issue
> > rather than a correctness issue.
> >
> > To me both commit messages seem to be equally understandable.
> >
> >>
> >> > Propagate the scan failure before changing table versions or writing
> >> > either copy. Keep explicit NAND_BBT_CREATE_EMPTY initialization unchanged.
> >> >
> >> > Fixes: eeada24da8bd ("[MTD] NAND: Read only OOB bytes during bad block scan")
> >> > Signed-off-by: James Hilliard <james.hilliard1@xxxxxxxxx>
> >>
> >> I would prefer if you could write your own commit logs, giving us more
> >> confidence on the work that has been asked. You can use an LLM to
> >> improve/enhance it obviously.
> >
> > To be honest I've never been great at writing commit messages in
> > general and have gotten a lot more complaints about commit messages
> > missing important details prior to using LLM's. I can barely tell the
>
> For me the code is getting better but the commit logs are unclear. Some
> details which don't need (imo) to be written are there, and the big
> picture is often missing or not clearly stated before the details.
>
> Don't get me wrong, using an LLM to refine/reword/reorganize is fine,
> but letting the LLM write it entirely is not working well.

I guess the issue is that I often don't have the best understanding
of what the big picture that needs explanation in the commit messages
is some of the time, if I did then I would then just instruct the LLM to
make those edits whenever I review patches before sending them.

I think I just don't have good intuition when it comes to predicting what is
important to explain in commit messages.

>
> Thanks,
> Miquèl