Re: [PATCH v3 2/5] mtd: mtdpart: Do not fail mtd probe when parsing partitions fails.

[Brian Norris]

Hi Michal,

Sorry for the very long waits here. Unfortunately, I've been saying that
a lot lately, as I haven't had a lot of time and have generated a lot of
backlog. I really like that you've been working on this though, since
there are definitely problems here.

On Tue, Aug 18, 2015 at 03:34:07PM -0000, Michal Suchanek wrote:
> Due to wrong assumption in ofpart ofpart fails on Exynos on SPI chips
> with no partitions because the subnode containing controller data
> confuses the ofpart parser.
> 
> Thus compiling in ofpart support automatically fails probing any SPI NOR
> flash without partitions on Exynos.
> 
> Compiling in a partitioning scheme should not cause probe of otherwise
> valid device to fail.
> 
> Remove that failure possibility when MTD_PARTITIONED_MASTER is set.
> 
> Signed-off-by: Michal Suchanek <hramrach@xxxxxxxxx>
> 
> ---
> v2:
> 
>  - only allow partition parsing failure when MTD_PARTITIONED_MASTER is
>    set

Why the special case for MTD_PARTITIONED_MASTER? I don't think this case
should be much different. In either case, we should actually be falling
back to just exposing the unpartitioned master device. We already do
that when no partitions are found (but no error codes) in either
MTD_PARTITIONED_MASTER={y,n}. We just need to do that for error cases
too, I think.

> ---
>  drivers/mtd/mtdpart.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/mtd/mtdpart.c b/drivers/mtd/mtdpart.c
> index 31888c2..6eafbe9 100644
> --- a/drivers/mtd/mtdpart.c
> +++ b/drivers/mtd/mtdpart.c
> @@ -774,10 +774,15 @@ int parse_mtd_partitions(struct mtd_info *master, const char *const *types,
>  		if (ret > 0) {
>  			printk(KERN_NOTICE "%d %s partitions found on MTD device %s\n",
>  			       ret, parser->name, master->name);
> -			break;
> +			return ret;
> +		}
> +		if (!IS_ENABLED(CONFIG_MTD_PARTITIONED_MASTER) && (ret < 0)) {
> +			pr_err("Error parsing %s partitions on %s\n",
> +			       parser->name, master->name);
> +			return ret;

I think this hunk should be more like the hunk from the previous
versions; don't do a separate CONFIG_xxx special case.

I also think this has one other flaw: it's best if we don't completely
ignore all failures. If we have one or more partitioning failures, this
should show up somehow in the log at least, and not only with the
pr_debug() messages from patch 1. I like the logic for the block
subsystem for failures (see check_partition() in
block/partitions/check.c). Skip a parser quietly if a later one
succeeds, but report an error with a warning printk if none succeed.

>  		}
>  	}
> -	return ret;
> +	return 0;
>  }
>  
>  int mtd_is_partition(const struct mtd_info *mtd)

All in all, I think my suggestions would look something like the
following alternative patch. I haven't tested it yet.

Brian

(git-format-patch pasted below)