Re: [PATCH] mtd: spi-nor: fix erase_type array to indicate current map conf

[Boris Brezillon <boris.brezillon@xxxxxxxxxxx>]

On Fri, 23 Nov 2018 09:42:55 +0000
"Sverdlin, Alexander (Nokia - DE/Ulm)" <alexander.sverdlin@xxxxxxxxx>
wrote:

> Hello Tudor,
> 
> On 22/11/2018 13:36, Tudor.Ambarus@xxxxxxxxxxxxx wrote:
> > From: Tudor Ambarus <tudor.ambarus@xxxxxxxxxxxxx>
> > 
> > Bug reported for the out-of-tree S25FS128S flash memory.
> > 
> > BFPT table advertises all the erase types supported by all the
> > possible map configurations. Update the erase_type array to indicate
> > which erase types are applicable to the current map configuration.
> > 
> > Backward compatibility test done on sst26vf064b.
> > 
> > Fixes: b038e8e3be72 ("mtd: spi-nor: parse SFDP Sector Map Parameter Table")
> > Reported-by: Alexander Sverdlin <alexander.sverdlin@xxxxxxxxx>
> > Signed-off-by: Tudor Ambarus <tudor.ambarus@xxxxxxxxxxxxx>  
> 
> I've tested this patch and it fixes the erasesize for S25FS128S and
> our 128k partitions are writeable again with this patch.
> 
> Nevertheless, I think this is coincidence. I don't think that it
> makes sense to OR all the erase types from all regions in one
> bitmask and derive any uniform erasesize out of it.
> This makes little sense for me in case of non-uniform maps.
> 
> I believe, the culprit here is one level higher, in the MTD partitioning
> code (mtdpart.c) which has to be taught about non-uniform maps
> but there is no infrastructure for this currently.

Keep in mind that mtdpart is only one part of the issue. As I said
previously, some MTD users (UBI) expect a single eraseblock size, so
it's not only mtdpart that you'd need to fix, but basically all MTD
users that don't check ->eraseregions.

> 
> What is possible to fix still is to choose smallest, not biggest
> erasesize for uniform case.

Yep, I agree. But we also have to be careful here, if some NORs were
already supported and were picking the biggest erasesize, we have to
preserve this behavior, otherwise we'll break things.

> I have such a patch, but I need hands
> on on a S25FS128S configured in uniform mode.
> 
> Non uniform case requires MTD layer rework.

There's already the concept of ->eraseregions. Maybe I'm wrong but I
thought it would fit the non-uniform erase scheme we have in SPI NOR. 

> We just cannot handle
> this hardware with just one erasesize in mind.

We can, if we decide to always use the biggest non-uniform erasesize.
Of course that only works if the biggest erase size is always a
multiple of smaller ones, but I'm pretty sure this is always true.

> 
> > ---
> >  drivers/mtd/spi-nor/spi-nor.c | 29 ++++++++++++++++++++++++++++-
> >  1 file changed, 28 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/mtd/spi-nor/spi-nor.c b/drivers/mtd/spi-nor/spi-nor.c
> > index 93c9bc8931fc..a71adcd6ddfa 100644
> > --- a/drivers/mtd/spi-nor/spi-nor.c
> > +++ b/drivers/mtd/spi-nor/spi-nor.c
> > @@ -3000,7 +3000,8 @@ static int spi_nor_init_non_uniform_erase_map(struct spi_nor *nor,
> >  	u64 offset;
> >  	u32 region_count;
> >  	int i, j;
> > -	u8 erase_type, uniform_erase_type;
> > +	u8 uniform_erase_type, save_uniform_erase_type;
> > +	u8 erase_type, regions_erase_type;
> >  
> >  	region_count = SMPT_MAP_REGION_COUNT(*smpt);
> >  	/*
> > @@ -3014,6 +3015,7 @@ static int spi_nor_init_non_uniform_erase_map(struct spi_nor *nor,
> >  	map->regions = region;
> >  
> >  	uniform_erase_type = 0xff;
> > +	regions_erase_type = 0;
> >  	offset = 0;
> >  	/* Populate regions. */
> >  	for (i = 0; i < region_count; i++) {
> > @@ -3030,13 +3032,38 @@ static int spi_nor_init_non_uniform_erase_map(struct spi_nor *nor,
> >  		 */
> >  		uniform_erase_type &= erase_type;
> >  
> > +		/*
> > +		 * regions_erase_type mask will indicate all the erase types
> > +		 * supported in this configuration map.
> > +		 */
> > +		regions_erase_type |= erase_type;
> > +
> >  		offset = (region[i].offset & ~SNOR_ERASE_FLAGS_MASK) +
> >  			 region[i].size;
> >  	}
> >  
> > +	save_uniform_erase_type = map->uniform_erase_type;
> >  	map->uniform_erase_type = spi_nor_sort_erase_mask(map,
> >  							  uniform_erase_type);
> >  
> > +	if (!regions_erase_type) {
> > +		/*
> > +		 * Roll back to the previous uniform_erase_type mask, SMPT is
> > +		 * broken.
> > +		 */
> > +		map->uniform_erase_type = save_uniform_erase_type;
> > +		return -EINVAL;
> > +	}
> > +
> > +	/*
> > +	 * BFPT table advertises all the erase types supported by all the
> > +	 * possible map configurations. Update the erase_type array to indicate
> > +	 * which erase types are applicable to the current map configuration.
> > +	 */
> > +	for (i = 0; i < SNOR_ERASE_TYPE_MAX; i++)
> > +		if (!(regions_erase_type & BIT(erase[i].idx)))
> > +			spi_nor_set_erase_type(&erase[i], 0, 0xFF);
> > +
> >  	spi_nor_region_mark_end(&region[i - 1]);
> >  
> >  	return 0;  
> 


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