Commit d020fc8e authored Jan 09, 2018 by Boris Brezillon

mtd: mtdpart: Make ECC stat handling consistent

part_read() and part_read_oob() were counting ECC failures and
bitflips differently. Adjust part_read_oob() to mimic what is done in
part_read(). This is needed to use ->_read_oob() as a fallback when
when ->_read() is not implemented.

Note that bitflips and ECC failure accounting on MTD partitions is
broken by design, because nothing prevents concurrent accesses to the
underlying master MTD device between the moment we save the stats in a
local variable and the moment master->_read[_oob]() returns. It's not
something that can easily be fixed, so leave it like that for now.

Suggested-by: Brian Norris <computersforpeace@gmail.com>
Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Tested-by: Ladislav Michl <ladis@linux-mips.org>

parent f72071b8

drivers/mtd/mtdpart.c

+8 −6

Original line number	Diff line number	Diff line
		@@ -105,6 +105,7 @@ static int part_read_oob(struct mtd_info *mtd, loff_t from,
		struct mtd_oob_ops *ops)
		{
		struct mtd_part *part = mtd_to_part(mtd);
		struct mtd_ecc_stats stats;
		int res;

		if (from >= mtd->size)
		@@ -126,13 +127,14 @@ static int part_read_oob(struct mtd_info *mtd, loff_t from,
		return -EINVAL;
		}

		stats = part->parent->ecc_stats;
		res = part->parent->_read_oob(part->parent, from + part->offset, ops);
		if (unlikely(res)) {
		if (mtd_is_bitflip(res))
		mtd->ecc_stats.corrected++;
		if (mtd_is_eccerr(res))
		mtd->ecc_stats.failed++;
		}
		if (unlikely(mtd_is_eccerr(res)))
		mtd->ecc_stats.failed +=
		part->parent->ecc_stats.failed - stats.failed;
		else
		mtd->ecc_stats.corrected +=
		part->parent->ecc_stats.corrected - stats.corrected;
		return res;
		}

Admin message