[PATCH 3/3] nandtest: Introduce multiple reads & check iterations
pekon at ti.com
Mon May 5 04:21:05 PDT 2014
>From: Artem Bityutskiy [mailto:dedekind1 at gmail.com]
>>On Mon, 2014-05-05 at 10:58 +0000, Gupta, Pekon wrote:
>> Therefore, I'm not sure having re-reads is a good test or not,
>> re-reads is changing the underlying testing scenario by introducing
>> bit-flips in neighboring regions pages because of read-disturbs.
>If the driver / HW cannot properly handle 4 re-reads, we do want the
>test to catch this, I think, irrespective of the true nature of the
>Therefore, I see this patch with a default 4 as an improvement.
I'm not against the patch, but my thought is that re-reads are
introducing read-disturb bit-flips.
Now in event of no UBIFS there would be no scrubbing. So these bit-flips
will keep on accumulating. And once these bit-flips cross beyond
ecc.strength then the nand_read is bound to fail.
So this makes this test un-deterministic.
Example: If I use this test with any driver using 1-bit ECC correction. Then
- test will most likely fail on weak or aged NAND devices, as they are more
prone to read-disturb errors.
- But _may_ not fail on fresh devices using same driver with 1-bit ECC correction.
So my argument is that this test is actually not testing the NAND driver,
it is actually testing the weakness of NAND device.
with regards, pekon
More information about the linux-mtd