On Mon, Jul 27, 2015 at 09:19:57PM +0200, Lucas Stach wrote:
> Am Dienstag, den 21.07.2015, 14:27 -0700 schrieb Brian Norris:
> I've checked this again and unfortunately the DEC_STATUS_ERR_COUNT seems
> to be completely bogus. It is certainly only set when there are
> bitflips, but it doesn't represent a valid value.
:( HW sucks.
> I have another register that seems to accurately reflect the maximum
> number of bitflips in any of the 512b ECC blocks, so I'm able to return
> the correct amount of bitflips to the upper level.
> The downside is that the hardware already does the calculation of max
> bitflips, so I'm unable to count bitflips correctly for the stats. All I
> can do is to return a pessimistic value that assumes that all 512b
> blocks had the returned number of bitflips. Could this be a problem?
As things currently stand: no, I don't think so. The ecc_stats are
actually mostly just that -- statistics. Some pieces of the codebase use
the fact they increased as a sign, but they don't care about the actual
number. The max_bitflips calculation, however, is very important so that
MTD can determine whether we've reached the reliability threshold.
So I guess you have some flexibility in how to report this, but I think
two main options are sane:
1. leave as-is; you can only guarantee that max_bitflips occurred in
total, so only increment ecc_stats.corrected by max_bitflips
2. change to
ecc_stats.corrected += max_bitflips * chip->ecc.steps;
This is the maximum that might have occured on this page read.
IOW, option 1 is optimistic and option 2 is pessimistic. Either way,
this probably deserves a comment in the code.
Regards,
Brian
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
