[U-Boot] [PATCH 1/5] nand: Use common read function instead of verify_buf()

Peter Tyser ptyser at xes-inc.com
Wed Jan 28 00:47:19 CET 2015


Hi Scott,


> > I waffled about removing it, but leaned towards leaving it in because:
> > - I didn't want to change the existing U-Boot behavior for other
> > users.  A google of 'u-boot "nand write"' shows a lot of examples that
> > don't include verification of writes, and they should if we remove
> > auto-verification.
> 
> How many configs actually enable this option?  I don't see many beyond
> the FSL PPC boards (which are so full of copy-and-paste that it probably
> wasn't deliberate).

Yeah, the majority are FSL 83xx and 85xx, with 2 or so random ARM boards.

> > - The reason it was removed in Linux was "Both UBI and JFFS2 are able
> > to read verify what they wrote already.  There are also MTD tests
> > which do this verification."  I thought U-Boot was more likely than
> > Linux to use raw NAND writes without a filesystem, so leaving it in U-
> > Boot made sense since the UBI/JFFS2 logic didn't apply as much here.
> 
> Right, though raw writes ought to be limited to blocks that aren't
> written often enough to fail.
> 
> > - I didn't think a lot of people would know they have to explicitly
> > verify NAND contents after a write, since they'd assume it was like
> > other memories that aren't as lossy.
> > 
> > - The penalty of slightly different code from Linux and a small
> > performance hit was worth the gain of auto-verification to me.  I
> > viewed consolidating it into one small chunk of code as a happy medium.
> 
> The davinci patches show that there can still be driver dependencies
> depending on what the driver overrides.  I'm not hugely opposed, but it
> seems like it would be better to do it at a higher level (e.g. in
> nand_util.c with a flag to enable, and either make support mandatory, or
> if you try to use that command variant without support it fails rather
> than silently not verifying).

That seems like a good idea.  How about:
- Remove all CONFIG_MTD_NAND_VERIFY_WRITE references

- Add a new flag WITH_WR_VERIFY and have nand_write_skip_bad() in 
nand_util.c verify writes only when it is set.

- Update the calls to nand_write_skip_bad() in cmd_nand.c to include
the new WITH_WR_VERIFY flag.  I'd vote to enable it for all boards,
but let me know if you disagree.

That would make all "nand write" commands verify writes, with the
exception of "nand write.raw".  Any opinion on if this should also
be verified?  I only use it for development/testing, so don't have
a strong opinion.

Regards,
Peter





More information about the U-Boot mailing list