[PATCH] mtd: nand: Disable subpage writes for drivers without ecc->hwctl

[Brian Norris]

On Sat, Dec 12, 2015 at 08:59:00AM +0100, Boris Brezillon wrote:
> On Fri, 11 Dec 2015 16:40:25 -0800
> Brian Norris <computersforpeace at gmail.com> wrote:
> 
> > On Thu, Dec 10, 2015 at 10:31:32AM +0100, Helmut Schaa wrote:
> > > On Thu, Dec 10, 2015 at 12:58 AM, Brian Norris
> > > <computersforpeace at gmail.com> wrote:
> > > > On Wed, Apr 09, 2014 at 11:13:24AM +0200, Helmut Schaa wrote:
> > > >> nand_write_subpage_hwecc causes a crash if the driver did not register
> > > >> ecc->hwctl or ecc->calculate. Fix this by disabling subpage writes if
> > > >> ecc->hwctl or ecc->calculate is not provided by the driver.
> > > >>
> > > >> This behavior was introduced in commit 837a6ba4f3b6d23026674e6af6b6849a4634fff9
> > > >> "mtd: nand: subpage write support for hardware based ECC schemes".
> > > >>
> > > >> This fixes a crash with fsl_elbc_nand and maybe others:
> > > >>
> > > > [...]
> > > >>
> > > >> --- a/drivers/mtd/nand/nand_base.c
> > > >> +++ b/drivers/mtd/nand/nand_base.c
> > > >> @@ -3768,7 +3768,7 @@ int nand_scan_tail(struct mtd_info *mtd)
> > > >>                       ecc->write_oob = nand_write_oob_std;
> > > >>               if (!ecc->read_subpage)
> > > >>                       ecc->read_subpage = nand_read_subpage;
> 
> Shouldn't we have the same kind of test for ->read_subpage.
> 
> 		if (!ecc->read_subpage && ecc->correct &&
> 		    ecc->calculate)
> 			ecc->read_subpage = nand_read_subpage;

We could do that, though right now we have a little less danger for
read_subpage() than we do with write_subpage(), because drivers have to
opt in with the NAND_SUBPAGE_READ flag before this function will
actually be used. Feel free to send a patch if you'd like.

> > > >> -             if (!ecc->write_subpage)
> > > >> +             if (!ecc->write_subpage && ecc->hwctl && ecc->calculate)
> > > >>                       ecc->write_subpage = nand_write_subpage_hwecc;
> > > >>
> > > >>       case NAND_ECC_HW_SYNDROME:
> > > >
> > > > I realize we've merged a fix for fsl_elbc_nand long ago (commit
> > > > f034d87def51 "mtd: eLBC NAND: fix subpage write support"), but this
> > > > change still looks sane, applies, and could possibly fix some other
> > > > drivers that are lurking somewhere. Any reason I shouldn't apply it?
> > > 
> > > I'm not carrying this patch in my tree anymore and did not see any more
> > > crashes with the flash configuration I'm using. So, I think the original patch
> > > is kind of superfluous.
> > 
> > There could easily be other drivers that don't have hwctl() or
> > calculate() callbacks yet use NAND_ECC_HW. By inspection, I see that
> > maybe the new hisi504_nand.c is susceptible, as well as sh_flctl.c and
> > maybe even sunxi_nand.c (?). Some of these might not be hit if they
> > don't use NAND that support subpage writes.
> 
> Yep, sunxi platforms could be hit by this bug, but it seems nobody
> decided to design a board with an SLC NAND (and MLCs are not supporting
> subpage writes).
> 
> > 
> > Anyway, I think this patch is helpful, because it prevents drivers from
> > having to fill in a 'subpage' write callback that looks just like their
> > write_page() callback.
> 
> I agree.

OK. Applied to l2-mtd.git, with additional commentary about how the
reported issue is already fixed, but there could be other drivers who
run across this problem.

Brian