Re: [PATCH] Fix writing mtdoops to nand flash.

[Boris Brezillon <boris.brezillon@free-electrons.com>]

On Mon, 30 Oct 2017 07:46:18 -0500
Brent Taylor <motobud@gmail.com> wrote:

> Hi Bors, thanks for your quick reply.
> 
> On Mon, Oct 30, 2017 at 3:23 AM, Boris Brezillon
> <boris.brezillon@free-electrons.com> wrote:
> > Hi Brent,
> >
> > Subject should be prefixed by "mtd: nand: ", so
> >
> > "mtd: nand: Fix writing mtdoops to nand flash"  
> 
> Oops, yes, will fix.
> 
> >
> > On Sun, 29 Oct 2017 23:23:43 -0500
> > motobud@gmail.com wrote:
> >  
> >> From: Brent Taylor <motobud@gmail.com>
> >>
> >> When mtdoops calls mtd_panic_write, it eventually calls
> >> panic_nand_write in nand_base.c.  In order to properly
> >> wait for the nand chip to be ready in panic_nand_wait,
> >> the chip must first be selected.
> >>
> >> When using the atmel nand flash controller, a panic
> >> would occur due to a NULL pointer exception.
> >>
> >> Signed-off-by: Brent Taylor <motobud@gmail.com>
> >> ---
> >>  drivers/mtd/nand/nand_base.c | 5 +++++
> >>  1 file changed, 5 insertions(+)
> >>
> >> diff --git a/drivers/mtd/nand/nand_base.c b/drivers/mtd/nand/nand_base.c
> >> index 12edaae17d81..0a8058a66d93 100644
> >> --- a/drivers/mtd/nand/nand_base.c
> >> +++ b/drivers/mtd/nand/nand_base.c
> >> @@ -2802,9 +2802,14 @@ static int panic_nand_write(struct mtd_info *mtd, loff_t to, size_t len,
> >>       struct mtd_oob_ops ops;
> >>       int ret;
> >>
> >> +     int chipnr = (int)(to >> chip->chip_shift);
> >> +     chip->select_chip(mtd, chipnr);
> >> +
> >>       /* Wait for the device to get ready */
> >>       panic_nand_wait(mtd, chip, 400);
> >>
> >> +     chip->select_chip(mtd, -1);
> >> +  
> >
> > Duh! Looks like a piece of code that is never tested. Did you face the
> > problem or did you find out by inspecting the code?  
> 
> I was playing with another driver on an atmel development board
> (at91sam9m10g45ek) and caused a panic with mtdoops enabled.  While
> writing the mtdoops to nand, another panic occurred which caused a
> storm of panics to generated.
> 
> >
> > Anyway, I fear the atmel driver is not the only one to rely on the chip
> > to be selected when ->dev_ready() is called so this should definitely
> > be fixed. Also, we should probably move the ->select_chip() and
> > panic_nand_wait() calls after panic_nand_get_device(), and I don't
> > think we need to unselect the chip (it will be taken care of by
> > nand_do_write_ops()).
> >  
> >>       /* Grab the device */
> >>       panic_nand_get_device(chip, mtd, FL_WRITING);
> >>  
> >  
> 
> After looking at this closer, a panic could happen at any point correct?

Yes.

> If
> that's the case, the kernel could have been in the middle of a nand read/write
> operation (which may or may not be on the same chip).  Would the chip the
> mtdoops is being written to need to be reset?

I'd prefer not. Actually, the code calls panic_nand_wait() to let the
NAND finish the operation it's currently doing, and most of the time it
will relax the busy pin within 400ms (the timeout passed to
panic_nand_wait()). If we do a RESET in the middle of a write operation
we'll just corrupt the data that is being written, and I'm not sure we
want that to happen.

> I haven't drilled down into the
> nand_reset function yet, but can that be called in a "panic" state?

Maybe, but I'm not sure we want to do that (see above).

Anyway, the write-on-panic path is fragile, and I'm not sure all
controllers will play well in this situation: just because the NAND is
ready does not mean the controller is, and there's nothing ensuring
that in the NAND framework.

Regards,

Boris