[PATCH v5 01/58] mtd: nand: denali: add missing nand_release() call in denali_remove()

[Boris Brezillon]

Hi Dinh,

On Fri, 11 Dec 2015 10:50:21 -0600
Dinh Nguyen <dinh.linux at gmail.com> wrote:

> Hi Boris,
> 
> On Fri, Dec 11, 2015 at 9:10 AM, Boris Brezillon
> <boris.brezillon at free-electrons.com> wrote:
> > + Dinh (who made commit 2a0a288ec258)
> >
> > Also added back the Fixes tag.
> >
> > On Fri, 11 Dec 2015 15:02:34 +0100
> > Boris Brezillon <boris.brezillon at free-electrons.com> wrote:
> >
> >> Unregister the NAND device from the NAND subsystem when removing a denali
> >> NAND controller, otherwise the MTD attached to the NAND device is still
> >> exposed by the MTD layer, and accesses to this device will likely crash
> >> the system.
> >>
> >> Signed-off-by: Boris Brezillon <boris.brezillon at free-electrons.com>
> >
> > Fixes: 2a0a288ec258 ("mtd: denali: split the generic driver and PCI layer")
> >
> >> ---
> >> Changes since v4:
> >> - remove Cc stable and fixes tags
> >> - calculate the dma buffer size before calling nand_release()
> >>
> >>  drivers/mtd/nand/denali.c | 6 ++++--
> >>  1 file changed, 4 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/mtd/nand/denali.c b/drivers/mtd/nand/denali.c
> >> index 67eb2be..fdfea05 100644
> >> --- a/drivers/mtd/nand/denali.c
> >> +++ b/drivers/mtd/nand/denali.c
> >> @@ -1622,9 +1622,11 @@ EXPORT_SYMBOL(denali_init);
> >>  /* driver exit point */
> >>  void denali_remove(struct denali_nand_info *denali)
> >>  {
> >> +     int bufsize = denali->mtd.writesize + denali->mtd.oobsize;
> >> +
> >> +     nand_release(&denali->mtd);
> >>       denali_irq_cleanup(denali->irq, denali);
> >> -     dma_unmap_single(denali->dev, denali->buf.dma_buf,
> >> -                      denali->mtd.writesize + denali->mtd.oobsize,
> >> +     dma_unmap_single(denali->dev, denali->buf.dma_buf, bufsize,
> >>                        DMA_BIDIRECTIONAL);
> 
> Not sure what is the need to add bufsize here, but the commit message
> doesn't reflect the change.

You were not in Cc of the first version (my fault), but Brian pointed
that the mtd fields could be in an unknown state after the
nand_release() call (this is currently not the case, but it change in
the future). The idea is to pre-compute the DMA buffer size before
releasing the mtd/nand device to prevent any future issues.

I don't think it is worth mentioning this in the commit message,
because these are just implementation details, but I can add the
following comment before the bufsize declaration:

	/*
	 * Pre-compute DMA buffer size to avoid any problems in case
	 * nand_release() ever changes in a way that mtd->writesize and
	 * mtd->oobsize are not reliable after this call.
	 */

What do you think?

Best Regards,

Boris

> 
> Dinh



-- 
Boris Brezillon, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com