Re: [PATCH] mtd: rawnand: denali: do not pass zero maxchips to nand_scan()

[Boris Brezillon <boris.brezillon@bootlin.com>]

On Sat, 25 Aug 2018 00:04:43 +0900
Masahiro Yamada <yamada.masahiro@socionext.com> wrote:

> Hi Boris,
> 
> 2018-08-24 21:55 GMT+09:00 Boris Brezillon <boris.brezillon@bootlin.com>:
> > Hi Masahiro,
> >
> > On Tue, 21 Aug 2018 17:23:19 +0900
> > Masahiro Yamada <yamada.masahiro@socionext.com> wrote:
> >  
> >> Commit 49aa76b16676 ("mtd: rawnand: do not execute nand_scan_ident()
> >> if maxchips is zero") gave a new meaning for calling nand_scan_ident()
> >> with maxchips=0.
> >>
> >> It is a special usage for some drivers such as docg4, but in fact
> >> the Denali driver may pass maxchips=0 to nand_scan() when the driver
> >> is enabled but no NAND chip is found on the board for some reasons.
> >>
> >> If nand_scan_with_ids() is called with maxchips=0, nand_scan_ident()
> >> is skipped, i.e. nand_set_defaults() is skipped.  Therefore, the
> >> driver must have set chip->controller beforehand.  Otherwise,
> >> nand_attach() causes NULL pointer dereference.
> >>
> >> In fact, the Denali controller knows the number of connected chips
> >> before calling nand_scan_ident(); if DEVICE_RESET fails, there is no
> >> chip in that chip select.  Then, denali_reset_banks() sets the maxchips
> >> to the number of detected chips.  If no chip is found, it is zero.
> >>
> >> The reason of this trick was, as commit f486287d2372 ("mtd: nand:
> >> denali: fix bank reset function to detect the number of chips")
> >> explained, nand_scan_ident() issued Set Features (0xEF) command
> >> to all CS lines, some of which may not be connected with a chip.
> >> Then, the driver would wait until R/B# response, which never happens.
> >>
> >> This problem was solved by commit 107b7d6a7ad4 ("mtd: rawnand: avoid
> >> setting again the timings to mode 0 after a reset").  In the current
> >> code, nand_setup_data_interface() is called from nand_scan_tail(),
> >> which is after the chip detection is done.
> >>
> >> Remove the code that is causing NULL pointer dereference.  Now, the
> >> maxchips passed to nand_scan() is the maximum number of chip selects
> >> supported by the IP (typically 4 or 8).  Leave all the chip detection
> >> process to nand_scan_ident().
> >>
> >> Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
> >> ---
> >>
> >>  drivers/mtd/nand/raw/denali.c | 1 -
> >>  1 file changed, 1 deletion(-)
> >>
> >> diff --git a/drivers/mtd/nand/raw/denali.c b/drivers/mtd/nand/raw/denali.c
> >> index ca18612..3e4b8e1 100644
> >> --- a/drivers/mtd/nand/raw/denali.c
> >> +++ b/drivers/mtd/nand/raw/denali.c
> >> @@ -1086,7 +1086,6 @@ static void denali_reset_banks(struct denali_nand_info *denali)
> >>       }
> >>
> >>       dev_dbg(denali->dev, "%d chips connected\n", i);
> >> -     denali->max_banks = i;  
> >
> > Shouldn't we instead avoid calling nand_scan() when
> > denali->max_banks=0? I mean, what's the point of calling this function
> > if you know for sure it will fail.  
> 
> 
> Right.  If no chip is found, it should error out with -ENODEV or something.
> 
> 
> 
> > Last question: do we still need this denali_reset_banks()? If it's only
> > about resetting the chip to detect how many are actually present,
> > that's already done by nand_scan().  
> 
> I thought this too.
> 
> Please give me time to answer this question.
> I need to check the datasheet and test on my boards.
> 
> If I can remove denali_reset_banks() entirely,
> it would be the best.

I'd like the fix to be as simple as possible, so that I can queue it
for -rc2. Please consider the solution where nand_scan() is skipped
when denali->max_banks=0 first. We can then decide to remove
denali_reset_banks() if that's appropriate.

Thanks,

Boris