Re: [PATCH] mtd: rawnand: marvell: select target before setting up data interface

[Miquel Raynal <miquel.raynal@bootlin.com>]

Hi Daniel,

Boris Brezillon <boris.brezillon@collabora.com> wrote on Sat, 30 Mar
2019 19:20:36 +0100:

> On Mon, 25 Mar 2019 17:21:42 +0100
> Daniel Mack <daniel@zonque.org> wrote:
> 
> > Hi Miquel,
> > 
> > On 25/3/2019 10:09 AM, Miquel Raynal wrote:  
> > > Daniel Mack <daniel@zonque.org> wrote on Fri, 22 Mar 2019 19:48:08
> > > +0100:
> > >     
> > >> Hi Miquel,
> > >>
> > >> On 22/3/2019 9:30 AM, Miquel Raynal wrote:    
> > >>> + Richard: this is fixes materials
> > >>>
> > >>> Daniel Mack <daniel@zonque.org> wrote on Thu, 21 Mar 2019 22:20:10
> > >>> +0100:
> > >>>       
> > >>>> Now that the nand drivers are responsible for selecting the target
> > >>>> prior to hardware access, a call to marvell_nfc_select_target() is
> > >>>> necessary from marvell_nfc_setup_data_interface().
> > >>>>
> > >>>> This is a regression introduced by commit b25251414f6e ("mtd:
> > >>>> rawnand: marvell: Stop implementing ->select_chip()").
> > >>>>
> > >>>> Fixes: b25251414f6e ("mtd: rawnand: marvell: Stop implementing ->select_chip()")
> > >>>> Cc: Boris Brezillon <boris.brezillon@collabora.com>
> > >>>> Cc: <stable@vger.kernel.org>
> > >>>> Signed-off-by: Daniel Mack <daniel@zonque.org>
> > >>>> ---
> > >>>>  drivers/mtd/nand/raw/marvell_nand.c | 2 ++
> > >>>>  1 file changed, 2 insertions(+)
> > >>>>
> > >>>> diff --git a/drivers/mtd/nand/raw/marvell_nand.c b/drivers/mtd/nand/raw/marvell_nand.c
> > >>>> index 84283c6bb0ff..93abe5be8995 100644
> > >>>> --- a/drivers/mtd/nand/raw/marvell_nand.c
> > >>>> +++ b/drivers/mtd/nand/raw/marvell_nand.c
> > >>>> @@ -2325,6 +2325,8 @@ static int marvell_nfc_setup_data_interface(struct nand_chip *chip, int chipnr,
> > >>>>  	struct marvell_nfc_timings nfc_tmg;
> > >>>>  	int read_delay;
> > >>>>  
> > >>>> +	marvell_nfc_select_target(chip, chip->cur_cs);
> > >>>> +
> > >>>>  	sdr = nand_get_sdr_timings(conf);
> > >>>>  	if (IS_ERR(sdr))
> > >>>>  		return PTR_ERR(sdr);      
> > >>>
> > >>> I had a short look yesterday and this is very likely to be the root
> > >>> cause of the regression, thank you very much!      
> > >>
> > >> Yes it is, it fixes the regression I have reported.
> > >>
> > >> Do you want me to resend with that nit addressed you mentioned? My
> > >> version works well, I know that much :)    
> > > 
> > > Yes please, resent with Boris' comment addressed and we'll be fine.    
> > 
> > I tried this now, and with 'chip->cur_cs' instead of 'chipnr', the
> > regression is back. This function is invoked twice during boot, and in
> > both cases, 'chip->cur_cs' is -1, while 'chipnr' is 0. The 2nd time this
> > function is invoked, the 'selected_die' cache value is 0, so
> > marvell_nfc_select_target() bails early, which seems to be a problem.
> > 
> > I can't quite make sense of that yet, all I know is that calling
> > marvell_nfc_select_target() with -1 from
> > marvell_nfc_setup_data_interface() makes the driver functional again. Hmm.  
> 
> Can you try with the following diff applied?
> --->8---  
> diff --git a/drivers/mtd/nand/raw/marvell_nand.c b/drivers/mtd/nand/raw/marvell_nand.c
> index f38e5c1b87e4..e8eccf6afc8f 100644
> --- a/drivers/mtd/nand/raw/marvell_nand.c
> +++ b/drivers/mtd/nand/raw/marvell_nand.c
> @@ -722,9 +722,6 @@ static void marvell_nfc_select_target(struct nand_chip *chip,
>         struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
>         u32 ndcr_generic;
>  
> -       if (chip == nfc->selected_chip && die_nr == marvell_nand->selected_die)
> -               return;
> -
>         writel_relaxed(marvell_nand->ndtr0, nfc->regs + NDTR0);
>         writel_relaxed(marvell_nand->ndtr1, nfc->regs + NDTR1);
>  

I tested on a pxa board and an armada board and I don't reproduce the
bug without this change. What is special with your setup?

I was trying to understand the fix to write a decent commit log but as
I cannot reproduce I am stuck. So if you need this fix please send the
patch with, hopefully, an explanation of what's wrong.

Thank you!
Miquèl