Re: [PATCH] mtd: rawnand: check return code of nand_reset() and nand_readid_op()

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

Hi Masahiro,

Masahiro Yamada <yamada.masahiro@socionext.com> wrote on Tue, 22 Jan
2019 00:57:43 +0900:

> On Mon, Jan 21, 2019 at 10:14 PM Boris Brezillon <bbrezillon@kernel.org> wrote:
> >
> > On Mon, 21 Jan 2019 22:05:34 +0900
> > Masahiro Yamada <yamada.masahiro@socionext.com> wrote:
> >  
> > > nand_scan_ident() iterates over maxchips to find as many homogeneous
> > > chips as possible.
> > >
> > > Currently, this loop bails out only when manufacturer or device ID
> > > unmatches. The reason of unmatch is most likely no chip is connected
> > > to that chip select. In this case, nand_reset() has already failed,
> > > and the following nand_readid_op() is pointless.  
> >
> > While I agree with the following diff, I'd also like to point out that
> > nand_scan() callers should know how many controller CS lines are
> > connected to the chip (board file or DT description). The check we do in
> > nand_scan_ident() should only be here to clamp this value if the board
> > desc is wrong (maybe we should even fail in that case instead of
> > silently fixing things).  
> 
> 
> I know this.
> This is a problem for denali because
> I have not decoupled chip/controller yet.
> 
> 
> Maybe, is the following better?
> 
> 
> ------------------>8-----------------------  
> nand_scan_ident() iterates over maxchips to find as many homogeneous
> chips as possible.
> 
> Since commit 2d472aba15ff ("mtd: nand: document the NAND controller/NAND
> chip DT representation"), new drivers should pass in the exact number of
> CS lines instead of possible max, but old platforms may still rely on
> nand_scan_ident() to detect the actual number of connected CS lines.
> 
> In that case, this loop bails out when manufacturer or device ID
> unmatches. The reason of unmatch is most likely no chip is connected
> to that CS line. If so, nand_reset() should already have failed,
> and the following nand_readid_op() is pointless.
> 
> Before ->exec_op hook was introduced, drivers had no way to tell
> the failure of NAND_CMD_RESET to the framework because the legacy
> ->cmdfunc() has void return type. Now drivers implementing ->exec_op  
> hook can return the error code. You can save nand_readid_op() by
> checking the return value of nand_reset(). The return value of
> nand_readid_op() should be checked as well. If it fails, probably
> id[0] and id[1] are undefined values.
> 
> Just for consistency, it should be sensible to check the return
> code in nand_do_write_oob() as well.
> ------------------------------>8--------------------------------  
> 

Patch applied to nand/next with the second commit log you proposed.

Thanks for all your work!
Miquèl