Question about nand_scan()

Question about nand_scan()

[Jan Glauber]

Hi all,

I'm working on a driver for the NAND controller on Cavium's ThunderX.

So far I implemented the low-level functions for using the controller
to access a NAND chip. I can read the ONFI ID and parameter page
with that.

Now I wanted to use nand_scan() instead of manually reading the chip
parameters, but it fails with "No NAND device found".

The hardware I'm using has one NAND device wired as chip 1 (the NAND
controller support chips 0..7).

The reason for the failure seems to be that nand_get_flash_type()
returns an error before all the chips are scanned. What I don't
understand is in that function chip 0 is selected before the loop
that would scan all chips:

        /* Select the device */
        chip->select_chip(mtd, 0);

My select_chip() stores the chip number (in that case 0) and uses
that for subsequent commands to the controller. Since there is no
chip 0 the read returns nothing and nand_scan() fails.

Probably I'm missing something, would be great if someone could
help me...

thanks,
Jan

Re: Question about nand_scan()

[Boris Brezillon]

Hi Jan,

On Fri, 23 Sep 2016 13:29:44 +0200
Jan Glauber <jan.glauber@caviumnetworks.com> wrote:

> Hi all,
> 
> I'm working on a driver for the NAND controller on Cavium's ThunderX.
> 
> So far I implemented the low-level functions for using the controller
> to access a NAND chip. I can read the ONFI ID and parameter page
> with that.
> 
> Now I wanted to use nand_scan() instead of manually reading the chip
> parameters, but it fails with "No NAND device found".
> 
> The hardware I'm using has one NAND device wired as chip 1 (the NAND
> controller support chips 0..7).
> 
> The reason for the failure seems to be that nand_get_flash_type()
> returns an error before all the chips are scanned. What I don't
> understand is in that function chip 0 is selected before the loop
> that would scan all chips:
> 
>         /* Select the device */
>         chip->select_chip(mtd, 0);
> 
> My select_chip() stores the chip number (in that case 0) and uses
> that for subsequent commands to the controller. Since there is no
> chip 0 the read returns nothing and nand_scan() fails.
> 
> Probably I'm missing something, would be great if someone could
> help me...

Actually, the chip parameter passed to ->select_chip() is not the
CS-id from the NAND controller PoV, but the one from the NAND chip PoV.

You have to store an association table between chip⁻CS and controller-CS
somewhere in your NAND controller private struct (a struct inheriting
from nand_control_hw).

And remember that nand_scan_ident() is supposed to detect a single NAND
chip, not all the chips connected to your controller (the multi-CS
logic is here to handle multi-dies NANDs, not the case where you have
multiple NAND chips connected to the same controller).

You can have a look at the sunxi_nand [1] driver if you want an example.

Regards,

Boris

[1]http://lxr.free-electrons.com/source/drivers/mtd/nand/sunxi_nand.c#L219

Re: Question about nand_scan()

[Jan Glauber]

On Fri, Sep 23, 2016 at 01:43:44PM +0200, Boris Brezillon wrote:
> Hi Jan,
> 
> On Fri, 23 Sep 2016 13:29:44 +0200
> Jan Glauber <jan.glauber@caviumnetworks.com> wrote:
> 
> > Hi all,
> > 
> > I'm working on a driver for the NAND controller on Cavium's ThunderX.
> > 
> > So far I implemented the low-level functions for using the controller
> > to access a NAND chip. I can read the ONFI ID and parameter page
> > with that.
> > 
> > Now I wanted to use nand_scan() instead of manually reading the chip
> > parameters, but it fails with "No NAND device found".
> > 
> > The hardware I'm using has one NAND device wired as chip 1 (the NAND
> > controller support chips 0..7).
> > 
> > The reason for the failure seems to be that nand_get_flash_type()
> > returns an error before all the chips are scanned. What I don't
> > understand is in that function chip 0 is selected before the loop
> > that would scan all chips:
> > 
> >         /* Select the device */
> >         chip->select_chip(mtd, 0);
> > 
> > My select_chip() stores the chip number (in that case 0) and uses
> > that for subsequent commands to the controller. Since there is no
> > chip 0 the read returns nothing and nand_scan() fails.
> > 
> > Probably I'm missing something, would be great if someone could
> > help me...
> 
> Actually, the chip parameter passed to ->select_chip() is not the
> CS-id from the NAND controller PoV, but the one from the NAND chip PoV.
> 
> You have to store an association table between chip⁻CS and controller-CS
> somewhere in your NAND controller private struct (a struct inheriting
> from nand_control_hw).

Thanks Boris. So far I only knew about controller-CS, so that confused
me. So I'll have:

- one nand_hw_control
- up to 8 chips (if present)

Then I probably don't need the chip->select_chip() at all.

> And remember that nand_scan_ident() is supposed to detect a single NAND
> chip, not all the chips connected to your controller (the multi-CS
> logic is here to handle multi-dies NANDs, not the case where you have
> multiple NAND chips connected to the same controller).
> 
> You can have a look at the sunxi_nand [1] driver if you want an example.

That's a much better example then what I've been looking at so far...

Thanks again,
Jan

> Regards,
> 
> Boris
> 
> [1]http://lxr.free-electrons.com/source/drivers/mtd/nand/sunxi_nand.c#L219