NAND Flash Framework for review

[Luiz Otavio O Souza]

On Mar 14, 2010, at 4:58 AM, Andrew Turner wrote:

> On Tue, 9 Mar 2010 00:28:53 +0100
> Rafal Jaworowski <raj at semihalf.com> wrote:
>> We are looking for review, comments and any other feedback.
> 
> Below is my first set of notes from reading the code. I haven't ported
> the driver to my hardware yet so these are only from using with nandsim.
> 
> Chip drivers:
> - lnand and snand have magic numbers to figure out which drive to use.
>  We should move these to a flag in the chip parameters.

We just need to add the chip size in nand_params and based on that we can calculate the number of address cycles (see below) and the type of chip (if chip >= 128MB and pagesize > 512 then you have a large page device).

> - nand_read_pages should be a device method so we can customise how we
>  handle reading of pages.
> - Split nand_generic into separate files with different device options
>  so we can compile only the one we need.
> - nand_generic.c should be split into multiple files to allow us to
>  support only the chips we have.

- Fixed number of address cycles (and wrong count for large page devices).
- Address cycles should not be sent by "driver" (this need to be moved up).
- No support to read the ready/busy nand pin.
- Bad block table for generic devices.

I'm not sure about slipt nand_generic.c, actually we can use the same kernel with different kind of chips (exactly what i'm doing here). Also there are only a few functions for each case.


> nandbus:
> - nandbus should not know what commands to send to nand chips. These
>  requests should be moved up to the chip driver.
> - Why is nandbus not exposing it's interface via NEWBUS?
> - Where is nandbus_destroy used?
> - Why is malloc called with M_NOWAIT when allocating the nandbus ivar?
> - What should happen when nandbus_send_command, nandbus_send_address,
>  nandbus_start_command, nandbus_read_buffer or nandbus_write_buffer
>  fail?
> - Why does nandbus need to know if we are an ONFi part or not? We
>  can push the check into the onand driver's probe function.

- nandbus have nandbus_select_cs() but not the equivalent nandbus_deselect_cs().

> 
> nandsim:
> - nandsim sample config has invalid time values (too small.
> - There is a kernel crash in nandsim_log.
> 
> General:
> - malloc with M_WAITOK will always succeed, no need to check return
>  value.
> - Should define our own malloc type.
> 
> Ideas:
> - Can we move ecc and bbt handling into GEOM? This will allow us to
>  bypass them when required.

This is a mandatory feature (disable ecc and may be the bbt checks) if you need to deal with some kind of unknown nand FS or unknown nand oob layout (like make a backup of your unknown nand data before erase it).

We also need userland tools to erase nand contents (erase all blocks of a device or slice), read and write raw pages (so you can make the full backup/restore of nand contents or write a yaffs like image - which already contains all the oob data).

Luiz