On Thursday 09 November 2006 16:48, Aubrey wrote:
> Hi Charles,
>
> Thanks for your reply.
> Now the kernel can be compiled.
>
> But I got some other issues:
>
> 1) everytime I mount the mtd-nand partition, everything works fine
> except it cost about 8 seconds to mount, I think it's too long. Is
> that correct? If I recall correctly, the former version cost no more
> than one second.
>     My nand part number is NAND128W3A, which is 128Mbit, 512bytes page.

This is the 32MBytes, 512-byte part. This will work with yaffs1.

The mount time does seem a bit excessive, but there are two things to 
consider:
1) The 512byte yaffs1 stuff does not support "checkpointing" so must scan from 
scratch each boot.
2) Still, this is pretty slow. I expect that your driver set-up is slow.

>
> root:> mount -t yaffs /dev/mtdblock2 /mnt
>
> 2) The same driver, If I change the nand chip to NAND01GW3B, which is
> 1Gbit, 2k byte page, I got every blocks are bad. I think there should
> be something wrong in my driver. Is there any existing mtd-nand driver
> support 2K page nand flash?

This part should work with yaffs2. yaffs2 supports checkpointing, so remounts 
should be way faster.

Did you erase the part first?

If you're seeing bad blocks then most likely:
1) You did not erase the part first.
- or -
2) Your driver is bad.

Most of the performance is due to the driver. People who want really serious 
performance write their own drivers.

I have not personally used yaffs on a 2.6.x board. I have used it on a 2.4.x 
ARM 200Mhz board with 128Mbyte (1G bit) 2k page device. With checkpointing I 
get sub-second boots. See 
http://aleph1.co.uk/lurker/message/20060508.194557.d9e329f9.en.html, and 
better numbers have been seen on similar hardware with a hot-rodded driver.

-- Charles