Hello,
I think I've identified a small bug in YAFFS - and instead of merely whingeing about it,
I'll offer a patch that fixes it, how about that!
The problem is related to the 'inband-tags' option. I had to switch this on, since the vendor's
MTD driver (Freescale) does not allow access to the NAND spare area at this point in time.
After switching this on, everything worked fine until a block had to be erased. That didn't work,
and the MTD driver complained about unaligned offsets and lengths.
It turns out that the mtd interface's NAND block erase function calculates the offsets and lengths
using dev->nDataBytesPerChunk. This will work fine as long as 'inband-tags' is not specified,
as dev->nDataBytesPerChunk == dev->totalBytesPerChunk in that case.
I've corrected this by using dev->totalBytesPerChunk for the offset/length calculation. I must
have done something right, as YAFFS works very well now.
Hope this is useful,
Sven
 
---
linux/fs/yaffs2/yaffs_mtdif.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/linux/fs/yaffs2/yaffs_mtdif.c b/linux/fs/yaffs2/yaffs_mtdif.c
index aab9a12..be61233 100644
--- a/linux/fs/yaffs2/yaffs_mtdif.c
+++ b/linux/fs/yaffs2/yaffs_mtdif.c
@@ -209,14 +209,14 @@ int nandmtd_EraseBlockInNAND(yaffs_Device *dev, int blockNumber)
{
struct mtd_info *mtd = (struct mtd_info *)(dev->genericDevice);
__u32 addr =
- ((loff_t) blockNumber) * dev->nDataBytesPerChunk
+ ((loff_t) blockNumber) * dev->totalBytesPerChunk
* dev->nChunksPerBlock;
struct erase_info ei;
int retval = 0;
ei.mtd = mtd;
ei.addr = addr;
- ei.len = dev->nDataBytesPerChunk * dev->nChunksPerBlock;
+ ei.len = dev->totalBytesPerChunk * dev->nChunksPerBlock;
ei.time = 1000;
ei.retries = 2;
ei.callback = NULL;
--
1.6.3.3