DataFlash & writesize

Hello Andrew,

AFAIR you are the guy who provided DataFlash drivers. May I ask you to fix them as I suspect they may be broken in mtd-2.6.git.

We've introduced a mtd->writesize field recently and it has to be filled by the flash driver. The field contains the minimal Flash's I/O unit size. For NOR it is 1, for NAND it is NAND page, or less if multiple writes to a NAND page are supported, etc. I once read a DataFlash datasheet and suspect this has to be DataFlash sector size, but I'm not sure.

So, I'm asking you to initialize mtd->writesize correctly.

Thanks.

Best Regards,
Artem B. Bityutskiy,
St.-Petersburg, Russia.

hi,

Since the writesize change is now in mainline git, here is a trivial patch to set mtd->writesize for DataFlash.
[writesize = erasesize = pagesize]

(It compiles, but not tested).

diff -urN linux-2.6.17-git5.orig/drivers/mtd/devices/mtd_dataflash.c linux-2.6.17-git5/drivers/mtd/devices/mtd_dataflash.c
@@ -478,6 +478,7 @@
 	device->name = (pdata && pdata->name) ? pdata->name : priv->name;
 	device->size = nr_pages * pagesize;
 	device->erasesize = pagesize;
+	device->writesize = pagesize;
 	device->owner = THIS_MODULE;
 	device->type = MTD_DATAFLASH;
 	device->flags = MTD_CAP_NORFLASH;

Andrew,

I made the same yesterday and it is in GIT, but thank you anyway!

Best Regards,
Artem B. Bityutskiy,
St.-Petersburg, Russia.

What are the intended semantics for "writesize" by the way?

I noticed I was getting that first BUG_ON in mtdcore:add_mtd_device since, with NOR CFI cmdset 1 flash, with current GIT. Seems like writesie should have been set up already.

- Dave

Which git tree and when? That was a bug that should have been fixed last week in the mtd-2.6.git tree. And it should also go to Linus ASAP.

josh

?

Kernel.org GIT tree, pulled yesterday.

Does not have the latest fixups. MTD git has. Sorry for the inconveniance.

tglx

Hello David,

Design-wise, it an attempt to generalize MTD. Different flashes have different minimal I/O unit size, like 1 byte in case of NOR, 16 bytes in case of ECCed NOR, 512/2K in case of NAND, etc. There were no such a field before and we had to do ugly things like

switch (mtd->type) {
	case MTD_NORFLASH:
		blah();
		break;
	case MTD_NANDFLASH:
		blah1();

etc.

Here is a comment from include/linux/mtd.h:

/* Minimal writable flash unit size. In case of NOR flash it is 1 (even
* though individual bits can be cleared), in case of NAND flash it is
* one NAND page (or half, or one-fourths of it), in case of ECC-ed NOR
* it is of ECC block size, etc. It is illegal to have writesize = 0.
* Any driver registering a struct mtd_info must ensure a writesize of
* 1 or larger.
*/

We fixed all drivers and now it should be fine - see commit 17ffc7ba6d7ea68b8d5f55a5ca1b87163e69720d

Best Regards,
Artem B. Bityutskiy,
St.-Petersburg, Russia.

David,

Design-wise, it an attempt to generalize MTD. Different flashes have different minimal I/O unit size, like 1 byte in case of NOR, 16 bytes in case of ECCed NOR, 512/2K in case of NAND, etc. There were no such a field before and we had to do ugly things like

switch (mtd->type) {
	case MTD_NORFLASH:
		blah();
		break;
	case MTD_NANDFLASH:
		blah1();

etc.

Here is a comment from include/linux/mtd.h:

/* Minimal writable flash unit size. In case of NOR flash it is 1 (even
* though individual bits can be cleared), in case of NAND flash it is
* one NAND page (or half, or one-fourths of it), in case of ECC-ed NOR
* it is of ECC block size, etc. It is illegal to have writesize = 0.
* Any driver registering a struct mtd_info must ensure a writesize of
* 1 or larger.
*/

We fixed all drivers and now it should be fine - see commit 17ffc7ba6d7ea68b8d5f55a5ca1b87163e69720d

Best Regards,
Artem B. Bityutskiy,
St.-Petersburg, Russia.

OK, thanks. For some reason the "writesize" I found didn't include the <linux/mtd/mtd.h> comment you mentioned... I guess that was the one in mtd-abi.h instead.

It's not fixed upstream yet.

- Dave