lock bit and erase errors

[Josh Boyer]

On Mon, 2005-03-14 at 13:08 -0600, Josh Boyer wrote:
> Caveat:  I am using some pretty old code in this case, but the current
> code is also questionable so bear with me.
> 
> I can't for the life of me figure out what the following code from
> do_erase_oneblock cfi_cmdset_0001.c is doing:
> 
> 	/* check for lock bit */
> 	if (status & CMD(0x3a)) {
> 		unsigned char chipstatus = status;
> 		if (status != CMD(status & 0xff)) {
> 			int i = cfi->interleave;
> 			for (i = 1; i<cfi->interleave; i++) {
> 				      chipstatus |= status >> (cfi->device_type * 8);
> 			}
> 			printk(KERN_WARNING "Status is not identical for all chips: 0x%x. Merging to give 0x%02x\n", status, chipstatus);
> 		}
> 
> 
> My device is a 16-bit device, the interleave is 1, and I have a few
> blocks with the lock bit on.  When I do a write to /dev/mtdblock/0/0, an
> erase is attempted.  When I run an older version of this code, I get:
> 
> "Status not identical for all chips: 0xa200.  Merging to give 0x00"
> 
> What I don't understand is how any of the above code is expected to
> work.  A 1 byte variable (chipstatus) is assigned a 4 byte value
> (status), which gets truncated.  Then status is shifted by 16 bits (in
> my case), which still means chipstatus is still 0.  When all is said and
> done, I get the merging printk but none of the checks below this code
> match.  That doesn't seem right...
> 
> I hacked the older code for my specific instance to do:
> 
> chipstatus = status >> 8;
> 
> and I get the proper return code for my situation (-EROFS).  I still
> don't get an error in the application writing to /dev/mtdblock/0/0, but
> at least the chip driver is actually returning an error.
> 
> Any ideas?
> 

Urm, I should also point out that I have intentionally locked the
eraseblocks on this device, so I _expect_ it to fail.  But the driver
didn't return a bad return code, which is what I'm confused about.

And I'm also confused as to why the app didn't get a bad return code
even when I "fixed" the driver to error out.  But one thing at a time.

josh