A couple of bugs in nand_write_ecc()

A couple of bugs in nand_write_ecc()

[Alex Samoutin]

I've  found a couple of bugs in nand.c driver, or may be I just don't
understand how it should works. Please, take a look.
Function nand_write_ecc() mast write buffer to NAND with any length  and
starting with any byte. But when the last byte
not finished at sector border   nand_write_ecc() write extra bytes out of
buffer :
 /*
                * Check, if we have a full page write, then we can
                * use the given buffer, else we have to copy
                */

               if (!col && ((len - written) >= mtd->oobblock))  {
                       this->data_poi = (u_char*) &buf[written];
                       cnt = mtd->oobblock;
               } else {
                       cnt = 0;
>>>>>>>>>>>                    for (i = col; i < len && i < mtd->oobblock;
i++) {
                               this->data_buf[i] = buf[written + i];
                               cnt++;
                       }
                       this->data_poi = this->data_buf;
               }


Market line should be
for (i = col; i < (len-written) && i < mtd->oobblock; i++) {

Then we invoke    nand_write_page() function

               if (eccbuf) {
                       ret = nand_write_page (mtd, this, page, col, cnt
,&eccbuf[oob], oobsel);
                       oob += mtd->oobsize;
               } else
                       ret = nand_write_page (mtd, this, page, col, cnt,
NULL, oobsel);

Where the "col" is offset form page start and "cnt" is leght of data. But
nand_write_page ()
defined as:

static int nand_write_page (struct mtd_info *mtd, struct nand_chip *this,
                           int page, int col, int last, u_char *oob_buf, int
oobsel)

where "col" mean the same, but "last" is offset of last byte of data!

So, we have to use:

   ret = nand_write_page (mtd, this, page, col, col+cnt, NULL, oobsel)

Am I right?

Alex.

A couple of bugs in nand_write_ecc()

[Thomas Gleixner]

On Friday 14 February 2003 18:43, Alex Samoutin wrote:
> Am I right?
Not really. 
But it's no more relevant. I removed the possiblibity to write not page 
aligned data to NAND. 

-- 
Thomas
________________________________________________________________________
linutronix - competence in embedded & realtime linux
http://www.linutronix.de
mail: tglx at linutronix.de

A couple of bugs in nand_write_ecc()

[Alex Samoutin]

From: "Thomas Gleixner" <tglx@linutronix.de>
> But it's no more relevant. I removed the possiblibity to write not page 
> aligned data to NAND. 


Ok, But what I have to do with this, 
 JFFS2 still trying to write not sector alligned data:
debian:~# rm /mnt/nand/a*
nand_read_ecc: from = 0x0000c800, len = 16
nand_read_ecc: from = 0x0000c834, len = 16
nand_read_ecc: from = 0x0000c868, len = 16
nand_read_ecc: from = 0x0000c89c, len = 16
nand_read_ecc: from = 0x0000c8d8, len = 16
nand_read_ecc: from = 0x0000c910, len = 16
nand_read_ecc: from = 0x00198000, len = 1052
nand_write_ecc: to = 0x0000c940, len = 1052
nand_write_ecc: Attempt to write not page aligned data
				nand_write_ecc()
				nand_write()
				part_write()
    call satck 			jffs2_flash_write()
        			jffs2_garbage_collect_pristine()
				jffs2_garbage_collect_live()
				jffs2_garbage_collect_pass()
				jffs2_garbage_collect_thread ()
							
Write of 1052 bytes at 0x0000c940 failed. returned -22, retlen 1052
nand_read_ecc: from = 0x00198000, len = 1052
nand_write_ecc: to = 0x0000cd5c, len = 1052
nand_write_ecc: Attempt to write not page aligned data
Write of 1052 bytes at 0x0000cd5c failed. returned -22, retlen 1052
nand_read_ecc: from = 0x0000d178, len = 16
jffs2_flash_writev(): Non-contiguous write to 0000d178
wbuf was previously 0000c800-0000c93f
kernel BUG at wbuf.c:282!
Oops: Exception in kernel mode, sig: 4

Call backtrace:

jffs2_flash_writev
jffs2_write_dirent
jffs2_do_unlink
jffs2_unlink
vfs_unlink
ret_from_syscall_1

Alex

A couple of bugs in nand_write_ecc()

[Thomas Gleixner]

On Saturday 15 February 2003 02:05, Alex Samoutin wrote:
> From: "Thomas Gleixner" <tglx@linutronix.de>
>
> Ok, But what I have to do with this,
>  JFFS2 still trying to write not sector alligned data:
I know, will be fixed today
-- 
Thomas
________________________________________________________________________
linutronix - competence in embedded & realtime linux
http://www.linutronix.de
mail: tglx at linutronix.de

A couple of bugs in nand_write_ecc()

[Thomas Gleixner]

On Saturday 15 February 2003 02:05, Alex Samoutin wrote:
> From: "Thomas Gleixner" <tglx@linutronix.de>
> Ok, But what I have to do with this,
>  JFFS2 still trying to write not sector alligned data:
Fixed in CVS. cvs up. 
Thanks for pointing it out.
-- 
Thomas
________________________________________________________________________
linutronix - competence in embedded & realtime linux
http://www.linutronix.de
mail: tglx at linutronix.de