JFFS2 wbuf non-contiguous write bug

[Alexander Belyakov]

On 6/29/06, Jörn Engel <joern at wohnheim.fh-wedel.de> wrote:
>
> How can that be?  The removed code was:
> -       if (jffs2_nor_ecc(c)) {
> -               if (((c->wbuf_ofs % c->sector_size) == 0) && !c->wbuf_len) {
> -                       c->wbuf_ofs = PAGE_DIV(to);
> -                       c->wbuf_len = PAGE_MOD(to);
> -                       memset(c->wbuf,0xff,c->wbuf_pagesize);
> -               }
> -       }
>
> And jffs2_nor_ecc(c) would have been false for Sibley (true for
> STMicro, though).  So if we have a bug for Sibley and adding this code
> solves the problem _now_, how could it have solved the problem
> _before_?
>

True. But actually that piece of code was (before migration to GIT)

	if (jffs2_nor_ecc(c) || jffs2_nor_wbuf_flash(c)) {
		if (((c->wbuf_ofs % c->sector_size) == 0) && !c->wbuf_len) {
			c->wbuf_ofs = PAGE_DIV(to);
			c->wbuf_len = PAGE_MOD(to);
			memset(c->wbuf,0xff,c->wbuf_pagesize);
		}
	}

Your commit just was one which removed that check completely. Probably
jffs2_nor_wbuf_flash(c) check here was just missed duirng migration to
GIT and "[JFFS2] Teach JFFS2 about Sibley flash" commit is just
incomplete.

http://kernel.org/git/?p=linux/kernel/git/dwmw2/mtd-2.6.git;a=commit;h=59da721a2288b8aec751a2716f7ab60f2ea0c925

Dunno.

Anyway I was speaking _not_ about reverting part of your patch but
about getting back following code for Sibley, since it indeed shows if
we moved to a new eraseblock.

		if (((c->wbuf_ofs % c->sector_size) == 0) && !c->wbuf_len) {
			c->wbuf_ofs = PAGE_DIV(to);
			c->wbuf_len = PAGE_MOD(to);
			memset(c->wbuf,0xff,c->wbuf_pagesize);
		}

> You must be missing something, but I don't know what.  Maybe Josh has
> some insight into this problem - he did a thorough review and asked
> many probing questions about this particular part of the patch.

It would be really nice, since current code seems to be broken somehow.

Thanks,
Alexander Belyakov