Re: 2.6.25-rc1/2 CD/DVD burning broken

[Kiyoshi Ueda <k-ueda@ct.jp.nec.com>]

Hi Andreas,

On Sat, 16 Feb 2008 21:52:21 +0100, Andreas Schwab wrote:
> Since commit aaa04c28cb9a1efd42541fdb7ab648231c2a2263 [blk_end_request:
> changing ide-cd (take 4)] I cannot burn any CD/DVD any more, getting the
> following error from wodim:
> 
> Errno: 0 (Success), write_g1 scsi sendcmd: no error
> CDB:  2A 00 00 00 00 00 00 00 1F 00
> status: 0x2 (CHECK CONDITION)
> Sense Bytes: 70 00 05 00 00 00 00 0E 00 00 00 00 21 02 00 00
> Sense Key: 0x5 Illegal Request, Segment 0
> Sense Code: 0x21 Qual 0x02 (invalid address for write) Fru 0x0
> Sense flags: Blk 0 (not valid) 
> resid: 63488

Could you try this patch?
I've only done a compile test, so this patch may not work.

During the conversion to blk_end_request, the code was changed
*not* to set rq->data_len = 0.
I removed that part because I thought it was just a trigger to
call post_transform_command().  However, since data_len can be
used as a residual length of the transfer, it might have to remain
there.
Actually, wodim seems checking the residual count how far it wrote
(e.g. wodim/wodim.c:write_track_data()).

This patch brings back the rq->data_len = 0.

--- 2.6.25-rc2/drivers/ide/ide-cd.c	2008-02-15 15:57:20.000000000 -0500
+++ ide-fix/drivers/ide/ide-cd.c	2008-02-18 01:23:40.000000000 -0500
@@ -1207,9 +1207,13 @@ static ide_startstop_t cdrom_newpc_intr(
 end_request:
 	if (blk_pc_request(rq)) {
 		unsigned long flags;
+		unsigned int dlen = rq->data_len;
+
+		if (dma)
+			rq->data_len = 0;
 
 		spin_lock_irqsave(&ide_lock, flags);
-		if (__blk_end_request(rq, 0, rq->data_len))
+		if (__blk_end_request(rq, 0, dlen))
 			BUG();
 		HWGROUP(drive)->rq = NULL;
 		spin_unlock_irqrestore(&ide_lock, flags);

Thanks,
Kiyoshi Ueda