From: Frans Pop <elendil@planet.nl>
To: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Subject: Re: cmd64x: irq 14: nobody cared - system is dreadfully slow
Date: Mon, 22 Jun 2009 21:01:37 +0200
User-Agent: KMail/1.9.9
Cc: David Miller <davem@davemloft.net>, sparclinux@vger.kernel.org,
       linux-ide@vger.kernel.org, linux-kernel@vger.kernel.org
References: <200906211446.38598.elendil@planet.nl> <200906221716.05983.elendil@planet.nl> <200906221938.27574.bzolnier@gmail.com>
In-Reply-To: <200906221938.27574.bzolnier@gmail.com>
MIME-Version: 1.0
Content-Disposition: inline
Content-Type: text/plain;
  charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
Message-Id: <200906222101.38586.elendil@planet.nl>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 3029
Lines: 75

On Monday 22 June 2009, Bartlomiej Zolnierkiewicz wrote:
> On Monday 22 June 2009 17:16:04 Frans Pop wrote:
> Thanks.  Please notice 0701 0701 words above -- it means that this
> device reports both SWDMA0 and MWDMA0 enabled at once (which results
> in IDE layer failing DMA tuning).
>
> The patch below should fix it

Yes, this gives back MWDMA2 for hdd.

> and it would be quite interesting to try 
> it on vanilla kernel to see if it helps with unexpected IRQ problem.

Will do later.

> However this still doesn't explain the regression fully -- we had
> ide_id_dma_bug() checks since Dec 2007 (and equivalent
> ide_dma_verbose() ones since almost forever) while 2.6.26 (which works
> fine) is much younger than that.  I suspect that there are some other
> kernel changes coming into the picture (Power Management?).  Would it
> be possible to try 2.6.2[78] and/or bisect this problem further?

I suspect commit 8d64fcd9 "ide: identify data word 53 bit 1 doesn't cover 
words 62 and 63 (take 3)":
@@ -396,15 +393,14 @@ int ide_id_dma_bug(ide_drive_t *drive)
 
 	if (id[ATA_ID_FIELD_VALID] & 4) {
 		if ((id[ATA_ID_UDMA_MODES] >> 8) &&
 		    (id[ATA_ID_MWDMA_MODES] >> 8))
 			goto err_out;
-	} else if (id[ATA_ID_FIELD_VALID] & 2) {
-		if ((id[ATA_ID_MWDMA_MODES] >> 8) &&
-		    (id[ATA_ID_SWDMA_MODES] >> 8))
-			goto err_out;
-	}
+	} else if ((id[ATA_ID_MWDMA_MODES] >> 8) &&
+		   (id[ATA_ID_SWDMA_MODES] >> 8))
+		goto err_out;


The logs I posted were from 2.6.30. I also tried 2.6.29 and that did *not* 
yet have the DMA problem. The commit above is from the 2.6.30 development 
cycle, so that fits. I expect you can verify it from the identify data.

> From: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
> Subject: [PATCH] ide: relax DMA info validity checking
>
> There are some broken devices that report multiple DMA xfer modes
> enabled at once (ATA spec doesn't allow it) but otherwise work fine
> with DMA so just delete ide_id_dma_bug().

The question is maybe: are there other devices that currently have dma 
disabled because of the (old) code and would stop working with 
ide_id_dma_bug() completely removed? The conservative thing to do I guess 
would be to reverse 8d64fcd9.


There is one thing I should mention here. I have been seeing the following 
error with this CD drive:
ide-cd: hdd: weird block size 2352
ide-cd: hdd: default to 2kb block size

This was present with 2.6.26 and also now with 2.6.31; not sure about 
older kernels. I initially saw it with a self-burned Debian installation 
CD. I also now see it with an audio CD. It does not seem to affect 
reading the disks: installations go fine and the audio CD plays without 
any problems.

Any risk this may be related to something we've been discussing so far, or 
is this a separate issue?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/