Date: Thu, 18 Jan 2007 18:09:50 -0600
From: Robert Hancock <hancockr@shaw.ca>
Subject: Re: SATA exceptions with 2.6.20-rc5
In-reply-to: <fa.O3RzvckSjB73Y0uL8P1nTXDRd6U@ifi.uio.no>
To: linux-kernel@vger.kernel.org
Cc: Larry Walton <lwalton@real.com>,
       =?ISO-8859-1?Q?Bj=F6rn_Steinbrink?= <B.Steinbrink@gmx.de>,
       jeff@garzik.org, htejun@gmail.com, jens.axboe@oracle.com
Message-id: <45B00C4E.4040803@shaw.ca>
MIME-version: 1.0
Content-type: text/plain; charset=ISO-8859-1; format=flowed
Content-transfer-encoding: 7bit
References: <fa.U/G88R1fWKOeQK3EBPHKK4MeRsQ@ifi.uio.no>
 <fa.2D0TIXbVTOgZmGg9ZJU+R7te70k@ifi.uio.no>
 <fa.hMhdefkReYJ4idUyqqEWJFnWUBE@ifi.uio.no>
 <fa.8TPWeOrcwkkHutPX5NOcJsTBO8Y@ifi.uio.no>
 <fa.b92BqwV090pDj7q0iBG6BChksbI@ifi.uio.no>
 <fa.O3RzvckSjB73Y0uL8P1nTXDRd6U@ifi.uio.no>
User-Agent: Thunderbird 1.5.0.9 (Windows/20061207)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1642
Lines: 36

I heard from Larry Walton who was apparently seeing this problem as 
well. He tried my recent "sata_nv: cleanup ADMA error handling v2" patch 
and originally thought it fixed the problem, but it turned out to only 
make it happen less often.

I wouldn't expect that patch to have an effect on this problem. If it 
seems to reduce the frequency that would tend to be further evidence of 
  some kind of timing-related issue where the code change just happens 
to make a difference.

I'll see if I can come up with a debug patch for people having this 
problem to try, which prints out when a flush command is issued and what 
interrupts happen when a flush is pending.

There is one important difference between ADMA and non-ADMA mode for 
non-DMA commands like flushes, which didn't come to mind before: ADMA 
mode uses MMIO registers on the controller whereas non-ADMA mode uses 
legacy IO registers. Posted write flushing is a concern with MMIO 
registers but not with PIO, the libata core is supposed to handle this 
but maybe it doesn't in some case(s). In fact, just looking at 
libata-sff.c there's this comment on the ata_exec_command_mmio function:

  *      FIXME: missing write posting for 400nS delay enforcement

That seems a bit suspicious..

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/

-
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/