Bug 12075 (Marvell) - Kernel 2.6 fails to install / run on Marvell 6145 SATA Controller
Summary: Kernel 2.6 fails to install / run on Marvell 6145 SATA Controller
Status: CLOSED CODE_FIX
Alias: Marvell
Product: Platform Specific/Hardware
Classification: Unclassified
Component: x86-64 (show other bugs)
Hardware: All Linux
: P1 blocking
Assignee: platform_x86_64@kernel-bugs.osdl.org
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2008-11-21 04:33 UTC by andrei
Modified: 2012-05-22 16:53 UTC (History)
5 users (show)

See Also:
Kernel Version: 2.6.27.5
Subsystem:
Regression: No
Bisected commit-id:


Attachments

Description andrei 2008-11-21 04:33:22 UTC
*** Distribution:
    openSUSE



*** Hardware Environment:
   Mainboard:
   Manufacturer :       Intel Corporation (Intel) 
   Product :    975X Bad Axe 2 
   Model :      D975XBX2 
   Version :    AAD53350-508 
   Serial Number :      BQB2736000SD 
   Support MP : Yes, 4 CPU(s) 
   Version MPS :        1.4 

   Processor:
   Type :       Intel Core 2 Quad 
   Specification :     Intel(R) Core(TM)2 Quad CPU    Q6600  @ 2.40GHz 
   Model Number :       Q6600 
   Codename :   Kentsfield 
   Revision :   G0 
   Technology : 0.065ยต 
   CPU ID :     6.F.B 
   CPU IDEx :   6.F.B 
   Microcode :  MU06FBB6

   Disk Controller :
   Marvell Semiconductor (Was: Galileo Technology Ltd) 6145 SATA2 Controller 
   Mode :       SATA 
   AHCI :       Yes 
   IDE Legacy : Yes 
   NCQ :        Yes 
   Port Multiplier :    Yes 
   IO Ports :   5 available on  5 
   Version :    1.00 
   Bandwidth :  3 GB/s 
   Port #0 :    ATA Device detected  -  (Generation 1) 
   Port #1 :    ATA Device detected  -  (Generation 2) 
   Port #2 :    ATA Device detected  -  (Generation 1) 
   Port #3 :    No Device 
   Port #4 :    No Device



*** Problem Description:
   I had recently a problem with D975XBX2 Intel motherboard. Exactly speaking of disk controller "Marvell 6145 SATA2 Controller".
   There is a (pseudo) driver in the kernel for handling that controller, called pata_marvell; it claims the device, but do nothing.
   There is another driver in the kernel for handling that controller, called ahci; it claims the device, but does not work.
   I have some screenshots...
Comment 1 andrei 2008-11-21 04:56:24 UTC
   I have 2 hdds and 1 dvd-drive attached to the Marvell 6145 controller. The
Windows (XP Home Edition SP2) runns perfectly, but the SUSE 11.0, 11.1b4, 11.1b5 fails to install: doesn't recognize neither the hdds, nor the dvd-drive.
   I have taken some screenshots. Hope it helps...

   https://bugzilla.novell.com/attachment.cgi?id=250910
   https://bugzilla.novell.com/attachment.cgi?id=250911
   https://bugzilla.novell.com/attachment.cgi?id=250912
Comment 2 andrei 2008-11-21 04:58:09 UTC
   On the Asus website (http://dlsvr01.asus.com/pub/ASUS/mb/socket775/P5K_SE/Marvell.zip), you can find Marvell SATA code source and documentation that explain how to build and install the linux driver for Marvell SATA controller. This code source is for 'Marvell Storage Controller 6121/6122/6141/6145'. The code source version is v1.0.0.9 for Marvell SATA Controller and was developed by a Marvell's developer.
Comment 3 Andrew Morton 2008-11-21 10:28:14 UTC
Reassigned to ata..
Comment 4 Alan 2008-12-01 08:52:32 UTC
IRQ routing problem, reassigned to x86
Comment 5 andrei 2010-04-15 20:37:29 UTC
   It is now a year and a half since I first reported this bug: the leak of a decent driver in the standard linux kernel for "Marvell 6145 SATA2 Controller".

This controller is on a mainstream Intel motherboard (975X Bad Axe 2), so one would expect linux to be able to utilize it.
Comment 6 Tejun Heo 2010-04-15 22:18:23 UTC
Alan, why did you think it was a IRQ routing problem?  andrei, have you tried newer distros?  Also, what is the original novell bug number?
Comment 7 Alan 2010-04-15 23:11:28 UTC
The traces seem to show the IRQ not being delivered. Also the chip is known to work on other setups, although Jeff still hasn't finished his extended PATA patches tio make it really fly.
Comment 8 andrei 2010-04-21 16:01:35 UTC
"The chip just isn't popular enough and is pretty tricky to support properly."

:)

https://bugzilla.novell.com/show_bug.cgi?id=398337
Comment 9 andrei 2010-04-21 16:05:14 UTC
"drivers are kernel business"

https://bugzilla.novell.com/show_bug.cgi?id=398337
Comment 10 andrei 2010-04-21 16:16:19 UTC
(In reply to comment #6)
> Alan, why did you think it was a IRQ routing problem?  andrei, have you tried
> newer distros?  Also, what is the original novell bug number?

Marvell 6145 controller situation @ Nov 2008:
   MS WinXP Home Edition works *** perfectly *** with proprietary drivers
   openSUSE 11.0 - fails to install
   openSUSE 11.1 beta 4 - fails to install
   openSUSE 11.1 beta 5 - fails to install

Marvell 6145 controller situation @ Apr 2010:
   Not known, because now I only use the Intel SATA controller.
Comment 11 Tejun Heo 2010-05-13 16:27:37 UTC
Okay, now I remember.  The NDA process took something like six months and I lost track of things.

Jeff, this is the weird ahci-like pata chip, right?  Are you gonna work on it?

Thanks.
Comment 12 egross 2011-06-20 16:19:40 UTC
I run debian kernel 2.6.32-5-amd64 on a Asus M2V motherboard. It has a Marvell sata controller for external hd (88SE6121 SATA II). Boot hangs a long time if the external hd is attached. I found a workaround putting "options ahci marvell_enable=1" in an archive in /etc/modprobe.d. I found a discussion on that issue at debian bug report (http://bugs.debian.org/515201), where details of my and another case are dealt with. There I was adviced to put the question here too. As I do not have enough technical knowledge, I ask to look the suggestions made there, especially by Jonathan Nieder.
Thanks.
egross@bol.com.br
Comment 13 Jonathan Nieder 2012-05-14 17:02:25 UTC
To summarize: various people using SATA drives attached to Marvell controllers
with a 2.6.32.y kernel have been having trouble (failed IDENTIFY, SRST taking
a long time).  Passing ahci.marvell_enable=1 avoids trouble.  If I understand
5b66c829bf5c (ahci, pata_marvell: play nicely together, 2008-09-03) correctly,
that is not supposed to be necessary:

| The actual fix for the moment is very simple. If the user has included
| the pata_marvell driver let it drive the ports. If they've only selected
| for SATA support give them the AHCI driver which will run the port a fraction
| faster. Allow the user to control this decision via ahci.marvell_enable as
| a module parameter so that distributions can ship 'it works' defaults and
| smarter users (or config tools) can then flip it over it desired.

What upstream commit might have fixed this?
Comment 14 Jonathan Nieder 2012-05-17 18:12:14 UTC
Reopening, but information on fixed versions would still of course be welcome.
Comment 15 Jonathan Nieder 2012-05-17 18:13:28 UTC
(In reply to comment #14)
> information on fixed versions would still of course be welcome.

(and status on 3.x.y kernels in general)
Comment 16 Alan 2012-05-22 15:27:34 UTC
See

 git log drivers/ata/ahci.c
Comment 17 Jonathan Nieder 2012-05-22 16:12:01 UTC
(In reply to comment #16)
> See
>
>  git log drivers/ata/ahci.c

Sorry, it's still not obvious which commit you're referring to.

The only commit after 2.6.32 I can see that touches this code
is 394d6e535f15 ("ahci: Factor out PCI specifics from
ahci_save_initial_config()", 2010-03-03) and I do not see how
it could have changed anything.

5b66c829bf5c ("ahci, pata_marvell: play nicely together",
2008-09-03) is the commit that introduced the marvell_enable
parameter.  That can't possibly be the fix, given that these
reports mention that setting marvell_enable works around the
problem with a cost of (apparently) stopping the PATA driver
from driving the PATA ports.
Comment 18 Alan 2012-05-22 16:30:28 UTC
The marvell_enable option forces the device to be handled by the AHCI driver, which doesn't support PATA mode. Otherwise if the Marvell driver is built in the pata_marvell driver handles all the ports, but within the limits of the hardware IDE support.

The other relevant change is

 cb6643e1c38b6bd5c1594f0a45d8cf6943a6f934
Comment 19 Jonathan Nieder 2012-05-22 16:35:28 UTC
Does "within the limits of the hardware IDE support" mean failed
IDENTIFY, SRST taking a long time, and so on?

If so, I think it would probably be best to (in distro builds)
make ahci.marvell_enable default to 1, even when the pata_marvell
driver has also been enabled as a module.
Comment 20 Alan 2012-05-22 16:46:09 UTC
the idea case would be someone wrote the needed support. However Marvell only ever gave the docs to one person and they never had time to do it as its very complicated to handle. Basically they have an AHCI-like PATA mode that is unique to their devices.
Comment 21 Jonathan Nieder 2012-05-22 16:53:21 UTC
(In reply to comment #20)
> the idea case would be someone wrote the needed support. However Marvell only
> ever gave the docs to one person and they never had time to do it as its very
> complicated to handle. Basically they have an AHCI-like PATA mode that is
> unique to their devices.

Makes sense.  And looks like cb6643e1c38b ([libata] pata_marvell:
CONFIG_AHCI is really CONFIG_SATA_AHCI) will do just what one would
want in the meantime.  Thanks for digging it up.

Note You need to log in before you can comment on or make changes to this bug.