Bug 14275

Summary: kernel>=2.6.31: ahci.c: do not force unconditionally sb600 to 32bit dma any more?
Product: IO/Storage Reporter: gabriele balducci (balducci)
Component: Serial ATAAssignee: Tejun Heo (tj)
Status: CLOSED CODE_FIX    
Severity: normal CC: rjw, shane.huang
Priority: P1    
Hardware: All   
OS: Linux   
Kernel Version: >=2.6.31 Subsystem:
Regression: Yes Bisected commit-id:
Bug Depends on:    
Bug Blocks: 13615    
Attachments: sb600-32bit-only-by-default.patch

Description gabriele balducci 2009-09-30 15:02:01 UTC
Hello,

apologies if I am missing some blatant point.

After upgrading from 2.6.30.5 to 2.6.31 one of my machines was not
booting any more, complaining like this (sata disks):

ata2.00: qc timeout (cmd 0xec)
ata2.00: failed to IDENTIFY (I/O error, err_mask=0x4)
ata1.00: qc timeout (cmd 0xec)
ata1.00: failed to IDENTIFY (I/O error, err_mask=0x4)
ata3.00: qc timeout (cmd 0xa1)
ata3.00: failed to IDENTIFY (I/O error, err_mask=0x4)

Same after installing 2.6.31.1.

It turned out that the problem is the sb600 SATA controller lying
about its 64bit dma capability.

As far as I seem to understand from ahci.c perusal, in >=2.6.31 the
sb600 is forced to 32bit dma *ONLY* on ASUS M2A-VM mother board, while
it was unconditionally (?) reset to 32bit dma before 2.6.31.

I am running on an ASROCK AliveXFire-eSATA2 mother board and
experience the fake 64bit dma advertising of the sb600 controller.

I know 0 (zero) about kernel and/or ATA coding; I can only say that
the following patch:

*** drivers/ata/ahci.c.ORIG     Wed Sep 30 15:37:54 2009
--- drivers/ata/ahci.c  Wed Sep 30 15:37:54 2009
***************
*** 433,438 ****
--- 433,439 ----
        [board_ahci_sb600] =
        {
                AHCI_HFLAGS     (AHCI_HFLAG_IGN_SERR_INTERNAL |
+                                AHCI_HFLAG_32BIT_ONLY | 
                                 AHCI_HFLAG_NO_MSI | AHCI_HFLAG_SECT255),
                .flags          = AHCI_FLAG_COMMON,
                .pio_mask       = ATA_PIO4,


fixes everything for me.

My impression is that the sb600 forcing should be done in more cases
than the single ASUS M2A-VM, or even unconditionally like it used to
be before 2.6.31 (but, as I said above, my understanding is poor and I
might be totally wrong).

I will be glad to send any other detail (dmesg, lspci, mobo specs, ...) if needed.

Hope that libata people will fix this problem easily.


thanks for all your work

ciao
gabriele
Comment 1 Rafael J. Wysocki 2009-10-01 20:41:30 UTC
Patch : http://bugzilla.kernel.org/show_bug.cgi?id=14275#c0
Comment 2 Tejun Heo 2009-10-03 08:23:37 UTC
bug#14311 is probably the same problem and you're right.  Enabling 64bit DMA on sb600 just doesn't seem to worth all these regressions.  All these machines have iommu anyway.  Shane, I'm disabling 64bit dma for all sb600's.

Thanks.
Comment 3 Tejun Heo 2009-10-03 09:28:41 UTC
Created attachment 23246 [details]
sb600-32bit-only-by-default.patch

Can you please verify this patch?
Comment 4 gabriele balducci 2009-10-05 06:53:24 UTC
(In reply to comment #3)
> Created an attachment (id=23246) [details]
> sb600-32bit-only-by-default.patch
> 
> Can you please verify this patch?

sorry for the delay: I was out for the weekend

Patched 2.6.32rc1 and all works fine (I mean: the system boots without problems and sb600 is forced to 32bit dma).


thanks and ciao
gabriele
Comment 5 Tejun Heo 2009-10-06 08:15:05 UTC
Patch included upstream.  Resolving as FIXED.  Oooh, already marked as FIXED.  Anyways, thanks.
Comment 6 Rafael J. Wysocki 2009-10-12 21:35:21 UTC
On Monday 12 October 2009, Chuck Ebbert wrote:
> On Mon, 12 Oct 2009 01:01:09 +0200 (CEST)
> "Rafael J. Wysocki" <rjw@sisk.pl> wrote:
> 
> > 
> > Bug-Entry   : http://bugzilla.kernel.org/show_bug.cgi?id=14275
> > Subject             : kernel>=2.6.31: ahci.c: do not force unconditionally
> sb600 to 32bit dma any more?
> > Submitter   : gabriele balducci <balducci@units.it>
> > Date                : 2009-09-30 15:02 (12 days old)
> > Patch               : http://bugzilla.kernel.org/show_bug.cgi?id=14275#c0
> > 
> 
> Already marked fixed in bugzilla.
> 
> Fixed by commit 2fcad9d271
> ("ahci: disable 64bit DMA by default on SB600s")
> 
> Not in 2.6.31-stable.