Promise IDE DMA issue

Promise IDE DMA issue

[John Wong]

I recently upgraded fron a nVidia nForce2 MCP-T based A7NX-DX
motherboard to an A8V DX, Via K8T800 Pro.  Now occassionally, I get 
DMA issues on a drive attached to a Promise 133 TX2 controller (20269).

Jan  1 11:14:51 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:16:01 gambit kernel: hdh: dma_timer_expiry: dma status == 0x64
Jan  1 11:16:01 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:16:01 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:16:01 gambit kernel: hde: lost interrupt
Jan  1 11:16:01 gambit kernel: PDC202XX: Secondary channel reset.
Jan  1 11:16:01 gambit kernel: hdh: DMA interrupt recovery
Jan  1 11:16:01 gambit kernel: hdh: lost interrupt
Jan  1 11:16:01 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:16:01 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:16:01 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:16:01 gambit kernel: hde: lost interrupt
Jan  1 11:16:01 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:16:01 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:18:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:18:31 gambit kernel: hde: lost interrupt
Jan  1 11:18:31 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:18:31 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:18:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:18:31 gambit kernel: hde: lost interrupt
Jan  1 11:18:31 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:18:31 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:18:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:18:31 gambit kernel: hde: lost interrupt
Jan  1 11:18:31 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:18:31 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:18:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:18:31 gambit kernel: hde: lost interrupt
Jan  1 11:18:31 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:19:31 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:19:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:19:31 gambit kernel: hde: lost interrupt
Jan  1 11:19:31 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:19:31 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:19:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:19:31 gambit kernel: hde: lost interrupt
Jan  1 11:19:31 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:19:31 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:19:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:19:31 gambit kernel: hde: lost interrupt
Jan  1 11:19:31 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:19:31 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:19:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:19:31 gambit kernel: hde: lost interrupt
Jan  1 11:19:51 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:20:01 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:20:01 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:20:01 gambit kernel: hde: lost interrupt
Jan  1 11:20:21 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:20:31 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:20:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:20:31 gambit kernel: hde: lost interrupt
Jan  1 11:20:51 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:21:01 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:21:01 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:21:01 gambit kernel: hde: lost interrupt
Jan  1 11:21:21 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:21:31 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:21:31 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:21:31 gambit kernel: hde: lost interrupt
Jan  1 11:21:51 gambit kernel: hde: dma_timer_expiry: dma status == 0x64
Jan  1 11:22:01 gambit kernel: PDC202XX: Primary channel reset.
Jan  1 11:22:01 gambit kernel: hde: DMA interrupt recovery
Jan  1 11:22:01 gambit kernel: hde: lost interrupt

hde and hdh are identical Seagate ST3200822A
hdf and hdg are Maxtor 6Y120P0 and 6Y120L0, however, they weren't
mounted and are manually mounted as needed as they are FAT32 and NTFS.

Jan  2 06:25:35 gambit kernel: hde: dma_intr: status=0x50 { DriveReady
SeekCompl
ete }
Jan  2 06:25:35 gambit kernel: 
Jan  2 06:25:35 gambit kernel: ide: failed opcode was: unknown

The above happened at 6:25 which coincides with Debian's cron.daily
runs (most likely the updating of the locate database).


I've seen http://seclists.org/lists/linux-kernel/2003/Oct/4754.html
where someone suggests disabling autotune to work around the problem.
Without DMA, the DMA problem shouldn't happen, but this would slow
things down.

http://www.uwsg.iu.edu/hypermail/linux/kernel/0412.2/1569.html
mentions about needing multiple independent PCI buses for more than 2
DMA transfers at a time.

My lspci output from the old A7NX (nVidia nForce2) does appear to have 
more PCI buses.  It looks to have a 0, 1, 2, 3 with 3 looking to be the 
AGP bus.  The A8V (Via K8T800 Pro) on the other hand looks to have just 
0, and 1 with 1 being the AGP bus.


John

Re: Promise IDE DMA issue

[Jeff Garzik]

John Wong wrote:
> I recently upgraded fron a nVidia nForce2 MCP-T based A7NX-DX
> motherboard to an A8V DX, Via K8T800 Pro.  Now occassionally, I get 
> DMA issues on a drive attached to a Promise 133 TX2 controller (20269).

I would try fiddling with BIOS settings, and make sure you have the 
latest BIOS.

	Jeff

Re: Promise IDE DMA issue

[John Wong]

Latest 1.009 BIOS flashed last night.  I'll try out some BIOS settings, 
but the settings work fine with Windows XP.  That's why I think it could 
be something with the driver.  This is with the PDC202XX_NEW on kernel 
2.6.10  The DMA timeout happens sporadically, but as of yet, has yet to
reoccur.  The change from 1.008 to 1.009 mentions nothing about the
Promide IDE.  

John

On Mon, Jan 03, 2005 at 01:00:59PM -0500, Jeff Garzik wrote:
> John Wong wrote:
> >I recently upgraded fron a nVidia nForce2 MCP-T based A7NX-DX
> >motherboard to an A8V DX, Via K8T800 Pro.  Now occassionally, I get 
> >DMA issues on a drive attached to a Promise 133 TX2 controller (20269).
> 
> I would try fiddling with BIOS settings, and make sure you have the 
> latest BIOS.
> 
> 	Jeff
> 
> 
> 
> 

Re: Promise IDE DMA issue

[Joel Cant]

John Wong wrote:

>Latest 1.009 BIOS flashed last night.  I'll try out some BIOS settings, 
>but the settings work fine with Windows XP.  That's why I think it could 
>be something with the driver.  This is with the PDC202XX_NEW on kernel 
>2.6.10  The DMA timeout happens sporadically, but as of yet, has yet to
>reoccur.  The change from 1.008 to 1.009 mentions nothing about the
>Promide IDE.  
>
>John
>
>On Mon, Jan 03, 2005 at 01:00:59PM -0500, Jeff Garzik wrote:
>  
>
>>John Wong wrote:
>>    
>>
>>>I recently upgraded fron a nVidia nForce2 MCP-T based A7NX-DX
>>>motherboard to an A8V DX, Via K8T800 Pro.  Now occassionally, I get 
>>>DMA issues on a drive attached to a Promise 133 TX2 controller (20269).
>>>      
>>>
>>I would try fiddling with BIOS settings, and make sure you have the 
>>latest BIOS.
>>
>>	Jeff
>>
>>
>>
>>
>>    
>>
Sounds simlar to the problems i'm having with the channels resetting 
under heavy load, and then fudging DMA, and i'm not the only one havign 
these issues, seems its a common problem with these cards, not sure if 
theres been some slight changes in the chip itself or wether there is a 
fault with the kernel driver, as you say, it seems that the problem does 
not occour under windows.

Joel

Re: Promise IDE DMA issue

[John Wong]

On Mon, Jan 03, 2005 at 10:22:58PM +0000, Joel Cant wrote:
> John Wong wrote:
> 
> >Latest 1.009 BIOS flashed last night.  I'll try out some BIOS settings, 
> >but the settings work fine with Windows XP.  That's why I think it could 
> >be something with the driver.  This is with the PDC202XX_NEW on kernel 
> >2.6.10  The DMA timeout happens sporadically, but as of yet, has yet to
> >reoccur.  The change from 1.008 to 1.009 mentions nothing about the
> >Promide IDE.  
> >
> >John
> >
> >On Mon, Jan 03, 2005 at 01:00:59PM -0500, Jeff Garzik wrote:
> > 
> >
> >>John Wong wrote:
> >>   
> >>
> >>>I recently upgraded fron a nVidia nForce2 MCP-T based A7NX-DX
> >>>motherboard to an A8V DX, Via K8T800 Pro.  Now occassionally, I get 
> >>>DMA issues on a drive attached to a Promise 133 TX2 controller (20269).
> >>>     
> >>>
> >>I would try fiddling with BIOS settings, and make sure you have the 
> >>latest BIOS.
> >>
> >>	Jeff
> >>
> >>
> >>
> >>
> >>   
> >>
> Sounds simlar to the problems i'm having with the channels resetting 
> under heavy load, and then fudging DMA, and i'm not the only one havign 
> these issues, seems its a common problem with these cards, not sure if 
> theres been some slight changes in the chip itself or wether there is a 
> fault with the kernel driver, as you say, it seems that the problem does 
> not occour under windows.
> 
> Joel

Before my motherboard upgrade, the Promise 133TX2 used to be a 100TX2.  
I didn't have this problem then, but then, I also had more PCI buses 
then according to the outputs of lspci.

The DMA problem does only appear to happen with me under load as well,
torrenting to one drive, and probably updatedb on another.

John