Re: support for marvell sata chips?

Ruslan Ermilov wrote:
> On Wed, Oct 18, 2006 at 09:18:58PM +0200, S?ren Schmidt wrote:
>   
>> Ruslan Ermilov wrote:
>>     
>>> BTW, we're having data corruption issues when using Marvell 88SX6081:
>>>
>>> # dmesg |grep ^ad16
>>> ad16: 476940MB <HDS725050KLA360 K2AOAB0A> at ata8-master SATA150
>>> # dmesg | grep ^ata8
>>> ata8: <ATA channel 6> on atapci0
>>> # dmesg | grep ^atapci0
>>> atapci0: <Marvell 88SX6081 SATA300 controller> port 0x9800-0x98ff mem 
>>> 0xfc900000-0xfc9fffff irq 20 at device 1.0 on pci2
>>> # dd if=/dev/ad16 bs=128k count=1024 | md5
>>> 1024+0 records in
>>> 1024+0 records out
>>> 134217728 bytes transferred in 2.091913 secs (64160282 bytes/sec)
>>> 669aa0b8c706f06d3d41aae95d4a3cf1
>>> # dd if=/dev/ad16 bs=128k count=1024 | md5
>>> 1024+0 records in
>>> 1024+0 records out
>>> 134217728 bytes transferred in 2.092598 secs (64139273 bytes/sec)
>>> 5b92918e4beaadf5a7ff74829b9ab9ce
>>> # dd if=/dev/ad16 bs=128k count=1024 | md5
>>> 1024+0 records in
>>> 1024+0 records out
>>> 134217728 bytes transferred in 2.096472 secs (64020758 bytes/sec)
>>> d56fbdee36d70c8d6fcae6ea7b7550a7
>>> # 
>>>
>>> I'm currently looking into adding errata quirks found in the Linux
>>> driver to see if that helps.  JFYI.
>>>  
>>>       
>> Right, there are a few mentioned in the docs, however I've not been able 
>> to reproduce any of them, but please let me know what you find out!!
>>
>>     
> By removing the riser card I can no longer easily reproduce
> the corruption but colleagues said it's still reproduceable
> even without the riser card.  I'm doing more testing now,
> but so far I couldn't reproduce it.  I tried several riser
> cards, two different models, the issue is still the same;
> reading 10 sectors gives one bit error (MSB bit) at one
> byte.  Do you have experience why using riser cards may be
> a bad idea?
>   
Hmm, I've seen bad behavior with PCI-X riser cards, I guess its because 
of poor design and the high speeds that the problems occur.
Not much to do about it other than finding one that works or lowering 
the PCI-X bus speed.

-Søren