Re: PATA port not detected when using AHCI on Intel 82801GB/GR/GH

[infernix <infernix@infernix.net>]

Hi again,

It's been a while because I got a bad set of hard disks to start testing 
with, but I'm all set now that I've got new ones. Both the PATA port 
with the IDE DOM as well as all the SATA ports are recognised and 
working now.

PATA_LEGACY Depends on: ATA [=y] && ATA_SFF [=y] && ISA [=n] && 
EXPERIMENTAL [=y]                                            │

To make it work, PATA_LEGACY needs to have the ISA dependancy removed 
(at least on x86_64). Not sure if that's been committed by anyone yet.


So, it boots up fine, but doesn't use DMA (it's ata7+8):

[    1.399227] ata1: SATA max UDMA/133 abar m1024@0xffa3fc00 port 
0xffa3fd00 irq 26
[    1.406607] ata2: SATA max UDMA/133 abar m1024@0xffa3fc00 port 
0xffa3fd80 irq 26
[    1.413981] ata3: SATA max UDMA/133 abar m1024@0xffa3fc00 port 
0xffa3fe00 irq 26
[    1.421355] ata4: SATA max UDMA/133 abar m1024@0xffa3fc00 port 
0xffa3fe80 irq 26
[    1.442719] ata5: SATA max UDMA/100 host m128@0xff8ffc00 port 
0xff8f8000 irq 18
[    1.450014] ata6: SATA max UDMA/100 host m128@0xff8ffc00 port 
0xff8fa000 irq 18
[    1.460718] ata7: PATA max PIO4 cmd 0x1f0 ctl 0x3f6 irq 14
[    1.628360] ata7.00: CFA: TRANSCEND, 20071207, max UDMA/66
[    1.633832] ata7.00: 7831152 sectors, multi 0: LBA
[    1.638708] ata7.00: configured for PIO
[    5.671978] ata8: PATA max PIO4 cmd 0x170 ctl 0x376 irq 15

Here's the PATA port performance with the BIOS set to Compatible (ATA_PIIX):

Timing cached reads:   1750 MB in  2.00 seconds = 875.19 MB/sec
Timing buffered disk reads:   80 MB in  3.01 seconds =  26.58 MB/sec

And here's with Enhanced (AHCI + PATA_LEGACY):

Timing cached reads:   770 MB in  2.00 seconds = 385.04 MB/sec
Timing buffered disk reads:    6 MB in  3.29 seconds =   1.82 MB/sec

On 23/01/10 14:38, Alan Cox wrote:
>> That would make me think it's actually wired to the ICH. Maybe also to
>> the LPC bus but I'm not knowledgeable enough to know if thats even possible.
>
> I'm not sure what occurs if you set up the address decode on the ICH for
> the port and then hide it. Seems a strange thing to do but it might be
> the case, and if so we can probably add a Linux quirk based on the DMI
> data to unhide it.

How can I try this? And would it yield any performance benefits? It's 
pretty slow now, and although usable in a NAS setup, it would be nice to 
try and get more out of it. As far as I can tell, the hardware can do 
it, judging by the hdparm performance when it's set to compatible mode.

Or is there no more performance to be gained in this mode?

Full dmesg, dmidecode and lspci can be found at 
http://dx.infernix.net/ss4200/

Thanks!