Topic: Can't enable DMA on ASUS A7v133

[Neil L]

I have installed e-smith 4.1.2 on and ASUS A7V133 w/ Athlon 1.4 GHz and 2x IBM 40 GB/7200 RPM drives.  The drives are connected to the on-board ATA100 ports (promise PDC20265 controller).  Everything boots and runs properly (although I get an occasional Squid error- investigating).

The problem is that the disks are S-L-O-W!  hdparm gives a speed of 8.6 Mb/sec.  Checking the disk params, DMA is not enabled.  When I send the command
hdparm -d1 /dev/hde, I get the messages "HDIO_SET_DMA failed: Operation not permitted.  using_dma = 0 (off)"

This seems to be a known problem with the 2.2.16 kernel, and I only find patches/fixes/solutions discussed in the 2.4.X kernels (maybe something in 2.2.19...?)

Anybody have any ideas?  I'm willing to re-compile a kernel (which I've already tried- I couldn't boot!) if I can get some guidance.  Will a 2.4.X kernel work, and are there any that are pre-compiled for e-smith?

TIA
Neil L
neil-spamfree@obi.com
(remove "spamfree" from e-mail address)

[Lloyd Keen]

Neil,
Hve you seen the howto relating to this? If not you can view it here http://www.e-smith.org/docs/howto/contrib/ata100.txt

[Neil L]

Thanks- I have seen this and I used it to get e-smith installed on a RAID-1 with the hard drives connected to the Promise ATA100 ports.  Worked fine EXCEPT that I get truly lousy performance- only about 8.5 MB/sec.  The problem is that I cannot enable DMA mode using this particular combination of the e-smith kernel (2.2.16) and hardware.  This is a known problem with this series of kernels, and is apparently fixed in versions 2.4.X.

Any advice on upgrading kernels?

[Neil L]

Wel  I didn't really recieve any usefull help from the board, but I've found a not-so-awful fix anyway.  I tried -ALOT- of different things and I will summarize them briefly:

1)  Install pre-compiled kernel 2.2.19- didn't help.  I still couldn't enable DMA on the hard drives and the speed was throttled to about 8.5 Mb/sec.  Also had module problems (couldn't do something with USB...)  Gave up.

2)  Installed Promise Ultra66 card- passed the right memory locations to the card at boot and got to the install screen.  Typed "accept" and it went through the whole install.  When it reacheds the post-install (i.e. time to make a boot disk)  it got a kernel panic and "exited abnormally".  *sigh*  Did the install from the Promise ATA100 ports again then moved the disks to the Ultra66 card.  Edited lilo to pass the right memory locations (ide2=... ide3=...) and it booted!  It would never get past loading usb_uhci module though and hung every time.  Couldn't figure it out and gave up...(ideas here?)

(Boss is looking over my shoulder and very skeptical of e-smith at this point.)

3)  Installed e-smith on disks attached to Promise ATA100 controller.  Compiled my own kernel on a seperate RH7.0 box.  See (1) above.

4)  Installed 2.4.6/2.4.7 kernel(both pre-compiled from rpm and home-made).  Too many problems to talk about.

5)  Finally!-  I put the CDROM and the tape drive on one on-board (VIA) IDE port (master/slave respectively) and the two hard drives in master/slave on the second VIA IDE channel.  Everything boots and DMA is enabled.  Each drive individually gives hdparm -T of about 35.8 MB/sec, the RAID device (/dev/md1) gives about 21 MB/sec.  Not great but not unexpected for this arrangement, and still better than 8.5 MB/sec attached to the promise controller.  The price we pay for redundancy...

So I'm still open to suggestions on how to get the  Promise PDC20265 chip working.  The only fix I've seen even suggested so far is to use a 2.4 series kernel (WHY won't they open-source their drivers!).

Any idea on when the next version of e-smith is coming out?!

Neil L