2003-03-22 11:41:41

by Justin Piszcz

[permalink] [raw]
Subject: Question about hdparm & dma.

root@war:~# hdparm -X69 /dev/hda
/dev/hda:
setting xfermode to 69 (UltraDMA mode5)
root@war:~# dmesg | tail -n 1
ide0: Speed warnings UDMA 3/4/5 is not functional.
root@war:~#

war@war:~$ dmesg|grep -i ST[0-9]
hda: ST3120024A, ATA DISK drive
war@war:~$

This is a Segeate 120GB 7200RPM drive.
On a:

SIS5513: IDE controller on PCI bus 00 dev 15
SIS5513: chipset revision 0
SIS5513: not 100% native mode: will probe irqs later

When I put the drive on a Promise Ultra ATA/133 board, it runs UDMA MODE
5 (ATA 100) just fine.

Next...

When I run hdparm -t /dev/hda on the SiS (with the settings I have shown):

root@war:~# hdparm -v /dev/hda

/dev/hda:
multcount = 16 (on)
IO_support = 0 (default 16-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 14593/255/63, sectors = 234441648, start = 0
root@war:~#

root@war:~# hdparm -t /dev/hda
/dev/hda:
Timing buffered disk reads: 64 MB in 1.60 seconds = 40.00 MB/sec

Now...

My question is, how is it possible to get > 33MB/s in only UDMA Mode 2
(the linux driver only supports up to UDMA2).

I haven't been able to figure it out.

With the same settings for the promise, and the promise, ide2=ata100
works on the command line, on the SiS/for the SiS, it does not, says it
is an invalid option, doing that or setting the dma on manually, I get
the same speed (MB/s), but is it really running at ATA/100? I mean, if
it is running in UDMA MODE 5 vs UDMA MODE 2, I would assume a little bit
of a speed boost, I remember with an older box, going from ATA/66 ->
ATA/100 increase about 2-3MB/s throughput with hdparm.

However, more importantly, when I am doing many things simultaenously, I
notice a slowdown, I did *NOT* notice this slow down on my older p3/866
+ ata/66 system, and I knew for a fact it was at ata/66, not only this,
it was a via/133 chipset and the ide[0|1]=ata66 worked as well.

So basically I am wondering if udma mode 5 will be supported for SIS
chipsets.
Secondly, I also have one of those Promise/Serial ATA raid on the
motherboard (2 serial ata/1 ata133), but that is not supported at all.

So what should I do if I want to run at UDMA MODE 5?
Should I buy another promise controller (ATA/133 PCI) and run it off of
that?

Anyone have any suggestions? Please let me know, thank you.
Please cc me as I am not on the list.




2003-03-22 13:33:45

by Alan

[permalink] [raw]
Subject: Re: Question about hdparm & dma.

On Sat, 2003-03-22 at 11:52, Justin Piszcz wrote:
> My question is, how is it possible to get > 33MB/s in only UDMA Mode 2
> (the linux driver only supports up to UDMA2).

You can overclock things if you have the setup wrong.

> So basically I am wondering if udma mode 5 will be supported for SIS
> chipsets.

I'm wondering if SiS are ever going to provide useful documentation.
SiS won't deal with individuals only companies which complicates matters

> Secondly, I also have one of those Promise/Serial ATA raid on the
> motherboard (2 serial ata/1 ata133), but that is not supported at all.

That may change if things go ok.

2003-03-22 14:30:19

by Murray J. Root

[permalink] [raw]
Subject: Re: Question about hdparm & dma.

On Sat, Mar 22, 2003 at 02:56:51PM +0000, Alan Cox wrote:
> On Sat, 2003-03-22 at 11:52, Justin Piszcz wrote:
> > My question is, how is it possible to get > 33MB/s in only UDMA Mode 2
> > (the linux driver only supports up to UDMA2).
>
> You can overclock things if you have the setup wrong.
>
> > So basically I am wondering if udma mode 5 will be supported for SIS
> > chipsets.
>
> I'm wondering if SiS are ever going to provide useful documentation.
> SiS won't deal with individuals only companies which complicates matters
>

Well, if you're not getting good docs from SiS I commend you on your abilities.
My P4S533 (SiS645DX chipset) is doing decently well with a Maxtor ATA/133 100G
drive. It's only a 5400rpm drive and nearly matches the 120G WD ATA/100 7200rpm
in my Dell for hdparm speeds. I was a little disappointed in the speed
comparisons till I saw that you're doing this blind.

Thanks.

--
Murray J. Root