[2.3.3x] ALI M15x3 chipset support (EXPERIMENTAL)

[Tom Crane]

In article <fa.d2j3fcv.f3ks2g en ifi.uio.no>, Luca Montecchiani <m.luca en teamfab.it> writes:
> Andrzej Krzysztofowicz wrote:

Thanks for all the followups.

>  
>> Unless your BIOS does set an apropriate PCI cinfiguration register properly.
>> Mine does not. Whatever PCI clock I set (jumpers) on the MoBo I always get 33
>> MHz from /proc/ide/ali ...
> 
> Agree, jumpers will never died ;)

I get 33MHz from /proc/ide/ali. AFAICT there are no jumpers on my MB for 
setting this.

>> Chih-Jen Tsai <cjtsai en ali.com.tw> suggested that UDMA problems are WD drives
>> specyfic and disappear when CRC errors are ignored. I'm not sure if it is
>> easy to ignore CRC errors for specyfic hardware combinations only, but if it
>> is - it could be a workaround of the problem.
> 
> Well, WDC disks are safely detected and driver fall back to dma2, for some
> strange reasons that I can't remember my first hack failed to detect a WDC
> and bonnie eat the filesystem, after that I've also put an extra hdparm -d0 /dev/hdc
> to my rc.local ;)
> 
>> Could you send me the patch you use for Rev.20/UDMA, please ?
> 
> patch ? Around line 413 (2.3.40-4+minor_patch):
> 
> -       if (m5229_revision <= 0x20) {
> +       if (m5229_revision < 0x20) {

I tried this, briefly, under a light I/O load, after a full-backup of my
root disc. It worked with the Maxtor 90680D4 HD going into UDMA at boot 
time. I did get lots of CRC errors, eg.

Jan 19 19:51:15 mklab kernel: hda: dma_intr: status=0x51 { DriveReady SeekComplete Error } 
Jan 19 19:51:15 mklab kernel: hda: dma_intr: error=0x84 { DriveStatusError BadCRC } 

I turned on some of the other optimisations with,
hdparm -c 1 -d 1 -m 16 -u 1 -a 8 

and get the following messages (presumably due to the m16??)

Jan 19 19:57:42 mklab kernel: hda: multwrite: count=8, current=0 
Jan 19 19:58:22 mklab last message repeated 53 times
Jan 19 19:59:23 mklab last message repeated 115 times

The root FS checked-out OK after this expt.

Here are some disc timing results.
With UDMA turned on the Maxtor gives,
 Timing buffer-cache reads:   128 MB in  2.97 seconds =43.10 MB/sec
 Timing buffered disk reads:  64 MB in  5.95 seconds =10.76 MB/sec

Turning it off gives,
 Timing buffer-cache reads:   128 MB in  3.18 seconds =40.25 MB/sec
 Timing buffered disk reads:  64 MB in 12.17 seconds = 5.26 MB/sec

Turning on the other enhancements (hdparm, as above)
 Timing buffer-cache reads:   128 MB in  3.04 seconds =42.11 MB/sec
 Timing buffered disk reads:  64 MB in  6.17 seconds =10.37 MB/sec

Turning off the hack and returning to MWDMA gives,
 Timing buffer-cache reads:   128 MB in  3.07 seconds =41.69 MB/sec
 Timing buffered disk reads:  64 MB in 11.76 seconds = 5.44 MB/sec

Staying with MWDMA and turning on the other optimisations gives,
 Timing buffer-cache reads:   128 MB in  3.12 seconds =41.03 MB/sec
 Timing buffered disk reads:  64 MB in 10.41 seconds = 6.15 MB/sec

If it is safe to run the Maxtor in UDMA, is there a recommended patch for the 
CRC errors? - something in ide.c? Any comments on the 'multwrite' diagnostics - 
Are these benign?

Best Regards
Tom.

-- 
Tom Crane, Dept. Physics, Royal Holloway, University of London, Egham Hill,
Egham, Surrey, TW20 0EX, England. 
Email:	uhap023 en vms.rhbnc.ac.uk
SPAN:   19.875
Fax:    01784 472794

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo en vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/