2001-10-30 04:58:10

by Skip Gaede

[permalink] [raw]
Subject: [PATCH] driver: ide-floppy.c kernel >=2.4.7

This patch fixes a lost interrupt problem with the Iomega ATAPI Zip 100 drive
on an Asus A7V133 board (uses South Bridge VIA VT82C686 chip). The problem
occurs when trying to format the drive using mke2fs /dev/hdx1. The patch
introduces an adjustable delay between the time the drive asserts DRQ and
deasserts BSY after issuing the packet command and before transferring the 12
byte packet.. With delays of 3-5 ticks, the filesystem creation occurs
without retries/resets. The delay can be adjusted through the proc interface
by adjusting the value assigned to the parameter ticks. (Without the patch, I
experienced 111 lost interrupts, resulting in an elapsed time of over 2 hours
to format the drive. With the patch, mke2fs completed in <15 seconds.)
I am looking for testing, by others who have the internal Zip drive and may
have experienced the same problem, as well as comments.

Thanks,
--Skip


Attachments:
ide-floppy-p7.patch.bz2 (2.86 kB)
README (74.00 B)
Download all attachments

2001-11-04 08:00:44

by Ken Moffat

[permalink] [raw]
Subject: Re: [PATCH] driver: ide-floppy.c kernel >=2.4.7

Skip,

I've just got around to testing this. I didn't have problems with the
current version of ide-floppy, but I thought I'd give it a go anyway. I'm
sorry to say that the performance as standard is atrocious. I was
previously using 2.4.12-ac3 with the preempt patch, for these tests I
tried 2.4.13-ac6. (All tests at the console, no other conscious use of the
box.)

hdparm -Tt /dev/hdd1 cache disk
2.4.12-ac3-preempt 52.67Mb/s 1.06Mb/s
2.4.13-ac6+your patch 52.89Mb/S 480.43kB/s
(above are worst of three, but there isn't wide variation)

time mke2fs /dev/hdd1 real user sys
2.4.12-ac3-preempt 0m4.252s 0m0.010s 0m0.090s
2.4.13-ac6+your patch 0m8.349s 0m0.000s 0m0.070s

Bonnie -s 85 -d /zip
---Sequential Output------- --Sequential Input--- Rnd Seek
Char Block Rewrite Char Block /sec
2.4.12-ac3 1246K/s 1504K/s 433K/s 702K/s 958K/s 30.0
2.4.13-ac6+ 575K/s 699K/s 211K/s 347K/s 447K/s 20.9

This is on a K6/500, and with your patch the disk feels _so_ slow. I
accept it can be tweaked through the /proc interface, but if this is the
default behaviour, it's dreadful. I do wonder if you've got a flakey
drive ? Sorry.

Ken

On Tue, 30 Oct 2001, Skip Gaede wrote:

> This patch fixes a lost interrupt problem with the Iomega ATAPI Zip 100 drive
> on an Asus A7V133 board (uses South Bridge VIA VT82C686 chip). The problem
> occurs when trying to format the drive using mke2fs /dev/hdx1. The patch
> introduces an adjustable delay between the time the drive asserts DRQ and
> deasserts BSY after issuing the packet command and before transferring the 12
> byte packet.. With delays of 3-5 ticks, the filesystem creation occurs
> without retries/resets. The delay can be adjusted through the proc interface
> by adjusting the value assigned to the parameter ticks. (Without the patch, I
> experienced 111 lost interrupts, resulting in an elapsed time of over 2 hours
> to format the drive. With the patch, mke2fs completed in <15 seconds.)
> I am looking for testing, by others who have the internal Zip drive and may
> have experienced the same problem, as well as comments.
>
> Thanks,
> --Skip

--
If a six turned out to be nine, I don't mind.

Home page : http://www.kenmoffat.uklinux.net