Bug 3420

Summary: CPU load is 100% on any read/write operations
Product: IO/Storage Reporter: Artem S. Tashkinov (aros)
Component: IDEAssignee: Bartlomiej Zolnierkiewicz (bzolnier)
Status: REJECTED DUPLICATE    
Severity: high    
Priority: P2    
Hardware: i386   
OS: Linux   
Kernel Version: 2.6.x Subsystem:
Regression: --- Bisected commit-id:
Attachments: different /proc information

Description Artem S. Tashkinov 2004-09-18 05:16:42 UTC 
Distribution:
Fedora Core 1

Hardware Environment:
Kernel: 2.6.x, vanilla, untainted, compiled with GCC 3.4.2 and GCC 2.95 vanilla
CPU: Intel Pentium 3 EB (FSB 133MHz)
MB: Asus P3V Via Apollo Pro 133 FSB 133MHz
RAM: Hyudai 256Mb PCI 133 running @ 133MHz
IDE0 channel: UDMA Mode 66Mbytes/sec (hda - Seagate 120Gb, hdb - Seagate 20Gb)
IDE1 channel: UDMA Mode 33Mbytes/sec (hdc - Sony CRX320E, hdd - Mitsumi)

Problem Description:
Steps to reproduce:
When reading files at maximum speed (no background processes/tasks running even
at runlevel 1 with only init and bash in memory) either from filesystem or
directly from /dev/hd? devices CPU load is 100% with IO wait around 60%.
The same happens when copying files within hda or hdb or from hda to hdb and
vica versa.

Tested with:
	dd if=/dev/hda of=/dev/null bs=10M
	dd if=/mnt/hda1/somefile of=/dev/null bs=10M
	cp /mnt/hda1/somefile /mnt/hda1/othername
	cp /mnt/hda1/somefile /mnt/hdb/

Windows XP Professional SP2 consumes around 92% CPU when copying files between
two HDDs on the same IDE cable and around 50% when reading/writing files.

Expected results:
CPU load as in Windows XP or less (otherwise what for is DMA intended for).

This situation happpens even if I leave the only hda device (with all others
unplugged). /dev/proc entries and lspci -vvv output are attached
Comment 1 Artem S. Tashkinov 2004-09-18 05:19:02 UTC 
Created attachment 3678 [details]
different /proc information
Comment 2 Bartlomiej Zolnierkiewicz 2004-09-25 07:36:06 UTC 
I can't reproduce these results.

Are you using 'top'? Do you have up-to-date procps package?
AFAIR Fedora Core 1 is not 2.6 ready ie. it has procps-2.0.17
while procps-3.2.0 is required.
Comment 3 Artem S. Tashkinov 2004-09-28 08:03:37 UTC 
I used both procps versions and they are all giving the same information. I also
rebooted in kernel 2.4.27-pre3 and CPU load in it is only 60% on most heavy I/O
operations
Comment 4 Bartlomiej Zolnierkiewicz 2004-10-05 12:56:16 UTC 

*** This bug has been marked as a duplicate of 3049 ***
Comment 5 Bartlomiej Zolnierkiewicz 2004-10-05 12:57:37 UTC 
damn, should be #3094 :)
Comment 6 Bartlomiej Zolnierkiewicz 2004-10-05 12:58:01 UTC 

*** This bug has been marked as a duplicate of 3094 ***
Comment 7 Artem S. Tashkinov 2004-10-23 08:29:27 UTC 
My bug is _not_ related to bug number 3420. I don't use i2c drivers at all. It
is POSSIBLY related to bug number 3094 but I don't see to be any similar.
Please, reopen
Comment 8 Bartlomiej Zolnierkiewicz 2004-10-23 09:23:03 UTC 
Isn't too high I/O load something similar?
Comment 9 Bartlomiej Zolnierkiewicz 2004-12-08 10:18:39 UTC 

*** This bug has been marked as a duplicate of 3094 ***