Skip to contentSkip to conversation

Disk-I/O and [email protected]% system (2.4.18-pre9)

DISCUSSION Florian Hars 5 participants 10 messages Feb 8, 2002 – Feb 13, 2002

lore
Share & tools
lore
#0 lore raw

I have an ext[23] filesystem (doesn't matter which), on an LVM Volume. Whenever I do some heavy disk-I/O (like untaring an archive with 13000 files that amount to 5GB), the CPU-state repeatedly goes to 99.9% system and stays there for a noticeable amount of time (1-2 seconds), during which the system doesn't respond very well to user action, to put it mildly.
Whenever this happens, top shows kupdated as one of the most active processes (sometimes it also claims that top uses 54% of the CPU, but I guess that is only marginally accurate). bdflush (which according to google was mentioned in connection to a similar problem some time ago) doesn't do anything.

The system is an Athlon XP 1800+ on a Gigabyte GA7-VTXE board with one IDE disk.
Kernel is 2.4.18-pre9, LVM utilities 1.0.1release, is there anything else that is relevant?

Please CC answers to me.

Yours, Florian.

#1 lore raw
Quoted 2 lines · Heinz Diehl

Downgrade to 2.4.18-pre8 and use Michael Cohen's patch from "ftp://ftp.kernel.org/pub/linux/kernel/people/mjc/linux-2.4/",

That doesn't help. I tried 2.4.16 with Debian modifications, said kernel patched to 2.4.18-pre9, stock 2.4.17, stock 2.4.18-pre9 and 2.4.18-pre8-mjc with preempt and lockbreak, and all behave alike.

On a plain ext2-filesystem on a primary partition I get (with -mjc):

During the sync the system is extremly sluggish, and once during my tests it froze completely (it did still return pings with a normal speed) so that I had to press reset.

The same operations on a slower computer running 2.2.20 are consideraby faster:

$ time tar -xzf linux-2.4.17.tar.gz; time sync
real    0m7.716s
user    0m5.430s
sys     0m2.110s
real    0m6.332s
user    0m0.000s
sys     0m0.120s
Quoted 1 line

Latency in stock 2.4.17/18-pre kernels is well known :\

this looks like more than a latency issue :-(.

Yours, Florian Hars.

#2 lore raw
In reply to Florian Hars

Soyy, I just hit the wrong key before pasting in the numbers, so here they are:

Quoted 1 line

On a plain ext2-filesystem on a primary partition I get (with -mjc):

# time tar -xzf linux-2.4.17.tar.gz; time sync
real    0m18.818s
user    0m1.890s
sys     0m14.740s
real    0m21.990s
user    0m0.000s
sys     0m11.320s

Yours, Florian Hars.

#3 lore raw
In reply to Florian Hars
Quoted 7 lines · Florian Hars

I have an ext[23] filesystem (doesn't matter which), on an LVM Volume. Whenever I do some heavy disk-I/O (like untaring an archive with 13000 files that amount to 5GB), the CPU-state repeatedly goes to 99.9% system and stays there for a noticeable amount of time (1-2 seconds), during which the system doesn't respond very well to user action, to put it mildly.

A kernel profile will presumably point us at the problem.

It's pretty simple. See
http://www.uwsg.iu.edu/hypermail/linux/kernel/0201.3/0773.html

-

#4 lore raw

2.5.4 error building vmlinuz

Hi Folks: I downloaded the 2.5.4 tar ball and applied a one line patch by Dave ??? I don't remember his last name. Most of the kernel compiled just fine until I got to the loader stage. I have tried the compile a number of times removing newer features I had included from the 2.5.4 configuration. It continually dies just after the ld stage. Here's the log:

ld -m elf_i386 -T /usr/src/linux-2.5.4/arch/i386/vmlinux.lds -e stext arch/i386/kernel/head.o arch/i386/kernel/init_task.o init/main.o init/version.o init/do_mounts.o \
	--start-group \
	arch/i386/kernel/kernel.o arch/i386/mm/mm.o kernel/kernel.o mm/mm.o fs/fs.o ipc/ipc.o \
	/usr/src/linux-2.5.4/arch/i386/lib/lib.a /usr/src/linux-2.5.4/lib/lib.a /usr/src/linux-2.5.4/arch/i386/lib/lib.a \
	 drivers/base/base.o drivers/char/char.o drivers/block/block.o drivers/misc/misc.o drivers/net/net.o drivers/media/media.o drivers/ide/idedriver.o drivers/cdrom/driver.o drivers/sound/sounddrivers.o drivers/pci/driver.o drivers/pnp/pnp.o drivers/video/video.o \
	net/network.o \
	--end-group \
	-o vmlinux

drivers/char/char.o(.data+0x46b4): undefined reference to local symbols in discarded section .text.exit' drivers/net/net.o(.data+0x174): undefined reference to local symbols in discarded section .text.exit'
make: *** [vmlinux] Error 1

Once again could someone help the clewless?
Kirk

Signature · 3 lines
Kirk Reiser				The Computer Braille Facility
e-mail: [email protected]		University of Western Ontario
phone: (519) 661-3061
#5 lore raw

Unknown Southbridge (was: Disk-I/O and [email protected]% system (2.4.18-pre9))

In reply to Florian Hars
Quoted 3 lines · I wrote:

Whenever I do some heavy disk-I/O (like untaring an archive with 13000 files that amount to 5GB), the CPU-state repeatedly goes to 99.9% system

Part of the problem could be alleviated by unmasking the interrupts, but now some part of the IDE system is crying for its master:

Uniform Multi-Platform E-IDE driver Revision: 6.31
ide: Assuming 33MHz system bus speed for PIO modes; override with idebus=xx
VP_IDE: IDE controller on PCI bus 00 dev 89
VP_IDE: chipset revision 6
VP_IDE: not 100% native mode: will probe irqs later
VP_IDE: Unknown VIA SouthBridge, contact Vojtech Pavlik <[email protected]>

I use a Gigabyte GA-7VTXE with a VIA KT266A chipset and a Southbridge called VT8233A, which does not look like one of the "FUTURE_BRIDGES" that are ifdefed out in the driver (or is it the same as the
{ "vt8233c", PCI_DEVICE_ID_VIA_8233C, 0x00, 0x2f, VIA_UDMA_100 } ?).

Yours, Florian Hars.

#6 lore raw

Re: Unknown Southbridge (was: Disk-I/O and [email protected]% system (2.4.18-pre9))

In reply to Florian Hars
Quoted 15 lines · Florian Hars
Quoted 3 lines · I wrote:

Whenever I do some heavy disk-I/O (like untaring an archive with 13000 files that amount to 5GB), the CPU-state repeatedly goes to 99.9% system

Part of the problem could be alleviated by unmasking the interrupts, but now some part of the IDE system is crying for its master:

Uniform Multi-Platform E-IDE driver Revision: 6.31
ide: Assuming 33MHz system bus speed for PIO modes; override with idebus=xx
VP_IDE: IDE controller on PCI bus 00 dev 89
VP_IDE: chipset revision 6
VP_IDE: not 100% native mode: will probe irqs later
VP_IDE: Unknown VIA SouthBridge, contact Vojtech Pavlik <[email protected]>

Quoted 6 lines

I use a Gigabyte GA-7VTXE with a VIA KT266A chipset and a Southbridge called VT8233A, which does not look like one of the "FUTURE_BRIDGES" that are ifdefed out in the driver (or is it the same as the
{ "vt8233c", PCI_DEVICE_ID_VIA_8233C, 0x00, 0x2f, VIA_UDMA_100 } ?).

Yours, Florian Hars.

2.5.2 (and later, and maybe some earlier versions as well) have support for this chipset. You can copy over the via82cxxx.c and ide-timing.h to your kernel and it should work.

Signature · 2 lines
Vojtech Pavlik
SuSE Labs
#7 lore raw

Re: Unknown Southbridge (was: Disk-I/O and [email protected]% system (2.4.18-pre9))

In reply to Florian Hars

begin Vojtech Pavlik quote:

Quoted 7 lines · Florian Hars
Quoted 2 lines

I use a Gigabyte GA-7VTXE with a VIA KT266A chipset and a Southbridge called VT8233A, which does not look like one of the "FUTURE_BRIDGES"

2.5.2 (and later, and maybe some earlier versions as well) have support for this chipset. You can copy over the via82cxxx.c and ide-timing.h to your kernel and it should work.

Not really:

via82cxxx.c: In function `ide_init_via82cxxx':
via82cxxx.c:548: structure has no member named `highmem' make[4]: *** [via82cxxx.o] Fehler 1

I had to remove the offending line (it isn't there in the 2.4 version of the file), and of course add the PCI ID of the SouthBridge in the appropriate places. Now the system boots and hdparm says that dma is activated. So lets see how well it behaves...

Yours, Florian Hars.

#8 lore raw

Re: Unknown Southbridge (was: Disk-I/O and [email protected]% system (2.4.18-pre9))

In reply to Florian Hars
Quoted 9 lines · Florian Hars

begin Vojtech Pavlik quote:

Quoted 7 lines · Florian Hars
Quoted 2 lines

I use a Gigabyte GA-7VTXE with a VIA KT266A chipset and a Southbridge called VT8233A, which does not look like one of the "FUTURE_BRIDGES"

2.5.2 (and later, and maybe some earlier versions as well) have support for this chipset. You can copy over the via82cxxx.c and ide-timing.h to your kernel and it should work.

Quoted 5 lines

Not really:

via82cxxx.c: In function `ide_init_via82cxxx':
via82cxxx.c:548: structure has no member named `highmem' make[4]: *** [via82cxxx.o] Fehler 1

Yes, it's for 2.5.

Quoted 4 lines

I had to remove the offending line (it isn't there in the 2.4 version of the file), and of course add the PCI ID of the SouthBridge in the appropriate places. Now the system boots and hdparm says that dma is activated. So lets see how well it behaves...

I've sent a patch to Jens Axboe for inclusion into 2.4, so it might be in 2.4.18. If you find any flaws, please tell me soon enough so I can stop the inclusion in time ...

Signature · 2 lines
Vojtech Pavlik
SuSE Labs
#9 lore raw

Re: Unknown Southbridge (was: Disk-I/O and [email protected]% system (2.4.18-pre9))

In reply to Florian Hars
Quoted 6 lines · Vojtech Pavlik
Quoted 1 line · Florian Hars

[Via VT8233A Southbridge]

I've sent a patch to Jens Axboe for inclusion into 2.4, so it might be in 2.4.18. If you find any flaws, please tell me soon enough so I can stop the inclusion in time ...

I made a naive stress test (untaring a 5GB archive into an ext3 filesystem on a LVM volume, while running fsx on the same filesystem (don't know if it is relevant for this problem) and throwing in an occasional hdparm -Tt) and everything looked good, even mozilla was still responsive.

Thanks for your help.

Yours, Florian.