2.6.13.2-nitro1 "Down with latency!'

Message

darklegion · Post by **darklegion** » Sun Oct 02, 2005 5:52 am

Seppe: Any chance you could patch in plugsched in a later release?
I've never had good performance with staircase but all of the current patchsets seem to be using it, unfortunately.Even vanilla with nicksched works better than any kernel with staircase (for me) but I miss the voluntary pre-emption patches which make performance that bit better.I could live with just using 2.6.12-rc1-love1 perpetually but even that has its issues.
Alternatively if anyone knows how to get the ck + plugsched working together, let me know

ankhcraft · Post by **ankhcraft** » Sun Oct 02, 2005 5:22 pm

Yeah.... the AMD64 patch is a bit lame. You really need to at least return a value from that function. What value are you hoping the thing returns most of the time? Why not just return a constant?

Here's the text of the patch so others can see what I'm talking about:

Code: Select all

diff -Nuar linux-2.6.13.2-nitro1/include/linux/dyn_pageflags.h linux-2.6.13.2-nitro1-modded/include/linux/dyn_pageflags.h
--- linux-2.6.13.2-nitro1/include/linux/dyn_pageflags.h 2005-09-23 05:01:37.000000000 +1000
+++ linux-2.6.13.2-nitro1-modded/include/linux/dyn_pageflags.h  2005-09-23 04:39:20.000000000 +1000
@@ -31,12 +31,14 @@

 static inline int __get_next_bit_on(dyn_pageflags_t bitmap, int counter)
 {
-       do {
-               counter++;
-       } while((counter < max_mapnr) &&
-               (!test_bit(PAGEBIT(pfn_to_page(counter)),
-                       PAGE_UL_PTR(bitmap, pfn_to_page(counter)))));
-       return counter;
+//     FIXME: Fix this dirty hack for AMD64.
+//
+//     do {
+//             counter++;
+//     } while((counter < max_mapnr) &&
+//             (!test_bit(PAGEBIT(pfn_to_page(counter)),
+//                     PAGE_UL_PTR(bitmap, pfn_to_page(counter)))));
+//     return counter;
 }

 /* With the above macros defined, you can do...

Alright, look, I'm just going to add the following line after the commented out return:

Code: Select all

return ++counter;

This is the absolute *smallest* value that the original function would've been able to return. I don't want to compile my kernel w/ a function returning undefined values, and neither should anyone else here.

Not that I am ungrateful for the patch or a solution to problem. Not in the slightest. Just trying to make improvements.

DROPTOTHETOP · Post by **DROPTOTHETOP** » Mon Oct 03, 2005 1:15 am

Hey,

I am getting very slow hard drive read speeds with hdparm:

Code: Select all

# hdparm -tT /dev/hda

/dev/hda:
 Timing cached reads:   1456 MB in  2.00 seconds = 727.38 MB/sec
 Timing buffered disk reads:   10 MB in  3.05 seconds =   3.28 MB/sec

I was using an mm2 patched vanilla source 2.6.13 and I was getting much higher speeds. I have the Nforce 2 Ultra chip. Before compiling the nitro sources I copied over my config from the vanilla sources so the device support should be correct. Does anyone have any suggestions?

Gnufsh · Post by **Gnufsh** » Mon Oct 03, 2005 3:27 am

Check to see if DMA is enabled (just use hdparm). If not, double check that the correct driver is, infact, selected.

ankhcraft · Post by **ankhcraft** » Mon Oct 03, 2005 8:57 pm

So, does nobody have any comments on the amd64 patch implementing an invalid C function? A function returning a value must always return a value, or behaviour is undefined.

Although in reality, on amd64 the behaviour is defined thus:

Whatever value was last in the RAX register is the "returned" value. This is completely random and is definitely *not* what one wants.

Gnufsh · Post by **Gnufsh** » Mon Oct 03, 2005 9:28 pm

I think that it is a valid concern, but since I'm not expierencing the problem on amd64, I didn't apply the patch, so it didn't effect me.

lazx888 · Post by **lazx888** » Tue Oct 04, 2005 8:10 am

Just updated to nitro sources...

I have 2 sata drives (sda = root + boot, sdb = storage + windows)

Using the sii3114 drivers

sda is coming up fine, but sdb doesn't show up... :-/ the drive is fine b/c I can boot windows off it and RIP detects it...

In dmesg, there is no output for sdb being detected/failing/etc

Udev rules seem to be fine...

Here is some grub info: kernel (hd0,0)/boot/bzImage root=/dev/sda2 gentoo=nodevfs udev doscsi usb-handoff

Tried with just kernel (hd0,0)/boot/bzImage root=/dev/sda2 and this did not make a difference

I have searched the forums already, does anyone have any ideas/links?

Thanks

m0p · Post by **m0p** » Tue Oct 04, 2005 3:00 pm

I'm still using 2.6.12-nitro5, is there any significant reason for me to upgrade yet? I know 2.6.13 adds alot of stuff but I'm sure that half of the stuff is unneccesary for me. You think it's really worth the upgrade?

satanskin · Post by **satanskin** » Tue Oct 04, 2005 6:34 pm

i had to revert back to 2.6.12-nitro5 as well since .13 was giving me a REALLY sluggish system. I'm still hoping for a 2.6.14 series release!

Tiger683 · Post by **Tiger683** » Tue Oct 04, 2005 8:08 pm

Some day.....

b3cks · Post by **b3cks** » Wed Oct 05, 2005 5:32 am

Tiger683 wrote:Some day.....

WOW, he's back! Welcome!

satanskin · Post by **satanskin** » Wed Oct 05, 2005 4:01 pm

Yes, Welcome back! Great to see you again! and maybe that someday will be sooner than later....?

brot · Post by **brot** » Wed Oct 05, 2005 6:19 pm

Tiger683 wrote:Some day.....

Nice to hear from you... Seppe did his job well,... I think *thank you seppe*

-

The problems with the swap prefetching should be fixed in ck7, which applies to 2.6.13, and patches to 2.6.13.3 .
Maybe this is worth releasing a new nitro

ThorstenHirsch · Post by **ThorstenHirsch** » Wed Oct 05, 2005 9:27 pm

hdparm has also some problems here with 2.6.13.2-nitro1, the speed is about 40% less than normal. The settings are the same of course.

# hdparm -tT /dev/hde

/dev/hde:
Timing cached reads: 1264 MB in 2.00 seconds = 631.15 MB/sec
Timing buffered disk reads: 102 MB in 3.08 seconds = 33.12 MB/sec
# hdparm /dev/hde

/dev/hde:
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 0 (off)
using_dma = 1 (on)
keepsettings = 1 (on)
readonly = 0 (off)
readahead = 256 (on)
geometry = 19457/255/63, sectors = 312581808, start = 0

The drive is connected to my second IDE controller chip, a HPT370/372.

iphitus · Post by **iphitus** » Thu Oct 06, 2005 12:02 am

Update to latest CK, -ck5 is old and has a few issues, including prefetch and another patch backported from mm.

Tiger683 · Post by **Tiger683** » Thu Oct 06, 2005 1:03 am

satanskin wrote:(...)and maybe that someday will be sooner than later....?

Maybe

artworcs · Post by **artworcs** » Thu Oct 06, 2005 2:55 pm

I have just upgraded my kernel and i have a little problem:
The root block is unspecified or not detected.
My grub.conf looks like this:

Code: Select all

# Boot automatically after 30 secs.
timeout 10

# By default, boot the first entry.
default 0

# Fallback to the second entry.
fallback 1

title Linux Gentoo 2.6.13
root (hd0,2)
kernel (hd0,2)/gentoo/kernel-2.6.13-gentoo root=/dev/hda5

title Linux Gentoo 2.6.11 (Kernel 2.6.11-nitro2)
root (hd0,2)
kernel (hd0,2)/gentoo/kernel-2.6.11-nitro2 root=/dev/hda5

title Linux Gentoo 2.6.13 (Kernel 2.6.13.2-nitro1)
root (hd0,2)
kernel (hd0,2)/kernel-genkernel-x86-2.6.13.2-nitro1 root=/dev/hda5
initrd (hd0,2)/initramfs-genkernel-x86-2.6.13.2-nitro1

If i type /dev/hda5(this is where gentoo is installed) it boots normally.

SchAmane · Post by **SchAmane** » Thu Oct 06, 2005 7:28 pm

Tiger683 wrote:Maybe

Hi Tiger, nice to see you again, how was on vacation ?

What we can wait for next ?

Tiger683 · Post by **Tiger683** » Thu Oct 06, 2005 8:01 pm

SchAmane wrote:Hi Tiger, nice to see you again, how was on vacation ?
What we can wait for next ?

1) awesome

2) a nitro eventually

to be more precise, 2.6.14-rc3-nitro1, an unstable release. I will use it to hook up with
the changes which happened since i was gone.
The list of mm-patches is already here:
mm-patches

cheers

T

satanskin · Post by **satanskin** » Thu Oct 06, 2005 8:35 pm

Tiger683 wrote:
SchAmane wrote:Hi Tiger, nice to see you again, how was on vacation ?
What we can wait for next ?
1) awesome

2) a nitro eventually

to be more precise, 2.6.14-rc3-nitro1, an unstable release. I will use it to hook up with
the changes which happened since i was gone.
The list of mm-patches is already here:
mm-patches

cheers

T

Nice! How long do you think it'll be before you have a release of it? And will it have reiser4 support? I'd be willing to help you test it out if needed. Let me know!

Tiger683 · Post by **Tiger683** » Fri Oct 07, 2005 11:28 am

satanskin wrote: How long do you think it'll be before you have a release of it?

If nothing breaks, approx. 5-6 hours after this post

satanskin wrote: And will it have reiser4 support?

Ofcourse.

T

SchAmane · Post by **SchAmane** » Fri Oct 07, 2005 2:00 pm

Tiger683 wrote:
satanskin wrote: And will it have reiser4 support?
Ofcourse.
T

As its already part of -mm.
But i readed somewhere that there are some problems in reiser4 in patches for 2.6.14 kernels ?

Tiger683 · Post by **Tiger683** » Fri Oct 07, 2005 2:44 pm

We will see....

joefish · Post by **joefish** » Wed Oct 12, 2005 10:15 am

This may sound like a silly question, but since I'm faily new to this nitro thing, please bear with me.

I'm having trouble with NFS, as described in this thread. Can someone please confirm whether or not this kernel version has the problem described there/here?

Thanks.

joefish · Post by **joefish** » Wed Oct 12, 2005 12:05 pm

Never mind, it seems my problems were caused by a bad line in my /etc/hosts file...