| View previous topic :: View next topic |
| Author |
Message |
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
 Posted: Mon Sep 20, 2004 9:27 pm Post subject: 2.6.9-rc2-oc2 "PEBKAC" |
 |
|
Well not much to say besides theres now staircase in this and voluntary preempt which should make some of you ck fanatics happy. I'll soon look into cpuse's pick-your-scheduler at runtime patches.
| Code: |
Kernel was patched in exact order of these patches.
Base Version
====================================================================================================
2.6.9-rc2 --------------------------------------------------> 2.6.9-rc2 kernel
MM Version
====================================================================================================
2.6.9-rc2-mm1 ----------------------------------------------> MM latest version
CK Version
====================================================================================================
2.6.9-rc2-mm1-to-staircase8.4-1 ----------------------------> Con Kolivas's staircase scheduler
schedrange.diff --------------------------------------------> Infrastructure for more policies
schedbatch2.4.diff -----------------------------------------> Batch scheduling
schediso2.6.diff -------------------------------------------> Isochronous scheduling
mapped_watermark4.diff -------------------------------------> Very unlikely to swap due to file caching
oc-sources
====================================================================================================
voluntary-preempt-2.6.9-rc2-mm1-S1 -------------------------> Ingo Molnar's great preempt work
rmap-unlock_page-fix.patch ---------------------------------> Race condition fix in rmap.c
1G-lowmem-2.diff -------------------------------------------> Configurable 1GB lowmem support
checkpoint-lat-fix.patch -----------------------------------> fs/checkpoint latency fix
config-nr-tty-devices.patch --------------------------------> Configure the ammount of tty devices in /dev
debug_slab-dumbass.patch -----------------------------------> Enabling debug_slab slows things down ( well no shit )
oc-sources.diff --------------------------------------------> Branding
stack-fix-i386.patch ---------------------------------------> Elf stack change
writeback-lat-fix.patch ------------------------------------> A couple cond_resched(); to improve latency
gcloop-2.6-20040527.patch ----------------------------------> A robust framework for building compressed partition image
layer-7-2.6.8-rc2-oc1.patch --------------------------------> Layer 7 packet filtering - very useful ( might not work )
lufs-0.9.7-2.6.0-test9.patch -------------------------------> "Linux userland file system" can mount almost anything
menuconfig-NAME-v2.1-dev5.patch ----------------------------> Display kernel name in menu
configureable-hid-mouse-polling-2.6.9-rc1.patch ------------> USB configureable mouse polling rate
squashfs-2.0.patch -----------------------------------------> Squashfs a well squashed file system
config_hz.diff ---------------------------------------------> Configure kernel internal frequency. Higher makes more overhead but lowers latencies. 1000 = 1ms
vesafb-tng-0.9-rc4-r3-2.6.9-rc2.patch ----------------------> High refresh rate framebuffer support
|
Ebuild
oc-sources_rc2-r2.ebuild
If you get this when you exit make menuconfig:
| Quote: |
arch/i386/defconfig:128: trying to assign nonexistent symbol PM_DISK
arch/i386/defconfig:175: trying to assign nonexistent symbol PCI_USE_VECTOR
arch/i386/defconfig:251: trying to assign nonexistent symbol BLK_DEV_CARMEL
arch/i386/defconfig:291: trying to assign nonexistent symbol BLK_DEV_ADMA
arch/i386/defconfig:364: trying to assign nonexistent symbol SCSI_MEGARAID
arch/i386/defconfig:569: trying to assign nonexistent symbol NET_FASTROUTE
arch/i386/defconfig:776: trying to assign nonexistent symbol QIC02_TAPE
arch/i386/defconfig:1247: trying to assign nonexistent symbol X86_STD_RESOURCES
|
Don't worry about it it's an old defconfig file that seems to try to enable options that aren't there anymore
EDIT - *don't* use "preempt softirqs" "preempt hardirqs" ( until ingo fixes it ) and "vesa-tng"
_________________
Great Resources
Last edited by nmcsween on Tue Sep 21, 2004 6:52 am; edited 1 time in total |
|
| Back to top |
|
 |
blue.sca
l33t
Joined: 28 Aug 2003
Posts: 680
Location: Mainz, Germany
|
| Posted: Mon Sep 20, 2004 11:10 pm Post subject: |
|
|
testing it out ;)
i havent found some options like hid_polling, config_hz for example. could be me...
//e compiled, but getting a hard-lock short after boot right after "agp aperture size...". could be a problem on my own, testing this on an iBook G4...
_________________
geek by nature, linux by choice
i want my avatar back... thank you
:wq |
|
| Back to top |
|
|
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
| Posted: Mon Sep 20, 2004 11:47 pm Post subject: |
|
|
I just went through the lkml and found that ingo's voluntary preempt for Andrew's MM patches seem to have some weird effects on seemingly random things try disabling Preempting softirq's and hardirqs.
| Quote: |
i havent found some options like hid_polling, config_hz for example. could be me...
|
1. Under Device drivers -> USB support -> Configurable USB HID mouse polling
2. Under Processor types and features -> kernel internal timer frequency
_________________
Great Resources |
|
| Back to top |
|
|
Cagnulein
l33t
Joined: 18 Sep 2003
Posts: 861
Location: Modena, Italy
|
| Posted: Tue Sep 21, 2004 4:55 am Post subject: |
|
|
testing 
| Code: | CC drivers/video/softcursor.o
CC drivers/video/vesafb-tng.o
drivers/video/vesafb-tng.c: In function `vesafb_init':
drivers/video/vesafb-tng.c:1066: error: too few arguments to function `fb_get_options'
drivers/video/vesafb-tng.c:1066: warning: passing arg 1 of `vesafb_setup' makes pointer from integer without a cast
make[2]: *** [drivers/video/vesafb-tng.o] Error 1
make[1]: *** [drivers/video] Error 2
make: *** [drivers] Error 2
|
|
|
| Back to top |
|
|
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
| Posted: Tue Sep 21, 2004 6:19 am Post subject: |
|
|
Im banging my head on the wall on this one.. theres the offending code in vesa-tng.c but everything seems fine....
| Code: |
int __init vesafb_init(void)
{
int ret;
vesafb_setup(fb_get_options("vesafb"));
ret = driver_register(&vesafb_driver);
if (!ret) {
ret = platform_device_register(&vesafb_device);
if (ret)
driver_unregister(&vesafb_driver);
}
return ret;
}
|
_________________
Great Resources |
|
| Back to top |
|
|
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
| Posted: Tue Sep 21, 2004 6:38 am Post subject: |
|
|
Well I figured out what it was, seems that the fbdev patches included with mm seem to do some nasty things to everything it touches. I was planing on removing this earlier but had other things to tend to. I'll work on this in oc3 tomorrow once I get fbdev out fbsplash and vesa-tng will be in. So for now *don't* use "preempt softirqs" "preempt hardirqs" ( until ingo fixes it ) and "vesa-tng"
_________________
Great Resources |
|
| Back to top |
|
|
Cagnulein
l33t
Joined: 18 Sep 2003
Posts: 861
Location: Modena, Italy
|
| Posted: Tue Sep 21, 2004 10:47 am Post subject: |
|
|
one question:
"preempt softirqs" "preempt hardirqs"
what are these? |
|
| Back to top |
|
|
AstralStorm
Tux's lil' helper
Joined: 06 Nov 2003
Posts: 80
Location: /dev/null
|
| Posted: Tue Sep 21, 2004 1:16 pm Post subject: |
|
|
"You lose" with CONFIG_HZ set to 2000...
Seems there's no shift value, I'm experimentally trying 11. (1 more than HZ = 1000)
You should either hide the option or fix the include/linux/timex.h. |
|
| Back to top |
|
|
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
| Posted: Tue Sep 21, 2004 6:14 pm Post subject: |
|
|
| Cagnulein wrote: | one question:
"preempt softirqs" "preempt hardirqs"
what are these? |
They thread softirqs and hardirqs threading softrqs might not be so dangerous but threading hardirqs is something I wouldn't do.
| Quote: |
"You lose" with CONFIG_HZ set to 2000...
Seems there's no shift value, I'm experimentally trying 11. (1 more than HZ = 1000)
You should either hide the option or fix the include/linux/timex.h.
|
I'll look into why you can't set the timer to 2000.
_________________
Great Resources |
|
| Back to top |
|
|
chickaroo
Tux's lil' helper
Joined: 21 Sep 2004
Posts: 102
Location: #!/usr/bin/girl
|
| Posted: Wed Sep 22, 2004 3:17 am Post subject: |
|
|
oooh this looks nice  i might try it when you release oc3
using rc2-mm1 with con's 8.4 straircase patches right now
_________________
Registered Linux user #364515 (Jun, 2004) |
|
| Back to top |
|
|
AstralStorm
Tux's lil' helper
Joined: 06 Nov 2003
Posts: 80
Location: /dev/null
|
| Posted: Wed Sep 22, 2004 6:08 pm Post subject: |
|
|
| nmcsween wrote: | | I'll look into why you can't set the timer to 2000. |
Easy, just enabled Extended RTC operation.
Still, the kernel's crashy a bit, maybe due to voluntary-preemption or softirq preemption. (Hardirq doesn't work with nvidia drivers)
I'm currently using 2.6.9-rc2-mm1 + ck patches - works fine. |
|
| Back to top |
|
|
|
Kernel & Hardware
