DIABLOTURBO comes back, with a vengeance! - Page 2 - maemo.org

Active Topics

Extra softwares in Sailfish using CLI, repositories, etc (117)
to SailfishOS by nieldk - 22 hrs, 55 mins ago
Firefox with Leste (6)
to Maemo 7 / Leste by teroyk - 1 day, 19 hrs ago
more...

Page 2 of 5

< Prev

Next >

Last »

Thread Tools

maacruz	2011-01-12 , 20:09
Posts: 1,101 \| Thanked: 1,184 times \| Joined on Aug 2008 @ Spain	#11

Originally Posted by wnd

I just installed and uninstalled cfs-ugroup-preemptnovermagic-256. Some extremely hasty and pessimistic observations. I think I'll try other configurations later; tonight I'm going to need a working device.

- Applications may have reacted faster to mouse events.

Good, that's the objective.

- Applications seem to receive mouse release events when they shouldn't. As strange as it sounds, panning in Maemo Mapper is much more difficult; half of the time panning stops after a fraction of a second. I don't remember any issues like with with SSU kernel.

I suffer this issue since long no matter the kernel. I think it is a hardware issue. May be the faster timer exacerbates it.

- When I power on, I can see Nokia splash screen gradually replaced by Hildon desktop. I've seen this before, but it's been months and months since the last time. This seems to happen every time I boot up. In related news, boot procedure (to desktop) seems to take significantly longer than before.

I've seen this in all preempted kernels, and I think it is more noticeable with the 256 Hz timer. This is to be expected, since there are more interrupts/second and preemption of processes even when making kernel calls, the cpu is more spread over all processes at boot, and the desktop appears more slowly. With slow timer and no preemption, the cpu spends more time rendering the desktop in one go while starving the other processes. Also, the different concept of fairness (how much cpu time a process should receive) among the different schedulers affects this.
With respect to boot time, I did a clock measure out of curiosity on sunday (I think it was on the same kernel), from start to 0 cpu (with metalayer-crawler disabled), and the time appeared to be basically the same. Anyway, as I already said, it is expected to be somewhat slower because of the increased context switches. More clock measures are needed to be sure of how much.

- During by second and third power-on sessions, ksoftirqd/0 would consume most of the CPU time. Battery life was projected to be hour hours instead of the usual seven. This may have been related to (usually harmless) JFFS warnings about JFFS internal storage space in kernel log.

I've also seen this issue. It doesn't go until you reboot, and happens sometimes. I think it could be related either to the CFS or to the faster HZ timer, but a google search shows many complaints of the same problem in some people's desktop hardware, and that it was eased with some very recent changes. Yesterday I was running O1-preempt-256 and today sd-preempt-256 and I haven't seen this issue, but I've rebooted very few times too. More testing needed with 128-256 and/or CFS-OtherScheduler to be sure where the problem lies.

- During my third session I could not run sudo gainroot; all four tries simply stopped after I pressed enter.

This surely happens if rootfs fills (because of big growing logs or any other reason). There may be other reasons, but take care the former doesn't happen, make sure you have always free space.

- Hildon UI clearly receives less CPU time. When the screen is powered after th
e device being idle, some screen elements may be missing at first. This happens nearly every single time with battely level status bar applet.

Since the device has been idle, this clearly isn't a cpu time issue. It could be a symptom of some kernel instability. Keep an eye on any other strange symptom to confirm or deny this.

- Installer script expects to find all (currently) installed modules in kbuilds
/foo. This means that if you have additional kernels such as those in kernel-diablo-modules-extra installed, you get very uninformative warning for every and each one of those.

Please elaborate a bit more. Is it an issue with the installer script? Please post one of such messages and your module's directory listing if so. Or are you referring to kernel log messages? I have other modules (fuser, netfilter, dmloop, rt73 and some more) and haven't seen any warning.
The script doesn't expect to find all installed modules in kbuilds/foo, it just looks for those existing in kbuilds/foo to replace the installed ones under /lib/modules/2.6.21-omap1. Extra modules should be ignored at this moment, except for the backup. Of course, there may be a small bug...

- WLAN/WiFi didn't work properly. I could scan available networks, but actually connecting one failed.

I haven't had any issue with wifi. Is the AP open, WEP or WPA?

Quote & Reply |

The Following User Says Thank You to maacruz For This Useful Post:
Kroll

bousch	2011-01-12 , 21:21
Posts: 120 \| Thanked: 69 times \| Joined on Nov 2007 @ NL	#12

Thanks for your effort.

Will this work when booting from MMC or only when booting from internal flash?

Quote & Reply |

maacruz	2011-01-12 , 22:11
Posts: 1,101 \| Thanked: 1,184 times \| Joined on Aug 2008 @ Spain	#13

I always boot from mmc, leaving internal flash as a recovery system to use when things go wrong.

Quote & Reply |

wnd	2011-01-12 , 23:47
Posts: 99 \| Thanked: 65 times \| Joined on Jan 2008 @ Finland	#14

First of all, I can't go into these things in more detail right now so you'll have to do with this quick reply.

Originally Posted by maacruz

(sudo gainroot not working) This surely happens if rootfs fills (because of big growing logs or any other reason). There may be other reasons, but take care the former doesn't happen, make sure you have always free space.

rootfs should've had around 20 MiB, so free disk space should not have been the issue. Right now I'll just ignore this and move on.

(installer) Please elaborate a bit more. Is it an issue with the installer script? Please post one of such messages and your module's directory listing if so. The script doesn't expect to find all installed modules in kbuilds/foo, it just looks for those existing in kbuilds/foo to replace the installed ones under /lib/modules/2.6.21-omap1.

I may be mistaken so correct me if I'm wrong. Although the installer only tries to replace modules currently in /lib/modules/$KVER/, I think the line following cp will echo an error if the module is not found in kbuilds. Note that "test -e foo && true" will set $? as non-zero if foo doesn't exist.

I haven't had any issue with wifi. Is the AP open, WEP or WPA?

WPA. I should be able to try open AP also. What I didn't remember to say before is that trying to join the AP instantly gave an error. In addition, I never had time to see system logs.

Quote & Reply |

The Following User Says Thank You to wnd For This Useful Post:
maacruz

maacruz	2011-01-13 , 16:36
Posts: 1,101 \| Thanked: 1,184 times \| Joined on Aug 2008 @ Spain	#15

Originally Posted by wnd

I may be mistaken so correct me if I'm wrong. Although the installer only tries to replace modules currently in /lib/modules/$KVER/, I think the line following cp will echo an error if the module is not found in kbuilds. Note that "test -e foo && true" will set $? as non-zero if foo doesn't exist.

Oh, you are right, thanks

I've attached a corrected install script, just replace the old one this one.
The bug is merely cosmetic and only affects those who have extra kernel modules not in subdirectories.

For those in doubt, I'll explain a little more about the different kernels:
The first set of letters after zImage is the scheduler used:
O1: Same as in stock kernel, the old (<2.6.22 kernel) scheduler.
sd: Staircase deadline scheduler, designed by Con Kolivas as a better replacement for O1. After Con presented SD, Ingo made CFS taking some ideas from SD design.
cfs: Completely fair scheduler, present in current kernels.
cfs-ugroup: cfs with groups. This cfs variant has task grouped, and divides the cpu fairly among groups, then among tasks inside each group.
The scheduler decides when and how much cpu time a task gets.
"preemptnovermagic": indicates the kernel has been built with voluntary preemption, so tasks can be interrupted by the scheduler even when making kernel calls, in certain predefined points. In non-preempted kernels tasks can't be interrupted when making kernels calls, so a task can monopolize the cpu for a long time if it makes an expensive kernel call.
The last number (128 or 256) is the interrupt timer. 128 is the same as in the stock kernel build. It affects the number of task switches per second.

For example, stock kernel would be named: zImage-O1-128HZ
O1 kernels won't differ much of stock kernels.
Most different are the cfs kernels. I suspect there may be some instability or subtle bug with them, but I need confirmation. Also, cfs with groups is the one which holds more promise with respect to optimization. Since kernel bugs can cause filesystem damage, it may be a good idea to make a backup before testing them.
I've run all 256HZ kernels
While running O1 and SD kernels I haven't had any issue. Currently I'm running SD 256, and I think it could be quite a bit better than O1.
sd and O1 kernel binary modules are identical, so stock modules should run on both with no problem. Preemption is not a problem since it's voluntary preemption. cfs modules are different, and I have had subtle problems with the stock ext3 modules running on cfs.

Attached Files

install.sh.gz (1.6 KB, 150 views)

Quote & Reply |

eneitten usr XD	2011-01-13 , 17:10
Posts: 59 \| Thanked: 10 times \| Joined on Dec 2010 @ / Puerto Rico/ Arecibo	#16

wooow im sooo happy someone still loves my N810 so i was wondering im not too sure about testing for few reasons, the main one being whatever i try if anything went wrong ?would JUST REFLASHING FIX IT ,,,OR WOULD I MESS IT UP AN TOAST IT? IM NOT SCARED TO TRY JUST DON WANNA LOOSE MY DEVICE ITS ALL I GOT ;-( EITHER WAY MAAAAAN U ARE AWESOME FOR WORKIN ON THIS IT IS A HUGE RELIEF TO C THIS "WORK IN PROGRESS" PROPS TO ALL THIS COMUNITY ILL TEST JUTS LET ME KKNOW IF IS NOT GONNA KILL MY BABY OR IF JUST FLASHING WOULD TAKE CARE!

__________________
the world beyond is what i discovered when i found Maemo,,,...© [°I__ I ] © Thank you LORD 4 ALL XD

About eneitten usr XD

Location
mississippi /\&\/ Puerto Rico
Operating System

Maemo4 (Penguin SMASHING windows lmbo]
Mobile Phone
it doesnt matter i still gotta wait to get a nokia what else...,,,,waiting..,,,,,,
Handhelds
15 months N810 all day everyyyyyy day thats it!!: g eek:

Quote & Reply |

maacruz	2011-01-13 , 20:08
Posts: 1,101 \| Thanked: 1,184 times \| Joined on Aug 2008 @ Spain	#17

If I believed that any kind of physical damage could be possible, I wouldn't be doing it myself.
The worst I think that can happen is having rootfs borked. And yes, a full reflash will leave the device in factory condition.

Quote & Reply |

eneitten usr XD	2011-01-13 , 21:48
Posts: 59 \| Thanked: 10 times \| Joined on Dec 2010 @ / Puerto Rico/ Arecibo	#18

so just plug it in reflash and. were back online !!! if so im not that scared anymore of trying all u guys hints and stuff i find interresting to do with my n810 i just need nubie walkthroughs kinda but ill try anything eversince i goth this tablet it oppened up a whoooole world i really wanna learn to maybe write a program or make a theme orsomething i find all this realy interesting. thanks to people like you that heeelps us out over this side an even inspire me like a couple others i c that put time and heart into these devices here on maemo many thumbs up

GOD bless u all+++

n§

P.S. muchas gracias por su gran trabajo!!

About eneitten usr XD

Location
mississippi /\&\/ Puerto Rico
Operating System

Quote & Reply |

wnd	2011-01-14 , 13:45
Posts: 99 \| Thanked: 65 times \| Joined on Jan 2008 @ Finland	#19

I now have O1-preemptnovermagic, cfs, cfs-preemptnovermagic and sd-preemptnovermagic, all with 128 Hz timer plus cfs-cgroup-preemptnovermagic with 256 Hz timer as mentioned before. Since I couldn't join WPA-PSK AP with any of them, I still didn't spend too much time playing around. There's very little in kernel and system log about this. However, connecting to an open WLAN AP works well.

Code:

syslog

<31>Jan 14 15:18:42 wlancond 1.3.10[1188]: SIOCGIWAP: 00:22:55:74:b6:54
<31>Jan 14 15:18:42 wlancond 1.3.10[1188]: ESSID:"<hidden/>"
<31>Jan 14 15:18:42 wlancond 1.3.10[1188]: Mode: Infra
<31>Jan 14 15:18:42 wlancond 1.3.10[1188]: Encryption: Yes
<31>Jan 14 15:18:42 wlancond 1.3.10[1188]: Channel: 1
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: RSSI: -80 dBm
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: RSN IE
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: IE includes PMKID data
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: WPA PSK supported
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: WPA AES supported for unicast
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: WPA TKIP supported for multicast

<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: AP (7) is <hidden/>, rssi:-48 channel:6 cap:60000141

<31>Jan 14 15:18:43 icd2 0.67[1201]: WLAN: Going up: type=WLAN_INFRA, nwattrs=0x06a1, iap="3f6e8d49-0bea-42d2-b31b-6654e5bfd490", up_cb=0xe139
<31>Jan 14 15:18:43 icd2 0.67[1201]: WLAN: IAP (3f6e8d49-0bea-42d2-b31b-6654e5bfd490) going up (state SEARCH_SSID)
<31>Jan 14 15:18:43 icd2 0.67[1201]: WLAN: 2816: state SEARCH_SSID (4) -> CONNECTING (1)
<31>Jan 14 15:18:43 icd2 0.67[1201]: WLAN: [ps-testlan] Configuring WLAN.
<31>Jan 14 15:18:43 icd2 0.67[1201]: WLAN: Sending WLANCOND_SETTINGS_AND_CONNECT_REQ: 0x8, "ps-testlan", 0x1, 0x60000140, 0x0
<31>Jan 14 15:18:43 icd2 0.67[1201]: WLAN: [3f6e8d49-0bea-42d2-b31b-6654e5bfd490] WLAN connecting...
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: Received settings_and_connect
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: Wlancond state change, old_state: WLAN_NOT_INITIALIZED, new_state: WLAN_INITIALIZED
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: Encryption setting: 60000140
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: WPA2 PSK selected
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: AES selected for unicast
<31>Jan 14 15:18:43 wlancond 1.3.10[1188]: TKIP Selected for multicast
<27>Jan 14 15:18:45 wlancond 1.3.10[1188]: WPA IE / SSID (ps-testlan) not valid
<31>Jan 14 15:18:45 wlancond 1.3.10[1188]: wlan0 is DOWN
<31>Jan 14 15:18:45 wlancond 1.3.10[1188]: Wlancond state change, old_state: WLAN_INITIALIZED, new_state: WLAN_NOT_INITIALIZED
<30>Jan 14 15:18:45 icd2 0.67[1201]: WLAN: [wlan0] Disassociated from network "3f6e8d49-0bea-42d2-b31b-6654e5bfd490"
<31>Jan 14 15:18:45 icd2 0.67[1201]: WLAN: [3f6e8d49-0bea-42d2-b31b-6654e5bfd490] Removing association timer because of disconnect (217)
<31>Jan 14 15:18:45 icd2 0.67[1201]: WLAN: [wlan0] calling link up failure for "3f6e8d49-0bea-42d2-b31b-6654e5bfd490", up=0xe139
<31>Jan 14 15:18:45 icd2 0.67[1201]: iap 0x40500 link_up callback did not set interface 'wlan0'
<30>Jan 14 15:18:45 icd2 0.67[1201]: iap 0x40500 callback in state ICD_IAP_STATE_LINK_UP with status 3, error '(null)', interface 'wlan0'
<30>Jan 14 15:18:45 icd2 0.67[1201]: IAP reports error, but error string NULL, set to 'com.nokia.icd.error.network_error'

Nothing in kernel log catches my eye. cx3110x scans the channels, prism_softmac_frame_tx() spews some warnings (as is customary) after which cx3110x shuts down softmac.

In addition to finding WLAN non-functional, I noticed the click-and-drag-problem with applications was gone when I was using O1 and plain CFS. I didn't use the device long enough to let the screen blank with every kernel, but I did notice the battely applet to behave badly with plain CFS kernel.

Quote & Reply |

The Following User Says Thank You to wnd For This Useful Post:
maacruz

maacruz	2011-01-14 , 17:59
Posts: 1,101 \| Thanked: 1,184 times \| Joined on Aug 2008 @ Spain	#20

Originally Posted by wnd

I now have O1-preemptnovermagic, cfs, cfs-preemptnovermagic and sd-preemptnovermagic, all with 128 Hz timer plus cfs-cgroup-preemptnovermagic with 256 Hz timer as mentioned before. Since I couldn't join WPA-PSK AP with any of them, I still didn't spend too much time playing around. There's very little in kernel and system log about this. However, connecting to an open WLAN AP works well.

Nothing in kernel log catches my eye. cx3110x scans the channels, prism_softmac_frame_tx() spews some warnings (as is customary) after which cx3110x shuts down softmac.

It must be the WPA-EAP patch. I'll remove it until I know more about it.

In addition to finding WLAN non-functional, I noticed the click-and-drag-problem with applications was gone when I was using O1 and plain CFS. I didn't use the device long enough to let the screen blank with every kernel, but I did notice the battely applet to behave badly with plain CFS kernel.

Do you mean that the problem goes away only in non-preempted kernels, or in O1 and plain CFS preempted goes away too?
Right now I'm working in a new touchscreen driver to try to solve this issue (it is the same as the repeated click that affects so many people)

Oh, I forgot to say this before.
A little warning: the cpu frequency governor doesn't work when the dsp is in use, so having /sys/power/op_dsp=0 means CPU pegged at 400 MHz when dsp is in use and battery life will be 30 % shorter when playing music if you do not set op_dsp=1 (CPU will be pegged at 333 MHz then, which is Diablo's default).

Last edited by maacruz; 2011-01-14 at 18:07.