Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Archived topics about LMDE 1 and LMDE 2
Locked
jasmineaura

Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by jasmineaura »

I'm still plagued with Total Freezes / Lockups. Nothing at all in syslog to point to what went wrong before the system froze, every time. :( :( :(

Prelude:
I was also having the same problem described in bug #1032980 ; LMDE MATE/Cinnamon 2012/04 installation always freezes at some random point of copying files. Similar hardware too here, i5-3210m (intel hd 4000) ivy bridge, Radeon HD 7670 -- Dell Inspiron 15R 5520. But I somehow eventually got it to install, but once I did, the display x-offsetting was wrong -- screen was shifted to the right (and wrapped around left) of the LCD, from boot (right after console switches to high res). The only thing that would mitigate that was disabling i915 modeset, which wasn't acceptable as I needed KMS to turn the radeon off via vgaswitcherro switch in debugfs (from /etc/rc.local). So for a while, I used liquorix kernel 3.4, and then switched back to the new kernel from incoming: 3.2.0-3-amd64 (3.2+45), and now freezing.

General info:

Only using LMDE incoming repos. Only foreign packages are from dell factory installed ubuntu for dell bcm43142 wlan/bluetooth card driver, and dell_laptop module -- both dkms installed. Freezes occur whether or not btusb module is loaded (when I don't often use). only using the "wl" module (for wifi).

fglrx not installed, radeon turned off at boot via debugfs vgaswitcheroo switch. Always using i915 video.

GRUB_CMDLINE_LINUX_DEFAULT="quiet pcie_aspm=force"

# cat /etc/modprobe.d/radeon-kms.conf
options radeon modeset=1

# cat /etc/modprobe.d/i915-kms.conf
options i915 modeset=1 i915_enable_rc6=1 i915_enable_fbc=1 lvds_downclock=1 semaphores=1

/home/jas# free -m
total used free shared buffers cached
Mem: 3868 1090 2777 0 121 416
-/+ buffers/cache: 552 3315
Swap: 4310 0 4310


Description of the problem:
Freezes can happen every other day, after several days, or twice or three times in a row, like this morning. Without any prior warning or signs. Last 3 times this morning while using firefox,first time when opening an image link (the image rendered half way and laptop froze). The second two times while hovering over left-side of browser for tree-style-tab (ff-extension) auto-hidden vertical tab list to show (w/ transparency). Only a terminal with 2-3 tabs and pluma open, otherwise. And only additional (mate) applets used are CPU Frequency Scaling Monitor 1.4.0 & MATE Sensors Applet 1.2.0.

Problem was exacerbated yesterday when I changed /etc/modprobe.d/i915-kms.conf
from:
options i915 modeset=1 i915_enable_rc6=1 i915_enable_fbc=1 lvds_downclock=1 semaphores=1
to:
options i915 modeset=1 powersave=1 semaphores=1
And so I switched back. Freezes didn't occur until this morning.

The one thing I've been doing more lately, is:
echo "min_power" | tee /sys/class/scsi_host/host*/link_power_management_policy

It seems to immediately quite down the laptop's fan, which is always otherwise running, regardless of AC/BAT status.

Is there anyway to make the kernel spit out anything to anywhere (local disk, remote syslogger, a network-connected vt) before freezing?
Any pointers to how to debug this? I feel I have absolutely nothing I can point my finger at with certainty to warrant filing a bug report.

From preliminary research, seems it is a problem with Ivy Bridge and Intel HD 4000 in 3.2.x Kernels:
Partially Sane developer (blog) - Ivy Bridge HD4000 Linux Freeze
Hardforum: Linux Bug Ivybridge
Ubuntu bug #999910 [IVB]12.04 64 bit System freezes (mouse, keyboard)

Guess I will have to deviate from LMDE's default (3.2.x) kernel, for now.

Any ideas?

Edit: Now running 3.5.0-4.dmz.1-liquorix-amd64 , to see if the freezes will cease. Will test thoroughly, put the i915 under various loads, and report back.
Last edited by LockBot on Wed Dec 28, 2022 7:16 am, edited 1 time in total.
Reason: Topic automatically closed 6 months after creation. New replies are no longer allowed.
abickerton

Re: Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by abickerton »

I'm running a 3.5.2 kernel and can report that the freezes are gone. I still cannot use compiz due to it crashing far too often, at least it's not a complete lock up anymore.
zerozero

Re: Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by zerozero »

[moved here]
jasmineaura

Re: Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by jasmineaura »

abickerton wrote:I'm running a 3.5.2 kernel and can report that the freezes are gone. I still cannot use compiz due to it crashing far too often, at least it's not a complete lock up anymore.
Nods. I'm only mostly on MATE till now. No lockups so far with liquorix kernel 3.5.0-5 , and I've been throwing various loads at it without being able to cause it to lock up.

Sadly, running powertop, after reactivating all my little power tweaks, one by one, I notice I don't get as much power savings as with 3.2.0-3 from LMDE UP5. I can dare to say I'm losing about 1-2Watts savings, which is a bummer.

Phoronix: Linux 3.5 Kernel Power Regression Spotted - (i915 & semaphores commit) - July 31

I've had semaphores enabled on i915 since 3.2.x. Disabling it via sysfs doesn't seem to make a difference in terms of power drain. :(

But all is not lost! 3.6-rc6 looks promising, with Ivy Bridge CPU-Idle support, i915 performance improvements, radeon PCIe-2.0 default enabled
Phoronix: Interesting Features For The Linux 3.6 Kernel - September 15
Intel Graphics Hit High Point With Linux 3.6 Kernel - August 16

Edit:
Ah, doh! Liquorix 3.5 kernel still uses BFS scheduler (CONFIG_SCHED_BFS=y), which isn't in mainline kernel (and never will be). Also, Liquorix sets CONFIG_HZ=1000 and enables PREEMPT, as opposed to 250Hz and disabled PREEMPT in LMDE Kernel 3.2.0-3.
Per the author's documentation
Running it on a real server I'd recommend the default rr_interval, 100Hz, no preempt and dynticks ON (for power saving).
jasmineaura

Re: Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by jasmineaura »

3.2.0-3 (LMDE UP5) , ~ 9.15W lowest idle , wlan on, screen on & dimmed
3.5.0-4 (Liquorix) , ~ 10.6 - 11.4W depending on its mood for idling, wlan on, screen on & dimmed

liquorix kernel power usage measured after switching from bfq to deadline for fair comparison ( echo deadline > /sys/block/sda/queue/scheduler )

:(((
watertavi

Re: Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by watertavi »

I use 3.4.4 without any problems and a similar setup. (No word on power though)
jasmineaura

Re: Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by jasmineaura »

Using 3.6-rc7 I can confirm that power regression in 3.4.x/3.5.x is FINALLY FIXED. That is, with 3.6-rc7, I'm getting idle power usage relative to that I was getting on 3.2.0-3 -- in the 9W range (comparable to win7, yay) -- and so far without freezes.

However, the dreaded, unpublished broadcom-sta-dkms driver (6.20.55.19) triggers new bugs with wpasupplicant 1.0-2 and this new (development) kernel.
As I already had a report to debian to add this new driver, I followed up on it and added linux-wireless to the list:
http://bugs.debian.org/cgi-bin/bugrepor ... =688823#10
Though I probably can't be helped given the proprietary nature of this driver.
Also on ubuntu: https://bugs.launchpad.net/ubuntu/+sour ... bug/923809

So:
3.2.x = Total lock-ups
3.3.x = EOL
3.4.x = power-regression
3.5.x = power-regression
3.6-rc7 = no wifi

I'm so doomed :D
jasmineaura

Re: Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by jasmineaura »

abickerton wrote:I'm running a 3.5.2 kernel and can report that the freezes are gone. I still cannot use compiz due to it crashing far too often, at least it's not a complete lock up anymore.
Curious which version of compiz you were using (pre-UP5?) when you were getting frequent crashes, and whether this was with metacity or marco. I was testing 0.8.4-5.1 yesterday, with marco, and it seemed to work pretty well, except that it needs to be explicitly told to use marco instead of metacity for window decoration.
Compositing is also working well in marco, apparently. Either by `marco --replace -c`, or Control Centre > Desktop Settings > Windows > Use Gnome Compositing, and restart desktop session.
jasmineaura

Re: Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by jasmineaura »

abickerton wrote:I'm running a 3.5.2 kernel and can report that the freezes are gone. I still cannot use compiz due to it crashing far too often, at least it's not a complete lock up anymore.
try compiz-mate 0.8.8 fork: http://forums.linuxmint.com/viewtopic.p ... 7&t=114613
ddurdle
Level 3
Level 3
Posts: 153
Joined: Sat Dec 17, 2011 10:03 am

Re: Linux 3.2.x, Ivy Bridge, Intel HD 4000 == total freezes?

Post by ddurdle »

I started getting the freezing once I enabled pcie_aspm=force i915.i915_enable_fbc=1 i915.lvds_downclock=1. These are typical settings I use on all notebooks, but on this machine, after enabling them today, I've had two freezes. It is annoying on the ultrabook because I have to disassemble the machine so I can pull the leads off the battery to power it off.

Happening on all kernels 3.5-3.7 that I've tested.
Locked

Return to “LMDE Archive”