The problem is described here: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/134856 Figured out that an entry in the set_max_cstate table in acpi/processor_idle.c fixes that problem.
Created attachment 17415 [details] working fix
Hi, Matthias Will you please try the boot option of "processor.max_cstate=2" and see whether the problem still exists? Please don't apply the patch in comment #1 and the processor (in the directory drivers/acpi/) should be compiled as built-in modules. Thanks.
Hello Yakui, I've tried that max_cstate=2 thing and first I thought "oh cool this is working too" because the slow-HD problem disappeared, but my system was unusual unstable (randomly freezes after minutes or hours) so actually its not a preferable solution. Similar freezes occur when running Linux without a max_cstate "limitation" and without "noapic" kernel-parameter, there the freezes occurs more frequently including the display turns black (AFAIK this is or was a known problem on many HP Laptops)... with max_cstate=1 none of these things happen even with apic enabled. I hope i could help greets, Matze
Hi, Matthias Maybe the system can work well only in case of "processor.max_cstate=1". Will you please attach the output of acpidump? Thanks.
Created attachment 17601 [details] dump created with acpidump Here it is :)
From the log in https://bugs.launchpad.net/ubuntu/+source/linux/+bug/134856 it seems that the issue can be fixed by adding the boot option of "nohz=off" or disabling tickless feature. Will you please try the boot option of 'nolapic_timer" and see whether the problem still exists? Thanks.
(c/p from my launchpad post to keep this up) Greetings, adding the "nolapic_timer" parameter leads to system freeze/blank screen during boot before getting to init :-(
hmmm, what if update your kernel to the latest kernel release, say 2.6.27?
currently using 2.6.27.5 but still need the patch. nohz=off is still an alternative but IMHO the less preferable workaroud
well, workaround is available, but we have no idea how to debug this problem. len and Venki, should we merge the patch in comment #1 or keep on searching for the root cause? re-assign to Venki.
Matthias, can you try powertop on this laptop? and see if there are a signaficant number of wakeup events.
Copying amd folks. Are C2 and C3 states supposed to work on this CPU? Thanks.
Here are my investigations with powertop: kernel 2.6.27.7, limited to C1 state: If I dont move anything, the most wakeup events come from the b43 w-lan driver which IMO seems normal. Type something on the keyboard or move mouse/touchpad causes many wakeups.. I guess this is normal too. with parameters "processor.max_cstate=9 nohz=off" : Powertop shows that C2 and C3 states are "available" but never uses them... I dont know when these states should be running but I also tried to run the Laptop from battery and turning of the ondemand gouvaneur and forced the CPU to run with 2GHZ (maximum).... only C1 state is used but again I dont know enough about these stuff so I could be totally wrong. with "processor.max_cstate=9" only: system boots slow, and freezes during init ... no chance to do investigations there. MAYBE the nohz feature forces the CPU to enter C2/C3 states, which cause the system to fail ? But strange that it reports that these states are available... thats all I could say now, I hope that helps in some way greetz, Matze
I can't tell if it's a Turion X2 or a Sempron. Could you dump /proc/cpuinfo, please?
Created attachment 19102 [details] dump of /proc/cpuinfo
Thank you. The revision guide clearly shows it to be a Mobile Semprom, and the BIOS guide indicates those should support C2/C3 and *not* C1e. It's possible that the BIOS isn't setting up the Cx SMAF codes correctly, but that seems unlikely. There is an erratum dealing with C3, but a) it causes a hang, not a slowdown b) there's a BIOS work-around. So as long as the BIOS is up-to-date, I am not sure what is happening here.
You're welcome, AMD-Guy :) With "hang" do you mean a short slowdown thats not even noticeable, or a hang "forever" ? Because this is actually happening when Linux is running with max_cstate=9 and nohz enabled ... BIOS is also up-to-date but I saw some people with HP Laptops who had similar problems running Linux... so maybe its HP's BIOSes for some machines who are fcked-up. If there is any more information/dumps/whatever I could supply to identify the problem, just let me know. Im not experienced enough to track the problem by myself but I find it stange that the C2/C3 states didnt cause problems with nohz disabled ...
Apparently, this is now fixed in 2.6.28: $ sudo hdparm -tT /dev/sda /dev/sda: Timing cached reads: 902 MB in 2.00 seconds = 450.42 MB/sec Timing buffered disk reads: 126 MB in 3.00 seconds = 41.98 MB/sec
HI, Fabio Good news. It seems that this is already fixed. Hi, Venki How about close this bug? Thanks.
close this bug as it's fixed in 2.6.28.