** No longer affects: null
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/491943
Title:
Kernel trace buffer should be set to less unrealistic value
To manage notifications about this bug go to:
http
This bug was fixed in the package ureadahead - 0.100.0-4.1.3
---
ureadahead (0.100.0-4.1.3) lucid-proposed; urgency=low
* Count the number of CPUs and divide buffer_size_kb by the number of CPUs.
Users should watch for instances of "mmiotrace has lost events" in dmesg to
make
** Tags added: verification-done
** Tags removed: verification-needed
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
--
ubuntu-bugs m
Installed ureadahead-0.100.0-4.1.3 from lucid-proposed.
oom-killer not invoked and no mmiotrace lost errors in dmesg.
Now we can reliably boot linux-image-2.6.32-24-generic-pae-2.6.32-24.42.
We had been using 2.6.32-22.36 which had allowed us to finish booting.
So, it works for us, thanks!
--
K
** Branch linked: lp:ubuntu/lucid-proposed/ureadahead
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
--
ubuntu-bugs mailing list
ubun
Ah, so buffer_size_kb is _per CPU_; that makes sense then, thanks!
** Changed in: ureadahead (Ubuntu Lucid)
Status: In Progress => Fix Committed
** Tags added: verification-needed
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You re
Martin - the value written to /sys/kernel/debug/tracing/buffer_size_kb
is the amount of _contiguous_ RAM allocated for each CPU. The initial
value of 128000 was somewhat arbitrary, so I decided to proceed likewise
and arbitrarily divide the value by the number of CPUs. I've tried it on
an Emerald c
Tim,
can you please explain why the buffer should get smaller with the number
of CPUs? Why are these figures related?
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
Bugs,
** Project changed: hwe-somerville => null
** Changed in: null
Status: In Progress => Invalid
** Changed in: null
Assignee: Bryan Wu (cooloney) => Chris Van Hoof (vanhoof)
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You recei
** Branch linked: lp:ubuntu/ureadahead
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
--
ubuntu-bugs mailing list
ubuntu-bugs@lists.u
This bug was fixed in the package ureadahead - 0.100.0-7
---
ureadahead (0.100.0-7) maverick; urgency=low
* Count the number of CPUs and divide buffer_size_kb by the number of CPUs.
Users should watch for instances of "mmiotrace has lost events" in dmesg to
make
sure the tr
SRU Justification
Impact: ureadahead can cause the kernel to allocate too much RAM for tracing on
a hyper-threaded system.
Symptoms vary, but can result in an unbootable machine.
Patch Description: Count the number of CPUs and divide the amount allocated of
trace buffer by that number.
ureadahe
Tim,
The testing machine is occupied by other team. As long as we got the
machine, I will check that mmiotrace warning.
But according to comment #23, it seems like that we cannot make sure
128M is enough for every machine, right? IMHO, kernel is supposed to
know such requirement, userspace applic
Bryan - Have you checked for "mmiotrace has lost events" in dmesg
according to comment #23 ? As for /proc getting mounted before
ureadahead starts, I assume upstart is in charge of prioritizing those
actions.
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.ne
Tim,
We tested on our buggy machine, this patch fixes that issue. Thanks a
lot.
Just one question, if the /proc filesystem is not ready before
ureadahead, how can we get the right CPU numbers of a system.
-Bryan
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launch
Tim,
We tested on our buggy machine, this patch fixes that issue. Thanks a
lot.
Just one question, if the /proc filesystem is not ready before
ureadahead, how can we get the right CPU numbers of a system.
-Bryan
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launch
** Tags added: patch
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
--
ubuntu-bugs mailing list
ubuntu-bugs@lists.ubuntu.com
https://
http://alinux.tv/Kernel-2.6.34/trace/mmiotrace.txt
Check that mmiotrace did not lose events due to a buffer filling up. Either
$ grep -i lost mydump.txt
which tells you exactly how many events were lost, or use
$ dmesg
to view your kernel log and look for "mmiotrace has lost events" warning. If
ev
Scott - how can I tell if the trace buffers are not big enough? Is there
some indication of overflow, or do you just get inadequate readahead?
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a
** Patch added: "Divide trace buffer size by number of CPUs"
https://bugs.edge.launchpad.net/ureadahead/+bug/491943/+attachment/1495969/+files/0001-Count-the-number-of-CPUs.patch
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You receive
** Also affects: hwe-somerville
Importance: Undecided
Status: New
** Changed in: hwe-somerville
Assignee: (unassigned) => Bryan Wu (cooloney)
** Changed in: hwe-somerville
Importance: Undecided => High
** Changed in: hwe-somerville
Status: New => In Progress
--
Kernel
This fix works for me as well. Thanks Tim.
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
--
ubuntu-bugs mailing list
ubuntu-bugs@li
Good news here.
I had the same issue as described by Bryan in comment #14 on my Sony
VPCF11C5E: Core i7 4 GB, the system cannot boot most of the times unless
/sbin/ureadahead is removed.
Now it works with ureadahead 0.100.0-4.1.3 from timg-tpi PPA.
Thanks,
stefano
--
Kernel trace buffer should
Tim,
Thanks, I will try it on our testing machine.
-Bryan
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
--
ubuntu-bugs mailing lis
Bryan - please try ureadahead 0.100.0-4.1.3 in my PPA at 'deb
http://ppa.launchpad.net/timg-tpi/ppa/ubuntu lucid main'. This version
divides buffer_size_kb by the number of CPUs so that the total
allocation does not exceed 128000.
--
Kernel trace buffer should be set to less unrealistic value
ht
** Changed in: ureadahead (Ubuntu)
Status: Triaged => In Progress
** Changed in: ureadahead (Ubuntu)
Assignee: (unassigned) => Tim Gardner (timg-tpi)
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notificati
An easy way to disable ureadahead without uninstalling is to just move
/etc/init/ureadahead* someplace else. eg. "mv /etc/init/ureadahead*
/root"
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you a
I have a 12 GB system with 2 quad-core Xeons running with lucid 32-bit
PAE kernel, and am getting this as well.
[ 13.068253] ureadahead invoked oom-killer: gfp_mask=0xd0, order=0, oom_adj=0
[ 13.068348] ureadahead cpuset=/ mems_allowed=0
[ 13.068433] Pid: 434, comm: ureadahead Not tainted 2.
We got a very critical issue on an Intel machine with 4 cores and 8
threads CPUs. Since we got 4G RAM, we are using 32bits -pae kernel.
ureadahead will try to get 128M trace buffer percpu, so it will allocate
128M * 8 = 1G RAM. Normally it will be killed by OOM, but system cann't
boot at all mostly
If you take a look at the forums you will notice that there are alot of
threads where I have recommended this for low RAM systems. The last time
ureadahead was on my laptop it took over 2 hours to get anywhere after a
reprofile. I just automatically remove it on low RAM systems now because
of this.
Also seen on a 2GB machine with the server kernel.
Note that you can do an:
apt-get --purge remove ureadahead
as a workaround, with largely no ill effects, if you aren't relying on
any of its functionality directly.
This will remove the meta-package ubuntu-minimal, so once this bug is
fixed
uname -r
2.6.35-2-generic-pae
[7.973371] Out of memory: kill process 375 (ureadahead) score 37 or a child
[7.973373] Killed process 375 (ureadahead) vsz:2372kB, anon-rss:132kB,
file-rss:560kB
[7.975277] ureadahead: page allocation failure. order:0, mode:0xd0
[7.975278] Pid: 375, c
Another system with 4GB of memory (Lenovo Thinkpad W510):
Jun 6 20:57:26 bigthink kernel: [4.392338] ureadahead invoked oom-killer:
gfp_mask=0xd0, order=0, oom_adj=0
Jun 6 20:57:26 bigthink kernel: [4.392342] ureadahead cpuset=/
mems_allowed=0
Jun 6 20:57:26 bigthink kernel: [4.39
same errors with 35-1-generic-pae on maverick a1 with 4 go ram
** Attachment added: "syslog"
http://launchpadlibrarian.net/49759331/syslog
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are
kernel 34-5 generic-pae on maverick:
ureadahead invoked oom-killer: gfp_mask=0xd0, order=0, oom_adj=0
kernel: [8.729062] ureadahead cpuset=/ mems_allowed=0
kernel: [8.741183] Pid: 343, comm: ureadahead Not tainted
2.6.34-5-generic-pae #12-Ubuntu
kernel: [8.753804] Call Trace:
kern
Same here: on every machine I have, all having 4GB or more memory
available. It always kills plymouthd process when called.
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
B
I get this in dmesg every boot (multiple times) and I have 2 GB:
[ 12.742072] ureadahead invoked oom-killer: gfp_mask=0xd0, order=0, oom_adj=0
[ 12.742076] ureadahead cpuset=/ mems_allowed=0-1
[ 12.742080] Pid: 578, comm: ureadahead Not tainted 2.6.32-18-server
#27-Ubuntu
[ 12.742082] Cal
Will use bug #501715 for that
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
--
ubuntu-bugs mailing list
ubuntu-bugs@lists.ubuntu.com
Another point is that the trace buffer is not shrunk after profiling;
this was for my own debugging, but I think by default we should reclaim
this kernel memory
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification
This is set to 128MB or something at the moment; that's ridiculously
high
--
Kernel trace buffer should be set to less unrealistic value
https://bugs.launchpad.net/bugs/491943
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
--
ubuntu-bu
This is set to 128MB or something at the moment; that's ridiculously
high
** Summary changed:
- When trying to profile on a system with 512MB -- get a kernel OOM
+ Kernel trace buffer should be set to less unrealistic value
** Changed in: ureadahead (Ubuntu)
Status: Confirmed => Triaged
41 matches
Mail list logo
