System freeze when memory is put on SWAP in Linux >4.10.x
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Linux |
Expired
|
Medium
|
|||
linux (Ubuntu) |
Confirmed
|
Undecided
|
Unassigned |
Bug Description
I'm reporting this since it's reproduceable the 70% of the time.
Summary:
In different circumstances, when the systems starts to swap out RAM memory, even small amounts, the system becomes completely unusuable and the screen freezes up, no mouse movement, no TTY access or SSH access can be made, only SYSRQ keys seem to do something (only reboot, so REISUB worked so far though, OOM is useless since the memory/swap is not even full)
The I/O Disk led is stuck to 100% in ALL the following cases when this happens.
So far:
- This happens even when only ZRAM is enabled, and no swap partition is used.
- Happens when ZSWAP is used with a swap partition
- Happens also when a partition without zram or zswap is used
- Maybe it's AMD specific?
However, I'm not experiencing this on my laptop using the same tests.
My laptop is an Intel one, while my desktop is an AMD Ryzen platform.
Here are the specs:
CPU: AMD Ryzen 5 1600 no OC
GPU: AMD RX 580 8GB
SSD: Crucial MX500 500GB
MOBO: MSI B350M Grenade
RAM: 8GB HyperX Kingston 2667Mhz
Ubuntu version: 18.04 LTS, backports repo enabled
Kernel version: 4.18.0-18, official ubuntu repo
Bios settings: Default
Additional info: Maybe I'm not 100% sure, but I noticed when using the 5.0.0-17 generic kernel, the lockups seem to still happen, but they recover eventually. Happened only a few times though...
But will always be frozen for at least 30 seconds, differently from my intel laptop where those do occur.
The SSD make is the same. I bought two of these, they got also the same amount of RAM.
In my laptop those do not occur at all. Swapping memory even huge quantities like 1GB or more, do not produce any issues.
Tests made:
For testing this behaviour I tried:
- Compiling the chromium-browser source code (takes up a lot of system RAM)
- Used the "stress" command, using a specific amount of memory to decide how many it will be swapped, and here I noticed that even small quantities like a couple of megabytes will cause the system to freeze the 70% of the times
Example: "stress --vm 1 --vm-bytes=7G"
What should happen:
I expect system slowdowns when swapping out memory since I do not have enough RAM, but unlikely when using Windows or my laptop with the same Linux version, not a completely unusuable environment. The swap partition is in both cases on an SSD.
Reproduceability: 70% of the times
Additional info again:
I'm not sure this is due to any hardware failure, my SSD health is fine, as my CPU and RAM. As I said swapping in Windows works fine...
---
ProblemType: Bug
ApportVersion: 2.20.9-0ubuntu7.6
Architecture: amd64
AudioDevicesInUse:
USER PID ACCESS COMMAND
/dev/snd/
/dev/snd/
/dev/snd/
CurrentDesktop: communitheme:
DistroRelease: Ubuntu 18.04
IwConfig:
enp24s0 no wireless extensions.
lo no wireless extensions.
MachineType: Micro-Star International Co., Ltd. MS-7A37
Package: linux (not installed)
ProcFB: 0 amdgpudrmfb
ProcKernelCmdLine: BOOT_IMAGE=
ProcVersionSign
RelatedPackageV
linux-
linux-
linux-firmware 1.173.6
RfKill:
0: hci0: Bluetooth
Soft blocked: no
Hard blocked: no
Tags: bionic
Uname: Linux 5.0.0-17-generic x86_64
UpgradeStatus: No upgrade log present (probably fresh install)
UserGroups: sudo video
WifiSyslog:
_MarkForUpload: True
dmi.bios.date: 01/22/2019
dmi.bios.vendor: American Megatrends Inc.
dmi.bios.version: 1.K0
dmi.board.
dmi.board.name: B350M MORTAR (MS-7A37)
dmi.board.vendor: Micro-Star International Co., Ltd.
dmi.board.version: 1.0
dmi.chassis.
dmi.chassis.type: 4
dmi.chassis.vendor: Micro-Star International Co., Ltd.
dmi.chassis.
dmi.modalias: dmi:bvnAmerican
dmi.product.family: To be filled by O.E.M.
dmi.product.name: MS-7A37
dmi.product.sku: To be filled by O.E.M.
dmi.product.
dmi.sys.vendor: Micro-Star International Co., Ltd.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#20 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#21 |
First - using kernel 4.10.17 - which does not show any issues in swapping:
I tried doing: swapoff /dev/sda2
Attached output as vmstat-
18:27:00 - Launched Cities: Skylines
18:27:30 - Started loading the saved game
18:28:25 - About this time the game started doing its thing. Started scrolling
around.
18:28:47 - System stopped responding and then the C:S was killed by the OOM
handler
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#22 |
Created attachment 258045
vmstat-
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#23 |
Created attachment 258047
vmstat-
Second test, same kernel with swap turned on:
I have attached the vmstat output that goes with the following timestamps for
system utilisation:
15:32:30 - Launch Skylines
15:33:00 - Load the saved game
15:34:11 - Saved game loaded ok.
15:35:00 - Launch Chrome.
15:35:36 - Chrome launched - System responding ok.
15:36:00 - Browsing a few web sites
15:36:50 - Exit Chrome
15:37:30 - Exit Cities: Skylines.
You'll note that there are very few missing vmstat lines - however I did
notice the following missing:
15:35:10
15:35:12
15:35:15
15:35:20
15:35:26
15:35:29
15:35:30
Attachment is vmstat-
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#24 |
Created attachment 258049
vmstat-20Gb.log (OK - all in RAM)
Now using kernel 4.11.x (same happens with 4.12.x) - and testing with 20Gb of RAM in the system - meaning no swapping.
Attached as: vmstat-20Gb.log
Timestamps of events:
21:57 - launch the game from within Steam.
21:58:00 - Load the saved game.
21:58:48 - Saved game is loaded and I'm scrolling around in the map.
22:00:00 - Hit the quit to desktop button.
22:00:31 - Am back to desktop with all RAM free again.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#25 |
Created attachment 258051
vmstat-10Gb.log (NOT OK - System Unresponsive)
I now drop back to 10Gb of RAM to test the swapping under 4.11.x kernel.
Log attached as vmstat-10Gb.log
Timestamps:
22:10:00 - Launched of the game from within Steam
22:11:00 - Load the same saved game from the previous log
22:12:01 - Saved game is loaded and I can scroll around. Noted a slight
pause when swapd went to 256 - but otherwise all is well.
22:13:00 - Launched Google Chrome browser to make the system swap.
After this point, the whole system went to hell. You'll note many missing
vmstat entries up until around 22:22 when I managed to exit from the game
back to desktop via the normal means (and not getting annoyed and doing a
pkill from tty2).
As such, the system went nuts for ~9 minutes until I was able to exit the
game to stop things going nuts.
I note that with 20Gb RAM - as the system never touches swap, I can still
play the game, browse the web with the Chrome browser, read / write
email, and even watch a DVB-T broadcast in VLC without having any more
than a minor pause in the game for less than a second.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#26 |
So overall, this seems to indicate a regression between kernel 4.10.x (I'm pretty sure I tested all ok with 4.10.15?) and the newer 4.11 and 4.12 builds.
I made contact with Rik van Riel and Ying Huang (which I will attempt to add to this as a CC for comment?) - they don't believe it is a swapping issue - however Rik seems to believe that:
> > There is ZERO swap space in use.
> >
> > In other words, it is not actually swapping,
> > but thrashing through the page cache.
You may want to email the people who worked on page cache
replacement stuff recently, and the linux-mm mailing list
as well.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#27 |
(switched to email. Please respond via emailed reply-to-all, not via the
bugzilla web interface).
On Tue, 22 Aug 2017 11:17:08 +0000 <email address hidden> wrote:
> https:/
>
> Bug ID: 196729
> Summary: System becomes unresponsive when swapping - Regression
> since 4.10.x
> Product: Memory Management
> Version: 2.5
> Kernel Version: 4.11.x / 4.12.x
> Hardware: All
> OS: Linux
> Tree: Mainline
> Status: NEW
> Severity: normal
> Priority: P1
> Component: Page Allocator
> Assignee: <email address hidden>
> Reporter: <email address hidden>
> Regression: No
So it's "Regression: yes". More info at the bugzilla link.
> I have 10Gb of RAM in this system and run Fedora 26. If I launch Cities:
> Skylines with no swap space, things run well performance wise until I get an
> OOM - and it all dies - which is expected.
>
> When I turn on swap to /dev/sda2 which resides on an SSD, I get complete
> system freezes while swap is being accessed.
>
> The first swap was after loading a saved game, then launching kmail in the
> background. This caused ~500Mb to be swapped to /dev/sda2 on an SSD. The
> system froze for about 8 minutes - barely being able to move the mouse. The
> HDD LED was on constantly during the entire time.
>
> To hopefully rule out the above glibc issue, I started the game via jemalloc
> -
> but experienced even more severe freezes while swapping. I gave up waiting
> after 13 minutes of non-responsiveness - not even being able to move the
> mouse
> properly.
>
> During these hangs, I could typed into a Konsole window, and some of the
> typing took 3+ minutes to display on the screen (yay for buffers?).
>
> I have tested this with both the default vm.swappiness values, as well as the
> following:
> vm.swappiness = 1
> vm.min_free_kbytes = 32768
> vm.vfs_
>
> I noticed that when I do eventually get screen updates, all 8 cpus (4 cores /
> 2 threads) show 100% CPU usage - and kswapd is right up there in the process
> list for CPU usage. Sadly I haven't been able to capture this information
> fully yet due to said unresponsiveness.
>
> (more to come in comments & attachments)
>
> --
> You are receiving this mail because:
> You are the assignee for the bug.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#28 |
Created attachment 258067
read_vmstat.c
On Tue 22-08-17 15:55:30, Andrew Morton wrote:
>
> (switched to email. Please respond via emailed reply-to-all, not via the
> bugzilla web interface).
>
> On Tue, 22 Aug 2017 11:17:08 +0000 <email address hidden> wrote:
[...]
> Sadly I haven't been able to capture this information
> > fully yet due to said unresponsiveness.
Please try to collect /proc/vmstat in the bacground and provide the
collected data. Something like
while true
do
cp /proc/vmstat > vmstat.$(date +%s)
sleep 1s
done
If the system turns out so busy that it won't be able to fork a process
or write the output (which you will see by checking timestamps of files
and looking for holes) then you can try the attached proggy
./read_vmstat output_file timeout output_size
Note you might need to increase the mlock rlimit to lock everything into
memory.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#29 |
Created attachment 258069
8Gb-noswap.tar.gz
On Wednesday, 23 August 2017 11:38:48 PM AEST Michal Hocko wrote:
> On Tue 22-08-17 15:55:30, Andrew Morton wrote:
> > (switched to email. Please respond via emailed reply-to-all, not via the
> > bugzilla web interface).
>
> > On Tue, 22 Aug 2017 11:17:08 +0000 <email address hidden>
wrote:
> [...]
>
> > Sadly I haven't been able to capture this information
> >
> > > fully yet due to said unresponsiveness.
>
> Please try to collect /proc/vmstat in the bacground and provide the
> collected data. Something like
>
> while true
> do
> cp /proc/vmstat > vmstat.$(date +%s)
> sleep 1s
> done
>
> If the system turns out so busy that it won't be able to fork a process
> or write the output (which you will see by checking timestamps of files
> and looking for holes) then you can try the attached proggy
> ./read_vmstat output_file timeout output_size
>
> Note you might need to increase the mlock rlimit to lock everything into
> memory.
Thanks Michal,
I have upgraded PCs since I initially put together this data - however I was
able to get strange behaviour by pulling out an 8Gb RAM stick in my new system
- leaving it with only 8Gb of RAM.
All these tests are performed with Fedora 26 and kernel 4.12.8-
I have attached 3 files with output.
8Gb-noswap.tar.gz contains the output of /proc/vmstat running on 8Gb of RAM
with no swap. Under this scenario, I was expecting the OOM reaper to just kill
the game when memory allocated became too high for the amount of physical RAM.
Interestingly, you'll notice a massive hang in the output before the game is
terminated. I didn't see this before.
8Gb-swap-
RAM - but creating a file with swap on the PCIe SSD /swapfile with size 8Gb
via:
# dd if=/dev/zero of=/swapfile bs=1G count=8
# mkswap /swapfile
# swapon /swapfile
Some times (all in UTC+10):
23:58:30 - Start loading the saved game
23:59:38 - Load ok, all running fine
00:00:15 - Load Chrome
00:01:00 - Quit the game
The game seemed to run ok with no real issue - and a lot was swapped to the
swap file. I'm wondering if it was purely the speed of the PCIe SSD that
caused this appearance - as the creation of the file with dd completed at
~1.4GB/sec.
8Gb-swap-
using the entire block device as swap via:
# mkswap -f /dev/sda
# swapon /dev/sda
There are many pauses and unresponsiveness issues while this was loading -
however we eventually got there.
Some timings (all in UTC+10 again):
00:06:33 - Load the saved game
00:11:22 - Saved game loaded - somewhat responsive
00:12:00 - Load Chrome
00:13:07 - Quit the game + chrome
For the sake of information, the following is a speed test on the SSD in
question:
# dd if=/dev/zero of=/dev/sda bs=1M count=8192 conv=fsync
8192+0 records in
8192+0 records out
8589934592 bytes (8.6 GB, 8.0 GiB) copied, 44.923 s, 191 MB/s
# dd if=/dev/sda of=/dev/null bs=1M count=8192 conv=fsync
dd: fsync failed for '/dev/null': Invalid argument
8192+0 records in
8192+0 records out
8589934592 bytes (8....
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#30 |
Created attachment 258071
8Gb-swap-
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#31 |
Created attachment 258073
8Gb-swap-
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#32 |
Created attachment 258075
signature.asc
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#33 |
On Thu 24-08-17 00:30:40, Steven Haigh wrote:
> On Wednesday, 23 August 2017 11:38:48 PM AEST Michal Hocko wrote:
> > On Tue 22-08-17 15:55:30, Andrew Morton wrote:
> > > (switched to email. Please respond via emailed reply-to-all, not via the
> > > bugzilla web interface).
> >
> > > On Tue, 22 Aug 2017 11:17:08 +0000 <email address hidden>
> wrote:
> > [...]
> >
> > > Sadly I haven't been able to capture this information
> > >
> > > > fully yet due to said unresponsiveness.
> >
> > Please try to collect /proc/vmstat in the bacground and provide the
> > collected data. Something like
> >
> > while true
> > do
> > cp /proc/vmstat > vmstat.$(date +%s)
> > sleep 1s
> > done
> >
> > If the system turns out so busy that it won't be able to fork a process
> > or write the output (which you will see by checking timestamps of files
> > and looking for holes) then you can try the attached proggy
> > ./read_vmstat output_file timeout output_size
> >
> > Note you might need to increase the mlock rlimit to lock everything into
> > memory.
>
> Thanks Michal,
>
> I have upgraded PCs since I initially put together this data - however I was
> able to get strange behaviour by pulling out an 8Gb RAM stick in my new
> system
> - leaving it with only 8Gb of RAM.
>
> All these tests are performed with Fedora 26 and kernel
> 4.12.8-
>
> I have attached 3 files with output.
>
> 8Gb-noswap.tar.gz contains the output of /proc/vmstat running on 8Gb of RAM
> with no swap. Under this scenario, I was expecting the OOM reaper to just
> kill
> the game when memory allocated became too high for the amount of physical
> RAM.
> Interestingly, you'll notice a massive hang in the output before the game is
> terminated. I didn't see this before.
I have checked few gaps. E.g. vmstat.1503496391 vmstat.1503496451 which
is one minute. The most notable thing is that there are only very few
pagecache pages
[base] [diff]
nr_active_file 1641 3345
nr_inactive_file 1630 4787
So there is not much to reclaim without swap. The more important thing
is that we keep reclaiming and refaulting that memory
workingset_activate 5905591 1616391
workingset_refault 33412538 10302135
pgactivate 42279686 13219593
pgdeactivate 48175757 14833350
pgscan_kswapd 379431778 126407849
pgsteal_kswapd 49751559 13322930
so we are effectivelly trashing over the very small amount of
reclaimable memory. This is something that we cannot detect right now.
It is even questionable whether the OOM killer would be an appropriate
action. Your system has recovered and then it is always hard to decide
whether a disruptive action is more appropriate. One minute of
unresponsiveness is certainly annoying though. Your system is obviously
under provisioned to load you want to run obviously.
It is quite interesting to see that we do not really have too many
direct reclaimers during this time period
allocstall_normal 30 1
allocstall_movable 490 88
pgscan_
pgsteal_direct 24434 4069
pgscan_direct 38678 5868
> 8Gb-swap-
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#34 |
Created attachment 258079
signature.asc
On Thursday, 24 August 2017 10:41:39 PM AEST Michal Hocko wrote:
> On Thu 24-08-17 00:30:40, Steven Haigh wrote:
> > On Wednesday, 23 August 2017 11:38:48 PM AEST Michal Hocko wrote:
> > > On Tue 22-08-17 15:55:30, Andrew Morton wrote:
> > > > (switched to email. Please respond via emailed reply-to-all, not via
> > > > the
> > > > bugzilla web interface).
> > > >
> > > > On Tue, 22 Aug 2017 11:17:08 +0000 <email address hidden>
> >
> > wrote:
> > > [...]
> > >
> > > > Sadly I haven't been able to capture this information
> > > >
> > > > > fully yet due to said unresponsiveness.
> > >
> > > Please try to collect /proc/vmstat in the bacground and provide the
> > > collected data. Something like
> > >
> > > while true
> > > do
> > >
> > > cp /proc/vmstat > vmstat.$(date +%s)
> > > sleep 1s
> > >
> > > done
> > >
> > > If the system turns out so busy that it won't be able to fork a process
> > > or write the output (which you will see by checking timestamps of files
> > > and looking for holes) then you can try the attached proggy
> > > ./read_vmstat output_file timeout output_size
> > >
> > > Note you might need to increase the mlock rlimit to lock everything into
> > > memory.
> >
> > Thanks Michal,
> >
> > I have upgraded PCs since I initially put together this data - however I
> > was able to get strange behaviour by pulling out an 8Gb RAM stick in my
> > new system - leaving it with only 8Gb of RAM.
> >
> > All these tests are performed with Fedora 26 and kernel
> > 4.12.8-
> >
> > I have attached 3 files with output.
> >
> > 8Gb-noswap.tar.gz contains the output of /proc/vmstat running on 8Gb of
> > RAM
> > with no swap. Under this scenario, I was expecting the OOM reaper to just
> > kill the game when memory allocated became too high for the amount of
> > physical RAM. Interestingly, you'll notice a massive hang in the output
> > before the game is terminated. I didn't see this before.
>
> I have checked few gaps. E.g. vmstat.1503496391 vmstat.1503496451 which
> is one minute. The most notable thing is that there are only very few
> pagecache pages
> [base] [diff]
> nr_active_file 1641 3345
> nr_inactive_file 1630 4787
>
> So there is not much to reclaim without swap. The more important thing
> is that we keep reclaiming and refaulting that memory
>
> workingset_activate 5905591 1616391
> workingset_refault 33412538 10302135
> pgactivate 42279686 13219593
> pgdeactivate 48175757 14833350
>
> pgscan_kswapd 379431778 126407849
> pgsteal_kswapd 49751559 13322930
>
> so we are effectivelly trashing over the very small amount of
> reclaimable memory. This is something that we cannot detect right now.
> It is even questionable whether the OOM killer would be an appropriate
> action. Your system has recovered and then it is always hard to decide
> whether a disruptive action is more appropriate. One minute of
> unresponsiveness is certainly annoying though. Your system is obviously
> under provisioned to loa...
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#35 |
Compared with
a) vmstat-
b) vmstat-10Gb.log (NOT OK - System Unresponsive)
The si/so is low in both files. And si/so in a) is higher than that of b), so the problem may be we swap less than before?
The bi is kept high in b). I guess we encountered thrashing for file pages.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#36 |
To give this a bit of a nudge, I've been seeing reports of others having similar issues. See:
https:/
Also lodged on the RH BZ a while ago:
https:/
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#37 |
Hi,
I’ve experienced what I believe is the same problem. The problem has gone away completely for me after I bumped vm.min_free_kbytes way up to 393216.
As soon as the system ran out of physical memory, the system would freeze for at least 2 minutes and often up to 45 minutes. GNOME desktop would stop. I could move the mouse cursor, and ping the system from a remote computer; but not connect over SSH or do anything other than wave the mouse about. The system clock on the top of GNOME would stop updating for 45 minutes. (Maaaybe it would move forward 1 minute after 20 minutes and still be 19 minutes out of sync.)
I've been having this issue for years on multiple different computer configurations with 8+ GiB of memory and large SWAP partitions. I never saw more than maybe 5 MiB in use on the SWAP partition. After tuning min_free_kbytes, the SWAP partition is now being used properly and the system only does the occasional (and expected) 1 second stutter when running low on physical memory.
I also run Fedora and have kept up with the latest stable release.
Aside 1: The issue would persist with SWAPOFF, just like Steven Haigh describes.
Aside 2: The problem happen much more frequently when I used BtrFS. After switching to XFS, this happen less frequently (weekly instead of daily).
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#38 |
Please refer also to this bug report. It is the same problem and has existed for eleven (11!) years if one can believe that.
https:/
I personally experience this on two 4GB laptops running live versions (no swap) of Debian 8.6 - 9.4, Fedora 25 - 28, Ubuntu, with a myriad of shells; Gnome, Mate, Cinnamon, KDE, etc. (One laptop I expanded to 8GB now but it doesn't matter- it just take a little longer to freeze the system.)
Various browsers from Firefox52 to current Developers 60, Chrome and Chromium
Certain combos eat up memory faster (Gnome has a memory leak for example in which it consumers memory for every window drawn and NEVER relinquishes that memory, without a restart to gnome-shell), new Firefox or Chrom* vs older ESR versions (54 and under) of FF eat up memory much more quickly.
Under the best combo/circumsta
In rare instances, some 10's of minutes later or even hours (4,8 12) later, the system *might* finally respond to the request to drop to the console. Keystrokes to issue the kill command can take minutes per key, but if successful, I've seen the load reported after the kill as high as 75.
Truly amazing.
It's difficult to fathom this critical a bug in memory management has gone un-addressed/
Also, on non-"live" installs, with swap configured, one can watch the hard drive light come on and remain solid to the same effect. Power cycle time. I've read from others, that they've determined swap isn't even really being used, so not sure what the "read" thrashing going on is (and it must be read thrashing because on Live versions there's no swap and the USB drive light is steady active also).
I just run Linux to not run Windows. Basic browsing, text document editing, file management, a few cli's and an instant messaging program typically opened simultaneously. Nothing computationally heavy but memory intensive (at least for the web browser) for sure.
STILL- difficult to believe the OS cannot handle this situation with some sort of message, or killing a window/throwing an error about an opened Firefox tab or something-- rather, it simply fills up the memory (I watch on gnome-system-
I really don't know technically how the Out Of Memory killer works/is supposed to work, but it sure isn't doing anything here.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#39 |
I experience this too. Ive tested using Kernel version Kernel 4.17 rc8, Kernel 4.16.8, Kernel 4.12.8, and Kernel 4.14 across Manjaro Linux, Ubuntu Linux, Opensuse leap 15 and Fedora 28.
Steps to trigger:
-Open firefox with many tabs, or any other high memory usage program
-Wait a second
-System freezes. Sometimes the only fix is a hard reboot
Other findings:
-I notice really high cpu load averages if the system unfreezes
-If the system is not frozen, it is highly unresponsive on high memory usage when swapping
-Hard drive indicator light stays solidly on when system is frozen (excessive hard disk use)
-The reason the system freezes is because it is swapping
Tested on a
Intel i5 520m with 4gb ram/ 4gb swap (Lenovo t410)
Intel E6400 with 3gb ram/ 3gb swap
This bug is really hard to deal with because it usually requires a hard restart. Please fix ASAP if possible
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#40 |
My fedora report here : https:/
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#41 |
If someone gives me some debugging instructions, ill gladly follow them
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#42 |
https:/
https:/
Over 5 (Non bug reporting) websites report this bug, with recent posts up to 2018. This memory issue is FATAL to people who rely on linux for gaming/working. Please don't let this issue go unnoticed for 11 more years. It is critical to system function that memory management is good. The issue actually doesn't only occur when swapping. Swapping however greatly aggravates this bug
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#43 |
@SlayerProof32 #19
> -The reason the system freezes is because it is swapping
I don't think this is the case at least not in the traditional sense. if you read my comment above yours, I'm running Live versions of Linux where no swap is configured.
I suppose it could be swapping portions of code into and out of regular memory, but there's no hard disk writes. On my USB sticks, I note the light comes on steady when the system freezes. It's reading (something).
I also noted the bug doesn't exhibit in 32-bit systems (I'd read previously somewhere).
Eleven years.
Shocking.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#44 |
@lou #20
I have confirmed that swapping is not the issue and you are correct.
On my linux system, I tried swapoff --a and to no surprise, it crashed on high memory usage still, and there was still excessive drive usage. Thanks for the insight
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#45 |
https:/
I filed another report.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#46 |
I'm just a user, haunted by similar issues.
1. If you have zswap compiled and enabled, try to disable it.
ZSwap uses some RAM in order to compress pages and hide them there. e.g. reserves 500MB with the idea it could put 1000MB swap in there.
2. I think that there is something very broken in "Transparent HugePage Support" or THP, especially at kernel-4.15 and later. A bug in 4.16 memory compaction led to very noticeable swap pressure. The issue has been fixed, but disabling THP also was able to workaround it.
If you have major issues, try to build a kernel with that option disabled and see if the problem remains as severe.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#47 |
What happens for me at least with basically any system I use, is that Linux runs out of memory first, then SLAMS my swap disk with writes, which I believe overloads my system’s I/O susbsystem, and caused a complete system freeze. I have swapiness set to 60.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#48 |
Update:
I'd commented in detail about this bug (comment 18.23 above). I run the live versions of Linux on a 4GB Core-i5 laptop (and another 4GB pentium laptop also.)
Just wanted to add:
I've added 4Gb of RAM to the Core-i5 laptop for 8Gb total now. This obviously helps a lot. Now some notes for 8GB RAM instances.
With Fedora 28, the system will still cease up with maybe 2 dozen (or less depending on what's happening (video, etc) ) FF tabs opened/active.
I came back here to note that, I'm currently using a Live Debian Stretch (9.5).
There are obviously significant differences in the way these variants of Linux manage memory.
Why?
Because under the same system conditions (Gnome, same s/w programs installed and/or running), I can open WAY more tabs in FF on Debian; open more simultaneous programs, without fear of a sudden system heart-attack.
In fact, it is much harder for me to cause the system freeze in Debian, even with approaching 50 tabs opened in FF developer 63...
I understand there are underlying Fedora vs Debian system differences like: systemd vs init, and Wayland vs Xorg, Gnome versions (3.28.1 vs 3.22.5) and kernel revisions (4.16.3-
Anyway, I just thought it's another data point to add to the mystery.
I still have to keep resource monitor opened even in Stretch, just in case, but I only crashed Stretch once over the past 3 months or so when I was in the 80's (mem % used) and let a video play for 2 hrs without checking up.
Normally anyway, that percentage isn't rising above the 70's in my typical "working" environment.
Finally, I'd like to mention to those asking for logs, etc., for this issue, realize that WHEN this issue occurs, it *is* essentially a heart-attack for the system. There is no recourse, and no way to gather logs. EVERYTHING ceases up- usually never to come back. A hard power-cycle is the only recourse, and NO logs which would shed light on the issue are written. EVERYTHING stops- including log writing.
This is the reality.
I *do* have a few logs from and old (non-live) Jesse 8.7.1 install-- for a few times when, the system did revive, after hours-- and there's nothing in there that would shed light on the issue. The few entries in the log that I've researched pointed to no other instances/causes of this same issue.
It would be nice after 11 or 12 years of this issue, if someone higher up and more knowledgeable in the development "food chain" would would simply replicate the issue, it's not really that hard to do so at all.
It honestly is a show-stopper.
Ciao.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#49 |
These kind of reports are all over the web. Please someone who knows how memory management works, fix this behavior
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#50 |
I can confirm this regression. My system does not freeze with 4.9.140 and is overall totally usable on high memory usage and swap (right now 7.5 out of 8 GB RAM is used and 2 GB is swapped), but it becomes unresponsive with 4.19.10 when memory consumption is close to my RAM limit of 8 GB.
My system is:
Lenovo Thinkpad X220 laptop, Intel Core i7-2640M CPU (Sandy Bridge), Intel HD 3000 GPU, 8GB RAM, 8GB SWAP, SSD, UEFI mode.
Fedora 29, kernels 4.9.140 (self-built), 4.19.10-
I'm using KDE Plasma 5.
You can easily reproduce this issue with the following steps:
1. Add "mem=2G" to the kernel command line. If you use GRUB, press "e" button in the bootloader menu and append "mem=2G" to the "linux" line. Your system would be limited to 2 GB RAM.
2. Boot the system (press F10 in bootloader menu after editing)
3. Run several heavy applications: web browser, word processor, GIMP. Open gmail web interface in browser, it's also heavy.
Expected result:
The system (at least the GUI, including mouse cursor) does not freeze/hang, but properly utilizes swap, what could be seen with 4.9.140 kernel (before regression). System is usable, all applications are still accessible, the music plays without hiccups, you can continue doing your work.
Actual result:
The system (at least the GUI, including mouse cursor) freezes/hangs with 4.19.10 kernel, with a very little chance to recover by itself. Disk activity LED is constantly lid. System is unusable.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#51 |
This issue appears to slightly improve in 4.20. Some of the issue is that things like klogin are getting swapped, as well as other parts of the GUI. We need a way to tell the kernel what to swap out first (like Firefox tabs, or open programs). The kernel should also start swapping at say 65% memory usage very slowly, and then increase swapping as memory fills, not slamming swap at 95% usage. We also should make sure the OOM killer is doing its job, instead of waiting for alt SysRq f. As ram frees, swap should be unloaded immediately, slowly like MacOs does.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#52 |
I know @SlayerProof32 posted that this was rectified w kernel >4.17.5 in https:/
Tested on a 3GB desktop (Core 2 Quad), running Live Ubuntu 18.10 LTS off of a flash drive (pendrivelinux.
FF 63.0. I set the download folder to a dir on the hard drive so as not to deliberately stress free RAM.
With the system showing about 1.5GB free (System Monitor), trying to d/l this 1.25GB ROM image from Mega ( https:/
I've mitigated this on the laptops because they now have 8GB of RAM. BUT-- even then, I can STILL crash those systems using Live Debian (or whatever flavor). It just takes more stressing (more open tabs, bigger d/l's, whatever) to get there, but it does.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#53 |
How to reproduce (light):
1. Open web browser
2. Run the following command:
stress --vm 1 --vm-hang 0 --vm-bytes "$(awk '/MemAvailable/ {print $2"000"}' /proc/meminfo)"
3. Navigate to https:/
Actual result:
The system is very slow, almost unresponsive. HDD LED is constantly lid.
How to reproduce (heavy):
1. Open web browser
2. Run the following command:
stress --vm 1 --vm-hang 0 --vm-bytes "$(awk '/MemAvailable/ {print $2"000"}' /proc/meminfo)"
3. Navigate to https:/
Actual result:
The system is unresponsive. HDD LED is constantly lid.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#54 |
Please pay special attention to the fact that you may need to know some knowledge and take a proper backup before trying the methods mentioned later.
I have tested and found that the kernel 4.19.32 may not have this problem,the problem in kernel 5.0.5 is obvious.
Under the kernel with this problem, I found a way, if the problem occurs, run the following command immediately under the appropriate permissions:
sync && echo 3 > /proc/sys/
This may reset some of the system's operations, which may affect the next few seconds of IO operation, but should alleviate the unresponsive situation.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#55 |
This problem is still happening as of kernel 5.0.17 (Fedora 30) - shouldn't this be reflected in "Kernel version" of this issue?
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#56 |
Those who experience the issue, try to set the following sysctl settings:
vm.swappiness=100
vm.watermark_
It greatly helps on my PC.
description: | updated |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Ubuntu Kernel Bot (ubuntu-kernel-bot) wrote : Missing required logs. | #1 |
This bug is missing log files that will aid in diagnosing the problem. While running an Ubuntu kernel (not a mainline or third-party kernel) please enter the following command in a terminal window:
apport-collect 1833281
and then change the status of the bug to 'Confirmed'.
If, due to the nature of the issue you have encountered, you are unable to run this command, please add a comment stating that fact and change the bug status to 'Confirmed'.
This change has been made by an automated script, maintained by the Ubuntu Kernel Team.
Changed in linux (Ubuntu): | |
status: | New → Incomplete |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : AlsaInfo.txt | #2 |
tags: | added: apport-collected bionic |
description: | updated |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : CRDA.txt | #3 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : CurrentDmesg.txt | #4 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : Lspci.txt | #5 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : Lsusb.txt | #6 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : ProcCpuinfo.txt | #7 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : ProcCpuinfoMinimal.txt | #8 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : ProcEnviron.txt | #9 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : ProcInterrupts.txt | #10 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : ProcModules.txt | #11 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : PulseList.txt | #12 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : UdevDb.txt | #13 |
Changed in linux (Ubuntu): | |
status: | Incomplete → Confirmed |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : Re: Complete system freeze even when swapping small amounts of memory | #14 |
A small update on this.
I tried another distribution with the same kernel version, and this does not happen... I don't really know why. 0 times out of 10 this does not happen.
Also, when this occurs, triggering the OOM with SYSRQ+F also fixes the issues, but shows that no swap memory was being used, like it couldn't swap properly...? To be correct, it swapped only 64MB in the last try I made, instead of 200MB
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Kevin Remisoski (kremisoski) wrote : | #15 |
Does this happen to you while using a web browser? This is my issue and it doesn't matter which browser or which kernel. Absolute garbage. Not sure what's happened with Ubuntu or why this is happening, but I'll try to figure out why since clearly I'm not the only person this is happening to and therefore it has nothing to do with my kernel updates.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : | #16 |
Not necessarly, I can reproduce this sometimes with other stress tests
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : | #17 |
Are you on a Ryzen platform too?
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : | #18 |
I tried on FreeBSD today, and this behavior is not present
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#57 |
I am experiencing the same on Ubuntu, when the system starts swapping even small quantities of RAM, it just locks up until I manually trigger the OOM via sysrq keys.
This does not happen on FreeBSD and I can easily swap gigabytes of memory without even a single slowdown
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : | #19 |
Update on this. Looks like it's an old upstream bug.
https:/
But now I noticed this also happens on my laptop, but it's more "recoverable" after something like 25/30 minutes.
Still, this problem is NOT present in older kernels than 4.10
summary: |
- Complete system freeze even when swapping small amounts of memory + System freeze when memory is put on SWAP in Linux >4.10.x |
Changed in linux: | |
importance: | Unknown → Medium |
status: | Unknown → Confirmed |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#58 |
(In reply to ValdikSS from comment #36)
> Those who experience the issue, try to set the following sysctl settings:
>
> vm.swappiness=100
> vm.watermark_
>
> It greatly helps on my PC.
I confirm this also works very nicely for me.
Thanks. Not it doesn't slow down at all.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#59 |
Actually, it did not solved the problem on my AMD desktop, but so far on my Intel laptop it seems fine
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Luca Osvaldo Mastromatteo (lukycrociato) wrote : | #60 |
As insane as it seems... Reverting back to EXT4 from BTRFS fixed the problem...
So for me, it is an issue when BTRFS is used as root partition.
I wasn't even using swap files
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#61 |
I switched from BTRFS to EXT4 and this did seem to solve the problem, at least for me.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#62 |
(In reply to Luca from comment #40)
> I switched from BTRFS to EXT4 and this did seem to solve the problem, at
> least for me.
Do you use swap file or swap partition? I use swap partition, so it shouldn't matter, but I indeed have btrfs for /.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#63 |
(In reply to ValdikSS from comment #41)
> (In reply to Luca from comment #40)
> > I switched from BTRFS to EXT4 and this did seem to solve the problem, at
> > least for me.
>
> Do you use swap file or swap partition? I use swap partition, so it
> shouldn't matter, but I indeed have btrfs for /.
I always used a swap partition.
That's why I was surprised when I switched back to ext4 this wasn't happening anymore.
That was also even happening when using zram only
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#64 |
I encountered the same or a similar bug on BTRFS + HDD 5400 RPM + swap on a separate partition. Unfortunately, that notebook is not mine and is far away from me, what makes it hard to make experiments, kernels 4.15 and 4.18 both did have this problem, which can be reliably reproduced by running
# stress --vm 2 --vm-bytes 2000M --vm-keep
on that notebook with 4 GB RAM.
After running stress Load Average bumps above 11.0 and the whole system freezes.
Information about its hardware: https:/
(HDD is in bad condition as you may see in smartctl by the URL above, but I still believe it's not the reason why swap is not being used properly).
Some other users complain about similar problems here: https:/
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#65 |
Did anybody try to reproduce it in a virtual environment? It would allow to bibisect the kernel automatically.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#66 |
its(In reply to Mikhail Novosyolov from comment #43)
> I encountered the same or a similar bug on BTRFS + HDD 5400 RPM + swap on a
> separate partition. Unfortunately, that notebook is not mine and is far away
> from me, what makes it hard to make experiments, kernels 4.15 and 4.18 both
> did have this problem, which can be reliably reproduced by running
>
> # stress --vm 2 --vm-bytes 2000M --vm-keep
>
> on that notebook with 4 GB RAM.
> After running stress Load Average bumps above 11.0 and the whole system
> freezes.
> Information about its hardware:
> https:/
> (HDD is in bad condition as you may see in smartctl by the URL above, but I
> still believe it's not the reason why swap is not being used properly).
>
> Some other users complain about similar problems here:
> https:/
> there also uses BTRFS. I see that here, in my case, in bug#199763 and also
> in another case of similar problem that I know there are too many people
> with BTRFS, but this may be a coincidence.
it's not a coincidence, I did a lot of tests and they were only happening on ALL my machines using a btrfs as root partition
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#67 |
Did anyone try to reproduce it on (open)SUSE, especially with their LTS kernel 4.12?
SUSE uses BTRFS by default and develops it, there is a chance that they might have caught and fixed or worked around this problem or maybe their default sheduler/kernel options/etc prevent this.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#68 |
This bug is being discussed on lkml:
https:/
I'm not going to participate there, but someone should point them to this bug and point out that everything worked fine until 4.10. Sometimes things that used to work and then got broken rate a higher priority.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#69 |
(In reply to Jim Rees from comment #47)
> This bug is being discussed on lkml:
> https:/
>
> I'm not going to participate there, but someone should point them to this
> bug and point out that everything worked fine until 4.10. Sometimes things
> that used to work and then got broken rate a higher priority.
I believe it is not relevant. They are discussing problem of memory allocation in general, that behaviour did not change much in recent kernels. But what we are discussing is a regression of using swap in kernels >= 4.10 (or >= 4.11?). In that thread topic starter suggests to swapoff, but what we are discussing is wise a versa not using an enabled swap.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#70 |
Just an idea, try reproducing with kernel patch `le9g.patch`:
```
diff --git a/mm/vmscan.c b/mm/vmscan.c
index dbdc46a84f63.
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -2445,6 +2445,13 @@ static void get_scan_
BUG();
}
+ if (NR_ACTIVE_FILE == lru) {
+ long long kib_active_
+ if (kib_active_
+ nr[lru] = 0; //don't reclaim any Active(file) (see /proc/meminfo) if they are under 256MiB
+ continue;
+ }
+ }
*lru_pages += size;
nr[lru] = scan;
}
```
see: https:/
(^ scroll a bit up for some details of what the patch does)
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#71 |
(In reply to ValdikSS from comment #36)
> Those who experience the issue, try to set the following sysctl settings:
>
> vm.swappiness=100
> vm.watermark_
>
> It greatly helps on my PC.
It did not change anything. Still only around ~15 MB are swapped while 3.5 out of 4 GB of RAM is used. This results to high load average (9-15) when loading new tabs in Chromium.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#72 |
Created attachment 284677
log of read_vmstat when filling RAM with new tabs in Chromium
(In reply to Michal Hocko from comment #8)
> Created attachment 258067 [details]
> read_vmstat.c
>
> On Tue 22-08-17 15:55:30, Andrew Morton wrote:
> >
> > (switched to email. Please respond via emailed reply-to-all, not via the
> > bugzilla web interface).
> >
> > On Tue, 22 Aug 2017 11:17:08 +0000 <email address hidden>
> wrote:
> [...]
> > Sadly I haven't been able to capture this information
> > > fully yet due to said unresponsiveness.
>
> Please try to collect /proc/vmstat in the bacground and provide the
> collected data. Something like
>
> while true
> do
> cp /proc/vmstat > vmstat.$(date +%s)
> sleep 1s
> done
>
> If the system turns out so busy that it won't be able to fork a process
> or write the output (which you will see by checking timestamps of files
> and looking for holes) then you can try the attached proggy
> ./read_vmstat output_file timeout output_size
>
> Note you might need to increase the mlock rlimit to lock everything into
> memory.
I am facing the following issue on this [1] hardware:
- when new tabs are openned in Chromium, swap (on SSD) is not used, 0K of swap is used. After about 2.5 GB out of total 4 GB RAM becomes used, about 15-50 MB of swap can be used, sometimes up to ~570 MB, but not more;
- this leads to that Load Everage bumps from normal ~0.9 to 9-15 when loading a new tab in Chromium;
- so the system in general freezes from time to time when working in the web browser and switching between tabs and/or openning new ones
I have run your program read_vmstat (./read_vmstat vmstat.log 5s) and then ran a script that openned many tabs in Chromium, a new tab each 5s; after all tabs were openned, about ~3.5 GB of RAM became used, but only about 15 MB were swapped; then I ctrl+c'ed ./read_vmstat. Collected log is attached.
Kernel was 4.15.0-58-generic in Ubuntu.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#73 |
Forgot to write, that if I ran
$ stress --vm 2 --vm-bytes 1000M --vm-keep
swap is eventually used normally
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#74 |
Recently, about since kernel 5.2.7, the issue is either gone or present to much less extent.
Right now I'm running kernel 5.2.11 and finally I can keel Firefox and VirtualBox running at the same time, with 3G+ in swap, and the system does not freeze.
Could anyone affected by this issue try newer kernels?
Mikhail, I have a spare laptop which I can setup for you for tests. Do you have time and wish to investigate this issue?
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#75 |
(In reply to ValdikSS from comment #53)
> Recently, about since kernel 5.2.7, the issue is either gone or present to
> much less extent.
> Right now I'm running kernel 5.2.11 and finally I can keel Firefox and
> VirtualBox running at the same time, with 3G+ in swap, and the system does
> not freeze.
>
> Could anyone affected by this issue try newer kernels?
First of all, did you reset all your custom sysctls to default values?
1) One user (ilfat@) reported that:
1.1) on kernels 4.19.57 and Ubuntu kernel 4.15.0-54 his system did not swap correctly: all memory was full but only a small part of the swap was used, that lead to freezes
1.2) on kernels 4.19.67 and Ubuntu kernels 4.15.0-60, 4.15.0-62 systemd swaps in general normally, but he tweaked vm.watermark_
1.3) Ilfat had exactly the same problem on both ext4 and btrfs
So, something was fixed in upstream, backported to LTS kernel 4.19 and to Ubuntu kernel. I don't know what. And that issue is 100% not in BTRFS but is another problem or another aspect of the problem.
2) I did not see much difference between Ubuntu kernels 54 and 60, 62 in what is described in comment#51. So, that mystereous fix did not help.
3) another user (anreas@) reports the same as in (2):
https:/
4) trying to overcome issues from comment#51, I built Ubuntu kernel 4.15.0 with 2 pathes:
* https:/
* https:/
and set kernel options:
# https:/
vm.watermark_
vm.unevictable_
#vm.swappiness=80
# https:/
# Disable swap read-ahead
vm.page-cluster=0
After that, I _thank_ it became a bit better (I cannot prove it by numbers, just was unbale to make the system become inresponsive, but LA remained the same), but unfortunately the main user of that notebook did not use it for some days and so right now I can't say that according to her feedback the situation has improved and the system does not microfreeze from time to time anymore. Let's wait a bit more. And still that may be a coincidence, not the result of patches and/or tweaked sysctls.
I was able to dead lock that system by openning too many tabs in Chromium, but that is not what those patches should have solved. nohang/earlymoon would have probably helped if it was used.
>
> Mikhail, I have a spare laptop which I can setup for you for tests. Do you
> have time and wish to investigate this issue?
I don't have ideas how to investigate it. And how to measure the result. Maybe PSI can tell something, I did not try to look at them. And even more, I don't understand what the problem is ;)
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#76 |
On an unrelated note(but since btrfs was thought to be a problem at some point), I've discovered that btrfs with zstd:5 (or worse zstd:15) can cause (at least) mouse cursor stuttering(like it was skipping frames), while zstd:1 doesn't(likely because of the low CPU usage during compression), regardless of how fast/many writes are happening on the SSD(2-6M/s with zstd:15, 38-50+M/s with zstd:1), apparently due to high CPU usage during the compression. (zstd unspecified means zstd:3 aka default)
Normal CPU usage by itself(eg. during compiling) doesn't cause such stuttering though. I've tested this on a Lenovo Ideapad Z575, 16G RAM, Kingston SSD SA400S37240G firmware SBFK71F1, and I've personally switched to zstd:1
ie.
```
diff --git a/fs/btrfs/zstd.c b/fs/btrfs/zstd.c
index 6b9e29d050f3.
--- a/fs/btrfs/zstd.c
+++ b/fs/btrfs/zstd.c
@@ -22,7 +22,7 @@
#define ZSTD_BTRFS_
#define ZSTD_BTRFS_
-#define ZSTD_BTRFS_
+#define ZSTD_BTRFS_
#define ZSTD_BTRFS_
/* 307s to avoid pathologically clashing with transaction commit */
#define ZSTD_BTRFS_
```
but zstd:1 in /etc/fstab should also work, unless using too old kernel that doesn't know about it (hence why I prefer using the patch anyway)
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#77 |
(In reply to howaboutsynergy from comment #55)
> On an unrelated note(but since btrfs was thought to be a problem at some
> point), I've discovered that btrfs with zstd:5 (or worse zstd:15) can cause
> (at least) mouse cursor stuttering(like it was skipping frames), while
> zstd:1 doesn't(likely because of the low CPU usage during compression),
> regardless of how fast/many writes are happening on the SSD(2-6M/s with
> zstd:15, 38-50+M/s with zstd:1), apparently due to high CPU usage during the
> compression. (zstd unspecified means zstd:3 aka default)
I tried to move Chromium cache from btrfs to tmpfs, nothing became better, seems that Load Average became even a bit bigger in peacks (regrading what is described in comment#51).
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#78 |
(In reply to Mikhail Novosyolov from comment #54)
>
> So, something was fixed in upstream, backported to LTS kernel 4.19 and to
> Ubuntu kernel. I don't know what. And that issue is 100% not in BTRFS but is
> another problem or another aspect of the problem.
>
I can suspect commit 2c012a4ad1a2cd3
https:/
It appeared in kernel 5.3 and was backported to 4.19.62 and Ubuntu kernel 4.15.0-59
(In reply to ValdikSS from comment #53)
> Recently, about since kernel 5.2.7, the issue is either gone or present to
> much less extent.
> Right now I'm running kernel 5.2.11 and finally I can keel Firefox and
> VirtualBox running at the same time, with 3G+ in swap, and the system does
> not freeze.
... and to 5.2.3
But that commit "fixes" commits from the era of kernels 3.8 and 3.9, so probably it is not the one I'm looking for.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#79 |
Same issue here, although I run Fedora 31 with KDE Plasma DE.
Swapping with 5.2.17 works fine, even when using 4GB of swap for several days. Swapping with 5.3.2 causes a complete freeze, i.e. screen freezes up, no mouse movement, no TTY access. I did not try SYSRQ keys.
System under test:
HP Elitebook 850 G4
CPU: Intel i5-7200U with embedded GPU
RAM: 4GB unbuffered, memtest OK
Disk: SSD Samsung PM961 (256GB), LVM+LUKS
Swap: swapping to file of size 20GB at path /swapfile
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Pål Bergström (palbergstrom) wrote : | #80 |
I have similar problem but more short freezes and lags. Is an update of the graphic drives a solution? For some it seems that way. If so why?
https:/
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#81 |
I have an idea why this bug is much worse with BTRFS than with EXT4: BTRFS has much bigger read/write amplification, up to 10x higher than EXT4.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#82 |
(In reply to ValdikSS from comment #58)
> I have an idea why this bug is much worse with BTRFS than with EXT4: BTRFS
> has much bigger read/write amplification, up to 10x higher than EXT4.
You mean that when e.g. Chromium browser writes its cache, it loads IO "up to 10x higher than EXT4", and, when IO is also loaded by swapping, it causes microfreezes? Are you sure that operations that cause write amplifications are not done in background?
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#83 |
(In reply to Mikhail Novosyolov from comment #59)
> You mean that when e.g. Chromium browser writes its cache, it loads IO "up
> to 10x higher than EXT4"
Yes. This is also true for read operations, not only for write.
> and, when IO is also loaded by swapping, it causes
> microfreezes?
Yes, probably
> Are you sure that operations that cause write amplifications
> are not done in background?
I'm not sure, but people on Rosa forum blamed BTRFS for this bug. I'm pretty sure it's not directly tight with BTRFS, but write and read amplification may explain why lags are more severe with this FS.
Check https:/
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#84 |
(In reply to ValdikSS from comment #60)
> I'm not sure, but people on Rosa forum blamed BTRFS for this bug. I'm pretty
> sure it's not directly tight with BTRFS, but write and read amplification
> may explain why lags are more severe with this FS.
People here and on ROSA forum blamed BTRFS for the problem that swap is being not used, but not for microfreezes...
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#85 |
I don't use btrfs, but only ext4 on an SSD. Since updating my system from kernel 5.1.7 to kernel 5.3.12 in Tumbleweed, I get regular ~1-2 second freezes (e.g. mouse pointer hangs in X11, or characters don't appear while typing in Konsole) while Blender renders and swaps out finished tiles. This didn't happen with the previous kernel.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Corben (tobias-krummen) wrote : | #86 |
I'm experiencing the same issue since I upgraded to Kernel 5.3.0 on Ubuntu 18.04 LTS via HWE stack.
This still does not happen with kernel 4.15 and iirc didn't happen with Kernel 5.0.0 (which got replaced with 5.3.0 through the HWE stack).
I see a heavy increase in the swap file usage after a freeze, the swap file is located on the internal drive of my surface pro 3. If the swapping starts, it uses up to 500 MB at once, while with kernel 4.15 swapping happens later and even then uses only some few MB.
I got 8 GB of RAM, I've set vm.swappiness to 40 instead of 60.
The freezes got a bit mitigated after I've reset the gnome settings via gnome-tweaks, and enabled trim on the swap partition, but still happen and tend to increase after a while.
I'll try to enable zram and see if that helps mitigating it even more, so it's not interrupting the workflow on this device.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Nico R (u-nico-c) wrote : | #87 |
Can confirm this on Kubuntu 19.10 on a Core i5-4310U. My RAM is fairly big enough for what I'm doing everyday, but as soon as the tiniest swapping occurs, it renders the machine quite unusable - stuttering window animations, stuttering/hanging mouse cursor, often for 1-2 minutes. As I can see, the bug is already marked as confirmed, so I just want to put this for the record.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#89 |
Christoph, your issue is different. Please fill separate bugreport.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#90 |
On HP Stream 7 Tablet (1GB RAM) there was similar regression, but it started since Linux 4.12 instead of 4.10. Last year I tried bisect and several workarounds such as autostart cleanup, sysctl tweaks, zram, etc. But in the end it's seems like Linux 5.5.8 solved issue, at least in this particular use-case (device with 1GB RAM) 5.5.8 makes swapping perform like it was with 4.11.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#91 |
I primarily test by building webkitgtk [1], and I experience the same loss of system responsiveness whether / is ext4 or Btrfs. But I do see a difference in top and iotop.
https:/
This is an extreme case of refaulting, it's out of memory and swap, and since kswapd and btrfs threads are using a lot of CPU I'm guessing the faults are a mix of anonymous pages and file pages. At this point the system is really lost which is why the UX is the same with ext4 and btrfs; but behind the scenes it does seem more is going on. There might be other workloads which aren't as extreme, thereby exposing the difference. Two possible sources of the heavy CPU for btrfs threads: decompression, and checksumming. If it's true there is near constant reclaim happening, it's not just a simple minimum 4K read but rather a 128K minimum because all Btrfs compressed files use 128K extent size; is then decompressed, and then requires reading csum tree and computing csum on the read to compare. Ordinarily this is cheap but in this situation possibly it's resulting in a lot of extra congestion, but this is the limit of my knowledge so it's just speculation.
Btrfs write amplification is a known issue (wandering trees problem). But that appears to not be the issue in this example.
It might be this problem is better dealt with by cgroupsv2 to protect certain tasks from reclaim, and thus reduce the problem on any file system. But Btrfs alone (for now) does have more sophisticated cgroupvs2 IO isolation control as well.
https:/
The upstream GNOME and KDE developers are aware of the loss of responsiveness problem and have done quite a lot of preliminary work in GNOME 3.34 with more work on the way.
https:/
You can today take advantage of this cgroupsv2 work by running resource hungry tasks as a systemd user unit in Fedora 31.
https:/
I expect in the next 6-12 months (it's a guesstimate) there will be additional work in GNOME to protect the user session or what I vaguely call the "GUI stack" from reclaim, and thus improve its responsiveness at the expense of the resource hungry process.
[1] first two lines; set -j to RAM in GiB +2 GiB; i.e. if you have 8G RAM, use -j 10; more jobs makes the problem happen faster.
https:/
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#92 |
Another resource, quite long but has a tl;dr, and reviewed by some of the cgroups/resource control folks:
https:/
I'll use somewhat technically sloppy language, but hopefully a useful metaphor: there's incidental swap and heavy swap. Incidental swap is when some file or anonymous page really isn't needed, and it's good to evict it to free up memory. But in the case of heavy swap (or even reclaim), not at all incidental, it becomes a serious performance impediment. The reality is that there are some tasks that just take gobs of memory and swap isn't a good substitute. But some swap is useful for freeing up memory for things that need to stay in memory.
I'm finding for the incidental swap need, swap-on-ZRAM is quite useful. You are exchanging an IO bound task for a memory+CPU task; also it forces such pages to be pinned into memory. So there's no free lunch. Conservative use would be a ZRAM device around 1/4 of RAM up to a max of 50% RAM. It's not bad or wrong to use more, it's just that some workloads, like the webkitgtk example, have such significant need that it'll actually turn 1/2 of memory into swap, which in effect means you have 50% less RAM for that task. It actually makes the problem worse. Really, the near term is a) build with flags that cause fewer resources to be used in the firist place, or b) build in a systemd user session and limit resources that way, or c) buy more RAM, or d) use conventional swap partition, possibly with zswap to leave the most frequent pages in a small RAM cache pool, that's big enough for the task to eventually complete and just suffer with the ensuing lack of GUI responsiveness.
Anyway, it's a bit complicated. And lots of moving parts. And probably needs more sophisticate use of perf and maybe even bpf could be useful in figuring where the various bottlenecks are.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Corben (tobias-krummen) wrote : | #88 |
It seems that zram mitigates the problem for me during a work day. As long as the system doesn't have to write swap to the disk, it seems fine. The longer I use the system, the sooner it is running out of (z)ram and swapping also to disk. As soon as this happens, I get these freezes again. I have the feeling, that the amount of swapped out data that is swapped at once is causing this. Seems like I/O is blocking to me.
I have the impression with a previous kernel, the blocks have been smaller so the system didn't block too long so it didn't really interrupt work.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Kai-Heng Feng (kaihengfeng) wrote : | #93 |
Possible duplicate of LP: #1861359.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#94 |
@lou +1
Sorry for my Google translate.
This bug is very many years old. This is *Bug 12309* !!! Why didn't anyone remember?
I'm tired of him. Was in the early 2010s for 512 MB of RAM. Was in the early 2010s for 512 MB of RAM. and 2 GB of RAM. Mid 2010s with 4GB of RAM. And for the past five years, 16 GB of RAM has not gone anywhere.
There is a funny article in Russian about this bug lurkmore.to/12309 with links to the original bug reports and "problem solutions" in each new kernel. "According to the anonymous author, this useful feature is LOVE AND CAREFULLY into a fresh kernel."
Earlier I tried to rebuild kernels according to the advice on the Internet (BFQ). Sometimes it helped a little. Some assemblies gave time to urgently close programs. But the most effective was vm.swappiness = 90 and a script with a notification when approaching this value to clean up memory (conky + dialog).
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
Kevin Wortman (kwortman) wrote : | #95 |
I am experiencing this bug on a stock Dell Inspiron 5000 (5482) with 8 GB RAM and the factory SSD. It is completely debilitating. Having 1-2 completely unpredictable, 30-minute-plus, hard freezes per day is a showstopper. I can't trust this environment for professional work, or even to take notes in gedit during a phone call.
I've been an Ubuntu user and advocate at work and home for 10+ years, but I guess I have to quit :(
I tried different flavors, and the bug manifested under all of them: Ubuntu LTS 20.04, Ubuntu 20.10, Ubuntu MATE 20.10, and Debian 10.
Fedora 33 does not suffer from this bug, so I'm using that. I get frequent "Gah! Your tab just crashed" errors in Firefox any time I have more than about 5 tabs open. I conjecture there is some difference between the Redhat-based and Debian-based kernels, that causes the Redhat-based ones to kill a tab process in the circumstance that causes the Debian-based ones to lock up. The Fedora experience is acceptable, though it's still disappointing that Linux performance is has regressed to be inadequate for basic web browsing.
Takeaway: Fedora seems to be an acceptable workaround for people experiencing this bug.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#96 |
Are there any changes?
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#97 |
Modern kernels handle SWAPPING situations much better.
Also, these days the Multi-LRU patchset should pretty much resolve the problem. It is not yet upstream, but is used in downstream kernels such as linux-zen and liquorix-kernel. There is hope it will be merged by 5.19¹
1: https:/
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#98 |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#99 |
(In reply to Konstantin Kharlamov from comment #70)
> Also, these days the Multi-LRU patchset should pretty much resolve the
> problem. It is not yet upstream, but is used in downstream kernels such as
> linux-zen and liquorix-kernel. There is hope it will be merged by 5.19¹
MLRU patches were merged long ago, should perhaps this issue be closed? Is Steven Haigh (the OP) by any chance still here?
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#100 |
I've commented here 6 years ago, having similar issues with swap trashing.
I can attest that despite web browsers managing to eat more memory than ram available, my system hasn't gone unresponsive recently.
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#101 |
Yeah - I'm still around.
These days, its become cheap enough to just have more RAM - so to be honest, I haven't seen an issue like this again in a number of years - but now I don't bother with less than 32Gb of RAM for a desktop.
Likely, this has just become obsolete now - so closing as such.
Changed in linux: | |
status: | Confirmed → Expired |
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#102 |
To future readers: if you're still seeing this, make sure you have the file `/sys/kernel/
![](/+icing/build/overlay/assets/skins/sam/images/close.gif)
|
#103 |
(In reply to Steven Haigh from comment #74)
> These days, its become cheap enough to just have more RAM - so to be honest,
> I haven't seen an issue like this again in a number of years - but now I
> don't bother with less than 32Gb of RAM for a desktop.
>
> Likely, this has just become obsolete now - so closing as such.
Sorry to object, but I'm reading this thread because the issue is still fresh enough and relevant to me. Some people are stuck with 4Gb to live everyday.
However
(In reply to Konstantin Kharlamov from comment #75)
This is helpful, like others above, the issue did not show up badly in the last few month using Mint 21.3 and kernel 6.8 (and perhaps since 5.15 or 5.19). I can not easily revert to an older one just for the test, but the machine currently reports `/sys/kernel/
This later is fair enough to mandate closure.
I have 10Gb of RAM in this system and run Fedora 26. If I launch Cities:
Skylines with no swap space, things run well performance wise until I get an
OOM - and it all dies - which is expected.
When I turn on swap to /dev/sda2 which resides on an SSD, I get complete
system freezes while swap is being accessed.
The first swap was after loading a saved game, then launching kmail in the
background. This caused ~500Mb to be swapped to /dev/sda2 on an SSD. The
system froze for about 8 minutes - barely being able to move the mouse. The
HDD LED was on constantly during the entire time.
To hopefully rule out the above glibc issue, I started the game via jemalloc -
but experienced even more severe freezes while swapping. I gave up waiting
after 13 minutes of non-responsiveness - not even being able to move the mouse
properly.
During these hangs, I could typed into a Konsole window, and some of the
typing took 3+ minutes to display on the screen (yay for buffers?).
I have tested this with both the default vm.swappiness values, as well as the cache_pressure = 60
following:
vm.swappiness = 1
vm.min_free_kbytes = 32768
vm.vfs_
I noticed that when I do eventually get screen updates, all 8 cpus (4 cores /
2 threads) show 100% CPU usage - and kswapd is right up there in the process
list for CPU usage. Sadly I haven't been able to capture this information
fully yet due to said unresponsiveness.
(more to come in comments & attachments)