1503533 - Upgrade to Firefox 63 made pages get stuck with loader gif

Status: RESOLVED FIXED

(Core :: Graphics, defect, P3)

Description

[Guy Sheffer]

Attachment: Screenshot_20181031_132637.png

User Agent: Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:63.0) Gecko/20100101 Firefox/63.0

Steps to reproduce:

Hey,
Upgraded Ubuntu from 18.04 to 18.10, including upgrading Firefox from 62 to 63.



Actual results:

Since then, pages randomly get stuck with a loader image attached. I can't open a developer console to see what has changed because its all blank.
This happens particularly often to pinned tabs.
Attaching the tab image


Expected results:

Pages should load correctly.

Comment 1

[Guy Sheffer]

Current workaround is to downgrade back to 62.0.3.

Comment 2

[Adrian Florinescu [:aflorinescu] [PTO until 04.23 ]]

Guy Sheffer, could you please try to open browser console and see if you can pull out additional information? (Ctrl+Shift+J)
Could you also confirm:
- is the firefox vesion you are using the cannonical one or the Mozilla one? 
- if this issue happens only to pinned tabs or it happens to normal tabs as well?


On Ubuntu 18.04 I couldn't reproduce this issue with Release  63.0.1 20181030165643, but I would assume that this issue is somehow a mix related to the profile (your current configurations) and the Ubuntu upgrade.

I am assuming that this report is on the cannonical build and not the Mozilla build. It would be of great help if you could test and reproduce this on the Mozilla build and then also try using a new profile. (https://support.mozilla.org/en-US/kb/profile-manager-create-and-remove-firefox-profiles)

Comment 3

[kit1980]

I have exactly the same problem since upgrading to 63.0 on Ubuntu 16.04.5 LTS, installed from Ubuntu repos.
The issue happens to normal tabs, both to already opened after switching and to fresh new tabs after typing an URL.
When the issue is happening, I can see 100% load for some cores from "Web Content".
Restarting the browser helps for some time.

I suspect that to reproduce this you need a lot of open tabs (I have 110 open tabs).

Comment 4

[Guy Sheffer]

Hey Adrian Florinescu [:adrian_sv],
Answers

- I am running Canonical Firefox 63, but downgraded to 62 from Mozilla's website where it does not happen.
- The issue seems to happen in other tabs as well, its just much more common on those tabs.
- It tends to happen a lot after coming back form suspend. Yesterday I worked all day fine. but when I came back from suspend this morning I had a large number of tabs stuck.
- Nothing special in the console, attaching all I see which is: [SW] on message <unavailable> messages.

Comment 5

[Guy Sheffer]

Attachment: [SW] on message <unavailable> mostly, some error that might be unrelated.

Comment 6

[Guy Sheffer]

Tested and this also happens in the Mozilla version.

Comment 7

[kristian.bendiksen]

Hi,

I am seeing the same problem on CentOS Linux release 7.5.1804 (Core) using 63.0.1 downloaded from mozilla.org. I have disabled all addons, and it still happens regularly. 10 open tabs (gmail, jira, microsoft teams, fastmail, github, gitlab, google docs). 

The machine has 24 cores and 128 gigs of memory.

Comment 8

[Adrian Florinescu [:aflorinescu] [PTO until 04.23 ]]

Can anyone who can reproduce this behavior assist with recording a performance profile? It might tell us at least what is stuck there, if not the cause. (https://perf-html.io/)

Comment 9

[Guy Sheffer]

Caught one, url was https://aersia.skie.me/#/VIP

Attaching.

IT was a pinned tab, after suspend.

I restarted the profiler after the suspend, so if anything happened before the suspend that caused it, it was not caught, I can try again regrading that.

Comment 10

[Guy Sheffer]

Will not attach because file is 85MB, so here:
http://unofficialpi.org/share/Firefox%202018-11-13%2010.35%20profile.sps.json

Comment 11

[Linus Nilsson]

I regularly get this problem. I got it just now by reloading a page where it often happens (a Facebook Messenger tab). When it happens, I must "kill" the content process that has 100% CPU (which obviously crashes the tabs that share the same content process). The offending content process does not die if I simply try to exit Firefox normally; it keeps spinning in the background until I "kill" it.

I ran strace, and also got a stack trace from gdb. It seems like it's stuck calling clock_gettime(). Not sure whether it adds anything to the profile that Guy Sheffer posted.

 $ strace -p 18750
 clock_gettime(CLOCK_MONOTONIC, {tv_sec=105673, tv_nsec=520376201}) = 0
 clock_gettime(CLOCK_MONOTONIC, {tv_sec=105673, tv_nsec=520393069}) = 0
 clock_gettime(CLOCK_MONOTONIC, {tv_sec=105673, tv_nsec=520406668}) = 0
 <thousands more of these, after which I quit strace>

 $ gdb -p 18750 /usr/bin/firefox
 <preamble et cetera removed>
 (gdb) bt
 #0  0x00007fb555643219 in syscall () at ../sysdeps/unix/sysv/linux/x86_64/syscall.S:38
 #1  0x00007fb5389e9445 in  () at /usr/lib/x86_64-linux-gnu/libGLX_nvidia.so.0
 #2  0x00007fb5389e90d3 in  () at /usr/lib/x86_64-linux-gnu/libGLX_nvidia.so.0
 #3  0x00007fb537a883fd in  () at /usr/lib/x86_64-linux-gnu/libnvidia-glcore.so.390.87
 #4  0x00007fb53797c941 in  () at /usr/lib/x86_64-linux-gnu/libnvidia-glcore.so.390.87
 #5  0x00007fb5375fb214 in  () at /usr/lib/x86_64-linux-gnu/libnvidia-glcore.so.390.87
 #6  0x00007fb54b554496 in  () at /usr/lib/firefox/libxul.so
 <19 similar lines removed>
 #26 0x00007fb54deaf754 in  () at /usr/lib/firefox/libxul.so
 #27 0x000022a5b9cfff64 in  ()
 <40 similar lines removed>
 #68 0x00007fb50e295ad8 in  ()
 #69 0x00007fb54ddc3a99 in  () at /usr/lib/firefox/libxul.so
 <6 similar lines removed>
 #75 0x00007fb54deaf754 in  () at /usr/lib/firefox/libxul.so
 #76 0x000022a5b9cfff64 in  ()
 <7 similar lines removed>
 #84 0x0000000000000000 in  ()

Comment 12

[kristian.bendiksen]

I also have a Nvidia graphics card (Quadro K2000). Driver version 390.48.

Comment 13

[Adrian Florinescu [:aflorinescu] [PTO until 04.23 ]]

I couldn't reproduce this issue at all on Ubuntu 16.04 / 18.04, but it seems to be also quite an intermittent behavior. 
From comment 4 seems like it might be a regression(works in 62/ doesn't work in 63?). Not entirely sure that comment 11 is the same bug.

I don't personally think the profile is of any help to move this further. I vaguely  recall that I seen something similar about tabs being stuck with loading gifs, but I don't think it's the same issue. Mike, any thoughts what we're looking at here or how to further proceed into investigating this?

Comment 14

[kristian.bendiksen]

I have upgraded my graphics card, and the Nvidia driver to 410.78. The symptoms still appears, but much less frequently. It used to happen 10 times a day, but after the upgrade it happens maybe once every other workday. So not unreasonable to suspect some issues with the nvidia drivers. I have not used firefox 62, but it did not happen in firefox 61.

Comment 15

[kristian.bendiksen]

Was able to trap one just now in gdb:
#0  0x00007f3d122d6ec9 in syscall () at /lib64/libc.so.6
#1  0x00007f3ce7e6d4b5 in  () at /lib64/libGLX_nvidia.so.0
#2  0x00007f3ce7e6d143 in  () at /lib64/libGLX_nvidia.so.0
#3  0x00007f3ce6e8931d in  () at /lib64/libnvidia-glcore.so.410.78
#4  0x00007f3ce6d76291 in  () at /lib64/libnvidia-glcore.so.410.78
#5  0x00007f3ce69e38d4 in  () at /lib64/libnvidia-glcore.so.410.78
#6  0x00007f3d01be0bb1 in  () at /usr/local/firefox/libxul.so
[snip]

So looks a bit like comment 11?

Comment 16

[Guy Sheffer]

Still happens in the updated 63.
I can try strace and gdb - when did you know when to connect and to which PID? (I see several PIDs).

Comment 17

[Linus Nilsson]

@Guy: I use top to see which process has 100% CPU when a tab is stuck loading.

I have installed the debug symbols for Firefox to see if I can get more information when running it in gdb. My machine hung the first time I tried though, so will see if it works better next time.

Comment 18

[Mike Conley (:mconley) (:⚙️)]

The profile posted in comment 10 is, unfortunately, not symbolicated, but it's clear from looking at the pseudostack frames and the libraries being called into that the content process is stuck talking to the graphics driver.

Putting this into Core :: Graphics to get on the graphics team's radar.

Comment 19

[kit1980]

For me, the issue now seems to be fixed in Firefox 64.0

Comment 20

[kristian.bendiksen]

I have also been using 64.0 a few days, and have not seen the problem after the upgrade.

Comment 21

[Guy Sheffer]

Also upgraded to Firefox 64.0 and it seems to solve the issue. I guess this narros it down to something introduced in 63 and removed/fixed in 64.

Comment 22

[Linus Nilsson]

I haven't had the issue since upgrading to 64.0 either.