Bug 210584

I still have this problem, and I'm seeing it multiple times each day. What can I
do to get more information for you to find and fix the problem?

The problem has been reported against Debian as well:

http://www.mail-archive.com/debian-bugs-dist@lists.debian.org/msg132215.html

Upstream of this bug seems to be https://bugs.freedesktop.org/show_bug.cgi?id=3168
and there is even (untested) workaround:

> A workaround is to disable a single acceleration in xorg.conf:
> 
> Option "XaaNoScreenToScreenCopy"
> 
> With this option set everything work fine for me, albeit
> scrool is quite slow.

Thanks Matej. From https://bugs.freedesktop.org/show_bug.cgi?id=2155, I see this
comment:

> A lot of users have this problem, it's been discussed on
> http://forums.gentoo.org/viewtopic-t-198023-highlight-nvrm.html (350 messages
> and 24924 reads). As you can see in the forums it doesn't matter if you use
> firefox or konqueror, ati or nvidia, 2.4 or 2.6 kernel, etc.
>
> Right before the freeze the following message is written in the log;
>
> NVRM: Xid: 13, 0000 02009700 00002597 00001528 004a016e 00400000

I see "NVRM: Xid" mentioned in multiple places in the threads I've referenced,
but I'm not seeing that myself:

bash-3.00$ sudo grep NVRM /var/log/messages*
Password:
bash-3.00$ dmesg | grep NVRM
bash-3.00$

> there is even (untested) workaround:

I could try that, but I'd rather do something more specific to localise the
problem. Can you provide instructions for connection to X with gdb, and
obtaining a backtrace? Do you have other investigation strategies to recommend?

The comments in https://bugs.freedesktop.org/show_bug.cgi?id=3168 suggest this
is an AGP problem. The frequency with which I'm seeing it does seem to vary from
week to week, and one thing which might be different is which kernel version I
am running. I'm also running vmware much of the time - I can't say that there is
any correlation with/without vmware running.

"NVRM" is the nvidia resource manager, which is part of the 'nvidia' X driver,
not the 'nv' X driver.  

All of the URLs under "Google knows of other reports of similar sounding
problem" above are for users with the nvidia X driver, and are basically
irrelevant here.

> A workaround is to disable a single acceleration in xorg.conf:
> 
> Option "XaaNoScreenToScreenCopy"

Note that Suse do this by default for some GeForce cards. They mention problems
with the nv driver but not the nVidia drivers.
http://suse.mirrors.tds.net/pub/suse/i386/9.3/docu/RELEASE-NOTES.en.html

Right, but they mentioned that for SuSE-9.3.  That bug was fixed about a year
ago in the nv driver.

I've had another lockup - attaching with gdb doesn't tell me a lot:


0x098cc3a8 in ?? ()
(gdb) bt
#0  0x098cc3a8 in ?? ()
#1  0x095f8858 in ?? ()
#2  0x099280c8 in ?? ()
#3  0xbfffc138 in ?? ()
#4  0x09935335 in ?? ()
#5  0x095f8858 in ?? ()
#6  0x09b7d900 in ?? ()
#7  0x09f1cff8 in ?? ()
#8  0x098fc718 in ?? ()
#9  0x00000000 in ?? ()
(gdb)

I've also found another problem with the current configuration. Ctl-Alt-Fn
causes an immediate apparent lockup of X. Mouse cursor no longer moves. strace
shows that stuff is still happening:

select(256, [1 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23], NULL, NULL,
{45, 949000}) = ? ERESTARTNOHAND
(To be restarted)
--- SIGALRM (Alarm clock) @ 0 (0) ---
setitimer(ITIMER_REAL, {it_interval={0, 0}, it_value={0, 0}}, NULL) = 0
sigreturn()                             = ? (mask now [IO])
gettimeofday({1162482444, 508321}, NULL) = 0
gettimeofday({1162482444, 508373}, NULL) = 0
select(256, [1 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23], NULL, NULL,
{45, 938000}) = 1 (in [19], left
{45, 914000})
setitimer(ITIMER_REAL, {it_interval={0, 20000}, it_value={0, 20000}}, NULL) = 0
gettimeofday({1162482444, 532563}, NULL) = 0
read(19, "5\20\4\0\370\2\200\1&\0\200\1\26\0.\0;\3\5\0(\0\200\1\0"..., 4096) = 264
read(19, 0x9ffa0b8, 4096)               = -1 EAGAIN (Resource temporarily
unavailable)
gettimeofday({1162482444, 532844}, NULL) = 0
select(256, [1 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23], NULL, NULL,
{45, 914000}) = ? ERESTARTNOHAND
(To be restarted)
--- SIGALRM (Alarm clock) @ 0 (0) ---
setitimer(ITIMER_REAL, {it_interval={0, 0}, it_value={0, 0}}, NULL) = 0
sigreturn()                             = ? (mask now [IO])
gettimeofday({1162482444, 553291}, NULL) = 0
gettimeofday({1162482444, 553341}, NULL) = 0
select(256, [1 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23], NULL, NULL,
{45, 893000}) = 1 (in [19], left
{45, 868000})
setitimer(ITIMER_REAL, {it_interval={0, 20000}, it_value={0, 20000}}, NULL) = 0
gettimeofday({1162482444, 578431}, NULL) = 0
read(19, "5\20\4\0\372\2\200\1&\0\200\1\26\0.\0;\3\5\0(\0\200\1\0"..., 4096) = 264
read(19, 0x9ffa0b8, 4096)               = -1 EAGAIN (Resource temporarily
unavailable)
gettimeofday({1162482444, 578682}, NULL) = 0
select(256, [1 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23], NULL, NULL,
{45, 868000}) = ? ERESTARTNOHAND
(To be restarted)
--- SIGALRM (Alarm clock) @ 0 (0) ---
setitimer(ITIMER_REAL, {it_interval={0, 0}, it_value={0, 0}}, NULL) = 0
sigreturn()                             = ? (mask now [IO])
gettimeofday({1162482444, 599287}, NULL) = 0
gettimeofday({1162482444, 599338}, NULL) = 0
select(256, [1 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23], NULL, NULL,
{45, 847000}) = 1 (in [19], left
{45, 822000})
setitimer(ITIMER_REAL, {it_interval={0, 20000}, it_value={0, 20000}}, NULL) = 0
gettimeofday({1162482444, 624430}, NULL) = 0
read(19, "5\20\4\0\374\2\200\1&\0\200\1\26\0.\0;\3\5\0(\0\200\1\0"..., 4096) = 264
...

> I don't know whether the first SIGTERM or the first
> SIGQUIT would have worked if I'd waited long enough.

A single SIGQUIT works to kill the errant X process. SIGINT and SIGTERM do not.

(In reply to comment #7)
> Right, but they mentioned that for SuSE-9.3.  That bug was fixed about a year
> ago in the nv driver.

I'll have to take your word for that, since there is no bug reference. All I see
is the same symptoms mentioned, and the same workaround. Looks like there may
have been two nv bugs with similar symptoms, one fixed a year ago, and one still
extant. Is that what you suspect?

Do you suspect that the driver is scheduling alarms with 0 timeout?

> A workaround is to disable a single acceleration in xorg.conf:
> 
> Option "XaaNoScreenToScreenCopy"

http://www.nvnews.net/vbulletin/showthread.php?p=1610 suggests a workaround with
less drastic performance downside - 'add Option "XaaNoOffscreenPixmaps"'.

to comment 4: The official instructions for X debugging are on
http://xorg.freedesktop.org/wiki/DebuggingTheXserver
However, if the X-server crashes (not just freezes) then it dumps backtrace to
/var/log/Xorg.0.log, which would be really useful if you send to us. You may try
to send killall -SEGV X from ssh session -- that may persuade X-server to
generate backtrace.

> However, if the X-server crashes (not just freezes) then it dumps backtrace to
> /var/log/Xorg.0.log, which would be really useful if you send to us.

It doesn't crash.

I doubt that -SEGV will help - the problem I believe is lack of symbols.

I'm rebuilding from source rpm so that I have debuginfo packages available.

> I'm rebuilding from source rpm so that I have debuginfo packages available.

Hmmm, there aren't any built. Here's why, from the spec file:

# xorg-x11 6.7.0, 6.8.0, 6.8.1
%define parallel_build          0
%define verbose_build           1
# Builds X with debug symbol info.  This results in *FRECKIN* *HUGE* packages.
# Enable this if you want to debug the modular X server.  As of gdb 5.1.1, this
# requires patches to gdb to be completely useful.  I MEAN HUGE, ie 2-3 times
# the size!  Don't complain - you've been warned!  gdb for X is
# available for download from ftp://people.redhat.com/mharris/gdb-xfree86
%define DebuggableBuild         0
# If enabled, this makes the libraries contain debug info.  In the future I
# would like to either have xorg-x11 debuginfoized or build both sets of libs.
%define with_debuglibs          0
# Pass -save-temps to gcc for debugging purposes
%define with_savetemps          0

You wont get debuginfo packages from any XFree86 build or monolithic X.Org
build in RHEL or Fedora Core.  Only the modularized X.Org server shipping
in Fedora Core 5 and later has debuginfo integration and direct gdb
debugging available.

Debugging the X server on older releases is much much more of a chore.

> Running strace on the X process shows it to be in a SIGALRM loop with
> 0 timeout:

Does this indicate a kernel bug? What is generating this apparently infinite set
of alarm signals? The server certainly doesn't appear to be asking for another
alarm in its signal handler.

I've hit the same symptoms several times.  Another co-worker has hit the same
problem and I was able to grab the following data from his workstation.

strace output:

--- SIGALRM (Alarm clock) @ 0 (0) ---
rt_sigreturn(0xe)                       = 28144
--- SIGALRM (Alarm clock) @ 0 (0) ---
rt_sigreturn(0xe)                       = 28144
--- SIGALRM (Alarm clock) @ 0 (0) ---

gdb backtrace:
(gdb) bt
#0  0x00002aaaac047c49 in NVSync ()
   from /usr/lib64/xorg/modules/drivers/nv_drv.so
#1  0x00002aaaae140bb6 in XAAGetFallbackOps ()
   from /usr/lib64/xorg/modules/libxaa.so
#2  0x0000000000512b04 in miInitializeCompositeWrapper ()
#3  0x000000000050e683 in DamageDamageRegion ()
#4  0x00000000004480ae in ProcCopyArea ()
#5  0x0000000000449c9a in Dispatch ()
#6  0x00000000004325d5 in main ()

xorg.conf:

# Xorg configuration created by system-config-display

Section "ServerLayout"
        Identifier     "single head configuration"
        Screen      0  "Screen0" 0 0
        InputDevice    "Keyboard0" "CoreKeyboard"
EndSection

Section "InputDevice"
        Identifier  "Keyboard0"
        Driver      "kbd"
        Option      "XkbModel" "pc105"
        Option      "XkbLayout" "us"
EndSection

Section "Device"
        Identifier  "Videocard0"
        Driver      "nv"
EndSection

Section "Screen"
        Identifier "Screen0"
        Device     "Videocard0"
        DefaultDepth     24
        SubSection "Display"
                Viewport   0 0
                Depth     24
        EndSubSection
EndSection

I should also add that we've hit this running on RHEL 5.0.

xorg-x11-server-Xorg-1.1.1-48.2.el5
xorg-x11-drv-nv-1.2.0-4.fc6

(In reply to comment #16)
> > Running strace on the X process shows it to be in a SIGALRM loop with
> > 0 timeout:
> 
> Does this indicate a kernel bug? What is generating this apparently infinite
> set of alarm signals? The server certainly doesn't appear to be asking
> for another alarm in its signal handler.

Not only that, but no other process is being scheduled to run often enough to
request an ALARM signal be sent to the X server.

Someone needs to start looking at what is happening in the kernel (unless strace
is lying about what is happening in the X server process).

I have no idea, why I haven't asked for /var/log/Xorg.*.log files being attached
as separate uncompressed attachments to this bug. So I am asking for that now.
Reporter, could you please attach these logs?

Thank you

Created attachment 159345 [details]
old Xorg log

Created attachment 159346 [details]
Xorg log

Actually, I think this duplicate of the famous nvsync bug. Sorry, there is not
much we can do about it (it is widely known among all Xorg distributions and
nobody found much solution for it yet after couple of years it goes around).

*** This bug has been marked as a duplicate of 219377 ***