280.11: 100% CPU and hung gnome-session

[iro]

Symptoms:
1. type username and password at gdm login window
2. gdm login window disappears but nothing happens
3. CPU Fan starts going at max speed

Switching to VT, and using top shows gnome-session is eating 100% CPU.
However perf top shows that all the time is spent in nvidia-driver:

   PerfTop:    1001 irqs/sec  kernel: 2.5%  exact:  0.0% [1000Hz cycles],  (all, 1 CPU)
-------------------------------------------------------------------------------

             samples  pcnt         RIP        function               DSO
             _______ _____ ________________ ______________________ _______________________

             3674.00 98.3% 0000000000000be0 _nv012tls              libnvidia-tls.so.280.11
               28.00  0.7% 00000000c101a630 read_hpet              [kernel]               
                5.00  0.1% 000000000003ee10 dso__load_sym.clone.10 perf                   
                5.00  0.1% 000000000001caa0 symbol_filter          perf

I'm using GNOME3, and it happens logging in as either "gnome-shell mode" or "fallback mode".

EDIT: I forgot to mention that 280.04 works fine.

[Аноним]

[bug report] 275.19 makes Gtk apps go into infinite loop when quitting, breaks Gnome
http://www.nvnews.net/vbulletin/showthread.php?t=164619

[jmak]

The (apparently) same bug is triggered by Clementine. Latest nVidia driver hits an infinite loop:

http://code.google.com/p/clementine-...detail?id=2113

Thread 1 (Thread 0x7fc4a51a88a0 (LWP 8135)):
#0 0x00007fc49b07aba4 in _nv012tls () from /usr/lib/tls/libnvidia-tls.so.280.11
#1 0x00007fc49e21eded in ?? () from /usr/lib/libGL.so.1
#2 0x00007fc49e1fe4a9 in ?? () from /usr/lib/libGL.so.1
#3 0x00007fc4a4fe321c in ?? () from /lib64/ld-linux-x86-64.so.2
#4 0x00007fc49e97dd82 in __run_exit_handlers (status=0, listp=0x7fc49ecc54c8, run_list_atexit=true) at exit.c:78
#5 0x00007fc49e97ddd5 in exit (status=-1630763112) at exit.c:100
#6 0x00007fc49e965eb4 in __libc_start_main (main=<value optimized out>, argc=<value optimized out>, ubp_av=<value optimized out>, init=<value optimized out>,
fini=<value optimized out>, rtld_fini=<value optimized out>, stack_end=0x7fff07ee40f8) at libc-start.c:260
#7 0x0000000000519c11 in _start ()
(gdb) disassemble 0x00007fc49b07aba4
Dump of assembler code for function _nv012tls:
0x00007fc49b07ab80 <+0>: sub $0x8,%rsp
0x00007fc49b07ab84 <+4>: callq 0x7fc49b07a990 <_nv009tls>
0x00007fc49b07ab89 <+9>: test %rax,%rax
0x00007fc49b07ab8c <+12>: je 0x7fc49b07ab9f <_nv012tls+31>
0x00007fc49b07ab8e <+14>: xchg %ax,%ax
0x00007fc49b07ab90 <+16>: mov 0x8(%rax),%edx
0x00007fc49b07ab93 <+19>: test %edx,%edx
0x00007fc49b07ab95 <+21>: jne 0x7fc49b07aba4 <_nv012tls+36>
0x00007fc49b07ab97 <+23>: mov (%rax),%rax
0x00007fc49b07ab9a <+26>: test %rax,%rax
0x00007fc49b07ab9d <+29>: jne 0x7fc49b07ab90 <_nv012tls+16>
0x00007fc49b07ab9f <+31>: add $0x8,%rsp
0x00007fc49b07aba3 <+35>: retq
=> 0x00007fc49b07aba4 <+36>: jmp 0x7fc49b07aba4 <_nv012tls+36>

[AaronP]

Thanks for reporting this. I filed bug 858349.

[iro]

From this gentoo bug I've found that compiling the cairo library without opengl support fixes the problem

[naib]

I can confirm this as well (can't even login via gdm... I enter username & passwd and the screen just goes to the gdm background while xserver eats an entire core...)

275.09.07 work well for me.

280.13 & 285.03 have this issue.

What is more worrying is with this new "long-lived" driver program (ie 275.28) which is just suppose to have bugfixes has this EXACT same issue.

Thus one of these so call bugfixes has the regression