2.5.48 hangs during boot
Hello all,
2.5.48 + Bill/Martin's noearlyirq patch hangs on boot on our NUMA-Q
machines. It boots normally up to
TCP: Hash tables configured (established 524288 bind 65536)
NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
VFS: Mounted root (ext2 filesystem) readonly. Freeing unused kernel memory: 268k freed
Then it *VERY* slowly proceeds to output a few more lines before hanging completely. The lines come out one at a time, with large time delays between each line. The last bit of output I get is the enabling swap line.
The -mm1 patch fixes this problem, and I'm in the process of determining exactly what fixes it. Any input/ideas would be greatly appreciated.
Thanks!
-Matt
Quoted 6 lines · Matthew Dobson
Hello all,
2.5.48 + Bill/Martin's noearlyirq patch hangs on boot on our NUMA-Q
machines. It boots normally up to
TCP: Hash tables configured (established 524288 bind 65536)
NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
VFS: Mounted root (ext2 filesystem) readonly.
Quoted 8 lines
Freeing unused kernel memory: 268k freed
Then it *VERY* slowly proceeds to output a few more lines before hanging completely. The lines come out one at a time, with large time delays between each line. The last bit of output I get is the enabling swap line. The -mm1 patch fixes this problem, and I'm in the process of determining exactly what fixes it. Any input/ideas would be greatly appreciated. Thanks!
-Matt
get the axboe/akpm fixes for the elevator deadlock and/or an intermediate bk tree. This is an io scheduling issue.
Bill
Quoted 15 lines · William Lee Irwin III
Quoted 11 lines · Matthew Dobson
Hello all,
2.5.48 + Bill/Martin's noearlyirq patch hangs on boot on our NUMA-Q
machines. It boots normally up to
TCP: Hash tables configured (established 524288 bind 65536)
NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
VFS: Mounted root (ext2 filesystem) readonly.
Freeing unused kernel memory: 268k freed
Then it *VERY* slowly proceeds to output a few more lines before hanging completely. The lines come out one at a time, with large time delays between each line. The last bit of output I get is the enabling swap line. The -mm1 patch fixes this problem, and I'm in the process of determining exactly what fixes it. Any input/ideas would be greatly appreciated.
Thanks!
-Matt
Quoted 5 lines
get the axboe/akpm fixes for the elevator deadlock and/or an intermediate bk tree. This is an io scheduling issue.
Bill
Yep.. the axboe-scsi patch from the mm1 tree fixes our problem...
Linus, you'll make a bunch of NUMA-Q developers (and likely many other people) really happy if you add that patch to the mainline.
Cheers!
-Matt
Quoted 6 lines · Matthew Dobson
...
Quoted 4 lines
get the axboe/akpm fixes for the elevator deadlock and/or an intermediate bk tree. This is an io scheduling issue.
Quoted 7 lines
Quoted 1 line
Bill
Yep.. the axboe-scsi patch from the mm1 tree fixes our problem...
Linus, you'll make a bunch of NUMA-Q developers (and likely many other people) really happy if you add that patch to the mainline.
A different fix was merged a couple of days ago.
You can get updates from
http://www.kernel.org/pub/linux/kernel/people/dwmw2/bk-2.5/ - the "Gzipped full patch" at the top. It's updated hourly (I think). I'd die without that web page.