Created attachment 379938 [details] 3.14.4 kernel config Hi, starting with kernel 3.15.0 i have the following, reproducable problem: When using something IO intensive (e.g. Installing a OS inside a VM) some processes start to block forever. Firefox, thunderbird, the VM itself and also some kernel threads (not necessarely related to IO activity). There is nothing in dmesg and a smarttest didn't find any faults. I've tested downloading a file inside a Windows 7 VM with Kernels 3.15.2, 3.15.0 and 3.14.4. After some minutes i can see blocked threads (ps aux D flag) on the 3.15.0 and the 3.15.2 kernel. A sync call never returns and the threads stay blocked forever. This does not happen on the 3.14.4 kernel. I'm using a Thinkpad x220 with a Samsung Evo SSD I've attached the 3.14.4 config. (I haven't found a a way to attach more than one file...) If further information is needed, i'd be happy to provied it. Regards, Robert
Created attachment 379940 [details] Kernel 3.15.1 config
echo w > /proc/sysrq-trigger; dmesg
Created attachment 379942 [details] dmesg output of blocked threads
I also have this problem, but it's been with at the very least the 3.12 kernels and up. Whenever I am writing heavily to a drive (like copying a file), it will be so slow to read from it that most programs being used from it becomes unusably slow, and you can't play things like video from the drive fast enough. Doesn't matter what disk/SSD I use, happens with all of them. I am using the CFQ I/O scheduler, although others like Deadline or no-op have the same problem. Is there some kind of setting I can tweak so it will actually fairly give disk usage? It doesn't happen with default Fedora. This has been a pretty bad problem for me. Thanks for any help.
The issue still exists in kernel 3.16.0. Also with 3.16.0 no high IO usage is needed to trigger the bug. Regards, Robert
Please take this bug upstream to http://bugzilla.kernel.org and post the bug # back here so we can track it.
