On 9/7/05, Mulyadi Santosa <mulyadi.santosa@xxxxxxxxx> wrote: > > I think I know why kswapd doesn't get killed. recall the fact that > kswapd is kernel thread (therefore running in kernel mode) and signal > handler (i mean the one which does the usual process termination) is > only executed when a process wants to return from kernel space to user > mode. read arch/i386/entry.S and search for "ENTRY(ret_from_sys_call)" > and "signal_return:". > > So, that's why kswapd just return back to TASK_RUNNING (previously > TASK_(UN)INTERRUPTIBLE) and doing loop again and again. I went through the arch/i386/kernel/entry.S file of kernel-2.6.11 and havn't found the entry for ret_from_sys_call ......... but i found that in kernel-2.4.28 but frankly speaking I m not able to recall assembly language stuffs and need to spend some time to look what .S files in kernel does, which I will sure do :) By the way the thing which you mentioned looks valid to me ..... > > I don't know clever way to stop kswapd, so I'll let other to give > commentaries. Personally I think it is better to directly patch > mm/vmscan.c rather than creating kswapd-alike kernel thread from kernel > module. > I agree with Mulyadi means better way is to patch the kernel's kswapd (or replace that with own function) rather creating a thread which works/helps in kernel/OS management stuffs from external module .... But I m interested to know any other way of doing this .... As I don't know a good way :) -- Fawad Lateef -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
