01:07 <agentzh> fche: is it a known issue that stap modules loading/unloading for many times might cause memory fragmentation which makes further ko loading fail with "Couldn't insert module 'xxx.ko': Operation not permitted"?

01:08 <agentzh> "echo 1 > /proc/sys/vm/compact_memory" always fixes this. but it is quite expensive and scary even.

01:20 <agentzh> does the kernel require contiguous physical memory address space for kernel loading?

01:20 <agentzh> it seems to me like that.

01:20 <agentzh> *for kernel module loading

01:55 <fche> agentzh, was not aware of that

01:55 <fche> well, some fragmentation is a problem, no matter the type of allocator

01:55 <fche> stap modules do release every bit of memory they allocated, so it should not accumulate damage

03:21 <agentzh> fche: right, it's not a leak, but failures due to fragmentations in long-running boxes.

03:46 <fche> hm too bad that reported rc from the fragmented case is just EPERM. staprun could attempt that /compact trigger and retry the load

03:46 <fche> 'expensive and scary' - tell me more

03:46 <fche> (or we could do it unconditionally at a module unload)