Hello, Sam from Coraid here. I have noticed something that I'm having a hard time explaining and after much wailing and gnashing of teeth I figured it's time to see if anyone else had a bright idea.
I have found a situation where I'm seeing page faults as a result of kproc work that is corresponding to an adverse performance case in our SR appliances. All the memory these kprocs touch is (presumably) in the kernel address space, though much of it is mallocz'd. FWIW, I've modified KZERO and memory.c to assign the bulk of RAM to kernel space. Perhaps this has broken an assumption somewhere. I'm monitoring faults through /dev/sysstat. Am I daft to think this is strange? Is there a good reason a kproc would page fault on kernel memory? Cheers, Sam
