On Thu, 2002-09-26 at 04:50, martin f krafft wrote: > i am repeatedly seeing the term "page fault" being used in Debian in > the wrong way.
(examples?) > A page fault, despite its name, has nothing to do with > memory corruption or an invalid access. It has quite a bit to do with an invalid access. As far as the MMU is concerned, it *is* an invalid access: There is no page mapped to the address, and thus it throws a (hardware) exception called a 'page fault'. Please check your friendly CPU data book ;-) > A page fault simply occurs > when a memory access causes the memory management system to have to > fetch the requested page from swap. Not quite. Page faults can be satisfied from things besides disk as well. Example: bringing in part of an ELF fragment from the page cache. Example: first write to newly allocated (in the kernel's view) memory. Example: High-mem support. Even things like copy-on-write pages (from, e.g., fork) generate something that might be called a page fault on the first write attempt. So does an attempt to write to null. > invalid accesses to memory, like > dereferencing null pointers or writing beyond the end of protected > storage cause segmentation faults or bus errors. Writing off of allocated memory causes a page fault as well (are there microprocessors on which it doesn't?), but the kernel notices that it can't do anything about it, and kills the program.
signature.asc
Description: This is a digitally signed message part