This is the current state of my audit of the sparc64 mm code. Some of these changes have to do with THP, but some do not. I've started to add some better checking of page table sanity, in the hopes that it will potentially help further diagnosis of any problems that still can be triggered. In particular, pmd_bad() and pud_bad() now actually do something :) If those exit_mmap() WARN_ON()'s still trigger, I'll have to code up a more sophisticated debugging patch I guess. My current idea is to have a debugging function which walks the page tables counting PTE tables (under suitable locks), and validates that mm->nr_ptes matches this count. Then we can sprinkle calls to this debugging function at various critical spots, such as fault entry and exit, and around major MM operations that potentially adjust page tables. Actually, if I implement that facility, the first iteration I'd have simply check only at free_pgtables() entry. That way we can learn quickly if we enter free_pgtables() with the counter already corrupted, or if it gets corrupted by the unmaps done by free_pgtables() itself. Anyways, please let me know how well these patches work and if they make any difference for your problems at all. THP enabled and THP disabled results are both extremely useful. Thanks! -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
