On Mon, May 09, 2022 at 06:56:41PM +0800, zhenwei pi wrote: > Originally, do num_poisoned_pages_inc() in memory failure routine, > use num_poisoned_pages_dec() to rollback the number if filtered/ > cancelled. > > Suggested by Naoya, do num_poisoned_pages_inc() only in > action_result(), this make this clear and simple. > > Signed-off-by: zhenwei pi <pizhenwei@xxxxxxxxxxxxx> I found that action_result(MF_MSG_UNKNOWN) in try_memory_failure_hugetlb() does not follow the rule, so that could break the counter. I don't think this is the issue in your patch, so I'm fine with this patch. I'll submit a fix later for this, which will add hugetlb_set_page_hwpoison() in the path. That will have a bit non-trivial change because we need do this in hugeltb_lock. Anyway, thank you for this work. Acked-by: Naoya Horiguchi <naoya.horiguchi@xxxxxxx>
