Re: [PATCH v7] mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()

[Miaohe Lin]

On 2022/4/8 9:56, HORIGUCHI NAOYA(堀口 直也) wrote:
> On Thu, Apr 07, 2022 at 09:38:26PM +0800, Miaohe Lin wrote:
>> On 2022/4/7 19:29, Naoya Horiguchi wrote:
> ...
>>> +int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
>>> +{
>>> +	struct page *page = pfn_to_page(pfn);
>>> +	struct page *head = compound_head(page);
>>> +	int ret = 2;	/* fallback to normal page handling */
>>> +	bool count_increased = false;
>>> +
>>> +	if (!PageHeadHuge(head))
>>> +		goto out;
>>> +
>>> +	if (flags & MF_COUNT_INCREASED) {
>>> +		ret = 1;
>>> +		count_increased = true;
>>> +	} else if (HPageFreed(head) || HPageMigratable(head)) {
>>> +		ret = get_page_unless_zero(head);
>>> +		if (ret)
>>> +			count_increased = true;
>>> +	} else {
>>> +		ret = -EBUSY;
>>> +		goto out;
>>> +	}
>>> +
>>> +	if (hwpoison_filter(page)) {
>>> +		ret = -EOPNOTSUPP;
>>> +		goto out;
>>> +	}
>>
>> Now hwpoison_filter is done without lock_page + unlock_page. Is this ok or
>> lock_page + unlock_page pair is indeed required?
> 
> Hmm, we had better call hwpoison_filter in page lock for hugepages.
> I'll move this too, thank you.
> 
>>> +
>>> +	if (TestSetPageHWPoison(head)) {
>>> +		ret = -EHWPOISON;
>>> +		goto out;
>>> +	}
>>
>> Without this patch, page refcnt is not decremented if MF_COUNT_INCREASED is set in flags
>> when PageHWPoison is already set. So I think this patch also fixes that issue. Thanks!
> 
> Good point, I even didn't notice that. And the issue still seems to exist
> for normal page's cases.  Maybe encountering "already hwpoisoned" case from
> madvise_inject_error() is rare but could happen when the first call failed
> to contain the error (which is still accessible from the calling process).

Oh, I missed normal page's issue. :) Will you fix this issue kindly or am I supposed
to fix it?

Many thanks.

> 
...