Re: [PATCH v1 1/2] mm/userfaultfd: don't place zeropages when zeropages are disallowed

[David Hildenbrand <david@redhat.com>]

On 21.03.24 23:46, Peter Xu wrote:
> On Thu, Mar 21, 2024 at 11:29:45PM +0100, David Hildenbrand wrote:
>> On 21.03.24 23:20, Peter Xu wrote:
>>> On Thu, Mar 21, 2024 at 10:59:53PM +0100, David Hildenbrand wrote:
>>>> s390x must disable shared zeropages for processes running VMs, because
>>>> the VMs could end up making use of "storage keys" or protected
>>>> virtualization, which are incompatible with shared zeropages.
>>>>
>>>> Yet, with userfaultfd it is possible to insert shared zeropages into
>>>> such processes. Let's fallback to simply allocating a fresh zeroed
>>>> anonymous folio and insert that instead.
>>>>
>>>> mm_forbids_zeropage() was introduced in commit 593befa6ab74 ("mm: introduce
>>>> mm_forbids_zeropage function"), briefly before userfaultfd went
>>>> upstream.
>>>>
>>>> Note that we don't want to fail the UFFDIO_ZEROPAGE request like we do
>>>> for hugetlb, it would be rather unexpected. Further, we also
>>>> cannot really indicated "not supported" to user space ahead of time: it
>>>> could be that the MM disallows zeropages after userfaultfd was already
>>>> registered.
>>>>
>>>> Fixes: c1a4de99fada ("userfaultfd: mcopy_atomic|mfill_zeropage: UFFDIO_COPY|UFFDIO_ZEROPAGE preparation")
>>>> Signed-off-by: David Hildenbrand <david@redhat.com>
>>>
>>> Reviewed-by: Peter Xu <peterx@redhat.com>
>>>
>>> Still, a few comments below.
>>>
>>>> ---
>>>>    mm/userfaultfd.c | 35 +++++++++++++++++++++++++++++++++++
>>>>    1 file changed, 35 insertions(+)
>>>>
>>>> diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
>>>> index 712160cd41eca..1d1061ccd1dea 100644
>>>> --- a/mm/userfaultfd.c
>>>> +++ b/mm/userfaultfd.c
>>>> @@ -316,6 +316,38 @@ static int mfill_atomic_pte_copy(pmd_t *dst_pmd,
>>>>    	goto out;
>>>>    }
>>>> +static int mfill_atomic_pte_zeroed_folio(pmd_t *dst_pmd,
>>>> +		 struct vm_area_struct *dst_vma, unsigned long dst_addr)
>>>> +{
>>>> +	struct folio *folio;
>>>> +	int ret;
>>>
>>> nitpick: we can set -ENOMEM here, then
>>>
>>>> +
>>>> +	folio = vma_alloc_zeroed_movable_folio(dst_vma, dst_addr);
>>>> +	if (!folio)
>>>> +		return -ENOMEM;
>>>
>>> return ret;
>>>
>>>> +
>>>> +	ret = -ENOMEM;
>>>
>>> drop.
>>
>> Sure!
>>
>>>
>>>> +	if (mem_cgroup_charge(folio, dst_vma->vm_mm, GFP_KERNEL))
>>>> +		goto out_put;
>>>> +
>>>> +	/*
>>>> +	 * The memory barrier inside __folio_mark_uptodate makes sure that
>>>> +	 * preceding stores to the page contents become visible before
>>>> +	 * the set_pte_at() write.
>>>> +	 */
>>>
>>> This comment doesn't apply.  We can drop it.
>>>
>>
>> I thought the same until I spotted that comment (where uffd originally
>> copied this from I strongly assume) in do_anonymous_page().
>>
>> "Preceding stores" here are: zeroing out the memory.
> 
> Ah.. that's okay then.
> 
> Considering that userfault used to be pretty cautious on such ordering, as
> its specialty to involve many user updates on the page, would you mind we
> mention those details out?
> 
> 	/*
> 	 * __folio_mark_uptodate contains the memory barrier to make sure
>           * the page updates to the zero page will be visible before
> 	 * installing the pgtable entries.  See do_anonymous_page().
> 	 */
> 
> Or anything better than my wordings.

Sure, I'd slightly reword it. The following on top:

diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index 1d1061ccd1dea..9d385696fb891 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c
@@ -320,20 +320,19 @@ static int mfill_atomic_pte_zeroed_folio(pmd_t *dst_pmd,
  		 struct vm_area_struct *dst_vma, unsigned long dst_addr)
  {
  	struct folio *folio;
-	int ret;
+	int ret = -ENOMEM;
  
  	folio = vma_alloc_zeroed_movable_folio(dst_vma, dst_addr);
  	if (!folio)
-		return -ENOMEM;
+		return ret;
  
-	ret = -ENOMEM;
  	if (mem_cgroup_charge(folio, dst_vma->vm_mm, GFP_KERNEL))
  		goto out_put;
  
  	/*
  	 * The memory barrier inside __folio_mark_uptodate makes sure that
-	 * preceding stores to the page contents become visible before
-	 * the set_pte_at() write.
+	 * zeroing out the folio become visible before mapping the page
+	 * using set_pte_at(). See do_anonymous_page().
  	 */
  	__folio_mark_uptodate(folio);
  

Thanks!

-- 
Cheers,

David / dhildenb