Re: [Xen-devel] xen: Can't insert balloon page into VM userspace (WAS Re: [linux-linus bisection] complete test-arm64-arm64-xl-xsm)

[Juergen Gross <jgross@suse.com>]

On 12/03/2019 20:46, David Hildenbrand wrote:
> On 12.03.19 19:23, David Hildenbrand wrote:
>> On 12.03.19 19:02, Boris Ostrovsky wrote:
>>> On 3/12/19 1:24 PM, Andrew Cooper wrote:
>>>> On 12/03/2019 17:18, David Hildenbrand wrote:
>>>>> On 12.03.19 18:14, Matthew Wilcox wrote:
>>>>>> On Tue, Mar 12, 2019 at 05:05:39PM +0000, Julien Grall wrote:
>>>>>>> On 3/12/19 3:59 PM, Julien Grall wrote:
>>>>>>>> It looks like all the arm test for linus [1] and next [2] tree
>>>>>>>> are now failing. x86 seems to be mostly ok.
>>>>>>>>
>>>>>>>> The bisector fingered the following commit:
>>>>>>>>
>>>>>>>> commit 0ee930e6cafa048c1925893d0ca89918b2814f2c
>>>>>>>> Author: Matthew Wilcox <willy@infradead.org>
>>>>>>>> Date:   Tue Mar 5 15:46:06 2019 -0800
>>>>>>>>
>>>>>>>>      mm/memory.c: prevent mapping typed pages to userspace
>>>>>>>>      Pages which use page_type must never be mapped to userspace as it would
>>>>>>>>      destroy their page type.  Add an explicit check for this instead of
>>>>>>>>      assuming that kernel drivers always get this right.
>>>>>> Oh good, it found a real problem.
>>>>>>
>>>>>>> It turns out the problem is because the balloon driver will call
>>>>>>> __SetPageOffline() on allocated page. Therefore the page has a type and
>>>>>>> vm_insert_pages will deny the insertion.
>>>>>>>
>>>>>>> My knowledge is quite limited in this area. So I am not sure how we can
>>>>>>> solve the problem.
>>>>>>>
>>>>>>> I would appreciate if someone could provide input of to fix the mapping.
>>>>>> I don't know the balloon driver, so I don't know why it was doing this,
>>>>>> but what it was doing was Wrong and has been since 2014 with:
>>>>>>
>>>>>> commit d6d86c0a7f8ddc5b38cf089222cb1d9540762dc2
>>>>>> Author: Konstantin Khlebnikov <k.khlebnikov@samsung.com>
>>>>>> Date:   Thu Oct 9 15:29:27 2014 -0700
>>>>>>
>>>>>>     mm/balloon_compaction: redesign ballooned pages management
>>>>>>
>>>>>> If ballooned pages are supposed to be mapped into userspace, you can't mark
>>>>>> them as ballooned pages using the mapcount field.
>>>>>>
>>>>> Asking myself why anybody would want to map balloon inflated pages into
>>>>> user space (this just sounds plain wrong but my understanding to what
>>>>> XEN balloon driver does might be limited), but I assume the easy fix
>>>>> would be to revert
>>>> I suspect the bug here is that the balloon driver is (ab)used for a
>>>> second purpose
>>>
>>> Yes. And its name is alloc_xenballooned_pages().
>>>
>>
>> Haven't had a look at the code yet, but would another temporary fix be
>> to clear/set PG_offline when allocating/freeing a ballooned page?
>> (assuming here that only such pages will be mapped to user space)
>>
> 
> I guess something like this could do the trick if I understood it correctly:
> 
> diff --git a/drivers/xen/balloon.c b/drivers/xen/balloon.c
> index 39b229f9e256..d37dd5bb7a8f 100644
> --- a/drivers/xen/balloon.c
> +++ b/drivers/xen/balloon.c
> @@ -604,6 +604,7 @@ int alloc_xenballooned_pages(int nr_pages, struct
> page **pages)
>         while (pgno < nr_pages) {
>                 page = balloon_retrieve(true);
>                 if (page) {
> +                       __ClearPageOffline(page);
>                         pages[pgno++] = page;
>  #ifdef CONFIG_XEN_HAVE_PVMMU
>                         /*
> @@ -645,8 +646,10 @@ void free_xenballooned_pages(int nr_pages, struct
> page **pages)
>         mutex_lock(&balloon_mutex);
> 
>         for (i = 0; i < nr_pages; i++) {
> -               if (pages[i])
> +               if (pages[i]) {
> +                       __SetPageOffline(pages[i]);
>                         balloon_append(pages[i]);
> +               }
>         }
> 
>         balloon_stats.target_unpopulated -= nr_pages;
> 
> 
> At least this way, the pages allocated (and thus eventually mapped to
> user space) would not be marked, but the other ones would remain marked
> and could be excluded by makedumptool.
> 

I think this patch should do the trick. Julien, could you give it a
try? On x86 I can't reproduce your problem easily as dom0 is PV with
plenty of unpopulated pages for grant memory not suffering from
missing "offline" bit.


Juergen