Re: [RFC PATCH] mm: Introduce new MADV_NOMOVABLE behavior

[Baolin Wang <baolin.wang@linux.alibaba.com>]

On 10/17/2022 4:41 PM, David Hildenbrand wrote:
> On 17.10.22 09:32, Baolin Wang wrote:
>> When creating a virtual machine, we will use memfd_create() to get
>> a file descriptor which can be used to create share memory mappings
>> using the mmap function, meanwhile the mmap() will set the MAP_POPULATE
>> flag to allocate physical pages for the virtual machine.
>>
>> When allocating physical pages for the guest, the host can fallback to
>> allocate some CMA pages for the guest when over half of the zone's free
>> memory is in the CMA area.
>>
>> In guest os, when the application wants to do some data transaction with
>> DMA, our QEMU will call VFIO_IOMMU_MAP_DMA ioctl to do longterm-pin and
>> create IOMMU mappings for the DMA pages. However, when calling
>> VFIO_IOMMU_MAP_DMA ioctl to pin the physical pages, we found it will be
>> failed to longterm-pin sometimes.
>>
>> After some invetigation, we found the pages used to do DMA mapping can
>> contain some CMA pages, and these CMA pages will cause a possible
>> failure of the longterm-pin, due to failed to migrate the CMA pages.
>> The reason of migration failure may be temporary reference count or
>> memory allocation failure. So that will cause the VFIO_IOMMU_MAP_DMA
>> ioctl returns error, which makes the application failed to start.
>>
>> To fix this issue, this patch introduces a new madvise behavior, named
>> as MADV_NOMOVABLE, to avoid allocating CMA pages and movable pages if
>> the users want to do longterm-pin, which can remove the possible failure
>> of movable or CMA pages migration.
> 
> Sorry to say, but that sounds like a hack to work around a kernel 
> implementation detail (how often we retry to migrate pages).

IMO, in our case one migration failure will make our application failed 
to start, which is not a trival problem. So mitigate the failure of 
migration can be important in this case.

> If there are CMA/ZONE_MOVABLE issue, please fix them instead, and avoid 
> leaking these details to user space.

Now we can not forbid the fallback to CMA allocation if there are enough 
free CMA in the zone, right? So adding a hint to help to diable 
ALLOC_CMA flag seems reasonable?

For CMA/ZONE_MOVABLE details, yes, not suitable to leak to user space. 
so how about rename the madvise as MADV_PINNABLE, which means we will do 
longterm-pin after allocation, and no CMA/ZONE_MOVABLE pages will be 
allocated.

Or do you have any good idea? Thanks.


> ALSO, with MAP_POPULATE as described by you this madvise flag doesn't 
> make too much sense, because it will gets et after all memory already 
> was allocated ...

This is not a problem I think, we can change to use MADV_POPULATE_XXX to 
preallocate the physical pages after MADV_NOMOVABLE madvise.