Re: [PATCH] mm/madvise: allow MADV_DONTNEED to free memory that is MLOCK_ONFAULT

[Jason Baron]

On 06/08/2018 03:57 PM, Andrew Morton wrote:
> On Fri,  8 Jun 2018 14:56:52 -0400 Jason Baron <jbaron@xxxxxxxxxx> wrote:
> 
>> In order to free memory that is marked MLOCK_ONFAULT, the memory region
>> needs to be first unlocked, before calling MADV_DONTNEED. And if the region
>> is to be reused as MLOCK_ONFAULT, we require another call to mlock2() with
>> the MLOCK_ONFAULT flag.
>>
>> Let's simplify freeing memory that is set MLOCK_ONFAULT, by allowing
>> MADV_DONTNEED to work directly for memory that is set MLOCK_ONFAULT. The
>> locked memory limits, tracked by mm->locked_vm do not need to be adjusted
>> in this case, since they were charged to the entire region when
>> MLOCK_ONFAULT was initially set.
> 
> Seems useful.
> 
> Is a manpage update planned?
> 

Yes, I will add a manpage update. I sort of wanted to see first if
people thought this patch was a reasonable thing to do.

> Various updates to tools/testing/selftests/vm/* seem appropriate.
> 

Indeed, I started updating tootls/testing/selftests/vm/mlock2-tests.c
with this new interface, but then I realized that that test is failing
before I made any changes. So I will go back and sort that out, and add
additional testing for this new interface.

>> Further, I don't think allowing MADV_FREE for MLOCK_ONFAULT regions makes
>> sense, since the point of MLOCK_ONFAULT is for userspace to know when pages
>> are locked in memory and thus to know when page faults will occur.
> 
> This sounds non-backward-compatible?
> 

I was making the point of why I think allowing 'MADV_DONTNEED' for
MLOCK_ONFAULT regions makes sense, while allowing 'MADV_FREE' for
MLOCK_ONFAULT regions really does not.

Thanks,

-Jason