Re: linux-next: Tree for Nov 7

[Michael Ellerman <mpe@ellerman.id.au>]

Michal Hocko <mhocko@kernel.org> writes:

> On Mon 13-11-17 13:00:57, Michal Hocko wrote:
> [...]
>> Yes, I have mentioned that in the previous email but the amount of code
>> would be even larger. Basically every arch which reimplements
>> arch_get_unmapped_area would have to special case new MAP_FIXED flag to
>> do vma lookup.
>
> It turned out that this might be much more easier than I thought after
> all. It seems we can really handle that in the common code.

Ah nice. I should have read this before replying to your previous mail.

> This would mean that we are exposing a new functionality to the userspace though.
> Myabe this would be useful on its own though.

Yes I think it would. At least jemalloc seems like it could use it:

  https://github.com/jemalloc/jemalloc/blob/9f455e2786685b443201c33119765c8093461174/src/pages.c#L65

And I have memories of some JIT code I read once which did a loop of
mmap()s or something to try and get allocations below 4GB or some other
limit - but I can't remember now what it was.

> Just a quick draft (not
> even compile tested) whether this makes sense in general. I would be
> worried about unexpected behavior when somebody set other bit without a
> good reason and we might fail with ENOMEM for such a call now.
>
> Elf loader would then use MAP_FIXED_SAFE rather than MAP_FIXED.
> ---
> diff --git a/arch/alpha/include/uapi/asm/mman.h b/arch/alpha/include/uapi/asm/mman.h
> index 3b26cc62dadb..d021c21f9b01 100644
> --- a/arch/alpha/include/uapi/asm/mman.h
> +++ b/arch/alpha/include/uapi/asm/mman.h
> @@ -31,6 +31,9 @@
>  #define MAP_STACK	0x80000		/* give out an address that is best suited for process/thread stacks */
>  #define MAP_HUGETLB	0x100000	/* create a huge page mapping */
>  
> +#define MAP_KEEP_MAPPING 0x2000000
> +#define MAP_FIXED_SAFE	MAP_FIXED|MAP_KEEP_MAPPING /* enforce MAP_FIXED without clobbering an existing mapping */


So bike-shedding a bit, but I think "SAFE" is too vague a name.

Perhaps MAP_NO_CLOBBER - which has the single semantic of "do not
clobber any existing mappings".

It would be a flag on its own, so you could pass it with or without
MAP_FIXED, but it would only change the behaviour when MAP_FIXED is
specified also.

cheers