Re: [PATCH] [RFC] mm: mmap: Allow mmap(MAP_STACK) to map growable stack

[Helge Deller]

On 9/12/24 03:08, Yang Shi wrote:
> On Wed, Sep 11, 2024 at 5:50 PM Helge Deller <deller@xxxxxx> wrote:
> > On 9/12/24 01:05, Liam R. Howlett wrote:
> > > * Yang Shi <shy828301@xxxxxxxxx> [240911 18:16]:
> > > > On Wed, Sep 11, 2024 at 12:49 PM Liam R. Howlett
> > > > <Liam.Howlett@xxxxxxxxxx> wrote:
> > > > > * Helge Deller <deller@xxxxxxxxxx> [240911 15:20]:
> > > > > > This is a RFC to change the behaviour of mmap(MAP_STACK) to be
> > > > > > sufficient to map memory for usage as stack on all architectures.
> > > > > > Currently MAP_STACK is a no-op on Linux, and instead MAP_GROWSDOWN
> > > > > > has to be used.
> > > > > > To clarify, here is the relevant info from the mmap() man page:
> > > > > >
> > > > > > MAP_GROWSDOWN
> > > > > >      This flag is used for stacks. It indicates to the kernel virtual
> > > > > >      memory system that the mapping should extend downward in memory.  The
> > > > > >      return address is one page lower than the memory area that is
> > > > > >      actually created in the process's virtual address space.  Touching an
> > > > > >      address in the "guard" page below the mapping will cause the mapping
> > > > > >      to grow by a page. This growth can be repeated until the mapping
> > > > > >      grows to within a page of the high end of the next lower mapping,
> > > > > >      at which point touching the "guard" page will result in a SIGSEGV
> > > > > >      signal.
> > > > > >
> > > > > > MAP_STACK (since Linux 2.6.27)
> > > > > >      Allocate the mapping at an address suitable for a process or thread
> > > > > >      stack.
> > > > > >
> > > > > >      This flag is currently a no-op on Linux. However, by employing this
> > > > > >      flag, applications can ensure that they transparently obtain support
> > > > > >      if the flag is implemented in the future. Thus, it is used in the
> > > > > >      glibc threading implementation to allow for the fact that
> > > > > >      some architectures may (later) require special treatment for
> > > > > >      stack allocations. A further reason to employ this flag is
> > > > > >      portability: MAP_STACK exists (and has an effect) on some
> > > > > >      other systems (e.g., some of the BSDs).
> > > > > >
> > > > > > The reason to suggest this change is, that on the parisc architecture the
> > > > > > stack grows upwards. As such, using solely the MAP_GROWSDOWN flag will not
> > > > > > work. Note that there exists no MAP_GROWSUP flag.
> > > > > > By changing the behaviour of MAP_STACK to mark the memory area with the
> > > > > > VM_STACK bit (which is VM_GROWSUP or VM_GROWSDOWN depending on the
> > > > > > architecture) the MAP_STACK flag does exactly what people would expect on
> > > > > > all platforms.
> > > > > >
> > > > > > This change should have no negative side-effect, as all code which
> > > > > > used mmap(MAP_GROWSDOWN | MAP_STACK) still work as before.
> > > > > >
> > > > > > Signed-off-by: Helge Deller <deller@xxxxxx>
> > > > > >
> > > > > > diff --git a/include/linux/mman.h b/include/linux/mman.h
> > > > > > index bcb201ab7a41..66bc72a0cb19 100644
> > > > > > --- a/include/linux/mman.h
> > > > > > +++ b/include/linux/mman.h
> > > > > > @@ -156,6 +156,7 @@ calc_vm_flag_bits(unsigned long flags)
> > > > > >         return _calc_vm_trans(flags, MAP_GROWSDOWN,  VM_GROWSDOWN ) |
> > > > > >                _calc_vm_trans(flags, MAP_LOCKED,     VM_LOCKED    ) |
> > > > > >                _calc_vm_trans(flags, MAP_SYNC,       VM_SYNC      ) |
> > > > > > +            _calc_vm_trans(flags, MAP_STACK,      VM_STACK     ) |
> > > > >
> > > > > Right now MAP_STACK can be used to set VM_NOHUGEPAGE, but this will
> > > > > change the user interface to create a vma that will grow.  I'm not
> > > > > entirely sure this is okay?
> > > >
> > > > AFAICT, I don't see this is a problem. Currently huge page also skips
> > > > the VMAs with VM_GROWS* flags set. See vma_is_temporary_stack().
> > > > __thp_vma_allowable_orders() returns 0 if the vma is a temporary
> > > > stack.
> > >
> > > If someone is using MAP_STACK to avoid having a huge page, they will
> > > also get a mapping that grows - which is different than what happens
> > > today.
> > >
> > > I'm not saying that's right, but someone could be abusing the existing
> > > flag and this will change the behaviour.
> >
> > Wouldn't a plain mmap() followed by madvise(MADV_NOHUGEPAGE) do exactly that?
> > Why abusing MAP_STACK for that?
>
> Different sources and reports showed having huge pages for stack
> mapping hurts performance. A lot of applications, for example, pthread
> lib, allocate stack with MAP_STACK and they don't call MADV_NOHUGEPAGE
> on stack mapping.

That's true, and my patch does not change the behaviour wrt huge pages.
Using MAP_STACK still tags the area VM_NOHUGEPAGE. See below...


> > > > > That is mmap(MAP_STACK) would set VM_NOHUGEPAGE right now, with this
> > > > > change you'd get VM_NOHUGEPAGE | VM_GROWS<something>
> > > > >
> > > > > >                _calc_vm_trans(flags, MAP_STACK,      VM_NOHUGEPAGE) |
> > > > > >                arch_calc_vm_flag_bits(flags);
> > > > > >    }