Re: [PATCH 8/8] x86/mm: Allow to have userspace mappings above 47-bits

["Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>]

On Thu, Apr 06, 2017 at 10:15:47PM +0300, Dmitry Safonov wrote:
> On 04/06/2017 09:43 PM, Dmitry Safonov wrote:
> > Hi Kirill,
> > 
> > On 04/06/2017 05:01 PM, Kirill A. Shutemov wrote:
> > > On x86, 5-level paging enables 56-bit userspace virtual address space.
> > > Not all user space is ready to handle wide addresses. It's known that
> > > at least some JIT compilers use higher bits in pointers to encode their
> > > information. It collides with valid pointers with 5-level paging and
> > > leads to crashes.
> > > 
> > > To mitigate this, we are not going to allocate virtual address space
> > > above 47-bit by default.
> > > 
> > > But userspace can ask for allocation from full address space by
> > > specifying hint address (with or without MAP_FIXED) above 47-bits.
> > > 
> > > If hint address set above 47-bit, but MAP_FIXED is not specified, we try
> > > to look for unmapped area by specified address. If it's already
> > > occupied, we look for unmapped area in *full* address space, rather than
> > > from 47-bit window.
> > 
> > Do you wish after the first over-47-bit mapping the following mmap()
> > calls return also over-47-bits if there is free space?
> > It so, you could simplify all this code by changing only mm->mmap_base
> > on the first over-47-bit mmap() call.
> > This will do simple trick.

No.

I want every allocation to explicitely opt-in large address space. It's
additional fail-safe: if a library can't handle large addresses it has
better chance to survive if its own allocation will stay within 47-bits.

> I just tried to define it like this:
> -#define DEFAULT_MAP_WINDOW     ((1UL << 47) - PAGE_SIZE)
> +#define DEFAULT_MAP_WINDOW     (test_thread_flag(TIF_ADDR32) ?         \
> +                               IA32_PAGE_OFFSET : ((1UL << 47) -
> PAGE_SIZE))
> 
> And it looks working better.

Okay, thanks. I'll send v2.

> > > +    if (addr > DEFAULT_MAP_WINDOW && !in_compat_syscall())
> > > +        info.high_limit += TASK_SIZE - DEFAULT_MAP_WINDOW;
> > 
> > Hmm, TASK_SIZE depends now on TIF_ADDR32, which is set during exec().
> > That means for ia32/x32 ELF which has TASK_SIZE < 4Gb as TIF_ADDR32
> > is set, which can do 64-bit syscalls - the subtraction will be
> > a negative..

With your proposed change to DEFAULT_MAP_WINDOW difinition it should be
okay, right?

-- 
 Kirill A. Shutemov