On 8/25/2020 12:19 PM, Dave Hansen wrote:

On 8/25/20 11:43 AM, Yu, Yu-cheng wrote:

quoted

quoted

quoted

arch_prctl(ARCH_X86_CET_MMAP_SHSTK, u64 *args)
      Allocate a new shadow stack.

      The parameter 'args' is a pointer to a user buffer.

      *args = desired size
      *(args + 1) = MAP_32BIT or MAP_POPULATE

      On returning, *args is the allocated shadow stack address.

This is hideous.  Would this be better as a new syscall?

Could you point out why this is hideous, so that I can modify the
arch_prctl?

Passing values in memory is hideous when we don't have to.  A syscall
would let you have separate arguments for size and flags and would also
let you have a nice return value instead of needing to do that in memory
too.

That is a good justification.

quoted

I think this is more arch-specific.  Even if it becomes a new syscall,
we still need to pass the same parameters.

Right, but without the copying in and out of memory.

Linux-api is already on the Cc list.  Do we need to add more people to 
get some agreements for the syscall?