Re: [PATCH -mm] elf: init pt_regs pointer later

[Alexey Dobriyan]

On Fri, Apr 19, 2019 at 01:08:26PM -0700, Andrew Morton wrote:
> On Fri, 19 Apr 2019 23:03:43 +0300 Alexey Dobriyan <adobriyan@xxxxxxxxx> wrote:
> 
> > Get "current_pt_regs" pointer right before usage.
> > 
> > Space savings on x86_64:
> > 
> > 	add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-180 (-180)
> > 	Function                           old     new   delta
> > 	load_elf_binary                   5806    5626    -180 !!!
> 
> -256 bytes with my setup.
> 
> > --- a/fs/binfmt_elf.c
> > +++ b/fs/binfmt_elf.c
> > @@ -704,12 +704,12 @@ static int load_elf_binary(struct linux_binprm *bprm)
> >  	unsigned long start_code, end_code, start_data, end_data;
> >  	unsigned long reloc_func_desc __maybe_unused = 0;
> >  	int executable_stack = EXSTACK_DEFAULT;
> > -	struct pt_regs *regs = current_pt_regs();
> >  	struct {
> >  		struct elfhdr elf_ex;
> >  		struct elfhdr interp_elf_ex;
> >  	} *loc;
> >  	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
> > +	struct pt_regs *regs;
> >  
> >  	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
> >  	if (!loc) {
> > @@ -1159,6 +1159,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
> >  				MAP_FIXED | MAP_PRIVATE, 0);
> >  	}
> >  
> > +	regs = current_pt_regs();
> >  #ifdef ELF_PLAT_INIT
> >  	/*
> >  	 * The ABI may specify that certain registers be set up in special
> 
> Why the heck does this make such a difference?

Good question. Looks like compiler doesn't know that "current_pt_regs" is
stable pointer (because it doesn't know ->stack isn't) even though it knows
that "current" is stable pointer. So it saves it in the very beginning and
then tries to carry it through a lot of code.

Here is what happens here:

load_elf_binary()
		...
	mov	rax,QWORD PTR gs:0x14c00
	mov	r13,QWORD PTR [rax+0x18]	r13 = current->stack
	call	kmem_cache_alloc		# first kmalloc

		[980 bytes later!]

	# let's spill that sucker because we need a register
	# for "load_bias" calculations at
	#
	#	if (interpreter) {
	#		load_bias = ELF_ET_DYN_BASE;
	#		if (current->flags & PF_RANDOMIZE)
	#			load_bias += arch_mmap_rnd();
	#		elf_flags |= elf_fixed;
	#	}
	mov	QWORD PTR [rsp+0x68],r13

If this is not _the_ root cause it is still eeeeh.

After the patch things become much simpler:

	mov	rax, QWORD PTR gs:0x14c00	# current
	mov	rdx, QWORD PTR [rax+0x18]	# current->stack
	movq	[rdx+0x3fb8], 0			# fill pt_regs
		...
	call finalize_exec