glibc/elf/tlsdeschtab.h
Szabolcs Nagy 08325735c2 [BZ 18034][AArch64] Lazy TLSDESC relocation data race fix
Lazy TLSDESC initialization needs to be synchronized with concurrent TLS
accesses.  The TLS descriptor contains a function pointer (entry) and an
argument that is accessed from the entry function.  With lazy initialization
the first call to the entry function updates the entry and the argument to
their final value.  A final entry function must make sure that it accesses an
initialized argument, this needs synchronization on systems with weak memory
ordering otherwise the writes of the first call can be observed out of order.

There are at least two issues with the current code:

tlsdesc.c (i386, x86_64, arm, aarch64) uses volatile memory accesses on the
write side (in the initial entry function) instead of C11 atomics.

And on systems with weak memory ordering (arm, aarch64) the read side
synchronization is missing from the final entry functions (dl-tlsdesc.S).

This patch only deals with aarch64.

* Write side:

Volatile accesses were replaced with C11 relaxed atomics, and a release
store was used for the initialization of entry so the read side can
synchronize with it.

* Read side:

TLS access generated by the compiler and an entry function code is roughly

  ldr x1, [x0]    // load the entry
  blr x1          // call it

entryfunc:
  ldr x0, [x0,#8] // load the arg
  ret

Various alternatives were considered to force the ordering in the entry
function between the two loads:

(1) barrier

entryfunc:
  dmb ishld
  ldr x0, [x0,#8]

(2) address dependency (if the address of the second load depends on the
result of the first one the ordering is guaranteed):

entryfunc:
  ldr x1,[x0]
  and x1,x1,#8
  orr x1,x1,#8
  ldr x0,[x0,x1]

(3) load-acquire (ARMv8 instruction that is ordered before subsequent
loads and stores)

entryfunc:
  ldar xzr,[x0]
  ldr x0,[x0,#8]

Option (1) is the simplest but slowest (note: this runs at every TLS
access), options (2) and (3) do one extra load from [x0] (same address
loads are ordered so it happens-after the load on the call site),
option (2) clobbers x1 which is problematic because existing gcc does
not expect that, so approach (3) was chosen.

A new _dl_tlsdesc_return_lazy entry function was introduced for lazily
relocated static TLS, so non-lazy static TLS can avoid the synchronization
cost.

	[BZ #18034]
	* sysdeps/aarch64/dl-tlsdesc.h (_dl_tlsdesc_return_lazy): Declare.
	* sysdeps/aarch64/dl-tlsdesc.S (_dl_tlsdesc_return_lazy): Define.
	(_dl_tlsdesc_undefweak): Guarantee TLSDESC entry and argument load-load
	ordering using ldar.
	(_dl_tlsdesc_dynamic): Likewise.
	(_dl_tlsdesc_return_lazy): Likewise.
	* sysdeps/aarch64/tlsdesc.c (_dl_tlsdesc_resolve_rela_fixup): Use
	relaxed atomics instead of volatile and synchronize with release store.
	(_dl_tlsdesc_resolve_hold_fixup): Use relaxed atomics instead of
	volatile.
	* elf/tlsdeschtab.h (_dl_tlsdesc_resolve_early_return_p): Likewise.
2015-06-17 12:41:01 +01:00

165 lines
4.5 KiB
C

/* Hash table for TLS descriptors.
Copyright (C) 2005-2015 Free Software Foundation, Inc.
This file is part of the GNU C Library.
Contributed by Alexandre Oliva <aoliva@redhat.com>
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#ifndef TLSDESCHTAB_H
# define TLSDESCHTAB_H 1
#include <atomic.h>
# ifdef SHARED
# include <inline-hashtab.h>
inline static int
hash_tlsdesc (void *p)
{
struct tlsdesc_dynamic_arg *td = p;
/* We know all entries are for the same module, so ti_offset is the
only distinguishing entry. */
return td->tlsinfo.ti_offset;
}
inline static int
eq_tlsdesc (void *p, void *q)
{
struct tlsdesc_dynamic_arg *tdp = p, *tdq = q;
return tdp->tlsinfo.ti_offset == tdq->tlsinfo.ti_offset;
}
inline static size_t
map_generation (struct link_map *map)
{
size_t idx = map->l_tls_modid;
struct dtv_slotinfo_list *listp = GL(dl_tls_dtv_slotinfo_list);
/* Find the place in the dtv slotinfo list. */
do
{
/* Does it fit in the array of this list element? */
if (idx < listp->len)
{
/* We should never get here for a module in static TLS, so
we can assume that, if the generation count is zero, we
still haven't determined the generation count for this
module. */
if (listp->slotinfo[idx].map == map && listp->slotinfo[idx].gen)
return listp->slotinfo[idx].gen;
else
break;
}
idx -= listp->len;
listp = listp->next;
}
while (listp != NULL);
/* If we get to this point, the module still hasn't been assigned an
entry in the dtv slotinfo data structures, and it will when we're
done with relocations. At that point, the module will get a
generation number that is one past the current generation, so
return exactly that. */
return GL(dl_tls_generation) + 1;
}
void *
internal_function
_dl_make_tlsdesc_dynamic (struct link_map *map, size_t ti_offset)
{
struct hashtab *ht;
void **entry;
struct tlsdesc_dynamic_arg *td, test;
/* FIXME: We could use a per-map lock here, but is it worth it? */
__rtld_lock_lock_recursive (GL(dl_load_lock));
ht = map->l_mach.tlsdesc_table;
if (! ht)
{
ht = htab_create ();
if (! ht)
{
__rtld_lock_unlock_recursive (GL(dl_load_lock));
return 0;
}
map->l_mach.tlsdesc_table = ht;
}
test.tlsinfo.ti_module = map->l_tls_modid;
test.tlsinfo.ti_offset = ti_offset;
entry = htab_find_slot (ht, &test, 1, hash_tlsdesc, eq_tlsdesc);
if (! entry)
{
__rtld_lock_unlock_recursive (GL(dl_load_lock));
return 0;
}
if (*entry)
{
td = *entry;
__rtld_lock_unlock_recursive (GL(dl_load_lock));
return td;
}
*entry = td = malloc (sizeof (struct tlsdesc_dynamic_arg));
/* This may be higher than the map's generation, but it doesn't
matter much. Worst case, we'll have one extra DTV update per
thread. */
td->gen_count = map_generation (map);
td->tlsinfo = test.tlsinfo;
__rtld_lock_unlock_recursive (GL(dl_load_lock));
return td;
}
# endif /* SHARED */
/* The idea of the following two functions is to stop multiple threads
from attempting to resolve the same TLS descriptor without busy
waiting. Ideally, we should be able to release the lock right
after changing td->entry, and then using say a condition variable
or a futex wake to wake up any waiting threads, but let's try to
avoid introducing such dependencies. */
static int
_dl_tlsdesc_resolve_early_return_p (struct tlsdesc volatile *td, void *caller)
{
if (caller != atomic_load_relaxed (&td->entry))
return 1;
__rtld_lock_lock_recursive (GL(dl_load_lock));
if (caller != atomic_load_relaxed (&td->entry))
{
__rtld_lock_unlock_recursive (GL(dl_load_lock));
return 1;
}
atomic_store_relaxed (&td->entry, _dl_tlsdesc_resolve_hold);
return 0;
}
static void
_dl_tlsdesc_wake_up_held_fixups (void)
{
__rtld_lock_unlock_recursive (GL(dl_load_lock));
}
#endif