glibc/sysdeps/aarch64/dl-tlsdesc.S

367 lines
10 KiB
ArmAsm
Raw Normal View History

2012-11-09 17:53:51 +00:00
/* Thread-local storage handling in the ELF dynamic linker.
AArch64 version.
Copyright (C) 2011-2016 Free Software Foundation, Inc.
2012-11-09 17:53:51 +00:00
This file is part of the GNU C Library.
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/>. */
#include <sysdep.h>
#include <tls.h>
#include "tlsdesc.h"
#define NSAVEDQREGPAIRS 16
#define SAVE_Q_REGISTERS \
stp q0, q1, [sp, #-32*NSAVEDQREGPAIRS]!; \
cfi_adjust_cfa_offset (32*NSAVEDQREGPAIRS); \
stp q2, q3, [sp, #32*1]; \
stp q4, q5, [sp, #32*2]; \
stp q6, q7, [sp, #32*3]; \
stp q8, q9, [sp, #32*4]; \
stp q10, q11, [sp, #32*5]; \
stp q12, q13, [sp, #32*6]; \
stp q14, q15, [sp, #32*7]; \
stp q16, q17, [sp, #32*8]; \
stp q18, q19, [sp, #32*9]; \
stp q20, q21, [sp, #32*10]; \
stp q22, q23, [sp, #32*11]; \
stp q24, q25, [sp, #32*12]; \
stp q26, q27, [sp, #32*13]; \
stp q28, q29, [sp, #32*14]; \
stp q30, q31, [sp, #32*15];
#define RESTORE_Q_REGISTERS \
ldp q2, q3, [sp, #32*1]; \
ldp q4, q5, [sp, #32*2]; \
ldp q6, q7, [sp, #32*3]; \
ldp q8, q9, [sp, #32*4]; \
ldp q10, q11, [sp, #32*5]; \
ldp q12, q13, [sp, #32*6]; \
ldp q14, q15, [sp, #32*7]; \
ldp q16, q17, [sp, #32*8]; \
ldp q18, q19, [sp, #32*9]; \
ldp q20, q21, [sp, #32*10]; \
ldp q22, q23, [sp, #32*11]; \
ldp q24, q25, [sp, #32*12]; \
ldp q26, q27, [sp, #32*13]; \
ldp q28, q29, [sp, #32*14]; \
ldp q30, q31, [sp, #32*15]; \
ldp q0, q1, [sp], #32*NSAVEDQREGPAIRS; \
cfi_adjust_cfa_offset (-32*NSAVEDQREGPAIRS);
.text
/* Compute the thread pointer offset for symbols in the static
TLS block. The offset is the same for all threads.
Prototype:
_dl_tlsdesc_return (tlsdesc *) ;
*/
.hidden _dl_tlsdesc_return
.global _dl_tlsdesc_return
.type _dl_tlsdesc_return,%function
cfi_startproc
.align 2
_dl_tlsdesc_return:
ldr x0, [x0, #8]
RET
cfi_endproc
.size _dl_tlsdesc_return, .-_dl_tlsdesc_return
[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 11:37:49 +00:00
/* Same as _dl_tlsdesc_return but with synchronization for
lazy relocation.
Prototype:
_dl_tlsdesc_return_lazy (tlsdesc *) ;
*/
.hidden _dl_tlsdesc_return_lazy
.global _dl_tlsdesc_return_lazy
.type _dl_tlsdesc_return_lazy,%function
cfi_startproc
.align 2
_dl_tlsdesc_return_lazy:
/* The ldar here happens after the load from [x0] at the call site
(that is generated by the compiler as part of the TLS access ABI),
so it reads the same value (this function is the final value of
td->entry) and thus it synchronizes with the release store to
td->entry in _dl_tlsdesc_resolve_rela_fixup ensuring that the load
from [x0,#8] here happens after the initialization of td->arg. */
ldar xzr, [x0]
ldr x0, [x0, #8]
RET
cfi_endproc
.size _dl_tlsdesc_return_lazy, .-_dl_tlsdesc_return_lazy
2012-11-09 17:53:51 +00:00
/* Handler for undefined weak TLS symbols.
Prototype:
_dl_tlsdesc_undefweak (tlsdesc *);
The second word of the descriptor contains the addend.
Return the addend minus the thread pointer. This ensures
that when the caller adds on the thread pointer it gets back
the addend. */
.hidden _dl_tlsdesc_undefweak
.global _dl_tlsdesc_undefweak
.type _dl_tlsdesc_undefweak,%function
cfi_startproc
.align 2
_dl_tlsdesc_undefweak:
str x1, [sp, #-16]!
cfi_adjust_cfa_offset (16)
[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 11:37:49 +00:00
/* The ldar here happens after the load from [x0] at the call site
(that is generated by the compiler as part of the TLS access ABI),
so it reads the same value (this function is the final value of
td->entry) and thus it synchronizes with the release store to
td->entry in _dl_tlsdesc_resolve_rela_fixup ensuring that the load
from [x0,#8] here happens after the initialization of td->arg. */
ldar xzr, [x0]
2012-11-09 17:53:51 +00:00
ldr x0, [x0, #8]
mrs x1, tpidr_el0
sub x0, x0, x1
ldr x1, [sp], #16
cfi_adjust_cfa_offset (-16)
2012-11-09 17:53:51 +00:00
RET
cfi_endproc
.size _dl_tlsdesc_undefweak, .-_dl_tlsdesc_undefweak
#ifdef SHARED
/* Handler for dynamic TLS symbols.
Prototype:
_dl_tlsdesc_dynamic (tlsdesc *) ;
The second word of the descriptor points to a
tlsdesc_dynamic_arg structure.
Returns the offset between the thread pointer and the
object referenced by the argument.
ptrdiff_t
__attribute__ ((__regparm__ (1)))
_dl_tlsdesc_dynamic (struct tlsdesc *tdp)
{
struct tlsdesc_dynamic_arg *td = tdp->arg;
dtv_t *dtv = *(dtv_t **)((char *)__thread_pointer + DTV_OFFSET);
if (__builtin_expect (td->gen_count <= dtv[0].counter
&& (dtv[td->tlsinfo.ti_module].pointer.val
!= TLS_DTV_UNALLOCATED),
1))
return dtv[td->tlsinfo.ti_module].pointer.val
+ td->tlsinfo.ti_offset
- __thread_pointer;
return ___tls_get_addr (&td->tlsinfo) - __thread_pointer;
}
*/
.hidden _dl_tlsdesc_dynamic
.global _dl_tlsdesc_dynamic
.type _dl_tlsdesc_dynamic,%function
cfi_startproc
.align 2
_dl_tlsdesc_dynamic:
# define NSAVEXREGPAIRS 2
stp x29, x30, [sp,#-(32+16*NSAVEXREGPAIRS)]!
cfi_adjust_cfa_offset (32+16*NSAVEXREGPAIRS)
mov x29, sp
/* Save just enough registers to support fast path, if we fall
into slow path we will save additional registers. */
stp x1, x2, [sp, #32+16*0]
stp x3, x4, [sp, #32+16*1]
mrs x4, tpidr_el0
[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 11:37:49 +00:00
/* The ldar here happens after the load from [x0] at the call site
(that is generated by the compiler as part of the TLS access ABI),
so it reads the same value (this function is the final value of
td->entry) and thus it synchronizes with the release store to
td->entry in _dl_tlsdesc_resolve_rela_fixup ensuring that the load
from [x0,#8] here happens after the initialization of td->arg. */
ldar xzr, [x0]
2012-11-09 17:53:51 +00:00
ldr x1, [x0,#8]
ldr x0, [x4]
ldr x3, [x1,#16]
ldr x2, [x0]
cmp x3, x2
b.hi 2f
ldr x2, [x1]
add x0, x0, x2, lsl #4
ldr x0, [x0]
cmn x0, #0x1
b.eq 2f
ldr x1, [x1,#8]
add x0, x0, x1
sub x0, x0, x4
1:
ldp x1, x2, [sp, #32+16*0]
ldp x3, x4, [sp, #32+16*1]
ldp x29, x30, [sp], #(32+16*NSAVEXREGPAIRS)
cfi_adjust_cfa_offset (-32-16*NSAVEXREGPAIRS)
2012-11-09 17:53:51 +00:00
# undef NSAVEXREGPAIRS
RET
2:
/* This is the slow path. We need to call __tls_get_addr() which
means we need to save and restore all the register that the
callee will trash. */
/* Save the remaining registers that we must treat as caller save. */
# define NSAVEXREGPAIRS 7
stp x5, x6, [sp, #-16*NSAVEXREGPAIRS]!
cfi_adjust_cfa_offset (16*NSAVEXREGPAIRS)
stp x7, x8, [sp, #16*1]
stp x9, x10, [sp, #16*2]
stp x11, x12, [sp, #16*3]
stp x13, x14, [sp, #16*4]
stp x15, x16, [sp, #16*5]
stp x17, x18, [sp, #16*6]
SAVE_Q_REGISTERS
mov x0, x1
bl __tls_get_addr
mrs x1, tpidr_el0
sub x0, x0, x1
RESTORE_Q_REGISTERS
ldp x7, x8, [sp, #16*1]
ldp x9, x10, [sp, #16*2]
ldp x11, x12, [sp, #16*3]
ldp x13, x14, [sp, #16*4]
ldp x15, x16, [sp, #16*5]
ldp x17, x18, [sp, #16*6]
ldp x5, x6, [sp], #16*NSAVEXREGPAIRS
cfi_adjust_cfa_offset (-16*NSAVEXREGPAIRS)
b 1b
cfi_endproc
.size _dl_tlsdesc_dynamic, .-_dl_tlsdesc_dynamic
# undef NSAVEXREGPAIRS
#endif
/* This function is a wrapper for a lazy resolver for TLS_DESC
RELA relocations.
When the actual resolver returns, it will have adjusted the
TLS descriptor such that we can tail-call it for it to return
the TP offset of the symbol. */
.hidden _dl_tlsdesc_resolve_rela
.global _dl_tlsdesc_resolve_rela
.type _dl_tlsdesc_resolve_rela,%function
cfi_startproc
.align 2
_dl_tlsdesc_resolve_rela:
#define NSAVEXREGPAIRS 9
stp x29, x30, [sp, #-(32+16*NSAVEXREGPAIRS)]!
cfi_adjust_cfa_offset (32+16*NSAVEXREGPAIRS)
mov x29, sp
stp x1, x4, [sp, #32+16*0]
stp x5, x6, [sp, #32+16*1]
stp x7, x8, [sp, #32+16*2]
stp x9, x10, [sp, #32+16*3]
stp x11, x12, [sp, #32+16*4]
stp x13, x14, [sp, #32+16*5]
stp x15, x16, [sp, #32+16*6]
stp x17, x18, [sp, #32+16*7]
str x0, [sp, #32+16*8]
SAVE_Q_REGISTERS
ldr x1, [x3, #8]
bl _dl_tlsdesc_resolve_rela_fixup
RESTORE_Q_REGISTERS
ldr x0, [sp, #32+16*8]
ldr x1, [x0]
blr x1
ldp x1, x4, [sp, #32+16*0]
ldp x5, x6, [sp, #32+16*1]
ldp x7, x8, [sp, #32+16*2]
ldp x9, x10, [sp, #32+16*3]
ldp x11, x12, [sp, #32+16*4]
ldp x13, x14, [sp, #32+16*5]
ldp x15, x16, [sp, #32+16*6]
ldp x17, x18, [sp, #32+16*7]
ldp x29, x30, [sp], #(32+16*NSAVEXREGPAIRS)
cfi_adjust_cfa_offset (-32-16*NSAVEXREGPAIRS)
2012-11-09 17:53:51 +00:00
ldp x2, x3, [sp], #16
cfi_adjust_cfa_offset (-16)
RET
#undef NSAVEXREGPAIRS
cfi_endproc
.size _dl_tlsdesc_resolve_rela, .-_dl_tlsdesc_resolve_rela
/* This function is a placeholder for lazy resolving of TLS
relocations. Once some thread starts resolving a TLS
relocation, it sets up the TLS descriptor to use this
resolver, such that other threads that would attempt to
resolve it concurrently may skip the call to the original lazy
resolver and go straight to a condition wait.
When the actual resolver returns, it will have adjusted the
TLS descriptor such that we can tail-call it for it to return
the TP offset of the symbol. */
.hidden _dl_tlsdesc_resolve_hold
.global _dl_tlsdesc_resolve_hold
.type _dl_tlsdesc_resolve_hold,%function
cfi_startproc
.align 2
_dl_tlsdesc_resolve_hold:
#define NSAVEXREGPAIRS 10
1:
stp x29, x30, [sp, #-(32+16*NSAVEXREGPAIRS)]!
cfi_adjust_cfa_offset (32+16*NSAVEXREGPAIRS)
mov x29, sp
stp x1, x2, [sp, #32+16*0]
stp x3, x4, [sp, #32+16*1]
stp x5, x6, [sp, #32+16*2]
stp x7, x8, [sp, #32+16*3]
stp x9, x10, [sp, #32+16*4]
stp x11, x12, [sp, #32+16*5]
stp x13, x14, [sp, #32+16*6]
stp x15, x16, [sp, #32+16*7]
stp x17, x18, [sp, #32+16*8]
str x0, [sp, #32+16*9]
SAVE_Q_REGISTERS
adr x1, 1b
bl _dl_tlsdesc_resolve_hold_fixup
RESTORE_Q_REGISTERS
ldr x0, [sp, #32+16*9]
ldr x1, [x0]
blr x1
ldp x1, x2, [sp, #32+16*0]
ldp x3, x4, [sp, #32+16*1]
ldp x5, x6, [sp, #32+16*2]
ldp x7, x8, [sp, #32+16*3]
ldp x9, x10, [sp, #32+16*4]
ldp x11, x12, [sp, #32+16*5]
ldp x13, x14, [sp, #32+16*6]
ldp x15, x16, [sp, #32+16*7]
ldp x17, x18, [sp, #32+16*8]
ldp x29, x30, [sp], #(32+16*NSAVEXREGPAIRS)
cfi_adjust_cfa_offset (-32-16*NSAVEXREGPAIRS)
2012-11-09 17:53:51 +00:00
RET
cfi_endproc
.size _dl_tlsdesc_resolve_hold, .-_dl_tlsdesc_resolve_hold
#undef NSAVEXREGPAIRS