this will allow the compiler to cache and reuse the result, meaning we
no longer have to take care not to load it more than once for the sake
of archs where the load may be expensive.
depends on commit 1c84c99913 for
correctness, since otherwise the compiler could hoist loads during
stage 3 of dynamic linking before the initial thread-pointer setup.
this cleans up what had become widespread direct inline use of "GNU C"
style attributes directly in the source, and lowers the barrier to
increased use of hidden visibility, which will be useful to recovering
some of the efficiency lost when the protected visibility hack was
dropped in commit dc2f368e56, especially
on archs where the PLT ABI is costly.
In TLS variant I the TLS is above TP (or above a fixed offset from TP)
but on some targets there is a reserved gap above TP before TLS starts.
This matters for the local-exec tls access model when the offsets of
TLS variables from the TP are hard coded by the linker into the
executable, so the libc must compute these offsets the same way as the
linker. The tls offset of the main module has to be
alignup(GAP_ABOVE_TP, main_tls_align).
If there is no TLS in the main module then the gap can be ignored
since musl does not use it and the tls access models of shared
libraries are not affected.
The previous setup only worked if (tls_align & -GAP_ABOVE_TP) == 0
(i.e. TLS did not require large alignment) because the gap was
treated as a fixed offset from TP. Now the TP points at the end
of the pthread struct (which is aligned) and there is a gap above
it (which may also need alignment).
The fix required changing TP_ADJ and __pthread_self on affected
targets (aarch64, arm and sh) and in the tlsdesc asm the offset to
access the dtv changed too.
__ARM_ARCH_6ZK__ is a gcc specific historical typo which may not be
defined by other compilers.
https://gcc.gnu.org/ml/gcc-patches/2015-07/msg02237.html
To avoid unexpected results when building for ARMv6KZ with clang, the
correct form of the macro (ie 6KZ) needs to be tested. The incorrect
form of the macro (ie 6ZK) still needs to be tested for compatibility
with pre-2015 versions of gcc.
three problems are addressed:
- use of pc arithmetic, which was difficult if not impossible to make
correct in thumb mode on all models, so that relative rather than
absolute pointers to the backends could be used. this was designed
back when there was no coherent model for the early stages of the
dynamic linker before relocations, and is no longer necessary.
- assumption that data (the relative pointers to the backends) can be
accessed at a constant displacement from the code. this will not be
possible on future fdpic subarchs (for cortex-m), so move
responsibility for loading the backend code address to the caller.
- hard-coded arm opcodes using the .word directive. instead, use the
.arch directive to work around the assembler's refusal to assemble
instructions not available (or in some cases, available but just
considered deprecated) in the target isa level. the obscure v6t2
arch is used for v6 code so as to (1) allow generation of thumb2
output if -mthumb is active, and (2) avoid warnings/errors for mcr
barriers that clang would produce if we just set arch to v7-a.
in addition, the __aeabi_read_tp function is moved out of the inner
workings and implemented as an asm wrapper around a C function, so
that asm code does not need to read global data. the asm wrapper
serves to satisfy the ABI calling convention requirements for this
function.
using the actual mcontext_t definition rather than an overlaid pointer
array both improves correctness/readability and eliminates some ugly
hacks for archs with 64-bit registers bit 32-bit program counter.
also fix UB due to comparison of pointers not in a common array
object.
this builds on commits a603a75a72 and
0ba35d69c0 to ensure that a compiler
cannot conclude that it's valid to reorder the asm to a point before
the thread pointer is set up, or to treat the inline function as if it
were declared with attribute((const)).
other archs already use volatile asm for thread pointer loading.
previously, builds for pre-armv6 targets hard-coded use of the "kuser
helper" system for atomics and thread-pointer access, resulting in
binaries that fail to run (crash) on systems where this functionality
has been disabled (as a security/hardening measure) in the kernel.
additionally, builds for armv6 hard-coded an outdated/deprecated
memory barrier instruction which may require emulation (extremely
slow) on future models.
this overhaul replaces the behavior for all pre-armv7 builds (both of
the above cases) to perform runtime detection of the appropriate
mechanisms for barrier, atomic compare-and-swap, and thread pointer
access. detection is based on information provided by the kernel in
auxv: presence of the HWCAP_TLS bit for AT_HWCAP and the architecture
version encoded in AT_PLATFORM. direct use of the instructions is
preferred when possible, since probing for the existence of the kuser
helper page would be difficult and would incur runtime cost.
for builds targeting armv7 or later, the runtime detection code is not
compiled at all, and much more efficient versions of the non-cas
atomic operations are provided by using ldrex/strex directly rather
than wrapping cas.
armv7/thumb2 provides a way to do atomics in thumb mode, but for armv6
we need a call to arm mode.
this commit is based on a patch by Stephen Thomas which fixed the
armv7 cases but not the armv6 ones.
all of this should be revisited if/when runtime selection of thread
pointer access and atomics are added.
this is perhaps not the optimal implementation; a_cas still compiles
to nested loops due to the different interface contracts of the kuser
helper cas function (whose contract this patch implements) and the
a_cas function (whose contract mimics the x86 cmpxchg). fixing this
may be possible, but it's more complicated and thus deferred until a
later time.
aside from improving performance and code size, this patch also
provides a means of producing binaries which can run on hardened
kernels where the kuser helpers have been disabled. however, at
present this requires producing binaries for armv6k or later, which
will not run on older cpus. a real solution to the problem of kernels
that omit the kuser helpers would be runtime detection, so that
universal binaries which run on all arm cpu models can also be
compatible with all kernel hardening profiles. robust detection
however is a much harder problem, and will be addressed at a later
time.
despite documentation that makes it sound a lot different, the only
ABI-constraint difference between TLS variants II and I seems to be
that variant II stores the initial TLS segment immediately below the
thread pointer (i.e. the thread pointer points to the end of it) and
variant I stores the initial TLS segment above the thread pointer,
requiring the thread descriptor to be stored below. the actual value
stored in the thread pointer register also tends to have per-arch
random offsets applied to it for silly micro-optimization purposes.
with these changes applied, TLS should be basically working on all
supported archs except microblaze. I'm still working on getting the
necessary information and a working toolchain that can build TLS
binaries for microblaze, but in theory, static-linked programs with
TLS and dynamic-linked programs where only the main executable uses
TLS should already work on microblaze.
alignment constraints have not yet been heavily tested, so it's
possible that this code does not always align TLS segments correctly
on archs that need TLS variant I.
actually this is just to avoid gcc being stupid and refusing to inline
the function version, even when the size cost is essentially identical
whether it's inlined or not.
this port assumes eabi calling conventions, eabi linux syscall
convention, and presence of the kernel helpers at 0xffff0f?0 needed
for threads support. otherwise it makes very few assumptions, and the
code should work even on armv4 without thumb support, as well as on
systems with thumb interworking. the bits headers declare this a
little endian system, but as far as i can tell the code should work
equally well on big endian.
some small details are probably broken; so far, testing has been
limited to qemu/aboriginal linux.
Rich Felker
Szabolcs Nagy
Andre McCurdy