It appears that starting with binutils 2.26, evaluating the current address (using the dot symbol) after a .align directive using data-based padding (e.g. .align 6, 0) makes it non-constant, resulting in errors such as: "Error: non-constant expression in ".if" statement" when evaluating "." inside an .if statement. This is on aarch64. The issue appeared when build the ARM Trusted Firmware: https://github.com/ARM-software/arm-trusted-firmware The related assembly is a combination of the vector_entry/vector_base (that does the alignment) and check_vector_size (that uses the dot symbol) macros, defined at: https://github.com/ARM-software/arm-trusted-firmware/blob/master/include/common/asm_macros.S#L69 and used at https://github.com/ARM-software/arm-trusted-firmware/blob/master/bl31/aarch64/runtime_exceptions.S#L171 The assembly is correct and is apparently not the problem. However, switching the ".align x, 0" directives to ".align x" solves the issue, but upstream is not going to make that change. The issue was discussed in depth at https://www.mail-archive.com/linaro-toolchain@lists.linaro.org/msg05685.html and an issue was opened on the ARM Trusted Firmware tracker at https://github.com/ARM-software/tf-issues/issues/401 A test case was provided at https://www.mail-archive.com/linaro-toolchain@lists.linaro.org/msg05688/tmp.s The issue was bisected down to (bad) commit c1baaddf8861aea666b84baeb4746caff51a579d commit c1baaddf8861aea666b84baeb4746caff51a579d Author: Renlin Li <renlin...@arm.com> Date: Thu Apr 2 14:59:45 2015 +0100 [AArch64] Emit DATA_MAP in order within text section 2015-03-27 Renlin Li <renlin...@arm.com> gas/ * config/tc-aarch64.c (mapping_state): Emit MAP_DATA within text section in order. (mapping_state_2): Don't emit MAP_DATA here. (s_aarch64_inst): Align frag during state transition. (md_assemble): Likewise.
Created attachment 9419 [details] Proposed patch Hi Paul, This is a subtle one. What is happening is that ".align <num>" in a code section (like .text or .vectors) will use the default NOP value to fill in the space. But ".align <num>, <fill>" will use the provided fill value. This matters because the AArch64 backend knows that NOP is an instruction, but it does not know about <fill> and it assumes that it is a *data* value, not an *instruction*. So it changes the mapping state to MAP_DATA, creates a mapping symbol ($d) indicate the change of state, and creates a new fragment to contain the data. All of this means that when the macro parser tries to evaluate the .if statement it finds that dot is one fragment and the symbol is in another fragment. Since fragments can grow or shrink the distance between them cannot be computed reliably (at the time that the .if expressions is evaluated) and so the assembler issues its error message. I have uploaded a small patch which I think will fix the problem. It tells the AArch64 backend not to assume that alignment directives in code sections contain data. Instead it should assume that they contain instructions, and so not change the mapping state. This causes a subtle problem with regard to literal pools, but the patch takes care of that. Please give the patch a try and let me know how you get on. Cheers Nick
The master branch has been updated by Nick Clifton <nickc@sourceware.org>: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=7ea12e5c3ad54da440c08f32da09534e63e515ca commit 7ea12e5c3ad54da440c08f32da09534e63e515ca Author: Nick Clifton <nickc@redhat.com> Date: Fri Aug 5 10:37:57 2016 +0100 Fix the generation of alignment frags in code sections for AArch64. PR gas/20364 * config/tc-aarch64.c (s_ltorg): Change the mapping state after aligning the frag. (aarch64_init): Treat rs_align frags in code sections as containing code, not data. * testsuite/gas/aarch64/pr20364.s: New test. * testsuite/gas/aarch64/pr20364.d: New test driver.
I have gone ahead and applied the patch.
Thanks for the explanation and the fix! It does solve the problem we had with the ARM Trusted Firmware codebase. I'm glad it was merged, too.
The linaro_binutils-2_27-branch branch has been updated by Christophe Lyon <clyon@sourceware.org>: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=eb65c97589b3c6a5c2954cea45c068b30c751b51 commit eb65c97589b3c6a5c2954cea45c068b30c751b51 Author: Nick Clifton <nickc@redhat.com> Date: Fri Aug 5 10:37:57 2016 +0100 Fix the generation of alignment frags in code sections for AArch64. PR gas/20364 * config/tc-aarch64.c (s_ltorg): Change the mapping state after aligning the frag. (aarch64_init): Treat rs_align frags in code sections as containing code, not data. * testsuite/gas/aarch64/pr20364.s: New test. * testsuite/gas/aarch64/pr20364.d: New test driver.