[PATCH][gold][aarch64][pr23870] Check whether symbols with MOVW_.ABS relocations require PLT entries

[Egeyar Bagcioglu]

ping

Adding Cary Coutant, since he could possibly review this patch as the 
gold maintainer.

Regards,
Egeyar


On 1/28/19 5:14 PM, Egeyar Bagcioglu wrote:
> Hello,
>
> The following patch on upstream glibc
>
> https://sourceware.org/ml/libc-alpha/2017-09/msg00282.html
>
> is causing crt1.o to contain R_AARCH64_MOVW_UABS_G* relocations for 
> the symbol "main". AArch64 backend of gold currently does not check 
> whether such symbols require a PLT entry.
>
> This is causing pr 23870:
> When a non-pie referring to a "main" symbol in a shared library is 
> being linked using a recent crt1.o, gold is failing to create a 
> dynamic symbol for "main". Therefore, upon calling "main", such an 
> executable seg-faults.
>
> I added a test case mimicking what _start and __libc_start_main does: 
> The former gives an undefined symbol as an argument to the latter 
> where it is called as a function. Therefore, the symbol does not 
> become a global symbol in either .o and the linker needs to arrange 
> the dynamic relocation.
>
> The attached patch checks if a symbol with R_AARCH64_MOVW_.ABS_G* 
> relocations needs a PLT entry and creates it if necessary. It is 
> tested on aarch64-unknown-linux-gnu and caused no regressions.
>
> gold/ChangeLog entry:
>
> 2019-01-28  Egeyar Bagcioglu  <egeyar.bagcioglu@oracle.com>
>
> Â Â Â  PR 23870
> Â Â Â  * aarch64.cc (Target_aarch64::Scan::global): Check if a symbol with
> Â Â Â  R_AARCH64_MOVW_.ABS_* relocations requires a PLT entry.
> Â Â Â  * testsuite/Makefile.am: Add aarch64_pr23870 test case.
> Â Â Â  * testsuite/Makefile.in: Regenerate.
> Â Â Â  * testsuite/aarch64_pr23870_bar.c: New file.
> Â Â Â  * testsuite/aarch64_pr23870_foo.c: New file.
> Â Â Â  * testsuite/aarch64_pr23870_main.S: New file.
>
>
> Please take a look and apply if legitimate.
>
> Regards,
> Egeyar