[Bug ld/31321] New: AVR: Remove PROVIDE from definition of symbol __flmap_init_label

[gjl at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31321

            Bug ID: 31321
           Summary: AVR: Remove PROVIDE from definition of symbol
                    __flmap_init_label
           Product: binutils
           Version: 2.42
            Status: NEW
          Severity: normal
          Priority: P2
         Component: ld
          Assignee: unassigned at sourceware dot org
          Reporter: gjl at gcc dot gnu.org
  Target Milestone: ---

PR31124 introduced symbol __flmap_init_label as:

PROVIDE (__flmap_init_label = DEFINED(__flmap_init_start) ? __flmap_init_start
: 0) ;

for emulations avrxmega2_flmap and avrxmega4_flmap.

The symbol is supposed to be used by the startup-code crt*.o from AVR-LibC: The
symbol resolves to:

* __flmap_init_start when NVMCTRL_CTRLB.FLMAP should be initialized according
to __flmap_value_with_lock, which is the case for emulations avrxmega2/4_flmap.

* __flmap_noinit_start when the startup code should bypass the initialization
of NVMCTRL_CTRLB.FLMAP, which is the case for emulations avrxmega2/4.

All this works as expected, except in the case when such a crt*.o is used with
a version of Binutils that does not implement PR31124.  This can happen when
crt*.o is deployed as part of a device pack that's build with PR31124 but used
with a toolchain that does not have PR31124.  In such a case, linking will
throw an undefined reference to __flmap_init_label.

The obvious fix would be to provide a default for that symbol in gcrt1.S like:

.weak __flmap_init_label
.set  __flmap_init_label, __flmap_noinit_start

which solves the undefined reference, however __flmap_init_label ceases to work
when PR31124 *IS* supported, because the PROVIDE (__flmap_init_label =
__flmap_init_start) won't touch the already defined __flmap_init_label, hence
init of NVMCTRL_CTRLB.FLMAP would be bypassed when it must not be bypassed.

Hence, remove PROVIDE from __flmap_init_label.

-- 
You are receiving this mail because:
You are on the CC list for the bug.