The fix for PR 13697
(https://sourceware.org/bugzilla/show_bug.cgi?id=13697) added a
KEEP(*(.data)) to prevent the .bss section from getting an incorrect
VMA.
The patch fixes the problem, but it also prevents the linker
from gc'ing .data sections. It also only keeps sections named
.data, so -fdata-sections could potentially cause the fix to break
(it works right now because the assembler generates an empty .data section).
The following patch fixes the problem by explicitly setting the
.bss section's VMA to the end of .data's VMA. It kinda looks redundant,
as both .data and .bss are assigned to the same region, but is needed because
the region's ORIGIN is hardcoded whereas the start of .data/.bss varies
by device (and is passed along as -Tdata=<address> by the compiler
driver).
Does this look ok? I verified that the testcase mentioned in the
bug report works fine.
ld/ChangeLog
2014-07-02 Senthil Kumar Selvaraj <senthil_kumar.selvaraj@atmel.com>
* scripttempl/avr.sc: Remove KEEP for .data and
force .bss VMA to end of .data VMA.
diff --git ld/scripttempl/avr.sc ld/scripttempl/avr.sc
index 10714e1..d356b71 100644
--- ld/scripttempl/avr.sc
+++ ld/scripttempl/avr.sc
@@ -166,10 +166,7 @@ SECTIONS
.data ${RELOCATING-0} :
{
${RELOCATING+ PROVIDE (__data_start = .) ; }
- /* --gc-sections will delete empty .data. This leads to wrong start
- addresses for subsequent sections because -Tdata= from the command
- line will have no effect, see PR13697. Thus, keep .data */
- KEEP (*(.data))
+ *(.data)
${RELOCATING+ *(.data*)}
*(.rodata) /* We need to include .rodata here if gcc is used */
${RELOCATING+ *(.rodata*)} /* with -fdata-sections. */
@@ -179,7 +176,7 @@ SECTIONS
${RELOCATING+ PROVIDE (__data_end = .) ; }
} ${RELOCATING+ > data ${RELOCATING+AT> text}}
- .bss ${RELOCATING-0} :${RELOCATING+ AT (ADDR (.bss))}
+ .bss ${RELOCATING+ ADDR(.data) + SIZEOF (.data)} ${RELOCATING-0} :${RELOCATING+ AT (ADDR (.bss))}
{
${RELOCATING+ PROVIDE (__bss_start = .) ; }
*(.bss)