Thumb, new EABI and STT_ARM_TFUNC

[Jonathan Larmour <jifl at eCosCentric dot com>]

Last November, a change was made to implement the new EABI that removes 
the STT_ARM_TFUNC symbol type from generated objects, and instead sets the 
low bit of the symbol address, leaving the symbol as STT_FUNC. This is 
implemented in elf32-arm.c:elf32_arm_swap_symbol_out()

Unfortunately this, like other EABI changes, is not controllable by a 
flag, and so the linker is now always producing symbols without 
STT_ARM_TFUNC. (I haven't tried experimenting as to whether these changes 
also break binary compatibility full stop, but being a "new ABI" that 
seems possible).

This change is in the released binutils 2.16. However the latest stable 
GDB release, 6.3 was released around the same time and its BFD does not 
have that change since it was branched a little earlier. GDB uses the 
STT_ARM_TFUNC symbol type to determine if a symbol is Thumb, so with 
binutils 2.16, GDB treats all functions as ARM. When it comes to 
debugging, particularly setting breakpoints, this causes one or two 
problems :-).

Personally, I'm going to kludge GDB to notice the low bit being set in the 
symbol address, so I'm okay. But I thought in general that old ABI 
behaviours are normally preserved with a flag (or at worst a CPP macro) 
for a while. Is this the only ABI incompatibility introduced? I'm prepared 
to put in a little time to add such a flag if people think it should 
really be there, but I'd need to know more about the ABI differences and 
incompatibilities.

Jifl
--
eCosCentric    http://www.eCosCentric.com/    The eCos and RedBoot experts
--["No sense being pessimistic, it wouldn't work anyway"]-- Opinions==mine