On 14.10.11 13:14, Georg-Johann Lay wrote:
> If I understand correctly, my changes in
> http://gcc.gnu.org/viewcvs/trunk/gcc/config/avr/libgcc.S?r1=179760&r2=179759&pathrev=179760
> are no good and I should rework the patch to use EI*, and instead of PUSH
> zero_reg there should either a push of EIND or, if no RET is used to indirect
> jump, EIJMP instead if IJMP?
With the linker inserting trampolines, and relaxation using them,
compatibility is provided by I*, with EI* working only because
EIND == 0, IIUC? Then emitting EI* only has a purpose if there are plans
to use EIND != 0, AIUI. (And then we've bypassed the trampolines, IIUC. [1]
In my mind, that raises the question: Since it's working now, why?)
In the OP, concern was raised that the use of EIND is not IRQ-safe, so
the trampolines should be used instead. Given that any interrupt handler
has to save any registers it clobbers, then I don't immediately see how
EIND should be IRQ-unsafe? (So the above could be done, but the goal is
still a bit unclear, at least at this end of the pipe.)
Erik
[1] Or, is there still a stub in this case, to set EIND, depending on
the address of the destination? (Because we still only have 16 bit
pointers in avr-gcc.)
--
If you understand what you're doing, you're not learning anything.
- A. L.
_______________________________________________
AVR-GCC-list mailing list
AVR-GCC-list@nongnu.org
https://lists.nongnu.org/mailman/listinfo/avr-gcc-list
