View Bug Activity | Format For Printing
This warning is emitted in line 648 of file gas/config/tc-avr.c. This warning has been confusing a lot of users in the past, because it is ultimately generated for legitimate assembly code generated by GCC as part of jump tables resulting out of a larger switch statement. Discussion with the author of the atmega256x patch, Björn Haase, resulted in an agreement that the warning should just be removed. There is indeed a minimal danger that a switch statement could cross an 128 KiB boundary where a linker stub could suddenly be placed within the table, but as the assembler cannot reliably detect this situation at all, it confuses more people with legitimate code (not crossing a 128 KiB boundary) than it would be really able to help out those who are in danger. In the end, the users are told to ignore the warning, so it doesn't do any good at all. In addition, Björn pointed out that it could only detect part of the problem but is not able to cover all possible dangerous cases, because some code might hide dangerous operations between other assembly code that is not covered by the warning. If at all, this warning needs to be emitted by GCC itself whenever it generates the possibly dangerous code.
Created an attachment (id=2177) Remove the warning
Hi Joerg, Please could you try out the uploaded patch and let me know if it resolves the problem for you ? Also - would it be possible for you to supply a small assembler source file that demonstrates this problem ? This could be turned in a gas test case to make sure that the problem remains fixed. Cheers Nick
I applied this patch to Binutils-2.18.50 I ran full avr gcc testsuite and there were no regressions. Importantly, it removed all the failures caused by the warning message removed by this patch. So its get my vote!
Hi Guys, I have checked the patch in. Cheers Nick gas/ChangeLog PR 5523 * config/tc-avr.c (avr_ldi_expression): Do not warn about unknown relocs here.