Jason Gunthorpe <[EMAIL PROTECTED]> writes: > On 21 Feb 2000, David Huggins-Daines wrote: > > > 3) C++ runtime support that reads the frame info in exception handlers > > (gcc/frame.o) > > Won't this also generate unaligned faults when they access the > information? Has anyone tested an actual exception throw to see how many > faults crop up?
No - read the code in gcc/frame.c - it already *expects* this information to be unaligned (in fact that's where I took the code to read unaligned quadwords from :). This is why I suspect that the ld.so workaround is the only way to fix this - gas knows it's unaligned, ld knows it, the runtime knows it, only ld.so doesn't...
