Dwarf views broken on ia64?

Jan Beulich JBeulich@suse.com
Mon May 28 08:42:00 GMT 2018


>>> On 26.05.18 at 02:43, <aoliva@redhat.com> wrote:
> On May 23, 2018, "Jan Beulich" <JBeulich@suse.com> wrote:
> 
>> Solution suggestions appreciated,
> 
> The problem seems to be caused by reuse of view symbols.  The same
> symbol may be copied to the debug_line associated with multiple slots.
> It looks like this fixes it,

Thanks, I'll give this a try (once I find time).

> but I'd appreciate if someone more familiar
> with the way ia64 does insn pack bundling could check that this makes
> sense.  I'm concerned that, even if this works as is, once GCC tries
> again to figure out on its own where view resets should be, and issue
> view asserts (view 0) at such points, placement of insns in bundles
> might cause views to be issued out of order, and thus end up failing
> asserts again.  Could this ever happen?

It's not clear to me to whom you raise this question. I'm certainly not in
the position to answer it, in particular without knowing what those views
are about and how they are supposed to work. From a general viewpoint,
I don't think the bundling of insns should affect any aspect of debug info
generation - individual insns still have separate (albeit fake) addresses
associated with them (while a bundle spans 16 bytes, the [up to] 3 insns
have addresses <bundle-base>, <bundle-base>+1, and <bundle-base>+2
- not sure if this helps).

Jan




More information about the Binutils mailing list