On 10/3/2021 7:03 PM, Eliot Moss via gem5-users wrote:
On 10/3/2021 2:34 PM, Eliot Moss via gem5-users wrote:
On 10/3/2021 1:43 PM, walt_90--- via gem5-users wrote:
I am seeing a similar fault. This is for gem5 21.0.0.0, for X86.
When trying to decode 0x1E in 64-bit mode, it tries to construct
a UD2 macroop (undefined instruction). Somewhere in doing that,
it attempts to "free" using a bad pointer.
- When compiled in 'fast' mode, it seg faults.
- When compiled in 'opt' mode, tcmalloc shows the bad "free"
- When compiled in 'debug' mode, it works (!).
My guess is that there's a subtle C++ compiler bug in there around
reference counted shared things, or a C++ library bug. I am using
gcc 7.5.0 and libc.so.6. I will ttry a more recent compiler and
see what happens ...
Using gcc 10.3.0 fixed the behavior for gem5.opt (or somehow avoided it).
I am trying now with gem5.fast and gem5.debug and will report back.
Confirming that using gcc 10.3.0 also causes the undesired behavior
to disappear for gem5.fast, and gem5.debug continues to work properly.
I suspect there was some optimization bug in g++, fixed in the intervening
versions.
Best - Eliot
_______________________________________________
gem5-users mailing list -- gem5-users@gem5.org
To unsubscribe send an email to gem5-users-le...@gem5.org
%(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
