On 27 Dec 2022 20:35, Dragan Mladjenovic wrote:
> On 25-Dec-22 01:26, Mike Frysinger wrote:
> > On 02 Feb 2022 11:17, Dragan Mladjenovic wrote:
> >> --- a/sim/mips/configure.ac
> >> +++ b/sim/mips/configure.ac
> >> @@ -216,6 +227,7 @@ if test ${sim_gen} = MULTI; then
> >>   #include "sim-main.h"
> >>   #include "multi-include.h"
> >>   #include "elf-bfd.h"
> >> +#include "elfxx-mips.h"
> >>   #include "elf/mips.h"
> >>   
> >>   #define SD sd
> >> @@ -235,7 +247,11 @@ sim_engine_run (SIM_DESC sd,
> >>   	   & EF_MIPS_ARCH_ASE_MICROMIPS)
> >>       mach = bfd_mach_mips_micromips;
> >>     else
> >> -    mach = STATE_ARCHITECTURE (SD)->mach;
> >> +  {
> >> +     mach = _bfd_elf_mips_mach (elf_elfheader (STATE_PROG_BFD (sd))->e_flags);
> >> +     if (!mach)
> >> +       mach = STATE_ARCHITECTURE (SD)->mach;
> >> +  }
> > 
> > why do you need this logic in the main engine run routine ?  why isn't the
> > STATE_ARCHITECTURE value already set to the correct value ?  if it isn't set
> > correctly all the time, this can lead to incorrect behavior elsewhere when
> > code checks it.
> > 
> > when the sim is opened, the bfd_mach is either explicitly passed in (which
> > gdb does), or initialized with a call to the bfd library APIs.  see the
> > sim/common/sim-utils.c:sim_analyze_program code.
> 
> Sorry but the exact reason for doing so eludes me. It is done on 
> internal jump for gdb 8 to gdb 10, permissibly to allow running
> binary even whet set arch forces older mips isa, but I see little
> use of that outside of testsuite maybe.

i'd argue that if you load a binary for one ISA, then force the runtime
to use a different one, and the result crashes, then the sim is WAI.  it
should behave as real hardware does, and if you load a binary that needs
a newer ISA than a CPU provides, it won't work.

do you know if there's a testsuite, sim or gdb, that exercises this ?
-mike