[Mono-dev] Mono AOT for 64-bit Windows

[Gavin Dodd]

I just dealt with AOT cross compiling from win64 to freebsd/amd64 and had something like this

mono_marshal_get_icall_wrapper emits the address of the function to call at the time of compilation.


As I was only concerned with AOT I just changed

mono_mb_emit_native_call (mb, csig2, (gpointer) func);

to

mono_mb_emit_byte (mb, MONO_CUSTOM_PREFIX);
mono_mb_emit_op (mb, CEE_MONO_ICALL, func);

I'm sure there was another case that also emitted the wrong address in the same way but can't find it at the moment.

Another issue that probably doesn't apply to you, but caused huge problems, was alignment of data structures between different compilers.
The mono vm compiled on freebsd had different alignment for bitfields than the compiler built with msvc.

Gavin




________________________________
 From: voldemarz <voldemarz at gmail.com>
To: mono-devel-list at lists.ximian.com 
Sent: Thursday, June 27, 2013 9:52 AM
Subject: Re: [Mono-dev] Mono AOT for 64-bit Windows
 

Thanks for input. By seasoned you mean someone who has good knowledge of Mono
internals or generally a guy with low level experience on various
platforms/architectures?

I slightly modified ASM generation 
- wrapped .type and .size within .def/.endef directives
- removed local symbol emission
- removed underscore in front of global symbols 

Now I can generate AOT images and runtime seems to find all data structures
it needs. In few first runs surprisingly some simple tests worked
successfully in aot and full-aot mode - console output, simple calculations,
events, catching exceptions, allocating bunch of objects and invoking
garbage collection.

Then I regenerated AOT image of mscorlib and it doesn't work any more.
Trying to figure out issues now. In full-aot mode it crashes after a jump to
a non-executable memory region from function
wrapper_managed_to_native_object___icall_wrapper_mono_object_new_ptrfree_box_intptr

The first few ops after the function prolog look like this (full function
asm  here <http://pastebin.com/GfTLeN1e>  ):
mov         rax,7FACDFD19E0h
mov         rcx,qword ptr [rbp-70h]
sub         rsp,20h
mov        qword ptr [rbp-38h],rsp
call         rax

The address loaded into rax looks very suspicios. As far I can see that
function is compiled from IL. Haven't tracked down were is it coming from?
Could someone point me to it? What is it attempting call internally?



--
View this message in context: http://mono.1490590.n4.nabble.com/Mono-AOT-for-64-bit-Windows-tp4659926p4660049.html
Sent from the Mono - Dev mailing list archive at Nabble.com.
_______________________________________________
Mono-devel-list mailing list
Mono-devel-list at lists.ximian.com
http://lists.ximian.com/mailman/listinfo/mono-devel-list
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ximian.com/pipermail/mono-devel-list/attachments/20130627/0456adea/attachment.html>