libffi on SPARC V9 have problem calling functions with floating point number arguments

[psumbera]

I see the problem on Solaris SPARC V9 with /usr/bin/gjs when it calls

clutter_actor_set_size (ClutterActor *self, gfloat width, gfloat height);

https://github.com/GNOME/mutter/blob/d643eb5c6fe50e7f1afffda0e8747a87f668a799/clutter/clutter/clutter-actor.c#L9473

Note the floating point number arguments. When I add some debug printfs there I see that it gets just zeros.

This is regression. Old version 3.2.1 worked well. The first bad commit was: 2b27890
(but I know it was major rewrite up and there were other follow up SPARC releated commits)

[psumbera]

The resolution for this particual issue seems to be add SPARC_FLAG_FP_ARGS into flags:

flags |= SPARC_FLAG_FP_ARGS;

for FFI_TYPE_FLOAT and FFI_TYPE_DOUBLE

libffi/src/sparc/ffi64.c

Line 385 in ac598b7

case FFI_TYPE_FLOAT:

But there seems to be some other SPARC problems too... (gnome-shell has still some issues even it's now able to start).

[psumbera]

I see that ffi_prep_cif_machdep_core does set SPARC_FLAG_FP_ARGS apropriatery. And libffi tests seems to pass (even with floating point arguments).

On the other hand gjs 1.70.2 doesn't seem to call ffi_prep_cif for clutter_actor_set at all.

[thesamesam]

Thanks for reporting this, I think I've seen this as well but I wasn't sure enough yet if it was a libffi bug (https://bugs.gentoo.org/882071) but I think it must be the same.

[thesamesam]

cc @rth7680 (commit author), although not sure if you're involved with libffi nowadays.

[psumbera]

I have looked at this again and now with my initial finding in #778 (comment) I don't see any other obvious problems (I believe originally there was problem with Gnome unlocking for example).

So, maybe following patch is enough?

--- libffi-3.4.4/src/sparc/ffi64.c
+++ libffi-3.4.4/src/sparc/ffi64.c
@@ -383,12 +383,16 @@
          break;
        case FFI_TYPE_UINT32:
        case FFI_TYPE_FLOAT:
+         flags |= SPARC_FLAG_FP_ARGS;
          *argp++ = *(UINT32 *)a;
          break;
+       case FFI_TYPE_DOUBLE:
+         flags |= SPARC_FLAG_FP_ARGS;
+         *argp++ = *(UINT64 *)a;
+         break;
        case FFI_TYPE_SINT64:
        case FFI_TYPE_UINT64:
        case FFI_TYPE_POINTER:
-       case FFI_TYPE_DOUBLE:
          *argp++ = *(UINT64 *)a;
          break;

@thesamesam can you please test whether it solves your problems?

[rth7680]

Setting FP_ARGS for FFI_TYPE_UINT32 is obviously wrong.
Need to separate the FFI_TYPE_FLOAT case.

[psumbera]

Ok, it sounds right. I have updated the patch. But with my test case of Gnome (gjs) it doesn't seem to make any difference...

--- libffi-3.4.4/src/sparc/ffi64.c
+++ libffi-3.4.4/src/sparc/ffi64.c
@@ -382,13 +382,19 @@
          *argp++ = *(SINT32 *)a;
          break;
        case FFI_TYPE_UINT32:
-       case FFI_TYPE_FLOAT:
          *argp++ = *(UINT32 *)a;
          break;
        case FFI_TYPE_SINT64:
        case FFI_TYPE_UINT64:
        case FFI_TYPE_POINTER:
+         *argp++ = *(UINT64 *)a;
+         break;
+       case FFI_TYPE_FLOAT:
+         flags |= SPARC_FLAG_FP_ARGS;
+         *argp++ = *(UINT32 *)a;
+         break;
        case FFI_TYPE_DOUBLE:
+         flags |= SPARC_FLAG_FP_ARGS;
          *argp++ = *(UINT64 *)a;
          break;

[thesamesam]

Thanks, I'll try the second patch first. Wrt no difference - you mean both patches fix your issue, or you mean the second patch doesn't fix your gjs problem? (Hopefully the former..)

[psumbera]

Thanks, I'll try the second patch first. Wrt no difference - you mean both patches fix your issue, or you mean the second patch doesn't fix your gjs problem? (Hopefully the former..)

Yes, the seond patch works for me too. I wasn't able to run gjs tests (which is probably your test case) as there is some unrelated Python problem (I think).

[thesamesam]

gjs tests before, with the failing GIMarshalling test:

 1/71 gjs:JS / GIMarshalling                       ERROR           7.93s   killed by signal 6 SIGABRT
>>> G_DEBUG=fatal-warnings,fatal-criticals TSAN_OPTIONS=history_size=5,force_seq_cst_atomics=1,suppressions=/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0/installed-tests/extra/tsan.supp GSETTINGS_SCHEM
A_DIR=/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/installed-tests/js GJS_DEBUG_TOPICS='' LD_LIBRARY_PATH=/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/:/var/tmp/portage/dev-libs/gj
s-1.78.0/work/gjs-1.78.0-build:/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/installed-tests/js:/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/installed-tests/js/libgjstesttools ENABL
E_GTK=yes LSAN_OPTIONS=fast_unwind_on_malloc=0,exitcode=23,suppressions=/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0/installed-tests/extra/lsan.supp G_FILENAME_ENCODING=latin1 LC_ALL=C.utf8 TOP_BUILDD
IR=/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build DYLD_FALLBACK_LIBRARY_PATH=/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build:/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/i
nstalled-tests/js:/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/installed-tests/js/libgjstesttools GJS_USE_UNINSTALLED_FILES=1 G_SLICE=always-malloc GI_TYPELIB_PATH=/var/tmp/portage/dev-libs/gjs-
1.78.0/work/gjs-1.78.0-build:/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/installed-tests/js:/var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/installed-tests/js/libgjstesttools GJS_PAT
H='' GSETTINGS_BACKEND=memory GJS_DEBUG_OUTPUT=stderr NO_AT_BRIDGE=1 MALLOC_PERTURB_=62 ASAN_OPTIONS=intercept_tls_get_addr=0 /var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/installed-tests/js/mini
jasmine /var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/../gjs-1.78.0/installed-tests/js/testGIMarshalling.js

stderr:
**
ERROR:/usr/share/gobject-introspection-1.0/tests/gimarshallingtests.c:904:gi_marshalling_tests_float_in: assertion failed (v == G_MAXFLOAT): (0 == 3.40282347e+38)

(test program exited with status code -6)


71/71 gjs:Scripts / CommandLine                    OK             16.65s   46 subtests passed

Summary of Failures:

 2/71 gjs:JS / GObjectValue                FAIL            2.79s   134/137 subtests passed
 3/71 gjs:JS / Regress                     FAIL            2.37s   275/276 subtests passed
 1/71 gjs:JS / GIMarshalling               ERROR           7.93s   killed by signal 6 SIGABRT

Ok:                 68
Expected Fail:      0
Fail:               3
Unexpected Pass:    0
Skipped:            0
Timeout:            0

After, all tests pass:

Ok:                 71
Expected Fail:      0
Fail:               0
Unexpected Pass:    0
Skipped:            0
Timeout:            0

Full log written to /var/tmp/portage/dev-libs/gjs-1.78.0/work/gjs-1.78.0-build/meson-logs/testlog.txt
>>> Completed testing dev-libs/gjs-1.78.0

Thanks a bunch! Would you mind making a PR for it?

[thesamesam]

Fixed by #802.