Luckily the crash is very deterministic. It happens in thread 0 and a partial stacktrace is this. 1 ED_region_do_draw area.c 479 0x10ce110 2 wm_method_draw_overlap_all wm_draw.c 289 0x9c5164 3 wm_draw_update wm_draw.c 953 0x9c8b84 4 WM_main wm.c 495 0x9bdbc8 5 main creator.c 2102 0x96b855 ED_region_do_draw() is called many times without problems but at a certain time it crashes. I'm still investigating to find the exact crashing spot. I can do source level debugging for blender itself but not for the crashing spot. Nevertheless it is also very deterministic and the disassembly looks like this: 0x7f3282a621aa 83 c7 0c add $0xc,%edi 0x7f3282a621ad 18 d2 sbb %dl,%dl 0x7f3282a621af c4 41 31 ef c9 vpxor %xmm9,%xmm9,%xmm9 0x7f3282a621b4 c4 41 29 ef d2 vpxor %xmm10,%xmm10,%xmm10 0x7f3282a621b9 44 39 f7 cmp %r14d,%edi 0x7f3282a621bc 77 6d ja 0x7f3282a6222b 0x7f3282a621be c4 c1 79 92 c7 kmovb %r15d,%k0 0x7f3282a621c3 c5 f8 93 f8 kmovw %k0,%edi 0x7f3282a621c7 81 e7 01 00 00 00 and $0x1,%edi 0x7f3282a621cd c4 41 29 ef d2 vpxor %xmm10,%xmm10,%xmm10 0x7f3282a621d2 40 84 ff test %dil,%dil 0x7f3282a621d5 75 54 jne 0x7f3282a6222b 0x7f3282a621d7 81 e3 01 00 00 00 and $0x1,%ebx 0x7f3282a621dd c5 f8 92 c3 kmovw %ebx,%k0 0x7f3282a621e1 c5 f8 93 f8 kmovw %k0,%edi 0x7f3282a621e5 81 e7 01 00 00 00 and $0x1,%edi 0x7f3282a621eb c4 41 29 ef d2 vpxor %xmm10,%xmm10,%xmm10 0x7f3282a621f0 40 84 ff test %dil,%dil 0x7f3282a621f3 75 36 jne 0x7f3282a6222b 0x7f3282a621f5 81 e2 01 00 00 00 and $0x1,%edx 0x7f3282a621fb c5 f8 92 c2 kmovw %edx,%k0 0x7f3282a621ff c5 f8 93 d0 kmovw %k0,%edx 0x7f3282a62203 81 e2 01 00 00 00 and $0x1,%edx 0x7f3282a62209 c4 41 29 ef d2 vpxor %xmm10,%xmm10,%xmm10 0x7f3282a6220e 84 d2 test %dl,%dl 0x7f3282a62210 75 19 jne 0x7f3282a6222b 0x7f3282a62212 48 98 cltq 0x7f3282a62214 62 d1 fe 08 7e 0c 01 vmovq (%r9,%rax,1),%xmm1 0x7f3282a6221b 62 d3 75 08 21 4c 01 02 20 vinsertps $0x20,0x8(%r9,%rax,1),%xmm1,%xmm1 The crash is always on line 7, "kmovb %r15d,%k0", which is an AVX-512 instruction.