can you reproduce the bug with baseq3?

A disassembly would help

on baseq3 right after /devmap in all maps I tried.


Program received signal SIGSEGV, Segmentation fault.
build_ycc_rgb_table (cinfo=<value optimized out>) at code/jpeg-6b/jdcolor.c:94
94          cconvert->Cr_r_tab[i] = (int)
(gdb) bt
#0  build_ycc_rgb_table (cinfo=<value optimized out>)
    at code/jpeg-6b/jdcolor.c:94
#1  0x0050917e in jinit_color_deconverter (cinfo=0x229738)
    at code/jpeg-6b/jdcolor.c:362
#2  0x0051132e in jinit_master_decompress (cinfo=0x229738)
    at code/jpeg-6b/jdmaster.c:367
#3  0x0050724b in jpeg_start_decompress (cinfo=0x229738)
    at code/jpeg-6b/jdapistd.c:42
#4  0x0052f42c in R_LoadJPG (filename=0xa02c00 "textures/sfx/logo512.jpg", 
    pic=0x229a3c, width=0x229a44, height=0x229a40)
    at code/renderer/tr_image_jpg.c:110
#5  0x00528b24 in R_LoadImage (name=0xa14600 "textures/sfx/logo512.tga", 
    pic=0x229a3c, width=0x229a44, height=0x229a40)
    at code/renderer/tr_image.c:906
#6  0x0052c190 in R_FindImageFile (name=0xa14600 "textures/sfx/logo512.tga", 
    mipmap=qfalse, allowPicmip=qfalse, glWrapClampMode=10497)
    at code/renderer/tr_image.c:966
#7  0x00549353 in ParseStage (stage=0xabde40, text=0x22a004)
    at code/renderer/tr_shader.c:660
#8  0x0054c9db in R_FindShader (name=0x62ff041 "menuback", lightmapIndex=-4, 
    mipRawImage=qfalse) at code/renderer/tr_shader.c:1437
#9  0x0054dd8f in RE_RegisterShaderNoMip (name=0x62ff041 "menuback")
    at code/renderer/tr_shader.c:2767
#10 0x00426353 in CL_UISystemCalls (args=0x63f8b18) at code/client/cl_ui.c:801
#11 0x0055d136 in CallAsmCall (syscallNum=20, programStack=1048404, 
    opStack=0x22a174) at code/qcommon/vm_x86.c:193
#12 0x0055d108 in AsmCall ()
#13 0x00000014 in ?? ()
#14 0x000fff54 in ?? ()
#15 0x0022a174 in ?? ()
#16 0x00000002 in ?? ()
#17 0x1c7ee3f2 in ?? ()
#18 0x0055d0f4 in AsmCall ()
#19 0x1c7aec96 in ?? ()
#20 0x0055d0f4 in AsmCall ()
#21 0x1c7900fd in ?? ()
#22 0x0055d24d in VM_CallCompiled (vm=0x1dd4598, args=0x22b204)
    at code/qcommon/vm_x86.c:1192
#23 0x00493c96 in VM_Call (vm=0x1dd4598, callnum=1) at code/qcommon/vm.c:771
#24 0x004253ea in CL_InitUI () at code/client/cl_ui.c:1123
#25 0x004204e9 in CL_DownloadsComplete () at code/client/cl_main.c:1878
#26 0x0042064c in CL_InitDownloads () at code/client/cl_main.c:2046
#27 0x00423375 in CL_ParseGamestate (msg=0x22f848)
    at code/client/cl_parse.c:532
#28 0x00423fa8 in CL_ParseServerMessage (msg=0x22f848)
    at code/client/cl_parse.c:894
#29 0x00422531 in CL_PacketEvent (from=
      {type = NA_LOOPBACK, ip = "\000\000\000", ip6 = '\0' <repeats 15 times>, port = 0, scope_id = 0}, msg=0x22f848) at code/client/cl_main.c:2486
#30 0x00443803 in Com_EventLoop () at code/qcommon/common.c:2208
#31 0x00443b3b in Com_Frame () at code/qcommon/common.c:2935
#32 0x0055cc69 in SDL_main (argc=4, argv=0x383158) at code/sys/sys_main.c:571
#33 0x00564ffb in console_main (argc=4, argv=0x383158)
    at ./src/main/win32/SDL_win32_main.c:217
#34 0x005651d9 in WinMain (hInst=0x400000, hPrev=0x0, 
    szCmdLine=0x22a2b4f "+set fs_basepath d:/games/UrbanTerror", sw=10)
    at ./src/main/win32/SDL_win32_main.c:353
#35 0x00564a58 in main ()
(gdb)

An alignment issue I guess. run "disassemble $eip $eip+16" in gdb to see which instruction causes the segv. I tried fixing the stack alignment in the compiled vm but that makes the problem only go away on some systems. Just don't use optimizations other than the default ones if you don't know how to fix problems.

Program received signal SIGSEGV, Segmentation fault.
R_SubdividePatchToGrid (width=3, height=3, points=0x21ef98)
    at code/renderer/tr_curve.c:411
411                                     VectorNormalize( dir );
(gdb) disassemble $eip $eip+16
Dump of assembler code from 0x523217 to 0x523227:
0x00523217 <R_SubdividePatchToGrid+1463>:       movaps %xmm0,-0x2db58(%ebp)
0x0052321e <R_SubdividePatchToGrid+1470>:       flds   0x8(%ebx)
0x00523221 <R_SubdividePatchToGrid+1473>:       flds   (%ebx)
0x00523223 <R_SubdividePatchToGrid+1475>:       fadds  (%esi)
0x00523225 <R_SubdividePatchToGrid+1477>:       flds   (%edi)
End of assembler dump.
(gdb) 


and on baseq3

Program received signal SIGSEGV, Segmentation fault.
build_ycc_rgb_table (cinfo=<value optimized out>) at code/jpeg-6b/jdcolor.c:94
94          cconvert->Cr_r_tab[i] = (int)
(gdb) disassemble $eip $eip+16
Dump of assembler code from 0x50839a to 0x5083aa:
0x0050839a <build_ycc_rgb_table+2362>:  movdqa %xmm7,-0x28(%ebp)
0x0050839f <build_ycc_rgb_table+2367>:  movdqa %xmm3,%xmm6
0x005083a3 <build_ycc_rgb_table+2371>:  movdqa %xmm4,%xmm7
0x005083a7 <build_ycc_rgb_table+2375>:  psrldq $0x4,%xmm6
End of assembler dump.
(gdb)

See if this fixes your problem:

Index: code/qcommon/vm_x86.c
===================================================================
--- code/qcommon/vm_x86.c
+++ code/qcommon/vm_x86.c
@@ -216,12 +216,17 @@
 	"0:\n\t" // system call
 	"notl  %eax\n\t"
 	"pushl %ecx\n\t"
+	"movl  %esp, %ecx\n\t"
+	"andl  $-16, %esp\n\t"
+	"pushl %ecx\n\t"
 	"pushl %edi\n\t" // opStack
 	"pushl %esi\n\t" // programStack
 	"pushl %eax\n\t" // syscallNum
 	"call  " CMANG(CallAsmCall) "\n\t"
 	"addl  $12, %esp\n\t"
 	"popl  %ecx\n\t"
+	"movl  %ecx, %esp\n\t"
+	"popl  %ecx\n\t"
 	"addl  $4, %edi\n\t"
 	"ret\n\t"
 #if defined __ELF__

It worked.

Patch fixed it.

That patch is by gimhael (Matthias Bentrup). It realigns the stack to 16 bytes on syscalls, which is what gcc expects.

(In reply to comment #8)
> That patch is by gimhael (Matthias Bentrup). It realigns the stack to 16 bytes
> on syscalls, which is what gcc expects.

Yeah, that's what I meant. Game still segfaults when compiled on openSUSE 11.0 though but doesn't on 11.1. Weird. I've commited the stack alignment fix anyways.