On a pentium3, libQt5core.so is installed successfully, and then exhibits a runtime failure (illegal instruction) first exposed when building qtwidgets as uic is linked against it and executed during the build. Temporary workaround by editing the ebuilds to disable automatic cpu feature detection for both packages. local myconf=( + -no-sse2 + -no-sse3 + -no-sse4.1 + -no-sse4.2 + -no-avx + -no-avx2 + -no-avx512 Upstream bug? See bug 552942 for similar issue on old version.
FYI: This isn't high priority for me - I don't actually use this machine and (unrelated) I think its hard drive may fail soon. But this looks like it is likely to have similar trouble on any non-SSE2 32-bit machine when compiled with gcc 6.4.0-r1, so I'll report what I've found: I see the same illegal instruction thing on my dual processor athlon-mp machine (-march=athlon-mp), as well as when trying to work around it with (-march=native) or brute force (-march=pentium and empty CPU_FLAGS_X86). I rebuild qtcore and attempt to build qtwidgets in each case, and in each case uic dies in qHash(QByteArray const&, unsigned int) inside /usr/lib/libQt5Core.so.5. After the failure, if I run the generated uic manually in gdb without arguments (after examining and bypassing uic_wrapper.sh) I get something like the following: ========================= [SNIP] # gdb /var/tmp/portage/dev-qt/qtwidgets-5.9.4-r1/work/qtbase-opensource-src-5.9.4/bin/uic GNU gdb (Gentoo 7.12.1 vanilla) 7.12.1 Copyright (C) 2017 Free Software Foundation, Inc. License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html> This is free software: you are free to change and redistribute it. There is NO WARRANTY, to the extent permitted by law. Type "show copying" and "show warranty" for details. This GDB was configured as "i486-pc-linux-gnu". Type "show configuration" for configuration details. For bug reporting instructions, please see: <https://bugs.gentoo.org/>. Find the GDB manual and other documentation resources online at: <http://www.gnu.org/software/gdb/documentation/>. For help, type "help". Type "apropos word" to search for commands related to "word"... Reading symbols from /var/tmp/portage/dev-qt/qtwidgets-5.9.4-r1/work/qtbase-opensource-src-5.9.4/bin/uic...(no debugging symbols found)...done. (gdb) run Starting program: /var/tmp/portage/dev-qt/qtwidgets-5.9.4-r1/work/qtbase-opensource-src-5.9.4/bin/uic [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib/libthread_db.so.1". Program received signal SIGILL, Illegal instruction. 0xb7b68d01 in qHash(QByteArray const&, unsigned int) () from /usr/lib/libQt5Core.so.5 (gdb) where #0 0xb7b68d01 in qHash(QByteArray const&, unsigned int) () from /usr/lib/libQt5Core.so.5 #1 0xb7d209e9 in QTextCodec::codecForName(QByteArray const&) () from /usr/lib/libQt5Core.so.5 #2 0xb7d20cdb in QTextCodec::codecForLocale() () from /usr/lib/libQt5Core.so.5 #3 0xb7b95cac in QString::fromLocal8Bit_helper(char const*, int) () from /usr/lib/libQt5Core.so.5 #4 0xb7cc88e8 in QCoreApplicationPrivate::appName() const () from /usr/lib/libQt5Core.so.5 #5 0xb7ccfea4 in QCoreApplicationPrivate::init() () from /usr/lib/libQt5Core.so.5 #6 0xb7cd0280 in QCoreApplication::QCoreApplication(int&, char**, int) () from /usr/lib/libQt5Core.so.5 #7 0x8006b258 in runUic(int, char**) () #8 0x8000b248 in main () (gdb) disassemble Dump of assembler code for function _Z5qHashRK10QByteArrayj: 0xb7b68cd0 <+0>: push %ebp 0xb7b68cd1 <+1>: push %edi 0xb7b68cd2 <+2>: push %esi 0xb7b68cd3 <+3>: push %ebx 0xb7b68cd4 <+4>: sub $0x1c,%esp 0xb7b68cd7 <+7>: call 0xb7b14a90 0xb7b68cdc <+12>: add $0x457324,%ebx 0xb7b68ce2 <+18>: mov 0x30(%esp),%eax 0xb7b68ce6 <+22>: mov 0x34(%esp),%ecx 0xb7b68cea <+26>: mov (%eax),%ebp 0xb7b68cec <+28>: mov 0xc(%ebp),%eax 0xb7b68cef <+31>: mov 0x4(%ebp),%edi 0xb7b68cf2 <+34>: add %eax,%ebp 0xb7b68cf4 <+36>: mov -0x25c(%ebx),%eax 0xb7b68cfa <+42>: mov %eax,(%esp) 0xb7b68cfd <+45>: movq (%eax),%xmm0 => 0xb7b68d01 <+49>: movq %xmm0,0x8(%esp) 0xb7b68d07 <+55>: mov 0xc(%esp),%edx 0xb7b68d0b <+59>: mov 0x8(%esp),%eax 0xb7b68d0f <+63>: mov %edx,%esi 0xb7b68d11 <+65>: or %eax,%esi 0xb7b68d13 <+67>: je 0xb7b68d60 <_Z5qHashRK10QByteArrayj+144> 0xb7b68d15 <+69>: xor %edx,%edx 0xb7b68d17 <+71>: and $0x100000,%eax 0xb7b68d1c <+76>: mov %edx,%esi 0xb7b68d1e <+78>: or %eax,%esi 0xb7b68d20 <+80>: jne 0xb7b68d50 <_Z5qHashRK10QByteArrayj+128> 0xb7b68d22 <+82>: test %edi,%edi 0xb7b68d24 <+84>: je 0xb7b68d80 <_Z5qHashRK10QByteArrayj+176> 0xb7b68d26 <+86>: add %ebp,%edi 0xb7b68d28 <+88>: mov %ecx,%eax 0xb7b68d2a <+90>: lea 0x0(%esi),%esi 0xb7b68d30 <+96>: mov %eax,%ecx 0xb7b68d32 <+98>: inc %ebp 0xb7b68d33 <+99>: shl $0x5,%ecx 0xb7b68d36 <+102>: sub %eax,%ecx 0xb7b68d38 <+104>: mov %ecx,%eax 0xb7b68d3a <+106>: xor %ecx,%ecx 0xb7b68d3c <+108>: mov -0x1(%ebp),%cl 0xb7b68d3f <+111>: add %ecx,%eax 0xb7b68d41 <+113>: cmp %ebp,%edi 0xb7b68d43 <+115>: jne 0xb7b68d30 <_Z5qHashRK10QByteArrayj+96> 0xb7b68d45 <+117>: add $0x1c,%esp 0xb7b68d48 <+120>: pop %ebx 0xb7b68d49 <+121>: pop %esi 0xb7b68d4a <+122>: pop %edi 0xb7b68d4b <+123>: pop %ebp 0xb7b68d4c <+124>: ret 0xb7b68d4d <+125>: lea 0x0(%esi),%esi 0xb7b68d50 <+128>: add $0x1c,%esp 0xb7b68d53 <+131>: mov %edi,%edx 0xb7b68d55 <+133>: mov %ebp,%eax 0xb7b68d57 <+135>: pop %ebx 0xb7b68d58 <+136>: pop %esi 0xb7b68d59 <+137>: pop %edi 0xb7b68d5a <+138>: pop %ebp 0xb7b68d5b <+139>: jmp 0xb7b68940 0xb7b68d60 <+144>: mov %ecx,0x4(%esp) 0xb7b68d64 <+148>: call 0xb7b8c770 <_Z18qDetectCpuFeaturesv> 0xb7b68d69 <+153>: mov (%esp),%eax 0xb7b68d6c <+156>: movq (%eax),%xmm1 0xb7b68d70 <+160>: mov 0x4(%esp),%ecx 0xb7b68d74 <+164>: movq %xmm1,0x8(%esp) 0xb7b68d7a <+170>: mov 0x8(%esp),%eax 0xb7b68d7e <+174>: jmp 0xb7b68d15 <_Z5qHashRK10QByteArrayj+69> 0xb7b68d80 <+176>: mov %ecx,%eax 0xb7b68d82 <+178>: jmp 0xb7b68d45 <_Z5qHashRK10QByteArrayj+117> End of assembler dump. (gdb) ========================= [SNIP] This file was compiled for package qtcore with (from log): i486-pc-linux-gnu-g++ -c -o qhash.o -march=pentium -O2 -pipe -fomit-frame-pointer -fno-stack-protector -std=c++11 -ffunction-sections -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/qmake -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/qmake/library -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/qmake/generators -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/qmake/generators/unix -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/qmake/generators/win32 -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/qmake/generators/mac -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/include -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/include/QtCore -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/include/QtCore/5.9.4 -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/include/QtCore/5.9.4/QtCore -I../src/corelib/global -I/var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/mkspecs/linux-g++ -DQT_VERSION_STR=\"5.9.4\" -DQT_VERSION_MAJOR=5 -DQT_VERSION_MINOR=9 -DQT_VERSION_PATCH=4 -DQT_BUILD_QMAKE -DQT_BOOTSTRAPPED -DPROEVALUATOR_FULL -DQT_NO_FOREACH /var/tmp/portage/dev-qt/qtcore-5.9.4-r2/work/qtbase-opensource-src-5.9.4/src/corelib/tools/qhash.cpp From the source code ( /var/tmp/portage/dev-qt/qtwidgets-5.9.4-r1/work/qtbase-opensource-src-5.9.4/src/corelib/tools/qhash.cpp ), the logic is split among various inline functions/defines/ifdefs/etc, but looks like depending on various #defines, it will typically try to dynamically detect if SSE4_2 is enabled, and either use some code based around an SSE4_2 CRC32 instruction, or fallback on the simple loop between +96 and +115 in the assembly listing. Parts of the listing varies by -march, but the bad "movq (%eax),%xmm0" instruction (SSE2, I think?) is always present. Speculation: I'm not sure, but maybe gcc (6.4.0-r1) is getting confused by calling a function that has SSE4_2 explicitly enabled (assuming that's what "QT_FUNCTION_TARGET(SSE4_2)" means), and allowing SSE2 instructions even in paths that don't call the SSE4_2 function? If so, that would probably imply this as a gcc bug. Or maybe Qt is erroneously detecting SSE2 and setting something in a configuration header file that is overriding the -march argument somehow? I haven't tried DKT's workaround (local ebuild change), but it looks likely to ifdef out the dynamic CPU tests and SSE4_2 stuff completely, which I could believe would also avoid the SSE2 movq.
Created attachment 530282 [details] emerge --info qtcore
Having reviewed this in more detail, this is definitely a duplicate of the issue in bug 552942, and the fix for that bug no longer works correctly. SSE is forcibly enabled by default for QT5 if the compiler supports it. The fix for bug 552942 in qt5-build.eclass could be updated to disable sse2 if the cpu does not support it, otherwise the resulting libraries will fail on non SSE2 processors. This appears to be a deliberate upstream policy decision, and they rely on distributions to either provide both sse2 and non-sse2 library, or explicitly build/configure with -no-sse2 when needed. There are no apparent runtime checks for SSE2 support in the cpu. https://lists.qt-project.org/pipermail/development/2013-November/014085.html qtbase-everywhere-src-5.11.1/mkspecs/features/qt_module.prf sse2:!contains(QT_CPU_FEATURES.$$QT_ARCH, sse2):!host_build:!if(static:qtConfig(shared)) { # If the compiler supports SSE2, enable it unconditionally in all of Qt shared libraries # (and only the libraries). This is not expected to be a problem because: # - on Windows, sharing of libraries is uncommon # - on Mac OS X, all x86 CPUs already have SSE2 support (we won't even reach here) # - on Linux, the dynamic loader can find the libraries on LIBDIR/sse2/ # The last guarantee does not apply to executables and plugins, so we can't enable for them. QT_CPU_FEATURES.$$QT_ARCH += sse sse2 QMAKE_CFLAGS += $$QMAKE_CFLAGS_SSE2 QMAKE_CXXFLAGS += $$QMAKE_CFLAGS_SSE2 } Updated workaround below in the ebuild for qtcore is sufficient to prevent sse2 support being forced on by QT for the older machine. I recognize that a minority are still running older non-sse2 processors, but we do exist! src_configure() { local myconf=( + -no-sse2 $(qt_use icu) $(qt_use !icu iconv) $(qt_use systemd journald) ) qt5-build_src_configure }
*** This bug has been marked as a duplicate of bug 648004 ***
