** Description changed: [Info] - Ghostty, a terminal emulator written in Zig, was first packaged for resolute. By default, the Zig compiler builds an application binary using all the available native CPU features. This means Ghostty binaries built on Launchpad builders include AVX and AVX2 instructions, making them incompatible with older x86_64 variants that do not include AVX. This manifests as a SIGILL crash on startup on those variants. This is a serious impediment to Ubuntu users wanting to install and use Ghostty on non-AVX variants. + Ghostty, a terminal emulator written in Zig, was first packaged for resolute. By default, the Zig compiler builds an application binary using native CPU features. This means Ghostty binaries built on Launchpad builders include AVX and AVX2 instructions, making them incompatible with older x86_64 variants without AVX. This manifests as a SIGILL crash on startup on those architectures. A simple solution to this problem is the use of the Zig compiler option - “-Dcpu=baseline”, which instructs Zig to compile for the architecture + “-Dcpu=baseline”, which tells the Zig to compile for the architecture baseline. - [Test plan] - A user running Ubuntu 26.04 on an older, non-AVX variant, should simply be able to install the Ghostty package ( version 1.3.0~us1-0ubuntu1) and reproduce the issue by simply clicking on the desktop launcher, or by invoking it from resolute’s default ptyxis terminal: + A user running Ubuntu 26.04 on an older, non-AVX variant, should be able to install the Ghostty package ( version 1.3.0~us1-0ubuntu1) and reproduce the issue by simply clicking on the desktop launcher, or by invoking ghostty from resolute’s default terminal: $ sudo apt update && sudo apt install ghostty … … $ ghostty Illegal instruction (core dumped) ghostty - One may also use qemu-x86_64 with a non-AVX cpu emulation to reproduce - the crash. For example: + On modern CPUs, one may use qemu-x86_64 with a non-AVX cpu emulation to + reproduce the crash. For example: # Nehalem is a cpu model without AVX support $ qemu-x86_64 -cpu Nehalem /usr/bin/ghostty +version qemu: uncaught target signal 4 (Illegal instruction) - core dumped # Haswell has AVX, but we can disable it to reproduce $ qemu-x86_64 -cpu Haswell,-avx,-avx2 /usr/bin/ghostty +version qemu: uncaught target signal 4 (Illegal instruction) - core dumped - Illegal instruction (core dumped) qemu-x86_64 -cpu - + Illegal instruction (core dumped) qemu-x86_64 -cpu [Where problems could occur] Functional regression possibility #1 — - Zig is a new programming language; the compiler hasn’t stabilized into a version 1.X yet. The number of real-world Zig applications/deployments is quite less, and these applications are likely to be built for, and deployed on, modern CPU models. As a result, compiling for the baseline will exercise relatively colder/infrequent code paths in the Zig compiler that are not tested enough on modern hardware. In the best case, this may lead to compiler failures. In the worst case, silent bugs may slip into the binary. + Zig is a new programming language; the compiler hasn’t stabilized to a version 1.X yet. The number of real-world Zig applications/deployments is quite less, and these applications are likely to be built for modern CPU models. As a result, compiling for the baseline will exercise rare code paths in the Zig compiler that are not tested enough on modern CPUs. In the best case, this may cause compilation failures. In the worst case, silent bugs may slip into the binary. Functional regression possibility #2 — - Zig is LLVM-based, and LLVM is a battle-tested compiler framework. However, the Zig community has a strategy to eventually move away from LLVM altogether. Such a move would further amplify the chances of possibility #1. + Zig is LLVM-based, and LLVM is a battle-tested compiler framework. However, the Zig maintainers are working towards making Zig drop this dependency. This is likely to further amplify the chances of possibility #1. Functional regression possibility #3 — - Zig code can actually test for CPU features and take a certain path if they are available vis-a-vis another path if they are not. As of version 1.3, Ghostty has no such checks for AVX,AVX2. However, if any such checks are introduced in the future, the Ubuntu binaries are likely to exercise the rare, cold paths; the upstream binaries are more likely to exercise and test the hot paths. This increases the possibility of finding regressions that upstream binaries cannot reproduce. + Zig code can test for CPU features and take a certain path if they are available vis-a-vis another path if they are not. As of version 1.3, Ghostty has no such checks for AVX,AVX2. However, if any such checks are introduced in the future, the Ubuntu binaries are likely to exercise the rare alternate paths; the upstream binaries are more likely to exercise and test the hot paths. This increases the possibility of hitting regressions that upstream binaries cannot reproduce. Performance regression possibility #1 — - Using SSE instructions on AVX-enabled architectures amounts to underutilization of the available CPU features. Unique, performance-intensive workloads on the terminal emulator may see a performance drop due to the compilation for baseline. + Using SSE instructions on AVX-enabled architectures amounts to under-utilization of the available CPU features. Performance-intensive workloads may see a performance drop due to the compilation for baseline. Functional/performance regression possibility #2 — The effect of -Dcpu=baseline for amd64v3 and arm64 is not fully understood, and it can have unforeseen consequences. [Other info] - Enforcing baseline compilation is something a future “debhelper for Zig” - must do. As of today there is no dh_zig in the archive. + Firstly, enforcing baseline compilation is something a future “debhelper + for Zig” must do. As of today there is no dh_zig. - The baseline for amd64 is different from the baseline for amd64v3. The - latter includes AVX, the former does not. The amd64v3 builds with - -Dcpu=baseline still contain AVX instructions. Consequently we had to - disable the dep8 test added for this fix on amd64v3 (LP: #2155110). + Secondly, the baselines for amd64 and amd64v3 are different. The latter + includes AVX, the former does not. The amd64v3 builds with + -Dcpu=baseline contain AVX instructions. Consequently we had to disable + the dep8 test added for this fix, on amd64v3 (LP: #2155110). ---- Ghostty on resolute crashes on start. Possibly zig uses all instruction sets available on the build machine by default? My machine here is a few years old and does not support some modern instructions (e.g. avx). Before the upgrade to resolute, I've been using a ghostty, which I had built from source on this exact machine. Worked fine. ~$ ghostty Illegal instruction (core dumped) ghostty $ sudo dmesg | tail -1 [70260.702796] traps: ghostty[20368] trap invalid opcode ip:5dc88ab52daa sp:7ffccb4a7590 error:0 in ghostty[1387daa,5dc88a9f2000+854000] $ apt policy ghostty ghostty: Installed: 1.3.0~us1-0ubuntu1 Candidate: 1.3.0~us1-0ubuntu1 Version table: *** 1.3.0~us1-0ubuntu1 500 500 http://de.archive.ubuntu.com/ubuntu resolute/universe amd64 Packages 100 /var/lib/dpkg/status $ cat /proc/cpuinfo processor : 0 vendor_id : GenuineIntel cpu family : 6 model : 60 model name : Intel(R) Pentium(R) CPU G3220 @ 3.00GHz stepping : 3 microcode : 0x28 cpu MHz : 2095.320 cache size : 3072 KB physical id : 0 siblings : 2 core id : 0 cpu cores : 2 apicid : 0 initial apicid : 0 fpu : yes fpu_exception : yes cpuid level : 13 wp : yes flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx est tm2 ssse3 sdbg cx16 xtpr pdcm pcid sse4_1 sse4_2 movbe popcnt tsc_deadline_timer xsave rdrand lahf_lm abm cpuid_fault epb pti ssbd ibrs ibpb stibp tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust erms invpcid xsaveopt dtherm arat pln pts vnmi md_clear flush_l1d vmx flags : vnmi preemption_timer invvpid ept_x_only ept_ad ept_1gb flexpriority tsc_offset vtpr mtf vapic ept vpid unrestricted_guest ple bugs : cpu_meltdown spectre_v1 spectre_v2 spec_store_bypass l1tf mds swapgs itlb_multihit srbds spectre_v2_user vmscape bogomips : 5986.67 clflush size : 64 cache_alignment : 64 address sizes : 39 bits physical, 48 bits virtual power management: ProblemType: Bug DistroRelease: Ubuntu 26.04 Package: ghostty 1.3.0~us1-0ubuntu1 ProcVersionSignature: Ubuntu 7.0.0-14.14-generic 7.0.0 Uname: Linux 7.0.0-14-generic x86_64 NonfreeKernelModules: zfs ApportVersion: 2.34.0-0ubuntu2 Architecture: amd64 CasperMD5CheckResult: unknown CurrentDesktop: ubuntu:GNOME Date: Sun Apr 19 11:45:01 2026 InstallationDate: Installed on 2019-05-24 (2522 days ago) InstallationMedia: Ubuntu 19.04 "Disco Dingo" - Release amd64 (20190416) SourcePackage: ghostty UpgradeStatus: Upgraded to resolute on 2026-04-18 (1 days ago)
** Description changed: [Info] Ghostty, a terminal emulator written in Zig, was first packaged for resolute. By default, the Zig compiler builds an application binary using native CPU features. This means Ghostty binaries built on Launchpad builders include AVX and AVX2 instructions, making them incompatible with older x86_64 variants without AVX. This manifests as a SIGILL crash on startup on those architectures. A simple solution to this problem is the use of the Zig compiler option “-Dcpu=baseline”, which tells the Zig to compile for the architecture baseline. [Test plan] A user running Ubuntu 26.04 on an older, non-AVX variant, should be able to install the Ghostty package ( version 1.3.0~us1-0ubuntu1) and reproduce the issue by simply clicking on the desktop launcher, or by invoking ghostty from resolute’s default terminal: $ sudo apt update && sudo apt install ghostty … … $ ghostty Illegal instruction (core dumped) ghostty On modern CPUs, one may use qemu-x86_64 with a non-AVX cpu emulation to reproduce the crash. For example: # Nehalem is a cpu model without AVX support $ qemu-x86_64 -cpu Nehalem /usr/bin/ghostty +version qemu: uncaught target signal 4 (Illegal instruction) - core dumped # Haswell has AVX, but we can disable it to reproduce $ qemu-x86_64 -cpu Haswell,-avx,-avx2 /usr/bin/ghostty +version qemu: uncaught target signal 4 (Illegal instruction) - core dumped Illegal instruction (core dumped) qemu-x86_64 -cpu [Where problems could occur] Functional regression possibility #1 — Zig is a new programming language; the compiler hasn’t stabilized to a version 1.X yet. The number of real-world Zig applications/deployments is quite less, and these applications are likely to be built for modern CPU models. As a result, compiling for the baseline will exercise rare code paths in the Zig compiler that are not tested enough on modern CPUs. In the best case, this may cause compilation failures. In the worst case, silent bugs may slip into the binary. Functional regression possibility #2 — Zig is LLVM-based, and LLVM is a battle-tested compiler framework. However, the Zig maintainers are working towards making Zig drop this dependency. This is likely to further amplify the chances of possibility #1. Functional regression possibility #3 — Zig code can test for CPU features and take a certain path if they are available vis-a-vis another path if they are not. As of version 1.3, Ghostty has no such checks for AVX,AVX2. However, if any such checks are introduced in the future, the Ubuntu binaries are likely to exercise the rare alternate paths; the upstream binaries are more likely to exercise and test the hot paths. This increases the possibility of hitting regressions that upstream binaries cannot reproduce. Performance regression possibility #1 — Using SSE instructions on AVX-enabled architectures amounts to under-utilization of the available CPU features. Performance-intensive workloads may see a performance drop due to the compilation for baseline. Functional/performance regression possibility #2 — The effect of -Dcpu=baseline for amd64v3 and arm64 is not fully understood, and it can have unforeseen consequences. [Other info] Firstly, enforcing baseline compilation is something a future “debhelper for Zig” must do. As of today there is no dh_zig. Secondly, the baselines for amd64 and amd64v3 are different. The latter includes AVX, the former does not. The amd64v3 builds with -Dcpu=baseline contain AVX instructions. Consequently we had to disable the dep8 test added for this fix, on amd64v3 (LP: #2155110). - ---- + ---- End of SRU template ---- Ghostty on resolute crashes on start. Possibly zig uses all instruction sets available on the build machine by default? My machine here is a few years old and does not support some modern instructions (e.g. avx). Before the upgrade to resolute, I've been using a ghostty, which I had built from source on this exact machine. Worked fine. ~$ ghostty Illegal instruction (core dumped) ghostty $ sudo dmesg | tail -1 [70260.702796] traps: ghostty[20368] trap invalid opcode ip:5dc88ab52daa sp:7ffccb4a7590 error:0 in ghostty[1387daa,5dc88a9f2000+854000] $ apt policy ghostty ghostty: Installed: 1.3.0~us1-0ubuntu1 Candidate: 1.3.0~us1-0ubuntu1 Version table: *** 1.3.0~us1-0ubuntu1 500 500 http://de.archive.ubuntu.com/ubuntu resolute/universe amd64 Packages 100 /var/lib/dpkg/status $ cat /proc/cpuinfo processor : 0 vendor_id : GenuineIntel cpu family : 6 model : 60 model name : Intel(R) Pentium(R) CPU G3220 @ 3.00GHz stepping : 3 microcode : 0x28 cpu MHz : 2095.320 cache size : 3072 KB physical id : 0 siblings : 2 core id : 0 cpu cores : 2 apicid : 0 initial apicid : 0 fpu : yes fpu_exception : yes cpuid level : 13 wp : yes flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx est tm2 ssse3 sdbg cx16 xtpr pdcm pcid sse4_1 sse4_2 movbe popcnt tsc_deadline_timer xsave rdrand lahf_lm abm cpuid_fault epb pti ssbd ibrs ibpb stibp tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust erms invpcid xsaveopt dtherm arat pln pts vnmi md_clear flush_l1d vmx flags : vnmi preemption_timer invvpid ept_x_only ept_ad ept_1gb flexpriority tsc_offset vtpr mtf vapic ept vpid unrestricted_guest ple bugs : cpu_meltdown spectre_v1 spectre_v2 spec_store_bypass l1tf mds swapgs itlb_multihit srbds spectre_v2_user vmscape bogomips : 5986.67 clflush size : 64 cache_alignment : 64 address sizes : 39 bits physical, 48 bits virtual power management: ProblemType: Bug DistroRelease: Ubuntu 26.04 Package: ghostty 1.3.0~us1-0ubuntu1 ProcVersionSignature: Ubuntu 7.0.0-14.14-generic 7.0.0 Uname: Linux 7.0.0-14-generic x86_64 NonfreeKernelModules: zfs ApportVersion: 2.34.0-0ubuntu2 Architecture: amd64 CasperMD5CheckResult: unknown CurrentDesktop: ubuntu:GNOME Date: Sun Apr 19 11:45:01 2026 InstallationDate: Installed on 2019-05-24 (2522 days ago) InstallationMedia: Ubuntu 19.04 "Disco Dingo" - Release amd64 (20190416) SourcePackage: ghostty UpgradeStatus: Upgraded to resolute on 2026-04-18 (1 days ago) ** Summary changed: - [sru] ghostty does not start: Illegal instruction + [SRU] ghostty does not start: Illegal instruction ** Also affects: ghostty (Ubuntu Resolute) Importance: Undecided Status: New ** Changed in: ghostty (Ubuntu Resolute) Status: New => In Progress -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2148769 Title: [SRU] ghostty does not start: Illegal instruction To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/ghostty/+bug/2148769/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
