|
From: | Richard Henderson |
Subject: | Re: [PATCH 07/13] cpu: Move common code to cpu-common |
Date: | Wed, 9 Feb 2022 09:29:35 +1100 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101 Thunderbird/91.5.0 |
On 2/9/22 02:22, Philippe Mathieu-Daudé wrote:
-void cpu_abort(CPUState *cpu, const char *fmt, ...) -{ - va_list ap; - va_list ap2; - - va_start(ap, fmt); - va_copy(ap2, ap); - fprintf(stderr, "qemu: fatal: "); - vfprintf(stderr, fmt, ap); - fprintf(stderr, "\n"); - cpu_dump_state(cpu, stderr, CPU_DUMP_FPU | CPU_DUMP_CCOP); - if (qemu_log_separate()) { - FILE *logfile = qemu_log_lock(); - qemu_log("qemu: fatal: "); - qemu_log_vprintf(fmt, ap2); - qemu_log("\n"); - log_cpu_state(cpu, CPU_DUMP_FPU | CPU_DUMP_CCOP); - qemu_log_flush(); - qemu_log_unlock(logfile); - qemu_log_close(); - } - va_end(ap2); - va_end(ap); - replay_finish(); -#if defined(CONFIG_USER_ONLY) - { - struct sigaction act; - sigfillset(&act.sa_mask); - act.sa_handler = SIG_DFL; - act.sa_flags = 0; - sigaction(SIGABRT, &act, NULL); - } -#endif - abort(); -}
This ifdef won't work moved to cpus-common.c.Coincidentally, Alex and I were a bit stumped early this morning about a g_assert_not_reached() failing to kill a plugin example, and I had totally forgotten about this.
We should probably rearrange user-only to *not* trap SIGABRT in general; only SIGABRT that can be raised by the guest. Perhaps the signal number should be remapped in the guest, to one of the host rt signals? Then if the signal makes its way to the SIG_DFL handler, we remap to a host abort().
r~
[Prev in Thread] | Current Thread | [Next in Thread] |