Re: [PATCH] Fix signal handler to detect crashes in qemu-linux-user

[Richard Henderson <richard.henderson@linaro.org>]

On 8/12/23 09:43, Helge Deller wrote:
> +/* _init and _fini are provided by the linker */
> +extern char _init;
> +extern char _fini;
> +
>   static void host_signal_handler(int host_sig, siginfo_t *info, void *puc)
>   {
>       CPUArchState *env = thread_cpu->env_ptr;
> @@ -819,6 +848,13 @@ static void host_signal_handler(int host_sig, siginfo_t *info, void *puc)
>           if (host_sig == SIGSEGV) {
>               bool maperr = true;
> 
> +            /* Did segfault happened in qemu source code? */
> +            if ((pc >= (uintptr_t) &_init && pc < (uintptr_t) &_fini) ||
> +                (TARGET_ABI_BITS == 32 && HOST_LONG_BITS == 64
> +                 && !h2g_valid(host_addr))) {
> +                qemu_show_backtrace(info);
> +            }

This is incorrect.

I did mention that you should look at adjust_signal_pc,
which has a lot of commentary on the subject.

(0) Place this after the write-protect and restart check
just below, since that is not an error of any sort.

(1) If helper_retaddr == 1, then this is a fault reading
for translation and is a guest SIGSEGV.

(2) If helper_retaddr != 0, then this is a fault in some
cpu_ldst.h operation and is a guest SIGSEGV.

(3) If in_code_gen_buffer(host_signal_pc()), then this
is a fault within generated code and is a guest SIGSEGV.

Otherwise it's a host SIGSEGV.


r~