Re: [PATCH 13/35] target/avr: call plugin trap callbacks

["Julian Ganz" <neither@nut.email>]

Hi Philippe,

sorry for the late reply. Did not realize I hadn't checked mails this
long.

August 17, 2026 at 1:14 PM, "Philippe Mathieu-Daudé" wrote:
> Expanding a bit ...
> 
>  77 void avr_cpu_do_interrupt(CPUState *cs)
>  78 {
>  79 CPUAVRState *env = cpu_env(cs);
>  80
>  81 uint32_t ret = env->pc_w;
>  82 int vector = 0;
>  83 int size = avr_feature(env, AVR_FEATURE_JMP_CALL) ? 2 : 1;
>  84 int base = 0;
>  85
>  86 if (cs->exception_index == EXCP_RESET) {
>  87 vector = 0;
>  88 } else if (env->intsrc != 0) {
>  89 vector = ctz64(env->intsrc) + 1;
>  90 }
> 
> Unconditionally recording exception vector as interrupt seems wrong,
> shouldn't we call qemu_plugin_vcpu_exception_cb() in that case?
> 
>  91
>  92 if (avr_feature(env, AVR_FEATURE_3_BYTE_PC)) {
>  93 do_stb(env, env->sp--, ret, 0);
>  94 do_stb(env, env->sp--, ret >> 8, 0);
>  95 do_stb(env, env->sp--, ret >> 16, 0);
>  96 } else if (avr_feature(env, AVR_FEATURE_2_BYTE_PC)) {
>  97 do_stb(env, env->sp--, ret, 0);
>  98 do_stb(env, env->sp--, ret >> 8, 0);
>  99 } else {
> 100 do_stb(env, env->sp--, ret, 0);
> 101 }
> 102
> 103 env->pc_w = base + vector * size;
> 104 env->sregI = 0; /* clear Global Interrupt Flag */
> 105
> 106 cs->exception_index = -1;
> 107
> 108 qemu_plugin_vcpu_interrupt_cb(cs, ret);
> 109 }

Yes. As should be apparent from _some_ commit message, I wrote this
under the impression that AVR did not have exceptions. Or rather, what
we call exceptions in the discon API, which are synchronous events or
traps resulting from specific instructions being executed (including
things such as invalid instructios and instruction load failures).

We _do_ want to make sure to call qemu_plugin_vcpu_interrupt_cb only for
interrupts. At the same time, we don't necessarily want to call
qemu_plugin_vcpu_exceptions_cb either, at least if AVR just resets on an
illegal instructions.

Regards,
Julian