On 4/28/24 15:14, Philippe Mathieu-Daudé wrote:
diff --git a/target/i386/helper.c b/target/i386/helper.c
index 48d1513a35..6c188ea94c 100644
--- a/target/i386/helper.c
+++ b/target/i386/helper.c
@@ -30,6 +30,7 @@
#include "qemu/log.h"
#ifdef CONFIG_TCG
#include "tcg/insn-start-words.h"
+#include "accel/tcg/vcpu-state.h" // ???
#endif
void cpu_sync_avx_hflag(CPUX86State *env)
@@ -518,7 +519,7 @@ static inline target_ulong get_memio_eip(CPUX86State *env)
uint64_t data[TARGET_INSN_START_WORDS];
CPUState *cs = env_cpu(env);
- if (!cpu_unwind_state_data(cs, cs->mem_io_pc, data)) {
+ if (!cpu_unwind_state_data(cs, cs->accel->mem_io_pc, data)) {
return env->eip;
}
Not good.
Although considering
https://lore.kernel.org/qemu-devel/20240416040609.1313605-3-richard.hender...@linaro.org/
we should probably merge mem_io_pc and plugin_ra. So the value to be accessed by i386
would not need to be in a tcg private data structure, but in CPUNegativeOffsetState.
r~
