Commit 27f347e6a1 for qemu.org

commit 27f347e6a1d269c533633c812321cabb249eada8
Author: Alex Bennée <alex.bennee@linaro.org>
Date:   Thu Jan 16 16:02:51 2025 +0000

    accel/tcg: also suppress asynchronous IRQs for cpu_io_recompile

    While it would be technically correct to allow an IRQ to happen (as
    the offending instruction never really completed) it messes up
    instrumentation. We already take care to only use memory
    instrumentation on the block, we should also suppress IRQs.

    Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
    Reviewed-by: Julian Ganz <neither@nut.email>
    Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
    Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
    Message-Id: <20250116160306.1709518-23-alex.bennee@linaro.org>

diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
index 453eb20ec9..d56ca13cdd 100644
--- a/accel/tcg/translate-all.c
+++ b/accel/tcg/translate-all.c
@@ -633,9 +633,10 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)
      * Exit the loop and potentially generate a new TB executing the
      * just the I/O insns. We also limit instrumentation to memory
      * operations only (which execute after completion) so we don't
-     * double instrument the instruction.
+     * double instrument the instruction. Also don't let an IRQ sneak
+     * in before we execute it.
      */
-    cpu->cflags_next_tb = curr_cflags(cpu) | CF_MEMI_ONLY | n;
+    cpu->cflags_next_tb = curr_cflags(cpu) | CF_MEMI_ONLY | CF_NOIRQ | n;

     if (qemu_loglevel_mask(CPU_LOG_EXEC)) {
         vaddr pc = cpu->cc->get_pc(cpu);