Hi, this is the fix for an issue I complained about a few days back on the IRC. Unfortunately my reproducer [1] does not work anymore, so I'm sending this separately and as an RFC.
The user-visible effect was that: - If a TB writes to itself; - The address of the write is before the TB start; - The newly written instruction causes an exception; - Then single-stepping the instruction that does a write gets you to an interrupt handler and not to the newly written "bad" instruction. You can still see that something is wrong with the softmmu reproducer in the debugger though. In helper_vstl(), probe_write_access() does not lead to cpu_loop_exit_noexc() - which I think it should, but cpu_stq_be_data_ra() still does, masking the issue. Best regards, Ilya [1] https://lists.gnu.org/archive/html/qemu-devel/2023-08/msg01081.html Ilya Leoshkevich (1): tcg: Always pass the full write size to notdirty_write() accel/tcg/cputlb.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) -- 2.41.0
