diff mbox series

[v2] target/arm: drop CF_LAST_IO/dc->condjump check

Message ID 20210416170207.12504-1-alex.bennee@linaro.org (mailing list archive)
State New
Headers show
Series [v2] target/arm: drop CF_LAST_IO/dc->condjump check | expand

Commit Message

Alex Bennée April 16, 2021, 5:02 p.m. UTC
This is a left over erroneous check from the days front-ends handled
io start/end themselves. Regardless just because IO could be performed
on the last instruction doesn't obligate the front end to do so.

This fixes an abort faced by the aspeed execute-in-place support which
will necessarily trigger this state (even before the one-shot
CF_LAST_IO fix). The test still seems to hang once it attempts to boot
the Linux kernel but I suspect this is an unrelated issue with icount
and the timer handling code.

The original intention of the cpu_abort (added in commit 2e70f6efa8b9
when the icount stuff was first added) seems to have been to act as
an assert() to catch an unhandled corner case where the generated code
would be something like:
    conditional branch to condlabel if its cc failed
    implementation of the insn (a conditional branch or trap)
    code emitted by gen_io_end()
 condlabel:
    gen_goto_tb or equivalent thing to go to next insn

At runtime the cc-failed case would skip over the code emitted by
gen_io_end(), leaving the can_do_io flag incorrectly set.

In commit ba3e7926691ed33 we switched to an implementation which
always clears can_do_io at the start of the following TB instead
of trying to clear it at the end of a TB that did IO. So the corner
case that this cpu_abort() was trying to flag is no longer possible,
because the gen_io_end() call has been deleted. We can therefore
safely remove the no-longer-valid assertion.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Cc: Cédric Le Goater <clg@kaod.org>

--
v2
  - added Peter's extended analysis to the commit message
---
 target/arm/translate.c | 5 -----
 1 file changed, 5 deletions(-)
diff mbox series

Patch

diff --git a/target/arm/translate.c b/target/arm/translate.c
index 62b1c2081b..7103da2d7a 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -9199,11 +9199,6 @@  static void arm_tr_tb_stop(DisasContextBase *dcbase, CPUState *cpu)
 {
     DisasContext *dc = container_of(dcbase, DisasContext, base);
 
-    if (tb_cflags(dc->base.tb) & CF_LAST_IO && dc->condjmp) {
-        /* FIXME: This can theoretically happen with self-modifying code. */
-        cpu_abort(cpu, "IO on conditional branch instruction");
-    }
-
     /* At this stage dc->condjmp will only be set when the skipped
        instruction was a conditional branch or trap, and the PC has
        already been written.  */