[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[PATCH for-6.1 0/2] accel/tcg: Fix hang when running in icount mode

From: Peter Maydell
Subject: [PATCH for-6.1 0/2] accel/tcg: Fix hang when running in icount mode
Date: Sun, 25 Jul 2021 18:44:03 +0100

This patchset fixes the intermittent hang seen when running a guest in
icount mode, as reported in
  https://gitlab.com/qemu-project/qemu/-/issues/499 .

The underlying cause of the hang is that code in cpu_loop_exec_tb()
was using CF_COUNT_MASK as the maximum possible number of instructions
it would try to execute from a TB when it set the icount_decr.u16.low
field. This is wrong, because (a) that field can validly be set to any
unsigned 16-bit integer and (b) now that CF_COUNT_MASK has been
reduced to 511 in commit 78ff82bb1b67c0d7, it might be less than the
number of insns in the TB.

Patch one fixes cpu_loop_exec_tb() to use the actual maximum valid
value for icount_decr.u16.low, which is 0xffff.  Patch two adjusts the
"should we ask for a TB with exactly this many insns in it?" condition
so that instead of testing "cpu->icount_extra == 0", which should be
always true if (insns_left > 0 && insns_left < tb->icount), we assert
it instead.  This assertion would have caught the bug fixed in patch

Tested using the same iterating loop test described in the bug report;
without the fix QEMU hangs within a handful of iterations. With the
fix it managed 175 successful iterations before I got bored and hit ^C.

-- PMM

Peter Maydell (2):
  accel/tcg: Don't use CF_COUNT_MASK as the max value of
  accel/tcg: Remove unnecessary check on icount_extra in

 accel/tcg/cpu-exec.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)


reply via email to

[Prev in Thread] Current Thread [Next in Thread]