Hi,
This is an alternative approach to fixing the hang that Emilio zeroed
in on with:
https://lists.nongnu.org/archive/html/qemu-devel/2017-06/msg03224.html
Instead of forcing the front-end to treat any MSRs differently we
shortcut the lookup_tb_ptr by checking for icount_decr and
cpu->interrupt_request conditions.
Fundamentally the problem was that an interrupt was pending
(interrupt_request was set) but the "msr daifclr" operations when the
kernel did local_irq/fiq_enable() never got handled because the
cpu_idle loop was being very efficiently chained. As a result we never
got around to exiting the TCG code and calling arm_cpu_do_interrupt
which would then raise the IRQ to move things on.
Emilio's fix is also correct - we should exit the loop whenever the
IRQ conditions may have changed. However by checking in the lookup_ptr
function we avoid churn in figuring out all the other cases in the
front ends. This may have a potential cost for code with lots of
calculated jumps although I would argue its fairly minimal given we've
already sucked up the cost of a helper function and I don't think the
difference between the helper function and a full exit is that
marginal.
I've also included Thomas's thread fix as it has yet to be merged.
I humbly submit my patches to the TCG gods to decide which is the best
approach ;-)
Alex Bennée (2):
tcg-runtime: light re-factor of lookup_tb_ptr
tcg-runtime: short-circuit lookup_tb_ptr on IRQs
Thomas Huth (1):
vl: Fix broken thread=xxx option of the --accel parameter
tcg-runtime.c | 52 +++++++++++++++++++++++++++++++---------------------
vl.c | 13 +++++--------
2 files changed, 36 insertions(+), 29 deletions(-)