On 2023-10-25 15:19, Steven Rostedt wrote:
On Wed, 25 Oct 2023 14:49:44 -0400
Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx> wrote:
No, I wouldn't say it's the same as priority inheritance, which is to help
with determinism and not performance. PI adds overhead but removes
unbounded latency. On average, a non PI mutex is faster than PI mutex, but
can suffer from unbounded priority inversion.
AFAIU, this is because PI mutex as implemented by sys futex only boosts the
priority of the lock owner. In my proposal here the owner would be able to
borrow scheduler slices from the waiters as well.
I would be worried that that could cause even more scheduling disruption.
Now we are taking time slices from other CPUs to run the current one?
AFAIU, as we look at the time slices of the waiting tasks, those have meaning
in the context of the runqueue they are currently attached to. We first consume
time slices from tasks sharing the same runqueue as the owner task. Else, I
guess we'd need to migrate the owner task to the runqueue where the time slice
is available so we can use it (but only if migration of the owner task is
allowed to that target cpu).
Thanks,
Mathieu
--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com
