On 2023-10-30 09:45, Steven Rostedt wrote:
On Mon, 30 Oct 2023 08:56:50 -0400
Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx> wrote:
This only works if "your" lock implementation is the only user of this
RSEQ feature within a process. RSEQ requires that multiple libraries can
share the facilities. Therefore, the rseq field should include the
nesting counter as part of the RSEQ ABI so various userspace libraries
can use it collaboratively.
Then I would suggest allowing bit 31 be an "on/off" switch, and the lower
bits to be a counter. When I first tried this with postgres, there was one
lwlock that looked to be held for the majority of the run, so I got rid of
the nesting. But I think a mixture of both would work, where you can have a
nesting counter and an on/off switch with the caveat that if you use it and
enable it, another library may disable it.
If you have the nesting counter, why do you need the explicit on/off
switch ?
Thanks,
Mathieu
--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com
