[PATCH v6 3/3] Cygwin: signal: Do not handle signal when __SIGFLUSHFAST is sent

[Corinna Vinschen]

On Jan 21 23:25, Takashi Yano wrote:
> On Tue, 21 Jan 2025 14:45:10 +0100
> Corinna Vinschen wrote:
> > This bugs me a bit.  While your solution nicely wraps the entire
> > timer problem into cygwait(), the downside is that each invocation
> > of cygwait() creates its own timer. Theoretically, given this is in a
> > loop with up to 100 iterations, you have up to 100 additional timer
> > create/destroy sequences.
> > 
> > So the question is, do you think this matters at all in this scenario,
> > given we're in a 10 ms wait state anyway?
> > 
> > If you think that's not an issue, feel free to apply the patch with
> > just the one-liner above.
> 
> Thansk for reviewing.
> cygwait (NULL, 10, cw_mask) is just waiting for resolving pipe full.
> Therefore, I think the overhead of creating and destroying a timer
> every 10 msec in the wait loop is small enough to be negligible.
> That is, the CPU load will be almost the same if we avoid it.
> 
> BTW, I'm happy if you could review also:
> [PATCH v2] Cygwin: signal: Avoid frequent TLS lock/unlock for SIGCONT processing

Will do tomorrow.  I'm a bit low on spare time ATM, sorry.


Corinna