Sepherosa Ziehau has implemented direct input support for polling, which will affect you most directly if you have an 10G ix(4) card. His commit message lists the performance improvements.
5 Replies to “Direct input support for polling”
Comments are closed.
I’m confused because this commit message below, Sepherosa says to NOT use direct input.
http://lists.dragonflybsd.org/pipermail/commits/2017-September/626407.html
Anonymous, you left out a very important piece of the commit message:
“…in critical section”
Full commit message is:
git: polling: Don’t do direct input in critical section.
Which is drastically different than:
Don’t do direct input.
Although a quick look at the patch referenced in your link, I think Sepherosa added to the polling so that higher priority interrupt sources don’t get starved.
How common is it to write code for “situation [that] should not be possible to even get into”, like the commit message states below:
http://lists.dragonflybsd.org/pipermail/commits/2017-September/626484.html
@flavio
i agree. definitely seems odd to include code in your codebase that should never be possible to run.
Sometimes it just means you haven’t thought of all the ways it could be run. “should never be possible” is not “impossible”. multiple cores, multiple cpus, faster cpus, sometimes what should never be possible happens.