<drewfustini>
We've noticed that null syscall latency takes a hit because of it
<drewfustini>
Maybe it could be made optional?
<palmer>
drewfustini: if you have a benchmark on HW then just send a patch to turn it off (maybe some DT key for performance?)
<drewfustini>
Bjorn
<drewfustini>
Oops
<palmer>
I haven't heard anything else I can quote ;)
<drewfustini>
Okay yeah I'll work on getting some public data
<palmer>
Thanks. It's not surprising it's going to be a perf problem, we just need something concrete
<drewfustini>
> DT key for performance
<drewfustini>
Do you mean creating a device tree property regarding this performance aspect?
<palmer>
Ya, there's going to be some microarchitectures it's slow on
<palmer>
There might also be some other zeroing scheme you need, or whatever
<drewfustini>
Okay, I'll post some data
<drewfustini>
Thanks
<palmer>
Cool, thanks
<palmer>
We're pretty far into microarchitecture specific stuff here, so it's impossible to tell exactly what's the right answer -- just that some wackiness is likely to crop up for someone at some point
<drewfustini>
I suppose some optionality would make sense
JRepin has quit [Quit: Konversation terminated!]
JRepin has joined #riscv
Andre_Z has quit [Ping timeout: 248 seconds]
Andre_Z has joined #riscv
vagrantc has joined #riscv
jobol has quit [Quit: Leaving]
Andre_Z has quit [Quit: Leaving.]
bsFFFFFF has quit [Remote host closed the connection]
mfny__ has quit [Read error: Connection reset by peer]