Intel Idle Support For Granite Rapids D Going Into Linux 6.13

[phoronix]

Phoronix: Intel Idle Support For Granite Rapids D Going Into Linux 6.13

The "intel_idle" driver provides CPU idle time management for Intel processors on Linux for helping to put the processor into low-power states in conjunction with the MWAIT instruction. With the upcoming Linux 6.13 kernel, the Granite Rapids D support for the Intel Idle driver is set to be merged...

https://www.phoronix.com/news/Intel-Idle-Granite-Rapids-D

[Espionage724]

Eh, as long as I can disable C-States entirely and even the cpuidle system, cool; I didn't buy hardware to run it slower

I've done cpuidle.off=1, and switched between idle=poll and halt to seemingly nothing notable (except high heat with poll) while already doing performance from x86_energy_perf_policy. I did less than an hour of research into the cpu idle system and came to some conclusion that mwait wasn't appealing, halt works but I don't understand what it does exactly, and poll works as-expected but why would that ever be preferred over halt? Then apparently disabling C-states with HT lowers performance vs leaving C-states enabled: https://community.intel.com/t5/Softw...x/td-p/1051623

Basically, the CPU idle system doesn't sound like something I want with max-perf locked-freq CPU, but I don't understand the exact specifics of idle/poll/mwait and just occasionally switch between them to see if I notice anything different

Edit: Heh, another article explained mwait for me, and now I know I really don't want it https://www.phoronix.com/news/Intel-...ke-Monitor-Bug

[jacob]

For a second I thought they were going to support Dlang.

[User29]

The "intel_idle" driver provides CPU idle time management for Intel processors on Linux for helping to put the processor into low-power states in conjunction with the MWAIT instruction. With the upcoming Linux 6.13 kernel, the Granite Rapids D support for the Intel Idle driver is set to be merged.

Does this mean that different CPU families/generations do this differently so 1-1 new driver is needed for each? Why is this good for any manufacturer? I'd think this is so basic stuff that shouldn't be touched for ages.

[intelfx]

Eh, as long as I can disable C-States entirely and even the cpuidle system, cool; I didn't buy hardware to run it slower

I've done cpuidle.off=1, and switched between idle=poll and halt to seemingly nothing notable (except high heat with poll) while already doing performance from x86_energy_perf_policy. I did less than an hour of research into the cpu idle system and came to some conclusion that mwait wasn't appealing, halt works but I don't understand what it does exactly, and poll works as-expected but why would that ever be preferred over halt? Then apparently disabling C-states with HT lowers performance vs leaving C-states enabled: https://community.intel.com/t5/Softw...x/td-p/1051623

Basically, the CPU idle system doesn't sound like something I want with max-perf locked-freq CPU, but I don't understand the exact specifics of idle/poll/mwait and just occasionally switch between them to see if I notice anything different

Edit: Heh, another article explained mwait for me, and now I know I really don't want it https://www.phoronix.com/news/Intel-...ke-Monitor-Bug

Are you on some sort of a contest to showcase as much incompetence as possible with all your comments?

[AdrianBc]

Does this mean that different CPU families/generations do this differently so 1-1 new driver is needed for each? Why is this good for any manufacturer? I'd think this is so basic stuff that shouldn't be touched for ages.

Each CPU supports some set of low-power states in which the CPU could wait while idle, where the states differ in the compromise between the power consumed while waiting to be woken and the time delay between receiving a wake-up signal and starting to run again the next instructions.

I assume that Granite Rapids D supports more such low-power states than the previous CPUs (or less, but that is less likely), so the idle driver must check the model of CPU that is used in order to choose or validate the type of low-power state that is requested for the CPU when halting it with MWAIT.

[Espionage724]

Are you on some sort of a contest to showcase as much incompetence as possible with all your comments?

Eh, kind-of mixed in a quest for others to share info that might be beneficial; got anything?