By the way, and just in case anyone is interested, I am using 5.12-rc1 with this change:
Code:
doug@s18:~/temp-k-git/linux$ git diff
diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c
index 3273360f30f7..3464defa004e 100644
--- a/drivers/idle/intel_idle.c
+++ b/drivers/idle/intel_idle.c
@@ -636,8 +636,8 @@ static struct cpuidle_state skl_cstates[] __initdata = {
.enter_s2idle = intel_idle_s2idle, },
{
.name = "C1E",
- .desc = "MWAIT 0x01",
- .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
+ .desc = "MWAIT 0x03",
+ .flags = MWAIT2flg(0x03) | CPUIDLE_FLAG_ALWAYS_ENABLE,
.exit_latency = 10,
.target_residency = 20,
.enter = &intel_idle,
As a potential solution to an HWP (HardWare Pstate) control problem in Intel processors. It has taken 9 months to figure out. Bug report.
EDIT: the patch is specific to my HWP capable processor, i5-9600K, and needs to be propagated for others.
EDIT 2: I have now entered a downstream bug report, on launchpad.