lkml.org 
[lkml]   [2021]   [Jun]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH][v2] intel_idle: Adjust the SKX C6 latency and residency if PC6 is disabled
From
Date
On Fri, 2021-05-28 at 11:20 +0800, Chen Yu wrote:
> Currently cpuidle assumes worst-case C-state parameters, and so C6
> is described with PC6 parameters, which is worst case for requesting
> CC6. When PC6 is enabled, this is appropriate. But if PC6 is disabled
> in BIOS, the exit latency and target_residency should be adjusted
> accordingly.
>
> Exit latency:
> Previously the C6 exit latency was measured when woken up from CC6/PC6.
> With PC6 disabled, the C6 exit latency should be CC6/PC0.
>
> Target residency:
> With PC6 disabled, idle duration within [CC6, PC6) would make the
> idle governor choose C1E over C6. This would cause low energy-efficiency.
> We should lower the bar to request C6 when PC6 is disabled.
>
> To fill this gap, check if PC6 is disabled in the BIOS in the
> MSR_PKG_CST_CONFIG_CONTROL(0xe2). If so, use CC6/PC0 parameters as the
> new exit latency. Meanwhile, update target_residency to 3 times of the new
> exit latency. This is consistent with how intel_idle driver uses _CST to
> calculate the target_residency. The consequence is that, the OS would
> be more offen to choose C6 over C1E when PC6 is disabled. This is reasonable
> because if the user is using C6, it implies that the user cares about energy,
> so choosing C6 more frequently is in accordance with user requirement.
>
> The new exit latency of CC6/PC0 92us was from wult[1] result on SKX, which was
> measured via NIC wakeup from 99.99th latency. Besides SKX, the CLX and CPX
> both have the same CPU model number. And since they have similar CC6 exit latency
> to SKX, 96us and 89us respectively, reuse the value of SKX.
>
> There is concern that if we should introduce a more generic solution
> rather than optimizing on each platforms. However consider the
> code complexity and different PC6 bit interpretation on different
> platforms, tune the code per platform seems to be an acceptable trade-off.
>
> [1] https://intel.github.io/wult/
>
> Suggested-by: Len Brown <len.brown@intel.com>
> Signed-off-by: Chen Yu <yu.c.chen@intel.com>
> ---
> v2: Simplify the commit log to not mention C3/PC3. (Artem)
>     Confirm the exit latency on CLX and CPX.(Artem)

Reviewed-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>


\
 
 \ /
  Last update: 2021-06-09 12:52    [W:0.100 / U:0.040 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site