R0dn3yS/android_kernel_xiaomi_sm7250
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

power: em: correct increasing freq/power ratio

Browse Source 
As freq increases freq/power should decrease, indicating nonlinearly
higher power is required for running at higher frequencies.

If freq/power increases, the cost associated with that
freq will be lower than its previous one, causing the energy
calculations to choose a cpu with that frequency over a cpu with
the previous lesser freq.

For some frequencies, if their voltage increment from the
previous ones is very small (or is the same), we could end up with
higher freq/power ratio. This is primarily because
of_dev_pm_opp_get_cpu_power() returns power in mW and looses precision.

But instead of addressing it, enforce the same cost as the previous
frequency.

The energy evaluation code prefers the previous cpu when the energy costs
are same as other candidates. By keeping the energy costs same in these
situations we are increasing the likely hood of selecting prev cpu even
if it results in a slight freq bump. In general, selecting prev cpu is
beneficial because it avoids warming up the caches at a different cpu.

Change-Id: Ic66ab18ba65f2917b73d9fbe92a39b743b9839a0
Signed-off-by: Abhijeet Dharmapurikar <adharmap@codeaurora.org>
This commit is contained in:
Abhijeet Dharmapurikar 2019-03-27 09:51:24 -07:00
parent 3d91e789ad
commit c208338efc
1 changed files with 5 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
kernel/power/energy_model.c
View File

@ -141,7 +141,7 @@ static struct em_perf_domain *em_create_pd(cpumask_t *span, int nr_states,
*/
opp_eff = freq / power;
if (opp_eff >= prev_opp_eff)
pr_warn("pd%d: hertz/watts ratio non-monotonically decreasing: em_cap_state %d >= em_cap_state%d\n",
pr_debug("pd%d: hertz/watts ratio non-monotonically decreasing: em_cap_state %d >= em_cap_state%d\n",
cpu, i, i - 1);
prev_opp_eff = opp_eff;
}
@ -151,6 +151,10 @@ static struct em_perf_domain *em_create_pd(cpumask_t *span, int nr_states,
for (i = 0; i < nr_states; i++) {
table[i].cost = div64_u64(fmax * table[i].power,
table[i].frequency);
if (i > 0 && (table[i].cost < table[i - 1].cost) &&
(table[i].power > table[i - 1].power)) {
table[i].cost = table[i - 1].cost;
}
}
pd->table = table;