gpu: nvgpu: set clk_rate on frequency change

Currently, in gk20a_scale_target, we set clock frequency
even if it is equivalent to the rate previously requested by
the governor. This change adds a check to bypass this in
case new_frequency == prev_frequency.

These clocking operations result in multiple BPMP calls, and add
significant overhead to submit time. So, we avoid these
operations when possible.

Bug 1795076

Change-Id: I0f180564e54581f0f4add4626c647e0b9a1bbe43
Signed-off-by: Sachit Kadle <skadle@nvidia.com>
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/1247913
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Aaron Huang <aaronh@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>

1 files changed, 4 insertions, 0 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/gk20a_scale.c b/drivers/gpu/nvgpu/gk20a/gk20a_scale.c
index d2229a6a..e3b21c5a 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a_scale.c
+++ b/drivers/gpu/nvgpu/gk20a/gk20a_scale.c
@@ -181,6 +181,10 @@ static int gk20a_scale_target(struct device *dev, unsigned long *freq,
         if (local_freq > max_freq)
                 local_freq = max_freq;
 
+        /* Check for duplicate request */
+        if (local_freq == g->devfreq->previous_freq)
+                return 0;
+
         /* set the final frequency */
         rounded_rate = platform->clk_round_rate(dev, local_freq);