gpu: nvgpu: Change quantize order in GPCPLL NA mode

When calculating fractional divider in GPCPLL NA mode quantize voltage
before (used to do it after) applying DFS_COEFF, to follow h/w order.

Bug 1555318

Change-Id: I37be2bc73cd1f849695b94acc4ff21caf26e8b97
Signed-off-by: Alex Frid <afrid@nvidia.com>
Reviewed-on: http://git-master/r/552741
Reviewed-by: Yu-Huan Hsu <yhsu@nvidia.com>

1 files changed, 4 insertions, 3 deletions

diff --git a/drivers/gpu/nvgpu/gm20b/clk_gm20b.c b/drivers/gpu/nvgpu/gm20b/clk_gm20b.c
index fbdf9368..16f88d62 100644
--- a/drivers/gpu/nvgpu/gm20b/clk_gm20b.c
+++ b/drivers/gpu/nvgpu/gm20b/clk_gm20b.c
@@ -302,10 +302,11 @@ static void clk_config_dvfs_ndiv(int mv, u32 n_eff, struct na_dvfs *d)
         u32 rem, rem_range;
         struct pll_parms *p = &gpc_pll_params;
 
-        det_delta = (mv * 1000 - d->uv_cal);
-        det_delta = min(det_delta, d->dfs_det_max * p->uvdet_slope);
+        det_delta = DIV_ROUND_CLOSEST(mv * 1000 - p->uvdet_offs,
+                                      p->uvdet_slope);
+        det_delta -= d->dfs_ext_cal;
+        det_delta = min(det_delta, d->dfs_det_max);
         det_delta = det_delta * d->dfs_coeff;
-        det_delta = DIV_ROUND_CLOSEST(det_delta, p->uvdet_slope);
 
         n = (int)(n_eff << DFS_DET_RANGE) - det_delta;
         BUG_ON((n < 0) || (n > (p->max_N << DFS_DET_RANGE)));