[PATCH 04/13] clk: qcom: fix 16 bit alpha support calculation

From: Abhishek Sahu
Date: Thu Sep 28 2017 - 13:54:20 EST


The alpha value calculation function has been written for 40 bit
alpha which is not coming properly for 16 bit

1. Alpha value is being calculated on the basis of
ALPHA_BITWIDTH to make the computation easy for 40 bit alpha.
After calculating the 32 bit alpha, It is being converted to 40
bit alpha by making making lower bits zero. But if actual alpha
register width is less than ALPHA_BITWIDTH, then the actual width
can be used for calculation

2. During 40 bit alpha pll set rate, the lower alpha register is
not being configured

Now the changes have been made to calculate the rate and register
values from alpha_width instead hardcoding it so that it can work
for all the cases.

Signed-off-by: Abhishek Sahu <absahu@xxxxxxxxxxxxxx>
---
drivers/clk/qcom/clk-alpha-pll.c | 55 ++++++++++++++++++++++++----------------
1 file changed, 33 insertions(+), 22 deletions(-)

diff --git a/drivers/clk/qcom/clk-alpha-pll.c b/drivers/clk/qcom/clk-alpha-pll.c
index d5bfc52..a12f7b4 100644
--- a/drivers/clk/qcom/clk-alpha-pll.c
+++ b/drivers/clk/qcom/clk-alpha-pll.c
@@ -45,8 +45,12 @@
* Even though 40 bits are present, use only 32 for ease of calculation.
*/
#define ALPHA_REG_BITWIDTH 40
+#define ALPHA_REG_16BIT_WIDTH 16
#define ALPHA_BITWIDTH 32
-#define ALPHA_16BIT_MASK 0xffff
+
+/* Returns the Alpha register width for pll */
+#define pll_alpha_width(pll) (pll->flags & SUPPORTS_16BIT_ALPHA ? \
+ ALPHA_REG_16BIT_WIDTH : ALPHA_REG_BITWIDTH)

/* Returns the alpha_pll_clk_ops for pll type */
#define pll_clk_ops(hw) (alpha_pll_props[to_clk_alpha_pll(hw)-> \
@@ -364,13 +368,16 @@ static void alpha_pll_default_disable(struct clk_hw *hw)
regmap_update_bits(pll->clkr.regmap, off, mask, 0);
}

-static unsigned long alpha_pll_calc_rate(u64 prate, u32 l, u32 a)
+static unsigned long
+alpha_pll_calc_rate(u64 prate, u32 l, u32 a, u32 alpha_width)
{
- return (prate * l) + ((prate * a) >> ALPHA_BITWIDTH);
+ return (prate * l) + ((prate * a) >>
+ (alpha_width < ALPHA_BITWIDTH ? alpha_width : ALPHA_BITWIDTH));
}

static unsigned long
-alpha_pll_round_rate(unsigned long rate, unsigned long prate, u32 *l, u64 *a)
+alpha_pll_round_rate(unsigned long rate, unsigned long prate, u32 *l, u64 *a,
+ u32 alpha_width)
{
u64 remainder;
u64 quotient;
@@ -385,14 +392,16 @@ static unsigned long alpha_pll_calc_rate(u64 prate, u32 l, u32 a)
}

/* Upper ALPHA_BITWIDTH bits of Alpha */
- quotient = remainder << ALPHA_BITWIDTH;
+ quotient = remainder << (alpha_width < ALPHA_BITWIDTH ?
+ alpha_width : ALPHA_BITWIDTH);
+
remainder = do_div(quotient, prate);

if (remainder)
quotient++;

*a = quotient;
- return alpha_pll_calc_rate(prate, *l, *a);
+ return alpha_pll_calc_rate(prate, *l, *a, alpha_width);
}

static const struct pll_vco *
@@ -414,7 +423,7 @@ static unsigned long alpha_pll_calc_rate(u64 prate, u32 l, u32 a)
u32 l, low, high, ctl;
u64 a = 0, prate = parent_rate;
struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
- u32 off = pll->offset;
+ u32 off = pll->offset, alpha_width = pll_alpha_width(pll);
u8 type = pll->pll_type;

regmap_read(pll->clkr.regmap, off + pll_l(type), &l);
@@ -422,17 +431,19 @@ static unsigned long alpha_pll_calc_rate(u64 prate, u32 l, u32 a)
regmap_read(pll->clkr.regmap, off + pll_user_ctl(type), &ctl);
if (ctl & PLL_ALPHA_EN) {
regmap_read(pll->clkr.regmap, off + pll_alpha(type), &low);
- if (pll->flags & SUPPORTS_16BIT_ALPHA) {
- a = low & ALPHA_16BIT_MASK;
- } else {
+ if (alpha_width > 32) {
regmap_read(pll->clkr.regmap, off + pll_alpha_u(type),
&high);
a = (u64)high << 32 | low;
- a >>= ALPHA_REG_BITWIDTH - ALPHA_BITWIDTH;
+ } else {
+ a = low & GENMASK(alpha_width - 1, 0);
}
+
+ if (alpha_width > ALPHA_BITWIDTH)
+ a >>= alpha_width - ALPHA_BITWIDTH;
}

- return alpha_pll_calc_rate(prate, l, a);
+ return alpha_pll_calc_rate(prate, l, a, alpha_width);
}

static int alpha_pll_default_set_rate(struct clk_hw *hw, unsigned long rate,
@@ -440,11 +451,11 @@ static int alpha_pll_default_set_rate(struct clk_hw *hw, unsigned long rate,
{
struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
const struct pll_vco *vco;
- u32 l, off = pll->offset;
+ u32 l, off = pll->offset, alpha_width = pll_alpha_width(pll);
u8 type = pll->pll_type;
u64 a;

- rate = alpha_pll_round_rate(rate, prate, &l, &a);
+ rate = alpha_pll_round_rate(rate, prate, &l, &a, alpha_width);
vco = alpha_pll_find_vco(pll, rate);
if (!vco) {
pr_err("alpha pll not in a valid vco range\n");
@@ -453,14 +464,14 @@ static int alpha_pll_default_set_rate(struct clk_hw *hw, unsigned long rate,

regmap_write(pll->clkr.regmap, off + pll_l(type), l);

- if (pll->flags & SUPPORTS_16BIT_ALPHA) {
- regmap_write(pll->clkr.regmap, off + pll_alpha(type),
- a & ALPHA_16BIT_MASK);
- } else {
- a <<= (ALPHA_REG_BITWIDTH - ALPHA_BITWIDTH);
+ if (alpha_width > ALPHA_BITWIDTH)
+ a <<= alpha_width - ALPHA_BITWIDTH;
+
+ if (alpha_width > 32)
regmap_write(pll->clkr.regmap, off + pll_alpha_u(type),
a >> 32);
- }
+
+ regmap_write(pll->clkr.regmap, off + pll_alpha(type), a);

regmap_update_bits(pll->clkr.regmap, off + pll_user_ctl(type),
PLL_VCO_MASK << PLL_VCO_SHIFT,
@@ -476,11 +487,11 @@ static long alpha_pll_default_round_rate(struct clk_hw *hw, unsigned long rate,
unsigned long *prate)
{
struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
- u32 l;
+ u32 l, alpha_width = pll_alpha_width(pll);
u64 a;
unsigned long min_freq, max_freq;

- rate = alpha_pll_round_rate(rate, *prate, &l, &a);
+ rate = alpha_pll_round_rate(rate, *prate, &l, &a, alpha_width);
if (alpha_pll_find_vco(pll, rate))
return rate;

--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation