Re: [PATCH v2 4/5] clk: qcom: clk-alpha-pll: Add support for Lucid 5LPE PLL

[Vinod Koul <vkoul@kernel.org>]

On 10-12-20, 12:36, Stephen Boyd wrote:
> Quoting Vinod Koul (2020-12-07 22:47:01)
> > diff --git a/drivers/clk/qcom/clk-alpha-pll.c b/drivers/clk/qcom/clk-alpha-pll.c
> > index 564431130a76..6a399663d564 100644
> > --- a/drivers/clk/qcom/clk-alpha-pll.c
> > +++ b/drivers/clk/qcom/clk-alpha-pll.c
> > @@ -146,6 +146,12 @@ EXPORT_SYMBOL_GPL(clk_alpha_pll_regs);
> >  /* LUCID PLL specific settings and offsets */
> >  #define LUCID_PCAL_DONE                BIT(27)
> >  
> > +/* LUCID 5LPE PLL specific settings and offsets */
> > +#define LUCID_5LPE_PCAL_DONE           BIT(11)
> > +#define LUCID_5LPE_ENABLE_VOTE_RUN     BIT(21)
> > +#define LUCID_5LPE_PLL_LATCH_INPUT     BIT(14)
> > +#define LUCID_5LPE_ALPHA_PLL_ACK_LATCH BIT(13)
> 
> Sort these by bit or define name?

Okay will sort by bit

> 
> > +
> >  #define pll_alpha_width(p)                                     \
> >                 ((PLL_ALPHA_VAL_U(p) - PLL_ALPHA_VAL(p) == 4) ? \
> >                                  ALPHA_REG_BITWIDTH : ALPHA_REG_16BIT_WIDTH)
> > @@ -1561,3 +1567,220 @@ const struct clk_ops clk_alpha_pll_postdiv_lucid_ops = {
> >         .set_rate = clk_alpha_pll_postdiv_fabia_set_rate,
> >  };
> >  EXPORT_SYMBOL_GPL(clk_alpha_pll_postdiv_lucid_ops);
> > +
> > +static int alpha_pll_lucid_5lpe_enable(struct clk_hw *hw)
> > +{
> > +       struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
> > +       u32 val;
> > +       int ret;
> > +
> > +       ret = regmap_read(pll->clkr.regmap, PLL_USER_CTL(pll), &val);
> > +       if (ret)
> > +               return ret;
> > +
> > +       /* If in FSM mode, just vote for it */
> > +       if (val & LUCID_5LPE_ENABLE_VOTE_RUN) {
> > +               ret = clk_enable_regmap(hw);
> > +               if (ret)
> > +                       return ret;
> > +               return wait_for_pll_enable_lock(pll);
> > +       }
> > +
> > +       /* Check if PLL is already enabled */
> 
> Yeah that's obvious, but then what?

then dont proceed :) will update

> > +       ret = trion_pll_is_enabled(pll, pll->clkr.regmap);
> > +       if (ret < 0)
> > +               return ret;
> > +
> > +       ret = regmap_update_bits(pll->clkr.regmap, PLL_MODE(pll), PLL_RESET_N, PLL_RESET_N);
> > +       if (ret)
> > +               return ret;
> > +
> > +       /* Set operation mode to RUN */
> 
> This comment is worthless.

Will drop

> 
> > +       regmap_write(pll->clkr.regmap, PLL_OPMODE(pll), PLL_RUN);
> > +
> > +       ret = wait_for_pll_enable_lock(pll);
> > +       if (ret)
> > +               return ret;
> > +
> > +       /* Enable the PLL outputs */
> > +       ret = regmap_update_bits(pll->clkr.regmap, PLL_USER_CTL(pll), PLL_OUT_MASK, PLL_OUT_MASK);
> > +       if (ret)
> > +               return ret;
> > +
> > +       /* Enable the global PLL outputs */
> > +       ret = regmap_update_bits(pll->clkr.regmap, PLL_MODE(pll), PLL_OUTCTRL, PLL_OUTCTRL);
> > +       if (ret)
> > +               return ret;
> > +
> > +       /* Ensure that the write above goes through before returning. */
> > +       mb();
> 
> Regmap has a memory barrier in writel. Drop this.

yes

> 
> > +       return ret;
> > +}
> > +
> > +static void alpha_pll_lucid_5lpe_disable(struct clk_hw *hw)
> > +{
> > +       struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
> > +       u32 val;
> > +       int ret;
> > +
> > +       ret = regmap_read(pll->clkr.regmap, PLL_USER_CTL(pll), &val);
> > +       if (ret)
> > +               return;
> > +
> > +       /* If in FSM mode, just unvote it */
> > +       if (val & LUCID_5LPE_ENABLE_VOTE_RUN) {
> > +               clk_disable_regmap(hw);
> > +               return;
> > +       }
> > +
> > +       /* Disable the global PLL output */
> > +       ret = regmap_update_bits(pll->clkr.regmap, PLL_MODE(pll), PLL_OUTCTRL, 0);
> > +       if (ret)
> > +               return;
> > +
> > +       /* Disable the PLL outputs */
> > +       ret = regmap_update_bits(pll->clkr.regmap, PLL_USER_CTL(pll), PLL_OUT_MASK, 0);
> > +       if (ret)
> > +               return;
> > +
> > +       /* Place the PLL mode in STANDBY */
> > +       regmap_write(pll->clkr.regmap, PLL_OPMODE(pll), PLL_STANDBY);
> > +}
> > +
> > +/*
> > + * The Lucid 5LPE PLL requires a power-on self-calibration which happens
> > + * when the PLL comes out of reset. Calibrate in case it is not completed.
> > + */
> > +static int alpha_pll_lucid_5lpe_prepare(struct clk_hw *hw)
> > +{
> > +       struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
> > +       struct clk_hw *p;
> > +       u32 regval;
> 
> Can you use u32 val? And also include a patch to replace the couple
> times where there is 'regval' in this file. The former is shorter and
> used far more in qcom clk code.

Will do

> 
> > +       int ret;
> > +
> > +       /* Return early if calibration is not needed. */
> > +       regmap_read(pll->clkr.regmap, PLL_MODE(pll), &regval);
> > +       if (regval & LUCID_5LPE_PCAL_DONE)
> > +               return 0;
> > +
> > +       p = clk_hw_get_parent(hw);
> > +       if (!p)
> > +               return -EINVAL;
> > +
> > +       ret = alpha_pll_lucid_5lpe_enable(hw);
> > +       if (ret)
> > +               return ret;
> > +
> > +       alpha_pll_lucid_5lpe_disable(hw);
> > +
> > +       return 0;
> > +}
> > +
> > +static int alpha_pll_lucid_5lpe_set_rate(struct clk_hw *hw, unsigned long rate,
> > +                                        unsigned long prate)
> > +{
> > +       struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
> > +       unsigned long rrate;
> > +       u32 regval, l;
> > +       u64 a;
> > +       int ret;
> > +
> > +       rrate = alpha_pll_round_rate(rate, prate, &l, &a, ALPHA_REG_16BIT_WIDTH);
> > +
> > +       /*
> > +        * Due to a limited number of bits for fractional rate programming, the
> > +        * rounded up rate could be marginally higher than the requested rate.
> > +        */
> > +       if (rrate > (rate + PLL_RATE_MARGIN) || rrate < rate) {
> > +               pr_err("Call set rate on the PLL with rounded rates!\n");
> > +               return -EINVAL;
> > +       }
> 
> Can we use alpha_pll_check_rate_margin()?

Ah a shiny new helper, looking at it yes we should

> 
> > +
> > +       regmap_write(pll->clkr.regmap, PLL_L_VAL(pll), l);
> > +       regmap_write(pll->clkr.regmap, PLL_ALPHA_VAL(pll), a);
> > +
> > +       /* Latch the PLL input */
> > +       ret = regmap_update_bits(pll->clkr.regmap, PLL_MODE(pll),
> > +                                LUCID_5LPE_PLL_LATCH_INPUT, LUCID_5LPE_PLL_LATCH_INPUT);
> > +       if (ret)
> > +               return ret;
> > +
> > +       /* Wait for 2 reference cycles before checking the ACK bit. */
> > +       udelay(1);
> > +       regmap_read(pll->clkr.regmap, PLL_MODE(pll), &regval);
> > +       if (!(regval & LUCID_5LPE_ALPHA_PLL_ACK_LATCH)) {
> > +               pr_err("Lucid 5LPE PLL latch failed. Output may be unstable!\n");
> > +               return -EINVAL;
> > +       }
> > +
> > +       /* Return the latch input to 0 */
> > +       ret = regmap_update_bits(pll->clkr.regmap, PLL_MODE(pll), LUCID_5LPE_PLL_LATCH_INPUT, 0);
> > +       if (ret)
> > +               return ret;
> > +
> > +       if (clk_hw_is_enabled(hw)) {
> > +               ret = wait_for_pll_enable_lock(pll);
> > +               if (ret)
> > +                       return ret;
> > +       }
> > +
> > +       /* Wait for PLL output to stabilize */
> > +       udelay(100);
> > +       return 0;
> > +}
> > +
> > +static int clk_lucid_5lpe_pll_postdiv_set_rate(struct clk_hw *hw, unsigned long rate,
> > +                                              unsigned long parent_rate)
> > +{
> > +       struct clk_alpha_pll_postdiv *pll = to_clk_alpha_pll_postdiv(hw);
> > +       int i, val = 0, div, ret;
> > +
> > +       /*
> > +        * If the PLL is in FSM mode, then treat set_rate callback as a
> > +        * no-operation.
> > +        */
> > +       ret = regmap_read(pll->clkr.regmap, PLL_USER_CTL(pll), &val);
> > +       if (ret)
> > +               return ret;
> > +
> > +       if (val & LUCID_5LPE_ENABLE_VOTE_RUN)
> > +               return 0;
> > +
> > +       if (!pll->post_div_table) {
> > +               pr_err("Missing the post_div_table for the PLL\n");
> 
> Can this be rolled into the loop below?

Yep

> > +               return -EINVAL;
> > +       }
> > +
> > +       div = DIV_ROUND_UP_ULL((u64)parent_rate, rate);
> > +       for (i = 0; i < pll->num_post_div; i++) {
> 
> So that this finds nothing.
> 
> > +               if (pll->post_div_table[i].div == div) {
> > +                       val = pll->post_div_table[i].val;
> > +                       break;
> > +               }
> > +       }
> 
> and then if val == -1 we return -EINVAL?

Correct, will update

> > +
> > +       return regmap_update_bits(pll->clkr.regmap, PLL_USER_CTL(pll),
> > +                               (BIT(pll->width) - 1) << pll->post_div_shift,
> 
> Use GENMASK?

Looks like this can be:
                GENMASK(pll->width + pll->post_div_shift - 1, pll->post_div_shift)

Not sure which one you like :)

-- 
~Vinod