Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 8CA4121720
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 8BIT
To:     Amit Nischal <anischal@codeaurora.org>,
        Michael Turquette <mturquette@baylibre.com>,
        Stephen Boyd <sboyd@codeaurora.org>
From:   Stephen Boyd <sboyd@kernel.org>
In-Reply-To: <1520493495-3084-3-git-send-email-anischal@codeaurora.org>
Cc:     Andy Gross <andy.gross@linaro.org>,
        David Brown <david.brown@linaro.org>,
        Rajendra Nayak <rnayak@codeaurora.org>,
        Odelu Kukatla <okukatla@codeaurora.org>,
        Taniya Das <tdas@codeaurora.org>,
        linux-arm-msm@vger.kernel.org, linux-soc@vger.kernel.org,
        linux-clk@vger.kernel.org, linux-kernel@vger.kernel.org,
        Amit Nischal <anischal@codeaurora.org>
References: <1520493495-3084-1-git-send-email-anischal@codeaurora.org>
 <1520493495-3084-3-git-send-email-anischal@codeaurora.org>
Message-ID: <152150582284.254778.15552235408861080239@swboyd.mtv.corp.google.com>
User-Agent: alot/0.7
Subject: Re: [PATCH v2 2/4] clk: qcom: Configure the RCGs to a safe source as needed
Date:   Mon, 19 Mar 2018 17:30:22 -0700
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Quoting Amit Nischal (2018-03-07 23:18:13)
> For some root clock generators, there could be child branches which are
> controlled by an entity other than application processor subsystem. For
> such RCGs, as per application processor subsystem clock driver, all of its
> downstream clocks are disabled and RCG is in disabled state but in actual
> downstream clocks can be left enabled before.

s/actual/reality?

> 
> So in this scenario, when RCG is disabled as per clock driver's point of
> view and when rate scaling request comes before downstream clock enable
> request, then RCG fails to update its configuration because in actual RCG

s/actual/reality?

> is on and it expects its new source to alredy in enable state  but in

s/alredy/already be/

> reality new source is in off state. In order to avoid letting the RCG to

s/is in off state/is off/
s/letting/having/

> go into an invalid state, add support to just cache the rate of RCG during

s/just//

> set_rate(), defer actual RCG configuration update to be done during
> clk_enable() as at this point of time, both its new parent and safe source
> will be already enabled and RCG can safely switch to new parent.
> 
> During clk_disable() request, configure it to safe source as both
> its parents, safe source and current parent will be enabled and RCG can
> safely execute a switch. Also add support to have safe configuration
> frequency table structure for each shared RCG.
> 
> Signed-off-by: Taniya Das <tdas@codeaurora.org>
> Signed-off-by: Amit Nischal <anischal@codeaurora.org>
> ---
> diff --git a/drivers/clk/qcom/clk-rcg.h b/drivers/clk/qcom/clk-rcg.h
> index 2a7489a..205fa34 100644
> --- a/drivers/clk/qcom/clk-rcg.h
> +++ b/drivers/clk/qcom/clk-rcg.h
> @@ -146,6 +146,9 @@ struct clk_dyn_rcg {
>   * @hid_width: number of bits in half integer divider
>   * @parent_map: map from software's parent index to hardware's src_sel field
>   * @freq_tbl: frequency table
> + * @current_freq: last cached frequency when using branches with shared RCGs
> + * @safe_src_freq_tbl : frequency table of safe source when using branches
> + *                     with shared RCGs
>   * @clkr: regmap clock handle
>   *
>   */
> @@ -155,6 +158,8 @@ struct clk_rcg2 {
>         u8                      hid_width;
>         const struct parent_map *parent_map;
>         const struct freq_tbl   *freq_tbl;
> +       unsigned long           current_freq;
> +       const struct freq_tbl   *safe_src_freq_tbl;

Can we store safe_src index instead? And then construct the frequency
table entry on the fly at the call site? I think it would greatly
clarify that we don't really care about the rate of the clk at all.
Instead, all we care about is making sure the mux is set to whatever
source selection can provide an always on signal.

>         struct clk_regmap       clkr;
>  };
> 
> @@ -167,5 +172,6 @@ struct clk_rcg2 {
>  extern const struct clk_ops clk_byte2_ops;
>  extern const struct clk_ops clk_pixel_ops;
>  extern const struct clk_ops clk_gfx3d_ops;
> +extern const struct clk_ops clk_rcg2_shared_ops;
> 
>  #endif
> diff --git a/drivers/clk/qcom/clk-rcg2.c b/drivers/clk/qcom/clk-rcg2.c
> index e63db10..a52de54 100644
> --- a/drivers/clk/qcom/clk-rcg2.c
> +++ b/drivers/clk/qcom/clk-rcg2.c
> @@ -790,3 +790,158 @@ static int clk_gfx3d_set_rate(struct clk_hw *hw, unsigned long rate,
>         .determine_rate = clk_gfx3d_determine_rate,
>  };
> +
> +static unsigned long
> +clk_rcg2_shared_recalc_rate(struct clk_hw *hw, unsigned long parent_rate)
> +{
> +       struct clk_rcg2 *rcg = to_clk_rcg2(hw);
> +
> +       if (!__clk_is_enabled(hw->clk)) {
> +               if (!rcg->current_freq) {
> +                       if (!clk_rcg2_get_parent(hw))

This is checking for 0 source selection of parent? So basically if
source is XO selected then return what we know is the XO speed? We
should be able to do that by returning parent_rate though?

> +                               rcg->current_freq =
> +                                       rcg->safe_src_freq_tbl->freq;
> +                       else
> +                               rcg->current_freq =
> +                                       clk_rcg2_recalc_rate(hw, parent_rate);
> +               }
> +
> +               return rcg->current_freq;
> +       }
> +
> +       return rcg->current_freq = clk_rcg2_recalc_rate(hw, parent_rate);

I don't get this function.

To simplify just return cached rate if it's set and clk is off,
otherwise read the hardware?

	if (!__clk_is_enabled(hw->clk) && rcg->current_freq)
		return rcg->current_freq;

	return clk_rcg2_recalc_rate(hw, parent_rate);

I would also rename current_freq to something like cached_freq and then
*only* assign it when someone calls clk_set_rate() (and save around the
frequency table pointer instead of raw frequency). Out of boot, we will
return the right frequency and clk_set_rate() called before enable will
cache the rate to what we want.

> +}
> +
> +static int clk_rcg2_shared_enable(struct clk_hw *hw)
> +{
> +       struct clk_rcg2 *rcg = to_clk_rcg2(hw);
> +
> +       if (rcg->current_freq == rcg->safe_src_freq_tbl->freq) {
> +               clk_rcg2_set_force_enable(hw);
> +               clk_rcg2_configure(rcg, rcg->safe_src_freq_tbl);
> +               clk_rcg2_clear_force_enable(hw);
> +
> +               return 0;
> +       }
> +
> +       /*
> +        * Switch from safe source to the stashed mux selection. The current
> +        * parent has already been prepared and enabled at this point, and
> +        * the safe source is always on while application processor subsystem
> +        * is online. Therefore, the RCG can safely switch its source.
> +        */
> +
> +       return clk_rcg2_shared_force_enable_clear(hw, rcg->current_freq);

Not much is different between clk_rcg2_shared_force_enable_clear() and
the three lines above inside that if condition. The only difference is
we don't search for a rate in the frequency table. Ugh, but we can't
change the frequency table because of hw clk control stuff that relies
on the table in software matching the table in hardware? Sad.

Can we at least change the if condition above to be something like:

	if (rcg->force_safe_src) {
		clk_rcg2_set_force_enable(...

And then set the force_safe_src bool when we want enable to skip normal
rate restoration behavior? We should be able to use that flag in other
places too, like where we check for the clk being enabled in software or
not.

> +}
> +
> +static void clk_rcg2_shared_disable(struct clk_hw *hw)
> +{
> +       struct clk_rcg2 *rcg = to_clk_rcg2(hw);
> +
> +       /*
> +        * Park the RCG at a safe configuration - sourced off from safe source.
> +        * Force enable and disable the RCG while configuring it to safeguard
> +        * against any update signal coming from the downstream clock.
> +        * The current parent is still prepared and enabled at this point, and
> +        * the safe source is always on while application processor subsystem
> +        * is online. Therefore, the RCG can safely switch its parent.
> +        */
> +       clk_rcg2_set_force_enable(hw);
> +       clk_rcg2_configure(rcg, rcg->safe_src_freq_tbl);
> +       clk_rcg2_clear_force_enable(hw);
> +}
> +

Yeah, all this stuff would get simpler if we stopped using frequencies.
Hopefully I made sense!