17.12.2019 04:29, Sowjanya Komatineni пишет:

On 12/7/19 11:20 AM, Sowjanya Komatineni wrote:

...

On 12/7/19 6:58 AM, Dmitry Osipenko wrote:

...

06.12.2019 05:48, Sowjanya Komatineni пишет:

...

Current ASoC driver uses extern1 as cdev1 clock from Tegra30 onwards through device tree.

Actual audio mclk is clk_out_1 and to use PLLA for mclk rate control, need to clk_out_1_mux parent to extern1 and extern1 parent to PLLA_OUT0.

Currently Tegra clock driver init sets the parents and enables both clk_out_1 and extern1 clocks. But these clocks parent and enables should be controlled by ASoC driver.

Clock parents can be specified in device tree using assigned-clocks and assigned-clock-parents.

To enable audio mclk, both clk_out_1 and extern1 clocks need to be enabled.

This patch configures parents for clk_out_1 and extern1 clocks if device tree does not specify clock parents inorder to support old device tree and controls mclk using both clk_out_1 and extern1 clocks.

Signed-off-by: Sowjanya Komatineni skomatineni@nvidia.com

sound/soc/tegra/tegra_asoc_utils.c | 66 ++++++++++++++++++++++++++++++++++++++   sound/soc/tegra/tegra_asoc_utils.h |  1 +   2 files changed, 67 insertions(+)

diff --git a/sound/soc/tegra/tegra_asoc_utils.c b/sound/soc/tegra/tegra_asoc_utils.c index 536a578e9512..8e3a3740df7c 100644 --- a/sound/soc/tegra/tegra_asoc_utils.c +++ b/sound/soc/tegra/tegra_asoc_utils.c @@ -60,6 +60,7 @@ int tegra_asoc_utils_set_rate(struct tegra_asoc_utils_data *data, int srate,       data->set_mclk = 0;         clk_disable_unprepare(data->clk_cdev1); +    clk_disable_unprepare(data->clk_extern1);       clk_disable_unprepare(data->clk_pll_a_out0);       clk_disable_unprepare(data->clk_pll_a);   @@ -89,6 +90,14 @@ int tegra_asoc_utils_set_rate(struct tegra_asoc_utils_data *data, int srate,           return err;       }   +    if (!IS_ERR_OR_NULL(data->clk_extern1)) { +        err = clk_prepare_enable(data->clk_extern1); +        if (err) { +            dev_err(data->dev, "Can't enable extern1: %d\n", err); +            return err; +        } +    }

err = clk_prepare_enable(data->clk_cdev1);       if (err) {           dev_err(data->dev, "Can't enable cdev1: %d\n", err); @@ -109,6 +118,7 @@ int tegra_asoc_utils_set_ac97_rate(struct tegra_asoc_utils_data *data)       int err;         clk_disable_unprepare(data->clk_cdev1); +    clk_disable_unprepare(data->clk_extern1);       clk_disable_unprepare(data->clk_pll_a_out0);       clk_disable_unprepare(data->clk_pll_a);   @@ -142,6 +152,14 @@ int tegra_asoc_utils_set_ac97_rate(struct tegra_asoc_utils_data *data)           return err;       }   +    if (!IS_ERR_OR_NULL(data->clk_extern1)) { +        err = clk_prepare_enable(data->clk_extern1); +        if (err) { +            dev_err(data->dev, "Can't enable extern1: %d\n", err); +            return err; +        } +    }

Why this is needed given that clk_extern1 is either a child of MCLK or MCLK itself (on T20)? The child clocks are enabled when the parent is enabled.

For T30 and later, clk_extern1 is one of the source for clk_out_1_mux. clk_extern1 is in CAR and it has its own gate and mux.

As audio mclk related clocks (clk_out_1, clk_out_1_mux, and extern1) are moved into ASoC driver from clock driver

need to enable extern1 gate as well along with clk_out1 for T30 through T210.

Just FYI, extern1 enable here happens only when data->clk_extern1 is available which is for T30 onwards.

...

...

err = clk_prepare_enable(data->clk_cdev1);       if (err) {           dev_err(data->dev, "Can't enable cdev1: %d\n", err); @@ -158,6 +176,7 @@ EXPORT_SYMBOL_GPL(tegra_asoc_utils_set_ac97_rate);   int tegra_asoc_utils_init(struct tegra_asoc_utils_data *data,                 struct device *dev)   { +    struct clk *clk_out_1_mux;       int ret;         data->dev = dev; @@ -196,6 +215,51 @@ int tegra_asoc_utils_init(struct tegra_asoc_utils_data *data,           goto err_put_pll_a_out0;       }

In a previous patch you added fallback to EXTPERIPH when clk_get(MCLK) fails. This will work perfectly fine for the older kernels which have all clocks in the same single CaR driver, but this may not work that great for the newer kernels because PMC driver isn't registered early during boot and thus it is possible to get a legit -EPROBE_DEFER which shouldn't be ignored. In other words, you need to add into this patch a check for the error code returned by clk_get(MCLK) and fallback only for -EINVAL.

yeah right, will add check in next version.

...

...

+    /* +     * If clock parents are not set in DT, configure here to use clk_out_1 +     * as mclk and extern1 as parent for Tegra30 and higher. +     */ +    if (!of_find_property(dev->of_node, "assigned-clock-parents", NULL) && +        data->soc > TEGRA_ASOC_UTILS_SOC_TEGRA20) { +        data->clk_extern1 = clk_get_sys("clk_out_1", "extern1"); +        if (IS_ERR(data->clk_extern1)) { +            dev_err(data->dev, "Can't retrieve clk extern1\n"); +            ret = PTR_ERR(data->clk_extern1); +            goto err_put_cdev1; +        }

+        ret = clk_set_parent(data->clk_extern1, data->clk_pll_a_out0); +        if (ret < 0) { +            dev_err(data->dev, +                "Set parent failed for clk extern1: %d\n", +                ret); +            goto err_put_cdev1; +        }

+        clk_out_1_mux = clk_get_sys(NULL, "clk_out_1_mux");

Note1: clk_get(dev, "clk_out_1_mux") should work here by letting clk core to fall back to the clk_get_sys() by itself. Either way should be good.

clk_get uses device rather and dev_id will be name of this device and when clk_get fall back to __clk_get_sys() it still will use dev id of this device rather than actual dev_id that pmc clocks are added to the lookup. So clk_get_sys() seems to be correct to use as we can specify exact dev_id and con_id.

It should be better to use something "resource managed", thus devm_clk_get() should be a better choice.

Also, clk_find retrieves clock from lookup only when it finds matching clock with both dev_id and con_id as pmc clocks are registered with both dev_id and con_id.

I see existing clock driver adds both extern and pmc clocks (clk_out) to lookup with same dev_id of clk_out_1/2/3 and con_id of extern1/2/3 and with this always extern clock will be retrieved and this is probably because old DT and audio driver always uses extern1 rather than actual clk_out_1

But this need to be fixed now as we changed to use clk_out directly rather than extern (even for other pmc clocks) to match actual hw design.

Will fix this as well to register pmc clocks using con_id as clk_out_1/2/3 in pmc driver and extern clocks using con_id of extern1/2/3 with dev_id being NULL so we can retrieve these clocks by just using con_id only using clk_get_sys as we switched to use clk_out_1 directly as pmc clock rather than extern from DT and no longer need to pair pmc clocks to extern clocks.

I'm not sure it's worth the effort to care about con_ids if implicit fallback to clk_get_sys(NULL, "...") does the right thing for the audio driver.

IIRC, CCF uses variant of matching clocks by names, although I'm not sure whether that applies to older stable kernels.

[snip]