Re: [PATCH v6 4/5] drm/msm/dp: add support for DP PLL driver

From: tanmay
Date: Thu Jun 11 2020 - 16:31:45 EST

Hi Stephen,

Thanks for reviews.
Please ignore previous response to this patch. Here, I have re-organized it.

Thanks,

On 2020-06-11 13:07, tanmay@xxxxxxxxxxxxxx wrote:
On 2020-06-09 19:06, Stephen Boyd wrote:
Quoting Tanmay Shah (2020-06-08 20:46:23)
diff --git a/drivers/gpu/drm/msm/dp/dp_catalog.c b/drivers/gpu/drm/msm/dp/dp_catalog.c
index d02f4eb..2b982f0 100644
--- a/drivers/gpu/drm/msm/dp/dp_catalog.c
+++ b/drivers/gpu/drm/msm/dp/dp_catalog.c
@@ -5,6 +5,7 @@

#define pr_fmt(fmt) "[drm-dp] %s: " fmt, __func__

+#include <linux/rational.h>
#include <linux/delay.h>
#include <linux/iopoll.h>
#include <drm/drm_dp_helper.h>
@@ -134,59 +135,61 @@ static inline void dp_write_ahb(struct dp_catalog_private *catalog,
writel(data, catalog->io->dp_controller.base + offset);
}

-static inline u32 dp_read_cc(struct dp_catalog_private *catalog, u32 offset)
-{
- return readl_relaxed(catalog->io->dp_cc_io.base + offset);
-}
-

Why was this added in the first place? Remove it from the place it came
in please.

Sure. I will remove it as part of DP base driver patch.
static inline void dp_write_phy(struct dp_catalog_private *catalog,
u32 offset, u32 data)
{
+ offset += DP_PHY_REG_OFFSET;
/*
* To make sure phy reg writes happens before any other operation,
[...]
@@ -568,17 +574,37 @@ void dp_catalog_ctrl_config_msa(struct dp_catalog *dp_catalog,
bool fixed_nvid)
{
u32 pixel_m, pixel_n;
- u32 mvid, nvid;
+ u32 mvid, nvid, div, pixel_div = 0, dispcc_input_rate;
u32 const nvid_fixed = DP_LINK_CONSTANT_N_VALUE;
u32 const link_rate_hbr2 = 540000;
u32 const link_rate_hbr3 = 810000;
+ unsigned long den, num;

struct dp_catalog_private *catalog = container_of(dp_catalog,
struct dp_catalog_private, dp_catalog);

- pixel_m = dp_read_cc(catalog, MMSS_DP_PIXEL_M);
- pixel_n = dp_read_cc(catalog, MMSS_DP_PIXEL_N);
- DRM_DEBUG_DP("pixel_m=0x%x, pixel_n=0x%x\n", pixel_m, pixel_n);
+ div = dp_read_phy(catalog, REG_DP_PHY_VCO_DIV);

Why do we need to read the phy? The pixel_div seems to match what the
clk driver is doing so presumably we can make this follow the link rate
being used vs. having to read the phy.


As mentioned in above diagram, there is an additional divider in the
PLL after the VCO (vco_divided_clk_src).
The input rate to the dispcc branch is (vco_rate * 10)/ vco_dividied_clk_src.
In order to know the MNDs at the dispcc, we need to know the input rate.
This register read is to figure out which divider value is currently
being set in the PLL.
This input rate is not the same as the link rate. When we move the
PHY/PLL to a separate driver,
we would have take care of finding a different way to get this input rate.
+ div &= 0x03;
+
+ if (div == 0)
+ pixel_div = 6;
+ else if (div == 1)
+ pixel_div = 2;
+ else if (div == 2)
+ pixel_div = 4;
+ else
+ DRM_ERROR("Invalid pixel mux divider\n");
+
+ dispcc_input_rate = (rate * 10) / pixel_div;
+
+ rational_best_approximation(dispcc_input_rate, stream_rate_khz,
+ (unsigned long)(1 << 16) - 1,
+ (unsigned long)(1 << 16) - 1, &den, &num);
+
+ den = ~(den - num);
+ den = den & 0xFFFF;
+ pixel_m = num;
+ pixel_n = den;

mvid = (pixel_m & 0xFFFF) * 5;
nvid = (0xFFFF & (~pixel_n)) + (pixel_m & 0xFFFF);
diff --git a/drivers/gpu/drm/msm/dp/dp_pll_10nm.c b/drivers/gpu/drm/msm/dp/dp_pll_10nm.c
new file mode 100644
index 0000000..998d659
--- /dev/null
+++ b/drivers/gpu/drm/msm/dp/dp_pll_10nm.c
@@ -0,0 +1,903 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2016-2020, The Linux Foundation. All rights reserved.
+ */
+
+/*
+ * Display Port PLL driver block diagram for branch clocks
+ *
+ * +------------------------------+
+ * | DP_VCO_CLK |
+ * | |
+ * | +-------------------+ |
+ * | | (DP PLL/VCO) | |
+ * | +---------+---------+ |
+ * | v |
+ * | +----------+-----------+ |
+ * | | hsclk_divsel_clk_src | |
+ * | +----------+-----------+ |
+ * +------------------------------+
+ * |
+ * +---------<---------v------------>----------+
+ * | |
+ * +--------v---------+ |
+ * | dp_phy_pll | |
+ * | link_clk | |
+ * +--------+---------+ |
+ * | |
+ * | |
+ * v v
+ * Input to DISPCC block |
+ * for link clk, crypto clk |
+ * and interface clock |
+ * |
+ * |
+ * +--------<------------+-----------------+---<---+
+ * | | |
+ * +----v---------+ +--------v-----+ +--------v------+
+ * | vco_divided | | vco_divided | | vco_divided |
+ * | _clk_src | | _clk_src | | _clk_src |
+ * | | | | | |
+ * |divsel_six | | divsel_two | | divsel_four |
+ * +-------+------+ +-----+--------+ +--------+------+
+ * | | |
+ * v---->----------v-------------<------v
+ * |
+ * +----------+---------+
+ * | dp_phy_pll_vco |
+ * | div_clk |
+ * +---------+----------+
+ * |
+ * v
+ * Input to DISPCC block
+ * for DP pixel clock
+ *
+ */
+
+#include <linux/clk.h>
+#include <linux/delay.h>
+#include <linux/err.h>
+#include <linux/kernel.h>
+#include <linux/regmap.h>
+#include <linux/iopoll.h>

Should be a clk-provider.h include here given that this is providing
clks.

Yes. currently it is included in dp_parser.h but I will include here as well.
+
+#include "dp_hpd.h"
+#include "dp_pll.h"
+#include "dp_pll_private.h"
+
+#define NUM_PROVIDED_CLKS 2
+
+#define DP_LINK_CLK_SRC 0
+#define DP_PIXEL_CLK_SRC 1
+
+static struct dp_pll_db *dp_pdb;
+
+static const struct clk_ops dp_10nm_vco_clk_ops = {
+ .recalc_rate = dp_vco_recalc_rate_10nm,
+ .set_rate = dp_vco_set_rate_10nm,
+ .round_rate = dp_vco_round_rate_10nm,
+ .prepare = dp_vco_prepare_10nm,
+ .unprepare = dp_vco_unprepare_10nm,
+};
+
+struct dp_pll_10nm_pclksel {
+ struct clk_hw hw;
+
+ /* divider params */
+ u8 shift;
+ u8 width;
+ u8 flags; /* same flags as used by clk_divider struct */
+
+ struct dp_pll_db *pll;
+};
+
+#define to_pll_10nm_pclksel(_hw) \
+ container_of(_hw, struct dp_pll_10nm_pclksel, hw)
+
+static const struct clk_parent_data disp_cc_parent_data_0[] = {
+ { .fw_name = "bi_tcxo" },
+ { .fw_name = "dp_phy_pll_link_clk", .name = "dp_phy_pll_link_clk" },
+ { .fw_name = "dp_phy_pll_vco_div_clk",
+ .name = "dp_phy_pll_vco_div_clk"},
+ { .fw_name = "core_bi_pll_test_se", .name = "core_bi_pll_test_se" },
+};
+
+static struct dp_pll_vco_clk dp_vco_clk = {
+ .min_rate = DP_VCO_HSCLK_RATE_1620MHZDIV1000,
+ .max_rate = DP_VCO_HSCLK_RATE_8100MHZDIV1000,
+};
+
+static int dp_pll_mux_set_parent_10nm(struct clk_hw *hw, u8 val)
+{
+ struct dp_pll_10nm_pclksel *pclksel = to_pll_10nm_pclksel(hw);
+ struct dp_pll_db *dp_res = pclksel->pll;
+ struct dp_io_pll *pll_io = &dp_res->base->pll_io;
+ u32 auxclk_div;
+
+ auxclk_div = PLL_REG_R(pll_io->phy_base, REG_DP_PHY_VCO_DIV);
+ auxclk_div &= ~0x03;
+
+ if (val == 0)
+ auxclk_div |= 1;
+ else if (val == 1)
+ auxclk_div |= 2;
+ else if (val == 2)
+ auxclk_div |= 0;
+
+ PLL_REG_W(pll_io->phy_base,
+ REG_DP_PHY_VCO_DIV, auxclk_div);
+ DRM_DEBUG_DP("%s: mux=%d auxclk_div=%x\n", __func__, val, auxclk_div);
+
+ return 0;
+}
+
+static u8 dp_pll_mux_get_parent_10nm(struct clk_hw *hw)
+{
+ u32 auxclk_div = 0;
+ struct dp_pll_10nm_pclksel *pclksel = to_pll_10nm_pclksel(hw);
+ struct dp_pll_db *dp_res = pclksel->pll;
+ struct dp_io_pll *pll_io = &dp_res->base->pll_io;
+ u8 val = 0;
+
+ auxclk_div = PLL_REG_R(pll_io->phy_base, REG_DP_PHY_VCO_DIV);
+ auxclk_div &= 0x03;
+
+ if (auxclk_div == 1) /* Default divider */
+ val = 0;
+ else if (auxclk_div == 2)
+ val = 1;
+ else if (auxclk_div == 0)
+ val = 2;
+
+ DRM_DEBUG_DP("%s: auxclk_div=%d, val=%d\n", __func__, auxclk_div, val);
+
+ return val;
+}
+
+static int dp_pll_clk_mux_determine_rate(struct clk_hw *hw,
+ struct clk_rate_request *req)
+{
+ unsigned long rate = 0;
+ int ret = 0;
+
+ rate = clk_get_rate(hw->clk);
+
+ if (rate <= 0) {
+ DRM_ERROR("Rate is not set properly\n");
+ return -EINVAL;
+ }
+
+ req->rate = rate;
+
+ DRM_DEBUG_DP("%s: rate=%ld\n", __func__, req->rate);
+ /* Set the new parent of mux if there is a new valid parent */
+ if (hw->clk && req->best_parent_hw->clk) {
+ ret = clk_set_parent(hw->clk, req->best_parent_hw->clk);

Why do we need to call clk consumer APIs from the clk provider ops? This
is pretty confusing what's going on here.

Sure. Use of clk_set_parent is redundant here and I will remove it.
+ if (ret) {
+ DRM_ERROR("%s: clk_set_parent failed: ret=%d\n",
+ __func__, ret);
+ return ret;
+ }
+ }
+ return 0;
+}
+
+static unsigned long dp_pll_mux_recalc_rate(struct clk_hw *hw,
+ unsigned long parent_rate)
+{
+ struct clk_hw *div_clk_hw = NULL, *vco_clk_hw = NULL;
+ struct dp_pll_vco_clk *vco;
+
+ div_clk_hw = clk_hw_get_parent(hw);
+ if (!div_clk_hw)
+ return 0;
+
+ vco_clk_hw = clk_hw_get_parent(div_clk_hw);
+ if (!vco_clk_hw)
+ return 0;
+
+ vco = to_dp_vco_hw(vco_clk_hw);
+ if (!vco)
+ return 0;
+
+ if (vco->rate == DP_VCO_HSCLK_RATE_8100MHZDIV1000)
+ return (vco->rate / 6);
+ else if (vco->rate == DP_VCO_HSCLK_RATE_5400MHZDIV1000)
+ return (vco->rate / 4);
+ else
+ return (vco->rate / 2);
+}
+
+static int dp_pll_10nm_get_provider(struct msm_dp_pll *pll,
+ struct clk **link_clk_provider,
+ struct clk **pixel_clk_provider)
+{
+ struct clk_hw_onecell_data *hw_data = pll->hw_data;
+
+ if (link_clk_provider)
+ *link_clk_provider = hw_data->hws[DP_LINK_CLK_SRC]->clk;
+ if (pixel_clk_provider)
+ *pixel_clk_provider = hw_data->hws[DP_PIXEL_CLK_SRC]->clk;
+
+ return 0;
+}
+
+static const struct clk_ops dp_10nm_pclksel_clk_ops = {
+ .get_parent = dp_pll_mux_get_parent_10nm,
+ .set_parent = dp_pll_mux_set_parent_10nm,
+ .recalc_rate = dp_pll_mux_recalc_rate,
+ .determine_rate = dp_pll_clk_mux_determine_rate,
+};
+
+static struct clk_hw *dp_pll_10nm_pixel_clk_sel(struct dp_pll_db *pll_10nm)
+{
+ struct device *dev = &pll_10nm->pdev->dev;
+ struct dp_pll_10nm_pclksel *pll_pclksel;
+ struct clk_init_data pclksel_init = {
+ .parent_data = disp_cc_parent_data_0,
+ .num_parents = 3,
+ .name = "dp_phy_pll_vco_div_clk",

So the dp_phy_pll_vco_div_clk has a potential parent that is
dp_phy_pll_vco_div_clk. Huh?

Thats right. I will remove dp_phy_pll_vco_div_clk from parent list
disp_cc_parent_data_0.

+ .ops = &dp_10nm_pclksel_clk_ops,
+ };
+ int ret;
+
+ pll_pclksel = devm_kzalloc(dev, sizeof(*pll_pclksel), GFP_KERNEL);
+ if (!pll_pclksel)
+ return ERR_PTR(-ENOMEM);
+
+ pll_pclksel->pll = pll_10nm;
+ pll_pclksel->shift = 0;
+ pll_pclksel->width = 4;
+ pll_pclksel->flags = CLK_DIVIDER_ONE_BASED;

Is this flag used?

No it is redundant. I will remove.
+ pll_pclksel->hw.init = &pclksel_init;
+
+ ret = clk_hw_register(dev, &pll_pclksel->hw);
+ if (ret)
+ return ERR_PTR(ret);
+
+ return &pll_pclksel->hw;
+}
+
+static int dp_pll_10nm_register(struct dp_pll_db *pll_10nm)
+{
+ struct clk_hw_onecell_data *hw_data;
+ int ret;
+ struct clk_hw *hw;
+
+ struct msm_dp_pll *pll = pll_10nm->base;
+ struct device *dev = &pll_10nm->pdev->dev;
+ struct clk_hw **hws = pll_10nm->hws;
+ int num = 0;
+ struct clk_init_data vco_init = {
+ .parent_data = &(const struct clk_parent_data){
+ .fw_name = "bi_tcxo",
+ },
+ .num_parents = 1,
+ .name = "dp_vco_clk",
+ .ops = &dp_10nm_vco_clk_ops,
+ };

I thought the plan was to not have a vco clk? Just expose the two clks
for the link and the vco divider. Furthermore, drop the divider
"parents" and implement a single clk that programs the right divider
value for the various link rates chosen.

This will be taken care at later point of time.

+
+ DRM_DEBUG_DP("DP->id = %d", pll_10nm->id);
+
+ hw_data = devm_kzalloc(dev, sizeof(*hw_data) +
+ NUM_PROVIDED_CLKS * sizeof(struct clk_hw *),
+ GFP_KERNEL);
+ if (!hw_data)
+ return -ENOMEM;
+
+ dp_vco_clk.hw.init = &vco_init;
+ ret = clk_hw_register(dev, &dp_vco_clk.hw);
+ if (ret)
+ return ret;
+ hws[num++] = &dp_vco_clk.hw;
+
+ hw = clk_hw_register_fixed_factor(dev, "dp_phy_pll_link_clk",
+ "dp_vco_clk", CLK_SET_RATE_PARENT, 1, 10);
+
+ if (IS_ERR(hw))
+ return PTR_ERR(hw);
+ hws[num++] = hw;
+ hw_data->hws[DP_LINK_CLK_SRC] = hw;
+
+ hw = clk_hw_register_fixed_factor(dev, "dp_vco_divsel_two_clk_src",
+ "dp_vco_clk", 0, 1, 2);
+ if (IS_ERR(hw))
+ return PTR_ERR(hw);
+ hws[num++] = hw;
+
+ hw = clk_hw_register_fixed_factor(dev, "dp_vco_divsel_four_clk_src",
+ "dp_vco_clk", 0, 1, 4);
+ if (IS_ERR(hw))
+ return PTR_ERR(hw);
+ hws[num++] = hw;
+
+ hw = clk_hw_register_fixed_factor(dev, "dp_vco_divsel_six_clk_src",
+ "dp_vco_clk", 0, 1, 6);
+ if (IS_ERR(hw))
+ return PTR_ERR(hw);
+ hws[num++] = hw;
+
+ hw = dp_pll_10nm_pixel_clk_sel(pll_10nm);
+ if (IS_ERR(hw))
+ return PTR_ERR(hw);
+
+ hws[num++] = hw;
+ hw_data->hws[DP_PIXEL_CLK_SRC] = hw;
+
+ pll_10nm->num_hws = num;
+
+ hw_data->num = NUM_PROVIDED_CLKS;
+ pll->hw_data = hw_data;
+
+ ret = of_clk_add_hw_provider(dev->of_node, of_clk_hw_onecell_get,
+ pll->hw_data);
+ if (ret) {
+ DRM_DEV_ERROR(dev, "failed to register clk provider: %d\n",
+ ret);
+ return ret;
+ }
+
+ return ret;
+}
+
+int msm_dp_pll_10nm_init(struct msm_dp_pll *pll, int id)
+{
+ struct dp_pll_db *dp_10nm_pll;
+ struct platform_device *pdev = pll->pdev;
+ int ret;
+
+ dp_10nm_pll = devm_kzalloc(&pdev->dev,
+ sizeof(*dp_10nm_pll), GFP_KERNEL);
+ if (!dp_10nm_pll)
+ return -ENOMEM;
+
+ DRM_DEBUG_DP("DP PLL%d", id);
+
+ dp_10nm_pll->base = pll;
+ dp_10nm_pll->pdev = pll->pdev;
+ dp_10nm_pll->id = id;
+ dp_pdb = dp_10nm_pll;
+ pll->priv = (void *)dp_10nm_pll;
+ dp_vco_clk.priv = pll;
+
+ ret = of_property_read_u32(pdev->dev.of_node, "cell-index",
+ &dp_10nm_pll->index);
+ if (ret) {
+ DRM_ERROR("Unable to get the cell-index ret=%d\n", ret);
+ dp_10nm_pll->index = 0;
+ }

Is the cell-index used for anything?

No it is redundant and will be removed in next patch.
+
+ ret = dp_pll_10nm_register(dp_10nm_pll);
+ if (ret) {
+ DRM_DEV_ERROR(&pdev->dev, "failed to register PLL: %d\n", ret);
+ return ret;
+ }
+
+ pll->get_provider = dp_pll_10nm_get_provider;
+
+ return ret;
+}
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel