[PATCH] memory: tegra210-emc: Support Device Tree EMC Tables

[PATCH] memory: tegra210-emc: Support Device Tree EMC Tables

[Aaron Kling via B4 Relay]

From: Aaron Kling <webgeek1234@gmail.com>

These are generated by the Tegra210 Android bootloader. This is similar
to the Tegra124 handling, so the support is based on that and modified
to match Tegra210 by referencing the downstream Nvidia 4.9 kernel.

Signed-off-by: Aaron Kling <webgeek1234@gmail.com>
---
 drivers/memory/tegra/tegra210-emc-core.c | 246 +++++++++++++++++++++++++++++--
 1 file changed, 236 insertions(+), 10 deletions(-)

diff --git a/drivers/memory/tegra/tegra210-emc-core.c b/drivers/memory/tegra/tegra210-emc-core.c
index e63f6269057106ded054dea94d92d96cb9c13c06..0b8c7cd09679dc64b3fb04acf2bb5963dd7544fc 100644
--- a/drivers/memory/tegra/tegra210-emc-core.c
+++ b/drivers/memory/tegra/tegra210-emc-core.c
@@ -1783,6 +1783,226 @@ static void tegra210_emc_detect(struct tegra210_emc *emc)
 		emc->num_channels = 1;
 }
 
+static struct device_node *
+tegra_emc_find_node_by_ram_code(struct device_node *node, u32 ram_code)
+{
+	struct device_node *np;
+	int err;
+
+	for_each_child_of_node(node, np) {
+		u32 value;
+
+		err = of_property_read_u32(np, "nvidia,ram-code", &value);
+		if (err || (value != ram_code))
+			continue;
+
+		return np;
+	}
+
+	return NULL;
+}
+
+static int load_one_timing_from_dt(struct tegra210_emc *emc,
+				   struct tegra210_emc_timing *timing,
+				   struct device_node *node)
+{
+	int err;
+
+#define EMC_READ_PROP(prop, dtprop) { \
+	err = of_property_read_u32(node, dtprop, &timing->prop); \
+	if (err) { \
+		dev_err(emc->dev, "timing %pOFn: failed to read " #prop ": %d\n", \
+			node, err); \
+		return err; \
+	} \
+}
+
+#define EMC_READ_PROP_STRING(prop, dtprop) { \
+	err = of_property_read_string(node, dtprop, (const char **)&timing->prop); \
+	if (err) { \
+		dev_err(emc->dev, "timing %pOFn: failed to read " #prop ": %d\n", \
+			node, err); \
+		return err; \
+	} \
+}
+
+#define EMC_READ_PROP_ARRAY(prop, dtprop, length) { \
+	err = of_property_read_u32_array(node, dtprop, timing->prop, length); \
+	if (err) { \
+		dev_err(emc->dev, "timing %pOFn: failed to read " #prop ": %d\n", \
+			node, err); \
+		return err; \
+	} \
+}
+
+	EMC_READ_PROP_STRING(clock_src, "nvidia,source")
+	EMC_READ_PROP_STRING(dvfs_ver, "nvidia,dvfs-version")
+
+	EMC_READ_PROP(revision, "nvidia,revision")
+	EMC_READ_PROP(rate, "clock-frequency")
+	EMC_READ_PROP(min_volt, "nvidia,emc-min-mv")
+	EMC_READ_PROP(gpu_min_volt, "nvidia,gk20a-min-mv")
+	EMC_READ_PROP(clk_src_emc, "nvidia,src-sel-reg")
+	EMC_READ_PROP(num_burst, "nvidia,burst-regs-num")
+	EMC_READ_PROP(emc_cfg_2, "nvidia,emc-cfg-2")
+	EMC_READ_PROP(emc_sel_dpd_ctrl, "nvidia,emc-sel-dpd-ctrl")
+	EMC_READ_PROP(emc_auto_cal_config, "nvidia,emc-auto-cal-config")
+	EMC_READ_PROP(emc_auto_cal_config2, "nvidia,emc-auto-cal-config2")
+	EMC_READ_PROP(emc_auto_cal_config3, "nvidia,emc-auto-cal-config3")
+	EMC_READ_PROP(latency, "nvidia,emc-clock-latency-change")
+	EMC_READ_PROP_ARRAY(burst_regs, "nvidia,emc-registers", timing->num_burst)
+	EMC_READ_PROP(needs_training, "nvidia,needs-training")
+	EMC_READ_PROP(trained, "nvidia,trained")
+
+	if (timing->revision >= 0x6) {
+		EMC_READ_PROP(periodic_training, "nvidia,periodic_training")
+		EMC_READ_PROP(trained_dram_clktree[C0D0U0], "nvidia,trained_dram_clktree_c0d0u0")
+		EMC_READ_PROP(trained_dram_clktree[C0D0U1], "nvidia,trained_dram_clktree_c0d0u1")
+		EMC_READ_PROP(trained_dram_clktree[C0D1U0], "nvidia,trained_dram_clktree_c0d1u0")
+		EMC_READ_PROP(trained_dram_clktree[C0D1U1], "nvidia,trained_dram_clktree_c0d1u1")
+		EMC_READ_PROP(trained_dram_clktree[C1D0U0], "nvidia,trained_dram_clktree_c1d0u0")
+		EMC_READ_PROP(trained_dram_clktree[C1D0U1], "nvidia,trained_dram_clktree_c1d0u1")
+		EMC_READ_PROP(trained_dram_clktree[C1D1U0], "nvidia,trained_dram_clktree_c1d1u0")
+		EMC_READ_PROP(trained_dram_clktree[C1D1U1], "nvidia,trained_dram_clktree_c1d1u1")
+		EMC_READ_PROP(current_dram_clktree[C0D0U0], "nvidia,current_dram_clktree_c0d0u0")
+		EMC_READ_PROP(current_dram_clktree[C0D0U1], "nvidia,current_dram_clktree_c0d0u1")
+		EMC_READ_PROP(current_dram_clktree[C0D1U0], "nvidia,current_dram_clktree_c0d1u0")
+		EMC_READ_PROP(current_dram_clktree[C0D1U1], "nvidia,current_dram_clktree_c0d1u1")
+		EMC_READ_PROP(current_dram_clktree[C1D0U0], "nvidia,current_dram_clktree_c1d0u0")
+		EMC_READ_PROP(current_dram_clktree[C1D0U1], "nvidia,current_dram_clktree_c1d0u1")
+		EMC_READ_PROP(current_dram_clktree[C1D1U0], "nvidia,current_dram_clktree_c1d1u0")
+		EMC_READ_PROP(current_dram_clktree[C1D1U1], "nvidia,current_dram_clktree_c1d1u1")
+		EMC_READ_PROP(run_clocks, "nvidia,run_clocks")
+		EMC_READ_PROP(tree_margin, "nvidia,tree_margin")
+	}
+
+	EMC_READ_PROP(num_burst_per_ch, "nvidia,burst-regs-per-ch-num")
+	EMC_READ_PROP(num_trim, "nvidia,trim-regs-num")
+	EMC_READ_PROP(num_trim_per_ch, "nvidia,trim-regs-per-ch-num")
+	EMC_READ_PROP(num_mc_regs, "nvidia,burst-mc-regs-num")
+	EMC_READ_PROP(num_up_down, "nvidia,la-scale-regs-num")
+	EMC_READ_PROP(vref_num, "nvidia,vref-regs-num")
+	EMC_READ_PROP(dram_timing_num, "nvidia,dram-timing-regs-num")
+	EMC_READ_PROP(min_mrs_wait, "nvidia,min-mrs-wait")
+	EMC_READ_PROP(emc_mrw, "nvidia,emc-mrw")
+	EMC_READ_PROP(emc_mrw2, "nvidia,emc-mrw2")
+	EMC_READ_PROP(emc_mrw3, "nvidia,emc-mrw3")
+	EMC_READ_PROP(emc_mrw4, "nvidia,emc-mrw4")
+	EMC_READ_PROP(emc_mrw9, "nvidia,emc-mrw9")
+	EMC_READ_PROP(emc_mrs, "nvidia,emc-mrs")
+	EMC_READ_PROP(emc_emrs, "nvidia,emc-emrs")
+	EMC_READ_PROP(emc_emrs2, "nvidia,emc-emrs2")
+	EMC_READ_PROP(emc_auto_cal_config4, "nvidia,emc-auto-cal-config4")
+	EMC_READ_PROP(emc_auto_cal_config5, "nvidia,emc-auto-cal-config5")
+	EMC_READ_PROP(emc_auto_cal_config6, "nvidia,emc-auto-cal-config6")
+	EMC_READ_PROP(emc_auto_cal_config7, "nvidia,emc-auto-cal-config7")
+	EMC_READ_PROP(emc_auto_cal_config8, "nvidia,emc-auto-cal-config8")
+	EMC_READ_PROP(emc_fdpd_ctrl_cmd_no_ramp, "nvidia,emc-fdpd-ctrl-cmd-no-ramp")
+	EMC_READ_PROP(dll_clk_src, "nvidia,dll-clk-src")
+	EMC_READ_PROP(clk_out_enb_x_0_clk_enb_emc_dll, "nvidia,clk-out-enb-x-0-clk-enb-emc-dll")
+
+	if (timing->revision >= 0x7)
+		EMC_READ_PROP_ARRAY(ptfv_list, "nvidia,ptfv", ARRAY_SIZE(timing->ptfv_list))
+
+	EMC_READ_PROP_ARRAY(burst_reg_per_ch, "nvidia,emc-burst-regs-per-ch",
+			timing->num_burst_per_ch)
+	EMC_READ_PROP_ARRAY(shadow_regs_ca_train, "nvidia,emc-shadow-regs-ca-train",
+			timing->num_burst)
+	EMC_READ_PROP_ARRAY(shadow_regs_quse_train, "nvidia,emc-shadow-regs-quse-train",
+			timing->num_burst)
+	EMC_READ_PROP_ARRAY(shadow_regs_rdwr_train, "nvidia,emc-shadow-regs-rdwr-train",
+			timing->num_burst)
+	EMC_READ_PROP_ARRAY(trim_regs, "nvidia,emc-trim-regs", timing->num_trim)
+	EMC_READ_PROP_ARRAY(trim_perch_regs, "nvidia,emc-trim-regs-per-ch", timing->num_trim_per_ch)
+	EMC_READ_PROP_ARRAY(vref_perch_regs, "nvidia,emc-vref-regs", timing->vref_num)
+	EMC_READ_PROP_ARRAY(dram_timings, "nvidia,emc-dram-timing-regs", timing->dram_timing_num)
+	EMC_READ_PROP_ARRAY(burst_mc_regs, "nvidia,emc-burst-mc-regs", timing->num_mc_regs)
+	EMC_READ_PROP_ARRAY(la_scale_regs, "nvidia,emc-la-scale-regs", timing->num_up_down)
+
+#undef EMC_READ_PROP
+#undef EMC_READ_STRING
+#undef EMC_READ_PROP_ARRAY
+
+	return 0;
+}
+
+#define NOMINAL_COMPATIBLE "nvidia,tegra21-emc-table"
+#define DERATED_COMPATIBLE "nvidia,tegra21-emc-table-derated"
+static int tegra210_emc_load_timings_from_dt(struct tegra210_emc *emc,
+					     struct device_node *node)
+{
+	struct tegra210_emc_timing *timing;
+	unsigned int num_nominal = 0, num_derated = 0;
+	int err;
+
+	emc->num_timings = 0;
+	for_each_child_of_node_scoped(node, child) {
+		if (of_device_is_compatible(child, NOMINAL_COMPATIBLE))
+			emc->num_timings++;
+		else if (of_device_is_compatible(child, DERATED_COMPATIBLE))
+			num_derated++;
+	}
+
+	if (!emc->num_timings || (num_derated && (emc->num_timings != num_derated)))
+		return -EINVAL;
+
+	emc->nominal = devm_kcalloc(emc->dev, emc->num_timings, sizeof(*timing),
+				    GFP_KERNEL);
+	if (!emc->nominal)
+		return -ENOMEM;
+
+	if (num_derated) {
+		num_derated = 0;
+		emc->derated = devm_kcalloc(emc->dev, emc->num_timings, sizeof(*timing),
+					    GFP_KERNEL);
+		if (!emc->derated)
+			return -ENOMEM;
+	}
+
+	for_each_child_of_node_scoped(node, child) {
+		if (of_device_is_compatible(child, NOMINAL_COMPATIBLE))
+			timing = &emc->nominal[num_nominal++];
+		else if (of_device_is_compatible(child, DERATED_COMPATIBLE))
+			timing = &emc->derated[num_derated++];
+		else
+			continue;
+
+		err = load_one_timing_from_dt(emc, timing, child);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+static int tegra210_emc_parse_dt(struct tegra210_emc *emc)
+{
+	struct device_node *node, *np = emc->dev->of_node;
+	int ram_code, ret = 0;
+
+	if (!np) {
+		dev_err(emc->dev, "Unable to find emc node\n");
+		return -ENODEV;
+	}
+
+	if (of_find_property(np, "nvidia,use-ram-code", NULL)) {
+		ram_code = tegra_read_ram_code();
+		node = tegra_emc_find_node_by_ram_code(np, ram_code);
+
+		if (!node) {
+			dev_warn(emc->dev, "can't find emc table for ram-code\n");
+			return -ENODEV;
+		}
+
+		ret = tegra210_emc_load_timings_from_dt(emc, node);
+		of_node_put(node);
+	} else {
+		ret = tegra210_emc_load_timings_from_dt(emc, np);
+	}
+
+	return ret;
+}
+
 static int tegra210_emc_validate_timings(struct tegra210_emc *emc,
 					 struct tegra210_emc_timing *timings,
 					 unsigned int num_timings)
@@ -1815,6 +2035,7 @@ static int tegra210_emc_probe(struct platform_device *pdev)
 	struct device_node *np;
 	unsigned int i;
 	int err;
+	bool have_dt_tables = false;
 
 	emc = devm_kzalloc(&pdev->dev, sizeof(*emc), GFP_KERNEL);
 	if (!emc)
@@ -1847,16 +2068,20 @@ static int tegra210_emc_probe(struct platform_device *pdev)
 	np = pdev->dev.of_node;
 
 	/* attach to the nominal and (optional) derated tables */
-	err = of_reserved_mem_device_init_by_name(emc->dev, np, "nominal");
-	if (err < 0) {
-		dev_err(emc->dev, "failed to get nominal EMC table: %d\n", err);
-		return err;
-	}
+	if (of_reserved_mem_device_init_by_name(emc->dev, np, "nominal") >= 0) {
+		err = of_reserved_mem_device_init_by_name(emc->dev, np, "derated");
+		if (err < 0 && err != -ENODEV) {
+			dev_err(emc->dev, "failed to get derated EMC table: %d\n", err);
+			goto release;
+		}
+	} else {
+		err = tegra210_emc_parse_dt(emc);
+		if (err < 0) {
+			dev_err(emc->dev, "failed to get EMC tables: %d\n", err);
+			return err;
+		}
 
-	err = of_reserved_mem_device_init_by_name(emc->dev, np, "derated");
-	if (err < 0 && err != -ENODEV) {
-		dev_err(emc->dev, "failed to get derated EMC table: %d\n", err);
-		goto release;
+		have_dt_tables = true;
 	}
 
 	/* validate the tables */
@@ -1980,7 +2205,8 @@ static int tegra210_emc_probe(struct platform_device *pdev)
 	debugfs_remove_recursive(emc->debugfs.root);
 	tegra210_clk_emc_detach(emc->clk);
 release:
-	of_reserved_mem_device_release(emc->dev);
+	if (!have_dt_tables)
+		of_reserved_mem_device_release(emc->dev);
 
 	return err;
 }

---
base-commit: 8bac8898fe398ffa3e09075ecea2be511725fb0b
change-id: 20250429-tegra210-emc-dt-97dce690ad4e

Best regards,
-- 
Aaron Kling <webgeek1234@gmail.com>

Re: [PATCH] memory: tegra210-emc: Support Device Tree EMC Tables

[Krzysztof Kozlowski]

On 30/04/2025 19:52, Aaron Kling via B4 Relay wrote:
> +#undef EMC_READ_PROP
> +#undef EMC_READ_STRING
> +#undef EMC_READ_PROP_ARRAY
> +
> +	return 0;
> +}
> +
> +#define NOMINAL_COMPATIBLE "nvidia,tegra21-emc-table"
> +#define DERATED_COMPATIBLE "nvidia,tegra21-emc-table-derated"

No, you cannot add undocumented compatibles. Missing bindings.

> +static int tegra210_emc_load_timings_from_dt(struct tegra210_emc *emc,
> +					     struct device_node *node)
> +{
> +	struct tegra210_emc_timing *timing;
> +	unsigned int num_nominal = 0, num_derated = 0;
> +	int err;
> +
> +	emc->num_timings = 0;
> +	for_each_child_of_node_scoped(node, child) {
> +		if (of_device_is_compatible(child, NOMINAL_COMPATIBLE))
> +			emc->num_timings++;
> +		else if (of_device_is_compatible(child, DERATED_COMPATIBLE))
> +			num_derated++;
> +	}
> +
> +	if (!emc->num_timings || (num_derated && (emc->num_timings != num_derated)))
> +		return -EINVAL;
> +
> +	emc->nominal = devm_kcalloc(emc->dev, emc->num_timings, sizeof(*timing),
> +				    GFP_KERNEL);
> +	if (!emc->nominal)
> +		return -ENOMEM;
> +
> +	if (num_derated) {
> +		num_derated = 0;
> +		emc->derated = devm_kcalloc(emc->dev, emc->num_timings, sizeof(*timing),
> +					    GFP_KERNEL);
> +		if (!emc->derated)
> +			return -ENOMEM;
> +	}
> +
> +	for_each_child_of_node_scoped(node, child) {
> +		if (of_device_is_compatible(child, NOMINAL_COMPATIBLE))
> +			timing = &emc->nominal[num_nominal++];
> +		else if (of_device_is_compatible(child, DERATED_COMPATIBLE))
> +			timing = &emc->derated[num_derated++];
> +		else
> +			continue;
> +
> +		err = load_one_timing_from_dt(emc, timing, child);
> +		if (err)
> +			return err;
> +	}
> +
> +	return 0;
> +}
> +
> +static int tegra210_emc_parse_dt(struct tegra210_emc *emc)
> +{
> +	struct device_node *node, *np = emc->dev->of_node;
> +	int ram_code, ret = 0;
> +
> +	if (!np) {
> +		dev_err(emc->dev, "Unable to find emc node\n");
> +		return -ENODEV;
> +	}
> +
> +	if (of_find_property(np, "nvidia,use-ram-code", NULL)) {

I cannot find the bindings for this. Where is your DTS? Was it tested?

It seems nothing here is documented.

Best regards,
Krzysztof

Re: [PATCH] memory: tegra210-emc: Support Device Tree EMC Tables

[Aaron Kling]

On Sun, May 4, 2025 at 8:38 AM Krzysztof Kozlowski <krzk@kernel.org> wrote:
>
> On 30/04/2025 19:52, Aaron Kling via B4 Relay wrote:
> > +#undef EMC_READ_PROP
> > +#undef EMC_READ_STRING
> > +#undef EMC_READ_PROP_ARRAY
> > +
> > +     return 0;
> > +}
> > +
> > +#define NOMINAL_COMPATIBLE "nvidia,tegra21-emc-table"
> > +#define DERATED_COMPATIBLE "nvidia,tegra21-emc-table-derated"
>
> No, you cannot add undocumented compatibles. Missing bindings.
>
> > +static int tegra210_emc_load_timings_from_dt(struct tegra210_emc *emc,
> > +                                          struct device_node *node)
> > +{
> > +     struct tegra210_emc_timing *timing;
> > +     unsigned int num_nominal = 0, num_derated = 0;
> > +     int err;
> > +
> > +     emc->num_timings = 0;
> > +     for_each_child_of_node_scoped(node, child) {
> > +             if (of_device_is_compatible(child, NOMINAL_COMPATIBLE))
> > +                     emc->num_timings++;
> > +             else if (of_device_is_compatible(child, DERATED_COMPATIBLE))
> > +                     num_derated++;
> > +     }
> > +
> > +     if (!emc->num_timings || (num_derated && (emc->num_timings != num_derated)))
> > +             return -EINVAL;
> > +
> > +     emc->nominal = devm_kcalloc(emc->dev, emc->num_timings, sizeof(*timing),
> > +                                 GFP_KERNEL);
> > +     if (!emc->nominal)
> > +             return -ENOMEM;
> > +
> > +     if (num_derated) {
> > +             num_derated = 0;
> > +             emc->derated = devm_kcalloc(emc->dev, emc->num_timings, sizeof(*timing),
> > +                                         GFP_KERNEL);
> > +             if (!emc->derated)
> > +                     return -ENOMEM;
> > +     }
> > +
> > +     for_each_child_of_node_scoped(node, child) {
> > +             if (of_device_is_compatible(child, NOMINAL_COMPATIBLE))
> > +                     timing = &emc->nominal[num_nominal++];
> > +             else if (of_device_is_compatible(child, DERATED_COMPATIBLE))
> > +                     timing = &emc->derated[num_derated++];
> > +             else
> > +                     continue;
> > +
> > +             err = load_one_timing_from_dt(emc, timing, child);
> > +             if (err)
> > +                     return err;
> > +     }
> > +
> > +     return 0;
> > +}
> > +
> > +static int tegra210_emc_parse_dt(struct tegra210_emc *emc)
> > +{
> > +     struct device_node *node, *np = emc->dev->of_node;
> > +     int ram_code, ret = 0;
> > +
> > +     if (!np) {
> > +             dev_err(emc->dev, "Unable to find emc node\n");
> > +             return -ENODEV;
> > +     }
> > +
> > +     if (of_find_property(np, "nvidia,use-ram-code", NULL)) {
>
> I cannot find the bindings for this. Where is your DTS? Was it tested?
>
> It seems nothing here is documented.

The relevant nodes are filled in by the bootloader. I had hoped that
the early stage bootloader would be able to copy that into the
mainline dt without needing to prime said mainline dt, but that's not
working as hoped. I'll send a v2 with dt bindings and the first usage
of it in p2371-2180.

Sincerely,
Aaron

Re: [PATCH] memory: tegra210-emc: Support Device Tree EMC Tables

[Krzysztof Kozlowski]

On 04/05/2025 17:58, Aaron Kling wrote:
>>> +static int tegra210_emc_parse_dt(struct tegra210_emc *emc)
>>> +{
>>> +     struct device_node *node, *np = emc->dev->of_node;
>>> +     int ram_code, ret = 0;
>>> +
>>> +     if (!np) {
>>> +             dev_err(emc->dev, "Unable to find emc node\n");
>>> +             return -ENODEV;
>>> +     }
>>> +
>>> +     if (of_find_property(np, "nvidia,use-ram-code", NULL)) {
>>
>> I cannot find the bindings for this. Where is your DTS? Was it tested?
>>
>> It seems nothing here is documented.
> 
> The relevant nodes are filled in by the bootloader. I had hoped that
> the early stage bootloader would be able to copy that into the

It does not matter. You cannot have compatibles and ABI here without
documenting that ABI.



Best regards,
Krzysztof

Re: [PATCH] memory: tegra210-emc: Support Device Tree EMC Tables

[Aaron Kling]

On Sun, May 4, 2025 at 11:13 AM Krzysztof Kozlowski <krzk@kernel.org> wrote:
>
> On 04/05/2025 17:58, Aaron Kling wrote:
> >>> +static int tegra210_emc_parse_dt(struct tegra210_emc *emc)
> >>> +{
> >>> +     struct device_node *node, *np = emc->dev->of_node;
> >>> +     int ram_code, ret = 0;
> >>> +
> >>> +     if (!np) {
> >>> +             dev_err(emc->dev, "Unable to find emc node\n");
> >>> +             return -ENODEV;
> >>> +     }
> >>> +
> >>> +     if (of_find_property(np, "nvidia,use-ram-code", NULL)) {
> >>
> >> I cannot find the bindings for this. Where is your DTS? Was it tested?
> >>
> >> It seems nothing here is documented.
> >
> > The relevant nodes are filled in by the bootloader. I had hoped that
> > the early stage bootloader would be able to copy that into the
>
> It does not matter. You cannot have compatibles and ABI here without
> documenting that ABI.

Understood, I will keep that in mind should the situation arise again.
And it will be rectified for this patch in v2.

Sincerely,
Aaron