ARM: OMAP2+: Fix populating the hwmod data from device tree

We have some device tree properties where the ti,hwmod has multiple values:

am33xx.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";
am4372.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";
dra7.dtsi: ti,hwmods = "l3_main_1", "l3_main_2";
omap3.dtsi: ti,hwmods = "mcbsp2", "mcbsp2_sidetone";
omap3.dtsi: ti,hwmods = "mcbsp3", "mcbsp3_sidetone";
omap4.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";
omap5.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";

So we need to handle the whole string array instead of just the first string to find the related hwmod entry.

Cc "Beno?t Cousson" <[email protected]>

Cc Paul Walmsley <[email protected]>

Signed-off-by Tony Lindgren <[email protected]>

arch/arm/mach-omap2/omap_hwmod.c +8 -3

@@ -2228,11 +2228,23 @@ static struct device_node *of_dev_hwmod_lookup(struct device_node *np,
 						struct omap_hwmod *oh)
 {
 	struct device_node *np0 = NULL, *np1 = NULL;
-	const char *p;
 
 	for_each_child_of_node(np, np0) {
-		if (of_find_property(np0, "ti,hwmods", NULL)) {
-			p = of_get_property(np0, "ti,hwmods", NULL);
+		int count, i;
+
+		count = of_property_count_strings(np0, "ti,hwmods");
+		if (count < 1)
+			continue;
+
+		for (i = 0; i < count; i++) {
+			const char *p;

Quoted 5 lines · Tony Lindgren

We have some device tree properties where the ti,hwmod has multiple values:

am33xx.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";
am4372.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";

Quoted 1 line

dra7.dtsi: ti,hwmods = "l3_main_1", "l3_main_2";

Quoted 2 lines

omap3.dtsi: ti,hwmods = "mcbsp2", "mcbsp2_sidetone";
omap3.dtsi: ti,hwmods = "mcbsp3", "mcbsp3_sidetone";

Quoted 2 lines

omap4.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";
omap5.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";

Hmm I think this also needs part two to it to populate the right IO space based on the index, this probably now wrongly populates the first IO space for all the hwmod instances in the group.

Quoted 6 lines · Tony Lindgren

Quoted 5 lines · Tony Lindgren

We have some device tree properties where the ti,hwmod has multiple values:

am33xx.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";
am4372.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";

I think these should be defined in the .dtsi files as separate devices and removed from the related omap_hwmod_*_data.c files.

Quoted 1 line

Quoted 1 line

dra7.dtsi: ti,hwmods = "l3_main_1", "l3_main_2";

These too should be defined in the .dtsi file. Looks like we're missing the related data in the omap_hwmod_7xx_data.c file anyways?

Quoted 2 lines

Quoted 2 lines

omap3.dtsi: ti,hwmods = "mcbsp2", "mcbsp2_sidetone";
omap3.dtsi: ti,hwmods = "mcbsp3", "mcbsp3_sidetone";

OK the sidetone data we can probably parse using the ti,hwmods as an index.

Quoted 2 lines

Quoted 2 lines

omap4.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";
omap5.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";

Benoit, I guess these mean that those devices should be created dynamically, right?

Yet we've removed the related data in commit 3b9b10 and are not passing it in the .dtsi file?

So these should be created as separate devices in the .dtsi file instead AFAIK.

Quoted 3 lines

Hmm I think this also needs part two to it to populate the right IO space based on the index, this probably now wrongly populates the first IO space for all the hwmod instances in the group.

Yeah this patch clearly is not the right fix to the problem of missing data.

Quoted 34 lines · Tony Lindgren

Quoted 6 lines · Tony Lindgren

Quoted 5 lines · Tony Lindgren

We have some device tree properties where the ti,hwmod has multiple values:

am33xx.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";
am4372.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";

I think these should be defined in the .dtsi files as separate devices and removed from the related omap_hwmod_*_data.c files.

Quoted 1 line

Quoted 1 line

dra7.dtsi: ti,hwmods = "l3_main_1", "l3_main_2";

These too should be defined in the .dtsi file. Looks like we're missing the related data in the omap_hwmod_7xx_data.c file anyways?

Quoted 2 lines

Quoted 2 lines

omap3.dtsi: ti,hwmods = "mcbsp2", "mcbsp2_sidetone";
omap3.dtsi: ti,hwmods = "mcbsp3", "mcbsp3_sidetone";

OK the sidetone data we can probably parse using the ti,hwmods as an index.

Quoted 2 lines

Quoted 2 lines

omap4.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";
omap5.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";

Benoit, I guess these mean that those devices should be created dynamically, right?

Yet we've removed the related data in commit 3b9b10 and are not passing it in the .dtsi file?

So these should be created as separate devices in the .dtsi file instead AFAIK.

Quoted 6 lines

Quoted 3 lines

Hmm I think this also needs part two to it to populate the right IO space based on the index, this probably now wrongly populates the first IO space for all the hwmod instances in the group.

Yeah this patch clearly is not the right fix to the problem of missing data.

In most of the cases with multiple hwmods per DT device, there's probably something wrong with the data. So we should try to remove those when possible.

For example if you look at the SIDETONE blocks, those should definitely be separate DT devices - they have separate address spaces, SYSCONFIG registers, etc. (section 21.6.2 "SIDETONE register mapping summary", OMAP36xx TRM)

Would strongly suspect that's the case also for the EDMA blocks you mention.

Am not 100% sure about the L3 cases; would have to dig deeper.

They at least had interrupts listed looking at commit 3b9b10. Probably the thing to do for now is to revert those changes, and see if we can just remove the L3 entries from the .dtsi files.

Quoted 4 lines · Tony Lindgren

They at least had interrupts listed looking at commit 3b9b10. Probably the thing to do for now is to revert those changes, and see if we can just remove the L3 entries from the .dtsi files.

Actually the patch I posted should be able to handle also the ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3" property in omap4.dtsi, but we're not currently parsing that as we only look at the children and not the properties of the OCP bus. Should be fixable, will take a look tomorrow if this approach makes sense to you.

Quoted 6 lines · Tony Lindgren

Quoted 4 lines · Tony Lindgren

They at least had interrupts listed looking at commit 3b9b10. Probably the thing to do for now is to revert those changes, and see if we can just remove the L3 entries from the .dtsi files.

Quoted 5 lines

Actually the patch I posted should be able to handle also the ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3" property in omap4.dtsi, but we're not currently parsing that as we only look at the children and not the properties of the OCP bus. Should be fixable, will take a look tomorrow if this approach makes sense to you.

OK this one seems to do the right thing for me.

Tony

From: Tony Lindgren <[email protected]>
Date: Wed, 20 Nov 2013 15:46:51 -0800
Subject: [PATCH] ARM: OMAP2+: Fix overwriting hwmod data with data from device tree
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

We have some device tree properties where the ti,hwmod have multiple values:

am33xx.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";
am4372.dtsi: ti,hwmods = "tpcc", "tptc0", "tptc1", "tptc2";
dra7.dtsi: ti,hwmods = "l3_main_1", "l3_main_2";
omap3.dtsi: ti,hwmods = "mcbsp2", "mcbsp2_sidetone";
omap3.dtsi: ti,hwmods = "mcbsp3", "mcbsp3_sidetone";
omap4.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";
omap5.dtsi: ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3";

That's not correct way of doing things in this case because these are separate devices with their own address space, interrupts, SYSCONFIG registers and can set their PM states independently.

So they should all be fixed up to be separate devices in the .dts files.

We also have the related data removed for at least omap4 in commit 3b9b10151c68 (ARM: OMAP4: hwmod data: Clean up the data file), so that data is wrongly initialized as null data.

So we need to fix two bugs:

1. We are only checking the first entry of the ti,hwmods property

This means that we're only initializing the first hwmods entry instead of the ones listed in the ti,hwmods property.

2. We are only checking the child nodes, not the nodes themselves

This means that anything listed at OCP level is currently just ignored and unitialized and at least the omap4 case, with the legacy data missing from the hwmod.

Fix both of the issues by using an index to the ti,hwmods property and changing the hwmod lookup function to also check the current node for ti,hwmods property instead of just the children.

While at it, let's also add some warnings for the bad data so it's easier to fix.

Cc "Beno?t Cousson" <[email protected]>

Cc Paul Walmsley <[email protected]>

Signed-off-by Tony Lindgren <[email protected]>

arch/arm/mach-omap2/omap_hwmod.c +78 -21

index e3f0eca..ee655da 100644
@@ -2326,38 +2326,80 @@ static int _shutdown(struct omap_hwmod *oh)
 	return 0;
 }
 
+static int of_dev_find_hwmod(struct device_node *np,
+			     struct omap_hwmod *oh)
+{
+	int count, i, res;
+	const char *p;
+
+	count = of_property_count_strings(np, "ti,hwmods");
+	if (count < 1)
+		return -ENODEV;
+
+	for (i = 0; i < count; i++) {
+		res = of_property_read_string_index(np, "ti,hwmods",
+						    i, &p);
+		if (res)
+			continue;
+		if (!strcmp(p, oh->name)) {
+			pr_debug("omap_hwmod: dt %s[%i] uses hwmod %s\n",
+				 np->name, i, oh->name);
+			return i;
+		}
+	}
+
+	return -ENODEV;
+}
+
 /**
  * of_dev_hwmod_lookup - look up needed hwmod from dt blob
  * @np: struct device_node *
  * @oh: struct omap_hwmod *
+ * @index: index of the entry found
+ * @found: struct device_node * found or NULL
  *
  * Parse the dt blob and find out needed hwmod. Recursive function is
  * implemented to take care hierarchical dt blob parsing.
- * Return: The device node on success or NULL on failure.
+ * Return: Returns 0 on success, -ENODEV when not found.
  */
-static struct device_node *of_dev_hwmod_lookup(struct device_node *np,
-						struct omap_hwmod *oh)
+static int of_dev_hwmod_lookup(struct device_node *np,
+			       struct omap_hwmod *oh,
+			       int *index,
+			       struct device_node **found)
 {
-	struct device_node *np0 = NULL, *np1 = NULL;
-	const char *p;
+	struct device_node *np0 = NULL;
+	int res;
+
+	res = of_dev_find_hwmod(np, oh);
+	if (res >= 0) {
+		*found = np;
+		*index = res;
+		return 0;
+	}
 
 	for_each_child_of_node(np, np0) {
-		if (of_find_property(np0, "ti,hwmods", NULL)) {
-			p = of_get_property(np0, "ti,hwmods", NULL);
-			if (!strcmp(p, oh->name))
-				return np0;
-			np1 = of_dev_hwmod_lookup(np0, oh);
-			if (np1)
-				return np1;
+		struct device_node *fc;
+		int i;
+
+		res = of_dev_hwmod_lookup(np0, oh, &i, &fc);
+		if (res == 0) {
+			*found = fc;
+			*index = i;
+			return 0;
 		}
 	}
-	return NULL;
+
+	*found = NULL;
+	*index = 0;
+
+	return -ENODEV;
 }
 
 /**
  * _init_mpu_rt_base - populate the virtual address for a hwmod
  * @oh: struct omap_hwmod * to locate the virtual address
  * @data: (unused, caller should pass NULL)
+ * @index: index of the reg entry iospace in device tree
  * @np: struct device_node * of the IP block's device node in the DT data
  *
  * Cache the virtual address used by the MPU to access this IP block's

@@ -2368,7 +2410,7 @@ static struct device_node *of_dev_hwmod_lookup(struct device_node *np,
  * -ENXIO on absent or invalid register target address space.
  */
 static int __init _init_mpu_rt_base(struct omap_hwmod *oh, void *data,
-				    struct device_node *np)
+				    int index, struct device_node *np)
 {
 	struct omap_hwmod_addr_space *mem;
 	void __iomem *va_start = NULL;

@@ -2390,13 +2432,17 @@ static int __init _init_mpu_rt_base(struct omap_hwmod *oh, void *data,
 		if (!np)
 			return -ENXIO;
 
-		va_start = of_iomap(np, oh->mpu_rt_idx);
+		va_start = of_iomap(np, index + oh->mpu_rt_idx);
 	} else {
 		va_start = ioremap(mem->pa_start, mem->pa_end - mem->pa_start);
 	}
 
 	if (!va_start) {
-		pr_err("omap_hwmod: %s: Could not ioremap\n", oh->name);
+		if (mem)
+			pr_err("omap_hwmod: %s: Could not ioremap\n", oh->name);
+		else
+			pr_err("omap_hwmod: %s: Missing dt reg%i for %s\n",
+			       oh->name, index, np->full_name);
 		return -ENXIO;
 	}
 

@@ -2422,17 +2468,29 @@ static int __init _init_mpu_rt_base(struct omap_hwmod *oh, void *data,
  */
 static int __init _init(struct omap_hwmod *oh, void *data)
 {
-	int r;
+	int r, index;
 	struct device_node *np = NULL;
 
 	if (oh->_state != _HWMOD_STATE_REGISTERED)
 		return 0;
 
-	if (of_have_populated_dt())
-		np = of_dev_hwmod_lookup(of_find_node_by_name(NULL, "ocp"), oh);
+	if (of_have_populated_dt()) {
+		struct device_node *bus;
+
+		bus = of_find_node_by_name(NULL, "ocp");
+		if (!bus)
+			return -ENODEV;
+
+		r = of_dev_hwmod_lookup(bus, oh, &index, &np);
+		if (r)
+			pr_debug("omap_hwmod: %s missing dt data\n", oh->name);
+		else if (np && index)
+			pr_warn("omap_hwmod: %s using broken dt data from %s\n",
+				oh->name, np->name);
+	}
 
 	if (oh->class->sysc) {
-		r = _init_mpu_rt_base(oh, NULL, np);
+		r = _init_mpu_rt_base(oh, NULL, index, np);
 		if (r < 0) {
 			WARN(1, "omap_hwmod: %s: doesn't have mpu register target base\n",
 			     oh->name);
--

To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

Quoted 13 lines · Tony Lindgren

Quoted 11 lines · Tony Lindgren

Quoted 4 lines · Tony Lindgren

They at least had interrupts listed looking at commit 3b9b10. Probably the thing to do for now is to revert those changes, and see if we can just remove the L3 entries from the .dtsi files.

Actually the patch I posted should be able to handle also the ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3" property in omap4.dtsi, but we're not currently parsing that as we only look at the children and not the properties of the OCP bus. Should be fixable, will take a look tomorrow if this approach makes sense to you.

Quoted 1 line

OK this one seems to do the right thing for me.

No comments? I'll queue the patch below to the fixes, please yell if you see any issues with that.

Quoted 16 lines · Tony Lindgren

Quoted 14 lines · Tony Lindgren

Quoted 11 lines · Tony Lindgren

Quoted 4 lines · Tony Lindgren

They at least had interrupts listed looking at commit 3b9b10. Probably the thing to do for now is to revert those changes, and see if we can just remove the L3 entries from the .dtsi files.

Actually the patch I posted should be able to handle also the ti,hwmods = "l3_main_1", "l3_main_2", "l3_main_3" property in omap4.dtsi, but we're not currently parsing that as we only look at the children and not the properties of the OCP bus. Should be fixable, will take a look tomorrow if this approach makes sense to you.

OK this one seems to do the right thing for me.

Quoted 2 lines

No comments? I'll queue the patch below to the fixes, please yell if you see any issues with that.

Looks reasonable to me based on a quick glance:

Acked-by Paul Walmsley <[email protected]>

Have not tested it though.

I like the warning message for the bad data.