Skip to contentSkip to conversation

Allow parsing more than one memory node

PATCH Superseded Clement Leger 5 participants 14 messages Mar 16, 2020 – Jun 18, 2020 Device tree

lore
Share & tools
lore
#0 lore raw

cover Allow parsing more than one memory node

Currently, barebox can only parse one memory node. This means that all other memory nodes (with device_type == "memory") will be ignored while parsing memory.
This patchset add a function of_probe_memory which will call of_add_memory for each found memory node. Additionally, of_add_memory has been modified to allow being called multiple times and correctly add all memory banks.

Clement Leger (2):
of: base: parse all available memory nodes
of: base: allow of_add_memory to be called multiple times

 drivers/of/base.c | 42 +++++++++++++++++++++++++++++++-----------
 1 file changed, 31 insertions(+), 11 deletions(-)
Signature · 7 lines
2.15.0.276.g89ea799


_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#1 lore raw

PATCH 1/2 of: base: parse all available memory nodes

Currently, barebox only parse one memory node which is either the "/memory" node or the first node with device_type == "memory". However, the use of multiple memory nodes with device_type = "memory" property is allowed by the device tree specification and already correctly parsed by Linux kernel.
In order to fix that, add of_probe_memory function which loop over all available memory nodes. In order to try to keep existing legacy search based on "memory" node name, try to find this node and add it. If the memory node contains the device_type property, then it will only be added once.

---
 drivers/of/base.c | 31 +++++++++++++++++++++++++------
 1 file changed, 25 insertions(+), 6 deletions(-)
drivers/of/base.c +25 -6
index 9ede05227..b1a96ee8f 100644
@@ -2061,9 +2061,32 @@ const struct of_device_id of_default_bus_match_table[] = {
 	}
 };
 
+static void of_probe_memory(void)
+{
+	struct device_node *memory = root_node, *legacy_memory;
+
+	/* Parse node based on name (for legacy dt) */
+	legacy_memory = of_find_node_by_path("/memory");
+	if (legacy_memory)
+		of_add_memory(legacy_memory, false);
+
+	/* Then, parse all available node with "memory" device_type */
+	while (1) {
+		memory = of_find_node_by_type(memory, "memory");
+		if (!memory)
+			break;
+
+		/* Skip potentially already added legacy memory node */
+		if (memory == legacy_memory)
+			continue;
+
+		of_add_memory(memory, false);
+	}
+}
+
 int of_probe(void)
 {
-	struct device_node *memory, *firmware;
+	struct device_node *firmware;
 
 	if(!root_node)
 		return -ENODEV;
@@ -2074,11 +2097,7 @@ int of_probe(void)
 	if (of_model)
 		barebox_set_model(of_model);
 
-	memory = of_find_node_by_path("/memory");
-	if (!memory)
-		memory = of_find_node_by_type(root_node, "memory");
-	if (memory)
-		of_add_memory(memory, false);
+	of_probe_memory();
 
 	firmware = of_find_node_by_path("/firmware");
 	if (firmware)
Signature · 7 lines
2.15.0.276.g89ea799


_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#2 lore raw

Re: [PATCH 1/2] of: base: parse all available memory nodes

Quoted 35 lines · Clement Leger

Currently, barebox only parse one memory node which is either the "/memory" node or the first node with device_type == "memory". However, the use of multiple memory nodes with device_type = "memory" property is allowed by the device tree specification and already correctly parsed by Linux kernel.
In order to fix that, add of_probe_memory function which loop over all available memory nodes. In order to try to keep existing legacy search based on "memory" node name, try to find this node and add it. If the memory node contains the device_type property, then it will only be added once.

---
 drivers/of/base.c | 31 +++++++++++++++++++++++++------
 1 file changed, 25 insertions(+), 6 deletions(-)
drivers/of/base.c +11 -0
index 9ede05227..b1a96ee8f 100644
@@ -2061,9 +2061,32 @@ const struct of_device_id of_default_bus_match_table[] = {
 	}
 };
 
+static void of_probe_memory(void)
+{
+	struct device_node *memory = root_node, *legacy_memory;
+
+	/* Parse node based on name (for legacy dt) */
+	legacy_memory = of_find_node_by_path("/memory");
+	if (legacy_memory)
+		of_add_memory(legacy_memory, false);
+
+	/* Then, parse all available node with "memory" device_type */
+	while (1) {

drivers/of/base.c const struct of_device_id of_default_bus_match_table[] = { · line 2084

Quoted 10 lines
drivers/of/base.c +10 -0
+		memory = of_find_node_by_type(memory, "memory");
+		if (!memory)
+			break;
+
+		/* Skip potentially already added legacy memory node */
+		if (memory == legacy_memory)
+			continue;
+
+		of_add_memory(memory, false);
+	}

AFAIK the device_type = "memory" property was mandatory in the early days as well, there shouldn't be any /memory nodes without this property. Given that, is the add-legacy-node-first still necessary?

Sascha

Signature · 9 lines
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#3 lore raw

Re: [PATCH 1/2] of: base: parse all available memory nodes

Hi Sascha,

Quoted 43 lines · ----- On 23 Mar, 2020, at 11:12, Sascha Hauer [email protected] wrote:
Quoted 42 lines · Clement Leger

Currently, barebox only parse one memory node which is either the "/memory" node or the first node with device_type == "memory". However, the use of multiple memory nodes with device_type = "memory" property is allowed by the device tree specification and already correctly parsed by Linux kernel.
In order to fix that, add of_probe_memory function which loop over all available memory nodes. In order to try to keep existing legacy search based on "memory" node name, try to find this node and add it. If the memory node contains the device_type property, then it will only be added once.

---
 drivers/of/base.c | 31 +++++++++++++++++++++++++------
 1 file changed, 25 insertions(+), 6 deletions(-)
drivers/of/base.c +17 -0
index 9ede05227..b1a96ee8f 100644
@@ -2061,9 +2061,32 @@ const struct of_device_id of_default_bus_match_table[] =
 {
 	}
 };
 
+static void of_probe_memory(void)
+{
+	struct device_node *memory = root_node, *legacy_memory;
+
+	/* Parse node based on name (for legacy dt) */
+	legacy_memory = of_find_node_by_path("/memory");
+	if (legacy_memory)
+		of_add_memory(legacy_memory, false);
+
+	/* Then, parse all available node with "memory" device_type */
+	while (1) {
+		memory = of_find_node_by_type(memory, "memory");
+		if (!memory)
+			break;
+
+		/* Skip potentially already added legacy memory node */
+		if (memory == legacy_memory)
Quoted 8 lines

drivers/of/base.c const struct of_device_id of_default_bus_match_table[] = { · line 2084

Quoted 4 lines
drivers/of/base.c +4 -0
+			continue;
+
+		of_add_memory(memory, false);
+	}

AFAIK the device_type = "memory" property was mandatory in the early days as well, there shouldn't be any /memory nodes without this property. Given that, is the add-legacy-node-first still necessary?

Agreed, I did that after speaking with someone on IRC which stated (rightfully), that I should keep the legacy behavior if possible. However, I agree that since the device_type = "memory" property should have always been there, there is no reason to keep this behavior.

If you are ok with that, I will remove this chunk of code.

Quoted 9 lines

Sascha

--

Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox

#4 lore raw

Re: [PATCH 1/2] of: base: parse all available memory nodes

Hi,

Quoted 6 lines · Clément Leger
Quoted 3 lines

AFAIK the device_type = "memory" property was mandatory in the early days as well, there shouldn't be any /memory nodes without this property. Given that, is the add-legacy-node-first still necessary?

Agreed, I did that after speaking with someone on IRC which stated (rightfully), that I should keep the legacy behavior if possible.

My concern was about HW that passes a device tree to barebox (PowerPC possibly?), where it might have worked before and now won't anymore.

If this is no real concern, it can be removed.

Cheers
Ahmad

Quoted 20 lines

However, I agree that since the device_type = "memory" property should have always been there, there is no reason to keep this behavior.

If you are ok with that, I will remove this chunk of code.

Quoted 9 lines

Sascha

--

Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox

Signature · 9 lines
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#5 lore raw

Re: [PATCH 1/2] of: base: parse all available memory nodes

Quoted 7 lines · Ahmad Fatoum

Hi,

Quoted 4 lines · Clément Leger
Quoted 3 lines

AFAIK the device_type = "memory" property was mandatory in the early days as well, there shouldn't be any /memory nodes without this property. Given that, is the add-legacy-node-first still necessary?

Quoted 7 lines
Quoted 2 lines

Agreed, I did that after speaking with someone on IRC which stated (rightfully), that I should keep the legacy behavior if possible.

My concern was about HW that passes a device tree to barebox (PowerPC possibly?), where it might have worked before and now won't anymore.

If this is no real concern, it can be removed.

There is no PowerPC hardware supported by barebox that passes a device tree to barebox. I don't think this is necessary.

Sascha

Signature · 9 lines
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#6 lore raw

Re: [PATCH 1/2] of: base: parse all available memory nodes

Quoted 11 lines · Sascha Hauer
Quoted 10 lines · Ahmad Fatoum

Hi,

Quoted 6 lines · Clément Leger
Quoted 3 lines

AFAIK the device_type = "memory" property was mandatory in the early days as well, there shouldn't be any /memory nodes without this property. Given that, is the add-legacy-node-first still necessary?

Agreed, I did that after speaking with someone on IRC which stated (rightfully), that I should keep the legacy behavior if possible.

Quoted 7 lines
Quoted 4 lines

My concern was about HW that passes a device tree to barebox (PowerPC possibly?), where it might have worked before and now won't anymore.

If this is no real concern, it can be removed.

There is no PowerPC hardware supported by barebox that passes a device tree to barebox. I don't think this is necessary.

Turns out we had quite a few device trees that lacked a device_type = "memory"
property: https://lists.infradead.org/pipermail/barebox/2020-June/042269.html

The property used to be there, but when upstream removed skeleton.dtsi and either there was no upstream memory node to extend or it was renamed, we ended up with an invalid memory node.

Quoted 3 lines

Sascha

Signature · 9 lines
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#7 lore raw

PATCH 2/2 of: base: allow of_add_memory to be called multiple times

In reply to Clement Leger

Currently, of_add_memory can't be called multiple times because it will always create memory banks by restarting at value 0. This means that, when adding a second memory bank by calling again of_add_memory, it will be named ram0 and overwrite the previous one. Fix that by using a static variable to store the current memory bank number in order to be preserved from one call to another.

---
 drivers/of/base.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
drivers/of/base.c +6 -5
index b1a96ee8f..12c58bfb6 100644
@@ -2019,6 +2019,8 @@ int of_set_property(struct device_node *np, const char *name, const void *val, i
 	return 0;
 }
 
+static int mem_bank_num;
+
 int of_add_memory(struct device_node *node, bool dump)
 {
 	const char *device_type;
@@ -2030,14 +2032,13 @@ int of_add_memory(struct device_node *node, bool dump)
 		return -ENXIO;
 
 	while (!of_address_to_resource(node, n, &res)) {
-		if (!resource_size(&res)) {
-			n++;
+		n++;
+		if (!resource_size(&res))
 			continue;
-		}
 
-		of_add_memory_bank(node, dump, n,
+		of_add_memory_bank(node, dump, mem_bank_num,
 				res.start, resource_size(&res));
-		n++;
+		mem_bank_num++;
 	}
 
 	return 0;
Signature · 7 lines
2.15.0.276.g89ea799


_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#8 lore raw

v2 cover Allow parsing more than one memory node

Currently, barebox can only parse one memory node. This means that all other memory nodes (with device_type == "memory") will be ignored while parsing memory.
This patchset add a function of_probe_memory which will call of_add_memory for each found memory node. Additionally, of_add_memory has been modified to allow being called multiple times and correctly add all memory banks.

Changes v1 -> v2
- Reorder patches
- Remove legacy "memory" node handling

Clement Leger (2):
of: base: allow of_add_memory to be called multiple times
of: base: parse all available memory nodes

 drivers/of/base.c | 33 ++++++++++++++++++++++-----------
 1 file changed, 22 insertions(+), 11 deletions(-)
Signature · 7 lines
2.17.1


_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#9 lore raw

PATCH v2 1/2 of: base: allow of_add_memory to be called multiple times

Currently, of_add_memory can't be called multiple times because it will always create memory banks by restarting at value 0. This means that, when adding a second memory bank by calling again of_add_memory, it will be named ram0 and overwrite the previous one. Fix that by using a static variable to store the current memory bank number in order to be preserved from one call to another.

---
 drivers/of/base.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
drivers/of/base.c +6 -5
index 9ede05227..2ea8d7516 100644
@@ -2019,6 +2019,8 @@ int of_set_property(struct device_node *np, const char *name, const void *val, i
 	return 0;
 }
 
+static int mem_bank_num;
+
 int of_add_memory(struct device_node *node, bool dump)
 {
 	const char *device_type;
@@ -2030,14 +2032,13 @@ int of_add_memory(struct device_node *node, bool dump)
 		return -ENXIO;
 
 	while (!of_address_to_resource(node, n, &res)) {
-		if (!resource_size(&res)) {
-			n++;
+		n++;
+		if (!resource_size(&res))
 			continue;
-		}
 
-		of_add_memory_bank(node, dump, n,
+		of_add_memory_bank(node, dump, mem_bank_num,
 				res.start, resource_size(&res));
-		n++;
+		mem_bank_num++;
 	}
 
 	return 0;
Signature · 7 lines
2.17.1


_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#10 lore raw

PATCH v2 2/2 of: base: parse all available memory nodes

In reply to Clement Leger

Currently, barebox only parse one memory node which is either the "/memory" node or the first node with device_type == "memory". However, the use of multiple memory nodes with device_type = "memory" property is allowed by the device tree specification and already correctly parsed by Linux kernel.
In order to fix that, add of_probe_memory function which loop over all available memory nodes matching device_type == "memory".

---
 drivers/of/base.c | 22 ++++++++++++++++------
 1 file changed, 16 insertions(+), 6 deletions(-)
drivers/of/base.c +16 -6
index 2ea8d7516..6d13b66a0 100644
@@ -2062,9 +2062,23 @@ const struct of_device_id of_default_bus_match_table[] = {
 	}
 };
 
+static void of_probe_memory(void)
+{
+	struct device_node *memory = root_node;
+
+	/* Parse all available node with "memory" device_type */
+	while (1) {
+		memory = of_find_node_by_type(memory, "memory");
+		if (!memory)
+			break;
+
+		of_add_memory(memory, false);
+	}
+}
+
 int of_probe(void)
 {
-	struct device_node *memory, *firmware;
+	struct device_node *firmware;
 
 	if(!root_node)
 		return -ENODEV;
@@ -2075,11 +2089,7 @@ int of_probe(void)
 	if (of_model)
 		barebox_set_model(of_model);
 
-	memory = of_find_node_by_path("/memory");
-	if (!memory)
-		memory = of_find_node_by_type(root_node, "memory");
-	if (memory)
-		of_add_memory(memory, false);
+	of_probe_memory();
 
 	firmware = of_find_node_by_path("/firmware");
 	if (firmware)
Signature · 7 lines
2.17.1


_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
Applied
#11 lore raw
In reply to Clement Leger
Quoted 8 lines · Clement Leger

Currently, barebox can only parse one memory node. This means that all other memory nodes (with device_type == "memory") will be ignored while parsing memory.
This patchset add a function of_probe_memory which will call of_add_memory for each found memory node. Additionally, of_add_memory has been modified to allow being called multiple times and correctly add all memory banks.

Quoted 7 lines

Changes v1 -> v2
- Reorder patches
- Remove legacy "memory" node handling

Clement Leger (2):
of: base: allow of_add_memory to be called multiple times
of: base: parse all available memory nodes

Applied, thanks

Sascha

Signature · 9 lines
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#12 lore raw
In reply to Clement Leger

Hi Clement.

Quoted 11 lines · Clement Leger

Currently, barebox can only parse one memory node. This means that all other memory nodes (with device_type == "memory") will be ignored while parsing memory.
This patchset add a function of_probe_memory which will call of_add_memory for each found memory node. Additionally, of_add_memory has been modified to allow being called multiple times and correctly add all memory banks.

Clement Leger (2):
of: base: parse all available memory nodes
of: base: allow of_add_memory to be called multiple times

With this order of the patches, after applying patch 1 the code will then use the last and not the first memory node. And only after the second patch it will work.

I think to improve bisectability the order of the two patches should be reversed.

Sam

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox

#13 lore raw

Hi Sam,

Quoted 15 lines · ----- On 17 Mar, 2020, at 08:35, Sam Ravnborg [email protected] wrote:

Hi Clement.

Quoted 11 lines · Clement Leger

Currently, barebox can only parse one memory node. This means that all other memory nodes (with device_type == "memory") will be ignored while parsing memory.
This patchset add a function of_probe_memory which will call of_add_memory for each found memory node. Additionally, of_add_memory has been modified to allow being called multiple times and correctly add all memory banks.

Clement Leger (2):
of: base: parse all available memory nodes
of: base: allow of_add_memory to be called multiple times

Quoted 6 lines

With this order of the patches, after applying patch 1 the code will then use the last and not the first memory node. And only after the second patch it will work.

I think to improve bisectability the order of the two patches should be reversed.

Oh yes indeed ! I will send a V2 after others have commented it.

Quoted 2 lines

Sam

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox