Skip to contentSkip to conversation

pstore: Support already existing reserved-memory node

PATCH In review Detlev Casanova 3 participants 8 messages Feb 7, 2022 – Apr 19, 2022

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 pstore: Support already existing reserved-memory node

The pstore command tries to create a reserved-memory node but fails if it is already present with:

Add 'reserved-memory' node failed: FDT_ERR_EXISTS

This patch creates the node only if it does not exist and adapts the reg values sizes depending on already present #address-cells and #size-cells values.

---

Changes in v2:
- Move declarations at beginning of function
- Use if instead of switch
- Reword Subject

 cmd/pstore.c | 39 ++++++++++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 5 deletions(-)
cmd/pstore.c +34 -5
index 9fac8c7218..cd6f6feb2f 100644
@@ -11,6 +11,7 @@
 #include <mapmem.h>
 #include <memalign.h>
 #include <part.h>
+#include <fdt_support.h>
 
 struct persistent_ram_buffer {
 	u32    sig;
@@ -485,6 +486,8 @@ void fdt_fixup_pstore(void *blob)
 {
 	char node[32];
 	int  nodeoffset;	/* node offset from libfdt */
+	u32 addr_cells;
+	u32 size_cells;
 
 	nodeoffset = fdt_path_offset(blob, "/");
 	if (nodeoffset < 0) {
@@ -493,14 +496,18 @@ void fdt_fixup_pstore(void *blob)
 		return;
 	}
 
-	nodeoffset = fdt_add_subnode(blob, nodeoffset, "reserved-memory");
+	nodeoffset = fdt_find_or_add_subnode(blob, nodeoffset, "reserved-memory");
 	if (nodeoffset < 0) {
 		log_err("Add 'reserved-memory' node failed: %s\n",
 				fdt_strerror(nodeoffset));
 		return;
 	}
-	fdt_setprop_u32(blob, nodeoffset, "#address-cells", 2);
-	fdt_setprop_u32(blob, nodeoffset, "#size-cells", 2);
+
+	addr_cells = fdt_getprop_u32_default_node(blob, nodeoffset, 0, "#address-cells", 2);
+	size_cells = fdt_getprop_u32_default_node(blob, nodeoffset, 0, "#size-cells", 2);
+	fdt_setprop_u32(blob, nodeoffset, "#address-cells", addr_cells);
+	fdt_setprop_u32(blob, nodeoffset, "#size-cells", size_cells);
+
 	fdt_setprop_empty(blob, nodeoffset, "ranges");
 
 	sprintf(node, "ramoops@%llx", (unsigned long long)pstore_addr);
@@ -509,14 +516,36 @@ void fdt_fixup_pstore(void *blob)
 		log_err("Add '%s' node failed: %s\n", node, fdt_strerror(nodeoffset));
 		return;
 	}
+
 	fdt_setprop_string(blob, nodeoffset, "compatible", "ramoops");
-	fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
-	fdt_appendprop_u64(blob, nodeoffset, "reg", pstore_length);
+
+	if (addr_cells == 1) {
+		fdt_setprop_u32(blob, nodeoffset, "reg", pstore_addr);
+	} else if (addr_cells == 2) {
+		fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
+	} else {
+		log_err("Unsupported #address-cells: %u\n", addr_cells);
+		goto clean_ramoops;
+	}
+
+	if (size_cells == 1) {
+		// Let's consider that the pstore_length fits in a 32 bits value
+		fdt_appendprop_u32(blob, nodeoffset, "reg", pstore_length);
+	} else if (size_cells == 2) {
+		fdt_appendprop_u64(blob, nodeoffset, "reg", pstore_length);
+	} else {
+		log_err("Unsupported #size-cells: %u\n", addr_cells);
+		goto clean_ramoops;
+	}
+
 	fdt_setprop_u32(blob, nodeoffset, "record-size", pstore_record_size);
 	fdt_setprop_u32(blob, nodeoffset, "console-size", pstore_console_size);
 	fdt_setprop_u32(blob, nodeoffset, "ftrace-size", pstore_ftrace_size);
 	fdt_setprop_u32(blob, nodeoffset, "pmsg-size", pstore_pmsg_size);
 	fdt_setprop_u32(blob, nodeoffset, "ecc-size", pstore_ecc_size);
+
+clean_ramoops:
+	fdt_del_node_and_alias(blob, node);
 }
 
 U_BOOT_CMD(pstore, 10, 0, do_pstore,
Signature · 1 line
2.35.1
#1 lore raw
Quoted 10 lines · Detlev Casanova

The pstore command tries to create a reserved-memory node but fails if it is already present with:

Add 'reserved-memory' node failed: FDT_ERR_EXISTS

This patch creates the node only if it does not exist and adapts the reg values sizes depending on already present #address-cells and #size-cells values.

Applied to u-boot/master, thanks!

Signature · 1 line
Tom
#2 lore raw
In reply to Detlev Casanova

Hi Detlev,

I've recently tried to update U-Boot from 2022.01 to 2022.04 and noticed a regression introduced by the commit below.

While the unwanted error message ("Add 'reserved-memory' node failed:
FDT_ERR_EXISTS") no longer appears, I now see a lengthy kernel stack trace in Linux instead:
...

[    0.008295] sysfs: cannot create duplicate filename '/devices/platform/42ff0000.ramoops'
[    0.008303] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G S                5.15.33 #0
[    0.008312] Hardware name: Bananapi BPI-R64 (DT)
[    0.008318] Call trace:
[    0.008322]  dump_backtrace+0x0/0x15c
[    0.008337]  show_stack+0x14/0x30
[    0.008345]  dump_stack_lvl+0x64/0x7c
[    0.008355]  dump_stack+0x14/0x2c
[    0.008362]  sysfs_warn_dup+0x5c/0x74
[    0.008373]  sysfs_create_dir_ns+0xb0/0xc0
[    0.008381]  kobject_add_internal+0xbc/0x330
[    0.008392]  kobject_add+0x80/0xe0
[    0.008400]  device_add+0xd4/0x82c
[    0.008410]  of_device_add+0x4c/0x5c
[    0.008420]  of_platform_device_create_pdata+0xb8/0xf0
[    0.008428]  of_platform_default_populate_init+0x58/0xcc
[    0.008437]  do_one_initcall+0x4c/0x1b0
[    0.008444]  kernel_init_freeable+0x230/0x294
[    0.008456]  kernel_init+0x20/0x120
[    0.008463]  ret_from_fork+0x10/0x20
[    0.008471] kobject_add_internal failed for 42ff0000.ramoops with -EEXIST, don't try to register things with the same name in the same directory.

...

I assume that fdt_find_or_add_subnode() doesn't behave as expected and apparently adds a duplicate node having the same name.
The pstore/ramoops reserved-memory is already defined in the fdt contained in the FIT image (for compatibility with older U-Boot which didn't care about pstore), so it should be not added again.

Quoted 105 lines · Detlev Casanova

The pstore command tries to create a reserved-memory node but fails if it is already present with:

Add 'reserved-memory' node failed: FDT_ERR_EXISTS

This patch creates the node only if it does not exist and adapts the reg values sizes depending on already present #address-cells and #size-cells values.

---

Changes in v2:
- Move declarations at beginning of function
- Use if instead of switch
- Reword Subject

 cmd/pstore.c | 39 ++++++++++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 5 deletions(-)
cmd/pstore.c +34 -5
index 9fac8c7218..cd6f6feb2f 100644
@@ -11,6 +11,7 @@
 #include <mapmem.h>
 #include <memalign.h>
 #include <part.h>
+#include <fdt_support.h>
 
 struct persistent_ram_buffer {
 	u32    sig;
@@ -485,6 +486,8 @@ void fdt_fixup_pstore(void *blob)
 {
 	char node[32];
 	int  nodeoffset;	/* node offset from libfdt */
+	u32 addr_cells;
+	u32 size_cells;
 
 	nodeoffset = fdt_path_offset(blob, "/");
 	if (nodeoffset < 0) {
@@ -493,14 +496,18 @@ void fdt_fixup_pstore(void *blob)
 		return;
 	}
 
-	nodeoffset = fdt_add_subnode(blob, nodeoffset, "reserved-memory");
+	nodeoffset = fdt_find_or_add_subnode(blob, nodeoffset, "reserved-memory");
 	if (nodeoffset < 0) {
 		log_err("Add 'reserved-memory' node failed: %s\n",
 				fdt_strerror(nodeoffset));
 		return;
 	}
-	fdt_setprop_u32(blob, nodeoffset, "#address-cells", 2);
-	fdt_setprop_u32(blob, nodeoffset, "#size-cells", 2);
+
+	addr_cells = fdt_getprop_u32_default_node(blob, nodeoffset, 0, "#address-cells", 2);
+	size_cells = fdt_getprop_u32_default_node(blob, nodeoffset, 0, "#size-cells", 2);
+	fdt_setprop_u32(blob, nodeoffset, "#address-cells", addr_cells);
+	fdt_setprop_u32(blob, nodeoffset, "#size-cells", size_cells);
+
 	fdt_setprop_empty(blob, nodeoffset, "ranges");
 
 	sprintf(node, "ramoops@%llx", (unsigned long long)pstore_addr);
@@ -509,14 +516,36 @@ void fdt_fixup_pstore(void *blob)
 		log_err("Add '%s' node failed: %s\n", node, fdt_strerror(nodeoffset));
 		return;
 	}
+
 	fdt_setprop_string(blob, nodeoffset, "compatible", "ramoops");
-	fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
-	fdt_appendprop_u64(blob, nodeoffset, "reg", pstore_length);
+
+	if (addr_cells == 1) {
+		fdt_setprop_u32(blob, nodeoffset, "reg", pstore_addr);
+	} else if (addr_cells == 2) {
+		fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
+	} else {
+		log_err("Unsupported #address-cells: %u\n", addr_cells);
+		goto clean_ramoops;
+	}
+
+	if (size_cells == 1) {
+		// Let's consider that the pstore_length fits in a 32 bits value
+		fdt_appendprop_u32(blob, nodeoffset, "reg", pstore_length);
+	} else if (size_cells == 2) {
+		fdt_appendprop_u64(blob, nodeoffset, "reg", pstore_length);
+	} else {
+		log_err("Unsupported #size-cells: %u\n", addr_cells);
+		goto clean_ramoops;
+	}
+
 	fdt_setprop_u32(blob, nodeoffset, "record-size", pstore_record_size);
 	fdt_setprop_u32(blob, nodeoffset, "console-size", pstore_console_size);
 	fdt_setprop_u32(blob, nodeoffset, "ftrace-size", pstore_ftrace_size);
 	fdt_setprop_u32(blob, nodeoffset, "pmsg-size", pstore_pmsg_size);
 	fdt_setprop_u32(blob, nodeoffset, "ecc-size", pstore_ecc_size);
+
+clean_ramoops:
+	fdt_del_node_and_alias(blob, node);
 }
 
 U_BOOT_CMD(pstore, 10, 0, do_pstore,
Signature · 1 line
2.35.1
#3 lore raw

Hi Daniel,

The patch only checks the existence of the reserved-memory node.

I guess that before, adding the reserved-memory node failed and so nothing else was happening. Now, the node is found and then, it adds the ramoops subnode to it, making it double because it is already there in your case.

But that should not happen, because the fdt_add_subnode() function should fail if the node is already present (like it was failing for reserved-memory).

I'm going to look into it. Can you send me the device tree taht you are using (or at least the reserved-memory part) and u-boot config ?

Detlev.

Quoted 119 lines · On Monday, April 11, 2022 7:00:20 P.M. EDT Daniel Golle wrote:

Hi Detlev,

I've recently tried to update U-Boot from 2022.01 to 2022.04 and noticed a regression introduced by the commit below.

While the unwanted error message ("Add 'reserved-memory' node failed:
FDT_ERR_EXISTS") no longer appears, I now see a lengthy kernel stack trace in Linux instead:
...

[    0.008295] sysfs: cannot create duplicate filename
'/devices/platform/42ff0000.ramoops' [    0.008303] CPU: 0 PID: 1 Comm:
swapper/0 Tainted: G S                5.15.33 #0 [    0.008312] Hardware
name: Bananapi BPI-R64 (DT)
[    0.008318] Call trace:
[    0.008322]  dump_backtrace+0x0/0x15c
[    0.008337]  show_stack+0x14/0x30
[    0.008345]  dump_stack_lvl+0x64/0x7c
[    0.008355]  dump_stack+0x14/0x2c
[    0.008362]  sysfs_warn_dup+0x5c/0x74
[    0.008373]  sysfs_create_dir_ns+0xb0/0xc0
[    0.008381]  kobject_add_internal+0xbc/0x330
[    0.008392]  kobject_add+0x80/0xe0
[    0.008400]  device_add+0xd4/0x82c
[    0.008410]  of_device_add+0x4c/0x5c
[    0.008420]  of_platform_device_create_pdata+0xb8/0xf0
[    0.008428]  of_platform_default_populate_init+0x58/0xcc
[    0.008437]  do_one_initcall+0x4c/0x1b0
[    0.008444]  kernel_init_freeable+0x230/0x294
[    0.008456]  kernel_init+0x20/0x120
[    0.008463]  ret_from_fork+0x10/0x20
[    0.008471] kobject_add_internal failed for 42ff0000.ramoops with

-EEXIST, don't try to register things with the same name in the same directory. ...

I assume that fdt_find_or_add_subnode() doesn't behave as expected and apparently adds a duplicate node having the same name.
The pstore/ramoops reserved-memory is already defined in the fdt contained in the FIT image (for compatibility with older U-Boot which didn't care about pstore), so it should be not added again.

Quoted 78 lines · Detlev Casanova

The pstore command tries to create a reserved-memory node but fails if

it is already present with:
Add 'reserved-memory' node failed: FDT_ERR_EXISTS

This patch creates the node only if it does not exist and adapts the reg values sizes depending on already present #address-cells and #size-cells values.

---

Changes in v2:
- Move declarations at beginning of function
- Use if instead of switch
- Reword Subject

 cmd/pstore.c | 39 ++++++++++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 5 deletions(-)
cmd/pstore.c +1 -0
index 9fac8c7218..cd6f6feb2f 100644
@@ -11,6 +11,7 @@
 
 #include <mapmem.h>
 #include <memalign.h>
 #include <part.h>
 
+#include <fdt_support.h>
 

struct persistent_ram_buffer {

cmd/pstore.c +2 -0
 	u32    sig;
@@ -485,6 +486,8 @@ void fdt_fixup_pstore(void *blob)
 
 {
 
 	char node[32];
 	int  nodeoffset;	/* node offset from libfdt */
 
+	u32 addr_cells;
+	u32 size_cells;
cmd/pstore.c +1 -1
 	nodeoffset = fdt_path_offset(blob, "/");
 	if (nodeoffset < 0) {
@@ -493,14 +496,18 @@ void fdt_fixup_pstore(void *blob)
 
 		return;
 
 	}
 
-	nodeoffset = fdt_add_subnode(blob, nodeoffset, "reserved-memory");
+	nodeoffset = fdt_find_or_add_subnode(blob, nodeoffset,

"reserved-memory");>

cmd/pstore.c +2 -2
 	if (nodeoffset < 0) {
 
 		log_err("Add 'reserved-memory' node failed: %s\n",
 
 				fdt_strerror(nodeoffset));
 
 		return;
 
 	}
 
-	fdt_setprop_u32(blob, nodeoffset, "#address-cells", 2);
-	fdt_setprop_u32(blob, nodeoffset, "#size-cells", 2);
+
+	addr_cells = fdt_getprop_u32_default_node(blob, nodeoffset, 0,

"#address-cells", 2); + size_cells = fdt_getprop_u32_default_node(blob,
nodeoffset, 0, "#size-cells", 2); + fdt_setprop_u32(blob, nodeoffset,
"#address-cells", addr_cells); + fdt_setprop_u32(blob, nodeoffset,
"#size-cells", size_cells);
+

Quoted 7 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 517

Quoted 7 lines
cmd/pstore.c +0 -0
 	fdt_setprop_empty(blob, nodeoffset, "ranges");
 
 	sprintf(node, "ramoops@%llx", (unsigned long long)pstore_addr);
@@ -509,14 +516,36 @@ void fdt_fixup_pstore(void *blob)
 
 		log_err("Add '%s' node failed: %s\n", node, 

fdt_strerror(nodeoffset));

Quoted 8 lines
Quoted 8 lines
cmd/pstore.c +1 -0
 		return;
 
 	}
 
+
 
 	fdt_setprop_string(blob, nodeoffset, "compatible", "ramoops");
 
Quoted 9 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 9 lines
cmd/pstore.c +7 -2
-	fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
-	fdt_appendprop_u64(blob, nodeoffset, "reg", pstore_length);
+
+	if (addr_cells == 1) {
+		fdt_setprop_u32(blob, nodeoffset, "reg", pstore_addr);
+	} else if (addr_cells == 2) {
+		fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
+	} else {
+		log_err("Unsupported #address-cells: %u\n", 

addr_cells);

Quoted 5 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 5 lines
cmd/pstore.c +5 -0
+		goto clean_ramoops;
+	}
+
+	if (size_cells == 1) {
+		// Let's consider that the pstore_length fits in a 32 

bits value

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 538

Quoted 1 line
cmd/pstore.c +1 -0
+		fdt_appendprop_u32(blob, nodeoffset, "reg", 

pstore_length);

Quoted 2 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 540

Quoted 2 lines
cmd/pstore.c +2 -0
+	} else if (size_cells == 2) {
+		fdt_appendprop_u64(blob, nodeoffset, "reg", 

pstore_length);

Quoted 7 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 7 lines
cmd/pstore.c +5 -0
+	} else {
+		log_err("Unsupported #size-cells: %u\n", addr_cells);
+		goto clean_ramoops;
+	}
+
 
 	fdt_setprop_u32(blob, nodeoffset, "record-size", 

pstore_record_size);

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 548

Quoted 1 line
cmd/pstore.c +0 -0
 	fdt_setprop_u32(blob, nodeoffset, "console-size", 

pstore_console_size);

Quoted 1 line
Quoted 1 line

fdt_setprop_u32(blob, nodeoffset, "ftrace-size",

pstore_ftrace_size);

Quoted 10 lines
Quoted 10 lines
cmd/pstore.c +3 -0
 	fdt_setprop_u32(blob, nodeoffset, "pmsg-size", pstore_pmsg_size);
 	fdt_setprop_u32(blob, nodeoffset, "ecc-size", pstore_ecc_size);
 
+
+clean_ramoops:
+	fdt_del_node_and_alias(blob, node);
 
 }
 
 U_BOOT_CMD(pstore, 10, 0, do_pstore,
#4 lore raw

Hi Detlev,

Quoted 8 lines · Detlev Casanova

Hi Daniel,

The patch only checks the existence of the reserved-memory node.

I guess that before, adding the reserved-memory node failed and so nothing else was happening. Now, the node is found and then, it adds the ramoops subnode to it, making it double because it is already there in your case.

Quoted 5 lines

But that should not happen, because the fdt_add_subnode() function should fail if the node is already present (like it was failing for reserved-memory).

I'm going to look into it. Can you send me the device tree taht you are using (or at least the reserved-memory part) and u-boot config ?

Thank you for your reply and for taking a look at what's going on.

reserved-memory bits in dts:
https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=package/boot/uboot-mediatek/patches/050-mt7622-enable-pstore.patch

u-boot config:
https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=package/boot/uboot-mediatek/patches/404-add-bananapi_bpi-r64_defconfigs.patch

Cheers

Daniel

Quoted 186 lines

Detlev.

Quoted 119 lines · On Monday, April 11, 2022 7:00:20 P.M. EDT Daniel Golle wrote:

Hi Detlev,

I've recently tried to update U-Boot from 2022.01 to 2022.04 and noticed a regression introduced by the commit below.

While the unwanted error message ("Add 'reserved-memory' node failed:
FDT_ERR_EXISTS") no longer appears, I now see a lengthy kernel stack trace in Linux instead:
...

[    0.008295] sysfs: cannot create duplicate filename
'/devices/platform/42ff0000.ramoops' [    0.008303] CPU: 0 PID: 1 Comm:
swapper/0 Tainted: G S                5.15.33 #0 [    0.008312] Hardware
name: Bananapi BPI-R64 (DT)
[    0.008318] Call trace:
[    0.008322]  dump_backtrace+0x0/0x15c
[    0.008337]  show_stack+0x14/0x30
[    0.008345]  dump_stack_lvl+0x64/0x7c
[    0.008355]  dump_stack+0x14/0x2c
[    0.008362]  sysfs_warn_dup+0x5c/0x74
[    0.008373]  sysfs_create_dir_ns+0xb0/0xc0
[    0.008381]  kobject_add_internal+0xbc/0x330
[    0.008392]  kobject_add+0x80/0xe0
[    0.008400]  device_add+0xd4/0x82c
[    0.008410]  of_device_add+0x4c/0x5c
[    0.008420]  of_platform_device_create_pdata+0xb8/0xf0
[    0.008428]  of_platform_default_populate_init+0x58/0xcc
[    0.008437]  do_one_initcall+0x4c/0x1b0
[    0.008444]  kernel_init_freeable+0x230/0x294
[    0.008456]  kernel_init+0x20/0x120
[    0.008463]  ret_from_fork+0x10/0x20
[    0.008471] kobject_add_internal failed for 42ff0000.ramoops with

-EEXIST, don't try to register things with the same name in the same directory. ...

I assume that fdt_find_or_add_subnode() doesn't behave as expected and apparently adds a duplicate node having the same name.
The pstore/ramoops reserved-memory is already defined in the fdt contained in the FIT image (for compatibility with older U-Boot which didn't care about pstore), so it should be not added again.

Quoted 78 lines · Detlev Casanova

The pstore command tries to create a reserved-memory node but fails if

it is already present with:
Add 'reserved-memory' node failed: FDT_ERR_EXISTS

This patch creates the node only if it does not exist and adapts the reg values sizes depending on already present #address-cells and #size-cells values.

---

Changes in v2:
- Move declarations at beginning of function
- Use if instead of switch
- Reword Subject

 cmd/pstore.c | 39 ++++++++++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 5 deletions(-)
cmd/pstore.c +1 -0
index 9fac8c7218..cd6f6feb2f 100644
@@ -11,6 +11,7 @@
 
 #include <mapmem.h>
 #include <memalign.h>
 #include <part.h>
 
+#include <fdt_support.h>
 

struct persistent_ram_buffer {

cmd/pstore.c +2 -0
 	u32    sig;
@@ -485,6 +486,8 @@ void fdt_fixup_pstore(void *blob)
 
 {
 
 	char node[32];
 	int  nodeoffset;	/* node offset from libfdt */
 
+	u32 addr_cells;
+	u32 size_cells;
cmd/pstore.c +1 -1
 	nodeoffset = fdt_path_offset(blob, "/");
 	if (nodeoffset < 0) {
@@ -493,14 +496,18 @@ void fdt_fixup_pstore(void *blob)
 
 		return;
 
 	}
 
-	nodeoffset = fdt_add_subnode(blob, nodeoffset, "reserved-memory");
+	nodeoffset = fdt_find_or_add_subnode(blob, nodeoffset,

"reserved-memory");>

cmd/pstore.c +2 -2
 	if (nodeoffset < 0) {
 
 		log_err("Add 'reserved-memory' node failed: %s\n",
 
 				fdt_strerror(nodeoffset));
 
 		return;
 
 	}
 
-	fdt_setprop_u32(blob, nodeoffset, "#address-cells", 2);
-	fdt_setprop_u32(blob, nodeoffset, "#size-cells", 2);
+
+	addr_cells = fdt_getprop_u32_default_node(blob, nodeoffset, 0,

"#address-cells", 2); + size_cells = fdt_getprop_u32_default_node(blob,
nodeoffset, 0, "#size-cells", 2); + fdt_setprop_u32(blob, nodeoffset,
"#address-cells", addr_cells); + fdt_setprop_u32(blob, nodeoffset,
"#size-cells", size_cells);
+

Quoted 7 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 517

Quoted 7 lines
cmd/pstore.c +0 -0
 	fdt_setprop_empty(blob, nodeoffset, "ranges");
 
 	sprintf(node, "ramoops@%llx", (unsigned long long)pstore_addr);
@@ -509,14 +516,36 @@ void fdt_fixup_pstore(void *blob)
 
 		log_err("Add '%s' node failed: %s\n", node, 

fdt_strerror(nodeoffset));

Quoted 8 lines
Quoted 8 lines
cmd/pstore.c +1 -0
 		return;
 
 	}
 
+
 
 	fdt_setprop_string(blob, nodeoffset, "compatible", "ramoops");
 
Quoted 9 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 9 lines
cmd/pstore.c +7 -2
-	fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
-	fdt_appendprop_u64(blob, nodeoffset, "reg", pstore_length);
+
+	if (addr_cells == 1) {
+		fdt_setprop_u32(blob, nodeoffset, "reg", pstore_addr);
+	} else if (addr_cells == 2) {
+		fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
+	} else {
+		log_err("Unsupported #address-cells: %u\n", 

addr_cells);

Quoted 5 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 5 lines
cmd/pstore.c +5 -0
+		goto clean_ramoops;
+	}
+
+	if (size_cells == 1) {
+		// Let's consider that the pstore_length fits in a 32 

bits value

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 538

Quoted 1 line
cmd/pstore.c +1 -0
+		fdt_appendprop_u32(blob, nodeoffset, "reg", 

pstore_length);

Quoted 2 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 540

Quoted 2 lines
cmd/pstore.c +2 -0
+	} else if (size_cells == 2) {
+		fdt_appendprop_u64(blob, nodeoffset, "reg", 

pstore_length);

Quoted 7 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 7 lines
cmd/pstore.c +5 -0
+	} else {
+		log_err("Unsupported #size-cells: %u\n", addr_cells);
+		goto clean_ramoops;
+	}
+
 
 	fdt_setprop_u32(blob, nodeoffset, "record-size", 

pstore_record_size);

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 548

Quoted 1 line
cmd/pstore.c +0 -0
 	fdt_setprop_u32(blob, nodeoffset, "console-size", 

pstore_console_size);

Quoted 1 line
Quoted 1 line

fdt_setprop_u32(blob, nodeoffset, "ftrace-size",

pstore_ftrace_size);

Quoted 10 lines
Quoted 10 lines
cmd/pstore.c +3 -0
 	fdt_setprop_u32(blob, nodeoffset, "pmsg-size", pstore_pmsg_size);
 	fdt_setprop_u32(blob, nodeoffset, "ecc-size", pstore_ecc_size);
 
+
+clean_ramoops:
+	fdt_del_node_and_alias(blob, node);
 
 }
 
 U_BOOT_CMD(pstore, 10, 0, do_pstore,
#5 lore raw

Hey Daniel,

I see that the node is called ramoops@0x42ff0000. We usually don't set the 0x in device tree nodes, and that's why the fdt_add_subnode function doesn't see it and add ramoops@42ff0000 (it just compares the text). For Linux, both nodes are the same and it complains.

I think the best course of action here is to remove that '0x' in your dts, so that u-boot sees it as the pstore node.

Regards,

Detlev.

Quoted 235 lines · On Monday, April 18, 2022 7:50:36 P.M. EDT Daniel Golle wrote:

Hi Detlev,

Quoted 8 lines · Detlev Casanova

Hi Daniel,

The patch only checks the existence of the reserved-memory node.

I guess that before, adding the reserved-memory node failed and so nothing else was happening. Now, the node is found and then, it adds the ramoops subnode to it, making it double because it is already there in your case.

Quoted 6 lines

But that should not happen, because the fdt_add_subnode() function should fail if the node is already present (like it was failing for reserved-memory).

I'm going to look into it. Can you send me the device tree taht you are using (or at least the reserved-memory part) and u-boot config ?

Thank you for your reply and for taking a look at what's going on.

reserved-memory bits in dts:
https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=package/boot/uboot-m
ediatek/patches/050-mt7622-enable-pstore.patch

u-boot config:
https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=package/boot/uboot-m
ediatek/patches/404-add-bananapi_bpi-r64_defconfigs.patch

Cheers

Daniel

Quoted 201 lines

Detlev.

Quoted 123 lines · On Monday, April 11, 2022 7:00:20 P.M. EDT Daniel Golle wrote:

Hi Detlev,

I've recently tried to update U-Boot from 2022.01 to 2022.04 and noticed a regression introduced by the commit below.

While the unwanted error message ("Add 'reserved-memory' node failed:
FDT_ERR_EXISTS") no longer appears, I now see a lengthy kernel stack trace in Linux instead:
...

[    0.008295] sysfs: cannot create duplicate filename
'/devices/platform/42ff0000.ramoops' [    0.008303] CPU: 0 PID: 1 Comm:
swapper/0 Tainted: G S                5.15.33 #0 [    0.008312] Hardware
name: Bananapi BPI-R64 (DT)
[    0.008318] Call trace:
[    0.008322]  dump_backtrace+0x0/0x15c
[    0.008337]  show_stack+0x14/0x30
[    0.008345]  dump_stack_lvl+0x64/0x7c
[    0.008355]  dump_stack+0x14/0x2c
[    0.008362]  sysfs_warn_dup+0x5c/0x74
[    0.008373]  sysfs_create_dir_ns+0xb0/0xc0
[    0.008381]  kobject_add_internal+0xbc/0x330
[    0.008392]  kobject_add+0x80/0xe0
[    0.008400]  device_add+0xd4/0x82c
[    0.008410]  of_device_add+0x4c/0x5c
[    0.008420]  of_platform_device_create_pdata+0xb8/0xf0
[    0.008428]  of_platform_default_populate_init+0x58/0xcc
[    0.008437]  do_one_initcall+0x4c/0x1b0
[    0.008444]  kernel_init_freeable+0x230/0x294
[    0.008456]  kernel_init+0x20/0x120
[    0.008463]  ret_from_fork+0x10/0x20
[    0.008471] kobject_add_internal failed for 42ff0000.ramoops with

-EEXIST, don't try to register things with the same name in the same directory. ...

I assume that fdt_find_or_add_subnode() doesn't behave as expected and apparently adds a duplicate node having the same name.
The pstore/ramoops reserved-memory is already defined in the fdt contained in the FIT image (for compatibility with older U-Boot which didn't care about pstore), so it should be not added again.

Quoted 82 lines · Detlev Casanova

The pstore command tries to create a reserved-memory node but fails if

it is already present with:
Add 'reserved-memory' node failed: FDT_ERR_EXISTS

This patch creates the node only if it does not exist and adapts the reg
values sizes depending on already present #address-cells and #size-cells
values.

---

Changes in v2:
- Move declarations at beginning of function
- Use if instead of switch
- Reword Subject

 cmd/pstore.c | 39 ++++++++++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 5 deletions(-)
cmd/pstore.c +1 -0
index 9fac8c7218..cd6f6feb2f 100644
@@ -11,6 +11,7 @@
 
 #include <mapmem.h>
 #include <memalign.h>
 #include <part.h>
 
+#include <fdt_support.h>
 

struct persistent_ram_buffer {

cmd/pstore.c +2 -0
 	u32    sig;
@@ -485,6 +486,8 @@ void fdt_fixup_pstore(void *blob)
 
 {
 
 	char node[32];
 	int  nodeoffset;	/* node offset from libfdt */
 
+	u32 addr_cells;
+	u32 size_cells;
cmd/pstore.c +1 -1
 	nodeoffset = fdt_path_offset(blob, "/");
 	if (nodeoffset < 0) {
@@ -493,14 +496,18 @@ void fdt_fixup_pstore(void *blob)
 
 		return;
 
 	}
 
-	nodeoffset = fdt_add_subnode(blob, nodeoffset, "reserved-memory");
+	nodeoffset = fdt_find_or_add_subnode(blob, nodeoffset,

"reserved-memory");>

cmd/pstore.c +2 -2
 	if (nodeoffset < 0) {
 
 		log_err("Add 'reserved-memory' node failed: %s\n",
 
 				fdt_strerror(nodeoffset));
 
 		return;
 
 	}
 
-	fdt_setprop_u32(blob, nodeoffset, "#address-cells", 2);
-	fdt_setprop_u32(blob, nodeoffset, "#size-cells", 2);
+
+	addr_cells = fdt_getprop_u32_default_node(blob, nodeoffset, 0,

"#address-cells", 2); + size_cells =
fdt_getprop_u32_default_node(blob,
nodeoffset, 0, "#size-cells", 2); + fdt_setprop_u32(blob, nodeoffset,
"#address-cells", addr_cells); + fdt_setprop_u32(blob, nodeoffset,
"#size-cells", size_cells);
+

Quoted 7 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 517

Quoted 7 lines
cmd/pstore.c +0 -0
 	fdt_setprop_empty(blob, nodeoffset, "ranges");
 
 	sprintf(node, "ramoops@%llx", (unsigned long long)pstore_addr);
@@ -509,14 +516,36 @@ void fdt_fixup_pstore(void *blob)
 
 		log_err("Add '%s' node failed: %s\n", node,

fdt_strerror(nodeoffset));

Quoted 8 lines
Quoted 8 lines
cmd/pstore.c +1 -0
 		return;
 
 	}
 
+
 
 	fdt_setprop_string(blob, nodeoffset, "compatible", "ramoops");
 
Quoted 9 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 9 lines
cmd/pstore.c +7 -2
-	fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
-	fdt_appendprop_u64(blob, nodeoffset, "reg", pstore_length);
+
+	if (addr_cells == 1) {
+		fdt_setprop_u32(blob, nodeoffset, "reg", pstore_addr);
+	} else if (addr_cells == 2) {
+		fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
+	} else {
+		log_err("Unsupported #address-cells: %u\n",

addr_cells);

Quoted 5 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 5 lines
cmd/pstore.c +5 -0
+		goto clean_ramoops;
+	}
+
+	if (size_cells == 1) {
+		// Let's consider that the pstore_length fits in a 32

bits value

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 538

Quoted 1 line
cmd/pstore.c +1 -0
+		fdt_appendprop_u32(blob, nodeoffset, "reg",

pstore_length);

Quoted 2 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 540

Quoted 2 lines
cmd/pstore.c +2 -0
+	} else if (size_cells == 2) {
+		fdt_appendprop_u64(blob, nodeoffset, "reg",

pstore_length);

Quoted 7 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 7 lines
cmd/pstore.c +5 -0
+	} else {
+		log_err("Unsupported #size-cells: %u\n", addr_cells);
+		goto clean_ramoops;
+	}
+
 
 	fdt_setprop_u32(blob, nodeoffset, "record-size",

pstore_record_size);

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 548

Quoted 1 line
cmd/pstore.c +0 -0
 	fdt_setprop_u32(blob, nodeoffset, "console-size",

pstore_console_size);

Quoted 1 line
Quoted 1 line

fdt_setprop_u32(blob, nodeoffset, "ftrace-size",

pstore_ftrace_size);

Quoted 10 lines
Quoted 10 lines
cmd/pstore.c +3 -0
 	fdt_setprop_u32(blob, nodeoffset, "pmsg-size", pstore_pmsg_size);
 	fdt_setprop_u32(blob, nodeoffset, "ecc-size", pstore_ecc_size);
 
+
+clean_ramoops:
+	fdt_del_node_and_alias(blob, node);
 
 }
 
 U_BOOT_CMD(pstore, 10, 0, do_pstore,
#6 lore raw

Hi Detlev,

Quoted 9 lines · Detlev Casanova

Hey Daniel,

I see that the node is called ramoops@0x42ff0000. We usually don't set the 0x in device tree nodes, and that's why the fdt_add_subnode function doesn't see it and add ramoops@42ff0000 (it just compares the text). For Linux, both nodes are the same and it complains.

I think the best course of action here is to remove that '0x' in your dts, so that u-boot sees it as the pstore node.

I tried the change you suggest and remove the '0x' prefix and indeed that works, the lengthy kernel warning about duplicate nodes and sysfs entries are gone and pstore works as expected.
I've hence committed that change to OpenWrt:

https://git.openwrt.org/?p=openwrt/openwrt.git;a=commit;h=fc245338d6e02e61fa7ecbd1a828aed97cdbef88

However, now there is another error message in U-Boot output:
Add 'ramoops@42ff0000' node failed: FDT_ERR_EXISTS

I figured that this is not a problem as previously (ie. before your patch) even the exitence of the 'reserved-memory' node would result in the 'ramoops' node not being added while now it really just errors out in case of exactly that 'ramoops' node already existing, which is fine.

Thank you again for resolving this issue!

Cheers

Daniel

Quoted 245 lines

Regards,

Detlev.

Quoted 235 lines · On Monday, April 18, 2022 7:50:36 P.M. EDT Daniel Golle wrote:

Hi Detlev,

Quoted 8 lines · Detlev Casanova

Hi Daniel,

The patch only checks the existence of the reserved-memory node.

I guess that before, adding the reserved-memory node failed and so nothing else was happening. Now, the node is found and then, it adds the ramoops subnode to it, making it double because it is already there in your case.

Quoted 6 lines

But that should not happen, because the fdt_add_subnode() function should fail if the node is already present (like it was failing for reserved-memory).

I'm going to look into it. Can you send me the device tree taht you are using (or at least the reserved-memory part) and u-boot config ?

Thank you for your reply and for taking a look at what's going on.

reserved-memory bits in dts:
https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=package/boot/uboot-m
ediatek/patches/050-mt7622-enable-pstore.patch

u-boot config:
https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=package/boot/uboot-m
ediatek/patches/404-add-bananapi_bpi-r64_defconfigs.patch

Cheers

Daniel

Quoted 201 lines

Detlev.

Quoted 123 lines · On Monday, April 11, 2022 7:00:20 P.M. EDT Daniel Golle wrote:

Hi Detlev,

I've recently tried to update U-Boot from 2022.01 to 2022.04 and noticed a regression introduced by the commit below.

While the unwanted error message ("Add 'reserved-memory' node failed:
FDT_ERR_EXISTS") no longer appears, I now see a lengthy kernel stack trace in Linux instead:
...

[    0.008295] sysfs: cannot create duplicate filename
'/devices/platform/42ff0000.ramoops' [    0.008303] CPU: 0 PID: 1 Comm:
swapper/0 Tainted: G S                5.15.33 #0 [    0.008312] Hardware
name: Bananapi BPI-R64 (DT)
[    0.008318] Call trace:
[    0.008322]  dump_backtrace+0x0/0x15c
[    0.008337]  show_stack+0x14/0x30
[    0.008345]  dump_stack_lvl+0x64/0x7c
[    0.008355]  dump_stack+0x14/0x2c
[    0.008362]  sysfs_warn_dup+0x5c/0x74
[    0.008373]  sysfs_create_dir_ns+0xb0/0xc0
[    0.008381]  kobject_add_internal+0xbc/0x330
[    0.008392]  kobject_add+0x80/0xe0
[    0.008400]  device_add+0xd4/0x82c
[    0.008410]  of_device_add+0x4c/0x5c
[    0.008420]  of_platform_device_create_pdata+0xb8/0xf0
[    0.008428]  of_platform_default_populate_init+0x58/0xcc
[    0.008437]  do_one_initcall+0x4c/0x1b0
[    0.008444]  kernel_init_freeable+0x230/0x294
[    0.008456]  kernel_init+0x20/0x120
[    0.008463]  ret_from_fork+0x10/0x20
[    0.008471] kobject_add_internal failed for 42ff0000.ramoops with

-EEXIST, don't try to register things with the same name in the same directory. ...

I assume that fdt_find_or_add_subnode() doesn't behave as expected and apparently adds a duplicate node having the same name.
The pstore/ramoops reserved-memory is already defined in the fdt contained in the FIT image (for compatibility with older U-Boot which didn't care about pstore), so it should be not added again.

Quoted 82 lines · Detlev Casanova

The pstore command tries to create a reserved-memory node but fails if

it is already present with:
Add 'reserved-memory' node failed: FDT_ERR_EXISTS

This patch creates the node only if it does not exist and adapts the reg
values sizes depending on already present #address-cells and #size-cells
values.

---

Changes in v2:
- Move declarations at beginning of function
- Use if instead of switch
- Reword Subject

 cmd/pstore.c | 39 ++++++++++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 5 deletions(-)
cmd/pstore.c +1 -0
index 9fac8c7218..cd6f6feb2f 100644
@@ -11,6 +11,7 @@
 
 #include <mapmem.h>
 #include <memalign.h>
 #include <part.h>
 
+#include <fdt_support.h>
 

struct persistent_ram_buffer {

cmd/pstore.c +2 -0
 	u32    sig;
@@ -485,6 +486,8 @@ void fdt_fixup_pstore(void *blob)
 
 {
 
 	char node[32];
 	int  nodeoffset;	/* node offset from libfdt */
 
+	u32 addr_cells;
+	u32 size_cells;
cmd/pstore.c +1 -1
 	nodeoffset = fdt_path_offset(blob, "/");
 	if (nodeoffset < 0) {
@@ -493,14 +496,18 @@ void fdt_fixup_pstore(void *blob)
 
 		return;
 
 	}
 
-	nodeoffset = fdt_add_subnode(blob, nodeoffset, "reserved-memory");
+	nodeoffset = fdt_find_or_add_subnode(blob, nodeoffset,

"reserved-memory");>

cmd/pstore.c +2 -2
 	if (nodeoffset < 0) {
 
 		log_err("Add 'reserved-memory' node failed: %s\n",
 
 				fdt_strerror(nodeoffset));
 
 		return;
 
 	}
 
-	fdt_setprop_u32(blob, nodeoffset, "#address-cells", 2);
-	fdt_setprop_u32(blob, nodeoffset, "#size-cells", 2);
+
+	addr_cells = fdt_getprop_u32_default_node(blob, nodeoffset, 0,

"#address-cells", 2); + size_cells =
fdt_getprop_u32_default_node(blob,
nodeoffset, 0, "#size-cells", 2); + fdt_setprop_u32(blob, nodeoffset,
"#address-cells", addr_cells); + fdt_setprop_u32(blob, nodeoffset,
"#size-cells", size_cells);
+

Quoted 7 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 517

Quoted 7 lines
cmd/pstore.c +0 -0
 	fdt_setprop_empty(blob, nodeoffset, "ranges");
 
 	sprintf(node, "ramoops@%llx", (unsigned long long)pstore_addr);
@@ -509,14 +516,36 @@ void fdt_fixup_pstore(void *blob)
 
 		log_err("Add '%s' node failed: %s\n", node,

fdt_strerror(nodeoffset));

Quoted 8 lines
Quoted 8 lines
cmd/pstore.c +1 -0
 		return;
 
 	}
 
+
 
 	fdt_setprop_string(blob, nodeoffset, "compatible", "ramoops");
 
Quoted 9 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 9 lines
cmd/pstore.c +7 -2
-	fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
-	fdt_appendprop_u64(blob, nodeoffset, "reg", pstore_length);
+
+	if (addr_cells == 1) {
+		fdt_setprop_u32(blob, nodeoffset, "reg", pstore_addr);
+	} else if (addr_cells == 2) {
+		fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
+	} else {
+		log_err("Unsupported #address-cells: %u\n",

addr_cells);

Quoted 5 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 5 lines
cmd/pstore.c +5 -0
+		goto clean_ramoops;
+	}
+
+	if (size_cells == 1) {
+		// Let's consider that the pstore_length fits in a 32

bits value

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 538

Quoted 1 line
cmd/pstore.c +1 -0
+		fdt_appendprop_u32(blob, nodeoffset, "reg",

pstore_length);

Quoted 2 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 540

Quoted 2 lines
cmd/pstore.c +2 -0
+	} else if (size_cells == 2) {
+		fdt_appendprop_u64(blob, nodeoffset, "reg",

pstore_length);

Quoted 7 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 7 lines
cmd/pstore.c +5 -0
+	} else {
+		log_err("Unsupported #size-cells: %u\n", addr_cells);
+		goto clean_ramoops;
+	}
+
 
 	fdt_setprop_u32(blob, nodeoffset, "record-size",

pstore_record_size);

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 548

Quoted 1 line
cmd/pstore.c +0 -0
 	fdt_setprop_u32(blob, nodeoffset, "console-size",

pstore_console_size);

Quoted 1 line
Quoted 1 line

fdt_setprop_u32(blob, nodeoffset, "ftrace-size",

pstore_ftrace_size);

Quoted 10 lines
Quoted 10 lines
cmd/pstore.c +3 -0
 	fdt_setprop_u32(blob, nodeoffset, "pmsg-size", pstore_pmsg_size);
 	fdt_setprop_u32(blob, nodeoffset, "ecc-size", pstore_ecc_size);
 
+
+clean_ramoops:
+	fdt_del_node_and_alias(blob, node);
 
 }
 
 U_BOOT_CMD(pstore, 10, 0, do_pstore,
#7 lore raw

Hi Daniel,

Quoted 21 lines · On Tuesday, April 19, 2022 12:14:07 P.M. EDT Daniel Golle wrote:

Hi Detlev,

Quoted 9 lines · Detlev Casanova

Hey Daniel,

I see that the node is called ramoops@0x42ff0000. We usually don't set the 0x in device tree nodes, and that's why the fdt_add_subnode function doesn't see it and add ramoops@42ff0000 (it just compares the text). For Linux, both nodes are the same and it complains.

I think the best course of action here is to remove that '0x' in your dts, so that u-boot sees it as the pstore node.

I tried the change you suggest and remove the '0x' prefix and indeed that works, the lengthy kernel warning about duplicate nodes and sysfs entries are gone and pstore works as expected.
I've hence committed that change to OpenWrt:

https://git.openwrt.org/?p=openwrt/openwrt.git;a=commit;h=fc245338d6e02e61fa
7ecbd1a828aed97cdbef88

Quoted 7 lines

However, now there is another error message in U-Boot output:
Add 'ramoops@42ff0000' node failed: FDT_ERR_EXISTS

I figured that this is not a problem as previously (ie. before your patch) even the exitence of the 'reserved-memory' node would result in the 'ramoops' node not being added while now it really just errors out in case of exactly that 'ramoops' node already existing, which is fine.

That's right, you can see that error message as "You activated pstore in the config to make me add the ramoops node, but it's already there. I cannot do my job" :)

Quoted 175 lines

Thank you again for resolving this issue!

Cheers

Daniel

Quoted 167 lines

Regards,

Detlev.

Quoted 162 lines · On Monday, April 18, 2022 7:50:36 P.M. EDT Daniel Golle wrote:

Hi Detlev,

Quoted 11 lines · Detlev Casanova

Hi Daniel,

The patch only checks the existence of the reserved-memory node.

I guess that before, adding the reserved-memory node failed and so nothing
else was happening. Now, the node is found and then, it adds the ramoops
subnode to it, making it double because it is already there in your case.

Quoted 8 lines

But that should not happen, because the fdt_add_subnode() function should
fail if the node is already present (like it was failing for reserved-memory).

I'm going to look into it. Can you send me the device tree taht you are
using (or at least the reserved-memory part) and u-boot config ?

Thank you for your reply and for taking a look at what's going on.

reserved-memory bits in dts:
https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=package/boot/ubo
ot-m ediatek/patches/050-mt7622-enable-pstore.patch

u-boot config:
https://git.openwrt.org/?p=openwrt/openwrt.git;a=blob;f=package/boot/ubo
ot-m ediatek/patches/404-add-bananapi_bpi-r64_defconfigs.patch

Cheers

Daniel

Quoted 123 lines

Detlev.

Quoted 120 lines · On Monday, April 11, 2022 7:00:20 P.M. EDT Daniel Golle wrote:

Hi Detlev,

I've recently tried to update U-Boot from 2022.01 to 2022.04 and noticed a regression introduced by the commit below.

While the unwanted error message ("Add 'reserved-memory' node failed:
FDT_ERR_EXISTS") no longer appears, I now see a lengthy kernel stack trace in Linux instead:
...

[    0.008295] sysfs: cannot create duplicate filename

'/devices/platform/42ff0000.ramoops' [ 0.008303] CPU: 0 PID: 1
Comm:
swapper/0 Tainted: G S 5.15.33 #0 [ 0.008312]
Hardware
name: Bananapi BPI-R64 (DT)

[    0.008318] Call trace:
[    0.008322]  dump_backtrace+0x0/0x15c
[    0.008337]  show_stack+0x14/0x30
[    0.008345]  dump_stack_lvl+0x64/0x7c
[    0.008355]  dump_stack+0x14/0x2c
[    0.008362]  sysfs_warn_dup+0x5c/0x74
[    0.008373]  sysfs_create_dir_ns+0xb0/0xc0
[    0.008381]  kobject_add_internal+0xbc/0x330
[    0.008392]  kobject_add+0x80/0xe0
[    0.008400]  device_add+0xd4/0x82c
[    0.008410]  of_device_add+0x4c/0x5c
[    0.008420]  of_platform_device_create_pdata+0xb8/0xf0
[    0.008428]  of_platform_default_populate_init+0x58/0xcc
[    0.008437]  do_one_initcall+0x4c/0x1b0
[    0.008444]  kernel_init_freeable+0x230/0x294
[    0.008456]  kernel_init+0x20/0x120
[    0.008463]  ret_from_fork+0x10/0x20
[    0.008471] kobject_add_internal failed for 42ff0000.ramoops with

-EEXIST, don't try to register things with the same name in the same directory. ...

I assume that fdt_find_or_add_subnode() doesn't behave as expected and
apparently adds a duplicate node having the same name. The pstore/ramoops reserved-memory is already defined in the fdt contained in the FIT image (for compatibility with older U-Boot which
didn't care about pstore), so it should be not added again.

Quoted 74 lines · Detlev Casanova

The pstore command tries to create a reserved-memory node but fails if

it is already present with:
Add 'reserved-memory' node failed: FDT_ERR_EXISTS

This patch creates the node only if it does not exist and adapts the
reg
values sizes depending on already present #address-cells and #size-cells
values.

---

Changes in v2:
- Move declarations at beginning of function
- Use if instead of switch
- Reword Subject

 cmd/pstore.c | 39 ++++++++++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 5 deletions(-)
cmd/pstore.c +1 -0
index 9fac8c7218..cd6f6feb2f 100644
@@ -11,6 +11,7 @@
 
 #include <mapmem.h>
 #include <memalign.h>
 #include <part.h>
 
+#include <fdt_support.h>
 

struct persistent_ram_buffer {

cmd/pstore.c +2 -0
 	u32    sig;
@@ -485,6 +486,8 @@ void fdt_fixup_pstore(void *blob)
 
 {
 
 	char node[32];
 	int  nodeoffset;	/* node offset from libfdt */
 
+	u32 addr_cells;
+	u32 size_cells;
cmd/pstore.c +1 -1
 	nodeoffset = fdt_path_offset(blob, "/");
 	if (nodeoffset < 0) {
@@ -493,14 +496,18 @@ void fdt_fixup_pstore(void *blob)
 
 		return;
 
 	}
 
-	nodeoffset = fdt_add_subnode(blob, nodeoffset,
 "reserved-memory");
+	nodeoffset = fdt_find_or_add_subnode(blob, nodeoffset,

"reserved-memory");>

cmd/pstore.c +0 -0
 	if (nodeoffset < 0) {
 
 		log_err("Add 'reserved-memory' node failed: %s\n",
 
 				fdt_strerror(nodeoffset));
 
 		return;
 
 	}
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
cmd/pstore.c +2 -2
-	fdt_setprop_u32(blob, nodeoffset, "#address-cells", 2);
-	fdt_setprop_u32(blob, nodeoffset, "#size-cells", 2);
+
+	addr_cells = fdt_getprop_u32_default_node(blob, nodeoffset, 0,

"#address-cells", 2); + size_cells =
fdt_getprop_u32_default_node(blob,
nodeoffset, 0, "#size-cells", 2); + fdt_setprop_u32(blob,
nodeoffset,
"#address-cells", addr_cells); + fdt_setprop_u32(blob,

nodeoffset,

Quoted 79 lines
Quoted 79 lines
Quoted 79 lines
Quoted 79 lines
Quoted 10 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 517

Quoted 10 lines

"#size-cells", size_cells);

cmd/pstore.c +1 -0
+
 
 	fdt_setprop_empty(blob, nodeoffset, "ranges");
 
 	sprintf(node, "ramoops@%llx", (unsigned long long)pstore_addr);
@@ -509,14 +516,36 @@ void fdt_fixup_pstore(void *blob)
 
 		log_err("Add '%s' node failed: %s\n", node,

fdt_strerror(nodeoffset));

Quoted 8 lines
Quoted 8 lines
cmd/pstore.c +1 -0
 		return;
 
 	}
 
+
 
 	fdt_setprop_string(blob, nodeoffset, "compatible", "ramoops");
 
Quoted 9 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 9 lines
cmd/pstore.c +7 -2
-	fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
-	fdt_appendprop_u64(blob, nodeoffset, "reg", pstore_length);
+
+	if (addr_cells == 1) {
+		fdt_setprop_u32(blob, nodeoffset, "reg", pstore_addr);
+	} else if (addr_cells == 2) {
+		fdt_setprop_u64(blob, nodeoffset, "reg", pstore_addr);
+	} else {
+		log_err("Unsupported #address-cells: %u\n",

addr_cells);

Quoted 5 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 5 lines
cmd/pstore.c +5 -0
+		goto clean_ramoops;
+	}
+
+	if (size_cells == 1) {
+		// Let's consider that the pstore_length fits in a 32

bits value

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 538

Quoted 1 line
cmd/pstore.c +1 -0
+		fdt_appendprop_u32(blob, nodeoffset, "reg",

pstore_length);

Quoted 2 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 540

Quoted 2 lines
cmd/pstore.c +2 -0
+	} else if (size_cells == 2) {
+		fdt_appendprop_u64(blob, nodeoffset, "reg",

pstore_length);

Quoted 7 lines

cmd/pstore.c void fdt_fixup_pstore(void *blob)

Quoted 7 lines
cmd/pstore.c +5 -0
+	} else {
+		log_err("Unsupported #size-cells: %u\n", addr_cells);
+		goto clean_ramoops;
+	}
+
 
 	fdt_setprop_u32(blob, nodeoffset, "record-size",

pstore_record_size);

Quoted 1 line

cmd/pstore.c void fdt_fixup_pstore(void *blob) · line 548

Quoted 1 line
cmd/pstore.c +0 -0
 	fdt_setprop_u32(blob, nodeoffset, "console-size",

pstore_console_size);

Quoted 1 line
Quoted 1 line

fdt_setprop_u32(blob, nodeoffset, "ftrace-size",

pstore_ftrace_size);

Quoted 11 lines
Quoted 11 lines
cmd/pstore.c +3 -0
 	fdt_setprop_u32(blob, nodeoffset, "pmsg-size",
 	pstore_pmsg_size);
 	fdt_setprop_u32(blob, nodeoffset, "ecc-size", pstore_ecc_size);
 
+
+clean_ramoops:
+	fdt_del_node_and_alias(blob, node);
 
 }
 
 U_BOOT_CMD(pstore, 10, 0, do_pstore,