Skip to contentSkip to conversation

Fix unbootable ARMv7 devices

PATCH In review Brian Sune 2 participants 5 messages Jan 30 Arm (32-bit)

lore
Share & tools
lore

Brian Sune

author
#0 lore raw

PATCH 1/1 Fix unbootable ARMv7 devices

The linker script u-boot-spl.lds now
is no longer aligned and -nodtb.bin ending shows the zeros are not 8 byte aligned. This result in the Makefile.xpl simply think the previous file is aligned and do not zero pad. simply fix the data align to 8 can ensure the -nodtb.bin is aligned, hence, the formation of u-boot-spl.bin properly align.

---
 arch/arm/cpu/u-boot-spl.lds | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/cpu/u-boot-spl.lds +1 -1
index c578c3ebf82..18b0f1e7284 100644
@@ -26,7 +26,7 @@ SECTIONS
 	. = ALIGN(4);
 	.rodata : { *(SORT_BY_ALIGNMENT(SORT_BY_NAME(.rodata*))) }
 
-	. = ALIGN(4);
+	. = ALIGN(8);
 	.data : {
 		*(.data*)
 	}
Signature · 1 line
2.34.1
#1 lore raw
Quoted 8 lines · Brian Sune

The linker script u-boot-spl.lds now
is no longer aligned and -nodtb.bin ending shows the zeros are not 8 byte aligned. This result in the Makefile.xpl simply think the previous file is aligned and do not zero pad. simply fix the data align to 8 can ensure the -nodtb.bin is aligned, hence, the formation of u-boot-spl.bin properly align.

Quoted 4 lines
---
 arch/arm/cpu/u-boot-spl.lds | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

This fixes your problem in that it results in the device tree being aligned still, but doesn't solve the underlying problem of why there were 4 extra bytes. Can you please try:

arch/arm/cpu/u-boot-spl.lds +5 -5
index c578c3ebf821..5f807323f856 100644
@@ -31,16 +31,16 @@ SECTIONS
 		*(.data*)
 	}
 
-	. = ALIGN(4);
-	__u_boot_list : {
-		KEEP(*(SORT(__u_boot_list*)));
-	}
-
 	. = ALIGN(4);
 	.binman_sym_table : {
 		__binman_sym_start = .;
 		KEEP(*(SORT(.binman_sym*)));
 		__binman_sym_end = .;
+	}
+
+	. = ALIGN(4);
+	__u_boot_list : {
+		KEEP(*(SORT(__u_boot_list*)));
 		. = ALIGN(8);
 	}
 

As the problem ends up being that when both of these are true:
- .binman_sym_table is empty of actual .binman_sym* symbols
- The inner '. = ALIGN(8);' increments the linker counter from the previous '. = ALIGN(4);' (in other words we were not already 8b aligned).

The linker sets the type for this section to "NOBITS" which in turn means that when we later call objcopy to create u-boot-spl-nodtb.bin we get a mismatch between our expected size and actual size for creating the -pad.bin file.

Signature · 1 line
Tom
author
#2 lore raw
Quoted 11 lines · Tom Rini
Quoted 8 lines · Brian Sune

The linker script u-boot-spl.lds now
is no longer aligned and -nodtb.bin ending shows the zeros are not 8 byte aligned. This result in the Makefile.xpl simply think the previous file is aligned and do not zero pad. simply fix the data align to 8 can ensure the -nodtb.bin is aligned, hence, the formation of u-boot-spl.bin properly align.

Quoted 7 lines
Quoted 4 lines
---
 arch/arm/cpu/u-boot-spl.lds | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

This fixes your problem in that it results in the device tree being aligned still, but doesn't solve the underlying problem of why there

No it is not 4 extra bytes but 4 bytes short on the latest master commits.

Quoted 18 lines

were 4 extra bytes. Can you please try:

arch/arm/cpu/u-boot-spl.lds +0 -5
index c578c3ebf821..5f807323f856 100644
@@ -31,16 +31,16 @@ SECTIONS
                *(.data*)
        }
 
-       . = ALIGN(4);
-       __u_boot_list : {
-               KEEP(*(SORT(__u_boot_list*)));
-       }
-
        . = ALIGN(4);
        .binman_sym_table : {
                __binman_sym_start = .;

arch/arm/cpu/u-boot-spl.lds SECTIONS · line 46

Quoted 10 lines
arch/arm/cpu/u-boot-spl.lds +5 -0
                KEEP(*(SORT(.binman_sym*)));
                __binman_sym_end = .;
+       }
+
+       . = ALIGN(4);
+       __u_boot_list : {
+               KEEP(*(SORT(__u_boot_list*)));
                . = ALIGN(8);
        }
 

This patch will create the same result but the actual problem never caused by the -pad.bin.

The bss size is 8 byte aligned hence it will not be the issue from first place at Makefile.xpl as it will not create any zero pad.

The issue here is that the nodtb size itself size is never 8 byte aligned at the end result in short or long cases while the Makefile.xpl only creates extra zero based on the bss size rather than the nodtb size.

Then they simply join nodtb + pad + bss but from first place nodtb is not sized correctly at the end aka 8 byte sized.

Any method that can make the end back to 8 byte size of the nodtb should do the job.

Quoted 5 lines

As the problem ends up being that when both of these are true:
- .binman_sym_table is empty of actual .binman_sym* symbols
- The inner '. = ALIGN(8);' increments the linker counter from the previous '. = ALIGN(4);' (in other words we were not already 8b aligned).

So what you are suggesting is that the section of binman_sym_table is empty so the ALIGN(8) just simply cannot do its 8 byte alignment job.

And reversing the order makes sure it will always ALIGN end with 8. But does "__u_boot_list" result in an empty case? If not then this indeed do what I also wanted to.

I intended not to change any ordering to fix the issue.

Brian

Quoted 8 lines

The linker sets the type for this section to "NOBITS" which in turn means that when we later call objcopy to create u-boot-spl-nodtb.bin we get a mismatch between our expected size and actual size for creating the -pad.bin file.

--
Tom

author
#4 lore raw
In reply to Tom Rini
Quoted 11 lines · Tom Rini
Quoted 8 lines · Brian Sune

The linker script u-boot-spl.lds now
is no longer aligned and -nodtb.bin ending shows the zeros are not 8 byte aligned. This result in the Makefile.xpl simply think the previous file is aligned and do not zero pad. simply fix the data align to 8 can ensure the -nodtb.bin is aligned, hence, the formation of u-boot-spl.bin properly align.

Quoted 7 lines
Quoted 4 lines
---
 arch/arm/cpu/u-boot-spl.lds | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

This fixes your problem in that it results in the device tree being aligned still, but doesn't solve the underlying problem of why there

Hi Tom,

Another follow up question here is even making the data node to 8 byte aligned will this create any harm or even fail case from first place? As for testing on a real platform it makes no difference at all.

If "binman_sym_table" is empty then the previous node data is 8 byte aligned. Where "__u_boot_list" is also gated by 8 byte end aligned unless it could be empty while "binman_sym_table" does not.

Just wondering and not really too good at these linker scripts.

brian

Quoted 41 lines

were 4 extra bytes. Can you please try:

arch/arm/cpu/u-boot-spl.lds +5 -5
index c578c3ebf821..5f807323f856 100644
@@ -31,16 +31,16 @@ SECTIONS
                *(.data*)
        }
 
-       . = ALIGN(4);
-       __u_boot_list : {
-               KEEP(*(SORT(__u_boot_list*)));
-       }
-
        . = ALIGN(4);
        .binman_sym_table : {
                __binman_sym_start = .;
                KEEP(*(SORT(.binman_sym*)));
                __binman_sym_end = .;
+       }
+
+       . = ALIGN(4);
+       __u_boot_list : {
+               KEEP(*(SORT(__u_boot_list*)));
                . = ALIGN(8);
        }
 

As the problem ends up being that when both of these are true:
- .binman_sym_table is empty of actual .binman_sym* symbols
- The inner '. = ALIGN(8);' increments the linker counter from the previous '. = ALIGN(4);' (in other words we were not already 8b aligned).

The linker sets the type for this section to "NOBITS" which in turn means that when we later call objcopy to create u-boot-spl-nodtb.bin we get a mismatch between our expected size and actual size for creating the -pad.bin file.

--
Tom

#3 lore raw
In reply to Sune Brian
Quoted 17 lines · Sune Brian
Quoted 16 lines · Tom Rini
Quoted 12 lines · Brian Sune

The linker script u-boot-spl.lds now
is no longer aligned and -nodtb.bin ending shows the zeros are not 8 byte aligned. This result in the Makefile.xpl simply think the previous file is aligned and do not zero pad. simply fix the data align to 8 can ensure the -nodtb.bin is aligned, hence, the formation of u-boot-spl.bin properly align.

---
 arch/arm/cpu/u-boot-spl.lds | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 6 lines
Quoted 2 lines

This fixes your problem in that it results in the device tree being aligned still, but doesn't solve the underlying problem of why there

No it is not 4 extra bytes but 4 bytes short on the latest master commits.

Quoted 1 line

were 4 extra bytes. Can you please try:

Yes, we're off by 4 bytes. Sometimes in one direction, sometimes in the other. I believe in either case it's because of not accounting for a non-empty section correctly.

Quoted 30 lines
Quoted 16 lines
arch/arm/cpu/u-boot-spl.lds +0 -5
index c578c3ebf821..5f807323f856 100644
@@ -31,16 +31,16 @@ SECTIONS
                *(.data*)
        }
 
-       . = ALIGN(4);
-       __u_boot_list : {
-               KEEP(*(SORT(__u_boot_list*)));
-       }
-
        . = ALIGN(4);
        .binman_sym_table : {
                __binman_sym_start = .;

arch/arm/cpu/u-boot-spl.lds SECTIONS · line 46

Quoted 10 lines
arch/arm/cpu/u-boot-spl.lds +5 -0
                KEEP(*(SORT(.binman_sym*)));
                __binman_sym_end = .;
+       }
+
+       . = ALIGN(4);
+       __u_boot_list : {
+               KEEP(*(SORT(__u_boot_list*)));
                . = ALIGN(8);
        }
 

This patch will create the same result but the actual problem never caused by the -pad.bin.

Quoted 2 lines

The bss size is 8 byte aligned hence it will not be the issue from first place at Makefile.xpl as it will not create any zero pad.

Right, the pad file is created the correct size.

Quoted 4 lines

The issue here is that the nodtb size itself size is never 8 byte aligned at the end result in short or long cases while the Makefile.xpl only creates extra zero based on the bss size rather than the nodtb size.

Yes, the nodtb isn't aligned, but it's supposed to be.

The ASSERT statements we have mean that the ELF object is correct, but when we transform it, we're breaking it.

Quoted 5 lines

Then they simply join nodtb + pad + bss but from first place nodtb is not sized correctly at the end aka 8 byte sized.

Any method that can make the end back to 8 byte size of the nodtb should do the job.

It's a two-part problem. The file has to be aligned, and we have to place the dtb at where __bss_end (in the case of SPL_SEPARATE_BSS=n) is at run-time.

Quoted 8 lines
Quoted 5 lines

As the problem ends up being that when both of these are true:
- .binman_sym_table is empty of actual .binman_sym* symbols
- The inner '. = ALIGN(8);' increments the linker counter from the previous '. = ALIGN(4);' (in other words we were not already 8b aligned).

So what you are suggesting is that the section of binman_sym_table is empty so the ALIGN(8) just simply cannot do its 8 byte alignment job.

Not quite. When we say "ALIGN(8)" we advance the linkers program counter but that's not the same as saying "add bytes of 0x0 until we reach the counter".

Quoted 4 lines

And reversing the order makes sure it will always ALIGN end with 8. But does "__u_boot_list" result in an empty case? If not then this indeed do what I also wanted to.

Quoted 9 lines

I intended not to change any ordering to fix the issue.

Brian

Quoted 5 lines

The linker sets the type for this section to "NOBITS" which in turn means that when we later call objcopy to create u-boot-spl-nodtb.bin we get a mismatch between our expected size and actual size for creating the -pad.bin file.

Which is what I'm explaining here. We have to make sure the last section before the BSS is something that will never be empty. That's what makes sure that the final ALIGN does what we expect.

Signature · 1 line
Tom