RTL9300 (mips) fails to boot with commit d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output section")

RTL9300 (mips) fails to boot with commit d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output section")

[Chris Packham]

Hi Rong,

I was just trying some of my in-flight changes against linux-next and 
found that my RTL9300* based board no longer boots (nothing on the 
serial console after U-Boot hands off to the kernel).

A bisect between Linus's tree and linux-next/master points to commit 
d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output 
section"). A build on d4a7f2510ff4^ boots successfully. That series is 
well beyond my area of knowledge, there may be something that the MIPS 
SoCs are doing (or not doing) that is being affected by the sections moving.

Thanks,
Chris

[*] - The RTL9300 is a Realtek Ethernet Switch with integrated MIPS SoC.

Re: RTL9300 (mips) fails to boot with commit d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output section")

[Chris Packham]

On 22/11/24 14:58, Chris Packham wrote:
> Hi Rong,
>
> I was just trying some of my in-flight changes against linux-next and 
> found that my RTL9300* based board no longer boots (nothing on the 
> serial console after U-Boot hands off to the kernel).
>
> A bisect between Linus's tree and linux-next/master points to commit 
> d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output 
> section"). A build on d4a7f2510ff4^ boots successfully. That series is 
> well beyond my area of knowledge, there may be something that the MIPS 
> SoCs are doing (or not doing) that is being affected by the sections 
> moving.

Based on the sparc change that accompanied this commit the following 
does seem to restore working order for me:

diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
index d575f945d422..e90b0f68290c 100644
--- a/arch/mips/kernel/vmlinux.lds.S
+++ b/arch/mips/kernel/vmlinux.lds.S
@@ -62,6 +62,8 @@ SECTIONS
         _text = .;      /* Text and read-only data */
         _stext = .;
         .text : {
+               /* Match text section symbols in head.S first */
+               *head.o(.text)
                 TEXT_TEXT
                 SCHED_TEXT
                 LOCK_TEXT

No idea if that's appropriate

Re: RTL9300 (mips) fails to boot with commit d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output section")

[Rong Xu]

Does the following patch fix your issue?
https://lore.kernel.org/lkml/CAK7LNAQSpM2hh2=Wgribb92nHF5_peRb+gxdO1Bt1MkeL8N+hw@mail.gmail.com/t/

Masahiro Yamada already applied the patch to kbuild-tree.

I also proposed a similar fix as yours. But we believe the above patch
should also work (and probably is better).

Please try and let us know.

Thanks!

-Rong

On Thu, Nov 21, 2024 at 6:29 PM Chris Packham
<chris.packham@alliedtelesis.co.nz> wrote:
>
>
> On 22/11/24 14:58, Chris Packham wrote:
> > Hi Rong,
> >
> > I was just trying some of my in-flight changes against linux-next and
> > found that my RTL9300* based board no longer boots (nothing on the
> > serial console after U-Boot hands off to the kernel).
> >
> > A bisect between Linus's tree and linux-next/master points to commit
> > d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output
> > section"). A build on d4a7f2510ff4^ boots successfully. That series is
> > well beyond my area of knowledge, there may be something that the MIPS
> > SoCs are doing (or not doing) that is being affected by the sections
> > moving.
>
> Based on the sparc change that accompanied this commit the following
> does seem to restore working order for me:
>
> diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
> index d575f945d422..e90b0f68290c 100644
> --- a/arch/mips/kernel/vmlinux.lds.S
> +++ b/arch/mips/kernel/vmlinux.lds.S
> @@ -62,6 +62,8 @@ SECTIONS
>          _text = .;      /* Text and read-only data */
>          _stext = .;
>          .text : {
> +               /* Match text section symbols in head.S first */
> +               *head.o(.text)
>                  TEXT_TEXT
>                  SCHED_TEXT
>                  LOCK_TEXT
>
> No idea if that's appropriate
>

Re: RTL9300 (mips) fails to boot with commit d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output section")

[Chris Packham]

Hi Rong,

On 22/11/24 18:02, Rong Xu wrote:
> Does the following patch fix your issue?
> https://lore.kernel.org/lkml/CAK7LNAQSpM2hh2=Wgribb92nHF5_peRb+gxdO1Bt1MkeL8N+hw@mail.gmail.com/t/
>
> Masahiro Yamada already applied the patch to kbuild-tree.
>
> I also proposed a similar fix as yours. But we believe the above patch
> should also work (and probably is better).
>
> Please try and let us know.

Looks like linux-next already has that patch and no it doesn't fix my 
issue. For the record here's the start of my System.map

$ head System.map
ffffffff80100000 T __split_text_end
ffffffff80100000 T __split_text_start
ffffffff80100000 T __unlikely_text_start
ffffffff80100000 T _stext
ffffffff80100000 T _text
ffffffff80100000 t run_init_process
ffffffff80100138 t try_to_run_init_process
ffffffff8010018c t put_page
ffffffff801001d0 t nr_blocks.isra.0
ffffffff80100208 T ack_bad_irq

It does look like whatever is supposed to put head.o first isn't kicking 
in for me.

>
> Thanks!
>
> -Rong
>
> On Thu, Nov 21, 2024 at 6:29 PM Chris Packham
> <chris.packham@alliedtelesis.co.nz> wrote:
>>
>> On 22/11/24 14:58, Chris Packham wrote:
>>> Hi Rong,
>>>
>>> I was just trying some of my in-flight changes against linux-next and
>>> found that my RTL9300* based board no longer boots (nothing on the
>>> serial console after U-Boot hands off to the kernel).
>>>
>>> A bisect between Linus's tree and linux-next/master points to commit
>>> d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output
>>> section"). A build on d4a7f2510ff4^ boots successfully. That series is
>>> well beyond my area of knowledge, there may be something that the MIPS
>>> SoCs are doing (or not doing) that is being affected by the sections
>>> moving.
>> Based on the sparc change that accompanied this commit the following
>> does seem to restore working order for me:
>>
>> diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
>> index d575f945d422..e90b0f68290c 100644
>> --- a/arch/mips/kernel/vmlinux.lds.S
>> +++ b/arch/mips/kernel/vmlinux.lds.S
>> @@ -62,6 +62,8 @@ SECTIONS
>>           _text = .;      /* Text and read-only data */
>>           _stext = .;
>>           .text : {
>> +               /* Match text section symbols in head.S first */
>> +               *head.o(.text)
>>                   TEXT_TEXT
>>                   SCHED_TEXT
>>                   LOCK_TEXT
>>
>> No idea if that's appropriate
>>

Re: RTL9300 (mips) fails to boot with commit d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output section")

[Rong Xu]

Thanks for confirming this! It looks like we do need to put all
symbols from head.o before any other symbols.
To be clear, the patch I mentioned was not trying to do that -- it
just makes sure the _stext is in the right places.

I'll draft a patch for this.

On Sun, Nov 24, 2024 at 12:40 PM Chris Packham
<Chris.Packham@alliedtelesis.co.nz> wrote:
>
> Hi Rong,
>
> On 22/11/24 18:02, Rong Xu wrote:
> > Does the following patch fix your issue?
> > https://lore.kernel.org/lkml/CAK7LNAQSpM2hh2=Wgribb92nHF5_peRb+gxdO1Bt1MkeL8N+hw@mail.gmail.com/t/
> >
> > Masahiro Yamada already applied the patch to kbuild-tree.
> >
> > I also proposed a similar fix as yours. But we believe the above patch
> > should also work (and probably is better).
> >
> > Please try and let us know.
>
> Looks like linux-next already has that patch and no it doesn't fix my
> issue. For the record here's the start of my System.map
>
> $ head System.map
> ffffffff80100000 T __split_text_end
> ffffffff80100000 T __split_text_start
> ffffffff80100000 T __unlikely_text_start
> ffffffff80100000 T _stext
> ffffffff80100000 T _text
> ffffffff80100000 t run_init_process
> ffffffff80100138 t try_to_run_init_process
> ffffffff8010018c t put_page
> ffffffff801001d0 t nr_blocks.isra.0
> ffffffff80100208 T ack_bad_irq
>
> It does look like whatever is supposed to put head.o first isn't kicking
> in for me.
>
> >
> > Thanks!
> >
> > -Rong
> >
> > On Thu, Nov 21, 2024 at 6:29 PM Chris Packham
> > <chris.packham@alliedtelesis.co.nz> wrote:
> >>
> >> On 22/11/24 14:58, Chris Packham wrote:
> >>> Hi Rong,
> >>>
> >>> I was just trying some of my in-flight changes against linux-next and
> >>> found that my RTL9300* based board no longer boots (nothing on the
> >>> serial console after U-Boot hands off to the kernel).
> >>>
> >>> A bisect between Linus's tree and linux-next/master points to commit
> >>> d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output
> >>> section"). A build on d4a7f2510ff4^ boots successfully. That series is
> >>> well beyond my area of knowledge, there may be something that the MIPS
> >>> SoCs are doing (or not doing) that is being affected by the sections
> >>> moving.
> >> Based on the sparc change that accompanied this commit the following
> >> does seem to restore working order for me:
> >>
> >> diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
> >> index d575f945d422..e90b0f68290c 100644
> >> --- a/arch/mips/kernel/vmlinux.lds.S
> >> +++ b/arch/mips/kernel/vmlinux.lds.S
> >> @@ -62,6 +62,8 @@ SECTIONS
> >>           _text = .;      /* Text and read-only data */
> >>           _stext = .;
> >>           .text : {
> >> +               /* Match text section symbols in head.S first */
> >> +               *head.o(.text)
> >>                   TEXT_TEXT
> >>                   SCHED_TEXT
> >>                   LOCK_TEXT
> >>
> >> No idea if that's appropriate
> >>

Re: RTL9300 (mips) fails to boot with commit d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output section")

[Chris Packham]

On 26/11/24 07:48, Rong Xu wrote:
> Thanks for confirming this! It looks like we do need to put all
> symbols from head.o before any other symbols.
> To be clear, the patch I mentioned was not trying to do that -- it
> just makes sure the _stext is in the right places.
Looking at other architectures it may just be a matter of adding __HEAD 
to an appropriate place in arch/mips/kernel/head.S and HEAD_TEXT_SECTION 
to arch/mips/kernel/vmlinux.lds.S. It might even be possible to drop 
mips from scripts/head-object-list.txt.

> I'll draft a patch for this.

I can give any patch a quick test on my setup.

>
> On Sun, Nov 24, 2024 at 12:40 PM Chris Packham
> <Chris.Packham@alliedtelesis.co.nz> wrote:
>> Hi Rong,
>>
>> On 22/11/24 18:02, Rong Xu wrote:
>>> Does the following patch fix your issue?
>>> https://lore.kernel.org/lkml/CAK7LNAQSpM2hh2=Wgribb92nHF5_peRb+gxdO1Bt1MkeL8N+hw@mail.gmail.com/t/
>>>
>>> Masahiro Yamada already applied the patch to kbuild-tree.
>>>
>>> I also proposed a similar fix as yours. But we believe the above patch
>>> should also work (and probably is better).
>>>
>>> Please try and let us know.
>> Looks like linux-next already has that patch and no it doesn't fix my
>> issue. For the record here's the start of my System.map
>>
>> $ head System.map
>> ffffffff80100000 T __split_text_end
>> ffffffff80100000 T __split_text_start
>> ffffffff80100000 T __unlikely_text_start
>> ffffffff80100000 T _stext
>> ffffffff80100000 T _text
>> ffffffff80100000 t run_init_process
>> ffffffff80100138 t try_to_run_init_process
>> ffffffff8010018c t put_page
>> ffffffff801001d0 t nr_blocks.isra.0
>> ffffffff80100208 T ack_bad_irq
>>
>> It does look like whatever is supposed to put head.o first isn't kicking
>> in for me.
>>
>>> Thanks!
>>>
>>> -Rong
>>>
>>> On Thu, Nov 21, 2024 at 6:29 PM Chris Packham
>>> <chris.packham@alliedtelesis.co.nz> wrote:
>>>> On 22/11/24 14:58, Chris Packham wrote:
>>>>> Hi Rong,
>>>>>
>>>>> I was just trying some of my in-flight changes against linux-next and
>>>>> found that my RTL9300* based board no longer boots (nothing on the
>>>>> serial console after U-Boot hands off to the kernel).
>>>>>
>>>>> A bisect between Linus's tree and linux-next/master points to commit
>>>>> d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output
>>>>> section"). A build on d4a7f2510ff4^ boots successfully. That series is
>>>>> well beyond my area of knowledge, there may be something that the MIPS
>>>>> SoCs are doing (or not doing) that is being affected by the sections
>>>>> moving.
>>>> Based on the sparc change that accompanied this commit the following
>>>> does seem to restore working order for me:
>>>>
>>>> diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
>>>> index d575f945d422..e90b0f68290c 100644
>>>> --- a/arch/mips/kernel/vmlinux.lds.S
>>>> +++ b/arch/mips/kernel/vmlinux.lds.S
>>>> @@ -62,6 +62,8 @@ SECTIONS
>>>>            _text = .;      /* Text and read-only data */
>>>>            _stext = .;
>>>>            .text : {
>>>> +               /* Match text section symbols in head.S first */
>>>> +               *head.o(.text)
>>>>                    TEXT_TEXT
>>>>                    SCHED_TEXT
>>>>                    LOCK_TEXT
>>>>
>>>> No idea if that's appropriate
>>>>

Re: RTL9300 (mips) fails to boot with commit d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output section")

[Rong Xu]

I sent out a patch using the __HEAD method. Compared to the
"*head.o(.text)", this only places
__kernel_entry at the beginning of the text section, and leaves others
reordered by the linker script.
I think it should work. Chris: Could you test it?

Thanks,

-Rong


On Mon, Nov 25, 2024 at 11:26 AM Chris Packham
<chris.packham@alliedtelesis.co.nz> wrote:
>
>
> On 26/11/24 07:48, Rong Xu wrote:
> > Thanks for confirming this! It looks like we do need to put all
> > symbols from head.o before any other symbols.
> > To be clear, the patch I mentioned was not trying to do that -- it
> > just makes sure the _stext is in the right places.
> Looking at other architectures it may just be a matter of adding __HEAD
> to an appropriate place in arch/mips/kernel/head.S and HEAD_TEXT_SECTION
> to arch/mips/kernel/vmlinux.lds.S. It might even be possible to drop
> mips from scripts/head-object-list.txt.
>
> > I'll draft a patch for this.
>
> I can give any patch a quick test on my setup.
>
> >
> > On Sun, Nov 24, 2024 at 12:40 PM Chris Packham
> > <Chris.Packham@alliedtelesis.co.nz> wrote:
> >> Hi Rong,
> >>
> >> On 22/11/24 18:02, Rong Xu wrote:
> >>> Does the following patch fix your issue?
> >>> https://lore.kernel.org/lkml/CAK7LNAQSpM2hh2=Wgribb92nHF5_peRb+gxdO1Bt1MkeL8N+hw@mail.gmail.com/t/
> >>>
> >>> Masahiro Yamada already applied the patch to kbuild-tree.
> >>>
> >>> I also proposed a similar fix as yours. But we believe the above patch
> >>> should also work (and probably is better).
> >>>
> >>> Please try and let us know.
> >> Looks like linux-next already has that patch and no it doesn't fix my
> >> issue. For the record here's the start of my System.map
> >>
> >> $ head System.map
> >> ffffffff80100000 T __split_text_end
> >> ffffffff80100000 T __split_text_start
> >> ffffffff80100000 T __unlikely_text_start
> >> ffffffff80100000 T _stext
> >> ffffffff80100000 T _text
> >> ffffffff80100000 t run_init_process
> >> ffffffff80100138 t try_to_run_init_process
> >> ffffffff8010018c t put_page
> >> ffffffff801001d0 t nr_blocks.isra.0
> >> ffffffff80100208 T ack_bad_irq
> >>
> >> It does look like whatever is supposed to put head.o first isn't kicking
> >> in for me.
> >>
> >>> Thanks!
> >>>
> >>> -Rong
> >>>
> >>> On Thu, Nov 21, 2024 at 6:29 PM Chris Packham
> >>> <chris.packham@alliedtelesis.co.nz> wrote:
> >>>> On 22/11/24 14:58, Chris Packham wrote:
> >>>>> Hi Rong,
> >>>>>
> >>>>> I was just trying some of my in-flight changes against linux-next and
> >>>>> found that my RTL9300* based board no longer boots (nothing on the
> >>>>> serial console after U-Boot hands off to the kernel).
> >>>>>
> >>>>> A bisect between Linus's tree and linux-next/master points to commit
> >>>>> d4a7f2510ff4 ("vmlinux.lds.h: Adjust symbol ordering in text output
> >>>>> section"). A build on d4a7f2510ff4^ boots successfully. That series is
> >>>>> well beyond my area of knowledge, there may be something that the MIPS
> >>>>> SoCs are doing (or not doing) that is being affected by the sections
> >>>>> moving.
> >>>> Based on the sparc change that accompanied this commit the following
> >>>> does seem to restore working order for me:
> >>>>
> >>>> diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
> >>>> index d575f945d422..e90b0f68290c 100644
> >>>> --- a/arch/mips/kernel/vmlinux.lds.S
> >>>> +++ b/arch/mips/kernel/vmlinux.lds.S
> >>>> @@ -62,6 +62,8 @@ SECTIONS
> >>>>            _text = .;      /* Text and read-only data */
> >>>>            _stext = .;
> >>>>            .text : {
> >>>> +               /* Match text section symbols in head.S first */
> >>>> +               *head.o(.text)
> >>>>                    TEXT_TEXT
> >>>>                    SCHED_TEXT
> >>>>                    LOCK_TEXT
> >>>>
> >>>> No idea if that's appropriate
> >>>>