Re: [xen-tip:linux-next 12/12] WARNING: modpost: vmlinux: section mismatch in reference: mc_debug_data+0x0 (section: .data) -> mc_debug_data_early (section: .init.data)

Re: [xen-tip:linux-next 12/12] WARNING: modpost: vmlinux: section mismatch in reference: mc_debug_data+0x0 (section: .data) -> mc_debug_data_early (section: .init.data)

[Borislav Petkov]

On Wed, Jul 24, 2024 at 11:55:39AM +0200, Jürgen Groß wrote:
> I'd prefer a general way to handle this problem, like e.g. some kind of
> __refdata tagging for percpu variables.

Any reason for not doing the trivial thing?

diff --git a/arch/x86/xen/multicalls.c b/arch/x86/xen/multicalls.c
index 10c660fae8b3..100bfaba3f4d 100644
--- a/arch/x86/xen/multicalls.c
+++ b/arch/x86/xen/multicalls.c
@@ -53,7 +53,7 @@ struct mc_debug_data {
 };
 
 static DEFINE_PER_CPU(struct mc_buffer, mc_buffer);
-static struct mc_debug_data mc_debug_data_early __initdata;
+static struct mc_debug_data mc_debug_data_early;
 static DEFINE_PER_CPU(struct mc_debug_data *, mc_debug_data) =
 	&mc_debug_data_early;
 static struct mc_debug_data __percpu *mc_debug_data_ptr;

---

since this breaks randbuilds:

WARNING: modpost: vmlinux: section mismatch in reference: mc_debug_data+0x0 (section: .data) -> mc_debug_data_early (section: .init.data)
ERROR: modpost: Section mismatches detected.
Set CONFIG_SECTION_MISMATCH_WARN_ONLY=y to allow them.
make[2]: *** [scripts/Makefile.modpost:147: Module.symvers] Error 1
make[2]: *** Deleting file 'Module.symvers'
make[1]: *** [/mnt/kernel/kernel/linux/Makefile:1947: modpost] Error 2
make: *** [Makefile:251: __sub-make] Error 2

?

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

Re: [xen-tip:linux-next 12/12] WARNING: modpost: vmlinux: section mismatch in reference: mc_debug_data+0x0 (section: .data) -> mc_debug_data_early (section: .init.data)

[Jürgen Groß]

[-- Attachment #1.1.1: Type: text/plain, Size: 965 bytes --]

On 27.03.25 15:13, Borislav Petkov wrote:
> On Wed, Jul 24, 2024 at 11:55:39AM +0200, Jürgen Groß wrote:
>> I'd prefer a general way to handle this problem, like e.g. some kind of
>> __refdata tagging for percpu variables.
> 
> Any reason for not doing the trivial thing?
> 
> diff --git a/arch/x86/xen/multicalls.c b/arch/x86/xen/multicalls.c
> index 10c660fae8b3..100bfaba3f4d 100644
> --- a/arch/x86/xen/multicalls.c
> +++ b/arch/x86/xen/multicalls.c
> @@ -53,7 +53,7 @@ struct mc_debug_data {
>   };
>   
>   static DEFINE_PER_CPU(struct mc_buffer, mc_buffer);
> -static struct mc_debug_data mc_debug_data_early __initdata;
> +static struct mc_debug_data mc_debug_data_early;
>   static DEFINE_PER_CPU(struct mc_debug_data *, mc_debug_data) =
>   	&mc_debug_data_early;
>   static struct mc_debug_data __percpu *mc_debug_data_ptr;

Well, that is wasting nearly 3kB of the data section.

Maybe not a big deal, but still...


Juergen

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3743 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 495 bytes --]

Re: [xen-tip:linux-next 12/12] WARNING: modpost: vmlinux: section mismatch in reference: mc_debug_data+0x0 (section: .data) -> mc_debug_data_early (section: .init.data)

[Borislav Petkov]

On Thu, Mar 27, 2025 at 03:21:45PM +0100, Jürgen Groß wrote:
> Well, that is wasting nearly 3kB of the data section.
> 
> Maybe not a big deal, but still...

We could do it until the proper fix is in place, no?

3K is meh, especially for the hypervisor kernel, I'd say...

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

Re: [xen-tip:linux-next 12/12] WARNING: modpost: vmlinux: section mismatch in reference: mc_debug_data+0x0 (section: .data) -> mc_debug_data_early (section: .init.data)

[Jürgen Groß]

[-- Attachment #1.1.1: Type: text/plain, Size: 825 bytes --]

On 27.03.25 15:40, Borislav Petkov wrote:
> On Thu, Mar 27, 2025 at 03:21:45PM +0100, Jürgen Groß wrote:
>> Well, that is wasting nearly 3kB of the data section.
>>
>> Maybe not a big deal, but still...
> 
> We could do it until the proper fix is in place, no?
> 
> 3K is meh, especially for the hypervisor kernel, I'd say...
> 

Yeah, that was my thinking.

Another approach could be to have:

-static DEFINE_PER_CPU(struct mc_debug_data *, mc_debug_data) =
-	&mc_debug_data_early;
+static DEFINE_PER_CPU(struct mc_debug_data *, mc_debug_data);

and to use an inline access function returning &mc_debug_data_early
if the percpu variable is NULL. This access function could be __ref.

It is a debug feature after all, so having a few additional instructions
isn't the end of the world.


Juergen

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3743 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 495 bytes --]

Re: [xen-tip:linux-next 12/12] WARNING: modpost: vmlinux: section mismatch in reference: mc_debug_data+0x0 (section: .data) -> mc_debug_data_early (section: .init.data)

[Borislav Petkov]

On Thu, Mar 27, 2025 at 04:15:11PM +0100, Jürgen Groß wrote:
> Another approach could be to have:
> 
> -static DEFINE_PER_CPU(struct mc_debug_data *, mc_debug_data) =
> -	&mc_debug_data_early;
> +static DEFINE_PER_CPU(struct mc_debug_data *, mc_debug_data);
> 
> and to use an inline access function returning &mc_debug_data_early
> if the percpu variable is NULL. This access function could be __ref.
> 
> It is a debug feature after all, so having a few additional instructions
> isn't the end of the world.

Yap, makes sense to me too.

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette