[v1] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

[PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by Boqun Feng 1 month, 3 weeks ago

Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
is disabled") fixed a build error by providing rust helpers when
CONFIG_PCI_MSI=n. However the rust helpers rely on the
pci_alloc_irq_vectors() function is defined, which is not true when
CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
could be just remove the calling of pci_alloc_irq_vectors() since it's
empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
more reasonable fix is to define pci_alloc_irq_vectors() when
CONFIG_PCI=n and this aligns with the situations of other primitives as
well.

Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
---
I hit a build error without this:

../rust/helpers/pci.c:36:2: error: call to undeclared function 'pci_free_irq_vectors'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
   36 |         pci_free_irq_vectors(dev);
      |         ^
../rust/helpers/pci.c:36:2: note: did you mean 'pci_alloc_irq_vectors'?
../include/linux/pci.h:2208:1: note: 'pci_alloc_irq_vectors' declared here
 2208 | pci_alloc_irq_vectors(struct pci_dev *dev, unsigned int min_vecs,
      | ^
1 error generated.

when ./tools/testing/kunit/kunit.py run --make_options LLVM=1 --arch arm64 --kconfig_add CONFIG_RUST=y  rust_doctests_kernel

 include/linux/pci.h | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/include/linux/pci.h b/include/linux/pci.h
index 864775651c6f..b5cc0c2b9906 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -2210,6 +2210,10 @@ pci_alloc_irq_vectors(struct pci_dev *dev, unsigned int min_vecs,
 {
 	return -ENOSPC;
 }
+
+static inline void pci_free_irq_vectors(struct pci_dev *dev)
+{
+}
 #endif /* CONFIG_PCI */
 
 /* Include architecture-dependent settings and functions */
-- 
2.51.0

Re: [PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by Miguel Ojeda 1 month, 2 weeks ago

On Mon, Dec 15, 2025 at 3:54 AM Boqun Feng <boqun.feng@gmail.com> wrote:
>
> Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
> is disabled") fixed a build error by providing rust helpers when
> CONFIG_PCI_MSI=n. However the rust helpers rely on the
> pci_alloc_irq_vectors() function is defined, which is not true when
> CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
> could be just remove the calling of pci_alloc_irq_vectors() since it's
> empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
> pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
> more reasonable fix is to define pci_alloc_irq_vectors() when
> CONFIG_PCI=n and this aligns with the situations of other primitives as
> well.
>
> Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>

Related: https://lore.kernel.org/rust-for-linux/20251209014312.575940-1-fujita.tomonori@gmail.com/

I guess it counts as a report, so we may want a Reported-by (Cc'ing Tomo).

Cheers,
Miguel

Re: [PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by FUJITA Tomonori 1 month, 2 weeks ago

On Thu, 25 Dec 2025 10:10:04 +0100
Miguel Ojeda <miguel.ojeda.sandonis@gmail.com> wrote:

> On Mon, Dec 15, 2025 at 3:54 AM Boqun Feng <boqun.feng@gmail.com> wrote:
>>
>> Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
>> is disabled") fixed a build error by providing rust helpers when
>> CONFIG_PCI_MSI=n. However the rust helpers rely on the
>> pci_alloc_irq_vectors() function is defined, which is not true when
>> CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
>> could be just remove the calling of pci_alloc_irq_vectors() since it's
>> empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
>> pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
>> more reasonable fix is to define pci_alloc_irq_vectors() when
>> CONFIG_PCI=n and this aligns with the situations of other primitives as
>> well.
>>
>> Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
>> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> 
> Related: https://lore.kernel.org/rust-for-linux/20251209014312.575940-1-fujita.tomonori@gmail.com/
> 
> I guess it counts as a report, so we may want a Reported-by (Cc'ing Tomo).

Since pci.rs is only compiled when CONFIG_PCI is enabled. So it seems
consistent to treat the PCI helpers the same way. That said, this
approach is also fine by me: it's already inconsistent that
pci_alloc_irq_vectors() has a stuf definition, while
pci_free_irq_vectors does not.

Re: [PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by Boqun Feng 1 month, 2 weeks ago

On Thu, Dec 25, 2025 at 06:36:31PM +0900, FUJITA Tomonori wrote:
> On Thu, 25 Dec 2025 10:10:04 +0100
> Miguel Ojeda <miguel.ojeda.sandonis@gmail.com> wrote:
> 
> > On Mon, Dec 15, 2025 at 3:54 AM Boqun Feng <boqun.feng@gmail.com> wrote:
> >>
> >> Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
> >> is disabled") fixed a build error by providing rust helpers when
> >> CONFIG_PCI_MSI=n. However the rust helpers rely on the
> >> pci_alloc_irq_vectors() function is defined, which is not true when
> >> CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
> >> could be just remove the calling of pci_alloc_irq_vectors() since it's
> >> empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
> >> pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
> >> more reasonable fix is to define pci_alloc_irq_vectors() when
> >> CONFIG_PCI=n and this aligns with the situations of other primitives as
> >> well.
> >>
> >> Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
> >> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> > 
> > Related: https://lore.kernel.org/rust-for-linux/20251209014312.575940-1-fujita.tomonori@gmail.com/
> > 
> > I guess it counts as a report, so we may want a Reported-by (Cc'ing Tomo).
> 
> Since pci.rs is only compiled when CONFIG_PCI is enabled. So it seems
> consistent to treat the PCI helpers the same way. That said, this
> approach is also fine by me: it's already inconsistent that
> pci_alloc_irq_vectors() has a stuf definition, while
> pci_free_irq_vectors does not.

Yes, I think providing a stub pci_free_irq_vectors() is the way to go.
If you are OK with it, I could send a v2 with your and Liang Jie's
Reported-by. Alternatively, since you're the first one sending the fix,
and I would have replied that with the suggestion if I didn't miss that,
feel free to send a v2 (providing the stub pci_free_irq_vectors()) from
you ;-)

Regards,
Boqun

Re: [PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by FUJITA Tomonori 1 month, 1 week ago

On Thu, 25 Dec 2025 18:19:08 +0800
Boqun Feng <boqun.feng@gmail.com> wrote:

> On Thu, Dec 25, 2025 at 06:36:31PM +0900, FUJITA Tomonori wrote:
>> On Thu, 25 Dec 2025 10:10:04 +0100
>> Miguel Ojeda <miguel.ojeda.sandonis@gmail.com> wrote:
>> 
>> > On Mon, Dec 15, 2025 at 3:54 AM Boqun Feng <boqun.feng@gmail.com> wrote:
>> >>
>> >> Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
>> >> is disabled") fixed a build error by providing rust helpers when
>> >> CONFIG_PCI_MSI=n. However the rust helpers rely on the
>> >> pci_alloc_irq_vectors() function is defined, which is not true when
>> >> CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
>> >> could be just remove the calling of pci_alloc_irq_vectors() since it's
>> >> empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
>> >> pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
>> >> more reasonable fix is to define pci_alloc_irq_vectors() when
>> >> CONFIG_PCI=n and this aligns with the situations of other primitives as
>> >> well.
>> >>
>> >> Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
>> >> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
>> > 
>> > Related: https://lore.kernel.org/rust-for-linux/20251209014312.575940-1-fujita.tomonori@gmail.com/
>> > 
>> > I guess it counts as a report, so we may want a Reported-by (Cc'ing Tomo).
>> 
>> Since pci.rs is only compiled when CONFIG_PCI is enabled. So it seems
>> consistent to treat the PCI helpers the same way. That said, this
>> approach is also fine by me: it's already inconsistent that
>> pci_alloc_irq_vectors() has a stuf definition, while
>> pci_free_irq_vectors does not.
> 
> Yes, I think providing a stub pci_free_irq_vectors() is the way to go.
> If you are OK with it, I could send a v2 with your and Liang Jie's
> Reported-by.

It works for me. Please go ahead.

Re: [PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by Boqun Feng 1 month, 2 weeks ago

On Thu, Dec 25, 2025 at 10:10:04AM +0100, Miguel Ojeda wrote:
> On Mon, Dec 15, 2025 at 3:54 AM Boqun Feng <boqun.feng@gmail.com> wrote:
> >
> > Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
> > is disabled") fixed a build error by providing rust helpers when
> > CONFIG_PCI_MSI=n. However the rust helpers rely on the
> > pci_alloc_irq_vectors() function is defined, which is not true when
> > CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
> > could be just remove the calling of pci_alloc_irq_vectors() since it's
> > empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
> > pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
> > more reasonable fix is to define pci_alloc_irq_vectors() when
> > CONFIG_PCI=n and this aligns with the situations of other primitives as
> > well.
> >
> > Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
> > Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> 
> Related: https://lore.kernel.org/rust-for-linux/20251209014312.575940-1-fujita.tomonori@gmail.com/
> 

Thanks! I was missing that, and I should have replied Tomo's patch.

> I guess it counts as a report, so we may want a Reported-by (Cc'ing Tomo).
> 

Tomo, how do you feel about two fix approaches? I think providing dummy
pci_free_irq_vectors() is better because it's easier and aligned with
other pci_*() APIs.

Regards,
Boqun

> Cheers,
> Miguel

Re: [PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by Joel Fernandes 1 month, 2 weeks ago


On 12/14/2025 9:54 PM, Boqun Feng wrote:
> Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
> is disabled") fixed a build error by providing rust helpers when
> CONFIG_PCI_MSI=n. However the rust helpers rely on the
> pci_alloc_irq_vectors() function is defined, which is not true when
> CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
> could be just remove the calling of pci_alloc_irq_vectors() since it's
> empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
> pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
> more reasonable fix is to define pci_alloc_irq_vectors() when
> CONFIG_PCI=n and this aligns with the situations of other primitives as
> well.
> 
> Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Joel Fernandes <joelagnelf@nvidia.com>

Re: [PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by David Gow 1 month, 2 weeks ago

On Mon, 15 Dec 2025 at 10:55, Boqun Feng <boqun.feng@gmail.com> wrote:
>
> Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
> is disabled") fixed a build error by providing rust helpers when
> CONFIG_PCI_MSI=n. However the rust helpers rely on the
> pci_alloc_irq_vectors() function is defined, which is not true when
> CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
> could be just remove the calling of pci_alloc_irq_vectors() since it's
> empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
> pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
> more reasonable fix is to define pci_alloc_irq_vectors() when
> CONFIG_PCI=n and this aligns with the situations of other primitives as
> well.
>
> Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> ---

Thanks very much!

Reviewed-by: David Gow <davidgow@google.com>

Cheers,
-- David

Re: [PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by Drew Fustini 1 month, 3 weeks ago

On Mon, Dec 15, 2025 at 11:54:44AM +0900, Boqun Feng wrote:
> Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
> is disabled") fixed a build error by providing rust helpers when
> CONFIG_PCI_MSI=n. However the rust helpers rely on the
> pci_alloc_irq_vectors() function is defined, which is not true when
> CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
> could be just remove the calling of pci_alloc_irq_vectors() since it's
> empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
> pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
> more reasonable fix is to define pci_alloc_irq_vectors() when
> CONFIG_PCI=n and this aligns with the situations of other primitives as
> well.
> 
> Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
> ---
> I hit a build error without this:
> 
> ../rust/helpers/pci.c:36:2: error: call to undeclared function 'pci_free_irq_vectors'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
>    36 |         pci_free_irq_vectors(dev);
>       |         ^
> ../rust/helpers/pci.c:36:2: note: did you mean 'pci_alloc_irq_vectors'?
> ../include/linux/pci.h:2208:1: note: 'pci_alloc_irq_vectors' declared here
>  2208 | pci_alloc_irq_vectors(struct pci_dev *dev, unsigned int min_vecs,
>       | ^
> 1 error generated.
> 
> when ./tools/testing/kunit/kunit.py run --make_options LLVM=1 --arch arm64 --kconfig_add CONFIG_RUST=y  rust_doctests_kernel
> 
>  include/linux/pci.h | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index 864775651c6f..b5cc0c2b9906 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -2210,6 +2210,10 @@ pci_alloc_irq_vectors(struct pci_dev *dev, unsigned int min_vecs,
>  {
>  	return -ENOSPC;
>  }
> +
> +static inline void pci_free_irq_vectors(struct pci_dev *dev)
> +{
> +}
>  #endif /* CONFIG_PCI */
>  
>  /* Include architecture-dependent settings and functions */
> -- 
> 2.51.0
> 

Reviewed-by: Drew Fustini <fustini@kernel.org>

Thanks for the fix. I ran into this when building next-20251217 for a
machine without PCIe and was happy to find this on lore.

Drew

Re: [PATCH] PCI: Provide pci_free_irq_vectors() for CONFIG_PCI=n

Posted by Danilo Krummrich 1 month, 3 weeks ago

On 12/15/25 3:54 AM, Boqun Feng wrote:
> Commit 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI
> is disabled") fixed a build error by providing rust helpers when
> CONFIG_PCI_MSI=n. However the rust helpers rely on the
> pci_alloc_irq_vectors() function is defined, which is not true when
> CONFIG_PCI=n. There are multiple ways to fix this, e.g. a possible fix
> could be just remove the calling of pci_alloc_irq_vectors() since it's
> empty when CONFIG_PCI_MSI=n anyway. However, since PCI irq APIs, such as
> pci_alloc_irq_vectors(), are already defined even when CONFIG_PCI=n, the
> more reasonable fix is to define pci_alloc_irq_vectors() when
> CONFIG_PCI=n and this aligns with the situations of other primitives as
> well.
> 
> Fixes: 473b9f331718 ("rust: pci: fix build failure when CONFIG_PCI_MSI is disabled")
> Signed-off-by: Boqun Feng <boqun.feng@gmail.com>

Reviewed-by: Danilo Krummrich <dakr@kernel.org>

Bjorn, I assume you will pick this up? Otherwise I can pick it up as well, I
will likely send a -fixes PR for -rc2 anyways.