If CONFIG_PCI=n, and gcc (e.g. 4.1.2) decides not to inline
get_pci_function_alias_group(), the build fails with:
drivers/iommu/iommu.o: In function `get_pci_function_alias_group':
iommu.c:(.text+0xfdc): undefined reference to `pci_acs_enabled'
Due to the various dummies for PCI calls in the CONFIG_PCI=n case,
pci_acs_enabled() isn't actually ever called, but not all versions of
gcc are smart enough to realize that.
While explicitly marking get_pci_function_alias_group() inline would fix
the build, this would inflate the code for the CONFIG_PCI=y case, as
get_pci_function_alias_group() is a not-so-small function called from
two places.
Hence fix the issue by introducing a dummy for pci_acs_enabled()
instead.
Signed-off-by: Geert Uytterhoeven <[email protected]>
---
include/linux/pci.h | 2 ++
1 file changed, 2 insertions(+)
diff --git a/include/linux/pci.h b/include/linux/pci.h
index f68c58a93dd045b9..f4f8ee5a7362e982 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -1685,6 +1685,8 @@ static inline int pci_get_new_domain_nr(void) { return -ENOSYS; }
#define dev_is_pci(d) (false)
#define dev_is_pf(d) (false)
+static inline bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
+{ return false; }
#endif /* CONFIG_PCI */
/* Include architecture-dependent settings and functions */
--
2.7.4
Hey Bjorn,
On Mon, Sep 11, 2017 at 02:29:15PM +0200, Geert Uytterhoeven wrote:
> If CONFIG_PCI=n, and gcc (e.g. 4.1.2) decides not to inline
> get_pci_function_alias_group(), the build fails with:
>
> drivers/iommu/iommu.o: In function `get_pci_function_alias_group':
> iommu.c:(.text+0xfdc): undefined reference to `pci_acs_enabled'
>
> Due to the various dummies for PCI calls in the CONFIG_PCI=n case,
> pci_acs_enabled() isn't actually ever called, but not all versions of
> gcc are smart enough to realize that.
>
> While explicitly marking get_pci_function_alias_group() inline would fix
> the build, this would inflate the code for the CONFIG_PCI=y case, as
> get_pci_function_alias_group() is a not-so-small function called from
> two places.
>
> Hence fix the issue by introducing a dummy for pci_acs_enabled()
> instead.
>
> Signed-off-by: Geert Uytterhoeven <[email protected]>
> ---
> include/linux/pci.h | 2 ++
> 1 file changed, 2 insertions(+)
I am sending iommu-fixes upstream this week, and this would fit into it.
Can you have a look please and ack it if you are okay with that?
Thanks,
Joerg
>
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index f68c58a93dd045b9..f4f8ee5a7362e982 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -1685,6 +1685,8 @@ static inline int pci_get_new_domain_nr(void) { return -ENOSYS; }
>
> #define dev_is_pci(d) (false)
> #define dev_is_pf(d) (false)
> +static inline bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
> +{ return false; }
> #endif /* CONFIG_PCI */
>
> /* Include architecture-dependent settings and functions */
> --
> 2.7.4
On Mon, 11 Sep 2017 14:29:15 +0200
Geert Uytterhoeven <[email protected]> wrote:
> If CONFIG_PCI=n, and gcc (e.g. 4.1.2) decides not to inline
> get_pci_function_alias_group(), the build fails with:
>
> drivers/iommu/iommu.o: In function `get_pci_function_alias_group':
> iommu.c:(.text+0xfdc): undefined reference to `pci_acs_enabled'
>
> Due to the various dummies for PCI calls in the CONFIG_PCI=n case,
> pci_acs_enabled() isn't actually ever called, but not all versions of
> gcc are smart enough to realize that.
>
> While explicitly marking get_pci_function_alias_group() inline would fix
> the build, this would inflate the code for the CONFIG_PCI=y case, as
> get_pci_function_alias_group() is a not-so-small function called from
> two places.
>
> Hence fix the issue by introducing a dummy for pci_acs_enabled()
> instead.
>
> Signed-off-by: Geert Uytterhoeven <[email protected]>
> ---
> include/linux/pci.h | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index f68c58a93dd045b9..f4f8ee5a7362e982 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -1685,6 +1685,8 @@ static inline int pci_get_new_domain_nr(void) { return -ENOSYS; }
>
> #define dev_is_pci(d) (false)
> #define dev_is_pf(d) (false)
> +static inline bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
> +{ return false; }
> #endif /* CONFIG_PCI */
>
> /* Include architecture-dependent settings and functions */
Reviewed-by: Alex Williamson <[email protected]>
On Mon, Sep 11, 2017 at 02:29:15PM +0200, Geert Uytterhoeven wrote:
> If CONFIG_PCI=n, and gcc (e.g. 4.1.2) decides not to inline
> get_pci_function_alias_group(), the build fails with:
>
> drivers/iommu/iommu.o: In function `get_pci_function_alias_group':
> iommu.c:(.text+0xfdc): undefined reference to `pci_acs_enabled'
>
> Due to the various dummies for PCI calls in the CONFIG_PCI=n case,
> pci_acs_enabled() isn't actually ever called, but not all versions of
> gcc are smart enough to realize that.
>
> While explicitly marking get_pci_function_alias_group() inline would fix
> the build, this would inflate the code for the CONFIG_PCI=y case, as
> get_pci_function_alias_group() is a not-so-small function called from
> two places.
>
> Hence fix the issue by introducing a dummy for pci_acs_enabled()
> instead.
>
> Signed-off-by: Geert Uytterhoeven <[email protected]>
Acked-by: Bjorn Helgaas <[email protected]>
Joerg, if you pick this up, would you mind capitalizing the subject
line to match the PCI convention, e.g.,
PCI: Add dummy pci_acs_enabled() for CONFIG_PCI=n build
If it's too late for you to pick this up this week, I can include it
next week. I assume this is not related to a specific change, i.e.,
it's not a regression? Should it be marked for stable?
> ---
> include/linux/pci.h | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index f68c58a93dd045b9..f4f8ee5a7362e982 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -1685,6 +1685,8 @@ static inline int pci_get_new_domain_nr(void) { return -ENOSYS; }
>
> #define dev_is_pci(d) (false)
> #define dev_is_pf(d) (false)
> +static inline bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
> +{ return false; }
> #endif /* CONFIG_PCI */
>
> /* Include architecture-dependent settings and functions */
> --
> 2.7.4
>
Hey Bjorn,
On Fri, Sep 22, 2017 at 10:56:05AM -0500, Bjorn Helgaas wrote:
> Joerg, if you pick this up, would you mind capitalizing the subject
> line to match the PCI convention, e.g.,
>
> PCI: Add dummy pci_acs_enabled() for CONFIG_PCI=n build
>
> If it's too late for you to pick this up this week, I can include it
> next week. I assume this is not related to a specific change, i.e.,
> it's not a regression? Should it be marked for stable?
I sent iommu fixes out today, this was not included. So if you are
sending fixes upstream next week anyway, please pick it up. I am not
sure if I have anything to send next week, so you are probably faster.
I think its stable material too.
Thanks,
Joerg
Hi Björn,
On Fri, Sep 22, 2017 at 5:56 PM, Bjorn Helgaas <[email protected]> wrote:
> On Mon, Sep 11, 2017 at 02:29:15PM +0200, Geert Uytterhoeven wrote:
>> If CONFIG_PCI=n, and gcc (e.g. 4.1.2) decides not to inline
>> get_pci_function_alias_group(), the build fails with:
>>
>> drivers/iommu/iommu.o: In function `get_pci_function_alias_group':
>> iommu.c:(.text+0xfdc): undefined reference to `pci_acs_enabled'
>>
>> Due to the various dummies for PCI calls in the CONFIG_PCI=n case,
>> pci_acs_enabled() isn't actually ever called, but not all versions of
>> gcc are smart enough to realize that.
>>
>> While explicitly marking get_pci_function_alias_group() inline would fix
>> the build, this would inflate the code for the CONFIG_PCI=y case, as
>> get_pci_function_alias_group() is a not-so-small function called from
>> two places.
>>
>> Hence fix the issue by introducing a dummy for pci_acs_enabled()
>> instead.
>>
>> Signed-off-by: Geert Uytterhoeven <[email protected]>
>
> Acked-by: Bjorn Helgaas <[email protected]>
>
> Joerg, if you pick this up, would you mind capitalizing the subject
> line to match the PCI convention, e.g.,
>
> PCI: Add dummy pci_acs_enabled() for CONFIG_PCI=n build
>
> If it's too late for you to pick this up this week, I can include it
> next week. I assume this is not related to a specific change, i.e.,
> it's not a regression? Should it be marked for stable?
It was introduced by commit 0ae349a0f33fb040 ("iommu/qcom: Add qcom_iommu"),
which enabled IOMMU support for compile-testing in e.g. allmodconfig on
platforms that don't have IOMMUs.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
On Fri, Sep 22, 2017 at 08:12:46PM +0200, Geert Uytterhoeven wrote:
> Hi Bj?rn,
>
> On Fri, Sep 22, 2017 at 5:56 PM, Bjorn Helgaas <[email protected]> wrote:
> > On Mon, Sep 11, 2017 at 02:29:15PM +0200, Geert Uytterhoeven wrote:
> >> If CONFIG_PCI=n, and gcc (e.g. 4.1.2) decides not to inline
> >> get_pci_function_alias_group(), the build fails with:
> >>
> >> drivers/iommu/iommu.o: In function `get_pci_function_alias_group':
> >> iommu.c:(.text+0xfdc): undefined reference to `pci_acs_enabled'
> >>
> >> Due to the various dummies for PCI calls in the CONFIG_PCI=n case,
> >> pci_acs_enabled() isn't actually ever called, but not all versions of
> >> gcc are smart enough to realize that.
> >>
> >> While explicitly marking get_pci_function_alias_group() inline would fix
> >> the build, this would inflate the code for the CONFIG_PCI=y case, as
> >> get_pci_function_alias_group() is a not-so-small function called from
> >> two places.
> >>
> >> Hence fix the issue by introducing a dummy for pci_acs_enabled()
> >> instead.
> >>
> >> Signed-off-by: Geert Uytterhoeven <[email protected]>
> >
> > Acked-by: Bjorn Helgaas <[email protected]>
> >
> > Joerg, if you pick this up, would you mind capitalizing the subject
> > line to match the PCI convention, e.g.,
> >
> > PCI: Add dummy pci_acs_enabled() for CONFIG_PCI=n build
> >
> > If it's too late for you to pick this up this week, I can include it
> > next week. I assume this is not related to a specific change, i.e.,
> > it's not a regression? Should it be marked for stable?
>
> It was introduced by commit 0ae349a0f33fb040 ("iommu/qcom: Add qcom_iommu"),
> which enabled IOMMU support for compile-testing in e.g. allmodconfig on
> platforms that don't have IOMMUs.
I put this on my for-linus branch for v4.14 with Alex's reviewed-by.
Since 0ae349a0f33f was merged in the v4.14 merge window (it first
appeared in v4.14-rc1), I don't think we need to mark it for stable.
Bjorn