KernelCI reports that bcm2835_defconfig is no longer booting since
commit ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING
forcibly") (https://lkml.org/lkml/2019/9/26/825).
I also received a regression report from Nicolas Saenz Julienne
(https://lkml.org/lkml/2019/9/27/263).
This problem has cropped up on bcm2835_defconfig because it enables
CONFIG_CC_OPTIMIZE_FOR_SIZE. The compiler tends to prefer not inlining
functions with -Os. I was able to reproduce it with other boards and
defconfig files by manually enabling CONFIG_CC_OPTIMIZE_FOR_SIZE.
The __get_user_check() specifically uses r0, r1, r2 registers.
So, uaccess_save_and_enable() and uaccess_restore() must be inlined.
Otherwise, those register assignments would be entirely dropped,
according to my analysis of the disassembly.
Prior to commit 9012d011660e ("compiler: allow all arches to enable
CONFIG_OPTIMIZE_INLINING"), the 'inline' marker was always enough for
inlining functions, except on x86.
Since that commit, all architectures can enable CONFIG_OPTIMIZE_INLINING.
So, __always_inline is now the only guaranteed way of forcible inlining.
I also added __always_inline to 4 functions in the call-graph from the
__get_user_check() macro.
Fixes: 9012d011660e ("compiler: allow all arches to enable CONFIG_OPTIMIZE_INLINING")
Reported-by: "kernelci.org bot" <[email protected]>
Reported-by: Nicolas Saenz Julienne <[email protected]>
Signed-off-by: Masahiro Yamada <[email protected]>
---
Changes in v2:
- Use __always_inline instead of changing the function call places
(per Russell King)
- The previous submission is: https://lore.kernel.org/patchwork/patch/1132459/
arch/arm/include/asm/domain.h | 8 ++++----
arch/arm/include/asm/uaccess.h | 4 ++--
2 files changed, 6 insertions(+), 6 deletions(-)
diff --git a/arch/arm/include/asm/domain.h b/arch/arm/include/asm/domain.h
index 567dbede4785..f1d0a7807cd0 100644
--- a/arch/arm/include/asm/domain.h
+++ b/arch/arm/include/asm/domain.h
@@ -82,7 +82,7 @@
#ifndef __ASSEMBLY__
#ifdef CONFIG_CPU_CP15_MMU
-static inline unsigned int get_domain(void)
+static __always_inline unsigned int get_domain(void)
{
unsigned int domain;
@@ -94,7 +94,7 @@ static inline unsigned int get_domain(void)
return domain;
}
-static inline void set_domain(unsigned val)
+static __always_inline void set_domain(unsigned int val)
{
asm volatile(
"mcr p15, 0, %0, c3, c0 @ set domain"
@@ -102,12 +102,12 @@ static inline void set_domain(unsigned val)
isb();
}
#else
-static inline unsigned int get_domain(void)
+static __always_inline unsigned int get_domain(void)
{
return 0;
}
-static inline void set_domain(unsigned val)
+static __always_inline void set_domain(unsigned int val)
{
}
#endif
diff --git a/arch/arm/include/asm/uaccess.h b/arch/arm/include/asm/uaccess.h
index 303248e5b990..98c6b91be4a8 100644
--- a/arch/arm/include/asm/uaccess.h
+++ b/arch/arm/include/asm/uaccess.h
@@ -22,7 +22,7 @@
* perform such accesses (eg, via list poison values) which could then
* be exploited for priviledge escalation.
*/
-static inline unsigned int uaccess_save_and_enable(void)
+static __always_inline unsigned int uaccess_save_and_enable(void)
{
#ifdef CONFIG_CPU_SW_DOMAIN_PAN
unsigned int old_domain = get_domain();
@@ -37,7 +37,7 @@ static inline unsigned int uaccess_save_and_enable(void)
#endif
}
-static inline void uaccess_restore(unsigned int flags)
+static __always_inline void uaccess_restore(unsigned int flags)
{
#ifdef CONFIG_CPU_SW_DOMAIN_PAN
/* Restore the user access mask */
--
2.17.1
On Tue, 2019-10-01 at 17:37 +0900, Masahiro Yamada wrote:
> KernelCI reports that bcm2835_defconfig is no longer booting since
> commit ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING
> forcibly") (https://lkml.org/lkml/2019/9/26/825).
>
> I also received a regression report from Nicolas Saenz Julienne
> (https://lkml.org/lkml/2019/9/27/263).
>
> This problem has cropped up on bcm2835_defconfig because it enables
> CONFIG_CC_OPTIMIZE_FOR_SIZE. The compiler tends to prefer not inlining
> functions with -Os. I was able to reproduce it with other boards and
> defconfig files by manually enabling CONFIG_CC_OPTIMIZE_FOR_SIZE.
>
> The __get_user_check() specifically uses r0, r1, r2 registers.
> So, uaccess_save_and_enable() and uaccess_restore() must be inlined.
> Otherwise, those register assignments would be entirely dropped,
> according to my analysis of the disassembly.
>
> Prior to commit 9012d011660e ("compiler: allow all arches to enable
> CONFIG_OPTIMIZE_INLINING"), the 'inline' marker was always enough for
> inlining functions, except on x86.
>
> Since that commit, all architectures can enable CONFIG_OPTIMIZE_INLINING.
> So, __always_inline is now the only guaranteed way of forcible inlining.
>
> I also added __always_inline to 4 functions in the call-graph from the
> __get_user_check() macro.
>
> Fixes: 9012d011660e ("compiler: allow all arches to enable
> CONFIG_OPTIMIZE_INLINING")
> Reported-by: "kernelci.org bot" <[email protected]>
> Reported-by: Nicolas Saenz Julienne <[email protected]>
> Signed-off-by: Masahiro Yamada <[email protected]>
Tested-by: Nicolas Saenz Julienne <[email protected]>
On Tue, Oct 1, 2019 at 1:37 AM Masahiro Yamada
<[email protected]> wrote:
>
> KernelCI reports that bcm2835_defconfig is no longer booting since
> commit ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING
> forcibly") (https://lkml.org/lkml/2019/9/26/825).
>
> I also received a regression report from Nicolas Saenz Julienne
> (https://lkml.org/lkml/2019/9/27/263).
>
> This problem has cropped up on bcm2835_defconfig because it enables
> CONFIG_CC_OPTIMIZE_FOR_SIZE. The compiler tends to prefer not inlining
> functions with -Os. I was able to reproduce it with other boards and
> defconfig files by manually enabling CONFIG_CC_OPTIMIZE_FOR_SIZE.
>
> The __get_user_check() specifically uses r0, r1, r2 registers.
> So, uaccess_save_and_enable() and uaccess_restore() must be inlined.
> Otherwise, those register assignments would be entirely dropped,
> according to my analysis of the disassembly.
>
> Prior to commit 9012d011660e ("compiler: allow all arches to enable
> CONFIG_OPTIMIZE_INLINING"), the 'inline' marker was always enough for
> inlining functions, except on x86.
>
> Since that commit, all architectures can enable CONFIG_OPTIMIZE_INLINING.
> So, __always_inline is now the only guaranteed way of forcible inlining.
No, the C preprocessor is the only guaranteed way of inlining. I
preferred v1; if you're going to <strikethrough>play with
fire</strikethrough>write assembly, don't get burned.
>
> I also added __always_inline to 4 functions in the call-graph from the
> __get_user_check() macro.
>
> Fixes: 9012d011660e ("compiler: allow all arches to enable CONFIG_OPTIMIZE_INLINING")
> Reported-by: "kernelci.org bot" <[email protected]>
> Reported-by: Nicolas Saenz Julienne <[email protected]>
> Signed-off-by: Masahiro Yamada <[email protected]>
> ---
>
> Changes in v2:
> - Use __always_inline instead of changing the function call places
> (per Russell King)
> - The previous submission is: https://lore.kernel.org/patchwork/patch/1132459/
>
> arch/arm/include/asm/domain.h | 8 ++++----
> arch/arm/include/asm/uaccess.h | 4 ++--
> 2 files changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/arch/arm/include/asm/domain.h b/arch/arm/include/asm/domain.h
> index 567dbede4785..f1d0a7807cd0 100644
> --- a/arch/arm/include/asm/domain.h
> +++ b/arch/arm/include/asm/domain.h
> @@ -82,7 +82,7 @@
> #ifndef __ASSEMBLY__
>
> #ifdef CONFIG_CPU_CP15_MMU
> -static inline unsigned int get_domain(void)
> +static __always_inline unsigned int get_domain(void)
> {
> unsigned int domain;
>
> @@ -94,7 +94,7 @@ static inline unsigned int get_domain(void)
> return domain;
> }
>
> -static inline void set_domain(unsigned val)
> +static __always_inline void set_domain(unsigned int val)
> {
> asm volatile(
> "mcr p15, 0, %0, c3, c0 @ set domain"
> @@ -102,12 +102,12 @@ static inline void set_domain(unsigned val)
> isb();
> }
> #else
> -static inline unsigned int get_domain(void)
> +static __always_inline unsigned int get_domain(void)
> {
> return 0;
> }
>
> -static inline void set_domain(unsigned val)
> +static __always_inline void set_domain(unsigned int val)
> {
> }
> #endif
> diff --git a/arch/arm/include/asm/uaccess.h b/arch/arm/include/asm/uaccess.h
> index 303248e5b990..98c6b91be4a8 100644
> --- a/arch/arm/include/asm/uaccess.h
> +++ b/arch/arm/include/asm/uaccess.h
> @@ -22,7 +22,7 @@
> * perform such accesses (eg, via list poison values) which could then
> * be exploited for priviledge escalation.
> */
> -static inline unsigned int uaccess_save_and_enable(void)
> +static __always_inline unsigned int uaccess_save_and_enable(void)
> {
> #ifdef CONFIG_CPU_SW_DOMAIN_PAN
> unsigned int old_domain = get_domain();
> @@ -37,7 +37,7 @@ static inline unsigned int uaccess_save_and_enable(void)
> #endif
> }
>
> -static inline void uaccess_restore(unsigned int flags)
> +static __always_inline void uaccess_restore(unsigned int flags)
> {
> #ifdef CONFIG_CPU_SW_DOMAIN_PAN
> /* Restore the user access mask */
> --
> 2.17.1
>
--
Thanks,
~Nick Desaulniers
On Tue, Oct 01, 2019 at 10:03:50AM -0700, Nick Desaulniers wrote:
> On Tue, Oct 1, 2019 at 1:37 AM Masahiro Yamada
> <[email protected]> wrote:
> >
> > KernelCI reports that bcm2835_defconfig is no longer booting since
> > commit ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING
> > forcibly") (https://lkml.org/lkml/2019/9/26/825).
> >
> > I also received a regression report from Nicolas Saenz Julienne
> > (https://lkml.org/lkml/2019/9/27/263).
> >
> > This problem has cropped up on bcm2835_defconfig because it enables
> > CONFIG_CC_OPTIMIZE_FOR_SIZE. The compiler tends to prefer not inlining
> > functions with -Os. I was able to reproduce it with other boards and
> > defconfig files by manually enabling CONFIG_CC_OPTIMIZE_FOR_SIZE.
> >
> > The __get_user_check() specifically uses r0, r1, r2 registers.
> > So, uaccess_save_and_enable() and uaccess_restore() must be inlined.
> > Otherwise, those register assignments would be entirely dropped,
> > according to my analysis of the disassembly.
> >
> > Prior to commit 9012d011660e ("compiler: allow all arches to enable
> > CONFIG_OPTIMIZE_INLINING"), the 'inline' marker was always enough for
> > inlining functions, except on x86.
> >
> > Since that commit, all architectures can enable CONFIG_OPTIMIZE_INLINING.
> > So, __always_inline is now the only guaranteed way of forcible inlining.
>
> No, the C preprocessor is the only guaranteed way of inlining. I
> preferred v1; if you're going to <strikethrough>play with
> fire</strikethrough>write assembly, don't get burned.
It seems we disagree on that.
Masahiro Yamada, please send this to the patch system, thanks.
>
> >
> > I also added __always_inline to 4 functions in the call-graph from the
> > __get_user_check() macro.
> >
> > Fixes: 9012d011660e ("compiler: allow all arches to enable CONFIG_OPTIMIZE_INLINING")
> > Reported-by: "kernelci.org bot" <[email protected]>
> > Reported-by: Nicolas Saenz Julienne <[email protected]>
> > Signed-off-by: Masahiro Yamada <[email protected]>
> > ---
> >
> > Changes in v2:
> > - Use __always_inline instead of changing the function call places
> > (per Russell King)
> > - The previous submission is: https://lore.kernel.org/patchwork/patch/1132459/
> >
> > arch/arm/include/asm/domain.h | 8 ++++----
> > arch/arm/include/asm/uaccess.h | 4 ++--
> > 2 files changed, 6 insertions(+), 6 deletions(-)
> >
> > diff --git a/arch/arm/include/asm/domain.h b/arch/arm/include/asm/domain.h
> > index 567dbede4785..f1d0a7807cd0 100644
> > --- a/arch/arm/include/asm/domain.h
> > +++ b/arch/arm/include/asm/domain.h
> > @@ -82,7 +82,7 @@
> > #ifndef __ASSEMBLY__
> >
> > #ifdef CONFIG_CPU_CP15_MMU
> > -static inline unsigned int get_domain(void)
> > +static __always_inline unsigned int get_domain(void)
> > {
> > unsigned int domain;
> >
> > @@ -94,7 +94,7 @@ static inline unsigned int get_domain(void)
> > return domain;
> > }
> >
> > -static inline void set_domain(unsigned val)
> > +static __always_inline void set_domain(unsigned int val)
> > {
> > asm volatile(
> > "mcr p15, 0, %0, c3, c0 @ set domain"
> > @@ -102,12 +102,12 @@ static inline void set_domain(unsigned val)
> > isb();
> > }
> > #else
> > -static inline unsigned int get_domain(void)
> > +static __always_inline unsigned int get_domain(void)
> > {
> > return 0;
> > }
> >
> > -static inline void set_domain(unsigned val)
> > +static __always_inline void set_domain(unsigned int val)
> > {
> > }
> > #endif
> > diff --git a/arch/arm/include/asm/uaccess.h b/arch/arm/include/asm/uaccess.h
> > index 303248e5b990..98c6b91be4a8 100644
> > --- a/arch/arm/include/asm/uaccess.h
> > +++ b/arch/arm/include/asm/uaccess.h
> > @@ -22,7 +22,7 @@
> > * perform such accesses (eg, via list poison values) which could then
> > * be exploited for priviledge escalation.
> > */
> > -static inline unsigned int uaccess_save_and_enable(void)
> > +static __always_inline unsigned int uaccess_save_and_enable(void)
> > {
> > #ifdef CONFIG_CPU_SW_DOMAIN_PAN
> > unsigned int old_domain = get_domain();
> > @@ -37,7 +37,7 @@ static inline unsigned int uaccess_save_and_enable(void)
> > #endif
> > }
> >
> > -static inline void uaccess_restore(unsigned int flags)
> > +static __always_inline void uaccess_restore(unsigned int flags)
> > {
> > #ifdef CONFIG_CPU_SW_DOMAIN_PAN
> > /* Restore the user access mask */
> > --
> > 2.17.1
> >
>
>
> --
> Thanks,
> ~Nick Desaulniers
>
--
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up
On Wed, Oct 2, 2019 at 5:25 PM Russell King - ARM Linux admin
<[email protected]> wrote:
> Masahiro Yamada, please send this to the patch system, thanks.
Done. (8908/1)
Thanks.
--
Best Regards
Masahiro Yamada
Hi Nick,
On Wed, Oct 2, 2019 at 2:04 AM Nick Desaulniers <[email protected]> wrote:
> > Since that commit, all architectures can enable CONFIG_OPTIMIZE_INLINING.
> > So, __always_inline is now the only guaranteed way of forcible inlining.
>
> No, the C preprocessor is the only guaranteed way of inlining.
I do not think this is fatal if I understood this correctly:
https://lore.kernel.org/patchwork/patch/1122097/#1331784
For GCC, we at least get a warning if a function with
__always_inline is not inlined.
I am fine with adding -Werror=attributes to the top Makefile
(unless we have unpleasant side-effects).
You filed the bug for Clang, and hopefully it will be OK
in the future releases?
--
Best Regards
Masahiro Yamada