[PATCH 5.4-rc1 BUILD FIX] s390: mark __cpacf_query() as __always_inline

[PATCH 5.4-rc1 BUILD FIX] s390: mark __cpacf_query() as __always_inline

[Jiri Kosina]

arch/s390/kvm/kvm-s390.c calls on several places __cpacf_query() directly, 
which makes it impossible to meet the "i" constraint for the asm operands 
(opcode in this case).

As we are now force-enabling CONFIG_OPTIMIZE_INLINING on all 
architectures, this causes a build failure on s390:

   In file included from arch/s390/kvm/kvm-s390.c:44:
   ./arch/s390/include/asm/cpacf.h: In function '__cpacf_query':
   ./arch/s390/include/asm/cpacf.h:179:2: warning: asm operand 3 probably doesn't match constraints
     179 |  asm volatile(
         |  ^~~
   ./arch/s390/include/asm/cpacf.h:179:2: error: impossible constraint in 'asm'

Mark __cpacf_query() as __always_inline in order to fix that, analogically 
how we fixes __cpacf_check_opcode(), cpacf_query_func() and scpacf_query() 
already.

Reported-and-tested-by: Michal Kubecek <mkubecek@suse.cz>
Fixes: d83623c5eab2 ("s390: mark __cpacf_check_opcode() and cpacf_query_func() as __always_inline")
Fixes: e60fb8bf68d4 ("s390/cpacf: mark scpacf_query() as __always_inline")
Fixes: ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING forcibly")
Fixes: 9012d011660e ("compiler: allow all arches to enable CONFIG_OPTIMIZE_INLINING")
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
---

I am wondering how is it possible that none of the build-testing 
infrastructure we have running against linux-next caught this? Not enough 
non-x86 coverage?

 arch/s390/include/asm/cpacf.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/s390/include/asm/cpacf.h b/arch/s390/include/asm/cpacf.h
index a092f63aac6a..c0f3bfeddcbe 100644
--- a/arch/s390/include/asm/cpacf.h
+++ b/arch/s390/include/asm/cpacf.h
@@ -171,7 +171,7 @@ typedef struct { unsigned char bytes[16]; } cpacf_mask_t;
  *
  * Returns 1 if @func is available for @opcode, 0 otherwise
  */
-static inline void __cpacf_query(unsigned int opcode, cpacf_mask_t *mask)
+static __always_inline void __cpacf_query(unsigned int opcode, cpacf_mask_t *mask)
 {
 	register unsigned long r0 asm("0") = 0;	/* query function */
 	register unsigned long r1 asm("1") = (unsigned long) mask;

-- 
Jiri Kosina
SUSE Labs

Re: [PATCH 5.4-rc1 BUILD FIX] s390: mark __cpacf_query() as __always_inline

[Jiri Kosina]

On Tue, 1 Oct 2019, Jiri Kosina wrote:

> arch/s390/kvm/kvm-s390.c calls on several places __cpacf_query() directly, 
> which makes it impossible to meet the "i" constraint for the asm operands 
> (opcode in this case).
> 
> As we are now force-enabling CONFIG_OPTIMIZE_INLINING on all 
> architectures, this causes a build failure on s390:
> 
>    In file included from arch/s390/kvm/kvm-s390.c:44:
>    ./arch/s390/include/asm/cpacf.h: In function '__cpacf_query':
>    ./arch/s390/include/asm/cpacf.h:179:2: warning: asm operand 3 probably doesn't match constraints
>      179 |  asm volatile(
>          |  ^~~
>    ./arch/s390/include/asm/cpacf.h:179:2: error: impossible constraint in 'asm'
> 
> Mark __cpacf_query() as __always_inline in order to fix that, analogically 
> how we fixes __cpacf_check_opcode(), cpacf_query_func() and scpacf_query() 
> already.
> 
> Reported-and-tested-by: Michal Kubecek <mkubecek@suse.cz>
> Fixes: d83623c5eab2 ("s390: mark __cpacf_check_opcode() and cpacf_query_func() as __always_inline")
> Fixes: e60fb8bf68d4 ("s390/cpacf: mark scpacf_query() as __always_inline")
> Fixes: ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING forcibly")
> Fixes: 9012d011660e ("compiler: allow all arches to enable CONFIG_OPTIMIZE_INLINING")
> Signed-off-by: Jiri Kosina <jkosina@suse.cz>

Gah, due to bug in my script the sigoff doesn't match the From:, so 
whoever is potentially applying it, please ammend it with

	From: Jiri Kosina <jkosina@suse.cz>

Sorry for the noise.

-- 
Jiri Kosina
SUSE Labs

Re: [PATCH 5.4-rc1 BUILD FIX] s390: mark __cpacf_query() as __always_inline

[Christian Borntraeger]

On 01.10.19 22:08, Jiri Kosina wrote:
> arch/s390/kvm/kvm-s390.c calls on several places __cpacf_query() directly, 
> which makes it impossible to meet the "i" constraint for the asm operands 
> (opcode in this case).
> 
> As we are now force-enabling CONFIG_OPTIMIZE_INLINING on all 
> architectures, this causes a build failure on s390:
> 
>    In file included from arch/s390/kvm/kvm-s390.c:44:
>    ./arch/s390/include/asm/cpacf.h: In function '__cpacf_query':
>    ./arch/s390/include/asm/cpacf.h:179:2: warning: asm operand 3 probably doesn't match constraints
>      179 |  asm volatile(
>          |  ^~~
>    ./arch/s390/include/asm/cpacf.h:179:2: error: impossible constraint in 'asm'
> 
> Mark __cpacf_query() as __always_inline in order to fix that, analogically 
> how we fixes __cpacf_check_opcode(), cpacf_query_func() and scpacf_query() 
> already.
> 
> Reported-and-tested-by: Michal Kubecek <mkubecek@suse.cz>
> Fixes: d83623c5eab2 ("s390: mark __cpacf_check_opcode() and cpacf_query_func() as __always_inline")
> Fixes: e60fb8bf68d4 ("s390/cpacf: mark scpacf_query() as __always_inline")
> Fixes: ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING forcibly")
> Fixes: 9012d011660e ("compiler: allow all arches to enable CONFIG_OPTIMIZE_INLINING")
> Signed-off-by: Jiri Kosina <jkosina@suse.cz>

Thanks applied. 


> 
> I am wondering how is it possible that none of the build-testing 
> infrastructure we have running against linux-next caught this? Not enough 
> non-x86 coverage?

We do build-test linux-next daily. Maybe our compiler just made a different
inlining decision.

Re: [PATCH 5.4-rc1 BUILD FIX] s390: mark __cpacf_query() as __always_inline

[Heiko Carstens]

On Tue, Oct 01, 2019 at 10:08:01PM +0200, Jiri Kosina wrote:
> arch/s390/kvm/kvm-s390.c calls on several places __cpacf_query() directly, 
> which makes it impossible to meet the "i" constraint for the asm operands 
> (opcode in this case).
> 
> As we are now force-enabling CONFIG_OPTIMIZE_INLINING on all 
> architectures, this causes a build failure on s390:
> 
>    In file included from arch/s390/kvm/kvm-s390.c:44:
>    ./arch/s390/include/asm/cpacf.h: In function '__cpacf_query':
>    ./arch/s390/include/asm/cpacf.h:179:2: warning: asm operand 3 probably doesn't match constraints
>      179 |  asm volatile(
>          |  ^~~
>    ./arch/s390/include/asm/cpacf.h:179:2: error: impossible constraint in 'asm'
> 
> Mark __cpacf_query() as __always_inline in order to fix that, analogically 
> how we fixes __cpacf_check_opcode(), cpacf_query_func() and scpacf_query() 
> already.
> 
> Reported-and-tested-by: Michal Kubecek <mkubecek@suse.cz>
> Fixes: d83623c5eab2 ("s390: mark __cpacf_check_opcode() and cpacf_query_func() as __always_inline")
> Fixes: e60fb8bf68d4 ("s390/cpacf: mark scpacf_query() as __always_inline")
> Fixes: ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING forcibly")
> Fixes: 9012d011660e ("compiler: allow all arches to enable CONFIG_OPTIMIZE_INLINING")
> Signed-off-by: Jiri Kosina <jkosina@suse.cz>
> ---
> 
> I am wondering how is it possible that none of the build-testing 
> infrastructure we have running against linux-next caught this? Not enough 
> non-x86 coverage?

Well, there is plenty of s390 coverage with respect to daily builds
(also here). It doesn't fail for me with gcc 9.1; so you may either
have a different gcc version or different config options(?) so the
compiler decided to not inline the function. I'll apply this in any
case, since your patch is obviously fine.

Thanks!

>  arch/s390/include/asm/cpacf.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/s390/include/asm/cpacf.h b/arch/s390/include/asm/cpacf.h
> index a092f63aac6a..c0f3bfeddcbe 100644
> --- a/arch/s390/include/asm/cpacf.h
> +++ b/arch/s390/include/asm/cpacf.h
> @@ -171,7 +171,7 @@ typedef struct { unsigned char bytes[16]; } cpacf_mask_t;
>   *
>   * Returns 1 if @func is available for @opcode, 0 otherwise
>   */
> -static inline void __cpacf_query(unsigned int opcode, cpacf_mask_t *mask)
> +static __always_inline void __cpacf_query(unsigned int opcode, cpacf_mask_t *mask)
>  {
>  	register unsigned long r0 asm("0") = 0;	/* query function */
>  	register unsigned long r1 asm("1") = (unsigned long) mask;
> 
> -- 
> Jiri Kosina
> SUSE Labs
> 

Re: [PATCH 5.4-rc1 BUILD FIX] s390: mark __cpacf_query() as __always_inline

[Heiko Carstens]

On Wed, Oct 02, 2019 at 08:46:05AM +0200, Heiko Carstens wrote:
> On Tue, Oct 01, 2019 at 10:08:01PM +0200, Jiri Kosina wrote:
> > I am wondering how is it possible that none of the build-testing 
> > infrastructure we have running against linux-next caught this? Not enough 
> > non-x86 coverage?
> 
> Well, there is plenty of s390 coverage with respect to daily builds
> (also here). It doesn't fail for me with gcc 9.1; so you may either
> have a different gcc version or different config options(?) so the
> compiler decided to not inline the function. I'll apply this in any
> case, since your patch is obviously fine.
> 
> Thanks!

Ok, Christian applied this already a couple of minutes earlier ;)

Re: [PATCH 5.4-rc1 BUILD FIX] s390: mark __cpacf_query() as __always_inline

[Michal Kubecek]

On Wed, Oct 02, 2019 at 08:46:05AM +0200, Heiko Carstens wrote:
> On Tue, Oct 01, 2019 at 10:08:01PM +0200, Jiri Kosina wrote:
> > 
> >    In file included from arch/s390/kvm/kvm-s390.c:44:
> >    ./arch/s390/include/asm/cpacf.h: In function '__cpacf_query':
> >    ./arch/s390/include/asm/cpacf.h:179:2: warning: asm operand 3 probably doesn't match constraints
> >      179 |  asm volatile(
> >          |  ^~~
> >    ./arch/s390/include/asm/cpacf.h:179:2: error: impossible constraint in 'asm'
> > 
> > ...
> > 
> > I am wondering how is it possible that none of the build-testing 
> > infrastructure we have running against linux-next caught this? Not enough 
> > non-x86 coverage?
> 
> Well, there is plenty of s390 coverage with respect to daily builds
> (also here). It doesn't fail for me with gcc 9.1; so you may either
> have a different gcc version or different config options(?) so the
> compiler decided to not inline the function.

I think I found the reason: we only hit the build failure with one
special config used for zfcpdump which has

  CONFIG_CC_OPTIMIZE_FOR_SIZE=y

When I switched to CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE=y (which we have
in other s390x configs and which most people probably prefer), the build
does not fail even without the patch.

Michal Kubecek

Re: [PATCH 5.4-rc1 BUILD FIX] s390: mark __cpacf_query() as __always_inline

[Heiko Carstens]

On Wed, Oct 02, 2019 at 09:03:33AM +0200, Michal Kubecek wrote:
> On Wed, Oct 02, 2019 at 08:46:05AM +0200, Heiko Carstens wrote:
> > On Tue, Oct 01, 2019 at 10:08:01PM +0200, Jiri Kosina wrote:
> > > 
> > >    In file included from arch/s390/kvm/kvm-s390.c:44:
> > >    ./arch/s390/include/asm/cpacf.h: In function '__cpacf_query':
> > >    ./arch/s390/include/asm/cpacf.h:179:2: warning: asm operand 3 probably doesn't match constraints
> > >      179 |  asm volatile(
> > >          |  ^~~
> > >    ./arch/s390/include/asm/cpacf.h:179:2: error: impossible constraint in 'asm'
> > > 
> > > ...
> > > 
> > > I am wondering how is it possible that none of the build-testing 
> > > infrastructure we have running against linux-next caught this? Not enough 
> > > non-x86 coverage?
> > 
> > Well, there is plenty of s390 coverage with respect to daily builds
> > (also here). It doesn't fail for me with gcc 9.1; so you may either
> > have a different gcc version or different config options(?) so the
> > compiler decided to not inline the function.
> 
> I think I found the reason: we only hit the build failure with one
> special config used for zfcpdump which has
> 
>   CONFIG_CC_OPTIMIZE_FOR_SIZE=y
> 
> When I switched to CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE=y (which we have
> in other s390x configs and which most people probably prefer), the build
> does not fail even without the patch.

Yes, with CONFIG_CC_OPTIMIZE_FOR_SIZE=y I can see plenty of _additional_
compile failures on s390 with "defconfig". Will fix them all...

Thanks for reporting!