[PATCH] x86/jailhouse: fix building without X86_X2APIC

[PATCH] x86/jailhouse: fix building without X86_X2APIC

[Arnd Bergmann]

When CONFIG_X86_X2APIC is disabled, jailhouse fails to build:

arch/x86/kernel/jailhouse.c: In function 'jailhouse_get_smp_config':
arch/x86/kernel/jailhouse.c:73:3: error: 'x2apic_phys' undeclared (first use in this function); did you mean 'x2apic_mode'?

The code is protected by an appropriate x2apic_enabled() check that leads
to the assignment being optimized out without a link-time reference to
x2apic_phys, so we just lack a declaration.

Let's move x2apic_mode and x2apic_phys outside of the #ifdef together,
for consistency.

Fixes: 11c8dc419bbc ("x86/jailhouse: Enable APIC and SMP support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 arch/x86/include/asm/apic.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
index 98722773391d..0317d635d9ba 100644
--- a/arch/x86/include/asm/apic.h
+++ b/arch/x86/include/asm/apic.h
@@ -188,6 +188,8 @@ static inline void lapic_assign_system_vectors(void) { }
 static inline void lapic_assign_legacy_vector(unsigned int i, bool r) { }
 #endif /* !CONFIG_X86_LOCAL_APIC */
 
+extern int x2apic_mode;
+extern int x2apic_phys;
 #ifdef CONFIG_X86_X2APIC
 /*
  * Make previous memory operations globally visible before
@@ -249,8 +251,6 @@ static inline u64 native_x2apic_icr_read(void)
 	return val;
 }
 
-extern int x2apic_mode;
-extern int x2apic_phys;
 extern void __init check_x2apic(void);
 extern void x2apic_setup(void);
 static inline int x2apic_enabled(void)
-- 
2.9.0

Re: [PATCH] x86/jailhouse: fix building without X86_X2APIC

[Dou Liyang]

Hi Arnd,

At 01/16/2018 04:23 AM, Arnd Bergmann wrote:
> When CONFIG_X86_X2APIC is disabled, jailhouse fails to build:
> 
> arch/x86/kernel/jailhouse.c: In function 'jailhouse_get_smp_config':
> arch/x86/kernel/jailhouse.c:73:3: error: 'x2apic_phys' undeclared (first use in this function); did you mean 'x2apic_mode'?
> 
> The code is protected by an appropriate x2apic_enabled() check that leads
> to the assignment being optimized out without a link-time reference to
> x2apic_phys, so we just lack a declaration.
> 
> Let's move x2apic_mode and x2apic_phys outside of the #ifdef together,
> for consistency.
> 
> Fixes: 11c8dc419bbc ("x86/jailhouse: Enable APIC and SMP support")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>   arch/x86/include/asm/apic.h | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
> index 98722773391d..0317d635d9ba 100644
> --- a/arch/x86/include/asm/apic.h
> +++ b/arch/x86/include/asm/apic.h
> @@ -188,6 +188,8 @@ static inline void lapic_assign_system_vectors(void) { }
>   static inline void lapic_assign_legacy_vector(unsigned int i, bool r) { }
>   #endif /* !CONFIG_X86_LOCAL_APIC */
>   
> +extern int x2apic_mode;
> +extern int x2apic_phys;
We can't do that, adding a macro for the X2APIC=n case is enough

Thanks,
	dou

-------------------------8<----------------------

Subject: [PATCH] x86/apic: Add a macro named x2apic_phys for the 
X2APIC=n case

The x86 system may build failed in the X2APIC=n case, due to the missing of
x2apic_phys.

So add a macro named x2apic_phys for the X2APIC=n case.

Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com>
---
  arch/x86/include/asm/apic.h | 3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
index a9e57f08bfa6..ca5940fb995a 100644
--- a/arch/x86/include/asm/apic.h
+++ b/arch/x86/include/asm/apic.h
@@ -264,7 +264,8 @@ static inline void x2apic_setup(void) { }
  static inline int x2apic_enabled(void) { return 0; }

  #define x2apic_mode		(0)
-#define	x2apic_supported()	(0)
+#define x2apic_phys		(0)
+#define x2apic_supported()	(0)
  #endif /* !CONFIG_X86_X2APIC */

  struct irq_data;
-- 

Re: [PATCH] x86/jailhouse: fix building without X86_X2APIC

[Dou Liyang]

Hi Arnd,

At 01/16/2018 09:25 AM, Dou Liyang wrote:
> Hi Arnd,
> 
> At 01/16/2018 04:23 AM, Arnd Bergmann wrote:
>> When CONFIG_X86_X2APIC is disabled, jailhouse fails to build:
>>
>> arch/x86/kernel/jailhouse.c: In function 'jailhouse_get_smp_config':
>> arch/x86/kernel/jailhouse.c:73:3: error: 'x2apic_phys' undeclared 
>> (first use in this function); did you mean 'x2apic_mode'?
>>
>> The code is protected by an appropriate x2apic_enabled() check that leads
>> to the assignment being optimized out without a link-time reference to
>> x2apic_phys, so we just lack a declaration.
>>
>> Let's move x2apic_mode and x2apic_phys outside of the #ifdef together,
>> for consistency.
>>
>> Fixes: 11c8dc419bbc ("x86/jailhouse: Enable APIC and SMP support")
>> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
>> ---
>>   arch/x86/include/asm/apic.h | 4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
>> index 98722773391d..0317d635d9ba 100644
>> --- a/arch/x86/include/asm/apic.h
>> +++ b/arch/x86/include/asm/apic.h
>> @@ -188,6 +188,8 @@ static inline void 
>> lapic_assign_system_vectors(void) { }
>>   static inline void lapic_assign_legacy_vector(unsigned int i, bool 
>> r) { }
>>   #endif /* !CONFIG_X86_LOCAL_APIC */
>> +extern int x2apic_mode;
>> +extern int x2apic_phys;
> We can't do that, adding a macro for the X2APIC=n case is enough
> 
I am sorry when I looked into your code in tip tree. I found this
measure is not true. please try the the following v2 patch.

The reason I don't want to expose the x2apic_mode and x2apic_phys is
that they may be misused in X2APIC=n case. So I create an interface to
wrap it. do you think so? ;-)

Thanks,
	dou.
---
  arch/x86/include/asm/apic.h | 10 +++++++++-
  arch/x86/kernel/apic/apic.c |  2 +-
  arch/x86/kernel/jailhouse.c |  2 +-
  3 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
index 98722773391d..ac25ac2e49af 100644
--- a/arch/x86/include/asm/apic.h
+++ b/arch/x86/include/asm/apic.h
@@ -251,6 +251,11 @@ static inline u64 native_x2apic_icr_read(void)

  extern int x2apic_mode;
  extern int x2apic_phys;
+static inline void apic_set_x2apic_phys(void)
+{
+	x2apic_phys = 1;
+}
+
  extern void __init check_x2apic(void);
  extern void x2apic_setup(void);
  static inline int x2apic_enabled(void)
@@ -265,7 +270,10 @@ static inline void x2apic_setup(void) { }
  static inline int x2apic_enabled(void) { return 0; }

  #define x2apic_mode		(0)
-#define	x2apic_supported()	(0)
+#define x2apic_phys		(0)
+#define x2apic_supported()	(0)
+
+static inline void apic_set_x2apic_phys(void){}
  #endif /* !CONFIG_X86_X2APIC */

  struct irq_data;
diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c
index 25ddf02598d2..440ff200104f 100644
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -1746,7 +1746,7 @@ static __init void try_to_enable_x2apic(int 
remap_mode)
  		 * without IR all CPUs can be addressed by IOAPIC/MSI
  		 * only in physical mode
  		 */
-		x2apic_phys = 1;
+		apic_set_x2apic_phys();
  	}
  	x2apic_enable();
  }
diff --git a/arch/x86/kernel/jailhouse.c b/arch/x86/kernel/jailhouse.c
index d6d5976a9b51..ce62162efcf2 100644
--- a/arch/x86/kernel/jailhouse.c
+++ b/arch/x86/kernel/jailhouse.c
@@ -70,7 +70,7 @@ static void __init jailhouse_get_smp_config(unsigned 
int early)
  		 * We do not have access to IR inside Jailhouse non-root cells.
  		 * So we have to run in physical mode.
  		 */
-		x2apic_phys = 1;
+		apic_set_x2apic_phys();

  		/*
  		 * This will trigger the switch to apic_x2apic_phys.
-- 
> Thanks,
>      dou
> 
> -------------------------8<----------------------
> 
> Subject: [PATCH] x86/apic: Add a macro named x2apic_phys for the 
> X2APIC=n case
> 
> The x86 system may build failed in the X2APIC=n case, due to the missing of
> x2apic_phys.
> 
> So add a macro named x2apic_phys for the X2APIC=n case.
> 
> Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com>
> ---
>   arch/x86/include/asm/apic.h | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
> index a9e57f08bfa6..ca5940fb995a 100644
> --- a/arch/x86/include/asm/apic.h
> +++ b/arch/x86/include/asm/apic.h
> @@ -264,7 +264,8 @@ static inline void x2apic_setup(void) { }
>   static inline int x2apic_enabled(void) { return 0; }
> 
>   #define x2apic_mode        (0)
> -#define    x2apic_supported()    (0)
> +#define x2apic_phys        (0)
> +#define x2apic_supported()    (0)
>   #endif /* !CONFIG_X86_X2APIC */
> 
>   struct irq_data;

Re: [PATCH] x86/jailhouse: fix building without X86_X2APIC

[Arnd Bergmann]

On Tue, Jan 16, 2018 at 3:34 AM, Dou Liyang <douly.fnst@cn.fujitsu.com> wrote:
> At 01/16/2018 09:25 AM, Dou Liyang wrote:

>>> diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
>>> index 98722773391d..0317d635d9ba 100644
>>> --- a/arch/x86/include/asm/apic.h
>>> +++ b/arch/x86/include/asm/apic.h
>>> @@ -188,6 +188,8 @@ static inline void lapic_assign_system_vectors(void)
>>> { }
>>>   static inline void lapic_assign_legacy_vector(unsigned int i, bool r) {
>>> }
>>>   #endif /* !CONFIG_X86_LOCAL_APIC */
>>> +extern int x2apic_mode;
>>> +extern int x2apic_phys;
>>
>> We can't do that, adding a macro for the X2APIC=n case is enough
>>
> I am sorry when I looked into your code in tip tree. I found this
> measure is not true. please try the the following v2 patch.
>
> The reason I don't want to expose the x2apic_mode and x2apic_phys is
> that they may be misused in X2APIC=n case. So I create an interface to
> wrap it. do you think so? ;-)

I'm not sure I follow what the intention of that is. If you want to hide
those two variables, maybe make them 'static' and remove the extern
declarations?

> diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
> index 98722773391d..ac25ac2e49af 100644
> --- a/arch/x86/include/asm/apic.h
> +++ b/arch/x86/include/asm/apic.h
> @@ -251,6 +251,11 @@ static inline u64 native_x2apic_icr_read(void)
>
>  extern int x2apic_mode;
>  extern int x2apic_phys;
> +static inline void apic_set_x2apic_phys(void)
> +{
> +       x2apic_phys = 1;
> +}
> +
>  extern void __init check_x2apic(void);
>  extern void x2apic_setup(void);
>  static inline int x2apic_enabled(void)
> @@ -265,7 +270,10 @@ static inline void x2apic_setup(void) { }
>  static inline int x2apic_enabled(void) { return 0; }
>
>  #define x2apic_mode            (0)
> -#define        x2apic_supported()      (0)
> +#define x2apic_phys            (0)
> +#define x2apic_supported()     (0)
> +
> +static inline void apic_set_x2apic_phys(void){}
>  #endif /* !CONFIG_X86_X2APIC */
>
>  struct irq_data;

I see nothing wrong it with this, but also don't see anything it does
that improves the interface.

      Arnd

Re: [PATCH] x86/jailhouse: fix building without X86_X2APIC

[Dou Liyang]

Hi Arnd,

[...]
>> The reason I don't want to expose the x2apic_mode and x2apic_phys is
>> that they may be misused in X2APIC=n case. So I create an interface to
>> wrap it. do you think so? ;-)
> 
> I'm not sure I follow what the intention of that is. If you want to hide

My purpose of that is hiding the variables in X2APIC=n case.

> those two variables, maybe make them 'static' and remove the extern
> declarations?
> 

Yes, that's what I really want, But due to the simple "x2apic_phys= 1"
operation, doing that may be too fat, I guess.

>> diff --git a/arch/x86/include/asm/apic.h b/arch/x86/include/asm/apic.h
>> index 98722773391d..ac25ac2e49af 100644
>> --- a/arch/x86/include/asm/apic.h
>> +++ b/arch/x86/include/asm/apic.h
>> @@ -251,6 +251,11 @@ static inline u64 native_x2apic_icr_read(void)
>>
>>   extern int x2apic_mode;
>>   extern int x2apic_phys;
>> +static inline void apic_set_x2apic_phys(void)
>> +{
>> +       x2apic_phys = 1;
>> +}
>> +
>>   extern void __init check_x2apic(void);
>>   extern void x2apic_setup(void);
>>   static inline int x2apic_enabled(void)
>> @@ -265,7 +270,10 @@ static inline void x2apic_setup(void) { }
>>   static inline int x2apic_enabled(void) { return 0; }
>>
>>   #define x2apic_mode            (0)
>> -#define        x2apic_supported()      (0)
>> +#define x2apic_phys            (0)
>> +#define x2apic_supported()     (0)
>> +
>> +static inline void apic_set_x2apic_phys(void){}
>>   #endif /* !CONFIG_X86_X2APIC */
>>
>>   struct irq_data;
> 
> I see nothing wrong it with this, but also don't see anything it does
> that improves the interface.
> 

Another way we can choice is wrap the code with "CONFIG_X86_X2APIC".

Thanks,
	dou
--------------------------------8<------------------------------------
diff --git a/arch/x86/kernel/jailhouse.c b/arch/x86/kernel/jailhouse.c
index d6d5976a9b51..d4aee43c8912 100644
--- a/arch/x86/kernel/jailhouse.c
+++ b/arch/x86/kernel/jailhouse.c
@@ -65,6 +65,7 @@ static void __init jailhouse_get_smp_config(unsigned 
int early)
         };
         unsigned int cpu;

+#ifdef CONFIG_X86_X2APIC
         if (x2apic_enabled()) {
                 /*
                  * We do not have access to IR inside Jailhouse 
non-root cells.
@@ -79,6 +80,7 @@ static void __init jailhouse_get_smp_config(unsigned 
int early)
                  */
                 default_acpi_madt_oem_check("", "");
         }
+#endif

         register_lapic_address(0xfee00000);

>        Arnd
> 
> 
> 

Re: [PATCH] x86/jailhouse: fix building without X86_X2APIC

[Arnd Bergmann]

On Tue, Jan 16, 2018 at 9:17 AM, Dou Liyang <douly.fnst@cn.fujitsu.com> wrote:
> Hi Arnd,
>
> [...]
>>>
>>> The reason I don't want to expose the x2apic_mode and x2apic_phys is
>>> that they may be misused in X2APIC=n case. So I create an interface to
>>> wrap it. do you think so? ;-)
>>
>>
>> I'm not sure I follow what the intention of that is. If you want to hide
>
>
> My purpose of that is hiding the variables in X2APIC=n case.

But why? I'd say either hide them all the time, or don't hide them at all.

>> I see nothing wrong it with this, but also don't see anything it does
>> that improves the interface.
>>
>
> Another way we can choice is wrap the code with "CONFIG_X86_X2APIC".
>

But why? That just makes perfectly reasonably code uglier. Generally
speaking, compiler conditionals are better than preprocessor conditionals
for this, as they are easier to read and provide better compile-time coverage
when things go wrong, such as the missing declaration.

     Arnd

> --------------------------------8<------------------------------------
> diff --git a/arch/x86/kernel/jailhouse.c b/arch/x86/kernel/jailhouse.c
> index d6d5976a9b51..d4aee43c8912 100644
> --- a/arch/x86/kernel/jailhouse.c
> +++ b/arch/x86/kernel/jailhouse.c
> @@ -65,6 +65,7 @@ static void __init jailhouse_get_smp_config(unsigned int
> early)
>         };
>         unsigned int cpu;
>
> +#ifdef CONFIG_X86_X2APIC
>         if (x2apic_enabled()) {
>                 /*
>                  * We do not have access to IR inside Jailhouse non-root
> cells.
> @@ -79,6 +80,7 @@ static void __init jailhouse_get_smp_config(unsigned int
> early)
>                  */
>                 default_acpi_madt_oem_check("", "");
>         }
> +#endif

Re: [PATCH] x86/jailhouse: fix building without X86_X2APIC

[Dou Liyang]

Hi Arnd,

At 01/16/2018 04:50 PM, Arnd Bergmann wrote:
> On Tue, Jan 16, 2018 at 9:17 AM, Dou Liyang <douly.fnst@cn.fujitsu.com> wrote:
>> Hi Arnd,
>>
>> [...]
>>>>
>>>> The reason I don't want to expose the x2apic_mode and x2apic_phys is
>>>> that they may be misused in X2APIC=n case. So I create an interface to
>>>> wrap it. do you think so? ;-)
>>>
>>>
>>> I'm not sure I follow what the intention of that is. If you want to hide
>>
>>
>> My purpose of that is hiding the variables in X2APIC=n case.
> 
> But why? I'd say either hide them all the time, or don't hide them at all.
> 

In this case, I prefer to hide them all the time.

>>> I see nothing wrong it with this, but also don't see anything it does
>>> that improves the interface.
>>>
>>
>> Another way we can choice is wrap the code with "CONFIG_X86_X2APIC".
>>
> 
> But why? That just makes perfectly reasonably code uglier. Generally
> speaking, compiler conditionals are better than preprocessor conditionals
> for this, as they are easier to read and provide better compile-time coverage
> when things go wrong, such as the missing declaration.
> 

Yes, Indeed. Thank you for telling me that.

Thanks,
	dou.