Skip to contentSkip to conversation

arch/x86/entry_fred: don't set up KVM IRQs if KVM is disabled

PATCH In review Max Kellermann 6 participants 15 messages Feb 15, 2024 – Feb 17, 2024

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 arch/x86/entry_fred: don't set up KVM IRQs if KVM is disabled

When KVM is disabled, the POSTED_INTR_* macros do not exist, and the build fails.

Fixes 14619d912b65 ("x86/fred: FRED entry/exit and dispatch code")
---
 arch/x86/entry/entry_fred.c | 2 ++
 1 file changed, 2 insertions(+)
arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
 
 	SYSVEC(IRQ_WORK_VECTOR,			irq_work),
 
+#if IS_ENABLED(CONFIG_KVM)
 	SYSVEC(POSTED_INTR_VECTOR,		kvm_posted_intr_ipi),
 	SYSVEC(POSTED_INTR_WAKEUP_VECTOR,	kvm_posted_intr_wakeup_ipi),
 	SYSVEC(POSTED_INTR_NESTED_VECTOR,	kvm_posted_intr_nested_ipi),
+#endif
 };
 
 static bool fred_setup_done __initdata;
Signature · 1 line
2.39.2
maintainer
#1 lore raw
Quoted 15 lines · Max Kellermann

When KVM is disabled, the POSTED_INTR_* macros do not exist, and the build fails.

Fixes 14619d912b65 ("x86/fred: FRED entry/exit and dispatch code")
---
  arch/x86/entry/entry_fred.c | 2 ++
  1 file changed, 2 insertions(+)
arch/x86/entry/entry_fred.c +0 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
 

arch/x86/entry/entry_fred.c static idtentry_t sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = { · line 124

Quoted 10 lines
arch/x86/entry/entry_fred.c +2 -0
  	SYSVEC(IRQ_WORK_VECTOR,			irq_work),
 
+#if IS_ENABLED(CONFIG_KVM)
  	SYSVEC(POSTED_INTR_VECTOR,		kvm_posted_intr_ipi),
  	SYSVEC(POSTED_INTR_WAKEUP_VECTOR,	kvm_posted_intr_wakeup_ipi),
  	SYSVEC(POSTED_INTR_NESTED_VECTOR,	kvm_posted_intr_nested_ipi),
+#endif
  };
 
  static bool fred_setup_done __initdata;

Hmm, we did test against !CONFIG_KVM.

The POSTED_INTR_* macros are under CONFIG_HAVE_KVM, which is 'selected' under CONFIG_X86.

Can you please send me your kernel config file?

Signature · 2 lines
Thanks!
    Xin
#2 lore raw
Quoted 2 lines · Xin Li

The POSTED_INTR_* macros are under CONFIG_HAVE_KVM, which is 'selected' under CONFIG_X86.

Sorry, I should have said that I found this bug on linux-next master, which has this commit:

https://kernel.googlesource.com/pub/scm/linux/kernel/git/next/linux-next/+/dcf0926e9b899eca754a07c4064de69815b85a38%5E%21/

.. which changes CONFIG_HAVE_KVM to CONFIG_KVM. I was not aware that this commit hadn't been merged upstream yet.

You can easily reproduce with with "defconfig" plus CONFIG_X86_FRED (on linux-next/master).

Max

#3 lore raw
In reply to Max Kellermann

+Paolo and Stephen

FYI, there's a build failure in -next due to a collision between kvm/next and tip/x86/fred. The above makes everything happy.

Quoted 27 lines · Max Kellermann

When KVM is disabled, the POSTED_INTR_* macros do not exist, and the build fails.

Fixes 14619d912b65 ("x86/fred: FRED entry/exit and dispatch code")
---
 arch/x86/entry/entry_fred.c | 2 ++
 1 file changed, 2 insertions(+)
arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
 
 	SYSVEC(IRQ_WORK_VECTOR,			irq_work),
 
+#if IS_ENABLED(CONFIG_KVM)
 	SYSVEC(POSTED_INTR_VECTOR,		kvm_posted_intr_ipi),
 	SYSVEC(POSTED_INTR_WAKEUP_VECTOR,	kvm_posted_intr_wakeup_ipi),
 	SYSVEC(POSTED_INTR_NESTED_VECTOR,	kvm_posted_intr_nested_ipi),
+#endif
 };
 
 static bool fred_setup_done __initdata;
Signature · 1 line
2.39.2
maintainer
#4 lore raw
Quoted 30 lines · Sean Christopherson

+Paolo and Stephen

FYI, there's a build failure in -next due to a collision between kvm/next and tip/x86/fred. The above makes everything happy.

Quoted 24 lines · Max Kellermann

When KVM is disabled, the POSTED_INTR_* macros do not exist, and the build fails.

Fixes 14619d912b65 ("x86/fred: FRED entry/exit and dispatch code")
---
  arch/x86/entry/entry_fred.c | 2 ++
  1 file changed, 2 insertions(+)
arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
 
  	SYSVEC(IRQ_WORK_VECTOR,			irq_work),
 
+#if IS_ENABLED(CONFIG_KVM)
  	SYSVEC(POSTED_INTR_VECTOR,		kvm_posted_intr_ipi),
  	SYSVEC(POSTED_INTR_WAKEUP_VECTOR,	kvm_posted_intr_wakeup_ipi),
  	SYSVEC(POSTED_INTR_NESTED_VECTOR,	kvm_posted_intr_nested_ipi),
+#endif
  };
 
Quoted 4 lines
Quoted 3 lines
arch/x86/entry/entry_fred.c +0 -0
  static bool fred_setup_done __initdata;
Signature · 1 line
2.39.2

We want to minimize #ifdeffery (which is why we didn't add any to
sysvec_table[]), would it be better to simply remove "#if IS_ENABLED(CONFIG_KVM)" around the the POSTED_INTR_* macros from the Linux-next tree?

BTW, kvm_posted_intr_*() are defined to NULL if !IS_ENABLED(CONFIG_KVM).

arch/x86/include/asm/irq_vectors.h +0 -2
index 3a19904c2db6..d18bfb238f66 100644
@@ -84,11 +84,9 @@
 #define HYPERVISOR_CALLBACK_VECTOR	0xf3
 
 /* Vector for KVM to deliver posted interrupt IPI */
-#if IS_ENABLED(CONFIG_KVM)
 #define POSTED_INTR_VECTOR		0xf2
 #define POSTED_INTR_WAKEUP_VECTOR	0xf1
 #define POSTED_INTR_NESTED_VECTOR	0xf0
-#endif
 
 #define MANAGED_IRQ_SHUTDOWN_VECTOR	0xef
 
maintainer
#5 lore raw
Quoted 37 lines · Xin Li
Quoted 15 lines · Sean Christopherson

+Paolo and Stephen

FYI, there's a build failure in -next due to a collision between kvm/next and tip/x86/fred.  The above makes everything happy.

Quoted 9 lines · Max Kellermann

When KVM is disabled, the POSTED_INTR_* macros do not exist, and the build fails.

Fixes 14619d912b65 ("x86/fred: FRED entry/exit and dispatch code")

---
  arch/x86/entry/entry_fred.c | 2 ++

  1 file changed, 2 insertions(+)
Quoted 16 lines
Quoted 16 lines
arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t 
 sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
       SYSVEC(IRQ_WORK_VECTOR,            irq_work),
+#if IS_ENABLED(CONFIG_KVM)
       SYSVEC(POSTED_INTR_VECTOR,        kvm_posted_intr_ipi),
       SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    kvm_posted_intr_wakeup_ipi),
       SYSVEC(POSTED_INTR_NESTED_VECTOR,    kvm_posted_intr_nested_ipi),
+#endif
   };
   static bool fred_setup_done __initdata;
Signature · 1 line
2.39.2

We want to minimize #ifdeffery (which is why we didn't add any to
sysvec_table[]), would it be better to simply remove "#if IS_ENABLED(CONFIG_KVM)" around the the POSTED_INTR_* macros from the Linux-next tree?

Quoted 1 line

BTW, kvm_posted_intr_*() are defined to NULL if !IS_ENABLED(CONFIG_KVM).

It is intentional that KVM-related things are compiled out completely if !IS_ENABLED(CONFIG_KVM), because then it's also not necessary to have

# define fred_sysvec_kvm_posted_intr_ipi                NULL
# define fred_sysvec_kvm_posted_intr_wakeup_ipi         NULL
# define fred_sysvec_kvm_posted_intr_nested_ipi         NULL

in arch/x86/include/asm/idtentry.h. The full conflict resultion is

arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
 
     SYSVEC(IRQ_WORK_VECTOR,            irq_work),
 
+#if IS_ENABLED(CONFIG_KVM)
     SYSVEC(POSTED_INTR_VECTOR,        kvm_posted_intr_ipi),
     SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    kvm_posted_intr_wakeup_ipi),
     SYSVEC(POSTED_INTR_NESTED_VECTOR,    kvm_posted_intr_nested_ipi),
+#endif
 };
 
 static bool fred_setup_done __initdata;
arch/x86/include/asm/idtentry.h +0 -4
index 749c7411d2f1..758f6a2838a8 100644
@@ -745,10 +745,6 @@ DECLARE_IDTENTRY_SYSVEC(IRQ_WORK_VECTOR,        sysvec_irq_work);
 DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_VECTOR,        sysvec_kvm_posted_intr_ipi);
 DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    sysvec_kvm_posted_intr_wakeup_ipi);
 DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_NESTED_VECTOR,    sysvec_kvm_posted_intr_nested_ipi);
-#else
-# define fred_sysvec_kvm_posted_intr_ipi        NULL
-# define fred_sysvec_kvm_posted_intr_wakeup_ipi        NULL
-# define fred_sysvec_kvm_posted_intr_nested_ipi        NULL
 #endif
 
 #if IS_ENABLED(CONFIG_HYPERV)

and it seems to be a net improvement to me. The #ifs match in the .h and .c files, and there are no unnecessary initializers in the sysvec_table.

Paolo

maintainer
#6 lore raw
Quoted 38 lines · Paolo Bonzini
Quoted 37 lines · Xin Li
Quoted 15 lines · Sean Christopherson

+Paolo and Stephen

FYI, there's a build failure in -next due to a collision between kvm/next and tip/x86/fred.  The above makes everything happy.

Quoted 9 lines · Max Kellermann

When KVM is disabled, the POSTED_INTR_* macros do not exist, and the build fails.

Fixes 14619d912b65 ("x86/fred: FRED entry/exit and dispatch code")

---
  arch/x86/entry/entry_fred.c | 2 ++

  1 file changed, 2 insertions(+)
Quoted 16 lines
Quoted 16 lines
arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t 
 sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
       SYSVEC(IRQ_WORK_VECTOR,            irq_work),
+#if IS_ENABLED(CONFIG_KVM)
       SYSVEC(POSTED_INTR_VECTOR,        kvm_posted_intr_ipi),
       SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    kvm_posted_intr_wakeup_ipi),
       SYSVEC(POSTED_INTR_NESTED_VECTOR,    kvm_posted_intr_nested_ipi),
+#endif
   };
   static bool fred_setup_done __initdata;
Signature · 1 line
2.39.2

We want to minimize #ifdeffery (which is why we didn't add any to
sysvec_table[]), would it be better to simply remove "#if IS_ENABLED(CONFIG_KVM)" around the the POSTED_INTR_* macros from the Linux-next tree?

Quoted 4 lines
Quoted 1 line

BTW, kvm_posted_intr_*() are defined to NULL if !IS_ENABLED(CONFIG_KVM).

It is intentional that KVM-related things are compiled out completely if !IS_ENABLED(CONFIG_KVM),

In arch/x86/include/asm/irq_vectors.h, most vector definitions are not under any #ifdeffery, e.g., THERMAL_APIC_VECTOR not under CONFIG_X86_THERMAL_VECTOR and IRQ_WORK_VECTOR not under CONFIG_IRQ_WORK.

We'd better make all of them consistent, and the question is that should we add #ifdefs or not.

Quoted 45 lines

because then it's also not necessary to have

# define fred_sysvec_kvm_posted_intr_ipi                NULL
# define fred_sysvec_kvm_posted_intr_wakeup_ipi         NULL
# define fred_sysvec_kvm_posted_intr_nested_ipi         NULL

in arch/x86/include/asm/idtentry.h. The full conflict resultion is

arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t sysvec_table[NR_SYSTEM_VECTORS] 
 __ro_after_init = {
 
      SYSVEC(IRQ_WORK_VECTOR,            irq_work),
 
+#if IS_ENABLED(CONFIG_KVM)
      SYSVEC(POSTED_INTR_VECTOR,        kvm_posted_intr_ipi),
      SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    kvm_posted_intr_wakeup_ipi),
      SYSVEC(POSTED_INTR_NESTED_VECTOR,    kvm_posted_intr_nested_ipi),
+#endif
  };
 

 static bool fred_setup_done __initdata;

arch/x86/include/asm/idtentry.h +0 -0
index 749c7411d2f1..758f6a2838a8 100644
@@ -745,10 +745,6 @@ DECLARE_IDTENTRY_SYSVEC(IRQ_WORK_VECTOR,        
 sysvec_irq_work);
  DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_VECTOR,        
 sysvec_kvm_posted_intr_ipi);
  DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    
 sysvec_kvm_posted_intr_wakeup_ipi);
  DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_NESTED_VECTOR,    

sysvec_kvm_posted_intr_nested_ipi);

arch/x86/include/asm/idtentry.h +0 -4
-#else
-# define fred_sysvec_kvm_posted_intr_ipi        NULL
-# define fred_sysvec_kvm_posted_intr_wakeup_ipi        NULL
-# define fred_sysvec_kvm_posted_intr_nested_ipi        NULL
  #endif
 
  #if IS_ENABLED(CONFIG_HYPERV)
Quoted 4 lines

and it seems to be a net improvement to me.  The #ifs match in the .h and .c files, and there are no unnecessary initializers in the sysvec_table.

I somehow get an impression that the x86 maintainers don't like #ifs in the .c files, but I could be just wrong.

Thanks!
Xin

maintainer
#7 lore raw
Quoted 93 lines · Xin Li
Quoted 38 lines · Paolo Bonzini
Quoted 37 lines · Xin Li
Quoted 15 lines · Sean Christopherson

+Paolo and Stephen

FYI, there's a build failure in -next due to a collision between kvm/next and tip/x86/fred.  The above makes everything happy.

Quoted 9 lines · Max Kellermann

When KVM is disabled, the POSTED_INTR_* macros do not exist, and the build fails.

Fixes 14619d912b65 ("x86/fred: FRED entry/exit and dispatch code")

---
  arch/x86/entry/entry_fred.c | 2 ++

  1 file changed, 2 insertions(+)
Quoted 16 lines
Quoted 16 lines
arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t 
 sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
       SYSVEC(IRQ_WORK_VECTOR,            irq_work),
+#if IS_ENABLED(CONFIG_KVM)
       SYSVEC(POSTED_INTR_VECTOR,        kvm_posted_intr_ipi),
       SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    kvm_posted_intr_wakeup_ipi),
       SYSVEC(POSTED_INTR_NESTED_VECTOR,    kvm_posted_intr_nested_ipi),
+#endif
   };
   static bool fred_setup_done __initdata;
Signature · 1 line
2.39.2

We want to minimize #ifdeffery (which is why we didn't add any to
sysvec_table[]), would it be better to simply remove "#if IS_ENABLED(CONFIG_KVM)" around the the POSTED_INTR_* macros from the Linux-next tree?

Quoted 4 lines
Quoted 1 line

BTW, kvm_posted_intr_*() are defined to NULL if !IS_ENABLED(CONFIG_KVM).

It is intentional that KVM-related things are compiled out completely if !IS_ENABLED(CONFIG_KVM),

In arch/x86/include/asm/irq_vectors.h, most vector definitions are not under any #ifdeffery, e.g., THERMAL_APIC_VECTOR not under CONFIG_X86_THERMAL_VECTOR and IRQ_WORK_VECTOR not under CONFIG_IRQ_WORK.

We'd better make all of them consistent, and the question is that should we add #ifdefs or not.

Quoted 42 lines

because then it's also not necessary to have

# define fred_sysvec_kvm_posted_intr_ipi                NULL
# define fred_sysvec_kvm_posted_intr_wakeup_ipi         NULL
# define fred_sysvec_kvm_posted_intr_nested_ipi         NULL

in arch/x86/include/asm/idtentry.h. The full conflict resultion is

arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t sysvec_table[NR_SYSTEM_VECTORS] 
 __ro_after_init = {
 
       SYSVEC(IRQ_WORK_VECTOR,            irq_work),
 
+#if IS_ENABLED(CONFIG_KVM)
       SYSVEC(POSTED_INTR_VECTOR,        kvm_posted_intr_ipi),
       SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    kvm_posted_intr_wakeup_ipi),
       SYSVEC(POSTED_INTR_NESTED_VECTOR,    kvm_posted_intr_nested_ipi),
+#endif
   };
 

  static bool fred_setup_done __initdata;

arch/x86/include/asm/idtentry.h +0 -4
index 749c7411d2f1..758f6a2838a8 100644
@@ -745,10 +745,6 @@ DECLARE_IDTENTRY_SYSVEC(IRQ_WORK_VECTOR, 
 sysvec_irq_work);
   DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_VECTOR, sysvec_kvm_posted_intr_ipi);
   DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_WAKEUP_VECTOR, sysvec_kvm_posted_intr_wakeup_ipi);
   DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_NESTED_VECTOR, sysvec_kvm_posted_intr_nested_ipi);
-#else
-# define fred_sysvec_kvm_posted_intr_ipi        NULL
-# define fred_sysvec_kvm_posted_intr_wakeup_ipi        NULL
-# define fred_sysvec_kvm_posted_intr_nested_ipi        NULL
   #endif
 

  #if IS_ENABLED(CONFIG_HYPERV)

Quoted 8 lines
Quoted 4 lines

and it seems to be a net improvement to me.  The #ifs match in the .h and .c files, and there are no unnecessary initializers in the sysvec_table.

I somehow get an impression that the x86 maintainers don't like #ifs in the .c files, but I could be just wrong.

Here is an example, but again my interpretation could just be wrong:

#ifdef CONFIG_X86_FRED
void fred_install_sysvec(unsigned int vector, const idtentry_t function); #else
static inline void fred_install_sysvec(unsigned int vector, const idtentry_t function) { } #endif

#define sysvec_install(vector, function) {                              \
        if (cpu_feature_enabled(X86_FEATURE_FRED))                      \
                fred_install_sysvec(vector, function);                  \
        else                                                            \
                idt_install_sysvec(vector, asm_##function);             \

}

#8 lore raw
In reply to Paolo Bonzini
Quoted 5 lines · Paolo Bonzini

On 2/16/24 03:10, Xin Li wrote:

It is intentional that KVM-related things are compiled out completely if !IS_ENABLED(CONFIG_KVM), because then it's also not necessary to have

That's a matter of taste. In both cases _ALL_ KVM related things are compiled out.

#ifdeffing out the vector numbers is silly to begin with because these vector numbers stay assigned to KVM whether KVM is enabled or not.

And no, I don't think it's a net win to have the #ifdeffery in that table. Look at apic_idts[] in arch/x86/kernel/idt.c how this ends up looking. It's unreadable gunk.

The few NULL defines in a header file next to the real stuff

#if IS_ENABLED(CONFIG_KVM)
.....
#else

# define fred_sysvec_kvm_posted_intr_ipi                NULL
# define fred_sysvec_kvm_posted_intr_wakeup_ipi         NULL
# define fred_sysvec_kvm_posted_intr_nested_ipi         NULL
#endif

are not hurting at all and they are at a place where #ifdeffery is required anyway. That's a very common pattern all over the kernel and it limits the #ifdef horror to _ONE_ place.

With your change you propagate the #ifdefffery to the multiple and the very wrong places for absolutely zero practical value. The resulting binary code is exactly the same for the price of tasteless #ifdeffery in places where it matters.

Please get rid of this #ifdef in the vector header and don't inflict bad taste on everyone.

Thanks,

tglx

#9 lore raw
Quoted 2 lines · Thomas Gleixner

#ifdeffing out the vector numbers is silly to begin with because these vector numbers stay assigned to KVM whether KVM is enabled or not.

There could be one non-silly use of this: if the macros are not defined in absence of the feature, any use of it will lead to a compiler error, which is good, because it may reveal certain kinds of bugs.
(Though I agree that this isn't worth the code ugliness. I prefer to avoid the preprocessor whenever possible. I hate how much the kernel uses macros instead of inline functions.)

#10 lore raw
Quoted 8 lines · Max Kellermann
Quoted 2 lines · Thomas Gleixner

#ifdeffing out the vector numbers is silly to begin with because these vector numbers stay assigned to KVM whether KVM is enabled or not.

There could be one non-silly use of this: if the macros are not defined in absence of the feature, any use of it will lead to a compiler error, which is good, because it may reveal certain kinds of bugs.

I generally agree with this sentiment, but for constants like those in the case at hand I really draw the line.

Quoted 3 lines

(Though I agree that this isn't worth the code ugliness. I prefer to avoid the preprocessor whenever possible. I hate how much the kernel uses macros instead of inline functions.)

No argument about that. I'm urging people to use inlines instead of macros where ever possible, but there are things which can only solved by macros.

I'm well aware that I wrote some of the more ugly ones myself. Though the end justifies the means. If the ugly macro from hell which you verify once safes you from the horrors of copy & pasta error hell then they are making the code better and there are plenty of options to make them reasonably (type) safe if done right.

Thanks,

tglx

maintainer
#11 lore raw
In reply to Thomas Gleixner
Quoted 8 lines · Thomas Gleixner
Quoted 5 lines · Paolo Bonzini

On 2/16/24 03:10, Xin Li wrote:

It is intentional that KVM-related things are compiled out completely if !IS_ENABLED(CONFIG_KVM), because then it's also not necessary to have

Quoted 5 lines

That's a matter of taste. In both cases _ALL_ KVM related things are compiled out.

#ifdeffing out the vector numbers is silly to begin with because these vector numbers stay assigned to KVM whether KVM is enabled or not.

No problem---it seems that I misunderstood or read too much into the usage of CONFIG_HAVE_KVM up to 6.8, so I'm happy to follow whatever FRED support did for thermal vector and the like, and remove the #ifdef for the vector numbers.

Paolo

#12 lore raw
Quoted 8 lines · Paolo Bonzini
Quoted 2 lines · Thomas Gleixner

#ifdeffing out the vector numbers is silly to begin with because these vector numbers stay assigned to KVM whether KVM is enabled or not.

No problem---it seems that I misunderstood or read too much into the usage of CONFIG_HAVE_KVM up to 6.8, so I'm happy to follow whatever FRED support did for thermal vector and the like, and remove the #ifdef for the vector numbers.

Thank you! Appreciated!

tglx

maintainer
#13 lore raw
In reply to Paolo Bonzini

+ Arnd for

https://lore.kernel.org/r/[email protected]

Quoted 79 lines · Paolo Bonzini
Quoted 39 lines · Xin Li
Quoted 16 lines · Sean Christopherson

+Paolo and Stephen

FYI, there's a build failure in -next due to a collision between kvm/next and
tip/x86/fred.  The above makes everything happy.

Quoted 9 lines · Max Kellermann

When KVM is disabled, the POSTED_INTR_* macros do not exist, and the build fails.

Fixes 14619d912b65 ("x86/fred: FRED entry/exit and dispatch code")

---
  arch/x86/entry/entry_fred.c | 2 ++

  1 file changed, 2 insertions(+)
Quoted 16 lines
Quoted 16 lines
arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t
 sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
       SYSVEC(IRQ_WORK_VECTOR,            irq_work),
+#if IS_ENABLED(CONFIG_KVM)
       SYSVEC(POSTED_INTR_VECTOR,        kvm_posted_intr_ipi),
       SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    kvm_posted_intr_wakeup_ipi),
       SYSVEC(POSTED_INTR_NESTED_VECTOR,    kvm_posted_intr_nested_ipi),
+#endif
   };
   static bool fred_setup_done __initdata;
Signature · 1 line
2.39.2

We want to minimize #ifdeffery (which is why we didn't add any to sysvec_table[]), would it be better to simply remove "#if IS_ENABLED(CONFIG_KVM)" around the the POSTED_INTR_* macros from the Linux-next tree?

Quoted 1 line

BTW, kvm_posted_intr_*() are defined to NULL if !IS_ENABLED(CONFIG_KVM).

It is intentional that KVM-related things are compiled out completely if !IS_ENABLED(CONFIG_KVM), because then it's also not necessary to have

# define fred_sysvec_kvm_posted_intr_ipi                NULL
# define fred_sysvec_kvm_posted_intr_wakeup_ipi         NULL
# define fred_sysvec_kvm_posted_intr_nested_ipi         NULL

in arch/x86/include/asm/idtentry.h. The full conflict resultion is

arch/x86/entry/entry_fred.c +2 -0
index ac120cbdaaf2..660b7f7f9a79 100644
@@ -114,9 +114,11 @@ static idtentry_t sysvec_table[NR_SYSTEM_VECTORS] __ro_after_init = {
     SYSVEC(IRQ_WORK_VECTOR,            irq_work),
+#if IS_ENABLED(CONFIG_KVM)
     SYSVEC(POSTED_INTR_VECTOR,        kvm_posted_intr_ipi),
     SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    kvm_posted_intr_wakeup_ipi),
     SYSVEC(POSTED_INTR_NESTED_VECTOR,    kvm_posted_intr_nested_ipi),
+#endif
 };
 static bool fred_setup_done __initdata;
arch/x86/include/asm/idtentry.h +0 -4
index 749c7411d2f1..758f6a2838a8 100644
@@ -745,10 +745,6 @@ DECLARE_IDTENTRY_SYSVEC(IRQ_WORK_VECTOR,        sysvec_irq_work);
 DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_VECTOR,        sysvec_kvm_posted_intr_ipi);
 DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_WAKEUP_VECTOR,    sysvec_kvm_posted_intr_wakeup_ipi);
 DECLARE_IDTENTRY_SYSVEC(POSTED_INTR_NESTED_VECTOR,    sysvec_kvm_posted_intr_nested_ipi);
-#else
-# define fred_sysvec_kvm_posted_intr_ipi        NULL
-# define fred_sysvec_kvm_posted_intr_wakeup_ipi        NULL
-# define fred_sysvec_kvm_posted_intr_nested_ipi        NULL
 #endif
 #if IS_ENABLED(CONFIG_HYPERV)
 
Quoted 3 lines

and it seems to be a net improvement to me. The #ifs match in the .h and .c files, and there are no unnecessary initializers in the sysvec_table.

Ok, I'll pick up Max' patch tomorrow and we must remember to tell Linus during the merge window about this.

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#14 lore raw
Quoted 7 lines · Borislav Petkov
Quoted 3 lines · Paolo Bonzini

and it seems to be a net improvement to me. The #ifs match in the .h and .c files, and there are no unnecessary initializers in the sysvec_table.

Ok, I'll pick up Max' patch tomorrow and we must remember to tell Linus during the merge window about this.

No. Don't.

This pointless #ifdeffery in the vector header needs to vanish from the KVM tree.

Why would you take the #ifdef mess into tasteful code just because someone decided that #ifdeffing out constants in a header which is maintained by other people is a brilliant idea?

The #ifdeffery in the idtentry header is unavoidable and the extra NULL defines are at the right place and not making the actual code unreadable.

Thanks,

tglx