linux-kernel - Re: [PATCH] arch/x86/entry_fred: don't set up KVM IRQs if KVM is disabled

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 16 Feb 2024 09:47:24 -0800
From: Xin Li <xin@...or.com>
To: Paolo Bonzini <pbonzini@...hat.com>,
        Sean Christopherson <seanjc@...gle.com>,
        Max Kellermann <max.kellermann@...os.com>
Cc: hpa@...or.com, x86@...nel.org, linux-kernel@...r.kernel.org,
        Stephen Rothwell <sfr@...b.auug.org.au>, kvm@...r.kernel.org
Subject: Re: [PATCH] arch/x86/entry_fred: don't set up KVM IRQs if KVM is
 disabled

On 2/16/2024 9:41 AM, Xin Li wrote:
> On 2/15/2024 10:31 PM, Paolo Bonzini wrote:
>> On 2/16/24 03:10, Xin Li wrote:
>>> On 2/15/2024 11:55 AM, Sean Christopherson wrote:
>>>> +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.
>>>>
>>>> On Thu, Feb 15, 2024, Max Kellermann wrote:
>>>>> 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")
>>>>> Signed-off-by: Max Kellermann <max.kellermann@...os.com>
>>>>> ---
>>>>>   arch/x86/entry/entry_fred.c | 2 ++
>>>>>   1 file changed, 2 insertions(+)
>>>>>
>>>>> diff --git a/arch/x86/entry/entry_fred.c b/arch/x86/entry/entry_fred.c
>>>>> index ac120cbdaaf2..660b7f7f9a79 100644
>>>>> --- a/arch/x86/entry/entry_fred.c
>>>>> +++ b/arch/x86/entry/entry_fred.c
>>>>> @@ -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;
>>>>> -- 
>>>>> 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).
>>
>> 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.
> 
>> 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
>>
>> diff --git a/arch/x86/entry/entry_fred.c b/arch/x86/entry/entry_fred.c
>> index ac120cbdaaf2..660b7f7f9a79 100644
>> --- a/arch/x86/entry/entry_fred.c
>> +++ b/arch/x86/entry/entry_fred.c
>> @@ -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;
>> diff --git a/arch/x86/include/asm/idtentry.h 
>> b/arch/x86/include/asm/idtentry.h
>> index 749c7411d2f1..758f6a2838a8 100644
>> --- a/arch/x86/include/asm/idtentry.h
>> +++ b/arch/x86/include/asm/idtentry.h
>> @@ -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.
>>
> 
> 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);             \
}