[PATCH] arm64/cpufeature: Convert hook_lock to raw_spin_lock_t in cpu_enable_ssbs()

[PATCH] arm64/cpufeature: Convert hook_lock to raw_spin_lock_t in cpu_enable_ssbs()

[Julien Grall]

cpu_enable_ssbs() is called via stop_machine() as part of the cpu_enable
callback. A spin lock is used to ensure the hook is registered before
the rest of the callback is executed.

On -RT spin_lock() may sleep. However, all the callees in stop_machine()
are expected to not sleep. Therefore a raw_spin_lock() is required here.

Given this is already done under stop_machine() and the work done under
the lock is quite small, the latency should not increase too much.

Signed-off-by: Julien Grall <julien.grall@arm.com>

---

It was noticed when looking at the current use of spin_lock in
arch/arm64. I don't have a platform calling that callback, so I have
hacked the code to reproduce the error and check it is now fixed.
---
 arch/arm64/kernel/cpufeature.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
index ca27e08e3d8a..2a7159fda3ce 100644
--- a/arch/arm64/kernel/cpufeature.c
+++ b/arch/arm64/kernel/cpufeature.c
@@ -1194,14 +1194,14 @@ static struct undef_hook ssbs_emulation_hook = {
 static void cpu_enable_ssbs(const struct arm64_cpu_capabilities *__unused)
 {
 	static bool undef_hook_registered = false;
-	static DEFINE_SPINLOCK(hook_lock);
+	static DEFINE_RAW_SPINLOCK(hook_lock);
 
-	spin_lock(&hook_lock);
+	raw_spin_lock(&hook_lock);
 	if (!undef_hook_registered) {
 		register_undef_hook(&ssbs_emulation_hook);
 		undef_hook_registered = true;
 	}
-	spin_unlock(&hook_lock);
+	raw_spin_unlock(&hook_lock);
 
 	if (arm64_get_ssbd_state() == ARM64_SSBD_FORCE_DISABLE) {
 		sysreg_clear_set(sctlr_el1, 0, SCTLR_ELx_DSSBS);
-- 
2.11.0

Re: [PATCH] arm64/cpufeature: Convert hook_lock to raw_spin_lock_t in cpu_enable_ssbs()

[Will Deacon]

On Thu, May 30, 2019 at 12:30:58PM +0100, Julien Grall wrote:
> cpu_enable_ssbs() is called via stop_machine() as part of the cpu_enable
> callback. A spin lock is used to ensure the hook is registered before
> the rest of the callback is executed.
> 
> On -RT spin_lock() may sleep. However, all the callees in stop_machine()
> are expected to not sleep. Therefore a raw_spin_lock() is required here.
> 
> Given this is already done under stop_machine() and the work done under
> the lock is quite small, the latency should not increase too much.
> 
> Signed-off-by: Julien Grall <julien.grall@arm.com>
> 
> ---
> 
> It was noticed when looking at the current use of spin_lock in
> arch/arm64. I don't have a platform calling that callback, so I have
> hacked the code to reproduce the error and check it is now fixed.
> ---
>  arch/arm64/kernel/cpufeature.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
> index ca27e08e3d8a..2a7159fda3ce 100644
> --- a/arch/arm64/kernel/cpufeature.c
> +++ b/arch/arm64/kernel/cpufeature.c
> @@ -1194,14 +1194,14 @@ static struct undef_hook ssbs_emulation_hook = {
>  static void cpu_enable_ssbs(const struct arm64_cpu_capabilities *__unused)
>  {
>  	static bool undef_hook_registered = false;
> -	static DEFINE_SPINLOCK(hook_lock);
> +	static DEFINE_RAW_SPINLOCK(hook_lock);
>  
> -	spin_lock(&hook_lock);
> +	raw_spin_lock(&hook_lock);
>  	if (!undef_hook_registered) {
>  		register_undef_hook(&ssbs_emulation_hook);
>  		undef_hook_registered = true;
>  	}
> -	spin_unlock(&hook_lock);
> +	raw_spin_unlock(&hook_lock);

Makes sense to me. We could probably avoid the lock entirely if we wanted
to (via atomic_dec_if_positive), but I'm not sure it's really worth it.

Will

Re: [PATCH] arm64/cpufeature: Convert hook_lock to raw_spin_lock_t in cpu_enable_ssbs()

[Julien Grall]

Hi Will,

On 5/30/19 1:01 PM, Will Deacon wrote:
> On Thu, May 30, 2019 at 12:30:58PM +0100, Julien Grall wrote:
>> cpu_enable_ssbs() is called via stop_machine() as part of the cpu_enable
>> callback. A spin lock is used to ensure the hook is registered before
>> the rest of the callback is executed.
>>
>> On -RT spin_lock() may sleep. However, all the callees in stop_machine()
>> are expected to not sleep. Therefore a raw_spin_lock() is required here.
>>
>> Given this is already done under stop_machine() and the work done under
>> the lock is quite small, the latency should not increase too much.
>>
>> Signed-off-by: Julien Grall <julien.grall@arm.com>
>>
>> ---
>>
>> It was noticed when looking at the current use of spin_lock in
>> arch/arm64. I don't have a platform calling that callback, so I have
>> hacked the code to reproduce the error and check it is now fixed.
>> ---
>>   arch/arm64/kernel/cpufeature.c | 6 +++---
>>   1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
>> index ca27e08e3d8a..2a7159fda3ce 100644
>> --- a/arch/arm64/kernel/cpufeature.c
>> +++ b/arch/arm64/kernel/cpufeature.c
>> @@ -1194,14 +1194,14 @@ static struct undef_hook ssbs_emulation_hook = {
>>   static void cpu_enable_ssbs(const struct arm64_cpu_capabilities *__unused)
>>   {
>>   	static bool undef_hook_registered = false;
>> -	static DEFINE_SPINLOCK(hook_lock);
>> +	static DEFINE_RAW_SPINLOCK(hook_lock);
>>   
>> -	spin_lock(&hook_lock);
>> +	raw_spin_lock(&hook_lock);
>>   	if (!undef_hook_registered) {
>>   		register_undef_hook(&ssbs_emulation_hook);
>>   		undef_hook_registered = true;
>>   	}
>> -	spin_unlock(&hook_lock);
>> +	raw_spin_unlock(&hook_lock);
> 
> Makes sense to me. We could probably avoid the lock entirely if we wanted
> to (via atomic_dec_if_positive), but I'm not sure it's really worth it.

I would prefer to remove the lock if it is possible. However, I was 
under the impression the lock is necessary so the hook is registered 
before any CPU attempt to configure the PSTATE.

Cheers,

-- 
Julien Grall

Re: [PATCH] arm64/cpufeature: Convert hook_lock to raw_spin_lock_t in cpu_enable_ssbs()

[Catalin Marinas]

On Thu, May 30, 2019 at 12:30:58PM +0100, Julien Grall wrote:
> cpu_enable_ssbs() is called via stop_machine() as part of the cpu_enable
> callback. A spin lock is used to ensure the hook is registered before
> the rest of the callback is executed.
> 
> On -RT spin_lock() may sleep. However, all the callees in stop_machine()
> are expected to not sleep. Therefore a raw_spin_lock() is required here.
> 
> Given this is already done under stop_machine() and the work done under
> the lock is quite small, the latency should not increase too much.
> 
> Signed-off-by: Julien Grall <julien.grall@arm.com>

Queued for 5.3. Thanks.

-- 
Catalin