[PATCH] cpufreq/amd-pstate: Fix per-policy boost flag incorrect when fail

[PATCH] cpufreq/amd-pstate: Fix per-policy boost flag incorrect when fail

[Lifeng Zheng]

Commit c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision
boost state") sets per-policy boost flag to false when boost fail.
However, this boost flag will be set to reverse value in
store_local_boost() and cpufreq_boost_trigger_state() in cpufreq.c. This
will cause the per-policy boost flag set to true when fail to set boost.
Remove the extra assignment in amd_pstate_set_boost() and keep all
operations on per-policy boost flag outside of set_boost() to fix this
problem.

Fixes: c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision boost state")
Signed-off-by: Lifeng Zheng <zhenglifeng1@huawei.com>
---
 drivers/cpufreq/amd-pstate.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
index 66e5dfc711c0..4ce923788f3a 100644
--- a/drivers/cpufreq/amd-pstate.c
+++ b/drivers/cpufreq/amd-pstate.c
@@ -730,7 +730,6 @@ static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
 	mutex_lock(&amd_pstate_driver_lock);
 	ret = amd_pstate_cpu_boost_update(policy, state);
 	WRITE_ONCE(cpudata->boost_state, !ret ? state : false);
-	policy->boost_enabled = !ret ? state : false;
 	refresh_frequency_limits(policy);
 	mutex_unlock(&amd_pstate_driver_lock);
 
-- 
2.33.0

Re: [PATCH] cpufreq/amd-pstate: Fix per-policy boost flag incorrect when fail

[Mario Limonciello]

On 1/3/2025 01:41, Lifeng Zheng wrote:
> Commit c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision
> boost state") sets per-policy boost flag to false when boost fail.
> However, this boost flag will be set to reverse value in
> store_local_boost() and cpufreq_boost_trigger_state() in cpufreq.c. This
> will cause the per-policy boost flag set to true when fail to set boost.
> Remove the extra assignment in amd_pstate_set_boost() and keep all
> operations on per-policy boost flag outside of set_boost() to fix this
> problem.
> 
> Fixes: c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision boost state")
> Signed-off-by: Lifeng Zheng <zhenglifeng1@huawei.com>
> ---

Hi There,

Thanks for the patch.  Unfortunately, it doesn't apply to the current 
linux-next branch at 
https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git

Although the issue you identified is still valid, there have been other 
contextual changes in the function [1].  Can you rebase on that branch, 
test it again and send a v2?

[1] 
https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git/tree/drivers/cpufreq/amd-pstate.c?h=linux-next#n750

Thanks!

>   drivers/cpufreq/amd-pstate.c | 1 -
>   1 file changed, 1 deletion(-)
> 
> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
> index 66e5dfc711c0..4ce923788f3a 100644
> --- a/drivers/cpufreq/amd-pstate.c
> +++ b/drivers/cpufreq/amd-pstate.c
> @@ -730,7 +730,6 @@ static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
>   	mutex_lock(&amd_pstate_driver_lock);
>   	ret = amd_pstate_cpu_boost_update(policy, state);
>   	WRITE_ONCE(cpudata->boost_state, !ret ? state : false);
> -	policy->boost_enabled = !ret ? state : false;
>   	refresh_frequency_limits(policy);
>   	mutex_unlock(&amd_pstate_driver_lock);
>   

Re: [PATCH] cpufreq/amd-pstate: Fix per-policy boost flag incorrect when fail

[zhenglifeng (A)]

On 2025/1/4 0:56, Mario Limonciello wrote:

> On 1/3/2025 01:41, Lifeng Zheng wrote:
>> Commit c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision
>> boost state") sets per-policy boost flag to false when boost fail.
>> However, this boost flag will be set to reverse value in
>> store_local_boost() and cpufreq_boost_trigger_state() in cpufreq.c. This
>> will cause the per-policy boost flag set to true when fail to set boost.
>> Remove the extra assignment in amd_pstate_set_boost() and keep all
>> operations on per-policy boost flag outside of set_boost() to fix this
>> problem.
>>
>> Fixes: c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision boost state")
>> Signed-off-by: Lifeng Zheng <zhenglifeng1@huawei.com>
>> ---
> 
> Hi There,
> 
> Thanks for the patch.  Unfortunately, it doesn't apply to the current linux-next branch at https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git
> 
> Although the issue you identified is still valid, there have been other contextual changes in the function [1].  Can you rebase on that branch, test it again and send a v2?
> 
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git/tree/drivers/cpufreq/amd-pstate.c?h=linux-next#n750
> 
> Thanks!

Thanks, I'll test on this branch and send a v2.

> 
>>   drivers/cpufreq/amd-pstate.c | 1 -
>>   1 file changed, 1 deletion(-)
>>
>> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
>> index 66e5dfc711c0..4ce923788f3a 100644
>> --- a/drivers/cpufreq/amd-pstate.c
>> +++ b/drivers/cpufreq/amd-pstate.c
>> @@ -730,7 +730,6 @@ static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
>>       mutex_lock(&amd_pstate_driver_lock);
>>       ret = amd_pstate_cpu_boost_update(policy, state);
>>       WRITE_ONCE(cpudata->boost_state, !ret ? state : false);
>> -    policy->boost_enabled = !ret ? state : false;
>>       refresh_frequency_limits(policy);
>>       mutex_unlock(&amd_pstate_driver_lock);
>>   
> 

Re: [PATCH] cpufreq/amd-pstate: Fix per-policy boost flag incorrect when fail

[zhenglifeng (A)]

On 2025/1/4 0:56, Mario Limonciello wrote:

> On 1/3/2025 01:41, Lifeng Zheng wrote:
>> Commit c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision
>> boost state") sets per-policy boost flag to false when boost fail.
>> However, this boost flag will be set to reverse value in
>> store_local_boost() and cpufreq_boost_trigger_state() in cpufreq.c. This
>> will cause the per-policy boost flag set to true when fail to set boost.
>> Remove the extra assignment in amd_pstate_set_boost() and keep all
>> operations on per-policy boost flag outside of set_boost() to fix this
>> problem.
>>
>> Fixes: c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision boost state")
>> Signed-off-by: Lifeng Zheng <zhenglifeng1@huawei.com>
>> ---
> 
> Hi There,
> 
> Thanks for the patch.  Unfortunately, it doesn't apply to the current linux-next branch at https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git

Hello Mario,

When I try to build on this branch, I got an error:

arch/x86/kernel/relocate_kernel_64.o: In function `virtual_mapped':
.../linux/arch/x86/kernel/relocate_kernel_64.S:249: undefined reference to `saved_context_gdt_desc'
scripts/Makefile.vmlinux:77: recipe for target 'vmlinux' failed

This error occurs when CONFIG_KEXEC_JUMP=y, and doesn't occur when build on
torvalds master branch with same config. Please check if there is any
problem whith this branch.

> 
> Although the issue you identified is still valid, there have been other contextual changes in the function [1].  Can you rebase on that branch, test it again and send a v2?
> 
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git/tree/drivers/cpufreq/amd-pstate.c?h=linux-next#n750
> 
> Thanks!
> 
>>   drivers/cpufreq/amd-pstate.c | 1 -
>>   1 file changed, 1 deletion(-)
>>
>> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
>> index 66e5dfc711c0..4ce923788f3a 100644
>> --- a/drivers/cpufreq/amd-pstate.c
>> +++ b/drivers/cpufreq/amd-pstate.c
>> @@ -730,7 +730,6 @@ static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
>>       mutex_lock(&amd_pstate_driver_lock);
>>       ret = amd_pstate_cpu_boost_update(policy, state);
>>       WRITE_ONCE(cpudata->boost_state, !ret ? state : false);
>> -    policy->boost_enabled = !ret ? state : false;
>>       refresh_frequency_limits(policy);
>>       mutex_unlock(&amd_pstate_driver_lock);
>>   
> 

Re: [PATCH] cpufreq/amd-pstate: Fix per-policy boost flag incorrect when fail

[Mario Limonciello]

On 1/7/2025 19:31, zhenglifeng (A) wrote:
> On 2025/1/4 0:56, Mario Limonciello wrote:
> 
>> On 1/3/2025 01:41, Lifeng Zheng wrote:
>>> Commit c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision
>>> boost state") sets per-policy boost flag to false when boost fail.
>>> However, this boost flag will be set to reverse value in
>>> store_local_boost() and cpufreq_boost_trigger_state() in cpufreq.c. This
>>> will cause the per-policy boost flag set to true when fail to set boost.
>>> Remove the extra assignment in amd_pstate_set_boost() and keep all
>>> operations on per-policy boost flag outside of set_boost() to fix this
>>> problem.
>>>
>>> Fixes: c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision boost state")
>>> Signed-off-by: Lifeng Zheng <zhenglifeng1@huawei.com>
>>> ---
>>
>> Hi There,
>>
>> Thanks for the patch.  Unfortunately, it doesn't apply to the current linux-next branch at https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git
> 
> Hello Mario,
> 
> When I try to build on this branch, I got an error:
> 
> arch/x86/kernel/relocate_kernel_64.o: In function `virtual_mapped':
> .../linux/arch/x86/kernel/relocate_kernel_64.S:249: undefined reference to `saved_context_gdt_desc'
> scripts/Makefile.vmlinux:77: recipe for target 'vmlinux' failed
> 
> This error occurs when CONFIG_KEXEC_JUMP=y, and doesn't occur when build on
> torvalds master branch with same config. Please check if there is any
> problem whith this branch.
> 
Hi,

It's because the branch is based on an earlier 6.13-rc.

Two ideas that can help you:

1) You can pull this patch manually on top of it to avoid that issue.
https://git.kernel.org/torvalds/c/aeb68937614f4

2) You can manually rebase the branch on newer 6.13-rc locally to make 
your commit.  That commit that fixed it landed in 6.13-rc3, so rc3 or 
later would be fine.

>>
>> Although the issue you identified is still valid, there have been other contextual changes in the function [1].  Can you rebase on that branch, test it again and send a v2?
>>
>> [1] https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git/tree/drivers/cpufreq/amd-pstate.c?h=linux-next#n750
>>
>> Thanks!
>>
>>>    drivers/cpufreq/amd-pstate.c | 1 -
>>>    1 file changed, 1 deletion(-)
>>>
>>> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
>>> index 66e5dfc711c0..4ce923788f3a 100644
>>> --- a/drivers/cpufreq/amd-pstate.c
>>> +++ b/drivers/cpufreq/amd-pstate.c
>>> @@ -730,7 +730,6 @@ static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
>>>        mutex_lock(&amd_pstate_driver_lock);
>>>        ret = amd_pstate_cpu_boost_update(policy, state);
>>>        WRITE_ONCE(cpudata->boost_state, !ret ? state : false);
>>> -    policy->boost_enabled = !ret ? state : false;
>>>        refresh_frequency_limits(policy);
>>>        mutex_unlock(&amd_pstate_driver_lock);
>>>    
>>
> 

Re: [PATCH] cpufreq/amd-pstate: Fix per-policy boost flag incorrect when fail

[zhenglifeng (A)]

On 2025/1/8 10:26, Mario Limonciello wrote:

> On 1/7/2025 19:31, zhenglifeng (A) wrote:
>> On 2025/1/4 0:56, Mario Limonciello wrote:
>>
>>> On 1/3/2025 01:41, Lifeng Zheng wrote:
>>>> Commit c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision
>>>> boost state") sets per-policy boost flag to false when boost fail.
>>>> However, this boost flag will be set to reverse value in
>>>> store_local_boost() and cpufreq_boost_trigger_state() in cpufreq.c. This
>>>> will cause the per-policy boost flag set to true when fail to set boost.
>>>> Remove the extra assignment in amd_pstate_set_boost() and keep all
>>>> operations on per-policy boost flag outside of set_boost() to fix this
>>>> problem.
>>>>
>>>> Fixes: c8c68c38b56f ("cpufreq: amd-pstate: initialize core precision boost state")
>>>> Signed-off-by: Lifeng Zheng <zhenglifeng1@huawei.com>
>>>> ---
>>>
>>> Hi There,
>>>
>>> Thanks for the patch.  Unfortunately, it doesn't apply to the current linux-next branch at https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git
>>
>> Hello Mario,
>>
>> When I try to build on this branch, I got an error:
>>
>> arch/x86/kernel/relocate_kernel_64.o: In function `virtual_mapped':
>> .../linux/arch/x86/kernel/relocate_kernel_64.S:249: undefined reference to `saved_context_gdt_desc'
>> scripts/Makefile.vmlinux:77: recipe for target 'vmlinux' failed
>>
>> This error occurs when CONFIG_KEXEC_JUMP=y, and doesn't occur when build on
>> torvalds master branch with same config. Please check if there is any
>> problem whith this branch.
>>
> Hi,
> 
> It's because the branch is based on an earlier 6.13-rc.
> 
> Two ideas that can help you:
> 
> 1) You can pull this patch manually on top of it to avoid that issue.
> https://git.kernel.org/torvalds/c/aeb68937614f4
> 
> 2) You can manually rebase the branch on newer 6.13-rc locally to make your commit.  That commit that fixed it landed in 6.13-rc3, so rc3 or later would be fine.

This solves the problem, thanks!

> 
>>>
>>> Although the issue you identified is still valid, there have been other contextual changes in the function [1].  Can you rebase on that branch, test it again and send a v2?
>>>
>>> [1] https://git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git/tree/drivers/cpufreq/amd-pstate.c?h=linux-next#n750
>>>
>>> Thanks!
>>>
>>>>    drivers/cpufreq/amd-pstate.c | 1 -
>>>>    1 file changed, 1 deletion(-)
>>>>
>>>> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
>>>> index 66e5dfc711c0..4ce923788f3a 100644
>>>> --- a/drivers/cpufreq/amd-pstate.c
>>>> +++ b/drivers/cpufreq/amd-pstate.c
>>>> @@ -730,7 +730,6 @@ static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
>>>>        mutex_lock(&amd_pstate_driver_lock);
>>>>        ret = amd_pstate_cpu_boost_update(policy, state);
>>>>        WRITE_ONCE(cpudata->boost_state, !ret ? state : false);
>>>> -    policy->boost_enabled = !ret ? state : false;
>>>>        refresh_frequency_limits(policy);
>>>>        mutex_unlock(&amd_pstate_driver_lock);
>>>>    
>>>
>>
>