Re: [PATCH] rust: kernel: Support more jump_label api

[Chen Miao]

On 11/5/2025 10:13 PM, Alice Ryhl wrote:
> On Wed, Nov 05, 2025 at 09:55:27PM +0800, Chen Miao wrote:
>> On 11/5/2025 9:38 PM, Alice Ryhl wrote:
>>> On Wed, Nov 05, 2025 at 09:28:39PM +0800, Chen Miao wrote:
>>>> On 11/4/2025 10:07 PM, Alice Ryhl wrote:
>>>>> On Tue, Nov 04, 2025 at 02:04:17AM +0000, chenmiao wrote:
>>>>>> The initial implementation of arch_static_branch was achieved by accessing
>>>>>> the offset from the original type. However, this approach extended the
>>>>>> path and introduced redundant calculations when dealing with types like
>>>>>> `static_key_true/false`, as shown below:
>>>>>>
>>>>>> ```
>>>>>> static_brach_unlikely(tp, tracepoint, key)
>>>>>>      => tracepoint->key->key
>>>>>>      => &tracepoint->key(static_key_false) == &tracepoint->key.key(static_key)
>>>>>>      => off: tracepoint->key - tracepoint
>>>>>> ```
>>>>>>
>>>>>> In practice, the implementation of `arch_static_branch` overlooked many
>>>>>> detailed descriptions. To improve clarity, additional comments have been
>>>>>> added to the original logic. The approach has been modified to directly
>>>>>> locate the corresponding `static_key` instead of using offsets, thereby
>>>>>> reducing  computational overhead.
>>>>>>
>>>>>> If finding the offset from the primitive type is necessary for this
>>>>>> implementation, I will abandon this change.
>>>>>>
>>>>>> Additionally, support for the `static_branch_enable/disable` APIs has been
>>>>>> introduced.
>>>>>>
>>>>>> Signed-off-by: chenmiao <chenmiao@xxxxxxxxxxxxx>
>>>>>> +    ($basety:ty, $branch:expr, $key:path, $keytyp:ty, $field:ident) => {{
>>>>>>             let _key: *const $keytyp = ::core::ptr::addr_of!($key);
>>>>>> -        let _key: *const $crate::bindings::static_key_false = ::core::ptr::addr_of!((*_key).$field);
>>>>>> +        let _key: *const $basety = ::core::ptr::addr_of!((*_key).$field);
>>>>>>             let _key: *const $crate::bindings::static_key = _key.cast();
>>>>>>     
>>>>>>             #[cfg(not(CONFIG_JUMP_LABEL))]
>>>>>> @@ -30,7 +71,88 @@ macro_rules! static_branch_unlikely {
>>>>>>             }
>>>>>>     
>>>>>>             #[cfg(CONFIG_JUMP_LABEL)]
>>>>>> -        $crate::jump_label::arch_static_branch! { $key, $keytyp, $field, false }
>>>>>> +        {
>>>>>> +            $crate::jump_label::arch_static_branch! { _key, $branch }
>>>>>> +        }
>>>>> So ... this is changing from $key to _key. That's replacing the global
>>>>> variable with a local variable holding a pointer to the global variable.
>>>>> However, the arch_static_branch! macro uses the `sym` operand which
>>>>> requires you to pass it the global directly.
>>>>>
>>>>> Did you try this code? I don't believe it will work.
>>>>>
>>>>> Alice
>>>> I'm very sorry for making a fatal mistake. My misunderstanding of sym led to
>>>> this issue, so I shouldn't make any changes to that part. However, regarding
>>>> the other modifications, I believe it is necessary to support the direct
>>>> passing of structures similar to `static_key_false`, just as in C language.
>>> It sounds like you are adding a new use-case for this macro. Can you
>>> provide more information for this new feature? It is currently unclear
>>> to me exactly how this will be used.
>>>
>>> Alice
>> If there's a binding-required driver implementation in the future where a key
>> function uses if (static_branch_unlikely(&zoned_enabled))— defined by
>> DEFINE_STATIC_KEY_FALSE(zoned_enabled);— then in Rust we can directly
>> implement it using if static_branch_unlikely!(zoned_enabled), without having
>> to call it via if static_branch_unlikely!(zoned_enabled,
>> bindings::static_key_false, key).
>>
>> static_branch_unlikely!(zoned_enabled) instead of
>> static_branch_unlikely!(zoned_enabled, bindings::static_key_false, key).
> In general, you would never use "static_key_false" as the second
> argument to static_branch_unlikely!. The second argument is the name of
> the struct *containing* a field of type static_key_false.
>
> I guess your point is that there's no way to use the macro right now if
> the global is a bare static_key_false that is not wrapped in a struct?
>
> Alice
Yes, you're right. But in fact, currently there are no other uses of 
static_branch_unlikely in Rust for Linux (except for tracepoint), so this is 
also a bit awkward.