On Fri, 19 Dec 2025 at 19:39, 'Bart Van Assche' via kasan-dev
<[email protected]> wrote:
> On 12/19/25 7:39 AM, Marco Elver wrote:
> > +#if defined(WARN_CONTEXT_ANALYSIS)
> > +
> > +/*
> > + * These attributes define new context lock (Clang: capability) types.
> > + * Internal only.
> > + */
>
> How can macros be "internal only" that are defined in a header file that
> will be included by almost all kernel code? Please consider changing
> "internal only" into something that is more clear, e.g. "should only be
> used in the macro definitions in this header file".
Sure, comment could be improved.
Let's say they aren't for general use by normal code that just enables
the analysis for checking; for that we define the shorter (retaining
previous names already in use) ones below. But some of these
attributes can and are used by implementing support for some of the
synchronization primitives.
> > +/*
> > + * The below are used to annotate code being checked. Internal only.
> > + */
>
> Same comment here about "internal only".
Sure, can be clarified.
> > +/**
> > + * context_lock_struct() - declare or define a context lock struct
> > + * @name: struct name
> > + *
> > + * Helper to declare or define a struct type that is also a context lock.
> > + *
> > + * .. code-block:: c
> > + *
> > + * context_lock_struct(my_handle) {
> > + * int foo;
> > + * long bar;
> > + * };
> > + *
> > + * struct some_state {
> > + * ...
> > + * };
> > + * // ... declared elsewhere ...
> > + * context_lock_struct(some_state);
> > + *
> > + * Note: The implementation defines several helper functions that can
> > acquire
> > + * and release the context lock.
> > + */
> > +# define context_lock_struct(name, ...)
> > \
> > + struct __ctx_lock_type(name) __VA_ARGS__ name;
> > \
> > + static __always_inline void __acquire_ctx_lock(const struct name
> > *var) \
> > + __attribute__((overloadable)) __no_context_analysis
> > __acquires_ctx_lock(var) { } \
> > + static __always_inline void __acquire_shared_ctx_lock(const struct
> > name *var) \
> > + __attribute__((overloadable)) __no_context_analysis
> > __acquires_shared_ctx_lock(var) { } \
> > + static __always_inline bool __try_acquire_ctx_lock(const struct name
> > *var, bool ret) \
> > + __attribute__((overloadable)) __no_context_analysis
> > __try_acquires_ctx_lock(1, var) \
> > + { return ret; }
> > \
> > + static __always_inline bool __try_acquire_shared_ctx_lock(const
> > struct name *var, bool ret) \
> > + __attribute__((overloadable)) __no_context_analysis
> > __try_acquires_shared_ctx_lock(1, var) \
> > + { return ret; }
> > \
> > + static __always_inline void __release_ctx_lock(const struct name
> > *var) \
> > + __attribute__((overloadable)) __no_context_analysis
> > __releases_ctx_lock(var) { } \
> > + static __always_inline void __release_shared_ctx_lock(const struct
> > name *var) \
> > + __attribute__((overloadable)) __no_context_analysis
> > __releases_shared_ctx_lock(var) { } \
> > + static __always_inline void __assume_ctx_lock(const struct name *var)
> > \
> > + __attribute__((overloadable)) __assumes_ctx_lock(var) { }
> > \
> > + static __always_inline void __assume_shared_ctx_lock(const struct
> > name *var) \
> > + __attribute__((overloadable)) __assumes_shared_ctx_lock(var)
> > { } \
> > + struct name
>
> I'm concerned that the context_lock_struct() macro will make code harder
> to read. Anyone who encounters the context_lock_struct() macro will have
> to look up its definition to learn what it does. I propose to split this
> macro into two macros:
> * One macro that expands into "__ctx_lock_type(name)".
> * A second macro that expands into the rest of the above macro.
>
> In other words, instead of having to write
> context_lock_struct(struct_name, { ... }); developers will have to write
>
> struct context_lock_type struct_name {
> ...;
> };
> context_struct_helper_functions(struct_name);
This doesn't necessarily help with not having to look up its
definition to learn what it does.
If this is the common pattern, it will blindly be repeated, and this
adds 1 more line and makes this a bit more verbose. Maybe the helper
functions aren't always needed, but I also think that context lock
types should remain relatively few. For all synchronization
primitives that were enabled in this series, the helpers are required.
The current usage is simply:
context_lock_struct(name) {
... struct goes here ...
}; // note no awkward ) brace
I don't know which way the current kernel style is leaning towards,
but if we take <linux/cleanup.h> as an example, a simple programming
model / API is actually preferred.
> My opinion is that the alternative that I'm proposing is easier to read.
> Additionally, it doesn't break existing tools that support jumping from
> the name of a struct to its definition, e.g. ctags and etags.
>
> > +config WARN_CONTEXT_ANALYSIS_ALL
> > + bool "Enable context analysis for all source files"
> > + depends on WARN_CONTEXT_ANALYSIS
> > + depends on EXPERT && !COMPILE_TEST
> > + help
> > + Enable tree-wide context analysis. This is likely to produce a
> > + large number of false positives - enable at your own risk.
> > +
> > + If unsure, say N.
>
> Why !COMPILE_TEST?
That's the idiomatic way to prevent this being enabled in allyesconfig
builds, and other compile-only random configs enabling this and then
stumbling over 1000s of warnings.
Thanks,
-- Marco
