On Fri, 18 Oct 2024 14:31:43 +0200
Przemek Kitszel <przemyslaw.kitszel@xxxxxxxxx> wrote:
> On 10/18/24 13:15, Peter Zijlstra wrote:
> > On Tue, Oct 01, 2024 at 05:30:18PM -0500, David Lechner wrote:
> >> Add a new if_not_cond_guard() macro to cleanup.h for handling
> >> conditional guards such as mutext_trylock().
> >>
> >> This is more ergonomic than scoped_cond_guard() for most use cases.
> >> Instead of hiding the error handling statement in the macro args, it
> >> works like a normal if statement and allow the error path to be indented
> >> while the normal code flow path is not indented. And it avoid unwanted
> >> side-effect from hidden for loop in scoped_cond_guard().
> >>
> >> Signed-off-by: David Lechner <dlechner@xxxxxxxxxxxx>
>
> So this is guard()() with error handler for cond class of locks.
> I would name such guard_or_err(), or guard_or_err_block(), to make it
> obvious why there is a block attached (so bad we could not ENFORCE that
> there is a block atached).
>
> Then, having it, it would make sense to not only limit guard_or_err() to
> cond class of locks, but also forbid plain guard() with cond locks
> (instead just discouraging it in the doc).
>
> >> ---
> >> include/linux/cleanup.h | 11 +++++++++++
> >> 1 file changed, 11 insertions(+)
> >>
> >> diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> >> index 038b2d523bf8..682bb3fadfc9 100644
> >> --- a/include/linux/cleanup.h
> >> +++ b/include/linux/cleanup.h
> >> @@ -273,6 +273,10 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> >> * an anonymous instance of the (guard) class, not recommended for
> >> * conditional locks.
> >> *
> >> + * if_not_cond_guard(name, args...) { <error handling> }:
> >> + * convenience macro for conditional guards that calls the statement that
> >> + * follows only if the lock was not acquired (typically an error return).
> >> + *
> >> * scoped_guard (name, args...) { }:
> >> * similar to CLASS(name, scope)(args), except the variable (with the
> >> * explicit name 'scope') is declard in a for-loop such that its scope is
> >> @@ -304,6 +308,13 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> >>
> >> #define __guard_ptr(_name) class_##_name##_lock_ptr
> >>
> >> +#define __if_not_cond_guard(_name, _id, args...) \
> >> + CLASS(_name, _id)(args); \
> >> + if (!__guard_ptr(_name)(&_id))
> >> +
> >> +#define if_not_cond_guard(_name, args...) \
> >> + __if_not_cond_guard(_name, __UNIQUE_ID(guard), args)
> >> +
> >> #define scoped_guard(_name, args...) \
> >> for (CLASS(_name, scope)(args), \
> >> *done = NULL; __guard_ptr(_name)(&scope) && !done; done = (void *)1)
> >
> >
> > So if I stick this on top of:
> >
> > https://lkml.kernel.org/r/20241011121535.28049-1-przemyslaw.kitszel@xxxxxxxxx
>
> I have v4 that fixes non-cond version. Apologies it took me that long.
> [v4]
> https://lore.kernel.org/netdev/20241018113823.171256-1-przemyslaw.kitszel@xxxxxxxxx
>
> I have tested it also with the unrechable() calls removed, as suggested
> by David Lechner here:
> https://lore.kernel.org/netdev/0f4786e9-d738-435d-afb9-8c0c4a028ddb@xxxxxxxxxxxx
>
> >
> > then I can add the below:
> >
> > --- a/include/linux/cleanup.h
> > +++ b/include/linux/cleanup.h
> > @@ -277,6 +277,8 @@ static inline class_##_name##_t class_##
> > * convenience macro for conditional guards that calls the statement that
> > * follows only if the lock was not acquired (typically an error return).
> > *
> > + * Only for conditional locks.
> > + *
> > * scoped_guard (name, args...) { }:
> > * similar to CLASS(name, scope)(args), except the variable (with the
> > * explicit name 'scope') is declard in a for-loop such that its scope is
> > @@ -290,7 +292,6 @@ static inline class_##_name##_t class_##
> > * acquire fails.
> > *
> > * Only for conditional locks.
> > - *
> > */
> >
> > #define __DEFINE_CLASS_IS_CONDITIONAL(_name, _is_cond) \
> > @@ -342,6 +343,7 @@ _label: \
> > __UNIQUE_ID(label), args)
> >
> > #define __if_not_guard(_name, _id, args...) \
> > + BUILD_BUG_ON(!__is_cond_ptr(_name)); \
> > CLASS(_name, _id)(args); \
> > if (!__guard_ptr(_name)(&_id))
> >
> >
> > That make sense to people?
>
> despite name, looks promising!
>
> >
> > I've queued these two patches:
> >
> > git://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git locking/core
> >
> > But lacking if_not_guard() users, the robot isn't really going to give
> > me much feedback there, I suppose...
>
> Couldn't you just pick the other patches, that use the newly introduced
> macro?
For a test, sure, but there is a lot of ad7380 work in flight and I'd rather
not push that back a cycle for this improvement (nice though it is!)
If it looks good, an immutable branch would be great, or I could just merge
from Peter's tree if that is stable.
Similarly there is a high risk of the CXL code changing for other reasons
this cycle, but same solution would work.
Jonathan
>
>
>
