2024-02-05 14:26:40

by Fabio M. De Francesco

[permalink] [raw]
Subject: [PATCH 0/2 v2] Add cond_guard() to conditional guards

Add cond_guard() macro to conditional guards and use it to replace an
open-coded up_read() in show_targetN() and remove a block marked by an
'out' label.

Cc: Peter Zijlstra <[email protected]>
Cc: Jonathan Cameron <[email protected]>
Suggested-by: Dan Williams <[email protected]>
Suggested-by: Ira Weiny <[email protected]>
Signed-off-by: Fabio M. De Francesco <[email protected]>
---

Changes from RFC v4:
Changed the interface of cond_guard() to take a variable to store
a return code, the succes code and failure code, to enable a
later check of the returned code in that variable.
Changes from RFC v5:
Changed the interface of cond_guard() to take one or more statements
in its second argument to conform to Dan's final suggestion
(thanks).
Changes from v1:
Fixed a grammar error in the commit message of 1/2; replaced the
name of the second argument of cond_guard() with '_fail'
according to Jonathan's comments (thanks).

Fabio M. De Francesco (2):
cleanup: Add cond_guard() to conditional guards
cxl/region: Use cond_guard() in show_targetN()

drivers/cxl/core/region.c | 16 ++++------------
include/linux/cleanup.h | 14 ++++++++++++++
2 files changed, 18 insertions(+), 12 deletions(-)

--
2.43.0



2024-02-05 14:26:59

by Fabio M. De Francesco

[permalink] [raw]
Subject: [PATCH 1/2 v2] cleanup: Add cond_guard() to conditional guards

Add cond_guard() macro to conditional guards.

cond_guard() is a guard to be used with the conditional variants of locks,
like down_read_trylock() or mutex_lock_interruptible().

It takes a statement (or more statements in a block) that is passed to its
second argument. That statement (or block) is executed if waiting for a
lock is interrupted or if a _trylock() fails in case of contention.

Usage example:

cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);

Consistently with the other guards, locks are unlocked at the exit of the
scope where cond_guard() is called.

Cc: Peter Zijlstra <[email protected]>
Suggested-by: Dan Williams <[email protected]>
Suggested-by: Ira Weiny <[email protected]>
Signed-off-by: Fabio M. De Francesco <[email protected]>
---
include/linux/cleanup.h | 14 ++++++++++++++
1 file changed, 14 insertions(+)

diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
index c2d09bc4f976..88af56600325 100644
--- a/include/linux/cleanup.h
+++ b/include/linux/cleanup.h
@@ -134,6 +134,16 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
* an anonymous instance of the (guard) class, not recommended for
* conditional locks.
*
+ * cond_guard(name, fail, args...):
+ * a guard to be used with the conditional variants of locks, like
+ * down_read_trylock() or mutex_lock_interruptible. 'fail' are one or more
+ * statements that are executed when waiting for a lock is interrupted or
+ * when a _trylock() fails in case of contention.
+ *
+ * Example:
+ *
+ * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
+ *
* 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
@@ -165,6 +175,10 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \

#define __guard_ptr(_name) class_##_name##_lock_ptr

+#define cond_guard(_name, _fail, args...) \
+ CLASS(_name, scope)(args); \
+ if (!__guard_ptr(_name)(&scope)) _fail
+
#define scoped_guard(_name, args...) \
for (CLASS(_name, scope)(args), \
*done = NULL; __guard_ptr(_name)(&scope) && !done; done = (void *)1)
--
2.43.0


2024-02-05 14:27:11

by Fabio M. De Francesco

[permalink] [raw]
Subject: [PATCH 2/2 v2] cxl/region: Use cond_guard() in show_targetN()

Use cond_guard() in show_target() to not open code an up_read() in an 'out'
block. If the down_read_interruptible() fails, the statement passed to the
second argument of cond_guard() returns -EINTR.

Cc: Peter Zijlstra <[email protected]>
Suggested-by: Dan Williams <[email protected]>
Suggested-by: Ira Weiny <[email protected]>
Reviewed-by: Jonathan Cameron <[email protected]>
Signed-off-by: Fabio M. De Francesco <[email protected]>
---
drivers/cxl/core/region.c | 16 ++++------------
1 file changed, 4 insertions(+), 12 deletions(-)

diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
index 0f05692bfec3..bd3236786a25 100644
--- a/drivers/cxl/core/region.c
+++ b/drivers/cxl/core/region.c
@@ -666,28 +666,20 @@ static size_t show_targetN(struct cxl_region *cxlr, char *buf, int pos)
{
struct cxl_region_params *p = &cxlr->params;
struct cxl_endpoint_decoder *cxled;
- int rc;

- rc = down_read_interruptible(&cxl_region_rwsem);
- if (rc)
- return rc;
+ cond_guard(rwsem_read_intr, return -EINTR, &cxl_region_rwsem);

if (pos >= p->interleave_ways) {
dev_dbg(&cxlr->dev, "position %d out of range %d\n", pos,
p->interleave_ways);
- rc = -ENXIO;
- goto out;
+ return -ENXIO;
}

cxled = p->targets[pos];
if (!cxled)
- rc = sysfs_emit(buf, "\n");
+ return sysfs_emit(buf, "\n");
else
- rc = sysfs_emit(buf, "%s\n", dev_name(&cxled->cxld.dev));
-out:
- up_read(&cxl_region_rwsem);
-
- return rc;
+ return sysfs_emit(buf, "%s\n", dev_name(&cxled->cxld.dev));
}

static int match_free_decoder(struct device *dev, void *data)
--
2.43.0


2024-02-05 14:29:12

by Jonathan Cameron

[permalink] [raw]
Subject: Re: [PATCH 1/2 v2] cleanup: Add cond_guard() to conditional guards

On Mon, 5 Feb 2024 15:26:12 +0100
"Fabio M. De Francesco" <[email protected]> wrote:

> Add cond_guard() macro to conditional guards.
>
> cond_guard() is a guard to be used with the conditional variants of locks,
> like down_read_trylock() or mutex_lock_interruptible().
>
> It takes a statement (or more statements in a block) that is passed to its
> second argument. That statement (or block) is executed if waiting for a
> lock is interrupted or if a _trylock() fails in case of contention.
>
> Usage example:
>
> cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
>
> Consistently with the other guards, locks are unlocked at the exit of the
> scope where cond_guard() is called.
>
> Cc: Peter Zijlstra <[email protected]>
> Suggested-by: Dan Williams <[email protected]>
> Suggested-by: Ira Weiny <[email protected]>
> Signed-off-by: Fabio M. De Francesco <[email protected]>

This version looks good to me, but these are still fairly new to me so good to get
inputs from others.

Reviewed-by: Jonathan Cameron <[email protected]>

> ---
> include/linux/cleanup.h | 14 ++++++++++++++
> 1 file changed, 14 insertions(+)
>
> diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> index c2d09bc4f976..88af56600325 100644
> --- a/include/linux/cleanup.h
> +++ b/include/linux/cleanup.h
> @@ -134,6 +134,16 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> * an anonymous instance of the (guard) class, not recommended for
> * conditional locks.
> *
> + * cond_guard(name, fail, args...):
> + * a guard to be used with the conditional variants of locks, like
> + * down_read_trylock() or mutex_lock_interruptible. 'fail' are one or more
> + * statements that are executed when waiting for a lock is interrupted or
> + * when a _trylock() fails in case of contention.
> + *
> + * Example:
> + *
> + * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> + *
> * 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
> @@ -165,6 +175,10 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
>
> #define __guard_ptr(_name) class_##_name##_lock_ptr
>
> +#define cond_guard(_name, _fail, args...) \
> + CLASS(_name, scope)(args); \
> + if (!__guard_ptr(_name)(&scope)) _fail
> +
> #define scoped_guard(_name, args...) \
> for (CLASS(_name, scope)(args), \
> *done = NULL; __guard_ptr(_name)(&scope) && !done; done = (void *)1)


2024-02-05 17:14:34

by Dave Jiang

[permalink] [raw]
Subject: Re: [PATCH 1/2 v2] cleanup: Add cond_guard() to conditional guards



On 2/5/24 7:26 AM, Fabio M. De Francesco wrote:
> Add cond_guard() macro to conditional guards.
>
> cond_guard() is a guard to be used with the conditional variants of locks,
> like down_read_trylock() or mutex_lock_interruptible().
>
> It takes a statement (or more statements in a block) that is passed to its
> second argument. That statement (or block) is executed if waiting for a
> lock is interrupted or if a _trylock() fails in case of contention.
>
> Usage example:
>
> cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
>
> Consistently with the other guards, locks are unlocked at the exit of the
> scope where cond_guard() is called.
>
> Cc: Peter Zijlstra <[email protected]>
> Suggested-by: Dan Williams <[email protected]>
> Suggested-by: Ira Weiny <[email protected]>
> Signed-off-by: Fabio M. De Francesco <[email protected]>

Reviewed-by: Dave Jiang <[email protected]>

> ---
> include/linux/cleanup.h | 14 ++++++++++++++
> 1 file changed, 14 insertions(+)
>
> diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> index c2d09bc4f976..88af56600325 100644
> --- a/include/linux/cleanup.h
> +++ b/include/linux/cleanup.h
> @@ -134,6 +134,16 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> * an anonymous instance of the (guard) class, not recommended for
> * conditional locks.
> *
> + * cond_guard(name, fail, args...):
> + * a guard to be used with the conditional variants of locks, like
> + * down_read_trylock() or mutex_lock_interruptible. 'fail' are one or more
> + * statements that are executed when waiting for a lock is interrupted or
> + * when a _trylock() fails in case of contention.
> + *
> + * Example:
> + *
> + * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> + *
> * 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
> @@ -165,6 +175,10 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
>
> #define __guard_ptr(_name) class_##_name##_lock_ptr
>
> +#define cond_guard(_name, _fail, args...) \
> + CLASS(_name, scope)(args); \
> + if (!__guard_ptr(_name)(&scope)) _fail
> +
> #define scoped_guard(_name, args...) \
> for (CLASS(_name, scope)(args), \
> *done = NULL; __guard_ptr(_name)(&scope) && !done; done = (void *)1)

2024-02-05 17:23:12

by Dave Jiang

[permalink] [raw]
Subject: Re: [PATCH 2/2 v2] cxl/region: Use cond_guard() in show_targetN()



On 2/5/24 7:26 AM, Fabio M. De Francesco wrote:
> Use cond_guard() in show_target() to not open code an up_read() in an 'out'
> block. If the down_read_interruptible() fails, the statement passed to the
> second argument of cond_guard() returns -EINTR.
>
> Cc: Peter Zijlstra <[email protected]>
> Suggested-by: Dan Williams <[email protected]>
> Suggested-by: Ira Weiny <[email protected]>
> Reviewed-by: Jonathan Cameron <[email protected]>
> Signed-off-by: Fabio M. De Francesco <[email protected]>
> ---
> drivers/cxl/core/region.c | 16 ++++------------
> 1 file changed, 4 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
> index 0f05692bfec3..bd3236786a25 100644
> --- a/drivers/cxl/core/region.c
> +++ b/drivers/cxl/core/region.c
> @@ -666,28 +666,20 @@ static size_t show_targetN(struct cxl_region *cxlr, char *buf, int pos)
> {
> struct cxl_region_params *p = &cxlr->params;
> struct cxl_endpoint_decoder *cxled;
> - int rc;
>
> - rc = down_read_interruptible(&cxl_region_rwsem);
> - if (rc)
> - return rc;
> + cond_guard(rwsem_read_intr, return -EINTR, &cxl_region_rwsem);
>
> if (pos >= p->interleave_ways) {
> dev_dbg(&cxlr->dev, "position %d out of range %d\n", pos,
> p->interleave_ways);
> - rc = -ENXIO;
> - goto out;
> + return -ENXIO;
> }
>
> cxled = p->targets[pos];
> if (!cxled)
> - rc = sysfs_emit(buf, "\n");
> + return sysfs_emit(buf, "\n");
> else

This else isn't needed because your if statement above returns. I think if you run checkpatch it should've flagged this.

> - rc = sysfs_emit(buf, "%s\n", dev_name(&cxled->cxld.dev));
> -out:
> - up_read(&cxl_region_rwsem);
> -
> - return rc;
> + return sysfs_emit(buf, "%s\n", dev_name(&cxled->cxld.dev));
> }
>
> static int match_free_decoder(struct device *dev, void *data)

2024-02-05 19:30:53

by Dan Williams

[permalink] [raw]
Subject: RE: [PATCH 1/2 v2] cleanup: Add cond_guard() to conditional guards

Fabio M. De Francesco wrote:
> Add cond_guard() macro to conditional guards.
>
> cond_guard() is a guard to be used with the conditional variants of locks,
> like down_read_trylock() or mutex_lock_interruptible().
>
> It takes a statement (or more statements in a block) that is passed to its
> second argument. That statement (or block) is executed if waiting for a
> lock is interrupted or if a _trylock() fails in case of contention.
>
> Usage example:
>
> cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
>
> Consistently with the other guards, locks are unlocked at the exit of the
> scope where cond_guard() is called.
>
> Cc: Peter Zijlstra <[email protected]>
> Suggested-by: Dan Williams <[email protected]>
> Suggested-by: Ira Weiny <[email protected]>
> Signed-off-by: Fabio M. De Francesco <[email protected]>
> ---
> include/linux/cleanup.h | 14 ++++++++++++++
> 1 file changed, 14 insertions(+)
>
> diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> index c2d09bc4f976..88af56600325 100644
> --- a/include/linux/cleanup.h
> +++ b/include/linux/cleanup.h
> @@ -134,6 +134,16 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> * an anonymous instance of the (guard) class, not recommended for
> * conditional locks.
> *
> + * cond_guard(name, fail, args...):
> + * a guard to be used with the conditional variants of locks, like
> + * down_read_trylock() or mutex_lock_interruptible. 'fail' are one or more
> + * statements that are executed when waiting for a lock is interrupted or
> + * when a _trylock() fails in case of contention.
> + *
> + * Example:
> + *
> + * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);

That _fail argument likely needs to be a statement expression for the
multi-statement case.

> + *
> * 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
> @@ -165,6 +175,10 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
>
> #define __guard_ptr(_name) class_##_name##_lock_ptr
>
> +#define cond_guard(_name, _fail, args...) \
> + CLASS(_name, scope)(args); \
> + if (!__guard_ptr(_name)(&scope)) _fail

No, as I stated before this is broken for usages of:

if () cond_guard() else if ()

The 'else' in the definition is critical, this builds for me (untested):

diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
index 88af56600325..665407498781 100644
--- a/include/linux/cleanup.h
+++ b/include/linux/cleanup.h
@@ -142,7 +142,7 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
*
* Example:
*
- * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
+ * cond_guard(rwsem_read_try, ({ printk(...); return 0; }), &semaphore);
*
* scoped_guard (name, args...) { }:
* similar to CLASS(name, scope)(args), except the variable (with the
@@ -177,7 +177,8 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \

#define cond_guard(_name, _fail, args...) \
CLASS(_name, scope)(args); \
- if (!__guard_ptr(_name)(&scope)) _fail
+ if (!__guard_ptr(_name)(&scope)) _fail; \
+ else

#define scoped_guard(_name, args...) \
for (CLASS(_name, scope)(args), \

2024-02-05 22:18:19

by Ira Weiny

[permalink] [raw]
Subject: RE: [PATCH 1/2 v2] cleanup: Add cond_guard() to conditional guards

Dan Williams wrote:
> Fabio M. De Francesco wrote:
> > Add cond_guard() macro to conditional guards.
> >
> > cond_guard() is a guard to be used with the conditional variants of locks,
> > like down_read_trylock() or mutex_lock_interruptible().
> >
> > It takes a statement (or more statements in a block) that is passed to its
> > second argument. That statement (or block) is executed if waiting for a
> > lock is interrupted or if a _trylock() fails in case of contention.
> >
> > Usage example:
> >
> > cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> >
> > Consistently with the other guards, locks are unlocked at the exit of the
> > scope where cond_guard() is called.
> >
> > Cc: Peter Zijlstra <[email protected]>
> > Suggested-by: Dan Williams <[email protected]>
> > Suggested-by: Ira Weiny <[email protected]>
> > Signed-off-by: Fabio M. De Francesco <[email protected]>
> > ---
> > include/linux/cleanup.h | 14 ++++++++++++++
> > 1 file changed, 14 insertions(+)
> >
> > diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> > index c2d09bc4f976..88af56600325 100644
> > --- a/include/linux/cleanup.h
> > +++ b/include/linux/cleanup.h
> > @@ -134,6 +134,16 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> > * an anonymous instance of the (guard) class, not recommended for
> > * conditional locks.
> > *
> > + * cond_guard(name, fail, args...):
> > + * a guard to be used with the conditional variants of locks, like
> > + * down_read_trylock() or mutex_lock_interruptible. 'fail' are one or more
> > + * statements that are executed when waiting for a lock is interrupted or
> > + * when a _trylock() fails in case of contention.
> > + *
> > + * Example:
> > + *
> > + * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
>
> That _fail argument likely needs to be a statement expression for the
> multi-statement case.

You mean ({ ... }) as discussed here?

https://lore.kernel.org/all/[email protected]/

>
> > + *
> > * 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
> > @@ -165,6 +175,10 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> >
> > #define __guard_ptr(_name) class_##_name##_lock_ptr
> >
> > +#define cond_guard(_name, _fail, args...) \
> > + CLASS(_name, scope)(args); \
> > + if (!__guard_ptr(_name)(&scope)) _fail
>
> No, as I stated before this is broken for usages of:
>
> if () cond_guard() else if ()
>
> The 'else' in the definition is critical, this builds for me (untested):

I did not test Fabios work directly but I don't understand this example.
It seems like your suggestion does nothing useful. The cond_guard()
becomes a single statement like...

if ()
cond_guard();
else ...

.. And can't protect anything. NOTE From my understanding of
cond_guard() as defined, the ';' must be used as part of cond_guard() and
should complete the internal macro 'if' statement.

I think this would work:

if () {
cond_guard();
... do locked stuff ...
} else ...

>
> diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> index 88af56600325..665407498781 100644
> --- a/include/linux/cleanup.h
> +++ b/include/linux/cleanup.h
> @@ -142,7 +142,7 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> *
> * Example:
> *
> - * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> + * cond_guard(rwsem_read_try, ({ printk(...); return 0; }), &semaphore);
> *
> * scoped_guard (name, args...) { }:
> * similar to CLASS(name, scope)(args), except the variable (with the
> @@ -177,7 +177,8 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
>
> #define cond_guard(_name, _fail, args...) \
> CLASS(_name, scope)(args); \
> - if (!__guard_ptr(_name)(&scope)) _fail
> + if (!__guard_ptr(_name)(&scope)) _fail; \

Building on what I found for scoped_cond_guard() this should be

> + if (!__guard_ptr(_name)(&scope)) { _fail; }

And drop the else. The else needs to clearly be part of an outside if in
your example.

Ira

> + else
>
> #define scoped_guard(_name, args...) \
> for (CLASS(_name, scope)(args), \

2024-02-05 23:30:49

by Dan Williams

[permalink] [raw]
Subject: RE: [PATCH 1/2 v2] cleanup: Add cond_guard() to conditional guards

Ira Weiny wrote:
> Dan Williams wrote:
> > Fabio M. De Francesco wrote:
> > > Add cond_guard() macro to conditional guards.
> > >
> > > cond_guard() is a guard to be used with the conditional variants of locks,
> > > like down_read_trylock() or mutex_lock_interruptible().
> > >
> > > It takes a statement (or more statements in a block) that is passed to its
> > > second argument. That statement (or block) is executed if waiting for a
> > > lock is interrupted or if a _trylock() fails in case of contention.
> > >
> > > Usage example:
> > >
> > > cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> > >
> > > Consistently with the other guards, locks are unlocked at the exit of the
> > > scope where cond_guard() is called.
> > >
> > > Cc: Peter Zijlstra <[email protected]>
> > > Suggested-by: Dan Williams <[email protected]>
> > > Suggested-by: Ira Weiny <[email protected]>
> > > Signed-off-by: Fabio M. De Francesco <[email protected]>
> > > ---
> > > include/linux/cleanup.h | 14 ++++++++++++++
> > > 1 file changed, 14 insertions(+)
> > >
> > > diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> > > index c2d09bc4f976..88af56600325 100644
> > > --- a/include/linux/cleanup.h
> > > +++ b/include/linux/cleanup.h
> > > @@ -134,6 +134,16 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> > > * an anonymous instance of the (guard) class, not recommended for
> > > * conditional locks.
> > > *
> > > + * cond_guard(name, fail, args...):
> > > + * a guard to be used with the conditional variants of locks, like
> > > + * down_read_trylock() or mutex_lock_interruptible. 'fail' are one or more
> > > + * statements that are executed when waiting for a lock is interrupted or
> > > + * when a _trylock() fails in case of contention.
> > > + *
> > > + * Example:
> > > + *
> > > + * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> >
> > That _fail argument likely needs to be a statement expression for the
> > multi-statement case.
>
> You mean ({ ... }) as discussed here?
>
> https://lore.kernel.org/all/[email protected]/

Yes.

> > > + *
> > > * 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
> > > @@ -165,6 +175,10 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> > >
> > > #define __guard_ptr(_name) class_##_name##_lock_ptr
> > >
> > > +#define cond_guard(_name, _fail, args...) \
> > > + CLASS(_name, scope)(args); \
> > > + if (!__guard_ptr(_name)(&scope)) _fail
> >
> > No, as I stated before this is broken for usages of:
> >
> > if () cond_guard() else if ()
> >
> > The 'else' in the definition is critical, this builds for me (untested):
>
> I did not test Fabios work directly but I don't understand this example.
> It seems like your suggestion does nothing useful. The cond_guard()
> becomes a single statement like...
>
> if ()
> cond_guard();
> else ...
>
> ... And can't protect anything.

A sequence to acquire and drop a lock is sometimes a barrier semantic.
Is it typical, no, is it possible, yes. I otherwise do not understand
the need to include the subtle side effect.

> cond_guard() as defined, the ';' must be used as part of cond_guard() and
> should complete the internal macro 'if' statement.
>
> I think this would work:
>
> if () {
> cond_guard();
> ... do locked stuff ...
> } else ...
>
> >
> > diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> > index 88af56600325..665407498781 100644
> > --- a/include/linux/cleanup.h
> > +++ b/include/linux/cleanup.h
> > @@ -142,7 +142,7 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> > *
> > * Example:
> > *
> > - * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> > + * cond_guard(rwsem_read_try, ({ printk(...); return 0; }), &semaphore);
> > *
> > * scoped_guard (name, args...) { }:
> > * similar to CLASS(name, scope)(args), except the variable (with the
> > @@ -177,7 +177,8 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> >
> > #define cond_guard(_name, _fail, args...) \
> > CLASS(_name, scope)(args); \
> > - if (!__guard_ptr(_name)(&scope)) _fail
> > + if (!__guard_ptr(_name)(&scope)) _fail; \
>
> Building on what I found for scoped_cond_guard() this should be
>
> > + if (!__guard_ptr(_name)(&scope)) { _fail; }

That's still a dangling if () statement.

>
> And drop the else. The else needs to clearly be part of an outside if in
> your example.

Please just rely on a statement-expression for the odd multi-statement _fail
use case and include the else in the definition to remove any room for
confusion.

2024-02-05 23:56:21

by Ira Weiny

[permalink] [raw]
Subject: RE: [PATCH 1/2 v2] cleanup: Add cond_guard() to conditional guards

Dan Williams wrote:
> Ira Weiny wrote:
> > Dan Williams wrote:
> > > Fabio M. De Francesco wrote:
> > > > Add cond_guard() macro to conditional guards.
> > > >
> > > > cond_guard() is a guard to be used with the conditional variants of locks,
> > > > like down_read_trylock() or mutex_lock_interruptible().
> > > >
> > > > It takes a statement (or more statements in a block) that is passed to its
> > > > second argument. That statement (or block) is executed if waiting for a
> > > > lock is interrupted or if a _trylock() fails in case of contention.
> > > >
> > > > Usage example:
> > > >
> > > > cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> > > >
> > > > Consistently with the other guards, locks are unlocked at the exit of the
> > > > scope where cond_guard() is called.
> > > >
> > > > Cc: Peter Zijlstra <[email protected]>
> > > > Suggested-by: Dan Williams <[email protected]>
> > > > Suggested-by: Ira Weiny <[email protected]>
> > > > Signed-off-by: Fabio M. De Francesco <[email protected]>
> > > > ---
> > > > include/linux/cleanup.h | 14 ++++++++++++++
> > > > 1 file changed, 14 insertions(+)
> > > >
> > > > diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> > > > index c2d09bc4f976..88af56600325 100644
> > > > --- a/include/linux/cleanup.h
> > > > +++ b/include/linux/cleanup.h
> > > > @@ -134,6 +134,16 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> > > > * an anonymous instance of the (guard) class, not recommended for
> > > > * conditional locks.
> > > > *
> > > > + * cond_guard(name, fail, args...):
> > > > + * a guard to be used with the conditional variants of locks, like
> > > > + * down_read_trylock() or mutex_lock_interruptible. 'fail' are one or more
> > > > + * statements that are executed when waiting for a lock is interrupted or
> > > > + * when a _trylock() fails in case of contention.
> > > > + *
> > > > + * Example:
> > > > + *
> > > > + * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> > >
> > > That _fail argument likely needs to be a statement expression for the
> > > multi-statement case.
> >
> > You mean ({ ... }) as discussed here?
> >
> > https://lore.kernel.org/all/[email protected]/
>
> Yes.
>
> > > > + *
> > > > * 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
> > > > @@ -165,6 +175,10 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> > > >
> > > > #define __guard_ptr(_name) class_##_name##_lock_ptr
> > > >
> > > > +#define cond_guard(_name, _fail, args...) \
> > > > + CLASS(_name, scope)(args); \
> > > > + if (!__guard_ptr(_name)(&scope)) _fail
> > >
> > > No, as I stated before this is broken for usages of:
> > >
> > > if () cond_guard() else if ()
> > >
> > > The 'else' in the definition is critical, this builds for me (untested):
> >
> > I did not test Fabios work directly but I don't understand this example.
> > It seems like your suggestion does nothing useful. The cond_guard()
> > becomes a single statement like...
> >
> > if ()
> > cond_guard();
> > else ...
> >
> > ... And can't protect anything.
>
> A sequence to acquire and drop a lock is sometimes a barrier semantic.
> Is it typical, no, is it possible, yes. I otherwise do not understand
> the need to include the subtle side effect.

I was not trying to include a subtle side effect. I was thinking that the
else block would be the only block covered by the lock. I've looked at
the preprocessor output again and I now see what you are saying. Also I
see I was thinking incorrectly. The else will be an empty statement and
the rest of the outer block will be covered by the lock...

Sorry for not seeing this before.

> > cond_guard() as defined, the ';' must be used as part of cond_guard() and
> > should complete the internal macro 'if' statement.
> >
> > I think this would work:
> >
> > if () {
> > cond_guard();
> > ... do locked stuff ...
> > } else ...
> >
> > >
> > > diff --git a/include/linux/cleanup.h b/include/linux/cleanup.h
> > > index 88af56600325..665407498781 100644
> > > --- a/include/linux/cleanup.h
> > > +++ b/include/linux/cleanup.h
> > > @@ -142,7 +142,7 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> > > *
> > > * Example:
> > > *
> > > - * cond_guard(rwsem_read_try, { printk(...); return 0; }, &semaphore);
> > > + * cond_guard(rwsem_read_try, ({ printk(...); return 0; }), &semaphore);
> > > *
> > > * scoped_guard (name, args...) { }:
> > > * similar to CLASS(name, scope)(args), except the variable (with the
> > > @@ -177,7 +177,8 @@ static inline class_##_name##_t class_##_name##ext##_constructor(_init_args) \
> > >
> > > #define cond_guard(_name, _fail, args...) \
> > > CLASS(_name, scope)(args); \
> > > - if (!__guard_ptr(_name)(&scope)) _fail
> > > + if (!__guard_ptr(_name)(&scope)) _fail; \
> >
> > Building on what I found for scoped_cond_guard() this should be
> >
> > > + if (!__guard_ptr(_name)(&scope)) { _fail; }
>
> That's still a dangling if () statement.
>
> >
> > And drop the else. The else needs to clearly be part of an outside if in
> > your example.
>
> Please just rely on a statement-expression for the odd multi-statement _fail
> use case and include the else in the definition to remove any room for
> confusion.

Yea ok I see it now,
Ira