Re: [PATCH 1/4] media: v4l2-subdev: Add cleanup macros for active state

From: Tomi Valkeinen
Date: Thu Sep 26 2024 - 11:26:55 EST

On 25/09/2024 19:35, Laurent Pinchart wrote:
Hi Tomi,

On Tue, Sep 24, 2024 at 08:53:38PM +0300, Tomi Valkeinen wrote:
On 24/09/2024 20:17, Laurent Pinchart wrote:
On Tue, Sep 17, 2024 at 05:09:29PM +0300, Tomi Valkeinen wrote:
From: Tomi Valkeinen <tomi.valkeinen+renesas@xxxxxxxxxxxxxxxx>

Add cleanup macros for active state. These can be used to call
v4l2_subdev_lock_and_get_active_state() and manage the unlocking either
in unscoped or scoped fashion:

This locks the state, gets it to the 'state' variable, and unlocks when
exiting the surrounding scope:

CLASS(v4l2_subdev_lock_and_get_active_state, state)(subdev);

This does the same, but inside the given scope:

scoped_v4l2_subdev_lock_and_get_active_state(subdev) {
}

Signed-off-by: Tomi Valkeinen <tomi.valkeinen+renesas@xxxxxxxxxxxxxxxx>
---
include/media/v4l2-subdev.h | 10 ++++++++++
1 file changed, 10 insertions(+)

diff --git a/include/media/v4l2-subdev.h b/include/media/v4l2-subdev.h
index bd235d325ff9..699007cfffd7 100644
--- a/include/media/v4l2-subdev.h
+++ b/include/media/v4l2-subdev.h
@@ -8,6 +8,7 @@
#ifndef _V4L2_SUBDEV_H
#define _V4L2_SUBDEV_H
+#include <linux/cleanup.h>
#include <linux/types.h>
#include <linux/v4l2-subdev.h>
#include <media/media-entity.h>
@@ -1854,6 +1855,15 @@ v4l2_subdev_lock_and_get_active_state(struct v4l2_subdev *sd)
return sd->active_state;
}
+DEFINE_CLASS(v4l2_subdev_lock_and_get_active_state, struct v4l2_subdev_state *,
+ v4l2_subdev_unlock_state(_T),
+ v4l2_subdev_lock_and_get_active_state(sd), struct v4l2_subdev *sd);
+
+#define scoped_v4l2_subdev_lock_and_get_active_state(sd) \
+ for (CLASS(v4l2_subdev_lock_and_get_active_state, state)(sd), \
+ *done = NULL; \
+ !done; done = (void *)1)

That a very long name :-S Could this be done using the scoped_guard()
macro instead ? For instance, with spinlocks you can do

scoped_guard(spinlock_irqsave, &dev->lock) {
...
}

It would be nice to be able to write

scoped_guard(v4l2_subdev_state, sd) {

This can be done but then you need to pass the state to it, not sd. Or
perhaps it would be possible to implicitly turn the sd into active
state, but I don't like that at all...

Or maybe:

scoped_guard(v4l2_subdev_state, v4l2_subdev_get_unlocked_active_state(sd))

Not very short either...

That's not very nice. Are there other examples in the kernel of
scoped_*() macros magically creating variables that are then used within
the scope ? I have a feeling we shouldn't do it here.

I'm not aware of any other scoped macros than scoped_guard. Or if there are similar macros creating a variable.

I think creating the variable inside the macro is one of the main points with scoped_v4l2_subdev_lock_and_get_active_state(), so that not only the locking of the state of scoped, also the variable is scoped. It hasn't been just once or twice when I have accidentally used a variable that was supposed to be only used inside a scope.

...
}

This being said, we would then end up with the state variable being
named scope, which wouldn't be great.

No, it wouldn't.

This is actually one of my issues with the above macros, and especially
scoped_v4l2_subdev_lock_and_get_active_state(). It creates a local state
variable in the scope, which makes the code less readable in my opinion.

It's trivial to add a variable name there, as mentioned in the intro letter.

You mentioned the const/non-const state issue in the other email. I
wonder if some macro-magic could be done for that... Or we can always
just add "scoped_v4l2_subdev_lock_and_get_active_state_const()"! =)

And that's supposed to be an improvement ? :D

Also, it's not like we have to use these _everywhere_. So maybe if you
want a const state, don't use the scoped or the class.

Looking at the rest of your series there are very few instances of
scoped_v4l2_subdev_lock_and_get_active_state(), so I'm tempted to simply

The changes in this series were just a few examples for a couple of drivers I've been working on lately. I didn't do any kind of thorough study how many users we have for these.

leave it out. When one writes

scoped_guard(spinlock_irqsave, &dev->lock) {
}

It's clear that you're locking the lock for the scope using
spinlock_irqsave. The scoped guard performs a scoped action on an
existing object. The V4L2 subdev active state is different, I don't
think scoped_guard() gives the right semantics.

Hmm so are you here referring specifically to using scoped_guard() for the active state locking, or do you refer to the scoped_v4l2_subdev_lock_and_get_active_state()?

We could keep the class and drop
scoped_v4l2_subdev_lock_and_get_active_state(). I think I would like to
shorten the class name then.

Another option is to use DEFINE_FREE() and __free() instead.

That can be added too. I had them, but I didn't consider them as useful
when I already added the class and scoped.

I have to say I don't particularly like the look of either the scoped or
the class, or even the free. But they're so useful wrt. error handling
that I don't care if the syntax annoys me =).

CLASS() is a bit better once we'll get used to it, as the name of the
variable is explicit. It however doesn't solve the const issue.

Again, we can add the variable name to scoped_v4l2_subdev_lock_and_get_active_state(), it's a trivial change.

I don't really see the const issue as a major thing. Missing the const here never breaks anything. You can pass the non-const state to functions that take const state. And you never have to use these helpers, you can do it all manually.

Adding const versions is possible. In some way the code has to tell the macros that "I want const state", so adding a __const() version of the macro sounds fine to me. Except the length of the name quickly becomes a problem =).

Furthermore, its semantics is meant to represent creation of an object
with automatic destruction when it leaves the scope, while with the
subdev active state you're not creating an object. That's why I think

The lock guards use CLASS(), and they're not creating any new objects. They do hide the use of CLASS, though. But we can do the same by just creating our own macro that directly uses CLASS.

that an explicit variable with a __free() annotation may be the best
option for this.

But __free() does a different thing. It can, of course, be used in a scoped fashion by adding a scope with { }, like:

{
struct v4l2_subdev_state *state __free(v4l2_subdev_state_lock) = v4l2_subdev_lock_and_get_active_state(sd);

// use state
}

but I don't very often see that style used. "__free()" there also makes me think the state will be freed... And the line above is definitely not short...

If we come up with a suitable short name, this is quite a bit cleaner looking:

hopefully_somewhat_short_name(sd, state) {
// use state
}

But all that said, I have to say I don't know if we have enough users for these to make these worthwhile. Looking at the single scoped_v4l2_subdev_lock_and_get_active_state() use I added to rzg2l-csi2.c... That's not really needed, as the original code is not right: the rzg2l_csi2_calc_mbps() shouldn't lock the active state, it should be passed the (locked) active state.

Tomi