RE: [PATCH v2 1/6] Add ancillary bus support

[Parav Pandit <parav@nvidia.com>]

> From: Leon Romanovsky <leon@kernel.org>
> Sent: Thursday, October 8, 2020 1:15 PM
> 
> On Thu, Oct 08, 2020 at 07:14:17AM +0000, Parav Pandit wrote:
> >
> >
> > > From: Leon Romanovsky <leon@kernel.org>
> > > Sent: Thursday, October 8, 2020 10:56 AM
> > >
> > > On Thu, Oct 08, 2020 at 04:56:01AM +0000, Parav Pandit wrote:
> > > >
> > > >
> > > > > From: Pierre-Louis Bossart
> > > > > <pierre-louis.bossart@linux.intel.com>
> > > > > Sent: Thursday, October 8, 2020 3:20 AM
> > > > >
> > > > >
> > > > > On 10/7/20 4:22 PM, Ertman, David M wrote:
> > > > > >> -----Original Message-----
> > > > > >> From: Pierre-Louis Bossart
> > > > > >> <pierre-louis.bossart@linux.intel.com>
> > > > > >> Sent: Wednesday, October 7, 2020 1:59 PM
> > > > > >> To: Ertman, David M <david.m.ertman@intel.com>; Parav Pandit
> > > > > >> <parav@nvidia.com>; Leon Romanovsky <leon@kernel.org>
> > > > > >> Cc: alsa-devel@alsa-project.org; parav@mellanox.com;
> > > > > >> tiwai@suse.de; netdev@vger.kernel.org;
> > > > > >> ranjani.sridharan@linux.intel.com;
> > > > > >> fred.oh@linux.intel.com; linux-rdma@vger.kernel.org;
> > > > > >> dledford@redhat.com; broonie@kernel.org; Jason Gunthorpe
> > > > > >> <jgg@nvidia.com>; gregkh@linuxfoundation.org;
> > > > > >> kuba@kernel.org; Williams, Dan J <dan.j.williams@intel.com>;
> > > > > >> Saleem, Shiraz <shiraz.saleem@intel.com>;
> > > > > >> davem@davemloft.net; Patil, Kiran <kiran.patil@intel.com>
> > > > > >> Subject: Re: [PATCH v2 1/6] Add ancillary bus support
> > > > > >>
> > > > > >>
> > > > > >>
> > > > > >>>> Below is most simple, intuitive and matching with core APIs
> > > > > >>>> for name and design pattern wise.
> > > > > >>>> init()
> > > > > >>>> {
> > > > > >>>> 	err = ancillary_device_initialize();
> > > > > >>>> 	if (err)
> > > > > >>>> 		return ret;
> > > > > >>>>
> > > > > >>>> 	err = ancillary_device_add();
> > > > > >>>> 	if (ret)
> > > > > >>>> 		goto err_unwind;
> > > > > >>>>
> > > > > >>>> 	err = some_foo();
> > > > > >>>> 	if (err)
> > > > > >>>> 		goto err_foo;
> > > > > >>>> 	return 0;
> > > > > >>>>
> > > > > >>>> err_foo:
> > > > > >>>> 	ancillary_device_del(adev);
> > > > > >>>> err_unwind:
> > > > > >>>> 	ancillary_device_put(adev->dev);
> > > > > >>>> 	return err;
> > > > > >>>> }
> > > > > >>>>
> > > > > >>>> cleanup()
> > > > > >>>> {
> > > > > >>>> 	ancillary_device_de(adev);
> > > > > >>>> 	ancillary_device_put(adev);
> > > > > >>>> 	/* It is common to have a one wrapper for this as
> > > > > >>>> ancillary_device_unregister().
> > > > > >>>> 	 * This will match with core device_unregister() that has
> > > > > >>>> precise documentation.
> > > > > >>>> 	 * but given fact that init() code need proper error
> > > > > >>>> unwinding, like above,
> > > > > >>>> 	 * it make sense to have two APIs, and no need to export
> > > > > >>>> another symbol for unregister().
> > > > > >>>> 	 * This pattern is very easy to audit and code.
> > > > > >>>> 	 */
> > > > > >>>> }
> > > > > >>>
> > > > > >>> I like this flow +1
> > > > > >>>
> > > > > >>> But ... since the init() function is performing both
> > > > > >>> device_init and device_add - it should probably be called
> > > > > >>> ancillary_device_register, and we are back to a single
> > > > > >>> exported API for both register and unregister.
> > > > > >>
> > > > > >> Kind reminder that we introduced the two functions to allow
> > > > > >> the caller to know if it needed to free memory when
> > > > > >> initialize() fails, and it didn't need to free memory when
> > > > > >> add() failed since
> > > > > >> put_device() takes care of it. If you have a single init()
> > > > > >> function it's impossible to know which behavior to select on error.
> > > > > >>
> > > > > >> I also have a case with SoundWire where it's nice to first
> > > > > >> initialize, then set some data and then add.
> > > > > >>
> > > > > >
> > > > > > The flow as outlined by Parav above does an initialize as the
> > > > > > first step, so every error path out of the function has to do
> > > > > > a put_device(), so you would never need to manually free the
> > > > > > memory in
> > > > > the setup function.
> > > > > > It would be freed in the release call.
> > > > >
> > > > > err = ancillary_device_initialize(); if (err)
> > > > > 	return ret;
> > > > >
> > > > > where is the put_device() here? if the release function does any
> > > > > sort of kfree, then you'd need to do it manually in this case.
> > > > Since device_initialize() failed, put_device() cannot be done here.
> > > > So yes, pseudo code should have shown, if (err) {
> > > > 	kfree(adev);
> > > > 	return err;
> > > > }
> > > >
> > > > If we just want to follow register(), unregister() pattern,
> > > >
> > > > Than,
> > > >
> > > > ancillar_device_register() should be,
> > > >
> > > > /**
> > > >  * ancillar_device_register() - register an ancillary device
> > > >  * NOTE: __never directly free @adev after calling this function,
> > > > even if it returned
> > > >  * an error. Always use ancillary_device_put() to give up the
> > > > reference
> > > initialized by this function.
> > > >  * This note matches with the core and caller knows exactly what
> > > > to be
> > > done.
> > > >  */
> > > > ancillary_device_register()
> > > > {
> > > > 	device_initialize(&adev->dev);
> > > > 	if (!dev->parent || !adev->name)
> > > > 		return -EINVAL;
> > > > 	if (!dev->release && !(dev->type && dev->type->release)) {
> > > > 		/* core is already capable and throws the warning when
> > > release callback is not set.
> > > > 		 * It is done at drivers/base/core.c:1798.
> > > > 		 * For NULL release it says, "does not have a release()
> > > function, it is broken and must be fixed"
> > > > 		 */
> > > > 		return -EINVAL;
> > > > 	}
> > > > 	err = dev_set_name(adev...);
> > > > 	if (err) {
> > > > 		/* kobject_release() -> kobject_cleanup() are capable to
> > > detect if name is set/ not set
> > > > 		  * and free the const if it was set.
> > > > 		  */
> > > > 		return err;
> > > > 	}
> > > > 	err = device_add(&adev->dev);
> > > > 	If (err)
> > > > 		return err;
> > > > }
> > > >
> > > > Caller code:
> > > > init()
> > > > {
> > > > 	adev = kzalloc(sizeof(*foo_adev)..);
> > > > 	if (!adev)
> > > > 		return -ENOMEM;
> > > > 	err = ancillary_device_register(&adev);
> > > > 	if (err)
> > > > 		goto err;
> > > >
> > > > err:
> > > > 	ancillary_device_put(&adev);
> > > > 	return err;
> > > > }
> > > >
> > > > cleanup()
> > > > {
> > > > 	ancillary_device_unregister(&adev);
> > > > }
> > > >
> > > > Above pattern is fine too matching the core.
> > > >
> > > > If I understand Leon correctly, he prefers simple register(),
> > > > unregister()
> > > pattern.
> > > > If, so it should be explicit register(), unregister() API.
> > >
> > > This is my summary
> > > https://lore.kernel.org/linux-rdma/20201008052137.GA13580@unreal
> > > The API should be symmetric.
> > >
> >
> > I disagree to your below point.
> > > 1. You are not providing driver/core API but simplification and
> > > obfuscation of basic primitives and structures. This is new layer.
> > > There is no room for a claim that we must to follow internal API.
> > If ancillary bus has
> > ancillary_device_add(), it cannot do device_initialize() and device_add() in
> both.
> >
> > I provided two examples and what really matters is a given patchset
> > uses (need to use) which pattern,
> > initialize() + add(), or register() + unregister().
> >
> > As we all know that API is not added for future. It is the future patch
> extends it.
> > So lets wait for Pierre to reply if soundwire can follow register(),
> unregister() sequence.
> > This way same APIs can service both use-cases.
> >
> > Regarding,
> > > 3. You can't "ask" from users to call internal calls (put_device)
> > > over internal fields in ancillary_device.
> > In that case if should be ancillary_device_put() ancillary_device_release().
> >
> > Or we should follow the patten of ib_alloc_device [1],
> > ancillary_device_alloc()
> >     -> kzalloc(adev + dev) with compile time assert check like rdma and vdpa
> subsystem.
> >     ->device_initialize()
> > ancillary_device_add()
> >
> > ancillar_device_de() <- balances with add
> > ancillary_device_dealloc() <-- balances with device_alloc(), which does the
> put_device() + free the memory allocated in alloc().
> >
> > This approach of [1] also eliminates exposing adev.dev.release =
> <drivers_release_method_to_free_adev> in drivers.
> > And container_of() benefit also continues..
> >
> > [1]
> > https://elixir.bootlin.com/linux/v5.9-rc8/source/include/rdma/ib_verbs
> > .h#L2791
> >
> 
> My code looks like this, probably yours looks the same.
> 
>   247                 priv->adev[i] = kzalloc(sizeof(*priv->adev[i]), GFP_KERNEL);
>   248                 if (!priv->adev[i])
>   249                         goto init_err;
>   250
>   251                 adev = &priv->adev[i]->adev;
>   252                 adev->id = idx;
>   253                 adev->name = mlx5_adev_devices[i].suffix;
>   254                 adev->dev.parent = dev->device;
>   255                 adev->dev.release = adev_release;
>   256                 priv->adev[i]->mdev = dev;
>   257
>   258                 ret = ancillary_device_initialize(adev);
>   259                 if (ret)
>   260                         goto init_err;
>   261
>   262                 ret = ancillary_device_add(adev);
>   263                 if (ret) {
>   264                         put_device(&adev->dev);
>   265                         goto add_err;
>   266                 }

Yes, subfunction code is also very similar.
You expressed concerned that you didn't like put_device() at [1].
But in above code is touching adev->dev.{parent, release} is ok?
>   254                 adev->dev.parent = dev->device;
>   255                 adev->dev.release = adev_release;

If not,

We can make it elegant by doing,

the patten of ib_alloc_device [1],
ancillary_device_alloc()
    -> kzalloc(adev + dev) with compile time assert check like rdma and vdpa subsystem.
    ->device_initialize()
ancillary_device_add()

ancillar_device_de() <- balances with add
ancillary_device_dealloc() <-- balances with device_alloc(), which does the put_device() + free the memory allocated in alloc().

This approach of [2] also eliminates exposing adev.dev.release = <drivers_release_method_to_free_adev> in drivers.
And container_of() benefit also continues..

[1] https://lore.kernel.org/linux-rdma/20201007192610.GD3964015@unreal/
[2] https://elixir.bootlin.com/linux/v5.9-rc8/source/include/rdma/ib_verbs.h#L2791