sysfs deferred_probe attribute

sysfs deferred_probe attribute

[Rob Herring]

I just noticed that we have a new device attribute 'deferred_probe'
added in 4.10 with this commit:

commit 6751667a29d6fd64afb9ce30567ad616b68ed789
Author: Ben Hutchings <ben.hutchings@codethink.co.uk>
Date:   Tue Aug 16 14:34:18 2016 +0100

    driver core: Add deferred_probe attribute to devices in sysfs

    It is sometimes useful to know that a device is on the deferred probe
    list rather than, say, not having a driver available.  Expose this
    information to user-space.

    Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
    Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>


It seems like a bad idea to add an ABI for an internal kernel feature.
When/if we replace deferred probe with something better based on
functional dependencies are we going to keep this attr around? Or
remove it and assume no userspace uses it? Perhaps it should be hidden
behind CONFIG_DEBUG or just make a debugfs file that lists the
deferred list. Then you wouldn't have to hunt for what got deferred.

Rob

Re: sysfs deferred_probe attribute

[Greg Kroah-Hartman]

On Thu, Jan 12, 2017 at 11:27:01AM -0600, Rob Herring wrote:
> I just noticed that we have a new device attribute 'deferred_probe'
> added in 4.10 with this commit:
> 
> commit 6751667a29d6fd64afb9ce30567ad616b68ed789
> Author: Ben Hutchings <ben.hutchings@codethink.co.uk>
> Date:   Tue Aug 16 14:34:18 2016 +0100
> 
>     driver core: Add deferred_probe attribute to devices in sysfs
> 
>     It is sometimes useful to know that a device is on the deferred probe
>     list rather than, say, not having a driver available.  Expose this
>     information to user-space.
> 
>     Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
>     Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> 
> 
> It seems like a bad idea to add an ABI for an internal kernel feature.
> When/if we replace deferred probe with something better based on
> functional dependencies are we going to keep this attr around? Or
> remove it and assume no userspace uses it? Perhaps it should be hidden
> behind CONFIG_DEBUG or just make a debugfs file that lists the
> deferred list. Then you wouldn't have to hunt for what got deferred.

Ah, debugfs would be nice, I'd much prefer that.  I don't know how Ben
is using this, but I think that would make more sense to me.

thanks,

greg k-h

Re: sysfs deferred_probe attribute

[Ben Hutchings]

On Thu, 2017-01-12 at 18:41 +0100, Greg Kroah-Hartman wrote:
> On Thu, Jan 12, 2017 at 11:27:01AM -0600, Rob Herring wrote:
> > I just noticed that we have a new device attribute 'deferred_probe'
> > added in 4.10 with this commit:
> > 
> > commit 6751667a29d6fd64afb9ce30567ad616b68ed789
> > Author: Ben Hutchings <ben.hutchings@codethink.co.uk>
> > Date:   Tue Aug 16 14:34:18 2016 +0100
> > 
> >     driver core: Add deferred_probe attribute to devices in sysfs
> > 
> >     It is sometimes useful to know that a device is on the deferred probe
> >     list rather than, say, not having a driver available.  Expose this
> >     information to user-space.
> > 
> >     Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
> >     Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > 
> > 
> > It seems like a bad idea to add an ABI for an internal kernel feature.
> > When/if we replace deferred probe with something better based on
> > functional dependencies are we going to keep this attr around? Or
> > remove it and assume no userspace uses it?

It should be removed then (and replaced with some kind of representation
of dependencies).

> > Perhaps it should be hidden
> > behind CONFIG_DEBUG or just make a debugfs file that lists the
> > deferred list. Then you wouldn't have to hunt for what got deferred.
> 
> Ah, debugfs would be nice, I'd much prefer that.  I don't know how Ben
> is using this, but I think that would make more sense to me.

I'm not using it any programmatic way, and don't intend to.  debugfs
would be OK, but attaching it to devices was easy to do and seemed to
make sense.

Ben.

-- 
Ben Hutchings
Software Developer, Codethink Ltd.

Re: sysfs deferred_probe attribute

[Tomeu Vizoso]

On 01/12/2017 07:26 PM, Ben Hutchings wrote:
> On Thu, 2017-01-12 at 18:41 +0100, Greg Kroah-Hartman wrote:
>> On Thu, Jan 12, 2017 at 11:27:01AM -0600, Rob Herring wrote:
>>> I just noticed that we have a new device attribute 'deferred_probe'
>>> added in 4.10 with this commit:
>>>
>>> commit 6751667a29d6fd64afb9ce30567ad616b68ed789
>>> Author: Ben Hutchings <ben.hutchings@codethink.co.uk>
>>> Date:   Tue Aug 16 14:34:18 2016 +0100
>>>
>>>     driver core: Add deferred_probe attribute to devices in sysfs
>>>
>>>     It is sometimes useful to know that a device is on the deferred probe
>>>     list rather than, say, not having a driver available.  Expose this
>>>     information to user-space.
>>>
>>>     Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
>>>     Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
>>>
>>>
>>> It seems like a bad idea to add an ABI for an internal kernel feature.
>>> When/if we replace deferred probe with something better based on
>>> functional dependencies are we going to keep this attr around? Or
>>> remove it and assume no userspace uses it?
> 
> It should be removed then (and replaced with some kind of representation
> of dependencies).
> 
>>> Perhaps it should be hidden
>>> behind CONFIG_DEBUG or just make a debugfs file that lists the
>>> deferred list. Then you wouldn't have to hunt for what got deferred.
>>
>> Ah, debugfs would be nice, I'd much prefer that.  I don't know how Ben
>> is using this, but I think that would make more sense to me.
> 
> I'm not using it any programmatic way, and don't intend to.  debugfs
> would be OK, but attaching it to devices was easy to do and seemed to
> make sense.

Russell King started work on printing those devices in the deferred
queue at late_initcall, not sure why it didn't land.

But note that without proper dependency information, you cannot know for
sure if a device deferred its probe just because a dependency doesn't
have a matching driver.

Regards,

Tomeu

Re: sysfs deferred_probe attribute

[Greg Kroah-Hartman]

On Thu, Jan 12, 2017 at 06:26:11PM +0000, Ben Hutchings wrote:
> On Thu, 2017-01-12 at 18:41 +0100, Greg Kroah-Hartman wrote:
> > On Thu, Jan 12, 2017 at 11:27:01AM -0600, Rob Herring wrote:
> > > I just noticed that we have a new device attribute 'deferred_probe'
> > > added in 4.10 with this commit:
> > > 
> > > commit 6751667a29d6fd64afb9ce30567ad616b68ed789
> > > Author: Ben Hutchings <ben.hutchings@codethink.co.uk>
> > > Date:   Tue Aug 16 14:34:18 2016 +0100
> > > 
> > >     driver core: Add deferred_probe attribute to devices in sysfs
> > > 
> > >     It is sometimes useful to know that a device is on the deferred probe
> > >     list rather than, say, not having a driver available.  Expose this
> > >     information to user-space.
> > > 
> > >     Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
> > >     Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > > 
> > > 
> > > It seems like a bad idea to add an ABI for an internal kernel feature.
> > > When/if we replace deferred probe with something better based on
> > > functional dependencies are we going to keep this attr around? Or
> > > remove it and assume no userspace uses it?
> 
> It should be removed then (and replaced with some kind of representation
> of dependencies).

Ok, now reverted.

thanks,

greg k-h

Re: sysfs deferred_probe attribute

[Geert Uytterhoeven]

Hi Rob,

On Thu, Jan 12, 2017 at 6:27 PM, Rob Herring <robh@kernel.org> wrote:
> I just noticed that we have a new device attribute 'deferred_probe'
> added in 4.10 with this commit:
>
> commit 6751667a29d6fd64afb9ce30567ad616b68ed789
> Author: Ben Hutchings <ben.hutchings@codethink.co.uk>
> Date:   Tue Aug 16 14:34:18 2016 +0100
>
>     driver core: Add deferred_probe attribute to devices in sysfs
>
>     It is sometimes useful to know that a device is on the deferred probe
>     list rather than, say, not having a driver available.  Expose this
>     information to user-space.
>
>     Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
>     Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
>
>
> It seems like a bad idea to add an ABI for an internal kernel feature.
> When/if we replace deferred probe with something better based on
> functional dependencies are we going to keep this attr around? Or
> remove it and assume no userspace uses it? Perhaps it should be hidden
> behind CONFIG_DEBUG or just make a debugfs file that lists the
> deferred list. Then you wouldn't have to hunt for what got deferred.

FWIW, I had just created a "check-deferred-probe" script that does

    find /sys -name deferred_probe -print0 | xargs -0 grep -v '^0$'

A list would be even better, from the point of view of the user.
As I haven't looked at the implementation, I don't know what impact
that would have on the system due to e.g. locking.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds