image.bbclass: USE_DEVFS is now useless

image.bbclass: USE_DEVFS is now useless

[Matthieu CRAPET]

[-- Attachment #1: Type: text/plain, Size: 459 bytes --]

Hi,

Since :
http://cgit.openembedded.org/openembedded-core/commit/meta/classes/image.bbclass?id=a83144bac8d67704ff66f5dc0fc56f5b63979694

USE_DEVFS is not considered anymore.

Setting IMAGE_DEVICE_TABLE or IMAGE_DEVICE_TABLES to empty string in an image recipe works (makedevs is not called)

Should be we drop USE_DEVFS variable (meta/classes/image.bbclass) or (re)add check in meta/lib/oe/rootfs.py?
I can propose a patch.

Regards,
Matthieu

[-- Attachment #2: Type: text/html, Size: 2975 bytes --]

Re: image.bbclass: USE_DEVFS is now useless

[Phil Blundell]

[-- Attachment #1: Type: text/plain, Size: 545 bytes --]

On Wed, 2014-02-26 at 11:11 +0000, Matthieu CRAPET wrote:
> Setting IMAGE_DEVICE_TABLE or IMAGE_DEVICE_TABLES to empty string in
> an image recipe works (makedevs is not called)
> 
>  
> 
> Should be we drop USE_DEVFS variable (meta/classes/image.bbclass) or
> (re)add check in meta/lib/oe/rootfs.py?


Having just tripped over this myself, I think we should re-add the
check.   I was going to post a reply to the original patch submission
email but for some reason I can't find it in either my inbox or the web
archive.

p.



[-- Attachment #2: Type: text/html, Size: 894 bytes --]

Re: image.bbclass: USE_DEVFS is now useless

[Phil Blundell]

[-- Attachment #1: Type: text/plain, Size: 1295 bytes --]

On Tue, 2014-04-01 at 18:44 +0100, Phil Blundell wrote:

> On Wed, 2014-02-26 at 11:11 +0000, Matthieu CRAPET wrote: 
> 
> > Setting IMAGE_DEVICE_TABLE or IMAGE_DEVICE_TABLES to empty string in
> > an image recipe works (makedevs is not called)
> > 
> >  
> > 
> > Should be we drop USE_DEVFS variable (meta/classes/image.bbclass) or
> > (re)add check in meta/lib/oe/rootfs.py?
> 
> 
> Having just tripped over this myself, I think we should re-add the
> check.   I was going to post a reply to the original patch submission
> email but for some reason I can't find it in either my inbox or the
> web archive.


Also note that the default for USE_DEVFS was (and is) 1, so the lack of
this check is actually causing a difference in the default behaviour.
If there's no appetite for reinstating the USE_DEVFS mechanism per se
then it seems like it would be a good idea to make the default
IMAGE_DEVICE_TABLE be blank in order to restore the previous default of
no /dev in the rootfs.  

At present you get a somewhat arbitrary-seeming smattering of devices
from meta/files/device_table-minimal.txt, including such anachronisms
as /dev/ttySA0 and /dev/apm_bios.  It's hard to imagine that anybody
actually wants this stuff in their rootfs in this day and age.

p.


[-- Attachment #2: Type: text/html, Size: 1715 bytes --]

Re: image.bbclass: USE_DEVFS is now useless

[Richard Purdie]

On Tue, 2014-04-01 at 18:48 +0100, Phil Blundell wrote:
> Also note that the default for USE_DEVFS was (and is) 1, so the lack
> of this check is actually causing a difference in the default
> behaviour.  If there's no appetite for reinstating the USE_DEVFS
> mechanism per se then it seems like it would be a good idea to make
> the default IMAGE_DEVICE_TABLE be blank in order to restore the
> previous default of no /dev in the rootfs.  
>
> At present you get a somewhat arbitrary-seeming smattering of devices
> from meta/files/device_table-minimal.txt, including such anachronisms
> as /dev/ttySA0 and /dev/apm_bios.  It's hard to imagine that anybody
> actually wants this stuff in their rootfs in this day and age.

Can we kill apmd at the same time? Please? :)

Cheers,

Richard

Re: image.bbclass: USE_DEVFS is now useless

[Mark Hatle]

On 4/1/14, 12:51 PM, Richard Purdie wrote:
> On Tue, 2014-04-01 at 18:48 +0100, Phil Blundell wrote:
>> Also note that the default for USE_DEVFS was (and is) 1, so the lack
>> of this check is actually causing a difference in the default
>> behaviour.  If there's no appetite for reinstating the USE_DEVFS
>> mechanism per se then it seems like it would be a good idea to make
>> the default IMAGE_DEVICE_TABLE be blank in order to restore the
>> previous default of no /dev in the rootfs.
>>
>> At present you get a somewhat arbitrary-seeming smattering of devices
>> from meta/files/device_table-minimal.txt, including such anachronisms
>> as /dev/ttySA0 and /dev/apm_bios.  It's hard to imagine that anybody
>> actually wants this stuff in their rootfs in this day and age.
>
> Can we kill apmd at the same time? Please? :)

Isn't this still used on some ARM and MIPS targets?  (they emulate apm for basic 
power management.)

If that's finally gone away -- I'd love to finally kill apmd and related.

--Mark

> Cheers,
>
> Richard
>

Re: image.bbclass: USE_DEVFS is now useless

[Richard Purdie]

On Tue, 2014-04-01 at 13:01 -0500, Mark Hatle wrote:
> On 4/1/14, 12:51 PM, Richard Purdie wrote:
> > On Tue, 2014-04-01 at 18:48 +0100, Phil Blundell wrote:
> >> Also note that the default for USE_DEVFS was (and is) 1, so the lack
> >> of this check is actually causing a difference in the default
> >> behaviour.  If there's no appetite for reinstating the USE_DEVFS
> >> mechanism per se then it seems like it would be a good idea to make
> >> the default IMAGE_DEVICE_TABLE be blank in order to restore the
> >> previous default of no /dev in the rootfs.
> >>
> >> At present you get a somewhat arbitrary-seeming smattering of devices
> >> from meta/files/device_table-minimal.txt, including such anachronisms
> >> as /dev/ttySA0 and /dev/apm_bios.  It's hard to imagine that anybody
> >> actually wants this stuff in their rootfs in this day and age.
> >
> > Can we kill apmd at the same time? Please? :)
> 
> Isn't this still used on some ARM and MIPS targets?  (they emulate apm for basic 
> power management.)

Its not gone as far as I know but whether we need the userspace portion,
I'm not so sure, it just does suspend/resume event notifuication afaik
these days and I doubt much uses that now. The battery interfaces were
long since replaced with proper sysfs ones.

Cheers,

Richard

Re: image.bbclass: USE_DEVFS is now useless

[Phil Blundell]

On Tue, 2014-04-01 at 18:51 +0100, Richard Purdie wrote:
> On Tue, 2014-04-01 at 18:48 +0100, Phil Blundell wrote:
> > Also note that the default for USE_DEVFS was (and is) 1, so the lack
> > of this check is actually causing a difference in the default
> > behaviour.  If there's no appetite for reinstating the USE_DEVFS
> > mechanism per se then it seems like it would be a good idea to make
> > the default IMAGE_DEVICE_TABLE be blank in order to restore the
> > previous default of no /dev in the rootfs.  
> >
> > At present you get a somewhat arbitrary-seeming smattering of devices
> > from meta/files/device_table-minimal.txt, including such anachronisms
> > as /dev/ttySA0 and /dev/apm_bios.  It's hard to imagine that anybody
> > actually wants this stuff in their rootfs in this day and age.
> 
> Can we kill apmd at the same time? Please? :)

I can't see why not.  I'd be surprised if there were many/any real users
left, beyond possibly some legacy stuff in meta-handheld, and it
definitely doesn't seem "core" by any reasonable interpretation of the
term.  

Apart from packagegrounds, the only recipe in oe-core that mentions a
dependency on apmd is matchbox-panel (for the battery applet
presumably); that dependency is conditional on having apm in
MACHINE_FEATURES (which it isn't by default), and it seems a bit fishy
anyway since I can't immediately think of a reason why matchbox-panel
would need the actual daemon.

So, I think it should probably be safe to just rescind support for both
apmd and the MACHINE_FEATURE apm in oe-core.  If meta-handheld, meta-oe
or anybody else wants to keep apmd then they can do so in their own
layers although I rather hope that nobody does.

p.

Re: image.bbclass: USE_DEVFS is now useless

[Laurentiu Palcu]

Hi Matthieu,

On Wed, Feb 26, 2014 at 11:11:28AM +0000, Matthieu CRAPET wrote:
> Hi,
> 
>  
> 
> Since :
> 
> http://cgit.openembedded.org/openembedded-core/commit/meta/classes/
> image.bbclass?id=a83144bac8d67704ff66f5dc0fc56f5b63979694
> 
>  
> 
> USE_DEVFS is not considered anymore.
> 
>  
> 
> Setting IMAGE_DEVICE_TABLE or IMAGE_DEVICE_TABLES to empty string in an image
> recipe works (makedevs is not called)
> 
>  
> 
> Should be we drop USE_DEVFS variable (meta/classes/image.bbclass) or (re)add
> check in meta/lib/oe/rootfs.py?
> 
> I can propose a patch.
Please do. It looks like I totally missed the USE_DEVFS check...

laurentiu
> 
>  
> 
> Regards,
> 
> Matthieu
> 

> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core@lists.openembedded.org
> http://lists.openembedded.org/mailman/listinfo/openembedded-core

Re: image.bbclass: USE_DEVFS is now useless

[Paul Eggleton]

On Tuesday 01 April 2014 22:29:02 Phil Blundell wrote:
> On Tue, 2014-04-01 at 18:51 +0100, Richard Purdie wrote:
> > On Tue, 2014-04-01 at 18:48 +0100, Phil Blundell wrote:
> > > Also note that the default for USE_DEVFS was (and is) 1, so the lack
> > > of this check is actually causing a difference in the default
> > > behaviour.  If there's no appetite for reinstating the USE_DEVFS
> > > mechanism per se then it seems like it would be a good idea to make
> > > the default IMAGE_DEVICE_TABLE be blank in order to restore the
> > > previous default of no /dev in the rootfs.
> > > 
> > > At present you get a somewhat arbitrary-seeming smattering of devices
> > > from meta/files/device_table-minimal.txt, including such anachronisms
> > > as /dev/ttySA0 and /dev/apm_bios.  It's hard to imagine that anybody
> > > actually wants this stuff in their rootfs in this day and age.
> > 
> > Can we kill apmd at the same time? Please? :)
> 
> I can't see why not.  I'd be surprised if there were many/any real users
> left, beyond possibly some legacy stuff in meta-handheld, and it
> definitely doesn't seem "core" by any reasonable interpretation of the
> term.
> 
> Apart from packagegrounds, the only recipe in oe-core that mentions a
> dependency on apmd is matchbox-panel (for the battery applet
> presumably); that dependency is conditional on having apm in
> MACHINE_FEATURES (which it isn't by default), and it seems a bit fishy
> anyway since I can't immediately think of a reason why matchbox-panel
> would need the actual daemon.
> 
> So, I think it should probably be safe to just rescind support for both
> apmd and the MACHINE_FEATURE apm in oe-core.  If meta-handheld, meta-oe
> or anybody else wants to keep apmd then they can do so in their own
> layers although I rather hope that nobody does.

I'm generally in favour of this too, however, one of the things that apmd 
provided was the ability to have scripts run on suspend/resume - is there a 
more modern mechanism that should replace this? zaurusd and opie currently 
rely on this (yeah, I know...)

Cheers,
Paul


-- 

Paul Eggleton
Intel Open Source Technology Centre