tmp/work-shared and sstate

tmp/work-shared and sstate

[Gary Thomas]

If I run a build where the kernel package is brought in via
sstate, tmp/work-shared (in particular the kernel-source tree)
is not populated.  This will break at least these recipes:
   meta-fsl-arm/recipes-multimedia/gstreamer/gst-fsl-plugin_4.0.2.bb
   meta-fsl-arm/recipes-multimedia/alsa/fsl-alsa-plugins_1.0.25.bb

These programs reference the kernel includes directly for some
ARM/i.MX specific headers (e.g. <linux/mxcfb.h>).  These headers
are not part of the mainline kernel which is used to create the
kernel headers that populates tmp/sysroots, so the build fails.
Note: I'm not sure of the mechanism that lets these programs
peek into the kernel build (I looked at them but nothing jumped
out), but they do build find if the kernel is actually built
and not just brought in by sstate.

Is this an error & if so, which recipe is at fault?  The FSL
recipes, or the new kernel build/classes?

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: [meta-freescale] tmp/work-shared and sstate

[Christopher Larson]

[-- Attachment #1: Type: text/plain, Size: 1414 bytes --]

On Tue, Feb 10, 2015 at 10:28 AM, Gary Thomas <gary@mlbassoc.com> wrote:

> If I run a build where the kernel package is brought in via
> sstate, tmp/work-shared (in particular the kernel-source tree)
> is not populated.  This will break at least these recipes:
>   meta-fsl-arm/recipes-multimedia/gstreamer/gst-fsl-plugin_4.0.2.bb
>   meta-fsl-arm/recipes-multimedia/alsa/fsl-alsa-plugins_1.0.25.bb
>
> These programs reference the kernel includes directly for some
> ARM/i.MX specific headers (e.g. <linux/mxcfb.h>).  These headers
> are not part of the mainline kernel which is used to create the
> kernel headers that populates tmp/sysroots, so the build fails.
> Note: I'm not sure of the mechanism that lets these programs
> peek into the kernel build (I looked at them but nothing jumped
> out), but they do build find if the kernel is actually built
> and not just brought in by sstate.
>
> Is this an error & if so, which recipe is at fault?  The FSL
> recipes, or the new kernel build/classes?
>

Per commit 46cdaf1c7bc597735d926af6a46f9483f7e57ce5
(oe-core 6a1ff0e7eacef595738f2fed086986fd622ec32a), you need to add this if
you depend on the sources:

    do_configure[depends] += "virtual/kernel:do_shared_workdir"
-- 
Christopher Larson
clarson at kergoth dot com
Founder - BitBake, OpenEmbedded, OpenZaurus
Maintainer - Tslib
Senior Software Engineer, Mentor Graphics

[-- Attachment #2: Type: text/html, Size: 2139 bytes --]

Re: tmp/work-shared and sstate

[Christopher Larson]

[-- Attachment #1: Type: text/plain, Size: 1414 bytes --]

On Tue, Feb 10, 2015 at 10:28 AM, Gary Thomas <gary@mlbassoc.com> wrote:

> If I run a build where the kernel package is brought in via
> sstate, tmp/work-shared (in particular the kernel-source tree)
> is not populated.  This will break at least these recipes:
>   meta-fsl-arm/recipes-multimedia/gstreamer/gst-fsl-plugin_4.0.2.bb
>   meta-fsl-arm/recipes-multimedia/alsa/fsl-alsa-plugins_1.0.25.bb
>
> These programs reference the kernel includes directly for some
> ARM/i.MX specific headers (e.g. <linux/mxcfb.h>).  These headers
> are not part of the mainline kernel which is used to create the
> kernel headers that populates tmp/sysroots, so the build fails.
> Note: I'm not sure of the mechanism that lets these programs
> peek into the kernel build (I looked at them but nothing jumped
> out), but they do build find if the kernel is actually built
> and not just brought in by sstate.
>
> Is this an error & if so, which recipe is at fault?  The FSL
> recipes, or the new kernel build/classes?
>

Per commit 46cdaf1c7bc597735d926af6a46f9483f7e57ce5
(oe-core 6a1ff0e7eacef595738f2fed086986fd622ec32a), you need to add this if
you depend on the sources:

    do_configure[depends] += "virtual/kernel:do_shared_workdir"
-- 
Christopher Larson
clarson at kergoth dot com
Founder - BitBake, OpenEmbedded, OpenZaurus
Maintainer - Tslib
Senior Software Engineer, Mentor Graphics

[-- Attachment #2: Type: text/html, Size: 2139 bytes --]

Re: [meta-freescale] tmp/work-shared and sstate

[Gary Thomas]

On 2015-02-10 11:13, Christopher Larson wrote:
>
> On Tue, Feb 10, 2015 at 10:28 AM, Gary Thomas <gary@mlbassoc.com <mailto:gary@mlbassoc.com>> wrote:
>
>     If I run a build where the kernel package is brought in via
>     sstate, tmp/work-shared (in particular the kernel-source tree)
>     is not populated.  This will break at least these recipes:
>        meta-fsl-arm/recipes-__multimedia/gstreamer/gst-fsl-__plugin_4.0.2.bb <http://gst-fsl-plugin_4.0.2.bb>
>        meta-fsl-arm/recipes-__multimedia/alsa/fsl-alsa-__plugins_1.0.25.bb <http://fsl-alsa-plugins_1.0.25.bb>
>
>     These programs reference the kernel includes directly for some
>     ARM/i.MX specific headers (e.g. <linux/mxcfb.h>).  These headers
>     are not part of the mainline kernel which is used to create the
>     kernel headers that populates tmp/sysroots, so the build fails.
>     Note: I'm not sure of the mechanism that lets these programs
>     peek into the kernel build (I looked at them but nothing jumped
>     out), but they do build find if the kernel is actually built
>     and not just brought in by sstate.
>
>     Is this an error & if so, which recipe is at fault?  The FSL
>     recipes, or the new kernel build/classes?
>
>
> Per commit 46cdaf1c7bc597735d926af6a46f9483f7e57ce5 (oe-core 6a1ff0e7eacef595738f2fed086986fd622ec32a), you need to add this if you depend on the sources:
>
>      do_configure[depends] += "virtual/kernel:do_shared_workdir"
> --

Thanks, I'm checking now to see if this fixes the problem.

One thing I noted is I added that line to the two recipes in
question.  When I [re]built my target image with these changes
in a tree that I had just built using only sstate, it kicked
off a ton of tasks (~6000!), and it seems that it's now rebuilding
everything from scratch, not just unpacking the kernel.  Once this
finishes, I'll try another rebuild from sstate to see how that works,
but it sure looks like it's doing a lot more than necessary.

Is this to be expected?


-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: [meta-freescale] tmp/work-shared and sstate

[Gary Thomas]

On 2015-02-10 11:35, Gary Thomas wrote:
> On 2015-02-10 11:13, Christopher Larson wrote:
>>
>> On Tue, Feb 10, 2015 at 10:28 AM, Gary Thomas <gary@mlbassoc.com <mailto:gary@mlbassoc.com>> wrote:
>>
>>     If I run a build where the kernel package is brought in via
>>     sstate, tmp/work-shared (in particular the kernel-source tree)
>>     is not populated.  This will break at least these recipes:
>>        meta-fsl-arm/recipes-__multimedia/gstreamer/gst-fsl-__plugin_4.0.2.bb <http://gst-fsl-plugin_4.0.2.bb>
>>        meta-fsl-arm/recipes-__multimedia/alsa/fsl-alsa-__plugins_1.0.25.bb <http://fsl-alsa-plugins_1.0.25.bb>
>>
>>     These programs reference the kernel includes directly for some
>>     ARM/i.MX specific headers (e.g. <linux/mxcfb.h>).  These headers
>>     are not part of the mainline kernel which is used to create the
>>     kernel headers that populates tmp/sysroots, so the build fails.
>>     Note: I'm not sure of the mechanism that lets these programs
>>     peek into the kernel build (I looked at them but nothing jumped
>>     out), but they do build find if the kernel is actually built
>>     and not just brought in by sstate.
>>
>>     Is this an error & if so, which recipe is at fault?  The FSL
>>     recipes, or the new kernel build/classes?
>>
>>
>> Per commit 46cdaf1c7bc597735d926af6a46f9483f7e57ce5 (oe-core 6a1ff0e7eacef595738f2fed086986fd622ec32a), you need to add this if you depend on the sources:
>>
>>      do_configure[depends] += "virtual/kernel:do_shared_workdir"
>> --
>
> Thanks, I'm checking now to see if this fixes the problem.
>
> One thing I noted is I added that line to the two recipes in
> question.  When I [re]built my target image with these changes
> in a tree that I had just built using only sstate, it kicked
> off a ton of tasks (~6000!), and it seems that it's now rebuilding
> everything from scratch, not just unpacking the kernel.  Once this
> finishes, I'll try another rebuild from sstate to see how that works,
> but it sure looks like it's doing a lot more than necessary.
>
> Is this to be expected?

I'm not sure why this complete rebuild happened (sadly, it took
more than 8 hours...)  When it completed, I tried another rebuild
from sstate only and things worked as expected with everything
coming in from sstate.

One concern though - the kernel package seems to have been completely
rebuilt, not just the work-shared tree being repopulated.  Is this
correct?

Thanks for the swift answer to this problem.

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------