Re: [linux-yocto] Removing DISTRO_FEATURES from a specific machine type

Re: [linux-yocto] Removing DISTRO_FEATURES from a specific machine type

[Richard Purdie]

On Mon, 2022-11-14 at 22:05 +0530, Pintu Agarwal wrote:
> Hi,
> 
> This question is related to Yocto Dunfell specifically.
> 
> I have certain distro features defined (by default) in our distro
> include/conf file.
> These distro features are required for certain machines and should not
> be enabled for only one machine.
> For example, we have a distro feature such as "nand-squashfs".
> So, in our distro.inc file we are including like this:
> DISTRO_FEATURES_append = " nand-squashfs"
> 
> Since this distro.inc file is shared across 2-3 machines, this feature
> will be available to all.
> But we don't want to include it in the "xyz" machine.
> 
> So, can I do like this:
> DISTRO_FEATURES_append = " nand-squashfs"
> DISTRO_FEATURES_remove_xyz = " nand-squashfs"
> 
> If there is some other better way please suggest.

Have you looked at COMBINED_FEATURES?

This allows you to only enable something if it is in both
DISTRO_FEATURES and MACHINE_FEATURES.

Cheers,

Richard

Re: [linux-yocto] Removing DISTRO_FEATURES from a specific machine type

[Jose Quaresma]

[-- Attachment #1: Type: text/plain, Size: 1514 bytes --]

Hi Pintu,

Pintu Agarwal <pintu.ping@gmail.com> escreveu no dia segunda, 14/11/2022
à(s) 16:35:

> Hi,
>
> This question is related to Yocto Dunfell specifically.
>
> I have certain distro features defined (by default) in our distro
> include/conf file.
> These distro features are required for certain machines and should not
> be enabled for only one machine.
>

For such cases Yocto have the COMBINED_FEATURES that can be more apropriate
https://docs.yoctoproject.org/3.1.20/ref-manual/ref-variables.html#term-COMBINED_FEATURES

But if the future is only used in some machine maybe the MACHINE_FEATURES
is more apropriate
https://docs.yoctoproject.org/3.1.20/ref-manual/ref-variables.html#term-MACHINE_FEATURES

In a perfect world the the DISTRO_FEATURES should be independent of the
bsp/machine used
https://docs.yoctoproject.org/3.1.20/ref-manual/ref-variables.html#term-DISTRO_FEATURES

Jose

For example, we have a distro feature such as "nand-squashfs".
> So, in our distro.inc file we are including like this:
> DISTRO_FEATURES_append = " nand-squashfs"
>
> Since this distro.inc file is shared across 2-3 machines, this feature
> will be available to all.
> But we don't want to include it in the "xyz" machine.
>
> So, can I do like this:
> DISTRO_FEATURES_append = " nand-squashfs"
> DISTRO_FEATURES_remove_xyz = " nand-squashfs"
>
> If there is some other better way please suggest.
>
> Thanks,
> Pintu
>
> 
>
>

-- 
Best regards,

José Quaresma

[-- Attachment #2: Type: text/html, Size: 2514 bytes --]

[poky] Replace a (not working) download by adding a local file as mirror?

[Steve]

Hi, maybe someone can help, although my problem occurs outside poky but
in Yocto general. But I think solution might be same.
I do have a problem with a download (see below the exact failure). I am
messing around with that for quite a while and couldn't find a solution.
I am wondering if I can put this file in a recipe as a local file:

FILESEXTRAPATH:prepend := "${THISDIR}/files/:"

after download that file externally in browser (that works)
Not sure will that work or do I have to add a real local mirror?


That happens while building tensorflow:
I don't find a solution to fix. No problem to download that file:
https://mirror.bazel.build/openjdk/azul-zulu11.50.19-ca-jdk11.0.12/zulu11.50.19-ca-jdk11.0.12-linux_x64.tar.gz
in my browser.
I found some hints to the certification error but nothing helped.
I am running a VM Ubuntu 20.04.


|     Fetching @remotejdk11_linux; fetching 13s
| INFO: Repository remotejdk11_linux instantiated at:
|   /DEFAULT.WORKSPACE.SUFFIX:41:6: in <toplevel>
|
/home/user/y/yocto40/my_visor/bazel/bazel_5ULmqVIS/out/external/bazel_tools/tools/build_defs/repo/utils.bzl:233:18:
in maybe
|
/home/user/y/yocto40/my_visor/bazel/bazel_5ULmqVIS/out/external/bazel_tools/tools/jdk/remote_java_repository.bzl:48:17:
in remote_java_repository
| Repository rule http_archive defined at:
|
/home/user/y/yocto40/my_visor/bazel/bazel_5ULmqVIS/out/external/bazel_tools/tools/build_defs/repo/http.bzl:353:31:
in <toplevel>
| Analyzing: target //src:bazel_nojdk (367 packages loaded, 9300 targets
configu\
| red)
|     Fetching @remotejdk11_linux; fetching 13s
| WARNING: Download from
https://mirror.bazel.build/openjdk/azul-zulu11.50.19-ca-jdk11.0.12/zulu11.50.19-ca-jdk11.0.12-linux_x64.tar.gz
failed: class javax.net.ssl.SSLHandshakeException PKIX path building
failed: sun.security.provider.certpath.SunCertPathBuilderException:
unable to find valid certification path to requested target
| Analyzing: target //src:bazel_nojdk (367 packages loaded, 9300 targets
configu\
| red)
| ERROR: An error occurred during the fetch of repository
'remotejdk11_linux':
|    Traceback (most recent call last):
|        File
"/home/user/y/yocto40/my_visor/bazel/bazel_5ULmqVIS/out/external/bazel_tools/tools/build_defs/repo/http.bzl",
line 100, column 45, in _http_archive_impl
|                download_info = ctx.download_and_extract(
| Error in download_and_extract: java.io.IOException: Error downloading
[https://mirror.bazel.build/openjdk/azul-zulu11.50.19-ca-jdk11.0.12/zulu11.50.19-ca-jdk11.0.12-linux_x64.tar.gz]
to
/home/user/y/yocto40/my_visor/bazel/bazel_5ULmqVIS/out/external/remotejdk11_linux/temp2967262093456757762/zulu11.50.19-ca-jdk11.0.12-linux_x64.tar.gz:
PKIX path building failed:
sun.security.provider.certpath.SunCertPathBuilderException: unable to
find valid certification path to requested target
| Analyzing: target //src:bazel_nojdk (367 packages loaded, 9300 targets
configu\
| red)

Re: [poky] Replace a (not working) download by adding a local file as mirror?

[Richard Purdie]

On Wed, 2022-11-16 at 17:39 +0100, Steve wrote:
> Hi, maybe someone can help, although my problem occurs outside poky but
> in Yocto general. But I think solution might be same.
> I do have a problem with a download (see below the exact failure). I am
> messing around with that for quite a while and couldn't find a solution.
> I am wondering if I can put this file in a recipe as a local file:
> 
> FILESEXTRAPATH:prepend := "${THISDIR}/files/:"
> 
> after download that file externally in browser (that works)
> Not sure will that work or do I have to add a real local mirror?
> 
> 
> That happens while building tensorflow:
> I don't find a solution to fix. No problem to download that file:
> https://mirror.bazel.build/openjdk/azul-zulu11.50.19-ca-jdk11.0.12/zulu11.50.19-ca-jdk11.0.12-linux_x64.tar.gz
> in my browser.
> I found some hints to the certification error but nothing helped.
> I am running a VM Ubuntu 20.04.
> 

Normally you'd place the file into DL_DIR and it should then work from
there?

Cheers,

Richard

Re: [poky] Replace a (not working) download by adding a local file as mirror?

[Steve]

Am 16.11.22 um 18:08 schrieb Richard Purdie:
> On Wed, 2022-11-16 at 17:39 +0100, Steve wrote:
>> Hi, maybe someone can help, although my problem occurs outside poky but
>> in Yocto general. But I think solution might be same.
>> I do have a problem with a download (see below the exact failure). I am
>> messing around with that for quite a while and couldn't find a solution.
>> I am wondering if I can put this file in a recipe as a local file:
>>
>> FILESEXTRAPATH:prepend := "${THISDIR}/files/:"
>>
>> after download that file externally in browser (that works)
>> Not sure will that work or do I have to add a real local mirror?
>>
>>
>> That happens while building tensorflow:
>> I don't find a solution to fix. No problem to download that file:
>> https://mirror.bazel.build/openjdk/azul-zulu11.50.19-ca-jdk11.0.12/zulu11.50.19-ca-jdk11.0.12-linux_x64.tar.gz
>> in my browser.
>> I found some hints to the certification error but nothing helped.
>> I am running a VM Ubuntu 20.04.
>>
>
> Normally you'd place the file into DL_DIR and it should then work from
> there?
>
> Cheers,
>
> Richard

Sounds both logically and easy. But if I remember right that doesn't
work (some time back I tried that)
Maybe fetcher remembers whether is has to download it or not.
I am not in how exactly that works, but isn't is stored in sstate?

Bad luck - can't test it right now as another build is running. The DL
folder is structured with subfolders git2, svn, uninitiative - but many
*tar.gz (or other archive-) files are just there, together with a
463byte binary file *.tar.gz.done.

Best regards, Steve

[poky] gcc 11.3 bug (gcc bugzilla 101723) on armv7

[Steve]

Hi,

I found that GCC 11.3 that is part of latest kirkstone 4.0.5 has a
problem building tensorflow-lite. (git.yoctoproject.org/meta-tensorflow)

Problem is similar as described here:
https://github.com/google/XNNPACK/issues/2003

There are options set to compile ( -march=armv8.2-a+dotprod
-mfpu=neon-fp-armv8) that doesn't match to the processor (armv7)

Following this issue
and points to  patch in
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101723

Which is related to gcc 9. But couldn't it be that it is not fixed in
11.3 too?

How is able to fix it or probably better ask: what can I do to get it
working? Honestly, I don't understand much of it.

Best regards, Steve

Re: [poky] gcc 11.3 bug (gcc bugzilla 101723) on armv7

[Khem Raj]

On Thu, Nov 17, 2022 at 10:27 PM Steve <star@gmx.li> wrote:
>
> Hi,
>
> I found that GCC 11.3 that is part of latest kirkstone 4.0.5 has a
> problem building tensorflow-lite. (git.yoctoproject.org/meta-tensorflow)
>
> Problem is similar as described here:
> https://github.com/google/XNNPACK/issues/2003
>
> There are options set to compile ( -march=armv8.2-a+dotprod
> -mfpu=neon-fp-armv8) that doesn't match to the processor (armv7)
>
> Following this issue
> and points to  patch in
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101723
>
> Which is related to gcc 9. But couldn't it be that it is not fixed in
> 11.3 too?

Its in gcc11 as well see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101723#c5

>
> How is able to fix it or probably better ask: what can I do to get it
> working? Honestly, I don't understand much of it.

are you seeing it on yocto master ?

>
> Best regards, Steve
>
>
> 
>

Re: [poky] gcc 11.3 bug (gcc bugzilla 101723) on armv7

[Steve]

Am 18.11.22 um 20:53 schrieb Khem Raj:
> On Thu, Nov 17, 2022 at 10:27 PM Steve <star@gmx.li> wrote:
>>
>> Hi,
>>
>> I found that GCC 11.3 that is part of latest kirkstone 4.0.5 has a
>> problem building tensorflow-lite. (git.yoctoproject.org/meta-tensorflow)
>>
>> Problem is similar as described here:
>> https://github.com/google/XNNPACK/issues/2003
>>
>> There are options set to compile ( -march=armv8.2-a+dotprod
>> -mfpu=neon-fp-armv8) that doesn't match to the processor (armv7)
>>
>> Following this issue
>> and points to  patch in
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101723
>>
>> Which is related to gcc 9. But couldn't it be that it is not fixed in
>> 11.3 too?
>
> Its in gcc11 as well see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101723#c5
>
>>
>> How is able to fix it or probably better ask: what can I do to get it
>> working? Honestly, I don't understand much of it.
>
> are you seeing it on yocto master ?

Good point. I need some time to check that. But even if it works, it is
not a real solution. I need to build a defined state.
The fix is not a new one. Wondering why it's not in (or I am wrong and
it is just similar).

Best regards,
Stece

Re: [poky] gcc 11.3 bug (gcc bugzilla 101723) on armv7

[Steve]

Am 20.11.22 um 17:59 schrieb Steve:
> Am 18.11.22 um 20:53 schrieb Khem Raj:
>> On Thu, Nov 17, 2022 at 10:27 PM Steve <star@gmx.li> wrote:
>>>
>>> Hi,
>>>
>>> I found that GCC 11.3 that is part of latest kirkstone 4.0.5 has a
>>> problem building tensorflow-lite. (git.yoctoproject.org/meta-tensorflow)
>>>
>>> Problem is similar as described here:
>>> https://github.com/google/XNNPACK/issues/2003
>>>
>>> There are options set to compile ( -march=armv8.2-a+dotprod
>>> -mfpu=neon-fp-armv8) that doesn't match to the processor (armv7)
>>>
>>> Following this issue
>>> and points to  patch in
>>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101723
>>>
>>> Which is related to gcc 9. But couldn't it be that it is not fixed in
>>> 11.3 too?
>>
>> Its in gcc11 as well see
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101723#c5
>>
>>>
>>> How is able to fix it or probably better ask: what can I do to get it
>>> working? Honestly, I don't understand much of it.
>>
>> are you seeing it on yocto master ?
>
> Good point. I need some time to check that. But even if it works, it is
> not a real solution. I need to build a defined state.
> The fix is not a new one. Wondering why it's not in (or I am wrong and
> it is just similar).
>
> Best regards,
> Steve

sorry, can't test Master right now, it is  another gcc (12.2 instead of
11.3) - but I need to be in 11.3. Please, if possible ad that patch to
kirkstone release.