[QUESTION] Minimal Xen Configuration

[QUESTION] Minimal Xen Configuration

[Kwon, Gihwan]

Dear Xen Community,

I hope this message finds you well.

I have heard that there exists a minimal Xen configuration optimized for 
safety-critical products, particularly in automotive applications, with the 
code size reduced to approximately 50k SLOC.

Could anyone provide guidance or point me to relevant resources regarding 
this minimal Xen configuration? Any insights or references would be greatly
appreciated.

Thank you in advance for your assistance.

Best regards,
Gihwan Kwon

Re: [QUESTION] Minimal Xen Configuration

[Nicola Vetrini]

On 2025-03-28 07:43, Kwon, Gihwan wrote:
> Dear Xen Community,
> 
> I hope this message finds you well.
> 
> I have heard that there exists a minimal Xen configuration optimized 
> for
> safety-critical products, particularly in automotive applications, with 
> the
> code size reduced to approximately 50k SLOC.
> 
> Could anyone provide guidance or point me to relevant resources 
> regarding
> this minimal Xen configuration? Any insights or references would be 
> greatly
> appreciated.
> 
> Thank you in advance for your assistance.
> 
> Best regards,
> Gihwan Kwon

Hello,

I am part of the team that provides static analysis for Xen, mainly 
towards the objective to make Xen compliant with MISRA C guidelines. As 
part of that effort, a minimal Xen configuration has been defined (for 
Arm and x86_64) at [1], which is the one currently analyzed for static 
analysis for each commit [2]. It starts from the default configuration 
for the architecture (i.e., make defconfig) and then turns on or off 
various Kconfig options, as specified by EXTRA_XEN_CONFIG.

@Stefano Stabellini (Cc'ed) is the main driving force behind the Xen 
certification effort for automotive, therefore he's probably the best 
person to approach to get more information on this matter.

Hope this was helpful as a starting point.

Best Regards,
  Nicola Vetrini

[1] 
https://gitlab.com/xen-project/xen/-/blob/staging/automation/gitlab-ci/analyze.yaml?ref_type=heads
[2] https://gitlab.com/xen-project/hardware/xen/-/pipelines

-- 
Nicola Vetrini, B.Sc.
Software Engineer
BUGSENG (https://bugseng.com)
LinkedIn: https://www.linkedin.com/in/nicola-vetrini-a42471253

Re: [QUESTION] Minimal Xen Configuration

[Stefano Stabellini]

On Wed, 2 Apr 2025, Nicola Vetrini wrote:
> On 2025-03-28 07:43, Kwon, Gihwan wrote:
> > Dear Xen Community,
> > 
> > I hope this message finds you well.
> > 
> > I have heard that there exists a minimal Xen configuration optimized for
> > safety-critical products, particularly in automotive applications, with the
> > code size reduced to approximately 50k SLOC.
> > 
> > Could anyone provide guidance or point me to relevant resources regarding
> > this minimal Xen configuration? Any insights or references would be greatly
> > appreciated.
> > 
> > Thank you in advance for your assistance.
> > 
> > Best regards,
> > Gihwan Kwon
> 
> Hello,
> 
> I am part of the team that provides static analysis for Xen, mainly towards
> the objective to make Xen compliant with MISRA C guidelines. As part of that
> effort, a minimal Xen configuration has been defined (for Arm and x86_64) at
> [1], which is the one currently analyzed for static analysis for each commit
> [2]. It starts from the default configuration for the architecture (i.e., make
> defconfig) and then turns on or off various Kconfig options, as specified by
> EXTRA_XEN_CONFIG.
> 
> @Stefano Stabellini (Cc'ed) is the main driving force behind the Xen
> certification effort for automotive, therefore he's probably the best person
> to approach to get more information on this matter.

Thanks Nicola. In addition to what Nicola wrote, there is an outstanding
activity to further reduce the code size by adding more kconfig options.
There are a couple of patch series on the list right now to that effect.

https://marc.info/?l=xen-devel&m=173745442831026

Last year we ran some experiments and for ARM we should be able to go below
35K LOC. The tools measuring LOC are very imprecise as they do not
account for #ifdefs and kconfig options. So we switched to text binary
size for measurements as it is more accurate.
do not account for kconfig options and #ifdefs. Hence, we switch to
using text size as size measurement for Xen. One of the figure was XXX.

Re: [QUESTION] Minimal Xen Configuration

[Kwon, Gihwan]

Stefano, Nicola

> On Wed, 2 Apr 2025, Nicola Vetrini wrote:
>> On 2025-03-28 07:43, Kwon, Gihwan wrote:
>>> Dear Xen Community,
>>> 
>>> I hope this message finds you well.
>>> 
>>> I have heard that there exists a minimal Xen configuration optimized
>>> for safety-critical products, particularly in automotive applications,
>>> with the code size reduced to approximately 50k SLOC.
>>> 
>>> Could anyone provide guidance or point me to relevant resources
>>> regarding this minimal Xen configuration? Any insights or references
>>> would be greatly appreciated.
>>> 
>>> Thank you in advance for your assistance.
>>> 
>>> Best regards,
>>> Gihwan Kwon
>> 
>> Hello,
>> 
>> I am part of the team that provides static analysis for Xen, mainly
>> towards the objective to make Xen compliant with MISRA C guidelines. As
>> part of that effort, a minimal Xen configuration has been defined (for
>> Arm and x86_64) at [1], which is the one currently analyzed for static
>> analysis for each commit [2]. It starts from the default configuration
>> for the architecture (i.e., make defconfig) and then turns on or off
>> various Kconfig options, as specified by EXTRA_XEN_CONFIG.
>> 
>> @Stefano Stabellini (Cc'ed) is the main driving force behind the Xen
>> certification effort for automotive, therefore he's probably the best
>> person to approach to get more information on this matter.
> 
> Thanks Nicola. In addition to what Nicola wrote, there is an outstanding
> activity to further reduce the code size by adding more kconfig options.
> There are a couple of patch series on the list right now to that effect.
> 
> https://marc.info/?l=xen-devel&m=173745442831026
> 
> Last year we ran some experiments and for ARM we should be able to go
> below 35K LOC. The tools measuring LOC are very imprecise as they do not
> account for #ifdefs and kconfig options. So we switched to text binary
> size for measurements as it is more accurate. do not account for kconfig
> options and #ifdefs. Hence, we switch to using text size as size
> measurement for Xen. One of the figure was XXX.

Thank you for sharing information regarding the current minimal config
and additional activities. I have reviewed the configuration applied to the
 CI and have a question.
 
Currently, `CONFIG_SCHED_RTDS` is enabled instead of `CONFIG_SCHED_CREDIT2`
and `CONFIG_SCHED_CREDIT`. Is this because ARINC653 scheduler cannot be
used as default scheduler, as well as, rt scheduler is smaller than 
credit2 or credit scheduler?

- Gihwan

RE: [QUESTION] Minimal Xen Configuration

[Kwon, Gihwan]

Stefano and Nicola,

> Stefano, Nicola
> 
>> On Wed, 2 Apr 2025, Nicola Vetrini wrote:
>>> On 2025-03-28 07:43, Kwon, Gihwan wrote:
>>>> Dear Xen Community,
>>>> 
>>>> I hope this message finds you well.
>>>> 
>>>> I have heard that there exists a minimal Xen configuration optimized
>>>> for safety-critical products, particularly in automotive applications,
>>>> with the code size reduced to approximately 50k SLOC.
>>>> 
>>>> Could anyone provide guidance or point me to relevant resources
>>>> regarding this minimal Xen configuration? Any insights or references
>>>> would be greatly appreciated.
>>>> 
>>>> Thank you in advance for your assistance.
>>>> 
>>>> Best regards,
>>>> Gihwan Kwon
>>> 
>>> Hello,
>>> 
>>> I am part of the team that provides static analysis for Xen, mainly
>>> towards the objective to make Xen compliant with MISRA C guidelines.
>>> As part of that effort, a minimal Xen configuration has been defined
>>> (for Arm and x86_64) at [1], which is the one currently analyzed for
>>> static analysis for each commit [2]. It starts from the default
>>> configuration for the architecture (i.e., make defconfig) and then
>>> turns on or off various Kconfig options, as specified by
>>> EXTRA_XEN_CONFIG.
>>> 
>>> @Stefano Stabellini (Cc'ed) is the main driving force behind the Xen
>>> certification effort for automotive, therefore he's probably the best
>>> person to approach to get more information on this matter.
>> 
>> Thanks Nicola. In addition to what Nicola wrote, there is an outstanding
>> activity to further reduce the code size by adding more kconfig options.
>> There are a couple of patch series on the list right now to that effect.
>> 
>> https://marc.info/?l=xen-devel&m=173745442831026
>> 
>> Last year we ran some experiments and for ARM we should be able to go
>> below 35K LOC. The tools measuring LOC are very imprecise as they do
>> not account for #ifdefs and kconfig options. So we switched to text
>> binary size for measurements as it is more accurate. do not account for
>> kconfig options and #ifdefs. Hence, we switch to using text size as
>> size measurement for Xen. One of the figure was XXX.
> 
> Thank you for sharing information regarding the current minimal config
> and additional activities. I have reviewed the configuration applied to the
>  CI and have a question.
> Currently, `CONFIG_SCHED_RTDS` is enabled instead of
> `CONFIG_SCHED_CREDIT2` and `CONFIG_SCHED_CREDIT`. Is this because
> ARINC653 scheduler cannot be used as default scheduler, as well as, rt
> scheduler is smaller than credit2 or credit scheduler?

Please disregard my previous message. I have double checked that 
mentioned configurations are set 'n'. Therefore,  `CONFIG_SCHED_NULL` 
and  `CONFIG_SCHED_CREDIT2` is being set and used in the minimal xen 
configuration.

> 
> - Gihwan
> 
>

- Gihwan