Re: AW: Xenalyze on ARM ( NXP S32G3 with Cortex-A53)

[Julien Grall <julien@xxxxxxx>]

On 13/01/2023 12:56, El Mesdadi Youssef ESK UILD7 wrote:
> Hello Julien,
Hi,

> > > > xentrace should work on upstream Xen. What did you version did you try?
> While building my image using the BSP-linux of NXP, the version that was 
> downloaded is Xen 4.14.
Do you know where the source are downloaded from?

> > > > Can you also clarify the error you are seen?
> The error I receive while tipping xentrace is: Command not found.

"Command not found" means the program hasn't been installed.

> I assume in this Xen version, Xentrace is not compatible with ARM architecture
The support for xentrace on Arm has been added around Xen 4.12. So it 
should work for Xen 4.14 (even though I don't recommend using older 
release).
I would suggest to check how you are building the tools and which 
package will be installed.
> My question is, is there any new version of Xen that supports Xentrace on ARM? If yes how 
> could I install it? Because Xen.4.14 was installed automatically by typing this ( 
> DISTRO_FEATURES_append += "xen") in the local.conf file while creating my image.
I am not familiar with the BSP-linux of NXP as this is not a project 
maintained by Xen Project.
If you need support for it, then I would suggest to discuss with NXP 
directly.
> Or is there any source on Xentrace that is compatible with ARM on GitHub, that 
> I could download and compile myself?
Even if the hypervisor is Xen, you seem to use code provided by an 
external entity. I can't advise on the next steps without knowing the 
modification that NXP made in the hypervisor.
> > > > Yes if you assign (or provide para-virtualized driver) the 
> > > > GPIO/LED/Can-Interface to the guest.
> Is there any tutorial that could help me create those drivers? And how 
> complicated is that, to create them?
I am not aware of any tutorial. Regarding the complexity, it all depends 
on what exactly you want to do.
> Or can they be assigned just with PCI-Passthrough?
PCI passthrough is not yet supported on Arm. That said, I was not 
expecting the GPIO/LED/Can-interface to be PCI devices.
If they are platform devices (i.e non-PCI devices), then you could 
potentially directly assign them to *one* guest (this would not work if 
you need to share them).
I wrote potentially because if the device is DMA-capabable, then the 
device must be behind an IOMMU.
Cheers,

--
Julien Grall