Re: PCI-Passthrough without "iommu=soft" possible?

[Joost Roeleveld <joost@xxxxxxxxxxxx>]

Hi Mathias,

The same wiki page lists that PCI passthrough is not possible using PVH. Unless that changed ?

--

Joost

Sent from Nine

---

From: Mathias Dufresne [mathias.dufresne@xxxxxxxxx]
Sent: Tuesday, 18 October 2022 16:59
To: J. Roeleveld
Cc: [xen-users@xxxxxxxxxxxxxxxxxxxx]
Subject: Re: PCI-Passthrough without "iommu=soft" possible?

Hi Joost,

I'm very not a Xen expert but according to https://wiki.xenproject.org/wiki/Linux_PVH which you gave in another mail, your kernel needs to be decently recent and configured with the following:

CONFIG_HYPERVISOR_GUEST=y
CONFIG_PARAVIRT=y
CONFIG_PARAVIRT_GUEST=y
CONFIG_PARAVIRT_SPINLOCKS=y
CONFIG_XEN=y
CONFIG_XEN_PVHVM=y
CONFIG_XEN_PVH=y

As you did not mention anything about the kernels you were testing except their version, as the one not working is the most recent, I'd suggest you check in 5.15.72 if these options are configured or not.

Most kernels are configured to give access to its configuration through /proc/config.gz, once the kernel is running on some system. A grep on that file once uncompressed should answer that question.

Cheers,

mathias

Le mar. 18 oct. 2022 à 13:48, J. Roeleveld <joost@xxxxxxxxxxxx> a écrit :

Hi all,

When trying to boot a guest domain with 2 HBA's assigned via PCI-passthrough I
get a kernel-panic with 5.15.72.
If I add "iommu=soft" to the kernel command line it works, but locks up
quickly with messages like:
[Sun Oct 16 20:33:36 2022] mpt3sas 0000:81:00.0: swiotlb buffer is full (sz:
319488 bytes), total 32768 (slots), used 68 (slots)
[Sun Oct 16 20:33:36 2022] sd 0:0:26:0: scsi_dma_map failed: request for
708608 bytes!

in the logs (dmesg)

If I stay with 5.15.59 (without "iommu=soft", it is running stable, but I
would like to upgrade to newer kernels when possible.

I am currently running Xen 4.15.3

Is there an additional setting I need to add or any other options I haven't
been able to find through a few days of google-searches that are advised?

Many thanks,

Joost Roeleveld