iommu must be enabled for PVH hardware domain

[Roger Pau Monné]

Hello,

El 13/05/15 a les 10.04, Ruslan Makhmatkhanov ha escrit:
> Roger Pau Monné wrote on 05/12/15 18:50:
>> Hello,
>>
>> El 12/05/15 a les 17.20, Ruslan Makhmatkhanov ha escrit:
>>> Hello,
>>>
>>> we have machine with VT-d and EPT-capable CPU:
>>> Intel(R) Xeon(R) CPU E5504 [1]
>>>
>>> The system is FreeBSD 11.0-CURRENT r282694 installed from
>>> ftp.freebsd.org iso.
>>>
>>> But dom0 fails to boot at this hardware with "iommu must be enabled for
>>> PVH hardware domain" [2]. VT-d support is indeed not detected at kernel
>>> level. Here is dmesg with VT-d enabled in BIOS [3] and dmesg with VT-d
>>> disabled in BIOS [4]. It's look like there is no difference on both
>>> outputs. Here is the proofpic of that the system is VT-d capable and the
>>> support is enabled in BIOS [5]. Here is acpidump -t output [6]
>>
>> My bet right now is be that Xen detects that your IOMMU contains erratas
>> and disables it. Can you add "iommu=debug,force" to your xen_cmdline and
>> try again?
>>
>> Also, can you paste the contents of your /boot/loader.conf?
>>
>> Roger.
> 
> You are right:
> 
> [VT-D] Disabling IOMMU due to Intel 5500/5520/X58 Chipset errata #47, #53
> 
> and after that:
> 
> ----------------------------------------------
> Panic on CPU 0:
> Couldn't enable IOMMU and iommu=required/force
> ----------------------------------------------
> 
> Reboot in five seconds.
> 
> But what this means practically? Some chipset bugs preventing us to use
> this feature. Can this circumvented somehow? Thanks.

AFAIK iommu=force should force the usage of the IOMMU even if erratas
are detected. You can also try with iommu=force,no-intremap.

Roger.