[dpdk-dev] Issue when the kernel parameter intel_iommu=on is being used

[François-Frédéric Ozog]

Hi,

To understand the issue, you may have a look at:
http://www.intel.com/content/www/us/en/intelligent-systems/intel-technology/
vt-directed-io-spec.html

When you have no IOMMU, "physical" address space is accessed directly by
hardware, so your core works.

When VT-d is active, there is DMA/IRQ remapping hardware layer between the
device and the memory/cpu. If you look at §3.4.3 of the spec, you that for
each device of each bus there is a context (enumerated at boot time,
leveraging BIOS/ACPI). For each device, you may have address translation
programmed so that DMA produced by hardware is actually mapped to a physical
address.

When you use the Linux kernel API for mapping DMA memory, Linux takes care
of the "details". 

For DPDK, documentation §5.6 Using Linux IOMMU Pass-Through to Run Intel®
DPDK with Intel® VT-d says that you should have iommu=pt kernel parameter
on. Do you have it ?

FF


> -----Message d'origine-----
> De : dev [mailto:dev-bounces at dpdk.org] De la part de Sridhar S
> Envoyé : dimanche 5 janvier 2014 13:38
> À : dev at dpdk.org
> Objet : [dpdk-dev] Issue when the kernel parameter intel_iommu=on is being
> used
> 
> Hello,
> 
> 
> 
> I am using DPDK 1.5 for development of host pmd for device “Connect X3”.
> 
> 
> 
> I am observing issue  while the ConnectX3 device DMA to a memory which is
> allocated with rte_memzone_reserve_aligned() API .
> 
> The issue(please refer ERROR below) has been observed if the system runs
> with the kernel parameter “intel_iommu=on”.
> 
> 
> 
> ERROR :
> 
> dmar: DRHD: handling fault status reg 302
> 
> dmar: DMAR:[DMA Write] Request device [01:00.0] fault addr 4f883000
> 
> DMAR:[fault reason 01] Present bit in root entry is clear
> 
> 
> The reported "fault Addr" is the physical address which was returned by
the
> Above API.
> 
> 
> 
> I don’t see any issue with the same code when the system up with kernel
> parameter intel_iommu=off.
> 
> 
> 
> 
> Can you share your comments on this issue?
> 
> 
> Thanks in advance
> 
> Sri