Re: Kdump kernel hang when set amd_iommu=off

Re: Kdump kernel hang when set amd_iommu=off

[Joerg Roedel]

Hi Baoquan,

On Fri, Jan 26, 2018 at 04:06:22PM +0800, Baoquan He wrote:
> Saw Huawei's bug report about kdump kernel hang when intel_iommu=off 
> is set. I met the similar problem in amd system, only set amd_iommu=off
> in kdump kernel, it means amd_iommu is on in 1st kernel.
> 
> I am reading doc and code, trying to debug. Please help have a look when
> convenient, see if there's hint or direction.

Yeah, I know that this is broken on the AMD-IOMMU side. Problem is that
the IOMMU hardware needs to be disabled in the kdump kernel when
amd_iommu=off is on the command-line. I'll look into fixing that.

I thought we fixed this a while ago already for VT-d, is it broken
again?


	Joerg

Re: Kdump kernel hang when set amd_iommu=off

[Baoquan He]

Hi Joerg,

On 02/13/18 at 02:25pm, Joerg Roedel wrote:
> Hi Baoquan,
> 
> On Fri, Jan 26, 2018 at 04:06:22PM +0800, Baoquan He wrote:
> > Saw Huawei's bug report about kdump kernel hang when intel_iommu=off 
> > is set. I met the similar problem in amd system, only set amd_iommu=off
> > in kdump kernel, it means amd_iommu is on in 1st kernel.
> > 
> > I am reading doc and code, trying to debug. Please help have a look when
> > convenient, see if there's hint or direction.
> 
> Yeah, I know that this is broken on the AMD-IOMMU side. Problem is that
> the IOMMU hardware needs to be disabled in the kdump kernel when
> amd_iommu=off is on the command-line. I'll look into fixing that.

Thanks for looking into this. This was found by our QE, I checked, and
found it only happened on two HPE systems. Other AMD systems have no
such issue. Since it's not a blocker, I just go to fix those urgent
bugs.

> 
> I thought we fixed this a while ago already for VT-d, is it broken
> again?

It's not seen in redhat's test machines, just saw Huawei engineers
reported this issue. They cc-ed lkml.

Thanks
Baoquan