Re: [Qemu-discuss] DMAR errors with iommu emulation enabled

[Alex Williamson]

On Tue, 7 Feb 2017 12:53:04 -0500
Jintack Lim <address@hidden> wrote:

> Thanks Alex, I appreciate your help.
> 
> On Tue, Feb 7, 2017 at 12:09 PM, Alex Williamson
> <address@hidden> wrote:
> > On Tue, 7 Feb 2017 11:46:57 -0500
> > Jintack Lim <address@hidden> wrote:
> >  
> >> Adding CC and some more information.  
> >
> > $ ./scripts/get_maintainer.pl -f ./hw/i386/intel_iommu.c
> > get_maintainer.pl: No maintainers found, printing recent contributors.
> > get_maintainer.pl: Do not blindly cc: them on patches!  Use common sense.  
> 
> Oh, I got it. Thanks!
> 
> >
> > "Michael S. Tsirkin" <address@hidden> (commit_signer:54/35=100%)
> > Peter Xu <address@hidden> (commit_signer:28/35=80%)
> > Paolo Bonzini <address@hidden> (commit_signer:6/35=17%)
> > Jason Wang <address@hidden> (commit_signer:5/35=14%)
> > "Radim Krčmář" <address@hidden> (commit_signer:5/35=14%)
> > address@hidden (open list:All patches CC here)
> >  
> >> On Tue, Feb 7, 2017 at 5:07 AM, Jintack Lim <address@hidden> wrote:  
> >> > Hi,
> >> >
> >> > I'm getting DMAR errors during VM booting when I enable the iommu
> >> > emulation for the VM. I was not able to complete booting since the VM
> >> > gets really slow and just keep printing the error message (sym0:
> >> > unexpected disconnect) at a speed of one character per second.
> >> >
> >> > I have enabled the iommu emulation, but didn't assign any device to the 
> >> > VM.
> >> >
> >> > This is the kernel log from the VM  
> >>
> >> I'm using 4.6.0-rc5+ kernel for the host and the VM.
> >>
> >> Here's the full kernel log from the VM.
> >> https://paste.ubuntu.com/23948597/
> >>  
> >> >
> >> > [    6.087794] sym0: SCSI BUS has been reset.
> >> > [    6.087960] DMAR: DRHD: handling fault status reg 2
> >> > [    6.088001] DMAR: DMAR:[DMA Read] Request device [04:03.0] fault
> >> > addr fe281000
> >> > [    6.088001] DMAR:[fault reason 06] PTE Read access is not set
> >> > [    6.090513] scsi host1: sym-2.2.3
> >> > [    6.090567] sym0: unexpected disconnect
> >> > [    8.814929] sym0: unexpected disconnect
> >> > [   11.670251] sym0: unexpected disconnect
> >> >
> >> > I enabled iommu in the host (intel_iommu=on). I also enabled iommu in
> >> > the guest AND gave this option to the qemu (-device intel-iommu). I'm
> >> > using qemu 2.8.0 and libvirt 3.0.0.
> >> > I used <qemu:commandline> in libvirt xml to enable iommu emulation.
> >> > Here's the full libvirt xml.
> >> > http://paste.ubuntu.com/23946803/
> >> >
> >> > I did lspci -vvv and 04:03:0 is scsi device. Unfortunately, I lost
> >> > that information, and can't boot the VM now. I'll add this information
> >> > later if necessary.  
> >>
> >> This is information about 04:03:0
> >> I got this from the another identical VM but not with the iommu emulation.
> >>
> >> address@hidden:~# lspci -vvs 04:03.0
> >> 04:03.0 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
> >> Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
> >> Stepping- SERR+ FastB2B- DisINTx-
> >> Status: Cap- 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort-
> >> <TAbort- <MAbort- >SERR- <PERR- INTx-
> >> Latency: 255, Cache Line Size: 64 bytes
> >> Interrupt: pin A routed to IRQ 23
> >> Region 0: I/O ports at c000 [size=256]
> >> Region 1: Memory at fe284000 (32-bit, non-prefetchable) [size=1K]
> >> Region 2: Memory at fe280000 (32-bit, non-prefetchable) [size=8K]
> >> Kernel driver in use: sym53c8xx
> >>  
> >> >
> >> > Any thoughts why this happens and how to fix?  
> >
> > Try a different disk controller in the VM?  Running virtual VT-d can't
> > automatically fix guest drivers that don't handle devices behind an
> > IOMMU correctly.  I don't know if that's the case, but I imagine 53c895a
> > has probably never been tested w/ VT-d emulation.  
> 
> Ok, I'll try another disk controller. If someone can suggest other
> disk controller which is working well with the iommu emulation, then
> it would be great. I guess I got the controller by using this option
> in virsh-install (--disk=/temp-space/guest50.img,size=10,bus=scsi),
> but not 100% sure.
> 
> > Note that when you
> > do get to the point off assigning a device to the VM, you're going to
> > need to build your own QEMU with patches from the mailing list, or
> > maybe wait a few days and they might make it to the git tree.  Thanks,  
> 
> I'm afraid I understand this.

lol ;)

> I was able to assign a network device to
> the VM without the iommu emulation. If I want to assign a network
> device to the VM with the iommu emulation, then I need to apply
> patches from the mailing list?

Yes, the version of intel-iommu in QEMU 2.8 does not work correctly
with assigned devices.  This is the latest posting:

https://lists.gnu.org/archive/html/qemu-devel/2017-02/msg01331.html