Re: PCI Passthrough Problems/Questions

[Weidong Han <weidong.han@intel.com>]

Konrad Rzeszutek Wilk wrote:
> On Mon, Oct 25, 2010 at 12:50:41PM -0600, Nick Couchman wrote:
>   
>> On Mon, 2010-10-25 at 14:48 -0400, Konrad Rzeszutek Wilk wrote:
>>     
>>> On Mon, Oct 25, 2010 at 12:33:09PM -0600, Nick Couchman wrote:
>>>       
>>>> On Mon, 2010-10-25 at 13:40 -0400, Konrad Rzeszutek Wilk wrote:
>>>>
>>>>         
>>>>> What do you see on your Xen serial output? I presume you cranked up logging:
>>>>> loglevel=all guest_lvl=all iommu=verbose on your Xen command line.
>>>>>
>>>>> Is there anything that shows up when you get the 'Failed to assign.." ?
>>>>>
>>>>>           
>>>> The only messages I get on the serial console after setting those
>>>> parameters on the xen.gz kernel line in grub (and rebooting, of course)
>>>> are the following:
>>>>
>>>> (XEN) [VT-D]iommu.c:1496: d0:PCI: unmap bdf = 2:0.0
>>>> (XEN) [VT-D]iommu.c:1364: d1:PCI: map bdf = 2:0.0
>>>> (XEN) domctl.c:848:d0 XEN_DOMCTL_assign_device: assign device (2:0.0)
>>>> failed
>>>> (XEN) event_channel.c:192:d0 EVTCHNOP failure: domain 1, error -22
>>>>         
Nick,

I think the issue is 02:00.0 was mapped twice. Could you try with below 
patch? Then post the xen log. Pls post all output of 'lspci -v' on your 
system.

diff -r eff592364826 xen/drivers/passthrough/vtd/iommu.c
--- a/xen/drivers/passthrough/vtd/iommu.c   Wed Sep 01 11:23:49 2010 +0100
+++ b/xen/drivers/passthrough/vtd/iommu.c   Wed Oct 27 14:33:23 2010 +0800
@@ -1225,7 +1225,11 @@ static int domain_context_mapping_one(
         if (!pdev)
             res = -ENODEV;
         else if (pdev->domain != domain)
-            res = -EINVAL;
+        {
+            dprintk(VTDPREFIX, "context_present: %x:%x.%x: 
pdev->domain=%d domain=%d\n",
+                    bus, PCI_SLOT(devfn), PCI_FUNC(devfn), 
pdev->domain, domain->domain_id);
+        //    res = -EINVAL;
+        }
         unmap_vtd_domain_page(context_entries);
         spin_unlock(&iommu->lock);
         return res;
@@ -1352,6 +1356,8 @@ static int domain_context_mapping(struct
         /* PCIe to PCI/PCIx bridge */
         if ( pdev_type(bus, devfn) == DEV_TYPE_PCIe2PCI_BRIDGE )
         {
+            dprintk(VTDPREFIX, "d%d:PCI: map PCIe2PCI bdf = %x:%x.%x\n",
+                    domain->domain_id, bus, PCI_SLOT(devfn), 
PCI_FUNC(devfn));
             ret = domain_context_mapping_one(domain, drhd->iommu, bus, 
devfn);
             if ( ret )
                 return ret;
@@ -1362,6 +1368,8 @@ static int domain_context_mapping(struct
              * on the secondary bus behind the bridge. Map that id
              * as well.
              */
+            dprintk(VTDPREFIX, "d%d:PCI: map secbus (%d) with devfn 0\n",
+                    domain->domain_id, secbus);
             ret = domain_context_mapping_one(domain, drhd->iommu, 
secbus, 0);
         }
         else /* Legacy PCI bridge */

>>> So, -EINVAL. How comfortable are you sticking a bunch of
>>> dprintk(VTDPREFIX, " in the drivers/passthrough/vtd/iommu.c file? Basically
>>> you need to figure which of the functions that are past line 1364
>>> are being called and return -EINVAL. 
>>>       
>> I'm happy to give it a shot...it'll take a while to get the devel
>> environment configured, as I'm using packages right now and I don't even
>> think I have a compiler on this system.  I'll report back once I get
>> that done and give that a try.
>>     
>
> Excellent. You might also want to CC Weidong (weidong.han@intel.com) in the future
> who is right now on travel and he might have better suggestions. CC-ing him on this e-mail.
>