RE: [Xen-devel] RE: [Xen-changelog] [xen-unstable]vt-d: Fixpanic in msi_msg_read_remap_rte with acpi=off

["Cui, Dexuan" <dexuan.cui@xxxxxxxxx>]

Cui, Dexuan wrote:
> Keir Fraser wrote:
>> On 19/10/2009 12:51, "Keir Fraser" <keir.fraser@xxxxxxxxxxxxx> wrote:
>> 
>>>> Got a backtrace and Xen boot params? If you pass acpi=off, then
>>>> disable_acpi() is invoked, and this sets acpi_disabled. If
>>>> acpi_disabled=1, then iommu_setup() sets iommu_enabled=0. If
>>>> iommu_enabled=0 then I think all the update_ire_from_* and similar
>>>> hooks get disabled in the callers. So something unexpected must be
>>>> happening.
>>> 
>>> Hmmm, perhaps this could be an early boot-time crash before
>>> iommu_setup() is even called? Perhaps moving the if(acpi_disabled)
>>> iommu_enabled=0 somewhere early-ish in setup.c:__start_xen(), with a
>>> warning message, would be the right thing to do in that case (e.g.,
>>> immediately after the call to cmdline_parse()).
>> 
>> After some more hmmm'ing I have to admit your original patch was
>> actually the best way to go. :-)
>> 
>> All other callers of acpi_find_matched_drhd_unit() that you didn't
>> patch already check for NULL return, so it's presumably expected as a
>> possibility. And it is okay for msi_msg_{read,write}_remap_rte() to
>> bail silently if they cannot determine there's any work to do, since
>> they operate as potential modifiers to operations which are
>> primarily implemented in their callers. 
>> 
>> So sorry about that. I'll revert Dexuan's patch.
>> 
>>  -- Keir
> 
> But, can't you reproduce the crash I mentioned before?
> Please see the attached crash log -- I'm using c/s 20341:ea34183c5c11
> and with "iommu=1 acpi=off" and I use a DQ35 host. 
> 
> Actually what I care is the " if ( acpi_disabled ) iommu_enabled = 0".

BTW: from my crash log, you can see the bogus info -- actually the host doesn't 
support SC, QI and IR.
(XEN) Intel VT-d Snoop Control supported.
(XEN) Intel VT-d DMA Passthrough not supported.
(XEN) Intel VT-d Queued Invalidation supported.
(XEN) Intel VT-d Interrupt Remapping supported.
(XEN) I/O virtualisation enabled

Thanks,
-- Dexuan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel