Re: [PATCH v2 06/12] VT-d: respect ACPI SATC's ATC_REQUIRED flag

[Jan Beulich <jbeulich@xxxxxxxx>]

On 06.05.2024 15:38, Roger Pau Monné wrote:
> On Thu, Feb 15, 2024 at 11:16:11AM +0100, Jan Beulich wrote:
>> When the flag is set, permit Dom0 to control the device (no worse than
>> what we had before and in line with other "best effort" behavior we use
>> when it comes to Dom0),
> 
> I think we should somehow be able to signal dom0 that this device
> might not operate as expected, otherwise dom0 might use it and the
> device could silently malfunction due to ATS not being enabled.

Whatever signaling we invented, no Dom0 would be required to respect it,
and for (perhaps quite) some time no Dom0 kernel would even exist to query
that property.

> Otherwise we should just hide the device from dom0.

This would feel wrong to me, almost like a regression from what we had
before.

> I assume setting the IOMMU context entry to passthrough mode would
> also be fine for such devices that require ATS?

I'm afraid I'm lacking the connection of the question to what is being
done here. Can you perhaps provide some more context? To provide some
context from my side: Using pass-through mode would be excluded when Dom0
is PVH. Hence why I'm not getting why we would want to even just consider
doing so.

Yet, looking at the spec, in pass-through mode translation requests are
treated as UR. So maybe your question was towards there needing to be
handling (whichever way) for the case where pass-through mode was
requested for PV Dom0? The only half-way sensible thing to do in that case
that I can think of right now would be to ignore that command line option,
just like we do when Dom0 is PVH. Yet that would equally apply to use of
"ats" on the command line, i.e. would likely first require yet another
separate patch. Except that in the "ats" case it may be reasonable to
instead panic(), for there being conflicting requests on the command line
(and it being unclear which one would be better to ignore).

>> --- a/xen/drivers/passthrough/vtd/iommu.c
>> +++ b/xen/drivers/passthrough/vtd/iommu.c
>> @@ -2364,6 +2364,26 @@ static int cf_check intel_iommu_add_devi
>>      if ( ret )
>>          dprintk(XENLOG_ERR VTDPREFIX, "%pd: context mapping failed\n",
>>                  pdev->domain);
>> +    else if ( !pdev->broken )
>> +    {
>> +        const struct acpi_drhd_unit *drhd = 
>> acpi_find_matched_drhd_unit(pdev);
>> +        const struct acpi_satc_unit *satc = 
>> acpi_find_matched_satc_unit(pdev);
>> +
>> +        /*
>> +         * Prevent the device from getting assigned to an unprivileged 
>> domain
>> +         * when firmware indicates ATS is required, but ATS could not be 
>> enabled
>> +         * or was not explicitly enabled via command line option.
>> +         */
>> +        if ( satc && satc->atc_required &&
>> +             (!drhd || ats_device(pdev, drhd) <= 0 ||
>> +              !pci_ats_enabled(pdev->seg, pdev->bus, pdev->devfn) ||
>> +              opt_ats < 0) )
> 
> Do you need the opt_ats check here?
> 
> I don't think it's possible for pci_ats_enabled() to return true if
> opt_ats is <= 0, and hence the opt_ats < 0 check can be dropped from
> the conditional?

In the present tristate mode of opt_ats a device can have ATS enabled when
opt_ats is -1 (i.e. no command line override): For devices with ATC_REQUIRED
set.

>> @@ -2375,12 +2395,26 @@ static int cf_check intel_iommu_enable_d
>>  
>>      pci_vtd_quirk(pdev);
>>  
>> -    if ( ret <= 0 )
>> -        return ret;
>> +    if ( ret <= 0 ||
>> +         (ret = enable_ats_device(pdev, &drhd->iommu->ats_devices)) < 0 ||
>> +         opt_ats < 0 )
> 
> Shouldn't this be opt_ats <= 0?

No, again not as long as this variable is a tristate one.

>> --- a/xen/drivers/passthrough/vtd/x86/ats.c
>> +++ b/xen/drivers/passthrough/vtd/x86/ats.c
>> @@ -45,8 +45,9 @@ int ats_device(const struct pci_dev *pde
>>  {
>>      struct acpi_drhd_unit *ats_drhd;
>>      unsigned int pos, expfl = 0;
>> +    const struct acpi_satc_unit *satc;
>>  
>> -    if ( opt_ats <= 0 || !iommu_qinval )
>> +    if ( !opt_ats || !iommu_qinval )
>>          return 0;
> 
> FWIW, I find this change confusing, hence my request earlier that
> opt_ats must be set to 0 or 1 by the point it gets used.

Right, but as said in particular on the subthread of patch 5, for now it has
to remain a full tristate. Whereas if the spec was changed, I expect the
variable could be switched to bool, and hence no overriding from -1 to 0/1
would be needed anymore at all.

Jan