Re: [PATCH] iommu/hyper-v: Allow hyperv irq remapping without x2apic

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 11/11/2022 9:58 AM, Michael Kelley (LINUX) wrote:
> From: Wei Liu <wei.liu@xxxxxxxxxx> Sent: Friday, November 11, 2022 9:27 AM
>>
>> On Fri, Nov 11, 2022 at 04:55:22PM +0000, Michael Kelley (LINUX) wrote:
>>> From: Wei Liu <wei.liu@xxxxxxxxxx> Sent: Friday, November 11, 2022 8:33 AM
>>>>
>>>> Hi Tianyu
>>>>
>>>> On Wed, Nov 09, 2022 at 11:07:33AM -0800, Nuno Das Neves wrote:
>>>>> If x2apic is not available, hyperv-iommu skips remapping
>>>>> irqs. This breaks root partition which always needs irqs
>>>>> remapped.
>>>>>
>>>>> Fix this by allowing irq remapping regardless of x2apic,
>>>>> and change hyperv_enable_irq_remapping() to return
>>>>> IRQ_REMAP_XAPIC_MODE in case x2apic is missing.
>>>>>
>>>>
>>>> Do you remember why it was x2apic only?
>>>>
>>>> We tested this patch on different VM SKUs and it worked fine. I'm just
>>>> wondering if there would be some subtle breakages that we couldn't
>>>> easily test.
>>>>
>>>> Thanks,
>>>> Wei.
>>>
>>> My recollection is that originally Hyper-V provided the x2apic in the
>>> guest only when the number of vCPUs exceeded 255, and that was
>>> the only case where IRQ remapping was needed.  The intent was to
>>> not disturb the case where # of vCPUs was < 255 and the xapic is used.
>>> I don't remember there being any potential for subtle breakages.
>>
>> Thanks for the information.
>>
>>>
>>> I think more recent versions of Hyper-V now provide the x2apic
>>> in the guest in some cases when # of vCPUs is < 255.
>>>
>>
>> On Azure the default for AMD SKUs is still xapic unless the number of
>> VCPUs exceeds 2XX (can't remember the exact number -- maybe it is 255).
> 
> I don't think Azure has any VM sizes based on AMD processors with more
> than 255 vCPUs.  All the >255 vCPUs VM sizes use Intel processors.
> 
> FWIW, I have a D2ds_v5 VM (2 CPUs & Intel processor) that shows an
> x2apic instead of an xapic.  My memory is vague, but I think that's the
> requirements to get an x2apic in a smaller VM:  must be a "v5" series and
> must be Intel processor.
> 
> Michael
>

Yes this seems to be the case, although I didn't realise it until now!
I sort of assumed since x2apic has been around so long, all the intel VMs
just had it enabled...

>>
>> Nuno, can you list the tests you've done? They will need to cover Linux
>> running as a normal guest on Azure and Hyper-V.
>>>> Thanks,
>> Wei.
>>

I've tested this patch on these Azure SKUs:
- Standard_D2S_v2 (intel xapic)
- Standard_D4ds_v4 (intel xapic)
- Standard_D4ds_v5 (intel x2apic)
- Standard_D4ads_v5 (amd xapic)

I've tested with linux Dom0 (nested hyperv root partition) and as a
regular L1 guest.

>>
>>> Michael
>>>
>>>>
>>>>> Tested with root and non-root hyperv partitions.
>>>>>
>>>>> Signed-off-by: Nuno Das Neves <nunodasneves@xxxxxxxxxxxxxxxxxxx>
>>>>> ---
>>>>>  drivers/iommu/Kconfig        | 6 +++---
>>>>>  drivers/iommu/hyperv-iommu.c | 7 ++++---
>>>>>  2 files changed, 7 insertions(+), 6 deletions(-)
>>>>>
>>>>> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
>>>>> index dc5f7a156ff5..cf7433652db0 100644
>>>>> --- a/drivers/iommu/Kconfig
>>>>> +++ b/drivers/iommu/Kconfig
>>>>> @@ -474,13 +474,13 @@ config QCOM_IOMMU
>>>>>  	  Support for IOMMU on certain Qualcomm SoCs.
>>>>>
>>>>>  config HYPERV_IOMMU
>>>>> -	bool "Hyper-V x2APIC IRQ Handling"
>>>>> +	bool "Hyper-V IRQ Handling"
>>>>>  	depends on HYPERV && X86
>>>>>  	select IOMMU_API
>>>>>  	default HYPERV
>>>>>  	help
>>>>> -	  Stub IOMMU driver to handle IRQs as to allow Hyper-V Linux
>>>>> -	  guests to run with x2APIC mode enabled.
>>>>> +	  Stub IOMMU driver to handle IRQs to support Hyper-V Linux
>>>>> +	  guest and root partitions.
>>>>>
>>>>>  config VIRTIO_IOMMU
>>>>>  	tristate "Virtio IOMMU driver"
>>>>> diff --git a/drivers/iommu/hyperv-iommu.c b/drivers/iommu/hyperv-iommu.c
>>>>> index e190bb8c225c..abd1826a9e63 100644
>>>>> --- a/drivers/iommu/hyperv-iommu.c
>>>>> +++ b/drivers/iommu/hyperv-iommu.c
>>>>> @@ -123,8 +123,7 @@ static int __init hyperv_prepare_irq_remapping(void)
>>>>>  	const struct irq_domain_ops *ops;
>>>>>
>>>>>  	if (!hypervisor_is_type(X86_HYPER_MS_HYPERV) ||
>>>>> -	    x86_init.hyper.msi_ext_dest_id() ||
>>>>> -	    !x2apic_supported())
>>>>> +	    x86_init.hyper.msi_ext_dest_id())
>>>>>  		return -ENODEV;
>>>>>
>>>>>  	if (hv_root_partition) {
>>>>> @@ -170,7 +169,9 @@ static int __init hyperv_prepare_irq_remapping(void)
>>>>>
>>>>>  static int __init hyperv_enable_irq_remapping(void)
>>>>>  {
>>>>> -	return IRQ_REMAP_X2APIC_MODE;
>>>>> +	if (x2apic_supported())
>>>>> +		return IRQ_REMAP_X2APIC_MODE;
>>>>> +	return IRQ_REMAP_XAPIC_MODE;
>>>>>  }
>>>>>
>>>>>  struct irq_remap_ops hyperv_irq_remap_ops = {
>>>>> --
>>>>> 2.25.1
>>>>>



[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux