ACK/cmnt: [Patch 0/2 v2] [impish/linux-azure] Azure: not enough RAM under 4GB for CVM

Philip Cox philip.cox at canonical.com
Mon Apr 4 17:05:26 UTC 2022


With bzolnier's comments addressed...


Acked-By: Philip Cox <philip.cox at canonical.com>


On 2022-04-04 12:49 p.m., Tim Gardner wrote:
>
>
> On 4/4/22 10:45, Bartlomiej Zolnierkiewicz wrote:
>> Acked-by: Bartlomiej Zolnierkiewicz 
>> <bartlomiej.zolnierkiewicz at canonical.com>
>>
>> This is under the assumption that the Azure host doesn't provide (on
>> any instance) devices with the addressing limitations which would
>> require a 4G limited swiotlb buffer.
>>
>> Also it seems that the patch #2 should have "UBUNTU: SAUCE: azure:" in
>> the patch subject (just like it is present in the patch #1).
>>
>
> Yup - I'll fix that at application time.
>
> rtg
>
>> On Mon, Apr 4, 2022 at 3:17 PM Tim Gardner 
>> <tim.gardner at canonical.com> wrote:
>>>
>>> v2 - attach the 2nd patch which was forgotten the first time.
>>>
>>> SRU Justification
>>>
>>> [Impact]
>>> Can you please include one more patchset for the "no enough RAM 
>>> under 4GB" issue
>>> for the v5.13 kernel (and v5.15 kernel):
>>> https://lwn.net/ml/linux-kernel/20220209122302.213882-2-ltykernel@gmail.com/ 
>>>
>>> https://lwn.net/ml/linux-kernel/20220209122302.213882-3-ltykernel@gmail.com/ 
>>>
>>>
>>> Without the two patches, currently the 5.13 kernel (and 5.15 kernel) 
>>> can't boot
>>> as a CVM guest on Azure. :-(
>>>
>>> The background is: currently an Azure host only provides about 900 
>>> MB memory below
>>> the 4GB boundary while the swiotlb code (which is required by the 
>>> CVM patchset)
>>> needs to allocate memory of a size of MIN(1GB, 6% of the total 
>>> memory). There will
>>> be an Azure host update, with which the host will provide about 3GB 
>>> memory below
>>> the 4GB boundary, but we still need to wait a few more weeks to know 
>>> the ETA.
>>> Currently the above two patches can work around the issue from the 
>>> guest side.
>>>
>>> [Test Case]
>>>
>>> Microsoft tested
>>>
>>> [Where things could go wrong]
>>>
>>> These patches could affect non-CVM instances.
>>>
>>> [Other Info]
>>>
>>> SF: #00323683
>



More information about the kernel-team mailing list