[SRU focal/linux-gcp 0/1] LP: #1958416 (IO failures on SEV GCP instances)

[Thadeu Lima de Souza Cascardo]

On Thu, Nov 10, 2022 at 10:00:31AM +0100, Stefan Bader wrote:
> On 09.11.22 18:06, Thadeu Lima de Souza Cascardo wrote:
> > [Impact]
> > Small SWIOTLB might cause DMA failures during High IO on GCP instances using SEV.
> > 
> > [Test case]
> > Boot n2d-standard-2 or n2d-standard-64 instances on GCP, issue a lot of IO,
> > and look for DMA overflow or full swiotlb messages on dmesg.
> > 
> > [Potential regression]
> > Boot failures should not be discarded when SWIOTLB cannot be allocated.
> > 
> > Ashish Kalra (1):
> >    x86,swiotlb: Adjust SWIOTLB bounce buffer size for SEV guests
> > 
> >   arch/x86/include/asm/mem_encrypt.h |  2 ++
> >   arch/x86/kernel/setup.c            |  6 ++++++
> >   arch/x86/mm/mem_encrypt.c          | 31 ++++++++++++++++++++++++++++++
> >   include/linux/swiotlb.h            |  8 ++++++++
> >   kernel/dma/swiotlb.c               | 20 +++++++++++++++++--
> >   5 files changed, 65 insertions(+), 2 deletions(-)
> > 
> 
> Hm, is this really only needed for the gcp kernel? Are we not also run tests
> for the focal:linux kernel on the GCP cloud?
> 
> -Stefan
> 

I just added in the LP bug the target to linux as Medium, while linux-gcp is
High. As linux-gcp has other changes to those files that make it a different
backport, I opted to prioritizing the fix in linux-gcp.

Cascardo.