Re: [RFC V1 3/5] x86: CVMs: Enable dynamic swiotlb by default for CVMs

[Vishal Annapurve]

On Thu, Feb 1, 2024 at 5:50 PM Jeremi Piotrowski
<jpiotrowski@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On 12/01/2024 06:52, Vishal Annapurve wrote:
> > CVMs used SWIOTLB for non-trusted IO using dma_map_* APIs. This series
> > will ensure that dma_alloc_* API to also allocate from SWIOTLB pools.
> > Initial size of SWIOTLB pool is decided using heuristics and has been
> > working with CVM usecases so far.
> >
> > It's better to allow SWIOTLB to scale up on demand rather than keeping
> > the size fixed to avoid failures with possibly increased usage of
> > SWIOTLB with dma_alloc_* APIs allocating from SWIOTLB pools. This should
> > also help in future with more devices getting used from CVMs for
> > non-trusted IO.
> >
> > Signed-off-by: Vishal Annapurve <vannapurve@xxxxxxxxxx>
> > ---
> >  arch/x86/Kconfig | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> > index 1566748f16c4..035c8a022c4c 100644
> > --- a/arch/x86/Kconfig
> > +++ b/arch/x86/Kconfig
> > @@ -884,6 +884,7 @@ config INTEL_TDX_GUEST
> >       select X86_MEM_ENCRYPT
> >       select X86_MCE
> >       select UNACCEPTED_MEMORY
> > +     select SWIOTLB_DYNAMIC
> >       help
> >         Support running as a guest under Intel TDX.  Without this support,
> >         the guest kernel can not boot or run under TDX.
> > @@ -1534,6 +1535,7 @@ config AMD_MEM_ENCRYPT
> >       select ARCH_HAS_CC_PLATFORM
> >       select X86_MEM_ENCRYPT
> >       select UNACCEPTED_MEMORY
> > +     select SWIOTLB_DYNAMIC
> >       help
> >         Say yes to enable support for the encryption of system memory.
> >         This requires an AMD processor that supports Secure Memory
>
> What this does is unconditionally enable SWIOTLB_DYNAMIC for every kernel compiled
> to support memory encryption, regardless of whether it runs inside a confidential guest.
> I don't think that is what you intended.
>

Right, I will have to add a way to toggle this support dynamically in
addition to having the support built in, based on more discussion
around 2M allocations and dynamic scaling.

> Best wishes,
> Jeremi
>
>