After commit 327d5b2fee91c ("iommu/vt-d: Allow 32bit devices to uses DMA
domain"), swiotbl could also be used for direct memory access if IOMMU
is enabled but a device is configured to pass through the DMA translation.
Keep swiotlb when IOMMU is forced on, otherwise, some devices won't work
if "iommu=pt" kernel parameter is used.
Fixes: 327d5b2fee91c ("iommu/vt-d: Allow 32bit devices to uses DMA domain")
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=210237
Reported-and-tested-by: Adrian Huang <[email protected]>
Signed-off-by: Lu Baolu <[email protected]>
---
arch/x86/kernel/tboot.c | 5 +----
1 file changed, 1 insertion(+), 4 deletions(-)
diff --git a/arch/x86/kernel/tboot.c b/arch/x86/kernel/tboot.c
index 420be871d9d4..ae64f98ec2ab 100644
--- a/arch/x86/kernel/tboot.c
+++ b/arch/x86/kernel/tboot.c
@@ -514,13 +514,10 @@ int tboot_force_iommu(void)
if (!tboot_enabled())
return 0;
- if (no_iommu || swiotlb || dmar_disabled)
+ if (no_iommu || dmar_disabled)
pr_warn("Forcing Intel-IOMMU to enabled\n");
dmar_disabled = 0;
-#ifdef CONFIG_SWIOTLB
- swiotlb = 0;
-#endif
no_iommu = 0;
return 1;
--
2.25.1
On Wed, 25 Nov 2020 09:41:24 +0800, Lu Baolu wrote:
> After commit 327d5b2fee91c ("iommu/vt-d: Allow 32bit devices to uses DMA
> domain"), swiotbl could also be used for direct memory access if IOMMU
> is enabled but a device is configured to pass through the DMA translation.
> Keep swiotlb when IOMMU is forced on, otherwise, some devices won't work
> if "iommu=pt" kernel parameter is used.
Applied to arm64 (for-next/iommu/fixes), thanks!
[1/1] x86/tboot: Don't disable swiotlb when iommu is forced on
https://git.kernel.org/arm64/c/e2be2a833ab5
Cheers,
--
Will
https://fixes.arm64.dev
https://next.arm64.dev
https://will.arm64.dev
On Wed, Nov 25, 2020 at 02:05:15PM +0000, Will Deacon wrote:
> On Wed, 25 Nov 2020 09:41:24 +0800, Lu Baolu wrote:
> > After commit 327d5b2fee91c ("iommu/vt-d: Allow 32bit devices to uses DMA
> > domain"), swiotbl could also be used for direct memory access if IOMMU
> > is enabled but a device is configured to pass through the DMA translation.
> > Keep swiotlb when IOMMU is forced on, otherwise, some devices won't work
> > if "iommu=pt" kernel parameter is used.
>
> Applied to arm64 (for-next/iommu/fixes), thanks!
>
> [1/1] x86/tboot: Don't disable swiotlb when iommu is forced on
> https://git.kernel.org/arm64/c/e2be2a833ab5
But tboot never ran on ARM. It is a Intel specifc.
I think either me or Thomas should take this patch.
On Wed, Nov 25, 2020 at 03:51:30PM +0000, Will Deacon wrote:
> Hi Konrad,
>
> On Wed, Nov 25, 2020 at 10:41:53AM -0500, Konrad Rzeszutek Wilk wrote:
> > On Wed, Nov 25, 2020 at 02:05:15PM +0000, Will Deacon wrote:
> > > On Wed, 25 Nov 2020 09:41:24 +0800, Lu Baolu wrote:
> > > > After commit 327d5b2fee91c ("iommu/vt-d: Allow 32bit devices to uses DMA
> > > > domain"), swiotbl could also be used for direct memory access if IOMMU
> > > > is enabled but a device is configured to pass through the DMA translation.
> > > > Keep swiotlb when IOMMU is forced on, otherwise, some devices won't work
> > > > if "iommu=pt" kernel parameter is used.
> > >
> > > Applied to arm64 (for-next/iommu/fixes), thanks!
> > >
> > > [1/1] x86/tboot: Don't disable swiotlb when iommu is forced on
> > > https://git.kernel.org/arm64/c/e2be2a833ab5
> >
> > But tboot never ran on ARM. It is a Intel specifc.
> >
> > I think either me or Thomas should take this patch.
>
> FWIW, I did check with Thomas before I picked it up. I know it looks weird
> going via arm64, but that's only because I'm temporarily handling the IOMMU
> tree there (including vt-d changes) while Joerg is away. Since this fixes a
> vt-d regression, I thought I'd pick it up along with the other IOMMU fixes I
> have queued for -rc6.
>
Aah, I missed the memo :-)
> That said, if you insist, then I can revert it. I'm really only trying to
> help here.
Nah. Enjoy picking up patches!
Thanks!
>
> Will
Hi Konrad,
On Wed, Nov 25, 2020 at 10:41:53AM -0500, Konrad Rzeszutek Wilk wrote:
> On Wed, Nov 25, 2020 at 02:05:15PM +0000, Will Deacon wrote:
> > On Wed, 25 Nov 2020 09:41:24 +0800, Lu Baolu wrote:
> > > After commit 327d5b2fee91c ("iommu/vt-d: Allow 32bit devices to uses DMA
> > > domain"), swiotbl could also be used for direct memory access if IOMMU
> > > is enabled but a device is configured to pass through the DMA translation.
> > > Keep swiotlb when IOMMU is forced on, otherwise, some devices won't work
> > > if "iommu=pt" kernel parameter is used.
> >
> > Applied to arm64 (for-next/iommu/fixes), thanks!
> >
> > [1/1] x86/tboot: Don't disable swiotlb when iommu is forced on
> > https://git.kernel.org/arm64/c/e2be2a833ab5
>
> But tboot never ran on ARM. It is a Intel specifc.
>
> I think either me or Thomas should take this patch.
FWIW, I did check with Thomas before I picked it up. I know it looks weird
going via arm64, but that's only because I'm temporarily handling the IOMMU
tree there (including vt-d changes) while Joerg is away. Since this fixes a
vt-d regression, I thought I'd pick it up along with the other IOMMU fixes I
have queued for -rc6.
That said, if you insist, then I can revert it. I'm really only trying to
help here.
Will