linux-stable.git/drivers/iommu, branch linux-4.6.y

iommu/amd: Fix unity mapping initialization race

2016-07-27T15:42:15+00:00

commit 522e5cb76d0663c88f96b6a8301451c8efa37207 upstream.

There is a race condition in the AMD IOMMU init code that
causes requested unity mappings to be blocked by the IOMMU
for a short period of time. This results on boot failures
and IO_PAGE_FAULTs on some machines.

Fix this by making sure the unity mappings are installed
before all other DMA is blocked.

Fixes: aafd8ba0ca74 ('iommu/amd: Implement add_device and remove_device')
Signed-off-by: Joerg Roedel 
Signed-off-by: Greg Kroah-Hartman

iommu/vt-d: Enable QI on all IOMMUs before setting root entry

2016-07-27T15:42:15+00:00

commit a4c34ff1c029e90e7d5f8dd8d29b0a93b31c3cb2 upstream.

This seems to be required on some X58 chipsets on systems
with more than one IOMMU. QI does not work until it is
enabled on all IOMMUs in the system.

Reported-by: Dheeraj CVR 
Tested-by: Dheeraj CVR 
Fixes: 5f0a7f7614a9 ('iommu/vt-d: Make root entry visible for hardware right after allocation')
Signed-off-by: Joerg Roedel 
Signed-off-by: Greg Kroah-Hartman

iommu/arm-smmu: Wire up map_sg for arm-smmu-v3

2016-07-27T15:42:15+00:00

commit 9aeb26cfc2abc96be42b9df2d0f2dc5d805084ff upstream.

The map_sg callback is missing from arm_smmu_ops, but is required by
iommu.h. Similarly to most other IOMMU drivers, connect it to
default_iommu_map_sg.

Signed-off-by: Jean-Philippe Brucker 
Signed-off-by: Will Deacon 
Signed-off-by: Joerg Roedel 
Signed-off-by: Greg Kroah-Hartman

iommu/rockchip: Fix zap cache during device attach

2016-07-27T15:42:15+00:00

commit ae8a7910fb0568531033bd6ebe65590f7a4fa6e2 upstream.

rk_iommu_command() takes a struct rk_iommu and iterates over the slave
MMUs, so this is doubly wrong in that we're passing in the wrong pointer
and talking to MMUs that we shouldn't be.

Fixes: cd6438c5f844 ("iommu/rockchip: Reconstruct to support multi slaves")
Signed-off-by: John Keeping 
Tested-by: Heiko Stuebner 
Reviewed-by: Heiko Stuebner 
Signed-off-by: Joerg Roedel 
Signed-off-by: Greg Kroah-Hartman

iommu/arm-smmu: Don't allocate resources for bypass domains

2016-04-21T14:47:32+00:00

Until we get fully plumbed into of_iommu_configure, our default
IOMMU_DOMAIN_DMA domains just bypass translation. Since we achieve that
by leaving the stream table entries set to bypass instead of pointing at
a translation context, the context bank we allocate for the domain is
completely wasted. Context banks are typically a rather limited
resource, so don't hog ones we don't need.

Reported-by: Eric Auger 
Tested-by: Eric Auger 
Signed-off-by: Robin Murphy 
Signed-off-by: Will Deacon 
Signed-off-by: Joerg Roedel

iommu/arm-smmu: Fix stream-match conflict with IOMMU_DOMAIN_DMA

2016-04-21T14:47:32+00:00

Commit cbf8277ef456 ("iommu/arm-smmu: Treat IOMMU_DOMAIN_DMA as bypass
for now") ignores requests to attach a device to the default domain
since, without IOMMU-basked DMA ops available everywhere, the default
domain will just lead to unexpected transaction faults being reported.

Unfortunately, the way this was implemented on SMMUv2 causes a
regression with VFIO PCI device passthrough under KVM on AMD Seattle.
On this system, the host controller device is associated with both a
pci_dev *and* a platform_device, and can therefore end up with duplicate
SMR entries, resulting in a stream-match conflict at runtime.

This patch amends the original fix so that attaching to IOMMU_DOMAIN_DMA
is rejected even before configuring the SMRs. This restores the old
behaviour for now, but we'll need to look at handing host controllers
specially when we come to supporting the default domain fully.

Reported-by: Eric Auger 
Tested-by: Eric Auger 
Tested-by: Yang Shi 
Signed-off-by: Will Deacon 
Signed-off-by: Joerg Roedel

iommu/amd: Fix checking of pci dma aliases

2016-04-11T14:07:51+00:00

Commit 61289cb ('iommu/amd: Remove old alias handling code')
removed the old alias handling code from the AMD IOMMU
driver because this is now handled by the IOMMU core code.

But this also removed the handling of PCI aliases, which is
not handled by the core code. This caused issues with PCI
devices that have hidden PCIe-to-PCI bridges that rewrite
the request-id.

Fix this bug by re-introducing some of the removed functions
from commit 61289cbaf6c8 and add a alias field
'struct iommu_dev_data'. This field carrys the return value
of the get_alias() function and uses that instead of the
amd_iommu_alias_table[] array in the code.

Fixes: 61289cbaf6c8 ('iommu/amd: Remove old alias handling code')
Cc: stable@vger.kernel.org # v4.4+
Tested-by: Tomasz Golinski 
Signed-off-by: Joerg Roedel

iommu/vt-d: Silence an uninitialized variable warning

2016-04-07T12:51:47+00:00

My static checker complains that "dma_alias" is uninitialized unless we
are dealing with a pci device.  This is true but harmless.  Anyway, we
can flip the condition around to silence the warning.

Signed-off-by: Dan Carpenter 
Signed-off-by: Joerg Roedel

iommu/rockchip: Fix "is stall active" check

2016-04-07T12:50:18+00:00

Since commit cd6438c5f844 ("iommu/rockchip: Reconstruct to support multi
slaves") rk_iommu_is_stall_active() always returns false because the
bitwise AND operates on the boolean flag promoted to an integer and a
value that is either zero or BIT(2).

Explicitly convert the right-hand value to a boolean so that both sides
are guaranteed to be either zero or one.

rk_iommu_is_paging_enabled() does not suffer from the same problem since
RK_MMU_STATUS_PAGING_ENABLED is BIT(0), but let's apply the same change
for consistency and to make it clear that it's correct without needing
to lookup the value.

Fixes: cd6438c5f844 ("iommu/rockchip: Reconstruct to support multi slaves")
Signed-off-by: John Keeping 
Reviewed-by: Heiko Stuebner 
Tested-by: Tomeu Vizoso 
Signed-off-by: Joerg Roedel

iommu: Don't overwrite domain pointer when there is no default_domain

2016-04-07T12:33:03+00:00

IOMMU drivers that do not support default domains, but make
use of the the group->domain pointer can get that pointer
overwritten with NULL on device add/remove.

Make sure this can't happen by only overwriting the domain
pointer when it is NULL.

Cc: stable@vger.kernel.org # v4.4+
Fixes: 1228236de5f9 ('iommu: Move default domain allocation to iommu_group_get_for_dev()')
Signed-off-by: Joerg Roedel