The -ENODEV error just means that the device is not
translated by an IOMMU. We shouldn't bail out of iommu
driver initialization when that happens, as this is a common
scenario on ARM.
Not returning -ENODEV in the drivers would be a bad idea, as
the IOMMU core would have no indication whether a device is
translated or not. This indication is not used at the
moment, but will probably be in the future.
Fixes: 19762d7 ("iommu: Propagate error in add_iommu_group")
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Tested-by: Eric Auger <eric.auger@linaro.org>
Tested-by: Heiko Stuebner <heiko@sntech.de>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The iommu_group_alloc() and iommu_group_get_for_dev()
functions return error pointers, they never return NULL.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This function can be called by an IOMMU driver to request
that a device's default domain is direct mapped.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Use the information exported by the IOMMU drivers to create
direct mapped regions in the default domains.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Add two new functions to the IOMMU-API to allow the IOMMU
drivers to export the requirements for direct mapped regions
per device.
This is useful for exporting the information in Intel VT-d's
RMRR entries or AMD-Vi's unity mappings.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Make use of the default domain and re-attach a device to it
when it is detached from another domain. Also enforce that a
device has to be in the default domain before it can be
attached to a different domain.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This patch changes the behavior of the iommu_attach_device
and iommu_detach_device functions. With this change these
functions only work on devices that have their own group.
For all other devices the iommu_group_attach/detach
functions must be used.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The default domain will be used (if supported by the iommu
driver) when the devices in the iommu group are not attached
to any other domain.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Make sure we call the ->remove_device call-back on all
devices already initialized with ->add_device when the bus
initialization fails.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Write a message to the kernel log when a device is added or
removed from a group and add debug messages to group
allocation and release routines.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Including the function name is only useful for debugging
messages. They don't belong into other messages from the
iommu core.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
iommu_group_alloc might be called very early in case of iommu controllers
activated from of_iommu, so ensure that this part of subsystem is ready
when devices are being populated from device-tree (core_initcall seems to
be okay for this case).
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Tested-by: Javier Martinez Canillas <javier.martinez@collabora.co.uk>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
All drivers have been converted to the new domain_alloc and
domain_free iommu-ops. So remove the old ones and get rid of
iommu_domain->priv too, as this is no longer needed when the
struct iommu_domain is embedded in the private structures of
the iommu drivers.
Tested-by: Thierry Reding <treding@nvidia.com>
Tested-by: Heiko Stuebner <heiko@sntech.de>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Check for the new __IOMMU_DOMAIN_PAGING flag before calling
into the iommu drivers ->map and ->unmap call-backs.
Tested-by: Thierry Reding <treding@nvidia.com>
Tested-by: Heiko Stuebner <heiko@sntech.de>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This allows to handle domains differently based on their
type in the future. An IOMMU driver can implement certain
optimizations for DMA-API domains for example.
The domain types can be extended later and some of the
existing domain attributes can be migrated to become domain
flags.
Tested-by: Thierry Reding <treding@nvidia.com>
Tested-by: Heiko Stuebner <heiko@sntech.de>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
These new call-backs defer the allocation and destruction of
'struct iommu_domain' to the iommu driver. This allows
drivers to embed this struct into their private domain
structures and to get rid of the domain_init and
domain_destroy call-backs when all drivers have been
converted.
Tested-by: Thierry Reding <treding@nvidia.com>
Tested-by: Heiko Stuebner <heiko@sntech.de>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
iommu_map() calls trace_map() with iova and size. trace_map()
should report original iova and original size as opposed to
iova and size after they get changed during mapping. size is
always zero at the end of mapping which is useless to report
and iova as it gets incremented, it is not as useful as the
original iova. Change iommu_map() to call trace_map() to
report original iova and original size.
Signed-off-by: Shuah Khan <shuahkh@osg.samsung.com>
Reported-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Currently map and unmap are implemented as events under a
common trace class declaration. The common class forces
trace_unmap() to require a bogus physical address argument
that it doesn't use. Changing unmap to report unmapped size
will provide useful information for debugging. Remove common
map_unmap trace class and change map and unmap into separate
events as opposed to events under the same class to allow for
differences in the reporting information. In addition, map and
unmap are changed to handle size value as size_t instead of int
to match the passed size value and avoid overflow.
Signed-off-by: Shuah Khan <shuahkh@osg.samsung.com>
Suggested-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
iommu_unmap() calls trace_unmap() with changed iova and original
size. trace_unmap() should report original iova instead. Change
iommu_unmap() to call trace_unmap() with original iova.
Signed-off-by: Shuah Khan <shuahkh@osg.samsung.com>
Reported-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The iomm-config branch contains work from Will Deacon, quoting his description:
This series adds automatic IOMMU and DMA-mapping configuration for
OF-based DMA masters described using the generic IOMMU devicetree
bindings. Although there is plenty of future work around splitting up
iommu_ops, adding default IOMMU domains and sorting out automatic IOMMU
group creation for the platform_bus, this is already useful enough for
people to port over their IOMMU drivers and start using the new probing
infrastructure (indeed, Marek has patches queued for the Exynos IOMMU).
The branch touches core ARM and IOMMU driver files, and the respective
maintainers (Russell King and Joerg Roedel) agreed to have the contents
merged through the arm-soc tree. The final version was ready just before
the merge window, so we ended up delaying it a bit longer than the rest,
but we don't expect to see regressions because this is just additional
infrastructure that will get used in drivers starting in 3.20 but is
unused so far.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)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=ELLq
-----END PGP SIGNATURE-----
Merge tag 'iommu-config-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Pull ARM SoC/iommu configuration update from Arnd Bergmann:
"The iomm-config branch contains work from Will Deacon, quoting his
description:
This series adds automatic IOMMU and DMA-mapping configuration for
OF-based DMA masters described using the generic IOMMU devicetree
bindings. Although there is plenty of future work around splitting up
iommu_ops, adding default IOMMU domains and sorting out automatic IOMMU
group creation for the platform_bus, this is already useful enough for
people to port over their IOMMU drivers and start using the new probing
infrastructure (indeed, Marek has patches queued for the Exynos IOMMU).
The branch touches core ARM and IOMMU driver files, and the respective
maintainers (Russell King and Joerg Roedel) agreed to have the
contents merged through the arm-soc tree.
The final version was ready just before the merge window, so we ended
up delaying it a bit longer than the rest, but we don't expect to see
regressions because this is just additional infrastructure that will
get used in drivers starting in 3.20 but is unused so far"
* tag 'iommu-config-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
iommu: store DT-probed IOMMU data privately
arm: dma-mapping: plumb our iommu mapping ops into arch_setup_dma_ops
arm: call iommu_init before of_platform_populate
dma-mapping: detect and configure IOMMU in of_dma_configure
iommu: fix initialization without 'add_device' callback
iommu: provide helper function to configure an IOMMU for an of master
iommu: add new iommu_ops callback for adding an OF device
dma-mapping: replace set_arch_dma_coherent_ops with arch_setup_dma_ops
iommu: provide early initialisation hook for IOMMU drivers
If the IOMMU supports pages smaller than the CPU page size, segments
which lie at offsets within the CPU page may be mapped based on the
finer-grained IOMMU page boundaries. This minimises the amount of
non-buffer memory between the CPU page boundary and the start of the
segment which must be mapped and therefore exposed to the device, and
brings the default iommu_map_sg implementation in line with
iommu_map/unmap with respect to alignment.
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
IOMMU drivers can be initialized from of_iommu helpers. Such drivers don't
need to provide device_add callbacks to operate properly, so there is no
need to fail initialization if the callback is missing.
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
When some part of bus_set_iommu fails it should undo any made changes
and not simply leave everything as is.
This includes unregistering the bus notifier in iommu_bus_init when
add_iommu_group fails and also setting the bus->iommu_ops back to NULL.
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The IOMMU-API works on page boundarys, unlike the DMA-API
which can work with sub-page buffers. The sg->offset
field does not make sense on the IOMMU level, so force it to
be 0. Do some error-path consolidation while at it.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Mapping and unmapping are more often than not in the critical path.
map_sg allows IOMMU driver implementations to optimize the process
of mapping buffers into the IOMMU page tables.
Instead of mapping a buffer one page at a time and requiring potentially
expensive TLB operations for each page, this function allows the driver
to map all pages in one go and defer TLB maintenance until after all
pages have been mapped.
Additionally, the mapping operation would be faster in general since
clients does not have to keep calling map API over and over again for
each physically contiguous chunk of memory that needs to be mapped to a
virtually contiguous region.
Signed-off-by: Olav Haugan <ohaugan@codeaurora.org>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
iommu_bus_init() registers a bus notifier on the given bus by using
a statically defined notifier block:
static struct notifier_block iommu_bus_nb = {
.notifier_call = iommu_bus_notifier,
};
This same notifier block is used for all busses. This causes a
problem for notifiers registered after iommu has registered this
callback on multiple busses. The problem is that a subsequent
notifier being registered on a bus which has this iommu notifier
will also get linked in to the notifier list of all other busses
which have this iommu notifier.
This patch fixes this by allocating the notifier_block at runtime.
Some error checking is also added to catch any allocation failure
or notifier registration error.
Signed-off-by: Mark Salter <msalter@redhat.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
It turns out that our assumption that aliases are always to the same
slot isn't true. One particular platform reports an IVRS alias of the
SATA controller (00:11.0) for the legacy IDE controller (00:14.1).
When we hit this, we attempt to use a single IOMMU group for
everything on the same bus, which in this case is the root complex.
We already have multiple groups defined for the root complex by this
point, resulting in multiple WARN_ON hits.
This patch makes these sorts of aliases work again with IOMMU groups
by reworking how we search through the PCI address space to find
existing groups. This should also now handle looped dependencies and
all sorts of crazy inter-dependencies that we'll likely never see.
The recursion used here should never be very deep. It's unlikely to
have individual aliases and only theoretical that we'd ever see a
chain where one alias causes us to search through to yet another
alias. We're also only dealing with PCIe device on a single bus,
which means we'll typically only see multiple slots in use on the root
complex. Loops are also a theoretically possibility, which I've
tested using fake DMA alias quirks and prevent from causing problems
using a bitmap of the devfn space that's been visited.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Cc: stable@vger.kernel.org # 3.17
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This function will replace the current iommu_domain_has_cap
function and clean up the interface while at it.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
When a non-PCI device is passed to that function it might
pass group == NULL to iommu_group_add_device() which then
dereferences it and cause a crash this way. Fix it by
just returning an error for non-PCI devices.
Fixes: 104a1c13ac
Cc: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This structure is read-only data and should never be modified.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Currently each IOMMU driver that supports IOMMU groups has its own
code for discovering the base device used in grouping. This code
is generally not specific to the IOMMU hardware, but to the bus of
the devices managed by the IOMMU. We can therefore create a common
interface for supporting devices on different buses.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Commit 6197ca82 (iommu: Use %pa and %zx instead of casting) introduced the
usage of '%pa', but still kept the '0x', which leads to printing '0x0x'.
Remove the '0x' when '%pa' is used.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
Change iommu driver to call unmap trace event. This iommu_map_unmap class
event can be enabled to trigger when iommu unmap iommu ops is called. Trace
information includes iova, physical address (map event only), and size.
Testing:
Added trace calls to iommu_prepare_identity_map() for testing some of the
conditions that are hard to trigger. Here is the trace from the testing:
swapper/0-1 [003] .... 1.854102: unmap: IOMMU: iova=0x00000000cb800000 size=0x400
Signed-off-by: Shuah Khan <shuah.kh@samsung.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
Change iommu driver to call map trace event. This iommu_map_unmap class event
can be enabled to trigger when iommu map iommu ops is called. Trace information
includes iova, physical address (map event only), and size.
Testing:
Added trace calls to iommu_prepare_identity_map() for testing some of the
conditions that are hard to trigger. Here is the trace from the testing:
swapper/0-1 [003] .... 1.854102: map: IOMMU: iova=0x00000000cb800000 paddr=0x00000000cf9fffff size=0x400
Signed-off-by: Shuah Khan <shuah.kh@samsung.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
Change iommu driver to call detach_device_to_domain trace event. This
iommu_device class event can be enabled to trigger when devices are detached
from a domain. Trace information includes device name.
Testing:
Added trace calls to iommu_prepare_identity_map() for testing some of the
conditions that are hard to trigger. Here is the trace from the testing:
swapper/0-1 [003] .... 1.854102: detach_device_from_domain: IOMMU: device=0000:00:02.0
Signed-off-by: Shuah Khan <shuah.kh@samsung.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
Change iommu driver to call attach_device_to_domain trace event. This
iommu_device class event can be enabled to trigger when devices are attached
to a domain. Trace information includes device name.
Testing:
Added trace calls to iommu_prepare_identity_map() for testing some of the
conditions that are hard to trigger. Here is the trace from the testing:
swapper/0-1 [003] .... 1.854102: attach_device_to_domain: IOMMU: device=0000:00:02.0
Signed-off-by: Shuah Khan <shuah.kh@samsung.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
Change iommu driver to call remove_device_to_group trace event. This
iommu_group class event can be enabled to trigger when devices get
removed from an iommu group. Trace information includes iommu group id and
device name.
Testing:
Added trace calls to iommu_prepare_identity_map() for testing some of the
conditions that are hard to trigger. Here is the trace from the testing:
swapper/0-1 [003] .... 1.854101: remove_device_from_group: IOMMU: groupID=0 device=0000:00:02.0
Signed-off-by: Shuah Khan <shuah.kh@samsung.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
Change iommu driver to call add_device_to_group trace event. This iommu_group
class event can be enabled to trigger when devices get added to an iommu group.
Trace information includes iommu group id and device name.
Testing:
The following is trace is generated when intel-iommu driver adds devices to
to iommu groups during boot-time during its initialization:
swapper/0-1 [003] .... 1.854793: add_device_to_group: IOMMU: groupID=0 device=0000:00:00.0
swapper/0-1 [003] .... 1.854797: add_device_to_group: IOMMU: groupID=1 device=0000:00:02.0
Signed-off-by: Shuah Khan <shuah.kh@samsung.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
Add tracing feature to iommu to report various iommu events. Classes
iommu_group, iommu_device, and iommu_map_unmap are defined.
iommu_group class events can be enabled to trigger when devices get added
to and removed from an iommu group. Trace information includes iommu group
id and device name.
iommu:add_device_to_group
iommu:remove_device_from_group
iommu_device class events can be enabled to trigger when devices are attached
to and detached from a domain. Trace information includes device name.
iommu:attach_device_to_domain
iommu:detach_device_from_domain
iommu_map_unmap class events can be enabled to trigger when iommu map and
unmap iommu ops. Trace information includes iova, physical address (map event
only), and size.
iommu:map
iommu:unmap
Signed-off-by: Shuah Khan <shuah.kh@samsung.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
printk supports using %pa for phys_addr_t and
%zx for size_t so use those instead of %lx and
casts to unsigned long.
Other miscellaneous changes around this:
Always use 0x%zx for size instead of one use of decimal.
Coalesce format and align arguments.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
iommu_map splits requests into pages that the iommu driver reports
that it can handle. The iommu_unmap path does not do the same. This
can cause problems not only from callers that might expect the same
behavior as the map path, but even from the failure path of iommu_map,
should it fail at a point where it has mapped and needs to unwind a
set of pages that the iommu driver cannot handle directly. amd_iommu,
for example, will BUG_ON if asked to unmap a non power of 2 size.
Fix this by extracting and generalizing the sizing code from the
iommu_map path and use it for both map and unmap.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
As IOMMU groups are exposed to the user space by their numbers,
the user space can use them in various kernel APIs so the kernel
might need an API to find a group by its ID.
As an example, QEMU VFIO on PPC64 platform needs it to associate
a logical bus number (LIOBN) with a specific IOMMU group in order
to support in-kernel handling of DMA map/unmap requests.
The patch adds the iommu_group_get_by_id(id) function which performs
such search.
v2: fixed reference counting.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
Each iommu window can have access permissions associated with it. Extended the
window_enable API to incorporate window access permissions.
In case of PAMU each window can have its specific set of permissions.
Signed-off-by: Varun Sethi <Varun.Sethi@freescale.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
This is required in case of PAMU, as it can support a window size of up
to 64G (even on 32bit).
Signed-off-by: Varun Sethi <Varun.Sethi@freescale.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
Add the iommu_domain_window_enable() and iommu_domain_window_disable()
functions to the IOMMU-API. These functions will be used to setup
domains that are based on subwindows and not on paging.
Signed-off-by: Joerg Roedel <joro@8bytes.org>
This attribute of a domain can be queried to find out if the
domain supports setting up page-tables using the iommu_map()
and iommu_unmap() functions.
Signed-off-by: Joerg Roedel <joro@8bytes.org>
In case the page-size bitmap is zero the code path in
iommu_map and iommu_unmap is undefined. Make it defined and
return -ENODEV in this case.
Signed-off-by: Joerg Roedel <joro@8bytes.org>
The iommu_init() initializes IOMMU internal structures and data
required for the IOMMU API as iommu_group_alloc().
It is registered as a subsys_initcall now.
One of the IOMMU users is going to be a PCI subsystem on POWER.
It discovers new IOMMU tables during the PCI scan so the logical
place to call iommu_group_alloc() is the moment when a new group
is discovered. However PCI scan is done from subsys_initcall hook
as IOMMU does so PCI hook can be (and is) called before the IOMMU one.
The patch moves IOMMU subsystem initialization one step earlier
to make sure that IOMMU is initialized before PCI scan begins.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
This patch introduces an extension to the iommu-api to get
and set attributes for an iommu_domain. Two functions are
introduced for this:
* iommu_domain_get_attr()
* iommu_domain_set_attr()
These functions will be used to make the iommu-api suitable
for GART-like IOMMUs and to implement hardware-specifc
api-extensions.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
IOMMU device groups are currently a rather vague associative notion
with assembly required by the user or user level driver provider to
do anything useful. This patch intends to grow the IOMMU group concept
into something a bit more consumable.
To do this, we first create an object representing the group, struct
iommu_group. This structure is allocated (iommu_group_alloc) and
filled (iommu_group_add_device) by the iommu driver. The iommu driver
is free to add devices to the group using it's own set of policies.
This allows inclusion of devices based on physical hardware or topology
limitations of the platform, as well as soft requirements, such as
multi-function trust levels or peer-to-peer protection of the
interconnects. Each device may only belong to a single iommu group,
which is linked from struct device.iommu_group. IOMMU groups are
maintained using kobject reference counting, allowing for automatic
removal of empty, unreferenced groups. It is the responsibility of
the iommu driver to remove devices from the group
(iommu_group_remove_device).
IOMMU groups also include a userspace representation in sysfs under
/sys/kernel/iommu_groups. When allocated, each group is given a
dynamically assign ID (int). The ID is managed by the core IOMMU group
code to support multiple heterogeneous iommu drivers, which could
potentially collide in group naming/numbering. This also keeps group
IDs to small, easily managed values. A directory is created under
/sys/kernel/iommu_groups for each group. A further subdirectory named
"devices" contains links to each device within the group. The iommu_group
file in the device's sysfs directory, which formerly contained a group
number when read, is now a link to the iommu group. Example:
$ ls -l /sys/kernel/iommu_groups/26/devices/
total 0
lrwxrwxrwx. 1 root root 0 Apr 17 12:57 0000:00:1e.0 ->
../../../../devices/pci0000:00/0000:00:1e.0
lrwxrwxrwx. 1 root root 0 Apr 17 12:57 0000:06:0d.0 ->
../../../../devices/pci0000:00/0000:00:1e.0/0000:06:0d.0
lrwxrwxrwx. 1 root root 0 Apr 17 12:57 0000:06:0d.1 ->
../../../../devices/pci0000:00/0000:00:1e.0/0000:06:0d.1
$ ls -l /sys/kernel/iommu_groups/26/devices/*/iommu_group
[truncating perms/owner/timestamp]
/sys/kernel/iommu_groups/26/devices/0000:00:1e.0/iommu_group ->
../../../kernel/iommu_groups/26
/sys/kernel/iommu_groups/26/devices/0000:06:0d.0/iommu_group ->
../../../../kernel/iommu_groups/26
/sys/kernel/iommu_groups/26/devices/0000:06:0d.1/iommu_group ->
../../../../kernel/iommu_groups/26
Groups also include several exported functions for use by user level
driver providers, for example VFIO. These include:
iommu_group_get(): Acquires a reference to a group from a device
iommu_group_put(): Releases reference
iommu_group_for_each_dev(): Iterates over group devices using callback
iommu_group_[un]register_notifier(): Allows notification of device add
and remove operations relevant to the group
iommu_group_id(): Return the group number
This patch also extends the IOMMU API to allow attaching groups to
domains. This is currently a simple wrapper for iterating through
devices within a group, but it's expected that the IOMMU API may
eventually make groups a more integral part of domains.
Groups intentionally do not try to manage group ownership. A user
level driver provider must independently acquire ownership for each
device within a group before making use of the group as a whole.
This may change in the future if group usage becomes more pervasive
across both DMA and IOMMU ops.
Groups intentionally do not provide a mechanism for driver locking
or otherwise manipulating driver matching/probing of devices within
the group. Such interfaces are generic to devices and beyond the
scope of IOMMU groups. If implemented, user level providers have
ready access via iommu_group_for_each_dev and group notifiers.
iommu_device_group() is removed here as it has no users. The
replacement is:
group = iommu_group_get(dev);
id = iommu_group_id(group);
iommu_group_put(group);
AMD-Vi & Intel VT-d support re-added in following patches.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Sometimes a single IOMMU user may have to deal with several
different IOMMU devices (e.g. remoteproc).
When an IOMMU fault happens, such users have to regain their
context in order to deal with the fault.
Users can't use the private fields of neither the iommu_domain nor
the IOMMU device, because those are already used by the IOMMU core
and low level driver (respectively).
This patch just simply allows users to pass a private token (most
notably their own context pointer) to iommu_set_fault_handler(),
and then makes sure it is provided back to the users whenever
an IOMMU fault happens.
The patch also adopts remoteproc to the new fault handling
interface, but the real functionality using this (recovery of
remote processors) will only be added later in a subsequent patch
set.
Cc: Fernando Guzman Lugo <fernando.lugo@ti.com>
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Since it is not guaranteed that an iommu driver initializes in its
domain_init() function, it must be initialized with NULL to prevent
calling a function in an arbitrary location when iommu fault occurred.
Signed-off-by: KyongHo Cho <pullip.cho@samsung.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
An IOMMU group is a set of devices for which the IOMMU cannot
distinguish transactions. For PCI devices, a group often occurs
when a PCI bridge is involved. Transactions from any device
behind the bridge appear to be sourced from the bridge itself.
We leave it to the IOMMU driver to define the grouping restraints
for their platform.
Using this new interface, the group for a device can be retrieved
using the iommu_device_group() callback. Users will compare the
value returned against the value returned for other devices to
determine whether they are part of the same group. Devices with
no group are not translated by the IOMMU. There should be no
expectations about the group numbers as they may be arbitrarily
assigned by the IOMMU driver and may not be persistent across boots.
We also provide a sysfs interface to the group numbers here so
that userspace can understand IOMMU dependencies between devices
for managing safe, userspace drivers.
[Some code changes by Joerg Roedel <joerg.roedel@amd.com>]
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Now that all IOMMU drivers are exporting their supported pgsizes,
we can remove the default pgsize settings in register_iommu().
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
When mapping a memory region, split it to page sizes as supported
by the iommu hardware. Always prefer bigger pages, when possible,
in order to reduce the TLB pressure.
The logic to do that is now added to the IOMMU core, so neither the iommu
drivers themselves nor users of the IOMMU API have to duplicate it.
This allows a more lenient granularity of mappings; traditionally the
IOMMU API took 'order' (of a page) as a mapping size, and directly let
the low level iommu drivers handle the mapping, but now that the IOMMU
core can split arbitrary memory regions into pages, we can remove this
limitation, so users don't have to split those regions by themselves.
Currently the supported page sizes are advertised once and they then
remain static. That works well for OMAP and MSM but it would probably
not fly well with intel's hardware, where the page size capabilities
seem to have the potential to be different between several DMA
remapping devices.
register_iommu() currently sets a default pgsize behavior, so we can convert
the IOMMU drivers in subsequent patches. After all the drivers
are converted, the temporary default settings will be removed.
Mainline users of the IOMMU API (kvm and omap-iovmm) are adopted
to deal with bytes instead of page order.
Many thanks to Joerg Roedel <Joerg.Roedel@amd.com> for significant review!
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Cc: David Brown <davidb@codeaurora.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Joerg Roedel <Joerg.Roedel@amd.com>
Cc: Stepan Moskovchenko <stepanm@codeaurora.org>
Cc: KyongHo Cho <pullip.cho@samsung.com>
Cc: Hiroshi DOYU <hdoyu@nvidia.com>
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Cc: kvm@vger.kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Express sizes in bytes rather than in page order, to eliminate the
size->order->size conversions we have whenever the IOMMU API is calling
the low level drivers' map/unmap methods.
Adopt all existing drivers.
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Cc: David Brown <davidb@codeaurora.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Joerg Roedel <Joerg.Roedel@amd.com>
Cc: Stepan Moskovchenko <stepanm@codeaurora.org>
Cc: KyongHo Cho <pullip.cho@samsung.com>
Cc: Hiroshi DOYU <hdoyu@nvidia.com>
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
With all IOMMU drivers being converted to bus_set_iommu the
global iommu_ops are no longer required. The same is true
for the deprecated register_iommu function.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
With per-bus iommu_ops the iommu_found function needs to
work on a bus_type too. This patch adds a bus_type parameter
to that function and converts all call-places.
The function is also renamed to iommu_present because the
function now checks if an iommu is present for a given bus
and does not check for a global iommu anymore.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
This is necessary to store a pointer to the bus-specific
iommu_ops in the iommu-domain structure. It will be used
later to call into bus-specific iommu-ops.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
This is the starting point to make the iommu_ops used for
the iommu-api a per-bus-type structure. It is required to
easily implement bus-specific setup in the iommu-layer.
The first user will be the iommu-group attribute in sysfs.
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Make report_iommu_fault() return -ENOSYS whenever an iommu fault
handler isn't installed, so IOMMU drivers can then do their own
platform-specific default behavior if they wanted.
Fault handlers can still return -ENOSYS in case they want to elicit the
default behavior of the IOMMU drivers.
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
commit 4f3f8d9 "iommu/core: Add fault reporting mechanism" added
the public iommu_set_fault_handler() symbol but forgot to export it.
Fix that.
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Add iommu fault report mechanism to the IOMMU API, so implementations
could report about mmu faults (translation errors, hardware errors,
etc..).
Fault reports can be used in several ways:
- mere logging
- reset the device that accessed the faulting address (may be necessary
in case the device is a remote processor for example)
- implement dynamic PTE/TLB loading
A dedicated iommu_set_fault_handler() API has been added to allow
users, who are interested to receive such reports, to provide
their handler.
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Replace the hard-coded 4kb by PAGE_SIZE to make iommu-api
implementations possible on architectures where
PAGE_SIZE != 4kb.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Replace iommu's alignment checks with the existing IS_ALIGNED macro,
to drop a few lines of code and utilize IS_ALIGNED's type safety.
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Create a dedicated folder for iommu drivers, and move the base
iommu implementation over there.
Grouping the various iommu drivers in a single location will help
finding similar problems shared by different platforms, so they
could be solved once, in the iommu framework, instead of solved
differently (or duplicated) in each driver.
Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Joerg Roedel