Commit 789a5913b2 ("iommu/amd: Use the generic iommu page table")
introduces the shared iommu page table for AMD IOMMU. Some bioses
contain an identity mapping for address 0x0, which is not parsed
properly (e.g., certain Strix Halo devices). This causes the DMA
components of the device to fail to initialize (e.g., the NVMe SSD
controller), leading to a failed post.
Specifically, on the GPD Win 5, the NVME and SSD GPU fail to mount,
making collecting errors difficult. While debugging, it was found that
a -EADDRINUSE error was emitted and its source was traced to
iommu_iova_to_phys(). After adding some debug prints, it was found that
phys_addr becomes 0, which causes the code to try to re-map the 0
address and fail, causing a cascade leading to a failed post. This is
because the GPD Win 5 contains a 0x0-0x1 identity mapping for DMA
devices, causing it to be repeated for each device.
The cause of this failure is the following check in
iommu_create_device_direct_mappings(), where address aliasing is handled
via the following check:
```
phys_addr = iommu_iova_to_phys(domain, addr);
if (!phys_addr) {
map_size += pg_size;
continue;
}
````
Obviously, the iommu_iova_to_phys() signature is faulty and aliases
unmapped and 0 together, causing the allocation code to try to
re-allocate the 0 address per device. However, it has too many
instantiations to fix. Therefore, use a ternary so that when addr
is 0, the check is done for address 1 instead.
Suggested-by: Robin Murphy <robin.murphy@arm.com>
Fixes: 789a5913b2 ("iommu/amd: Use the generic iommu page table")
Signed-off-by: Antheas Kapenekakis <lkml@antheas.dev>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
This is the exact same thing as the 'alloc_obj()' version, only much
smaller because there are a lot fewer users of the *alloc_flex()
interface.
As with alloc_obj() version, this was done entirely with mindless brute
force, using the same script, except using 'flex' in the pattern rather
than 'objs*'.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This was done entirely with mindless brute force, using
git grep -l '\<k[vmz]*alloc_objs*(.*, GFP_KERNEL)' |
xargs sed -i 's/\(alloc_objs*(.*\), GFP_KERNEL)/\1)/'
to convert the new alloc_obj() users that had a simple GFP_KERNEL
argument to just drop that argument.
Note that due to the extreme simplicity of the scripting, any slightly
more complex cases spread over multiple lines would not be triggered:
they definitely exist, but this covers the vast bulk of the cases, and
the resulting diff is also then easier to check automatically.
For the same reason the 'flex' versions will be done as a separate
conversion.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the result of running the Coccinelle script from
scripts/coccinelle/api/kmalloc_objs.cocci. The script is designed to
avoid scalar types (which need careful case-by-case checking), and
instead replace kmalloc-family calls that allocate struct or union
object instances:
Single allocations: kmalloc(sizeof(TYPE), ...)
are replaced with: kmalloc_obj(TYPE, ...)
Array allocations: kmalloc_array(COUNT, sizeof(TYPE), ...)
are replaced with: kmalloc_objs(TYPE, COUNT, ...)
Flex array allocations: kmalloc(struct_size(PTR, FAM, COUNT), ...)
are replaced with: kmalloc_flex(*PTR, FAM, COUNT, ...)
(where TYPE may also be *VAR)
The resulting allocations no longer return "void *", instead returning
"TYPE *".
Signed-off-by: Kees Cook <kees@kernel.org>
Use LIST_HEAD() to declare and initialize the 'mappings' list head in
iommu_create_device_direct_mappings() instead of separate declaration and
INIT_LIST_HEAD(). This simplifies the code by combining declaration and
initialization into a single idiomatic form, improving readability without
changing functionality.
Signed-off-by: Can Peng <pengcan@kylinos.cn>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Add calls for the new iommu debug config IOMMU_DEBUG_PAGEALLOC:
- iommu_debug_init: Enable the debug mode if configured by the user.
- iommu_debug_map: Track iommu pages mapped, using physical address.
- iommu_debug_unmap_begin: Track start of iommu unmap operation, with
IOVA and size.
- iommu_debug_unmap_end: Track the end of unmap operation, passing the
actual unmapped size versus the tracked one at unmap_begin.
We have to do the unmap_begin/end as once pages are unmapped we lose
the information of the physical address.
This is racy, but the API is racy by construction as it uses refcounts
and doesn't attempt to lock/synchronize with the IOMMU API as that will
be costly, meaning that possibility of false negative exists.
Reviewed-by: Samiullah Khawaja <skhawaja@google.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Pranjal Shrivastava <praan@google.com>
Signed-off-by: Mostafa Saleh <smostafa@google.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
PCIe permits a device to ignore ATS invalidation TLPs while processing a
reset. This creates a problem visible to the OS where an ATS invalidation
command will time out. E.g. an SVA domain will have no coordination with a
reset event and can racily issue ATS invalidations to a resetting device.
The OS should do something to mitigate this as we do not want production
systems to be reporting critical ATS failures, especially in a hypervisor
environment. Broadly, OS could arrange to ignore the timeouts, block page
table mutations to prevent invalidations, or disable and block ATS.
The PCIe r6.0, sec 10.3.1 IMPLEMENTATION NOTE recommends SW to disable and
block ATS before initiating a Function Level Reset. It also mentions that
other reset methods could have the same vulnerability as well.
Provide a callback from the PCI subsystem that will enclose the reset and
have the iommu core temporarily change all the attached RID/PASID domains
group->blocking_domain so that the IOMMU hardware would fence any incoming
ATS queries. And IOMMU drivers should also synchronously stop issuing new
ATS invalidations and wait for all ATS invalidations to complete. This can
avoid any ATS invaliation timeouts.
However, if there is a domain attachment/replacement happening during an
ongoing reset, ATS routines may be re-activated between the two function
calls. So, introduce a new resetting_domain in the iommu_group structure
to reject any concurrent attach_dev/set_dev_pasid call during a reset for
a concern of compatibility failure. Since this changes the behavior of an
attach operation, update the uAPI accordingly.
Note that there are two corner cases:
1. Devices in the same iommu_group
Since an attachment is always per iommu_group, this means that any
sibling devices in the iommu_group cannot change domain, to prevent
race conditions.
2. An SR-IOV PF that is being reset while its VF is not
In such case, the VF itself is already broken. So, there is no point
in preventing PF from going through the iommu reset.
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Tested-by: Dheeraj Kumar Srivastava <dheerajkumar.srivastava@amd.com>
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
There is a need to stage a resetting PCI device to temporarily the blocked
domain and then attach back to its previously attached domain after reset.
This can be simply done by keeping the "previously attached domain" in the
iommu_group->domain pointer while adding an iommu_group->resetting_domain,
which gives troubles to IOMMU drivers using the iommu_get_domain_for_dev()
for a device's physical domain in order to program IOMMU hardware.
And in such for-driver use cases, the iommu_group->mutex must be held, so
it doesn't fit in external callers that don't hold the iommu_group->mutex.
Introduce a new iommu_driver_get_domain_for_dev() helper, exclusively for
driver use cases that hold the iommu_group->mutex, to separate from those
external use cases.
Add a lockdep_assert_not_held to the existing iommu_get_domain_for_dev()
and highlight that in a kdoc.
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Tested-by: Dheeraj Kumar Srivastava <dheerajkumar.srivastava@amd.com>
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
This function can only be called on the default_domain. Trivally pass it
in. In all three existing cases, the default domain was just attached to
the device.
This avoids iommu_setup_dma_ops() calling iommu_get_domain_for_dev() that
will be used by external callers.
Suggested-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Tested-by: Dheeraj Kumar Srivastava <dheerajkumar.srivastava@amd.com>
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
The iommu_deferred_attach() function invokes __iommu_attach_device(), but
doesn't hold the group->mutex like other __iommu_attach_device() callers.
Though there is no pratical bug being triggered so far, it would be better
to apply the same locking to this __iommu_attach_device(), since the IOMMU
drivers nowaday are more aware of the group->mutex -- some of them use the
iommu_group_mutex_assert() function that could be potentially in the path
of an attach_dev callback function invoked by the __iommu_attach_device().
Worth mentioning that the iommu_deferred_attach() will soon need to check
group->resetting_domain that must be locked also.
Thus, grab the mutex to guard __iommu_attach_device() like other callers.
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Tested-by: Dheeraj Kumar Srivastava <dheerajkumar.srivastava@amd.com>
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-by: Samiullah Khawaja <skhawaja@google.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
The IOMMU core attaches each device to a default domain on probe(). Then,
every new "attach" operation has a fundamental meaning of two-fold:
- detach from its currently attached (old) domain
- attach to a given new domain
Modern IOMMU drivers following this pattern usually want to clean up the
things related to the old domain, so they call iommu_get_domain_for_dev()
to fetch the old domain.
Pass in the old domain pointer from the core to drivers, aligning with the
set_dev_pasid op that does so already.
Ensure all low-level attach fcuntions in the core can forward the correct
old domain pointer. Thus, rework those functions as well.
Suggested-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
The last gdev is the device that failed the __iommu_device_set_domain().
So, it doesn't need to be reverted, given it's attached to group->domain
already.
This is not a problem currently, since it's a simply re-attach. However,
the core will need to pass in the old domain to __iommu_device_set_domain
so the old domain pointers would be inconsistent between a failed device
and all its prior succeeded devices, as all the prior devices need to be
reverted.
Avoid the re-attach for the last gdev, by breaking before the revert.
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Generally an IOMMU driver should leave the translation as BLOCKED until the
translation entry is probed onto a struct device. When the struct device is
removed, the translation should be put back to BLOCKED.
Drivers that are able to work like this can set their release_domain to the
blocking domain, and the core code handles this work.
The exception is when the device has an IOMMU_RESV_DIRECT region, in which
case the OS should continuously allow translations for the given range. And
the core code generally prevents using a BLOCKED domain with this device.
Continue this logic for the device release and hoist some open coding from
drivers. If the device has dev->iommu->require_direct and the driver uses a
BLOCKED release_domain, override it to IDENTITY to preserve the semantics.
The only remaining required driver code for IOMMU_RESV_DIRECT should preset
an IDENTITY translation during early IOMMU startup for those devices.
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Since the bus ops were retired the iommu subsystem changed to using fwspec
to match the iommu driver to the iommu device. If a device has a NULL
fwspec then it is matched to the first iommu driver with a NULL fwspec,
effectively disabling support for systems with more than one non-fwspec
iommu driver.
Thus, if the iommufd selfest are run in an x86 system that registers a
non-fwspec iommu driver they fail to bind their mock devices to the mock
iommu driver.
Fix this by allocating a software fwnode for mock iommu driver's
iommu_device, and set it to the device which mock iommu driver created.
This is done by adding a new helper iommu_mock_device_add() which abuses
the internals of the fwspec system to establish a fwspec before the device
is added and is careful not to leak it. A matching dummy fwspec is
automatically added to the mock iommu driver.
Test by "make -C toosl/testing/selftests TARGETS=iommu run_tests":
PASSED: 229 / 229 tests passed.
In addition, this issue is also can be found on amd platform, and
also tested on a amd machine.
Link: https://patch.msgid.link/r/20250925054730.3877-1-kanie@linux.alibaba.com
Fixes: 17de3f5fdd ("iommu: Retire bus ops")
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Tested-by: Qinyun Tan <qinyuntan@linux.alibaba.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Pull iommu updates from Joerg Roedel:
"Core:
- Introduction of iommu-pages infrastructure to consolitate
page-table allocation code among hardware drivers. This is
ground-work for more generalization in the future
- Remove IOMMU_DEV_FEAT_SVA and IOMMU_DEV_FEAT_IOPF feature flags
- Convert virtio-iommu to domain_alloc_paging()
- KConfig cleanups
- Some small fixes for possible overflows and race conditions
Intel VT-d driver:
- Restore WO permissions on second-level paging entries
- Use ida to manage domain id
- Miscellaneous cleanups
AMD-Vi:
- Make sure notifiers finish running before module unload
- Add support for HTRangeIgnore feature
- Allow matching ACPI HID devices without matching UIDs
ARM-SMMU:
- SMMUv2:
- Recognise the compatible string for SAR2130P MDSS in the
Qualcomm driver, as this device requires an identity domain
- Fix Adreno stall handling so that GPU debugging is more robust
and doesn't e.g. result in deadlock
- SMMUv3:
- Fix ->attach_dev() error reporting for unrecognised domains
- IO-pgtable:
- Allow clients (notably, drivers that process requests from
userspace) to silence warnings when mapping an already-mapped
IOVA
S390:
- Add support for additional table regions
Mediatek:
- Add support for MT6893 MM IOMMU
And some smaller fixes and improvements in various other drivers"
* tag 'iommu-updates-v6.16' of git://git.kernel.org/pub/scm/linux/kernel/git/iommu/linux: (75 commits)
iommu/vt-d: Restore context entry setup order for aliased devices
iommu/mediatek: Fix compatible typo for mediatek,mt6893-iommu-mm
iommu/arm-smmu-qcom: Make set_stall work when the device is on
iommu/arm-smmu: Move handing of RESUME to the context fault handler
iommu/arm-smmu-qcom: Enable threaded IRQ for Adreno SMMUv2/MMU500
iommu/io-pgtable-arm: Add quirk to quiet WARN_ON()
iommu: Clear the freelist after iommu_put_pages_list()
iommu/vt-d: Change dmar_ats_supported() to return boolean
iommu/vt-d: Eliminate pci_physfn() in dmar_find_matched_satc_unit()
iommu/vt-d: Replace spin_lock with mutex to protect domain ida
iommu/vt-d: Use ida to manage domain id
iommu/vt-d: Restore WO permissions on second-level paging entries
iommu/amd: Allow matching ACPI HID devices without matching UIDs
iommu: make inclusion of arm/arm-smmu-v3 directory conditional
iommu: make inclusion of riscv directory conditional
iommu: make inclusion of amd directory conditional
iommu: make inclusion of intel directory conditional
iommu: remove duplicate selection of DMAR_TABLE
iommu/fsl_pamu: remove trailing space after \n
iommu/arm-smmu-qcom: Add SAR2130P MDSS compatible
...
Pull dma-mapping updates from Marek Szyprowski:
"New two step DMA mapping API, which is is a first step to a long path
to provide alternatives to scatterlist and to remove hacks, abuses and
design mistakes related to scatterlists.
This new approach optimizes some calls to DMA-IOMMU layer and cache
maintenance by batching them, reduces memory usage as it is no need to
store mapped DMA addresses to unmap them, and reduces some function
call overhead. It is a combination effort of many people, lead and
developed by Christoph Hellwig and Leon Romanovsky"
* tag 'dma-mapping-6.16-2025-05-26' of git://git.kernel.org/pub/scm/linux/kernel/git/mszyprowski/linux:
docs: core-api: document the IOVA-based API
dma-mapping: add a dma_need_unmap helper
dma-mapping: Implement link/unlink ranges API
iommu/dma: Factor out a iommu_dma_map_swiotlb helper
dma-mapping: Provide an interface to allow allocate IOVA
iommu: add kernel-doc for iommu_unmap_fast
iommu: generalize the batched sync after map interface
dma-mapping: move the PCI P2PDMA mapping helpers to pci-p2pdma.h
PCI/P2PDMA: Refactor the p2pdma mapping helpers
Generally PASID support requires ACS settings that usually create
single device groups, but there are some niche cases where we can get
multi-device groups and still have working PASID support. The primary
issue is that PCI switches are not required to treat PASID tagged TLPs
specially so appropriate ACS settings are required to route all TLPs to
the host bridge if PASID is going to work properly.
pci_enable_pasid() does check that each device that will use PASID has
the proper ACS settings to achieve this routing.
However, no-PASID devices can be combined with PASID capable devices
within the same topology using non-uniform ACS settings. In this case
the no-PASID devices may not have strict route to host ACS flags and
end up being grouped with the PASID devices.
This configuration fails to allow use of the PASID within the iommu
core code which wrongly checks if the no-PASID device supports PASID.
Fix this by ignoring no-PASID devices during the PASID validation. They
will never issue a PASID TLP anyhow so they can be ignored.
Fixes: c404f55c26 ("iommu: Validate the PASID in iommu_attach_device_pasid()")
Cc: stable@vger.kernel.org
Signed-off-by: Tushar Dave <tdave@nvidia.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Link: https://lore.kernel.org/r/20250520011937.3230557-1-tdave@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
For the upcoming IOVA-based DMA API we want to batch the
ops->iotlb_sync_map() call after mapping multiple IOVAs from
dma-iommu without having a scatterlist. Improve the API.
Add a wrapper for the map_sync as iommu_sync_map() so that callers
don't need to poke into the methods directly.
Formalize __iommu_map() into iommu_map_nosync() which requires the
caller to call iommu_sync_map() after all maps are completed.
Refactor the existing sanity checks from all the different layers
into iommu_map_nosync().
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Will Deacon <will@kernel.org>
Tested-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
On a 32 bit system calling:
iommu_map(0, 0x40000000)
When using the AMD V1 page table type with a domain->pgsize of 0xfffff000
causes iommu_pgsize() to miscalculate a result of:
size=0x40000000 count=2
count should be 1. This completely corrupts the mapping process.
This is because the final test to adjust the pagesize malfunctions when
the addition overflows. Use check_add_overflow() to prevent this.
Fixes: b1d99dc5f9 ("iommu: Hook up '->unmap_pages' driver callback")
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/0-v1-3ad28fc2e3a3+163327-iommu_overflow_pgsize_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Next up on our list of race windows to close is another one during
iommu_device_register() - it's now OK again for multiple instances to
run their bus_iommu_probe() in parallel, but an iommu_probe_device() can
still also race against a running bus_iommu_probe(). As Johan has
managed to prove, this has now become a lot more visible on DT platforms
wth driver_async_probe where a client driver is attempting to probe in
parallel with its IOMMU driver - although commit b46064a188 ("iommu:
Handle race with default domain setup") resolves this from the client
driver's point of view, this isn't before of_iommu_configure() has had
the chance to attempt to "replay" a probe that the bus walk hasn't even
tried yet, and so still cause the out-of-order group allocation
behaviour that we're trying to clean up (and now warning about).
The most reliable thing to do here is to explicitly keep track of the
"iommu_device_register() is still running" state, so we can then
special-case the ops lookup for the replay path (based on dev->iommu
again) to let that think it's still waiting for the IOMMU driver to
appear at all. This still leaves the longstanding theoretical case of
iommu_bus_notifier() being triggered during bus_iommu_probe(), but it's
not so simple to defer a notifier, and nobody's ever reported that being
a visible issue, so let's quietly kick that can down the road for now...
Reported-by: Johan Hovold <johan@kernel.org>
Fixes: bcb81ac6ae ("iommu: Get DT/ACPI parsing into the proper probe path")
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/88d54c1b48fed8279aa47d30f3d75173685bb26a.1745516488.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The idxd driver attaches the default domain to a PASID of the device to
perform kernel DMA using that PASID. The domain is attached to the
device's PASID through iommu_attach_device_pasid(), which checks if the
domain->owner matches the iommu_ops retrieved from the device. If they
do not match, it returns a failure.
if (ops != domain->owner || pasid == IOMMU_NO_PASID)
return -EINVAL;
The static identity domain implemented by the intel iommu driver doesn't
specify the domain owner. Therefore, kernel DMA with PASID doesn't work
for the idxd driver if the device translation mode is set to passthrough.
Generally the owner field of static domains are not set because they are
already part of iommu ops. Add a helper domain_iommu_ops_compatible()
that checks if a domain is compatible with the device's iommu ops. This
helper explicitly allows the static blocked and identity domains associated
with the device's iommu_ops to be considered compatible.
Fixes: 2031c469f8 ("iommu/vt-d: Add support for static identity domain")
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220031
Cc: stable@vger.kernel.org
Suggested-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/linux-iommu/20250422191554.GC1213339@ziepe.ca/
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/20250424034123.2311362-1-baolu.lu@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
fsl_pamu is the last user of domain_alloc(), and it is using it to create
something weird that doesn't really fit into the iommu subsystem
architecture. It is a not a paging domain since it doesn't have any
map/unmap ops. It may be some special kind of identity domain.
For now just leave it as is. Wrap it's definition in CONFIG_FSL_PAMU to
discourage any new drivers from attempting to use it.
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/5-v4-ff5fb6b03bd1+288-iommu_virtio_domains_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
virtio has the complication that it sometimes wants to return a paging
domain for IDENTITY which makes this conversion a little different than
other drivers.
Add a viommu_domain_alloc_paging() that combines viommu_domain_alloc() and
viommu_domain_finalise() to always return a fully initialized and
finalized paging domain.
Use viommu_domain_alloc_identity() to implement the special non-bypass
IDENTITY flow by calling viommu_domain_alloc_paging() then
viommu_domain_map_identity().
Remove support for deferred finalize and the vdomain->mutex.
Remove core support for domain_alloc() IDENTITY as virtio was the last
driver using it.
Reviewed-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/3-v4-ff5fb6b03bd1+288-iommu_virtio_domains_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
virtio-iommu has a mode where the IDENTITY domain is actually a paging
domain with an identity mapping covering some of the system address
space manually created.
To support this add a new domain_alloc_identity() op that accepts
the struct device so that virtio can allocate and fully finalize a
paging domain to return.
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Link: https://lore.kernel.org/r/2-v4-ff5fb6b03bd1+288-iommu_virtio_domains_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Although the lock-juggling is only a temporary workaround, we don't want
it to make things avoidably worse. Jason was right to be nervous, since
bus_iommu_probe() doesn't care *which* IOMMU instance it's probing for,
so it probably is possible for one walk to finish a probe which a
different walk started, thus we do want to check for that.
Also there's no need to drop the lock just to have of_iommu_configure()
do nothing when a fwspec already exists; check that directly and avoid
opening a window at all in that (still somewhat likely) case.
Suggested-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/09d901ad11b3a410fbb6e27f7d04ad4609c3fe4a.1741706365.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The following crash is observed while handling an IOMMU fault with a
recent kernel:
kernel tried to execute NX-protected page - exploit attempt? (uid: 0)
BUG: unable to handle page fault for address: ffff8c708299f700
PGD 19ee01067 P4D 19ee01067 PUD 101c10063 PMD 80000001028001e3
Oops: Oops: 0011 [#1] SMP NOPTI
CPU: 4 UID: 0 PID: 139 Comm: irq/25-AMD-Vi Not tainted 6.15.0-rc1+ #20 PREEMPT(lazy)
Hardware name: LENOVO 21D0/LNVNB161216, BIOS J6CN50WW 09/27/2024
RIP: 0010:0xffff8c708299f700
Call Trace:
<TASK>
? report_iommu_fault+0x78/0xd3
? amd_iommu_report_page_fault+0x91/0x150
? amd_iommu_int_thread+0x77/0x180
? __pfx_irq_thread_fn+0x10/0x10
? irq_thread_fn+0x23/0x60
? irq_thread+0xf9/0x1e0
? __pfx_irq_thread_dtor+0x10/0x10
? __pfx_irq_thread+0x10/0x10
? kthread+0xfc/0x240
? __pfx_kthread+0x10/0x10
? ret_from_fork+0x34/0x50
? __pfx_kthread+0x10/0x10
? ret_from_fork_asm+0x1a/0x30
</TASK>
report_iommu_fault() checks for an installed handler comparing the
corresponding field to NULL. It can (and could before) be called for a
domain with a different cookie type - IOMMU_COOKIE_DMA_IOVA, specifically.
Cookie is represented as a union so we may end up with a garbage value
treated there if this happens for a domain with another cookie type.
Formerly there were two exclusive cookie types in the union.
IOMMU_DOMAIN_SVA has a dedicated iommu_report_device_fault().
Call the fault handler only if the passed domain has a required cookie
type.
Found by Linux Verification Center (linuxtesting.org).
Fixes: 6aa63a4ec9 ("iommu: Sort out domain user data")
Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/20250408213342.285955-1-pchelkin@ispras.ru
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Pull iommufd updates from Jason Gunthorpe:
"Two significant new items:
- Allow reporting IOMMU HW events to userspace when the events are
clearly linked to a device.
This is linked to the VIOMMU object and is intended to be used by a
VMM to forward HW events to the virtual machine as part of
emulating a vIOMMU. ARM SMMUv3 is the first driver to use this
mechanism. Like the existing fault events the data is delivered
through a simple FD returning event records on read().
- PASID support in VFIO.
The "Process Address Space ID" is a PCI feature that allows the
device to tag all PCI DMA operations with an ID. The IOMMU will
then use the ID to select a unique translation for those DMAs. This
is part of Intel's vIOMMU support as VT-D HW requires the
hypervisor to manage each PASID entry.
The support is generic so any VFIO user could attach any
translation to a PASID, and the support should work on ARM SMMUv3
as well. AMD requires additional driver work.
Some minor updates, along with fixes:
- Prevent using nested parents with fault's, no driver support today
- Put a single "cookie_type" value in the iommu_domain to indicate
what owns the various opaque owner fields"
* tag 'for-linus-iommufd' of git://git.kernel.org/pub/scm/linux/kernel/git/jgg/iommufd: (49 commits)
iommufd: Test attach before detaching pasid
iommufd: Fix iommu_vevent_header tables markup
iommu: Convert unreachable() to BUG()
iommufd: Balance veventq->num_events inc/dec
iommufd: Initialize the flags of vevent in iommufd_viommu_report_event()
iommufd/selftest: Add coverage for reporting max_pasid_log2 via IOMMU_HW_INFO
iommufd: Extend IOMMU_GET_HW_INFO to report PASID capability
vfio: VFIO_DEVICE_[AT|DE]TACH_IOMMUFD_PT support pasid
vfio-iommufd: Support pasid [at|de]tach for physical VFIO devices
ida: Add ida_find_first_range()
iommufd/selftest: Add coverage for iommufd pasid attach/detach
iommufd/selftest: Add test ops to test pasid attach/detach
iommufd/selftest: Add a helper to get test device
iommufd/selftest: Add set_dev_pasid in mock iommu
iommufd: Allow allocating PASID-compatible domain
iommu/vt-d: Add IOMMU_HWPT_ALLOC_PASID support
iommufd: Enforce PASID-compatible domain for RID
iommufd: Support pasid attach/replace
iommufd: Enforce PASID-compatible domain in PASID path
iommufd/device: Add pasid_attach array to track per-PASID attach
...
When DMA/MSI cookies were made first-class citizens back in commit
46983fcd67 ("iommu: Pull IOVA cookie management into the core"), there
was no real need to further expose the two different cookie types.
However, now that IOMMUFD wants to add a third type of MSI-mapping
cookie, we do have a nicely compelling reason to properly dismabiguate
things at the domain level beyond just vaguely guessing from the domain
type.
Meanwhile, we also effectively have another "cookie" in the form of the
anonymous union for other user data, which isn't much better in terms of
being vague and unenforced. The fact is that all these cookie types are
mutually exclusive, in the sense that combining them makes zero sense
and/or would be catastrophic (iommu_set_fault_handler() on an SVA
domain, anyone?) - the only combination which *might* be reasonable is
perhaps a fault handler and an MSI cookie, but nobody's doing that at
the moment, so let's rule it out as well for the sake of being clear and
robust. To that end, we pull DMA and MSI cookies apart a little more,
mostly to clear up the ambiguity at domain teardown, then for clarity
(and to save a little space), move them into the union, whose ownership
we can then properly describe and enforce entirely unambiguously.
[nicolinc: rebase on latest tree; use prefix IOMMU_COOKIE_; merge unions
in iommu_domain; add IOMMU_COOKIE_IOMMUFD for iommufd_hwpt]
Link: https://patch.msgid.link/r/1ace9076c95204bbe193ee77499d395f15f44b23.1742871535.git.nicolinc@nvidia.com
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In hindsight, there were some crucial subtleties overlooked when moving
{of,acpi}_dma_configure() to driver probe time to allow waiting for
IOMMU drivers with -EPROBE_DEFER, and these have become an
ever-increasing source of problems. The IOMMU API has some fundamental
assumptions that iommu_probe_device() is called for every device added
to the system, in the order in which they are added. Calling it in a
random order or not at all dependent on driver binding leads to
malformed groups, a potential lack of isolation for devices with no
driver, and all manner of unexpected concurrency and race conditions.
We've attempted to mitigate the latter with point-fix bodges like
iommu_probe_device_lock, but it's a losing battle and the time has come
to bite the bullet and address the true source of the problem instead.
The crux of the matter is that the firmware parsing actually serves two
distinct purposes; one is identifying the IOMMU instance associated with
a device so we can check its availability, the second is actually
telling that instance about the relevant firmware-provided data for the
device. However the latter also depends on the former, and at the time
there was no good place to defer and retry that separately from the
availability check we also wanted for client driver probe.
Nowadays, though, we have a proper notion of multiple IOMMU instances in
the core API itself, and each one gets a chance to probe its own devices
upon registration, so we can finally make that work as intended for
DT/IORT/VIOT platforms too. All we need is for iommu_probe_device() to
be able to run the iommu_fwspec machinery currently buried deep in the
wrong end of {of,acpi}_dma_configure(). Luckily it turns out to be
surprisingly straightforward to bootstrap this transformation by pretty
much just calling the same path twice. At client driver probe time,
dev->driver is obviously set; conversely at device_add(), or a
subsequent bus_iommu_probe(), any device waiting for an IOMMU really
should *not* have a driver already, so we can use that as a condition to
disambiguate the two cases, and avoid recursing back into the IOMMU core
at the wrong times.
Obviously this isn't the nicest thing, but for now it gives us a
functional baseline to then unpick the layers in between without many
more awkward cross-subsystem patches. There are some minor side-effects
like dma_range_map potentially being created earlier, and some debug
prints being repeated, but these aren't significantly detrimental. Let's
make things work first, then deal with making them nice.
With the basic flow finally in the right order again, the next step is
probably turning the bus->dma_configure paths inside-out, since all we
really need from bus code is its notion of which device and input ID(s)
to parse the common firmware properties with...
Acked-by: Bjorn Helgaas <bhelgaas@google.com> # pci-driver.c
Acked-by: Rob Herring (Arm) <robh@kernel.org> # of/device.c
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Lorenzo Pieralisi <lpieralisi@kernel.org>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/e3b191e6fd6ca9a1e84c5e5e40044faf97abb874.1740753261.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
At the moment, if of_iommu_configure() allocates dev->iommu itself via
iommu_fwspec_init(), then suffers a DT parsing failure, it cleans up the
fwspec but leaves the empty dev_iommu hanging around. So far this is
benign (if a tiny bit wasteful), but we'd like to be able to reason
about dev->iommu having a consistent and unambiguous lifecycle. Thus
make sure that the of_iommu cleanup undoes precisely whatever it did.
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/d219663a3f23001f23d520a883ac622d70b4e642.1740753261.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
It turns out that deferred default domain creation leaves a subtle
race window during iommu_device_register() wherein a client driver may
asynchronously probe in parallel and get as far as performing DMA API
operations with dma-direct, only to be switched to iommu-dma underfoot
once the default domain attachment finally happens, with obviously
disastrous consequences. Even the wonky of_iommu_configure() path is at
risk, since iommu_fwspec_init() will no longer defer client probe as the
instance ops are (necessarily) already registered, and the "replay"
iommu_probe_device() call can see dev->iommu_group already set and so
think there's nothing to do either.
Fortunately we already have the right tool in the right place in the
form of iommu_device_use_default_domain(), which just needs to ensure
that said default domain is actually ready to *be* used. Deferring the
client probe shouldn't have too much impact, given that this only
happens while the IOMMU driver is probing, and thus due to kick the
deferred probe list again once it finishes.
Reported-by: Charan Teja Kalla <quic_charante@quicinc.com>
Fixes: 98ac73f99b ("iommu: Require a default_domain for all iommu drivers")
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/e88b94c9b575034a2c98a48b3d383654cbda7902.1740753261.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The drivers doing their own fwspec parsing have no need to call
iommu_fwspec_free() since fwspecs were moved into dev_iommu, as
returning an error from .probe_device will tear down the whole lot
anyway. Move it into the private interface now that it only serves
for of_iommu to clean up in an error case.
I have no idea what mtk_v1 was doing in effectively guaranteeing
a NULL fwspec would be dereferenced if no "iommus" DT property was
found, so add a check for that to at least make the code look sane.
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/36e245489361de2d13db22a510fa5c79e7126278.1740667667.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
iommu shared branch with iommufd
The three dependent series on a shared branch:
- Change the iommufd fault handle into an always present hwpt handle in
the domain
- Give iommufd its own SW_MSI implementation along with some IRQ layer
rework
- Improvements to the handle attach API
The current implementation stores entry to the group->pasid_array before
the underlying iommu driver has successfully set the new domain. This can
lead to issues where PRIs are received on the new domain before the attach
operation is completed.
This patch swaps the order of operations to ensure that the domain is set
in the underlying iommu driver before updating the group->pasid_array.
Link: https://patch.msgid.link/r/20250226011849.5102-5-yi.l.liu@intel.com
Suggested-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Yi Liu <yi.l.liu@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
iommu_attach_device_pasid() only stores handle to group->pasid_array
when there is a valid handle input. However, it makes the
iommu_attach_device_pasid() unable to detect if the pasid has been
attached or not previously.
To be complete, let the iommu_attach_device_pasid() store the domain
to group->pasid_array if no valid handle. The other users of the
group->pasid_array should be updated to be consistent. e.g. the
iommu_attach_group_handle() and iommu_replace_group_handle().
Link: https://patch.msgid.link/r/20250226011849.5102-4-yi.l.liu@intel.com
Suggested-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Yi Liu <yi.l.liu@intel.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
SW_MSI supports IOMMU to translate an MSI message before the MSI message
is delivered to the interrupt controller. On such systems, an iommu_domain
must have a translation for the MSI message for interrupts to work.
The IRQ subsystem will call into IOMMU to request that a physical page be
set up to receive MSI messages, and the IOMMU then sets an IOVA that maps
to that physical page. Ultimately the IOVA is programmed into the device
via the msi_msg.
Generalize this by allowing iommu_domain owners to provide implementations
of this mapping. Add a function pointer in struct iommu_domain to allow a
domain owner to provide its own implementation.
Have dma-iommu supply its implementation for IOMMU_DOMAIN_DMA types during
the iommu_get_dma_cookie() path. For IOMMU_DOMAIN_UNMANAGED types used by
VFIO (and iommufd for now), have the same iommu_dma_sw_msi set as well in
the iommu_get_msi_cookie() path.
Hold the group mutex while in iommu_dma_prepare_msi() to ensure the domain
doesn't change or become freed while running. Races with IRQ operations
from VFIO and domain changes from iommufd are possible here.
Replace the msi_prepare_lock with a lockdep assertion for the group mutex
as documentation. For the dmau_iommu.c each iommu_domain is unique to a
group.
Link: https://patch.msgid.link/r/4ca696150d2baee03af27c4ddefdb7b0b0280e7b.1740014950.git.nicolinc@nvidia.com
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
