Sleeping while atomic = bad. Let's fix an obvious typo to try to avoid it.
The warning that was seen (on a downstream kernel with the problematic
patch backported):
BUG: sleeping function called from invalid context at mm/page_alloc.c:4726
in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 9, name: ksoftirqd/0
CPU: 0 PID: 9 Comm: ksoftirqd/0 Not tainted 5.4.93-12508-gc10c93e28e39 #1
Call trace:
dump_backtrace+0x0/0x154
show_stack+0x20/0x2c
dump_stack+0xa0/0xfc
___might_sleep+0x11c/0x12c
__might_sleep+0x50/0x84
__alloc_pages_nodemask+0xf8/0x2bc
__arm_lpae_alloc_pages+0x48/0x1b4
__arm_lpae_map+0x124/0x274
__arm_lpae_map+0x1cc/0x274
arm_lpae_map+0x140/0x170
arm_smmu_map+0x78/0xbc
__iommu_map+0xd4/0x210
_iommu_map+0x4c/0x84
iommu_map_atomic+0x44/0x58
__iommu_dma_map+0x8c/0xc4
iommu_dma_map_page+0xac/0xf0
Fixes: d8c1df02ac ("iommu: Move iotlb_sync_map out from __iommu_map")
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Yong Wu <yong.wu@mediatek.com>
Acked-by: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20210201170611.1.I64a7b62579287d668d7c89e105dcedf45d641063@changeid
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Currently, because domain attach allows to be deferred from iommu
driver to device driver, and when iommu initializes, the devices
on the bus will be scanned and the default groups will be allocated.
Due to the above changes, some devices could be added to the same
group as below:
[ 3.859417] pci 0000:01:00.0: Adding to iommu group 16
[ 3.864572] pci 0000:01:00.1: Adding to iommu group 16
[ 3.869738] pci 0000:02:00.0: Adding to iommu group 17
[ 3.874892] pci 0000:02:00.1: Adding to iommu group 17
But when attaching these devices, it doesn't allow that a group has
more than one device, otherwise it will return an error. This conflicts
with the deferred attaching. Unfortunately, it has two devices in the
same group for my side, for example:
[ 9.627014] iommu_group_device_count(): device name[0]:0000:01:00.0
[ 9.633545] iommu_group_device_count(): device name[1]:0000:01:00.1
...
[ 10.255609] iommu_group_device_count(): device name[0]:0000:02:00.0
[ 10.262144] iommu_group_device_count(): device name[1]:0000:02:00.1
Finally, which caused the failure of tg3 driver when tg3 driver calls
the dma_alloc_coherent() to allocate coherent memory in the tg3_test_dma().
[ 9.660310] tg3 0000:01:00.0: DMA engine test failed, aborting
[ 9.754085] tg3: probe of 0000:01:00.0 failed with error -12
[ 9.997512] tg3 0000:01:00.1: DMA engine test failed, aborting
[ 10.043053] tg3: probe of 0000:01:00.1 failed with error -12
[ 10.288905] tg3 0000:02:00.0: DMA engine test failed, aborting
[ 10.334070] tg3: probe of 0000:02:00.0 failed with error -12
[ 10.578303] tg3 0000:02:00.1: DMA engine test failed, aborting
[ 10.622629] tg3: probe of 0000:02:00.1 failed with error -12
In addition, the similar situations also occur in other drivers such
as the bnxt_en driver. That can be reproduced easily in kdump kernel
when SME is active.
Let's move the handling currently in iommu_dma_deferred_attach() into
the iommu core code so that it can call the __iommu_attach_device()
directly instead of the iommu_attach_device(). The external interface
iommu_attach_device() is not suitable for handling this situation.
Signed-off-by: Lianbo Jiang <lijiang@redhat.com>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/20210126115337.20068-3-lijiang@redhat.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
iotlb_sync_map allow IOMMU drivers tlb sync after completing the whole
mapping. This patch adds iova and size as the parameters in it. then the
IOMMU driver could flush tlb with the whole range once after iova mapping
to improve performance.
Signed-off-by: Yong Wu <yong.wu@mediatek.com>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Acked-by: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20210107122909.16317-3-yong.wu@mediatek.com
Signed-off-by: Will Deacon <will@kernel.org>
In the end of __iommu_map, It alway call iotlb_sync_map.
This patch moves iotlb_sync_map out from __iommu_map since it is
unnecessary to call this for each sg segment especially iotlb_sync_map
is flush tlb all currently. Add a little helper _iommu_map for this.
Signed-off-by: Yong Wu <yong.wu@mediatek.com>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Acked-by: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20210107122909.16317-2-yong.wu@mediatek.com
Signed-off-by: Will Deacon <will@kernel.org>
Function iommu_dev_has_feature() has never been referenced in the tree,
and there does not appear to be anything coming soon to use it, so delete
it.
Signed-off-by: John Garry <john.garry@huawei.com>
Acked-by: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/1609940111-28563-7-git-send-email-john.garry@huawei.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Function iommu_map_sg_atomic() is only referenced in dma-iommu.c, which
can only be built-in, so stop exporting.
Signed-off-by: John Garry <john.garry@huawei.com>
Acked-by: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/1609940111-28563-5-git-send-email-john.garry@huawei.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Merge in IOMMU fixes for 5.10 in order to resolve conflicts against the
queue for 5.11.
* for-next/iommu/fixes:
iommu/amd: Set DTE[IntTabLen] to represent 512 IRTEs
iommu/vt-d: Don't read VCCAP register unless it exists
x86/tboot: Don't disable swiotlb when iommu is forced on
iommu: Check return of __iommu_attach_device()
arm-smmu-qcom: Ensure the qcom_scm driver has finished probing
iommu/amd: Enforce 4k mapping for certain IOMMU data structures
MAINTAINERS: Temporarily add myself to the IOMMU entry
iommu/vt-d: Fix compile error with CONFIG_PCI_ATS not set
iommu/vt-d: Avoid panic if iommu init fails in tboot system
iommu/vt-d: Cure VF irqdomain hickup
x86/platform/uv: Fix copied UV5 output archtype
x86/platform/uv: Drop last traces of uv_flush_tlb_others
Miscellaneous IOMMU changes for 5.11. Largely cosmetic, apart from a
change to the way in which identity-mapped domains are configured so
that the requests are now batched and can potentially use larger pages
for the mapping.
* for-next/iommu/misc:
iommu/io-pgtable-arm: Remove unused 'level' parameter from iopte_type() macro
iommu: Defer the early return in arm_(v7s/lpae)_map
iommu: Improve the performance for direct_mapping
iommu: return error code when it can't get group
iommu: Modify the description of iommu_sva_unbind_device
Currently direct_mapping always use the smallest pgsize which is SZ_4K
normally to mapping. This is unnecessary. we could gather the size, and
call iommu_map then, iommu_map could decide how to map better with the
just right pgsize.
>From the original comment, we should take care overlap, otherwise,
iommu_map may return -EEXIST. In this overlap case, we should map the
previous region before overlap firstly. then map the left part.
Each a iommu device will call this direct_mapping when its iommu
initialize, This patch is effective to improve the boot/initialization
time especially while it only needs level 1 mapping.
Signed-off-by: Anan Sun <anan.sun@mediatek.com>
Signed-off-by: Yong Wu <yong.wu@mediatek.com>
Link: https://lore.kernel.org/r/20201207093553.8635-1-yong.wu@mediatek.com
Signed-off-by: Will Deacon <will@kernel.org>
Although iommu_group_get() in iommu_probe_device() will always succeed
thanks to __iommu_probe_device() creating the group if it's not present,
it's still worth initialising 'ret' to -ENODEV in case this path is
reachable in the future.
For now, this patch results in no functional change.
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Link: https://lore.kernel.org/r/20201126133825.3643852-1-yangyingliang@huawei.com
Signed-off-by: Will Deacon <will@kernel.org>
"/sys/kernel/iommu_groups/<grp_id>/type" file could be read to find out the
default domain type of an iommu group. The default domain of an iommu group
doesn't change after booting and hence could be read directly. But,
after addding support to dynamically change iommu group default domain, the
above assumption no longer stays valid.
iommu group default domain type could be changed at any time by writing to
"/sys/kernel/iommu_groups/<grp_id>/type". So, take group mutex before
reading iommu group default domain type so that the user wouldn't see stale
values or iommu_group_show_type() doesn't try to derefernce stale pointers.
Signed-off-by: Sai Praneeth Prakhya <sai.praneeth.prakhya@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Ashok Raj <ashok.raj@intel.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Sohil Mehta <sohil.mehta@intel.com>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Link: https://lore.kernel.org/r/20201124130604.2912899-4-baolu.lu@linux.intel.com
Signed-off-by: Will Deacon <will@kernel.org>
Presently, the default domain of an iommu group is allocated during boot
time and it cannot be changed later. So, the device would typically be
either in identity (also known as pass_through) mode or the device would be
in DMA mode as long as the machine is up and running. There is no way to
change the default domain type dynamically i.e. after booting, a device
cannot switch between identity mode and DMA mode.
But, assume a use case wherein the user trusts the device and believes that
the OS is secure enough and hence wants *only* this device to bypass IOMMU
(so that it could be high performing) whereas all the other devices to go
through IOMMU (so that the system is protected). Presently, this use case
is not supported. It will be helpful if there is some way to change the
default domain of an iommu group dynamically. Hence, add such support.
A privileged user could request the kernel to change the default domain
type of a iommu group by writing to
"/sys/kernel/iommu_groups/<grp_id>/type" file. Presently, only three values
are supported
1. identity: all the DMA transactions from the device in this group are
*not* translated by the iommu
2. DMA: all the DMA transactions from the device in this group are
translated by the iommu
3. auto: change to the type the device was booted with
Note:
1. Default domain of an iommu group with two or more devices cannot be
changed.
2. The device in the iommu group shouldn't be bound to any driver.
3. The device shouldn't be assigned to user for direct access.
4. The change request will fail if any device in the group has a mandatory
default domain type and the requested one conflicts with that.
Please see "Documentation/ABI/testing/sysfs-kernel-iommu_groups" for more
information.
Signed-off-by: Sai Praneeth Prakhya <sai.praneeth.prakhya@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Ashok Raj <ashok.raj@intel.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Sohil Mehta <sohil.mehta@intel.com>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Link: https://lore.kernel.org/r/20201124130604.2912899-3-baolu.lu@linux.intel.com
Signed-off-by: Will Deacon <will@kernel.org>
Currently iommu_create_device_direct_mappings() is called
without checking the return of __iommu_attach_device(). This
may result in failures in iommu driver if dev attach returns
error.
Fixes: ce574c27ae ("iommu: Move iommu_group_create_direct_mappings() out of iommu_group_add_device()")
Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com>
Link: https://lore.kernel.org/r/20201119165846.34180-1-shameerali.kolothum.thodi@huawei.com
Signed-off-by: Will Deacon <will@kernel.org>
The "data->flags" variable is a u64 so if one of the high 32 bits is
set the original code will allow it, but it should be rejected. The
fix is to declare "mask" as a u64 instead of a u32.
Fixes: d90573812e ("iommu/uapi: Handle data and argsz filled by users")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Link: https://lore.kernel.org/r/20201103101623.GA1127762@mwanda
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Including:
- ARM-SMMU Updates from Will:
- Continued SVM enablement, where page-table is shared with
CPU
- Groundwork to support integrated SMMU with Adreno GPU
- Allow disabling of MSI-based polling on the kernel
command-line
- Minor driver fixes and cleanups (octal permissions, error
messages, ...)
- Secure Nested Paging Support for AMD IOMMU. The IOMMU will
fault when a device tries DMA on memory owned by a guest. This
needs new fault-types as well as a rewrite of the IOMMU memory
semaphore for command completions.
- Allow broken Intel IOMMUs (wrong address widths reported) to
still be used for interrupt remapping.
- IOMMU UAPI updates for supporting vSVA, where the IOMMU can
access address spaces of processes running in a VM.
- Support for the MT8167 IOMMU in the Mediatek IOMMU driver.
- Device-tree updates for the Renesas driver to support r8a7742.
- Several smaller fixes and cleanups all over the place.
-----BEGIN PGP SIGNATURE-----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=C0Sg
-----END PGP SIGNATURE-----
Merge tag 'iommu-updates-v5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu
Pull iommu updates from Joerg Roedel:
- ARM-SMMU Updates from Will:
- Continued SVM enablement, where page-table is shared with CPU
- Groundwork to support integrated SMMU with Adreno GPU
- Allow disabling of MSI-based polling on the kernel command-line
- Minor driver fixes and cleanups (octal permissions, error
messages, ...)
- Secure Nested Paging Support for AMD IOMMU. The IOMMU will fault when
a device tries DMA on memory owned by a guest. This needs new
fault-types as well as a rewrite of the IOMMU memory semaphore for
command completions.
- Allow broken Intel IOMMUs (wrong address widths reported) to still be
used for interrupt remapping.
- IOMMU UAPI updates for supporting vSVA, where the IOMMU can access
address spaces of processes running in a VM.
- Support for the MT8167 IOMMU in the Mediatek IOMMU driver.
- Device-tree updates for the Renesas driver to support r8a7742.
- Several smaller fixes and cleanups all over the place.
* tag 'iommu-updates-v5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu: (57 commits)
iommu/vt-d: Gracefully handle DMAR units with no supported address widths
iommu/vt-d: Check UAPI data processed by IOMMU core
iommu/uapi: Handle data and argsz filled by users
iommu/uapi: Rename uapi functions
iommu/uapi: Use named union for user data
iommu/uapi: Add argsz for user filled data
docs: IOMMU user API
iommu/qcom: add missing put_device() call in qcom_iommu_of_xlate()
iommu/arm-smmu-v3: Add SVA device feature
iommu/arm-smmu-v3: Check for SVA features
iommu/arm-smmu-v3: Seize private ASID
iommu/arm-smmu-v3: Share process page tables
iommu/arm-smmu-v3: Move definitions to a header
iommu/io-pgtable-arm: Move some definitions to a header
iommu/arm-smmu-v3: Ensure queue is read after updating prod pointer
iommu/amd: Re-purpose Exclusion range registers to support SNP CWWB
iommu/amd: Add support for RMP_PAGE_FAULT and RMP_HW_ERR
iommu/amd: Use 4K page for completion wait write-back semaphore
iommu/tegra-smmu: Allow to group clients in same swgroup
iommu/tegra-smmu: Fix iova->phys translation
...
IOMMU user APIs are responsible for processing user data. This patch
changes the interface such that user pointers can be passed into IOMMU
code directly. Separate kernel APIs without user pointers are introduced
for in-kernel users of the UAPI functionality.
IOMMU UAPI data has a user filled argsz field which indicates the data
length of the structure. User data is not trusted, argsz must be
validated based on the current kernel data size, mandatory data size,
and feature flags.
User data may also be extended, resulting in possible argsz increase.
Backward compatibility is ensured based on size and flags (or
the functional equivalent fields) checking.
This patch adds sanity checks in the IOMMU layer. In addition to argsz,
reserved/unused fields in padding, flags, and version are also checked.
Details are documented in Documentation/userspace-api/iommu.rst
Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Reviewed-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Link: https://lore.kernel.org/r/1601051567-54787-6-git-send-email-jacob.jun.pan@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
User APIs such as iommu_sva_unbind_gpasid() may also be used by the
kernel. Since we introduced user pointer to the UAPI functions,
in-kernel callers cannot share the same APIs. In-kernel callers are also
trusted, there is no need to validate the data.
We plan to have two flavors of the same API functions, one called
through ioctls, carrying a user pointer and one called directly with
valid IOMMU UAPI structs. To differentiate both, let's rename existing
functions with an iommu_uapi_ prefix.
Suggested-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Link: https://lore.kernel.org/r/1601051567-54787-5-git-send-email-jacob.jun.pan@linux.intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
PASID is defined as a few different types in iommu including "int",
"u32", and "unsigned int". To be consistent and to match with uapi
definitions, define PASID and its variations (e.g. max PASID) as "u32".
"u32" is also shorter and a little more explicit than "unsigned int".
No PASID type change in uapi although it defines PASID as __u64 in
some places.
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Joerg Roedel <jroedel@suse.de>
Link: https://lkml.kernel.org/r/1600187413-163670-2-git-send-email-fenghua.yu@intel.com
To keep naming consistent we should stick with *iotlb*. This patch
renames a few remaining functions.
Signed-off-by: Tom Murphy <murphyt7@tcd.ie>
Link: https://lore.kernel.org/r/20200817210051.13546-1-murphyt7@tcd.ie
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The sparse tool complains as follows:
drivers/iommu/iommu.c:386:5: warning:
symbol 'iommu_insert_resv_region' was not declared. Should it be static?
drivers/iommu/iommu.c:2182:5: warning:
symbol '__iommu_map' was not declared. Should it be static?
Those functions are not used outside of iommu.c, so mark them static.
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Link: https://lore.kernel.org/r/20200713142542.50294-1-weiyongjun1@huawei.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Some PCIe devices do not expect a PASID value in PRI Page Responses.
If the "PRG Response PASID Required" bit in the PRI capability is zero,
then the OS should not set the PASID field. Similarly on Arm SMMU,
responses to stall events do not have a PASID.
Currently iommu_page_response() systematically checks that the PASID in
the page response corresponds to the one in the page request. This can't
work with virtualization because a page response coming from a guest OS
won't have a PASID if the passed-through device does not require one.
Add a flag to page requests that declares whether the corresponding
response needs to have a PASID. When this flag isn't set, allow page
responses without PASID.
Reported-by: Shameerali Kolothum Thodi <shameerali.kolothum.thodi@huawei.com>
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Link: https://lore.kernel.org/r/20200616144712.748818-1-jean-philippe@linaro.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The iommu_group_do_dma_attach() must not attach devices which have
deferred_attach set. Otherwise devices could cause IOMMU faults when
re-initialized in a kdump kernel.
Fixes: deac0b3bed ("iommu: Split off default domain allocation from group assignment")
Reported-by: Jerry Snitselaar <jsnitsel@redhat.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Jerry Snitselaar <jsnitsel@redhat.com>
Reviewed-by: Jerry Snitselaar <jsnitsel@redhat.com>
Link: https://lore.kernel.org/r/20200604091944.26402-1-joro@8bytes.org
kobject_init_and_add() takes reference even when it fails.
Thus, when kobject_init_and_add() returns an error,
kobject_put() must be called to properly clean up the kobject.
Fixes: d72e31c937 ("iommu: IOMMU Groups")
Signed-off-by: Qiushi Wu <wu000273@umn.edu>
Link: https://lore.kernel.org/r/20200527210020.6522-1-wu000273@umn.edu
Signed-off-by: Joerg Roedel <jroedel@suse.de>
After binding a device to an mm, device drivers currently need to
register a mm_exit handler. This function is called when the mm exits,
to gracefully stop DMA targeting the address space and flush page faults
to the IOMMU.
This is deemed too complex for the MMU release() notifier, which may be
triggered by any mmput() invocation, from about 120 callsites [1]. The
upcoming SVA module has an example of such complexity: the I/O Page
Fault handler would need to call mmput_async() instead of mmput() after
handling an IOPF, to avoid triggering the release() notifier which would
in turn drain the IOPF queue and lock up.
Another concern is the DMA stop function taking too long, up to several
minutes [2]. For some mmput() callers this may disturb other users. For
example, if the OOM killer picks the mm bound to a device as the victim
and that mm's memory is locked, if the release() takes too long, it
might choose additional innocent victims to kill.
To simplify the MMU release notifier, don't forward the notification to
device drivers. Since they don't stop DMA on mm exit anymore, the PASID
lifetime is extended:
(1) The device driver calls bind(). A PASID is allocated.
Here any DMA fault is handled by mm, and on error we don't print
anything to dmesg. Userspace can easily trigger errors by issuing DMA
on unmapped buffers.
(2) exit_mmap(), for example the process took a SIGKILL. This step
doesn't happen during normal operations. Remove the pgd from the
PASID table, since the page tables are about to be freed. Invalidate
the IOTLBs.
Here the device may still perform DMA on the address space. Incoming
transactions are aborted but faults aren't printed out. ATS
Translation Requests return Successful Translation Completions with
R=W=0. PRI Page Requests return with Invalid Request.
(3) The device driver stops DMA, possibly following release of a fd, and
calls unbind(). PASID table is cleared, IOTLB invalidated if
necessary. The page fault queues are drained, and the PASID is
freed.
If DMA for that PASID is still running here, something went seriously
wrong and errors should be reported.
For now remove iommu_sva_ops entirely. We might need to re-introduce
them at some point, for example to notify device drivers of unhandled
IOPF.
[1] https://lore.kernel.org/linux-iommu/20200306174239.GM31668@ziepe.ca/
[2] https://lore.kernel.org/linux-iommu/4d68da96-0ad5-b412-5987-2f7a6aa796c3@amd.com/
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Acked-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200423125329.782066-3-jean-philippe@linaro.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The iommu_alloc_default_domain() function takes a reference to an IOMMU
group without releasing it. This causes the group to never be released,
with undefined side effects.
The function has only one call-site, which takes a group reference on
its own, so to fix this leak, do not take another reference in
iommu_alloc_default_domain() and pass the group as a function parameter
instead.
Fixes: 6e1aa20491 ("iommu: Move default domain allocation to iommu_probe_device()")
Reported-by: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org>
Cc: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org>
Link: https://lore.kernel.org/r/20200525130122.380-1-joro@8bytes.org
Reference: https://lore.kernel.org/lkml/20200522130145.30067-1-saiprakash.ranjan@codeaurora.org/
The .probe_finalize() call-back of some IOMMU drivers calls into
arm_iommu_attach_device(). This function will call back into the
IOMMU core code, where it tries to take group->mutex again, resulting
in a deadlock.
As there is no reason why .probe_finalize() needs to be called under
that mutex, move it after the lock has been released to fix the
deadlock.
Fixes: deac0b3bed ("iommu: Split off default domain allocation from group assignment")
Reported-by: Yong Wu <yong.wu@mediatek.com>
Tested-by: Yong Wu <yong.wu@mediatek.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Cc: Yong Wu <yong.wu@mediatek.com>
Link: https://lore.kernel.org/r/20200519132824.15163-1-joro@8bytes.org
The IOMMU core code has support for deferring the attachment of a domain
to a device. This is needed in kdump kernels where the new domain must
not be attached to a device before the device driver takes it over.
When the AMD IOMMU driver got converted to use the dma-iommu
implementation, the deferred attaching got lost. The code in
dma-iommu.c has support for deferred attaching, but it calls into
iommu_attach_device() to actually do it. But iommu_attach_device()
will check if the device should be deferred in it code-path and do
nothing, breaking deferred attachment.
Move the is_deferred_attach() check out of the attach_device path and
into iommu_group_add_device() to make deferred attaching work from the
dma-iommu code.
Fixes: 795bbbb9b6 ("iommu/dma-iommu: Handle deferred devices")
Reported-by: Jerry Snitselaar <jsnitsel@redhat.com>
Suggested-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Jerry Snitselaar <jsnitsel@redhat.com>
Cc: Jerry Snitselaar <jsnitsel@redhat.com>
Cc: Tom Murphy <murphyt7@tcd.ie>
Cc: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/20200519130340.14564-1-joro@8bytes.org
After moving iommu_group setup to iommu core code [1][2] and removing
private domain support in vt-d [3], there are no users for functions such
as iommu_request_dm_for_dev(), iommu_request_dma_domain_for_dev() and
request_default_domain_for_dev(). So, remove these functions.
[1] commit dce8d6964e ("iommu/amd: Convert to probe/release_device()
call-backs")
[2] commit e5d1841f18 ("iommu/vt-d: Convert to probe/release_device()
call-backs")
[3] commit 327d5b2fee ("iommu/vt-d: Allow 32bit devices to uses DMA
domain")
Signed-off-by: Sai Praneeth Prakhya <sai.praneeth.prakhya@intel.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20200513224721.20504-1-sai.praneeth.prakhya@intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The host1x bus implemented on Tegra SoCs is primarily an abstraction to
create logical device from multiple platform devices. Since the devices
in such a setup are typically hierarchical, DMA setup still needs to be
done so that DMA masks can be properly inherited, but we don't actually
want to attach the host1x logical devices to any IOMMU. The platform
devices that make up the logical device are responsible for memory bus
transactions, so it is them that will need to be attached to the IOMMU.
Add a check to __iommu_probe_device() that aborts IOMMU setup early for
busses that don't have the IOMMU operations pointer set since they will
cause a crash otherwise.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200511161000.3853342-1-thierry.reding@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The function is now only used in IOMMU core code and shouldn't be used
outside of it anyway, so remove the export for it.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-35-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Move the calls to dev_iommu_get() and try_module_get() into
__iommu_probe_device(), so that the callers don't have to do it on
their own.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-34-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
All drivers are converted to use the probe/release_device()
call-backs, so the add_device/remove_device() pointers are unused and
the code using them can be removed.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-33-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Add a check to the bus_iommu_probe() call-path to make sure it ignores
devices which have already been successfully probed. Then export the
bus_iommu_probe() function so it can be used by IOMMU drivers.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-14-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
After the previous changes the iommu group may not have a default
domain when iommu_group_add_device() is called. With no default domain
iommu_group_create_direct_mappings() will do nothing and no direct
mappings will be created.
Rename iommu_group_create_direct_mappings() to
iommu_create_device_direct_mappings() to better reflect that the
function creates direct mappings only for one device and not for all
devices in the group. Then move the call to the places where a default
domain actually exists.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-13-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
When a bus is initialized with iommu-ops, all devices on the bus are
scanned and iommu-groups are allocated for them, and each groups will
also get a default domain allocated.
Until now this happened as soon as the group was created and the first
device added to it. When other devices with different default domain
requirements were added to the group later on, the default domain was
re-allocated, if possible.
This resulted in some back and forth and unnecessary allocations, so
change the flow to defer default domain allocation until all devices
have been added to their respective IOMMU groups.
The default domains are allocated for newly allocated groups after
each device on the bus is handled and was probed by the IOMMU driver.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-12-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This makes it easier to remove to old code-path when all drivers are
converted. As a side effect that it also fixes the error cleanup
path.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-11-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This is needed to defer default_domain allocation for new IOMMU groups
until all devices have been added to the group.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-10-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Well, not really. The call to iommu_alloc_default_domain() in
iommu_group_get_for_dev() has to stay around as long as there are
IOMMU drivers using the add/remove_device() call-backs instead of
probe/release_device().
Those drivers expect that iommu_group_get_for_dev() returns the device
attached to a group and the group set up with a default domain (and
the device attached to the groups current domain).
But when all drivers are converted this compatability mess can be
removed.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-9-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Add call-backs to 'struct iommu_ops' as an alternative to the
add_device() and remove_device() call-backs, which will be removed when
all drivers are converted.
The new call-backs will not setup IOMMU groups and domains anymore,
so also add a probe_finalize() call-back where the IOMMU driver can do
per-device setup work which require the device to be set up with a
group and a domain.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-8-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Some devices are reqired to use a specific type (identity or dma)
of default domain when they are used with a vendor iommu. When the
system level default domain type is different from it, the vendor
iommu driver has to request a new default domain with
iommu_request_dma_domain_for_dev() and iommu_request_dm_for_dev()
in the add_dev() callback. Unfortunately, these two helpers only
work when the group hasn't been assigned to any other devices,
hence, some vendor iommu driver has to use a private domain if
it fails to request a new default one.
This adds def_domain_type() callback in the iommu_ops, so that
any special requirement of default domain for a device could be
aware by the iommu generic layer.
Signed-off-by: Sai Praneeth Prakhya <sai.praneeth.prakhya@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
[ jroedel@suse.de: Added iommu_get_def_domain_type() function and use
it to allocate the default domain ]
Co-developed-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20200429133712.31431-3-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Joerg Roedel
Douglas Anderson
Lianbo Jiang
Yong Wu
John Garry
Will Deacon
Yang Yingliang
Sai Praneeth Prakhya
Lu Baolu
Shameer Kolothum
Chen Jun
Dan Carpenter
Linus Torvalds
Jacob Pan
Fenghua Yu
Tom Murphy
Baolin Wang
Wei Yongjun
Qian Cai
Jean-Philippe Brucker
Qiushi Wu
Thierry Reding