The bus_set_iommu() in tegra_smmu_probe() enumerates all clients
to call in tegra_smmu_probe_device() where each client searches
its DT node for smmu pointer and swgroup ID, so as to configure
an fwspec. But this requires a valid smmu pointer even before mc
and smmu drivers are probed. So in tegra_smmu_probe() we added a
line of code to fill mc->smmu, marking "a bit of a hack".
This works for most of clients in the DTB, however, doesn't work
for a client that doesn't exist in DTB, a PCI device for example.
Actually, if we return ERR_PTR(-ENODEV) in ->probe_device() when
it's called from bus_set_iommu(), iommu core will let everything
carry on. Then when a client gets probed, of_iommu_configure() in
iommu core will search DTB for swgroup ID and call ->of_xlate()
to prepare an fwspec, similar to tegra_smmu_probe_device() and
tegra_smmu_configure(). Then it'll call tegra_smmu_probe_device()
again, and this time we shall return smmu->iommu pointer properly.
So we can get rid of tegra_smmu_find() and tegra_smmu_configure()
along with DT polling code by letting the iommu core handle every
thing, except a problem that we search iommus property in DTB not
only for swgroup ID but also for mc node to get mc->smmu pointer
to call dev_iommu_priv_set() and return the smmu->iommu pointer.
So we'll need to find another way to get smmu pointer.
Referencing the implementation of sun50i-iommu driver, of_xlate()
has client's dev pointer, mc node and swgroup ID. This means that
we can call dev_iommu_priv_set() in of_xlate() instead, so we can
simply get smmu pointer in ->probe_device().
This patch reworks tegra_smmu_probe_device() by:
1) Removing mc->smmu hack in tegra_smmu_probe() so as to return
ERR_PTR(-ENODEV) in tegra_smmu_probe_device() during stage of
tegra_smmu_probe/tegra_mc_probe().
2) Moving dev_iommu_priv_set() to of_xlate() so we can get smmu
pointer in tegra_smmu_probe_device() to replace DTB polling.
3) Removing tegra_smmu_configure() accordingly since iommu core
takes care of it.
This also fixes a problem that previously we could add clients to
iommu groups before iommu core initializes its default domain:
ubuntu@jetson:~$ dmesg | grep iommu
platform 50000000.host1x: Adding to iommu group 1
platform 57000000.gpu: Adding to iommu group 2
iommu: Default domain type: Translated
platform 54200000.dc: Adding to iommu group 3
platform 54240000.dc: Adding to iommu group 3
platform 54340000.vic: Adding to iommu group 4
Though it works fine with IOMMU_DOMAIN_UNMANAGED, but will have
warnings if switching to IOMMU_DOMAIN_DMA:
iommu: Failed to allocate default IOMMU domain of type 0 for
group (null) - Falling back to IOMMU_DOMAIN_DMA
iommu: Failed to allocate default IOMMU domain of type 0 for
group (null) - Falling back to IOMMU_DOMAIN_DMA
Now, bypassing the first probe_device() call from bus_set_iommu()
fixes the sequence:
ubuntu@jetson:~$ dmesg | grep iommu
iommu: Default domain type: Translated
tegra-host1x 50000000.host1x: Adding to iommu group 0
tegra-dc 54200000.dc: Adding to iommu group 1
tegra-dc 54240000.dc: Adding to iommu group 1
tegra-vic 54340000.vic: Adding to iommu group 2
nouveau 57000000.gpu: Adding to iommu group 3
Note that dmesg log above is testing with IOMMU_DOMAIN_UNMANAGED.
Signed-off-by: Nicolin Chen <nicoleotsuka@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20201125101013.14953-5-nicoleotsuka@gmail.com
Signed-off-by: Will Deacon <will@kernel.org>
In tegra_smmu_(de)attach_dev() functions, we poll DTB for each
client's iommus property to get swgroup ID in order to prepare
"as" and enable smmu. Actually tegra_smmu_configure() prepared
an fwspec for each client, and added to the fwspec all swgroup
IDs of client DT node in DTB.
So this patch uses fwspec in tegra_smmu_(de)attach_dev() so as
to replace the redundant DT polling code.
Signed-off-by: Nicolin Chen <nicoleotsuka@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20201125101013.14953-4-nicoleotsuka@gmail.com
Signed-off-by: Will Deacon <will@kernel.org>
This is used to protect potential race condition at use_count.
since probes of client drivers, calling attach_dev(), may run
concurrently.
Signed-off-by: Nicolin Chen <nicoleotsuka@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20201125101013.14953-3-nicoleotsuka@gmail.com
Signed-off-by: Will Deacon <will@kernel.org>
The tegra_smmu_group_get was added to group devices in different
SWGROUPs and it'd return a NULL group pointer upon a mismatch at
tegra_smmu_find_group(), so for most of clients/devices, it very
likely would mismatch and need a fallback generic_device_group().
But now tegra_smmu_group_get handles devices in same SWGROUP too,
which means that it would allocate a group for every new SWGROUP
or would directly return an existing one upon matching a SWGROUP,
i.e. any device will go through this function.
So possibility of having a NULL group pointer in device_group()
is upon failure of either devm_kzalloc() or iommu_group_alloc().
In either case, calling generic_device_group() no longer makes a
sense. Especially for devm_kzalloc() failing case, it'd cause a
problem if it fails at devm_kzalloc() yet succeeds at a fallback
generic_device_group(), because it does not create a group->list
for other devices to match.
This patch simply unwraps the function to clean it up.
Signed-off-by: Nicolin Chen <nicoleotsuka@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20201125101013.14953-2-nicoleotsuka@gmail.com
Signed-off-by: Will Deacon <will@kernel.org>
There can be clients using the same swgroup in DT, for example i2c0
and i2c1. The current driver will add them to separate IOMMU groups,
though it has implemented device_group() callback which is to group
devices using different swgroups like DC and DCB.
All clients having the same swgroup should be also added to the same
IOMMU group so as to share an asid. Otherwise, the asid register may
get overwritten every time a new device is attached.
Signed-off-by: Nicolin Chen <nicoleotsuka@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200911071643.17212-4-nicoleotsuka@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
IOVA might not be always 4KB aligned. So tegra_smmu_iova_to_phys
function needs to add on the lower 12-bit offset from input iova.
Signed-off-by: Nicolin Chen <nicoleotsuka@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200911071643.17212-3-nicoleotsuka@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
PAGE_SHIFT and PAGE_MASK are defined corresponding to the page size
for CPU virtual addresses, which means PAGE_SHIFT could be a number
other than 12, but tegra-smmu maintains fixed 4KB IOVA pages and has
fixed [21:12] bit range for PTE entries.
So this patch replaces all PAGE_SHIFT/PAGE_MASK references with the
macros defined with SMMU_PTE_SHIFT.
Signed-off-by: Nicolin Chen <nicoleotsuka@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200911071643.17212-2-nicoleotsuka@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The "num_tlb_lines" might not be a power-of-2 value, being 48 on
Tegra210 for example. So the current way of calculating tlb_mask
using the num_tlb_lines is not correct: tlb_mask=0x5f in case of
num_tlb_lines=48, which will trim a setting of 0x30 (48) to 0x10.
Signed-off-by: Nicolin Chen <nicoleotsuka@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200917113155.13438-2-nicoleotsuka@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The mapping operations of the Tegra SMMU driver are subjected to a race
condition issues because SMMU Address Space isn't allocated and freed
atomically, while it should be. This patch makes the mapping operations
atomic, it fixes an accidentally released Host1x Address Space problem
which happens while running multiple graphics tests in parallel on
Tegra30, i.e. by having multiple threads racing with each other in the
Host1x's submission and completion code paths, performing IOVA mappings
and unmappings in parallel.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Thierry Reding <treding@nvidia.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200901203730.27865-1-digetx@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
In order to share groups between multiple devices we keep track of them
in a per-SMMU list. When an IOMMU group is released, a dangling pointer
to it stays around in that list. Fix this by implementing an IOMMU data
release callback for groups where the dangling pointer can be removed.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200806155404.3936074-4-thierry.reding@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
For groups that are shared between multiple devices, care must be taken
to acquire a reference for each device, otherwise the IOMMU core ends up
dropping the last reference too early, which will cause the group to be
released while consumers may still be thinking that they're holding a
reference to it.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200806155404.3936074-3-thierry.reding@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Remove the use of dev->archdata.iommu and use the private per-device
pointer provided by IOMMU core code instead.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Reviewed-by: Jerry Snitselaar <jsnitsel@redhat.com>
Link: https://lore.kernel.org/r/20200625130836.1916-7-joro@8bytes.org
Convert the Tegra IOMMU drivers to use the probe_device() and
release_device() call-backs of iommu_ops, so that the iommu core code
does the group and sysfs setup.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20200429133712.31431-27-joro@8bytes.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Page tables that reside in physical memory beyond the 4 GiB boundary are
currently not working properly. The reason is that when the physical
address for page directory entries is read, it gets truncated at 32 bits
and can cause crashes when passing that address to the DMA API.
Fix this by first casting the PDE value to a dma_addr_t and then using
the page frame number mask for the SMMU instance to mask out the invalid
bits, which are typically used for mapping attributes, etc.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Use PTB_ASID instead of SMMU_CONFIG to flush smmu.
PTB_ASID can be accessed from non-secure mode, SMMU_CONFIG cannot be.
Using SMMU_CONFIG could pose a problem when kernel doesn't have secure
mode access enabled from boot.
Signed-off-by: Navneet Kumar <navneetk@nvidia.com>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Add a gfp_t parameter to the iommu_ops::map function.
Remove the needless locking in the AMD iommu driver.
The iommu_ops::map function (or the iommu_map function which calls it)
was always supposed to be sleepable (according to Joerg's comment in
this thread: https://lore.kernel.org/patchwork/patch/977520/ ) and so
should probably have had a "might_sleep()" since it was written. However
currently the dma-iommu api can call iommu_map in an atomic context,
which it shouldn't do. This doesn't cause any problems because any iommu
driver which uses the dma-iommu api uses gfp_atomic in it's
iommu_ops::map function. But doing this wastes the memory allocators
atomic pools.
Signed-off-by: Tom Murphy <murphyt7@tcd.ie>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
To allow IOMMU drivers to batch up TLB flushing operations and postpone
them until ->iotlb_sync() is called, extend the prototypes for the
->unmap() and ->iotlb_sync() IOMMU ops callbacks to take a pointer to
the current iommu_iotlb_gather structure.
All affected IOMMU drivers are updated, but there should be no
functional change since the extra parameter is ignored for now.
Signed-off-by: Will Deacon <will@kernel.org>
Based on 2 normalized pattern(s):
this program is free software you can redistribute it and or modify
it under the terms of the gnu general public license version 2 as
published by the free software foundation
this program is free software you can redistribute it and or modify
it under the terms of the gnu general public license version 2 as
published by the free software foundation #
extracted by the scancode license scanner the SPDX license identifier
GPL-2.0-only
has been chosen to replace the boilerplate/reference in 4122 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Enrico Weigelt <info@metux.net>
Reviewed-by: Kate Stewart <kstewart@linuxfoundation.org>
Reviewed-by: Allison Randal <allison@lohutok.net>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190604081206.933168790@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Set PTE read/write attributes accordingly to the the protections requested
by IOMMU API.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Release all memory allocations associated with a released domain and emit
warning if domain is in-use at the time of destruction.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Both Tegra30 and Tegra114 have 4 ASID's and the corresponding bitfield of
the TLB_FLUSH register differs from later Tegra generations that have 128
ASID's.
In a result the PTE's are now flushed correctly from TLB and this fixes
problems with graphics (randomly failing tests) on Tegra30.
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Tegra20 doesn't have SMMU. Move out checking of the SMMU presence from
the SMMU driver into the Memory Controller driver. This change makes code
consistent in regards to how GART/SMMU presence checking is performed.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Use the new helpers dev_iommu_fwspec_get()/set() to access
the dev->iommu_fwspec pointer. This makes it easier to move
that pointer later into another struct.
Cc: Thierry Reding <thierry.reding@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Use DEFINE_SHOW_ATTRIBUTE macro to simplify the code.
Signed-off-by: Yangtao Li <tiny.windzz@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
All iommu drivers use the default_iommu_map_sg implementation, and there
is no good reason to ever override it. Just expose it as iommu_map_sg
directly and remove the indirection, specially in our post-spectre world
where indirect calls are horribly expensive.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
In case of error, the function iommu_group_alloc() returns ERR_PTR() and
never returns NULL. The NULL test in the return value check should be
replaced with IS_ERR().
Fixes: 7f4c9176f7 ("iommu/tegra: Allow devices to be grouped")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Implement the ->device_group() and ->of_xlate() callbacks which are used
in order to group devices. Each group can then share a single domain.
This is implemented primarily in order to achieve the same semantics on
Tegra210 and earlier as on Tegra186 where the Tegra SMMU was replaced by
an ARM SMMU. Users of the IOMMU API can now use the same code to share
domains between devices, whereas previously they used to attach each
device individually.
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
The bus_set_iommu() function will call the add_device()
call-back which needs the iommu to be registered.
Reported-by: Jon Hunter <jonathanh@nvidia.com>
Fixes: 0b480e4470 ('iommu/tegra: Add support for struct iommu_device')
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Add a struct iommu_device to each tegra-smmu and register it
with the iommu-core. Also link devices added to the driver
to their respective hardware iommus.
Acked-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
As the last step to making groups mandatory, clean up the remaining
drivers by adding basic support. Whilst it may not perfectly reflect
the isolation capabilities of the hardware (tegra_smmu_swgroup sounds
suspiciously like something that might warrant representing at the
iommu_group level), using generic_device_group() should at least
maintain existing behaviour with respect to the API.
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Tested-by: Mikko Perttunen <mperttunen@nvidia.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The include file does not need any PCI specifics, so remove
that include. Also fix the places that relied on it.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The number of TLB lines was increased from 16 on Tegra30 to 32 on
Tegra114 and later. Parameterize the value so that the initial default
can be set accordingly.
On Tegra30, initializing the value to 32 would effectively disable the
TLB and hence cause massive latencies for memory accesses translated
through the SMMU. This is especially noticeable for isochronuous clients
such as display, whose FIFOs would continuously underrun.
Fixes: 8918465163 ("memory: Add NVIDIA Tegra memory controller support")
Signed-off-by: Thierry Reding <treding@nvidia.com>
This code is used both when creating a new page directory entry and when
tearing it down, with only the PDE value changing between both cases.
Factor the code out so that it can be reused.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
[treding@nvidia.com: make commit message more accurate]
Signed-off-by: Thierry Reding <treding@nvidia.com>
Extract the use count reference accounting into a separate function and
separate it from allocating the PTE.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
[treding@nvidia.com: extract and write commit message]
Signed-off-by: Thierry Reding <treding@nvidia.com>
Rather than explicitly zeroing pages allocated via alloc_page(), add
__GFP_ZERO to the gfp mask to ask the allocator for zeroed pages.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Remove the unnecessary manipulation of the PageReserved flags in the
Tegra SMMU driver. None of this is required as the page(s) remain
private to the SMMU driver.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Use the DMA API instead of calling architecture internal functions in
the Tegra SMMU driver.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Pass smmu_flush_ptc() the device address rather than struct page
pointer.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
smmu_flush_ptc() is used in two modes: one is to flush an individual
entry, the other is to flush all entries. We know at the call site
which we require. Split the function into smmu_flush_ptc_all() and
smmu_flush_ptc().
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Drivers should not be using __cpuc_* functions nor outer_cache_flush()
directly. This change partly cleans up tegra-smmu.c.
The only difference between cache handling of the tegra variants is
Denver, which omits the call to outer_cache_flush(). This is due to
Denver being an ARM64 CPU, and the ARM64 architecture does not provide
this function. (This, in itself, is a good reason why these should not
be used.)
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
[treding@nvidia.com: fix build failure on 64-bit ARM]
Signed-off-by: Thierry Reding <treding@nvidia.com>
Use kcalloc() to allocate the use-counter array for the page directory
entries/page tables. Using kcalloc() allows us to be provided with
zero-initialised memory from the allocators, rather than initialising
it ourselves.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Store the struct page pointer for the second level page tables, rather
than working back from the page directory entry. This is necessary as
we want to eliminate the use of physical addresses used with
arch-private functions, switching instead to use the streaming DMA API.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Fix the page table lookup in the unmap and iova_to_phys methods.
Neither of these methods should allocate a page table; a missing page
table should be treated the same as no mapping present.
More importantly, using as_get_pte() for an IOVA corresponding with a
non-present page table entry increments the use-count for the page
table, on the assumption that the caller of as_get_pte() is going to
setup a mapping. This is an incorrect assumption.
Fix both of these bugs by providing a separate helper which only looks
up the page table, but never allocates it. This is akin to pte_offset()
for CPU page tables.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Add a pair of helpers to get the page directory and page table indexes.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Factor out the common PTE setting code into a separate function.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Nicolin Chen