This partially reverts the DMA API support that was recently merged
because it was causing performance regressions on older Tegra devices.
Unfortunately, the cache maintenance performed by dma_map_sg() and
dma_unmap_sg() causes performance to drop by a factor of 10.
The right solution for this would be to cache mappings for buffers per
consumer device, but that's a bit involved. Instead, we simply revert to
the old behaviour of sharing IOVA mappings when we know that devices can
do so (i.e. they share the same IOMMU domain).
Cc: <stable@vger.kernel.org> # v5.5
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
drm-misc-next for v5.6:
UAPI Changes:
- Add support for DMA-BUF HEAPS.
Cross-subsystem Changes:
- mipi dsi definition updates, pulled into drm-intel as well.
- Add lockdep annotations for dma_resv vs mmap_sem and fs_reclaim.
- Remove support for dma-buf kmap/kunmap.
- Constify fb_ops in all fbdev drivers, including drm drivers and drm-core, and media as well.
Core Changes:
- Small cleanups to ttm.
- Fix SCDC definition.
- Assorted cleanups to core.
- Add todo to remove load/unload hooks, and use generic fbdev emulation.
- Assorted documentation updates.
- Use blocking ww lock in ttm fault handler.
- Remove drm_fb_helper_fbdev_setup/teardown.
- Warning fixes with W=1 for atomic.
- Use drm_debug_enabled() instead of drm_debug flag testing in various drivers.
- Fallback to nontiled mode in fbdev emulation when not all tiles are present. (Later on reverted)
- Various kconfig indentation fixes in core and drivers.
- Fix freeing transactions in dp-mst correctly.
- Sean Paul is steping down as core maintainer. :-(
- Add lockdep annotations for atomic locks vs dma-resv.
- Prevent use-after-free for a bad job in drm_scheduler.
- Fill out all block sizes in the P01x and P210 definitions.
- Avoid division by zero in drm/rect, and fix bounds.
- Add drm/rect selftests.
- Add aspect ratio and alternate clocks for HDMI 4k modes.
- Add todo for drm_framebuffer_funcs and fb_create cleanup.
- Drop DRM_AUTH for prime import/export ioctls.
- Clear DP-MST payload id tables downstream when initializating.
- Fix for DSC throughput definition.
- Add extra FEC definitions.
- Fix fake offset in drm_gem_object_funs.mmap.
- Stop using encoder->bridge in core directly
- Handle bridge chaining slightly better.
- Add backlight support to drm/panel, and use it in many panel drivers.
- Increase max number of y420 modes from 128 to 256, as preparation to add the new modes.
Driver Changes:
- Small fixes all over.
- Fix documentation in vkms.
- Fix mmap_sem vs dma_resv in nouveau.
- Small cleanup in komeda.
- Add page flip support in gma500 for psb/cdv.
- Add ddc symlink in the connector sysfs directory for many drivers.
- Add support for analogic an6345, and fix small bugs in it.
- Add atomic modesetting support to ast.
- Fix radeon fault handler VMA race.
- Switch udl to use generic shmem helpers.
- Unconditional vblank handling for mcde.
- Miscellaneous fixes to mcde.
- Tweak debug output from komeda using debugfs.
- Add gamma and color transform support to komeda for DOU-IPS.
- Add support for sony acx424AKP panel.
- Various small cleanups to gma500.
- Use generic fbdev emulation in udl, and replace udl_framebuffer with generic implementation.
- Add support for Logic PD Type 28 panel.
- Use drm_panel_* wrapper functions in exynos/tegra/msm.
- Add devicetree bindings for generic DSI panels.
- Don't include drm_pci.h directly in many drivers.
- Add support for begin/end_cpu_access in udmabuf.
- Stop using drm_get_pci_dev in gma500 and mga200.
- Fixes to UDL damage handling, and use dma_buf_begin/end_cpu_access.
- Add devfreq thermal support to panfrost.
- Fix hotplug with daisy chained monitors by removing VCPI when disabling topology manager.
- meson: Add support for OSD1 plane AFBC commit.
- Stop displaying garbage when toggling ast primary plane on/off.
- More cleanups and fixes to UDL.
- Add D32 suport to komeda.
- Remove globle copy of drm_dev in gma500.
- Add support for Boe Himax8279d MIPI-DSI LCD panel.
- Add support for ingenic JZ4770 panel.
- Small null pointer deference fix in ingenic.
- Remove support for the special tfp420 driver, as there is a generic way to do it.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/ba73535a-9334-5302-2e1f-5208bd7390bd@linux.intel.com
All the display related blocks on Tegra require contiguous memory. Using
the DMA API, there is no knowing at import time which device will end up
using the buffer, so it's not known whether or not an IOMMU will be used
to map the buffer.
Move the check for non-contiguous buffers/mappings to the tegra_dc_pin()
function which is now the earliest point where it is known if a DMA BUF
can be used by the given device or not.
v2: add check for contiguous buffer/mapping in tegra_dc_pin()
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Buffers that are imported from a DMA-BUF don't have pages allocated with
them. At the same time an SG table for them can't be derived using the
DMA API helpers because the necessary information doesn't exist. However
there's already an SG table that was created during import, so this can
simply be duplicated.
Signed-off-by: Thierry Reding <treding@nvidia.com>
If host1x_bo_pin() returns an SG table, create a DMA mapping for the
buffer. For buffers that the host1x client has already mapped itself,
host1x_bo_pin() returns NULL and the existing DMA address is used.
Signed-off-by: Thierry Reding <treding@nvidia.com>
The host1x_bo_pin() and host1x_bo_unpin() APIs are used to pin and unpin
buffers during host1x job submission. Pinning currently returns the SG
table and the DMA address (an IOVA if an IOMMU is used or a physical
address if no IOMMU is used) of the buffer. The DMA address is only used
for buffers that are relocated, whereas the host1x driver will map
gather buffers into its own IOVA space so that they can be processed by
the CDMA engine.
This approach has a couple of issues. On one hand it's not very useful
to return a DMA address for the buffer if host1x doesn't need it. On the
other hand, returning the SG table of the buffer is suboptimal because a
single SG table cannot be shared for multiple mappings, because the DMA
address is stored within the SG table, and the DMA address may be
different for different devices.
Subsequent patches will move the host1x driver over to the DMA API which
doesn't work with a single shared SG table. Fix this by returning a new
SG table each time a buffer is pinned. This allows the buffer to be
referenced by multiple jobs for different engines.
Change the prototypes of host1x_bo_pin() and host1x_bo_unpin() to take a
struct device *, specifying the device for which the buffer should be
pinned. This is required in order to be able to properly construct the
SG table. While at it, make host1x_bo_pin() return the SG table because
that allows us to return an ERR_PTR()-encoded error code if we need to,
or return NULL to signal that we don't need the SG table to be remapped
and can simply use the DMA address as-is. At the same time, returning
the DMA address is made optional because in the example of command
buffers, host1x doesn't need to know the DMA address since it will have
to create its own mapping anyway.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Instead of manually creating the SG table for a discontiguous buffer,
use the existing sg_alloc_table_from_pages(). Note that this is not safe
to be used with the ARM DMA/IOMMU integration code because that will not
ensure that the whole buffer is mapped contiguously. Depending on the
size of the individual entries the mapping may end up containing holes
to ensure alignment.
However, we only ever use these buffers with explicit IOMMU API usage
and know how to avoid these holes.
Signed-off-by: Thierry Reding <treding@nvidia.com>
When an importer wants to map a DMA-BUF, make sure to always actually
map it, irrespective of whether the buffer is contiguous or not.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Rather than manually creating an SG table in an incorrect way, let the
standard dma_get_sgtable() function do it.
Signed-off-by: Thierry Reding <treding@nvidia.com>
The address can refer to either physical memory or IO virtual memory.
If referring to IO virtual memory, there will always be an associated
physical memory address. Rename this variable to "iova" to clarify in
all cases that this is the IO virtual memory, which in the absence of
an IOMMU is identical to the physical address.
Signed-off-by: Thierry Reding <treding@nvidia.com>
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>
The allocated pages need to be invalidated in CPU caches. On ARM32 the
DMA_BIDIRECTIONAL flag only ensures that data is written-back to DRAM and
the data stays in CPU cache lines. While the DMA_FROM_DEVICE flag ensures
that the corresponding CPU cache lines are getting invalidated and nothing
more, that's exactly what is needed for a newly allocated pages.
This fixes randomly failing rendercheck tests on Tegra30 using the
Opentegra driver for tests that use small-sized pixmaps (10x10 and less,
i.e. 1-2 memory pages) because apparently CPU reads out stale data from
caches and/or that data is getting evicted to DRAM at the time of HW job
execution.
Fixes: bd43c9f0fa ("drm/tegra: gem: Map pages via the DMA API")
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
We got a few conflicts in drm_atomic.c after merging the DRM writeback support,
now we need a backmerge to unlock develop development on drm-misc-next.
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.com>
Set the owner and name of the exported DMA-BUF in addition to the
already filled-in fields.
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Use new return type vm_fault_t for fault handler. For now, this is just
documenting that the function returns a VM_FAULT value rather than an
errno. Once all instances are converted, vm_fault_t will become a
distinct type.
Reference id -> 1c8f422059 ("mm: change return type to vm_fault_t")
Previously vm_insert_page() returns err which driver mapped into
VM_FAULT_* type. The new function vmf_insert_page() will replace this
inefficiency by returning VM_FAULT_* type.
Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
These callbacks allow the exporter to swap in and pin the backing
storage for buffers as well as invalidate the cache in preparation for
accessing the buffer from the CPU, and flush the cache and unpin the
backing storage when the CPU is done modifying the buffer.
Signed-off-by: Thierry Reding <treding@nvidia.com>
When allocating pages, map them with the DMA API in order to invalidate
caches. This is the correct usage of the API and works just as well as
faking up the SG table and using the dma_sync_sg_for_device() function.
Signed-off-by: Thierry Reding <treding@nvidia.com>
This function allows mapping a GEM object into a virtual memory address
space, which makes it useful outside of the GEM code.
While at it, rename the function so it doesn't clash with the function
that implements the DRM_TEGRA_GEM_MMAP IOCTL.
Signed-off-by: Thierry Reding <treding@nvidia.com>
iommu_map_sg() doesn't return a error value, but a size of the requested
IOMMU mapping or zero in case of error.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
drm/tegra: Changes for v4.14-rc1
This contains a couple of fixes and improvements for host1x, with some
preparatory work for Tegra186 support.
The remainder is cleanup and minor bugfixes for Tegra DRM along with
enhancements to debuggability.
There have also been some enhancements to the kernel interfaces for
host1x job submissions and support for mmap'ing PRIME buffers directly,
all of which get the interfaces very close to ready for serious work.
* tag 'drm/tegra/for-4.14-rc1' of git://anongit.freedesktop.org/tegra/linux: (21 commits)
drm/tegra: Prevent BOs from being freed during job submission
drm/tegra: gem: Implement mmap() for PRIME buffers
drm/tegra: Support render node
drm/tegra: sor: Trace register accesses
drm/tegra: dpaux: Trace register accesses
drm/tegra: dsi: Trace register accesses
drm/tegra: hdmi: Trace register accesses
drm/tegra: dc: Trace register accesses
drm/tegra: sor: Use unsigned int for register offsets
drm/tegra: hdmi: Use unsigned int for register offsets
drm/tegra: dsi: Use unsigned int for register offsets
drm/tegra: dpaux: Use unsigned int for register offsets
drm/tegra: dc: Use unsigned int for register offsets
drm/tegra: Fix NULL deref in debugfs/iova
drm/tegra: switch to drm_*_get(), drm_*_put() helpers
drm/tegra: Set MODULE_FIRMWARE for the VIC
drm/tegra: Add CONFIG_OF dependency
gpu: host1x: Support sub-devices recursively
gpu: host1x: fix error return code in host1x_probe()
gpu: host1x: Fix bitshift/mask multipliers
...
The mapping of PRIME buffers can reuse much of the GEM mapping code, so
extract the common bits into a new tegra_gem_mmap() helper.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Use drm_*_get() and drm_*_put() helpers instead of drm_*_reference()
and drm_*_unreference() helpers.
drm_*_reference() and drm_*_unreference() functions are just
compatibility alias for drm_*_get() and drm_*_put() and should not be
used by new code. So convert all users of compatibility functions to
use the new APIs.
Generated by: scripts/coccinelle/api/drm-get-put.cocci
Signed-off-by: Cihangir Akturk <cakturk@gmail.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
If commands buffer claims a number of words that is higher than its BO can
fit, a kernel OOPS will be fired on the out-of-bounds BO access. This was
triggered by an opentegra Xorg driver that erroneously pushed too many
commands to the pushbuf.
The CDMA commands buffer address is 4 bytes aligned, so check its
alignment.
The maximum number of the CDMA gather fetches is 16383, add a check for
it.
Add a sanity check for the relocations in a same way.
[ 46.829393] Unable to handle kernel paging request at virtual address f09b2000
...
[<c04a3ba4>] (host1x_job_pin) from [<c04dfcd0>] (tegra_drm_submit+0x474/0x510)
[<c04dfcd0>] (tegra_drm_submit) from [<c04deea0>] (tegra_submit+0x50/0x6c)
[<c04deea0>] (tegra_submit) from [<c04c07c0>] (drm_ioctl+0x1e4/0x3ec)
[<c04c07c0>] (drm_ioctl) from [<c02541a0>] (do_vfs_ioctl+0x9c/0x8e4)
[<c02541a0>] (do_vfs_ioctl) from [<c0254a1c>] (SyS_ioctl+0x34/0x5c)
[<c0254a1c>] (SyS_ioctl) from [<c0107640>] (ret_fast_syscall+0x0/0x3c)
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Reviewed-by: Erik Faye-Lund <kusmabite@gmail.com>
Reviewed-by: Mikko Perttunen <mperttunen@nvidia.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
drm/tegra: Changes for v4.12-rc1
This contains various fixes to the host1x driver as well as a plug for a
leak of kernel pointers to userspace.
A fairly big addition this time around is the Video Image Composer (VIC)
support that can be used to accelerate some 2D and image compositing
operations.
Furthermore the driver now supports FB modifiers, so we no longer rely
on a custom IOCTL to set those.
Finally this contains a few preparatory patches for Tegra186 support
which unfortunately didn't quite make it this time, but will hopefully
be ready for v4.13.
* tag 'drm/tegra/for-4.12-rc1' of git://anongit.freedesktop.org/tegra/linux:
gpu: host1x: Fix host1x driver shutdown
gpu: host1x: Support module reset
gpu: host1x: Sort includes alphabetically
drm/tegra: Add VIC support
dt-bindings: Add bindings for the Tegra VIC
drm/tegra: Add falcon helper library
drm/tegra: Add Tegra DRM allocation API
drm/tegra: Add tiling FB modifiers
drm/tegra: Don't leak kernel pointer to userspace
drm/tegra: Protect IOMMU operations by mutex
drm/tegra: Enable IOVA API when IOMMU support is enabled
gpu: host1x: Add IOMMU support
gpu: host1x: Fix potential out-of-bounds access
iommu/iova: Fix compile error with CONFIG_IOMMU_IOVA=m
iommu: Add dummy implementations for !IOMMU_IOVA
MAINTAINERS: Add related headers to IOMMU section
iommu/iova: Consolidate code for adding new node to iovad domain rbtree
IOMMU support is currently not thread-safe, which can cause crashes,
amongst other things, under certain workloads.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Fix tegra_bo_pin() to set the parameter sgt pointer. host1x job pinning
requires the sgt to determine physical memory addresses of gathers.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
host1x command buffer patching requires that the buffer object can be
mapped into kernel address space. However, the recent addition of IOMMU
support did not account for this requirement. Therefore host1x engines
cannot be used if IOMMU is enabled.
This patch implements kmap, kunmap, mmap and munmap functions to host1x
buffer objects.
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
dma_buf_map_attachment() never returns NULL, so there is no need to
check for it.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Thierry Reding <treding@nvidia.com>
This only grabs the mutex when really needed, but still has a might-
acquire lockdep check to make sure that's always possible. With this
patch Tegra DRM is officially struct_mutex free, yay!
v2: refernce_unlocked doesn't exist as kbuild spotted.
Cc: Thierry Reding <thierry.reding@gmail.com>
Acked-by: Thierry Reding <thierry.reding@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
[treding@nvidia.com: remove unused variables]
Signed-off-by: Thierry Reding <treding@nvidia.com>
Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).
v2: Finally get rid of the copypasta from another commit in this
commit message. And convert to _unlocked like we need to (Patrik).
Cc: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>
Cc: Thierry Reding <thierry.reding@gmail.com>
Acked-by: Thierry Reding <thierry.reding@gmail.com>
Reviewed-by: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Pages allocated from shmemfs don't end up being cleared and flushed on
ARMv7, so they must be flushed explicitly. Use the DMA mapping API for
that purpose, even though it's not used for anything else.
Signed-off-by: Thierry Reding <treding@nvidia.com>
At present, dma_buf_export() takes a series of parameters, which
makes it difficult to add any new parameters for exporters, if required.
Make it simpler by moving all these parameters into a struct, and pass
the struct * as parameter to dma_buf_export().
While at it, unite dma_buf_export_named() with dma_buf_export(), and
change all callers accordingly.
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Reviewed-by: Daniel Thompson <daniel.thompson@linaro.org>
Acked-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
Acked-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Sumit Semwal <sumit.semwal@linaro.org>
If the requested buffer size wasn't a multiple of the page size, the
IOMMU code would round down the size to the next multiple of the page
size, thereby causing translation errors. To fix this we no longer pass
around the requested size but reuse the computed size of the GEM object.
This is already rounded to the next page boundary, so mapping that size
works out fine.
Signed-off-by: Thierry Reding <treding@nvidia.com>
