As the name says global memory and bo accounting is global. So it doesn't
make to much sense having pointers to global structures all around the code.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NThomas Hellström <thellstrom@vmware.com>
Link: https://patchwork.freedesktop.org/patch/332879/

This allows blocking for BOs to become available
in the memory management.

Amdgpu is doing this for quite a while now during CS. Now
apply the new behavior to all drivers using TTM.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NThomas Hellstrom <thellstrom@vmware.com>
Link: https://patchwork.freedesktop.org/patch/332878/

This feature is only used by vmwgfx and superfluous for everybody else.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Co-developed-by: NThomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: NThomas Hellstrom <thellstrom@vmware.com>
Tested-by: NThomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Link: https://patchwork.freedesktop.org/patch/333650/

Rename ttm_fbdev_mmap to ttm_bo_mmap_obj.  Move the vm_pgoff sanity
check to amdgpu_bo_fbdev_mmap (only ttm_fbdev_mmap user in tree).

The ttm_bo_mmap_obj function can now be used to map any buffer object.
This allows to implement &drm_gem_object_funcs.mmap in gem ttm helpers.

v3: patch added to series
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Acked-by: NThomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Acked-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20191016115203.20095-8-kraxel@redhat.com

Factor out ttm vma setup to a new function.
Reduces code duplication a bit.

v2: don't change vm_flags (moved to separate patch).
v4: make ttm_bo_mmap_vma_setup static.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Acked-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20191016115203.20095-7-kraxel@redhat.com

We should not add the BO to the swap LRU when the new mem is fixed and
the TTM object about to be destroyed.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NKevin Wang <kevin1.wang@amd.com>
Link: https://patchwork.freedesktop.org/patch/335246/

Commit 4daa4fba ("gpu: drm: ttm: Adding new return type vm_fault_t")
broke TTM prefaulting. Since vmf_insert_mixed() typically always returns
VM_FAULT_NOPAGE, prefaulting stops after the second PTE.

Restore (almost) the original behaviour. Unfortunately we can no longer
with the new vm_fault_t return type determine whether a prefaulting
PTE insertion hit an already populated PTE, and terminate the insertion
loop. Instead we continue with the pre-determined number of prefaults.

Fixes: 4daa4fba ("gpu: drm: ttm: Adding new return type vm_fault_t")
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: NThomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org # v4.19+
Signed-off-by: NChristian König <christian.koenig@amd.com>
Link: https://patchwork.freedesktop.org/patch/330387/

The busy BO might actually be already deleted,
so grab only a list reference.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NThomas Hellström <thellstrom@vmware.com>
Link: https://patchwork.freedesktop.org/patch/332877/

No users left.  Drivers either setup vma_offset_manager themself
(vmwgfx) or pass the gem vma_offset_manager to ttm_bo_device_init
(all other drivers).
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190905070509.22407-9-kraxel@redhat.com

Rename the embedded struct vma_offset_manager, new name is _vma_manager.
ttm_bo_device.vma_manager changed to a pointer.

The ttm_bo_device_init() function gets an additional vma_manager
argument which allows to initialize ttm with a different vma manager.
When passing NULL the embedded _vma_manager is used.

All callers are updated to pass NULL, so the behavior doesn't change.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190905070509.22407-2-kraxel@redhat.com

Be more consistent with the naming of the other DMA-buf objects.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/323401/

All users moved to ttm_buffer_object->base.resv
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190805140119.7337-18-kraxel@redhat.com

Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190805140119.7337-11-kraxel@redhat.com

Initialize both ttm_buffer_object->resv and ttm_buffer_object->base.resv
pointers.  This allows to move users from the former to the latter.  When
all users are moved we can drop ttm_buffer_object->resv.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190805140119.7337-10-kraxel@redhat.com

Drop vma_node from ttm_buffer_object, use the gem struct
(base.vma_node) instead.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190805140119.7337-9-kraxel@redhat.com

Drop ttm_resv from ttm_buffer_object, use the gem reservation object
(base._resv) instead.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190805140119.7337-8-kraxel@redhat.com

Complete the abstraction of the ww_mutex inside the reservation object.

This allows us to add more handling and debugging to the reservation
object in the future.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/320761/

This notifies the driver that a BO is about to be released.

Releasing a BO also invokes the move_notify callback from
ttm_bo_cleanup_memtype_use, but that happens too late for anything
that would add fences to the BO and require a delayed delete.
Signed-off-by: NFelix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

In function __ttm_dma_alloc_page(), d_page->addr is allocated
by dma_alloc_attrs() but freed with use dma_free_coherent() in
__ttm_dma_free_page().
Use the correct dma_free_attrs() to free d_page->vaddr.
Signed-off-by: NFuqian Huang <huangfq.daxian@gmail.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

In function __ttm_dma_alloc_page(), d_page->addr is allocated
by dma_alloc_attrs() but freed with use dma_free_coherent() in
__ttm_dma_free_page().
Use the correct dma_free_attrs() to free d_page->vaddr.
Signed-off-by: NFuqian Huang <huangfq.daxian@gmail.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This reverts commit 031e610a, reversing
changes made to 52d2d44e.

The mm changes in there we premature and not fully ack or reviewed by core mm folks,
I dropped the ball by merging them via this tree, so lets take em all back out.
Signed-off-by: NDave Airlie <airlied@redhat.com>

Returning -EAGAIN prevents ttm_bo_mem_space from trying alternate
placements and can lead to live-locks in amdgpu_cs, retrying
indefinitely and never succeeding.

Fixes: d367bd2a ("drm/ttm: fix busy memory to fail other user v10")
Signed-off-by: NFelix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Since commit 4b050ba7 ("MIPS: pgtable.h: Implement the
pgprot_writecombine function for MIPS") and commit c4687b15 ("MIPS: Fix
definition of pgprot_writecombine()") write-combine vma mapping is
available to be used by kernel subsystems for MIPS. In particular the
uncached accelerated attribute is requested to be set by ioremap_wc()
method and by generic PCI memory pages/ranges mapping methods. The same
is done by the drm_io_prot()/ttm_io_prot() functions in case if
write-combine flag is set for vma's passed for mapping. But for some
reason the pgprot_writecombine() method calling is ifdefed to be a
platform-specific with MIPS system being marked as lacking of one. At the
very least it doesn't reflect the current MIPS platform implementation.
So in order to improve the DRM subsystem performance on MIPS with UCA
mapping enabled, we need to have pgprot_writecombine() called for buffers,
which need store operations being combined. In case if particular MIPS
chip doesn't support the UCA attribute, the mapping will fall back to
noncached.

Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paul Burton <paul.burton@mips.com>
Cc: James Hogan <jhogan@kernel.org>
Acked-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NPaul Burton <paul.burton@mips.com>
Signed-off-by: NVadim V. Vlasov <vadim.vlasov@t-platforms.ru>
Signed-off-by: NSerge Semin <fancer.lancer@gmail.com>
Signed-off-by: NSean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20190423123122.32573-1-fancer.lancer@gmail.com

With the vmwgfx dirty tracking, the default TTM fault handler is not
completely sufficient (vmwgfx need to modify the vma->vm_flags member,
and also needs to restrict the number of prefaults).

We also want to replicate the new ttm_bo_vm_reserve() functionality

So start turning the TTM vm code into helpers: ttm_bo_vm_fault_reserved()
and ttm_bo_vm_reserve(), and provide a default TTM fault handler for other
drivers to use.

Cc: "Christian König" <christian.koenig@amd.com>
Signed-off-by: NThomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: "Christian König" <christian.koenig@amd.com> #v1

Add a pointer to the struct vm_operations_struct in the bo_device, and
assign that pointer to the default value currently used.

The driver can then optionally modify that pointer and the new value
can be used for each new vma created.

Cc: "Christian König" <christian.koenig@amd.com>
Signed-off-by: NThomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>

BOs on the LRU might be blocked during command submission
and cause OOM situations.

Avoid this by blocking for the first busy BO not locked by
the same ticket as the BO we are searching space for.

v10: completely start over with the patch since we didn't
     handled a whole bunch of corner cases.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NChunming Zhou <david1.zhou@amd.com>
Tested-by: NPierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Move BOs which are currently in a lower domain to the new
LRU before allocating backing space while validating.

This makes sure that we always have enough entries on the
LRU to allow for other processes to wait for an operation
to complete.

v2: generalize the test
v3: fix rebase error
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NChunming Zhou <david1.zhou@amd.com>
Tested-by: NPierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We tried this once before, but that turned out to be more
complicated than thought. With all the right prerequisites
it looks like we can do this now.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NChunming Zhou <david1.zhou@amd.com>
Tested-by: NPierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

If drivers don't prefer a system memory placement
they should not but it into the placement list first.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NChunming Zhou <david1.zhou@amd.com>
Tested-by: NPierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

When a signal arrives we should return immediately for
handling it and not try other placements first.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NChunming Zhou <david1.zhou@amd.com>
Tested-by: NPierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We are already doing this for DMA-buf imports and also for
amdgpu VM BOs for quite a while now.

If this doesn't run into any problems we are probably going
to stop removing BOs from the LRU altogether.

v2: drop BUG_ON from ttm_bo_add_to_lru
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NChunming Zhou <david1.zhou@amd.com>
Tested-by: NPierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

When a driver unloads without unloading TTM we don't correctly
clear the global structures leading to errors on re-init.

Next step should probably be to remove the global structures and
kobjs all together, but this is tricky since we need to maintain
backward compatibility.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NKarol Herbst <kherbst@redhat.com>
Tested-by: NKarol Herbst <kherbst@redhat.com>
CC: stable@vger.kernel.org # 5.0.x
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

the ttm_bo_add_move_fence takes a reference to the struct dma_fence, but
failed to release it on the error path, leading to a memory leak.
add dma_fence_put before return when error occur.
Signed-off-by: NLin Yi <teroincn@163.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

When we increment the counter we need to increment the pointer as well.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Fixes: e16858a7e6e7 drm/ttm: fix start page for huge page check in ttm_put_pages()
Reviewed-by: NMichel Dänzer <michel.daenzer@amd.com>
Acked-by: NHuang Rui <ray.huang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

The first page entry is always the same with itself.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NMichel Dänzer <michel.daenzer@amd.com>
Reviewed-by: NJunwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: NHuang Rui <ray.huang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

When ttm_put_pages() tries to figure out whether it's dealing with
transparent hugepages, it just reads past the bounds of the pages array
without a check.

v2: simplify the test if enough pages are left in the array (Christian).
Signed-off-by: NJann Horn <jannh@google.com>
Signed-off-by: NChristian König <christian.koenig@amd.com>
Fixes: 5c42c64f ("drm/ttm: fix the fix for huge compound pages")
Cc: stable@vger.kernel.org
Reviewed-by: NMichel Dänzer <michel.daenzer@amd.com>
Reviewed-by: NJunwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: NHuang Rui <ray.huang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

The symbol for binary prefix kibi is "Ki", with uppercase K.
(In contrast, the symbol for decimal kilo is lowercase "k".)
Signed-off-by: NJakub Wilk <jwilk@jwilk.net>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/gpu/drm/ttm/ttm_execbuf_util.c: In function 'ttm_eu_fence_buffer_objects':
drivers/gpu/drm/ttm/ttm_execbuf_util.c:191:24: warning:
 variable 'bdev' set but not used [-Wunused-but-set-variable]

It's not used any more and can be removed.
Signed-off-by: NYueHaibing <yuehaibing@huawei.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Don't account for them in other zones such as dma32. The kernel page
allocator has its own heuristics to avoid exhausting special zones
for regular kernel allocations.
Signed-off-by: NFelix Kuehling <Felix.Kuehling@amd.com>
Acked-by: NThomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

GEM defines DRM_FILE_PAGE_OFFSET_{START,SIZE} constants for the
mmap-able range of addresses. TTM can use them as well.
Signed-off-by: NThomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Acked-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>