drm/i915/vma: fix struct i915_vma_bindinfo kernel-doc

You can't document both a sub-struct type and a struct member at the
same time. Separate them.

drivers/gpu/drm/i915/i915_vma_resource.h:91: warning: Incorrect use of kernel-doc format: * struct i915_vma_bindinfo - Information needed for async bind
drivers/gpu/drm/i915/i915_vma_resource.h:129: warning: Function parameter or member 'bi' not described in 'i915_vma_resource'

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/89ab5cf179566c429383cc57db746038f75cba0d.1683041799.git.jani.nikula@intel.com

drm/i915/vma: document struct i915_vma_resource wakeref member

drivers/gpu/drm/i915/i915_vma_resource.h:129: warning: Function parameter or member 'wakeref' not described in 'i915_vma_resource'

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/50389365e095dc564ab5f1f1e3647934163ffefa.1683041799.git.jani.nikula@intel.com

drm/i915: Introduce guard pages to i915_vma

Introduce the concept of padding the i915_vma with guard pages before
and after. The major consequence is that all ordinary uses of i915_vma
must use i915_vma_offset/i915_vma_size and not i915_vma.node.start/size
directly, as the drm_mm_node will include the guard pages that surround
our object.

The biggest connundrum is how exactly to mix requesting a fixed address
with guard pages, particularly through the existing uABI. The user does
not know about guard pages, so such must be transparent to the user, and
so the execobj.offset must be that of the object itself excluding the
guard. So a PIN_OFFSET_FIXED must then be exclusive of the guard pages.
The caveat is that some placements will be impossible with guard pages,
as wrap arounds need to be avoided, and the vma itself will require a
larger node. We must not report EINVAL but ENOSPC as these are unavailable
locations within the GTT rather than conflicting user requirements.

In the next patch, we start using guard pages for scanout objects. While
these are limited to GGTT vma, on a few platforms these vma (or at least
an alias of the vma) is shared with userspace, so we may leak the
existence of such guards if we are not careful to ensure that the
execobj.offset is transparent and excludes the guards. (On such platforms
like ivb, without full-ppgtt, userspace has to use relocations so the
presence of more untouchable regions within its GTT such be of no further
issue.)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Tejas Upadhyay <tejaskumarx.surendrakumar.upadhyay@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221201203912.346110-1-andi.shyti@linux.intel.com

drm/i915: Wrap all access to i915_vma.node.start|size

We already wrap i915_vma.node.start for use with the GGTT, as there we
can perform additional sanity checks that the node belongs to the GGTT
and fits within the 32b registers. In the next couple of patches, we
will introduce guard pages around the objects _inside_ the drm_mm_node
allocation. That is we will offset the vma->pages so that the first page
is at drm_mm_node.start + vma->guard (not 0 as is currently the case).
All users must then not use i915_vma.node.start directly, but compute
the guard offset, thus all users are converted to use a
i915_vma_offset() wrapper.

The notable exceptions are the selftests that are testing exact
behaviour of i915_vma_pin/i915_vma_insert.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Tejas Upadhyay <tejaskumarx.surendrakumar.upadhyay@intel.com>
Co-developed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221130235805.221010-3-andi.shyti@linux.intel.com

drm/i915/gt: Batch TLB invalidations

Invalidate TLB in batches, in order to reduce performance regressions.

Currently, every caller performs a full barrier around a TLB
invalidation, ignoring all other invalidations that may have already
removed their PTEs from the cache. As this is a synchronous operation
and can be quite slow, we cause multiple threads to contend on the TLB
invalidate mutex blocking userspace.

We only need to invalidate the TLB once after replacing our PTE to
ensure that there is no possible continued access to the physical
address before releasing our pages. By tracking a seqno for each full
TLB invalidate we can quickly determine if one has been performed since
rewriting the PTE, and only if necessary trigger one for ourselves.

That helps to reduce the performance regression introduced by TLB
invalidate logic.

[mchehab: rebased to not require moving the code to a separate file]

Cc: stable@vger.kernel.org
Fixes: 7938d61591d3 ("drm/i915: Flush TLBs before releasing backing store")
Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Cc: Fei Yang <fei.yang@intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/4e97ef5deb6739cadaaf40aa45620547e9c4ec06.1658924372.git.mchehab@kernel.org

drm/i915/gt: Batch TLB invalidations

Invalidate TLB in batches, in order to reduce performance regressions.

Currently, every caller performs a full barrier around a TLB
invalidation, ignoring all other invalidations that may have already
removed their PTEs from the cache. As this is a synchronous operation
and can be quite slow, we cause multiple threads to contend on the TLB
invalidate mutex blocking userspace.

We only need to invalidate the TLB once after replacing our PTE to
ensure that there is no possible continued access to the physical
address before releasing our pages. By tracking a seqno for each full
TLB invalidate we can quickly determine if one has been performed since
rewriting the PTE, and only if necessary trigger one for ourselves.

That helps to reduce the performance regression introduced by TLB
invalidate logic.

[mchehab: rebased to not require moving the code to a separate file]

Cc: stable@vger.kernel.org
Fixes: 7938d61591d3 ("drm/i915: Flush TLBs before releasing backing store")
Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Cc: Fei Yang <fei.yang@intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/4e97ef5deb6739cadaaf40aa45620547e9c4ec06.1658924372.git.mchehab@kernel.org
(cherry picked from commit 5d36acb7198b0e5eb88e6b701f9ad7b9448f8df9)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/i915: Remove the vm open count

vms are not getting properly closed. Rather than fixing that,
Remove the vm open count and instead rely on the vm refcount.

The vm open count existed solely to break the strong references the
vmas had on the vms. Now instead make those references weak and
ensure vmas are destroyed when the vm is destroyed.

Unfortunately if the vm destructor and the object destructor both
wants to destroy a vma, that may lead to a race in that the vm
destructor just unbinds the vma and leaves the actual vma destruction
to the object destructor. However in order for the object destructor
to ensure the vma is unbound it needs to grab the vm mutex. In order
to keep the vm mutex alive until the object destructor is done with
it, somewhat hackishly grab a vm_resv refcount that is released late
in the vma destruction process, when the vm mutex is no longer needed.

v2: Address review-comments from Niranjana
- Clarify that the struct i915_address_space::skip_pte_rewrite is a hack
and should ideally be replaced in an upcoming patch.
- Remove an unneeded continue in clear_vm_list and update comment.

v3:
- Documentation update
- Commit message formatting

Co-developed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220304082641.308069-2-thomas.hellstrom@linux.intel.com

drm/i915: Use struct vma_resource instead of struct vma_snapshot

There is always a struct vma_resource guaranteed to be alive when we
access a corresponding struct vma_snapshot.

So ditch the latter and instead of allocating vma_snapshots, reference
the already existning vma_resource.

This requires a couple of extra members in struct vma_resource but that's
a small price to pay for the simplification.

v2:
- Fix a missing include and declaration (kernel test robot <lkp@intel.com>)

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220110172219.107131-7-thomas.hellstrom@linux.intel.com

drm/i915: Use vma resources for async unbinding

Implement async (non-blocking) unbinding by not syncing the vma before
calling unbind on the vma_resource.
Add the resulting unbind fence to the object's dma_resv from where it is
picked up by the ttm migration code.
Ideally these unbind fences should be coalesced with the migration blit
fence to avoid stalling the migration blit waiting for unbind, as they
can certainly go on in parallel, but since we don't yet have a
reasonable data structure to use to coalesce fences and attach the
resulting fence to a timeline, we defer that for now.

Note that with async unbinding, even while the unbind waits for the
preceding bind to complete before unbinding, the vma itself might have been
destroyed in the process, clearing the vma pages. Therefore we can
only allow async unbinding if we have a refcounted sg-list and keep a
refcount on that for the vma resource pages to stay intact until
binding occurs. If this condition is not met, a request for an async
unbind is diverted to a sync unbind.

v2:
- Use a separate kmem_cache for vma resources for now to isolate their
memory allocation and aid debugging.
- Move the check for vm closed to the actual unbinding thread. Regardless
of whether the vm is closed, we need the unbind fence to properly wait
for capture.
- Clear vma_res::vm on unbind and update its documentation.
v4:
- Take cache coloring into account when searching for vma resources
pending unbind. (Matthew Auld)
v5:
- Fix timeout and error check in i915_vma_resource_bind_dep_await().
- Avoid taking a reference on the object for async binding if
async unbind capable.
- Fix braces around a single-line if statement.
v6:
- Fix up the cache coloring adjustment. (Kernel test robot <lkp@intel.com>)
- Don't allow async unbinding if the vma_res pages are not the same as
the object pages. (Matthew Auld)
v7:
- s/unsigned long/u64/ in a number of places (Matthew Auld)

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220110172219.107131-5-thomas.hellstrom@linux.intel.com

drm/i915: Use the vma resource as argument for gtt binding / unbinding

When introducing asynchronous unbinding, the vma itself may no longer
be alive when the actual binding or unbinding takes place.

Update the gtt i915_vma_ops accordingly to take a struct i915_vma_resource
instead of a struct i915_vma for the bind_vma() and unbind_vma() ops.
Similarly change the insert_entries() op for struct i915_address_space.

Replace a couple of i915_vma_snapshot members with their newly introduced
i915_vma_resource counterparts, since they have the same lifetime.

Also make sure to avoid changing the struct i915_vma_flags (in particular
the bind flags) async. That should now only be done sync under the
vm mutex.

v2:
- Update the vma_res::bound_flags when binding to the aliased ggtt
v6:
- Remove I915_VMA_ALLOC_BIT (Matthew Auld)
- Change some members of struct i915_vma_resource from unsigned long to u64
(Matthew Auld)
v7:
- Fix vma resource size parameters to be u64 rather than unsigned long
(Matthew Auld)

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220110172219.107131-3-thomas.hellstrom@linux.intel.com

drm/i915: Initial introduction of vma resources

Introduce vma resources, sort of similar to TTM resources, needed for
asynchronous bind management. Initially we will use them to hold
completion of unbinding when we capture data from a vma, but they will
be used extensively in upcoming patches for asynchronous vma unbinding.

v6:
- Some documentation updates

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220110172219.107131-2-thomas.hellstrom@linux.intel.com