This was the competing idea long ago, but it was only with the rewrite
of the idr as an radixtree and using the radixtree directly ourselves,
along with the realisation that we can store the vma directly in the
radixtree and only need a list for the reverse mapping, that made the
patch performant enough to displace using a hashtable. Though the vma ht
is fast and doesn't require any extra allocation (as we can embed the node
inside the vma), it does require a thread for resizing and serialization
and will have the occasional slow lookup. That is hairy enough to
investigate alternatives and favour them if equivalent in peak performance.
One advantage of allocating an indirection entry is that we can support a
single shared bo between many clients, something that was done on a
first-come first-serve basis for shared GGTT vma previously. To offset
the extra allocations, we create yet another kmem_cache for them.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170816085210.4199-5-chris@chris-wilson.co.uk
During execbuf, a mandatory step is that we add this request (this
fence) to each object's reservation_object. Inside execbuf, we track the
vma, and to add the fence to the reservation_object then means having to
first chase the obj, incurring another cache miss. We can reduce the
number of cache misses by stashing a pointer to the reservation_object
in the vma itself.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170616140525.6394-1-chris@chris-wilson.co.uk
If we take a reference to the object/vma when it is first used in an
execbuf, we can keep that reference until the object's file-local handle
is closed. Thereby saving a frequent ref/unref pair.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
The major scaling bottleneck in execbuffer is the processing of the
execobjects. Creating an auxiliary list is inefficient when compared to
using the execobject array we already have allocated.
Reservation is then split into phases. As we lookup up the VMA, we
try and bind it back into active location. Only if that fails, do we add
it to the unbound list for phase 2. In phase 2, we try and add all those
objects that could not fit into their previous location, with fallback
to retrying all objects and evicting the VM in case of severe
fragmentation. (This is the same as before, except that phase 1 is now
done inline with looking up the VMA to avoid an iteration over the
execobject array. In the ideal case, we eliminate the separate reservation
phase). During the reservation phase, we only evict from the VM between
passes (rather than currently as we try to fit every new VMA). In
testing with Unreal Engine's Atlantis demo which stresses the eviction
logic on gen7 class hardware, this speed up the framerate by a factor of
2.
The second loop amalgamation is between move_to_gpu and move_to_active.
As we always submit the request, even if incomplete, we can use the
current request to track active VMA as we perform the flushes and
synchronisation required.
The next big advancement is to avoid copying back to the user any
execobjects and relocations that are not changed.
v2: Add a Theory of Operation spiel.
v3: Fall back to slow relocations in preparation for flushing userptrs.
v4: Document struct members, factor out eb_validate_vma(), add a few
more comments to explain some magic and hide other magic behind macros.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
The advent of full-ppgtt lead to an extra indirection between the object
and its binding. That extra indirection has a noticeable impact on how
fast we can convert from the user handles to our internal vma for
execbuffer. In order to bypass the extra indirection, we use a
resizable hashtable to jump from the object to the per-ctx vma.
rhashtable was considered but we don't need the online resizing feature
and the extra complexity proved to undermine its usefulness. Instead, we
simply reallocate the hastable on demand in a background task and
serialize it before iterating.
In non-full-ppgtt modes, multiple files and multiple contexts can share
the same vma. This leads to having multiple possible handle->vma links,
so we only use the first to establish the fast path. The majority of
buffers are not shared and so we should still be able to realise
speedups with multiple clients.
v2: Prettier names, more magic.
v3: Many style tweaks, most notably hiding the misuse of execobj[].rsvd2
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Since tweaking i915_vma_compare() we allowed constructors to skip
clearing the ggtt_view believing that we didn't access the unused
members. That, as it turns out, was not entirely true. In particular,
i915_gem_fault() uses
ret = remap_io_mapping(area,
area->vm_start + (vma->ggtt_view.partial.offset << PAGE_SHIFT),
(ggtt->mappable_base + vma->node.start) >> PAGE_SHIFT,
min_t(u64, vma->size, area->vm_end - area->vm_start),
&ggtt->mappable);
i.e. the ggtt_view.partial for both normal and partial views. If we
allowed garbage into the normal vma->ggtt_view and then try userspace
tried to mmap it, we could explode in an unobvious fashion.
Fixes: 7b92c047ba ("drm/i915: Eliminate superfluous i915_ggtt_view_rotated")
Fixes: 3bf4d57519 ("drm/i915: Stop clearing i915_ggtt_view")
Reported-by: Matthew Auld <matthew.william.auld@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170123145245.3972-1-chris@chris-wilson.co.uk
Tested-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
With the introduce of i915_vma_instance() for obtaining the VMA
singleton for a (obj, vm, view) tuple, we can remove the
i915_vma_create() in favour of a single entry point. We do incur a
lookup onto an empty tree, but the i915_vma_create() were being called
infrequently and during initialisation, so the small overhead is
negligible.
v2: Drop the i915_ prefix from the now static vma_create() function
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170116152131.18089-4-chris@chris-wilson.co.uk
Reading the ggtt_views is much more pleasant without the extra
characters from specifying the union (i.e. ggtt_view.partial rather than
ggtt_view.params.partial). To make this work inside i915_vma_compare()
with only a single memcmp requires us to ensure that there are no
uninitialised bytes within each branch of the union (we make sure the
structs are packed) and we need to store the size of each branch.
v4: Rewrite changelog and add comments explaining the assert.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20170114002827.31315-5-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
When we evict from the GTT to make room for an object, the hole we
create is put onto the MRU stack inside the drm_mm range manager. On the
next search pass, we can speed up a PIN_HIGH allocation by referencing
that stack for the new hole.
v2: Pull together the 3 identical implements (ahem, a couple were
outdated) into a common routine for allocating a node and evicting as
necessary.
v3: Detect invalid calls to i915_gem_gtt_insert()
v4: kerneldoc
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170111112312.31493-1-chris@chris-wilson.co.uk
Start converting over from the byte count to its semantic macro, either
we want to allocate the size of a physical page in main memory or we
want the size of a virtual page in the GTT. 4096 could mean either, but
PAGE_SIZE and I915_GTT_PAGE_SIZE are explicit and should help improve
code comprehension and future changes. In the future, we may want to use
variable GTT page sizes and so have the challenge of knowing which
hardcoded values were used to represent a physical page vs the virtual
page.
v2: Look for a few more 4096s to convert, discover IS_ALIGNED().
v3: 4096ul paranoia, make fence alignment a distinct value of 4096, keep
bdw stolen w/a as 4096 until we know better.
v4: Add asserts that i915_vma_insert() start/end are aligned to GTT page
sizes.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170110144734.26052-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
The fence size/alignment is a combination of the vma size plus object
tiling parameters. Those parameters are rarely changed, making the fence
size/alignemnt roughly constant for the lifetime of the VMA. We can
simplify subsequent calculations by precalculating the size/alignment
required for GGTT vma taking fencing into account (with an update if we
do change the tiling or stride).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170109161613.11881-4-chris@chris-wilson.co.uk
Directly merge drm-misc into drm-intel since Dave is on vacation and
we need the various drm-misc patches (fb format rework, drm mm fixes,
selftest framework and others). Also pulled back -rc2 in first to
resync with drm-intel-fixes and make sure I can reuse the exact rerere
solutions from drm-tip for safety, and because I'm lazy.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Insulate users from changes to the internal hole tracking within
struct drm_mm_node by using an accessor for hole_follows.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
[danvet: resolve conflicts in i915_vma.c]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Soft-pinning depends upon being able to check for availabilty of an
interval and evict overlapping object from a drm_mm range manager very
quickly. Currently it uses a linear list, and so performance is dire and
not suitable as a general replacement. Worse, the current code will oops
if it tries to evict an active buffer.
It also helps if the routine reports the correct error codes as expected
by its callers and emits a tracepoint upon use.
For posterity since the wrong patch was pushed (i.e. that missed these
key points and had known bugs), this is the changelog that should have
been on commit 506a8e87d8 ("drm/i915: Add soft-pinning API for
execbuffer"):
Userspace can pass in an offset that it presumes the object is located
at. The kernel will then do its utmost to fit the object into that
location. The assumption is that userspace is handling its own object
locations (for example along with full-ppgtt) and that the kernel will
rarely have to make space for the user's requests.
This extends the DRM_IOCTL_I915_GEM_EXECBUFFER2 to do the following:
* if the user supplies a virtual address via the execobject->offset
*and* sets the EXEC_OBJECT_PINNED flag in execobject->flags, then
that object is placed at that offset in the address space selected
by the context specifier in execbuffer.
* the location must be aligned to the GTT page size, 4096 bytes
* as the object is placed exactly as specified, it may be used by this
execbuffer call without relocations pointing to it
It may fail to do so if:
* EINVAL is returned if the object does not have a 4096 byte aligned
address
* the object conflicts with another pinned object (either pinned by
hardware in that address space, e.g. scanouts in the aliasing ppgtt)
or within the same batch.
EBUSY is returned if the location is pinned by hardware
EINVAL is returned if the location is already in use by the batch
* EINVAL is returned if the object conflicts with its own alignment (as meets
the hardware requirements) or if the placement of the object does not fit
within the address space
All other execbuffer errors apply.
Presence of this execbuf extension may be queried by passing
I915_PARAM_HAS_EXEC_SOFTPIN to DRM_IOCTL_I915_GETPARAM and checking for
a reported value of 1 (or greater).
v2: Combine the hole/adjusted-hole ENOSPC checks
v3: More color, more splitting, more blurb.
Fixes: 506a8e87d8 ("drm/i915: Add soft-pinning API for execbuffer")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161205142941.21965-2-chris@chris-wilson.co.uk
