Even though FENCE_SUBMIT is only documented to wait until the request in
the in-fence starts instead of waiting until it completes, it has a bit
more magic than that. If FENCE_SUBMIT is used to submit something to a
balanced engine, we would wait to assign engines until the primary
request was ready to start and then attempt to assign it to a different
engine than the primary. There is an IGT test (the bonded-slice subtest
of gem_exec_balancer) which exercises this by submitting a primary batch
to a specific VCS and then using FENCE_SUBMIT to submit a secondary
which can run on any VCS and have i915 figure out which VCS to run it on
such that they can run in parallel.
However, this functionality has never been used in the real world. The
media driver (the only user of FENCE_SUBMIT) always picks exactly two
physical engines to bond and never asks us to pick which to use.
v2 (Daniel Vetter):
- Mention the exact IGT test this breaks
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210708154835.528166-11-jason@jlekstrand.net
This adds a bunch of complexity which the media driver has never
actually used. The media driver does technically bond a balanced engine
to another engine but the balanced engine only has one engine in the
sibling set. This doesn't actually result in a virtual engine.
This functionality was originally added to handle cases where we may
have more than two video engines and media might want to load-balance
their bonded submits by, for instance, submitting to a balanced vcs0-1
as the primary and then vcs2-3 as the secondary. However, no such
hardware has shipped thus far and, if we ever want to enable such
use-cases in the future, we'll use the up-and-coming parallel submit API
which targets GuC submission.
This makes I915_CONTEXT_ENGINES_EXT_BOND a total no-op. We leave the
validation code in place in case we ever decide we want to do something
interesting with the bonding information.
v2 (Jason Ekstrand):
- Don't delete quite as much code.
v3 (Tvrtko Ursulin):
- Add some history to the commit message
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210708154835.528166-10-jason@jlekstrand.net
This has never been used by any userspace except IGT and provides no
real functionality beyond parroting back parameters userspace passed in
as part of context creation or via setparam. If the context is in
legacy mode (where you use I915_EXEC_RENDER and friends), it returns
success with zero data so it's not useful for discovering what engines
are in the context. It's also not a replacement for the recently
removed I915_CONTEXT_CLONE_ENGINES because it doesn't return any of the
balancing or bonding information.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210708154835.528166-9-jason@jlekstrand.net
This API is entirely unnecessary and I'd love to get rid of it. If
userspace wants a single timeline across multiple contexts, they can
either use implicit synchronization or a syncobj, both of which existed
at the time this feature landed. The justification given at the time
was that it would help GL drivers which are inherently single-timeline.
However, neither of our GL drivers actually wanted the feature. i965
was already in maintenance mode at the time and iris uses syncobj for
everything.
Unfortunately, as much as I'd love to get rid of it, it is used by the
media driver so we can't do that. We can, however, do the next-best
thing which is to embed a syncobj in the context and do exactly what
we'd expect from userspace internally. This isn't an entirely identical
implementation because it's no longer atomic if userspace races with
itself by calling execbuffer2 twice simultaneously from different
threads. It won't crash in that case; it just doesn't guarantee any
ordering between those two submits. It also means that sync files
exported from different engines on a SINGLE_TIMELINE context will have
different fence contexts. This is visible to userspace if it looks at
the obj_name field of sync_fence_info.
Moving SINGLE_TIMELINE to a syncobj emulation has a couple of technical
advantages beyond mere annoyance. One is that intel_timeline is no
longer an api-visible object and can remain entirely an implementation
detail. This may be advantageous as we make scheduler changes going
forward. Second is that, together with deleting the CLONE_CONTEXT API,
we should now have a 1:1 mapping between intel_context and
intel_timeline which may help us reduce locking.
v2 (Tvrtko Ursulin):
- Update the comment on i915_gem_context::syncobj to mention that it's
an emulation and the possible race if userspace calls execbuffer2
twice on the same context concurrently.
v2 (Jason Ekstrand):
- Wrap the checks for eb.gem_context->syncobj in unlikely()
- Drop the dma_fence reference
- Improved commit message
v3 (Jason Ekstrand):
- Move the dma_fence_put() to before the error exit
v4 (Tvrtko Ursulin):
- Add a comment about fence contexts to the commit message
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210708154835.528166-8-jason@jlekstrand.net
This API allows one context to grab bits out of another context upon
creation. It can be used as a short-cut for setparam(getparam()) for
things like I915_CONTEXT_PARAM_VM. However, it's never been used by any
real userspace. It's used by a few IGT tests and that's it. Since it
doesn't add any real value (most of the stuff you can CLONE you can copy
in other ways), drop it.
There is one thing that this API allows you to clone which you cannot
clone via getparam/setparam: timelines. However, timelines are an
implementation detail of i915 and not really something that needs to be
exposed to userspace. Also, sharing timelines between contexts isn't
obviously useful and supporting it has the potential to complicate i915
internally. It also doesn't add any functionality that the client can't
get in other ways. If a client really wants a shared timeline, they can
use a syncobj and set it as an in and out fence on every submit.
v2 (Jason Ekstrand):
- More detailed commit message
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210708154835.528166-7-jason@jlekstrand.net
Instead of handling it like a context param, unconditionally set it when
intel_contexts are created. For years we've had the idea of a watchdog
uAPI floating about. The aim was for media, so that they could set very
tight deadlines for their transcodes jobs, so that if you have a corrupt
bitstream (especially for decoding) you don't hang your desktop too
hard. But it's been stuck in limbo since forever, and this simplifies
things a bit in preparation for the proto-context work. If we decide to
actually make said uAPI a reality, we can do it through the proto-
context easily enough.
This does mean that we move from reading the request_timeout_ms param
once per engine when engines are created instead of once at context
creation. If someone changes request_timeout_ms between creating a
context and setting engines, it will mean that they get the new timeout.
If someone races setting request_timeout_ms and context creation, they
can theoretically end up with different timeouts. However, since both
of these are fairly harmless and require changing kernel params, we
don't care.
v2 (Tvrtko Ursulin):
- Add a comment about races with request_timeout_ms
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210708154835.528166-5-jason@jlekstrand.net
Previously, we were storing the ring size in the ring pointer before it
was actually allocated. We would then guard setting the ring size on
checking for CONTEXT_ALLOC_BIT. This is error-prone at best and really
only saves us a few bytes on something that already burns at least 4K.
Instead, this patch adds a new ring_size field and makes everything use
that.
v2 (Daniel Vetter):
- Replace 512 * SZ_4K with SZ_2M
v2 (Jason Ekstrand):
- Rebase on top of page migration code
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210708154835.528166-3-jason@jlekstrand.net
This reverts commit 88be76cdaf ("drm/i915: Allow userspace to specify
ringsize on construction"). This API was originally added for OpenCL
but the compute-runtime PR has sat open for a year without action so we
can still pull it out if we want. I argue we should drop it for three
reasons:
1. If the compute-runtime PR has sat open for a year, this clearly
isn't that important.
2. It's a very leaky API. Ring size is an implementation detail of the
current execlist scheduler and really only makes sense there. It
can't apply to the older ring-buffer scheduler on pre-execlist
hardware because that's shared across all contexts and it won't
apply to the GuC scheduler that's in the pipeline.
3. Having userspace set a ring size in bytes is a bad solution to the
problem of having too small a ring. There is no way that userspace
has the information to know how to properly set the ring size so
it's just going to detect the feature and always set it to the
maximum of 512K. This is what the compute-runtime PR does. The
scheduler in i915, on the other hand, does have the information to
make an informed choice. It could detect if the ring size is a
problem and grow it itself. Or, if that's too hard, we could just
increase the default size from 16K to 32K or even 64K instead of
relying on userspace to do it.
Let's drop this API for now and, if someone decides they really care
about solving this problem, they can do it properly.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210708154835.528166-2-jason@jlekstrand.net
46 bit addressing enables you to use 4 bits to support some
MKTME features, and 3 more bits for Optane support that uses
a subset of MTKME for persistent memory.
But GTT addressing sticking to 39 bit addressing, thus setting
dma_mask_size to 39 fixes below tests :
igt@i915_selftest@live@mman
igt@kms_big_fb@linear-32bpp-rotate-0
igt@gem_create@create-clear
igt@gem_mmap_offset@clear
igt@gem_mmap_gtt@cpuset-big-copy
In a way solves Gitlab#3142
https://gitlab.freedesktop.org/drm/intel/-/issues/3142, which had
following errors :
DMAR: DRHD: handling fault status reg 2
DMAR: [DMA Write] Request device [00:02.0] PASID ffffffff fault addr
7effff9000 [fault reason 05] PTE Write access is not set
0x7effff9000 is suspiciously exactly 39 bits, so it seems likely that
the HW just ends up masking off those extra bits hence DMA errors.
Changes since V2 :
- dim checkpatch error solved
Changes since V1 :
- Added more details to commit message - Matthew Auld
Signed-off-by: Tejas Upadhyay <tejaskumarx.surendrakumar.upadhyay@intel.com>
Acked-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210708071222.955455-1-tejaskumarx.surendrakumar.upadhyay@intel.com
For some specialised objects we might need something larger than the
regions min_page_size due to some hw restriction, and slightly more
hairy is needing something smaller with the guarantee that such objects
will never be inserted into any GTT, which is the case for the paging
structures.
This also fixes how we setup the BO page_alignment, if we later migrate
the object somewhere else. For example if the placements are {SMEM,
LMEM}, then we might get this wrong. Pushing the min_page_size behaviour
into the manager should fix this.
v2(Thomas): push the default page size behaviour into buddy_man, and let
the user override it with the page-alignment, which looks cleaner
v3: rebase on ttm sys changes
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210625103824.558481-1-matthew.auld@intel.com
Introduce an interface to migrate objects between regions.
This is primarily intended to migrate objects to LMEM for display and
to SYSTEM for dma-buf, but might be reused in one form or another for
performance-based migration.
v2:
- Verify that the memory region given as an id really exists.
(Reported by Matthew Auld)
- Call i915_gem_object_{init,release}_memory_region() when switching region
to handle also switching region lists. (Reported by Matthew Auld)
v3:
- Fix i915_gem_object_can_migrate() to return true if object is already in
the correct region, even if the object ops doesn't have a migrate()
callback.
- Update typo in commit message.
- Fix kerneldoc of i915_gem_object_wait_migration().
v4:
- Improve documentation (Suggested by Mattew Auld and Michael Ruhl)
- Always assume TTM migration hits a TTM move and unsets the pages through
move_notify. (Reported by Matthew Auld)
- Add a dma_fence_might_wait() annotation to
i915_gem_object_wait_migration() (Suggested by Daniel Vetter)
v5:
- Re-add might_sleep() instead of __dma_fence_might_wait(), Sent
v4 with the wrong version, didn't compile and __dma_fence_might_wait()
is not exported.
- Added an R-B.
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210629151203.209465-2-thomas.hellstrom@linux.intel.com
For discrete, use TTM for both cached and WC system memory. That means
we currently rely on the TTM memory accounting / shrinker. For cached
system memory we should consider remaining shmem-backed, which can be
implemented from our ttm_tt_populate callback. We can then also reuse our
own very elaborate shrinker for that memory.
If an object is evicted to a gem allowable region, we will now consider
the object migrated, and we flip the gem region and move the object to a
different region list. Since we are now changing gem regions, we can't
any longer rely on the CONTIGUOUS flag being set based on the region
min page size, so remove that flag update. If we want to reintroduce it,
we need to put it in the mutable flags.
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210624084240.270219-4-thomas.hellstrom@linux.intel.com
After a TTM move or object init we need to update the i915 gem flags and
caching settings to reflect the new placement. Currently caching settings
are not changed during the lifetime of an object, although that might
change moving forward if we run into performance issues or issues with
WC system page allocations.
Also introduce gpu_binds_iomem() and cpu_maps_iomem() to clean up the
various ways we previously used to detect this.
Finally, initialize the TTM object reserved to be able to update
flags and caching before anyone else gets hold of the object.
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210624084240.270219-3-thomas.hellstrom@linux.intel.com
In
commit ebc0808fa2
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Tue Oct 18 13:02:51 2016 +0100
drm/i915: Restrict pagefault disabling to just around copy_from_user()
we entirely missed that there's a slow path call to eb_relocate_entry
(or i915_gem_execbuffer_relocate_entry as it was called back then)
which was left fully wrapped by pagefault_disable/enable() calls.
Previously any issues with blocking calls where handled by the
following code:
/* we can't wait for rendering with pagefaults disabled */
if (pagefault_disabled() && !object_is_idle(obj))
return -EFAULT;
Now at this point the prefaulting was still around, which means in
normal applications it was very hard to hit this bug. No idea why the
regressions in igts weren't caught.
Now this all changed big time with 2 patches merged closely together.
First
commit 2889caa923
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Fri Jun 16 15:05:19 2017 +0100
drm/i915: Eliminate lots of iterations over the execobjects array
removes the prefaulting from the first relocation path, pushing it into
the first slowpath (of which this patch added a total of 3 escalation
levels). This would have really quickly uncovered the above bug, were
it not for immediate adding a duct-tape on top with
commit 7dd4f6729f
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Fri Jun 16 15:05:24 2017 +0100
drm/i915: Async GPU relocation processing
by pushing all all the relocation patching to the gpu if the buffer
was busy, which avoided all the possible blocking calls.
The entire slowpath was then furthermore ditched in
commit 7dc8f11437
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Wed Mar 11 16:03:10 2020 +0000
drm/i915/gem: Drop relocation slowpath
and resurrected in
commit fd1500fcd4
Author: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Date: Wed Aug 19 16:08:43 2020 +0200
Revert "drm/i915/gem: Drop relocation slowpath".
but this did not further impact what's going on.
Since pagefault_disable/enable is an atomic section, any sleeping in
there is prohibited, and we definitely do that without gpu relocations
since we have to wait for the gpu usage to finish before we can patch
up the relocations.
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Cc: Jon Bloomfield <jon.bloomfield@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: Dave Airlie <airlied@redhat.com>
Cc: Jason Ekstrand <jason@jlekstrand.net>
Link: https://patchwork.freedesktop.org/patch/msgid/20210618214503.1773805-1-daniel.vetter@ffwll.ch
Introduce i915_sched_engine object which is lower level data structure
that i915_scheduler / generic code can operate on without touching
execlist specific structures. This allows additional submission backends
to be added without breaking the layering. Currently the execlists
backend uses 1 of these object per each engine (physical or virtual) but
future backends like the GuC will point to less instances utilizing the
reference counting.
This is a bit of detour to integrating the i915 with the DRM scheduler
but this object will still exist when the DRM scheduler lands in the
i915. It will however look a bit different. It will encapsulate the
drm_gpu_scheduler object plus and common variables (to the backends)
related to scheduling. Regardless this is a step in the right direction.
This patch starts the aforementioned transition by moving the priolist
into the i915_sched_engine object.
v3:
(Jason Ekstrand)
Update comment next to intel_engine_cs.virtual
Add kernel doc
(Checkpatch)
Fix double the in commit message
v4:
(Daniele)
Update comment message.
Add comment about subclass field
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210618010638.98941-2-matthew.brost@intel.com
We have assumed that if the current placement was not the requested
placement, but instead one of the busy placements, a TTM move would have
been triggered. That is not the case.
So when we initially place LMEM objects in "Limbo", (that is system
placement without any pages allocated), to be able to defer clearing
objects until first get_pages(), the first get_pages() would happily keep
objects in system memory if that is one of the allowed placements. And
since we don't yet support i915 GEM system memory from TTM, everything
breaks apart.
So make sure we try the requested placement first, if no eviction is
needed. If that fails, retry with all allowed placements also allowing
evictions. Also make sure we handle TTM failure codes correctly.
Also temporarily (until we support i915 GEM system on TTM), restrict
allowed placements to the requested placement to avoid things falling
apart should LMEM be full.
Fixes: 38f28c0695 ("drm/i915/ttm: Calculate the object placement at get_pages time")
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210618132515.163277-1-thomas.hellstrom@linux.intel.com
Because Render Power Gating restricts us to just a single subslice as a
valid steering target for reads of multicast registers in a SUBSLICE
range, the default steering we setup at init may not lead to a suitable
target for L3BANK multicast register. In cases where it does not, use
explicit runtime steering whenever an L3BANK multicast register is read.
While we're at it, let's simplify the function a little bit and drop its
support for gen10/CNL since no such platforms ever materialized for real
use. Multicast register steering is already an area that causes enough
confusion; no need to complicate it with what's effectively dead code.
v2:
- Use gt->uncore instead of gt->i915->uncore. (Tvrtko)
- Use {} as table terminator. (Rodrigo)
v3:
- L3bank fuse register is a disable mask rather than an enable mask.
We need to invert it before use. (CI)
v4:
- L3bank ID goes in the subslice field, not the slice field. (CI)
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210617211425.1943662-4-matthew.d.roper@intel.com
