Looking through the attributes for DMA mappings, it appears that by
default dma_map_sg will try and create a kernel accessible map of the
page. We never access this, as we either have a struct page already or
an iomap, so we can request that the dma mapper does not create one.
Without a kernel map in place, one presumes the rest of the memory
control attributes do not apply. We also explicitly control the caches
around the mappings, so we can ask it not to bother synchronising itself.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200706224308.22636-1-chris@chris-wilson.co.uk
Converts various instances of the printk drm logging macros to the
struct drm_device based logging macros in the drm/i915 folder using the
following coccinelle script that transforms based on the existence of
the struct drm_i915_private device pointer:
@@
identifier fn, T;
@@
fn(...) {
...
struct drm_i915_private *T = ...;
<+...
(
-DRM_INFO(
+drm_info(&T->drm,
...)
|
-DRM_ERROR(
+drm_err(&T->drm,
...)
|
-DRM_WARN(
+drm_warn(&T->drm,
...)
|
-DRM_DEBUG_KMS(
+drm_dbg_kms(&T->drm,
...)
|
-DRM_DEBUG_DRIVER(
+drm_dbg(&T->drm,
...)
|
-DRM_DEBUG_ATOMIC(
+drm_dbg_atomic(&T->drm,
...)
)
...+>
}
@@
identifier fn, T;
@@
fn(...,struct drm_i915_private *T,...) {
<+...
(
-DRM_INFO(
+drm_info(&T->drm,
...)
|
-DRM_ERROR(
+drm_err(&T->drm,
...)
|
-DRM_WARN(
+drm_warn(&T->drm,
...)
|
-DRM_DEBUG_DRIVER(
+drm_dbg(&T->drm,
...)
|
-DRM_DEBUG_KMS(
+drm_dbg_kms(&T->drm,
...)
|
-DRM_DEBUG_ATOMIC(
+drm_dbg_atomic(&T->drm,
...)
)
...+>
}
Checkpatch warnings were fixed manually.
Instances of the DRM_DEBUG macro were not converted due to lack of a
consensus of an analogous struct drm_device based macro.
References: https://lists.freedesktop.org/archives/dri-devel/2020-January/253381.html
Signed-off-by: Wambui Karuga <wambui.karugax@gmail.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200131093416.28431-2-wambui.karugax@gmail.com
Allocate only an internal intel_context for the kernel_context, forgoing
a global GEM context for internal use as we only require a separate
address space (for our own protection).
Now having weaned GT from requiring ce->gem_context, we can stop
referencing it entirely. This also means we no longer have to create random
and unnecessary GEM contexts for internal use.
GEM contexts are now entirely for tracking GEM clients, and intel_context
the execution environment on the GPU.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Andi Shyti <andi.shyti@intel.com>
Acked-by: Andi Shyti <andi.shyti@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191221160324.1073045-1-chris@chris-wilson.co.uk
After much hair pulling, resort to preallocating the ppGTT entries on
init to circumvent the apparent lack of PD invalidate following the
write to PP_DCLV upon switching mm between contexts (and here the same
context after binding new objects). However, the details of that PP_DCLV
invalidate are still unknown, and it appears we need to reload the mm
twice to cover over a timing issue. Worrying.
Fixes: 3dc007fe9b ("drm/i915/gtt: Downgrade gen7 (ivb, byt, hsw) back to aliasing-ppgtt")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191129201328.1398583-1-chris@chris-wilson.co.uk
Forgo the struct_mutex serialisation for i915_active, and interpose its
own mutex handling for active/retire.
This is a multi-layered sleight-of-hand. First, we had to ensure that no
active/retire callbacks accidentally inverted the mutex ordering rules,
nor assumed that they were themselves serialised by struct_mutex. More
challenging though, is the rule over updating elements of the active
rbtree. Instead of the whole i915_active now being serialised by
struct_mutex, allocations/rotations of the tree are serialised by the
i915_active.mutex and individual nodes are serialised by the caller
using the i915_timeline.mutex (we need to use nested spinlocks to
interact with the dma_fence callback lists).
The pain point here is that instead of a single mutex around execbuf, we
now have to take a mutex for active tracker (one for each vma, context,
etc) and a couple of spinlocks for each fence update. The improvement in
fine grained locking allowing for multiple concurrent clients
(eventually!) should be worth it in typical loads.
v2: Add some comments that barely elucidate anything :(
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/20191004134015.13204-6-chris@chris-wilson.co.uk
Replace the struct_mutex requirement for pinning the i915_vma with the
local vm->mutex instead. Note that the vm->mutex is tainted by the
shrinker (we require unbinding from inside fs-reclaim) and so we cannot
allocate while holding that mutex. Instead we have to preallocate
workers to do allocate and apply the PTE updates after we have we
reserved their slot in the drm_mm (using fences to order the PTE writes
with the GPU work and with later unbind).
In adding the asynchronous vma binding, one subtle requirement is to
avoid coupling the binding fence into the backing object->resv. That is
the asynchronous binding only applies to the vma timeline itself and not
to the pages as that is a more global timeline (the binding of one vma
does not need to be ordered with another vma, nor does the implicit GEM
fencing depend on a vma, only on writes to the backing store). Keeping
the vma binding distinct from the backing store timelines is verified by
a number of async gem_exec_fence and gem_exec_schedule tests. The way we
do this is quite simple, we keep the fence for the vma binding separate
and only wait on it as required, and never add it to the obj->resv
itself.
Another consequence in reducing the locking around the vma is the
destruction of the vma is no longer globally serialised by struct_mutex.
A natural solution would be to add a kref to i915_vma, but that requires
decoupling the reference cycles, possibly by introducing a new
i915_mm_pages object that is own by both obj->mm and vma->pages.
However, we have not taken that route due to the overshadowing lmem/ttm
discussions, and instead play a series of complicated games with
trylocks to (hopefully) ensure that only one destruction path is called!
v2: Add some commentary, and some helpers to reduce patch churn.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
Since we cannot allocate underneath the vm->mutex (it is used in the
direct-reclaim paths), we need to shift the allocations off into a
mutexless worker with fence recursion prevention. To know when we need
this protection, we mark up the address spaces that do allocate before
insertion. In the future, we may wish to extend the async bind scheme to
more than just allocations.
v2: s/vm->bind_alloc/vm->bind_async_flags/
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-3-chris@chris-wilson.co.uk
The sg_table for our backing store might contain addresses from
stolen-memory or in the future local-memory, at which point this is no
longer a dma-iterator. As a consequence we should now break on NULL
iter.sgp, instead of dmap == 0 which is considered an invalid dma
address.
As a bonus, gcc much prefers this construct,
Function old new delta
gen8_ggtt_insert_entries 211 192 -19
gen6_ggtt_insert_entries 292 262 -30
i915_error_object_create 996 954 -42
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190829201919.21493-1-matthew.auld@intel.com
In order for the Braswell top-level PD to remain the same from the time
of request construction to its submission onto HW, as we may be
asynchronously rewriting the page tables (thus changing the expected
register state after having already stored the old addresses in the
request), the top level PD must be preallocated.
So wave goodbye to our lazy allocation of those 4x2 pages.
v2: A little bit of write-flushing required (presumably it always has
been required, but now we are more susceptible and it is showing up!)
v3: Put back the forced-PD-reload on every batch, we can't survive
without it and explicitly marking the context for PD reload makes
Braswell turn nasty.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190823141421.2398-1-chris@chris-wilson.co.uk
When under severe stress for GTT mappable space, the LRU eviction model
falls off a cliff. We spend all our time scanning the much larger
non-mappable area searching for something within the mappable zone we can
evict. Turn this on its head by only using the full vma for the object if
it is already pinned in the mappable zone or there is sufficient *free*
space to accommodate it (prioritizing speedy reuse). If there is not,
immediately fall back to using small chunks (tilerow for GTT mmap, single
pages for pwrite/relocation) and using random eviction before doing a full
search.
Testcase: igt/gem_concurrent_blt
References: https://bugs.freedesktop.org/show_bug.cgi?id=110848
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190821123234.19194-1-chris@chris-wilson.co.uk
The current assertion tries to make sure that we do not over count the
number of used PDE inside a page directory -- that is with an array of
512 pde, we do not expect more than 512 elements used! However, our
assertion has to take into account that as we pin an element into the
page directory, the caller first pins the page directory so the usage
count is one higher. However, this should be one extra pin per thread,
and the upper bound is that we may have one thread for each entry.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190820141218.14714-1-chris@chris-wilson.co.uk
