A starting point to counter the pervasive struct_mutex. For the goal of
avoiding (or at least blocking under them!) global locks during user
request submission, a simple but important step is being able to manage
each clients GTT separately. For which, we want to replace using the
struct_mutex as the guard for all things GTT/VM and switch instead to a
specific mutex inside i915_address_space.
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/20190128102356.15037-2-chris@chris-wilson.co.uk
Our goal is to remove struct_mutex and replace it with fine grained
locking. One of the thorny issues is our eviction logic for reclaiming
space for an execbuffer (or GTT mmaping, among a few other examples).
While eviction itself is easy to move under a per-VM mutex, performing
the activity tracking is less agreeable. One solution is not to do any
MRU tracking and do a simple coarse evaluation during eviction of
active/inactive, with a loose temporal ordering of last
insertion/evaluation. That keeps all the locking constrained to when we
are manipulating the VM itself, neatly avoiding the tricky handling of
possible recursive locking during execbuf and elsewhere.
Note that discarding the MRU (currently implemented as a pair of lists,
to avoid scanning the active list for a NONBLOCKING search) is unlikely
to impact upon our efficiency to reclaim VM space (where we think a LRU
model is best) as our current strategy is to use random idle replacement
first before doing a search, and over time the use of softpinned 48b
per-ppGTT is growing (thereby eliminating any need to perform any eviction
searches, in theory at least) with the remaining users being found on
much older devices (gen2-gen6).
v2: Changelog and commentary rewritten to elaborate on the duality of a
single list being both an inactive and active list.
v3: Consolidate bool parameters into a single set of flags; don't
comment on the duality of a single variable being a multiplicity of
bits.
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/20190128102356.15037-1-chris@chris-wilson.co.uk
The majority of runtime-pm operations are bounded and scoped within a
function; these are easy to verify that the wakeref are handled
correctly. We can employ the compiler to help us, and reduce the number
of wakerefs tracked when debugging, by passing around cookies provided
by the various rpm_get functions to their rpm_put counterpart. This
makes the pairing explicit, and given the required wakeref cookie the
compiler can verify that we pass an initialised value to the rpm_put
(quite handy for double checking error paths).
For regular builds, the compiler should be able to eliminate the unused
local variables and the program growth should be minimal. Fwiw, it came
out as a net improvement as gcc was able to refactor rpm_get and
rpm_get_if_in_use together,
v2: Just s/rpm_put/rpm_put_unchecked/ everywhere, leaving the manual
mark up for smaller more targeted patches.
v3: Mention the cookie in Returns
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190114142129.24398-2-chris@chris-wilson.co.uk
The wait-for-idle used from within the shrinker_lock_uninterruptible
depends on the struct_mutex locking state being known and declared to
i915_request_wait(). As it is conceivable that we reach the vmap
notifier from underneath struct_mutex (and so keep on relying on the
mutex_trylock_recursive), we should not blindly call i915_request_wait.
In the process we can remove the dubious polling to acquire
struct_mutex, and simply act, or not, on a successful trylock.
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/20190109164204.23935-2-chris@chris-wilson.co.uk
If the current process is being killed (it was interrupted with SIGKILL
or equivalent), it will not make any progress in page allocation and we
can abort performing the shrinking on its behalf. So we can use
mutex_lock_killable() instead (although this path should only be
reachable from kswapd currently).
Tvrtko pointed out that it should also be reachable from debugfs, which
he would prefer retain its interruptiblity. As a compromise, killable is a
step in the right direction!
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/20190109164204.23935-1-chris@chris-wilson.co.uk
Ignore trying to shrink from i915 if we fail to acquire the struct_mutex
in the shrinker while performing direct-reclaim. The trade-off being
(much) lower latency for non-i915 clients at an increased risk of being
unable to obtain a page from direct-reclaim without hitting the
oom-notifier. The proviso being that we still keep trying to hard
obtain the lock for kswapd so that we can reap under heavy memory
pressure.
v2: Taint all mutexes taken within the shrinker with the struct_mutex
subclass as an early warning system, and drop I915_SHRINK_ACTIVE from
vmap to reduce the number of dangerous paths. We also have to drop
I915_SHRINK_ACTIVE from oom-notifier to be able to make the same claim
that ACTIVE is only used from outside context, which fits in with a
longer strategy of avoiding stalls due to scanning active during
shrinking.
The danger in using the subclass struct_mutex is that we declare
ourselves more knowledgable than lockdep and deprive ourselves of
automatic coverage. Instead, we require ourselves to mark up any mutex
taken inside the shrinker in order to detect lock-inversion, and if we
miss any we are doomed to a deadlock at the worst possible moment.
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/20190107115509.12523-1-chris@chris-wilson.co.uk
Add a mutex into struct i915_address_space to be used while operating on
the vma and their lists for a particular vm. As this may be called from
the shrinker, we taint the mutex with fs_reclaim so that from the start
lockdep warns us if we are caught holding the mutex across an
allocation. (With such small steps we will eventually rid ourselves of
struct_mutex recursion!)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20180711073608.20286-2-chris@chris-wilson.co.uk
Usually we have no idea about the upper bound we need to wait to catch
up with userspace when idling the device, but in a few situations we
know the system was idle beforehand and can provide a short timeout in
order to very quickly catch a failure, long before hangcheck kicks in.
In the following patches, we will use the timeout to curtain two overly
long waits, where we know we can expect the GPU to complete within a
reasonable time or declare it broken.
In particular, with a broken GPU we expect it to fail during the initial
GPU setup where do a couple of context switches to record the defaults.
This is a task that takes a few milliseconds even on the slowest of
devices, but we may have to wait 60s for hangcheck to give in and
declare the machine inoperable. In this a case where any gpu hang is
unacceptable, both from a timeliness and practical standpoint.
The other improvement is that in selftests, we do not need to arm an
independent timer to inject a wedge, as we can just limit the timeout on
the wait directly.
v2: Include the timeout parameter in the trace.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180709122044.7028-1-chris@chris-wilson.co.uk
The handling of contexts are peculiar. Instead of tieing their vma to
activity, we pin the context. This means that we cannot simply unbind
the context object itself at will (which would normally cause us to wait
for the vma to be idle), but must manually idle the GPU and retire
requests first.
A consequence of this peculiarity is when doing a last desperate attempt
to recover memory. If the memory is tied up inside active context
objects, we will fail to recover any memory simply by trying to unbind
the objects without first doing a wait-for-idle.
A side-effect of removing the call to shrinker_lock_uninterruptible()
from i915_gem_shrinker_oom() was that we removed an unlocked
wait-for-idle, and so lost the "natural" shrinkage of context objects.
By replacing that with a locked wait from inside i915_gem_shrink(), we
not only replace it with the ability to recover all context objects, but
do so for all i915_gem_shrink_all() callers.
v2: Switching requires request allocation, which is not permitted from
inside the shrinker as it only uses ordinary allocations.
References: https://bugs.freedesktop.org/show_bug.cgi?id=102936
Fixes: f2123818ff ("drm/i915: Move dev_priv->mm.[un]bound_list to its own lock")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171108094400.1386-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Prefer to defer activating our GEM shrinker until we have a few
megabytes to free; or we have accumulated sufficient mempressure by
deferring the reclaim to force a shrink. The intent is that because our
objects may typically be large, we are too effective at shrinking and
are not rewarded for freeing more pages than the batch. It will also
defer the initial shrinking to hopefully put it at a lower priority than
say the buffer cache (although it will balance out over a number of
reclaims, with GEM being more bursty).
v2: Give it a feedback system to try and tune the batch size towards
an effective size for the available objects.
v3: Start keeping track of shrinker stats in debugfs
v4: Protect against finding no shrinkable objects (div-by-zero)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171013202621.7276-7-chris@chris-wilson.co.uk
shrink_slab() allows us to report back the number of objects we
successfully scanned (out of the target shrinkctl->nr_to_scan). As
report the number of pages owned by each GEM object as a separate item
to the shrinker, we cannot precisely control the number of shrinker
objects we scan on each pass; and indeed may free more than requested.
If we fail to tell the shrinker about the number of objects we process,
it will continue to hold a grudge against us as any objects left
unscanned are added to the next reclaim -- and so we will keep on
"unfairly" shrinking our own slab in comparison to other slabs.
v2: fixup the misplaced addition, we want to count everything we scan
(to match the number we reported earlier) not just the objects we
successfully validated and freed.
References: 912d572d63 ("drm/i915: wire up shrinkctl->nr_scanned")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171013202621.7276-6-chris@chris-wilson.co.uk
Remove the struct_mutex requirement around dev_priv->mm.bound_list and
dev_priv->mm.unbound_list by giving it its own spinlock. This reduces
one more requirement for struct_mutex and in the process gives us
slightly more accurate unbound_list tracking, which should improve the
shrinker - but the drawback is that we drop the retirement before
counting so i915_gem_object_is_active() may be stale and lead us to
underestimate the number of objects that may be shrunk (see commit
bed50aea61 ("drm/i915/shrinker: Flush active on objects before
counting")).
v2: Crosslink the spinlock to the lists it protects, and btw this
changes s/obj->global_link/obj->mm.link/
v3: Fix decoupling of old links in i915_gem_object_attach_phys()
v3.1: Fix the fix, only unlink if it was linked
v3.2: Use a local for to_i915(obj->base.dev)->mm.obj_lock
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171016114037.5556-1-chris@chris-wilson.co.uk
In the next patch, we want to reduce the lock coverage within the
shrinker, and one of the dangerous walks we have is over obj->vma_list.
We are only walking the obj->vma_list in order to check whether it has
been permanently pinned by HW access, typically via use on the scanout.
But we have a couple of other long term pins, the context objects for
which we currently have to check the individual vma pin_count. If we
instead mark these using obj->pin_display, we can forgo the dangerous
and sometimes slow list iteration.
v2: Rearrange code to try and avoid confusion from false associations
due to arrangement of whitespace along with rebasing on obj->pin_global.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171013202621.7276-4-chris@chris-wilson.co.uk
Having resolved whether or not we would deadlock upon a call to
mutex_lock(&dev->struct_mutex), we can then spin for the contended
struct_mutex if we are not the owner. We cannot afford to simply block
and wait for the mutex, as the owner may itself be waiting for the
allocator -- i.e. a cyclic deadlock. This should significantly improve
the chance of running the shrinker for other processes whilst the GPU is
busy.
A more balanced approach would be to optimistically spin whilst the
mutex owner was on the cpu and there was an opportunity to acquire the
mutex for ourselves quickly. However, that requires support from
kernel/locking/ and a new mutex_spin_trylock() primitive.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170609110350.1767-4-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
In commit 5763ff04dc ("drm/i915: Avoid GPU stalls from kswapd") we
stopped direct reclaim and kswapd from triggering GPU/client stalls
whilst running (by restricting the objects they could reap to be idle).
However with abusive GPU usage, it becomes quite easy to starve kswapd
of memory and prevent it from making forward progress towards obtaining
enough free memory (thus driving the system closer to swap exhaustion).
Relax the previous restriction to allow kswapd (but not direct reclaim)
to stall the device whilst reaping purgeable pages.
v2: Also acquire the rpm wakelock to allow kswapd to unbind buffers.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170601133331.5973-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Only call synchronize_rcu_expedited after unlocking struct_mutex to
avoid deadlock because the workqueues depend on struct_mutex.
>From original patch by Andrea:
synchronize_rcu/synchronize_sched/synchronize_rcu_expedited() will
hang until its own workqueues are run. The i915 gem workqueues will
wait on the struct_mutex to be released. So we cannot wait for a
quiescent state using those rcu primitives while holding the
struct_mutex or it creates a circular lock dependency resulting in
kernel hangs (which is reproducible but goes undetected by lockdep).
kswapd0 D 0 700 2 0x00000000
Call Trace:
? __schedule+0x1a5/0x660
? schedule+0x36/0x80
? _synchronize_rcu_expedited.constprop.65+0x2ef/0x300
? wake_up_bit+0x20/0x20
? rcu_stall_kick_kthreads.part.54+0xc0/0xc0
? rcu_exp_wait_wake+0x530/0x530
? i915_gem_shrink+0x34b/0x4b0
? i915_gem_shrinker_scan+0x7c/0x90
? i915_gem_shrinker_scan+0x7c/0x90
? shrink_slab.part.61.constprop.72+0x1c1/0x3a0
? shrink_zone+0x154/0x160
? kswapd+0x40a/0x720
? kthread+0xf4/0x130
? try_to_free_pages+0x450/0x450
? kthread_create_on_node+0x40/0x40
? ret_from_fork+0x23/0x30
plasmashell D 0 4657 4614 0x00000000
Call Trace:
? __schedule+0x1a5/0x660
? schedule+0x36/0x80
? schedule_preempt_disabled+0xe/0x10
? __mutex_lock.isra.4+0x1c9/0x790
? i915_gem_close_object+0x26/0xc0
? i915_gem_close_object+0x26/0xc0
? drm_gem_object_release_handle+0x48/0x90
? drm_gem_handle_delete+0x50/0x80
? drm_ioctl+0x1fa/0x420
? drm_gem_handle_create+0x40/0x40
? pipe_write+0x391/0x410
? __vfs_write+0xc6/0x120
? do_vfs_ioctl+0x8b/0x5d0
? SyS_ioctl+0x3b/0x70
? entry_SYSCALL_64_fastpath+0x13/0x94
kworker/0:0 D 0 29186 2 0x00000000
Workqueue: events __i915_gem_free_work
Call Trace:
? __schedule+0x1a5/0x660
? schedule+0x36/0x80
? schedule_preempt_disabled+0xe/0x10
? __mutex_lock.isra.4+0x1c9/0x790
? del_timer_sync+0x44/0x50
? update_curr+0x57/0x110
? __i915_gem_free_objects+0x31/0x300
? __i915_gem_free_objects+0x31/0x300
? __i915_gem_free_work+0x2d/0x40
? process_one_work+0x13a/0x3b0
? worker_thread+0x4a/0x460
? kthread+0xf4/0x130
? process_one_work+0x3b0/0x3b0
? kthread_create_on_node+0x40/0x40
? ret_from_fork+0x23/0x30
Fixes: 3d3d18f086 ("drm/i915: Avoid rcu_barrier() from reclaim paths (shrinker)")
Reported-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Pull drm updates from Dave Airlie:
"This is the main pull request for drm for 4.10 kernel.
New drivers:
- ZTE VOU display driver (zxdrm)
- Amlogic Meson Graphic Controller GXBB/GXL/GXM SoCs (meson)
- MXSFB support (mxsfb)
Core:
- Format handling has been reworked
- Better atomic state debugging
- drm_mm leak debugging
- Atomic explicit fencing support
- fbdev helper ops
- Documentation updates
- MST fbcon fixes
Bridge:
- Silicon Image SiI8620 driver
Panel:
- Add support for new simple panels
i915:
- GVT Device model
- Better HDMI2.0 support on skylake
- More watermark fixes
- GPU idling rework for suspend/resume
- DP Audio workarounds
- Scheduler prep-work
- Opregion CADL handling
- GPU scheduler and priority boosting
amdgfx/radeon:
- Support for virtual devices
- New VM manager for non-contig VRAM buffers
- UVD powergating
- SI register header cleanup
- Cursor fixes
- Powermanagement fixes
nouveau:
- Powermangement reworks for better voltage/clock changes
- Atomic modesetting support
- Displayport Multistream (MST) support.
- GP102/104 hang and cursor fixes
- GP106 support
hisilicon:
- hibmc support (BMC chip for aarch64 servers)
armada:
- add tracing support for overlay change
- refactor plane support
- de-midlayer the driver
omapdrm:
- Timing code cleanups
rcar-du:
- R8A7792/R8A7796 support
- Misc fixes.
sunxi:
- A31 SoC display engine support
imx-drm:
- YUV format support
- Cleanup plane atomic update
mali-dp:
- Misc fixes
dw-hdmi:
- Add support for HDMI i2c master controller
tegra:
- IOMMU support fixes
- Error handling fixes
tda998x:
- Fix connector registration
- Improved robustness
- Fix infoframe/audio compliance
virtio:
- fix busid issues
- allocate more vbufs
qxl:
- misc fixes and cleanups.
vc4:
- Fragment shader threading
- ETC1 support
- VEC (tv-out) support
msm:
- A5XX GPU support
- Lots of atomic changes
tilcdc:
- Misc fixes and cleanups.
etnaviv:
- Fix dma-buf export path
- DRAW_INSTANCED support
- fix driver on i.MX6SX
exynos:
- HDMI refactoring
fsl-dcu:
- fbdev changes"
* tag 'drm-for-v4.10' of git://people.freedesktop.org/~airlied/linux: (1343 commits)
drm/nouveau/kms/nv50: fix atomic regression on original G80
drm/nouveau/bl: Do not register interface if Apple GMUX detected
drm/nouveau/bl: Assign different names to interfaces
drm/nouveau/bios/dp: fix handling of LevelEntryTableIndex on DP table 4.2
drm/nouveau/ltc: protect clearing of comptags with mutex
drm/nouveau/gr/gf100-: handle GPC/TPC/MPC trap
drm/nouveau/core: recognise GP106 chipset
drm/nouveau/ttm: wait for bo fence to signal before unmapping vmas
drm/nouveau/gr/gf100-: FECS intr handling is not relevant on proprietary ucode
drm/nouveau/gr/gf100-: properly ack all FECS error interrupts
drm/nouveau/fifo/gf100-: recover from host mmu faults
drm: Add fake controlD* symlinks for backwards compat
drm/vc4: Don't use drm_put_dev
drm/vc4: Document VEC DT binding
drm/vc4: Add support for the VEC (Video Encoder) IP
drm: Add TV connector states to drm_connector_state
drm: Turn DRM_MODE_SUBCONNECTOR_xx definitions into an enum
drm/vc4: Fix ->clock_select setting for the VEC encoder
drm/amdgpu/dce6: Set MASTER_UPDATE_MODE to 0 in resume_mc_access as well
drm/amdgpu: use pin rather than pin_restricted in a few cases
...
The shrinker may appear to recurse into obj->mm.lock as the shrinker may
be called from a direct reclaim path whilst handling get_pages. We
filter out recursing on the same obj->mm.lock by inspecting
obj->mm.pages, but we do want to take the lock on a second object in
order to reap their pages. lockdep spots the recursion on the same
lockclass and needs annotation to avoid a false positive. To keep the
two paths distinct, create an enum to indicate which subclass of
obj->mm.lock we are using. This removes the false positive and avoids
masking real bugs.
Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161101121134.27504-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
During shrinking, we walk over the list of objects searching for
victims. Any that are not removed are put back into the global list.
Currently, they are put back in order (at the front) which means they
will be first to be scanned again. If we instead move them to the rear
of the list, we will scan new potential victims on the next pass and
waste less time rescanning unshrinkable objects. Normally the lists are
kept in rough order to shrinking (with object least frequently used at
the start), by moving just scanned objects to the rear we are
acknowledging that they are still in use.
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/20161101084843.3961-3-chris@chris-wilson.co.uk
We want to hide the latency of releasing objects and their backing
storage from the submission, so we move the actual free to a worker.
This allows us to switch to struct_mutex freeing of the object in the
next patch.
Furthermore, if we know that the object we are dereferencing remains valid
for the duration of our access, we can forgo the usual synchronisation
barriers and atomic reference counting. To ensure this we defer freeing
an object til after an RCU grace period, such that any lookup of the
object within an RCU read critical section will remain valid until
after we exit that critical section. We also employ this delay for
rate-limiting the serialisation on reallocation - we have to slow down
object creation in order to prevent resource starvation (in particular,
files).
v2: Return early in i915_gem_tiling() ioctl to skip over superfluous
work on error.
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/20161028125858.23563-19-chris@chris-wilson.co.uk
Break the allocation of the backing storage away from struct_mutex into
a per-object lock. This allows parallel page allocation, provided we can
do so outside of struct_mutex (i.e. set-domain-ioctl, pwrite, GTT
fault), i.e. before execbuf! The increased cost of the atomic counters
are hidden behind i915_vma_pin() for the typical case of execbuf, i.e.
as the object is typically bound between execbufs, the page_pin_count is
static. The cost will be felt around set-domain and pwrite, but offset
by the improvement from reduced struct_mutex contention.
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/20161028125858.23563-14-chris@chris-wilson.co.uk
In the next patch we want to handle reset directly by a locked waiter in
order to avoid issues with returning before the reset is handled. To
handle the reset, we must first know whether we hold the struct_mutex.
If we do not hold the struct_mtuex we can not perform the reset, but we do
not block the reset worker either (and so we can just continue to wait for
request completion) - otherwise we must relinquish the mutex.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160909131201.16673-10-chris@chris-wilson.co.uk
