We can now wait for the GPU (all engines) to become idle without
requiring the struct_mutex. Inside the shrinker, we need to currently
take the struct_mutex in order to purge objects and to purge the objects
we need the GPU to be idle - causing a stall whilst we hold the
struct_mutex. We can hide most of that stall by performing the wait
before taking the struct_mutex and only doing essential waits for
new rendering on objects to be freed.
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/1470388464-28458-8-git-send-email-chris@chris-wilson.co.uk
The principal motivation for this was to try and eliminate the
struct_mutex from i915_gem_suspend - but we still need to hold the mutex
current for the i915_gem_context_lost(). (The issue there is that there
may be an indirect lockdep cycle between cpu_hotplug (i.e. suspend) and
struct_mutex via the stop_machine().) For the moment, enabling last
request tracking for the engine, allows us to do busyness checking and
waiting without requiring the struct_mutex - which is useful in its own
right.
As a side-effect of having a robust means for tracking engine busyness,
we can replace our other busyness heuristic, that of comparing against
the last submitted seqno. For paranoid reasons, we have a semi-ordered
check of that seqno inside the hangchecker, which we can now improve to
an ordered check of the engine's busyness (removing a locked xchg in the
process).
v2: Pass along "bool interruptible" as being unlocked we cannot rely on
i915->mm.interruptible being stable or even under our control.
v3: Replace check Ironlake i915_gpu_busy() with the common precalculated value
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/1470388464-28458-6-git-send-email-chris@chris-wilson.co.uk
If we enable RCU for the requests (providing a grace period where we can
inspect a "dead" request before it is freed), we can allow callers to
carefully perform lockless lookup of an active request.
However, by enabling deferred freeing of requests, we can potentially
hog a lot of memory when dealing with tens of thousands of requests per
second - with a quick insertion of a synchronize_rcu() inside our
shrinker callback, that issue disappears.
v2: Currently, it is our responsibility to handle reclaim i.e. to avoid
hogging memory with the delayed slab frees. At the moment, we wait for a
grace period in the shrinker, and block for all RCU callbacks on oom.
Suggested alternatives focus on flushing our RCU callback when we have a
certain number of outstanding request frees, and blocking on that flush
after a second high watermark. (So rather than wait for the system to
run out of memory, we stop issuing requests - both are nondeterministic.)
Paul E. McKenney wrote:
Another approach is synchronize_rcu() after some largish number of
requests. The advantage of this approach is that it throttles the
production of callbacks at the source. The corresponding disadvantage
is that it slows things up.
Another approach is to use call_rcu(), but if the previous call_rcu()
is still in flight, block waiting for it. Yet another approach is
the get_state_synchronize_rcu() / cond_synchronize_rcu() pair. The
idea is to do something like this:
cond_synchronize_rcu(cookie);
cookie = get_state_synchronize_rcu();
You would of course do an initial get_state_synchronize_rcu() to
get things going. This would not block unless there was less than
one grace period's worth of time between invocations. But this
assumes a busy system, where there is almost always a grace period
in flight. But you can make that happen as follows:
cond_synchronize_rcu(cookie);
cookie = get_state_synchronize_rcu();
call_rcu(&my_rcu_head, noop_function);
Note that you need additional code to make sure that the old callback
has completed before doing a new one. Setting and clearing a flag
with appropriate memory ordering control suffices (e.g,. smp_load_acquire()
and smp_store_release()).
v3: More comments on compiler and processor order of operations within
the RCU lookup and discover we can use rcu_access_pointer() here instead.
v4: Wrap i915_gem_active_get_rcu() to take the rcu_read_lock itself.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: "Goel, Akash" <akash.goel@intel.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1470324762-2545-25-git-send-email-chris@chris-wilson.co.uk
We are motivated to avoid using a bitfield for obj->active for a couple
of reasons. Firstly, we wish to document our lockless read of obj->active
using READ_ONCE inside i915_gem_busy_ioctl() and that requires an
integral type (i.e. not a bitfield). Secondly, gcc produces abysmal code
when presented with a bitfield and that shows up high on the profiles of
request tracking (mainly due to excess memory traffic as it converts
the bitfield to a register and back and generates frequent AGI in the
process).
v2: BIT, break up a long line in compute the other engines, new paint
for i915_gem_object_is_active (now i915_gem_object_get_active).
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/1470324762-2545-23-git-send-email-chris@chris-wilson.co.uk
When we call i915_vma_unbind(), we will wait upon outstanding rendering.
This will also trigger a retirement phase, which may update the object
lists. If, we extend request tracking to the VMA itself (rather than
keep it at the encompassing object), then there is a potential that the
obj->vma_list be modified for other elements upon i915_vma_unbind(). As
a result, if we walk over the object list and call i915_vma_unbind(), we
need to be prepared for that list to change.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1470293567-10811-8-git-send-email-chris@chris-wilson.co.uk
Since drm_i915_private is now a subclass of drm_device we do not need to
chase the drm_i915_private->dev backpointer and can instead simply
access drm_i915_private->drm directly.
text data bss dec hex filename
1068757 4565 416 1073738 10624a drivers/gpu/drm/i915/i915.ko
1066949 4565 416 1071930 105b3a drivers/gpu/drm/i915/i915.ko
Created by the coccinelle script:
@@
struct drm_i915_private *d;
identifier i;
@@
(
- d->dev->i
+ d->drm.i
|
- d->dev
+ &d->drm
)
and for good measure the dev_priv->dev backpointer was removed entirely.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1467711623-2905-4-git-send-email-chris@chris-wilson.co.uk
By tracking the iomapping on the VMA itself, we can share that area
between multiple users. Also by only revoking the iomapping upon
unbinding from the mappable portion of the GGTT, we can keep that iomap
across multiple invocations (e.g. execlists context pinning).
Note that by moving the iounnmap tracking to the VMA, we actually end up
fixing a leak of the iomapping in intel_fbdev.
v1.5: Rebase prompted by Tvrtko
v2: Drop dev_priv parameter, we can recover the i915_ggtt from the vma.
v3: Move handling of ioremap space exhaustion to vmap_purge and also
allow vmallocs to recover old iomaps. Add Tvrtko's kerneldoc.
v4: Fix a use-after-free in shrinker and rearrange i915_vma_iomap
v5: Back to i915_vm_to_ggtt
v6: Use i915_vma_pin_iomap and i915_vma_unpin_iomap to mark critical
sections and ensure the VMA cannot be reaped whilst mapped.
v7: Move i915_vma_iounmap so that consumers of the API are not tempted,
and add iomem annotations
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>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1461833819-3991-5-git-send-email-chris@chris-wilson.co.uk
Elsewhere we have adopted the convention of using '_link' to denote
elements in the list (and '_list' for the actual list_head itself), and
that the name should indicate which list the link belongs to (and
preferrably not just where the link is being stored).
s/vma_link/obj_link/ (we iterate over obj->vma_list)
s/mm_list/vm_link/ (we iterate over vm->[in]active_list)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
If the system has no available swap pages, we cannot make forward
progress in the shrinker by releasing active pages, only by releasing
purgeable pages which are immediately reaped. Take total_swap_pages into
account when counting up available objects to be shrunk and subsequently
shrinking them. By doing so, we avoid unbinding objects that cannot be
shrunk and so wasting CPU cycles flushing those objects from the GPU to
the system and then immediately back again (as they will more than
likely be reused shortly after).
Based on a patch by Akash Goel.
v2: frontswap registers extra swap pages available for the system, so it
is already include in the count of available swap pages.
v3: Use get_nr_swap_pages() to query the currently available amount of
swap space. This should also stop us from shrinking the GPU buffers if
we ever run out of swap space. Though at that point, we would expect the
oom-notifier to be running and failing miserably...
Reported-by: Akash Goel <akash.goel@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: linux-mm@kvack.org
Cc: Akash Goel <akash.goel@intel.com>
Cc: sourab.gupta@intel.com
Link: http://patchwork.freedesktop.org/patch/msgid/1449244734-25733-2-git-send-email-chris@chris-wilson.co.uk
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Exclude active GPU pages from the purview of the background shrinker
(kswapd), as these cause uncontrollable GPU stalls. Given that the
shrinker is rerun until the freelists are satisfied, we should have
opportunity in subsequent passes to recover the pages once idle. If the
machine does run out of memory entirely, we have the forced idling in the
oom-notifier as a means of releasing all the pages we can before an oom
is prematurely executed.
Note that this relies upon an up-front retire_requests to keep the
inactive list in shape, which was added in a previous patch, mostly as
execlist ctx pinning band-aids.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Add note about retire_requests.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We can forgo an evict-everything here as the shrinker operation itself
will unbind any vma as required. If we explicitly idle the GPU through a
switch to the default context, we not only create a request in an
illegal context (e.g. whilst shrinking during execbuf with a request
already allocated), but switching to the default context will not free
up the memory backing the active contexts - unless in the unlikely
situation that context had already been closed (and just kept arrive by
being the current context). The saving is near zero and the danger real.
To compensate for the loss of the forced retire, add a couple of
retire-requests to i915_gem_shirnk() - this should help free up any
transitive cache from the requests.
Note that the second retire_requests is for the benefit of the
hand-rolled execlist ctx active tracking: We need to manually kick
requests to get those unpinned again. Once that's fixed we can try to
remove this again.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add summary of why we need a pile of retire_requests.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Often it is very useful to know why we suddenly purge vast tracts of
memory and surprisingly up until now we didn't even have a tracepoint
for when we shrink our memory.
Note that there are slab_start/end tracepoints already, but those
don't cover the internal recursion when we directly call into our
shrinker code. Hence a separate tracepoint seems justified. Also note
that we don't really need a separate tracepoint for the actual amount
of pages freed since we already have an unbind tracpoint for that.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add a note that there's also slab_start/end and why they're
insufficient.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
As the shrinker_control now passes us unsigned long targets, update our
shrinker functions to match.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
I've botched this, so let's fix it.
Botched in
commit eb0b44adc0
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Wed Mar 18 14:47:59 2015 +0100
drm/i915: kerneldoc for i915_gem_shrinker.c
v2: Be a good citizen^Wmaintainer and add the proper commit citation.
Noticed by Jani.
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
When looking for viable candidates to shrink, we only want objects that
are not pinned. However to do so we performed a double iteration over
the vma in the objects, first looking for the pin-count, then looking
for allocations. We can do both at once and be slightly more explicit in
our validity test.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
And remove one bogus * from i915_gem_gtt.c since that's not a
kerneldoc there.
v2: Review from Chris:
- Clarify memory space to better distinguish from address space.
- Add note that shrink doesn't guarantee the freed memory and that
users must fall back to shrink_all.
- Explain how pinning ties in with eviction/shrinker.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Two code changes:
- Extract i915_gem_shrinker_init.
- Inline i915_gem_object_is_purgeable since we open-code it everywhere
else too.
This already has the benefit of pulling all the shrinker code
together, next patch adds a bit of kerneldoc.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
