Apply backpressure to hogs that emit requests faster than the GPU can
process them by waiting for their ring to be less than half-full before
proceeding with taking the struct_mutex.
This is a gross hack to apply throttling backpressure, the long term
goal is to remove the struct_mutex contention so that each client
naturally waits, preferably in an asynchronous, nonblocking fashion
(pipelined operations for the win), for their own resources and never
blocks another client within the driver at least. (Realtime priority
goals would extend to ensuring that resource contention favours high
priority clients as well.)
This patch only limits excessive request production and does not attempt
to throttle clients that block waiting for eviction (either global GTT or
system memory) or any other global resources, see above for the long term
goal.
No microbenchmarks are harmed (to the best of my knowledge).
Testcase: igt/gem_exec_schedule/pi-ringfull-*
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190207071829.5574-1-chris@chris-wilson.co.uk
A few years ago, see commit 688e6c7258 ("drm/i915: Slaughter the
thundering i915_wait_request herd"), the issue of handling multiple
clients waiting in parallel was brought to our attention. The
requirement was that every client should be woken immediately upon its
request being signaled, without incurring any cpu overhead.
To handle certain fragility of our hw meant that we could not do a
simple check inside the irq handler (some generations required almost
unbounded delays before we could be sure of seqno coherency) and so
request completion checking required delegation.
Before commit 688e6c7258, the solution was simple. Every client
waiting on a request would be woken on every interrupt and each would do
a heavyweight check to see if their request was complete. Commit
688e6c7258 introduced an rbtree so that only the earliest waiter on
the global timeline would woken, and would wake the next and so on.
(Along with various complications to handle requests being reordered
along the global timeline, and also a requirement for kthread to provide
a delegate for fence signaling that had no process context.)
The global rbtree depends on knowing the execution timeline (and global
seqno). Without knowing that order, we must instead check all contexts
queued to the HW to see which may have advanced. We trim that list by
only checking queued contexts that are being waited on, but still we
keep a list of all active contexts and their active signalers that we
inspect from inside the irq handler. By moving the waiters onto the fence
signal list, we can combine the client wakeup with the dma_fence
signaling (a dramatic reduction in complexity, but does require the HW
being coherent, the seqno must be visible from the cpu before the
interrupt is raised - we keep a timer backup just in case).
Having previously fixed all the issues with irq-seqno serialisation (by
inserting delays onto the GPU after each request instead of random delays
on the CPU after each interrupt), we can rely on the seqno state to
perfom direct wakeups from the interrupt handler. This allows us to
preserve our single context switch behaviour of the current routine,
with the only downside that we lose the RT priority sorting of wakeups.
In general, direct wakeup latency of multiple clients is about the same
(about 10% better in most cases) with a reduction in total CPU time spent
in the waiter (about 20-50% depending on gen). Average herd behaviour is
improved, but at the cost of not delegating wakeups on task_prio.
v2: Capture fence signaling state for error state and add comments to
warm even the most cold of hearts.
v3: Check if the request is still active before busywaiting
v4: Reduce the amount of pointer misdirection with list_for_each_safe
and using a local i915_request variable inside the loops
v5: Add a missing pluralisation to a purely informative selftest message.
References: 688e6c7258 ("drm/i915: Slaughter the thundering i915_wait_request herd")
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/20190129205230.19056-2-chris@chris-wilson.co.uk
To allow requests to forgo a common execution timeline, one question we
need to be able to answer is "is this request running?". To track
whether a request has started on HW, we can emit a breadcrumb at the
beginning of the request and check its timeline's HWSP to see if the
breadcrumb has advanced past the start of this request. (This is in
contrast to the global timeline where we need only ask if we are on the
global timeline and if the timeline has advanced past the end of the
previous request.)
There is still confusion from a preempted request, which has already
started but relinquished the HW to a high priority request. For the
common case, this discrepancy should be negligible. However, for
identification of hung requests, knowing which one was running at the
time of the hang will be much more important.
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/20190129185452.20989-2-chris@chris-wilson.co.uk
Now that we have allocated ourselves a cacheline to store a breadcrumb,
we can emit a write from the GPU into the timeline's HWSP of the
per-context seqno as we complete each request. This drops the mirroring
of the per-engine HWSP and allows each context to operate independently.
We do not need to unwind the per-context timeline, and so requests are
always consistent with the timeline breadcrumb, greatly simplifying the
completion checks as we no longer need to be concerned about the
global_seqno changing mid check.
One complication though is that we have to be wary that the request may
outlive the HWSP and so avoid touching the potentially danging pointer
after we have retired the fence. We also have to guard our access of the
HWSP with RCU, the release of the obj->mm.pages should already be RCU-safe.
At this point, we are emitting both per-context and global seqno and
still using the single per-engine execution timeline for resolving
interrupts.
v2: s/fake_complete/mark_complete/
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/20190128181812.22804-5-chris@chris-wilson.co.uk
Allocate a page for use as a status page by a group of timelines, as we
only need a dword of storage for each (rounded up to the cacheline for
safety) we can pack multiple timelines into the same page. Each timeline
will then be able to track its own HW seqno.
v2: Reuse the common per-engine HWSP for the solitary ringbuffer
timeline, so that we do not have to emit (using per-gen specialised
vfuncs) the breadcrumb into the distinct timeline HWSP and instead can
keep on using the common MI_STORE_DWORD_INDEX. However, to maintain the
sleight-of-hand for the global/per-context seqno switchover, we will
store both temporarily (and so use a custom offset for the shared timeline
HWSP until the switch over).
v3: Keep things simple and allocate a page for each timeline, page
sharing comes next.
v4: I was caught repeating the same MI_STORE_DWORD_IMM over and over
again in selftests.
v5: And caught red handed copying create timeline + check.
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/20190128181812.22804-3-chris@chris-wilson.co.uk
Currently we only allocate an object and vma if we are using a GGTT
virtual HWSP, and a plain struct page for a physical HWSP. For
convenience later on with global timelines, it will be useful to always
have the status page being tracked by a struct i915_vma. Make it so.
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/20190128102356.15037-4-chris@chris-wilson.co.uk
Now that the submission backends are controlled via their own spinlocks,
with a wave of a magic wand we can lift the struct_mutex requirement
around GPU reset. That is we allow the submission frontend (userspace)
to keep on submitting while we process the GPU reset as we can suspend
the backend independently.
The major change is around the backoff/handoff strategy for performing
the reset. With no mutex deadlock, we no longer have to coordinate with
any waiter, and just perform the reset immediately.
Testcase: igt/gem_mmap_gtt/hang # regresses
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125132230.22221-3-chris@chris-wilson.co.uk
Instead of tediously and fragilely counting up the number of dwords
required to emit the breadcrumb to seal a request, fake a request and
measure it automatically once during engine setup.
The downside is that this requires a fair amount of mocking to create a
proper breadcrumb. Still, should be less error prone in future as the
breadcrumb size fluctuates!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125100520.20163-1-chris@chris-wilson.co.uk
Needs just a few additional includes here and there.
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190108082709.3748-1-jani.nikula@intel.com
Haswell also requires the RING_IMR flush for its unique vebox setup to
avoid losing interrupts, as per 476af9c260 ("drm/i915/gen6: Flush
RING_IMR changes before changing the global GT IMR"):
On Baytail, notably, we can still detect missed interrupt syndrome
(where we never spot a completed request). In this case, it can be
alleviated by always keeping the interrupt unmasked, implying that the
interrupt is being lost in the window after modifying the IMR. (This is
the reason we still have the posting reads on enable_irq, if we remove
them we miss interrupts!) Having narrowed the issue down to the IMR,
rather than keeping it always enabled, applying the usual posting
read/flush of the RING_IMR before unmasking the GT IMR also seems to
prevent the missed interrupt. So be it.
References: 476af9c260 ("drm/i915/gen6: Flush RING_IMR changes before changing the global GT IMR")
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/20190105115647.4970-1-chris@chris-wilson.co.uk
On Baytail, notably, we can still detect missed interrupt syndrome
(where we never spot a completed request). In this case, it can be
alleviated by always keeping the interrupt unmasked, implying that the
interrupt is being lost in the window after modifying the IMR. (This is
the reason we still have the posting reads on enable_irq, if we remove
them we miss interrupts!) Having narrowed the issue down to the IMR,
rather than keeping it always enabled, applying the usual posting
read/flush of the RING_IMR before unmasking the GT IMR also seems to
prevent the missed interrupt. So be it.
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/20190102163524.19353-1-chris@chris-wilson.co.uk
Now that we have eliminated the CPU-side irq_seqno_barrier by moving the
delays on the GPU before emitting the MI_USER_INTERRUPT, we can remove
the engine->irq_seqno_barrier infrastructure. Though intentionally
slowing down the GPU is nasty, so is the code we can now remove!
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/20181228171641.16531-6-chris@chris-wilson.co.uk
The irq_seqno_barrier is a tradeoff between doing work on every request
(on the GPU) and doing work after every interrupt (on the CPU). We
presume we have many more requests than interrupts! However, for
Ironlake, the workaround is a pretty hideous usleep() and so even though
it was found we need to repeat the MI_STORE_DWORD_IMM 8 times, or about
1us of GPU time, doing so is preferrable than requiring a sleep of
125-250us on the CPU where we desire to respond immediately (ideally from
within the interrupt handler)!
The additional MI_STORE_DWORD_IMM also have the side-effect of flushing
MI operations from userspace which are not caught by MI_FLUSH!
Testcase: igt/gem_sync
Testcase: igt/gem_exec_whisper
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/20181228171641.16531-5-chris@chris-wilson.co.uk
The irq_seqno_barrier is a tradeoff between doing work on every request
(on the GPU) and doing work after every interrupt (on the CPU). We
presume we have many more requests than interrupts! However, the current
w/a for Ivybridge is an implicit delay that currently fails sporadically
and consistently if we move the w/a into the irq handler itself. This
makes the CPU barrier untenable for upcoming interrupt handler changes
and so we need to replace it with a delay on the GPU before we send the
MI_USER_INTERRUPT. As it turns out that delay is 32x MI_STORE_DWORD_IMM,
or about 0.6us per request! Quite nasty, but the lesser of two evils
looking to the future.
Testcase: igt/gem_sync
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/20181228171641.16531-4-chris@chris-wilson.co.uk
The MI_FLUSH_DW does appear coherent with the following
MI_USER_INTERRUPT, but only on Sandybridge. Ivybridge requires a heavier
hammer, but on Sandybridge we can stop requiring the irq_seqno barrier.
Testcase: igt/gem_sync
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/20181228171641.16531-3-chris@chris-wilson.co.uk
Having transitioned to using PIPECONTROL to combine the flush with the
breadcrumb write using their post-sync functions, assume that this will
resolve the serialisation with the subsequent MI_USER_INTERRUPT. That is
when inspecting the breadcrumb after an interrupt we can rely on the write
being posted (i.e. the HWSP will be coherent).
Testing using gem_sync shows that the PIPECONTROL + CS stall does
serialise the command streamer sufficient that the breadcrumb lands
before the MI_USER_INTERRUPT. The same is not true for MI_FLUSH_DW.
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/20181228171641.16531-2-chris@chris-wilson.co.uk
Now that we perform the request flushing inline with emitting the
breadcrumb, we can remove the now redundant manual flush. And we can
also remove the infrastructure that remained only for its purpose.
v2: emit_breadcrumb_sz is in dwords, but rq->reserved_space is in bytes
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/20181228171641.16531-1-chris@chris-wilson.co.uk
In preparation for removing the manual EMIT_FLUSH prior to emitting the
breadcrumb implement the flush inline with writing the breadcrumb for
ringbuffer emission.
With a combined flush+breadcrumb, we can use a single operation to both
flush and after the flush is complete (post-sync) write the breadcrumb.
This gives us a strongly ordered operation that should be sufficient to
serialise the write before we emit the interrupt; and therefore we may
take the opportunity to remove the irq_seqno_barrier w/a for gen6+.
Although using the PIPECONTROL to write the breadcrumb is slower than
MI_STORE_DWORD_IMM, by combining the operations into one and removing the
extra flush (next patch) it is faster
For gen2-5, we simply combine the MI_FLUSH into the breadcrumb emission,
though maybe we could find a solution here to the seqno-vs-interrupt
issue on Ironlake by mixing up the flush? The answer is no, adding an
MI_FLUSH before the interrupt is insufficient.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181228153114.4948-2-chris@chris-wilson.co.uk
The writing is on the wall for the existence of a single execution queue
along each engine, and as a consequence we will not be able to track
dependencies along the HW queue itself, i.e. we will not be able to use
HW semaphores on gen7 as they use a global set of registers (and unlike
gen8+ we can not effectively target memory to keep per-context seqno and
dependencies).
On the positive side, when we implement request reordering for gen7 we
also can not presume a simple execution queue and would also require
removing the current semaphore generation code. So this bring us another
step closer to request reordering for ringbuffer submission!
The negative side is that using interrupts to drive inter-engine
synchronisation is much slower (4us -> 15us to do a nop on each of the 3
engines on ivb). This is much better than it was at the time of introducing
the HW semaphores and equally important userspace weaned itself off
intermixing dependent BLT/RENDER operations (the prime culprit was glyph
rendering in UXA). So while we regress the microbenchmarks, it should not
impact the user.
References: https://bugs.freedesktop.org/show_bug.cgi?id=108888
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181228140736.32606-2-chris@chris-wilson.co.uk
After we found a workaround for a hang on context load, Ben Widawsky
found confirmation that it was for an issue with waking from rc6 and
loading a context image.
The workaround from on high suggests that we should
I915_WRITE(RING_WAIT_FOR_RC6_EXIT(engine->mmio_base),
_MASKED_FIELD(RING_RC6_SEL_WRITE_ADDR_MASK,
RING_RC6_SEL_WRITE_ADDR_UPPER_LEFT));
in our rc6 setup for Haswell GT1, but on applying that we find instead
that the machine encounters a GT forcewake error and locks up.
As we are removing HW semaphore usage in the next patch, and the
suggested workaround is no improvement, we need to
decouple the PSMI workaround from HAS_SEMAPHORES to IS_HSW_GT1.
References: 2c55018347 ("drm/i915: Disable PSMI sleep messages on all rings around context switches")
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/20181228140736.32606-1-chris@chris-wilson.co.uk
Having completed a test run of gem_eio across all machines in CI we also
observe the phenomenon (of lost interrupts after resetting the GPU) on
gen3 machines as well as the previously sighted gen6/gen7. Let's apply
the same HWSTAM workaround that was effective for gen6+ for all, as
although we haven't seen the same failure on gen4/5 it seems prudent to
keep the code the same.
As a consequence we can remove the extra setting of HWSTAM and apply the
register from a single site.
v2: Delazy and move the HWSTAM into its own function
v3: Mask off all HWSP writes on driver unload and engine cleanup.
v4: And what about the physical hwsp?
v5: No, engine->init_hw() is not called from driver_init_hw(), don't be
daft. Really scrub HWSTAM as early as we can in driver_init_mmio()
v6: Rename set_hwsp as it was setting the mask not the hwsp register.
v7: Ville pointed out that although vcs(bsd) was introduced for g4x/ilk,
per-engine HWSTAM was not introduced until gen6!
References: https://bugs.freedesktop.org/show_bug.cgi?id=108735
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181218102712.11058-1-chris@chris-wilson.co.uk
Define IS_GEN() similarly to our IS_GEN_RANGE(). but use gen instead of
gen_mask to do the comparison. Now callers can pass then gen as a parameter,
so we don't require one macro for each gen.
The following spatch was used to convert the users of these macros:
@@
expression e;
@@
(
- IS_GEN2(e)
+ IS_GEN(e, 2)
|
- IS_GEN3(e)
+ IS_GEN(e, 3)
|
- IS_GEN4(e)
+ IS_GEN(e, 4)
|
- IS_GEN5(e)
+ IS_GEN(e, 5)
|
- IS_GEN6(e)
+ IS_GEN(e, 6)
|
- IS_GEN7(e)
+ IS_GEN(e, 7)
|
- IS_GEN8(e)
+ IS_GEN(e, 8)
|
- IS_GEN9(e)
+ IS_GEN(e, 9)
|
- IS_GEN10(e)
+ IS_GEN(e, 10)
|
- IS_GEN11(e)
+ IS_GEN(e, 11)
)
v2: use IS_GEN rather than GT_GEN and compare to info.gen rather than
using the bitmask
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181212181044.15886-2-lucas.demarchi@intel.com
RANGE makes it longer, but clearer. We are also going to add a macro to
check an individual gen, so add the _RANGE prefix here.
Diff generated with:
sed 's/IS_GEN(/IS_GEN_RANGE(/g' drivers/gpu/drm/i915/{*/,}*.{c,h} -i
v2: use IS_GEN rather than GT_GEN
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181212181044.15886-1-lucas.demarchi@intel.com
Adding an extra MI_STORE_DWORD_IMM to the gpu relocation path for gen3
was good, but still not good enough. To survive 24+ hours under test we
needed to perform not one, not two but three extra store-dw. Doing so
for each GPU relocation was a little unsightly and since we need to
worry about userspace hitting the same issues, we should apply the dummy
store-dw into the EMIT_FLUSH.
Fixes: 7dd4f6729f ("drm/i915: Async GPU relocation processing")
References: 7fa28e1469 ("drm/i915: Write GPU relocs harder with gen3")
Testcase: igt/gem_tiled_fence_blits # blb/pnv
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181207134037.11848-1-chris@chris-wilson.co.uk
Currently we allocate a scratch page for each engine, but since we only
ever write into it for post-sync operations, it is not exposed to
userspace nor do we care for coherency. As we then do not care about its
contents, we can use one page for all, reducing our allocations and
avoid complications by not assuming per-engine isolation.
For later use, it simplifies engine initialisation (by removing the
allocation that required struct_mutex!) and means that we can always rely
on there being a scratch page.
v2: Check that we allocated a large enough scratch for I830 w/a
Fixes: 06e562e7f515 ("drm/i915/ringbuffer: Delay after EMIT_INVALIDATE for gen4/gen5") # v4.18.20
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108850
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@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/20181204141522.13640-1-chris@chris-wilson.co.uk
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: <stable@vger.kernel.org> # v4.18.20+
Convert the per context workaround handling code to run against the newly
introduced common workaround framework and fuse the two to use the
existing smarter list add helper, the one which does the sorted insert and
merges registers where possible.
This completes migration of all four classes of workarounds onto the
common framework.
Existing macros are kept untouched for smaller code churn.
v2:
* Rename to list name ctx_wa_list and move from dev_priv to engine.
v3:
* API rename and parameters tweaking. (Chris Wilson)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20181203133357.10341-1-tvrtko.ursulin@linux.intel.com
Exercising the gpu reloc path strenuously revealed an issue where the
updated relocations (from MI_STORE_DWORD_IMM) were not being observed
upon execution. After some experiments with adding pipecontrols (a lot
of pipecontrols (32) as gen4/5 do not have a bit to wait on earlier pipe
controls or even the current on), it was discovered that we merely
needed to delay the EMIT_INVALIDATE by several flushes. It is important
to note that it is the EMIT_INVALIDATE as opposed to the EMIT_FLUSH that
needs the delay as opposed to what one might first expect -- that the
delay is required for the TLB invalidation to take effect (one presumes
to purge any CS buffers) as opposed to a delay after flushing to ensure
the writes have landed before triggering invalidation.
Testcase: igt/gem_tiled_fence_blits
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181105094305.5767-1-chris@chris-wilson.co.uk
We mix hexa- and decimal which is confusing when reading the logs. So make
the single odd one out instance decimal for consistency.
v2:
* Do the intel_ringbuffer.c as well. (Chris Wilson)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20180926145033.16318-1-tvrtko.ursulin@linux.intel.com
Continuing the fun of trying to find exactly the delay that is
sufficient to ensure that the page directory is fully loaded between
context switches, move the extra flush added in commit 70b73f9ac1
("drm/i915/ringbuffer: Delay after invalidating gen6+ xcs") to just
after we flush the pd. Entirely based on the empirical data of running
failing tests in a loop until we survive a day (before the mtbf is 10-30
minutes).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107769
References: 70b73f9ac1 ("drm/i915/ringbuffer: Delay after invalidating gen6+ xcs")
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/20180904063802.13880-1-chris@chris-wilson.co.uk
Older gen use a physical address for the hardware status page, for which
we use cache-coherent writes. As the writes are into the cpu cache, we use
a normal WB mapped page to read the HWS, used for our seqno tracking.
Anecdotally, I observed lost breadcrumbs writes into the HWS on i965gm,
which so far have not reoccurred with this patch. How reliable that
evidence is remains to be seen.
v2: Explicitly pass the expected physical address to the hw
v3: Also remember the wild writes we once had for HWS above 4G.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20180903152304.31589-2-chris@chris-wilson.co.uk
During stress testing of full-ppgtt (on Baytrail at least), we found
that the invalidation around a context/mm switch was insufficient (writes
would go astray). Adding a second MI_FLUSH_DW barrier prevents this, but
it is unclear as to whether this is merely a delaying tactic or if it is
truly serialising with the TLB invalidation. Either way, it is
empirically required.
v2: Avoid the loop for readability;
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107715
References: https://bugs.freedesktop.org/show_bug.cgi?id=107759
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: Matthew Auld <matthew.william.auld@gmail.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180830161042.29193-1-chris@chris-wilson.co.uk
This reapplies commit 39f3be162c ("drm/i915: Kick waiters on resetting
legacy rings") after the improved gem_eio was run across all machines we
found that gen3 and early gen4 still lost the immediate interrupt
following reset, and the HWSTAM w/a applied to gen6+ is inadequate.
Unlike the later gen, on gen3/4 the principle (and only tests to fail so
far) are the wait vs reset test cases, whereas the reset stress case
works fine (which was the predominantly failing case for gen6+). That is
enough to suggest the underlying issue is sufficiently different to
support the difference in HWSTAM efficacy.
Testcase: igt/gem_eio/wait-10ms
References: 39f3be162c ("drm/i915: Kick waiters on resetting legacy rings")
References: a69ab52b03 ("drm/i915: Remove extra waiter kick on legacy resets")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@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/20180814104056.27001-1-chris@chris-wilson.co.uk
Now with a more efficacious workaround for the lost interrupts after
reset, we can remove the hack of kicking the waiters after reset. The
issue was that the kick only worked for the immediate window after the
reset (those seqno that would complete in the time it took for the
waiter thread to perform its check) but miss any seqno that lacked an
interrupt afterwards.
References: 39f3be162c ("drm/i915: Kick waiters on resetting legacy rings")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180808105101.913-3-chris@chris-wilson.co.uk
An oddity occurs on Sandybridge, Ivybridge and Haswell (and presumably
Valleyview) in that for the period following the GPU restart after a
reset, there are no GT interrupts received. From Ville's notes, bit 0 in
the HWSTAM corresponds to the render interrupt, and if we unmask it we
do see immediate resumption of GT interrupt delivery (via the master irq
handler) after the reset.
v2: Limit the w/a to the render interrupt from rcs
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107500
Fixes: c549808946 ("drm/i915: Mask everything in ring HWSTAM on gen6+ in ringbuffer mode")
References: d420a50c21 ("drm/i915: Clean up the HWSTAM mess")
Testcase: igt/gem_eio/reset-stress
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180808105101.913-2-chris@chris-wilson.co.uk
After disabling resource streamer on ICL (due to it actually not
existing there), I got feedback that there have been some experimental
patches for mesa to use RS years ago, but nothing ever landed or shipped
because there was no performance improvement.
This removes it from kernel keeping the uapi defines around for
compatibility.
v2: - re-add the inadvertent removal of CTX_CTRL_INHIBIT_SYN_CTX_SWITCH
- don't bother trying to document removed params on uapi header:
applications should know that from the query.
(from Chris)
v3: - disable CTX_CTRL_RS_CTX_ENABLE istead of removing it
- reword commit message after Daniele confirmed no performance
regression on his machine
- reword commit message to make clear RS is being removed due to
never been used
v4: - move I915_EXEC_RESOURCE_STREAMER to __I915_EXEC_ILLEGAL_FLAGS so
the check on ioctl() is made much earlier by
i915_gem_check_execbuffer() (suggested by Tvrtko)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Acked-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20180803232443.17193-1-lucas.demarchi@intel.com
Since ggtt_offset_bias is now stored in ggtt.pin_bias, it is duplicated
inside i915_gem_context, and can instead be accessed directly from ggtt.
v3:
Added a helper function to retrieve the ggtt.pin_bias from the vma.
v4:
Moved the helper function to the previous patch in the series.
Dropped the bias from intel_ring_pin. This introduces a slight functional
change since we are always pinning the ring a bit higher if GuC is present
even though we don't really need to.
v8:
Fixed patch not applying on the most recent upstream.
Signed-off-by: Jakub Bartmiński <jakub.bartminski@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
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/20180727141148.30874-4-jakub.bartminski@intel.com
Using PAGE_SIZE for virtual offset alignment is superfluous as it is
equal to the minimum gtt alignment and so equivalent to 0. It is also
the wrong value to use as we stopped using physical page constructs for
the virtual GTT, i.e. it would be preferrable to use I915_GTT_PAGE_SIZE
and in these cases merely imply I915_GTT_MIN_ALIGNMENT.
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/20180727091855.1879-1-chris@chris-wilson.co.uk
