- 01 Jul, 2016 20 commits
-
-
Chris Wilson authored
If we flag the seqno as potentially stale upon receiving an interrupt, we can use that information to reduce the frequency that we apply the heavyweight coherent seqno read (i.e. if we wake up a chain of waiters). v2: Use cmpxchg to replace READ_ONCE/WRITE_ONCE for more explicit control of the ordering wrt to interrupt generation and interrupt checking in the bottom-half. Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by:
Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1467390209-3576-14-git-send-email-chris@chris-wilson.co.uk
-
Chris Wilson authored
If we have multiple waiters, we may find that many complete on the same wake up. If we first inspect the seqno from the CPU cache, we may reduce the number of heavyweight coherent seqno reads we require. Signed-off-by:
-
Chris Wilson authored
On Ironlake, there is no command nor register to ensure that the write from a MI_STORE command is completed (and coherent on the CPU) before the command parser continues. This means that the ordering between the seqno write and the subsequent user interrupt is undefined (like gen6+). So to ensure that the seqno write is completed after the final user interrupt we need to delay the read sufficiently to allow the write to complete. This delay is undefined by the bspec, and empirically requires 75us even though a register read combined with a clflush is less than 500ns. Hence, the delay is due to an on-chip buffer rather than the latency of the write to memory. Note that the render ring controls this by filling the PIPE_CONTROL fifo with stalling commands that force the earliest pipe-control with the seqno to be completed before the command parser continues. Given that we need a barrier operation for BSD, we may as well forgo the extra per-batch latency by using a common per-interrupt barrier. Studying the impact of adding the usleep shows that in both sequences of and individual synchronous no-op batches is negligible for the media engine (where the write now is unordered with the interrupt). Converting the render engine over from the current glutton of pie-controls over to the per-interrupt delays speeds up both the sequential and individual synchronous no-ops by 20% and 60%, respectively. This speed up holds even when looking at the throughput of small copies (4KiB->4MiB), both serial and synchronous, by about 20%. This is because despite adding a significant delay to the interrupt, in all likelihood we will see the seqno write without having to apply the barrier (only in the rare corner cases where the write is delayed on the last required is the delay necessary). Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94307 Testcase: igt/gem_sync #ilk Signed-off-by:
-
Chris Wilson authored
The gen2 w/a buffer is stuffed into the same slot as the gen5+ scratch buffer. If we pass in the size we want to allocate for the scratch buffer, both callers can use the same routine. Signed-off-by:
-
Chris Wilson authored
With the last direct CPU access to the scratch page removed, we can now allocate it from our small amount of reserved system pages (stolen memory). Signed-off-by:
-
Chris Wilson authored
After the elimination of using the scratch page for Ironlake's breadcrumb, we no longer need to kmap the object. We therefore can move it into the high unmappable space and do not need to force the object to be coherent (i.e. snooped on !llc platforms). Signed-off-by:
-
Chris Wilson authored
By using the same address for storing the HWS on every platform, we can remove the platform specific vfuncs and reduce the get-seqno routine to a single read of a cached memory location. v2: Fix semaphore_passed() to look at the signaling engine (not the waiter's) Signed-off-by:
-
Chris Wilson authored
When waiting for an interrupt (waiting for the engine to complete some work), we know we are the only waiter to be woken on this engine. We also know when the GPU has nearly completed our request (or at least started processing it), so after being woken and we detect that the GPU is active and working on our request, allow us the bottom-half (the first waiter who wakes up to handle checking the seqno after the interrupt) to spin for a very short while to reduce client latencies. The impact is minimal, there was an improvement to the realtime-vs-many clients case, but exporting the function proves useful later. However, it is tempting to adjust irq_seqno_barrier to include the spin. The problem is first ensuring that the "start-of-request" seqno is coherent as we use that as our basis for judging when it is ok to spin. If we could, spinning there could dramatically shorten some sleeps, and allow us to make the barriers more conservative to handle missed seqno writes on more platforms (all gen7+ are known to have the occasional issue, at least). Signed-off-by:
-
Chris Wilson authored
One particularly stressful scenario consists of many independent tasks all competing for GPU time and waiting upon the results (e.g. realtime transcoding of many, many streams). One bottleneck in particular is that each client waits on its own results, but every client is woken up after every batchbuffer - hence the thunder of hooves as then every client must do its heavyweight dance to read a coherent seqno to see if it is the lucky one. Ideally, we only want one client to wake up after the interrupt and check its request for completion. Since the requests must retire in order, we can select the first client on the oldest request to be woken. Once that client has completed his wait, we can then wake up the next client and so on. However, all clients then incur latency as every process in the chain may be delayed for scheduling - this may also then cause some priority inversion. To reduce the latency, when a client is added or removed from the list, we scan the tree for completed seqno and wake up all the completed waiters in parallel. Using igt/benchmarks/gem_latency, we can demonstrate this effect. The benchmark measures the number of GPU cycles between completion of a batch and the client waking up from a call to wait-ioctl. With many concurrent waiters, with each on a different request, we observe that the wakeup latency before the patch scales nearly linearly with the number of waiters (before external factors kick in making the scaling much worse). After applying the patch, we can see that only the single waiter for the request is being woken up, providing a constant wakeup latency for every operation. However, the situation is not quite as rosy for many waiters on the same request, though to the best of my knowledge this is much less likely in practice. Here, we can observe that the concurrent waiters incur extra latency from being woken up by the solitary bottom-half, rather than directly by the interrupt. This appears to be scheduler induced (having discounted adverse effects from having a rbtree walk/erase in the wakeup path), each additional wake_up_process() costs approximately 1us on big core. Another effect of performing the secondary wakeups from the first bottom-half is the incurred delay this imposes on high priority threads - rather than immediately returning to userspace and leaving the interrupt handler to wake the others. To offset the delay incurred with additional waiters on a request, we could use a hybrid scheme that did a quick read in the interrupt handler and dequeued all the completed waiters (incurring the overhead in the interrupt handler, not the best plan either as we then incur GPU submission latency) but we would still have to wake up the bottom-half every time to do the heavyweight slow read. Or we could only kick the waiters on the seqno with the same priority as the current task (i.e. in the realtime waiter scenario, only it is woken up immediately by the interrupt and simply queues the next waiter before returning to userspace, minimising its delay at the expense of the chain, and also reducing contention on its scheduler runqueue). This is effective at avoid long pauses in the interrupt handler and at avoiding the extra latency in realtime/high-priority waiters. v2: Convert from a kworker per engine into a dedicated kthread for the bottom-half. v3: Rename request members and tweak comments. v4: Use a per-engine spinlock in the breadcrumbs bottom-half. v5: Fix race in locklessly checking waiter status and kicking the task on adding a new waiter. v6: Fix deciding when to force the timer to hide missing interrupts. v7: Move the bottom-half from the kthread to the first client process. v8: Reword a few comments v9: Break the busy loop when the interrupt is unmasked or has fired. v10: Comments, unnecessary churn, better debugging from Tvrtko v11: Wake all completed waiters on removing the current bottom-half to reduce the latency of waking up a herd of clients all waiting on the same request. v12: Rearrange missed-interrupt fault injection so that it works with igt/drv_missed_irq_hang v13: Rename intel_breadcrumb and friends to intel_wait in preparation for signal handling. v14: RCU commentary, assert_spin_locked v15: Hide BUG_ON behind the compiler; report on gem_latency findings. v16: Sort seqno-groups by priority so that first-waiter has the highest task priority (and so avoid priority inversion). v17: Add waiters to post-mortem GPU hang state. v18: Return early for a completed wait after acquiring the spinlock. Avoids adding ourselves to the tree if the is already complete, and skips the awkward question of why we don't do completion wakeups for waits earlier than or equal to ourselves. v19: Prepare for init_breadcrumbs to fail. Later patches may want to allocate during init, so be prepared to propagate back the error code. Testcase: igt/gem_concurrent_blit Testcase: igt/benchmarks/gem_latency Signed-off-by:
-
Chris Wilson authored
Currently __i915_wait_request uses a per-engine wait_queue_t for the dual purpose of waking after the GPU advances or for waking after an error. In the future, we may add even more wake sources and require greater separation, but for now we can conceptually simplify wakeups by separating the two sources. In particular, this allows us to use different wait-queues (e.g. one on the engine advancement, a global one for errors and one on each requests) without any hassle. Signed-off-by:
-
Chris Wilson authored
Since the function is a small wrapper around schedule_delayed_work(), move it inline to remove the function call overhead for the principle caller. Signed-off-by:
-
Chris Wilson authored
The queue only ever contains at most one item and has no special flags. It is just a very simple wrapper around the system-wq - a complication with no benefits. v2: Use the system_long_wq as we may wish to capture the error state after detecting the hang - which may take a bit of time. Signed-off-by:
-
Chris Wilson authored
We can forgo queuing the hangcheck from the start of every request to until we wait upon a request. This reduces the overhead of every request, but may increase the latency of detecting a hang. However, if nothing every waits upon a hang, did it ever hang? It also improves the robustness of the wait-request by ensuring that the hangchecker is indeed running before we sleep indefinitely (and thereby ensuring that we never actually sleep forever waiting for a dead GPU). As pointed out by Tvrtko, it is possible for a GPU hang to go unnoticed for as long as nobody is waiting for the GPU. Though this rare, during that time we may be consuming more power than if we had promptly recovered, and in the most extreme case we may exhaust all memory before forcing the hangcheck. Something to be wary off in future. Signed-off-by:
-
Chris Wilson authored
As we inspect obj->active to decide how many objects we can shrink (we only shrink idle objects), it helps to flush the active lists first in order to have a more accurate count of available objects. Signed-off-by:
-
Imre Deak authored
Broxton is now part of CI which doesn't indicate any major problems so enable the driver by default. Signed-off-by:
Imre Deak <imre.deak@intel.com> Reviewed-by:
Rodrigo Vivi <rodrigo.vivi@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1467384045-17028-1-git-send-email-imre.deak@intel.com
-
arun.siluvery@linux.intel.com authored
Pooled EU is a bxt only feature and kernel changes are already merged. This feature is not yet exposed to userspace as the support was not yet available. Beignet team expressed interest and added patches to use this. Since we now have a user and patches to use them, expose them from the kernel side as well. v2: fix compile error [1] https://lists.freedesktop.org/archives/beignet/2016-June/007698.html [2] https://lists.freedesktop.org/archives/beignet/2016-June/007699.html Cc: Winiarski, Michal <michal.winiarski@intel.com> Cc: Zou, Nanhai <nanhai.zou@intel.com> Cc: Yang, Rong R <rong.r.yang@intel.com> Cc: Tim Gore <tim.gore@intel.com> Cc: Jeff McGee <jeff.mcgee@intel.com> Signed-off-by:
Arun Siluvery <arun.siluvery@linux.intel.com> Acked-by:
Jani Nikula <jani.nikula@intel.com>
-
Imre Deak authored
BXT BIOS has two options related to GPU power management: "RC6(Render Standby)" and "GT PM Support". The assumption so far was that disabling either of these options would leave RC6 uninitialized. According to my tests this isn't so: for a proper RC6 setup we only need the "GT PM Support" option to be enabled while the "RC6" option only controls whether RC6 is left enabled or not by BIOS. OTOH we were missing a few checks to ensure a proper RC6 setup. Add these now and don't fail the sanity check if RC6 is disabled. This fixes a problem where RC6 remains disabled after reloading the driver, since we explicitly disable RC6 during unloading. v2: - Print a debug message about the BIOS enabled RC state. (Sagar) CC: Sagar Arun Kamble <sagar.a.kamble@intel.com> Signed-off-by:
Sagar Arun Kamble <sagar.a.kamble@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1467216835-1086-2-git-send-email-imre.deak@intel.com
-
Imre Deak authored
RC6 isn't really a KMS feature, so use the more proper DRIVER log type for RC6 related debug messages. CC: Sagar Arun Kamble <sagar.a.kamble@intel.com> Signed-off-by:
-
Chris Wilson authored
Just plonk all the default irq vfuncs together in one function to keep the initialisers of reasonable size. Signed-off-by:
-
Chris Wilson authored
Consolidate the block of default vfuncs for dispatching the batchbuffer. Just a minor tweak on top of Tvrtko's great job of tidying up the vfunc initialisation. Signed-off-by:
-
- 30 Jun, 2016 20 commits
-
-
Tvrtko Ursulin authored
Just a bit of cleanup after the previous refactoring. Signed-off-by:
-
Tvrtko Ursulin authored
Replace per-engine initialization with a common half-programatic, half-data driven code for ease of maintenance and compactness. Signed-off-by:
-
Tvrtko Ursulin authored
Store the semaphore offset in a temporary variable to avoid having to get the VMA offset twice. Signed-off-by:
-
Tvrtko Ursulin authored
Replace the macro initializer with a programatic loop which results in smaller code and hopefully just as clear. v2: Rebase. Signed-off-by:
-
Tvrtko Ursulin authored
The object needs to be created before semaphores can be initialized on any ring and it makes sense to pull it out to this semaphore dedicated helper. Signed-off-by:
-
Tvrtko Ursulin authored
Signed-off-by: -
Tvrtko Ursulin authored
v2: Put dispatch_execbuffer before add_request. (Chris Wilson) v3: Fix add_request and irq_seqno_barrier for gen8+. Signed-off-by:
-
Tvrtko Ursulin authored
Signed-off-by:
-
Tvrtko Ursulin authored
Signed-off-by:
-
Tvrtko Ursulin authored
v2: Consistent INTEL_GEN vs IS_GEN usage. (Chris Wilson) Signed-off-by:
-
Tvrtko Ursulin authored
Signed-off-by:
-
Tvrtko Ursulin authored
All engines apart from render select this based on Gen. Move it to the common helper as well. Signed-off-by:
-
Tvrtko Ursulin authored
Introduce a function which initializes vfuncs mostly common across engines and move write_tail initialization in it since only one engine overrides the default. Signed-off-by:
-
Chris Wilson authored
By using the out-of-line intel_wait_for_register() not only do we can efficiency from using the hybrid wait_for() contained within, but we avoid code bloat from the numerous inlined loops, in total (all patches): text data bss dec hex filename 1078551 4557 416 1083524 108884 drivers/gpu/drm/i915/i915.ko 1070775 4557 416 1075748 106a24 drivers/gpu/drm/i915/i915.ko Signed-off-by:
-
Chris Wilson authored
By using the out-of-line intel_wait_for_register() not only do we can efficiency from using the hybrid wait_for() contained within, but we avoid code bloat from the numerous inlined loops, in total (all patches): text data bss dec hex filename 1078551 4557 416 1083524 108884 drivers/gpu/drm/i915/i915.ko 1070775 4557 416 1075748 106a24 drivers/gpu/drm/i915/i915.ko Signed-off-by:
-
Chris Wilson authored
Since we have a sequence of register reads and writes, we can reduce the latency of starting the BSD ring by performing all the mmio operations under the same forcewake wakeref. Signed-off-by:
-
Chris Wilson authored
By using the out-of-line intel_wait_for_register() not only do we can efficiency from using the hybrid wait_for() contained within, but we avoid code bloat from the numerous inlined loops, in total (all patches): text data bss dec hex filename 1078551 4557 416 1083524 108884 drivers/gpu/drm/i915/i915.ko 1070775 4557 416 1075748 106a24 drivers/gpu/drm/i915/i915.ko Signed-off-by:
-
Chris Wilson authored
By using the out-of-line intel_wait_for_register() not only do we can efficiency from using the hybrid wait_for() contained within, but we avoid code bloat from the numerous inlined loops, in total (all patches): text data bss dec hex filename 1078551 4557 416 1083524 108884 drivers/gpu/drm/i915/i915.ko 1070775 4557 416 1075748 106a24 drivers/gpu/drm/i915/i915.ko Signed-off-by:
-
Chris Wilson authored
By using the out-of-line intel_wait_for_register() not only do we can efficiency from using the hybrid wait_for() contained within, but we avoid code bloat from the numerous inlined loops, in total (all patches): text data bss dec hex filename 1078551 4557 416 1083524 108884 drivers/gpu/drm/i915/i915.ko 1070775 4557 416 1075748 106a24 drivers/gpu/drm/i915/i915.ko Signed-off-by:
-
Chris Wilson authored
By using the out-of-line intel_wait_for_register() not only do we can efficiency from using the hybrid wait_for() contained within, but we avoid code bloat from the numerous inlined loops, in total (all patches): text data bss dec hex filename 1078551 4557 416 1083524 108884 drivers/gpu/drm/i915/i915.ko 1070775 4557 416 1075748 106a24 drivers/gpu/drm/i915/i915.ko Signed-off-by:
-
