- 01 Mar, 2019 7 commits
-
-
Chris Wilson authored
We don't want to busywait on the GPU if we have other work to do. If we give non-busywaiting workloads higher (initial) priority than workloads that require a busywait, we will prioritise work that is ready to run immediately. We then also have to be careful that we don't give earlier semaphores an accidental boost because later work doesn't wait on other rings, hence we keep a history of semaphore usage of the dependency chain. v2: Stop rolling the bits into a chain and just use a flag in case this request or any of our dependencies use a semaphore. The rolling around was contagious as Tvrtko was heard to fall off his chair. Testcase: igt/gem_exec_schedule/semaphore 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/20190301170901.8340-4-chris@chris-wilson.co.uk
-
Chris Wilson authored
Having introduced per-context seqno, we now have a means to identity progress across the system without feel of rollback as befell the global_seqno. That is we can program a MI_SEMAPHORE_WAIT operation in advance of submission safe in the knowledge that our target seqno and address is stable. However, since we are telling the GPU to busy-spin on the target address until it matches the signaling seqno, we only want to do so when we are sure that busy-spin will be completed quickly. To achieve this we only submit the request to HW once the signaler is itself executing (modulo preemption causing us to wait longer), and we only do so for default and above priority requests (so that idle priority tasks never themselves hog the GPU waiting for others). As might be reasonably expected, HW semaphores excel in inter-engine synchronisation microbenchmarks (where the 3x reduced latency / increased throughput more than offset the power cost of spinning on a second ring) and have significant improvement (can be up to ~10%, most see no change) for single clients that utilize multiple engines (typically media players and transcoders), without regressing multiple clients that can saturate the system or changing the power envelope dramatically. v3: Drop the older NEQ branch, now we pin the signaler's HWSP anyway. v4: Tell the world and include it as part of scheduler caps. Testcase: igt/gem_exec_whisper Testcase: igt/benchmarks/gem_wsim Signed-off-by:
-
Chris Wilson authored
In preparation for enabling HW semaphores, we need to keep in flight timeline HWSP alive until its use across entire system has completed, as any other timeline active on the GPU may still refer back to the already retired timeline. We both have to delay recycling available cachelines and unpinning old HWSP until the next idle point. An easy option would be to simply keep all used HWSP until the system as a whole was idle, i.e. we could release them all at once on parking. However, on a busy system, we may never see a global idle point, essentially meaning the resource will be leaked until we are forced to do a GC pass. We already employ a fine-grained idle detection mechanism for vma, which we can reuse here so that each cacheline can be freed immediately after the last request using it is retired. v3: Keep track of the activity of each cacheline. v4: cacheline_free() on canceling the seqno tracking v5: Finally with a testcase to exercise wraparound v6: Pack cacheline into empty bits of page-aligned vaddr v7: Use i915_utils to hide the pointer casting around bit manipulation Signed-off-by:
-
Chris Wilson authored
On unwinding the active request we give it a small (limited to internal priority levels) boost to prevent it from being gazumped a second time. However, this means that it can be promoted to above the request that triggered the preemption request, causing a preempt-to-idle cycle for no change. We can avoid this if we take the boost into account when checking if the preemption request is valid. v2: After preemption the active request will be after the preemptee if they end up with equal priority. v3: Tvrtko pointed out that this, the existing logic, makes I915_PRIORITY_WAIT non-preemptible. Document this interesting quirk! v4: Prove Tvrtko was right about WAIT being non-preemptible and test it. v5: Except not all priorities were made equal, and the WAIT not preempting is only if we start off as !NEWCLIENT. v6: More commentary after coming to an understanding about what I had forgotten to say. Signed-off-by:
-
Chris Wilson authored
There is no point in whitelisting a register that the user then cannot write to, so check the register exists before merging such patches. v2: Mark SLICE_COMMON_ECO_CHICKEN1 [731c] as write-only v3: Use different variables for different meanings! Signed-off-by:
Michał Winiarski <michal.winiarski@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190301140404.26690-6-chris@chris-wilson.co.uk Link: https://patchwork.freedesktop.org/patch/msgid/20190301160108.19039-1-chris@chris-wilson.co.uk
-
Ville Syrjälä authored
The icl wm1+ underrun w/a has been added to the spec. It changed slightly from the previous incarnation by requiring that we mirror the lines watermark and the ignore lines bit from WM0 into WM1. Signed-off-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190228173639.18422-1-ville.syrjala@linux.intel.comReviewed-by:
Matt Roper <matthew.d.roper@intel.com> Tested-by:
Clint Taylor <Clinton.A.Taylor@intel.com>
-
Chris Wilson authored
A simple mutex used for guarding the flow of requests in and out of the timeline. In the short-term, it will be used only to guard the addition of requests into the timeline, taken on alloc and released on commit so that only one caller can construct a request into the timeline (important as the seqno and ring pointers must be serialised). This will be used by observers to ensure that the seqno/hwsp is stable. Later, when we have reduced retiring to only operate on a single timeline at a time, we can then use the mutex as the sole guard required for retiring. Signed-off-by:
-
- 28 Feb, 2019 8 commits
-
-
Chris Wilson authored
WAIT is occasionally suppressed by virtue of preempted requests being promoted to NEWCLIENT if they have not all ready received that boost. Make this consistent for all WAIT boosts that they are not allowed to preempt executing contexts and are merely granted the right to be at the front of the queue for the next execution slot. This is in keeping with the desire that the WAIT boost be a minor tweak that does not give excessive promotion to its user and open ourselves to trivial abuse. The problem with the inconsistent WAIT preemption becomes more apparent as the preemption is propagated across the engines, where one engine may preempt and the other not, and we be relying on the exact execution order being consistent across engines (e.g. using HW semaphores to coordinate parallel execution). v2: Also protect GuC submission from false preemption loops. v3: Build bug safeguards and better debug messages for st. v4: Do the priority bumping in unsubmit (i.e. on preemption/reset unwind), applying it earlier during submit causes out-of-order execution combined with execute fences. v5: Call sw_fence_fini for our dummy request (Matthew) Signed-off-by:
-
Chris Wilson authored
Gcc has a slight preference if we use __ffs() to subtract one from the index once rather than each use: __execlists_submission_tasklet 2867 2847 -20 Signed-off-by:
-
Chris Wilson authored
We currently use a worker queued from an rcu callback to determine when a how grace period has elapsed while we remained idle. We use this idle delay to infer that we will be idle for a while and this is a suitable point at which we can trim our global memory caches. Since we wrote that, this mechanism now exists as rcu_work, and having converted the idle shrinkers over to using that, we can remove our own variant. v2: Say goodbye to gt.epoch as well. v3: Remove the misplaced and redundant comment before parking globals Signed-off-by:
-
Chris Wilson authored
As our allocations are not device specific, we can move our slab caches to a global scope. Signed-off-by:
-
Chris Wilson authored
As kmem_caches share the same properties (size, allocation/free behaviour) for all potential devices, we can use global caches. While this potential has worse fragmentation behaviour (one can argue that different devices would have different activity lifetimes, but you can also argue that activity is temporal across the system) it is the default behaviour of the system at large to amalgamate matching caches. The benefit for us is much reduced pointer dancing along the frequent allocation paths. v2: Defer shrinking until after a global grace period for futureproofing multiple consumers of the slab caches, similar to the current strategy for avoiding shrinking too early. Signed-off-by:
-
Chris Wilson authored
If we have parked, then we must have passed an idleness test and still be idle. We chose not to use this shortcut in the past so that we could use the idleness test at any time and inspect HW. However, some HW like Sandybridge, doesn't like being woken up frivolously, so avoid doing so. References: 0b702dca ("drm/i915: Avoid waking the engines just to check if they are idle") Signed-off-by:
Mika Kuoppala <mika.kuoppala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190227214159.7946-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
This reverts commit 0b702dca. CI reports that this is not as reliable as it first appears, with failures starting to sporadically occur in selftests. Fixes: 0b702dca ("drm/i915: Avoid waking the engines just to check if they are idle") Signed-off-by:
-
Chris Wilson authored
Do a pass over all the engines upon starting to determine the global scheduler capability flags (those that are agreed upon by all). Signed-off-by:
-
- 27 Feb, 2019 2 commits
-
-
Chris Wilson authored
Exploit that reads of the ring registers return 0 from the engine when it is idle and we do not apply forcewake to know that if the engine is idle then both reads will be identical (and so we interpret the ring as idle). The ulterior motive is to try and reduce the number of spurious wakeups to avoid untimely death, such as: <3> [85.046836] [drm:fw_domains_get [i915]] *ERROR* render: timed out waiting for forcewake ack request. <4> [85.051916] ------------[ cut here ]------------ <4> [85.051917] GT thread status wait timed out <4> [85.051963] WARNING: CPU: 2 PID: 2195 at drivers/gpu/drm/i915/intel_uncore.c:303 __gen6_gt_wait_for_thread_c0+0x6e/0xa0 [i915] <4> [85.051964] Modules linked in: snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic i915 x86_pkg_temp_thermal coretemp mei_hdcp crct10dif_pclmul crc32_pclmul snd_hda_intel ghash_clmulni_intel snd_hda_codec broadcom bcm_phy_lib i2c_i801 snd_hwdep snd_hda_core tg3 snd_pcm ptp pps_core mei_me mei prime_numbers lpc_ich <4> [85.051980] CPU: 2 PID: 2195 Comm: drm_read Tainted: G U 5.0.0-rc8-CI-CI_DRM_5662+ #1 <4> [85.051981] Hardware name: Dell Inc. XPS 8300 /0Y2MRG, BIOS A06 10/17/2011 <4> [85.052012] RIP: 0010:__gen6_gt_wait_for_thread_c0+0x6e/0xa0 [i915] <4> [85.052015] Code: 8b 92 5c 80 13 00 83 e2 07 75 d5 5b 5d c3 80 3d 5b 6a 1a 00 00 75 f4 48 c7 c7 38 21 31 a0 c6 05 4b 6a 1a 00 01 e8 e2 84 ea e0 <0f> 0b eb dd 80 3d 3a 6a 1a 00 00 75 98 48 c7 c6 08 21 31 a0 48 c7 <4> [85.052016] RSP: 0018:ffffc9000043bd00 EFLAGS: 00010086 <4> [85.052019] RAX: 0000000000000000 RBX: ffff888217c50000 RCX: 0000000000000000 <4> [85.052020] RDX: 0000000000000007 RSI: ffffffff820cb141 RDI: 00000000ffffffff <4> [85.052022] RBP: 00000013cd30f2fb R08: 0000000000000000 R09: 0000000000000001 <4> [85.052024] R10: ffffc9000043bce0 R11: 0000000000000000 R12: ffff888217c50ee0 <4> [85.052025] R13: 0000000000000001 R14: 00000000ffffffff R15: ffff888218076530 <4> [85.052028] FS: 00007fc79d049980(0000) GS:ffff888227a80000(0000) knlGS:0000000000000000 <4> [85.052029] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 <4> [85.052031] CR2: 00007f782e2940f8 CR3: 000000022458e006 CR4: 00000000000606e0 <4> [85.052033] Call Trace: <4> [85.052064] gen6_read32+0x14e/0x250 [i915] <4> [85.052096] intel_engine_is_idle+0x7d/0x180 [i915] <4> [85.052126] intel_engines_are_idle+0x29/0x50 [i915] <4> [85.052153] i915_drop_caches_set+0x21c/0x290 [i915] <4> [85.052160] simple_attr_write+0xb0/0xd0 <4> [85.052165] full_proxy_write+0x51/0x80 <4> [85.052170] __vfs_write+0x31/0x190 <4> [85.052176] ? rcu_read_lock_sched_held+0x6f/0x80 <4> [85.052178] ? rcu_sync_lockdep_assert+0x29/0x50 <4> [85.052181] ? __sb_start_write+0x152/0x1f0 <4> [85.052183] ? __sb_start_write+0x163/0x1f0 <4> [85.052187] vfs_write+0xbd/0x1b0 <4> [85.052191] ksys_write+0x50/0xc0 <4> [85.052196] do_syscall_64+0x55/0x190 <4> [85.052200] entry_SYSCALL_64_after_hwframe+0x49/0xbe <4> [85.052202] RIP: 0033:0x7fc79c9d3281 <4> [85.052204] Code: c3 0f 1f 84 00 00 00 00 00 48 8b 05 59 8d 20 00 c3 0f 1f 84 00 00 00 00 00 8b 05 8a d1 20 00 85 c0 75 16 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 57 f3 c3 0f 1f 44 00 00 41 54 55 49 89 d4 53 <4> [85.052206] RSP: 002b:00007fffa4a0a7f8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 <4> [85.052208] RAX: ffffffffffffffda RBX: 0000000000000001 RCX: 00007fc79c9d3281 <4> [85.052210] RDX: 0000000000000005 RSI: 00007fffa4a0a880 RDI: 0000000000000008 <4> [85.052212] RBP: 00007fffa4a0a820 R08: 0000000000000000 R09: 0000000000000000 <4> [85.052213] R10: 0000000000000000 R11: 0000000000000246 R12: 00007fc79c9bc718 <4> [85.052215] R13: 0000000000000003 R14: 00007fc79c9c1628 R15: 00007fc79c9bdd80 <4> [85.052223] irq event stamp: 71630 <4> [85.052226] hardirqs last enabled at (71629): [<ffffffff8197b64c>] _raw_spin_unlock_irqrestore+0x4c/0x60 <4> [85.052228] hardirqs last disabled at (71630): [<ffffffff8197b4bd>] _raw_spin_lock_irqsave+0xd/0x50 <4> [85.052231] softirqs last enabled at (70444): [<ffffffff81c0033a>] __do_softirq+0x33a/0x4b9 <4> [85.052234] softirqs last disabled at (70433): [<ffffffff810b51b1>] irq_exit+0xd1/0xe0 <4> [85.052264] WARNING: CPU: 2 PID: 2195 at drivers/gpu/drm/i915/intel_uncore.c:303 __gen6_gt_wait_for_thread_c0+0x6e/0xa0 [i915] Signed-off-by:
Mika Kuoppala <mika.kuoppala@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190227114958.32438-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
When a request has its priority changed, we traverse the graph of all of its signalers to raise their priorities to match (priority inheritance). If the request has already started executing its payload, we know that all of its signalers must have signaled and we do not need to process our list of signalers. Signed-off-by:
-
- 26 Feb, 2019 14 commits
-
-
Ville Syrjälä authored
Dump out the infoframes in the normal crtc state dump. TODO: Try to better integrate the infoframe dumps with drm state dumps Signed-off-by:Daniel Vetter <daniel.vetter@ffwll.ch> Link: https://patchwork.freedesktop.org/patch/msgid/20190225174106.2163-10-ville.syrjala@linux.intel.com
-
Ville Syrjälä authored
Check the infoframes and infoframe enable state when comparing two crtc states. We'll use the infoframe logging functions from video/hdmi.c to show the infoframes as part of the state dump. TODO: Try to better integrate the infoframe dumps with drm state dumps v2: drm_printk() is no more Signed-off-by: -
Ville Syrjälä authored
Read the HDMI infoframes from the hbuf and unpack them into the crtc state. Well, actually just AVI infoframe for now but let's write the infoframe readout code in a more generic fashion in case we expand this later. Note that Daniel was sceptical about the benefit if this and also concerned about the potential for crappy sdvo encoders not implementing the hbuf read commands. My (admittedly limited) experience is that such encoders don't implement even the get/set hdmi encoding commands and thus would always be treated as dvi only. Hence I believe this is safe, and also IMO preferable having quirks to deal with missing readout support. The readout support is neatly isolated in the sdvo code whereas the quirk would leak to other parts of the driver (state checker, fastboot, etc.) thus complicating the lives of other people. Signed-off-by:
-
Ville Syrjälä authored
As with regular HDMI encoders, let's precompute the infoframes (actually just AVI infoframe for the time being) with SDVO HDMI encoders. v2: Drop the WARN_ON() from drm_hdmi_avi_infoframe_from_display_mode() return since that could genuinely fail due to user asking for incompatible aspect ratio v3: .compute_config() now returns int Signed-off-by: -
Ville Syrjälä authored
Add code to read the infoframes from the video DIP and unpack them into the crtc state. v2: Make the read funcs return void (Daniel) Drop the duplicate infoframe enabled checks (Daniel) Add a FIXME for lspcon infoframe readout Signed-off-by: -
Ville Syrjälä authored
Store the infoframes in the crtc state and precompute them in .compute_config(). While precomputing we'll also fill out the inforames.enable bitmask appropriately. v2: Drop the null packet stuff (Daniel) Add a FIXME for lspcon v3: .compute_config() now returns int Signed-off-by: -
Ville Syrjälä authored
Store the mask of enabled infoframes in the crtc state. We'll start with just the readout for HDMI encoder, and we'll expand this to compute the bitmask in .compute_config() later. SDVO will also follow later. Signed-off-by:
-
Ville Syrjälä authored
We want to start tracking which infoframes are enabled, so let's replace the boolean flag with a bitmask. We'll abstract the bitmask so that it's not platform dependent. That will allow us to examine the bitmask later in platform independent code. v2: Don't map VIDEO_DIP_ENABLE to the null packet (Daniel) Put a FIXME in the lspcon function Signed-off-by: -
Ville Syrjälä authored
We have definitions and low level code for everything except the gamut metadata HDMI packet. Add the missing bits. Signed-off-by:
-
Abdiel Janulgue authored
This simplifies adding new query item objects. v2: Use query_hdr (Tvrtko, Chris). int instead of u32 in return (Tvrtko) v3: More naming fixes (Tvrtko) Signed-off-by:Abdiel Janulgue <abdiel.janulgue@linux.intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by:
-
Chris Wilson authored
In selftests/live_hangcheck, we have a lot of tests for resetting simple spinners, but nothing quite prepared us for how the GPU reacted to triggering a reset outside of the safe spinner. These two subtests fill the ring with plain old empty, non-spinning requests, and then triggers a reset. Without a user-payload to blame, these requests will exercise the 'non-started' paths and mostly be replayed verbatim. Signed-off-by:
-
Chris Wilson authored
Having weaned the interrupt handling off using a single global execution queue, we no longer need to emit a global_seqno. Note that we still have a few assumptions about execution order along engine timelines, but this removes the most obvious artefact! Signed-off-by:
-
Chris Wilson authored
Stop accessing the HWSP to read the global seqno, and stop tracking the mirror in the engine's execution timeline -- it is unused. Signed-off-by:
-
Chris Wilson authored
To determine whether an engine has 'stuck', we simply check whether or not is still on the same seqno for several seconds. To keep this simple mechanism intact over the loss of a global seqno, we can simply add a new global heartbeat seqno instead. As we cannot know the sequence in which requests will then be completed, we use a primitive random number generator instead (with a cycle long enough to not matter over an interval of a few thousand requests between hangcheck samples). The alternative to using a dedicated seqno on every request is to issue a heartbeat request and query its progress through the system. Sadly this requires us to reduce struct_mutex so that we can issue requests without requiring that bkl. v2: And without the extra CS_STALL for the hangcheck seqno -- we don't need strict serialisation with what comes later, we just need to be sure we don't write the hangcheck seqno before our batch is flushed. Signed-off-by:
-
- 25 Feb, 2019 1 commit
-
-
José Roberto de Souza authored
The commit that this patch fixes changed the order of the parameters of MG_DP_MODE() but din't update the callers, breaking type-c on ICL. Fixes: 58106b7d ("drm/i915: Make MG PHY macros semantically consistent") Cc: Clint Taylor <clinton.a.taylor@intel.com> Cc: Imre Deak <imre.deak@intel.com> Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: Aditya Swarup <aditya.swarup@intel.com> Cc: Manasi navare <manasi.d.navare@intel.com> Cc: Jani Nikula <jani.nikula@intel.com> Signed-off-by:
José Roberto de Souza <jose.souza@intel.com> Reviewed-by:
Lucas De Marchi <lucas.demarchi@intel.com> Signed-off-by:
Jani Nikula <jani.nikula@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190222202437.6575-1-jose.souza@intel.com
-
- 23 Feb, 2019 1 commit
-
-
Chris Wilson authored
As we no longer have a precise indication of requests queued to an engine, make no presumptions and just sample the ring registers to see if the engine is busy. v2: Report busy while the ring is idling on a semaphore/event. v3: Give the struct a name! v4: Always 0 outside the powerwell; trusting the powerwell is accurate enough for our sampling pmu. v5: Protect against gen7 mmio madness and try to improve grammar Signed-off-by:
-
- 22 Feb, 2019 4 commits
-
-
Rodrigo Vivi authored
Other than LPT, no other PCH needed to differentiate between LP and HP. So let's remove this before we spread this mistake to future platforms. Cc: Anusha Srivatsa <anusha.srivatsa@intel.com> Cc: Lucas De Marchi <lucas.demarchi@intel.com> Signed-off-by:
Rodrigo Vivi <rodrigo.vivi@intel.com> Reviewed-by:
-
Rodrigo Vivi authored
No functional change. Just a reorg to match the preferred behavior. v2: missing else (Ville) Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Cc: Lucas De Marchi <lucas.demarchi@intel.com> Signed-off-by:
-
Rodrigo Vivi authored
No functional change. Just a reorg to match the preferred behavior. Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Cc: Lucas De Marchi <lucas.demarchi@intel.com> Signed-off-by:
-
Chris Wilson authored
As we disable the log capture events, flush any residual interrupt before we flush and disable the worker. v2: Mika pointed out that it wasn't the worker re-queueing itself, but a rogue irq. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109716Signed-off-by:
-
- 21 Feb, 2019 3 commits
-
-
Chengguang Xu authored
unlikely has already included in IS_ERR(), so just remove redundant likely/unlikely annotation. Signed-off-by:
Chengguang Xu <cgxu519@gmx.com> Reviewed-by:
-
Chris Wilson authored
Annoyingly, struct_mutex was not entirely eliminated from the reset pathway; for reasons of its own, intel_display_resume() requires struct_mutex to prepare the planes it already captured. To avoid the immediate problem of a deadlock between the struct_mutex and the reset srcu, we have to acquire the reset_lock before struct_mutex in i915_gem_fault(). Now any wait underneath struct_mutex will result us in having to forcibly reset all inflight rendering, less than ideal, but better than a deadlock (and will do for the short term). Signed-off-by:
-
Uma Shankar authored
gamma mode mask was not considering the 30th and 31st bits. Due to this state readout was masking these bits, causing a mismatch and false warning, even though the registers were updated correctly. Dropped the gamma mode mask as it is redundant and ideally entire register content should be matching. This resolves the state mismatch warnings. Signed-off-by:
Uma Shankar <uma.shankar@intel.com> Signed-off-by:
-
