- 10 Nov, 2017 16 commits
-
-
Chris Wilson authored
Move all of the i915->mm initialisation to a private function that can be reused by the mock i915 device to save forgetting any more steps. For example, <7>[ 1542.046332] [IGT] drv_selftest: starting subtest mock_objects <4>[ 1542.123924] Setting dangerous option mock_selftests - tainting kernel <6>[ 1542.167941] i915: Performing mock selftests with st_random_seed=0x246f5ab5 st_timeout=1000 <4>[ 1542.178012] INFO: trying to register non-static key. <4>[ 1542.178027] the code is fine but needs lockdep annotation. <4>[ 1542.178032] turning off the locking correctness validator. <4>[ 1542.178041] CPU: 3 PID: 6008 Comm: kworker/3:7 Tainted: G U 4.14.0-rc8-CI-CI_DRM_3332+ #1 <4>[ 1542.178049] Hardware name: /NUC6CAYB, BIOS AYAPLCEL.86A.0040.2017.0619.1722 06/19/2017 <4>[ 1542.178144] Workqueue: events __i915_gem_free_work [i915] <4>[ 1542.178152] Call Trace: <4>[ 1542.178163] dump_stack+0x68/0x9f <4>[ 1542.178170] register_lock_class+0x3fd/0x580 <4>[ 1542.178177] ? unwind_next_frame+0x14/0x20 <4>[ 1542.178184] ? __save_stack_trace+0x73/0xd0 <4>[ 1542.178191] __lock_acquire+0xa4/0x1b00 <4>[ 1542.178254] ? __i915_gem_free_work+0x28/0xa0 [i915] <4>[ 1542.178261] ? __lock_acquire+0x4ab/0x1b00 <4>[ 1542.178268] lock_acquire+0xb0/0x200 <4>[ 1542.178273] ? lock_acquire+0xb0/0x200 <4>[ 1542.178336] ? __i915_gem_free_work+0x28/0xa0 [i915] <4>[ 1542.178344] _raw_spin_lock+0x32/0x50 <4>[ 1542.178405] ? __i915_gem_free_work+0x28/0xa0 [i915] <4>[ 1542.178468] __i915_gem_free_work+0x28/0xa0 [i915] <4>[ 1542.178476] process_one_work+0x221/0x650 <4>[ 1542.178483] worker_thread+0x4e/0x3c0 <4>[ 1542.178489] kthread+0x114/0x150 <4>[ 1542.178494] ? process_one_work+0x650/0x650 <4>[ 1542.178499] ? kthread_create_on_node+0x40/0x40 <4>[ 1542.178506] ret_from_fork+0x27/0x40 v2: Fish out i915->mm.object_stat_lock which was being inited over in i915_drv.c (Matthew) Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Cc: Matthew Auld <matthew.auld@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20171110232447.21618-1-chris@chris-wilson.co.ukReviewed-by:
Matthew Auld <matthew.auld@intel.com>
-
Chris Wilson authored
Update the kerneldoc parameter name to match the real parameter name. Signed-off-by:
-
Chris Wilson authored
As we now record the default HW state and so only emit the "golden" renderstate once to prepare the HW, there is no advantage in keeping the renderstate batch around as it will never be used again. Signed-off-by:
Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20171110142634.10551-8-chris@chris-wilson.co.uk
-
Chris Wilson authored
Take a copy of the HW state after a reset upon module loading by executing a context switch from a blank context to the kernel context, thus saving the default hw state over the blank context image. We can then use the default hw state to initialise any future context, ensuring that each starts with the default view of hw state. v2: Unmap our default state from the GTT after stealing it from the context. This should stop us from accidentally overwriting it via the GTT (and frees up some precious GTT space). Testcase: igt/gem_ctx_isolation Signed-off-by:
-
Chris Wilson authored
In the next few patches, we will want to both copy out of the context image and write a valid image into a new context. To be completely safe, we should then couple in our domain tracking to ensure that we don't have any issues with stale data remaining in unwanted cachelines. Historically, we omitted the .write=true from the call to set-gtt-domain in i915_switch_context() in order to avoid a stall between every request as we would want to wait for the previous context write from the gpu. Since then, we limit the set-gtt-domain to only occur when we first bind the vma, so once in use we will never stall, and we are sure to flush the context following a load from swap. Equally we never applied the lessons learnt from ringbuffer submission to execlists; so time to apply the flush of the lrc after load as well. Signed-off-by:
Mika Kuoppala <mika.kuoppala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20171110142634.10551-6-chris@chris-wilson.co.uk
-
Chris Wilson authored
intel_modeset_gem_init() now only sets up the legacy overlay, so let's remove the function and call the setup directly during driver load. This should help us find a better point in the initialisation sequence for it later. Signed-off-by:
-
Chris Wilson authored
Despite its name intel_init_clock_gating applies both display clock gating workarounds; GT mmio workarounds and the occasional GT power context workaround. Worse, sometimes it includes a context register workaround which we need to apply before we record the default HW state for all contexts. Signed-off-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20171110142634.10551-4-chris@chris-wilson.co.uk
-
Chris Wilson authored
GT powersaving is tightly coupled to the request infrastructure. To avoid complications with the order of initialisation in the next patch (where we want to send requests to hw during GEM init) move the powersaving initialisation into the purview of i915_gem_init(). Signed-off-by:
-
Chris Wilson authored
In the next few patches, we will have a hard requirement that we emit a context-switch to the perma-pinned i915->kernel_context (so that we can save the HW state using that context-switch). As the first context itself may be classed as a kernel context, we want to be explicit in our comparison. For an extra-layer of finesse, we can check the last unretired context on the engine; as well as the last retired context when idle. v2: verbose verbosity v3: Always force the switch, even when the engine is idle, and update the assert that this happens before suspend. Signed-off-by:
-
Tvrtko Ursulin authored
We want to be able to report back to userspace details about an engine's class, and in return for userspace to be able to request actions regarding certain classes of engines. To isolate the uABI from any variations between hw generations, we define an abstract class for the engines and internally map onto the hw. v2: Remove MAX from the uABI; keep it internal if we need it, but don't let userspace make the mistake of using it themselves. v3: s/OTHER/INVALID/ The use of OTHER is ill-defined, so remove it from the uABI as any future new type of engine can define a class to suit it. But keep a reserved value for an invalid class, so that we can always unambiguously express when something doesn't belong to the classification. Signed-off-by:
Tvrtko Ursulin <tvrtko.ursulin@intel.com> Signed-off-by:
Lionel Landwerlin <lionel.g.landwerlin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20171110142634.10551-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
lockdep spotted that the mock tests were using the i915->mm.obj_lock without first initialiasing it: >[ 1303.217043] [IGT] drv_selftest: starting subtest mock_objects <4>[ 1303.240898] Setting dangerous option mock_selftests - tainting kernel <6>[ 1303.253665] i915: Performing mock selftests with st_random_seed=0xd87ea6c6 st_timeout=1000 <4>[ 1303.254812] INFO: trying to register non-static key. <4>[ 1303.254816] the code is fine but needs lockdep annotation. <4>[ 1303.254818] turning off the locking correctness validator. <4>[ 1303.254820] CPU: 4 PID: 13112 Comm: drv_selftest Tainted: G U W 4.14.0-rc8-CI-Patchwork_7058+ #1 <4>[ 1303.254823] Hardware name: MSI MS-7924/Z97M-G43(MS-7924), BIOS V1.12 02/15/2016 <4>[ 1303.254825] Call Trace: <4>[ 1303.254829] dump_stack+0x68/0x9f <4>[ 1303.254832] register_lock_class+0x3fd/0x580 <4>[ 1303.254835] ? ___slab_alloc.constprop.29+0x157/0x3d0 <4>[ 1303.254837] ? ___slab_alloc.constprop.29+0x157/0x3d0 <4>[ 1303.254840] ? sg_kmalloc+0x1e/0x50 <4>[ 1303.254842] ? debug_smp_processor_id+0x17/0x20 <4>[ 1303.254845] __lock_acquire+0xa4/0x1b00 <4>[ 1303.254884] ? __i915_gem_object_set_pages+0x116/0x1f0 [i915] <4>[ 1303.254887] ? __this_cpu_preempt_check+0x13/0x20 <4>[ 1303.254889] ? sg_kmalloc+0x1e/0x50 <4>[ 1303.254891] lock_acquire+0xb0/0x200 <4>[ 1303.254893] ? lock_acquire+0xb0/0x200 <4>[ 1303.254917] ? __i915_gem_object_set_pages+0x116/0x1f0 [i915] <4>[ 1303.254920] _raw_spin_lock+0x32/0x50 <4>[ 1303.254944] ? __i915_gem_object_set_pages+0x116/0x1f0 [i915] <4>[ 1303.254967] __i915_gem_object_set_pages+0x116/0x1f0 [i915] <4>[ 1303.254991] i915_gem_object_get_pages_phys+0x286/0x2b0 [i915] <4>[ 1303.255015] ____i915_gem_object_get_pages+0x20/0x60 [i915] <4>[ 1303.255039] i915_gem_object_attach_phys+0x137/0x1a0 [i915] <4>[ 1303.255063] igt_phys_object+0x45/0x120 [i915] <4>[ 1303.255094] __i915_subtests+0x40/0xd0 [i915] <4>[ 1303.255099] ? work_on_cpu_safe+0x60/0x60 <4>[ 1303.255131] i915_gem_object_mock_selftests+0x34/0x50 [i915] Signed-off-by:
Matthew Auld <matthew.william.auld@gmail.com> Reviewed-by:
-
Chris Wilson authored
So it appears that commit 5427f207 ("drm/i915: Bump wait-times for the final CS interrupt before parking") was a little over optimistic in its belief that it had successfully waited for all residual activity on the engines before parking. Numerous sightings in CI since then of <7>[ 52.542886] [IGT] core_auth: executing <3>[ 52.561013] [drm:intel_engines_park [i915]] *ERROR* vcs0 is not idle before parking <7>[ 52.561215] intel_engines_park vcs0 <7>[ 52.561229] intel_engines_park current seqno 98, last 98, hangcheck 0 [-247449 ms], inflight 0 <7>[ 52.561238] intel_engines_park Reset count: 0 <7>[ 52.561266] intel_engines_park Requests: <7>[ 52.561363] intel_engines_park RING_START: 0x00000000 [0x00000000] <7>[ 52.561377] intel_engines_park RING_HEAD: 0x00000000 [0x00000000] <7>[ 52.561390] intel_engines_park RING_TAIL: 0x00000000 [0x00000000] <7>[ 52.561406] intel_engines_park RING_CTL: 0x00000000 <7>[ 52.561422] intel_engines_park RING_MODE: 0x00000200 [idle] <7>[ 52.561442] intel_engines_park ACTHD: 0x00000000_00000000 <7>[ 52.561459] intel_engines_park BBADDR: 0x00000000_00000000 <7>[ 52.561474] intel_engines_park Execlist status: 0x00000301 00000000 <7>[ 52.561489] intel_engines_park Execlist CSB read 5 [5 cached], write 5 [5 from hws], interrupt posted? no <7>[ 52.561500] intel_engines_park ELSP[0] idle <7>[ 52.561510] intel_engines_park ELSP[1] idle <7>[ 52.561519] intel_engines_park HW active? 0x0 <7>[ 52.561608] intel_engines_park Idle? yes <7>[ 52.561617] intel_engines_park on Braswell, which indicates that the engine just needs that little bit longer after flushing the tasklet to settle. So give it a few more milliseconds before declaring an err and applying the emergency brake. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103479Signed-off-by:
-
Hans de Goede authored
intel_uncore_forcewake_reset() does forcewake puts and gets as such we need to make sure that no-one tries to access the PUNIT->PMIC bus (on systems where this bus is shared) while it runs, otherwise bad things happen. Normally this is taken care of by the i915_pmic_bus_access_notifier() which does an intel_uncore_forcewake_get(FORCEWAKE_ALL) when some other driver tries to access the PMIC bus, so that later forcewake gets are no-ops (for the duration of the bus access). But intel_uncore_forcewake_reset gets called in 3 cases: 1) Before registering the pmic_bus_access_notifier 2) After unregistering the pmic_bus_access_notifier 3) To reset forcewake state on a GPU reset In all 3 cases the i915_pmic_bus_access_notifier() protection is insufficient. This commit fixes this race by calling iosf_mbi_punit_acquire() before calling intel_uncore_forcewake_reset(). In the case where it is called directly after unregistering the pmic_bus_access_notifier, we need to hold the punit-lock over both calls to avoid a race where intel_uncore_fw_release_timer() may execute between the 2 calls. Reviewed-by:
Imre Deak <imre.deak@intel.com> Acked-by:
Ingo Molnar <mingo@kernel.org> Signed-off-by:
Hans de Goede <hdegoede@redhat.com> Link: https://patchwork.freedesktop.org/patch/msgid/20171019111620.26761-3-hdegoede@redhat.com
-
Hans de Goede authored
For race free unregistration drivers may need to acquire PMIC bus access through iosf_mbi_punit_acquire() and then (un)register the notifier without dropping the lock. This commit adds an unlocked variant of iosf_mbi_unregister_pmic_bus_access_notifier for this use case. Acked-by:
-
Chris Wilson authored
In order to allow the mock breadcrumbs tests to run without device irqs being enabled, move the intel_irqs_enabled() assert deeper to just before we commit to enabling the HW irq. v2: Add a FIXME explaining that placing the assertion so deep is not ideal, but a compromise for mock breadcrumbs. Signed-off-by:
-
Chris Wilson authored
Originally it was anticipated that timeouts would be a rare event, and so merit a warning that the test was incomplete. However, for igt we keep the timeout low, and hitting the timeout is intentional. It no longer necessitates a warning, but to be expected. Signed-off-by:
-
- 09 Nov, 2017 16 commits
-
-
Rodrigo Vivi authored
Signed-off-by:Rodrigo Vivi <rodrigo.vivi@intel.com>
-
Rodrigo Vivi authored
Signed-off-by: -
Chris Wilson authored
Whenever we want to unbind a vma, we must wait on all GPU activity to complete first. (This is what gives us the ability to do fine grained eviction and purging by only having to wait on the VMA that we need to unbind to proceed; though if pushed we can make it a rule that we are only allowed to unbind already idle VMA and move the burden of the work and organising the sleep onto the caller.) Currently, we might only sleep if the vma is still active on the GPU, but in principle i915_vma_unbind() always implies a sleep, so mark it up with a might_sleep(). Suggested-by:
-
Chris Wilson authored
vm_free_page() may call down into set_pages_array_wb() (which itself sleeps, on x86 at least) but only if on !llc and the caches overflow. Since this is unlikely, we only rarely trigger the error in practice, and so to make CI detection of this sleeping bug possible we want to mark the common vm_free_page() as a potential sleep. Suggested-by:
Daniel Vetter <daniel.vetter@ffwll.ch> Link: https://patchwork.freedesktop.org/patch/msgid/20171109213450.13875-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
Trying to enable printk debugging for GEM is fraught with the issue of spam; interactions with HW are very frequent and often boring. However, one instance where they are not so boring is just before a BUG; here ftrace provides a facility to dump its ringbuffer on an oops. So for CI let's enable trace_printk() to capture the last exchanges with HW as a death rattle. For example, [ 79.234110] ------------[ cut here ]------------ [ 79.234137] kernel BUG at drivers/gpu/drm/i915/intel_lrc.c:907! [ 79.234145] invalid opcode: 0000 [#1] SMP [ 79.234153] Dumping ftrace buffer: [ 79.234158] --------------------------------- ... [ 79.314044] gem_conc-1059 1..s1 79203443us : intel_lrc_irq_handler: bcs0 out[0]: ctx=5.2, seqno=145 [ 79.314089] gem_conc-1059 1..s. 79220800us : intel_lrc_irq_handler: bcs0 csb[1/1]: status=0x00000018:0x00000005 [ 79.314133] gem_conc-1059 1..s. 79220803us : intel_lrc_irq_handler: bcs0 out[0]: ctx=5.1, seqno=145 [ 79.314177] gem_conc-1062 2..s1 79230458us : intel_lrc_irq_handler: bcs0 in[0]: ctx=8.1, seqno=146 [ 79.314220] gem_conc-1062 2..s1 79230515us : intel_lrc_irq_handler: bcs0 in[0]: ctx=8.2, seqno=147 [ 79.314265] gem_conc-1059 1..s1 79230951us : intel_lrc_irq_handler: bcs0 csb[2/3]: status=0x00000012:0x00000008 [ 79.314309] gem_conc-1059 1..s1 79230954us : intel_lrc_irq_handler: bcs0 out[0]: ctx=8.2, seqno=147 [ 79.314353] gem_conc-1059 1..s1 79230954us : intel_lrc_irq_handler: bcs0 csb[3/3]: status=0x00008002:0x00000008 [ 79.314396] gem_conc-1059 1..s1 79230955us : intel_lrc_irq_handler: bcs0 out[0]: ctx=8.1, seqno=147 [ 79.314402] --------------------------------- v2: Tweak the formatting to be more consistent between in/out. v3: do {} while (0) stub macro protection Suggested-by:Michał Winiarski <michal.winiarski@intel.com> Signed-off-by:
-
Ville Syrjälä authored
Eliminate a ton of pointless 'dev' variables in the DP code, and pass around 'dev_priv' instead of 'dev'. v2: Rebase Acked-by:
Jani Nikula <jani.nikula@intel.com> Reviewed-by:
Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com> Signed-off-by:
-
Ville Syrjälä authored
No need to pass 'dev' or 'dev_priv' when the function already takes 'intel_dp'. Also let's prefer passing 'dev_priv' instead of 'dev' when we have to pass one or the other. Signed-off-by:
-
Ville Syrjälä authored
Remove intel_digital_port->port and replace its users with intel_encoder->port. intel_encoder->port is a superset of intel_digital_port->port, and it works correctly even for MST encoders. v2: Eliminate a few dp_to_dig_port()->base.port cases too (DK) Performed with cocci: @@ @@ struct intel_digital_port { ... - enum port port; ... } @@ struct intel_digital_port *D; expression E; @@ - D->port = E; @@ struct intel_digital_port *D; @@ - D->port + D->base.port @ expression E; @@ ( - dp_to_dig_port(E)->port + dp_to_dig_port(E)->base.port | - enc_to_dig_port(E)->port + to_intel_encoder(E)->port ) @@ expression E; @@ - to_intel_encoder(&E->base) + E @@ struct intel_digital_port *D; identifier I, M; @@ I = &D->base <... ( - D->base.M + I->M | - &D->base + I ) ...> @@ identifier D; expression E; identifier M; @@ D = enc_to_dig_port(&E->base) <... ( - D->base.M + E->M | - &D->base + E ) ...> @@ identifier D, DP; expression E; identifier M; @@ DP = enc_to_intel_dp(&E->base) <... ( - dp_to_dig_port(DP)->base.M + E->M | - &dp_to_dig_port(DP)->base + E ) ...> @@ expression E; identifier M; @@ ( - enc_to_dig_port(&E->base)->base.M + E->M | - enc_to_dig_port(&E->base)->base + E | - enc_to_mst(&E->base)->primary->base.port + E->port ) @@ expression E; identifier D; @@ - struct intel_digital_port *D = E; ... when != D Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com> Reviewed-by: -
Ville Syrjälä authored
Replace dig_port->port with encoder->port in the BXT DPLL selection. We can do this because both the master encoder and the fake MST encoders have the same encoder->port value, whereas using dig_port->port only worked for the master encoder since the fake encoders were't derived from intel_digital_port. This eliminates the DP MST special case. Do this by hand because spatch is having problems with the control flow due to the dig_port assignment happening in two different branches. Signed-off-by:
-
Ville Syrjälä authored
Replace crtc->config usage with the passed down crtc state. Also take the opportunity for some s/pipe_config/crtc_state/ while at it. Signed-off-by:
-
Ville Syrjälä authored
Rather than digging through encoder->crtc and crtc->config in the DPIO PHY functions, pass down the correct crtc state from the caller. Signed-off-by:
-
Ville Syrjälä authored
Extract the current crtc from the crtc state rather than via the legacy encoder->crtc pointer whenever possible. Signed-off-by:
-
Ville Syrjälä authored
Extract the current crtc from the crtc state rather than via the legacy encoder->crtc pointer whenever possible. Signed-off-by:
-
Ville Syrjälä authored
Extract the current crtc from the crtc state rather than via the legacy encoder->crtc pointer whenever possible. Signed-off-by:
-
Ville Syrjälä authored
Extract the current crtc from the crtc state rather than via the legacy encoder->crtc pointer whenever possible. Signed-off-by:
-
Chris Wilson authored
When we close the VMA, we unbind it from the ppgtt and tear down the page directory pointing at it. That may trigger us to return WC pages back to the system, requiring conversion back to WB which itself may sleep. That makes i915_vma_close() unsuitable for use inside the RCU read lock, which we need to hold to iterate the radixtree. The fix is quite simple, we can close all the VMA as we close the ppgtt, we only need to do that instead of closing them during destruction of the LUT. v2: Order between closing the LUT and the ppgtt is important; we use the vma inside the LUT as a means of retrieving the object, and so we must clear the LUT before freeing the VMA when closing the ppgtt. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103638 Fixes: 547da76b ("drm/i915: Hold rcu_read_lock when iterating over the radixtree (vma idr)") Fixes: d1b48c1e ("drm/i915: Replace execbuf vma ht with an idr") Signed-off-by:
-
- 08 Nov, 2017 8 commits
-
-
Michal Wajdeczko authored
We can program GUC_SHIM_CONTROL register with all expected bits without use of extra macro defined in fwif.h v2: rebased without pre-prod code v3: fixed typo Signed-off-by:
Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Sagar Arun Kamble <sagar.a.kamble@intel.com> Reviewed-by:
Sagar Arun Kamble <sagar.a.kamble@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20171103151816.62048-4-michal.wajdeczko@intel.comSigned-off-by:
-
Michal Wajdeczko authored
We don't keep the workarounds for pre-production hardware (see intel_detect_preproduction_hw) thus we can drop some extra steps during firmware upload needed only for unsupported platforms. Suggested-by:
-
Michal Wajdeczko authored
We silently assumed that DMA transfer will be completed within assumed timeout and thus we were waiting only at last step for GuC to become ready. Add intermediate wait to catch unexpected delays in DMA transfer. Signed-off-by:
-
Michal Wajdeczko authored
Transfer of GuC firmware requires few steps that currently are implemented in two large functions. Split this code into smaller functions to make these steps small and clear. Also be prepared for potential DMA xfer step failure. v2: rename function steps (Sagar) Signed-off-by:
-
Rafael Antognolli authored
The workaround for this is described as: "if RenderSurfaceState.Num_Multisamples > 1, disable RCC clock gating if RenderSurfaceState.Num_Multisamples == 1, set 0x7010[14] = 1" Further documentation in the internal bug referenced by the bspec suggest that any of the above suggestions should suffice to fix the issue. We are going with disabling RCC clock gating. Unfortunately, what we are doing doesn't match the name of the workaround, but at least it matches its description. This change improves CNL stability by avoiding some of the hangs seen in the platform. v2: Only disable RCC clock gating. Signed-off-by:
Rafael Antognolli <rafael.antognolli@intel.com> Reviewed-by:
-
Ville Syrjälä authored
Apparently setting up a bunch of GT registers before we've properly initialized the rest of the GT hardware leads to these setting being lost. So looks like I broke HSW with commit b7048ea1 ("drm/i915: Do .init_clock_gating() earlier to avoid it clobbering watermarks") by doing init_clock_gating() too early. This should actually affect other platforms as well, but apparently not to such a great degree. What I was ultimately after in that commit was to move the ilk_init_lp_watermarks() call earlier. So let's undo the damage and move init_clock_gating() back to where it was, and call ilk_init_lp_watermarks() just before the watermark state readout. This highlights how fragile and messed up our init order really is. I wonder why we even initialize the display before gem. The opposite order would make much more sense to me... v2: Keep WaRsPkgCStateDisplayPMReq:hsw early as it really must be done before all planes might get disabled. Cc: stable@vger.kernel.org Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Mark Janes <mark.a.janes@intel.com> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Oscar Mateo <oscar.mateo@intel.com> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Reported-by:
Mark Janes <mark.a.janes@intel.com> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103549 Fixes: b7048ea1 ("drm/i915: Do .init_clock_gating() earlier to avoid it clobbering watermarks") References: https://lists.freedesktop.org/archives/intel-gfx/2017-November/145432.htmlSigned-off-by:
-
Chris Wilson authored
The shared fence array is not autopruning and may continue to grow as an object is shared between new timelines. Take the opportunity when we think the object is idle (we have to confirm that any external fence is also signaled) to decouple all the fences. We apply a similar trick after waiting on an object, see commit e54ca977 ("drm/i915: Remove completed fences after a wait") v2: No longer need to handle the batch pool as a special case. v3: Need to trylock from within i915_vma_retire as this may be called form the shrinker - and we may later try to allocate underneath the reservation lock, so a deadlock is possible. References: https://bugs.freedesktop.org/show_bug.cgi?id=102936 Fixes: d07f0e59 ("drm/i915: Move GEM activity tracking into a common struct reservation_object") Fixes: 80b204bc ("drm/i915: Enable multiple timelines") Signed-off-by:
-
Chris Wilson authored
The handling of contexts are peculiar. Instead of tieing their vma to activity, we pin the context. This means that we cannot simply unbind the context object itself at will (which would normally cause us to wait for the vma to be idle), but must manually idle the GPU and retire requests first. A consequence of this peculiarity is when doing a last desperate attempt to recover memory. If the memory is tied up inside active context objects, we will fail to recover any memory simply by trying to unbind the objects without first doing a wait-for-idle. A side-effect of removing the call to shrinker_lock_uninterruptible() from i915_gem_shrinker_oom() was that we removed an unlocked wait-for-idle, and so lost the "natural" shrinkage of context objects. By replacing that with a locked wait from inside i915_gem_shrink(), we not only replace it with the ability to recover all context objects, but do so for all i915_gem_shrink_all() callers. v2: Switching requires request allocation, which is not permitted from inside the shrinker as it only uses ordinary allocations. References: https://bugs.freedesktop.org/show_bug.cgi?id=102936 Fixes: f2123818 ("drm/i915: Move dev_priv->mm.[un]bound_list to its own lock") Signed-off-by:
-
