- 16 Jan, 2019 3 commits
-
-
Chris Wilson authored
Make i915_gem_set_wedged() and i915_gem_unset_wedged() behaviour more consistent if called concurrently, and only do the wedging and reporting once, curtailing any possible race where we start unwedging in the middle of a wedge. 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/20190114210408.4561-2-chris@chris-wilson.co.uk
-
Lyude Paul authored
Something that I completely missed when implementing the new MST VCPI atomic helpers is that with those helpers, there's technically a chance of us having to grab additional modeset locks in ->compute_config() and furthermore, that means we have the potential to hit a normal modeset deadlock. However, because ->compute_config() only returns a bool this means we can't return -EDEADLK when we need to drop locks and try again which means we end up just failing the atomic check permanently. Whoops. So, fix this by modifying ->compute_config() to pass down an actual error code instead of a bool so that the atomic check can be restarted on modeset deadlocks. Thanks to Ville Syrjälä for pointing this out! Changes since v1: * Add some newlines * Return only -EINVAL from hsw_crt_compute_config() * Propogate return code from intel_dp_compute_dsc_params() * Change all of the intel_dp_compute_link_config*() variants * Don't miss if (hdmi_port_clock_valid()) branch in intel_hdmi_compute_config() [Cherry-picked from drm-misc-next to drm-intel-next-queued to fix linux-next & drm-tip conflict, while waiting for proper propagation of the DP MST series that this commit fixes. In hindsight, a topic branch might have been a better approach for it.] Signed-off-by:
Lyude Paul <lyude@redhat.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Fixes: eceae147 ("drm/dp_mst: Start tracking per-port VCPI allocations") Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109320Reviewed-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190115200800.3121-1-lyude@redhat.com (cherry picked from commit 96550555) Signed-off-by:
Jani Nikula <jani.nikula@intel.com> Acked-by:
Daniel Vetter <daniel@ffwll.ch>
-
Chris Wilson authored
mutex_lock_killable() returns -EINTR on failure, not the anticipate bool return like trylock. (Oh no, not again.) Fixes: 484d9a84 ("drm/i915/userptr: Avoid struct_mutex recursion for mmu_invalidate_range_start") Signed-off-by:
Matthew Auld <matthew.auld@intel.com>
-
- 15 Jan, 2019 9 commits
-
-
Chris Wilson authored
Move the debug pretty printer into a standalone routine prior to extending it in upcoming feature work. Signed-off-by:
Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190115212948.10423-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
Since commit 93065ac7 ("mm, oom: distinguish blockable mode for mmu notifiers") we have been able to report failure from mmu_invalidate_range_start which allows us to use a trylock on the struct_mutex to avoid potential recursion and report -EBUSY instead. Furthermore, this allows us to pull the work into the main callback and avoid the sleight-of-hand in using a workqueue to avoid lockdep. However, not all paths to mmu_invalidate_range_start are prepared to handle failure, so instead of reporting the recursion, deal with it by propagating the failure upwards, who can decide themselves to handle it or report it. v2: Mark up the recursive lock behaviour and comment on the various weak points. v3: Follow commit 3824e419 ("drm/i915: Use mutex_lock_killable() from inside the shrinker") and also use mutex_lock_killable(). v3.1: No leak on EINTR. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108375 References: 93065ac7 ("mm, oom: distinguish blockable mode for mmu notifiers") Signed-off-by:
-
Imre Deak authored
Registering an output for a non-existent port (on a given SKU) can lead to problems when trying to use the port, for instance timeouts during power well enabling. Since there are no strap bits for port detection we have to rely on VBT for this, so do that here. There are no known SKUs where any of the A-E ports are non-existent, so to reduce the likelihood of breakage due to incorrect VBT information, do this detection only for port F (which is known to be missing on some ICL SKUs). Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108915 Cc: Mika Kahola <mika.kahola@intel.com> Cc: Jani Nikula <jani.nikula@linux.intel.com> Signed-off-by:
Imre Deak <imre.deak@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20181220132604.25222-2-imre.deak@intel.comReviewed-by:
-
Imre Deak authored
We have already a function to detect DDI ports using VBT, so instead of opencoding the DDI specific version of this, move the opencoded part to the existing helper. Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: Mika Kahola <mika.kahola@intel.com> Signed-off-by:
-
Chris Wilson authored
As we may frequently mark the device as wedged to flush requests off it during the normal course of events, quite often we have a large state dump that is of no interest. Don't bother dumping it all if the engines are all idle. Signed-off-by:
-
Chris Wilson authored
drivers/gpu/drm/i915/i915_drv.h:1375: warning: Function parameter or member 'wakeref' not described in 'i915_perf_stream' Reported-by: kbuild-all@01.org Fixes: 6619c007 ("drm/i915/perf: Track the rpm wakeref") Signed-off-by:
-
Radhakrishna Sripada authored
intel_dp->dsc_dpcd is defined as an array making the if check redundant. Fixes: e845f099 ("drm/i915/dsc: Add Per connector debugfs node for DSC support/enable") Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Reported-by:
Nathan Chancellor <natechancellor@gmail.com> Signed-off-by:
Radhakrishna Sripada <radhakrishna.sripada@intel.com> Reviewed-by:
Manasi Navare <manasi.d.navare@intel.com> Reviewed-by:
-
Aditya Swarup authored
CNL macros for register groups CNL_PORT_TX_DW2_* / CNL_PORT_TX_DW5_* are configured incorrectly wrt definition of _CNL_PORT_TX_DW_GRP. v2: Jani suggested to keep the macros organized semantically i.e., by function, secondarily by port/pipe/transcoder.->(dw, port) Fixes: 4e53840f ("drm/i915/icl: Introduce new macros to get combophy registers") Cc: Clint Taylor <clinton.a.taylor@intel.com> Cc: Imre Deak <imre.deak@intel.com> Cc: Jani Nikula <jani.nikula@linux.intel.com> Signed-off-by:
Aditya Swarup <aditya.swarup@intel.com> Signed-off-by:
-
Chris Wilson authored
On Braswell, under heavy stress, if we update the GGTT while simultaneously accessing another region inside the GTT, we are returned the wrong values. To prevent this we stop the machine to update the GGTT entries so that no memory traffic can occur at the same time. This was first spotted in commit 5bab6f60 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Fri Oct 23 18:43:32 2015 +0100 drm/i915: Serialise updates to GGTT with access through GGTT on Braswell but removed again in forlorn hope with commit 4509276e Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Mon Feb 20 12:47:18 2017 +0000 drm/i915: Remove Braswell GGTT update w/a However, gem_concurrent_blit is once again only stable with the patch applied and CI is detecting the odd failure in forked gem_mmap_gtt tests (which smell like the same issue). Fwiw, a wide variety of CPU memory barriers (around GGTT flushing, fence updates, PTE updates) and GPU flushes/invalidates (between requests, after PTE updates) were tried as part of the investigation to find an alternate cause, nothing comes close to serialised GGTT updates. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105591 Testcase: igt/gem_concurrent_blit Testcase: igt/gem_mmap_gtt/*forked* References: 5bab6f60 ("drm/i915: Serialise updates to GGTT with access through GGTT on Braswell") References: 4509276e ("drm/i915: Remove Braswell GGTT update w/a") Signed-off-by:
-
- 14 Jan, 2019 23 commits
-
-
Chris Wilson authored
We have two classes of VM, global GTT and per-process GTT. In order to allow ourselves the freedom to mix both along call chains, distinguish the two classes with regards to their mutex and lockdep maps. Signed-off-by:
-
Juha-Pekka Heikkila authored
Primary and sprite plane enable on ILK-IVB may take two frames to complete Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103925Signed-off-by:
Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com> Signed-off-by:
-
Chris Wilson authored
Currently Ironlake operates under the assumption that rpm awake (and its error checking is disabled). As such, we have missed a few places where we access registers without taking the rpm wakeref and thus trigger warnings. intel_ips being one culprit. As this involved adding a potentially sleeping rpm_get, we have to rearrange the spinlocks slightly and so switch to acquiring a device-ref under the spinlock rather than hold the spinlock for the whole operation. To be consistent, we make the change in pattern common to the intel_ips interface even though this adds a few more atomic operations than necessary in a few cases. v2: Sagar noted the mb around setting mch_dev were overkill as we only need ordering there, and that i915_emon_status was still using struct_mutex for no reason, but lacked rpm. Signed-off-by:
John Harrison <John.C.Harrison@Intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190114142129.24398-21-chris@chris-wilson.co.uk
-
Chris Wilson authored
As the GT_IRQ power domain implies a wakeref, we can use it inplace of our existing redundant rpm grab. v2: Drop papering over forgetting to take the runtime wakeref in selftests Signed-off-by:
-
Chris Wilson authored
As we only release each power well once, we assume that each transcoder maps to a different domain. Complain if this is not so. Signed-off-by:
-
Chris Wilson authored
Track where and when we acquire and release the power well for pps access along the dp aux link, with a view to detecting if we leak any wakerefs. Signed-off-by:
-
Chris Wilson authored
On module load and unload, we grab the POWER_DOMAIN_INIT powerwells and transfer them to the runtime-pm code. We can use our wakeref tracking to verify that the wakeref is indeed passed from init to enable, and disable to fini; and across suspend. Signed-off-by:
-
Chris Wilson authored
The majority of runtime-pm operations are bounded and scoped within a function; these are easy to verify that the wakeref are handled correctly. We can employ the compiler to help us, and reduce the number of wakerefs tracked when debugging, by passing around cookies provided by the various rpm_get functions to their rpm_put counterpart. This makes the pairing explicit, and given the required wakeref cookie the compiler can verify that we pass an initialised value to the rpm_put (quite handy for double checking error paths). Signed-off-by:
-
Chris Wilson authored
Frequently, we use intel_runtime_pm_get/_put around a small block. Formalise that usage by providing a macro to define such a block with an automatic closure to scope the intel_runtime_pm wakeref to that block, i.e. macro abuse smelling of python. Signed-off-by:
-
Chris Wilson authored
Track the temporary wakerefs used within the selftests so that leaks are clear. v2: Add a couple of coarse annotations for mock selftests as we now loudly warn about the errors. Signed-off-by:
-
Chris Wilson authored
Keep track of the temporary rpm wakeref used for panel backlight access, so that we can cancel it immediately upon release and so more clearly identify leaks. Signed-off-by:
-
Chris Wilson authored
Keep track of the temporary rpm wakeref inside hotplug detection, so that we can cancel it immediately upon release and so clearly identify leaks. Signed-off-by:
-
Chris Wilson authored
Keep track of the rpm wakeref used for framebuffer access so that we can cancel upon release and so more clearly identify leaks. Signed-off-by:
-
Chris Wilson authored
Keep track of the temporary rpm wakerefs used for user access to the device, so that we can cancel them upon release and clearly identify any leaks. Signed-off-by:
-
Chris Wilson authored
Keep track of our acquired wakeref for interacting with the guc, so that we can cancel it upon release and so clearly identify leaks. Signed-off-by:
-
Chris Wilson authored
Track the wakeref used for temporary access to the device, and discard it upon release so that leaks can be identified. Signed-off-by:
-
Chris Wilson authored
Keep track of our wakeref used to keep the device awake so we can catch any leak. Signed-off-by:
-
Chris Wilson authored
As debugfs has a simple pattern of taking a rpm wakeref around the user access, we can track the local reference and drop it as soon as possible. Signed-off-by:
-
Chris Wilson authored
As sysfs has a simple pattern of taking a rpm wakeref around the user access, we can track the local reference and drop it as soon as possible. Signed-off-by:
-
Chris Wilson authored
Keep hold of the local wakeref used in error handling, to cancel the tracking upon release so that leaks can be identified. Signed-off-by:
-
Chris Wilson authored
Record the wakeref used for keeping the device awake as the GPU is executing requests and be sure to cancel the tracking upon parking. Signed-off-by:
-
Chris Wilson authored
The majority of runtime-pm operations are bounded and scoped within a function; these are easy to verify that the wakeref are handled correctly. We can employ the compiler to help us, and reduce the number of wakerefs tracked when debugging, by passing around cookies provided by the various rpm_get functions to their rpm_put counterpart. This makes the pairing explicit, and given the required wakeref cookie the compiler can verify that we pass an initialised value to the rpm_put (quite handy for double checking error paths). For regular builds, the compiler should be able to eliminate the unused local variables and the program growth should be minimal. Fwiw, it came out as a net improvement as gcc was able to refactor rpm_get and rpm_get_if_in_use together, v2: Just s/rpm_put/rpm_put_unchecked/ everywhere, leaving the manual mark up for smaller more targeted patches. v3: Mention the cookie in Returns Signed-off-by:
-
Chris Wilson authored
Everytime we take a wakeref, record the stack trace of where it was taken; clearing the set if we ever drop back to no owners. For debugging a rpm leak, we can look at all the current wakerefs and check if they have a matching rpm_put. v2: Use skip=0 for unwinding the stack as it appears our noinline function doesn't appear on the stack (nor does save_stack_trace itself!) v3: Allow rpm->debug_count to disappear between inspections and so avoid calling krealloc(0) as that may return a ZERO_PTR not NULL! (Mika) v4: Show who last acquire/released the runtime pm Signed-off-by:
-
- 10 Jan, 2019 5 commits
-
-
Rodrigo Vivi authored
Signed-off-by:Rodrigo Vivi <rodrigo.vivi@intel.com>
-
Chris Wilson authored
The wait-for-idle used from within the shrinker_lock_uninterruptible depends on the struct_mutex locking state being known and declared to i915_request_wait(). As it is conceivable that we reach the vmap notifier from underneath struct_mutex (and so keep on relying on the mutex_trylock_recursive), we should not blindly call i915_request_wait. In the process we can remove the dubious polling to acquire struct_mutex, and simply act, or not, on a successful trylock. Signed-off-by:
-
Chris Wilson authored
If the current process is being killed (it was interrupted with SIGKILL or equivalent), it will not make any progress in page allocation and we can abort performing the shrinking on its behalf. So we can use mutex_lock_killable() instead (although this path should only be reachable from kswapd currently). Tvrtko pointed out that it should also be reachable from debugfs, which he would prefer retain its interruptiblity. As a compromise, killable is a step in the right direction! Signed-off-by:
-
Chris Wilson authored
If we find an incompletely setup vma inside the request/engine at the time of a hang, it may not have vma->pages initialised, so skip capturing the object before we iterate over NULL. Spotted by Matthew in preparation for using unpinned vma to track engine state. Signed-off-by:
-
Daniele Ceraolo Spurio authored
The only gen8+ platform that has the feature is BDW, but we don't define the feature flag on any BDW platform and we only have partial support in the gen8 path (irq enabling code, but no handler). The only thing we could do in the irq handler is report the error to userspace, but no one asked/cared about that since BDW was released so it is relatively safe to assume that even if we added the message no one would look at it. Just drop the dead code from the driver instead. Signed-off-by:
Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Reviewed-by:
-
