- 20 Jun, 2019 9 commits
-
-
Chris Wilson authored
Our intel_rings are always flushed as they are continually used to submit commands to the GPU, and so do not need to be flushed on unpinning. This avoids pulling in the flush_ggtt_writes locking into our context unpin, which we want to allow from atomic context (for simplicity). 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/20190619203504.4220-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
If we have multiple contexts of equal priority pending execution, activate a timer to demote the currently executing context in favour of the next in the queue when that timeslice expires. This enforces fairness between contexts (so long as they allow preemption -- forced preemption, in the future, will kick those who do not obey) and allows us to avoid userspace blocking forward progress with e.g. unbounded MI_SEMAPHORE_WAIT. For the starting point here, we use the jiffie as our timeslice so that we should be reasonably efficient wrt frequent CPU wakeups. Testcase: igt/gem_exec_scheduler/semaphore-resolve Signed-off-by:
Mika Kuoppala <mika.kuoppala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190620142052.19311-2-chris@chris-wilson.co.uk
-
Chris Wilson authored
When using a global seqno, we required a precise stop-the-workd event to handle preemption and unwind the global seqno counter. To accomplish this, we would preempt to a special out-of-band context and wait for the machine to report that it was idle. Given an idle machine, we could very precisely see which requests had completed and which we needed to feed back into the run queue. However, now that we have scrapped the global seqno, we no longer need to precisely unwind the global counter and only track requests by their per-context seqno. This allows us to loosely unwind inflight requests while scheduling a preemption, with the enormous caveat that the requests we put back on the run queue are still _inflight_ (until the preemption request is complete). This makes request tracking much more messy, as at any point then we can see a completed request that we believe is not currently scheduled for execution. We also have to be careful not to rewind RING_TAIL past RING_HEAD on preempting to the running context, and for this we use a semaphore to prevent completion of the request before continuing. To accomplish this feat, we change how we track requests scheduled to the HW. Instead of appending our requests onto a single list as we submit, we track each submission to ELSP as its own block. Then upon receiving the CS preemption event, we promote the pending block to the inflight block (discarding what was previously being tracked). As normal CS completion events arrive, we then remove stale entries from the inflight tracker. v2: Be a tinge paranoid and ensure we flush the write into the HWS page for the GPU semaphore to pick in a timely fashion. Signed-off-by:
-
Daniele Ceraolo Spurio authored
With multiple uncore to initialize (GT vs Display), it makes little sense to have the vgpu_check inside uncore_init(). We also have a catch-22 scenario where the uncore is required to read the vgpu capabilities while the vgpu capabilities are required to decide if we need to initialize forcewake support. To remove this circular dependency, we can perform the required MMIO access by mmapping just the vgtif shared page in mmio space and use raw accessors. v2: rename check_vgpu to detect_vgpu (Chris) Signed-off-by:
Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Cc: Zhenyu Wang <zhenyuw@linux.intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by:
-
Daniele Ceraolo Spurio authored
We'd like to introduce a display uncore with no forcewake domains, so let's avoid wasting memory and be ready to allocate only what we need. Even without multiple uncore, we still don't need all the domains on all gens. v2: avoid hidden control flow, improve checks (Tvrtko), fix IVB special case, add failure injection point Signed-off-by:
-
Daniele Ceraolo Spurio authored
We always call some of the setup/cleanup functions for forcewake, even if the feature is not actually available. Skipping these operations if forcewake is not available saves us some operations on older gens and prepares us for having a forcewake-less display uncore. v2: do not make suspend/resume functions forcewake-specific (Chris, Tvrtko), use GEM_BUG_ON in internal forcewake-only functions (Tvrtko) Signed-off-by:
-
Daniele Ceraolo Spurio authored
Let's get rid of it before it proliferates, since with split GT/Display uncores the container_of won't work anymore. I've kept the rpm pointer as well to minimize the pointer chasing in the MMIO accessors. v2: swap parameter order for intel_uncore_init_early (Tvrtko) Signed-off-by:
-
Daniele Ceraolo Spurio authored
uncore_sanitize performs no action on the uncore structure and just calls intel_sanitize_gt_powersave, so we can just call the latter directly. Signed-off-by:
Jani Nikula <jani.nikula@intel.com> Signed-off-by:
-
Daniele Ceraolo Spurio authored
Instead of going through the if-else chain every time, let's save the function in the uncore structure. Note that the new functions are purposely not used from the reg read/write functions to keep the inlining there. While at it, use the new macro to call the old ones to clean the code a bit. v2: Rename macros for no-forcewake function assignment (Tvrtko) Signed-off-by:
-
- 19 Jun, 2019 8 commits
-
-
Chris Wilson authored
Remember to keep the rings pinned as well as the context image until the GPU is no longer active. v2: Introduce a ring->pin_count primarily to hide the mock_ring that doesn't fit into the normal GGTT vma picture. v3: Order is important in teardown, ringbuffer submission needs to drop the pin count on the engine->kernel_context before it can gleefully free its ring. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110946 Fixes: ce476c80 ("drm/i915: Keep contexts pinned until after the next kernel context switch") Signed-off-by:
-
Matt Roper authored
EHL has a mux on combo PHY A that allows it to be driven either by an internal display (DDI-A or DSI DPHY) or by an external display (DDI-D). This is a motherboard design decision that can not be changed on the fly. Unfortunately there are no strap registers that allow us to detect the board configuration directly, so let's use the VBT to try to figure it out and program the mux accordingly. For now if we run across a broken VBT that tries to claim that PHY A is attached to both internal and external displays at the same time, we'll resolve the conflict in favor of the internal display. To help debug these kind of bad VBT's, let's also add a quick DRM_DEBUG message during child device parsing so that it's easier to understand these cases if they show up in bug reports. v2: - Confirmed that VBT's dvo port refers to the DDI and not the PHY. Thus we can check more explicitly for (ddi_d && !(ddi_a || dsi)). If a bad VBT contradicts itself, let internal display win. (Ville) v3: - Switch condition from !IS_ICELAKE to IS_ELKHARTLAKE. Although the convention is usually to assume that future platforms will inherit all current platform behavior, this feels more like a one-platform quirk. (Ville) - Update commit message to describe what we do if/when we encounter broken VBT's, and note that the new debug print during child device parsing is intentional. Cc: Clint Taylor <Clinton.A.Taylor@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by:
Matt Roper <matthew.d.roper@intel.com> Reviewed-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190618175131.9139-1-matthew.d.roper@intel.com
-
Chris Wilson authored
In the unlikely case the request completes while we regard it as not even executing on the GPU (see the next patch!), we have to flush any pending execution callbacks at retirement and ensure that we do not add any more. Signed-off-by:
-
Chris Wilson authored
With the upcoming change to automanaged i915_active, the intent is that whenever we wait on the set of active fences, they are signaled and collected. The requirement is that all successful returns from i915_request_wait() signal the fence, so fixup the one remaining path where we may return before the interrupt has been run. Signed-off-by:
-
Jani Nikula authored
Signed-off-by: -
Chris Wilson authored
Since commit eb8d0f5a ("drm/i915: Remove GPU reset dependence on struct_mutex"), the I915_WAIT_LOCKED flags passed to i915_request_wait() has been defunct. Now go ahead and remove it from all callers. References: eb8d0f5a ("drm/i915: Remove GPU reset dependence on struct_mutex") Signed-off-by:
-
Chris Wilson authored
The process_csb routine from execlists_submission is incompatible with the GuC backend. Add a warning to detect if we accidentally end up in the wrong spot. Signed-off-by:
-
Chris Wilson authored
The idea behind keeping the saturation mask local to a context backfired spectacularly. The premise with the local mask was that we would be more proactive in attempting to use semaphores after each time the context idled, and that all new contexts would attempt to use semaphores ignoring the current state of the system. This turns out to be horribly optimistic. If the system state is still oversaturated and the existing workloads have all stopped using semaphores, the new workloads would attempt to use semaphores and be deprioritised behind real work. The new contexts would not switch off using semaphores until their initial batch of low priority work had completed. Given sufficient backload load of equal user priority, this would completely starve the new work of any GPU time. To compensate, remove the local tracking in favour of keeping it as global state on the engine -- once the system is saturated and semaphores are disabled, everyone stops attempting to use semaphores until the system is idle again. One of the reason for preferring local context tracking was that it worked with virtual engines, so for switching to global state we could either do a complete check of all the virtual siblings or simply disable semaphores for those requests. This takes the simpler approach of disabling semaphores on virtual engines. The downside is that the decision that the engine is saturated is a local measure -- we are only checking whether or not this context was scheduled in a timely fashion, it may be legitimately delayed due to user priorities. We still have the same dilemma though, that we do not want to employ the semaphore poll unless it will be used. v2: Explain why we need to assume the worst wrt virtual engines. Fixes: ca6e56f6 ("drm/i915: Disable semaphore busywaits on saturated systems") Signed-off-by:
-
- 18 Jun, 2019 16 commits
-
-
José Roberto de Souza authored
To do frontbuffer tracking we are depending on Display WA #0884 to exit PSR when there is a frontbuffer modification but according to user reports a write to CURSURFLIVE do not cause PSR to exit in older gens so lets force a PSR exit. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110799 Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by:
José Roberto de Souza <jose.souza@intel.com> Reviewed-by:
Rodrigo Vivi <rodrigo.vivi@intel.com> Tested-by:
Thomas Rohwer <trohwer85@gmail.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190617195154.30292-1-jose.souza@intel.com
-
Chris Wilson authored
This has caught me out on countless occasions, when we retrieve a pointer from the submission/execlists backend, it does not carry a reference to the context or ring. Those are only pinned while the request is active, so if we see the request is already completed, it may be in the process of being retired and those pointers defunct. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110938 Fixes: 3a068721 ("drm/i915: Show ring->start for the ELSP context/request queue") Signed-off-by:
-
Chris Wilson authored
Be sure to cleanup after live_evict by flushing any residual state off the GPU using igt_flush_test. Tvrtko mentioned that it is probably wise to stop repeating this ad hoc around the tests and implement a live test runner. Signed-off-by:
-
Chris Wilson authored
Previously, we wanted to shrink the pages of freed objects before they were finally RCU collected. However, by removing the struct_mutex serialisation around the active reference, we need to acquire an extra reference around the wait. Unfortunately this means that we have to skip objects that are waiting RCU collection. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110937Signed-off-by:
-
Robert M. Fosha authored
Updates the live_workarounds selftest to handle whitelisted registers that are flagged as read only. Signed-off-by:
Robert M. Fosha <robert.m.fosha@intel.com> Signed-off-by:
John Harrison <John.C.Harrison@Intel.com> Reviewed-by:
-
John Harrison authored
Updated whitelist table for ICL. v2: Reduce changes to just those required for media driver until the selftest can be updated to support the new features of the other entries. Signed-off-by:
-
John Harrison authored
Newer hardware requires setting up whitelists on engines other than render. So, extend the whitelist code to support all engines. Signed-off-by:
-
John Harrison authored
Newer hardware adds flags to the whitelist work-around register. These allow per access direction privileges and ranges. Signed-off-by:
-
Ville Syrjälä authored
We have full infoframe readout now so we can replace the PIPE_CONF_CHECK_BOOL_INCOMPLETE(has_infoframe) with the normal PIPE_CONF_CHECK_BOOL(has_infoframe). Signed-off-by:
Imre Deak <imre.deak@intel.com>
-
Ville Syrjälä authored
Rename pipe_config_err() to pipe_config_mismatch(), and also print whether we're doing the fastset check or the sw vs. hw state readout check. Should make the logs a bit less confusing when they're not filled with what looks like a real error. Also rename the 'adjust' variable to 'fastset' to make it clear what it means. Signed-off-by:
-
Ville Syrjälä authored
Now that intel_pipe_config_compare() no longer clobbers the passed in state we can make both crtc states const. And while at we simplify the calling convention, and clean up intel_compare_link_m_n() a bit. Signed-off-by:
-
Ville Syrjälä authored
We're now calling intel_pipe_config_compare(..., true) uncoditionally which means we're always going clobber the calculated M/N values with the old values if the fuzzy M/N check passes. That causes problems because the fuzzy check allows for a huge difference in the values. I'm actually tempted to just make the M/N checks exact, but that might prevent fastboot from kicking in when people want it. So for now let's overwrite the computed values with the old values only if decide to skip the modeset. v2: Copy has_drrs along with M/N M2/N2 values Cc: stable@vger.kernel.org Cc: Blubberbub@protonmail.com Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Cc: Hans de Goede <hdegoede@redhat.com> Tested-by: Blubberbub@protonmail.com Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110782 Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110675 Fixes: d19f958d ("drm/i915: Enable fastset for non-boot modesets.") Signed-off-by:
-
Chris Wilson authored
Since commit 1ba62714 ("drm: Add reservation_object to drm_gem_object"), struct drm_gem_object grew its own builtin reservation_object rendering our own private one bloat. Remove our redundant reservation_object and point into obj->base.resv instead. References: 1ba62714 ("drm: Add reservation_object to drm_gem_object") Signed-off-by:
Matthew Auld <matthew.auld@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190618125858.7295-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
Though we pin the context first before taking the pm wakeref, during retire we need to unpin before dropping the pm wakeref (breaking the "natural" onion). During the unpin, we may need to attach a cleanup operation on to the engine wakeref, ergo we want to keep the engine awake until after the unpin. v2: Push the engine wakeref into the barrier so we keep the onion unwind ordering in the request itself Fixes: ce476c80 ("drm/i915: Keep contexts pinned until after the next kernel context switch") Signed-off-by:
-
Chris Wilson authored
If the user is clearing the log buffer too slowly, we overflow. As this is an expected condition, and the driver tries to handle it, reduce the error message down to a notice. Michal mentioned that another cause would be incorrect reset handling, so we don't want to lose the notification entirely. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110817Signed-off-by:
-
Jani Nikula authored
Fix the plethora of Sphinx build errors after moving the display files under a subdirectory. Fixes: 379bc100 ("drm/i915: move modesetting output/encoder code under display/") Fixes: df0566a6 ("drm/i915: move modesetting core code under display/") Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Reviewed-by:
-
- 17 Jun, 2019 7 commits
-
-
Matt Roper authored
Although EHL introduces a new PCH, the South Display part of the PCH that we care about is nearly identical to ICP, just with some pins remapped. Most notably, Port C is mapped to the pins that ICP uses for TC Port 1. Signed-off-by:
-
Chris Wilson authored
Currently, we perform a locked update of the shadow entry when allocating a page directory entry such that if two clients are concurrently allocating neighbouring ranges we only insert one new entry for the pair of them. However, we also need to serialise both clients wrt to the actual entry in the HW table, or else we may allow one client or even a third client to proceed ahead of the HW write. My handwave before was that under the _pathological_ condition we would see the scratch entry instead of the expected entry, causing a temporary glitch. That starvation condition will eventually show up in practice, so fix it. The reason for the previous cheat was to avoid having to free the extra allocation while under the spinlock. Now, we keep the extra entry allocated until the end instead. v2: Fix error paths for gen6 Fixes: 1d1b5490 ("drm/i915/gtt: Replace struct_mutex serialisation for allocation") Signed-off-by:
Mika Kuoppala <mika.kuoppala@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190617140426.7203-1-chris@chris-wilson.co.uk
-
Lucas De Marchi authored
In intel_package_header version 2 there's a new field in the fw_info table that must be 0, otherwise it's not the correct DMC firmware. Add a check for version 2 or later. Signed-off-by:
Lucas De Marchi <lucas.demarchi@intel.com> Reviewed-by:
Anusha Srivatsa <anusha.srivatsa@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190607091230.1489-10-lucas.demarchi@intel.com
-
Lucas De Marchi authored
parse_csr_fw() is responsible to set up several fields in struct intel_csr, including the payload. We don't need to assign it again. Signed-off-by:
-
Lucas De Marchi authored
Main difference is that now there are up to 20 MMIOs that can be set and a lot of noise due to the struct changing the fields in the middle. Signed-off-by:
-
Lucas De Marchi authored
Complete the extraction of functions to parse specific parts of the firmware. The return of the function parse_csr_fw() is now redundant since it already sets the dmc_payload field. Changing it is left for later to avoid noise in the commit. Signed-off-by:
-
Lucas De Marchi authored
Like parse_csr_fw_css() this parses the package_header from firmware and saves the relevant fields in the csr struct. In this function we also lookup the fw_info we are interested in. Signed-off-by:
-
