- 18 Mar, 2016 3 commits
-
-
Tvrtko Ursulin authored
By reading the CSB (slow MMIO accesses) into a temporary local buffer we can decrease the duration of holding the execlist lock. Main advantage is that during heavy batch buffer submission we reduce the execlist lock contention, which should decrease the latency and CPU usage between the submitting userspace process and interrupt handling. Downside is that we need to grab and relase the forcewake twice, but as the below numbers will show this is completely hidden by the primary gains. Testing with "gem_latency -n 100" (submit batch buffers with a hundred nops each) shows more than doubling of the throughput and more than halving of the dispatch latency, overall latency and CPU time spend in the submitting process. Submitting empty batches ("gem_latency -n 0") does not seem significantly affected by this change with throughput and CPU time improving by half a percent, and overall latency worsening by the same amount. Above tests were done in a hundred runs on a big core Broadwell. v2: * Overflow protection to local CSB buffer. * Use closer dev_priv in execlists_submit_requests. (Chris Wilson) v3: Rebase. v4: Added commend about irq needed to be disabled in execlists_submit_request. (Chris Wilson) Signed-off-by:Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by:
Chris Wilsno <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1458219586-20452-1-git-send-email-tvrtko.ursulin@linux.intel.com
-
Tvrtko Ursulin authored
Where we have a request we can use req->i915 directly instead of going through the engine and device. Coccinelle script: @@ function f; identifier r; @@ f(..., struct drm_i915_gem_request *r, ...) { ... - engine->dev->dev_private + r->i915 ... } @@ struct drm_i915_gem_request *req; @@ ( req-> - engine->dev->dev_private + i915 ) Signed-off-by: -
Tvrtko Ursulin authored
Signed-off-by:
-
- 17 Mar, 2016 26 commits
-
-
Imre Deak authored
Add support for forcing an error at selected places in the driver. As an example add 4 options to fail during driver loading. Requested by Chris. v2: - Add fault point for modeset initialization - Print debug message when injecting an error v3: - Rename inject_fault to inject_load_failure, rename the related macros and helper accordingly (Chris) - Use a counter instead of a mask to identify the failure point (Daniel) - Mark the module option as _unsafe and keep i915_params ordered (Joonas) v4: - Rebase on latest -nightly v5: - Use DRM_INFO instead of DRM_DEBUG_DRIVER, making it clearer in CI reports that a following error message is expected (IRC r-b from Chris on v5) CC: Chris Wilson <chris@chris-wilson.co.uk> CC: Daniel Vetter <daniel.vetter@ffwll.ch> CC: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by:
Imre Deak <imre.deak@intel.com> Reviewed-by:
-
Imre Deak authored
Move the cleanup of the power domain HW state on the error path to the same function where the corresponding init call was called from. I noticed this problem when loading the module with load failure injection enabled, making i915_load_modeset_init() fail. CC: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Imre Deak authored
According to the new init phases scheme we should register the device making it available via some kernel internal or user space interface as the last step in the init sequence, so move the corresponding code to a separate function. Also add a TODO comment about code that still needs to be moved around to one of the init phases functions depending on what the role and effect of that code is. No functional change, except for the reordering of the unload time unregistration steps of sysfs wrt. acpi and opregion. Suggested by Chris. v3: - rename i915_driver_init_register to i915_driver_init_frameworks (Chris) - rename i915_driver_init_frameworks to i915_driver_register (Daniel) CC: Chris Wilson <chris@chris-wilson.co.uk> CC: Daniel Vetter <daniel.vetter@ffwll.ch> Signed-off-by:
-
Imre Deak authored
According to the new init phases scheme we should have a definite step in the init sequence where we setup things requiring accessing the device, so move the corresponding code to separate function. The steps in this init phase should avoid exposing the driver via some interface, which is done in the last registration init phase. This changae also has the benefit of making the error path cleaner both in the new function and i915_driver_load()/unload(). No functional change. Suggested by Chris. CC: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Imre Deak authored
According to the new init phases scheme we should have a definite step in the init sequence where MMIO access is setup, so move the corresponding code to a separate function. This also has the benefit of making the error path cleaner both in the new function and in i915_driver_load()/unload(). No functional change. Suggested by Chris. Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Imre Deak authored
According to the new init phases scheme we should initialize "SW-only" state not requiring accessing the device as the very first step, so that the reasoning about dependencies of later steps becomes easier. So move these init steps into a separate function. This also has the benefit of making the error path cleaner both in the new function and int i915_driver_load()/unload(). No functional change. Suggested by Chris. CC: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Imre Deak authored
Move the opregion unregistration earlier to match its corresponding registration order. Signed-off-by:
-
Imre Deak authored
Move the GTT,MSI IRQ cleanup later so that it matches their corresponding init order. Also fix the order of these calls wrt. each other to match their corresponding init order. Signed-off-by:
-
Imre Deak authored
Move the power domain uninitialization later so that it matches its corresponding init order. Since we access the HW during the later unitialization steps keep a wake reference until after the last such step. Signed-off-by:
-
Imre Deak authored
We should register all the interfaces before we enable runtime PM. Signed-off-by:
-
Imre Deak authored
According to the new init phases scheme we should register the driver with frameworks/userspace only one the device is setup fully. So move the shrinker registration later accordingly. Also fix the shrinker unregistration order wrt. the acpi unregistration to fix the corresponding init order. Signed-off-by:
-
Imre Deak authored
We require the device to be powered only before accessing it, so we can move this call later. Signed-off-by:
-
Imre Deak authored
The only steps requiring device access is the fence and swizzling initialization, so split these out keeping them in their current place and move the rest of init steps earlier. v2-v3: - unchanged v4: - move call to i915_gem_detect_bit_6_swizzle() to i915_gem_load_init_fences() and preserve the original order of the detection of HW fence capailities wrt. swizzling (Chris) CC: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Imre Deak authored
This init step accesses the device, but doesn't have any device specific side effect. It also sets up some platform specific attributes that may be required early, so move it earlier. Signed-off-by:
-
Imre Deak authored
Split out the part initing the clock gating hooks and move it earlier. Add a new NOP hook for platforms without the need to apply clockgating or workaround settings, so that the hook can be called unconditionally. Also add a WARN for future platforms that forget to add a hook. The rest of the hooks in intel_init_pm() should be inited in the same way, but atm some of the hooks are set only conditionally, so before doing this we need to make the setup unconditional and use instead some flags. v2: - add a NOP hook and WARN if no hook is set for the platform (Chris) - use the term hook instead of callback for these functions (Jani) v3: - remove the GEN4() check it's already covered by earlier platform checks (Chris) CC: Jani Nikula <jani.nikula@intel.com> CC: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
-
Imre Deak authored
All of this is SW only initialization so we can move them earlier. Move the mutex init where the rest of the locks are inited. While at it also convert dev to dev_priv. v2: - use the term hook instead of callback for these functions (Jani) CC: Jani Nikula <jani.nikula@intel.com> Signed-off-by:
-
Imre Deak authored
Most of the IRQ init is setting up hooks so move that part earlier. Leave the pm_qos_add_request() call in place. Signed-off-by:
-
Imre Deak authored
These are all SW only init steps not accessing the device and they only need the platform identification macros to work, which are already available earlier, so move these init steps earlier. Signed-off-by:
-
Imre Deak authored
MCHBAR is cleaned up in i915_mmio_cleanup(), so the separate call in i915_driver_load() is incorrect. CC: David Weinehall <david.weinehall@intel.com> Fixes: ad5c3d3f ("drm/i915: Move MCHBAR setup earlier during init") Signed-off-by:
David Weinehall <david.weinehall@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1458128348-15730-2-git-send-email-imre.deak@intel.com
-
Mika Kuoppala authored
In full gpu reset we prime all engines and reset domains corresponding to each engine. Per engine reset is just a special case of this process wherein only a single engine is reset. This change is aimed to modify relevant functions to achieve this. There are some other steps we carry out in case of engine reset which are addressed in later patches. Reset func now accepts a mask of all engines that need to be reset. Where per engine resets are supported, error handler populates the mask accordingly otherwise all engines are specified. v2: ALL_ENGINES mask fixup, better for_each_ring_masked (Chris) v3: Whitespace fixes (Chris) v4: Rebase due to s/ring/engine Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by:
Arun Siluvery <arun.siluvery@linux.intel.com> Reviewed-by:
-
Jani Nikula authored
We've been accumulating code across the driver that depends on the VBT specific structures and defines. The VBT is an uncontrollable beast. Encourage encapsulation of the VBT data by hiding the structures and defines in a private header only to be included from intel_bios.c. Reviewed-by:
Daniel Vetter <daniel.vetter@ffwll.ch> Signed-off-by:
Jani Nikula <jani.nikula@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1458125015-7931-7-git-send-email-jani.nikula@intel.com
-
Jani Nikula authored
drivers/gpu/drm/i915/intel_dpll_mgr.c:1200:32: warning: Using plain integer as NULL pointer Fixes: 304b65cb ("drm/i915: Move SKL/KLB pll selection logic to intel_dpll_mgr.c") Reviewed-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by:
-
-
Maarten Lankhorst authored
Warn for the wrong mask in enable only. Disable will have the wrong mask now because the new state is committed before disabling the old state. Changes since v1: - Use crtc_mask (Durgadoss) - Rebase. Signed-off-by:
Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1457944075-14123-3-git-send-email-maarten.lankhorst@linux.intel.comReviewed-by:
Ander Conselvan de Oliveira <conselvan2@gmail.com>
-
Maarten Lankhorst authored
This makes it easier to verify correct dpll setup with only a single crtc. It is also useful to detect double dpll enable/disable. Changes since v1: - Rebase on top of Ander's dpll rework. - Change debugfs active to a mask. - Change enabled_crtcs and active_crtcs to unsigned. Signed-off-by:
-
Lyude authored
After unplugging a DP MST display from the system, we have to go through and destroy all of the DRM connectors associated with it since none of them are valid anymore. Unfortunately, intel_dp_destroy_mst_connector() doesn't do a good enough job of ensuring that throughout the destruction process that no modesettings can be done with the connectors. As it is right now, intel_dp_destroy_mst_connector() works like this: * Take all modeset locks * Clear the configuration of the crtc on the connector, if there is one * Drop all modeset locks, this is required because of circular dependency issues that arise with trying to remove the connector from sysfs with modeset locks held * Unregister the connector * Take all modeset locks, again * Do the rest of the required cleaning for destroying the connector * Finally drop all modeset locks for good This only works sometimes. During the destruction process, it's very possible that a userspace application will attempt to do a modesetting using the connector. When we drop the modeset locks, an ioctl handler such as drm_mode_setcrtc has the oppurtunity to take all of the modeset locks from us. When this happens, one thing leads to another and eventually we end up committing a mode with the non-existent connector: [drm:intel_dp_link_training_clock_recovery [i915]] *ERROR* failed to enable link training [drm:intel_dp_aux_ch] dp_aux_ch timeout status 0x7cf0001f [drm:intel_dp_start_link_train [i915]] *ERROR* failed to start channel equalization [drm:intel_dp_aux_ch] dp_aux_ch timeout status 0x7cf0001f [drm:intel_mst_pre_enable_dp [i915]] *ERROR* failed to allocate vcpi And in some cases, such as with the T460s using an MST dock, this results in breaking modesetting and/or panicking the system. To work around this, we now unregister the connector at the very beginning of intel_dp_destroy_mst_connector(), grab all the modesetting locks, and then hold them until we finish the rest of the function. CC: stable@vger.kernel.org Signed-off-by:
Lyude <cpaul@redhat.com> Signed-off-by:
Rob Clark <rclark@redhat.com> Reviewed-by:
-
- 16 Mar, 2016 11 commits
-
-
Alex Dai authored
This version of GuC firmware fixes the engine reset issue where golden context LRC address is treated as page index by mistake. It also fixes the problem that scheduler stops submiting to one engine when the other engine work queue is full. Signed-off-by:
Alex Dai <yu.dai@intel.com> Reviewed-by:
Dave Gordon <david.s.gordon@intel.com> Signed-off-by:
Rodrigo Vivi <rodrigo.vivi@intel.com>
-
Jani Nikula authored
Favor a single point of truth instead of duplicating the information. The change also filters out unsupported DSI ports at this stage, accepting only ports A and C, instead of waiting until the port checks. Reviewed-by:
-
Jani Nikula authored
Use the connector type instead of VBT directly to decide which backlight mechanism to use on VLV/CHV. (Indirectly, this is the same thing, but hides the VBT use.) Reviewed-by:
-
Jani Nikula authored
Hide knowledge about VBT child devices in intel_bios.c. v2: Move port check to intel_bios.c (Sivakumar) Reviewed-by:
-
Jani Nikula authored
Hide knowledge about VBT child devices in intel_bios.c. Reviewed-by:
-
Jani Nikula authored
Hide knowledge about VBT child devices in intel_bios.c. Reviewed-by:
-
Jani Nikula authored
Hide knowledge about VBT child devices in intel_bios.c. v2: also move int_tv_support check to intel_bios.c (Sivakumar) Reviewed-by:
-
Jani Nikula authored
BXT isn't as limited as BYT and CHT regarding DSI pipes and ports. Reviewed-by:
-
Jani Nikula authored
Make the code easier to read and update. No functional changes. Reviewed-by:
-
Jani Nikula authored
Nicer for eDP (actually "EDP" instead of "D"), and makes future expansion for DSI transcoders easier. Reviewed-by:
-
Jani Nikula authored
Same as for_each_dsi_port, but for general use. Leave the for_each_dsi_port version around as an "alias" for now to not cause too much churn. No functional changes. Reviewed-by:
-
