- 08 Jun, 2017 6 commits
-
-
Chris Wilson authored
Originally we would enable and disable the breadcrumb interrupt immediately on demand. This was slow enough to have a large impact (>30%) on tasks that hopped between engines. However, by using a shadow to keep the irq alive for an extra interrupt (see commit 67b807a8 ("drm/i915: Delay disabling the user interrupt for breadcrumbs")) and by recently reducing the cost in adding ourselves to the signal tree, we no longer need to spin-request during await_request to avoid delays in throughput tests. Without the earlier patches to stop the wakeup when signaling if the irq was already active, we saw no improvement in execbuf overhead (and corresponding contention in other clients) despite the removal of the spinner in a simple test like glxgears. This means there will be scenarios where now we spend longer enabling the interrupt than we would have spent spinning, but these are not likely to have as noticeable an impact as the high frequency test cases (where there should not be any regression). Ulterior motive: generalising the engine->sync_to to handle different types of semaphores and non-semaphores. Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Oscar Mateo <oscar.mateo@intel.com> Reviewed-by:
Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by:
Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170608111405.16466-4-chris@chris-wilson.co.uk
-
Chris Wilson authored
Enabling the interrupt for the signaler takes a finite amount of time (a few microseconds) during which it is possible for the request to complete. Check afterwards and skip adding the request to the signal rbtree if it complete. Signed-off-by:
-
Chris Wilson authored
The important condition that we need to check after enabling the interrupt for signaling is whether the request completed in the process (and so we missed that interrupt). A large cost in enabling the signaling (rather than waiters) is in waking up the auxiliary signaling thread, but we only need to do so to catch that missed interrupt. If we know we didn't miss any interrupts (because we didn't arm the interrupt) then we can skip waking the auxiliary thread. Signed-off-by:
Mika Kuoppala <mika.kuoppala@intel.com> Reviewed-by:
-
Chris Wilson authored
Setting up the irq to signal the request completion takes a finite amount of time, during which it is possible that the request already completed. Check afterwards, just in case, so that we can respond immediately. Signed-off-by:
-
Michel Thierry authored
And prevent calling i915_ggtt_disable_guc twice (the first when GuC init failed, and the second time during driver unload / intel_uc_fini_hw), and hitting the GEM_BUG_ON. v2: Clear enable_guc_loading unconditionally (Michal) Make sure guc_free_load_err_log is still called (Daniele) Don't shoot the messenger (Chris) Fixes: 3950bf3d ("drm/i915/guc: Add onion teardown to the GuC setup") Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by:Michel Thierry <michel.thierry@intel.com> Reviewed-by:
Michal Wajdeczko <michal.wajdeczko@intel.com> Signed-off-by:
-
Chris Wilson authored
Replace the large comment about requiring the powerwell for intel_uncore_arm_unclaimed_mmio_detection() by moving the arming of the mmio error detection into the powerwell held for modesetting. Thereby also accomplishing the goal of only arming the mmio detection after a full modeset. Signed-off-by:
Rodrigo Vivi <rodrigo.vivi@intel.com>
-
- 07 Jun, 2017 13 commits
-
-
Rodrigo Vivi authored
This patch clean up a bit the platform definition block in a way to avoid duplications and to let clear that GT3 for the current platform only have the extra Media engine (BSD2). v2: Kabylake IS_KABYLAKE as Anusha noticed. v3: Avoid EXTRA_ENGINE_MASK and list rings out on GT3 to make it more clear. Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Anusha Srivatsa <anusha.srivatsa@intel.com> Signed-off-by:Anusha Srivatsa <anusha.srivatsa@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1496765166-7068-1-git-send-email-rodrigo.vivi@intel.com
-
Rodrigo Vivi authored
Let's be picky and just use PICK directly. So we can extend this later without creating a new PORT_X por every new number of ports we have to handle. Cc: Manasi Navare <manasi.d.navare@intel.com> Signed-off-by:
Manasi Navare <manasi.d.navare@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1496700722-13755-1-git-send-email-rodrigo.vivi@intel.com
-
Rodrigo Vivi authored
The workaround added in commit c6782b76 ("drm/i915/gen9: Reset secondary power well equests left on by DMC/KVMR") needs to be applied on Cannonlake as well. So let's assume any platform using this power well setup will also need and let's just go ahead and remove if condition. Cc: Imre Deak <imre.deak@intel.com> Signed-off-by:
Imre Deak <imre.deak@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1496781040-20888-11-git-send-email-rodrigo.vivi@intel.com
-
Ville Syrjälä authored
CNL power wells are very similar to SKL, with the exception that the misc IO well has been split into separate AUX IO wells. Not sure if DMC is supposed to manage the AUX wells for us or not. Let's assume so for now. v2: DDI A power well wants DDI A domains, not DDI B domains v3: s/BIT/BIT_ULL and add proper Aux IO domains. (Rodrigo) v4: Remove PW_DDI_E. Not supported on Current CNL SKUs. (Rodrigo). v5: Removed DDI_E_IO_DOMAINS and moved PORT_DDI_E_IO to DDI_A_IO for the same reasons as v4 when we found out that current CNL SKUs don't have the full port E split. Signed-off-by:Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by:
-
Michel Thierry authored
Indirect Context Offset Pointer has changed for Cannonlake. INDIRECT_CTX_OFFSET[15:6] valid value for CNL is 19h per Spec. v2: rebased to intel_lr_indirect_ctx_offset v3: Commit message added per Tvrtko request. Signed-off-by:
-
Rodrigo Vivi authored
All registers and default configuration are the same for Skylake and Cannonlake. v2: Don't apply Wa for platforms without MOCS. (Paulo) v3: Removed WaDisableSkipCaching that Joonas noticed that according to spec it is not applicable to CNL. Cc: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by:
-
Rodrigo Vivi authored
Cannonlake also supports slice power gating on devices with more than one slice as SKL. Let's assume that this is the same for SKL+ and exclude BXT only. v2: Also remove KBL. Signed-off-by:
-
James Irwin authored
Issue: VIZ-4525 Reviewed-by:
Damien Lespiau <damien.lespiau@intel.com> Signed-off-by:
James Irwin <james.irwin@intel.com> Signed-off-by:
Ander Conselvan de Oliveira <conselvan2@gmail.com> Link: http://patchwork.freedesktop.org/patch/msgid/1496781040-20888-6-git-send-email-rodrigo.vivi@intel.com
-
Paulo Zanoni authored
We're going to use it in the next commits. Signed-off-by:
Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by:
Jim Bride <jim.bride@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1496781040-20888-5-git-send-email-rodrigo.vivi@intel.com
-
Rodrigo Vivi authored
By the Spec all CNL Y skus are 2+2, i.e. GT2. v2: Really include the PCI IDs to the picidlist[]; Reviewed-by:
-
Rodrigo Vivi authored
Platform enabling and its power-on are organized in different skus (U x Y x S x H, etc). So instead of organizing it in GT1 x GT2 x GT3 let's also use the platform sku. This is also the new Spec style what makes the review much more easy and straightforward. v2: Really include the PCI IDs to the picidlist[]; v3: Remove PCI IDs not present in spec. v4: Rebase. Signed-off-by:
Clinton Taylor <clinton.a.taylor@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1496781040-20888-3-git-send-email-rodrigo.vivi@intel.com
-
Rodrigo Vivi authored
Avoid warning when CNP is detected with CNL. Also let's force it on the virtual detection. Signed-off-by:
-
Rodrigo Vivi authored
Cannonlake is a Intel
® Processor containing Intel® HD Graphics following Kabylake. It is Gen10. Let's start by adding the platform definition based on previous platforms but yet as alpha_support. On following patches we will start adding PCI IDs and the platform specific changes. CNL has an increased DDB size as Damien had previously noticed and provided a separated patch that got squashed here. v2: Squash DDB size here per Ander request. Credits-to: Damien Lespiau <damien.lespiau@intel.com> Signed-off-by:
-
- 06 Jun, 2017 8 commits
-
-
Ville Syrjälä authored
The clipped src coordinates have already been rotated by 270 degrees for when the plane rotation is 90/270 degrees, hence the FBC code should no longer swap the width and height. Cc: stable@vger.kernel.org Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Paulo Zanoni <paulo.r.zanoni@intel.com> Fixes: b63a16f6 ("drm/i915: Compute display surface offset in the plane check hook for SKL+") Signed-off-by:
Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
-
Ville Syrjälä authored
skl_check_plane_surface() already rotates the clipped plane source coordinates to match the scanout direction because that's the way the GTT mapping is set up. Thus we no longer need to rotate the coordinates in the watermark code. For cursors we use the non-clipped coordinates which are not rotated appropriately, but that doesn't actually matter since cursors don't even support 90/270 degree rotation. v2: Resolve conflicts from SKL+ wm rework Cc: stable@vger.kernel.org Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Fixes: b63a16f6 ("drm/i915: Compute display surface offset in the plane check hook for SKL+") Signed-off-by:
-
Ville Syrjälä authored
Starting from commit b63a16f6 ("drm/i915: Compute display surface offset in the plane check hook for SKL+") we've already rotated the src coordinates by 270 degrees by the time we check if a scaler is needed or not, so we must not account for the rotation a second time. Previously we did these steps in the opposite order and hence the scaler check had to deal with rotation itself. The double rotation handling causes us to enable a scaler pretty much every time 90/270 degree plane rotation is requested, leading to fuzzier fonts and whatnot. v2: s/unsigned/unsigned int/ to appease checkpatch v3: s/DRM_ROTATE_0/DRM_MODE_ROTATE_0/ Cc: stable@vger.kernel.org Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reported-by:
-
Ville Syrjälä authored
Restore the lost has_fbc flag for mobile ILK. Cc: Carlos Santa <carlos.santa@intel.com> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Fixes: a1323380 ("drm/i915: Introduce GEN5_FEATURES for device info") Signed-off-by:
-
Ville Syrjälä authored
We're not that short on characters that we can't spell out "false_color". Saves me from figuring out what "fc" means the next time look at the code. Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by:
-
Ville Syrjälä authored
The number of compressed segments has been available ever since FBC2 was introduced in g4x, it just moved from the STATUS register into STATUS2 on IVB. For FBC1 if we really wanted the number of compressed segments we'd have to trawl through the tags, but in this case since the code just uses the number of compressed segments as an indicator whether compression has occurred we can just check the state of the COMPRESSING and COMPRESSED bits. IIRC the hardware will try to periodically recompress all uncompressed lines even if they haven't changed and the COMPRESSED bit will be cleared while the compressor is running, so just checking the COMPRESSED bit might not give us the right answer. Hence it seems better to check for both COMPRESSED and COMPRESSING as that should tell us that the compressor is at least trying to do something. While at it move the IVB+ register define to the right place, unify the naming convention of the compressed segment count masks, and fix up the mask for g4x. v2: s/ILK_DPFC_STATUS2/IVB_FBC_STATUS2/ (Paulo) Cc: Paulo Zanoni <paulo.r.zanoni@intel.com> Cc: Gabriel Krisman Bertazi <krisman@collabora.co.uk> Signed-off-by:
-
Ville Syrjälä authored
The scanline counter is bonkers on VLV/CHV DSI. The scanline counter increment is not lined up with the start of vblank like it is on every other platform and output type. This causes problems for both the vblank timestamping and atomic update vblank evasion. On my FFRD8 machine at least, the scanline counter increment happens about 1/3 of a scanline ahead of the start of vblank (which is where all register latching happens still). That means we can't trust the scanline counter to tell us whether we're in vblank or not while we're on that particular line. In order to keep vblank timestamping in working condition when called from the vblank irq, we'll leave scanline_offset at one, which means that the entire line containing the start of vblank is considered to be inside the vblank. For the vblank evasion we'll need to consider that entire line to be bad, since we can't tell whether the registers already got latched or not. And we can't actually use the start of vblank interrupt to get us past that line as the interrupt would fire too soon, and then we'd up waiting for the next start of vblank instead. One way around that would using the frame start interrupt instead since that wouldn't fire until the next scanline, but that would require some bigger changes in the interrupt code. So for simplicity we'll just poll until we get past the bad line. v2: Adjust the comments a bit Cc: stable@vger.kernel.org Cc: Jonas Aaberg <cja@gmx.net> Tested-by:
Jonas Aaberg <cja@gmx.net> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99086Signed-off-by:
Mika Kahola <mika.kahola@intel.com> Reviewed-by:
-
Ville Syrjälä authored
Remove the SNB PCH refclock init call from the runtime resume handler. I don't think it was actually needed even when we had SNB runtime PM, and if definitely isn't needed ever since SNB runtime PM was nuked in commit d4c5636e ("drm/i915: Remove runtime PM for SNB"). Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Cc: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by:
-
- 02 Jun, 2017 8 commits
-
-
Rodrigo Vivi authored
Panel Power sequences for CNP is similar to Broxton, but with only one sequencer. Main difference from SPT is that PP_DIVISOR was removed and power cycle delay has been moved to PP_CONTROL. v2: Add missed pp_div write, that is now part of PP_CONTROL[8:4] as on Broxton. (Found by DK) v3: Improve commit message. (By DK) Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com> Cc: Jani Nikula <jani.nikula@intel.com> Signed-off-by:Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1496434004-29812-6-git-send-email-rodrigo.vivi@intel.com
-
Rodrigo Vivi authored
On CNP PCH based platforms the gmbus is on the south display that is on PCH. The existing implementation for previous platforms already covers the need for CNP expect for the pin pair configuration that follows similar definitions that we had on BXT. v2: Don't drop "_BXT" as the indicator of the first platform supporting this pin numbers. Suggested by Daniel. v3: Add missing else and fix register table since CNP GPIO_CTL starts on 0xC5014. v4: Fix pin number and map according to the current available VBT. Re-add pin 4 for port D. Lost during some rebase. v5: Use table as spec. If VBT is wrong it should be ignored. Cc: Daniel Vetter <daniel.vetter@intel.com> Cc: Jani Nikula <jani.nikula@intel.com> Signed-off-by: -
Rodrigo Vivi authored
Split out BXT and CNP's setup_backlight(),enable_backlight(), disable_backlight() and hz_to_pwm() into two separate functions instead of reusing BXT function. Reuse set_backlight() and get_backlight() since they have no reference to the utility pin. v2: Reuse BXT functions with controller 0 instead of redefining it. (Jani). Use dev_priv->rawclk_freq instead of getting the value from SFUSE_STRAP. v3: Avoid setup backligh controller along with hooks and fully reuse hooks setup as suggested by Jani. v4: Clean up commit message. v5: Implement per PCH instead per platform. v6: Introduce a new function for CNP.(Jani and Ville) v7: Squash the all CNP Backlight support patches into a single patch. (Jani) v8: Correct indentation, remove unneeded blank lines and correct mail address (Jani). v9: Remove unused enum pipe. (by CI) v10: Remove comment mentioning SFUSE_STRAP in a part of the code that we don't use it. (Jani) Make controller = 0 since current CNP has only one controller and put a comment mentioning why we reuse the BXT definitions and are keeping the controller = 0. (DK) v11: Remove spurious line. (DK) Reviewed-by:Jani Nikula <jani.nikula@intel.com> Suggested-by:
Ville Syrjala <ville.syrjala@intel.com> Cc: Ville Syrjala <ville.syrjala@linux.intel.com> Cc: Jani Nikula <jani.nikula@intel.com> Signed-off-by:
-
Rodrigo Vivi authored
RAWCLK_FREQ register has changed for platforms with CNP+. [29:26] This field provides the denominator for the fractional part of the microsecond counter divider. The numerator is fixed at 1. Program this field to the denominator of the fractional portion of reference frequency minus one. If the fraction is 0, program to 0. 0100b = Fraction .2 MHz = Fraction 1/5. 0000b = Fraction .0 MHz. [25:16] This field provides the integer part of the microsecond counter divider. Program this field to the integer portion of the reference frequenct minus one. Also this register tells us that proper raw clock should be read from SFUSE_STRAP and programmed to this register. Up to this point on other platforms we are reading instead of programming it so probably relying on whatever BIOS had configured here. Now on let's follow the spec and also program this register fetching the right value from SFUSE_STRAP as Spec tells us to do. v2: Read from SFUSE_STRAP and Program RAWCLK_FREQ instead of reading the value relying someone else will program that for us. v3: Add missing else. (Jani) v4: Addressing all Ville's catches: Use macro for shift bits instead of defining shift. Remove shift from the cleaning bits with mask that already has it. Add missing I915_WRITE to actually write the reg. Stop using useless DIV_ROUND_* on divider that is exact dividion and use DIV_ROUND_CLOSEST for the fraction part. v5: Remove useless Read-Modify-Write on raclk_freq reg. (Ville). v6: Change is per PCH instead of per platform. Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Cc: Jani Nikula <jani.nikula@intel.com> Signed-off-by: -
Dhinakaran Pandiyan authored
The first two bytes of PCI ID for CNP_LP PCH are the same as that of SPT_LP. We should really be looking at the first 9 bits instead of the first 8 to identify platforms, although this seems to have not caused any problems on earlier platforms. Introduce a 9 bit extended mask for SPT and CNP while not touching the code for any of the other platforms. v2: (Rodrigo) Make platform agnostic and fix commit message. Signed-off-by:
-
Rodrigo Vivi authored
Most of south engine display that is in PCH is still the same as SPT and KBP, except for this key differences: - Backlight: Backlight programming changed in CNP PCH. - Panel Power: Sligh programming changed in CNP PCH. - GMBUS and GPIO: The pin mapping has changed in CNP PCH. All of these changes follow more the BXT style. v2: Update definition to use dev_priv isntead of dev (Tvrtko). Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com> Signed-off-by:
-
Chris Wilson authored
In commit 5763ff04 ("drm/i915: Avoid GPU stalls from kswapd") we stopped direct reclaim and kswapd from triggering GPU/client stalls whilst running (by restricting the objects they could reap to be idle). However with abusive GPU usage, it becomes quite easy to starve kswapd of memory and prevent it from making forward progress towards obtaining enough free memory (thus driving the system closer to swap exhaustion). Relax the previous restriction to allow kswapd (but not direct reclaim) to stall the device whilst reaping purgeable pages. v2: Also acquire the rpm wakelock to allow kswapd to unbind buffers. Signed-off-by:
-
Weinan Li authored
I915_GEM_GET_APERTURE ioctl is used to probe aperture size from userspace. In gvt environment, each vm only use the ballooned part of aperture, so we should return the correct available aperture size exclude the reserved part by balloon. v2: add 'reserved' in struct i915_address_space to record the reserved size in ggtt (Chris) v3: remain aper_size as total, adjust aper_available_size exclude reserved and pinned. UMD driver need to adjust the max allocation size according to the available aperture size but not total size. KMD return the correct usable aperture size any time (Chris, Joonas) v4: decrease reserved in deballoon (Joonas) v5: add onion teardown in balloon, add vgt_deballoon_space (Joonas) v6: change title name (Zhenyu) v7: code style refine (Joonas) Suggested-by:
Weinan Li <weinan.z.li@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1496198152-14175-1-git-send-email-weinan.z.li@intel.comReviewed-by:
-
- 01 Jun, 2017 5 commits
-
-
Chris Wilson authored
The assertion that we want to make before disabling the pin of the pages for the unknown swizzling quirk is that the quirk is indeed active, and that the quirk is disabled before we do apply it to the pages. Fixes: 2c3a3f44 ("drm/i915: Fix pages pin counting around swizzle quirk") Fixes: 957870f9 ("drm/i915: Split out i915_gem_object_set_tiling()") Signed-off-by:
-
Chris Wilson authored
As another precaution when testing whether the CS engine is actually idle, also inspect the ring's HEAD/TAIL registers, which should be equal when there are no commands left to execute by the GPU. Signed-off-by:
-
Chris Wilson authored
When we enable the GuC, we enable an alternative mechanism for doing post-GGTT update invalidation. Likewise, when we disable the GuC, we restore the previous method. Assert that we change between known endpoints, so that we can catch if we accidentally clobber some other gen and if we change the invalidate routine without updating guc. Signed-off-by:
-
Jani Nikula authored
Print DID not VID on the DID error path. Looks like a copy-paste error from the VID error path. Clarify and clean up error logging, making them distinguishable from each other, while at it. Reported-by:
Petru Mihancea <petrum@gmail.com> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101243Reviewed-by:
-
Imre Deak authored
Atm disabling either DP or eDP outputs can generate a spurious short pulse interrupt. The reason is that after disabling the port the source will stop sending a valid stream data, while the sink expects either a valid stream or the idle pattern. Since neither of this is sent the sink assumes (after an arbitrary delay) that the link is lost and requests for link retraining with a short pulse. The spurious pulse is a real problem at least for eDP panels with long power-off / power-cycle delays: as part of disabling the output we disable the panel power. The subsequent spurious short pulse handling will have to turn the power back on, which means the driver has to do a redundant wait for the power-off and power-cycle delays. During system suspend this leads to an unnecessary delay up to ~1s on systems with such panels as reported by Rui. To fix this put the sink to DPMS D3 state before turning off the port. According to the DP spec in this state the sink should not request retraining. This is also what we do already on pre-ddi platforms. As an alternative I also tried configuring the port to send idle pattern - which is against BSPec - and leave the port in normal mode before turning off the port. Neither of these resolved the problem. Cc: Zhang Rui <rui.zhang@intel.com> Cc: David Weinehall <david.weinehall@linux.intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Reported-and-tested-by:
Zhang Rui <rui.zhang@intel.com> Signed-off-by:
-
