- 11 Sep, 2018 2 commits
-
-
Chris Wilson authored
Given that we are now reasonably confident in our ability to detect and reserve the stolen memory (physical memory reserved for graphics by the BIOS) for ourselves on most machines, we can put it to use. In this case, we need a page to hold the overlay registers. On an i915g running MythTv, H Buus noticed that commit 6a2c4232 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Nov 4 04:51:40 2014 -0800 drm/i915: Make the physical object coherent with GTT introduced stuttering into his video playback. After discarding the likely suspect of it being the physical cursor updates, we were left with the use of the phys object for the overlay. And lo, if we completely avoid using the phys object (allocated just once on module load!) by switching to stolen memory, the stuttering goes away. For lack of a better explanation, claim victory and kill two birds with one stone. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107600 Fixes: 6a2c4232 ("drm/i915: Make the physical object coherent with GTT") Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20180906190144.1272-1-chris@chris-wilson.co.uk
-
P Raviraj Sitaram authored
During modeset, previously configured csc coefficient matrix,if any, will not persist. This can result in blank screen as csc mode will be programmed while loading LUT but csc coefficient matrix remains unprogrammed. Changes since V1: - Removed platform check Signed-off-by:
P Raviraj Sitaram <raviraj.p.sitaram@intel.com> Signed-off-by:
-
- 10 Sep, 2018 1 commit
-
-
Michal Wajdeczko authored
We should update GuC power domain states also when GuC submission is disabled, otherwise GuC might complain or ignore our requests. This seems to be required for all currently released GuC firmwares. v2: it is only needed by pre-Gen11 firmwares Signed-off-by:
Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: John Spotswood <john.a.spotswood@intel.com> Cc: Anusha Srivatsa <anusha.srivatsa@intel.com> Cc: Tomasz Lis <tomasz.lis@intel.com> Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Reviewed-by:
Anusha Srivatsa <anusha.srivatsa@intel.com> Signed-off-by:
-
- 07 Sep, 2018 1 commit
-
-
Chris Wilson authored
Using the guc, we cannot disable the user interrupt generation as we use it for driving submission. And from Icelake, we no longer have the ability to individually mask interrupt generation from each engine, disabling our ability to fake missed interrupts. In both cases, report back to userspace that the missed interrupt generator is no longer available. Signed-off-by:
Mika Kuoppala <mika.kuoppala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20180907112856.28242-1-chris@chris-wilson.co.uk
-
- 06 Sep, 2018 4 commits
-
-
Joonas Lahtinen authored
Signed-off-by:Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
-
https://github.com/intel/gvt-linuxJoonas Lahtinen authored
gvt-next-2018-09-04 - guest context shadow optimization for restore inhibit one (Yan) - cmd parser optimization (Yan) - W=1 warning fixes (Zhenyu) Signed-off-by:
-
Joonas Lahtinen authored
Signed-off-by: -
Joonas Lahtinen authored
Signed-off-by:
-
- 05 Sep, 2018 4 commits
-
-
Chris Wilson authored
Introduce a complementary function to i915_driver_create() to undo all that is created. Suggested-by:
-
Chris Wilson authored
Attach our device_info to the our i915 private on creation so that it is always available for inspection. Signed-off-by:
-
Imre Deak authored
During IPS disabling the current 42ms timeout value leads to occasional timeouts, increase it to 100ms which seems to get rid of the problem. References: https://bugs.freedesktop.org/show_bug.cgi?id=107494 Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107562Reported-by:
Diego Viola <diego.viola@gmail.com> Tested-by:
Imre Deak <imre.deak@intel.com> Reviewed-by:
-
Chris Wilson authored
Future gen reduce the number of bits we will have available to differentiate between contexts, so reduce the lifetime of the ID assignment from that of the context to its current active cycle (i.e. only while it is pinned for use by the HW, will it have a constant ID). This means that instead of a max of 2k allocated contexts (worst case before fun with bit twiddling), we instead have a limit of 2k in flight contexts (minus a few that have been pinned by the kernel or by perf). To reduce the number of contexts id we require, we allocate a context id on first and mark it as pinned for as long as the GEM context itself is, that is we keep it pinned it while active on each engine. If we exhaust our context id space, then we try to reclaim an id from an idle context. In the extreme case where all context ids are pinned by active contexts, we force the system to idle in order to recover ids. We cannot reduce the scope of an HW-ID to an engine (allowing the same gem_context to have different ids on each engine) as in the future we will need to preassign an id before we know which engine the context is being executed on. v2: Improved commentary (Tvrtko) [I tried at least] References: https://bugs.freedesktop.org/show_bug.cgi?id=107788Signed-off-by:
Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20180904153117.3907-1-chris@chris-wilson.co.uk
-
- 04 Sep, 2018 6 commits
-
-
Chris Wilson authored
If the previous modeset commit has completed and is no longer part of the crtc state, skip waiting for it. Ville pointed out that, in fact, the commit is never removed after a modeset so the only way we could see a NULL here should be if there was never a commit attached. Nevertheless, we have the evidence it can be NULL and it has been defended against elsewhere, for example commit 93313538 ("drm/i915: Pass idle crtc_state to intel_dp_sink_crc"). Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107792 Fixes: c44301fc ("drm/i915: Allow control of PSR at runtime through debugfs, v6") Signed-off-by:
-
Chris Wilson authored
Elsewhere we manipulate uncore.unclaimed_mmio_check and i915_param.mmio_debug under the irq lock (e.g. preserving the current value across a user forcewake grab), but do not protect the manipulation inside intel_uncore_arm_unclaimed_mmio_detection() from concurrent access, even from itself. This is an issue as we do call arm_unclaimed_mmio_detection from multiple threads without coordination. Suggested-by:
Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by:
Mika Kuoppala <mika.kuoppala@linux.intelcom> Link: https://patchwork.freedesktop.org/patch/msgid/20180904131207.17563-1-chris@chris-wilson.co.uk
-
Tvrtko Ursulin authored
There are two issues with the current RPCS programming for Icelake: Expansion of the slice count bitfield has been missed, as well as the required programming workaround for the subslice count bitfield size limitation. 1) Bitfield width for configuring the active slice count has grown so we need to program the GEN8_R_PWR_CLK_STATE accordingly. Current code was always requesting eight times the number of slices (due writing to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations. Due both of the above there was no impact from this incorrect programming but we should still fix it. 2) Due subslice count bitfield being only three bits wide and furthermore capped to a maximum documented value of four, special programming workaround is needed to enable more than four subslices. With this programming driver has to consider the GT configuration as 2x4x8, while the hardware internally translates this to 1x8x8. A limitation stemming from this is that either a subslice count between one and four can be selected, or a subslice count equaling the total number of subslices in all selected slices. In other words, odd subslice counts greater than four are impossible, as are odd subslice counts greater than a single slice subslice count. This also had no impact in the current code base due breakage from 1) always reqesting more than one slice. While fixing this we also add some asserts to flag up any future bitfield overflows. v2: * Use a local in all branches for clarity. (Lionel) Signed-off-by:
Lionel Landwerlin <lionel.g.landwerlin@intel.com> Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com> Cc: tony.ye@intel.com Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Reviewed-by:
-
Chris Wilson authored
Continuing the fun of trying to find exactly the delay that is sufficient to ensure that the page directory is fully loaded between context switches, move the extra flush added in commit 70b73f9a ("drm/i915/ringbuffer: Delay after invalidating gen6+ xcs") to just after we flush the pd. Entirely based on the empirical data of running failing tests in a loop until we survive a day (before the mtbf is 10-30 minutes). Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107769 References: 70b73f9a ("drm/i915/ringbuffer: Delay after invalidating gen6+ xcs") Signed-off-by:
-
Chris Wilson authored
Currently, if the user has enabled mmio-debug around each register access, we presume that we have then checked them all. However, it is still possible through omission (raw register access) or external interaction that the unclaimed access was not highlighted. Signed-off-by:
-
Ville Syrjälä authored
Copy the 38.4 vs. 19.2 MHz ref clock exception from the dpll mgr into the clock readout function as well. v2: Refactor the code into a common function s/is_icl/gen11+/ (Rodrigo) Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107722Signed-off-by:Rodrigo Vivi <rodrigo.vivi@intel.com>
-
- 03 Sep, 2018 10 commits
-
-
Chris Wilson authored
Older gen use a physical address for the hardware status page, for which we use cache-coherent writes. As the writes are into the cpu cache, we use a normal WB mapped page to read the HWS, used for our seqno tracking. Anecdotally, I observed lost breadcrumbs writes into the HWS on i965gm, which so far have not reoccurred with this patch. How reliable that evidence is remains to be seen. v2: Explicitly pass the expected physical address to the hw v3: Also remember the wild writes we once had for HWS above 4G. Signed-off-by:
Daniel Vetter <daniel@ffwll.ch> Link: https://patchwork.freedesktop.org/patch/msgid/20180903152304.31589-2-chris@chris-wilson.co.uk
-
Chris Wilson authored
Pull the physical status page cleanup into a common cleanup_status_page() for caller simplicity. Signed-off-by:
Matthew Auld <matthew.auld@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20180903152304.31589-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
We currently assert that if the target is in a CPU write domain, we use a CPU reloc path rather than the GPU reloc path. However, we have a debug override to force the GPU path and that unfortunately hits the assert. Include the async clflush under the debug option to ensure correct behaviour even when debugging, and strict when not. Signed-off-by:
-
Joonas Lahtinen authored
Signed-off-by: -
Chris Wilson authored
Add a mode to debugfs/drop-caches to flush unwanted requests off the GPU (by wedging the device and resetting). This is very useful if a test terminated leaving a long queue of hanging batches that would ordinarily require a round trip through hangcheck for each. It reduces the inter-test operation to just a write into drop-caches to reset driver/GPU state between tests. Signed-off-by:
-
Chris Wilson authored
We currently try to pin and allocate the whole buffer at a time. If that object is larger than RAM, we will try to pin the whole of physical memory, force the machine into oom, and then still fail the allocation. If the request is obviously too large, error out early. We opt to do this in the backend to make it easy to use alternate paths that do not require the entire object pinned, or may easily handle proxy objects that are larger than physical memory. Signed-off-by:
-
Chris Wilson authored
If we fail to write the user relocation back when it is changed, force ourselves to take the slow relocation path where we can handle faults in the write path. There is still an element of dubiousness as having patched up the batch to use the correct offset, it no longer matches the presumed_offset in the relocation, so a second pass may miss any changes in layout. Signed-off-by:
-
Chris Wilson authored
We do not explicitly mark the PTE for the user's GTT mmap as being wrprotect, so we don't get a refault when we would need to change a read-only mmapping into read-write. As such, we must presume that if the vma has PROT_WRITE it may be written to, although this is supposed to be indicated by set-domain there are cases (e.g. after swap) where userspace may not be aware of the implicit domain change. Signed-off-by:
-
Chris Wilson authored
We only call unset_wedged on the global reset path (since it's a global operation), so if we are terminally wedged and wish to reset, take the full device reset path rather than the quicker individual engine resets. Signed-off-by:
-
Chris Wilson authored
Rather than inspect the global module parameter for whether full-ppgtt maybe enabled, we can inspect the context directly as to whether it has its own vm. Signed-off-by:
-
- 01 Sep, 2018 2 commits
-
-
Tvrtko Ursulin authored
So far we have been relying on vm->file pointer being NULL to declare something GGTT. This has the unfortunate consequence that the default kernel context is also declared GGTT and interferes with the following patch which wants to instantiate VMA's and execute requests against the kernel context. Change the is_ggtt test to use an explicit flag in struct address_space to solve this issue. Note that the bit used is free since there is an alignment hole in the struct. v2: * Mark mock ggtt. Signed-off-by:
-
Imre Deak authored
commit afb2c443 ("drm/i915/ddi: Push pipe clock enabling to encoders") inadvertently stopped enabling the pipe clock for any DP-MST stream after the first one. It also rearranged the pipe clock enabling wrt. initial MST payload allocation step (which may or may not be a problem, but it's contrary to the spec.). Fix things by making the above commit truly a non-functional change. Fixes: afb2c443 ("drm/i915/ddi: Push pipe clock enabling to encoders") Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107365Reported-by:
Lyude Paul <lyude@redhat.com> Reported-by: dmummenschanz@web.de Tested-by: dmummenschanz@web.de Tested-by:
-
- 31 Aug, 2018 5 commits
-
-
Lionel Landwerlin authored
Abstract the context image access a bit. Signed-off-by:
-
Lionel Landwerlin authored
We don't need any special treatment on error so just return as soon as possible. Signed-off-by:
-
Jyoti Yadav authored
This patch resolves the DMC FW loading issue. Earlier DMC FW package have only one DMC FW for one stepping. But as such there is no such restriction from Package side. For ICL icl_dmc_ver1_07.bin binary package has DMC FW for 2 steppings. So while reading the dmc_offset from package header, for 1st stepping offset used to come 0x0 and was working fine till now. But for second stepping and other steppings, offset is non zero number and is in dwords. So we need to convert into bytes to fetch correct DMC FW from correct place. v2 : Added check for DMC FW max size for various gen. (Imre Deak) v3 : Corrected naming convention for various gen. (Imre Deak) v4 : Initialized max_fw_size to 0 v5 : Corrected DMC FW MAX_SIZE for various gen. (Imre Deak) v6 : Fixed the typo issues. Reviewed-by:
Jyoti Yadav <jyoti.r.yadav@intel.com> Signed-off-by:
-
Chris Wilson authored
Although we cannot do a full system-level test of suspend/hibernate from deep with the kernel selftests, we can exercise the GEM subsystem in isolation and simulate the external effects (such as losing stolen contents and trashing the register state). v2: Don't forget to hold rpm v3: Suspend the GTT mappings, and more rpm! References: https://bugs.freedesktop.org/show_bug.cgi?id=96526 References: 5ab57c70 ("drm/i915: Flush logical context image out to memory upon suspend") Signed-off-by:
Jakub Bartmiński <jakub.bartminski@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20180830134806.21939-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
The optimisation inherent in commit 6a2c4232 ("drm/i915: Make the physical object coherent with GTT") relies on that once we allocated a cursor we would have coherent, zero overhead access to the scanout plane holding the cursor. That is we could then do the very frequent cursor updates X enjoys with no indirection or kernel involvement. However, that all hinges on the GGTT mmap of the cursor being pinned and not require refaulting on each access -- handling such a page fault likely requires the busy GGTT to be rearranged causing a stall. A very simple fix is then to handle the physical cursor exactly like other cursors and keep its vma pinned while active. References: https://bugs.freedesktop.org/show_bug.cgi?id=107600 References: 6a2c4232 ("drm/i915: Make the physical object coherent with GTT") Signed-off-by:
-
- 30 Aug, 2018 4 commits
-
-
Chris Wilson authored
During stress testing of full-ppgtt (on Baytrail at least), we found that the invalidation around a context/mm switch was insufficient (writes would go astray). Adding a second MI_FLUSH_DW barrier prevents this, but it is unclear as to whether this is merely a delaying tactic or if it is truly serialising with the TLB invalidation. Either way, it is empirically required. v2: Avoid the loop for readability; Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107715 References: https://bugs.freedesktop.org/show_bug.cgi?id=107759Signed-off-by:
-
Lionel Landwerlin authored
We need to clear the register in order to get correct value after the next potential hang. v2: Centralize error register clearing in i915_irq.c (Chris) v3: Don't read gen8 register on < gen6 (Chris) v4: Don't swap gen8+ & gen6+ code... (Chris) Signed-off-by:
-
Joonas Lahtinen authored
Signed-off-by: -
Joonas Lahtinen authored
Signed-off-by:
-
- 29 Aug, 2018 1 commit
-
-
Chris Wilson authored
On finishing the reset, the intention is to restart the GPU before we relinquish the forcewake taken to handle the reset - the goal being the GPU reloads a context before it is allowed to sleep. For this purpose, we used tasklet_flush() which although it accomplished the goal of restarting the GPU, carried with it a sting in its tail: it cleared the TASKLET_STATE_SCHED bit. This meant that if another CPU queued a new request to this engine, we would clear the flag and later attempt to requeue the tasklet on the local CPU, breaking the per-cpu softirq lists. Remove the dangerous tasklet_kill() and just run the tasklet func directly as we know it is safe to do so (the tasklets are internally locked to allow mixed usage from direct submission). Signed-off-by:
-
