- 27 Mar, 2017 6 commits
-
-
Chris Wilson authored
Not all of our target platforms have clflush. For those without, just assume the status page is sufficiently coherent that we do not need our paranoia. Reported-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Fixes: 14a6bbf9 ("drm/i915: Replace irq_seqno_barrier on hws write with a clflush") Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Cc: Mika Kuoppala <mika.kuoppala@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170324163540.31981-1-chris@chris-wilson.co.ukTested-by:
Mika Kuoppala <mika.kuoppala@intel.com>
-
Chris Wilson authored
Old devices have quite severe restrictions for using fences, and unlike more recent device (anything from Pineview onwards) we need to enforce those restrictions even for unfenced tiled access from the render pipeline. Fixes: 944397f0 ("drm/i915: Store required fence size/alignment for GGTT vma") Reported-by:
Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by:
Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Tested-by:
-
Chris Wilson authored
We have only 8bits of precise timestamps in which to complete our upper/load reads, along with the switch between precision. This is not always enough time to read the upper counter twice within the same time slice, leading to hard lockups. Limit the number of times to prevent an inifite loop (my fault for assuming we would have no trouble doing the write + reads fast enough). Fixes: 47c21d9a ("drm/i915: Extend vlv/chv residency resolution") Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100377Signed-off-by:
-
Chris Wilson authored
We can assume that if the device is asleep then all pending GTT writes will have been posted, and so we can defer the flush from i915_gem_object_flush_gtt_write_domain() [ 1957.462568] WARNING: CPU: 0 PID: 6132 at drivers/gpu/drm/i915/intel_drv.h:1742 fwtable_read32+0x123/0x150 [i915] [ 1957.462582] RPM wakelock ref not held during HW access [ 1957.462583] Modules linked in: i915 intel_gtt drm_kms_helper prime_numbers [ 1957.462607] CPU: 0 PID: 6132 Comm: gem_concurrent_ Tainted: G U 4.11.0-rc1+ #464 [ 1957.462619] Hardware name: / , BIOS PYBSWCEL.86A.0027.2015.0507.1758 05/07/2015 [ 1957.462630] Call Trace: [ 1957.462646] dump_stack+0x4d/0x6f [ 1957.462657] __warn+0xc1/0xe0 [ 1957.462667] warn_slowpath_fmt+0x4a/0x50 [ 1957.462709] fwtable_read32+0x123/0x150 [i915] [ 1957.462750] i915_gem_object_flush_gtt_write_domain+0x43/0x70 [i915] [ 1957.462791] i915_gem_object_set_to_cpu_domain+0x46/0xa0 [i915] [ 1957.462831] i915_gem_set_domain_ioctl+0x15d/0x220 [i915] [ 1957.462843] drm_ioctl+0x1d7/0x440 [ 1957.462885] ? i915_gem_obj_prepare_shmem_write+0x1d0/0x1d0 [i915] [ 1957.462896] ? pick_next_task_fair+0x436/0x440 [ 1957.462906] ? mntput+0x1f/0x30 [ 1957.462915] do_vfs_ioctl+0x8f/0x5c0 [ 1957.462925] ? __schedule+0x16f/0x5f0 [ 1957.462935] ? ____fput+0x9/0x10 [ 1957.462943] SyS_ioctl+0x3c/0x70 [ 1957.462952] entry_SYSCALL_64_fastpath+0x17/0x98 [ 1957.462961] RIP: 0033:0x7fc542179ca7 [ 1957.462968] RSP: 002b:00007ffeef12ff98 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 [ 1957.462982] RAX: ffffffffffffffda RBX: 00007ffeef1301d0 RCX: 00007fc542179ca7 [ 1957.462990] RDX: 00007ffeef12ffd0 RSI: 00000000400c645f RDI: 0000000000000003 [ 1957.462999] RBP: 0000000000000003 R08: 000055f433bc7c40 R09: 000000000000002c [ 1957.463006] R10: 0000000000000073 R11: 0000000000000246 R12: 0000000000000018 [ 1957.463015] R13: 000055f432c89d20 R14: 000055f432c87690 R15: 0000000000000000 Fixes: 3b5724d7 ("drm/i915: Wait for writes through the GTT to land before reading back") Signed-off-by:
-
Chris Wilson authored
I noticed that gcc was spilling the CSB to the stack, so rearrange the code to be more compact. Spilling in this function is slightly more interesting due to the mmio reads acting as memory barriers and so end up flushing the stack spills. Still miniscule to having to do at least the pair of uncached reads :( function old new delta intel_lrc_irq_handler 1039 878 -161 Signed-off-by:
Tvrtko Ursulin <tvrtko.ursulin@intel.com>
-
Michal Wajdeczko authored
There is no need to expose this function as it is called from one function only. Also move it up to avoid forward declaration. v2: drop intel_ prefix (Oscar) and rename to fetch_uc_fw (Michal) Signed-off-by:
Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Arkadiusz Hiler <arkadiusz.hiler@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170327094510.167400-1-michal.wajdeczko@intel.comReviewed-by:
-
- 24 Mar, 2017 4 commits
-
-
Chris Wilson authored
The required number of dwords for semaphore emission on BDW RCS is 8, not 6 - leading to ring buffer corruption and immediate GPU hangs when using ringbuffer submission. Signed-off-by:
Michał Winiarski <michal.winiarski@intel.com>
-
Chris Wilson authored
The current w/a for the gen7 psmi related hangs doesn't apply to bdw, so disable it if using bdw ringbuffer submission. Signed-off-by:
-
Daniele Ceraolo Spurio authored
The forcewake_get call in the guc_send_mmio function was added to avoid getting and releasing forcewake on each register access. While this makes sense, all GuC registers are in the blitter range so no need to wake all the wells. Signed-off-by:
Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1490366919-34715-1-git-send-email-daniele.ceraolospurio@intel.comReviewed-by:
Sagar Arun Kamble <sagar.a.kamble@intel.com> Signed-off-by:
-
Chris Wilson authored
Move the common "client->vaddr + client->proc_desc_offset" to its own function, __get_process_desc() to match the newly established pattern. Signed-off-by:
-
- 23 Mar, 2017 30 commits
-
-
Chris Wilson authored
We only need to care about the ordering of the clearing of the bit with the uncached CSB read in order to correctly detect a new interrupt before the read completes. The uncached read itself acts as a full memory barrier, so we do not need to enforce another in the form of a locked clear_bit. v2: Clarify why the split and unlocked test/clear is harmless. Signed-off-by:
-
Ville Syrjälä authored
Pull the code to calculate the cursor control register value into separate functions. Allows us to pre-compute them in the future. Signed-off-by:
-
Ville Syrjälä authored
Pull the code to calculate the ILK-SNB sprite control register value into a separate function. Allows us to pre-compute it in the future. Signed-off-by:
-
Ville Syrjälä authored
Pull the code to calculate the IVB-BDW sprite control register value into a separate function. Allows us to pre-compute it in the future. Signed-off-by:
-
Ville Syrjälä authored
Pull the code to calculate the VLV/CHV sprite control register value into a separate function. Allows us to pre-compute it in the future. Signed-off-by:
-
Ville Syrjälä authored
On SKL the planes are uniform so the "sprites" can use the primary plane code perfectly fine. The only difference we have is the color key handling, but since we never enable that for the primary plane the same code works just fine. Signed-off-by:
-
Ville Syrjälä authored
Pull the code to calculate the SKL plane control register value into a separate function. Allows us to pre-compute it in the future. Signed-off-by:
-
Oscar Mateo authored
They go better together. Signed-off-by:
Oscar Mateo <oscar.mateo@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by:
-
Oscar Mateo authored
We are going to need it for future platforms. Signed-off-by:
-
Oscar Mateo authored
A GuC context and a HW context are in no way related, so the name "GuC context descriptor" is very unfortunate, because a new reader of the code gets overwhelmed very quickly with a lot of things called "context" that refer to different things. We can improve legibility a lot by simply renaming a few objects in the GuC code. v2: - Rebased - s/ctx_desc_pool/stage_desc_pool - Move some explanations to the definition of the guc_stage_desc struct (Chris) v3: - Calculate gemsize with less intermediate steps (Joonas) - Use BIT() macro (Joonas) Signed-off-by:
-
Oscar Mateo authored
Some recent refactoring patches have left the doorbell creation outside the GuC client allocation, which does not make a lot of sense (a client without a doorbell is something useless). Move it back there, and refactor the init_doorbell_hw consequently. Thanks to this, we can do some other improvements, like hoisting the check for GuC submission enabled out of the enable function. v2: Rebased. Signed-off-by:
-
Oscar Mateo authored
Doorbell release flow requires that we wait for GEN8_DRB_VALID bit to go to zero after updating db_status before we call the GuC to release the doorbell. Kudos to Daniele for finding this out. v2: WARN instead of DRM_ERROR (Joonas) Signed-off-by:
-
Oscar Mateo authored
While at it, fix a typo (s/ring_lcra/ring_lrca) and improve the naming of one firware interface field (s/ring_tail/submit_element_info, since it can contain more than just the ring tail). No change in functionality. v2: - Remove reference to "unique user" of the GuC (Daniele) - Keep mention to renaming from "GuC context" to "client" (Daniele) Signed-off-by:
-
Oscar Mateo authored
Prepare for an alternate GuC communication interface. v2: Make a few functions static and name them correctly while we are at it (Oscar), but leave an intel_guc_send_mmio interface for users that require old-style communication. v3: Send intel_uc_init_early back to the top (Michal). Signed-off-by:
Michel Thierry <michel.thierry@intel.com> Signed-off-by:
-
Oscar Mateo authored
When initializing the GuC log struct, there is an object we need to allocate always, since the GuC needs its address at fw load time. The rest is only needed during runtime, in the sense that we only create if we actually enable GuC logging. Make that distinction explicit by subdividing further the intel_guc_log struct. v2: Call the new struct "runtime", instead of "extras" (Joonas) v3: Check indent (Joonas) Signed-off-by:
-
Oscar Mateo authored
It's mandatory and it gets created if and only if GuC submission is enabled, so that should be the condition for informing the GuC about it. Also s/guc_addon_create/guc_ads_create and s/guc_addon_destroy/guc_ads_destroy and, while at it, add an explanation of what things go inside the ADS object. Signed-off-by:
-
Oscar Mateo authored
Starting with intel_guc_loader, down to intel_guc_submission and finally to intel_guc_log. v2: - Null execbuf client outside guc_client_free (Daniele) - Assert if things try to get allocated twice (Daniele/Joonas) - Null guc->log.buf_addr when destroyed (Daniele) - Newline between returning success and error labels (Joonas) - Remove some unnecessary comments (Joonas) - Keep guc_log_create_extras naming convention (Joonas) - Helper function guc_log_has_extras (Joonas) - No need for separate relay_channel create/destroy. It's just another extra. - No need to nullify guc->log.flush_wq when destroyed (Joonas) - Hoist the check for has_extras out of guc_log_create_extras (Joonas) - Try to do i915_guc_log_register/unregister calls (kind of) symmetric (Daniele) - Make sure initel_guc_fini is not called before init is ever called (Daniele) v3: - Remove unnecessary parenthesis (Joonas) - Check for logs enabled on debugfs registration - Rebase on top of Tvrtko's "Fix request re-submission after reset" v4: - Rebased - Comment around enabling/disabling interrupts inside GuC logging (Joonas) Signed-off-by:
-
Oscar Mateo authored
The GuC descriptor is big in size. If we use a local definition of guc_desc we have a chance to overflow stack, so avoid it. Also, Chris abhors scatterlists :) v2: Rebased, helper function to retrieve the context descriptor, s/ctx_pool_vma/ctx_pool/ v3: Zero out guc_context_desc before initialization v4: Do not do arithmetic on void pointers (Daniele) v5: Nicer than arithmetic on pointers (Chris, Joonas) Signed-off-by:
-
Joonas Lahtinen authored
Started adding proper teardown to guc_client_alloc, ended up removing quite a few dead ends where errors communicating with the GuC were silently ignored. There also seemed to be quite a few erronous teardown actions performed in case of an error (ordering wrong). v2: - Increase function symmetry/proximity (Michal/Daniele) - Fix __reserve_doorbell accounting for high priority (Daniele) - Call __update_doorbell_desc! (Daniele) - Isolate __guc_{,de}allocate_doorbell (Michal/Daniele) v3: - "Select" a cacheline is a more accurate verb than "reserve" (Daniele). - We cannot update & create the doorbell without reserving it first, so move the whole doorbell creation for execbuf_client to the submission enable (Oscar).i - Add a fixme for ignoring possible doorbell destroy errors. v4: - Remove comment about is_high_priority (Daniele) - Debug message typo (Daniele) - Reuse __get_doorbell in more places (Daniele) - Do not do arithmetic on void pointers (Daniele) - Add comment to __reset_doorbell (Daniele) v5: - gccisms like arithmetic on void pointers are not frowned upon (Oscar) Signed-off-by: -
Chris Wilson authored
Ensure that before we overwrite the reservation_object with our exclusive fence for the pending clflush operation, that we do wait upon all the fences in the current reservation_object. Fixes: 57822dc6 ("drm/i915: Perform object clflushing asynchronously") Signed-off-by:
-
Chris Wilson authored
The set of available structs is not protected by the spinlock, and for the single read we can use READ_ONCE instead. Signed-off-by:
-
Chris Wilson authored
Since we reuse the same values for each fw_domain, move them onto uncore. Signed-off-by:
-
Chris Wilson authored
We can relax the requirement upon ourselves that the forcewake is released immediately and just allow it to occur naturally following our mmio request. Signed-off-by:
-
Chris Wilson authored
Use find-first-set bitop to quickly scan through the fw_domains mask and skip iterating over unused domains. v2: Move the WARN into the caller, to prevent compiler warnings in normal builds. Signed-off-by:
-
Chris Wilson authored
In the next patch we will begin to sanity check that we do not attempt to obtain the forcewake on an unsupport domain. However, that is exactly what we do during reset of the fw_domains - rectify it before it explodes. Signed-off-by:
Mika Kuoppala <mika.kuoppala@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170323101944.21627-3-chris@chris-wilson.co.uk
-
Chris Wilson authored
In the next patch we will begin to sanity check that we do not attempt to obtain the forcewake on an unsupport domain. However, that is exactly what we do during our actual initialisation of fw_domains - rectify it before it explodes. Signed-off-by:
-
Chris Wilson authored
Pass along the drm_i915_private pointer from the caller, rather than looking it up from each fw_domain during fw_domains_get/_put. This allows us to then eliminate the backpointer, in exchange for a more complicated unwrapping procedure in the rare intel_uncore_fw_release_timer(). Signed-off-by:
-
Chris Wilson authored
Commit e8a9c58f ("drm/i915: Unify active context tracking between legacy/execlists/guc") converted the legacy intel_ringbuffer submission to the same context pinning mechanism as execlists - that is to pin the context until the subsequent request is retired. Previously it used the vma retirement of the context object to keep itself pinned until the next request (after i915_vma_move_to_active()). In the conversion, I missed that the vma retirement was also responsible for marking the object as dirty. Mark the context object as dirty when pinning (equivalent to execlists) which ensures that if the context is swapped out due to mempressure or suspend/hibernation, when it is loaded back in it does so with the previous state (and not all zero). Fixes: e8a9c58f ("drm/i915: Unify active context tracking between legacy/execlists/guc") Reported-by:
Dennis Gilmore <dennis@ausil.us> Reported-by:
Mathieu Marquer <mathieu.marquer@gmail.com> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99993 Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100181Signed-off-by:
-
Chris Wilson authored
Why have both hw_flags and flags, when just one will do? Signed-off-by:
-
Chris Wilson authored
Words cannot describe the embarrassment at creating a new gfp_t relaim to only prevent the oomkiller but allow direct|kswapd reclaim, and then not use it in the shmem_read_mapping_page_gfp(). Fixes: 24f8e00a ("drm/i915: Prefer to report ENOMEM rather than incur the oom for gfx allocations") Signed-off-by:
-
