- 14 Aug, 2016 2 commits
-
-
Chris Wilson authored
Closed vma are removed from the obj->vma_list so that they cannot be found by userspace. However, this means that when forcibly unbinding an object, we have to wait upon all rendering to that object first in order for the closed, but active, vma to be reaped and their bindings removed. Reported-by:
Matthew Auld <matthew.auld@intel.com> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97343 Fixes: aa653a68 ("drm/i915: Be more careful when unbinding vma") Fixes: 8a3b3d57 (" drm/i915: Convert non-blocking userptr waits...") Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Cc: Matthew Auld <matthew.auld@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1471196681-30043-2-git-send-email-chris@chris-wilson.co.ukReviewed-by:
-
Chris Wilson authored
If the obj->vma_list is empty, we immediately return ret. However, we are doing so having never set it to any value, it should be zero! Reported-by:
-
- 12 Aug, 2016 2 commits
-
-
Chris Wilson authored
This patch provides the infrastructure for performing a 16-byte aligned read from WC memory using non-temporal instructions introduced with sse4.1. Using movntdqa we can bypass the CPU caches and read directly from memory and ignoring the page attributes set on the CPU PTE i.e. negating the impact of an otherwise UC access. Copying using movntdqa from WC is almost as fast as reading from WB memory, modulo the possibility of both hitting the CPU cache or leaving the data in the CPU cache for the next consumer. (The CPU cache itself my be flushed for the region of the movntdqa and on later access the movntdqa reads from a separate internal buffer for the cacheline.) The write back to the memory is however cached. This will be used in later patches to accelerate accessing WC memory. v2: Report whether the accelerated copy is successful/possible. v3: Function alignment override was only necessary when using the function target("sse4.1") - which is not necessary for emitting movntdqa from __asm__. v4: Improve notes on CPU cache behaviour vs non-temporal stores. v5: Fix byte offsets for unrolled moves. v6: Find all remaining typos of "movntqda", use kernel_fpu_begin. Signed-off-by:Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1471001999-17787-2-git-send-email-chris@chris-wilson.co.uk
-
Chris Wilson authored
vmaps has a provision for controlling the page protection bits, with which we can use to control the mapping type, e.g. WB, WC, UC or even WT. To allow the caller to choose their mapping type, we add a parameter to i915_gem_object_pin_map - but we still only allow one vmap to be cached per object. If the object is currently not pinned, then we recreate the previous vmap with the new access type, but if it was pinned we report an error. This effectively limits the access via i915_gem_object_pin_map to a single mapping type for the lifetime of the object. Not usually a problem, but something to be aware of when setting up the object's vmap. We will want to vary the access type to enable WC mappings of ringbuffer and context objects on !llc platforms, as well as other objects where we need coherent access to the GPU's pages without going through the GTT v2: Remove the redundant braces around pin count check and fix the marker in documentation (Chris) v3: - Add a new enum for the vmalloc mapping type & pass that as an argument to i915_object_pin_map. (Tvrtko) - Use PAGE_MASK to extract or filter the mapping type info and remove a superfluous BUG_ON.(Tvrtko) v4: - Rename the enums and clean up the pin_map function. (Chris) v5: Drop the VM_NO_GUARD, minor cosmetics. Signed-off-by:Akash Goel <akash.goel@intel.com> Reviewed-by:
Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1471001999-17787-1-git-send-email-chris@chris-wilson.co.uk
-
- 11 Aug, 2016 14 commits
-
-
Ville Syrjälä authored
With NV12 we have two color planes to deal with so we must compute the surface and x/y offsets for the second plane as well. What makes this a bit nasty is that the hardware expects the surface offset to be specified as a distance from the main surface offset. What's worse, the distance must be non-negative (no neat wraparound or anything). So we must make sure that the main surface offset is always less or equal to the AUX surface offset. We do that by computing the AUX offset first and the main surface offset second. If the main surface offset ends up being above the AUX offset, we just push it down as far as is required while still maintaining the required alignment etc. Fortunately the AUX offset only reuqires 4K alignment, so we don't need to do any of the backwards searching for an acceptable offset that we must do for the main surface. And X tiled + NV12 isn't a supported combination anyway. Note that this just computes aux surface offsets, we do not yet program them into the actual hardware registers, and hence we can't yet expose NV12. v2: Rebase due to drm_plane_state src/dst rects s/TODO.../something else/ in the commit message/ (Daniel) Signed-off-by:Ville Syrjälä <ville.syrjala@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1470821001-25272-12-git-send-email-ville.syrjala@linux.intel.comReviewed-by:
Daniel Vetter <daniel.vetter@ffwll.ch>
-
Ville Syrjälä authored
SKL has nasty limitations with the display surface offsets: * source x offset + width must be less than the stride for X tiled surfaces or the display engine falls over * the surface offset requires lots of alignment (256K or 1M) These facts mean that we can't just pick any suitably aligned tile boundary as the offset and expect the resulting x offset to be useable. The solution is to start with the closest boundary as before, but then keep searching backwards until we find one that works, or don't. This means we must be prepared to fail, hence the whole surface offset calculation needs to be moved to the .check_plane() hook from the .update_plane() hook. While at it we can check that the source width/height don't exceed maximum plane size limits. We'll store the results of the computation in the plane state to make it easy for the .update_plane() hook to do its thing. v2: Replace for+break loop with while loop Rebase due to drm_plane_state src/dst rects Rebase due to plane_check_state() Signed-off-by:Sivakumar Thulasimani <sivakumar.thulasimani@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1470821001-25272-11-git-send-email-ville.syrjala@linux.intel.com
-
Ville Syrjälä authored
To make life less surprising we can make intel_adjust_tile_offset() deal with linear buffers as well. Currently it doesn't seem like there's a real need for this since only X tiling and NV12 (which would always be tiled currently) should need it. But I've used it for some debug hacks already so seems like a reasonable thing to have. Signed-off-by:
-
Ville Syrjälä authored
Minimize the resulting X coordinate after intel_adjust_tile_offset() is done with it's offset adjustment. This allows calling intel_adjust_tile_offset() multiple times in case we need to adjust the offset several times. Signed-off-by:
-
Ville Syrjälä authored
If there's a fence on the object it will be aligned to the start of the object, and hence CPU rendering to any fb that straddles the fence edge will come out wrong due to lines wrapping at the wrong place. We have no API to manage fences on a sub-object level, so we can't really fix this in any way. Additonally gen2/3 fences are rather coarse grained so adjusting the offset migth not even be possible. Avoid these problems by requiring the fb layout to agree with the fence layout (if present). v2: Rebase due to i915_gem_object_get_tiling() & co. Signed-off-by:
-
Ville Syrjälä authored
Currently we require the object to be X tiled if the fb is X tiled. The argument is supposedly FBC GTT tracking. But actually that no longer holds water since FBC supports Y tiling as well on SKL+. A better rule IMO is to require that if there is a fence, the fb modifier match the object tiling mode. But if the object is linear, we can allow the fb modifier to be anything. The idea being that if the user set the tiling mode on the object, presumably the intention is to actually use the fence for CPU access. But if the tiling mode is not set, the user has no intention of using a fence (and can't actually since we disallow tiling mode changes when there are framebuffers associated with the object). On gen2/3 we must keep to the rule that the object and fb must be either both linear or both X tiled. No mixing allowed since the display engine itself will use the fence if it's present. v2: Fix typos v3: Rebase due to i915_gem_object_get_tiling() & co. Reviewed-by:
-
Ville Syrjälä authored
Soon the fence tiling mode may not always match the fb modifier even for X tiled buffers. So let's use the fb modifier consistently for all display tiling decisions. v2: Rebased due to s/ring/engine/ v3: Rebased due to s/engine/ring/ O_o v4: Rebase due to i915_gem_object_get_tiling() & co. Reviewed-by: Matthew Auld <matthew.auld@intel.com> (v1) Signed-off-by:
-
Ville Syrjälä authored
intel_compute_tile_offset() and intel_add_fb_offsets() get passed the fb and the rotation. As both of those come from the plane state we can just pass that in instead. For extra consitency pass the plane state to intel_fb_xy_to_linear() as well even though it only really needs the fb. Signed-off-by:
-
Ville Syrjälä authored
We repeat the SKL stride register value calculations a several places. Move it into a small helper function. v2: Rebase due to drm_plane_state src/dst rects Signed-off-by:
-
Ville Syrjälä authored
intel_compute_page_offset() can dig up the correct pitch from the fb itself, no need for the caller to pass it in. A bit of extra care is needed for the lower level _intel_compute_page_offset() since that one gets called before the rotated pitch under intel_fb is populated. Note that we don't actually call it with anything but DRM_ROTATE_0 there so we wouldn't actually look up the rotated pitch there, but still, leave the pitch as something the caller has to pass to _intel_compute_page_offset() as an indicator that something is a bit special. This leaves 'stride_div' in the skl plane update hooks as a mostly useless variable so just get rid of it. v2: Add a note why stride_div got nuked v3: Extract intel_fb_pitch() since it can be useful later Signed-off-by:
-
Ville Syrjälä authored
Redo the fb rotation handling in order to: - eliminate the NV12 special casing - handle fb->offsets[] properly - make the rotation handling easier for the plane code To achieve these goals we reduce intel_rotation_info to only contain (for each plane) the rotated view width,height,stride in tile units, and the page offset into the object where the plane starts. Each plane is handled exactly the same way, no special casing for NV12 or other formats. We then store the computed rotation_info under intel_framebuffer so that we don't have to recompute it again. To handle fb->offsets[] we treat them as a linear offsets and convert them to x/y offsets from the start of the relevant GTT mapping (either normal or rotated). We store the x/y offsets under intel_framebuffer, and for some extra convenience we also store the rotated pitch (ie. tile aligned plane height). So for each plane we have the normal x/y offsets, rotated x/y offsets, and the rotated pitch. The normal pitch is available already in fb->pitches[]. While we're gathering up all that extra information, we can also easily compute the storage requirements for the framebuffer, so that we can check that the object is big enough to hold it. When it comes time to deal with the plane source coordinates, we first rotate the clipped src coordinates to match the relevant GTT view orientation, then add to them the fb x/y offsets. Next we compute the aligned surface page offset, and as a result we're left with some residual x/y offsets. Finally, if required by the hardware, we convert the remaining x/y offsets into a linear offset. For gen2/3 we simply skip computing the final page offset, and just convert the src+fb x/y offsets directly into a linear offset since that's what the hardware wants. After this all platforms, incluing SKL+, compute these things in exactly the same way (excluding alignemnt differences). v2: Use BIT(DRM_ROTATE_270) instead of ROTATE_270 when rotating plane src coordinates Drop some spurious changes that got left behind during development v3: Split out more changes to prep patches (Daniel) s/intel_fb->plane[].foo.bar/intel_fb->foo[].bar/ for brevity Rename intel_surf_gtt_offset to intel_fb_gtt_offset Kill the pointless 'plane' parameter from intel_fb_gtt_offset() v4: Fix alignment vs. alignment-1 when calling _intel_compute_tile_offset() from intel_fill_fb_info() Pass the pitch in tiles in stad of pixels to intel_adjust_tile_offset() from intel_fill_fb_info() Pass the full width/height of the rotated area to drm_rect_rotate() for clarity Use u32 for more offsets v5: Preserve the upper_32_bits()/lower_32_bits() handling for the fb ggtt offset (Sivakumar) v6: Rebase due to drm_plane_state src/dst rects Cc: Sivakumar Thulasimani <sivakumar.thulasimani@intel.com> Signed-off-by: -
Ville Syrjälä authored
Looks like the TSEG lives just above TOUD, stolen comes after TSEG. The spec seems somewhat self-contradictory in places, in the ESMRAMC register desctription it says: TSEG Size: 10=(TOUD + 512 KB) to TOUD 11 =(TOUD + 1 MB) to TOUD so that agrees with TSEG being at TOUD. But the example given elsehwere in the spec says: TOUD equals 62.5 MB = 03E7FFFFh TSEG selected as 512 KB in size, Graphics local memory selected as 1 MB in size General System RAM available in system = 62.5 MB General system RAM range00000000h to 03E7FFFFh TSEG address range03F80000h to 03FFFFFFh TSEG pre-allocated from03F80000h to 03FFFFFFh Graphics local memory pre-allocated from03E80000h to 03F7FFFFh so here we have TSEG above stolen. Real world evidence agrees with the TOUD->TSEG->stolen order however, so let's fix up the code to account for the TSEG size. Cc: Taketo Kabe <fdporg@vega.pgw.jp> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: x86@kernel.org Cc: stable@vger.kernel.org Fixes: 0ad98c74 ("drm/i915: Determine the stolen memory base address on gen2") Fixes: a4dff769 ("x86/gpu: Add Intel graphics stolen memory quirk for gen2 platforms") Reported-by:
Taketo Kabe <fdporg@vega.pgw.jp> Tested-by:
-
Tvrtko Ursulin authored
Currently to change the firmware one has to update the exported module firmware string and the major-minor versions used for verification after load. Consolidate that to a single place defining correct major and minor versions per platform. v2: Rebased for KBL. Signed-off-by:
Dave Gordon <david.s.gordon@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1470842206-35685-1-git-send-email-tvrtko.ursulin@linux.intel.com
-
Tvrtko Ursulin authored
Until now code was calling hweight32 to figure out the number from device_info->ring_mask at runtime. Instead we can cache it at engine init time and use directly. Signed-off-by:
-
- 10 Aug, 2016 17 commits
-
-
Chris Wilson authored
request->batch_obj is only set by execbuffer for the convenience of debugging hangs. By moving that operation to the callsite, we can simplify all other callers and future patches. We also move the complications of reference handling of the request->batch_obj next to where the active tracking is set up for the request. Signed-off-by:
Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1470832906-13972-2-git-send-email-chris@chris-wilson.co.uk
-
Chris Wilson authored
We allocate a few objects into the GGTT that we never need to access via the mappable aperture (such as contexts, status pages). We can request that these are bound high in the VM to increase the amount of mappable aperture available. However, anything that may be frequently pinned (such as logical contexts) we want to use the fast search & insert. Signed-off-by:
-
Chris Wilson authored
During intel_gt_powersave_init() we take the RPS mutex to ensure that all locking requirements are met as we talk to the punit, but we also require the struct_mutex for allocating a slice of the global GTT for a power context on Valleyview. struct_mutex must be the outer lock here, as we nest rps.mutex inside later on. Reported-by:
-
Imre Deak authored
In the preceding patches we made sure that: - the LVDS encoder takes care of reiniting both the LVDS register and its PPS - the eDP encoder takes care of reiniting its PPS - the PPS register unlocking workaround is applied explicitly whenever the PPS context is lost Based on the above we can safely remove the opaque LVDS and PPS save / restore from generic code. Signed-off-by:
Imre Deak <imre.deak@intel.com> Reviewed-by:
-
Imre Deak authored
Atm, we apply this workaround somewhat inconsistently at the following points: driver loading, LVDS init, eDP PPS init, system resume. As this workaround also affects registers other than PPS (timing, PLL) a more consistent way is to apply it early after the PPS HW context is known to be lost: driver loading, system resume and on VLV/CHV/BXT when turning on power domains. This is needed by the next patch that removes saving/restoring of the PP_CONTROL register. This also removes the incorrect programming of the workaround on HSW+ PCH platforms which don't have the register locking mechanism. v2: (Ville) - Don't apply the workaround on BXT. - Simplify platform checks using HAS_DDI(). v3: - Move the call of intel_pps_unlock_regs_wa() to the more logical vlv_display_power_well_init() (also fixing CHV) (Ville). Signed-off-by:
-
Imre Deak authored
Similarly to the previous patch, initialize the PPS from the DP encoder's resume hook. Note that as opposed to LVDS we can't do this during encoder enabling, since we need the PPS for DP detection as well. The PPS init code is now the same for init and resume, so factor out a new intel_dp_pps_init() helper for this. v2: - Factor out intel_dp_pps_init() (Ville). Signed-off-by:
-
Imre Deak authored
Atm the LVDS encoder depends on the PPS HW context being saved/restored from generic suspend/resume code. Since the PPS is specific to the LVDS and eDP encoders a cleaner way is to reinitialize it during encoder enabling, so do this here for LVDS. Follow-up patches will init the PPS for the eDP encoder similarly and remove the suspend/resume time save / restore. v2: - Apply BSpec +1 offset and use DIV_ROUND_UP() when programming the power cycle delay. (Ville) v3: (Ville) - Fix +1 vs. round-up order. - s/reset_on_powerdown/powerdown_on_reset/ Signed-off-by:
-
Imre Deak authored
These two flags mean the same thing, so remove the duplication. Signed-off-by:
-
Imre Deak authored
The PPS registers are pretty much the same everywhere, the differences being: - Register fields appearing, disappearing from one platform to the next: panel-reset-on-powerdown, backlight-on, panel-port, register-unlock - Different register base addresses - Different number of PPS instances: 2 on VLV/CHV/BXT, 1 everywhere else. We can merge the separate set of PPS definitions by extending the PPS instance argument to all platforms and using instance 0 on platforms with a single instance. This means we'll need to calculate the register addresses dynamically based on the given platform and PPS instance. v2: - Simplify if ladder in intel_pps_get_registers(). (Ville) Signed-off-by:
-
Dave Gordon authored
As we're tweaking the GuC-related code in debugfs, we can drop the no-longer-used 'q_fail' and repack the structure. Signed-off-by:
-
Dave Gordon authored
Now that host structures are indexed by host engine-id rather than guc_id, we can usefully convert some for_each_engine() loops to use for_each_engine_id() and avoid multiple dereferences of engine->id. Also a few related tweaks to cache structure members locally wherever they're used more than once or twice, hopefully eliminating memory references. Signed-off-by:
-
Dave Gordon authored
The Context Descriptor passed by the kernel to the GuC contains a field specifying which engine(s) the context will use. Historically, this was always set to "all of them", but if we had a separate client for each engine, we could be more precise, and set only the bit for the engine that the client was associated with. So this patch enables this usage, in preparation for having multiple clients, though at this point there is still only a single client used for all supported engines. Signed-off-by:
-
Dave Gordon authored
We have essentially the same code in each of two different loops, so we can refactor it into a little helper function. This also reduces the amount of work done during startup, as we now only reprogram h/w found to be in a state other than that expected, and so avoid the overhead of setting doorbell registers to the state they're already in. Signed-off-by:
-
Dave Gordon authored
guc_init_doorbell_hw() borrows the (currently single) GuC client to use in reinitialising ALL the doorbell registers (as the hardware doesn't reset them when the GuC is reset). As a prerequisite for accommodating multiple clients, it should only reset doorbells that are supposed to be disabled, avoiding those that are marked as in use by any client. Signed-off-by:
-
Chris Wilson authored
The bottom-half we use for processing the breadcrumb interrupt is a task, which is an RCU protected struct. When accessing this struct, we need to be holding the RCU read lock to prevent it disappearing beneath us. We can use the RCU annotation to mark our irq_seqno_bh pointer as being under RCU guard and then use the RCU accessors to both provide correct ordering of access through the pointer. Most notably, this fixes the access from hard irq context to use the RCU read lock, which both Daniel and Tvrtko complained about. Signed-off-by:
-
Chris Wilson authored
In commit 2529d570 ("drm/i915: Drop racy markup of missed-irqs from idle-worker") the racy detection of missed interrupts was removed when we went idle. This however opened up the issue that the stuck waiters were not being reported, causing a test case failure. If we move the stuck waiter detection out of hangcheck and into the breadcrumb mechanims (i.e. the waiter) itself, we can avoid this issue entirely. This leaves hangcheck looking for a stuck GPU (inspecting for request advancement and HEAD motion), and breadcrumbs looking for a stuck waiter - hopefully make both easier to understand by their segregation. v2: Reduce the error message as we now run independently of hangcheck, and the hanging batch used by igt also counts as a stuck waiter causing extra warnings in dmesg. v3: Move the breadcrumb's hangcheck kickstart to the first missed wait. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97104 Fixes: 2529d570 (waiter"drm/i915: Drop racy markup of missed-irqs...") Testcase: igt/drv_missed_irq Signed-off-by:
Mika Kuoppala <mika.kuoppala@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1470761272-1245-2-git-send-email-chris@chris-wilson.co.uk
-
Chris Wilson authored
One of the few guarantees we want the busy ioctl to provide is that the reported busy writer is included in the set of busy read engines. This should be provided by the ordering of setting and retiring the active trackers, but we can do better by explicitly setting the busy read engine flag for the last writer. v2: More comments inside __busy_write_id() to explain why both fields are set. Fixes: 3fdc13c7 ("drm/i915: Remove (struct_mutex) locking for busy-ioctl") Signed-off-by:
-
- 09 Aug, 2016 2 commits
-
-
Dave Gordon authored
Recent versions of gcc say this: include/drm/i915_drm.h:96:34: warning: result of ‘65535 << 20’ requires 37 bits to represent, but ‘int’ only has 32 bits [-Wshift-overflow=] Reported-by:
David Binderman <linuxdev.baldrick@gmail.com> Signed-off-by:
-
Chris Wilson authored
I mistyped and added an extra _request_ to __i915_gem_active_get_rcu() Also, the same happened to another comment for i915_gem_active_get_rcu() Reported-by:
Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by:
-
- 05 Aug, 2016 3 commits
-
-
Ville Syrjälä authored
We don't have GPU reset support for gen2, which means the display hardware is unaffected when a GPU hang is handled. However as the ring has in fact stopped, any flips still in the ring will never complete, and thus the display base address updates will never happen. So we really need to fix that up manually just like we do on g4x+. In fact, let's just use intel_has_gpu_reset() instead of IS_GEN2() since that'll also handle cases where someone would disable the GPU reset support on gen3/4 for whatever reason. Signed-off-by:
-
Ville Syrjälä authored
Factor out the "does the GPU reset clobber the display?" check into a small helper. Signed-off-by:
-
Maarten Lankhorst authored
Add force_reset_modeset_test as a parameter to force the modeset path during gpu reset. This allows a IGT test to set the knob and trigger a hang to force the gpu reset, even on platforms that wouldn't otherwise require it. Changes since v1: - Split out fix to separate commit. Changes since v2: - This commit is purely about force_reset_modeset_test now. Signed-off-by:
Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Testcase: drv_hangman.reset-with-forced-modeset Tested-by:
-
