Hi Janusz, On Wed, Nov 08, 2023 at 05:29:06PM +0100, Janusz Krzysztofik wrote: > Object debugging tools were sporadically reporting illegal attempts to > free a still active i915 VMA object from when parking a GPU tile believed > to be idle. > > [161.359441] ODEBUG: free active (active state 0) object: ffff88811643b958 object type: i915_active hint: __i915_vma_active+0x0/0x50 [i915] > [161.360082] WARNING: CPU: 5 PID: 276 at lib/debugobjects.c:514 debug_print_object+0x80/0xb0 > ... > [161.360304] CPU: 5 PID: 276 Comm: kworker/5:2 Not tainted 6.5.0-rc1-CI_DRM_13375-g003f860e5577+ #1 > [161.360314] Hardware name: Intel Corporation Rocket Lake Client Platform/RocketLake S UDIMM 6L RVP, BIOS RKLSFWI1.R00.3173.A03.2204210138 04/21/2022 > [161.360322] Workqueue: i915-unordered __intel_wakeref_put_work [i915] > [161.360592] RIP: 0010:debug_print_object+0x80/0xb0 > ... > [161.361347] debug_object_free+0xeb/0x110 > [161.361362] i915_active_fini+0x14/0x130 [i915] > [161.361866] release_references+0xfe/0x1f0 [i915] > [161.362543] i915_vma_parked+0x1db/0x380 [i915] > [161.363129] __gt_park+0x121/0x230 [i915] > [161.363515] ____intel_wakeref_put_last+0x1f/0x70 [i915] > > That has been tracked down to be happening when another thread was > deactivating the VMA inside __active_retire() helper, after the VMA's > active counter was already decremented to 0, but before deactivation of > the VMA's object was reported to the object debugging tools. Root cause > has been identified as premature release of last wakeref for the GPU tile > to which the active VMA belonged. > > In case of single-tile platforms, an engine associated with a request that > uses the VMA is usually keeping the tile's wakeref long enough for that > VMA to be deactivated on time, before it is going to be freed on last put > of that wakeref. However, on multi-tile platforms, a request may use a > VMA from a tile other than the one that hosts the request's engine, then, > not protected with the engine's wakeref. > > Get an extra wakeref for the VMA's tile when activating it, and put that > wakeref only after the VMA is deactivated. However, exclude GGTT from > that processing path, otherwise the GPU never goes idle. Since > __i915_vma_retire() may be called from atomic contexts, use async variant > of wakeref put. > > CI reports indicate that single-tile platforms also suffer sporadically > from the same race, however, unlike in case of multi-tile, exact scenario > when that happens hasn't been discovered yet. Then, while I submit this > patch as fix for multi-tile cases, and in hope it also addresses single- > tile, I'm not able to blame any particular commit for that issue. > However, I'm going to ask i915 maintainers to include this fix, if > accepted, in the current rc cycle (6.7-rc) as important for the first > supported multi-tile platform -- Meteor Lake. > > v2: Get the wakeref before vm mutex to avoid circular locking dependency, > - drop questionable Fixes: tag. > > Closes: https://gitlab.freedesktop.org/drm/intel/issues/8875 > Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@xxxxxxxxxxxxxxx> Can it be: Fixes: 12c255b5dad1 ("drm/i915: Provide an i915_active.acquire callback") Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> Cc: <stable@xxxxxxxxxxxxxxx> # v5.4+ It's not exactly the one fix we need for as this patch is fixing something on its own but not explicitely stated (maybe it was a precautionary measure). But if the fix has to be applied, it has to date back to that period, I guess. Reviewed-by: Andi Shyti <andi.shyti@xxxxxxxxxxxxxxx> Andi
