drm/i915: Move dev_priv->mm.[un]bound_list to its own lock
authorChris Wilson <chris@chris-wilson.co.uk>
Mon, 16 Oct 2017 11:40:37 +0000 (12:40 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Mon, 16 Oct 2017 19:44:19 +0000 (20:44 +0100)
commitf2123818ffad0332e03c266ca73fe116e8ea5354
tree5af7c3453f8d6049588b2fe5ba556507011fdade
parent3d574a6bbb12a6c3bbeea807a1724b44a0f6ebbb
drm/i915: Move dev_priv->mm.[un]bound_list to its own lock

Remove the struct_mutex requirement around dev_priv->mm.bound_list and
dev_priv->mm.unbound_list by giving it its own spinlock. This reduces
one more requirement for struct_mutex and in the process gives us
slightly more accurate unbound_list tracking, which should improve the
shrinker - but the drawback is that we drop the retirement before
counting so i915_gem_object_is_active() may be stale and lead us to
underestimate the number of objects that may be shrunk (see commit
bed50aea61df ("drm/i915/shrinker: Flush active on objects before
counting")).

v2: Crosslink the spinlock to the lists it protects, and btw this
changes s/obj->global_link/obj->mm.link/
v3: Fix decoupling of old links in i915_gem_object_attach_phys()
v3.1: Fix the fix, only unlink if it was linked
v3.2: Use a local for to_i915(obj->base.dev)->mm.obj_lock

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171016114037.5556-1-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/i915_debugfs.c
drivers/gpu/drm/i915/i915_drv.h
drivers/gpu/drm/i915/i915_gem.c
drivers/gpu/drm/i915/i915_gem_gtt.c
drivers/gpu/drm/i915/i915_gem_object.h
drivers/gpu/drm/i915/i915_gem_shrinker.c
drivers/gpu/drm/i915/i915_gem_stolen.c
drivers/gpu/drm/i915/i915_vma.c
drivers/gpu/drm/i915/selftests/i915_gem_context.c
drivers/gpu/drm/i915/selftests/i915_gem_evict.c