Chris Wilson <[email protected]> writes: > If userspace is asynchronously streaming into the batch or other > execobjects, we may not flush those writes along with a change in cache > domain (as there is no change). Therefore those writes may end up in > internal chipset buffers and not visible to the GPU upon execution. We > must issue a flush command or otherwise we encounter incoherency in the > batchbuffers and the GPU executing invalid commands (i.e. hanging) quite > regularly. > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90841 > Fixes: 1816f9236303 ("drm/i915: Support creation of unbound wc user...") > Signed-off-by: Chris Wilson <[email protected]> > Cc: Akash Goel <[email protected]> > Cc: Daniel Vetter <[email protected]> > Cc: Tvrtko Ursulin <[email protected]> > Tested-by: Matti Hämäläinen <[email protected]> > Cc: [email protected] > --- > drivers/gpu/drm/i915/i915_gem_execbuffer.c | 13 +++---------- > 1 file changed, 3 insertions(+), 10 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c > b/drivers/gpu/drm/i915/i915_gem_execbuffer.c > index 699315304748..bce587abc601 100644 > --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c > +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c > @@ -1015,8 +1015,6 @@ i915_gem_execbuffer_move_to_gpu(struct > drm_i915_gem_request *req, > { > const unsigned int other_rings = eb_other_engines(req); > struct i915_vma *vma; > - uint32_t flush_domains = 0; > - bool flush_chipset = false; > int ret; > > list_for_each_entry(vma, vmas, exec_list) { > @@ -1029,16 +1027,11 @@ i915_gem_execbuffer_move_to_gpu(struct > drm_i915_gem_request *req, > } > > if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) > - flush_chipset |= i915_gem_clflush_object(obj, false); > - > - flush_domains |= obj->base.write_domain; > + i915_gem_clflush_object(obj, false); > } > > - if (flush_chipset) > - i915_gem_chipset_flush(req->engine->i915); > - > - if (flush_domains & I915_GEM_DOMAIN_GTT) > - wmb();
Was a bit worried about this vanishing. But after chatting with Chris and also founding this: https://lwn.net/Articles/174655/ [in I386 AND X86_64 SPECIFIC NOTES] I am convinced that the uncached write to force the chipset_flush will be strong barrier enough. Reviewed-by: Mika Kuoppala <[email protected]> > + /* Unconditionally flush any chipset caches (for streaming writes). */ > + i915_gem_chipset_flush(req->engine->i915); > > /* Unconditionally invalidate GPU caches and TLBs. */ > return req->engine->emit_flush(req, EMIT_INVALIDATE); > -- > 2.9.3 > > _______________________________________________ > Intel-gfx mailing list > [email protected] > https://lists.freedesktop.org/mailman/listinfo/intel-gfx _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
