Quoting Chris Wilson (2020-04-02 12:59:08) > Add a per-engine request mempool so that we should always have a couple > of requests available for powermanagement allocations from tricky > contexts. These reserves are expected to be only used for kernel > contexts when barriers must be emitted [almost] without fail. > > When using the mempool, requests are first allocated from the global > slab cache (utilising all the per-cpu lockless freelists and caches) and > only if that is empty and cannot be filled under the gfp_t do we > fallback to using the per-engine cache of recently freed requests. For > our use cases, this will never be empty for long as there will always be > at least the previous powermanagent request to reuse. > > The downside is that this is quite a bulky addition and abstraction to > use, but it will ensure that we never fail to park the engine due to > oom. Strictly speaking, mempool_alloc() does not have the semantics I ascribe it. Which is annoying. mempool_alloc() will always dip into the reserves if a NORETRY alloc fails, and so we should ourselves reserve dipping into the mempool if we can. -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
