For the storage track, I would like to propose a topic for differentiated blk-mq hardware contexts. Today, blk-mq considers all hardware contexts equal, and are selected based on the software's CPU context. There are use cases that benefit from having hardware context selection criteria beyond which CPU a process happens to be running on.
One example is exlusive polling for the latency sensitive use cases. Mixing polled and non-polled requests into the same context loses part of the benefit when interrupts unnecessarilly occur, and coalescing tricks to mitigate this have undesirable side effects during times when HIPRI commands are not issued. Another example is for hardware priority queues, where not all command queues the hardware provides may be equal to another. Many newer storage controllers provide such queues with different QoS guarantees, and Linux currently does not make use of this feature. The talk would like to discuss potential new blk-mq APIs needed for a block driver to register its different priority queues, changes to blk-mq hwctx selection, and implications for low level drivers that utilize IRQ affinity to set up current mappings. Thanks, Keith
