On Mon, 2017-05-01 at 13:40 -0500, Mike Christie wrote: > On 04/30/2017 06:29 AM, Xiubo Li wrote: > > [...]
<SNIP> > >> To avoid starvation, I think you want a second list/fifo that holds the > >> watiers. In tcmu_get_empty_block if the list is not empty, record how > >> many pages we needed and then add the device to the list and wait in > >> tcmu_queue_cmd_ring. > >> > >> Above if we freed enough pages for the device at head then wake up the > >> device. > >> > >> I think you also need a wake_up call in the completion path in case the > >> initial call could not free enough pages. It could probably check if the > >> completion was going to free enough pages for a waiter and then call > >> wake. > >> > > Yes, I meant to introduce this later after this series to not let the > > patches too > > complex to review. > > > > If you agree I will do this later, or in V7 series ? > > > Yeah, I am ok with adding it after the initial patches go in. Btw, adding your Acked-by for the initial merge of these. If that's a problem, please make some noise. ;)