David Kahn wrote: > one more question. It's not clear to me how > the irm pool stuff works. > > Who actually allocates MSI/X address/data values > when these pools are used/defined? > > Did I miss that part of the design or interface? > > -David
The "Design and Implementation" spec has more about the mechanics of defining interrupt pools, mapping requests from devices into the pools, and keeping the pools balanced. Although in spite of all of its excruciating detail, I can see how the main point you are asking isn't explained there in a simple way. The interrupt pool is just an abstraction to represent some supply of interrupt vectors in the system. It describes how many they are, of what type, and who supplied them. And the request structures are an abstraction of how many interrupts a particular driver wants and gets from its respective interrupt pool. By mapping the two types of structures together, IRM can generically compute a fair number of interrupts to give to each device from its pool, in a way which is balancing together the needs of all the requests in the same pool. IRM then uses callbacks to tell the drivers when there is a change in those computations. During the callbacks, when the drivers then respond by calling ddi_intr_alloc(9F) or ddi_intr_free(9F), those functions are implemented by sending INTROPs to whichever nexus driver owns and manages the actual vectors represented by the interrupt pool. And there are private implementations in those nexus drivers to implement the actual vector assignments at that level. -- Scott
