On Wed, 22 May 2002, Benoit PAPILLAULT wrote: > Ok. But how do I manage 12 or 16 BT878 chips on a machine with several PCI busses (but > most cards are on the same bus) using the ServerWorks South Bridge (can't find data sheet > for it, and already modify a lot the linux code to make it work [resolving IRQ conflict!])? You might want to set them all lowish. The PCI latency timer is the max number of clocks a device can perform busmaster transfer. After the timer expires, the device must let the bus go and grab it again if it still has more work to do. PCI busmaster devices have two registers, min grant and max latency, that tell about what kind of access they need. Min grant is the shortest time a PCI device can access the bus. Therefore, the latency must be greater than or equal to the min grant (note that PCI latency is in clock ticks, while min gnt is in some other units (quarter microseconds?)). The max latency is the longest the device can go without access to the PCI bus. If you add up the latency of all the other PCI devices, it should be less than the max latency the device can handle. Of course, this is probably not possible, as the latency of all other PCI devices will probably be much greater than max latency a given device supports. You also want the latency as high as possible, as fewer larger transfers make more efficient use of the PCI bus.
