On Tuesday, September 9, 2008 5:57 pm Yang, Sheng wrote: > On Wednesday 10 September 2008 00:45:12 Jesse Barnes wrote: > > On Wednesday, August 27, 2008 10:38 pm Yang, Sheng wrote: > > > + /* Wait for Transaction Pending bit clean */ > > > + msleep(100); > > > > Looks pretty straightforward, but I wonder if these 100ms sleeps will > > generally be enough? Or maybe we can make them shorter? Is there some > > specified time we need to wait after reset (sorry I don't have my spec > > handy). > > The spec (6.6 FLR - Implementation Note) said: > > " Software polls the Transactions Pending bit in the Device Status register > either until it is clear or until it has been long enough that software is > reasonably certain that Completions associated with any remaining > outstanding Transactions will never arrive. On many platforms, the > Transactions Pending bit will usually clear within *a few milliseconds*, so > software might choose to poll during this initial period using a tight > software loop. On rare cases when the Transactions Pending bit does not > clear by this time, software will need to poll for a much longer > platform-specific period (potentially seconds), so software might choose to > conduct this polling using a timer-based interrupt polling mechanism." > > Compared to "a few ms", I think 100ms should be OK here. Maybe we can get > it shorter. but the spec is not that precise in this point, maybe we can > have a test? But it's not that general... > > But for after reset, we need to wait for 100ms. The spec make it clear this > time. :) Ok, thanks for clarifying Sheng. I'll queue it up for linux-next assuming there are no other objections. -- Jesse Barnes, Intel Open Source Technology Center -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
