On Tue, 2011-04-19 at 00:54 -0700, Lennert Buytenhek wrote: > On Tue, Apr 19, 2011 at 11:12:50AM +0530, Nishant Sarmukadam wrote: > > > From: Pradeep Nemavat <pnemavat@xxxxxxxxxxx> > > > > Timestamp tx packets using a HW micro-second timer. > > This timestamp will be compared to the current timestamp > > in the hardware and if the difference is greater than 500ms, > > the packet will be dropped. > > Why don't you do this entirely in the firmware? > > When a packet is put into the TX ring by the host, give it a timestamp > of zero. > > When the firmware gets a TX interrupt, have it walk the TX ring from > the last TX entry where a timestamp was written to (i.e. keep a separate > index for this), and timestamp all TX ring entries that are marked as > being FW OWNED but don't have a timestamp yet (and update your index). > > Then when the time comes to process the TX packet (i.e. it has reached > the head of the TX queue), compare the timestamp with the current time. Since firmware cannot access host memory directly, it needs additional dma operations to read all descriptors, then check the ownership bit and update the timestamp for the descriptors. It can easily take about 1us per descriptor for this and it can hit the peak throughput by about 2.5%. Also, when the firmware is processing the packets from the host tx queues, it keeps the host interrupt for tx packets disabled to avoid unnecessary interruptions. Hence, the timestamp in the firmware can deviate from when the packet was queued in the host queue. > > As an added bonus, this'll also work over USB/SPI/whatever other bus > type where you cannot read some H/W microsecond timestamp register > directly like you can on PCI(e) -- which sticks out like a sore thumb > because it goes around the clean command / ring based host<->firmware > interface. We agree that this approach will not work without additional changes for non-PCI devices. However, we do not foresee such devices being supported by mwl8k in the near future. -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
