On Mon, Dec 10, 2018 at 5:27 AM Lucas Stach <l.stach@xxxxxxxxxxxxxx> wrote: > > From: Andrey Smirnov <andrew.smirnov@xxxxxxxxx> > > When operating at hight baudrates (> 1Mbaud) on a system that perfoms > several polling tasks, it is often the case that xy_read_block() > errors out due to the fact that incoming data overran serial FIFO and > some of the payload got lost. For those kind of situations it is not > very beneficital to wait for 3 seconds for every block lost this way, > and decreasing it in order to force a quick NAK to the host is > beneficial to overall throughput. > > This patch changes the timeout to be the bigger of 50ms or 10 times > the about of time it'd take to transfer a single payload block for a > given baudrate. > > Signed-off-by: Andrey Smirnov <andrew.smirnov@xxxxxxxxx> > Signed-off-by: Lucas Stach <l.stach@xxxxxxxxxxxxxx> I don't think this patch is really needed anymore. It was initially created to alleviate problems there were encountered when doing X/Y modem transfers on a system with a serdev driver running, but the final version of the serdev code that landed in the tree allows polling to be temporarily disabled by setting polling interval to 0: https://git.pengutronix.de/cgit/barebox/tree/common/serdev.c#n78 IMHO, in light of that, a user would be better off disabling polling, performing X/Y modem and re-enabling polling back rather than relying on a faster failed transfer recovery due to this patch. Thanks, Andrey Smirnov _______________________________________________ barebox mailing list barebox@xxxxxxxxxxxxxxxxxxx http://lists.infradead.org/mailman/listinfo/barebox
