On Fri, 21 Feb 2020 at 04:24, Gavin Shan <gs...@redhat.com> wrote: > If above analysis is correct and the first approach doesn't work out. We have > to > consider the 2nd approach - adding option to backend to allow losing data. I'm > going to add "allow-data-lost" option for TYPE_CHARDEV_SOCKET. With the > option, > a back-off algorithm in tcp_chr_write(): The channel is consider as broken if > it fails to transmit data in last continuous 5 times. The transmission is > still > issued when the channel is in broken state and recovered to normal state if > transmission succeeds for once.
Before you do that, I would suggest investigating: * is this a problem we've already had on x86 and that there is a standard solution for? * should this be applicable to more than just the socket chardev? What's special about the socket chardev? I've added the chardev backend maintainers to the cc list. thanks -- PMM
