On Thu, Aug 31, 2017 at 11:10 PM, Alexei Starovoitov <alexei.starovoi...@gmail.com> wrote: > On Thu, Aug 31, 2017 at 11:04:41PM -0400, Willem de Bruijn wrote: >> On Thu, Aug 31, 2017 at 10:10 PM, Alexei Starovoitov >> <alexei.starovoi...@gmail.com> wrote: >> > On Thu, Aug 31, 2017 at 05:00:13PM -0400, Willem de Bruijn wrote: >> >> From: Willem de Bruijn <will...@google.com> >> >> >> >> Documentation for this feature was missing from the patchset. >> >> Copied a lot from the netdev 2.1 paper, addressing some small >> >> interface changes since then. >> >> >> >> Signed-off-by: Willem de Bruijn <will...@google.com> >> > ... >> >> +Notification Batching >> >> +~~~~~~~~~~~~~~~~~~~~~ >> >> + >> >> +Multiple outstanding packets can be read at once using the recvmmsg >> >> +call. This is often not needed. In each message the kernel returns not >> >> +a single value, but a range. It coalesces consecutive notifications >> >> +while one is outstanding for reception on the error queue. >> >> + >> >> +When a new notification is about to be queued, it checks whether the >> >> +new value extends the range of the notification at the tail of the >> >> +queue. If so, it drops the new notification packet and instead increases >> >> +the range upper value of the outstanding notification. >> > >> > Would it make sense to mention that max notification range is 32-bit? >> > So each 4Gbyte of xmit bytes there will be a notification. >> > In modern 40Gbps NICs it's not a lot. Means that there will be >> > at least one notification every second. >> > Or I misread the code? >> >> You're right. The doc does mention that the counter and range >> are 32-bit. I can state more explicitly that that bounds the working >> set size to 4GB. Do you expect this to be problematic? Processing >> a single notification per 4GB of data should not be a significant >> cost in itself.
Actually, the counter is not a byte counter. It is incremented on each system call that sends data with MSG_ZEROCOPY. So the 4GB limit would only hold if a caller sends single byte requests at a time. I will make this more clear in v2. > > I think 4GB is fine. Just there was an idea that in cases when > notification of transmission can be known by other means the user space > could have skipped reading errqeuee completely, but looks like it > still needs to poll. That's fine. >> > Thanks for the doc! >> >> Thanks for reviewing :) >> >> > >> > Acked-by: Alexei Starovoitov <a...@kernel.org> >> >
