On Sun, 26 Apr 2020 22:26:54 -0700 John Fastabend <john.fastabend@xxxxxxxxx> wrote: > Jesper Dangaard Brouer wrote: > > Clearing memory of tail when grow happens, because it is too easy > > to write a XDP_PASS program that extend the tail, which expose > > this memory to users that can run tcpdump. > > > > Signed-off-by: Jesper Dangaard Brouer <brouer@xxxxxxxxxx> > > --- > > Hi Jesper, Thanks for the series any idea what the cost of doing > this is? If you have some data I would be curious to know a > baseline measurment, a grow with memset, then a grow with memset. > I'm guess this can be relatively expensive? I have a "time_bench" memset kernel module[1] that I use to understand that is the best-case/minimum overhead with a hot-cache. But in this case, the memory will be in L3-cache (at least on Intel with DDIO). For legitimate use-cases, the BPF-programmer will write her tail data into this memory area anyhow. Thus, I'm not convinced this will be a performance issue for real use-cases. When we have a real use-case that need this tail extend and does XDP_TX, I say we can revisit this. [1] https://github.com/netoptimizer/prototype-kernel/blob/master/kernel/lib/time_bench_memset.c -- Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
