On Wed, Jul 30, 2014 at 10:36 AM, Frank Ch. Eigler <fche@xxxxxxxxxx> wrote: >> > What kind of locking/serialization is provided by the ebpf runtime >> > over shared variables such as my_map? >> >> it's traditional rcu scheme. > > OK, that protects the table structure, but: > >> [...] In such case concurrent write access to map value can be done >> with bpf_xadd instruction, though using normal read/write is also >> allowed. In some cases the speed of racy var++ is preferred over >> 'lock xadd'. > > ... so concurrency control over shared values is left up to the > programmer. yes. It has to be flexible and fast. One of our main use cases is network analytics where a lot of packets are going through ebpf programs, so every cycle counts. Mandatory locks in critical path are not acceptable. If we add locks they will be optional. >> There are no lock/unlock function helpers available to ebpf >> programs, since program may terminate early with div by zero >> for example, so in-kernel lock helper implementation would >> be complicated and slow. It's possible to do, but for the use >> cases so far there is no need. > > OK, I hope that works out. I've been told that dtrace does something > similiar (!) by eschewing protection on global variables such as > strings. In their case it's less bad than it sounds because they are > used to offloading computation to userspace or to store only > thread-local state, and accept the corollary limitations on control. interesting. btw, things like global variables, per-cpu storage are potential ebpf features. So far they're 'nice to have' instead of 'mandatory'. The maps are powerful enough to do the same: Global storage is map of one element. Per-cpu storage is map of num_cpu elements. -- To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html