On Sun, Mar 10, 2024 at 3:05 PM Puranjay Mohan <puranjay12@xxxxxxxxx> wrote: > > Hi Alexei, > > On Sat, Mar 9, 2024 at 5:38 PM Alexei Starovoitov > <alexei.starovoitov@xxxxxxxxx> wrote: > > > > Puranjay, > > > > Looks like I have to drop this patch for now, > > since PMD_SIZE is not a constant on all archs > > that can be evaluated by a preprocessor. > > > > We need to find a different way. > > > > How about we define it like: > -#define BPF_PROG_PACK_SIZE (PMD_SIZE * num_possible_nodes()) > +/* > + * PMD_SIZE is really big for some archs. It doesn't make sense to > + * reserve too much memory in one allocation. Cap BPF_PROG_PACK_SIZE to > + * 2MiB * num_possible_nodes(). > + */ > +#define BPF_PROG_PACK_SIZE ((PMD_SIZE <= (1 << 21)) ? (PMD_SIZE * > num_possible_nodes()) \ > + > : ((1 << 21) * num_possible_nodes())) > > So, it will be computed at runtime. This adds more performance > overhead but this is called very infrequently > so shouldn't matter. > > Or we can hardcode it as `2MB * num_possible_nodes()` because > PMD size in most architectures will be 2MB or larger but never smaller. I think hard coding is cleaner and less surprising. I'd like to hear what Song thinks about it.
