On Sun, Oct 20, 2019 at 10:51:10PM +0000, Christopher Lameter wrote: > On Thu, 17 Oct 2019, Roman Gushchin wrote: > > > Currently s8 type is used for per-cpu caching of per-node statistics. > > It works fine because the overfill threshold can't exceed 125. > > > > But if some counters are in bytes (and the next commit in the series > > will convert slab counters to bytes), it's not gonna work: > > value in bytes can easily exceed s8 without exceeding the threshold > > converted to bytes. So to avoid overfilling per-cpu caches and breaking > > vmstats correctness, let's use s32 instead. > > Actually this is inconsistenct since the other counters are all used to > account for pages. Some of the functions in use for the page counters will > no longer make sense. inc/dec_node_state() etc. Actually I tried to implement what Johannes suggested earlier and convert all counters to bytes, but it looked like it can lead to a significant performance hit on 32bit platforms, as I'd need to replace all atomic_long_t with atomic64_t. Also, to make the code look good, I'd need to convert all counters to bytes (and atomic64_t): zone stats, vmevents, etc. So I gave up relatively quickly. Maybe it's a good long-term plan, but as now it doesn't really look as an obviously good think to do. I'm fully open to any suggestions here.
