[09:31:23] <zmatt> so, multicore cpu caches often do this sort of thing too, but with cacheline granularity
[09:32:26] <ds2> you mean what Linux uses or the hw can do?
[09:33:53] <zmatt> a net-shmem would have its pages readonly by default, on write you'd need to negotiate an exclusive lock which involves making the page read/write for that node and unreadable for all others. an access by someone else would then grab the updated version from the current page owner (which would revert to read-only or no-access depending on whether the other node wants to read or write)
[09:34:58] <ds2> IIRC - there are bits in the page entries to specify different shift or something like that
[09:35:22] <zmatt> there are usually a few page sizes
[09:35:58] <zmatt> though on ARM some of them just require 16 consecutive identical page table entries (but having the benefit of taking up only 1 TLB entry, thus reducing TLB pressure)
[09:36:50] <zmatt> some ancient ARM cores supported "tiny pages": 1 KB
[09:37:03] <zmatt> but other than that I've never seen smaller than 4 KB
[09:37:47] <zmatt> did you try the unshare btw ? it really is funny to see a user namespace in action
[09:38:38] <zmatt> creating only a new user namespace is so much weirder than a full container
[09:54:27] * Cats_home (~Thunderbi@dsl.198.58.173.185.ebox.ca) has joined #beagle
[09:57:43] * contempt (contempt@unaffiliated/contempt) Quit (Disconnected by services)
[09:58:03] * contempt (contempt@unaffiliated/contempt) has joined #beagleboard
[09:58:03] * contempt (contempt@unaffiliated/contempt) has joined #beagle