asciilifeform: so elementary final solution is - instrumented ram.
asciilifeform: btw at some point somebody will ask, 'why not use existing gadget with dma, e.g. usb3, to diddle memory'. so answer, intel et al are implementing 'iommu', so that only nsa-approved diddles go through dma controller. plus os can see if it has usb3 etc hanging off
asciilifeform: it dun win much, though, linux is a hog
asciilifeform: if you know something about the page table, can sorta cheat
asciilifeform: ( and that was a toy box, with 1G !! )
asciilifeform: i had this , handmade and orcish, using sageprobe . but it took ~2h per bringup...
asciilifeform: the ultimate win, of course, is that it also gives you a pc that you can magic into a known state on demand.
asciilifeform: bonus if you have a cpu that lets you kill cache
asciilifeform: stick also sees the address and data lanes, so can also record all ram fetches ( such that don't hit the cache )
asciilifeform: no matter how crippled ( see amd thread from september )
asciilifeform: anyway for n00bz : a two-hole ram stick will give you snapshots of physical memory, incl. 'prohibited' (i.e. SMM) pages, undetectably to ring-whatever soft. and likewise write - gives you much of what debug probe gives, with the difference that cpu could not refuse
asciilifeform: will fail boot test, but you can't have everything!1
asciilifeform: lol mircea_popescu can make this with a sharp knife right now!1111
asciilifeform: ( re pc vivisection : one item on asciilifeform's wish list, that does not currently exist : a doctored stick of sdram that can be read/written 'out of band', i.e. via another hole )
asciilifeform: still not quite pc bus speed. but massive improvement over the crapola that litters asciilifeform's bench, at least on paper.
asciilifeform: this'll have to wait for my own napkin.
asciilifeform: whereas as i understand it, we do not have this.
asciilifeform: well, there - one bit of input per one of output. whereas here is a hash: flipping ANY bit of the input should have an impractical-to-compute effect on ~all~ of the output
asciilifeform: my current reading of the algo incidentally is that it IS possible for the last bit of S to flip only 1 bit of R.
asciilifeform: transform string S into a larger S' where -- no matter what substructure of S enemy might know, he knows nothing at all re any part of S', nor any relation between any particular substrings of S'.
asciilifeform: see the linked thread. original subj was (the eternally misnamed) 'rsa padding' problem.
asciilifeform: that nevertheless decompresses in something like O(N)
asciilifeform: and while we're on subj, i'd still like to see the thing applied in for the purpose i originally suggested -- a 'enemy knows 0 structure ahead of time' bitstring transform
asciilifeform: ( i have not yet tried, hands -- very full atm )
asciilifeform: incidentally mircea_popescu's hash may well be susceptible to hashlifeization
asciilifeform: i don't see any theoretical basis for this statement
asciilifeform: observation is that 'gnarly slow hash with bitwise sliders' is not so much of a pill against theenemy as running on pc hardware suggests
asciilifeform: whereas for item that ~you wrote~, you already have the kind of picture that a sampling profiler is giving you -- in your head☟︎
asciilifeform: phf: imho sampling profilers are a wholly useless thing, 'horse with pedals', unless you're working a honeywagon (e.g. virginal trb) and have deeply nfi what the hell the program is doing
asciilifeform: the plan 'for broken rng' is to have 7 running from 7 batteries.
asciilifeform: mircea_popescu: one nuance is -- it is folly to plan FOR a broken rng. 'what to do if a shell lands in my trench? jump twenty metres and scatter yerself around'
asciilifeform: mircea_popescu: i can't imagine it wouldn't be well known. but i was digging for a 'schoolbook' description of it, which i thought existed.
