15:53 babyfaceold: elimination strategy is reasonably easy, you add to the focal value a common value between inverse and focal to the focal value, multiply the value by two and subtract from bound, then you add contiguous indexes, you index into one value in the bank where addition is done from the times two the buffer and index, you then eliminate the combined bound, then you subtract times one buffer, then you add index inverse of index bound and add
15:53 babyfaceold: times one buffer, you eliminate rest of the index bounds and there it is the value, it suffers under weak crypto compared to polynomial functions like elliptic curve though. but unrolled version is very fast and compresses all the memory very densely as if you had a giant register file.
16:06 babyfaceold: but the real issue with those are not the crypto but NIC MAC address attacks , they can intrude well using a certain vulnerability of PCI interconnect, so this is some work that i try to do, you offer a window of NIC through the mac hash like io pagetables, and would want to serialize the access to the devices, so that memory race would not happen on the buses. Cause pci is multi-master, isa buses were securer.
16:10 babyfaceold: I personally got fed up of having my computer as an open door day for everyone, but this is an attack that uses hw vulnerability its arp and ndp header collision attack, where they can get to an interrupt handler, cause of multi-master processing.
16:13 babyfaceold: so you would want to hold all other devices idle through remapping the base address of configuration space into a shadow io table in memory that serializes one device one cpu sequentially, then there is no opportunity for the race
16:16 babyfaceold: that would end up being similar to isa bus, there is only one master, and one slave, and cpu does not process in parallel with io
16:19 babyfaceold: my friends employers company was hit by ransomware and this thing is dead serious, if there is no way to secure the nic there is no business, cause internet needs to be used at times on the work computers too.
16:29 babyfaceold: i find the bus securing thingy more complex to performance code, to serialize the devices where cpu is also one on the bus, there is few info, but i read that parity errors and latency timers and such things can be forced to the processing queues
16:42 babyfaceold: http://hyc.io/spacetime/README.html that was one of the first lattice based solver stuff, miden-vm is seeming more advanced and friendly though, it's apache licensed but i can not understand the docs nor code for the last, miden-vm i can read.