#
✕ nick:marcan
#asahi-re · 2023-12-29 11:30
marcanbut there may well not be too
#asahi-re · 2023-12-29 11:30
marcanif it isn't and there is some folding into an e.g. 128-bit bus, there *may* be some higher-level structure, e.g. it can be split into two halves, but it depends
#asahi-re · 2023-12-29 11:29
marcanthe bus is probably 256 bits wide, as large as the sbox
#asahi-re · 2023-12-29 11:28
marcanI saw someone comment that there's no way the s-box is encoded in hardware like that and there must be a simpler boolean reduction, but that's not true; it really is arbitrary and encoded in hardware like that because it literally is a mapping of XOR gates to the cache SRAM input bus
#asahi-re · 2023-12-29 11:28
marcansame with the skipped entries, it's also effectively arbitrary
#asahi-re · 2023-12-29 11:27
marcanthere's probably some underlying hardware reason for that order, but it might well be "someone hand-optimized the XORs" or "the synthesis engine optimized the physical layout"
#asahi-re · 2023-12-29 11:27
marcansame way the order of bits going into / out of an SRAM array does not matter as long as it's the same
#asahi-re · 2023-12-29 11:27
marcanit does not matter at all, it's literally just the order of the wires in the circuit, and as long as the ECC computation is always consistent it does not matter
#asahi-re · 2023-12-29 11:26
marcanI suspect there is no particular logic to the entry order, since it is truly arbitrary
#asahi-re · 2023-12-29 11:26
marcannicolas17: the order of the s-box entries is the bit order; the values are basically the indices, except skipping powers of two and some unused indices, and plus that parity bit thing
#asahi-re · 2023-12-18 04:33
marcan13.5 is the current target
#asahi-re · 2023-11-14 16:42
marcanvbindiffing DCP I see the usual address changes everywhere, and a small fragment of code added/changed (a few instructions), plus a couple instructions removed elsewhere.
#asahi-re · 2023-11-14 16:39
marcanDCP firmware, not so much
#asahi-re · 2023-11-14 16:39
marcanonly build IDs change
#asahi-re · 2023-11-14 16:39
marcanthe kernel is obviously the same if you vbindiff it after extraction
#asahi-re · 2023-11-14 16:37
marcanI see changes there
#asahi-re · 2023-11-14 16:34
marcantry DCP firmware
#asahi-re · 2023-11-10 03:55
marcantheir AMCC driver already has the ability to log ECC errors there
#asahi-re · 2023-11-10 03:54
marcanit could be to log ECC error rates in general, for stuff like the internal caches or the internal LPDDR5 stuff, who knows
#asahi-re · 2023-11-10 02:23
marcanit can't be
#asahi-re · 2023-11-10 02:23
marcan"traditional" ECC isn't a thing in embedded like this
#asahi-re · 2023-11-10 02:23
marcanand we know the internal caches have ECC (at least MCC does)
#asahi-re · 2023-11-10 02:23
marcanamarioguy: I thought LPDDR5 had embedded ECC anyway
#asahi-re · 2023-08-04 09:05
marcanyes, a lot of keys are just derived
#asahi-re · 2023-08-04 08:51
marcan(from looking at all that battery/charge control stuff)
#asahi-re · 2023-08-04 08:51
marcanchadmed: https://mrcn.st/t/Screenshot_20230804_175004.png I actually have a decent chunk of it worked out, at least the overall structure
#asahi-re · 2023-06-13 13:14
marcan:D
#asahi-re · 2023-06-13 10:06
marcanwonder how that happened
#asahi-re · 2023-06-13 10:06
marcanSEP though.... fun.
#asahi-re · 2023-06-13 10:05
marcannicolas17: I have the iboot keys up to 12.0 betasomething, that's how I found the new SMC charge threshold feature
#asahi-re · 2023-06-05 21:49
marcanhttps://github.com/AsahiLinux/macvdmtool
#asahi-re · 2023-06-05 21:49
marcanroxfan: or you know, any Apple Silicon Mac
#asahi-re · 2023-06-02 12:35
marcantracing with the hypervisor
#asahi-re · 2023-05-09 15:00
marcan(protip: if you don't use that feature you can turn it off with the configurator, I do that for all of mine)
#asahi-re · 2023-05-09 14:57
marcanfredmb: yubikey? :p
#asahi-re · 2023-04-23 09:46
marcanyeah, as I said, I don't expect rotation to be a thing unfortunately
#asahi-re · 2023-04-23 09:40
marcansince it has to read backwards
#asahi-re · 2023-04-23 09:40
marcanprobably the width/height needed for initialization of xflip/yflip :p
#asahi-re · 2023-04-23 09:31
marcanI also recommend `from m1n1.shell import run_shell` and `run_shell(globals(), msg="Have fun!")` at the end
#asahi-re · 2023-04-23 09:30
marcanhttps://mrcn.st/p/FAb0vjLc
#asahi-re · 2023-04-23 09:29
marcannote: this is on t8103
#asahi-re · 2023-04-23 09:29
marcanChaosPrincess: with just that one line changed I get a flipped rainbow
#asahi-re · 2023-04-23 09:16
marcan2 is Xflip and 4 is Yflip
#asahi-re · 2023-04-23 09:15
marcanit makes two side by side rainbows so you can actually tell when it Y flips
#asahi-re · 2023-04-23 09:12
marcanChaosPrincess: pushed a thing
#asahi-re · 2023-04-23 09:12
marcanI think your framebuffer generator is a bit broken
#asahi-re · 2023-04-23 09:06
marcanyes
#asahi-re · 2023-04-23 09:05
marcanyes
#asahi-re · 2023-04-23 09:05
marcanChaosPrincess: bit 0x2 is xflip, works for me
#asahi-re · 2023-04-23 08:55
marcankeep in mind that rotation is probably going to explode badly if the dimensions are wrong, you'll probably want to try with a square layer

plus ›