#
#asahi-re · 2022-12-02 23:53
eilnthank you! :))
#asahi-re · 2022-12-02 23:52
amarioguyread that writeup loved it
#asahi-re · 2022-12-02 23:52
amarioguyeiln: you're doing great tho
#asahi-re · 2022-12-02 23:52
amarioguyyea geohot is not what i'd call a paragon of ethical work...
#asahi-re · 2022-12-02 23:34
eilnsomething something dunning
#asahi-re · 2022-12-02 23:33
nicolas17still laughing at his recent over-confident statements in a live stream about how few people are needed to run Twitter, immediately followed by him reading a basic GraphQL tutorial
#asahi-re · 2022-12-02 23:31
svena decade or so ago he took our ps3 research, used that to release keys (which we deliberately avoided) without any credit and then got us sued by Sony
#asahi-re · 2022-12-02 23:30
nicolas17I missed who we're talking about *checks IRC log* OH yeah that makes sense
#asahi-re · 2022-12-02 23:29
sven“Ripping off others people research” is kinda his thing unfortunately…
#asahi-re · 2022-12-02 23:29
nicolas17today I found there's 4 different C tools to decompress Apple "pbzx" compressed streams and they *all* misinterpret what the fields actually mean, probably because they copied code off each other
#asahi-re · 2022-12-02 23:28
nicolas17eiln: ah overconfident misinterpretations...
#asahi-re · 2022-12-02 23:19
eilnsorry i had to pull the https://xkcd.com/386/
#asahi-re · 2022-12-02 23:18
eilni did not use any of his work in doing mine.
#asahi-re · 2022-12-02 23:18
eilnsounds like he's just puking out the strings of the kext binaries.
#asahi-re · 2022-12-02 23:17
eilni can take almost every sentence in that readme and nitpick it but i'm not.
#asahi-re · 2022-12-02 23:17
eilnstuff about L2 and the strides are completely wrong.
#asahi-re · 2022-12-02 23:17
eilnand that's ignoring whatever his view of an "op" is
#asahi-re · 2022-12-02 23:17
eilnthis value is *integral* because it's an input to the function to one of the "trigger" regs. the request will *not* go through without that 0x274.
#asahi-re · 2022-12-02 23:17
eilnnot 0x300
#asahi-re · 2022-12-02 23:17
eiln0xf4 + 0x3c + 0x6c + 0x44 + 0x0c + 0x10 + 0x18 + (0x7 * 8) + (0x28 or 0x32) = 0x274 or 0x278
#asahi-re · 2022-12-02 23:17
eilnwhich turns out to be
#asahi-re · 2022-12-02 23:17
eilnyou can see i went through the patent, figured out the encoding format ("Register Count" and "Register Address", specifically [31:24] and [23:0]), and derived the calculations behind the size of the "op"
#asahi-re · 2022-12-02 23:17
eilnhttps://gitlab.com/eiln/m1n1/-/tree/ane/proxyclient/m1n1/ane
#asahi-re · 2022-12-02 23:17
eilnif you look @ the TD section of my writeup
#asahi-re · 2022-12-02 23:17
eiln"At its core, it runs through 0x300 ops"
#asahi-re · 2022-12-02 23:16
eilnlet's take the second sentence for example,
#asahi-re · 2022-12-02 23:16
eilnwhich i'd get if it's concatted into something that works, but i really can't build _anything_
#asahi-re · 2022-12-02 23:16
eilnnot only is his work a very very small subsection of the npu but his entire work is ripped off of others' research without a full understanding of it
#asahi-re · 2022-12-02 23:16
eilni dont know how to say this without being an asshole but
#asahi-re · 2022-12-02 23:16
eilni have been summoned by the keyword ane :P
#asahi-re · 2022-12-02 19:13
amarioguyalright time for decoding random endpoints episode 10
#asahi-re · 2022-12-02 18:55
amarioguywhich ends very badly wrt any restore attempts
#asahi-re · 2022-12-02 18:55
amarioguynuking this means iboot won't be able to load the devicetree with the right properties
#asahi-re · 2022-12-02 18:55
amarioguydsharshakov: here's the short of it - the NOR region's name is called "SysCfg" it's what holds the serial number, device color, calibration data, etc (I haven't looked at it myself but that's part of it)
#asahi-re · 2022-12-02 14:47
Dcowlooks like Geohot done some ANE RE, not sure if it's relevant, but https://github.com/geohot/tinygrad/tree/master/accel/ane
#asahi-re · 2022-11-29 14:29
marcanBut there's probably a much easier trick: just change the reg to 0 and add 0x4000. DAPF is always one page after DART and instance 0 should match up with the first reg.
#asahi-re · 2022-11-29 14:27
marcaneiln: ah yes, the instancing thing is kind of complicated. I think what's supposed to happen is you count the instances from the length of the `instance` property and then the DAPFs come after the # instances
#asahi-re · 2022-11-29 14:14
amarioguyeiln: nice
#asahi-re · 2022-11-29 00:54
eilnokay nvm, wrote the dart stuff back when i was fussing w/ asc. none of the dapf stuff is necessary except for ttbr
#asahi-re · 2022-11-29 00:14
eilnyea filter-data uses the 3rd prop (for the ane at least)
#asahi-re · 2022-11-29 00:12
eilnand filter-data-instance-0 to the 3rd
#asahi-re · 2022-11-29 00:11
eilndart-tunables-instance-0,1,2 goes to the 0,1,2 regs
#asahi-re · 2022-11-29 00:11
eilnoh wait I see
#asahi-re · 2022-11-29 00:00
eilneverything else is handled well
#asahi-re · 2022-11-28 23:56
eilnremap for the 2nd isn't messed up
#asahi-re · 2022-11-28 23:54
eiln0x26b810000+00008c REMAP[3] = 0x2 (MAP3=0x0, MAP2=0x0, MAP1=0x0, MAP0=0x2)
#asahi-re · 2022-11-28 23:53
eiln0x26b810000+000088 REMAP[2] = 0xb050000 (MAP3=0xb, MAP2=0x5, MAP1=0x0, MAP0=0x0)
#asahi-re · 2022-11-28 23:53
eiln0x26b810000+000084 REMAP[1] = 0x1 (MAP3=0x0, MAP2=0x0, MAP1=0x0, MAP0=0x1)
#asahi-re · 2022-11-28 23:53
eiln0x26b810000+000080 REMAP[0] = 0x1 (MAP3=0x0, MAP2=0x0, MAP1=0x0, MAP0=0x1)
#asahi-re · 2022-11-28 23:53
eilnadding dapf:

plus ›