14:52 <CounterPillow> ethz projects and not having users, name a more iconic pair

15:42 <tomeu> hey, I played with Oberon and Modula-2 as a kid :)

17:20 <rah> is there (likely to be) any effort to reverse engineer the proprietary firmware blob?

17:35 <robmur01> TBH if you don't trust Arm to write part of their GPU in C, you've little reason to trust the part written in Verilog either...

17:37 <rah> robmur01: that's not helpful

17:37 <robmur01> no, but it is pragmatic

17:38 <robclark> and true

17:38 <rah> but why are you telling me?

17:38 <rah> you're not being helpful

17:39 <alyssa> it *is* helpful

17:39 <alyssa> IMHO

17:39 <rah> *facepalm*

17:40 <robmur01> I'm implying that reverse-engineering CSF firmware is unlikely to be worthwhile for any reason other than extreme curiosity

17:41 <rah> robmur01: why not state such plainly then?

17:44 <rah> regardless, just because something is done for reasons of "curiosity" doesn't mean the result won't be useful to others

17:44 <robmur01> because that's not my style :)

17:48 <rah> also, the freedom of users isn't about trust, it's about freedom

17:51 <robclark> if the fw is redistributable, then how does it matter? It's like saying "I can't use this hw because I don't have the vhdl".. (not that I'm against open hw, just trying to point out that the anti-fw crowd is a bit silly)

17:52 <rah> that's a straw man

17:52 <rah> I don't really know what you're referring to as "the anti-fw crowd", as if there is just one group all with the same views

17:54 <rah> for me, what you've said is a mischaracterisation, it's a straw man; I wouldn't say "I can't use this hw because I don't have the vhdl" but I would say "I perfer hw for which I can get the vhdl", as I say "I prefer hw which does not require proprietary firmware blobs in order to operate"

17:57 <robmur01> and you're free to prefer that, and I for one support your choice, but it's also pretty much where the meaningful "freedom" argument ends

17:57 <rah> that's a one true scotsman

17:58 <rah> your concern is "meaningful freedom"

17:59 <alyssa> please take this elsewhere

17:59 <alyssa> this has veered pretty far from open mali drivers

18:00 <robmur01> +1

18:00 <rah> where, presumably, "meaningful" is malleable and means whatever you want it to mean, such that a true scotsman only concerns themselves with "meaningful freedom"

18:00 <alyssa> i've spent a lot of time on Franklin St, i don't fundamentally disagree

18:01 <alyssa> and obviously I would also prefer more open hw

18:02 <alyssa> but job manager mali, which has the extra hw baked into silicon in verilog you don't get, is not meaningfully more free in any sense than csf mali, which has a fw blob for part of the same

18:03 <alyssa> and the linux-libre policy of supporting JM Mali but not CSF Mali amounts to "if I can't see you, you can't see me"

18:03 <alyssa> and.. I think I've said more than I wish to

18:03 <robmur01> you did say it well, though :)

18:03 <alyssa> later this year we may publish some tooling that would help with CSF firmware r/e for those with extreme curiousity

18:04 <alyssa> we have a pretty good idea of the physical hw interface and could write a free fw if made sense to do so, but ATM it doesn't look like it makes sense to do so

18:04 <alyssa> that could change, of course

18:05 <alyssa> I will say that the CSF firmware is frighteningly boring

18:05 <alyssa> when I last looked (in dec 2021?!) it was like 64kb and 2% of that was their libc's printf

18:06 <alyssa> those of us working on the kernel mode driver for CSF Mali have had purely technical reasons to prefer using either Arm's fw blob or writing our own

18:06 <alyssa> (I'm not working on that driver so won't speak to them in detail)

18:07 <alyssa> but I don't think those reasons should concern end users too much

18:07 <alyssa> IDK

18:07 <alyssa> Maybe I've sold out and need to go back to the 2018 Alyssa who pm'd robclark from her desk on Franklin St asking about the Adreno blobs

18:08 <alyssa> robclark: I believe you told me roughly what I parrotted above, it just took me a few years to get where you were coming from :-p

18:10 <alyssa> Maybe I should've said anything. Keep discussing if you want. I'm too jaded at this point to care.

18:11 <alyssa> no ethical gfx under capitalism or something

18:12 <alyssa> *shouldn't have

18:13 <alyssa> What You Own - Rent

18:16 <robmur01> I believe the very very rough summary of CSF FW is: 1) scheduling gubbins that is made sufficiently complex by Vulkan requirements that it would be overly bug-prone to do in Verilog, 2) abstract low-level power management etc. details that the driver doesn't want to have to know about, 3) hide hardware bugs that the driver doesn't want to have to know about

18:16 <robclark> yeah, the fw basically babysits the hw and sequences work on the gpu.. not super exciting. On the adreno side we have tools to r/e (disasm/asm, etc.. unlike csf, sqe is a custom isa) for the curious.. but it is mostly kinda boring so meh. You could burn the fw into the hw, which would somehow make some people happy, but fundamentally changes nothing

18:17 <alyssa> fwiw, the CSF firmware runs on a Cortex-M7, according to the chip id register I dumped :~P

18:20 <robclark> I kinda like that sqe has a custom isa.. keeps it simple.. it doesn't even really have a stack.. and mostly doesn't have load/store instructions.. but it's kinda a neat isa.. but I guess that is getting more off topic ;-)

18:22 <rah> robclark: requiring firmware that isn't baked into hardware gives the vendor the power over who can and cannot legally use their hardware; that's pretty fundamental IMHO

18:22 <alyssa> only if non-redistributable?

18:23 <robclark> right, that is why I mentioned the redistributable part earlier

18:23 <rah> refraining from exerting one's power over someone does not mean that one does not have that power

18:23 <alyssa> (redistributability is a red line of course)