Re: [PATCH v4 2/7] MIPS: Loongson64: Distinguish firmware dependencies DTB/LEFI

Am 10.03.21 um 16:58 schrieb Jiaxun Yang:

在 2021/3/10 21:26, Oleksij Rempel 写道:
[...]

quoted

I would like to understand, why it is impossible. Do fw_arg0 provide memory address or some kind of
count/size? Can it be negative?

We already had same situation with ARM and it was fixed. Why this can't be done for MIPS or LS?

Well we can fix it for future MIPS systems but not present Loongson.

Actually I've talked to several Chinese MIPS chip vendors and they
all agreed to unify the booting protcol.

I raised the same question to Loongson ~3 years ago but got
negative response.
The idea got reject in a recent private discussions with Loongson
about 2K1000 upstream stuff.

Also I realized Loongson's chip comes with too many ISA level
custom designs which makes it impossible to use MIPS generic
kernel.
 

quoted

quoted

quoted

This protocol is described here on page 15, "3. Boot protocols"
https://docplayer.net/62444141-Unified-hosting-interface-md01069-reference-manual.html

According to this protocol, you should have:
fw_arg0 = -2
fw_arg1 = Virtual (kseg0) address of Device Tree Blob

This would made LS a first grade resident for many boot loaders and
save a lot of needles headaches.

Loongson is stepping away from MIPS and it seems like they're going to use EDK-II for their Loongarch.

It seems to be UEFI related, it seems to be not related to the CPU arch, or do i'm missing something?

I meant they won't invest much to MIPS based chips :-(

quoted

In any case, if this is true, then it means, that Loongsoon is about to drop support for old boot
loaders (PMON?) and do new thing (one more boot protocol?). So argumentation, we upstream old own
protocol, but will drop it to make some thing new is not really good example :)

Yes, actually many new Loongson desktop machines (3A3000/3A4000) are shipped
with UEFI and relies on second stage bootloader to convert to old boot
protocol.

I'm againt to upstraming the "new" protocol as it's non-standard UEFI
with many
nonsense design.

quoted

quoted

TBH I've checked Loongson's PMON code and realized it can't be ported to other projects easily.
Tons of unregonized assembly code.

No need to port it. Here is example of working clean code:
https://git.pengutronix.de/cgit/barebox/tree/arch/mips/boards/loongson-ls1b/lowlevel.S

I've ported U-Boot to LS1C as well. But LS1B/LS1C is Loongson32.
Loongson64 based systems like LS2K is much more complex than it.
The bootloader is fullfilled with assembly and mostly undocumented.
LS2K's PMON contains ~50k lines of assembly.

OK, i hope i'll never have customers using this chips. Are there any
reason why do we wont to keep it alive? :D


--
Regards,
Oleksij