I see that DBG2/SPCR spec (https://docs.microsoft.com/en-us/windows-hardware/drivers/bringup/acpi-debug-port-table) defines 0x10 for Serial
type for BCM2835:
|
0x0010 |
BCM2835 |
Should the spec be updated to make this more explicit? i.e. “BCM283x/BCM27xx MiniUART”
From: devel@edk2.groups.io <devel@edk2.groups.io>
On Behalf Of Andrei Warkentin via groups.io
Sent: Tuesday, December 14, 2021 1:21 AM
To: Jeremy Linton <Jeremy.Linton@arm.com>; devel@edk2.groups.io; Ard Biesheuvel <ardb@kernel.org>; Adrien Thierry <athierry@redhat.com>; Pete Batard <pete@akeo.ie>
Cc: Ard Biesheuvel <ardb+tianocore@kernel.org>; Leif Lindholm <leif@nuviainc.com>
Subject: Re: [edk2-devel] [edk2-platforms PATCH] Platform/RaspberryPi: Fix miniuart base address and length
The Raspberry Pi support in edk2-platforms, including ACPI, is a direct ancestor of the original ms-iot tree (https://github.com/ms-iot/RPi-UEFI,
by way of https://github.com/andreiw/RaspberryPiPkg).
The way the miniUART is described in ACPI came from Microsoft. Microsoft introduced DBG2/SPCR type 0x10 (https://docs.microsoft.com/en-us/windows-hardware/drivers/bringup/acpi-debug-port-table)
and the BCM2836 _HID to describe the miniUART, and the contract is that the base address includes all those crazy registers.
To the best of my knowledge, today there isn't any other way to correctly describe the miniUART in a DBG2, SPCR or DSDT. Moreover, because there's code out there in at least two operating systems coded against these specific definitions,
you don't get to change how a _HID == BCM2836 device or SPCR/DBG2 type 0x10 is described.
If you wanted to introduce an alternate mechanism to describe the miniUART - great. You'd have to pick a new _HID. And re-use one of the generic DBG2/SPCR types or cajole for a new one (I'm guessing in the ASWG but I really don't know).
But you surely can't haphazardly change an existing firmware<->OS contract. Moreover, you can't deprecate the existing contract overnight as well, so you'd have to add an option to expose the miniUART using a presumably more-Linux friendly option.
If you do introduce a new mechanism to describe the miniUART in ACPI, I'm happy to add support for it in ESXi, paving the way for eventual deprecation of the current mechanism (assuming you get all the other OSes to play ball too...)
Still a NAK. It's not a fix because it's not broken. And it's not considered broken just because you don't like the definitions. I don't like the definitions either, but that's all we got.
A
From: Jeremy Linton <jeremy.linton@arm.com>
Sent: Monday, December 13, 2021 11:39 PM
To: devel@edk2.groups.io <devel@edk2.groups.io>; Andrei Warkentin <awarkentin@vmware.com>; Ard Biesheuvel <ardb@kernel.org>;
Adrien Thierry <athierry@redhat.com>; Pete Batard <pete@akeo.ie>
Cc: Ard Biesheuvel <ardb+tianocore@kernel.org>; Leif Lindholm <leif@nuviainc.com>
Subject: Re: [edk2-devel] [edk2-platforms PATCH] Platform/RaspberryPi: Fix miniuart base address and length
Hi,
On 12/13/21 13:17, Andrei Warkentin via groups.io wrote:
> If I understand correctly, you want to describe the UART at 0x00215000 to be at 0x00215040.
>
> This will break SPCR and DBG2 - so that's a regression for Windows, ESXi and possibly the NetBSDs.
>
> I guess that's a NAK unless I misunderstood something.
Presumably the end goal is to get BT working, or are we trying to get
the console working too?
Either way, the historical SPCR definition is less than ideal because it
covers those AUX_IRQ/AUX_ENABLE registers which include information for
the SPI which isn't included in the "uart" definition here. So, IMHO it
is wrong, but its stuck that way unless we define another uart. Which if
all we wanted it for was BT then we could just create another device
under BCM2836 which is only the 8250 like registers. That is sorta ugly,
but not having a standard uart is ugly too. The other ugly thing is to
just use the address as is, and offset it by 0x40 in linux as part of
the clock and ACPI bindings linux patch. (i've got a patch to make it
work someone wants to bite into it. Lol).
For linux the base clock-rate is going to have to be added as a _DSD
too. Which I assume is a large part of why it has a custom SPCR id? Put
another way, is anyone using the extra AUX_ registers, and what else are
people (windows/etc) "quirking" with the SPCR id?
For linux I've not particularly felt the need to fix this because I had
BT working (although unreliably) this time last year when I was working
on the SD/SDIO drivers, and my answer at the time was that one either
gets a serial console using the pl011 or one gets BT with the pl011. But
it looks like at a minimum the linux-firmware project and the brcmfmac
firmware loader have been tweaked over the past year and getting BT
working isn't as simple as just taking the miniuart-bt line out of
config.txt as I have in my not particularly good notes from that time
period.
So, while its behaving like it did when it had bad firmware, it could be
something in the lower level firmware since attempting to roll back to
an older firmware/kernel I had on another disk didn't immediately fix it.
> ________________________________
> From: Ard Biesheuvel <ardb@kernel.org>
> Sent: Monday, December 13, 2021 9:14 AM
> To: Adrien Thierry <athierry@redhat.com>; Andrei Warkentin <awarkentin@vmware.com>; Pete Batard <pete@akeo.ie>
> Cc: edk2-devel-groups-io <devel@edk2.groups.io>; Ard Biesheuvel <ardb+tianocore@kernel.org>; Leif Lindholm <leif@nuviainc.com>
> Subject: Re: [edk2-platforms PATCH] Platform/RaspberryPi: Fix miniuart base address and length
>
> On Mon, 13 Dec 2021 at 15:54, Adrien Thierry <athierry@redhat.com> wrote:
>>
>> Hi Ard, Leif, Pete
>>
>> Do you have any feedback on this patch ?
>>
>
> No objections from me but I'd like an ack from someone else as well.
>
>
>
>
>
>