Re: [OE-core] [RFC PATCH] Add genericarm64 MACHINE using upstream defconfig
From: Paul Gortmaker [[RFC PATCH] Add genericarm64 MACHINE using upstream defconfig] On 21/02/2024 (Wed 10:57) ross.bur...@arm.com wrote: > From: Ross Burton > > This is a new 64-bit "generic" Arm machine, that expects the hardware to > be SystemReady IR compatible. This is slightly forward-leaning as there's > not a _lot_ of SystemReady hardware in the wild, but most modern boards > are and the number will only grow. Also, this is the only way to have a > 'generic' machine as without standardised bootloaders and firmware it > would be impossible. > > The base machine configuration isn't that exciting: it's a fully featured > machine that supports most things, booting via UEFI and an initramfs. > > However, the kernel is more interesting. This RFC uses the upstream defconfig > because unlike some other platforms, the arm64 defconfig is actively > maintained with the goal of being a 'boots on most hardware' configuration. > My argument is: why would we duplicate that effort? I have no problem with the idea of a genericarm64, but the defconfig approach is not the way to go. > The "linux-yocto way" is configuration fragments and after a week of > hair-pulling I do actually have fragments that boot on a BeaglePlay, but > to say this was a tiresome and frustrating exercise would be understating it. Yes, if you start with a giant "defconfig" file and try and slowly chip away at it, it can be frustrating and time consuming. I never do that. It is the same problem space as trying to turn core-image-full-cmdline into core-image-minimal by removing one package at a time. Always easier to start with the bare minimum and then add stuff. > So, a request for comments: is it acceptable to use the upstream defconfig in > a reference BSP? Personally I'm torn: the Yocto way is fragments not > monolithic In a word, the answer is "no". Let me insert a bit of history here - darn near 20y ago now - when we were just starting out with linux, we had BSPs on different kernel versions, and worse - the use of flat monolithic "defconfig" type files on a per BSP basis. The latter was to be blunt, just crap. There was no uniformity across BSPs in the same "family" - where family would be something like CGL back in the 2005 era. One BSP might have built iso9660 fs, another one not. Or one enabled some debug option and another one didn't. The inconsistency was just horrible and unbearable. BSPs were screwing with options they had no business touching. After a couple years of that mess, I believe several of us (which included Bruce and myself) had just about enough. What we came up with was isolating the board config to a select subset of hardware/driver Kconfig options and the "platform" would handle "non-harware" options. That largely is still what exists in Yocto today, even though we were not using Yocto waaay back then. It immediately brought consistency across platforms, and a stark clarity to exactly what the board required for Kconfig settings - typically less than a screen full - choose the CPU and the driver for video and disk/flash and USB chip etc. Compare that to the unreadable 10,000+ lines in a defconfig now. So, yes I'll grant you that making that one screen full of core hardware settings can be frustrating. But at the same time, it has also served as a valuable filter. If a person can't distill a BSP down into the core options it *really* needs, then they typically either don't understand the hardware, or are just not invested enough to go that extra mile. I'm not saying either is the case here - just in general. On top of all that - we've had similar discussions before along similar lines - where Yocto/OE are not RedHat or Ubuntu - where we try and deliver a distro with "one config works everywhere" kind of stuff. I'm fine with genericarm64 - but please please don't encourage the chaos and madness associated with random unmaintainable defconfigs! Just because it exists upstream doesn't mean it is good. Thanks, Paul. > configs, but repeating the effort to fragmentise the configuration and then > also have it sufficiently modular that it can be used in pieces - instead of > just being a large file split up into smaller files - is a lot of effort for > what might end up being minimal gain. My fear is we end up with a fragmented > configuration that can't be easily modified without breaking some platforms, > and badly copies what the defconfig already does. > > Ross > --- > meta-yocto-bsp/README.hardware.md | 7 + > meta-yocto-bsp/conf/machine/genericarm64.conf | 26 +++ > .../linux/linux-yocto_6.6.bbappend| 9 +++ > meta-yocto-bsp/wic/genericarm64.wks.in| 11 > 4 files changed, 53 insertions(+) > create mode 100644 meta-yocto-bsp/conf/machine/genericarm64.conf > create mode 100644 meta-yocto-bsp/wic/genericarm64.wks.in > > diff --git a/meta-yocto-bsp/README.hardware.md > b/meta-yocto-bsp/README.hardware.md >
[OE-core] [RFC PATCH] Add genericarm64 MACHINE using upstream defconfig
From: Ross Burton
This is a new 64-bit "generic" Arm machine, that expects the hardware to
be SystemReady IR compatible. This is slightly forward-leaning as there's
not a _lot_ of SystemReady hardware in the wild, but most modern boards
are and the number will only grow. Also, this is the only way to have a
'generic' machine as without standardised bootloaders and firmware it
would be impossible.
The base machine configuration isn't that exciting: it's a fully featured
machine that supports most things, booting via UEFI and an initramfs.
However, the kernel is more interesting. This RFC uses the upstream defconfig
because unlike some other platforms, the arm64 defconfig is actively
maintained with the goal of being a 'boots on most hardware' configuration.
My argument is: why would we duplicate that effort?
The "linux-yocto way" is configuration fragments and after a week of
hair-pulling I do actually have fragments that boot on a BeaglePlay, but
to say this was a tiresome and frustrating exercise would be understating it.
So, a request for comments: is it acceptable to use the upstream defconfig in
a reference BSP? Personally I'm torn: the Yocto way is fragments not monolithic
configs, but repeating the effort to fragmentise the configuration and then
also have it sufficiently modular that it can be used in pieces - instead of
just being a large file split up into smaller files - is a lot of effort for
what might end up being minimal gain. My fear is we end up with a fragmented
configuration that can't be easily modified without breaking some platforms,
and badly copies what the defconfig already does.
Ross
---
meta-yocto-bsp/README.hardware.md | 7 +
meta-yocto-bsp/conf/machine/genericarm64.conf | 26 +++
.../linux/linux-yocto_6.6.bbappend| 9 +++
meta-yocto-bsp/wic/genericarm64.wks.in| 11
4 files changed, 53 insertions(+)
create mode 100644 meta-yocto-bsp/conf/machine/genericarm64.conf
create mode 100644 meta-yocto-bsp/wic/genericarm64.wks.in
diff --git a/meta-yocto-bsp/README.hardware.md
b/meta-yocto-bsp/README.hardware.md
index a8f38cb21a6..58ebc328b56 100644
--- a/meta-yocto-bsp/README.hardware.md
+++ b/meta-yocto-bsp/README.hardware.md
@@ -29,6 +29,7 @@ The following boards are supported by the meta-yocto-bsp
layer:
* Texas Instruments Beaglebone (beaglebone-yocto)
* General IA platforms (genericx86 and genericx86-64)
+ * General 64-bit Arm SystemReady platforms (genericarm64)
For more information see the board's section below. The appropriate MACHINE
variable value corresponding to the board is given in brackets.
@@ -126,6 +127,12 @@ USB Device:
dd command to write the image to a USB stick.
+SystemReady Arm Platforms
+=
+
+TODO
+
+
Texas Instruments Beaglebone (beaglebone-yocto)
===
diff --git a/meta-yocto-bsp/conf/machine/genericarm64.conf
b/meta-yocto-bsp/conf/machine/genericarm64.conf
new file mode 100644
index 000..2ea270d8b06
--- /dev/null
+++ b/meta-yocto-bsp/conf/machine/genericarm64.conf
@@ -0,0 +1,26 @@
+#@TYPE: Machine
+#@NAME: genericarm64
+#@DESCRIPTION: Generic Arm64 machine for typical SystemReady platforms, which
+#have working firmware and boot via EFI.
+
+require conf/machine/include/arm/arch-armv8a.inc
+
+# Arm Base System Architecture says v8.0+ is allowed, but FEAT_CRC32 is
required
+DEFAULTTUNE = "armv8a-crc"
+
+MACHINE_FEATURES = "acpi alsa bluetooth efi keyboard pci qemu-usermode rtc
screen usbhost vfat wifi"
+
+# Install all the kernel modules and all the firmware
+MACHINE_EXTRA_RRECOMMENDS += "kernel-modules linux-firmware"
+
+KERNEL_IMAGETYPE = "Image"
+PREFERRED_PROVIDER_virtual/kernel ?= "linux-yocto"
+INITRAMFS_IMAGE ?= "core-image-initramfs-boot"
+
+IMAGE_FSTYPES ?= "wic"
+WKS_FILE ?= "genericarm64.wks.in"
+
+EFI_PROVIDER ?= "${@bb.utils.contains("DISTRO_FEATURES", "systemd",
"systemd-boot", "grub-efi", d)}"
+
+# Try to bring up one physical serial console, or a virtualized serial console
+SERIAL_CONSOLES ?= "115200;ttyAMA0 115200;hvc0"
diff --git a/meta-yocto-bsp/recipes-kernel/linux/linux-yocto_6.6.bbappend
b/meta-yocto-bsp/recipes-kernel/linux/linux-yocto_6.6.bbappend
index 8e465c241e8..18f95de348f 100644
--- a/meta-yocto-bsp/recipes-kernel/linux/linux-yocto_6.6.bbappend
+++ b/meta-yocto-bsp/recipes-kernel/linux/linux-yocto_6.6.bbappend
@@ -1,19 +1,28 @@
KBRANCH:genericx86 = "v6.6/standard/base"
+KBRANCH:genericarm64 = "v6.6/standard/base"
KBRANCH:genericx86-64 = "v6.6/standard/base"
KBRANCH:beaglebone-yocto = "v6.6/standard/beaglebone"
+KMACHINE:genericarm64 ?= "genericarm64"
KMACHINE:genericx86 ?= "common-pc"
KMACHINE:genericx86-64 ?= "common-pc-64"
KMACHINE:beaglebone-yocto ?= "beaglebone"
+SRCREV_machine:genericarm64 ?= "332d4668fcc32826907d4f3c4938845206006089"
SRCREV_machine:genericx86 ?= "332d4668fcc32826907d4f3c4938845206006089"
