mirror of
git://projects.qi-hardware.com/xburst-tools.git
synced 2024-11-01 14:14:38 +02:00
qi-doc-README-update.patch
Signed-off-by: Andy Green <andy@openmoko.com>
This commit is contained in:
parent
8ce5b645e6
commit
e856eed04a
102
qiboot/README
102
qiboot/README
@ -1,5 +1,97 @@
|
|||||||
Backlight turns on right away, but screen remains blank until kernel load,
|
Qi
|
||||||
and kernel does NOT output gobs of text to the screen.
|
==
|
||||||
Will ALWAYS boot from uSD if first partition is ext2 and contains
|
|
||||||
/boot/uImage.bin, otherwise boots from NAND.
|
Qi (named by Alan Cox on Openmoko kernel list) is a minimal bootloader that
|
||||||
No splash screen.
|
"breathes life" into Linux. Its goal is to stay close to the minimum needed
|
||||||
|
to "load" and then "boot" Linux -- no boot menus, additional peripheral init
|
||||||
|
or private states.
|
||||||
|
|
||||||
|
|
||||||
|
What's wrong with U-Boot, they keep telling people to not reinvent the wheel?
|
||||||
|
=============================================================================
|
||||||
|
|
||||||
|
U-Boot is gradually becoming a simplified knockoff of Linux. After using it a
|
||||||
|
while, it became clear we were cutting and pasting drivers into U-Boot from
|
||||||
|
Linux, cutting them down and not having a plan to maintain the U-Boot versions
|
||||||
|
as the Linux ones were changed.
|
||||||
|
|
||||||
|
We decided that we would use full Linux for things that Linux is good at and
|
||||||
|
only have the bootloader do the device init that is absolutely required before
|
||||||
|
Linux can be pulled into memory and started. In practice since we had a working
|
||||||
|
U-Boot implementation it meant cutting that right down to the bone (start.S
|
||||||
|
mainly for s3c2442) and then building it up from scratch optimized to just do
|
||||||
|
load and boot.
|
||||||
|
|
||||||
|
|
||||||
|
Samsung - specific boot sequence
|
||||||
|
================================
|
||||||
|
|
||||||
|
Right now Qi supports Samsung "steppingstone" scheme devices, but in fact it's
|
||||||
|
the same in processors like iMX31 that there is a small area of SRAM that is
|
||||||
|
prepped with NAND content via ROM on the device. There's nothing that stops Qi
|
||||||
|
use on processors without steppingstone, although the ATAG stuff assumes we deal
|
||||||
|
with ARM based processor.
|
||||||
|
|
||||||
|
|
||||||
|
Per-CPU Qi
|
||||||
|
==========
|
||||||
|
|
||||||
|
Qi has a concept of a single bootloader binary created per CPU type. The
|
||||||
|
different devices that use that CPU are all supported in the same binary. At
|
||||||
|
runtime after the common init is done, Qi asks each supported device code in
|
||||||
|
turn if it recognizes the device as being handled by it, the first one to reply
|
||||||
|
that it knows the device gets to control the rest of the process.
|
||||||
|
|
||||||
|
Consequently, there is NO build-time configuration file as found on U-Boot
|
||||||
|
except a make env var that sets the CPU type being built, eg:
|
||||||
|
|
||||||
|
make CPU=s3c6410
|
||||||
|
|
||||||
|
|
||||||
|
Booting Heuristics
|
||||||
|
==================
|
||||||
|
|
||||||
|
Qi has one or more ways to fetch a kernel depending on the device it finds it is
|
||||||
|
running on, for example on GTA02 it can use NAND and SD card devices. It goes
|
||||||
|
through these device-specific storage devices in order and tries to boot the
|
||||||
|
first viable kernel it finds, usually /boot/uImage.bin.
|
||||||
|
|
||||||
|
The kernel commandline used is associated with the storage device, this allows
|
||||||
|
the correct root= line to be arrived at without any work. The inability to set
|
||||||
|
the Qi kernel commandline externally is deliberate, two otherwise identical
|
||||||
|
devices differing by the kernel commandline or other "environment" is not good.
|
||||||
|
A whole class of bugs and support issues around private bootloader state are
|
||||||
|
therefore avoided.
|
||||||
|
|
||||||
|
If no kernel image can be found, Qi falls back to doing a memory test.
|
||||||
|
|
||||||
|
|
||||||
|
Initrd support
|
||||||
|
==============
|
||||||
|
|
||||||
|
Initrd or initramfs in separate file is supported to be loaded at given
|
||||||
|
memory address in addition to kernel image. The ATAGs are issued accordingly.
|
||||||
|
|
||||||
|
|
||||||
|
Functional Differences from U-Boot on GTA02
|
||||||
|
===========================================
|
||||||
|
|
||||||
|
- Backlight is not enabled until Linux starts after a few seconds
|
||||||
|
|
||||||
|
- kernel loglevel is set to NOT output gobs of text to the screen
|
||||||
|
|
||||||
|
- On GTA02 will ALWAYS boot from uSD if first partition is ext2 and contains
|
||||||
|
/boot/uImage.bin, otherwise boots from NAND
|
||||||
|
|
||||||
|
- On GTA03 will ALWAYS boot from uSD second partition if /boot/uImage.bin is
|
||||||
|
present otherwise try the third / backup partition
|
||||||
|
|
||||||
|
- No startup splash screen
|
||||||
|
|
||||||
|
- Way faster
|
||||||
|
|
||||||
|
- There is no concept of "staying in the bootloader". The bootloader exits to
|
||||||
|
Linux as fast as possible, that's all it does.
|
||||||
|
|
||||||
|
- USB is not started until Linux starts around 5 seconds after boot, there is
|
||||||
|
no DFU.
|
||||||
|
Loading…
Reference in New Issue
Block a user