This makes a new behaviour for Qi, when there are no valid kernels then it
performs the memory test. It means you can either move kernels out of the way
to get the memory test, or make an SD Card just with normal Qi on it and no
kernels to get a memory test.
Signed-off-by: Andy Green <andy@openmoko.com>
Some of utils.c isn't used until the full Qi image has been loaded into
memory, to save space in 4K steppingstone case on 2442, we split utils.c now
so only the interesting routines for steppingstone time take up space there.
Signed-off-by: Andy Green <andy@openmoko.com>
We need putc even when we are operating entirely from steppingstone.
Arrange that the board-specific putc code is in a section that goes into
steppingstone, and adapt the utils.c putc() so that it no longer needs to
indirect through the board_api struct that is in main memory.
Signed-off-by: Andy Green <andy@openmoko.com>
This adds the memory testing code to Qi. It tests the range of memory with
several constants and then a 32-bit walking 1 pattern, and repeats forever.
The entire main SDRAM can be tested due to the fact this runs out of
steppingstone only and does not need to store anything outside of it.
It introduces a steppingstone-based stack for use entirely in steppingstone.
Signed-off-by: Andy Green <andy@openmoko.com>
Improve glamo-mmc so that it uses MMC_READ_MULTIPLE_BLOCK and copes with
64 blocks in one command (and STOP sent after each set), instead of sending
a new command per-block
Signed-off-by: Andy Green <andy@openmoko.com>
Little cleanup and new code that collects and defers contiguous sector reads
into one potentially more efficient larger sequential read action.
This matters for example on SD Card protocol.
Signed-off-by: Andy Green <andy@openmoko.com>
Naughty bug bug U-Boot ext2 implementation -- cut and paste of stanza above for
indir1 check used for indir2 but not all the check was not updated.
This removes hundreds of repeated reads of same sector during kernel load.
Signed-off-by: Andy Green <andy@openmoko.com>
There was a bug about dealing with last ext2 sector of file
because we failed to return the number of sectors pulled
Signed-off-by: Andy Green <andy@openmoko.com>
This heavily adapts the Samsung U-Boot hs_mmc code and combines it with the
SD / SDHC startup code written for glamo-mci stuff that is known to work OK
with common SD and SDHC.
tla01 is changed to use the implementation.
Signed-off-by: Andy Green <andy@openmoko.com>
We want utils stuffs in steppingstone part, it's too expensive to have
CRC32 in there. Bust it out into its own file that appears only in
everything_else section in second bootloader stage.
Signed-off-by: Andy Green <andy@openmoko.com>
This patch makes qi source structure cpu-centric, and allows multiple CPUs
to be handled with the board definitions inside the CPU dirs.
You have to make a particular CPU version of Qi now, which you can do by
a shell command like this:
make clean ; make CPU=s3c2442 && make CPU=s3c6410
which gets you
$ ls -l image/
total 744
-rwxrwxr-x 1 agreen agreen 25372 2008-10-17 18:25 qi-s3c2442-andy_77c1fcdddc3e2cbf
-rw-rw-r-- 1 agreen agreen 237100 2008-10-17 18:25 qi-s3c2442-andy_77c1fcdddc3e2cbf.dis
-rw-rw-r-- 1 agreen agreen 25388 2008-10-17 18:25 qi-s3c2442-andy_77c1fcdddc3e2cbf.udfu
-rwxrwxr-x 1 agreen agreen 22736 2008-10-17 18:25 qi-s3c6410-andy_77c1fcdddc3e2cbf
-rw-rw-r-- 1 agreen agreen 216294 2008-10-17 18:25 qi-s3c6410-andy_77c1fcdddc3e2cbf.dis
-rw-rw-r-- 1 agreen agreen 22752 2008-10-17 18:25 qi-s3c6410-andy_77c1fcdddc3e2cbf.udfu
The 6410 support in there is enough to send a character "U" on the 6410 SMDK
Because the product naming is not defined, currently the device targeted for 6410
is called "TLA01"
Signed-off-by: Andy Green <andy@openmoko.com>
Before leaving for Linux, gratuitously clear down the totrst / timeout counter
to help stop us dying partway through boot on effective power-off.
In the case we have no battery or battery < ~3V, we still somehow need to do
more in kernel because it can switch itself off more than 8 seconds
after this point.
Signed-off-by: Andy Green <andy@openmoko.com>
Part of the effort to save power during booting so we can do it inside the
500mW budget from un-enumerated USB connection.
Signed-off-by: Andy Green <andy@openmoko.com>
On the last read, master must NAK, on intermediate reads it must ACK.
If you get this wrong, communication with device is screwed even in Linux.
Signed-off-by: Andy Green <andy@openmoko.com>
We need to set a few more things up in pcf50633 to get suspend and resume
to work -- without them suspend actually goes OFF. Take the opportunity
to optimize this init significantly.
Signed-off-by: Andy Green <andy@openmoko.com>
With large SDHC cards, the ext2/3 filesystem put on large partitions
by mkfs.ext2/3 is not how it used to be, it has a dynamic multiplier for
its block addressing. Without these changes, based on a patch by Ryan
Chen
http://lists.denx.de/pipermail/u-boot/2008-July/037786.html
we cannot parse large ext3 filesystem.
Signed-off-by: Andy Green <andy@openmoko.com>
Now cards larger than 4GB are normal, we can't use the byte addressing
internally any more for SDHC type card. This changes us to use block
(512 byte) addressing internally always.
Signed-off-by: Andy Green <andy@openmoko.com>
This patch performs the equivalent init actions to the SDHC slow
unitl first bulk patch in kernel, it allows us to work with large
SDHC cards which exceed Glamo timeout capability at 16MHz for first
access.
Signed-off-by: Andy Green <andy@openmoko.com>
This patch increases the timeout for card initializaton so it works with
8GB Sandisk uSDHC, and adds the size computation for SDHC as well.
Signed-off-by: Andy Green <andy@openmoko.com>
This is the script for gta03 load by openocd. We use the single qi
image for all purposes now so U-Boot lowlevel_foo style crap is dead.
We have to run the first part of init before copying qi image to
TEXT_BASE (0x33000000) because DRAM is not functional until then.
Fixed branchthrough at 0x8 is used as a place to hang the breakpoint.
Signed-off-by: Andy Green <andy@openmoko.com>
This adds capability to use single qi image for JTAG load action as well
as execution from NAND. It requires JTAG script to load the image at
0x0 and 0x33000000 addresses, then set data at 0x04 address to 0xffffffff.
This eliminates the lowlevel_foo stuff from U-Boot world.
Signed-off-by: Andy Green <andy@openmoko.com>
This adds ext2 support from U-Boot and stitches it into the
partition stuff. It also upgrades the board definitions so they
can define the path to look for in the ext2 filesystem being
mounted. I used /boot/uImage.bin because this is already in use
by the packaged kernel.
We now mount, open and pull the kernel from ext2 in phase2.c if
the kernel source defines it.
Signed-off-by: Andy Green <andy@openmoko.com>
We have to minimally support DOS partition scheme.
Partitions are counted from 1+ now, and 0 means no
partition table instead of -1.
Signed-off-by: Andy Green <andy@openmoko.com>
We leave a lot of funny clocks up in things like camera unit. This
pares it down to just what we use in Qi.
Signed-off-by: Andy Green <andy@openmoko.com>
Moves various .h into include from drivers
Gets glamo-mmc.h working so we can detect
card ID / Size if the card is plugged in.
Adds FAT / MMC kernel source entry for GTA02 that
is first before the NAND one. When it works it will
favour to boot off SD Card if a kernel is found there, but
right now we don't have a working filesystem parser and
partition handling hooked up.
Signed-off-by: Andy Green <andy@openmoko.com>
Move serial.c into drivers/serial-s3c24xx.c and qi-serial.h into include/serial-s3c24xx.h
making things a bit cleaner for being s3c24xx-specific. This needed a lot of meddling
additionally, ending up with a new puts() callback that belongs in the board structure
and removal of the debug uart member, since the puts() action was the only user.
Also change serial init API name and function to only fix to 115kbps and adapt to
PCLK.
Signed-off-by: Andy Green <andy@openmoko.com>
Introduce generic bitbang I2C system, and a s3c24xx-specific implementation of
it that knows which GPIO pins and how to control them.
The generic bitbang stuff exposes synchronous (ie, it will return when it is
done) and asynchronous read and write APIs, allowing the delay between bits to
be hidden in other slow, looping code if necessary.
Signed-off-by: Andy Green <andy@openmoko.com>
Create ./src/drivers and move s3c24xx mmc thing in there with more
specific name. Move fat.h into ./include
Signed-off-by: Andy Green <andy@openmoko.com>
Giant patch:
- renames everything from kboot to qi
- changes filenames accordingly in several places
- fixes the linker script so stuff that does not execute
from steppingstone context has real linked addresses
in the relocated region, it means all code and pointers
work now outside first 4KBytes
- adds src/gta02/gta02.c to contain board-specific init and
other functions
- adds sophisticated structs to define most features in the
board-specific files, including board type detection,
board revision detection, and multiple kernel source
definition (NAND, SD FAT, SD ext2, etc), including auto
sequencing of trying the kernel sources in order (filesystems
and partition support not done yet)
- GTA02 detects itself by NOR presence and reports A5 / A6
- commandlines for kernel also come from board-specific
kernel source definitions so correct kernel commandlines
are provided depending on boot device -- on GTA02 now
boots NAND kernel into NAND jffs2 filesystem
- CRC32 is checked on loaded kernel image to make sure we
know about corruption in bootloader
Signed-off-by: Andy Green <andy@openmoko.com>
Huge patch boils down massive kernel image parsing and boot action
to a modest sequence of code that actually boots the kernel.
Signed-off-by: Andy Green <andy@openmoko.com>
NAND stuff wasn't going to do anything until the controller in
the CPU was reset. NAND code was cleaned and other minor meddlings
Signed-off-by: Andy Green <andy@openmoko.com>
Start meddling with linkser script, more changes came later
that actually got it working by pushing all .rodata into first
4K so the linked addresses matched reality even when ran from
the SDRAM copy.
Signed-off-by: Andy Green <andy@openmoko.com>
Add various debugging code for serial, this got changed around a
lot in subsequent patches and printk() / vsprintf() was taken out
in the end.
Signed-off-by: Andy Green <andy@openmoko.com>
