Hi,
I'm doing some tricks with Linux kernel and I'm dependent on bootloader atags passing.
On PandaBoard the u-boot is always passing DTF/Atags pointer in r2 register, and I'm kind of depending on it. I need to emulate this behavior by -initrd qemu's argument. However it seems that qemu-system-arm zeros the registers on the start, and places atags in some hardcoded memory address no matter what.
Is this behaviour really OK? Isn't r2 register a proper way to pass atags on ARM platform? If so, would you merge into mainline a patch that implement this? I could write it, but I don't want end-users of my work to have to patch and compile Qemu manually.
Please CC me, I'm not following actively these mailinglists.
Regards,
On 6 February 2012 12:19, dpc@ucore.info dpc@ucore.info wrote:
On PandaBoard the u-boot is always passing DTF/Atags pointer in r2 register, and I'm kind of depending on it. I need to emulate this behavior by -initrd qemu's argument. However it seems that qemu-system-arm zeros the registers on the start, and places atags in some hardcoded memory address no matter what.
No, qemu will correctly pass the atags in r2 if it is booting a linux kernel (ie if you pass it a non ELF file via the -kernel argument). Otherwise we wouldn't be able to boot Linux. How are you starting qemu ?
-- PMM
On Mon, Feb 6, 2012 at 5:35 PM, Peter Maydell peter.maydell@linaro.org wrote:
No, qemu will correctly pass the atags in r2 if it is booting a linux kernel (ie if you pass it a non ELF file via the -kernel argument). Otherwise we wouldn't be able to boot Linux. How are you starting qemu ?
Well it's a bit complex:
sudo "$QEMU" -daemonize -no-quit -pidfile "$PIDFILE" $* \ -serial telnet:localhost:${SERIAL_PORT_BASE},server,nowait,ipv4 \ -serial telnet:localhost:$(($SERIAL_PORT_BASE + 1)),server,nowait,ipv4 \ -serial telnet:localhost:$(($SERIAL_PORT_BASE + 2)),server,nowait,ipv4 \ -serial telnet:localhost:$(($SERIAL_PORT_BASE + 3)),server,nowait,ipv4 \ $QEMU_LOCAL_ARCH_ARGS \ -net nic -net tap,script="$ROOT/tools/qemu-ifup,downscript=$ROOT/tools/qemu-ifdown" \ -gdb tcp:127.0.0.1:${GDB_PORT},server,ipv4 \ -monitor telnet:127.0.0.1:${MONITOR_PORT},server,ipv4,nowait \ -kernel "$FINAL_IMAGE" \ $QEMU_ARCH_ARGS -S || { if [ -f "$PIDFILE" ]; then sudo rm -f "$PIDFILE" fi exit 1 }
FINAL_IMAGE is elf or uImage (tried both), the uImage is supposed to work. -initrd along with -sd and some architecture specific arguments are passed in QEMU_ARCH_ARGS
When I'm attaching gdb and doing "info registers" I get 0x0 in all registers except PC.
I've analyzed the source, added some printfs and the image is correctly detected as Linux.
In the sourcode (current release) I couldn't find any place where any register except PC is being modified.
I've found some comments about some Linux header file that no longer exists and the code suggest that atags data is put into some pre-defined memory location, with no pointer being passed in r2.
On Mon, Feb 6, 2012 at 6:48 PM, dpc@ucore.info dpc@ucore.info wrote:
In the sourcode (current release) I couldn't find any place where any register except PC is being modified.
While looking for some data to paste here, I've notice that the pointers should be set by "the smallest bootlader in the world". I wasn't expecting it and checked status of the registers right after emulation start.
I now see that registers are OK if I step 4 times in gdb.
Please disregard my email.
Regards,
linaro-toolchain@lists.linaro.org
-
dpc＠ucore.info -
Peter Maydell