u-boot chain-loading LineageOS bootimg
Subsystems (2)
cover u-boot chain-loading LineageOS bootimg
This is a series of patches to enable chainloading LineageOS on qcom SOC.
First patch is to workaround kernel/ramdisk invalid addr by identify its physical memory address out-of-range. Since qcom SOC usually have 0x80000000 as start/base/real memory address but androidboot img specified to around 0x0. If other vendor bootloader behave similar then this patch can also workaround it as well.
Second patch is enable bootmeth-android to have chance for extra mem block. Usually the fastboot mem block, to hold androidboot img, and loadaddr for unzipped kernel. If other SOC have extra mem block available but fastboot block, we can add extra logic to take care of it.
Third patch is optional to enable snapdragon board to have extra cmdline found from original bootloader that is required for LineageOS boot init. An alternate method is to put the append string into a dummy device-tree file, as /chosen/bootarg ofnode porperty, that also contain msm-id. Then encapsulate as androidboot img and let u-boot as kernel binary. Below is an example for Xiaomi Miatoll device.
/dts-v1/;
/ {
qcom,msm-id = <443 0x0>;
qcom,board-id = <0 0>,
<0x10022 1>,
<0x20022 1>,
<0x30022 1>,
<0x40022 1>,
<0x50022 1>;
#address-cells = <2>;
#size-cells = <2>;
memory {
ddr_device_type = <0x07>;
/* We expect the bootloader to fill in the size */
reg = <0 0 0 0>;
};
chosen {
bootargs = "";
};
};---
George Chan (3):
boot/image-android.c: Workaround androidboot kernel/ramdisk addr boot/bootmeth-android.c: Reuse fastboot memory block for unzip kernel
mach-snapdragon: Add support to append string to kernel cmdline
arch/arm/mach-snapdragon/Kconfig | 11 +++++ arch/arm/mach-snapdragon/board.c | 97 ++++++++++++++++++++++++++++++++++++++++ boot/bootmeth_android.c | 12 +++++ boot/image-android.c | 10 +++++ 4 files changed, 130 insertions(+)
---
base-commit: 5a0a93a768487e55ebe50a34cc90d751bf99cc56
change-id: 20250427-android-boot-ecbb768cda72
Best regards,
Signature · 1 line
George Chan <[email protected]>
PATCH 1/3 boot/image-android.c: Workaround androidboot kernel/ramdisk addr
Some vendor bootloader ignored kernel/ramdisk addr and use their own addr. Even those addr are pointing to out-of-reach memory block at 0, and available address start at 0x80000000.
So we also need to use our own available addr for loadaddr.
--- boot/image-android.c | 10 ++++++++++ 1 file changed, 10 insertions(+)
boot/image-android.c +10 -0
@@ -12,6 +12,7 @@#include <asm/unaligned.h> #include <mapmem.h> #include <linux/libfdt.h> +#include <dm/of.h> #define ANDROID_IMAGE_DEFAULT_KERNEL_ADDR 0x10008000#define ANDROID_IMAGE_DEFAULT_RAMDISK_ADDR 0x11000000
@@ -281,6 +282,13 @@ static ulong android_image_get_kernel_addr(struct andr_image_data *img_data,if (img_data->kernel_addr ==0 && img_data->ramdisk_addr ==0 )return env_get_ulong ("kernel_addr_r" ,16 ,0 ); +/* +* Some vendor bootloader ignore kernel and ramdisk address also +* default value are out of physical mem base... so we use ours. +*/ +if (img_data->kernel_addr < gd->ram_base) +return env_get_ulong ("kernel_addr_r" ,16 ,0 ); +return img_data->kernel_addr; }
@@ -490,6 +498,8 @@ int android_image_get_ramdisk(const void *hdr, const void *vendor_boot_img,if (img_data.ramdisk_addr ==0 || img_data.ramdisk_addr ==ANDROID_IMAGE_DEFAULT_RAMDISK_ADDR ) { *rd_data = img_data.ramdisk_ptr; + }else if (img_data.ramdisk_addr < gd->ram_base) { + *rd_data = img_data.ramdisk_ptr; }else { ramdisk_ptr = img_data.ramdisk_addr; *rd_data = ramdisk_ptr;
Signature · 1 line
2.43.0
PATCH 2/3 boot/bootmeth-android.c: Reuse fastboot memory block for unzip kernel
In reply to George ChanSome androidboot images have gzipped kernel so we need to reuse extra mem block for holding gzipped boot.img, and let loadaddr to hold unzipped kernel data. Here we choose fastboot memory for reuse and avoid dramatically increase memory footprint.
--- boot/bootmeth_android.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
boot/bootmeth_android.c +12 -0
@@ -518,6 +518,18 @@ static int boot_android_normal(struct bootflow *bflow)int ret;ulong loadaddr =env_get_hex ("loadaddr" ,0 );ulong vloadaddr =env_get_hex ("vendor_boot_comp_addr_r" ,0 ); +ulong fastboot_addr_r =env_get_hex ("fastboot_addr_r" ,0 ); + +/* +* Since there are some soc with very constrain mem footprint +* so dont borther to allocate extra block for androidboot img +* instead, reuse fastboot mem block for androidboot img +* +* The idea is to let loadaddr as source boot.img for unzip, +* and the original loadaddr is for holding unzipped kernel. +*/ +if (fastboot_addr_r) + loadaddr = fastboot_addr_r; ret =run_avb_verification (bflow);if (ret <0 )
Signature · 1 line
2.43.0
PATCH 3/3 mach-snapdragon: Add support to append string to kernel cmdline
In reply to George ChanAdd support for blindly appending string to bootargs env_param and let boot process take care of it.
--- arch/arm/mach-snapdragon/Kconfig | 11 +++++ arch/arm/mach-snapdragon/board.c | 97 ++++++++++++++++++++++++++++++++++++++++ 2 files changed, 108 insertions(+)
arch/arm/mach-snapdragon/Kconfig +11 -0
@@ -45,4 +45,15 @@ config SYS_CONFIG_NAME Based on this option include/configs/<CONFIG_SYS_CONFIG_NAME>.h header will be used for board configuration. +config SYS_BOARD_CMDLINE_APPEND + bool "Snapdragon SoCs based board cmdline append string" + help + Allows to specify the Snapdragon SoCs based board kernel cmdline override. + will be used as the custom board bootloader cmdline booting OS like Android. + +config SYS_BOARD_CMDLINE_APPEND_STRING + string "String to append" + default "" + depends on SYS_BOARD_CMDLINE_APPEND + endif
arch/arm/mach-snapdragon/board.c +97 -0
103 lines not on this page · raw message
Signature · 1 line
2.43.0
Hi George,
Thanks a lot for the series, it's super exciting to see support for booting Android on top of U-Boot :D
Quoted 7 lines · On 4/27/25 13:25, George Chan via B4 Relay wrote:
This is a series of patches to enable chainloading LineageOS on qcom SOC.
First patch is to workaround kernel/ramdisk invalid addr by identify its physical memory address out-of-range. Since qcom SOC usually have
0x80000000as start/base/real memory address but androidboot img specified to around0x0. If other vendor bootloader behave similar then this patch can also workaround it as well.
I'm curious to know what values are there, tbh I think U-Boot should entirely ignore the values in the boot image but I guess that could break some other platforms.
How about adding a config option CONFIG_ANDROID_BOOTIMG_IGNORE_ADDRS and just ORing it in all the places that have checks like "img_data->kernel_addr == 0". Then you can enable this for mach-snapdragon by selecting it in the ARCH_SNAPDRAGON config definition.
Quoted 5 lines
Second patch is enable bootmeth-android to have chance for extra mem block. Usually the fastboot mem block, to hold androidboot img, and loadaddr for unzipped kernel. If other SOC have extra mem block available but fastboot block, we can add extra logic to take care of it.
There is a small hack in mach-snapdragon that we put kernel_addr_r and loadaddr at the same address to avoid using up too much memory on devices that only have 1GB which explains this crash because we try to decompress the kernel to the same address it got loaded too.
I think we can get away with giving loadaddr its own allocation of 64M or so, maybe 96. This would make your second patch redundant (and feels like the right fix to me)
Quoted 7 lines
Third patch is optional to enable snapdragon board to have extra cmdline found from original bootloader that is required for LineageOS boot init. An alternate method is to put the append string into a dummy device-tree file, as /chosen/bootarg ofnode porperty, that also contain msm-id. Then encapsulate as androidboot img and let u-boot as kernel binary. Below is an example for Xiaomi Miatoll device.
I assume you're chainloading U-Boot on this device? If so, isn't it enough to just copy the boot args that ABL gives us and maybe tweak a few?
I'd also rather keep this constrained, maybe an android specific board callback to set boot args (if there isn't one already) and have the FDT fixup code be alongside some other generic FDT fixup stuff. I'd rather not have this stuff in board specific code.
I hope this makes sense, feel free to ask for clarifications on anything.
Thanks and kind regards,
Quoted 43 lines
/dts-v1/;
/ { qcom,msm-id = <443 0x0>; qcom,board-id = <0 0>, <0x10022 1>, <0x20022 1>, <0x30022 1>, <0x40022 1>, <0x50022 1>; #address-cells = <2>; #size-cells = <2>; memory { ddr_device_type = <0x07>; /* We expect the bootloader to fill in the size */ reg = <0 0 0 0>; }; chosen { bootargs = ""; }; };Signed-off-by George Chan <[email protected]>---
George Chan (3):
boot/image-android.c: Workaround androidboot kernel/ramdisk addrboot/bootmeth-android.c: Reuse fastboot memory block for unzip kernel
mach-snapdragon: Add support to append string to kernel cmdlinearch/arm/mach-snapdragon/Kconfig | 11 +++++ arch/arm/mach-snapdragon/board.c | 97 ++++++++++++++++++++++++++++++++++++++++ boot/bootmeth_android.c | 12 +++++ boot/image-android.c | 10 +++++ 4 files changed, 130 insertions(+)---
base-commit:5a0a93a768487e55ebe50a34cc90d751bf99cc56
change-id: 20250427-android-boot-ecbb768cda72Best regards,
Signature · 1 line
Casey (she/they)
Hi Casey,
Thx for reply. Sorry for a bit long message for explain the problem.
在 2025年4月28日週一 21:53,Casey Connolly <[email protected]> 寫道:
Quoted 8 lines
Hi George,
Thanks a lot for the series, it's super exciting to see support for booting Android on top of U-Boot :D
Quoted 2 lines · On 4/27/25 13:25, George Chan via B4 Relay wrote:
This is a series of patches to enable chainloading LineageOS on qcom SOC.
Quoted 7 lines
Quoted 5 lines
First patch is to workaround kernel/ramdisk invalid addr by identify its physical memory address out-of-range. Since qcom SOC usually have
0x80000000as start/base/real memory address but androidboot img specified to around0x0. If other vendor bootloader behave similar then this patch can also workaround it as well.I'm curious to know what values are there, tbh I think U-Boot should
I don't know either. I figured it out a failure memcpy kernel to addr 0, and failure at right after first 16k succes. Since kernel is relocatable so I ignored it. I know you are dealing with memory mapping on xbl and chain loading case with some kind of smem value. Good job!
Quoted 2 lines
entirely ignore the values in the boot image but I guess that could break some other platforms.
I agreed your below suggested alternative approach apply can ease for maintaining, and less concern. I am willing to help draft a patch worked as your mentioned approach.
Maybe it is more optimistic than what you see here. Below are some reasons might be worth to rethink.
(1) This is specific to bootm dealing androidboot img only. (2)A similar approach applied to check specific kernel address already. (3)The proposed logic is to cater extra case that is never works. (4)unless there is a need for soc that mmap() high (physical) memory addr to low (nonexistent) addr, aka 0, and use low address as kernel data. Sounds like hyperizer iommu work. Looks redundance.
(5)if checking gd->base_mem is not safe then we can add extra checking for gd null. The logic only affect valid gd and base_mem value. Either null or 0 won't have any side effect.
So it less likely breaks. In the end, if apply your suggested approach, then mine and existing one are redundance and could be removed for clean up. I am not sure what happened and concluded for existing code but the code exists. Maybe some senior contributor knows more about it.
Removing code would be a big behavioral change require broader acknowledgement. Or just leave existing logic and apply only your suggested approach anyway. Both are fine for me. I prefer to wait for some feedback before new revision roll.
Quoted 5 lines
How about adding a config option
CONFIG_ANDROID_BOOTIMG_IGNORE_ADDRSand just ORing it in all the places that have checks like "img_data->kernel_addr== 0". Then you can enable this for mach-snapdragon by selecting it in theARCH_SNAPDRAGONconfig definition.
As stated as above.
Quoted 12 lines
Quoted 2 lines
Second patch is enable bootmeth-android to have chance for extra mem
block.
Quoted 3 lines
Usually the fastboot mem block, to hold androidboot img, and loadaddr for unzipped kernel. If other SOC have extra mem block available but fastboot block, we can add extra logic to take care of it.
There is a small hack in mach-snapdragon that we put
kernel_addr_rand loadaddr at the same address to avoid using up too much memory on devices that only have 1GB which explains this crash because we try to decompress the kernel to the same address it got loaded too.
Reasonable.
I think we can get away with giving loadaddr its own allocation of 64M
Quoted 3 lines
or so, maybe 96. This would make your second patch redundant (and feels like the right fix to me)
I don't know how memory constrain affected, and have virtually no knowledge on memory arrangement. So I can't comment on this.
There is a third way. Add one more env_var like androidboot_addr_r and pointed to same as fastboot_addr_r then use it in bootmeth-android if this new env exists. If var not exists, fallback to existing logic. That balance with minimal logic change and readability.
Quoted 11 lines
Quoted 7 lines
Third patch is optional to enable snapdragon board to have extra cmdline found from original bootloader that is required for LineageOS boot init. An alternate method is to put the append string into a dummy device-tree file, as /chosen/bootarg ofnode porperty, that also contain msm-id. Then encapsulate as androidboot img and let u-boot as kernel binary. Below is an example for Xiaomi Miatoll device.
I assume you're chainloading U-Boot on this device? If so, isn't it enough to just copy the boot args that ABL gives us and maybe tweak a few?
Yes, I am doing chain loading u-boot. And yes, patch 3 is doing exactly you mentioned.
I mentioned here is 2nd approach that I feels cleaner for chain loading case. I really have no idea on how prev_blob and target_blob works all the way fdt handling (difference of replace xbl and chain loading). If non-chain-loading case there have no prev_blob, then 1st approach is preferable.
Quoted 5 lines
I'd also rather keep this constrained, maybe an android specific board callback to set boot args (if there isn't one already) and have the FDT fixup code be alongside some other generic FDT fixup stuff. I'd rather not have this stuff in board specific code.
Reasonably. I can't figure out the difference board specific fixup and android specific fixup. My limited knowledge is fdt_support.c have weak board_fdt_chosen_bootargs func to board.c as my 1st approach do as patch 3. I wonder if you are talking about adding one more fdt-support.c weak func like board_fdt_chosen_bootargs_android. And leave board_fdt_chosen_bootargs empty for further use case on purpose. I also curious if you are targeting to make it generic enough for mach-* and/or enhanced API with full fdt for fixup because the existing one is incapable. And this is reasonable too.
I wanna to express some though about additional fixups here. As you knows those androidboot param are half baked and soc/machine/device/os dependent. So it can't be one-sized-fit-all. Currently we have some approaches to put it in.
(1) In upstream dts, block the merge.
(2) In a dummy dts, might not work for xbl case.
(3) In board.c/Kconfig compile-time config. Not flexible. Can't do it with constrained way unless a string parser is needed (I wrote one similar thing but found useless
https://github.com/99degree/u-boot/commit/3240054aa85f4cb70daa8671150bf4a9ca885a61
)
Since all added Androidboot param are android specific, my approach is very simple. Let the android init to get rid of it.
Here is how to do so. Android Init treat Androidboot.* param as ro.boot.* and set-once-read-only and first-come-first-serve so any taylor-made param we put in the head will mandate the value even the same appears in the tail. In other words, params in tail are failsafe as result. This behavior is good enough to make an approach to balance flexible and function. We can have wanted param in chainloaded Androidboot (boot.) img dtb, and 'default' param live in board specific fixup logic. Since env_bootargs have param from Android boot img and board can do fixup by append android specific param to end of env_bootargs. Then we do not need to have constrained param (to append env_bootargs, my own understanding) unless overflow bootargs length. Well, this is the core idea of my impl in patch 3. Also patch 3 have Kconfig string to override SOC specific fixup so param value can set by config file.
One more thing worth mentioning, I made boot.img repacked with dtb from /sys/firmware/fdt and do u-boot chain load because of well known dtbo issue. Maybe u-boot support dtbo partition but I have not tried. I cant assumed people get used to familiar with android param so a dummy way is do less-is-more. It is easy to repack boot.img with new fdt (by tool like e.g. magisk) but not easy to figure which param to append bootargs is needed. Even duplicated param as result.
Quoted 2 lines
I hope this makes sense, feel free to ask for clarifications on anything.
Yes your suggestions are valuable
Quoted 2 lines
Thanks and kind regards,
kind regards,
George
Quoted 49 lines
Quoted 26 lines
/dts-v1/;
/ { qcom,msm-id = <443 0x0>; qcom,board-id = <0 0>, <0x10022 1>, <0x20022 1>, <0x30022 1>, <0x40022 1>, <0x50022 1>; #address-cells = <2>; #size-cells = <2>; memory { ddr_device_type = <0x07>; /* We expect the bootloader to fill in the size */ reg = <0 0 0 0>; }; chosen { bootargs = ""; }; };Quoted 5 lines
Signed-off-by George Chan <[email protected]>---
George Chan (3):
boot/image-android.c: Workaround androidboot kernel/ramdisk addrboot/bootmeth-android.c: Reuse fastboot memory block for unzipkernel
Quoted 4 lines
mach-snapdragon: Add support to append string to kernel cmdline
arch/arm/mach-snapdragon/Kconfig | 11 +++++ arch/arm/mach-snapdragon/board.c | 97++++++++++++++++++++++++++++++++++++++++
Quoted 8 lines
boot/bootmeth_android.c | 12 +++++ boot/image-android.c | 10 +++++ 4 files changed, 130 insertions(+)---
base-commit:5a0a93a768487e55ebe50a34cc90d751bf99cc56
change-id: 20250427-android-boot-ecbb768cda72Best regards,
--
Casey (she/they)
Hi George,
Thank you for contributing.
Quoted 14 lines · Casey Connolly
Hi George,
Thanks a lot for the series, it's super exciting to see support for booting Android on top of U-Boot :D
Quoted 7 lines · On 4/27/25 13:25, George Chan via B4 Relay wrote:
This is a series of patches to enable chainloading LineageOS on qcom SOC.
First patch is to workaround kernel/ramdisk invalid addr by identify its physical memory address out-of-range. Since qcom SOC usually have
0x80000000as start/base/real memory address but androidboot img specified to around0x0. If other vendor bootloader behave similar then this patch can also workaround it as well.
Quoted 3 lines
I'm curious to know what values are there, tbh I think U-Boot should entirely ignore the values in the boot image but I guess that could break some other platforms.
Yes, some platforms actually use the values from the boot.img. We had some breakage on Khadas VIM3 in this area. See:
https://lore.kernel.org/all/[email protected]/
The above thread also mentions how to repack a boot.img, where we can specify a different ramdisk address.
It's a bit unclear to me why it's impractical to repack the boot.img and specify the appropriate address. Could you elaborate ?
Quoted 5 lines
How about adding a config option
CONFIG_ANDROID_BOOTIMG_IGNORE_ADDRSand just ORing it in all the places that have checks like "img_data->kernel_addr== 0". Then you can enable this for mach-snapdragon by selecting it in theARCH_SNAPDRAGONconfig definition.
If we need to add this in U-Boot, I'd prefer this option as suggested by Casey.
Quoted 7 lines
Quoted 6 lines
Second patch is enable bootmeth-android to have chance for extra mem block. Usually the fastboot mem block, to hold androidboot img, and loadaddr for unzipped kernel. If other SOC have extra mem block available but fastboot block, we can add extra logic to take care of it.
Quoted 8 lines
There is a small hack in mach-snapdragon that we put
kernel_addr_rand loadaddr at the same address to avoid using up too much memory on devices that only have 1GB which explains this crash because we try to decompress the kernel to the same address it got loaded too.I think we can get away with giving loadaddr its own allocation of 64M or so, maybe 96. This would make your second patch redundant (and feels like the right fix to me)
I was not aware of this hack, but I agree with Casey.
Quoted 9 lines
Quoted 8 lines
Third patch is optional to enable snapdragon board to have extra cmdline found from original bootloader that is required for LineageOS boot init. An alternate method is to put the append string into a dummy device-tree file, as /chosen/bootarg ofnode porperty, that also contain msm-id. Then encapsulate as androidboot img and let u-boot as kernel binary. Below is an example for Xiaomi Miatoll device.
Quoted 7 lines
I assume you're chainloading U-Boot on this device? If so, isn't it enough to just copy the boot args that ABL gives us and maybe tweak a few?
I'd also rather keep this constrained, maybe an android specific board callback to set boot args (if there isn't one already) and have the FDT fixup code be alongside some other generic FDT fixup stuff. I'd rather not have this stuff in board specific code.
I also agree with Casey here. Looking at patch: "[PATCH 3/3]
mach-snapdragon: Add support to append string to kernel cmdline", the android bootmethod already takes care of some androidboot.* arguments. See:
https://source.denx.de/u-boot/u-boot/-/blob/master/boot/bootmeth_android.c?ref_type=heads#L504
For androidboot.verifiedbootstate=orange, this is done via avb_append_commandline_arg() when AVB is enabled in U-Boot.
Quoted 50 lines
I hope this makes sense, feel free to ask for clarifications on anything.
Thanks and kind regards,
Quoted 43 lines
/dts-v1/;
/ { qcom,msm-id = <443 0x0>; qcom,board-id = <0 0>, <0x10022 1>, <0x20022 1>, <0x30022 1>, <0x40022 1>, <0x50022 1>; #address-cells = <2>; #size-cells = <2>; memory { ddr_device_type = <0x07>; /* We expect the bootloader to fill in the size */ reg = <0 0 0 0>; }; chosen { bootargs = ""; }; };Signed-off-by George Chan <[email protected]>---
George Chan (3):
boot/image-android.c: Workaround androidboot kernel/ramdisk addrboot/bootmeth-android.c: Reuse fastboot memory block for unzip kernel
mach-snapdragon: Add support to append string to kernel cmdlinearch/arm/mach-snapdragon/Kconfig | 11 +++++ arch/arm/mach-snapdragon/board.c | 97 ++++++++++++++++++++++++++++++++++++++++ boot/bootmeth_android.c | 12 +++++ boot/image-android.c | 10 +++++ 4 files changed, 130 insertions(+)---
base-commit:5a0a93a768487e55ebe50a34cc90d751bf99cc56
change-id: 20250427-android-boot-ecbb768cda72Best regards,
Signature · 1 line
Casey (she/they)
Hi Mattijs,
CC: Casey,
Thx for your reply.
在 2025年4月29日週二 16:30,Mattijs Korpershoek <[email protected]> 寫道:
Quoted 29 lines
Hi George,
Thank you for contributing.
Quoted 7 lines · Casey Connolly
Hi George,
Thanks a lot for the series, it's super exciting to see support for booting Android on top of U-Boot :D
Quoted 1 line · On 4/27/25 13:25, George Chan via B4 Relay wrote:
This is a series of patches to enable chainloading LineageOS on qcom
SOC.
Quoted 10 lines
Quoted 6 lines
First patch is to workaround kernel/ramdisk invalid addr by identify its physical memory address out-of-range. Since qcom SOC usually have
0x80000000as start/base/real memory address but androidboot img specified to around0x0. If other vendor bootloader behave similar then this patch can also workaround it as well.I'm curious to know what values are there, tbh I think U-Boot should entirely ignore the values in the boot image but I guess that could break some other platforms.
Yes, some platforms actually use the values from the boot.img. We had some breakage on Khadas VIM3 in this area. See:
Quoted 7 lines
https://lore.kernel.org/all/[email protected]/
The above thread also mentions how to repack a boot.img, where we can specify a different ramdisk address.
It's a bit unclear to me why it's impractical to repack the boot.img and specify the appropriate address. Could you elaborate ?
It is hard to ask less experience people to modify the boot img....I do prefer no mod is the best. Why get life so complicated, my friend?
Quoted 9 lines
Quoted 5 lines
How about adding a config option
CONFIG_ANDROID_BOOTIMG_IGNORE_ADDRSand just ORing it in all the places that have checks like "img_data->kernel_addr== 0". Then you can enable this for mach-snapdragon by selecting it in theARCH_SNAPDRAGONconfig definition.If we need to add this in U-Boot, I'd prefer this option as suggested by Casey.
No problem. If this approach have wider audience and also benefit other soc/device, I am glad to know. Lets wait for user of below code and see if any willingness to migrate.
And
Quoted 14 lines
Quoted 3 lines
Quoted 3 lines
Second patch is enable bootmeth-android to have chance for extra mem
block.
Quoted 1 line
Quoted 1 line
Usually the fastboot mem block, to hold androidboot img, and loadaddr
for
Quoted 8 lines
Quoted 2 lines
unzipped kernel. If other SOC have extra mem block available but fastboot block, we can add extra logic to take care of it.
There is a small hack in mach-snapdragon that we put
kernel_addr_rand loadaddr at the same address to avoid using up too much memory on devices that only have 1GB which explains this crash because we try to decompress the kernel to the same address it got loaded too.
Quoted 6 lines
Quoted 3 lines
I think we can get away with giving loadaddr its own allocation of 64M or so, maybe 96. This would make your second patch redundant (and feels like the right fix to me)
I was not aware of this hack, but I agree with Casey.
Hmm... I have no idea/experience about it. Casey will you follow this issue with a proper patch?
Quoted 24 lines
Quoted 11 lines
Quoted 8 lines
Third patch is optional to enable snapdragon board to have extra cmdline found from original bootloader that is required for LineageOS boot init. An alternate method is to put the append string into a dummy device-tree file, as /chosen/bootarg ofnode porperty, that also contain msm-id. Then encapsulate as androidboot img and let u-boot as kernel binary. Below is an example for Xiaomi Miatoll device.
I assume you're chainloading U-Boot on this device? If so, isn't it enough to just copy the boot args that ABL gives us and maybe tweak a
few?
Quoted 5 lines
I'd also rather keep this constrained, maybe an android specific board callback to set boot args (if there isn't one already) and have the FDT fixup code be alongside some other generic FDT fixup stuff. I'd rather not have this stuff in board specific code.
I also agree with Casey here. Looking at patch: "[PATCH 3/3]
mach-snapdragon: Add support to append string to kernel cmdline", the android bootmethod already takes care of some androidboot.* arguments. See:
Quoted 5 lines
https://source.denx.de/u-boot/u-boot/-/blob/master/boot/bootmeth_android.c?ref_type=heads#L504
For androidboot.verifiedbootstate=orange, this is done via
avb_append_commandline_arg()when AVB is enabled in U-Boot.
Reasonable. My plan is remove all default params. I still not decided to let kconfig do it or leave it as a new env var. And put that post-pending logic into fdt-support.c
Thx again,
George
Quoted 53 lines
Quoted 36 lines
I hope this makes sense, feel free to ask for clarifications on anything.
Thanks and kind regards,
Quoted 31 lines
/dts-v1/;
/ { qcom,msm-id = <443 0x0>; qcom,board-id = <0 0>, <0x10022 1>, <0x20022 1>, <0x30022 1>, <0x40022 1>, <0x50022 1>; #address-cells = <2>; #size-cells = <2>; memory { ddr_device_type = <0x07>; /* We expect the bootloader to fill in the size */ reg = <0 0 0 0>; }; chosen { bootargs = ""; }; };Signed-off-by George Chan <[email protected]>---
George Chan (3):
boot/image-android.c: Workaround androidboot kernel/ramdisk addrboot/bootmeth-android.c: Reuse fastboot memory block for unzipkernel
Quoted 4 lines
Quoted 4 lines
mach-snapdragon: Add support to append string to kernel cmdline
arch/arm/mach-snapdragon/Kconfig | 11 +++++ arch/arm/mach-snapdragon/board.c | 97++++++++++++++++++++++++++++++++++++++++
Quoted 10 lines
Quoted 8 lines
boot/bootmeth_android.c | 12 +++++ boot/image-android.c | 10 +++++ 4 files changed, 130 insertions(+)---
base-commit:5a0a93a768487e55ebe50a34cc90d751bf99cc56
change-id: 20250427-android-boot-ecbb768cda72Best regards,
--
Casey (she/they)
Hi George,
Quoted 42 lines · george chan
Hi Mattijs,
CC: Casey,Thx for your reply.
在 2025年4月29日週二 16:30,Mattijs Korpershoek <[email protected]> 寫道:
Quoted 34 lines
Hi George,
Thank you for contributing.
Quoted 7 lines · Casey Connolly
Hi George,
Thanks a lot for the series, it's super exciting to see support for booting Android on top of U-Boot :D
Quoted 1 line · On 4/27/25 13:25, George Chan via B4 Relay wrote:
This is a series of patches to enable chainloading LineageOS on qcom
SOC.
Quoted 10 lines
Quoted 6 lines
First patch is to workaround kernel/ramdisk invalid addr by identify its physical memory address out-of-range. Since qcom SOC usually have
0x80000000as start/base/real memory address but androidboot img specified to around0x0. If other vendor bootloader behave similar then this patch can also workaround it as well.I'm curious to know what values are there, tbh I think U-Boot should entirely ignore the values in the boot image but I guess that could break some other platforms.
Yes, some platforms actually use the values from the boot.img. We had some breakage on Khadas VIM3 in this area. See:
https://lore.kernel.org/all/[email protected]/
The above thread also mentions how to repack a boot.img, where we can specify a different ramdisk address.
Quoted 6 lines
Quoted 2 lines
It's a bit unclear to me why it's impractical to repack the boot.img and specify the appropriate address. Could you elaborate ?
It is hard to ask less experience people to modify the boot img....I do prefer no mod is the best. Why get life so complicated, my friend?
Sorry, In my opinion this is not a valid argument. Why is patching the bootloader more complicated than repacking the boot.img ?
Can you maybe illustrate with a detailed example?
The offsets can also be configured via the Android build system directly by modifying BOARD_MKBOOTIMG_ARGS, see:
https://cs.android.com/android/platform/superproject/main/+/main:device/amlogic/yukawa/BoardConfig.mk;l=114?q=BOARD_MKBOOTIMG_ARGS&start=1
https://cs.android.com/android/platform/superproject/main/+/main:device/linaro/dragonboard/linaro_swr/BoardConfig.mk;l=17?q=BOARD_MKBOOTIMG_ARGS&start=21
Quoted 139 lines
Quoted 11 lines
Quoted 7 lines
How about adding a config option
CONFIG_ANDROID_BOOTIMG_IGNORE_ADDRSand just ORing it in all the places that have checks like "img_data->kernel_addr== 0". Then you can enable this for mach-snapdragon by selecting it in theARCH_SNAPDRAGONconfig definition.If we need to add this in U-Boot, I'd prefer this option as suggested by Casey.
No problem. If this approach have wider audience and also benefit other soc/device, I am glad to know. Lets wait for user of below code and see if any willingness to migrate.
And
Quoted 14 lines
Quoted 3 lines
Quoted 3 lines
Second patch is enable bootmeth-android to have chance for extra mem
block.
Quoted 1 line
Quoted 1 line
Usually the fastboot mem block, to hold androidboot img, and loadaddr
for
Quoted 8 lines
Quoted 2 lines
unzipped kernel. If other SOC have extra mem block available but fastboot block, we can add extra logic to take care of it.
There is a small hack in mach-snapdragon that we put
kernel_addr_rand loadaddr at the same address to avoid using up too much memory on devices that only have 1GB which explains this crash because we try to decompress the kernel to the same address it got loaded too.Quoted 6 lines
Quoted 3 lines
I think we can get away with giving loadaddr its own allocation of 64M or so, maybe 96. This would make your second patch redundant (and feels like the right fix to me)
I was not aware of this hack, but I agree with Casey.
Hmm... I have no idea/experience about it. Casey will you follow this issue with a proper patch?
Quoted 24 lines
Quoted 11 lines
Quoted 8 lines
Third patch is optional to enable snapdragon board to have extra cmdline found from original bootloader that is required for LineageOS boot init. An alternate method is to put the append string into a dummy device-tree file, as /chosen/bootarg ofnode porperty, that also contain msm-id. Then encapsulate as androidboot img and let u-boot as kernel binary. Below is an example for Xiaomi Miatoll device.
I assume you're chainloading U-Boot on this device? If so, isn't it enough to just copy the boot args that ABL gives us and maybe tweak a
few?
Quoted 5 lines
I'd also rather keep this constrained, maybe an android specific board callback to set boot args (if there isn't one already) and have the FDT fixup code be alongside some other generic FDT fixup stuff. I'd rather not have this stuff in board specific code.
I also agree with Casey here. Looking at patch: "[PATCH 3/3]
mach-snapdragon: Add support to append string to kernel cmdline", the android bootmethod already takes care of some androidboot.* arguments. See:Quoted 5 lines
https://source.denx.de/u-boot/u-boot/-/blob/master/boot/bootmeth_android.c?ref_type=heads#L504
For androidboot.verifiedbootstate=orange, this is done via
avb_append_commandline_arg()when AVB is enabled in U-Boot.Reasonable. My plan is remove all default params. I still not decided to let kconfig do it or leave it as a new env var. And put that post-pending logic into
fdt-support.cThx again,
GeorgeQuoted 53 lines
Quoted 36 lines
I hope this makes sense, feel free to ask for clarifications on anything.
Thanks and kind regards,
Quoted 31 lines
/dts-v1/;
/ { qcom,msm-id = <443 0x0>; qcom,board-id = <0 0>, <0x10022 1>, <0x20022 1>, <0x30022 1>, <0x40022 1>, <0x50022 1>; #address-cells = <2>; #size-cells = <2>; memory { ddr_device_type = <0x07>; /* We expect the bootloader to fill in the size */ reg = <0 0 0 0>; }; chosen { bootargs = ""; }; };Signed-off-by George Chan <[email protected]>---
George Chan (3):
boot/image-android.c: Workaround androidboot kernel/ramdisk addrboot/bootmeth-android.c: Reuse fastboot memory block for unzipkernel
Quoted 4 lines
Quoted 4 lines
mach-snapdragon: Add support to append string to kernel cmdline
arch/arm/mach-snapdragon/Kconfig | 11 +++++ arch/arm/mach-snapdragon/board.c | 97++++++++++++++++++++++++++++++++++++++++
Quoted 10 lines
Quoted 8 lines
boot/bootmeth_android.c | 12 +++++ boot/image-android.c | 10 +++++ 4 files changed, 130 insertions(+)---
base-commit:5a0a93a768487e55ebe50a34cc90d751bf99cc56
change-id: 20250427-android-boot-ecbb768cda72Best regards,
--
Casey (she/they)
Hi Mattijs,
Thx for reply. I like your attitude and appreciate it. There are many topic get silent in the end.
Quoted 10 lines
Quoted 6 lines
Quoted 2 lines
It's a bit unclear to me why it's impractical to repack the boot.img and specify the appropriate address. Could you elaborate ?
It is hard to ask less experience people to modify the boot img....I do prefer no mod is the best. Why get life so complicated, my friend?
Sorry, In my opinion this is not a valid argument. Why is patching the bootloader more complicated than repacking the boot.img ?
No, i mean repacking or modding boot.img is a very complicated thing for less experience(not me! well, lazy, in other words, thats me!) people. Sorry to get you confused.
Lets get it clear, for not going back and forth, in case of typo.
If you are talking about modding/repacking is easier than changing uboot logic. Or there is already a similar logic exists, only make a small mod for bootimg will solve the chain loading kind of thing. And then my opinion is, it might be a short term solution for engineer and as the shorter the better. I prefer to make "uboot for dummy" in mind. And it benefits those lineage os phone user (including me) who also wanted to dual boot other os. And easily update official los bootimg without any modding for every time; that's why I hate dtbo partition.
If you agreed making uboot change is easier than modding bootimg then we are same side, man.
Btw my proposed change of the loadaddr and kernel_addr_r conflict issue can work like a band-aid short term solution and easily get reverted. I forsee the proper fix is hard to get ready in short term.
Thx for letting me explain how I thought. Appreciated!
Regards,
George