Skip to contentSkip to conversation

mips: ralink: allow zboot

PATCH RESEND Superseded George Hilliard 3 participants 6 messages Mar 21, 2019 – Mar 22, 2019 MIPS

lore
Share & tools
lore
#0 lore raw

cover mips: ralink: allow zboot

Re-submitting this patch to the correct linux-mips mailing list; I just learned the linux-mips.org one is not used. Sorry for duplicates.

Do I need to do a make olddefconfig or similar for existing defconfigs? I am unsure of what the procedure is here, and I couldn't find anything in the kernel docs about what to do when you add a Kconfig option.

Thanks,
George

#1 lore raw

PATCH RESEND 1/1 mips: ralink: allow zboot

Architecturally, there's nothing preventing compressed images from working. Bootloaders built with support for the various compression methods can decompress and run the kernel. In practice, many bootloaders do not support compressed images, but kernels for those boards should just not be compressed.

Tested on an MT7688 with U-Boot doing LZMA decompression of uImage.

---
 arch/mips/Kconfig | 1 +
 1 file changed, 1 insertion(+)
arch/mips/Kconfig +1 -0
index 4a5f5b0ee9a9..b286fbbd9699 100644
@@ -619,6 +619,7 @@ config RALINK
 	select SYS_SUPPORTS_32BIT_KERNEL
 	select SYS_SUPPORTS_LITTLE_ENDIAN
 	select SYS_SUPPORTS_MIPS16
+	select SYS_SUPPORTS_ZBOOT
 	select SYS_HAS_EARLY_PRINTK
 	select CLKDEV_LOOKUP
 	select ARCH_HAS_RESET_CONTROLLER
Signature · 1 line
2.21.0
#2 lore raw

Hi,

Quoted 5 lines · George Hilliard

Architecturally, there's nothing preventing compressed images from working. Bootloaders built with support for the various compression methods can decompress and run the kernel. In practice, many bootloaders do not support compressed images, but kernels for those boards should just not be compressed.

With zboot the decompressor is included in the image so no bootloader support is needed.

A.

#3 lore raw

My version of U-Boot complains if I compile out LZMA support:
---
=> tftpboot 0x81000000 uImage; tftpboot 0x84000000 hawkeye.dtb; bootm 0x81000000 - 0x84000000
(snip)
Bytes transferred = 7349 (1cb5 hex)
## Booting kernel from Legacy Image at 81000000 ...

   Image Name:   Linux-5.1.0-rc1
   Image Type:   MIPS Linux Kernel Image (lzma compressed)
   Data Size:    1465871 Bytes = 1.4 MiB

Load Address: 80000000
Entry Point: 80344338
Verifying Checksum ... OK
## Flattened Device Tree blob at 84000000
Booting using the fdt blob at 0x84000000 Uncompressing Kernel Image ... Unimplemented compression type 3
exit not allowed from main input shell.
=>
---

Thanks,
George

Quoted 14 lines · Aaro Koskinen

Hi,

Quoted 5 lines · George Hilliard

Architecturally, there's nothing preventing compressed images from working. Bootloaders built with support for the various compression methods can decompress and run the kernel. In practice, many bootloaders do not support compressed images, but kernels for those boards should just not be compressed.

With zboot the decompressor is included in the image so no bootloader support is needed.

A.

#4 lore raw

Hi George,

Quoted 10 lines · George Hilliard

My version of U-Boot complains if I compile out LZMA support:
---
=> tftpboot 0x81000000 uImage; tftpboot 0x84000000 hawkeye.dtb; bootm 0x81000000 - 0x84000000
(snip)
Bytes transferred = 7349 (1cb5 hex)
## Booting kernel from Legacy Image at 81000000 ...

   Image Name:   Linux-5.1.0-rc1
   Image Type:   MIPS Linux Kernel Image (lzma compressed)
   Data Size:    1465871 Bytes = 1.4 MiB
Quoted 9 lines

Load Address: 80000000
Entry Point: 80344338
Verifying Checksum ... OK
## Flattened Device Tree blob at 84000000
Booting using the fdt blob at 0x84000000 Uncompressing Kernel Image ... Unimplemented compression type 3
exit not allowed from main input shell.
=>
---

There you're using a uImage though, which is different to CONFIG_SYS_SUPPORTS_ZBOOT. Even without CONFIG_SYS_SUPPORTS_ZBOOT enabled, you can build either a compressed or uncompressed uImage.

Some platforms enable one of those targets by default but it doesn't appear that ralink does, so I guess you're specifying a uImage.lzma target when building the kernel? Try doing that when CONFIG_SYS_SUPPORTS_ZBOOT isn't enabled - it'll still generate the same type of LZMA-compressed uImage.

What CONFIG_SYS_SUPPORTS_ZBOOT does is produce a program (vmlinuz) which contains a compressed version of vmlinux.bin along with some code to decompress it. The decompression is performed by vmlinuz itself, not by U-Boot.

A uImage in contrast is just the compressed (or uncompressed) vmlinux.bin with a header prepended that U-Boot uses to understand what it's loading. If the header says the data is compressed with LZMA then U-Boot will need to support LZMA decompression.

What U-Boot is complaining about there is that you gave it a uImage that's LZMA compressed & your build of U-Boot itself doesn't support LZMA. That has nothing to do with CONFIG_SYS_SUPPORTS_ZBOOT or whether the kernel binary itself supports any particular compression algorithm - it's purely down to the configuration of your copy of U-Boot.

Options would be to rebuild U-Boot with LZMA support enabled, or to use a different compression for your uImage that U-Boot already supports. eg. you could specify that you want to use gzip:

  $ make ARCH=mips uImage.gz

That would give you an image that will work so long as U-Boot supports gzip decompression, or:

  $ make ARCH=mips uImage.bin

That would give you an uncompressed uImage which U-Boot always supports.

Now it could still be valid to select CONFIG_SYS_SUPPORTS_ZBOOT if you had a need for a better compression algorithm & couldn't update your bootloader to support it, but that would be solving a different problem than your commit message claims & it wouldn't matter at all whether the bootloader supports a particular compression algorithm.

Thanks,
Paul

#5 lore raw
Quoted 25 lines · Paul Burton
Quoted 10 lines · George Hilliard

My version of U-Boot complains if I compile out LZMA support:
---
=> tftpboot 0x81000000 uImage; tftpboot 0x84000000 hawkeye.dtb; bootm 0x81000000 - 0x84000000
(snip)
Bytes transferred = 7349 (1cb5 hex)
## Booting kernel from Legacy Image at 81000000 ...

   Image Name:   Linux-5.1.0-rc1
   Image Type:   MIPS Linux Kernel Image (lzma compressed)
   Data Size:    1465871 Bytes = 1.4 MiB
Quoted 9 lines

Load Address: 80000000
Entry Point: 80344338
Verifying Checksum ... OK
## Flattened Device Tree blob at 84000000
Booting using the fdt blob at 0x84000000 Uncompressing Kernel Image ... Unimplemented compression type 3
exit not allowed from main input shell.
=>
---

There you're using a uImage though, which is different to CONFIG_SYS_SUPPORTS_ZBOOT. Even without CONFIG_SYS_SUPPORTS_ZBOOT enabled, you can build either a compressed or uncompressed uImage.

Quoted 5 lines

Some platforms enable one of those targets by default but it doesn't appear that ralink does, so I guess you're specifying a uImage.lzma target when building the kernel? Try doing that when CONFIG_SYS_SUPPORTS_ZBOOT isn't enabled - it'll still generate the same type of LZMA-compressed uImage.

OK, Buildroot was obfuscating this a little from me. I selected its LZMA compression option, and nothing happened. Asking for LZMA does not make it build uImage.lzma. Selecting the *kernel's* KERNEL_LZMA option caused Buildroot's make uImage to make the same sort of image as made by make uImage.lzma as you suggest, for a reason I have not figured out.

Quoted 4 lines

What CONFIG_SYS_SUPPORTS_ZBOOT does is produce a program (vmlinuz) which contains a compressed version of vmlinux.bin along with some code to decompress it. The decompression is performed by vmlinuz itself, not by U-Boot.

OK, that makes more sense. I have not tested zboot then :)

Quoted 5 lines

Now it could still be valid to select CONFIG_SYS_SUPPORTS_ZBOOT if you had a need for a better compression algorithm & couldn't update your bootloader to support it, but that would be solving a different problem than your commit message claims & it wouldn't matter at all whether the bootloader supports a particular compression algorithm.

Yeah, I agree. I will just make a uImage.lzma, which is what I was after anyway.

Thank you very much for the explanation!

George