linux-custom: Fix decompressed kernel name for arm64

Message ID 20210122171354.28330-1-amikan@ilbers.de
State Superseded, archived
Headers show
Series linux-custom: Fix decompressed kernel name for arm64 | expand

Commit Message

Anton Mikanovich Jan. 22, 2021, 7:13 a.m. UTC
We are performing gunzip from vmlinuz on arm64, so output should have
correct naming.

Signed-off-by: Anton Mikanovich <amikan@ilbers.de>
---
 meta/conf/bitbake.conf                                   | 1 +
 meta/recipes-kernel/linux/files/debian/isar/install.tmpl | 6 +++---
 2 files changed, 4 insertions(+), 3 deletions(-)

Comments

Jan Kiszka Jan. 25, 2021, 2:32 a.m. UTC | #1
On 22.01.21 18:13, Anton Mikanovich wrote:
> We are performing gunzip from vmlinuz on arm64, so output should have
> correct naming.
> 

What's the practical impact of this change? How did you notice it? Can
anything break that expected the old incorrect name?

Jan

> Signed-off-by: Anton Mikanovich <amikan@ilbers.de>
> ---
>  meta/conf/bitbake.conf                                   | 1 +
>  meta/recipes-kernel/linux/files/debian/isar/install.tmpl | 6 +++---
>  2 files changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/meta/conf/bitbake.conf b/meta/conf/bitbake.conf
> index acce55d..9859456 100644
> --- a/meta/conf/bitbake.conf
> +++ b/meta/conf/bitbake.conf
> @@ -61,6 +61,7 @@ CACHE = "${TMPDIR}/cache"
>  KERNEL_FILE ?= "vmlinuz"
>  KERNEL_FILE_mipsel ?= "vmlinux"
>  KERNEL_FILE_riscv64 ?= "vmlinux"
> +KERNEL_FILE_arm64 ?= "vmlinux"
>  
>  OVERRIDES = "${DISTRO_ARCH}:${COMPAT_OVERRIDE}:${MACHINE}:${DISTRO}:forcevariable"
>  FILESOVERRIDES = "${DISTRO_ARCH}:${MACHINE}"
> diff --git a/meta/recipes-kernel/linux/files/debian/isar/install.tmpl b/meta/recipes-kernel/linux/files/debian/isar/install.tmpl
> index 8eca4c7..69fb88d 100644
> --- a/meta/recipes-kernel/linux/files/debian/isar/install.tmpl
> +++ b/meta/recipes-kernel/linux/files/debian/isar/install.tmpl
> @@ -26,9 +26,9 @@ do_install() {
>      fi
>      krel="$(${MAKE} O=${O} -s --no-print-directory kernelrelease)"
>      case "${ARCH}" in
> -        mips|powerpc|riscv) kimage_path="boot/vmlinux-${krel}"    ;;
> -                        um) kimage_path="usr/bin/vmlinux-${krel}" ;;
> -                         *) kimage_path="boot/vmlinuz-${krel}"    ;;
> +        mips|powerpc|riscv|arm64) kimage_path="boot/vmlinux-${krel}"    ;;
> +                              um) kimage_path="usr/bin/vmlinux-${krel}" ;;
> +                               *) kimage_path="boot/vmlinuz-${krel}"    ;;
>      esac
>  
>      print_settings
>
Anton Mikanovich Jan. 25, 2021, 2:51 a.m. UTC | #2
25.01.2021 15:32, Jan Kiszka wrote:
> On 22.01.21 18:13, Anton Mikanovich wrote:
>> We are performing gunzip from vmlinuz on arm64, so output should have
>> correct naming.
>>
> What's the practical impact of this change? How did you notice it? Can
> anything break that expected the old incorrect name?
>
> Jan

I've noticed that when building custom kernel for imx8 (which is arm64) 
on one of the private projects. Naming decompressed image as compressed 
one cause bootloader configuring issues and general misunderstanding.
Moreover, the same logic is already fixed for riscv, so arm64 also 
should work in the same way.
All additional logic assuming incorrect name should be fixed of course.
Jan Kiszka Jan. 25, 2021, 3:11 a.m. UTC | #3
On 25.01.21 13:51, Anton Mikanovich wrote:
> 25.01.2021 15:32, Jan Kiszka wrote:
>> On 22.01.21 18:13, Anton Mikanovich wrote:
>>> We are performing gunzip from vmlinuz on arm64, so output should have
>>> correct naming.
>>>
>> What's the practical impact of this change? How did you notice it? Can
>> anything break that expected the old incorrect name?
>>
>> Jan
> 
> I've noticed that when building custom kernel for imx8 (which is arm64)
> on one of the private projects. Naming decompressed image as compressed
> one cause bootloader configuring issues and general misunderstanding.
> Moreover, the same logic is already fixed for riscv, so arm64 also
> should work in the same way.
> All additional logic assuming incorrect name should be fixed of course.
> 

We have tons of arm64 systems out there, that's why I'm asking
cautiously. If you can't specify the impact, whether if it requires a
RECIPE-API-CHANGELOG entry e.g., I guess we need to test it...

Jan

Patch

diff --git a/meta/conf/bitbake.conf b/meta/conf/bitbake.conf
index acce55d..9859456 100644
--- a/meta/conf/bitbake.conf
+++ b/meta/conf/bitbake.conf
@@ -61,6 +61,7 @@  CACHE = "${TMPDIR}/cache"
 KERNEL_FILE ?= "vmlinuz"
 KERNEL_FILE_mipsel ?= "vmlinux"
 KERNEL_FILE_riscv64 ?= "vmlinux"
+KERNEL_FILE_arm64 ?= "vmlinux"
 
 OVERRIDES = "${DISTRO_ARCH}:${COMPAT_OVERRIDE}:${MACHINE}:${DISTRO}:forcevariable"
 FILESOVERRIDES = "${DISTRO_ARCH}:${MACHINE}"
diff --git a/meta/recipes-kernel/linux/files/debian/isar/install.tmpl b/meta/recipes-kernel/linux/files/debian/isar/install.tmpl
index 8eca4c7..69fb88d 100644
--- a/meta/recipes-kernel/linux/files/debian/isar/install.tmpl
+++ b/meta/recipes-kernel/linux/files/debian/isar/install.tmpl
@@ -26,9 +26,9 @@  do_install() {
     fi
     krel="$(${MAKE} O=${O} -s --no-print-directory kernelrelease)"
     case "${ARCH}" in
-        mips|powerpc|riscv) kimage_path="boot/vmlinux-${krel}"    ;;
-                        um) kimage_path="usr/bin/vmlinux-${krel}" ;;
-                         *) kimage_path="boot/vmlinuz-${krel}"    ;;
+        mips|powerpc|riscv|arm64) kimage_path="boot/vmlinux-${krel}"    ;;
+                              um) kimage_path="usr/bin/vmlinux-${krel}" ;;
+                               *) kimage_path="boot/vmlinuz-${krel}"    ;;
     esac
 
     print_settings