[5.2-rc6] GX6605S testing

[5.2-rc6] GX6605S testing

[Andreas Färber]

Hello,

Newbie alert. I've tailored a Kconfig for the GX6605S dev board based on
defconfig, which required to drop "if COMPILE_TEST" for GX6605S_TIMER
since nothing was selecting it.

I've forward-ported the board's binary dtb to match the upstream
bindings (still need to do some cleanups before I can post it here).
Mainly compatibles and one clock-frequency -> clocks.

Here's how far I've come with the uImage:

...
[   60.977460] sd 0:0:0:0: [sda] Attached SCSI removable disk
[   61.357214] random: crng init done
[   64.539130] EXT4-fs (sda2): mounted filesystem with ordered data
mode. Opts: (null)
[   64.763086] VFS: Mounted root (ext4 filesystem) on device 8:2.
[   66.946712] devtmpfs: mounted
[   67.058804] Freeing unused kernel memory: 100k freed
[   67.205338] This architecture does not have kernel memory protection.
[   67.386763] Run /sbin/init as init process
[   77.719086] Kernel panic - not syncing: Attempted to kill init!
exitcode=0x0000000b
[   77.927527] ---[ end Kernel panic - not syncing: Attempted to kill
init! exitcode=0x0000000b ]---

Note how the time is quite fast with the downstream frequency of 1 MHz.
Sadly there's no public manual for the GX6605S SoC, only for CK610 CPU.
24 MHz matches the wall clock much better, indicating about 3 seconds.

I've also tried using /bin/sh with the equivalent result.
The rootfs is the one from usb.img from the C-Sky GitHub pages (modified
to spawn getty for ttyS0), which booted okay with its binary 4.9 kernel.

So, does this mean that syscalls or something changed incompatibly
between downstream 4.9 and upstream 5.2 that I would need a new rootfs?
Or does this mean that I'm still missing some Kconfig setting or that
some fix/feature for CK610 hasn't made it into the mainline kernel yet?
Did anyone here succeed with mainline on GX6605S yet?

Thanks for any directions,

Andreas


https://hackweek.suse.com/18/projects/investigate-c-sky-architecture

-- 
SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Felix Imendörffer, Mary Higgins, Sri Rasiah
HRB 21284 (AG Nürnberg)

Re: [5.2-rc6] GX6605S testing

[Andreas Färber]

Am 24.06.19 um 20:12 schrieb Andreas Färber:
> Hello,
> 
> Newbie alert. I've tailored a Kconfig for the GX6605S dev board based on
> defconfig, which required to drop "if COMPILE_TEST" for GX6605S_TIMER
> since nothing was selecting it.

Another observation: earlycon didn't seem to work. After I sorted the
missing timer out I found this:

[    0.000000] earlycon: earlycon_map: Couldn't map 0x00403000
[    0.000000] earlycon: ns16550a0 at MMIO 0x00403000 (options '115200n8')

Having it working would be extremely helpful for diagnosing early-boot
issues such as a missing timer driver. ;)

Cheers,
Andreas

-- 
SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Felix Imendörffer, Mary Higgins, Sri Rasiah
HRB 21284 (AG Nürnberg)

Re: [5.2-rc6] GX6605S testing

[Andreas Färber]

Am 24.06.19 um 20:12 schrieb Andreas Färber:
> Hello,
> 
> Newbie alert. I've tailored a Kconfig for the GX6605S dev board based on
> defconfig, which required to drop "if COMPILE_TEST" for GX6605S_TIMER
> since nothing was selecting it.

Turns out this Kconfig issue is already fixed in linux-next.git; the
other two reported issues remain.

Regards,
Andreas

-- 
SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Felix Imendörffer, Mary Higgins, Sri Rasiah
HRB 21284 (AG Nürnberg)

Re: [5.2-rc6] GX6605S testing

[Guo Ren]

Hi Andreas,

On Tue, Jun 25, 2019 at 5:19 AM Andreas Färber <afaerber@suse.de> wrote:
>
> Hello,
>
> Newbie alert. I've tailored a Kconfig for the GX6605S dev board based on
> defconfig, which required to drop "if COMPILE_TEST" for GX6605S_TIMER
> since nothing was selecting it.
>
> I've forward-ported the board's binary dtb to match the upstream
> bindings (still need to do some cleanups before I can post it here).
> Mainly compatibles and one clock-frequency -> clocks.
It's 27Mhz:
+ dummy_apb: apb-clock {
+ compatible = "fixed-clock";
+ clock-frequency = <27000000>;


>
> Here's how far I've come with the uImage:
>
> ...
> [   60.977460] sd 0:0:0:0: [sda] Attached SCSI removable disk
> [   61.357214] random: crng init done
> [   64.539130] EXT4-fs (sda2): mounted filesystem with ordered data
> mode. Opts: (null)
> [   64.763086] VFS: Mounted root (ext4 filesystem) on device 8:2.
> [   66.946712] devtmpfs: mounted
> [   67.058804] Freeing unused kernel memory: 100k freed
> [   67.205338] This architecture does not have kernel memory protection.
> [   67.386763] Run /sbin/init as init process
> [   77.719086] Kernel panic - not syncing: Attempted to kill init!
> exitcode=0x0000000b
> [   77.927527] ---[ end Kernel panic - not syncing: Attempted to kill
> init! exitcode=0x0000000b ]---
>
> Note how the time is quite fast with the downstream frequency of 1 MHz.
> Sadly there's no public manual for the GX6605S SoC, only for CK610 CPU.
> 24 MHz matches the wall clock much better, indicating about 3 seconds.
>
> I've also tried using /bin/sh with the equivalent result.
> The rootfs is the one from usb.img from the C-Sky GitHub pages (modified
> to spawn getty for ttyS0), which booted okay with its binary 4.9 kernel.
>
> So, does this mean that syscalls or something changed incompatibly
> between downstream 4.9 and upstream 5.2 that I would need a new rootfs?
> Or does this mean that I'm still missing some Kconfig setting or that
> some fix/feature for CK610 hasn't made it into the mainline kernel yet?
> Did anyone here succeed with mainline on GX6605S yet?
The syscall.h is changed for newest arch version. You need new rootfs
and I recommend you try this buildroot patch:
https://lore.kernel.org/linux-csky/1561376581-19568-1-git-send-email-guoren@kernel.org/T/#u

Use that rootfs.ext2 in output image.

-- 
Best Regards
 Guo Ren

ML: https://lore.kernel.org/linux-csky/

Re: [5.2-rc6] GX6605S testing

[Andreas Färber]

Hi Guo,

Am 25.06.19 um 02:53 schrieb Guo Ren:
> On Tue, Jun 25, 2019 at 5:19 AM Andreas Färber <afaerber@suse.de> wrote:
>> [   67.205338] This architecture does not have kernel memory protection.
>> [   67.386763] Run /sbin/init as init process
>> [   77.719086] Kernel panic - not syncing: Attempted to kill init!
>> exitcode=0x0000000b
>> [   77.927527] ---[ end Kernel panic - not syncing: Attempted to kill
>> init! exitcode=0x0000000b ]---
[...]
>> So, does this mean that syscalls or something changed incompatibly
>> between downstream 4.9 and upstream 5.2 that I would need a new rootfs?
[...]
> The syscall.h is changed for newest arch version. You need new rootfs
> and I recommend you try this buildroot patch:
> https://lore.kernel.org/linux-csky/1561376581-19568-1-git-send-email-guoren@kernel.org/T/#u
> 
> Use that rootfs.ext2 in output image.

Given that Buildroot is not building for me (host-gmp - both upstream
and downstream), is such a mainline-compatible rootfs tarball somewhere
on GitLab? The closest job I found there was 4.14 gx6622, nothing 5.x
and gx6605s.

Thanks,
Andreas

-- 
SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Felix Imendörffer, Mary Higgins, Sri Rasiah
HRB 21284 (AG Nürnberg)