defconfig'ed v4.13 does not boot on ThunderX

defconfig'ed v4.13 does not boot on ThunderX

[Itaru Kitayama]

Hi,

defconfig'ed kernel v4.13 does not boot on ThunderX, leaving only
messages below:

EFI stub: Booting Linux Kernel...
EFI stub: Using DTB from configuration table
EFI stub: Exiting boot services and installing virtual address map...

Our HW is by GIGABYTE, the model is R150 T62 rev 100:

http://b2b.gigabyte.com/ARM-Server/R150-T62-rev-100#ov

The BIOS we're on is T43. Has anyone testing upstream kernels seen a 
similar
issue like this?

Itar

defconfig'ed v4.13 does not boot on ThunderX

[Mark Rutland]

On Wed, Oct 04, 2017 at 04:25:58PM +0900, Itaru Kitayama wrote:
> Hi,

Hi,

> defconfig'ed kernel v4.13 does not boot on ThunderX, leaving only
> messages below:
> 
> EFI stub: Booting Linux Kernel...
> EFI stub: Using DTB from configuration table
> EFI stub: Exiting boot services and installing virtual address map...

These are all message from the EFI stub. It's likely that we get into
the kernel proper, but it dies before bringing up the usual console.

Can you try booting with 'earlycon' on the kernel commandline? That
might get some early boot-time console output.

Thanks,
Mark.

defconfig'ed v4.13 does not boot on ThunderX

[Itaru Kitayama]

Mark,
I tried to bring up it adding earlycon=p1011,0x87e024000000
but no luck. Not sure if that's the right value to feed either though.

Itaru

On 2017/10/04 19:08, Mark Rutland wrote:
> On Wed, Oct 04, 2017 at 04:25:58PM +0900, Itaru Kitayama wrote:
>> Hi,
> 
> Hi,
> 
>> defconfig'ed kernel v4.13 does not boot on ThunderX, leaving only
>> messages below:
>>
>> EFI stub: Booting Linux Kernel...
>> EFI stub: Using DTB from configuration table
>> EFI stub: Exiting boot services and installing virtual address map...
> 
> These are all message from the EFI stub. It's likely that we get into
> the kernel proper, but it dies before bringing up the usual console.
> 
> Can you try booting with 'earlycon' on the kernel commandline? That
> might get some early boot-time console output.
> 
> Thanks,
> Mark.
> 

defconfig'ed v4.13 does not boot on ThunderX

[Ard Biesheuvel]

On 4 October 2017 at 13:49, Itaru Kitayama <itaru.kitayama@riken.jp> wrote:
> Mark,
> I tried to bring up it adding earlycon=p1011,0x87e024000000
> but no luck. Not sure if that's the right value to feed either though.
>

Please try just 'earlycon' without = or any arguments, and if that
does not work, use pl011 not p1011 (L not 1)

defconfig'ed v4.13 does not boot on ThunderX

[Itaru Kitayama]

Neither just earlycon nor earlycon=pl011,0x87e24000000 printed out anything.

On 2017/10/04 21:53, Ard Biesheuvel wrote:
> On 4 October 2017 at 13:49, Itaru Kitayama <itaru.kitayama@riken.jp> wrote:
>> Mark,
>> I tried to bring up it adding earlycon=p1011,0x87e024000000
>> but no luck. Not sure if that's the right value to feed either though.
>>
> 
> Please try just 'earlycon' without = or any arguments, and if that
> does not work, use pl011 not p1011 (L not 1)
> 

defconfig'ed v4.13 does not boot on ThunderX

[Itaru Kitayama]

Booting up together also with acpi=force did nothing.

On 2017/10/04 21:53, Ard Biesheuvel wrote:
> On 4 October 2017 at 13:49, Itaru Kitayama <itaru.kitayama@riken.jp> wrote:
>> Mark,
>> I tried to bring up it adding earlycon=p1011,0x87e024000000
>> but no luck. Not sure if that's the right value to feed either though.
>>
> 
> Please try just 'earlycon' without = or any arguments, and if that
> does not work, use pl011 not p1011 (L not 1)
> 

defconfig'ed v4.13 does not boot on ThunderX

[Robin Murphy]

On 04/10/17 08:25, Itaru Kitayama wrote:
> Hi,
> 
> defconfig'ed kernel v4.13 does not boot on ThunderX, leaving only
> messages below:
> 
> EFI stub: Booting Linux Kernel...
> EFI stub: Using DTB from configuration table
> EFI stub: Exiting boot services and installing virtual address map...
> 
> Our HW is by GIGABYTE, the model is R150 T62 rev 100:
> 
> http://b2b.gigabyte.com/ARM-Server/R150-T62-rev-100#ov
> 
> The BIOS we're on is T43. Has anyone testing upstream kernels seen a
> similar
> issue like this?

FWIW defconfig does *not* enable most of the ThunderX drivers (most
importantly, PCI and networking will be missing).

I've just built the v4.13 tag with the appropriate drivers added and
booted it on a different ThunderX box - the net core throws some
warnings from __dev_xdp_attached() but it gets to userspace OK for me.

Robin.

defconfig'ed v4.13 does not boot on ThunderX

[Itaru Kitayama]

Robin,
Could I have your .config off the list?

Thanks,

Itaru

On 2017/10/04 22:38, Robin Murphy wrote:
> On 04/10/17 08:25, Itaru Kitayama wrote:
>> Hi,
>>
>> defconfig'ed kernel v4.13 does not boot on ThunderX, leaving only
>> messages below:
>>
>> EFI stub: Booting Linux Kernel...
>> EFI stub: Using DTB from configuration table
>> EFI stub: Exiting boot services and installing virtual address map...
>>
>> Our HW is by GIGABYTE, the model is R150 T62 rev 100:
>>
>> http://b2b.gigabyte.com/ARM-Server/R150-T62-rev-100#ov
>>
>> The BIOS we're on is T43. Has anyone testing upstream kernels seen a
>> similar
>> issue like this?
> 
> FWIW defconfig does *not* enable most of the ThunderX drivers (most
> importantly, PCI and networking will be missing).
> 
> I've just built the v4.13 tag with the appropriate drivers added and
> booted it on a different ThunderX box - the net core throws some
> warnings from __dev_xdp_attached() but it gets to userspace OK for me.
> 
> Robin.
> 

defconfig'ed v4.13 does not boot on ThunderX

[Mark Rutland]

On Wed, Oct 04, 2017 at 02:38:58PM +0100, Robin Murphy wrote:
> On 04/10/17 08:25, Itaru Kitayama wrote:
> > Hi,
> > 
> > defconfig'ed kernel v4.13 does not boot on ThunderX, leaving only
> > messages below:
> > 
> > EFI stub: Booting Linux Kernel...
> > EFI stub: Using DTB from configuration table
> > EFI stub: Exiting boot services and installing virtual address map...
> > 
> > Our HW is by GIGABYTE, the model is R150 T62 rev 100:
> > 
> > http://b2b.gigabyte.com/ARM-Server/R150-T62-rev-100#ov
> > 
> > The BIOS we're on is T43. Has anyone testing upstream kernels seen a
> > similar issue like this?
> 
> FWIW defconfig does *not* enable most of the ThunderX drivers (most
> importantly, PCI and networking will be missing).

.... it's a bit worrying if earlycon doesn't work without those, though.
It shouldn't be dependendent on either. :/

> I've just built the v4.13 tag with the appropriate drivers added and
> booted it on a different ThunderX box - the net core throws some
> warnings from __dev_xdp_attached() but it gets to userspace OK for me.

Given these seem to exist out in the field, we should probably enable
those drivers in defconfig.

Thanks,
Mark.

defconfig'ed v4.13 does not boot on ThunderX

[Robin Murphy]

On 04/10/17 14:44, Mark Rutland wrote:
> On Wed, Oct 04, 2017 at 02:38:58PM +0100, Robin Murphy wrote:
>> On 04/10/17 08:25, Itaru Kitayama wrote:
>>> Hi,
>>>
>>> defconfig'ed kernel v4.13 does not boot on ThunderX, leaving only
>>> messages below:
>>>
>>> EFI stub: Booting Linux Kernel...
>>> EFI stub: Using DTB from configuration table
>>> EFI stub: Exiting boot services and installing virtual address map...
>>>
>>> Our HW is by GIGABYTE, the model is R150 T62 rev 100:
>>>
>>> http://b2b.gigabyte.com/ARM-Server/R150-T62-rev-100#ov
>>>
>>> The BIOS we're on is T43. Has anyone testing upstream kernels seen a
>>> similar issue like this?
>>
>> FWIW defconfig does *not* enable most of the ThunderX drivers (most
>> importantly, PCI and networking will be missing).
> 
> .... it's a bit worrying if earlycon doesn't work without those, though.
> It shouldn't be dependendent on either. :/

Well, my earlycon was fine too, but that was an a different board with
different firmware, so who knows...

>> I've just built the v4.13 tag with the appropriate drivers added and
>> booted it on a different ThunderX box - the net core throws some
>> warnings from __dev_xdp_attached() but it gets to userspace OK for me.
> 
> Given these seem to exist out in the field, we should probably enable
> those drivers in defconfig.

True, I think the below ought to suffice - arm-soc take defconfig
patches now, right?

Robin.

----->8-----
From: Robin Murphy <robin.murphy@arm.com>
Subject: [PATCH] arm64: Add ThunderX drivers to defconfig

ThunderX needs its PCI host drivers to do anything useful, and it's
probably helpful to have networking by default too.

Signed-off-by: Robin Murphy <robin.murphy@arm.com>
---
 arch/arm64/configs/defconfig | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/arch/arm64/configs/defconfig b/arch/arm64/configs/defconfig
index 34480e9af2e7..1c8e0d77deee 100644
--- a/arch/arm64/configs/defconfig
+++ b/arch/arm64/configs/defconfig
@@ -76,6 +76,8 @@ CONFIG_PCIE_RCAR=y
 CONFIG_PCIE_ROCKCHIP=m
 CONFIG_PCI_HOST_GENERIC=y
 CONFIG_PCI_XGENE=y
+CONFIG_PCI_HOST_THUNDER_PEM=y
+CONFIG_PCI_HOST_THUNDER_ECAM=y
 CONFIG_ARM64_VA_BITS_48=y
 CONFIG_SCHED_MC=y
 CONFIG_NUMA=y
@@ -188,6 +190,7 @@ CONFIG_VIRTIO_NET=y
 CONFIG_AMD_XGBE=y
 CONFIG_NET_XGENE=y
 CONFIG_MACB=y
+CONFIG_THUNDER_NIC_PF=y
 CONFIG_HNS_DSAF=y
 CONFIG_HNS_ENET=y
 CONFIG_E1000E=y
-- 
2.13.4.dirty

defconfig'ed v4.13 does not boot on ThunderX

[Itaru Kitayama]

Rebuilding v4.13 kernel using the config-4.5.0-25.el7.aarch64 gets me to 
the prompt
at least.

I'll be happy to test the updated defconfig once ThunderX changes 
accepted upstream.

On 10/4/17 10:44 PM, Mark Rutland wrote:
> On Wed, Oct 04, 2017 at 02:38:58PM +0100, Robin Murphy wrote:
>> On 04/10/17 08:25, Itaru Kitayama wrote:
>>> Hi,
>>>
>>> defconfig'ed kernel v4.13 does not boot on ThunderX, leaving only
>>> messages below:
>>>
>>> EFI stub: Booting Linux Kernel...
>>> EFI stub: Using DTB from configuration table
>>> EFI stub: Exiting boot services and installing virtual address map...
>>>
>>> Our HW is by GIGABYTE, the model is R150 T62 rev 100:
>>>
>>> http://b2b.gigabyte.com/ARM-Server/R150-T62-rev-100#ov
>>>
>>> The BIOS we're on is T43. Has anyone testing upstream kernels seen a
>>> similar issue like this?
>> FWIW defconfig does *not* enable most of the ThunderX drivers (most
>> importantly, PCI and networking will be missing).
> .... it's a bit worrying if earlycon doesn't work without those, though.
> It shouldn't be dependendent on either. :/
>
>> I've just built the v4.13 tag with the appropriate drivers added and
>> booted it on a different ThunderX box - the net core throws some
>> warnings from __dev_xdp_attached() but it gets to userspace OK for me.
> Given these seem to exist out in the field, we should probably enable
> those drivers in defconfig.
>
> Thanks,
> Mark.