"clk: sunxi: Add a simple gates driver" breaks kernel with older DTB

"clk: sunxi: Add a simple gates driver" breaks kernel with older DTB

[Ian Campbell]

Booting a recent kernel with the DTB supplied with Debian Jessie (3.16
based) breaks on Cubietruck because that DTB lacks the clock-indices nodes
which the new driver from the commit below adds (replacing the hardcoding
which used to be in clk-sunxi.c).

It is panicing in drivers/clocksource/timer-sun5i.c with:

[    0.015413] clocksource: timer: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 79635851949 ns
[    0.025049] Kernel panic - not syncing: Can't get timer clock
[    0.030794] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.3.0-rc2-arm-native+ #6
[    0.038002] Hardware name: Allwinner sun7i (A20) Family
[    0.043253] [<c0219794>] (unwind_backtrace) from [<c0214e44>] (show_stack+0x10/0x14)
[    0.050992] [<c0214e44>] (show_stack) from [<c048159c>] (dump_stack+0x88/0x98)
[    0.058213] [<c048159c>] (dump_stack) from [<c02caa98>] (panic+0xa4/0x22c)
[    0.065090] [<c02caa98>] (panic) from [<c0d7c518>] (sun5i_timer_init+0x80/0x384)
[    0.072482] [<c0d7c518>] (sun5i_timer_init) from [<c0d7aad0>] (clocksource_of_init+0x4c/0x8c)
[    0.081001] [<c0d7aad0>] (clocksource_of_init) from [<c0d39b48>] (start_kernel+0x28c/0x3c4)
[    0.089343] [<c0d39b48>] (start_kernel) from [<4020807c>] (0x4020807c)
[    0.095866] ---[ end Kernel panic - not syncing: Can't get timer clock

Reverting ee38b2698ae2 fixes this specific issue for me (it boots further,
but there seems to be other problems later when earlycon hands over to
proper console, which I've not yet looked into).

Is this considered acceptable linkage between the kernel and the dtbs?

I suspect that even if anyone does care this is going to be an uphill
struggle for that minority so I'm going to adjust our test infrastructure
to pickup the dtbs from the kernel it is trying to test rather then reusing
the one from the OS install.

Ian.

commit ee38b2698ae234c03f65ccafa1811d4dda3c316d
Author: Maxime Ripard <maxime.ripard@free-electrons.com>
Date:   Fri Jul 31 19:46:22 2015 +0200

    clk: sunxi: Add a simple gates driver
    
    The gates were handled with a common piece of framework that was
    registering all gates array, that was not using the CLK_OF_DECLARE logic,
    and was not using clock-indices but some private masks that were pretty
    much equivalent.
    
    Move this code in a new driver that handles all the gates array and solves
    both these issues.
    
    Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
    Signed-off-by: Michael Turquette <mturquette@baylibre.com>
    [sboyd@codeaurora.org: Include clk.h for consumer API usage]
    Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>

Re: "clk: sunxi: Add a simple gates driver" breaks kernel with older DTB

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 2297 bytes --]

Hi Ian,

On Thu, Oct 01, 2015 at 09:47:11AM +0100, Ian Campbell wrote:
> Booting a recent kernel with the DTB supplied with Debian Jessie (3.16
> based) breaks on Cubietruck because that DTB lacks the clock-indices nodes
> which the new driver from the commit below adds (replacing the hardcoding
> which used to be in clk-sunxi.c).
> 
> It is panicing in drivers/clocksource/timer-sun5i.c with:
> 
> [    0.015413] clocksource: timer: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 79635851949 ns
> [    0.025049] Kernel panic - not syncing: Can't get timer clock
> [    0.030794] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.3.0-rc2-arm-native+ #6
> [    0.038002] Hardware name: Allwinner sun7i (A20) Family
> [    0.043253] [<c0219794>] (unwind_backtrace) from [<c0214e44>] (show_stack+0x10/0x14)
> [    0.050992] [<c0214e44>] (show_stack) from [<c048159c>] (dump_stack+0x88/0x98)
> [    0.058213] [<c048159c>] (dump_stack) from [<c02caa98>] (panic+0xa4/0x22c)
> [    0.065090] [<c02caa98>] (panic) from [<c0d7c518>] (sun5i_timer_init+0x80/0x384)
> [    0.072482] [<c0d7c518>] (sun5i_timer_init) from [<c0d7aad0>] (clocksource_of_init+0x4c/0x8c)
> [    0.081001] [<c0d7aad0>] (clocksource_of_init) from [<c0d39b48>] (start_kernel+0x28c/0x3c4)
> [    0.089343] [<c0d39b48>] (start_kernel) from [<4020807c>] (0x4020807c)
> [    0.095866] ---[ end Kernel panic - not syncing: Can't get timer clock
> 
> Reverting ee38b2698ae2 fixes this specific issue for me (it boots further,
> but there seems to be other problems later when earlycon hands over to
> proper console, which I've not yet looked into).
> 
> Is this considered acceptable linkage between the kernel and the dtbs?
> 
> I suspect that even if anyone does care this is going to be an uphill
> struggle for that minority so I'm going to adjust our test infrastructure
> to pickup the dtbs from the kernel it is trying to test rather then reusing
> the one from the OS install.

It's never been something we supported (or even claim to support), so
it's actually surprising that it's the only commit that need to be
reverted, but yeah, you should keep your DTB in sync.

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: "clk: sunxi: Add a simple gates driver" breaks kernel with older DTB

[Julien Grall]

Hi Maxime,

On 01/10/15 21:45, Maxime Ripard wrote:
> On Thu, Oct 01, 2015 at 09:47:11AM +0100, Ian Campbell wrote:
>> Booting a recent kernel with the DTB supplied with Debian Jessie (3.16
>> based) breaks on Cubietruck because that DTB lacks the clock-indices nodes
>> which the new driver from the commit below adds (replacing the hardcoding
>> which used to be in clk-sunxi.c).
>>
>> It is panicing in drivers/clocksource/timer-sun5i.c with:
>>
>> [    0.015413] clocksource: timer: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 79635851949 ns
>> [    0.025049] Kernel panic - not syncing: Can't get timer clock
>> [    0.030794] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.3.0-rc2-arm-native+ #6
>> [    0.038002] Hardware name: Allwinner sun7i (A20) Family
>> [    0.043253] [<c0219794>] (unwind_backtrace) from [<c0214e44>] (show_stack+0x10/0x14)
>> [    0.050992] [<c0214e44>] (show_stack) from [<c048159c>] (dump_stack+0x88/0x98)
>> [    0.058213] [<c048159c>] (dump_stack) from [<c02caa98>] (panic+0xa4/0x22c)
>> [    0.065090] [<c02caa98>] (panic) from [<c0d7c518>] (sun5i_timer_init+0x80/0x384)
>> [    0.072482] [<c0d7c518>] (sun5i_timer_init) from [<c0d7aad0>] (clocksource_of_init+0x4c/0x8c)
>> [    0.081001] [<c0d7aad0>] (clocksource_of_init) from [<c0d39b48>] (start_kernel+0x28c/0x3c4)
>> [    0.089343] [<c0d39b48>] (start_kernel) from [<4020807c>] (0x4020807c)
>> [    0.095866] ---[ end Kernel panic - not syncing: Can't get timer clock
>>
>> Reverting ee38b2698ae2 fixes this specific issue for me (it boots further,
>> but there seems to be other problems later when earlycon hands over to
>> proper console, which I've not yet looked into).
>>
>> Is this considered acceptable linkage between the kernel and the dtbs?
>>
>> I suspect that even if anyone does care this is going to be an uphill
>> struggle for that minority so I'm going to adjust our test infrastructure
>> to pickup the dtbs from the kernel it is trying to test rather then reusing
>> the one from the OS install.
> 
> It's never been something we supported (or even claim to support), so
> it's actually surprising that it's the only commit that need to be
> reverted, but yeah, you should keep your DTB in sync.

I would have though the bindings are stable, i.e the compatibility with
old DT would have been kept in newer kernel.

So is it something specific to cubietruck? If no, that's very
unfortunate because it means that you can't re-use the same DT across
multiple OS and the DT provided by the firmware (if it's built-in).

Regards,

-- 
Julien Grall

Re: "clk: sunxi: Add a simple gates driver" breaks kernel with older DTB

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 3622 bytes --]

Hi,

On Fri, Oct 02, 2015 at 11:07:35AM +0100, Julien Grall wrote:
> Hi Maxime,
> 
> On 01/10/15 21:45, Maxime Ripard wrote:
> > On Thu, Oct 01, 2015 at 09:47:11AM +0100, Ian Campbell wrote:
> >> Booting a recent kernel with the DTB supplied with Debian Jessie (3.16
> >> based) breaks on Cubietruck because that DTB lacks the clock-indices nodes
> >> which the new driver from the commit below adds (replacing the hardcoding
> >> which used to be in clk-sunxi.c).
> >>
> >> It is panicing in drivers/clocksource/timer-sun5i.c with:
> >>
> >> [    0.015413] clocksource: timer: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 79635851949 ns
> >> [    0.025049] Kernel panic - not syncing: Can't get timer clock
> >> [    0.030794] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.3.0-rc2-arm-native+ #6
> >> [    0.038002] Hardware name: Allwinner sun7i (A20) Family
> >> [    0.043253] [<c0219794>] (unwind_backtrace) from [<c0214e44>] (show_stack+0x10/0x14)
> >> [    0.050992] [<c0214e44>] (show_stack) from [<c048159c>] (dump_stack+0x88/0x98)
> >> [    0.058213] [<c048159c>] (dump_stack) from [<c02caa98>] (panic+0xa4/0x22c)
> >> [    0.065090] [<c02caa98>] (panic) from [<c0d7c518>] (sun5i_timer_init+0x80/0x384)
> >> [    0.072482] [<c0d7c518>] (sun5i_timer_init) from [<c0d7aad0>] (clocksource_of_init+0x4c/0x8c)
> >> [    0.081001] [<c0d7aad0>] (clocksource_of_init) from [<c0d39b48>] (start_kernel+0x28c/0x3c4)
> >> [    0.089343] [<c0d39b48>] (start_kernel) from [<4020807c>] (0x4020807c)
> >> [    0.095866] ---[ end Kernel panic - not syncing: Can't get timer clock
> >>
> >> Reverting ee38b2698ae2 fixes this specific issue for me (it boots further,
> >> but there seems to be other problems later when earlycon hands over to
> >> proper console, which I've not yet looked into).
> >>
> >> Is this considered acceptable linkage between the kernel and the dtbs?
> >>
> >> I suspect that even if anyone does care this is going to be an uphill
> >> struggle for that minority so I'm going to adjust our test infrastructure
> >> to pickup the dtbs from the kernel it is trying to test rather then reusing
> >> the one from the OS install.
> > 
> > It's never been something we supported (or even claim to support), so
> > it's actually surprising that it's the only commit that need to be
> > reverted, but yeah, you should keep your DTB in sync.
> 
> I would have though the bindings are stable, i.e the compatibility with
> old DT would have been kept in newer kernel.
> 
> So is it something specific to cubietruck?

Nope. All the platforms I've been using on a regular basis do not
guarantee this. Allwinner, obviously, but also:

Atmel:
  * https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/Documentation/arm/Atmel/README#n110

OMAP:
  * https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=8b9a2810b02e3d9806ba2bf307c8e8dcedaf902d

Berlin:
  * https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=fd26031ba6356d2b0f7aa214ebff4b12736b6529

Rockchip:
  * https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=26ab69cb4c1f77060bece483f2ec210163867782

etc...

> If no, that's very unfortunate because it means that you can't
> re-use the same DT across multiple OS and the DT provided by the
> firmware (if it's built-in).

Is there such OS yet (and by that, I mean one that actually shares our
DT, instead of rewriting its own) ?

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: "clk: sunxi: Add a simple gates driver" breaks kernel with older DTB

[Julien Grall]

On 02/10/15 11:53, Maxime Ripard wrote:
>> If no, that's very unfortunate because it means that you can't
>> re-use the same DT across multiple OS and the DT provided by the
>> firmware (if it's built-in).
> 
> Is there such OS yet (and by that, I mean one that actually shares our
> DT, instead of rewriting its own) ?

Yes, FreeBSD started to support the DT provided by vendors which are
based on Linux bindings. The ARM64 port is only using DTB provided by
the hardware. For ARM32, there is still some platform using a specific
DT for FreeBSD but they are trying to get a ride.

Regards,

-- 
Julien Grall

Re: "clk: sunxi: Add a simple gates driver" breaks kernel with older DTB

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 1000 bytes --]

1;3803;0c
On Fri, Oct 02, 2015 at 12:15:52PM +0100, Julien Grall wrote:
> On 02/10/15 11:53, Maxime Ripard wrote:
> >> If no, that's very unfortunate because it means that you can't
> >> re-use the same DT across multiple OS and the DT provided by the
> >> firmware (if it's built-in).
> > 
> > Is there such OS yet (and by that, I mean one that actually shares our
> > DT, instead of rewriting its own) ?
> 
> Yes, FreeBSD started to support the DT provided by vendors which are
> based on Linux bindings. The ARM64 port is only using DTB provided by
> the hardware. For ARM32, there is still some platform using a specific
> DT for FreeBSD but they are trying to get a ride.

Well, it would have been good if they contacted us in the first place
then, because this has never been considered seriously afaik, and we
never stated that it actually was an ABI.

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]