[REGRESSION] Arndale Octa panics when booting 3.14-rc1

[Sylwester Nawrocki]

Hi,

On 03/03/2014 08:04 PM, Tomasz Figa wrote:
> On 03.03.2014 19:57, Javi Merino wrote:
>> Hi,
>>
>> Commit fcb0ee6a3d (clk: Implement clk_unregister) added calls to
>> kref_get() and kref_put() to __clk_get() and __clk_put() without
>> checking if clk is not NULL:
>>
>>> @@ -1987,6 +2097,7 @@ int __clk_get(struct clk *clk)
>>>          if (clk && !try_module_get(clk->owner))
>>>                  return 0;
>>>
>>> +       kref_get(&clk->ref);
>>>          return 1;
>>>   }
>>>
>>> @@ -1995,6 +2106,10 @@ void __clk_put(struct clk *clk)
>>>          if (WARN_ON_ONCE(IS_ERR(clk)))
>>>                  return;
>>>
>>> +       clk_prepare_lock();
>>> +       kref_put(&clk->ref, __clk_release);
>>> +       clk_prepare_unlock();
>>> +
>>>          if (clk)
>>>                  module_put(clk->owner);
>>>   }
>>
>> Before this commit, these functions allowed clk to be NULL.  In
>> particular, the "if (clk)" in __clk_put() is now useless, as clk has
>> been dereferenced by the time you reach it.
>>
>> This causes Arndale Octa to panic on boot:
>>
>> 8<---
>>      [    7.430000] Creating amba device /amba/pdma at 121B0000
>>      [    7.435000]    create child: /amba/mdma at 10800000
>>      [    7.440000] Creating amba device /amba/mdma at 10800000
>>      [    7.445000] Unable to handle kernel NULL pointer dereference at virtual address 00000050
>>
>>      [    7.450000] pgd = c0004000
>>      [    7.450000] [00000050] *pgd=00000000
>>      [    7.455000] Internal error: Oops: 5 [#1] PREEMPT SMP ARM
>>      [    7.455000] Modules linked in:
>>      [    7.455000] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.14.0-rc3-00003-gb6026fe-dirty #23
>>
>>      [    7.455000] task: ee0a8000 ti: ee0a6000 task.ti: ee0a6000
>>      [    7.455000] PC is at __clk_get+0x24/0x84
>>      [    7.455000] LR is at of_clk_get+0x5c/0x74
[...]
>>
>>      [    7.455000] [<c02e5cc8>] (__clk_get) from [<c02e34b8>] (of_clk_get+0x5c/0x74)
>>
>>      [    7.455000] [<c02e34b8>] (of_clk_get) from [<c02e350c>] (of_clk_get_by_name+0x3c/0xb0)
>>
>>      [    7.455000] [<c02e350c>] (of_clk_get_by_name) from [<c02e35a4>] (clk_get+0x24/0x44)
>>
>>      [    7.455000] [<c02e35a4>] (clk_get) from [<c01ee6fc>] (amba_get_enable_pclk+0x10/0x60)
>>
>>      [    7.455000] [<c01ee6fc>] (amba_get_enable_pclk) from [<c01ee7ec>] (amba_device_add+0xa0/0x1e0)
>>
>>      [    7.455000] [<c01ee7ec>] (amba_device_add) from [<c02e0f94>] (of_platform_bus_create+0x190/0x308)
>>
>>      [    7.455000] [<c02e0f94>] (of_platform_bus_create) from [<c02e103c>] (of_platform_bus_create+0x238
>>
>>      [    7.455000] [<c02e103c>] (of_platform_bus_create) from [<c02e116c>] (of_platform_populate+0x60/0x
>>
>>      [    7.455000] [<c02e116c>] (of_platform_populate) from [<c04cc280>] (customize_machine+0x1c/0x40)
>>
>>      [    7.455000] [<c04cc280>] (customize_machine) from [<c00088e8>] (do_one_initcall+0xe4/0x140)
>>
>>      [    7.455000] [<c00088e8>] (do_one_initcall) from [<c04c9c50>] (kernel_init_freeable+0xfc/0x1c8)
>>
>>      [    7.455000] [<c04c9c50>] (kernel_init_freeable) from [<c038447c>] (kernel_init+0x8/0xe4)
>>
>>      [    7.455000] [<c038447c>] (kernel_init) from [<c000e4b8>] (ret_from_fork+0x14/0x3c)
>>
>>      [    7.455000] Code: e3500000 08bd8010 e2843050 f57ff05b (e1932f9f)
>>      [    7.460000] ---[ end trace fe5cfa405506a77d ]---
>>      [    7.460000] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
>>
>> 8<---
>>
>> I've tried putting "if (clk)" around the calls to kref_get() and
>> kref_put(), but that makes the Octa fail to boot like 3.13 did: when
>> tries to mount root.
>>
>> What works for me so far is to remove the offending nodes from the DT:
>>
>> 8<---
>> diff --git a/arch/arm/boot/dts/exynos5420.dtsi b/arch/arm/boot/dts/exynos5420.dtsi
>>
>> index 8db792b..c4ab992 100644
>> --- a/arch/arm/boot/dts/exynos5420.dtsi
>> +++ b/arch/arm/boot/dts/exynos5420.dtsi
>> @@ -303,27 +303,6 @@
>>                          #dma-requests = <32>;
>>                  };
>>
>> -               mdma0: mdma at 10800000 {
>> -                       compatible = "arm,pl330", "arm,primecell";
>> -                       reg = <0x10800000 0x1000>;
>> -                       interrupts = <0 33 0>;
>> -                       clocks = <&clock 473>;
>> -                       clock-names = "apb_pclk";
>> -                       #dma-cells = <1>;
>> -                       #dma-channels = <8>;
>> -                       #dma-requests = <1>;
>> -               };
>> -
>> -               mdma1: mdma at 11C10000 {
>> -                       compatible = "arm,pl330", "arm,primecell";
>> -                       reg = <0x11C10000 0x1000>;
>> -                       interrupts = <0 124 0>;
>> -                       clocks = <&clock 442>;
>> -                       clock-names = "apb_pclk";
>> -                       #dma-cells = <1>;
>> -                       #dma-channels = <8>;
>> -                       #dma-requests = <1>;
>> -               };
>>          };
>>
>>          spi_0: spi at 12d20000 {
>> 8<---
>>
>> With this the Arndale Octa boots.  This issue is present from 3.14-rc1
>> to 3.14-rc5.  Any idea how this should be properly fixed in mainline?
>
> This should be fixed with [1], which was supposed to be merged for
> 3.14-rc5. Mike?
>
> [1] https://git.linaro.org/people/mike.turquette/linux.git/commit/00efcb1c8e1c3c5e5d3ce6f0682d66402911a84f

You may also want to consider this patch:
https://patchwork.kernel.org/patch/3597711

It looks like there is no clock registered for the mdm0 PL330 device
in exynos5420 clocks driver (drivers/clk/samsung/clk-exynos5420.c).

So a proper fix, except of a patch pointed out by Tomasz, would be to
add a corresponding entry to exynos5420_gate_clocks[] array (the mdm0
gate clock seems to be already documented in Documentation/devicetree/
bindings/clock/exynos5420-clock.txt).

AFAIU, you're still having problems even with patch [1] applied ?

--
Regards,
Sylwester