Re: [PATCH 5/6] ARM: ux500: Enable HIGHMEM on all mop500 platforms

[Lee Jones <lee.jones@linaro.org>]

On 01/08/12 09:41, Russell King - ARM Linux wrote:
> On Wed, Aug 01, 2012 at 08:56:14AM +0100, Lee Jones wrote:
>> On 31/07/12 23:01, Russell King - ARM Linux wrote:
>>> On Tue, Jul 31, 2012 at 08:50:02PM +0000, Arnd Bergmann wrote:
>>>> On Tuesday 31 July 2012, Russell King - ARM Linux wrote:
>>>>> I still fail to see how not having highmem enabled would ever cause memory
>>>>> corruption errors (unless something dealing with memory in a very very
>>>>> wrong way - iow, not using one of the reservation or memory allocation
>>>>> methods provided by the kernel.)
>>>>
>>>> The problem is that all users of ux500 systems pass a command line like
>>>>
>>>> vmalloc=256M mem=128M@0 mali.mali_mem=32M@128M hwmem=168M@160M mem=48M@328M mem_issw=1M@383M mem=640M@384M
>>>>
>>>> This is of course totally bogus and should not be done. If I understand
>>>> Lee correctly, one of the issues resulting from passing a command
>>>> line like this without enabling highmem is memory corruption.
>>>
>>> But the question is _why_ does that corruption happen.
>>>
>>>   From the above, we will end up with the kernel getting:
>>>
>>> 0x00000000 - 0x07ffffff (128M @ 0)
>>> 0x14800000 - 0x177fffff (48M  @ 328M)
>>> 0x18000000 - 0x3fffffff (640M @ 384M)
>>>
>>> with:
>>>
>>> 0x08000000 - 0x081fffff used for mali
>>> 0x0a000000 - 0x147fffff used for hwmem
>>> 0x17f00000 - 0x17ffffff used for mem_issw
>>>
>>> Now, with highmem disabled, the kernel should still map exactly the
>>> regions: 0x00000000 - 0x07ffffff, 0x14800000 - 0x177fffff, into the
>>> direct mapped region, and truncate the 0x18000000 - 0x3fffffff
>>> region appropriately, reducing the amount of memory available such
>>> that it won't overlap the vmalloc area (which you've specified to be
>>> a minimum of 256M.)
>>>
>>> This should _NOT_ cause any memory corruption.
>>>
>>> So, come on guys.  Debugging is *mandatory* for this kind of problem.
>>> Papering over it is obscene.
>>
>> Actually I didn't go any further with it, as I changed to another
>> identical piece of hardware and couldn't reproduce the issue.
>>
>> FYI, here's the boot log from the broken board:
>>
>> http://paste.ubuntu.com/1102017/
>
> Well, the good thing is this:
>
>     8 Truncating RAM at 18000000-3fffffff to -2c3fffff (vmalloc region overlap).
>
> which means the RAM was properly truncated before it is passed to
> memblock, etc.
>
> That oops dump looks very much like an ASoC problem, where
> dapm_widget_power_check() recurses into dapm_supply_check_power()
> which then recurses back into dapm_widget_power_check(), and it
> eventually overflows the kernel stack, corrupting the thread_info
> and the pages below.
>
> Given the address of the stack pointer (ebc480a8) I don't think
> we can be too sure where it was supposed to be, and where the top
> of stack should have been, so we don't know how many pages have
> been stomped on and corrupted.
>
> Stopping that recursion is the first thing that needs to be done
> so that the cause of it can then be properly debugged without the
> kernel itself corrupting memory below the kernel stack.

Those were my thoughts.

Here was my cry for help: https://lkml.org/lkml/2012/7/23/181

-- 
Lee Jones
Linaro ST-Ericsson Landing Team Lead
Linaro.org │ Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog