Re: [RFC] m68k: Enable memtest kernel parameter

[Michael Schmitz <schmitzmic@gmail.com>]

Hi Mike,

thanks - from what I can see, the easiest way to fix the problem we see 
with memtest clobbering the ramdisk data then is to either move the 
memblock_reserve() for the ramdisk image into paging_init(), or else 
move early_memtest() into setup_mm.c:setup_arch().

(Tried excepting the ramdisk range in when calling early_memtest() from 
paging_init() but that requires exporting m68k_ramdisk, Geert might not 
be overly keen on that).

Cheers,

	Michael


On 13/09/21 17:42, Mike Rapoport wrote:
> Hi Michael,
>
> On Mon, Sep 13, 2021 at 03:40:12PM +1200, Michael Schmitz wrote:
>> Hi Finn,
>>
>> On 13/09/21 13:34, Finn Thain wrote:
>>>
>>> This appears to work on Aranym, QEMU and the Quadra 630 I have here.
>>> It is completely untested on Coldfire etc. I don't even know whether the
>>> right memory mappings are in place for this to actually work as intended.
>>
>> The comment just above the section from paging_init() in your patch states
>> that all physical memory is mapped into kernel virtual address space. I
>> don't think that mapping is contiguous though.
>
> early_memtest() operates on physical ranges, so it does not matter what is
> the virtual mapping.
>
>>> Would someone take a look please?
>>>
>>> diff --git a/arch/m68k/Kconfig b/arch/m68k/Kconfig
>>> index 6e40f7f29ebc..d87e12b4855b 100644
>>> --- a/arch/m68k/Kconfig
>>> +++ b/arch/m68k/Kconfig
>>> @@ -9,6 +9,7 @@ config M68K
>>>  	select ARCH_HAVE_NMI_SAFE_CMPXCHG if RMW_INSNS
>>>  	select ARCH_MIGHT_HAVE_PC_PARPORT if ISA
>>>  	select ARCH_NO_PREEMPT if !COLDFIRE
>>> +	select ARCH_USE_MEMTEST
>>>  	select ARCH_WANT_IPC_PARSE_VERSION
>>>  	select BINFMT_FLAT_ARGVP_ENVP_ON_STACK
>>>  	select DMA_DIRECT_REMAP if HAS_DMA && MMU && !COLDFIRE
>>> diff --git a/arch/m68k/mm/motorola.c b/arch/m68k/mm/motorola.c
>>> index 3a653f0a4188..d6301a094fc1 100644
>>> --- a/arch/m68k/mm/motorola.c
>>> +++ b/arch/m68k/mm/motorola.c
>>> @@ -455,6 +455,8 @@ void __init paging_init(void)
>>>
>>>  	flush_tlb_all();
>>>
>>> +	early_memtest(min_addr, max_addr);
>>> +
>>>  	/*
>>>  	 * initialize the bad page table and bad page to point
>>>  	 * to a couple of allocated pages
>>>
>>
>> I'd rather start that from availmem, not from min_addr ... I'm amazed this
>> works - I'd expect the kernel text segment to get trashed by the memory test
>> routines.
>
> early_memtest() checks only free memory ranges, i.e. those that were not
> memblock_reserve()d, so as long as we reserve the range [min_addr,
> availmem] the memtest will skip it.
>