Re: [Qemu-devel] [RFC] s390: dump-guest-memory implementation for s390x arch.

[Christian Borntraeger <borntraeger@de.ibm.com>]

>>>> diff --git a/target-s390x/arch_memory_mapping.c b/target-s390x/arch_memory_mapping.c
>>>> new file mode 100644
>>>> index 0000000..3dad3b9
>>>> --- /dev/null
>>>> +++ b/target-s390x/arch_memory_mapping.c
>>>> @@ -0,0 +1,26 @@
>>>> +/*
>>>> + * s390x memory mapping
>>>> + *
>>>> + * Copyright IBM Corp. 2012
>>>> + *
>>>> + * Authors:
>>>> + *     Ekaterina Tumanova <tumanova@linux.vnet.ibm.com>
>>>> + *
>>>> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
>>>> + * See the COPYING file in the top-level directory.
>>>> + *
>>>> + */
>>>> +
>>>> +#include "cpu.h"
>>>> +#include "exec/cpu-all.h"
>>>> +#include "sysemu/memory_mapping.h"
>>>> +
>>>> +int cpu_get_memory_mapping(MemoryMappingList *list, CPUArchState *env)
>>>> +{
>>>> +    return 0;
>>>> +}
>>>> +
>>>> +bool cpu_paging_enabled(CPUArchState *env)
>>>> +{
>>>> +    return false;
>>>
>>> Why? :)
>>
>> To make things compile :-)
>> Why no paging? The idea was to have some code that allows getting an ELF dump of a Linux kernel.
>> All existing tooling for s390 expects real memory dumps. 
>>
>> Actually having CONFIG_HAVE_GET_MEMORY_MAPPING=n would be a sane thing todo, but IIRC the common
>> code in dump.c references some functions that are only available with 
>> CONFIG_HAVE_GET_MEMORY_MAPPING=y. 
>>
>> /home/cborntra/REPOS/qemu/dump.c:84: undefined reference to `memory_mapping_list_free'
>> dump.o: In function `dump_init':
>> /home/cborntra/REPOS/qemu/dump.c:753: undefined reference to `memory_mapping_list_init'
>> /home/cborntra/REPOS/qemu/dump.c:757: undefined reference to `qemu_get_guest_simple_memory_mapping'
>> /home/cborntra/REPOS/qemu/dump.c:761: undefined reference to `memory_mapping_filter'
> 
> It might be better to just fix those places and not claim that you can fetch memory mappings then. QEMU has all the knowledge of s390 virtual memory you would need to implement it. I can see how you wouldn't need it for core dumps of the kernel, but it's confusing to claim dump support for VM and then say you don't.
> 
> 
> Alex

So you suggest to implementing stubs for those 4 functions in memory_mapping-stub.c and see if dumping
of a kernel still works with CONFIG_HAVE_GET_MEMORY_MAPPING=n, right?

Christian