Re: [PATCH v2 qemu 3/3] dump: Add qmp argument "reassembled"

[Stephen Brennan]

Daniel P. Berrangé <berrange@redhat.com> writes:
> On Wed, Sep 13, 2023 at 06:03:15PM -0700, Stephen Brennan wrote:
>> This can be used from QMP command line as "-R" to mirror the
>> corresponding flag for makedumpfile. This enables the kdump_reassembled
>> flag introduced in the previous patch.
>> 
>> Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
>> ---
>>  dump/dump-hmp-cmds.c |  8 +++++++-
>>  dump/dump.c          | 12 +++++++++++-
>>  hmp-commands.hx      |  7 +++++--
>>  qapi/dump.json       | 14 +++++++++++++-
>>  4 files changed, 36 insertions(+), 5 deletions(-)
>
>> diff --git a/qapi/dump.json b/qapi/dump.json
>> index 4ae1f722a9..9cc7c3ea93 100644
>> --- a/qapi/dump.json
>> +++ b/qapi/dump.json
>> @@ -69,6 +69,18 @@
>>  #     to dump all guest's memory, please specify the start @begin and
>>  #     @length
>>  #
>> +# @reassembled: if false (the default), the kdump output formats will use 
>> the
>> +#     "makedumpfile flattened" variant of the format, which is less broadly
>> +#     compatible with analysis tools. The flattened dump can be reassembled
>> +#     after the fact using the command "makedumpfile -R". If true, Qemu
>> +#     attempts to generate the standard kdump format. This requires a
>> +#     seekable file as output -- if the output file is not seekable, then
>> +#     the flattened format is still generated. The standard format is more
>> +#     broadly compatible with debugging tools, but generating it requires a
>> +#     seekable output file descriptor, and could use more system memory due
>> +#     to page cache utilization. This should be left unspecified for 
>> non-kdump
>> +#     output formats.
>> +#
>>  # @format: if specified, the format of guest memory dump.  But non-elf
>>  #     format is conflict with paging and filter, ie.  @paging, @begin
>>  #     and @length is not allowed to be specified with non-elf @format
>> @@ -89,7 +101,7 @@
>>  { 'command': 'dump-guest-memory',
>>    'data': { 'paging': 'bool', 'protocol': 'str', '*detach': 'bool',
>>              '*begin': 'int', '*length': 'int',
>> -            '*format': 'DumpGuestMemoryFormat'} }
>> +            '*reassembled': 'bool', '*format': 'DumpGuestMemoryFormat'} }
>
> The 'reassembled' flag is changing the meaning of 3 out of the 5
> 'format' enum values. IMHO, we should just be adding new formats
> instead of changing the meaning of existing formats. It is a shame
> we have the current 'kdump' naming prefix, but we're stuck with
> that for backwards compat, we need a new prefix. I'd suggest
> 'kdump-raw'. eg
>
> #
> # @DumpGuestMemoryFormat:
> #
> # An enumeration of guest-memory-dump's format.
> #
> # @elf: elf format
> #
> # @kdump-zlib: makedumpfile flattened, kdump-compressed format with 
> zlib-compressed
> #
> # @kdump-lzo: makedumpfile flattened, kdump-compressed format with 
> lzo-compressed
> #
> # @kdump-snappy: makedumpfile flattened, kdump-compressed format with 
> snappy-compressed
> #
> # @kdump-raw-zlib: raw assembled kdump-compressed format with zlib-compressed 
> (since 8.2)
> #
> # @kdump-raw-lzo: raw assembled kdump-compressed format with lzo-compressed 
> (since 8.2)
> #
> # @kdump-raw-snappy: raw assembled kdump-compressed format with 
> snappy-compressed (since 8.2)
> #
> # @win-dmp: Windows full crashdump format, can be used instead of ELF
> #     converting (since 2.13)
> #
> # Since: 2.0
> ##
> { 'enum': 'DumpGuestMemoryFormat',
>   'data': [ 'elf',
>             'kdump-zlib', 'kdump-lzo', 'kdump-snappy',
>             'kdump-raw-zlib', 'kdump-raw-lzo', 'kdump-raw-snappy',
>             'win-dmp' ] }

Hi Daniel,

Sure, I'll go ahead and use this approach instead. One question: I see
that this generates the enumeration DumpGuestMemoryFormat in
qapi-types-dump.h. I just wanted to double-check if there's any ABI
considerations for the numbering of this enum? Inserting kdump-raw-* at
this point would result in 'win-dmp' getting a different numbering, and
it seems possible that the API/ABI which libvirt uses might depend on
the enumeration values not changing. E.G. if libvirt is built against
one version of Qemu and then used with a different one.

Thanks,
Stephen

> With regards,
> Daniel
> -- 
> |: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
> |: https://libvirt.org         -o-            https://fstop138.berrange.com :|
> |: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|