Re: [Qemu-devel] [PATCH] migration: Make analyze-migration script target-page-size aware

[Alexander Graf <agraf@suse.de>]

On 07.07.17 17:00, Dr. David Alan Gilbert wrote:
> * Alexander Graf (agraf@suse.de) wrote:
>> The configuration section has a new subsection to transmit the target page
>> size along with the migration stream. The analyze migration script needs
>> to learn about that to read configuration streams that were triggering
>> this subsection to get transmitted.
>>
>> With this patch applied, I can successfully analyze migration streams
>> on AArch64 again.
>>
>> Signed-off-by: Alexander Graf <agraf@suse.de>
>> ---
>>   scripts/analyze-migration.py | 13 +++++++++++++
>>   1 file changed, 13 insertions(+)
>>
>> diff --git a/scripts/analyze-migration.py b/scripts/analyze-migration.py
>> index 1455387..02784f2 100755
>> --- a/scripts/analyze-migration.py
>> +++ b/scripts/analyze-migration.py
>> @@ -254,12 +254,25 @@ class HTABSection(object):
> 
> (Note I'm not a particularly python person, so take lightly)
> 
>>   class ConfigurationSection(object):
>> +    QEMU_VM_SUBSECTION    = 0x05
>> +
> 
> It's odd, you already have this constant defined twice in this script.

Yes, it lives once per class. I am not sure how to easily make it a global.

> 
>>       def __init__(self, file):
>>           self.file = file
>>   
>>       def read(self):
>>           name_len = self.file.read32()
>>           name = self.file.readstr(len = name_len)
>> +        oldpos = self.file.tell()
>> +        if self.file.read8() == self.QEMU_VM_SUBSECTION:
>> +            name = self.file.readstr()
>> +            version_id = self.file.read32()
>> +            if name == "configuration/target-page-bits":
>> +                target_page_size = self.file.read32()
> 
> All of your other references to target_page_size in the script
> are self.TARGET_PAGE_SIZE.

Well, all other ones are actual page sizes :). This variable is never 
used anywhere - we just want to stash it somewhere.

> You might want to make the conditional subsection check into a function
> somewhere, but that's OK for now.

I really hope we don't need it again. All subsections should be 
described via the in-stream json description. We just missed out the 
configuration one because it's not part of the object model.


Alex