performance: 0.824
other: 0.673
device: 0.608
semantic: 0.600
KVM: 0.591
socket: 0.570
network: 0.569
permissions: 0.554
graphic: 0.542
PID: 0.532
debug: 0.475
vnc: 0.406
boot: 0.399
files: 0.236

qemu became more picky parsing -m option

With qemu-kvm-2.9.0-3.fc26.x86_64 I am no longer to specify the memory size using something like "-m 1.00000GiB" but with qemu-kvm-2.7.1-7.fc25.x86_64 I could without any problem.  I now get an error message like:

qemu-system-x86_64: -m 1.00000GiB: Parameter 'size' expects a non-negative number below 2^64
Optional suffix k, M, G, T, P or E means kilo-, mega-, giga-, tera-, peta-
and exabytes, respectively.


Is this expected or a regression?

On 07/31/2017 03:34 AM, Markus Armbruster wrote:
> John Florian <email address hidden> writes:
> 
>> Public bug reported:
>>
>> With qemu-kvm-2.9.0-3.fc26.x86_64 I am no longer to specify the memory
>> size using something like "-m 1.00000GiB" but with qemu-
>> kvm-2.7.1-7.fc25.x86_64 I could without any problem.  I now get an error
>> message like:
>>
>> qemu-system-x86_64: -m 1.00000GiB: Parameter 'size' expects a non-negative number below 2^64
>> Optional suffix k, M, G, T, P or E means kilo-, mega-, giga-, tera-, peta-
>> and exabytes, respectively.
>>
>>
>> Is this expected or a regression?
> 
> We recognize suffix "G".  Before commit 75cdcd1 (v2.9.0), trailing
> garbage after a recognized suffix was silently ignored.  "1.0G",
> "1.0GiB", "1.0Garbage-trucks-of-RAM" were all the same to QEMU.  No
> more.
> 
> All clear?

That said, virsh from libvirt manages to recognize 'G' and 'GiB' as
synonyms (powers of 2), as well as 'GB' (powers of 10); we could justify
patching qemu's parser to accept more valid suffixes, particularly since
'GiB' is a typical suffix that has seen use (in spite of it not being
documented).

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org



Not sure why I can only see Markus' comment here as part of Eric's but anyway... the behavior change *is* expected.

Can qemu behave more like virsh then?  That would be ideal IMHO.  I prefer to specify my RAM in powers of 2 and disk in powers of 10 so that when I test virtually using qemu I more closely match the exact constraints of real hardware.  For the embedded work I do fitting in tight confines, it can make a significant difference.

(I actually to this with a wrapper I have around qemu, which is why you see a floating point value for GiB in my example above.  My wrapper behaves like virsh and takes any *B, *iB format and regurgitates it into something qemu accepts.)

Looks like nobody cared to implement this within 3 years ... and IMHO it's maybe even better to not overload the CLI syntax too much ... so I'm closing this ticket now.