Re: [Qemu-devel] [PATCH v3 0/2] Introduce default ram size in MachineClass

[Nikunj A Dadhania]

Markus Armbruster <address@hidden> writes:

> Nikunj A Dadhania <address@hidden> writes:
>
>> Markus Armbruster <address@hidden> writes:
>>
>>> Nikunj A Dadhania <address@hidden> writes:
>>>
>>>> Current DEFAULT_RAM_SIZE(128MB) enforced by QEMU would not work for
>>>> all machines. Introduce a default_ram_size as part of MachineClass.
>>>>
>>>> The below patches has following behaviour:
>>>>
>>>> 1) If the user does not provide "-m" option, machine's default ram
>>>>    size will be picked.
>>>>
>>>> 2) In case the user provides memory that is lesser than machine's
>>>>    default ram size, we upscale the ram_size to machine's
>>>>    default_ram_size. A warning is displayed for the change that qemu
>>>>    has done.
>>>
>>> Please do not "improve" the user's explicit order that way.  Either
>>> execute the order as is, or reject it as invalid.
>>
>> If there is consensus for doing this, I can change the patches
>> accordingly.
>>
>> Rejection is also change of behaviour. Because till now, a VM would
>> start with any memory size, even if it's less that 128MB
>> (default_ram_size). With rejection, all those VMs would fail booting
>> displaying the warning. Is this OK?
>
> I'd stick to "don't reject".  

Agree, i have already sent v4 with those changes, as there were multiple
opinions against changing the behaviour.

> Yes, the failure mode is ugly.  But protecting the user from it is
> also somewhat problematic, because we don't generally know how much
> RAM the actual guest requires, and it's an incompatible change.  Seems
> not worth it.
>
> Back in 2012, we discussed rejecting RAM size less than 1MiB for PC
> machines, because SeaBIOS requires at least that much, and decided
> against it.  See
> http://www.seabios.org/pipermail/seabios/2012-August/004343.html
>
> If you want to pursue "reject" anyway, please make sure to split this
> into two separate patches: one patch to make the default ram size
> machine-specific, and another patch to reject user requests for less.
>
> [...]

Thanks,
Nikunj