Re: [Qemu-devel] [PATCH] vl: move rom_load_all after machine init done

[Eric Auger]

On 07/07/2015 11:02 AM, Paolo Bonzini wrote:
> 
> 
> On 07/07/2015 11:00, Eric Auger wrote:
>> Hi Paolo, Peter,
>> On 06/22/2015 11:58 AM, Eric Auger wrote:
>>> On 06/22/2015 11:53 AM, Paolo Bonzini wrote:
>>>>
>>>>
>>>> On 22/06/2015 11:49, Eric Auger wrote:
>>>>>>> It seems safe because rom_load_all really doesn't load anything, it only
>>>>>>> does an overlap check.  Is this right?
>>>>> it does the check + isrom field setting
>>>>>>>
>>>>>>> Is the bug that some overlapping ROMs are not detected?  The commit
>>>>>>> message is not clear.
>>>>> The regression is that the both overlap check and isrom setting are not
>>>>> done since ROM are inserted in the roms list afterwards, at machine init
>>>>> done time. The bug was not really observed yet I think.
>>>>
>>>> isrom is just an optimization though, right?  What is it useful for?
>>> My understanding is it serves 2 purposes:
>>>
>>> - report info in the monitor (hmp_info_roms)
>>> - decide whether the rom->data can be freed on ROM reset notifier
>>> (rom_reset).
>>>
>>> Hope I didn't miss anything else.
>>>
>>> Eric
>>
>> What do we decide then about this regression on arm. Do we fix it in 2.4
>> or later?
> 
> Yes, it should be fixed in 2.4.
Do you want me to resend it with a new commit message or is the context
clearer now?

Thanks

Eric
> 
> Paolo
>