Re: [PATCH] memory: Display MemoryRegion name in read/write ops trace events

[Laurent Vivier]

Le 07/06/2021 à 10:28, Philippe Mathieu-Daudé a écrit :
> On 6/7/21 9:33 AM, Laurent Vivier wrote:
>> Le 02/06/2021 à 12:49, Philippe Mathieu-Daudé a écrit :
>>> Cc'ing qemu-trivial@
>>>
>>> On 3/18/21 4:39 PM, Philippe Mathieu-Daudé wrote:
>>>> ping?
>>>>
>>>> On 3/7/21 8:48 AM, Philippe Mathieu-Daudé wrote:
>>>>> MemoryRegion names is cached on first call to memory_region_name(),
>>
>> It is cached on first call but now that it is used in the trace function, 
>> does it mean it will be
>> always allocated in memory?
> 
> Yes, this is how memory_region_name() works:
> 
> const char *memory_region_name(const MemoryRegion *mr)
> {
>     if (!mr->name) {
>         ((MemoryRegion *)mr)->name =
>             g_strdup(object_get_canonical_path_component(OBJECT(mr)));
>     }
>     return mr->name;
> }

OK, in fact I didn't see it was called from inside a "if
(trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_XXX))" and was worrying 
about the memory for
the string always allocated.

So it looks good.

Thanks,
Laurent