Re: [Qemu-devel] [PATCH] memory: Set notdirty_mem_ops validator

[Eric Blake]

On 9/6/19 3:28 AM, Philippe Mathieu-Daudé wrote:
> On 9/2/19 3:26 AM, Tony Nguyen wrote:
>> Existing read rejecting validator was mistakenly cleared.
>>
>> Reads dispatched to io_mem_notdirty then segfaults as there is no read
>> handler.
>>
>> Signed-off-by: Tony Nguyen <address@hidden>
>> ---
>>  exec.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/exec.c b/exec.c
>> index 1df966d17a..05d664541f 100644
>> --- a/exec.c
>> +++ b/exec.c
>> @@ -2796,12 +2796,12 @@ static bool notdirty_mem_accepts(void *opaque, 
>> hwaddr addr,
>>  
>>  static const MemoryRegionOps notdirty_mem_ops = {
>>      .write = notdirty_mem_write,
>> -    .valid.accepts = notdirty_mem_accepts,
>>      .endianness = DEVICE_NATIVE_ENDIAN,
>>      .valid = {
>>          .min_access_size = 1,
>>          .max_access_size = 8,
>>          .unaligned = false,
>> +        .accepts = notdirty_mem_accepts,
> 
> I'm surprised the compiler doesn't emit any warning...

Same here.

But reading
https://en.cppreference.com/w/c/language/struct_initialization, this is
compliant behavior:

"However, when an initializer begins with a left open brace, its current
object is fully re-initialized and any prior explicit initializers for
any of its subobjects are ignored:"

so it is worth filing a gcc bug asking for a QoI improvement in adding a
warning (since the code does not violate the C standard, but does cause
surprises in the reinitialization of omitted members in the later {} to
go back to 0 in spite of the earlier initialization by nested name).

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org