On 08/27/2013 04:30 AM, Russell King - ARM Linux wrote:
On Tue, Aug 27, 2013 at 01:14:58PM +0200, Jean Delvare wrote:
On Mon, 26 Aug 2013 15:48:43 -0700, Guenter Roeck wrote:
On Mon, Aug 26, 2013 at 10:47:34PM +0200, Jean Delvare wrote:
On Wed, 7 Aug 2013 15:17:13 +0530, Sachin Kamat wrote:
__initdata should be placed between the variable name and equal
sign for the variable to be placed in the intended section.
Really? With gcc 4.7.2 of openSUSE 12.3/x86-64, I see no difference
with and without this change. pm_dmi_table is in section .init.data in
both cases. So when/where/how does it actually matter?
I see that there are hundreds of other occurrences of this in the
kernel tree, so I admit I have a hard time believing it is actually
wrong, and I would appreciate extra explanations.
There is this:
http://permalink.gmane.org/gmane.linux.ports.arm.kernel/258149
Maybe target and/or compiler version specific ?
Russell, can you explain? Is this an old gcc misbehavior which has been
fixed meanwhile, or...?
Me, I can't see how a section attribute could apply to a variable type
in the first place, so I'd expect gcc to either transparently apply it
to the variable instead (which it apparently does for me) or emit a
warning.
OTOH if the problem is real then a check for it should be added to
checkpatch.pl because I don't think a lot of people know about it. A
quick grep suggests that 29% of the use cases (1260 occurrences) have
it wrong.
If it's gcc misbehaviour, then it's documented gcc misbehaviour, because
my description in the above link comes from the GCC info pages. See:
C Extensions -> Attribute Syntax
This is because attributes can be either function attributes, variable
attributes or type attributes. Which kind they are depends on where
they are placed.
When they are placed immediately after 'struct', 'enum' or 'union', or
immediately after the closing brace of such a declaration, they are a
type attribute. It may be possible to argue that a section-specifying
attribute should be allowable as part of the type, but that doesn't
appear to be the case (and it doesn't error out either.)
The version online at: http://gcc.gnu.org/onlinedocs/gcc/Attribute-Syntax.html
looks very similar to the version I've been reading for my gcc version,
so I assume nothing has changed in this regard.
If I understand the document correctly,
static struct dmi_system_id __initdata pm_dmi_table[] = {
applies the attribute to the type used to declare the variable,
thus the variable will be of type 'struct dmi_system_id __initdata'
and be placed in initdata.
static struct dmi_system_id pm_dmi_table[] __initdata = {
applies the attribute to the variable, with the same result.
So I wonder - is this a real problem ? The result should be the same
in either case.
Guenter
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
