Thread (10 messages) flat view 10 messages, 5 authors, 2014-12-29

Re: [GIT PULL] fbdev fixes for 3.19

From: Rafael J. Wysocki <hidden>
Date: 2014-12-12 15:03:53
Also in: lkml

On Thursday, December 11, 2014 06:59:34 PM Rob Herring wrote:
On Thu, Dec 11, 2014 at 4:58 PM, Rafael J. Wysocki [off-list ref] wrote:
quoted
On Thursday, December 11, 2014 12:55:28 PM Linus Torvalds wrote:
quoted
On Thu, Dec 11, 2014 at 5:41 AM, Tomi Valkeinen [off-list ref] wrote:
quoted
Please pull fbdev changes for 3.19.
Ugh. I've pulled it, but I'm not a huge fan of commit 6d09dc6b74ca
("of.h: Keep extern declaration of of_* variables when !CONFIG_OF").

I guess it works, but it ends up delaying any failures from compile
time to link time, and the linker error messages likely won't be
great.

So I'm cc'ing the OF people and Rafael (who worked on OF/ACPI
unification) to see if they have comments, preferences, or other ways
to solve this.

Quite frankly, to me it smells like non-OF builds should just have an
empty "for_each_child_of_node ()" define instead of playing games with
externs that then end up not existing, and depending on the compiler
DTRT.

Comments?
"for_each_child_of_node ()" is not a problem here AFAICS, because in the
example from the commit changelog the compiler would complain on the
of_chosen instance in the "if ()" statement already anyway if I'm not mistaken.
Really, the "if" should be completely removed here.
for_each_child_of_node should do the right thing for !OF or if
of_chosen is NULL already.
This still is an example of a common pattern:

<statement optimized away for CONFIG_BLAH unset> {
	<some code>
}

Now, if you use variables defined only for CONFIG_BLAH set in the above (block
or statement), you'll get compile errors for !CONFIG_BLAH.

There is a limited choice of what can be done to avoid those errors:
 (a) Put the whole code in question under #ifdef CONFIG_BLAH (defeats the
     purpose of the entire exercise).
 (b) Define the variables unconditionally (wasted memory for !CONFIG_BLAH).
 (c) Wrap the variables in static inline functions that will be empty for
     !CONFIG_BLAH ("one more level of indirection" approach).
 (d) Do dirty tricks like in commit 6d09dc6b74ca.

Are there any more options that I'm overlooking?  If not, (a) and (b) are
unattractive, so the real choice seems to be between (c) and (d) to me.

Rafael
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help