Re: [PATCH] tracing: properly align linker defined symbols

[Sam Ravnborg <sam@ravnborg.org>]

On Sat, Jul 10, 2010 at 03:25:02PM -0700, Linus Torvalds wrote:
> On Fri, Jul 9, 2010 at 11:35 PM, Sam Ravnborg <sam@ravnborg.org> wrote:
> >
> > We define a number of symbols in the linker scipt like this:
> >
> >    __start_syscalls_metadata = .;
> >    *(__syscalls_metadata)
> >
> > But we do not know the alignment of "." when we assign
> > the __start_syscalls_metadata symbol.
> > gcc started to uses bigger alignment for structs (32 bytes),
> > so we saw situations where the linker due to alignment
> > constraints increased the value of "." after the symbol assignment.
> 
> Ok, why not clean this up a bit more, and use a helper macro for this
> pattern. There's a fair number of users of that kind of pattern, so
> that actually removes a few lines.
> 
> Here's an example patch. Untested. Whatever. But just this part
> 
>  1 files changed, 21 insertions(+), 31 deletions(-)
> 
> says to me that it's a good idea, and there are other cases that could
> use the new SYMBOL_SECTION() helper.
> 
> What do people think?

Looks good.
I especially like how we with this standardize on the alignment.
I will make sure a working version hits next merge window.

A few comments.

+#define SYMBOL_SECTION(name, section)          \
+       . = ALIGN(32);                          \
+       VMLINUX_SYMBOL(__start_##section) = .;  \
+       *(name)                                 \
+       VMLINUX_SYMBOL(__stop_##section) = .;

The arguments to this macro is confusing.
Something like this:

    #define SYMBOL_SECTION(section, symbol_suffix)

To encourage people to use the section name as suffix
the __start / __stop variables we could
introduce an additional define:

#define SYMBOL_SECTION(section) SYMBOL_SECTION_SUFFIX(section, section)

#define SYMBOL_SECTION_SUFFIX(section, symbol_suffix) \
+       . = ALIGN(32);                                \
+       VMLINUX_SYMBOL(__start_##symbol_suffix) = .;  \
+       *(section)                                    \
+       VMLINUX_SYMBOL(__stop_##symbol_suffix) = .;


I will update the patch to reflect this (+ the fix you pointed out).
But it will wait until Steven has decided what patch to forward
to fix the discussed regression.

	Sam