* percpu section failure with Gold linker
@ 2012-11-23 20:44 Jan Engelhardt
2012-11-23 21:25 ` Tejun Heo
0 siblings, 1 reply; 2+ messages in thread
From: Jan Engelhardt @ 2012-11-23 20:44 UTC (permalink / raw)
To: tj, cl; +Cc: linux-mm, Linux Kernel Mailing List
Hi,
when compiling a kernel with the gold linker (3.7.0-rc6 26d29d06ea0204,
gcc-4.7 and binutils-2.23 in my case), certain pcpu symbols are
seemingly errneously copied over from .o files to .ko files, leading to
a hard warning during depmod:
gold$ make -j8 LD=gold HOSTLD=gold
gold$ make modules_install INSTALL_MOD_PATH=/tmp/foo
[...]
WARNING: /tmp/foo/lib/modules/3.7.0-rc6-jng6-default+
/kernel/net/rds/rds_tcp.ko needs unknown symbol
__pcpu_scope_rds_tcp_stats
This happens with many modules using percpu; looking at things with nm
reveals:
gold/net/ipv6$ nm ipv6.o | grep __pcpu_
0000000000000000 D __pcpu_unique_ipv6_cookie_scratch
gold/net/ipv6$ nm ipv6.ko | grep __pcpu_
U __pcpu_unique_ipv6_cookie_scratch
On the other hand, in a linux tree built with the original ld (ld.bfd),
things look like:
bfd$ make -j8
[...]
bfd/net/ipv6$ nm ipv6.o | grep pcpu
0000000000000000 D __pcpu_unique_ipv6_cookie_scratch
bfd/net/ipv6$ nm ipv6.ko | grep pcpu
(no result)
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: percpu section failure with Gold linker
2012-11-23 20:44 percpu section failure with Gold linker Jan Engelhardt
@ 2012-11-23 21:25 ` Tejun Heo
0 siblings, 0 replies; 2+ messages in thread
From: Tejun Heo @ 2012-11-23 21:25 UTC (permalink / raw)
To: Jan Engelhardt; +Cc: cl, linux-mm, Linux Kernel Mailing List
Hello, Jan.
On Fri, Nov 23, 2012 at 09:44:02PM +0100, Jan Engelhardt wrote:
> when compiling a kernel with the gold linker (3.7.0-rc6 26d29d06ea0204,
> gcc-4.7 and binutils-2.23 in my case), certain pcpu symbols are
> seemingly errneously copied over from .o files to .ko files, leading to
> a hard warning during depmod:
>
> gold$ make -j8 LD=gold HOSTLD=gold
> gold$ make modules_install INSTALL_MOD_PATH=/tmp/foo
> [...]
> WARNING: /tmp/foo/lib/modules/3.7.0-rc6-jng6-default+
> /kernel/net/rds/rds_tcp.ko needs unknown symbol
> __pcpu_scope_rds_tcp_stats
Yeah, this is from the nasty tricks percpu plays to get percpu decls
working on s390 and alpha. Take a look at
DECLARE/DEFINE_PER_CPU_SECTION() definitions in
include/linux/percpu-defs.h for details.
> This happens with many modules using percpu; looking at things with nm
> reveals:
>
> gold/net/ipv6$ nm ipv6.o | grep __pcpu_
> 0000000000000000 D __pcpu_unique_ipv6_cookie_scratch
> gold/net/ipv6$ nm ipv6.ko | grep __pcpu_
> U __pcpu_unique_ipv6_cookie_scratch
>
> On the other hand, in a linux tree built with the original ld (ld.bfd),
> things look like:
>
> bfd$ make -j8
> [...]
> bfd/net/ipv6$ nm ipv6.o | grep pcpu
> 0000000000000000 D __pcpu_unique_ipv6_cookie_scratch
> bfd/net/ipv6$ nm ipv6.ko | grep pcpu
> (no result)
So, those __pcpu_unique and __pcpu_scope variables are only used so
that multiple copy of them collide with each other and trigger compile
failure (as the actual percpu variable needs to be declared __weak).
They're all put into .discard section, so should go away once it
passes through the linker. gold left reference to that symbol while
discarding the definition. Hmmm... different interpretation of
.discard? Still weird tho. There's nothing dereferencing that
symbol.
Thanks.
--
tejun
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2012-11-23 21:26 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-11-23 20:44 percpu section failure with Gold linker Jan Engelhardt
2012-11-23 21:25 ` Tejun Heo
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).