[Linux kernel] static variable with forward declared type

[llvmbot]

Bugzilla Link	4065
Resolution	WONTFIX
Resolved on	Feb 22, 2010 12:47
Version	unspecified
OS	Linux
Blocks	llvm/llvm-bugzilla-archive#4068
Attachments	pre-processed source code
Reporter	LLVM Bugzilla Contributor
CC	@DougGregor,@edwintorok,@efriedma-quic

Extended Description

The Linux kernel source code file kernel/sched.c builds fine with gcc, but fails under clang:

clang -MD -MF kernel/.sched.o.d -nostdinc -isystem include -Iinclude -I/spare/repo/linux-2.6/arch/x86/include -include include/linux/autoconf.h -D__KERNEL__ -Wall -Wundef -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -Werror-implicit-function-declaration -O2 -m64 -march=core2 -mno-red-zone -mcmodel=kernel -funit-at-a-time -maccumulate-outgoing-args -DCONFIG_AS_CFI=1 -DCONFIG_AS_CFI_SIGNAL_FRAME=1 -pipe -Wno-sign-compare -fno-asynchronous-unwind-tables -mno-sse -mno-mmx -mno-sse2 -mno-3dnow -Wframe-larger-than=2048 -fno-stack-protector -fno-omit-frame-pointer -fno-optimize-sibling-calls -pg -Wdeclaration-after-statement -Wno-pointer-sign -fwrapv -fno-dwarf2-cfi-asm -D"KBUILD_STR(s)=#s" -D"KBUILD_BASENAME=KBUILD_STR(sched)" -D"KBUILD_MODNAME=KBUILD_STR(sched)" -c -o kernel/sched.o kernel/sched.c
clang: warning: the clang compiler does not yet support '-pg'
warning: unknown warning option: -Wframe-larger-than=2048
kernel/sched.c:325:8: error: variable has incomplete type 'typeof(struct cfs_rq)' (aka 'struct cfs_rq')
static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp;
^
include/linux/percpu-defs.h:37:2: note: instantiated from:
DEFINE_PER_CPU_SECTION(type, name, "")
^
:44:1: note: instantiated from:
per_cpu__init_cfs_rq
^
In file included from kernel/sched.c:31:
In file included from include/linux/nmi.h:7:
include/linux/sched.h:144:8: note: forward declaration of 'struct cfs_rq'
struct cfs_rq;
^
kernel/sched.c:7760:21: warning: field of variable sized type 'sg' not at the end of a struct or class is a GNU extension
struct sched_group sg;
^
kernel/sched.c:7765:22: warning: field of variable sized type 'sd' not at the end of a struct or class is a GNU extension
struct sched_domain sd;
^
kernel/sched.c:8982:6: error: indirection requires pointer operand ('void' invalid)
&per_cpu(init_cfs_rq, i),
^~~~~~~~~~~~~~~~~~~~~~~
include/asm-generic/percpu.h:53:3: note: instantiated from:
(*SHIFT_PERCPU_PTR(&per_cpu_var(var), per_cpu_offset(cpu)))
^
5 diagnostics generated.
make[1]: *** [kernel/sched.o] Error 1
make: *** [kernel/] Error 2

[lattner]

Reduced testcase for the first issue:

struct foo;
static struct foo x;
struct foo {
int a;
};

It looks like static should be handled the same as tentative definitions. No idea if this is standard or a GNU extension.

[lattner]

The "void" problem is a bogus diagnostic that happens because per_cpu__init_cfs_rq is invalid. The only issue here is the one in comment #​2. I'll fix the bogus error though.

[lattner]

I filed:
http://llvm.org/bugs/show_bug.cgi?id=4070

about the bogus diagnostic.

[efriedma-quic]

It looks like static should be handled the same as tentative definitions. No
idea if this is standard or a GNU extension.

I'm pretty sure the testcase in comment 2 violates C99 6.9.2p3.

[llvmbot]

In file scope (i.e. not in C function), a tentative definition may have an incomplete type.

[DougGregor]

I think Eli is right. C99 sayeth:

If the declaration of an identifier for an object is a tentative definition
and has internal linkage, the declared type shall not be an incomplete type.

static objects have internal linkage.

[llvmbot]

It does appear so in C99 6.9.2.3, but it is also true that gcc "-std=c99 -Wall -pedantic" does not complain about this code :)

[efriedma-quic]

It does appear so in C99 6.9.2.3, but it is also true that gcc "-std=c99 -Wall
-pedantic" does not complain about this code :)

See http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26581#c5 .

[lattner]

Can we reasonably support this as an extension?

[DougGregor]

It should be reasonably easy to support as an extension.

[llvmbot]

Might want to hold off a bit; this might get fixed in the kernel.

[edwintorok]

*** Bug llvm/llvm-bugzilla-archive#4131 has been marked as a duplicate of this bug. ***

[llvmbot]

Code is too grotty, will fix in kernel.

[edwintorok]

mentioned in issue llvm/llvm-bugzilla-archive#4068

[lattner]

mentioned in issue llvm/llvm-bugzilla-archive#4070

[edwintorok]

mentioned in issue llvm/llvm-bugzilla-archive#4131