Bug 19113

That sounds like its probably VMware's problem. Like most vendors we ship
kernels with modified
VM to do things like 4Gig ram. Do provide some info on the redefinitions you see
though, just in case its our mess.

Here's some further information about the redefinition warnings.

I'm only providing information for compilation of a single .c file in
the first of the three vmware modules, but I've attached the vmmon
module source code to this bug report, so you can try to compile
everything for yourself. I've replaced CC=gcc with CC=kgcc in the
Makefile.

I've compiled driver.c in the vmmon module (make driver.o). Below I've
quoted the compilation command and the compiler output, using the
redhat-kernel source and using a plain 2.2.16 kernel source.

compilation command (make driver.o):
kgcc -O2 -DMODULE -D__KERNEL__ -fno-strength-reduce -pipe -m486 -malign-
loops=2 -malign-jumps=2 -malign-functions=2 -fno-omit-frame-pointer -DCPU=586 -
Wall -Wstrict-prototypes -DVMWARE__FIX_IO_APIC_BASE=FIX_IO_APIC_BASE_0 -
I./include -I./common -I./linux -I./export/include -I/usr/src/linux/include -c 
linux/driver.c

############ Compiler output with redhat-2.2.16-22 kernel source: #########

In file included from /usr/src/linux/include/linux/sched.h:20,
                 from linux/driver.c:21:
/usr/src/linux/include/linux/smp.h:77: warning: `smp_num_cpus' redefined
/usr/src/linux/include/linux/modules/i386_ksyms.ver:28: warning: this is the 
location of the previous definition
/usr/src/linux/include/linux/smp.h:83: warning: `smp_call_function' redefined
/usr/src/linux/include/linux/modules/i386_ksyms.ver:118: warning: this is the 
location of the previous definition
In file included from /usr/src/linux/include/linux/sched.h:74,
                 from linux/driver.c:21:
/usr/src/linux/include/asm/processor.h:96: warning: `cpu_data' redefined
/usr/src/linux/include/linux/modules/i386_ksyms.ver:6: warning: this is the 
location of the previous definition
In file included from /usr/src/linux/include/linux/interrupt.h:51,
                 from linux/driver.c:31:
/usr/src/linux/include/asm/hardirq.h:23: warning: `synchronize_irq' redefined
/usr/src/linux/include/linux/modules/i386_ksyms.ver:138: warning: this is the 
location of the previous definition
In file included from /usr/src/linux/include/linux/interrupt.h:52,
                 from linux/driver.c:31:
/usr/src/linux/include/asm/softirq.h:75: warning: `synchronize_bh' redefined
/usr/src/linux/include/linux/modules/i386_ksyms.ver:142: warning: this is the 
location of the previous definition
{standard input}: Assembler messages:
{standard input}:9: Warning: Ignoring changed section attributes for .modinfo

######## END OF Compiler output with redhat-2.2.16-22 kernel source: #########


############ Compiler output with plain 2.2.16 kernel source: #########

{standard input}: Assembler messages:
{standard input}:9: Warning: Ignoring changed section attributes for .modinfo

######## END OF Compiler output with plain 2.2.16 kernel source: ######

Created attachment 4169 [details]
vmmon kernel module from VMware 2.0.2, Makefile modif. to use kgcc

Justed wanted to note, that I can compile the modules against the redhat 2.2.16-
22 kernel source if I enable SMP support in the kernel source. Maybe this can 
narrow the bug search area down.

With respect to the previous comment: I forgot to mention that I also added -
DSUPPORT_SMP to the command line.

New discoveries:
If I configure the kernel *without* CONFIG_MODVERSIONS and add the switch -
DMODVERSIONS to the compilation command for the vmware module it compiles 
without warnings. Isn't that very strange and counter-intuitive?

Another thing that I'd like to point out, is that the symbols defined in 
i386_ksyms.ver that are being redefined are these (manually retyped here, 
beware of typos):

#define cpu_data _set_ver(cpu_data)

#define smp_num_cpus _set_ver(smp_num_cpus)

#define smp_call_function _set_ver()

#define synchronize_irq _set_ver(synchronize_irq)

#define synchronize_bh _set_ver(synchronize_bh)

I know nothing about it, and haven't looked any further into the source, but it 
seems weird that something like smp_num_cpus should depend on module 
versioning, doesn't it?

The sources come with the kernel set up for an SMP build. If you switch to non
SMP you need to
save the .config file, do a 'make mrproper' then put the .config back, make
oldconfig and build the
tree.

Its a funny of the way the kernel build process works. Its a prob;lem in the
2.2.x 'official' tarballs too and very hard to cure