sparse error: unable to open 'stdarg.h'

sparse error: unable to open 'stdarg.h'

[Timur Tabi]

I'm trying to run sparse on my external module, but sparse complains about not being able 
to find stdarg.h.  I know this bug was supposed to have been fixed back in January, but 
I'm using the latest code, so I can't explain what's wrong.  I've tried this on a couple 
different 2.6 kernels.

Here's the output I get:

make -C /lib/modules/2.6.8-24-smp/source 
SUBDIRS=/root/AMSO1100/software/host/linux/sys/devccil  C=1 V=2
make[2]: Entering directory `/usr/src/linux-2.6.8-24'
   CHECK   /root/AMSO1100/software/host/linux/sys/devccil/devnet.c
include/linux/kernel.h:10:11: error: unable to open 'stdarg.h'
make[3]: *** [/root/AMSO1100/software/host/linux/sys/devccil/devnet.o] Error 1
make[2]: *** [_module_/root/AMSO1100/software/host/linux/sys/devccil] Error 2
make[2]: Leaving directory `/usr/src/linux-2.6.8-24'
make[1]: *** [all] Error 2
make[1]: Leaving directory `/root/AMSO1100/software/host/linux/sys/devccil'
make: *** [build] Error 2

-- 
Timur Tabi
Staff Software Engineer
timur.tabi@ammasso.com

One thing a Southern boy will never say is,
"I don't think duct tape will fix it."
      -- Ed Smylie, NASA engineer for Apollo 13

Re: sparse error: unable to open 'stdarg.h'

[Christopher Li]

It is missing the gcc default include path.

Check your pre-processor.h in sparse to see that match your gcc
include path or not.

If not, remove pre-processor.h and recompile sparse should solve
the problem.

Chris

On Tue, May 17, 2005 at 04:46:04PM -0500, Timur Tabi wrote:
> I'm trying to run sparse on my external module, but sparse complains about 
> not being able to find stdarg.h.  I know this bug was supposed to have been 
> fixed back in January, but I'm using the latest code, so I can't explain 
> what's wrong.  I've tried this on a couple different 2.6 kernels.
> 
> Here's the output I get:
> 
> make -C /lib/modules/2.6.8-24-smp/source 
> SUBDIRS=/root/AMSO1100/software/host/linux/sys/devccil  C=1 V=2
> make[2]: Entering directory `/usr/src/linux-2.6.8-24'
>   CHECK   /root/AMSO1100/software/host/linux/sys/devccil/devnet.c
> include/linux/kernel.h:10:11: error: unable to open 'stdarg.h'
> make[3]: *** [/root/AMSO1100/software/host/linux/sys/devccil/devnet.o] 
> Error 1
> make[2]: *** [_module_/root/AMSO1100/software/host/linux/sys/devccil] Error 
> 2
> make[2]: Leaving directory `/usr/src/linux-2.6.8-24'
> make[1]: *** [all] Error 2
> make[1]: Leaving directory `/root/AMSO1100/software/host/linux/sys/devccil'
> make: *** [build] Error 2
> 
> -- 
> Timur Tabi
> Staff Software Engineer
> timur.tabi@ammasso.com
> 
> One thing a Southern boy will never say is,
> "I don't think duct tape will fix it."
>      -- Ed Smylie, NASA engineer for Apollo 13
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: sparse error: unable to open 'stdarg.h'

[Timur Tabi]

Christopher Li wrote:
> It is missing the gcc default include path.
> 
> Check your pre-processor.h in sparse to see that match your gcc
> include path or not.
> 
> If not, remove pre-processor.h and recompile sparse should solve
> the problem.

Nope, that didn't fix it.  I deleted pre-process.h and re-ran "make", and it created a new 
one:

#define GCC_INTERNAL_INCLUDE "/usr/lib/gcc-lib/i586-suse-linux/3.3.4/include"

vic1:~/sparse-bk # ll /usr/lib/gcc-lib/i586-suse-linux/3.3.4/include/stdarg.h
-rw-r--r--  1 root root 4325 Oct  1  2004 
/usr/lib/gcc-lib/i586-suse-linux/3.3.4/include/stdarg.h

-- 
Timur Tabi
Staff Software Engineer
timur.tabi@ammasso.com

One thing a Southern boy will never say is,
"I don't think duct tape will fix it."
      -- Ed Smylie, NASA engineer for Apollo 13

Re: sparse error: unable to open 'stdarg.h'

[Christopher Li]

On Wed, May 18, 2005 at 09:08:39AM -0500, Timur Tabi wrote:
> 
> Nope, that didn't fix it.  I deleted pre-process.h and re-ran "make", and 
> it created a new one:
> 
> #define GCC_INTERNAL_INCLUDE 
> "/usr/lib/gcc-lib/i586-suse-linux/3.3.4/include"
> 
> vic1:~/sparse-bk # ll 
> /usr/lib/gcc-lib/i586-suse-linux/3.3.4/include/stdarg.h
> -rw-r--r--  1 root root 4325 Oct  1  2004 
> /usr/lib/gcc-lib/i586-suse-linux/3.3.4/include/stdarg.h
> 
That is wired.  Can you try to edit a test.c contain just one line:

#include <stdarg.h>

run sparse on that test.c and see if you get any complain or not?

If you still get complain about file not found. Can you run
"strace -e trace=file ./check test.c" and show me the out put?

Chris

PS, here is what I get:

[chrisl@64m sparse-be]$ strace -e trace=file ./check test.h 
execve("./check", ["./check", "test.h"], [/* 26 vars */]) = 0
open("/etc/ld.so.preload", O_RDONLY)    = -1 ENOENT (No such file or directory)
open("/etc/ld.so.cache", O_RDONLY)      = 3
fstat64(3, {st_mode=S_IFREG|0644, st_size=58983, ...}) = 0
open("/lib/tls/libc.so.6", O_RDONLY)    = 3
fstat64(3, {st_mode=S_IFREG|0755, st_size=1539996, ...}) = 0
open("test.h", O_RDONLY)                = 3
open("/usr/include/stdarg.h", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/usr/local/include/stdarg.h", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/usr/lib/gcc-lib/i386-redhat-linux/3.2.2/include/stdarg.h", O_RDONLY) = 3
[chrisl@64m sparse-be]$ 

Re: sparse error: unable to open 'stdarg.h'

[Timur Tabi]

Christopher Li wrote:

> That is wired.  Can you try to edit a test.c contain just one line:
> 
> #include <stdarg.h>
> 
> run sparse on that test.c and see if you get any complain or not?

I did "sparse test.c" and got no output whatsoever.  No files were created, either.

> 
> If you still get complain about file not found. Can you run
> "strace -e trace=file ./check test.c" and show me the out put?

I get the same thing:

execve("sparse-bk/check", ["sparse-bk/check", "test.c"], [/* 57 vars */]) = 0
access("/etc/ld.so.preload", R_OK)      = -1 ENOENT (No such file or directory)
open("/etc/ld.so.cache", O_RDONLY)      = 3
fstat64(3, {st_mode=S_IFREG|0644, st_size=74426, ...}) = 0
open("/lib/tls/libc.so.6", O_RDONLY)    = 3
fstat64(3, {st_mode=S_IFREG|0755, st_size=1359489, ...}) = 0
open("test.c", O_RDONLY)                = 3
open("/usr/include/stdarg.h", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/usr/local/include/stdarg.h", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/usr/lib/gcc-lib/i586-suse-linux/3.3.4/include/stdarg.h", O_RDONLY) = 3

There must be something specific about how kbuild calls sparse.

-- 
Timur Tabi
Staff Software Engineer
timur.tabi@ammasso.com

One thing a Southern boy will never say is,
"I don't think duct tape will fix it."
      -- Ed Smylie, NASA engineer for Apollo 13

Re: sparse error: unable to open 'stdarg.h'

[Christopher Li]

On Wed, May 18, 2005 at 10:51:08AM -0500, Timur Tabi wrote:
> Christopher Li wrote:
> 
> >That is wired.  Can you try to edit a test.c contain just one line:
> >
> >#include <stdarg.h>
> >
> >run sparse on that test.c and see if you get any complain or not?
> 
> I did "sparse test.c" and got no output whatsoever.  No files were created, 
> either.

The sparse checker does not create files. That means your sparse is
fine. It should locate the stdarg.h

> There must be something specific about how kbuild calls sparse.

I think I know that it is. There is a "-nostdinc" in the sparse
options, which I saw it in the other email you send out. It
drop the internal include path. Gcc is does the same thing.

gcc -c -nostdinc /tmp/test.c
/tmp/test.c:1:22: no include path in which to find stdarg.h

Chris

Re: sparse error: unable to open 'stdarg.h'

[Timur Tabi]

Christopher Li wrote:

> I think I know that it is. There is a "-nostdinc" in the sparse
> options, which I saw it in the other email you send out. It
> drop the internal include path. Gcc is does the same thing.
> 
> gcc -c -nostdinc /tmp/test.c
> /tmp/test.c:1:22: no include path in which to find stdarg.h

That option is set in the a_flags variable.  I'm looking through the kbuild files 
(Makefile, etc) to see why a_flags is being used to build my driver.

As far as I'm concerned, this is a bug in kbuild, and I think it only shows up with 
external modules.

-- 
Timur Tabi
Staff Software Engineer
timur.tabi@ammasso.com

One thing a Southern boy will never say is,
"I don't think duct tape will fix it."
      -- Ed Smylie, NASA engineer for Apollo 13

Re: sparse error: unable to open 'stdarg.h'

[Sam Ravnborg]

On Wed, May 18, 2005 at 11:45:55AM -0500, Timur Tabi wrote:
> Christopher Li wrote:
> 
> >I think I know that it is. There is a "-nostdinc" in the sparse
> >options, which I saw it in the other email you send out. It
> >drop the internal include path. Gcc is does the same thing.
> >
> >gcc -c -nostdinc /tmp/test.c
> >/tmp/test.c:1:22: no include path in which to find stdarg.h
> 
> That option is set in the a_flags variable.  I'm looking through the kbuild 
> files (Makefile, etc) to see why a_flags is being used to build my driver.

Can you post a copy of you makefile. Then I may be able to tell you why.

	Sam

Re: sparse error: unable to open 'stdarg.h'

[Timur Tabi]

Sam Ravnborg wrote:

> Can you post a copy of you makefile. Then I may be able to tell you why.

Ok, but it's really huge and ugly:

# Set some global variables

ifndef SOFTWARE
     # Specifying 'export' turns SOFTWARE into a real environment variable,
     # not just a makefile variable.  This means that when make calls itself
     # recursively in Section 1, the child process will already have SOFTWARE
     # defined, and won't try to re-define it.  In Section 2, the current
     # directory is ${KERNEL_SOURCE}, so calculating SOFTWARE is impossible.
     export SOFTWARE = ${shell cd ../../../..; /bin/pwd}
endif

# Only if the Config.mk file exists will the next line include it.
# (Inside Ammasso, the needed information is derived differently.)
ifneq (${wildcard ${SOFTWARE}/../Config.mk},)
     include ${SOFTWARE}/../Config.mk
endif

ifndef KERNEL_SOURCE
     ${error KERNEL_SOURCE environment variable not defined.}
endif

# O points to where a 2.6 kernel tree put its build targets.  Normally, the
# targets are placed in the same directory as the source, but you can use the
# "O=" option on the 'make' command line to override that.  If the kernel is
# built with the "O=" option, then all external modules must also be built
# with that option.  However, we also use the O= path here to determine where
# to find certain files.

ifdef O
     KERNEL_BUILD=${O}
else
     KERNEL_BUILD=${KERNEL_SOURCE}
endif

# Determine the kernel version.  We only really care about 2.4 vs 2.6, so this
# simple code will work with version.h files that have multiple UTS_RELEASEs.
# Fail if version.h doesn't exist.

ifeq (${wildcard ${KERNEL_BUILD}/include/linux/version.h},)
     ${error ${KERNEL_BUILD}/include/linux/version.h does not exist.}
endif

KERNEL_VERSION=${shell grep -m 1 UTS_RELEASE ${KERNEL_BUILD}/include/linux/version.h | cut 
-f 2 -d'"'}

ifndef KERNEL_VERSION
     ${error Kernel version not found in ${KERNEL_BUILD}/include/linux/version.h.}
endif

ifndef KERNEL_CODE
     ${error KERNEL_CODE environment variable not defined.}
endif

# This makefile is divided into three sections.
# Section 1 is the first pass of a kbuild-style makefile
# Section 2 is the second pass of a kbuild-style makefile
# Section 3 is the tradition 2.4-compatible makefile

# The concept of a two-pass kbuild-style makefile is taken from
# "Driver porting: compiling external modules" http://lwn.net/Articles/21823/

ifneq (${shell expr ${KERNEL_VERSION} : '2.4'}, 0)
     SECTION = 3
else
     ifeq (${KERNELRELEASE},)
         SECTION = 1
     else
         SECTION = 2
     endif
endif

# Check which type of build we want: release, debug, or trace.

ifeq (${CCRELEASE}, 1)
     # Release build
     BUILD_VERSION=release
     EXTRA_CFLAGS += -DCCNOPRINTF
else
     ifeq (${CCDEBUGFAST}, 1)
         # Trace build
         BUILD_VERSION=trace
         EXTRA_CFLAGS += -DCCDEBUG -DCCNOPRINTF
     else
         # Debug build
         BUILD_VERSION=debug
         EXTRA_CFLAGS += -g -DCCDEBUG
     endif
endif

# Set the target directory for output files

TARGET_DIR = obj_${KERNEL_CODE}_${BUILD_VERSION}

# --------------------------  SECTION 1  -----------------------------

# Kbuild pass #1

ifeq (${SECTION}, 1)

# If O= is specified on the make command line, then you must have write
# access to KERNEL_SOURCE, otherwise the build will fail.  So we only pass
# O= if it is specified in Config.mk.

ifdef O
     KO=O=${KERNEL_BUILD}
endif

all:
# We need to manually copy files to the object directory
# because kbuild always dumps the targets in the current directory
	@rm -rf .tmp_versions ${TARGET_DIR}
	@mkdir -p ${TARGET_DIR}
	@${MAKE} -C ${KERNEL_SOURCE} SUBDIRS=${shell pwd} ${KO}
	@mv *.o *.ko *.lst ${TARGET_DIR}
	@rm -f .*.cmd *.mod.c

clean:
	@rm -rf .tmp_versions ${TARGET_DIR}

endif # Section 1

# ----------------------------  COMMON  ------------------------------

# Here we calculate variables that are common to sections 2 and 3

ifneq (${SECTION}, 1)

# Determine where include/asm points to.

TEMP_KERNEL_ARCHDIR=$(subst -, ,${shell cd ${KERNEL_BUILD}/include/asm && pwd -P})
KERNEL_ARCHDIR=${word ${words ${TEMP_KERNEL_ARCHDIR}}, ${TEMP_KERNEL_ARCHDIR}}

# Specify the gcc parameters for this hardware platform.

ifeq (${PLATFORM}, x86_64)
     EXTRA_CFLAGS += -DX86_64 -mno-red-zone -mcmodel=kernel
else
     ifeq (${PLATFORM}, x86_32)
         EXTRA_CFLAGS += -DX86_32
     else
         ${error Unsupported platform}
     endif
endif

# Set the compiler paramaters. AMSO_CFLAGS is an environment variable

EXTRA_CFLAGS += -DEXPORT_SYMTAB -Wall ${AMSO_CFLAGS}
EXTRA_CFLAGS += -I${SOFTWARE}/host/linux/include -I${SOFTWARE}/host/common/include
EXTRA_CFLAGS += -I${SOFTWARE}/common/include -I${SOFTWARE}/common/include/clustercore
EXTRA_CFLAGS += -I${SOFTWARE}/host/linux/common -I${SOFTWARE}/host/linux/sys/devccil

# Define DO_MUNMAP_API_CHANGE if this kernel uses the version of do_unmap()
# that has four parameters instead of just three.

ifneq (${shell grep -c -m 1 do_munmap.*acct ${KERNEL_SOURCE}/include/linux/mm.h}, 0)
     EXTRA_CFLAGS += -DDO_MUNMAP_API_CHANGE
endif

# Define REMAP_API_CHANGE if this kernel uses the version of remap_page_range()
# that has five parameters

ifneq (${shell grep -c -m 1 remap_page_range.*vm_area_struct 
${KERNEL_SOURCE}/include/linux/mm.h}, 0)
     EXTRA_CFLAGS += -DREMAP_API_CHANGE
endif

# Define REMAP_PFN_RANGE if the function remap_pfn_range() exists in mm.h
# This function deprecates remap_page_range().  remap_pfn_range() uses a page
# index rather than a physical address, which allows it to support >4GB of RAM
# on 32-bit systems.

ifneq (${shell grep -c -m 1 remap_pfn_range ${KERNEL_SOURCE}/include/linux/mm.h}, 0)
     EXTRA_CFLAGS += -DREMAP_PFN_RANGE
endif

# Define INCLUDE_SYSCALL if linux/syscall.h exists.

ifneq (${wildcard ${KERNEL_SOURCE}/include/linux/syscall.h},)
     EXTRA_CFLAGS += -DINCLUDE_SYSCALL
endif

# Define INCLUDE_SYSTEM if interrupt.h does not include asm/system.h
# Red Hat 8.0 (and maybe others) has a bug in interrupt.h where it forgets
# to include this header file.

ifeq (${shell grep -c asm/system\.h ${KERNEL_SOURCE}/include/linux/interrupt.h}, 0)
     EXTRA_CFLAGS += -DINCLUDE_SYSTEM
endif

# Define INCLUDE_CURRENT if hw_irq.h does not include sched.h and current.h
# Red Hat 7.3 (and maybe others) has a bug in hw_irq.h where it forgets
# to include these header files.

ifeq (${shell grep -c asm/current\.h 
${KERNEL_SOURCE}/include/asm-${KERNEL_ARCHDIR}/hw_irq.h}, 0)
     EXTRA_CFLAGS += -DINCLUDE_CURRENT
endif

# Define NET_DEVICE_HAS_IW if the net_device structure has a field called "wireless_handlers"

ifneq (${shell grep -cw wireless_handlers ${KERNEL_SOURCE}/include/linux/netdevice.h}, 0)
     EXTRA_CFLAGS += -DNET_DEVICE_HAS_IW
endif

# Define PCI_SAVE_STATE_BUFFER if function pci_save_state() has a parameter called "buffer"

ifneq (${shell grep -c pci_save_state.*buffer ${KERNEL_SOURCE}/include/linux/pci.h}, 0)
     EXTRA_CFLAGS += -DPCI_SAVE_STATE_BUFFER
endif

# Define QDISC_LIST_HEAD if the net_device.qdisc_list structure is of type
# "struct list_head" instead of "struct Qdisc *".  This change was made in 2.6.8

ifneq (${shell grep -cw list_head.*qdisc_list ${KERNEL_SOURCE}/include/linux/netdevice.h}, 0)
     EXTRA_CFLAGS += -DQDISC_LIST_HEAD
endif

# Define PCI_DMA_CPU if the function pci_dma_sync_single_for_cpu() exists.
# If so, then we need to call this function instead of pci_dma_sync_single().

ifneq (${wildcard ${KERNEL_SOURCE}/include/asm-generic/pci-dma-compat.h},)
     ifneq (${shell grep -c pci_dma_sync_single_for_cpu 
${KERNEL_SOURCE}/include/asm-generic/pci-dma-compat.h}, 0)
         EXTRA_CFLAGS += -DPCI_DMA_CPU
     endif
endif

# Define SIGNAL_RLIM if rlim[] is a member of signal_struct instead of
# task_struct.  Since both signal_struct and task_struct are defined in
# sched.h, we need to grep the code for an actual usage of rlim[], i.e.
# "current->signal->rlim[]" as opposed to "current->rlim[]".  It looks like
# mm.h is a good candidate for this check.
ifneq (${shell grep -c -m 1 "signal->rlim" ${KERNEL_SOURCE}/include/linux/mm.h}, 0)
     EXTRA_CFLAGS += -DSIGNAL_RLIM
endif

# Define USE_GUP if get_user_pages() truly pins down pages.  Kernels prior to
# 2.6.7 had a bug in get_user_pages() that would unpin a page under extreme
# memory pressure.
ifneq (${wildcard ${KERNEL_SOURCE}/mm/rmap.c},)
     ifneq (${shell grep -c -m 1 try_to_unmap_one.*vm_area_struct 
${KERNEL_SOURCE}/mm/rmap.c}, 0)
         EXTRA_CFLAGS += -DUSE_GUP
         USE_GUP=1
     endif
endif

ifndef USE_GUP
     # If we can't use get_user_pages(), then we should use mlock().  Normally,
     # we would need a kernel that support non-root mlock(), but the driver
     # actually overrides any limitiations on mlock().
     EXTRA_CFLAGS += -DUSE_MLOCK
     USE_MLOCK=1

     # Define USE_MLOCK if this kernel supports mlock for non-root processes.
     # If so, then we libccil needs to call mlock instead of having the driver
     # do it.  We also skip all the code to determine the sys_mlock() calling method.
#    ifneq (${shell grep -c -m 1 can_do_mlock ${KERNEL_SOURCE}/include/linux/mm.h}, 0)
#        EXTRA_CFLAGS += -DUSE_MLOCK
#        USE_MLOCK=1
#    endif
endif

# If neither USE_MLOCK nor USE_GUP is defined, then we want the driver to
# call sys_mlock for us.
ifndef USE_MLOCK
     ifndef USE_GUP
         EXTRA_CFLAGS += -DUSE_SYSCALL
         USE_SYSCALL=1
     endif
endif

ifdef USE_MLOCK
     SYSCALL_METHOD = "Using mlock() system call"
endif
ifdef USE_GUP
     SYSCALL_METHOD = "Using get_user_pages()"
endif
ifdef USE_SYSCALL
     # Here we determine which method we will use to call sys_mlock().
     # The rule is:

     # If /proc/k[all]syms exists, then scan it for a list of syscalls
     # If /proc/k[all]syms doesn't exists or we can't find at least 2 syscalls,
     #    then scan the kernel source tree for a list of syscalls
     # If that doesn't work either, then look up the sys_mlock entry point in
     #    /boot/System.map-`uname -r`
     # If that also doesn't work, and if we're x86-32, then use KERNEL_SYSCALLS
     # Otherwise, we can't call sys_mlock().  Exit with failure.

     ifneq (${wildcard /proc/ksyms},)
         SYMFILE = /proc/ksyms
     else
         ifneq (${wildcard /proc/kallsyms},)
             SYMFILE = /proc/kallsyms
         endif
     endif

     # A list of system calls we look for
     SYSCALLS = open close read write lseek wait4

     ifdef SYMFILE
         # The kernel symbol file is readable, so let's use it
         SYMLIST = ${shell x=1; for i in ${SYSCALLS}; do if [ "`grep -sh -m 1 --mmap 
\"[^t] sys_$$i\(_R[_[:alnum:]]*\|\)$$\" ${SYMFILE}`" ]; then echo "-DSYSCALL$$x=$$i"; 
x=`expr $$x + 1`; fi; done}
     else
         # The kernel symbol file is not readable, so we need to scan the source tree
         # We also need to search the architecture-specific trees
         ARCH=${KERNEL_SOURCE}/arch/${KERNEL_ARCHDIR}/kernel/*
         SYMLIST = ${shell x=1; for i in ${SYSCALLS}; do if [ "`grep -sh -m 1 --mmap 
EXPORT_SYMBOL[_GPL]*\(sys_$$i ${KERNEL_SOURCE}/fs/* ${KERNEL_SOURCE}/kernel/* ${ARCH}`" ]; 
then echo "-DSYSCALL$$x=$$i"; x=`expr $$x + 1`; fi; done}
     endif
     ifeq (${shell expr `echo ${SYMLIST} | wc -w` \>= 2}, 1)
         # There are enough syscalls, so let's use the list
         EXTRA_CFLAGS += ${SYMLIST}
         SYSCALL_METHOD = "Using system call table method"
     else
         SYMFILE = /boot/System.map-${shell uname -r}
         # There aren't enough syscalls, so let's try System.map-`uname -r
         ifeq (${shell grep -cw "\(sys_mlock\|sys_munlock\)" ${SYMFILE}}, 2)
             # We found sys_mlock and sys_munlock, so we'll use their addresses.
             # The addresses will be passed in on the insmod command-line,
             # by our loader script.
             EXTRA_CFLAGS += -DSYSTEM_MAP
             SYSCALL_METHOD = "Using System.map method"
         else
             # We couldn't find sys_m[un]lock, so search for other syscalls in
             # System.map.  We are assuming that System.map contains entries
             # only for exported functions, not every function in the kernel.
             SYMLIST = ${shell x=1; for i in ${SYSCALLS}; do if [ "`grep -sh -m 1 --mmap 
\"[^t] sys_$$i\(_R[_[:alnum:]]*\|\)$$\" ${SYMFILE}`" ]; then echo "-DSYSCALL$$x=$$i"; 
x=`expr $$x + 1`; fi; done}
             ifeq (${shell expr `echo ${SYMLIST} | wc -w` \>= 2}, 1)
                 # There are enough syscalls, so let's use the list
                 EXTRA_CFLAGS += ${SYMLIST}
                 SYSCALL_METHOD = "Using system call table method from System.map"
             else
                 # Nothing so far has worked, so if we're x86-32 we can use
                 # kernel syscalls, otherwise we give up.
                 ifeq (${PLATFORM}, x86_32)
                     EXTRA_CFLAGS += -DKERNEL_SYSCALLS
                     SYSCALL_METHOD = "Using kernel syscall method"
                 else
                     # None of our options are going to work, so just give up
                     ${error Could not determine system call method}
                 endif
             endif
         endif
     endif
endif

# Support for KDAPL
EXTRA_CFLAGS += -DCCIL_KDAPL -DCCTHREADSAFE

# Source files

CFILES = \
         devnet.c \
         ccilnet.c \
         devccil.c \
         devccil_adapter.c \
         devccil_rnic.c \
         devccil_mem.c \
         devccil_vq.c \
         devccil_eh.c \
         devccil_cq.c \
         devccil_mq.c \
         devccil_pd.c \
         devccil_srq.c \
         devccil_qp.c \
         devccil_mm.c \
         devccil_ep.c \
         devccil_wrappers.c \
         devccil_ae.c \
         devccil_logging.c

COMMON_CFILES   = \
		cc_cq_common.c \
                 cc_mq_common.c \
                 cc_qp_common.c

# --------------------------  SECTION 2  -----------------------------

# Kbuild pass #2

ifeq (${SECTION}, 2)

SOFTWARE = ${shell cd ${SUBDIRS}/../../../..; pwd}

# Generate an assembly listing for each file

EXTRA_CFLAGS += -Wa,-aldh=$*.lst

# Fix the paths for the common .c files
CFILES += $(addprefix ../../common/, ${COMMON_CFILES})

# Linker parameters

obj-m := ccil.o

ccil-objs := ${CFILES:.c=.o}

syscall:
	@echo ${SYSCALL_METHOD}

endif # Section 2

# --------------------------  SECTION 3  -----------------------------

# Kernel 2.4-compatible makefile

ifeq (${SECTION}, 3)

include ${SOFTWARE}/header_gcc.mk

COMMON_OFILES:=$(patsubst %.c, ${SOFTWARE}/host/linux/sys/devccil/${TARGET_DIR}/%.o, 
${COMMON_CFILES})
COMMON_CFILES:=$(addprefix ${SOFTWARE}/host/linux/common/,${COMMON_CFILES})

# Here we try to identify the Scyld kernel, because it has peculiarities.
# If linux/fs.h does not have a prototype for sys_read, but asm/unistd.h does,
# and it's X86_64, then we assume that this is the Scyld kernel.
# This kernel needs to have __KERNEL_SYSCALLS__ defined in order to pick up
# the sys_xxx prototypes.  It also needs MODVERSIONS defined and modversions.h
# needs to be included in every source file.

ifeq (${PLATFORM}, x86_64)
     ifeq (${shell grep -cw sys_read ${KERNEL_SOURCE}/include/linux/fs.h}, 0)
         ifeq (${shell grep -cw -m 1 sys_read 
${KERNEL_SOURCE}/include/asm-${KERNEL_ARCHDIR}/unistd.h}, 1)
             EXTRA_CFLAGS += -D__KERNEL_SYSCALLS__ -DMODVERSIONS -include linux/modversions.h
         endif
     endif
endif

# Reset CFLAGS because we don't like the way header_gcc.mk initializes it

CFLAGS = ${EXTRA_CFLAGS} -O2 -fno-strict-aliasing -D__KERNEL__ -DMODULE 
-I${KERNEL_SOURCE}/include

# Generate an assembly listing for each file

COMMON_CFLAGS := ${CFLAGS}
CFLAGS += -Wa,-aldh=${TARGET_DIR}/$*.lst

# Linker parameters

LDFLAGS += -r -E -d --whole-archive

all: syscall ${TARGET_DIR} ${COMMON_OFILES} ${TARGET_DIR}/ccil.o

syscall:
	@echo ${SYSCALL_METHOD}

${COMMON_OFILES}: ${COMMON_CFILES}
	${CC} -c ${COMMON_CFLAGS} -Wa,-aldh=/$*.lst -o $@ ${SOFTWARE}/host/linux/common/$(notdir 
$*).c

include ${SOFTWARE}/footer_gcc.mk

${TARGET_DIR}/ccil.o: ${OBJECTS} ${COMMON_OFILES}
	${LD} ${LDFLAGS} -o $@ ${OBJECTS} ${COMMON_OFILES}

clean::
	@rm -fr ${TARGET_DIR}

unload:
	@echo "Unloading ccil.o"
	-@rmmod ccil

load: unload
	@echo "Loading ${TARGET_DIR}/ccil.o"
	@./loadccil.bash ${TARGET_DIR}/ccil.o

endif # Section 3

endif # ifneq (${SECTION}, 1)


-- 
Timur Tabi
Staff Software Engineer
timur.tabi@ammasso.com

One thing a Southern boy will never say is,
"I don't think duct tape will fix it."
      -- Ed Smylie, NASA engineer for Apollo 13

Kbuild trick

[Sam Ravnborg]

Hi Timur.
Some tricks/hints.

For both kernel 2.4 and 2.6 you can split up your makefile like this:
makefile <= all the external modules specific part
Makefile <= the kbuild specific part

Recent 2.6 kernels looks for a fine named Kbuild before Makefile
so you can use the file Kbuild for a 2.6 compatible kbuild file, and
Makefile for a 2.4 compatible kbuild file.

 
> # Only if the Config.mk file exists will the next line include it.
> # (Inside Ammasso, the needed information is derived differently.)
> ifneq (${wildcard ${SOFTWARE}/../Config.mk},)
>     include ${SOFTWARE}/../Config.mk
> endif
-include does the same

 
> # Determine the kernel version.  We only really care about 2.4 vs 2.6, so 
> this
> # simple code will work with version.h files that have multiple 
> UTS_RELEASEs.

In the kbuild part of the file you can check for KERNELRELEASE.
And a lot of what is below ought to be in the kbuild part of the file.

 
> # If O= is specified on the make command line, then you must have write
> # access to KERNEL_SOURCE, otherwise the build will fail.  So we only pass
> # O= if it is specified in Config.mk.
O= does not require write access to KERNEL_SOURCE. Thats one one the
main concpets. KERNEL_SOURCE may be owned by whoever.

 
> # Define DO_MUNMAP_API_CHANGE if this kernel uses the version of do_unmap()
> # that has four parameters instead of just three.
> 
> ifneq (${shell grep -c -m 1 do_munmap.*acct 
> ${KERNEL_SOURCE}/include/linux/mm.h}, 0)
>     EXTRA_CFLAGS += -DDO_MUNMAP_API_CHANGE
> endif

A macro can simplify this:

feature = $(if $(shell grep -m 1 $(1) $(srctree)/include/$(2)), $(3))

Usage:
EXTRA_CFLAGS += $(call feature, do_munmap.*acct, linux/mm.h, -DDO_MUNMAP_API_CHANGE)
EXTRA_CFLAGS += $(call feature, remap_page_range.*vm_area_struct, \
                  linux/mm.h, -DREMAP_API_CHANGE)

etc..

> 
>     ifneq (${wildcard /proc/ksyms},)
>         SYMFILE = /proc/ksyms
>     else
>         ifneq (${wildcard /proc/kallsyms},)
>             SYMFILE = /proc/kallsyms
>         endif
>     endif
This looks wrong. Either you shall check kernel source or running
kernel.
But mixing it like this is to call for troubles.


 
> # Source files
> 
> CFILES = \
>         devnet.c \
>         ccilnet.c \
>         devccil.c \
>         devccil_adapter.c \
>         devccil_rnic.c \
>         devccil_mem.c \
>         devccil_vq.c \
>         devccil_eh.c \
>         devccil_cq.c \
>         devccil_mq.c \
>         devccil_pd.c \
>         devccil_srq.c \
>         devccil_qp.c \
>         devccil_mm.c \
>         devccil_ep.c \
>         devccil_wrappers.c \
>         devccil_ae.c \
>         devccil_logging.c
> 
> COMMON_CFILES   = \
> 		cc_cq_common.c \
>                 cc_mq_common.c \
>                 cc_qp_common.c
> 
In kbuild files the usual pattern is to specify the .o files not the
source files. So for anyone familiar with kbuild files this looks wrong.
And be explicit. No reason to hide directories behind the addprefix
macro below. This is not an obscufating contest.
> # Fix the paths for the common .c files
> CFILES += $(addprefix ../../common/, ${COMMON_CFILES})

> # Generate an assembly listing for each file
> 
> EXTRA_CFLAGS += -Wa,-aldh=$*.lst
If needed you can use: make cc_mp_common.lst to let kbuild generate
them. Dunno if it works for external modules btw.

> 
> 
> # Linker parameters
> 
> obj-m := ccil.o
> 
> ccil-objs := ${CFILES:.c=.o}
> 
> syscall:
> 	@echo ${SYSCALL_METHOD}

As you have noticed in another mail - this does not work.
An untested approch would be to use:
.PHONY: syscall
$(obj)/ccil.o: syscall

	Sam

Re: Kbuild trick

[Joel Becker]

On Fri, May 20, 2005 at 09:37:06PM +0200, Sam Ravnborg wrote:
> For both kernel 2.4 and 2.6 you can split up your makefile like this:
> makefile <= all the external modules specific part
> Makefile <= the kbuild specific part

	You could also use our fake-2.6-kbuild-for-2.4 makefile,
retrievable via:

svn cat -r 2205 http://oss.oracle.com/projects/ocfs2/src/trunk/Kbuild-24.make

	Just include it and set your target to build-modules,
clean-modules, or install-modules.

Joel

-- 

To spot the expert, pick the one who predicts the job will take the
longest and cost the most.

Joel Becker
Senior Member of Technical Staff
Oracle
E-mail: joel.becker@oracle.com
Phone: (650) 506-8127

Re: Kbuild trick

[Sam Ravnborg]

On Fri, May 20, 2005 at 04:43:53PM -0700, Joel Becker wrote:
> On Fri, May 20, 2005 at 09:37:06PM +0200, Sam Ravnborg wrote:
> > For both kernel 2.4 and 2.6 you can split up your makefile like this:
> > makefile <= all the external modules specific part
> > Makefile <= the kbuild specific part
> 
> 	You could also use our fake-2.6-kbuild-for-2.4 makefile,
> retrievable via:
> 
> svn cat -r 2205 http://oss.oracle.com/projects/ocfs2/src/trunk/Kbuild-24.make

$ which svn
which: no svn in (/bin:/usr/bin:/.....)

Care to post a copy?

	Sam

Re: Kbuild trick

[randy_dunlap]

On Sat, 21 May 2005 07:12:17 +0200 Sam Ravnborg wrote:

| On Fri, May 20, 2005 at 04:43:53PM -0700, Joel Becker wrote:
| > On Fri, May 20, 2005 at 09:37:06PM +0200, Sam Ravnborg wrote:
| > > For both kernel 2.4 and 2.6 you can split up your makefile like this:
| > > makefile <= all the external modules specific part
| > > Makefile <= the kbuild specific part
| > 
| > 	You could also use our fake-2.6-kbuild-for-2.4 makefile,
| > retrievable via:
| > 
| > svn cat -r 2205 http://oss.oracle.com/projects/ocfs2/src/trunk/Kbuild-24.make
| 
| $ which svn
| which: no svn in (/bin:/usr/bin:/.....)
| 
| Care to post a copy?

svn == Subversion, at http://subversion.tigris.org/

oh, you mean post a copy of Kbuild-24.make; yes, that makes sense.

---
~Randy

Re: Kbuild trick

[Joel Becker]

[-- Attachment #1: Type: text/plain, Size: 712 bytes --]

On Sat, May 21, 2005 at 07:12:17AM +0200, Sam Ravnborg wrote:
> > svn cat -r 2205 http://oss.oracle.com/projects/ocfs2/src/trunk/Kbuild-24.make
> 
> Care to post a copy?

	That would be too logical.  Here you go.  The first file is a
sample Makefile for both 2.6 and 2.4, stripped out of our history.  Any
mistakes are mine.  The second is Kbuild-24.make.  The files expect
KERNELDIR and KERNELINC to be set, and a 2.4 build needs to set GCCINC
as well.  Use your favorite autoconf/anti-autoconf system.

Joel

-- 

"Well-timed silence hath more eloquence than speech."  
         - Martin Fraquhar Tupper

Joel Becker
Senior Member of Technical Staff
Oracle
E-mail: joel.becker@oracle.com
Phone: (650) 506-8127

[-- Attachment #2: Makefile --]
[-- Type: text/plain, Size: 583 bytes --]


obj-m	+= myfs.o
myfs-objs := inode.o super.o

EXTRA_CFLAGS += -DSOMETHING

ifeq ($(KERNELRELEASE),)
# Called as "make" part of makefile.  This executes kbuild

all: build-myfs

clean: clean-myfs

install: install-myfs

ifneq ($(KERNEL_26),)
# This is a 2.6 kenrel, forward to kbuild

build-myfs:
	$(MAKE) -C $(KERNELDIR) M=`pwd` modules

clean-myfs:
	$(MAKE) -C $(KERNELDIR) M=`pwd` clean

install-myfs:
	$(MAKE) -C $(KERNELDIR) M=`pwd` modules_install
else
# This is a 2.4 kernel, use our own

include Kbuild-24.make

endif

else
# Called from 2.6 kbuild

STAMP_DIR = `pwd`

endif

[-- Attachment #3: Kbuild-24.make --]
[-- Type: text/plain, Size: 2610 bytes --]


#
# This is a Makefile for 2.4 external module builds.  Red Hat
# Enterprise Linux does not support the usual:
#
#     make -C /lib/modules/`uname -r`/build SUBDIRS=`pwd` modules
#
# So we must hand-roll our compile lines.  This is in a seperate
# file to keep the regular Makefile readable.
#
# This uses the Makebo schema.  So, the compile parts are filled in.
# All Makefiles using this code should set MODULE_CFLAGS and
# MODULE_DEFINES for any extra cflags/defines needed.
#

WARNINGS = -Wall -Wstrict-prototypes
OPTIMIZE = -O2

KERNEL_DEFINES = -D__KERNEL__ -DMODULE -DLINUX -DEXPORT_SYMTAB
KERNEL_CFLAGS = -pipe -nostdinc -fno-strict-aliasing -fno-common \
	-fomit-frame-pointer
KERNEL_LDADD = -nostdlib

#
# These arch-specific flags would normally be provided by the
# kbuild system.  Hope we got them right.
#

ifeq ($(PROCESSOR),ppc64)
  MACH_CFLAGS += -m64 -fsigned-char -fno-builtin -msoft-float \
	-mminimal-toc
  MACH_LDADD += -m elf64ppc
endif
ifeq ($(PROCESSOR),x86_64)
  MACH_CFLAGS += -m64 -mcmodel=kernel
endif
ifeq ($(PROCESSOR),s390x)
  MACH_CFLAGS += -fno-strength-reduce -fpic
  MACH_LDADD += -m elf64_s390
endif



INCLUDES = -I. -I$(KERNELINC) -I$(GCCINC)
DEFINES = $(KERNEL_DEFINES)
CFLAGS = $(OPTIMIZE) $(KERNEL_CFLAGS) $(MACH_CFLAGS) $(MODVERSIONS) \
	$(WARNINGS) $(EXTRA_CFLAGS) $(CFLAGS_$(*F).o)
LDADD = $(KERNEL_LDADD) $(MACH_LDADD)

#
# Ripping off some kbuild 2.6 tricks
#

# Modules that have multiple parts.
multi-used-m := $(sort $(foreach m, $(obj-m), $(if $(strip $($(m:.o=-objs))), $(m))))
multi-objs-m := $(foreach m, $(multi-used-m), $($(m:.o=-objs)))

# Modules that don't.
single-used-m := $(sort $(filter-out $(multi-used-m), $(obj-m)))
link_multi_deps = $($(@:.o=-objs))

#
# It's impossible, AFAICT, to generate the dependancies for each
# individual object:parts set.  In fact, kbuild 2.6 doesn't try either.
# They use a simpler trick...they force the rebuild like this, but
# call if_changed on each target to avoid the useless rebuilds.
# This is a 2.4 hack, so I don't care that much :-)
#
$(multi-used-m): $(multi-objs-m)
	$(CC) -Wl,-r -o $@ $(LDADD) $(link_multi_deps)

build-modules: $(single-used-m) $(multi-used-m)


# Kbuild uses INSTALL_MOD_PATH for rooted installs.  Makebo uses
# DESTDIR.  Folks really should set INSTALL_MOD_PATH, but in case they
# don't...
INSTALL_MOD_PATH ?= $(DESTDIR)
MODLIB = $(INSTALL_MOD_PATH)$(MODULEDIR)

install-modules: $(obj-m)
	$(TOPDIR)/mkinstalldirs $(MODLIB)/$(INSTALL_MOD_DIR)
	for file in $(obj-m); do \
	  $(INSTALL_DATA) $$file $(MODLIB)/$(INSTALL_MOD_DIR)/$$file; \
	done

clean-modules:
	rm -f *.o *.p *.s

Re: Kbuild trick

[Kedar Sovani]

One of the question that I haven't yet managed to solve properly is
how do we manage kernel modules which have its C files in multiple
underlying directories.

say :
/src/Makefile
/src/main.c
/src/module1/Makefile
/src/module1/module1.c
/src/module1/module_stuff.c

And the 3 C files should be built into a single module at the top level.
I tried something like this in the top level Makefile, but it does not work.

obj-m += mymodule.o
main-objs=main.o module1/

I could use,
main-objs=main.o module1/module1.o module1/module_stuff.o

But I don't think that is a good idea. I looked at
Documentation/kbuild/, but no luck.

Any suggestions?

Thanks,
Kedar.

On 5/21/05, Sam Ravnborg <sam@ravnborg.org> wrote:
> 
> Hi Timur.
> Some tricks/hints.
> 
> For both kernel 2.4 and 2.6 you can split up your makefile like this:
> makefile <= all the external modules specific part
> Makefile <= the kbuild specific part
> 
> Recent 2.6 kernels looks for a fine named Kbuild before Makefile
> so you can use the file Kbuild for a 2.6 compatible kbuild file, and
> Makefile for a 2.4 compatible kbuild file.
> 
> 
> > # Only if the Config.mk file exists will the next line include it.
> > # (Inside Ammasso, the needed information is derived differently.)
> > ifneq (${wildcard ${SOFTWARE}/../Config.mk},)
> >     include ${SOFTWARE}/../Config.mk
> > endif
> -include does the same
> 
> 
> > # Determine the kernel version.  We only really care about 2.4 vs 2.6, so
> > this
> > # simple code will work with version.h files that have multiple
> > UTS_RELEASEs.
> 
> In the kbuild part of the file you can check for KERNELRELEASE.
> And a lot of what is below ought to be in the kbuild part of the file.
> 
> 
> > # If O= is specified on the make command line, then you must have write
> > # access to KERNEL_SOURCE, otherwise the build will fail.  So we only pass
> > # O= if it is specified in Config.mk.
> O= does not require write access to KERNEL_SOURCE. Thats one one the
> main concpets. KERNEL_SOURCE may be owned by whoever.
> 
> 
> > # Define DO_MUNMAP_API_CHANGE if this kernel uses the version of do_unmap()
> > # that has four parameters instead of just three.
> >
> > ifneq (${shell grep -c -m 1 do_munmap.*acct
> > ${KERNEL_SOURCE}/include/linux/mm.h}, 0)
> >     EXTRA_CFLAGS += -DDO_MUNMAP_API_CHANGE
> > endif
> 
> A macro can simplify this:
> 
> feature = $(if $(shell grep -m 1 $(1) $(srctree)/include/$(2)), $(3))
> 
> Usage:
> EXTRA_CFLAGS += $(call feature, do_munmap.*acct, linux/mm.h, -DDO_MUNMAP_API_CHANGE)
> EXTRA_CFLAGS += $(call feature, remap_page_range.*vm_area_struct, \
>                   linux/mm.h, -DREMAP_API_CHANGE)
> 
> etc..
> 
> >
> >     ifneq (${wildcard /proc/ksyms},)
> >         SYMFILE = /proc/ksyms
> >     else
> >         ifneq (${wildcard /proc/kallsyms},)
> >             SYMFILE = /proc/kallsyms
> >         endif
> >     endif
> This looks wrong. Either you shall check kernel source or running
> kernel.
> But mixing it like this is to call for troubles.
> 
> 
> 
> > # Source files
> >
> > CFILES = \
> >         devnet.c \
> >         ccilnet.c \
> >         devccil.c \
> >         devccil_adapter.c \
> >         devccil_rnic.c \
> >         devccil_mem.c \
> >         devccil_vq.c \
> >         devccil_eh.c \
> >         devccil_cq.c \
> >         devccil_mq.c \
> >         devccil_pd.c \
> >         devccil_srq.c \
> >         devccil_qp.c \
> >         devccil_mm.c \
> >         devccil_ep.c \
> >         devccil_wrappers.c \
> >         devccil_ae.c \
> >         devccil_logging.c
> >
> > COMMON_CFILES   = \
> >               cc_cq_common.c \
> >                 cc_mq_common.c \
> >                 cc_qp_common.c
> >
> In kbuild files the usual pattern is to specify the .o files not the
> source files. So for anyone familiar with kbuild files this looks wrong.
> And be explicit. No reason to hide directories behind the addprefix
> macro below. This is not an obscufating contest.
> > # Fix the paths for the common .c files
> > CFILES += $(addprefix ../../common/, ${COMMON_CFILES})
> 
> > # Generate an assembly listing for each file
> >
> > EXTRA_CFLAGS += -Wa,-aldh=$*.lst
> If needed you can use: make cc_mp_common.lst to let kbuild generate
> them. Dunno if it works for external modules btw.
> 
> >
> >
> > # Linker parameters
> >
> > obj-m := ccil.o
> >
> > ccil-objs := ${CFILES:.c=.o}
> >
> > syscall:
> >       @echo ${SYSCALL_METHOD}
> 
> As you have noticed in another mail - this does not work.
> An untested approch would be to use:
> .PHONY: syscall
> $(obj)/ccil.o: syscall
> 
>         Sam
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>

Re: Kbuild trick

[Sam Ravnborg]

On Sat, May 21, 2005 at 04:29:35PM +0530, Kedar Sovani wrote:
> One of the question that I haven't yet managed to solve properly is
> how do we manage kernel modules which have its C files in multiple
> underlying directories.
> 
> say :
> /src/Makefile
> /src/main.c
> /src/module1/Makefile
> /src/module1/module1.c
> /src/module1/module_stuff.c
> 
> And the 3 C files should be built into a single module at the top level.
> I tried something like this in the top level Makefile, but it does not work.
> 
> obj-m += mymodule.o
> main-objs=main.o module1/

This create a number of modules - sometimes a good way to do it.
We have several examples in the kernel using this method.

> 
> I could use,
> main-objs=main.o module1/module1.o module1/module_stuff.o
> 
> But I don't think that is a good idea. I looked at
> Documentation/kbuild/, but no luck.
This is one way to do it.
The other way as used by the kernel is to avoid spreading the .c
files in multiple directories.

	Sam

Re: sparse error: unable to open 'stdarg.h'

[Timur Tabi]

Timur Tabi wrote:

> make -C /lib/modules/2.6.8-24-smp/source 
> SUBDIRS=/root/AMSO1100/software/host/linux/sys/devccil  C=1 V=2

When I replace V=2 with V=1 (don't know how that happened), I get this output:

sparse  -D__i386__=1 -Wp,-MD,/root/AMSO1100/software/host/linux/sys/devccil/.devnet.o.d 
-nostdinc -iwithprefix include -D__KERNEL__ -Iinclude  -Wall -Wstrict-prototypes 
-Wno-trigraphs -fno-strict-aliasing -fno-common -O2 -fomit-frame-pointer  -pipe 
-msoft-float -mpreferred-stack-boundary=2 -funit-at-a-time -fno-unit-at-a-time -march=i586 
-mregparm=3 -mcpu=i686 -Iinclude/asm-i386/mach-generic -Iinclude/asm-i386/mach-default 
-DCCNOPRINTF -DX86_32 -DEXPORT_SYMTAB -Wall  -I/root/AMSO1100/software/host/linux/include 
-I/root/AMSO1100/software/host/common/include  -I/root/AMSO1100/software/common/include 
-I/root/AMSO1100/software/common/include/clustercore 
-I/root/AMSO1100/software/host/linux/common 
-I/root/AMSO1100/software/host/linux/sys/devccil  -DREMAP_API_CHANGE -DINCLUDE_CURRENT 
-DNET_DEVICE_HAS_IW -DPCI_SAVE_STATE_BUFFER -DQDISC_LIST_HEAD -DPCI_DMA_CPU -DUSE_GUP 
-DCCIL_KDAPL -DCCTHREADSAFE 
-Wa,-aldh=/root/AMSO1100/software/host/linux/sys/devccil/devnet.lst  -DMODULE 
-DKBUILD_BASENAME=devnet -DKBUILD_MODNAME=ccil 
/root/AMSO1100/software/host/linux/sys/devccil/devnet.c ;


-- 
Timur Tabi
Staff Software Engineer
timur.tabi@ammasso.com

One thing a Southern boy will never say is,
"I don't think duct tape will fix it."
      -- Ed Smylie, NASA engineer for Apollo 13

Re: sparse error: unable to open 'stdarg.h'

[Sam Ravnborg]

On Tue, May 17, 2005 at 05:39:26PM -0500, Timur Tabi wrote:
> Timur Tabi wrote:
> 
> >make -C /lib/modules/2.6.8-24-smp/source 
> >SUBDIRS=/root/AMSO1100/software/host/linux/sys/devccil  C=1 V=2
> 
> When I replace V=2 with V=1 (don't know how that happened), I get this 
> output:
> 
> sparse  -D__i386__=1 
> -Wp,-MD,/root/AMSO1100/software/host/linux/sys/devccil/.devnet.o.d 
> -nostdinc -iwithprefix include -D__KERNEL__ -Iinclude  -Wall 

Newer kbuild's do not use the -nostdinc -iwithprefix include trick.
Instead they use -nostdinc -isystem `gcc --print-file-name=include`

Wich is a more reliable way to find stdarg.h. Newer sparse understands
this too.

Please post make V=1 output with a newer kernel.

	Sam

Re: sparse error: unable to open 'stdarg.h'

[Timur Tabi]

Sam Ravnborg wrote:

> Newer kbuild's do not use the -nostdinc -iwithprefix include trick.
> Instead they use -nostdinc -isystem `gcc --print-file-name=include`
> 
> Wich is a more reliable way to find stdarg.h. Newer sparse understands
> this too.
> 
> Please post make V=1 output with a newer kernel.

I downloaded and install 2.6.11.10, and everything works.  Thanks!

-- 
Timur Tabi
Staff Software Engineer
timur.tabi@ammasso.com

One thing a Southern boy will never say is,
"I don't think duct tape will fix it."
      -- Ed Smylie, NASA engineer for Apollo 13