On Thu, Jan 21, 2016 at 12:03:42AM +0000, Paul Burton wrote:

On Tue, Jan 19, 2016 at 01:37:50PM +0000, James Hogan wrote:

quoted

When a header file is removed from generic-y (often accompanied by the
addition of an arch specific header), the generated wrapper file will
persist, and in some cases may still take precedence over the new arch
header.

For example commit f1fe2d21f4e1 ("MIPS: Add definitions for extended
context") removed ucontext.h from generic-y in arch/mips/include/asm/,
and added an arch/mips/include/uapi/asm/ucontext.h. The continued use of
the wrapper when reusing a dirty build tree resulted in build failures
in arch/mips/kernel/signal.c:

arch/mips/kernel/signal.c: In function ‘sc_to_extcontext’:
arch/mips/kernel/signal.c:142:12: error: ‘struct ucontext’ has no member named ‘uc_extcontext’
  return &uc->uc_extcontext;
            ^

Fix by detecting and removing wrapper headers in generated header
directories that do not correspond to a filename in generic-y, genhdr-y,
or the newly introduced generated-y.

Reported-by: Jacek Anaszewski <redacted>
Reported-by: Hauke Mehrtens <hauke@hauke-m.de>
Reported-by: Heinrich Schuchardt <redacted>
Signed-off-by: James Hogan <redacted>
Cc: Michal Marek <redacted>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Ralf Baechle <redacted>
Cc: Paul Burton <redacted>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Cc: linux-kbuild@vger.kernel.org
Cc: linux-arch@vger.kernel.org
Cc: linux-mips@linux-mips.org
---
Changes in v2:
- Rewrite a bit, drawing inspiration from Makefile.headersinst.
- Exclude genhdr-y and generated-y (thanks to kbuild test robot).
---
 scripts/Makefile.asm-generic | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/scripts/Makefile.asm-generic b/scripts/Makefile.asm-generic
index 045e0098e962..24c29f16f029 100644
--- a/scripts/Makefile.asm-generic
+++ b/scripts/Makefile.asm-generic

@@ -13,11 +13,26 @@ include scripts/Kbuild.include
 # Create output directory if not already present
 _dummy := $(shell [ -d $(obj) ] || mkdir -p $(obj))
 
+# Stale wrappers when the corresponding files are removed from generic-y
+# need removing.
+generated-y   := $(generic-y) $(genhdr-y) $(generated-y)
+all-files     := $(patsubst %, $(obj)/%, $(generated-y))
+old-headers   := $(wildcard $(obj)/*.h)
+unwanted      := $(filter-out $(all-files),$(old-headers))

Hi James,

Thanks a bunch for fixing this!

FTR, I noticed yesterday it fixes a similar case when switching v4.3 to
v4.4 too:

arch/mips/kernel/../../../fs/binfmt_elf.c In function ‘create_elf_tables’:
./arch/mips/include/asm/elf.h +425 :14: error: ‘AT_SYSINFO_EHDR’ undeclared (first use in this function)
  NEW_AUX_ENT(AT_SYSINFO_EHDR,     \
              ^

Due to commit ebb5e78cc634 ("MIPS: Initial implementation of a VDSO")
adding uapi/asm/auxvec.h and changing generic-y to header-y. Should
ucontext.h be exported via header-y too?

With these patches, it removes the stale file:
  REMOVE  arch/mips/include/generated/uapi/asm/auxvec.h

Though is it my sleepy self or are all-files & old-headers misnomers?
That is, isn't all-files actually a list of headers to be kept, and
old-headers actually the list of all (header) files?

I've followed the naming in Makefile.headersinst. I read all-files as
"all the files we care about" (i.e. its a combination of several sets of
generated files, hence "all") and old-headers as in "existing headers"
(since it won't include files which haven't been generated yet).

all-files could perhaps be renamed new-headers, but that could be
misleading too.

Cheers
James

Thanks,
    Paul

quoted

+
 quiet_cmd_wrap = WRAP    $@
 cmd_wrap = echo "\#include <asm-generic/$*.h>" >$@
 
-all: $(patsubst %, $(obj)/%, $(generic-y))
+quiet_cmd_remove = REMOVE  $(unwanted)
+cmd_remove = rm -f $(unwanted)
+
+all: $(patsubst %, $(obj)/%, $(generic-y)) FORCE
+	$(if $(unwanted),$(call cmd,remove),)
 	@:
 
 $(obj)/%.h:
 	$(call cmd,wrap)
+
+.PHONY: $(PHONY)
+PHONY += FORCE
+FORCE: ;
-- 
2.4.10