On Tue, Sep 29, 2020 at 10:44:31AM -0700, John Hubbard wrote:

On 9/29/20 9:35 AM, Jason Gunthorpe wrote:

quoted

On Mon, Sep 28, 2020 at 01:10:24PM -0700, John Hubbard wrote:

quoted

On 9/28/20 5:57 AM, Jason Gunthorpe wrote:

quoted

On Sun, Sep 27, 2020 at 11:21:53PM -0700, John Hubbard wrote:

quoted

diff --git a/tools/testing/selftests/vm/Makefile b/tools/testing/selftests/vm/Makefile
index d1ae706d9927..9cc6bc087461 100644
+++ b/tools/testing/selftests/vm/Makefile

@@ -130,3 +130,5 @@ endif
   $(OUTPUT)/userfaultfd: LDLIBS += -lpthread
   $(OUTPUT)/mlock-random-test: LDLIBS += -lcap
+
+$(OUTPUT)/gup_test: ../../../../mm/gup_test.h

There is no reason to do this, the auto depends will pick up header
files, and gup_test.h isn't a generated file

Oh, I misread your comment! You were talking about this Makefile
dependency that I'm adding, rather than the ../'s in the path.

Well, for that though, it also has to stay as shown in this patch,
because of this:

I don't see any "gcc -m" type of dependency generation pass happening
in this relatively simple Make system. 

It happens with -MD, all the deps are stored in files like mm/.init-mm.o.cmd
and sucked into the build.

And so, without including an explicit header file dependency (at
least, that's the simplest way), changes to gup_test.h are not
detected.

Shouldn't be

Both the Makefile code and the observed behavior back this up. (I
expect that this is because there is less use of header files in
this area, because most unit tests are self-contained within a
single .c file.)

Something else is very wrong then.

Jason