On 10/17/24 11:49 PM, Lorenzo Stoakes wrote:
On Thu, Oct 17, 2024 at 02:45:43PM -0700, John Hubbard wrote:
On 10/17/24 2:05 PM, Lorenzo Stoakes wrote:
...
Your include path above actually refers to:
$(top_srcdir)/include/uapi/linux/fcntl.h
...but what I was intending was to copy a snapshot of that file (or a
snapshot from the one generated by "make headers"), to here:
$(top_srcdir)/tools/include/uapi/linux/fcntl.h
Yeah my first version of this used the uapi one but I thought doing that
might conflict with snapshotting? Also it'd mean you'd absolutely have to
have the $(TOOLS_INCLUDES) earlier in the include priority list and better
maybe to special case in this instance.
Actually, I think the goal is to just stop using KHDR_INCLUDES (./usr/include)
entirely!
More below...
...and then use $(TOOLS_INCLUDES), which is already in selftests/lib.mk,
for that reason: to be available to all of the kselftests:
TOOLS_INCLUDES := -isystem $(top_srcdir)/tools/include/uapi
The reasoning for this directory is further explained here:
tools/include/uapi/README
(And I see that selftests/proc has started using $(TOOLS_INCLUDES), that's
progress.)
And now, it's possible to change fcntl.h in place, instead of using a wrapper.
Although either way seems OK to me. (I'm sort of ignoring the details of
the actual header file conflict itself, for now.)
The fcntl.h and linux/fcntl.h conflict is apparently a rather well-known
horror show. It's a difficult one to resolve as the UAPI pidfd.h header
needs O_xxx defines but we also need to include this header in kernel code.
An #ifdef __KERNEL__ block might be a solution here but fixing that is out
of scope for these changes.
Certainly out of scope! Your patch already avoids the biggest issue: it no
longer requires "make headers", in order to build it. That's fine for now. Sorry
to put you into the middle of a pre-existing kselftests debate.
And the #ifdef __KERNEL__ sounds like a potential solution, or at least a
building block for one. I need to take a closer look at this particular header
file mess, the fcntl.h situation is new to me.
+#endif /* _TOOLS_LINUX_PIDFD_H */
diff --git a/tools/testing/selftests/pidfd/Makefile b/tools/testing/selftests/pidfd/Makefile
index d731e3e76d5b..f5038c9dae14 100644
--- a/tools/testing/selftests/pidfd/Makefile
+++ b/tools/testing/selftests/pidfd/Makefile
@@ -1,8 +1,7 @@
# SPDX-License-Identifier: GPL-2.0-only
-CFLAGS += -g $(KHDR_INCLUDES) -pthread -Wall
+CFLAGS += -g -isystem $(top_srcdir)/tools/include $(KHDR_INCLUDES) -pthread -Wall
Instead, it would look like this, which now mostly matches selftests/mm/Makefile,
which is also helpful, because eventually this can be factored into a common
piece for all selftests:
CFLAGS += -g -isystem $(KHDR_INCLUDES) $(TOOLS_INCLUDES) -pthread -Wall
I apologize for just now noticing this! And these kselftests shouldn't require
so much fussing around, I know. But once we get this just right, it will work
well and last a long time. :)
Yeah I know, but this won't work due to the header conflict, I was doing
this previously.
Also doing it this way means that uapi snapshot doesn't override the usr/
one if you have that, which I guess you want?
Actually, given that we want (or should want, so I claim) to build without first
running "make headers", and given that "make headers" populates ./usr/include/,
which in turn is what $(KHDR_INCLUDES) points to, this means that eventually we
should end up with approximately:
CFLAGS += -g -isystem $(TOOLS_INCLUDES) -pthread -Wall
And I just checked, today's selftests/mm builds just fine, with a similar
diff applied, so I'm not totally crazy:
diff --git a/tools/testing/selftests/mm/Makefile b/tools/testing/selftests/mm/Makefile
index 02e1204971b0..b004a8edcba5 100644
--- a/tools/testing/selftests/mm/Makefile
+++ b/tools/testing/selftests/mm/Makefile
@@ -33,7 +33,7 @@ endif
# LDLIBS.
MAKEFLAGS += --no-builtin-rules
-CFLAGS = -Wall -I $(top_srcdir) $(EXTRA_CFLAGS) $(KHDR_INCLUDES) $(TOOLS_INCLUDES)
+CFLAGS = -Wall -I $(top_srcdir) $(EXTRA_CFLAGS) $(TOOLS_INCLUDES)
LDLIBS = -lrt -lpthread -lm
TEST_GEN_FILES = cow
thanks,
--
John Hubbard
