Skip to contentSkip to conversation

tools: ynl: work around stale system headers

PATCH Applied Jakub Kicinski 2 participants 2 messages Jun 14, 2023 – Jun 15, 2023

Subsystems (3)
lore
Share & tools
lore

Jakub Kicinski

author maintainer
#0 lore raw

PATCH 1/1 tools: ynl: work around stale system headers

The inability to include the uAPI headers directly in tools/ is one of the bigger annoyances of compiling user space code. Most projects trade the pain for smaller inconvenience of having to copy the headers under tools/include.

In case of netlink headers I think that we can avoid both. Netlink family headers are simple and should be self-contained. We can try to twiddle the Makefile a little to force-include just the family header, and use system headers for the rest.

This works fairly well. There are two warts - for some reason if we specify -include $path/family.h as a compilation flag, the #ifdef header guard does not seem to work. So we need to throw the guard in on the command line as well. Seems like GCC detects that the header is different and tries to include both. Second problem is that make wants hash sign to be escaped or not depending on the version. Sigh.

---
 tools/net/ynl/Makefile.deps      | 28 ++++++++++++++++++++++++++++
 tools/net/ynl/generated/Makefile |  6 ++++--
 tools/net/ynl/samples/Makefile   |  4 +++-
 3 files changed, 35 insertions(+), 3 deletions(-)
 create mode 100644 tools/net/ynl/Makefile.deps
tools/net/ynl/Makefile.deps new file +28 -0
new file mode 100644
index 000000000000..524fc4bb586b
@@ -0,0 +1,28 @@
+# SPDX-License-Identifier: GPL-2.0
+
+# Try to include uAPI headers from the kernel uapi/ path.
+# Most code under tools/ requires the respective kernel uAPI headers
+# to be copied to tools/include. The duplication is annoying.
+# All the family headers should be self-contained. We avoid the copying
+# by selectively including just the uAPI header of the family directly
+# from the kernel sources.
+
+UAPI_PATH:=../../../../include/uapi/
+
+# If the header does not exist at all in the system path - let the
+# compiler fall back to the kernel header via -Idirafter.
+# GCC seems to ignore header guard if the header is different, so we need
+# to specify the -D$(hdr_guard).
+# And we need to define HASH indirectly because GNU Make 4.2 wants it escaped
+# and Gnu Make 4.4 wants it without escaping.
+
+HASH := \#
+
+get_hdr_inc=$(if $(shell echo "$(HASH)include <linux/$(2)>" | \
+			 cpp >>/dev/null 2>/dev/null && echo yes),\
+		-D$(1) -include $(UAPI_PATH)/linux/$(2))
+
+CFLAGS_devlink:=$(call get_hdr_inc,_UAPI_LINUX_DEVLINK_H_,devlink.h)
+CFLAGS_ethtool:=$(call get_hdr_inc,_LINUX_ETHTOOL_NETLINK_H_,ethtool_netlink.h)
+CFLAGS_handshake:=$(call get_hdr_inc,_UAPI_LINUX_HANDSHAKE_H,handshake.h)
+CFLAGS_netdev:=$(call get_hdr_inc,_UAPI_LINUX_NETDEV_H,netdev.h)
tools/net/ynl/generated/Makefile +4 -2
index f15c24893296..f8817d2e56e4 100644
@@ -2,11 +2,13 @@
 
 CC=gcc
 CFLAGS=-std=gnu11 -O2 -W -Wall -Wextra -Wno-unused-parameter -Wshadow \
-	-I../lib/
+	-I../lib/ -idirafter $(UAPI_PATH)
 ifeq ("$(DEBUG)","1")
   CFLAGS += -g -fsanitize=address -fsanitize=leak -static-libasan
 endif
 
+include ../Makefile.deps
+
 YNL_GEN_ARG_ethtool:=--user-header linux/ethtool_netlink.h \
 	--exclude-op stats-get
 
@@ -33,7 +35,7 @@ protos.a: $(OBJS)
 
 %-user.o: %-user.c %-user.h
 	@echo -e "\tCC $@"
-	@$(COMPILE.c) -c -o $@ $<
+	@$(COMPILE.c) $(CFLAGS_$*) -o $@ $<
 
 clean:
 	rm -f *.o
tools/net/ynl/samples/Makefile +3 -1
index 714316cad45f..f2db8bb78309 100644
@@ -1,8 +1,10 @@
 # SPDX-License-Identifier: GPL-2.0
 
+include ../Makefile.deps
+
 CC=gcc
 CFLAGS=-std=gnu11 -O2 -W -Wall -Wextra -Wno-unused-parameter -Wshadow \
-	-I../lib/ -I../generated/
+	-I../lib/ -I../generated/ -idirafter $(UAPI_PATH)
 ifeq ("$(DEBUG)","1")
   CFLAGS += -g -fsanitize=address -fsanitize=leak -static-libasan
 endif
Signature · 1 line
2.40.1
#1 lore raw
This patch was applied to netdev/net-next.git (main) by David S. Miller <[email protected]>:

Hello:

This patch was applied to netdev/net-next.git (main) by David S. Miller <[email protected]>:

Quoted 11 lines · On Tue, 13 Jun 2023 17:28:00 -0700 you wrote:

The inability to include the uAPI headers directly in tools/ is one of the bigger annoyances of compiling user space code. Most projects trade the pain for smaller inconvenience of having to copy the headers under tools/include.

In case of netlink headers I think that we can avoid both. Netlink family headers are simple and should be self-contained. We can try to twiddle the Makefile a little to force-include just the family header, and use system headers for the rest.

[...]

Here is the summary with links:
- [net-next] tools: ynl: work around stale system headers

    https://git.kernel.org/netdev/net-next/c/f0ec58d557d6

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html