Skip to contentSkip to conversation

semctl: Fix 32 bit build

PATCH New Petr Vorel 2 participants 8 messages Jul 30, 2020 – Jul 31, 2020

lore
Share & tools
lore

Petr Vorel

author
#0 lore raw

PATCH 1/1 semctl: Fix 32 bit build

Since dfe45f3cf libltpnewipc started to depend on safe macros from libltp.a, thus order for linking must be -lltpnewipc -lltp.

Default order (where -lltp goes first) caused build failures on 32 bit:

cd testcases/kernel/syscalls/ipc/semctl
gcc -m32 -Wformat -Werror=format-security -Werror=implicit-function-declaration -Werror=return-type -fno-common -g -O2 -fno-strict-aliasing -pipe -Wall -W -Wold-style-definition -I../../../../../include -I../../../../../include -I../../../../../include/old/ -m32 -L../../../../../libs/libltpipc -L../../../../../libs/libltpnewipc -L../../../../../lib semctl08.c -lltp -lltpnewipc -o semctl08 /usr/bin/ld: ../../../../../libs/libltpnewipc/libltpnewipc.a(libnewipc.o): in function `probe_free_addr':
libs/libltpnewipc/libnewipc.c:79: undefined reference to `safe_shmget'
/usr/bin/ld: libs/libltpnewipc/libnewipc.c:81: undefined reference to `safe_shmat'
/usr/bin/ld: libs/libltpnewipc/libnewipc.c:82: undefined reference to `safe_shmdt'
/usr/bin/ld: libs/libltpnewipc/libnewipc.c:83: undefined reference to `safe_shmctl'

NOTE: other uses of -lltpnewipc (msg{ctl,get,snd}, shm{at,ctl}) are not affected.

Fixes dfe45f3cf ("libs/libltpnewipc: Use safe macros")

---
Hi,

Reason why I don't push this fix directly is that
I wonder why msg{ctl,get,snd}, shm{at,ctl} aren't affected and whether we can avoid directly specifying -lltp.

Kind regards,
Petr

 testcases/kernel/syscalls/ipc/semctl/Makefile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/kernel/syscalls/ipc/semctl/Makefile +1 -1
index 99971a7db..2559b4c28 100644
@@ -8,6 +8,6 @@ LTPLIBS = ltpipc ltpnewipc
 include $(top_srcdir)/include/mk/testcases.mk
 
 semctl01 semctl02 semctl03 semctl04 semctl05 semctl06 semctl07: LDLIBS += -lltpipc
-semctl08: LDLIBS += -lltpnewipc
+semctl08: LDLIBS = -lltpnewipc -lltp
 
 include $(top_srcdir)/include/mk/generic_leaf_target.mk
Signature · 1 line
2.27.0
#1 lore raw

PATCH 1/1 semctl: Fix 32 bit build

Hi!

testcases/kernel/syscalls/ipc/semctl/Makefile LTPLIBS = ltpipc ltpnewipc · line 11

Quoted 13 lines
 testcases/kernel/syscalls/ipc/semctl/Makefile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/kernel/syscalls/ipc/semctl/Makefile +1 -1
index 99971a7db..2559b4c28 100644
@@ -8,6 +8,6 @@ LTPLIBS = ltpipc ltpnewipc
 include $(top_srcdir)/include/mk/testcases.mk
 
 semctl01 semctl02 semctl03 semctl04 semctl05 semctl06 semctl07: LDLIBS += -lltpipc
-semctl08: LDLIBS += -lltpnewipc
+semctl08: LDLIBS = -lltpnewipc -lltp

If nothing else this may break things if user passed something in LDLIBS, so it should be:

LDLIBS = -lltpnewipc $(LDLIBS)

And I guess the safest rule would be to add the -lltp* libraries first, because naturally none of the code in LTP but the test depends on these.

Signature · 2 lines
Cyril Hrubis
[email protected]
author
#2 lore raw

PATCH 1/1 semctl: Fix 32 bit build

Hi Cyril,

Quoted 11 lines

Hi!

Quoted 9 lines
 testcases/kernel/syscalls/ipc/semctl/Makefile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/kernel/syscalls/ipc/semctl/Makefile +0 -0
index 99971a7db..2559b4c28 100644
@@ -8,6 +8,6 @@ LTPLIBS = ltpipc ltpnewipc
 include $(top_srcdir)/include/mk/testcases.mk
Quoted 8 lines

testcases/kernel/syscalls/ipc/semctl/Makefile LTPLIBS = ltpipc ltpnewipc · line 11

Quoted 3 lines
testcases/kernel/syscalls/ipc/semctl/Makefile +1 -1
 semctl01 semctl02 semctl03 semctl04 semctl05 semctl06 semctl07: LDLIBS += -lltpipc
-semctl08: LDLIBS += -lltpnewipc
+semctl08: LDLIBS = -lltpnewipc -lltp

If nothing else this may break things if user passed something in LDLIBS, so it should be:

LDLIBS = -lltpnewipc $(LDLIBS)

Thanks! I tried that before, but without ':' before '=':
semctl08: LDLIBS = -lltpnewipc $(LDLIBS)
Makefile:12: *** Recursive variable 'LDLIBS' references itself (eventually). Stop.

Assigning as := fixes that:
-semctl08: LDLIBS += -lltpnewipc
+semctl08: LDLIBS := -lltpnewipc $(LDLIBS)

Sorry for overlooking obvious error.

Quoted 2 lines

And I guess the safest rule would be to add the -lltp* libraries first, because naturally none of the code in LTP but the test depends on these.

Are you're going to fix by changing order somewhere in include/mk/? Or shell I push the fix with your ack?
I'd prefer proper fix so commits like this or 22f510de8 ("Fix static linking with musl-fts") aren't needed any more.

Kind regards,
Petr

#3 lore raw

PATCH 1/1 semctl: Fix 32 bit build

Hi!

Quoted 13 lines
Quoted 4 lines

If nothing else this may break things if user passed something in LDLIBS, so it should be:

LDLIBS = -lltpnewipc $(LDLIBS)

Thanks! I tried that before, but without ':' before '=':
semctl08: LDLIBS = -lltpnewipc $(LDLIBS)
Makefile:12: *** Recursive variable 'LDLIBS' references itself (eventually). Stop.

Assigning as := fixes that:

(quoted diff) +1 -1
-semctl08: LDLIBS += -lltpnewipc
+semctl08: LDLIBS := -lltpnewipc $(LDLIBS)
Quoted 8 lines

Sorry for overlooking obvious error.

Quoted 2 lines

And I guess the safest rule would be to add the -lltp* libraries first, because naturally none of the code in LTP but the test depends on these.

Are you're going to fix by changing order somewhere in include/mk/? Or shell I push the fix with your ack?
I'd prefer proper fix so commits like this or 22f510de8 ("Fix static linking with musl-fts") aren't needed any more.

I wonder what would be the easiest solution here.

The main problem is that these flags are per-testcase defined and are not expanded before we enter rule to build a test. And as we are using implicit rules to compile C code we cannot easily change that.

I guess that we can write down our rules and do whatever we want there though.

Signature · 2 lines
Cyril Hrubis
[email protected]
author
#4 lore raw

PATCH 1/1 semctl: Fix 32 bit build

Hi Cyril,

Quoted 7 lines
Quoted 6 lines
Quoted 2 lines

And I guess the safest rule would be to add the -lltp* libraries first, because naturally none of the code in LTP but the test depends on these.

Are you're going to fix by changing order somewhere in include/mk/? Or shell I push the fix with your ack?
I'd prefer proper fix so commits like this or 22f510de8 ("Fix static linking with musl-fts") aren't needed any more.

Quoted 8 lines

I wonder what would be the easiest solution here.

The main problem is that these flags are per-testcase defined and are not expanded before we enter rule to build a test. And as we are using implicit rules to compile C code we cannot easily change that.

I guess that we can write down our rules and do whatever we want there though.

Thanks for info. Well, I'll probably merge the original fix then.

Kind regards,
Petr

#5 lore raw

PATCH 1/1 semctl: Fix 32 bit build

Hi!

Quoted 9 lines
Quoted 8 lines

I wonder what would be the easiest solution here.

The main problem is that these flags are per-testcase defined and are not expanded before we enter rule to build a test. And as we are using implicit rules to compile C code we cannot easily change that.

I guess that we can write down our rules and do whatever we want there though.

Thanks for info. Well, I'll probably merge the original fix then.

So I've been looking into the problem for a while and due to a make limitations the best bet for a solution would be adding a special variable for the LTP libraries as:

include/mk/testcases.mk +3 -0
index bb22be82e..03937516a 100644
@@ -59,5 +59,8 @@ LDFLAGS += $(addprefix -L$(top_builddir)/libs/lib, $(LTPLIBS))
 
 endif
 
+%: %.c
+	$(CC) $(CFLAGS) $(CPPFLAGS) $(LDFLAGS) $< $(LTPLDLIBS) $(LDLIBS) -o $@
+
 $(LTPLIBS_DIRS) $(APICMDS_DIR) $(LIBLTP_DIR): %:
 	mkdir -p "$@"
testcases/kernel/syscalls/ipc/shmctl/Makefile +2 -2
index 0172bb495..2e0ed0ceb 100644
@@ -10,7 +10,7 @@ shmctl05: LDLIBS += -lrt
 
 include $(top_srcdir)/include/mk/testcases.mk
 
-shmctl01 shmctl02 shmctl03 shmctl04 shmctl05: LDLIBS += -lltpipc
-shmctl06: LDLIBS += -lltpnewipc
+shmctl01 shmctl02 shmctl03 shmctl04 shmctl05: LTPLDLIBS = -lltpipc
+shmctl06: LTPLDLIBS = -lltpnewipc
 
 include $(top_srcdir)/include/mk/generic_leaf_target.mk
Signature · 2 lines
Cyril Hrubis
[email protected]
author
#6 lore raw

PATCH 1/1 semctl: Fix 32 bit build

Hi Metan,

Quoted 32 lines

Hi! > > > I wonder what would be the easiest solution here.

Quoted 7 lines
Quoted 6 lines

The main problem is that these flags are per-testcase defined and are not expanded before we enter rule to build a test. And as we are using implicit rules to compile C code we cannot easily change that.

I guess that we can write down our rules and do whatever we want there though.

Thanks for info. Well, I'll probably merge the original fix then.

So I've been looking into the problem for a while and due to a make limitations the best bet for a solution would be adding a special variable for the LTP libraries as:

include/mk/testcases.mk +3 -0
index bb22be82e..03937516a 100644
@@ -59,5 +59,8 @@ LDFLAGS += $(addprefix -L$(top_builddir)/libs/lib, $(LTPLIBS))
 
 endif
 
+%: %.c
+	$(CC) $(CFLAGS) $(CPPFLAGS) $(LDFLAGS) $< $(LTPLDLIBS) $(LDLIBS) -o $@
+
 $(LTPLIBS_DIRS) $(APICMDS_DIR) $(LIBLTP_DIR): %:
 	mkdir -p "$@"
testcases/kernel/syscalls/ipc/shmctl/Makefile +0 -0
index 0172bb495..2e0ed0ceb 100644

testcases/kernel/syscalls/ipc/shmctl/Makefile shmctl05: LDLIBS += -lrt · line 16

Quoted 9 lines
testcases/kernel/syscalls/ipc/shmctl/Makefile +2 -2
@@ -10,7 +10,7 @@ shmctl05: LDLIBS += -lrt
 
 include $(top_srcdir)/include/mk/testcases.mk
 
-shmctl01 shmctl02 shmctl03 shmctl04 shmctl05: LDLIBS += -lltpipc
-shmctl06: LDLIBS += -lltpnewipc
+shmctl01 shmctl02 shmctl03 shmctl04 shmctl05: LTPLDLIBS = -lltpipc
+shmctl06: LTPLDLIBS = -lltpnewipc
 
 include $(top_srcdir)/include/mk/generic_leaf_target.mk

Nice, it fixes problem here and in testcases/kernel/syscalls/ipc/semctl/Makefile (the original problem, if I also change to LTPLDLIBS).

But various other dependencies across LTP fail:
https://travis-ci.org/github/pevik/ltp/builds/713370406

e.g.
https://travis-ci.org/github/pevik/ltp/jobs/713370415
gcc -L/usr/src/ltp/utils/sctp/func_tests/../lib -L/usr/src/ltp/utils/sctp/func_tests/../testlib -L../../../lib test_fragments_v6.o -lltp -lsctputil -lsctp -lpthread -o test_fragments_v6 /usr/lib/gcc/x86_64-alpine-linux-musl/9.3.0/../../../../x86_64-alpine-linux-musl/bin/ld: /tmp/ccBpdLjB.o: in function `setup':
/usr/src/ltp/testcases/kernel/syscalls/bpf/bpf_map01.c:133: undefined reference to rlimit_bump_memlock' /usr/lib/gcc/x86_64-alpine-linux-musl/9.3.0/../../../../x86_64-alpine-linux-musl/bin/ld: /tmp/ccBpdLjB.o: in function run':
/usr/src/ltp/testcases/kernel/syscalls/bpf/bpf_map01.c:53: undefined reference to `bpf_map_create'
collect2: error: ld returned 1 exit status

https://travis-ci.org/github/pevik/ltp/jobs/713370407
/usr/bin/ld: /tmp/ccaGt1eJ.o: in function `main':
/usr/src/ltp/testcases/network/nfsv4/locks/locktests.c:1012: undefined reference to configureClient' /usr/bin/ld: /usr/src/ltp/testcases/network/nfsv4/locks/locktests.c:1015: undefined reference to getConfiguration' /usr/bin/ld: /usr/src/ltp/testcases/network/nfsv4/locks/locktests.c:1008: undefined reference to configureServer' /usr/bin/ld: /usr/src/ltp/testcases/network/nfsv4/locks/locktests.c:1009: undefined reference to setupClients'

Obviously using LTPLDLIBS doesn't apply here. But I don't see what is missing / wrong.

Kind regards,
Petr

#7 lore raw

PATCH 1/1 semctl: Fix 32 bit build

Hi!

Quoted 9 lines

https://travis-ci.org/github/pevik/ltp/jobs/713370407
/usr/bin/ld: /tmp/ccaGt1eJ.o: in function `main':
/usr/src/ltp/testcases/network/nfsv4/locks/locktests.c:1012: undefined reference to configureClient' /usr/bin/ld: /usr/src/ltp/testcases/network/nfsv4/locks/locktests.c:1015: undefined reference to getConfiguration' /usr/bin/ld: /usr/src/ltp/testcases/network/nfsv4/locks/locktests.c:1008: undefined reference to configureServer' /usr/bin/ld: /usr/src/ltp/testcases/network/nfsv4/locks/locktests.c:1009: undefined reference to setupClients'

Obviously using LTPLDLIBS doesn't apply here. But I don't see what is missing / wrong.

Well it matched our explicit rule instead of (different) implicit one, looks like we have to define a full set, I will send a patchset later on.

Signature · 2 lines
Cyril Hrubis
[email protected]