Date:   Tue, 28 Jun 2022 08:55:24 +0800
From:   Zhouyi Zhou <zhouzhouyi@...il.com>
To:     Masahiro Yamada <masahiroy@...nel.org>,
        linux-kernel <linux-kernel@...r.kernel.org>,
        rcu <rcu@...r.kernel.org>,
        "Paul E. McKenney" <paulmck@...nel.org>,
        linux-kbuild@...r.kernel.org, linux-modules@...r.kernel.org,
        linux@...ck-us.net
Subject: Possible problem in check-local-export during the kernel build
 process (RCU torture)

Dear Masahiro:

1. The cause of the problem
When I am doing RCU torture test:

#git clone https://kernel.source.codeaurora.cn/pub/scm/linux/kernel/git/paulmck/linux-rcu.git
#cd linux-rcu
#git checkout remotes/origin/pmladek.2022.06.15a
#./tools/testing/selftests/rcutorture/bin/torture.sh

The kernel building report error something in both Dell PowerEdge R720
and Thinkpad T14 (Amd).

For example:
/mnt/rcu/linux-rcu/tools/testing/selftests/rcutorture/res/2022.06.27-10.42.37-torture#
find . -name Make.out|xargs grep Error
./results-rcutorture-kasan/RUDE01/Make.out:make[2]: ***
[scripts/Makefile.build:257: kernel/trace/power-traces.o] Error 255

2. I trace the problem to check-local-export
I add some echo statement in Makefile.build
diff --git a/scripts/Makefile.build b/scripts/Makefile.build
index 1f01ac65c0cd..0d48a2d3efff 100644
--- a/scripts/Makefile.build
+++ b/scripts/Makefile.build
@@ -227,13 +227,21 @@ cmd_check_local_export =
$(srctree)/scripts/check-local-export $@

 define rule_cc_o_c
        $(call cmd_and_fixdep,cc_o_c)
+       echo "after fixdep $$? $<"
        $(call cmd,gen_ksymdeps)
+       echo "after gen_ksymdeps $$? $<"
        $(call cmd,check_local_export)
+       echo "after check_local_export $$? $<"
        $(call cmd,checksrc)
+       echo "after checksrc $$? $<"
        $(call cmd,checkdoc)
+       echo "after checkdoc $$? $<"
        $(call cmd,gen_objtooldep)
+       echo "after gen_objtooldep $$? $<"
        $(call cmd,gen_symversions_c)
+       echo "after gen_symversions $$? $<"
        $(call cmd,record_mcount)
+       echo "after record_mcount $$? $<"
 endef

Then I rerun the torture.sh

The result show it is check_local_export did not continue in all failed builds.

3. I trace into wait statement in check-local-export
diff --git a/scripts/check-local-export b/scripts/check-local-export
index da745e2743b7..d35477d95bdc 100755
--- a/scripts/check-local-export
+++ b/scripts/check-local-export
@@ -12,7 +12,7 @@ declare -A symbol_types
 declare -a export_symbols

 exit_code=0
-
+echo "check-local-export L15 ${0} ${1}"
 while read value type name
 do
        # Skip the line if the number of fields is less than 3.
@@ -50,9 +50,10 @@ do
        #   done < <(${NM} --quiet ${1})
 done < <(${NM} ${1} 2>/dev/null || { echo "${0}: ${NM} failed" >&2; false; } )

+echo "check-local-export L53 ${0} ${1}"
 # Catch error in the process substitution
 wait $!
-
+echo "check-local-export L56 ${0} ${1} $! $?"
 for name in "${export_symbols[@]}"
 do
        # nm(3) says "If lowercase, the symbol is usually local"
@@ -61,5 +62,9 @@ do
                exit_code=1
        fi
 done
-
+if [ ${exit_code} -ne 0 ] ; then
+    echo "Zhouyi Zhou"
+    echo ${exit_code}
+fi
+echo "check-local-export L69 $? ${exit_code}"
 exit ${exit_code}

Then I rerun the torture.sh

The result show it is wait $! in all failed builds because in all failed cases,
there is L53, but no L56

4. I look into source code of wait command
#wget http://ftp.gnu.org/gnu/bash/bash-5.0.tar.gz
#tar zxf tar zxf bash-5.0.tar.gz
I found that there are QUIT statements in realization of function wait_for

5. My Guess
wait statement in check-local-export may cause bash to quit
I am very interested in this problem, but I am a rookie, I am very
glad to proceed the investigation with your further directions.

Sorry to have brought you so much trouble.


Kind Regard
Thank you very much
Zhouyi

Powered by blists - more mailing lists