Re: [Fuego] [PATCH] Benchmark.lmbench2: fix lmbench compile issue

See comments below.

> -----Original Message-----
> From: [email protected] <[email protected]>
> 
> From: venkata pyla <[email protected]>
> 
> The test is failing in fuego buster during build phase, it is due to
> the llseek function is no more available from glibc version 2.28,
> instead use the lseek64 function.
> 
> ```
> /usr/bin/ld: /tmp/ccUq4xrV.o: in function `seekto':
> disk.c:(.text+0x2f): undefined reference to `llseek'
> collect2: error: ld returned 1 exit status
> ```
> created a patch for lmbench3 source to fix the above issue,
> and applied during build phase.

Nice description!  It's very clear what you are fixing and why.

I was worried that lseek64 might not be available for some of
my older boards and build environments.  However, according
to the man page, it was introduced in glibc version 2.1
The older version of glibc on any of my boards is 2.23, so
it looks like I'm OK.


> 
> Signed-off-by: venkata pyla <[email protected]>
> ---
>  .../fix-undefined-reference-llseek.patch      | 24 +++++++++++++++++++
>  tests/Benchmark.lmbench2/fuego_test.sh        |  5 ++++
>  2 files changed, 29 insertions(+)
>  create mode 100755 tests/Benchmark.lmbench2/fix-undefined-reference-llseek.patch
> 
> diff --git a/tests/Benchmark.lmbench2/fix-undefined-reference-llseek.patch b/tests/Benchmark.lmbench2/fix-undefined-reference-
> llseek.patch
> new file mode 100755
> index 0000000..99070f8
> --- /dev/null
> +++ b/tests/Benchmark.lmbench2/fix-undefined-reference-llseek.patch
> @@ -0,0 +1,24 @@
> +--- disk.c	2021-08-02 23:59:04.976884974 +0530
> ++++ disk_new.c	2021-08-03 00:01:11.952696624 +0530
> +@@ -7,6 +7,10 @@
> +  * Bits of this are derived from work by Ethan Solomita.
> +  */
> +
> ++#ifdef __linux__
> ++#define _LARGEFILE64_SOURCE
> ++#endif
> ++
> + #include	<stdio.h>
> + #include	<sys/types.h>
> + #include	<unistd.h>
> +@@ -289,9 +293,7 @@
> + seekto(int fd, uint64 off)
> + {
> + #ifdef	__linux__
> +-	extern	loff_t llseek(int, loff_t, int);
> +-
> +-	if (llseek(fd, (loff_t)off, SEEK_SET) == (loff_t)-1) {
> ++	if (lseek64(fd, (off64_t)off, SEEK_SET) == (off64_t)-1) {
> + 		return(-1);
> + 	}
> + 	return (0);
> diff --git a/tests/Benchmark.lmbench2/fuego_test.sh b/tests/Benchmark.lmbench2/fuego_test.sh
> index 9bdf0d2..d07606c 100755
> --- a/tests/Benchmark.lmbench2/fuego_test.sh
> +++ b/tests/Benchmark.lmbench2/fuego_test.sh
> @@ -9,6 +9,7 @@ function test_build {
>     patch -p0 < $TEST_HOME/lmbench3.mem64.patch
>     cd ../src
>     patch -p0 < $TEST_HOME/bench.h.patch
> +   patch -p0 < $TEST_HOME/fix-undefined-reference-llseek.patch
>     cd ..
>     CFLAGS+=" -g -O"
>     export OS=$PREFIX
> @@ -25,7 +26,11 @@ function test_run {
>     if [ -n "$PREFIX" ] ; then
>        LMBENCH_OS=`ls ./bin`
>     else
> +      LMBENCH_OS=`ls ./bin`
> +      echo "lmbench_os: $LMBENCH_OS"
> +      echo "curr path: $(pwd)"
>        LMBENCH_OS=$(scripts/os)
> +      echo "lmbench_os: $LMBENCH_OS"

Nice catch.  This `ls ./bin` construct is IMHO quite weird.

>     fi
>     safe_cmd "rm -rf $BOARD_TESTDIR/fuego.$TESTDIR/results"
>     safe_cmd "cd $BOARD_TESTDIR/fuego.$TESTDIR/scripts; OS=$LMBENCH_OS ./config-run"
> --
> 2.20.1
> 

Thanks for this fix.  This is the type of thing that is a real pain to hunt down
and fix, and is nice to keep the test from bit-rotting.  So I appreciate this
fix a lot.

I found a couple of other issues with the lmbench build and results parsing,
that I'll fix up in the next few days.  These were due to weird kernel configs
and toolchain settings that caused certain bugs to surface.  When I'm done
with the fixes, I'll let you know so you can re-test lmbench with your boards
and make sure I didn't break anything.

Thanks again.  This is applied and pushed.
 -- Tim