On Thu, 19 Mar 2026 at 18:14, Bruce Richardson
<[email protected]> wrote:
>
> Fix the test_debug unit test on FreeBSD by:
> * passing no-telemetry to the additional process calls
> * ensuring that we always use the no-huge and no-shconf params
> * removing the dependency on procfs to get the exe and using argv[0]
> instead
>
> Bugzilla ID: 761 (partial fix)
> Fixes: 764bf26873b9 ("add FreeBSD support")
> Cc: [email protected]
>
> Signed-off-by: Bruce Richardson <[email protected]>
> ---
> app/test/process.h | 11 ++++++++---
> app/test/test_debug.c | 6 +++---
> 2 files changed, 11 insertions(+), 6 deletions(-)
>
> diff --git a/app/test/process.h b/app/test/process.h
> index df43966a2a..46b69ce330 100644
> --- a/app/test/process.h
> +++ b/app/test/process.h
> @@ -96,7 +96,7 @@ process_dup(const char *const argv[], int numargs, const
> char *env_value)
> int driver_path_num;
> int argv_num;
> int i, status;
> - char path[32];
> + char *path;
> #ifdef RTE_LIB_PDUMP
> #ifdef RTE_NET_RING
> rte_thread_t thread;
> @@ -188,8 +188,13 @@ process_dup(const char *const argv[], int numargs, const
> char *env_value)
> /* set the environment variable */
> if (setenv(RECURSIVE_ENV_VAR, env_value, 1) != 0)
> rte_panic("Cannot export environment variable\n");
> -
> - strlcpy(path, "/proc/" self "/" exe, sizeof(path));
> +#ifdef RTE_EXEC_ENV_LINUX
> + char path_buf[32];
> + strlcpy(path_buf, "/proc/" self "/" exe, sizeof(path_buf));
> + path = path_buf;
> +#else
> + path = argv_cpy[0];
> +#endif
You can remove the self and exe macros that makes no sense anymore
(the only left users are Linux only).
> if (execv(path, argv_cpy) < 0) {
IIRC, all callers pass an array with args[0] == prgname == original
argv[0] from main.
To convince myself, I ran unit tests in GHA with a dumb change:
diff --git a/app/test/process.h b/app/test/process.h
index df43966a2a..c9205992d2 100644
--- a/app/test/process.h
+++ b/app/test/process.h
@@ -104,6 +104,9 @@ process_dup(const char *const argv[], int numargs,
const char *env_value)
#endif
#endif
+ if (argv[0] != prgname)
+ return -1;
+
pid_t pid = fork();
if (pid < 0)
return -1;
So it seems that we can remove this whole exe path discovery and just
pass argv_cpy[0].
> if (errno == ENOENT) {
> printf("Could not find '%s', is procfs
> mounted?\n",
Mention of procfs makes no sense for FreeBSD.
> diff --git a/app/test/test_debug.c b/app/test/test_debug.c
> index bfda4d1aa3..eda9c3e0e1 100644
> --- a/app/test/test_debug.c
> +++ b/app/test/test_debug.c
> @@ -154,15 +154,15 @@ test_debug(void)
> test_args[2] = "-l";
> test_args[3] = core;
>
> - if (rte_eal_has_hugepages()) {
> + if (rte_eal_has_hugepages() && RTE_EXEC_ENV_IS_LINUX) {
I miss the reason.
Why can't we run with hugepages on FreeBSD?
> test_args[4] = "";
> test_args[5] = "";
> test_args[6] = "";
> test_args[7] = "";
> } else {
> test_args[4] = "--no-huge";
> - test_args[5] = "-m";
> - test_args[6] = "2048";
> + test_args[5] = "-m2048";
> + test_args[6] = "--no-telemetry";
> #ifdef RTE_ARCH_PPC_64
> /* iova=pa is the default, but fails on ppc64 with --no-huge
> */
> test_args[7] = "--iova-mode=va";
> --
> 2.51.0
>
--
David Marchand
