The ptrace_v_not_enabled test expects the child to reach its ebreak
before it has used the vector extension. That is not guaranteed when
using fork(), because libc may run child atfork handlers before
returning to the test code. In those cases PTRACE_GETREGSET for
NT_RISCV_VECTOR then succeeds instead of returning ENODATA for
inactive vector state.
Use the raw clone syscall with SIGCHLD to keep fork-like semantics
while bypassing libc's fork wrapper and atfork handler chain.
Cc: Andy Chiu <tchiu@tenstorrent.com>
Assisted-by: Claude:claude-sonnet-4-6
Signed-off-by: Andrew Jones <andrew.jones@oss.qualcomm.com>
Link: https://patch.msgid.link/20260707153827.175245-1-andrew.jones@oss.qualcomm.com
Signed-off-by: Paul Walmsley <pjw@kernel.org>
// SPDX-License-Identifier: GPL-2.0-only
#include <sys/ptrace.h>
+#include <sys/syscall.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <sys/uio.h>
SKIP(return, "Vector not supported");
chld_lock = 1;
- pid = fork();
+ pid = (pid_t)syscall(SYS_clone, SIGCHLD, 0, NULL, 0, NULL);
ASSERT_LE(0, pid)
- TH_LOG("fork: %m");
+ TH_LOG("clone: %m");
if (pid == 0) {
while (chld_lock == 1)