On 1/29/2021 10:56 AM, Yu, Yu-cheng wrote:

On 1/29/2021 9:07 AM, Dave Hansen wrote:

quoted

On 1/27/21 1:25 PM, Yu-cheng Yu wrote:

quoted

arch_prctl(ARCH_X86_CET_STATUS, u64 *args)
     Get CET feature status.

     The parameter 'args' is a pointer to a user buffer.  The kernel 
returns
     the following information:

     *args = shadow stack/IBT status
     *(args + 1) = shadow stack base address
     *(args + 2) = shadow stack size

[...]

quoted

quoted

+int prctl_cet(int option, u64 arg2)
+{
+    struct cet_status *cet;
+    unsigned int features;
+
+    /*
+     * GLIBC's ENOTSUPP == EOPNOTSUPP == 95, and it does not recognize
+     * the kernel's ENOTSUPP (524).  So return EOPNOTSUPP here.
+     */
+    if (!IS_ENABLED(CONFIG_X86_CET))
+        return -EOPNOTSUPP;

Let's ignore glibc for a moment.  What error code *should* the kernel be
returning here?  errno(3) says:

        EOPNOTSUPP      Operation not supported on socket (POSIX.1)
...
        ENOTSUP         Operation not supported (POSIX.1)

Yeah, other places in kernel use ENOTSUPP.  This seems to be out of 
line.  And since the issue is long-existing, applications already know 
how to deal with it.  I should have made that argument.  Change it to 
ENOTSUPP.

When I make the change, checkpatch says...

WARNING: ENOTSUPP is not a SUSV4 error code, prefer EOPNOTSUPP
#128: FILE: arch/x86/kernel/cet_prctl.c:33:
+		return -ENOTSUPP;

Do we want to reconsider?

[...]