On Tue, 22 May 2018 19:33:33 +0200
Sebastian Andrzej Siewior [off-list ref] wrote:

On 2018-05-22 13:24:29 [-0400], Steven Rostedt wrote:

quoted

On Tue, 22 May 2018 19:21:16 +0200
Sebastian Andrzej Siewior [off-list ref] wrote:
  

quoted

On 2018-05-22 13:10:04 [-0400], Steven Rostedt wrote:  

quoted

On Thu, 17 May 2018 14:40:06 +0200
Sebastian Andrzej Siewior [off-list ref] wrote:
    

quoted

+static DEFINE_LOCAL_IRQ_LOCK(fpsimd_lock);
 /*
  * Update current's FPSIMD/SVE registers from thread_struct.
  *

@@ -594,6 +595,7 @@ int sve_set_vector_length(struct task_struct *task,
 	 * non-SVE thread.
 	 */
 	if (task == current) {
+		local_lock(fpsimd_lock);
 		local_bh_disable();    

I'm surprised that we don't have a "local_lock_bh()"?    

right. Like the last time when we introduced a global lock with no
locking context? 
  

I meant, we could have local_lock_bh(fpsimd_lock); that would turn into
a local_bh_disable() when !PREEMPT_RT.  

Oh that part. That could be possible I guess. I need to look into the
second part which disables preemption while the FPU is taken.

Did you decide to create a local_lock_bh(lock) function? I don't see it.

And should this be backported to 4.14-rt too? You state you saw this in
4.16-rt, but did you start doing something different then, or did the
kernel change?

Thanks!

-- Steve