[Bug 262587] panic: lock (sleep mutex) umtxql not locked @ /usr/src/sys/sys/umtxvar.h:262 on 14-current master-n253798-8cdecdecb43

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=262587

--- Comment #18 from Alexander Motin <mav@FreeBSD.org> ---
OK.  I still don't understand how can it be safe to change the lock when there
seems to be a window between umtxq_getchain() and mtx_lock() inside the
umtxq_lock(), but what would you say about this patch:

--- a/sys/kern/kern_umtx.c
+++ b/sys/kern/kern_umtx.c
@@ -808,8 +808,10 @@ umtxq_sleep(struct umtx_q *uq, const char *wmesg,
                        if (error != 0)
                                break;
                }
-               error = msleep_sbt(uq, &uc->uc_lock, PCATCH, wmesg,
+               error = msleep_sbt(uq, &uc->uc_lock, PCATCH | PDROP, wmesg,
                    sbt, 0, flags);
+               uc = umtxq_getchain(&uq->uq_key);
+               mtx_lock(&uc->uc_lock);
                if (error == EINTR || error == ERESTART)
                        break;
                if (error == EWOULDBLOCK && (flags & C_ABSOLUTE) != 0) {

It seems even before my change code could try on the old lock due to uc never
being updated, so I did it this way instead of using umtxq_lock().

-- 
You are receiving this mail because:
You are the assignee for the bug.