[src/trunk]: src/sys/dev/ic add debug code to check for completion queue corr...

[jdolecek <jdolecek%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/dc9bb36aae6a
branches:  trunk
changeset: 348468:dc9bb36aae6a
user:      jdolecek <jdolecek%NetBSD.org@localhost>
date:      Wed Oct 19 19:34:31 2016 +0000

description:
add debug code to check for completion queue corruption

diffstat:

 sys/dev/ic/nvme.c    |  25 ++++++++++++++++++++++---
 sys/dev/ic/nvmevar.h |   3 ++-
 2 files changed, 24 insertions(+), 4 deletions(-)

diffs (78 lines):

diff -r b55f63619e94 -r dc9bb36aae6a sys/dev/ic/nvme.c
--- a/sys/dev/ic/nvme.c Wed Oct 19 19:31:23 2016 +0000
+++ b/sys/dev/ic/nvme.c Wed Oct 19 19:34:31 2016 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: nvme.c,v 1.17 2016/10/19 19:31:23 jdolecek Exp $       */
+/*     $NetBSD: nvme.c,v 1.18 2016/10/19 19:34:31 jdolecek Exp $       */
 /*     $OpenBSD: nvme.c,v 1.49 2016/04/18 05:59:50 dlg Exp $ */
 
 /*
@@ -18,7 +18,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: nvme.c,v 1.17 2016/10/19 19:31:23 jdolecek Exp $");
+__KERNEL_RCSID(0, "$NetBSD: nvme.c,v 1.18 2016/10/19 19:34:31 jdolecek Exp $");
 
 #include <sys/param.h>
 #include <sys/systm.h>
@@ -1069,6 +1069,18 @@
                        q->q_cq_phase ^= NVME_CQE_PHASE;
                }
 
+#ifdef DEBUG
+               /*
+                * If we get spurious completion notification, something
+                * is seriously hosed up. Very likely DMA to some random
+                * memory place happened, so just bail out.
+                */
+               if ((intptr_t)ccb->ccb_cookie == NVME_CCB_FREE) {
+                       panic("%s: invalid ccb detected",
+                           device_xname(sc->sc_dev));
+                       /* NOTREACHED */
+               }
+#endif
                rv = 1;
 
                /*
@@ -1314,8 +1326,12 @@
 
        mutex_enter(&q->q_ccb_mtx);
        ccb = SIMPLEQ_FIRST(&q->q_ccb_list);
-       if (ccb != NULL)
+       if (ccb != NULL) {
                SIMPLEQ_REMOVE_HEAD(&q->q_ccb_list, ccb_entry);
+#ifdef DEBUG
+               ccb->ccb_cookie = NULL;
+#endif
+       }
        mutex_exit(&q->q_ccb_mtx);
 
        return ccb;
@@ -1326,6 +1342,9 @@
 {
 
        mutex_enter(&q->q_ccb_mtx);
+#ifdef DEBUG
+       ccb->ccb_cookie = (void *)NVME_CCB_FREE;
+#endif
        SIMPLEQ_INSERT_HEAD(&q->q_ccb_list, ccb, ccb_entry);
        mutex_exit(&q->q_ccb_mtx);
 }
diff -r b55f63619e94 -r dc9bb36aae6a sys/dev/ic/nvmevar.h
--- a/sys/dev/ic/nvmevar.h      Wed Oct 19 19:31:23 2016 +0000
+++ b/sys/dev/ic/nvmevar.h      Wed Oct 19 19:34:31 2016 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: nvmevar.h,v 1.7 2016/10/19 19:31:23 jdolecek Exp $     */
+/*     $NetBSD: nvmevar.h,v 1.8 2016/10/19 19:34:31 jdolecek Exp $     */
 /*     $OpenBSD: nvmevar.h,v 1.8 2016/04/14 11:18:32 dlg Exp $ */
 
 /*
@@ -53,6 +53,7 @@
        /* command context */
        uint16_t                ccb_id;
        void                    *ccb_cookie;
+#define NVME_CCB_FREE  0xbeefdeed
        void                    (*ccb_done)(struct nvme_queue *,
                                    struct nvme_ccb *, struct nvme_cqe *);