On 28/10/17(Sat) 22:18, [email protected] wrote:
> Hello,
>
> Since my bug report, I read the beginning of the man page for crash. I now
> know that the ddb log can be retrieved with the dmesg command at the
> following boot. The log I attach today must be more accurate than the one I
> copied by hand last week. By the way, do I have to launch a trace command for
> every cpu?
>
> I experienced another way to crash my machine with the same error. While
> watching youtube videos with firefox, if I wait for the screen to blank for
> inactivity, the sound still works, but if I want to get the screen up again,
> the system crashes.
>
> I will read the crash man page ahead, and try to find more information from
> the crash dump.
>
> Thanks a lot for your excellent work.
Does the diff below help?
Index: linux_ww_mutex.h
===================================================================
RCS file: /cvs/src/sys/dev/pci/drm/linux_ww_mutex.h,v
retrieving revision 1.1
diff -u -p -r1.1 linux_ww_mutex.h
--- linux_ww_mutex.h 1 Jul 2017 16:14:10 -0000 1.1
+++ linux_ww_mutex.h 28 Oct 2017 22:41:54 -0000
@@ -96,7 +96,7 @@ ww_mutex_is_locked(struct ww_mutex *lock
* Return 1 if lock could be acquired, else 0 (contended).
*/
static inline int
-ww_mutex_trylock(struct ww_mutex *lock) {
+ww_mutex_trylock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) {
int res = 0;
mtx_enter(&lock->lock);
@@ -106,9 +106,16 @@ ww_mutex_trylock(struct ww_mutex *lock)
if (lock->acquired == 0) {
KASSERT(lock->ctx == NULL);
lock->acquired = 1;
+ lock->ctx = ctx;
lock->owner = curproc;
res = 1;
}
+ /*
+ * In case we already hold the ww_mutex, increase a count.
+ */
+ else if (lock->owner == curproc) {
+ res = 1;
+ }
mtx_leave(&lock->lock);
return res;
}
@@ -155,7 +162,7 @@ __ww_mutex_lock(struct ww_mutex *lock, s
* - We are in the slow-path (first lock to obtain).
*
* - No context was specified. We assume a single
- * resouce, so there is no danger of a deadlock.
+ * resource, so there is no danger of a deadlock.
*
* - An `older` process (`ctx`) tries to acquire a
* lock already held by a `younger` process.
Index: drm_modeset_lock.c
===================================================================
RCS file: /cvs/src/sys/dev/pci/drm/drm_modeset_lock.c,v
retrieving revision 1.1
diff -u -p -r1.1 drm_modeset_lock.c
--- drm_modeset_lock.c 1 Jul 2017 16:14:10 -0000 1.1
+++ drm_modeset_lock.c 28 Oct 2017 22:42:19 -0000
@@ -309,7 +309,7 @@ static inline int modeset_lock(struct dr
if (ctx->trylock_only) {
lockdep_assert_held(&ctx->ww_ctx);
- if (!ww_mutex_trylock(&lock->mutex))
+ if (!ww_mutex_trylock(&lock->mutex, &ctx->ww_ctx))
return -EBUSY;
else
return 0;
