* Re: 2.6.13-rt3
@ 2005-09-01 16:24 Steven Rostedt
2005-09-01 16:46 ` 2.6.13-rt3 Steven Rostedt
0 siblings, 1 reply; 5+ messages in thread
From: Steven Rostedt @ 2005-09-01 16:24 UTC (permalink / raw)
To: Ingo Molnar; +Cc: LKML
Ingo,
Here's a patch to fix some of the problems when defining ALL_TASKS_PI.
The pi_setprio logic is currently incorrect. This should fix that. I
converted ALL_TASKS_PI to a constant number, so that it can be used in
if statements.
-- Steve
Note: I compiled this, but I haven't run it yet. I'll run it right after
I send this note and respond how it worked.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Index: linux_realtime_goliath/kernel/rt.c
===================================================================
--- linux_realtime_goliath/kernel/rt.c (revision 314)
+++ linux_realtime_goliath/kernel/rt.c (working copy)
@@ -104,7 +104,7 @@
* in the system fall under PI handling. Normally only SCHED_FIFO/RR
* tasks are PI-handled:
*/
-#define ALL_TASKS_PI
+#define ALL_TASKS_PI 1
#ifdef CONFIG_RT_DEADLOCK_DETECT
# define __EIP_DECL__ , unsigned long eip
@@ -663,7 +663,7 @@
#endif
-#if defined(ALL_TASKS_PI) && defined(CONFIG_RT_DEADLOCK_DETECT)
+#if ALL_TASKS_PI && defined(CONFIG_RT_DEADLOCK_DETECT)
static void
check_pi_list_present(struct rt_mutex *lock, struct rt_mutex_waiter *waiter,
@@ -674,7 +674,6 @@
__raw_spin_lock(&old_owner->task->pi_lock);
TRACE_WARN_ON_LOCKED(plist_empty(&waiter->pi_list));
- TRACE_WARN_ON_LOCKED(lock_owner(lock));
plist_for_each(curr1, &old_owner->task->pi_waiters) {
w = plist_entry(curr1, struct rt_mutex_waiter, pi_list);
@@ -851,7 +850,7 @@
__raw_spin_lock(&l->wait_lock);
TRACE_BUG_ON_LOCKED(!lock_owner(l));
- if (rt_task(p) && plist_empty(&w->pi_list)) {
+ if ((ALL_TASKS_PI || rt_task(p)) && plist_empty(&w->pi_list)) {
TRACE_BUG_ON_LOCKED(was_rt);
plist_init(&w->pi_list, prio);
plist_add(&w->pi_list, &lock_owner(l)->task->pi_waiters);
@@ -868,7 +867,7 @@
* (TODO: this can be unfair to SCHED_NORMAL tasks if they
* get PI handled.)
*/
- if (!rt_task(p) && !plist_empty(&w->pi_list)) {
+ if (!ALL_TASKS_PI && !rt_task(p) && !plist_empty(&w->pi_list)) {
TRACE_BUG_ON_LOCKED(!was_rt);
plist_del(&w->pi_list, &lock_owner(l)->task->pi_waiters);
plist_del(&w->list, &l->wait_list);
@@ -970,7 +969,7 @@
*/
TRACE_BUG_ON_LOCKED(!spin_is_locked(&task->pi_lock));
TRACE_BUG_ON_LOCKED(!spin_is_locked(&lock->wait_lock));
-#ifndef ALL_TASKS_PI
+#if !ALL_TASKS_PI
if (!rt_task(task)) {
plist_add(&waiter->list, &lock->wait_list);
set_lock_owner_pending(lock);
@@ -1083,7 +1082,7 @@
#endif
trace_special_pid(waiter->ti->task->pid, waiter->ti->task->prio, 0);
-#ifdef ALL_TASKS_PI
+#if ALL_TASKS_PI
check_pi_list_present(lock, waiter, old_owner);
#endif
new_owner = waiter->ti;
@@ -1543,7 +1542,7 @@
list_del_init(&lock->held_list);
#endif
-#ifdef ALL_TASKS_PI
+#if ALL_TASKS_PI
if (plist_empty(&lock->wait_list))
check_pi_list_empty(lock, lock_owner(lock));
#endif
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: 2.6.13-rt3
2005-09-01 16:24 2.6.13-rt3 Steven Rostedt
@ 2005-09-01 16:46 ` Steven Rostedt
2005-09-01 19:38 ` 2.6.13-rt3 Steven Rostedt
0 siblings, 1 reply; 5+ messages in thread
From: Steven Rostedt @ 2005-09-01 16:46 UTC (permalink / raw)
To: Ingo Molnar; +Cc: LKML
On Thu, 2005-09-01 at 12:24 -0400, Steven Rostedt wrote:
>
> Note: I compiled this, but I haven't run it yet. I'll run it right after
> I send this note and respond how it worked.
>
I'm currently sending this message while running the kernel with this
patch. But I haven't compiled with RT_DEADLOCK_DETECT. Instead I'm
recompiling with my changes to turn on all TRACE_{WARN,BUG} to be real
WARN and BUG without the deadlock detect code. This way I can run it
without the trace_lock helping.
-- Steve
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: 2.6.13-rt3
2005-09-01 16:46 ` 2.6.13-rt3 Steven Rostedt
@ 2005-09-01 19:38 ` Steven Rostedt
2005-09-01 20:28 ` 2.6.13-rt3 Ingo Molnar
0 siblings, 1 reply; 5+ messages in thread
From: Steven Rostedt @ 2005-09-01 19:38 UTC (permalink / raw)
To: Ingo Molnar; +Cc: LKML
Ingo,
I just found a __MAJOR__ bug in my code. Below is the patch that fixes
this bug, zaps the WARN_ON in check_pi_list_present, and changes
ALL_TASKS_PI to a booleon instead of just a define.
The major bug was in __down_trylock. See anything wrong with this
code :-) I'm surprised that this worked as well as it did!
if (likely(!old_owner) || __grab_lock(lock, task, old_owner->task)) {
/* granted */
TRACE_WARN_ON_LOCKED(!plist_empty(&lock->wait_list) && !old_owner);
if (old_owner) {
__raw_spin_lock(&old_owner->task->pi_lock);
set_new_owner(lock, old_owner, ti __EIP__);
__raw_spin_unlock(&old_owner->task->pi_lock);
} else
set_new_owner(lock, old_owner, ti __EIP__);
__raw_spin_unlock(&lock->wait_lock);
__raw_spin_unlock(&task->pi_lock);
ret = 1;
}
__raw_spin_unlock(&lock->wait_lock);
__raw_spin_unlock(&task->pi_lock);
Here's the patch:
-- Steve
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Index: linux_realtime_goliath/kernel/rt.c
===================================================================
--- linux_realtime_goliath/kernel/rt.c (revision 314)
+++ linux_realtime_goliath/kernel/rt.c (working copy)
@@ -104,7 +104,7 @@
* in the system fall under PI handling. Normally only SCHED_FIFO/RR
* tasks are PI-handled:
*/
-#define ALL_TASKS_PI
+#define ALL_TASKS_PI 1
#ifdef CONFIG_RT_DEADLOCK_DETECT
# define __EIP_DECL__ , unsigned long eip
@@ -663,7 +663,7 @@
#endif
-#if defined(ALL_TASKS_PI) && defined(CONFIG_RT_DEADLOCK_DETECT)
+#if ALL_TASKS_PI && defined(CONFIG_RT_DEADLOCK_DETECT)
static void
check_pi_list_present(struct rt_mutex *lock, struct rt_mutex_waiter *waiter,
@@ -674,7 +674,6 @@
__raw_spin_lock(&old_owner->task->pi_lock);
TRACE_WARN_ON_LOCKED(plist_empty(&waiter->pi_list));
- TRACE_WARN_ON_LOCKED(lock_owner(lock));
plist_for_each(curr1, &old_owner->task->pi_waiters) {
w = plist_entry(curr1, struct rt_mutex_waiter, pi_list);
@@ -851,7 +850,7 @@
__raw_spin_lock(&l->wait_lock);
TRACE_BUG_ON_LOCKED(!lock_owner(l));
- if (rt_task(p) && plist_empty(&w->pi_list)) {
+ if ((ALL_TASKS_PI || rt_task(p)) && plist_empty(&w->pi_list)) {
TRACE_BUG_ON_LOCKED(was_rt);
plist_init(&w->pi_list, prio);
plist_add(&w->pi_list, &lock_owner(l)->task->pi_waiters);
@@ -868,7 +867,7 @@
* (TODO: this can be unfair to SCHED_NORMAL tasks if they
* get PI handled.)
*/
- if (!rt_task(p) && !plist_empty(&w->pi_list)) {
+ if (!ALL_TASKS_PI && !rt_task(p) && !plist_empty(&w->pi_list)) {
TRACE_BUG_ON_LOCKED(!was_rt);
plist_del(&w->pi_list, &lock_owner(l)->task->pi_waiters);
plist_del(&w->list, &l->wait_list);
@@ -970,7 +969,7 @@
*/
TRACE_BUG_ON_LOCKED(!spin_is_locked(&task->pi_lock));
TRACE_BUG_ON_LOCKED(!spin_is_locked(&lock->wait_lock));
-#ifndef ALL_TASKS_PI
+#if !ALL_TASKS_PI
if (!rt_task(task)) {
plist_add(&waiter->list, &lock->wait_list);
set_lock_owner_pending(lock);
@@ -1071,6 +1070,7 @@
struct rt_mutex_waiter *waiter = NULL;
struct thread_info *new_owner;
+ TRACE_BUG_ON_LOCKED(!spin_is_locked(&lock->wait_lock));
/*
* Get the highest prio one:
*
@@ -1083,7 +1083,7 @@
#endif
trace_special_pid(waiter->ti->task->pid, waiter->ti->task->prio, 0);
-#ifdef ALL_TASKS_PI
+#if ALL_TASKS_PI
check_pi_list_present(lock, waiter, old_owner);
#endif
new_owner = waiter->ti;
@@ -1543,7 +1543,7 @@
list_del_init(&lock->held_list);
#endif
-#ifdef ALL_TASKS_PI
+#if ALL_TASKS_PI
if (plist_empty(&lock->wait_list))
check_pi_list_empty(lock, lock_owner(lock));
#endif
@@ -1906,8 +1906,6 @@
__raw_spin_unlock(&old_owner->task->pi_lock);
} else
set_new_owner(lock, old_owner, ti __EIP__);
- __raw_spin_unlock(&lock->wait_lock);
- __raw_spin_unlock(&task->pi_lock);
ret = 1;
}
__raw_spin_unlock(&lock->wait_lock);
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: 2.6.13-rt3
2005-09-01 19:38 ` 2.6.13-rt3 Steven Rostedt
@ 2005-09-01 20:28 ` Ingo Molnar
2005-09-08 13:41 ` 2.6.13-rt3 Stephane Couture
0 siblings, 1 reply; 5+ messages in thread
From: Ingo Molnar @ 2005-09-01 20:28 UTC (permalink / raw)
To: Steven Rostedt; +Cc: LKML
* Steven Rostedt <rostedt@goodmis.org> wrote:
> Ingo,
>
> I just found a __MAJOR__ bug in my code. Below is the patch that
> fixes this bug, zaps the WARN_ON in check_pi_list_present, and changes
> ALL_TASKS_PI to a booleon instead of just a define.
>
> The major bug was in __down_trylock. See anything wrong with this
> code :-) I'm surprised that this worked as well as it did!
ok, i've released -rt4 with this fix included. The 8-way box boots fine
now.
Ingo
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: 2.6.13-rt3
2005-09-01 20:28 ` 2.6.13-rt3 Ingo Molnar
@ 2005-09-08 13:41 ` Stephane Couture
0 siblings, 0 replies; 5+ messages in thread
From: Stephane Couture @ 2005-09-08 13:41 UTC (permalink / raw)
To: LKML; +Cc: Ingo Molnar
Ingo Molnar wrote:
> * Steven Rostedt <rostedt@goodmis.org> wrote:
>
>
>>Ingo,
>>
>>I just found a __MAJOR__ bug in my code. Below is the patch that
>>fixes this bug, zaps the WARN_ON in check_pi_list_present, and changes
>>ALL_TASKS_PI to a booleon instead of just a define.
>>
>>The major bug was in __down_trylock. See anything wrong with this
>>code :-) I'm surprised that this worked as well as it did!
>
>
> ok, i've released -rt4 with this fix included. The 8-way box boots fine
> now.
>
> Ingo
On 2.6.13-rt4, I get this error when HIGH_RES_TIMERS is disabled :
arch/ppc/kernel/built-in.o(.text+0x1dbc): In function `timer_interrupt':
arch/ppc/kernel/time.c:241: undefined reference to `do_hr_timer_int'
There is some #ifdef CONFIG_HIGH_RES_TIMERS missing in
arch/ppc/kernel/time.c
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2005-09-08 13:41 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2005-09-01 16:24 2.6.13-rt3 Steven Rostedt
2005-09-01 16:46 ` 2.6.13-rt3 Steven Rostedt
2005-09-01 19:38 ` 2.6.13-rt3 Steven Rostedt
2005-09-01 20:28 ` 2.6.13-rt3 Ingo Molnar
2005-09-08 13:41 ` 2.6.13-rt3 Stephane Couture
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox