> Doing them as wait queues is the obvious answer: turning
>
> cli();
> if (cond)
> sleep_on(&cond_list);
>
> into
> if (cond) {
> current->state = TASK_UNINTERRUPTIBLE;
> add_wait_queue(&wait, &cond_list);
> if (cond)
> schedule();
> remove_wait_queue(..)
> }
ok, next try to fix this issue cleanly.
i've attached a draft 'waiting for events' framework, and i've converted
printk.c to the 'new' usage. This now is both raceless and doesnt lock up
on SMP. The user-visible interface is:
wait_event(wq, condition);
or
wait_async_event(wq, condition);
or
error = wait_event_interruptible(wq, condition);
or
error = wait_async_event_interruptible(wq, condition);
which visibly simplifies usage. The downside is that these are macros. I
tried to write them as safely as possible, but they are still very ugly
and might be considered a showstopper? ...
the 'async' versions are really just calls to the 'simple' versions, they
do a cli(); condition check; sti(); thing to ensure correctness of
volatile conditions. Smarter code can still define arbitrary conditions
similarly to the _async_ variants.
this is now running on my box, and syslog works just like before.
is this about the way how it should be done? I always found the explicit
waitqueue manipulations to be a complexity overkill, and people stumble
across this periodically. Also, if in the future we change the way we wait
for events, we can just fix up the framework instead of all the usages.
-- mingo
--- linux/kernel/printk.c.orig Mon Mar 1 10:18:56 1999
+++ linux/kernel/printk.c Tue Mar 2 20:33:45 1999
@@ -137,15 +137,9 @@
error = verify_area(VERIFY_WRITE,buf,len);
if (error)
goto out;
- cli();
- error = -ERESTARTSYS;
- while (!log_size) {
- if (signal_pending(current)) {
- sti();
- goto out;
- }
- interruptible_sleep_on(&log_wait);
- }
+ error = wait_async_event_interruptible(log_wait, log_size);
+ if (error)
+ goto out;
i = 0;
while (log_size && i < len) {
c = *((char *) log_buf+log_start);
--- linux/include/linux/sched.h.orig Sun Feb 28 14:41:28 1999
+++ linux/include/linux/sched.h Tue Mar 2 20:31:17 1999
@@ -671,6 +671,72 @@
__remove_wait_queue(p, wait);
write_unlock_irqrestore(&waitqueue_lock, flags);
}
+
+#define wait_event(wq, condition) \
+do { \
+ struct wait_queue __wait; \
+ \
+ if (condition) \
+ break; \
+ \
+ __wait.task = current; \
+ add_wait_queue(&wq, &__wait); \
+ for (;;) { \
+ current->state = TASK_UNINTERRUPTIBLE; \
+ if (condition) \
+ break; \
+ schedule(); \
+ } \
+ current->state = TASK_RUNNING; \
+ remove_wait_queue(&wq, &__wait); \
+} while (0)
+
+#define wait_async_event(wq, condition) \
+ wait_event(wq, \
+ ({ \
+ int __ret; \
+ cli(); \
+ __ret = condition; \
+ sti(); \
+ __ret; \
+ }))
+
+#define wait_event_interruptible(wq, condition) \
+({ \
+ int __ret = 0; \
+ struct wait_queue __wait; \
+ \
+ if (condition) \
+ goto __out; \
+ \
+ __wait.task = current; \
+ add_wait_queue(&wq, &__wait); \
+ for (;;) { \
+ current->state = TASK_INTERRUPTIBLE; \
+ if (condition) \
+ break; \
+ if (signal_pending(current)) { \
+ __ret = -ERESTARTSYS; \
+ break; \
+ } \
+ schedule(); \
+ } \
+ current->state = TASK_RUNNING; \
+ remove_wait_queue(&wq, &__wait); \
+__out: \
+ __ret; \
+})
+
+#define wait_async_event_interruptible(wq, condition) \
+ wait_event_interruptible(wq, \
+ ({ \
+ int __ret; \
+ cli(); \
+ __ret = condition; \
+ sti(); \
+ __ret; \
+ }))
+
#define REMOVE_LINKS(p) do { \
(p)->next_task->prev_task = (p)->prev_task; \
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/