fallout from atomic_long_t patch
PATCH 1/1 fallout from atomic_long_t patch
---
drivers/connector/connector.c +1 -2
@@ -135,8 +135,7 @@ static int cn_call_callback(struct cn_msspin_lock_bh (&dev->cbdev->queue_lock);list_for_each_entry (__cbq, &dev->cbdev->queue_list, callback_entry) {if (cn_cb_equal (&__cbq->id.id, &msg->id)) { -if (likely (!test_bit (WORK_STRUCT_PENDING , - &__cbq->work.work.management) && +if (likely (!work_pending (&__cbq->work.work) && __cbq->data.ddata ==NULL )) { __cbq->data.callback_priv = msg;
Review
Quoted 3 lines · Al Viro
(quoted diff) +1
-2- if (likely(!test_bit(WORK_STRUCT_PENDING, - &__cbq->work.work.management) && + if (likely(!work_pending(&__cbq->work.work) &&
That should properly be
if (likely(!delayed_work_pending(&__cbq->work) && ...
and why the heck was it doing that open-coded int he first place?
HOWEVER, looking even more, why is that thing a "delayed work" at all? All the queuing seems to happen with a timeout of zero..
So I _think_ that the proper patch is actually the following, but somebody who knows and uses the connector thing should double-check. Please?
Linus
---
drivers/connector/cn_queue.c +2 -3
@@ -34,7 +34,7 @@void cn_queue_wrapper (struct work_struct *work) {struct cn_callback_entry *cbq = - container_of(work, struct cn_callback_entry, work.work); + container_of(work, struct cn_callback_entry, work);struct cn_callback_data *d = &cbq->data; d->callback (d->callback_priv);
@@ -59,13 +59,12 @@ static struct cn_callback_entry *cn_queue_alloc_callback_entry(char *name, strucmemcpy (&cbq->id.id, id,sizeof (struct cb_id )); cbq->data.callback = callback; -INIT_DELAYED_WORK (&cbq->work, &cn_queue_wrapper); +INIT_WORK (&cbq->work, &cn_queue_wrapper);return cbq; }static void cn_queue_free_callback (struct cn_callback_entry *cbq) { -cancel_delayed_work (&cbq->work);flush_workqueue (cbq->pdev->cn_queue);kfree (cbq);
drivers/connector/connector.c +6 -7
@@ -135,17 +135,16 @@ static int cn_call_callback(struct cn_msg *msg, void (*destruct_data)(void *), vspin_lock_bh (&dev->cbdev->queue_lock);list_for_each_entry (__cbq, &dev->cbdev->queue_list, callback_entry) {if (cn_cb_equal (&__cbq->id.id, &msg->id)) { -if (likely (!test_bit (WORK_STRUCT_PENDING , - &__cbq->work.work.management) && +if (likely (!work_pending (&__cbq->work) && __cbq->data.ddata ==NULL )) { __cbq->data.callback_priv = msg; __cbq->data.ddata = data; __cbq->data.destruct_data = destruct_data; -if (queue_delayed_work ( +if (queue_work ( dev->cbdev->cn_queue, - &__cbq->work,0 )) + &__cbq->work)) err =0 ; }else {struct cn_callback_data *d;
@@ -159,12 +158,12 @@ static int cn_call_callback(struct cn_msg *msg, void (*destruct_data)(void *), v d->destruct_data = destruct_data; d->free = __cbq; -INIT_DELAYED_WORK (&__cbq->work, +INIT_WORK (&__cbq->work, &cn_queue_wrapper); -if (queue_delayed_work ( +if (queue_work ( dev->cbdev->cn_queue, - &__cbq->work,0 )) + &__cbq->work)) err =0 ;else {kfree (__cbq);
include/linux/connector.h +1 -1
@@ -133,7 +133,7 @@ struct cn_callback_data {struct cn_callback_entry {struct list_head callback_entry;struct cn_callback *cb; -struct delayed_work work; +struct work_struct work;struct cn_queue_dev *pdev;struct cn_callback_id id;
Quoted 11 lines · Linus Torvalds
drivers/connector/connector.c static int cn_call_callback(struct cn_msg *msg, void (*destruct_data)(void *), v
Quoted 3 lines · Al Viro
drivers/connector/connector.c +1
-2- if (likely(!test_bit(WORK_STRUCT_PENDING, - &__cbq->work.work.management) && + if (likely(!work_pending(&__cbq->work.work) &&That should properly be
if (likely(!delayed_work_pending(&__cbq->work) && ...
Quoted 7 lines
and why the heck was it doing that open-coded int he first place?
HOWEVER, looking even more, why is that thing a "delayed work" at all? All the queuing seems to happen with a timeout of zero..
So I _think_ that the proper patch is actually the following, but somebody who knows and uses the connector thing should double-check. Please?
Delayed work was used to play with different timeouts and thus allow to smooth performance peaks, but then I dropped that idea, so timeout is always zero.
I posted similar patch today to netdev@, which directly used work_pending instead of delayed_work_pending(), but if you will figure this out itself, I'm ok with proposed patch.
Quoted 1 line
Linus
Signature · 1 line
Evgeniy Polyakov
Quoted 4 lines · Evgeniy Polyakov
Delayed work was used to play with different timeouts and thus allow to smooth performance peaks, but then I dropped that idea, so timeout is always zero.
Ok, thanks for the explanation.
Quoted 3 lines
I posted similar patch today to netdev@, which directly used
work_pendinginstead ofdelayed_work_pending(), but if you will figure this out itself, I'm ok with proposed patch.
If I'm going to get the proper patch from the proper network trees, I'll just drop my patch. Whether you replace "delayed_work" with "work_struct" or not is not something I really care about - if you think you may want to play with the timeout idea in the future, please feel free to continue using delayed_work.
But if you do use delayed work, please use the "delayed_work_pending(&x)" function, rather than doing "work_pending(&x->work)" and knowing about the internals of how the delayed-work structure looks.
So with that out of the way, I'll just expect that I'll get whatever you decide on through Davem's git tree, once his drunken holiday revelry is over ;)
Linus
Quoted 3 lines · Linus Torvalds
So with that out of the way, I'll just expect that I'll get whatever you decide on through Davem's git tree, once his drunken holiday revelry is over ;)
This is important process - never interrupt it for things like patches from external developers :)
I will push it through his tree.
Quoted 1 line
Linus
Signature · 1 line
Evgeniy Polyakov