2007-02-14 16:45:07

by Hoang-Nam Nguyen

Subject: [PATCH 2.6.21-rc1 4/5] ehca: replace yield() by wait_for_completion()

Hi,
this patch removes yield() and uses wait_for_completion() in order
to wait for running completion handlers finished before destroying
associated completion queue.
Thanks
Nam


Signed-off-by: Hoang-Nam Nguyen <[email protected]>
---


ehca_classes.h | 3 +++
ehca_cq.c | 3 ++-
ehca_irq.c | 6 +++++-
3 files changed, 10 insertions(+), 2 deletions(-)


diff -Nurp infiniband_orig/drivers/infiniband/hw/ehca/ehca_classes.h infiniband_work/drivers/infiniband/hw/ehca/ehca_classes.h
--- infiniband_orig/drivers/infiniband/hw/ehca/ehca_classes.h 2007-02-14 13:52:49.000000000 +0100
+++ infiniband_work/drivers/infiniband/hw/ehca/ehca_classes.h 2007-02-14 13:52:06.000000000 +0100
@@ -52,6 +52,8 @@ struct ehca_mw;
struct ehca_pd;
struct ehca_av;

+#include <linux/completion.h>
+
#include <rdma/ib_verbs.h>
#include <rdma/ib_user_verbs.h>

@@ -154,6 +156,7 @@ struct ehca_cq {
struct hlist_head qp_hashtab[QP_HASHTAB_LEN];
struct list_head entry;
u32 nr_callbacks;
+ struct completion zero_callbacks;
spinlock_t task_lock;
u32 ownpid;
/* mmap counter for resources mapped into user space */
diff -Nurp infiniband_orig/drivers/infiniband/hw/ehca/ehca_cq.c infiniband_work/drivers/infiniband/hw/ehca/ehca_cq.c
--- infiniband_orig/drivers/infiniband/hw/ehca/ehca_cq.c 2007-02-14 13:52:49.000000000 +0100
+++ infiniband_work/drivers/infiniband/hw/ehca/ehca_cq.c 2007-02-14 13:52:06.000000000 +0100
@@ -147,6 +147,7 @@ struct ib_cq *ehca_create_cq(struct ib_d
spin_lock_init(&my_cq->spinlock);
spin_lock_init(&my_cq->cb_lock);
spin_lock_init(&my_cq->task_lock);
+ init_completion(&my_cq->zero_callbacks);
my_cq->ownpid = current->tgid;

cq = &my_cq->ib_cq;
@@ -332,7 +333,7 @@ int ehca_destroy_cq(struct ib_cq *cq)
spin_lock_irqsave(&ehca_cq_idr_lock, flags);
while (my_cq->nr_callbacks) {
spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
- yield();
+ wait_for_completion(&my_cq->zero_callbacks);
spin_lock_irqsave(&ehca_cq_idr_lock, flags);
}

diff -Nurp infiniband_orig/drivers/infiniband/hw/ehca/ehca_irq.c infiniband_work/drivers/infiniband/hw/ehca/ehca_irq.c
--- infiniband_orig/drivers/infiniband/hw/ehca/ehca_irq.c 2007-02-14 13:52:49.000000000 +0100
+++ infiniband_work/drivers/infiniband/hw/ehca/ehca_irq.c 2007-02-14 13:52:06.000000000 +0100
@@ -605,6 +605,7 @@ static void run_comp_task(struct ehca_cp
spin_lock_irqsave(&cct->task_lock, flags);

while (!list_empty(&cct->cq_list)) {
+ int is_complete = 0;
cq = list_entry(cct->cq_list.next, struct ehca_cq, entry);
spin_unlock_irqrestore(&cct->task_lock, flags);
comp_event_callback(cq);
@@ -612,11 +613,14 @@ static void run_comp_task(struct ehca_cp

spin_lock(&cq->task_lock);
cq->nr_callbacks--;
- if (cq->nr_callbacks == 0) {
+ is_complete = (cq->nr_callbacks == 0);
+ if (is_complete) {
list_del_init(cct->cq_list.next);
cct->cq_jobs--;
}
spin_unlock(&cq->task_lock);
+ if (is_complete) /* wake up waiting destroy_cq() */
+ complete(&cq->zero_callbacks);
}

spin_unlock_irqrestore(&cct->task_lock, flags);

2007-02-14 22:28:31

by Christoph Hellwig

Subject: Re: [PATCH 2.6.21-rc1 4/5] ehca: replace yield() by wait_for_completion()

> @@ -332,7 +333,7 @@ int ehca_destroy_cq(struct ib_cq *cq)
> spin_lock_irqsave(&ehca_cq_idr_lock, flags);
> while (my_cq->nr_callbacks) {
> spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
> - yield();
> + wait_for_completion(&my_cq->zero_callbacks);
> spin_lock_irqsave(&ehca_cq_idr_lock, flags);
> }

A while loop around wait_for_completion doesn't make all that much sense.
I suspect a simple

if (my_cq->nr_callbacks)
wait_for_completion(&my_cq->zero_callbacks);

Is what you need.

2007-02-15 00:50:23

by Roland Dreier

Subject: Re: [PATCH 2.6.21-rc1 4/5] ehca: replace yield() by wait_for_completion()

I agree with Christoph -- the use of wait_for_completion() in a loop
makes no sense. When you send a new copy of this patch without
whitespace damage, please fix that up too...