[dpdk-dev] [PATCH] event/sw: fix cq idx check for unlink usecases

[dpdk-dev] [PATCH] event/sw: fix cq idx check for unlink usecases

[Harry van Haaren]

This commit fixes the cq index checks when unlinking
ports/queues while the scheduler core is running.
Previously, the == comparison could be "skipped" if
in particular corner cases. With the check being changed
to >= this is resolved as the cq idx gets reset to zero.

Bugzilla ID: 60
Fixes: 617995dfc5b2 ("event/sw: add scheduling logic")

Suggested-by: Matias Elo <matias.elo@nokia.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>

---

Cc: stable@dpdk.org

@Matias,

When testing this patch with your provided test case as per attachment
to bug #60 in Bugzilla, I don't see any events arriving to port 0.
Hence, I believe this to be the correct behaviour, if you can confirm
that'd be awesome!

Regards, -Harry

---
 drivers/event/sw/sw_evdev_scheduler.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/drivers/event/sw/sw_evdev_scheduler.c b/drivers/event/sw/sw_evdev_scheduler.c
index e3a41e02f..fb5d44630 100644
--- a/drivers/event/sw/sw_evdev_scheduler.c
+++ b/drivers/event/sw/sw_evdev_scheduler.c
@@ -51,9 +51,11 @@ sw_schedule_atomic_to_cq(struct sw_evdev *sw, struct sw_qid * const qid,
 		int cq = fid->cq;
 
 		if (cq < 0) {
-			uint32_t cq_idx = qid->cq_next_tx++;
-			if (qid->cq_next_tx == qid->cq_num_mapped_cqs)
+			uint32_t cq_idx;
+			if (qid->cq_next_tx >= qid->cq_num_mapped_cqs)
 				qid->cq_next_tx = 0;
+			cq_idx = qid->cq_next_tx++;
+
 			cq = qid->cq_map[cq_idx];
 
 			/* find least used */
@@ -140,9 +142,10 @@ sw_schedule_parallel_to_cq(struct sw_evdev *sw, struct sw_qid * const qid,
 		do {
 			if (++cq_check_count > qid->cq_num_mapped_cqs)
 				goto exit;
-			cq = qid->cq_map[cq_idx];
-			if (++cq_idx == qid->cq_num_mapped_cqs)
+			if (cq_idx >= qid->cq_num_mapped_cqs)
 				cq_idx = 0;
+			cq = qid->cq_map[cq_idx++];
+
 		} while (rte_event_ring_free_count(
 				sw->ports[cq].cq_worker_ring) == 0 ||
 				sw->ports[cq].inflights == SW_PORT_HIST_LIST);
@@ -220,7 +223,7 @@ sw_schedule_qid_to_cq(struct sw_evdev *sw)
 		int iq_num = PKT_MASK_TO_IQ(qid->iq_pkt_mask);
 
 		/* zero mapped CQs indicates directed */
-		if (iq_num >= SW_IQS_MAX)
+		if (iq_num >= SW_IQS_MAX || qid->cq_num_mapped_cqs == 0)
 			continue;
 
 		uint32_t pkts_done = 0;
-- 
2.17.1

Re: [dpdk-dev] [PATCH] event/sw: fix cq idx check for unlink usecases

[Elo, Matias (Nokia - FI/Espoo)]

> On 21 Sep 2018, at 13:25, Harry van Haaren <harry.van.haaren@intel.com> wrote:
> 
> This commit fixes the cq index checks when unlinking
> ports/queues while the scheduler core is running.
> Previously, the == comparison could be "skipped" if
> in particular corner cases. With the check being changed
> to >= this is resolved as the cq idx gets reset to zero.
> 
> Bugzilla ID: 60
> Fixes: 617995dfc5b2 ("event/sw: add scheduling logic")
> 
> Suggested-by: Matias Elo <matias.elo@nokia.com>
> Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
> 
> ---
> 
> Cc: stable@dpdk.org
> 
> @Matias,
> 
> When testing this patch with your provided test case as per attachment
> to bug #60 in Bugzilla, I don't see any events arriving to port 0.
> Hence, I believe this to be the correct behaviour, if you can confirm
> that'd be awesome!
> 

I can confirm that this patch fixes the problem also on my system. Thanks!

I can’t remember if I ever saw this problem with ordered queues. At least now they seem to be also working correctly.

Regards,
Matias

Re: [dpdk-dev] [PATCH] event/sw: fix cq idx check for unlink usecases

[Jerin Jacob]

-----Original Message-----
> Date: Fri, 21 Sep 2018 12:17:38 +0000
> From: "Elo, Matias (Nokia - FI/Espoo)" <matias.elo@nokia.com>
> To: Harry van Haaren <harry.van.haaren@intel.com>
> CC: "dev@dpdk.org" <dev@dpdk.org>, "stable@dpdk.org" <stable@dpdk.org>
> Subject: Re: [dpdk-dev] [PATCH] event/sw: fix cq idx check for unlink
>  usecases
> x-mailer: Apple Mail (2.3445.9.1)
> 
> 
> > On 21 Sep 2018, at 13:25, Harry van Haaren <harry.van.haaren@intel.com> wrote:
> >
> > This commit fixes the cq index checks when unlinking
> > ports/queues while the scheduler core is running.
> > Previously, the == comparison could be "skipped" if
> > in particular corner cases. With the check being changed
> > to >= this is resolved as the cq idx gets reset to zero.
> >
> > Bugzilla ID: 60
> > Fixes: 617995dfc5b2 ("event/sw: add scheduling logic")
> >
> > Suggested-by: Matias Elo <matias.elo@nokia.com>
> > Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
> >
> > ---
> >
> > Cc: stable@dpdk.org
> >
> > @Matias,
> >
> > When testing this patch with your provided test case as per attachment
> > to bug #60 in Bugzilla, I don't see any events arriving to port 0.
> > Hence, I believe this to be the correct behaviour, if you can confirm
> > that'd be awesome!
> >
> 
> I can confirm that this patch fixes the problem also on my system. Thanks!
> 
> I can’t remember if I ever saw this problem with ordered queues. At least now they seem to be also working correctly.
> 
> Regards,
> Matias

Cc: stable@dpdk.org

Applied to dpdk-next-eventdev/master. Thanks.

>