[dpdk-dev] [PATCH 1/2] raw/ioat: fix ring space checks

[dpdk-dev] [PATCH 1/2] raw/ioat: fix ring space checks

[Bruce Richardson]

When enqueuing a descriptor, when checking that there is at least one
slot free for the current descriptor and a later batch descriptor, we
need to test for both two free and one free, in case the last write
was a batch descriptor which is allowed to use the "spare" slot.

Similarly, when computing the free space in the ring to return to the
user, we need to take account of the same condition, so that we do not
return a "-1" ring space value, by blindly subtracting "2".

Fixes: 245efe544d8e ("raw/ioat: report status of completed jobs")

Reported-by: Jiayu Hu <jiayu.hu@intel.com>
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/raw/ioat/rte_idxd_rawdev_fns.h | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/drivers/raw/ioat/rte_idxd_rawdev_fns.h b/drivers/raw/ioat/rte_idxd_rawdev_fns.h
index 862e0eb41..6ed67d77e 100644
--- a/drivers/raw/ioat/rte_idxd_rawdev_fns.h
+++ b/drivers/raw/ioat/rte_idxd_rawdev_fns.h
@@ -132,7 +132,7 @@ __idxd_burst_capacity(int dev_id)
 	struct rte_idxd_rawdev *idxd =
 			(struct rte_idxd_rawdev *)rte_rawdevs[dev_id].dev_private;
 	uint16_t write_idx = idxd->batch_start + idxd->batch_size;
-	uint16_t used_space;
+	uint16_t used_space, free_space;
 
 	/* Check for space in the batch ring */
 	if ((idxd->batch_idx_read == 0 && idxd->batch_idx_write == idxd->max_batches) ||
@@ -147,7 +147,10 @@ __idxd_burst_capacity(int dev_id)
 	/* Return amount of free space in the descriptor ring
 	 * subtract 1 for space for batch descriptor and 1 for possible null desc
 	 */
-	return idxd->desc_ring_mask - used_space - 2;
+	free_space = idxd->desc_ring_mask - used_space;
+	if (free_space < 2)
+		return 0;
+	return free_space - 2;
 }
 
 static __rte_always_inline rte_iova_t
@@ -174,7 +177,8 @@ __idxd_write_desc(int dev_id,
 			idxd->batch_idx_write + 1 == idxd->batch_idx_read)
 		goto failed;
 	/* for descriptor ring, we always need a slot for batch completion */
-	if (((write_idx + 2) & mask) == idxd->hdls_read)
+	if (((write_idx + 2) & mask) == idxd->hdls_read ||
+			((write_idx + 1) & mask) == idxd->hdls_read)
 		goto failed;
 
 	/* write desc and handle. Note, descriptors don't wrap */
-- 
2.30.2

[dpdk-dev] [PATCH 2/2] raw/ioat: remove special case for no status reporting

[Bruce Richardson]

The special fast-path for returning completed descriptors without
reporting status or user-handles returns the number of completed ring
slots used, rather than the number of actual user-submitted jobs. This
means that the counts returned are too high, as the batch descriptor
slots would be included in the total. Therefore remove this special
case, and use the normal status-processing path so that the returned
count is correct in all cases.

Fixes: 245efe544d8e ("raw/ioat: report status of completed jobs")

Reported-by: Jiayu Hu <jiayu.hu@intel.com>
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/raw/ioat/rte_idxd_rawdev_fns.h | 9 ---------
 1 file changed, 9 deletions(-)

diff --git a/drivers/raw/ioat/rte_idxd_rawdev_fns.h b/drivers/raw/ioat/rte_idxd_rawdev_fns.h
index 6ed67d77e..6c5334cb3 100644
--- a/drivers/raw/ioat/rte_idxd_rawdev_fns.h
+++ b/drivers/raw/ioat/rte_idxd_rawdev_fns.h
@@ -343,14 +343,6 @@ __idxd_completed_ops(int dev_id, uint8_t max_ops, uint32_t *status, uint8_t *num
 			idxd->batch_idx_read = 0;
 	}
 
-	if (idxd->cfg.hdls_disable && status == NULL) {
-		n = (idxd->hdls_avail < idxd->hdls_read) ?
-				(idxd->hdls_avail + idxd->desc_ring_mask + 1 - idxd->hdls_read) :
-				(idxd->hdls_avail - idxd->hdls_read);
-		idxd->hdls_read = idxd->hdls_avail;
-		goto out;
-	}
-
 	n = 0;
 	h_idx = idxd->hdls_read;
 	while (h_idx != idxd->hdls_avail) {
@@ -386,7 +378,6 @@ __idxd_completed_ops(int dev_id, uint8_t max_ops, uint32_t *status, uint8_t *num
 	}
 	idxd->hdls_read = h_idx;
 
-out:
 	idxd->xstats.completed += n;
 	return n;
 }
-- 
2.30.2

Re: [dpdk-dev] [PATCH 1/2] raw/ioat: fix ring space checks

[Laatz, Kevin]

> When enqueuing a descriptor, when checking that there is at least one
> slot free for the current descriptor and a later batch descriptor, we
> need to test for both two free and one free, in case the last write
> was a batch descriptor which is allowed to use the "spare" slot.
> 
> Similarly, when computing the free space in the ring to return to the
> user, we need to take account of the same condition, so that we do not
> return a "-1" ring space value, by blindly subtracting "2".
> 
> Fixes: 245efe544d8e ("raw/ioat: report status of completed jobs")
> 
> Reported-by: Jiayu Hu <jiayu.hu@intel.com>
> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> ---
<...>

Acked-by: Kevin Laatz <kevin.laatz@intel.com>

Re: [dpdk-dev] [PATCH 2/2] raw/ioat: remove special case for no status reporting

[Laatz, Kevin]

> The special fast-path for returning completed descriptors without
> reporting status or user-handles returns the number of completed ring
> slots used, rather than the number of actual user-submitted jobs. This
> means that the counts returned are too high, as the batch descriptor
> slots would be included in the total. Therefore remove this special
> case, and use the normal status-processing path so that the returned
> count is correct in all cases.
> 
> Fixes: 245efe544d8e ("raw/ioat: report status of completed jobs")
> 
> Reported-by: Jiayu Hu <jiayu.hu@intel.com>
> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> ---
<...>

Acked-by: Kevin Laatz <kevin.laatz@intel.com>

Re: [dpdk-dev] [PATCH 2/2] raw/ioat: remove special case for no status reporting

[Thomas Monjalon]

> > The special fast-path for returning completed descriptors without
> > reporting status or user-handles returns the number of completed ring
> > slots used, rather than the number of actual user-submitted jobs. This
> > means that the counts returned are too high, as the batch descriptor
> > slots would be included in the total. Therefore remove this special
> > case, and use the normal status-processing path so that the returned
> > count is correct in all cases.
> > 
> > Fixes: 245efe544d8e ("raw/ioat: report status of completed jobs")
> > 
> > Reported-by: Jiayu Hu <jiayu.hu@intel.com>
> > Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> 
> Acked-by: Kevin Laatz <kevin.laatz@intel.com>

Series applied, thanks