[PATCH v2] dmaengine: idxd: fix interrupt completion after unmasking

[PATCH v2] dmaengine: idxd: fix interrupt completion after unmasking

[Dave Jiang]

The current implementation may miss completions after we unmask the
interrupt. In order to make sure we process all competions, we need to:
1. Do an MMIO read from the device as a barrier to ensure that all PCI
   writes for completions have arrived.
2. Check for any additional completions that we missed.

Fixes: 8f47d1a5e545 ("dmaengine: idxd: connect idxd to dmaengine subsystem")

Reported-by: Sanjay Kumar <sanjay.k.kumar@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
---

v2:
- Initialize processed counter

 drivers/dma/idxd/device.c |    7 +++++++
 drivers/dma/idxd/irq.c    |   26 +++++++++++++++++++-------
 2 files changed, 26 insertions(+), 7 deletions(-)

diff --git a/drivers/dma/idxd/device.c b/drivers/dma/idxd/device.c
index f6f49f0f6fae..8d79a8787104 100644
--- a/drivers/dma/idxd/device.c
+++ b/drivers/dma/idxd/device.c
@@ -62,6 +62,13 @@ int idxd_unmask_msix_vector(struct idxd_device *idxd, int vec_id)
 	perm.ignore = 0;
 	iowrite32(perm.bits, idxd->reg_base + offset);
 
+	/*
+	 * A readback from the device ensures that any previously generated
+	 * completion record writes are visible to software based on PCI
+	 * ordering rules.
+	 */
+	perm.bits = ioread32(idxd->reg_base + offset);
+
 	return 0;
 }
 
diff --git a/drivers/dma/idxd/irq.c b/drivers/dma/idxd/irq.c
index d6fcd2e60103..6510791b9921 100644
--- a/drivers/dma/idxd/irq.c
+++ b/drivers/dma/idxd/irq.c
@@ -173,6 +173,7 @@ static int irq_process_pending_llist(struct idxd_irq_entry *irq_entry,
 	struct llist_node *head;
 	int queued = 0;
 
+	*processed = 0;
 	head = llist_del_all(&irq_entry->pending_llist);
 	if (!head)
 		return 0;
@@ -197,6 +198,7 @@ static int irq_process_work_list(struct idxd_irq_entry *irq_entry,
 	struct list_head *node, *next;
 	int queued = 0;
 
+	*processed = 0;
 	if (list_empty(&irq_entry->work_list))
 		return 0;
 
@@ -218,10 +220,9 @@ static int irq_process_work_list(struct idxd_irq_entry *irq_entry,
 	return queued;
 }
 
-irqreturn_t idxd_wq_thread(int irq, void *data)
+static int idxd_desc_process(struct idxd_irq_entry *irq_entry)
 {
-	struct idxd_irq_entry *irq_entry = data;
-	int rc, processed = 0, retry = 0;
+	int rc, processed, total = 0;
 
 	/*
 	 * There are two lists we are processing. The pending_llist is where
@@ -244,15 +245,26 @@ irqreturn_t idxd_wq_thread(int irq, void *data)
 	 */
 	do {
 		rc = irq_process_work_list(irq_entry, &processed);
-		if (rc != 0) {
-			retry++;
+		total += processed;
+		if (rc != 0)
 			continue;
-		}
 
 		rc = irq_process_pending_llist(irq_entry, &processed);
-	} while (rc != 0 && retry != 10);
+		total += processed;
+	} while (rc != 0);
+
+	return total;
+}
+
+irqreturn_t idxd_wq_thread(int irq, void *data)
+{
+	struct idxd_irq_entry *irq_entry = data;
+	int processed;
 
+	processed = idxd_desc_process(irq_entry);
 	idxd_unmask_msix_vector(irq_entry->idxd, irq_entry->id);
+	/* catch anything unprocessed after unmasking */
+	processed += idxd_desc_process(irq_entry);
 
 	if (processed == 0)
 		return IRQ_NONE;

Re: [PATCH v2] dmaengine: idxd: fix interrupt completion after unmasking

[Vinod Koul]

On 01-05-20, 08:21, Dave Jiang wrote:
> The current implementation may miss completions after we unmask the
> interrupt. In order to make sure we process all competions, we need to:
> 1. Do an MMIO read from the device as a barrier to ensure that all PCI
>    writes for completions have arrived.
> 2. Check for any additional completions that we missed.

Applied, thanks

-- 
~Vinod