qed: fix handling of concurrent ramrods.
Concurrent non-blocking slowpath ramrods can be completed
out-of-order on the completion chain. Recycling completed elements,
while previously sent elements are still completion pending,
can lead to overriding of active elements on the chain. Furthermore,
sending pending slowpath ramrods currently lacks the update of the
chain element physical pointer.
This patch:
* Ensures that ramrods are sent to the FW with
consecutive echo values.
* Handles out-of-order completions by freeing only first
successive completed entries.
* Updates the chain element physical pointer when copying
a pending element into a free element for sending.
Signed-off-by: Tomer Tayar <Tomer.Tayar@qlogic.com>
Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/net/ethernet/qlogic/qed/qed_spq.c b/drivers/net/ethernet/qlogic/qed/qed_spq.c
index 7c0b845..3dd548a 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_spq.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_spq.c
@@ -112,8 +112,6 @@
qed_spq_fill_entry(struct qed_hwfn *p_hwfn,
struct qed_spq_entry *p_ent)
{
- p_ent->elem.hdr.echo = 0;
- p_hwfn->p_spq->echo_idx++;
p_ent->flags = 0;
switch (p_ent->comp_mode) {
@@ -195,10 +193,12 @@
struct qed_spq *p_spq,
struct qed_spq_entry *p_ent)
{
- struct qed_chain *p_chain = &p_hwfn->p_spq->chain;
+ struct qed_chain *p_chain = &p_hwfn->p_spq->chain;
+ u16 echo = qed_chain_get_prod_idx(p_chain);
struct slow_path_element *elem;
struct core_db_data db;
+ p_ent->elem.hdr.echo = cpu_to_le16(echo);
elem = qed_chain_produce(p_chain);
if (!elem) {
DP_NOTICE(p_hwfn, "Failed to produce from SPQ chain\n");
@@ -437,7 +437,9 @@
p_spq->comp_count = 0;
p_spq->comp_sent_count = 0;
p_spq->unlimited_pending_count = 0;
- p_spq->echo_idx = 0;
+
+ bitmap_zero(p_spq->p_comp_bitmap, SPQ_RING_SIZE);
+ p_spq->comp_bitmap_idx = 0;
/* SPQ cid, cannot fail */
qed_cxt_acquire_cid(p_hwfn, PROTOCOLID_CORE, &p_spq->cid);
@@ -582,26 +584,32 @@
struct qed_spq *p_spq = p_hwfn->p_spq;
if (p_ent->queue == &p_spq->unlimited_pending) {
- struct qed_spq_entry *p_en2;
if (list_empty(&p_spq->free_pool)) {
list_add_tail(&p_ent->list, &p_spq->unlimited_pending);
p_spq->unlimited_pending_count++;
return 0;
+ } else {
+ struct qed_spq_entry *p_en2;
+
+ p_en2 = list_first_entry(&p_spq->free_pool,
+ struct qed_spq_entry,
+ list);
+ list_del(&p_en2->list);
+
+ /* Copy the ring element physical pointer to the new
+ * entry, since we are about to override the entire ring
+ * entry and don't want to lose the pointer.
+ */
+ p_ent->elem.data_ptr = p_en2->elem.data_ptr;
+
+ *p_en2 = *p_ent;
+
+ kfree(p_ent);
+
+ p_ent = p_en2;
}
-
- p_en2 = list_first_entry(&p_spq->free_pool,
- struct qed_spq_entry,
- list);
- list_del(&p_en2->list);
-
- /* Strcut assignment */
- *p_en2 = *p_ent;
-
- kfree(p_ent);
-
- p_ent = p_en2;
}
/* entry is to be placed in 'pending' queue */
@@ -777,13 +785,38 @@
list_for_each_entry_safe(p_ent, tmp, &p_spq->completion_pending,
list) {
if (p_ent->elem.hdr.echo == echo) {
+ u16 pos = le16_to_cpu(echo) % SPQ_RING_SIZE;
+
list_del(&p_ent->list);
- qed_chain_return_produced(&p_spq->chain);
+ /* Avoid overriding of SPQ entries when getting
+ * out-of-order completions, by marking the completions
+ * in a bitmap and increasing the chain consumer only
+ * for the first successive completed entries.
+ */
+ bitmap_set(p_spq->p_comp_bitmap, pos, SPQ_RING_SIZE);
+
+ while (test_bit(p_spq->comp_bitmap_idx,
+ p_spq->p_comp_bitmap)) {
+ bitmap_clear(p_spq->p_comp_bitmap,
+ p_spq->comp_bitmap_idx,
+ SPQ_RING_SIZE);
+ p_spq->comp_bitmap_idx++;
+ qed_chain_return_produced(&p_spq->chain);
+ }
+
p_spq->comp_count++;
found = p_ent;
break;
}
+
+ /* This is relatively uncommon - depends on scenarios
+ * which have mutliple per-PF sent ramrods.
+ */
+ DP_VERBOSE(p_hwfn, QED_MSG_SPQ,
+ "Got completion for echo %04x - doesn't match echo %04x in completion pending list\n",
+ le16_to_cpu(echo),
+ le16_to_cpu(p_ent->elem.hdr.echo));
}
/* Release lock before callback, as callback may post