cxgb4/iw_cxgb4: Doorbell Drop Avoidance Bug Fixes

The current logic suffers from a slow response time to disable user DB
usage, and also fails to avoid DB FIFO drops under heavy load. This commit
fixes these deficiencies and makes the avoidance logic more optimal.
This is done by more efficiently notifying the ULDs of potential DB
problems, and implements a smoother flow control algorithm in iw_cxgb4,
which is the ULD that puts the most load on the DB fifo.

Design:

cxgb4:

Direct ULD callback from the DB FULL/DROP interrupt handler.  This allows
the ULD to stop doing user DB writes as quickly as possible.

While user DB usage is disabled, the LLD will accumulate DB write events
for its queues.  Then once DB usage is reenabled, a single DB write is
done for each queue with its accumulated write count.  This reduces the
load put on the DB fifo when reenabling.

iw_cxgb4:

Instead of marking each qp to indicate DB writes are disabled, we create
a device-global status page that each user process maps.  This allows
iw_cxgb4 to only set this single bit to disable all DB writes for all
user QPs vs traversing the idr of all the active QPs.  If the libcxgb4
doesn't support this, then we fall back to the old approach of marking
each QP.  Thus we allow the new driver to work with an older libcxgb4.

When the LLD upcalls iw_cxgb4 indicating DB FULL, we disable all DB writes
via the status page and transition the DB state to STOPPED.  As user
processes see that DB writes are disabled, they call into iw_cxgb4
to submit their DB write events.  Since the DB state is in STOPPED,
the QP trying to write gets enqueued on a new DB "flow control" list.
As subsequent DB writes are submitted for this flow controlled QP, the
amount of writes are accumulated for each QP on the flow control list.
So all the user QPs that are actively ringing the DB get put on this
list and the number of writes they request are accumulated.

When the LLD upcalls iw_cxgb4 indicating DB EMPTY, which is in a workq
context, we change the DB state to FLOW_CONTROL, and begin resuming all
the QPs that are on the flow control list.  This logic runs on until
the flow control list is empty or we exit FLOW_CONTROL mode (due to
a DB DROP upcall, for example).  QPs are removed from this list, and
their accumulated DB write counts written to the DB FIFO.  Sets of QPs,
called chunks in the code, are removed at one time. The chunk size is 64.
So 64 QPs are resumed at a time, and before the next chunk is resumed, the
logic waits (blocks) for the DB FIFO to drain.  This prevents resuming to
quickly and overflowing the FIFO.  Once the flow control list is empty,
the db state transitions back to NORMAL and user QPs are again allowed
to write directly to the user DB register.

The algorithm is designed such that if the DB write load is high enough,
then all the DB writes get submitted by the kernel using this flow
controlled approach to avoid DB drops.  As the load lightens though, we
resume to normal DB writes directly by user applications.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 58 insertions, 38 deletions

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
index 50abe1d61287..32db37709263 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
@@ -500,6 +500,7 @@ struct sge_txq {
         spinlock_t db_lock;
         int db_disabled;
         unsigned short db_pidx;
+        unsigned short db_pidx_inc;
         u64 udb;
 };
 
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
index 0ac53dd84c61..cc04d090354c 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
@@ -3578,14 +3578,25 @@ static void drain_db_fifo(struct adapter *adap, int usecs)
 
 static void disable_txq_db(struct sge_txq *q)
 {
-        spin_lock_irq(&q->db_lock);
+        unsigned long flags;
+
+        spin_lock_irqsave(&q->db_lock, flags);
         q->db_disabled = 1;
-        spin_unlock_irq(&q->db_lock);
+        spin_unlock_irqrestore(&q->db_lock, flags);
 }
 
-static void enable_txq_db(struct sge_txq *q)
+static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
 {
         spin_lock_irq(&q->db_lock);
+        if (q->db_pidx_inc) {
+                /* Make sure that all writes to the TX descriptors
+                 * are committed before we tell HW about them.
+                 */
+                wmb();
+                t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
+                             QID(q->cntxt_id) | PIDX(q->db_pidx_inc));
+                q->db_pidx_inc = 0;
+        }
         q->db_disabled = 0;
         spin_unlock_irq(&q->db_lock);
 }
@@ -3607,11 +3618,32 @@ static void enable_dbs(struct adapter *adap)
         int i;
 
         for_each_ethrxq(&adap->sge, i)
-                enable_txq_db(&adap->sge.ethtxq[i].q);
+                enable_txq_db(adap, &adap->sge.ethtxq[i].q);
         for_each_ofldrxq(&adap->sge, i)
-                enable_txq_db(&adap->sge.ofldtxq[i].q);
+                enable_txq_db(adap, &adap->sge.ofldtxq[i].q);
         for_each_port(adap, i)
-                enable_txq_db(&adap->sge.ctrlq[i].q);
+                enable_txq_db(adap, &adap->sge.ctrlq[i].q);
+}
+
+static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
+{
+        if (adap->uld_handle[CXGB4_ULD_RDMA])
+                ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
+                                cmd);
+}
+
+static void process_db_full(struct work_struct *work)
+{
+        struct adapter *adap;
+
+        adap = container_of(work, struct adapter, db_full_task);
+
+        drain_db_fifo(adap, dbfifo_drain_delay);
+        enable_dbs(adap);
+        notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
+        t4_set_reg_field(adap, SGE_INT_ENABLE3,
+                         DBFIFO_HP_INT | DBFIFO_LP_INT,
+                         DBFIFO_HP_INT | DBFIFO_LP_INT);
 }
 
 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
@@ -3619,7 +3651,7 @@ static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
         u16 hw_pidx, hw_cidx;
         int ret;
 
-        spin_lock_bh(&q->db_lock);
+        spin_lock_irq(&q->db_lock);
         ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
         if (ret)
                 goto out;
@@ -3636,7 +3668,8 @@ static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
         }
 out:
         q->db_disabled = 0;
-        spin_unlock_bh(&q->db_lock);
+        q->db_pidx_inc = 0;
+        spin_unlock_irq(&q->db_lock);
         if (ret)
                 CH_WARN(adap, "DB drop recovery failed.\n");
 }
@@ -3652,29 +3685,6 @@ static void recover_all_queues(struct adapter *adap)
                 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
 }
 
-static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
-{
-        mutex_lock(&uld_mutex);
-        if (adap->uld_handle[CXGB4_ULD_RDMA])
-                ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
-                                cmd);
-        mutex_unlock(&uld_mutex);
-}
-
-static void process_db_full(struct work_struct *work)
-{
-        struct adapter *adap;
-
-        adap = container_of(work, struct adapter, db_full_task);
-
-        notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
-        drain_db_fifo(adap, dbfifo_drain_delay);
-        t4_set_reg_field(adap, SGE_INT_ENABLE3,
-                         DBFIFO_HP_INT | DBFIFO_LP_INT,
-                         DBFIFO_HP_INT | DBFIFO_LP_INT);
-        notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
-}
-
 static void process_db_drop(struct work_struct *work)
 {
         struct adapter *adap;
@@ -3682,11 +3692,13 @@ static void process_db_drop(struct work_struct *work)
         adap = container_of(work, struct adapter, db_drop_task);
 
         if (is_t4(adap->params.chip)) {
-                disable_dbs(adap);
+                drain_db_fifo(adap, dbfifo_drain_delay);
                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
-                drain_db_fifo(adap, 1);
+                drain_db_fifo(adap, dbfifo_drain_delay);
                 recover_all_queues(adap);
+                drain_db_fifo(adap, dbfifo_drain_delay);
                 enable_dbs(adap);
+                notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
         } else {
                 u32 dropped_db = t4_read_reg(adap, 0x010ac);
                 u16 qid = (dropped_db >> 15) & 0x1ffff;
@@ -3727,6 +3739,8 @@ static void process_db_drop(struct work_struct *work)
 void t4_db_full(struct adapter *adap)
 {
         if (is_t4(adap->params.chip)) {
+                disable_dbs(adap);
+                notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
                 t4_set_reg_field(adap, SGE_INT_ENABLE3,
                                  DBFIFO_HP_INT | DBFIFO_LP_INT, 0);
                 queue_work(workq, &adap->db_full_task);
@@ -3735,8 +3749,11 @@ void t4_db_full(struct adapter *adap)
 
 void t4_db_dropped(struct adapter *adap)
 {
-        if (is_t4(adap->params.chip))
-                queue_work(workq, &adap->db_drop_task);
+        if (is_t4(adap->params.chip)) {
+                disable_dbs(adap);
+                notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
+        }
+        queue_work(workq, &adap->db_drop_task);
 }
 
 static void uld_attach(struct adapter *adap, unsigned int uld)
diff --git a/drivers/net/ethernet/chelsio/cxgb4/sge.c b/drivers/net/ethernet/chelsio/cxgb4/sge.c
index 46429f9d0592..d4db382ff8c7 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/sge.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/sge.c
@@ -860,9 +860,10 @@ static void cxgb_pio_copy(u64 __iomem *dst, u64 *src)
 static inline void ring_tx_db(struct adapter *adap, struct sge_txq *q, int n)
 {
         unsigned int *wr, index;
+        unsigned long flags;
 
         wmb();            /* write descriptors before telling HW */
-        spin_lock(&q->db_lock);
+        spin_lock_irqsave(&q->db_lock, flags);
         if (!q->db_disabled) {
                 if (is_t4(adap->params.chip)) {
                         t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
@@ -878,9 +879,10 @@ static inline void ring_tx_db(struct adapter *adap, struct sge_txq *q, int n)
                                 writel(n,  adap->bar2 + q->udb + 8);
                         wmb();
                 }
-        }
+        } else
+                q->db_pidx_inc += n;
         q->db_pidx = q->pidx;
-        spin_unlock(&q->db_lock);
+        spin_unlock_irqrestore(&q->db_lock, flags);
 }
 
 /**