RDMA/cxgb4: Add DB Overflow Avoidance
Get FULL/EMPTY/DROP events from LLD. On FULL event, disable normal
user mode DB rings.
Add modify_qp semantics to allow user processes to call into the
kernel to ring doobells without overflowing.
Add DB Full/Empty/Drop stats.
Mark queues when created indicating the doorbell state.
If we're in the middle of db overflow avoidance, then newly created
queues should start out in this mode.
Bump the C4IW_UVERBS_ABI_VERSION to 2 so the user mode library can
know if the driver supports the kernel mode db ringing.
Signed-off-by: Vipul Pandya <vipul@chelsio.com>
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
diff --git a/drivers/infiniband/hw/cxgb4/device.c b/drivers/infiniband/hw/cxgb4/device.c
index 8483111..9062ed9 100644
--- a/drivers/infiniband/hw/cxgb4/device.c
+++ b/drivers/infiniband/hw/cxgb4/device.c
@@ -44,6 +44,12 @@
MODULE_LICENSE("Dual BSD/GPL");
MODULE_VERSION(DRV_VERSION);
+struct uld_ctx {
+ struct list_head entry;
+ struct cxgb4_lld_info lldi;
+ struct c4iw_dev *dev;
+};
+
static LIST_HEAD(uld_ctx_list);
static DEFINE_MUTEX(dev_mutex);
@@ -263,6 +269,9 @@
seq_printf(seq, " OCQPMEM: %10llu %10llu %10llu\n",
dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
dev->rdev.stats.ocqp.max);
+ seq_printf(seq, " DB FULL: %10llu\n", dev->rdev.stats.db_full);
+ seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
+ seq_printf(seq, " DB DROP: %10llu\n", dev->rdev.stats.db_drop);
return 0;
}
@@ -283,6 +292,9 @@
dev->rdev.stats.pbl.max = 0;
dev->rdev.stats.rqt.max = 0;
dev->rdev.stats.ocqp.max = 0;
+ dev->rdev.stats.db_full = 0;
+ dev->rdev.stats.db_empty = 0;
+ dev->rdev.stats.db_drop = 0;
mutex_unlock(&dev->rdev.stats.lock);
return count;
}
@@ -443,12 +455,6 @@
c4iw_destroy_resource(&rdev->resource);
}
-struct uld_ctx {
- struct list_head entry;
- struct cxgb4_lld_info lldi;
- struct c4iw_dev *dev;
-};
-
static void c4iw_dealloc(struct uld_ctx *ctx)
{
c4iw_rdev_close(&ctx->dev->rdev);
@@ -514,6 +520,7 @@
idr_init(&devp->mmidr);
spin_lock_init(&devp->lock);
mutex_init(&devp->rdev.stats.lock);
+ mutex_init(&devp->db_mutex);
if (c4iw_debugfs_root) {
devp->debugfs_root = debugfs_create_dir(
@@ -659,11 +666,76 @@
return 0;
}
+static int disable_qp_db(int id, void *p, void *data)
+{
+ struct c4iw_qp *qp = p;
+
+ t4_disable_wq_db(&qp->wq);
+ return 0;
+}
+
+static void stop_queues(struct uld_ctx *ctx)
+{
+ spin_lock_irq(&ctx->dev->lock);
+ ctx->dev->db_state = FLOW_CONTROL;
+ idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
+ spin_unlock_irq(&ctx->dev->lock);
+}
+
+static int enable_qp_db(int id, void *p, void *data)
+{
+ struct c4iw_qp *qp = p;
+
+ t4_enable_wq_db(&qp->wq);
+ return 0;
+}
+
+static void resume_queues(struct uld_ctx *ctx)
+{
+ spin_lock_irq(&ctx->dev->lock);
+ ctx->dev->db_state = NORMAL;
+ idr_for_each(&ctx->dev->qpidr, enable_qp_db, NULL);
+ spin_unlock_irq(&ctx->dev->lock);
+}
+
+static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
+{
+ struct uld_ctx *ctx = handle;
+
+ switch (control) {
+ case CXGB4_CONTROL_DB_FULL:
+ stop_queues(ctx);
+ mutex_lock(&ctx->dev->rdev.stats.lock);
+ ctx->dev->rdev.stats.db_full++;
+ mutex_unlock(&ctx->dev->rdev.stats.lock);
+ break;
+ case CXGB4_CONTROL_DB_EMPTY:
+ resume_queues(ctx);
+ mutex_lock(&ctx->dev->rdev.stats.lock);
+ ctx->dev->rdev.stats.db_empty++;
+ mutex_unlock(&ctx->dev->rdev.stats.lock);
+ break;
+ case CXGB4_CONTROL_DB_DROP:
+ printk(KERN_WARNING MOD "%s: Fatal DB DROP\n",
+ pci_name(ctx->lldi.pdev));
+ mutex_lock(&ctx->dev->rdev.stats.lock);
+ ctx->dev->rdev.stats.db_drop++;
+ mutex_unlock(&ctx->dev->rdev.stats.lock);
+ break;
+ default:
+ printk(KERN_WARNING MOD "%s: unknown control cmd %u\n",
+ pci_name(ctx->lldi.pdev), control);
+ break;
+ }
+ return 0;
+}
+
static struct cxgb4_uld_info c4iw_uld_info = {
.name = DRV_NAME,
.add = c4iw_uld_add,
.rx_handler = c4iw_uld_rx_handler,
.state_change = c4iw_uld_state_change,
+ .control = c4iw_uld_control,
};
static int __init c4iw_init_module(void)