RDMA/ucma: Allow tuning the max listen backlog

For iWARP connections, the connect request is carried in a TCP payload
on an already established TCP connection.  So if the ucma's backlog is
full, the connection request is transmitted and acked at the TCP level
by the time the connect request gets dropped in the ucma.  The end
result is the connection gets rejected by the iWARP provider.
Further, a 32 node 256NP OpenMPI job will generate > 128 connect
requests on some ranks.

This patch increases the default max backlog to 1024, and adds a
sysctl variable so the backlog can be adjusted at run time.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

1 files changed, 33 insertions, 6 deletions

diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c
index ac7edc24165c..b2748a673989 100644
--- a/drivers/infiniband/core/ucma.c
+++ b/drivers/infiniband/core/ucma.c
@@ -40,6 +40,7 @@
 #include <linux/in6.h>
 #include <linux/miscdevice.h>
 #include <linux/slab.h>
+#include <linux/sysctl.h>
 
 #include <rdma/rdma_user_cm.h>
 #include <rdma/ib_marshall.h>
@@ -50,8 +51,24 @@ MODULE_AUTHOR("Sean Hefty");
 MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access");
 MODULE_LICENSE("Dual BSD/GPL");
 
-enum {
-	UCMA_MAX_BACKLOG	= 128
+static unsigned int max_backlog = 1024;
+
+static struct ctl_table_header *ucma_ctl_table_hdr;
+static ctl_table ucma_ctl_table[] = {
+	{
+		.procname	= "max_backlog",
+		.data		= &max_backlog,
+		.maxlen		= sizeof max_backlog,
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{ }
+};
+
+static struct ctl_path ucma_ctl_path[] = {
+	{ .procname = "net" },
+	{ .procname = "rdma_ucm" },
+	{ }
 };
 
 struct ucma_file {
@@ -686,8 +703,8 @@ static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
-	ctx->backlog = cmd.backlog > 0 && cmd.backlog < UCMA_MAX_BACKLOG ?
-		       cmd.backlog : UCMA_MAX_BACKLOG;
+	ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
+		       cmd.backlog : max_backlog;
 	ret = rdma_listen(ctx->cm_id, ctx->backlog);
 	ucma_put_ctx(ctx);
 	return ret;
@@ -1279,16 +1296,26 @@ static int __init ucma_init(void)
 	ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
 	if (ret) {
 		printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
-		goto err;
+		goto err1;
+	}
+
+	ucma_ctl_table_hdr = register_sysctl_paths(ucma_ctl_path, ucma_ctl_table);
+	if (!ucma_ctl_table_hdr) {
+		printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n");
+		ret = -ENOMEM;
+		goto err2;
 	}
 	return 0;
-err:
+err2:
+	device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
+err1:
 	misc_deregister(&ucma_misc);
 	return ret;
 }
 
 static void __exit ucma_cleanup(void)
 {
+	unregister_sysctl_table(ucma_ctl_table_hdr);
 	device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
 	misc_deregister(&ucma_misc);
 	idr_destroy(&ctx_idr);