* Re: 2.6.31-rc6-git5: Reported regressions from 2.6.30
From: Rafael J. Wysocki @ 2009-08-21 22:02 UTC (permalink / raw)
To: Larry Finger
Cc: Linux Kernel Mailing List, Adrian Bunk, Andrew Morton,
Linus Torvalds, Natalie Protasevich, Kernel Testers List,
Network Development, Linux ACPI, Linux PM List, Linux SCSI List,
Linux Wireless List, DRI
In-Reply-To: <4A8F12CE.9090601@lwfinger.net>
On Friday 21 August 2009, Larry Finger wrote:
> Rafael J. Wysocki wrote:
>
> > Bug-Entry : http://bugzilla.kernel.org/show_bug.cgi?id=13960
> > Subject : rtl8187 not connect to wifi
> > Submitter : okias <d.okias@gmail.com>
> > Date : 2009-08-10 19:16 (10 days old)
>
> The patch for this one was sent from Linville to DaveM earlier today,
> and should be sent to mainline in the near future.
>
> AFAIK, the OP has not yet tested the patch, but I think I was able to
> reproduce the problem, and the patch did fit it for me.
Thanks for the update.
Can you please close the bug when the patch is merged?
Rafael
^ permalink raw reply
* [PATCH net-next 0/5] RDS: TCP transport support
From: Andy Grover @ 2009-08-21 22:28 UTC (permalink / raw)
To: netdev; +Cc: rds-devel
Hello netdev,
This patch series adds a TCP transport for RDS. Instead of requiring
an IB queue pair for communication between two nodes, it may instead
tunnel traffic over a TCP connection. This makes RDS usable by
people without IB hardware, or in a VM.
In addition, this patchset modularizes RDS into an rds core module,
an rds_rdma module (for Infiniband and iWARP) and rds_tcp module. RDS
itself no longer has a dependency on IB (the rds_rdma module does).
This code has been tested and verified to work and perform acceptably,
but has not yet had intensive performance analysis.
Thanks -- Regards -- Andy
^ permalink raw reply
* [PATCH 2/5] RDS: Export symbols from core RDS
From: Andy Grover @ 2009-08-21 22:28 UTC (permalink / raw)
To: netdev; +Cc: rds-devel
In-Reply-To: <1250893715-21939-1-git-send-email-andy.grover@oracle.com>
Now that rdma and tcp transports will be modularized,
we need to export a number of functions so they can call them.
Signed-off-by: Andy Grover <andy.grover@oracle.com>
---
net/rds/cong.c | 1 +
net/rds/connection.c | 5 +++++
net/rds/info.c | 3 +++
net/rds/message.c | 6 ++++++
net/rds/page.c | 1 +
net/rds/recv.c | 4 ++++
net/rds/send.c | 3 +++
net/rds/stats.c | 2 ++
net/rds/threads.c | 2 ++
net/rds/transport.c | 2 ++
10 files changed, 29 insertions(+), 0 deletions(-)
diff --git a/net/rds/cong.c b/net/rds/cong.c
index 710e459..dd2711d 100644
--- a/net/rds/cong.c
+++ b/net/rds/cong.c
@@ -254,6 +254,7 @@ void rds_cong_map_updated(struct rds_cong_map *map, uint64_t portmask)
read_unlock_irqrestore(&rds_cong_monitor_lock, flags);
}
}
+EXPORT_SYMBOL_GPL(rds_cong_map_updated);
int rds_cong_updated_since(unsigned long *recent)
{
diff --git a/net/rds/connection.c b/net/rds/connection.c
index b420a20..cc8b568 100644
--- a/net/rds/connection.c
+++ b/net/rds/connection.c
@@ -255,12 +255,14 @@ struct rds_connection *rds_conn_create(__be32 laddr, __be32 faddr,
{
return __rds_conn_create(laddr, faddr, trans, gfp, 0);
}
+EXPORT_SYMBOL_GPL(rds_conn_create);
struct rds_connection *rds_conn_create_outgoing(__be32 laddr, __be32 faddr,
struct rds_transport *trans, gfp_t gfp)
{
return __rds_conn_create(laddr, faddr, trans, gfp, 1);
}
+EXPORT_SYMBOL_GPL(rds_conn_create_outgoing);
void rds_conn_destroy(struct rds_connection *conn)
{
@@ -303,6 +305,7 @@ void rds_conn_destroy(struct rds_connection *conn)
rds_conn_count--;
}
+EXPORT_SYMBOL_GPL(rds_conn_destroy);
static void rds_conn_message_info(struct socket *sock, unsigned int len,
struct rds_info_iterator *iter,
@@ -406,6 +409,7 @@ void rds_for_each_conn_info(struct socket *sock, unsigned int len,
spin_unlock_irqrestore(&rds_conn_lock, flags);
}
+EXPORT_SYMBOL_GPL(rds_for_each_conn_info);
static int rds_conn_info_visitor(struct rds_connection *conn,
void *buffer)
@@ -481,6 +485,7 @@ void rds_conn_drop(struct rds_connection *conn)
atomic_set(&conn->c_state, RDS_CONN_ERROR);
queue_work(rds_wq, &conn->c_down_w);
}
+EXPORT_SYMBOL_GPL(rds_conn_drop);
/*
* An error occurred on the connection
diff --git a/net/rds/info.c b/net/rds/info.c
index 62aeef3..814a91a 100644
--- a/net/rds/info.c
+++ b/net/rds/info.c
@@ -79,6 +79,7 @@ void rds_info_register_func(int optname, rds_info_func func)
rds_info_funcs[offset] = func;
spin_unlock(&rds_info_lock);
}
+EXPORT_SYMBOL_GPL(rds_info_register_func);
void rds_info_deregister_func(int optname, rds_info_func func)
{
@@ -91,6 +92,7 @@ void rds_info_deregister_func(int optname, rds_info_func func)
rds_info_funcs[offset] = NULL;
spin_unlock(&rds_info_lock);
}
+EXPORT_SYMBOL_GPL(rds_info_deregister_func);
/*
* Typically we hold an atomic kmap across multiple rds_info_copy() calls
@@ -137,6 +139,7 @@ void rds_info_copy(struct rds_info_iterator *iter, void *data,
}
}
}
+EXPORT_SYMBOL_GPL(rds_info_copy);
/*
* @optval points to the userspace buffer that the information snapshot
diff --git a/net/rds/message.c b/net/rds/message.c
index 5a15dc8..ca50a8e 100644
--- a/net/rds/message.c
+++ b/net/rds/message.c
@@ -50,6 +50,7 @@ void rds_message_addref(struct rds_message *rm)
rdsdebug("addref rm %p ref %d\n", rm, atomic_read(&rm->m_refcount));
atomic_inc(&rm->m_refcount);
}
+EXPORT_SYMBOL_GPL(rds_message_addref);
/*
* This relies on dma_map_sg() not touching sg[].page during merging.
@@ -92,6 +93,7 @@ void rds_message_put(struct rds_message *rm)
kfree(rm);
}
}
+EXPORT_SYMBOL_GPL(rds_message_put);
void rds_message_inc_free(struct rds_incoming *inc)
{
@@ -108,6 +110,7 @@ void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
hdr->h_sequence = cpu_to_be64(seq);
hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
}
+EXPORT_SYMBOL_GPL(rds_message_populate_header);
int rds_message_add_extension(struct rds_header *hdr,
unsigned int type, const void *data, unsigned int len)
@@ -133,6 +136,7 @@ int rds_message_add_extension(struct rds_header *hdr,
dst[len] = RDS_EXTHDR_NONE;
return 1;
}
+EXPORT_SYMBOL_GPL(rds_message_add_extension);
/*
* If a message has extension headers, retrieve them here.
@@ -208,6 +212,7 @@ int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 o
ext_hdr.h_rdma_offset = cpu_to_be32(offset);
return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
}
+EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
struct rds_message *rds_message_alloc(unsigned int nents, gfp_t gfp)
{
@@ -399,4 +404,5 @@ void rds_message_unmapped(struct rds_message *rm)
if (waitqueue_active(&rds_message_flush_waitq))
wake_up(&rds_message_flush_waitq);
}
+EXPORT_SYMBOL_GPL(rds_message_unmapped);
diff --git a/net/rds/page.c b/net/rds/page.c
index c460743..55c21ef 100644
--- a/net/rds/page.c
+++ b/net/rds/page.c
@@ -81,6 +81,7 @@ int rds_page_copy_user(struct page *page, unsigned long offset,
return 0;
}
+EXPORT_SYMBOL_GPL(rds_page_copy_user);
/*
* Message allocation uses this to build up regions of a message.
diff --git a/net/rds/recv.c b/net/rds/recv.c
index 86bc1a0..fdff33c 100644
--- a/net/rds/recv.c
+++ b/net/rds/recv.c
@@ -46,12 +46,14 @@ void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn,
inc->i_saddr = saddr;
inc->i_rdma_cookie = 0;
}
+EXPORT_SYMBOL_GPL(rds_inc_init);
void rds_inc_addref(struct rds_incoming *inc)
{
rdsdebug("addref inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
atomic_inc(&inc->i_refcount);
}
+EXPORT_SYMBOL_GPL(rds_inc_addref);
void rds_inc_put(struct rds_incoming *inc)
{
@@ -62,6 +64,7 @@ void rds_inc_put(struct rds_incoming *inc)
inc->i_conn->c_trans->inc_free(inc);
}
}
+EXPORT_SYMBOL_GPL(rds_inc_put);
static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk,
struct rds_cong_map *map,
@@ -237,6 +240,7 @@ out:
if (rs)
rds_sock_put(rs);
}
+EXPORT_SYMBOL_GPL(rds_recv_incoming);
/*
* be very careful here. This is being called as the condition in
diff --git a/net/rds/send.c b/net/rds/send.c
index a4a7f42..28c88ff 100644
--- a/net/rds/send.c
+++ b/net/rds/send.c
@@ -439,6 +439,7 @@ void rds_rdma_send_complete(struct rds_message *rm, int status)
sock_put(rds_rs_to_sk(rs));
}
}
+EXPORT_SYMBOL_GPL(rds_rdma_send_complete);
/*
* This is the same as rds_rdma_send_complete except we
@@ -494,6 +495,7 @@ out:
return found;
}
+EXPORT_SYMBOL_GPL(rds_send_get_message);
/*
* This removes messages from the socket's list if they're on it. The list
@@ -610,6 +612,7 @@ void rds_send_drop_acked(struct rds_connection *conn, u64 ack,
/* now remove the messages from the sock list as needed */
rds_send_remove_from_sock(&list, RDS_RDMA_SUCCESS);
}
+EXPORT_SYMBOL_GPL(rds_send_drop_acked);
void rds_send_drop_to(struct rds_sock *rs, struct sockaddr_in *dest)
{
diff --git a/net/rds/stats.c b/net/rds/stats.c
index 91d8c58..7598eb0 100644
--- a/net/rds/stats.c
+++ b/net/rds/stats.c
@@ -37,6 +37,7 @@
#include "rds.h"
DEFINE_PER_CPU_SHARED_ALIGNED(struct rds_statistics, rds_stats);
+EXPORT_PER_CPU_SYMBOL_GPL(rds_stats);
/* :.,$s/unsigned long\>.*\<s_\(.*\);/"\1",/g */
@@ -90,6 +91,7 @@ void rds_stats_info_copy(struct rds_info_iterator *iter,
rds_info_copy(iter, &ctr, sizeof(ctr));
}
}
+EXPORT_SYMBOL_GPL(rds_stats_info_copy);
/*
* This gives global counters across all the transports. The strings
diff --git a/net/rds/threads.c b/net/rds/threads.c
index 828a1bf..dd7e0ca 100644
--- a/net/rds/threads.c
+++ b/net/rds/threads.c
@@ -68,6 +68,7 @@
* (TCP, IB/RDMA) to provide the necessary synchronisation.
*/
struct workqueue_struct *rds_wq;
+EXPORT_SYMBOL_GPL(rds_wq);
void rds_connect_complete(struct rds_connection *conn)
{
@@ -89,6 +90,7 @@ void rds_connect_complete(struct rds_connection *conn)
queue_delayed_work(rds_wq, &conn->c_send_w, 0);
queue_delayed_work(rds_wq, &conn->c_recv_w, 0);
}
+EXPORT_SYMBOL_GPL(rds_connect_complete);
/*
* This random exponential backoff is relied on to eventually resolve racing
diff --git a/net/rds/transport.c b/net/rds/transport.c
index 767da61..56a5309 100644
--- a/net/rds/transport.c
+++ b/net/rds/transport.c
@@ -53,6 +53,7 @@ int rds_trans_register(struct rds_transport *trans)
return 0;
}
+EXPORT_SYMBOL_GPL(rds_trans_register);
void rds_trans_unregister(struct rds_transport *trans)
{
@@ -63,6 +64,7 @@ void rds_trans_unregister(struct rds_transport *trans)
up_write(&rds_trans_sem);
}
+EXPORT_SYMBOL_GPL(rds_trans_unregister);
struct rds_transport *rds_trans_get_preferred(__be32 addr)
{
--
1.6.0.4
^ permalink raw reply related
* [PATCH 1/5] RDS: Add TCP transport to RDS
From: Andy Grover @ 2009-08-21 22:28 UTC (permalink / raw)
To: netdev; +Cc: rds-devel
In-Reply-To: <1250893715-21939-1-git-send-email-andy.grover@oracle.com>
This code allows RDS to be tunneled over a TCP connection.
RDMA operations are disabled when using TCP transport,
but this frees RDS from the IB/RDMA stack dependency, and allows
it to be used with standard Ethernet adapters, or in a VM.
Signed-off-by: Andy Grover <andy.grover@oracle.com>
---
include/linux/rds.h | 12 ++
net/rds/tcp.c | 319 +++++++++++++++++++++++++++++++++++++++++++
net/rds/tcp.h | 93 +++++++++++++
net/rds/tcp_connect.c | 153 +++++++++++++++++++++
net/rds/tcp_listen.c | 199 +++++++++++++++++++++++++++
net/rds/tcp_recv.c | 356 +++++++++++++++++++++++++++++++++++++++++++++++++
net/rds/tcp_send.c | 263 ++++++++++++++++++++++++++++++++++++
net/rds/tcp_stats.c | 74 ++++++++++
8 files changed, 1469 insertions(+), 0 deletions(-)
create mode 100644 net/rds/tcp.c
create mode 100644 net/rds/tcp.h
create mode 100644 net/rds/tcp_connect.c
create mode 100644 net/rds/tcp_listen.c
create mode 100644 net/rds/tcp_recv.c
create mode 100644 net/rds/tcp_send.c
create mode 100644 net/rds/tcp_stats.c
diff --git a/include/linux/rds.h b/include/linux/rds.h
index d91dc91..89d46e1 100644
--- a/include/linux/rds.h
+++ b/include/linux/rds.h
@@ -147,6 +147,18 @@ struct rds_info_socket {
u_int64_t inum;
} __attribute__((packed));
+struct rds_info_tcp_socket {
+ __be32 local_addr;
+ __be16 local_port;
+ __be32 peer_addr;
+ __be16 peer_port;
+ u_int64_t hdr_rem;
+ u_int64_t data_rem;
+ u_int32_t last_sent_nxt;
+ u_int32_t last_expected_una;
+ u_int32_t last_seen_una;
+} __attribute__((packed));
+
#define RDS_IB_GID_LEN 16
struct rds_info_rdma_connection {
__be32 src_addr;
diff --git a/net/rds/tcp.c b/net/rds/tcp.c
new file mode 100644
index 0000000..e0ac900
--- /dev/null
+++ b/net/rds/tcp.c
@@ -0,0 +1,319 @@
+/*
+ * Copyright (c) 2006 Oracle. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/kernel.h>
+#include <linux/in.h>
+#include <net/tcp.h>
+
+#include "rds.h"
+#include "tcp.h"
+
+/* only for info exporting */
+static DEFINE_SPINLOCK(rds_tcp_tc_list_lock);
+static LIST_HEAD(rds_tcp_tc_list);
+unsigned int rds_tcp_tc_count;
+
+/* Track rds_tcp_connection structs so they can be cleaned up */
+static DEFINE_SPINLOCK(rds_tcp_conn_lock);
+static LIST_HEAD(rds_tcp_conn_list);
+
+static struct kmem_cache *rds_tcp_conn_slab;
+
+#define RDS_TCP_DEFAULT_BUFSIZE (128 * 1024)
+
+/* doing it this way avoids calling tcp_sk() */
+void rds_tcp_nonagle(struct socket *sock)
+{
+ mm_segment_t oldfs = get_fs();
+ int val = 1;
+
+ set_fs(KERNEL_DS);
+ sock->ops->setsockopt(sock, SOL_TCP, TCP_NODELAY, (char __user *)&val,
+ sizeof(val));
+ set_fs(oldfs);
+}
+
+void rds_tcp_tune(struct socket *sock)
+{
+ struct sock *sk = sock->sk;
+
+ rds_tcp_nonagle(sock);
+
+ /*
+ * We're trying to saturate gigabit with the default,
+ * see svc_sock_setbufsize().
+ */
+ lock_sock(sk);
+ sk->sk_sndbuf = RDS_TCP_DEFAULT_BUFSIZE;
+ sk->sk_rcvbuf = RDS_TCP_DEFAULT_BUFSIZE;
+ sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
+ release_sock(sk);
+}
+
+u32 rds_tcp_snd_nxt(struct rds_tcp_connection *tc)
+{
+ return tcp_sk(tc->t_sock->sk)->snd_nxt;
+}
+
+u32 rds_tcp_snd_una(struct rds_tcp_connection *tc)
+{
+ return tcp_sk(tc->t_sock->sk)->snd_una;
+}
+
+void rds_tcp_restore_callbacks(struct socket *sock,
+ struct rds_tcp_connection *tc)
+{
+ rdsdebug("restoring sock %p callbacks from tc %p\n", sock, tc);
+ write_lock_bh(&sock->sk->sk_callback_lock);
+
+ /* done under the callback_lock to serialize with write_space */
+ spin_lock(&rds_tcp_tc_list_lock);
+ list_del_init(&tc->t_list_item);
+ rds_tcp_tc_count--;
+ spin_unlock(&rds_tcp_tc_list_lock);
+
+ tc->t_sock = NULL;
+
+ sock->sk->sk_write_space = tc->t_orig_write_space;
+ sock->sk->sk_data_ready = tc->t_orig_data_ready;
+ sock->sk->sk_state_change = tc->t_orig_state_change;
+ sock->sk->sk_user_data = NULL;
+
+ write_unlock_bh(&sock->sk->sk_callback_lock);
+}
+
+/*
+ * This is the only path that sets tc->t_sock. Send and receive trust that
+ * it is set. The RDS_CONN_CONNECTED bit protects those paths from being
+ * called while it isn't set.
+ */
+void rds_tcp_set_callbacks(struct socket *sock, struct rds_connection *conn)
+{
+ struct rds_tcp_connection *tc = conn->c_transport_data;
+
+ rdsdebug("setting sock %p callbacks to tc %p\n", sock, tc);
+ write_lock_bh(&sock->sk->sk_callback_lock);
+
+ /* done under the callback_lock to serialize with write_space */
+ spin_lock(&rds_tcp_tc_list_lock);
+ list_add_tail(&tc->t_list_item, &rds_tcp_tc_list);
+ rds_tcp_tc_count++;
+ spin_unlock(&rds_tcp_tc_list_lock);
+
+ /* accepted sockets need our listen data ready undone */
+ if (sock->sk->sk_data_ready == rds_tcp_listen_data_ready)
+ sock->sk->sk_data_ready = sock->sk->sk_user_data;
+
+ tc->t_sock = sock;
+ tc->conn = conn;
+ tc->t_orig_data_ready = sock->sk->sk_data_ready;
+ tc->t_orig_write_space = sock->sk->sk_write_space;
+ tc->t_orig_state_change = sock->sk->sk_state_change;
+
+ sock->sk->sk_user_data = conn;
+ sock->sk->sk_data_ready = rds_tcp_data_ready;
+ sock->sk->sk_write_space = rds_tcp_write_space;
+ sock->sk->sk_state_change = rds_tcp_state_change;
+
+ write_unlock_bh(&sock->sk->sk_callback_lock);
+}
+
+static void rds_tcp_tc_info(struct socket *sock, unsigned int len,
+ struct rds_info_iterator *iter,
+ struct rds_info_lengths *lens)
+{
+ struct rds_info_tcp_socket tsinfo;
+ struct rds_tcp_connection *tc;
+ unsigned long flags;
+ struct sockaddr_in sin;
+ int sinlen;
+
+ spin_lock_irqsave(&rds_tcp_tc_list_lock, flags);
+
+ if (len / sizeof(tsinfo) < rds_tcp_tc_count)
+ goto out;
+
+ list_for_each_entry(tc, &rds_tcp_tc_list, t_list_item) {
+
+ sock->ops->getname(sock, (struct sockaddr *)&sin, &sinlen, 0);
+ tsinfo.local_addr = sin.sin_addr.s_addr;
+ tsinfo.local_port = sin.sin_port;
+ sock->ops->getname(sock, (struct sockaddr *)&sin, &sinlen, 1);
+ tsinfo.peer_addr = sin.sin_addr.s_addr;
+ tsinfo.peer_port = sin.sin_port;
+
+ tsinfo.hdr_rem = tc->t_tinc_hdr_rem;
+ tsinfo.data_rem = tc->t_tinc_data_rem;
+ tsinfo.last_sent_nxt = tc->t_last_sent_nxt;
+ tsinfo.last_expected_una = tc->t_last_expected_una;
+ tsinfo.last_seen_una = tc->t_last_seen_una;
+
+ rds_info_copy(iter, &tsinfo, sizeof(tsinfo));
+ }
+
+out:
+ lens->nr = rds_tcp_tc_count;
+ lens->each = sizeof(tsinfo);
+
+ spin_unlock_irqrestore(&rds_tcp_tc_list_lock, flags);
+}
+
+static int rds_tcp_laddr_check(__be32 addr)
+{
+ if (inet_addr_type(&init_net, addr) == RTN_LOCAL)
+ return 0;
+ return -EADDRNOTAVAIL;
+}
+
+static int rds_tcp_conn_alloc(struct rds_connection *conn, gfp_t gfp)
+{
+ struct rds_tcp_connection *tc;
+
+ tc = kmem_cache_alloc(rds_tcp_conn_slab, gfp);
+ if (tc == NULL)
+ return -ENOMEM;
+
+ tc->t_sock = NULL;
+ tc->t_tinc = NULL;
+ tc->t_tinc_hdr_rem = sizeof(struct rds_header);
+ tc->t_tinc_data_rem = 0;
+
+ conn->c_transport_data = tc;
+
+ spin_lock_irq(&rds_tcp_conn_lock);
+ list_add_tail(&tc->t_tcp_node, &rds_tcp_conn_list);
+ spin_unlock_irq(&rds_tcp_conn_lock);
+
+ rdsdebug("alloced tc %p\n", conn->c_transport_data);
+ return 0;
+}
+
+static void rds_tcp_conn_free(void *arg)
+{
+ struct rds_tcp_connection *tc = arg;
+ rdsdebug("freeing tc %p\n", tc);
+ kmem_cache_free(rds_tcp_conn_slab, tc);
+}
+
+static void rds_tcp_destroy_conns(void)
+{
+ struct rds_tcp_connection *tc, *_tc;
+ LIST_HEAD(tmp_list);
+
+ /* avoid calling conn_destroy with irqs off */
+ spin_lock_irq(&rds_tcp_conn_lock);
+ list_splice(&rds_tcp_conn_list, &tmp_list);
+ INIT_LIST_HEAD(&rds_tcp_conn_list);
+ spin_unlock_irq(&rds_tcp_conn_lock);
+
+ list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) {
+ if (tc->conn->c_passive)
+ rds_conn_destroy(tc->conn->c_passive);
+ rds_conn_destroy(tc->conn);
+ }
+}
+
+void rds_tcp_exit(void)
+{
+ rds_info_deregister_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info);
+ rds_tcp_listen_stop();
+ rds_tcp_destroy_conns();
+ rds_trans_unregister(&rds_tcp_transport);
+ rds_tcp_recv_exit();
+ kmem_cache_destroy(rds_tcp_conn_slab);
+}
+module_exit(rds_tcp_exit);
+
+struct rds_transport rds_tcp_transport = {
+ .laddr_check = rds_tcp_laddr_check,
+ .xmit_prepare = rds_tcp_xmit_prepare,
+ .xmit_complete = rds_tcp_xmit_complete,
+ .xmit_cong_map = rds_tcp_xmit_cong_map,
+ .xmit = rds_tcp_xmit,
+ .recv = rds_tcp_recv,
+ .conn_alloc = rds_tcp_conn_alloc,
+ .conn_free = rds_tcp_conn_free,
+ .conn_connect = rds_tcp_conn_connect,
+ .conn_shutdown = rds_tcp_conn_shutdown,
+ .inc_copy_to_user = rds_tcp_inc_copy_to_user,
+ .inc_purge = rds_tcp_inc_purge,
+ .inc_free = rds_tcp_inc_free,
+ .stats_info_copy = rds_tcp_stats_info_copy,
+ .exit = rds_tcp_exit,
+ .t_owner = THIS_MODULE,
+ .t_name = "tcp",
+ .t_prefer_loopback = 1,
+};
+
+int __init rds_tcp_init(void)
+{
+ int ret;
+
+ rds_tcp_conn_slab = kmem_cache_create("rds_tcp_connection",
+ sizeof(struct rds_tcp_connection),
+ 0, 0, NULL);
+ if (rds_tcp_conn_slab == NULL) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ ret = rds_tcp_recv_init();
+ if (ret)
+ goto out_slab;
+
+ ret = rds_trans_register(&rds_tcp_transport);
+ if (ret)
+ goto out_recv;
+
+ ret = rds_tcp_listen_init();
+ if (ret)
+ goto out_register;
+
+ rds_info_register_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info);
+
+ goto out;
+
+out_register:
+ rds_trans_unregister(&rds_tcp_transport);
+out_recv:
+ rds_tcp_recv_exit();
+out_slab:
+ kmem_cache_destroy(rds_tcp_conn_slab);
+out:
+ return ret;
+}
+module_init(rds_tcp_init);
+
+MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
+MODULE_DESCRIPTION("RDS: TCP transport");
+MODULE_LICENSE("Dual BSD/GPL");
+
diff --git a/net/rds/tcp.h b/net/rds/tcp.h
new file mode 100644
index 0000000..844fa6b
--- /dev/null
+++ b/net/rds/tcp.h
@@ -0,0 +1,93 @@
+#ifndef _RDS_TCP_H
+#define _RDS_TCP_H
+
+#define RDS_TCP_PORT 16385
+
+struct rds_tcp_incoming {
+ struct rds_incoming ti_inc;
+ struct sk_buff_head ti_skb_list;
+};
+
+struct rds_tcp_connection {
+
+ struct list_head t_tcp_node;
+ struct rds_connection *conn;
+ struct socket *t_sock;
+ void *t_orig_write_space;
+ void *t_orig_data_ready;
+ void *t_orig_state_change;
+
+ struct rds_tcp_incoming *t_tinc;
+ size_t t_tinc_hdr_rem;
+ size_t t_tinc_data_rem;
+
+ /* XXX error report? */
+ struct work_struct t_conn_w;
+ struct work_struct t_send_w;
+ struct work_struct t_down_w;
+ struct work_struct t_recv_w;
+
+ /* for info exporting only */
+ struct list_head t_list_item;
+ u32 t_last_sent_nxt;
+ u32 t_last_expected_una;
+ u32 t_last_seen_una;
+};
+
+struct rds_tcp_statistics {
+ uint64_t s_tcp_data_ready_calls;
+ uint64_t s_tcp_write_space_calls;
+ uint64_t s_tcp_sndbuf_full;
+ uint64_t s_tcp_connect_raced;
+ uint64_t s_tcp_listen_closed_stale;
+};
+
+/* tcp.c */
+int __init rds_tcp_init(void);
+void rds_tcp_exit(void);
+void rds_tcp_tune(struct socket *sock);
+void rds_tcp_nonagle(struct socket *sock);
+void rds_tcp_set_callbacks(struct socket *sock, struct rds_connection *conn);
+void rds_tcp_restore_callbacks(struct socket *sock,
+ struct rds_tcp_connection *tc);
+u32 rds_tcp_snd_nxt(struct rds_tcp_connection *tc);
+u32 rds_tcp_snd_una(struct rds_tcp_connection *tc);
+u64 rds_tcp_map_seq(struct rds_tcp_connection *tc, u32 seq);
+extern struct rds_transport rds_tcp_transport;
+
+/* tcp_connect.c */
+int rds_tcp_conn_connect(struct rds_connection *conn);
+void rds_tcp_conn_shutdown(struct rds_connection *conn);
+void rds_tcp_state_change(struct sock *sk);
+
+/* tcp_listen.c */
+int __init rds_tcp_listen_init(void);
+void rds_tcp_listen_stop(void);
+void rds_tcp_listen_data_ready(struct sock *sk, int bytes);
+
+/* tcp_recv.c */
+int __init rds_tcp_recv_init(void);
+void rds_tcp_recv_exit(void);
+void rds_tcp_data_ready(struct sock *sk, int bytes);
+int rds_tcp_recv(struct rds_connection *conn);
+void rds_tcp_inc_purge(struct rds_incoming *inc);
+void rds_tcp_inc_free(struct rds_incoming *inc);
+int rds_tcp_inc_copy_to_user(struct rds_incoming *inc, struct iovec *iov,
+ size_t size);
+
+/* tcp_send.c */
+void rds_tcp_xmit_prepare(struct rds_connection *conn);
+void rds_tcp_xmit_complete(struct rds_connection *conn);
+int rds_tcp_xmit(struct rds_connection *conn, struct rds_message *rm,
+ unsigned int hdr_off, unsigned int sg, unsigned int off);
+void rds_tcp_write_space(struct sock *sk);
+int rds_tcp_xmit_cong_map(struct rds_connection *conn,
+ struct rds_cong_map *map, unsigned long offset);
+
+/* tcp_stats.c */
+DECLARE_PER_CPU(struct rds_tcp_statistics, rds_tcp_stats);
+#define rds_tcp_stats_inc(member) rds_stats_inc_which(rds_tcp_stats, member)
+unsigned int rds_tcp_stats_info_copy(struct rds_info_iterator *iter,
+ unsigned int avail);
+
+#endif
diff --git a/net/rds/tcp_connect.c b/net/rds/tcp_connect.c
new file mode 100644
index 0000000..211522f
--- /dev/null
+++ b/net/rds/tcp_connect.c
@@ -0,0 +1,153 @@
+/*
+ * Copyright (c) 2006 Oracle. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/kernel.h>
+#include <linux/in.h>
+#include <net/tcp.h>
+
+#include "rds.h"
+#include "tcp.h"
+
+void rds_tcp_state_change(struct sock *sk)
+{
+ void (*state_change)(struct sock *sk);
+ struct rds_connection *conn;
+ struct rds_tcp_connection *tc;
+
+ read_lock(&sk->sk_callback_lock);
+ conn = sk->sk_user_data;
+ if (conn == NULL) {
+ state_change = sk->sk_state_change;
+ goto out;
+ }
+ tc = conn->c_transport_data;
+ state_change = tc->t_orig_state_change;
+
+ rdsdebug("sock %p state_change to %d\n", tc->t_sock, sk->sk_state);
+
+ switch(sk->sk_state) {
+ /* ignore connecting sockets as they make progress */
+ case TCP_SYN_SENT:
+ case TCP_SYN_RECV:
+ break;
+ case TCP_ESTABLISHED:
+ rds_connect_complete(conn);
+ break;
+ case TCP_CLOSE:
+ rds_conn_drop(conn);
+ default:
+ break;
+ }
+out:
+ read_unlock(&sk->sk_callback_lock);
+ state_change(sk);
+}
+
+int rds_tcp_conn_connect(struct rds_connection *conn)
+{
+ struct socket *sock = NULL;
+ struct sockaddr_in src, dest;
+ int ret;
+
+ ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
+ if (ret < 0)
+ goto out;
+
+ rds_tcp_tune(sock);
+
+ src.sin_family = AF_INET;
+ src.sin_addr.s_addr = (__force u32)conn->c_laddr;
+ src.sin_port = (__force u16)htons(0);
+
+ ret = sock->ops->bind(sock, (struct sockaddr *)&src, sizeof(src));
+ if (ret) {
+ rdsdebug("bind failed with %d at address %u.%u.%u.%u\n",
+ ret, NIPQUAD(conn->c_laddr));
+ goto out;
+ }
+
+ dest.sin_family = AF_INET;
+ dest.sin_addr.s_addr = (__force u32)conn->c_faddr;
+ dest.sin_port = (__force u16)htons(RDS_TCP_PORT);
+
+ /*
+ * once we call connect() we can start getting callbacks and they
+ * own the socket
+ */
+ rds_tcp_set_callbacks(sock, conn);
+ ret = sock->ops->connect(sock, (struct sockaddr *)&dest, sizeof(dest),
+ O_NONBLOCK);
+ sock = NULL;
+
+ rdsdebug("connect to address %u.%u.%u.%u returned %d\n",
+ NIPQUAD(conn->c_faddr), ret);
+ if (ret == -EINPROGRESS)
+ ret = 0;
+
+out:
+ if (sock)
+ sock_release(sock);
+ return ret;
+}
+
+/*
+ * Before killing the tcp socket this needs to serialize with callbacks. The
+ * caller has already grabbed the sending sem so we're serialized with other
+ * senders.
+ *
+ * TCP calls the callbacks with the sock lock so we hold it while we reset the
+ * callbacks to those set by TCP. Our callbacks won't execute again once we
+ * hold the sock lock.
+ */
+void rds_tcp_conn_shutdown(struct rds_connection *conn)
+{
+ struct rds_tcp_connection *tc = conn->c_transport_data;
+ struct socket *sock = tc->t_sock;
+
+ rdsdebug("shutting down conn %p tc %p sock %p\n", conn, tc, sock);
+
+ if (sock) {
+ sock->ops->shutdown(sock, RCV_SHUTDOWN | SEND_SHUTDOWN);
+ lock_sock(sock->sk);
+ rds_tcp_restore_callbacks(sock, tc); /* tc->tc_sock = NULL */
+
+ release_sock(sock->sk);
+ sock_release(sock);
+ };
+
+ if (tc->t_tinc) {
+ rds_inc_put(&tc->t_tinc->ti_inc);
+ tc->t_tinc = NULL;
+ }
+ tc->t_tinc_hdr_rem = sizeof(struct rds_header);
+ tc->t_tinc_data_rem = 0;
+}
diff --git a/net/rds/tcp_listen.c b/net/rds/tcp_listen.c
new file mode 100644
index 0000000..24b743e
--- /dev/null
+++ b/net/rds/tcp_listen.c
@@ -0,0 +1,199 @@
+/*
+ * Copyright (c) 2006 Oracle. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/kernel.h>
+#include <linux/in.h>
+#include <net/tcp.h>
+
+#include "rds.h"
+#include "tcp.h"
+
+/*
+ * cheesy, but simple..
+ */
+static void rds_tcp_accept_worker(struct work_struct *work);
+static DECLARE_WORK(rds_tcp_listen_work, rds_tcp_accept_worker);
+static struct socket *rds_tcp_listen_sock;
+
+static int rds_tcp_accept_one(struct socket *sock)
+{
+ struct socket *new_sock = NULL;
+ struct rds_connection *conn;
+ int ret;
+ struct inet_sock *inet;
+
+ ret = sock_create_lite(sock->sk->sk_family, sock->sk->sk_type,
+ sock->sk->sk_protocol, &new_sock);
+ if (ret)
+ goto out;
+
+ new_sock->type = sock->type;
+ new_sock->ops = sock->ops;
+ ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
+ if (ret < 0)
+ goto out;
+
+ rds_tcp_tune(new_sock);
+
+ inet = inet_sk(new_sock->sk);
+
+ rdsdebug("accepted tcp %u.%u.%u.%u:%u -> %u.%u.%u.%u:%u\n",
+ NIPQUAD(inet->saddr), ntohs(inet->sport),
+ NIPQUAD(inet->daddr), ntohs(inet->dport));
+
+ conn = rds_conn_create(inet->saddr, inet->daddr, &rds_tcp_transport,
+ GFP_KERNEL);
+ if (IS_ERR(conn)) {
+ ret = PTR_ERR(conn);
+ goto out;
+ }
+
+ /*
+ * see the comment above rds_queue_delayed_reconnect()
+ */
+ if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) {
+ if (rds_conn_state(conn) == RDS_CONN_UP)
+ rds_tcp_stats_inc(s_tcp_listen_closed_stale);
+ else
+ rds_tcp_stats_inc(s_tcp_connect_raced);
+ rds_conn_drop(conn);
+ ret = 0;
+ goto out;
+ }
+
+ rds_tcp_set_callbacks(new_sock, conn);
+ rds_connect_complete(conn);
+ new_sock = NULL;
+ ret = 0;
+
+out:
+ if (new_sock)
+ sock_release(new_sock);
+ return ret;
+}
+
+static void rds_tcp_accept_worker(struct work_struct *work)
+{
+ while (rds_tcp_accept_one(rds_tcp_listen_sock) == 0)
+ cond_resched();
+}
+
+void rds_tcp_listen_data_ready(struct sock *sk, int bytes)
+{
+ void (*ready)(struct sock *sk, int bytes);
+
+ rdsdebug("listen data ready sk %p\n", sk);
+
+ read_lock(&sk->sk_callback_lock);
+ ready = sk->sk_user_data;
+ if (ready == NULL) { /* check for teardown race */
+ ready = sk->sk_data_ready;
+ goto out;
+ }
+
+ /*
+ * ->sk_data_ready is also called for a newly established child socket
+ * before it has been accepted and the accepter has set up their
+ * data_ready.. we only want to queue listen work for our listening
+ * socket
+ */
+ if (sk->sk_state == TCP_LISTEN)
+ queue_work(rds_wq, &rds_tcp_listen_work);
+
+out:
+ read_unlock(&sk->sk_callback_lock);
+ ready(sk, bytes);
+}
+
+int __init rds_tcp_listen_init(void)
+{
+ struct sockaddr_in sin;
+ struct socket *sock = NULL;
+ int ret;
+
+ ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
+ if (ret < 0)
+ goto out;
+
+ sock->sk->sk_reuse = 1;
+ rds_tcp_nonagle(sock);
+
+ write_lock_bh(&sock->sk->sk_callback_lock);
+ sock->sk->sk_user_data = sock->sk->sk_data_ready;
+ sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
+ write_unlock_bh(&sock->sk->sk_callback_lock);
+
+ sin.sin_family = PF_INET,
+ sin.sin_addr.s_addr = (__force u32)htonl(INADDR_ANY);
+ sin.sin_port = (__force u16)htons(RDS_TCP_PORT);
+
+ ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
+ if (ret < 0)
+ goto out;
+
+ ret = sock->ops->listen(sock, 64);
+ if (ret < 0)
+ goto out;
+
+ rds_tcp_listen_sock = sock;
+ sock = NULL;
+out:
+ if (sock)
+ sock_release(sock);
+ return ret;
+}
+
+void rds_tcp_listen_stop(void)
+{
+ struct socket *sock = rds_tcp_listen_sock;
+ struct sock *sk;
+
+ if (sock == NULL)
+ return;
+
+ sk = sock->sk;
+
+ /* serialize with and prevent further callbacks */
+ lock_sock(sk);
+ write_lock_bh(&sk->sk_callback_lock);
+ if (sk->sk_user_data) {
+ sk->sk_data_ready = sk->sk_user_data;
+ sk->sk_user_data = NULL;
+ }
+ write_unlock_bh(&sk->sk_callback_lock);
+ release_sock(sk);
+
+ /* wait for accepts to stop and close the socket */
+ flush_workqueue(rds_wq);
+ sock_release(sock);
+ rds_tcp_listen_sock = NULL;
+}
diff --git a/net/rds/tcp_recv.c b/net/rds/tcp_recv.c
new file mode 100644
index 0000000..c00daff
--- /dev/null
+++ b/net/rds/tcp_recv.c
@@ -0,0 +1,356 @@
+/*
+ * Copyright (c) 2006 Oracle. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/kernel.h>
+#include <net/tcp.h>
+
+#include "rds.h"
+#include "tcp.h"
+
+static struct kmem_cache *rds_tcp_incoming_slab;
+
+void rds_tcp_inc_purge(struct rds_incoming *inc)
+{
+ struct rds_tcp_incoming *tinc;
+ tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
+ rdsdebug("purging tinc %p inc %p\n", tinc, inc);
+ skb_queue_purge(&tinc->ti_skb_list);
+}
+
+void rds_tcp_inc_free(struct rds_incoming *inc)
+{
+ struct rds_tcp_incoming *tinc;
+ tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
+ rds_tcp_inc_purge(inc);
+ rdsdebug("freeing tinc %p inc %p\n", tinc, inc);
+ kmem_cache_free(rds_tcp_incoming_slab, tinc);
+}
+
+/*
+ * this is pretty lame, but, whatever.
+ */
+int rds_tcp_inc_copy_to_user(struct rds_incoming *inc, struct iovec *first_iov,
+ size_t size)
+{
+ struct rds_tcp_incoming *tinc;
+ struct iovec *iov, tmp;
+ struct sk_buff *skb;
+ unsigned long to_copy, skb_off;
+ int ret = 0;
+
+ if (size == 0)
+ goto out;
+
+ tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
+ iov = first_iov;
+ tmp = *iov;
+
+ skb_queue_walk(&tinc->ti_skb_list, skb) {
+ skb_off = 0;
+ while (skb_off < skb->len) {
+ while (tmp.iov_len == 0) {
+ iov++;
+ tmp = *iov;
+ }
+
+ to_copy = min(tmp.iov_len, size);
+ to_copy = min(to_copy, skb->len - skb_off);
+
+ rdsdebug("ret %d size %zu skb %p skb_off %lu "
+ "skblen %d iov_base %p iov_len %zu cpy %lu\n",
+ ret, size, skb, skb_off, skb->len,
+ tmp.iov_base, tmp.iov_len, to_copy);
+
+ /* modifies tmp as it copies */
+ if (skb_copy_datagram_iovec(skb, skb_off, &tmp,
+ to_copy)) {
+ ret = -EFAULT;
+ goto out;
+ }
+
+ size -= to_copy;
+ ret += to_copy;
+ skb_off += to_copy;
+ if (size == 0)
+ goto out;
+ }
+ }
+out:
+ return ret;
+}
+
+/*
+ * We have a series of skbs that have fragmented pieces of the congestion
+ * bitmap. They must add up to the exact size of the congestion bitmap. We
+ * use the skb helpers to copy those into the pages that make up the in-memory
+ * congestion bitmap for the remote address of this connection. We then tell
+ * the congestion core that the bitmap has been changed so that it can wake up
+ * sleepers.
+ *
+ * This is racing with sending paths which are using test_bit to see if the
+ * bitmap indicates that their recipient is congested.
+ */
+
+static void rds_tcp_cong_recv(struct rds_connection *conn,
+ struct rds_tcp_incoming *tinc)
+{
+ struct sk_buff *skb;
+ unsigned int to_copy, skb_off;
+ unsigned int map_off;
+ unsigned int map_page;
+ struct rds_cong_map *map;
+ int ret;
+
+ /* catch completely corrupt packets */
+ if (be32_to_cpu(tinc->ti_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES)
+ return;
+
+ map_page = 0;
+ map_off = 0;
+ map = conn->c_fcong;
+
+ skb_queue_walk(&tinc->ti_skb_list, skb) {
+ skb_off = 0;
+ while (skb_off < skb->len) {
+ to_copy = min_t(unsigned int, PAGE_SIZE - map_off,
+ skb->len - skb_off);
+
+ BUG_ON(map_page >= RDS_CONG_MAP_PAGES);
+
+ /* only returns 0 or -error */
+ ret = skb_copy_bits(skb, skb_off,
+ (void *)map->m_page_addrs[map_page] + map_off,
+ to_copy);
+ BUG_ON(ret != 0);
+
+ skb_off += to_copy;
+ map_off += to_copy;
+ if (map_off == PAGE_SIZE) {
+ map_off = 0;
+ map_page++;
+ }
+ }
+ }
+
+ rds_cong_map_updated(map, ~(u64) 0);
+}
+
+struct rds_tcp_desc_arg {
+ struct rds_connection *conn;
+ gfp_t gfp;
+ enum km_type km;
+};
+
+static int rds_tcp_data_recv(read_descriptor_t *desc, struct sk_buff *skb,
+ unsigned int offset, size_t len)
+{
+ struct rds_tcp_desc_arg *arg = desc->arg.data;
+ struct rds_connection *conn = arg->conn;
+ struct rds_tcp_connection *tc = conn->c_transport_data;
+ struct rds_tcp_incoming *tinc = tc->t_tinc;
+ struct sk_buff *clone;
+ size_t left = len, to_copy;
+
+ rdsdebug("tcp data tc %p skb %p offset %u len %zu\n", tc, skb, offset,
+ len);
+
+ /*
+ * tcp_read_sock() interprets partial progress as an indication to stop
+ * processing.
+ */
+ while (left) {
+ if (tinc == NULL) {
+ tinc = kmem_cache_alloc(rds_tcp_incoming_slab,
+ arg->gfp);
+ if (tinc == NULL) {
+ desc->error = -ENOMEM;
+ goto out;
+ }
+ tc->t_tinc = tinc;
+ rdsdebug("alloced tinc %p\n", tinc);
+ rds_inc_init(&tinc->ti_inc, conn, conn->c_faddr);
+ /*
+ * XXX * we might be able to use the __ variants when
+ * we've already serialized at a higher level.
+ */
+ skb_queue_head_init(&tinc->ti_skb_list);
+ }
+
+ if (left && tc->t_tinc_hdr_rem) {
+ to_copy = min(tc->t_tinc_hdr_rem, left);
+ rdsdebug("copying %zu header from skb %p\n", to_copy,
+ skb);
+ skb_copy_bits(skb, offset,
+ (char *)&tinc->ti_inc.i_hdr +
+ sizeof(struct rds_header) -
+ tc->t_tinc_hdr_rem,
+ to_copy);
+ tc->t_tinc_hdr_rem -= to_copy;
+ left -= to_copy;
+ offset += to_copy;
+
+ if (tc->t_tinc_hdr_rem == 0) {
+ /* could be 0 for a 0 len message */
+ tc->t_tinc_data_rem =
+ be32_to_cpu(tinc->ti_inc.i_hdr.h_len);
+ }
+ }
+
+ if (left && tc->t_tinc_data_rem) {
+ clone = skb_clone(skb, arg->gfp);
+ if (clone == NULL) {
+ desc->error = -ENOMEM;
+ goto out;
+ }
+
+ to_copy = min(tc->t_tinc_data_rem, left);
+ pskb_pull(clone, offset);
+ pskb_trim(clone, to_copy);
+ skb_queue_tail(&tinc->ti_skb_list, clone);
+
+ rdsdebug("skb %p data %p len %d off %u to_copy %zu -> "
+ "clone %p data %p len %d\n",
+ skb, skb->data, skb->len, offset, to_copy,
+ clone, clone->data, clone->len);
+
+ tc->t_tinc_data_rem -= to_copy;
+ left -= to_copy;
+ offset += to_copy;
+ }
+
+ if (tc->t_tinc_hdr_rem == 0 && tc->t_tinc_data_rem == 0) {
+ if (tinc->ti_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP)
+ rds_tcp_cong_recv(conn, tinc);
+ else
+ rds_recv_incoming(conn, conn->c_faddr,
+ conn->c_laddr, &tinc->ti_inc,
+ arg->gfp, arg->km);
+
+ tc->t_tinc_hdr_rem = sizeof(struct rds_header);
+ tc->t_tinc_data_rem = 0;
+ tc->t_tinc = NULL;
+ rds_inc_put(&tinc->ti_inc);
+ tinc = NULL;
+ }
+ }
+out:
+ rdsdebug("returning len %zu left %zu skb len %d rx queue depth %d\n",
+ len, left, skb->len,
+ skb_queue_len(&tc->t_sock->sk->sk_receive_queue));
+ return len - left;
+}
+
+/* the caller has to hold the sock lock */
+int rds_tcp_read_sock(struct rds_connection *conn, gfp_t gfp, enum km_type km)
+{
+ struct rds_tcp_connection *tc = conn->c_transport_data;
+ struct socket *sock = tc->t_sock;
+ read_descriptor_t desc;
+ struct rds_tcp_desc_arg arg;
+
+ /* It's like glib in the kernel! */
+ arg.conn = conn;
+ arg.gfp = gfp;
+ arg.km = km;
+ desc.arg.data = &arg;
+ desc.error = 0;
+ desc.count = 1; /* give more than one skb per call */
+
+ tcp_read_sock(sock->sk, &desc, rds_tcp_data_recv);
+ rdsdebug("tcp_read_sock for tc %p gfp 0x%x returned %d\n", tc, gfp,
+ desc.error);
+
+ return desc.error;
+}
+
+/*
+ * We hold the sock lock to serialize our rds_tcp_recv->tcp_read_sock from
+ * data_ready.
+ *
+ * if we fail to allocate we're in trouble.. blindly wait some time before
+ * trying again to see if the VM can free up something for us.
+ */
+int rds_tcp_recv(struct rds_connection *conn)
+{
+ struct rds_tcp_connection *tc = conn->c_transport_data;
+ struct socket *sock = tc->t_sock;
+ int ret = 0;
+
+ rdsdebug("recv worker conn %p tc %p sock %p\n", conn, tc, sock);
+
+ lock_sock(sock->sk);
+ ret = rds_tcp_read_sock(conn, GFP_KERNEL, KM_USER0);
+ release_sock(sock->sk);
+
+ return ret;
+}
+
+void rds_tcp_data_ready(struct sock *sk, int bytes)
+{
+ void (*ready)(struct sock *sk, int bytes);
+ struct rds_connection *conn;
+ struct rds_tcp_connection *tc;
+
+ rdsdebug("data ready sk %p bytes %d\n", sk, bytes);
+
+ read_lock(&sk->sk_callback_lock);
+ conn = sk->sk_user_data;
+ if (conn == NULL) { /* check for teardown race */
+ ready = sk->sk_data_ready;
+ goto out;
+ }
+
+ tc = conn->c_transport_data;
+ ready = tc->t_orig_data_ready;
+ rds_tcp_stats_inc(s_tcp_data_ready_calls);
+
+ if (rds_tcp_read_sock(conn, GFP_ATOMIC, KM_SOFTIRQ0) == -ENOMEM)
+ queue_delayed_work(rds_wq, &conn->c_recv_w, 0);
+out:
+ read_unlock(&sk->sk_callback_lock);
+ ready(sk, bytes);
+}
+
+int __init rds_tcp_recv_init(void)
+{
+ rds_tcp_incoming_slab = kmem_cache_create("rds_tcp_incoming",
+ sizeof(struct rds_tcp_incoming),
+ 0, 0, NULL);
+ if (rds_tcp_incoming_slab == NULL)
+ return -ENOMEM;
+ return 0;
+}
+
+void rds_tcp_recv_exit(void)
+{
+ kmem_cache_destroy(rds_tcp_incoming_slab);
+}
diff --git a/net/rds/tcp_send.c b/net/rds/tcp_send.c
new file mode 100644
index 0000000..ab545e0
--- /dev/null
+++ b/net/rds/tcp_send.c
@@ -0,0 +1,263 @@
+/*
+ * Copyright (c) 2006 Oracle. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/kernel.h>
+#include <linux/in.h>
+#include <net/tcp.h>
+
+#include "rds.h"
+#include "tcp.h"
+
+static void rds_tcp_cork(struct socket *sock, int val)
+{
+ mm_segment_t oldfs;
+
+ oldfs = get_fs();
+ set_fs(KERNEL_DS);
+ sock->ops->setsockopt(sock, SOL_TCP, TCP_CORK, (char __user *)&val,
+ sizeof(val));
+ set_fs(oldfs);
+}
+
+void rds_tcp_xmit_prepare(struct rds_connection *conn)
+{
+ struct rds_tcp_connection *tc = conn->c_transport_data;
+
+ rds_tcp_cork(tc->t_sock, 1);
+}
+
+void rds_tcp_xmit_complete(struct rds_connection *conn)
+{
+ struct rds_tcp_connection *tc = conn->c_transport_data;
+
+ rds_tcp_cork(tc->t_sock, 0);
+}
+
+/* the core send_sem serializes this with other xmit and shutdown */
+int rds_tcp_sendmsg(struct socket *sock, void *data, unsigned int len)
+{
+ struct kvec vec = {
+ .iov_base = data,
+ .iov_len = len,
+ };
+ struct msghdr msg = {
+ .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL,
+ };
+
+ return kernel_sendmsg(sock, &msg, &vec, 1, vec.iov_len);
+}
+
+/* the core send_sem serializes this with other xmit and shutdown */
+int rds_tcp_xmit_cong_map(struct rds_connection *conn,
+ struct rds_cong_map *map, unsigned long offset)
+{
+ static struct rds_header rds_tcp_map_header = {
+ .h_flags = RDS_FLAG_CONG_BITMAP,
+ };
+ struct rds_tcp_connection *tc = conn->c_transport_data;
+ unsigned long i;
+ int ret;
+ int copied = 0;
+
+ /* Some problem claims cpu_to_be32(constant) isn't a constant. */
+ rds_tcp_map_header.h_len = cpu_to_be32(RDS_CONG_MAP_BYTES);
+
+ if (offset < sizeof(struct rds_header)) {
+ ret = rds_tcp_sendmsg(tc->t_sock,
+ (void *)&rds_tcp_map_header + offset,
+ sizeof(struct rds_header) - offset);
+ if (ret <= 0)
+ return ret;
+ offset += ret;
+ copied = ret;
+ if (offset < sizeof(struct rds_header))
+ return ret;
+ }
+
+ offset -= sizeof(struct rds_header);
+ i = offset / PAGE_SIZE;
+ offset = offset % PAGE_SIZE;
+ BUG_ON(i >= RDS_CONG_MAP_PAGES);
+
+ do {
+ ret = tc->t_sock->ops->sendpage(tc->t_sock,
+ virt_to_page(map->m_page_addrs[i]),
+ offset, PAGE_SIZE - offset,
+ MSG_DONTWAIT);
+ if (ret <= 0)
+ break;
+ copied += ret;
+ offset += ret;
+ if (offset == PAGE_SIZE) {
+ offset = 0;
+ i++;
+ }
+ } while (i < RDS_CONG_MAP_PAGES);
+
+ return copied ? copied : ret;
+}
+
+/* the core send_sem serializes this with other xmit and shutdown */
+int rds_tcp_xmit(struct rds_connection *conn, struct rds_message *rm,
+ unsigned int hdr_off, unsigned int sg, unsigned int off)
+{
+ struct rds_tcp_connection *tc = conn->c_transport_data;
+ int done = 0;
+ int ret = 0;
+
+ if (hdr_off == 0) {
+ /*
+ * m_ack_seq is set to the sequence number of the last byte of
+ * header and data. see rds_tcp_is_acked().
+ */
+ tc->t_last_sent_nxt = rds_tcp_snd_nxt(tc);
+ rm->m_ack_seq = tc->t_last_sent_nxt +
+ sizeof(struct rds_header) +
+ be32_to_cpu(rm->m_inc.i_hdr.h_len) - 1;
+ smp_mb__before_clear_bit();
+ set_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags);
+ tc->t_last_expected_una = rm->m_ack_seq + 1;
+
+ rdsdebug("rm %p tcp nxt %u ack_seq %llu\n",
+ rm, rds_tcp_snd_nxt(tc),
+ (unsigned long long)rm->m_ack_seq);
+ }
+
+ if (hdr_off < sizeof(struct rds_header)) {
+ /* see rds_tcp_write_space() */
+ set_bit(SOCK_NOSPACE, &tc->t_sock->sk->sk_socket->flags);
+
+ ret = rds_tcp_sendmsg(tc->t_sock,
+ (void *)&rm->m_inc.i_hdr + hdr_off,
+ sizeof(rm->m_inc.i_hdr) - hdr_off);
+ if (ret < 0)
+ goto out;
+ done += ret;
+ if (hdr_off + done != sizeof(struct rds_header))
+ goto out;
+ }
+
+ while (sg < rm->m_nents) {
+ ret = tc->t_sock->ops->sendpage(tc->t_sock,
+ sg_page(&rm->m_sg[sg]),
+ rm->m_sg[sg].offset + off,
+ rm->m_sg[sg].length - off,
+ MSG_DONTWAIT|MSG_NOSIGNAL);
+ rdsdebug("tcp sendpage %p:%u:%u ret %d\n", (void *)sg_page(&rm->m_sg[sg]),
+ rm->m_sg[sg].offset + off, rm->m_sg[sg].length - off,
+ ret);
+ if (ret <= 0)
+ break;
+
+ off += ret;
+ done += ret;
+ if (off == rm->m_sg[sg].length) {
+ off = 0;
+ sg++;
+ }
+ }
+
+out:
+ if (ret <= 0) {
+ /* write_space will hit after EAGAIN, all else fatal */
+ if (ret == -EAGAIN) {
+ rds_tcp_stats_inc(s_tcp_sndbuf_full);
+ ret = 0;
+ } else {
+ printk(KERN_WARNING "RDS/tcp: send to %u.%u.%u.%u "
+ "returned %d, disconnecting and reconnecting\n",
+ NIPQUAD(conn->c_faddr), ret);
+ rds_conn_drop(conn);
+ }
+ }
+ if (done == 0)
+ done = ret;
+ return done;
+}
+
+/*
+ * rm->m_ack_seq is set to the tcp sequence number that corresponds to the
+ * last byte of the message, including the header. This means that the
+ * entire message has been received if rm->m_ack_seq is "before" the next
+ * unacked byte of the TCP sequence space. We have to do very careful
+ * wrapping 32bit comparisons here.
+ */
+static int rds_tcp_is_acked(struct rds_message *rm, uint64_t ack)
+{
+ if (!test_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags))
+ return 0;
+ return (__s32)((u32)rm->m_ack_seq - (u32)ack) < 0;
+}
+
+void rds_tcp_write_space(struct sock *sk)
+{
+ void (*write_space)(struct sock *sk);
+ struct rds_connection *conn;
+ struct rds_tcp_connection *tc;
+
+ read_lock(&sk->sk_callback_lock);
+ conn = sk->sk_user_data;
+ if (conn == NULL) {
+ write_space = sk->sk_write_space;
+ goto out;
+ }
+
+ tc = conn->c_transport_data;
+ rdsdebug("write_space for tc %p\n", tc);
+ write_space = tc->t_orig_write_space;
+ rds_tcp_stats_inc(s_tcp_write_space_calls);
+
+ rdsdebug("tcp una %u\n", rds_tcp_snd_una(tc));
+ tc->t_last_seen_una = rds_tcp_snd_una(tc);
+ rds_send_drop_acked(conn, rds_tcp_snd_una(tc), rds_tcp_is_acked);
+
+ queue_delayed_work(rds_wq, &conn->c_send_w, 0);
+out:
+ read_unlock(&sk->sk_callback_lock);
+
+ /*
+ * write_space is only called when data leaves tcp's send queue if
+ * SOCK_NOSPACE is set. We set SOCK_NOSPACE every time we put
+ * data in tcp's send queue because we use write_space to parse the
+ * sequence numbers and notice that rds messages have been fully
+ * received.
+ *
+ * tcp's write_space clears SOCK_NOSPACE if the send queue has more
+ * than a certain amount of space. So we need to set it again *after*
+ * we call tcp's write_space or else we might only get called on the
+ * first of a series of incoming tcp acks.
+ */
+ write_space(sk);
+
+ if (sk->sk_socket)
+ set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
+}
diff --git a/net/rds/tcp_stats.c b/net/rds/tcp_stats.c
new file mode 100644
index 0000000..d5898d0
--- /dev/null
+++ b/net/rds/tcp_stats.c
@@ -0,0 +1,74 @@
+/*
+ * Copyright (c) 2006 Oracle. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/percpu.h>
+#include <linux/seq_file.h>
+#include <linux/proc_fs.h>
+
+#include "rds.h"
+#include "tcp.h"
+
+DEFINE_PER_CPU(struct rds_tcp_statistics, rds_tcp_stats)
+ ____cacheline_aligned;
+
+static const char const *rds_tcp_stat_names[] = {
+ "tcp_data_ready_calls",
+ "tcp_write_space_calls",
+ "tcp_sndbuf_full",
+ "tcp_connect_raced",
+ "tcp_listen_closed_stale",
+};
+
+unsigned int rds_tcp_stats_info_copy(struct rds_info_iterator *iter,
+ unsigned int avail)
+{
+ struct rds_tcp_statistics stats = {0, };
+ uint64_t *src;
+ uint64_t *sum;
+ size_t i;
+ int cpu;
+
+ if (avail < ARRAY_SIZE(rds_tcp_stat_names))
+ goto out;
+
+ for_each_online_cpu(cpu) {
+ src = (uint64_t *)&(per_cpu(rds_tcp_stats, cpu));
+ sum = (uint64_t *)&stats;
+ for (i = 0; i < sizeof(stats) / sizeof(uint64_t); i++)
+ *(sum++) += *(src++);
+ }
+
+ rds_stats_info_copy(iter, (uint64_t *)&stats, rds_tcp_stat_names,
+ ARRAY_SIZE(rds_tcp_stat_names));
+out:
+ return ARRAY_SIZE(rds_tcp_stat_names);
+}
--
1.6.0.4
^ permalink raw reply related
* [PATCH 3/5] RDS: Modularize RDMA and TCP transports
From: Andy Grover @ 2009-08-21 22:28 UTC (permalink / raw)
To: netdev; +Cc: rds-devel
In-Reply-To: <1250893715-21939-1-git-send-email-andy.grover@oracle.com>
Enable the building of transports as modules.
Also, improve consistency of Kconfig messages in relation to other
protocols, and move build dependency on IB from the RDS core code
to the rds_rdma module.
Signed-off-by: Andy Grover <andy.grover@oracle.com>
---
net/rds/Kconfig | 26 ++++++++++++++++++++------
net/rds/Makefile | 11 +++++++++--
net/rds/af_rds.c | 8 --------
net/rds/rdma_transport.c | 6 ++++++
4 files changed, 35 insertions(+), 16 deletions(-)
diff --git a/net/rds/Kconfig b/net/rds/Kconfig
index 796773b..ec753b3 100644
--- a/net/rds/Kconfig
+++ b/net/rds/Kconfig
@@ -1,14 +1,28 @@
config RDS
- tristate "Reliable Datagram Sockets (RDS) (EXPERIMENTAL)"
- depends on INET && INFINIBAND_IPOIB && EXPERIMENTAL
- depends on INFINIBAND && INFINIBAND_ADDR_TRANS
+ tristate "The RDS Protocol (EXPERIMENTAL)"
+ depends on INET && EXPERIMENTAL
---help---
- RDS provides reliable, sequenced delivery of datagrams
- over Infiniband.
+ The RDS (Reliable Datagram Sockets) protocol provides reliable,
+ sequenced delivery of datagrams over Infiniband, iWARP,
+ or TCP.
+
+config RDS_RDMA
+ tristate "RDS over Infiniband and iWARP"
+ depends on RDS && INFINIBAND && INFINIBAND_ADDR_TRANS
+ ---help---
+ Allow RDS to use Infiniband and iWARP as a transport.
+ This transport supports RDMA operations.
+
+config RDS_TCP
+ tristate "RDS over TCP"
+ depends on RDS
+ ---help---
+ Allow RDS to use TCP as a transport.
+ This transport does not support RDMA operations.
config RDS_DEBUG
- bool "Debugging messages"
+ bool "RDS debugging messages"
depends on RDS
default n
diff --git a/net/rds/Makefile b/net/rds/Makefile
index 51f2758..b46eca1 100644
--- a/net/rds/Makefile
+++ b/net/rds/Makefile
@@ -1,13 +1,20 @@
obj-$(CONFIG_RDS) += rds.o
rds-y := af_rds.o bind.o cong.o connection.o info.o message.o \
recv.o send.o stats.o sysctl.o threads.o transport.o \
- loop.o page.o rdma.o \
- rdma_transport.o \
+ loop.o page.o rdma.o
+
+obj-$(CONFIG_RDS_RDMA) += rds_rdma.o
+rds_rdma-objs := rdma_transport.o \
ib.o ib_cm.o ib_recv.o ib_ring.o ib_send.o ib_stats.o \
ib_sysctl.o ib_rdma.o \
iw.o iw_cm.o iw_recv.o iw_ring.o iw_send.o iw_stats.o \
iw_sysctl.o iw_rdma.o
+
+obj-$(CONFIG_RDS_TCP) += rds_tcp.o
+rds_tcp-objs := tcp.o tcp_connect.o tcp_listen.o tcp_recv.o \
+ tcp_send.o tcp_stats.o
+
ifeq ($(CONFIG_RDS_DEBUG), y)
EXTRA_CFLAGS += -DDEBUG
endif
diff --git a/net/rds/af_rds.c b/net/rds/af_rds.c
index b11e7e5..108ed2e 100644
--- a/net/rds/af_rds.c
+++ b/net/rds/af_rds.c
@@ -39,7 +39,6 @@
#include "rds.h"
#include "rdma.h"
-#include "rdma_transport.h"
/* this is just used for stats gathering :/ */
static DEFINE_SPINLOCK(rds_sock_lock);
@@ -509,7 +508,6 @@ out:
static void __exit rds_exit(void)
{
- rds_rdma_exit();
sock_unregister(rds_family_ops.family);
proto_unregister(&rds_proto);
rds_conn_exit();
@@ -549,14 +547,8 @@ static int __init rds_init(void)
rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
- /* ib/iwarp transports currently compiled-in */
- ret = rds_rdma_init();
- if (ret)
- goto out_sock;
goto out;
-out_sock:
- sock_unregister(rds_family_ops.family);
out_proto:
proto_unregister(&rds_proto);
out_stats:
diff --git a/net/rds/rdma_transport.c b/net/rds/rdma_transport.c
index 7a6c748..9ece910 100644
--- a/net/rds/rdma_transport.c
+++ b/net/rds/rdma_transport.c
@@ -203,6 +203,7 @@ err_iw_init:
out:
return ret;
}
+module_init(rds_rdma_init);
void rds_rdma_exit(void)
{
@@ -211,4 +212,9 @@ void rds_rdma_exit(void)
rds_ib_exit();
rds_iw_exit();
}
+module_exit(rds_rdma_exit);
+
+MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
+MODULE_DESCRIPTION("RDS: IB/iWARP transport");
+MODULE_LICENSE("Dual BSD/GPL");
--
1.6.0.4
^ permalink raw reply related
* [PATCH 4/5] RDS: Track transports via an array, not a list
From: Andy Grover @ 2009-08-21 22:28 UTC (permalink / raw)
To: netdev; +Cc: rds-devel
In-Reply-To: <1250893715-21939-1-git-send-email-andy.grover@oracle.com>
Now that transports can be loaded in arbitrary order,
it is important for rds_trans_get_preferred() to look
for them in a particular order, instead of walking the list
until it finds a transport that works for a given address.
Now, each transport registers for a specific transport slot,
and these are ordered so that preferred transports come first,
and then if they are not loaded, other transports are queried.
Signed-off-by: Andy Grover <andy.grover@oracle.com>
---
net/rds/ib.c | 1 +
net/rds/iw.c | 1 +
net/rds/rds.h | 6 ++++++
net/rds/tcp.c | 1 +
net/rds/transport.c | 29 +++++++++++++++++++----------
5 files changed, 28 insertions(+), 10 deletions(-)
diff --git a/net/rds/ib.c b/net/rds/ib.c
index 868559a..536ebe5 100644
--- a/net/rds/ib.c
+++ b/net/rds/ib.c
@@ -282,6 +282,7 @@ struct rds_transport rds_ib_transport = {
.flush_mrs = rds_ib_flush_mrs,
.t_owner = THIS_MODULE,
.t_name = "infiniband",
+ .t_type = RDS_TRANS_IB
};
int __init rds_ib_init(void)
diff --git a/net/rds/iw.c b/net/rds/iw.c
index f5e9a29..db224f7 100644
--- a/net/rds/iw.c
+++ b/net/rds/iw.c
@@ -284,6 +284,7 @@ struct rds_transport rds_iw_transport = {
.flush_mrs = rds_iw_flush_mrs,
.t_owner = THIS_MODULE,
.t_name = "iwarp",
+ .t_type = RDS_TRANS_IWARP,
.t_prefer_loopback = 1,
};
diff --git a/net/rds/rds.h b/net/rds/rds.h
index 290566c..85d6f89 100644
--- a/net/rds/rds.h
+++ b/net/rds/rds.h
@@ -311,11 +311,17 @@ struct rds_notifier {
* flag and header.
*/
+#define RDS_TRANS_IB 0
+#define RDS_TRANS_IWARP 1
+#define RDS_TRANS_TCP 2
+#define RDS_TRANS_COUNT 3
+
struct rds_transport {
char t_name[TRANSNAMSIZ];
struct list_head t_item;
struct module *t_owner;
unsigned int t_prefer_loopback:1;
+ unsigned int t_type;
int (*laddr_check)(__be32 addr);
int (*conn_alloc)(struct rds_connection *conn, gfp_t gfp);
diff --git a/net/rds/tcp.c b/net/rds/tcp.c
index e0ac900..b5198ae 100644
--- a/net/rds/tcp.c
+++ b/net/rds/tcp.c
@@ -271,6 +271,7 @@ struct rds_transport rds_tcp_transport = {
.exit = rds_tcp_exit,
.t_owner = THIS_MODULE,
.t_name = "tcp",
+ .t_type = RDS_TRANS_TCP,
.t_prefer_loopback = 1,
};
diff --git a/net/rds/transport.c b/net/rds/transport.c
index 56a5309..7e10679 100644
--- a/net/rds/transport.c
+++ b/net/rds/transport.c
@@ -37,7 +37,7 @@
#include "rds.h"
#include "loop.h"
-static LIST_HEAD(rds_transports);
+static struct rds_transport *transports[RDS_TRANS_COUNT];
static DECLARE_RWSEM(rds_trans_sem);
int rds_trans_register(struct rds_transport *trans)
@@ -46,8 +46,13 @@ int rds_trans_register(struct rds_transport *trans)
down_write(&rds_trans_sem);
- list_add_tail(&trans->t_item, &rds_transports);
- printk(KERN_INFO "Registered RDS/%s transport\n", trans->t_name);
+ if (transports[trans->t_type])
+ printk(KERN_ERR "RDS Transport type %d already registered\n",
+ trans->t_type);
+ else {
+ transports[trans->t_type] = trans;
+ printk(KERN_INFO "Registered RDS/%s transport\n", trans->t_name);
+ }
up_write(&rds_trans_sem);
@@ -59,7 +64,7 @@ void rds_trans_unregister(struct rds_transport *trans)
{
down_write(&rds_trans_sem);
- list_del_init(&trans->t_item);
+ transports[trans->t_type] = NULL;
printk(KERN_INFO "Unregistered RDS/%s transport\n", trans->t_name);
up_write(&rds_trans_sem);
@@ -68,16 +73,17 @@ EXPORT_SYMBOL_GPL(rds_trans_unregister);
struct rds_transport *rds_trans_get_preferred(__be32 addr)
{
- struct rds_transport *trans;
struct rds_transport *ret = NULL;
+ int i;
if (IN_LOOPBACK(ntohl(addr)))
return &rds_loop_transport;
down_read(&rds_trans_sem);
- list_for_each_entry(trans, &rds_transports, t_item) {
- if (trans->laddr_check(addr) == 0) {
- ret = trans;
+ for (i = 0; i < RDS_TRANS_COUNT; i++)
+ {
+ if (transports[i] && (transports[i]->laddr_check(addr) == 0)) {
+ ret = transports[i];
break;
}
}
@@ -99,12 +105,15 @@ unsigned int rds_trans_stats_info_copy(struct rds_info_iterator *iter,
struct rds_transport *trans;
unsigned int total = 0;
unsigned int part;
+ int i;
rds_info_iter_unmap(iter);
down_read(&rds_trans_sem);
- list_for_each_entry(trans, &rds_transports, t_item) {
- if (trans->stats_info_copy == NULL)
+ for (i = 0; i < RDS_TRANS_COUNT; i++)
+ {
+ trans = transports[i];
+ if (!trans || !trans->stats_info_copy)
continue;
part = trans->stats_info_copy(iter, avail);
--
1.6.0.4
^ permalink raw reply related
* [PATCH 5/5] RDS: Add a debug message suggesting to load transport modules
From: Andy Grover @ 2009-08-21 22:28 UTC (permalink / raw)
To: netdev; +Cc: rds-devel
In-Reply-To: <1250893715-21939-1-git-send-email-andy.grover@oracle.com>
Now that RDS transports are no longer compiled-in to RDS core,
there is now the possibility that they will not be loaded. This
adds a helpful suggestion when rds_bind() fails to find a transport.
Signed-off-by: Andy Grover <andy.grover@oracle.com>
---
net/rds/bind.c | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/net/rds/bind.c b/net/rds/bind.c
index c17cc39..5d95fc0 100644
--- a/net/rds/bind.c
+++ b/net/rds/bind.c
@@ -187,6 +187,9 @@ int rds_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
if (trans == NULL) {
ret = -EADDRNOTAVAIL;
rds_remove_bound(rs);
+ if (printk_ratelimit())
+ printk(KERN_INFO "RDS: rds_bind() could not find a transport, "
+ "load rds_tcp or rds_rdma?\n");
goto out;
}
--
1.6.0.4
^ permalink raw reply related
* Re: [PATCH] pxaficp-ir - remove incorrect net_device_ops
From: Marek Vasut @ 2009-08-21 22:54 UTC (permalink / raw)
To: Alexander Beregalov
Cc: Eric Miao, linux-arm-kernel, Russell King - ARM Linux, samuel,
netdev, linux-arm-kernel
In-Reply-To: <200908051602.34501.marek.vasut@gmail.com>
Dne St 5. srpna 2009 16:02:34 Marek Vasut napsal(a):
> Dne St 29. července 2009 13:16:25 Alexander Beregalov napsal(a):
> > 2009/7/28 Eric Miao <eric.y.miao@gmail.com>:
> > > Marek Vasut wrote:
> > >> Hi!
> > >>
> > >> This patch fixes broken pxaficp-ir. The problem was in incorrect
> > >> net_device_ops being specified which prevented the driver from
> > >> operating. The symptoms were:
> > >> - failing ifconfig for IrLAN, resulting in
> > >> SIOCSIFFLAGS: Cannot assign requested address
> > >> - irattach working for IrCOMM, but the port stayed disabled
> > >>
> > >> Moreover this patch corrects missing sysfs device link.
> > >>
> > >> btw. guys, be honest, when did you last tested pxaficp-ir on real
> > >> hardware? ;-)
> > >
> > > Well, this seems to be brought by the net_device_ops change, which
> > > seems to happen silently without any of us being notified.
> > >
> > > OK, netdev and Alex are copied, so that we can look into this issue a
> > > bit deeper:
> > >
> > > 1. it looks to me that SIOCSIFFLAGS actually returned -EADDRNOTAVAIL,
> > > which is likely caused by eth_validate_addr, the default eth_addr comes
> > > with irda should be "00:00:00:00:00:00" if not explicitly specified
> > > (kzalloc), and this should be the problem, solution ? Either give a
> > > valid address to the irda net_device or remove this
> > > 'ndo_validate_addr'. And which is a correct fix will impact on the
> > > .ndo_set_mac_address
> > >
> > > 2. '.ndo_change_mtu' ? It looks to me that Irda device doesn't care too
> > > much about the MTU, eth_change_mtu is supposed to work just fine and
> > > not to cause any side effects, and may just benefit later irda device
> > > drivers if there is a weird device happens to care about MTU
> > >
> > > - eric
> > >
> > > Marek's original patch in attachment.
> >
> > Sorry about that and thanks.
> > Then we should fix the rest of irda drivers in the same way.
>
> Hi!
> Was my patch applied or what's the current status? Thanks
Hi!
any updates ? Are we letting this into .32 in current broken state?
Thanks!
^ permalink raw reply
* Re: sd8686 linux system hang: not always
From: Benoît Vaillant @ 2009-08-21 22:55 UTC (permalink / raw)
To: Dan Williams; +Cc: Benoît Vaillant, netdev
In-Reply-To: <1250859856.2876.11.camel@localhost.localdomain>
[-- Attachment #1: Type: text/plain, Size: 1705 bytes --]
Hi,
On Fri, Aug 21, 2009 at 08:04:16AM -0500, Dan Williams wrote:
> Well, besides the fact that the Poulsbo SDHC driver hasn't actually
> passed review for the kernel yet...
And other parts of Poulsbo drivers were removed from -at least-
debian, for the xorg part of it. I had that in mind.
> That doesn't mean the Poulsbo driver isn't a fine driver, but if
> Pierre hasn't acked it, I don't really trust it. The largest # of
> problems people have had with Libertas SDIO have been with the host
> controller with things like transfer widths, clocking, interrupt
> vs. polled mode, chunk size, etc. When the SDHC driver and hardware
> they are using actually starts working well, then libertas actually
> starts working well.
Ok. I'll check on parameters when I get spare time. I did consider
such things could be part (all? :) ) of the issue.
Pointers welcomed, if any other than browsing the web!
> So I don't mean to brush you off, but we know the chip works pretty well
> with the Ricoh line (which I'm typing this on right now with an sd8686)
> and a few other embedded-type controllers. So I'd start diving down
> into the SDHC driver code for Poulsbo and try to figure out what's going
> on there and in the MMC layer.
Hehe, no worry, I know the chip works fine under midinux. Yet
compiling such working drivers would either lead me to revert to an
older kernel version (and not provide testing with newer ones, thus)
or adapt the marvell driver to newer kernel, task which I'm rather not
sure to be able to carry by some quite good amount of time :)
You didn't brush me off, I wasn't expecting a straight working
solution :)
TFYA,
--
Benoît
[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 197 bytes --]
^ permalink raw reply
* Re: [PATCH 2.6.30-rc4] r8169: avoid losing MSI interrupts
From: David Dillow @ 2009-08-21 22:59 UTC (permalink / raw)
To: Eric W. Biederman
Cc: Michael Riepe, Michael Buesch, Francois Romieu, Rui Santos,
Michael Büker, linux-kernel, netdev
In-Reply-To: <m1skfkrik2.fsf@fess.ebiederm.org>
On Fri, 2009-08-21 at 13:57 -0700, Eric W. Biederman wrote:
> David Dillow <dave@thedillows.org> writes:
> I have what at first glance looks like a problem caused by this
> patch. For the last month since upgrading one of my machines from
> 2.6.28 to 2.6.30 it has been becomming inaccessible from the
> network and I have a few:
>
> NETDEV WATCHDOG: eth0 (r8169): transmit timed out
>
> in my logs and a lot soft lockups that always have rtl8169_interrupt
> as the thing that is running. I suspect your patch has introduced
> a near infinite loop in the interrupt handler and is causing these
> soft lockups.
>
> Any ideas?
I would be surprised, but I suppose it is not out of the realm of
possibility. Can you send me a full dmesg, please?
^ permalink raw reply
* Re: [PATCH 2.6.30-rc4] r8169: avoid losing MSI interrupts
From: David Dillow @ 2009-08-21 23:34 UTC (permalink / raw)
To: Eric W. Biederman
Cc: Michael Riepe, Michael Buesch, Francois Romieu, Rui Santos,
Michael Büker, linux-kernel, netdev
In-Reply-To: <1250895567.23419.1.camel@obelisk.thedillows.org>
On Fri, 2009-08-21 at 18:59 -0400, David Dillow wrote:
> On Fri, 2009-08-21 at 13:57 -0700, Eric W. Biederman wrote:
> > David Dillow <dave@thedillows.org> writes:
> > I have what at first glance looks like a problem caused by this
> > patch. For the last month since upgrading one of my machines from
> > 2.6.28 to 2.6.30 it has been becomming inaccessible from the
> > network and I have a few:
> >
> > NETDEV WATCHDOG: eth0 (r8169): transmit timed out
> >
> > in my logs and a lot soft lockups that always have rtl8169_interrupt
> > as the thing that is running. I suspect your patch has introduced
> > a near infinite loop in the interrupt handler and is causing these
> > soft lockups.
> >
> > Any ideas?
>
> I would be surprised, but I suppose it is not out of the realm of
> possibility. Can you send me a full dmesg, please?
Re-looking at the code, I'd guess that some IRQ status line is getting
stuck high, but I don't see why -- we should acknowledge all outstanding
interrupts each time through the loop, whether we care about them or
not.
Could reproduce a problem with the following patch applied, and send the
full dmesg, please?
diff --git a/drivers/net/r8169.c b/drivers/net/r8169.c
index b82780d..46cb05a 100644
--- a/drivers/net/r8169.c
+++ b/drivers/net/r8169.c
@@ -3556,6 +3556,7 @@ static irqreturn_t rtl8169_interrupt(int irq, void *dev_instance)
void __iomem *ioaddr = tp->mmio_addr;
int handled = 0;
int status;
+ int count = 0;
/* loop handling interrupts until we have no new ones or
* we hit a invalid/hotplug case.
@@ -3564,6 +3565,15 @@ static irqreturn_t rtl8169_interrupt(int irq, void *dev_instance)
while (status && status != 0xffff) {
handled = 1;
+ if (count++ > 100) {
+ printk_once("r8169 screaming irq status %08x "
+ "mask %08x event %08x napi %08x\n",
+ status, tp->intr_mask, tp->intr_event,
+ tp->napi_event);
+ break;
+ }
+
+
/* Handle all of the error cases first. These will reset
* the chip, so just exit the loop.
*/
^ permalink raw reply related
* Re: [Patch] Fix commit 63d9950b08184e6531adceb65f64b429909cc101 (ipv6: Make v4-mapped bindings consistent with IPv4)
From: Bruno Prémont @ 2009-08-21 23:45 UTC (permalink / raw)
To: Vlad Yasevich, David S. Miller; +Cc: netdev
In-Reply-To: <20090822011541.7b95dbcc@neptune.home>
Commit 63d9950b08184e6531adceb65f64b429909cc101
(ipv6: Make v4-mapped bindings consistent with IPv4)
changes behavior of inet6_bind() for v4-mapped addresses so it should
behave the same way as inet_bind().
During this change setting of err to -EADDRNOTAVAIL got lost:
af_inet.c:469 inet_bind()
err = -EADDRNOTAVAIL;
if (!sysctl_ip_nonlocal_bind &&
!(inet->freebind || inet->transparent) &&
addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
chk_addr_ret != RTN_LOCAL &&
chk_addr_ret != RTN_MULTICAST &&
chk_addr_ret != RTN_BROADCAST)
goto out;
af_inet6.c:463 inet6_bind()
if (addr_type == IPV6_ADDR_MAPPED) {
int chk_addr_ret;
/* Binding to v4-mapped address on a v6-only socket
* makes no sense
*/
if (np->ipv6only) {
err = -EINVAL;
goto out;
}
/* Reproduce AF_INET checks to make the bindings consitant */
v4addr = addr->sin6_addr.s6_addr32[3];
chk_addr_ret = inet_addr_type(net, v4addr);
if (!sysctl_ip_nonlocal_bind &&
!(inet->freebind || inet->transparent) &&
v4addr != htonl(INADDR_ANY) &&
chk_addr_ret != RTN_LOCAL &&
chk_addr_ret != RTN_MULTICAST &&
chk_addr_ret != RTN_BROADCAST)
goto out;
} else {
Signed-off-by Bruno Prémont <bonbons@linux-vserver.org>
---
diff --git a/net/ipv6/af_inet6.c b/net/ipv6/af_inet6.c
index caa0278..45f9a2a 100644
--- a/net/ipv6/af_inet6.c
+++ b/net/ipv6/af_inet6.c
@@ -306,8 +306,10 @@ int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
v4addr != htonl(INADDR_ANY) &&
chk_addr_ret != RTN_LOCAL &&
chk_addr_ret != RTN_MULTICAST &&
- chk_addr_ret != RTN_BROADCAST)
+ chk_addr_ret != RTN_BROADCAST) {
+ err = -EADDRNOTAVAIL;
goto out;
+ }
} else {
if (addr_type != IPV6_ADDR_ANY) {
struct net_device *dev = NULL;
^ permalink raw reply related
* [PATCH 0/2]: Convert packet scheduler hrtimers to tasklet_hrtimer
From: David Miller @ 2009-08-22 0:03 UTC (permalink / raw)
To: netdev; +Cc: tglx
tasklet_hrtimers were created as a response to how we noticed that
we had hrtimers but they expected to run in softirq context but
actually hrtimers run in hw IRQ context
This converts the two hrtimers in the packet scheduler over to
be tasklet_hrtimers.
The CAN protocol has a few hrtimers, but they fire off tasklets to
do softirq work. So they're OK.
So after this we should be clean in the networking.
^ permalink raw reply
* [PATCH 1/2]: pkt_sched: Convert qdisc_watchdog to tasklet_hrtimer
From: David Miller @ 2009-08-22 0:03 UTC (permalink / raw)
To: netdev; +Cc: tglx
None of this stuff should execute in hw IRQ context, therefore
use a tasklet_hrtimer so that it runs in softirq context.
Signed-off-by: David S. Miller <davem@davemloft.net>
---
include/net/pkt_sched.h | 4 ++--
net/sched/sch_api.c | 8 ++++----
2 files changed, 6 insertions(+), 6 deletions(-)
diff --git a/include/net/pkt_sched.h b/include/net/pkt_sched.h
index 82a3191..7eafb8d 100644
--- a/include/net/pkt_sched.h
+++ b/include/net/pkt_sched.h
@@ -61,8 +61,8 @@ psched_tdiff_bounded(psched_time_t tv1, psched_time_t tv2, psched_time_t bound)
}
struct qdisc_watchdog {
- struct hrtimer timer;
- struct Qdisc *qdisc;
+ struct tasklet_hrtimer timer;
+ struct Qdisc *qdisc;
};
extern void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc);
diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
index 24d17ce..e1c2bf7 100644
--- a/net/sched/sch_api.c
+++ b/net/sched/sch_api.c
@@ -468,8 +468,8 @@ static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
{
- hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
- wd->timer.function = qdisc_watchdog;
+ tasklet_hrtimer_init(&wd->timer, qdisc_watchdog,
+ CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
wd->qdisc = qdisc;
}
EXPORT_SYMBOL(qdisc_watchdog_init);
@@ -485,13 +485,13 @@ void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
wd->qdisc->flags |= TCQ_F_THROTTLED;
time = ktime_set(0, 0);
time = ktime_add_ns(time, PSCHED_TICKS2NS(expires));
- hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
+ tasklet_hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
}
EXPORT_SYMBOL(qdisc_watchdog_schedule);
void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
{
- hrtimer_cancel(&wd->timer);
+ tasklet_hrtimer_cancel(&wd->timer);
wd->qdisc->flags &= ~TCQ_F_THROTTLED;
}
EXPORT_SYMBOL(qdisc_watchdog_cancel);
--
1.6.3.4
^ permalink raw reply related
* [PATCH 2/2]: pkt_sched: Convert CBQ to tasklet_hrtimer.
From: David Miller @ 2009-08-22 0:03 UTC (permalink / raw)
To: netdev; +Cc: tglx
This code expects to run in softirq context, and bare hrtimers
run in hw IRQ context.
Signed-off-by: David S. Miller <davem@davemloft.net>
---
net/sched/sch_cbq.c | 23 +++++++++++++----------
1 files changed, 13 insertions(+), 10 deletions(-)
diff --git a/net/sched/sch_cbq.c b/net/sched/sch_cbq.c
index d5798e1..81652d6 100644
--- a/net/sched/sch_cbq.c
+++ b/net/sched/sch_cbq.c
@@ -163,7 +163,7 @@ struct cbq_sched_data
psched_time_t now_rt; /* Cached real time */
unsigned pmask;
- struct hrtimer delay_timer;
+ struct tasklet_hrtimer delay_timer;
struct qdisc_watchdog watchdog; /* Watchdog timer,
started when CBQ has
backlog, but cannot
@@ -503,6 +503,8 @@ static void cbq_ovl_delay(struct cbq_class *cl)
cl->undertime = q->now + delay;
if (delay > 0) {
+ struct hrtimer *ht;
+
sched += delay + cl->penalty;
cl->penalized = sched;
cl->cpriority = TC_CBQ_MAXPRIO;
@@ -510,12 +512,12 @@ static void cbq_ovl_delay(struct cbq_class *cl)
expires = ktime_set(0, 0);
expires = ktime_add_ns(expires, PSCHED_TICKS2NS(sched));
- if (hrtimer_try_to_cancel(&q->delay_timer) &&
- ktime_to_ns(ktime_sub(
- hrtimer_get_expires(&q->delay_timer),
- expires)) > 0)
- hrtimer_set_expires(&q->delay_timer, expires);
- hrtimer_restart(&q->delay_timer);
+ ht = &q->delay_timer.timer;
+ if (hrtimer_try_to_cancel(ht) &&
+ ktime_to_ns(ktime_sub(hrtimer_get_expires(ht),
+ expires)) > 0)
+ hrtimer_set_expires(ht, expires);
+ hrtimer_restart(ht);
cl->delayed = 1;
cl->xstats.overactions++;
return;
@@ -621,7 +623,7 @@ static enum hrtimer_restart cbq_undelay(struct hrtimer *timer)
time = ktime_set(0, 0);
time = ktime_add_ns(time, PSCHED_TICKS2NS(now + delay));
- hrtimer_start(&q->delay_timer, time, HRTIMER_MODE_ABS);
+ tasklet_hrtimer_start(&q->delay_timer, time, HRTIMER_MODE_ABS);
}
sch->flags &= ~TCQ_F_THROTTLED;
@@ -1214,7 +1216,7 @@ cbq_reset(struct Qdisc* sch)
q->tx_class = NULL;
q->tx_borrowed = NULL;
qdisc_watchdog_cancel(&q->watchdog);
- hrtimer_cancel(&q->delay_timer);
+ tasklet_hrtimer_cancel(&q->delay_timer);
q->toplevel = TC_CBQ_MAXLEVEL;
q->now = psched_get_time();
q->now_rt = q->now;
@@ -1397,7 +1399,8 @@ static int cbq_init(struct Qdisc *sch, struct nlattr *opt)
q->link.minidle = -0x7FFFFFFF;
qdisc_watchdog_init(&q->watchdog, sch);
- hrtimer_init(&q->delay_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
+ tasklet_hrtimer_init(&q->delay_timer, cbq_undelay,
+ CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
q->delay_timer.function = cbq_undelay;
q->toplevel = TC_CBQ_MAXLEVEL;
q->now = psched_get_time();
--
1.6.3.4
^ permalink raw reply related
* Re: [PATCH 2.6.30-rc4] r8169: avoid losing MSI interrupts
From: Eric W. Biederman @ 2009-08-22 0:24 UTC (permalink / raw)
To: David Dillow
Cc: Michael Riepe, Michael Buesch, Francois Romieu, Rui Santos,
Michael Büker, linux-kernel, netdev
In-Reply-To: <1250897657.23419.5.camel@obelisk.thedillows.org>
David Dillow <dave@thedillows.org> writes:
> On Fri, 2009-08-21 at 18:59 -0400, David Dillow wrote:
>> On Fri, 2009-08-21 at 13:57 -0700, Eric W. Biederman wrote:
>> > David Dillow <dave@thedillows.org> writes:
>> > I have what at first glance looks like a problem caused by this
>> > patch. For the last month since upgrading one of my machines from
>> > 2.6.28 to 2.6.30 it has been becomming inaccessible from the
>> > network and I have a few:
>> >
>> > NETDEV WATCHDOG: eth0 (r8169): transmit timed out
>> >
>> > in my logs and a lot soft lockups that always have rtl8169_interrupt
>> > as the thing that is running. I suspect your patch has introduced
>> > a near infinite loop in the interrupt handler and is causing these
>> > soft lockups.
>> >
>> > Any ideas?
>>
>> I would be surprised, but I suppose it is not out of the realm of
>> possibility. Can you send me a full dmesg, please?
>
> Re-looking at the code, I'd guess that some IRQ status line is getting
> stuck high, but I don't see why -- we should acknowledge all outstanding
> interrupts each time through the loop, whether we care about them or
> not.
>
> Could reproduce a problem with the following patch applied, and send the
> full dmesg, please?
Will do. This looks like a good way to test my hypothesis thanks.
I can't quite reproduce this problem so it may be a few days before
I know.
Eric
> diff --git a/drivers/net/r8169.c b/drivers/net/r8169.c
> index b82780d..46cb05a 100644
> --- a/drivers/net/r8169.c
> +++ b/drivers/net/r8169.c
> @@ -3556,6 +3556,7 @@ static irqreturn_t rtl8169_interrupt(int irq, void *dev_instance)
> void __iomem *ioaddr = tp->mmio_addr;
> int handled = 0;
> int status;
> + int count = 0;
>
> /* loop handling interrupts until we have no new ones or
> * we hit a invalid/hotplug case.
> @@ -3564,6 +3565,15 @@ static irqreturn_t rtl8169_interrupt(int irq, void *dev_instance)
> while (status && status != 0xffff) {
> handled = 1;
>
> + if (count++ > 100) {
> + printk_once("r8169 screaming irq status %08x "
> + "mask %08x event %08x napi %08x\n",
> + status, tp->intr_mask, tp->intr_event,
> + tp->napi_event);
> + break;
> + }
> +
> +
> /* Handle all of the error cases first. These will reset
> * the chip, so just exit the loop.
> */
^ permalink raw reply
* RFC 5386
From: steve baker @ 2009-08-22 1:09 UTC (permalink / raw)
To: netdev
Hello,
I was wondering if there are any plans in the works to implement RFC
5386 (better than nothing security), and who might have the lead on
it?
Thanks,
Steve
^ permalink raw reply
* [PATCH -next 3/8] bnx2: Close device if MTU change or ring size change fails.
From: Michael Chan @ 2009-08-22 2:20 UTC (permalink / raw)
To: davem; +Cc: netdev
In-Reply-To: <1250907650-24672-1-git-send-email-mchan@broadcom.com>
When unable to allocate memory for new MTU or new ring size, we need
to close the device to prevent it from crashing.
Signed-off-by: Michael Chan <mchan@broadcom.com>
---
drivers/net/bnx2.c | 9 +++++++--
1 files changed, 7 insertions(+), 2 deletions(-)
diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c
index 9cfd460..1d502e6 100644
--- a/drivers/net/bnx2.c
+++ b/drivers/net/bnx2.c
@@ -6992,9 +6992,14 @@ bnx2_change_ring_size(struct bnx2 *bp, u32 rx, u32 tx)
int rc;
rc = bnx2_alloc_mem(bp);
- if (rc)
+ if (!rc)
+ rc = bnx2_init_nic(bp, 0);
+
+ if (rc) {
+ bnx2_napi_enable(bp);
+ dev_close(bp->dev);
return rc;
- bnx2_init_nic(bp, 0);
+ }
bnx2_netif_start(bp);
}
return 0;
--
1.5.6.GIT
^ permalink raw reply related
* [PATCH -next 4/8] bnx2: Apply BROKEN_STATS workaround to 5706 and 5708.
From: Michael Chan @ 2009-08-22 2:20 UTC (permalink / raw)
To: davem; +Cc: netdev
In-Reply-To: <1250907650-24672-1-git-send-email-mchan@broadcom.com>
Add flag to expand the workaround to both chips.
Signed-off-by: Michael Chan <mchan@broadcom.com>
---
drivers/net/bnx2.c | 7 ++++---
drivers/net/bnx2.h | 1 +
2 files changed, 5 insertions(+), 3 deletions(-)
diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c
index 1d502e6..085c2dd 100644
--- a/drivers/net/bnx2.c
+++ b/drivers/net/bnx2.c
@@ -4900,7 +4900,7 @@ bnx2_init_chip(struct bnx2 *bp)
REG_WR(bp, BNX2_HC_CMD_TICKS,
(bp->cmd_ticks_int << 16) | bp->cmd_ticks);
- if (CHIP_NUM(bp) == CHIP_NUM_5708)
+ if (bp->flags & BNX2_FLAG_BROKEN_STATS)
REG_WR(bp, BNX2_HC_STATS_TICKS, 0);
else
REG_WR(bp, BNX2_HC_STATS_TICKS, bp->stats_ticks);
@@ -6025,7 +6025,7 @@ bnx2_timer(unsigned long data)
bnx2_reg_rd_ind(bp, BNX2_FW_RX_DROP_COUNT);
/* workaround occasional corrupted counters */
- if (CHIP_NUM(bp) == CHIP_NUM_5708 && bp->stats_ticks)
+ if ((bp->flags & BNX2_FLAG_BROKEN_STATS) && bp->stats_ticks)
REG_WR(bp, BNX2_HC_COMMAND, bp->hc_cmd |
BNX2_HC_COMMAND_STATS_NOW);
@@ -6941,7 +6941,7 @@ bnx2_set_coalesce(struct net_device *dev, struct ethtool_coalesce *coal)
0xff;
bp->stats_ticks = coal->stats_block_coalesce_usecs;
- if (CHIP_NUM(bp) == CHIP_NUM_5708) {
+ if (bp->flags & BNX2_FLAG_BROKEN_STATS) {
if (bp->stats_ticks != 0 && bp->stats_ticks != USEC_PER_SEC)
bp->stats_ticks = USEC_PER_SEC;
}
@@ -7722,6 +7722,7 @@ bnx2_init_board(struct pci_dev *pdev, struct net_device *dev)
rc = -EIO;
goto err_out_unmap;
}
+ bp->flags |= BNX2_FLAG_BROKEN_STATS;
}
if (CHIP_NUM(bp) == CHIP_NUM_5709 && CHIP_REV(bp) != CHIP_REV_Ax) {
diff --git a/drivers/net/bnx2.h b/drivers/net/bnx2.h
index f1edfaa..3c004b4 100644
--- a/drivers/net/bnx2.h
+++ b/drivers/net/bnx2.h
@@ -6718,6 +6718,7 @@ struct bnx2 {
BNX2_FLAG_USING_MSIX)
#define BNX2_FLAG_JUMBO_BROKEN 0x00000800
#define BNX2_FLAG_CAN_KEEP_VLAN 0x00001000
+#define BNX2_FLAG_BROKEN_STATS 0x00002000
struct bnx2_napi bnx2_napi[BNX2_MAX_MSIX_VEC];
--
1.5.6.GIT
^ permalink raw reply related
* [PATCH -next 5/8] bnx2: Report FTQ discard counter.
From: Michael Chan @ 2009-08-22 2:20 UTC (permalink / raw)
To: davem; +Cc: netdev
In-Reply-To: <1250907650-24672-1-git-send-email-mchan@broadcom.com>
Report this counter to ethtool -S and include it in netstat.
Signed-off-by: Michael Chan <mchan@broadcom.com>
---
drivers/net/bnx2.c | 20 ++++++++++++--------
1 files changed, 12 insertions(+), 8 deletions(-)
diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c
index 085c2dd..e025833 100644
--- a/drivers/net/bnx2.c
+++ b/drivers/net/bnx2.c
@@ -6485,7 +6485,8 @@ bnx2_get_stats(struct net_device *dev)
stats_blk->stat_EtherStatsOverrsizePkts);
net_stats->rx_over_errors =
- (unsigned long) stats_blk->stat_IfInMBUFDiscards;
+ (unsigned long) (stats_blk->stat_IfInFTQDiscards +
+ stats_blk->stat_IfInMBUFDiscards);
net_stats->rx_frame_errors =
(unsigned long) stats_blk->stat_Dot3StatsAlignmentErrors;
@@ -6518,8 +6519,8 @@ bnx2_get_stats(struct net_device *dev)
net_stats->tx_carrier_errors;
net_stats->rx_missed_errors =
- (unsigned long) (stats_blk->stat_IfInMBUFDiscards +
- stats_blk->stat_FwRxDrop);
+ (unsigned long) (stats_blk->stat_IfInFTQDiscards +
+ stats_blk->stat_IfInMBUFDiscards + stats_blk->stat_FwRxDrop);
return net_stats;
}
@@ -7090,11 +7091,9 @@ bnx2_set_tso(struct net_device *dev, u32 data)
return 0;
}
-#define BNX2_NUM_STATS 46
-
static struct {
char string[ETH_GSTRING_LEN];
-} bnx2_stats_str_arr[BNX2_NUM_STATS] = {
+} bnx2_stats_str_arr[] = {
{ "rx_bytes" },
{ "rx_error_bytes" },
{ "tx_bytes" },
@@ -7139,10 +7138,14 @@ static struct {
{ "tx_xoff_frames" },
{ "rx_mac_ctrl_frames" },
{ "rx_filtered_packets" },
+ { "rx_ftq_discards" },
{ "rx_discards" },
{ "rx_fw_discards" },
};
+#define BNX2_NUM_STATS (sizeof(bnx2_stats_str_arr)/\
+ sizeof(bnx2_stats_str_arr[0]))
+
#define STATS_OFFSET32(offset_name) (offsetof(struct statistics_block, offset_name) / 4)
static const unsigned long bnx2_stats_offset_arr[BNX2_NUM_STATS] = {
@@ -7190,6 +7193,7 @@ static const unsigned long bnx2_stats_offset_arr[BNX2_NUM_STATS] = {
STATS_OFFSET32(stat_OutXoffSent),
STATS_OFFSET32(stat_MacControlFramesReceived),
STATS_OFFSET32(stat_IfInFramesL2FilterDiscards),
+ STATS_OFFSET32(stat_IfInFTQDiscards),
STATS_OFFSET32(stat_IfInMBUFDiscards),
STATS_OFFSET32(stat_FwRxDrop),
};
@@ -7202,7 +7206,7 @@ static u8 bnx2_5706_stats_len_arr[BNX2_NUM_STATS] = {
4,0,4,4,4,4,4,4,4,4,
4,4,4,4,4,4,4,4,4,4,
4,4,4,4,4,4,4,4,4,4,
- 4,4,4,4,4,4,
+ 4,4,4,4,4,4,4,
};
static u8 bnx2_5708_stats_len_arr[BNX2_NUM_STATS] = {
@@ -7210,7 +7214,7 @@ static u8 bnx2_5708_stats_len_arr[BNX2_NUM_STATS] = {
4,4,4,4,4,4,4,4,4,4,
4,4,4,4,4,4,4,4,4,4,
4,4,4,4,4,4,4,4,4,4,
- 4,4,4,4,4,4,
+ 4,4,4,4,4,4,4,
};
#define BNX2_NUM_TESTS 6
--
1.5.6.GIT
^ permalink raw reply related
* [PATCH -next 1/8] bnx2: Zero out status block before chip reset.
From: Michael Chan @ 2009-08-22 2:20 UTC (permalink / raw)
To: davem; +Cc: netdev
In case IRQs are shared, we will not mistakenly start processing
the ring based on old status block indices.
Signed-off-by: Michael Chan <mchan@broadcom.com>
---
drivers/net/bnx2.c | 1 +
1 files changed, 1 insertions(+), 0 deletions(-)
diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c
index 756d4b4..e75b6e9 100644
--- a/drivers/net/bnx2.c
+++ b/drivers/net/bnx2.c
@@ -4858,6 +4858,7 @@ bnx2_init_chip(struct bnx2 *bp)
bnx2_reg_wr_ind(bp, BNX2_RBUF_CONFIG2, BNX2_RBUF_CONFIG2_VAL(mtu));
bnx2_reg_wr_ind(bp, BNX2_RBUF_CONFIG3, BNX2_RBUF_CONFIG3_VAL(mtu));
+ memset(bp->bnx2_napi[0].status_blk.msi, 0, bp->status_stats_size);
for (i = 0; i < BNX2_MAX_MSIX_VEC; i++)
bp->bnx2_napi[i].last_status_idx = 0;
--
1.5.6.GIT
^ permalink raw reply related
* [PATCH -next 2/8] bnx2: Check if_running() before touching chip registers.
From: Michael Chan @ 2009-08-22 2:20 UTC (permalink / raw)
To: davem; +Cc: netdev
In-Reply-To: <1250907650-24672-1-git-send-email-mchan@broadcom.com>
Add this check to bnx2_netif_stop() and bnx2_vlan_rx_register() to
prevent bus lockups on some systems when the chip is in low power state.
Signed-off-by: Michael Chan <mchan@broadcom.com>
---
drivers/net/bnx2.c | 10 +++++++++-
1 files changed, 9 insertions(+), 1 deletions(-)
diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c
index e75b6e9..9cfd460 100644
--- a/drivers/net/bnx2.c
+++ b/drivers/net/bnx2.c
@@ -619,6 +619,9 @@ bnx2_disable_int_sync(struct bnx2 *bp)
int i;
atomic_inc(&bp->intr_sem);
+ if (!netif_running(bp->dev))
+ return;
+
bnx2_disable_int(bp);
for (i = 0; i < bp->irq_nvecs; i++)
synchronize_irq(bp->irq_tbl[i].vector);
@@ -6254,9 +6257,14 @@ bnx2_vlan_rx_register(struct net_device *dev, struct vlan_group *vlgrp)
{
struct bnx2 *bp = netdev_priv(dev);
- bnx2_netif_stop(bp);
+ if (netif_running(dev))
+ bnx2_netif_stop(bp);
bp->vlgrp = vlgrp;
+
+ if (!netif_running(dev))
+ return;
+
bnx2_set_rx_mode(dev);
if (bp->flags & BNX2_FLAG_CAN_KEEP_VLAN)
bnx2_fw_sync(bp, BNX2_DRV_MSG_CODE_KEEP_VLAN_UPDATE, 0, 1);
--
1.5.6.GIT
^ permalink raw reply related
* [PATCH -next 6/8] bnx2: Refine coalescing parameters.
From: Michael Chan @ 2009-08-22 2:20 UTC (permalink / raw)
To: davem; +Cc: netdev
In-Reply-To: <1250907650-24672-1-git-send-email-mchan@broadcom.com>
- Set the USE_INT_PARAM bit so the rx-frames-irq and tx-frames-irq will take
effect on 5709.
- Increase the default rx-frames to reduce interrupt count.
- Decrease the default rx-frames-irq and tx-frames-irq to catch more events
during NAPI poll.
All these will reduce interrupts without affecting latency.
Signed-off-by: Michael Chan <mchan@broadcom.com>
---
drivers/net/bnx2.c | 10 +++++-----
1 files changed, 5 insertions(+), 5 deletions(-)
diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c
index e025833..ea6e6b7 100644
--- a/drivers/net/bnx2.c
+++ b/drivers/net/bnx2.c
@@ -4921,7 +4921,7 @@ bnx2_init_chip(struct bnx2 *bp)
}
if (bp->flags & BNX2_FLAG_ONE_SHOT_MSI)
- val |= BNX2_HC_CONFIG_ONE_SHOT;
+ val |= BNX2_HC_CONFIG_ONE_SHOT | BNX2_HC_CONFIG_USE_INT_PARAM;
REG_WR(bp, BNX2_HC_CONFIG, val);
@@ -7858,13 +7858,13 @@ bnx2_init_board(struct pci_dev *pdev, struct net_device *dev)
bp->rx_csum = 1;
- bp->tx_quick_cons_trip_int = 20;
+ bp->tx_quick_cons_trip_int = 2;
bp->tx_quick_cons_trip = 20;
- bp->tx_ticks_int = 80;
+ bp->tx_ticks_int = 18;
bp->tx_ticks = 80;
- bp->rx_quick_cons_trip_int = 6;
- bp->rx_quick_cons_trip = 6;
+ bp->rx_quick_cons_trip_int = 2;
+ bp->rx_quick_cons_trip = 12;
bp->rx_ticks_int = 18;
bp->rx_ticks = 18;
--
1.5.6.GIT
^ permalink raw reply related
* [PATCH -next 8/8] bnx2: Update version to 2.0.2.
From: Michael Chan @ 2009-08-22 2:20 UTC (permalink / raw)
To: davem; +Cc: netdev
In-Reply-To: <1250907650-24672-1-git-send-email-mchan@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
---
drivers/net/bnx2.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c
index 58ab3f4..4450177 100644
--- a/drivers/net/bnx2.c
+++ b/drivers/net/bnx2.c
@@ -59,8 +59,8 @@
#define DRV_MODULE_NAME "bnx2"
#define PFX DRV_MODULE_NAME ": "
-#define DRV_MODULE_VERSION "2.0.1"
-#define DRV_MODULE_RELDATE "May 6, 2009"
+#define DRV_MODULE_VERSION "2.0.2"
+#define DRV_MODULE_RELDATE "Aug 21, 2009"
#define FW_MIPS_FILE_06 "bnx2/bnx2-mips-06-4.6.16.fw"
#define FW_RV2P_FILE_06 "bnx2/bnx2-rv2p-06-4.6.16.fw"
#define FW_MIPS_FILE_09 "bnx2/bnx2-mips-09-4.6.17.fw"
--
1.5.6.GIT
^ permalink raw reply related
* [PATCH -next 7/8] bnx2: Use const on flash_table structure.
From: Michael Chan @ 2009-08-22 2:20 UTC (permalink / raw)
To: davem; +Cc: netdev
In-Reply-To: <1250907650-24672-1-git-send-email-mchan@broadcom.com>
The structure, once initialized, never changes.
Signed-off-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
---
drivers/net/bnx2.c | 6 +++---
drivers/net/bnx2.h | 2 +-
2 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c
index ea6e6b7..58ab3f4 100644
--- a/drivers/net/bnx2.c
+++ b/drivers/net/bnx2.c
@@ -145,7 +145,7 @@ static DEFINE_PCI_DEVICE_TABLE(bnx2_pci_tbl) = {
{ 0, }
};
-static struct flash_spec flash_table[] =
+static const struct flash_spec flash_table[] =
{
#define BUFFERED_FLAGS (BNX2_NV_BUFFERED | BNX2_NV_TRANSLATE)
#define NONBUFFERED_FLAGS (BNX2_NV_WREN)
@@ -234,7 +234,7 @@ static struct flash_spec flash_table[] =
"Buffered flash (256kB)"},
};
-static struct flash_spec flash_5709 = {
+static const struct flash_spec flash_5709 = {
.flags = BNX2_NV_BUFFERED,
.page_bits = BCM5709_FLASH_PAGE_BITS,
.page_size = BCM5709_FLASH_PAGE_SIZE,
@@ -4227,7 +4227,7 @@ bnx2_init_nvram(struct bnx2 *bp)
{
u32 val;
int j, entry_count, rc = 0;
- struct flash_spec *flash;
+ const struct flash_spec *flash;
if (CHIP_NUM(bp) == CHIP_NUM_5709) {
bp->flash_info = &flash_5709;
diff --git a/drivers/net/bnx2.h b/drivers/net/bnx2.h
index 3c004b4..7544188 100644
--- a/drivers/net/bnx2.h
+++ b/drivers/net/bnx2.h
@@ -6889,7 +6889,7 @@ struct bnx2 {
int pm_cap;
int pcix_cap;
- struct flash_spec *flash_info;
+ const struct flash_spec *flash_info;
u32 flash_size;
int status_stats_size;
--
1.5.6.GIT
^ permalink raw reply related
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox