Netdev List
 help / color / mirror / Atom feed
* Re: Major network performance regression in 3.7
From: Eric Dumazet @ 2013-01-06  2:22 UTC (permalink / raw)
  To: Willy Tarreau; +Cc: netdev, linux-kernel
In-Reply-To: <20130106021824.GW16031@1wt.eu>

On Sun, 2013-01-06 at 03:18 +0100, Willy Tarreau wrote:
> On Sat, Jan 05, 2013 at 06:16:31PM -0800, Eric Dumazet wrote:
> > On Sat, 2013-01-05 at 17:51 -0800, Eric Dumazet wrote:
> > > On Sat, 2013-01-05 at 17:40 -0800, Eric Dumazet wrote:
> > > > On Sun, 2013-01-06 at 02:30 +0100, Willy Tarreau wrote:
> > > > 
> > > > > Ah interesting because these were some of the mm patches that I had
> > > > > tried to revert.
> > > > 
> > > > Hmm, or we should fix __skb_splice_bits()
> > > > 
> > > > I'll send a patch.
> > > > 
> > > 
> > > Could you try the following ?
> > 
> > Or more exactly...
> 
> The first one did not change a iota unfortunately. I'm about to
> spot the commit causing the loopback regression. It's a few patches
> before the first one you pointed. It's almost finished and I test
> your patch below immediately after.

I bet you are going to find commit
69b08f62e17439ee3d436faf0b9a7ca6fffb78db
(net: use bigger pages in __netdev_alloc_frag )

Am I wrong ?

^ permalink raw reply

* Re: [PATCH] tcp: make sysctl_tcp_ecn namespace aware
From: Eric Dumazet @ 2013-01-06  2:19 UTC (permalink / raw)
  To: Hannes Frederic Sowa; +Cc: Stephen Hemminger, netdev, davem
In-Reply-To: <20130106021048.GA32202@order.stressinduktion.org>

On Sun, 2013-01-06 at 03:10 +0100, Hannes Frederic Sowa wrote:
> As per suggestion from Eric Dumazet this patch makes tcp_ecn sysctl
> namespace aware.  The reason behind this patch is to ease the testing
> of ecn problems on the internet and allows applications to tune their
> own use of ecn.
> 
> Cc: Eric Dumazet <eric.dumazet@gmail.com>
> Cc: David Miller <davem@davemloft.net>
> Cc: Stephen Hemminger <shemminger@vyatta.com>
> Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
> ---

Seems good to me, thanks !

Acked-by: Eric Dumazet <edumazet@google.com>

^ permalink raw reply

* [PATCH 1/1] net: fec: report correct hardware stamping info to ethtool
From: Frank Li @ 2013-01-06  1:38 UTC (permalink / raw)
  To: lznuaa, shawn.guo, richardcochran, davem, linux-arm-kernel,
	netdev
  Cc: s.hauer, Frank Li

Report correct hardware stamping capibilty by ethtool interface.
The v1.0 ptp4l check it.

Signed-off-by: Frank Li <Frank.Li@freescale.com>
---
 drivers/net/ethernet/freescale/fec.c |   28 +++++++++++++++++++++++++++-
 1 files changed, 27 insertions(+), 1 deletions(-)

diff --git a/drivers/net/ethernet/freescale/fec.c b/drivers/net/ethernet/freescale/fec.c
index a379319..c713db5 100644
--- a/drivers/net/ethernet/freescale/fec.c
+++ b/drivers/net/ethernet/freescale/fec.c
@@ -1173,12 +1173,38 @@ static void fec_enet_get_drvinfo(struct net_device *ndev,
 	strcpy(info->bus_info, dev_name(&ndev->dev));
 }
 
+static int fec_enet_get_ts_info(struct net_device *ndev,
+				struct ethtool_ts_info *info)
+{
+	struct fec_enet_private *fep = netdev_priv(ndev);
+	int ret;
+	ret = ethtool_op_get_ts_info(ndev, info);
+
+	if (fep->bufdesc_ex) {
+
+		info->so_timestamping |= SOF_TIMESTAMPING_TX_HARDWARE |
+					 SOF_TIMESTAMPING_RX_HARDWARE |
+					 SOF_TIMESTAMPING_RAW_HARDWARE;
+		if (fep->ptp_clock)
+			info->phc_index = ptp_clock_index(fep->ptp_clock);
+		else
+			info->phc_index = -1;
+
+		info->tx_types = (1 << HWTSTAMP_TX_OFF) |
+				 (1 << HWTSTAMP_TX_ON);
+
+		info->rx_filters = (1 << HWTSTAMP_FILTER_NONE) |
+				   (1 << HWTSTAMP_FILTER_ALL);
+	}
+	return ret;
+}
+
 static const struct ethtool_ops fec_enet_ethtool_ops = {
 	.get_settings		= fec_enet_get_settings,
 	.set_settings		= fec_enet_set_settings,
 	.get_drvinfo		= fec_enet_get_drvinfo,
 	.get_link		= ethtool_op_get_link,
-	.get_ts_info		= ethtool_op_get_ts_info,
+	.get_ts_info		= fec_enet_get_ts_info,
 };
 
 static int fec_enet_ioctl(struct net_device *ndev, struct ifreq *rq, int cmd)
-- 
1.7.1

^ permalink raw reply related

* Re: Major network performance regression in 3.7
From: Willy Tarreau @ 2013-01-06  2:18 UTC (permalink / raw)
  To: Eric Dumazet; +Cc: netdev, linux-kernel
In-Reply-To: <1357438591.1678.5205.camel@edumazet-glaptop>

On Sat, Jan 05, 2013 at 06:16:31PM -0800, Eric Dumazet wrote:
> On Sat, 2013-01-05 at 17:51 -0800, Eric Dumazet wrote:
> > On Sat, 2013-01-05 at 17:40 -0800, Eric Dumazet wrote:
> > > On Sun, 2013-01-06 at 02:30 +0100, Willy Tarreau wrote:
> > > 
> > > > Ah interesting because these were some of the mm patches that I had
> > > > tried to revert.
> > > 
> > > Hmm, or we should fix __skb_splice_bits()
> > > 
> > > I'll send a patch.
> > > 
> > 
> > Could you try the following ?
> 
> Or more exactly...

The first one did not change a iota unfortunately. I'm about to
spot the commit causing the loopback regression. It's a few patches
before the first one you pointed. It's almost finished and I test
your patch below immediately after.

Thanks,
Willy

> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index 3ab989b..01f222c 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -1736,11 +1736,8 @@ static bool __splice_segment(struct page *page, unsigned int poff,
>  		return false;
>  	}
>  
> -	/* ignore any bits we already processed */
> -	if (*off) {
> -		__segment_seek(&page, &poff, &plen, *off);
> -		*off = 0;
> -	}
> +	__segment_seek(&page, &poff, &plen, *off);
> +	*off = 0;
>  
>  	do {
>  		unsigned int flen = min(*len, plen);
> @@ -1768,14 +1765,15 @@ static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
>  			      struct splice_pipe_desc *spd, struct sock *sk)
>  {
>  	int seg;
> +	struct page *page = virt_to_page(skb->data);
> +	unsigned int poff = skb->data - (unsigned char *)page_address(page);
>  
>  	/* map the linear part :
>  	 * If skb->head_frag is set, this 'linear' part is backed by a
>  	 * fragment, and if the head is not shared with any clones then
>  	 * we can avoid a copy since we own the head portion of this page.
>  	 */
> -	if (__splice_segment(virt_to_page(skb->data),
> -			     (unsigned long) skb->data & (PAGE_SIZE - 1),
> +	if (__splice_segment(page, poff,
>  			     skb_headlen(skb),
>  			     offset, len, skb, spd,
>  			     skb_head_is_locked(skb),
> 

^ permalink raw reply

* Re: Major network performance regression in 3.7
From: Eric Dumazet @ 2013-01-06  2:16 UTC (permalink / raw)
  To: Willy Tarreau; +Cc: netdev, linux-kernel
In-Reply-To: <1357437086.1678.5135.camel@edumazet-glaptop>

On Sat, 2013-01-05 at 17:51 -0800, Eric Dumazet wrote:
> On Sat, 2013-01-05 at 17:40 -0800, Eric Dumazet wrote:
> > On Sun, 2013-01-06 at 02:30 +0100, Willy Tarreau wrote:
> > 
> > > Ah interesting because these were some of the mm patches that I had
> > > tried to revert.
> > 
> > Hmm, or we should fix __skb_splice_bits()
> > 
> > I'll send a patch.
> > 
> 
> Could you try the following ?

Or more exactly...

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 3ab989b..01f222c 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -1736,11 +1736,8 @@ static bool __splice_segment(struct page *page, unsigned int poff,
 		return false;
 	}
 
-	/* ignore any bits we already processed */
-	if (*off) {
-		__segment_seek(&page, &poff, &plen, *off);
-		*off = 0;
-	}
+	__segment_seek(&page, &poff, &plen, *off);
+	*off = 0;
 
 	do {
 		unsigned int flen = min(*len, plen);
@@ -1768,14 +1765,15 @@ static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
 			      struct splice_pipe_desc *spd, struct sock *sk)
 {
 	int seg;
+	struct page *page = virt_to_page(skb->data);
+	unsigned int poff = skb->data - (unsigned char *)page_address(page);
 
 	/* map the linear part :
 	 * If skb->head_frag is set, this 'linear' part is backed by a
 	 * fragment, and if the head is not shared with any clones then
 	 * we can avoid a copy since we own the head portion of this page.
 	 */
-	if (__splice_segment(virt_to_page(skb->data),
-			     (unsigned long) skb->data & (PAGE_SIZE - 1),
+	if (__splice_segment(page, poff,
 			     skb_headlen(skb),
 			     offset, len, skb, spd,
 			     skb_head_is_locked(skb),

^ permalink raw reply related

* Re: [PATCH] tcp: make sysctl_tcp_ecn namespace aware
From: Hannes Frederic Sowa @ 2013-01-06  2:10 UTC (permalink / raw)
  To: Eric Dumazet; +Cc: Stephen Hemminger, netdev, davem
In-Reply-To: <1357435646.1678.5082.camel@edumazet-glaptop>

As per suggestion from Eric Dumazet this patch makes tcp_ecn sysctl
namespace aware.  The reason behind this patch is to ease the testing
of ecn problems on the internet and allows applications to tune their
own use of ecn.

Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: David Miller <davem@davemloft.net>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
---
 include/net/netns/ipv4.h   |  2 ++
 include/net/tcp.h          |  9 +++++----
 net/ipv4/syncookies.c      |  7 ++++---
 net/ipv4/sysctl_net_ipv4.c | 16 +++++++++-------
 net/ipv4/tcp_input.c       |  2 --
 net/ipv4/tcp_ipv4.c        |  3 ++-
 net/ipv4/tcp_output.c      |  2 +-
 net/ipv6/syncookies.c      |  2 +-
 net/ipv6/tcp_ipv6.c        |  2 +-
 9 files changed, 25 insertions(+), 20 deletions(-)

diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
index 2ae2b83..9b78862 100644
--- a/include/net/netns/ipv4.h
+++ b/include/net/netns/ipv4.h
@@ -61,6 +61,8 @@ struct netns_ipv4 {
 	int sysctl_icmp_ratemask;
 	int sysctl_icmp_errors_use_inbound_ifaddr;
 
+	int sysctl_tcp_ecn;
+
 	kgid_t sysctl_ping_group_range[2];
 	long sysctl_tcp_mem[3];
 
diff --git a/include/net/tcp.h b/include/net/tcp.h
index aed42c7..614af8b 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -266,7 +266,6 @@ extern int sysctl_tcp_abort_on_overflow;
 extern int sysctl_tcp_max_orphans;
 extern int sysctl_tcp_fack;
 extern int sysctl_tcp_reordering;
-extern int sysctl_tcp_ecn;
 extern int sysctl_tcp_dsack;
 extern int sysctl_tcp_wmem[3];
 extern int sysctl_tcp_rmem[3];
@@ -504,7 +503,8 @@ static inline __u32 cookie_v4_init_sequence(struct sock *sk,
 #endif
 
 extern __u32 cookie_init_timestamp(struct request_sock *req);
-extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
+extern bool cookie_check_timestamp(struct tcp_options_received *opt,
+				struct net *net, bool *ecn_ok);
 
 /* From net/ipv6/syncookies.c */
 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
@@ -728,11 +728,12 @@ struct tcp_skb_cb {
  * notifications, we disable TCP ECN negociation.
  */
 static inline void
-TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
+TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb,
+		struct net *net)
 {
 	const struct tcphdr *th = tcp_hdr(skb);
 
-	if (sysctl_tcp_ecn && th->ece && th->cwr &&
+	if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr &&
 	    INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
 		inet_rsk(req)->ecn_ok = 1;
 }
diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
index b236ef0..ef54377 100644
--- a/net/ipv4/syncookies.c
+++ b/net/ipv4/syncookies.c
@@ -232,7 +232,8 @@ static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
  *
  * return false if we decode an option that should not be.
  */
-bool cookie_check_timestamp(struct tcp_options_received *tcp_opt, bool *ecn_ok)
+bool cookie_check_timestamp(struct tcp_options_received *tcp_opt,
+			struct net *net, bool *ecn_ok)
 {
 	/* echoed timestamp, lowest bits contain options */
 	u32 options = tcp_opt->rcv_tsecr & TSMASK;
@@ -247,7 +248,7 @@ bool cookie_check_timestamp(struct tcp_options_received *tcp_opt, bool *ecn_ok)
 
 	tcp_opt->sack_ok = (options & (1 << 4)) ? TCP_SACK_SEEN : 0;
 	*ecn_ok = (options >> 5) & 1;
-	if (*ecn_ok && !sysctl_tcp_ecn)
+	if (*ecn_ok && !net->ipv4.sysctl_tcp_ecn)
 		return false;
 
 	if (tcp_opt->sack_ok && !sysctl_tcp_sack)
@@ -295,7 +296,7 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
 	memset(&tcp_opt, 0, sizeof(tcp_opt));
 	tcp_parse_options(skb, &tcp_opt, &hash_location, 0, NULL);
 
-	if (!cookie_check_timestamp(&tcp_opt, &ecn_ok))
+	if (!cookie_check_timestamp(&tcp_opt, sock_net(sk), &ecn_ok))
 		goto out;
 
 	ret = NULL;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index d84400b..7716d40 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -538,13 +538,6 @@ static struct ctl_table ipv4_table[] = {
 		.proc_handler	= proc_dointvec
 	},
 	{
-		.procname	= "tcp_ecn",
-		.data		= &sysctl_tcp_ecn,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= proc_dointvec
-	},
-	{
 		.procname	= "tcp_dsack",
 		.data		= &sysctl_tcp_dsack,
 		.maxlen		= sizeof(int),
@@ -850,6 +843,13 @@ static struct ctl_table ipv4_net_table[] = {
 		.proc_handler	= ipv4_ping_group_range,
 	},
 	{
+		.procname	= "tcp_ecn",
+		.data		= &init_net.ipv4.sysctl_tcp_ecn,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec
+	},
+	{
 		.procname	= "tcp_mem",
 		.maxlen		= sizeof(init_net.ipv4.sysctl_tcp_mem),
 		.mode		= 0644,
@@ -882,6 +882,8 @@ static __net_init int ipv4_sysctl_init_net(struct net *net)
 			&net->ipv4.sysctl_icmp_ratemask;
 		table[6].data =
 			&net->ipv4.sysctl_ping_group_range;
+		table[7].data =
+			&net->ipv4.sysctl_tcp_ecn;
 
 		/* Don't export sysctls to unprivileged users */
 		if (net->user_ns != &init_user_ns)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index a28e4db..38e1184 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -81,8 +81,6 @@ int sysctl_tcp_sack __read_mostly = 1;
 int sysctl_tcp_fack __read_mostly = 1;
 int sysctl_tcp_reordering __read_mostly = TCP_FASTRETRANS_THRESH;
 EXPORT_SYMBOL(sysctl_tcp_reordering);
-int sysctl_tcp_ecn __read_mostly = 2;
-EXPORT_SYMBOL(sysctl_tcp_ecn);
 int sysctl_tcp_dsack __read_mostly = 1;
 int sysctl_tcp_app_win __read_mostly = 31;
 int sysctl_tcp_adv_win_scale __read_mostly = 1;
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 54139fa..c6ce9ca 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -1568,7 +1568,7 @@ int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 		goto drop_and_free;
 
 	if (!want_cookie || tmp_opt.tstamp_ok)
-		TCP_ECN_create_request(req, skb);
+		TCP_ECN_create_request(req, skb, sock_net(sk));
 
 	if (want_cookie) {
 		isn = cookie_v4_init_sequence(sk, skb, &req->mss);
@@ -2888,6 +2888,7 @@ EXPORT_SYMBOL(tcp_prot);
 
 static int __net_init tcp_sk_init(struct net *net)
 {
+	net->ipv4.sysctl_tcp_ecn = 2;
 	return 0;
 }
 
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 5d45159..667a6ad 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -314,7 +314,7 @@ static inline void TCP_ECN_send_syn(struct sock *sk, struct sk_buff *skb)
 	struct tcp_sock *tp = tcp_sk(sk);
 
 	tp->ecn_flags = 0;
-	if (sysctl_tcp_ecn == 1) {
+	if (sock_net(sk)->ipv4.sysctl_tcp_ecn == 1) {
 		TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
 		tp->ecn_flags = TCP_ECN_OK;
 	}
diff --git a/net/ipv6/syncookies.c b/net/ipv6/syncookies.c
index 4016197..8a0848b 100644
--- a/net/ipv6/syncookies.c
+++ b/net/ipv6/syncookies.c
@@ -179,7 +179,7 @@ struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb)
 	memset(&tcp_opt, 0, sizeof(tcp_opt));
 	tcp_parse_options(skb, &tcp_opt, &hash_location, 0, NULL);
 
-	if (!cookie_check_timestamp(&tcp_opt, &ecn_ok))
+	if (!cookie_check_timestamp(&tcp_opt, sock_net(sk), &ecn_ok))
 		goto out;
 
 	ret = NULL;
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 93825dd..3164ad2 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -1027,7 +1027,7 @@ static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 	treq->rmt_addr = ipv6_hdr(skb)->saddr;
 	treq->loc_addr = ipv6_hdr(skb)->daddr;
 	if (!want_cookie || tmp_opt.tstamp_ok)
-		TCP_ECN_create_request(req, skb);
+		TCP_ECN_create_request(req, skb, sock_net(sk));
 
 	treq->iif = sk->sk_bound_dev_if;
 
-- 
1.7.11.7

^ permalink raw reply related

* Re: Major network performance regression in 3.7
From: Eric Dumazet @ 2013-01-06  1:51 UTC (permalink / raw)
  To: Willy Tarreau; +Cc: netdev, linux-kernel
In-Reply-To: <1357436430.1678.5111.camel@edumazet-glaptop>

On Sat, 2013-01-05 at 17:40 -0800, Eric Dumazet wrote:
> On Sun, 2013-01-06 at 02:30 +0100, Willy Tarreau wrote:
> 
> > Ah interesting because these were some of the mm patches that I had
> > tried to revert.
> 
> Hmm, or we should fix __skb_splice_bits()
> 
> I'll send a patch.
> 

Could you try the following ?

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 3ab989b..c5246be 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -1768,14 +1768,15 @@ static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
 			      struct splice_pipe_desc *spd, struct sock *sk)
 {
 	int seg;
+	struct page *page = virt_to_page(skb->data);
+	unsigned int poff = skb->data - (unsigned char *)page_address(page);
 
 	/* map the linear part :
 	 * If skb->head_frag is set, this 'linear' part is backed by a
 	 * fragment, and if the head is not shared with any clones then
 	 * we can avoid a copy since we own the head portion of this page.
 	 */
-	if (__splice_segment(virt_to_page(skb->data),
-			     (unsigned long) skb->data & (PAGE_SIZE - 1),
+	if (__splice_segment(page, poff,
 			     skb_headlen(skb),
 			     offset, len, skb, spd,
 			     skb_head_is_locked(skb),

^ permalink raw reply related

* Re: Major network performance regression in 3.7
From: Eric Dumazet @ 2013-01-06  1:40 UTC (permalink / raw)
  To: Willy Tarreau; +Cc: netdev, linux-kernel
In-Reply-To: <20130106013027.GV16031@1wt.eu>

On Sun, 2013-01-06 at 02:30 +0100, Willy Tarreau wrote:

> Ah interesting because these were some of the mm patches that I had
> tried to revert.

Hmm, or we should fix __skb_splice_bits()

I'll send a patch.

^ permalink raw reply

* Re: [PATCH v2] tcp: split tcp_ecn sysctl knob to distinguish between IPv4 and IPv6
From: Hannes Frederic Sowa @ 2013-01-06  1:31 UTC (permalink / raw)
  To: David Miller; +Cc: netdev
In-Reply-To: <20130103.045350.1322196416782379673.davem@davemloft.net>

On Thu, Jan 03, 2013 at 04:53:50AM -0800, David Miller wrote:
> > On Thu, Jan 03, 2013 at 01:00:09AM +0100, Hannes Frederic Sowa wrote:
> >> ECN could be more reliable when used with IPv6 (I don't have proofs). For
> >> people who want to try ECN with IPv6 but still have problems connecting
> >> to destinations because of broken IPv4 routers this switch allows one
> >> to enable ECN just for IPv6.
> >> 
> >> Perhaps ECN could be enabled by default in future.
> >> 
> >> No code changes since initial submission, just added documentation.
> > 
> > May I ask what lead to the rejection of the patch?
> 
> I don't see any value in splitting up ipv4 and ipv6.
> 
> I think we should turn the existing knob on unliaterally
> by default.  That would be a much better patch.

I wonder if we really should go down this route to turn on ecn by
default or if I should try to come up with a patch which marks non
working ecn paths in the destination cache and suppress ecn for those
connections. Any advice?

^ permalink raw reply

* Re: Major network performance regression in 3.7
From: Willy Tarreau @ 2013-01-06  1:30 UTC (permalink / raw)
  To: Eric Dumazet; +Cc: netdev, linux-kernel
In-Reply-To: <1357435276.1678.5067.camel@edumazet-glaptop>

On Sat, Jan 05, 2013 at 05:21:16PM -0800, Eric Dumazet wrote:
> On Sun, 2013-01-06 at 01:50 +0100, Willy Tarreau wrote:
> 
> > Yes, I've removed all zero counters in this short view for easier
> > reading (complete version appended at the end of this email). This
> > was after around 140 GB were transferred :
> 
> OK I only wanted to make sure skb were not linearized in xmit.
> 
> Could you try to disable CONFIG_COMPACTION ?

It's already disabled.

> ( This is the other thread mentioning this : "ppoll() stuck on POLLIN
> while TCP peer is sending" )

Ah interesting because these were some of the mm patches that I had
tried to revert.

Willy

^ permalink raw reply

* Re: [PATCH] tcp: make sysctl_tcp_ecn namespace aware
From: Eric Dumazet @ 2013-01-06  1:27 UTC (permalink / raw)
  To: Hannes Frederic Sowa; +Cc: Stephen Hemminger, netdev, davem
In-Reply-To: <20130106011849.GA28545@order.stressinduktion.org>

On Sun, 2013-01-06 at 02:18 +0100, Hannes Frederic Sowa wrote:

> 
> I just verified that req->sk is indeed not initialized at that point. I
> had a kernel panic on the first syn packet. So I would propose to keep
> the patch as it is.
> 

Please pass the "struct net" pointer instead, so that the tcp_ecn
dereference is done only if really needed.

Naming a function parameter, or an automatic variable, sysctl_tcp_ecn is
not very nice.

^ permalink raw reply

* Re: Major network performance regression in 3.7
From: Eric Dumazet @ 2013-01-06  1:21 UTC (permalink / raw)
  To: Willy Tarreau; +Cc: netdev, linux-kernel
In-Reply-To: <20130106005053.GS16031@1wt.eu>

On Sun, 2013-01-06 at 01:50 +0100, Willy Tarreau wrote:

> Yes, I've removed all zero counters in this short view for easier
> reading (complete version appended at the end of this email). This
> was after around 140 GB were transferred :

OK I only wanted to make sure skb were not linearized in xmit.

Could you try to disable CONFIG_COMPACTION ?

( This is the other thread mentioning this : "ppoll() stuck on POLLIN
while TCP peer is sending" )

^ permalink raw reply

* Re: [PATCH] tcp: make sysctl_tcp_ecn namespace aware
From: Hannes Frederic Sowa @ 2013-01-06  1:18 UTC (permalink / raw)
  To: Stephen Hemminger, netdev, eric.dumazet, davem
In-Reply-To: <20130105190508.GE4031@order.stressinduktion.org>

On Sat, Jan 05, 2013 at 08:05:08PM +0100, Hannes Frederic Sowa wrote:
> On Sat, Jan 05, 2013 at 11:00:05AM -0800, Stephen Hemminger wrote:
> > Rather than passing sysctl_tcp_ecn around as a parameter, I think it
> > would be clearer and more efficient to make the ECN functions namespace
> > aware.
> > 
> > Instead of:
> > > @@ -728,7 +728,8 @@ struct tcp_skb_cb {
> > >   * notifications, we disable TCP ECN negociation.
> > >   */
> > >  static inline void
> > > -TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
> > > +TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb,
> > > +		int sysctl_tcp_ecn)
> > >  {
> > >  	const struct tcphdr *th = tcp_hdr(skb);
> > >  
> > 
> > @@ -731,8 +730,9 @@ static inline void
> >  TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
> >  {
> >         const struct tcphdr *th = tcp_hdr(skb);
> > -
> > -       if (sysctl_tcp_ecn && th->ece && th->cwr &&
> > +       
> > +       if (sock_net(req->sk)->ipv4.sysctl_tcp_ecn && 
> > +           th->ece && th->cwr &&
> >             INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
> >                 inet_rsk(req)->ecn_ok = 1;
> 
> Having a quick look, I think req->sk is not initialized at that moment, but
> I'll verify later.

I just verified that req->sk is indeed not initialized at that point. I
had a kernel panic on the first syn packet. So I would propose to keep
the patch as it is.

^ permalink raw reply

* Re: Major network performance regression in 3.7
From: Willy Tarreau @ 2013-01-06  0:50 UTC (permalink / raw)
  To: Eric Dumazet; +Cc: netdev, linux-kernel
In-Reply-To: <1357430523.1678.4901.camel@edumazet-glaptop>

On Sat, Jan 05, 2013 at 04:02:03PM -0800, Eric Dumazet wrote:
> On Sun, 2013-01-06 at 00:29 +0100, Willy Tarreau wrote:
> 
> > > 2) Another possibility would be that Myri card/driver doesnt like very
> > > well high order pages.
> > 
> > It looks like it has not changed much since 3.6 :-/ I really suspect
> > something is wrong with memory allocation. I have tried reverting many
> > patches affecting the mm/ directory just in case but I did not come to
> > anything useful yet.
> > 
> 
> Hmm, I was referring to TCP stack now using order-3 pages instead of
> order-0 ones
> 
> See commit 5640f7685831e088fe6c2e1f863a6805962f8e81
> (net: use a per task frag allocator)

OK, so you think there are two distinct problems ?

I have tried to revert this one but it did not change the performance, I'm
still saturating at ~6.9 Gbps.

> Could you please post :
> 
> ethtool -S eth0

Yes, I've removed all zero counters in this short view for easier
reading (complete version appended at the end of this email). This
was after around 140 GB were transferred :

# ethtool -S eth1|grep -vw 0
NIC statistics:
     rx_packets: 8001500
     tx_packets: 10015409
     rx_bytes: 480115998
     tx_bytes: 148825674976
     tx_boundary: 2048
     WC: 1
     irq: 45
     MSI: 1
     read_dma_bw_MBs: 1200
     write_dma_bw_MBs: 1614
     read_write_dma_bw_MBs: 2101
     serial_number: 320061
     link_changes: 2
     link_up: 1
     tx_pkt_start: 10015409
     tx_pkt_done: 10015409
     tx_req: 93407411
     tx_done: 93407411
     rx_small_cnt: 8001500
     wake_queue: 187727
     stop_queue: 187727
     LRO aggregated: 146
     LRO flushed: 146
     LRO avg aggr: 1
     LRO no_desc: 80

Quite honnestly, this is typically the pattern what I'm used to
observe here. I'm now trying to bisect, hopefully we'll get
something exploitable.

Cheers,
Willy

----- full ethtool -S ----

NIC statistics:
     rx_packets: 8001500
     tx_packets: 10015409
     rx_bytes: 480115998
     tx_bytes: 148825674976
     rx_errors: 0
     tx_errors: 0
     rx_dropped: 0
     tx_dropped: 0
     multicast: 0
     collisions: 0
     rx_length_errors: 0
     rx_over_errors: 0
     rx_crc_errors: 0
     rx_frame_errors: 0
     rx_fifo_errors: 0
     rx_missed_errors: 0
     tx_aborted_errors: 0
     tx_carrier_errors: 0
     tx_fifo_errors: 0
     tx_heartbeat_errors: 0
     tx_window_errors: 0
     tx_boundary: 2048
     WC: 1
     irq: 45
     MSI: 1
     MSIX: 0
     read_dma_bw_MBs: 1200
     write_dma_bw_MBs: 1614
     read_write_dma_bw_MBs: 2101
     serial_number: 320061
     watchdog_resets: 0
     link_changes: 2
     link_up: 1
     dropped_link_overflow: 0
     dropped_link_error_or_filtered: 0
     dropped_pause: 0
     dropped_bad_phy: 0
     dropped_bad_crc32: 0
     dropped_unicast_filtered: 0
     dropped_multicast_filtered: 0
     dropped_runt: 0
     dropped_overrun: 0
     dropped_no_small_buffer: 0
     dropped_no_big_buffer: 0
     ----------- slice ---------: 0
     tx_pkt_start: 10015409
     tx_pkt_done: 10015409
     tx_req: 93407411
     tx_done: 93407411
     rx_small_cnt: 8001500
     rx_big_cnt: 0
     wake_queue: 187727
     stop_queue: 187727
     tx_linearized: 0
     LRO aggregated: 146
     LRO flushed: 146
     LRO avg aggr: 1
     LRO no_desc: 80

^ permalink raw reply

* [PATCH V4] bgmac: driver for GBit MAC core on BCMA bus
From: Rafał Miłecki @ 2013-01-06  0:20 UTC (permalink / raw)
  To: netdev, David S. Miller
  Cc: Francois Romieu, Joe Perches, Rafał Miłecki

BCMA is a Broadcom specific bus with devices AKA cores. All recent BCMA
based SoCs have gigabit ethernet provided by the GBit MAC core. This
patch adds driver for such a cores registering itself as a netdev. It
has been tested on a BCM4706 and BCM4718 chipsets.

In the kernel tree there is already b44 driver which has some common
things with bgmac, however there are many differences that has led to
the decision or writing a new driver:
1) GBit MAC cores appear on BCMA bus (not SSB as in case of b44)
2) There is 64bit DMA engine which differs from 32bit one
3) There is no CAM (Content Addressable Memory) in GBit MAC
4) We have 4 TX queues on GBit MAC devices (instead of 1)
5) Many registers have different addresses/values
6) RX header flags are also different

The driver in it's state is functional how, however there is of course
place for improvements:
1) Supporting more net_device_ops
2) SUpporting more ethtool_ops
3) Unaligned addressing in DMA
4) Writing separated PHY driver

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
---
V2: fix typos
     improve stat_xmit (netif_stop_queue)
     change scope of some variables
     switch to dev_kfree_skb
     change net_dev->stats.rx_dropped
     drop "bool tx;" from DMA ring struct
     bgmac_open: fail on IRQ request error
     use dev<level> for printing (this is great!)
V3: fix strings (long lines)
    fix returned values in xmit
    some extra info message about PHY addr
V4: stop queue when ring is full
---
 drivers/bcma/driver_chipcommon_pmu.c        |    3 +-
 drivers/net/ethernet/broadcom/Kconfig       |    9 +
 drivers/net/ethernet/broadcom/Makefile      |    1 +
 drivers/net/ethernet/broadcom/bgmac.c       | 1420 +++++++++++++++++++++++++++
 drivers/net/ethernet/broadcom/bgmac.h       |  456 +++++++++
 include/linux/bcma/bcma_driver_chipcommon.h |    2 +
 6 files changed, 1890 insertions(+), 1 deletions(-)
 create mode 100644 drivers/net/ethernet/broadcom/bgmac.c
 create mode 100644 drivers/net/ethernet/broadcom/bgmac.h

diff --git a/drivers/bcma/driver_chipcommon_pmu.c b/drivers/bcma/driver_chipcommon_pmu.c
index c62c788..932b101 100644
--- a/drivers/bcma/driver_chipcommon_pmu.c
+++ b/drivers/bcma/driver_chipcommon_pmu.c
@@ -264,7 +264,7 @@ static u32 bcma_pmu_pll_clock_bcm4706(struct bcma_drv_cc *cc, u32 pll0, u32 m)
 }
 
 /* query bus clock frequency for PMU-enabled chipcommon */
-static u32 bcma_pmu_get_bus_clock(struct bcma_drv_cc *cc)
+u32 bcma_pmu_get_bus_clock(struct bcma_drv_cc *cc)
 {
 	struct bcma_bus *bus = cc->core->bus;
 
@@ -293,6 +293,7 @@ static u32 bcma_pmu_get_bus_clock(struct bcma_drv_cc *cc)
 	}
 	return BCMA_CC_PMU_HT_CLOCK;
 }
+EXPORT_SYMBOL_GPL(bcma_pmu_get_bus_clock);
 
 /* query cpu clock frequency for PMU-enabled chipcommon */
 u32 bcma_pmu_get_cpu_clock(struct bcma_drv_cc *cc)
diff --git a/drivers/net/ethernet/broadcom/Kconfig b/drivers/net/ethernet/broadcom/Kconfig
index f552673..42c0913 100644
--- a/drivers/net/ethernet/broadcom/Kconfig
+++ b/drivers/net/ethernet/broadcom/Kconfig
@@ -121,4 +121,13 @@ config BNX2X
 	  To compile this driver as a module, choose M here: the module
 	  will be called bnx2x.  This is recommended.
 
+config BGMAC
+	tristate "BCMA bus GBit core support"
+	depends on BCMA_HOST_SOC && HAS_DMA
+	---help---
+	  This driver supports GBit MAC and BCM4706 GBit MAC cores on BCMA bus.
+	  They can be found on BCM47xx SoCs and provide gigabit ethernet.
+	  In case of using this driver on BCM4706 it's also requires to enable
+	  BCMA_DRIVER_GMAC_CMN to make it work.
+
 endif # NET_VENDOR_BROADCOM
diff --git a/drivers/net/ethernet/broadcom/Makefile b/drivers/net/ethernet/broadcom/Makefile
index b789605..68efa1a 100644
--- a/drivers/net/ethernet/broadcom/Makefile
+++ b/drivers/net/ethernet/broadcom/Makefile
@@ -9,3 +9,4 @@ obj-$(CONFIG_CNIC) += cnic.o
 obj-$(CONFIG_BNX2X) += bnx2x/
 obj-$(CONFIG_SB1250_MAC) += sb1250-mac.o
 obj-$(CONFIG_TIGON3) += tg3.o
+obj-$(CONFIG_BGMAC) += bgmac.o
diff --git a/drivers/net/ethernet/broadcom/bgmac.c b/drivers/net/ethernet/broadcom/bgmac.c
new file mode 100644
index 0000000..3073196
--- /dev/null
+++ b/drivers/net/ethernet/broadcom/bgmac.c
@@ -0,0 +1,1420 @@
+/*
+ * Driver for (BCM4706)? GBit MAC core on BCMA bus.
+ *
+ * Copyright (C) 2012 Rafał Miłecki <zajec5@gmail.com>
+ *
+ * Licensed under the GNU/GPL. See COPYING for details.
+ */
+
+#include "bgmac.h"
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/delay.h>
+#include <linux/etherdevice.h>
+#include <linux/mii.h>
+#include <linux/interrupt.h>
+#include <linux/dma-mapping.h>
+#include <asm/mach-bcm47xx/nvram.h>
+
+static const struct bcma_device_id bgmac_bcma_tbl[] = {
+	BCMA_CORE(BCMA_MANUF_BCM, BCMA_CORE_4706_MAC_GBIT, BCMA_ANY_REV, BCMA_ANY_CLASS),
+	BCMA_CORE(BCMA_MANUF_BCM, BCMA_CORE_MAC_GBIT, BCMA_ANY_REV, BCMA_ANY_CLASS),
+	BCMA_CORETABLE_END
+};
+MODULE_DEVICE_TABLE(bcma, bgmac_bcma_tbl);
+
+static bool bgmac_wait_value(struct bcma_device *core, u16 reg, u32 mask,
+			     u32 value, int timeout)
+{
+	u32 val;
+	int i;
+
+	for (i = 0; i < timeout / 10; i++) {
+		val = bcma_read32(core, reg);
+		if ((val & mask) == value)
+			return true;
+		udelay(10);
+	}
+	pr_err("Timeout waiting for reg 0x%X\n", reg);
+	return false;
+}
+
+/**************************************************
+ * DMA
+ **************************************************/
+
+static void bgmac_dma_tx_reset(struct bgmac *bgmac, struct bgmac_dma_ring *ring)
+{
+	u32 val;
+	int i;
+
+	if (!ring->mmio_base)
+		return;
+
+	/* Suspend DMA TX ring first.
+	 * bgmac_wait_value doesn't support waiting for any of few values, so
+	 * implement whole loop here.
+	 */
+	bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_TX_CTL,
+		    BGMAC_DMA_TX_SUSPEND);
+	for (i = 0; i < 10000 / 10; i++) {
+		val = bgmac_read(bgmac, ring->mmio_base + BGMAC_DMA_TX_STATUS);
+		val &= BGMAC_DMA_TX_STAT;
+		if (val == BGMAC_DMA_TX_STAT_DISABLED ||
+		    val == BGMAC_DMA_TX_STAT_IDLEWAIT ||
+		    val == BGMAC_DMA_TX_STAT_STOPPED) {
+			i = 0;
+			break;
+		}
+		udelay(10);
+	}
+	if (i)
+		bgmac_err(bgmac, "Timeout suspending DMA TX ring 0x%X (BGMAC_DMA_TX_STAT: 0x%08X)\n",
+			  ring->mmio_base, val);
+
+	/* Remove SUSPEND bit */
+	bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_TX_CTL, 0);
+	if (!bgmac_wait_value(bgmac->core,
+			      ring->mmio_base + BGMAC_DMA_TX_STATUS,
+			      BGMAC_DMA_TX_STAT, BGMAC_DMA_TX_STAT_DISABLED,
+			      10000)) {
+		bgmac_warn(bgmac, "DMA TX ring 0x%X wasn't disabled on time, waiting additional 300us\n",
+			   ring->mmio_base);
+		udelay(300);
+		val = bgmac_read(bgmac, ring->mmio_base + BGMAC_DMA_TX_STATUS);
+		if ((val & BGMAC_DMA_TX_STAT) != BGMAC_DMA_TX_STAT_DISABLED)
+			bgmac_err(bgmac, "Reset of DMA TX ring 0x%X failed\n",
+				  ring->mmio_base);
+	}
+}
+
+static void bgmac_dma_tx_enable(struct bgmac *bgmac,
+				struct bgmac_dma_ring *ring)
+{
+	u32 ctl;
+
+	ctl = bgmac_read(bgmac, ring->mmio_base + BGMAC_DMA_TX_CTL);
+	ctl |= BGMAC_DMA_TX_ENABLE;
+	ctl |= BGMAC_DMA_TX_PARITY_DISABLE;
+	bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_TX_CTL, ctl);
+}
+
+static netdev_tx_t bgmac_dma_tx_add(struct bgmac *bgmac,
+				    struct bgmac_dma_ring *ring,
+				    struct sk_buff *skb)
+{
+	struct device *dma_dev = bgmac->core->dma_dev;
+	struct bgmac_dma_desc *dma_desc;
+	struct bgmac_slot_info *slot;
+	u32 ctl0, ctl1;
+	int free_slots;
+
+	if (skb->len > BGMAC_DESC_CTL1_LEN) {
+		bgmac_err(bgmac, "Too long skb (%d)\n", skb->len);
+		goto err_stop_drop;
+	}
+
+	if (ring->start <= ring->end)
+		free_slots = ring->start - ring->end + BGMAC_TX_RING_SLOTS;
+	else
+		free_slots = ring->start - ring->end;
+	if (free_slots == 1) { /* Should never happen (netif_stop_queue) */
+		bgmac_err(bgmac, "TX after stopping queue!\n");
+		return NETDEV_TX_BUSY;
+	}
+
+	slot = &ring->slots[ring->end];
+	slot->skb = skb;
+	slot->dma_addr = dma_map_single(dma_dev, skb->data, skb->len,
+					DMA_TO_DEVICE);
+	if (dma_mapping_error(dma_dev, slot->dma_addr)) {
+		bgmac_err(bgmac, "Mapping error of skb on ring 0x%X\n",
+			  ring->mmio_base);
+		goto err_stop_drop;
+	}
+
+	ctl0 = BGMAC_DESC_CTL0_IOC | BGMAC_DESC_CTL0_SOF | BGMAC_DESC_CTL0_EOF;
+	if (ring->end == ring->num_slots - 1)
+		ctl0 |= BGMAC_DESC_CTL0_EOT;
+	ctl1 = skb->len & BGMAC_DESC_CTL1_LEN;
+
+	dma_desc = ring->cpu_base;
+	dma_desc += ring->end;
+	dma_desc->addr_low = cpu_to_le32(lower_32_bits(slot->dma_addr));
+	dma_desc->addr_high = cpu_to_le32(upper_32_bits(slot->dma_addr));
+	dma_desc->ctl0 = cpu_to_le32(ctl0);
+	dma_desc->ctl1 = cpu_to_le32(ctl1);
+
+	wmb();
+
+	/* Increase ring->end to point empty slot. We tell hardware the first
+	 * slot it should *not* read.
+	 */
+	if (++ring->end >= BGMAC_TX_RING_SLOTS)
+		ring->end = 0;
+	bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_TX_INDEX,
+		    ring->end * sizeof(struct bgmac_dma_desc));
+
+	/* Always keep one slot free to allow detecting bugged calls. */
+	if (--free_slots == 1)
+		netif_stop_queue(bgmac->net_dev);
+
+	return NETDEV_TX_OK;
+
+err_stop_drop:
+	netif_stop_queue(bgmac->net_dev);
+	dev_kfree_skb(skb);
+	return NETDEV_TX_OK;
+}
+
+/* Free transmitted packets */
+static void bgmac_dma_tx_free(struct bgmac *bgmac, struct bgmac_dma_ring *ring)
+{
+	struct device *dma_dev = bgmac->core->dma_dev;
+	int empty_slot;
+	bool freed = false;
+
+	/* The last slot that hardware didn't consume yet */
+	empty_slot = bgmac_read(bgmac, ring->mmio_base + BGMAC_DMA_TX_STATUS);
+	empty_slot &= BGMAC_DMA_TX_STATDPTR;
+	empty_slot /= sizeof(struct bgmac_dma_desc);
+
+	while (ring->start != empty_slot) {
+		struct bgmac_slot_info *slot = &ring->slots[ring->start];
+
+		if (slot->skb) {
+			/* Unmap no longer used buffer */
+			dma_unmap_single(dma_dev, slot->dma_addr,
+					 slot->skb->len, DMA_TO_DEVICE);
+			slot->dma_addr = 0;
+
+			/* Free memory! :) */
+			dev_kfree_skb(slot->skb);
+			slot->skb = NULL;
+		} else {
+			bgmac_err(bgmac, "Hardware reported transmission for empty TX ring slot %d! End of ring: %d\n",
+				  ring->start, ring->end);
+		}
+
+		if (++ring->start >= BGMAC_TX_RING_SLOTS)
+			ring->start = 0;
+		freed = true;
+	}
+
+	if (freed && netif_queue_stopped(bgmac->net_dev))
+		netif_wake_queue(bgmac->net_dev);
+}
+
+static void bgmac_dma_rx_reset(struct bgmac *bgmac, struct bgmac_dma_ring *ring)
+{
+	if (!ring->mmio_base)
+		return;
+
+	bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_RX_CTL, 0);
+	if (!bgmac_wait_value(bgmac->core,
+			      ring->mmio_base + BGMAC_DMA_RX_STATUS,
+			      BGMAC_DMA_RX_STAT, BGMAC_DMA_RX_STAT_DISABLED,
+			      10000))
+		bgmac_err(bgmac, "Reset of ring 0x%X RX failed\n",
+			  ring->mmio_base);
+}
+
+static void bgmac_dma_rx_enable(struct bgmac *bgmac,
+				struct bgmac_dma_ring *ring)
+{
+	u32 ctl;
+
+	ctl = bgmac_read(bgmac, ring->mmio_base + BGMAC_DMA_RX_CTL);
+	ctl &= BGMAC_DMA_RX_ADDREXT_MASK;
+	ctl |= BGMAC_DMA_RX_ENABLE;
+	ctl |= BGMAC_DMA_RX_PARITY_DISABLE;
+	ctl |= BGMAC_DMA_RX_OVERFLOW_CONT;
+	ctl |= BGMAC_RX_FRAME_OFFSET << BGMAC_DMA_RX_FRAME_OFFSET_SHIFT;
+	bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_RX_CTL, ctl);
+}
+
+static int bgmac_dma_rx_skb_for_slot(struct bgmac *bgmac,
+				     struct bgmac_slot_info *slot)
+{
+	struct device *dma_dev = bgmac->core->dma_dev;
+	struct bgmac_rx_header *rx;
+
+	/* Alloc skb */
+	slot->skb = netdev_alloc_skb(bgmac->net_dev, BGMAC_RX_BUF_SIZE);
+	if (!slot->skb) {
+		bgmac_err(bgmac, "Allocation of skb failed!\n");
+		return -ENOMEM;
+	}
+
+	/* Poison - if everything goes fine, hardware will overwrite it */
+	rx = (struct bgmac_rx_header *)slot->skb->data;
+	rx->len = cpu_to_le16(0xdead);
+	rx->flags = cpu_to_le16(0xbeef);
+
+	/* Map skb for the DMA */
+	slot->dma_addr = dma_map_single(dma_dev, slot->skb->data,
+					BGMAC_RX_BUF_SIZE, DMA_FROM_DEVICE);
+	if (dma_mapping_error(dma_dev, slot->dma_addr)) {
+		bgmac_err(bgmac, "DMA mapping error\n");
+		return -ENOMEM;
+	}
+	if (slot->dma_addr & 0xC0000000)
+		bgmac_warn(bgmac, "DMA address using 0xC0000000 bit(s), it may need translation trick\n");
+
+	return 0;
+}
+
+static int bgmac_dma_rx_read(struct bgmac *bgmac, struct bgmac_dma_ring *ring,
+			     int weight)
+{
+	u32 end_slot;
+	int handled = 0;
+
+	end_slot = bgmac_read(bgmac, ring->mmio_base + BGMAC_DMA_RX_STATUS);
+	end_slot &= BGMAC_DMA_RX_STATDPTR;
+	end_slot /= sizeof(struct bgmac_dma_desc);
+
+	ring->end = end_slot;
+
+	while (ring->start != ring->end) {
+		struct device *dma_dev = bgmac->core->dma_dev;
+		struct bgmac_slot_info *slot = &ring->slots[ring->start];
+		struct sk_buff *skb = slot->skb;
+		struct sk_buff *new_skb;
+		struct bgmac_rx_header *rx;
+		u16 len, flags;
+
+		/* Unmap buffer to make it accessible to the CPU */
+		dma_sync_single_for_cpu(dma_dev, slot->dma_addr,
+					BGMAC_RX_BUF_SIZE, DMA_FROM_DEVICE);
+
+		/* Get info from the header */
+		rx = (struct bgmac_rx_header *)skb->data;
+		len = le16_to_cpu(rx->len);
+		flags = le16_to_cpu(rx->flags);
+
+		/* Check for poison and drop or pass the packet */
+		if (len == 0xdead && flags == 0xbeef) {
+			bgmac_err(bgmac, "Found poisoned packet at slot %d, DMA issue!\n",
+				  ring->start);
+		} else {
+			new_skb = netdev_alloc_skb(bgmac->net_dev, len);
+			if (new_skb) {
+				skb_put(new_skb, len);
+				skb_copy_from_linear_data_offset(skb, BGMAC_RX_FRAME_OFFSET,
+								 new_skb->data,
+								 len);
+				new_skb->protocol =
+					eth_type_trans(new_skb, bgmac->net_dev);
+				netif_receive_skb(new_skb);
+				handled++;
+			} else {
+				bgmac->net_dev->stats.rx_dropped++;
+				bgmac_err(bgmac, "Allocation of skb for copying packet failed!\n");
+			}
+
+			/* Poison the old skb */
+			rx->len = cpu_to_le16(0xdead);
+			rx->flags = cpu_to_le16(0xbeef);
+		}
+
+		/* Make it back accessible to the hardware */
+		dma_sync_single_for_device(dma_dev, slot->dma_addr,
+					   BGMAC_RX_BUF_SIZE, DMA_FROM_DEVICE);
+
+		if (++ring->start >= BGMAC_RX_RING_SLOTS)
+			ring->start = 0;
+
+		if (handled >= weight) /* Should never be greater */
+			break;
+	}
+
+	return handled;
+}
+
+/* Does ring support unaligned addressing? */
+static bool bgmac_dma_unaligned(struct bgmac *bgmac,
+				struct bgmac_dma_ring *ring,
+				enum bgmac_dma_ring_type ring_type)
+{
+	switch (ring_type) {
+	case BGMAC_DMA_RING_TX:
+		bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_TX_RINGLO,
+			    0xff0);
+		if (bgmac_read(bgmac, ring->mmio_base + BGMAC_DMA_TX_RINGLO))
+			return true;
+		break;
+	case BGMAC_DMA_RING_RX:
+		bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_RX_RINGLO,
+			    0xff0);
+		if (bgmac_read(bgmac, ring->mmio_base + BGMAC_DMA_RX_RINGLO))
+			return true;
+		break;
+	}
+	return false;
+}
+
+static void bgmac_dma_ring_free(struct bgmac *bgmac,
+				struct bgmac_dma_ring *ring)
+{
+	struct device *dma_dev = bgmac->core->dma_dev;
+	struct bgmac_slot_info *slot;
+	int size;
+	int i;
+
+	for (i = 0; i < ring->num_slots; i++) {
+		slot = &ring->slots[i];
+		if (slot->skb) {
+			if (slot->dma_addr)
+				dma_unmap_single(dma_dev, slot->dma_addr,
+						 slot->skb->len, DMA_TO_DEVICE);
+			dev_kfree_skb(slot->skb);
+		}
+	}
+
+	if (ring->cpu_base) {
+		/* Free ring of descriptors */
+		size = ring->num_slots * sizeof(struct bgmac_dma_desc);
+		dma_free_coherent(dma_dev, size, ring->cpu_base,
+				  ring->dma_base);
+	}
+}
+
+static void bgmac_dma_free(struct bgmac *bgmac)
+{
+	int i;
+
+	for (i = 0; i < BGMAC_MAX_TX_RINGS; i++)
+		bgmac_dma_ring_free(bgmac, &bgmac->tx_ring[i]);
+	for (i = 0; i < BGMAC_MAX_RX_RINGS; i++)
+		bgmac_dma_ring_free(bgmac, &bgmac->rx_ring[i]);
+}
+
+static int bgmac_dma_alloc(struct bgmac *bgmac)
+{
+	struct device *dma_dev = bgmac->core->dma_dev;
+	struct bgmac_dma_ring *ring;
+	static const u16 ring_base[] = { BGMAC_DMA_BASE0, BGMAC_DMA_BASE1,
+					 BGMAC_DMA_BASE2, BGMAC_DMA_BASE3, };
+	int size; /* ring size: different for Tx and Rx */
+	int err;
+	int i;
+
+	BUILD_BUG_ON(BGMAC_MAX_TX_RINGS > ARRAY_SIZE(ring_base));
+	BUILD_BUG_ON(BGMAC_MAX_RX_RINGS > ARRAY_SIZE(ring_base));
+
+	if (!(bcma_aread32(bgmac->core, BCMA_IOST) & BCMA_IOST_DMA64)) {
+		bgmac_err(bgmac, "Core does not report 64-bit DMA\n");
+		return -ENOTSUPP;
+	}
+
+	for (i = 0; i < BGMAC_MAX_TX_RINGS; i++) {
+		ring = &bgmac->tx_ring[i];
+		ring->num_slots = BGMAC_TX_RING_SLOTS;
+		ring->mmio_base = ring_base[i];
+		if (bgmac_dma_unaligned(bgmac, ring, BGMAC_DMA_RING_TX))
+			bgmac_warn(bgmac, "TX on ring 0x%X supports unaligned addressing but this feature is not implemented\n",
+				   ring->mmio_base);
+
+		/* Alloc ring of descriptors */
+		size = ring->num_slots * sizeof(struct bgmac_dma_desc);
+		ring->cpu_base = dma_zalloc_coherent(dma_dev, size,
+						     &ring->dma_base,
+						     GFP_KERNEL);
+		if (!ring->cpu_base) {
+			bgmac_err(bgmac, "Allocation of TX ring 0x%X failed\n",
+				  ring->mmio_base);
+			goto err_dma_free;
+		}
+		if (ring->dma_base & 0xC0000000)
+			bgmac_warn(bgmac, "DMA address using 0xC0000000 bit(s), it may need translation trick\n");
+
+		/* No need to alloc TX slots yet */
+	}
+
+	for (i = 0; i < BGMAC_MAX_RX_RINGS; i++) {
+		ring = &bgmac->rx_ring[i];
+		ring->num_slots = BGMAC_RX_RING_SLOTS;
+		ring->mmio_base = ring_base[i];
+		if (bgmac_dma_unaligned(bgmac, ring, BGMAC_DMA_RING_RX))
+			bgmac_warn(bgmac, "RX on ring 0x%X supports unaligned addressing but this feature is not implemented\n",
+				   ring->mmio_base);
+
+		/* Alloc ring of descriptors */
+		size = ring->num_slots * sizeof(struct bgmac_dma_desc);
+		ring->cpu_base = dma_zalloc_coherent(dma_dev, size,
+						     &ring->dma_base,
+						     GFP_KERNEL);
+		if (!ring->cpu_base) {
+			bgmac_err(bgmac, "Allocation of RX ring 0x%X failed\n",
+				  ring->mmio_base);
+			err = -ENOMEM;
+			goto err_dma_free;
+		}
+		if (ring->dma_base & 0xC0000000)
+			bgmac_warn(bgmac, "DMA address using 0xC0000000 bit(s), it may need translation trick\n");
+
+		/* Alloc RX slots */
+		for (i = 0; i < ring->num_slots; i++) {
+			err = bgmac_dma_rx_skb_for_slot(bgmac, &ring->slots[i]);
+			if (err) {
+				bgmac_err(bgmac, "Can't allocate skb for slot in RX ring\n");
+				goto err_dma_free;
+			}
+		}
+	}
+
+	return 0;
+
+err_dma_free:
+	bgmac_dma_free(bgmac);
+	return -ENOMEM;
+}
+
+static void bgmac_dma_init(struct bgmac *bgmac)
+{
+	struct bgmac_dma_ring *ring;
+	struct bgmac_dma_desc *dma_desc;
+	u32 ctl0, ctl1;
+	int i;
+
+	for (i = 0; i < BGMAC_MAX_TX_RINGS; i++) {
+		ring = &bgmac->tx_ring[i];
+
+		/* We don't implement unaligned addressing, so enable first */
+		bgmac_dma_tx_enable(bgmac, ring);
+		bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_TX_RINGLO,
+			    lower_32_bits(ring->dma_base));
+		bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_TX_RINGHI,
+			    upper_32_bits(ring->dma_base));
+
+		ring->start = 0;
+		ring->end = 0;	/* Points the slot that should *not* be read */
+	}
+
+	for (i = 0; i < BGMAC_MAX_RX_RINGS; i++) {
+		ring = &bgmac->rx_ring[i];
+
+		/* We don't implement unaligned addressing, so enable first */
+		bgmac_dma_rx_enable(bgmac, ring);
+		bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_RX_RINGLO,
+			    lower_32_bits(ring->dma_base));
+		bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_RX_RINGHI,
+			    upper_32_bits(ring->dma_base));
+
+		for (i = 0, dma_desc = ring->cpu_base; i < ring->num_slots;
+		     i++, dma_desc++) {
+			ctl0 = ctl1 = 0;
+
+			if (i == ring->num_slots - 1)
+				ctl0 |= BGMAC_DESC_CTL0_EOT;
+			ctl1 |= BGMAC_RX_BUF_SIZE & BGMAC_DESC_CTL1_LEN;
+			/* Is there any BGMAC device that requires extension? */
+			/* ctl1 |= (addrext << B43_DMA64_DCTL1_ADDREXT_SHIFT) &
+			 * B43_DMA64_DCTL1_ADDREXT_MASK;
+			 */
+
+			dma_desc->addr_low = cpu_to_le32(lower_32_bits(ring->slots[i].dma_addr));
+			dma_desc->addr_high = cpu_to_le32(upper_32_bits(ring->slots[i].dma_addr));
+			dma_desc->ctl0 = cpu_to_le32(ctl0);
+			dma_desc->ctl1 = cpu_to_le32(ctl1);
+		}
+
+		bgmac_write(bgmac, ring->mmio_base + BGMAC_DMA_RX_INDEX,
+			    ring->num_slots * sizeof(struct bgmac_dma_desc));
+
+		ring->start = 0;
+		ring->end = 0;
+	}
+}
+
+/**************************************************
+ * PHY ops
+ **************************************************/
+
+u16 bgmac_phy_read(struct bgmac *bgmac, u8 phyaddr, u8 reg)
+{
+	struct bcma_device *core;
+	u16 phy_access_addr;
+	u16 phy_ctl_addr;
+	u32 tmp;
+
+	BUILD_BUG_ON(BGMAC_PA_DATA_MASK != BCMA_GMAC_CMN_PA_DATA_MASK);
+	BUILD_BUG_ON(BGMAC_PA_ADDR_MASK != BCMA_GMAC_CMN_PA_ADDR_MASK);
+	BUILD_BUG_ON(BGMAC_PA_ADDR_SHIFT != BCMA_GMAC_CMN_PA_ADDR_SHIFT);
+	BUILD_BUG_ON(BGMAC_PA_REG_MASK != BCMA_GMAC_CMN_PA_REG_MASK);
+	BUILD_BUG_ON(BGMAC_PA_REG_SHIFT != BCMA_GMAC_CMN_PA_REG_SHIFT);
+	BUILD_BUG_ON(BGMAC_PA_WRITE != BCMA_GMAC_CMN_PA_WRITE);
+	BUILD_BUG_ON(BGMAC_PA_START != BCMA_GMAC_CMN_PA_START);
+	BUILD_BUG_ON(BGMAC_PC_EPA_MASK != BCMA_GMAC_CMN_PC_EPA_MASK);
+	BUILD_BUG_ON(BGMAC_PC_MCT_MASK != BCMA_GMAC_CMN_PC_MCT_MASK);
+	BUILD_BUG_ON(BGMAC_PC_MCT_SHIFT != BCMA_GMAC_CMN_PC_MCT_SHIFT);
+	BUILD_BUG_ON(BGMAC_PC_MTE != BCMA_GMAC_CMN_PC_MTE);
+
+	if (bgmac->core->id.id == BCMA_CORE_4706_MAC_GBIT) {
+		core = bgmac->core->bus->drv_gmac_cmn.core;
+		phy_access_addr = BCMA_GMAC_CMN_PHY_ACCESS;
+		phy_ctl_addr = BCMA_GMAC_CMN_PHY_CTL;
+	} else {
+		core = bgmac->core;
+		phy_access_addr = BGMAC_PHY_ACCESS;
+		phy_ctl_addr = BGMAC_PHY_CNTL;
+	}
+
+	tmp = bcma_read32(core, phy_ctl_addr);
+	tmp &= ~BGMAC_PC_EPA_MASK;
+	tmp |= phyaddr;
+	bcma_write32(core, phy_ctl_addr, tmp);
+
+	tmp = BGMAC_PA_START;
+	tmp |= phyaddr << BGMAC_PA_ADDR_SHIFT;
+	tmp |= reg << BGMAC_PA_REG_SHIFT;
+	bcma_write32(core, phy_access_addr, tmp);
+
+	if (!bgmac_wait_value(core, phy_access_addr, BGMAC_PA_START, 0, 1000)) {
+		bgmac_err(bgmac, "Reading PHY %d register 0x%X failed\n",
+			  phyaddr, reg);
+		return 0xffff;
+	}
+
+	return bcma_read32(core, phy_access_addr) & BGMAC_PA_DATA_MASK;
+}
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/chipphywr */
+void bgmac_phy_write(struct bgmac *bgmac, u8 phyaddr, u8 reg, u16 value)
+{
+	struct bcma_device *core;
+	u16 phy_access_addr;
+	u16 phy_ctl_addr;
+	u32 tmp;
+
+	if (bgmac->core->id.id == BCMA_CORE_4706_MAC_GBIT) {
+		core = bgmac->core->bus->drv_gmac_cmn.core;
+		phy_access_addr = BCMA_GMAC_CMN_PHY_ACCESS;
+		phy_ctl_addr = BCMA_GMAC_CMN_PHY_CTL;
+	} else {
+		core = bgmac->core;
+		phy_access_addr = BGMAC_PHY_ACCESS;
+		phy_ctl_addr = BGMAC_PHY_CNTL;
+	}
+
+	tmp = bcma_read32(core, phy_ctl_addr);
+	tmp &= ~BGMAC_PC_EPA_MASK;
+	tmp |= phyaddr;
+	bcma_write32(core, phy_ctl_addr, tmp);
+
+	bgmac_write(bgmac, BGMAC_INT_STATUS, BGMAC_IS_MDIO);
+	if (bgmac_read(bgmac, BGMAC_INT_STATUS) & BGMAC_IS_MDIO)
+		bgmac_warn(bgmac, "Error setting MDIO int\n");
+
+	tmp = BGMAC_PA_START;
+	tmp |= BGMAC_PA_WRITE;
+	tmp |= phyaddr << BGMAC_PA_ADDR_SHIFT;
+	tmp |= reg << BGMAC_PA_REG_SHIFT;
+	tmp |= value;
+	bcma_write32(core, phy_access_addr, tmp);
+
+	if (!bgmac_wait_value(core, phy_access_addr, BGMAC_PA_START, 0, 1000))
+		bgmac_err(bgmac, "Writing to PHY %d register 0x%X failed\n",
+			  phyaddr, reg);
+}
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/chipphyforce */
+static void bgmac_phy_force(struct bgmac *bgmac)
+{
+	u16 ctl;
+	u16 mask = ~(BGMAC_PHY_CTL_SPEED | BGMAC_PHY_CTL_SPEED_MSB |
+		     BGMAC_PHY_CTL_ANENAB | BGMAC_PHY_CTL_DUPLEX);
+
+	if (bgmac->phyaddr == BGMAC_PHY_NOREGS)
+		return;
+
+	if (bgmac->autoneg)
+		return;
+
+	ctl = bgmac_phy_read(bgmac, bgmac->phyaddr, BGMAC_PHY_CTL);
+	ctl &= mask;
+	if (bgmac->full_duplex)
+		ctl |= BGMAC_PHY_CTL_DUPLEX;
+	if (bgmac->speed == BGMAC_SPEED_100)
+		ctl |= BGMAC_PHY_CTL_SPEED_100;
+	else if (bgmac->speed == BGMAC_SPEED_1000)
+		ctl |= BGMAC_PHY_CTL_SPEED_1000;
+	bgmac_phy_write(bgmac, bgmac->phyaddr, BGMAC_PHY_CTL, ctl);
+}
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/chipphyadvertise */
+static void bgmac_phy_advertise(struct bgmac *bgmac)
+{
+	u16 adv;
+
+	if (bgmac->phyaddr == BGMAC_PHY_NOREGS)
+		return;
+
+	if (!bgmac->autoneg)
+		return;
+
+	/* Adv selected 10/100 speeds */
+	adv = bgmac_phy_read(bgmac, bgmac->phyaddr, BGMAC_PHY_ADV);
+	adv &= ~(BGMAC_PHY_ADV_10HALF | BGMAC_PHY_ADV_10FULL |
+		 BGMAC_PHY_ADV_100HALF | BGMAC_PHY_ADV_100FULL);
+	if (!bgmac->full_duplex && bgmac->speed & BGMAC_SPEED_10)
+		adv |= BGMAC_PHY_ADV_10HALF;
+	if (!bgmac->full_duplex && bgmac->speed & BGMAC_SPEED_100)
+		adv |= BGMAC_PHY_ADV_100HALF;
+	if (bgmac->full_duplex && bgmac->speed & BGMAC_SPEED_10)
+		adv |= BGMAC_PHY_ADV_10FULL;
+	if (bgmac->full_duplex && bgmac->speed & BGMAC_SPEED_100)
+		adv |= BGMAC_PHY_ADV_100FULL;
+	bgmac_phy_write(bgmac, bgmac->phyaddr, BGMAC_PHY_ADV, adv);
+
+	/* Adv selected 1000 speeds */
+	adv = bgmac_phy_read(bgmac, bgmac->phyaddr, BGMAC_PHY_ADV2);
+	adv &= ~(BGMAC_PHY_ADV2_1000HALF | BGMAC_PHY_ADV2_1000FULL);
+	if (!bgmac->full_duplex && bgmac->speed & BGMAC_SPEED_1000)
+		adv |= BGMAC_PHY_ADV2_1000HALF;
+	if (bgmac->full_duplex && bgmac->speed & BGMAC_SPEED_1000)
+		adv |= BGMAC_PHY_ADV2_1000FULL;
+	bgmac_phy_write(bgmac, bgmac->phyaddr, BGMAC_PHY_ADV2, adv);
+
+	/* Restart */
+	bgmac_phy_write(bgmac, bgmac->phyaddr, BGMAC_PHY_CTL,
+			bgmac_phy_read(bgmac, bgmac->phyaddr, BGMAC_PHY_CTL) |
+			BGMAC_PHY_CTL_RESTART);
+}
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/chipphyinit */
+static void bgmac_phy_init(struct bgmac *bgmac)
+{
+	struct bcma_chipinfo *ci = &bgmac->core->bus->chipinfo;
+	struct bcma_drv_cc *cc = &bgmac->core->bus->drv_cc;
+	u8 i;
+
+	if (ci->id == BCMA_CHIP_ID_BCM5356) {
+		for (i = 0; i < 5; i++) {
+			bgmac_phy_write(bgmac, i, 0x1f, 0x008b);
+			bgmac_phy_write(bgmac, i, 0x15, 0x0100);
+			bgmac_phy_write(bgmac, i, 0x1f, 0x000f);
+			bgmac_phy_write(bgmac, i, 0x12, 0x2aaa);
+			bgmac_phy_write(bgmac, i, 0x1f, 0x000b);
+		}
+	}
+	if ((ci->id == BCMA_CHIP_ID_BCM5357 && ci->pkg != 10) ||
+	    (ci->id == BCMA_CHIP_ID_BCM4749 && ci->pkg != 10) ||
+	    (ci->id == BCMA_CHIP_ID_BCM53572 && ci->pkg != 9)) {
+		bcma_chipco_chipctl_maskset(cc, 2, ~0xc0000000, 0);
+		bcma_chipco_chipctl_maskset(cc, 4, ~0x80000000, 0);
+		for (i = 0; i < 5; i++) {
+			bgmac_phy_write(bgmac, i, 0x1f, 0x000f);
+			bgmac_phy_write(bgmac, i, 0x16, 0x5284);
+			bgmac_phy_write(bgmac, i, 0x1f, 0x000b);
+			bgmac_phy_write(bgmac, i, 0x17, 0x0010);
+			bgmac_phy_write(bgmac, i, 0x1f, 0x000f);
+			bgmac_phy_write(bgmac, i, 0x16, 0x5296);
+			bgmac_phy_write(bgmac, i, 0x17, 0x1073);
+			bgmac_phy_write(bgmac, i, 0x17, 0x9073);
+			bgmac_phy_write(bgmac, i, 0x16, 0x52b6);
+			bgmac_phy_write(bgmac, i, 0x17, 0x9273);
+			bgmac_phy_write(bgmac, i, 0x1f, 0x000b);
+		}
+	}
+}
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/chipphyreset */
+static void bgmac_phy_reset(struct bgmac *bgmac)
+{
+	if (bgmac->phyaddr == BGMAC_PHY_NOREGS)
+		return;
+
+	bgmac_phy_write(bgmac, bgmac->phyaddr, BGMAC_PHY_CTL,
+			BGMAC_PHY_CTL_RESET);
+	udelay(100);
+	if (bgmac_phy_read(bgmac, bgmac->phyaddr, BGMAC_PHY_CTL) &
+	    BGMAC_PHY_CTL_RESET)
+		bgmac_err(bgmac, "PHY reset failed\n");
+	bgmac_phy_init(bgmac);
+}
+
+/**************************************************
+ * Chip ops
+ **************************************************/
+
+/* TODO: can we just drop @force? Can we don't reset MAC at all if there is
+ * nothing to change? Try if after stabilizng driver.
+ */
+static void bgmac_cmdcfg_maskset(struct bgmac *bgmac, u32 mask, u32 set,
+				 bool force)
+{
+	u32 cmdcfg = bgmac_read(bgmac, BGMAC_CMDCFG);
+	u32 new_val = (cmdcfg & mask) | set;
+
+	bgmac_set(bgmac, BGMAC_CMDCFG, BGMAC_CMDCFG_SR);
+	udelay(2);
+
+	if (new_val != cmdcfg || force)
+		bgmac_write(bgmac, BGMAC_CMDCFG, new_val);
+
+	bgmac_mask(bgmac, BGMAC_CMDCFG, ~BGMAC_CMDCFG_SR);
+	udelay(2);
+}
+
+#if 0 /* We don't use that regs yet */
+static void bgmac_chip_stats_update(struct bgmac *bgmac)
+{
+	int i;
+
+	if (bgmac->core->id.id != BCMA_CORE_4706_MAC_GBIT) {
+		for (i = 0; i < BGMAC_NUM_MIB_TX_REGS; i++)
+			bgmac->mib_tx_regs[i] =
+				bgmac_read(bgmac,
+					   BGMAC_TX_GOOD_OCTETS + (i * 4));
+		for (i = 0; i < BGMAC_NUM_MIB_RX_REGS; i++)
+			bgmac->mib_rx_regs[i] =
+				bgmac_read(bgmac,
+					   BGMAC_RX_GOOD_OCTETS + (i * 4));
+	}
+
+	/* TODO: what else? how to handle BCM4706? Specs are needed */
+}
+#endif
+
+static void bgmac_clear_mib(struct bgmac *bgmac)
+{
+	int i;
+
+	if (bgmac->core->id.id == BCMA_CORE_4706_MAC_GBIT)
+		return;
+
+	bgmac_set(bgmac, BGMAC_DEV_CTL, BGMAC_DC_MROR);
+	for (i = 0; i < BGMAC_NUM_MIB_TX_REGS; i++)
+		bgmac_read(bgmac, BGMAC_TX_GOOD_OCTETS + (i * 4));
+	for (i = 0; i < BGMAC_NUM_MIB_RX_REGS; i++)
+		bgmac_read(bgmac, BGMAC_RX_GOOD_OCTETS + (i * 4));
+}
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/gmac_speed */
+static void bgmac_speed(struct bgmac *bgmac, int speed)
+{
+	u32 mask = ~(BGMAC_CMDCFG_ES_MASK | BGMAC_CMDCFG_HD);
+	u32 set = 0;
+
+	if (speed & BGMAC_SPEED_10)
+		set |= BGMAC_CMDCFG_ES_10;
+	if (speed & BGMAC_SPEED_100)
+		set |= BGMAC_CMDCFG_ES_100;
+	if (speed & BGMAC_SPEED_1000)
+		set |= BGMAC_CMDCFG_ES_1000;
+	if (!bgmac->full_duplex)
+		set |= BGMAC_CMDCFG_HD;
+	bgmac_cmdcfg_maskset(bgmac, mask, set, true);
+}
+
+static void bgmac_miiconfig(struct bgmac *bgmac)
+{
+	u8 imode = (bgmac_read(bgmac, BGMAC_DEV_STATUS) & BGMAC_DS_MM_MASK) >>
+			BGMAC_DS_MM_SHIFT;
+	if (imode == 0 || imode == 1) {
+		if (bgmac->autoneg)
+			bgmac_speed(bgmac, BGMAC_SPEED_100);
+		else
+			bgmac_speed(bgmac, bgmac->speed);
+	}
+}
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/chipreset */
+static void bgmac_chip_reset(struct bgmac *bgmac)
+{
+	struct bcma_device *core = bgmac->core;
+	struct bcma_bus *bus = core->bus;
+	struct bcma_chipinfo *ci = &bus->chipinfo;
+	u32 flags = 0;
+	u32 iost;
+	int i;
+
+	if (bcma_core_is_enabled(core)) {
+		if (!bgmac->stats_grabbed) {
+			/* bgmac_chip_stats_update(bgmac); */
+			bgmac->stats_grabbed = true;
+		}
+
+		for (i = 0; i < BGMAC_MAX_TX_RINGS; i++)
+			bgmac_dma_tx_reset(bgmac, &bgmac->tx_ring[i]);
+
+		bgmac_cmdcfg_maskset(bgmac, ~0, BGMAC_CMDCFG_ML, false);
+		udelay(1);
+
+		for (i = 0; i < BGMAC_MAX_RX_RINGS; i++)
+			bgmac_dma_rx_reset(bgmac, &bgmac->rx_ring[i]);
+
+		/* TODO: Clear software multicast filter list */
+	}
+
+	iost = bcma_aread32(core, BCMA_IOST);
+	if ((ci->id == BCMA_CHIP_ID_BCM5357 && ci->pkg == 10) ||
+	    (ci->id == BCMA_CHIP_ID_BCM4749 && ci->pkg == 10) ||
+	    (ci->id == BCMA_CHIP_ID_BCM53572 && ci->pkg == 9))
+		iost &= ~BGMAC_BCMA_IOST_ATTACHED;
+
+	if (iost & BGMAC_BCMA_IOST_ATTACHED) {
+		flags = BGMAC_BCMA_IOCTL_SW_CLKEN;
+		if (!bgmac->has_robosw)
+			flags |= BGMAC_BCMA_IOCTL_SW_RESET;
+	}
+
+	bcma_core_enable(core, flags);
+
+	if (core->id.rev > 2) {
+		bgmac_set(bgmac, BCMA_CLKCTLST, 1 << 8);
+		bgmac_wait_value(bgmac->core, BCMA_CLKCTLST, 1 << 24, 1 << 24,
+				 1000);
+	}
+
+	if (ci->id == BCMA_CHIP_ID_BCM5357 || ci->id == BCMA_CHIP_ID_BCM4749 ||
+	    ci->id == BCMA_CHIP_ID_BCM53572) {
+		struct bcma_drv_cc *cc = &bgmac->core->bus->drv_cc;
+		u8 et_swtype = 0;
+		u8 sw_type = BGMAC_CHIPCTL_1_SW_TYPE_EPHY |
+			     BGMAC_CHIPCTL_1_IF_TYPE_RMII;
+		char buf[2];
+
+		if (nvram_getenv("et_swtype", buf, 1) > 0) {
+			if (kstrtou8(buf, 0, &et_swtype))
+				bgmac_err(bgmac, "Failed to parse et_swtype (%s)\n",
+					  buf);
+			et_swtype &= 0x0f;
+			et_swtype <<= 4;
+			sw_type = et_swtype;
+		} else if (ci->id == BCMA_CHIP_ID_BCM5357 && ci->pkg == 9) {
+			sw_type = BGMAC_CHIPCTL_1_SW_TYPE_EPHYRMII;
+		} else if (0) {
+			/* TODO */
+		}
+		bcma_chipco_chipctl_maskset(cc, 1,
+					    ~(BGMAC_CHIPCTL_1_IF_TYPE_MASK |
+					      BGMAC_CHIPCTL_1_SW_TYPE_MASK),
+					    sw_type);
+	}
+
+	if (iost & BGMAC_BCMA_IOST_ATTACHED && !bgmac->has_robosw)
+		bcma_awrite32(core, BCMA_IOCTL,
+			      bcma_aread32(core, BCMA_IOCTL) &
+			      ~BGMAC_BCMA_IOCTL_SW_RESET);
+
+	/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/gmac_reset
+	 * Specs don't say about using BGMAC_CMDCFG_SR, but in this routine
+	 * BGMAC_CMDCFG is read _after_ putting chip in a reset. So it has to
+	 * be keps until taking MAC out of the reset.
+	 */
+	bgmac_cmdcfg_maskset(bgmac,
+			     ~(BGMAC_CMDCFG_TE |
+			       BGMAC_CMDCFG_RE |
+			       BGMAC_CMDCFG_RPI |
+			       BGMAC_CMDCFG_TAI |
+			       BGMAC_CMDCFG_HD |
+			       BGMAC_CMDCFG_ML |
+			       BGMAC_CMDCFG_CFE |
+			       BGMAC_CMDCFG_RL |
+			       BGMAC_CMDCFG_RED |
+			       BGMAC_CMDCFG_PE |
+			       BGMAC_CMDCFG_TPI |
+			       BGMAC_CMDCFG_PAD_EN |
+			       BGMAC_CMDCFG_PF),
+			     BGMAC_CMDCFG_PROM |
+			     BGMAC_CMDCFG_NLC |
+			     BGMAC_CMDCFG_CFE |
+			     BGMAC_CMDCFG_SR,
+			     false);
+
+	bgmac_clear_mib(bgmac);
+	if (core->id.id == BCMA_CORE_4706_MAC_GBIT)
+		bcma_maskset32(bgmac->cmn, BCMA_GMAC_CMN_PHY_CTL, ~0,
+			       BCMA_GMAC_CMN_PC_MTE);
+	else
+		bgmac_set(bgmac, BGMAC_PHY_CNTL, BGMAC_PC_MTE);
+	bgmac_miiconfig(bgmac);
+	bgmac_phy_init(bgmac);
+
+	bgmac->int_status = 0;
+}
+
+static void bgmac_chip_intrs_on(struct bgmac *bgmac)
+{
+	bgmac_write(bgmac, BGMAC_INT_MASK, bgmac->int_mask);
+}
+
+static void bgmac_chip_intrs_off(struct bgmac *bgmac)
+{
+	bgmac_write(bgmac, BGMAC_INT_MASK, 0);
+}
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/gmac_enable */
+static void bgmac_enable(struct bgmac *bgmac)
+{
+	struct bcma_chipinfo *ci = &bgmac->core->bus->chipinfo;
+	u32 cmdcfg;
+	u32 mode;
+	u32 rxq_ctl;
+	u32 fl_ctl;
+	u16 bp_clk;
+	u8 mdp;
+
+	cmdcfg = bgmac_read(bgmac, BGMAC_CMDCFG);
+	bgmac_cmdcfg_maskset(bgmac, ~(BGMAC_CMDCFG_TE | BGMAC_CMDCFG_RE),
+			     BGMAC_CMDCFG_SR, true);
+	udelay(2);
+	cmdcfg |= BGMAC_CMDCFG_TE | BGMAC_CMDCFG_RE;
+	bgmac_write(bgmac, BGMAC_CMDCFG, cmdcfg);
+
+	mode = (bgmac_read(bgmac, BGMAC_DEV_STATUS) & BGMAC_DS_MM_MASK) >>
+		BGMAC_DS_MM_SHIFT;
+	if (ci->id != BCMA_CHIP_ID_BCM47162 || mode != 0)
+		bgmac_set(bgmac, BCMA_CLKCTLST, BCMA_CLKCTLST_FORCEHT);
+	if (ci->id == BCMA_CHIP_ID_BCM47162 && mode == 2)
+		bcma_chipco_chipctl_maskset(&bgmac->core->bus->drv_cc, 1, ~0,
+					    BGMAC_CHIPCTL_1_RXC_DLL_BYPASS);
+
+	switch (ci->id) {
+	case BCMA_CHIP_ID_BCM5357:
+	case BCMA_CHIP_ID_BCM4749:
+	case BCMA_CHIP_ID_BCM53572:
+	case BCMA_CHIP_ID_BCM4716:
+	case BCMA_CHIP_ID_BCM47162:
+		fl_ctl = 0x03cb04cb;
+		if (ci->id == BCMA_CHIP_ID_BCM5357 ||
+		    ci->id == BCMA_CHIP_ID_BCM4749 ||
+		    ci->id == BCMA_CHIP_ID_BCM53572)
+			fl_ctl = 0x2300e1;
+		bgmac_write(bgmac, BGMAC_FLOW_CTL_THRESH, fl_ctl);
+		bgmac_write(bgmac, BGMAC_PAUSE_CTL, 0x27fff);
+		break;
+	}
+
+	rxq_ctl = bgmac_read(bgmac, BGMAC_RXQ_CTL);
+	rxq_ctl &= ~BGMAC_RXQ_CTL_MDP_MASK;
+	bp_clk = bcma_pmu_get_bus_clock(&bgmac->core->bus->drv_cc) / 1000000;
+	mdp = (bp_clk * 128 / 1000) - 3;
+	rxq_ctl |= (mdp << BGMAC_RXQ_CTL_MDP_SHIFT);
+	bgmac_write(bgmac, BGMAC_RXQ_CTL, rxq_ctl);
+}
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/chipinit */
+static void bgmac_chip_init(struct bgmac *bgmac, bool full_init)
+{
+	struct bgmac_dma_ring *ring;
+	u8 *mac = bgmac->net_dev->dev_addr;
+	u32 tmp;
+	int i;
+
+	/* 1 interrupt per received frame */
+	bgmac_write(bgmac, BGMAC_INT_RECV_LAZY, 1 << BGMAC_IRL_FC_SHIFT);
+
+	/* Enable 802.3x tx flow control (honor received PAUSE frames) */
+	bgmac_cmdcfg_maskset(bgmac, ~BGMAC_CMDCFG_RPI, 0, true);
+
+	if (bgmac->net_dev->flags & IFF_PROMISC)
+		bgmac_cmdcfg_maskset(bgmac, ~0, BGMAC_CMDCFG_PROM, false);
+	else
+		bgmac_cmdcfg_maskset(bgmac, ~BGMAC_CMDCFG_PROM, 0, false);
+
+	/* Set MAC addr */
+	tmp = (mac[0] << 24) | (mac[1] << 16) | (mac[2] << 8) | mac[3];
+	bgmac_write(bgmac, BGMAC_MACADDR_HIGH, tmp);
+	tmp = (mac[4] << 8) | mac[5];
+	bgmac_write(bgmac, BGMAC_MACADDR_LOW, tmp);
+
+	if (bgmac->loopback)
+		bgmac_cmdcfg_maskset(bgmac, ~0, BGMAC_CMDCFG_ML, true);
+	else
+		bgmac_cmdcfg_maskset(bgmac, ~BGMAC_CMDCFG_ML, 0, true);
+
+	bgmac_write(bgmac, BGMAC_RXMAX_LENGTH, 32 + ETHER_MAX_LEN);
+
+	if (!bgmac->autoneg) {
+		bgmac_speed(bgmac, bgmac->speed);
+		bgmac_phy_force(bgmac);
+	} else if (bgmac->speed) { /* if there is anything to adv */
+		bgmac_phy_advertise(bgmac);
+	}
+
+	if (full_init) {
+		bgmac_dma_init(bgmac);
+		if (1) /* FIXME: is there any case we don't want IRQs? */
+			bgmac_chip_intrs_on(bgmac);
+	} else {
+		for (i = 0; i < BGMAC_MAX_RX_RINGS; i++) {
+			ring = &bgmac->rx_ring[i];
+			bgmac_dma_rx_enable(bgmac, ring);
+		}
+	}
+
+	bgmac_enable(bgmac);
+}
+
+static irqreturn_t bgmac_interrupt(int irq, void *dev_id)
+{
+	struct bgmac *bgmac = netdev_priv(dev_id);
+
+	u32 int_status = bgmac_read(bgmac, BGMAC_INT_STATUS);
+	int_status &= bgmac->int_mask;
+
+	if (!int_status)
+		return IRQ_NONE;
+
+	/* Ack */
+	bgmac_write(bgmac, BGMAC_INT_STATUS, int_status);
+
+	/* Disable new interrupts until handling existing ones */
+	bgmac_chip_intrs_off(bgmac);
+
+	bgmac->int_status = int_status;
+
+	napi_schedule(&bgmac->napi);
+
+	return IRQ_HANDLED;
+}
+
+static int bgmac_poll(struct napi_struct *napi, int weight)
+{
+	struct bgmac *bgmac = container_of(napi, struct bgmac, napi);
+	struct bgmac_dma_ring *ring;
+	int handled = 0;
+
+	if (bgmac->int_status & BGMAC_IS_TX0) {
+		ring = &bgmac->tx_ring[0];
+		bgmac_dma_tx_free(bgmac, ring);
+		bgmac->int_status &= ~BGMAC_IS_TX0;
+	}
+
+	if (bgmac->int_status & BGMAC_IS_RX) {
+		ring = &bgmac->rx_ring[0];
+		handled += bgmac_dma_rx_read(bgmac, ring, weight);
+		bgmac->int_status &= ~BGMAC_IS_RX;
+	}
+
+	if (bgmac->int_status) {
+		bgmac_err(bgmac, "Unknown IRQs: 0x%08X\n", bgmac->int_status);
+		bgmac->int_status = 0;
+	}
+
+	if (handled < weight)
+		napi_complete(napi);
+
+	bgmac_chip_intrs_on(bgmac);
+
+	return handled;
+}
+
+/**************************************************
+ * net_device_ops
+ **************************************************/
+
+static int bgmac_open(struct net_device *net_dev)
+{
+	struct bgmac *bgmac = netdev_priv(net_dev);
+	int err = 0;
+
+	bgmac_chip_reset(bgmac);
+	/* Specs say about reclaiming rings here, but we do that in DMA init */
+	bgmac_chip_init(bgmac, true);
+
+	err = request_irq(bgmac->core->irq, bgmac_interrupt, IRQF_SHARED,
+			  KBUILD_MODNAME, net_dev);
+	if (err < 0) {
+		bgmac_err(bgmac, "IRQ request error: %d!\n", err);
+		goto err_out;
+	}
+	napi_enable(&bgmac->napi);
+
+	netif_carrier_on(net_dev);
+
+err_out:
+	return err;
+}
+
+static int bgmac_stop(struct net_device *net_dev)
+{
+	struct bgmac *bgmac = netdev_priv(net_dev);
+
+	netif_carrier_off(net_dev);
+
+	napi_disable(&bgmac->napi);
+	bgmac_chip_intrs_off(bgmac);
+	free_irq(bgmac->core->irq, net_dev);
+
+	bgmac_chip_reset(bgmac);
+
+	return 0;
+}
+
+static netdev_tx_t bgmac_start_xmit(struct sk_buff *skb,
+				    struct net_device *net_dev)
+{
+	struct bgmac *bgmac = netdev_priv(net_dev);
+	struct bgmac_dma_ring *ring;
+
+	/* No QOS support yet */
+	ring = &bgmac->tx_ring[0];
+	return bgmac_dma_tx_add(bgmac, ring, skb);
+}
+
+static int bgmac_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
+{
+	struct bgmac *bgmac = netdev_priv(net_dev);
+	struct mii_ioctl_data *data = if_mii(ifr);
+
+	switch (cmd) {
+	case SIOCGMIIPHY:
+		data->phy_id = bgmac->phyaddr;
+		/* fallthru */
+	case SIOCGMIIREG:
+		if (!netif_running(net_dev))
+			return -EAGAIN;
+		data->val_out = bgmac_phy_read(bgmac, bgmac->phyaddr,
+					       data->reg_num & 0x1f);
+		return 0;
+	case SIOCSMIIREG:
+		if (!netif_running(net_dev))
+			return -EAGAIN;
+		bgmac_phy_write(bgmac, bgmac->phyaddr, data->reg_num & 0x1f,
+				data->val_in);
+		return 0;
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+static const struct net_device_ops bgmac_netdev_ops = {
+	.ndo_open		= bgmac_open,
+	.ndo_stop		= bgmac_stop,
+	.ndo_start_xmit		= bgmac_start_xmit,
+	.ndo_set_mac_address	= eth_mac_addr, /* generic, sets dev_addr */
+	.ndo_do_ioctl           = bgmac_ioctl,
+};
+
+/**************************************************
+ * ethtool_ops
+ **************************************************/
+
+static int bgmac_get_settings(struct net_device *net_dev,
+			      struct ethtool_cmd *cmd)
+{
+	struct bgmac *bgmac = netdev_priv(net_dev);
+
+	cmd->supported = SUPPORTED_10baseT_Half |
+			 SUPPORTED_10baseT_Full |
+			 SUPPORTED_100baseT_Half |
+			 SUPPORTED_100baseT_Full |
+			 SUPPORTED_1000baseT_Half |
+			 SUPPORTED_1000baseT_Full |
+			 SUPPORTED_Autoneg;
+
+	if (bgmac->autoneg) {
+		WARN_ON(cmd->advertising);
+		if (bgmac->full_duplex) {
+			if (bgmac->speed & BGMAC_SPEED_10)
+				cmd->advertising |= ADVERTISED_10baseT_Full;
+			if (bgmac->speed & BGMAC_SPEED_100)
+				cmd->advertising |= ADVERTISED_100baseT_Full;
+			if (bgmac->speed & BGMAC_SPEED_1000)
+				cmd->advertising |= ADVERTISED_1000baseT_Full;
+		} else {
+			if (bgmac->speed & BGMAC_SPEED_10)
+				cmd->advertising |= ADVERTISED_10baseT_Half;
+			if (bgmac->speed & BGMAC_SPEED_100)
+				cmd->advertising |= ADVERTISED_100baseT_Half;
+			if (bgmac->speed & BGMAC_SPEED_1000)
+				cmd->advertising |= ADVERTISED_1000baseT_Half;
+		}
+	} else {
+		switch (bgmac->speed) {
+		case BGMAC_SPEED_10:
+			ethtool_cmd_speed_set(cmd, SPEED_10);
+			break;
+		case BGMAC_SPEED_100:
+			ethtool_cmd_speed_set(cmd, SPEED_100);
+			break;
+		case BGMAC_SPEED_1000:
+			ethtool_cmd_speed_set(cmd, SPEED_1000);
+			break;
+		}
+	}
+
+	cmd->duplex = bgmac->full_duplex ? DUPLEX_FULL : DUPLEX_HALF;
+
+	cmd->autoneg = bgmac->autoneg;
+
+	return 0;
+}
+
+#if 0
+static int bgmac_set_settings(struct net_device *net_dev,
+			      struct ethtool_cmd *cmd)
+{
+	struct bgmac *bgmac = netdev_priv(net_dev);
+
+	return -1;
+}
+#endif
+
+static void bgmac_get_drvinfo(struct net_device *net_dev,
+			      struct ethtool_drvinfo *info)
+{
+	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
+	strlcpy(info->bus_info, "BCMA", sizeof(info->bus_info));
+}
+
+static const struct ethtool_ops bgmac_ethtool_ops = {
+	.get_settings		= bgmac_get_settings,
+	.get_drvinfo		= bgmac_get_drvinfo,
+};
+
+/**************************************************
+ * BCMA bus ops
+ **************************************************/
+
+/* http://bcm-v4.sipsolutions.net/mac-gbit/gmac/chipattach */
+static int bgmac_probe(struct bcma_device *core)
+{
+	struct net_device *net_dev;
+	struct bgmac *bgmac;
+	struct ssb_sprom *sprom = &core->bus->sprom;
+	u8 *mac = core->core_unit ? sprom->et1mac : sprom->et0mac;
+	int err;
+
+	/* We don't support 2nd, 3rd, ... units, SPROM has to be adjusted */
+	if (core->core_unit > 1) {
+		pr_err("Unsupported core_unit %d\n", core->core_unit);
+		return -ENOTSUPP;
+	}
+
+	/* Allocation and references */
+	net_dev = alloc_etherdev(sizeof(*bgmac));
+	if (!net_dev)
+		return -ENOMEM;
+	net_dev->netdev_ops = &bgmac_netdev_ops;
+	net_dev->irq = core->irq;
+	SET_ETHTOOL_OPS(net_dev, &bgmac_ethtool_ops);
+	bgmac = netdev_priv(net_dev);
+	bgmac->net_dev = net_dev;
+	bgmac->core = core;
+	bcma_set_drvdata(core, bgmac);
+
+	/* Defaults */
+	bgmac->autoneg = true;
+	bgmac->full_duplex = true;
+	bgmac->speed = BGMAC_SPEED_10 | BGMAC_SPEED_100 | BGMAC_SPEED_1000;
+	memcpy(bgmac->net_dev->dev_addr, mac, ETH_ALEN);
+
+	/* On BCM4706 we need common core to access PHY */
+	if (core->id.id == BCMA_CORE_4706_MAC_GBIT &&
+	    !core->bus->drv_gmac_cmn.core) {
+		bgmac_err(bgmac, "GMAC CMN core not found (required for BCM4706)\n");
+		err = -ENODEV;
+		goto err_netdev_free;
+	}
+	bgmac->cmn = core->bus->drv_gmac_cmn.core;
+
+	bgmac->phyaddr = core->core_unit ? sprom->et1phyaddr :
+			 sprom->et0phyaddr;
+	bgmac->phyaddr &= BGMAC_PHY_MASK;
+	if (bgmac->phyaddr == BGMAC_PHY_MASK) {
+		bgmac_err(bgmac, "No PHY found\n");
+		err = -ENODEV;
+		goto err_netdev_free;
+	}
+	bgmac_info(bgmac, "Found PHY addr: %d%s\n", bgmac->phyaddr,
+		   bgmac->phyaddr == BGMAC_PHY_NOREGS ? " (NOREGS)" : "");
+
+	if (core->bus->hosttype == BCMA_HOSTTYPE_PCI) {
+		bgmac_err(bgmac, "PCI setup not implemented\n");
+		err = -ENOTSUPP;
+		goto err_netdev_free;
+	}
+
+	bgmac_chip_reset(bgmac);
+
+	err = bgmac_dma_alloc(bgmac);
+	if (err) {
+		bgmac_err(bgmac, "Unable to alloc memory for DMA\n");
+		goto err_netdev_free;
+	}
+
+	bgmac->int_mask = BGMAC_IS_ERRMASK | BGMAC_IS_RX | BGMAC_IS_TX_MASK;
+	if (nvram_getenv("et0_no_txint", NULL, 0) == 0)
+		bgmac->int_mask &= ~BGMAC_IS_TX_MASK;
+
+	/* TODO: reset the external phy. Specs are needed */
+	bgmac_phy_reset(bgmac);
+
+	bgmac->has_robosw = !!(core->bus->sprom.boardflags_lo &
+			       BGMAC_BFL_ENETROBO);
+	if (bgmac->has_robosw)
+		bgmac_warn(bgmac, "Support for Roboswitch not implemented\n");
+
+	if (core->bus->sprom.boardflags_lo & BGMAC_BFL_ENETADM)
+		bgmac_warn(bgmac, "Support for ADMtek ethernet switch not implemented\n");
+
+	err = register_netdev(bgmac->net_dev);
+	if (err) {
+		bgmac_err(bgmac, "Cannot register net device\n");
+		err = -ENOTSUPP;
+		goto err_dma_free;
+	}
+
+	netif_carrier_off(net_dev);
+
+	netif_napi_add(net_dev, &bgmac->napi, bgmac_poll, BGMAC_WEIGHT);
+
+	return 0;
+
+err_dma_free:
+	bgmac_dma_free(bgmac);
+
+err_netdev_free:
+	bcma_set_drvdata(core, NULL);
+	free_netdev(net_dev);
+
+	return err;
+}
+
+static void bgmac_remove(struct bcma_device *core)
+{
+	struct bgmac *bgmac = bcma_get_drvdata(core);
+
+	netif_napi_del(&bgmac->napi);
+	unregister_netdev(bgmac->net_dev);
+	bgmac_dma_free(bgmac);
+	bcma_set_drvdata(core, NULL);
+	free_netdev(bgmac->net_dev);
+}
+
+static struct bcma_driver bgmac_bcma_driver = {
+	.name		= KBUILD_MODNAME,
+	.id_table	= bgmac_bcma_tbl,
+	.probe		= bgmac_probe,
+	.remove		= bgmac_remove,
+};
+
+static int __init bgmac_init(void)
+{
+	int err;
+
+	err = bcma_driver_register(&bgmac_bcma_driver);
+	if (err)
+		return err;
+	pr_info("Broadcom 47xx GBit MAC driver loaded\n");
+
+	return 0;
+}
+
+static void __exit bgmac_exit(void)
+{
+	bcma_driver_unregister(&bgmac_bcma_driver);
+}
+
+module_init(bgmac_init)
+module_exit(bgmac_exit)
+
+MODULE_AUTHOR("Rafał Miłecki");
+MODULE_LICENSE("GPL");
diff --git a/drivers/net/ethernet/broadcom/bgmac.h b/drivers/net/ethernet/broadcom/bgmac.h
new file mode 100644
index 0000000..1299470
--- /dev/null
+++ b/drivers/net/ethernet/broadcom/bgmac.h
@@ -0,0 +1,456 @@
+#ifndef _BGMAC_H
+#define _BGMAC_H
+
+#define pr_fmt(fmt)		KBUILD_MODNAME ": " fmt
+
+#define bgmac_err(bgmac, fmt, ...) \
+	dev_err(&(bgmac)->core->dev, fmt, ##__VA_ARGS__)
+#define bgmac_warn(bgmac, fmt, ...) \
+	dev_warn(&(bgmac)->core->dev, fmt,  ##__VA_ARGS__)
+#define bgmac_info(bgmac, fmt, ...) \
+	dev_info(&(bgmac)->core->dev, fmt,  ##__VA_ARGS__)
+#define bgmac_dbg(bgmac, fmt, ...) \
+	dev_dbg(&(bgmac)->core->dev, fmt, ##__VA_ARGS__)
+
+#include <linux/bcma/bcma.h>
+#include <linux/netdevice.h>
+
+#define BGMAC_DEV_CTL				0x000
+#define  BGMAC_DC_TSM				0x00000002
+#define  BGMAC_DC_CFCO				0x00000004
+#define  BGMAC_DC_RLSS				0x00000008
+#define  BGMAC_DC_MROR				0x00000010
+#define  BGMAC_DC_FCM_MASK			0x00000060
+#define  BGMAC_DC_FCM_SHIFT			5
+#define  BGMAC_DC_NAE				0x00000080
+#define  BGMAC_DC_TF				0x00000100
+#define  BGMAC_DC_RDS_MASK			0x00030000
+#define  BGMAC_DC_RDS_SHIFT			16
+#define  BGMAC_DC_TDS_MASK			0x000c0000
+#define  BGMAC_DC_TDS_SHIFT			18
+#define BGMAC_DEV_STATUS			0x004		/* Configuration of the interface */
+#define  BGMAC_DS_RBF				0x00000001
+#define  BGMAC_DS_RDF				0x00000002
+#define  BGMAC_DS_RIF				0x00000004
+#define  BGMAC_DS_TBF				0x00000008
+#define  BGMAC_DS_TDF				0x00000010
+#define  BGMAC_DS_TIF				0x00000020
+#define  BGMAC_DS_PO				0x00000040
+#define  BGMAC_DS_MM_MASK			0x00000300	/* Mode of the interface */
+#define  BGMAC_DS_MM_SHIFT			8
+#define BGMAC_BIST_STATUS			0x00c
+#define BGMAC_INT_STATUS			0x020		/* Interrupt status */
+#define  BGMAC_IS_MRO				0x00000001
+#define  BGMAC_IS_MTO				0x00000002
+#define  BGMAC_IS_TFD				0x00000004
+#define  BGMAC_IS_LS				0x00000008
+#define  BGMAC_IS_MDIO				0x00000010
+#define  BGMAC_IS_MR				0x00000020
+#define  BGMAC_IS_MT				0x00000040
+#define  BGMAC_IS_TO				0x00000080
+#define  BGMAC_IS_DESC_ERR			0x00000400	/* Descriptor error */
+#define  BGMAC_IS_DATA_ERR			0x00000800	/* Data error */
+#define  BGMAC_IS_DESC_PROT_ERR			0x00001000	/* Descriptor protocol error */
+#define  BGMAC_IS_RX_DESC_UNDERF		0x00002000	/* Receive descriptor underflow */
+#define  BGMAC_IS_RX_F_OVERF			0x00004000	/* Receive FIFO overflow */
+#define  BGMAC_IS_TX_F_UNDERF			0x00008000	/* Transmit FIFO underflow */
+#define  BGMAC_IS_RX				0x00010000	/* Interrupt for RX queue 0 */
+#define  BGMAC_IS_TX0				0x01000000	/* Interrupt for TX queue 0 */
+#define  BGMAC_IS_TX1				0x02000000	/* Interrupt for TX queue 1 */
+#define  BGMAC_IS_TX2				0x04000000	/* Interrupt for TX queue 2 */
+#define  BGMAC_IS_TX3				0x08000000	/* Interrupt for TX queue 3 */
+#define  BGMAC_IS_TX_MASK			0x0f000000
+#define  BGMAC_IS_INTMASK			0x0f01fcff
+#define  BGMAC_IS_ERRMASK			0x0000fc00
+#define BGMAC_INT_MASK				0x024		/* Interrupt mask */
+#define BGMAC_GP_TIMER				0x028
+#define BGMAC_INT_RECV_LAZY			0x100
+#define  BGMAC_IRL_TO_MASK			0x00ffffff
+#define  BGMAC_IRL_FC_MASK			0xff000000
+#define  BGMAC_IRL_FC_SHIFT			24		/* Shift the number of interrupts triggered per received frame */
+#define BGMAC_FLOW_CTL_THRESH			0x104		/* Flow control thresholds */
+#define BGMAC_WRRTHRESH				0x108
+#define BGMAC_GMAC_IDLE_CNT_THRESH		0x10c
+#define BGMAC_PHY_ACCESS			0x180		/* PHY access address */
+#define  BGMAC_PA_DATA_MASK			0x0000ffff
+#define  BGMAC_PA_ADDR_MASK			0x001f0000
+#define  BGMAC_PA_ADDR_SHIFT			16
+#define  BGMAC_PA_REG_MASK			0x1f000000
+#define  BGMAC_PA_REG_SHIFT			24
+#define  BGMAC_PA_WRITE				0x20000000
+#define  BGMAC_PA_START				0x40000000
+#define BGMAC_PHY_CNTL				0x188		/* PHY control address */
+#define  BGMAC_PC_EPA_MASK			0x0000001f
+#define  BGMAC_PC_MCT_MASK			0x007f0000
+#define  BGMAC_PC_MCT_SHIFT			16
+#define  BGMAC_PC_MTE				0x00800000
+#define BGMAC_TXQ_CTL				0x18c
+#define  BGMAC_TXQ_CTL_DBT_MASK			0x00000fff
+#define  BGMAC_TXQ_CTL_DBT_SHIFT		0
+#define BGMAC_RXQ_CTL				0x190
+#define  BGMAC_RXQ_CTL_DBT_MASK			0x00000fff
+#define  BGMAC_RXQ_CTL_DBT_SHIFT		0
+#define  BGMAC_RXQ_CTL_PTE			0x00001000
+#define  BGMAC_RXQ_CTL_MDP_MASK			0x3f000000
+#define  BGMAC_RXQ_CTL_MDP_SHIFT		24
+#define BGMAC_GPIO_SELECT			0x194
+#define BGMAC_GPIO_OUTPUT_EN			0x198
+/* For 0x1e0 see BCMA_CLKCTLST */
+#define BGMAC_HW_WAR				0x1e4
+#define BGMAC_PWR_CTL				0x1e8
+#define BGMAC_DMA_BASE0				0x200		/* Tx and Rx controller */
+#define BGMAC_DMA_BASE1				0x240		/* Tx controller only */
+#define BGMAC_DMA_BASE2				0x280		/* Tx controller only */
+#define BGMAC_DMA_BASE3				0x2C0		/* Tx controller only */
+#define BGMAC_TX_GOOD_OCTETS			0x300
+#define BGMAC_TX_GOOD_OCTETS_HIGH		0x304
+#define BGMAC_TX_GOOD_PKTS			0x308
+#define BGMAC_TX_OCTETS				0x30c
+#define BGMAC_TX_OCTETS_HIGH			0x310
+#define BGMAC_TX_PKTS				0x314
+#define BGMAC_TX_BROADCAST_PKTS			0x318
+#define BGMAC_TX_MULTICAST_PKTS			0x31c
+#define BGMAC_TX_LEN_64				0x320
+#define BGMAC_TX_LEN_65_TO_127			0x324
+#define BGMAC_TX_LEN_128_TO_255			0x328
+#define BGMAC_TX_LEN_256_TO_511			0x32c
+#define BGMAC_TX_LEN_512_TO_1023		0x330
+#define BGMAC_TX_LEN_1024_TO_1522		0x334
+#define BGMAC_TX_LEN_1523_TO_2047		0x338
+#define BGMAC_TX_LEN_2048_TO_4095		0x33c
+#define BGMAC_TX_LEN_4095_TO_8191		0x340
+#define BGMAC_TX_LEN_8192_TO_MAX		0x344
+#define BGMAC_TX_JABBER_PKTS			0x348		/* Error */
+#define BGMAC_TX_OVERSIZE_PKTS			0x34c		/* Error */
+#define BGMAC_TX_FRAGMENT_PKTS			0x350
+#define BGMAC_TX_UNDERRUNS			0x354		/* Error */
+#define BGMAC_TX_TOTAL_COLS			0x358
+#define BGMAC_TX_SINGLE_COLS			0x35c
+#define BGMAC_TX_MULTIPLE_COLS			0x360
+#define BGMAC_TX_EXCESSIVE_COLS			0x364		/* Error */
+#define BGMAC_TX_LATE_COLS			0x368		/* Error */
+#define BGMAC_TX_DEFERED			0x36c
+#define BGMAC_TX_CARRIER_LOST			0x370
+#define BGMAC_TX_PAUSE_PKTS			0x374
+#define BGMAC_TX_UNI_PKTS			0x378
+#define BGMAC_TX_Q0_PKTS			0x37c
+#define BGMAC_TX_Q0_OCTETS			0x380
+#define BGMAC_TX_Q0_OCTETS_HIGH			0x384
+#define BGMAC_TX_Q1_PKTS			0x388
+#define BGMAC_TX_Q1_OCTETS			0x38c
+#define BGMAC_TX_Q1_OCTETS_HIGH			0x390
+#define BGMAC_TX_Q2_PKTS			0x394
+#define BGMAC_TX_Q2_OCTETS			0x398
+#define BGMAC_TX_Q2_OCTETS_HIGH			0x39c
+#define BGMAC_TX_Q3_PKTS			0x3a0
+#define BGMAC_TX_Q3_OCTETS			0x3a4
+#define BGMAC_TX_Q3_OCTETS_HIGH			0x3a8
+#define BGMAC_RX_GOOD_OCTETS			0x3b0
+#define BGMAC_RX_GOOD_OCTETS_HIGH		0x3b4
+#define BGMAC_RX_GOOD_PKTS			0x3b8
+#define BGMAC_RX_OCTETS				0x3bc
+#define BGMAC_RX_OCTETS_HIGH			0x3c0
+#define BGMAC_RX_PKTS				0x3c4
+#define BGMAC_RX_BROADCAST_PKTS			0x3c8
+#define BGMAC_RX_MULTICAST_PKTS			0x3cc
+#define BGMAC_RX_LEN_64				0x3d0
+#define BGMAC_RX_LEN_65_TO_127			0x3d4
+#define BGMAC_RX_LEN_128_TO_255			0x3d8
+#define BGMAC_RX_LEN_256_TO_511			0x3dc
+#define BGMAC_RX_LEN_512_TO_1023		0x3e0
+#define BGMAC_RX_LEN_1024_TO_1522		0x3e4
+#define BGMAC_RX_LEN_1523_TO_2047		0x3e8
+#define BGMAC_RX_LEN_2048_TO_4095		0x3ec
+#define BGMAC_RX_LEN_4095_TO_8191		0x3f0
+#define BGMAC_RX_LEN_8192_TO_MAX		0x3f4
+#define BGMAC_RX_JABBER_PKTS			0x3f8		/* Error */
+#define BGMAC_RX_OVERSIZE_PKTS			0x3fc		/* Error */
+#define BGMAC_RX_FRAGMENT_PKTS			0x400
+#define BGMAC_RX_MISSED_PKTS			0x404		/* Error */
+#define BGMAC_RX_CRC_ALIGN_ERRS			0x408		/* Error */
+#define BGMAC_RX_UNDERSIZE			0x40c		/* Error */
+#define BGMAC_RX_CRC_ERRS			0x410		/* Error */
+#define BGMAC_RX_ALIGN_ERRS			0x414		/* Error */
+#define BGMAC_RX_SYMBOL_ERRS			0x418		/* Error */
+#define BGMAC_RX_PAUSE_PKTS			0x41c
+#define BGMAC_RX_NONPAUSE_PKTS			0x420
+#define BGMAC_RX_SACHANGES			0x424
+#define BGMAC_RX_UNI_PKTS			0x428
+#define BGMAC_UNIMAC_VERSION			0x800
+#define BGMAC_HDBKP_CTL				0x804
+#define BGMAC_CMDCFG				0x808		/* Configuration */
+#define  BGMAC_CMDCFG_TE			0x00000001	/* Set to activate TX */
+#define  BGMAC_CMDCFG_RE			0x00000002	/* Set to activate RX */
+#define  BGMAC_CMDCFG_ES_MASK			0x0000000c	/* Ethernet speed see gmac_speed */
+#define   BGMAC_CMDCFG_ES_10			0x00000000
+#define   BGMAC_CMDCFG_ES_100			0x00000004
+#define   BGMAC_CMDCFG_ES_1000			0x00000008
+#define  BGMAC_CMDCFG_PROM			0x00000010	/* Set to activate promiscuous mode */
+#define  BGMAC_CMDCFG_PAD_EN			0x00000020
+#define  BGMAC_CMDCFG_CF			0x00000040
+#define  BGMAC_CMDCFG_PF			0x00000080
+#define  BGMAC_CMDCFG_RPI			0x00000100	/* Unset to enable 802.3x tx flow control */
+#define  BGMAC_CMDCFG_TAI			0x00000200
+#define  BGMAC_CMDCFG_HD			0x00000400	/* Set if in half duplex mode */
+#define  BGMAC_CMDCFG_HD_SHIFT			10
+#define  BGMAC_CMDCFG_SR			0x00000800	/* Set to reset mode */
+#define  BGMAC_CMDCFG_ML			0x00008000	/* Set to activate mac loopback mode */
+#define  BGMAC_CMDCFG_AE			0x00400000
+#define  BGMAC_CMDCFG_CFE			0x00800000
+#define  BGMAC_CMDCFG_NLC			0x01000000
+#define  BGMAC_CMDCFG_RL			0x02000000
+#define  BGMAC_CMDCFG_RED			0x04000000
+#define  BGMAC_CMDCFG_PE			0x08000000
+#define  BGMAC_CMDCFG_TPI			0x10000000
+#define  BGMAC_CMDCFG_AT			0x20000000
+#define BGMAC_MACADDR_HIGH			0x80c		/* High 4 octets of own mac address */
+#define BGMAC_MACADDR_LOW			0x810		/* Low 2 octets of own mac address */
+#define BGMAC_RXMAX_LENGTH			0x814		/* Max receive frame length with vlan tag */
+#define BGMAC_PAUSEQUANTA			0x818
+#define BGMAC_MAC_MODE				0x844
+#define BGMAC_OUTERTAG				0x848
+#define BGMAC_INNERTAG				0x84c
+#define BGMAC_TXIPG				0x85c
+#define BGMAC_PAUSE_CTL				0xb30
+#define BGMAC_TX_FLUSH				0xb34
+#define BGMAC_RX_STATUS				0xb38
+#define BGMAC_TX_STATUS				0xb3c
+
+#define BGMAC_PHY_CTL				0x00
+#define  BGMAC_PHY_CTL_SPEED_MSB		0x0040
+#define  BGMAC_PHY_CTL_DUPLEX			0x0100		/* duplex mode */
+#define  BGMAC_PHY_CTL_RESTART			0x0200		/* restart autonegotiation */
+#define  BGMAC_PHY_CTL_ANENAB			0x1000		/* enable autonegotiation */
+#define  BGMAC_PHY_CTL_SPEED			0x2000
+#define  BGMAC_PHY_CTL_LOOP			0x4000		/* loopback */
+#define  BGMAC_PHY_CTL_RESET			0x8000		/* reset */
+/* Helpers */
+#define  BGMAC_PHY_CTL_SPEED_10			0
+#define  BGMAC_PHY_CTL_SPEED_100		BGMAC_PHY_CTL_SPEED
+#define  BGMAC_PHY_CTL_SPEED_1000		BGMAC_PHY_CTL_SPEED_MSB
+#define BGMAC_PHY_ADV				0x04
+#define  BGMAC_PHY_ADV_10HALF			0x0020		/* advertise 10MBits/s half duplex */
+#define  BGMAC_PHY_ADV_10FULL			0x0040		/* advertise 10MBits/s full duplex */
+#define  BGMAC_PHY_ADV_100HALF			0x0080		/* advertise 100MBits/s half duplex */
+#define  BGMAC_PHY_ADV_100FULL			0x0100		/* advertise 100MBits/s full duplex */
+#define BGMAC_PHY_ADV2				0x09
+#define  BGMAC_PHY_ADV2_1000HALF		0x0100		/* advertise 1000MBits/s half duplex */
+#define  BGMAC_PHY_ADV2_1000FULL		0x0200		/* advertise 1000MBits/s full duplex */
+
+/* BCMA GMAC core specific IO Control (BCMA_IOCTL) flags */
+#define BGMAC_BCMA_IOCTL_SW_CLKEN		0x00000004	/* PHY Clock Enable */
+#define BGMAC_BCMA_IOCTL_SW_RESET		0x00000008	/* PHY Reset */
+
+/* BCMA GMAC core specific IO status (BCMA_IOST) flags */
+#define BGMAC_BCMA_IOST_ATTACHED		0x00000800
+
+#define BGMAC_NUM_MIB_TX_REGS	\
+		(((BGMAC_TX_Q3_OCTETS_HIGH - BGMAC_TX_GOOD_OCTETS) / 4) + 1)
+#define BGMAC_NUM_MIB_RX_REGS	\
+		(((BGMAC_RX_UNI_PKTS - BGMAC_RX_GOOD_OCTETS) / 4) + 1)
+
+#define BGMAC_DMA_TX_CTL			0x00
+#define  BGMAC_DMA_TX_ENABLE			0x00000001
+#define  BGMAC_DMA_TX_SUSPEND			0x00000002
+#define  BGMAC_DMA_TX_LOOPBACK			0x00000004
+#define  BGMAC_DMA_TX_FLUSH			0x00000010
+#define  BGMAC_DMA_TX_PARITY_DISABLE		0x00000800
+#define  BGMAC_DMA_TX_ADDREXT_MASK		0x00030000
+#define  BGMAC_DMA_TX_ADDREXT_SHIFT		16
+#define BGMAC_DMA_TX_INDEX			0x04
+#define BGMAC_DMA_TX_RINGLO			0x08
+#define BGMAC_DMA_TX_RINGHI			0x0C
+#define BGMAC_DMA_TX_STATUS			0x10
+#define  BGMAC_DMA_TX_STATDPTR			0x00001FFF
+#define  BGMAC_DMA_TX_STAT			0xF0000000
+#define   BGMAC_DMA_TX_STAT_DISABLED		0x00000000
+#define   BGMAC_DMA_TX_STAT_ACTIVE		0x10000000
+#define   BGMAC_DMA_TX_STAT_IDLEWAIT		0x20000000
+#define   BGMAC_DMA_TX_STAT_STOPPED		0x30000000
+#define   BGMAC_DMA_TX_STAT_SUSP		0x40000000
+#define BGMAC_DMA_TX_ERROR			0x14
+#define  BGMAC_DMA_TX_ERRDPTR			0x0001FFFF
+#define  BGMAC_DMA_TX_ERR			0xF0000000
+#define   BGMAC_DMA_TX_ERR_NOERR		0x00000000
+#define   BGMAC_DMA_TX_ERR_PROT			0x10000000
+#define   BGMAC_DMA_TX_ERR_UNDERRUN		0x20000000
+#define   BGMAC_DMA_TX_ERR_TRANSFER		0x30000000
+#define   BGMAC_DMA_TX_ERR_DESCREAD		0x40000000
+#define   BGMAC_DMA_TX_ERR_CORE			0x50000000
+#define BGMAC_DMA_RX_CTL			0x20
+#define  BGMAC_DMA_RX_ENABLE			0x00000001
+#define  BGMAC_DMA_RX_FRAME_OFFSET_MASK		0x000000FE
+#define  BGMAC_DMA_RX_FRAME_OFFSET_SHIFT	1
+#define  BGMAC_DMA_RX_DIRECT_FIFO		0x00000100
+#define  BGMAC_DMA_RX_OVERFLOW_CONT		0x00000400
+#define  BGMAC_DMA_RX_PARITY_DISABLE		0x00000800
+#define  BGMAC_DMA_RX_ADDREXT_MASK		0x00030000
+#define  BGMAC_DMA_RX_ADDREXT_SHIFT		16
+#define BGMAC_DMA_RX_INDEX			0x24
+#define BGMAC_DMA_RX_RINGLO			0x28
+#define BGMAC_DMA_RX_RINGHI			0x2C
+#define BGMAC_DMA_RX_STATUS			0x30
+#define  BGMAC_DMA_RX_STATDPTR			0x00001FFF
+#define  BGMAC_DMA_RX_STAT			0xF0000000
+#define   BGMAC_DMA_RX_STAT_DISABLED		0x00000000
+#define   BGMAC_DMA_RX_STAT_ACTIVE		0x10000000
+#define   BGMAC_DMA_RX_STAT_IDLEWAIT		0x20000000
+#define   BGMAC_DMA_RX_STAT_STOPPED		0x30000000
+#define   BGMAC_DMA_RX_STAT_SUSP		0x40000000
+#define BGMAC_DMA_RX_ERROR			0x34
+#define  BGMAC_DMA_RX_ERRDPTR			0x0001FFFF
+#define  BGMAC_DMA_RX_ERR			0xF0000000
+#define   BGMAC_DMA_RX_ERR_NOERR		0x00000000
+#define   BGMAC_DMA_RX_ERR_PROT			0x10000000
+#define   BGMAC_DMA_RX_ERR_UNDERRUN		0x20000000
+#define   BGMAC_DMA_RX_ERR_TRANSFER		0x30000000
+#define   BGMAC_DMA_RX_ERR_DESCREAD		0x40000000
+#define   BGMAC_DMA_RX_ERR_CORE			0x50000000
+
+#define BGMAC_DESC_CTL0_EOT			0x10000000	/* End of ring */
+#define BGMAC_DESC_CTL0_IOC			0x20000000	/* IRQ on complete */
+#define BGMAC_DESC_CTL0_SOF			0x40000000	/* Start of frame */
+#define BGMAC_DESC_CTL0_EOF			0x80000000	/* End of frame */
+#define BGMAC_DESC_CTL1_LEN			0x00001FFF
+
+#define BGMAC_PHY_NOREGS			0x1E
+#define BGMAC_PHY_MASK				0x1F
+
+#define BGMAC_MAX_TX_RINGS			4
+#define BGMAC_MAX_RX_RINGS			1
+
+#define BGMAC_TX_RING_SLOTS			128
+#define BGMAC_RX_RING_SLOTS			512 - 1		/* Why -1? Well, Broadcom does that... */
+
+#define BGMAC_RX_HEADER_LEN			28		/* Last 24 bytes are unused. Well... */
+#define BGMAC_RX_FRAME_OFFSET			30		/* There are 2 unused bytes between header and real data */
+#define BGMAC_RX_MAX_FRAME_SIZE			1536		/* Copied from b44/tg3 */
+#define BGMAC_RX_BUF_SIZE			(BGMAC_RX_FRAME_OFFSET + BGMAC_RX_MAX_FRAME_SIZE)
+
+#define BGMAC_BFL_ENETROBO			0x0010		/* has ephy roboswitch spi */
+#define BGMAC_BFL_ENETADM			0x0080		/* has ADMtek switch */
+#define BGMAC_BFL_ENETVLAN			0x0100		/* can do vlan */
+
+#define BGMAC_CHIPCTL_1_IF_TYPE_MASK		0x00000030
+#define BGMAC_CHIPCTL_1_IF_TYPE_RMII		0x00000000
+#define BGMAC_CHIPCTL_1_IF_TYPE_MI		0x00000010
+#define BGMAC_CHIPCTL_1_IF_TYPE_RGMII		0x00000020
+#define BGMAC_CHIPCTL_1_SW_TYPE_MASK		0x000000C0
+#define BGMAC_CHIPCTL_1_SW_TYPE_EPHY		0x00000000
+#define BGMAC_CHIPCTL_1_SW_TYPE_EPHYMII		0x00000040
+#define BGMAC_CHIPCTL_1_SW_TYPE_EPHYRMII	0x00000080
+#define BGMAC_CHIPCTL_1_SW_TYPE_RGMI		0x000000C0
+#define BGMAC_CHIPCTL_1_RXC_DLL_BYPASS		0x00010000
+
+#define BGMAC_SPEED_10				0x0001
+#define BGMAC_SPEED_100				0x0002
+#define BGMAC_SPEED_1000			0x0004
+
+#define BGMAC_WEIGHT	64
+
+#define ETHER_MAX_LEN   1518
+
+struct bgmac_slot_info {
+	struct sk_buff *skb;
+	dma_addr_t dma_addr;
+};
+
+struct bgmac_dma_desc {
+	__le32 ctl0;
+	__le32 ctl1;
+	__le32 addr_low;
+	__le32 addr_high;
+} __packed;
+
+enum bgmac_dma_ring_type {
+	BGMAC_DMA_RING_TX,
+	BGMAC_DMA_RING_RX,
+};
+
+/**
+ * bgmac_dma_ring - contains info about DMA ring (either TX or RX one)
+ * @start: index of the first slot containing data
+ * @end: index of a slot that can *not* be read (yet)
+ *
+ * Be really aware of the specific @end meaning. It's an index of a slot *after*
+ * the one containing data that can be read. If @start equals @end the ring is
+ * empty.
+ */
+struct bgmac_dma_ring {
+	u16 num_slots;
+	u16 start;
+	u16 end;
+
+	u16 mmio_base;
+	struct bgmac_dma_desc *cpu_base;
+	dma_addr_t dma_base;
+
+	struct bgmac_slot_info slots[BGMAC_RX_RING_SLOTS];
+};
+
+struct bgmac_rx_header {
+	__le16 len;
+	__le16 flags;
+	__le16 pad[12];
+};
+
+struct bgmac {
+	struct bcma_device *core;
+	struct bcma_device *cmn; /* Reference to CMN core for BCM4706 */
+	struct net_device *net_dev;
+	struct napi_struct napi;
+
+	/* DMA */
+	struct bgmac_dma_ring tx_ring[BGMAC_MAX_TX_RINGS];
+	struct bgmac_dma_ring rx_ring[BGMAC_MAX_RX_RINGS];
+
+	/* Stats */
+	bool stats_grabbed;
+	u32 mib_tx_regs[BGMAC_NUM_MIB_TX_REGS];
+	u32 mib_rx_regs[BGMAC_NUM_MIB_RX_REGS];
+
+	/* Int */
+	u32 int_mask;
+	u32 int_status;
+
+	/* Speed-related */
+	int speed;
+	bool autoneg;
+	bool full_duplex;
+
+	u8 phyaddr;
+	bool has_robosw;
+
+	bool loopback;
+};
+
+static inline u32 bgmac_read(struct bgmac *bgmac, u16 offset)
+{
+	return bcma_read32(bgmac->core, offset);
+}
+
+static inline void bgmac_write(struct bgmac *bgmac, u16 offset, u32 value)
+{
+	bcma_write32(bgmac->core, offset, value);
+}
+
+static inline void bgmac_maskset(struct bgmac *bgmac, u16 offset, u32 mask,
+				   u32 set)
+{
+	bgmac_write(bgmac, offset, (bgmac_read(bgmac, offset) & mask) | set);
+}
+
+static inline void bgmac_mask(struct bgmac *bgmac, u16 offset, u32 mask)
+{
+	bgmac_maskset(bgmac, offset, mask, 0);
+}
+
+static inline void bgmac_set(struct bgmac *bgmac, u16 offset, u32 set)
+{
+	bgmac_maskset(bgmac, offset, ~0, set);
+}
+
+u16 bgmac_phy_read(struct bgmac *bgmac, u8 phyaddr, u8 reg);
+void bgmac_phy_write(struct bgmac *bgmac, u8 phyaddr, u8 reg, u16 value);
+
+#endif /* _BGMAC_H */
diff --git a/include/linux/bcma/bcma_driver_chipcommon.h b/include/linux/bcma/bcma_driver_chipcommon.h
index 9a0e3fa..ee332fa 100644
--- a/include/linux/bcma/bcma_driver_chipcommon.h
+++ b/include/linux/bcma/bcma_driver_chipcommon.h
@@ -634,4 +634,6 @@ extern void bcma_chipco_regctl_maskset(struct bcma_drv_cc *cc,
 				       u32 offset, u32 mask, u32 set);
 extern void bcma_pmu_spuravoid_pllupdate(struct bcma_drv_cc *cc, int spuravoid);
 
+extern u32 bcma_pmu_get_bus_clock(struct bcma_drv_cc *cc);
+
 #endif /* LINUX_BCMA_DRIVER_CC_H_ */
-- 
1.7.7

^ permalink raw reply related

* Re: Major network performance regression in 3.7
From: Eric Dumazet @ 2013-01-06  0:02 UTC (permalink / raw)
  To: Willy Tarreau; +Cc: netdev, linux-kernel
In-Reply-To: <20130105232913.GQ16031@1wt.eu>

On Sun, 2013-01-06 at 00:29 +0100, Willy Tarreau wrote:

> > 2) Another possibility would be that Myri card/driver doesnt like very
> > well high order pages.
> 
> It looks like it has not changed much since 3.6 :-/ I really suspect
> something is wrong with memory allocation. I have tried reverting many
> patches affecting the mm/ directory just in case but I did not come to
> anything useful yet.
> 

Hmm, I was referring to TCP stack now using order-3 pages instead of
order-0 ones

See commit 5640f7685831e088fe6c2e1f863a6805962f8e81
(net: use a per task frag allocator)

Could you please post :

ethtool -S eth0

^ permalink raw reply

* Re: Major network performance regression in 3.7
From: Willy Tarreau @ 2013-01-05 23:29 UTC (permalink / raw)
  To: Eric Dumazet; +Cc: netdev, linux-kernel
In-Reply-To: <1357427926.1678.4771.camel@edumazet-glaptop>

Hi Eric,

On Sat, Jan 05, 2013 at 03:18:46PM -0800, Eric Dumazet wrote:
> Hi Willy, another good finding during the week end ! ;)

Yes, I wanted to experiment with TFO and stopped on this :-)

> 1) This looks like interrupts are spreaded on multiple cpus, and this
> give Out Of Order problems with TCP stack.

No, I forgot to mention this, I have tried to bind IRQs to a single
core, with the server either on the same or another one, but the
problem remained.

Also, the loopback is much more affected and doesn't use IRQs. And
BTW tcpdump on the loopback shouldn't drop that many packets (up to
90% even at low rate). I just noticed something, transferring data
using netcat on the loopback doesn't affect tcpdump. So it's likely
only the spliced data that are affected.

> 2) Another possibility would be that Myri card/driver doesnt like very
> well high order pages.

It looks like it has not changed much since 3.6 :-/ I really suspect
something is wrong with memory allocation. I have tried reverting many
patches affecting the mm/ directory just in case but I did not come to
anything useful yet.

I'm continuing to dig.

Thanks,
Willy

^ permalink raw reply

* Re: Major network performance regression in 3.7
From: Eric Dumazet @ 2013-01-05 23:18 UTC (permalink / raw)
  To: Willy Tarreau; +Cc: netdev, linux-kernel
In-Reply-To: <20130105214958.GA19814@1wt.eu>

On Sat, 2013-01-05 at 22:49 +0100, Willy Tarreau wrote:
> Hi,
> 
> I'm observing multiple apparently unrelated network performance
> issues in 3.7, to the point that I'm doubting it comes from the
> network stack.
> 
> My setup involves 3 machines connected point-to-point with myri
> 10GE NICs (the middle machine has 2 NICs). The middle machine
> normally runs haproxy, the other two run either an HTTP load
> generator or a dummy web server :
> 
> 
>   [ client ] <--------> [ haproxy ] <--------> [ server ]
> 
> Usually transferring HTTP objects from the server to the client
> via haproxy causes no problem at 10 Gbps for moderately large
> objects.
> 
> This time I observed that it was not possible to go beyond 6.8 Gbps,
> with all the chain idling a lot. I tried to change the IRQ rate, CPU
> affinity, tcp_rmem/tcp_wmem, disabling flow control, etc... the usual
> knobs, nothing managed to go beyond.
> 
> So I removed haproxy from the equation, and simply started the client
> on the middle machine. Same issue. I thought about concurrency issues,
> so I reduced to a single connection, and nothing changed (usually I
> achieve 10G even with a single connection with large enough TCP windows).
> I tried to start tcpdump and the transfer immediately stalled and did not
> come back after I stopped tcpdump. This was reproducible several times
> but not always.
> 
> So I first thought about an issue in the myri10ge driver and wanted to
> confirm that everything was OK on the middle machine.
> 
> I started the server on it and aimed the client at it via the loopback.
> The transfer rate was even worse : randomly oscillating between 10 and
> 100 MB/s ! Normally on the loop back, I get several GB/s here.
> 
> Running tcpdump on the loopback showed be several very concerning issues :
> 
> 1) lots of packets are lost before reaching tcpdump. The trace shows that
>    these segments are ACKed so they're correctly received, but tcpdump
>    does not get them. Tcpdump stats at the end report impressive numbers,
>    around 90% packet dropped from the capture!
> 
> 2) ACKs seem to be immediately delivered but do not trigger sending, the
>    system seems to be running with delayed ACKs, as it waits 40 or 200ms
>    before restarting, and this is visible even in the first round trips :
> 
>    - connection setup :
> 
>    18:32:08.071602 IP 127.0.0.1.26792 > 127.0.0.1.8000: S 2036886615:2036886615(0) win 8030 <mss 65495,nop,nop,sackOK,nop,wscale 9>
>    18:32:08.071605 IP 127.0.0.1.8000 > 127.0.0.1.26792: S 126397113:126397113(0) ack 2036886616 win 8030 <mss 65495,nop,nop,sackOK,nop,wscale 9>
>    18:32:08.071614 IP 127.0.0.1.26792 > 127.0.0.1.8000: . ack 126397114 win 16
> 
>    - GET /?s=1g HTTP/1.0
> 
>    18:32:08.071649 IP 127.0.0.1.26792 > 127.0.0.1.8000: P 2036886616:2036886738(122) ack 126397114 win 16
> 
>    - HTTP/1.1 200 OK with the beginning of the response :
> 
>    18:32:08.071672 IP 127.0.0.1.8000 > 127.0.0.1.26792: . 126397114:126401210(4096) ack 2036886738 win 16
>    18:32:08.071676 IP 127.0.0.1.26792 > 127.0.0.1.8000: . ack 126401210 win 250
>    ==> 200ms pause here
>    18:32:08.275493 IP 127.0.0.1.8000 > 127.0.0.1.26792: P 126401210:126463006(61796) ack 2036886738 win 16
>    ==> 40ms pause here
>    18:32:08.315493 IP 127.0.0.1.26792 > 127.0.0.1.8000: . ack 126463006 win 256
>    18:32:08.315498 IP 127.0.0.1.8000 > 127.0.0.1.26792: . 126463006:126527006(64000) ack 2036886738 win 16
> 
>    ... and so on
> 
>    My server is using splice() with the SPLICE_F_MORE flag to send data.
>    I noticed that not using splice and relying on send(MSG_MORE) instead
>    I don't get the issue.
> 
> 3) I wondered if this had something to do with the 64k MTU on the loopback
>    so I lowered it to 16kB. The performance was even worse (about 5MB/s).
>    Starting tcpdump managed to make my transfer stall, just like with the
>    myri10ge. In this last test, I noticed that there were some real drops,
>    because there were some SACKs :
> 
>    18:45:16.699951 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 956153186:956169530(16344) ack 131668746 win 16
>    18:45:16.699956 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 956169530 win 64
>    18:45:16.904119 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 957035762:957052106(16344) ack 131668746 win 16
>    18:45:16.904122 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 957052106 win 703
>    18:45:16.904124 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 957052106:957099566(47460) ack 131668746 win 16
>    18:45:17.108117 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 957402550:957418894(16344) ack 131668746 win 16
>    18:45:17.108119 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 957418894 win 1846
>    18:45:17.312115 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 957672806:957689150(16344) ack 131668746 win 16
>    18:45:17.312117 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 957689150 win 2902
>    18:45:17.516114 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 958962966:958979310(16344) ack 131668746 win 16
>    18:45:17.516116 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 958979310 win 7941
>    18:45:17.516150 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 959503678 win 9926 <nop,nop,sack 1 {959405614:959421958}>
>    18:45:17.516151 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 959503678 win 9926 <nop,nop,sack 1 {959421958:959438302}>
> 
> Please note that the Myri card is running with the normal MTU of 1500,
> jumbo frames were not used.
> 
> From what I could test, only the Tx path seems affected, because after
> rebooting the server on 3.5.7, I can transfer at 10 Gbps via the myri10ge
> again to the 3.7.1 client. I tried to disable GSO, TSO, etc... but nothing
> worked. Also, on 3.5.7, there are almost no drops between the kernel and
> tcpdump.
> 
> I really suspect that all these issues are related. Maybe something about
> send buffers recycling, I don't know. The fact that tcpdump gets very few
> packets also makes me think about something about memory allocation. And
> that could perhaps explain why starting it causes the traffic to stall.
> 
> Sadly, it will be hard to bisect this bug because I had a hard time running
> 3.7-rc up to rc5 due to several bugs that affected this workload (and were
> fortunately fixed since).
> 
> I think it's better to only focus on the issue with the loopback since it's
> easier to reproduce and involves less moving blocks.
> 
> I've just tried to remove the SPLICE_F_MORE flag on the sender but it does
> not change anything.
> 
> If anyone has any idea of things to test to narrow the problem down, I'm
> interested. I have not tried 3.8-rc2 yet, and reading the archives there
> does not seem to be any similar reports. Note that I've been using 3.5.7
> as a fallback, but the same lab has run 3.6.6 without any issues 2 months
> ago, so I'm really confident that this is a 3.7 regression.
> 
> Regards,
> Willy
> 
> --

Hi Willy, another good finding during the week end ! ;)

1) This looks like interrupts are spreaded on multiple cpus, and this
give Out Of Order problems with TCP stack.

2) Another possibility would be that Myri card/driver doesnt like very
well high order pages.

^ permalink raw reply

* Re: [PATCH] ipv4: fix NULL checking in devinet_ioctl()
From: Eric Dumazet @ 2013-01-05 23:10 UTC (permalink / raw)
  To: Xi Wang; +Cc: David S. Miller, netdev
In-Reply-To: <1357420764-12381-1-git-send-email-xi.wang@gmail.com>

On Sat, 2013-01-05 at 16:19 -0500, Xi Wang wrote:
> The NULL pointer check `!ifa' should come before its first use.
> 
> Signed-off-by: Xi Wang <xi.wang@gmail.com>
> ---
>  net/ipv4/devinet.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/net/ipv4/devinet.c b/net/ipv4/devinet.c
> index cc06a47..a8e4f26 100644
> --- a/net/ipv4/devinet.c
> +++ b/net/ipv4/devinet.c
> @@ -823,9 +823,9 @@ int devinet_ioctl(struct net *net, unsigned int cmd, void __user *arg)
>  		if (!ifa) {
>  			ret = -ENOBUFS;
>  			ifa = inet_alloc_ifa();
> -			INIT_HLIST_NODE(&ifa->hash);
>  			if (!ifa)
>  				break;
> +			INIT_HLIST_NODE(&ifa->hash);
>  			if (colon)
>  				memcpy(ifa->ifa_label, ifr.ifr_name, IFNAMSIZ);
>  			else

Acked-by: Eric Dumazet <edumazet@google.com>

Bug origin : commit fd23c3b31107e2fc483301ee923d8a1db14e53f4
(ipv4: Add hash table of interface addresses) 

in linux-2.6.39

Thanks

^ permalink raw reply

* Major network performance regression in 3.7
From: Willy Tarreau @ 2013-01-05 21:49 UTC (permalink / raw)
  To: netdev, linux-kernel

Hi,

I'm observing multiple apparently unrelated network performance
issues in 3.7, to the point that I'm doubting it comes from the
network stack.

My setup involves 3 machines connected point-to-point with myri
10GE NICs (the middle machine has 2 NICs). The middle machine
normally runs haproxy, the other two run either an HTTP load
generator or a dummy web server :


  [ client ] <--------> [ haproxy ] <--------> [ server ]

Usually transferring HTTP objects from the server to the client
via haproxy causes no problem at 10 Gbps for moderately large
objects.

This time I observed that it was not possible to go beyond 6.8 Gbps,
with all the chain idling a lot. I tried to change the IRQ rate, CPU
affinity, tcp_rmem/tcp_wmem, disabling flow control, etc... the usual
knobs, nothing managed to go beyond.

So I removed haproxy from the equation, and simply started the client
on the middle machine. Same issue. I thought about concurrency issues,
so I reduced to a single connection, and nothing changed (usually I
achieve 10G even with a single connection with large enough TCP windows).
I tried to start tcpdump and the transfer immediately stalled and did not
come back after I stopped tcpdump. This was reproducible several times
but not always.

So I first thought about an issue in the myri10ge driver and wanted to
confirm that everything was OK on the middle machine.

I started the server on it and aimed the client at it via the loopback.
The transfer rate was even worse : randomly oscillating between 10 and
100 MB/s ! Normally on the loop back, I get several GB/s here.

Running tcpdump on the loopback showed be several very concerning issues :

1) lots of packets are lost before reaching tcpdump. The trace shows that
   these segments are ACKed so they're correctly received, but tcpdump
   does not get them. Tcpdump stats at the end report impressive numbers,
   around 90% packet dropped from the capture!

2) ACKs seem to be immediately delivered but do not trigger sending, the
   system seems to be running with delayed ACKs, as it waits 40 or 200ms
   before restarting, and this is visible even in the first round trips :

   - connection setup :

   18:32:08.071602 IP 127.0.0.1.26792 > 127.0.0.1.8000: S 2036886615:2036886615(0) win 8030 <mss 65495,nop,nop,sackOK,nop,wscale 9>
   18:32:08.071605 IP 127.0.0.1.8000 > 127.0.0.1.26792: S 126397113:126397113(0) ack 2036886616 win 8030 <mss 65495,nop,nop,sackOK,nop,wscale 9>
   18:32:08.071614 IP 127.0.0.1.26792 > 127.0.0.1.8000: . ack 126397114 win 16

   - GET /?s=1g HTTP/1.0

   18:32:08.071649 IP 127.0.0.1.26792 > 127.0.0.1.8000: P 2036886616:2036886738(122) ack 126397114 win 16

   - HTTP/1.1 200 OK with the beginning of the response :

   18:32:08.071672 IP 127.0.0.1.8000 > 127.0.0.1.26792: . 126397114:126401210(4096) ack 2036886738 win 16
   18:32:08.071676 IP 127.0.0.1.26792 > 127.0.0.1.8000: . ack 126401210 win 250
   ==> 200ms pause here
   18:32:08.275493 IP 127.0.0.1.8000 > 127.0.0.1.26792: P 126401210:126463006(61796) ack 2036886738 win 16
   ==> 40ms pause here
   18:32:08.315493 IP 127.0.0.1.26792 > 127.0.0.1.8000: . ack 126463006 win 256
   18:32:08.315498 IP 127.0.0.1.8000 > 127.0.0.1.26792: . 126463006:126527006(64000) ack 2036886738 win 16

   ... and so on

   My server is using splice() with the SPLICE_F_MORE flag to send data.
   I noticed that not using splice and relying on send(MSG_MORE) instead
   I don't get the issue.

3) I wondered if this had something to do with the 64k MTU on the loopback
   so I lowered it to 16kB. The performance was even worse (about 5MB/s).
   Starting tcpdump managed to make my transfer stall, just like with the
   myri10ge. In this last test, I noticed that there were some real drops,
   because there were some SACKs :

   18:45:16.699951 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 956153186:956169530(16344) ack 131668746 win 16
   18:45:16.699956 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 956169530 win 64
   18:45:16.904119 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 957035762:957052106(16344) ack 131668746 win 16
   18:45:16.904122 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 957052106 win 703
   18:45:16.904124 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 957052106:957099566(47460) ack 131668746 win 16
   18:45:17.108117 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 957402550:957418894(16344) ack 131668746 win 16
   18:45:17.108119 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 957418894 win 1846
   18:45:17.312115 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 957672806:957689150(16344) ack 131668746 win 16
   18:45:17.312117 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 957689150 win 2902
   18:45:17.516114 IP 127.0.0.1.8000 > 127.0.0.1.8002: P 958962966:958979310(16344) ack 131668746 win 16
   18:45:17.516116 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 958979310 win 7941
   18:45:17.516150 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 959503678 win 9926 <nop,nop,sack 1 {959405614:959421958}>
   18:45:17.516151 IP 127.0.0.1.8002 > 127.0.0.1.8000: . ack 959503678 win 9926 <nop,nop,sack 1 {959421958:959438302}>

Please note that the Myri card is running with the normal MTU of 1500,
jumbo frames were not used.

>From what I could test, only the Tx path seems affected, because after
rebooting the server on 3.5.7, I can transfer at 10 Gbps via the myri10ge
again to the 3.7.1 client. I tried to disable GSO, TSO, etc... but nothing
worked. Also, on 3.5.7, there are almost no drops between the kernel and
tcpdump.

I really suspect that all these issues are related. Maybe something about
send buffers recycling, I don't know. The fact that tcpdump gets very few
packets also makes me think about something about memory allocation. And
that could perhaps explain why starting it causes the traffic to stall.

Sadly, it will be hard to bisect this bug because I had a hard time running
3.7-rc up to rc5 due to several bugs that affected this workload (and were
fortunately fixed since).

I think it's better to only focus on the issue with the loopback since it's
easier to reproduce and involves less moving blocks.

I've just tried to remove the SPLICE_F_MORE flag on the sender but it does
not change anything.

If anyone has any idea of things to test to narrow the problem down, I'm
interested. I have not tried 3.8-rc2 yet, and reading the archives there
does not seem to be any similar reports. Note that I've been using 3.5.7
as a fallback, but the same lab has run 3.6.6 without any issues 2 months
ago, so I'm really confident that this is a 3.7 regression.

Regards,
Willy

^ permalink raw reply

* Re: [PATCH 1/2] ipv6: avoid blackhole and prohibited entries upon prefix purge [v2]
From: Romain KUNTZ @ 2013-01-05 21:44 UTC (permalink / raw)
  To: netdev@vger.kernel.org; +Cc: Eric Dumazet, yoshfuji, davem
In-Reply-To: <1357415941.1678.4163.camel@edumazet-glaptop>

Mobile IPv6 provokes a kernel Oops since commit 64c6d08e (ipv6:
del unreachable route when an addr is deleted on lo), because
ip6_route_lookup() may also return blackhole and prohibited
entry. However, these entries have a NULL rt6i_table argument,
which provokes an Oops in __ip6_del_rt() when trying to lock
rt6i_table->tb6_lock.

Beside, when purging a prefix, blakhole and prohibited entries
should not be selected because they are not what we are looking
for.

We fix this by adding two new lookup flags (RT6_LOOKUP_F_NO_BLK_HOLE
and RT6_LOOKUP_F_NO_PROHIBIT) in order to ensure that such entries
are skipped during lookup and that the correct entry is returned.

[v2]: use 'goto out;' instead of 'goto again;' to avoid unnecessary
oprations on rt (as suggested by Eric Dumazet).

Signed-off-by: Romain Kuntz <r.kuntz@ipflavors.com>
---
 include/net/ip6_route.h |    2 ++
 net/ipv6/addrconf.c     |    4 +++-
 net/ipv6/fib6_rules.c   |    4 ++++
 3 files changed, 9 insertions(+), 1 deletions(-)

diff --git a/include/net/ip6_route.h b/include/net/ip6_route.h
index 27d8318..3c93743 100644
--- a/include/net/ip6_route.h
+++ b/include/net/ip6_route.h
@@ -30,6 +30,8 @@ struct route_info {
 #define RT6_LOOKUP_F_SRCPREF_TMP	0x00000008
 #define RT6_LOOKUP_F_SRCPREF_PUBLIC	0x00000010
 #define RT6_LOOKUP_F_SRCPREF_COA	0x00000020
+#define RT6_LOOKUP_F_NO_BLK_HOLE	0x00000040
+#define RT6_LOOKUP_F_NO_PROHIBIT	0x00000080
 
 /*
  * rt6_srcprefs2flags() and rt6_flags2srcprefs() translate
diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
index 408cac4a..1891e23 100644
--- a/net/ipv6/addrconf.c
+++ b/net/ipv6/addrconf.c
@@ -948,7 +948,9 @@ static void ipv6_del_addr(struct inet6_ifaddr *ifp)
 		fl6.flowi6_oif = ifp->idev->dev->ifindex;
 		fl6.daddr = prefix;
 		rt = (struct rt6_info *)ip6_route_lookup(net, &fl6,
-							 RT6_LOOKUP_F_IFACE);
+						RT6_LOOKUP_F_IFACE |
+						RT6_LOOKUP_F_NO_BLK_HOLE |
+						RT6_LOOKUP_F_NO_PROHIBIT);
 
 		if (rt != net->ipv6.ip6_null_entry &&
 		    addrconf_is_prefix_route(rt)) {
diff --git a/net/ipv6/fib6_rules.c b/net/ipv6/fib6_rules.c
index 2e1a432..9b48128 100644
--- a/net/ipv6/fib6_rules.c
+++ b/net/ipv6/fib6_rules.c
@@ -64,9 +64,13 @@ static int fib6_rule_action(struct fib_rule *rule, struct flowi *flp,
 		goto discard_pkt;
 	default:
 	case FR_ACT_BLACKHOLE:
+		if (flags & RT6_LOOKUP_F_NO_BLK_HOLE)
+			goto out;
 		rt = net->ipv6.ip6_blk_hole_entry;
 		goto discard_pkt;
 	case FR_ACT_PROHIBIT:
+		if (flags & RT6_LOOKUP_F_NO_PROHIBIT)
+			goto out;
 		rt = net->ipv6.ip6_prohibit_entry;
 		goto discard_pkt;
 	}
-- 
1.7.2.5

^ permalink raw reply related

* [PATCH] ipv4: fix NULL checking in devinet_ioctl()
From: Xi Wang @ 2013-01-05 21:19 UTC (permalink / raw)
  To: David S. Miller; +Cc: netdev, Xi Wang

The NULL pointer check `!ifa' should come before its first use.

Signed-off-by: Xi Wang <xi.wang@gmail.com>
---
 net/ipv4/devinet.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/ipv4/devinet.c b/net/ipv4/devinet.c
index cc06a47..a8e4f26 100644
--- a/net/ipv4/devinet.c
+++ b/net/ipv4/devinet.c
@@ -823,9 +823,9 @@ int devinet_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 		if (!ifa) {
 			ret = -ENOBUFS;
 			ifa = inet_alloc_ifa();
-			INIT_HLIST_NODE(&ifa->hash);
 			if (!ifa)
 				break;
+			INIT_HLIST_NODE(&ifa->hash);
 			if (colon)
 				memcpy(ifa->ifa_label, ifr.ifr_name, IFNAMSIZ);
 			else
-- 
1.7.10.4

^ permalink raw reply related

* Re: [patch net-next] ethtool: fix drvinfo strings set in drivers
From: Jiri Pirko @ 2013-01-05 20:40 UTC (permalink / raw)
  To: netdev, davem, edumazet, bhutchings, shemminger, sathya.perla,
	subbu.seetharaman, ajit.khaparde, roland, sean.hefty,
	hal.rosenstock
In-Reply-To: <20130105202356.GA1967@obelix.rh>

Sat, Jan 05, 2013 at 09:23:57PM CET, fbl@redhat.com wrote:
>On Sat, Jan 05, 2013 at 08:48:56PM +0100, Jiri Pirko wrote:
>> Use strlcpy where possible to ensure the string is \0 terminated.
>> Use always sizeof(string) instead of 32 and custom defines.
>> Also use snprintf instead of strcpy in nes_nic.
>> Also do not do unnecessary init of drvinfo in spider_net_ethtool.
>> 
>
>Looks good. You can add some more fixes to your patch
>BTW, I was using ethtool defines, but I think it's cleaner
>to use sizeof() instead.

Okay, I will add these 4.


>
>fbl
>
>diff --git a/drivers/net/cris/eth_v10.c b/drivers/net/cris/eth_v10.c
>index 021d69c..29e272c 100644
>--- a/drivers/net/cris/eth_v10.c
>+++ b/drivers/net/cris/eth_v10.c
>@@ -1448,10 +1448,10 @@ static int e100_set_settings(struct net_device *dev,
> static void e100_get_drvinfo(struct net_device *dev,
> 			     struct ethtool_drvinfo *info)
> {
>-	strncpy(info->driver, "ETRAX 100LX", sizeof(info->driver) - 1);
>-	strncpy(info->version, "$Revision: 1.31 $", sizeof(info->version) - 1);
>-	strncpy(info->fw_version, "N/A", sizeof(info->fw_version) - 1);
>-	strncpy(info->bus_info, "N/A", sizeof(info->bus_info) - 1);
>+	strlcpy(info->driver, "ETRAX 100LX", sizeof(info->driver));
>+	strlcpy(info->version, "$Revision: 1.31 $", sizeof(info->version));
>+	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
>+	strlcpy(info->bus_info, "N/A", sizeof(info->bus_info));
> }
> 
> static int e100_nway_reset(struct net_device *dev)
>diff --git a/drivers/net/ethernet/3com/3c501.c b/drivers/net/ethernet/3com/3c501.c
>index 2038eaa..c6dfbb6 100644
>--- a/drivers/net/ethernet/3com/3c501.c
>+++ b/drivers/net/ethernet/3com/3c501.c
>@@ -823,9 +823,10 @@ static void set_multicast_list(struct net_device *dev)
> static void netdev_get_drvinfo(struct net_device *dev,
> 			       struct ethtool_drvinfo *info)
> {
>-	strcpy(info->driver, DRV_NAME);
>-	strcpy(info->version, DRV_VERSION);
>-	sprintf(info->bus_info, "ISA 0x%lx", dev->base_addr);
>+	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
>+	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
>+	snprintf(info->bus_info, ETHTOOL_BUSINFO_LEN, "ISA 0x%lx",
>+		 dev->base_addr);
> }
> 
> static u32 netdev_get_msglevel(struct net_device *dev)
>diff --git a/drivers/net/ethernet/3com/3c509.c b/drivers/net/ethernet/3com/3c509.c
>index 633c709..f36ff99 100644
>--- a/drivers/net/ethernet/3com/3c509.c
>+++ b/drivers/net/ethernet/3com/3c509.c
>@@ -1161,8 +1161,8 @@ el3_netdev_set_ecmd(struct net_device *dev, struct ethtool_cmd *ecmd)
> 
> static void el3_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
> {
>-	strcpy(info->driver, DRV_NAME);
>-	strcpy(info->version, DRV_VERSION);
>+	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
>+	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
> }
> 
> static int el3_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
>diff --git a/drivers/net/ethernet/3com/3c515.c b/drivers/net/ethernet/3com/3c515.c
>index 59e1e00..9e9fef3 100644
>--- a/drivers/net/ethernet/3com/3c515.c
>+++ b/drivers/net/ethernet/3com/3c515.c
>@@ -1542,9 +1542,10 @@ static void set_rx_mode(struct net_device *dev)
> static void netdev_get_drvinfo(struct net_device *dev,
> 			       struct ethtool_drvinfo *info)
> {
>-	strcpy(info->driver, DRV_NAME);
>-	strcpy(info->version, DRV_VERSION);
>-	sprintf(info->bus_info, "ISA 0x%lx", dev->base_addr);
>+	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
>+	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
>+	snprintf(info->bus_info, ETHTOOL_BUSINFO_LEN, "ISA 0x%lx",
>+		 dev->base_addr);
> }
> 
> static u32 netdev_get_msglevel(struct net_device *dev)
>

^ permalink raw reply

* Re: [patch net-next] ethtool: fix drvinfo strings set in drivers
From: Flavio Leitner @ 2013-01-05 20:23 UTC (permalink / raw)
  To: Jiri Pirko
  Cc: netdev, davem, edumazet, bhutchings, shemminger, sathya.perla,
	subbu.seetharaman, ajit.khaparde, roland, sean.hefty,
	hal.rosenstock
In-Reply-To: <1357415336-17807-1-git-send-email-jiri@resnulli.us>

On Sat, Jan 05, 2013 at 08:48:56PM +0100, Jiri Pirko wrote:
> Use strlcpy where possible to ensure the string is \0 terminated.
> Use always sizeof(string) instead of 32 and custom defines.
> Also use snprintf instead of strcpy in nes_nic.
> Also do not do unnecessary init of drvinfo in spider_net_ethtool.
> 

Looks good. You can add some more fixes to your patch
BTW, I was using ethtool defines, but I think it's cleaner
to use sizeof() instead.

fbl

diff --git a/drivers/net/cris/eth_v10.c b/drivers/net/cris/eth_v10.c
index 021d69c..29e272c 100644
--- a/drivers/net/cris/eth_v10.c
+++ b/drivers/net/cris/eth_v10.c
@@ -1448,10 +1448,10 @@ static int e100_set_settings(struct net_device *dev,
 static void e100_get_drvinfo(struct net_device *dev,
 			     struct ethtool_drvinfo *info)
 {
-	strncpy(info->driver, "ETRAX 100LX", sizeof(info->driver) - 1);
-	strncpy(info->version, "$Revision: 1.31 $", sizeof(info->version) - 1);
-	strncpy(info->fw_version, "N/A", sizeof(info->fw_version) - 1);
-	strncpy(info->bus_info, "N/A", sizeof(info->bus_info) - 1);
+	strlcpy(info->driver, "ETRAX 100LX", sizeof(info->driver));
+	strlcpy(info->version, "$Revision: 1.31 $", sizeof(info->version));
+	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
+	strlcpy(info->bus_info, "N/A", sizeof(info->bus_info));
 }
 
 static int e100_nway_reset(struct net_device *dev)
diff --git a/drivers/net/ethernet/3com/3c501.c b/drivers/net/ethernet/3com/3c501.c
index 2038eaa..c6dfbb6 100644
--- a/drivers/net/ethernet/3com/3c501.c
+++ b/drivers/net/ethernet/3com/3c501.c
@@ -823,9 +823,10 @@ static void set_multicast_list(struct net_device *dev)
 static void netdev_get_drvinfo(struct net_device *dev,
 			       struct ethtool_drvinfo *info)
 {
-	strcpy(info->driver, DRV_NAME);
-	strcpy(info->version, DRV_VERSION);
-	sprintf(info->bus_info, "ISA 0x%lx", dev->base_addr);
+	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
+	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
+	snprintf(info->bus_info, ETHTOOL_BUSINFO_LEN, "ISA 0x%lx",
+		 dev->base_addr);
 }
 
 static u32 netdev_get_msglevel(struct net_device *dev)
diff --git a/drivers/net/ethernet/3com/3c509.c b/drivers/net/ethernet/3com/3c509.c
index 633c709..f36ff99 100644
--- a/drivers/net/ethernet/3com/3c509.c
+++ b/drivers/net/ethernet/3com/3c509.c
@@ -1161,8 +1161,8 @@ el3_netdev_set_ecmd(struct net_device *dev, struct ethtool_cmd *ecmd)
 
 static void el3_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
 {
-	strcpy(info->driver, DRV_NAME);
-	strcpy(info->version, DRV_VERSION);
+	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
+	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 }
 
 static int el3_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
diff --git a/drivers/net/ethernet/3com/3c515.c b/drivers/net/ethernet/3com/3c515.c
index 59e1e00..9e9fef3 100644
--- a/drivers/net/ethernet/3com/3c515.c
+++ b/drivers/net/ethernet/3com/3c515.c
@@ -1542,9 +1542,10 @@ static void set_rx_mode(struct net_device *dev)
 static void netdev_get_drvinfo(struct net_device *dev,
 			       struct ethtool_drvinfo *info)
 {
-	strcpy(info->driver, DRV_NAME);
-	strcpy(info->version, DRV_VERSION);
-	sprintf(info->bus_info, "ISA 0x%lx", dev->base_addr);
+	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
+	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
+	snprintf(info->bus_info, ETHTOOL_BUSINFO_LEN, "ISA 0x%lx",
+		 dev->base_addr);
 }
 
 static u32 netdev_get_msglevel(struct net_device *dev)

^ permalink raw reply related

* Re: [PATCH 1/2] ipv6: avoid blackhole and prohibited entries upon prefix purge
From: Eric Dumazet @ 2013-01-05 19:59 UTC (permalink / raw)
  To: Romain KUNTZ; +Cc: netdev, yoshfuji, davem
In-Reply-To: <0CC79564-4AF2-42F9-8D06-1BCC912A1AF7@ipflavors.com>

On Sat, 2013-01-05 at 17:13 +0100, Romain KUNTZ wrote:
> Mobile IPv6 provokes a kernel Oops since commit 64c6d08e (ipv6:
> del unreachable route when an addr is deleted on lo), because
> ip6_route_lookup() may also return blackhole and prohibited
> entry. However, these entries have a NULL rt6i_table argument,
> which provokes an Oops in __ip6_del_rt() when trying to lock
> rt6i_table->tb6_lock.
> 
> Beside, when purging a prefix, blakhole and prohibited entries
> should not be selected because they are not what we are looking
> for.
> 
> We fix this by adding two new lookup flags (RT6_LOOKUP_F_NO_BLK_HOLE
> and RT6_LOOKUP_F_NO_PROHIBIT) in order to ensure that such entries
> are skipped during lookup and that the correct entry is returned.
> 
> Signed-off-by: Romain Kuntz <r.kuntz@ipflavors.com>
> ---
>  include/net/ip6_route.h |    2 ++
>  net/ipv6/addrconf.c     |    4 +++-
>  net/ipv6/fib6_rules.c   |    4 ++++
>  3 files changed, 9 insertions(+), 1 deletions(-)
> 
> diff --git a/include/net/ip6_route.h b/include/net/ip6_route.h
> index 27d8318..3c93743 100644
> --- a/include/net/ip6_route.h
> +++ b/include/net/ip6_route.h
> @@ -30,6 +30,8 @@ struct route_info {
>  #define RT6_LOOKUP_F_SRCPREF_TMP	0x00000008
>  #define RT6_LOOKUP_F_SRCPREF_PUBLIC	0x00000010
>  #define RT6_LOOKUP_F_SRCPREF_COA	0x00000020
> +#define RT6_LOOKUP_F_NO_BLK_HOLE	0x00000040
> +#define RT6_LOOKUP_F_NO_PROHIBIT	0x00000080
>  
>  /*
>   * rt6_srcprefs2flags() and rt6_flags2srcprefs() translate
> diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
> index 408cac4a..1891e23 100644
> --- a/net/ipv6/addrconf.c
> +++ b/net/ipv6/addrconf.c
> @@ -948,7 +948,9 @@ static void ipv6_del_addr(struct inet6_ifaddr *ifp)
>  		fl6.flowi6_oif = ifp->idev->dev->ifindex;
>  		fl6.daddr = prefix;
>  		rt = (struct rt6_info *)ip6_route_lookup(net, &fl6,
> -							 RT6_LOOKUP_F_IFACE);
> +						RT6_LOOKUP_F_IFACE |
> +						RT6_LOOKUP_F_NO_BLK_HOLE |
> +						RT6_LOOKUP_F_NO_PROHIBIT);
>  
>  		if (rt != net->ipv6.ip6_null_entry &&
>  		    addrconf_is_prefix_route(rt)) {
> diff --git a/net/ipv6/fib6_rules.c b/net/ipv6/fib6_rules.c
> index 2e1a432..d290da5 100644
> --- a/net/ipv6/fib6_rules.c
> +++ b/net/ipv6/fib6_rules.c
> @@ -64,9 +64,13 @@ static int fib6_rule_action(struct fib_rule *rule, struct flowi *flp,
>  		goto discard_pkt;
>  	default:
>  	case FR_ACT_BLACKHOLE:
> +		if (flags & RT6_LOOKUP_F_NO_BLK_HOLE)
> +			goto again;

		goto out; ???

>  		rt = net->ipv6.ip6_blk_hole_entry;
>  		goto discard_pkt;
>  	case FR_ACT_PROHIBIT:
> +		if (flags & RT6_LOOKUP_F_NO_PROHIBIT)
> +			goto again;
	
		goto out;

>  		rt = net->ipv6.ip6_prohibit_entry;
>  		goto discard_pkt;
>  	}

^ permalink raw reply


This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox