12 years agoaf_key: fix SADB_X_SPDDELETE response
Tobias Brunner [Fri, 10 Oct 2008 21:07:03 +0000]
af_key: fix SADB_X_SPDDELETE response

When deleting an SPD entry using SADB_X_SPDDELETE, c.data.byid is not
initialized to zero in pfkey_spddelete(). Thus, key_notify_policy()
responds with a PF_KEY message of type SADB_X_SPDDELETE2 instead of
SADB_X_SPDDELETE.

Signed-off-by: Tobias Brunner <tobias.brunner@strongswan.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoath9k: Fix return code when ath9k_hw_setpower() fails on reset
Luis R. Rodriguez [Fri, 10 Oct 2008 19:26:24 +0000]
ath9k: Fix return code when ath9k_hw_setpower() fails on reset

We were not reporting a status code back ath9k_hw_setpower() failed
during reset so lets correct this.

Reported-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoath9k: remove nasty FAIL macro from ath9k_hw_reset()
Luis R. Rodriguez [Fri, 10 Oct 2008 19:25:45 +0000]
ath9k: remove nasty FAIL macro from ath9k_hw_reset()

This is fucking horribe crap code so nuke it. There I cursed too in a commit log.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agogre: minor cleanups in netlink interface
Patrick McHardy [Fri, 10 Oct 2008 19:11:06 +0000]
gre: minor cleanups in netlink interface

- use typeful helpers for IFLA_GRE_LOCAL/IFLA_GRE_REMOTE
- replace magic value by FIELD_SIZEOF
- use MODULE_ALIAS_RTNL_LINK macro

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agogre: fix copy and paste error
Patrick McHardy [Fri, 10 Oct 2008 19:10:30 +0000]
gre: fix copy and paste error

The flags are dumped twice, the keys not at all.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoenic: Attempt to fix build in 32-bit such as i386.
David S. Miller [Fri, 10 Oct 2008 05:13:53 +0000]
enic: Attempt to fix build in 32-bit such as i386.

Such platforms lack readq/writeq but this driver want to call them.

Noticed by Andrew Morton.

Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agotcpv6: fix error with CONFIG_TCP_MD5SIG disabled
Guo-Fu Tseng [Fri, 10 Oct 2008 04:11:56 +0000]
tcpv6: fix error with CONFIG_TCP_MD5SIG disabled

This patch fix error with CONFIG_TCP_MD5SIG disabled.

Signed-off-by: Guo-Fu Tseng <cooldavid@cooldavid.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agonf_conntrack_ecache.h: Fix missing braces
Guo-Fu Tseng [Fri, 10 Oct 2008 04:10:36 +0000]
nf_conntrack_ecache.h: Fix missing braces

This patch add missing braces of today's net-next-2.6:
include/net/netfilter/nf_conntrack_ecache.h

Signed-off-by: Guo-Fu Tseng <cooldavid@cooldavid.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoppp: Use skb_queue_walk() in ppp_mp_insert().
David S. Miller [Thu, 9 Oct 2008 23:40:29 +0000]
ppp: Use skb_queue_walk() in ppp_mp_insert().

Instead of open-coded version.

Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoudp: complete port availability checking
Eric Dumazet [Thu, 9 Oct 2008 21:51:27 +0000]
udp: complete port availability checking

While looking at UDP port randomization, I noticed it
was litle bit pessimistic, not looking at type of sockets
(IPV6/IPV4) and not looking at bound addresses if any.

We should perform same tests than when binding to a
specific port.

This permits a cleanup of udp_lib_get_port()

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agotcpv6: combine tcp_v6_send_(reset|ack)
Ilpo Järvinen [Thu, 9 Oct 2008 21:42:40 +0000]
tcpv6: combine tcp_v6_send_(reset|ack)

$ codiff tcp_ipv6.o.old tcp_ipv6.o.new
net/ipv6/tcp_ipv6.c:
  tcp_v6_md5_hash_hdr | -144
  tcp_v6_send_ack     | -585
  tcp_v6_send_reset   | -540
 3 functions changed, 1269 bytes removed, diff: -1269

net/ipv6/tcp_ipv6.c:
  tcp_v6_send_response | +791
 1 function changed, 791 bytes added, diff: +791

tcp_ipv6.o.new:
 4 functions changed, 791 bytes added, 1269 bytes removed, diff: -478

I choose to leave the reset related netns comment in place (not
the one that is killed) as I cannot understand its English so
it's a bit hard for me to evaluate its usefulness :-).

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agotcpv6: convert opt[] -> topt in tcp_v6_send_reset
Ilpo Järvinen [Thu, 9 Oct 2008 21:42:01 +0000]
tcpv6: convert opt[] -> topt in tcp_v6_send_reset

after this I get:

$ diff-funcs tcp_v6_send_reset tcp_ipv6.c tcp_ipv6.c tcp_v6_send_ack
 --- tcp_ipv6.c:tcp_v6_send_reset()
 +++ tcp_ipv6.c:tcp_v6_send_ack()
@@ -1,4 +1,5 @@
-static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
+static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
u32 ts,
+                           struct tcp_md5sig_key *key)
 {
        struct tcphdr *th = tcp_hdr(skb), *t1;
        struct sk_buff *buff;
@@ -7,31 +8,14 @@
        struct sock *ctl_sk = net->ipv6.tcp_sk;
        unsigned int tot_len = sizeof(struct tcphdr);
        __be32 *topt;
-#ifdef CONFIG_TCP_MD5SIG
-       struct tcp_md5sig_key *key;
-#endif
-
-       if (th->rst)
-               return;
-
-       if (!ipv6_unicast_destination(skb))
-               return;

+       if (ts)
+               tot_len += TCPOLEN_TSTAMP_ALIGNED;
 #ifdef CONFIG_TCP_MD5SIG
-       if (sk)
-               key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
-       else
-               key = NULL;
-
        if (key)
                tot_len += TCPOLEN_MD5SIG_ALIGNED;
 #endif

-       /*
-        * We need to grab some memory, and put together an RST,
-        * and then put it into the queue to be sent.
-        */
-
        buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
                         GFP_ATOMIC);
        if (buff == NULL)
@@ -46,18 +30,20 @@
        t1->dest = th->source;
        t1->source = th->dest;
        t1->doff = tot_len / 4;
-       t1->rst = 1;
-
-       if(th->ack) {
-               t1->seq = th->ack_seq;
-       } else {
-               t1->ack = 1;
-               t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
-                                   + skb->len - (th->doff<<2));
-       }
+       t1->seq = htonl(seq);
+       t1->ack_seq = htonl(ack);
+       t1->ack = 1;
+       t1->window = htons(win);

        topt = (__be32 *)(t1 + 1);

+       if (ts) {
+               *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
+                               (TCPOPT_TIMESTAMP << 8) |
TCPOLEN_TIMESTAMP);
+               *topt++ = htonl(tcp_time_stamp);
+               *topt++ = htonl(ts);
+       }
+
 #ifdef CONFIG_TCP_MD5SIG
        if (key) {
                *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
@@ -84,15 +70,10 @@
        fl.fl_ip_sport = t1->source;
        security_skb_classify_flow(skb, &fl);

-       /* Pass a socket to ip6_dst_lookup either it is for RST
-        * Underlying function will use this to retrieve the network
-        * namespace
-        */
        if (!ip6_dst_lookup(ctl_sk, &buff->dst, &fl)) {
                if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
                        ip6_xmit(ctl_sk, buff, &fl, NULL, 0);
                        TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
-                       TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
                        return;
                }
        }

...which starts to be trivial to combine.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agotcpv6: trivial formatting changes to send_(ack|reset)
Ilpo Järvinen [Thu, 9 Oct 2008 21:41:38 +0000]
tcpv6: trivial formatting changes to send_(ack|reset)

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agotcpv[46]: fix md5 pseudoheader address field ordering
Ilpo Järvinen [Thu, 9 Oct 2008 21:37:47 +0000]
tcpv[46]: fix md5 pseudoheader address field ordering

Maybe it's just me but I guess those md5 people made a mess
out of it by having *_md5_hash_* to use daddr, saddr order
instead of the one that is natural (and equal to what csum
functions use). For the segment were sending, the original
addresses are reversed so buff's saddr == skb's daddr and
vice-versa.

Maybe I can finally proceed with unification of some code
after fixing it first... :-)

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agor8169: fix early spinlock use
Francois Romieu [Thu, 9 Oct 2008 21:35:58 +0000]
r8169: fix early spinlock use

rtl8169_init_one
-> rtl_init_mac_address
   -> rtl_rar_set
      -> spin_lock_irq(&tp->lock);
[...]
-> spin_lock_init(&tp->lock);

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agosmsc95xx: fix usb usage on big endian platforms
Steve Glendinning [Thu, 9 Oct 2008 21:34:47 +0000]
smsc95xx: fix usb usage on big endian platforms

Fix warnings caught by David Miller on sparc64

Signed-off-by: Steve Glendinning <steve.glendinning@smsc.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agosctp: update SNMP statiscts when T5 timer expired.
Vlad Yasevich [Thu, 9 Oct 2008 21:33:26 +0000]
sctp: update SNMP statiscts when T5 timer expired.

The T5 timer is the timer for the over-all shutdown procedure.  If
this timer expires, then shutdown procedure has not completed and we
ABORT the association.  We should update SCTP_MIB_ABORTED and
SCTP_MIB_CURRESTAB  when aborting.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agosctp: Fix SNMP number of SCTP_MIB_ABORTED during violation handling.
Vlad Yasevich [Thu, 9 Oct 2008 21:33:01 +0000]
sctp: Fix SNMP number of SCTP_MIB_ABORTED during violation handling.

If ABORT chunks require authentication and a protocol violation
is triggered, we do not tear down the association.  Subsequently,
we should not increment SCTP_MIB_ABORTED.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agosctp: Fix the SNMP number of SCTP_MIB_CURRESTAB
Wei Yongjun [Thu, 9 Oct 2008 21:32:24 +0000]
sctp: Fix the SNMP number of SCTP_MIB_CURRESTAB

RFC3873 defined SCTP_MIB_CURRESTAB:
  sctpCurrEstab OBJECT-TYPE
    SYNTAX         Gauge32
    MAX-ACCESS     read-only
    STATUS         current
    DESCRIPTION
         "The number of associations for which the current state is
         either ESTABLISHED, SHUTDOWN-RECEIVED or SHUTDOWN-PENDING."
    REFERENCE
         "Section 4 in RFC2960 covers the SCTP   Association state
         diagram."

If the T4 RTO timer expires many times(timeout), the association will enter
CLOSED state, so we should dec the number of SCTP_MIB_CURRESTAB, not inc the
number of SCTP_MIB_CURRESTAB.

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agomyri10ge: remove over-paranoid queue number modulus
Brice Goglin [Thu, 9 Oct 2008 21:31:42 +0000]
myri10ge: remove over-paranoid queue number modulus

No need to modulus the queue number in ->hard_start_xmit() since the
core is going to do that for you modulus ->real_num_tx_queues.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoe1000: don't generate bad checksums for tcp packets with 0 csum
Dave Graham [Thu, 9 Oct 2008 21:29:26 +0000]
e1000: don't generate bad checksums for tcp packets with 0 csum

When offloading transmit checksums only, the driver was not
correctly configuring the hardware to handle the case of a zero
checksum.  For UDP the correct behavior is to leave it alone, but
for tcp the checksum must be changed from 0x0000 to 0xFFFF.  The
hardware takes care of this case but only if it is told the
packet is tcp.

same patch as e1000e

Signed-off-by: Dave Graham <david.graham@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoe1000e: don't generate bad checksums for tcp packets with 0 csum
Dave Graham [Thu, 9 Oct 2008 21:28:58 +0000]
e1000e: don't generate bad checksums for tcp packets with 0 csum

When offloading transmit checksums only, the driver was not
correctly configuring the hardware to handle the case of a zero
checksum.  For UDP the correct behavior is to leave it alone, but
for tcp the checksum must be changed from 0x0000 to 0xFFFF.  The
hardware takes care of this case but only if it is told the
packet is tcp.

Signed-off-by: Dave Graham <david.graham@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agobnx2: Update version to 1.8.1.
Michael Chan [Thu, 9 Oct 2008 19:27:06 +0000]
bnx2: Update version to 1.8.1.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agobnx2: Handle DMA mapping errors.
Benjamin Li [Thu, 9 Oct 2008 19:26:41 +0000]
bnx2: Handle DMA mapping errors.

Before, the driver would not care about the return codes from pci_map_*
functions.  This could be potentially dangerous if a mapping failed.
Now, we will check all pci_map_* calls.  On the transmit side, we switch
to use the new function skb_dma_map().  On the receive side, we add
pci_dma_mapping_error().

Signed-off-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agobnx2: Eliminate TSO header modifications.
Michael Chan [Thu, 9 Oct 2008 19:24:39 +0000]
bnx2: Eliminate TSO header modifications.

This is now possible with updated firmware.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agobnx2: Update 5706/5708 firmware.
Michael Chan [Thu, 9 Oct 2008 19:22:38 +0000]
bnx2: Update 5706/5708 firmware.

With this new firmware, the driver no longer has to modify the
TCP/IP header fields when transmitting TSO packets.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agobnx2: Check netif_running() in all ethtool operations.
Michael Chan [Thu, 9 Oct 2008 19:21:46 +0000]
bnx2: Check netif_running() in all ethtool operations.

We need to check netif_running() state in most ethtool operations
and properly handle the !netif_running() state where the chip is
in an uninitailzed state or low power state that may not accept
any MMIO.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agobnx2: Add bnx2_shutdown_chip().
Michael Chan [Thu, 9 Oct 2008 19:21:08 +0000]
bnx2: Add bnx2_shutdown_chip().

This logic is used in bnx2_close() and bnx2_suspend() and
so should be separated out into a separate function.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Benjamin Li <benli@broadcom.com>
Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoinet: Make tunnel RX/TX byte counters more consistent
Herbert Xu [Thu, 9 Oct 2008 19:03:17 +0000]
inet: Make tunnel RX/TX byte counters more consistent

This patch makes the RX/TX byte counters for IPIP, GRE and SIT more
consistent.  Previously we included the external IP headers on the
way out but not when the packet is inbound.

The new scheme is to count payload only in both directions.  For
IPIP and SIT this simply means the exclusion of the external IP
header.  For GRE this means that we exclude the GRE header as
well.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agogre: Add Transparent Ethernet Bridging
Herbert Xu [Thu, 9 Oct 2008 19:00:17 +0000]
gre: Add Transparent Ethernet Bridging

This patch adds support for Ethernet over GRE encapsulation.
This is exposed to user-space with a new link type of "gretap"
instead of "gre".  It will create an ARPHRD_ETHER device in
lieu of the usual ARPHRD_IPGRE.

Note that to preserver backwards compatibility all Transparent
Ethernet Bridging packets are passed to an ARPHRD_IPGRE tunnel
if its key matches and there is no ARPHRD_ETHER device whose
key matches more closely.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agogre: Add netlink interface
Herbert Xu [Thu, 9 Oct 2008 18:59:55 +0000]
gre: Add netlink interface

This patch adds a netlink interface that will eventually displace
the existing ioctl interface.  It utilises the elegant rtnl_link_ops
mechanism.

This also means that user-space no longer needs to rely on the
tunnel interface being of type GRE to identify GRE tunnels.  The
identification can now occur using rtnl_link_ops.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agogre: Move MTU setting out of ipgre_tunnel_bind_dev
Herbert Xu [Thu, 9 Oct 2008 18:59:32 +0000]
gre: Move MTU setting out of ipgre_tunnel_bind_dev

This patch moves the dev->mtu setting out of ipgre_tunnel_bind_dev.
This is in prepartion of using rtnl_link where we'll need to make
the MTU setting conditional on whether the user has supplied an
MTU.  This also requires the move of the ipgre_tunnel_bind_dev
call out of the dev->init function so that we can access the user
parameters later.

This patch also adds a check to prevent setting the MTU below
the minimum of 68.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agogre: Use needed_headroom
Herbert Xu [Thu, 9 Oct 2008 18:58:54 +0000]
gre: Use needed_headroom

Now that we have dev->needed_headroom, we can use it instead of
having a bogus dev->hard_header_len.  This also allows us to
include dev->hard_header_len in the MTU computation so that when
we do have a meaningful hard_harder_len in future it is included
automatically in figuring out the MTU.

Incidentally, this fixes a bug where we ignored the needed_headroom
field of the underlying device in calculating our own hard_header_len.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agophylib: two dynamic mii_bus allocation fallout fixes
Lennert Buytenhek [Thu, 9 Oct 2008 16:45:04 +0000]
phylib: two dynamic mii_bus allocation fallout fixes

1. arch/powerpc/platforms/pasemi/gpio_mdio.c also needs to be
   converted over to mdiobus_{alloc,free}().

2. drivers/net/phy/fixed.c used to embed a struct mii_bus into its
   struct fixed_mdio_bus and then use container_of() to go from the
   former to the latter.  Since mii bus structures are no longer
   embedded, we need to do something like use the mii bus private
   pointer to go from mii_bus to fixed_mdio_bus instead.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agojme: Fix warnings with CONFIG_PM disabled.
David S. Miller [Thu, 9 Oct 2008 02:54:31 +0000]
jme: Fix warnings with CONFIG_PM disabled.

drivers/net/jme.c:1598: warning: ‘jme_set_100m_half’ defined but not used
drivers/net/jme.c:1618: warning: ‘jme_wait_link’ defined but not used

Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agojme: Advances version number
Guo-Fu Tseng [Thu, 9 Oct 2008 02:50:44 +0000]
jme: Advances version number

Advances the driver version after modification.

Signed-off-by: Guo-Fu Tseng <cooldavid@cooldavid.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agojme: Faulty IRQ handle bug fix
akeemting [Thu, 9 Oct 2008 02:50:03 +0000]
jme: Faulty IRQ handle bug fix

Fix IRQ handle bug when interrupt mode.

The driver was incorrectly handled and returned IRQ_HANDLED
while the device is not generating the interrupt.
It happened due to faulty determination of interrupt status register.

Found by: "Ethan" <ethanhsiao@jmicron.com>
Fixed by: "akeemting" <akeem@jmicron.com>
Signed-off-by: Guo-Fu Tseng <cooldavid@cooldavid.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agojme: Added half-duplex mode and IPv6 RSS fix
Guo-Fu Tseng [Thu, 9 Oct 2008 02:48:58 +0000]
jme: Added half-duplex mode and IPv6 RSS fix

1. Set bit 5 of GPREG1 to 1 to enable hardware workaround for half-duplex
   mode. Which the MAC processor generates CRS/COL by itself instead of
   receive it from PHY processor.

2. Set bit 6 of GPREG1 to 1 to enable hardware workaround that masks the
   MAC processor working right while calculating IPv6 RSS in 10/100
   mode.

3. Group the workaround codes all together.

Signed-off-by: Guo-Fu Tseng <cooldavid@cooldavid.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agocxgb3: Add 1G fiber support
Divy Le Ray [Thu, 9 Oct 2008 00:40:28 +0000]
cxgb3: Add 1G fiber support

Add support for 1G optical Vitesse PHY.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agocxgb3: Support for Aeluros 2005 PHY
Divy Le Ray [Thu, 9 Oct 2008 00:40:07 +0000]
cxgb3: Support for Aeluros 2005 PHY

Add support for SR PHY.
Auto-detect phy module type, and report type changes.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agocxgb3: commnonize LASI phy code
Divy Le Ray [Thu, 9 Oct 2008 00:39:31 +0000]
cxgb3: commnonize LASI phy code

Add generic code to manage interrupt driven PHYs.
Do not reset the phy after link parameters update,
the new values might get lost.
Return early from link change notification
when the link parameters remain unchanged.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agocxgb3: More flexible support for PHY interrupts.
Divy Le Ray [Thu, 9 Oct 2008 00:39:00 +0000]
cxgb3: More flexible support for PHY interrupts.

Do not require PHY interrupts to be connected to GPIs in ascending order.
Base interrupt availability both on PHYs supporting them and on GPIs being
hooked up.  Allows boards to specify interrupt GPIs though the PHYs don't
use them.

Remove spurious PHY interrupts due to clearing T3DBG interrupts before
setting their polarity.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agocxgb3: simplify port type struct and usage
Divy Le Ray [Thu, 9 Oct 2008 00:38:29 +0000]
cxgb3: simplify port type struct and usage

Second step in overall phy layer reorganization.
Clean up the port_type_info structure.
Support coextistence of clause 22 and clause 45 MDIO devices.
Select the type of MDIO transaction on a per transaction basis.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agocxgb3: allow for PHY reset status
Divy Le Ray [Thu, 9 Oct 2008 00:38:01 +0000]
cxgb3: allow for PHY reset status

First step towards overall PHY layering re-organization.
Allow a status return when a PHY is reset.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agocxgb3: Allocate multiqueues at init time
Divy Le Ray [Thu, 9 Oct 2008 00:37:33 +0000]
cxgb3: Allocate multiqueues at init time

Allocate a queue set per core, up to the maximum of available qsets.
Share the queue sets on multi port adapters.
Rename MSI-X interrupt vectors ethX-N, N being the queue set number.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agocxgb3: reset the adapter on fatal error
Divy Le Ray [Thu, 9 Oct 2008 00:36:03 +0000]
cxgb3: reset the adapter on fatal error

when a fatal error occurs, bring ports down, reset the chip,
and bring ports back up.

Factorize code used for both EEH and fatal error recovery.
Fix timer usage when bringing up/resetting sge queue sets.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agodsa: Need to select PHYLIB.
David S. Miller [Thu, 9 Oct 2008 00:33:01 +0000]
dsa: Need to select PHYLIB.

Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agodsa: add support for the Marvell 88E6060 switch chip
Lennert Buytenhek [Tue, 7 Oct 2008 13:46:22 +0000]
dsa: add support for the Marvell 88E6060 switch chip

Add support for the Marvell 88E6060 switch chip.  This chip only
supports the Header and Trailer tagging formats, and we use it in
Trailer mode since that mode is slightly easier to handle than
Header mode.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Tested-by: Byron Bradley <byron.bbradley@gmail.com>
Tested-by: Tim Ellis <tim.ellis@mac.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agodsa: add support for Trailer tagging format
Lennert Buytenhek [Tue, 7 Oct 2008 13:46:07 +0000]
dsa: add support for Trailer tagging format

This adds support for the Trailer switch tagging format.  This is
another tagging that doesn't explicitly mark tagged packets with a
distinct ethertype, so that we need to add a similar hack in the
receive path as for the Original DSA tagging format.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Tested-by: Byron Bradley <byron.bbradley@gmail.com>
Tested-by: Tim Ellis <tim.ellis@mac.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agodsa: add support for the Marvell 88E6131 switch chip
Lennert Buytenhek [Tue, 7 Oct 2008 13:45:18 +0000]
dsa: add support for the Marvell 88E6131 switch chip

Add support for the Marvell 88E6131 switch chip.  This chip only
supports the original (ethertype-less) DSA tagging format.

On the 88E6131, there is a PHY Polling Unit (PPU) which has exclusive
access to each of the PHYs's MII management registers.  If we want to
talk to the PHYs from software, we have to disable the PPU and wait
for it to complete its current transaction before we can do so, and we
need to re-enable the PPU afterwards to make sure that the switch will
notice changes in link state and speed on the individual ports as they
occur.

Since disabling the PPU is rather slow, and since MII management
accesses are typically done in bursts, this patch keeps the PPU disabled
for 10ms after a software access completes.  This makes handling the
PPU slightly more complex, but speeds up something like running ethtool
on one of the switch slave interfaces from ~300ms to ~30ms on typical
hardware.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Tested-by: Nicolas Pitre <nico@marvell.com>
Tested-by: Peter van Valderen <linux@ddcrew.com>
Tested-by: Dirk Teurlings <dirk@upexia.nl>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agodsa: add support for original DSA tagging format
Lennert Buytenhek [Tue, 7 Oct 2008 13:45:02 +0000]
dsa: add support for original DSA tagging format

Most of the DSA switches currently in the field do not support the
Ethertype DSA tagging format that one of the previous patches added
support for, but only the original DSA tagging format.

The original DSA tagging format carries the same information as the
Ethertype DSA tagging format, but with the difference that it does not
have an ethertype field.  In other words, when receiving a packet that
is tagged with an original DSA tag, there is no way of telling in
eth_type_trans() that this packet is in fact a DSA-tagged packet.

This patch adds a hook into eth_type_trans() which is only compiled in
if support for a switch chip that doesn't support Ethertype DSA is
selected, and which checks whether there is a DSA switch driver
instance attached to this network device which uses the old tag format.
If so, it sets the protocol field to ETH_P_DSA without looking at the
packet, so that the packet ends up in the right place.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Tested-by: Nicolas Pitre <nico@marvell.com>
Tested-by: Peter van Valderen <linux@ddcrew.com>
Tested-by: Dirk Teurlings <dirk@upexia.nl>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agonet: Distributed Switch Architecture protocol support
Lennert Buytenhek [Tue, 7 Oct 2008 13:44:02 +0000]
net: Distributed Switch Architecture protocol support

Distributed Switch Architecture is a protocol for managing hardware
switch chips.  It consists of a set of MII management registers and
commands to configure the switch, and an ethernet header format to
signal which of the ports of the switch a packet was received from
or is intended to be sent to.

The switches that this driver supports are typically embedded in
access points and routers, and a typical setup with a DSA switch
looks something like this:

+-----------+       +-----------+
|           | RGMII |           |
|           +-------+           +------ 1000baseT MDI ("WAN")
|           |       |  6-port   +------ 1000baseT MDI ("LAN1")
|    CPU    |       |  ethernet +------ 1000baseT MDI ("LAN2")
|           |MIImgmt|  switch   +------ 1000baseT MDI ("LAN3")
|           +-------+  w/5 PHYs +------ 1000baseT MDI ("LAN4")
|           |       |           |
+-----------+       +-----------+

The switch driver presents each port on the switch as a separate
network interface to Linux, polls the switch to maintain software
link state of those ports, forwards MII management interface
accesses to those network interfaces (e.g. as done by ethtool) to
the switch, and exposes the switch's hardware statistics counters
via the appropriate Linux kernel interfaces.

This initial patch supports the MII management interface register
layout of the Marvell 88E6123, 88E6161 and 88E6165 switch chips, and
supports the "Ethertype DSA" packet tagging format.

(There is no officially registered ethertype for the Ethertype DSA
packet format, so we just grab a random one.  The ethertype to use
is programmed into the switch, and the switch driver uses the value
of ETH_P_EDSA for this, so this define can be changed at any time in
the future if the one we chose is allocated to another protocol or
if Ethertype DSA gets its own officially registered ethertype, and
everything will continue to work.)

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Tested-by: Nicolas Pitre <nico@marvell.com>
Tested-by: Byron Bradley <byron.bbradley@gmail.com>
Tested-by: Tim Ellis <tim.ellis@mac.com>
Tested-by: Peter van Valderen <linux@ddcrew.com>
Tested-by: Dirk Teurlings <dirk@upexia.nl>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years ago8139too: move wmb before TX DMA start
Andreas Oberritter [Wed, 8 Oct 2008 06:14:16 +0000]
8139too: move wmb before TX DMA start

The write barrier should be used before starting a DMA transfer. This fixes
a problem, where almost all packets received on another machine had garbled
content. Tested with an RTL8100C on a MIPS machine.

Signed-off-by: Andreas Oberritter <obi@linuxtv.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agonetdev: uniquify the tx_timeout name
Arjan van de Ven [Sun, 5 Oct 2008 07:35:05 +0000]
netdev: uniquify the tx_timeout name

there's several drivers that have use "tx_timeout" for the .. tx
timeout function.  All fine with that, they're static, however for
doing stats on how often which driver hits the timeout it's a tad
unfortunate. The patch below gives the ones I found in the
kerneloops.org database unique names.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agodrivers/net/usb/Makefile - shrink logspam
David Brownell [Sat, 4 Oct 2008 22:11:12 +0000]
drivers/net/usb/Makefile - shrink logspam

When building with CONFIG_USB_DEBUG, don't create logspam from
the USB networking drivers.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agor8169: WoL fixes, part 2.
Bruno Prémont [Thu, 9 Oct 2008 00:06:25 +0000]
r8169: WoL fixes, part 2.

Since recent kernel (2.6.26 or 2.6.27) the PCI wakeup functions are
influenced by generic device ability and configuration when enabling
PCI-device triggered wake-up.

This patch causes WoL setting to enable/disable device's wish to
be permitted to wake-up the host when changing WoL options and
also during device probing.

Without this patch one has write 'enabled' to
  /sys/bus/pci/devices/0000:02:08.0/power/wakeup

Signed-off-by: Bruno Prémont <bonbons@linux-vserver.org>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agor8169: WoL fixes, part 1.
Bruno Prémont [Thu, 9 Oct 2008 00:05:03 +0000]
r8169: WoL fixes, part 1.

When probing the chip and handling it's power management settings
also remember wether WoL feature is enabled.

Without this patch one has to call ethtool to change WoL settings
for this flag to be set and any WoL being enabled on suspend to
RAM.

Signed-off-by: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agogianfar: Create net device with carrier down
Trent Piepho [Thu, 2 Oct 2008 11:12:24 +0000]
gianfar: Create net device with carrier down

The device's carrier status is controlled via the functions
netif_carrier_on() and netif_carrier_off().  These set or clear a bit
indicating the carrier (aka lower level link) is down, and if the state
changed, they fire off a routing netlink event.

Add a call to netif_carrier_off() before register_netdev() so that the
newly created device will be set to carrier down.  Then when the carrier
comes up for the first time, a netlink event will be generated, as the
carrier changed from down to up.  Otherwise the initial carrier up will
appear to be changing the status from up to up, and so no event is
generated since that's not a change.

Signed-off-by: Trent Piepho <tpiepho@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agomv643xx_eth: include linux/ip.h to fix build
Lennert Buytenhek [Thu, 2 Oct 2008 00:52:39 +0000]
mv643xx_eth: include linux/ip.h to fix build

mv643xx_eth uses ip_hdr() (defined in linux/ip.h), but relied on
another header file to include the needed header file indirectly.
In latest net-next this indirect include chain is gone, so the
driver fails to build.  Include linux/ip.h explicitly to fix this.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agolib8390: Fix locking in ei_poll (poll controller)
Jarek Poplawski [Tue, 30 Sep 2008 20:58:25 +0000]
lib8390: Fix locking in ei_poll (poll controller)

This lockdep warning:
  =================================
  [ INFO: inconsistent lock state ]
  2.6.27-rc7 #3
  ---------------------------------
  inconsistent {in-softirq-W} -> {softirq-on-W} usage.
  syslogd/2474 [HC0[0]:SC0[0]:HE1:SE1] takes:
   (_xmit_ETHER#2){-+..}, at: [<c0265562>] netpoll_send_skb+0x132/0x190
  ...

is caused by unconditional local_irq_disable()/local_irq_enable() in
disable_irq_lockdep()/enable_irq_lockdep() used by __ei_poll(). Since
netconsole/netpoll always calls dev->poll_controller() with local irqs
disabled, disable_irq()/enable_irq() instead is safe and enough (like
e.g. in 3c509 or 8139xx drivers).

Reported-and-tested-by: Bernard Pidoux F6BVP <f6bvp@free.fr>
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoixgb: fix bug when freeing resources
Brandeburg, Jesse [Tue, 30 Sep 2008 13:08:48 +0000]
ixgb: fix bug when freeing resources

It was pointed out by Breno Leitao <leitao@linux.vnet.ibm.com> that
ixgb would crash on PPC when an IOMMU was in use, if change_mtu was
called.

It appears to be a pretty simple issue in the driver that wasn't discovered
because most systems don't run with an IOMMU.  The driver needs to only unmap
buffers that are mapped (duh).

CC: Breno Leitao <leitao@linux.vnet.ibm.com>

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoe1000e: release hw semaphore after successfully writing EEPROM
Arthur Jones [Mon, 29 Sep 2008 11:51:09 +0000]
e1000e: release hw semaphore after successfully writing EEPROM

Since e1000e has been existance in linux-2.6, we've
never released the hardware semaphore after a successful
write to the SPI EEPROM.  I guess we don't write to
SPI EEPROM much -- but those few of us that do appreciate
it when we can later read from the EEPROM without having
to reboot.

Found-by: Nick Van Fossen <Nick.VanFossen@riverbed.com>
Signed-off-by: Arthur Jones <ajones@riverbed.com>
Reviewed-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agonet: Add SMSC LAN9500 USB2.0 10/100 ethernet adapter driver
Steve Glendinning [Thu, 2 Oct 2008 05:27:57 +0000]
net: Add SMSC LAN9500 USB2.0 10/100 ethernet adapter driver

Attached is a driver for SMSC's LAN9500 USB2.0 10/100 ethernet
adapter.

Signed-off-by: Steve Glendinning <steve.glendinning@smsc.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agophylib: add mdiobus_{read,write}
Lennert Buytenhek [Mon, 29 Sep 2008 17:12:35 +0000]
phylib: add mdiobus_{read,write}

Add mdiobus_{read,write} routines to allow direct reading/writing
of registers on an mii bus without having to go through the PHY
abstraction, and make phy_{read,write} use these primitives.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agophylib: give mdio buses a device tree presence
Lennert Buytenhek [Wed, 8 Oct 2008 23:33:40 +0000]
phylib: give mdio buses a device tree presence

Introduce the mdio_bus class, and give each 'struct mii_bus' its own
'struct device', so that mii_bus objects are represented in the device
tree and can be found by querying the device tree.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Acked-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agophylib: move to dynamic allocation of struct mii_bus
Lennert Buytenhek [Wed, 8 Oct 2008 23:29:57 +0000]
phylib: move to dynamic allocation of struct mii_bus

This patch introduces mdiobus_alloc() and mdiobus_free(), and
makes all mdio bus drivers use these functions to allocate their
struct mii_bus'es dynamically.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Andy Fleming <afleming@freescale.com>

12 years agophylib: rename mii_bus::dev to mii_bus::parent
Lennert Buytenhek [Wed, 1 Oct 2008 15:41:33 +0000]
phylib: rename mii_bus::dev to mii_bus::parent

In preparation of giving mii_bus objects a device tree presence of
their own, rename struct mii_bus's ->dev argument to ->parent, since
having a 'struct device *dev' that points to our parent device
conflicts with introducing a 'struct device dev' representing our own
device.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Andy Fleming <afleming@freescale.com>

12 years agomyri10ge: add multiqueue TX
Brice Goglin [Sun, 28 Sep 2008 15:34:21 +0000]
myri10ge: add multiqueue TX

Add multiqueue TX support to myri10ge.

[ Removed reference to deprecated CONFIG_NETDEVICES_MULTIQUEUE and
  NETIF_F_MULTI_QUEUE ]

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoatl1: update introductory comments
Jay Cliburn [Sat, 27 Sep 2008 04:17:23 +0000]
atl1: update introductory comments

Update the driver's introductory comments.

Signed-off-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoatl1: remove EXPERIMENTAL label
Jay Cliburn [Sat, 27 Sep 2008 04:17:22 +0000]
atl1: remove EXPERIMENTAL label

Remove the EXPERIMENTAL label from the atl1 driver and change the vendor
name to include Attansic's successor, Atheros.  We'll leave Attansic in
the name since Attansic's PCI ID (1969) is encoded in the PCI config and
is what users encounter on their systems.

Signed-off-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoatl1: remove LLTX
Jay Cliburn [Sat, 27 Sep 2008 04:17:21 +0000]
atl1: remove LLTX

NETIF_F_LLTX is deprecated. Remove private TX locking from the driver
and remove the NETIF_F_LLTX feature flag.

Signed-off-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoatl1: fix transmit timeout bug
Jay Cliburn [Sat, 27 Sep 2008 04:17:20 +0000]
atl1: fix transmit timeout bug

See http://marc.info/?l=linux-netdev&m=121931988219314&w=2

Stop the queue and turn off carrier to prevent transmit timeouts
when the cable is unplugged/replugged.

Signed-off-by: Jay Cliburn <jacliburn@bellsouth.net>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoskge: Fix skge_set_ring_param() losing error return
Wang Chen [Fri, 26 Sep 2008 08:20:32 +0000]
skge: Fix skge_set_ring_param() losing error return

The error return is useful to caller, driver shouldn't miss it.

Signed-off-by: Wang Chen <wangchen@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years ago[10/21] driver/net/skge.c: restart the interface when it's options or
Xiaoming.Zhang [Thu, 25 Sep 2008 20:28:05 +0000]
[10/21] driver/net/skge.c: restart the interface when it's options or
pauseparam is set

On Wednesday 24 September 2008 07:47, Stephen Hemminger wrote:
> On Mon, 22 Sep 2008 14:52:17 -0700
>
> akpm@linux-foundation.org wrote:
> > From: "Xiaoming.Zhang" <Xiaoming.Zhang@resilience.com>
> >
> > We have an issue of the skge driver: The card won't work when it's
> > options are changed.  Here's the hardware info:
> >
> > # lspci -v
> > 05:04.0 Ethernet controller: Marvell Technology Group Ltd. 88E8001
> > Gigabit Ethernet Controller (rev 13) Subsystem: Marvell Technology Group
> > Ltd. Marvell RDK-8001 Flags: bus master, 66MHz, medium devsel, latency
> > 32, IRQ 16 Memory at d042c000 (32-bit, non-prefetchable) [size=16K] I/O
> > ports at d000 [size=256]
> >         [virtual] Expansion ROM at 20400000 [disabled] [size=128K]
> >         Capabilities: [48] Power Management version 2
> >         Capabilities: [50] Vital Product Data
> >
> > The happens in both Linux-2.6.26(skge version 1.23) and RHEL5.2(skge
> > version 1.6).
> >
> > For example, at first it is set to "speed 1000 duplex full auto-neg on"
> > and it works, then run
> >
> >        ethtool -s <ethx> autoneg off
> > or     ethtool -s <ethx> speed 100 duplex full autoneg off
> >
> > Then it will stop working. After that if we restart the interface:
> >
> >  ifconifg <ethx> down
> >  ifconfig <ethx> up
> >
> > It will work again. And `ethtool -A' has the same issue.
> >
> > So we think after setting the options, the interface should be restarted.
> >
> > Signed-off-by: Zhang Xiaoming <xiaoming.zhang@resilience.com>
> > Cc: Stephen Hemminger <shemminger@vyatta.com>
> > Cc: Jeff Garzik <jeff@garzik.org>
> > Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> > ---
> >
> >  drivers/net/skge.c |   12 ++++++++----
> >  1 file changed, 8 insertions(+), 4 deletions(-)
> >
> > diff -puN
> > drivers/net/skge.c~driver-net-skgec-restart-the-interface-when-its-option
> >s-or-pauseparam-is-set drivers/net/skge.c ---
> > a/drivers/net/skge.c~driver-net-skgec-restart-the-interface-when-its-opti
> >ons-or-pauseparam-is-set +++ a/drivers/net/skge.c
> > @@ -353,8 +353,10 @@ static int skge_set_settings(struct net_
> >   skge->autoneg = ecmd->autoneg;
> >   skge->advertising = ecmd->advertising;
> >
> > - if (netif_running(dev))
> > - skge_phy_reset(skge);
> > + if (netif_running(dev)) {
> > + skge_down(dev);
> > + skge_up(dev);
> > + }
> >
> >   return (0);
> >  }
> > @@ -595,8 +597,10 @@ static int skge_set_pauseparam(struct ne
> >   skge->flow_control = FLOW_MODE_NONE;
> >   }
> >
> > - if (netif_running(dev))
> > - skge_phy_reset(skge);
> > + if (netif_running(dev)) {
> > + skge_down(dev);
> > + skge_up(dev);
> > + }
> >
> >   return 0;
> >  }
>
> Since skge_up can fail because of out of memory, this code needs to
> check the return value. And then if it fails the "limbo state" needs
> to be handled in skge_down.

How about like this? It is tested.

Thank you.

Signed-off-by: Zhang Xiaoming <xiaoming.zhang@resilience.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years ago[2.6.28,1/1] cxgb3 - fix race in EEH
Divy Le Ray [Thu, 25 Sep 2008 14:05:28 +0000]
[2.6.28,1/1] cxgb3 - fix race in EEH

A SGE queue set timer might access registers while in EEH recovery,
triggering an EEH error loop. Stop all timers early in EEH process.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agovia-velocity: Fix warnings on sparc64.
Francois Romieu [Wed, 8 Oct 2008 22:55:15 +0000]
via-velocity: Fix warnings on sparc64.

As reported by Meelis Roos.

Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agonet: remove LLTX in atl2 driver
Kevin Hao [Thu, 25 Sep 2008 16:20:11 +0000]
net: remove LLTX in atl2 driver

When NETIF_F_LLTX is set, the atlx driver will use a private lock.
But in recent kernels this implementation seems redundant and
can cause problems where AF_PACKET sees things twice. Since
NETIF_F_LLTX is marked as deprecated and shouldn't be used in
new driver, this patch removes NETIF_F_LLTX and adds a mmiowb
before sending packet. I have tested this driver on a Eee PC.
It works well.

Signed-off-by: Kevin Hao <kexin.hao@windriver.com>
Acked-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agophylib: Wake PHY state machine on interrupt
Trent Piepho [Wed, 8 Oct 2008 22:46:57 +0000]
phylib: Wake PHY state machine on interrupt

This way the phy layer will respond to a change in phy state immediately,
instead of up to one second later when the state machine timer runs.

Signed-off-by: Trent Piepho <tpiepho@freescale.com>
Acked-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoPHY: Avoid unnecessary aneg restarts
Trent Piepho [Wed, 24 Sep 2008 10:55:46 +0000]
PHY: Avoid unnecessary aneg restarts

The PHY's aneg is configured and restarted whenever the link is brought up,
e.g. when DHCP is started after the kernel has booted.  This can take the
link down for several seconds while auto-negotiation is redone.

If the advertised features haven't changed, then it shouldn't be necessary
to bring down the link and start auto-negotiation over again.

genphy_config_advert() is enhanced to return 0 when the advertised features
haven't been changed and >0 when they have been.

genphy_config_aneg() then uses this information to not call
genphy_restart_aneg() if there has been no change.

Signed-off-by: Trent Piepho <tpiepho@freescale.com>
Acked-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agor8169: read MAC address from EEPROM on init
Ivan Vecera [Tue, 23 Sep 2008 22:46:29 +0000]
r8169: read MAC address from EEPROM on init

Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agonet: add net poll support for atl2 driver
Kevin Hao [Fri, 19 Sep 2008 21:56:44 +0000]
net: add net poll support for atl2 driver

Add netconsole support for Atheros L2 10/100 network device.

Signed-off-by: Kevin Hao <kexin.hao@windriver.com>
Acked-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoibm_newemac: Fix EMAC soft reset on 460EX/GT
Victor Gallardo [Thu, 18 Sep 2008 12:41:26 +0000]
ibm_newemac: Fix EMAC soft reset on 460EX/GT

This patch fixes EMAC soft reset on 460EX/GT when no external clock is
available.

Signed-off-by: Victor Gallardo <vgallardo@amcc.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agophylib: phy_mii_ioctl() fixes
Lennert Buytenhek [Thu, 18 Sep 2008 03:06:52 +0000]
phylib: phy_mii_ioctl() fixes

Make the SIOCGMIIPHY case fall through properly (it is supposed
to not only return the ID of the default PHY but also to read from
that PHY), and make phy_mii_ioctl() return the same error code as
generic_mii_ioctl() in case of an unsupported operation.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Acked-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoMerge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
David S. Miller [Wed, 8 Oct 2008 21:56:41 +0000]
Merge branch 'master' of /linux/kernel/git/davem/net-2.6

Conflicts:

drivers/net/e1000e/ich8lan.c
drivers/net/e1000e/netdev.c

12 years agoipvs: Remove stray file left over from ipvs move
Sven Wegener [Wed, 8 Oct 2008 21:41:35 +0000]
ipvs: Remove stray file left over from ipvs move

Commit cb7f6a7b716e801097b564dec3ccb58d330aef56 ("IPVS: Move IPVS to
net/netfilter/ipvs") has left a stray file in the old location of ipvs.

Signed-off-by: Sven Wegener <sven.wegener@stealer.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agotcpv6: fix option space offsets with md5
Ilpo Järvinen [Wed, 8 Oct 2008 21:36:33 +0000]
tcpv6: fix option space offsets with md5

More breakage :-), part of timestamps just were previously
overwritten.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoMerge branch 'lvs-next-2.6' of git://git.kernel.org/pub/scm/linux/kernel/git/horms...
David S. Miller [Wed, 8 Oct 2008 21:26:36 +0000]
Merge branch 'lvs-next-2.6' of git://git./linux/kernel/git/horms/lvs-2.6

Conflicts:

net/netfilter/Kconfig

12 years agosctp: shrink sctp_tsnmap some more by removing gabs array
Vlad Yasevich [Wed, 8 Oct 2008 21:19:01 +0000]
sctp: shrink sctp_tsnmap some more by removing gabs array

The gabs array in the sctp_tsnmap structure is only used
in one place, sctp_make_sack().  As such, carrying the
array around in the sctp_tsnmap and thus directly in
the sctp_association is rather pointless since most
of the time it's just taking up space.  Now, let
sctp_make_sack create and populate it and then throw
it away when it's done.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agosctp: Rework the tsn map to use generic bitmap.
Vlad Yasevich [Wed, 8 Oct 2008 21:18:39 +0000]
sctp: Rework the tsn map to use generic bitmap.

The tsn map currently use is 4K large and is stuck inside
the sctp_association structure making memory references REALLY
expensive.  What we really need is at most 4K worth of bits
so the biggest map we would have is 512 bytes.   Also, the
map is only really usefull when we have gaps to store and
report.  As such, starting with minimal map of say 32 TSNs (bits)
should be enough for normal low-loss operations.  We can grow
the map by some multiple of 32 along with some extra room any
time we receive the TSN which would put us outside of the map
boundry.  As we close gaps, we can shift the map to rebase
it on the latest TSN we've seen.  This saves 4088 bytes per
association just in the map alone along savings from the now
unnecessary structure members.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoinet: cleanup of local_port_range
Eric Dumazet [Wed, 8 Oct 2008 21:18:04 +0000]
inet: cleanup of local_port_range

I noticed sysctl_local_port_range[] and its associated seqlock
sysctl_local_port_range_lock were on separate cache lines.
Moreover, sysctl_local_port_range[] was close to unrelated
variables, highly modified, leading to cache misses.

Moving these two variables in a structure can help data
locality and moving this structure to read_mostly section
helps sharing of this data among cpus.

Cleanup of extern declarations (moved in include file where
they belong), and use of inet_get_local_port_range()
accessor instead of direct access to ports values.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoudp: Improve port randomization
Eric Dumazet [Wed, 8 Oct 2008 18:44:17 +0000]
udp: Improve port randomization

Current UDP port allocation is suboptimal.
We select the shortest chain to chose a port (out of 512)
that will hash in this shortest chain.

First, it can lead to give not so ramdom ports and ease
give attackers more opportunities to break the system.

Second, it can consume a lot of CPU to scan all table
in order to find the shortest chain.

Third, in some pathological cases we can fail to find
a free port even if they are plenty of them.

This patch zap the search for a short chain and only
use one random seed. Problem of getting long chains
should be addressed in another way, since we can
obtain long chains with non random ports.

Based on a report and patch from Vitaly Mayatskikh

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Wed, 8 Oct 2008 18:41:10 +0000]
Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] Sibyte: Register PIO PATA device only for Swarm and Litte Sur

12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Wed, 8 Oct 2008 18:40:19 +0000]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  tcp: Fix tcp_hybla zero congestion window growth with small rho and large cwnd.
  net: Fix netdev_run_todo dead-lock
  tcp: Fix possible double-ack w/ user dma
  net: only invoke dev->change_rx_flags when device is UP
  netrom: Fix sock_orphan() use in nr_release
  ax25: Quick fix for making sure unaccepted sockets get destroyed.
  Revert "ax25: Fix std timer socket destroy handling."
  [Bluetooth] Add reset quirk for A-Link BlueUSB21 dongle
  [Bluetooth] Add reset quirk for new Targus and Belkin dongles
  [Bluetooth] Fix double frees on error paths of btusb and bpa10x drivers

12 years agopkt_sched: Update qdisc requeue stats in dev_requeue_skb()
Jarek Poplawski [Wed, 8 Oct 2008 18:36:22 +0000]
pkt_sched: Update qdisc requeue stats in dev_requeue_skb()

After the last change of requeuing there is no info about such
incidents in tc stats. This patch updates the counter, but we should
consider this should differ from previous stats because of additional
checks preventing to repeat this. On the other hand, previous stats
didn't include requeuing of gso_segmented skbs.

Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agotcp: fix length used for checksum in a reset
Ilpo Järvinen [Wed, 8 Oct 2008 18:34:06 +0000]
tcp: fix length used for checksum in a reset

While looking for some common code I came across difference
in checksum calculation between tcp_v6_send_(reset|ack) I
couldn't explain. I checked both v4 and v6 and found out that
both seem to have the same "feature". I couldn't find anything
in rfc nor anywhere else which would state that md5 option
should be ignored like it was in case of reset so I came to
a conclusion that this is probably a genuine bug. I suspect
that addition of md5 just was fooled by the excessive
copy-paste code in those functions and the reset part was
never tested well enough to find out the problem.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years ago[MIPS] Sibyte: Register PIO PATA device only for Swarm and Litte Sur
Ralf Baechle [Wed, 8 Oct 2008 18:14:33 +0000]
[MIPS] Sibyte: Register PIO PATA device only for Swarm and Litte Sur

Symbol name spaghetti which is too complicated to cleanup on this stage
of the release cycle breaks the build on BCM1480 platforms.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

12 years agoipv6: remove unused not init_ipv6_mibs/cleanup_ipv6_mibs
Denis V. Lunev [Wed, 8 Oct 2008 17:36:24 +0000]
ipv6: remove unused not init_ipv6_mibs/cleanup_ipv6_mibs

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoipv6: making ip and icmp statistics per/namespace
Denis V. Lunev [Wed, 8 Oct 2008 17:36:03 +0000]
ipv6: making ip and icmp statistics per/namespace

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoipv6: added net argument to _DEVINC/_DEVADD
Denis V. Lunev [Wed, 8 Oct 2008 17:35:11 +0000]
ipv6: added net argument to _DEVINC/_DEVADD

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoipv6: added net argument to ICMP6MSGIN_INC_STATS_BH
Denis V. Lunev [Wed, 8 Oct 2008 17:34:54 +0000]
ipv6: added net argument to ICMP6MSGIN_INC_STATS_BH

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>