openwrt/staging/blogic.git
9 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Thu, 13 Aug 2015 23:23:11 +0000 (16:23 -0700)]
Merge git://git./linux/kernel/git/davem/net

Conflicts:
drivers/net/ethernet/cavium/Kconfig

The cavium conflict was overlapping dependency
changes.

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'dm-4.2-fixes-5' of git://git.kernel.org/pub/scm/linux/kernel/git/device...
Linus Torvalds [Thu, 13 Aug 2015 20:52:46 +0000 (13:52 -0700)]
Merge tag 'dm-4.2-fixes-5' of git://git./linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - two stable fixes for corruption seen in a snapshot of thinp metadata;
   metadata snapshots aren't widely used but help provide a consistent
   view of the metadata associated with an active thin-pool.

 - a dm-cache fix for the 4.2 "default" policy switch from "mq" to "smq"

* tag 'dm-4.2-fixes-5' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm cache policy smq: move 'dm-cache-default' module alias to SMQ
  dm btree: add ref counting ops for the leaves of top level btrees
  dm thin metadata: delete btrees when releasing metadata snapshot

9 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-block
Linus Torvalds [Thu, 13 Aug 2015 20:44:32 +0000 (13:44 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-block

Pull xen block driver fixes from Jens Axboe:
 "A few small bug fixes for xen-blk{front,back} that have been sitting
  over my vacation"

* 'for-linus' of git://git.kernel.dk/linux-block:
  xen-blkback: replace work_pending with work_busy in purge_persistent_gnt()
  xen-blkfront: don't add indirect pages to list when !feature_persistent
  xen-blkfront: introduce blkfront_gather_backend_features()

9 years agoMerge tag 'for-linus-4.2-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 13 Aug 2015 20:36:22 +0000 (13:36 -0700)]
Merge tag 'for-linus-4.2-rc6-tag' of git://git./linux/kernel/git/xen/tip

Pull xen bug fixes from David Vrabel:

 - revert a fix from 4.2-rc5 that was causing lots of WARNING spam.

 - fix a memory leak affecting backends in HVM guests.

 - fix PV domU hang with certain configurations.

* tag 'for-linus-4.2-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  xen/xenbus: Don't leak memory when unmapping the ring on HVM backend
  Revert "xen/events/fifo: Handle linked events when closing a port"
  x86/xen: build "Xen PV" APIC driver for domU as well

9 years agoRevert x86 sigcontext cleanups
Linus Torvalds [Thu, 13 Aug 2015 15:25:20 +0000 (08:25 -0700)]
Revert x86 sigcontext cleanups

This reverts commits 9a036b93a344 ("x86/signal/64: Remove 'fs' and 'gs'
from sigcontext") and c6f2062935c8 ("x86/signal/64: Fix SS handling for
signals delivered to 64-bit programs").

They were cleanups, but they break dosemu by changing the signal return
behavior (and removing 'fs' and 'gs' from the sigcontext struct - while
not actually changing any behavior - causes build problems).

Reported-and-tested-by: Stas Sergeev <stsp@list.ru>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Thu, 13 Aug 2015 17:46:39 +0000 (10:46 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Workaround hw bug when acquiring PCI bos ownership of iwlwifi
    devices, from Emmanuel Grumbach.

 2) Falling back to vmalloc in conntrack should not emit a warning, from
    Pablo Neira Ayuso.

 3) Fix NULL deref when rtlwifi driver is used as an AP, from Luis
    Felipe Dominguez Vega.

 4) Rocker doesn't free netdev on device removal, from Ido Schimmel.

 5) UDP multicast early sock demux has route handling races, from Eric
    Dumazet.

 6) Fix L4 checksum handling in openvswitch, from Glenn Griffin.

 7) Fix use-after-free in skb_set_peeked, from Herbert Xu.

 8) Don't advertize NETIF_F_FRAGLIST in virtio_net driver, this can lead
    to fraglists longer than the driver can support.  From Jason Wang.

 9) Fix mlx5 on non-4k-pagesize systems, from Carol L Soto.

10) Fix interrupt storm in bna driver, from Ivan Vecera.

11) Don't propagate -EBUSY from netlink_insert(), from Daniel Borkmann.

12) Fix inet request sock leak, from Eric Dumazet.

13) Fix TX interrupt masking and marking in TX descriptors of fs_enet
    driver, from LEROY Christophe.

14) Get rid of rule optimizer in gianfar driver, it's buggy and unlikely
    to get fixed any time soon.  From Jakub Kicinski

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (61 commits)
  cosa: missing error code on failure in probe()
  gianfar: remove faulty filer optimizer
  gianfar: correct list membership accounting
  gianfar: correct filer table writing
  bonding: Gratuitous ARP gets dropped when first slave added
  net: dsa: Do not override PHY interface if already configured
  net: fs_enet: mask interrupts for TX partial frames.
  net: fs_enet: explicitly remove I flag on TX partial frames
  inet: fix possible request socket leak
  inet: fix races with reqsk timers
  mkiss: Fix error handling in mkiss_open()
  bnx2x: Free NVRAM lock at end of each page
  bnx2x: Prevent null pointer dereference on SKB release
  cxgb4: missing curly braces in t4_setup_debugfs()
  net-timestamp: Update skb_complete_tx_timestamp comment
  ipv6: don't reject link-local nexthop on other interface
  netlink: make sure -EBUSY won't escape from netlink_insert
  bna: fix interrupts storm caused by erroneous packets
  net: mvpp2: replace TX coalescing interrupts with hrtimer
  net: mvpp2: enable proper per-CPU TX buffers unmapping
  ...

9 years agoMerge tag 'edac_fix_for_4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/bp/bp
Linus Torvalds [Thu, 13 Aug 2015 17:22:11 +0000 (10:22 -0700)]
Merge tag 'edac_fix_for_4.2' of git://git./linux/kernel/git/bp/bp

Pull EDAC fix from Borislav Petkov:
 "A ppc4xx_edac fix for accessing ->csrows properly.  This driver was
  missed during the conversion a couple of years ago"

* tag 'edac_fix_for_4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/bp/bp:
  EDAC, ppc4xx: Access mci->csrows array elements properly

9 years agoEDAC, ppc4xx: Access mci->csrows array elements properly
Michael Walle [Tue, 21 Jul 2015 09:00:53 +0000 (11:00 +0200)]
EDAC, ppc4xx: Access mci->csrows array elements properly

The commit

  de3910eb79ac ("edac: change the mem allocation scheme to
 make Documentation/kobject.txt happy")

changed the memory allocation for the csrows member. But ppc4xx_edac was
forgotten in the patch. Fix it.

Signed-off-by: Michael Walle <michael@walle.cc>
Cc: <stable@vger.kernel.org>
Cc: linux-edac <linux-edac@vger.kernel.org>
Cc: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
Link: http://lkml.kernel.org/r/1437469253-8611-1-git-send-email-michael@walle.cc
Signed-off-by: Borislav Petkov <bp@suse.de>
9 years agocosa: missing error code on failure in probe()
Dan Carpenter [Wed, 12 Aug 2015 21:08:01 +0000 (00:08 +0300)]
cosa: missing error code on failure in probe()

If register_hdlc_device() fails, the current code returns 0 but we
should return an error code instead.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agodocumentation: bring vxlan documentation more up-to-date
Rick Jones [Wed, 12 Aug 2015 17:23:14 +0000 (10:23 -0700)]
documentation: bring vxlan documentation more up-to-date

A few things have changed since the previous version of the vxlan
documentation was written, so update it and correct some grammar and
such while we are at it.

Signed-off-by: Rick Jones <rick.jones2@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: fec: Remove unneeded use of IS_ERR_VALUE() macro
Fabio Estevam [Wed, 12 Aug 2015 15:10:23 +0000 (12:10 -0300)]
net: fec: Remove unneeded use of IS_ERR_VALUE() macro

There is no need to use the IS_ERR_VALUE() macro for checking
the return value from pm_runtime_* functions.

Just do a simple negative test instead.

The semantic patch that makes this change is available
in scripts/coccinelle/api/pm_runtime.cocci.

Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: fix bpf_perf_event_read() loop upper bound
Wei-Chun Chao [Wed, 12 Aug 2015 14:57:12 +0000 (07:57 -0700)]
bpf: fix bpf_perf_event_read() loop upper bound

Verifier rejects programs incorrectly.

Fixes: 35578d798400 ("bpf: Implement function bpf_perf_event_read()")
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: Wei-Chun Chao <weichunc@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'cxgb4-more-debug-info'
David S. Miller [Wed, 12 Aug 2015 23:42:12 +0000 (16:42 -0700)]
Merge branch 'cxgb4-more-debug-info'

Hariprasad Shenai says:

====================
Add some more debug info

This patch series adds the following.
Add more info for sge_qinfo dump
Differentiate tid and stids between different regions, and add a debugfs
entry to dump all the tid info

This patch series has been created against net-next tree and includes
patches on cxgb4 driver.

We have included all the maintainers of respective drivers. Kindly review
the change and let us know in case of any review comments.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add debugfs support to dump tid info
Hariprasad Shenai [Wed, 12 Aug 2015 11:25:07 +0000 (16:55 +0530)]
cxgb4: Add debugfs support to dump tid info

Add debugfs support to dump tid info like stid, sftid, tids, atid and
hwtids

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Differentiate between stids between server and filter region
Hariprasad Shenai [Wed, 12 Aug 2015 11:25:06 +0000 (16:55 +0530)]
cxgb4: Differentiate between stids between server and filter region

For T4 adapter, offloaded servers tid for IPv4 connections are
allocated from filter region. So add a new field for server filter tid if
server tid is allocated from filter region.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Differentiates between TIDs being used in TCAM and HASH
Hariprasad Shenai [Wed, 12 Aug 2015 11:25:05 +0000 (16:55 +0530)]
cxgb4: Differentiates between TIDs being used in TCAM and HASH

For the tid info, differentiate from which region the TID is allocated
from. It can be from TCAM region or HASH region.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: Add some more details to sge qinfo
Hariprasad Shenai [Wed, 12 Aug 2015 11:25:04 +0000 (16:55 +0530)]
cxgb4: Add some more details to sge qinfo

Adding more details to sge qinfo for debugging purpose.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: ipv4: increase dhcp inter device timeout
Mugunthan V N [Wed, 12 Aug 2015 10:01:43 +0000 (15:31 +0530)]
net: ipv4: increase dhcp inter device timeout

When a system has multiple ethernet devices and during DHCP
request (for using NFS), the system waits only for HZ/2 which is
500mS before switching to another interface for DHCP.

There are some routers (Ex: Trendnet routers) which responds to
DHCP request at about 560mS. When the system has only one
ethernet interface there is no issue as the timeout is 2S and the
dev xid doesn't changes and only retries.

But when the system has multiple Ethernet like DRA74x with CPSW
in dual EMAC mode, the DHCP response is dropped as the dev xid
changes while shifting to the next device. So changing inter
device timeout to HZ (which is 1S).

Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: fix build warnings and add function read_trace_pipe()
Kaixu Xia [Wed, 12 Aug 2015 09:37:53 +0000 (09:37 +0000)]
bpf: fix build warnings and add function read_trace_pipe()

There are two improvements in this patch:
 1. Fix the build warnings;
 2. Add function read_trace_pipe() to print the result on
    the screen;

Before this patch, we can get the result through /sys/kernel/de
bug/tracing/trace_pipe and get nothing on the screen.
By applying this patch, the result can be printed on the screen.
  $ ./tracex6
...
         tracex6-705   [003] d..1   131.428593: : CPU-3   19981414
            sshd-683   [000] d..1   131.428727: : CPU-0   221682321
            sshd-683   [000] d..1   131.428821: : CPU-0   221808766
            sshd-683   [000] d..1   131.428950: : CPU-0   221982984
            sshd-683   [000] d..1   131.429045: : CPU-0   222111851
         tracex6-705   [003] d..1   131.429168: : CPU-3   20757551
            sshd-683   [000] d..1   131.429170: : CPU-0   222281240
            sshd-683   [000] d..1   131.429261: : CPU-0   222403340
            sshd-683   [000] d..1   131.429378: : CPU-0   222561024
...

Signed-off-by: Kaixu Xia <xiakaixu@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: atl1c: add BQL support
Ron Angeles [Wed, 12 Aug 2015 06:01:20 +0000 (23:01 -0700)]
net: atl1c: add BQL support

This BQL implementation is mostly derived from its related driver, alx.
Tested on AR8131 (rev c0) [1969:1063]. Saturated a 100mbps link with 5
concurrent runs of netperf. Ping latency dropped from 14ms to 3ms.

Signed-off-by: Ron Angeles <ronangeles@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'gianfar-fixes'
David S. Miller [Wed, 12 Aug 2015 21:47:06 +0000 (14:47 -0700)]
Merge branch 'gianfar-fixes'

Jakub Kicinski says:

====================
gianfar: filer changes

respinning with examples as requested.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogianfar: remove faulty filer optimizer
Jakub Kicinski [Wed, 12 Aug 2015 00:41:57 +0000 (02:41 +0200)]
gianfar: remove faulty filer optimizer

Current filer rule optimization is broken in several ways:
 (1) Can perform reads/writes beyond end of allocated tables.
     (gianfar_ethtool.c:1326).

(2) It breaks badly for rules with more than 2 specifiers
     (e.g. matching ip, port, tos).

Example:
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.1 dst-port 1 tos 1 action 1
Added rule with ID 254
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.2 dst-port 2 tos 2 action 9
Added rule with ID 253
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.3 dst-port 3 tos 3 action 17
Added rule with ID 252
# ./filer_decode /sys/kernel/debug/gfar1/filer_raw
00: MASK == 00000210 AND         Q:00           ctrl:00000080 prop:00000210
01: FPR  == 00000210 AND CLE     Q:00           ctrl:00000281 prop:00000210
02: MASK == ffffffff AND         Q:00           ctrl:00000080 prop:ffffffff
03: DPT  == 00000003 AND         Q:00           ctrl:0000008e prop:00000003
04: TOS  == 00000003 AND         Q:00           ctrl:0000008a prop:00000003
05: DIA  == 0a000003 AND         Q:11           ctrl:0000448c prop:0a000003
06: DPT  == 00000002 AND         Q:00           ctrl:0000008e prop:00000002
07: TOS  == 00000002 AND         Q:00           ctrl:0000008a prop:00000002
08: DIA  == 0a000002 AND         Q:09           ctrl:0000248c prop:0a000002
09: DIA  == 0a000001 AND         Q:00           ctrl:0000008c prop:0a000001
0a: DPT  == 00000001 AND         Q:00           ctrl:0000008e prop:00000001
0b: TOS  == 00000001     CLE     Q:01           ctrl:0000060a prop:00000001
ff: MASK >= 00000000             Q:00           ctrl:00000020 prop:00000000

(Entire cluster gets AND-ed together).

 (3) We observed that the masking rules it generates do not
     play well with clustering on P2020.  Only first rule
     of the cluster would ever fire.  Given that optimizer
     relies heavily on masking this is very hard to fix.

Example:
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.1 dst-port 1  action 1
Added rule with ID 254
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.2 dst-port 2  action 9
Added rule with ID 253
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.3 dst-port 3  action 17
Added rule with ID 252
# ./filer_decode /sys/kernel/debug/gfar1/filer_raw
00: MASK == 00000210 AND         Q:00           ctrl:00000080 prop:00000210
01: FPR  == 00000210 AND CLE     Q:00           ctrl:00000281 prop:00000210
02: MASK == ffffffff AND         Q:00           ctrl:00000080 prop:ffffffff
03: DPT  == 00000003 AND         Q:00           ctrl:0000008e prop:00000003
04: DIA  == 0a000003             Q:11           ctrl:0000440c prop:0a000003
05: DPT  == 00000002 AND         Q:00           ctrl:0000008e prop:00000002
06: DIA  == 0a000002             Q:09           ctrl:0000240c prop:0a000002
07: DIA  == 0a000001 AND         Q:00           ctrl:0000008c prop:0a000001
08: DPT  == 00000001     CLE     Q:01           ctrl:0000060e prop:00000001
ff: MASK >= 00000000             Q:00           ctrl:00000020 prop:00000000

Which looks correct according to the spec but only the first
(eth id 252)/last added rule for 10.0.0.3 will ever trigger.
As if filer did not treat the AND CLE as cluster start but
also kept AND-ing the rules.  We found no errata covering this.

The fact that nobody noticed (2) or (3) makes me think
that this feature is not very widely used and we should just
remove it.

Reported-by: Aleksander Dutkowski <adutkowski@gmail.com>
Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Acked-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogianfar: correct list membership accounting
Jakub Kicinski [Wed, 12 Aug 2015 00:41:56 +0000 (02:41 +0200)]
gianfar: correct list membership accounting

At a cost of one line let's make sure .count is correct
when calling gfar_process_filer_changes().

Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogianfar: correct filer table writing
Jakub Kicinski [Wed, 12 Aug 2015 00:41:55 +0000 (02:41 +0200)]
gianfar: correct filer table writing

MAX_FILER_IDX is the last usable index.  Using less-than
will already guarantee that one entry for catch-all rule
will be left, no need to subtract 1 here.

Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agohv_netvsc: Implement set_channels ethtool op
Andrew Schwartzmeyer [Wed, 12 Aug 2015 00:14:32 +0000 (17:14 -0700)]
hv_netvsc: Implement set_channels ethtool op

This enables the use of ethtool --set-channels devname combined N to
change the number of vRSS queues. Separate rx, tx, and other parameters
are not supported. The maximum is rsscap.num_recv_que. It passes the
given value to rndis_filter_device_add through the device_info->num_chn
field.

If the procedure fails, it attempts to recover to the prior state. If
the recovery fails, it logs an error and aborts.

Current num_chn is saved and restored when changing the MTU.

Signed-off-by: Andrew Schwartzmeyer <andschwa@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agohv_netvsc: Set vRSS with num_chn in RNDIS filter
Andrew Schwartzmeyer [Wed, 12 Aug 2015 00:14:31 +0000 (17:14 -0700)]
hv_netvsc: Set vRSS with num_chn in RNDIS filter

Uses device_info->num_chn to pass user provided number of vRSS
queues (from ethtool --set-channels) to rndis_filter_device_add. If
nonzero and less than the maximum, set net_device->num_chn to the given
value; else default to prior algorithm.

Always initialize struct device_info to 0, otherwise not all its fields
are guaranteed to be 0, which is necessary when checking if num_chn has
been purposefully set.

Signed-off-by: Andrew Schwartzmeyer <andschwa@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agolan78xx: Remove BUG_ON()
Woojung.Huh@microchip.com [Tue, 11 Aug 2015 15:23:33 +0000 (15:23 +0000)]
lan78xx: Remove BUG_ON()

Removing BUG_ON()

Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agolan78xx: Fix Smatch Warnings
Woojung.Huh@microchip.com [Tue, 11 Aug 2015 15:21:41 +0000 (15:21 +0000)]
lan78xx: Fix Smatch Warnings

lan78xx.c:2282 tx_complete() warn: variable dereferenced before check 'skb' (see line 2249)
lan78xx.c:2885 lan78xx_bh() info: ignoring unreachable code.
lan78xx.c:3159 lan78xx_probe() info: ignoring unreachable code.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobonding: Gratuitous ARP gets dropped when first slave added
Venkat Venkatsubra [Tue, 11 Aug 2015 14:57:23 +0000 (07:57 -0700)]
bonding: Gratuitous ARP gets dropped when first slave added

When the first slave is added (such as during bootup) the first
gratuitous ARP gets dropped. We don't see this drop during a failover.
The packet gets dropped in qdisc (noop_enqueue).

The fix is to delay the sending of gratuitous ARPs till the bond dev's
carrier is present.

It can also be worked around by setting num_grat_arp to more than 1.

Signed-off-by: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: eth: altera: Remove sgdmadesclen member from altera_tse_private
Tobias Klauser [Mon, 10 Aug 2015 10:26:32 +0000 (12:26 +0200)]
net: eth: altera: Remove sgdmadesclen member from altera_tse_private

altera_tse_private->sgdmadesclen is always assigned assigned the same
value and never changes during runtime.  Remove the struct member and
use a new define for sizeof(struct sgdma_descrip) instead.

Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: Do not override PHY interface if already configured
Florian Fainelli [Sat, 8 Aug 2015 19:58:57 +0000 (12:58 -0700)]
net: dsa: Do not override PHY interface if already configured

In case we need to divert reads/writes using the slave MII bus, we may have
already fetched a valid PHY interface property from Device Tree, and that
mode is used by the PHY driver to make configuration decisions.

If we could not fetch the "phy-mode" property, we will assign p->phy_interface
to PHY_INTERFACE_MODE_NA, such that we can actually check for that condition as
to whether or not we should override the interface value.

Fixes: 19334920eaf7 ("net: dsa: Set valid phy interface type")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Wed, 12 Aug 2015 18:25:01 +0000 (11:25 -0700)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fix from Catalin Marinas:
 "Fix coarse clock monotonicity (VDSO timestamp off by one jiffy
  compared to the syscall one)"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: VDSO: fix coarse clock monotonicity regression

9 years agoMerge branch 'drm-fixes-4.2' of git://people.freedesktop.org/~agd5f/linux
Linus Torvalds [Wed, 12 Aug 2015 18:13:54 +0000 (11:13 -0700)]
Merge branch 'drm-fixes-4.2' of git://people.freedesktop.org/~agd5f/linux

Pull amd drm fixes from Alex Deucher:
 "Dave is on vacation at the moment, so please pull these radeon and
  amdgpu fixes directly.

  Just a few minor things for 4.2:

   - add a new radeon pci id
   - fix a power management regression in amdgpu
   - fix HEVC command buffer validation in amdgpu"

* 'drm-fixes-4.2' of git://people.freedesktop.org/~agd5f/linux:
  drm/radeon: add new OLAND pci id
  Revert "drm/amdgpu: Configure doorbell to maximum slots"
  drm/amdgpu: add context buffer size check for HEVC

9 years agodrm/radeon: add new OLAND pci id
Alex Deucher [Mon, 10 Aug 2015 19:28:49 +0000 (15:28 -0400)]
drm/radeon: add new OLAND pci id

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
9 years agoRevert "drm/amdgpu: Configure doorbell to maximum slots"
Alex Deucher [Mon, 10 Aug 2015 15:08:31 +0000 (11:08 -0400)]
Revert "drm/amdgpu: Configure doorbell to maximum slots"

This reverts commit 78ad5cdd21f0d614983fc397338944e797ec70b9.
This commit breaks dpm and suspend/resume on CZ.

9 years agodrm/amdgpu: add context buffer size check for HEVC
Boyuan Zhang [Wed, 5 Aug 2015 18:03:48 +0000 (14:03 -0400)]
drm/amdgpu: add context buffer size check for HEVC

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
9 years agoMerge tag 'regmap-fix-v4.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Wed, 12 Aug 2015 16:06:39 +0000 (09:06 -0700)]
Merge tag 'regmap-fix-v4.2-rc6' of git://git./linux/kernel/git/broonie/regmap

Pull regmap fix from Mark Brown:
 "regmap: Fix handling of present bits on rbtree cache block resize

  When expanding a cache block we use krealloc() to resize the register
  present bitmap without initialising the newly allocated data (the
  original code was written for kzalloc()).  Add an appropraite memset()
  to fix that"

* tag 'regmap-fix-v4.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: regcache-rbtree: Clean new present bits on present bitmap resize

9 years agodm cache policy smq: move 'dm-cache-default' module alias to SMQ
Yi Zhang [Wed, 12 Aug 2015 11:22:43 +0000 (19:22 +0800)]
dm cache policy smq: move 'dm-cache-default' module alias to SMQ

When creating dm-cache with the default policy, it will call
request_module("dm-cache-default") to register the default policy.
But the "dm-cache-default" alias was left referring to the MQ policy.
Fix this by moving the module alias to SMQ.

Fixes: bccab6a0 (dm cache: switch the "default" cache replacement policy from mq to smq)
Signed-off-by: Yi Zhang <yizhan@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
9 years agodm btree: add ref counting ops for the leaves of top level btrees
Joe Thornber [Wed, 12 Aug 2015 14:12:09 +0000 (15:12 +0100)]
dm btree: add ref counting ops for the leaves of top level btrees

When using nested btrees, the top leaves of the top levels contain
block addresses for the root of the next tree down.  If we shadow a
shared leaf node the leaf values (sub tree roots) should be incremented
accordingly.

This is only an issue if there is metadata sharing in the top levels.
Which only occurs if metadata snapshots are being used (as is possible
with dm-thinp).  And could result in a block from the thinp metadata
snap being reused early, thus corrupting the thinp metadata snap.

Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
9 years agodm thin metadata: delete btrees when releasing metadata snapshot
Joe Thornber [Wed, 12 Aug 2015 14:10:21 +0000 (15:10 +0100)]
dm thin metadata: delete btrees when releasing metadata snapshot

The device details and mapping trees were just being decremented
before.  Now btree_del() is called to do a deep delete.

Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
9 years agoMerge tag 'localmodconfig-v4.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Tue, 11 Aug 2015 22:13:41 +0000 (15:13 -0700)]
Merge tag 'localmodconfig-v4.2-rc6' of git://git./linux/kernel/git/rostedt/linux-kconfig

Pull localmodconfig fix from Steven Rostedt:
 "Leonidas Spyropoulos found that modules like nouveau were being
  unselected by make localmodconfig even though their configs were set
  and the module was loaded and visible by lsmod.

  The reason for this was because streamline-config.pl only looks at
  Makefiles, and not Kbuild files.  As these modules use Kbuild for
  their names, they too need to be checked by localmodconfig.  This was
  fixed by Richard Weinberger"

* tag 'localmodconfig-v4.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-kconfig:
  localmodconfig: Use Kbuild files too

9 years agolocalmodconfig: Use Kbuild files too
Richard Weinberger [Sun, 26 Jul 2015 22:06:55 +0000 (00:06 +0200)]
localmodconfig: Use Kbuild files too

In kbuild it is allowed to define objects in files named "Makefile"
and "Kbuild".
Currently localmodconfig reads objects only from "Makefile"s and misses
modules like nouveau.

Link: http://lkml.kernel.org/r/1437948415-16290-1-git-send-email-richard@nod.at
Cc: stable@vger.kernel.org
Reported-and-tested-by: Leonidas Spyropoulos <artafinde@gmail.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agoMerge branch 'for-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetoot...
David S. Miller [Tue, 11 Aug 2015 21:16:07 +0000 (14:16 -0700)]
Merge branch 'for-upstream' of git://git./linux/kernel/git/bluetooth/bluetooth

Johan Hedberg says:

====================
pull request: bluetooth 2015-08-11

Here's an important regression fix for the 4.2-rc series that ensures
user space isn't given invalid LTK values. The bug essentially prevents
the encryption of subsequent LE connections, i.e. makes it impossible to
pair devices over LE.

Let me know if there are any issues pulling. Thanks.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: fs_enet: mask interrupts for TX partial frames.
LEROY Christophe [Tue, 11 Aug 2015 10:11:03 +0000 (12:11 +0200)]
net: fs_enet: mask interrupts for TX partial frames.

We are not interested in interrupts for partially transmitted frames.
Unlike SCC and FCC, the FEC doesn't handle the I bit in buffer
descriptors, instead it defines two interrupt bits, TXB and TXF.

We have to mask TXB in order to only get interrupts once the
frame is fully transmitted.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: fs_enet: explicitly remove I flag on TX partial frames
LEROY Christophe [Tue, 11 Aug 2015 10:11:00 +0000 (12:11 +0200)]
net: fs_enet: explicitly remove I flag on TX partial frames

We are not interested in interrupts for partially transmitted frames,
we have to clear BD_ENET_TX_INTR explicitly otherwise it may remain
from a previously used descriptor.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'mv88e6xxx-switchdev-fdb'
David S. Miller [Tue, 11 Aug 2015 19:03:20 +0000 (12:03 -0700)]
Merge branch 'mv88e6xxx-switchdev-fdb'

Vivien Didelot says:

====================
net: dsa: mv88e6xxx: support switchdev FDB objects

This patchset refactors the FDB management in the mv88e6xxx code and adds the
glue in DSA to use the switchdev FDB objects.

Below is an usage example (ports 0-2 belongs to br0, ports 3-4 belongs to br1):

    # bridge fdb add 3c:97:0e:11:30:6e dev swp2
    # bridge fdb add 3c:97:0e:11:40:78 dev swp3
    # bridge fdb add 3c:97:0e:11:50:86 dev swp4
    # bridge fdb del 3c:97:0e:11:40:78 dev swp3
    # bridge fdb
    01:00:5e:00:00:01 dev eth0 self permanent
    01:00:5e:00:00:01 dev eth1 self permanent
    00:50:d2:10:78:15 dev swp0 master br0 permanent
    3c:97:0e:11:30:6e dev swp2 self static
    00:50:d2:10:78:15 dev swp3 master br1 permanent
    3c:97:0e:11:50:86 dev swp4 self static
    # cat /sys/kernel/debug/dsa0/atu
    # DB   T/P  Vec State Addr
    # 001  Port 004   e   3c:97:0e:11:30:6e
    # 004  Port 010   e   3c:97:0e:11:50:86

For the 88E6xxx switches, FIDs 1 to num_ports will be reserved for non-bridged
ports and bridge groups, and the remaining will be later used by VLANs.

This change is necessary to welcome the support for hardware VLANs (which will
follow soon).

Changes in v3:

 - reorder commits to improve bisectability and minimize diffs

 - add an ndm_state member in switchdev_fdb_obj instead of an is_static boolean

 - drop the need to convert unsigned char *addr to u8 addr[ETH_ALEN]
   (it is casted to char pointer anyway)

Changes in v2:

 - remove ndo_bridge_{get,set,del}link from switchdev/DSA glue code

 - use ether_addr_copy instead of memcpy for MAC addresses

 - constify MAC address in port_fdb_{add,del}

 - split the mv88e6xxx code refactoring into several patches
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: add support for switchdev FDB objects
Vivien Didelot [Mon, 10 Aug 2015 13:09:53 +0000 (09:09 -0400)]
net: dsa: add support for switchdev FDB objects

Implement the switchdev_port_obj_{add,del,dump} functions in DSA to
support the SWITCHDEV_OBJ_PORT_FDB objects.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: switchdev: support static FDB addresses
Vivien Didelot [Mon, 10 Aug 2015 13:09:52 +0000 (09:09 -0400)]
net: switchdev: support static FDB addresses

This patch adds an ndm_state member to the switchdev_obj_fdb structure,
in order to support static FDB addresses.

Set Rocker ndm_state to NUD_REACHABLE.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: rework FDB Get Next operation
Vivien Didelot [Mon, 10 Aug 2015 13:09:51 +0000 (09:09 -0400)]
net: dsa: mv88e6xxx: rework FDB Get Next operation

Add a low level _mv88e6xxx_atu_getnext function for convenient access to
the hardware, and rework the FDB Get Next operation.

This will ease the future integration with VLAN IDs.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: rework FDB add/del operations
Vivien Didelot [Mon, 10 Aug 2015 13:09:50 +0000 (09:09 -0400)]
net: dsa: mv88e6xxx: rework FDB add/del operations

Add a mv88e6xxx_atu_entry structure and a low level function for the ATU
Load operation, and provide FDB add and delete wrappers functions.

This implementation handles the eventual trunk mapping. If the related
bit is set, then the ATU data register would contain the trunk ID, and
not the port vector.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: change FDB routines prototypes
Vivien Didelot [Mon, 10 Aug 2015 13:09:49 +0000 (09:09 -0400)]
net: dsa: change FDB routines prototypes

Change the prototype of port_getnext to include a vid parameter.

This is necessary to introduce the support for VLAN.

Also rename the fdb_{add,del,getnext} function pointers to
port_fdb_{add,del,getnext} since they are specific to a given port.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: rename ATU MAC accessors
Vivien Didelot [Mon, 10 Aug 2015 13:09:48 +0000 (09:09 -0400)]
net: dsa: mv88e6xxx: rename ATU MAC accessors

Rename the __mv88e6xxx_{read,write}_addr functions to more explicit
_mv88e6xxx_atu_mac_{read,write} functions, which also respect the single
underscore convention used in the file (meaning SMI lock must be held).

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: extend fid mask
Vivien Didelot [Mon, 10 Aug 2015 13:09:47 +0000 (09:09 -0400)]
net: dsa: mv88e6xxx: extend fid mask

The driver currently manages one FID per port (or bridge group), with a
mask of DSA_MAX_PORTS bits, where 0 means that the FID is in use.

The Marvell 88E6xxx switches support up to 4094 FIDs (from 1 to 0xfff;
FID 0 means that multiple address databases are not being used).

This patch changes the fid_mask for an fid_bitmap of 4096 bits.

>From now on, FIDs 1 to num_ports are reserved for non-bridged ports and
bridge groups (a bridge group gets the FID of its first member). The
remaining bits will be reserved for VLAN entries.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6xxx: define GLOBAL_ATU_FID
Vivien Didelot [Mon, 10 Aug 2015 13:09:46 +0000 (09:09 -0400)]
net: dsa: mv88e6xxx: define GLOBAL_ATU_FID

Define register GLOBAL_ATU_FID instead of the raw value 0x01.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRevert "Merge branch 'mv88e6xxx-switchdev-fdb'"
David S. Miller [Tue, 11 Aug 2015 19:00:37 +0000 (12:00 -0700)]
Revert "Merge branch 'mv88e6xxx-switchdev-fdb'"

This reverts commit f1d5ca434413b20cd3f8c18ff2b634b7782149a5, reversing
changes made to 4933d85c5173832ebd261756522095837583c458.

I applied v2 instead of v3.

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: s390: Fix build error caused by the struct bpf_array member name changed
Kaixu Xia [Tue, 11 Aug 2015 08:56:51 +0000 (08:56 +0000)]
bpf: s390: Fix build error caused by the struct bpf_array member name changed

There is a build error that "'struct bpf_array' has no member
named 'prog'" on s390. In commit 2a36f0b92eb6 ("bpf: Make the
bpf_prog_array_map more generic"), the member 'prog' of struct
bpf_array is replaced by 'ptrs'. So this patch fixes it.

Fixes: 2a36f0b92eb6 ("bpf: Make the bpf_prog_array_map more generic")
Reported-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Kaixu Xia <xiakaixu@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'thunder-acpi'
David S. Miller [Tue, 11 Aug 2015 18:47:57 +0000 (11:47 -0700)]
Merge branch 'thunder-acpi'

David Daney says:

====================
net: thunder: Add ACPI support.

Change from v1:  Drop PHY binding part, use fwnode_property* APIs.

The first patch (1/2) rearranges the existing code a little with no
functional change to get ready for the second.  The second (2/2) does
the actual work of adding support to extract the needed information
from the ACPI tables.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet, thunder, bgx: Add support to get MAC address from ACPI.
David Daney [Tue, 11 Aug 2015 00:58:37 +0000 (17:58 -0700)]
net, thunder, bgx: Add support to get MAC address from ACPI.

Currently there is no way to get the MAC address in a firmware
independent manner, so set the MAC address of the device directly from
the ACPI tables.

The binding agrees with the proposed standard here:

http://www.uefi.org/sites/default/files/resources/nic-request-v2.pdf

Based on code from: Narinder Dhillon <ndhillon@cavium.com>
                    Tomasz Nowicki <tomasz.nowicki@linaro.org>
                    Robert Richter <rrichter@cavium.com>

Signed-off-by: Tomasz Nowicki <tomasz.nowicki@linaro.org>
Signed-off-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: thunder: Factor out DT specific code in BGX
Robert Richter [Tue, 11 Aug 2015 00:58:36 +0000 (17:58 -0700)]
net: thunder: Factor out DT specific code in BGX

Separate DT code in preparation for follow-on ACPI integration.

Based on code from: Tomasz Nowicki <tomasz.nowicki@linaro.org>

Signed-off-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'fbdev-fixes-4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba...
Linus Torvalds [Tue, 11 Aug 2015 17:23:59 +0000 (10:23 -0700)]
Merge tag 'fbdev-fixes-4.2' of git://git./linux/kernel/git/tomba/linux

Pull fbdev fixes from Tomi Valkeinen:
 - fix display regression on Versatile boards
 - fix OF node refcount bugs on omapdss
 - fix WARN about clock prepare on pxa3xx_gcu
 - fix mem leak in videomode helpers
 - fix fbconsole related boot problem on sun7i-a20-olinuxino-micro

* tag 'fbdev-fixes-4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux:
  fbcon: unconditionally initialize cursor blink interval
  video: Fix possible leak in of_get_videomode()
  video: fbdev: pxa3xx_gcu: prepare the clocks
  OMAPDSS: Fix omap_dss_find_output_by_port_node() port refcount decrement
  OMAPDSS: Fix node refcount leak in omapdss_of_get_next_port()
  fbdev: select versatile helpers for the integrator

9 years agoxen/xenbus: Don't leak memory when unmapping the ring on HVM backend
Julien Grall [Mon, 10 Aug 2015 18:10:38 +0000 (19:10 +0100)]
xen/xenbus: Don't leak memory when unmapping the ring on HVM backend

The commit ccc9d90a9a8b5c4ad7e9708ec41f75ff9e98d61d "xenbus_client:
Extend interface to support multi-page ring" removes the call to
free_xenballooned_pages() in xenbus_unmap_ring_vfree_hvm(), leaking a
page for every shared ring.

Only with backends running in HVM domains were affected.

Signed-off-by: Julien Grall <julien.grall@citrix.com>
Cc: <stable@vger.kernel.org>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Reviewed-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
9 years agoRevert "xen/events/fifo: Handle linked events when closing a port"
David Vrabel [Mon, 10 Aug 2015 17:11:06 +0000 (18:11 +0100)]
Revert "xen/events/fifo: Handle linked events when closing a port"

This reverts commit fcdf31a7c162de0c93a2bee51df4688ab0a348f8.

This was causing a WARNING whenever a PIRQ was closed since
shutdown_pirq() is called with irqs disabled.

Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Cc: <stable@vger.kernel.org>
9 years agoinet: fix possible request socket leak
Eric Dumazet [Mon, 10 Aug 2015 22:07:34 +0000 (15:07 -0700)]
inet: fix possible request socket leak

In commit b357a364c57c9 ("inet: fix possible panic in
reqsk_queue_unlink()"), I missed fact that tcp_check_req()
can return the listener socket in one case, and that we must
release the request socket refcount or we leak it.

Tested:

 Following packetdrill test template shows the issue

0     socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0    setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0    bind(3, ..., ...) = 0
+0    listen(3, 1) = 0

+0    < S 0:0(0) win 2920 <mss 1460,sackOK,nop,nop>
+0    > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK>
+.002 < . 1:1(0) ack 21 win 2920
+0    > R 21:21(0)

Fixes: b357a364c57c9 ("inet: fix possible panic in reqsk_queue_unlink()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoinet: fix races with reqsk timers
Eric Dumazet [Mon, 10 Aug 2015 16:09:13 +0000 (09:09 -0700)]
inet: fix races with reqsk timers

reqsk_queue_destroy() and reqsk_queue_unlink() should use
del_timer_sync() instead of del_timer() before calling reqsk_put(),
otherwise we could free a req still used by another cpu.

But before doing so, reqsk_queue_destroy() must release syn_wait_lock
spinlock or risk a dead lock, as reqsk_timer_handler() might
need to take this same spinlock from reqsk_queue_unlink() (called from
inet_csk_reqsk_queue_drop())

Fixes: fa76ce7328b2 ("inet: get rid of central tcp/dccp listener timer")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agovxlan: fix fdb_dump index calculation
Atzm Watanabe [Mon, 10 Aug 2015 14:39:09 +0000 (23:39 +0900)]
vxlan: fix fdb_dump index calculation

When too many remotes are bound to an FDB entry, index may not be increased.
This problem will be caused on the large scale environment that is based on
the unicast default destination, for instance.

Signed-off-by: Atzm Watanabe <atzm@iij.ad.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomellanox: mlxsw: Use '%zx' to print size_t format
Fabio Estevam [Mon, 10 Aug 2015 12:54:28 +0000 (09:54 -0300)]
mellanox: mlxsw: Use '%zx' to print size_t format

Use '%zx' to print size_t format in order to fix the following build warning:

drivers/net/ethernet/mellanox/mlxsw/item.h:65:3: warning: format '%lx' expects argument of type 'long unsigned int', but argument 6 has type 'size_t' [-Wformat=]

Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomkiss: Fix error handling in mkiss_open()
Fabio Estevam [Mon, 10 Aug 2015 17:22:43 +0000 (14:22 -0300)]
mkiss: Fix error handling in mkiss_open()

If register_netdev() fails we are not propagating the error and
we return success because ax_open() succeeded previously.

Fix this by checking the return value of ax_open() and
register_netdev() and propagate the error in case of failure.

Reported-by: RUC_Soft_Sec <zy900702@163.com>
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
David S. Miller [Tue, 11 Aug 2015 04:03:25 +0000 (21:03 -0700)]
Merge git://git./pub/scm/linux/kernel/git/pablo/nf

Pablo Neira Ayuso says:

====================
Netfilter fixes for net

The following patchset contains five Netfilter fixes for your net tree,
they are:

1) Silence a warning on falling back to vmalloc(). Since 88eab472ec21, we can
   easily hit this warning message, that gets users confused. So let's get rid
   of it.

2) Recently when porting the template object allocation on top of kmalloc to
   fix the netns dependencies between x_tables and conntrack, the error
   checks where left unchanged. Remove IS_ERR() and check for NULL instead.
   Patch from Dan Carpenter.

3) Don't ignore gfp_flags in the new nf_ct_tmpl_alloc() function, from
   Joe Stringer.

4) Fix a crash due to NULL pointer dereference in ip6t_SYNPROXY, patch from
   Phil Sutter.

5) The sequence number of the Syn+ack that is sent from SYNPROXY to clients is
   not adjusted through our NAT infrastructure, as a result the client may
   ignore this TCP packet and TCP flow hangs until the client probes us.  Also
   from Phil Sutter.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid
Linus Torvalds [Mon, 10 Aug 2015 22:16:48 +0000 (15:16 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/jikos/hid

Pull HID fixes from Jiri Kosina:

 - fix for bounds limit calculation in uclogic driver, by Dan Carpenter

 - fix for use-after-free during device removal, by Krzysztof Kozlowski

 - fix for userspace regression (that became apparent only with shiny
   new libinput, so it's not that bad, but I still consider it 4.2
   material), in wacom driver, by Jason Gerecke

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid:
  HID: wacom: Report correct device resolution when using the wireless adapater
  HID: hid-input: Fix accessing freed memory during device disconnect
  HID: uclogic: fix limit in uclogic_tablet_enable()

9 years agoHID: wacom: Report correct device resolution when using the wireless adapater
Jason Gerecke [Wed, 5 Aug 2015 22:44:53 +0000 (15:44 -0700)]
HID: wacom: Report correct device resolution when using the wireless adapater

The 'wacom_wireless_work' function does not recalculate the tablet's
resolution, causing the value contained in the 'features' struct to
always be reported to userspace. This value is valid only for the pen
interface, meaning that the value will be incorrect for the touchpad (if
present). This in particular causes problems for libinput which relies
on the reported resolution being correct.

This patch adds the necessary calls to recalculate the resolution for
each interface. This requires a little bit of code shuffling since both
the 'wacom_set_default_phy' and 'wacom_calculate_res' are declared below
their new first point of use in 'wacom_wireless_work'.

Signed-off-by: Jason Gerecke <jason.gerecke@wacom.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
9 years agoMerge branch 'bnx2x-fixes'
David S. Miller [Mon, 10 Aug 2015 21:31:59 +0000 (14:31 -0700)]
Merge branch 'bnx2x-fixes'

Yuval Mintz says:

====================
bnx2x: small fixes

This adds 2 small fixes, one to error flows during memory release
and the other to flash writes via ethtool API.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobnx2x: Free NVRAM lock at end of each page
Yuval Mintz [Mon, 10 Aug 2015 09:49:36 +0000 (12:49 +0300)]
bnx2x: Free NVRAM lock at end of each page

Writing each 4Kb page into flash might take up-to ~100 miliseconds,
during which time management firmware cannot acces the nvram for its
own uses.

Firmware upgrade utility use the ethtool API to burn new flash images
for the device via the ethtool API, doing so by writing several page-worth
of data on each command. Such action might create problems for the
management firmware, as the nvram might not be accessible for a long time.

This patch changes the write implementation, releasing the nvram lock on
the completion of each page, allowing the management firmware time to
claim it and perform its own required actions.

Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: Ariel Elior <Ariel.Elior@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobnx2x: Prevent null pointer dereference on SKB release
Yuval Mintz [Mon, 10 Aug 2015 09:49:35 +0000 (12:49 +0300)]
bnx2x: Prevent null pointer dereference on SKB release

On error flows its possible to free an SKB even if it was not allocated.

Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: Ariel Elior <Ariel.Elior@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoisdn: Remove reverse_bits(), use revbit8()
yalin wang [Mon, 10 Aug 2015 09:15:57 +0000 (17:15 +0800)]
isdn: Remove reverse_bits(), use revbit8()

This change isdn driver, remove reverse_bits() function,
use the generic revbit8() function instead.

Signed-off-by: yalin wang <yalin.wang2010@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: cleanup some indenting
Dan Carpenter [Sat, 8 Aug 2015 19:15:59 +0000 (22:15 +0300)]
cxgb4: cleanup some indenting

Add or remove some tabs so that statements line up correctly.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agocxgb4: missing curly braces in t4_setup_debugfs()
Dan Carpenter [Sat, 8 Aug 2015 19:15:25 +0000 (22:15 +0300)]
cxgb4: missing curly braces in t4_setup_debugfs()

There were missing curly braces so it means we call add_debugfs_mem()
unintentionally.

Fixes: 3ccc6cf74d8c ('cxgb4: Adds support for T6 adapter')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agodsa: Support multiple MDIO busses
Andrew Lunn [Sat, 8 Aug 2015 15:09:14 +0000 (17:09 +0200)]
dsa: Support multiple MDIO busses

When using a cluster of switches, some topologies will have an MDIO
bus per switch, not one for the whole cluster. Allow this to be
represented in the device tree, by adding an optional mii-bus property
at the switch level. The old platform_device method of instantiation
supports this already, so only the device tree binding needs extending
with an additional optional phandle.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: dsa: mv88e6352: Use mnemonics for EEPROM registers and bits
Andrew Lunn [Sat, 8 Aug 2015 15:04:50 +0000 (17:04 +0200)]
net: dsa: mv88e6352: Use mnemonics for EEPROM registers and bits

Add register definitions #defines for accessing the EEPROM.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'ovs-gre'
David S. Miller [Mon, 10 Aug 2015 21:03:55 +0000 (14:03 -0700)]
Merge branch 'ovs-gre'

Pravin B Shelar says:

====================
GRE: Use flow based tunneling for OVS GRE vport.

Following patches make use of new Using GRE tunnel meta data
collection feature. This allows us to directly use netdev
based GRE tunnel implementation. While doing so I have
removed GRE demux API which were targeted for OVS. Most
of GRE protocol code is now consolidated in ip_gre module.

v5-v4:
Fixed Kconfig dependency for vport-gre module.

v3-v4:
Added interface to ip-gre device to enable meta data collection.
While doing this I split second patch into two patches.

v2-v3:
Add API to create GRE flow based device.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogre: Remove support for sharing GRE protocol hook.
Pravin B Shelar [Sat, 8 Aug 2015 06:51:52 +0000 (23:51 -0700)]
gre: Remove support for sharing GRE protocol hook.

Support for sharing GREPROTO_CISCO port was added so that
OVS gre port and kernel GRE devices can co-exist. After
flow-based tunneling patches OVS GRE protocol processing
is completely moved to ip_gre module. so there is no need
for GRE protocol hook. Following patch consolidates
GRE protocol related functions into ip_gre module.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoopenvswitch: Use regular GRE net_device instead of vport
Pravin B Shelar [Sat, 8 Aug 2015 06:51:47 +0000 (23:51 -0700)]
openvswitch: Use regular GRE net_device instead of vport

Using GRE tunnel meta data collection feature, we can implement
OVS GRE vport. This patch removes all of the OVS
specific GRE code and make OVS use a ip_gre net_device.
Minimal GRE vport is kept to handle compatibility with
current userspace application.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoip_gre: Add support to collect tunnel metadata.
Pravin B Shelar [Sat, 8 Aug 2015 06:51:42 +0000 (23:51 -0700)]
ip_gre: Add support to collect tunnel metadata.

Following patch create new tunnel flag which enable
tunnel metadata collection on given device.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoopenvswitch: Move tunnel destroy function to oppenvswitch module.
Pravin B Shelar [Sat, 8 Aug 2015 06:51:33 +0000 (23:51 -0700)]
openvswitch: Move tunnel destroy function to oppenvswitch module.

This function will be used in gre and geneve vport implementations.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: add explicit logging and stat for neighbour table overflow
Rick Jones [Fri, 7 Aug 2015 18:10:37 +0000 (11:10 -0700)]
net: add explicit logging and stat for neighbour table overflow

Add an explicit neighbour table overflow message (ratelimited) and
statistic to make diagnosing neighbour table overflows tractable in
the wild.

Diagnosing a neighbour table overflow can be quite difficult in the wild
because there is no explicit dmesg logged.  Callers to neighbour code
seem to use net_dbg_ratelimit when the neighbour call fails which means
the "base message" is not emitted and the callback suppressed messages
from the ratelimiting can end-up juxtaposed with unrelated messages.
Further, a forced garbage collection will increment a stat on each call
whether it was successful in freeing-up a table entry or not, so that
statistic is only a hint.  So, add a net_info_ratelimited message and
explicit statistic to the neighbour code.

Signed-off-by: Rick Jones <rick.jones2@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobridge: netlink: add support for vlan_filtering attribute
Nikolay Aleksandrov [Fri, 7 Aug 2015 16:40:45 +0000 (19:40 +0300)]
bridge: netlink: add support for vlan_filtering attribute

This patch adds the ability to toggle the vlan filtering support via
netlink. Since we're already running with rtnl in .changelink() we don't
need to take any additional locks.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet-timestamp: Update skb_complete_tx_timestamp comment
Benjamin Poirier [Fri, 7 Aug 2015 16:32:21 +0000 (09:32 -0700)]
net-timestamp: Update skb_complete_tx_timestamp comment

After "62bccb8 net-timestamp: Make the clone operation stand-alone from phy
timestamping" the hwtstamps parameter of skb_complete_tx_timestamp() may no
longer be NULL.

Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Cc: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'qlcnic-enhancements'
David S. Miller [Mon, 10 Aug 2015 20:34:29 +0000 (13:34 -0700)]
Merge branch 'qlcnic-enhancements'

Shahed Shaikh says:

====================
qlcnic: enhancements

This series adds few enhancements.

  o Patch from Harish reorders the sequence of header files inclusion,
    keeping kernel's header files on top.

  o Firmware introduced a new feature which allows driver to increases
    the size of firmware dump of iSCSI function which is being collected
    by NIC driver.

  o Print buffer address which is holding a firmware dump.

  o Use vzalloc() instead kzalloc() for allocating large chunk of memory
    which will avoid potential memory allocation failure.

  o Add new device ID for 0x8C30 which is a 83xx series based VF function.

Please apply this series to net-next.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoqlcnic: Update version to 5.3.63
Shahed Shaikh [Fri, 7 Aug 2015 11:17:07 +0000 (07:17 -0400)]
qlcnic: Update version to 5.3.63

Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoqlcnic: Don't use kzalloc unncecessarily for allocating large chunk of memory
Shahed Shaikh [Fri, 7 Aug 2015 11:17:06 +0000 (07:17 -0400)]
qlcnic: Don't use kzalloc unncecessarily for allocating large chunk of memory

Driver allocates a large chunk of temporary buffer using kzalloc
to copy FW image. As there is no real need of this memory to be
physically contiguous, use vzalloc instead.

Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoqlcnic: Add new VF device ID 0x8C30
Shahed Shaikh [Fri, 7 Aug 2015 11:17:05 +0000 (07:17 -0400)]
qlcnic: Add new VF device ID 0x8C30

This is a 83xx series based VF device

Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoqlcnic: Print firmware minidump buffer and template header addresses
Shahed Shaikh [Fri, 7 Aug 2015 11:17:04 +0000 (07:17 -0400)]
qlcnic: Print firmware minidump buffer and template header addresses

Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoqlcnic: Add support to enable capability to extend minidump for iSCSI
Shahed Shaikh [Fri, 7 Aug 2015 11:17:03 +0000 (07:17 -0400)]
qlcnic: Add support to enable capability to extend minidump for iSCSI

In some cases it is required to capture minidump for iSCSI functions
as part of default minidump collection process. To enable this, firmware
exports it's capability and driver need to enable that capability
by issuing a mailbox command.

With this feature, firmware can provide additional iSCSI function's
minidump with smaller minidump capture mask (0x1f).

Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoqlcnic: Rearrange ordering of header files inclusion
Harish Patil [Fri, 7 Aug 2015 11:17:02 +0000 (07:17 -0400)]
qlcnic: Rearrange ordering of header files inclusion

Include local headers files after kernel's header files.

Signed-off-by: Harish Patil <harish.patil@qlogic.com>
Signed-off-by: Shahed Shaikh <shahed.shaikh@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: phy: select copper mode when Marvel 88e1111 in SGMII
Madalin Bucur [Fri, 7 Aug 2015 09:07:50 +0000 (17:07 +0800)]
net: phy: select copper mode when Marvel 88e1111 in SGMII

For the Marvel 88e1111 PHY only two SGMII modes are available, both
allowing only SGMII to copper mode (with or without clock). SGMII
to fiber mode is not supported. Make sure the fiber/copper registers
selector bits are cleared for selecting copper mode.

Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
Signed-off-by: Shaohui Xie <Shaohui.Xie@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: don't reject link-local nexthop on other interface
Florian Westphal [Fri, 7 Aug 2015 08:54:28 +0000 (10:54 +0200)]
ipv6: don't reject link-local nexthop on other interface

48ed7b26faa7 ("ipv6: reject locally assigned nexthop addresses") is too
strict; it rejects following corner-case:

ip -6 route add default via fe80::1:2:3 dev eth1

[ where fe80::1:2:3 is assigned to a local interface, but not eth1 ]

Fix this by restricting search to given device if nh is linklocal.

Joint work with Hannes Frederic Sowa.

Fixes: 48ed7b26faa7 ("ipv6: reject locally assigned nexthop addresses")
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: fec: fix the race between xmit and bdp reclaiming path
Kevin Hao [Fri, 7 Aug 2015 05:52:37 +0000 (13:52 +0800)]
net: fec: fix the race between xmit and bdp reclaiming path

When we transmit a fragmented skb, we may run into a race like the
following scenario (assume txq->cur_tx is next to txq->dirty_tx):
           cpu 0                                          cpu 1
  fec_enet_txq_submit_skb
    reserve a bdp for the first fragment
    fec_enet_txq_submit_frag_skb
       update the bdp for the other fragment
       update txq->cur_tx
                                                   fec_enet_tx_queue
                                                     bdp = fec_enet_get_nextdesc(txq->dirty_tx, fep, queue_id);
                                                     This bdp is the bdp reserved for the first segment. Given
                                                     that this bdp BD_ENET_TX_READY bit is not set and txq->cur_tx
                                                     is already pointed to a bdp beyond this one. We think this is a
                                                     completed bdp and try to reclaim it.
    update the bdp for the first segment
    update txq->cur_tx

So we shouldn't update the txq->cur_tx until all the update to the
bdps used for fragments are performed. Also add the corresponding
memory barrier to guarantee that the update to the bdps, dirty_tx and
cur_tx performed in the proper order.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonetlink: make sure -EBUSY won't escape from netlink_insert
Daniel Borkmann [Thu, 6 Aug 2015 22:26:41 +0000 (00:26 +0200)]
netlink: make sure -EBUSY won't escape from netlink_insert

Linus reports the following deadlock on rtnl_mutex; triggered only
once so far (extract):

[12236.694209] NetworkManager  D 0000000000013b80     0  1047      1 0x00000000
[12236.694218]  ffff88003f902640 0000000000000000 ffffffff815d15a9 0000000000000018
[12236.694224]  ffff880119538000 ffff88003f902640 ffffffff81a8ff84 00000000ffffffff
[12236.694230]  ffffffff81a8ff88 ffff880119c47f00 ffffffff815d133a ffffffff81a8ff80
[12236.694235] Call Trace:
[12236.694250]  [<ffffffff815d15a9>] ? schedule_preempt_disabled+0x9/0x10
[12236.694257]  [<ffffffff815d133a>] ? schedule+0x2a/0x70
[12236.694263]  [<ffffffff815d15a9>] ? schedule_preempt_disabled+0x9/0x10
[12236.694271]  [<ffffffff815d2c3f>] ? __mutex_lock_slowpath+0x7f/0xf0
[12236.694280]  [<ffffffff815d2cc6>] ? mutex_lock+0x16/0x30
[12236.694291]  [<ffffffff814f1f90>] ? rtnetlink_rcv+0x10/0x30
[12236.694299]  [<ffffffff8150ce3b>] ? netlink_unicast+0xfb/0x180
[12236.694309]  [<ffffffff814f5ad3>] ? rtnl_getlink+0x113/0x190
[12236.694319]  [<ffffffff814f202a>] ? rtnetlink_rcv_msg+0x7a/0x210
[12236.694331]  [<ffffffff8124565c>] ? sock_has_perm+0x5c/0x70
[12236.694339]  [<ffffffff814f1fb0>] ? rtnetlink_rcv+0x30/0x30
[12236.694346]  [<ffffffff8150d62c>] ? netlink_rcv_skb+0x9c/0xc0
[12236.694354]  [<ffffffff814f1f9f>] ? rtnetlink_rcv+0x1f/0x30
[12236.694360]  [<ffffffff8150ce3b>] ? netlink_unicast+0xfb/0x180
[12236.694367]  [<ffffffff8150d344>] ? netlink_sendmsg+0x484/0x5d0
[12236.694376]  [<ffffffff810a236f>] ? __wake_up+0x2f/0x50
[12236.694387]  [<ffffffff814cad23>] ? sock_sendmsg+0x33/0x40
[12236.694396]  [<ffffffff814cb05e>] ? ___sys_sendmsg+0x22e/0x240
[12236.694405]  [<ffffffff814cab75>] ? ___sys_recvmsg+0x135/0x1a0
[12236.694415]  [<ffffffff811a9d12>] ? eventfd_write+0x82/0x210
[12236.694423]  [<ffffffff811a0f9e>] ? fsnotify+0x32e/0x4c0
[12236.694429]  [<ffffffff8108cb70>] ? wake_up_q+0x60/0x60
[12236.694434]  [<ffffffff814cba09>] ? __sys_sendmsg+0x39/0x70
[12236.694440]  [<ffffffff815d4797>] ? entry_SYSCALL_64_fastpath+0x12/0x6a

It seems so far plausible that the recursive call into rtnetlink_rcv()
looks suspicious. One way, where this could trigger is that the senders
NETLINK_CB(skb).portid was wrongly 0 (which is rtnetlink socket), so
the rtnl_getlink() request's answer would be sent to the kernel instead
to the actual user process, thus grabbing rtnl_mutex() twice.

One theory would be that netlink_autobind() triggered via netlink_sendmsg()
internally overwrites the -EBUSY error to 0, but where it is wrongly
originating from __netlink_insert() instead. That would reset the
socket's portid to 0, which is then filled into NETLINK_CB(skb).portid
later on. As commit d470e3b483dc ("[NETLINK]: Fix two socket hashing bugs.")
also puts it, -EBUSY should not be propagated from netlink_insert().

It looks like it's very unlikely to reproduce. We need to trigger the
rhashtable_insert_rehash() handler under a situation where rehashing
currently occurs (one /rare/ way would be to hit ht->elasticity limits
while not filled enough to expand the hashtable, but that would rather
require a specifically crafted bind() sequence with knowledge about
destination slots, seems unlikely). It probably makes sense to guard
__netlink_insert() in any case and remap that error. It was suggested
that EOVERFLOW might be better than an already overloaded ENOMEM.

Reference: http://thread.gmane.org/gmane.linux.network/372676
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobna: fix interrupts storm caused by erroneous packets
Ivan Vecera [Thu, 6 Aug 2015 20:48:23 +0000 (22:48 +0200)]
bna: fix interrupts storm caused by erroneous packets

The commit "e29aa33 bna: Enable Multi Buffer RX" moved packets counter
increment from the beginning of the NAPI processing loop after the check
for erroneous packets so they are never accounted. This counter is used
to inform firmware about number of processed completions (packets).
As these packets are never acked the firmware fires IRQs for them again
and again.

Fixes: e29aa33 ("bna: Enable Multi Buffer RX")
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Acked-by: Rasesh Mody <rasesh.mody@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'mvpp2-fixes'
David S. Miller [Mon, 10 Aug 2015 17:57:01 +0000 (10:57 -0700)]
Merge branch 'mvpp2-fixes'

Marcin Wojtas says:

====================
Fixes for the network driver of Marvell Armada 375 SoC

This is a set of three patches that fix long-lasting problems implemented in
the initial support for the Armada 375 network controller.

Due to an inappropriate concept of handling the per-CPU sent packets'
processing on TX path the driver numerous problems occured, such as RCU
stalls. Those have been fixed, of which details you can find in the commit
logs. The patches were intensively tested on top of v4.2-rc5.

I'm looking forward to any comments or remarks.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: mvpp2: replace TX coalescing interrupts with hrtimer
Marcin Wojtas [Thu, 6 Aug 2015 17:00:30 +0000 (19:00 +0200)]
net: mvpp2: replace TX coalescing interrupts with hrtimer

The PP2 controller is capable of per-CPU TX processing, which means there are
per-CPU banked register sets and queues. Current version of the driver supports
TX packet coalescing - once on given CPU sent packets amount reaches a threshold
value, an IRQ occurs. However, there is a single interrupt line responsible for
CPU0/1 TX and RX events (the latter is not per-CPU, the hardware does not
support RSS).

When the top-half executes the interrupt cause is not known. This is why in
NAPI poll function, along with RX processing, IRQ cause register on both
CPU's is accessed in order to determine on which of them the TX coalescing
threshold might have been reached. Thus the egress processing and releasing the
buffers is able to take place on the corresponding CPU. Hitherto approach lead
to an illegal usage of on_each_cpu function in softirq context.

The problem is solved by resigning from TX coalescing interrupts and separating
egress finalization from NAPI processing. For that purpose a method of using
hrtimer is introduced. In main transmit function (mvpp2_tx) buffers are released
once a software coalescing threshold is reached. In case not all the data is
processed a timer is set on this CPU - in its interrupt context a tasklet is
scheduled in which all queues are processed. At once only one timer per-CPU can
be running, which is controlled by a dedicated flag.

This commit removes TX processing from NAPI polling function, disables hardware
coalescing and enables hrtimer with tasklet, using new per-CPU port structure
(mvpp2_port_pcpu).

Signed-off-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>