openwrt/staging/blogic.git
13 years agoigb: Remove multi_tx_table and simplify igb_xmit_frame
Alexander Duyck [Fri, 26 Aug 2011 07:44:10 +0000 (07:44 +0000)]
igb: Remove multi_tx_table and simplify igb_xmit_frame

Instead of using the multi_tx_table to map possible Tx queues to Tx rings
we can just do simple subtraction for the unlikely event that the Tx queue
provided exceeds the number of Tx rings.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoigb: Replace E1000_XX_DESC_ADV with IGB_XX_DESC
Alexander Duyck [Fri, 26 Aug 2011 07:44:05 +0000 (07:44 +0000)]
igb: Replace E1000_XX_DESC_ADV with IGB_XX_DESC

Since igb only uses advanced descriptors we might as well just use an IGB
specific define and drop the _ADV suffix for the descriptor declarations.
In addition this can be further reduced by assuming that it will be working
on pointers since that is normally how the Tx descriptors are handled.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoigb: drop the "adv" off function names relating to descriptors
Alexander Duyck [Fri, 26 Aug 2011 07:43:59 +0000 (07:43 +0000)]
igb: drop the "adv" off function names relating to descriptors

Many of the function names in the hot path are carrying an extra "_adv"
suffix on the end of them to represent the fact that they are using
advanced descriptors instead of legacy descriptors.  However since all igb
uses are advanced descriptors adding the extra suffix doesn't really add
any additional data.  Since this is the case it is easiest to just drop the
suffix and save us from having to store the extra characters.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoigb: Refactor clean_rx_irq to reduce overhead and improve performance
Alexander Duyck [Fri, 26 Aug 2011 07:43:54 +0000 (07:43 +0000)]
igb: Refactor clean_rx_irq to reduce overhead and improve performance

This change is meant to be a general cleanup and performance improvement
for clean_rx_irq.  The previous patch should have updated the allocation so
that the rings can be treated as read-only within the clean_rx_irq
function.  In addition I am re-ordering the operations such that several
goals are accomplished including reducing the overhead for packet
accounting, reducing the number of items on the stack, and improving
overall performance.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoigb: update ring and adapter structure to improve performance
Alexander Duyck [Fri, 26 Aug 2011 07:43:48 +0000 (07:43 +0000)]
igb: update ring and adapter structure to improve performance

This change is meant to improve performance by splitting the Tx and Rx
rings into 3 sections.  The first is primarily a read only section
containing basic things like the indexes, a pointer to the dev and netdev
structures, and basic information.  The second section contains the stats
and next_to_use and next_to_clean values.  The third section is primarily
unused values that can just be placed at the end of the ring and are not
used in the hot path.

The adapter structure has several sections that are read in the hot path.
In order to improve performance there I am combining the frequent read
hot path items into a single cache line.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoigb: streamline Rx buffer allocation and cleanup
Alexander Duyck [Fri, 26 Aug 2011 07:43:43 +0000 (07:43 +0000)]
igb: streamline Rx buffer allocation and cleanup

This change is meant to streamline the Rx buffer allocation and cleanup.
This is accomplished by reducing the number of writes by only having the Rx
descriptor ring written by software during allocation, and it will only be
read during cleanup.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoigb: drop support for single buffer mode
Alexander Duyck [Fri, 26 Aug 2011 07:43:38 +0000 (07:43 +0000)]
igb: drop support for single buffer mode

This change removes support for single buffer mode from igb and makes the
driver function in packet split always.  The advantage to doing this is
that we can reduce total memory allocation overhead significantly as we
will only need to allocate one 1K slab per packet and then make use of a
reusable half page instead of allocating a 2K slab per packet.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoigb: Update max_frame_size to account for an optional VLAN tag if present
Alexander Duyck [Fri, 26 Aug 2011 07:43:32 +0000 (07:43 +0000)]
igb: Update max_frame_size to account for an optional VLAN tag if present

This patch modifies the max_frame_size in order account for an optional
VLAN tag.  In order to support this we must also increase the
MAX_STD_JUMBO_FRAME_SIZE to account for the 4 extra bytes.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoigb: Update RXDCTL/TXDCTL configurations
Alexander Duyck [Fri, 26 Aug 2011 07:43:27 +0000 (07:43 +0000)]
igb: Update RXDCTL/TXDCTL configurations

This change cleans up the RXDCTL and TXDCTL configurations and optimizes RX
performance by allowing back write-backs on all hardware other than 82576.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agotcp: md5: remove one indirection level in tcp_md5sig_pool
Eric Dumazet [Mon, 12 Sep 2011 20:28:37 +0000 (20:28 +0000)]
tcp: md5: remove one indirection level in tcp_md5sig_pool

tcp_md5sig_pool is currently an 'array' (a percpu object) of pointers to
struct tcp_md5sig_pool. Only the pointers are NUMA aware, but objects
themselves are all allocated on a single node.

Remove this extra indirection to get proper percpu memory (NUMA aware)
and make code simpler.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Eliminate Small Race Condition Window in RX Path
Rasesh Mody [Fri, 16 Sep 2011 15:06:48 +0000 (15:06 +0000)]
bna: Eliminate Small Race Condition Window in RX Path

Change details:
 - In a continuous sequence of ifconfig up/down operations, there is a small
   window of race between bnad_set_rx_mode() and bnad_cleanup_rx() while the
   former tries to access rx_info->rx & the latter sets it to NULL. This race
   could lead to bna_rx_mode_set() being called with a NULL (rx_info->rx)
   pointer and a crash.
 - Hold bnad->bna_lock while setting / unsetting rx_info->rx in bnad_setup_rx()
   & bnad_cleanup_rx(), thereby eliminating the race described above.

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Set Ring Param Fix
Rasesh Mody [Fri, 16 Sep 2011 15:06:47 +0000 (15:06 +0000)]
bna: Set Ring Param Fix

When Rx queue size is changed, queues are torn down and setup with the new queue
size. During this operation, clear promiscuous mode and restore the original
VLAN filter.

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Semaphore Lock Fix
Rasesh Mody [Fri, 16 Sep 2011 15:06:46 +0000 (15:06 +0000)]
bna: Semaphore Lock Fix

Remove a BUG_ON() as it is not required.

Change the unconditional write to release a semaphore to read sem first
and then write. This will eliminate the possibility of sem getting locked
while trying to release it in case if previous sem_get operation failed.

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoixgbe: remove duplicate netif_tx_start_all_queues
Emil Tantilov [Thu, 28 Jul 2011 06:17:09 +0000 (06:17 +0000)]
ixgbe: remove duplicate netif_tx_start_all_queues

netif_tx_start_all_queues() is already called in ixgbe_up_complete, no need
to do it twice.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: fix FCRTL/H register dump for X540
Emil Tantilov [Wed, 27 Jul 2011 04:16:29 +0000 (04:16 +0000)]
ixgbe: fix FCRTL/H register dump for X540

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: cleanup some register reads
Emil Tantilov [Tue, 26 Jul 2011 07:51:41 +0000 (07:51 +0000)]
ixgbe: cleanup some register reads

Remove duplicate inc of hwstats->ruc
Introduce separate loops for 8 and 16 register reads.
Consolidate mac checks under one case.
Make sure registers are cleared on read.

Reported-by: Jonathan Lynch <jonathan.lynch@thenowfactory.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
CC: Jonathan Lynch <jonathan.lynch@thenowfactory.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: Make better use of memory allocations in one-buffer mode w/ RSC
Alexander Duyck [Fri, 26 Aug 2011 09:52:38 +0000 (09:52 +0000)]
ixgbe: Make better use of memory allocations in one-buffer mode w/ RSC

This patch improves the memory utilization with RSC when in one-buffer
mode.  This is accomplished by making the default buffer sizes match up
with the standard memory allocation sizes minus 1K for shared info and
padding overhead.  By doing this CPU utilization when doing large receives
can be reduced by as much as 8%.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: drop adapter from ixgbe_fso call documentation
Alexander Duyck [Thu, 21 Jul 2011 00:40:56 +0000 (00:40 +0000)]
ixgbe: drop adapter from ixgbe_fso call documentation

The adapter structure was removed from the call so it can be dropped from
the ixgbe_fso documentation.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: Add SFP support for missed 82598 PHY
Alexander Duyck [Thu, 21 Jul 2011 00:40:51 +0000 (00:40 +0000)]
ixgbe: Add SFP support for missed 82598 PHY

One of the 82598 phys was not being correctly identified as being SFP.
This change corrects that.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: Add missing code for enabling overheat sensor interrupt
Alexander Duyck [Thu, 21 Jul 2011 00:40:45 +0000 (00:40 +0000)]
ixgbe: Add missing code for enabling overheat sensor interrupt

This change adds a small bit of missing code for enabling the overheat sensor

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: make ixgbe_up and ixgbe_up_complete void functions
Alexander Duyck [Thu, 21 Jul 2011 00:40:40 +0000 (00:40 +0000)]
ixgbe: make ixgbe_up and ixgbe_up_complete void functions

ixgbe_up and ixgbe_up_complete will always return 0.  Since this doesn't
provide any useful information we might as well just make them both void
and save ourselves from having to return an unused value.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agov2 ixgbe: Update packet buffer reservation to correct fdir headroom size
Alexander Duyck [Thu, 21 Jul 2011 00:40:35 +0000 (00:40 +0000)]
v2 ixgbe: Update packet buffer reservation to correct fdir headroom size

This change fixes an issue in which the incorrect amount of headroom was
being reserved for flow director filters.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: remove redundant configuration of tx_sample_rate
Alexander Duyck [Thu, 21 Jul 2011 00:40:30 +0000 (00:40 +0000)]
ixgbe: remove redundant configuration of tx_sample_rate

This change fixes a minor redundancy in that tx_sample_rate was set twice.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgb: convert to ndo_fix_features
Michał Mirosław [Wed, 8 Jun 2011 08:39:40 +0000 (08:39 +0000)]
ixgb: convert to ndo_fix_features

Private rx_csum flags are now duplicate of netdev->features & NETIF_F_RXCSUM.
Removing this needs deeper surgery.

Things noticed:
 - ixgb has RX csum disabled by default
 - HW VLAN acceleration probably can be toggled, but it's left as is
 - the resets on RX csum offload change can probably be avoided
 - there is A LOT of copy-and-pasted code here

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoRevert "Scm: Remove unnecessary pid & credential references in Unix socket's send...
David S. Miller [Fri, 16 Sep 2011 23:34:00 +0000 (19:34 -0400)]
Revert "Scm: Remove unnecessary pid & credential references in Unix socket's send and receive path"

This reverts commit 0856a304091b33a8e8f9f9c98e776f425af2b625.

As requested by Eric Dumazet, it has various ref-counting
problems and has introduced regressions.  Eric will add
a more suitable version of this performance fix.

Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoethtool: Update ethtool_rxnfc::rule_cnt on return from ETHTOOL_GRXCLSRLALL
Ben Hutchings [Tue, 6 Sep 2011 13:52:47 +0000 (13:52 +0000)]
ethtool: Update ethtool_rxnfc::rule_cnt on return from ETHTOOL_GRXCLSRLALL

A user-space process must use ETHTOOL_GRXCLSRLCNT to find the number
of classification rules, then allocate a buffer of the right size,
then use ETHTOOL_GRXCLSRLALL to fill the buffer.  If some other
process inserts or deletes a rule between those two operations,
the user buffer might turn out to be the wrong size.

If it's too small, the return value will be -EMSGSIZE.  But if it's
too large, there is no indication of this.  Fix this by updating
the rule_cnt field on return.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoethtool: Clean up definitions of rule location arrays in RX NFC
Ben Hutchings [Tue, 6 Sep 2011 13:49:12 +0000 (13:49 +0000)]
ethtool: Clean up definitions of rule location arrays in RX NFC

Correct the description of ethtool_rxnfc::rule_locs; it is an array
of currently used locations, not all possible valid locations.

Add note that drivers must not use ethtool_rxnfc::rule_locs.

The rule_locs argument to ethtool_ops::get_rxnfc is either NULL or a
pointer to an array of u32, so change the parameter type accordingly.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoethtool: Explicitly state that RX NFC rule locations are priorities
Ben Hutchings [Tue, 6 Sep 2011 13:48:56 +0000 (13:48 +0000)]
ethtool: Explicitly state that RX NFC rule locations are priorities

The location of an RX flow classification rule is needed to identify
it for retrieval, replacement or deletion.  However it also defines
the priority of the rule in the case that a flow is matched by
multiple rules.  This is what I intended to imply by referring to the
use of a TCAM, commonly used to implement that behaviour.

However there are other ways this can be done, and it is better to
specify this explicitly.  Further, I want to add the option for
automatic selection of rule locations.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoethtool: Make struct ethtool_rxnfc kernel-doc more self-consistent
Ben Hutchings [Tue, 6 Sep 2011 13:48:20 +0000 (13:48 +0000)]
ethtool: Make struct ethtool_rxnfc kernel-doc more self-consistent

Refer consistently to 'classification rules' or just 'rules' rather
than 'filter specifications' or 'filter rules'.

Refer consistently to rule 'locations' and not 'indices'.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agovxge: make function table const
stephen hemminger [Fri, 16 Sep 2011 11:10:01 +0000 (11:10 +0000)]
vxge: make function table const

All tables of function pointers should be const.
The pre-existing code has lots of needless indirection...

Inspired by similar change in PAX.
Compile tested only.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: make function tables cont
stephen hemminger [Fri, 16 Sep 2011 11:09:51 +0000 (11:09 +0000)]
bna: make function tables cont

To prevent malicious usage, all tables of pointers must be const.

Compile tested only.
Gleaned for PAX.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agopcnet32: constify function table
stephen hemminger [Fri, 16 Sep 2011 11:06:26 +0000 (11:06 +0000)]
pcnet32: constify function table

Function tables need to be const to prevent malicious use.

This is compile tested only.
Gleaned from PAX.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agowan: make LAPB callbacks const
stephen hemminger [Fri, 16 Sep 2011 11:04:29 +0000 (11:04 +0000)]
wan: make LAPB callbacks const

This is compile tested only.
Suggested by dumpster diving in PAX.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoipv6: Send ICMPv6 RSes only when RAs are accepted
Tore Anderson [Sun, 28 Aug 2011 23:47:33 +0000 (23:47 +0000)]
ipv6: Send ICMPv6 RSes only when RAs are accepted

This patch improves the logic determining when to send ICMPv6 Router
Solicitations, so that they are 1) always sent when the kernel is
accepting Router Advertisements, and 2) never sent when the kernel is
not accepting RAs. In other words, the operational setting of the
"accept_ra" sysctl is used.

The change also makes the special "Hybrid Router" forwarding mode
("forwarding" sysctl set to 2) operate exactly the same as the standard
Router mode (forwarding=1). The only difference between the two was
that RSes was being sent in the Hybrid Router mode only. The sysctl
documentation describing the special Hybrid Router mode has therefore
been removed.

Rationale for the change:

Currently, the value of forwarding sysctl is the only thing determining
whether or not to send RSes. If it has the value 0 or 2, they are sent,
otherwise they are not. This leads to inconsistent behaviour in the
following cases:

* accept_ra=0, forwarding=0
* accept_ra=0, forwarding=2
* accept_ra=1, forwarding=2
* accept_ra=2, forwarding=1

In the first three cases, the kernel will send RSes, even though it will
not accept any RAs received in reply. In the last case, it will not send
any RSes, even though it will accept and process any RAs received. (Most
routers will send unsolicited RAs periodically, so suppressing RSes in
the last case will merely delay auto-configuration, not prevent it.)

Also, it is my opinion that having the forwarding sysctl control RS
sending behaviour (completely independent of whether RAs are being
accepted or not) is simply not what most users would intuitively expect
to be the case.

Signed-off-by: Tore Anderson <tore@fud.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agocan-gw: add netlink based CAN routing
Oliver Hartkopp [Thu, 1 Sep 2011 04:23:23 +0000 (04:23 +0000)]
can-gw: add netlink based CAN routing

This patch adds a CAN Gateway/Router to route (and modify) CAN frames.

It is based on the PF_CAN core infrastructure for msg filtering and msg
sending and can optionally modify routed CAN frames on the fly.
CAN frames can *only* be routed between CAN network interfaces (one hop).
They can be modified with AND/OR/XOR/SET operations as configured by the
netlink configuration interface known e.g. from iptables. From the netlink
view this can-gw implements RTM_{NEW|DEL|GET}ROUTE for PF_CAN.

The CAN specific userspace tool to manage CAN routing entries can be found in
the CAN utils http://svn.berlios.de/wsvn/socketcan/trunk/can-utils/cangw.c
at the SocketCAN SVN on BerliOS.

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agosfc: Use correct fields of struct ethtool_coalesce
Ben Hutchings [Mon, 5 Sep 2011 07:43:49 +0000 (07:43 +0000)]
sfc: Use correct fields of struct ethtool_coalesce

An earlier developer misunderstood the meaning of the 'irq' fields and
the driver did not support the standard fields.  To avoid invalidating
existing user documentation, we report and accept changes through
either the standard or 'irq' fields.  If both are changed at the same
time, we prefer the standard field.

Also explain why we don't currently use the 'max_frames' fields.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agosfc: Validate IRQ moderation parameters in efx_init_irq_moderation()
Ben Hutchings [Mon, 5 Sep 2011 07:43:04 +0000 (07:43 +0000)]
sfc: Validate IRQ moderation parameters in efx_init_irq_moderation()

Add a range check, and move the check that RX and TX are consistent
from efx_ethtool_set_coalesce().

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agosfc: Correct reporting and validation of TX interrupt coalescing
Ben Hutchings [Mon, 5 Sep 2011 07:42:25 +0000 (07:42 +0000)]
sfc: Correct reporting and validation of TX interrupt coalescing

The reported TX IRQ moderation is generated in a completely crazy way.
Make it simple and correct.

When channels are shared between RX and TX, TX IRQ moderation must be
the same as RX IRQ moderation, but must be specified as 0!  Allow it
to be either specified as the same, or left at its previous value
in which case it will be quietly overridden.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agosfc: Use consistent types for interrupt coalescing parameters
Ben Hutchings [Mon, 5 Sep 2011 07:41:44 +0000 (07:41 +0000)]
sfc: Use consistent types for interrupt coalescing parameters

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agosfc: Correct error code for unsupported interrupt coalescing parameters
Ben Hutchings [Mon, 5 Sep 2011 07:41:27 +0000 (07:41 +0000)]
sfc: Correct error code for unsupported interrupt coalescing parameters

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoMerge git://github.com/Jkirsher/net-next
David S. Miller [Fri, 16 Sep 2011 19:18:02 +0000 (15:18 -0400)]
Merge git://github.com/Jkirsher/net-next

13 years agoMerge branch 'for-davem' of git://git.infradead.org/users/linville/wireless-next
David S. Miller [Fri, 16 Sep 2011 19:14:19 +0000 (15:14 -0400)]
Merge branch 'for-davem' of git://git.infradead.org/users/linville/wireless-next

13 years agoMerge branch 'master' of ../netdev-next/
David S. Miller [Fri, 16 Sep 2011 06:58:54 +0000 (02:58 -0400)]
Merge branch 'master' of ../netdev-next/

13 years agoixgbe: Correctly name and handle MSI-X other interrupt
Alexander Duyck [Fri, 15 Jul 2011 07:29:55 +0000 (07:29 +0000)]
ixgbe: Correctly name and handle MSI-X other interrupt

It was possible to inadvertently add additional interrupt causes to the
MSI-X other interrupt.  This occurred when things such as RX buffer overrun
events were being triggered at the same time as an event such as a Flow
Director table reinit request.  In order to avoid this we should be
explicitly programming only the interrupts that we want enabled.  In
addition I am renaming the ixgbe_msix_lsc function and interrupt to drop
any implied meaning of this being a link status only interrupt.

Unfortunately the patch is a bit ugly due to the fact that ixgbe_irq_enable
needed to be moved up before ixgbe_msix_other in order to have things
defined in the correct order.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: cleanup configuration of EITRSEL and VF reset path
Alexander Duyck [Fri, 15 Jul 2011 07:29:49 +0000 (07:29 +0000)]
ixgbe: cleanup configuration of EITRSEL and VF reset path

This change is meant to cleanup some of the code related to SR-IOV and the
interrupt registers.  Specifically I am moving the EITRSEL configuration
into the MSI-X configuration section instead of enablement.  Also I am
fixing the VF shutdown path since it had operations in the incorrect order.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: cleanup reset paths
Alexander Duyck [Fri, 15 Jul 2011 07:29:44 +0000 (07:29 +0000)]
ixgbe: cleanup reset paths

The reset paths are overly complicated and are either missing steps or
contain extra unnecessary steps such as reading MAC address twice.  This
change is meant to help clean up the reset paths an get things functioning
correctly.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: Update TXDCTL configuration to correctly handle WTHRESH
Alexander Duyck [Fri, 15 Jul 2011 03:06:12 +0000 (03:06 +0000)]
ixgbe: Update TXDCTL configuration to correctly handle WTHRESH

This change updated the TXDCTL configuration.  The main goal is to be much
more explicit about the configuration and avoid a possible fake TX hang
when the interrupt throttle rate is set to 0.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: combine PCI_VDEVICE and board declaration to same line
Alexander Duyck [Fri, 15 Jul 2011 03:06:06 +0000 (03:06 +0000)]
ixgbe: combine PCI_VDEVICE and board declaration to same line

This patch is a minor whitespace cleanup to compress the device ID
declaration and board type declaration onto the same line.  It seems to
make sense since all of the combinations of the two are less than 80
characters and it makes the overall layout a bit more readable.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: Drop unnecessary adapter->hw dereference in loopback test setup
Alexander Duyck [Fri, 15 Jul 2011 03:06:01 +0000 (03:06 +0000)]
ixgbe: Drop unnecessary adapter->hw dereference in loopback test setup

This patch drops a set of unnecessary dereferences to the hardware structure
since we already have a local copy of the hardware pointer.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: commonize ixgbe_map_rings_to_vectors to work for all interrupt types
Alexander Duyck [Fri, 15 Jul 2011 03:05:51 +0000 (03:05 +0000)]
ixgbe: commonize ixgbe_map_rings_to_vectors to work for all interrupt types

This patch makes it so that the map_rings_to_vectors call will work with
all interrupt types.  The advantage to this is that there will now be a
predictable mapping for all given interrupt types.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: Use ring->dev instead of adapter->pdev->dev when updating DCA
Alexander Duyck [Fri, 15 Jul 2011 03:05:46 +0000 (03:05 +0000)]
ixgbe: Use ring->dev instead of adapter->pdev->dev when updating DCA

This change switches us over to using the ring->dev pointer instead of
having to use the adapter->pdev->dev reference.  The advantage to this is
that it is a much shorter route to get the to final needed value.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: cleanup allocation and freeing of IRQ affinity hint
Alexander Duyck [Fri, 15 Jul 2011 03:05:37 +0000 (03:05 +0000)]
ixgbe: cleanup allocation and freeing of IRQ affinity hint

The allocation and freeing of the IRQ affinity hint needs some updates
since there are a number of spots where we run into possible issues with
the hint not being correctly updated.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agov2 ixgbe: consolidate all MSI-X ring interrupts and poll routines into one
Alexander Duyck [Wed, 31 Aug 2011 00:01:11 +0000 (00:01 +0000)]
v2 ixgbe: consolidate all MSI-X ring interrupts and poll routines into one

This change consolidates all of the MSI-X interrupt and polling routines
into two single functions.  One for the interrupt and one for the code.
The main advantage to doing this is that the compiler can optimize the
routines into single monolithic functions which should allow all of them
function to occupy a single block of memory and as such avoid jumping
around.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agoixgbe: Change default Tx work limit size to 256 buffers
Alexander Duyck [Wed, 31 Aug 2011 00:01:06 +0000 (00:01 +0000)]
ixgbe: Change default Tx work limit size to 256 buffers

This change makes it so that the default Tx work limit is 256 buffers or
1/2 of an entire ring instead of a full ring size so that it is much more
likely that we will be able to actually reach the work limit value.
Previously with the value set to an entire ring it would not have been
possible for us to trigger an event due to the fact that the Tx work is
stopped at the point where we cannot place one more buffer on the ring and
it is not restarted until cleanup is complete.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
13 years agonet: consolidate and fix ethtool_ops->get_settings calling
Jiri Pirko [Sat, 3 Sep 2011 03:34:30 +0000 (03:34 +0000)]
net: consolidate and fix ethtool_ops->get_settings calling

This patch does several things:
- introduces __ethtool_get_settings which is called from ethtool code and
  from drivers as well. Put ASSERT_RTNL there.
- dev_ethtool_get_settings() is replaced by __ethtool_get_settings()
- changes calling in drivers so rtnl locking is respected. In
  iboe_get_rate was previously ->get_settings() called unlocked. This
  fixes it. Also prb_calc_retire_blk_tmo() in af_packet.c had the same
  problem. Also fixed by calling __dev_get_by_index() instead of
  dev_get_by_index() and holding rtnl_lock for both calls.
- introduces rtnl_lock in bnx2fc_vport_create() and fcoe_vport_create()
  so bnx2fc_if_create() and fcoe_if_create() are called locked as they
  are from other places.
- use __ethtool_get_settings() in bonding code

Signed-off-by: Jiri Pirko <jpirko@redhat.com>
v2->v3:
-removed dev_ethtool_get_settings()
-added ASSERT_RTNL into __ethtool_get_settings()
-prb_calc_retire_blk_tmo - use __dev_get_by_index() and lock
 around it and __ethtool_get_settings() call
v1->v2:
        add missing export_symbol
Reviewed-by: Ben Hutchings <bhutchings@solarflare.com> [except FCoE bits]
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agonet/irda: sh_irda: add PM support
Kuninori Morimoto [Wed, 31 Aug 2011 20:08:48 +0000 (20:08 +0000)]
net/irda: sh_irda: add PM support

Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agonet/irda: sh_irda: update author's email address
Kuninori Morimoto [Wed, 31 Aug 2011 20:08:37 +0000 (20:08 +0000)]
net/irda: sh_irda: update author's email address

it also cleanup white space

Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agonet/irda: sh_irda: add sh_irda_ prefix to all functions
Kuninori Morimoto [Wed, 31 Aug 2011 20:08:22 +0000 (20:08 +0000)]
net/irda: sh_irda: add sh_irda_ prefix to all functions

Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agotg3: Code movement
Matt Carlson [Wed, 31 Aug 2011 11:44:54 +0000 (11:44 +0000)]
tg3: Code movement

This patch just moves some code around for better organization.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agotg3: Eliminate tg3_halt_cpu() prototype
Matt Carlson [Wed, 31 Aug 2011 11:44:53 +0000 (11:44 +0000)]
tg3: Eliminate tg3_halt_cpu() prototype

This patch moves the implementatino of tg3_halt_cpu() earlier in the
file to eliminate its prototype.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agotg3: Eliminate tg3_write_sig_post_reset() prototype
Matt Carlson [Wed, 31 Aug 2011 11:44:52 +0000 (11:44 +0000)]
tg3: Eliminate tg3_write_sig_post_reset() prototype

This patch moves the implementation of tg3_write_sig_post_reset()
earlier to eliminate its prototype.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agotg3: Eliminate tg3_stop_fw() prototype
Matt Carlson [Wed, 31 Aug 2011 11:44:51 +0000 (11:44 +0000)]
tg3: Eliminate tg3_stop_fw() prototype

This patch moves tg3_stop_fw() earlier in the file to eliminate its
prototype.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agotg3: Add ability to turn off 1shot MSI
Matt Carlson [Wed, 31 Aug 2011 11:44:50 +0000 (11:44 +0000)]
tg3: Add ability to turn off 1shot MSI

In older devices, 1-shot MSI mode had to be enabled by the code.  In
newer devices however, 1-shot MSI mode is enabled by default; code would
be needed to disable it.

Disabling 1-shot MSI mode is useful when debugging.  This patch changes
the code so that the TG3_FLAG_1SHOT_MSI accurately reflects (and
controls) the state of 1-shot MSI mode.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agotg3: Remove tp->rx_offset term when unneeded
Matt Carlson [Wed, 31 Aug 2011 11:44:49 +0000 (11:44 +0000)]
tg3: Remove tp->rx_offset term when unneeded

This patch removes the tp->rx_offset term if NET_IP_ALIGN is defined to
zero.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agotg3: Fix missed MSI workaround
Matt Carlson [Wed, 31 Aug 2011 11:44:48 +0000 (11:44 +0000)]
tg3: Fix missed MSI workaround

This patch fixes a minor counter initialization bug and makes the MSI
workaround slightly more efficient by attempting to service pending
interrupts before applying the workaround.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agotg3: Check all adv bits when checking config
Matt Carlson [Wed, 31 Aug 2011 11:44:47 +0000 (11:44 +0000)]
tg3: Check all adv bits when checking config

This patch makes sure the driver checks all advertisement bits when
checking the current hw advertisements.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Benjamin Li <benli@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoMII: fix Kconfig dependencies for MII
Jeff Kirsher [Wed, 14 Sep 2011 21:23:14 +0000 (21:23 +0000)]
MII: fix Kconfig dependencies for MII

MII Kconfig option is apart of the core networking drivers and
by default NET_CORE is enabled so drivers selecting MII will
have MII enabled as well.  It was found using the randconfig
option during testing, MII would be selected but NET_CORE
could be disabled.  This caused a dependency error.

Resolved the dependency by selecting NET_CORE when MII is
selected.

Reported-by: Emil Tantilov <emil.s.tantilov@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agostmmac: update the doc with new info about the driver's debug (v3)
Giuseppe CAVALLARO [Thu, 1 Sep 2011 21:51:42 +0000 (21:51 +0000)]
stmmac: update the doc with new info about the driver's debug (v3)

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agostmmac: update the driver version (Aug_2011) (v3)
Giuseppe CAVALLARO [Thu, 1 Sep 2011 21:51:43 +0000 (21:51 +0000)]
stmmac: update the driver version (Aug_2011) (v3)

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agostmmac: add HW DMA feature register (v3)
Giuseppe CAVALLARO [Thu, 1 Sep 2011 21:51:41 +0000 (21:51 +0000)]
stmmac: add HW DMA feature register (v3)

New GMAC chips have an extra register to indicate
the presence of the optional features/functions of
the DMA core.

This patch adds this support and all the HW cap
are exported via debugfs.

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agostmmac: rework the code to get the Synopsys ID (v3)
Giuseppe CAVALLARO [Thu, 1 Sep 2011 21:51:40 +0000 (21:51 +0000)]
stmmac: rework the code to get the Synopsys ID (v3)

The Synopsys ID is now passed from the MAC core
to the main. This info will be used for managing
the HW cap register (supported in the new GMAC
generations).

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agostmmac: export DMA TX/RX rings via debugfs (v3)
Giuseppe CAVALLARO [Thu, 1 Sep 2011 21:51:39 +0000 (21:51 +0000)]
stmmac: export DMA TX/RX rings via debugfs (v3)

This patch adds the following debugFs entry to dump the
RX/TX DMA rings:

/sys/kernel/debug/stmmaceth/descriptors_status

This is an example:

=======================
  RX descriptor ring
=======================
[0] DES0=0x85ee0320 DES1=0x1fff1fff BUF1=0x5fae2022 BUF2=0x0
[1] DES0=0x85ee0320 DES1=0x1fff1fff BUF1=0x5fae0022 BUF2=0x0
[2] DES0=0x81460320 DES1=0x1fff1fff BUF1=0x5f9dd022 BUF2=0x0

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agostmmac: add MMC support exported via ethtool (v3)
Giuseppe CAVALLARO [Thu, 1 Sep 2011 21:51:38 +0000 (21:51 +0000)]
stmmac: add MMC support exported via ethtool (v3)

This patch adds the MMC management counters support.
MMC module is an extension of the register address
space and all the hardware counters can be accessed
via ethtoo -S ethX.

Note that, the MMC interrupts remain masked and the logic
to handle this kind of interrupt will be added later (if
actually useful).

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agostmmac: support wake up irq from external sources (v3)
Deepak Sikri [Thu, 1 Sep 2011 21:51:37 +0000 (21:51 +0000)]
stmmac: support wake up irq from external sources (v3)

On some platforms e.g. SPEAr the wake up irq differs from the
GMAC interrupt source.
With this patch an external wake up irq can be passed through the
platform code and named as "eth_wake_irq".

In case the wake up interrupt is not passed from the platform
so the driver will continue to use the mac irq (ndev->irq)

Signed-off-by: Deepak Sikri <deepak.sikri@st.com>
Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agostmmac: remove the mmc code (v3)
Giuseppe CAVALLARO [Thu, 1 Sep 2011 21:51:36 +0000 (21:51 +0000)]
stmmac: remove the mmc code (v3)

DWMAC Management Counters (MMC) are not fully support.
The minimal support added in the past allowed to
only disable counters (if present) and mask their
interrupts.
This patch prepares the driver to support the MMC
removing obsolete code.

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agostmmac: remove the STBus bridge setting from the GMAC code (v3)
Giuseppe CAVALLARO [Thu, 1 Sep 2011 21:51:35 +0000 (21:51 +0000)]
stmmac: remove the STBus bridge setting from the GMAC code (v3)

This patch removes a piece of code (actually commented)
only useful for some ST platforms in the past.

This kind of setting now can be done by using the platform
callbacks provided in linux/stmmac.h (see the stmmac.txt for
further details).

Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years ago6LoWPAN: fix skb_copy call
alex.bluesman.smirnov@gmail.com [Thu, 1 Sep 2011 03:55:15 +0000 (03:55 +0000)]
6LoWPAN: fix skb_copy call

This patch fixes 2 issues in lowpan_skb_deliver function:
1. Check for return status of skb_copy call;
2. Use skb_copy with proper GFP flag, drop check for non-interrupt
context.

Signed-off-by: Alexander Smirnov <alex.bluesman.smirnov@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobr: remove redundant check and init
Jiri Pirko [Thu, 1 Sep 2011 03:29:38 +0000 (03:29 +0000)]
br: remove redundant check and init

Since these checks and initialization are done in
dev_ethtool_get_settings called later on, remove this redundancy.

Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agomii: Remove references to DP83840 PHY in mii.h
Mark Einon [Wed, 31 Aug 2011 23:22:17 +0000 (23:22 +0000)]
mii: Remove references to DP83840 PHY in mii.h

There are references to this PHY chip in the generic mii.h header, so removing them.
Re-jiggle the changed comments, in response to points raised by Ben Hutchings.

Signed-off-by: Mark Einon <mark.einon@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agomii: Convert spaces to tabs in mii.h
Mark Einon [Wed, 31 Aug 2011 23:22:16 +0000 (23:22 +0000)]
mii: Convert spaces to tabs in mii.h

Whitespace changes - spaces converted to tabs after each define name and value

Signed-off-by: Mark Einon <mark.einon@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agonet: linkwatch: allow vlans to get carrier changes faster
Eric Dumazet [Tue, 30 Aug 2011 23:31:58 +0000 (23:31 +0000)]
net: linkwatch: allow vlans to get carrier changes faster

There is a time-lag of IFF_RUNNING flag consistency between vlan and
real devices when the real devices are in problem such as link or cable
broken.

This leads to a degradation of Availability such as a delay of failover
in HA systems using vlan since the detection of the problem at real
device is delayed.

We can avoid the linkwatch delay (~1 sec) for devices linked to another
ones, since delay is already done for the realdev.

Based on a previous patch from Mitsuo Hayasaka

Reported-by: Mitsuo Hayasaka <mitsuo.hayasaka.hu@hitachi.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Patrick McHardy <kaber@trash.net>
Cc: "Michał Mirosław" <mirq-linux@rere.qmqm.pl>
Cc: Tom Herbert <therbert@google.com>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Cc: Jesse Gross <jesse@nicira.com>
Tested-by: Mitsuo Hayasaka <mitsuo.hayasaka.hu@hitachi.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Driver Version changed to 3.0.2.1
Rasesh Mody [Tue, 30 Aug 2011 15:27:48 +0000 (15:27 +0000)]
bna: Driver Version changed to 3.0.2.1

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: SKB PCI UNMAP Fix
Rasesh Mody [Tue, 30 Aug 2011 15:27:47 +0000 (15:27 +0000)]
bna: SKB PCI UNMAP Fix

Change details:
 - Found a leak in sk_buff unmapping of PCI dma addresses where boundary
   conditions are not properly handled in freeing all Tx buffers. Freeing
   of all Tx buffers is done considering sk_buffs data and fragments can
   be mapped at the boundary.

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: TX Queue Depth Fix
Rasesh Mody [Tue, 30 Aug 2011 15:27:46 +0000 (15:27 +0000)]
bna: TX Queue Depth Fix

sk_buff unmap_array grows greater than 65536 (x2) with Tx ring of 65536.
Reducing TXQ depth and safe(max) acking of Tx events to 32768 (same as Rx).
Add defines for TX and RX queue depths.

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: MBOX IRQ Flag Check after Locking
Rasesh Mody [Tue, 30 Aug 2011 15:27:45 +0000 (15:27 +0000)]
bna: MBOX IRQ Flag Check after Locking

Change details:
 - Check the BNAD_RF_MBOX_IRQ_DISABLED flag after acquiring the bna_lock,
   since checking the flag and executing bna_mbox_handler needs to be atomic.
   If not, it opens up window where flag is reset when it was checked, but got
   set while spinning on the lock by the other thread which is actually
   holding the lock

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Async Mode Tx Rx Init Fix
Rasesh Mody [Tue, 30 Aug 2011 15:27:44 +0000 (15:27 +0000)]
bna: Async Mode Tx Rx Init Fix

Change details:
 - Async mode of Tx/Rx queue initialization in BNAD from a task queue context
   runs into non-unique taskq allocation issues. Get rid of Tx/Rx
   initialization from task q context
 - In the attach function, wait for IOC enable, then do Tx/Rx queue
   initialization. Default BNA attributes are used when IOC enable from attach
   fails and values are set to:
   1 TxQ, 1 RxQ, 1 Unicast MAC, 1 RIT entry

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Ethtool Enhancements and Fix
Rasesh Mody [Tue, 30 Aug 2011 15:27:43 +0000 (15:27 +0000)]
bna: Ethtool Enhancements and Fix

Change details:
 - Add tx_skb counters and NAPI debug counters to ethtool stats.
 - Add rlb stats strings to bnad_net_stats_strings{} array. rlb_stats field
   was added to struct bfi_enet_stats {} but the corresponding name structure
   array for ethtool was not initialized with right strings, even though the
   actual name structure array got expanded. This caused a NULL pointer
   violation and a crash when doing ehtool -S <if_name>.
 - Modify dim timer stop logic to make it dependent on cfg and run flags
 - While setting the ring parameter restore the rx, vlan configuration and
   set rx mode
 - Indentation fix

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Initialization and Locking Fix
Rasesh Mody [Tue, 30 Aug 2011 15:27:42 +0000 (15:27 +0000)]
bna: Initialization and Locking Fix

Change details:
 - Initialize rx_id to 0 for bnad_cleanup_rx
 - Return -ENOMEM in case if bna_rx_create fails
 - Count the Rx buffer allocation failures in bnad_alloc_n_post_rxbufs()
 - Remove unnecessary initialization of using_dac to false in bnad_pci_probe
 - Release lock if error while doing bna_num_txq_set in bnad_pci_probe

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Formatting and Code Cleanup
Rasesh Mody [Tue, 30 Aug 2011 15:27:41 +0000 (15:27 +0000)]
bna: Formatting and Code Cleanup

Change details:
 - Print log messages when running with reduced number of MSI-X vectors
   and when defaulting to INTx mode.
 - Remove BUG_ONs and header file inclusion that are not needed
 - Comments addition/cleanup
 - Unused code cleanup
 - Add New Line to Print msg in bfa_sm_fault
 - Formatting fix

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: TX Path and RX Path Changes
Rasesh Mody [Tue, 30 Aug 2011 15:27:40 +0000 (15:27 +0000)]
bna: TX Path and RX Path Changes

Change details:
 - Add bnad_pci_unmap_skb() API to unmap skb from transmit path and update the
   unmap index. Add more checks for illegal skbs in transmit path. Add tx_skb
   counters for dropped skbs.
 - The unmap_cons index used in bnad_free_txbufs() is incorrectly declared as
   u16. It quickly wraps around and accesses null sk_buff ptr. So using u32 to
   handle unmap_array.
 - Disable and enable interrupts from the same polling context to prevent
   reordering in Rx path.
 - Add Rx NAPI debug counters.
 - Make NAPI budget check more generic.
 - Modify dim timer stop logic to make it dependent on cfg and run flags
 - Handle reduced MSI-X vectors case in bnad_enable_msix.
 - Check for single frame TSO skbs and send them out as non-TSO.
 - Put memory barrier after bna_txq_prod_indx_doorbell().

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Interrupt Polling and NAPI Init Changes
Rasesh Mody [Tue, 30 Aug 2011 15:27:39 +0000 (15:27 +0000)]
bna: Interrupt Polling and NAPI Init Changes

Change details:
 - Remove unnecessary ccb check from bnad_poll_cq
 - Add bnad pointer to rx_ctrl structure, so that bnad can be accessed directly
   from rx_ctrl in the NAPI poll routines, even if ccb is NULL
 - Validate ccb before referencing to it in bnad_msix_rx and bnad_napi_poll_rx
 - Fix the order of NAPI init / uninit in Tx / Rx setup / teardown path:
   a. Kill bnad tx free tasklet ahead of call to bna_tx_destroy()
   b. Call NAPI disable only after call to Rx free_irq(). This makes sure Rx
      interrupt does not schedule a poll when NAPI is already disabled
 - NAPI poll runs before the h/w has completed configuration. This causes a
   crash. Delay enabling NAPI till after bna_rx_enable(). Split NAPI
   initialization into 2 steps, bnad_napi_init() & bnad_napi_enable().

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: PCI Probe Fix
Rasesh Mody [Tue, 30 Aug 2011 15:27:38 +0000 (15:27 +0000)]
bna: PCI Probe Fix

Change details:
 - Return error as -EIO if bnad_res_alloc fails
 - Release the configuration lock before registering with net_device layer.

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agobna: Naming Change and Minor Macro Fix
Rasesh Mody [Tue, 30 Aug 2011 15:27:37 +0000 (15:27 +0000)]
bna: Naming Change and Minor Macro Fix

Naming changes: rename devid, BNAD_MAX_TXS, BNAD_MAX_RXS,
BNAD_MAX_RXPS_PER_RX to device, BNAD_MAX_TX, BNAD_MAX_RX,
BNAD_MAX_RXP_PER_RX respectively and change all the references.

Macro Fix: Add ioc_isr_mod_set check to bfa_nw_ioc_mbox_regisr macro

Signed-off-by: Gurunatha Karaje <gkaraje@brocade.com>
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agonet, rds, Replace xlist in net/rds/xlist.h with llist
Huang Ying [Tue, 30 Aug 2011 15:21:30 +0000 (15:21 +0000)]
net, rds, Replace xlist in net/rds/xlist.h with llist

The functionality of xlist and llist is almost same.  This patch
replace xlist with llist to avoid code duplication.

Known issues: don't know how to test this, need special hardware?

Signed-off-by: Huang Ying <ying.huang@intel.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Andy Grover <andy.grover@oracle.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years ago6LoWPAN: call dev_put() on error in lowpan_newlink()
Dan Carpenter [Tue, 30 Aug 2011 03:51:09 +0000 (03:51 +0000)]
6LoWPAN: call dev_put() on error in lowpan_newlink()

We should release the dev_hold() on error before returning here.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years ago6LoWPAN: use the _safe version of list_for_each
Dan Carpenter [Tue, 30 Aug 2011 03:46:40 +0000 (03:46 +0000)]
6LoWPAN: use the _safe version of list_for_each

When we kfree(entry) that causes a use-after-free bug so we have to
use list_for_each_entry_safe() safe here.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years ago6LoWPAN: use kfree_skb() instead of kfree()
Dan Carpenter [Tue, 30 Aug 2011 03:45:52 +0000 (03:45 +0000)]
6LoWPAN: use kfree_skb() instead of kfree()

Use kfree_skb() to free sbk_buffs.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agor8169: convert to SKB paged frag API.
Ian Campbell [Wed, 31 Aug 2011 00:47:06 +0000 (00:47 +0000)]
r8169: convert to SKB paged frag API.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Realtek linux nic maintainers <nic_swsd@realtek.com>
Cc: Francois Romieu <romieu@fr.zoreil.com>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoqlge: convert to SKB paged frag API.
Ian Campbell [Wed, 31 Aug 2011 00:47:05 +0000 (00:47 +0000)]
qlge: convert to SKB paged frag API.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Cc: Jitendra Kalsaria <jitendra.kalsaria@qlogic.com>
Cc: Ron Mercer <ron.mercer@qlogic.com>
Cc: linux-driver@qlogic.com
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
13 years agoqlcnic: convert to SKB paged frag API.
Ian Campbell [Wed, 31 Aug 2011 00:47:04 +0000 (00:47 +0000)]
qlcnic: convert to SKB paged frag API.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Cc: Sony Chacko <sony.chacko@qlogic.com>
Cc: linux-driver@qlogic.com
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>