- 22 May, 2020 40 commits
-
-
Bartosz Golaszewski authored
We'll soon by adding a second MediaTek Ethernet driver so modify the Kconfig prompt. Signed-off-by:
Bartosz Golaszewski <bgolaszewski@baylibre.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Bartosz Golaszewski authored
This adds yaml DT bindings for the MediaTek STAR Ethernet MAC present on the mt8* family of SoCs. Signed-off-by:
Bartosz Golaszewski <bgolaszewski@baylibre.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Bartosz Golaszewski authored
The PERICFG controller is present on the MT8516 SoC. Add an appropriate compatible variant. Signed-off-by:
Bartosz Golaszewski <bgolaszewski@baylibre.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Bartosz Golaszewski authored
Convert the DT binding .txt file for MediaTek's peripheral configuration controller to YAML. There's one special case where the compatible has three positions. Otherwise, it's a pretty normal syscon. Signed-off-by:
Bartosz Golaszewski <bgolaszewski@baylibre.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Arthur Kiyanovski says: ==================== ENA features and cosmetic changes Diff from V1 of this patchset: Removed error prints patch This patchset includes: 1. new rx offset feature 2. reduction of the driver load time 3. multiple cosmetic changes to the code ==================== Acked-by:
Jakub Kicinski <kuba@kernel.org> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
This commit reduces the driver load time by using usec resolution instead of msec when polling for hardware state change. Also add back-off mechanism to handle cases where minimal sleep time is not enough. Signed-off-by:
Josh Triplett <josh@joshtriplett.org> Signed-off-by:
Sameeh Jubran <sameehj@amazon.com> Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
1. Use BIT macro instead of shift operator for code clarity 2. Replace multiple flag assignments to a single assignment of multiple flags in ena_com_add_single_rx_desc() 3. Move ENA_HASH_KEY_SIZE from ena_netdev.h to ena_com.h Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
1. Add leading and trailing spaces to several comments for better readability 2. Make tabs and spaces uniform in enum defines in ena_admin_defs.h Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
1. Reorder sanity checks in get_comp_ctxt() to make more sense 2. Reorder variables in ena_com_fill_hash_function() and ena_calc_io_queue_size() in reverse christmas tree. 3. Move around member initializations. Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
1. Remove unused definition of DRV_MODULE_VERSION 2. Remove {} from single line-of-code ifs 3. Remove unnecessary comments from ena_get/set_coalesce() 4. Remove unnecessary extra spaces and newlines Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
1. Join unnecessarily broken short lines in ena_com.c ena_netdev.c 2. Fix Indentations of broken lines Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
fix spelling and grammar mistakes in comments in ena_com.h, ena_com.c and ena_netdev.c Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
Make all types of variables that convey the number and sizeof queues to be u32, for consistency with the API between the driver and device via ena_admin_defs.h:ena_admin_get_feat_resp.max_queue_ext fields. Current code sometimes uses int and there are multiple assignments between these variables with different types. Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
Rename ena_update_tx/rx_rings_intr_moderation() to ena_update_tx/rx_rings_nonadaptive_intr_moderation() to distinguish between adaptive and non adaptive interrupt moderaion. Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
Initialize prev_intr_delay_resolution with ena_dev->intr_delay_resolution unconditionally, since it is initialized with ENA_DEFAULT_INTR_DELAY_RESOLUTION in ena_probe(). This approach makes much more sense than handling errors of not initializing it. Also added unlikely to if condition. Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
Default return value should be -EINVAL since the input in this case was unexpected. Also remove the now redundant check in the beginning of the function. Signed-off-by:
Sameeh Jubran <sameehj@amazon.com> Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
Use u64 instead of unsigned long long for clarity Signed-off-by:
Shai Brandes <shaibran@amazon.com> Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
Rename ena_com_free_desc to ena_com_free_q_entries to match the LLQ mode. In non-LLQ mode, an entry in an IO ring corresponds to a a descriptor. In LLQ mode an entry may correspond to several descriptors (per LLQ definition). Signed-off-by:
Igor Chauskin <igorch@amazon.com> Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Arthur Kiyanovski authored
Newer ENA devices can write data to rx buffers with an offset from the beginning of the buffer. This commit adds support for this feature in the driver. Signed-off-by:
Sameeh Jubran <sameehj@amazon.com> Signed-off-by:
Arthur Kiyanovski <akiyano@amazon.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Igor Russkikh says: ==================== net: atlantic: QoS implementation This patch series adds support for mqprio rate limiters and multi-TC: * max_rate is supported on both A1 and A2; * min_rate is supported on A2 only; This is a joint work of Mark and Dmitry. To implement this feature, a couple of rearrangements and code improvements were done, in areas of TC/ring management, allocation control, etc. One of the problems we faced is conflicting ptp functionality, which consumes a whole traffic class due to hardware limitations. Patches below have a more detailed description on how PTP and multi-TC co-exist right now. v2: * accommodated review comments (-Wmissing-prototypes and -Wunused-but-set-variable findings); * added user notification in case of conflicting multi-TC<->PTP configuration; * added automatic PTP disabling, if a conflicting configuration is detected; * removed module param, which was used for PTP disabling in v1; v1: https://patchwork.ozlabs.org/cover/1294380/ ==================== Acked-by:
Jakub Kicinski <kuba@kernel.org> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Mark Starovoytov authored
This patch fixes an inconsistency between code and spec, which was found while working on the QoS implementation. When 8TCs are used, 2 is the maximum supported number of index bits. In a 4TC mode, we do support 3, but we shouldn't really use the bytes, which are intended for the 8TC mode. Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Mark Starovoytov authored
This patch adds support for mqprio min_rate limiters. A2 HW supports Weighted Strict Priority (WSP) arbitration for Tx Descriptor Queue scheduling among TCs, which can be used for min_rate shaping. Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Mark Starovoytov authored
This patch changes the order of arguments for TC weight/credit setter functions. Having the "value to be set" on the right is slightly more robust in a sense that it's more natural for the humans, so it's a bit more error-proof this way. Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Mark Starovoytov authored
This patch changes the TC-queue mapping mechanism used on A2. Configure the A2 HW in such a way that we can keep queue index mapping exactly as it was on A1. Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Mark Starovoytov authored
This patch adds support for automatic queue number downgrade. On A2: this is a must have, because only TC0/TC1 support more than 4Q. Other TCs support 4Qs maximum. Thus, on A2 we must downgrade the number of queues per TC to 4, if more than 2 TCs are requested. On A1: this allows using 8TCs even on systems with cpu count >= 8, when we have 8 queues by default. We will just automatically switch to 8TCx4Q mode in this case. Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Mark Starovoytov authored
This patch adds initial support for mqprio rate limiters (max_rate only). Atlantic HW supports Rate-Shaping for time-sensitive traffic at per Traffic Class (TC) granularity. Target rate is defined by: * nominal link rate (always 10G); * rate factor (ratio between nominal rate and max allowed). Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Mark Starovoytov authored
This patch updates TCVEC2RING to accept nic_cfg, which is needed to be able to use it from hw_atl. The name is updated to reflect the changes. Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Mark Starovoytov authored
This patch adds support for per-TC queue statistics. By default (single TC), the output is the same as it used to be, e.g.: Queue[0] InPackets: 2 Queue[0] OutPackets: 8 Queue[0] Restarts: 0 Queue[0] InJumboPackets: 0 Queue[0] InLroPackets: 0 Queue[0] InErrors: 0 If several TCs are enabled, then each queue statistics line is prefixed with TC number, e.g.: TC0 Queue[0] InPackets: 6 TC0 Queue[0] OutPackets: 11 Queue numbering is end-to-end, so: TC1 Queue[4] InPackets: 0 TC1 Queue[4] OutPackets: 22 Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Dmitry Bezrukov authored
This patch adds multi-TC support. PTP is automatically disabled when the user enables more than 2 TCs, otherwise traffic on TC2 won't quite work, because it's reserved for PTP. Signed-off-by:
Dmitry Bezrukov <dbezrukov@marvell.com> Co-developed-by:
Dmitry Bogdanov <dbogdanov@marvell.com> Signed-off-by:
Dmitry Bogdanov <dbogdanov@marvell.com> Co-developed-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Dmitry Bezrukov authored
This patch contains the following changes: * add cfg->is_ptp (used for PTP enable/disable switch, which is described in more details below); * add cfg->tc_mode (A1 supports 2 HW modes only); * setup queue to TC mapping based on TC mode on A2; * remove hw_tx_tc_mode_get / hw_rx_tc_mode_get hw_ops. In the first generation of our hardware (A1), a whole traffic class is consumed for PTP handling in FW (FW uses it to send the ptp data and to send back timestamps). The 'is_ptp' flag introduced in this patch will be used in to automatically disable PTP when a conflicting configuration is detected, e.g. when multiple TCs are enabled. Signed-off-by:
Dmitry Bezrukov <dbezrukov@marvell.com> Co-developed-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Dmitry Bezrukov authored
This patch moves the PTP TC initialization into a separate function. Signed-off-by:
Dmitry Bezrukov <dbezrukov@marvell.com> Co-developed-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Dmitry Bezrukov authored
This patch contains the following changes: * access cfg via aq_nic_get_cfg() in aq_nic_start() and aq_nic_map_skb(); * call aq_nic_get_dev() just once in aq_nic_map_skb(); * move ring allocation/deallocation out of aq_vec_alloc()/aq_vec_free(); * add the missing aq_nic_deinit() in atl_resume_common(); * rename 'tcs' field to 'tcs_max' in aq_hw_caps_s to differentiate it from the 'tcs' field in aq_nic_cfg_s, which is used for the current number of TCs; * update _TC_MAX defines to the actual number of supported TCs; * move tx_tc_mode register defines slightly higher (just to keep the order of definitions); * separate variables for TX/RX buff_size in hw_atl*_hw_qos_set(); * use AQ_HW_*_TC instead of hardcoded magic numbers; * actually use the 'ret' value in aq_mdo_add_secy(); Signed-off-by:
Dmitry Bezrukov <dbezrukov@marvell.com> Co-developed-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Mark Starovoytov <mstarovoitov@marvell.com> Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queueDavid S. Miller authored
Jeff Kirsher says: ==================== 100GbE Intel Wired LAN Driver Updates 2020-05-21 This series contains updates to ice driver only. Several of the changes are fixes, which could be backported to stable, of which, only one was marked for stable because of the memory leak potential. Jake exposes the information in the flash memory used for link management, which is called the netlist module. Henry and Tony add support for tunnel offloads. Brett adds promiscuous support in VF's which is based on VF trust and the new vf-true-promisc flag. Avinash fixes an issue where a transmit timeout for a queue that belongs to a PFC enabled TC is not a true transmit timeout, but because the PFC is in action. Dave fixes the check for contiguous TCs to allow for various UP2TC mapping configurations. Also fixed an issue when changing the pause parameters would could multiple link drop/down's in succession, which in turn caused the firmware to not generate a link interrupt for the driver to respond to. Anirudh (Ani) fixed a potential race condition in probe/open due to a bit being cleared too early. Lihong updates an error message to make it more meaningful instead of just printing out the numerical value of the status/error code. Also fixed an incorrect return value if deleting a filter does not find a match to delete or when adding a filter that already exists. Karol fixes casting issues and precision loss in the driver. Jesse make the sign usage more consistent in the driver by making sure all instances of vf_id are unsigned, since it can never be negative. Eric fixes a potential memory leak in ice_add_prof_id_vsig() where was not cleaning up resources properly when an error occurs. Michal to help organize the filtering code in the driver, refactor the code into a separate file and add functions to prepare the filter information. Bruce cleaned up a conditional statement that always resulted in true and provided a comment to make it more obvious. Also cleaned up redundant code checks. Tony helps with potential namespace issues by renaming a 'ice' specific function with the driver name prepended. ==================== Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Roopa Prabhu says: ==================== Support for fdb ECMP nexthop groups This series introduces ecmp nexthops and nexthop groups for mac fdb entries. In subsequent patches this is used by the vxlan driver fdb entries. The use case is E-VPN multihoming [1,2,3] which requires bridged vxlan traffic to be load balanced to remote switches (vteps) belonging to the same multi-homed ethernet segment (This is analogous to a multi-homed LAG but over vxlan). Changes include new nexthop flag NHA_FDB for nexthops referenced by fdb entries. These nexthops only have ip. The patches make sure that routes dont reference such nexthops. example: $ip nexthop add id 12 via 172.16.1.2 fdb $ip nexthop add id 13 via 172.16.1.3 fdb $ip nexthop add id 102 group 12/13 fdb $bridge fdb add 02:02:00:00:00:13 dev vxlan1000 nhid 101 self [1] E-VPN https://tools.ietf.org/html/rfc7432 [2] E-VPN VxLAN: https://tools.ietf.org/html/rfc8365 [3] LPC talk with mention of nexthop groups for L2 ecmp http://vger.kernel.org/lpc_net2018_talks/scaling_bridge_fdb_database_slidesV3.pdf v4 - - fix error path free_skb in vxlan_xmit_nh - fix atomic notifier initialization issue (Reported-by: kernel test robot <rong.a.chen@intel.com>) The reported error was easy to locate and fix, but i was not able to re-test with the robot reproducer script due to some other issues with running the script on my test system. v3 - fix wording in selftest print as pointed out by davidA v2 - - dropped nikolays fixes for nexthop multipath null pointer deref (he will send those separately) - added negative tests for route add with fdb nexthop + a few more - Fixes for a few fdb replace conditions found during more testing - Moved to rcu_dereference_rtnl in vxlan_fdb_info and consolidate rcu dereferences - Fixes to build failures Reported-by: kbuild test robot <lkp@intel.com> - DavidA, I am going to send a separate patch for the neighbor code validation for NDA_NH_ID if thats ok. ==================== Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Roopa Prabhu authored
This commit adds ipv4 and ipv6 fdb nexthop api tests to fib_nexthops.sh. Signed-off-by:
Roopa Prabhu <roopa@cumulusnetworks.com> Reviewed-by:
David Ahern <dsahern@gmail.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Roopa Prabhu authored
vxlan driver registers for nexthop add/del notifiers to cleanup fdb entries pointing to such nexthops. Signed-off-by:
Roopa Prabhu <roopa@cumulusnetworks.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Roopa Prabhu authored
This patch adds nexthop add/del notifiers. To be used by vxlan driver in a later patch. Could possibly be used by switchdev drivers in the future. Signed-off-by:
Roopa Prabhu <roopa@cumulusnetworks.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Roopa Prabhu authored
Todays vxlan mac fdb entries can point to multiple remote ips (rdsts) with the sole purpose of replicating broadcast-multicast and unknown unicast packets to those remote ips. E-VPN multihoming [1,2,3] requires bridged vxlan traffic to be load balanced to remote switches (vteps) belonging to the same multi-homed ethernet segment (E-VPN multihoming is analogous to multi-homed LAG implementations, but with the inter-switch peerlink replaced with a vxlan tunnel). In other words it needs support for mac ecmp. Furthermore, for faster convergence, E-VPN multihoming needs the ability to update fdb ecmp nexthops independent of the fdb entries. New route nexthop API is perfect for this usecase. This patch extends the vxlan fdb code to take a nexthop id pointing to an ecmp nexthop group. Changes include: - New NDA_NH_ID attribute for fdbs - Use the newly added fdb nexthop groups - makes vxlan rdsts and nexthop handling code mutually exclusive - since this is a new use-case and the requirement is for ecmp nexthop groups, the fdb add and update path checks that the nexthop is really an ecmp nexthop group. This check can be relaxed in the future, if we want to introduce replication fdb nexthop groups and allow its use in lieu of current rdst lists. - fdb update requests with nexthop id's only allowed for existing fdb's that have nexthop id's - learning will not override an existing fdb entry with nexthop group - I have wrapped the switchdev offload code around the presence of rdst [1] E-VPN RFC https://tools.ietf.org/html/rfc7432 [2] E-VPN with vxlan https://tools.ietf.org/html/rfc8365 [3] http://vger.kernel.org/lpc_net2018_talks/scaling_bridge_fdb_database_slidesV3.pdf Includes a null check fix in vxlan_xmit from Nikolay v2 - Fixed build issue: Reported-by:
kbuild test robot <lkp@intel.com> Signed-off-by:
Roopa Prabhu <roopa@cumulusnetworks.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Roopa Prabhu authored
This patch introduces ecmp nexthops and nexthop groups for mac fdb entries. In subsequent patches this is used by the vxlan driver fdb entries. The use case is E-VPN multihoming [1,2,3] which requires bridged vxlan traffic to be load balanced to remote switches (vteps) belonging to the same multi-homed ethernet segment (This is analogous to a multi-homed LAG but over vxlan). Changes include new nexthop flag NHA_FDB for nexthops referenced by fdb entries. These nexthops only have ip. This patch includes appropriate checks to avoid routes referencing such nexthops. example: $ip nexthop add id 12 via 172.16.1.2 fdb $ip nexthop add id 13 via 172.16.1.3 fdb $ip nexthop add id 102 group 12/13 fdb $bridge fdb add 02:02:00:00:00:13 dev vxlan1000 nhid 101 self [1] E-VPN https://tools.ietf.org/html/rfc7432 [2] E-VPN VxLAN: https://tools.ietf.org/html/rfc8365 [3] LPC talk with mention of nexthop groups for L2 ecmp http://vger.kernel.org/lpc_net2018_talks/scaling_bridge_fdb_database_slidesV3.pdf v4 - fixed uninitialized variable reported by kernel test robot Reported-by:
kernel test robot <rong.a.chen@intel.com> Signed-off-by:
Roopa Prabhu <roopa@cumulusnetworks.com> Reviewed-by:
David Ahern <dsahern@gmail.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queueDavid S. Miller authored
Jeff Kirsher says: ==================== 1GbE Intel Wired LAN Driver Updates 2020-05-21 This series contains updates to igc and e1000. Andre cleans up code that was left over from the igb driver that handled MAC address filters based on the source address, which is not currently supported. Simplifies the MAC address filtering code and prepare the igc driver for future source address support. Updated the MAC address filter internal APIs to support filters based on source address. Added support for Network Flow Classification (NFC) rules based on source MAC address. Cleaned up the 'cookie' field which is not used anywhere in the code and cleaned up a wrapper function that was not needed. Simplified the filtering code for readability and aligned the ethtool functions, so that function names were consistent. Alex provides a fix for e1000 to resolve a deadlock issue when NAPI is being disabled. Sasha does additional cleanup of the igc driver of dead code that is not used or needed. v2: Fix the function header comment in patch 3 of the series, based on the feedback from Jakub Kicinski. ==================== Reviewed-by:
Jakub Kicinski <kuba@kernel.org> Signed-off-by:
David S. Miller <davem@davemloft.net>
-