- 08 Dec, 2018 1 commit
-
-
David Ahern authored
The existing garbage collection algorithm has a number of problems: 1. The gc algorithm will not evict PERMANENT entries as those entries are managed by userspace, yet the existing algorithm walks the entire hash table which means it always considers PERMANENT entries when looking for entries to evict. In some use cases (e.g., EVPN) there can be tens of thousands of PERMANENT entries leading to wasted CPU cycles when gc kicks in. As an example, with 32k permanent entries, neigh_alloc has been observed taking more than 4 msec per invocation. 2. Currently, when the number of neighbor entries hits gc_thresh2 and the last flush for the table was more than 5 seconds ago gc kicks in walks the entire hash table evicting *all* entries not in PERMANENT or REACHABLE state and not marked as externally learned. There is no discriminator on when the neigh entry was created or if it just moved from REACHABLE to another NUD_VALID state (e.g., NUD_STALE). It is possible for entries to be created or for established neighbor entries to be moved to STALE (e.g., an external node sends an ARP request) right before the 5 second window lapses: -----|---------x|----------|----- t-5 t t+5 If that happens those entries are evicted during gc causing unnecessary thrashing on neighbor entries and userspace caches trying to track them. Further, this contradicts the description of gc_thresh2 which says "Entries older than 5 seconds will be cleared". One workaround is to make gc_thresh2 == gc_thresh3 but that negates the whole point of having separate thresholds. 3. Clearing *all* neigh non-PERMANENT/REACHABLE/externally learned entries when gc_thresh2 is exceeded is over kill and contributes to trashing especially during startup. This patch addresses these problems as follows: 1. Use of a separate list_head to track entries that can be garbage collected along with a separate counter. PERMANENT entries are not added to this list. The gc_thresh parameters are only compared to the new counter, not the total entries in the table. The forced_gc function is updated to only walk this new gc_list looking for entries to evict. 2. Entries are added to the list head at the tail and removed from the front. 3. Entries are only evicted if they were last updated more than 5 seconds ago, adhering to the original intent of gc_thresh2. 4. Forced gc is stopped once the number of gc_entries drops below gc_thresh2. 5. Since gc checks do not apply to PERMANENT entries, gc levels are skipped when allocating a new neighbor for a PERMANENT entry. By extension this means there are no explicit limits on the number of PERMANENT entries that can be created, but this is no different than FIB entries or FDB entries. Signed-off-by:David Ahern <dsahern@gmail.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
- 07 Dec, 2018 33 commits
-
-
David S. Miller authored
Salil Mehta says: ==================== net: hns3: Additions/optimizations related to HNS3 H/W err handling This patch set primarily does following addtions and optimizations related to error handling in HNS3 Ethernet driver: 1. Name changes for enable and process functions and minor loop optimizations. [PATCH 1-6] 2. Modify query and clearing of RAS errors using new set of commands because modules specific commands for clearing RCB PPP PF, SSU are obselete. [PATCH 7] 3. Deletes logging 1-bit errors for RAS in HNS3 driver as these never get reported to the driver. [PATCH 8] 4. Add handling of NIC hw errors reported through MSIx rather than PCIe AER channel. [PATCH 9] 5. Add handling for the HW RAS and MSIx errors in the modules MAC, PPP PF, MSIx SRAM, RCB and SSU. [PATCH 10-13] 6. Add handling of RoCEE RAS errors. [PATCH 14] ==================== Signed-off-by: -
Shiju Jose authored
This patch handles the RDMA RAS errors. 1. Enable RAS interrupt, print error detail info and clear error status. 2. Do CORE reset to recovery when these non-fatal errors happened. Signed-off-by:
Xiaofei Tan <tanxiaofei@huawei.com> Signed-off-by:
Shiju Jose <shiju.jose@huawei.com> Signed-off-by:
Salil Mehta <salil.mehta@huawei.com> Signed-off-by:
-
Shiju Jose authored
This patch enables and handles hw errors of the Storage Switch Unit(SSU). Signed-off-by:
-
Shiju Jose authored
This patch enables and handles hw RAS and MSIx errors of PPU(RCB). Signed-off-by:
-
Shiju Jose authored
This patch handles PF hw errors of PPP(Programmable Packet Processor). Signed-off-by:
-
Shiju Jose authored
This patch adds enable and handling of hw errors of the MAC block. Signed-off-by:
-
Salil Mehta authored
This patch adds handling for HNS3 hardware errors(non-standard) which are reported through MSIX interrupts and not through PCIe AER channel. These MSIX reported hardware errors are handled using common misc. interrupt handler. Hardware error related registers cannot be cleared in context to the interrupt received as they require *heavy* access to hardware using IMP(Integrated Mangement Processor) commands. Hence, we defer the clearing of such error events till later time. Since, we have defered exact identification of errors we will have to defer the level of receovery/reset which might be required. Hence, a new reset type UNKNOWN reset has been introduced which effectively defers the assertion of the reset till we get hold of kind of errors at later time. Signed-off-by:
-
Shiju Jose authored
This patch deletes logging 1 bit errors for the following reasons. 1. AER does not notify 1 bit errors to the device drivers. However AER reports 1 bit errors to the userspace through the trace_aer_event for logging in the rasdaemon. 2. Firmware clears the status of 1 bit errors in the hw registers. Signed-off-by:
-
Shiju Jose authored
1. This patch adds handling of hw ras errors using new set of common commands. 2. Updated the error message tables to match the register's name and error status returned by the commands. Signed-off-by:
-
Shiju Jose authored
1. This patch adds minor loop optimization in the hclge_hw_error_set_state function. 2. Adds logging module's name if it fails to configure the error interrupts. Signed-off-by:
-
Shiju Jose authored
This patch renames process_hw_error function to handle_hw_ras_error function to match the purpose of the function. This is because hw errors reported through ras and msix interrupts will be handled separately. Signed-off-by:
-
Shiju Jose authored
This patch deletes unnecessary setting of the descriptor data to 0 for disabling error interrupts because it is already done by the hclge_cmd_setup_basic_desc function. Signed-off-by:
-
Shiju Jose authored
This patch adds calling hclge_hw_error_set_state function to re-enable the error interrupts those will be disabled on the hw reset. Signed-off-by:
-
Shiju Jose authored
This patch - renames the enable error interrupt functions. The reason is that these functions are used for both enable and disable error interrupts. - removes redundant logs from the enable error interrupt functions. Signed-off-by:
-
Shiju Jose authored
1.The command interface for queryng and clearing hw errors is changed, which requires the new process error functions to be added. This patch removes all the current process error functions and associated definitions. The new functions to handle ras errors would be added in this patch set. 2. Fixed order issue of the hw_blk table. Signed-off-by:
-
David S. Miller authored
Ido Schimmel says: ==================== mlxsw: Un/offload FDB on NVE detach/attach Petr says: When a VXLAN device is attached to a bridge of a driver capable of offloading such, or upped, the FDB entries already present at the device need to be offloaded. Similarly when an offloaded VXLAN device ceases being interesting (it is downed, or detached, or a front-panel port netdevice is detached from the bridge that the VXLAN device is attached to), any offloaded FDB entries need to be unoffloaded and unmarked. This attach / detach processing is implemented in this patchset. In patch #1, a code pattern is extracted into a named function for easier reuse. In patch #2, vxlan_fdb_replay() is added to send SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE for each FDB entry with a given VNI. The intention is that the offloading driver will interpret these events like any other and thus offload the FDB entries that existed prior to VXLAN attach. In patches #3 and #4, the functions vxlan_fdb_clear_offload() resp. br_fdb_clear_offload() are added. These clear the offloaded flag at matching FDB entries. In patches #5-#9, we introduce FID-type-specific and NVE-type-specific ops necessary to properly abstract invocations of the replay/clear functions. Finally patch #10 implements the FDB management. In patch #11, the mlxsw-specific test case is extended to check that the management of offload marks under the newly-supported situations is correct. Patch #12, from Ido, exercises the new code paths in actual functional test. v2: - Patch #1: - Modify vxlan_fdb_switchdev_notifier_info() to initialize the structure through a passed-in pointer argument, instead of returning it as a value. - Patch #2: - Adapt to API change in vxlan_fdb_switchdev_notifier_info() ==================== Signed-off-by: -
Ido Schimmel authored
When using VLAN-aware bridges with VXLAN, the VLAN that is mapped to the VNI of the VXLAN device is that which is configured as "pvid untagged" on the corresponding bridge port. When these flags are toggled or when the VLAN is deleted entirely, remote hosts should not be able to receive packets from the VTEP. Add a test case for above mentioned scenarios. Signed-off-by:
Petr Machata <petrm@mellanox.com> Signed-off-by:
Ido Schimmel <idosch@mellanox.com> Signed-off-by:
-
Petr Machata authored
When a VXLAN device is attached to an offloaded bridge, or when a front-panel port is attached to a bridge that already has a VXLAN device, mlxsw should offload the existing offloadable FDB entries. Similarly when VXLAN device is downed, the FDB entries are unoffloaded, and the marks thus need to be cleared. Similarly when a front-panel port device is attached to a bridge with a VXLAN device, or when VLAN flags are tweaked on a VXLAN port attached to a VLAN-aware bridge. Test that the replaying / clearing logic works by observing transitions in presence of offload marks under different scenarios. Signed-off-by:
-
Petr Machata authored
Any existing NVE FDB entries need to be offloaded when NVE is enabled for a given FID. Recent patches have added fdb_replay op for this, so just invoke it from mlxsw_sp_nve_fid_enable(). When NVE is disabled on a FID, any existing FDB offloaded marks need to be cleared on NVE device as well as on its bridge master. An op to handle this, fdb_clear_offload, has been added to FID ops and NVE ops in previous patches. Add code to resolve the NVE device, NVE type, and dispatch to both fdb_clear_offload ops. Signed-off-by:
-
Petr Machata authored
If there are any offloaded FDB entries at bridge master of an NVE device at the time that it's un-offloaded, their offloaded marks need to be cleared. How that is done depends on whether the bridge in question is vlan aware. Therefore add a per-FID-type operation. Implement the operation for the 802.1q and 802.1d bridges. Add and publish a function mlxsw_sp_fid_fdb_clear_offload() to dispatch to the new operation according to FID type. Signed-off-by:
-
Petr Machata authored
If there are any offloaded FDB entries at an NVE device at the time that it's un-offloaded, their offloaded marks need to be cleared. How that is done depends on NVE device type, and therefore add a per-NVE-type operation. Implement the operation for the sole NVE device type currently supported by mlxsw, VXLAN. Signed-off-by:
-
Petr Machata authored
A replay of FDB needs to be performed so that the FDB entries existing at the NVE device are offloaded. How the replay is done depends on NVE device type, and therefore add a per-NVE-type operation. Implement the operation for the sole NVE device type currently supported by mlxsw, VXLAN. Signed-off-by:
-
Petr Machata authored
The notifier block will need to be passed to vxlan_fdb_replay() in a follow-up patch. Signed-off-by:
-
Petr Machata authored
A follow-up patch will add support for replay and for clearing of offload marks. These are NVE type-sensitive operations, and to be able to dispatch them properly, a FID needs to know what NVE type is attached to it. Therefore, track the NVE type at struct mlxsw_sp_fid. Extend mlxsw_sp_fid_vni_set() to take it as an argument, and add mlxsw_sp_fid_nve_type(). Signed-off-by:
-
Petr Machata authored
When a driver unoffloads all FDB entries en bloc, it's inefficient to send the switchdev notification one by one. Add a helper that unsets the offload flag on FDB entries on a given bridge port and VLAN. Signed-off-by:
Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Signed-off-by:
-
Petr Machata authored
When a driver unoffloads all FDB entries en bloc, it's inefficient to send the switchdev notification one by one. Add a helper that walks the FDB table, unsetting the offload flag on RDST with a given VNI. Signed-off-by:
-
Petr Machata authored
When a VXLAN device becomes relevant to a driver (such as when it is attached to an offloaded bridge), the driver will generally need to walk the existing FDB entries and offload them. Add a function vxlan_fdb_replay() to call a given notifier block for each FDB entry with a given VNI. Signed-off-by:
-
Petr Machata authored
There are currently two places that need to initialize the notifier info structure, and one more is coming next when vxlan_fdb_replay() is introduced. These three instances have / will have very similar code that is easy to abstract away into a named function. Add such function, vxlan_fdb_switchdev_notifier_info(), and call it from vxlan_fdb_switchdev_call_notifiers() and vxlan_fdb_find_uc(). Signed-off-by:
-
David S. Miller authored
Igor Russkikh says: ==================== net: aquantia: add RSS configuration In this patchset few bugs related to RSS are fixed and RSS table and hash key configuration is added. We also do increase max number of HW rings upto 8. v2: removed extra arg check ==================== Signed-off-by: -
Dmitry Bogdanov authored
Add support of configuration of RSS hash key and RSS indirection table. Signed-off-by:
Dmitry Bogdanov <dmitry.bogdanov@aquantia.com> Signed-off-by:
Igor Russkikh <igor.russkikh@aquantia.com> Signed-off-by:
-
Dmitry Bogdanov authored
Now RSS indirection table is initialized before setting up the number of hw queues, consequently the table may be filled by non existing queues. This patch moves the initialization when the number of hw queues is known. Signed-off-by:
-
Dmitry Bogdanov authored
Increase the upper limit of the hw queues up to 8. This makes RSS better on multiheaded cpus. This is a maximum AQC hardware supports in one traffic class. The actual value is still limited by a number of available cpu cores. Signed-off-by:
-
Dmitry Bogdanov authored
Set RSS indirection table and RSS hash key sizes to their real size. Signed-off-by:
-
- 06 Dec, 2018 6 commits
-
-
David S. Miller authored
Petr Machata says: ==================== Pass extack to NETDEV_PRE_UP Drivers may need to validate configuration of a device that's about to be upped. An example is mlxsw, which needs to check the configuration of a VXLAN device attached to an offloaded bridge. Should the validation fail, there's currently no way to communicate details of the failure to the user, beyond an error number. Therefore this patch set extends the NETDEV_PRE_UP event to include extack, if available. There are three vectors through which NETDEV_PRE_UP invocation can be reached. The two major ones are dev_open() and dev_change_flags(), the last is then __dev_change_flags(). In patch #1, the first access vector, dev_open() is addressed. An extack parameter is added and all users converted to use it. Before addressing the second vector, two preparatory patches propagate extack argument to the proximity of the dev_change_flags() call in VRF and IPVLAN drivers. That happens in patches #2 and #3. Then in patch #4, dev_change_flags() is treated similarly to dev_open(). Likewise in patch #5, __dev_change_flags() is extended. Then in patches #6 and #7, the extack is finally propagated all the way to the point where the notification is emitted. This change allows particularly mlxsw (which already has code to leverage extack if available) to communicate to the user error messages regarding VXLAN configuration. In patch #8, add a test case that exercises this code and checks that an error message is propagated. For example: local 192.0.2.17 remote 192.0.2.18 \ dstport 4789 nolearning noudpcsum tos inherit ttl 100 local 192.0.2.17 remote 192.0.2.18 \ dstport 4789 nolearning noudpcsum tos inherit ttl 100 Error: mlxsw_spectrum: Conflicting NVE tunnels configuration. v2: - Add David Ahern's tags. ==================== Signed-off-by: -
Petr Machata authored
Add a testsuite dedicated to testing extack propagation and related functionality. Signed-off-by:
Jiri Pirko <jiri@mellanox.com> Reviewed-by:
-
Petr Machata authored
Drivers may need to validate configuration of a device that's about to be upped. Should the validation fail, there's currently no way to communicate details of the failure to the user, beyond an error number. To mend that, change __dev_open() to take an extack argument and pass it from __dev_change_flags() and dev_open(), where it was propagated in the previous patches. Change __dev_open() to call call_netdevice_notifiers_extack() so that the passed-in extack is attached to the NETDEV_PRE_UP notifier. Signed-off-by:
-
Petr Machata authored
In order to propagate extack through NETDEV_PRE_UP, add a new function call_netdevice_notifiers_extack() that primes the extack field of the notifier info. Convert call_netdevice_notifiers() to a simple wrapper around the new function that passes NULL for extack. Signed-off-by:
-
Petr Machata authored
In order to pass extack together with NETDEV_PRE_UP notifications, it's necessary to route the extack to __dev_open() from diverse (possibly indirect) callers. The last missing API is __dev_change_flags(). Therefore extend __dev_change_flags() with and extra extack argument and update the two existing users. Since the function declaration line is changed anyway, name the struct net_device argument to placate checkpatch. Signed-off-by:
-
Petr Machata authored
In order to pass extack together with NETDEV_PRE_UP notifications, it's necessary to route the extack to __dev_open() from diverse (possibly indirect) callers. One prominent API through which the notification is invoked is dev_change_flags(). Therefore extend dev_change_flags() with and extra extack argument and update all users. Most of the calls end up just encoding NULL, but several sites (VLAN, ipvlan, VRF, rtnetlink) do have extack available. Since the function declaration line is changed anyway, name the other function arguments to placate checkpatch. Signed-off-by:
-
