- 01 Aug, 2024 19 commits
-
-
Anand Khoje authored
In non FLR context, at times CX-5 requests release of ~8 million FW pages. This needs humongous number of cmd mailboxes, which to be released once the pages are reclaimed. Release of humongous number of cmd mailboxes is consuming cpu time running into many seconds. Which with non preemptible kernels is leading to critical process starving on that cpu’s RQ. On top of it, the FW does not use all the mailbox messages as it has a limit of releasing 50K pages at once per MLX5_CMD_OP_MANAGE_PAGES + MLX5_PAGES_TAKE device command. Hence, the allocation of these many mailboxes is extra and adds unnecessary overhead. To alleviate this, this change restricts the total number of pages a worker will try to reclaim to maximum 50K pages in one go. Our tests have shown significant benefit of this change in terms of time consumed by dma_pool_free(). During a test where an event was raised by HCA to release 1.3 Million pages, following observations were made: - Without this change: Number of mailbox messages allocated was around 20K, to accommodate the DMA addresses of 1.3 million pages. The average time spent by dma_pool_free() to free the DMA pool is between 16 usec to 32 usec. value ------------- Distribution ------------- count 256 | 0 512 |@ 287 1024 |@@@ 1332 2048 |@ 656 4096 |@@@@@ 2599 8192 |@@@@@@@@@@ 4755 16384 |@@@@@@@@@@@@@@@ 7545 32768 |@@@@@ 2501 65536 | 0 - With this change: Number of mailbox messages allocated was around 800; this was to accommodate DMA addresses of only 50K pages. The average time spent by dma_pool_free() to free the DMA pool in this case lies between 1 usec to 2 usec. value ------------- Distribution ------------- count 256 | 0 512 |@@@@@@@@@@@@@@@@@@ 346 1024 |@@@@@@@@@@@@@@@@@@@@@@ 435 2048 | 0 4096 | 0 8192 | 1 16384 | 0 Signed-off-by:Anand Khoje <anand.a.khoje@oracle.com> Reviewed-by:
Leon Romanovsky <leonro@nvidia.com> Reviewed-by:
Zhu Yanjun <yanjun.zhu@linux.dev> Acked-by:
Saeed Mahameed <saeedm@nvidia.com> Link: https://patch.msgid.link/20240730073634.114407-1-anand.a.khoje@oracle.comSigned-off-by:
Jakub Kicinski <kuba@kernel.org>
-
Breno Leitao authored
This patch modifies the skb_unref function to skip the early return optimization when CONFIG_DEBUG_NET is enabled. The change ensures that the reference count decrement always occurs in debug builds, allowing for more thorough checking of SKB reference counting. Previously, when the SKB's reference count was 1 and CONFIG_DEBUG_NET was not set, the function would return early after a memory barrier (smp_rmb()) without decrementing the reference count. This optimization assumes it's safe to proceed with freeing the SKB without the overhead of an atomic decrement from 1 to 0. With this change: - In non-debug builds (CONFIG_DEBUG_NET not set), behavior remains unchanged, preserving the performance optimization. - In debug builds (CONFIG_DEBUG_NET set), the reference count is always decremented, even when it's 1, allowing for consistent behavior and potentially catching subtle SKB management bugs. This modification enhances debugging capabilities for networking code without impacting performance in production kernels. It helps kernel developers identify and diagnose issues related to SKB management and reference counting in the network stack. Cc: Chris Mason <clm@fb.com> Suggested-by:
Breno Leitao <leitao@debian.org> Reviewed-by:
Eric Dumazet <edumazet@google.com> Link: https://patch.msgid.link/20240729104741.370327-1-leitao@debian.orgSigned-off-by:
Paolo Abeni <pabeni@redhat.com>
-
Jakub Kicinski authored
Allen Pais says: ==================== ethernet: Convert from tasklet to BH workqueue [part] The only generic interface to execute asynchronously in the BH context is tasklet; however, it's marked deprecated and has some design flaws. To replace tasklets, BH workqueue support was recently added. A BH workqueue behaves similarly to regular workqueues except that the queued work items are executed in the BH context. This patch converts a few drivers in drivers/ethernet/* from tasklet to BH workqueue. The next set will be sent out after the next -rc is out. v2: https://lore.kernel.org/20240621183947.4105278-1-allen.lkml@gmail.com v1: https://lore.kernel.org/20240507190111.16710-2-apais@linux.microsoft.com ==================== Link: https://patch.msgid.link/20240730183403.4176544-1-allen.lkml@gmail.comSigned-off-by:
-
Allen Pais authored
Migrate tasklet APIs to the new bottom half workqueue mechanism. It replaces all occurrences of tasklet usage with the appropriate workqueue APIs throughout the macb driver. This transition ensures compatibility with the latest design and enhances performance. Signed-off-by:
Allen Pais <allen.lkml@gmail.com> Link: https://patch.msgid.link/20240730183403.4176544-5-allen.lkml@gmail.comSigned-off-by:
-
Allen Pais authored
Migrate tasklet APIs to the new bottom half workqueue mechanism. It replaces all occurrences of tasklet usage with the appropriate workqueue APIs throughout the cnic driver. This transition ensures compatibility with the latest design and enhances performance. Signed-off-by:
-
Allen Pais authored
Migrate tasklet APIs to the new bottom half workqueue mechanism. It replaces all occurrences of tasklet usage with the appropriate workqueue APIs throughout the xgbe driver. This transition ensures compatibility with the latest design and enhances performance. Signed-off-by:
-
Allen Pais authored
Migrate tasklet APIs to the new bottom half workqueue mechanism. It replaces all occurrences of tasklet usage with the appropriate workqueue APIs throughout the alteon driver. This transition ensures compatibility with the latest design and enhances performance. Signed-off-by:
-
Jakub Kicinski authored
Petr Machata says: ==================== mlxsw: core_thermal: Small cleanups Ido Schimmel says: Clean up various issues which I noticed while addressing feedback on a different patchset. ==================== Link: https://patch.msgid.link/cover.1722345311.git.petrm@nvidia.comSigned-off-by:
-
Ido Schimmel authored
The name of a thermal zone device cannot be longer than 19 characters ('THERMAL_NAME_LENGTH - 1'). The format string 'mlxsw-lc%d-module%d' can exceed this limitation if the maximum number of line cards cannot be represented using a single digit and the maximum number of transceiver modules cannot be represented using two digits. This is not the case with current systems nor future ones. Therefore, increase the size of the result buffer beyond 'THERMAL_NAME_LENGTH' and suppress the following build warning [1]. If this limitation is ever exceeded, we will know about it since the thermal core validates the thermal device's name during registration. [1] drivers/net/ethernet/mellanox/mlxsw/core_thermal.c: In function ‘mlxsw_thermal_modules_init.part.0’: drivers/net/ethernet/mellanox/mlxsw/core_thermal.c:418:70: error: ‘%d’ directive output may be truncated writing between 1 and 3 bytes into a region of size between 2 and 4 [-Werror=format-truncation=] 418 | snprintf(tz_name, sizeof(tz_name), "mlxsw-lc%d-module%d", | ^~ In function ‘mlxsw_thermal_module_tz_init’, inlined from ‘mlxsw_thermal_module_init’ at drivers/net/ethernet/mellanox/mlxsw/core_thermal.c:465:9, inlined from ‘mlxsw_thermal_modules_init.part.0’ at drivers/net/ethernet/mellanox/mlxsw/core_thermal.c:500:9: drivers/net/ethernet/mellanox/mlxsw/core_thermal.c:418:52: note: directive argument in the range [1, 256] 418 | snprintf(tz_name, sizeof(tz_name), "mlxsw-lc%d-module%d", | ^~~~~~~~~~~~~~~~~~~~~ drivers/net/ethernet/mellanox/mlxsw/core_thermal.c:418:17: note: ‘snprintf’ output between 18 and 22 bytes into a destination of size 20 418 | snprintf(tz_name, sizeof(tz_name), "mlxsw-lc%d-module%d", | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 419 | module_tz->slot_index, module_tz->module + 1); | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Signed-off-by:Ido Schimmel <idosch@nvidia.com> Reviewed-by:
Vadim Pasternak <vadimp@nvidia.com> Signed-off-by:
Petr Machata <petrm@nvidia.com> Reviewed-by:
Wojciech Drewek <wojciech.drewek@intel.com> Link: https://patch.msgid.link/583a70c6dbe75e6bf0c2c58abbb3470a860d2dc3.1722345311.git.petrm@nvidia.comSigned-off-by:
-
Ido Schimmel authored
Setting both pointers to NULL is unnecessary since the code never checks whether these pointers are NULL or not. Remove the assignments. Signed-off-by:
-
Ido Schimmel authored
mlxsw_thermal_module_fini() cannot be invoked with a thermal module which is NULL or which is not associated with a thermal zone, so remove these checks. Signed-off-by:
-
Ido Schimmel authored
During rollback, instead of calling mlxsw_thermal_module_fini() for all the modules, only call it for modules that were successfully initialized. This is not a bug fix since mlxsw_thermal_module_fini() first checks that the module was initialized. Signed-off-by:
-
Ido Schimmel authored
mlxsw_thermal_module_fini() de-initializes the module's thermal zone, but mlxsw_thermal_module_init() does not initialize it. Make both functions symmetric by moving the initialization of the module's thermal zone to mlxsw_thermal_module_init(). Signed-off-by:
-
Ido Schimmel authored
'dev' and 'core' arguments are not used by mlxsw_thermal_module_init(). Remove them. Signed-off-by:
-
Ido Schimmel authored
There is no need to traverse the same array twice. Do it once by folding both loops into one. Signed-off-by:
-
Ido Schimmel authored
mlxsw_thermal_modules_init() allocates an array of modules and then initializes each entry by calling mlxsw_thermal_module_init() which among other things initializes the 'parent' pointer of the entry. mlxsw_thermal_modules_init() then traverses over the array again, but skips over entries that do not have their 'parent' pointer set which is impossible given the above. Therefore, remove the unnecessary check. Signed-off-by:
-
Ido Schimmel authored
mlxsw_thermal_modules_init() allocates an array of modules and then calls mlxsw_thermal_module_init() to initialize each entry in the array. It is therefore impossible for mlxsw_thermal_module_init() to encounter an entry that is already initialized and has its 'parent' pointer set. Remove the unnecessary check. Signed-off-by:
-
Ido Schimmel authored
The function returns immediately if the thermal zone pointer is NULL so there is no need to check it before calling the function. Remove the check. Signed-off-by:
-
Krzysztof Olędzki authored
Enable reading additional EEPROM information from high pages such as thresholds and alarms on QSFP/QSFP+/QSFP28 modules. "This is similar to commit a708fb7b ("net/mlx5e: ethtool, Add support for EEPROM high pages query") but given all the required logic already exists in mlx4_qsfp_eeprom_params_set() only s/_LEN/MAX_LEN/ is needed. Tested-by:
Dan Merillat <git@dan.merillat.org> Signed-off-by:
Krzysztof Piotr Oledzki <ole@ans.pl> Reviewed-by:
-
- 31 Jul, 2024 21 commits
-
-
Patrick Rohr authored
draft-ietf-6man-pio-pflag is adding a new flag to the Prefix Information Option to signal that the network can allocate a unique IPv6 prefix per client via DHCPv6-PD (see draft-ietf-v6ops-dhcp-pd-per-device). When ra_honor_pio_pflag is enabled, the presence of a P-flag causes SLAAC autoconfiguration to be disabled for that particular PIO. An automated test has been added in Android (r.android.com/3195335) to go along with this change. Cc: Maciej Żenczykowski <maze@google.com> Cc: Lorenzo Colitti <lorenzo@google.com> Cc: David Lamparter <equinox@opensourcerouting.org> Cc: Simon Horman <horms@kernel.org> Signed-off-by:
Patrick Rohr <prohr@google.com> Reviewed-by:
Maciej Żenczykowski <maze@google.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Zhengchao Shao says: ==================== net/smc: do some cleanups in smc module Do some cleanups in smc module. ==================== Signed-off-by: -
Zhengchao Shao authored
The input parameter "is_rmb" of the smcr_new_buf_create function has never been used, remove it. Signed-off-by:
Zhengchao Shao <shaozhengchao@huawei.com> Reviewed-by:
Simon Horman <horms@kernel.org> Signed-off-by:
-
Zhengchao Shao authored
When the SMC client perform CLC handshake, it will check whether the clc header type is correct in receiving SMC_CLC_ACCEPT packet. The specific invoking path is as follows: __smc_connect smc_connect_clc smc_clc_wait_msg smc_clc_msg_hdr_valid smc_clc_msg_acc_conf_valid Therefore, the smc_connect_check_aclc interface invoked by __smc_connect does not need to check type again. Signed-off-by: -
Zhengchao Shao authored
When the SMC client begins to connect to server, smcd_version is set to SMC_V1 + SMC_V2. If fail to get VLAN ID, only SMC_V2 information is left in smcd_version. And smcd_version will not be changed to 0. Therefore, remove the fallback caused by the failure to get VLAN ID. Signed-off-by:
-
Zhengchao Shao authored
Because linux/err.h is unreferenced in smc_loopback.h file, so remove it. Signed-off-by:
D. Wythe <alibuda@linux.alibaba.com> Signed-off-by:
-
Pawel Dembicki authored
Add a schema validator to vitesse,vsc73xx.yaml for MAC-level RGMII delays in the CPU port. Additionally, valid values for VSC73XX were defined, and a common definition for the RX and TX valid range was created. Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
Pawel Dembicki <paweldembicki@gmail.com> Reviewed-by:
Rob Herring (Arm) <robh@kernel.org> Signed-off-by:
-
Pawel Dembicki authored
This patch switches hardcoded RGMII transmit/receive delay to a configurable value. Delay values are taken from the properties of the CPU port: 'tx-internal-delay-ps' and 'rx-internal-delay-ps'. The default value is configured to 2.0 ns to maintain backward compatibility with existing code. Signed-off-by:
-
David S. Miller authored
James Chapman says: ==================== l2tp: simplify tunnel and session cleanup This series simplifies and improves l2tp tunnel and session cleanup. * refactor l2tp management code to not use the tunnel socket's sk_user_data. This allows the tunnel and its socket to be closed and freed without sequencing the two using the socket's sk_destruct hook. * export ip_flush_pending_frames and use it when closing l2tp_ip sockets. * move the work of closing all sessions in the tunnel to the work queue so that sessions are deleted using the same codepath whether they are closed by user API request or their parent tunnel is closing. * refactor l2tp_ppp pppox socket / session relationship to have the session keep the socket alive, not the other way around. Previously the pppox socket held a ref on the session, which complicated session delete by having to go through the pppox socket destructor. * free sessions and pppox sockets by rcu. * fix a possible tunnel refcount underflow. * avoid using rcu_barrier in net exit handler. ==================== Signed-off-by: -
James Chapman authored
Move the work of closing all tunnels from the pernet exit hook to pre_exit since the core does rcu synchronisation between these steps and we can therefore remove rcu_barrier from l2tp code. Signed-off-by:
James Chapman <jchapman@katalix.com> Signed-off-by:
Tom Parkin <tparkin@katalix.com> Signed-off-by:
-
James Chapman authored
l2tp eth/ppp pseudowire setup/cleanup uses kfree() in some error paths. Drop the refcount instead such that the session object is always freed when the refcount reaches 0. Signed-off-by:
-
James Chapman authored
l2tp_session_register uses an idr_alloc then idr_replace pattern to insert sessions into the session IDR. To catch invalid locking, add a WARN_ON_ONCE if the IDR entry is modified by another thread between alloc and replace steps. Also add comments to make expectations clear. Signed-off-by:
-
James Chapman authored
l2tp_v3_session_htable and tunnel->session_list are read by lockless getters using RCU. Use rcu list variants when adding or removing list items. Signed-off-by:
-
James Chapman authored
When a session is created, it sets a backpointer to its tunnel. When the session refcount drops to 0, l2tp_session_free drops the tunnel refcount if session->tunnel is non-NULL. However, session->tunnel is set in l2tp_session_create, before the tunnel refcount is incremented by l2tp_session_register, which leaves a small window where session->tunnel is non-NULL when the tunnel refcount hasn't been bumped. Moving the assignment to l2tp_session_register is trivial but l2tp_session_create calls l2tp_session_set_header_len which uses session->tunnel to get the tunnel's encap. Add an encap arg to l2tp_session_set_header_len to avoid using session->tunnel. If l2tpv3 sessions have colliding IDs, it is possible for l2tp_v3_session_get to race with l2tp_session_register and fetch a session which doesn't yet have session->tunnel set. Add a check for this case. Signed-off-by:
-
James Chapman authored
Each l2tp ppp session has an associated pppox socket. l2tp_ppp uses the session's pppox socket refcount to manage session lifetimes; the pppox socket holds a ref on the session which is dropped by the socket destructor. This complicates session cleanup. Given l2tp sessions are refcounted, it makes more sense to reverse this relationship such that the session keeps the socket alive, not the other way around. So refactor l2tp_ppp to have the session hold a ref on its socket while it references it. When the session is closed, it drops its socket ref when it detaches from its socket. If the socket is closed first, it initiates the closing of its session, if one is attached. The socket/session can then be freed asynchronously when their refcounts drop to 0. Use the session's session_close callback to detach the pppox socket since this will be done on the work queue together with the rest of the session cleanup via l2tp_session_delete. Also, since l2tp_ppp uses the pppox socket's sk_user_data, use the rcu sk_user_data access helpers when accessing it and set the socket's SOCK_RCU_FREE flag to have pppox sockets freed by rcu. Signed-off-by:
-
James Chapman authored
l2tp sessions may be accessed under an rcu read lock. Have them freed via rcu and remove the now unneeded synchronize_rcu when a session is removed. Signed-off-by:
-
James Chapman authored
When a tunnel is closed, l2tp_tunnel_closeall closes all sessions in the tunnel. Move the work of deleting each session to the work queue so that sessions are deleted using the same codepath whether they are closed by user API request or their parent tunnel is closing. This also avoids the locking dance in l2tp_tunnel_closeall where the tunnel's session list lock was unlocked and relocked in the loop. In l2tp_exit_net, use drain_workqueue instead of flush_workqueue because the processing of tunnel_delete work may queue session_delete work items which must also be processed. Signed-off-by:
-
James Chapman authored
When the l2tp tunnel socket used sk_user_data to point to its associated l2tp tunnel, socket and tunnel cleanup had to make use of the socket's destructor to free the tunnel only when the socket could no longer be accessed. Now that sk_user_data is no longer used, we can simplify socket and tunnel cleanup: * If the tunnel closes first, it cleans up and drops its socket ref when the tunnel refcount drops to zero. If its socket was provided by userspace, the socket is closed and freed asynchronously, when userspace closes it. If its socket is a kernel socket, the tunnel closes the socket itself during cleanup and drops its socket ref when the tunnel's refcount drops to zero. * If the socket closes first, we initiate the closing of its associated tunnel. For UDP sockets, this is via the socket's encap_destroy hook. For L2TPIP sockets, this is via the socket's destroy callback. The tunnel holds a socket ref while it references the sock. When the tunnel is freed, it drops its socket ref and the socket will be cleaned up when its own refcount drops to zero, asynchronous to the tunnel free. * The tunnel socket destructor is no longer needed since the tunnel is no longer freed through the socket destructor. Signed-off-by: -
James Chapman authored
Since l2tp no longer derives tunnel pointers directly via sk_user_data, it is no longer useful for l2tp to check tunnel pointers using a magic feather. Drop the tunnel's magic field. Signed-off-by:
-
James Chapman authored
l2tp no longer uses the tunnel socket's sk_user_data so drop the code which sets it. In l2tp_validate_socket use l2tp_sk_to_tunnel to check whether a given socket is already attached to an l2tp tunnel since we can no longer use non-null sk_user_data to indicate this. Signed-off-by:
-
James Chapman authored
l2tp's ppp procfs output can be used to show internal state of pppol2tp. It includes a 'user-data-ok' field, which is derived from the tunnel socket's sk_user_data being non-NULL. Use tunnel->sock being non-NULL to indicate this instead. Signed-off-by:
-
