- 21 Nov, 2023 5 commits
-
-
Radhey Shyam Pandey authored
Add dmaengine framework to communicate with the xilinx DMAengine driver(AXIDMA). Axi ethernet driver uses separate channels for transmit and receive. Add support for these channels to handle TX and RX with skb and appropriate callbacks. Also add axi ethernet core interrupt for dmaengine framework support. The dmaengine framework was extended for metadata API support. However it still needs further enhancements to make it well suited for ethernet usecases. The ethernet features i.e ethtool set/get of DMA IP properties, ndo_poll_controller,(mentioned in TODO) are not supported and it requires follow-up discussions. dmaengine support has a dependency on xilinx_dma as it uses xilinx_vdma_channel_set_config() API to reset the DMA IP which internally reset MAC prior to accessing MDIO. Benchmark with netperf: xilinx-zcu102-20232:~$ netperf -H 192.168.10.20 -t TCP_STREAM MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.10.20 () port 0 AF_INET Recv Send Send Socket Socket Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec 131072 16384 16384 10.02 886.69 xilinx-zcu102-20232:~$ netperf -H 192.168.10.20 -t UDP_STREAM MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.10.20 () port 0 AF_INET Socket Message Elapsed Messages Size Size Time Okay Errors Throughput bytes bytes secs # # 10^6bits/sec 212992 65507 10.00 15851 0 830.66 212992 10.00 15851 830.66 Signed-off-by:
Radhey Shyam Pandey <radhey.shyam.pandey@amd.com> Link: https://lore.kernel.org/r/1700074613-1977070-4-git-send-email-radhey.shyam.pandey@amd.comSigned-off-by:
Jakub Kicinski <kuba@kernel.org>
-
Sarath Babu Naidu Gaddam authored
The axiethernet driver has inbuilt dma programming. In order to add dmaengine support and make it's integration seamless the current axidma inbuilt programming code is put under use_dmaengine check. It also performs minor code reordering to minimize conditional use_dmaengine checks and there is no functional change. It uses "dmas" property to identify whether it should use a dmaengine framework or inbuilt axidma programming. Signed-off-by:
Sarath Babu Naidu Gaddam <sarath.babu.naidu.gaddam@amd.com> Signed-off-by:
-
Radhey Shyam Pandey authored
Xilinx 1G/2.5G Ethernet Subsystem provides 32-bit AXI4-Stream buses to move transmit and receive Ethernet data to and from the subsystem. These buses are designed to be used with an AXI Direct Memory Access(DMA) IP or AXI Multichannel Direct Memory Access (MCDMA) IP core, AXI4-Stream Data FIFO, or any other custom logic in any supported device. Primary high-speed DMA data movement between system memory and stream target is through the AXI4 Read Master to AXI4 memory-mapped to stream (MM2S) Master, and AXI stream to memory-mapped (S2MM) Slave to AXI4 Write Master. AXI DMA/MCDMA enables channel of data movement on both MM2S and S2MM paths in scatter/gather mode. AXI DMA has two channels where as MCDMA has 16 Tx and 16 Rx channels. To uniquely identify each channel use 'chan' suffix. Depending on the usecase AXI ethernet driver can request any combination of multichannel DMA channels using generic dmas, dma-names properties. Example: dma-names = tx_chan0, rx_chan0, tx_chan1, rx_chan1; Signed-off-by:
Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> Link: https://lore.kernel.org/r/1700074613-1977070-2-git-send-email-radhey.shyam.pandey@amd.comSigned-off-by:
-
Willem de Bruijn authored
Commit 29f834aa ("net_sched: sch_fq: add 3 bands and WRR scheduling") introduces multiple traffic bands, and per-band maximum packet count. Per-band limits ensures that packets in one class cannot fill the entire qdisc and so cause DoS to the traffic in the other classes. Verify this behavior: 1. set the limit to 10 per band 2. send 20 pkts on band A: verify that 10 are queued, 10 dropped 3. send 20 pkts on band A: verify that 0 are queued, 20 dropped 4. send 20 pkts on band B: verify that 10 are queued, 10 dropped Packets must remain queued for a period to trigger this behavior. Use SO_TXTIME to store packets for 100 msec. The test reuses existing upstream test infra. The script is a fork of cmsg_time.sh. The scripts call cmsg_sender. The test extends cmsg_sender with two arguments: * '-P' SO_PRIORITY There is a subtle difference between IPv4 and IPv6 stack behavior: PF_INET/IP_TOS sets IP header bits and sk_priority PF_INET6/IPV6_TCLASS sets IP header bits BUT NOT sk_priority * '-n' num pkts Send multiple packets in quick succession. I first attempted a for loop in the script, but this is too slow in virtualized environments, causing flakiness as the 100ms timeout is reached and packets are dequeued. Also do not wait for timestamps to be queued unless timestamps are requested. Signed-off-by:
Willem de Bruijn <willemb@google.com> Reviewed-by:
Simon Horman <horms@kernel.org> Reviewed-by:
Eric Dumazet <edumazet@google.com> Link: https://lore.kernel.org/r/20231116203449.2627525-1-willemdebruijn.kernel@gmail.comSigned-off-by:
-
Vishvambar Panth S authored
The LAN743x/PCI11xxx DMA descriptors are always 4 dwords long, but the device supports placing the descriptors in memory back to back or reserving space in between them using its DMA_DESCRIPTOR_SPACE (DSPACE) configurable hardware setting. Currently DSPACE is unnecessarily set to match the host's L1 cache line size, resulting in space reserved in between descriptors in most platforms and causing a suboptimal behavior (single PCIe Mem transaction per descriptor). By changing the setting to DSPACE=16 many descriptors can be packed in a single PCIe Mem transaction resulting in a massive performance improvement in bidirectional tests without any negative effects. Tested and verified improvements on x64 PC and several ARM platforms (typical data below) Test setup 1: x64 PC with LAN7430 ---> x64 PC iperf3 UDP bidirectional with DSPACE set to L1 CACHE Size: - - - - - - - - - - - - - - - - - - - - - - - - - [ ID][Role] Interval Transfer Bitrate [ 5][TX-C] 0.00-10.00 sec 170 MBytes 143 Mbits/sec sender [ 5][TX-C] 0.00-10.04 sec 169 MBytes 141 Mbits/sec receiver [ 7][RX-C] 0.00-10.00 sec 1.02 GBytes 876 Mbits/sec sender [ 7][RX-C] 0.00-10.04 sec 1.02 GBytes 870 Mbits/sec receiver iperf3 UDP bidirectional with DSPACE set to 16 Bytes - - - - - - - - - - - - - - - - - - - - - - - - - [ ID][Role] Interval Transfer Bitrate [ 5][TX-C] 0.00-10.00 sec 1.11 GBytes 956 Mbits/sec sender [ 5][TX-C] 0.00-10.04 sec 1.11 GBytes 951 Mbits/sec receiver [ 7][RX-C] 0.00-10.00 sec 1.10 GBytes 948 Mbits/sec sender [ 7][RX-C] 0.00-10.04 sec 1.10 GBytes 942 Mbits/sec receiver Test setup 2 : RK3399 with LAN7430 ---> x64 PC RK3399 Spec: The SOM-RK3399 is ARM module designed and developed by FriendlyElec. Cores: 64-bit Dual Core Cortex-A72 + Quad Core Cortex-A53 Frequency: Cortex-A72(up to 2.0GHz), Cortex-A53(up to 1.5GHz) PCIe: PCIe x4, compatible with PCIe 2.1, Dual operation mode iperf3 UDP bidirectional with DSPACE set to L1 CACHE Size: - - - - - - - - - - - - - - - - - - - - - - - - - [ ID][Role] Interval Transfer Bitrate [ 5][TX-C] 0.00-10.00 sec 534 MBytes 448 Mbits/sec sender [ 5][TX-C] 0.00-10.05 sec 534 MBytes 446 Mbits/sec receiver [ 7][RX-C] 0.00-10.00 sec 1.12 GBytes 961 Mbits/sec sender [ 7][RX-C] 0.00-10.05 sec 1.11 GBytes 946 Mbits/sec receiver iperf3 UDP bidirectional with DSPACE set to 16 Bytes - - - - - - - - - - - - - - - - - - - - - - - - - [ ID][Role] Interval Transfer Bitrate [ 5][TX-C] 0.00-10.00 sec 966 MBytes 810 Mbits/sec sender [ 5][TX-C] 0.00-10.04 sec 965 MBytes 806 Mbits/sec receiver [ 7][RX-C] 0.00-10.00 sec 1.11 GBytes 956 Mbits/sec sender [ 7][RX-C] 0.00-10.04 sec 1.07 GBytes 919 Mbits/sec receiver Signed-off-by:
Vishvambar Panth S <vishvambarpanth.s@microchip.com> Reviewed-by:
Florian Fainelli <florian.fainelli@broadcom.com> Reviewed-by:
Jacob Keller <jacob.e.keller@intel.com> Link: https://lore.kernel.org/r/20231116054350.620420-1-vishvambarpanth.s@microchip.comSigned-off-by:
-
- 19 Nov, 2023 13 commits
-
-
Lorenzo Bianconi authored
Simplify the code and use __dev_alloc_page() instead of __dev_alloc_pages() with order 0 in mtk_wed_tx_buffer_alloc routine Signed-off-by:
Lorenzo Bianconi <lorenzo@kernel.org> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Roger Quadros says: =================== net: eth: am65-cpsw: add ethtool MAC stats Gets 'ethtool -S eth0 --groups eth-mac' command to work. Also set default TX channels to maximum available and does cleanup in am65_cpsw_nuss_common_open() error path. Changelog: v2: - add __iomem to *stats, to prevent sparse warning - clean up RX descriptors and free up SKB in error handling of am65_cpsw_nuss_common_open() - Re-arrange some funcitons to avoid forward declaration ==================== Signed-off-by: -
Roger Quadros authored
k3_udma_glue_enable_rx/tx_chn returns error code on failure. Bail out on error while enabling TX/RX channel. In the error path, clean up the RX descriptors and SKBs. Get rid of kmemleak_not_leak() as it seems unnecessary now. Fixes: 93a76530 ("net: ethernet: ti: introduce am65x/j721e gigabit eth subsystem driver") Signed-off-by:
Roger Quadros <rogerq@kernel.org> Signed-off-by:
-
Roger Quadros authored
am65-cpsw supports 8 TX hardware queues. Set this as default. The rationale is that some am65-cpsw devices can have up to 4 ethernet ports. If the number of TX channels have to be changed then all interfaces have to be brought down and up as the old default of 1 TX channel is too restrictive for any mqprio/taprio usage. Another reason for this change is to allow testing using kselftest:net/forwarding:ethtool_mm.sh out of the box. Signed-off-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
-
Roger Quadros authored
Re-arrange am65_cpsw_nuss_rx_cleanup(), am65_cpsw_nuss_xmit_free() and am65_cpsw_nuss_tx_cleanup() to avoid forward declaration. No functional change. Signed-off-by:
-
Roger Quadros authored
Gets 'ethtool -S eth0 --groups eth-mac' command to work. Signed-off-by:
-
Li RongQing authored
if a PF has 256 or more VFs, ip link command will allocate an order 3 memory or more, and maybe trigger OOM due to memory fragment, the VFs needed memory size is computed in rtnl_vfinfo_size. so introduce nlmsg_new_large which calls netlink_alloc_large_skb in which vmalloc is used for large memory, to avoid the failure of allocating memory ip invoked oom-killer: gfp_mask=0xc2cc0(GFP_KERNEL|__GFP_NOWARN|\ __GFP_COMP|__GFP_NOMEMALLOC), order=3, oom_score_adj=0 CPU: 74 PID: 204414 Comm: ip Kdump: loaded Tainted: P OE Call Trace: dump_stack+0x57/0x6a dump_header+0x4a/0x210 oom_kill_process+0xe4/0x140 out_of_memory+0x3e8/0x790 __alloc_pages_slowpath.constprop.116+0x953/0xc50 __alloc_pages_nodemask+0x2af/0x310 kmalloc_large_node+0x38/0xf0 __kmalloc_node_track_caller+0x417/0x4d0 __kmalloc_reserve.isra.61+0x2e/0x80 __alloc_skb+0x82/0x1c0 rtnl_getlink+0x24f/0x370 rtnetlink_rcv_msg+0x12c/0x350 netlink_rcv_skb+0x50/0x100 netlink_unicast+0x1b2/0x280 netlink_sendmsg+0x355/0x4a0 sock_sendmsg+0x5b/0x60 ____sys_sendmsg+0x1ea/0x250 ___sys_sendmsg+0x88/0xd0 __sys_sendmsg+0x5e/0xa0 do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7f95a65a5b70 Cc: Yunsheng Lin <linyunsheng@huawei.com> Signed-off-by:Li RongQing <lirongqing@baidu.com> Link: https://lore.kernel.org/r/20231115120108.3711-1-lirongqing@baidu.comSigned-off-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queueJakub Kicinski authored
Tony Nguyen says: ==================== ice: one by one port representors creation Michal Swiatkowski says: Currently ice supports creating port representors only for VFs. For that use case they can be created and removed in one step. This patchset is refactoring current flow to support port representor creation also for subfunctions and SIOV. In this case port representors need to be created and removed one by one. Also, they can be added and removed while other port representors are running. To achieve that we need to change the switchdev configuration flow. Three first patches are only cosmetic (renaming, removing not used code). Next few ones are preparation for new flow. The most important one is "add VF representor one by one". It fully implements new flow. New type of port representor (for subfunction) will be introduced in follow up patchset. * '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue: ice: reserve number of CP queues ice: adjust switchdev rebuild path ice: add VF representors one by one ice: realloc VSI stats arrays ice: set Tx topology every time new repr is added ice: allow changing SWITCHDEV_CTRL VSI queues ice: return pointer to representor ice: make representor code generic ice: remove VF pointer reference in eswitch code ice: track port representors in xarray ice: use repr instead of vf->repr ice: track q_id in representor ice: remove unused control VSI parameter ice: remove redundant max_vsi_num variable ice: rename switchdev to eswitch ==================== Link: https://lore.kernel.org/r/20231114181449.1290117-1-anthony.l.nguyen@intel.comSigned-off-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queueJakub Kicinski authored
Tony Nguyen says: ==================== igc: Add support for physical + free-running timers Vinicius Costa Gomes says: The objective is to allow having functionality that depends on the physical timer (taprio and ETF offloads, for example) and vclocks operating together. The "big" missing piece is the implementation of the .getcyclesx64() function in igc, as i225/i226 have multiple timers, we use one of those timers (timer 1) as a free-running (non adjustable) timer. The complication is that only implementing .getcyclesx64() and nothing else will break synchronization when using vclocks, as reading the clock will retrieve the free-running value but timnestamps will come from the adjustable timer. The solution is to modify "in one go" the timestamping code to be able to retrieve the timestamp from the correct timer (if a socket is "phc_bound" to a vclock the timestamp will come from the free-running timer). I was debating whether or not to do the adjustments for the internal latencies for the free-running timestamps, decided to do the adjustments so the path delay when using vclocks is similar to the one when using the physical clock. One future improvement is to implement the .getcrosscycles() function. * '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue: igc: Add support for PTP .getcyclesx64() igc: Simplify setting flags in the TX data descriptor ==================== Link: https://lore.kernel.org/r/20231114183640.1303163-1-anthony.l.nguyen@intel.comSigned-off-by:
-
Jakub Kicinski authored
Pedro Tammela says: ==================== net/sched: cls_u32: use proper refcounts In u32 we are open coding refcounts of hashtables with integers which is far from ideal. Update those with proper refcount and add a couple of tests to tdc that exercise the refcounts explicitly. ==================== Link: https://lore.kernel.org/r/20231114141856.974326-1-pctammela@mojatatu.comSigned-off-by:
-
Pedro Tammela authored
Add tests to specifically check for the refcount interactions of hashtables created by u32. These tables should not be deleted when referenced and the flush order should respect a tree like composition. Signed-off-by:
Pedro Tammela <pctammela@mojatatu.com> Acked-by:
Jamal Hadi Salim <jhs@mojatatu.com> Link: https://lore.kernel.org/r/20231114141856.974326-3-pctammela@mojatatu.comSigned-off-by:
-
Pedro Tammela authored
Proper refcounts will always warn splat when something goes wrong, be it underflow, saturation or object resurrection. As these are always a source of bugs, use it in cls_u32 as a safeguard to prevent/catch issues. Another benefit is that the refcount API self documents the code, making clear when transitions to dead are expected. For such an update we had to make minor adaptations on u32 to fit the refcount API. First we set explicitly to '1' when objects are created, then the objects are alive until a 1 -> 0 happens, which is then released appropriately. The above made clear some redundant operations in the u32 code around the root_ht handling that were removed. The root_ht is created with a refcnt set to 1. Then when it's associated with tcf_proto it increments the refcnt to 2. Throughout the entire code the root_ht is an exceptional case and can never be referenced, therefore the refcnt never incremented/decremented. Its lifetime is always bound to tcf_proto, meaning if you delete tcf_proto the root_ht is deleted as well. The code made up for the fact that root_ht refcnt is 2 and did a double decrement to free it, which is not a fit for the refcount API. Even though refcount_t is implemented using atomics, we should observe a negligible control plane impact. Signed-off-by:
-
Jakub Kicinski authored
Revert following commits: commit acec05fb ("net_tstamp: Add TIMESTAMPING SOFTWARE and HARDWARE mask") commit 11d55be0 ("net: ethtool: Add a command to expose current time stamping layer") commit bb8645b0 ("netlink: specs: Introduce new netlink command to get current timestamp") commit d905f9c7 ("net: ethtool: Add a command to list available time stamping layers") commit aed5004e ("netlink: specs: Introduce new netlink command to list available time stamping layers") commit 51bdf316 ("net: Replace hwtstamp_source by timestamping layer") commit 0f7f463d ("net: Change the API of PHY default timestamp to MAC") commit 091fab12 ("net: ethtool: ts: Update GET_TS to reply the current selected timestamp") commit 152c75e1 ("net: ethtool: ts: Let the active time stamping layer be selectable") commit ee60ea6b ("netlink: specs: Introduce time stamping set command") They need more time for reviews. Link: https://lore.kernel.org/all/20231118183529.6e67100c@kernel.org/Signed-off-by:
-
- 18 Nov, 2023 22 commits
-
-
Heiner Kallweit authored
Mirsad proposed a patch to reduce the number of spinlock lock/unlock operations and the function code size. This can be further improved because the function sets a consecutive register block. Suggested-by:
Mirsad Todorovac <mirsad.todorovac@alu.unizg.hr> Signed-off-by:
Heiner Kallweit <hkallweit1@gmail.com> Reviewed-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linuxDavid S. Miller authored
mlx5-updates-2023-11-13 1) Cleanup patches, leftovers from previous cycle 2) Allow sync reset flow when BF MGT interface device is present 3) Trivial ptp refactorings and improvements 4) Add local loopback counter to vport rep stats Signed-off-by: -
git://git.open-mesh.org/linux-mergeDavid S. Miller authored
This feature/cleanup patchset includes the following patches: - bump version strings, by Simon Wunderlich - Implement new multicast packet type, including its transmission, forwarding and parsing, by Linus Lüssing (3 patches) - Switch to new headers for sprintf and array size, by Sven Eckelmann (2 patches) Signed-off-by: -
David S. Miller authored
Petr Machata says: ==================== mlxsw: Add support for new reset flow Ido Schimmel writes: This patchset changes mlxsw to issue a PCI reset during probe and devlink reload so that the PCI firmware could be upgraded without a reboot. Unlike the old version of this patchset [1], in this version the driver no longer tries to issue a PCI reset by triggering a PCI link toggle on its own, but instead calls the PCI core to issue the reset. The PCI APIs require the device lock to be held which is why patches Patches #7 adds reset method quirk for NVIDIA Spectrum devices. Patch #8 adds a debug level print in PCI core so that device ready delay will be printed even if it is shorter than one second. Patches #9-#11 are straightforward preparations in mlxsw. Patch #12 finally implements the new reset flow in mlxsw. Patch #13 adds PCI reset handlers in mlxsw to avoid user space from resetting the device from underneath an unaware driver. Instead, the driver is gracefully de-initialized before the PCI reset and then initialized again after it. Patch #14 adds a PCI reset selftest to make sure this code path does not regress. [1] https://lore.kernel.org/netdev/cover.1679502371.git.petrm@nvidia.com/ ==================== Signed-off-by:
-
Ido Schimmel authored
Test that PCI reset works correctly by verifying that only the expected reset methods are supported and that after issuing the reset the ifindex of the port changes. Signed-off-by:
Ido Schimmel <idosch@nvidia.com> Reviewed-by:
Petr Machata <petrm@nvidia.com> Signed-off-by:
-
Ido Schimmel authored
Implement reset_prepare() and reset_done() handlers that are invoked by the PCI core before and after issuing a PCI reset, respectively. Specifically, implement reset_prepare() by calling mlxsw_core_bus_device_unregister() and reset_done() by calling mlxsw_core_bus_device_register(). This is the same implementation as the reload_{down,up}() devlink operations with the following differences: 1. The devlink instance is unregistered and then registered again after the reset. 2. A reset via the device's command interface (using MRSR register) is not issued during reset_done() as PCI core already issued a PCI reset. Tested: # for i in $(seq 1 10); do echo 1 > /sys/bus/pci/devices/0000\:01\:00.0/reset; done Reviewed-by: -
Ido Schimmel authored
The driver resets the device during probe and during a devlink reload. The current reset method reloads the current firmware version or a pending one, if one was previously flashed using devlink. However, the current reset method does not result in a PCI hot reset, preventing the PCI firmware from being upgraded, unless the system is rebooted. To solve this problem, a new reset command (6) was implemented in the firmware. Unlike the current command (1), after issuing the new command the device will not start the reset immediately, but only after a PCI hot reset. Implement the new reset method by first verifying that it is supported by the current firmware version by querying the Management Capabilities Mask (MCAM) register. If supported, issue the new reset command (6) via MRSR register followed by a PCI reset by calling __pci_reset_function_locked(). Once the PCI firmware is operational, go back to the regular reset flow and wait for the entire device to become ready. That is, repeatedly read the "system_status" register from the BAR until a value of "FW_READY" (0x5E) appears. Tested: # for i in $(seq 1 10); do devlink dev reload pci/0000:01:00.0; done Signed-off-by:
-
Amit Cohen authored
In general, the existing flow of software reset in the driver is: 1. Wait for system ready status. 2. Send MRSR command, to start the reset. 3. Wait for system ready status. This flow will be extended once a new reset command is supported. As a preparation, move step #2 to a separate function. Signed-off-by:
Amit Cohen <amcohen@nvidia.com> Reviewed-by:
-
Amit Cohen authored
In the next patches, mlxsw_pci_sw_reset() will be extended to support more reset types and will not necessarily issue a software reset. Rename the function to reflect that. Signed-off-by:
-
Amit Cohen authored
Currently mlxsw_reg_mrsr_pack() always sets 'command=1'. As preparation for support of new reset flow, pass the command as an argument to the function and add an enum for this field. For now, always pass 'command=1' to the pack() function. Signed-off-by:
-
Ido Schimmel authored
Currently, the time it took a PCI device to become ready after reset is only printed if it was longer than 1000ms ('PCI_RESET_WAIT'). However, for debugging purposes it is useful to know this time even if it was shorter. For example, with the device I am working on, hardware engineers asked to verify that it becomes ready on the first try (no delay). To that end, add a debug level print that can be enabled using dynamic debug. Example: # echo 1 > /sys/bus/pci/devices/0000\:01\:00.0/reset # dmesg -c | grep ready # echo "file drivers/pci/pci.c +p" > /sys/kernel/debug/dynamic_debug/control # echo 1 > /sys/bus/pci/devices/0000\:01\:00.0/reset # dmesg -c | grep ready [ 396.060335] mlxsw_spectrum4 0000:01:00.0: ready 0ms after bus reset # echo "file drivers/pci/pci.c -p" > /sys/kernel/debug/dynamic_debug/control # echo 1 > /sys/bus/pci/devices/0000\:01\:00.0/reset # dmesg -c | grep ready Signed-off-by:Bjorn Helgaas <bhelgaas@google.com> Signed-off-by:
-
Ido Schimmel authored
Spectrum-{1,2,3,4} devices report that a D3hot->D0 transition causes a reset (i.e., they advertise NoSoftRst-). However, this transition does not have any effect on the device: It continues to be operational and network ports remain up. Advertising this support makes it seem as if a PM reset is viable for these devices. Mark it as unavailable to skip it when testing reset methods. Before: # cat /sys/bus/pci/devices/0000\:03\:00.0/reset_method pm bus After: # cat /sys/bus/pci/devices/0000\:03\:00.0/reset_method bus Signed-off-by: -
Ido Schimmel authored
Add an assert to verify that the device lock is always held throughout reload operations. Tested the following flows with netdevsim and mlxsw while lockdep is enabled: netdevsim: # echo "10 1" > /sys/bus/netdevsim/new_device # devlink dev reload netdevsim/netdevsim10 # ip netns add bla # devlink dev reload netdevsim/netdevsim10 netns bla # ip netns del bla # echo 10 > /sys/bus/netdevsim/del_device mlxsw: # devlink dev reload pci/0000:01:00.0 # ip netns add bla # devlink dev reload pci/0000:01:00.0 netns bla # ip netns del bla # echo 1 > /sys/bus/pci/devices/0000\:01\:00.0/remove # echo 1 > /sys/bus/pci/rescan Signed-off-by:
Jiri Pirko <jiri@nvidia.com> Signed-off-by:
-
Ido Schimmel authored
Device drivers register with devlink from their probe routines (under the device lock) by acquiring the devlink instance lock and calling devl_register(). Drivers that support a devlink reload usually implement the reload_{down, up}() operations in a similar fashion to their remove and probe routines, respectively. However, while the remove and probe routines are invoked with the device lock held, the reload operations are only invoked with the devlink instance lock held. It is therefore impossible for drivers to acquire the device lock from their reload operations, as this would result in lock inversion. The motivating use case for invoking the reload operations with the device lock held is in mlxsw which needs to trigger a PCI reset as part of the reload. The driver cannot call pci_reset_function() as this function acquires the device lock. Instead, it needs to call __pci_reset_function_locked which expects the device lock to be held. To that end, adjust devlink to always acquire the device lock before the devlink instance lock when performing a reload. Do that when reload is explicitly triggered by user space by specifying the 'DEVLINK_NL_FLAG_NEED_DEV_LOCK' flag in the pre_doit and post_doit operations of the reload command. A previous patch already handled the case where reload is invoked as part of netns dismantle. Signed-off-by: -
Ido Schimmel authored
Introduce a new private flag ('DEVLINK_NL_FLAG_NEED_DEV_LOCK') to allow netlink commands to specify that they need to acquire the device lock in their pre_doit operation and release it in their post_doit operation. The reload command will use this flag in the subsequent patch. No functional changes intended. Signed-off-by: -
Ido Schimmel authored
Currently, private flags (e.g., 'DEVLINK_NL_FLAG_NEED_PORT') are only used in pre_doit operations, but a subsequent patch will need to conditionally lock and unlock the device lock in pre and post doit operations, respectively. As a preparation, enable the use of private flags in post_doit operations in a similar fashion to how it is done for pre_doit operations. No functional changes intended. Signed-off-by:
-
Ido Schimmel authored
Device drivers register with devlink from their probe routines (under the device lock) by acquiring the devlink instance lock and calling devl_register(). Drivers that support a devlink reload usually implement the reload_{down, up}() operations in a similar fashion to their remove and probe routines, respectively. However, while the remove and probe routines are invoked with the device lock held, the reload operations are only invoked with the devlink instance lock held. It is therefore impossible for drivers to acquire the device lock from their reload operations, as this would result in lock inversion. The motivating use case for invoking the reload operations with the device lock held is in mlxsw which needs to trigger a PCI reset as part of the reload. The driver cannot call pci_reset_function() as this function acquires the device lock. Instead, it needs to call __pci_reset_function_locked which expects the device lock to be held. To that end, adjust devlink to always acquire the device lock before the devlink instance lock when performing a reload. For now, only do that when reload is triggered as part of netns dismantle. Subsequent patches will handle the case where reload is explicitly triggered by user space. Signed-off-by: -
Ido Schimmel authored
The flags are not used outside of the C file so move them there. Suggested-by:
-
David S. Miller authored
Patrick Williams says: ==================== net/ncsi: Add NC-SI 1.2 Get MC MAC Address command NC-SI 1.2 has now been published[1] and adds a new command for "Get MC MAC Address". This is often used by BMCs to get the assigned MAC address for the channel used by the BMC. This change set has been tested on a Broadcomm 200G NIC with updated firmware for NC-SI 1.2 and at least one other non-public NIC design. 1. https://www.dmtf.org/sites/default/files/standards/documents/DSP0222_1.2.0.pdf ==================== Signed-off-by:
-
Peter Delevoryas authored
This change adds support for the NC-SI 1.2 Get MC MAC Address command, specified here: https://www.dmtf.org/sites/default/files/standards/documents/DSP0222_1.2.0.pdf It serves the exact same function as the existing OEM Get MAC Address commands, so if a channel reports that it supports NC-SI 1.2, we prefer to use the standard command rather than the OEM command. Verified with an invalid MAC address and 2 valid ones: [ 55.137072] ftgmac100 1e690000.ftgmac eth0: NCSI: Received 3 provisioned MAC addresses [ 55.137614] ftgmac100 1e690000.ftgmac eth0: NCSI: MAC address 0: 00:00:00:00:00:00 [ 55.138026] ftgmac100 1e690000.ftgmac eth0: NCSI: MAC address 1: fa:ce:b0:0c:20:22 [ 55.138528] ftgmac100 1e690000.ftgmac eth0: NCSI: MAC address 2: fa:ce:b0:0c:20:23 [ 55.139241] ftgmac100 1e690000.ftgmac eth0: NCSI: Unable to assign 00:00:00:00:00:00 to device [ 55.140098] ftgmac100 1e690000.ftgmac eth0: NCSI: Set MAC address to fa:ce:b0:0c:20:22 Signed-off-by:
Peter Delevoryas <peter@pjd.dev> Signed-off-by:
Patrick Williams <patrick@stwcx.xyz> Signed-off-by:
-
Peter Delevoryas authored
The netlink interface for major and minor version numbers doesn't actually return the major and minor version numbers. It reports a u32 that contains the (major, minor, update, alpha1) components as the major version number, and then alpha2 as the minor version number. For whatever reason, the u32 byte order was reversed (ntohl): maybe it was assumed that the encoded value was a single big-endian u32, and alpha2 was the minor version. The correct way to get the supported NC-SI version from the network controller is to parse the Get Version ID response as described in 8.4.44 of the NC-SI spec[1]. Get Version ID Response Packet Format Bits +--------+--------+--------+--------+ Bytes | 31..24 | 23..16 | 15..8 | 7..0 | +-------+--------+--------+--------+--------+ | 0..15 | NC-SI Header | +-------+--------+--------+--------+--------+ | 16..19| Response code | Reason code | +-------+--------+--------+--------+--------+ |20..23 | Major | Minor | Update | Alpha1 | +-------+--------+--------+--------+--------+ |24..27 | reserved | Alpha2 | +-------+--------+--------+--------+--------+ | .... other stuff .... | The major, minor, and update fields are all binary-coded decimal (BCD) encoded [2]. The spec provides examples below the Get Version ID response format in section 8.4.44.1, but for practical purposes, this is an example from a live network card: root@bmc:~# ncsi-util 0x15 NC-SI Command Response: cmd: GET_VERSION_ID(0x15) Response: COMMAND_COMPLETED(0x0000) Reason: NO_ERROR(0x0000) Payload length = 40 20: 0xf1 0xf1 0xf0 0x00 <<<<<<<<< (major, minor, update, alpha1) 24: 0x00 0x00 0x00 0x00 <<<<<<<<< (_, _, _, alpha2) 28: 0x6d 0x6c 0x78 0x30 32: 0x2e 0x31 0x00 0x00 36: 0x00 0x00 0x00 0x00 40: 0x16 0x1d 0x07 0xd2 44: 0x10 0x1d 0x15 0xb3 48: 0x00 0x17 0x15 0xb3 52: 0x00 0x00 0x81 0x19 This should be parsed as "1.1.0". "f" in the upper-nibble means to ignore it, contributing zero. If both nibbles are "f", I think the whole field is supposed to be ignored. Major and minor are "required", meaning they're not supposed to be "ff", but the update field is "optional" so I think it can be ff. I think the simplest thing to do is just set the major and minor to zero instead of juggling some conditional logic or something. bcd2bin() from "include/linux/bcd.h" seems to assume both nibbles are 0-9, so I've provided a custom BCD decoding function. Alpha1 and alpha2 are ISO/IEC 8859-1 encoded, which just means ASCII characters as far as I can tell, although the full encoding table for non-alphabetic characters is slightly different (I think). I imagine the alpha fields are just supposed to be alphabetic characters, but I haven't seen any network cards actually report a non-zero value for either. If people wrote software against this netlink behavior, and were parsing the major and minor versions themselves from the u32, then this would definitely break their code. [1] https://www.dmtf.org/sites/default/files/standards/documents/DSP0222_1.0.0.pdf [2] https://en.wikipedia.org/wiki/Binary-coded_decimal [2] https://en.wikipedia.org/wiki/ISO/IEC_8859-1Signed-off-by: -
Peter Delevoryas authored
Background: 1. CONFIG_NCSI_OEM_CMD_KEEP_PHY If this is enabled, we send an extra OEM Intel command in the probe sequence immediately after discovering a channel (e.g. after "Clear Initial State"). 2. CONFIG_NCSI_OEM_CMD_GET_MAC If this is enabled, we send one of 3 OEM "Get MAC Address" commands from Broadcom, Mellanox (Nvidida), and Intel in the *configuration* sequence for a channel. 3. mellanox,multi-host (or mlx,multi-host) Introduced by this patch: https://lore.kernel.org/all/20200108234341.2590674-1-vijaykhemka@fb.com/ Which was actually originally from cosmo.chou@quantatw.com: https://github.com/facebook/openbmc-linux/commit/9f132a10ec48db84613519258cd8a317fb9c8f1b Cosmo claimed that the Nvidia ConnectX-4 and ConnectX-6 NIC's don't respond to Get Version ID, et. al in the probe sequence unless you send the Set MC Affinity command first. Problem Statement: We've been using a combination of #ifdef code blocks and IS_ENABLED() conditions to conditionally send these OEM commands. It makes adding any new code around these commands hard to understand. Solution: In this patch, I just want to remove the conditionally compiled blocks of code, and always use IS_ENABLED(...) to do dynamic control flow. I don't think the small amount of code this adds to non-users of the OEM Kconfigs is a big deal. Signed-off-by:
-
