aboutsummaryrefslogtreecommitdiff
path: root/drivers
AgeCommit message (Collapse)AuthorFilesLines
2019-04-26s390/qeth: extract helper to determine L2 cast typeJulian Wiedmann3-18/+14
This de-duplicates the L2 and L3 cast-type code, and makes the L2 code a bit more robust by removing the fragile assumption that skb->data always points to the Ethernet Header. This would break in code paths where we pushed the HW header onto the skb. Signed-off-by: Julian Wiedmann <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26s390/qeth: cache max number of available buffer elementsJulian Wiedmann3-24/+22
The QETH_MAX_BUFFER_ELEMENTS() macro effectively returns a constant value. To avoid some redundant pointer chasing and computations in the xmit hot path, cache this value in the queue struct. Take this as opportunity to shrink some of the queue struct's fields to their appropriate value range, slightly reducing its total size. Signed-off-by: Julian Wiedmann <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26s390/qeth: don't clear Output buffers on every queue initJulian Wiedmann1-6/+2
On the first initialization of a queue, its Output Buffers are in a clean state with no attached resources. On every subsequent initialization, qeth_l?_stop_card() has previously put them in a clean state via qeth_drain_output_queues(). So the call to qeth_clear_output_buffer() is redundant and can be removed. While at it, move the initialization of the queue's card pointer into the queue allocation. It never changes afterwards. Signed-off-by: Julian Wiedmann <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26s390/qeth: use IS_* helpers for checking device typeJulian Wiedmann7-78/+54
We have helper macros for all possible device types, replace all remaining open-coded accesses to the type fields. Signed-off-by: Julian Wiedmann <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26s390/qeth: clean up stale buffer state documentationJulian Wiedmann1-28/+13
We don't keep track of Input Buffer states, so remove the comments that make it sound like the qeth_qdio_buffer_states enum applies to Input Buffers. Signed-off-by: Julian Wiedmann <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26s390/qeth: remove RX seqno in skb->cbJulian Wiedmann2-3/+0
It's unclear what exact purpose this seqno may have served in the past. But it's certainly no longer used anymore, as the following napi_gro_receive() will straight away clear this part of the cb again. Suggested-by: Karsten Graul <[email protected]> Signed-off-by: Julian Wiedmann <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26s390: qeth: address type mismatch warningArnd Bergmann1-3/+7
clang produces a harmless warning for each use for the qeth_adp_supported macro: drivers/s390/net/qeth_l2_main.c:559:31: warning: implicit conversion from enumeration type 'enum qeth_ipa_setadp_cmd' to different enumeration type 'enum qeth_ipa_funcs' [-Wenum-conversion] if (qeth_adp_supported(card, IPA_SETADP_SET_PROMISC_MODE)) ~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~ drivers/s390/net/qeth_core.h:179:41: note: expanded from macro 'qeth_adp_supported' qeth_is_ipa_supported(&c->options.adp, f) ~~~~~~~~~~~~~~~~~~~~~ ^ Add a version of this macro that uses the correct types, and remove the unused qeth_adp_enabled() macro that has the same problem. Reviewed-by: Nathan Chancellor <[email protected]> Signed-off-by: Arnd Bergmann <[email protected]> Signed-off-by: Julian Wiedmann <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26mac80211_hwsim: Extended Key ID supportAlexander Wetzel1-0/+6
Allow Extended Key ID to be used with hwsim. Hwsim can only communicate with other hwsim cards, allowing it to bypass creation of A-MPDUs in the first place. Mixing keyIDs in an A-MPDU is therefore impossible and can never cause interoperability issues with other cards. Signed-off-by: Alexander Wetzel <[email protected]> [reword comment slightly] Signed-off-by: Johannes Berg <[email protected]>
2019-04-26mac80211_hwsim: set p2p device interface support indicationAvraham Stern1-0/+2
P2P device interface type was not indicated in the supported interface types even when hwsim was configured with p2p device support. Fix it. Signed-off-by: Avraham Stern <[email protected]> Signed-off-by: Luca Coelho <[email protected]> Signed-off-by: Johannes Berg <[email protected]>
2019-04-26netdevsim: implement ndo_get_devlink_portJiri Pirko1-0/+8
Implement ndo_get_devlink_port and allow switch_id and port_name to be handled by devlink. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: move netdev creation/destruction to dev probeJiri Pirko6-118/+85
Remove the existing way to create netdevsim over rtnetlink and move the netdev creation/destruction to dev probe, so for every probed port, a netdevsim-netdev instance is created. Adjust selftests to work with new interface. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: extend device attrs to support port addition and deletionJiri Pirko3-4/+92
In order to test flows in core, it is beneficial to maintain previously supported possibility to add and delete ports during netdevsim lifetime. Do it by extending device sysfs attrs by "new_port" and "del_port". Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: implement dev probe/remove skeleton with port initializationJiri Pirko4-30/+160
Implement netdevsim bus probing of netdevsim devices. For every probed device create a devlink instance. According to the user-passed value, create a number of ports represented by devlink port instances. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: change debugfs tree topologyJiri Pirko3-15/+11
With the model where dev is represented by devlink and ports are represented by devlink ports, make debugfs file names independent on netdev names. Change the topology to the one illustrated by the following example: $ ls /sys/kernel/debug/netdevsim/ netdevsim1 $ ls /sys/kernel/debug/netdevsim/netdevsim1/ bpf_bind_accept bpf_bind_verifier_delay bpf_bound_progs ports $ ls /sys/kernel/debug/netdevsim/netdevsim1/ports/ 0 1 $ ls /sys/kernel/debug/netdevsim/netdevsim1/ports/0/ bpf_map_accept bpf_offloaded_id bpf_tc_accept bpf_tc_non_bound_accept bpf_xdpdrv_accept bpf_xdpoffload_accept dev ipsec $ ls /sys/kernel/debug/netdevsim/netdevsim1/ports/0/dev -l lrwxrwxrwx 1 root root 0 Apr 13 15:58 /sys/kernel/debug/netdevsim/netdevsim1/ports/0/dev -> ../../../netdevsim1 Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: generate random switch id instead of using dev idJiri Pirko3-2/+5
Current implementation of parent_id/switch_id does not follow the original idea of being unique. The values are "0", "1", etc. Instead of that, generate 32 random bytes. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: merge sdev into devJiri Pirko5-115/+116
As previously introduce dev which is mapped 1:1 to a bus device covers the purpose of the original shared device, merge the sdev code into dev. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: rename dev_init/exit() functions and make them independent on nsJiri Pirko3-28/+30
These functions are going to be called from bus probe/release(), therefore make them independent on ns struct and rename accordingly. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: add bus attributes to add new and delete devicesJiri Pirko3-4/+99
Add a way to add new netdevsim device on netdevsim bus and also to delete existing netdevsim device from the bus. Track the bus devices in using a list. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: use ida for bus device idsJiri Pirko1-3/+10
Instead of increments of u32 value, use ida to manage bus device ids. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: add stub netdevsim driver implementationJiri Pirko1-1/+20
In order to bus probing to work correctly, register a simple netdevsim driver implementation. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: move device registration and related code to bus.cJiri Pirko4-163/+207
Move netdevsim device registration into bus.c and alongside with that the related sysfs attributes. Introduce new struct nsim_bus_dev to represent a netdevsim device on netdevsim bus. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: put netdevsim bus code into separate fileJiri Pirko4-13/+38
As the code related to netdevsim bus is going to get bigger, move the existing code to a separate file. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: rename devlink.c to dev.c to contain per-dev(asic) itemsJiri Pirko4-59/+56
The existing devlink.c code is going to be extended to represent asic device on a bus. As this is about more than just devlink, rename the file. Do appropriate prefix renaming alongside with that. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: create devlink instance per netdevsim instanceJiri Pirko4-202/+124
Currently there is one devlink instance created per network namespace. That is quite odd considering the fact that devlink instance should represent an ASIC. The following patches are going to move the devlink instance even more down to a bus device, but until then, have one devlink instance per netdevsim instance. Struct nsim_devlink is introduced to hold fib setting. The changes in the fib code are only related to holding the configuration per devlink instance instead of network namespace. Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26netdevsim: move device registration on bus to be done earlier in initJiri Pirko1-13/+13
As a dependency of the subsequent patch, mode device registration to be done earlier, directly in nsim_newlink(). Signed-off-by: Jiri Pirko <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-26Merge branch 'for-upstream' of ↵David S. Miller11-44/+1205
git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth-next Johan Hedberg says: ==================== pull request: bluetooth-next 2019-04-25 Here's the main bluetooth-next pull request for the 5.2 kernel. - Added support for Mediatek SDIO controllers - Added support for Broadcom BCM2076B1 UART controller - Added support for Marvel SD8987 chipset - Fix buffer overflow bug in hidp protocol - Various other smaller fixes & improvements Please let me know if there are any issues pulling. Thanks. ==================== Signed-off-by: David S. Miller <[email protected]>
2019-04-25Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/netDavid S. Miller91-245/+470
Two easy cases of overlapping changes. Signed-off-by: David S. Miller <[email protected]>
2019-04-24Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/netLinus Torvalds23-33/+81
Pull networking fixes from David Miller: "Just the usual assortment of small'ish fixes: 1) Conntrack timeout is sometimes not initialized properly, from Alexander Potapenko. 2) Add a reasonable range limit to tcp_min_rtt_wlen to avoid undefined behavior. From ZhangXiaoxu. 3) des1 field of descriptor in stmmac driver is initialized with the wrong variable. From Yue Haibing. 4) Increase mlxsw pci sw reset timeout a little bit more, from Ido Schimmel. 5) Match IOT2000 stmmac devices more accurately, from Su Bao Cheng. 6) Fallback refcount fix in TLS code, from Jakub Kicinski. 7) Fix max MTU check when using XDP in mlx5, from Maxim Mikityanskiy. 8) Fix recursive locking in team driver, from Hangbin Liu. 9) Fix tls_set_device_offload_Rx() deadlock, from Jakub Kicinski. 10) Don't use napi_alloc_frag() outside of softiq context of socionext driver, from Ilias Apalodimas. 11) MAC address increment overflow in ncsi, from Tao Ren. 12) Fix a regression in 8K/1M pool switching of RDS, from Zhu Yanjun. 13) ipv4_link_failure has to validate the headers that are actually there because RAW sockets can pass in arbitrary garbage, from Eric Dumazet" * git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (43 commits) ipv4: add sanity checks in ipv4_link_failure() net/rose: fix unbound loop in rose_loopback_timer() rxrpc: fix race condition in rxrpc_input_packet() net: rds: exchange of 8K and 1M pool net: vrf: Fix operation not supported when set vrf mac net/ncsi: handle overflow when incrementing mac address net: socionext: replace napi_alloc_frag with the netdev variant on init net: atheros: fix spelling mistake "underun" -> "underrun" spi: ST ST95HF NFC: declare missing of table spi: Micrel eth switch: declare missing of table net: stmmac: move stmmac_check_ether_addr() to driver probe netfilter: fix nf_l4proto_log_invalid to log invalid packets netfilter: never get/set skb->tstamp netfilter: ebtables: CONFIG_COMPAT: drop a bogus WARN_ON Documentation: decnet: remove reference to CONFIG_DECNET_ROUTE_FWMARK dt-bindings: add an explanation for internal phy-mode net/tls: don't leak IV and record seq when offload fails net/tls: avoid potential deadlock in tls_set_device_offload_rx() selftests/net: correct the return value for run_afpackettests team: fix possible recursive locking when add slaves ...
2019-04-24net: mvneta: Switch to using devm_alloc_etherdev_mqsRosen Penev1-8/+4
It allows some of the code to be simplified. Tested on Turris Omnia. Signed-off-by: Rosen Penev <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-24net: vrf: Fix operation not supported when set vrf macMiaohe Lin1-0/+2
Vrf device is not able to change mac address now because lack of ndo_set_mac_address. Complete this in case some apps need to do this. Reported-by: Hui Wang <[email protected]> Signed-off-by: Miaohe Lin <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-23Merge tag 'drm-fixes-2019-04-24' of git://anongit.freedesktop.org/drm/drmLinus Torvalds4-7/+25
Pull drm regression fixes from Dave Airlie: "We interrupt your regularly scheduled drm fixes for a regression special. The first is for a fix in i915 that had unexpected side effects fallout in the userspace X.org modesetting driver where X would no longer start. I got tired of the nitpicking and issued a large hammer on it. The X.org driver is buggy, but blackscreen regressions are worse. The second was an oversight that myself and Gerd should have noticed better, Gerd is trying to fix this properly, but the regression is too large to leave, even if the original behaviour is bad in some cases, it's clearly bad to break a bunch of working use cases. I'll likely have a regular fixes pull later, but I really wanted to highlight these" * tag 'drm-fixes-2019-04-24' of git://anongit.freedesktop.org/drm/drm: Revert "drm/virtio: drop prime import/export callbacks" Revert "drm/i915/fbdev: Actually configure untiled displays"
2019-04-23mlxsw: spectrum_router: Prevent ipv6 gateway with v4 route via replace and ↵David Ahern1-0/+2
append mlxsw currently does not support v6 gateways with v4 routes. Commit 19a9d136f198 ("ipv4: Flag fib_info with a fib_nh using IPv6 gateway") prevents a route from being added, but nothing stops the replace or append. Add a catch for them too. $ ip ro add 172.16.2.0/24 via 10.99.1.2 $ ip ro replace 172.16.2.0/24 via inet6 fe80::202:ff:fe00:b dev swp1s0 Error: mlxsw_spectrum: IPv6 gateway with IPv4 route is not supported. $ ip ro append 172.16.2.0/24 via inet6 fe80::202:ff:fe00:b dev swp1s0 Error: mlxsw_spectrum: IPv6 gateway with IPv4 route is not supported. Signed-off-by: David Ahern <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-23net: socionext: replace napi_alloc_frag with the netdev variant on initIlias Apalodimas1-4/+7
The netdev variant is usable on any context since it disables interrupts. The napi variant of the call should only be used within softirq context. Replace napi_alloc_frag on driver init with the correct netdev_alloc_frag call Changes since v1: - Adjusted commit message Acked-by: Ard Biesheuvel <[email protected]> Acked-by: Jassi Brar <[email protected]> Fixes: 4acb20b46214 ("net: socionext: different approach on DMA") Signed-off-by: Ilias Apalodimas <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-23net: atheros: fix spelling mistake "underun" -> "underrun"Colin Ian King4-5/+5
There are spelling mistakes in structure elements, fix these. Signed-off-by: Colin Ian King <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-23ravb: Avoid unsupported internal delay mode for R-Car E3/D3Simon Horman1-2/+13
According to the R-Car Gen3 Hardware Manual Rev 1.50 of Nov 30, 2018, the TX clock internal delay mode isn't supported on R-Car E3 (r8a77990) or D3 (r8a77995). And by extension it is also not supported by RZ/G2E (r9a774c0). This matches all ES versions of the affected SoCs as it is not clear if this problem will be resolved in newer chips. This can be revisited, as necessary. This patch does not error-out if PHY_INTERFACE_MODE_RGMII_ID or PHY_INTERFACE_MODE_RGMII_TXID are used on SoCs where TX clock delay mode is not supported as there is a risk of introducing a regression when used in conjunction with older DT blobs present in the field. Rather, a warning is logged in such cases. Based on work by Kazuya Mizuguchi. Signed-off-by: Simon Horman <[email protected]> Reviewed-by: Andrew Lunn <[email protected]> Reviewed-by: Sergei Shtylyov <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-23isdn:mISDN: fix misuse of %x in hfcpci.cFuqian Huang1-3/+3
Pointers should be printed with %p or %px rather than cast to (u_long) and printed with %lx. Change %lx to %p to print the pointer. Change %lx to %pad to print the dma_addr_t. Signed-off-by: Fuqian Huang <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-23isdn: hisax: Fix misuse of %x in config.cFuqian Huang1-3/+3
Pointers should be printed with %p or %px rather than cast to (u_long) type and printed with %lX. As the function seems to be for debug purpose. Change %lX to %px to print the pointer value. Signed-off-by: Fuqian Huang <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-24Revert "drm/virtio: drop prime import/export callbacks"Dave Airlie3-0/+20
This patch does more harm than good, as it breaks both Xwayland and gnome-shell with X11. Xwayland requires DRI3 & DRI3 requires PRIME. X11 crash for obscure double-free reason which are hard to debug (starting X11 by hand doesn't trigger the crash). I don't see an apparent problem implementing those stub prime functions, they may return an error at run-time, and it seems to be handled fine by GNOME at least. This reverts commit b318e3ff7ca065d6b107e424c85a63d7a6798a69. [airlied: This broke userspace for virtio-gpus, and regressed things from DRI3 to DRI2. This brings back the original problem, but it's better than regressions.] Fixes: b318e3ff7ca065d6b107e424c85a63d7a6798a ("drm/virtio: drop prime import/export callbacks") Signed-off-by: Marc-André Lureau <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
2019-04-24Revert "drm/i915/fbdev: Actually configure untiled displays"Dave Airlie1-7/+5
This reverts commit d179b88deb3bf6fed4991a31fd6f0f2cad21fab5. This commit is documented to break userspace X.org modesetting driver in certain configurations. The X.org modesetting userspace driver is broken. No fixes are available yet. In order for this patch to be applied it either needs a config option or a workaround developed. This has been reported a few times, saying it's a userspace problem is clearly against the regression rules. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109806 Signed-off-by: Dave Airlie <[email protected]> Cc: <[email protected]> # v3.19+
2019-04-23Merge tag 'mlx5-updates-2019-04-22' of ↵David S. Miller14-202/+399
git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux Saeed Mahameed says: ==================== mlx5-updates-2019-04-22 This series includes updates to mlx5e driver RX data path and some significant XDP RX/TX improvements to overcome/mitigate HW and PCIE bottlenecks. From Tariq: 1) Some Enhancements in rq->flags 2) Stabilize RX packet rate (on Striding RQ) with multiple outstanding UMR posts In this patch, we add support for multiple outstanding UMR posts, to allow faster gap closure between consuming MPWQEs and reposting them back into the WQ. Performance test: As expected, huge improvement in large-scale (48 cores). xdp_redirect_map, 64B UDP multi-stream. Redirect from ConnectX-5 100Gbps to ConnectX-6 100Gbps. CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz. Before: Unstable, 7 to 30 Mpps After: Stable, at 70.5 Mpps From Shay: 3) XDP, Inline small packets into the TX MPWQE in XDP xmit flow Upon high packet rate with multiple CPUs TX workloads, much of the HCA's resources are spent on prefetching TX descriptors, thus affecting transmission rates. This patch comes to mitigate this problem by moving some workload to the CPU and reducing the HW data prefetch overhead for small packets (<= 256B). When forwarding packets with XDP, a packet that is smaller than a certain size (set to ~256 bytes) would be sent inline within its WQE TX descrptor (mem-copied), when the hardware tx queue is congested beyond a pre-defined water-mark. Performance: Tested packet rate for UDP 64Byte multi-stream over two dual port ConnectX-5 100Gbps NICs. CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz * Tested with hyper-threading disabled XDP_TX: | | before | after | | | 24 rings | 51Mpps | 116Mpps | +126% | | 1 ring | 12Mpps | 12Mpps | same | XDP_REDIRECT: ** Below is the transmit rate, not the redirection rate which might be larger, and is not affected by this patch. | | before | after | | | 32 rings | 64Mpps | 92Mpps | +43% | | 1 ring | 6.4Mpps | 6.4Mpps | same | As we can see, feature significantly improves scaling, without hurting single ring performance. From Maxim: 4) Some trivial refactoring and code improvements prior to a larger series to support AF_XDP. ==================== Acked-by: Jakub Kicinski <[email protected]> Signed-off-by: David S. Miller <[email protected]>
2019-04-23Bluetooth: hci_qca: Fix crash with non-serdev devicesMatthias Kaehlcke1-1/+2
qca_set_baudrate() calls serdev_device_wait_until_sent() assuming that the HCI is always associated with a serdev device. This isn't true for ROME controllers instantiated through ldisc, where the call causes a crash due to a NULL pointer dereferentiation. Only call the function when we have a serdev device. The timeout for ROME devices at the end of qca_set_baudrate() is long enough to be reasonably sure that the command was sent. Fixes: fa9ad876b8e0 ("Bluetooth: hci_qca: Add support for Qualcomm Bluetooth chip wcn3990") Reported-by: Balakrishna Godavarthi <[email protected]> Reported-by: Rocky Liao <[email protected]> Signed-off-by: Matthias Kaehlcke <[email protected]> Reviewed-by: Rocky Liao <[email protected]> Tested-by: Rocky Liao <[email protected]> Reviewed-by: Balakrishna Godavarthi <[email protected]> Signed-off-by: Marcel Holtmann <[email protected]>
2019-04-23net/mlx5e: Use #define for the WQE wait timeout constantMaxim Mikityanskiy1-3/+7
Create a #define for the timeout of mlx5e_wait_for_min_rx_wqes to clarify the meaning of a magic number. Signed-off-by: Maxim Mikityanskiy <[email protected]> Reviewed-by: Tariq Toukan <[email protected]> Signed-off-by: Saeed Mahameed <[email protected]>
2019-04-23net/mlx5e: Remove unused rx_page_reuse statMaxim Mikityanskiy2-5/+0
Remove the no longer used page_reuse stat of RQs. Signed-off-by: Maxim Mikityanskiy <[email protected]> Reviewed-by: Tariq Toukan <[email protected]> Signed-off-by: Saeed Mahameed <[email protected]>
2019-04-23net/mlx5e: Take HW interrupt trigger into a functionMaxim Mikityanskiy3-9/+13
mlx5e_trigger_irq posts a NOP to the ICO SQ just to trigger an IRQ and enter the NAPI poll on the right CPU according to the affinity. Use it in mlx5e_activate_rq. Signed-off-by: Maxim Mikityanskiy <[email protected]> Reviewed-by: Tariq Toukan <[email protected]> Signed-off-by: Saeed Mahameed <[email protected]>
2019-04-23net/mlx5e: Remove unused parameterMaxim Mikityanskiy3-11/+9
mdev is unused in mlx5e_rx_is_linear_skb. Signed-off-by: Maxim Mikityanskiy <[email protected]> Reviewed-by: Tariq Toukan <[email protected]> Signed-off-by: Saeed Mahameed <[email protected]>
2019-04-23net/mlx5e: Add an underflow warning commentMaxim Mikityanskiy1-2/+5
mlx5e_mpwqe_get_log_rq_size calculates the number of WQEs (N) based on the requested number of frames in the RQ (F) and the number of packets per WQE (P). It ensures that N is not less than the minimum number of WQEs in an RQ (N_min). Arithmetically, it means that F / P >= N_min should be true. This function deals with logarithms, so it should check that log(F) - log(P) >= log(N_min). However, if F < P, this expression will cause an unsigned underflow. Check log(F) >= log(P) + log(N_min) instead. Signed-off-by: Maxim Mikityanskiy <[email protected]> Reviewed-by: Tariq Toukan <[email protected]> Signed-off-by: Saeed Mahameed <[email protected]>
2019-04-23net/mlx5e: Move parameter calculation functions to en/params.cMaxim Mikityanskiy4-99/+128
This commit moves the parameter calculation functions to a separate file for better modularity and code sharing with future features. Signed-off-by: Maxim Mikityanskiy <[email protected]> Reviewed-by: Tariq Toukan <[email protected]> Signed-off-by: Saeed Mahameed <[email protected]>
2019-04-23net/mlx5e: Report mlx5e_xdp_set errorsMaxim Mikityanskiy1-1/+1
If the channels fail to reopen after setting an XDP program, return the error code instead of 0. A proper fix is still needed, as now any error while reopening the channels brings the interface down. This patch only adds error reporting. Signed-off-by: Maxim Mikityanskiy <[email protected]> Reviewed-by: Tariq Toukan <[email protected]> Signed-off-by: Saeed Mahameed <[email protected]>
2019-04-23net/mlx5e: Remove unused parameterMaxim Mikityanskiy1-2/+1
params is unused in mlx5e_init_di_list. Signed-off-by: Maxim Mikityanskiy <[email protected]> Reviewed-by: Tariq Toukan <[email protected]> Signed-off-by: Saeed Mahameed <[email protected]>
2019-04-23net/mlx5e: XDP, Inline small packets into the TX MPWQE in XDP xmit flowShay Agroskin6-15/+82
Upon high packet rate with multiple CPUs TX workloads, much of the HCA's resources are spent on prefetching TX descriptors, thus affecting transmission rates. This patch comes to mitigate this problem by moving some workload to the CPU and reducing the HW data prefetch overhead for small packets (<= 256B). When forwarding packets with XDP, a packet that is smaller than a certain size (set to ~256 bytes) would be sent inline within its WQE TX descrptor (mem-copied), when the hardware tx queue is congested beyond a pre-defined water-mark. This is added to better utilize the HW resources (which now makes one less packet data prefetch) and allow better scalability, on the account of CPU usage (which now 'memcpy's the packet into the WQE). To load balance between HW and CPU and get max packet rate, we use watermarks to detect how much the HW is congested and move the work loads back and forth between HW and CPU. Performance: Tested packet rate for UDP 64Byte multi-stream over two dual port ConnectX-5 100Gbps NICs. CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz * Tested with hyper-threading disabled XDP_TX: | | before | after | | | 24 rings | 51Mpps | 116Mpps | +126% | | 1 ring | 12Mpps | 12Mpps | same | XDP_REDIRECT: ** Below is the transmit rate, not the redirection rate which might be larger, and is not affected by this patch. | | before | after | | | 32 rings | 64Mpps | 92Mpps | +43% | | 1 ring | 6.4Mpps | 6.4Mpps | same | As we can see, feature significantly improves scaling, without hurting single ring performance. Signed-off-by: Shay Agroskin <[email protected]> Signed-off-by: Tariq Toukan <[email protected]> Signed-off-by: Saeed Mahameed <[email protected]>
Powered by cgit, Gruvboxed by Blaster4385.