- 10 6月, 2021 11 次提交
-
-
由 Colin Ian King 提交于
There are missing { } around a block of code on an if statement. Fix this by adding them in. Addresses-Coverity: ("Nesting level does not match indentation") Fixes: 46682cb8 ("net: stmmac: enable Intel mGbE 2.5Gbps link speed") Signed-off-by: NColin Ian King <colin.king@canonical.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Yang Yingliang 提交于
Use the devm_platform_ioremap_resource_byname() helper instead of calling platform_get_resource_byname() and devm_ioremap_resource() separately. Signed-off-by: NYang Yingliang <yangyingliang@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Matteo Croce 提交于
Most of the time during the RX is caused by the compound_head() call done at the end of the RX loop: │ build_skb(): [...] │ static inline struct page *compound_head(struct page *page) │ { │ unsigned long head = READ_ONCE(page->compound_head); 65.23 │ ldr x2, [x1, #8] Prefetch the page struct as soon as possible, to speedup the RX path noticeabily by a ~3-4% packet rate in a drop test. │ build_skb(): [...] │ static inline struct page *compound_head(struct page *page) │ { │ unsigned long head = READ_ONCE(page->compound_head); 17.92 │ ldr x2, [x1, #8] Signed-off-by: NMatteo Croce <mcroce@microsoft.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Matteo Croce 提交于
In the RX buffer, the received data starts after a headroom used to align the IP header and to allow prepending headers efficiently. The prefetch() should take this into account, and prefetch from the very start of the received data. We can see that ether_addr_equal_64bits(), which is the first function to access the data, drops from the top of the perf top output. prefetch(data): Overhead Shared Object Symbol 11.64% [kernel] [k] eth_type_trans prefetch(data + MVPP2_MH_SIZE + MVPP2_SKB_HEADROOM): Overhead Shared Object Symbol 13.42% [kernel] [k] build_skb 10.35% [mvpp2] [k] mvpp2_rx 9.35% [kernel] [k] __netif_receive_skb_core 8.24% [kernel] [k] kmem_cache_free 7.97% [kernel] [k] dev_gro_receive 7.68% [kernel] [k] page_pool_put_page 7.32% [kernel] [k] kmem_cache_alloc 7.09% [mvpp2] [k] mvpp2_bm_pool_put 3.36% [kernel] [k] eth_type_trans Also, move the eth_type_trans() call a bit down, to give the RAM more time to prefetch the data. Signed-off-by: NMatteo Croce <mcroce@microsoft.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Yang Yingliang 提交于
Use the devm_platform_ioremap_resource_byname() helper instead of calling platform_get_resource_byname() and devm_ioremap_resource() separately. Signed-off-by: NYang Yingliang <yangyingliang@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Yang Yingliang 提交于
Use the devm_platform_ioremap_resource_byname() helper instead of calling platform_get_resource_byname() and devm_ioremap_resource() separately. Signed-off-by: NYang Yingliang <yangyingliang@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Yang Yingliang 提交于
It will cause null-ptr-deref if platform_get_resource() returns NULL, we need check the return value. Signed-off-by: NYang Yingliang <yangyingliang@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Baokun Li 提交于
Using list_move_tail() instead of list_del() + list_add_tail() in hclge_main.c. Reported-by: NHulk Robot <hulkci@huawei.com> Signed-off-by: NBaokun Li <libaokun1@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Baokun Li 提交于
Using list_move_tail() instead of list_del() + list_add_tail() in hclgevf_main.c. Reported-by: NHulk Robot <hulkci@huawei.com> Signed-off-by: NBaokun Li <libaokun1@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Baokun Li 提交于
Using list_move() instead of list_del() + list_add() in nfp_cppcore.c. Reported-by: NHulk Robot <hulkci@huawei.com> Signed-off-by: NBaokun Li <libaokun1@huawei.com> Reviewed-by: NSimon Horman <simon.horman@corigine.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Yang Yingliang 提交于
Use devm_platform_get_and_ioremap_resource() to simplify code. Signed-off-by: NYang Yingliang <yangyingliang@huawei.com> Reviewed-by: NSergei Shtylyov <sergei.shtylyov@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
- 09 6月, 2021 27 次提交
-
-
由 Matthew Hagan 提交于
We are currently assuming that GMAC_AHB_RESET will already be deasserted by the bootloader. However if this has not been done, probing of the GMAC will fail. To remedy this we must ensure GMAC_AHB_RESET has been deasserted prior to probing. v2 changes: - remove NULL condition check for stmmac_ahb_rst in stmmac_main.c - unwrap dev_err() message in stmmac_main.c - add PTR_ERR() around plat->stmmac_ahb_rst in stmmac_platform.c v3 changes: - add error pointer to dev_err() output - add reset_control_assert(stmmac_ahb_rst) in stmmac_dvr_remove - revert PTR_ERR() around plat->stmmac_ahb_rst since this is performed on the returned value of ret by the calling function Signed-off-by: NMatthew Hagan <mnhagan88@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Yang Yingliang 提交于
Use devm_platform_get_and_ioremap_resource() to simplify code. Signed-off-by: NYang Yingliang <yangyingliang@huawei.com> Reviewed-by: NSergei Shtylyov <sergei.shtylyov@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Yang Yingliang 提交于
Use devm_platform_get_and_ioremap_resource() and devm_platform_ioremap_resource_byname to simplify code. Signed-off-by: NYang Yingliang <yangyingliang@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Wang Hai 提交于
Convert list_for_each() to list_for_each_entry() where applicable. This simplifies the code. Reported-by: NHulk Robot <hulkci@huawei.com> Signed-off-by: NWang Hai <wanghai38@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Vladimir Oltean 提交于
priv->plat->mdio_bus_data is optional, some platforms may not set it, however we proceed to look straight at priv->plat->mdio_bus_data->has_xpcs. Since the xpcs is instantiated based on the has_xpcs property, we can avoid looking at the priv->plat->mdio_bus_data structure altogether and just check for the presence of the xpcs pointer. Fixes: 11059740 ("net: pcs: xpcs: convert to phylink_pcs_ops") Reported-by: NDan Carpenter <dan.carpenter@oracle.com> Signed-off-by: NVladimir Oltean <vladimir.oltean@nxp.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Wang Hai 提交于
Convert list_for_each() to list_for_each_entry() where applicable. This simplifies the code. Reported-by: NHulk Robot <hulkci@huawei.com> Signed-off-by: NWang Hai <wanghai38@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiaran Zhang 提交于
During initialization, the driver logs and clears the hw errors that already occurred. For device supports imp-handle ras capability, it needs handle different error status, otherwise it may cause wrong reset. So fix it by adding a new processing branch. Signed-off-by:
Jiaran Zhang <zhangjiaran@huawei.com> Signed-off-by: NGuangbin Huang <huangguangbin2@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiaran Zhang 提交于
Update error recovery module and type for RoCE. The enumeration values of module names and error types are not sorted in sequence. If use the current printing mode, they cannot be correctly printed. Use the index mode, If mod_id and type_id match the enumerated value, display the corresponding information. Signed-off-by:
Jiaran Zhang <zhangjiaran@huawei.com> Signed-off-by: NWeihang Li <liweihang@huawei.com> Signed-off-by: NGuangbin Huang <huangguangbin2@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiaran Zhang 提交于
IMP(Intelligent Management Processor) firmware add a new feature to handle and consolidate RAS information for new devices, NIC driver only needs to query the reported RAS information. NIC driver adds support for this feature. Driver queries device capability to check whether IMP support this feature, If yes, execute the new RAS processing branch. In order to add a method to check whether PF supports imp-handle RAS feature, add dumping this info in debugfs. Signed-off-by:
Jiaran Zhang <zhangjiaran@huawei.com> Signed-off-by: NGuangbin Huang <huangguangbin2@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiaran Zhang 提交于
To adapt to hardware modification and ensure that the driver is compatible with the original error handling content, we need to add the RAS compatibility adaptation solution. Add a processing branch to the driver during error handling. In the new processing branch, NIC fault information is integrated by the IMP. An interaction command is added between the driver and IMP to query and clear the fault source and interrupt source. The IMP integrates error information and reports the highest reset level to the driver. Signed-off-by:
Jiaran Zhang <zhangjiaran@huawei.com> Signed-off-by: NGuangbin Huang <huangguangbin2@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Yufeng Mo 提交于
Currently, hardware errors can be reported through AER or MSI-X mode. However, the AER mode is intended to handle only bus errors, but not hardware errors. On the other hand, virtual machines cannot handle AER errors. When an AER error is reported, virtual machines will be suspended. So add support for handling all these hardware errors through MSI-X mode which depends on a newer version of firmware, and reserve the handler of the AER mode for compatibility. Signed-off-by: NYufeng Mo <moyufeng@huawei.com> Signed-off-by:
Jiaran Zhang <zhangjiaran@huawei.com> Signed-off-by: NGuangbin Huang <huangguangbin2@huawei.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shay Agroskin 提交于
Restructure some ethtool to a switch-case blocks to make it more uniform with other similar functions. Also restructure variable declaration to create reversed x-mas tree. Signed-off-by: NArthur Kiyanovski <akiyano@amazon.com> Signed-off-by: NShay Agroskin <shayagr@amazon.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shay Agroskin 提交于
Use dev_alloc() when allocating RX buffers instead of specifying the allocation flags explicitly. This result in same behaviour with less code. Also move the page allocation and its DMA mapping into a function. This creates a logical block, which may help understanding the code. Signed-off-by: NShay Agroskin <shayagr@amazon.com> Reported-by: Nkernel test robot <lkp@intel.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shay Agroskin 提交于
The ena_ring_tx_doorbell() is introduced to call the doorbell and increase the driver's corresponding stat. Signed-off-by: NIdo Segev <idose@amazon.com> Signed-off-by: NShay Agroskin <shayagr@amazon.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shay Agroskin 提交于
Remove the module param 'debug' which allows to specify the message level of the driver. This value can be specified using ethtool command. Also reduce the message level of LLQ support to be a warning since it is not an indication of an error. Signed-off-by: NArthur Kiyanovski <akiyano@amazon.com> Signed-off-by: NShay Agroskin <shayagr@amazon.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shay Agroskin 提交于
There are instances when we want to know when the last napi was called for debugging. On stuck / heavy loaded CPUs, the ena napi handler might not be called for a long period of time. This stat can help us to determine how much time passed since the last execution of napi. Signed-off-by: NSameeh Jubran <sameehj@amazon.com> Signed-off-by: NShay Agroskin <shayagr@amazon.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shay Agroskin 提交于
This patch converts the RX path to use build_skb() for packets larger than copybreak (set to 256 by default). This function makes the first descriptor's page to be the linear part of the sk_buff struct buffer. Also remove the SKB description from the README since most of it no longer relevant and the parts that are left don't add information. Signed-off-by: NShay Agroskin <shayagr@amazon.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shay Agroskin 提交于
Add prints to improve logging of driver's errors. Signed-off-by: NArthur Kiyanovski <akiyano@amazon.com> Signed-off-by: NShay Agroskin <shayagr@amazon.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shay Agroskin 提交于
The ENA_DEFAULT_MIN_RX_BUFF_ALLOC_SIZE macro, ena_xdp_queues_present() function and SUSPEND_RESUME enums aren't used in the driver, and so not needed. Signed-off-by: NArthur Kiyanovski <akiyano@amazon.com> Signed-off-by: NGal Pressman <galpress@amazon.com> Signed-off-by: NSameeh Jubran <sameehj@amazon.com> Signed-off-by: NShay Agroskin <shayagr@amazon.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shay Agroskin 提交于
This tweaks several small places to improve the data access in fast path: * Remove duplicates of first_interrupt flag and surround it with WRITE/READ_ONCE macros: The flag is used to detect HW disorders in its interrupt communication with the driver. The flag is set when an interrupt is received and used in the health check function (ena_timer_service()) to help it find irregularities. * Reorder some fields in ena_napi struct to take better advantage of cache access pattern. * Move XDP TX queue number to a variable to save its calculation for every packet. * Use likely in a condition to improve branch prediction The 'first_interrupt' and 'interrupt_masked' flags were moved to reside in the same cache line as the first fields of 'napi' struct. This placement ensures that all memory accessed during upper-half handler reside in the same cacheline (napi_schedule_irqoff() only accesses 'state' and 'poll_list' fields which are at the beginning of napi struct). Signed-off-by: NSameeh Jubran <sameehj@amazon.com> Signed-off-by: NShay Agroskin <shayagr@amazon.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Mykola Kostenok 提交于
mlxsw_thermal_module_trips_update() is used to update the trip points of the module's thermal zone. Currently, this is done by querying the thresholds from the module's EEPROM via MCIA register. This data does not pass validation and in some cases can be unreliable. For example, due to some problem with transceiver module. Previous patch made it possible to read module's temperature and thresholds via MTMP register. Therefore, extend mlxsw_thermal_module_trips_update() to use the thresholds queried from MTMP, if valid. This is both more reliable and more efficient than current method, as temperature and thresholds are queried in one transaction instead of three. This is significant when working over a slow bus such as I2C. Signed-off-by: NMykola Kostenok <c_mykolak@nvidia.com> Acked-by: NVadim Pasternak <vadimp@nvidia.com> Signed-off-by: NIdo Schimmel <idosch@nvidia.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Mykola Kostenok 提交于
Provide new function mlxsw_thermal_module_temp_and_thresholds_get() for reading temperature and temperature thresholds by a single operation. The motivation is to reduce the number of transactions with the device which is important when operating over a slow bus such as I2C. Currently, the sole caller of the function is only using it to read the module's temperature. The next patch will also use it to query the module's temperature thresholds. Signed-off-by: NMykola Kostenok <c_mykolak@nvidia.com> Acked-by: NVadim Pasternak <vadimp@nvidia.com> Signed-off-by: NIdo Schimmel <idosch@nvidia.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Mykola Kostenok 提交于
Currently, module temperature thresholds are obtained from Management Cable Info Access (MCIA) register by specifying the thresholds offsets within module EEPROM layout. This data does not pass validation and in some cases can be unreliable. For example, due to some problem with the module. Add support for a new feature provided by Management Temperature (MTMP) register for sanitization of temperature thresholds values. Extend mlxsw_env_module_temp_thresholds_get() to get temperature thresholds through MTMP field 'max_operational_temperature' - if it is not zero, feature is supported. Otherwise fallback to old method and get the thresholds through MCIA. Signed-off-by: NMykola Kostenok <c_mykolak@nvidia.com> Acked-by: NVadim Pasternak <vadimp@nvidia.com> Signed-off-by: NIdo Schimmel <idosch@nvidia.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Mykola Kostenok 提交于
Extend Management Temperature (MTMP) register with new field specifying the maximum temperature threshold. Extend mlxsw_reg_mtmp_unpack() function with two extra arguments, providing high and maximum temperature thresholds. For modules, these thresholds correspond to critical and emergency thresholds that are read from the module's EEPROM. Signed-off-by: NMykola Kostenok <c_mykolak@nvidia.com> Acked-by: NVadim Pasternak <vadimp@nvidia.com> Signed-off-by: NIdo Schimmel <idosch@nvidia.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Amit Cohen 提交于
The abort mechanism was introduced in commit 8e05fd71 ("fib: hook IPv4 fib for hardware offload") with the purpose of falling back to software-based routing in case of a route programming error in hardware. The process is irreversible and requires users to reload the offloading driver or reboot the machine. While this approach might make sense in theory, it makes very little sense in practice. In the case of high speed ASICs such as the Spectrum ASIC, the abort mechanism effectively kills the machine upon a non-fatal error such as a route programming error. Such an extreme policy does not belong in the kernel, especially when user space can simply try to reprogram the route following the RTM_NEWROUTE failure notification. Therefore, remove the abort mechanism. Signed-off-by: NAmit Cohen <amcohen@nvidia.com> Reviewed-by: NPetr Machata <petrm@nvidia.com> Signed-off-by: NIdo Schimmel <idosch@nvidia.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Voon Weifeng 提交于
The Intel mGbE supports 2.5Gbps link speed by increasing the clock rate by 2.5 times of the original rate. In this mode, the serdes/PHY operates at a serial baud rate of 3.125 Gbps and the PCS data path and GMII interface of the MAC operate at 312.5 MHz instead of 125 MHz. For Intel mGbE, the overclocking of 2.5 times clock rate to support 2.5G is only able to be configured in the BIOS during boot time. Kernel driver has no access to modify the clock rate for 1Gbps/2.5G mode. The way to determined the current 1G/2.5G mode is by reading a dedicated adhoc register through mdio bus. In short, after the system boot up, it is either in 1G mode or 2.5G mode which not able to be changed on the fly. Compared to 1G mode, the 2.5G mode selects the 2500BASEX as PHY interface and disables the xpcs_an_inband. This is to cater for some PHYs that only supports 2500BASEX PHY interface with no autonegotiation. v2: remove MAC supported link speed masking v3: Restructure to introduce intel_speed_mode_2500() to read serdes registers for max speed supported and select the appropritate configuration. Use max_speed to determine the supported link speed mask. Signed-off-by: NVoon Weifeng <weifeng.voon@intel.com> Signed-off-by: NMichael Sit Wei Hong <michael.wei.hong.sit@intel.com> Reviewed-by: NVladimir Oltean <vladimir.oltean@nxp.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Voon Weifeng 提交于
This patch is a preparation patch for the enabling of Intel mGbE 2.5Gbps link speed. The Intel mGbR link speed configuration (1G/2.5G) is depends on a mdio ADHOC register which can be configured in the bios menu. As PHY interface might be different for 1G and 2.5G, the mdio bus need be ready to check the link speed and select the PHY interface before probing the xPCS. Signed-off-by: NVoon Weifeng <weifeng.voon@intel.com> Signed-off-by: NMichael Sit Wei Hong <michael.wei.hong.sit@intel.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
- 08 6月, 2021 2 次提交
-
-
由 Matteo Croce 提交于
Use the new recycling API for page_pool. In a drop rate test, the packet rate increased by 10%, from 296 Kpps to 326 Kpps. perf top on a stock system shows: Overhead Shared Object Symbol 23.66% [kernel] [k] __pi___inval_dcache_area 22.85% [mvneta] [k] mvneta_rx_swbm 7.54% [kernel] [k] kmem_cache_alloc 6.49% [kernel] [k] eth_type_trans 3.94% [kernel] [k] dev_gro_receive 3.91% [kernel] [k] __netif_receive_skb_core 3.91% [kernel] [k] kmem_cache_free 3.76% [kernel] [k] page_pool_release_page 3.56% [kernel] [k] free_unref_page 2.40% [kernel] [k] build_skb 1.49% [kernel] [k] skb_release_data 1.45% [kernel] [k] __alloc_pages_bulk 1.30% [kernel] [k] page_frag_free And this is the same output with recycling enabled: Overhead Shared Object Symbol 26.41% [kernel] [k] __pi___inval_dcache_area 25.00% [mvneta] [k] mvneta_rx_swbm 8.14% [kernel] [k] kmem_cache_alloc 6.84% [kernel] [k] eth_type_trans 4.44% [kernel] [k] __netif_receive_skb_core 4.38% [kernel] [k] kmem_cache_free 4.16% [kernel] [k] dev_gro_receive 3.21% [kernel] [k] page_pool_put_page 2.41% [kernel] [k] build_skb 1.82% [kernel] [k] skb_release_data 1.61% [kernel] [k] napi_gro_receive 1.25% [kernel] [k] page_pool_refill_alloc_cache 1.16% [kernel] [k] __netif_receive_skb_list_core We can see that page_pool_release_page(), free_unref_page() and __alloc_pages_bulk() are no longer on top of the list when receiving traffic. The test was done with mausezahn on the TX side with 64 byte raw ethernet frames. Signed-off-by: NMatteo Croce <mcroce@microsoft.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Matteo Croce 提交于
Use the new recycling API for page_pool. In a drop rate test, the packet rate is almost doubled, from 1110 Kpps to 2128 Kpps. perf top on a stock system shows: Overhead Shared Object Symbol 34.88% [kernel] [k] page_pool_release_page 8.06% [kernel] [k] free_unref_page 6.42% [mvpp2] [k] mvpp2_rx 6.07% [kernel] [k] eth_type_trans 5.18% [kernel] [k] __netif_receive_skb_core 4.95% [kernel] [k] build_skb 4.88% [kernel] [k] kmem_cache_free 3.97% [kernel] [k] kmem_cache_alloc 3.45% [kernel] [k] dev_gro_receive 2.73% [kernel] [k] page_frag_free 2.07% [kernel] [k] __alloc_pages_bulk 1.99% [kernel] [k] arch_local_irq_save 1.84% [kernel] [k] skb_release_data 1.20% [kernel] [k] netif_receive_skb_list_internal With packet rate stable at 1100 Kpps: tx: 0 bps 0 pps rx: 532.7 Mbps 1110 Kpps tx: 0 bps 0 pps rx: 532.6 Mbps 1110 Kpps tx: 0 bps 0 pps rx: 532.4 Mbps 1109 Kpps tx: 0 bps 0 pps rx: 532.1 Mbps 1109 Kpps tx: 0 bps 0 pps rx: 531.9 Mbps 1108 Kpps tx: 0 bps 0 pps rx: 531.9 Mbps 1108 Kpps And this is the same output with recycling enabled: Overhead Shared Object Symbol 12.91% [kernel] [k] eth_type_trans 12.54% [mvpp2] [k] mvpp2_rx 9.67% [kernel] [k] build_skb 9.63% [kernel] [k] __netif_receive_skb_core 8.44% [kernel] [k] page_pool_put_page 8.07% [kernel] [k] kmem_cache_free 7.79% [kernel] [k] kmem_cache_alloc 6.86% [kernel] [k] dev_gro_receive 3.19% [kernel] [k] skb_release_data 2.41% [kernel] [k] netif_receive_skb_list_internal 2.18% [kernel] [k] page_pool_refill_alloc_cache 1.76% [kernel] [k] napi_gro_receive 1.61% [kernel] [k] kfree_skb 1.20% [kernel] [k] dma_sync_single_for_device 1.16% [mvpp2] [k] mvpp2_poll 1.12% [mvpp2] [k] mvpp2_read With packet rate above 2100 Kpps: tx: 0 bps 0 pps rx: 1021 Mbps 2128 Kpps tx: 0 bps 0 pps rx: 1021 Mbps 2127 Kpps tx: 0 bps 0 pps rx: 1021 Mbps 2128 Kpps tx: 0 bps 0 pps rx: 1021 Mbps 2128 Kpps tx: 0 bps 0 pps rx: 1022 Mbps 2128 Kpps tx: 0 bps 0 pps rx: 1022 Mbps 2129 Kpps The major performance increase is explained by the fact that the most CPU consuming functions (page_pool_release_page, page_frag_free and free_unref_page) are no longer called on a per packet basis. The test was done by sending to the macchiatobin 64 byte ethernet frames with an invalid ethertype, so the packets are dropped early in the RX path. Signed-off-by: NMatteo Croce <mcroce@microsoft.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-