11751cc36SAlex Maftei (amaftei) // SPDX-License-Identifier: GPL-2.0-only 21751cc36SAlex Maftei (amaftei) /**************************************************************************** 31751cc36SAlex Maftei (amaftei) * Driver for Solarflare network controllers and boards 41751cc36SAlex Maftei (amaftei) * Copyright 2018 Solarflare Communications Inc. 51751cc36SAlex Maftei (amaftei) * 61751cc36SAlex Maftei (amaftei) * This program is free software; you can redistribute it and/or modify it 71751cc36SAlex Maftei (amaftei) * under the terms of the GNU General Public License version 2 as published 81751cc36SAlex Maftei (amaftei) * by the Free Software Foundation, incorporated herein by reference. 91751cc36SAlex Maftei (amaftei) */ 101751cc36SAlex Maftei (amaftei) 111751cc36SAlex Maftei (amaftei) #include "net_driver.h" 121751cc36SAlex Maftei (amaftei) #include <linux/module.h> 133d95b884SAlex Maftei (amaftei) #include <linux/iommu.h> 141751cc36SAlex Maftei (amaftei) #include "efx.h" 151751cc36SAlex Maftei (amaftei) #include "nic.h" 161751cc36SAlex Maftei (amaftei) #include "rx_common.h" 171751cc36SAlex Maftei (amaftei) 181751cc36SAlex Maftei (amaftei) /* This is the percentage fill level below which new RX descriptors 191751cc36SAlex Maftei (amaftei) * will be added to the RX descriptor ring. 201751cc36SAlex Maftei (amaftei) */ 211751cc36SAlex Maftei (amaftei) static unsigned int rx_refill_threshold; 221751cc36SAlex Maftei (amaftei) module_param(rx_refill_threshold, uint, 0444); 231751cc36SAlex Maftei (amaftei) MODULE_PARM_DESC(rx_refill_threshold, 241751cc36SAlex Maftei (amaftei) "RX descriptor ring refill threshold (%)"); 251751cc36SAlex Maftei (amaftei) 263d95b884SAlex Maftei (amaftei) /* Number of RX buffers to recycle pages for. When creating the RX page recycle 273d95b884SAlex Maftei (amaftei) * ring, this number is divided by the number of buffers per page to calculate 283d95b884SAlex Maftei (amaftei) * the number of pages to store in the RX page recycle ring. 293d95b884SAlex Maftei (amaftei) */ 303d95b884SAlex Maftei (amaftei) #define EFX_RECYCLE_RING_SIZE_IOMMU 4096 313d95b884SAlex Maftei (amaftei) #define EFX_RECYCLE_RING_SIZE_NOIOMMU (2 * EFX_RX_PREFERRED_BATCH) 323d95b884SAlex Maftei (amaftei) 331751cc36SAlex Maftei (amaftei) /* RX maximum head room required. 341751cc36SAlex Maftei (amaftei) * 351751cc36SAlex Maftei (amaftei) * This must be at least 1 to prevent overflow, plus one packet-worth 361751cc36SAlex Maftei (amaftei) * to allow pipelined receives. 371751cc36SAlex Maftei (amaftei) */ 381751cc36SAlex Maftei (amaftei) #define EFX_RXD_HEAD_ROOM (1 + EFX_RX_MAX_FRAGS) 391751cc36SAlex Maftei (amaftei) 403d95b884SAlex Maftei (amaftei) /* Check the RX page recycle ring for a page that can be reused. */ 413d95b884SAlex Maftei (amaftei) static struct page *efx_reuse_page(struct efx_rx_queue *rx_queue) 423d95b884SAlex Maftei (amaftei) { 433d95b884SAlex Maftei (amaftei) struct efx_nic *efx = rx_queue->efx; 443d95b884SAlex Maftei (amaftei) struct efx_rx_page_state *state; 453d95b884SAlex Maftei (amaftei) unsigned int index; 463d95b884SAlex Maftei (amaftei) struct page *page; 473d95b884SAlex Maftei (amaftei) 483d95b884SAlex Maftei (amaftei) index = rx_queue->page_remove & rx_queue->page_ptr_mask; 493d95b884SAlex Maftei (amaftei) page = rx_queue->page_ring[index]; 503d95b884SAlex Maftei (amaftei) if (page == NULL) 513d95b884SAlex Maftei (amaftei) return NULL; 523d95b884SAlex Maftei (amaftei) 533d95b884SAlex Maftei (amaftei) rx_queue->page_ring[index] = NULL; 543d95b884SAlex Maftei (amaftei) /* page_remove cannot exceed page_add. */ 553d95b884SAlex Maftei (amaftei) if (rx_queue->page_remove != rx_queue->page_add) 563d95b884SAlex Maftei (amaftei) ++rx_queue->page_remove; 573d95b884SAlex Maftei (amaftei) 583d95b884SAlex Maftei (amaftei) /* If page_count is 1 then we hold the only reference to this page. */ 593d95b884SAlex Maftei (amaftei) if (page_count(page) == 1) { 603d95b884SAlex Maftei (amaftei) ++rx_queue->page_recycle_count; 613d95b884SAlex Maftei (amaftei) return page; 623d95b884SAlex Maftei (amaftei) } else { 633d95b884SAlex Maftei (amaftei) state = page_address(page); 643d95b884SAlex Maftei (amaftei) dma_unmap_page(&efx->pci_dev->dev, state->dma_addr, 653d95b884SAlex Maftei (amaftei) PAGE_SIZE << efx->rx_buffer_order, 663d95b884SAlex Maftei (amaftei) DMA_FROM_DEVICE); 673d95b884SAlex Maftei (amaftei) put_page(page); 683d95b884SAlex Maftei (amaftei) ++rx_queue->page_recycle_failed; 693d95b884SAlex Maftei (amaftei) } 703d95b884SAlex Maftei (amaftei) 713d95b884SAlex Maftei (amaftei) return NULL; 723d95b884SAlex Maftei (amaftei) } 733d95b884SAlex Maftei (amaftei) 743d95b884SAlex Maftei (amaftei) /* Attempt to recycle the page if there is an RX recycle ring; the page can 753d95b884SAlex Maftei (amaftei) * only be added if this is the final RX buffer, to prevent pages being used in 763d95b884SAlex Maftei (amaftei) * the descriptor ring and appearing in the recycle ring simultaneously. 773d95b884SAlex Maftei (amaftei) */ 783d95b884SAlex Maftei (amaftei) static void efx_recycle_rx_page(struct efx_channel *channel, 793d95b884SAlex Maftei (amaftei) struct efx_rx_buffer *rx_buf) 803d95b884SAlex Maftei (amaftei) { 813d95b884SAlex Maftei (amaftei) struct efx_rx_queue *rx_queue = efx_channel_get_rx_queue(channel); 823d95b884SAlex Maftei (amaftei) struct efx_nic *efx = rx_queue->efx; 833d95b884SAlex Maftei (amaftei) struct page *page = rx_buf->page; 843d95b884SAlex Maftei (amaftei) unsigned int index; 853d95b884SAlex Maftei (amaftei) 863d95b884SAlex Maftei (amaftei) /* Only recycle the page after processing the final buffer. */ 873d95b884SAlex Maftei (amaftei) if (!(rx_buf->flags & EFX_RX_BUF_LAST_IN_PAGE)) 883d95b884SAlex Maftei (amaftei) return; 893d95b884SAlex Maftei (amaftei) 903d95b884SAlex Maftei (amaftei) index = rx_queue->page_add & rx_queue->page_ptr_mask; 913d95b884SAlex Maftei (amaftei) if (rx_queue->page_ring[index] == NULL) { 923d95b884SAlex Maftei (amaftei) unsigned int read_index = rx_queue->page_remove & 933d95b884SAlex Maftei (amaftei) rx_queue->page_ptr_mask; 943d95b884SAlex Maftei (amaftei) 953d95b884SAlex Maftei (amaftei) /* The next slot in the recycle ring is available, but 963d95b884SAlex Maftei (amaftei) * increment page_remove if the read pointer currently 973d95b884SAlex Maftei (amaftei) * points here. 983d95b884SAlex Maftei (amaftei) */ 993d95b884SAlex Maftei (amaftei) if (read_index == index) 1003d95b884SAlex Maftei (amaftei) ++rx_queue->page_remove; 1013d95b884SAlex Maftei (amaftei) rx_queue->page_ring[index] = page; 1023d95b884SAlex Maftei (amaftei) ++rx_queue->page_add; 1033d95b884SAlex Maftei (amaftei) return; 1043d95b884SAlex Maftei (amaftei) } 1053d95b884SAlex Maftei (amaftei) ++rx_queue->page_recycle_full; 1063d95b884SAlex Maftei (amaftei) efx_unmap_rx_buffer(efx, rx_buf); 1073d95b884SAlex Maftei (amaftei) put_page(rx_buf->page); 1083d95b884SAlex Maftei (amaftei) } 1093d95b884SAlex Maftei (amaftei) 1103d95b884SAlex Maftei (amaftei) /* Recycle the pages that are used by buffers that have just been received. */ 1113d95b884SAlex Maftei (amaftei) void efx_recycle_rx_pages(struct efx_channel *channel, 1123d95b884SAlex Maftei (amaftei) struct efx_rx_buffer *rx_buf, 1133d95b884SAlex Maftei (amaftei) unsigned int n_frags) 1143d95b884SAlex Maftei (amaftei) { 1153d95b884SAlex Maftei (amaftei) struct efx_rx_queue *rx_queue = efx_channel_get_rx_queue(channel); 1163d95b884SAlex Maftei (amaftei) 1173d95b884SAlex Maftei (amaftei) do { 1183d95b884SAlex Maftei (amaftei) efx_recycle_rx_page(channel, rx_buf); 1193d95b884SAlex Maftei (amaftei) rx_buf = efx_rx_buf_next(rx_queue, rx_buf); 1203d95b884SAlex Maftei (amaftei) } while (--n_frags); 1213d95b884SAlex Maftei (amaftei) } 1223d95b884SAlex Maftei (amaftei) 1233d95b884SAlex Maftei (amaftei) void efx_discard_rx_packet(struct efx_channel *channel, 1243d95b884SAlex Maftei (amaftei) struct efx_rx_buffer *rx_buf, 1253d95b884SAlex Maftei (amaftei) unsigned int n_frags) 1263d95b884SAlex Maftei (amaftei) { 1273d95b884SAlex Maftei (amaftei) struct efx_rx_queue *rx_queue = efx_channel_get_rx_queue(channel); 1283d95b884SAlex Maftei (amaftei) 1293d95b884SAlex Maftei (amaftei) efx_recycle_rx_pages(channel, rx_buf, n_frags); 1303d95b884SAlex Maftei (amaftei) 1313d95b884SAlex Maftei (amaftei) efx_free_rx_buffers(rx_queue, rx_buf, n_frags); 1323d95b884SAlex Maftei (amaftei) } 1333d95b884SAlex Maftei (amaftei) 1343d95b884SAlex Maftei (amaftei) static void efx_init_rx_recycle_ring(struct efx_rx_queue *rx_queue) 1353d95b884SAlex Maftei (amaftei) { 1363d95b884SAlex Maftei (amaftei) unsigned int bufs_in_recycle_ring, page_ring_size; 1373d95b884SAlex Maftei (amaftei) struct efx_nic *efx = rx_queue->efx; 1383d95b884SAlex Maftei (amaftei) 1393d95b884SAlex Maftei (amaftei) /* Set the RX recycle ring size */ 1403d95b884SAlex Maftei (amaftei) #ifdef CONFIG_PPC64 1413d95b884SAlex Maftei (amaftei) bufs_in_recycle_ring = EFX_RECYCLE_RING_SIZE_IOMMU; 1423d95b884SAlex Maftei (amaftei) #else 1433d95b884SAlex Maftei (amaftei) if (iommu_present(&pci_bus_type)) 1443d95b884SAlex Maftei (amaftei) bufs_in_recycle_ring = EFX_RECYCLE_RING_SIZE_IOMMU; 1453d95b884SAlex Maftei (amaftei) else 1463d95b884SAlex Maftei (amaftei) bufs_in_recycle_ring = EFX_RECYCLE_RING_SIZE_NOIOMMU; 1473d95b884SAlex Maftei (amaftei) #endif /* CONFIG_PPC64 */ 1483d95b884SAlex Maftei (amaftei) 1493d95b884SAlex Maftei (amaftei) page_ring_size = roundup_pow_of_two(bufs_in_recycle_ring / 1503d95b884SAlex Maftei (amaftei) efx->rx_bufs_per_page); 1513d95b884SAlex Maftei (amaftei) rx_queue->page_ring = kcalloc(page_ring_size, 1523d95b884SAlex Maftei (amaftei) sizeof(*rx_queue->page_ring), GFP_KERNEL); 153*bdf1b5c3SJiasheng Jiang if (!rx_queue->page_ring) 154*bdf1b5c3SJiasheng Jiang rx_queue->page_ptr_mask = 0; 155*bdf1b5c3SJiasheng Jiang else 1563d95b884SAlex Maftei (amaftei) rx_queue->page_ptr_mask = page_ring_size - 1; 1573d95b884SAlex Maftei (amaftei) } 1583d95b884SAlex Maftei (amaftei) 1593d95b884SAlex Maftei (amaftei) static void efx_fini_rx_recycle_ring(struct efx_rx_queue *rx_queue) 1603d95b884SAlex Maftei (amaftei) { 1613d95b884SAlex Maftei (amaftei) struct efx_nic *efx = rx_queue->efx; 1623d95b884SAlex Maftei (amaftei) int i; 1633d95b884SAlex Maftei (amaftei) 1643d95b884SAlex Maftei (amaftei) /* Unmap and release the pages in the recycle ring. Remove the ring. */ 1653d95b884SAlex Maftei (amaftei) for (i = 0; i <= rx_queue->page_ptr_mask; i++) { 1663d95b884SAlex Maftei (amaftei) struct page *page = rx_queue->page_ring[i]; 1673d95b884SAlex Maftei (amaftei) struct efx_rx_page_state *state; 1683d95b884SAlex Maftei (amaftei) 1693d95b884SAlex Maftei (amaftei) if (page == NULL) 1703d95b884SAlex Maftei (amaftei) continue; 1713d95b884SAlex Maftei (amaftei) 1723d95b884SAlex Maftei (amaftei) state = page_address(page); 1733d95b884SAlex Maftei (amaftei) dma_unmap_page(&efx->pci_dev->dev, state->dma_addr, 1743d95b884SAlex Maftei (amaftei) PAGE_SIZE << efx->rx_buffer_order, 1753d95b884SAlex Maftei (amaftei) DMA_FROM_DEVICE); 1763d95b884SAlex Maftei (amaftei) put_page(page); 1773d95b884SAlex Maftei (amaftei) } 1783d95b884SAlex Maftei (amaftei) kfree(rx_queue->page_ring); 1793d95b884SAlex Maftei (amaftei) rx_queue->page_ring = NULL; 1803d95b884SAlex Maftei (amaftei) } 1813d95b884SAlex Maftei (amaftei) 1821751cc36SAlex Maftei (amaftei) static void efx_fini_rx_buffer(struct efx_rx_queue *rx_queue, 1831751cc36SAlex Maftei (amaftei) struct efx_rx_buffer *rx_buf) 1841751cc36SAlex Maftei (amaftei) { 1851751cc36SAlex Maftei (amaftei) /* Release the page reference we hold for the buffer. */ 1861751cc36SAlex Maftei (amaftei) if (rx_buf->page) 1871751cc36SAlex Maftei (amaftei) put_page(rx_buf->page); 1881751cc36SAlex Maftei (amaftei) 1891751cc36SAlex Maftei (amaftei) /* If this is the last buffer in a page, unmap and free it. */ 1901751cc36SAlex Maftei (amaftei) if (rx_buf->flags & EFX_RX_BUF_LAST_IN_PAGE) { 1911751cc36SAlex Maftei (amaftei) efx_unmap_rx_buffer(rx_queue->efx, rx_buf); 1921751cc36SAlex Maftei (amaftei) efx_free_rx_buffers(rx_queue, rx_buf, 1); 1931751cc36SAlex Maftei (amaftei) } 1941751cc36SAlex Maftei (amaftei) rx_buf->page = NULL; 1951751cc36SAlex Maftei (amaftei) } 1961751cc36SAlex Maftei (amaftei) 1971751cc36SAlex Maftei (amaftei) int efx_probe_rx_queue(struct efx_rx_queue *rx_queue) 1981751cc36SAlex Maftei (amaftei) { 1991751cc36SAlex Maftei (amaftei) struct efx_nic *efx = rx_queue->efx; 2001751cc36SAlex Maftei (amaftei) unsigned int entries; 2011751cc36SAlex Maftei (amaftei) int rc; 2021751cc36SAlex Maftei (amaftei) 2031751cc36SAlex Maftei (amaftei) /* Create the smallest power-of-two aligned ring */ 2041751cc36SAlex Maftei (amaftei) entries = max(roundup_pow_of_two(efx->rxq_entries), EFX_MIN_DMAQ_SIZE); 2051751cc36SAlex Maftei (amaftei) EFX_WARN_ON_PARANOID(entries > EFX_MAX_DMAQ_SIZE); 2061751cc36SAlex Maftei (amaftei) rx_queue->ptr_mask = entries - 1; 2071751cc36SAlex Maftei (amaftei) 2081751cc36SAlex Maftei (amaftei) netif_dbg(efx, probe, efx->net_dev, 2091751cc36SAlex Maftei (amaftei) "creating RX queue %d size %#x mask %#x\n", 2101751cc36SAlex Maftei (amaftei) efx_rx_queue_index(rx_queue), efx->rxq_entries, 2111751cc36SAlex Maftei (amaftei) rx_queue->ptr_mask); 2121751cc36SAlex Maftei (amaftei) 2131751cc36SAlex Maftei (amaftei) /* Allocate RX buffers */ 2141751cc36SAlex Maftei (amaftei) rx_queue->buffer = kcalloc(entries, sizeof(*rx_queue->buffer), 2151751cc36SAlex Maftei (amaftei) GFP_KERNEL); 2161751cc36SAlex Maftei (amaftei) if (!rx_queue->buffer) 2171751cc36SAlex Maftei (amaftei) return -ENOMEM; 2181751cc36SAlex Maftei (amaftei) 2191751cc36SAlex Maftei (amaftei) rc = efx_nic_probe_rx(rx_queue); 2201751cc36SAlex Maftei (amaftei) if (rc) { 2211751cc36SAlex Maftei (amaftei) kfree(rx_queue->buffer); 2221751cc36SAlex Maftei (amaftei) rx_queue->buffer = NULL; 2231751cc36SAlex Maftei (amaftei) } 2241751cc36SAlex Maftei (amaftei) 2251751cc36SAlex Maftei (amaftei) return rc; 2261751cc36SAlex Maftei (amaftei) } 2271751cc36SAlex Maftei (amaftei) 2281751cc36SAlex Maftei (amaftei) void efx_init_rx_queue(struct efx_rx_queue *rx_queue) 2291751cc36SAlex Maftei (amaftei) { 2301751cc36SAlex Maftei (amaftei) unsigned int max_fill, trigger, max_trigger; 2311751cc36SAlex Maftei (amaftei) struct efx_nic *efx = rx_queue->efx; 2321751cc36SAlex Maftei (amaftei) int rc = 0; 2331751cc36SAlex Maftei (amaftei) 2341751cc36SAlex Maftei (amaftei) netif_dbg(rx_queue->efx, drv, rx_queue->efx->net_dev, 2351751cc36SAlex Maftei (amaftei) "initialising RX queue %d\n", efx_rx_queue_index(rx_queue)); 2361751cc36SAlex Maftei (amaftei) 2371751cc36SAlex Maftei (amaftei) /* Initialise ptr fields */ 2381751cc36SAlex Maftei (amaftei) rx_queue->added_count = 0; 2391751cc36SAlex Maftei (amaftei) rx_queue->notified_count = 0; 2401751cc36SAlex Maftei (amaftei) rx_queue->removed_count = 0; 2411751cc36SAlex Maftei (amaftei) rx_queue->min_fill = -1U; 2421751cc36SAlex Maftei (amaftei) efx_init_rx_recycle_ring(rx_queue); 2431751cc36SAlex Maftei (amaftei) 2441751cc36SAlex Maftei (amaftei) rx_queue->page_remove = 0; 2451751cc36SAlex Maftei (amaftei) rx_queue->page_add = rx_queue->page_ptr_mask + 1; 2461751cc36SAlex Maftei (amaftei) rx_queue->page_recycle_count = 0; 2471751cc36SAlex Maftei (amaftei) rx_queue->page_recycle_failed = 0; 2481751cc36SAlex Maftei (amaftei) rx_queue->page_recycle_full = 0; 2491751cc36SAlex Maftei (amaftei) 2501751cc36SAlex Maftei (amaftei) /* Initialise limit fields */ 2511751cc36SAlex Maftei (amaftei) max_fill = efx->rxq_entries - EFX_RXD_HEAD_ROOM; 2521751cc36SAlex Maftei (amaftei) max_trigger = 2531751cc36SAlex Maftei (amaftei) max_fill - efx->rx_pages_per_batch * efx->rx_bufs_per_page; 2541751cc36SAlex Maftei (amaftei) if (rx_refill_threshold != 0) { 2551751cc36SAlex Maftei (amaftei) trigger = max_fill * min(rx_refill_threshold, 100U) / 100U; 2561751cc36SAlex Maftei (amaftei) if (trigger > max_trigger) 2571751cc36SAlex Maftei (amaftei) trigger = max_trigger; 2581751cc36SAlex Maftei (amaftei) } else { 2591751cc36SAlex Maftei (amaftei) trigger = max_trigger; 2601751cc36SAlex Maftei (amaftei) } 2611751cc36SAlex Maftei (amaftei) 2621751cc36SAlex Maftei (amaftei) rx_queue->max_fill = max_fill; 2631751cc36SAlex Maftei (amaftei) rx_queue->fast_fill_trigger = trigger; 2641751cc36SAlex Maftei (amaftei) rx_queue->refill_enabled = true; 2651751cc36SAlex Maftei (amaftei) 2661751cc36SAlex Maftei (amaftei) /* Initialise XDP queue information */ 2671751cc36SAlex Maftei (amaftei) rc = xdp_rxq_info_reg(&rx_queue->xdp_rxq_info, efx->net_dev, 268b02e5a0eSBjörn Töpel rx_queue->core_index, 0); 2691751cc36SAlex Maftei (amaftei) 2701751cc36SAlex Maftei (amaftei) if (rc) { 2711751cc36SAlex Maftei (amaftei) netif_err(efx, rx_err, efx->net_dev, 2721751cc36SAlex Maftei (amaftei) "Failure to initialise XDP queue information rc=%d\n", 2731751cc36SAlex Maftei (amaftei) rc); 2741751cc36SAlex Maftei (amaftei) efx->xdp_rxq_info_failed = true; 2751751cc36SAlex Maftei (amaftei) } else { 2761751cc36SAlex Maftei (amaftei) rx_queue->xdp_rxq_info_valid = true; 2771751cc36SAlex Maftei (amaftei) } 2781751cc36SAlex Maftei (amaftei) 2791751cc36SAlex Maftei (amaftei) /* Set up RX descriptor ring */ 2801751cc36SAlex Maftei (amaftei) efx_nic_init_rx(rx_queue); 2811751cc36SAlex Maftei (amaftei) } 2821751cc36SAlex Maftei (amaftei) 2831751cc36SAlex Maftei (amaftei) void efx_fini_rx_queue(struct efx_rx_queue *rx_queue) 2841751cc36SAlex Maftei (amaftei) { 2851751cc36SAlex Maftei (amaftei) struct efx_rx_buffer *rx_buf; 2861751cc36SAlex Maftei (amaftei) int i; 2871751cc36SAlex Maftei (amaftei) 2881751cc36SAlex Maftei (amaftei) netif_dbg(rx_queue->efx, drv, rx_queue->efx->net_dev, 2891751cc36SAlex Maftei (amaftei) "shutting down RX queue %d\n", efx_rx_queue_index(rx_queue)); 2901751cc36SAlex Maftei (amaftei) 2911751cc36SAlex Maftei (amaftei) del_timer_sync(&rx_queue->slow_fill); 2921751cc36SAlex Maftei (amaftei) 2931751cc36SAlex Maftei (amaftei) /* Release RX buffers from the current read ptr to the write ptr */ 2941751cc36SAlex Maftei (amaftei) if (rx_queue->buffer) { 2951751cc36SAlex Maftei (amaftei) for (i = rx_queue->removed_count; i < rx_queue->added_count; 2961751cc36SAlex Maftei (amaftei) i++) { 2971751cc36SAlex Maftei (amaftei) unsigned int index = i & rx_queue->ptr_mask; 2981751cc36SAlex Maftei (amaftei) 2991751cc36SAlex Maftei (amaftei) rx_buf = efx_rx_buffer(rx_queue, index); 3001751cc36SAlex Maftei (amaftei) efx_fini_rx_buffer(rx_queue, rx_buf); 3011751cc36SAlex Maftei (amaftei) } 3021751cc36SAlex Maftei (amaftei) } 3031751cc36SAlex Maftei (amaftei) 3043d95b884SAlex Maftei (amaftei) efx_fini_rx_recycle_ring(rx_queue); 3051751cc36SAlex Maftei (amaftei) 3061751cc36SAlex Maftei (amaftei) if (rx_queue->xdp_rxq_info_valid) 3071751cc36SAlex Maftei (amaftei) xdp_rxq_info_unreg(&rx_queue->xdp_rxq_info); 3081751cc36SAlex Maftei (amaftei) 3091751cc36SAlex Maftei (amaftei) rx_queue->xdp_rxq_info_valid = false; 3101751cc36SAlex Maftei (amaftei) } 3111751cc36SAlex Maftei (amaftei) 3121751cc36SAlex Maftei (amaftei) void efx_remove_rx_queue(struct efx_rx_queue *rx_queue) 3131751cc36SAlex Maftei (amaftei) { 3141751cc36SAlex Maftei (amaftei) netif_dbg(rx_queue->efx, drv, rx_queue->efx->net_dev, 3151751cc36SAlex Maftei (amaftei) "destroying RX queue %d\n", efx_rx_queue_index(rx_queue)); 3161751cc36SAlex Maftei (amaftei) 3171751cc36SAlex Maftei (amaftei) efx_nic_remove_rx(rx_queue); 3181751cc36SAlex Maftei (amaftei) 3191751cc36SAlex Maftei (amaftei) kfree(rx_queue->buffer); 3201751cc36SAlex Maftei (amaftei) rx_queue->buffer = NULL; 3211751cc36SAlex Maftei (amaftei) } 3221751cc36SAlex Maftei (amaftei) 3231751cc36SAlex Maftei (amaftei) /* Unmap a DMA-mapped page. This function is only called for the final RX 3241751cc36SAlex Maftei (amaftei) * buffer in a page. 3251751cc36SAlex Maftei (amaftei) */ 3261751cc36SAlex Maftei (amaftei) void efx_unmap_rx_buffer(struct efx_nic *efx, 3271751cc36SAlex Maftei (amaftei) struct efx_rx_buffer *rx_buf) 3281751cc36SAlex Maftei (amaftei) { 3291751cc36SAlex Maftei (amaftei) struct page *page = rx_buf->page; 3301751cc36SAlex Maftei (amaftei) 3311751cc36SAlex Maftei (amaftei) if (page) { 3321751cc36SAlex Maftei (amaftei) struct efx_rx_page_state *state = page_address(page); 3331751cc36SAlex Maftei (amaftei) 3341751cc36SAlex Maftei (amaftei) dma_unmap_page(&efx->pci_dev->dev, 3351751cc36SAlex Maftei (amaftei) state->dma_addr, 3361751cc36SAlex Maftei (amaftei) PAGE_SIZE << efx->rx_buffer_order, 3371751cc36SAlex Maftei (amaftei) DMA_FROM_DEVICE); 3381751cc36SAlex Maftei (amaftei) } 3391751cc36SAlex Maftei (amaftei) } 3401751cc36SAlex Maftei (amaftei) 3411751cc36SAlex Maftei (amaftei) void efx_free_rx_buffers(struct efx_rx_queue *rx_queue, 3421751cc36SAlex Maftei (amaftei) struct efx_rx_buffer *rx_buf, 3431751cc36SAlex Maftei (amaftei) unsigned int num_bufs) 3441751cc36SAlex Maftei (amaftei) { 3451751cc36SAlex Maftei (amaftei) do { 3461751cc36SAlex Maftei (amaftei) if (rx_buf->page) { 3471751cc36SAlex Maftei (amaftei) put_page(rx_buf->page); 3481751cc36SAlex Maftei (amaftei) rx_buf->page = NULL; 3491751cc36SAlex Maftei (amaftei) } 3501751cc36SAlex Maftei (amaftei) rx_buf = efx_rx_buf_next(rx_queue, rx_buf); 3511751cc36SAlex Maftei (amaftei) } while (--num_bufs); 3521751cc36SAlex Maftei (amaftei) } 3531751cc36SAlex Maftei (amaftei) 3541751cc36SAlex Maftei (amaftei) void efx_rx_slow_fill(struct timer_list *t) 3551751cc36SAlex Maftei (amaftei) { 3561751cc36SAlex Maftei (amaftei) struct efx_rx_queue *rx_queue = from_timer(rx_queue, t, slow_fill); 3571751cc36SAlex Maftei (amaftei) 3581751cc36SAlex Maftei (amaftei) /* Post an event to cause NAPI to run and refill the queue */ 3591751cc36SAlex Maftei (amaftei) efx_nic_generate_fill_event(rx_queue); 3601751cc36SAlex Maftei (amaftei) ++rx_queue->slow_fill_count; 3611751cc36SAlex Maftei (amaftei) } 3621751cc36SAlex Maftei (amaftei) 3631751cc36SAlex Maftei (amaftei) void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue) 3641751cc36SAlex Maftei (amaftei) { 3651751cc36SAlex Maftei (amaftei) mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(10)); 3661751cc36SAlex Maftei (amaftei) } 3671751cc36SAlex Maftei (amaftei) 3681751cc36SAlex Maftei (amaftei) /* efx_init_rx_buffers - create EFX_RX_BATCH page-based RX buffers 3691751cc36SAlex Maftei (amaftei) * 3701751cc36SAlex Maftei (amaftei) * @rx_queue: Efx RX queue 3711751cc36SAlex Maftei (amaftei) * 3721751cc36SAlex Maftei (amaftei) * This allocates a batch of pages, maps them for DMA, and populates 3731751cc36SAlex Maftei (amaftei) * struct efx_rx_buffers for each one. Return a negative error code or 3741751cc36SAlex Maftei (amaftei) * 0 on success. If a single page can be used for multiple buffers, 3751751cc36SAlex Maftei (amaftei) * then the page will either be inserted fully, or not at all. 3761751cc36SAlex Maftei (amaftei) */ 3771751cc36SAlex Maftei (amaftei) static int efx_init_rx_buffers(struct efx_rx_queue *rx_queue, bool atomic) 3781751cc36SAlex Maftei (amaftei) { 3791751cc36SAlex Maftei (amaftei) unsigned int page_offset, index, count; 3801751cc36SAlex Maftei (amaftei) struct efx_nic *efx = rx_queue->efx; 3811751cc36SAlex Maftei (amaftei) struct efx_rx_page_state *state; 3821751cc36SAlex Maftei (amaftei) struct efx_rx_buffer *rx_buf; 3831751cc36SAlex Maftei (amaftei) dma_addr_t dma_addr; 3841751cc36SAlex Maftei (amaftei) struct page *page; 3851751cc36SAlex Maftei (amaftei) 3861751cc36SAlex Maftei (amaftei) count = 0; 3871751cc36SAlex Maftei (amaftei) do { 3881751cc36SAlex Maftei (amaftei) page = efx_reuse_page(rx_queue); 3891751cc36SAlex Maftei (amaftei) if (page == NULL) { 3901751cc36SAlex Maftei (amaftei) page = alloc_pages(__GFP_COMP | 3911751cc36SAlex Maftei (amaftei) (atomic ? GFP_ATOMIC : GFP_KERNEL), 3921751cc36SAlex Maftei (amaftei) efx->rx_buffer_order); 3931751cc36SAlex Maftei (amaftei) if (unlikely(page == NULL)) 3941751cc36SAlex Maftei (amaftei) return -ENOMEM; 3951751cc36SAlex Maftei (amaftei) dma_addr = 3961751cc36SAlex Maftei (amaftei) dma_map_page(&efx->pci_dev->dev, page, 0, 3971751cc36SAlex Maftei (amaftei) PAGE_SIZE << efx->rx_buffer_order, 3981751cc36SAlex Maftei (amaftei) DMA_FROM_DEVICE); 3991751cc36SAlex Maftei (amaftei) if (unlikely(dma_mapping_error(&efx->pci_dev->dev, 4001751cc36SAlex Maftei (amaftei) dma_addr))) { 4011751cc36SAlex Maftei (amaftei) __free_pages(page, efx->rx_buffer_order); 4021751cc36SAlex Maftei (amaftei) return -EIO; 4031751cc36SAlex Maftei (amaftei) } 4041751cc36SAlex Maftei (amaftei) state = page_address(page); 4051751cc36SAlex Maftei (amaftei) state->dma_addr = dma_addr; 4061751cc36SAlex Maftei (amaftei) } else { 4071751cc36SAlex Maftei (amaftei) state = page_address(page); 4081751cc36SAlex Maftei (amaftei) dma_addr = state->dma_addr; 4091751cc36SAlex Maftei (amaftei) } 4101751cc36SAlex Maftei (amaftei) 4111751cc36SAlex Maftei (amaftei) dma_addr += sizeof(struct efx_rx_page_state); 4121751cc36SAlex Maftei (amaftei) page_offset = sizeof(struct efx_rx_page_state); 4131751cc36SAlex Maftei (amaftei) 4141751cc36SAlex Maftei (amaftei) do { 4151751cc36SAlex Maftei (amaftei) index = rx_queue->added_count & rx_queue->ptr_mask; 4161751cc36SAlex Maftei (amaftei) rx_buf = efx_rx_buffer(rx_queue, index); 4171751cc36SAlex Maftei (amaftei) rx_buf->dma_addr = dma_addr + efx->rx_ip_align + 41886e85bf6SJesper Dangaard Brouer EFX_XDP_HEADROOM; 4191751cc36SAlex Maftei (amaftei) rx_buf->page = page; 4201751cc36SAlex Maftei (amaftei) rx_buf->page_offset = page_offset + efx->rx_ip_align + 42186e85bf6SJesper Dangaard Brouer EFX_XDP_HEADROOM; 4221751cc36SAlex Maftei (amaftei) rx_buf->len = efx->rx_dma_len; 4231751cc36SAlex Maftei (amaftei) rx_buf->flags = 0; 4241751cc36SAlex Maftei (amaftei) ++rx_queue->added_count; 4251751cc36SAlex Maftei (amaftei) get_page(page); 4261751cc36SAlex Maftei (amaftei) dma_addr += efx->rx_page_buf_step; 4271751cc36SAlex Maftei (amaftei) page_offset += efx->rx_page_buf_step; 4281751cc36SAlex Maftei (amaftei) } while (page_offset + efx->rx_page_buf_step <= PAGE_SIZE); 4291751cc36SAlex Maftei (amaftei) 4301751cc36SAlex Maftei (amaftei) rx_buf->flags = EFX_RX_BUF_LAST_IN_PAGE; 4311751cc36SAlex Maftei (amaftei) } while (++count < efx->rx_pages_per_batch); 4321751cc36SAlex Maftei (amaftei) 4331751cc36SAlex Maftei (amaftei) return 0; 4341751cc36SAlex Maftei (amaftei) } 4351751cc36SAlex Maftei (amaftei) 4361751cc36SAlex Maftei (amaftei) void efx_rx_config_page_split(struct efx_nic *efx) 4371751cc36SAlex Maftei (amaftei) { 4381751cc36SAlex Maftei (amaftei) efx->rx_page_buf_step = ALIGN(efx->rx_dma_len + efx->rx_ip_align + 43986e85bf6SJesper Dangaard Brouer EFX_XDP_HEADROOM + EFX_XDP_TAILROOM, 4401751cc36SAlex Maftei (amaftei) EFX_RX_BUF_ALIGNMENT); 4411751cc36SAlex Maftei (amaftei) efx->rx_bufs_per_page = efx->rx_buffer_order ? 1 : 4421751cc36SAlex Maftei (amaftei) ((PAGE_SIZE - sizeof(struct efx_rx_page_state)) / 4431751cc36SAlex Maftei (amaftei) efx->rx_page_buf_step); 4441751cc36SAlex Maftei (amaftei) efx->rx_buffer_truesize = (PAGE_SIZE << efx->rx_buffer_order) / 4451751cc36SAlex Maftei (amaftei) efx->rx_bufs_per_page; 4461751cc36SAlex Maftei (amaftei) efx->rx_pages_per_batch = DIV_ROUND_UP(EFX_RX_PREFERRED_BATCH, 4471751cc36SAlex Maftei (amaftei) efx->rx_bufs_per_page); 4481751cc36SAlex Maftei (amaftei) } 4491751cc36SAlex Maftei (amaftei) 4501751cc36SAlex Maftei (amaftei) /* efx_fast_push_rx_descriptors - push new RX descriptors quickly 4511751cc36SAlex Maftei (amaftei) * @rx_queue: RX descriptor queue 4521751cc36SAlex Maftei (amaftei) * 4531751cc36SAlex Maftei (amaftei) * This will aim to fill the RX descriptor queue up to 4541751cc36SAlex Maftei (amaftei) * @rx_queue->@max_fill. If there is insufficient atomic 4551751cc36SAlex Maftei (amaftei) * memory to do so, a slow fill will be scheduled. 4561751cc36SAlex Maftei (amaftei) * 4571751cc36SAlex Maftei (amaftei) * The caller must provide serialisation (none is used here). In practise, 4581751cc36SAlex Maftei (amaftei) * this means this function must run from the NAPI handler, or be called 4591751cc36SAlex Maftei (amaftei) * when NAPI is disabled. 4601751cc36SAlex Maftei (amaftei) */ 4611751cc36SAlex Maftei (amaftei) void efx_fast_push_rx_descriptors(struct efx_rx_queue *rx_queue, bool atomic) 4621751cc36SAlex Maftei (amaftei) { 4631751cc36SAlex Maftei (amaftei) struct efx_nic *efx = rx_queue->efx; 4641751cc36SAlex Maftei (amaftei) unsigned int fill_level, batch_size; 4651751cc36SAlex Maftei (amaftei) int space, rc = 0; 4661751cc36SAlex Maftei (amaftei) 4671751cc36SAlex Maftei (amaftei) if (!rx_queue->refill_enabled) 4681751cc36SAlex Maftei (amaftei) return; 4691751cc36SAlex Maftei (amaftei) 4701751cc36SAlex Maftei (amaftei) /* Calculate current fill level, and exit if we don't need to fill */ 4711751cc36SAlex Maftei (amaftei) fill_level = (rx_queue->added_count - rx_queue->removed_count); 4721751cc36SAlex Maftei (amaftei) EFX_WARN_ON_ONCE_PARANOID(fill_level > rx_queue->efx->rxq_entries); 4731751cc36SAlex Maftei (amaftei) if (fill_level >= rx_queue->fast_fill_trigger) 4741751cc36SAlex Maftei (amaftei) goto out; 4751751cc36SAlex Maftei (amaftei) 4761751cc36SAlex Maftei (amaftei) /* Record minimum fill level */ 4771751cc36SAlex Maftei (amaftei) if (unlikely(fill_level < rx_queue->min_fill)) { 4781751cc36SAlex Maftei (amaftei) if (fill_level) 4791751cc36SAlex Maftei (amaftei) rx_queue->min_fill = fill_level; 4801751cc36SAlex Maftei (amaftei) } 4811751cc36SAlex Maftei (amaftei) 4821751cc36SAlex Maftei (amaftei) batch_size = efx->rx_pages_per_batch * efx->rx_bufs_per_page; 4831751cc36SAlex Maftei (amaftei) space = rx_queue->max_fill - fill_level; 4841751cc36SAlex Maftei (amaftei) EFX_WARN_ON_ONCE_PARANOID(space < batch_size); 4851751cc36SAlex Maftei (amaftei) 4861751cc36SAlex Maftei (amaftei) netif_vdbg(rx_queue->efx, rx_status, rx_queue->efx->net_dev, 4871751cc36SAlex Maftei (amaftei) "RX queue %d fast-filling descriptor ring from" 4881751cc36SAlex Maftei (amaftei) " level %d to level %d\n", 4891751cc36SAlex Maftei (amaftei) efx_rx_queue_index(rx_queue), fill_level, 4901751cc36SAlex Maftei (amaftei) rx_queue->max_fill); 4911751cc36SAlex Maftei (amaftei) 4921751cc36SAlex Maftei (amaftei) do { 4931751cc36SAlex Maftei (amaftei) rc = efx_init_rx_buffers(rx_queue, atomic); 4941751cc36SAlex Maftei (amaftei) if (unlikely(rc)) { 4951751cc36SAlex Maftei (amaftei) /* Ensure that we don't leave the rx queue empty */ 4961751cc36SAlex Maftei (amaftei) efx_schedule_slow_fill(rx_queue); 4971751cc36SAlex Maftei (amaftei) goto out; 4981751cc36SAlex Maftei (amaftei) } 4991751cc36SAlex Maftei (amaftei) } while ((space -= batch_size) >= batch_size); 5001751cc36SAlex Maftei (amaftei) 5011751cc36SAlex Maftei (amaftei) netif_vdbg(rx_queue->efx, rx_status, rx_queue->efx->net_dev, 5021751cc36SAlex Maftei (amaftei) "RX queue %d fast-filled descriptor ring " 5031751cc36SAlex Maftei (amaftei) "to level %d\n", efx_rx_queue_index(rx_queue), 5041751cc36SAlex Maftei (amaftei) rx_queue->added_count - rx_queue->removed_count); 5051751cc36SAlex Maftei (amaftei) 5061751cc36SAlex Maftei (amaftei) out: 5071751cc36SAlex Maftei (amaftei) if (rx_queue->notified_count != rx_queue->added_count) 5081751cc36SAlex Maftei (amaftei) efx_nic_notify_rx_desc(rx_queue); 5091751cc36SAlex Maftei (amaftei) } 5103d95b884SAlex Maftei (amaftei) 5113d95b884SAlex Maftei (amaftei) /* Pass a received packet up through GRO. GRO can handle pages 5123d95b884SAlex Maftei (amaftei) * regardless of checksum state and skbs with a good checksum. 5133d95b884SAlex Maftei (amaftei) */ 5143d95b884SAlex Maftei (amaftei) void 5153d95b884SAlex Maftei (amaftei) efx_rx_packet_gro(struct efx_channel *channel, struct efx_rx_buffer *rx_buf, 5164d9c0a2dSEdward Cree unsigned int n_frags, u8 *eh, __wsum csum) 5173d95b884SAlex Maftei (amaftei) { 5183d95b884SAlex Maftei (amaftei) struct napi_struct *napi = &channel->napi_str; 5193d95b884SAlex Maftei (amaftei) struct efx_nic *efx = channel->efx; 5203d95b884SAlex Maftei (amaftei) struct sk_buff *skb; 5213d95b884SAlex Maftei (amaftei) 5223d95b884SAlex Maftei (amaftei) skb = napi_get_frags(napi); 5233d95b884SAlex Maftei (amaftei) if (unlikely(!skb)) { 5243d95b884SAlex Maftei (amaftei) struct efx_rx_queue *rx_queue; 5253d95b884SAlex Maftei (amaftei) 5263d95b884SAlex Maftei (amaftei) rx_queue = efx_channel_get_rx_queue(channel); 5273d95b884SAlex Maftei (amaftei) efx_free_rx_buffers(rx_queue, rx_buf, n_frags); 5283d95b884SAlex Maftei (amaftei) return; 5293d95b884SAlex Maftei (amaftei) } 5303d95b884SAlex Maftei (amaftei) 53106888543SEdward Cree if (efx->net_dev->features & NETIF_F_RXHASH && 53206888543SEdward Cree efx_rx_buf_hash_valid(efx, eh)) 5333d95b884SAlex Maftei (amaftei) skb_set_hash(skb, efx_rx_buf_hash(efx, eh), 5343d95b884SAlex Maftei (amaftei) PKT_HASH_TYPE_L3); 5354d9c0a2dSEdward Cree if (csum) { 5364d9c0a2dSEdward Cree skb->csum = csum; 5374d9c0a2dSEdward Cree skb->ip_summed = CHECKSUM_COMPLETE; 5384d9c0a2dSEdward Cree } else { 5393d95b884SAlex Maftei (amaftei) skb->ip_summed = ((rx_buf->flags & EFX_RX_PKT_CSUMMED) ? 5403d95b884SAlex Maftei (amaftei) CHECKSUM_UNNECESSARY : CHECKSUM_NONE); 5414d9c0a2dSEdward Cree } 5423d95b884SAlex Maftei (amaftei) skb->csum_level = !!(rx_buf->flags & EFX_RX_PKT_CSUM_LEVEL); 5433d95b884SAlex Maftei (amaftei) 5443d95b884SAlex Maftei (amaftei) for (;;) { 5453d95b884SAlex Maftei (amaftei) skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, 5463d95b884SAlex Maftei (amaftei) rx_buf->page, rx_buf->page_offset, 5473d95b884SAlex Maftei (amaftei) rx_buf->len); 5483d95b884SAlex Maftei (amaftei) rx_buf->page = NULL; 5493d95b884SAlex Maftei (amaftei) skb->len += rx_buf->len; 5503d95b884SAlex Maftei (amaftei) if (skb_shinfo(skb)->nr_frags == n_frags) 5513d95b884SAlex Maftei (amaftei) break; 5523d95b884SAlex Maftei (amaftei) 5533d95b884SAlex Maftei (amaftei) rx_buf = efx_rx_buf_next(&channel->rx_queue, rx_buf); 5543d95b884SAlex Maftei (amaftei) } 5553d95b884SAlex Maftei (amaftei) 5563d95b884SAlex Maftei (amaftei) skb->data_len = skb->len; 5573d95b884SAlex Maftei (amaftei) skb->truesize += n_frags * efx->rx_buffer_truesize; 5583d95b884SAlex Maftei (amaftei) 5593d95b884SAlex Maftei (amaftei) skb_record_rx_queue(skb, channel->rx_queue.core_index); 5603d95b884SAlex Maftei (amaftei) 5613d95b884SAlex Maftei (amaftei) napi_gro_frags(napi); 5623d95b884SAlex Maftei (amaftei) } 563960f1627SAlex Maftei (amaftei) 564960f1627SAlex Maftei (amaftei) /* RSS contexts. We're using linked lists and crappy O(n) algorithms, because 565960f1627SAlex Maftei (amaftei) * (a) this is an infrequent control-plane operation and (b) n is small (max 64) 566960f1627SAlex Maftei (amaftei) */ 567960f1627SAlex Maftei (amaftei) struct efx_rss_context *efx_alloc_rss_context_entry(struct efx_nic *efx) 568960f1627SAlex Maftei (amaftei) { 569960f1627SAlex Maftei (amaftei) struct list_head *head = &efx->rss_context.list; 570960f1627SAlex Maftei (amaftei) struct efx_rss_context *ctx, *new; 571960f1627SAlex Maftei (amaftei) u32 id = 1; /* Don't use zero, that refers to the master RSS context */ 572960f1627SAlex Maftei (amaftei) 573960f1627SAlex Maftei (amaftei) WARN_ON(!mutex_is_locked(&efx->rss_lock)); 574960f1627SAlex Maftei (amaftei) 575960f1627SAlex Maftei (amaftei) /* Search for first gap in the numbering */ 576960f1627SAlex Maftei (amaftei) list_for_each_entry(ctx, head, list) { 577960f1627SAlex Maftei (amaftei) if (ctx->user_id != id) 578960f1627SAlex Maftei (amaftei) break; 579960f1627SAlex Maftei (amaftei) id++; 580960f1627SAlex Maftei (amaftei) /* Check for wrap. If this happens, we have nearly 2^32 581960f1627SAlex Maftei (amaftei) * allocated RSS contexts, which seems unlikely. 582960f1627SAlex Maftei (amaftei) */ 583960f1627SAlex Maftei (amaftei) if (WARN_ON_ONCE(!id)) 584960f1627SAlex Maftei (amaftei) return NULL; 585960f1627SAlex Maftei (amaftei) } 586960f1627SAlex Maftei (amaftei) 587960f1627SAlex Maftei (amaftei) /* Create the new entry */ 588960f1627SAlex Maftei (amaftei) new = kmalloc(sizeof(*new), GFP_KERNEL); 589960f1627SAlex Maftei (amaftei) if (!new) 590960f1627SAlex Maftei (amaftei) return NULL; 591f7226e0fSAlex Maftei (amaftei) new->context_id = EFX_MCDI_RSS_CONTEXT_INVALID; 592960f1627SAlex Maftei (amaftei) new->rx_hash_udp_4tuple = false; 593960f1627SAlex Maftei (amaftei) 594960f1627SAlex Maftei (amaftei) /* Insert the new entry into the gap */ 595960f1627SAlex Maftei (amaftei) new->user_id = id; 596960f1627SAlex Maftei (amaftei) list_add_tail(&new->list, &ctx->list); 597960f1627SAlex Maftei (amaftei) return new; 598960f1627SAlex Maftei (amaftei) } 599960f1627SAlex Maftei (amaftei) 600960f1627SAlex Maftei (amaftei) struct efx_rss_context *efx_find_rss_context_entry(struct efx_nic *efx, u32 id) 601960f1627SAlex Maftei (amaftei) { 602960f1627SAlex Maftei (amaftei) struct list_head *head = &efx->rss_context.list; 603960f1627SAlex Maftei (amaftei) struct efx_rss_context *ctx; 604960f1627SAlex Maftei (amaftei) 605960f1627SAlex Maftei (amaftei) WARN_ON(!mutex_is_locked(&efx->rss_lock)); 606960f1627SAlex Maftei (amaftei) 607960f1627SAlex Maftei (amaftei) list_for_each_entry(ctx, head, list) 608960f1627SAlex Maftei (amaftei) if (ctx->user_id == id) 609960f1627SAlex Maftei (amaftei) return ctx; 610960f1627SAlex Maftei (amaftei) return NULL; 611960f1627SAlex Maftei (amaftei) } 612960f1627SAlex Maftei (amaftei) 613960f1627SAlex Maftei (amaftei) void efx_free_rss_context_entry(struct efx_rss_context *ctx) 614960f1627SAlex Maftei (amaftei) { 615960f1627SAlex Maftei (amaftei) list_del(&ctx->list); 616960f1627SAlex Maftei (amaftei) kfree(ctx); 617960f1627SAlex Maftei (amaftei) } 618960f1627SAlex Maftei (amaftei) 619960f1627SAlex Maftei (amaftei) void efx_set_default_rx_indir_table(struct efx_nic *efx, 620960f1627SAlex Maftei (amaftei) struct efx_rss_context *ctx) 621960f1627SAlex Maftei (amaftei) { 622960f1627SAlex Maftei (amaftei) size_t i; 623960f1627SAlex Maftei (amaftei) 624960f1627SAlex Maftei (amaftei) for (i = 0; i < ARRAY_SIZE(ctx->rx_indir_table); i++) 625960f1627SAlex Maftei (amaftei) ctx->rx_indir_table[i] = 626960f1627SAlex Maftei (amaftei) ethtool_rxfh_indir_default(i, efx->rss_spread); 627960f1627SAlex Maftei (amaftei) } 628f7226e0fSAlex Maftei (amaftei) 629f7226e0fSAlex Maftei (amaftei) /** 630f7226e0fSAlex Maftei (amaftei) * efx_filter_is_mc_recipient - test whether spec is a multicast recipient 631f7226e0fSAlex Maftei (amaftei) * @spec: Specification to test 632f7226e0fSAlex Maftei (amaftei) * 633f7226e0fSAlex Maftei (amaftei) * Return: %true if the specification is a non-drop RX filter that 634f7226e0fSAlex Maftei (amaftei) * matches a local MAC address I/G bit value of 1 or matches a local 635f7226e0fSAlex Maftei (amaftei) * IPv4 or IPv6 address value in the respective multicast address 636f7226e0fSAlex Maftei (amaftei) * range. Otherwise %false. 637f7226e0fSAlex Maftei (amaftei) */ 638f7226e0fSAlex Maftei (amaftei) bool efx_filter_is_mc_recipient(const struct efx_filter_spec *spec) 639f7226e0fSAlex Maftei (amaftei) { 640f7226e0fSAlex Maftei (amaftei) if (!(spec->flags & EFX_FILTER_FLAG_RX) || 641f7226e0fSAlex Maftei (amaftei) spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP) 642f7226e0fSAlex Maftei (amaftei) return false; 643f7226e0fSAlex Maftei (amaftei) 644f7226e0fSAlex Maftei (amaftei) if (spec->match_flags & 645f7226e0fSAlex Maftei (amaftei) (EFX_FILTER_MATCH_LOC_MAC | EFX_FILTER_MATCH_LOC_MAC_IG) && 646f7226e0fSAlex Maftei (amaftei) is_multicast_ether_addr(spec->loc_mac)) 647f7226e0fSAlex Maftei (amaftei) return true; 648f7226e0fSAlex Maftei (amaftei) 649f7226e0fSAlex Maftei (amaftei) if ((spec->match_flags & 650f7226e0fSAlex Maftei (amaftei) (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) == 651f7226e0fSAlex Maftei (amaftei) (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) { 652f7226e0fSAlex Maftei (amaftei) if (spec->ether_type == htons(ETH_P_IP) && 653f7226e0fSAlex Maftei (amaftei) ipv4_is_multicast(spec->loc_host[0])) 654f7226e0fSAlex Maftei (amaftei) return true; 655f7226e0fSAlex Maftei (amaftei) if (spec->ether_type == htons(ETH_P_IPV6) && 656f7226e0fSAlex Maftei (amaftei) ((const u8 *)spec->loc_host)[0] == 0xff) 657f7226e0fSAlex Maftei (amaftei) return true; 658f7226e0fSAlex Maftei (amaftei) } 659f7226e0fSAlex Maftei (amaftei) 660f7226e0fSAlex Maftei (amaftei) return false; 661f7226e0fSAlex Maftei (amaftei) } 662f7226e0fSAlex Maftei (amaftei) 663f7226e0fSAlex Maftei (amaftei) bool efx_filter_spec_equal(const struct efx_filter_spec *left, 664f7226e0fSAlex Maftei (amaftei) const struct efx_filter_spec *right) 665f7226e0fSAlex Maftei (amaftei) { 666f7226e0fSAlex Maftei (amaftei) if ((left->match_flags ^ right->match_flags) | 667f7226e0fSAlex Maftei (amaftei) ((left->flags ^ right->flags) & 668f7226e0fSAlex Maftei (amaftei) (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX))) 669f7226e0fSAlex Maftei (amaftei) return false; 670f7226e0fSAlex Maftei (amaftei) 671f7226e0fSAlex Maftei (amaftei) return memcmp(&left->outer_vid, &right->outer_vid, 672f7226e0fSAlex Maftei (amaftei) sizeof(struct efx_filter_spec) - 673f7226e0fSAlex Maftei (amaftei) offsetof(struct efx_filter_spec, outer_vid)) == 0; 674f7226e0fSAlex Maftei (amaftei) } 675f7226e0fSAlex Maftei (amaftei) 676f7226e0fSAlex Maftei (amaftei) u32 efx_filter_spec_hash(const struct efx_filter_spec *spec) 677f7226e0fSAlex Maftei (amaftei) { 678f7226e0fSAlex Maftei (amaftei) BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3); 679f7226e0fSAlex Maftei (amaftei) return jhash2((const u32 *)&spec->outer_vid, 680f7226e0fSAlex Maftei (amaftei) (sizeof(struct efx_filter_spec) - 681f7226e0fSAlex Maftei (amaftei) offsetof(struct efx_filter_spec, outer_vid)) / 4, 682f7226e0fSAlex Maftei (amaftei) 0); 683f7226e0fSAlex Maftei (amaftei) } 684f7226e0fSAlex Maftei (amaftei) 685f7226e0fSAlex Maftei (amaftei) #ifdef CONFIG_RFS_ACCEL 686f7226e0fSAlex Maftei (amaftei) bool efx_rps_check_rule(struct efx_arfs_rule *rule, unsigned int filter_idx, 687f7226e0fSAlex Maftei (amaftei) bool *force) 688f7226e0fSAlex Maftei (amaftei) { 689f7226e0fSAlex Maftei (amaftei) if (rule->filter_id == EFX_ARFS_FILTER_ID_PENDING) { 690f7226e0fSAlex Maftei (amaftei) /* ARFS is currently updating this entry, leave it */ 691f7226e0fSAlex Maftei (amaftei) return false; 692f7226e0fSAlex Maftei (amaftei) } 693f7226e0fSAlex Maftei (amaftei) if (rule->filter_id == EFX_ARFS_FILTER_ID_ERROR) { 694f7226e0fSAlex Maftei (amaftei) /* ARFS tried and failed to update this, so it's probably out 695f7226e0fSAlex Maftei (amaftei) * of date. Remove the filter and the ARFS rule entry. 696f7226e0fSAlex Maftei (amaftei) */ 697f7226e0fSAlex Maftei (amaftei) rule->filter_id = EFX_ARFS_FILTER_ID_REMOVING; 698f7226e0fSAlex Maftei (amaftei) *force = true; 699f7226e0fSAlex Maftei (amaftei) return true; 700f7226e0fSAlex Maftei (amaftei) } else if (WARN_ON(rule->filter_id != filter_idx)) { /* can't happen */ 701f7226e0fSAlex Maftei (amaftei) /* ARFS has moved on, so old filter is not needed. Since we did 702f7226e0fSAlex Maftei (amaftei) * not mark the rule with EFX_ARFS_FILTER_ID_REMOVING, it will 703f7226e0fSAlex Maftei (amaftei) * not be removed by efx_rps_hash_del() subsequently. 704f7226e0fSAlex Maftei (amaftei) */ 705f7226e0fSAlex Maftei (amaftei) *force = true; 706f7226e0fSAlex Maftei (amaftei) return true; 707f7226e0fSAlex Maftei (amaftei) } 708f7226e0fSAlex Maftei (amaftei) /* Remove it iff ARFS wants to. */ 709f7226e0fSAlex Maftei (amaftei) return true; 710f7226e0fSAlex Maftei (amaftei) } 711f7226e0fSAlex Maftei (amaftei) 712f7226e0fSAlex Maftei (amaftei) static 713f7226e0fSAlex Maftei (amaftei) struct hlist_head *efx_rps_hash_bucket(struct efx_nic *efx, 714f7226e0fSAlex Maftei (amaftei) const struct efx_filter_spec *spec) 715f7226e0fSAlex Maftei (amaftei) { 716f7226e0fSAlex Maftei (amaftei) u32 hash = efx_filter_spec_hash(spec); 717f7226e0fSAlex Maftei (amaftei) 718f7226e0fSAlex Maftei (amaftei) lockdep_assert_held(&efx->rps_hash_lock); 719f7226e0fSAlex Maftei (amaftei) if (!efx->rps_hash_table) 720f7226e0fSAlex Maftei (amaftei) return NULL; 721f7226e0fSAlex Maftei (amaftei) return &efx->rps_hash_table[hash % EFX_ARFS_HASH_TABLE_SIZE]; 722f7226e0fSAlex Maftei (amaftei) } 723f7226e0fSAlex Maftei (amaftei) 724f7226e0fSAlex Maftei (amaftei) struct efx_arfs_rule *efx_rps_hash_find(struct efx_nic *efx, 725f7226e0fSAlex Maftei (amaftei) const struct efx_filter_spec *spec) 726f7226e0fSAlex Maftei (amaftei) { 727f7226e0fSAlex Maftei (amaftei) struct efx_arfs_rule *rule; 728f7226e0fSAlex Maftei (amaftei) struct hlist_head *head; 729f7226e0fSAlex Maftei (amaftei) struct hlist_node *node; 730f7226e0fSAlex Maftei (amaftei) 731f7226e0fSAlex Maftei (amaftei) head = efx_rps_hash_bucket(efx, spec); 732f7226e0fSAlex Maftei (amaftei) if (!head) 733f7226e0fSAlex Maftei (amaftei) return NULL; 734f7226e0fSAlex Maftei (amaftei) hlist_for_each(node, head) { 735f7226e0fSAlex Maftei (amaftei) rule = container_of(node, struct efx_arfs_rule, node); 736f7226e0fSAlex Maftei (amaftei) if (efx_filter_spec_equal(spec, &rule->spec)) 737f7226e0fSAlex Maftei (amaftei) return rule; 738f7226e0fSAlex Maftei (amaftei) } 739f7226e0fSAlex Maftei (amaftei) return NULL; 740f7226e0fSAlex Maftei (amaftei) } 741f7226e0fSAlex Maftei (amaftei) 742f7226e0fSAlex Maftei (amaftei) struct efx_arfs_rule *efx_rps_hash_add(struct efx_nic *efx, 743f7226e0fSAlex Maftei (amaftei) const struct efx_filter_spec *spec, 744f7226e0fSAlex Maftei (amaftei) bool *new) 745f7226e0fSAlex Maftei (amaftei) { 746f7226e0fSAlex Maftei (amaftei) struct efx_arfs_rule *rule; 747f7226e0fSAlex Maftei (amaftei) struct hlist_head *head; 748f7226e0fSAlex Maftei (amaftei) struct hlist_node *node; 749f7226e0fSAlex Maftei (amaftei) 750f7226e0fSAlex Maftei (amaftei) head = efx_rps_hash_bucket(efx, spec); 751f7226e0fSAlex Maftei (amaftei) if (!head) 752f7226e0fSAlex Maftei (amaftei) return NULL; 753f7226e0fSAlex Maftei (amaftei) hlist_for_each(node, head) { 754f7226e0fSAlex Maftei (amaftei) rule = container_of(node, struct efx_arfs_rule, node); 755f7226e0fSAlex Maftei (amaftei) if (efx_filter_spec_equal(spec, &rule->spec)) { 756f7226e0fSAlex Maftei (amaftei) *new = false; 757f7226e0fSAlex Maftei (amaftei) return rule; 758f7226e0fSAlex Maftei (amaftei) } 759f7226e0fSAlex Maftei (amaftei) } 760f7226e0fSAlex Maftei (amaftei) rule = kmalloc(sizeof(*rule), GFP_ATOMIC); 761f7226e0fSAlex Maftei (amaftei) *new = true; 762f7226e0fSAlex Maftei (amaftei) if (rule) { 763f7226e0fSAlex Maftei (amaftei) memcpy(&rule->spec, spec, sizeof(rule->spec)); 764f7226e0fSAlex Maftei (amaftei) hlist_add_head(&rule->node, head); 765f7226e0fSAlex Maftei (amaftei) } 766f7226e0fSAlex Maftei (amaftei) return rule; 767f7226e0fSAlex Maftei (amaftei) } 768f7226e0fSAlex Maftei (amaftei) 769f7226e0fSAlex Maftei (amaftei) void efx_rps_hash_del(struct efx_nic *efx, const struct efx_filter_spec *spec) 770f7226e0fSAlex Maftei (amaftei) { 771f7226e0fSAlex Maftei (amaftei) struct efx_arfs_rule *rule; 772f7226e0fSAlex Maftei (amaftei) struct hlist_head *head; 773f7226e0fSAlex Maftei (amaftei) struct hlist_node *node; 774f7226e0fSAlex Maftei (amaftei) 775f7226e0fSAlex Maftei (amaftei) head = efx_rps_hash_bucket(efx, spec); 776f7226e0fSAlex Maftei (amaftei) if (WARN_ON(!head)) 777f7226e0fSAlex Maftei (amaftei) return; 778f7226e0fSAlex Maftei (amaftei) hlist_for_each(node, head) { 779f7226e0fSAlex Maftei (amaftei) rule = container_of(node, struct efx_arfs_rule, node); 780f7226e0fSAlex Maftei (amaftei) if (efx_filter_spec_equal(spec, &rule->spec)) { 781f7226e0fSAlex Maftei (amaftei) /* Someone already reused the entry. We know that if 782f7226e0fSAlex Maftei (amaftei) * this check doesn't fire (i.e. filter_id == REMOVING) 783f7226e0fSAlex Maftei (amaftei) * then the REMOVING mark was put there by our caller, 784f7226e0fSAlex Maftei (amaftei) * because caller is holding a lock on filter table and 785f7226e0fSAlex Maftei (amaftei) * only holders of that lock set REMOVING. 786f7226e0fSAlex Maftei (amaftei) */ 787f7226e0fSAlex Maftei (amaftei) if (rule->filter_id != EFX_ARFS_FILTER_ID_REMOVING) 788f7226e0fSAlex Maftei (amaftei) return; 789f7226e0fSAlex Maftei (amaftei) hlist_del(node); 790f7226e0fSAlex Maftei (amaftei) kfree(rule); 791f7226e0fSAlex Maftei (amaftei) return; 792f7226e0fSAlex Maftei (amaftei) } 793f7226e0fSAlex Maftei (amaftei) } 794f7226e0fSAlex Maftei (amaftei) /* We didn't find it. */ 795f7226e0fSAlex Maftei (amaftei) WARN_ON(1); 796f7226e0fSAlex Maftei (amaftei) } 797f7226e0fSAlex Maftei (amaftei) #endif 798f7226e0fSAlex Maftei (amaftei) 799f7226e0fSAlex Maftei (amaftei) int efx_probe_filters(struct efx_nic *efx) 800f7226e0fSAlex Maftei (amaftei) { 801f7226e0fSAlex Maftei (amaftei) int rc; 802f7226e0fSAlex Maftei (amaftei) 803f7226e0fSAlex Maftei (amaftei) mutex_lock(&efx->mac_lock); 804f7226e0fSAlex Maftei (amaftei) down_write(&efx->filter_sem); 805f7226e0fSAlex Maftei (amaftei) rc = efx->type->filter_table_probe(efx); 806f7226e0fSAlex Maftei (amaftei) if (rc) 807f7226e0fSAlex Maftei (amaftei) goto out_unlock; 808f7226e0fSAlex Maftei (amaftei) 809f7226e0fSAlex Maftei (amaftei) #ifdef CONFIG_RFS_ACCEL 810f7226e0fSAlex Maftei (amaftei) if (efx->type->offload_features & NETIF_F_NTUPLE) { 811f7226e0fSAlex Maftei (amaftei) struct efx_channel *channel; 812f7226e0fSAlex Maftei (amaftei) int i, success = 1; 813f7226e0fSAlex Maftei (amaftei) 814f7226e0fSAlex Maftei (amaftei) efx_for_each_channel(channel, efx) { 815f7226e0fSAlex Maftei (amaftei) channel->rps_flow_id = 816f7226e0fSAlex Maftei (amaftei) kcalloc(efx->type->max_rx_ip_filters, 817f7226e0fSAlex Maftei (amaftei) sizeof(*channel->rps_flow_id), 818f7226e0fSAlex Maftei (amaftei) GFP_KERNEL); 819f7226e0fSAlex Maftei (amaftei) if (!channel->rps_flow_id) 820f7226e0fSAlex Maftei (amaftei) success = 0; 821f7226e0fSAlex Maftei (amaftei) else 822f7226e0fSAlex Maftei (amaftei) for (i = 0; 823f7226e0fSAlex Maftei (amaftei) i < efx->type->max_rx_ip_filters; 824f7226e0fSAlex Maftei (amaftei) ++i) 825f7226e0fSAlex Maftei (amaftei) channel->rps_flow_id[i] = 826f7226e0fSAlex Maftei (amaftei) RPS_FLOW_ID_INVALID; 827f7226e0fSAlex Maftei (amaftei) channel->rfs_expire_index = 0; 828f7226e0fSAlex Maftei (amaftei) channel->rfs_filter_count = 0; 829f7226e0fSAlex Maftei (amaftei) } 830f7226e0fSAlex Maftei (amaftei) 831f7226e0fSAlex Maftei (amaftei) if (!success) { 832f7226e0fSAlex Maftei (amaftei) efx_for_each_channel(channel, efx) 833f7226e0fSAlex Maftei (amaftei) kfree(channel->rps_flow_id); 834f7226e0fSAlex Maftei (amaftei) efx->type->filter_table_remove(efx); 835f7226e0fSAlex Maftei (amaftei) rc = -ENOMEM; 836f7226e0fSAlex Maftei (amaftei) goto out_unlock; 837f7226e0fSAlex Maftei (amaftei) } 838f7226e0fSAlex Maftei (amaftei) } 839f7226e0fSAlex Maftei (amaftei) #endif 840f7226e0fSAlex Maftei (amaftei) out_unlock: 841f7226e0fSAlex Maftei (amaftei) up_write(&efx->filter_sem); 842f7226e0fSAlex Maftei (amaftei) mutex_unlock(&efx->mac_lock); 843f7226e0fSAlex Maftei (amaftei) return rc; 844f7226e0fSAlex Maftei (amaftei) } 845f7226e0fSAlex Maftei (amaftei) 846f7226e0fSAlex Maftei (amaftei) void efx_remove_filters(struct efx_nic *efx) 847f7226e0fSAlex Maftei (amaftei) { 848f7226e0fSAlex Maftei (amaftei) #ifdef CONFIG_RFS_ACCEL 849f7226e0fSAlex Maftei (amaftei) struct efx_channel *channel; 850f7226e0fSAlex Maftei (amaftei) 851f7226e0fSAlex Maftei (amaftei) efx_for_each_channel(channel, efx) { 852f7226e0fSAlex Maftei (amaftei) cancel_delayed_work_sync(&channel->filter_work); 853f7226e0fSAlex Maftei (amaftei) kfree(channel->rps_flow_id); 854788f920aSEdward Cree channel->rps_flow_id = NULL; 855f7226e0fSAlex Maftei (amaftei) } 856f7226e0fSAlex Maftei (amaftei) #endif 857f7226e0fSAlex Maftei (amaftei) down_write(&efx->filter_sem); 858f7226e0fSAlex Maftei (amaftei) efx->type->filter_table_remove(efx); 859f7226e0fSAlex Maftei (amaftei) up_write(&efx->filter_sem); 860f7226e0fSAlex Maftei (amaftei) } 86128abe825SEdward Cree 86228abe825SEdward Cree #ifdef CONFIG_RFS_ACCEL 86328abe825SEdward Cree 86428abe825SEdward Cree static void efx_filter_rfs_work(struct work_struct *data) 86528abe825SEdward Cree { 86628abe825SEdward Cree struct efx_async_filter_insertion *req = container_of(data, struct efx_async_filter_insertion, 86728abe825SEdward Cree work); 86828abe825SEdward Cree struct efx_nic *efx = netdev_priv(req->net_dev); 86928abe825SEdward Cree struct efx_channel *channel = efx_get_channel(efx, req->rxq_index); 87028abe825SEdward Cree int slot_idx = req - efx->rps_slot; 87128abe825SEdward Cree struct efx_arfs_rule *rule; 87228abe825SEdward Cree u16 arfs_id = 0; 87328abe825SEdward Cree int rc; 87428abe825SEdward Cree 87528abe825SEdward Cree rc = efx->type->filter_insert(efx, &req->spec, true); 87628abe825SEdward Cree if (rc >= 0) 87728abe825SEdward Cree /* Discard 'priority' part of EF10+ filter ID (mcdi_filters) */ 87828abe825SEdward Cree rc %= efx->type->max_rx_ip_filters; 87928abe825SEdward Cree if (efx->rps_hash_table) { 88028abe825SEdward Cree spin_lock_bh(&efx->rps_hash_lock); 88128abe825SEdward Cree rule = efx_rps_hash_find(efx, &req->spec); 88228abe825SEdward Cree /* The rule might have already gone, if someone else's request 88328abe825SEdward Cree * for the same spec was already worked and then expired before 88428abe825SEdward Cree * we got around to our work. In that case we have nothing 88528abe825SEdward Cree * tying us to an arfs_id, meaning that as soon as the filter 88628abe825SEdward Cree * is considered for expiry it will be removed. 88728abe825SEdward Cree */ 88828abe825SEdward Cree if (rule) { 88928abe825SEdward Cree if (rc < 0) 89028abe825SEdward Cree rule->filter_id = EFX_ARFS_FILTER_ID_ERROR; 89128abe825SEdward Cree else 89228abe825SEdward Cree rule->filter_id = rc; 89328abe825SEdward Cree arfs_id = rule->arfs_id; 89428abe825SEdward Cree } 89528abe825SEdward Cree spin_unlock_bh(&efx->rps_hash_lock); 89628abe825SEdward Cree } 89728abe825SEdward Cree if (rc >= 0) { 89828abe825SEdward Cree /* Remember this so we can check whether to expire the filter 89928abe825SEdward Cree * later. 90028abe825SEdward Cree */ 90128abe825SEdward Cree mutex_lock(&efx->rps_mutex); 90228abe825SEdward Cree if (channel->rps_flow_id[rc] == RPS_FLOW_ID_INVALID) 90328abe825SEdward Cree channel->rfs_filter_count++; 90428abe825SEdward Cree channel->rps_flow_id[rc] = req->flow_id; 90528abe825SEdward Cree mutex_unlock(&efx->rps_mutex); 90628abe825SEdward Cree 90728abe825SEdward Cree if (req->spec.ether_type == htons(ETH_P_IP)) 90828abe825SEdward Cree netif_info(efx, rx_status, efx->net_dev, 90928abe825SEdward Cree "steering %s %pI4:%u:%pI4:%u to queue %u [flow %u filter %d id %u]\n", 91028abe825SEdward Cree (req->spec.ip_proto == IPPROTO_TCP) ? "TCP" : "UDP", 91128abe825SEdward Cree req->spec.rem_host, ntohs(req->spec.rem_port), 91228abe825SEdward Cree req->spec.loc_host, ntohs(req->spec.loc_port), 91328abe825SEdward Cree req->rxq_index, req->flow_id, rc, arfs_id); 91428abe825SEdward Cree else 91528abe825SEdward Cree netif_info(efx, rx_status, efx->net_dev, 91628abe825SEdward Cree "steering %s [%pI6]:%u:[%pI6]:%u to queue %u [flow %u filter %d id %u]\n", 91728abe825SEdward Cree (req->spec.ip_proto == IPPROTO_TCP) ? "TCP" : "UDP", 91828abe825SEdward Cree req->spec.rem_host, ntohs(req->spec.rem_port), 91928abe825SEdward Cree req->spec.loc_host, ntohs(req->spec.loc_port), 92028abe825SEdward Cree req->rxq_index, req->flow_id, rc, arfs_id); 92128abe825SEdward Cree channel->n_rfs_succeeded++; 92228abe825SEdward Cree } else { 92328abe825SEdward Cree if (req->spec.ether_type == htons(ETH_P_IP)) 92428abe825SEdward Cree netif_dbg(efx, rx_status, efx->net_dev, 92528abe825SEdward Cree "failed to steer %s %pI4:%u:%pI4:%u to queue %u [flow %u rc %d id %u]\n", 92628abe825SEdward Cree (req->spec.ip_proto == IPPROTO_TCP) ? "TCP" : "UDP", 92728abe825SEdward Cree req->spec.rem_host, ntohs(req->spec.rem_port), 92828abe825SEdward Cree req->spec.loc_host, ntohs(req->spec.loc_port), 92928abe825SEdward Cree req->rxq_index, req->flow_id, rc, arfs_id); 93028abe825SEdward Cree else 93128abe825SEdward Cree netif_dbg(efx, rx_status, efx->net_dev, 93228abe825SEdward Cree "failed to steer %s [%pI6]:%u:[%pI6]:%u to queue %u [flow %u rc %d id %u]\n", 93328abe825SEdward Cree (req->spec.ip_proto == IPPROTO_TCP) ? "TCP" : "UDP", 93428abe825SEdward Cree req->spec.rem_host, ntohs(req->spec.rem_port), 93528abe825SEdward Cree req->spec.loc_host, ntohs(req->spec.loc_port), 93628abe825SEdward Cree req->rxq_index, req->flow_id, rc, arfs_id); 93728abe825SEdward Cree channel->n_rfs_failed++; 93828abe825SEdward Cree /* We're overloading the NIC's filter tables, so let's do a 93928abe825SEdward Cree * chunk of extra expiry work. 94028abe825SEdward Cree */ 94128abe825SEdward Cree __efx_filter_rfs_expire(channel, min(channel->rfs_filter_count, 94228abe825SEdward Cree 100u)); 94328abe825SEdward Cree } 94428abe825SEdward Cree 94528abe825SEdward Cree /* Release references */ 94628abe825SEdward Cree clear_bit(slot_idx, &efx->rps_slot_map); 94728abe825SEdward Cree dev_put(req->net_dev); 94828abe825SEdward Cree } 94928abe825SEdward Cree 95028abe825SEdward Cree int efx_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb, 95128abe825SEdward Cree u16 rxq_index, u32 flow_id) 95228abe825SEdward Cree { 95328abe825SEdward Cree struct efx_nic *efx = netdev_priv(net_dev); 95428abe825SEdward Cree struct efx_async_filter_insertion *req; 95528abe825SEdward Cree struct efx_arfs_rule *rule; 95628abe825SEdward Cree struct flow_keys fk; 95728abe825SEdward Cree int slot_idx; 95828abe825SEdward Cree bool new; 95928abe825SEdward Cree int rc; 96028abe825SEdward Cree 96128abe825SEdward Cree /* find a free slot */ 96228abe825SEdward Cree for (slot_idx = 0; slot_idx < EFX_RPS_MAX_IN_FLIGHT; slot_idx++) 96328abe825SEdward Cree if (!test_and_set_bit(slot_idx, &efx->rps_slot_map)) 96428abe825SEdward Cree break; 96528abe825SEdward Cree if (slot_idx >= EFX_RPS_MAX_IN_FLIGHT) 96628abe825SEdward Cree return -EBUSY; 96728abe825SEdward Cree 96828abe825SEdward Cree if (flow_id == RPS_FLOW_ID_INVALID) { 96928abe825SEdward Cree rc = -EINVAL; 97028abe825SEdward Cree goto out_clear; 97128abe825SEdward Cree } 97228abe825SEdward Cree 97328abe825SEdward Cree if (!skb_flow_dissect_flow_keys(skb, &fk, 0)) { 97428abe825SEdward Cree rc = -EPROTONOSUPPORT; 97528abe825SEdward Cree goto out_clear; 97628abe825SEdward Cree } 97728abe825SEdward Cree 97828abe825SEdward Cree if (fk.basic.n_proto != htons(ETH_P_IP) && fk.basic.n_proto != htons(ETH_P_IPV6)) { 97928abe825SEdward Cree rc = -EPROTONOSUPPORT; 98028abe825SEdward Cree goto out_clear; 98128abe825SEdward Cree } 98228abe825SEdward Cree if (fk.control.flags & FLOW_DIS_IS_FRAGMENT) { 98328abe825SEdward Cree rc = -EPROTONOSUPPORT; 98428abe825SEdward Cree goto out_clear; 98528abe825SEdward Cree } 98628abe825SEdward Cree 98728abe825SEdward Cree req = efx->rps_slot + slot_idx; 98828abe825SEdward Cree efx_filter_init_rx(&req->spec, EFX_FILTER_PRI_HINT, 98928abe825SEdward Cree efx->rx_scatter ? EFX_FILTER_FLAG_RX_SCATTER : 0, 99028abe825SEdward Cree rxq_index); 99128abe825SEdward Cree req->spec.match_flags = 99228abe825SEdward Cree EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_IP_PROTO | 99328abe825SEdward Cree EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT | 99428abe825SEdward Cree EFX_FILTER_MATCH_REM_HOST | EFX_FILTER_MATCH_REM_PORT; 99528abe825SEdward Cree req->spec.ether_type = fk.basic.n_proto; 99628abe825SEdward Cree req->spec.ip_proto = fk.basic.ip_proto; 99728abe825SEdward Cree 99828abe825SEdward Cree if (fk.basic.n_proto == htons(ETH_P_IP)) { 99928abe825SEdward Cree req->spec.rem_host[0] = fk.addrs.v4addrs.src; 100028abe825SEdward Cree req->spec.loc_host[0] = fk.addrs.v4addrs.dst; 100128abe825SEdward Cree } else { 100228abe825SEdward Cree memcpy(req->spec.rem_host, &fk.addrs.v6addrs.src, 100328abe825SEdward Cree sizeof(struct in6_addr)); 100428abe825SEdward Cree memcpy(req->spec.loc_host, &fk.addrs.v6addrs.dst, 100528abe825SEdward Cree sizeof(struct in6_addr)); 100628abe825SEdward Cree } 100728abe825SEdward Cree 100828abe825SEdward Cree req->spec.rem_port = fk.ports.src; 100928abe825SEdward Cree req->spec.loc_port = fk.ports.dst; 101028abe825SEdward Cree 101128abe825SEdward Cree if (efx->rps_hash_table) { 101228abe825SEdward Cree /* Add it to ARFS hash table */ 101328abe825SEdward Cree spin_lock(&efx->rps_hash_lock); 101428abe825SEdward Cree rule = efx_rps_hash_add(efx, &req->spec, &new); 101528abe825SEdward Cree if (!rule) { 101628abe825SEdward Cree rc = -ENOMEM; 101728abe825SEdward Cree goto out_unlock; 101828abe825SEdward Cree } 101928abe825SEdward Cree if (new) 102028abe825SEdward Cree rule->arfs_id = efx->rps_next_id++ % RPS_NO_FILTER; 102128abe825SEdward Cree rc = rule->arfs_id; 102228abe825SEdward Cree /* Skip if existing or pending filter already does the right thing */ 102328abe825SEdward Cree if (!new && rule->rxq_index == rxq_index && 102428abe825SEdward Cree rule->filter_id >= EFX_ARFS_FILTER_ID_PENDING) 102528abe825SEdward Cree goto out_unlock; 102628abe825SEdward Cree rule->rxq_index = rxq_index; 102728abe825SEdward Cree rule->filter_id = EFX_ARFS_FILTER_ID_PENDING; 102828abe825SEdward Cree spin_unlock(&efx->rps_hash_lock); 102928abe825SEdward Cree } else { 103028abe825SEdward Cree /* Without an ARFS hash table, we just use arfs_id 0 for all 103128abe825SEdward Cree * filters. This means if multiple flows hash to the same 103228abe825SEdward Cree * flow_id, all but the most recently touched will be eligible 103328abe825SEdward Cree * for expiry. 103428abe825SEdward Cree */ 103528abe825SEdward Cree rc = 0; 103628abe825SEdward Cree } 103728abe825SEdward Cree 103828abe825SEdward Cree /* Queue the request */ 103928abe825SEdward Cree dev_hold(req->net_dev = net_dev); 104028abe825SEdward Cree INIT_WORK(&req->work, efx_filter_rfs_work); 104128abe825SEdward Cree req->rxq_index = rxq_index; 104228abe825SEdward Cree req->flow_id = flow_id; 104328abe825SEdward Cree schedule_work(&req->work); 104428abe825SEdward Cree return rc; 104528abe825SEdward Cree out_unlock: 104628abe825SEdward Cree spin_unlock(&efx->rps_hash_lock); 104728abe825SEdward Cree out_clear: 104828abe825SEdward Cree clear_bit(slot_idx, &efx->rps_slot_map); 104928abe825SEdward Cree return rc; 105028abe825SEdward Cree } 105128abe825SEdward Cree 105228abe825SEdward Cree bool __efx_filter_rfs_expire(struct efx_channel *channel, unsigned int quota) 105328abe825SEdward Cree { 105428abe825SEdward Cree bool (*expire_one)(struct efx_nic *efx, u32 flow_id, unsigned int index); 105528abe825SEdward Cree struct efx_nic *efx = channel->efx; 105628abe825SEdward Cree unsigned int index, size, start; 105728abe825SEdward Cree u32 flow_id; 105828abe825SEdward Cree 105928abe825SEdward Cree if (!mutex_trylock(&efx->rps_mutex)) 106028abe825SEdward Cree return false; 106128abe825SEdward Cree expire_one = efx->type->filter_rfs_expire_one; 106228abe825SEdward Cree index = channel->rfs_expire_index; 106328abe825SEdward Cree start = index; 106428abe825SEdward Cree size = efx->type->max_rx_ip_filters; 106528abe825SEdward Cree while (quota) { 106628abe825SEdward Cree flow_id = channel->rps_flow_id[index]; 106728abe825SEdward Cree 106828abe825SEdward Cree if (flow_id != RPS_FLOW_ID_INVALID) { 106928abe825SEdward Cree quota--; 107028abe825SEdward Cree if (expire_one(efx, flow_id, index)) { 107128abe825SEdward Cree netif_info(efx, rx_status, efx->net_dev, 107228abe825SEdward Cree "expired filter %d [channel %u flow %u]\n", 107328abe825SEdward Cree index, channel->channel, flow_id); 107428abe825SEdward Cree channel->rps_flow_id[index] = RPS_FLOW_ID_INVALID; 107528abe825SEdward Cree channel->rfs_filter_count--; 107628abe825SEdward Cree } 107728abe825SEdward Cree } 107828abe825SEdward Cree if (++index == size) 107928abe825SEdward Cree index = 0; 108028abe825SEdward Cree /* If we were called with a quota that exceeds the total number 108128abe825SEdward Cree * of filters in the table (which shouldn't happen, but could 108228abe825SEdward Cree * if two callers race), ensure that we don't loop forever - 108328abe825SEdward Cree * stop when we've examined every row of the table. 108428abe825SEdward Cree */ 108528abe825SEdward Cree if (index == start) 108628abe825SEdward Cree break; 108728abe825SEdward Cree } 108828abe825SEdward Cree 108928abe825SEdward Cree channel->rfs_expire_index = index; 109028abe825SEdward Cree mutex_unlock(&efx->rps_mutex); 109128abe825SEdward Cree return true; 109228abe825SEdward Cree } 109328abe825SEdward Cree 109428abe825SEdward Cree #endif /* CONFIG_RFS_ACCEL */ 1095