1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
3 * Copyright 2015-2020 Amazon.com, Inc. or its affiliates. All rights reserved.
9 #include <linux/bitops.h>
10 #include <linux/dim.h>
11 #include <linux/etherdevice.h>
12 #include <linux/if_vlan.h>
13 #include <linux/inetdevice.h>
14 #include <linux/interrupt.h>
15 #include <linux/netdevice.h>
16 #include <linux/skbuff.h>
19 #include "ena_eth_com.h"
21 #define DRV_MODULE_GEN_MAJOR 2
22 #define DRV_MODULE_GEN_MINOR 1
23 #define DRV_MODULE_GEN_SUBMINOR 0
25 #define DRV_MODULE_NAME "ena"
27 #define DEVICE_NAME "Elastic Network Adapter (ENA)"
29 /* 1 for AENQ + ADMIN */
30 #define ENA_ADMIN_MSIX_VEC 1
31 #define ENA_MAX_MSIX_VEC(io_queues) (ENA_ADMIN_MSIX_VEC + (io_queues))
33 /* The ENA buffer length fields is 16 bit long. So when PAGE_SIZE == 64kB the
35 * Since the max packet size the ENA handles is ~9kB limit the buffer length to
38 #if PAGE_SIZE > SZ_16K
39 #define ENA_PAGE_SIZE (_AC(SZ_16K, UL))
41 #define ENA_PAGE_SIZE PAGE_SIZE
44 #define ENA_MIN_MSIX_VEC 2
48 #define ENA_BAR_MASK (BIT(ENA_REG_BAR) | BIT(ENA_MEM_BAR))
50 #define ENA_DEFAULT_RING_SIZE (1024)
51 #define ENA_MIN_RING_SIZE (256)
53 #define ENA_MIN_NUM_IO_QUEUES (1)
55 #define ENA_TX_WAKEUP_THRESH (MAX_SKB_FRAGS + 2)
56 #define ENA_DEFAULT_RX_COPYBREAK (256 - NET_IP_ALIGN)
58 #define ENA_MIN_MTU 128
60 #define ENA_NAME_MAX_LEN 20
61 #define ENA_IRQNAME_SIZE 40
63 #define ENA_PKT_MAX_BUFS 19
65 #define ENA_RX_RSS_TABLE_LOG_SIZE 7
66 #define ENA_RX_RSS_TABLE_SIZE (1 << ENA_RX_RSS_TABLE_LOG_SIZE)
68 /* The number of tx packet completions that will be handled each NAPI poll
69 * cycle is ring_size / ENA_TX_POLL_BUDGET_DIVIDER.
71 #define ENA_TX_POLL_BUDGET_DIVIDER 4
73 /* Refill Rx queue when number of required descriptors is above
74 * QUEUE_SIZE / ENA_RX_REFILL_THRESH_DIVIDER or ENA_RX_REFILL_THRESH_PACKET
76 #define ENA_RX_REFILL_THRESH_DIVIDER 8
77 #define ENA_RX_REFILL_THRESH_PACKET 256
79 /* Number of queues to check for missing queues per timer service */
80 #define ENA_MONITORED_TX_QUEUES 4
81 /* Max timeout packets before device reset */
82 #define MAX_NUM_OF_TIMEOUTED_PACKETS 128
84 #define ENA_TX_RING_IDX_NEXT(idx, ring_size) (((idx) + 1) & ((ring_size) - 1))
86 #define ENA_RX_RING_IDX_NEXT(idx, ring_size) (((idx) + 1) & ((ring_size) - 1))
87 #define ENA_RX_RING_IDX_ADD(idx, n, ring_size) \
88 (((idx) + (n)) & ((ring_size) - 1))
90 #define ENA_IO_TXQ_IDX(q) (2 * (q))
91 #define ENA_IO_RXQ_IDX(q) (2 * (q) + 1)
92 #define ENA_IO_TXQ_IDX_TO_COMBINED_IDX(q) ((q) / 2)
93 #define ENA_IO_RXQ_IDX_TO_COMBINED_IDX(q) (((q) - 1) / 2)
95 #define ENA_MGMNT_IRQ_IDX 0
96 #define ENA_IO_IRQ_FIRST_IDX 1
97 #define ENA_IO_IRQ_IDX(q) (ENA_IO_IRQ_FIRST_IDX + (q))
99 #define ENA_ADMIN_POLL_DELAY_US 100
101 /* ENA device should send keep alive msg every 1 sec.
102 * We wait for 6 sec just to be on the safe side.
104 #define ENA_DEVICE_KALIVE_TIMEOUT (6 * HZ)
105 #define ENA_MAX_NO_INTERRUPT_ITERATIONS 3
107 #define ENA_MMIO_DISABLE_REG_READ BIT(0)
109 /* The max MTU size is configured to be the ethernet frame size without
110 * the overhead of the ethernet header, which can have a VLAN header, and
111 * a frame check sequence (FCS).
112 * The buffer size we share with the device is defined to be ENA_PAGE_SIZE
115 #define ENA_XDP_MAX_MTU (ENA_PAGE_SIZE - ETH_HLEN - ETH_FCS_LEN - \
116 VLAN_HLEN - XDP_PACKET_HEADROOM - \
117 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
119 #define ENA_IS_XDP_INDEX(adapter, index) (((index) >= (adapter)->xdp_first_ring) && \
120 ((index) < (adapter)->xdp_first_ring + (adapter)->xdp_num_queues))
123 irq_handler_t handler;
127 cpumask_t affinity_hint_mask;
128 char name[ENA_IRQNAME_SIZE];
132 u8 first_interrupt ____cacheline_aligned;
133 u8 interrupts_masked;
134 struct napi_struct napi;
135 struct ena_ring *tx_ring;
136 struct ena_ring *rx_ring;
137 struct ena_ring *xdp_ring;
142 struct ena_calc_queue_size_ctx {
143 struct ena_com_dev_get_features_ctx *get_feat_ctx;
144 struct ena_com_dev *ena_dev;
145 struct pci_dev *pdev;
148 u32 max_tx_queue_size;
149 u32 max_rx_queue_size;
154 struct ena_tx_buffer {
156 /* num of ena desc for this specific skb
157 * (includes data desc and metadata desc)
160 /* num of buffers used by this skb */
163 /* XDP buffer structure which is used for sending packets in
166 struct xdp_frame *xdpf;
168 /* Indicate if bufs[0] map the linear data of the skb. */
171 /* Used for detect missing tx packets to limit the number of prints */
173 /* Save the last jiffies to detect missing tx packets
175 * sets to non zero value on ena_start_xmit and set to zero on
176 * napi and timer_Service_routine.
178 * while this value is not protected by lock,
179 * a given packet is not expected to be handled by ena_start_xmit
180 * and by napi/timer_service at the same time.
182 unsigned long last_jiffies;
183 struct ena_com_buf bufs[ENA_PKT_MAX_BUFS];
184 } ____cacheline_aligned;
186 struct ena_rx_buffer {
190 struct ena_com_buf ena_buf;
191 } ____cacheline_aligned;
193 struct ena_stats_tx {
201 u64 linearize_failed;
208 u64 unmask_interrupt;
209 u64 last_napi_jiffies;
212 struct ena_stats_rx {
215 u64 rx_copybreak_pkt;
235 /* Holds the empty requests for TX/RX
236 * out of order completions
241 struct ena_tx_buffer *tx_buffer_info;
242 struct ena_rx_buffer *rx_buffer_info;
245 /* cache ptr to avoid using the adapter */
247 struct pci_dev *pdev;
248 struct napi_struct *napi;
249 struct net_device *netdev;
250 struct ena_com_dev *ena_dev;
251 struct ena_adapter *adapter;
252 struct ena_com_io_cq *ena_com_io_cq;
253 struct ena_com_io_sq *ena_com_io_sq;
254 struct bpf_prog *xdp_bpf_prog;
255 struct xdp_rxq_info xdp_rxq;
256 spinlock_t xdp_tx_lock; /* synchronize XDP TX/Redirect traffic */
257 /* Used for rx queues only to point to the xdp tx ring, to
258 * which traffic should be redirected from this rx ring.
260 struct ena_ring *xdp_ring;
270 /* The maximum header length the device can handle */
271 u8 tx_max_header_size;
273 bool disable_meta_caching;
274 u16 no_interrupt_event_cnt;
278 /* number of tx/rx_buffer_info's entries */
281 enum ena_admin_placement_policy_type tx_mem_queue_type;
283 struct ena_com_rx_buf_info ena_bufs[ENA_PKT_MAX_BUFS];
284 u32 smoothed_interval;
285 u32 per_napi_packets;
286 u16 non_empty_napi_events;
287 struct u64_stats_sync syncp;
289 struct ena_stats_tx tx_stats;
290 struct ena_stats_rx rx_stats;
293 u8 *push_buf_intermediate_buf;
295 } ____cacheline_aligned;
297 struct ena_stats_dev {
310 ENA_FLAG_DEVICE_RUNNING,
313 ENA_FLAG_MSIX_ENABLED,
314 ENA_FLAG_TRIGGER_RESET,
315 ENA_FLAG_ONGOING_RESET
318 /* adapter specific private data structure */
320 struct ena_com_dev *ena_dev;
321 /* OS defined structs */
322 struct net_device *netdev;
323 struct pci_dev *pdev;
325 /* rx packets that shorter that this len will be copied to the skb
332 u32 max_num_io_queues;
336 u32 missing_tx_completion_threshold;
338 u32 requested_tx_ring_size;
339 u32 requested_rx_ring_size;
341 u32 max_tx_ring_size;
342 u32 max_rx_ring_size;
349 u8 mac_addr[ETH_ALEN];
351 unsigned long keep_alive_timeout;
352 unsigned long missing_tx_completion_to;
354 char name[ENA_NAME_MAX_LEN];
358 struct ena_ring tx_ring[ENA_MAX_NUM_IO_QUEUES]
359 ____cacheline_aligned_in_smp;
362 struct ena_ring rx_ring[ENA_MAX_NUM_IO_QUEUES]
363 ____cacheline_aligned_in_smp;
365 struct ena_napi ena_napi[ENA_MAX_NUM_IO_QUEUES];
367 struct ena_irq irq_tbl[ENA_MAX_MSIX_VEC(ENA_MAX_NUM_IO_QUEUES)];
370 struct work_struct reset_task;
371 struct timer_list timer_service;
374 bool dev_up_before_reset;
375 bool disable_meta_caching;
376 unsigned long last_keep_alive_jiffies;
378 struct u64_stats_sync syncp;
379 struct ena_stats_dev dev_stats;
380 struct ena_admin_eni_stats eni_stats;
381 bool eni_stats_supported;
383 /* last queue index that was checked for uncompleted tx packets */
384 u32 last_monitored_tx_qid;
386 enum ena_regs_reset_reason_types reset_reason;
388 struct bpf_prog *xdp_bpf_prog;
393 void ena_set_ethtool_ops(struct net_device *netdev);
395 void ena_dump_stats_to_dmesg(struct ena_adapter *adapter);
397 void ena_dump_stats_to_buf(struct ena_adapter *adapter, u8 *buf);
399 int ena_update_hw_stats(struct ena_adapter *adapter);
401 int ena_update_queue_sizes(struct ena_adapter *adapter,
405 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count);
407 int ena_get_sset_count(struct net_device *netdev, int sset);
409 enum ena_xdp_errors_t {
411 ENA_XDP_CURRENT_MTU_TOO_LARGE,
412 ENA_XDP_NO_ENOUGH_QUEUES,
415 static inline bool ena_xdp_present(struct ena_adapter *adapter)
417 return !!adapter->xdp_bpf_prog;
420 static inline bool ena_xdp_present_ring(struct ena_ring *ring)
422 return !!ring->xdp_bpf_prog;
425 static inline bool ena_xdp_legal_queue_count(struct ena_adapter *adapter,
428 return 2 * queues <= adapter->max_num_io_queues;
431 static inline enum ena_xdp_errors_t ena_xdp_allowed(struct ena_adapter *adapter)
433 enum ena_xdp_errors_t rc = ENA_XDP_ALLOWED;
435 if (adapter->netdev->mtu > ENA_XDP_MAX_MTU)
436 rc = ENA_XDP_CURRENT_MTU_TOO_LARGE;
437 else if (!ena_xdp_legal_queue_count(adapter, adapter->num_io_queues))
438 rc = ENA_XDP_NO_ENOUGH_QUEUES;
443 #endif /* !(ENA_H) */