2 * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
37 #include <linux/compiler.h>
38 #include <linux/list.h>
39 #include <linux/mutex.h>
40 #include <linux/netdevice.h>
42 #include <linux/mlx4/device.h>
43 #include <linux/mlx4/qp.h>
44 #include <linux/mlx4/cq.h>
45 #include <linux/mlx4/srq.h>
46 #include <linux/mlx4/doorbell.h>
47 #include <linux/mlx4/cmd.h>
51 #define DRV_NAME "mlx4_en"
52 #define DRV_VERSION "1.5.1.6"
53 #define DRV_RELDATE "August 2010"
55 #define MLX4_EN_MSG_LEVEL (NETIF_MSG_LINK | NETIF_MSG_IFDOWN)
62 #define MLX4_EN_PAGE_SHIFT 12
63 #define MLX4_EN_PAGE_SIZE (1 << MLX4_EN_PAGE_SHIFT)
64 #define MAX_RX_RINGS 16
65 #define MIN_RX_RINGS 4
67 #define HEADROOM (2048 / TXBB_SIZE + 1)
68 #define STAMP_STRIDE 64
69 #define STAMP_DWORDS (STAMP_STRIDE / 4)
70 #define STAMP_SHIFT 31
71 #define STAMP_VAL 0x7fffffff
72 #define STATS_DELAY (HZ / 4)
74 /* Typical TSO descriptor with 16 gather entries is 352 bytes... */
75 #define MAX_DESC_SIZE 512
76 #define MAX_DESC_TXBBS (MAX_DESC_SIZE / TXBB_SIZE)
79 * OS related constants and tunables
82 #define MLX4_EN_WATCHDOG_TIMEOUT (15 * HZ)
84 #define MLX4_EN_ALLOC_ORDER 2
85 #define MLX4_EN_ALLOC_SIZE (PAGE_SIZE << MLX4_EN_ALLOC_ORDER)
87 #define MLX4_EN_MAX_LRO_DESCRIPTORS 32
89 /* Receive fragment sizes; we use at most 4 fragments (for 9600 byte MTU
90 * and 4K allocations) */
92 FRAG_SZ0
= 512 - NET_IP_ALIGN
,
95 FRAG_SZ3
= MLX4_EN_ALLOC_SIZE
97 #define MLX4_EN_MAX_RX_FRAGS 4
99 /* Maximum ring sizes */
100 #define MLX4_EN_MAX_TX_SIZE 8192
101 #define MLX4_EN_MAX_RX_SIZE 8192
103 /* Minimum ring size for our page-allocation sceme to work */
104 #define MLX4_EN_MIN_RX_SIZE (MLX4_EN_ALLOC_SIZE / SMP_CACHE_BYTES)
105 #define MLX4_EN_MIN_TX_SIZE (4096 / TXBB_SIZE)
107 #define MLX4_EN_SMALL_PKT_SIZE 64
108 #define MLX4_EN_NUM_TX_RINGS 8
109 #define MLX4_EN_NUM_PPP_RINGS 8
110 #define MAX_TX_RINGS (MLX4_EN_NUM_TX_RINGS + MLX4_EN_NUM_PPP_RINGS)
111 #define MLX4_EN_DEF_TX_RING_SIZE 512
112 #define MLX4_EN_DEF_RX_RING_SIZE 1024
114 /* Target number of packets to coalesce with interrupt moderation */
115 #define MLX4_EN_RX_COAL_TARGET 44
116 #define MLX4_EN_RX_COAL_TIME 0x10
118 #define MLX4_EN_TX_COAL_PKTS 5
119 #define MLX4_EN_TX_COAL_TIME 0x80
121 #define MLX4_EN_RX_RATE_LOW 400000
122 #define MLX4_EN_RX_COAL_TIME_LOW 0
123 #define MLX4_EN_RX_RATE_HIGH 450000
124 #define MLX4_EN_RX_COAL_TIME_HIGH 128
125 #define MLX4_EN_RX_SIZE_THRESH 1024
126 #define MLX4_EN_RX_RATE_THRESH (1000000 / MLX4_EN_RX_COAL_TIME_HIGH)
127 #define MLX4_EN_SAMPLE_INTERVAL 0
128 #define MLX4_EN_AVG_PKT_SMALL 256
130 #define MLX4_EN_AUTO_CONF 0xffff
132 #define MLX4_EN_DEF_RX_PAUSE 1
133 #define MLX4_EN_DEF_TX_PAUSE 1
135 /* Interval between successive polls in the Tx routine when polling is used
136 instead of interrupts (in per-core Tx rings) - should be power of 2 */
137 #define MLX4_EN_TX_POLL_MODER 16
138 #define MLX4_EN_TX_POLL_TIMEOUT (HZ / 4)
140 #define ETH_LLC_SNAP_SIZE 8
142 #define SMALL_PACKET_SIZE (256 - NET_IP_ALIGN)
143 #define HEADER_COPY_SIZE (128 - NET_IP_ALIGN)
144 #define MLX4_LOOPBACK_TEST_PAYLOAD (HEADER_COPY_SIZE - ETH_HLEN)
146 #define MLX4_EN_MIN_MTU 46
147 #define ETH_BCAST 0xffffffffffffULL
149 #define MLX4_EN_LOOPBACK_RETRIES 5
150 #define MLX4_EN_LOOPBACK_TIMEOUT 100
152 #ifdef MLX4_EN_PERF_STAT
153 /* Number of samples to 'average' */
155 #define AVG_FACTOR 1024
156 #define NUM_PERF_STATS NUM_PERF_COUNTERS
158 #define INC_PERF_COUNTER(cnt) (++(cnt))
159 #define ADD_PERF_COUNTER(cnt, add) ((cnt) += (add))
160 #define AVG_PERF_COUNTER(cnt, sample) \
161 ((cnt) = ((cnt) * (AVG_SIZE - 1) + (sample) * AVG_FACTOR) / AVG_SIZE)
162 #define GET_PERF_COUNTER(cnt) (cnt)
163 #define GET_AVG_PERF_COUNTER(cnt) ((cnt) / AVG_FACTOR)
167 #define NUM_PERF_STATS 0
168 #define INC_PERF_COUNTER(cnt) do {} while (0)
169 #define ADD_PERF_COUNTER(cnt, add) do {} while (0)
170 #define AVG_PERF_COUNTER(cnt, sample) do {} while (0)
171 #define GET_PERF_COUNTER(cnt) (0)
172 #define GET_AVG_PERF_COUNTER(cnt) (0)
173 #endif /* MLX4_EN_PERF_STAT */
188 #define ROUNDUP_LOG2(x) ilog2(roundup_pow_of_two(x))
189 #define XNOR(x, y) (!(x) == !(y))
190 #define ILLEGAL_MAC(addr) (addr == 0xffffffffffffULL || addr == 0x0)
193 struct mlx4_en_tx_info
{
202 #define MLX4_EN_BIT_DESC_OWN 0x80000000
203 #define CTRL_SIZE sizeof(struct mlx4_wqe_ctrl_seg)
204 #define MLX4_EN_MEMTYPE_PAD 0x100
205 #define DS_SIZE sizeof(struct mlx4_wqe_data_seg)
208 struct mlx4_en_tx_desc
{
209 struct mlx4_wqe_ctrl_seg ctrl
;
211 struct mlx4_wqe_data_seg data
; /* at least one data segment */
212 struct mlx4_wqe_lso_seg lso
;
213 struct mlx4_wqe_inline_seg inl
;
217 #define MLX4_EN_USE_SRQ 0x01000000
219 struct mlx4_en_rx_alloc
{
224 struct mlx4_en_tx_ring
{
225 struct mlx4_hwq_resources wqres
;
226 u32 size
; /* number of TXBBs */
229 u16 cqn
; /* index of port CQ associated with this ring */
237 struct mlx4_en_tx_info
*tx_info
;
241 struct mlx4_qp_context context
;
243 enum mlx4_qp_state qp_state
;
244 struct mlx4_srq dummy
;
246 unsigned long packets
;
247 spinlock_t comp_lock
;
250 struct mlx4_en_rx_desc
{
251 /* actual number of entries depends on rx ring stride */
252 struct mlx4_wqe_data_seg data
[0];
255 struct mlx4_en_rx_ring
{
256 struct mlx4_hwq_resources wqres
;
257 struct mlx4_en_rx_alloc page_alloc
[MLX4_EN_MAX_RX_FRAGS
];
258 u32 size
; /* number of Rx descs*/
263 u16 cqn
; /* index of port CQ associated with this ring */
270 unsigned long packets
;
274 static inline int mlx4_en_can_lro(__be16 status
)
276 return (status
& cpu_to_be16(MLX4_CQE_STATUS_IPV4
|
277 MLX4_CQE_STATUS_IPV4F
|
278 MLX4_CQE_STATUS_IPV6
|
279 MLX4_CQE_STATUS_IPV4OPT
|
280 MLX4_CQE_STATUS_TCP
|
281 MLX4_CQE_STATUS_UDP
|
282 MLX4_CQE_STATUS_IPOK
)) ==
283 cpu_to_be16(MLX4_CQE_STATUS_IPV4
|
284 MLX4_CQE_STATUS_IPOK
|
285 MLX4_CQE_STATUS_TCP
);
290 struct mlx4_hwq_resources wqres
;
293 struct net_device
*dev
;
294 struct napi_struct napi
;
295 /* Per-core Tx cq processing support */
296 struct timer_list timer
;
303 struct mlx4_cqe
*buf
;
304 #define MLX4_EN_OPCODE_ERROR 0x1e
307 struct mlx4_en_port_profile
{
319 struct mlx4_en_profile
{
327 struct mlx4_en_port_profile prof
[MLX4_MAX_PORTS
+ 1];
331 struct mlx4_dev
*dev
;
332 struct pci_dev
*pdev
;
333 struct mutex state_lock
;
334 struct net_device
*pndev
[MLX4_MAX_PORTS
+ 1];
337 struct mlx4_en_profile profile
;
339 struct workqueue_struct
*workqueue
;
340 struct device
*dma_device
;
341 void __iomem
*uar_map
;
342 struct mlx4_uar priv_uar
;
346 u8 mac_removed
[MLX4_MAX_PORTS
+ 1];
350 struct mlx4_en_rss_map
{
352 struct mlx4_qp qps
[MAX_RX_RINGS
];
353 enum mlx4_qp_state state
[MAX_RX_RINGS
];
354 struct mlx4_qp indir_qp
;
355 enum mlx4_qp_state indir_state
;
358 struct mlx4_en_rss_context
{
368 struct mlx4_en_port_state
{
374 struct mlx4_en_pkt_stats
{
375 unsigned long broadcast
;
376 unsigned long rx_prio
[8];
377 unsigned long tx_prio
[8];
378 #define NUM_PKT_STATS 17
381 struct mlx4_en_port_stats
{
382 unsigned long tso_packets
;
383 unsigned long queue_stopped
;
384 unsigned long wake_queue
;
385 unsigned long tx_timeout
;
386 unsigned long rx_alloc_failed
;
387 unsigned long rx_chksum_good
;
388 unsigned long rx_chksum_none
;
389 unsigned long tx_chksum_offload
;
390 #define NUM_PORT_STATS 8
393 struct mlx4_en_perf_stats
{
400 #define NUM_PERF_COUNTERS 6
403 struct mlx4_en_frag_info
{
405 u16 frag_prefix_size
;
412 struct mlx4_en_priv
{
413 struct mlx4_en_dev
*mdev
;
414 struct mlx4_en_port_profile
*prof
;
415 struct net_device
*dev
;
416 struct vlan_group
*vlgrp
;
417 struct net_device_stats stats
;
418 struct net_device_stats ret_stats
;
419 struct mlx4_en_port_state port_state
;
420 spinlock_t stats_lock
;
422 unsigned long last_moder_packets
;
423 unsigned long last_moder_tx_packets
;
424 unsigned long last_moder_bytes
;
425 unsigned long last_moder_jiffies
;
436 u16 adaptive_rx_coal
;
439 u32 validate_loopback
;
441 struct mlx4_hwq_resources res
;
455 struct mlx4_en_rss_map rss_map
;
457 #define MLX4_EN_FLAG_PROMISC 0x1
461 struct mlx4_en_frag_info frag_info
[MLX4_EN_MAX_RX_FRAGS
];
465 struct mlx4_en_tx_ring tx_ring
[MAX_TX_RINGS
];
467 struct mlx4_en_rx_ring rx_ring
[MAX_RX_RINGS
];
468 struct mlx4_en_cq tx_cq
[MAX_TX_RINGS
];
469 struct mlx4_en_cq rx_cq
[MAX_RX_RINGS
];
470 struct work_struct mcast_task
;
471 struct work_struct mac_task
;
472 struct work_struct watchdog_task
;
473 struct work_struct linkstate_task
;
474 struct delayed_work stats_task
;
475 struct mlx4_en_perf_stats pstats
;
476 struct mlx4_en_pkt_stats pkstats
;
477 struct mlx4_en_port_stats port_stats
;
480 struct mlx4_en_stat_out_mbox hw_stats
;
485 void mlx4_en_destroy_netdev(struct net_device
*dev
);
486 int mlx4_en_init_netdev(struct mlx4_en_dev
*mdev
, int port
,
487 struct mlx4_en_port_profile
*prof
);
489 int mlx4_en_start_port(struct net_device
*dev
);
490 void mlx4_en_stop_port(struct net_device
*dev
);
492 void mlx4_en_free_resources(struct mlx4_en_priv
*priv
, bool reserve_vectors
);
493 int mlx4_en_alloc_resources(struct mlx4_en_priv
*priv
);
495 int mlx4_en_create_cq(struct mlx4_en_priv
*priv
, struct mlx4_en_cq
*cq
,
496 int entries
, int ring
, enum cq_type mode
);
497 void mlx4_en_destroy_cq(struct mlx4_en_priv
*priv
, struct mlx4_en_cq
*cq
,
498 bool reserve_vectors
);
499 int mlx4_en_activate_cq(struct mlx4_en_priv
*priv
, struct mlx4_en_cq
*cq
);
500 void mlx4_en_deactivate_cq(struct mlx4_en_priv
*priv
, struct mlx4_en_cq
*cq
);
501 int mlx4_en_set_cq_moder(struct mlx4_en_priv
*priv
, struct mlx4_en_cq
*cq
);
502 int mlx4_en_arm_cq(struct mlx4_en_priv
*priv
, struct mlx4_en_cq
*cq
);
504 void mlx4_en_poll_tx_cq(unsigned long data
);
505 void mlx4_en_tx_irq(struct mlx4_cq
*mcq
);
506 u16
mlx4_en_select_queue(struct net_device
*dev
, struct sk_buff
*skb
);
507 netdev_tx_t
mlx4_en_xmit(struct sk_buff
*skb
, struct net_device
*dev
);
509 int mlx4_en_create_tx_ring(struct mlx4_en_priv
*priv
, struct mlx4_en_tx_ring
*ring
,
510 u32 size
, u16 stride
);
511 void mlx4_en_destroy_tx_ring(struct mlx4_en_priv
*priv
, struct mlx4_en_tx_ring
*ring
);
512 int mlx4_en_activate_tx_ring(struct mlx4_en_priv
*priv
,
513 struct mlx4_en_tx_ring
*ring
,
515 void mlx4_en_deactivate_tx_ring(struct mlx4_en_priv
*priv
,
516 struct mlx4_en_tx_ring
*ring
);
518 int mlx4_en_create_rx_ring(struct mlx4_en_priv
*priv
,
519 struct mlx4_en_rx_ring
*ring
,
520 u32 size
, u16 stride
);
521 void mlx4_en_destroy_rx_ring(struct mlx4_en_priv
*priv
,
522 struct mlx4_en_rx_ring
*ring
);
523 int mlx4_en_activate_rx_rings(struct mlx4_en_priv
*priv
);
524 void mlx4_en_deactivate_rx_ring(struct mlx4_en_priv
*priv
,
525 struct mlx4_en_rx_ring
*ring
);
526 int mlx4_en_process_rx_cq(struct net_device
*dev
,
527 struct mlx4_en_cq
*cq
,
529 int mlx4_en_poll_rx_cq(struct napi_struct
*napi
, int budget
);
530 void mlx4_en_fill_qp_context(struct mlx4_en_priv
*priv
, int size
, int stride
,
531 int is_tx
, int rss
, int qpn
, int cqn
,
532 struct mlx4_qp_context
*context
);
533 void mlx4_en_sqp_event(struct mlx4_qp
*qp
, enum mlx4_event event
);
534 int mlx4_en_map_buffer(struct mlx4_buf
*buf
);
535 void mlx4_en_unmap_buffer(struct mlx4_buf
*buf
);
537 void mlx4_en_calc_rx_buf(struct net_device
*dev
);
538 int mlx4_en_config_rss_steer(struct mlx4_en_priv
*priv
);
539 void mlx4_en_release_rss_steer(struct mlx4_en_priv
*priv
);
540 int mlx4_en_free_tx_buf(struct net_device
*dev
, struct mlx4_en_tx_ring
*ring
);
541 void mlx4_en_rx_irq(struct mlx4_cq
*mcq
);
543 int mlx4_SET_MCAST_FLTR(struct mlx4_dev
*dev
, u8 port
, u64 mac
, u64 clear
, u8 mode
);
544 int mlx4_SET_VLAN_FLTR(struct mlx4_dev
*dev
, u8 port
, struct vlan_group
*grp
);
545 int mlx4_SET_PORT_general(struct mlx4_dev
*dev
, u8 port
, int mtu
,
546 u8 pptx
, u8 pfctx
, u8 pprx
, u8 pfcrx
);
547 int mlx4_SET_PORT_qpn_calc(struct mlx4_dev
*dev
, u8 port
, u32 base_qpn
,
550 int mlx4_en_DUMP_ETH_STATS(struct mlx4_en_dev
*mdev
, u8 port
, u8 reset
);
551 int mlx4_en_QUERY_PORT(struct mlx4_en_dev
*mdev
, u8 port
);
553 #define MLX4_EN_NUM_SELF_TEST 5
554 void mlx4_en_ex_selftest(struct net_device
*dev
, u32
*flags
, u64
*buf
);
555 u64
mlx4_en_mac_to_u64(u8
*addr
);
560 extern const struct ethtool_ops mlx4_en_ethtool_ops
;
565 * printk / logging functions
568 int en_print(const char *level
, const struct mlx4_en_priv
*priv
,
569 const char *format
, ...) __attribute__ ((format (printf
, 3, 4)));
571 #define en_dbg(mlevel, priv, format, arg...) \
573 if (NETIF_MSG_##mlevel & priv->msg_enable) \
574 en_print(KERN_DEBUG, priv, format, ##arg); \
576 #define en_warn(priv, format, arg...) \
577 en_print(KERN_WARNING, priv, format, ##arg)
578 #define en_err(priv, format, arg...) \
579 en_print(KERN_ERR, priv, format, ##arg)
580 #define en_info(priv, format, arg...) \
581 en_print(KERN_INFO, priv, format, ## arg)
583 #define mlx4_err(mdev, format, arg...) \
584 pr_err("%s %s: " format, DRV_NAME, \
585 dev_name(&mdev->pdev->dev), ##arg)
586 #define mlx4_info(mdev, format, arg...) \
587 pr_info("%s %s: " format, DRV_NAME, \
588 dev_name(&mdev->pdev->dev), ##arg)
589 #define mlx4_warn(mdev, format, arg...) \
590 pr_warning("%s %s: " format, DRV_NAME, \
591 dev_name(&mdev->pdev->dev), ##arg)