2 * Copyright 2008-2010 Cisco Systems, Inc. All rights reserved.
3 * Copyright 2007 Nuova Systems, Inc. All rights reserved.
5 * Copyright (c) 2014, Cisco Systems, Inc.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in
17 * the documentation and/or other materials provided with the
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31 * POSSIBILITY OF SUCH DAMAGE.
41 #include <rte_memzone.h>
43 /* Work queue control */
45 u64 ring_base; /* 0x00 */
46 u32 ring_size; /* 0x08 */
48 u32 posted_index; /* 0x10 */
50 u32 cq_index; /* 0x18 */
52 u32 enable; /* 0x20 */
54 u32 running; /* 0x28 */
56 u32 fetch_index; /* 0x30 */
58 u32 dca_value; /* 0x38 */
60 u32 error_interrupt_enable; /* 0x40 */
62 u32 error_interrupt_offset; /* 0x48 */
64 u32 error_status; /* 0x50 */
75 struct vnic_dev *vdev;
76 struct vnic_wq_ctrl __iomem *ctrl; /* memory-mapped */
77 struct vnic_dev_ring ring;
78 struct vnic_wq_buf *bufs;
79 unsigned int head_idx;
80 unsigned int tail_idx;
81 unsigned int pkts_outstanding;
82 unsigned int socket_id;
83 const struct rte_memzone *cqmsg_rz;
84 uint16_t last_completed_index;
87 static inline unsigned int vnic_wq_desc_avail(struct vnic_wq *wq)
89 /* how many does SW own? */
90 return wq->ring.desc_avail;
93 static inline unsigned int vnic_wq_desc_used(struct vnic_wq *wq)
95 /* how many does HW own? */
96 return wq->ring.desc_count - wq->ring.desc_avail - 1;
99 #define PI_LOG2_CACHE_LINE_SIZE 5
100 #define PI_INDEX_BITS 12
101 #define PI_INDEX_MASK ((1U << PI_INDEX_BITS) - 1)
102 #define PI_PREFETCH_LEN_MASK ((1U << PI_LOG2_CACHE_LINE_SIZE) - 1)
103 #define PI_PREFETCH_LEN_OFF 16
104 #define PI_PREFETCH_ADDR_BITS 43
105 #define PI_PREFETCH_ADDR_MASK ((1ULL << PI_PREFETCH_ADDR_BITS) - 1)
106 #define PI_PREFETCH_ADDR_OFF 21
108 /** How many cache lines are touched by buffer (addr, len). */
109 static inline unsigned int num_cache_lines_touched(dma_addr_t addr,
112 const unsigned long mask = PI_PREFETCH_LEN_MASK;
113 const unsigned long laddr = (unsigned long)addr;
114 unsigned long lines, equiv_len;
115 /* A. If addr is aligned, our solution is just to round up len to the
118 e.g. addr = 0, len = 48
119 +--------------------+
120 |XXXXXXXXXXXXXXXXXXXX| 32-byte cacheline a
121 +--------------------+
122 |XXXXXXXXXX | cacheline b
123 +--------------------+
125 B. If addr is not aligned, however, we may use an extra
126 cacheline. e.g. addr = 12, len = 22
128 +--------------------+
130 +--------------------+
132 +--------------------+
134 Our solution is to make the problem equivalent to case A
135 above by adding the empty space in the first cacheline to the length:
138 +--------------------+
139 |eeeeeeeXXXXXXXXXXXXX| "e" is empty space, which we add to len
140 +--------------------+
142 +--------------------+
145 equiv_len = len + (laddr & mask);
147 /* Now we can just round up this len to the next 32-byte boundary. */
148 lines = (equiv_len + mask) & (~mask);
150 /* Scale bytes -> cachelines. */
151 return lines >> PI_LOG2_CACHE_LINE_SIZE;
154 static inline u64 vnic_cached_posted_index(dma_addr_t addr, unsigned int len,
157 unsigned int num_cache_lines = num_cache_lines_touched(addr, len);
158 /* Wish we could avoid a branch here. We could have separate
159 * vnic_wq_post() and vinc_wq_post_inline(), the latter
160 * only supporting < 1k (2^5 * 2^5) sends, I suppose. This would
161 * eliminate the if (eop) branch as well.
163 if (num_cache_lines > PI_PREFETCH_LEN_MASK)
165 return (index & PI_INDEX_MASK) |
166 ((num_cache_lines & PI_PREFETCH_LEN_MASK) << PI_PREFETCH_LEN_OFF) |
167 (((addr >> PI_LOG2_CACHE_LINE_SIZE) &
168 PI_PREFETCH_ADDR_MASK) << PI_PREFETCH_ADDR_OFF);
171 static inline uint32_t
172 buf_idx_incr(uint32_t n_descriptors, uint32_t idx)
175 if (unlikely(idx == n_descriptors))
180 void vnic_wq_free(struct vnic_wq *wq);
181 int vnic_wq_alloc(struct vnic_dev *vdev, struct vnic_wq *wq, unsigned int index,
182 unsigned int desc_count, unsigned int desc_size);
183 void vnic_wq_init_start(struct vnic_wq *wq, unsigned int cq_index,
184 unsigned int fetch_index, unsigned int posted_index,
185 unsigned int error_interrupt_enable,
186 unsigned int error_interrupt_offset);
187 void vnic_wq_init(struct vnic_wq *wq, unsigned int cq_index,
188 unsigned int error_interrupt_enable,
189 unsigned int error_interrupt_offset);
190 void vnic_wq_error_out(struct vnic_wq *wq, unsigned int error);
191 unsigned int vnic_wq_error_status(struct vnic_wq *wq);
192 void vnic_wq_enable(struct vnic_wq *wq);
193 int vnic_wq_disable(struct vnic_wq *wq);
194 void vnic_wq_clean(struct vnic_wq *wq,
195 void (*buf_clean)(struct vnic_wq *wq, struct vnic_wq_buf *buf));
196 int vnic_wq_mem_size(struct vnic_wq *wq, unsigned int desc_count,
197 unsigned int desc_size);
199 #endif /* _VNIC_WQ_H_ */