1 #ifndef _HFI1_USER_SDMA_H
2 #define _HFI1_USER_SDMA_H
3 /*
4  * Copyright(c) 2015 - 2018 Intel Corporation.
5  *
6  * This file is provided under a dual BSD/GPLv2 license.  When using or
7  * redistributing this file, you may do so under either license.
8  *
9  * GPL LICENSE SUMMARY
10  *
11  * This program is free software; you can redistribute it and/or modify
12  * it under the terms of version 2 of the GNU General Public License as
13  * published by the Free Software Foundation.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * BSD LICENSE
21  *
22  * Redistribution and use in source and binary forms, with or without
23  * modification, are permitted provided that the following conditions
24  * are met:
25  *
26  *  - Redistributions of source code must retain the above copyright
27  *    notice, this list of conditions and the following disclaimer.
28  *  - Redistributions in binary form must reproduce the above copyright
29  *    notice, this list of conditions and the following disclaimer in
30  *    the documentation and/or other materials provided with the
31  *    distribution.
32  *  - Neither the name of Intel Corporation nor the names of its
33  *    contributors may be used to endorse or promote products derived
34  *    from this software without specific prior written permission.
35  *
36  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
37  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
38  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
39  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
40  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
41  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
42  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
43  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
44  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
45  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
46  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
47  *
48  */
49 #include <linux/device.h>
50 #include <linux/wait.h>
51 
52 #include "common.h"
53 #include "iowait.h"
54 #include "user_exp_rcv.h"
55 
56 /* The maximum number of Data io vectors per message/request */
57 #define MAX_VECTORS_PER_REQ 8
58 /*
59  * Maximum number of packet to send from each message/request
60  * before moving to the next one.
61  */
62 #define MAX_PKTS_PER_QUEUE 16
63 
64 #define num_pages(x) (1 + ((((x) - 1) & PAGE_MASK) >> PAGE_SHIFT))
65 
66 #define req_opcode(x) \
67 	(((x) >> HFI1_SDMA_REQ_OPCODE_SHIFT) & HFI1_SDMA_REQ_OPCODE_MASK)
68 #define req_version(x) \
69 	(((x) >> HFI1_SDMA_REQ_VERSION_SHIFT) & HFI1_SDMA_REQ_OPCODE_MASK)
70 #define req_iovcnt(x) \
71 	(((x) >> HFI1_SDMA_REQ_IOVCNT_SHIFT) & HFI1_SDMA_REQ_IOVCNT_MASK)
72 
73 /* Number of BTH.PSN bits used for sequence number in expected rcvs */
74 #define BTH_SEQ_MASK 0x7ffull
75 
76 #define AHG_KDETH_INTR_SHIFT 12
77 #define AHG_KDETH_SH_SHIFT   13
78 #define AHG_KDETH_ARRAY_SIZE  9
79 
80 #define PBC2LRH(x) ((((x) & 0xfff) << 2) - 4)
81 #define LRH2PBC(x) ((((x) >> 2) + 1) & 0xfff)
82 
83 /**
84  * Build an SDMA AHG header update descriptor and save it to an array.
85  * @arr        - Array to save the descriptor to.
86  * @idx        - Index of the array at which the descriptor will be saved.
87  * @array_size - Size of the array arr.
88  * @dw         - Update index into the header in DWs.
89  * @bit        - Start bit.
90  * @width      - Field width.
91  * @value      - 16 bits of immediate data to write into the field.
92  * Returns -ERANGE if idx is invalid. If successful, returns the next index
93  * (idx + 1) of the array to be used for the next descriptor.
94  */
95 static inline int ahg_header_set(u32 *arr, int idx, size_t array_size,
96 				 u8 dw, u8 bit, u8 width, u16 value)
97 {
98 	if ((size_t)idx >= array_size)
99 		return -ERANGE;
100 	arr[idx++] = sdma_build_ahg_descriptor(value, dw, bit, width);
101 	return idx;
102 }
103 
104 /* Tx request flag bits */
105 #define TXREQ_FLAGS_REQ_ACK   BIT(0)      /* Set the ACK bit in the header */
106 #define TXREQ_FLAGS_REQ_DISABLE_SH BIT(1) /* Disable header suppression */
107 
108 #define SDMA_PKT_Q_INACTIVE BIT(0)
109 #define SDMA_PKT_Q_ACTIVE   BIT(1)
110 #define SDMA_PKT_Q_DEFERRED BIT(2)
111 
112 /*
113  * Maximum retry attempts to submit a TX request
114  * before putting the process to sleep.
115  */
116 #define MAX_DEFER_RETRY_COUNT 1
117 
118 #define SDMA_IOWAIT_TIMEOUT 1000 /* in milliseconds */
119 
120 #define SDMA_DBG(req, fmt, ...)				     \
121 	hfi1_cdbg(SDMA, "[%u:%u:%u:%u] " fmt, (req)->pq->dd->unit, \
122 		 (req)->pq->ctxt, (req)->pq->subctxt, (req)->info.comp_idx, \
123 		 ##__VA_ARGS__)
124 
125 struct hfi1_user_sdma_pkt_q {
126 	u16 ctxt;
127 	u16 subctxt;
128 	u16 n_max_reqs;
129 	atomic_t n_reqs;
130 	u16 reqidx;
131 	struct hfi1_devdata *dd;
132 	struct kmem_cache *txreq_cache;
133 	struct user_sdma_request *reqs;
134 	unsigned long *req_in_use;
135 	struct iowait busy;
136 	unsigned state;
137 	wait_queue_head_t wait;
138 	unsigned long unpinned;
139 	struct mmu_rb_handler *handler;
140 	atomic_t n_locked;
141 	struct mm_struct *mm;
142 };
143 
144 struct hfi1_user_sdma_comp_q {
145 	u16 nentries;
146 	struct hfi1_sdma_comp_entry *comps;
147 };
148 
149 struct sdma_mmu_node {
150 	struct mmu_rb_node rb;
151 	struct hfi1_user_sdma_pkt_q *pq;
152 	atomic_t refcount;
153 	struct page **pages;
154 	unsigned int npages;
155 };
156 
157 struct user_sdma_iovec {
158 	struct list_head list;
159 	struct iovec iov;
160 	/* number of pages in this vector */
161 	unsigned int npages;
162 	/* array of pinned pages for this vector */
163 	struct page **pages;
164 	/*
165 	 * offset into the virtual address space of the vector at
166 	 * which we last left off.
167 	 */
168 	u64 offset;
169 	struct sdma_mmu_node *node;
170 };
171 
172 /* evict operation argument */
173 struct evict_data {
174 	u32 cleared;	/* count evicted so far */
175 	u32 target;	/* target count to evict */
176 };
177 
178 struct user_sdma_request {
179 	/* This is the original header from user space */
180 	struct hfi1_pkt_header hdr;
181 
182 	/* Read mostly fields */
183 	struct hfi1_user_sdma_pkt_q *pq ____cacheline_aligned_in_smp;
184 	struct hfi1_user_sdma_comp_q *cq;
185 	/*
186 	 * Pointer to the SDMA engine for this request.
187 	 * Since different request could be on different VLs,
188 	 * each request will need it's own engine pointer.
189 	 */
190 	struct sdma_engine *sde;
191 	struct sdma_req_info info;
192 	/* TID array values copied from the tid_iov vector */
193 	u32 *tids;
194 	/* total length of the data in the request */
195 	u32 data_len;
196 	/* number of elements copied to the tids array */
197 	u16 n_tids;
198 	/*
199 	 * We copy the iovs for this request (based on
200 	 * info.iovcnt). These are only the data vectors
201 	 */
202 	u8 data_iovs;
203 	s8 ahg_idx;
204 
205 	/* Writeable fields shared with interrupt */
206 	u64 seqcomp ____cacheline_aligned_in_smp;
207 	u64 seqsubmitted;
208 	/* status of the last txreq completed */
209 	int status;
210 
211 	/* Send side fields */
212 	struct list_head txps ____cacheline_aligned_in_smp;
213 	u64 seqnum;
214 	/*
215 	 * KDETH.OFFSET (TID) field
216 	 * The offset can cover multiple packets, depending on the
217 	 * size of the TID entry.
218 	 */
219 	u32 tidoffset;
220 	/*
221 	 * KDETH.Offset (Eager) field
222 	 * We need to remember the initial value so the headers
223 	 * can be updated properly.
224 	 */
225 	u32 koffset;
226 	u32 sent;
227 	/* TID index copied from the tid_iov vector */
228 	u16 tididx;
229 	/* progress index moving along the iovs array */
230 	u8 iov_idx;
231 	u8 done;
232 	u8 has_error;
233 
234 	struct user_sdma_iovec iovs[MAX_VECTORS_PER_REQ];
235 } ____cacheline_aligned_in_smp;
236 
237 /*
238  * A single txreq could span up to 3 physical pages when the MTU
239  * is sufficiently large (> 4K). Each of the IOV pointers also
240  * needs it's own set of flags so the vector has been handled
241  * independently of each other.
242  */
243 struct user_sdma_txreq {
244 	/* Packet header for the txreq */
245 	struct hfi1_pkt_header hdr;
246 	struct sdma_txreq txreq;
247 	struct list_head list;
248 	struct user_sdma_request *req;
249 	u16 flags;
250 	unsigned int busycount;
251 	u64 seqnum;
252 };
253 
254 int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
255 				struct hfi1_filedata *fd);
256 int hfi1_user_sdma_free_queues(struct hfi1_filedata *fd,
257 			       struct hfi1_ctxtdata *uctxt);
258 int hfi1_user_sdma_process_request(struct hfi1_filedata *fd,
259 				   struct iovec *iovec, unsigned long dim,
260 				   unsigned long *count);
261 
262 #endif /* _HFI1_USER_SDMA_H */
263