xref: /linux/drivers/infiniband/hw/hfi1/pio.h (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
1 #ifndef _PIO_H
2 #define _PIO_H
3 /*
4  * Copyright(c) 2015-2017 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 
50 /* send context types */
51 #define SC_KERNEL 0
52 #define SC_VL15   1
53 #define SC_ACK    2
54 #define SC_USER   3	/* must be the last one: it may take all left */
55 #define SC_MAX    4	/* count of send context types */
56 
57 /*
58  * SC_VNIC types are allocated (dynamically) from the user context pool,
59  * (SC_USER) and used by kernel driver as kernel contexts (SC_KERNEL).
60  */
61 #define SC_VNIC   SC_MAX
62 
63 /* invalid send context index */
64 #define INVALID_SCI 0xff
65 
66 /* PIO buffer release callback function */
67 typedef void (*pio_release_cb)(void *arg, int code);
68 
69 /* PIO release codes - in bits, as there could more than one that apply */
70 #define PRC_OK		0	/* no known error */
71 #define PRC_STATUS_ERR	0x01	/* credit return due to status error */
72 #define PRC_PBC		0x02	/* credit return due to PBC */
73 #define PRC_THRESHOLD	0x04	/* credit return due to threshold */
74 #define PRC_FILL_ERR	0x08	/* credit return due fill error */
75 #define PRC_FORCE	0x10	/* credit return due credit force */
76 #define PRC_SC_DISABLE	0x20	/* clean-up after a context disable */
77 
78 /* byte helper */
79 union mix {
80 	u64 val64;
81 	u32 val32[2];
82 	u8  val8[8];
83 };
84 
85 /* an allocated PIO buffer */
86 struct pio_buf {
87 	struct send_context *sc;/* back pointer to owning send context */
88 	pio_release_cb cb;	/* called when the buffer is released */
89 	void *arg;		/* argument for cb */
90 	void __iomem *start;	/* buffer start address */
91 	void __iomem *end;	/* context end address */
92 	unsigned long sent_at;	/* buffer is sent when <= free */
93 	union mix carry;	/* pending unwritten bytes */
94 	u16 qw_written;		/* QW written so far */
95 	u8 carry_bytes;	/* number of valid bytes in carry */
96 };
97 
98 /* cache line aligned pio buffer array */
99 union pio_shadow_ring {
100 	struct pio_buf pbuf;
101 } ____cacheline_aligned;
102 
103 /* per-NUMA send context */
104 struct send_context {
105 	/* read-only after init */
106 	struct hfi1_devdata *dd;		/* device */
107 	union pio_shadow_ring *sr;	/* shadow ring */
108 	void __iomem *base_addr;	/* start of PIO memory */
109 	u32 __percpu *buffers_allocated;/* count of buffers allocated */
110 	u32 size;			/* context size, in bytes */
111 
112 	int node;			/* context home node */
113 	u32 sr_size;			/* size of the shadow ring */
114 	u16 flags;			/* flags */
115 	u8  type;			/* context type */
116 	u8  sw_index;			/* software index number */
117 	u8  hw_context;			/* hardware context number */
118 	u8  group;			/* credit return group */
119 
120 	/* allocator fields */
121 	spinlock_t alloc_lock ____cacheline_aligned_in_smp;
122 	u32 sr_head;			/* shadow ring head */
123 	unsigned long fill;		/* official alloc count */
124 	unsigned long alloc_free;	/* copy of free (less cache thrash) */
125 	u32 fill_wrap;			/* tracks fill within ring */
126 	u32 credits;			/* number of blocks in context */
127 	/* adding a new field here would make it part of this cacheline */
128 
129 	/* releaser fields */
130 	spinlock_t release_lock ____cacheline_aligned_in_smp;
131 	u32 sr_tail;			/* shadow ring tail */
132 	unsigned long free;		/* official free count */
133 	volatile __le64 *hw_free;	/* HW free counter */
134 	/* list for PIO waiters */
135 	struct list_head piowait  ____cacheline_aligned_in_smp;
136 	spinlock_t credit_ctrl_lock ____cacheline_aligned_in_smp;
137 	u32 credit_intr_count;		/* count of credit intr users */
138 	u64 credit_ctrl;		/* cache for credit control */
139 	wait_queue_head_t halt_wait;    /* wait until kernel sees interrupt */
140 	struct work_struct halt_work;	/* halted context work queue entry */
141 };
142 
143 /* send context flags */
144 #define SCF_ENABLED 0x01
145 #define SCF_IN_FREE 0x02
146 #define SCF_HALTED  0x04
147 #define SCF_FROZEN  0x08
148 
149 struct send_context_info {
150 	struct send_context *sc;	/* allocated working context */
151 	u16 allocated;			/* has this been allocated? */
152 	u16 type;			/* context type */
153 	u16 base;			/* base in PIO array */
154 	u16 credits;			/* size in PIO array */
155 };
156 
157 /* DMA credit return, index is always (context & 0x7) */
158 struct credit_return {
159 	volatile __le64 cr[8];
160 };
161 
162 /* NUMA indexed credit return array */
163 struct credit_return_base {
164 	struct credit_return *va;
165 	dma_addr_t dma;
166 };
167 
168 /* send context configuration sizes (one per type) */
169 struct sc_config_sizes {
170 	short int size;
171 	short int count;
172 };
173 
174 /*
175  * The diagram below details the relationship of the mapping structures
176  *
177  * Since the mapping now allows for non-uniform send contexts per vl, the
178  * number of send contexts for a vl is either the vl_scontexts[vl] or
179  * a computation based on num_kernel_send_contexts/num_vls:
180  *
181  * For example:
182  * nactual = vl_scontexts ? vl_scontexts[vl] : num_kernel_send_contexts/num_vls
183  *
184  * n = roundup to next highest power of 2 using nactual
185  *
186  * In the case where there are num_kernel_send_contexts/num_vls doesn't divide
187  * evenly, the extras are added from the last vl downward.
188  *
189  * For the case where n > nactual, the send contexts are assigned
190  * in a round robin fashion wrapping back to the first send context
191  * for a particular vl.
192  *
193  *               dd->pio_map
194  *                    |                                   pio_map_elem[0]
195  *                    |                                +--------------------+
196  *                    v                                |       mask         |
197  *               pio_vl_map                            |--------------------|
198  *      +--------------------------+                   | ksc[0] -> sc 1     |
199  *      |    list (RCU)            |                   |--------------------|
200  *      |--------------------------|                 ->| ksc[1] -> sc 2     |
201  *      |    mask                  |              --/  |--------------------|
202  *      |--------------------------|            -/     |        *           |
203  *      |    actual_vls (max 8)    |          -/       |--------------------|
204  *      |--------------------------|       --/         | ksc[n-1] -> sc n   |
205  *      |    vls (max 8)           |     -/            +--------------------+
206  *      |--------------------------|  --/
207  *      |    map[0]                |-/
208  *      |--------------------------|                   +--------------------+
209  *      |    map[1]                |---                |       mask         |
210  *      |--------------------------|   \----           |--------------------|
211  *      |           *              |        \--        | ksc[0] -> sc 1+n   |
212  *      |           *              |           \----   |--------------------|
213  *      |           *              |                \->| ksc[1] -> sc 2+n   |
214  *      |--------------------------|                   |--------------------|
215  *      |   map[vls - 1]           |-                  |         *          |
216  *      +--------------------------+ \-                |--------------------|
217  *                                     \-              | ksc[m-1] -> sc m+n |
218  *                                       \             +--------------------+
219  *                                        \-
220  *                                          \
221  *                                           \-        +----------------------+
222  *                                             \-      |       mask           |
223  *                                               \     |----------------------|
224  *                                                \-   | ksc[0] -> sc 1+m+n   |
225  *                                                  \- |----------------------|
226  *                                                    >| ksc[1] -> sc 2+m+n   |
227  *                                                     |----------------------|
228  *                                                     |         *            |
229  *                                                     |----------------------|
230  *                                                     | ksc[o-1] -> sc o+m+n |
231  *                                                     +----------------------+
232  *
233  */
234 
235 /* Initial number of send contexts per VL */
236 #define INIT_SC_PER_VL 2
237 
238 /*
239  * struct pio_map_elem - mapping for a vl
240  * @mask - selector mask
241  * @ksc - array of kernel send contexts for this vl
242  *
243  * The mask is used to "mod" the selector to
244  * produce index into the trailing array of
245  * kscs
246  */
247 struct pio_map_elem {
248 	u32 mask;
249 	struct send_context *ksc[0];
250 };
251 
252 /*
253  * struct pio_vl_map - mapping for a vl
254  * @list - rcu head for free callback
255  * @mask - vl mask to "mod" the vl to produce an index to map array
256  * @actual_vls - number of vls
257  * @vls - numbers of vls rounded to next power of 2
258  * @map - array of pio_map_elem entries
259  *
260  * This is the parent mapping structure. The trailing members of the
261  * struct point to pio_map_elem entries, which in turn point to an
262  * array of kscs for that vl.
263  */
264 struct pio_vl_map {
265 	struct rcu_head list;
266 	u32 mask;
267 	u8 actual_vls;
268 	u8 vls;
269 	struct pio_map_elem *map[0];
270 };
271 
272 int pio_map_init(struct hfi1_devdata *dd, u8 port, u8 num_vls,
273 		 u8 *vl_scontexts);
274 void free_pio_map(struct hfi1_devdata *dd);
275 struct send_context *pio_select_send_context_vl(struct hfi1_devdata *dd,
276 						u32 selector, u8 vl);
277 struct send_context *pio_select_send_context_sc(struct hfi1_devdata *dd,
278 						u32 selector, u8 sc5);
279 
280 /* send context functions */
281 int init_credit_return(struct hfi1_devdata *dd);
282 void free_credit_return(struct hfi1_devdata *dd);
283 int init_sc_pools_and_sizes(struct hfi1_devdata *dd);
284 int init_send_contexts(struct hfi1_devdata *dd);
285 int init_credit_return(struct hfi1_devdata *dd);
286 int init_pervl_scs(struct hfi1_devdata *dd);
287 struct send_context *sc_alloc(struct hfi1_devdata *dd, int type,
288 			      uint hdrqentsize, int numa);
289 void sc_free(struct send_context *sc);
290 int sc_enable(struct send_context *sc);
291 void sc_disable(struct send_context *sc);
292 int sc_restart(struct send_context *sc);
293 void sc_return_credits(struct send_context *sc);
294 void sc_flush(struct send_context *sc);
295 void sc_drop(struct send_context *sc);
296 void sc_stop(struct send_context *sc, int bit);
297 struct pio_buf *sc_buffer_alloc(struct send_context *sc, u32 dw_len,
298 				pio_release_cb cb, void *arg);
299 void sc_release_update(struct send_context *sc);
300 void sc_return_credits(struct send_context *sc);
301 void sc_group_release_update(struct hfi1_devdata *dd, u32 hw_context);
302 void sc_add_credit_return_intr(struct send_context *sc);
303 void sc_del_credit_return_intr(struct send_context *sc);
304 void sc_set_cr_threshold(struct send_context *sc, u32 new_threshold);
305 u32 sc_percent_to_threshold(struct send_context *sc, u32 percent);
306 u32 sc_mtu_to_threshold(struct send_context *sc, u32 mtu, u32 hdrqentsize);
307 void hfi1_sc_wantpiobuf_intr(struct send_context *sc, u32 needint);
308 void sc_wait(struct hfi1_devdata *dd);
309 void set_pio_integrity(struct send_context *sc);
310 
311 /* support functions */
312 void pio_reset_all(struct hfi1_devdata *dd);
313 void pio_freeze(struct hfi1_devdata *dd);
314 void pio_kernel_unfreeze(struct hfi1_devdata *dd);
315 
316 /* global PIO send control operations */
317 #define PSC_GLOBAL_ENABLE 0
318 #define PSC_GLOBAL_DISABLE 1
319 #define PSC_GLOBAL_VLARB_ENABLE 2
320 #define PSC_GLOBAL_VLARB_DISABLE 3
321 #define PSC_CM_RESET 4
322 #define PSC_DATA_VL_ENABLE 5
323 #define PSC_DATA_VL_DISABLE 6
324 
325 void __cm_reset(struct hfi1_devdata *dd, u64 sendctrl);
326 void pio_send_control(struct hfi1_devdata *dd, int op);
327 
328 /* PIO copy routines */
329 void pio_copy(struct hfi1_devdata *dd, struct pio_buf *pbuf, u64 pbc,
330 	      const void *from, size_t count);
331 void seg_pio_copy_start(struct pio_buf *pbuf, u64 pbc,
332 			const void *from, size_t nbytes);
333 void seg_pio_copy_mid(struct pio_buf *pbuf, const void *from, size_t nbytes);
334 void seg_pio_copy_end(struct pio_buf *pbuf);
335 
336 #endif /* _PIO_H */
337