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 /* invalid send context index */
58 #define INVALID_SCI 0xff
59 
60 /* PIO buffer release callback function */
61 typedef void (*pio_release_cb)(void *arg, int code);
62 
63 /* PIO release codes - in bits, as there could more than one that apply */
64 #define PRC_OK		0	/* no known error */
65 #define PRC_STATUS_ERR	0x01	/* credit return due to status error */
66 #define PRC_PBC		0x02	/* credit return due to PBC */
67 #define PRC_THRESHOLD	0x04	/* credit return due to threshold */
68 #define PRC_FILL_ERR	0x08	/* credit return due fill error */
69 #define PRC_FORCE	0x10	/* credit return due credit force */
70 #define PRC_SC_DISABLE	0x20	/* clean-up after a context disable */
71 
72 /* byte helper */
73 union mix {
74 	u64 val64;
75 	u32 val32[2];
76 	u8  val8[8];
77 };
78 
79 /* an allocated PIO buffer */
80 struct pio_buf {
81 	struct send_context *sc;/* back pointer to owning send context */
82 	pio_release_cb cb;	/* called when the buffer is released */
83 	void *arg;		/* argument for cb */
84 	void __iomem *start;	/* buffer start address */
85 	void __iomem *end;	/* context end address */
86 	unsigned long sent_at;	/* buffer is sent when <= free */
87 	union mix carry;	/* pending unwritten bytes */
88 	u16 qw_written;		/* QW written so far */
89 	u8 carry_bytes;	/* number of valid bytes in carry */
90 };
91 
92 /* cache line aligned pio buffer array */
93 union pio_shadow_ring {
94 	struct pio_buf pbuf;
95 } ____cacheline_aligned;
96 
97 /* per-NUMA send context */
98 struct send_context {
99 	/* read-only after init */
100 	struct hfi1_devdata *dd;		/* device */
101 	union pio_shadow_ring *sr;	/* shadow ring */
102 	void __iomem *base_addr;	/* start of PIO memory */
103 	u32 __percpu *buffers_allocated;/* count of buffers allocated */
104 	u32 size;			/* context size, in bytes */
105 
106 	int node;			/* context home node */
107 	u32 sr_size;			/* size of the shadow ring */
108 	u16 flags;			/* flags */
109 	u8  type;			/* context type */
110 	u8  sw_index;			/* software index number */
111 	u8  hw_context;			/* hardware context number */
112 	u8  group;			/* credit return group */
113 
114 	/* allocator fields */
115 	spinlock_t alloc_lock ____cacheline_aligned_in_smp;
116 	u32 sr_head;			/* shadow ring head */
117 	unsigned long fill;		/* official alloc count */
118 	unsigned long alloc_free;	/* copy of free (less cache thrash) */
119 	u32 fill_wrap;			/* tracks fill within ring */
120 	u32 credits;			/* number of blocks in context */
121 	/* adding a new field here would make it part of this cacheline */
122 
123 	/* releaser fields */
124 	spinlock_t release_lock ____cacheline_aligned_in_smp;
125 	u32 sr_tail;			/* shadow ring tail */
126 	unsigned long free;		/* official free count */
127 	volatile __le64 *hw_free;	/* HW free counter */
128 	/* list for PIO waiters */
129 	struct list_head piowait  ____cacheline_aligned_in_smp;
130 	seqlock_t waitlock;
131 
132 	spinlock_t credit_ctrl_lock ____cacheline_aligned_in_smp;
133 	u32 credit_intr_count;		/* count of credit intr users */
134 	u64 credit_ctrl;		/* cache for credit control */
135 	wait_queue_head_t halt_wait;    /* wait until kernel sees interrupt */
136 	struct work_struct halt_work;	/* halted context work queue entry */
137 };
138 
139 /* send context flags */
140 #define SCF_ENABLED 0x01
141 #define SCF_IN_FREE 0x02
142 #define SCF_HALTED  0x04
143 #define SCF_FROZEN  0x08
144 #define SCF_LINK_DOWN 0x10
145 
146 struct send_context_info {
147 	struct send_context *sc;	/* allocated working context */
148 	u16 allocated;			/* has this been allocated? */
149 	u16 type;			/* context type */
150 	u16 base;			/* base in PIO array */
151 	u16 credits;			/* size in PIO array */
152 };
153 
154 /* DMA credit return, index is always (context & 0x7) */
155 struct credit_return {
156 	volatile __le64 cr[8];
157 };
158 
159 /* NUMA indexed credit return array */
160 struct credit_return_base {
161 	struct credit_return *va;
162 	dma_addr_t dma;
163 };
164 
165 /* send context configuration sizes (one per type) */
166 struct sc_config_sizes {
167 	short int size;
168 	short int count;
169 };
170 
171 /*
172  * The diagram below details the relationship of the mapping structures
173  *
174  * Since the mapping now allows for non-uniform send contexts per vl, the
175  * number of send contexts for a vl is either the vl_scontexts[vl] or
176  * a computation based on num_kernel_send_contexts/num_vls:
177  *
178  * For example:
179  * nactual = vl_scontexts ? vl_scontexts[vl] : num_kernel_send_contexts/num_vls
180  *
181  * n = roundup to next highest power of 2 using nactual
182  *
183  * In the case where there are num_kernel_send_contexts/num_vls doesn't divide
184  * evenly, the extras are added from the last vl downward.
185  *
186  * For the case where n > nactual, the send contexts are assigned
187  * in a round robin fashion wrapping back to the first send context
188  * for a particular vl.
189  *
190  *               dd->pio_map
191  *                    |                                   pio_map_elem[0]
192  *                    |                                +--------------------+
193  *                    v                                |       mask         |
194  *               pio_vl_map                            |--------------------|
195  *      +--------------------------+                   | ksc[0] -> sc 1     |
196  *      |    list (RCU)            |                   |--------------------|
197  *      |--------------------------|                 ->| ksc[1] -> sc 2     |
198  *      |    mask                  |              --/  |--------------------|
199  *      |--------------------------|            -/     |        *           |
200  *      |    actual_vls (max 8)    |          -/       |--------------------|
201  *      |--------------------------|       --/         | ksc[n-1] -> sc n   |
202  *      |    vls (max 8)           |     -/            +--------------------+
203  *      |--------------------------|  --/
204  *      |    map[0]                |-/
205  *      |--------------------------|                   +--------------------+
206  *      |    map[1]                |---                |       mask         |
207  *      |--------------------------|   \----           |--------------------|
208  *      |           *              |        \--        | ksc[0] -> sc 1+n   |
209  *      |           *              |           \----   |--------------------|
210  *      |           *              |                \->| ksc[1] -> sc 2+n   |
211  *      |--------------------------|                   |--------------------|
212  *      |   map[vls - 1]           |-                  |         *          |
213  *      +--------------------------+ \-                |--------------------|
214  *                                     \-              | ksc[m-1] -> sc m+n |
215  *                                       \             +--------------------+
216  *                                        \-
217  *                                          \
218  *                                           \-        +----------------------+
219  *                                             \-      |       mask           |
220  *                                               \     |----------------------|
221  *                                                \-   | ksc[0] -> sc 1+m+n   |
222  *                                                  \- |----------------------|
223  *                                                    >| ksc[1] -> sc 2+m+n   |
224  *                                                     |----------------------|
225  *                                                     |         *            |
226  *                                                     |----------------------|
227  *                                                     | ksc[o-1] -> sc o+m+n |
228  *                                                     +----------------------+
229  *
230  */
231 
232 /* Initial number of send contexts per VL */
233 #define INIT_SC_PER_VL 2
234 
235 /*
236  * struct pio_map_elem - mapping for a vl
237  * @mask - selector mask
238  * @ksc - array of kernel send contexts for this vl
239  *
240  * The mask is used to "mod" the selector to
241  * produce index into the trailing array of
242  * kscs
243  */
244 struct pio_map_elem {
245 	u32 mask;
246 	struct send_context *ksc[0];
247 };
248 
249 /*
250  * struct pio_vl_map - mapping for a vl
251  * @list - rcu head for free callback
252  * @mask - vl mask to "mod" the vl to produce an index to map array
253  * @actual_vls - number of vls
254  * @vls - numbers of vls rounded to next power of 2
255  * @map - array of pio_map_elem entries
256  *
257  * This is the parent mapping structure. The trailing members of the
258  * struct point to pio_map_elem entries, which in turn point to an
259  * array of kscs for that vl.
260  */
261 struct pio_vl_map {
262 	struct rcu_head list;
263 	u32 mask;
264 	u8 actual_vls;
265 	u8 vls;
266 	struct pio_map_elem *map[0];
267 };
268 
269 int pio_map_init(struct hfi1_devdata *dd, u8 port, u8 num_vls,
270 		 u8 *vl_scontexts);
271 void free_pio_map(struct hfi1_devdata *dd);
272 struct send_context *pio_select_send_context_vl(struct hfi1_devdata *dd,
273 						u32 selector, u8 vl);
274 struct send_context *pio_select_send_context_sc(struct hfi1_devdata *dd,
275 						u32 selector, u8 sc5);
276 
277 /* send context functions */
278 int init_credit_return(struct hfi1_devdata *dd);
279 void free_credit_return(struct hfi1_devdata *dd);
280 int init_sc_pools_and_sizes(struct hfi1_devdata *dd);
281 int init_send_contexts(struct hfi1_devdata *dd);
282 int init_credit_return(struct hfi1_devdata *dd);
283 int init_pervl_scs(struct hfi1_devdata *dd);
284 struct send_context *sc_alloc(struct hfi1_devdata *dd, int type,
285 			      uint hdrqentsize, int numa);
286 void sc_free(struct send_context *sc);
287 int sc_enable(struct send_context *sc);
288 void sc_disable(struct send_context *sc);
289 int sc_restart(struct send_context *sc);
290 void sc_return_credits(struct send_context *sc);
291 void sc_flush(struct send_context *sc);
292 void sc_drop(struct send_context *sc);
293 void sc_stop(struct send_context *sc, int bit);
294 struct pio_buf *sc_buffer_alloc(struct send_context *sc, u32 dw_len,
295 				pio_release_cb cb, void *arg);
296 void sc_release_update(struct send_context *sc);
297 void sc_return_credits(struct send_context *sc);
298 void sc_group_release_update(struct hfi1_devdata *dd, u32 hw_context);
299 void sc_add_credit_return_intr(struct send_context *sc);
300 void sc_del_credit_return_intr(struct send_context *sc);
301 void sc_set_cr_threshold(struct send_context *sc, u32 new_threshold);
302 u32 sc_percent_to_threshold(struct send_context *sc, u32 percent);
303 u32 sc_mtu_to_threshold(struct send_context *sc, u32 mtu, u32 hdrqentsize);
304 void hfi1_sc_wantpiobuf_intr(struct send_context *sc, u32 needint);
305 void sc_wait(struct hfi1_devdata *dd);
306 void set_pio_integrity(struct send_context *sc);
307 
308 /* support functions */
309 void pio_reset_all(struct hfi1_devdata *dd);
310 void pio_freeze(struct hfi1_devdata *dd);
311 void pio_kernel_unfreeze(struct hfi1_devdata *dd);
312 void pio_kernel_linkup(struct hfi1_devdata *dd);
313 
314 /* global PIO send control operations */
315 #define PSC_GLOBAL_ENABLE 0
316 #define PSC_GLOBAL_DISABLE 1
317 #define PSC_GLOBAL_VLARB_ENABLE 2
318 #define PSC_GLOBAL_VLARB_DISABLE 3
319 #define PSC_CM_RESET 4
320 #define PSC_DATA_VL_ENABLE 5
321 #define PSC_DATA_VL_DISABLE 6
322 
323 void __cm_reset(struct hfi1_devdata *dd, u64 sendctrl);
324 void pio_send_control(struct hfi1_devdata *dd, int op);
325 
326 /* PIO copy routines */
327 void pio_copy(struct hfi1_devdata *dd, struct pio_buf *pbuf, u64 pbc,
328 	      const void *from, size_t count);
329 void seg_pio_copy_start(struct pio_buf *pbuf, u64 pbc,
330 			const void *from, size_t nbytes);
331 void seg_pio_copy_mid(struct pio_buf *pbuf, const void *from, size_t nbytes);
332 void seg_pio_copy_end(struct pio_buf *pbuf);
333 
334 void seqfile_dump_sci(struct seq_file *s, u32 i,
335 		      struct send_context_info *sci);
336 
337 #endif /* _PIO_H */
338