1/***********************license start***************
2 * Copyright (c) 2003-2010  Cavium Networks (support@cavium.com). All rights
3 * reserved.
4 *
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions are
8 * met:
9 *
10 *   * Redistributions of source code must retain the above copyright
11 *     notice, this list of conditions and the following disclaimer.
12 *
13 *   * Redistributions in binary form must reproduce the above
14 *     copyright notice, this list of conditions and the following
15 *     disclaimer in the documentation and/or other materials provided
16 *     with the distribution.
17
18 *   * Neither the name of Cavium Networks nor the names of
19 *     its contributors may be used to endorse or promote products
20 *     derived from this software without specific prior written
21 *     permission.
22
23 * This Software, including technical data, may be subject to U.S. export  control
24 * laws, including the U.S. Export Administration Act and its  associated
25 * regulations, and may be subject to export or import  regulations in other
26 * countries.
27
28 * TO THE MAXIMUM EXTENT PERMITTED BY LAW, THE SOFTWARE IS PROVIDED "AS IS"
29 * AND WITH ALL FAULTS AND CAVIUM  NETWORKS MAKES NO PROMISES, REPRESENTATIONS OR
30 * WARRANTIES, EITHER EXPRESS, IMPLIED, STATUTORY, OR OTHERWISE, WITH RESPECT TO
31 * THE SOFTWARE, INCLUDING ITS CONDITION, ITS CONFORMITY TO ANY REPRESENTATION OR
32 * DESCRIPTION, OR THE EXISTENCE OF ANY LATENT OR PATENT DEFECTS, AND CAVIUM
33 * SPECIFICALLY DISCLAIMS ALL IMPLIED (IF ANY) WARRANTIES OF TITLE,
34 * MERCHANTABILITY, NONINFRINGEMENT, FITNESS FOR A PARTICULAR PURPOSE, LACK OF
35 * VIRUSES, ACCURACY OR COMPLETENESS, QUIET ENJOYMENT, QUIET POSSESSION OR
36 * CORRESPONDENCE TO DESCRIPTION. THE ENTIRE  RISK ARISING OUT OF USE OR
37 * PERFORMANCE OF THE SOFTWARE LIES WITH YOU.
38 ***********************license end**************************************/
39
40
41
42
43
44
45
46/**
47 * @file
48 *
49 * Interface to the PCI / PCIe DMA engines. These are only avialable
50 * on chips with PCI / PCIe.
51 *
52 * <hr>$Revision: 49448 $<hr>
53 */
54
55#ifndef __CVMX_DMA_ENGINES_H__
56#define __CVMX_DMA_ENGINES_H__
57
58#ifdef	__cplusplus
59extern "C" {
60#endif
61
62typedef enum
63{
64    CVMX_DMA_ENGINE_TRANSFER_OUTBOUND = 0,  /**< OUTBOUND (read from L2/DRAM, write into PCI / PCIe memory space) */
65    CVMX_DMA_ENGINE_TRANSFER_INBOUND = 1,   /**< INBOUND (read from PCI / PCIe memory space, write into L2/DRAM) */
66    CVMX_DMA_ENGINE_TRANSFER_INTERNAL = 2,  /**< INTERNAL-ONLY (read from L2/DRAM, write into L2/DRAM). Only available on chips with PCIe */
67    CVMX_DMA_ENGINE_TRANSFER_EXTERNAL = 3,  /**< EXTERNAL-ONLY (read from PCIe memory space, write into PCIe memory space). Only available on chips with PCIe */
68} cvmx_dma_engine_transfer_t;
69
70typedef union
71{
72    uint64_t u64;
73    struct
74    {
75        uint64_t    reserved_60_63 :   4;   /**< Must be zero */
76        uint64_t    fport           :  2;   /**< First port. FPort indicates the physical PCIe port used for the
77                                                PCIe memory space pointers in the FIRST POINTERS block in the
78                                                EXTERNAL-ONLY case. Must be zero in the OUTBOUND, INBOUND and
79                                                INTERNAL-ONLY cases. Must be zero on chips with PCI */
80        uint64_t    lport           :  2;   /**< Last port. LPort indicates the physical PCIe port used for the
81                                                PCIe memory space pointers in the LAST POINTERS block in the
82                                                OUTBOUND, INBOUND, and EXTERNAL-ONLY cases. Must be zero in the
83                                                INTERNAL-ONLY case. Must be zero on chips with PCI */
84        cvmx_dma_engine_transfer_t type :  2; /**< Type � A given PCI DMA transfer is either OUTBOUND (read from L2/DRAM,
85                                                write into PCI / PCIe memory space), INBOUND (read from PCI / PCIe memory space, write
86                                                into L2/DRAM), INTERNAL-ONLY (read from L2/DRAM, write into L2/DRAM), or
87                                                EXTERNAL-ONLY (read from PCIe memory space, write into PCIe memory space). */
88        uint64_t    wqp             :  1;   /**< Work-queue pointer. When WQP = 1, PTR (if non-zero) is a pointer to a
89                                                work-queue entry that is submitted by the hardware after completing the DMA;
90                                                when WQP = 0, PTR (if non-zero) is a pointer to a byte in local memory that
91                                                is written to 0 by the hardware after completing the DMA. */
92        uint64_t    c               :  1;   /**< C � Counter. 1 = use counter 1, 0 = use counter 0.
93                                                The C bit selects between the two counters (NPEI_DMA_CNTS[DMA0,DMA1])
94                                                that can optionally be updated after an OUTBOUND or EXTERNAL-ONLY
95                                                transfer, and also selects between the two forced-interrupt bits
96                                                (NPEI_INT_SUMn[DMA0_FI, DMA1_FI]) that can optionally be set after an
97                                                OUTBOUND or EXTERNAL-ONLY transfer. C must be zero for INBOUND or
98                                                INTERNAL-ONLY transfers. */
99        uint64_t    ca              :  1;   /**< CA � Counter add.
100                                                When CA = 1, the hardware updates the selected counter after it completes the
101                                                PCI DMA OUTBOUND or EXTERNAL-ONLY Instruction.
102                                                    - If C = 0, PCIE_DMA_CNT0 is updated
103                                                    - If C = 1, PCIE_DMA_CNT1 is updated.
104                                                Note that this update may indirectly cause
105                                                NPEI_INT_SUM[DCNT0,DCNT1,DTIME0,DTIME1] to become set (depending
106                                                on the NPEI_DMA*_INT_LEVEL settings), so may cause interrupts to occur on a
107                                                remote PCI host.
108                                                    - If NPEI_DMA_CONTROL[O_ADD1] = 1, the counter is updated by 1.
109                                                    - If NPEI_DMA_CONTROL[O_ADD1] = 0, the counter is updated by the total
110                                                    bytes in the transfer.
111                                                When CA = 0, the hardware does not update any counters.
112                                                For an INBOUND or INTERNAL-ONLY PCI DMA transfer, CA must never be
113                                                set, and the hardware never adds to the counters. */
114        uint64_t    fi              :  1;   /**< FI � Force interrupt.
115                                                When FI is set for an OUTBOUND or EXTERNAL-ONLY transfer, the hardware
116                                                sets a forced interrupt bit after it completes the PCI DMA Instruction. If C = 0,
117                                                NPEI_INT_SUMn[DMA0_FI] is set, else NPEI_INT_SUMn[DMA1_FI] is set. For
118                                                an INBOUND or INTERNAL-ONLY PCI DMA operation, FI must never be set,
119                                                and the hardware never generates interrupts. */
120        uint64_t    ii              :  1;   /**< II� Ignore the I bit (i.e. the I bit of the PCI DMA instruction local pointer).
121                                                For OUTBOUND transfers when II = 1, ignore the I bit and the FL bit in the
122                                                DMA HDR alone determines whether the hardware frees any/all of the local
123                                                buffers in the FIRST POINTERS area:
124                                                    - when FL = 1, the hardware frees the local buffer when II=1.
125                                                    - when FL = 0, the hardware does not free the local buffer when II=1.
126                                                For OUTBOUND transfers when II = 0, the I bit in the local pointer selects
127                                                whether local buffers are freed on a pointer-by-pointer basis:
128                                                    - when (FL  I) is true, the hardware frees the local buffer when II=0.
129                                                For INBOUND, INTERNAL-ONLY, and EXTERNAL-ONLY PCI DMA transfers,
130                                                II must never be set, and local buffers are never freed. */
131        uint64_t    fl              :  1;   /**< FL � Free local buffer.
132                                                When FL = 1, for an OUTBOUND operation, it indicates that the local buffers in
133                                                the FIRST BUFFERS area should be freed.
134                                                If II = 1, the FL bit alone indicates whether the local buffer should be freed:
135                                                    - when FL = 1, the hardware frees the local buffer when II=1.
136                                                    - when FL = 0, the hardware does not free the local buffer when II=1.
137                                                If II = 0, the I bit in the local pointer (refer to Section 9.5.2) determines whether
138                                                the local buffer is freed:
139                                                    - when (FL  I) is true, the hardware frees the local buffer when II=0.
140                                                For an INBOUND, INTERNAL-ONLY, or EXTERNAL-ONLY PCI DMA transfer,
141                                                FL must never be set, and local buffers are never freed. */
142        uint64_t    nlst            :  4;   /**< NLST � Number Last pointers.
143                                                The number of pointers in the LAST POINTERS area.
144                                                In the INBOUND, OUTBOUND, and EXTERNAL-ONLY cases, the LAST
145                                                POINTERS area contains PCI components, and the number of 64-bit words
146                                                required in the LAST POINTERS area is:
147                                                    - HDR.NLST + ((HDR.NLST + 3)/4) where the division removes the fraction.
148                                                In the INTERNAL-ONLY case, the LAST POINTERS area contains local
149                                                pointers, and the number of 64-bit words required in the LAST POINTERS area is:
150                                                    - HDR.NLST
151                                                Note that the sum of the number of 64-bit words in the LAST POINTERS and
152                                                FIRST POINTERS area must never exceed 31. */
153        uint64_t    nfst            :  4;   /**< NFST � Number First pointers.
154                                                The number of pointers in the FIRST POINTERS area.
155                                                In the INBOUND, OUTBOUND, and INTERNAL-ONLY cases, the FIRST
156                                                POINTERS area contains local pointers, and the number of 64-bit words required
157                                                in the FIRST POINTERS area is:
158                                                    - HDR.NFST
159                                                In the EXTERNAL-ONLY case, the FIRST POINTERS area contains PCI
160                                                components, and the number of 64-bit words required in the FIRST POINTERS
161                                                area is:
162                                                    - HDR.NFST + ((HDR.NFST + 3)/4) where the division removes the fraction. */
163        uint64_t    addr            : 40;   /**< PTR � Pointer, either a work-queue-entry pointer (when WQP = 1) or a local
164                                                memory pointer (WQP = 0).
165                                                When WQP = 1 and PTR  0x0, the hardware inserts the work-queue entry
166                                                indicated by PTR into a POW input queue after the PCI DMA operation is
167                                                complete. (Section 5.4 describes the work queue entry requirements in this
168                                                case.) When WQP = 1, PTR<2:0> must be 0x0.
169                                                When WQP = 0 and PTR  0x0, the hardware writes the single byte in local
170                                                memory indicated by PTR to 0x0 after the PCI DMA operation is complete.
171                                                NPEI_DMA_CONTROL[B0_LEND] selects the endian-ness of PTR in this
172                                                case.
173                                                When PTR = 0x0, the hardware performs no operation after the PCI DMA
174                                                operation is complete. */
175    } s;
176} cvmx_dma_engine_header_t;
177
178typedef union
179{
180    uint64_t u64;
181    struct
182    {
183        uint64_t    i               :  1;   /**< I � Invert free.
184                                                This bit gives the software the ability to free buffers independently for an
185                                                OUTBOUND PCI DMA transfer. I is not used by the hardware when II is set. I
186                                                must not be set, and buffers are never freed, for INBOUND, INTERNAL-ONLY,
187                                                and EXTERNAL-ONLY PCI DMA transfers. */
188        uint64_t    back            :  4;   /**< Back � Backup amount.
189                                                Allows the start of a buffer that is to be freed during an OUTBOUND transfer to
190                                                be different from the ptr value. Back specifies the amount to subtract from the
191                                                pointer to reach the start when freeing a buffer.
192                                                The address that is the start of the buffer being freed is:
193                                                    - Buffer start address = ((ptr >> 7) - Back) << 7.
194                                                Back is only used by the hardware when the buffer corresponding to ptr is freed.
195                                                Back must be 0x0, and buffers are never freed, for INBOUND, INTERNAL-ONLY,
196                                                and EXTERNAL-ONLY PCI DMA transfers. */
197        uint64_t    pool            :  3;   /**< Pool � Free pool.
198                                                Specifies which pool (of the eight hardware-managed FPA free pools) receives the
199                                                buffer associated with ptr when freed during an OUTBOUND transfer.
200                                                Pool is only used when the buffer corresponding to ptr is freed. Pool must be 0x0,
201                                                and buffers are never freed, for INBOUND, INTERNAL-ONLY, and EXTERNAL-ONLY
202                                                PCI DMA transfers. */
203        uint64_t    f               :  1;   /**< F � Full-block writes are allowed.
204                                                When set, the hardware is permitted to write all the bytes in the cache blocks
205                                                covered by ptr, ptr + Size - 1. This can improve memory system performance
206                                                when the write misses in the L2 cache.
207                                                F can only be set for local pointers that can be written to:
208                                                    - The local pointers in the FIRST POINTERS area that are write pointers for
209                                                    INBOUND transfers.
210                                                    - The local pointers in the LAST POINTERS area that are always write
211                                                    pointers (when present for INTERNAL-ONLY transfers).
212                                                F must not be set for local pointers that are not written to:
213                                                    - The local pointers in the FIRST POINTERS area for OUTBOUND and
214                                                    INTERNAL-ONLY transfers. */
215        uint64_t    a               :  1;   /**< A � Allocate L2.
216                                                This is a hint to the hardware that the cache blocks should be allocated in the L2
217                                                cache (if they were not already). */
218        uint64_t    l               :  1;   /**< L � Little-endian.
219                                                When L is set, the data at ptr is in little-endian format rather than big-endian. */
220        uint64_t    size            : 13;   /**< Size � Size in bytes of the contiguous space specified by ptr. A Size value of 0 is
221                                                illegal. Note that the sum of the sizes in the FIRST POINTERS area must always
222                                                exactly equal the sum of the sizes/lengths in the LAST POINTERS area:
223                                                    - In the OUTBOUND and INBOUND cases, the HDR.NFST size fields in the
224                                                    local pointers in the FIRST POINTERS area must exactly equal the lengths
225                                                    of the HDR.NLST fragments in the PCI components in the LAST POINTERS
226                                                    area.
227                                                    - In the INTERNAL-ONLY case, the HDR.NFST size fields in the local
228                                                    pointers in the FIRST POINTERS area must equal the HDR.NLST size
229                                                    fields in the local pointers in the LAST POINTERS area. */
230        uint64_t    reserved_36_39  :  4;   /**< Must be zero */
231        uint64_t    addr            : 36;   /**< L2/DRAM byte pointer. Points to where the packet data starts.
232                                                Ptr can be any byte alignment. Note that ptr is interpreted as a big-endian byte
233                                                pointer when L is clear, a little-endian byte pointer when L is set. */
234    } internal;
235    struct
236    {
237        uint64_t    len0            : 16;   /**< Length of PCI / PCIe memory for address 0 */
238        uint64_t    len1            : 16;   /**< Length of PCI / PCIe memory for address 1 */
239        uint64_t    len2            : 16;   /**< Length of PCI / PCIe memory for address 2 */
240        uint64_t    len3            : 16;   /**< Length of PCI / PCIe memory for address 3 */
241    } pcie_length;
242} cvmx_dma_engine_buffer_t;
243
244/**
245 * Initialize the DMA engines for use
246 *
247 * @return Zero on success, negative on failure
248 */
249int cvmx_dma_engine_initialize(void);
250
251/**
252 * Shutdown all DMA engines. The engeines must be idle when this
253 * function is called.
254 *
255 * @return Zero on success, negative on failure
256 */
257int cvmx_dma_engine_shutdown(void);
258
259/**
260 * Return the number of DMA engimes supported by this chip
261 *
262 * @return Number of DMA engines
263 */
264int cvmx_dma_engine_get_num(void);
265
266/**
267 * Submit a series of DMA comamnd to the DMA engines.
268 *
269 * @param engine  Engine to submit to (0 to cvmx_dma_engine_get_num()-1)
270 * @param header  Command header
271 * @param num_buffers
272 *                The number of data pointers
273 * @param buffers Comamnd data pointers
274 *
275 * @return Zero on success, negative on failure
276 */
277int cvmx_dma_engine_submit(int engine, cvmx_dma_engine_header_t header, int num_buffers, cvmx_dma_engine_buffer_t buffers[]);
278
279/**
280 * Build the first and last pointers based on a DMA engine header
281 * and submit them to the engine. The purpose of this function is
282 * to simplify the building of DMA engine commands by automatically
283 * converting a simple address and size into the apropriate internal
284 * or PCI / PCIe address list. This function does not support gather lists,
285 * so you will need to build your own lists in that case.
286 *
287 * @param engine Engine to submit to (0 to cvmx_dma_engine_get_num()-1)
288 * @param header DMA Command header. Note that the nfst and nlst fields do not
289 *               need to be filled in. All other fields must be set properly.
290 * @param first_address
291 *               Address to use for the first pointers. In the case of INTERNAL,
292 *               INBOUND, and OUTBOUND this is an Octeon memory address. In the
293 *               case of EXTERNAL, this is the source PCI / PCIe address.
294 * @param last_address
295 *               Address to use for the last pointers. In the case of EXTERNAL,
296 *               INBOUND, and OUTBOUND this is a PCI / PCIe address. In the
297 *               case of INTERNAL, this is the Octeon memory destination address.
298 * @param size   Size of the transfer to perform.
299 *
300 * @return Zero on success, negative on failure
301 */
302int cvmx_dma_engine_transfer(int engine, cvmx_dma_engine_header_t header,
303                             uint64_t first_address, uint64_t last_address,
304                             int size);
305
306/**
307 * Simplified interface to the DMA engines to emulate memcpy()
308 *
309 * @param engine Engine to submit to (0 to cvmx_dma_engine_get_num()-1)
310 * @param dest   Pointer to the destination memory. cvmx_ptr_to_phys() will be
311 *               used to turn this into a physical address. It cannot be a local
312 *               or CVMX_SHARED block.
313 * @param source Pointer to the source memory.
314 *               cvmx_ptr_to_phys() will be used to turn this
315 *               into a physical address. It cannot be a local
316 *               or CVMX_SHARED block.
317 * @param length Number of bytes to copy
318 *
319 * @return Zero on success, negative on failure
320 */
321static inline int cvmx_dma_engine_memcpy(int engine, void *dest, void *source, int length)
322{
323    cvmx_dma_engine_header_t header;
324    header.u64 = 0;
325    header.s.type = CVMX_DMA_ENGINE_TRANSFER_INTERNAL;
326    return cvmx_dma_engine_transfer(engine, header, cvmx_ptr_to_phys(source),
327                                    cvmx_ptr_to_phys(dest), length);
328}
329
330#ifdef	__cplusplus
331}
332#endif
333
334#endif // __CVMX_CMD_QUEUE_H__
335