1 /***********************license start***************
2  * Author: Cavium Networks
3  *
4  * Contact: support@caviumnetworks.com
5  * This file is part of the OCTEON SDK
6  *
7  * Copyright (c) 2003-2008 Cavium Networks
8  *
9  * This file is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License, Version 2, as
11  * published by the Free Software Foundation.
12  *
13  * This file is distributed in the hope that it will be useful, but
14  * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty
15  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or
16  * NONINFRINGEMENT.  See the GNU General Public License for more
17  * details.
18  *
19  * You should have received a copy of the GNU General Public License
20  * along with this file; if not, write to the Free Software
21  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
22  * or visit http://www.gnu.org/licenses/.
23  *
24  * This file may also be available under a different license from Cavium.
25  * Contact Cavium Networks for more information
26  ***********************license end**************************************/
27 
28 /**
29  *
30  * Interface to the hardware Packet Output unit.
31  *
32  * Starting with SDK 1.7.0, the PKO output functions now support
33  * two types of locking. CVMX_PKO_LOCK_ATOMIC_TAG continues to
34  * function similarly to previous SDKs by using POW atomic tags
35  * to preserve ordering and exclusivity. As a new option, you
36  * can now pass CVMX_PKO_LOCK_CMD_QUEUE which uses a ll/sc
37  * memory based locking instead. This locking has the advantage
38  * of not affecting the tag state but doesn't preserve packet
39  * ordering. CVMX_PKO_LOCK_CMD_QUEUE is appropriate in most
40  * generic code while CVMX_PKO_LOCK_CMD_QUEUE should be used
41  * with hand tuned fast path code.
42  *
43  * Some of other SDK differences visible to the command queuing:
44  * - PKO indexes are no longer stored in the FAU. A large
45  *   percentage of the FAU register block used to be tied up
46  *   maintaining PKO queue pointers. These are now stored in a
47  *   global named block.
48  * - The PKO <b>use_locking</b> parameter can now have a global
49  *   effect. Since all application use the same named block,
50  *   queue locking correctly applies across all operating
51  *   systems when using CVMX_PKO_LOCK_CMD_QUEUE.
52  * - PKO 3 word commands are now supported. Use
53  *   cvmx_pko_send_packet_finish3().
54  *
55  */
56 
57 #ifndef __CVMX_PKO_H__
58 #define __CVMX_PKO_H__
59 
60 #include <asm/octeon/cvmx-fpa.h>
61 #include <asm/octeon/cvmx-pow.h>
62 #include <asm/octeon/cvmx-cmd-queue.h>
63 #include <asm/octeon/cvmx-pko-defs.h>
64 
65 /* Adjust the command buffer size by 1 word so that in the case of using only
66  * two word PKO commands no command words stradle buffers.  The useful values
67  * for this are 0 and 1. */
68 #define CVMX_PKO_COMMAND_BUFFER_SIZE_ADJUST (1)
69 
70 #define CVMX_PKO_MAX_OUTPUT_QUEUES_STATIC 256
71 #define CVMX_PKO_MAX_OUTPUT_QUEUES	((OCTEON_IS_MODEL(OCTEON_CN31XX) || \
72 	OCTEON_IS_MODEL(OCTEON_CN3010) || OCTEON_IS_MODEL(OCTEON_CN3005) || \
73 	OCTEON_IS_MODEL(OCTEON_CN50XX)) ? 32 : \
74 		(OCTEON_IS_MODEL(OCTEON_CN58XX) || \
75 		OCTEON_IS_MODEL(OCTEON_CN56XX)) ? 256 : 128)
76 #define CVMX_PKO_NUM_OUTPUT_PORTS	40
77 /* use this for queues that are not used */
78 #define CVMX_PKO_MEM_QUEUE_PTRS_ILLEGAL_PID 63
79 #define CVMX_PKO_QUEUE_STATIC_PRIORITY	9
80 #define CVMX_PKO_ILLEGAL_QUEUE	0xFFFF
81 #define CVMX_PKO_MAX_QUEUE_DEPTH 0
82 
83 typedef enum {
84 	CVMX_PKO_SUCCESS,
85 	CVMX_PKO_INVALID_PORT,
86 	CVMX_PKO_INVALID_QUEUE,
87 	CVMX_PKO_INVALID_PRIORITY,
88 	CVMX_PKO_NO_MEMORY,
89 	CVMX_PKO_PORT_ALREADY_SETUP,
90 	CVMX_PKO_CMD_QUEUE_INIT_ERROR
91 } cvmx_pko_status_t;
92 
93 /**
94  * This enumeration represents the differnet locking modes supported by PKO.
95  */
96 typedef enum {
97 	/*
98 	 * PKO doesn't do any locking. It is the responsibility of the
99 	 * application to make sure that no other core is accessing
100 	 * the same queue at the same time
101 	 */
102 	CVMX_PKO_LOCK_NONE = 0,
103 	/*
104 	 * PKO performs an atomic tagswitch to insure exclusive access
105 	 * to the output queue. This will maintain packet ordering on
106 	 * output.
107 	 */
108 	CVMX_PKO_LOCK_ATOMIC_TAG = 1,
109 	/*
110 	 * PKO uses the common command queue locks to insure exclusive
111 	 * access to the output queue. This is a memory based
112 	 * ll/sc. This is the most portable locking mechanism.
113 	 */
114 	CVMX_PKO_LOCK_CMD_QUEUE = 2,
115 } cvmx_pko_lock_t;
116 
117 typedef struct {
118 	uint32_t packets;
119 	uint64_t octets;
120 	uint64_t doorbell;
121 } cvmx_pko_port_status_t;
122 
123 /**
124  * This structure defines the address to use on a packet enqueue
125  */
126 typedef union {
127 	uint64_t u64;
128 	struct {
129 #ifdef __BIG_ENDIAN_BITFIELD
130 		/* Must CVMX_IO_SEG */
131 		uint64_t mem_space:2;
132 		/* Must be zero */
133 		uint64_t reserved:13;
134 		/* Must be one */
135 		uint64_t is_io:1;
136 		/* The ID of the device on the non-coherent bus */
137 		uint64_t did:8;
138 		/* Must be zero */
139 		uint64_t reserved2:4;
140 		/* Must be zero */
141 		uint64_t reserved3:18;
142 		/*
143 		 * The hardware likes to have the output port in
144 		 * addition to the output queue,
145 		 */
146 		uint64_t port:6;
147 		/*
148 		 * The output queue to send the packet to (0-127 are
149 		 * legal)
150 		 */
151 		uint64_t queue:9;
152 		/* Must be zero */
153 		uint64_t reserved4:3;
154 #else
155 	        uint64_t reserved4:3;
156 	        uint64_t queue:9;
157 	        uint64_t port:9;
158 	        uint64_t reserved3:15;
159 	        uint64_t reserved2:4;
160 	        uint64_t did:8;
161 	        uint64_t is_io:1;
162 	        uint64_t reserved:13;
163 	        uint64_t mem_space:2;
164 #endif
165 	} s;
166 } cvmx_pko_doorbell_address_t;
167 
168 /**
169  * Structure of the first packet output command word.
170  */
171 union cvmx_pko_command_word0 {
172 	uint64_t u64;
173 	struct {
174 #ifdef __BIG_ENDIAN_BITFIELD
175 		/*
176 		 * The size of the reg1 operation - could be 8, 16,
177 		 * 32, or 64 bits.
178 		 */
179 		uint64_t size1:2;
180 		/*
181 		 * The size of the reg0 operation - could be 8, 16,
182 		 * 32, or 64 bits.
183 		 */
184 		uint64_t size0:2;
185 		/*
186 		 * If set, subtract 1, if clear, subtract packet
187 		 * size.
188 		 */
189 		uint64_t subone1:1;
190 		/*
191 		 * The register, subtract will be done if reg1 is
192 		 * non-zero.
193 		 */
194 		uint64_t reg1:11;
195 		/* If set, subtract 1, if clear, subtract packet size */
196 		uint64_t subone0:1;
197 		/* The register, subtract will be done if reg0 is non-zero */
198 		uint64_t reg0:11;
199 		/*
200 		 * When set, interpret segment pointer and segment
201 		 * bytes in little endian order.
202 		 */
203 		uint64_t le:1;
204 		/*
205 		 * When set, packet data not allocated in L2 cache by
206 		 * PKO.
207 		 */
208 		uint64_t n2:1;
209 		/*
210 		 * If set and rsp is set, word3 contains a pointer to
211 		 * a work queue entry.
212 		 */
213 		uint64_t wqp:1;
214 		/* If set, the hardware will send a response when done */
215 		uint64_t rsp:1;
216 		/*
217 		 * If set, the supplied pkt_ptr is really a pointer to
218 		 * a list of pkt_ptr's.
219 		 */
220 		uint64_t gather:1;
221 		/*
222 		 * If ipoffp1 is non zero, (ipoffp1-1) is the number
223 		 * of bytes to IP header, and the hardware will
224 		 * calculate and insert the UDP/TCP checksum.
225 		 */
226 		uint64_t ipoffp1:7;
227 		/*
228 		 * If set, ignore the I bit (force to zero) from all
229 		 * pointer structures.
230 		 */
231 		uint64_t ignore_i:1;
232 		/*
233 		 * If clear, the hardware will attempt to free the
234 		 * buffers containing the packet.
235 		 */
236 		uint64_t dontfree:1;
237 		/*
238 		 * The total number of segs in the packet, if gather
239 		 * set, also gather list length.
240 		 */
241 		uint64_t segs:6;
242 		/* Including L2, but no trailing CRC */
243 		uint64_t total_bytes:16;
244 #else
245 	        uint64_t total_bytes:16;
246 	        uint64_t segs:6;
247 	        uint64_t dontfree:1;
248 	        uint64_t ignore_i:1;
249 	        uint64_t ipoffp1:7;
250 	        uint64_t gather:1;
251 	        uint64_t rsp:1;
252 	        uint64_t wqp:1;
253 	        uint64_t n2:1;
254 	        uint64_t le:1;
255 	        uint64_t reg0:11;
256 	        uint64_t subone0:1;
257 	        uint64_t reg1:11;
258 	        uint64_t subone1:1;
259 	        uint64_t size0:2;
260 	        uint64_t size1:2;
261 #endif
262 	} s;
263 };
264 
265 /* CSR typedefs have been moved to cvmx-csr-*.h */
266 
267 /**
268  * Definition of internal state for Packet output processing
269  */
270 typedef struct {
271 	/* ptr to start of buffer, offset kept in FAU reg */
272 	uint64_t *start_ptr;
273 } cvmx_pko_state_elem_t;
274 
275 /**
276  * Call before any other calls to initialize the packet
277  * output system.
278  */
279 extern void cvmx_pko_initialize_global(void);
280 
281 /**
282  * Enables the packet output hardware. It must already be
283  * configured.
284  */
285 extern void cvmx_pko_enable(void);
286 
287 /**
288  * Disables the packet output. Does not affect any configuration.
289  */
290 extern void cvmx_pko_disable(void);
291 
292 /**
293  * Shutdown and free resources required by packet output.
294  */
295 
296 extern void cvmx_pko_shutdown(void);
297 
298 /**
299  * Configure a output port and the associated queues for use.
300  *
301  * @port:	Port to configure.
302  * @base_queue: First queue number to associate with this port.
303  * @num_queues: Number of queues t oassociate with this port
304  * @priority:	Array of priority levels for each queue. Values are
305  *		     allowed to be 1-8. A value of 8 get 8 times the traffic
306  *		     of a value of 1. There must be num_queues elements in the
307  *		     array.
308  */
309 extern cvmx_pko_status_t cvmx_pko_config_port(uint64_t port,
310 					      uint64_t base_queue,
311 					      uint64_t num_queues,
312 					      const uint64_t priority[]);
313 
314 /**
315  * Ring the packet output doorbell. This tells the packet
316  * output hardware that "len" command words have been added
317  * to its pending list.	 This command includes the required
318  * CVMX_SYNCWS before the doorbell ring.
319  *
320  * @port:   Port the packet is for
321  * @queue:  Queue the packet is for
322  * @len:    Length of the command in 64 bit words
323  */
cvmx_pko_doorbell(uint64_t port,uint64_t queue,uint64_t len)324 static inline void cvmx_pko_doorbell(uint64_t port, uint64_t queue,
325 				     uint64_t len)
326 {
327 	cvmx_pko_doorbell_address_t ptr;
328 
329 	ptr.u64 = 0;
330 	ptr.s.mem_space = CVMX_IO_SEG;
331 	ptr.s.did = CVMX_OCT_DID_PKT_SEND;
332 	ptr.s.is_io = 1;
333 	ptr.s.port = port;
334 	ptr.s.queue = queue;
335 	/*
336 	 * Need to make sure output queue data is in DRAM before
337 	 * doorbell write.
338 	 */
339 	CVMX_SYNCWS;
340 	cvmx_write_io(ptr.u64, len);
341 }
342 
343 /**
344  * Prepare to send a packet.  This may initiate a tag switch to
345  * get exclusive access to the output queue structure, and
346  * performs other prep work for the packet send operation.
347  *
348  * cvmx_pko_send_packet_finish() MUST be called after this function is called,
349  * and must be called with the same port/queue/use_locking arguments.
350  *
351  * The use_locking parameter allows the caller to use three
352  * possible locking modes.
353  * - CVMX_PKO_LOCK_NONE
354  *	- PKO doesn't do any locking. It is the responsibility
355  *	    of the application to make sure that no other core
356  *	    is accessing the same queue at the same time.
357  * - CVMX_PKO_LOCK_ATOMIC_TAG
358  *	- PKO performs an atomic tagswitch to insure exclusive
359  *	    access to the output queue. This will maintain
360  *	    packet ordering on output.
361  * - CVMX_PKO_LOCK_CMD_QUEUE
362  *	- PKO uses the common command queue locks to insure
363  *	    exclusive access to the output queue. This is a
364  *	    memory based ll/sc. This is the most portable
365  *	    locking mechanism.
366  *
367  * NOTE: If atomic locking is used, the POW entry CANNOT be
368  * descheduled, as it does not contain a valid WQE pointer.
369  *
370  * @port:   Port to send it on
371  * @queue:  Queue to use
372  * @use_locking: CVMX_PKO_LOCK_NONE, CVMX_PKO_LOCK_ATOMIC_TAG, or
373  *		 CVMX_PKO_LOCK_CMD_QUEUE
374  */
375 
cvmx_pko_send_packet_prepare(uint64_t port,uint64_t queue,cvmx_pko_lock_t use_locking)376 static inline void cvmx_pko_send_packet_prepare(uint64_t port, uint64_t queue,
377 						cvmx_pko_lock_t use_locking)
378 {
379 	if (use_locking == CVMX_PKO_LOCK_ATOMIC_TAG) {
380 		/*
381 		 * Must do a full switch here to handle all cases.  We
382 		 * use a fake WQE pointer, as the POW does not access
383 		 * this memory.	 The WQE pointer and group are only
384 		 * used if this work is descheduled, which is not
385 		 * supported by the
386 		 * cvmx_pko_send_packet_prepare/cvmx_pko_send_packet_finish
387 		 * combination.	 Note that this is a special case in
388 		 * which these fake values can be used - this is not a
389 		 * general technique.
390 		 */
391 		uint32_t tag =
392 		    CVMX_TAG_SW_BITS_INTERNAL << CVMX_TAG_SW_SHIFT |
393 		    CVMX_TAG_SUBGROUP_PKO << CVMX_TAG_SUBGROUP_SHIFT |
394 		    (CVMX_TAG_SUBGROUP_MASK & queue);
395 		cvmx_pow_tag_sw_full((struct cvmx_wqe *) cvmx_phys_to_ptr(0x80), tag,
396 				     CVMX_POW_TAG_TYPE_ATOMIC, 0);
397 	}
398 }
399 
400 /**
401  * Complete packet output. cvmx_pko_send_packet_prepare() must be
402  * called exactly once before this, and the same parameters must be
403  * passed to both cvmx_pko_send_packet_prepare() and
404  * cvmx_pko_send_packet_finish().
405  *
406  * @port:   Port to send it on
407  * @queue:  Queue to use
408  * @pko_command:
409  *		 PKO HW command word
410  * @packet: Packet to send
411  * @use_locking: CVMX_PKO_LOCK_NONE, CVMX_PKO_LOCK_ATOMIC_TAG, or
412  *		 CVMX_PKO_LOCK_CMD_QUEUE
413  *
414  * Returns: CVMX_PKO_SUCCESS on success, or error code on
415  * failure of output
416  */
cvmx_pko_send_packet_finish(uint64_t port,uint64_t queue,union cvmx_pko_command_word0 pko_command,union cvmx_buf_ptr packet,cvmx_pko_lock_t use_locking)417 static inline cvmx_pko_status_t cvmx_pko_send_packet_finish(
418 	uint64_t port,
419 	uint64_t queue,
420 	union cvmx_pko_command_word0 pko_command,
421 	union cvmx_buf_ptr packet,
422 	cvmx_pko_lock_t use_locking)
423 {
424 	cvmx_cmd_queue_result_t result;
425 	if (use_locking == CVMX_PKO_LOCK_ATOMIC_TAG)
426 		cvmx_pow_tag_sw_wait();
427 	result = cvmx_cmd_queue_write2(CVMX_CMD_QUEUE_PKO(queue),
428 				       (use_locking == CVMX_PKO_LOCK_CMD_QUEUE),
429 				       pko_command.u64, packet.u64);
430 	if (likely(result == CVMX_CMD_QUEUE_SUCCESS)) {
431 		cvmx_pko_doorbell(port, queue, 2);
432 		return CVMX_PKO_SUCCESS;
433 	} else if ((result == CVMX_CMD_QUEUE_NO_MEMORY)
434 		   || (result == CVMX_CMD_QUEUE_FULL)) {
435 		return CVMX_PKO_NO_MEMORY;
436 	} else {
437 		return CVMX_PKO_INVALID_QUEUE;
438 	}
439 }
440 
441 /**
442  * Complete packet output. cvmx_pko_send_packet_prepare() must be
443  * called exactly once before this, and the same parameters must be
444  * passed to both cvmx_pko_send_packet_prepare() and
445  * cvmx_pko_send_packet_finish().
446  *
447  * @port:   Port to send it on
448  * @queue:  Queue to use
449  * @pko_command:
450  *		 PKO HW command word
451  * @packet: Packet to send
452  * @addr: Plysical address of a work queue entry or physical address
453  *	  to zero on complete.
454  * @use_locking: CVMX_PKO_LOCK_NONE, CVMX_PKO_LOCK_ATOMIC_TAG, or
455  *		 CVMX_PKO_LOCK_CMD_QUEUE
456  *
457  * Returns: CVMX_PKO_SUCCESS on success, or error code on
458  * failure of output
459  */
cvmx_pko_send_packet_finish3(uint64_t port,uint64_t queue,union cvmx_pko_command_word0 pko_command,union cvmx_buf_ptr packet,uint64_t addr,cvmx_pko_lock_t use_locking)460 static inline cvmx_pko_status_t cvmx_pko_send_packet_finish3(
461 	uint64_t port,
462 	uint64_t queue,
463 	union cvmx_pko_command_word0 pko_command,
464 	union cvmx_buf_ptr packet,
465 	uint64_t addr,
466 	cvmx_pko_lock_t use_locking)
467 {
468 	cvmx_cmd_queue_result_t result;
469 	if (use_locking == CVMX_PKO_LOCK_ATOMIC_TAG)
470 		cvmx_pow_tag_sw_wait();
471 	result = cvmx_cmd_queue_write3(CVMX_CMD_QUEUE_PKO(queue),
472 				       (use_locking == CVMX_PKO_LOCK_CMD_QUEUE),
473 				       pko_command.u64, packet.u64, addr);
474 	if (likely(result == CVMX_CMD_QUEUE_SUCCESS)) {
475 		cvmx_pko_doorbell(port, queue, 3);
476 		return CVMX_PKO_SUCCESS;
477 	} else if ((result == CVMX_CMD_QUEUE_NO_MEMORY)
478 		   || (result == CVMX_CMD_QUEUE_FULL)) {
479 		return CVMX_PKO_NO_MEMORY;
480 	} else {
481 		return CVMX_PKO_INVALID_QUEUE;
482 	}
483 }
484 
485 /**
486  * Return the pko output queue associated with a port and a specific core.
487  * In normal mode (PKO lockless operation is disabled), the value returned
488  * is the base queue.
489  *
490  * @port:   Port number
491  * @core:   Core to get queue for
492  *
493  * Returns Core-specific output queue
494  */
cvmx_pko_get_base_queue_per_core(int port,int core)495 static inline int cvmx_pko_get_base_queue_per_core(int port, int core)
496 {
497 #ifndef CVMX_HELPER_PKO_MAX_PORTS_INTERFACE0
498 #define CVMX_HELPER_PKO_MAX_PORTS_INTERFACE0 16
499 #endif
500 #ifndef CVMX_HELPER_PKO_MAX_PORTS_INTERFACE1
501 #define CVMX_HELPER_PKO_MAX_PORTS_INTERFACE1 16
502 #endif
503 
504 	if (port < CVMX_PKO_MAX_PORTS_INTERFACE0)
505 		return port * CVMX_PKO_QUEUES_PER_PORT_INTERFACE0 + core;
506 	else if (port >= 16 && port < 16 + CVMX_PKO_MAX_PORTS_INTERFACE1)
507 		return CVMX_PKO_MAX_PORTS_INTERFACE0 *
508 		    CVMX_PKO_QUEUES_PER_PORT_INTERFACE0 + (port -
509 							   16) *
510 		    CVMX_PKO_QUEUES_PER_PORT_INTERFACE1 + core;
511 	else if ((port >= 32) && (port < 36))
512 		return CVMX_PKO_MAX_PORTS_INTERFACE0 *
513 		    CVMX_PKO_QUEUES_PER_PORT_INTERFACE0 +
514 		    CVMX_PKO_MAX_PORTS_INTERFACE1 *
515 		    CVMX_PKO_QUEUES_PER_PORT_INTERFACE1 + (port -
516 							   32) *
517 		    CVMX_PKO_QUEUES_PER_PORT_PCI;
518 	else if ((port >= 36) && (port < 40))
519 		return CVMX_PKO_MAX_PORTS_INTERFACE0 *
520 		    CVMX_PKO_QUEUES_PER_PORT_INTERFACE0 +
521 		    CVMX_PKO_MAX_PORTS_INTERFACE1 *
522 		    CVMX_PKO_QUEUES_PER_PORT_INTERFACE1 +
523 		    4 * CVMX_PKO_QUEUES_PER_PORT_PCI + (port -
524 							36) *
525 		    CVMX_PKO_QUEUES_PER_PORT_LOOP;
526 	else
527 		/* Given the limit on the number of ports we can map to
528 		 * CVMX_MAX_OUTPUT_QUEUES_STATIC queues (currently 256,
529 		 * divided among all cores), the remaining unmapped ports
530 		 * are assigned an illegal queue number */
531 		return CVMX_PKO_ILLEGAL_QUEUE;
532 }
533 
534 /**
535  * For a given port number, return the base pko output queue
536  * for the port.
537  *
538  * @port:   Port number
539  * Returns Base output queue
540  */
cvmx_pko_get_base_queue(int port)541 static inline int cvmx_pko_get_base_queue(int port)
542 {
543 	if (OCTEON_IS_MODEL(OCTEON_CN68XX))
544 		return port;
545 
546 	return cvmx_pko_get_base_queue_per_core(port, 0);
547 }
548 
549 /**
550  * For a given port number, return the number of pko output queues.
551  *
552  * @port:   Port number
553  * Returns Number of output queues
554  */
cvmx_pko_get_num_queues(int port)555 static inline int cvmx_pko_get_num_queues(int port)
556 {
557 	if (port < 16)
558 		return CVMX_PKO_QUEUES_PER_PORT_INTERFACE0;
559 	else if (port < 32)
560 		return CVMX_PKO_QUEUES_PER_PORT_INTERFACE1;
561 	else if (port < 36)
562 		return CVMX_PKO_QUEUES_PER_PORT_PCI;
563 	else if (port < 40)
564 		return CVMX_PKO_QUEUES_PER_PORT_LOOP;
565 	else
566 		return 0;
567 }
568 
569 /**
570  * Get the status counters for a port.
571  *
572  * @port_num: Port number to get statistics for.
573  * @clear:    Set to 1 to clear the counters after they are read
574  * @status:   Where to put the results.
575  */
cvmx_pko_get_port_status(uint64_t port_num,uint64_t clear,cvmx_pko_port_status_t * status)576 static inline void cvmx_pko_get_port_status(uint64_t port_num, uint64_t clear,
577 					    cvmx_pko_port_status_t *status)
578 {
579 	union cvmx_pko_reg_read_idx pko_reg_read_idx;
580 	union cvmx_pko_mem_count0 pko_mem_count0;
581 	union cvmx_pko_mem_count1 pko_mem_count1;
582 
583 	pko_reg_read_idx.u64 = 0;
584 	pko_reg_read_idx.s.index = port_num;
585 	cvmx_write_csr(CVMX_PKO_REG_READ_IDX, pko_reg_read_idx.u64);
586 
587 	pko_mem_count0.u64 = cvmx_read_csr(CVMX_PKO_MEM_COUNT0);
588 	status->packets = pko_mem_count0.s.count;
589 	if (clear) {
590 		pko_mem_count0.s.count = port_num;
591 		cvmx_write_csr(CVMX_PKO_MEM_COUNT0, pko_mem_count0.u64);
592 	}
593 
594 	pko_mem_count1.u64 = cvmx_read_csr(CVMX_PKO_MEM_COUNT1);
595 	status->octets = pko_mem_count1.s.count;
596 	if (clear) {
597 		pko_mem_count1.s.count = port_num;
598 		cvmx_write_csr(CVMX_PKO_MEM_COUNT1, pko_mem_count1.u64);
599 	}
600 
601 	if (OCTEON_IS_MODEL(OCTEON_CN3XXX)) {
602 		union cvmx_pko_mem_debug9 debug9;
603 		pko_reg_read_idx.s.index = cvmx_pko_get_base_queue(port_num);
604 		cvmx_write_csr(CVMX_PKO_REG_READ_IDX, pko_reg_read_idx.u64);
605 		debug9.u64 = cvmx_read_csr(CVMX_PKO_MEM_DEBUG9);
606 		status->doorbell = debug9.cn38xx.doorbell;
607 	} else {
608 		union cvmx_pko_mem_debug8 debug8;
609 		pko_reg_read_idx.s.index = cvmx_pko_get_base_queue(port_num);
610 		cvmx_write_csr(CVMX_PKO_REG_READ_IDX, pko_reg_read_idx.u64);
611 		debug8.u64 = cvmx_read_csr(CVMX_PKO_MEM_DEBUG8);
612 		status->doorbell = debug8.cn50xx.doorbell;
613 	}
614 }
615 
616 /**
617  * Rate limit a PKO port to a max packets/sec. This function is only
618  * supported on CN57XX, CN56XX, CN55XX, and CN54XX.
619  *
620  * @port:      Port to rate limit
621  * @packets_s: Maximum packet/sec
622  * @burst:     Maximum number of packets to burst in a row before rate
623  *		    limiting cuts in.
624  *
625  * Returns Zero on success, negative on failure
626  */
627 extern int cvmx_pko_rate_limit_packets(int port, int packets_s, int burst);
628 
629 /**
630  * Rate limit a PKO port to a max bits/sec. This function is only
631  * supported on CN57XX, CN56XX, CN55XX, and CN54XX.
632  *
633  * @port:   Port to rate limit
634  * @bits_s: PKO rate limit in bits/sec
635  * @burst:  Maximum number of bits to burst before rate
636  *		 limiting cuts in.
637  *
638  * Returns Zero on success, negative on failure
639  */
640 extern int cvmx_pko_rate_limit_bits(int port, uint64_t bits_s, int burst);
641 
642 #endif /* __CVMX_PKO_H__ */
643