2 * Copyright 2012 Tilera Corporation. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation, version 2.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
11 * NON INFRINGEMENT. See the GNU General Public License for
15 #ifndef _GXIO_MPIPE_H_
16 #define _GXIO_MPIPE_H_
20 * An API for allocating, configuring, and manipulating mPIPE hardware
24 #include <gxio/common.h>
25 #include <gxio/dma_queue.h>
27 #include <linux/time.h>
29 #include <arch/mpipe_def.h>
30 #include <arch/mpipe_shm.h>
32 #include <hv/drv_mpipe_intf.h>
37 * The TILE-Gx mPIPE&tm; shim provides Ethernet connectivity, packet
38 * classification, and packet load balancing services. The
39 * gxio_mpipe_ API, declared in <gxio/mpipe.h>, allows applications to
40 * allocate mPIPE IO channels, configure packet distribution
41 * parameters, and send and receive Ethernet packets. The API is
42 * designed to be a minimal wrapper around the mPIPE hardware, making
43 * system calls only where necessary to preserve inter-process
44 * protection guarantees.
46 * The APIs described below allow the programmer to allocate and
47 * configure mPIPE resources. As described below, the mPIPE is a
48 * single shared hardware device that provides partitionable resources
49 * that are shared between all applications in the system. The
50 * gxio_mpipe_ API allows userspace code to make resource request
51 * calls to the hypervisor, which in turns keeps track of the
52 * resources in use by all applications, maintains protection
53 * guarantees, and resets resources upon application shutdown.
55 * We strongly recommend reading the mPIPE section of the IO Device
56 * Guide (UG404) before working with this API. Most functions in the
57 * gxio_mpipe_ API are directly analogous to hardware interfaces and
58 * the documentation assumes that the reader understands those
59 * hardware interfaces.
61 * @section mpipe__ingress mPIPE Ingress Hardware Resources
63 * The mPIPE ingress hardware provides extensive hardware offload for
64 * tasks like packet header parsing, load balancing, and memory
65 * management. This section provides a brief introduction to the
66 * hardware components and the gxio_mpipe_ calls used to manage them;
67 * see the IO Device Guide for a much more detailed description of the
68 * mPIPE's capabilities.
70 * When a packet arrives at one of the mPIPE's Ethernet MACs, it is
71 * assigned a channel number indicating which MAC received it. It
72 * then proceeds through the following hardware pipeline:
74 * @subsection mpipe__classification Classification
76 * A set of classification processors run header parsing code on each
77 * incoming packet, extracting information including the destination
78 * MAC address, VLAN, Ethernet type, and five-tuple hash. Some of
79 * this information is then used to choose which buffer stack will be
80 * used to hold the packet, and which bucket will be used by the load
81 * balancer to determine which application will receive the packet.
83 * The rules by which the buffer stack and bucket are chosen can be
84 * configured via the @ref gxio_mpipe_classifier API. A given app can
85 * specify multiple rules, each one specifying a bucket range, and a
86 * set of buffer stacks, to be used for packets matching the rule.
87 * Each rule can optionally specify a restricted set of channels,
88 * VLANs, and/or dMACs, in which it is interested. By default, a
89 * given rule starts out matching all channels associated with the
90 * mPIPE context's set of open links; all VLANs; and all dMACs.
91 * Subsequent restrictions can then be added.
93 * @subsection mpipe__load_balancing Load Balancing
95 * The mPIPE load balancer is responsible for choosing the NotifRing
96 * to which the packet will be delivered. This decision is based on
97 * the bucket number indicated by the classification program. In
98 * general, the bucket number is based on some number of low bits of
99 * the packet's flow hash (applications that aren't interested in flow
100 * hashing use a single bucket). Each load balancer bucket keeps a
101 * record of the NotifRing to which packets directed to that bucket
102 * are currently being delivered. Based on the bucket's load
103 * balancing mode (@ref gxio_mpipe_bucket_mode_t), the load balancer
104 * either forwards the packet to the previously assigned NotifRing or
105 * decides to choose a new NotifRing. If a new NotifRing is required,
106 * the load balancer chooses the least loaded ring in the NotifGroup
107 * associated with the bucket.
109 * The load balancer is a shared resource. Each application needs to
110 * explicitly allocate NotifRings, NotifGroups, and buckets, using
111 * gxio_mpipe_alloc_notif_rings(), gxio_mpipe_alloc_notif_groups(),
112 * and gxio_mpipe_alloc_buckets(). Then the application needs to
113 * configure them using gxio_mpipe_init_notif_ring() and
114 * gxio_mpipe_init_notif_group_and_buckets().
116 * @subsection mpipe__buffers Buffer Selection and Packet Delivery
118 * Once the load balancer has chosen the destination NotifRing, the
119 * mPIPE DMA engine pops at least one buffer off of the 'buffer stack'
120 * chosen by the classification program and DMAs the packet data into
121 * that buffer. Each buffer stack provides a hardware-accelerated
122 * stack of data buffers with the same size. If the packet data is
123 * larger than the buffers provided by the chosen buffer stack, the
124 * mPIPE hardware pops off multiple buffers and chains the packet data
125 * through a multi-buffer linked list. Once the packet data is
126 * delivered to the buffer(s), the mPIPE hardware writes the
127 * ::gxio_mpipe_idesc_t metadata object (calculated by the classifier)
128 * into the NotifRing and increments the number of packets delivered
131 * Applications can push buffers onto a buffer stack by calling
132 * gxio_mpipe_push_buffer() or by egressing a packet with the
133 * ::gxio_mpipe_edesc_t::hwb bit set, indicating that the egressed
134 * buffers should be returned to the stack.
136 * Applications can allocate and initialize buffer stacks with the
137 * gxio_mpipe_alloc_buffer_stacks() and gxio_mpipe_init_buffer_stack()
140 * The application must also register the memory pages that will hold
141 * packets. This requires calling gxio_mpipe_register_page() for each
142 * memory page that will hold packets allocated by the application for
143 * a given buffer stack. Since each buffer stack is limited to 16
144 * registered pages, it may be necessary to use huge pages, or even
145 * extremely huge pages, to hold all the buffers.
147 * @subsection mpipe__iqueue NotifRings
149 * Each NotifRing is a region of shared memory, allocated by the
150 * application, to which the mPIPE delivers packet descriptors
151 * (::gxio_mpipe_idesc_t). The application can allocate them via
152 * gxio_mpipe_alloc_notif_rings(). The application can then either
153 * explicitly initialize them with gxio_mpipe_init_notif_ring() and
154 * then read from them manually, or can make use of the convenience
155 * wrappers provided by @ref gxio_mpipe_wrappers.
157 * @section mpipe__egress mPIPE Egress Hardware
159 * Applications use eDMA rings to queue packets for egress. The
160 * application can allocate them via gxio_mpipe_alloc_edma_rings().
161 * The application can then either explicitly initialize them with
162 * gxio_mpipe_init_edma_ring() and then write to them manually, or
163 * can make use of the convenience wrappers provided by
164 * @ref gxio_mpipe_wrappers.
166 * @section gxio__shortcomings Plans for Future API Revisions
168 * The API defined here is only an initial version of the mPIPE API.
169 * Future plans include:
171 * - Higher level wrapper functions to provide common initialization
172 * patterns. This should help users start writing mPIPE programs
173 * without having to learn the details of the hardware.
175 * - Support for reset and deallocation of resources, including
176 * cleanup upon application shutdown.
178 * - Support for calling these APIs in the BME.
180 * - Support for IO interrupts.
182 * - Clearer definitions of thread safety guarantees.
184 * @section gxio__mpipe_examples Examples
186 * See the following mPIPE example programs for more information about
187 * allocating mPIPE resources and using them in real applications:
189 * - @ref mpipe/ingress/app.c : Receiving packets.
191 * - @ref mpipe/forward/app.c : Forwarding packets.
193 * Note that there are several more examples.
196 /* Flags that can be passed to resource allocation functions. */
197 enum gxio_mpipe_alloc_flags_e
{
198 /* Require an allocation to start at a specified resource index. */
199 GXIO_MPIPE_ALLOC_FIXED
= HV_MPIPE_ALLOC_FIXED
,
202 /* Flags that can be passed to memory registration functions. */
203 enum gxio_mpipe_mem_flags_e
{
204 /* Do not fill L3 when writing, and invalidate lines upon egress. */
205 GXIO_MPIPE_MEM_FLAG_NT_HINT
= IORPC_MEM_BUFFER_FLAG_NT_HINT
,
207 /* L3 cache fills should only populate IO cache ways. */
208 GXIO_MPIPE_MEM_FLAG_IO_PIN
= IORPC_MEM_BUFFER_FLAG_IO_PIN
,
211 /* An ingress packet descriptor. When a packet arrives, the mPIPE
212 * hardware generates this structure and writes it into a NotifRing.
214 typedef MPIPE_PDESC_t gxio_mpipe_idesc_t
;
216 /* An egress command descriptor. Applications write this structure
217 * into eDMA rings and the hardware performs the indicated operation
218 * (normally involving egressing some bytes). Note that egressing a
219 * single packet may involve multiple egress command descriptors.
221 typedef MPIPE_EDMA_DESC_t gxio_mpipe_edesc_t
;
224 * Max # of mpipe instances. 2 currently.
226 #define GXIO_MPIPE_INSTANCE_MAX HV_MPIPE_INSTANCE_MAX
228 #define NR_MPIPE_MAX GXIO_MPIPE_INSTANCE_MAX
230 /* Get the "va" field from an "idesc".
232 * This is the address at which the ingress hardware copied the first
233 * byte of the packet.
235 * If the classifier detected a custom header, then this will point to
236 * the custom header, and gxio_mpipe_idesc_get_l2_start() will point
237 * to the actual L2 header.
239 * Note that this value may be misleading if "idesc->be" is set.
241 * @param idesc An ingress packet descriptor.
243 static inline unsigned char *gxio_mpipe_idesc_get_va(gxio_mpipe_idesc_t
*idesc
)
245 return (unsigned char *)(long)idesc
->va
;
248 /* Get the "xfer_size" from an "idesc".
250 * This is the actual number of packet bytes transferred into memory
253 * Note that this value may be misleading if "idesc->be" is set.
255 * @param idesc An ingress packet descriptor.
257 * ISSUE: Is this the best name for this?
258 * FIXME: Add more docs about chaining, clipping, etc.
260 static inline unsigned int gxio_mpipe_idesc_get_xfer_size(gxio_mpipe_idesc_t
263 return idesc
->l2_size
;
266 /* Get the "l2_offset" from an "idesc".
268 * Extremely customized classifiers might not support this function.
270 * This is the number of bytes between the "va" and the L2 header.
272 * The L2 header consists of a destination mac address, a source mac
273 * address, and an initial ethertype. Various initial ethertypes
274 * allow encoding extra information in the L2 header, often including
275 * a vlan, and/or a new ethertype.
277 * Note that the "l2_offset" will be non-zero if (and only if) the
278 * classifier processed a custom header for the packet.
280 * @param idesc An ingress packet descriptor.
282 static inline uint8_t gxio_mpipe_idesc_get_l2_offset(gxio_mpipe_idesc_t
*idesc
)
284 return (idesc
->custom1
>> 32) & 0xFF;
287 /* Get the "l2_start" from an "idesc".
289 * This is simply gxio_mpipe_idesc_get_va() plus
290 * gxio_mpipe_idesc_get_l2_offset().
292 * @param idesc An ingress packet descriptor.
294 static inline unsigned char *gxio_mpipe_idesc_get_l2_start(gxio_mpipe_idesc_t
297 unsigned char *va
= gxio_mpipe_idesc_get_va(idesc
);
298 return va
+ gxio_mpipe_idesc_get_l2_offset(idesc
);
301 /* Get the "l2_length" from an "idesc".
303 * This is simply gxio_mpipe_idesc_get_xfer_size() minus
304 * gxio_mpipe_idesc_get_l2_offset().
306 * @param idesc An ingress packet descriptor.
308 static inline unsigned int gxio_mpipe_idesc_get_l2_length(gxio_mpipe_idesc_t
311 unsigned int xfer_size
= idesc
->l2_size
;
312 return xfer_size
- gxio_mpipe_idesc_get_l2_offset(idesc
);
315 /* A context object used to manage mPIPE hardware resources. */
318 /* File descriptor for calling up to Linux (and thus the HV). */
321 /* Corresponding mpipe instance #. */
324 /* The VA at which configuration registers are mapped. */
327 /* The VA at which IDMA, EDMA, and buffer manager are mapped. */
328 char *mmio_fast_base
;
330 /* The "initialized" buffer stacks. */
331 gxio_mpipe_rules_stacks_t __stacks
;
333 } gxio_mpipe_context_t
;
335 /* This is only used internally, but it's most easily made visible here. */
336 typedef gxio_mpipe_context_t gxio_mpipe_info_context_t
;
338 /* Initialize an mPIPE context.
340 * This function allocates an mPIPE "service domain" and maps the MMIO
341 * registers into the caller's VA space.
343 * @param context Context object to be initialized.
344 * @param mpipe_instance Instance number of mPIPE shim to be controlled via
347 extern int gxio_mpipe_init(gxio_mpipe_context_t
*context
,
348 unsigned int mpipe_instance
);
350 /* Destroy an mPIPE context.
352 * This function frees the mPIPE "service domain" and unmaps the MMIO
353 * registers from the caller's VA space.
355 * If a user process exits without calling this routine, the kernel
356 * will destroy the mPIPE context as part of process teardown.
358 * @param context Context object to be destroyed.
360 extern int gxio_mpipe_destroy(gxio_mpipe_context_t
*context
);
362 /*****************************************************************
364 ******************************************************************/
366 /* Allocate a set of buffer stacks.
368 * The return value is NOT interesting if count is zero.
370 * @param context An initialized mPIPE context.
371 * @param count Number of stacks required.
372 * @param first Index of first stack if ::GXIO_MPIPE_ALLOC_FIXED flag is set,
374 * @param flags Flag bits from ::gxio_mpipe_alloc_flags_e.
375 * @return Index of first allocated buffer stack, or
376 * ::GXIO_MPIPE_ERR_NO_BUFFER_STACK if allocation failed.
378 extern int gxio_mpipe_alloc_buffer_stacks(gxio_mpipe_context_t
*context
,
383 /* Enum codes for buffer sizes supported by mPIPE. */
385 /* 128 byte packet data buffer. */
386 GXIO_MPIPE_BUFFER_SIZE_128
= MPIPE_BSM_INIT_DAT_1__SIZE_VAL_BSZ_128
,
387 /* 256 byte packet data buffer. */
388 GXIO_MPIPE_BUFFER_SIZE_256
= MPIPE_BSM_INIT_DAT_1__SIZE_VAL_BSZ_256
,
389 /* 512 byte packet data buffer. */
390 GXIO_MPIPE_BUFFER_SIZE_512
= MPIPE_BSM_INIT_DAT_1__SIZE_VAL_BSZ_512
,
391 /* 1024 byte packet data buffer. */
392 GXIO_MPIPE_BUFFER_SIZE_1024
= MPIPE_BSM_INIT_DAT_1__SIZE_VAL_BSZ_1024
,
393 /* 1664 byte packet data buffer. */
394 GXIO_MPIPE_BUFFER_SIZE_1664
= MPIPE_BSM_INIT_DAT_1__SIZE_VAL_BSZ_1664
,
395 /* 4096 byte packet data buffer. */
396 GXIO_MPIPE_BUFFER_SIZE_4096
= MPIPE_BSM_INIT_DAT_1__SIZE_VAL_BSZ_4096
,
397 /* 10368 byte packet data buffer. */
398 GXIO_MPIPE_BUFFER_SIZE_10368
=
399 MPIPE_BSM_INIT_DAT_1__SIZE_VAL_BSZ_10368
,
400 /* 16384 byte packet data buffer. */
401 GXIO_MPIPE_BUFFER_SIZE_16384
= MPIPE_BSM_INIT_DAT_1__SIZE_VAL_BSZ_16384
402 } gxio_mpipe_buffer_size_enum_t
;
404 /* Convert a buffer size in bytes into a buffer size enum. */
405 extern gxio_mpipe_buffer_size_enum_t
406 gxio_mpipe_buffer_size_to_buffer_size_enum(size_t size
);
408 /* Convert a buffer size enum into a buffer size in bytes. */
410 gxio_mpipe_buffer_size_enum_to_buffer_size(gxio_mpipe_buffer_size_enum_t
413 /* Calculate the number of bytes required to store a given number of
414 * buffers in the memory registered with a buffer stack via
415 * gxio_mpipe_init_buffer_stack().
417 extern size_t gxio_mpipe_calc_buffer_stack_bytes(unsigned long buffers
);
419 /* Initialize a buffer stack. This function binds a region of memory
420 * to be used by the hardware for storing buffer addresses pushed via
421 * gxio_mpipe_push_buffer() or as the result of sending a buffer out
422 * the egress with the 'push to stack when done' bit set. Once this
423 * function returns, the memory region's contents may be arbitrarily
424 * modified by the hardware at any time and software should not access
425 * the memory region again.
427 * @param context An initialized mPIPE context.
428 * @param stack The buffer stack index.
429 * @param buffer_size_enum The size of each buffer in the buffer stack,
431 * @param mem The address of the buffer stack. This memory must be
432 * physically contiguous and aligned to a 64kB boundary.
433 * @param mem_size The size of the buffer stack, in bytes.
434 * @param mem_flags ::gxio_mpipe_mem_flags_e memory flags.
435 * @return Zero on success, ::GXIO_MPIPE_ERR_INVAL_BUFFER_SIZE if
436 * buffer_size_enum is invalid, ::GXIO_MPIPE_ERR_BAD_BUFFER_STACK if
437 * stack has not been allocated.
439 extern int gxio_mpipe_init_buffer_stack(gxio_mpipe_context_t
*context
,
441 gxio_mpipe_buffer_size_enum_t
442 buffer_size_enum
, void *mem
,
444 unsigned int mem_flags
);
446 /* Push a buffer onto a previously initialized buffer stack.
448 * The size of the buffer being pushed must match the size that was
449 * registered with gxio_mpipe_init_buffer_stack(). All packet buffer
450 * addresses are 128-byte aligned; the low 7 bits of the specified
451 * buffer address will be ignored.
453 * @param context An initialized mPIPE context.
454 * @param stack The buffer stack index.
455 * @param buffer The buffer (the low seven bits are ignored).
457 static inline void gxio_mpipe_push_buffer(gxio_mpipe_context_t
*context
,
458 unsigned int stack
, void *buffer
)
460 MPIPE_BSM_REGION_ADDR_t offset
= { {0} };
461 MPIPE_BSM_REGION_VAL_t val
= { {0} };
464 * The mmio_fast_base region starts at the IDMA region, so subtract
465 * off that initial offset.
468 MPIPE_MMIO_ADDR__REGION_VAL_BSM
-
469 MPIPE_MMIO_ADDR__REGION_VAL_IDMA
;
470 offset
.stack
= stack
;
472 #if __SIZEOF_POINTER__ == 4
473 val
.va
= ((ulong
) buffer
) >> MPIPE_BSM_REGION_VAL__VA_SHIFT
;
475 val
.va
= ((long)buffer
) >> MPIPE_BSM_REGION_VAL__VA_SHIFT
;
478 __gxio_mmio_write(context
->mmio_fast_base
+ offset
.word
, val
.word
);
481 /* Pop a buffer off of a previously initialized buffer stack.
483 * @param context An initialized mPIPE context.
484 * @param stack The buffer stack index.
485 * @return The buffer, or NULL if the stack is empty.
487 static inline void *gxio_mpipe_pop_buffer(gxio_mpipe_context_t
*context
,
490 MPIPE_BSM_REGION_ADDR_t offset
= { {0} };
493 * The mmio_fast_base region starts at the IDMA region, so subtract
494 * off that initial offset.
497 MPIPE_MMIO_ADDR__REGION_VAL_BSM
-
498 MPIPE_MMIO_ADDR__REGION_VAL_IDMA
;
499 offset
.stack
= stack
;
503 * Case 1: val.c == ..._UNCHAINED, va is non-zero.
504 * Case 2: val.c == ..._INVALID, va is zero.
505 * Case 3: val.c == ..._NOT_RDY, va is zero.
507 MPIPE_BSM_REGION_VAL_t val
;
509 __gxio_mmio_read(context
->mmio_fast_base
+
513 * Handle case 1 and 2 by returning the buffer (or NULL).
514 * Handle case 3 by waiting for the prefetch buffer to refill.
516 if (val
.c
!= MPIPE_EDMA_DESC_WORD1__C_VAL_NOT_RDY
)
517 return (void *)((unsigned long)val
.
518 va
<< MPIPE_BSM_REGION_VAL__VA_SHIFT
);
522 /*****************************************************************
524 ******************************************************************/
526 /* Allocate a set of NotifRings.
528 * The return value is NOT interesting if count is zero.
530 * Note that NotifRings are allocated in chunks, so allocating one at
531 * a time is much less efficient than allocating several at once.
533 * @param context An initialized mPIPE context.
534 * @param count Number of NotifRings required.
535 * @param first Index of first NotifRing if ::GXIO_MPIPE_ALLOC_FIXED flag
536 * is set, otherwise ignored.
537 * @param flags Flag bits from ::gxio_mpipe_alloc_flags_e.
538 * @return Index of first allocated buffer NotifRing, or
539 * ::GXIO_MPIPE_ERR_NO_NOTIF_RING if allocation failed.
541 extern int gxio_mpipe_alloc_notif_rings(gxio_mpipe_context_t
*context
,
542 unsigned int count
, unsigned int first
,
545 /* Initialize a NotifRing, using the given memory and size.
547 * @param context An initialized mPIPE context.
548 * @param ring The NotifRing index.
549 * @param mem A physically contiguous region of memory to be filled
550 * with a ring of ::gxio_mpipe_idesc_t structures.
551 * @param mem_size Number of bytes in the ring. Must be 128, 512,
552 * 2048, or 65536 * sizeof(gxio_mpipe_idesc_t).
553 * @param mem_flags ::gxio_mpipe_mem_flags_e memory flags.
555 * @return 0 on success, ::GXIO_MPIPE_ERR_BAD_NOTIF_RING or
556 * ::GXIO_ERR_INVAL_MEMORY_SIZE on failure.
558 extern int gxio_mpipe_init_notif_ring(gxio_mpipe_context_t
*context
,
560 void *mem
, size_t mem_size
,
561 unsigned int mem_flags
);
563 /* Configure an interrupt to be sent to a tile on incoming NotifRing
564 * traffic. Once an interrupt is sent for a particular ring, no more
565 * will be sent until gxio_mica_enable_notif_ring_interrupt() is called.
567 * @param context An initialized mPIPE context.
568 * @param x X coordinate of interrupt target tile.
569 * @param y Y coordinate of interrupt target tile.
570 * @param i Index of the IPI register which will receive the interrupt.
571 * @param e Specific event which will be set in the target IPI register when
572 * the interrupt occurs.
573 * @param ring The NotifRing index.
574 * @return Zero on success, GXIO_ERR_INVAL if params are out of range.
576 extern int gxio_mpipe_request_notif_ring_interrupt(gxio_mpipe_context_t
577 *context
, int x
, int y
,
581 /* Enable an interrupt on incoming NotifRing traffic.
583 * @param context An initialized mPIPE context.
584 * @param ring The NotifRing index.
585 * @return Zero on success, GXIO_ERR_INVAL if params are out of range.
587 extern int gxio_mpipe_enable_notif_ring_interrupt(gxio_mpipe_context_t
588 *context
, unsigned int ring
);
590 /* Map all of a client's memory via the given IOTLB.
591 * @param context An initialized mPIPE context.
592 * @param iotlb IOTLB index.
593 * @param pte Page table entry.
594 * @param flags Flags.
595 * @return Zero on success, or a negative error code.
597 extern int gxio_mpipe_register_client_memory(gxio_mpipe_context_t
*context
,
598 unsigned int iotlb
, HV_PTE pte
,
601 /*****************************************************************
603 ******************************************************************/
605 /* Allocate a set of NotifGroups.
607 * The return value is NOT interesting if count is zero.
609 * @param context An initialized mPIPE context.
610 * @param count Number of NotifGroups required.
611 * @param first Index of first NotifGroup if ::GXIO_MPIPE_ALLOC_FIXED flag
612 * is set, otherwise ignored.
613 * @param flags Flag bits from ::gxio_mpipe_alloc_flags_e.
614 * @return Index of first allocated buffer NotifGroup, or
615 * ::GXIO_MPIPE_ERR_NO_NOTIF_GROUP if allocation failed.
617 extern int gxio_mpipe_alloc_notif_groups(gxio_mpipe_context_t
*context
,
622 /* Add a NotifRing to a NotifGroup. This only sets a bit in the
623 * application's 'group' object; the hardware NotifGroup can be
624 * initialized by passing 'group' to gxio_mpipe_init_notif_group() or
625 * gxio_mpipe_init_notif_group_and_buckets().
628 gxio_mpipe_notif_group_add_ring(gxio_mpipe_notif_group_bits_t
*bits
, int ring
)
630 bits
->ring_mask
[ring
/ 64] |= (1ull << (ring
% 64));
633 /* Set a particular NotifGroup bitmask. Since the load balancer
634 * makes decisions based on both bucket and NotifGroup state, most
635 * applications should use gxio_mpipe_init_notif_group_and_buckets()
636 * rather than using this function to configure just a NotifGroup.
638 extern int gxio_mpipe_init_notif_group(gxio_mpipe_context_t
*context
,
640 gxio_mpipe_notif_group_bits_t bits
);
642 /*****************************************************************
644 ******************************************************************/
646 /* Allocate a set of load balancer buckets.
648 * The return value is NOT interesting if count is zero.
650 * Note that buckets are allocated in chunks, so allocating one at
651 * a time is much less efficient than allocating several at once.
653 * Note that the buckets are actually divided into two sub-ranges, of
654 * different sizes, and different chunk sizes, and the range you get
655 * by default is determined by the size of the request. Allocations
656 * cannot span the two sub-ranges.
658 * @param context An initialized mPIPE context.
659 * @param count Number of buckets required.
660 * @param first Index of first bucket if ::GXIO_MPIPE_ALLOC_FIXED flag is set,
662 * @param flags Flag bits from ::gxio_mpipe_alloc_flags_e.
663 * @return Index of first allocated buffer bucket, or
664 * ::GXIO_MPIPE_ERR_NO_BUCKET if allocation failed.
666 extern int gxio_mpipe_alloc_buckets(gxio_mpipe_context_t
*context
,
667 unsigned int count
, unsigned int first
,
670 /* The legal modes for gxio_mpipe_bucket_info_t and
671 * gxio_mpipe_init_notif_group_and_buckets().
673 * All modes except ::GXIO_MPIPE_BUCKET_ROUND_ROBIN expect that the user
674 * will allocate a power-of-two number of buckets and initialize them
675 * to the same mode. The classifier program then uses the appropriate
676 * number of low bits from the incoming packet's flow hash to choose a
677 * load balancer bucket. Based on that bucket's load balancing mode,
678 * reference count, and currently active NotifRing, the load balancer
679 * chooses the NotifRing to which the packet will be delivered.
682 /* All packets for a bucket go to the same NotifRing unless the
683 * NotifRing gets full, in which case packets will be dropped. If
684 * the bucket reference count ever reaches zero, a new NotifRing may
687 GXIO_MPIPE_BUCKET_DYNAMIC_FLOW_AFFINITY
=
688 MPIPE_LBL_INIT_DAT_BSTS_TBL__MODE_VAL_DFA
,
690 /* All packets for a bucket always go to the same NotifRing.
692 GXIO_MPIPE_BUCKET_STATIC_FLOW_AFFINITY
=
693 MPIPE_LBL_INIT_DAT_BSTS_TBL__MODE_VAL_FIXED
,
695 /* All packets for a bucket go to the least full NotifRing in the
696 * group, providing load balancing round robin behavior.
698 GXIO_MPIPE_BUCKET_ROUND_ROBIN
=
699 MPIPE_LBL_INIT_DAT_BSTS_TBL__MODE_VAL_ALWAYS_PICK
,
701 /* All packets for a bucket go to the same NotifRing unless the
702 * NotifRing gets full, at which point the bucket starts using the
703 * least full NotifRing in the group. If all NotifRings in the
704 * group are full, packets will be dropped.
706 GXIO_MPIPE_BUCKET_STICKY_FLOW_LOCALITY
=
707 MPIPE_LBL_INIT_DAT_BSTS_TBL__MODE_VAL_STICKY
,
709 /* All packets for a bucket go to the same NotifRing unless the
710 * NotifRing gets full, or a random timer fires, at which point the
711 * bucket starts using the least full NotifRing in the group. If
712 * all NotifRings in the group are full, packets will be dropped.
713 * WARNING: This mode is BROKEN on chips with fewer than 64 tiles.
715 GXIO_MPIPE_BUCKET_PREFER_FLOW_LOCALITY
=
716 MPIPE_LBL_INIT_DAT_BSTS_TBL__MODE_VAL_STICKY_RAND
,
718 } gxio_mpipe_bucket_mode_t
;
720 /* Copy a set of bucket initialization values into the mPIPE
721 * hardware. Since the load balancer makes decisions based on both
722 * bucket and NotifGroup state, most applications should use
723 * gxio_mpipe_init_notif_group_and_buckets() rather than using this
724 * function to configure a single bucket.
726 * @param context An initialized mPIPE context.
727 * @param bucket Bucket index to be initialized.
728 * @param bucket_info Initial reference count, NotifRing index, and mode.
729 * @return 0 on success, ::GXIO_MPIPE_ERR_BAD_BUCKET on failure.
731 extern int gxio_mpipe_init_bucket(gxio_mpipe_context_t
*context
,
733 gxio_mpipe_bucket_info_t bucket_info
);
735 /* Initializes a group and range of buckets and range of rings such
736 * that the load balancer runs a particular load balancing function.
738 * First, the group is initialized with the given rings.
740 * Second, each bucket is initialized with the mode and group, and a
741 * ring chosen round-robin from the given rings.
743 * Normally, the classifier picks a bucket, and then the load balancer
744 * picks a ring, based on the bucket's mode, group, and current ring,
745 * possibly updating the bucket's ring.
747 * @param context An initialized mPIPE context.
748 * @param group The group.
749 * @param ring The first ring.
750 * @param num_rings The number of rings.
751 * @param bucket The first bucket.
752 * @param num_buckets The number of buckets.
753 * @param mode The load balancing mode.
755 * @return 0 on success, ::GXIO_MPIPE_ERR_BAD_BUCKET,
756 * ::GXIO_MPIPE_ERR_BAD_NOTIF_GROUP, or
757 * ::GXIO_MPIPE_ERR_BAD_NOTIF_RING on failure.
759 extern int gxio_mpipe_init_notif_group_and_buckets(gxio_mpipe_context_t
763 unsigned int num_rings
,
765 unsigned int num_buckets
,
766 gxio_mpipe_bucket_mode_t
769 /* Return credits to a NotifRing and/or bucket.
771 * @param context An initialized mPIPE context.
772 * @param ring The NotifRing index, or -1.
773 * @param bucket The bucket, or -1.
774 * @param count The number of credits to return.
776 static inline void gxio_mpipe_credit(gxio_mpipe_context_t
*context
,
777 int ring
, int bucket
, unsigned int count
)
779 /* NOTE: Fancy struct initialization would break "C89" header test. */
781 MPIPE_IDMA_RELEASE_REGION_ADDR_t offset
= { {0} };
782 MPIPE_IDMA_RELEASE_REGION_VAL_t val
= { {0} };
785 * The mmio_fast_base region starts at the IDMA region, so subtract
786 * off that initial offset.
789 MPIPE_MMIO_ADDR__REGION_VAL_IDMA
-
790 MPIPE_MMIO_ADDR__REGION_VAL_IDMA
;
792 offset
.bucket
= bucket
;
793 offset
.ring_enable
= (ring
>= 0);
794 offset
.bucket_enable
= (bucket
>= 0);
797 __gxio_mmio_write(context
->mmio_fast_base
+ offset
.word
, val
.word
);
800 /*****************************************************************
802 ******************************************************************/
804 /* Allocate a set of eDMA rings.
806 * The return value is NOT interesting if count is zero.
808 * @param context An initialized mPIPE context.
809 * @param count Number of eDMA rings required.
810 * @param first Index of first eDMA ring if ::GXIO_MPIPE_ALLOC_FIXED flag
811 * is set, otherwise ignored.
812 * @param flags Flag bits from ::gxio_mpipe_alloc_flags_e.
813 * @return Index of first allocated buffer eDMA ring, or
814 * ::GXIO_MPIPE_ERR_NO_EDMA_RING if allocation failed.
816 extern int gxio_mpipe_alloc_edma_rings(gxio_mpipe_context_t
*context
,
817 unsigned int count
, unsigned int first
,
820 /* Initialize an eDMA ring, using the given memory and size.
822 * @param context An initialized mPIPE context.
823 * @param ering The eDMA ring index.
824 * @param channel The channel to use. This must be one of the channels
825 * associated with the context's set of open links.
826 * @param mem A physically contiguous region of memory to be filled
827 * with a ring of ::gxio_mpipe_edesc_t structures.
828 * @param mem_size Number of bytes in the ring. Must be 512, 2048,
829 * 8192 or 65536, times 16 (i.e. sizeof(gxio_mpipe_edesc_t)).
830 * @param mem_flags ::gxio_mpipe_mem_flags_e memory flags.
832 * @return 0 on success, ::GXIO_MPIPE_ERR_BAD_EDMA_RING or
833 * ::GXIO_ERR_INVAL_MEMORY_SIZE on failure.
835 extern int gxio_mpipe_init_edma_ring(gxio_mpipe_context_t
*context
,
836 unsigned int ering
, unsigned int channel
,
837 void *mem
, size_t mem_size
,
838 unsigned int mem_flags
);
840 /* Set the "max_blks", "min_snf_blks", and "db" fields of
841 * ::MPIPE_EDMA_RG_INIT_DAT_THRESH_t for a given edma ring.
843 * The global pool of dynamic blocks will be automatically adjusted.
845 * This function should not be called after any egress has been done
848 * Most applications should just use gxio_mpipe_equeue_set_snf_size().
850 * @param context An initialized mPIPE context.
851 * @param ering The eDMA ring index.
852 * @param max_blks The number of blocks to dedicate to the ring
853 * (normally min_snf_blks + 1). Must be greater than min_snf_blocks.
854 * @param min_snf_blks The number of blocks which must be stored
855 * prior to starting to send the packet (normally 12).
856 * @param db Whether to allow use of dynamic blocks by the ring
859 * @return 0 on success, negative on error.
861 extern int gxio_mpipe_config_edma_ring_blks(gxio_mpipe_context_t
*context
,
863 unsigned int max_blks
,
864 unsigned int min_snf_blks
,
867 /*****************************************************************
868 * Classifier Program *
869 ******************************************************************/
873 * Functions for loading or configuring the mPIPE classifier program.
875 * The mPIPE classification processors all run a special "classifier"
876 * program which, for each incoming packet, parses the packet headers,
877 * encodes some packet metadata in the "idesc", and either drops the
878 * packet, or picks a notif ring to handle the packet, and a buffer
879 * stack to contain the packet, usually based on the channel, VLAN,
880 * dMAC, flow hash, and packet size, under the guidance of the "rules"
881 * API described below.
883 * @section gxio_mpipe_classifier_default Default Classifier
885 * The MDE provides a simple "default" classifier program. It is
886 * shipped as source in "$TILERA_ROOT/src/sys/mpipe/classifier.c",
887 * which serves as its official documentation. It is shipped as a
888 * binary program in "$TILERA_ROOT/tile/boot/classifier", which is
889 * automatically included in bootroms created by "tile-monitor", and
890 * is automatically loaded by the hypervisor at boot time.
892 * The L2 analysis handles LLC packets, SNAP packets, and "VLAN
893 * wrappers" (keeping the outer VLAN).
895 * The L3 analysis handles IPv4 and IPv6, dropping packets with bad
896 * IPv4 header checksums, requesting computation of a TCP/UDP checksum
897 * if appropriate, and hashing the dest and src IP addresses, plus the
898 * ports for TCP/UDP packets, into the flow hash. No special analysis
899 * is done for "fragmented" packets or "tunneling" protocols. Thus,
900 * the first fragment of a fragmented TCP/UDP packet is hashed using
901 * src/dest IP address and ports and all subsequent fragments are only
902 * hashed according to src/dest IP address.
904 * The L3 analysis handles other packets too, hashing the dMAC
905 * smac into a flow hash.
907 * The channel, VLAN, and dMAC used to pick a "rule" (see the
908 * "rules" APIs below), which in turn is used to pick a buffer stack
909 * (based on the packet size) and a bucket (based on the flow hash).
911 * To receive traffic matching a particular (channel/VLAN/dMAC
912 * pattern, an application should allocate its own buffer stacks and
913 * load balancer buckets, and map traffic to those stacks and buckets,
914 * as decribed by the "rules" API below.
916 * Various packet metadata is encoded in the idesc. The flow hash is
917 * four bytes at 0x0C. The VLAN is two bytes at 0x10. The ethtype is
918 * two bytes at 0x12. The l3 start is one byte at 0x14. The l4 start
919 * is one byte at 0x15 for IPv4 and IPv6 packets, and otherwise zero.
920 * The protocol is one byte at 0x16 for IPv4 and IPv6 packets, and
923 * @section gxio_mpipe_classifier_custom Custom Classifiers.
925 * A custom classifier may be created using "tile-mpipe-cc" with a
926 * customized version of the default classifier sources.
928 * The custom classifier may be included in bootroms using the
929 * "--classifier" option to "tile-monitor", or loaded dynamically
930 * using gxio_mpipe_classifier_load_from_file().
932 * Be aware that "extreme" customizations may break the assumptions of
933 * the "rules" APIs described below, but simple customizations, such
934 * as adding new packet metadata, should be fine.
937 /* A set of classifier rules, plus a context. */
941 gxio_mpipe_context_t
*context
;
943 /* The actual rules. */
944 gxio_mpipe_rules_list_t list
;
946 } gxio_mpipe_rules_t
;
948 /* Initialize a classifier program rules list.
950 * This function can be called on a previously initialized rules list
951 * to discard any previously added rules.
953 * @param rules Rules list to initialize.
954 * @param context An initialized mPIPE context.
956 extern void gxio_mpipe_rules_init(gxio_mpipe_rules_t
*rules
,
957 gxio_mpipe_context_t
*context
);
959 /* Begin a new rule on the indicated rules list.
961 * Note that an empty rule matches all packets, but an empty rule list
962 * matches no packets.
964 * @param rules Rules list to which new rule is appended.
965 * @param bucket First load balancer bucket to which packets will be
967 * @param num_buckets Number of buckets (must be a power of two) across
968 * which packets will be distributed based on the "flow hash".
969 * @param stacks Either NULL, to assign each packet to the smallest
970 * initialized buffer stack which does not induce chaining (and to
971 * drop packets which exceed the largest initialized buffer stack
972 * buffer size), or an array, with each entry indicating which buffer
973 * stack should be used for packets up to that size (with 255
974 * indicating that those packets should be dropped).
975 * @return 0 on success, or a negative error code on failure.
977 extern int gxio_mpipe_rules_begin(gxio_mpipe_rules_t
*rules
,
979 unsigned int num_buckets
,
980 gxio_mpipe_rules_stacks_t
*stacks
);
982 /* Set the headroom of the current rule.
984 * @param rules Rules list whose current rule will be modified.
985 * @param headroom The headroom.
986 * @return 0 on success, or a negative error code on failure.
988 extern int gxio_mpipe_rules_set_headroom(gxio_mpipe_rules_t
*rules
,
991 /* Indicate that packets from a particular channel can be delivered
992 * to the buckets and buffer stacks associated with the current rule.
994 * Channels added must be associated with links opened by the mPIPE context
995 * used in gxio_mpipe_rules_init(). A rule with no channels is equivalent
996 * to a rule naming all such associated channels.
998 * @param rules Rules list whose current rule will be modified.
999 * @param channel The channel to add.
1000 * @return 0 on success, or a negative error code on failure.
1002 extern int gxio_mpipe_rules_add_channel(gxio_mpipe_rules_t
*rules
,
1003 unsigned int channel
);
1007 * The rules are sent to the hypervisor, where they are combined with
1008 * the rules from other apps, and used to program the hardware classifier.
1010 * Note that if this function returns an error, then the rules will NOT
1011 * have been committed, even if the error is due to interactions with
1012 * rules from another app.
1014 * @param rules Rules list to commit.
1015 * @return 0 on success, or a negative error code on failure.
1017 extern int gxio_mpipe_rules_commit(gxio_mpipe_rules_t
*rules
);
1019 /*****************************************************************
1020 * Ingress Queue Wrapper *
1021 ******************************************************************/
1025 * Convenience functions for receiving packets from a NotifRing and
1026 * sending packets via an eDMA ring.
1028 * The mpipe ingress and egress hardware uses shared memory packet
1029 * descriptors to describe packets that have arrived on ingress or
1030 * are destined for egress. These descriptors are stored in shared
1031 * memory ring buffers and written or read by hardware as necessary.
1032 * The gxio library provides wrapper functions that manage the head and
1033 * tail pointers for these rings, allowing the user to easily read or
1034 * write packet descriptors.
1036 * The initialization interface for ingress and egress rings is quite
1037 * similar. For example, to create an ingress queue, the user passes
1038 * a ::gxio_mpipe_iqueue_t state object, a ring number from
1039 * gxio_mpipe_alloc_notif_rings(), and the address of memory to hold a
1040 * ring buffer to the gxio_mpipe_iqueue_init() function. The function
1041 * returns success when the state object has been initialized and the
1042 * hardware configured to deliver packets to the specified ring
1043 * buffer. Similarly, gxio_mpipe_equeue_init() takes a
1044 * ::gxio_mpipe_equeue_t state object, a ring number from
1045 * gxio_mpipe_alloc_edma_rings(), and a shared memory buffer.
1047 * @section gxio_mpipe_iqueue Working with Ingress Queues
1049 * Once initialized, the gxio_mpipe_iqueue_t API provides two flows
1050 * for getting the ::gxio_mpipe_idesc_t packet descriptor associated
1051 * with incoming packets. The simplest is to call
1052 * gxio_mpipe_iqueue_get() or gxio_mpipe_iqueue_try_get(). These
1053 * functions copy the oldest packet descriptor out of the NotifRing and
1054 * into a descriptor provided by the caller. They also immediately
1055 * inform the hardware that a descriptor has been processed.
1057 * For applications with stringent performance requirements, higher
1058 * efficiency can be achieved by avoiding the packet descriptor copy
1059 * and processing multiple descriptors at once. The
1060 * gxio_mpipe_iqueue_peek() and gxio_mpipe_iqueue_try_peek() functions
1061 * allow such optimizations. These functions provide a pointer to the
1062 * next valid ingress descriptor in the NotifRing's shared memory ring
1063 * buffer, and a count of how many contiguous descriptors are ready to
1064 * be processed. The application can then process any number of those
1065 * descriptors in place, calling gxio_mpipe_iqueue_consume() to inform
1066 * the hardware after each one has been processed.
1068 * @section gxio_mpipe_equeue Working with Egress Queues
1070 * Similarly, the egress queue API provides a high-performance
1071 * interface plus a simple wrapper for use in posting
1072 * ::gxio_mpipe_edesc_t egress packet descriptors. The simple
1073 * version, gxio_mpipe_equeue_put(), allows the programmer to wait for
1074 * an eDMA ring slot to become available and write a single descriptor
1077 * Alternatively, you can reserve slots in the eDMA ring using
1078 * gxio_mpipe_equeue_reserve() or gxio_mpipe_equeue_try_reserve(), and
1079 * then fill in each slot using gxio_mpipe_equeue_put_at(). This
1080 * capability can be used to amortize the cost of reserving slots
1081 * across several packets. It also allows gather operations to be
1082 * performed on a shared equeue, by ensuring that the edescs for all
1083 * the fragments are all contiguous in the eDMA ring.
1085 * The gxio_mpipe_equeue_reserve() and gxio_mpipe_equeue_try_reserve()
1086 * functions return a 63-bit "completion slot", which is actually a
1087 * sequence number, the low bits of which indicate the ring buffer
1088 * index and the high bits the number of times the application has
1089 * gone around the egress ring buffer. The extra bits allow an
1090 * application to check for egress completion by calling
1091 * gxio_mpipe_equeue_is_complete() to see whether a particular 'slot'
1092 * number has finished. Given the maximum packet rates of the Gx
1093 * processor, the 63-bit slot number will never wrap.
1095 * In practice, most applications use the ::gxio_mpipe_edesc_t::hwb
1096 * bit to indicate that the buffers containing egress packet data
1097 * should be pushed onto a buffer stack when egress is complete. Such
1098 * applications generally do not need to know when an egress operation
1099 * completes (since there is no need to free a buffer post-egress),
1100 * and thus can use the optimized gxio_mpipe_equeue_reserve_fast() or
1101 * gxio_mpipe_equeue_try_reserve_fast() functions, which return a 24
1102 * bit "slot", instead of a 63-bit "completion slot".
1104 * Once a slot has been "reserved", it MUST be filled. If the
1105 * application reserves a slot and then decides that it does not
1106 * actually need it, it can set the ::gxio_mpipe_edesc_t::ns (no send)
1107 * bit on the descriptor passed to gxio_mpipe_equeue_put_at() to
1108 * indicate that no data should be sent. This technique can also be
1109 * used to drop an incoming packet, instead of forwarding it, since
1110 * any buffer will still be pushed onto the buffer stack when the
1111 * egress descriptor is processed.
1114 /* A convenient interface to a NotifRing, for use by a single thread.
1119 gxio_mpipe_context_t
*context
;
1121 /* The actual NotifRing. */
1122 gxio_mpipe_idesc_t
*idescs
;
1124 /* The number of entries. */
1125 unsigned long num_entries
;
1127 /* The number of entries minus one. */
1128 unsigned long mask_num_entries
;
1130 /* The log2() of the number of entries. */
1131 unsigned long log2_num_entries
;
1133 /* The next entry. */
1136 /* The NotifRing id. */
1139 #ifdef __BIG_ENDIAN__
1140 /* The number of byteswapped entries. */
1141 unsigned int swapped
;
1144 } gxio_mpipe_iqueue_t
;
1146 /* Initialize an "iqueue".
1148 * Takes the iqueue plus the same args as gxio_mpipe_init_notif_ring().
1150 extern int gxio_mpipe_iqueue_init(gxio_mpipe_iqueue_t
*iqueue
,
1151 gxio_mpipe_context_t
*context
,
1153 void *mem
, size_t mem_size
,
1154 unsigned int mem_flags
);
1156 /* Advance over some old entries in an iqueue.
1158 * Please see the documentation for gxio_mpipe_iqueue_consume().
1160 * @param iqueue An ingress queue initialized via gxio_mpipe_iqueue_init().
1161 * @param count The number of entries to advance over.
1163 static inline void gxio_mpipe_iqueue_advance(gxio_mpipe_iqueue_t
*iqueue
,
1166 /* Advance with proper wrap. */
1167 int head
= iqueue
->head
+ count
;
1169 (head
& iqueue
->mask_num_entries
) +
1170 (head
>> iqueue
->log2_num_entries
);
1172 #ifdef __BIG_ENDIAN__
1173 /* HACK: Track swapped entries. */
1174 iqueue
->swapped
-= count
;
1178 /* Release the ring and bucket for an old entry in an iqueue.
1180 * Releasing the ring allows more packets to be delivered to the ring.
1182 * Releasing the bucket allows flows using the bucket to be moved to a
1183 * new ring when using GXIO_MPIPE_BUCKET_DYNAMIC_FLOW_AFFINITY.
1185 * This function is shorthand for "gxio_mpipe_credit(iqueue->context,
1186 * iqueue->ring, idesc->bucket_id, 1)", and it may be more convenient
1187 * to make that underlying call, using those values, instead of
1188 * tracking the entire "idesc".
1190 * If packet processing is deferred, optimal performance requires that
1191 * the releasing be deferred as well.
1193 * Please see the documentation for gxio_mpipe_iqueue_consume().
1195 * @param iqueue An ingress queue initialized via gxio_mpipe_iqueue_init().
1196 * @param idesc The descriptor which was processed.
1198 static inline void gxio_mpipe_iqueue_release(gxio_mpipe_iqueue_t
*iqueue
,
1199 gxio_mpipe_idesc_t
*idesc
)
1201 gxio_mpipe_credit(iqueue
->context
, iqueue
->ring
, idesc
->bucket_id
, 1);
1204 /* Consume a packet from an "iqueue".
1206 * After processing packets peeked at via gxio_mpipe_iqueue_peek()
1207 * or gxio_mpipe_iqueue_try_peek(), you must call this function, or
1208 * gxio_mpipe_iqueue_advance() plus gxio_mpipe_iqueue_release(), to
1209 * advance over those entries, and release their rings and buckets.
1211 * You may call this function as each packet is processed, or you can
1212 * wait until several packets have been processed.
1214 * Note that if you are using a single bucket, and you are handling
1215 * batches of N packets, then you can replace several calls to this
1216 * function with calls to "gxio_mpipe_iqueue_advance(iqueue, N)" and
1217 * "gxio_mpipe_credit(iqueue->context, iqueue->ring, bucket, N)".
1219 * Note that if your classifier sets "idesc->nr", then you should
1220 * explicitly call "gxio_mpipe_iqueue_advance(iqueue, idesc)" plus
1221 * "gxio_mpipe_credit(iqueue->context, iqueue->ring, -1, 1)", to
1222 * avoid incorrectly crediting the (unused) bucket.
1224 * @param iqueue An ingress queue initialized via gxio_mpipe_iqueue_init().
1225 * @param idesc The descriptor which was processed.
1227 static inline void gxio_mpipe_iqueue_consume(gxio_mpipe_iqueue_t
*iqueue
,
1228 gxio_mpipe_idesc_t
*idesc
)
1230 gxio_mpipe_iqueue_advance(iqueue
, 1);
1231 gxio_mpipe_iqueue_release(iqueue
, idesc
);
1234 /* Peek at the next packet(s) in an "iqueue", without waiting.
1236 * If no packets are available, fills idesc_ref with NULL, and then
1237 * returns ::GXIO_MPIPE_ERR_IQUEUE_EMPTY. Otherwise, fills idesc_ref
1238 * with the address of the next valid packet descriptor, and returns
1239 * the maximum number of valid descriptors which can be processed.
1240 * You may process fewer descriptors if desired.
1242 * Call gxio_mpipe_iqueue_consume() on each packet once it has been
1243 * processed (or dropped), to allow more packets to be delivered.
1245 * @param iqueue An ingress queue initialized via gxio_mpipe_iqueue_init().
1246 * @param idesc_ref A pointer to a packet descriptor pointer.
1247 * @return The (positive) number of packets which can be processed,
1248 * or ::GXIO_MPIPE_ERR_IQUEUE_EMPTY if no packets are available.
1250 static inline int gxio_mpipe_iqueue_try_peek(gxio_mpipe_iqueue_t
*iqueue
,
1251 gxio_mpipe_idesc_t
**idesc_ref
)
1253 gxio_mpipe_idesc_t
*next
;
1255 uint64_t head
= iqueue
->head
;
1256 uint64_t tail
= __gxio_mmio_read(iqueue
->idescs
);
1258 /* Available entries. */
1260 (tail
>= head
) ? (tail
- head
) : (iqueue
->num_entries
- head
);
1264 return GXIO_MPIPE_ERR_IQUEUE_EMPTY
;
1267 next
= &iqueue
->idescs
[head
];
1269 /* ISSUE: Is this helpful? */
1270 __insn_prefetch(next
);
1272 #ifdef __BIG_ENDIAN__
1273 /* HACK: Swap new entries directly in memory. */
1276 for (i
= iqueue
->swapped
; i
< avail
; i
++) {
1277 for (j
= 0; j
< 8; j
++)
1279 __builtin_bswap64(next
[i
].words
[j
]);
1281 iqueue
->swapped
= avail
;
1290 /* Drop a packet by pushing its buffer (if appropriate).
1292 * NOTE: The caller must still call gxio_mpipe_iqueue_consume() if idesc
1293 * came from gxio_mpipe_iqueue_try_peek() or gxio_mpipe_iqueue_peek().
1295 * @param iqueue An ingress queue initialized via gxio_mpipe_iqueue_init().
1296 * @param idesc A packet descriptor.
1298 static inline void gxio_mpipe_iqueue_drop(gxio_mpipe_iqueue_t
*iqueue
,
1299 gxio_mpipe_idesc_t
*idesc
)
1301 /* FIXME: Handle "chaining" properly. */
1304 unsigned char *va
= gxio_mpipe_idesc_get_va(idesc
);
1305 gxio_mpipe_push_buffer(iqueue
->context
, idesc
->stack_idx
, va
);
1309 /*****************************************************************
1310 * Egress Queue Wrapper *
1311 ******************************************************************/
1313 /* A convenient, thread-safe interface to an eDMA ring. */
1316 /* State object for tracking head and tail pointers. */
1317 __gxio_dma_queue_t dma_queue
;
1319 /* The ring entries. */
1320 gxio_mpipe_edesc_t
*edescs
;
1322 /* The number of entries minus one. */
1323 unsigned long mask_num_entries
;
1325 /* The log2() of the number of entries. */
1326 unsigned long log2_num_entries
;
1329 gxio_mpipe_context_t
*context
;
1335 unsigned int channel
;
1337 } gxio_mpipe_equeue_t
;
1339 /* Initialize an "equeue".
1341 * This function uses gxio_mpipe_init_edma_ring() to initialize the
1342 * underlying edma_ring using the provided arguments.
1344 * @param equeue An egress queue to be initialized.
1345 * @param context An initialized mPIPE context.
1346 * @param ering The eDMA ring index.
1347 * @param channel The channel to use. This must be one of the channels
1348 * associated with the context's set of open links.
1349 * @param mem A physically contiguous region of memory to be filled
1350 * with a ring of ::gxio_mpipe_edesc_t structures.
1351 * @param mem_size Number of bytes in the ring. Must be 512, 2048,
1352 * 8192 or 65536, times 16 (i.e. sizeof(gxio_mpipe_edesc_t)).
1353 * @param mem_flags ::gxio_mpipe_mem_flags_e memory flags.
1355 * @return 0 on success, ::GXIO_MPIPE_ERR_BAD_EDMA_RING or
1356 * ::GXIO_ERR_INVAL_MEMORY_SIZE on failure.
1358 extern int gxio_mpipe_equeue_init(gxio_mpipe_equeue_t
*equeue
,
1359 gxio_mpipe_context_t
*context
,
1361 unsigned int channel
,
1362 void *mem
, unsigned int mem_size
,
1363 unsigned int mem_flags
);
1365 /* Reserve completion slots for edescs.
1367 * Use gxio_mpipe_equeue_put_at() to actually populate the slots.
1369 * This function is slower than gxio_mpipe_equeue_reserve_fast(), but
1370 * returns a full 64 bit completion slot, which can be used with
1371 * gxio_mpipe_equeue_is_complete().
1373 * @param equeue An egress queue initialized via gxio_mpipe_equeue_init().
1374 * @param num Number of slots to reserve (must be non-zero).
1375 * @return The first reserved completion slot, or a negative error code.
1377 static inline int64_t gxio_mpipe_equeue_reserve(gxio_mpipe_equeue_t
*equeue
,
1380 return __gxio_dma_queue_reserve_aux(&equeue
->dma_queue
, num
, true);
1383 /* Reserve completion slots for edescs, if possible.
1385 * Use gxio_mpipe_equeue_put_at() to actually populate the slots.
1387 * This function is slower than gxio_mpipe_equeue_try_reserve_fast(),
1388 * but returns a full 64 bit completion slot, which can be used with
1389 * gxio_mpipe_equeue_is_complete().
1391 * @param equeue An egress queue initialized via gxio_mpipe_equeue_init().
1392 * @param num Number of slots to reserve (must be non-zero).
1393 * @return The first reserved completion slot, or a negative error code.
1395 static inline int64_t gxio_mpipe_equeue_try_reserve(gxio_mpipe_equeue_t
1396 *equeue
, unsigned int num
)
1398 return __gxio_dma_queue_reserve_aux(&equeue
->dma_queue
, num
, false);
1401 /* Reserve slots for edescs.
1403 * Use gxio_mpipe_equeue_put_at() to actually populate the slots.
1405 * This function is faster than gxio_mpipe_equeue_reserve(), but
1406 * returns a 24 bit slot (instead of a 64 bit completion slot), which
1407 * thus cannot be used with gxio_mpipe_equeue_is_complete().
1409 * @param equeue An egress queue initialized via gxio_mpipe_equeue_init().
1410 * @param num Number of slots to reserve (should be non-zero).
1411 * @return The first reserved slot, or a negative error code.
1413 static inline int64_t gxio_mpipe_equeue_reserve_fast(gxio_mpipe_equeue_t
1414 *equeue
, unsigned int num
)
1416 return __gxio_dma_queue_reserve(&equeue
->dma_queue
, num
, true, false);
1419 /* Reserve slots for edescs, if possible.
1421 * Use gxio_mpipe_equeue_put_at() to actually populate the slots.
1423 * This function is faster than gxio_mpipe_equeue_try_reserve(), but
1424 * returns a 24 bit slot (instead of a 64 bit completion slot), which
1425 * thus cannot be used with gxio_mpipe_equeue_is_complete().
1427 * @param equeue An egress queue initialized via gxio_mpipe_equeue_init().
1428 * @param num Number of slots to reserve (should be non-zero).
1429 * @return The first reserved slot, or a negative error code.
1431 static inline int64_t gxio_mpipe_equeue_try_reserve_fast(gxio_mpipe_equeue_t
1435 return __gxio_dma_queue_reserve(&equeue
->dma_queue
, num
, false, false);
1439 * HACK: This helper function tricks gcc 4.6 into avoiding saving
1440 * a copy of "edesc->words[0]" on the stack for no obvious reason.
1443 static inline void gxio_mpipe_equeue_put_at_aux(gxio_mpipe_equeue_t
*equeue
,
1447 unsigned long edma_slot
= slot
& equeue
->mask_num_entries
;
1448 gxio_mpipe_edesc_t
*edesc_p
= &equeue
->edescs
[edma_slot
];
1451 * ISSUE: Could set eDMA ring to be on generation 1 at start, which
1452 * would avoid the negation here, perhaps allowing "__insn_bfins()".
1454 ew
[0] |= !((slot
>> equeue
->log2_num_entries
) & 1);
1457 * NOTE: We use "__gxio_mpipe_write()", plus the fact that the eDMA
1458 * queue alignment restrictions ensure that these two words are on
1459 * the same cacheline, to force proper ordering between the stores.
1461 __gxio_mmio_write64(&edesc_p
->words
[1], ew
[1]);
1462 __gxio_mmio_write64(&edesc_p
->words
[0], ew
[0]);
1465 /* Post an edesc to a given slot in an equeue.
1467 * This function copies the supplied edesc into entry "slot mod N" in
1468 * the underlying ring, setting the "gen" bit to the appropriate value
1469 * based on "(slot mod N*2)", where "N" is the size of the ring. Note
1470 * that the higher bits of slot are unused, and thus, this function
1471 * can handle "slots" as well as "completion slots".
1473 * Normally this function is used to fill in slots reserved by
1474 * gxio_mpipe_equeue_try_reserve(), gxio_mpipe_equeue_reserve(),
1475 * gxio_mpipe_equeue_try_reserve_fast(), or
1476 * gxio_mpipe_equeue_reserve_fast(),
1478 * This function can also be used without "reserving" slots, if the
1479 * application KNOWS that the ring can never overflow, for example, by
1480 * pushing fewer buffers into the buffer stacks than there are total
1481 * slots in the equeue, but this is NOT recommended.
1483 * @param equeue An egress queue initialized via gxio_mpipe_equeue_init().
1484 * @param edesc The egress descriptor to be posted.
1485 * @param slot An egress slot (only the low bits are actually used).
1487 static inline void gxio_mpipe_equeue_put_at(gxio_mpipe_equeue_t
*equeue
,
1488 gxio_mpipe_edesc_t edesc
,
1491 gxio_mpipe_equeue_put_at_aux(equeue
, edesc
.words
, slot
);
1494 /* Post an edesc to the next slot in an equeue.
1496 * This is a convenience wrapper around
1497 * gxio_mpipe_equeue_reserve_fast() and gxio_mpipe_equeue_put_at().
1499 * @param equeue An egress queue initialized via gxio_mpipe_equeue_init().
1500 * @param edesc The egress descriptor to be posted.
1501 * @return 0 on success.
1503 static inline int gxio_mpipe_equeue_put(gxio_mpipe_equeue_t
*equeue
,
1504 gxio_mpipe_edesc_t edesc
)
1506 int64_t slot
= gxio_mpipe_equeue_reserve_fast(equeue
, 1);
1510 gxio_mpipe_equeue_put_at(equeue
, edesc
, slot
);
1515 /* Ask the mPIPE hardware to egress outstanding packets immediately.
1517 * This call is not necessary, but may slightly reduce overall latency.
1519 * Technically, you should flush all gxio_mpipe_equeue_put_at() writes
1520 * to memory before calling this function, to ensure the descriptors
1521 * are visible in memory before the mPIPE hardware actually looks for
1522 * them. But this should be very rare, and the only side effect would
1523 * be increased latency, so it is up to the caller to decide whether
1524 * or not to flush memory.
1526 * @param equeue An egress queue initialized via gxio_mpipe_equeue_init().
1528 static inline void gxio_mpipe_equeue_flush(gxio_mpipe_equeue_t
*equeue
)
1530 /* Use "ring_idx = 0" and "count = 0" to "wake up" the eDMA ring. */
1531 MPIPE_EDMA_POST_REGION_VAL_t val
= { {0} };
1532 /* Flush the write buffers. */
1534 __gxio_mmio_write(equeue
->dma_queue
.post_region_addr
, val
.word
);
1537 /* Determine if a given edesc has been completed.
1539 * Note that this function requires a "completion slot", and thus may
1540 * NOT be used with a "slot" from gxio_mpipe_equeue_reserve_fast() or
1541 * gxio_mpipe_equeue_try_reserve_fast().
1543 * @param equeue An egress queue initialized via gxio_mpipe_equeue_init().
1544 * @param completion_slot The completion slot used by the edesc.
1545 * @param update If true, and the desc does not appear to have completed
1546 * yet, then update any software cache of the hardware completion counter,
1547 * and check again. This should normally be true.
1548 * @return True iff the given edesc has been completed.
1550 static inline int gxio_mpipe_equeue_is_complete(gxio_mpipe_equeue_t
*equeue
,
1551 int64_t completion_slot
,
1554 return __gxio_dma_queue_is_complete(&equeue
->dma_queue
,
1555 completion_slot
, update
);
1558 /* Set the snf (store and forward) size for an equeue.
1560 * The snf size for an equeue defaults to 1536, and encodes the size
1561 * of the largest packet for which egress is guaranteed to avoid
1562 * transmission underruns and/or corrupt checksums under heavy load.
1564 * The snf size affects a global resource pool which cannot support,
1565 * for example, all 24 equeues each requesting an snf size of 8K.
1567 * To ensure that jumbo packets can be egressed properly, the snf size
1568 * should be set to the size of the largest possible packet, which
1569 * will usually be limited by the size of the app's largest buffer.
1571 * This is a convenience wrapper around
1572 * gxio_mpipe_config_edma_ring_blks().
1574 * This function should not be called after any egress has been done
1577 * @param equeue An egress queue initialized via gxio_mpipe_equeue_init().
1578 * @param size The snf size, in bytes.
1579 * @return Zero on success, negative error otherwise.
1581 static inline int gxio_mpipe_equeue_set_snf_size(gxio_mpipe_equeue_t
*equeue
,
1584 int blks
= (size
+ 127) / 128;
1585 return gxio_mpipe_config_edma_ring_blks(equeue
->context
, equeue
->ering
,
1589 /*****************************************************************
1591 ******************************************************************/
1595 * Functions for manipulating and sensing the state and configuration
1596 * of physical network links.
1598 * @section gxio_mpipe_link_perm Link Permissions
1600 * Opening a link (with gxio_mpipe_link_open()) requests a set of link
1601 * permissions, which control what may be done with the link, and potentially
1602 * what permissions may be granted to other processes.
1604 * Data permission allows the process to receive packets from the link by
1605 * specifying the link's channel number in mPIPE packet distribution rules,
1606 * and to send packets to the link by using the link's channel number as
1607 * the target for an eDMA ring.
1609 * Stats permission allows the process to retrieve link attributes (such as
1610 * the speeds it is capable of running at, or whether it is currently up), and
1611 * to read and write certain statistics-related registers in the link's MAC.
1613 * Control permission allows the process to retrieve and modify link attributes
1614 * (so that it may, for example, bring the link up and take it down), and
1615 * read and write many registers in the link's MAC and PHY.
1617 * Any permission may be requested as shared, which allows other processes
1618 * to also request shared permission, or exclusive, which prevents other
1619 * processes from requesting it. In keeping with GXIO's typical usage in
1620 * an embedded environment, the defaults for all permissions are shared.
1622 * Permissions are granted on a first-come, first-served basis, so if two
1623 * applications request an exclusive permission on the same link, the one
1624 * to run first will win. Note, however, that some system components, like
1625 * the kernel Ethernet driver, may get an opportunity to open links before
1626 * any applications run.
1628 * @section gxio_mpipe_link_names Link Names
1630 * Link names are of the form gbe<em>number</em> (for Gigabit Ethernet),
1631 * xgbe<em>number</em> (for 10 Gigabit Ethernet), loop<em>number</em> (for
1632 * internal mPIPE loopback), or ilk<em>number</em>/<em>channel</em>
1633 * (for Interlaken links); for instance, gbe0, xgbe1, loop3, and
1634 * ilk0/12 are all possible link names. The correspondence between
1635 * the link name and an mPIPE instance number or mPIPE channel number is
1636 * system-dependent; all links will not exist on all systems, and the set
1637 * of numbers used for a particular link type may not start at zero and may
1638 * not be contiguous. Use gxio_mpipe_link_enumerate() to retrieve the set of
1639 * links which exist on a system, and always use gxio_mpipe_link_instance()
1640 * to determine which mPIPE controls a particular link.
1642 * Note that in some cases, links may share hardware, such as PHYs, or
1643 * internal mPIPE buffers; in these cases, only one of the links may be
1644 * opened at a time. This is especially common with xgbe and gbe ports,
1645 * since each xgbe port uses 4 SERDES lanes, each of which may also be
1646 * configured as one gbe port.
1648 * @section gxio_mpipe_link_states Link States
1650 * The mPIPE link management model revolves around three different states,
1651 * which are maintained for each link:
1653 * 1. The <em>current</em> link state: is the link up now, and if so, at
1656 * 2. The <em>desired</em> link state: what do we want the link state to be?
1657 * The system is always working to make this state the current state;
1658 * thus, if the desired state is up, and the link is down, we'll be
1659 * constantly trying to bring it up, automatically.
1661 * 3. The <em>possible</em> link state: what speeds are valid for this
1662 * particular link? Or, in other words, what are the capabilities of
1663 * the link hardware?
1665 * These link states are not, strictly speaking, related to application
1666 * state; they may be manipulated at any time, whether or not the link
1667 * is currently being used for data transfer. However, for convenience,
1668 * gxio_mpipe_link_open() and gxio_mpipe_link_close() (or application exit)
1669 * can affect the link state. These implicit link management operations
1670 * may be modified or disabled by the use of link open flags.
1672 * From an application, you can use gxio_mpipe_link_get_attr()
1673 * and gxio_mpipe_link_set_attr() to manipulate the link states.
1674 * gxio_mpipe_link_get_attr() with ::GXIO_MPIPE_LINK_POSSIBLE_STATE
1675 * gets you the possible link state. gxio_mpipe_link_get_attr() with
1676 * ::GXIO_MPIPE_LINK_CURRENT_STATE gets you the current link state.
1677 * Finally, gxio_mpipe_link_set_attr() and gxio_mpipe_link_get_attr()
1678 * with ::GXIO_MPIPE_LINK_DESIRED_STATE allow you to modify or retrieve
1679 * the desired link state.
1681 * If you want to manage a link from a part of your application which isn't
1682 * involved in packet processing, you can use the ::GXIO_MPIPE_LINK_NO_DATA
1683 * flags on a gxio_mpipe_link_open() call. This opens the link, but does
1684 * not request data permission, so it does not conflict with any exclusive
1685 * permissions which may be held by other processes. You can then can use
1686 * gxio_mpipe_link_get_attr() and gxio_mpipe_link_set_attr() on this link
1687 * object to bring up or take down the link.
1689 * Some links support link state bits which support various loopback
1690 * modes. ::GXIO_MPIPE_LINK_LOOP_MAC tests datapaths within the Tile
1691 * Processor itself; ::GXIO_MPIPE_LINK_LOOP_PHY tests the datapath between
1692 * the Tile Processor and the external physical layer interface chip; and
1693 * ::GXIO_MPIPE_LINK_LOOP_EXT tests the entire network datapath with the
1694 * aid of an external loopback connector. In addition to enabling hardware
1695 * testing, such configuration can be useful for software testing, as well.
1697 * When LOOP_MAC or LOOP_PHY is enabled, packets transmitted on a channel
1698 * will be received by that channel, instead of being emitted on the
1699 * physical link, and packets received on the physical link will be ignored.
1700 * Other than that, all standard GXIO operations work as you might expect.
1701 * Note that loopback operation requires that the link be brought up using
1702 * one or more of the GXIO_MPIPE_LINK_SPEED_xxx link state bits.
1704 * Those familiar with previous versions of the MDE on TILEPro hardware
1705 * will notice significant similarities between the NetIO link management
1706 * model and the mPIPE link management model. However, the NetIO model
1707 * was developed in stages, and some of its features -- for instance,
1708 * the default setting of certain flags -- were shaped by the need to be
1709 * compatible with previous versions of NetIO. Since the features provided
1710 * by the mPIPE hardware and the mPIPE GXIO library are significantly
1711 * different than those provided by NetIO, in some cases, we have made
1712 * different choices in the mPIPE link management API. Thus, please read
1713 * this documentation carefully before assuming that mPIPE link management
1714 * operations are exactly equivalent to their NetIO counterparts.
1717 /* An object used to manage mPIPE link state and resources. */
1719 /* The overall mPIPE context. */
1720 gxio_mpipe_context_t
*context
;
1722 /* The channel number used by this link. */
1725 /* The MAC index used by this link. */
1727 } gxio_mpipe_link_t
;
1729 /* Translate a link name to the instance number of the mPIPE shim which is
1730 * connected to that link. This call does not verify whether the link is
1731 * currently available, and does not reserve any link resources;
1732 * gxio_mpipe_link_open() must be called to perform those functions.
1734 * Typically applications will call this function to translate a link name
1735 * to an mPIPE instance number; call gxio_mpipe_init(), passing it that
1736 * instance number, to initialize the mPIPE shim; and then call
1737 * gxio_mpipe_link_open(), passing it the same link name plus the mPIPE
1738 * context, to configure the link.
1740 * @param link_name Name of the link; see @ref gxio_mpipe_link_names.
1741 * @return The mPIPE instance number which is associated with the named
1742 * link, or a negative error code (::GXIO_ERR_NO_DEVICE) if the link does
1745 extern int gxio_mpipe_link_instance(const char *link_name
);
1747 /* Retrieve one of this system's legal link names, and its MAC address.
1749 * @param index Link name index. If a system supports N legal link names,
1750 * then indices between 0 and N - 1, inclusive, each correspond to one of
1751 * those names. Thus, to retrieve all of a system's legal link names,
1752 * call this function in a loop, starting with an index of zero, and
1753 * incrementing it once per iteration until -1 is returned.
1754 * @param link_name Pointer to the buffer which will receive the retrieved
1755 * link name. The buffer should contain space for at least
1756 * ::GXIO_MPIPE_LINK_NAME_LEN bytes; the returned name, including the
1757 * terminating null byte, will be no longer than that.
1758 * @param link_name Pointer to the buffer which will receive the retrieved
1759 * MAC address. The buffer should contain space for at least 6 bytes.
1760 * @return Zero if a link name was successfully retrieved; -1 if one was
1763 extern int gxio_mpipe_link_enumerate_mac(int index
, char *link_name
,
1766 /* Open an mPIPE link.
1768 * A link must be opened before it may be used to send or receive packets,
1769 * and before its state may be examined or changed. Depending up on the
1770 * link's intended use, one or more link permissions may be requested via
1771 * the flags parameter; see @ref gxio_mpipe_link_perm. In addition, flags
1772 * may request that the link's state be modified at open time. See @ref
1773 * gxio_mpipe_link_states and @ref gxio_mpipe_link_open_flags for more detail.
1775 * @param link A link state object, which will be initialized if this
1776 * function completes successfully.
1777 * @param context An initialized mPIPE context.
1778 * @param link_name Name of the link.
1779 * @param flags Zero or more @ref gxio_mpipe_link_open_flags, ORed together.
1780 * @return 0 if the link was successfully opened, or a negative error code.
1783 extern int gxio_mpipe_link_open(gxio_mpipe_link_t
*link
,
1784 gxio_mpipe_context_t
*context
,
1785 const char *link_name
, unsigned int flags
);
1787 /* Close an mPIPE link.
1789 * Closing a link makes it available for use by other processes. Once
1790 * a link has been closed, packets may no longer be sent on or received
1791 * from the link, and its state may not be examined or changed.
1793 * @param link A link state object, which will no longer be initialized
1794 * if this function completes successfully.
1795 * @return 0 if the link was successfully closed, or a negative error code.
1798 extern int gxio_mpipe_link_close(gxio_mpipe_link_t
*link
);
1800 /* Return a link's channel number.
1802 * @param link A properly initialized link state object.
1803 * @return The channel number for the link.
1805 static inline int gxio_mpipe_link_channel(gxio_mpipe_link_t
*link
)
1807 return link
->channel
;
1810 /* Set a link attribute.
1812 * @param link A properly initialized link state object.
1813 * @param attr An attribute from the set of @ref gxio_mpipe_link_attrs.
1814 * @param val New value of the attribute.
1815 * @return 0 if the attribute was successfully set, or a negative error
1818 extern int gxio_mpipe_link_set_attr(gxio_mpipe_link_t
*link
, uint32_t attr
,
1821 ///////////////////////////////////////////////////////////////////
1823 ///////////////////////////////////////////////////////////////////
1825 /* Get the timestamp of mPIPE when this routine is called.
1827 * @param context An initialized mPIPE context.
1828 * @param ts A timespec structure to store the current clock.
1829 * @return If the call was successful, zero; otherwise, a negative error
1832 extern int gxio_mpipe_get_timestamp(gxio_mpipe_context_t
*context
,
1833 struct timespec
*ts
);
1835 /* Set the timestamp of mPIPE.
1837 * @param context An initialized mPIPE context.
1838 * @param ts A timespec structure to store the requested clock.
1839 * @return If the call was successful, zero; otherwise, a negative error
1842 extern int gxio_mpipe_set_timestamp(gxio_mpipe_context_t
*context
,
1843 const struct timespec
*ts
);
1845 /* Adjust the timestamp of mPIPE.
1847 * @param context An initialized mPIPE context.
1848 * @param delta A signed time offset to adjust, in nanoseconds.
1849 * The absolute value of this parameter must be less than or
1850 * equal to 1000000000.
1851 * @return If the call was successful, zero; otherwise, a negative error
1854 extern int gxio_mpipe_adjust_timestamp(gxio_mpipe_context_t
*context
,
1857 /** Adjust the mPIPE timestamp clock frequency.
1859 * @param context An initialized mPIPE context.
1860 * @param ppb A 32-bit signed PPB (Parts Per Billion) value to adjust.
1861 * The absolute value of ppb must be less than or equal to 1000000000.
1862 * Values less than about 30000 will generally cause a GXIO_ERR_INVAL
1863 * return due to the granularity of the hardware that converts reference
1864 * clock cycles into seconds and nanoseconds.
1865 * @return If the call was successful, zero; otherwise, a negative error
1868 extern int gxio_mpipe_adjust_timestamp_freq(gxio_mpipe_context_t
* context
,
1871 #endif /* !_GXIO_MPIPE_H_ */