2 * Copyright (c) 2005-2011 Atheros Communications Inc.
3 * Copyright (c) 2011-2013 Qualcomm Atheros, Inc.
5 * Permission to use, copy, modify, and/or distribute this software for any
6 * purpose with or without fee is hereby granted, provided that the above
7 * copyright notice and this permission notice appear in all copies.
9 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
10 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
11 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
12 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
13 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
14 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
15 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
24 * Support for Copy Engine hardware, which is mainly used for
25 * communication between Host and Target over a PCIe interconnect.
29 * A single CopyEngine (CE) comprises two "rings":
33 * Each ring consists of a number of descriptors which specify
34 * an address, length, and meta-data.
36 * Typically, one side of the PCIe interconnect (Host or Target)
37 * controls one ring and the other side controls the other ring.
38 * The source side chooses when to initiate a transfer and it
39 * chooses what to send (buffer address, length). The destination
40 * side keeps a supply of "anonymous receive buffers" available and
41 * it handles incoming data as it arrives (when the destination
42 * recieves an interrupt).
44 * The sender may send a simple buffer (address/length) or it may
45 * send a small list of buffers. When a small list is sent, hardware
46 * "gathers" these and they end up in a single destination buffer
47 * with a single interrupt.
49 * There are several "contexts" managed by this layer -- more, it
50 * may seem -- than should be needed. These are provided mainly for
51 * maximum flexibility and especially to facilitate a simpler HIF
52 * implementation. There are per-CopyEngine recv, send, and watermark
53 * contexts. These are supplied by the caller when a recv, send,
54 * or watermark handler is established and they are echoed back to
55 * the caller when the respective callbacks are invoked. There is
56 * also a per-transfer context supplied by the caller when a buffer
57 * (or sendlist) is sent and when a buffer is enqueued for recv.
58 * These per-transfer contexts are echoed back to the caller when
59 * the buffer is sent/received.
62 static inline void ath10k_ce_dest_ring_write_index_set(struct ath10k *ar,
66 ath10k_pci_write32(ar, ce_ctrl_addr + DST_WR_INDEX_ADDRESS, n);
69 static inline u32 ath10k_ce_dest_ring_write_index_get(struct ath10k *ar,
72 return ath10k_pci_read32(ar, ce_ctrl_addr + DST_WR_INDEX_ADDRESS);
75 static inline void ath10k_ce_src_ring_write_index_set(struct ath10k *ar,
79 ath10k_pci_write32(ar, ce_ctrl_addr + SR_WR_INDEX_ADDRESS, n);
82 static inline u32 ath10k_ce_src_ring_write_index_get(struct ath10k *ar,
85 return ath10k_pci_read32(ar, ce_ctrl_addr + SR_WR_INDEX_ADDRESS);
88 static inline u32 ath10k_ce_src_ring_read_index_get(struct ath10k *ar,
91 return ath10k_pci_read32(ar, ce_ctrl_addr + CURRENT_SRRI_ADDRESS);
94 static inline void ath10k_ce_src_ring_base_addr_set(struct ath10k *ar,
98 ath10k_pci_write32(ar, ce_ctrl_addr + SR_BA_ADDRESS, addr);
101 static inline void ath10k_ce_src_ring_size_set(struct ath10k *ar,
105 ath10k_pci_write32(ar, ce_ctrl_addr + SR_SIZE_ADDRESS, n);
108 static inline void ath10k_ce_src_ring_dmax_set(struct ath10k *ar,
112 u32 ctrl1_addr = ath10k_pci_read32((ar),
113 (ce_ctrl_addr) + CE_CTRL1_ADDRESS);
115 ath10k_pci_write32(ar, ce_ctrl_addr + CE_CTRL1_ADDRESS,
116 (ctrl1_addr & ~CE_CTRL1_DMAX_LENGTH_MASK) |
117 CE_CTRL1_DMAX_LENGTH_SET(n));
120 static inline void ath10k_ce_src_ring_byte_swap_set(struct ath10k *ar,
124 u32 ctrl1_addr = ath10k_pci_read32(ar, ce_ctrl_addr + CE_CTRL1_ADDRESS);
126 ath10k_pci_write32(ar, ce_ctrl_addr + CE_CTRL1_ADDRESS,
127 (ctrl1_addr & ~CE_CTRL1_SRC_RING_BYTE_SWAP_EN_MASK) |
128 CE_CTRL1_SRC_RING_BYTE_SWAP_EN_SET(n));
131 static inline void ath10k_ce_dest_ring_byte_swap_set(struct ath10k *ar,
135 u32 ctrl1_addr = ath10k_pci_read32(ar, ce_ctrl_addr + CE_CTRL1_ADDRESS);
137 ath10k_pci_write32(ar, ce_ctrl_addr + CE_CTRL1_ADDRESS,
138 (ctrl1_addr & ~CE_CTRL1_DST_RING_BYTE_SWAP_EN_MASK) |
139 CE_CTRL1_DST_RING_BYTE_SWAP_EN_SET(n));
142 static inline u32 ath10k_ce_dest_ring_read_index_get(struct ath10k *ar,
145 return ath10k_pci_read32(ar, ce_ctrl_addr + CURRENT_DRRI_ADDRESS);
148 static inline void ath10k_ce_dest_ring_base_addr_set(struct ath10k *ar,
152 ath10k_pci_write32(ar, ce_ctrl_addr + DR_BA_ADDRESS, addr);
155 static inline void ath10k_ce_dest_ring_size_set(struct ath10k *ar,
159 ath10k_pci_write32(ar, ce_ctrl_addr + DR_SIZE_ADDRESS, n);
162 static inline void ath10k_ce_src_ring_highmark_set(struct ath10k *ar,
166 u32 addr = ath10k_pci_read32(ar, ce_ctrl_addr + SRC_WATERMARK_ADDRESS);
168 ath10k_pci_write32(ar, ce_ctrl_addr + SRC_WATERMARK_ADDRESS,
169 (addr & ~SRC_WATERMARK_HIGH_MASK) |
170 SRC_WATERMARK_HIGH_SET(n));
173 static inline void ath10k_ce_src_ring_lowmark_set(struct ath10k *ar,
177 u32 addr = ath10k_pci_read32(ar, ce_ctrl_addr + SRC_WATERMARK_ADDRESS);
179 ath10k_pci_write32(ar, ce_ctrl_addr + SRC_WATERMARK_ADDRESS,
180 (addr & ~SRC_WATERMARK_LOW_MASK) |
181 SRC_WATERMARK_LOW_SET(n));
184 static inline void ath10k_ce_dest_ring_highmark_set(struct ath10k *ar,
188 u32 addr = ath10k_pci_read32(ar, ce_ctrl_addr + DST_WATERMARK_ADDRESS);
190 ath10k_pci_write32(ar, ce_ctrl_addr + DST_WATERMARK_ADDRESS,
191 (addr & ~DST_WATERMARK_HIGH_MASK) |
192 DST_WATERMARK_HIGH_SET(n));
195 static inline void ath10k_ce_dest_ring_lowmark_set(struct ath10k *ar,
199 u32 addr = ath10k_pci_read32(ar, ce_ctrl_addr + DST_WATERMARK_ADDRESS);
201 ath10k_pci_write32(ar, ce_ctrl_addr + DST_WATERMARK_ADDRESS,
202 (addr & ~DST_WATERMARK_LOW_MASK) |
203 DST_WATERMARK_LOW_SET(n));
206 static inline void ath10k_ce_copy_complete_inter_enable(struct ath10k *ar,
209 u32 host_ie_addr = ath10k_pci_read32(ar,
210 ce_ctrl_addr + HOST_IE_ADDRESS);
212 ath10k_pci_write32(ar, ce_ctrl_addr + HOST_IE_ADDRESS,
213 host_ie_addr | HOST_IE_COPY_COMPLETE_MASK);
216 static inline void ath10k_ce_copy_complete_intr_disable(struct ath10k *ar,
219 u32 host_ie_addr = ath10k_pci_read32(ar,
220 ce_ctrl_addr + HOST_IE_ADDRESS);
222 ath10k_pci_write32(ar, ce_ctrl_addr + HOST_IE_ADDRESS,
223 host_ie_addr & ~HOST_IE_COPY_COMPLETE_MASK);
226 static inline void ath10k_ce_watermark_intr_disable(struct ath10k *ar,
229 u32 host_ie_addr = ath10k_pci_read32(ar,
230 ce_ctrl_addr + HOST_IE_ADDRESS);
232 ath10k_pci_write32(ar, ce_ctrl_addr + HOST_IE_ADDRESS,
233 host_ie_addr & ~CE_WATERMARK_MASK);
236 static inline void ath10k_ce_error_intr_enable(struct ath10k *ar,
239 u32 misc_ie_addr = ath10k_pci_read32(ar,
240 ce_ctrl_addr + MISC_IE_ADDRESS);
242 ath10k_pci_write32(ar, ce_ctrl_addr + MISC_IE_ADDRESS,
243 misc_ie_addr | CE_ERROR_MASK);
246 static inline void ath10k_ce_error_intr_disable(struct ath10k *ar,
249 u32 misc_ie_addr = ath10k_pci_read32(ar,
250 ce_ctrl_addr + MISC_IE_ADDRESS);
252 ath10k_pci_write32(ar, ce_ctrl_addr + MISC_IE_ADDRESS,
253 misc_ie_addr & ~CE_ERROR_MASK);
256 static inline void ath10k_ce_engine_int_status_clear(struct ath10k *ar,
260 ath10k_pci_write32(ar, ce_ctrl_addr + HOST_IS_ADDRESS, mask);
265 * Guts of ath10k_ce_send, used by both ath10k_ce_send and
266 * ath10k_ce_sendlist_send.
267 * The caller takes responsibility for any needed locking.
269 int ath10k_ce_send_nolock(struct ath10k_ce_pipe *ce_state,
270 void *per_transfer_context,
273 unsigned int transfer_id,
276 struct ath10k *ar = ce_state->ar;
277 struct ath10k_ce_ring *src_ring = ce_state->src_ring;
278 struct ce_desc *desc, *sdesc;
279 unsigned int nentries_mask = src_ring->nentries_mask;
280 unsigned int sw_index = src_ring->sw_index;
281 unsigned int write_index = src_ring->write_index;
282 u32 ctrl_addr = ce_state->ctrl_addr;
286 if (nbytes > ce_state->src_sz_max)
287 ath10k_warn("%s: send more we can (nbytes: %d, max: %d)\n",
288 __func__, nbytes, ce_state->src_sz_max);
290 if (unlikely(CE_RING_DELTA(nentries_mask,
291 write_index, sw_index - 1) <= 0)) {
296 desc = CE_SRC_RING_TO_DESC(src_ring->base_addr_owner_space,
298 sdesc = CE_SRC_RING_TO_DESC(src_ring->shadow_base, write_index);
300 desc_flags |= SM(transfer_id, CE_DESC_FLAGS_META_DATA);
302 if (flags & CE_SEND_FLAG_GATHER)
303 desc_flags |= CE_DESC_FLAGS_GATHER;
304 if (flags & CE_SEND_FLAG_BYTE_SWAP)
305 desc_flags |= CE_DESC_FLAGS_BYTE_SWAP;
307 sdesc->addr = __cpu_to_le32(buffer);
308 sdesc->nbytes = __cpu_to_le16(nbytes);
309 sdesc->flags = __cpu_to_le16(desc_flags);
313 src_ring->per_transfer_context[write_index] = per_transfer_context;
315 /* Update Source Ring Write Index */
316 write_index = CE_RING_IDX_INCR(nentries_mask, write_index);
319 if (!(flags & CE_SEND_FLAG_GATHER))
320 ath10k_ce_src_ring_write_index_set(ar, ctrl_addr, write_index);
322 src_ring->write_index = write_index;
327 void __ath10k_ce_send_revert(struct ath10k_ce_pipe *pipe)
329 struct ath10k *ar = pipe->ar;
330 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
331 struct ath10k_ce_ring *src_ring = pipe->src_ring;
332 u32 ctrl_addr = pipe->ctrl_addr;
334 lockdep_assert_held(&ar_pci->ce_lock);
337 * This function must be called only if there is an incomplete
338 * scatter-gather transfer (before index register is updated)
339 * that needs to be cleaned up.
341 if (WARN_ON_ONCE(src_ring->write_index == src_ring->sw_index))
344 if (WARN_ON_ONCE(src_ring->write_index ==
345 ath10k_ce_src_ring_write_index_get(ar, ctrl_addr)))
348 src_ring->write_index--;
349 src_ring->write_index &= src_ring->nentries_mask;
351 src_ring->per_transfer_context[src_ring->write_index] = NULL;
354 int ath10k_ce_send(struct ath10k_ce_pipe *ce_state,
355 void *per_transfer_context,
358 unsigned int transfer_id,
361 struct ath10k *ar = ce_state->ar;
362 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
365 spin_lock_bh(&ar_pci->ce_lock);
366 ret = ath10k_ce_send_nolock(ce_state, per_transfer_context,
367 buffer, nbytes, transfer_id, flags);
368 spin_unlock_bh(&ar_pci->ce_lock);
373 int ath10k_ce_num_free_src_entries(struct ath10k_ce_pipe *pipe)
375 struct ath10k *ar = pipe->ar;
376 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
379 spin_lock_bh(&ar_pci->ce_lock);
380 delta = CE_RING_DELTA(pipe->src_ring->nentries_mask,
381 pipe->src_ring->write_index,
382 pipe->src_ring->sw_index - 1);
383 spin_unlock_bh(&ar_pci->ce_lock);
388 int ath10k_ce_recv_buf_enqueue(struct ath10k_ce_pipe *ce_state,
389 void *per_recv_context,
392 struct ath10k_ce_ring *dest_ring = ce_state->dest_ring;
393 u32 ctrl_addr = ce_state->ctrl_addr;
394 struct ath10k *ar = ce_state->ar;
395 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
396 unsigned int nentries_mask = dest_ring->nentries_mask;
397 unsigned int write_index;
398 unsigned int sw_index;
401 spin_lock_bh(&ar_pci->ce_lock);
402 write_index = dest_ring->write_index;
403 sw_index = dest_ring->sw_index;
405 if (CE_RING_DELTA(nentries_mask, write_index, sw_index - 1) > 0) {
406 struct ce_desc *base = dest_ring->base_addr_owner_space;
407 struct ce_desc *desc = CE_DEST_RING_TO_DESC(base, write_index);
409 /* Update destination descriptor */
410 desc->addr = __cpu_to_le32(buffer);
413 dest_ring->per_transfer_context[write_index] =
416 /* Update Destination Ring Write Index */
417 write_index = CE_RING_IDX_INCR(nentries_mask, write_index);
418 ath10k_ce_dest_ring_write_index_set(ar, ctrl_addr, write_index);
419 dest_ring->write_index = write_index;
425 spin_unlock_bh(&ar_pci->ce_lock);
430 * Guts of ath10k_ce_completed_recv_next.
431 * The caller takes responsibility for any necessary locking.
433 static int ath10k_ce_completed_recv_next_nolock(struct ath10k_ce_pipe *ce_state,
434 void **per_transfer_contextp,
436 unsigned int *nbytesp,
437 unsigned int *transfer_idp,
438 unsigned int *flagsp)
440 struct ath10k_ce_ring *dest_ring = ce_state->dest_ring;
441 unsigned int nentries_mask = dest_ring->nentries_mask;
442 unsigned int sw_index = dest_ring->sw_index;
444 struct ce_desc *base = dest_ring->base_addr_owner_space;
445 struct ce_desc *desc = CE_DEST_RING_TO_DESC(base, sw_index);
446 struct ce_desc sdesc;
449 /* Copy in one go for performance reasons */
452 nbytes = __le16_to_cpu(sdesc.nbytes);
455 * This closes a relatively unusual race where the Host
456 * sees the updated DRRI before the update to the
457 * corresponding descriptor has completed. We treat this
458 * as a descriptor that is not yet done.
465 /* Return data from completed destination descriptor */
466 *bufferp = __le32_to_cpu(sdesc.addr);
468 *transfer_idp = MS(__le16_to_cpu(sdesc.flags), CE_DESC_FLAGS_META_DATA);
470 if (__le16_to_cpu(sdesc.flags) & CE_DESC_FLAGS_BYTE_SWAP)
471 *flagsp = CE_RECV_FLAG_SWAPPED;
475 if (per_transfer_contextp)
476 *per_transfer_contextp =
477 dest_ring->per_transfer_context[sw_index];
480 dest_ring->per_transfer_context[sw_index] = NULL;
482 /* Update sw_index */
483 sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
484 dest_ring->sw_index = sw_index;
489 int ath10k_ce_completed_recv_next(struct ath10k_ce_pipe *ce_state,
490 void **per_transfer_contextp,
492 unsigned int *nbytesp,
493 unsigned int *transfer_idp,
494 unsigned int *flagsp)
496 struct ath10k *ar = ce_state->ar;
497 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
500 spin_lock_bh(&ar_pci->ce_lock);
501 ret = ath10k_ce_completed_recv_next_nolock(ce_state,
502 per_transfer_contextp,
504 transfer_idp, flagsp);
505 spin_unlock_bh(&ar_pci->ce_lock);
510 int ath10k_ce_revoke_recv_next(struct ath10k_ce_pipe *ce_state,
511 void **per_transfer_contextp,
514 struct ath10k_ce_ring *dest_ring;
515 unsigned int nentries_mask;
516 unsigned int sw_index;
517 unsigned int write_index;
520 struct ath10k_pci *ar_pci;
522 dest_ring = ce_state->dest_ring;
528 ar_pci = ath10k_pci_priv(ar);
530 spin_lock_bh(&ar_pci->ce_lock);
532 nentries_mask = dest_ring->nentries_mask;
533 sw_index = dest_ring->sw_index;
534 write_index = dest_ring->write_index;
535 if (write_index != sw_index) {
536 struct ce_desc *base = dest_ring->base_addr_owner_space;
537 struct ce_desc *desc = CE_DEST_RING_TO_DESC(base, sw_index);
539 /* Return data from completed destination descriptor */
540 *bufferp = __le32_to_cpu(desc->addr);
542 if (per_transfer_contextp)
543 *per_transfer_contextp =
544 dest_ring->per_transfer_context[sw_index];
547 dest_ring->per_transfer_context[sw_index] = NULL;
549 /* Update sw_index */
550 sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
551 dest_ring->sw_index = sw_index;
557 spin_unlock_bh(&ar_pci->ce_lock);
563 * Guts of ath10k_ce_completed_send_next.
564 * The caller takes responsibility for any necessary locking.
566 static int ath10k_ce_completed_send_next_nolock(struct ath10k_ce_pipe *ce_state,
567 void **per_transfer_contextp,
569 unsigned int *nbytesp,
570 unsigned int *transfer_idp)
572 struct ath10k_ce_ring *src_ring = ce_state->src_ring;
573 u32 ctrl_addr = ce_state->ctrl_addr;
574 struct ath10k *ar = ce_state->ar;
575 unsigned int nentries_mask = src_ring->nentries_mask;
576 unsigned int sw_index = src_ring->sw_index;
577 struct ce_desc *sdesc, *sbase;
578 unsigned int read_index;
580 if (src_ring->hw_index == sw_index) {
582 * The SW completion index has caught up with the cached
583 * version of the HW completion index.
584 * Update the cached HW completion index to see whether
585 * the SW has really caught up to the HW, or if the cached
586 * value of the HW index has become stale.
589 read_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
590 if (read_index == 0xffffffff)
593 read_index &= nentries_mask;
594 src_ring->hw_index = read_index;
597 read_index = src_ring->hw_index;
599 if (read_index == sw_index)
602 sbase = src_ring->shadow_base;
603 sdesc = CE_SRC_RING_TO_DESC(sbase, sw_index);
605 /* Return data from completed source descriptor */
606 *bufferp = __le32_to_cpu(sdesc->addr);
607 *nbytesp = __le16_to_cpu(sdesc->nbytes);
608 *transfer_idp = MS(__le16_to_cpu(sdesc->flags),
609 CE_DESC_FLAGS_META_DATA);
611 if (per_transfer_contextp)
612 *per_transfer_contextp =
613 src_ring->per_transfer_context[sw_index];
616 src_ring->per_transfer_context[sw_index] = NULL;
618 /* Update sw_index */
619 sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
620 src_ring->sw_index = sw_index;
625 /* NB: Modeled after ath10k_ce_completed_send_next */
626 int ath10k_ce_cancel_send_next(struct ath10k_ce_pipe *ce_state,
627 void **per_transfer_contextp,
629 unsigned int *nbytesp,
630 unsigned int *transfer_idp)
632 struct ath10k_ce_ring *src_ring;
633 unsigned int nentries_mask;
634 unsigned int sw_index;
635 unsigned int write_index;
638 struct ath10k_pci *ar_pci;
640 src_ring = ce_state->src_ring;
646 ar_pci = ath10k_pci_priv(ar);
648 spin_lock_bh(&ar_pci->ce_lock);
650 nentries_mask = src_ring->nentries_mask;
651 sw_index = src_ring->sw_index;
652 write_index = src_ring->write_index;
654 if (write_index != sw_index) {
655 struct ce_desc *base = src_ring->base_addr_owner_space;
656 struct ce_desc *desc = CE_SRC_RING_TO_DESC(base, sw_index);
658 /* Return data from completed source descriptor */
659 *bufferp = __le32_to_cpu(desc->addr);
660 *nbytesp = __le16_to_cpu(desc->nbytes);
661 *transfer_idp = MS(__le16_to_cpu(desc->flags),
662 CE_DESC_FLAGS_META_DATA);
664 if (per_transfer_contextp)
665 *per_transfer_contextp =
666 src_ring->per_transfer_context[sw_index];
669 src_ring->per_transfer_context[sw_index] = NULL;
671 /* Update sw_index */
672 sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
673 src_ring->sw_index = sw_index;
679 spin_unlock_bh(&ar_pci->ce_lock);
684 int ath10k_ce_completed_send_next(struct ath10k_ce_pipe *ce_state,
685 void **per_transfer_contextp,
687 unsigned int *nbytesp,
688 unsigned int *transfer_idp)
690 struct ath10k *ar = ce_state->ar;
691 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
694 spin_lock_bh(&ar_pci->ce_lock);
695 ret = ath10k_ce_completed_send_next_nolock(ce_state,
696 per_transfer_contextp,
699 spin_unlock_bh(&ar_pci->ce_lock);
705 * Guts of interrupt handler for per-engine interrupts on a particular CE.
707 * Invokes registered callbacks for recv_complete,
708 * send_complete, and watermarks.
710 void ath10k_ce_per_engine_service(struct ath10k *ar, unsigned int ce_id)
712 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
713 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
714 u32 ctrl_addr = ce_state->ctrl_addr;
716 spin_lock_bh(&ar_pci->ce_lock);
718 /* Clear the copy-complete interrupts that will be handled here. */
719 ath10k_ce_engine_int_status_clear(ar, ctrl_addr,
720 HOST_IS_COPY_COMPLETE_MASK);
722 spin_unlock_bh(&ar_pci->ce_lock);
724 if (ce_state->recv_cb)
725 ce_state->recv_cb(ce_state);
727 if (ce_state->send_cb)
728 ce_state->send_cb(ce_state);
730 spin_lock_bh(&ar_pci->ce_lock);
733 * Misc CE interrupts are not being handled, but still need
736 ath10k_ce_engine_int_status_clear(ar, ctrl_addr, CE_WATERMARK_MASK);
738 spin_unlock_bh(&ar_pci->ce_lock);
742 * Handler for per-engine interrupts on ALL active CEs.
743 * This is used in cases where the system is sharing a
744 * single interrput for all CEs
747 void ath10k_ce_per_engine_service_any(struct ath10k *ar)
752 intr_summary = CE_INTERRUPT_SUMMARY(ar);
754 for (ce_id = 0; intr_summary && (ce_id < CE_COUNT); ce_id++) {
755 if (intr_summary & (1 << ce_id))
756 intr_summary &= ~(1 << ce_id);
758 /* no intr pending on this CE */
761 ath10k_ce_per_engine_service(ar, ce_id);
766 * Adjust interrupts for the copy complete handler.
767 * If it's needed for either send or recv, then unmask
768 * this interrupt; otherwise, mask it.
770 * Called with ce_lock held.
772 static void ath10k_ce_per_engine_handler_adjust(struct ath10k_ce_pipe *ce_state)
774 u32 ctrl_addr = ce_state->ctrl_addr;
775 struct ath10k *ar = ce_state->ar;
776 bool disable_copy_compl_intr = ce_state->attr_flags & CE_ATTR_DIS_INTR;
778 if ((!disable_copy_compl_intr) &&
779 (ce_state->send_cb || ce_state->recv_cb))
780 ath10k_ce_copy_complete_inter_enable(ar, ctrl_addr);
782 ath10k_ce_copy_complete_intr_disable(ar, ctrl_addr);
784 ath10k_ce_watermark_intr_disable(ar, ctrl_addr);
787 int ath10k_ce_disable_interrupts(struct ath10k *ar)
791 for (ce_id = 0; ce_id < CE_COUNT; ce_id++) {
792 u32 ctrl_addr = ath10k_ce_base_address(ce_id);
794 ath10k_ce_copy_complete_intr_disable(ar, ctrl_addr);
795 ath10k_ce_error_intr_disable(ar, ctrl_addr);
796 ath10k_ce_watermark_intr_disable(ar, ctrl_addr);
802 void ath10k_ce_enable_interrupts(struct ath10k *ar)
804 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
807 for (ce_id = 0; ce_id < CE_COUNT; ce_id++)
808 ath10k_ce_per_engine_handler_adjust(&ar_pci->ce_states[ce_id]);
811 static int ath10k_ce_init_src_ring(struct ath10k *ar,
813 const struct ce_attr *attr)
815 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
816 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
817 struct ath10k_ce_ring *src_ring = ce_state->src_ring;
818 u32 nentries, ctrl_addr = ath10k_ce_base_address(ce_id);
820 nentries = roundup_pow_of_two(attr->src_nentries);
822 memset(src_ring->per_transfer_context, 0,
823 nentries * sizeof(*src_ring->per_transfer_context));
825 src_ring->sw_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
826 src_ring->sw_index &= src_ring->nentries_mask;
827 src_ring->hw_index = src_ring->sw_index;
829 src_ring->write_index =
830 ath10k_ce_src_ring_write_index_get(ar, ctrl_addr);
831 src_ring->write_index &= src_ring->nentries_mask;
833 ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr,
834 src_ring->base_addr_ce_space);
835 ath10k_ce_src_ring_size_set(ar, ctrl_addr, nentries);
836 ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, attr->src_sz_max);
837 ath10k_ce_src_ring_byte_swap_set(ar, ctrl_addr, 0);
838 ath10k_ce_src_ring_lowmark_set(ar, ctrl_addr, 0);
839 ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, nentries);
841 ath10k_dbg(ATH10K_DBG_BOOT,
842 "boot init ce src ring id %d entries %d base_addr %p\n",
843 ce_id, nentries, src_ring->base_addr_owner_space);
848 static int ath10k_ce_init_dest_ring(struct ath10k *ar,
850 const struct ce_attr *attr)
852 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
853 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
854 struct ath10k_ce_ring *dest_ring = ce_state->dest_ring;
855 u32 nentries, ctrl_addr = ath10k_ce_base_address(ce_id);
857 nentries = roundup_pow_of_two(attr->dest_nentries);
859 memset(dest_ring->per_transfer_context, 0,
860 nentries * sizeof(*dest_ring->per_transfer_context));
862 dest_ring->sw_index = ath10k_ce_dest_ring_read_index_get(ar, ctrl_addr);
863 dest_ring->sw_index &= dest_ring->nentries_mask;
864 dest_ring->write_index =
865 ath10k_ce_dest_ring_write_index_get(ar, ctrl_addr);
866 dest_ring->write_index &= dest_ring->nentries_mask;
868 ath10k_ce_dest_ring_base_addr_set(ar, ctrl_addr,
869 dest_ring->base_addr_ce_space);
870 ath10k_ce_dest_ring_size_set(ar, ctrl_addr, nentries);
871 ath10k_ce_dest_ring_byte_swap_set(ar, ctrl_addr, 0);
872 ath10k_ce_dest_ring_lowmark_set(ar, ctrl_addr, 0);
873 ath10k_ce_dest_ring_highmark_set(ar, ctrl_addr, nentries);
875 ath10k_dbg(ATH10K_DBG_BOOT,
876 "boot ce dest ring id %d entries %d base_addr %p\n",
877 ce_id, nentries, dest_ring->base_addr_owner_space);
882 static struct ath10k_ce_ring *
883 ath10k_ce_alloc_src_ring(struct ath10k *ar, unsigned int ce_id,
884 const struct ce_attr *attr)
886 struct ath10k_ce_ring *src_ring;
887 u32 nentries = attr->src_nentries;
888 dma_addr_t base_addr;
890 nentries = roundup_pow_of_two(nentries);
892 src_ring = kzalloc(sizeof(*src_ring) +
894 sizeof(*src_ring->per_transfer_context)),
896 if (src_ring == NULL)
897 return ERR_PTR(-ENOMEM);
899 src_ring->nentries = nentries;
900 src_ring->nentries_mask = nentries - 1;
903 * Legacy platforms that do not support cache
904 * coherent DMA are unsupported
906 src_ring->base_addr_owner_space_unaligned =
907 dma_alloc_coherent(ar->dev,
908 (nentries * sizeof(struct ce_desc) +
910 &base_addr, GFP_KERNEL);
911 if (!src_ring->base_addr_owner_space_unaligned) {
913 return ERR_PTR(-ENOMEM);
916 src_ring->base_addr_ce_space_unaligned = base_addr;
918 src_ring->base_addr_owner_space = PTR_ALIGN(
919 src_ring->base_addr_owner_space_unaligned,
921 src_ring->base_addr_ce_space = ALIGN(
922 src_ring->base_addr_ce_space_unaligned,
926 * Also allocate a shadow src ring in regular
927 * mem to use for faster access.
929 src_ring->shadow_base_unaligned =
930 kmalloc((nentries * sizeof(struct ce_desc) +
931 CE_DESC_RING_ALIGN), GFP_KERNEL);
932 if (!src_ring->shadow_base_unaligned) {
933 dma_free_coherent(ar->dev,
934 (nentries * sizeof(struct ce_desc) +
936 src_ring->base_addr_owner_space,
937 src_ring->base_addr_ce_space);
939 return ERR_PTR(-ENOMEM);
942 src_ring->shadow_base = PTR_ALIGN(
943 src_ring->shadow_base_unaligned,
949 static struct ath10k_ce_ring *
950 ath10k_ce_alloc_dest_ring(struct ath10k *ar, unsigned int ce_id,
951 const struct ce_attr *attr)
953 struct ath10k_ce_ring *dest_ring;
955 dma_addr_t base_addr;
957 nentries = roundup_pow_of_two(attr->dest_nentries);
959 dest_ring = kzalloc(sizeof(*dest_ring) +
961 sizeof(*dest_ring->per_transfer_context)),
963 if (dest_ring == NULL)
964 return ERR_PTR(-ENOMEM);
966 dest_ring->nentries = nentries;
967 dest_ring->nentries_mask = nentries - 1;
970 * Legacy platforms that do not support cache
971 * coherent DMA are unsupported
973 dest_ring->base_addr_owner_space_unaligned =
974 dma_alloc_coherent(ar->dev,
975 (nentries * sizeof(struct ce_desc) +
977 &base_addr, GFP_KERNEL);
978 if (!dest_ring->base_addr_owner_space_unaligned) {
980 return ERR_PTR(-ENOMEM);
983 dest_ring->base_addr_ce_space_unaligned = base_addr;
986 * Correctly initialize memory to 0 to prevent garbage
987 * data crashing system when download firmware
989 memset(dest_ring->base_addr_owner_space_unaligned, 0,
990 nentries * sizeof(struct ce_desc) + CE_DESC_RING_ALIGN);
992 dest_ring->base_addr_owner_space = PTR_ALIGN(
993 dest_ring->base_addr_owner_space_unaligned,
995 dest_ring->base_addr_ce_space = ALIGN(
996 dest_ring->base_addr_ce_space_unaligned,
1003 * Initialize a Copy Engine based on caller-supplied attributes.
1004 * This may be called once to initialize both source and destination
1005 * rings or it may be called twice for separate source and destination
1006 * initialization. It may be that only one side or the other is
1007 * initialized by software/firmware.
1009 int ath10k_ce_init_pipe(struct ath10k *ar, unsigned int ce_id,
1010 const struct ce_attr *attr,
1011 void (*send_cb)(struct ath10k_ce_pipe *),
1012 void (*recv_cb)(struct ath10k_ce_pipe *))
1014 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
1015 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
1019 * Make sure there's enough CE ringbuffer entries for HTT TX to avoid
1020 * additional TX locking checks.
1022 * For the lack of a better place do the check here.
1024 BUILD_BUG_ON(2*TARGET_NUM_MSDU_DESC >
1025 (CE_HTT_H2T_MSG_SRC_NENTRIES - 1));
1026 BUILD_BUG_ON(2*TARGET_10X_NUM_MSDU_DESC >
1027 (CE_HTT_H2T_MSG_SRC_NENTRIES - 1));
1029 spin_lock_bh(&ar_pci->ce_lock);
1031 ce_state->id = ce_id;
1032 ce_state->ctrl_addr = ath10k_ce_base_address(ce_id);
1033 ce_state->attr_flags = attr->flags;
1034 ce_state->src_sz_max = attr->src_sz_max;
1035 if (attr->src_nentries)
1036 ce_state->send_cb = send_cb;
1037 if (attr->dest_nentries)
1038 ce_state->recv_cb = recv_cb;
1039 spin_unlock_bh(&ar_pci->ce_lock);
1041 if (attr->src_nentries) {
1042 ret = ath10k_ce_init_src_ring(ar, ce_id, attr);
1044 ath10k_err("Failed to initialize CE src ring for ID: %d (%d)\n",
1050 if (attr->dest_nentries) {
1051 ret = ath10k_ce_init_dest_ring(ar, ce_id, attr);
1053 ath10k_err("Failed to initialize CE dest ring for ID: %d (%d)\n",
1062 static void ath10k_ce_deinit_src_ring(struct ath10k *ar, unsigned int ce_id)
1064 u32 ctrl_addr = ath10k_ce_base_address(ce_id);
1066 ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr, 0);
1067 ath10k_ce_src_ring_size_set(ar, ctrl_addr, 0);
1068 ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, 0);
1069 ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, 0);
1072 static void ath10k_ce_deinit_dest_ring(struct ath10k *ar, unsigned int ce_id)
1074 u32 ctrl_addr = ath10k_ce_base_address(ce_id);
1076 ath10k_ce_dest_ring_base_addr_set(ar, ctrl_addr, 0);
1077 ath10k_ce_dest_ring_size_set(ar, ctrl_addr, 0);
1078 ath10k_ce_dest_ring_highmark_set(ar, ctrl_addr, 0);
1081 void ath10k_ce_deinit_pipe(struct ath10k *ar, unsigned int ce_id)
1083 ath10k_ce_deinit_src_ring(ar, ce_id);
1084 ath10k_ce_deinit_dest_ring(ar, ce_id);
1087 int ath10k_ce_alloc_pipe(struct ath10k *ar, int ce_id,
1088 const struct ce_attr *attr)
1090 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
1091 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
1094 if (attr->src_nentries) {
1095 ce_state->src_ring = ath10k_ce_alloc_src_ring(ar, ce_id, attr);
1096 if (IS_ERR(ce_state->src_ring)) {
1097 ret = PTR_ERR(ce_state->src_ring);
1098 ath10k_err("failed to allocate copy engine source ring %d: %d\n",
1100 ce_state->src_ring = NULL;
1105 if (attr->dest_nentries) {
1106 ce_state->dest_ring = ath10k_ce_alloc_dest_ring(ar, ce_id,
1108 if (IS_ERR(ce_state->dest_ring)) {
1109 ret = PTR_ERR(ce_state->dest_ring);
1110 ath10k_err("failed to allocate copy engine destination ring %d: %d\n",
1112 ce_state->dest_ring = NULL;
1120 void ath10k_ce_free_pipe(struct ath10k *ar, int ce_id)
1122 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
1123 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
1125 if (ce_state->src_ring) {
1126 kfree(ce_state->src_ring->shadow_base_unaligned);
1127 dma_free_coherent(ar->dev,
1128 (ce_state->src_ring->nentries *
1129 sizeof(struct ce_desc) +
1130 CE_DESC_RING_ALIGN),
1131 ce_state->src_ring->base_addr_owner_space,
1132 ce_state->src_ring->base_addr_ce_space);
1133 kfree(ce_state->src_ring);
1136 if (ce_state->dest_ring) {
1137 dma_free_coherent(ar->dev,
1138 (ce_state->dest_ring->nentries *
1139 sizeof(struct ce_desc) +
1140 CE_DESC_RING_ALIGN),
1141 ce_state->dest_ring->base_addr_owner_space,
1142 ce_state->dest_ring->base_addr_ce_space);
1143 kfree(ce_state->dest_ring);
1146 ce_state->src_ring = NULL;
1147 ce_state->dest_ring = NULL;