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);
264 * Guts of ath10k_ce_send, used by both ath10k_ce_send and
265 * ath10k_ce_sendlist_send.
266 * The caller takes responsibility for any needed locking.
268 int ath10k_ce_send_nolock(struct ath10k_ce_pipe *ce_state,
269 void *per_transfer_context,
272 unsigned int transfer_id,
275 struct ath10k *ar = ce_state->ar;
276 struct ath10k_ce_ring *src_ring = ce_state->src_ring;
277 struct ce_desc *desc, *sdesc;
278 unsigned int nentries_mask = src_ring->nentries_mask;
279 unsigned int sw_index = src_ring->sw_index;
280 unsigned int write_index = src_ring->write_index;
281 u32 ctrl_addr = ce_state->ctrl_addr;
285 if (nbytes > ce_state->src_sz_max)
286 ath10k_warn(ar, "%s: send more we can (nbytes: %d, max: %d)\n",
287 __func__, nbytes, ce_state->src_sz_max);
289 if (unlikely(CE_RING_DELTA(nentries_mask,
290 write_index, sw_index - 1) <= 0)) {
295 desc = CE_SRC_RING_TO_DESC(src_ring->base_addr_owner_space,
297 sdesc = CE_SRC_RING_TO_DESC(src_ring->shadow_base, write_index);
299 desc_flags |= SM(transfer_id, CE_DESC_FLAGS_META_DATA);
301 if (flags & CE_SEND_FLAG_GATHER)
302 desc_flags |= CE_DESC_FLAGS_GATHER;
303 if (flags & CE_SEND_FLAG_BYTE_SWAP)
304 desc_flags |= CE_DESC_FLAGS_BYTE_SWAP;
306 sdesc->addr = __cpu_to_le32(buffer);
307 sdesc->nbytes = __cpu_to_le16(nbytes);
308 sdesc->flags = __cpu_to_le16(desc_flags);
312 src_ring->per_transfer_context[write_index] = per_transfer_context;
314 /* Update Source Ring Write Index */
315 write_index = CE_RING_IDX_INCR(nentries_mask, write_index);
318 if (!(flags & CE_SEND_FLAG_GATHER))
319 ath10k_ce_src_ring_write_index_set(ar, ctrl_addr, write_index);
321 src_ring->write_index = write_index;
326 void __ath10k_ce_send_revert(struct ath10k_ce_pipe *pipe)
328 struct ath10k *ar = pipe->ar;
329 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
330 struct ath10k_ce_ring *src_ring = pipe->src_ring;
331 u32 ctrl_addr = pipe->ctrl_addr;
333 lockdep_assert_held(&ar_pci->ce_lock);
336 * This function must be called only if there is an incomplete
337 * scatter-gather transfer (before index register is updated)
338 * that needs to be cleaned up.
340 if (WARN_ON_ONCE(src_ring->write_index == src_ring->sw_index))
343 if (WARN_ON_ONCE(src_ring->write_index ==
344 ath10k_ce_src_ring_write_index_get(ar, ctrl_addr)))
347 src_ring->write_index--;
348 src_ring->write_index &= src_ring->nentries_mask;
350 src_ring->per_transfer_context[src_ring->write_index] = NULL;
353 int ath10k_ce_send(struct ath10k_ce_pipe *ce_state,
354 void *per_transfer_context,
357 unsigned int transfer_id,
360 struct ath10k *ar = ce_state->ar;
361 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
364 spin_lock_bh(&ar_pci->ce_lock);
365 ret = ath10k_ce_send_nolock(ce_state, per_transfer_context,
366 buffer, nbytes, transfer_id, flags);
367 spin_unlock_bh(&ar_pci->ce_lock);
372 int ath10k_ce_num_free_src_entries(struct ath10k_ce_pipe *pipe)
374 struct ath10k *ar = pipe->ar;
375 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
378 spin_lock_bh(&ar_pci->ce_lock);
379 delta = CE_RING_DELTA(pipe->src_ring->nentries_mask,
380 pipe->src_ring->write_index,
381 pipe->src_ring->sw_index - 1);
382 spin_unlock_bh(&ar_pci->ce_lock);
387 int __ath10k_ce_rx_num_free_bufs(struct ath10k_ce_pipe *pipe)
389 struct ath10k *ar = pipe->ar;
390 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
391 struct ath10k_ce_ring *dest_ring = pipe->dest_ring;
392 unsigned int nentries_mask = dest_ring->nentries_mask;
393 unsigned int write_index = dest_ring->write_index;
394 unsigned int sw_index = dest_ring->sw_index;
396 lockdep_assert_held(&ar_pci->ce_lock);
398 return CE_RING_DELTA(nentries_mask, write_index, sw_index - 1);
401 int __ath10k_ce_rx_post_buf(struct ath10k_ce_pipe *pipe, void *ctx, u32 paddr)
403 struct ath10k *ar = pipe->ar;
404 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
405 struct ath10k_ce_ring *dest_ring = pipe->dest_ring;
406 unsigned int nentries_mask = dest_ring->nentries_mask;
407 unsigned int write_index = dest_ring->write_index;
408 unsigned int sw_index = dest_ring->sw_index;
409 struct ce_desc *base = dest_ring->base_addr_owner_space;
410 struct ce_desc *desc = CE_DEST_RING_TO_DESC(base, write_index);
411 u32 ctrl_addr = pipe->ctrl_addr;
413 lockdep_assert_held(&ar_pci->ce_lock);
415 if (CE_RING_DELTA(nentries_mask, write_index, sw_index - 1) == 0)
418 desc->addr = __cpu_to_le32(paddr);
421 dest_ring->per_transfer_context[write_index] = ctx;
422 write_index = CE_RING_IDX_INCR(nentries_mask, write_index);
423 ath10k_ce_dest_ring_write_index_set(ar, ctrl_addr, write_index);
424 dest_ring->write_index = write_index;
429 int ath10k_ce_rx_post_buf(struct ath10k_ce_pipe *pipe, void *ctx, u32 paddr)
431 struct ath10k *ar = pipe->ar;
432 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
435 spin_lock_bh(&ar_pci->ce_lock);
436 ret = __ath10k_ce_rx_post_buf(pipe, ctx, paddr);
437 spin_unlock_bh(&ar_pci->ce_lock);
443 * Guts of ath10k_ce_completed_recv_next.
444 * The caller takes responsibility for any necessary locking.
446 int ath10k_ce_completed_recv_next_nolock(struct ath10k_ce_pipe *ce_state,
447 void **per_transfer_contextp,
449 unsigned int *nbytesp,
450 unsigned int *transfer_idp,
451 unsigned int *flagsp)
453 struct ath10k_ce_ring *dest_ring = ce_state->dest_ring;
454 unsigned int nentries_mask = dest_ring->nentries_mask;
455 unsigned int sw_index = dest_ring->sw_index;
457 struct ce_desc *base = dest_ring->base_addr_owner_space;
458 struct ce_desc *desc = CE_DEST_RING_TO_DESC(base, sw_index);
459 struct ce_desc sdesc;
462 /* Copy in one go for performance reasons */
465 nbytes = __le16_to_cpu(sdesc.nbytes);
468 * This closes a relatively unusual race where the Host
469 * sees the updated DRRI before the update to the
470 * corresponding descriptor has completed. We treat this
471 * as a descriptor that is not yet done.
478 /* Return data from completed destination descriptor */
479 *bufferp = __le32_to_cpu(sdesc.addr);
481 *transfer_idp = MS(__le16_to_cpu(sdesc.flags), CE_DESC_FLAGS_META_DATA);
483 if (__le16_to_cpu(sdesc.flags) & CE_DESC_FLAGS_BYTE_SWAP)
484 *flagsp = CE_RECV_FLAG_SWAPPED;
488 if (per_transfer_contextp)
489 *per_transfer_contextp =
490 dest_ring->per_transfer_context[sw_index];
493 dest_ring->per_transfer_context[sw_index] = NULL;
495 /* Update sw_index */
496 sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
497 dest_ring->sw_index = sw_index;
502 int ath10k_ce_completed_recv_next(struct ath10k_ce_pipe *ce_state,
503 void **per_transfer_contextp,
505 unsigned int *nbytesp,
506 unsigned int *transfer_idp,
507 unsigned int *flagsp)
509 struct ath10k *ar = ce_state->ar;
510 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
513 spin_lock_bh(&ar_pci->ce_lock);
514 ret = ath10k_ce_completed_recv_next_nolock(ce_state,
515 per_transfer_contextp,
517 transfer_idp, flagsp);
518 spin_unlock_bh(&ar_pci->ce_lock);
523 int ath10k_ce_revoke_recv_next(struct ath10k_ce_pipe *ce_state,
524 void **per_transfer_contextp,
527 struct ath10k_ce_ring *dest_ring;
528 unsigned int nentries_mask;
529 unsigned int sw_index;
530 unsigned int write_index;
533 struct ath10k_pci *ar_pci;
535 dest_ring = ce_state->dest_ring;
541 ar_pci = ath10k_pci_priv(ar);
543 spin_lock_bh(&ar_pci->ce_lock);
545 nentries_mask = dest_ring->nentries_mask;
546 sw_index = dest_ring->sw_index;
547 write_index = dest_ring->write_index;
548 if (write_index != sw_index) {
549 struct ce_desc *base = dest_ring->base_addr_owner_space;
550 struct ce_desc *desc = CE_DEST_RING_TO_DESC(base, sw_index);
552 /* Return data from completed destination descriptor */
553 *bufferp = __le32_to_cpu(desc->addr);
555 if (per_transfer_contextp)
556 *per_transfer_contextp =
557 dest_ring->per_transfer_context[sw_index];
560 dest_ring->per_transfer_context[sw_index] = NULL;
562 /* Update sw_index */
563 sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
564 dest_ring->sw_index = sw_index;
570 spin_unlock_bh(&ar_pci->ce_lock);
576 * Guts of ath10k_ce_completed_send_next.
577 * The caller takes responsibility for any necessary locking.
579 int ath10k_ce_completed_send_next_nolock(struct ath10k_ce_pipe *ce_state,
580 void **per_transfer_contextp,
582 unsigned int *nbytesp,
583 unsigned int *transfer_idp)
585 struct ath10k_ce_ring *src_ring = ce_state->src_ring;
586 u32 ctrl_addr = ce_state->ctrl_addr;
587 struct ath10k *ar = ce_state->ar;
588 unsigned int nentries_mask = src_ring->nentries_mask;
589 unsigned int sw_index = src_ring->sw_index;
590 struct ce_desc *sdesc, *sbase;
591 unsigned int read_index;
593 if (src_ring->hw_index == sw_index) {
595 * The SW completion index has caught up with the cached
596 * version of the HW completion index.
597 * Update the cached HW completion index to see whether
598 * the SW has really caught up to the HW, or if the cached
599 * value of the HW index has become stale.
602 read_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
603 if (read_index == 0xffffffff)
606 read_index &= nentries_mask;
607 src_ring->hw_index = read_index;
610 read_index = src_ring->hw_index;
612 if (read_index == sw_index)
615 sbase = src_ring->shadow_base;
616 sdesc = CE_SRC_RING_TO_DESC(sbase, sw_index);
618 /* Return data from completed source descriptor */
619 *bufferp = __le32_to_cpu(sdesc->addr);
620 *nbytesp = __le16_to_cpu(sdesc->nbytes);
621 *transfer_idp = MS(__le16_to_cpu(sdesc->flags),
622 CE_DESC_FLAGS_META_DATA);
624 if (per_transfer_contextp)
625 *per_transfer_contextp =
626 src_ring->per_transfer_context[sw_index];
629 src_ring->per_transfer_context[sw_index] = NULL;
631 /* Update sw_index */
632 sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
633 src_ring->sw_index = sw_index;
638 /* NB: Modeled after ath10k_ce_completed_send_next */
639 int ath10k_ce_cancel_send_next(struct ath10k_ce_pipe *ce_state,
640 void **per_transfer_contextp,
642 unsigned int *nbytesp,
643 unsigned int *transfer_idp)
645 struct ath10k_ce_ring *src_ring;
646 unsigned int nentries_mask;
647 unsigned int sw_index;
648 unsigned int write_index;
651 struct ath10k_pci *ar_pci;
653 src_ring = ce_state->src_ring;
659 ar_pci = ath10k_pci_priv(ar);
661 spin_lock_bh(&ar_pci->ce_lock);
663 nentries_mask = src_ring->nentries_mask;
664 sw_index = src_ring->sw_index;
665 write_index = src_ring->write_index;
667 if (write_index != sw_index) {
668 struct ce_desc *base = src_ring->base_addr_owner_space;
669 struct ce_desc *desc = CE_SRC_RING_TO_DESC(base, sw_index);
671 /* Return data from completed source descriptor */
672 *bufferp = __le32_to_cpu(desc->addr);
673 *nbytesp = __le16_to_cpu(desc->nbytes);
674 *transfer_idp = MS(__le16_to_cpu(desc->flags),
675 CE_DESC_FLAGS_META_DATA);
677 if (per_transfer_contextp)
678 *per_transfer_contextp =
679 src_ring->per_transfer_context[sw_index];
682 src_ring->per_transfer_context[sw_index] = NULL;
684 /* Update sw_index */
685 sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
686 src_ring->sw_index = sw_index;
692 spin_unlock_bh(&ar_pci->ce_lock);
697 int ath10k_ce_completed_send_next(struct ath10k_ce_pipe *ce_state,
698 void **per_transfer_contextp,
700 unsigned int *nbytesp,
701 unsigned int *transfer_idp)
703 struct ath10k *ar = ce_state->ar;
704 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
707 spin_lock_bh(&ar_pci->ce_lock);
708 ret = ath10k_ce_completed_send_next_nolock(ce_state,
709 per_transfer_contextp,
712 spin_unlock_bh(&ar_pci->ce_lock);
718 * Guts of interrupt handler for per-engine interrupts on a particular CE.
720 * Invokes registered callbacks for recv_complete,
721 * send_complete, and watermarks.
723 void ath10k_ce_per_engine_service(struct ath10k *ar, unsigned int ce_id)
725 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
726 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
727 u32 ctrl_addr = ce_state->ctrl_addr;
729 spin_lock_bh(&ar_pci->ce_lock);
731 /* Clear the copy-complete interrupts that will be handled here. */
732 ath10k_ce_engine_int_status_clear(ar, ctrl_addr,
733 HOST_IS_COPY_COMPLETE_MASK);
735 spin_unlock_bh(&ar_pci->ce_lock);
737 if (ce_state->recv_cb)
738 ce_state->recv_cb(ce_state);
740 if (ce_state->send_cb)
741 ce_state->send_cb(ce_state);
743 spin_lock_bh(&ar_pci->ce_lock);
746 * Misc CE interrupts are not being handled, but still need
749 ath10k_ce_engine_int_status_clear(ar, ctrl_addr, CE_WATERMARK_MASK);
751 spin_unlock_bh(&ar_pci->ce_lock);
755 * Handler for per-engine interrupts on ALL active CEs.
756 * This is used in cases where the system is sharing a
757 * single interrput for all CEs
760 void ath10k_ce_per_engine_service_any(struct ath10k *ar)
765 intr_summary = CE_INTERRUPT_SUMMARY(ar);
767 for (ce_id = 0; intr_summary && (ce_id < CE_COUNT); ce_id++) {
768 if (intr_summary & (1 << ce_id))
769 intr_summary &= ~(1 << ce_id);
771 /* no intr pending on this CE */
774 ath10k_ce_per_engine_service(ar, ce_id);
779 * Adjust interrupts for the copy complete handler.
780 * If it's needed for either send or recv, then unmask
781 * this interrupt; otherwise, mask it.
783 * Called with ce_lock held.
785 static void ath10k_ce_per_engine_handler_adjust(struct ath10k_ce_pipe *ce_state)
787 u32 ctrl_addr = ce_state->ctrl_addr;
788 struct ath10k *ar = ce_state->ar;
789 bool disable_copy_compl_intr = ce_state->attr_flags & CE_ATTR_DIS_INTR;
791 if ((!disable_copy_compl_intr) &&
792 (ce_state->send_cb || ce_state->recv_cb))
793 ath10k_ce_copy_complete_inter_enable(ar, ctrl_addr);
795 ath10k_ce_copy_complete_intr_disable(ar, ctrl_addr);
797 ath10k_ce_watermark_intr_disable(ar, ctrl_addr);
800 int ath10k_ce_disable_interrupts(struct ath10k *ar)
804 for (ce_id = 0; ce_id < CE_COUNT; ce_id++) {
805 u32 ctrl_addr = ath10k_ce_base_address(ce_id);
807 ath10k_ce_copy_complete_intr_disable(ar, ctrl_addr);
808 ath10k_ce_error_intr_disable(ar, ctrl_addr);
809 ath10k_ce_watermark_intr_disable(ar, ctrl_addr);
815 void ath10k_ce_enable_interrupts(struct ath10k *ar)
817 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
820 /* Skip the last copy engine, CE7 the diagnostic window, as that
821 * uses polling and isn't initialized for interrupts.
823 for (ce_id = 0; ce_id < CE_COUNT - 1; ce_id++)
824 ath10k_ce_per_engine_handler_adjust(&ar_pci->ce_states[ce_id]);
827 static int ath10k_ce_init_src_ring(struct ath10k *ar,
829 const struct ce_attr *attr)
831 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
832 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
833 struct ath10k_ce_ring *src_ring = ce_state->src_ring;
834 u32 nentries, ctrl_addr = ath10k_ce_base_address(ce_id);
836 nentries = roundup_pow_of_two(attr->src_nentries);
838 memset(src_ring->per_transfer_context, 0,
839 nentries * sizeof(*src_ring->per_transfer_context));
841 src_ring->sw_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
842 src_ring->sw_index &= src_ring->nentries_mask;
843 src_ring->hw_index = src_ring->sw_index;
845 src_ring->write_index =
846 ath10k_ce_src_ring_write_index_get(ar, ctrl_addr);
847 src_ring->write_index &= src_ring->nentries_mask;
849 ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr,
850 src_ring->base_addr_ce_space);
851 ath10k_ce_src_ring_size_set(ar, ctrl_addr, nentries);
852 ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, attr->src_sz_max);
853 ath10k_ce_src_ring_byte_swap_set(ar, ctrl_addr, 0);
854 ath10k_ce_src_ring_lowmark_set(ar, ctrl_addr, 0);
855 ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, nentries);
857 ath10k_dbg(ar, ATH10K_DBG_BOOT,
858 "boot init ce src ring id %d entries %d base_addr %p\n",
859 ce_id, nentries, src_ring->base_addr_owner_space);
864 static int ath10k_ce_init_dest_ring(struct ath10k *ar,
866 const struct ce_attr *attr)
868 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
869 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
870 struct ath10k_ce_ring *dest_ring = ce_state->dest_ring;
871 u32 nentries, ctrl_addr = ath10k_ce_base_address(ce_id);
873 nentries = roundup_pow_of_two(attr->dest_nentries);
875 memset(dest_ring->per_transfer_context, 0,
876 nentries * sizeof(*dest_ring->per_transfer_context));
878 dest_ring->sw_index = ath10k_ce_dest_ring_read_index_get(ar, ctrl_addr);
879 dest_ring->sw_index &= dest_ring->nentries_mask;
880 dest_ring->write_index =
881 ath10k_ce_dest_ring_write_index_get(ar, ctrl_addr);
882 dest_ring->write_index &= dest_ring->nentries_mask;
884 ath10k_ce_dest_ring_base_addr_set(ar, ctrl_addr,
885 dest_ring->base_addr_ce_space);
886 ath10k_ce_dest_ring_size_set(ar, ctrl_addr, nentries);
887 ath10k_ce_dest_ring_byte_swap_set(ar, ctrl_addr, 0);
888 ath10k_ce_dest_ring_lowmark_set(ar, ctrl_addr, 0);
889 ath10k_ce_dest_ring_highmark_set(ar, ctrl_addr, nentries);
891 ath10k_dbg(ar, ATH10K_DBG_BOOT,
892 "boot ce dest ring id %d entries %d base_addr %p\n",
893 ce_id, nentries, dest_ring->base_addr_owner_space);
898 static struct ath10k_ce_ring *
899 ath10k_ce_alloc_src_ring(struct ath10k *ar, unsigned int ce_id,
900 const struct ce_attr *attr)
902 struct ath10k_ce_ring *src_ring;
903 u32 nentries = attr->src_nentries;
904 dma_addr_t base_addr;
906 nentries = roundup_pow_of_two(nentries);
908 src_ring = kzalloc(sizeof(*src_ring) +
910 sizeof(*src_ring->per_transfer_context)),
912 if (src_ring == NULL)
913 return ERR_PTR(-ENOMEM);
915 src_ring->nentries = nentries;
916 src_ring->nentries_mask = nentries - 1;
919 * Legacy platforms that do not support cache
920 * coherent DMA are unsupported
922 src_ring->base_addr_owner_space_unaligned =
923 dma_alloc_coherent(ar->dev,
924 (nentries * sizeof(struct ce_desc) +
926 &base_addr, GFP_KERNEL);
927 if (!src_ring->base_addr_owner_space_unaligned) {
929 return ERR_PTR(-ENOMEM);
932 src_ring->base_addr_ce_space_unaligned = base_addr;
934 src_ring->base_addr_owner_space = PTR_ALIGN(
935 src_ring->base_addr_owner_space_unaligned,
937 src_ring->base_addr_ce_space = ALIGN(
938 src_ring->base_addr_ce_space_unaligned,
942 * Also allocate a shadow src ring in regular
943 * mem to use for faster access.
945 src_ring->shadow_base_unaligned =
946 kmalloc((nentries * sizeof(struct ce_desc) +
947 CE_DESC_RING_ALIGN), GFP_KERNEL);
948 if (!src_ring->shadow_base_unaligned) {
949 dma_free_coherent(ar->dev,
950 (nentries * sizeof(struct ce_desc) +
952 src_ring->base_addr_owner_space,
953 src_ring->base_addr_ce_space);
955 return ERR_PTR(-ENOMEM);
958 src_ring->shadow_base = PTR_ALIGN(
959 src_ring->shadow_base_unaligned,
965 static struct ath10k_ce_ring *
966 ath10k_ce_alloc_dest_ring(struct ath10k *ar, unsigned int ce_id,
967 const struct ce_attr *attr)
969 struct ath10k_ce_ring *dest_ring;
971 dma_addr_t base_addr;
973 nentries = roundup_pow_of_two(attr->dest_nentries);
975 dest_ring = kzalloc(sizeof(*dest_ring) +
977 sizeof(*dest_ring->per_transfer_context)),
979 if (dest_ring == NULL)
980 return ERR_PTR(-ENOMEM);
982 dest_ring->nentries = nentries;
983 dest_ring->nentries_mask = nentries - 1;
986 * Legacy platforms that do not support cache
987 * coherent DMA are unsupported
989 dest_ring->base_addr_owner_space_unaligned =
990 dma_alloc_coherent(ar->dev,
991 (nentries * sizeof(struct ce_desc) +
993 &base_addr, GFP_KERNEL);
994 if (!dest_ring->base_addr_owner_space_unaligned) {
996 return ERR_PTR(-ENOMEM);
999 dest_ring->base_addr_ce_space_unaligned = base_addr;
1002 * Correctly initialize memory to 0 to prevent garbage
1003 * data crashing system when download firmware
1005 memset(dest_ring->base_addr_owner_space_unaligned, 0,
1006 nentries * sizeof(struct ce_desc) + CE_DESC_RING_ALIGN);
1008 dest_ring->base_addr_owner_space = PTR_ALIGN(
1009 dest_ring->base_addr_owner_space_unaligned,
1010 CE_DESC_RING_ALIGN);
1011 dest_ring->base_addr_ce_space = ALIGN(
1012 dest_ring->base_addr_ce_space_unaligned,
1013 CE_DESC_RING_ALIGN);
1019 * Initialize a Copy Engine based on caller-supplied attributes.
1020 * This may be called once to initialize both source and destination
1021 * rings or it may be called twice for separate source and destination
1022 * initialization. It may be that only one side or the other is
1023 * initialized by software/firmware.
1025 int ath10k_ce_init_pipe(struct ath10k *ar, unsigned int ce_id,
1026 const struct ce_attr *attr)
1030 if (attr->src_nentries) {
1031 ret = ath10k_ce_init_src_ring(ar, ce_id, attr);
1033 ath10k_err(ar, "Failed to initialize CE src ring for ID: %d (%d)\n",
1039 if (attr->dest_nentries) {
1040 ret = ath10k_ce_init_dest_ring(ar, ce_id, attr);
1042 ath10k_err(ar, "Failed to initialize CE dest ring for ID: %d (%d)\n",
1051 static void ath10k_ce_deinit_src_ring(struct ath10k *ar, unsigned int ce_id)
1053 u32 ctrl_addr = ath10k_ce_base_address(ce_id);
1055 ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr, 0);
1056 ath10k_ce_src_ring_size_set(ar, ctrl_addr, 0);
1057 ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, 0);
1058 ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, 0);
1061 static void ath10k_ce_deinit_dest_ring(struct ath10k *ar, unsigned int ce_id)
1063 u32 ctrl_addr = ath10k_ce_base_address(ce_id);
1065 ath10k_ce_dest_ring_base_addr_set(ar, ctrl_addr, 0);
1066 ath10k_ce_dest_ring_size_set(ar, ctrl_addr, 0);
1067 ath10k_ce_dest_ring_highmark_set(ar, ctrl_addr, 0);
1070 void ath10k_ce_deinit_pipe(struct ath10k *ar, unsigned int ce_id)
1072 ath10k_ce_deinit_src_ring(ar, ce_id);
1073 ath10k_ce_deinit_dest_ring(ar, ce_id);
1076 int ath10k_ce_alloc_pipe(struct ath10k *ar, int ce_id,
1077 const struct ce_attr *attr,
1078 void (*send_cb)(struct ath10k_ce_pipe *),
1079 void (*recv_cb)(struct ath10k_ce_pipe *))
1081 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
1082 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
1086 * Make sure there's enough CE ringbuffer entries for HTT TX to avoid
1087 * additional TX locking checks.
1089 * For the lack of a better place do the check here.
1091 BUILD_BUG_ON(2*TARGET_NUM_MSDU_DESC >
1092 (CE_HTT_H2T_MSG_SRC_NENTRIES - 1));
1093 BUILD_BUG_ON(2*TARGET_10X_NUM_MSDU_DESC >
1094 (CE_HTT_H2T_MSG_SRC_NENTRIES - 1));
1097 ce_state->id = ce_id;
1098 ce_state->ctrl_addr = ath10k_ce_base_address(ce_id);
1099 ce_state->attr_flags = attr->flags;
1100 ce_state->src_sz_max = attr->src_sz_max;
1102 if (attr->src_nentries)
1103 ce_state->send_cb = send_cb;
1105 if (attr->dest_nentries)
1106 ce_state->recv_cb = recv_cb;
1108 if (attr->src_nentries) {
1109 ce_state->src_ring = ath10k_ce_alloc_src_ring(ar, ce_id, attr);
1110 if (IS_ERR(ce_state->src_ring)) {
1111 ret = PTR_ERR(ce_state->src_ring);
1112 ath10k_err(ar, "failed to allocate copy engine source ring %d: %d\n",
1114 ce_state->src_ring = NULL;
1119 if (attr->dest_nentries) {
1120 ce_state->dest_ring = ath10k_ce_alloc_dest_ring(ar, ce_id,
1122 if (IS_ERR(ce_state->dest_ring)) {
1123 ret = PTR_ERR(ce_state->dest_ring);
1124 ath10k_err(ar, "failed to allocate copy engine destination ring %d: %d\n",
1126 ce_state->dest_ring = NULL;
1134 void ath10k_ce_free_pipe(struct ath10k *ar, int ce_id)
1136 struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
1137 struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
1139 if (ce_state->src_ring) {
1140 kfree(ce_state->src_ring->shadow_base_unaligned);
1141 dma_free_coherent(ar->dev,
1142 (ce_state->src_ring->nentries *
1143 sizeof(struct ce_desc) +
1144 CE_DESC_RING_ALIGN),
1145 ce_state->src_ring->base_addr_owner_space,
1146 ce_state->src_ring->base_addr_ce_space);
1147 kfree(ce_state->src_ring);
1150 if (ce_state->dest_ring) {
1151 dma_free_coherent(ar->dev,
1152 (ce_state->dest_ring->nentries *
1153 sizeof(struct ce_desc) +
1154 CE_DESC_RING_ALIGN),
1155 ce_state->dest_ring->base_addr_owner_space,
1156 ce_state->dest_ring->base_addr_ce_space);
1157 kfree(ce_state->dest_ring);
1160 ce_state->src_ring = NULL;
1161 ce_state->dest_ring = NULL;