1 /******************************************************************************
2  *
3  * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
4  * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
5  * Copyright(c) 2016 - 2017 Intel Deutschland GmbH
6  * Copyright(c) 2018 Intel Corporation
7  *
8  * Portions of this file are derived from the ipw3945 project, as well
9  * as portions of the ieee80211 subsystem header files.
10  *
11  * This program is free software; you can redistribute it and/or modify it
12  * under the terms of version 2 of the GNU General Public License as
13  * published by the Free Software Foundation.
14  *
15  * This program is distributed in the hope that it will be useful, but WITHOUT
16  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
17  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
18  * more details.
19  *
20  * You should have received a copy of the GNU General Public License along with
21  * this program.
22  *
23  * The full GNU General Public License is included in this distribution in the
24  * file called LICENSE.
25  *
26  * Contact Information:
27  *  Intel Linux Wireless <linuxwifi@intel.com>
28  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
29  *
30  *****************************************************************************/
31 #include <linux/sched.h>
32 #include <linux/wait.h>
33 #include <linux/gfp.h>
34 
35 #include "iwl-prph.h"
36 #include "iwl-io.h"
37 #include "internal.h"
38 #include "iwl-op-mode.h"
39 #include "iwl-context-info-gen3.h"
40 
41 /******************************************************************************
42  *
43  * RX path functions
44  *
45  ******************************************************************************/
46 
47 /*
48  * Rx theory of operation
49  *
50  * Driver allocates a circular buffer of Receive Buffer Descriptors (RBDs),
51  * each of which point to Receive Buffers to be filled by the NIC.  These get
52  * used not only for Rx frames, but for any command response or notification
53  * from the NIC.  The driver and NIC manage the Rx buffers by means
54  * of indexes into the circular buffer.
55  *
56  * Rx Queue Indexes
57  * The host/firmware share two index registers for managing the Rx buffers.
58  *
59  * The READ index maps to the first position that the firmware may be writing
60  * to -- the driver can read up to (but not including) this position and get
61  * good data.
62  * The READ index is managed by the firmware once the card is enabled.
63  *
64  * The WRITE index maps to the last position the driver has read from -- the
65  * position preceding WRITE is the last slot the firmware can place a packet.
66  *
67  * The queue is empty (no good data) if WRITE = READ - 1, and is full if
68  * WRITE = READ.
69  *
70  * During initialization, the host sets up the READ queue position to the first
71  * INDEX position, and WRITE to the last (READ - 1 wrapped)
72  *
73  * When the firmware places a packet in a buffer, it will advance the READ index
74  * and fire the RX interrupt.  The driver can then query the READ index and
75  * process as many packets as possible, moving the WRITE index forward as it
76  * resets the Rx queue buffers with new memory.
77  *
78  * The management in the driver is as follows:
79  * + A list of pre-allocated RBDs is stored in iwl->rxq->rx_free.
80  *   When the interrupt handler is called, the request is processed.
81  *   The page is either stolen - transferred to the upper layer
82  *   or reused - added immediately to the iwl->rxq->rx_free list.
83  * + When the page is stolen - the driver updates the matching queue's used
84  *   count, detaches the RBD and transfers it to the queue used list.
85  *   When there are two used RBDs - they are transferred to the allocator empty
86  *   list. Work is then scheduled for the allocator to start allocating
87  *   eight buffers.
88  *   When there are another 6 used RBDs - they are transferred to the allocator
89  *   empty list and the driver tries to claim the pre-allocated buffers and
90  *   add them to iwl->rxq->rx_free. If it fails - it continues to claim them
91  *   until ready.
92  *   When there are 8+ buffers in the free list - either from allocation or from
93  *   8 reused unstolen pages - restock is called to update the FW and indexes.
94  * + In order to make sure the allocator always has RBDs to use for allocation
95  *   the allocator has initial pool in the size of num_queues*(8-2) - the
96  *   maximum missing RBDs per allocation request (request posted with 2
97  *    empty RBDs, there is no guarantee when the other 6 RBDs are supplied).
98  *   The queues supplies the recycle of the rest of the RBDs.
99  * + A received packet is processed and handed to the kernel network stack,
100  *   detached from the iwl->rxq.  The driver 'processed' index is updated.
101  * + If there are no allocated buffers in iwl->rxq->rx_free,
102  *   the READ INDEX is not incremented and iwl->status(RX_STALLED) is set.
103  *   If there were enough free buffers and RX_STALLED is set it is cleared.
104  *
105  *
106  * Driver sequence:
107  *
108  * iwl_rxq_alloc()            Allocates rx_free
109  * iwl_pcie_rx_replenish()    Replenishes rx_free list from rx_used, and calls
110  *                            iwl_pcie_rxq_restock.
111  *                            Used only during initialization.
112  * iwl_pcie_rxq_restock()     Moves available buffers from rx_free into Rx
113  *                            queue, updates firmware pointers, and updates
114  *                            the WRITE index.
115  * iwl_pcie_rx_allocator()     Background work for allocating pages.
116  *
117  * -- enable interrupts --
118  * ISR - iwl_rx()             Detach iwl_rx_mem_buffers from pool up to the
119  *                            READ INDEX, detaching the SKB from the pool.
120  *                            Moves the packet buffer from queue to rx_used.
121  *                            Posts and claims requests to the allocator.
122  *                            Calls iwl_pcie_rxq_restock to refill any empty
123  *                            slots.
124  *
125  * RBD life-cycle:
126  *
127  * Init:
128  * rxq.pool -> rxq.rx_used -> rxq.rx_free -> rxq.queue
129  *
130  * Regular Receive interrupt:
131  * Page Stolen:
132  * rxq.queue -> rxq.rx_used -> allocator.rbd_empty ->
133  * allocator.rbd_allocated -> rxq.rx_free -> rxq.queue
134  * Page not Stolen:
135  * rxq.queue -> rxq.rx_free -> rxq.queue
136  * ...
137  *
138  */
139 
140 /*
141  * iwl_rxq_space - Return number of free slots available in queue.
142  */
143 static int iwl_rxq_space(const struct iwl_rxq *rxq)
144 {
145 	/* Make sure rx queue size is a power of 2 */
146 	WARN_ON(rxq->queue_size & (rxq->queue_size - 1));
147 
148 	/*
149 	 * There can be up to (RX_QUEUE_SIZE - 1) free slots, to avoid ambiguity
150 	 * between empty and completely full queues.
151 	 * The following is equivalent to modulo by RX_QUEUE_SIZE and is well
152 	 * defined for negative dividends.
153 	 */
154 	return (rxq->read - rxq->write - 1) & (rxq->queue_size - 1);
155 }
156 
157 /*
158  * iwl_dma_addr2rbd_ptr - convert a DMA address to a uCode read buffer ptr
159  */
160 static inline __le32 iwl_pcie_dma_addr2rbd_ptr(dma_addr_t dma_addr)
161 {
162 	return cpu_to_le32((u32)(dma_addr >> 8));
163 }
164 
165 /*
166  * iwl_pcie_rx_stop - stops the Rx DMA
167  */
168 int iwl_pcie_rx_stop(struct iwl_trans *trans)
169 {
170 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560) {
171 		/* TODO: remove this for 22560 once fw does it */
172 		iwl_write_prph(trans, RFH_RXF_DMA_CFG_GEN3, 0);
173 		return iwl_poll_prph_bit(trans, RFH_GEN_STATUS_GEN3,
174 					 RXF_DMA_IDLE, RXF_DMA_IDLE, 1000);
175 	} else if (trans->cfg->mq_rx_supported) {
176 		iwl_write_prph(trans, RFH_RXF_DMA_CFG, 0);
177 		return iwl_poll_prph_bit(trans, RFH_GEN_STATUS,
178 					   RXF_DMA_IDLE, RXF_DMA_IDLE, 1000);
179 	} else {
180 		iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
181 		return iwl_poll_direct_bit(trans, FH_MEM_RSSR_RX_STATUS_REG,
182 					   FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE,
183 					   1000);
184 	}
185 }
186 
187 /*
188  * iwl_pcie_rxq_inc_wr_ptr - Update the write pointer for the RX queue
189  */
190 static void iwl_pcie_rxq_inc_wr_ptr(struct iwl_trans *trans,
191 				    struct iwl_rxq *rxq)
192 {
193 	u32 reg;
194 
195 	lockdep_assert_held(&rxq->lock);
196 
197 	/*
198 	 * explicitly wake up the NIC if:
199 	 * 1. shadow registers aren't enabled
200 	 * 2. there is a chance that the NIC is asleep
201 	 */
202 	if (!trans->cfg->base_params->shadow_reg_enable &&
203 	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
204 		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
205 
206 		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
207 			IWL_DEBUG_INFO(trans, "Rx queue requesting wakeup, GP1 = 0x%x\n",
208 				       reg);
209 			iwl_set_bit(trans, CSR_GP_CNTRL,
210 				    BIT(trans->cfg->csr->flag_mac_access_req));
211 			rxq->need_update = true;
212 			return;
213 		}
214 	}
215 
216 	rxq->write_actual = round_down(rxq->write, 8);
217 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560)
218 		iwl_write32(trans, HBUS_TARG_WRPTR,
219 			    (rxq->write_actual |
220 			     ((FIRST_RX_QUEUE + rxq->id) << 16)));
221 	else if (trans->cfg->mq_rx_supported)
222 		iwl_write32(trans, RFH_Q_FRBDCB_WIDX_TRG(rxq->id),
223 			    rxq->write_actual);
224 	else
225 		iwl_write32(trans, FH_RSCSR_CHNL0_WPTR, rxq->write_actual);
226 }
227 
228 static void iwl_pcie_rxq_check_wrptr(struct iwl_trans *trans)
229 {
230 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
231 	int i;
232 
233 	for (i = 0; i < trans->num_rx_queues; i++) {
234 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
235 
236 		if (!rxq->need_update)
237 			continue;
238 		spin_lock(&rxq->lock);
239 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
240 		rxq->need_update = false;
241 		spin_unlock(&rxq->lock);
242 	}
243 }
244 
245 static void iwl_pcie_restock_bd(struct iwl_trans *trans,
246 				struct iwl_rxq *rxq,
247 				struct iwl_rx_mem_buffer *rxb)
248 {
249 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560) {
250 		struct iwl_rx_transfer_desc *bd = rxq->bd;
251 
252 		bd[rxq->write].type_n_size =
253 			cpu_to_le32((IWL_RX_TD_TYPE & IWL_RX_TD_TYPE_MSK) |
254 			((IWL_RX_TD_SIZE_2K >> 8) & IWL_RX_TD_SIZE_MSK));
255 		bd[rxq->write].addr = cpu_to_le64(rxb->page_dma);
256 		bd[rxq->write].rbid = cpu_to_le16(rxb->vid);
257 	} else {
258 		__le64 *bd = rxq->bd;
259 
260 		bd[rxq->write] = cpu_to_le64(rxb->page_dma | rxb->vid);
261 	}
262 }
263 
264 /*
265  * iwl_pcie_rxmq_restock - restock implementation for multi-queue rx
266  */
267 static void iwl_pcie_rxmq_restock(struct iwl_trans *trans,
268 				  struct iwl_rxq *rxq)
269 {
270 	struct iwl_rx_mem_buffer *rxb;
271 
272 	/*
273 	 * If the device isn't enabled - no need to try to add buffers...
274 	 * This can happen when we stop the device and still have an interrupt
275 	 * pending. We stop the APM before we sync the interrupts because we
276 	 * have to (see comment there). On the other hand, since the APM is
277 	 * stopped, we cannot access the HW (in particular not prph).
278 	 * So don't try to restock if the APM has been already stopped.
279 	 */
280 	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
281 		return;
282 
283 	spin_lock(&rxq->lock);
284 	while (rxq->free_count) {
285 		/* Get next free Rx buffer, remove from free list */
286 		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
287 				       list);
288 		list_del(&rxb->list);
289 		rxb->invalid = false;
290 		/* 12 first bits are expected to be empty */
291 		WARN_ON(rxb->page_dma & DMA_BIT_MASK(12));
292 		/* Point to Rx buffer via next RBD in circular buffer */
293 		iwl_pcie_restock_bd(trans, rxq, rxb);
294 		rxq->write = (rxq->write + 1) & MQ_RX_TABLE_MASK;
295 		rxq->free_count--;
296 	}
297 	spin_unlock(&rxq->lock);
298 
299 	/*
300 	 * If we've added more space for the firmware to place data, tell it.
301 	 * Increment device's write pointer in multiples of 8.
302 	 */
303 	if (rxq->write_actual != (rxq->write & ~0x7)) {
304 		spin_lock(&rxq->lock);
305 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
306 		spin_unlock(&rxq->lock);
307 	}
308 }
309 
310 /*
311  * iwl_pcie_rxsq_restock - restock implementation for single queue rx
312  */
313 static void iwl_pcie_rxsq_restock(struct iwl_trans *trans,
314 				  struct iwl_rxq *rxq)
315 {
316 	struct iwl_rx_mem_buffer *rxb;
317 
318 	/*
319 	 * If the device isn't enabled - not need to try to add buffers...
320 	 * This can happen when we stop the device and still have an interrupt
321 	 * pending. We stop the APM before we sync the interrupts because we
322 	 * have to (see comment there). On the other hand, since the APM is
323 	 * stopped, we cannot access the HW (in particular not prph).
324 	 * So don't try to restock if the APM has been already stopped.
325 	 */
326 	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
327 		return;
328 
329 	spin_lock(&rxq->lock);
330 	while ((iwl_rxq_space(rxq) > 0) && (rxq->free_count)) {
331 		__le32 *bd = (__le32 *)rxq->bd;
332 		/* The overwritten rxb must be a used one */
333 		rxb = rxq->queue[rxq->write];
334 		BUG_ON(rxb && rxb->page);
335 
336 		/* Get next free Rx buffer, remove from free list */
337 		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
338 				       list);
339 		list_del(&rxb->list);
340 		rxb->invalid = false;
341 
342 		/* Point to Rx buffer via next RBD in circular buffer */
343 		bd[rxq->write] = iwl_pcie_dma_addr2rbd_ptr(rxb->page_dma);
344 		rxq->queue[rxq->write] = rxb;
345 		rxq->write = (rxq->write + 1) & RX_QUEUE_MASK;
346 		rxq->free_count--;
347 	}
348 	spin_unlock(&rxq->lock);
349 
350 	/* If we've added more space for the firmware to place data, tell it.
351 	 * Increment device's write pointer in multiples of 8. */
352 	if (rxq->write_actual != (rxq->write & ~0x7)) {
353 		spin_lock(&rxq->lock);
354 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
355 		spin_unlock(&rxq->lock);
356 	}
357 }
358 
359 /*
360  * iwl_pcie_rxq_restock - refill RX queue from pre-allocated pool
361  *
362  * If there are slots in the RX queue that need to be restocked,
363  * and we have free pre-allocated buffers, fill the ranks as much
364  * as we can, pulling from rx_free.
365  *
366  * This moves the 'write' index forward to catch up with 'processed', and
367  * also updates the memory address in the firmware to reference the new
368  * target buffer.
369  */
370 static
371 void iwl_pcie_rxq_restock(struct iwl_trans *trans, struct iwl_rxq *rxq)
372 {
373 	if (trans->cfg->mq_rx_supported)
374 		iwl_pcie_rxmq_restock(trans, rxq);
375 	else
376 		iwl_pcie_rxsq_restock(trans, rxq);
377 }
378 
379 /*
380  * iwl_pcie_rx_alloc_page - allocates and returns a page.
381  *
382  */
383 static struct page *iwl_pcie_rx_alloc_page(struct iwl_trans *trans,
384 					   gfp_t priority)
385 {
386 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
387 	struct page *page;
388 	gfp_t gfp_mask = priority;
389 
390 	if (trans_pcie->rx_page_order > 0)
391 		gfp_mask |= __GFP_COMP;
392 
393 	/* Alloc a new receive buffer */
394 	page = alloc_pages(gfp_mask, trans_pcie->rx_page_order);
395 	if (!page) {
396 		if (net_ratelimit())
397 			IWL_DEBUG_INFO(trans, "alloc_pages failed, order: %d\n",
398 				       trans_pcie->rx_page_order);
399 		/*
400 		 * Issue an error if we don't have enough pre-allocated
401 		  * buffers.
402 `		 */
403 		if (!(gfp_mask & __GFP_NOWARN) && net_ratelimit())
404 			IWL_CRIT(trans,
405 				 "Failed to alloc_pages\n");
406 		return NULL;
407 	}
408 	return page;
409 }
410 
411 /*
412  * iwl_pcie_rxq_alloc_rbs - allocate a page for each used RBD
413  *
414  * A used RBD is an Rx buffer that has been given to the stack. To use it again
415  * a page must be allocated and the RBD must point to the page. This function
416  * doesn't change the HW pointer but handles the list of pages that is used by
417  * iwl_pcie_rxq_restock. The latter function will update the HW to use the newly
418  * allocated buffers.
419  */
420 void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans, gfp_t priority,
421 			    struct iwl_rxq *rxq)
422 {
423 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
424 	struct iwl_rx_mem_buffer *rxb;
425 	struct page *page;
426 
427 	while (1) {
428 		spin_lock(&rxq->lock);
429 		if (list_empty(&rxq->rx_used)) {
430 			spin_unlock(&rxq->lock);
431 			return;
432 		}
433 		spin_unlock(&rxq->lock);
434 
435 		/* Alloc a new receive buffer */
436 		page = iwl_pcie_rx_alloc_page(trans, priority);
437 		if (!page)
438 			return;
439 
440 		spin_lock(&rxq->lock);
441 
442 		if (list_empty(&rxq->rx_used)) {
443 			spin_unlock(&rxq->lock);
444 			__free_pages(page, trans_pcie->rx_page_order);
445 			return;
446 		}
447 		rxb = list_first_entry(&rxq->rx_used, struct iwl_rx_mem_buffer,
448 				       list);
449 		list_del(&rxb->list);
450 		spin_unlock(&rxq->lock);
451 
452 		BUG_ON(rxb->page);
453 		rxb->page = page;
454 		/* Get physical address of the RB */
455 		rxb->page_dma =
456 			dma_map_page(trans->dev, page, 0,
457 				     PAGE_SIZE << trans_pcie->rx_page_order,
458 				     DMA_FROM_DEVICE);
459 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
460 			rxb->page = NULL;
461 			spin_lock(&rxq->lock);
462 			list_add(&rxb->list, &rxq->rx_used);
463 			spin_unlock(&rxq->lock);
464 			__free_pages(page, trans_pcie->rx_page_order);
465 			return;
466 		}
467 
468 		spin_lock(&rxq->lock);
469 
470 		list_add_tail(&rxb->list, &rxq->rx_free);
471 		rxq->free_count++;
472 
473 		spin_unlock(&rxq->lock);
474 	}
475 }
476 
477 void iwl_pcie_free_rbs_pool(struct iwl_trans *trans)
478 {
479 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
480 	int i;
481 
482 	for (i = 0; i < RX_POOL_SIZE; i++) {
483 		if (!trans_pcie->rx_pool[i].page)
484 			continue;
485 		dma_unmap_page(trans->dev, trans_pcie->rx_pool[i].page_dma,
486 			       PAGE_SIZE << trans_pcie->rx_page_order,
487 			       DMA_FROM_DEVICE);
488 		__free_pages(trans_pcie->rx_pool[i].page,
489 			     trans_pcie->rx_page_order);
490 		trans_pcie->rx_pool[i].page = NULL;
491 	}
492 }
493 
494 /*
495  * iwl_pcie_rx_allocator - Allocates pages in the background for RX queues
496  *
497  * Allocates for each received request 8 pages
498  * Called as a scheduled work item.
499  */
500 static void iwl_pcie_rx_allocator(struct iwl_trans *trans)
501 {
502 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
503 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
504 	struct list_head local_empty;
505 	int pending = atomic_xchg(&rba->req_pending, 0);
506 
507 	IWL_DEBUG_RX(trans, "Pending allocation requests = %d\n", pending);
508 
509 	/* If we were scheduled - there is at least one request */
510 	spin_lock(&rba->lock);
511 	/* swap out the rba->rbd_empty to a local list */
512 	list_replace_init(&rba->rbd_empty, &local_empty);
513 	spin_unlock(&rba->lock);
514 
515 	while (pending) {
516 		int i;
517 		LIST_HEAD(local_allocated);
518 		gfp_t gfp_mask = GFP_KERNEL;
519 
520 		/* Do not post a warning if there are only a few requests */
521 		if (pending < RX_PENDING_WATERMARK)
522 			gfp_mask |= __GFP_NOWARN;
523 
524 		for (i = 0; i < RX_CLAIM_REQ_ALLOC;) {
525 			struct iwl_rx_mem_buffer *rxb;
526 			struct page *page;
527 
528 			/* List should never be empty - each reused RBD is
529 			 * returned to the list, and initial pool covers any
530 			 * possible gap between the time the page is allocated
531 			 * to the time the RBD is added.
532 			 */
533 			BUG_ON(list_empty(&local_empty));
534 			/* Get the first rxb from the rbd list */
535 			rxb = list_first_entry(&local_empty,
536 					       struct iwl_rx_mem_buffer, list);
537 			BUG_ON(rxb->page);
538 
539 			/* Alloc a new receive buffer */
540 			page = iwl_pcie_rx_alloc_page(trans, gfp_mask);
541 			if (!page)
542 				continue;
543 			rxb->page = page;
544 
545 			/* Get physical address of the RB */
546 			rxb->page_dma = dma_map_page(trans->dev, page, 0,
547 					PAGE_SIZE << trans_pcie->rx_page_order,
548 					DMA_FROM_DEVICE);
549 			if (dma_mapping_error(trans->dev, rxb->page_dma)) {
550 				rxb->page = NULL;
551 				__free_pages(page, trans_pcie->rx_page_order);
552 				continue;
553 			}
554 
555 			/* move the allocated entry to the out list */
556 			list_move(&rxb->list, &local_allocated);
557 			i++;
558 		}
559 
560 		pending--;
561 		if (!pending) {
562 			pending = atomic_xchg(&rba->req_pending, 0);
563 			IWL_DEBUG_RX(trans,
564 				     "Pending allocation requests = %d\n",
565 				     pending);
566 		}
567 
568 		spin_lock(&rba->lock);
569 		/* add the allocated rbds to the allocator allocated list */
570 		list_splice_tail(&local_allocated, &rba->rbd_allocated);
571 		/* get more empty RBDs for current pending requests */
572 		list_splice_tail_init(&rba->rbd_empty, &local_empty);
573 		spin_unlock(&rba->lock);
574 
575 		atomic_inc(&rba->req_ready);
576 	}
577 
578 	spin_lock(&rba->lock);
579 	/* return unused rbds to the allocator empty list */
580 	list_splice_tail(&local_empty, &rba->rbd_empty);
581 	spin_unlock(&rba->lock);
582 }
583 
584 /*
585  * iwl_pcie_rx_allocator_get - returns the pre-allocated pages
586 .*
587 .* Called by queue when the queue posted allocation request and
588  * has freed 8 RBDs in order to restock itself.
589  * This function directly moves the allocated RBs to the queue's ownership
590  * and updates the relevant counters.
591  */
592 static void iwl_pcie_rx_allocator_get(struct iwl_trans *trans,
593 				      struct iwl_rxq *rxq)
594 {
595 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
596 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
597 	int i;
598 
599 	lockdep_assert_held(&rxq->lock);
600 
601 	/*
602 	 * atomic_dec_if_positive returns req_ready - 1 for any scenario.
603 	 * If req_ready is 0 atomic_dec_if_positive will return -1 and this
604 	 * function will return early, as there are no ready requests.
605 	 * atomic_dec_if_positive will perofrm the *actual* decrement only if
606 	 * req_ready > 0, i.e. - there are ready requests and the function
607 	 * hands one request to the caller.
608 	 */
609 	if (atomic_dec_if_positive(&rba->req_ready) < 0)
610 		return;
611 
612 	spin_lock(&rba->lock);
613 	for (i = 0; i < RX_CLAIM_REQ_ALLOC; i++) {
614 		/* Get next free Rx buffer, remove it from free list */
615 		struct iwl_rx_mem_buffer *rxb =
616 			list_first_entry(&rba->rbd_allocated,
617 					 struct iwl_rx_mem_buffer, list);
618 
619 		list_move(&rxb->list, &rxq->rx_free);
620 	}
621 	spin_unlock(&rba->lock);
622 
623 	rxq->used_count -= RX_CLAIM_REQ_ALLOC;
624 	rxq->free_count += RX_CLAIM_REQ_ALLOC;
625 }
626 
627 void iwl_pcie_rx_allocator_work(struct work_struct *data)
628 {
629 	struct iwl_rb_allocator *rba_p =
630 		container_of(data, struct iwl_rb_allocator, rx_alloc);
631 	struct iwl_trans_pcie *trans_pcie =
632 		container_of(rba_p, struct iwl_trans_pcie, rba);
633 
634 	iwl_pcie_rx_allocator(trans_pcie->trans);
635 }
636 
637 static int iwl_pcie_free_bd_size(struct iwl_trans *trans, bool use_rx_td)
638 {
639 	struct iwl_rx_transfer_desc *rx_td;
640 
641 	if (use_rx_td)
642 		return sizeof(*rx_td);
643 	else
644 		return trans->cfg->mq_rx_supported ? sizeof(__le64) :
645 			sizeof(__le32);
646 }
647 
648 static void iwl_pcie_free_rxq_dma(struct iwl_trans *trans,
649 				  struct iwl_rxq *rxq)
650 {
651 	struct device *dev = trans->dev;
652 	bool use_rx_td = (trans->cfg->device_family >=
653 			  IWL_DEVICE_FAMILY_22560);
654 	int free_size = iwl_pcie_free_bd_size(trans, use_rx_td);
655 
656 	if (rxq->bd)
657 		dma_free_coherent(trans->dev,
658 				  free_size * rxq->queue_size,
659 				  rxq->bd, rxq->bd_dma);
660 	rxq->bd_dma = 0;
661 	rxq->bd = NULL;
662 
663 	if (rxq->rb_stts)
664 		dma_free_coherent(trans->dev,
665 				  use_rx_td ? sizeof(__le16) :
666 				  sizeof(struct iwl_rb_status),
667 				  rxq->rb_stts, rxq->rb_stts_dma);
668 	rxq->rb_stts_dma = 0;
669 	rxq->rb_stts = NULL;
670 
671 	if (rxq->used_bd)
672 		dma_free_coherent(trans->dev,
673 				  (use_rx_td ? sizeof(*rxq->cd) :
674 				   sizeof(__le32)) * rxq->queue_size,
675 				  rxq->used_bd, rxq->used_bd_dma);
676 	rxq->used_bd_dma = 0;
677 	rxq->used_bd = NULL;
678 
679 	if (trans->cfg->device_family < IWL_DEVICE_FAMILY_22560)
680 		return;
681 
682 	if (rxq->tr_tail)
683 		dma_free_coherent(dev, sizeof(__le16),
684 				  rxq->tr_tail, rxq->tr_tail_dma);
685 	rxq->tr_tail_dma = 0;
686 	rxq->tr_tail = NULL;
687 
688 	if (rxq->cr_tail)
689 		dma_free_coherent(dev, sizeof(__le16),
690 				  rxq->cr_tail, rxq->cr_tail_dma);
691 	rxq->cr_tail_dma = 0;
692 	rxq->cr_tail = NULL;
693 }
694 
695 static int iwl_pcie_alloc_rxq_dma(struct iwl_trans *trans,
696 				  struct iwl_rxq *rxq)
697 {
698 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
699 	struct device *dev = trans->dev;
700 	int i;
701 	int free_size;
702 	bool use_rx_td = (trans->cfg->device_family >=
703 			  IWL_DEVICE_FAMILY_22560);
704 
705 	spin_lock_init(&rxq->lock);
706 	if (trans->cfg->mq_rx_supported)
707 		rxq->queue_size = MQ_RX_TABLE_SIZE;
708 	else
709 		rxq->queue_size = RX_QUEUE_SIZE;
710 
711 	free_size = iwl_pcie_free_bd_size(trans, use_rx_td);
712 
713 	/*
714 	 * Allocate the circular buffer of Read Buffer Descriptors
715 	 * (RBDs)
716 	 */
717 	rxq->bd = dma_zalloc_coherent(dev,
718 				      free_size * rxq->queue_size,
719 				      &rxq->bd_dma, GFP_KERNEL);
720 	if (!rxq->bd)
721 		goto err;
722 
723 	if (trans->cfg->mq_rx_supported) {
724 		rxq->used_bd = dma_zalloc_coherent(dev,
725 						   (use_rx_td ?
726 						   sizeof(*rxq->cd) :
727 						   sizeof(__le32)) *
728 						   rxq->queue_size,
729 						   &rxq->used_bd_dma,
730 						   GFP_KERNEL);
731 		if (!rxq->used_bd)
732 			goto err;
733 	}
734 
735 	/* Allocate the driver's pointer to receive buffer status */
736 	rxq->rb_stts = dma_zalloc_coherent(dev, use_rx_td ?
737 					   sizeof(__le16) :
738 					   sizeof(struct iwl_rb_status),
739 					   &rxq->rb_stts_dma,
740 					   GFP_KERNEL);
741 	if (!rxq->rb_stts)
742 		goto err;
743 
744 	if (!use_rx_td)
745 		return 0;
746 
747 	/* Allocate the driver's pointer to TR tail */
748 	rxq->tr_tail = dma_zalloc_coherent(dev, sizeof(__le16),
749 					   &rxq->tr_tail_dma,
750 					   GFP_KERNEL);
751 	if (!rxq->tr_tail)
752 		goto err;
753 
754 	/* Allocate the driver's pointer to CR tail */
755 	rxq->cr_tail = dma_zalloc_coherent(dev, sizeof(__le16),
756 					   &rxq->cr_tail_dma,
757 					   GFP_KERNEL);
758 	if (!rxq->cr_tail)
759 		goto err;
760 	/*
761 	 * W/A 22560 device step Z0 must be non zero bug
762 	 * TODO: remove this when stop supporting Z0
763 	 */
764 	*rxq->cr_tail = cpu_to_le16(500);
765 
766 	return 0;
767 
768 err:
769 	for (i = 0; i < trans->num_rx_queues; i++) {
770 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
771 
772 		iwl_pcie_free_rxq_dma(trans, rxq);
773 	}
774 	kfree(trans_pcie->rxq);
775 
776 	return -ENOMEM;
777 }
778 
779 static int iwl_pcie_rx_alloc(struct iwl_trans *trans)
780 {
781 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
782 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
783 	int i, ret;
784 
785 	if (WARN_ON(trans_pcie->rxq))
786 		return -EINVAL;
787 
788 	trans_pcie->rxq = kcalloc(trans->num_rx_queues, sizeof(struct iwl_rxq),
789 				  GFP_KERNEL);
790 	if (!trans_pcie->rxq)
791 		return -EINVAL;
792 
793 	spin_lock_init(&rba->lock);
794 
795 	for (i = 0; i < trans->num_rx_queues; i++) {
796 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
797 
798 		ret = iwl_pcie_alloc_rxq_dma(trans, rxq);
799 		if (ret)
800 			return ret;
801 	}
802 	return 0;
803 }
804 
805 static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
806 {
807 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
808 	u32 rb_size;
809 	unsigned long flags;
810 	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
811 
812 	switch (trans_pcie->rx_buf_size) {
813 	case IWL_AMSDU_4K:
814 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
815 		break;
816 	case IWL_AMSDU_8K:
817 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
818 		break;
819 	case IWL_AMSDU_12K:
820 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_12K;
821 		break;
822 	default:
823 		WARN_ON(1);
824 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
825 	}
826 
827 	if (!iwl_trans_grab_nic_access(trans, &flags))
828 		return;
829 
830 	/* Stop Rx DMA */
831 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
832 	/* reset and flush pointers */
833 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
834 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
835 	iwl_write32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
836 
837 	/* Reset driver's Rx queue write index */
838 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
839 
840 	/* Tell device where to find RBD circular buffer in DRAM */
841 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
842 		    (u32)(rxq->bd_dma >> 8));
843 
844 	/* Tell device where in DRAM to update its Rx status */
845 	iwl_write32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
846 		    rxq->rb_stts_dma >> 4);
847 
848 	/* Enable Rx DMA
849 	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
850 	 *      the credit mechanism in 5000 HW RX FIFO
851 	 * Direct rx interrupts to hosts
852 	 * Rx buffer size 4 or 8k or 12k
853 	 * RB timeout 0x10
854 	 * 256 RBDs
855 	 */
856 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
857 		    FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
858 		    FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
859 		    FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
860 		    rb_size |
861 		    (RX_RB_TIMEOUT << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS) |
862 		    (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));
863 
864 	iwl_trans_release_nic_access(trans, &flags);
865 
866 	/* Set interrupt coalescing timer to default (2048 usecs) */
867 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
868 
869 	/* W/A for interrupt coalescing bug in 7260 and 3160 */
870 	if (trans->cfg->host_interrupt_operation_mode)
871 		iwl_set_bit(trans, CSR_INT_COALESCING, IWL_HOST_INT_OPER_MODE);
872 }
873 
874 void iwl_pcie_enable_rx_wake(struct iwl_trans *trans, bool enable)
875 {
876 	if (trans->cfg->device_family != IWL_DEVICE_FAMILY_9000)
877 		return;
878 
879 	if (CSR_HW_REV_STEP(trans->hw_rev) != SILICON_A_STEP)
880 		return;
881 
882 	if (!trans->cfg->integrated)
883 		return;
884 
885 	/*
886 	 * Turn on the chicken-bits that cause MAC wakeup for RX-related
887 	 * values.
888 	 * This costs some power, but needed for W/A 9000 integrated A-step
889 	 * bug where shadow registers are not in the retention list and their
890 	 * value is lost when NIC powers down
891 	 */
892 	iwl_set_bit(trans, CSR_MAC_SHADOW_REG_CTRL,
893 		    CSR_MAC_SHADOW_REG_CTRL_RX_WAKE);
894 	iwl_set_bit(trans, CSR_MAC_SHADOW_REG_CTL2,
895 		    CSR_MAC_SHADOW_REG_CTL2_RX_WAKE);
896 }
897 
898 static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
899 {
900 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
901 	u32 rb_size, enabled = 0;
902 	unsigned long flags;
903 	int i;
904 
905 	switch (trans_pcie->rx_buf_size) {
906 	case IWL_AMSDU_2K:
907 		rb_size = RFH_RXF_DMA_RB_SIZE_2K;
908 		break;
909 	case IWL_AMSDU_4K:
910 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
911 		break;
912 	case IWL_AMSDU_8K:
913 		rb_size = RFH_RXF_DMA_RB_SIZE_8K;
914 		break;
915 	case IWL_AMSDU_12K:
916 		rb_size = RFH_RXF_DMA_RB_SIZE_12K;
917 		break;
918 	default:
919 		WARN_ON(1);
920 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
921 	}
922 
923 	if (!iwl_trans_grab_nic_access(trans, &flags))
924 		return;
925 
926 	/* Stop Rx DMA */
927 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG, 0);
928 	/* disable free amd used rx queue operation */
929 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, 0);
930 
931 	for (i = 0; i < trans->num_rx_queues; i++) {
932 		/* Tell device where to find RBD free table in DRAM */
933 		iwl_write_prph64_no_grab(trans,
934 					 RFH_Q_FRBDCB_BA_LSB(i),
935 					 trans_pcie->rxq[i].bd_dma);
936 		/* Tell device where to find RBD used table in DRAM */
937 		iwl_write_prph64_no_grab(trans,
938 					 RFH_Q_URBDCB_BA_LSB(i),
939 					 trans_pcie->rxq[i].used_bd_dma);
940 		/* Tell device where in DRAM to update its Rx status */
941 		iwl_write_prph64_no_grab(trans,
942 					 RFH_Q_URBD_STTS_WPTR_LSB(i),
943 					 trans_pcie->rxq[i].rb_stts_dma);
944 		/* Reset device indice tables */
945 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_WIDX(i), 0);
946 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_RIDX(i), 0);
947 		iwl_write_prph_no_grab(trans, RFH_Q_URBDCB_WIDX(i), 0);
948 
949 		enabled |= BIT(i) | BIT(i + 16);
950 	}
951 
952 	/*
953 	 * Enable Rx DMA
954 	 * Rx buffer size 4 or 8k or 12k
955 	 * Min RB size 4 or 8
956 	 * Drop frames that exceed RB size
957 	 * 512 RBDs
958 	 */
959 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG,
960 			       RFH_DMA_EN_ENABLE_VAL | rb_size |
961 			       RFH_RXF_DMA_MIN_RB_4_8 |
962 			       RFH_RXF_DMA_DROP_TOO_LARGE_MASK |
963 			       RFH_RXF_DMA_RBDCB_SIZE_512);
964 
965 	/*
966 	 * Activate DMA snooping.
967 	 * Set RX DMA chunk size to 64B for IOSF and 128B for PCIe
968 	 * Default queue is 0
969 	 */
970 	iwl_write_prph_no_grab(trans, RFH_GEN_CFG,
971 			       RFH_GEN_CFG_RFH_DMA_SNOOP |
972 			       RFH_GEN_CFG_VAL(DEFAULT_RXQ_NUM, 0) |
973 			       RFH_GEN_CFG_SERVICE_DMA_SNOOP |
974 			       RFH_GEN_CFG_VAL(RB_CHUNK_SIZE,
975 					       trans->cfg->integrated ?
976 					       RFH_GEN_CFG_RB_CHUNK_SIZE_64 :
977 					       RFH_GEN_CFG_RB_CHUNK_SIZE_128));
978 	/* Enable the relevant rx queues */
979 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, enabled);
980 
981 	iwl_trans_release_nic_access(trans, &flags);
982 
983 	/* Set interrupt coalescing timer to default (2048 usecs) */
984 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
985 
986 	iwl_pcie_enable_rx_wake(trans, true);
987 }
988 
989 void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq)
990 {
991 	lockdep_assert_held(&rxq->lock);
992 
993 	INIT_LIST_HEAD(&rxq->rx_free);
994 	INIT_LIST_HEAD(&rxq->rx_used);
995 	rxq->free_count = 0;
996 	rxq->used_count = 0;
997 }
998 
999 int iwl_pcie_dummy_napi_poll(struct napi_struct *napi, int budget)
1000 {
1001 	WARN_ON(1);
1002 	return 0;
1003 }
1004 
1005 static int _iwl_pcie_rx_init(struct iwl_trans *trans)
1006 {
1007 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1008 	struct iwl_rxq *def_rxq;
1009 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1010 	int i, err, queue_size, allocator_pool_size, num_alloc;
1011 
1012 	if (!trans_pcie->rxq) {
1013 		err = iwl_pcie_rx_alloc(trans);
1014 		if (err)
1015 			return err;
1016 	}
1017 	def_rxq = trans_pcie->rxq;
1018 
1019 	cancel_work_sync(&rba->rx_alloc);
1020 
1021 	spin_lock(&rba->lock);
1022 	atomic_set(&rba->req_pending, 0);
1023 	atomic_set(&rba->req_ready, 0);
1024 	INIT_LIST_HEAD(&rba->rbd_allocated);
1025 	INIT_LIST_HEAD(&rba->rbd_empty);
1026 	spin_unlock(&rba->lock);
1027 
1028 	/* free all first - we might be reconfigured for a different size */
1029 	iwl_pcie_free_rbs_pool(trans);
1030 
1031 	for (i = 0; i < RX_QUEUE_SIZE; i++)
1032 		def_rxq->queue[i] = NULL;
1033 
1034 	for (i = 0; i < trans->num_rx_queues; i++) {
1035 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1036 
1037 		rxq->id = i;
1038 
1039 		spin_lock(&rxq->lock);
1040 		/*
1041 		 * Set read write pointer to reflect that we have processed
1042 		 * and used all buffers, but have not restocked the Rx queue
1043 		 * with fresh buffers
1044 		 */
1045 		rxq->read = 0;
1046 		rxq->write = 0;
1047 		rxq->write_actual = 0;
1048 		memset(rxq->rb_stts, 0,
1049 		       (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560) ?
1050 		       sizeof(__le16) : sizeof(struct iwl_rb_status));
1051 
1052 		iwl_pcie_rx_init_rxb_lists(rxq);
1053 
1054 		if (!rxq->napi.poll)
1055 			netif_napi_add(&trans_pcie->napi_dev, &rxq->napi,
1056 				       iwl_pcie_dummy_napi_poll, 64);
1057 
1058 		spin_unlock(&rxq->lock);
1059 	}
1060 
1061 	/* move the pool to the default queue and allocator ownerships */
1062 	queue_size = trans->cfg->mq_rx_supported ?
1063 		     MQ_RX_NUM_RBDS : RX_QUEUE_SIZE;
1064 	allocator_pool_size = trans->num_rx_queues *
1065 		(RX_CLAIM_REQ_ALLOC - RX_POST_REQ_ALLOC);
1066 	num_alloc = queue_size + allocator_pool_size;
1067 	BUILD_BUG_ON(ARRAY_SIZE(trans_pcie->global_table) !=
1068 		     ARRAY_SIZE(trans_pcie->rx_pool));
1069 	for (i = 0; i < num_alloc; i++) {
1070 		struct iwl_rx_mem_buffer *rxb = &trans_pcie->rx_pool[i];
1071 
1072 		if (i < allocator_pool_size)
1073 			list_add(&rxb->list, &rba->rbd_empty);
1074 		else
1075 			list_add(&rxb->list, &def_rxq->rx_used);
1076 		trans_pcie->global_table[i] = rxb;
1077 		rxb->vid = (u16)(i + 1);
1078 		rxb->invalid = true;
1079 	}
1080 
1081 	iwl_pcie_rxq_alloc_rbs(trans, GFP_KERNEL, def_rxq);
1082 
1083 	return 0;
1084 }
1085 
1086 int iwl_pcie_rx_init(struct iwl_trans *trans)
1087 {
1088 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1089 	int ret = _iwl_pcie_rx_init(trans);
1090 
1091 	if (ret)
1092 		return ret;
1093 
1094 	if (trans->cfg->mq_rx_supported)
1095 		iwl_pcie_rx_mq_hw_init(trans);
1096 	else
1097 		iwl_pcie_rx_hw_init(trans, trans_pcie->rxq);
1098 
1099 	iwl_pcie_rxq_restock(trans, trans_pcie->rxq);
1100 
1101 	spin_lock(&trans_pcie->rxq->lock);
1102 	iwl_pcie_rxq_inc_wr_ptr(trans, trans_pcie->rxq);
1103 	spin_unlock(&trans_pcie->rxq->lock);
1104 
1105 	return 0;
1106 }
1107 
1108 int iwl_pcie_gen2_rx_init(struct iwl_trans *trans)
1109 {
1110 	/*
1111 	 * We don't configure the RFH.
1112 	 * Restock will be done at alive, after firmware configured the RFH.
1113 	 */
1114 	return _iwl_pcie_rx_init(trans);
1115 }
1116 
1117 void iwl_pcie_rx_free(struct iwl_trans *trans)
1118 {
1119 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1120 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1121 	int i;
1122 
1123 	/*
1124 	 * if rxq is NULL, it means that nothing has been allocated,
1125 	 * exit now
1126 	 */
1127 	if (!trans_pcie->rxq) {
1128 		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
1129 		return;
1130 	}
1131 
1132 	cancel_work_sync(&rba->rx_alloc);
1133 
1134 	iwl_pcie_free_rbs_pool(trans);
1135 
1136 	for (i = 0; i < trans->num_rx_queues; i++) {
1137 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1138 
1139 		iwl_pcie_free_rxq_dma(trans, rxq);
1140 
1141 		if (rxq->napi.poll)
1142 			netif_napi_del(&rxq->napi);
1143 	}
1144 	kfree(trans_pcie->rxq);
1145 }
1146 
1147 /*
1148  * iwl_pcie_rx_reuse_rbd - Recycle used RBDs
1149  *
1150  * Called when a RBD can be reused. The RBD is transferred to the allocator.
1151  * When there are 2 empty RBDs - a request for allocation is posted
1152  */
1153 static void iwl_pcie_rx_reuse_rbd(struct iwl_trans *trans,
1154 				  struct iwl_rx_mem_buffer *rxb,
1155 				  struct iwl_rxq *rxq, bool emergency)
1156 {
1157 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1158 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1159 
1160 	/* Move the RBD to the used list, will be moved to allocator in batches
1161 	 * before claiming or posting a request*/
1162 	list_add_tail(&rxb->list, &rxq->rx_used);
1163 
1164 	if (unlikely(emergency))
1165 		return;
1166 
1167 	/* Count the allocator owned RBDs */
1168 	rxq->used_count++;
1169 
1170 	/* If we have RX_POST_REQ_ALLOC new released rx buffers -
1171 	 * issue a request for allocator. Modulo RX_CLAIM_REQ_ALLOC is
1172 	 * used for the case we failed to claim RX_CLAIM_REQ_ALLOC,
1173 	 * after but we still need to post another request.
1174 	 */
1175 	if ((rxq->used_count % RX_CLAIM_REQ_ALLOC) == RX_POST_REQ_ALLOC) {
1176 		/* Move the 2 RBDs to the allocator ownership.
1177 		 Allocator has another 6 from pool for the request completion*/
1178 		spin_lock(&rba->lock);
1179 		list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
1180 		spin_unlock(&rba->lock);
1181 
1182 		atomic_inc(&rba->req_pending);
1183 		queue_work(rba->alloc_wq, &rba->rx_alloc);
1184 	}
1185 }
1186 
1187 static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
1188 				struct iwl_rxq *rxq,
1189 				struct iwl_rx_mem_buffer *rxb,
1190 				bool emergency)
1191 {
1192 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1193 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1194 	bool page_stolen = false;
1195 	int max_len = PAGE_SIZE << trans_pcie->rx_page_order;
1196 	u32 offset = 0;
1197 
1198 	if (WARN_ON(!rxb))
1199 		return;
1200 
1201 	dma_unmap_page(trans->dev, rxb->page_dma, max_len, DMA_FROM_DEVICE);
1202 
1203 	while (offset + sizeof(u32) + sizeof(struct iwl_cmd_header) < max_len) {
1204 		struct iwl_rx_packet *pkt;
1205 		u16 sequence;
1206 		bool reclaim;
1207 		int index, cmd_index, len;
1208 		struct iwl_rx_cmd_buffer rxcb = {
1209 			._offset = offset,
1210 			._rx_page_order = trans_pcie->rx_page_order,
1211 			._page = rxb->page,
1212 			._page_stolen = false,
1213 			.truesize = max_len,
1214 		};
1215 
1216 		pkt = rxb_addr(&rxcb);
1217 
1218 		if (pkt->len_n_flags == cpu_to_le32(FH_RSCSR_FRAME_INVALID)) {
1219 			IWL_DEBUG_RX(trans,
1220 				     "Q %d: RB end marker at offset %d\n",
1221 				     rxq->id, offset);
1222 			break;
1223 		}
1224 
1225 		WARN((le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1226 			FH_RSCSR_RXQ_POS != rxq->id,
1227 		     "frame on invalid queue - is on %d and indicates %d\n",
1228 		     rxq->id,
1229 		     (le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1230 			FH_RSCSR_RXQ_POS);
1231 
1232 		IWL_DEBUG_RX(trans,
1233 			     "Q %d: cmd at offset %d: %s (%.2x.%2x, seq 0x%x)\n",
1234 			     rxq->id, offset,
1235 			     iwl_get_cmd_string(trans,
1236 						iwl_cmd_id(pkt->hdr.cmd,
1237 							   pkt->hdr.group_id,
1238 							   0)),
1239 			     pkt->hdr.group_id, pkt->hdr.cmd,
1240 			     le16_to_cpu(pkt->hdr.sequence));
1241 
1242 		len = iwl_rx_packet_len(pkt);
1243 		len += sizeof(u32); /* account for status word */
1244 		trace_iwlwifi_dev_rx(trans->dev, trans, pkt, len);
1245 		trace_iwlwifi_dev_rx_data(trans->dev, trans, pkt, len);
1246 
1247 		/* Reclaim a command buffer only if this packet is a response
1248 		 *   to a (driver-originated) command.
1249 		 * If the packet (e.g. Rx frame) originated from uCode,
1250 		 *   there is no command buffer to reclaim.
1251 		 * Ucode should set SEQ_RX_FRAME bit if ucode-originated,
1252 		 *   but apparently a few don't get set; catch them here. */
1253 		reclaim = !(pkt->hdr.sequence & SEQ_RX_FRAME);
1254 		if (reclaim && !pkt->hdr.group_id) {
1255 			int i;
1256 
1257 			for (i = 0; i < trans_pcie->n_no_reclaim_cmds; i++) {
1258 				if (trans_pcie->no_reclaim_cmds[i] ==
1259 							pkt->hdr.cmd) {
1260 					reclaim = false;
1261 					break;
1262 				}
1263 			}
1264 		}
1265 
1266 		sequence = le16_to_cpu(pkt->hdr.sequence);
1267 		index = SEQ_TO_INDEX(sequence);
1268 		cmd_index = iwl_pcie_get_cmd_index(txq, index);
1269 
1270 		if (rxq->id == 0)
1271 			iwl_op_mode_rx(trans->op_mode, &rxq->napi,
1272 				       &rxcb);
1273 		else
1274 			iwl_op_mode_rx_rss(trans->op_mode, &rxq->napi,
1275 					   &rxcb, rxq->id);
1276 
1277 		if (reclaim) {
1278 			kzfree(txq->entries[cmd_index].free_buf);
1279 			txq->entries[cmd_index].free_buf = NULL;
1280 		}
1281 
1282 		/*
1283 		 * After here, we should always check rxcb._page_stolen,
1284 		 * if it is true then one of the handlers took the page.
1285 		 */
1286 
1287 		if (reclaim) {
1288 			/* Invoke any callbacks, transfer the buffer to caller,
1289 			 * and fire off the (possibly) blocking
1290 			 * iwl_trans_send_cmd()
1291 			 * as we reclaim the driver command queue */
1292 			if (!rxcb._page_stolen)
1293 				iwl_pcie_hcmd_complete(trans, &rxcb);
1294 			else
1295 				IWL_WARN(trans, "Claim null rxb?\n");
1296 		}
1297 
1298 		page_stolen |= rxcb._page_stolen;
1299 		if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560)
1300 			break;
1301 		offset += ALIGN(len, FH_RSCSR_FRAME_ALIGN);
1302 	}
1303 
1304 	/* page was stolen from us -- free our reference */
1305 	if (page_stolen) {
1306 		__free_pages(rxb->page, trans_pcie->rx_page_order);
1307 		rxb->page = NULL;
1308 	}
1309 
1310 	/* Reuse the page if possible. For notification packets and
1311 	 * SKBs that fail to Rx correctly, add them back into the
1312 	 * rx_free list for reuse later. */
1313 	if (rxb->page != NULL) {
1314 		rxb->page_dma =
1315 			dma_map_page(trans->dev, rxb->page, 0,
1316 				     PAGE_SIZE << trans_pcie->rx_page_order,
1317 				     DMA_FROM_DEVICE);
1318 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
1319 			/*
1320 			 * free the page(s) as well to not break
1321 			 * the invariant that the items on the used
1322 			 * list have no page(s)
1323 			 */
1324 			__free_pages(rxb->page, trans_pcie->rx_page_order);
1325 			rxb->page = NULL;
1326 			iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1327 		} else {
1328 			list_add_tail(&rxb->list, &rxq->rx_free);
1329 			rxq->free_count++;
1330 		}
1331 	} else
1332 		iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1333 }
1334 
1335 static struct iwl_rx_mem_buffer *iwl_pcie_get_rxb(struct iwl_trans *trans,
1336 						  struct iwl_rxq *rxq, int i)
1337 {
1338 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1339 	struct iwl_rx_mem_buffer *rxb;
1340 	u16 vid;
1341 
1342 	if (!trans->cfg->mq_rx_supported) {
1343 		rxb = rxq->queue[i];
1344 		rxq->queue[i] = NULL;
1345 		return rxb;
1346 	}
1347 
1348 	/* used_bd is a 32/16 bit but only 12 are used to retrieve the vid */
1349 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560)
1350 		vid = le16_to_cpu(rxq->cd[i].rbid) & 0x0FFF;
1351 	else
1352 		vid = le32_to_cpu(rxq->bd_32[i]) & 0x0FFF;
1353 
1354 	if (!vid || vid > ARRAY_SIZE(trans_pcie->global_table))
1355 		goto out_err;
1356 
1357 	rxb = trans_pcie->global_table[vid - 1];
1358 	if (rxb->invalid)
1359 		goto out_err;
1360 
1361 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560)
1362 		rxb->size = le32_to_cpu(rxq->cd[i].size) & IWL_RX_CD_SIZE;
1363 
1364 	rxb->invalid = true;
1365 
1366 	return rxb;
1367 
1368 out_err:
1369 	WARN(1, "Invalid rxb from HW %u\n", (u32)vid);
1370 	iwl_force_nmi(trans);
1371 	return NULL;
1372 }
1373 
1374 /*
1375  * iwl_pcie_rx_handle - Main entry function for receiving responses from fw
1376  */
1377 static void iwl_pcie_rx_handle(struct iwl_trans *trans, int queue)
1378 {
1379 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1380 	struct iwl_rxq *rxq = &trans_pcie->rxq[queue];
1381 	u32 r, i, count = 0;
1382 	bool emergency = false;
1383 
1384 restart:
1385 	spin_lock(&rxq->lock);
1386 	/* uCode's read index (stored in shared DRAM) indicates the last Rx
1387 	 * buffer that the driver may process (last buffer filled by ucode). */
1388 	r = le16_to_cpu(iwl_get_closed_rb_stts(trans, rxq)) & 0x0FFF;
1389 	i = rxq->read;
1390 
1391 	/* W/A 9000 device step A0 wrap-around bug */
1392 	r &= (rxq->queue_size - 1);
1393 
1394 	/* Rx interrupt, but nothing sent from uCode */
1395 	if (i == r)
1396 		IWL_DEBUG_RX(trans, "Q %d: HW = SW = %d\n", rxq->id, r);
1397 
1398 	while (i != r) {
1399 		struct iwl_rx_mem_buffer *rxb;
1400 
1401 		if (unlikely(rxq->used_count == rxq->queue_size / 2))
1402 			emergency = true;
1403 
1404 		rxb = iwl_pcie_get_rxb(trans, rxq, i);
1405 		if (!rxb)
1406 			goto out;
1407 
1408 		IWL_DEBUG_RX(trans, "Q %d: HW = %d, SW = %d\n", rxq->id, r, i);
1409 		iwl_pcie_rx_handle_rb(trans, rxq, rxb, emergency);
1410 
1411 		i = (i + 1) & (rxq->queue_size - 1);
1412 
1413 		/*
1414 		 * If we have RX_CLAIM_REQ_ALLOC released rx buffers -
1415 		 * try to claim the pre-allocated buffers from the allocator.
1416 		 * If not ready - will try to reclaim next time.
1417 		 * There is no need to reschedule work - allocator exits only
1418 		 * on success
1419 		 */
1420 		if (rxq->used_count >= RX_CLAIM_REQ_ALLOC)
1421 			iwl_pcie_rx_allocator_get(trans, rxq);
1422 
1423 		if (rxq->used_count % RX_CLAIM_REQ_ALLOC == 0 && !emergency) {
1424 			struct iwl_rb_allocator *rba = &trans_pcie->rba;
1425 
1426 			/* Add the remaining empty RBDs for allocator use */
1427 			spin_lock(&rba->lock);
1428 			list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
1429 			spin_unlock(&rba->lock);
1430 		} else if (emergency) {
1431 			count++;
1432 			if (count == 8) {
1433 				count = 0;
1434 				if (rxq->used_count < rxq->queue_size / 3)
1435 					emergency = false;
1436 
1437 				rxq->read = i;
1438 				spin_unlock(&rxq->lock);
1439 				iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1440 				iwl_pcie_rxq_restock(trans, rxq);
1441 				goto restart;
1442 			}
1443 		}
1444 	}
1445 out:
1446 	/* Backtrack one entry */
1447 	rxq->read = i;
1448 	/* update cr tail with the rxq read pointer */
1449 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560)
1450 		*rxq->cr_tail = cpu_to_le16(r);
1451 	spin_unlock(&rxq->lock);
1452 
1453 	/*
1454 	 * handle a case where in emergency there are some unallocated RBDs.
1455 	 * those RBDs are in the used list, but are not tracked by the queue's
1456 	 * used_count which counts allocator owned RBDs.
1457 	 * unallocated emergency RBDs must be allocated on exit, otherwise
1458 	 * when called again the function may not be in emergency mode and
1459 	 * they will be handed to the allocator with no tracking in the RBD
1460 	 * allocator counters, which will lead to them never being claimed back
1461 	 * by the queue.
1462 	 * by allocating them here, they are now in the queue free list, and
1463 	 * will be restocked by the next call of iwl_pcie_rxq_restock.
1464 	 */
1465 	if (unlikely(emergency && count))
1466 		iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1467 
1468 	if (rxq->napi.poll)
1469 		napi_gro_flush(&rxq->napi, false);
1470 
1471 	iwl_pcie_rxq_restock(trans, rxq);
1472 }
1473 
1474 static struct iwl_trans_pcie *iwl_pcie_get_trans_pcie(struct msix_entry *entry)
1475 {
1476 	u8 queue = entry->entry;
1477 	struct msix_entry *entries = entry - queue;
1478 
1479 	return container_of(entries, struct iwl_trans_pcie, msix_entries[0]);
1480 }
1481 
1482 /*
1483  * iwl_pcie_rx_msix_handle - Main entry function for receiving responses from fw
1484  * This interrupt handler should be used with RSS queue only.
1485  */
1486 irqreturn_t iwl_pcie_irq_rx_msix_handler(int irq, void *dev_id)
1487 {
1488 	struct msix_entry *entry = dev_id;
1489 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
1490 	struct iwl_trans *trans = trans_pcie->trans;
1491 
1492 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, false, 0, 0);
1493 
1494 	if (WARN_ON(entry->entry >= trans->num_rx_queues))
1495 		return IRQ_NONE;
1496 
1497 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1498 
1499 	local_bh_disable();
1500 	iwl_pcie_rx_handle(trans, entry->entry);
1501 	local_bh_enable();
1502 
1503 	iwl_pcie_clear_irq(trans, entry);
1504 
1505 	lock_map_release(&trans->sync_cmd_lockdep_map);
1506 
1507 	return IRQ_HANDLED;
1508 }
1509 
1510 /*
1511  * iwl_pcie_irq_handle_error - called for HW or SW error interrupt from card
1512  */
1513 static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
1514 {
1515 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1516 	int i;
1517 
1518 	/* W/A for WiFi/WiMAX coex and WiMAX own the RF */
1519 	if (trans->cfg->internal_wimax_coex &&
1520 	    !trans->cfg->apmg_not_supported &&
1521 	    (!(iwl_read_prph(trans, APMG_CLK_CTRL_REG) &
1522 			     APMS_CLK_VAL_MRB_FUNC_MODE) ||
1523 	     (iwl_read_prph(trans, APMG_PS_CTRL_REG) &
1524 			    APMG_PS_CTRL_VAL_RESET_REQ))) {
1525 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1526 		iwl_op_mode_wimax_active(trans->op_mode);
1527 		wake_up(&trans_pcie->wait_command_queue);
1528 		return;
1529 	}
1530 
1531 	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
1532 		if (!trans_pcie->txq[i])
1533 			continue;
1534 		del_timer(&trans_pcie->txq[i]->stuck_timer);
1535 	}
1536 
1537 	/* The STATUS_FW_ERROR bit is set in this function. This must happen
1538 	 * before we wake up the command caller, to ensure a proper cleanup. */
1539 	iwl_trans_fw_error(trans);
1540 
1541 	clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1542 	wake_up(&trans_pcie->wait_command_queue);
1543 }
1544 
1545 static u32 iwl_pcie_int_cause_non_ict(struct iwl_trans *trans)
1546 {
1547 	u32 inta;
1548 
1549 	lockdep_assert_held(&IWL_TRANS_GET_PCIE_TRANS(trans)->irq_lock);
1550 
1551 	trace_iwlwifi_dev_irq(trans->dev);
1552 
1553 	/* Discover which interrupts are active/pending */
1554 	inta = iwl_read32(trans, CSR_INT);
1555 
1556 	/* the thread will service interrupts and re-enable them */
1557 	return inta;
1558 }
1559 
1560 /* a device (PCI-E) page is 4096 bytes long */
1561 #define ICT_SHIFT	12
1562 #define ICT_SIZE	(1 << ICT_SHIFT)
1563 #define ICT_COUNT	(ICT_SIZE / sizeof(u32))
1564 
1565 /* interrupt handler using ict table, with this interrupt driver will
1566  * stop using INTA register to get device's interrupt, reading this register
1567  * is expensive, device will write interrupts in ICT dram table, increment
1568  * index then will fire interrupt to driver, driver will OR all ICT table
1569  * entries from current index up to table entry with 0 value. the result is
1570  * the interrupt we need to service, driver will set the entries back to 0 and
1571  * set index.
1572  */
1573 static u32 iwl_pcie_int_cause_ict(struct iwl_trans *trans)
1574 {
1575 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1576 	u32 inta;
1577 	u32 val = 0;
1578 	u32 read;
1579 
1580 	trace_iwlwifi_dev_irq(trans->dev);
1581 
1582 	/* Ignore interrupt if there's nothing in NIC to service.
1583 	 * This may be due to IRQ shared with another device,
1584 	 * or due to sporadic interrupts thrown from our NIC. */
1585 	read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1586 	trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index, read);
1587 	if (!read)
1588 		return 0;
1589 
1590 	/*
1591 	 * Collect all entries up to the first 0, starting from ict_index;
1592 	 * note we already read at ict_index.
1593 	 */
1594 	do {
1595 		val |= read;
1596 		IWL_DEBUG_ISR(trans, "ICT index %d value 0x%08X\n",
1597 				trans_pcie->ict_index, read);
1598 		trans_pcie->ict_tbl[trans_pcie->ict_index] = 0;
1599 		trans_pcie->ict_index =
1600 			((trans_pcie->ict_index + 1) & (ICT_COUNT - 1));
1601 
1602 		read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1603 		trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index,
1604 					   read);
1605 	} while (read);
1606 
1607 	/* We should not get this value, just ignore it. */
1608 	if (val == 0xffffffff)
1609 		val = 0;
1610 
1611 	/*
1612 	 * this is a w/a for a h/w bug. the h/w bug may cause the Rx bit
1613 	 * (bit 15 before shifting it to 31) to clear when using interrupt
1614 	 * coalescing. fortunately, bits 18 and 19 stay set when this happens
1615 	 * so we use them to decide on the real state of the Rx bit.
1616 	 * In order words, bit 15 is set if bit 18 or bit 19 are set.
1617 	 */
1618 	if (val & 0xC0000)
1619 		val |= 0x8000;
1620 
1621 	inta = (0xff & val) | ((0xff00 & val) << 16);
1622 	return inta;
1623 }
1624 
1625 void iwl_pcie_handle_rfkill_irq(struct iwl_trans *trans)
1626 {
1627 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1628 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1629 	bool hw_rfkill, prev, report;
1630 
1631 	mutex_lock(&trans_pcie->mutex);
1632 	prev = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1633 	hw_rfkill = iwl_is_rfkill_set(trans);
1634 	if (hw_rfkill) {
1635 		set_bit(STATUS_RFKILL_OPMODE, &trans->status);
1636 		set_bit(STATUS_RFKILL_HW, &trans->status);
1637 	}
1638 	if (trans_pcie->opmode_down)
1639 		report = hw_rfkill;
1640 	else
1641 		report = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1642 
1643 	IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
1644 		 hw_rfkill ? "disable radio" : "enable radio");
1645 
1646 	isr_stats->rfkill++;
1647 
1648 	if (prev != report)
1649 		iwl_trans_pcie_rf_kill(trans, report);
1650 	mutex_unlock(&trans_pcie->mutex);
1651 
1652 	if (hw_rfkill) {
1653 		if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
1654 				       &trans->status))
1655 			IWL_DEBUG_RF_KILL(trans,
1656 					  "Rfkill while SYNC HCMD in flight\n");
1657 		wake_up(&trans_pcie->wait_command_queue);
1658 	} else {
1659 		clear_bit(STATUS_RFKILL_HW, &trans->status);
1660 		if (trans_pcie->opmode_down)
1661 			clear_bit(STATUS_RFKILL_OPMODE, &trans->status);
1662 	}
1663 }
1664 
1665 irqreturn_t iwl_pcie_irq_handler(int irq, void *dev_id)
1666 {
1667 	struct iwl_trans *trans = dev_id;
1668 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1669 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1670 	u32 inta = 0;
1671 	u32 handled = 0;
1672 
1673 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1674 
1675 	spin_lock(&trans_pcie->irq_lock);
1676 
1677 	/* dram interrupt table not set yet,
1678 	 * use legacy interrupt.
1679 	 */
1680 	if (likely(trans_pcie->use_ict))
1681 		inta = iwl_pcie_int_cause_ict(trans);
1682 	else
1683 		inta = iwl_pcie_int_cause_non_ict(trans);
1684 
1685 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1686 		IWL_DEBUG_ISR(trans,
1687 			      "ISR inta 0x%08x, enabled 0x%08x(sw), enabled(hw) 0x%08x, fh 0x%08x\n",
1688 			      inta, trans_pcie->inta_mask,
1689 			      iwl_read32(trans, CSR_INT_MASK),
1690 			      iwl_read32(trans, CSR_FH_INT_STATUS));
1691 		if (inta & (~trans_pcie->inta_mask))
1692 			IWL_DEBUG_ISR(trans,
1693 				      "We got a masked interrupt (0x%08x)\n",
1694 				      inta & (~trans_pcie->inta_mask));
1695 	}
1696 
1697 	inta &= trans_pcie->inta_mask;
1698 
1699 	/*
1700 	 * Ignore interrupt if there's nothing in NIC to service.
1701 	 * This may be due to IRQ shared with another device,
1702 	 * or due to sporadic interrupts thrown from our NIC.
1703 	 */
1704 	if (unlikely(!inta)) {
1705 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
1706 		/*
1707 		 * Re-enable interrupts here since we don't
1708 		 * have anything to service
1709 		 */
1710 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1711 			_iwl_enable_interrupts(trans);
1712 		spin_unlock(&trans_pcie->irq_lock);
1713 		lock_map_release(&trans->sync_cmd_lockdep_map);
1714 		return IRQ_NONE;
1715 	}
1716 
1717 	if (unlikely(inta == 0xFFFFFFFF || (inta & 0xFFFFFFF0) == 0xa5a5a5a0)) {
1718 		/*
1719 		 * Hardware disappeared. It might have
1720 		 * already raised an interrupt.
1721 		 */
1722 		IWL_WARN(trans, "HARDWARE GONE?? INTA == 0x%08x\n", inta);
1723 		spin_unlock(&trans_pcie->irq_lock);
1724 		goto out;
1725 	}
1726 
1727 	/* Ack/clear/reset pending uCode interrupts.
1728 	 * Note:  Some bits in CSR_INT are "OR" of bits in CSR_FH_INT_STATUS,
1729 	 */
1730 	/* There is a hardware bug in the interrupt mask function that some
1731 	 * interrupts (i.e. CSR_INT_BIT_SCD) can still be generated even if
1732 	 * they are disabled in the CSR_INT_MASK register. Furthermore the
1733 	 * ICT interrupt handling mechanism has another bug that might cause
1734 	 * these unmasked interrupts fail to be detected. We workaround the
1735 	 * hardware bugs here by ACKing all the possible interrupts so that
1736 	 * interrupt coalescing can still be achieved.
1737 	 */
1738 	iwl_write32(trans, CSR_INT, inta | ~trans_pcie->inta_mask);
1739 
1740 	if (iwl_have_debug_level(IWL_DL_ISR))
1741 		IWL_DEBUG_ISR(trans, "inta 0x%08x, enabled 0x%08x\n",
1742 			      inta, iwl_read32(trans, CSR_INT_MASK));
1743 
1744 	spin_unlock(&trans_pcie->irq_lock);
1745 
1746 	/* Now service all interrupt bits discovered above. */
1747 	if (inta & CSR_INT_BIT_HW_ERR) {
1748 		IWL_ERR(trans, "Hardware error detected.  Restarting.\n");
1749 
1750 		/* Tell the device to stop sending interrupts */
1751 		iwl_disable_interrupts(trans);
1752 
1753 		isr_stats->hw++;
1754 		iwl_pcie_irq_handle_error(trans);
1755 
1756 		handled |= CSR_INT_BIT_HW_ERR;
1757 
1758 		goto out;
1759 	}
1760 
1761 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1762 		/* NIC fires this, but we don't use it, redundant with WAKEUP */
1763 		if (inta & CSR_INT_BIT_SCD) {
1764 			IWL_DEBUG_ISR(trans,
1765 				      "Scheduler finished to transmit the frame/frames.\n");
1766 			isr_stats->sch++;
1767 		}
1768 
1769 		/* Alive notification via Rx interrupt will do the real work */
1770 		if (inta & CSR_INT_BIT_ALIVE) {
1771 			IWL_DEBUG_ISR(trans, "Alive interrupt\n");
1772 			isr_stats->alive++;
1773 			if (trans->cfg->gen2) {
1774 				/*
1775 				 * We can restock, since firmware configured
1776 				 * the RFH
1777 				 */
1778 				iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
1779 			}
1780 		}
1781 	}
1782 
1783 	/* Safely ignore these bits for debug checks below */
1784 	inta &= ~(CSR_INT_BIT_SCD | CSR_INT_BIT_ALIVE);
1785 
1786 	/* HW RF KILL switch toggled */
1787 	if (inta & CSR_INT_BIT_RF_KILL) {
1788 		iwl_pcie_handle_rfkill_irq(trans);
1789 		handled |= CSR_INT_BIT_RF_KILL;
1790 	}
1791 
1792 	/* Chip got too hot and stopped itself */
1793 	if (inta & CSR_INT_BIT_CT_KILL) {
1794 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
1795 		isr_stats->ctkill++;
1796 		handled |= CSR_INT_BIT_CT_KILL;
1797 	}
1798 
1799 	/* Error detected by uCode */
1800 	if (inta & CSR_INT_BIT_SW_ERR) {
1801 		IWL_ERR(trans, "Microcode SW error detected. "
1802 			" Restarting 0x%X.\n", inta);
1803 		isr_stats->sw++;
1804 		iwl_pcie_irq_handle_error(trans);
1805 		handled |= CSR_INT_BIT_SW_ERR;
1806 	}
1807 
1808 	/* uCode wakes up after power-down sleep */
1809 	if (inta & CSR_INT_BIT_WAKEUP) {
1810 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
1811 		iwl_pcie_rxq_check_wrptr(trans);
1812 		iwl_pcie_txq_check_wrptrs(trans);
1813 
1814 		isr_stats->wakeup++;
1815 
1816 		handled |= CSR_INT_BIT_WAKEUP;
1817 	}
1818 
1819 	/* All uCode command responses, including Tx command responses,
1820 	 * Rx "responses" (frame-received notification), and other
1821 	 * notifications from uCode come through here*/
1822 	if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX |
1823 		    CSR_INT_BIT_RX_PERIODIC)) {
1824 		IWL_DEBUG_ISR(trans, "Rx interrupt\n");
1825 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX)) {
1826 			handled |= (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX);
1827 			iwl_write32(trans, CSR_FH_INT_STATUS,
1828 					CSR_FH_INT_RX_MASK);
1829 		}
1830 		if (inta & CSR_INT_BIT_RX_PERIODIC) {
1831 			handled |= CSR_INT_BIT_RX_PERIODIC;
1832 			iwl_write32(trans,
1833 				CSR_INT, CSR_INT_BIT_RX_PERIODIC);
1834 		}
1835 		/* Sending RX interrupt require many steps to be done in the
1836 		 * the device:
1837 		 * 1- write interrupt to current index in ICT table.
1838 		 * 2- dma RX frame.
1839 		 * 3- update RX shared data to indicate last write index.
1840 		 * 4- send interrupt.
1841 		 * This could lead to RX race, driver could receive RX interrupt
1842 		 * but the shared data changes does not reflect this;
1843 		 * periodic interrupt will detect any dangling Rx activity.
1844 		 */
1845 
1846 		/* Disable periodic interrupt; we use it as just a one-shot. */
1847 		iwl_write8(trans, CSR_INT_PERIODIC_REG,
1848 			    CSR_INT_PERIODIC_DIS);
1849 
1850 		/*
1851 		 * Enable periodic interrupt in 8 msec only if we received
1852 		 * real RX interrupt (instead of just periodic int), to catch
1853 		 * any dangling Rx interrupt.  If it was just the periodic
1854 		 * interrupt, there was no dangling Rx activity, and no need
1855 		 * to extend the periodic interrupt; one-shot is enough.
1856 		 */
1857 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX))
1858 			iwl_write8(trans, CSR_INT_PERIODIC_REG,
1859 				   CSR_INT_PERIODIC_ENA);
1860 
1861 		isr_stats->rx++;
1862 
1863 		local_bh_disable();
1864 		iwl_pcie_rx_handle(trans, 0);
1865 		local_bh_enable();
1866 	}
1867 
1868 	/* This "Tx" DMA channel is used only for loading uCode */
1869 	if (inta & CSR_INT_BIT_FH_TX) {
1870 		iwl_write32(trans, CSR_FH_INT_STATUS, CSR_FH_INT_TX_MASK);
1871 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
1872 		isr_stats->tx++;
1873 		handled |= CSR_INT_BIT_FH_TX;
1874 		/* Wake up uCode load routine, now that load is complete */
1875 		trans_pcie->ucode_write_complete = true;
1876 		wake_up(&trans_pcie->ucode_write_waitq);
1877 	}
1878 
1879 	if (inta & ~handled) {
1880 		IWL_ERR(trans, "Unhandled INTA bits 0x%08x\n", inta & ~handled);
1881 		isr_stats->unhandled++;
1882 	}
1883 
1884 	if (inta & ~(trans_pcie->inta_mask)) {
1885 		IWL_WARN(trans, "Disabled INTA bits 0x%08x were pending\n",
1886 			 inta & ~trans_pcie->inta_mask);
1887 	}
1888 
1889 	spin_lock(&trans_pcie->irq_lock);
1890 	/* only Re-enable all interrupt if disabled by irq */
1891 	if (test_bit(STATUS_INT_ENABLED, &trans->status))
1892 		_iwl_enable_interrupts(trans);
1893 	/* we are loading the firmware, enable FH_TX interrupt only */
1894 	else if (handled & CSR_INT_BIT_FH_TX)
1895 		iwl_enable_fw_load_int(trans);
1896 	/* Re-enable RF_KILL if it occurred */
1897 	else if (handled & CSR_INT_BIT_RF_KILL)
1898 		iwl_enable_rfkill_int(trans);
1899 	spin_unlock(&trans_pcie->irq_lock);
1900 
1901 out:
1902 	lock_map_release(&trans->sync_cmd_lockdep_map);
1903 	return IRQ_HANDLED;
1904 }
1905 
1906 /******************************************************************************
1907  *
1908  * ICT functions
1909  *
1910  ******************************************************************************/
1911 
1912 /* Free dram table */
1913 void iwl_pcie_free_ict(struct iwl_trans *trans)
1914 {
1915 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1916 
1917 	if (trans_pcie->ict_tbl) {
1918 		dma_free_coherent(trans->dev, ICT_SIZE,
1919 				  trans_pcie->ict_tbl,
1920 				  trans_pcie->ict_tbl_dma);
1921 		trans_pcie->ict_tbl = NULL;
1922 		trans_pcie->ict_tbl_dma = 0;
1923 	}
1924 }
1925 
1926 /*
1927  * allocate dram shared table, it is an aligned memory
1928  * block of ICT_SIZE.
1929  * also reset all data related to ICT table interrupt.
1930  */
1931 int iwl_pcie_alloc_ict(struct iwl_trans *trans)
1932 {
1933 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1934 
1935 	trans_pcie->ict_tbl =
1936 		dma_zalloc_coherent(trans->dev, ICT_SIZE,
1937 				   &trans_pcie->ict_tbl_dma,
1938 				   GFP_KERNEL);
1939 	if (!trans_pcie->ict_tbl)
1940 		return -ENOMEM;
1941 
1942 	/* just an API sanity check ... it is guaranteed to be aligned */
1943 	if (WARN_ON(trans_pcie->ict_tbl_dma & (ICT_SIZE - 1))) {
1944 		iwl_pcie_free_ict(trans);
1945 		return -EINVAL;
1946 	}
1947 
1948 	return 0;
1949 }
1950 
1951 /* Device is going up inform it about using ICT interrupt table,
1952  * also we need to tell the driver to start using ICT interrupt.
1953  */
1954 void iwl_pcie_reset_ict(struct iwl_trans *trans)
1955 {
1956 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1957 	u32 val;
1958 
1959 	if (!trans_pcie->ict_tbl)
1960 		return;
1961 
1962 	spin_lock(&trans_pcie->irq_lock);
1963 	_iwl_disable_interrupts(trans);
1964 
1965 	memset(trans_pcie->ict_tbl, 0, ICT_SIZE);
1966 
1967 	val = trans_pcie->ict_tbl_dma >> ICT_SHIFT;
1968 
1969 	val |= CSR_DRAM_INT_TBL_ENABLE |
1970 	       CSR_DRAM_INIT_TBL_WRAP_CHECK |
1971 	       CSR_DRAM_INIT_TBL_WRITE_POINTER;
1972 
1973 	IWL_DEBUG_ISR(trans, "CSR_DRAM_INT_TBL_REG =0x%x\n", val);
1974 
1975 	iwl_write32(trans, CSR_DRAM_INT_TBL_REG, val);
1976 	trans_pcie->use_ict = true;
1977 	trans_pcie->ict_index = 0;
1978 	iwl_write32(trans, CSR_INT, trans_pcie->inta_mask);
1979 	_iwl_enable_interrupts(trans);
1980 	spin_unlock(&trans_pcie->irq_lock);
1981 }
1982 
1983 /* Device is going down disable ict interrupt usage */
1984 void iwl_pcie_disable_ict(struct iwl_trans *trans)
1985 {
1986 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1987 
1988 	spin_lock(&trans_pcie->irq_lock);
1989 	trans_pcie->use_ict = false;
1990 	spin_unlock(&trans_pcie->irq_lock);
1991 }
1992 
1993 irqreturn_t iwl_pcie_isr(int irq, void *data)
1994 {
1995 	struct iwl_trans *trans = data;
1996 
1997 	if (!trans)
1998 		return IRQ_NONE;
1999 
2000 	/* Disable (but don't clear!) interrupts here to avoid
2001 	 * back-to-back ISRs and sporadic interrupts from our NIC.
2002 	 * If we have something to service, the tasklet will re-enable ints.
2003 	 * If we *don't* have something, we'll re-enable before leaving here.
2004 	 */
2005 	iwl_write32(trans, CSR_INT_MASK, 0x00000000);
2006 
2007 	return IRQ_WAKE_THREAD;
2008 }
2009 
2010 irqreturn_t iwl_pcie_msix_isr(int irq, void *data)
2011 {
2012 	return IRQ_WAKE_THREAD;
2013 }
2014 
2015 irqreturn_t iwl_pcie_irq_msix_handler(int irq, void *dev_id)
2016 {
2017 	struct msix_entry *entry = dev_id;
2018 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
2019 	struct iwl_trans *trans = trans_pcie->trans;
2020 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
2021 	u32 inta_fh, inta_hw;
2022 
2023 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
2024 
2025 	spin_lock(&trans_pcie->irq_lock);
2026 	inta_fh = iwl_read32(trans, CSR_MSIX_FH_INT_CAUSES_AD);
2027 	inta_hw = iwl_read32(trans, CSR_MSIX_HW_INT_CAUSES_AD);
2028 	/*
2029 	 * Clear causes registers to avoid being handling the same cause.
2030 	 */
2031 	iwl_write32(trans, CSR_MSIX_FH_INT_CAUSES_AD, inta_fh);
2032 	iwl_write32(trans, CSR_MSIX_HW_INT_CAUSES_AD, inta_hw);
2033 	spin_unlock(&trans_pcie->irq_lock);
2034 
2035 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, true, inta_fh, inta_hw);
2036 
2037 	if (unlikely(!(inta_fh | inta_hw))) {
2038 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
2039 		lock_map_release(&trans->sync_cmd_lockdep_map);
2040 		return IRQ_NONE;
2041 	}
2042 
2043 	if (iwl_have_debug_level(IWL_DL_ISR))
2044 		IWL_DEBUG_ISR(trans, "ISR inta_fh 0x%08x, enabled 0x%08x\n",
2045 			      inta_fh,
2046 			      iwl_read32(trans, CSR_MSIX_FH_INT_MASK_AD));
2047 
2048 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_NON_RX) &&
2049 	    inta_fh & MSIX_FH_INT_CAUSES_Q0) {
2050 		local_bh_disable();
2051 		iwl_pcie_rx_handle(trans, 0);
2052 		local_bh_enable();
2053 	}
2054 
2055 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS) &&
2056 	    inta_fh & MSIX_FH_INT_CAUSES_Q1) {
2057 		local_bh_disable();
2058 		iwl_pcie_rx_handle(trans, 1);
2059 		local_bh_enable();
2060 	}
2061 
2062 	/* This "Tx" DMA channel is used only for loading uCode */
2063 	if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM) {
2064 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
2065 		isr_stats->tx++;
2066 		/*
2067 		 * Wake up uCode load routine,
2068 		 * now that load is complete
2069 		 */
2070 		trans_pcie->ucode_write_complete = true;
2071 		wake_up(&trans_pcie->ucode_write_waitq);
2072 	}
2073 
2074 	/* Error detected by uCode */
2075 	if ((inta_fh & MSIX_FH_INT_CAUSES_FH_ERR) ||
2076 	    (inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR) ||
2077 	    (inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR_V2)) {
2078 		IWL_ERR(trans,
2079 			"Microcode SW error detected. Restarting 0x%X.\n",
2080 			inta_fh);
2081 		isr_stats->sw++;
2082 		iwl_pcie_irq_handle_error(trans);
2083 	}
2084 
2085 	/* After checking FH register check HW register */
2086 	if (iwl_have_debug_level(IWL_DL_ISR))
2087 		IWL_DEBUG_ISR(trans,
2088 			      "ISR inta_hw 0x%08x, enabled 0x%08x\n",
2089 			      inta_hw,
2090 			      iwl_read32(trans, CSR_MSIX_HW_INT_MASK_AD));
2091 
2092 	/* Alive notification via Rx interrupt will do the real work */
2093 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_ALIVE) {
2094 		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
2095 		isr_stats->alive++;
2096 		if (trans->cfg->gen2) {
2097 			/* We can restock, since firmware configured the RFH */
2098 			iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
2099 		}
2100 	}
2101 
2102 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560 &&
2103 	    inta_hw & MSIX_HW_INT_CAUSES_REG_IPC) {
2104 		/* Reflect IML transfer status */
2105 		int res = iwl_read32(trans, CSR_IML_RESP_ADDR);
2106 
2107 		IWL_DEBUG_ISR(trans, "IML transfer status: %d\n", res);
2108 		if (res == IWL_IMAGE_RESP_FAIL) {
2109 			isr_stats->sw++;
2110 			iwl_pcie_irq_handle_error(trans);
2111 		}
2112 	} else if (inta_hw & MSIX_HW_INT_CAUSES_REG_WAKEUP) {
2113 		/* uCode wakes up after power-down sleep */
2114 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
2115 		iwl_pcie_rxq_check_wrptr(trans);
2116 		iwl_pcie_txq_check_wrptrs(trans);
2117 
2118 		isr_stats->wakeup++;
2119 	}
2120 
2121 	/* Chip got too hot and stopped itself */
2122 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_CT_KILL) {
2123 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
2124 		isr_stats->ctkill++;
2125 	}
2126 
2127 	/* HW RF KILL switch toggled */
2128 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RF_KILL)
2129 		iwl_pcie_handle_rfkill_irq(trans);
2130 
2131 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_HW_ERR) {
2132 		IWL_ERR(trans,
2133 			"Hardware error detected. Restarting.\n");
2134 
2135 		isr_stats->hw++;
2136 		iwl_pcie_irq_handle_error(trans);
2137 	}
2138 
2139 	iwl_pcie_clear_irq(trans, entry);
2140 
2141 	lock_map_release(&trans->sync_cmd_lockdep_map);
2142 
2143 	return IRQ_HANDLED;
2144 }
2145