1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /*
3  * Copyright (C) 2003-2014, 2018-2022 Intel Corporation
4  * Copyright (C) 2013-2015 Intel Mobile Communications GmbH
5  * Copyright (C) 2016-2017 Intel Deutschland GmbH
6  */
7 #include <linux/sched.h>
8 #include <linux/wait.h>
9 #include <linux/gfp.h>
10 
11 #include "iwl-prph.h"
12 #include "iwl-io.h"
13 #include "internal.h"
14 #include "iwl-op-mode.h"
15 #include "iwl-context-info-gen3.h"
16 
17 /******************************************************************************
18  *
19  * RX path functions
20  *
21  ******************************************************************************/
22 
23 /*
24  * Rx theory of operation
25  *
26  * Driver allocates a circular buffer of Receive Buffer Descriptors (RBDs),
27  * each of which point to Receive Buffers to be filled by the NIC.  These get
28  * used not only for Rx frames, but for any command response or notification
29  * from the NIC.  The driver and NIC manage the Rx buffers by means
30  * of indexes into the circular buffer.
31  *
32  * Rx Queue Indexes
33  * The host/firmware share two index registers for managing the Rx buffers.
34  *
35  * The READ index maps to the first position that the firmware may be writing
36  * to -- the driver can read up to (but not including) this position and get
37  * good data.
38  * The READ index is managed by the firmware once the card is enabled.
39  *
40  * The WRITE index maps to the last position the driver has read from -- the
41  * position preceding WRITE is the last slot the firmware can place a packet.
42  *
43  * The queue is empty (no good data) if WRITE = READ - 1, and is full if
44  * WRITE = READ.
45  *
46  * During initialization, the host sets up the READ queue position to the first
47  * INDEX position, and WRITE to the last (READ - 1 wrapped)
48  *
49  * When the firmware places a packet in a buffer, it will advance the READ index
50  * and fire the RX interrupt.  The driver can then query the READ index and
51  * process as many packets as possible, moving the WRITE index forward as it
52  * resets the Rx queue buffers with new memory.
53  *
54  * The management in the driver is as follows:
55  * + A list of pre-allocated RBDs is stored in iwl->rxq->rx_free.
56  *   When the interrupt handler is called, the request is processed.
57  *   The page is either stolen - transferred to the upper layer
58  *   or reused - added immediately to the iwl->rxq->rx_free list.
59  * + When the page is stolen - the driver updates the matching queue's used
60  *   count, detaches the RBD and transfers it to the queue used list.
61  *   When there are two used RBDs - they are transferred to the allocator empty
62  *   list. Work is then scheduled for the allocator to start allocating
63  *   eight buffers.
64  *   When there are another 6 used RBDs - they are transferred to the allocator
65  *   empty list and the driver tries to claim the pre-allocated buffers and
66  *   add them to iwl->rxq->rx_free. If it fails - it continues to claim them
67  *   until ready.
68  *   When there are 8+ buffers in the free list - either from allocation or from
69  *   8 reused unstolen pages - restock is called to update the FW and indexes.
70  * + In order to make sure the allocator always has RBDs to use for allocation
71  *   the allocator has initial pool in the size of num_queues*(8-2) - the
72  *   maximum missing RBDs per allocation request (request posted with 2
73  *    empty RBDs, there is no guarantee when the other 6 RBDs are supplied).
74  *   The queues supplies the recycle of the rest of the RBDs.
75  * + A received packet is processed and handed to the kernel network stack,
76  *   detached from the iwl->rxq.  The driver 'processed' index is updated.
77  * + If there are no allocated buffers in iwl->rxq->rx_free,
78  *   the READ INDEX is not incremented and iwl->status(RX_STALLED) is set.
79  *   If there were enough free buffers and RX_STALLED is set it is cleared.
80  *
81  *
82  * Driver sequence:
83  *
84  * iwl_rxq_alloc()            Allocates rx_free
85  * iwl_pcie_rx_replenish()    Replenishes rx_free list from rx_used, and calls
86  *                            iwl_pcie_rxq_restock.
87  *                            Used only during initialization.
88  * iwl_pcie_rxq_restock()     Moves available buffers from rx_free into Rx
89  *                            queue, updates firmware pointers, and updates
90  *                            the WRITE index.
91  * iwl_pcie_rx_allocator()     Background work for allocating pages.
92  *
93  * -- enable interrupts --
94  * ISR - iwl_rx()             Detach iwl_rx_mem_buffers from pool up to the
95  *                            READ INDEX, detaching the SKB from the pool.
96  *                            Moves the packet buffer from queue to rx_used.
97  *                            Posts and claims requests to the allocator.
98  *                            Calls iwl_pcie_rxq_restock to refill any empty
99  *                            slots.
100  *
101  * RBD life-cycle:
102  *
103  * Init:
104  * rxq.pool -> rxq.rx_used -> rxq.rx_free -> rxq.queue
105  *
106  * Regular Receive interrupt:
107  * Page Stolen:
108  * rxq.queue -> rxq.rx_used -> allocator.rbd_empty ->
109  * allocator.rbd_allocated -> rxq.rx_free -> rxq.queue
110  * Page not Stolen:
111  * rxq.queue -> rxq.rx_free -> rxq.queue
112  * ...
113  *
114  */
115 
116 /*
117  * iwl_rxq_space - Return number of free slots available in queue.
118  */
119 static int iwl_rxq_space(const struct iwl_rxq *rxq)
120 {
121 	/* Make sure rx queue size is a power of 2 */
122 	WARN_ON(rxq->queue_size & (rxq->queue_size - 1));
123 
124 	/*
125 	 * There can be up to (RX_QUEUE_SIZE - 1) free slots, to avoid ambiguity
126 	 * between empty and completely full queues.
127 	 * The following is equivalent to modulo by RX_QUEUE_SIZE and is well
128 	 * defined for negative dividends.
129 	 */
130 	return (rxq->read - rxq->write - 1) & (rxq->queue_size - 1);
131 }
132 
133 /*
134  * iwl_dma_addr2rbd_ptr - convert a DMA address to a uCode read buffer ptr
135  */
136 static inline __le32 iwl_pcie_dma_addr2rbd_ptr(dma_addr_t dma_addr)
137 {
138 	return cpu_to_le32((u32)(dma_addr >> 8));
139 }
140 
141 /*
142  * iwl_pcie_rx_stop - stops the Rx DMA
143  */
144 int iwl_pcie_rx_stop(struct iwl_trans *trans)
145 {
146 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
147 		/* TODO: remove this once fw does it */
148 		iwl_write_umac_prph(trans, RFH_RXF_DMA_CFG_GEN3, 0);
149 		return iwl_poll_umac_prph_bit(trans, RFH_GEN_STATUS_GEN3,
150 					      RXF_DMA_IDLE, RXF_DMA_IDLE, 1000);
151 	} else if (trans->trans_cfg->mq_rx_supported) {
152 		iwl_write_prph(trans, RFH_RXF_DMA_CFG, 0);
153 		return iwl_poll_prph_bit(trans, RFH_GEN_STATUS,
154 					   RXF_DMA_IDLE, RXF_DMA_IDLE, 1000);
155 	} else {
156 		iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
157 		return iwl_poll_direct_bit(trans, FH_MEM_RSSR_RX_STATUS_REG,
158 					   FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE,
159 					   1000);
160 	}
161 }
162 
163 /*
164  * iwl_pcie_rxq_inc_wr_ptr - Update the write pointer for the RX queue
165  */
166 static void iwl_pcie_rxq_inc_wr_ptr(struct iwl_trans *trans,
167 				    struct iwl_rxq *rxq)
168 {
169 	u32 reg;
170 
171 	lockdep_assert_held(&rxq->lock);
172 
173 	/*
174 	 * explicitly wake up the NIC if:
175 	 * 1. shadow registers aren't enabled
176 	 * 2. there is a chance that the NIC is asleep
177 	 */
178 	if (!trans->trans_cfg->base_params->shadow_reg_enable &&
179 	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
180 		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
181 
182 		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
183 			IWL_DEBUG_INFO(trans, "Rx queue requesting wakeup, GP1 = 0x%x\n",
184 				       reg);
185 			iwl_set_bit(trans, CSR_GP_CNTRL,
186 				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
187 			rxq->need_update = true;
188 			return;
189 		}
190 	}
191 
192 	rxq->write_actual = round_down(rxq->write, 8);
193 	if (!trans->trans_cfg->mq_rx_supported)
194 		iwl_write32(trans, FH_RSCSR_CHNL0_WPTR, rxq->write_actual);
195 	else if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ)
196 		iwl_write32(trans, HBUS_TARG_WRPTR, rxq->write_actual |
197 			    HBUS_TARG_WRPTR_RX_Q(rxq->id));
198 	else
199 		iwl_write32(trans, RFH_Q_FRBDCB_WIDX_TRG(rxq->id),
200 			    rxq->write_actual);
201 }
202 
203 static void iwl_pcie_rxq_check_wrptr(struct iwl_trans *trans)
204 {
205 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
206 	int i;
207 
208 	for (i = 0; i < trans->num_rx_queues; i++) {
209 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
210 
211 		if (!rxq->need_update)
212 			continue;
213 		spin_lock_bh(&rxq->lock);
214 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
215 		rxq->need_update = false;
216 		spin_unlock_bh(&rxq->lock);
217 	}
218 }
219 
220 static void iwl_pcie_restock_bd(struct iwl_trans *trans,
221 				struct iwl_rxq *rxq,
222 				struct iwl_rx_mem_buffer *rxb)
223 {
224 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
225 		struct iwl_rx_transfer_desc *bd = rxq->bd;
226 
227 		BUILD_BUG_ON(sizeof(*bd) != 2 * sizeof(u64));
228 
229 		bd[rxq->write].addr = cpu_to_le64(rxb->page_dma);
230 		bd[rxq->write].rbid = cpu_to_le16(rxb->vid);
231 	} else {
232 		__le64 *bd = rxq->bd;
233 
234 		bd[rxq->write] = cpu_to_le64(rxb->page_dma | rxb->vid);
235 	}
236 
237 	IWL_DEBUG_RX(trans, "Assigned virtual RB ID %u to queue %d index %d\n",
238 		     (u32)rxb->vid, rxq->id, rxq->write);
239 }
240 
241 /*
242  * iwl_pcie_rxmq_restock - restock implementation for multi-queue rx
243  */
244 static void iwl_pcie_rxmq_restock(struct iwl_trans *trans,
245 				  struct iwl_rxq *rxq)
246 {
247 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
248 	struct iwl_rx_mem_buffer *rxb;
249 
250 	/*
251 	 * If the device isn't enabled - no need to try to add buffers...
252 	 * This can happen when we stop the device and still have an interrupt
253 	 * pending. We stop the APM before we sync the interrupts because we
254 	 * have to (see comment there). On the other hand, since the APM is
255 	 * stopped, we cannot access the HW (in particular not prph).
256 	 * So don't try to restock if the APM has been already stopped.
257 	 */
258 	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
259 		return;
260 
261 	spin_lock_bh(&rxq->lock);
262 	while (rxq->free_count) {
263 		/* Get next free Rx buffer, remove from free list */
264 		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
265 				       list);
266 		list_del(&rxb->list);
267 		rxb->invalid = false;
268 		/* some low bits are expected to be unset (depending on hw) */
269 		WARN_ON(rxb->page_dma & trans_pcie->supported_dma_mask);
270 		/* Point to Rx buffer via next RBD in circular buffer */
271 		iwl_pcie_restock_bd(trans, rxq, rxb);
272 		rxq->write = (rxq->write + 1) & (rxq->queue_size - 1);
273 		rxq->free_count--;
274 	}
275 	spin_unlock_bh(&rxq->lock);
276 
277 	/*
278 	 * If we've added more space for the firmware to place data, tell it.
279 	 * Increment device's write pointer in multiples of 8.
280 	 */
281 	if (rxq->write_actual != (rxq->write & ~0x7)) {
282 		spin_lock_bh(&rxq->lock);
283 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
284 		spin_unlock_bh(&rxq->lock);
285 	}
286 }
287 
288 /*
289  * iwl_pcie_rxsq_restock - restock implementation for single queue rx
290  */
291 static void iwl_pcie_rxsq_restock(struct iwl_trans *trans,
292 				  struct iwl_rxq *rxq)
293 {
294 	struct iwl_rx_mem_buffer *rxb;
295 
296 	/*
297 	 * If the device isn't enabled - not need to try to add buffers...
298 	 * This can happen when we stop the device and still have an interrupt
299 	 * pending. We stop the APM before we sync the interrupts because we
300 	 * have to (see comment there). On the other hand, since the APM is
301 	 * stopped, we cannot access the HW (in particular not prph).
302 	 * So don't try to restock if the APM has been already stopped.
303 	 */
304 	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
305 		return;
306 
307 	spin_lock_bh(&rxq->lock);
308 	while ((iwl_rxq_space(rxq) > 0) && (rxq->free_count)) {
309 		__le32 *bd = (__le32 *)rxq->bd;
310 		/* The overwritten rxb must be a used one */
311 		rxb = rxq->queue[rxq->write];
312 		BUG_ON(rxb && rxb->page);
313 
314 		/* Get next free Rx buffer, remove from free list */
315 		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
316 				       list);
317 		list_del(&rxb->list);
318 		rxb->invalid = false;
319 
320 		/* Point to Rx buffer via next RBD in circular buffer */
321 		bd[rxq->write] = iwl_pcie_dma_addr2rbd_ptr(rxb->page_dma);
322 		rxq->queue[rxq->write] = rxb;
323 		rxq->write = (rxq->write + 1) & RX_QUEUE_MASK;
324 		rxq->free_count--;
325 	}
326 	spin_unlock_bh(&rxq->lock);
327 
328 	/* If we've added more space for the firmware to place data, tell it.
329 	 * Increment device's write pointer in multiples of 8. */
330 	if (rxq->write_actual != (rxq->write & ~0x7)) {
331 		spin_lock_bh(&rxq->lock);
332 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
333 		spin_unlock_bh(&rxq->lock);
334 	}
335 }
336 
337 /*
338  * iwl_pcie_rxq_restock - refill RX queue from pre-allocated pool
339  *
340  * If there are slots in the RX queue that need to be restocked,
341  * and we have free pre-allocated buffers, fill the ranks as much
342  * as we can, pulling from rx_free.
343  *
344  * This moves the 'write' index forward to catch up with 'processed', and
345  * also updates the memory address in the firmware to reference the new
346  * target buffer.
347  */
348 static
349 void iwl_pcie_rxq_restock(struct iwl_trans *trans, struct iwl_rxq *rxq)
350 {
351 	if (trans->trans_cfg->mq_rx_supported)
352 		iwl_pcie_rxmq_restock(trans, rxq);
353 	else
354 		iwl_pcie_rxsq_restock(trans, rxq);
355 }
356 
357 /*
358  * iwl_pcie_rx_alloc_page - allocates and returns a page.
359  *
360  */
361 static struct page *iwl_pcie_rx_alloc_page(struct iwl_trans *trans,
362 					   u32 *offset, gfp_t priority)
363 {
364 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
365 	unsigned int rbsize = iwl_trans_get_rb_size(trans_pcie->rx_buf_size);
366 	unsigned int allocsize = PAGE_SIZE << trans_pcie->rx_page_order;
367 	struct page *page;
368 	gfp_t gfp_mask = priority;
369 
370 	if (trans_pcie->rx_page_order > 0)
371 		gfp_mask |= __GFP_COMP;
372 
373 	if (trans_pcie->alloc_page) {
374 		spin_lock_bh(&trans_pcie->alloc_page_lock);
375 		/* recheck */
376 		if (trans_pcie->alloc_page) {
377 			*offset = trans_pcie->alloc_page_used;
378 			page = trans_pcie->alloc_page;
379 			trans_pcie->alloc_page_used += rbsize;
380 			if (trans_pcie->alloc_page_used >= allocsize)
381 				trans_pcie->alloc_page = NULL;
382 			else
383 				get_page(page);
384 			spin_unlock_bh(&trans_pcie->alloc_page_lock);
385 			return page;
386 		}
387 		spin_unlock_bh(&trans_pcie->alloc_page_lock);
388 	}
389 
390 	/* Alloc a new receive buffer */
391 	page = alloc_pages(gfp_mask, trans_pcie->rx_page_order);
392 	if (!page) {
393 		if (net_ratelimit())
394 			IWL_DEBUG_INFO(trans, "alloc_pages failed, order: %d\n",
395 				       trans_pcie->rx_page_order);
396 		/*
397 		 * Issue an error if we don't have enough pre-allocated
398 		  * buffers.
399 		 */
400 		if (!(gfp_mask & __GFP_NOWARN) && net_ratelimit())
401 			IWL_CRIT(trans,
402 				 "Failed to alloc_pages\n");
403 		return NULL;
404 	}
405 
406 	if (2 * rbsize <= allocsize) {
407 		spin_lock_bh(&trans_pcie->alloc_page_lock);
408 		if (!trans_pcie->alloc_page) {
409 			get_page(page);
410 			trans_pcie->alloc_page = page;
411 			trans_pcie->alloc_page_used = rbsize;
412 		}
413 		spin_unlock_bh(&trans_pcie->alloc_page_lock);
414 	}
415 
416 	*offset = 0;
417 	return page;
418 }
419 
420 /*
421  * iwl_pcie_rxq_alloc_rbs - allocate a page for each used RBD
422  *
423  * A used RBD is an Rx buffer that has been given to the stack. To use it again
424  * a page must be allocated and the RBD must point to the page. This function
425  * doesn't change the HW pointer but handles the list of pages that is used by
426  * iwl_pcie_rxq_restock. The latter function will update the HW to use the newly
427  * allocated buffers.
428  */
429 void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans, gfp_t priority,
430 			    struct iwl_rxq *rxq)
431 {
432 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
433 	struct iwl_rx_mem_buffer *rxb;
434 	struct page *page;
435 
436 	while (1) {
437 		unsigned int offset;
438 
439 		spin_lock_bh(&rxq->lock);
440 		if (list_empty(&rxq->rx_used)) {
441 			spin_unlock_bh(&rxq->lock);
442 			return;
443 		}
444 		spin_unlock_bh(&rxq->lock);
445 
446 		page = iwl_pcie_rx_alloc_page(trans, &offset, priority);
447 		if (!page)
448 			return;
449 
450 		spin_lock_bh(&rxq->lock);
451 
452 		if (list_empty(&rxq->rx_used)) {
453 			spin_unlock_bh(&rxq->lock);
454 			__free_pages(page, trans_pcie->rx_page_order);
455 			return;
456 		}
457 		rxb = list_first_entry(&rxq->rx_used, struct iwl_rx_mem_buffer,
458 				       list);
459 		list_del(&rxb->list);
460 		spin_unlock_bh(&rxq->lock);
461 
462 		BUG_ON(rxb->page);
463 		rxb->page = page;
464 		rxb->offset = offset;
465 		/* Get physical address of the RB */
466 		rxb->page_dma =
467 			dma_map_page(trans->dev, page, rxb->offset,
468 				     trans_pcie->rx_buf_bytes,
469 				     DMA_FROM_DEVICE);
470 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
471 			rxb->page = NULL;
472 			spin_lock_bh(&rxq->lock);
473 			list_add(&rxb->list, &rxq->rx_used);
474 			spin_unlock_bh(&rxq->lock);
475 			__free_pages(page, trans_pcie->rx_page_order);
476 			return;
477 		}
478 
479 		spin_lock_bh(&rxq->lock);
480 
481 		list_add_tail(&rxb->list, &rxq->rx_free);
482 		rxq->free_count++;
483 
484 		spin_unlock_bh(&rxq->lock);
485 	}
486 }
487 
488 void iwl_pcie_free_rbs_pool(struct iwl_trans *trans)
489 {
490 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
491 	int i;
492 
493 	if (!trans_pcie->rx_pool)
494 		return;
495 
496 	for (i = 0; i < RX_POOL_SIZE(trans_pcie->num_rx_bufs); i++) {
497 		if (!trans_pcie->rx_pool[i].page)
498 			continue;
499 		dma_unmap_page(trans->dev, trans_pcie->rx_pool[i].page_dma,
500 			       trans_pcie->rx_buf_bytes, DMA_FROM_DEVICE);
501 		__free_pages(trans_pcie->rx_pool[i].page,
502 			     trans_pcie->rx_page_order);
503 		trans_pcie->rx_pool[i].page = NULL;
504 	}
505 }
506 
507 /*
508  * iwl_pcie_rx_allocator - Allocates pages in the background for RX queues
509  *
510  * Allocates for each received request 8 pages
511  * Called as a scheduled work item.
512  */
513 static void iwl_pcie_rx_allocator(struct iwl_trans *trans)
514 {
515 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
516 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
517 	struct list_head local_empty;
518 	int pending = atomic_read(&rba->req_pending);
519 
520 	IWL_DEBUG_TPT(trans, "Pending allocation requests = %d\n", pending);
521 
522 	/* If we were scheduled - there is at least one request */
523 	spin_lock_bh(&rba->lock);
524 	/* swap out the rba->rbd_empty to a local list */
525 	list_replace_init(&rba->rbd_empty, &local_empty);
526 	spin_unlock_bh(&rba->lock);
527 
528 	while (pending) {
529 		int i;
530 		LIST_HEAD(local_allocated);
531 		gfp_t gfp_mask = GFP_KERNEL;
532 
533 		/* Do not post a warning if there are only a few requests */
534 		if (pending < RX_PENDING_WATERMARK)
535 			gfp_mask |= __GFP_NOWARN;
536 
537 		for (i = 0; i < RX_CLAIM_REQ_ALLOC;) {
538 			struct iwl_rx_mem_buffer *rxb;
539 			struct page *page;
540 
541 			/* List should never be empty - each reused RBD is
542 			 * returned to the list, and initial pool covers any
543 			 * possible gap between the time the page is allocated
544 			 * to the time the RBD is added.
545 			 */
546 			BUG_ON(list_empty(&local_empty));
547 			/* Get the first rxb from the rbd list */
548 			rxb = list_first_entry(&local_empty,
549 					       struct iwl_rx_mem_buffer, list);
550 			BUG_ON(rxb->page);
551 
552 			/* Alloc a new receive buffer */
553 			page = iwl_pcie_rx_alloc_page(trans, &rxb->offset,
554 						      gfp_mask);
555 			if (!page)
556 				continue;
557 			rxb->page = page;
558 
559 			/* Get physical address of the RB */
560 			rxb->page_dma = dma_map_page(trans->dev, page,
561 						     rxb->offset,
562 						     trans_pcie->rx_buf_bytes,
563 						     DMA_FROM_DEVICE);
564 			if (dma_mapping_error(trans->dev, rxb->page_dma)) {
565 				rxb->page = NULL;
566 				__free_pages(page, trans_pcie->rx_page_order);
567 				continue;
568 			}
569 
570 			/* move the allocated entry to the out list */
571 			list_move(&rxb->list, &local_allocated);
572 			i++;
573 		}
574 
575 		atomic_dec(&rba->req_pending);
576 		pending--;
577 
578 		if (!pending) {
579 			pending = atomic_read(&rba->req_pending);
580 			if (pending)
581 				IWL_DEBUG_TPT(trans,
582 					      "Got more pending allocation requests = %d\n",
583 					      pending);
584 		}
585 
586 		spin_lock_bh(&rba->lock);
587 		/* add the allocated rbds to the allocator allocated list */
588 		list_splice_tail(&local_allocated, &rba->rbd_allocated);
589 		/* get more empty RBDs for current pending requests */
590 		list_splice_tail_init(&rba->rbd_empty, &local_empty);
591 		spin_unlock_bh(&rba->lock);
592 
593 		atomic_inc(&rba->req_ready);
594 
595 	}
596 
597 	spin_lock_bh(&rba->lock);
598 	/* return unused rbds to the allocator empty list */
599 	list_splice_tail(&local_empty, &rba->rbd_empty);
600 	spin_unlock_bh(&rba->lock);
601 
602 	IWL_DEBUG_TPT(trans, "%s, exit.\n", __func__);
603 }
604 
605 /*
606  * iwl_pcie_rx_allocator_get - returns the pre-allocated pages
607 .*
608 .* Called by queue when the queue posted allocation request and
609  * has freed 8 RBDs in order to restock itself.
610  * This function directly moves the allocated RBs to the queue's ownership
611  * and updates the relevant counters.
612  */
613 static void iwl_pcie_rx_allocator_get(struct iwl_trans *trans,
614 				      struct iwl_rxq *rxq)
615 {
616 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
617 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
618 	int i;
619 
620 	lockdep_assert_held(&rxq->lock);
621 
622 	/*
623 	 * atomic_dec_if_positive returns req_ready - 1 for any scenario.
624 	 * If req_ready is 0 atomic_dec_if_positive will return -1 and this
625 	 * function will return early, as there are no ready requests.
626 	 * atomic_dec_if_positive will perofrm the *actual* decrement only if
627 	 * req_ready > 0, i.e. - there are ready requests and the function
628 	 * hands one request to the caller.
629 	 */
630 	if (atomic_dec_if_positive(&rba->req_ready) < 0)
631 		return;
632 
633 	spin_lock(&rba->lock);
634 	for (i = 0; i < RX_CLAIM_REQ_ALLOC; i++) {
635 		/* Get next free Rx buffer, remove it from free list */
636 		struct iwl_rx_mem_buffer *rxb =
637 			list_first_entry(&rba->rbd_allocated,
638 					 struct iwl_rx_mem_buffer, list);
639 
640 		list_move(&rxb->list, &rxq->rx_free);
641 	}
642 	spin_unlock(&rba->lock);
643 
644 	rxq->used_count -= RX_CLAIM_REQ_ALLOC;
645 	rxq->free_count += RX_CLAIM_REQ_ALLOC;
646 }
647 
648 void iwl_pcie_rx_allocator_work(struct work_struct *data)
649 {
650 	struct iwl_rb_allocator *rba_p =
651 		container_of(data, struct iwl_rb_allocator, rx_alloc);
652 	struct iwl_trans_pcie *trans_pcie =
653 		container_of(rba_p, struct iwl_trans_pcie, rba);
654 
655 	iwl_pcie_rx_allocator(trans_pcie->trans);
656 }
657 
658 static int iwl_pcie_free_bd_size(struct iwl_trans *trans)
659 {
660 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210)
661 		return sizeof(struct iwl_rx_transfer_desc);
662 
663 	return trans->trans_cfg->mq_rx_supported ?
664 			sizeof(__le64) : sizeof(__le32);
665 }
666 
667 static int iwl_pcie_used_bd_size(struct iwl_trans *trans)
668 {
669 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ)
670 		return sizeof(struct iwl_rx_completion_desc_bz);
671 
672 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210)
673 		return sizeof(struct iwl_rx_completion_desc);
674 
675 	return sizeof(__le32);
676 }
677 
678 static void iwl_pcie_free_rxq_dma(struct iwl_trans *trans,
679 				  struct iwl_rxq *rxq)
680 {
681 	int free_size = iwl_pcie_free_bd_size(trans);
682 
683 	if (rxq->bd)
684 		dma_free_coherent(trans->dev,
685 				  free_size * rxq->queue_size,
686 				  rxq->bd, rxq->bd_dma);
687 	rxq->bd_dma = 0;
688 	rxq->bd = NULL;
689 
690 	rxq->rb_stts_dma = 0;
691 	rxq->rb_stts = NULL;
692 
693 	if (rxq->used_bd)
694 		dma_free_coherent(trans->dev,
695 				  iwl_pcie_used_bd_size(trans) *
696 					rxq->queue_size,
697 				  rxq->used_bd, rxq->used_bd_dma);
698 	rxq->used_bd_dma = 0;
699 	rxq->used_bd = NULL;
700 }
701 
702 static int iwl_pcie_alloc_rxq_dma(struct iwl_trans *trans,
703 				  struct iwl_rxq *rxq)
704 {
705 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
706 	struct device *dev = trans->dev;
707 	int i;
708 	int free_size;
709 	bool use_rx_td = (trans->trans_cfg->device_family >=
710 			  IWL_DEVICE_FAMILY_AX210);
711 	size_t rb_stts_size = use_rx_td ? sizeof(__le16) :
712 			      sizeof(struct iwl_rb_status);
713 
714 	spin_lock_init(&rxq->lock);
715 	if (trans->trans_cfg->mq_rx_supported)
716 		rxq->queue_size = trans->cfg->num_rbds;
717 	else
718 		rxq->queue_size = RX_QUEUE_SIZE;
719 
720 	free_size = iwl_pcie_free_bd_size(trans);
721 
722 	/*
723 	 * Allocate the circular buffer of Read Buffer Descriptors
724 	 * (RBDs)
725 	 */
726 	rxq->bd = dma_alloc_coherent(dev, free_size * rxq->queue_size,
727 				     &rxq->bd_dma, GFP_KERNEL);
728 	if (!rxq->bd)
729 		goto err;
730 
731 	if (trans->trans_cfg->mq_rx_supported) {
732 		rxq->used_bd = dma_alloc_coherent(dev,
733 						  iwl_pcie_used_bd_size(trans) *
734 							rxq->queue_size,
735 						  &rxq->used_bd_dma,
736 						  GFP_KERNEL);
737 		if (!rxq->used_bd)
738 			goto err;
739 	}
740 
741 	rxq->rb_stts = (u8 *)trans_pcie->base_rb_stts + rxq->id * rb_stts_size;
742 	rxq->rb_stts_dma =
743 		trans_pcie->base_rb_stts_dma + rxq->id * rb_stts_size;
744 
745 	return 0;
746 
747 err:
748 	for (i = 0; i < trans->num_rx_queues; i++) {
749 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
750 
751 		iwl_pcie_free_rxq_dma(trans, rxq);
752 	}
753 
754 	return -ENOMEM;
755 }
756 
757 static int iwl_pcie_rx_alloc(struct iwl_trans *trans)
758 {
759 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
760 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
761 	int i, ret;
762 	size_t rb_stts_size = trans->trans_cfg->device_family >=
763 				IWL_DEVICE_FAMILY_AX210 ?
764 			      sizeof(__le16) : sizeof(struct iwl_rb_status);
765 
766 	if (WARN_ON(trans_pcie->rxq))
767 		return -EINVAL;
768 
769 	trans_pcie->rxq = kcalloc(trans->num_rx_queues, sizeof(struct iwl_rxq),
770 				  GFP_KERNEL);
771 	trans_pcie->rx_pool = kcalloc(RX_POOL_SIZE(trans_pcie->num_rx_bufs),
772 				      sizeof(trans_pcie->rx_pool[0]),
773 				      GFP_KERNEL);
774 	trans_pcie->global_table =
775 		kcalloc(RX_POOL_SIZE(trans_pcie->num_rx_bufs),
776 			sizeof(trans_pcie->global_table[0]),
777 			GFP_KERNEL);
778 	if (!trans_pcie->rxq || !trans_pcie->rx_pool ||
779 	    !trans_pcie->global_table) {
780 		ret = -ENOMEM;
781 		goto err;
782 	}
783 
784 	spin_lock_init(&rba->lock);
785 
786 	/*
787 	 * Allocate the driver's pointer to receive buffer status.
788 	 * Allocate for all queues continuously (HW requirement).
789 	 */
790 	trans_pcie->base_rb_stts =
791 			dma_alloc_coherent(trans->dev,
792 					   rb_stts_size * trans->num_rx_queues,
793 					   &trans_pcie->base_rb_stts_dma,
794 					   GFP_KERNEL);
795 	if (!trans_pcie->base_rb_stts) {
796 		ret = -ENOMEM;
797 		goto err;
798 	}
799 
800 	for (i = 0; i < trans->num_rx_queues; i++) {
801 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
802 
803 		rxq->id = i;
804 		ret = iwl_pcie_alloc_rxq_dma(trans, rxq);
805 		if (ret)
806 			goto err;
807 	}
808 	return 0;
809 
810 err:
811 	if (trans_pcie->base_rb_stts) {
812 		dma_free_coherent(trans->dev,
813 				  rb_stts_size * trans->num_rx_queues,
814 				  trans_pcie->base_rb_stts,
815 				  trans_pcie->base_rb_stts_dma);
816 		trans_pcie->base_rb_stts = NULL;
817 		trans_pcie->base_rb_stts_dma = 0;
818 	}
819 	kfree(trans_pcie->rx_pool);
820 	trans_pcie->rx_pool = NULL;
821 	kfree(trans_pcie->global_table);
822 	trans_pcie->global_table = NULL;
823 	kfree(trans_pcie->rxq);
824 	trans_pcie->rxq = NULL;
825 
826 	return ret;
827 }
828 
829 static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
830 {
831 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
832 	u32 rb_size;
833 	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
834 
835 	switch (trans_pcie->rx_buf_size) {
836 	case IWL_AMSDU_4K:
837 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
838 		break;
839 	case IWL_AMSDU_8K:
840 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
841 		break;
842 	case IWL_AMSDU_12K:
843 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_12K;
844 		break;
845 	default:
846 		WARN_ON(1);
847 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
848 	}
849 
850 	if (!iwl_trans_grab_nic_access(trans))
851 		return;
852 
853 	/* Stop Rx DMA */
854 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
855 	/* reset and flush pointers */
856 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
857 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
858 	iwl_write32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
859 
860 	/* Reset driver's Rx queue write index */
861 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
862 
863 	/* Tell device where to find RBD circular buffer in DRAM */
864 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
865 		    (u32)(rxq->bd_dma >> 8));
866 
867 	/* Tell device where in DRAM to update its Rx status */
868 	iwl_write32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
869 		    rxq->rb_stts_dma >> 4);
870 
871 	/* Enable Rx DMA
872 	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
873 	 *      the credit mechanism in 5000 HW RX FIFO
874 	 * Direct rx interrupts to hosts
875 	 * Rx buffer size 4 or 8k or 12k
876 	 * RB timeout 0x10
877 	 * 256 RBDs
878 	 */
879 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
880 		    FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
881 		    FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
882 		    FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
883 		    rb_size |
884 		    (RX_RB_TIMEOUT << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS) |
885 		    (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));
886 
887 	iwl_trans_release_nic_access(trans);
888 
889 	/* Set interrupt coalescing timer to default (2048 usecs) */
890 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
891 
892 	/* W/A for interrupt coalescing bug in 7260 and 3160 */
893 	if (trans->cfg->host_interrupt_operation_mode)
894 		iwl_set_bit(trans, CSR_INT_COALESCING, IWL_HOST_INT_OPER_MODE);
895 }
896 
897 static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
898 {
899 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
900 	u32 rb_size, enabled = 0;
901 	int i;
902 
903 	switch (trans_pcie->rx_buf_size) {
904 	case IWL_AMSDU_2K:
905 		rb_size = RFH_RXF_DMA_RB_SIZE_2K;
906 		break;
907 	case IWL_AMSDU_4K:
908 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
909 		break;
910 	case IWL_AMSDU_8K:
911 		rb_size = RFH_RXF_DMA_RB_SIZE_8K;
912 		break;
913 	case IWL_AMSDU_12K:
914 		rb_size = RFH_RXF_DMA_RB_SIZE_12K;
915 		break;
916 	default:
917 		WARN_ON(1);
918 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
919 	}
920 
921 	if (!iwl_trans_grab_nic_access(trans))
922 		return;
923 
924 	/* Stop Rx DMA */
925 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG, 0);
926 	/* disable free amd used rx queue operation */
927 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, 0);
928 
929 	for (i = 0; i < trans->num_rx_queues; i++) {
930 		/* Tell device where to find RBD free table in DRAM */
931 		iwl_write_prph64_no_grab(trans,
932 					 RFH_Q_FRBDCB_BA_LSB(i),
933 					 trans_pcie->rxq[i].bd_dma);
934 		/* Tell device where to find RBD used table in DRAM */
935 		iwl_write_prph64_no_grab(trans,
936 					 RFH_Q_URBDCB_BA_LSB(i),
937 					 trans_pcie->rxq[i].used_bd_dma);
938 		/* Tell device where in DRAM to update its Rx status */
939 		iwl_write_prph64_no_grab(trans,
940 					 RFH_Q_URBD_STTS_WPTR_LSB(i),
941 					 trans_pcie->rxq[i].rb_stts_dma);
942 		/* Reset device indice tables */
943 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_WIDX(i), 0);
944 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_RIDX(i), 0);
945 		iwl_write_prph_no_grab(trans, RFH_Q_URBDCB_WIDX(i), 0);
946 
947 		enabled |= BIT(i) | BIT(i + 16);
948 	}
949 
950 	/*
951 	 * Enable Rx DMA
952 	 * Rx buffer size 4 or 8k or 12k
953 	 * Min RB size 4 or 8
954 	 * Drop frames that exceed RB size
955 	 * 512 RBDs
956 	 */
957 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG,
958 			       RFH_DMA_EN_ENABLE_VAL | rb_size |
959 			       RFH_RXF_DMA_MIN_RB_4_8 |
960 			       RFH_RXF_DMA_DROP_TOO_LARGE_MASK |
961 			       RFH_RXF_DMA_RBDCB_SIZE_512);
962 
963 	/*
964 	 * Activate DMA snooping.
965 	 * Set RX DMA chunk size to 64B for IOSF and 128B for PCIe
966 	 * Default queue is 0
967 	 */
968 	iwl_write_prph_no_grab(trans, RFH_GEN_CFG,
969 			       RFH_GEN_CFG_RFH_DMA_SNOOP |
970 			       RFH_GEN_CFG_VAL(DEFAULT_RXQ_NUM, 0) |
971 			       RFH_GEN_CFG_SERVICE_DMA_SNOOP |
972 			       RFH_GEN_CFG_VAL(RB_CHUNK_SIZE,
973 					       trans->trans_cfg->integrated ?
974 					       RFH_GEN_CFG_RB_CHUNK_SIZE_64 :
975 					       RFH_GEN_CFG_RB_CHUNK_SIZE_128));
976 	/* Enable the relevant rx queues */
977 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, enabled);
978 
979 	iwl_trans_release_nic_access(trans);
980 
981 	/* Set interrupt coalescing timer to default (2048 usecs) */
982 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
983 }
984 
985 void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq)
986 {
987 	lockdep_assert_held(&rxq->lock);
988 
989 	INIT_LIST_HEAD(&rxq->rx_free);
990 	INIT_LIST_HEAD(&rxq->rx_used);
991 	rxq->free_count = 0;
992 	rxq->used_count = 0;
993 }
994 
995 static int iwl_pcie_rx_handle(struct iwl_trans *trans, int queue, int budget);
996 
997 static int iwl_pcie_napi_poll(struct napi_struct *napi, int budget)
998 {
999 	struct iwl_rxq *rxq = container_of(napi, struct iwl_rxq, napi);
1000 	struct iwl_trans_pcie *trans_pcie;
1001 	struct iwl_trans *trans;
1002 	int ret;
1003 
1004 	trans_pcie = container_of(napi->dev, struct iwl_trans_pcie, napi_dev);
1005 	trans = trans_pcie->trans;
1006 
1007 	ret = iwl_pcie_rx_handle(trans, rxq->id, budget);
1008 
1009 	IWL_DEBUG_ISR(trans, "[%d] handled %d, budget %d\n",
1010 		      rxq->id, ret, budget);
1011 
1012 	if (ret < budget) {
1013 		spin_lock(&trans_pcie->irq_lock);
1014 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1015 			_iwl_enable_interrupts(trans);
1016 		spin_unlock(&trans_pcie->irq_lock);
1017 
1018 		napi_complete_done(&rxq->napi, ret);
1019 	}
1020 
1021 	return ret;
1022 }
1023 
1024 static int iwl_pcie_napi_poll_msix(struct napi_struct *napi, int budget)
1025 {
1026 	struct iwl_rxq *rxq = container_of(napi, struct iwl_rxq, napi);
1027 	struct iwl_trans_pcie *trans_pcie;
1028 	struct iwl_trans *trans;
1029 	int ret;
1030 
1031 	trans_pcie = container_of(napi->dev, struct iwl_trans_pcie, napi_dev);
1032 	trans = trans_pcie->trans;
1033 
1034 	ret = iwl_pcie_rx_handle(trans, rxq->id, budget);
1035 	IWL_DEBUG_ISR(trans, "[%d] handled %d, budget %d\n", rxq->id, ret,
1036 		      budget);
1037 
1038 	if (ret < budget) {
1039 		int irq_line = rxq->id;
1040 
1041 		/* FIRST_RSS is shared with line 0 */
1042 		if (trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS &&
1043 		    rxq->id == 1)
1044 			irq_line = 0;
1045 
1046 		spin_lock(&trans_pcie->irq_lock);
1047 		iwl_pcie_clear_irq(trans, irq_line);
1048 		spin_unlock(&trans_pcie->irq_lock);
1049 
1050 		napi_complete_done(&rxq->napi, ret);
1051 	}
1052 
1053 	return ret;
1054 }
1055 
1056 static int _iwl_pcie_rx_init(struct iwl_trans *trans)
1057 {
1058 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1059 	struct iwl_rxq *def_rxq;
1060 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1061 	int i, err, queue_size, allocator_pool_size, num_alloc;
1062 
1063 	if (!trans_pcie->rxq) {
1064 		err = iwl_pcie_rx_alloc(trans);
1065 		if (err)
1066 			return err;
1067 	}
1068 	def_rxq = trans_pcie->rxq;
1069 
1070 	cancel_work_sync(&rba->rx_alloc);
1071 
1072 	spin_lock_bh(&rba->lock);
1073 	atomic_set(&rba->req_pending, 0);
1074 	atomic_set(&rba->req_ready, 0);
1075 	INIT_LIST_HEAD(&rba->rbd_allocated);
1076 	INIT_LIST_HEAD(&rba->rbd_empty);
1077 	spin_unlock_bh(&rba->lock);
1078 
1079 	/* free all first - we overwrite everything here */
1080 	iwl_pcie_free_rbs_pool(trans);
1081 
1082 	for (i = 0; i < RX_QUEUE_SIZE; i++)
1083 		def_rxq->queue[i] = NULL;
1084 
1085 	for (i = 0; i < trans->num_rx_queues; i++) {
1086 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1087 
1088 		spin_lock_bh(&rxq->lock);
1089 		/*
1090 		 * Set read write pointer to reflect that we have processed
1091 		 * and used all buffers, but have not restocked the Rx queue
1092 		 * with fresh buffers
1093 		 */
1094 		rxq->read = 0;
1095 		rxq->write = 0;
1096 		rxq->write_actual = 0;
1097 		memset(rxq->rb_stts, 0,
1098 		       (trans->trans_cfg->device_family >=
1099 			IWL_DEVICE_FAMILY_AX210) ?
1100 		       sizeof(__le16) : sizeof(struct iwl_rb_status));
1101 
1102 		iwl_pcie_rx_init_rxb_lists(rxq);
1103 
1104 		spin_unlock_bh(&rxq->lock);
1105 
1106 		if (!rxq->napi.poll) {
1107 			int (*poll)(struct napi_struct *, int) = iwl_pcie_napi_poll;
1108 
1109 			if (trans_pcie->msix_enabled)
1110 				poll = iwl_pcie_napi_poll_msix;
1111 
1112 			netif_napi_add(&trans_pcie->napi_dev, &rxq->napi,
1113 				       poll);
1114 			napi_enable(&rxq->napi);
1115 		}
1116 
1117 	}
1118 
1119 	/* move the pool to the default queue and allocator ownerships */
1120 	queue_size = trans->trans_cfg->mq_rx_supported ?
1121 			trans_pcie->num_rx_bufs - 1 : RX_QUEUE_SIZE;
1122 	allocator_pool_size = trans->num_rx_queues *
1123 		(RX_CLAIM_REQ_ALLOC - RX_POST_REQ_ALLOC);
1124 	num_alloc = queue_size + allocator_pool_size;
1125 
1126 	for (i = 0; i < num_alloc; i++) {
1127 		struct iwl_rx_mem_buffer *rxb = &trans_pcie->rx_pool[i];
1128 
1129 		if (i < allocator_pool_size)
1130 			list_add(&rxb->list, &rba->rbd_empty);
1131 		else
1132 			list_add(&rxb->list, &def_rxq->rx_used);
1133 		trans_pcie->global_table[i] = rxb;
1134 		rxb->vid = (u16)(i + 1);
1135 		rxb->invalid = true;
1136 	}
1137 
1138 	iwl_pcie_rxq_alloc_rbs(trans, GFP_KERNEL, def_rxq);
1139 
1140 	return 0;
1141 }
1142 
1143 int iwl_pcie_rx_init(struct iwl_trans *trans)
1144 {
1145 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1146 	int ret = _iwl_pcie_rx_init(trans);
1147 
1148 	if (ret)
1149 		return ret;
1150 
1151 	if (trans->trans_cfg->mq_rx_supported)
1152 		iwl_pcie_rx_mq_hw_init(trans);
1153 	else
1154 		iwl_pcie_rx_hw_init(trans, trans_pcie->rxq);
1155 
1156 	iwl_pcie_rxq_restock(trans, trans_pcie->rxq);
1157 
1158 	spin_lock_bh(&trans_pcie->rxq->lock);
1159 	iwl_pcie_rxq_inc_wr_ptr(trans, trans_pcie->rxq);
1160 	spin_unlock_bh(&trans_pcie->rxq->lock);
1161 
1162 	return 0;
1163 }
1164 
1165 int iwl_pcie_gen2_rx_init(struct iwl_trans *trans)
1166 {
1167 	/* Set interrupt coalescing timer to default (2048 usecs) */
1168 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
1169 
1170 	/*
1171 	 * We don't configure the RFH.
1172 	 * Restock will be done at alive, after firmware configured the RFH.
1173 	 */
1174 	return _iwl_pcie_rx_init(trans);
1175 }
1176 
1177 void iwl_pcie_rx_free(struct iwl_trans *trans)
1178 {
1179 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1180 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1181 	int i;
1182 	size_t rb_stts_size = trans->trans_cfg->device_family >=
1183 				IWL_DEVICE_FAMILY_AX210 ?
1184 			      sizeof(__le16) : sizeof(struct iwl_rb_status);
1185 
1186 	/*
1187 	 * if rxq is NULL, it means that nothing has been allocated,
1188 	 * exit now
1189 	 */
1190 	if (!trans_pcie->rxq) {
1191 		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
1192 		return;
1193 	}
1194 
1195 	cancel_work_sync(&rba->rx_alloc);
1196 
1197 	iwl_pcie_free_rbs_pool(trans);
1198 
1199 	if (trans_pcie->base_rb_stts) {
1200 		dma_free_coherent(trans->dev,
1201 				  rb_stts_size * trans->num_rx_queues,
1202 				  trans_pcie->base_rb_stts,
1203 				  trans_pcie->base_rb_stts_dma);
1204 		trans_pcie->base_rb_stts = NULL;
1205 		trans_pcie->base_rb_stts_dma = 0;
1206 	}
1207 
1208 	for (i = 0; i < trans->num_rx_queues; i++) {
1209 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1210 
1211 		iwl_pcie_free_rxq_dma(trans, rxq);
1212 
1213 		if (rxq->napi.poll) {
1214 			napi_disable(&rxq->napi);
1215 			netif_napi_del(&rxq->napi);
1216 		}
1217 	}
1218 	kfree(trans_pcie->rx_pool);
1219 	kfree(trans_pcie->global_table);
1220 	kfree(trans_pcie->rxq);
1221 
1222 	if (trans_pcie->alloc_page)
1223 		__free_pages(trans_pcie->alloc_page, trans_pcie->rx_page_order);
1224 }
1225 
1226 static void iwl_pcie_rx_move_to_allocator(struct iwl_rxq *rxq,
1227 					  struct iwl_rb_allocator *rba)
1228 {
1229 	spin_lock(&rba->lock);
1230 	list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
1231 	spin_unlock(&rba->lock);
1232 }
1233 
1234 /*
1235  * iwl_pcie_rx_reuse_rbd - Recycle used RBDs
1236  *
1237  * Called when a RBD can be reused. The RBD is transferred to the allocator.
1238  * When there are 2 empty RBDs - a request for allocation is posted
1239  */
1240 static void iwl_pcie_rx_reuse_rbd(struct iwl_trans *trans,
1241 				  struct iwl_rx_mem_buffer *rxb,
1242 				  struct iwl_rxq *rxq, bool emergency)
1243 {
1244 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1245 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1246 
1247 	/* Move the RBD to the used list, will be moved to allocator in batches
1248 	 * before claiming or posting a request*/
1249 	list_add_tail(&rxb->list, &rxq->rx_used);
1250 
1251 	if (unlikely(emergency))
1252 		return;
1253 
1254 	/* Count the allocator owned RBDs */
1255 	rxq->used_count++;
1256 
1257 	/* If we have RX_POST_REQ_ALLOC new released rx buffers -
1258 	 * issue a request for allocator. Modulo RX_CLAIM_REQ_ALLOC is
1259 	 * used for the case we failed to claim RX_CLAIM_REQ_ALLOC,
1260 	 * after but we still need to post another request.
1261 	 */
1262 	if ((rxq->used_count % RX_CLAIM_REQ_ALLOC) == RX_POST_REQ_ALLOC) {
1263 		/* Move the 2 RBDs to the allocator ownership.
1264 		 Allocator has another 6 from pool for the request completion*/
1265 		iwl_pcie_rx_move_to_allocator(rxq, rba);
1266 
1267 		atomic_inc(&rba->req_pending);
1268 		queue_work(rba->alloc_wq, &rba->rx_alloc);
1269 	}
1270 }
1271 
1272 static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
1273 				struct iwl_rxq *rxq,
1274 				struct iwl_rx_mem_buffer *rxb,
1275 				bool emergency,
1276 				int i)
1277 {
1278 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1279 	struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id];
1280 	bool page_stolen = false;
1281 	int max_len = trans_pcie->rx_buf_bytes;
1282 	u32 offset = 0;
1283 
1284 	if (WARN_ON(!rxb))
1285 		return;
1286 
1287 	dma_unmap_page(trans->dev, rxb->page_dma, max_len, DMA_FROM_DEVICE);
1288 
1289 	while (offset + sizeof(u32) + sizeof(struct iwl_cmd_header) < max_len) {
1290 		struct iwl_rx_packet *pkt;
1291 		bool reclaim;
1292 		int len;
1293 		struct iwl_rx_cmd_buffer rxcb = {
1294 			._offset = rxb->offset + offset,
1295 			._rx_page_order = trans_pcie->rx_page_order,
1296 			._page = rxb->page,
1297 			._page_stolen = false,
1298 			.truesize = max_len,
1299 		};
1300 
1301 		pkt = rxb_addr(&rxcb);
1302 
1303 		if (pkt->len_n_flags == cpu_to_le32(FH_RSCSR_FRAME_INVALID)) {
1304 			IWL_DEBUG_RX(trans,
1305 				     "Q %d: RB end marker at offset %d\n",
1306 				     rxq->id, offset);
1307 			break;
1308 		}
1309 
1310 		WARN((le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1311 			FH_RSCSR_RXQ_POS != rxq->id,
1312 		     "frame on invalid queue - is on %d and indicates %d\n",
1313 		     rxq->id,
1314 		     (le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1315 			FH_RSCSR_RXQ_POS);
1316 
1317 		IWL_DEBUG_RX(trans,
1318 			     "Q %d: cmd at offset %d: %s (%.2x.%2x, seq 0x%x)\n",
1319 			     rxq->id, offset,
1320 			     iwl_get_cmd_string(trans,
1321 						WIDE_ID(pkt->hdr.group_id, pkt->hdr.cmd)),
1322 			     pkt->hdr.group_id, pkt->hdr.cmd,
1323 			     le16_to_cpu(pkt->hdr.sequence));
1324 
1325 		len = iwl_rx_packet_len(pkt);
1326 		len += sizeof(u32); /* account for status word */
1327 
1328 		offset += ALIGN(len, FH_RSCSR_FRAME_ALIGN);
1329 
1330 		/* check that what the device tells us made sense */
1331 		if (len < sizeof(*pkt) || offset > max_len)
1332 			break;
1333 
1334 		trace_iwlwifi_dev_rx(trans->dev, trans, pkt, len);
1335 		trace_iwlwifi_dev_rx_data(trans->dev, trans, pkt, len);
1336 
1337 		/* Reclaim a command buffer only if this packet is a response
1338 		 *   to a (driver-originated) command.
1339 		 * If the packet (e.g. Rx frame) originated from uCode,
1340 		 *   there is no command buffer to reclaim.
1341 		 * Ucode should set SEQ_RX_FRAME bit if ucode-originated,
1342 		 *   but apparently a few don't get set; catch them here. */
1343 		reclaim = !(pkt->hdr.sequence & SEQ_RX_FRAME);
1344 		if (reclaim && !pkt->hdr.group_id) {
1345 			int i;
1346 
1347 			for (i = 0; i < trans_pcie->n_no_reclaim_cmds; i++) {
1348 				if (trans_pcie->no_reclaim_cmds[i] ==
1349 							pkt->hdr.cmd) {
1350 					reclaim = false;
1351 					break;
1352 				}
1353 			}
1354 		}
1355 
1356 		if (rxq->id == trans_pcie->def_rx_queue)
1357 			iwl_op_mode_rx(trans->op_mode, &rxq->napi,
1358 				       &rxcb);
1359 		else
1360 			iwl_op_mode_rx_rss(trans->op_mode, &rxq->napi,
1361 					   &rxcb, rxq->id);
1362 
1363 		/*
1364 		 * After here, we should always check rxcb._page_stolen,
1365 		 * if it is true then one of the handlers took the page.
1366 		 */
1367 
1368 		if (reclaim) {
1369 			u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1370 			int index = SEQ_TO_INDEX(sequence);
1371 			int cmd_index = iwl_txq_get_cmd_index(txq, index);
1372 
1373 			kfree_sensitive(txq->entries[cmd_index].free_buf);
1374 			txq->entries[cmd_index].free_buf = NULL;
1375 
1376 			/* Invoke any callbacks, transfer the buffer to caller,
1377 			 * and fire off the (possibly) blocking
1378 			 * iwl_trans_send_cmd()
1379 			 * as we reclaim the driver command queue */
1380 			if (!rxcb._page_stolen)
1381 				iwl_pcie_hcmd_complete(trans, &rxcb);
1382 			else
1383 				IWL_WARN(trans, "Claim null rxb?\n");
1384 		}
1385 
1386 		page_stolen |= rxcb._page_stolen;
1387 		if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210)
1388 			break;
1389 	}
1390 
1391 	/* page was stolen from us -- free our reference */
1392 	if (page_stolen) {
1393 		__free_pages(rxb->page, trans_pcie->rx_page_order);
1394 		rxb->page = NULL;
1395 	}
1396 
1397 	/* Reuse the page if possible. For notification packets and
1398 	 * SKBs that fail to Rx correctly, add them back into the
1399 	 * rx_free list for reuse later. */
1400 	if (rxb->page != NULL) {
1401 		rxb->page_dma =
1402 			dma_map_page(trans->dev, rxb->page, rxb->offset,
1403 				     trans_pcie->rx_buf_bytes,
1404 				     DMA_FROM_DEVICE);
1405 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
1406 			/*
1407 			 * free the page(s) as well to not break
1408 			 * the invariant that the items on the used
1409 			 * list have no page(s)
1410 			 */
1411 			__free_pages(rxb->page, trans_pcie->rx_page_order);
1412 			rxb->page = NULL;
1413 			iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1414 		} else {
1415 			list_add_tail(&rxb->list, &rxq->rx_free);
1416 			rxq->free_count++;
1417 		}
1418 	} else
1419 		iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1420 }
1421 
1422 static struct iwl_rx_mem_buffer *iwl_pcie_get_rxb(struct iwl_trans *trans,
1423 						  struct iwl_rxq *rxq, int i,
1424 						  bool *join)
1425 {
1426 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1427 	struct iwl_rx_mem_buffer *rxb;
1428 	u16 vid;
1429 
1430 	BUILD_BUG_ON(sizeof(struct iwl_rx_completion_desc) != 32);
1431 	BUILD_BUG_ON(sizeof(struct iwl_rx_completion_desc_bz) != 4);
1432 
1433 	if (!trans->trans_cfg->mq_rx_supported) {
1434 		rxb = rxq->queue[i];
1435 		rxq->queue[i] = NULL;
1436 		return rxb;
1437 	}
1438 
1439 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) {
1440 		struct iwl_rx_completion_desc_bz *cd = rxq->used_bd;
1441 
1442 		vid = le16_to_cpu(cd[i].rbid);
1443 		*join = cd[i].flags & IWL_RX_CD_FLAGS_FRAGMENTED;
1444 	} else if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
1445 		struct iwl_rx_completion_desc *cd = rxq->used_bd;
1446 
1447 		vid = le16_to_cpu(cd[i].rbid);
1448 		*join = cd[i].flags & IWL_RX_CD_FLAGS_FRAGMENTED;
1449 	} else {
1450 		__le32 *cd = rxq->used_bd;
1451 
1452 		vid = le32_to_cpu(cd[i]) & 0x0FFF; /* 12-bit VID */
1453 	}
1454 
1455 	if (!vid || vid > RX_POOL_SIZE(trans_pcie->num_rx_bufs))
1456 		goto out_err;
1457 
1458 	rxb = trans_pcie->global_table[vid - 1];
1459 	if (rxb->invalid)
1460 		goto out_err;
1461 
1462 	IWL_DEBUG_RX(trans, "Got virtual RB ID %u\n", (u32)rxb->vid);
1463 
1464 	rxb->invalid = true;
1465 
1466 	return rxb;
1467 
1468 out_err:
1469 	WARN(1, "Invalid rxb from HW %u\n", (u32)vid);
1470 	iwl_force_nmi(trans);
1471 	return NULL;
1472 }
1473 
1474 /*
1475  * iwl_pcie_rx_handle - Main entry function for receiving responses from fw
1476  */
1477 static int iwl_pcie_rx_handle(struct iwl_trans *trans, int queue, int budget)
1478 {
1479 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1480 	struct iwl_rxq *rxq;
1481 	u32 r, i, count = 0, handled = 0;
1482 	bool emergency = false;
1483 
1484 	if (WARN_ON_ONCE(!trans_pcie->rxq || !trans_pcie->rxq[queue].bd))
1485 		return budget;
1486 
1487 	rxq = &trans_pcie->rxq[queue];
1488 
1489 restart:
1490 	spin_lock(&rxq->lock);
1491 	/* uCode's read index (stored in shared DRAM) indicates the last Rx
1492 	 * buffer that the driver may process (last buffer filled by ucode). */
1493 	r = le16_to_cpu(iwl_get_closed_rb_stts(trans, rxq)) & 0x0FFF;
1494 	i = rxq->read;
1495 
1496 	/* W/A 9000 device step A0 wrap-around bug */
1497 	r &= (rxq->queue_size - 1);
1498 
1499 	/* Rx interrupt, but nothing sent from uCode */
1500 	if (i == r)
1501 		IWL_DEBUG_RX(trans, "Q %d: HW = SW = %d\n", rxq->id, r);
1502 
1503 	while (i != r && ++handled < budget) {
1504 		struct iwl_rb_allocator *rba = &trans_pcie->rba;
1505 		struct iwl_rx_mem_buffer *rxb;
1506 		/* number of RBDs still waiting for page allocation */
1507 		u32 rb_pending_alloc =
1508 			atomic_read(&trans_pcie->rba.req_pending) *
1509 			RX_CLAIM_REQ_ALLOC;
1510 		bool join = false;
1511 
1512 		if (unlikely(rb_pending_alloc >= rxq->queue_size / 2 &&
1513 			     !emergency)) {
1514 			iwl_pcie_rx_move_to_allocator(rxq, rba);
1515 			emergency = true;
1516 			IWL_DEBUG_TPT(trans,
1517 				      "RX path is in emergency. Pending allocations %d\n",
1518 				      rb_pending_alloc);
1519 		}
1520 
1521 		IWL_DEBUG_RX(trans, "Q %d: HW = %d, SW = %d\n", rxq->id, r, i);
1522 
1523 		rxb = iwl_pcie_get_rxb(trans, rxq, i, &join);
1524 		if (!rxb)
1525 			goto out;
1526 
1527 		if (unlikely(join || rxq->next_rb_is_fragment)) {
1528 			rxq->next_rb_is_fragment = join;
1529 			/*
1530 			 * We can only get a multi-RB in the following cases:
1531 			 *  - firmware issue, sending a too big notification
1532 			 *  - sniffer mode with a large A-MSDU
1533 			 *  - large MTU frames (>2k)
1534 			 * since the multi-RB functionality is limited to newer
1535 			 * hardware that cannot put multiple entries into a
1536 			 * single RB.
1537 			 *
1538 			 * Right now, the higher layers aren't set up to deal
1539 			 * with that, so discard all of these.
1540 			 */
1541 			list_add_tail(&rxb->list, &rxq->rx_free);
1542 			rxq->free_count++;
1543 		} else {
1544 			iwl_pcie_rx_handle_rb(trans, rxq, rxb, emergency, i);
1545 		}
1546 
1547 		i = (i + 1) & (rxq->queue_size - 1);
1548 
1549 		/*
1550 		 * If we have RX_CLAIM_REQ_ALLOC released rx buffers -
1551 		 * try to claim the pre-allocated buffers from the allocator.
1552 		 * If not ready - will try to reclaim next time.
1553 		 * There is no need to reschedule work - allocator exits only
1554 		 * on success
1555 		 */
1556 		if (rxq->used_count >= RX_CLAIM_REQ_ALLOC)
1557 			iwl_pcie_rx_allocator_get(trans, rxq);
1558 
1559 		if (rxq->used_count % RX_CLAIM_REQ_ALLOC == 0 && !emergency) {
1560 			/* Add the remaining empty RBDs for allocator use */
1561 			iwl_pcie_rx_move_to_allocator(rxq, rba);
1562 		} else if (emergency) {
1563 			count++;
1564 			if (count == 8) {
1565 				count = 0;
1566 				if (rb_pending_alloc < rxq->queue_size / 3) {
1567 					IWL_DEBUG_TPT(trans,
1568 						      "RX path exited emergency. Pending allocations %d\n",
1569 						      rb_pending_alloc);
1570 					emergency = false;
1571 				}
1572 
1573 				rxq->read = i;
1574 				spin_unlock(&rxq->lock);
1575 				iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1576 				iwl_pcie_rxq_restock(trans, rxq);
1577 				goto restart;
1578 			}
1579 		}
1580 	}
1581 out:
1582 	/* Backtrack one entry */
1583 	rxq->read = i;
1584 	spin_unlock(&rxq->lock);
1585 
1586 	/*
1587 	 * handle a case where in emergency there are some unallocated RBDs.
1588 	 * those RBDs are in the used list, but are not tracked by the queue's
1589 	 * used_count which counts allocator owned RBDs.
1590 	 * unallocated emergency RBDs must be allocated on exit, otherwise
1591 	 * when called again the function may not be in emergency mode and
1592 	 * they will be handed to the allocator with no tracking in the RBD
1593 	 * allocator counters, which will lead to them never being claimed back
1594 	 * by the queue.
1595 	 * by allocating them here, they are now in the queue free list, and
1596 	 * will be restocked by the next call of iwl_pcie_rxq_restock.
1597 	 */
1598 	if (unlikely(emergency && count))
1599 		iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1600 
1601 	iwl_pcie_rxq_restock(trans, rxq);
1602 
1603 	return handled;
1604 }
1605 
1606 static struct iwl_trans_pcie *iwl_pcie_get_trans_pcie(struct msix_entry *entry)
1607 {
1608 	u8 queue = entry->entry;
1609 	struct msix_entry *entries = entry - queue;
1610 
1611 	return container_of(entries, struct iwl_trans_pcie, msix_entries[0]);
1612 }
1613 
1614 /*
1615  * iwl_pcie_rx_msix_handle - Main entry function for receiving responses from fw
1616  * This interrupt handler should be used with RSS queue only.
1617  */
1618 irqreturn_t iwl_pcie_irq_rx_msix_handler(int irq, void *dev_id)
1619 {
1620 	struct msix_entry *entry = dev_id;
1621 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
1622 	struct iwl_trans *trans = trans_pcie->trans;
1623 	struct iwl_rxq *rxq = &trans_pcie->rxq[entry->entry];
1624 
1625 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, false, 0, 0);
1626 
1627 	if (WARN_ON(entry->entry >= trans->num_rx_queues))
1628 		return IRQ_NONE;
1629 
1630 	if (!rxq) {
1631 		if (net_ratelimit())
1632 			IWL_ERR(trans,
1633 				"[%d] Got MSI-X interrupt before we have Rx queues\n",
1634 				entry->entry);
1635 		return IRQ_NONE;
1636 	}
1637 
1638 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1639 	IWL_DEBUG_ISR(trans, "[%d] Got interrupt\n", entry->entry);
1640 
1641 	local_bh_disable();
1642 	if (napi_schedule_prep(&rxq->napi))
1643 		__napi_schedule(&rxq->napi);
1644 	else
1645 		iwl_pcie_clear_irq(trans, entry->entry);
1646 	local_bh_enable();
1647 
1648 	lock_map_release(&trans->sync_cmd_lockdep_map);
1649 
1650 	return IRQ_HANDLED;
1651 }
1652 
1653 /*
1654  * iwl_pcie_irq_handle_error - called for HW or SW error interrupt from card
1655  */
1656 static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
1657 {
1658 	int i;
1659 
1660 	/* W/A for WiFi/WiMAX coex and WiMAX own the RF */
1661 	if (trans->cfg->internal_wimax_coex &&
1662 	    !trans->cfg->apmg_not_supported &&
1663 	    (!(iwl_read_prph(trans, APMG_CLK_CTRL_REG) &
1664 			     APMS_CLK_VAL_MRB_FUNC_MODE) ||
1665 	     (iwl_read_prph(trans, APMG_PS_CTRL_REG) &
1666 			    APMG_PS_CTRL_VAL_RESET_REQ))) {
1667 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1668 		iwl_op_mode_wimax_active(trans->op_mode);
1669 		wake_up(&trans->wait_command_queue);
1670 		return;
1671 	}
1672 
1673 	for (i = 0; i < trans->trans_cfg->base_params->num_of_queues; i++) {
1674 		if (!trans->txqs.txq[i])
1675 			continue;
1676 		del_timer(&trans->txqs.txq[i]->stuck_timer);
1677 	}
1678 
1679 	/* The STATUS_FW_ERROR bit is set in this function. This must happen
1680 	 * before we wake up the command caller, to ensure a proper cleanup. */
1681 	iwl_trans_fw_error(trans, false);
1682 
1683 	clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1684 	wake_up(&trans->wait_command_queue);
1685 }
1686 
1687 static u32 iwl_pcie_int_cause_non_ict(struct iwl_trans *trans)
1688 {
1689 	u32 inta;
1690 
1691 	lockdep_assert_held(&IWL_TRANS_GET_PCIE_TRANS(trans)->irq_lock);
1692 
1693 	trace_iwlwifi_dev_irq(trans->dev);
1694 
1695 	/* Discover which interrupts are active/pending */
1696 	inta = iwl_read32(trans, CSR_INT);
1697 
1698 	/* the thread will service interrupts and re-enable them */
1699 	return inta;
1700 }
1701 
1702 /* a device (PCI-E) page is 4096 bytes long */
1703 #define ICT_SHIFT	12
1704 #define ICT_SIZE	(1 << ICT_SHIFT)
1705 #define ICT_COUNT	(ICT_SIZE / sizeof(u32))
1706 
1707 /* interrupt handler using ict table, with this interrupt driver will
1708  * stop using INTA register to get device's interrupt, reading this register
1709  * is expensive, device will write interrupts in ICT dram table, increment
1710  * index then will fire interrupt to driver, driver will OR all ICT table
1711  * entries from current index up to table entry with 0 value. the result is
1712  * the interrupt we need to service, driver will set the entries back to 0 and
1713  * set index.
1714  */
1715 static u32 iwl_pcie_int_cause_ict(struct iwl_trans *trans)
1716 {
1717 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1718 	u32 inta;
1719 	u32 val = 0;
1720 	u32 read;
1721 
1722 	trace_iwlwifi_dev_irq(trans->dev);
1723 
1724 	/* Ignore interrupt if there's nothing in NIC to service.
1725 	 * This may be due to IRQ shared with another device,
1726 	 * or due to sporadic interrupts thrown from our NIC. */
1727 	read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1728 	trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index, read);
1729 	if (!read)
1730 		return 0;
1731 
1732 	/*
1733 	 * Collect all entries up to the first 0, starting from ict_index;
1734 	 * note we already read at ict_index.
1735 	 */
1736 	do {
1737 		val |= read;
1738 		IWL_DEBUG_ISR(trans, "ICT index %d value 0x%08X\n",
1739 				trans_pcie->ict_index, read);
1740 		trans_pcie->ict_tbl[trans_pcie->ict_index] = 0;
1741 		trans_pcie->ict_index =
1742 			((trans_pcie->ict_index + 1) & (ICT_COUNT - 1));
1743 
1744 		read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1745 		trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index,
1746 					   read);
1747 	} while (read);
1748 
1749 	/* We should not get this value, just ignore it. */
1750 	if (val == 0xffffffff)
1751 		val = 0;
1752 
1753 	/*
1754 	 * this is a w/a for a h/w bug. the h/w bug may cause the Rx bit
1755 	 * (bit 15 before shifting it to 31) to clear when using interrupt
1756 	 * coalescing. fortunately, bits 18 and 19 stay set when this happens
1757 	 * so we use them to decide on the real state of the Rx bit.
1758 	 * In order words, bit 15 is set if bit 18 or bit 19 are set.
1759 	 */
1760 	if (val & 0xC0000)
1761 		val |= 0x8000;
1762 
1763 	inta = (0xff & val) | ((0xff00 & val) << 16);
1764 	return inta;
1765 }
1766 
1767 void iwl_pcie_handle_rfkill_irq(struct iwl_trans *trans)
1768 {
1769 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1770 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1771 	bool hw_rfkill, prev, report;
1772 
1773 	mutex_lock(&trans_pcie->mutex);
1774 	prev = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1775 	hw_rfkill = iwl_is_rfkill_set(trans);
1776 	if (hw_rfkill) {
1777 		set_bit(STATUS_RFKILL_OPMODE, &trans->status);
1778 		set_bit(STATUS_RFKILL_HW, &trans->status);
1779 	}
1780 	if (trans_pcie->opmode_down)
1781 		report = hw_rfkill;
1782 	else
1783 		report = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1784 
1785 	IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
1786 		 hw_rfkill ? "disable radio" : "enable radio");
1787 
1788 	isr_stats->rfkill++;
1789 
1790 	if (prev != report)
1791 		iwl_trans_pcie_rf_kill(trans, report);
1792 	mutex_unlock(&trans_pcie->mutex);
1793 
1794 	if (hw_rfkill) {
1795 		if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
1796 				       &trans->status))
1797 			IWL_DEBUG_RF_KILL(trans,
1798 					  "Rfkill while SYNC HCMD in flight\n");
1799 		wake_up(&trans->wait_command_queue);
1800 	} else {
1801 		clear_bit(STATUS_RFKILL_HW, &trans->status);
1802 		if (trans_pcie->opmode_down)
1803 			clear_bit(STATUS_RFKILL_OPMODE, &trans->status);
1804 	}
1805 }
1806 
1807 irqreturn_t iwl_pcie_irq_handler(int irq, void *dev_id)
1808 {
1809 	struct iwl_trans *trans = dev_id;
1810 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1811 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1812 	u32 inta = 0;
1813 	u32 handled = 0;
1814 	bool polling = false;
1815 
1816 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1817 
1818 	spin_lock_bh(&trans_pcie->irq_lock);
1819 
1820 	/* dram interrupt table not set yet,
1821 	 * use legacy interrupt.
1822 	 */
1823 	if (likely(trans_pcie->use_ict))
1824 		inta = iwl_pcie_int_cause_ict(trans);
1825 	else
1826 		inta = iwl_pcie_int_cause_non_ict(trans);
1827 
1828 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1829 		IWL_DEBUG_ISR(trans,
1830 			      "ISR inta 0x%08x, enabled 0x%08x(sw), enabled(hw) 0x%08x, fh 0x%08x\n",
1831 			      inta, trans_pcie->inta_mask,
1832 			      iwl_read32(trans, CSR_INT_MASK),
1833 			      iwl_read32(trans, CSR_FH_INT_STATUS));
1834 		if (inta & (~trans_pcie->inta_mask))
1835 			IWL_DEBUG_ISR(trans,
1836 				      "We got a masked interrupt (0x%08x)\n",
1837 				      inta & (~trans_pcie->inta_mask));
1838 	}
1839 
1840 	inta &= trans_pcie->inta_mask;
1841 
1842 	/*
1843 	 * Ignore interrupt if there's nothing in NIC to service.
1844 	 * This may be due to IRQ shared with another device,
1845 	 * or due to sporadic interrupts thrown from our NIC.
1846 	 */
1847 	if (unlikely(!inta)) {
1848 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
1849 		/*
1850 		 * Re-enable interrupts here since we don't
1851 		 * have anything to service
1852 		 */
1853 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1854 			_iwl_enable_interrupts(trans);
1855 		spin_unlock_bh(&trans_pcie->irq_lock);
1856 		lock_map_release(&trans->sync_cmd_lockdep_map);
1857 		return IRQ_NONE;
1858 	}
1859 
1860 	if (unlikely(inta == 0xFFFFFFFF || (inta & 0xFFFFFFF0) == 0xa5a5a5a0)) {
1861 		/*
1862 		 * Hardware disappeared. It might have
1863 		 * already raised an interrupt.
1864 		 */
1865 		IWL_WARN(trans, "HARDWARE GONE?? INTA == 0x%08x\n", inta);
1866 		spin_unlock_bh(&trans_pcie->irq_lock);
1867 		goto out;
1868 	}
1869 
1870 	/* Ack/clear/reset pending uCode interrupts.
1871 	 * Note:  Some bits in CSR_INT are "OR" of bits in CSR_FH_INT_STATUS,
1872 	 */
1873 	/* There is a hardware bug in the interrupt mask function that some
1874 	 * interrupts (i.e. CSR_INT_BIT_SCD) can still be generated even if
1875 	 * they are disabled in the CSR_INT_MASK register. Furthermore the
1876 	 * ICT interrupt handling mechanism has another bug that might cause
1877 	 * these unmasked interrupts fail to be detected. We workaround the
1878 	 * hardware bugs here by ACKing all the possible interrupts so that
1879 	 * interrupt coalescing can still be achieved.
1880 	 */
1881 	iwl_write32(trans, CSR_INT, inta | ~trans_pcie->inta_mask);
1882 
1883 	if (iwl_have_debug_level(IWL_DL_ISR))
1884 		IWL_DEBUG_ISR(trans, "inta 0x%08x, enabled 0x%08x\n",
1885 			      inta, iwl_read32(trans, CSR_INT_MASK));
1886 
1887 	spin_unlock_bh(&trans_pcie->irq_lock);
1888 
1889 	/* Now service all interrupt bits discovered above. */
1890 	if (inta & CSR_INT_BIT_HW_ERR) {
1891 		IWL_ERR(trans, "Hardware error detected.  Restarting.\n");
1892 
1893 		/* Tell the device to stop sending interrupts */
1894 		iwl_disable_interrupts(trans);
1895 
1896 		isr_stats->hw++;
1897 		iwl_pcie_irq_handle_error(trans);
1898 
1899 		handled |= CSR_INT_BIT_HW_ERR;
1900 
1901 		goto out;
1902 	}
1903 
1904 	/* NIC fires this, but we don't use it, redundant with WAKEUP */
1905 	if (inta & CSR_INT_BIT_SCD) {
1906 		IWL_DEBUG_ISR(trans,
1907 			      "Scheduler finished to transmit the frame/frames.\n");
1908 		isr_stats->sch++;
1909 	}
1910 
1911 	/* Alive notification via Rx interrupt will do the real work */
1912 	if (inta & CSR_INT_BIT_ALIVE) {
1913 		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
1914 		isr_stats->alive++;
1915 		if (trans->trans_cfg->gen2) {
1916 			/*
1917 			 * We can restock, since firmware configured
1918 			 * the RFH
1919 			 */
1920 			iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
1921 		}
1922 
1923 		handled |= CSR_INT_BIT_ALIVE;
1924 	}
1925 
1926 	/* Safely ignore these bits for debug checks below */
1927 	inta &= ~(CSR_INT_BIT_SCD | CSR_INT_BIT_ALIVE);
1928 
1929 	/* HW RF KILL switch toggled */
1930 	if (inta & CSR_INT_BIT_RF_KILL) {
1931 		iwl_pcie_handle_rfkill_irq(trans);
1932 		handled |= CSR_INT_BIT_RF_KILL;
1933 	}
1934 
1935 	/* Chip got too hot and stopped itself */
1936 	if (inta & CSR_INT_BIT_CT_KILL) {
1937 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
1938 		isr_stats->ctkill++;
1939 		handled |= CSR_INT_BIT_CT_KILL;
1940 	}
1941 
1942 	/* Error detected by uCode */
1943 	if (inta & CSR_INT_BIT_SW_ERR) {
1944 		IWL_ERR(trans, "Microcode SW error detected. "
1945 			" Restarting 0x%X.\n", inta);
1946 		isr_stats->sw++;
1947 		iwl_pcie_irq_handle_error(trans);
1948 		handled |= CSR_INT_BIT_SW_ERR;
1949 	}
1950 
1951 	/* uCode wakes up after power-down sleep */
1952 	if (inta & CSR_INT_BIT_WAKEUP) {
1953 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
1954 		iwl_pcie_rxq_check_wrptr(trans);
1955 		iwl_pcie_txq_check_wrptrs(trans);
1956 
1957 		isr_stats->wakeup++;
1958 
1959 		handled |= CSR_INT_BIT_WAKEUP;
1960 	}
1961 
1962 	/* All uCode command responses, including Tx command responses,
1963 	 * Rx "responses" (frame-received notification), and other
1964 	 * notifications from uCode come through here*/
1965 	if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX |
1966 		    CSR_INT_BIT_RX_PERIODIC)) {
1967 		IWL_DEBUG_ISR(trans, "Rx interrupt\n");
1968 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX)) {
1969 			handled |= (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX);
1970 			iwl_write32(trans, CSR_FH_INT_STATUS,
1971 					CSR_FH_INT_RX_MASK);
1972 		}
1973 		if (inta & CSR_INT_BIT_RX_PERIODIC) {
1974 			handled |= CSR_INT_BIT_RX_PERIODIC;
1975 			iwl_write32(trans,
1976 				CSR_INT, CSR_INT_BIT_RX_PERIODIC);
1977 		}
1978 		/* Sending RX interrupt require many steps to be done in the
1979 		 * device:
1980 		 * 1- write interrupt to current index in ICT table.
1981 		 * 2- dma RX frame.
1982 		 * 3- update RX shared data to indicate last write index.
1983 		 * 4- send interrupt.
1984 		 * This could lead to RX race, driver could receive RX interrupt
1985 		 * but the shared data changes does not reflect this;
1986 		 * periodic interrupt will detect any dangling Rx activity.
1987 		 */
1988 
1989 		/* Disable periodic interrupt; we use it as just a one-shot. */
1990 		iwl_write8(trans, CSR_INT_PERIODIC_REG,
1991 			    CSR_INT_PERIODIC_DIS);
1992 
1993 		/*
1994 		 * Enable periodic interrupt in 8 msec only if we received
1995 		 * real RX interrupt (instead of just periodic int), to catch
1996 		 * any dangling Rx interrupt.  If it was just the periodic
1997 		 * interrupt, there was no dangling Rx activity, and no need
1998 		 * to extend the periodic interrupt; one-shot is enough.
1999 		 */
2000 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX))
2001 			iwl_write8(trans, CSR_INT_PERIODIC_REG,
2002 				   CSR_INT_PERIODIC_ENA);
2003 
2004 		isr_stats->rx++;
2005 
2006 		local_bh_disable();
2007 		if (napi_schedule_prep(&trans_pcie->rxq[0].napi)) {
2008 			polling = true;
2009 			__napi_schedule(&trans_pcie->rxq[0].napi);
2010 		}
2011 		local_bh_enable();
2012 	}
2013 
2014 	/* This "Tx" DMA channel is used only for loading uCode */
2015 	if (inta & CSR_INT_BIT_FH_TX) {
2016 		iwl_write32(trans, CSR_FH_INT_STATUS, CSR_FH_INT_TX_MASK);
2017 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
2018 		isr_stats->tx++;
2019 		handled |= CSR_INT_BIT_FH_TX;
2020 		/* Wake up uCode load routine, now that load is complete */
2021 		trans_pcie->ucode_write_complete = true;
2022 		wake_up(&trans_pcie->ucode_write_waitq);
2023 		/* Wake up IMR write routine, now that write to SRAM is complete */
2024 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2025 			trans_pcie->imr_status = IMR_D2S_COMPLETED;
2026 			wake_up(&trans_pcie->ucode_write_waitq);
2027 		}
2028 	}
2029 
2030 	if (inta & ~handled) {
2031 		IWL_ERR(trans, "Unhandled INTA bits 0x%08x\n", inta & ~handled);
2032 		isr_stats->unhandled++;
2033 	}
2034 
2035 	if (inta & ~(trans_pcie->inta_mask)) {
2036 		IWL_WARN(trans, "Disabled INTA bits 0x%08x were pending\n",
2037 			 inta & ~trans_pcie->inta_mask);
2038 	}
2039 
2040 	if (!polling) {
2041 		spin_lock_bh(&trans_pcie->irq_lock);
2042 		/* only Re-enable all interrupt if disabled by irq */
2043 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
2044 			_iwl_enable_interrupts(trans);
2045 		/* we are loading the firmware, enable FH_TX interrupt only */
2046 		else if (handled & CSR_INT_BIT_FH_TX)
2047 			iwl_enable_fw_load_int(trans);
2048 		/* Re-enable RF_KILL if it occurred */
2049 		else if (handled & CSR_INT_BIT_RF_KILL)
2050 			iwl_enable_rfkill_int(trans);
2051 		/* Re-enable the ALIVE / Rx interrupt if it occurred */
2052 		else if (handled & (CSR_INT_BIT_ALIVE | CSR_INT_BIT_FH_RX))
2053 			iwl_enable_fw_load_int_ctx_info(trans);
2054 		spin_unlock_bh(&trans_pcie->irq_lock);
2055 	}
2056 
2057 out:
2058 	lock_map_release(&trans->sync_cmd_lockdep_map);
2059 	return IRQ_HANDLED;
2060 }
2061 
2062 /******************************************************************************
2063  *
2064  * ICT functions
2065  *
2066  ******************************************************************************/
2067 
2068 /* Free dram table */
2069 void iwl_pcie_free_ict(struct iwl_trans *trans)
2070 {
2071 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2072 
2073 	if (trans_pcie->ict_tbl) {
2074 		dma_free_coherent(trans->dev, ICT_SIZE,
2075 				  trans_pcie->ict_tbl,
2076 				  trans_pcie->ict_tbl_dma);
2077 		trans_pcie->ict_tbl = NULL;
2078 		trans_pcie->ict_tbl_dma = 0;
2079 	}
2080 }
2081 
2082 /*
2083  * allocate dram shared table, it is an aligned memory
2084  * block of ICT_SIZE.
2085  * also reset all data related to ICT table interrupt.
2086  */
2087 int iwl_pcie_alloc_ict(struct iwl_trans *trans)
2088 {
2089 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2090 
2091 	trans_pcie->ict_tbl =
2092 		dma_alloc_coherent(trans->dev, ICT_SIZE,
2093 				   &trans_pcie->ict_tbl_dma, GFP_KERNEL);
2094 	if (!trans_pcie->ict_tbl)
2095 		return -ENOMEM;
2096 
2097 	/* just an API sanity check ... it is guaranteed to be aligned */
2098 	if (WARN_ON(trans_pcie->ict_tbl_dma & (ICT_SIZE - 1))) {
2099 		iwl_pcie_free_ict(trans);
2100 		return -EINVAL;
2101 	}
2102 
2103 	return 0;
2104 }
2105 
2106 /* Device is going up inform it about using ICT interrupt table,
2107  * also we need to tell the driver to start using ICT interrupt.
2108  */
2109 void iwl_pcie_reset_ict(struct iwl_trans *trans)
2110 {
2111 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2112 	u32 val;
2113 
2114 	if (!trans_pcie->ict_tbl)
2115 		return;
2116 
2117 	spin_lock_bh(&trans_pcie->irq_lock);
2118 	_iwl_disable_interrupts(trans);
2119 
2120 	memset(trans_pcie->ict_tbl, 0, ICT_SIZE);
2121 
2122 	val = trans_pcie->ict_tbl_dma >> ICT_SHIFT;
2123 
2124 	val |= CSR_DRAM_INT_TBL_ENABLE |
2125 	       CSR_DRAM_INIT_TBL_WRAP_CHECK |
2126 	       CSR_DRAM_INIT_TBL_WRITE_POINTER;
2127 
2128 	IWL_DEBUG_ISR(trans, "CSR_DRAM_INT_TBL_REG =0x%x\n", val);
2129 
2130 	iwl_write32(trans, CSR_DRAM_INT_TBL_REG, val);
2131 	trans_pcie->use_ict = true;
2132 	trans_pcie->ict_index = 0;
2133 	iwl_write32(trans, CSR_INT, trans_pcie->inta_mask);
2134 	_iwl_enable_interrupts(trans);
2135 	spin_unlock_bh(&trans_pcie->irq_lock);
2136 }
2137 
2138 /* Device is going down disable ict interrupt usage */
2139 void iwl_pcie_disable_ict(struct iwl_trans *trans)
2140 {
2141 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2142 
2143 	spin_lock_bh(&trans_pcie->irq_lock);
2144 	trans_pcie->use_ict = false;
2145 	spin_unlock_bh(&trans_pcie->irq_lock);
2146 }
2147 
2148 irqreturn_t iwl_pcie_isr(int irq, void *data)
2149 {
2150 	struct iwl_trans *trans = data;
2151 
2152 	if (!trans)
2153 		return IRQ_NONE;
2154 
2155 	/* Disable (but don't clear!) interrupts here to avoid
2156 	 * back-to-back ISRs and sporadic interrupts from our NIC.
2157 	 * If we have something to service, the tasklet will re-enable ints.
2158 	 * If we *don't* have something, we'll re-enable before leaving here.
2159 	 */
2160 	iwl_write32(trans, CSR_INT_MASK, 0x00000000);
2161 
2162 	return IRQ_WAKE_THREAD;
2163 }
2164 
2165 irqreturn_t iwl_pcie_msix_isr(int irq, void *data)
2166 {
2167 	return IRQ_WAKE_THREAD;
2168 }
2169 
2170 irqreturn_t iwl_pcie_irq_msix_handler(int irq, void *dev_id)
2171 {
2172 	struct msix_entry *entry = dev_id;
2173 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
2174 	struct iwl_trans *trans = trans_pcie->trans;
2175 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
2176 	u32 inta_fh_msk = ~MSIX_FH_INT_CAUSES_DATA_QUEUE;
2177 	u32 inta_fh, inta_hw;
2178 	bool polling = false;
2179 	bool sw_err;
2180 
2181 	if (trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_NON_RX)
2182 		inta_fh_msk |= MSIX_FH_INT_CAUSES_Q0;
2183 
2184 	if (trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS)
2185 		inta_fh_msk |= MSIX_FH_INT_CAUSES_Q1;
2186 
2187 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
2188 
2189 	spin_lock_bh(&trans_pcie->irq_lock);
2190 	inta_fh = iwl_read32(trans, CSR_MSIX_FH_INT_CAUSES_AD);
2191 	inta_hw = iwl_read32(trans, CSR_MSIX_HW_INT_CAUSES_AD);
2192 	/*
2193 	 * Clear causes registers to avoid being handling the same cause.
2194 	 */
2195 	iwl_write32(trans, CSR_MSIX_FH_INT_CAUSES_AD, inta_fh & inta_fh_msk);
2196 	iwl_write32(trans, CSR_MSIX_HW_INT_CAUSES_AD, inta_hw);
2197 	spin_unlock_bh(&trans_pcie->irq_lock);
2198 
2199 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, true, inta_fh, inta_hw);
2200 
2201 	if (unlikely(!(inta_fh | inta_hw))) {
2202 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
2203 		lock_map_release(&trans->sync_cmd_lockdep_map);
2204 		return IRQ_NONE;
2205 	}
2206 
2207 	if (iwl_have_debug_level(IWL_DL_ISR)) {
2208 		IWL_DEBUG_ISR(trans,
2209 			      "ISR[%d] inta_fh 0x%08x, enabled (sw) 0x%08x (hw) 0x%08x\n",
2210 			      entry->entry, inta_fh, trans_pcie->fh_mask,
2211 			      iwl_read32(trans, CSR_MSIX_FH_INT_MASK_AD));
2212 		if (inta_fh & ~trans_pcie->fh_mask)
2213 			IWL_DEBUG_ISR(trans,
2214 				      "We got a masked interrupt (0x%08x)\n",
2215 				      inta_fh & ~trans_pcie->fh_mask);
2216 	}
2217 
2218 	inta_fh &= trans_pcie->fh_mask;
2219 
2220 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_NON_RX) &&
2221 	    inta_fh & MSIX_FH_INT_CAUSES_Q0) {
2222 		local_bh_disable();
2223 		if (napi_schedule_prep(&trans_pcie->rxq[0].napi)) {
2224 			polling = true;
2225 			__napi_schedule(&trans_pcie->rxq[0].napi);
2226 		}
2227 		local_bh_enable();
2228 	}
2229 
2230 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS) &&
2231 	    inta_fh & MSIX_FH_INT_CAUSES_Q1) {
2232 		local_bh_disable();
2233 		if (napi_schedule_prep(&trans_pcie->rxq[1].napi)) {
2234 			polling = true;
2235 			__napi_schedule(&trans_pcie->rxq[1].napi);
2236 		}
2237 		local_bh_enable();
2238 	}
2239 
2240 	/* This "Tx" DMA channel is used only for loading uCode */
2241 	if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM &&
2242 	    trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2243 		IWL_DEBUG_ISR(trans, "IMR Complete interrupt\n");
2244 		isr_stats->tx++;
2245 
2246 		/* Wake up IMR routine once write to SRAM is complete */
2247 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2248 			trans_pcie->imr_status = IMR_D2S_COMPLETED;
2249 			wake_up(&trans_pcie->ucode_write_waitq);
2250 		}
2251 	} else if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM) {
2252 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
2253 		isr_stats->tx++;
2254 		/*
2255 		 * Wake up uCode load routine,
2256 		 * now that load is complete
2257 		 */
2258 		trans_pcie->ucode_write_complete = true;
2259 		wake_up(&trans_pcie->ucode_write_waitq);
2260 
2261 		/* Wake up IMR routine once write to SRAM is complete */
2262 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2263 			trans_pcie->imr_status = IMR_D2S_COMPLETED;
2264 			wake_up(&trans_pcie->ucode_write_waitq);
2265 		}
2266 	}
2267 
2268 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ)
2269 		sw_err = inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR_BZ;
2270 	else
2271 		sw_err = inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR;
2272 
2273 	/* Error detected by uCode */
2274 	if ((inta_fh & MSIX_FH_INT_CAUSES_FH_ERR) || sw_err) {
2275 		IWL_ERR(trans,
2276 			"Microcode SW error detected. Restarting 0x%X.\n",
2277 			inta_fh);
2278 		isr_stats->sw++;
2279 		/* during FW reset flow report errors from there */
2280 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2281 			trans_pcie->imr_status = IMR_D2S_ERROR;
2282 			wake_up(&trans_pcie->imr_waitq);
2283 		} else if (trans_pcie->fw_reset_state == FW_RESET_REQUESTED) {
2284 			trans_pcie->fw_reset_state = FW_RESET_ERROR;
2285 			wake_up(&trans_pcie->fw_reset_waitq);
2286 		} else {
2287 			iwl_pcie_irq_handle_error(trans);
2288 		}
2289 	}
2290 
2291 	/* After checking FH register check HW register */
2292 	if (iwl_have_debug_level(IWL_DL_ISR)) {
2293 		IWL_DEBUG_ISR(trans,
2294 			      "ISR[%d] inta_hw 0x%08x, enabled (sw) 0x%08x (hw) 0x%08x\n",
2295 			      entry->entry, inta_hw, trans_pcie->hw_mask,
2296 			      iwl_read32(trans, CSR_MSIX_HW_INT_MASK_AD));
2297 		if (inta_hw & ~trans_pcie->hw_mask)
2298 			IWL_DEBUG_ISR(trans,
2299 				      "We got a masked interrupt 0x%08x\n",
2300 				      inta_hw & ~trans_pcie->hw_mask);
2301 	}
2302 
2303 	inta_hw &= trans_pcie->hw_mask;
2304 
2305 	/* Alive notification via Rx interrupt will do the real work */
2306 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_ALIVE) {
2307 		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
2308 		isr_stats->alive++;
2309 		if (trans->trans_cfg->gen2) {
2310 			/* We can restock, since firmware configured the RFH */
2311 			iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
2312 		}
2313 	}
2314 
2315 	/*
2316 	 * In some rare cases when the HW is in a bad state, we may
2317 	 * get this interrupt too early, when prph_info is still NULL.
2318 	 * So make sure that it's not NULL to prevent crashing.
2319 	 */
2320 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_WAKEUP && trans_pcie->prph_info) {
2321 		u32 sleep_notif =
2322 			le32_to_cpu(trans_pcie->prph_info->sleep_notif);
2323 		if (sleep_notif == IWL_D3_SLEEP_STATUS_SUSPEND ||
2324 		    sleep_notif == IWL_D3_SLEEP_STATUS_RESUME) {
2325 			IWL_DEBUG_ISR(trans,
2326 				      "Sx interrupt: sleep notification = 0x%x\n",
2327 				      sleep_notif);
2328 			trans_pcie->sx_complete = true;
2329 			wake_up(&trans_pcie->sx_waitq);
2330 		} else {
2331 			/* uCode wakes up after power-down sleep */
2332 			IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
2333 			iwl_pcie_rxq_check_wrptr(trans);
2334 			iwl_pcie_txq_check_wrptrs(trans);
2335 
2336 			isr_stats->wakeup++;
2337 		}
2338 	}
2339 
2340 	/* Chip got too hot and stopped itself */
2341 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_CT_KILL) {
2342 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
2343 		isr_stats->ctkill++;
2344 	}
2345 
2346 	/* HW RF KILL switch toggled */
2347 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RF_KILL)
2348 		iwl_pcie_handle_rfkill_irq(trans);
2349 
2350 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_HW_ERR) {
2351 		IWL_ERR(trans,
2352 			"Hardware error detected. Restarting.\n");
2353 
2354 		isr_stats->hw++;
2355 		trans->dbg.hw_error = true;
2356 		iwl_pcie_irq_handle_error(trans);
2357 	}
2358 
2359 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RESET_DONE) {
2360 		IWL_DEBUG_ISR(trans, "Reset flow completed\n");
2361 		trans_pcie->fw_reset_state = FW_RESET_OK;
2362 		wake_up(&trans_pcie->fw_reset_waitq);
2363 	}
2364 
2365 	if (!polling)
2366 		iwl_pcie_clear_irq(trans, entry->entry);
2367 
2368 	lock_map_release(&trans->sync_cmd_lockdep_map);
2369 
2370 	return IRQ_HANDLED;
2371 }
2372