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  *
7  * Portions of this file are derived from the ipw3945 project, as well
8  * as portions of the ieee80211 subsystem header files.
9  *
10  * This program is free software; you can redistribute it and/or modify it
11  * under the terms of version 2 of the GNU General Public License as
12  * published by the Free Software Foundation.
13  *
14  * This program is distributed in the hope that it will be useful, but WITHOUT
15  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
17  * more details.
18  *
19  * You should have received a copy of the GNU General Public License along with
20  * this program; if not, write to the Free Software Foundation, Inc.,
21  * 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA
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 
40 /******************************************************************************
41  *
42  * RX path functions
43  *
44  ******************************************************************************/
45 
46 /*
47  * Rx theory of operation
48  *
49  * Driver allocates a circular buffer of Receive Buffer Descriptors (RBDs),
50  * each of which point to Receive Buffers to be filled by the NIC.  These get
51  * used not only for Rx frames, but for any command response or notification
52  * from the NIC.  The driver and NIC manage the Rx buffers by means
53  * of indexes into the circular buffer.
54  *
55  * Rx Queue Indexes
56  * The host/firmware share two index registers for managing the Rx buffers.
57  *
58  * The READ index maps to the first position that the firmware may be writing
59  * to -- the driver can read up to (but not including) this position and get
60  * good data.
61  * The READ index is managed by the firmware once the card is enabled.
62  *
63  * The WRITE index maps to the last position the driver has read from -- the
64  * position preceding WRITE is the last slot the firmware can place a packet.
65  *
66  * The queue is empty (no good data) if WRITE = READ - 1, and is full if
67  * WRITE = READ.
68  *
69  * During initialization, the host sets up the READ queue position to the first
70  * INDEX position, and WRITE to the last (READ - 1 wrapped)
71  *
72  * When the firmware places a packet in a buffer, it will advance the READ index
73  * and fire the RX interrupt.  The driver can then query the READ index and
74  * process as many packets as possible, moving the WRITE index forward as it
75  * resets the Rx queue buffers with new memory.
76  *
77  * The management in the driver is as follows:
78  * + A list of pre-allocated RBDs is stored in iwl->rxq->rx_free.
79  *   When the interrupt handler is called, the request is processed.
80  *   The page is either stolen - transferred to the upper layer
81  *   or reused - added immediately to the iwl->rxq->rx_free list.
82  * + When the page is stolen - the driver updates the matching queue's used
83  *   count, detaches the RBD and transfers it to the queue used list.
84  *   When there are two used RBDs - they are transferred to the allocator empty
85  *   list. Work is then scheduled for the allocator to start allocating
86  *   eight buffers.
87  *   When there are another 6 used RBDs - they are transferred to the allocator
88  *   empty list and the driver tries to claim the pre-allocated buffers and
89  *   add them to iwl->rxq->rx_free. If it fails - it continues to claim them
90  *   until ready.
91  *   When there are 8+ buffers in the free list - either from allocation or from
92  *   8 reused unstolen pages - restock is called to update the FW and indexes.
93  * + In order to make sure the allocator always has RBDs to use for allocation
94  *   the allocator has initial pool in the size of num_queues*(8-2) - the
95  *   maximum missing RBDs per allocation request (request posted with 2
96  *    empty RBDs, there is no guarantee when the other 6 RBDs are supplied).
97  *   The queues supplies the recycle of the rest of the RBDs.
98  * + A received packet is processed and handed to the kernel network stack,
99  *   detached from the iwl->rxq.  The driver 'processed' index is updated.
100  * + If there are no allocated buffers in iwl->rxq->rx_free,
101  *   the READ INDEX is not incremented and iwl->status(RX_STALLED) is set.
102  *   If there were enough free buffers and RX_STALLED is set it is cleared.
103  *
104  *
105  * Driver sequence:
106  *
107  * iwl_rxq_alloc()            Allocates rx_free
108  * iwl_pcie_rx_replenish()    Replenishes rx_free list from rx_used, and calls
109  *                            iwl_pcie_rxq_restock.
110  *                            Used only during initialization.
111  * iwl_pcie_rxq_restock()     Moves available buffers from rx_free into Rx
112  *                            queue, updates firmware pointers, and updates
113  *                            the WRITE index.
114  * iwl_pcie_rx_allocator()     Background work for allocating pages.
115  *
116  * -- enable interrupts --
117  * ISR - iwl_rx()             Detach iwl_rx_mem_buffers from pool up to the
118  *                            READ INDEX, detaching the SKB from the pool.
119  *                            Moves the packet buffer from queue to rx_used.
120  *                            Posts and claims requests to the allocator.
121  *                            Calls iwl_pcie_rxq_restock to refill any empty
122  *                            slots.
123  *
124  * RBD life-cycle:
125  *
126  * Init:
127  * rxq.pool -> rxq.rx_used -> rxq.rx_free -> rxq.queue
128  *
129  * Regular Receive interrupt:
130  * Page Stolen:
131  * rxq.queue -> rxq.rx_used -> allocator.rbd_empty ->
132  * allocator.rbd_allocated -> rxq.rx_free -> rxq.queue
133  * Page not Stolen:
134  * rxq.queue -> rxq.rx_free -> rxq.queue
135  * ...
136  *
137  */
138 
139 /*
140  * iwl_rxq_space - Return number of free slots available in queue.
141  */
142 static int iwl_rxq_space(const struct iwl_rxq *rxq)
143 {
144 	/* Make sure rx queue size is a power of 2 */
145 	WARN_ON(rxq->queue_size & (rxq->queue_size - 1));
146 
147 	/*
148 	 * There can be up to (RX_QUEUE_SIZE - 1) free slots, to avoid ambiguity
149 	 * between empty and completely full queues.
150 	 * The following is equivalent to modulo by RX_QUEUE_SIZE and is well
151 	 * defined for negative dividends.
152 	 */
153 	return (rxq->read - rxq->write - 1) & (rxq->queue_size - 1);
154 }
155 
156 /*
157  * iwl_dma_addr2rbd_ptr - convert a DMA address to a uCode read buffer ptr
158  */
159 static inline __le32 iwl_pcie_dma_addr2rbd_ptr(dma_addr_t dma_addr)
160 {
161 	return cpu_to_le32((u32)(dma_addr >> 8));
162 }
163 
164 /*
165  * iwl_pcie_rx_stop - stops the Rx DMA
166  */
167 int iwl_pcie_rx_stop(struct iwl_trans *trans)
168 {
169 	if (trans->cfg->mq_rx_supported) {
170 		iwl_write_prph(trans, RFH_RXF_DMA_CFG, 0);
171 		return iwl_poll_prph_bit(trans, RFH_GEN_STATUS,
172 					   RXF_DMA_IDLE, RXF_DMA_IDLE, 1000);
173 	} else {
174 		iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
175 		return iwl_poll_direct_bit(trans, FH_MEM_RSSR_RX_STATUS_REG,
176 					   FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE,
177 					   1000);
178 	}
179 }
180 
181 /*
182  * iwl_pcie_rxq_inc_wr_ptr - Update the write pointer for the RX queue
183  */
184 static void iwl_pcie_rxq_inc_wr_ptr(struct iwl_trans *trans,
185 				    struct iwl_rxq *rxq)
186 {
187 	u32 reg;
188 
189 	lockdep_assert_held(&rxq->lock);
190 
191 	/*
192 	 * explicitly wake up the NIC if:
193 	 * 1. shadow registers aren't enabled
194 	 * 2. there is a chance that the NIC is asleep
195 	 */
196 	if (!trans->cfg->base_params->shadow_reg_enable &&
197 	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
198 		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
199 
200 		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
201 			IWL_DEBUG_INFO(trans, "Rx queue requesting wakeup, GP1 = 0x%x\n",
202 				       reg);
203 			iwl_set_bit(trans, CSR_GP_CNTRL,
204 				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
205 			rxq->need_update = true;
206 			return;
207 		}
208 	}
209 
210 	rxq->write_actual = round_down(rxq->write, 8);
211 	if (trans->cfg->mq_rx_supported)
212 		iwl_write32(trans, RFH_Q_FRBDCB_WIDX_TRG(rxq->id),
213 			    rxq->write_actual);
214 	else
215 		iwl_write32(trans, FH_RSCSR_CHNL0_WPTR, rxq->write_actual);
216 }
217 
218 static void iwl_pcie_rxq_check_wrptr(struct iwl_trans *trans)
219 {
220 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
221 	int i;
222 
223 	for (i = 0; i < trans->num_rx_queues; i++) {
224 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
225 
226 		if (!rxq->need_update)
227 			continue;
228 		spin_lock(&rxq->lock);
229 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
230 		rxq->need_update = false;
231 		spin_unlock(&rxq->lock);
232 	}
233 }
234 
235 /*
236  * iwl_pcie_rxmq_restock - restock implementation for multi-queue rx
237  */
238 static void iwl_pcie_rxmq_restock(struct iwl_trans *trans,
239 				  struct iwl_rxq *rxq)
240 {
241 	struct iwl_rx_mem_buffer *rxb;
242 
243 	/*
244 	 * If the device isn't enabled - no need to try to add buffers...
245 	 * This can happen when we stop the device and still have an interrupt
246 	 * pending. We stop the APM before we sync the interrupts because we
247 	 * have to (see comment there). On the other hand, since the APM is
248 	 * stopped, we cannot access the HW (in particular not prph).
249 	 * So don't try to restock if the APM has been already stopped.
250 	 */
251 	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
252 		return;
253 
254 	spin_lock(&rxq->lock);
255 	while (rxq->free_count) {
256 		__le64 *bd = (__le64 *)rxq->bd;
257 
258 		/* Get next free Rx buffer, remove from free list */
259 		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
260 				       list);
261 		list_del(&rxb->list);
262 		rxb->invalid = false;
263 		/* 12 first bits are expected to be empty */
264 		WARN_ON(rxb->page_dma & DMA_BIT_MASK(12));
265 		/* Point to Rx buffer via next RBD in circular buffer */
266 		bd[rxq->write] = cpu_to_le64(rxb->page_dma | rxb->vid);
267 		rxq->write = (rxq->write + 1) & MQ_RX_TABLE_MASK;
268 		rxq->free_count--;
269 	}
270 	spin_unlock(&rxq->lock);
271 
272 	/*
273 	 * If we've added more space for the firmware to place data, tell it.
274 	 * Increment device's write pointer in multiples of 8.
275 	 */
276 	if (rxq->write_actual != (rxq->write & ~0x7)) {
277 		spin_lock(&rxq->lock);
278 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
279 		spin_unlock(&rxq->lock);
280 	}
281 }
282 
283 /*
284  * iwl_pcie_rxsq_restock - restock implementation for single queue rx
285  */
286 static void iwl_pcie_rxsq_restock(struct iwl_trans *trans,
287 				  struct iwl_rxq *rxq)
288 {
289 	struct iwl_rx_mem_buffer *rxb;
290 
291 	/*
292 	 * If the device isn't enabled - not need to try to add buffers...
293 	 * This can happen when we stop the device and still have an interrupt
294 	 * pending. We stop the APM before we sync the interrupts because we
295 	 * have to (see comment there). On the other hand, since the APM is
296 	 * stopped, we cannot access the HW (in particular not prph).
297 	 * So don't try to restock if the APM has been already stopped.
298 	 */
299 	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
300 		return;
301 
302 	spin_lock(&rxq->lock);
303 	while ((iwl_rxq_space(rxq) > 0) && (rxq->free_count)) {
304 		__le32 *bd = (__le32 *)rxq->bd;
305 		/* The overwritten rxb must be a used one */
306 		rxb = rxq->queue[rxq->write];
307 		BUG_ON(rxb && rxb->page);
308 
309 		/* Get next free Rx buffer, remove from free list */
310 		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
311 				       list);
312 		list_del(&rxb->list);
313 		rxb->invalid = false;
314 
315 		/* Point to Rx buffer via next RBD in circular buffer */
316 		bd[rxq->write] = iwl_pcie_dma_addr2rbd_ptr(rxb->page_dma);
317 		rxq->queue[rxq->write] = rxb;
318 		rxq->write = (rxq->write + 1) & RX_QUEUE_MASK;
319 		rxq->free_count--;
320 	}
321 	spin_unlock(&rxq->lock);
322 
323 	/* If we've added more space for the firmware to place data, tell it.
324 	 * Increment device's write pointer in multiples of 8. */
325 	if (rxq->write_actual != (rxq->write & ~0x7)) {
326 		spin_lock(&rxq->lock);
327 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
328 		spin_unlock(&rxq->lock);
329 	}
330 }
331 
332 /*
333  * iwl_pcie_rxq_restock - refill RX queue from pre-allocated pool
334  *
335  * If there are slots in the RX queue that need to be restocked,
336  * and we have free pre-allocated buffers, fill the ranks as much
337  * as we can, pulling from rx_free.
338  *
339  * This moves the 'write' index forward to catch up with 'processed', and
340  * also updates the memory address in the firmware to reference the new
341  * target buffer.
342  */
343 static
344 void iwl_pcie_rxq_restock(struct iwl_trans *trans, struct iwl_rxq *rxq)
345 {
346 	if (trans->cfg->mq_rx_supported)
347 		iwl_pcie_rxmq_restock(trans, rxq);
348 	else
349 		iwl_pcie_rxsq_restock(trans, rxq);
350 }
351 
352 /*
353  * iwl_pcie_rx_alloc_page - allocates and returns a page.
354  *
355  */
356 static struct page *iwl_pcie_rx_alloc_page(struct iwl_trans *trans,
357 					   gfp_t priority)
358 {
359 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
360 	struct page *page;
361 	gfp_t gfp_mask = priority;
362 
363 	if (trans_pcie->rx_page_order > 0)
364 		gfp_mask |= __GFP_COMP;
365 
366 	/* Alloc a new receive buffer */
367 	page = alloc_pages(gfp_mask, trans_pcie->rx_page_order);
368 	if (!page) {
369 		if (net_ratelimit())
370 			IWL_DEBUG_INFO(trans, "alloc_pages failed, order: %d\n",
371 				       trans_pcie->rx_page_order);
372 		/*
373 		 * Issue an error if we don't have enough pre-allocated
374 		  * buffers.
375 `		 */
376 		if (!(gfp_mask & __GFP_NOWARN) && net_ratelimit())
377 			IWL_CRIT(trans,
378 				 "Failed to alloc_pages\n");
379 		return NULL;
380 	}
381 	return page;
382 }
383 
384 /*
385  * iwl_pcie_rxq_alloc_rbs - allocate a page for each used RBD
386  *
387  * A used RBD is an Rx buffer that has been given to the stack. To use it again
388  * a page must be allocated and the RBD must point to the page. This function
389  * doesn't change the HW pointer but handles the list of pages that is used by
390  * iwl_pcie_rxq_restock. The latter function will update the HW to use the newly
391  * allocated buffers.
392  */
393 static void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans, gfp_t priority,
394 				   struct iwl_rxq *rxq)
395 {
396 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
397 	struct iwl_rx_mem_buffer *rxb;
398 	struct page *page;
399 
400 	while (1) {
401 		spin_lock(&rxq->lock);
402 		if (list_empty(&rxq->rx_used)) {
403 			spin_unlock(&rxq->lock);
404 			return;
405 		}
406 		spin_unlock(&rxq->lock);
407 
408 		/* Alloc a new receive buffer */
409 		page = iwl_pcie_rx_alloc_page(trans, priority);
410 		if (!page)
411 			return;
412 
413 		spin_lock(&rxq->lock);
414 
415 		if (list_empty(&rxq->rx_used)) {
416 			spin_unlock(&rxq->lock);
417 			__free_pages(page, trans_pcie->rx_page_order);
418 			return;
419 		}
420 		rxb = list_first_entry(&rxq->rx_used, struct iwl_rx_mem_buffer,
421 				       list);
422 		list_del(&rxb->list);
423 		spin_unlock(&rxq->lock);
424 
425 		BUG_ON(rxb->page);
426 		rxb->page = page;
427 		/* Get physical address of the RB */
428 		rxb->page_dma =
429 			dma_map_page(trans->dev, page, 0,
430 				     PAGE_SIZE << trans_pcie->rx_page_order,
431 				     DMA_FROM_DEVICE);
432 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
433 			rxb->page = NULL;
434 			spin_lock(&rxq->lock);
435 			list_add(&rxb->list, &rxq->rx_used);
436 			spin_unlock(&rxq->lock);
437 			__free_pages(page, trans_pcie->rx_page_order);
438 			return;
439 		}
440 
441 		spin_lock(&rxq->lock);
442 
443 		list_add_tail(&rxb->list, &rxq->rx_free);
444 		rxq->free_count++;
445 
446 		spin_unlock(&rxq->lock);
447 	}
448 }
449 
450 static void iwl_pcie_free_rbs_pool(struct iwl_trans *trans)
451 {
452 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
453 	int i;
454 
455 	for (i = 0; i < RX_POOL_SIZE; i++) {
456 		if (!trans_pcie->rx_pool[i].page)
457 			continue;
458 		dma_unmap_page(trans->dev, trans_pcie->rx_pool[i].page_dma,
459 			       PAGE_SIZE << trans_pcie->rx_page_order,
460 			       DMA_FROM_DEVICE);
461 		__free_pages(trans_pcie->rx_pool[i].page,
462 			     trans_pcie->rx_page_order);
463 		trans_pcie->rx_pool[i].page = NULL;
464 	}
465 }
466 
467 /*
468  * iwl_pcie_rx_allocator - Allocates pages in the background for RX queues
469  *
470  * Allocates for each received request 8 pages
471  * Called as a scheduled work item.
472  */
473 static void iwl_pcie_rx_allocator(struct iwl_trans *trans)
474 {
475 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
476 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
477 	struct list_head local_empty;
478 	int pending = atomic_xchg(&rba->req_pending, 0);
479 
480 	IWL_DEBUG_RX(trans, "Pending allocation requests = %d\n", pending);
481 
482 	/* If we were scheduled - there is at least one request */
483 	spin_lock(&rba->lock);
484 	/* swap out the rba->rbd_empty to a local list */
485 	list_replace_init(&rba->rbd_empty, &local_empty);
486 	spin_unlock(&rba->lock);
487 
488 	while (pending) {
489 		int i;
490 		LIST_HEAD(local_allocated);
491 		gfp_t gfp_mask = GFP_KERNEL;
492 
493 		/* Do not post a warning if there are only a few requests */
494 		if (pending < RX_PENDING_WATERMARK)
495 			gfp_mask |= __GFP_NOWARN;
496 
497 		for (i = 0; i < RX_CLAIM_REQ_ALLOC;) {
498 			struct iwl_rx_mem_buffer *rxb;
499 			struct page *page;
500 
501 			/* List should never be empty - each reused RBD is
502 			 * returned to the list, and initial pool covers any
503 			 * possible gap between the time the page is allocated
504 			 * to the time the RBD is added.
505 			 */
506 			BUG_ON(list_empty(&local_empty));
507 			/* Get the first rxb from the rbd list */
508 			rxb = list_first_entry(&local_empty,
509 					       struct iwl_rx_mem_buffer, list);
510 			BUG_ON(rxb->page);
511 
512 			/* Alloc a new receive buffer */
513 			page = iwl_pcie_rx_alloc_page(trans, gfp_mask);
514 			if (!page)
515 				continue;
516 			rxb->page = page;
517 
518 			/* Get physical address of the RB */
519 			rxb->page_dma = dma_map_page(trans->dev, page, 0,
520 					PAGE_SIZE << trans_pcie->rx_page_order,
521 					DMA_FROM_DEVICE);
522 			if (dma_mapping_error(trans->dev, rxb->page_dma)) {
523 				rxb->page = NULL;
524 				__free_pages(page, trans_pcie->rx_page_order);
525 				continue;
526 			}
527 
528 			/* move the allocated entry to the out list */
529 			list_move(&rxb->list, &local_allocated);
530 			i++;
531 		}
532 
533 		pending--;
534 		if (!pending) {
535 			pending = atomic_xchg(&rba->req_pending, 0);
536 			IWL_DEBUG_RX(trans,
537 				     "Pending allocation requests = %d\n",
538 				     pending);
539 		}
540 
541 		spin_lock(&rba->lock);
542 		/* add the allocated rbds to the allocator allocated list */
543 		list_splice_tail(&local_allocated, &rba->rbd_allocated);
544 		/* get more empty RBDs for current pending requests */
545 		list_splice_tail_init(&rba->rbd_empty, &local_empty);
546 		spin_unlock(&rba->lock);
547 
548 		atomic_inc(&rba->req_ready);
549 	}
550 
551 	spin_lock(&rba->lock);
552 	/* return unused rbds to the allocator empty list */
553 	list_splice_tail(&local_empty, &rba->rbd_empty);
554 	spin_unlock(&rba->lock);
555 }
556 
557 /*
558  * iwl_pcie_rx_allocator_get - returns the pre-allocated pages
559 .*
560 .* Called by queue when the queue posted allocation request and
561  * has freed 8 RBDs in order to restock itself.
562  * This function directly moves the allocated RBs to the queue's ownership
563  * and updates the relevant counters.
564  */
565 static void iwl_pcie_rx_allocator_get(struct iwl_trans *trans,
566 				      struct iwl_rxq *rxq)
567 {
568 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
569 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
570 	int i;
571 
572 	lockdep_assert_held(&rxq->lock);
573 
574 	/*
575 	 * atomic_dec_if_positive returns req_ready - 1 for any scenario.
576 	 * If req_ready is 0 atomic_dec_if_positive will return -1 and this
577 	 * function will return early, as there are no ready requests.
578 	 * atomic_dec_if_positive will perofrm the *actual* decrement only if
579 	 * req_ready > 0, i.e. - there are ready requests and the function
580 	 * hands one request to the caller.
581 	 */
582 	if (atomic_dec_if_positive(&rba->req_ready) < 0)
583 		return;
584 
585 	spin_lock(&rba->lock);
586 	for (i = 0; i < RX_CLAIM_REQ_ALLOC; i++) {
587 		/* Get next free Rx buffer, remove it from free list */
588 		struct iwl_rx_mem_buffer *rxb =
589 			list_first_entry(&rba->rbd_allocated,
590 					 struct iwl_rx_mem_buffer, list);
591 
592 		list_move(&rxb->list, &rxq->rx_free);
593 	}
594 	spin_unlock(&rba->lock);
595 
596 	rxq->used_count -= RX_CLAIM_REQ_ALLOC;
597 	rxq->free_count += RX_CLAIM_REQ_ALLOC;
598 }
599 
600 void iwl_pcie_rx_allocator_work(struct work_struct *data)
601 {
602 	struct iwl_rb_allocator *rba_p =
603 		container_of(data, struct iwl_rb_allocator, rx_alloc);
604 	struct iwl_trans_pcie *trans_pcie =
605 		container_of(rba_p, struct iwl_trans_pcie, rba);
606 
607 	iwl_pcie_rx_allocator(trans_pcie->trans);
608 }
609 
610 static int iwl_pcie_rx_alloc(struct iwl_trans *trans)
611 {
612 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
613 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
614 	struct device *dev = trans->dev;
615 	int i;
616 	int free_size = trans->cfg->mq_rx_supported ? sizeof(__le64) :
617 						      sizeof(__le32);
618 
619 	if (WARN_ON(trans_pcie->rxq))
620 		return -EINVAL;
621 
622 	trans_pcie->rxq = kcalloc(trans->num_rx_queues, sizeof(struct iwl_rxq),
623 				  GFP_KERNEL);
624 	if (!trans_pcie->rxq)
625 		return -EINVAL;
626 
627 	spin_lock_init(&rba->lock);
628 
629 	for (i = 0; i < trans->num_rx_queues; i++) {
630 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
631 
632 		spin_lock_init(&rxq->lock);
633 		if (trans->cfg->mq_rx_supported)
634 			rxq->queue_size = MQ_RX_TABLE_SIZE;
635 		else
636 			rxq->queue_size = RX_QUEUE_SIZE;
637 
638 		/*
639 		 * Allocate the circular buffer of Read Buffer Descriptors
640 		 * (RBDs)
641 		 */
642 		rxq->bd = dma_zalloc_coherent(dev,
643 					     free_size * rxq->queue_size,
644 					     &rxq->bd_dma, GFP_KERNEL);
645 		if (!rxq->bd)
646 			goto err;
647 
648 		if (trans->cfg->mq_rx_supported) {
649 			rxq->used_bd = dma_zalloc_coherent(dev,
650 							   sizeof(__le32) *
651 							   rxq->queue_size,
652 							   &rxq->used_bd_dma,
653 							   GFP_KERNEL);
654 			if (!rxq->used_bd)
655 				goto err;
656 		}
657 
658 		/*Allocate the driver's pointer to receive buffer status */
659 		rxq->rb_stts = dma_zalloc_coherent(dev, sizeof(*rxq->rb_stts),
660 						   &rxq->rb_stts_dma,
661 						   GFP_KERNEL);
662 		if (!rxq->rb_stts)
663 			goto err;
664 	}
665 	return 0;
666 
667 err:
668 	for (i = 0; i < trans->num_rx_queues; i++) {
669 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
670 
671 		if (rxq->bd)
672 			dma_free_coherent(dev, free_size * rxq->queue_size,
673 					  rxq->bd, rxq->bd_dma);
674 		rxq->bd_dma = 0;
675 		rxq->bd = NULL;
676 
677 		if (rxq->rb_stts)
678 			dma_free_coherent(trans->dev,
679 					  sizeof(struct iwl_rb_status),
680 					  rxq->rb_stts, rxq->rb_stts_dma);
681 
682 		if (rxq->used_bd)
683 			dma_free_coherent(dev, sizeof(__le32) * rxq->queue_size,
684 					  rxq->used_bd, rxq->used_bd_dma);
685 		rxq->used_bd_dma = 0;
686 		rxq->used_bd = NULL;
687 	}
688 	kfree(trans_pcie->rxq);
689 
690 	return -ENOMEM;
691 }
692 
693 static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
694 {
695 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
696 	u32 rb_size;
697 	unsigned long flags;
698 	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
699 
700 	switch (trans_pcie->rx_buf_size) {
701 	case IWL_AMSDU_4K:
702 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
703 		break;
704 	case IWL_AMSDU_8K:
705 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
706 		break;
707 	case IWL_AMSDU_12K:
708 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_12K;
709 		break;
710 	default:
711 		WARN_ON(1);
712 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
713 	}
714 
715 	if (!iwl_trans_grab_nic_access(trans, &flags))
716 		return;
717 
718 	/* Stop Rx DMA */
719 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
720 	/* reset and flush pointers */
721 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
722 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
723 	iwl_write32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
724 
725 	/* Reset driver's Rx queue write index */
726 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
727 
728 	/* Tell device where to find RBD circular buffer in DRAM */
729 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
730 		    (u32)(rxq->bd_dma >> 8));
731 
732 	/* Tell device where in DRAM to update its Rx status */
733 	iwl_write32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
734 		    rxq->rb_stts_dma >> 4);
735 
736 	/* Enable Rx DMA
737 	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
738 	 *      the credit mechanism in 5000 HW RX FIFO
739 	 * Direct rx interrupts to hosts
740 	 * Rx buffer size 4 or 8k or 12k
741 	 * RB timeout 0x10
742 	 * 256 RBDs
743 	 */
744 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
745 		    FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
746 		    FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
747 		    FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
748 		    rb_size |
749 		    (RX_RB_TIMEOUT << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS) |
750 		    (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));
751 
752 	iwl_trans_release_nic_access(trans, &flags);
753 
754 	/* Set interrupt coalescing timer to default (2048 usecs) */
755 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
756 
757 	/* W/A for interrupt coalescing bug in 7260 and 3160 */
758 	if (trans->cfg->host_interrupt_operation_mode)
759 		iwl_set_bit(trans, CSR_INT_COALESCING, IWL_HOST_INT_OPER_MODE);
760 }
761 
762 void iwl_pcie_enable_rx_wake(struct iwl_trans *trans, bool enable)
763 {
764 	if (trans->cfg->device_family != IWL_DEVICE_FAMILY_9000)
765 		return;
766 
767 	if (CSR_HW_REV_STEP(trans->hw_rev) != SILICON_A_STEP)
768 		return;
769 
770 	if (!trans->cfg->integrated)
771 		return;
772 
773 	/*
774 	 * Turn on the chicken-bits that cause MAC wakeup for RX-related
775 	 * values.
776 	 * This costs some power, but needed for W/A 9000 integrated A-step
777 	 * bug where shadow registers are not in the retention list and their
778 	 * value is lost when NIC powers down
779 	 */
780 	iwl_set_bit(trans, CSR_MAC_SHADOW_REG_CTRL,
781 		    CSR_MAC_SHADOW_REG_CTRL_RX_WAKE);
782 	iwl_set_bit(trans, CSR_MAC_SHADOW_REG_CTL2,
783 		    CSR_MAC_SHADOW_REG_CTL2_RX_WAKE);
784 }
785 
786 static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
787 {
788 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
789 	u32 rb_size, enabled = 0;
790 	unsigned long flags;
791 	int i;
792 
793 	switch (trans_pcie->rx_buf_size) {
794 	case IWL_AMSDU_4K:
795 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
796 		break;
797 	case IWL_AMSDU_8K:
798 		rb_size = RFH_RXF_DMA_RB_SIZE_8K;
799 		break;
800 	case IWL_AMSDU_12K:
801 		rb_size = RFH_RXF_DMA_RB_SIZE_12K;
802 		break;
803 	default:
804 		WARN_ON(1);
805 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
806 	}
807 
808 	if (!iwl_trans_grab_nic_access(trans, &flags))
809 		return;
810 
811 	/* Stop Rx DMA */
812 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG, 0);
813 	/* disable free amd used rx queue operation */
814 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, 0);
815 
816 	for (i = 0; i < trans->num_rx_queues; i++) {
817 		/* Tell device where to find RBD free table in DRAM */
818 		iwl_write_prph64_no_grab(trans,
819 					 RFH_Q_FRBDCB_BA_LSB(i),
820 					 trans_pcie->rxq[i].bd_dma);
821 		/* Tell device where to find RBD used table in DRAM */
822 		iwl_write_prph64_no_grab(trans,
823 					 RFH_Q_URBDCB_BA_LSB(i),
824 					 trans_pcie->rxq[i].used_bd_dma);
825 		/* Tell device where in DRAM to update its Rx status */
826 		iwl_write_prph64_no_grab(trans,
827 					 RFH_Q_URBD_STTS_WPTR_LSB(i),
828 					 trans_pcie->rxq[i].rb_stts_dma);
829 		/* Reset device indice tables */
830 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_WIDX(i), 0);
831 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_RIDX(i), 0);
832 		iwl_write_prph_no_grab(trans, RFH_Q_URBDCB_WIDX(i), 0);
833 
834 		enabled |= BIT(i) | BIT(i + 16);
835 	}
836 
837 	/*
838 	 * Enable Rx DMA
839 	 * Rx buffer size 4 or 8k or 12k
840 	 * Min RB size 4 or 8
841 	 * Drop frames that exceed RB size
842 	 * 512 RBDs
843 	 */
844 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG,
845 			       RFH_DMA_EN_ENABLE_VAL | rb_size |
846 			       RFH_RXF_DMA_MIN_RB_4_8 |
847 			       RFH_RXF_DMA_DROP_TOO_LARGE_MASK |
848 			       RFH_RXF_DMA_RBDCB_SIZE_512);
849 
850 	/*
851 	 * Activate DMA snooping.
852 	 * Set RX DMA chunk size to 64B for IOSF and 128B for PCIe
853 	 * Default queue is 0
854 	 */
855 	iwl_write_prph_no_grab(trans, RFH_GEN_CFG,
856 			       RFH_GEN_CFG_RFH_DMA_SNOOP |
857 			       RFH_GEN_CFG_VAL(DEFAULT_RXQ_NUM, 0) |
858 			       RFH_GEN_CFG_SERVICE_DMA_SNOOP |
859 			       RFH_GEN_CFG_VAL(RB_CHUNK_SIZE,
860 					       trans->cfg->integrated ?
861 					       RFH_GEN_CFG_RB_CHUNK_SIZE_64 :
862 					       RFH_GEN_CFG_RB_CHUNK_SIZE_128));
863 	/* Enable the relevant rx queues */
864 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, enabled);
865 
866 	iwl_trans_release_nic_access(trans, &flags);
867 
868 	/* Set interrupt coalescing timer to default (2048 usecs) */
869 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
870 
871 	iwl_pcie_enable_rx_wake(trans, true);
872 }
873 
874 static void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq)
875 {
876 	lockdep_assert_held(&rxq->lock);
877 
878 	INIT_LIST_HEAD(&rxq->rx_free);
879 	INIT_LIST_HEAD(&rxq->rx_used);
880 	rxq->free_count = 0;
881 	rxq->used_count = 0;
882 }
883 
884 static int iwl_pcie_dummy_napi_poll(struct napi_struct *napi, int budget)
885 {
886 	WARN_ON(1);
887 	return 0;
888 }
889 
890 static int _iwl_pcie_rx_init(struct iwl_trans *trans)
891 {
892 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
893 	struct iwl_rxq *def_rxq;
894 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
895 	int i, err, queue_size, allocator_pool_size, num_alloc;
896 
897 	if (!trans_pcie->rxq) {
898 		err = iwl_pcie_rx_alloc(trans);
899 		if (err)
900 			return err;
901 	}
902 	def_rxq = trans_pcie->rxq;
903 
904 	spin_lock(&rba->lock);
905 	atomic_set(&rba->req_pending, 0);
906 	atomic_set(&rba->req_ready, 0);
907 	INIT_LIST_HEAD(&rba->rbd_allocated);
908 	INIT_LIST_HEAD(&rba->rbd_empty);
909 	spin_unlock(&rba->lock);
910 
911 	/* free all first - we might be reconfigured for a different size */
912 	iwl_pcie_free_rbs_pool(trans);
913 
914 	for (i = 0; i < RX_QUEUE_SIZE; i++)
915 		def_rxq->queue[i] = NULL;
916 
917 	for (i = 0; i < trans->num_rx_queues; i++) {
918 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
919 
920 		rxq->id = i;
921 
922 		spin_lock(&rxq->lock);
923 		/*
924 		 * Set read write pointer to reflect that we have processed
925 		 * and used all buffers, but have not restocked the Rx queue
926 		 * with fresh buffers
927 		 */
928 		rxq->read = 0;
929 		rxq->write = 0;
930 		rxq->write_actual = 0;
931 		memset(rxq->rb_stts, 0, sizeof(*rxq->rb_stts));
932 
933 		iwl_pcie_rx_init_rxb_lists(rxq);
934 
935 		if (!rxq->napi.poll)
936 			netif_napi_add(&trans_pcie->napi_dev, &rxq->napi,
937 				       iwl_pcie_dummy_napi_poll, 64);
938 
939 		spin_unlock(&rxq->lock);
940 	}
941 
942 	/* move the pool to the default queue and allocator ownerships */
943 	queue_size = trans->cfg->mq_rx_supported ?
944 		     MQ_RX_NUM_RBDS : RX_QUEUE_SIZE;
945 	allocator_pool_size = trans->num_rx_queues *
946 		(RX_CLAIM_REQ_ALLOC - RX_POST_REQ_ALLOC);
947 	num_alloc = queue_size + allocator_pool_size;
948 	BUILD_BUG_ON(ARRAY_SIZE(trans_pcie->global_table) !=
949 		     ARRAY_SIZE(trans_pcie->rx_pool));
950 	for (i = 0; i < num_alloc; i++) {
951 		struct iwl_rx_mem_buffer *rxb = &trans_pcie->rx_pool[i];
952 
953 		if (i < allocator_pool_size)
954 			list_add(&rxb->list, &rba->rbd_empty);
955 		else
956 			list_add(&rxb->list, &def_rxq->rx_used);
957 		trans_pcie->global_table[i] = rxb;
958 		rxb->vid = (u16)(i + 1);
959 		rxb->invalid = true;
960 	}
961 
962 	iwl_pcie_rxq_alloc_rbs(trans, GFP_KERNEL, def_rxq);
963 
964 	return 0;
965 }
966 
967 int iwl_pcie_rx_init(struct iwl_trans *trans)
968 {
969 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
970 	int ret = _iwl_pcie_rx_init(trans);
971 
972 	if (ret)
973 		return ret;
974 
975 	if (trans->cfg->mq_rx_supported)
976 		iwl_pcie_rx_mq_hw_init(trans);
977 	else
978 		iwl_pcie_rx_hw_init(trans, trans_pcie->rxq);
979 
980 	iwl_pcie_rxq_restock(trans, trans_pcie->rxq);
981 
982 	spin_lock(&trans_pcie->rxq->lock);
983 	iwl_pcie_rxq_inc_wr_ptr(trans, trans_pcie->rxq);
984 	spin_unlock(&trans_pcie->rxq->lock);
985 
986 	return 0;
987 }
988 
989 int iwl_pcie_gen2_rx_init(struct iwl_trans *trans)
990 {
991 	/*
992 	 * We don't configure the RFH.
993 	 * Restock will be done at alive, after firmware configured the RFH.
994 	 */
995 	return _iwl_pcie_rx_init(trans);
996 }
997 
998 void iwl_pcie_rx_free(struct iwl_trans *trans)
999 {
1000 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1001 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1002 	int free_size = trans->cfg->mq_rx_supported ? sizeof(__le64) :
1003 					      sizeof(__le32);
1004 	int i;
1005 
1006 	/*
1007 	 * if rxq is NULL, it means that nothing has been allocated,
1008 	 * exit now
1009 	 */
1010 	if (!trans_pcie->rxq) {
1011 		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
1012 		return;
1013 	}
1014 
1015 	cancel_work_sync(&rba->rx_alloc);
1016 
1017 	iwl_pcie_free_rbs_pool(trans);
1018 
1019 	for (i = 0; i < trans->num_rx_queues; i++) {
1020 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1021 
1022 		if (rxq->bd)
1023 			dma_free_coherent(trans->dev,
1024 					  free_size * rxq->queue_size,
1025 					  rxq->bd, rxq->bd_dma);
1026 		rxq->bd_dma = 0;
1027 		rxq->bd = NULL;
1028 
1029 		if (rxq->rb_stts)
1030 			dma_free_coherent(trans->dev,
1031 					  sizeof(struct iwl_rb_status),
1032 					  rxq->rb_stts, rxq->rb_stts_dma);
1033 		else
1034 			IWL_DEBUG_INFO(trans,
1035 				       "Free rxq->rb_stts which is NULL\n");
1036 
1037 		if (rxq->used_bd)
1038 			dma_free_coherent(trans->dev,
1039 					  sizeof(__le32) * rxq->queue_size,
1040 					  rxq->used_bd, rxq->used_bd_dma);
1041 		rxq->used_bd_dma = 0;
1042 		rxq->used_bd = NULL;
1043 
1044 		if (rxq->napi.poll)
1045 			netif_napi_del(&rxq->napi);
1046 	}
1047 	kfree(trans_pcie->rxq);
1048 }
1049 
1050 /*
1051  * iwl_pcie_rx_reuse_rbd - Recycle used RBDs
1052  *
1053  * Called when a RBD can be reused. The RBD is transferred to the allocator.
1054  * When there are 2 empty RBDs - a request for allocation is posted
1055  */
1056 static void iwl_pcie_rx_reuse_rbd(struct iwl_trans *trans,
1057 				  struct iwl_rx_mem_buffer *rxb,
1058 				  struct iwl_rxq *rxq, bool emergency)
1059 {
1060 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1061 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1062 
1063 	/* Move the RBD to the used list, will be moved to allocator in batches
1064 	 * before claiming or posting a request*/
1065 	list_add_tail(&rxb->list, &rxq->rx_used);
1066 
1067 	if (unlikely(emergency))
1068 		return;
1069 
1070 	/* Count the allocator owned RBDs */
1071 	rxq->used_count++;
1072 
1073 	/* If we have RX_POST_REQ_ALLOC new released rx buffers -
1074 	 * issue a request for allocator. Modulo RX_CLAIM_REQ_ALLOC is
1075 	 * used for the case we failed to claim RX_CLAIM_REQ_ALLOC,
1076 	 * after but we still need to post another request.
1077 	 */
1078 	if ((rxq->used_count % RX_CLAIM_REQ_ALLOC) == RX_POST_REQ_ALLOC) {
1079 		/* Move the 2 RBDs to the allocator ownership.
1080 		 Allocator has another 6 from pool for the request completion*/
1081 		spin_lock(&rba->lock);
1082 		list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
1083 		spin_unlock(&rba->lock);
1084 
1085 		atomic_inc(&rba->req_pending);
1086 		queue_work(rba->alloc_wq, &rba->rx_alloc);
1087 	}
1088 }
1089 
1090 static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
1091 				struct iwl_rxq *rxq,
1092 				struct iwl_rx_mem_buffer *rxb,
1093 				bool emergency)
1094 {
1095 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1096 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1097 	bool page_stolen = false;
1098 	int max_len = PAGE_SIZE << trans_pcie->rx_page_order;
1099 	u32 offset = 0;
1100 
1101 	if (WARN_ON(!rxb))
1102 		return;
1103 
1104 	dma_unmap_page(trans->dev, rxb->page_dma, max_len, DMA_FROM_DEVICE);
1105 
1106 	while (offset + sizeof(u32) + sizeof(struct iwl_cmd_header) < max_len) {
1107 		struct iwl_rx_packet *pkt;
1108 		u16 sequence;
1109 		bool reclaim;
1110 		int index, cmd_index, len;
1111 		struct iwl_rx_cmd_buffer rxcb = {
1112 			._offset = offset,
1113 			._rx_page_order = trans_pcie->rx_page_order,
1114 			._page = rxb->page,
1115 			._page_stolen = false,
1116 			.truesize = max_len,
1117 		};
1118 
1119 		pkt = rxb_addr(&rxcb);
1120 
1121 		if (pkt->len_n_flags == cpu_to_le32(FH_RSCSR_FRAME_INVALID)) {
1122 			IWL_DEBUG_RX(trans,
1123 				     "Q %d: RB end marker at offset %d\n",
1124 				     rxq->id, offset);
1125 			break;
1126 		}
1127 
1128 		WARN((le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1129 			FH_RSCSR_RXQ_POS != rxq->id,
1130 		     "frame on invalid queue - is on %d and indicates %d\n",
1131 		     rxq->id,
1132 		     (le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1133 			FH_RSCSR_RXQ_POS);
1134 
1135 		IWL_DEBUG_RX(trans,
1136 			     "Q %d: cmd at offset %d: %s (%.2x.%2x, seq 0x%x)\n",
1137 			     rxq->id, offset,
1138 			     iwl_get_cmd_string(trans,
1139 						iwl_cmd_id(pkt->hdr.cmd,
1140 							   pkt->hdr.group_id,
1141 							   0)),
1142 			     pkt->hdr.group_id, pkt->hdr.cmd,
1143 			     le16_to_cpu(pkt->hdr.sequence));
1144 
1145 		len = iwl_rx_packet_len(pkt);
1146 		len += sizeof(u32); /* account for status word */
1147 		trace_iwlwifi_dev_rx(trans->dev, trans, pkt, len);
1148 		trace_iwlwifi_dev_rx_data(trans->dev, trans, pkt, len);
1149 
1150 		/* Reclaim a command buffer only if this packet is a response
1151 		 *   to a (driver-originated) command.
1152 		 * If the packet (e.g. Rx frame) originated from uCode,
1153 		 *   there is no command buffer to reclaim.
1154 		 * Ucode should set SEQ_RX_FRAME bit if ucode-originated,
1155 		 *   but apparently a few don't get set; catch them here. */
1156 		reclaim = !(pkt->hdr.sequence & SEQ_RX_FRAME);
1157 		if (reclaim && !pkt->hdr.group_id) {
1158 			int i;
1159 
1160 			for (i = 0; i < trans_pcie->n_no_reclaim_cmds; i++) {
1161 				if (trans_pcie->no_reclaim_cmds[i] ==
1162 							pkt->hdr.cmd) {
1163 					reclaim = false;
1164 					break;
1165 				}
1166 			}
1167 		}
1168 
1169 		sequence = le16_to_cpu(pkt->hdr.sequence);
1170 		index = SEQ_TO_INDEX(sequence);
1171 		cmd_index = iwl_pcie_get_cmd_index(txq, index);
1172 
1173 		if (rxq->id == 0)
1174 			iwl_op_mode_rx(trans->op_mode, &rxq->napi,
1175 				       &rxcb);
1176 		else
1177 			iwl_op_mode_rx_rss(trans->op_mode, &rxq->napi,
1178 					   &rxcb, rxq->id);
1179 
1180 		if (reclaim) {
1181 			kzfree(txq->entries[cmd_index].free_buf);
1182 			txq->entries[cmd_index].free_buf = NULL;
1183 		}
1184 
1185 		/*
1186 		 * After here, we should always check rxcb._page_stolen,
1187 		 * if it is true then one of the handlers took the page.
1188 		 */
1189 
1190 		if (reclaim) {
1191 			/* Invoke any callbacks, transfer the buffer to caller,
1192 			 * and fire off the (possibly) blocking
1193 			 * iwl_trans_send_cmd()
1194 			 * as we reclaim the driver command queue */
1195 			if (!rxcb._page_stolen)
1196 				iwl_pcie_hcmd_complete(trans, &rxcb);
1197 			else
1198 				IWL_WARN(trans, "Claim null rxb?\n");
1199 		}
1200 
1201 		page_stolen |= rxcb._page_stolen;
1202 		offset += ALIGN(len, FH_RSCSR_FRAME_ALIGN);
1203 	}
1204 
1205 	/* page was stolen from us -- free our reference */
1206 	if (page_stolen) {
1207 		__free_pages(rxb->page, trans_pcie->rx_page_order);
1208 		rxb->page = NULL;
1209 	}
1210 
1211 	/* Reuse the page if possible. For notification packets and
1212 	 * SKBs that fail to Rx correctly, add them back into the
1213 	 * rx_free list for reuse later. */
1214 	if (rxb->page != NULL) {
1215 		rxb->page_dma =
1216 			dma_map_page(trans->dev, rxb->page, 0,
1217 				     PAGE_SIZE << trans_pcie->rx_page_order,
1218 				     DMA_FROM_DEVICE);
1219 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
1220 			/*
1221 			 * free the page(s) as well to not break
1222 			 * the invariant that the items on the used
1223 			 * list have no page(s)
1224 			 */
1225 			__free_pages(rxb->page, trans_pcie->rx_page_order);
1226 			rxb->page = NULL;
1227 			iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1228 		} else {
1229 			list_add_tail(&rxb->list, &rxq->rx_free);
1230 			rxq->free_count++;
1231 		}
1232 	} else
1233 		iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1234 }
1235 
1236 /*
1237  * iwl_pcie_rx_handle - Main entry function for receiving responses from fw
1238  */
1239 static void iwl_pcie_rx_handle(struct iwl_trans *trans, int queue)
1240 {
1241 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1242 	struct iwl_rxq *rxq = &trans_pcie->rxq[queue];
1243 	u32 r, i, count = 0;
1244 	bool emergency = false;
1245 
1246 restart:
1247 	spin_lock(&rxq->lock);
1248 	/* uCode's read index (stored in shared DRAM) indicates the last Rx
1249 	 * buffer that the driver may process (last buffer filled by ucode). */
1250 	r = le16_to_cpu(READ_ONCE(rxq->rb_stts->closed_rb_num)) & 0x0FFF;
1251 	i = rxq->read;
1252 
1253 	/* W/A 9000 device step A0 wrap-around bug */
1254 	r &= (rxq->queue_size - 1);
1255 
1256 	/* Rx interrupt, but nothing sent from uCode */
1257 	if (i == r)
1258 		IWL_DEBUG_RX(trans, "Q %d: HW = SW = %d\n", rxq->id, r);
1259 
1260 	while (i != r) {
1261 		struct iwl_rx_mem_buffer *rxb;
1262 
1263 		if (unlikely(rxq->used_count == rxq->queue_size / 2))
1264 			emergency = true;
1265 
1266 		if (trans->cfg->mq_rx_supported) {
1267 			/*
1268 			 * used_bd is a 32 bit but only 12 are used to retrieve
1269 			 * the vid
1270 			 */
1271 			u16 vid = le32_to_cpu(rxq->used_bd[i]) & 0x0FFF;
1272 
1273 			if (WARN(!vid ||
1274 				 vid > ARRAY_SIZE(trans_pcie->global_table),
1275 				 "Invalid rxb index from HW %u\n", (u32)vid)) {
1276 				iwl_force_nmi(trans);
1277 				goto out;
1278 			}
1279 			rxb = trans_pcie->global_table[vid - 1];
1280 			if (WARN(rxb->invalid,
1281 				 "Invalid rxb from HW %u\n", (u32)vid)) {
1282 				iwl_force_nmi(trans);
1283 				goto out;
1284 			}
1285 			rxb->invalid = true;
1286 		} else {
1287 			rxb = rxq->queue[i];
1288 			rxq->queue[i] = NULL;
1289 		}
1290 
1291 		IWL_DEBUG_RX(trans, "Q %d: HW = %d, SW = %d\n", rxq->id, r, i);
1292 		iwl_pcie_rx_handle_rb(trans, rxq, rxb, emergency);
1293 
1294 		i = (i + 1) & (rxq->queue_size - 1);
1295 
1296 		/*
1297 		 * If we have RX_CLAIM_REQ_ALLOC released rx buffers -
1298 		 * try to claim the pre-allocated buffers from the allocator.
1299 		 * If not ready - will try to reclaim next time.
1300 		 * There is no need to reschedule work - allocator exits only
1301 		 * on success
1302 		 */
1303 		if (rxq->used_count >= RX_CLAIM_REQ_ALLOC)
1304 			iwl_pcie_rx_allocator_get(trans, rxq);
1305 
1306 		if (rxq->used_count % RX_CLAIM_REQ_ALLOC == 0 && !emergency) {
1307 			struct iwl_rb_allocator *rba = &trans_pcie->rba;
1308 
1309 			/* Add the remaining empty RBDs for allocator use */
1310 			spin_lock(&rba->lock);
1311 			list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
1312 			spin_unlock(&rba->lock);
1313 		} else if (emergency) {
1314 			count++;
1315 			if (count == 8) {
1316 				count = 0;
1317 				if (rxq->used_count < rxq->queue_size / 3)
1318 					emergency = false;
1319 
1320 				rxq->read = i;
1321 				spin_unlock(&rxq->lock);
1322 				iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1323 				iwl_pcie_rxq_restock(trans, rxq);
1324 				goto restart;
1325 			}
1326 		}
1327 	}
1328 out:
1329 	/* Backtrack one entry */
1330 	rxq->read = i;
1331 	spin_unlock(&rxq->lock);
1332 
1333 	/*
1334 	 * handle a case where in emergency there are some unallocated RBDs.
1335 	 * those RBDs are in the used list, but are not tracked by the queue's
1336 	 * used_count which counts allocator owned RBDs.
1337 	 * unallocated emergency RBDs must be allocated on exit, otherwise
1338 	 * when called again the function may not be in emergency mode and
1339 	 * they will be handed to the allocator with no tracking in the RBD
1340 	 * allocator counters, which will lead to them never being claimed back
1341 	 * by the queue.
1342 	 * by allocating them here, they are now in the queue free list, and
1343 	 * will be restocked by the next call of iwl_pcie_rxq_restock.
1344 	 */
1345 	if (unlikely(emergency && count))
1346 		iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1347 
1348 	if (rxq->napi.poll)
1349 		napi_gro_flush(&rxq->napi, false);
1350 
1351 	iwl_pcie_rxq_restock(trans, rxq);
1352 }
1353 
1354 static struct iwl_trans_pcie *iwl_pcie_get_trans_pcie(struct msix_entry *entry)
1355 {
1356 	u8 queue = entry->entry;
1357 	struct msix_entry *entries = entry - queue;
1358 
1359 	return container_of(entries, struct iwl_trans_pcie, msix_entries[0]);
1360 }
1361 
1362 static inline void iwl_pcie_clear_irq(struct iwl_trans *trans,
1363 				      struct msix_entry *entry)
1364 {
1365 	/*
1366 	 * Before sending the interrupt the HW disables it to prevent
1367 	 * a nested interrupt. This is done by writing 1 to the corresponding
1368 	 * bit in the mask register. After handling the interrupt, it should be
1369 	 * re-enabled by clearing this bit. This register is defined as
1370 	 * write 1 clear (W1C) register, meaning that it's being clear
1371 	 * by writing 1 to the bit.
1372 	 */
1373 	iwl_write32(trans, CSR_MSIX_AUTOMASK_ST_AD, BIT(entry->entry));
1374 }
1375 
1376 /*
1377  * iwl_pcie_rx_msix_handle - Main entry function for receiving responses from fw
1378  * This interrupt handler should be used with RSS queue only.
1379  */
1380 irqreturn_t iwl_pcie_irq_rx_msix_handler(int irq, void *dev_id)
1381 {
1382 	struct msix_entry *entry = dev_id;
1383 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
1384 	struct iwl_trans *trans = trans_pcie->trans;
1385 
1386 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, false, 0, 0);
1387 
1388 	if (WARN_ON(entry->entry >= trans->num_rx_queues))
1389 		return IRQ_NONE;
1390 
1391 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1392 
1393 	local_bh_disable();
1394 	iwl_pcie_rx_handle(trans, entry->entry);
1395 	local_bh_enable();
1396 
1397 	iwl_pcie_clear_irq(trans, entry);
1398 
1399 	lock_map_release(&trans->sync_cmd_lockdep_map);
1400 
1401 	return IRQ_HANDLED;
1402 }
1403 
1404 /*
1405  * iwl_pcie_irq_handle_error - called for HW or SW error interrupt from card
1406  */
1407 static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
1408 {
1409 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1410 	int i;
1411 
1412 	/* W/A for WiFi/WiMAX coex and WiMAX own the RF */
1413 	if (trans->cfg->internal_wimax_coex &&
1414 	    !trans->cfg->apmg_not_supported &&
1415 	    (!(iwl_read_prph(trans, APMG_CLK_CTRL_REG) &
1416 			     APMS_CLK_VAL_MRB_FUNC_MODE) ||
1417 	     (iwl_read_prph(trans, APMG_PS_CTRL_REG) &
1418 			    APMG_PS_CTRL_VAL_RESET_REQ))) {
1419 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1420 		iwl_op_mode_wimax_active(trans->op_mode);
1421 		wake_up(&trans_pcie->wait_command_queue);
1422 		return;
1423 	}
1424 
1425 	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
1426 		if (!trans_pcie->txq[i])
1427 			continue;
1428 		del_timer(&trans_pcie->txq[i]->stuck_timer);
1429 	}
1430 
1431 	/* The STATUS_FW_ERROR bit is set in this function. This must happen
1432 	 * before we wake up the command caller, to ensure a proper cleanup. */
1433 	iwl_trans_fw_error(trans);
1434 
1435 	clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1436 	wake_up(&trans_pcie->wait_command_queue);
1437 }
1438 
1439 static u32 iwl_pcie_int_cause_non_ict(struct iwl_trans *trans)
1440 {
1441 	u32 inta;
1442 
1443 	lockdep_assert_held(&IWL_TRANS_GET_PCIE_TRANS(trans)->irq_lock);
1444 
1445 	trace_iwlwifi_dev_irq(trans->dev);
1446 
1447 	/* Discover which interrupts are active/pending */
1448 	inta = iwl_read32(trans, CSR_INT);
1449 
1450 	/* the thread will service interrupts and re-enable them */
1451 	return inta;
1452 }
1453 
1454 /* a device (PCI-E) page is 4096 bytes long */
1455 #define ICT_SHIFT	12
1456 #define ICT_SIZE	(1 << ICT_SHIFT)
1457 #define ICT_COUNT	(ICT_SIZE / sizeof(u32))
1458 
1459 /* interrupt handler using ict table, with this interrupt driver will
1460  * stop using INTA register to get device's interrupt, reading this register
1461  * is expensive, device will write interrupts in ICT dram table, increment
1462  * index then will fire interrupt to driver, driver will OR all ICT table
1463  * entries from current index up to table entry with 0 value. the result is
1464  * the interrupt we need to service, driver will set the entries back to 0 and
1465  * set index.
1466  */
1467 static u32 iwl_pcie_int_cause_ict(struct iwl_trans *trans)
1468 {
1469 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1470 	u32 inta;
1471 	u32 val = 0;
1472 	u32 read;
1473 
1474 	trace_iwlwifi_dev_irq(trans->dev);
1475 
1476 	/* Ignore interrupt if there's nothing in NIC to service.
1477 	 * This may be due to IRQ shared with another device,
1478 	 * or due to sporadic interrupts thrown from our NIC. */
1479 	read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1480 	trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index, read);
1481 	if (!read)
1482 		return 0;
1483 
1484 	/*
1485 	 * Collect all entries up to the first 0, starting from ict_index;
1486 	 * note we already read at ict_index.
1487 	 */
1488 	do {
1489 		val |= read;
1490 		IWL_DEBUG_ISR(trans, "ICT index %d value 0x%08X\n",
1491 				trans_pcie->ict_index, read);
1492 		trans_pcie->ict_tbl[trans_pcie->ict_index] = 0;
1493 		trans_pcie->ict_index =
1494 			((trans_pcie->ict_index + 1) & (ICT_COUNT - 1));
1495 
1496 		read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1497 		trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index,
1498 					   read);
1499 	} while (read);
1500 
1501 	/* We should not get this value, just ignore it. */
1502 	if (val == 0xffffffff)
1503 		val = 0;
1504 
1505 	/*
1506 	 * this is a w/a for a h/w bug. the h/w bug may cause the Rx bit
1507 	 * (bit 15 before shifting it to 31) to clear when using interrupt
1508 	 * coalescing. fortunately, bits 18 and 19 stay set when this happens
1509 	 * so we use them to decide on the real state of the Rx bit.
1510 	 * In order words, bit 15 is set if bit 18 or bit 19 are set.
1511 	 */
1512 	if (val & 0xC0000)
1513 		val |= 0x8000;
1514 
1515 	inta = (0xff & val) | ((0xff00 & val) << 16);
1516 	return inta;
1517 }
1518 
1519 void iwl_pcie_handle_rfkill_irq(struct iwl_trans *trans)
1520 {
1521 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1522 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1523 	bool hw_rfkill, prev, report;
1524 
1525 	mutex_lock(&trans_pcie->mutex);
1526 	prev = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1527 	hw_rfkill = iwl_is_rfkill_set(trans);
1528 	if (hw_rfkill) {
1529 		set_bit(STATUS_RFKILL_OPMODE, &trans->status);
1530 		set_bit(STATUS_RFKILL_HW, &trans->status);
1531 	}
1532 	if (trans_pcie->opmode_down)
1533 		report = hw_rfkill;
1534 	else
1535 		report = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1536 
1537 	IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
1538 		 hw_rfkill ? "disable radio" : "enable radio");
1539 
1540 	isr_stats->rfkill++;
1541 
1542 	if (prev != report)
1543 		iwl_trans_pcie_rf_kill(trans, report);
1544 	mutex_unlock(&trans_pcie->mutex);
1545 
1546 	if (hw_rfkill) {
1547 		if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
1548 				       &trans->status))
1549 			IWL_DEBUG_RF_KILL(trans,
1550 					  "Rfkill while SYNC HCMD in flight\n");
1551 		wake_up(&trans_pcie->wait_command_queue);
1552 	} else {
1553 		clear_bit(STATUS_RFKILL_HW, &trans->status);
1554 		if (trans_pcie->opmode_down)
1555 			clear_bit(STATUS_RFKILL_OPMODE, &trans->status);
1556 	}
1557 }
1558 
1559 irqreturn_t iwl_pcie_irq_handler(int irq, void *dev_id)
1560 {
1561 	struct iwl_trans *trans = dev_id;
1562 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1563 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1564 	u32 inta = 0;
1565 	u32 handled = 0;
1566 
1567 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1568 
1569 	spin_lock(&trans_pcie->irq_lock);
1570 
1571 	/* dram interrupt table not set yet,
1572 	 * use legacy interrupt.
1573 	 */
1574 	if (likely(trans_pcie->use_ict))
1575 		inta = iwl_pcie_int_cause_ict(trans);
1576 	else
1577 		inta = iwl_pcie_int_cause_non_ict(trans);
1578 
1579 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1580 		IWL_DEBUG_ISR(trans,
1581 			      "ISR inta 0x%08x, enabled 0x%08x(sw), enabled(hw) 0x%08x, fh 0x%08x\n",
1582 			      inta, trans_pcie->inta_mask,
1583 			      iwl_read32(trans, CSR_INT_MASK),
1584 			      iwl_read32(trans, CSR_FH_INT_STATUS));
1585 		if (inta & (~trans_pcie->inta_mask))
1586 			IWL_DEBUG_ISR(trans,
1587 				      "We got a masked interrupt (0x%08x)\n",
1588 				      inta & (~trans_pcie->inta_mask));
1589 	}
1590 
1591 	inta &= trans_pcie->inta_mask;
1592 
1593 	/*
1594 	 * Ignore interrupt if there's nothing in NIC to service.
1595 	 * This may be due to IRQ shared with another device,
1596 	 * or due to sporadic interrupts thrown from our NIC.
1597 	 */
1598 	if (unlikely(!inta)) {
1599 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
1600 		/*
1601 		 * Re-enable interrupts here since we don't
1602 		 * have anything to service
1603 		 */
1604 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1605 			_iwl_enable_interrupts(trans);
1606 		spin_unlock(&trans_pcie->irq_lock);
1607 		lock_map_release(&trans->sync_cmd_lockdep_map);
1608 		return IRQ_NONE;
1609 	}
1610 
1611 	if (unlikely(inta == 0xFFFFFFFF || (inta & 0xFFFFFFF0) == 0xa5a5a5a0)) {
1612 		/*
1613 		 * Hardware disappeared. It might have
1614 		 * already raised an interrupt.
1615 		 */
1616 		IWL_WARN(trans, "HARDWARE GONE?? INTA == 0x%08x\n", inta);
1617 		spin_unlock(&trans_pcie->irq_lock);
1618 		goto out;
1619 	}
1620 
1621 	/* Ack/clear/reset pending uCode interrupts.
1622 	 * Note:  Some bits in CSR_INT are "OR" of bits in CSR_FH_INT_STATUS,
1623 	 */
1624 	/* There is a hardware bug in the interrupt mask function that some
1625 	 * interrupts (i.e. CSR_INT_BIT_SCD) can still be generated even if
1626 	 * they are disabled in the CSR_INT_MASK register. Furthermore the
1627 	 * ICT interrupt handling mechanism has another bug that might cause
1628 	 * these unmasked interrupts fail to be detected. We workaround the
1629 	 * hardware bugs here by ACKing all the possible interrupts so that
1630 	 * interrupt coalescing can still be achieved.
1631 	 */
1632 	iwl_write32(trans, CSR_INT, inta | ~trans_pcie->inta_mask);
1633 
1634 	if (iwl_have_debug_level(IWL_DL_ISR))
1635 		IWL_DEBUG_ISR(trans, "inta 0x%08x, enabled 0x%08x\n",
1636 			      inta, iwl_read32(trans, CSR_INT_MASK));
1637 
1638 	spin_unlock(&trans_pcie->irq_lock);
1639 
1640 	/* Now service all interrupt bits discovered above. */
1641 	if (inta & CSR_INT_BIT_HW_ERR) {
1642 		IWL_ERR(trans, "Hardware error detected.  Restarting.\n");
1643 
1644 		/* Tell the device to stop sending interrupts */
1645 		iwl_disable_interrupts(trans);
1646 
1647 		isr_stats->hw++;
1648 		iwl_pcie_irq_handle_error(trans);
1649 
1650 		handled |= CSR_INT_BIT_HW_ERR;
1651 
1652 		goto out;
1653 	}
1654 
1655 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1656 		/* NIC fires this, but we don't use it, redundant with WAKEUP */
1657 		if (inta & CSR_INT_BIT_SCD) {
1658 			IWL_DEBUG_ISR(trans,
1659 				      "Scheduler finished to transmit the frame/frames.\n");
1660 			isr_stats->sch++;
1661 		}
1662 
1663 		/* Alive notification via Rx interrupt will do the real work */
1664 		if (inta & CSR_INT_BIT_ALIVE) {
1665 			IWL_DEBUG_ISR(trans, "Alive interrupt\n");
1666 			isr_stats->alive++;
1667 			if (trans->cfg->gen2) {
1668 				/*
1669 				 * We can restock, since firmware configured
1670 				 * the RFH
1671 				 */
1672 				iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
1673 			}
1674 		}
1675 	}
1676 
1677 	/* Safely ignore these bits for debug checks below */
1678 	inta &= ~(CSR_INT_BIT_SCD | CSR_INT_BIT_ALIVE);
1679 
1680 	/* HW RF KILL switch toggled */
1681 	if (inta & CSR_INT_BIT_RF_KILL) {
1682 		iwl_pcie_handle_rfkill_irq(trans);
1683 		handled |= CSR_INT_BIT_RF_KILL;
1684 	}
1685 
1686 	/* Chip got too hot and stopped itself */
1687 	if (inta & CSR_INT_BIT_CT_KILL) {
1688 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
1689 		isr_stats->ctkill++;
1690 		handled |= CSR_INT_BIT_CT_KILL;
1691 	}
1692 
1693 	/* Error detected by uCode */
1694 	if (inta & CSR_INT_BIT_SW_ERR) {
1695 		IWL_ERR(trans, "Microcode SW error detected. "
1696 			" Restarting 0x%X.\n", inta);
1697 		isr_stats->sw++;
1698 		iwl_pcie_irq_handle_error(trans);
1699 		handled |= CSR_INT_BIT_SW_ERR;
1700 	}
1701 
1702 	/* uCode wakes up after power-down sleep */
1703 	if (inta & CSR_INT_BIT_WAKEUP) {
1704 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
1705 		iwl_pcie_rxq_check_wrptr(trans);
1706 		iwl_pcie_txq_check_wrptrs(trans);
1707 
1708 		isr_stats->wakeup++;
1709 
1710 		handled |= CSR_INT_BIT_WAKEUP;
1711 	}
1712 
1713 	/* All uCode command responses, including Tx command responses,
1714 	 * Rx "responses" (frame-received notification), and other
1715 	 * notifications from uCode come through here*/
1716 	if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX |
1717 		    CSR_INT_BIT_RX_PERIODIC)) {
1718 		IWL_DEBUG_ISR(trans, "Rx interrupt\n");
1719 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX)) {
1720 			handled |= (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX);
1721 			iwl_write32(trans, CSR_FH_INT_STATUS,
1722 					CSR_FH_INT_RX_MASK);
1723 		}
1724 		if (inta & CSR_INT_BIT_RX_PERIODIC) {
1725 			handled |= CSR_INT_BIT_RX_PERIODIC;
1726 			iwl_write32(trans,
1727 				CSR_INT, CSR_INT_BIT_RX_PERIODIC);
1728 		}
1729 		/* Sending RX interrupt require many steps to be done in the
1730 		 * the device:
1731 		 * 1- write interrupt to current index in ICT table.
1732 		 * 2- dma RX frame.
1733 		 * 3- update RX shared data to indicate last write index.
1734 		 * 4- send interrupt.
1735 		 * This could lead to RX race, driver could receive RX interrupt
1736 		 * but the shared data changes does not reflect this;
1737 		 * periodic interrupt will detect any dangling Rx activity.
1738 		 */
1739 
1740 		/* Disable periodic interrupt; we use it as just a one-shot. */
1741 		iwl_write8(trans, CSR_INT_PERIODIC_REG,
1742 			    CSR_INT_PERIODIC_DIS);
1743 
1744 		/*
1745 		 * Enable periodic interrupt in 8 msec only if we received
1746 		 * real RX interrupt (instead of just periodic int), to catch
1747 		 * any dangling Rx interrupt.  If it was just the periodic
1748 		 * interrupt, there was no dangling Rx activity, and no need
1749 		 * to extend the periodic interrupt; one-shot is enough.
1750 		 */
1751 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX))
1752 			iwl_write8(trans, CSR_INT_PERIODIC_REG,
1753 				   CSR_INT_PERIODIC_ENA);
1754 
1755 		isr_stats->rx++;
1756 
1757 		local_bh_disable();
1758 		iwl_pcie_rx_handle(trans, 0);
1759 		local_bh_enable();
1760 	}
1761 
1762 	/* This "Tx" DMA channel is used only for loading uCode */
1763 	if (inta & CSR_INT_BIT_FH_TX) {
1764 		iwl_write32(trans, CSR_FH_INT_STATUS, CSR_FH_INT_TX_MASK);
1765 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
1766 		isr_stats->tx++;
1767 		handled |= CSR_INT_BIT_FH_TX;
1768 		/* Wake up uCode load routine, now that load is complete */
1769 		trans_pcie->ucode_write_complete = true;
1770 		wake_up(&trans_pcie->ucode_write_waitq);
1771 	}
1772 
1773 	if (inta & ~handled) {
1774 		IWL_ERR(trans, "Unhandled INTA bits 0x%08x\n", inta & ~handled);
1775 		isr_stats->unhandled++;
1776 	}
1777 
1778 	if (inta & ~(trans_pcie->inta_mask)) {
1779 		IWL_WARN(trans, "Disabled INTA bits 0x%08x were pending\n",
1780 			 inta & ~trans_pcie->inta_mask);
1781 	}
1782 
1783 	spin_lock(&trans_pcie->irq_lock);
1784 	/* only Re-enable all interrupt if disabled by irq */
1785 	if (test_bit(STATUS_INT_ENABLED, &trans->status))
1786 		_iwl_enable_interrupts(trans);
1787 	/* we are loading the firmware, enable FH_TX interrupt only */
1788 	else if (handled & CSR_INT_BIT_FH_TX)
1789 		iwl_enable_fw_load_int(trans);
1790 	/* Re-enable RF_KILL if it occurred */
1791 	else if (handled & CSR_INT_BIT_RF_KILL)
1792 		iwl_enable_rfkill_int(trans);
1793 	spin_unlock(&trans_pcie->irq_lock);
1794 
1795 out:
1796 	lock_map_release(&trans->sync_cmd_lockdep_map);
1797 	return IRQ_HANDLED;
1798 }
1799 
1800 /******************************************************************************
1801  *
1802  * ICT functions
1803  *
1804  ******************************************************************************/
1805 
1806 /* Free dram table */
1807 void iwl_pcie_free_ict(struct iwl_trans *trans)
1808 {
1809 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1810 
1811 	if (trans_pcie->ict_tbl) {
1812 		dma_free_coherent(trans->dev, ICT_SIZE,
1813 				  trans_pcie->ict_tbl,
1814 				  trans_pcie->ict_tbl_dma);
1815 		trans_pcie->ict_tbl = NULL;
1816 		trans_pcie->ict_tbl_dma = 0;
1817 	}
1818 }
1819 
1820 /*
1821  * allocate dram shared table, it is an aligned memory
1822  * block of ICT_SIZE.
1823  * also reset all data related to ICT table interrupt.
1824  */
1825 int iwl_pcie_alloc_ict(struct iwl_trans *trans)
1826 {
1827 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1828 
1829 	trans_pcie->ict_tbl =
1830 		dma_zalloc_coherent(trans->dev, ICT_SIZE,
1831 				   &trans_pcie->ict_tbl_dma,
1832 				   GFP_KERNEL);
1833 	if (!trans_pcie->ict_tbl)
1834 		return -ENOMEM;
1835 
1836 	/* just an API sanity check ... it is guaranteed to be aligned */
1837 	if (WARN_ON(trans_pcie->ict_tbl_dma & (ICT_SIZE - 1))) {
1838 		iwl_pcie_free_ict(trans);
1839 		return -EINVAL;
1840 	}
1841 
1842 	return 0;
1843 }
1844 
1845 /* Device is going up inform it about using ICT interrupt table,
1846  * also we need to tell the driver to start using ICT interrupt.
1847  */
1848 void iwl_pcie_reset_ict(struct iwl_trans *trans)
1849 {
1850 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1851 	u32 val;
1852 
1853 	if (!trans_pcie->ict_tbl)
1854 		return;
1855 
1856 	spin_lock(&trans_pcie->irq_lock);
1857 	_iwl_disable_interrupts(trans);
1858 
1859 	memset(trans_pcie->ict_tbl, 0, ICT_SIZE);
1860 
1861 	val = trans_pcie->ict_tbl_dma >> ICT_SHIFT;
1862 
1863 	val |= CSR_DRAM_INT_TBL_ENABLE |
1864 	       CSR_DRAM_INIT_TBL_WRAP_CHECK |
1865 	       CSR_DRAM_INIT_TBL_WRITE_POINTER;
1866 
1867 	IWL_DEBUG_ISR(trans, "CSR_DRAM_INT_TBL_REG =0x%x\n", val);
1868 
1869 	iwl_write32(trans, CSR_DRAM_INT_TBL_REG, val);
1870 	trans_pcie->use_ict = true;
1871 	trans_pcie->ict_index = 0;
1872 	iwl_write32(trans, CSR_INT, trans_pcie->inta_mask);
1873 	_iwl_enable_interrupts(trans);
1874 	spin_unlock(&trans_pcie->irq_lock);
1875 }
1876 
1877 /* Device is going down disable ict interrupt usage */
1878 void iwl_pcie_disable_ict(struct iwl_trans *trans)
1879 {
1880 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1881 
1882 	spin_lock(&trans_pcie->irq_lock);
1883 	trans_pcie->use_ict = false;
1884 	spin_unlock(&trans_pcie->irq_lock);
1885 }
1886 
1887 irqreturn_t iwl_pcie_isr(int irq, void *data)
1888 {
1889 	struct iwl_trans *trans = data;
1890 
1891 	if (!trans)
1892 		return IRQ_NONE;
1893 
1894 	/* Disable (but don't clear!) interrupts here to avoid
1895 	 * back-to-back ISRs and sporadic interrupts from our NIC.
1896 	 * If we have something to service, the tasklet will re-enable ints.
1897 	 * If we *don't* have something, we'll re-enable before leaving here.
1898 	 */
1899 	iwl_write32(trans, CSR_INT_MASK, 0x00000000);
1900 
1901 	return IRQ_WAKE_THREAD;
1902 }
1903 
1904 irqreturn_t iwl_pcie_msix_isr(int irq, void *data)
1905 {
1906 	return IRQ_WAKE_THREAD;
1907 }
1908 
1909 irqreturn_t iwl_pcie_irq_msix_handler(int irq, void *dev_id)
1910 {
1911 	struct msix_entry *entry = dev_id;
1912 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
1913 	struct iwl_trans *trans = trans_pcie->trans;
1914 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1915 	u32 inta_fh, inta_hw;
1916 
1917 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1918 
1919 	spin_lock(&trans_pcie->irq_lock);
1920 	inta_fh = iwl_read32(trans, CSR_MSIX_FH_INT_CAUSES_AD);
1921 	inta_hw = iwl_read32(trans, CSR_MSIX_HW_INT_CAUSES_AD);
1922 	/*
1923 	 * Clear causes registers to avoid being handling the same cause.
1924 	 */
1925 	iwl_write32(trans, CSR_MSIX_FH_INT_CAUSES_AD, inta_fh);
1926 	iwl_write32(trans, CSR_MSIX_HW_INT_CAUSES_AD, inta_hw);
1927 	spin_unlock(&trans_pcie->irq_lock);
1928 
1929 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, true, inta_fh, inta_hw);
1930 
1931 	if (unlikely(!(inta_fh | inta_hw))) {
1932 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
1933 		lock_map_release(&trans->sync_cmd_lockdep_map);
1934 		return IRQ_NONE;
1935 	}
1936 
1937 	if (iwl_have_debug_level(IWL_DL_ISR))
1938 		IWL_DEBUG_ISR(trans, "ISR inta_fh 0x%08x, enabled 0x%08x\n",
1939 			      inta_fh,
1940 			      iwl_read32(trans, CSR_MSIX_FH_INT_MASK_AD));
1941 
1942 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_NON_RX) &&
1943 	    inta_fh & MSIX_FH_INT_CAUSES_Q0) {
1944 		local_bh_disable();
1945 		iwl_pcie_rx_handle(trans, 0);
1946 		local_bh_enable();
1947 	}
1948 
1949 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS) &&
1950 	    inta_fh & MSIX_FH_INT_CAUSES_Q1) {
1951 		local_bh_disable();
1952 		iwl_pcie_rx_handle(trans, 1);
1953 		local_bh_enable();
1954 	}
1955 
1956 	/* This "Tx" DMA channel is used only for loading uCode */
1957 	if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM) {
1958 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
1959 		isr_stats->tx++;
1960 		/*
1961 		 * Wake up uCode load routine,
1962 		 * now that load is complete
1963 		 */
1964 		trans_pcie->ucode_write_complete = true;
1965 		wake_up(&trans_pcie->ucode_write_waitq);
1966 	}
1967 
1968 	/* Error detected by uCode */
1969 	if ((inta_fh & MSIX_FH_INT_CAUSES_FH_ERR) ||
1970 	    (inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR)) {
1971 		IWL_ERR(trans,
1972 			"Microcode SW error detected. Restarting 0x%X.\n",
1973 			inta_fh);
1974 		isr_stats->sw++;
1975 		iwl_pcie_irq_handle_error(trans);
1976 	}
1977 
1978 	/* After checking FH register check HW register */
1979 	if (iwl_have_debug_level(IWL_DL_ISR))
1980 		IWL_DEBUG_ISR(trans,
1981 			      "ISR inta_hw 0x%08x, enabled 0x%08x\n",
1982 			      inta_hw,
1983 			      iwl_read32(trans, CSR_MSIX_HW_INT_MASK_AD));
1984 
1985 	/* Alive notification via Rx interrupt will do the real work */
1986 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_ALIVE) {
1987 		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
1988 		isr_stats->alive++;
1989 		if (trans->cfg->gen2) {
1990 			/* We can restock, since firmware configured the RFH */
1991 			iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
1992 		}
1993 	}
1994 
1995 	/* uCode wakes up after power-down sleep */
1996 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_WAKEUP) {
1997 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
1998 		iwl_pcie_rxq_check_wrptr(trans);
1999 		iwl_pcie_txq_check_wrptrs(trans);
2000 
2001 		isr_stats->wakeup++;
2002 	}
2003 
2004 	/* Chip got too hot and stopped itself */
2005 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_CT_KILL) {
2006 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
2007 		isr_stats->ctkill++;
2008 	}
2009 
2010 	/* HW RF KILL switch toggled */
2011 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RF_KILL)
2012 		iwl_pcie_handle_rfkill_irq(trans);
2013 
2014 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_HW_ERR) {
2015 		IWL_ERR(trans,
2016 			"Hardware error detected. Restarting.\n");
2017 
2018 		isr_stats->hw++;
2019 		iwl_pcie_irq_handle_error(trans);
2020 	}
2021 
2022 	iwl_pcie_clear_irq(trans, entry);
2023 
2024 	lock_map_release(&trans->sync_cmd_lockdep_map);
2025 
2026 	return IRQ_HANDLED;
2027 }
2028