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 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 		struct 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 		INIT_LIST_HEAD(&local_allocated);
498 
499 		for (i = 0; i < RX_CLAIM_REQ_ALLOC;) {
500 			struct iwl_rx_mem_buffer *rxb;
501 			struct page *page;
502 
503 			/* List should never be empty - each reused RBD is
504 			 * returned to the list, and initial pool covers any
505 			 * possible gap between the time the page is allocated
506 			 * to the time the RBD is added.
507 			 */
508 			BUG_ON(list_empty(&local_empty));
509 			/* Get the first rxb from the rbd list */
510 			rxb = list_first_entry(&local_empty,
511 					       struct iwl_rx_mem_buffer, list);
512 			BUG_ON(rxb->page);
513 
514 			/* Alloc a new receive buffer */
515 			page = iwl_pcie_rx_alloc_page(trans, gfp_mask);
516 			if (!page)
517 				continue;
518 			rxb->page = page;
519 
520 			/* Get physical address of the RB */
521 			rxb->page_dma = dma_map_page(trans->dev, page, 0,
522 					PAGE_SIZE << trans_pcie->rx_page_order,
523 					DMA_FROM_DEVICE);
524 			if (dma_mapping_error(trans->dev, rxb->page_dma)) {
525 				rxb->page = NULL;
526 				__free_pages(page, trans_pcie->rx_page_order);
527 				continue;
528 			}
529 
530 			/* move the allocated entry to the out list */
531 			list_move(&rxb->list, &local_allocated);
532 			i++;
533 		}
534 
535 		pending--;
536 		if (!pending) {
537 			pending = atomic_xchg(&rba->req_pending, 0);
538 			IWL_DEBUG_RX(trans,
539 				     "Pending allocation requests = %d\n",
540 				     pending);
541 		}
542 
543 		spin_lock(&rba->lock);
544 		/* add the allocated rbds to the allocator allocated list */
545 		list_splice_tail(&local_allocated, &rba->rbd_allocated);
546 		/* get more empty RBDs for current pending requests */
547 		list_splice_tail_init(&rba->rbd_empty, &local_empty);
548 		spin_unlock(&rba->lock);
549 
550 		atomic_inc(&rba->req_ready);
551 	}
552 
553 	spin_lock(&rba->lock);
554 	/* return unused rbds to the allocator empty list */
555 	list_splice_tail(&local_empty, &rba->rbd_empty);
556 	spin_unlock(&rba->lock);
557 }
558 
559 /*
560  * iwl_pcie_rx_allocator_get - returns the pre-allocated pages
561 .*
562 .* Called by queue when the queue posted allocation request and
563  * has freed 8 RBDs in order to restock itself.
564  * This function directly moves the allocated RBs to the queue's ownership
565  * and updates the relevant counters.
566  */
567 static void iwl_pcie_rx_allocator_get(struct iwl_trans *trans,
568 				      struct iwl_rxq *rxq)
569 {
570 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
571 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
572 	int i;
573 
574 	lockdep_assert_held(&rxq->lock);
575 
576 	/*
577 	 * atomic_dec_if_positive returns req_ready - 1 for any scenario.
578 	 * If req_ready is 0 atomic_dec_if_positive will return -1 and this
579 	 * function will return early, as there are no ready requests.
580 	 * atomic_dec_if_positive will perofrm the *actual* decrement only if
581 	 * req_ready > 0, i.e. - there are ready requests and the function
582 	 * hands one request to the caller.
583 	 */
584 	if (atomic_dec_if_positive(&rba->req_ready) < 0)
585 		return;
586 
587 	spin_lock(&rba->lock);
588 	for (i = 0; i < RX_CLAIM_REQ_ALLOC; i++) {
589 		/* Get next free Rx buffer, remove it from free list */
590 		struct iwl_rx_mem_buffer *rxb =
591 			list_first_entry(&rba->rbd_allocated,
592 					 struct iwl_rx_mem_buffer, list);
593 
594 		list_move(&rxb->list, &rxq->rx_free);
595 	}
596 	spin_unlock(&rba->lock);
597 
598 	rxq->used_count -= RX_CLAIM_REQ_ALLOC;
599 	rxq->free_count += RX_CLAIM_REQ_ALLOC;
600 }
601 
602 static void iwl_pcie_rx_allocator_work(struct work_struct *data)
603 {
604 	struct iwl_rb_allocator *rba_p =
605 		container_of(data, struct iwl_rb_allocator, rx_alloc);
606 	struct iwl_trans_pcie *trans_pcie =
607 		container_of(rba_p, struct iwl_trans_pcie, rba);
608 
609 	iwl_pcie_rx_allocator(trans_pcie->trans);
610 }
611 
612 static int iwl_pcie_rx_alloc(struct iwl_trans *trans)
613 {
614 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
615 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
616 	struct device *dev = trans->dev;
617 	int i;
618 	int free_size = trans->cfg->mq_rx_supported ? sizeof(__le64) :
619 						      sizeof(__le32);
620 
621 	if (WARN_ON(trans_pcie->rxq))
622 		return -EINVAL;
623 
624 	trans_pcie->rxq = kcalloc(trans->num_rx_queues, sizeof(struct iwl_rxq),
625 				  GFP_KERNEL);
626 	if (!trans_pcie->rxq)
627 		return -EINVAL;
628 
629 	spin_lock_init(&rba->lock);
630 
631 	for (i = 0; i < trans->num_rx_queues; i++) {
632 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
633 
634 		spin_lock_init(&rxq->lock);
635 		if (trans->cfg->mq_rx_supported)
636 			rxq->queue_size = MQ_RX_TABLE_SIZE;
637 		else
638 			rxq->queue_size = RX_QUEUE_SIZE;
639 
640 		/*
641 		 * Allocate the circular buffer of Read Buffer Descriptors
642 		 * (RBDs)
643 		 */
644 		rxq->bd = dma_zalloc_coherent(dev,
645 					     free_size * rxq->queue_size,
646 					     &rxq->bd_dma, GFP_KERNEL);
647 		if (!rxq->bd)
648 			goto err;
649 
650 		if (trans->cfg->mq_rx_supported) {
651 			rxq->used_bd = dma_zalloc_coherent(dev,
652 							   sizeof(__le32) *
653 							   rxq->queue_size,
654 							   &rxq->used_bd_dma,
655 							   GFP_KERNEL);
656 			if (!rxq->used_bd)
657 				goto err;
658 		}
659 
660 		/*Allocate the driver's pointer to receive buffer status */
661 		rxq->rb_stts = dma_zalloc_coherent(dev, sizeof(*rxq->rb_stts),
662 						   &rxq->rb_stts_dma,
663 						   GFP_KERNEL);
664 		if (!rxq->rb_stts)
665 			goto err;
666 	}
667 	return 0;
668 
669 err:
670 	for (i = 0; i < trans->num_rx_queues; i++) {
671 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
672 
673 		if (rxq->bd)
674 			dma_free_coherent(dev, free_size * rxq->queue_size,
675 					  rxq->bd, rxq->bd_dma);
676 		rxq->bd_dma = 0;
677 		rxq->bd = NULL;
678 
679 		if (rxq->rb_stts)
680 			dma_free_coherent(trans->dev,
681 					  sizeof(struct iwl_rb_status),
682 					  rxq->rb_stts, rxq->rb_stts_dma);
683 
684 		if (rxq->used_bd)
685 			dma_free_coherent(dev, sizeof(__le32) * rxq->queue_size,
686 					  rxq->used_bd, rxq->used_bd_dma);
687 		rxq->used_bd_dma = 0;
688 		rxq->used_bd = NULL;
689 	}
690 	kfree(trans_pcie->rxq);
691 
692 	return -ENOMEM;
693 }
694 
695 static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
696 {
697 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
698 	u32 rb_size;
699 	unsigned long flags;
700 	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
701 
702 	switch (trans_pcie->rx_buf_size) {
703 	case IWL_AMSDU_4K:
704 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
705 		break;
706 	case IWL_AMSDU_8K:
707 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
708 		break;
709 	case IWL_AMSDU_12K:
710 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_12K;
711 		break;
712 	default:
713 		WARN_ON(1);
714 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
715 	}
716 
717 	if (!iwl_trans_grab_nic_access(trans, &flags))
718 		return;
719 
720 	/* Stop Rx DMA */
721 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
722 	/* reset and flush pointers */
723 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
724 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
725 	iwl_write32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
726 
727 	/* Reset driver's Rx queue write index */
728 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
729 
730 	/* Tell device where to find RBD circular buffer in DRAM */
731 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
732 		    (u32)(rxq->bd_dma >> 8));
733 
734 	/* Tell device where in DRAM to update its Rx status */
735 	iwl_write32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
736 		    rxq->rb_stts_dma >> 4);
737 
738 	/* Enable Rx DMA
739 	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
740 	 *      the credit mechanism in 5000 HW RX FIFO
741 	 * Direct rx interrupts to hosts
742 	 * Rx buffer size 4 or 8k or 12k
743 	 * RB timeout 0x10
744 	 * 256 RBDs
745 	 */
746 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
747 		    FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
748 		    FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
749 		    FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
750 		    rb_size |
751 		    (RX_RB_TIMEOUT << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS) |
752 		    (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));
753 
754 	iwl_trans_release_nic_access(trans, &flags);
755 
756 	/* Set interrupt coalescing timer to default (2048 usecs) */
757 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
758 
759 	/* W/A for interrupt coalescing bug in 7260 and 3160 */
760 	if (trans->cfg->host_interrupt_operation_mode)
761 		iwl_set_bit(trans, CSR_INT_COALESCING, IWL_HOST_INT_OPER_MODE);
762 }
763 
764 void iwl_pcie_enable_rx_wake(struct iwl_trans *trans, bool enable)
765 {
766 	/*
767 	 * Turn on the chicken-bits that cause MAC wakeup for RX-related
768 	 * values.
769 	 * This costs some power, but needed for W/A 9000 integrated A-step
770 	 * bug where shadow registers are not in the retention list and their
771 	 * value is lost when NIC powers down
772 	 */
773 	if (trans->cfg->integrated) {
774 		iwl_set_bit(trans, CSR_MAC_SHADOW_REG_CTRL,
775 			    CSR_MAC_SHADOW_REG_CTRL_RX_WAKE);
776 		iwl_set_bit(trans, CSR_MAC_SHADOW_REG_CTL2,
777 			    CSR_MAC_SHADOW_REG_CTL2_RX_WAKE);
778 	}
779 }
780 
781 static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
782 {
783 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
784 	u32 rb_size, enabled = 0;
785 	unsigned long flags;
786 	int i;
787 
788 	switch (trans_pcie->rx_buf_size) {
789 	case IWL_AMSDU_4K:
790 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
791 		break;
792 	case IWL_AMSDU_8K:
793 		rb_size = RFH_RXF_DMA_RB_SIZE_8K;
794 		break;
795 	case IWL_AMSDU_12K:
796 		rb_size = RFH_RXF_DMA_RB_SIZE_12K;
797 		break;
798 	default:
799 		WARN_ON(1);
800 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
801 	}
802 
803 	if (!iwl_trans_grab_nic_access(trans, &flags))
804 		return;
805 
806 	/* Stop Rx DMA */
807 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG, 0);
808 	/* disable free amd used rx queue operation */
809 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, 0);
810 
811 	for (i = 0; i < trans->num_rx_queues; i++) {
812 		/* Tell device where to find RBD free table in DRAM */
813 		iwl_write_prph64_no_grab(trans,
814 					 RFH_Q_FRBDCB_BA_LSB(i),
815 					 trans_pcie->rxq[i].bd_dma);
816 		/* Tell device where to find RBD used table in DRAM */
817 		iwl_write_prph64_no_grab(trans,
818 					 RFH_Q_URBDCB_BA_LSB(i),
819 					 trans_pcie->rxq[i].used_bd_dma);
820 		/* Tell device where in DRAM to update its Rx status */
821 		iwl_write_prph64_no_grab(trans,
822 					 RFH_Q_URBD_STTS_WPTR_LSB(i),
823 					 trans_pcie->rxq[i].rb_stts_dma);
824 		/* Reset device indice tables */
825 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_WIDX(i), 0);
826 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_RIDX(i), 0);
827 		iwl_write_prph_no_grab(trans, RFH_Q_URBDCB_WIDX(i), 0);
828 
829 		enabled |= BIT(i) | BIT(i + 16);
830 	}
831 
832 	/*
833 	 * Enable Rx DMA
834 	 * Rx buffer size 4 or 8k or 12k
835 	 * Min RB size 4 or 8
836 	 * Drop frames that exceed RB size
837 	 * 512 RBDs
838 	 */
839 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG,
840 			       RFH_DMA_EN_ENABLE_VAL | rb_size |
841 			       RFH_RXF_DMA_MIN_RB_4_8 |
842 			       RFH_RXF_DMA_DROP_TOO_LARGE_MASK |
843 			       RFH_RXF_DMA_RBDCB_SIZE_512);
844 
845 	/*
846 	 * Activate DMA snooping.
847 	 * Set RX DMA chunk size to 64B for IOSF and 128B for PCIe
848 	 * Default queue is 0
849 	 */
850 	iwl_write_prph_no_grab(trans, RFH_GEN_CFG, RFH_GEN_CFG_RFH_DMA_SNOOP |
851 			       (DEFAULT_RXQ_NUM <<
852 				RFH_GEN_CFG_DEFAULT_RXQ_NUM_POS) |
853 			       RFH_GEN_CFG_SERVICE_DMA_SNOOP |
854 			       (trans->cfg->integrated ?
855 				RFH_GEN_CFG_RB_CHUNK_SIZE_64 :
856 				RFH_GEN_CFG_RB_CHUNK_SIZE_128) <<
857 			       RFH_GEN_CFG_RB_CHUNK_SIZE_POS);
858 	/* Enable the relevant rx queues */
859 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, enabled);
860 
861 	iwl_trans_release_nic_access(trans, &flags);
862 
863 	/* Set interrupt coalescing timer to default (2048 usecs) */
864 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
865 
866 	iwl_pcie_enable_rx_wake(trans, true);
867 }
868 
869 static void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq)
870 {
871 	lockdep_assert_held(&rxq->lock);
872 
873 	INIT_LIST_HEAD(&rxq->rx_free);
874 	INIT_LIST_HEAD(&rxq->rx_used);
875 	rxq->free_count = 0;
876 	rxq->used_count = 0;
877 }
878 
879 static int iwl_pcie_dummy_napi_poll(struct napi_struct *napi, int budget)
880 {
881 	WARN_ON(1);
882 	return 0;
883 }
884 
885 int iwl_pcie_rx_init(struct iwl_trans *trans)
886 {
887 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
888 	struct iwl_rxq *def_rxq;
889 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
890 	int i, err, queue_size, allocator_pool_size, num_alloc;
891 
892 	if (!trans_pcie->rxq) {
893 		err = iwl_pcie_rx_alloc(trans);
894 		if (err)
895 			return err;
896 	}
897 	def_rxq = trans_pcie->rxq;
898 	if (!rba->alloc_wq)
899 		rba->alloc_wq = alloc_workqueue("rb_allocator",
900 						WQ_HIGHPRI | WQ_UNBOUND, 1);
901 	INIT_WORK(&rba->rx_alloc, iwl_pcie_rx_allocator_work);
902 
903 	spin_lock(&rba->lock);
904 	atomic_set(&rba->req_pending, 0);
905 	atomic_set(&rba->req_ready, 0);
906 	INIT_LIST_HEAD(&rba->rbd_allocated);
907 	INIT_LIST_HEAD(&rba->rbd_empty);
908 	spin_unlock(&rba->lock);
909 
910 	/* free all first - we might be reconfigured for a different size */
911 	iwl_pcie_free_rbs_pool(trans);
912 
913 	for (i = 0; i < RX_QUEUE_SIZE; i++)
914 		def_rxq->queue[i] = NULL;
915 
916 	for (i = 0; i < trans->num_rx_queues; i++) {
917 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
918 
919 		rxq->id = i;
920 
921 		spin_lock(&rxq->lock);
922 		/*
923 		 * Set read write pointer to reflect that we have processed
924 		 * and used all buffers, but have not restocked the Rx queue
925 		 * with fresh buffers
926 		 */
927 		rxq->read = 0;
928 		rxq->write = 0;
929 		rxq->write_actual = 0;
930 		memset(rxq->rb_stts, 0, sizeof(*rxq->rb_stts));
931 
932 		iwl_pcie_rx_init_rxb_lists(rxq);
933 
934 		if (!rxq->napi.poll)
935 			netif_napi_add(&trans_pcie->napi_dev, &rxq->napi,
936 				       iwl_pcie_dummy_napi_poll, 64);
937 
938 		spin_unlock(&rxq->lock);
939 	}
940 
941 	/* move the pool to the default queue and allocator ownerships */
942 	queue_size = trans->cfg->mq_rx_supported ?
943 		     MQ_RX_NUM_RBDS : RX_QUEUE_SIZE;
944 	allocator_pool_size = trans->num_rx_queues *
945 		(RX_CLAIM_REQ_ALLOC - RX_POST_REQ_ALLOC);
946 	num_alloc = queue_size + allocator_pool_size;
947 	BUILD_BUG_ON(ARRAY_SIZE(trans_pcie->global_table) !=
948 		     ARRAY_SIZE(trans_pcie->rx_pool));
949 	for (i = 0; i < num_alloc; i++) {
950 		struct iwl_rx_mem_buffer *rxb = &trans_pcie->rx_pool[i];
951 
952 		if (i < allocator_pool_size)
953 			list_add(&rxb->list, &rba->rbd_empty);
954 		else
955 			list_add(&rxb->list, &def_rxq->rx_used);
956 		trans_pcie->global_table[i] = rxb;
957 		rxb->vid = (u16)(i + 1);
958 		rxb->invalid = true;
959 	}
960 
961 	iwl_pcie_rxq_alloc_rbs(trans, GFP_KERNEL, def_rxq);
962 
963 	if (trans->cfg->mq_rx_supported)
964 		iwl_pcie_rx_mq_hw_init(trans);
965 	else
966 		iwl_pcie_rx_hw_init(trans, def_rxq);
967 
968 	iwl_pcie_rxq_restock(trans, def_rxq);
969 
970 	spin_lock(&def_rxq->lock);
971 	iwl_pcie_rxq_inc_wr_ptr(trans, def_rxq);
972 	spin_unlock(&def_rxq->lock);
973 
974 	return 0;
975 }
976 
977 void iwl_pcie_rx_free(struct iwl_trans *trans)
978 {
979 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
980 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
981 	int free_size = trans->cfg->mq_rx_supported ? sizeof(__le64) :
982 					      sizeof(__le32);
983 	int i;
984 
985 	/*
986 	 * if rxq is NULL, it means that nothing has been allocated,
987 	 * exit now
988 	 */
989 	if (!trans_pcie->rxq) {
990 		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
991 		return;
992 	}
993 
994 	cancel_work_sync(&rba->rx_alloc);
995 	if (rba->alloc_wq) {
996 		destroy_workqueue(rba->alloc_wq);
997 		rba->alloc_wq = NULL;
998 	}
999 
1000 	iwl_pcie_free_rbs_pool(trans);
1001 
1002 	for (i = 0; i < trans->num_rx_queues; i++) {
1003 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1004 
1005 		if (rxq->bd)
1006 			dma_free_coherent(trans->dev,
1007 					  free_size * rxq->queue_size,
1008 					  rxq->bd, rxq->bd_dma);
1009 		rxq->bd_dma = 0;
1010 		rxq->bd = NULL;
1011 
1012 		if (rxq->rb_stts)
1013 			dma_free_coherent(trans->dev,
1014 					  sizeof(struct iwl_rb_status),
1015 					  rxq->rb_stts, rxq->rb_stts_dma);
1016 		else
1017 			IWL_DEBUG_INFO(trans,
1018 				       "Free rxq->rb_stts which is NULL\n");
1019 
1020 		if (rxq->used_bd)
1021 			dma_free_coherent(trans->dev,
1022 					  sizeof(__le32) * rxq->queue_size,
1023 					  rxq->used_bd, rxq->used_bd_dma);
1024 		rxq->used_bd_dma = 0;
1025 		rxq->used_bd = NULL;
1026 
1027 		if (rxq->napi.poll)
1028 			netif_napi_del(&rxq->napi);
1029 	}
1030 	kfree(trans_pcie->rxq);
1031 }
1032 
1033 /*
1034  * iwl_pcie_rx_reuse_rbd - Recycle used RBDs
1035  *
1036  * Called when a RBD can be reused. The RBD is transferred to the allocator.
1037  * When there are 2 empty RBDs - a request for allocation is posted
1038  */
1039 static void iwl_pcie_rx_reuse_rbd(struct iwl_trans *trans,
1040 				  struct iwl_rx_mem_buffer *rxb,
1041 				  struct iwl_rxq *rxq, bool emergency)
1042 {
1043 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1044 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1045 
1046 	/* Move the RBD to the used list, will be moved to allocator in batches
1047 	 * before claiming or posting a request*/
1048 	list_add_tail(&rxb->list, &rxq->rx_used);
1049 
1050 	if (unlikely(emergency))
1051 		return;
1052 
1053 	/* Count the allocator owned RBDs */
1054 	rxq->used_count++;
1055 
1056 	/* If we have RX_POST_REQ_ALLOC new released rx buffers -
1057 	 * issue a request for allocator. Modulo RX_CLAIM_REQ_ALLOC is
1058 	 * used for the case we failed to claim RX_CLAIM_REQ_ALLOC,
1059 	 * after but we still need to post another request.
1060 	 */
1061 	if ((rxq->used_count % RX_CLAIM_REQ_ALLOC) == RX_POST_REQ_ALLOC) {
1062 		/* Move the 2 RBDs to the allocator ownership.
1063 		 Allocator has another 6 from pool for the request completion*/
1064 		spin_lock(&rba->lock);
1065 		list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
1066 		spin_unlock(&rba->lock);
1067 
1068 		atomic_inc(&rba->req_pending);
1069 		queue_work(rba->alloc_wq, &rba->rx_alloc);
1070 	}
1071 }
1072 
1073 static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
1074 				struct iwl_rxq *rxq,
1075 				struct iwl_rx_mem_buffer *rxb,
1076 				bool emergency)
1077 {
1078 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1079 	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1080 	bool page_stolen = false;
1081 	int max_len = PAGE_SIZE << trans_pcie->rx_page_order;
1082 	u32 offset = 0;
1083 
1084 	if (WARN_ON(!rxb))
1085 		return;
1086 
1087 	dma_unmap_page(trans->dev, rxb->page_dma, max_len, DMA_FROM_DEVICE);
1088 
1089 	while (offset + sizeof(u32) + sizeof(struct iwl_cmd_header) < max_len) {
1090 		struct iwl_rx_packet *pkt;
1091 		u16 sequence;
1092 		bool reclaim;
1093 		int index, cmd_index, len;
1094 		struct iwl_rx_cmd_buffer rxcb = {
1095 			._offset = offset,
1096 			._rx_page_order = trans_pcie->rx_page_order,
1097 			._page = rxb->page,
1098 			._page_stolen = false,
1099 			.truesize = max_len,
1100 		};
1101 
1102 		pkt = rxb_addr(&rxcb);
1103 
1104 		if (pkt->len_n_flags == cpu_to_le32(FH_RSCSR_FRAME_INVALID))
1105 			break;
1106 
1107 		WARN_ON((le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1108 			FH_RSCSR_RXQ_POS != rxq->id);
1109 
1110 		IWL_DEBUG_RX(trans,
1111 			     "cmd at offset %d: %s (0x%.2x, seq 0x%x)\n",
1112 			     rxcb._offset,
1113 			     iwl_get_cmd_string(trans,
1114 						iwl_cmd_id(pkt->hdr.cmd,
1115 							   pkt->hdr.group_id,
1116 							   0)),
1117 			     pkt->hdr.cmd, le16_to_cpu(pkt->hdr.sequence));
1118 
1119 		len = iwl_rx_packet_len(pkt);
1120 		len += sizeof(u32); /* account for status word */
1121 		trace_iwlwifi_dev_rx(trans->dev, trans, pkt, len);
1122 		trace_iwlwifi_dev_rx_data(trans->dev, trans, pkt, len);
1123 
1124 		/* Reclaim a command buffer only if this packet is a response
1125 		 *   to a (driver-originated) command.
1126 		 * If the packet (e.g. Rx frame) originated from uCode,
1127 		 *   there is no command buffer to reclaim.
1128 		 * Ucode should set SEQ_RX_FRAME bit if ucode-originated,
1129 		 *   but apparently a few don't get set; catch them here. */
1130 		reclaim = !(pkt->hdr.sequence & SEQ_RX_FRAME);
1131 		if (reclaim) {
1132 			int i;
1133 
1134 			for (i = 0; i < trans_pcie->n_no_reclaim_cmds; i++) {
1135 				if (trans_pcie->no_reclaim_cmds[i] ==
1136 							pkt->hdr.cmd) {
1137 					reclaim = false;
1138 					break;
1139 				}
1140 			}
1141 		}
1142 
1143 		sequence = le16_to_cpu(pkt->hdr.sequence);
1144 		index = SEQ_TO_INDEX(sequence);
1145 		cmd_index = get_cmd_index(&txq->q, index);
1146 
1147 		if (rxq->id == 0)
1148 			iwl_op_mode_rx(trans->op_mode, &rxq->napi,
1149 				       &rxcb);
1150 		else
1151 			iwl_op_mode_rx_rss(trans->op_mode, &rxq->napi,
1152 					   &rxcb, rxq->id);
1153 
1154 		if (reclaim) {
1155 			kzfree(txq->entries[cmd_index].free_buf);
1156 			txq->entries[cmd_index].free_buf = NULL;
1157 		}
1158 
1159 		/*
1160 		 * After here, we should always check rxcb._page_stolen,
1161 		 * if it is true then one of the handlers took the page.
1162 		 */
1163 
1164 		if (reclaim) {
1165 			/* Invoke any callbacks, transfer the buffer to caller,
1166 			 * and fire off the (possibly) blocking
1167 			 * iwl_trans_send_cmd()
1168 			 * as we reclaim the driver command queue */
1169 			if (!rxcb._page_stolen)
1170 				iwl_pcie_hcmd_complete(trans, &rxcb);
1171 			else
1172 				IWL_WARN(trans, "Claim null rxb?\n");
1173 		}
1174 
1175 		page_stolen |= rxcb._page_stolen;
1176 		offset += ALIGN(len, FH_RSCSR_FRAME_ALIGN);
1177 	}
1178 
1179 	/* page was stolen from us -- free our reference */
1180 	if (page_stolen) {
1181 		__free_pages(rxb->page, trans_pcie->rx_page_order);
1182 		rxb->page = NULL;
1183 	}
1184 
1185 	/* Reuse the page if possible. For notification packets and
1186 	 * SKBs that fail to Rx correctly, add them back into the
1187 	 * rx_free list for reuse later. */
1188 	if (rxb->page != NULL) {
1189 		rxb->page_dma =
1190 			dma_map_page(trans->dev, rxb->page, 0,
1191 				     PAGE_SIZE << trans_pcie->rx_page_order,
1192 				     DMA_FROM_DEVICE);
1193 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
1194 			/*
1195 			 * free the page(s) as well to not break
1196 			 * the invariant that the items on the used
1197 			 * list have no page(s)
1198 			 */
1199 			__free_pages(rxb->page, trans_pcie->rx_page_order);
1200 			rxb->page = NULL;
1201 			iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1202 		} else {
1203 			list_add_tail(&rxb->list, &rxq->rx_free);
1204 			rxq->free_count++;
1205 		}
1206 	} else
1207 		iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1208 }
1209 
1210 /*
1211  * iwl_pcie_rx_handle - Main entry function for receiving responses from fw
1212  */
1213 static void iwl_pcie_rx_handle(struct iwl_trans *trans, int queue)
1214 {
1215 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1216 	struct iwl_rxq *rxq = &trans_pcie->rxq[queue];
1217 	u32 r, i, count = 0;
1218 	bool emergency = false;
1219 
1220 restart:
1221 	spin_lock(&rxq->lock);
1222 	/* uCode's read index (stored in shared DRAM) indicates the last Rx
1223 	 * buffer that the driver may process (last buffer filled by ucode). */
1224 	r = le16_to_cpu(ACCESS_ONCE(rxq->rb_stts->closed_rb_num)) & 0x0FFF;
1225 	i = rxq->read;
1226 
1227 	/* W/A 9000 device step A0 wrap-around bug */
1228 	r &= (rxq->queue_size - 1);
1229 
1230 	/* Rx interrupt, but nothing sent from uCode */
1231 	if (i == r)
1232 		IWL_DEBUG_RX(trans, "Q %d: HW = SW = %d\n", rxq->id, r);
1233 
1234 	while (i != r) {
1235 		struct iwl_rx_mem_buffer *rxb;
1236 
1237 		if (unlikely(rxq->used_count == rxq->queue_size / 2))
1238 			emergency = true;
1239 
1240 		if (trans->cfg->mq_rx_supported) {
1241 			/*
1242 			 * used_bd is a 32 bit but only 12 are used to retrieve
1243 			 * the vid
1244 			 */
1245 			u16 vid = le32_to_cpu(rxq->used_bd[i]) & 0x0FFF;
1246 
1247 			if (WARN(!vid ||
1248 				 vid > ARRAY_SIZE(trans_pcie->global_table),
1249 				 "Invalid rxb index from HW %u\n", (u32)vid)) {
1250 				iwl_force_nmi(trans);
1251 				goto out;
1252 			}
1253 			rxb = trans_pcie->global_table[vid - 1];
1254 			if (WARN(rxb->invalid,
1255 				 "Invalid rxb from HW %u\n", (u32)vid)) {
1256 				iwl_force_nmi(trans);
1257 				goto out;
1258 			}
1259 			rxb->invalid = true;
1260 		} else {
1261 			rxb = rxq->queue[i];
1262 			rxq->queue[i] = NULL;
1263 		}
1264 
1265 		IWL_DEBUG_RX(trans, "Q %d: HW = %d, SW = %d\n", rxq->id, r, i);
1266 		iwl_pcie_rx_handle_rb(trans, rxq, rxb, emergency);
1267 
1268 		i = (i + 1) & (rxq->queue_size - 1);
1269 
1270 		/*
1271 		 * If we have RX_CLAIM_REQ_ALLOC released rx buffers -
1272 		 * try to claim the pre-allocated buffers from the allocator.
1273 		 * If not ready - will try to reclaim next time.
1274 		 * There is no need to reschedule work - allocator exits only
1275 		 * on success
1276 		 */
1277 		if (rxq->used_count >= RX_CLAIM_REQ_ALLOC)
1278 			iwl_pcie_rx_allocator_get(trans, rxq);
1279 
1280 		if (rxq->used_count % RX_CLAIM_REQ_ALLOC == 0 && !emergency) {
1281 			struct iwl_rb_allocator *rba = &trans_pcie->rba;
1282 
1283 			/* Add the remaining empty RBDs for allocator use */
1284 			spin_lock(&rba->lock);
1285 			list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
1286 			spin_unlock(&rba->lock);
1287 		} else if (emergency) {
1288 			count++;
1289 			if (count == 8) {
1290 				count = 0;
1291 				if (rxq->used_count < rxq->queue_size / 3)
1292 					emergency = false;
1293 
1294 				rxq->read = i;
1295 				spin_unlock(&rxq->lock);
1296 				iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1297 				iwl_pcie_rxq_restock(trans, rxq);
1298 				goto restart;
1299 			}
1300 		}
1301 	}
1302 out:
1303 	/* Backtrack one entry */
1304 	rxq->read = i;
1305 	spin_unlock(&rxq->lock);
1306 
1307 	/*
1308 	 * handle a case where in emergency there are some unallocated RBDs.
1309 	 * those RBDs are in the used list, but are not tracked by the queue's
1310 	 * used_count which counts allocator owned RBDs.
1311 	 * unallocated emergency RBDs must be allocated on exit, otherwise
1312 	 * when called again the function may not be in emergency mode and
1313 	 * they will be handed to the allocator with no tracking in the RBD
1314 	 * allocator counters, which will lead to them never being claimed back
1315 	 * by the queue.
1316 	 * by allocating them here, they are now in the queue free list, and
1317 	 * will be restocked by the next call of iwl_pcie_rxq_restock.
1318 	 */
1319 	if (unlikely(emergency && count))
1320 		iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1321 
1322 	if (rxq->napi.poll)
1323 		napi_gro_flush(&rxq->napi, false);
1324 
1325 	iwl_pcie_rxq_restock(trans, rxq);
1326 }
1327 
1328 static struct iwl_trans_pcie *iwl_pcie_get_trans_pcie(struct msix_entry *entry)
1329 {
1330 	u8 queue = entry->entry;
1331 	struct msix_entry *entries = entry - queue;
1332 
1333 	return container_of(entries, struct iwl_trans_pcie, msix_entries[0]);
1334 }
1335 
1336 static inline void iwl_pcie_clear_irq(struct iwl_trans *trans,
1337 				      struct msix_entry *entry)
1338 {
1339 	/*
1340 	 * Before sending the interrupt the HW disables it to prevent
1341 	 * a nested interrupt. This is done by writing 1 to the corresponding
1342 	 * bit in the mask register. After handling the interrupt, it should be
1343 	 * re-enabled by clearing this bit. This register is defined as
1344 	 * write 1 clear (W1C) register, meaning that it's being clear
1345 	 * by writing 1 to the bit.
1346 	 */
1347 	iwl_write32(trans, CSR_MSIX_AUTOMASK_ST_AD, BIT(entry->entry));
1348 }
1349 
1350 /*
1351  * iwl_pcie_rx_msix_handle - Main entry function for receiving responses from fw
1352  * This interrupt handler should be used with RSS queue only.
1353  */
1354 irqreturn_t iwl_pcie_irq_rx_msix_handler(int irq, void *dev_id)
1355 {
1356 	struct msix_entry *entry = dev_id;
1357 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
1358 	struct iwl_trans *trans = trans_pcie->trans;
1359 
1360 	if (WARN_ON(entry->entry >= trans->num_rx_queues))
1361 		return IRQ_NONE;
1362 
1363 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1364 
1365 	local_bh_disable();
1366 	iwl_pcie_rx_handle(trans, entry->entry);
1367 	local_bh_enable();
1368 
1369 	iwl_pcie_clear_irq(trans, entry);
1370 
1371 	lock_map_release(&trans->sync_cmd_lockdep_map);
1372 
1373 	return IRQ_HANDLED;
1374 }
1375 
1376 /*
1377  * iwl_pcie_irq_handle_error - called for HW or SW error interrupt from card
1378  */
1379 static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
1380 {
1381 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1382 	int i;
1383 
1384 	/* W/A for WiFi/WiMAX coex and WiMAX own the RF */
1385 	if (trans->cfg->internal_wimax_coex &&
1386 	    !trans->cfg->apmg_not_supported &&
1387 	    (!(iwl_read_prph(trans, APMG_CLK_CTRL_REG) &
1388 			     APMS_CLK_VAL_MRB_FUNC_MODE) ||
1389 	     (iwl_read_prph(trans, APMG_PS_CTRL_REG) &
1390 			    APMG_PS_CTRL_VAL_RESET_REQ))) {
1391 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1392 		iwl_op_mode_wimax_active(trans->op_mode);
1393 		wake_up(&trans_pcie->wait_command_queue);
1394 		return;
1395 	}
1396 
1397 	iwl_pcie_dump_csr(trans);
1398 	iwl_dump_fh(trans, NULL);
1399 
1400 	local_bh_disable();
1401 	/* The STATUS_FW_ERROR bit is set in this function. This must happen
1402 	 * before we wake up the command caller, to ensure a proper cleanup. */
1403 	iwl_trans_fw_error(trans);
1404 	local_bh_enable();
1405 
1406 	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++)
1407 		del_timer(&trans_pcie->txq[i].stuck_timer);
1408 
1409 	clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1410 	wake_up(&trans_pcie->wait_command_queue);
1411 }
1412 
1413 static u32 iwl_pcie_int_cause_non_ict(struct iwl_trans *trans)
1414 {
1415 	u32 inta;
1416 
1417 	lockdep_assert_held(&IWL_TRANS_GET_PCIE_TRANS(trans)->irq_lock);
1418 
1419 	trace_iwlwifi_dev_irq(trans->dev);
1420 
1421 	/* Discover which interrupts are active/pending */
1422 	inta = iwl_read32(trans, CSR_INT);
1423 
1424 	/* the thread will service interrupts and re-enable them */
1425 	return inta;
1426 }
1427 
1428 /* a device (PCI-E) page is 4096 bytes long */
1429 #define ICT_SHIFT	12
1430 #define ICT_SIZE	(1 << ICT_SHIFT)
1431 #define ICT_COUNT	(ICT_SIZE / sizeof(u32))
1432 
1433 /* interrupt handler using ict table, with this interrupt driver will
1434  * stop using INTA register to get device's interrupt, reading this register
1435  * is expensive, device will write interrupts in ICT dram table, increment
1436  * index then will fire interrupt to driver, driver will OR all ICT table
1437  * entries from current index up to table entry with 0 value. the result is
1438  * the interrupt we need to service, driver will set the entries back to 0 and
1439  * set index.
1440  */
1441 static u32 iwl_pcie_int_cause_ict(struct iwl_trans *trans)
1442 {
1443 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1444 	u32 inta;
1445 	u32 val = 0;
1446 	u32 read;
1447 
1448 	trace_iwlwifi_dev_irq(trans->dev);
1449 
1450 	/* Ignore interrupt if there's nothing in NIC to service.
1451 	 * This may be due to IRQ shared with another device,
1452 	 * or due to sporadic interrupts thrown from our NIC. */
1453 	read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1454 	trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index, read);
1455 	if (!read)
1456 		return 0;
1457 
1458 	/*
1459 	 * Collect all entries up to the first 0, starting from ict_index;
1460 	 * note we already read at ict_index.
1461 	 */
1462 	do {
1463 		val |= read;
1464 		IWL_DEBUG_ISR(trans, "ICT index %d value 0x%08X\n",
1465 				trans_pcie->ict_index, read);
1466 		trans_pcie->ict_tbl[trans_pcie->ict_index] = 0;
1467 		trans_pcie->ict_index =
1468 			((trans_pcie->ict_index + 1) & (ICT_COUNT - 1));
1469 
1470 		read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1471 		trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index,
1472 					   read);
1473 	} while (read);
1474 
1475 	/* We should not get this value, just ignore it. */
1476 	if (val == 0xffffffff)
1477 		val = 0;
1478 
1479 	/*
1480 	 * this is a w/a for a h/w bug. the h/w bug may cause the Rx bit
1481 	 * (bit 15 before shifting it to 31) to clear when using interrupt
1482 	 * coalescing. fortunately, bits 18 and 19 stay set when this happens
1483 	 * so we use them to decide on the real state of the Rx bit.
1484 	 * In order words, bit 15 is set if bit 18 or bit 19 are set.
1485 	 */
1486 	if (val & 0xC0000)
1487 		val |= 0x8000;
1488 
1489 	inta = (0xff & val) | ((0xff00 & val) << 16);
1490 	return inta;
1491 }
1492 
1493 irqreturn_t iwl_pcie_irq_handler(int irq, void *dev_id)
1494 {
1495 	struct iwl_trans *trans = dev_id;
1496 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1497 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1498 	u32 inta = 0;
1499 	u32 handled = 0;
1500 
1501 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1502 
1503 	spin_lock(&trans_pcie->irq_lock);
1504 
1505 	/* dram interrupt table not set yet,
1506 	 * use legacy interrupt.
1507 	 */
1508 	if (likely(trans_pcie->use_ict))
1509 		inta = iwl_pcie_int_cause_ict(trans);
1510 	else
1511 		inta = iwl_pcie_int_cause_non_ict(trans);
1512 
1513 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1514 		IWL_DEBUG_ISR(trans,
1515 			      "ISR inta 0x%08x, enabled 0x%08x(sw), enabled(hw) 0x%08x, fh 0x%08x\n",
1516 			      inta, trans_pcie->inta_mask,
1517 			      iwl_read32(trans, CSR_INT_MASK),
1518 			      iwl_read32(trans, CSR_FH_INT_STATUS));
1519 		if (inta & (~trans_pcie->inta_mask))
1520 			IWL_DEBUG_ISR(trans,
1521 				      "We got a masked interrupt (0x%08x)\n",
1522 				      inta & (~trans_pcie->inta_mask));
1523 	}
1524 
1525 	inta &= trans_pcie->inta_mask;
1526 
1527 	/*
1528 	 * Ignore interrupt if there's nothing in NIC to service.
1529 	 * This may be due to IRQ shared with another device,
1530 	 * or due to sporadic interrupts thrown from our NIC.
1531 	 */
1532 	if (unlikely(!inta)) {
1533 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
1534 		/*
1535 		 * Re-enable interrupts here since we don't
1536 		 * have anything to service
1537 		 */
1538 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1539 			_iwl_enable_interrupts(trans);
1540 		spin_unlock(&trans_pcie->irq_lock);
1541 		lock_map_release(&trans->sync_cmd_lockdep_map);
1542 		return IRQ_NONE;
1543 	}
1544 
1545 	if (unlikely(inta == 0xFFFFFFFF || (inta & 0xFFFFFFF0) == 0xa5a5a5a0)) {
1546 		/*
1547 		 * Hardware disappeared. It might have
1548 		 * already raised an interrupt.
1549 		 */
1550 		IWL_WARN(trans, "HARDWARE GONE?? INTA == 0x%08x\n", inta);
1551 		spin_unlock(&trans_pcie->irq_lock);
1552 		goto out;
1553 	}
1554 
1555 	/* Ack/clear/reset pending uCode interrupts.
1556 	 * Note:  Some bits in CSR_INT are "OR" of bits in CSR_FH_INT_STATUS,
1557 	 */
1558 	/* There is a hardware bug in the interrupt mask function that some
1559 	 * interrupts (i.e. CSR_INT_BIT_SCD) can still be generated even if
1560 	 * they are disabled in the CSR_INT_MASK register. Furthermore the
1561 	 * ICT interrupt handling mechanism has another bug that might cause
1562 	 * these unmasked interrupts fail to be detected. We workaround the
1563 	 * hardware bugs here by ACKing all the possible interrupts so that
1564 	 * interrupt coalescing can still be achieved.
1565 	 */
1566 	iwl_write32(trans, CSR_INT, inta | ~trans_pcie->inta_mask);
1567 
1568 	if (iwl_have_debug_level(IWL_DL_ISR))
1569 		IWL_DEBUG_ISR(trans, "inta 0x%08x, enabled 0x%08x\n",
1570 			      inta, iwl_read32(trans, CSR_INT_MASK));
1571 
1572 	spin_unlock(&trans_pcie->irq_lock);
1573 
1574 	/* Now service all interrupt bits discovered above. */
1575 	if (inta & CSR_INT_BIT_HW_ERR) {
1576 		IWL_ERR(trans, "Hardware error detected.  Restarting.\n");
1577 
1578 		/* Tell the device to stop sending interrupts */
1579 		iwl_disable_interrupts(trans);
1580 
1581 		isr_stats->hw++;
1582 		iwl_pcie_irq_handle_error(trans);
1583 
1584 		handled |= CSR_INT_BIT_HW_ERR;
1585 
1586 		goto out;
1587 	}
1588 
1589 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1590 		/* NIC fires this, but we don't use it, redundant with WAKEUP */
1591 		if (inta & CSR_INT_BIT_SCD) {
1592 			IWL_DEBUG_ISR(trans,
1593 				      "Scheduler finished to transmit the frame/frames.\n");
1594 			isr_stats->sch++;
1595 		}
1596 
1597 		/* Alive notification via Rx interrupt will do the real work */
1598 		if (inta & CSR_INT_BIT_ALIVE) {
1599 			IWL_DEBUG_ISR(trans, "Alive interrupt\n");
1600 			isr_stats->alive++;
1601 		}
1602 	}
1603 
1604 	/* Safely ignore these bits for debug checks below */
1605 	inta &= ~(CSR_INT_BIT_SCD | CSR_INT_BIT_ALIVE);
1606 
1607 	/* HW RF KILL switch toggled */
1608 	if (inta & CSR_INT_BIT_RF_KILL) {
1609 		bool hw_rfkill;
1610 
1611 		hw_rfkill = iwl_is_rfkill_set(trans);
1612 		IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
1613 			 hw_rfkill ? "disable radio" : "enable radio");
1614 
1615 		isr_stats->rfkill++;
1616 
1617 		mutex_lock(&trans_pcie->mutex);
1618 		iwl_trans_pcie_rf_kill(trans, hw_rfkill);
1619 		mutex_unlock(&trans_pcie->mutex);
1620 		if (hw_rfkill) {
1621 			set_bit(STATUS_RFKILL, &trans->status);
1622 			if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
1623 					       &trans->status))
1624 				IWL_DEBUG_RF_KILL(trans,
1625 						  "Rfkill while SYNC HCMD in flight\n");
1626 			wake_up(&trans_pcie->wait_command_queue);
1627 		} else {
1628 			clear_bit(STATUS_RFKILL, &trans->status);
1629 		}
1630 
1631 		handled |= CSR_INT_BIT_RF_KILL;
1632 	}
1633 
1634 	/* Chip got too hot and stopped itself */
1635 	if (inta & CSR_INT_BIT_CT_KILL) {
1636 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
1637 		isr_stats->ctkill++;
1638 		handled |= CSR_INT_BIT_CT_KILL;
1639 	}
1640 
1641 	/* Error detected by uCode */
1642 	if (inta & CSR_INT_BIT_SW_ERR) {
1643 		IWL_ERR(trans, "Microcode SW error detected. "
1644 			" Restarting 0x%X.\n", inta);
1645 		isr_stats->sw++;
1646 		iwl_pcie_irq_handle_error(trans);
1647 		handled |= CSR_INT_BIT_SW_ERR;
1648 	}
1649 
1650 	/* uCode wakes up after power-down sleep */
1651 	if (inta & CSR_INT_BIT_WAKEUP) {
1652 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
1653 		iwl_pcie_rxq_check_wrptr(trans);
1654 		iwl_pcie_txq_check_wrptrs(trans);
1655 
1656 		isr_stats->wakeup++;
1657 
1658 		handled |= CSR_INT_BIT_WAKEUP;
1659 	}
1660 
1661 	/* All uCode command responses, including Tx command responses,
1662 	 * Rx "responses" (frame-received notification), and other
1663 	 * notifications from uCode come through here*/
1664 	if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX |
1665 		    CSR_INT_BIT_RX_PERIODIC)) {
1666 		IWL_DEBUG_ISR(trans, "Rx interrupt\n");
1667 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX)) {
1668 			handled |= (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX);
1669 			iwl_write32(trans, CSR_FH_INT_STATUS,
1670 					CSR_FH_INT_RX_MASK);
1671 		}
1672 		if (inta & CSR_INT_BIT_RX_PERIODIC) {
1673 			handled |= CSR_INT_BIT_RX_PERIODIC;
1674 			iwl_write32(trans,
1675 				CSR_INT, CSR_INT_BIT_RX_PERIODIC);
1676 		}
1677 		/* Sending RX interrupt require many steps to be done in the
1678 		 * the device:
1679 		 * 1- write interrupt to current index in ICT table.
1680 		 * 2- dma RX frame.
1681 		 * 3- update RX shared data to indicate last write index.
1682 		 * 4- send interrupt.
1683 		 * This could lead to RX race, driver could receive RX interrupt
1684 		 * but the shared data changes does not reflect this;
1685 		 * periodic interrupt will detect any dangling Rx activity.
1686 		 */
1687 
1688 		/* Disable periodic interrupt; we use it as just a one-shot. */
1689 		iwl_write8(trans, CSR_INT_PERIODIC_REG,
1690 			    CSR_INT_PERIODIC_DIS);
1691 
1692 		/*
1693 		 * Enable periodic interrupt in 8 msec only if we received
1694 		 * real RX interrupt (instead of just periodic int), to catch
1695 		 * any dangling Rx interrupt.  If it was just the periodic
1696 		 * interrupt, there was no dangling Rx activity, and no need
1697 		 * to extend the periodic interrupt; one-shot is enough.
1698 		 */
1699 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX))
1700 			iwl_write8(trans, CSR_INT_PERIODIC_REG,
1701 				   CSR_INT_PERIODIC_ENA);
1702 
1703 		isr_stats->rx++;
1704 
1705 		local_bh_disable();
1706 		iwl_pcie_rx_handle(trans, 0);
1707 		local_bh_enable();
1708 	}
1709 
1710 	/* This "Tx" DMA channel is used only for loading uCode */
1711 	if (inta & CSR_INT_BIT_FH_TX) {
1712 		iwl_write32(trans, CSR_FH_INT_STATUS, CSR_FH_INT_TX_MASK);
1713 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
1714 		isr_stats->tx++;
1715 		handled |= CSR_INT_BIT_FH_TX;
1716 		/* Wake up uCode load routine, now that load is complete */
1717 		trans_pcie->ucode_write_complete = true;
1718 		wake_up(&trans_pcie->ucode_write_waitq);
1719 	}
1720 
1721 	if (inta & ~handled) {
1722 		IWL_ERR(trans, "Unhandled INTA bits 0x%08x\n", inta & ~handled);
1723 		isr_stats->unhandled++;
1724 	}
1725 
1726 	if (inta & ~(trans_pcie->inta_mask)) {
1727 		IWL_WARN(trans, "Disabled INTA bits 0x%08x were pending\n",
1728 			 inta & ~trans_pcie->inta_mask);
1729 	}
1730 
1731 	spin_lock(&trans_pcie->irq_lock);
1732 	/* only Re-enable all interrupt if disabled by irq */
1733 	if (test_bit(STATUS_INT_ENABLED, &trans->status))
1734 		_iwl_enable_interrupts(trans);
1735 	/* we are loading the firmware, enable FH_TX interrupt only */
1736 	else if (handled & CSR_INT_BIT_FH_TX)
1737 		iwl_enable_fw_load_int(trans);
1738 	/* Re-enable RF_KILL if it occurred */
1739 	else if (handled & CSR_INT_BIT_RF_KILL)
1740 		iwl_enable_rfkill_int(trans);
1741 	spin_unlock(&trans_pcie->irq_lock);
1742 
1743 out:
1744 	lock_map_release(&trans->sync_cmd_lockdep_map);
1745 	return IRQ_HANDLED;
1746 }
1747 
1748 /******************************************************************************
1749  *
1750  * ICT functions
1751  *
1752  ******************************************************************************/
1753 
1754 /* Free dram table */
1755 void iwl_pcie_free_ict(struct iwl_trans *trans)
1756 {
1757 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1758 
1759 	if (trans_pcie->ict_tbl) {
1760 		dma_free_coherent(trans->dev, ICT_SIZE,
1761 				  trans_pcie->ict_tbl,
1762 				  trans_pcie->ict_tbl_dma);
1763 		trans_pcie->ict_tbl = NULL;
1764 		trans_pcie->ict_tbl_dma = 0;
1765 	}
1766 }
1767 
1768 /*
1769  * allocate dram shared table, it is an aligned memory
1770  * block of ICT_SIZE.
1771  * also reset all data related to ICT table interrupt.
1772  */
1773 int iwl_pcie_alloc_ict(struct iwl_trans *trans)
1774 {
1775 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1776 
1777 	trans_pcie->ict_tbl =
1778 		dma_zalloc_coherent(trans->dev, ICT_SIZE,
1779 				   &trans_pcie->ict_tbl_dma,
1780 				   GFP_KERNEL);
1781 	if (!trans_pcie->ict_tbl)
1782 		return -ENOMEM;
1783 
1784 	/* just an API sanity check ... it is guaranteed to be aligned */
1785 	if (WARN_ON(trans_pcie->ict_tbl_dma & (ICT_SIZE - 1))) {
1786 		iwl_pcie_free_ict(trans);
1787 		return -EINVAL;
1788 	}
1789 
1790 	return 0;
1791 }
1792 
1793 /* Device is going up inform it about using ICT interrupt table,
1794  * also we need to tell the driver to start using ICT interrupt.
1795  */
1796 void iwl_pcie_reset_ict(struct iwl_trans *trans)
1797 {
1798 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1799 	u32 val;
1800 
1801 	if (!trans_pcie->ict_tbl)
1802 		return;
1803 
1804 	spin_lock(&trans_pcie->irq_lock);
1805 	_iwl_disable_interrupts(trans);
1806 
1807 	memset(trans_pcie->ict_tbl, 0, ICT_SIZE);
1808 
1809 	val = trans_pcie->ict_tbl_dma >> ICT_SHIFT;
1810 
1811 	val |= CSR_DRAM_INT_TBL_ENABLE |
1812 	       CSR_DRAM_INIT_TBL_WRAP_CHECK |
1813 	       CSR_DRAM_INIT_TBL_WRITE_POINTER;
1814 
1815 	IWL_DEBUG_ISR(trans, "CSR_DRAM_INT_TBL_REG =0x%x\n", val);
1816 
1817 	iwl_write32(trans, CSR_DRAM_INT_TBL_REG, val);
1818 	trans_pcie->use_ict = true;
1819 	trans_pcie->ict_index = 0;
1820 	iwl_write32(trans, CSR_INT, trans_pcie->inta_mask);
1821 	_iwl_enable_interrupts(trans);
1822 	spin_unlock(&trans_pcie->irq_lock);
1823 }
1824 
1825 /* Device is going down disable ict interrupt usage */
1826 void iwl_pcie_disable_ict(struct iwl_trans *trans)
1827 {
1828 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1829 
1830 	spin_lock(&trans_pcie->irq_lock);
1831 	trans_pcie->use_ict = false;
1832 	spin_unlock(&trans_pcie->irq_lock);
1833 }
1834 
1835 irqreturn_t iwl_pcie_isr(int irq, void *data)
1836 {
1837 	struct iwl_trans *trans = data;
1838 
1839 	if (!trans)
1840 		return IRQ_NONE;
1841 
1842 	/* Disable (but don't clear!) interrupts here to avoid
1843 	 * back-to-back ISRs and sporadic interrupts from our NIC.
1844 	 * If we have something to service, the tasklet will re-enable ints.
1845 	 * If we *don't* have something, we'll re-enable before leaving here.
1846 	 */
1847 	iwl_write32(trans, CSR_INT_MASK, 0x00000000);
1848 
1849 	return IRQ_WAKE_THREAD;
1850 }
1851 
1852 irqreturn_t iwl_pcie_msix_isr(int irq, void *data)
1853 {
1854 	return IRQ_WAKE_THREAD;
1855 }
1856 
1857 irqreturn_t iwl_pcie_irq_msix_handler(int irq, void *dev_id)
1858 {
1859 	struct msix_entry *entry = dev_id;
1860 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
1861 	struct iwl_trans *trans = trans_pcie->trans;
1862 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1863 	u32 inta_fh, inta_hw;
1864 
1865 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1866 
1867 	spin_lock(&trans_pcie->irq_lock);
1868 	inta_fh = iwl_read32(trans, CSR_MSIX_FH_INT_CAUSES_AD);
1869 	inta_hw = iwl_read32(trans, CSR_MSIX_HW_INT_CAUSES_AD);
1870 	/*
1871 	 * Clear causes registers to avoid being handling the same cause.
1872 	 */
1873 	iwl_write32(trans, CSR_MSIX_FH_INT_CAUSES_AD, inta_fh);
1874 	iwl_write32(trans, CSR_MSIX_HW_INT_CAUSES_AD, inta_hw);
1875 	spin_unlock(&trans_pcie->irq_lock);
1876 
1877 	if (unlikely(!(inta_fh | inta_hw))) {
1878 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
1879 		lock_map_release(&trans->sync_cmd_lockdep_map);
1880 		return IRQ_NONE;
1881 	}
1882 
1883 	if (iwl_have_debug_level(IWL_DL_ISR))
1884 		IWL_DEBUG_ISR(trans, "ISR inta_fh 0x%08x, enabled 0x%08x\n",
1885 			      inta_fh,
1886 			      iwl_read32(trans, CSR_MSIX_FH_INT_MASK_AD));
1887 
1888 	/* This "Tx" DMA channel is used only for loading uCode */
1889 	if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM) {
1890 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
1891 		isr_stats->tx++;
1892 		/*
1893 		 * Wake up uCode load routine,
1894 		 * now that load is complete
1895 		 */
1896 		trans_pcie->ucode_write_complete = true;
1897 		wake_up(&trans_pcie->ucode_write_waitq);
1898 	}
1899 
1900 	/* Error detected by uCode */
1901 	if ((inta_fh & MSIX_FH_INT_CAUSES_FH_ERR) ||
1902 	    (inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR)) {
1903 		IWL_ERR(trans,
1904 			"Microcode SW error detected. Restarting 0x%X.\n",
1905 			inta_fh);
1906 		isr_stats->sw++;
1907 		iwl_pcie_irq_handle_error(trans);
1908 	}
1909 
1910 	/* After checking FH register check HW register */
1911 	if (iwl_have_debug_level(IWL_DL_ISR))
1912 		IWL_DEBUG_ISR(trans,
1913 			      "ISR inta_hw 0x%08x, enabled 0x%08x\n",
1914 			      inta_hw,
1915 			      iwl_read32(trans, CSR_MSIX_HW_INT_MASK_AD));
1916 
1917 	/* Alive notification via Rx interrupt will do the real work */
1918 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_ALIVE) {
1919 		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
1920 		isr_stats->alive++;
1921 	}
1922 
1923 	/* uCode wakes up after power-down sleep */
1924 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_WAKEUP) {
1925 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
1926 		iwl_pcie_rxq_check_wrptr(trans);
1927 		iwl_pcie_txq_check_wrptrs(trans);
1928 
1929 		isr_stats->wakeup++;
1930 	}
1931 
1932 	/* Chip got too hot and stopped itself */
1933 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_CT_KILL) {
1934 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
1935 		isr_stats->ctkill++;
1936 	}
1937 
1938 	/* HW RF KILL switch toggled */
1939 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RF_KILL) {
1940 		bool hw_rfkill;
1941 
1942 		hw_rfkill = iwl_is_rfkill_set(trans);
1943 		IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
1944 			 hw_rfkill ? "disable radio" : "enable radio");
1945 
1946 		isr_stats->rfkill++;
1947 
1948 		mutex_lock(&trans_pcie->mutex);
1949 		iwl_trans_pcie_rf_kill(trans, hw_rfkill);
1950 		mutex_unlock(&trans_pcie->mutex);
1951 		if (hw_rfkill) {
1952 			set_bit(STATUS_RFKILL, &trans->status);
1953 			if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
1954 					       &trans->status))
1955 				IWL_DEBUG_RF_KILL(trans,
1956 						  "Rfkill while SYNC HCMD in flight\n");
1957 			wake_up(&trans_pcie->wait_command_queue);
1958 		} else {
1959 			clear_bit(STATUS_RFKILL, &trans->status);
1960 		}
1961 	}
1962 
1963 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_HW_ERR) {
1964 		IWL_ERR(trans,
1965 			"Hardware error detected. Restarting.\n");
1966 
1967 		isr_stats->hw++;
1968 		iwl_pcie_irq_handle_error(trans);
1969 	}
1970 
1971 	iwl_pcie_clear_irq(trans, entry);
1972 
1973 	lock_map_release(&trans->sync_cmd_lockdep_map);
1974 
1975 	return IRQ_HANDLED;
1976 }
1977