1 /******************************************************************************
2  *
3  * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
4  * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
5  *
6  * Portions of this file are derived from the ipw3945 project, as well
7  * as portions of the ieee80211 subsystem header files.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of version 2 of the GNU General Public License as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful, but WITHOUT
14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
16  * more details.
17  *
18  * You should have received a copy of the GNU General Public License along with
19  * this program; if not, write to the Free Software Foundation, Inc.,
20  * 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA
21  *
22  * The full GNU General Public License is included in this distribution in the
23  * file called LICENSE.
24  *
25  * Contact Information:
26  *  Intel Linux Wireless <linuxwifi@intel.com>
27  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
28  *
29  *****************************************************************************/
30 #include <linux/sched.h>
31 #include <linux/wait.h>
32 #include <linux/gfp.h>
33 
34 #include "iwl-prph.h"
35 #include "iwl-io.h"
36 #include "internal.h"
37 #include "iwl-op-mode.h"
38 
39 /******************************************************************************
40  *
41  * RX path functions
42  *
43  ******************************************************************************/
44 
45 /*
46  * Rx theory of operation
47  *
48  * Driver allocates a circular buffer of Receive Buffer Descriptors (RBDs),
49  * each of which point to Receive Buffers to be filled by the NIC.  These get
50  * used not only for Rx frames, but for any command response or notification
51  * from the NIC.  The driver and NIC manage the Rx buffers by means
52  * of indexes into the circular buffer.
53  *
54  * Rx Queue Indexes
55  * The host/firmware share two index registers for managing the Rx buffers.
56  *
57  * The READ index maps to the first position that the firmware may be writing
58  * to -- the driver can read up to (but not including) this position and get
59  * good data.
60  * The READ index is managed by the firmware once the card is enabled.
61  *
62  * The WRITE index maps to the last position the driver has read from -- the
63  * position preceding WRITE is the last slot the firmware can place a packet.
64  *
65  * The queue is empty (no good data) if WRITE = READ - 1, and is full if
66  * WRITE = READ.
67  *
68  * During initialization, the host sets up the READ queue position to the first
69  * INDEX position, and WRITE to the last (READ - 1 wrapped)
70  *
71  * When the firmware places a packet in a buffer, it will advance the READ index
72  * and fire the RX interrupt.  The driver can then query the READ index and
73  * process as many packets as possible, moving the WRITE index forward as it
74  * resets the Rx queue buffers with new memory.
75  *
76  * The management in the driver is as follows:
77  * + A list of pre-allocated RBDs is stored in iwl->rxq->rx_free.
78  *   When the interrupt handler is called, the request is processed.
79  *   The page is either stolen - transferred to the upper layer
80  *   or reused - added immediately to the iwl->rxq->rx_free list.
81  * + When the page is stolen - the driver updates the matching queue's used
82  *   count, detaches the RBD and transfers it to the queue used list.
83  *   When there are two used RBDs - they are transferred to the allocator empty
84  *   list. Work is then scheduled for the allocator to start allocating
85  *   eight buffers.
86  *   When there are another 6 used RBDs - they are transferred to the allocator
87  *   empty list and the driver tries to claim the pre-allocated buffers and
88  *   add them to iwl->rxq->rx_free. If it fails - it continues to claim them
89  *   until ready.
90  *   When there are 8+ buffers in the free list - either from allocation or from
91  *   8 reused unstolen pages - restock is called to update the FW and indexes.
92  * + In order to make sure the allocator always has RBDs to use for allocation
93  *   the allocator has initial pool in the size of num_queues*(8-2) - the
94  *   maximum missing RBDs per allocation request (request posted with 2
95  *    empty RBDs, there is no guarantee when the other 6 RBDs are supplied).
96  *   The queues supplies the recycle of the rest of the RBDs.
97  * + A received packet is processed and handed to the kernel network stack,
98  *   detached from the iwl->rxq.  The driver 'processed' index is updated.
99  * + If there are no allocated buffers in iwl->rxq->rx_free,
100  *   the READ INDEX is not incremented and iwl->status(RX_STALLED) is set.
101  *   If there were enough free buffers and RX_STALLED is set it is cleared.
102  *
103  *
104  * Driver sequence:
105  *
106  * iwl_rxq_alloc()            Allocates rx_free
107  * iwl_pcie_rx_replenish()    Replenishes rx_free list from rx_used, and calls
108  *                            iwl_pcie_rxq_restock.
109  *                            Used only during initialization.
110  * iwl_pcie_rxq_restock()     Moves available buffers from rx_free into Rx
111  *                            queue, updates firmware pointers, and updates
112  *                            the WRITE index.
113  * iwl_pcie_rx_allocator()     Background work for allocating pages.
114  *
115  * -- enable interrupts --
116  * ISR - iwl_rx()             Detach iwl_rx_mem_buffers from pool up to the
117  *                            READ INDEX, detaching the SKB from the pool.
118  *                            Moves the packet buffer from queue to rx_used.
119  *                            Posts and claims requests to the allocator.
120  *                            Calls iwl_pcie_rxq_restock to refill any empty
121  *                            slots.
122  *
123  * RBD life-cycle:
124  *
125  * Init:
126  * rxq.pool -> rxq.rx_used -> rxq.rx_free -> rxq.queue
127  *
128  * Regular Receive interrupt:
129  * Page Stolen:
130  * rxq.queue -> rxq.rx_used -> allocator.rbd_empty ->
131  * allocator.rbd_allocated -> rxq.rx_free -> rxq.queue
132  * Page not Stolen:
133  * rxq.queue -> rxq.rx_free -> rxq.queue
134  * ...
135  *
136  */
137 
138 /*
139  * iwl_rxq_space - Return number of free slots available in queue.
140  */
141 static int iwl_rxq_space(const struct iwl_rxq *rxq)
142 {
143 	/* Make sure RX_QUEUE_SIZE is a power of 2 */
144 	BUILD_BUG_ON(RX_QUEUE_SIZE & (RX_QUEUE_SIZE - 1));
145 
146 	/*
147 	 * There can be up to (RX_QUEUE_SIZE - 1) free slots, to avoid ambiguity
148 	 * between empty and completely full queues.
149 	 * The following is equivalent to modulo by RX_QUEUE_SIZE and is well
150 	 * defined for negative dividends.
151 	 */
152 	return (rxq->read - rxq->write - 1) & (RX_QUEUE_SIZE - 1);
153 }
154 
155 /*
156  * iwl_dma_addr2rbd_ptr - convert a DMA address to a uCode read buffer ptr
157  */
158 static inline __le32 iwl_pcie_dma_addr2rbd_ptr(dma_addr_t dma_addr)
159 {
160 	return cpu_to_le32((u32)(dma_addr >> 8));
161 }
162 
163 /*
164  * iwl_pcie_rx_stop - stops the Rx DMA
165  */
166 int iwl_pcie_rx_stop(struct iwl_trans *trans)
167 {
168 	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
169 	return iwl_poll_direct_bit(trans, FH_MEM_RSSR_RX_STATUS_REG,
170 				   FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE, 1000);
171 }
172 
173 /*
174  * iwl_pcie_rxq_inc_wr_ptr - Update the write pointer for the RX queue
175  */
176 static void iwl_pcie_rxq_inc_wr_ptr(struct iwl_trans *trans)
177 {
178 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
179 	struct iwl_rxq *rxq = &trans_pcie->rxq;
180 	u32 reg;
181 
182 	lockdep_assert_held(&rxq->lock);
183 
184 	/*
185 	 * explicitly wake up the NIC if:
186 	 * 1. shadow registers aren't enabled
187 	 * 2. there is a chance that the NIC is asleep
188 	 */
189 	if (!trans->cfg->base_params->shadow_reg_enable &&
190 	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
191 		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
192 
193 		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
194 			IWL_DEBUG_INFO(trans, "Rx queue requesting wakeup, GP1 = 0x%x\n",
195 				       reg);
196 			iwl_set_bit(trans, CSR_GP_CNTRL,
197 				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
198 			rxq->need_update = true;
199 			return;
200 		}
201 	}
202 
203 	rxq->write_actual = round_down(rxq->write, 8);
204 	iwl_write32(trans, FH_RSCSR_CHNL0_WPTR, rxq->write_actual);
205 }
206 
207 static void iwl_pcie_rxq_check_wrptr(struct iwl_trans *trans)
208 {
209 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
210 	struct iwl_rxq *rxq = &trans_pcie->rxq;
211 
212 	spin_lock(&rxq->lock);
213 
214 	if (!rxq->need_update)
215 		goto exit_unlock;
216 
217 	iwl_pcie_rxq_inc_wr_ptr(trans);
218 	rxq->need_update = false;
219 
220  exit_unlock:
221 	spin_unlock(&rxq->lock);
222 }
223 
224 /*
225  * iwl_pcie_rxq_restock - refill RX queue from pre-allocated pool
226  *
227  * If there are slots in the RX queue that need to be restocked,
228  * and we have free pre-allocated buffers, fill the ranks as much
229  * as we can, pulling from rx_free.
230  *
231  * This moves the 'write' index forward to catch up with 'processed', and
232  * also updates the memory address in the firmware to reference the new
233  * target buffer.
234  */
235 static void iwl_pcie_rxq_restock(struct iwl_trans *trans)
236 {
237 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
238 	struct iwl_rxq *rxq = &trans_pcie->rxq;
239 	struct iwl_rx_mem_buffer *rxb;
240 
241 	/*
242 	 * If the device isn't enabled - not need to try to add buffers...
243 	 * This can happen when we stop the device and still have an interrupt
244 	 * pending. We stop the APM before we sync the interrupts because we
245 	 * have to (see comment there). On the other hand, since the APM is
246 	 * stopped, we cannot access the HW (in particular not prph).
247 	 * So don't try to restock if the APM has been already stopped.
248 	 */
249 	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
250 		return;
251 
252 	spin_lock(&rxq->lock);
253 	while ((iwl_rxq_space(rxq) > 0) && (rxq->free_count)) {
254 		/* The overwritten rxb must be a used one */
255 		rxb = rxq->queue[rxq->write];
256 		BUG_ON(rxb && rxb->page);
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 
263 		/* Point to Rx buffer via next RBD in circular buffer */
264 		rxq->bd[rxq->write] = iwl_pcie_dma_addr2rbd_ptr(rxb->page_dma);
265 		rxq->queue[rxq->write] = rxb;
266 		rxq->write = (rxq->write + 1) & RX_QUEUE_MASK;
267 		rxq->free_count--;
268 	}
269 	spin_unlock(&rxq->lock);
270 
271 	/* If we've added more space for the firmware to place data, tell it.
272 	 * Increment device's write pointer in multiples of 8. */
273 	if (rxq->write_actual != (rxq->write & ~0x7)) {
274 		spin_lock(&rxq->lock);
275 		iwl_pcie_rxq_inc_wr_ptr(trans);
276 		spin_unlock(&rxq->lock);
277 	}
278 }
279 
280 /*
281  * iwl_pcie_rx_alloc_page - allocates and returns a page.
282  *
283  */
284 static struct page *iwl_pcie_rx_alloc_page(struct iwl_trans *trans,
285 					   gfp_t priority)
286 {
287 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
288 	struct iwl_rxq *rxq = &trans_pcie->rxq;
289 	struct page *page;
290 	gfp_t gfp_mask = priority;
291 
292 	if (rxq->free_count > RX_LOW_WATERMARK)
293 		gfp_mask |= __GFP_NOWARN;
294 
295 	if (trans_pcie->rx_page_order > 0)
296 		gfp_mask |= __GFP_COMP;
297 
298 	/* Alloc a new receive buffer */
299 	page = alloc_pages(gfp_mask, trans_pcie->rx_page_order);
300 	if (!page) {
301 		if (net_ratelimit())
302 			IWL_DEBUG_INFO(trans, "alloc_pages failed, order: %d\n",
303 				       trans_pcie->rx_page_order);
304 		/* Issue an error if the hardware has consumed more than half
305 		 * of its free buffer list and we don't have enough
306 		 * pre-allocated buffers.
307 `		 */
308 		if (rxq->free_count <= RX_LOW_WATERMARK &&
309 		    iwl_rxq_space(rxq) > (RX_QUEUE_SIZE / 2) &&
310 		    net_ratelimit())
311 			IWL_CRIT(trans,
312 				 "Failed to alloc_pages with GFP_KERNEL. Only %u free buffers remaining.\n",
313 				 rxq->free_count);
314 		return NULL;
315 	}
316 	return page;
317 }
318 
319 /*
320  * iwl_pcie_rxq_alloc_rbs - allocate a page for each used RBD
321  *
322  * A used RBD is an Rx buffer that has been given to the stack. To use it again
323  * a page must be allocated and the RBD must point to the page. This function
324  * doesn't change the HW pointer but handles the list of pages that is used by
325  * iwl_pcie_rxq_restock. The latter function will update the HW to use the newly
326  * allocated buffers.
327  */
328 static void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans, gfp_t priority)
329 {
330 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
331 	struct iwl_rxq *rxq = &trans_pcie->rxq;
332 	struct iwl_rx_mem_buffer *rxb;
333 	struct page *page;
334 
335 	while (1) {
336 		spin_lock(&rxq->lock);
337 		if (list_empty(&rxq->rx_used)) {
338 			spin_unlock(&rxq->lock);
339 			return;
340 		}
341 		spin_unlock(&rxq->lock);
342 
343 		/* Alloc a new receive buffer */
344 		page = iwl_pcie_rx_alloc_page(trans, priority);
345 		if (!page)
346 			return;
347 
348 		spin_lock(&rxq->lock);
349 
350 		if (list_empty(&rxq->rx_used)) {
351 			spin_unlock(&rxq->lock);
352 			__free_pages(page, trans_pcie->rx_page_order);
353 			return;
354 		}
355 		rxb = list_first_entry(&rxq->rx_used, struct iwl_rx_mem_buffer,
356 				       list);
357 		list_del(&rxb->list);
358 		spin_unlock(&rxq->lock);
359 
360 		BUG_ON(rxb->page);
361 		rxb->page = page;
362 		/* Get physical address of the RB */
363 		rxb->page_dma =
364 			dma_map_page(trans->dev, page, 0,
365 				     PAGE_SIZE << trans_pcie->rx_page_order,
366 				     DMA_FROM_DEVICE);
367 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
368 			rxb->page = NULL;
369 			spin_lock(&rxq->lock);
370 			list_add(&rxb->list, &rxq->rx_used);
371 			spin_unlock(&rxq->lock);
372 			__free_pages(page, trans_pcie->rx_page_order);
373 			return;
374 		}
375 		/* dma address must be no more than 36 bits */
376 		BUG_ON(rxb->page_dma & ~DMA_BIT_MASK(36));
377 		/* and also 256 byte aligned! */
378 		BUG_ON(rxb->page_dma & DMA_BIT_MASK(8));
379 
380 		spin_lock(&rxq->lock);
381 
382 		list_add_tail(&rxb->list, &rxq->rx_free);
383 		rxq->free_count++;
384 
385 		spin_unlock(&rxq->lock);
386 	}
387 }
388 
389 static void iwl_pcie_rxq_free_rbs(struct iwl_trans *trans)
390 {
391 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
392 	struct iwl_rxq *rxq = &trans_pcie->rxq;
393 	int i;
394 
395 	lockdep_assert_held(&rxq->lock);
396 
397 	for (i = 0; i < RX_QUEUE_SIZE; i++) {
398 		if (!rxq->pool[i].page)
399 			continue;
400 		dma_unmap_page(trans->dev, rxq->pool[i].page_dma,
401 			       PAGE_SIZE << trans_pcie->rx_page_order,
402 			       DMA_FROM_DEVICE);
403 		__free_pages(rxq->pool[i].page, trans_pcie->rx_page_order);
404 		rxq->pool[i].page = NULL;
405 	}
406 }
407 
408 /*
409  * iwl_pcie_rx_replenish - Move all used buffers from rx_used to rx_free
410  *
411  * When moving to rx_free an page is allocated for the slot.
412  *
413  * Also restock the Rx queue via iwl_pcie_rxq_restock.
414  * This is called only during initialization
415  */
416 static void iwl_pcie_rx_replenish(struct iwl_trans *trans)
417 {
418 	iwl_pcie_rxq_alloc_rbs(trans, GFP_KERNEL);
419 
420 	iwl_pcie_rxq_restock(trans);
421 }
422 
423 /*
424  * iwl_pcie_rx_allocator - Allocates pages in the background for RX queues
425  *
426  * Allocates for each received request 8 pages
427  * Called as a scheduled work item.
428  */
429 static void iwl_pcie_rx_allocator(struct iwl_trans *trans)
430 {
431 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
432 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
433 	struct list_head local_empty;
434 	int pending = atomic_xchg(&rba->req_pending, 0);
435 
436 	IWL_DEBUG_RX(trans, "Pending allocation requests = %d\n", pending);
437 
438 	/* If we were scheduled - there is at least one request */
439 	spin_lock(&rba->lock);
440 	/* swap out the rba->rbd_empty to a local list */
441 	list_replace_init(&rba->rbd_empty, &local_empty);
442 	spin_unlock(&rba->lock);
443 
444 	while (pending) {
445 		int i;
446 		struct list_head local_allocated;
447 
448 		INIT_LIST_HEAD(&local_allocated);
449 
450 		for (i = 0; i < RX_CLAIM_REQ_ALLOC;) {
451 			struct iwl_rx_mem_buffer *rxb;
452 			struct page *page;
453 
454 			/* List should never be empty - each reused RBD is
455 			 * returned to the list, and initial pool covers any
456 			 * possible gap between the time the page is allocated
457 			 * to the time the RBD is added.
458 			 */
459 			BUG_ON(list_empty(&local_empty));
460 			/* Get the first rxb from the rbd list */
461 			rxb = list_first_entry(&local_empty,
462 					       struct iwl_rx_mem_buffer, list);
463 			BUG_ON(rxb->page);
464 
465 			/* Alloc a new receive buffer */
466 			page = iwl_pcie_rx_alloc_page(trans, GFP_KERNEL);
467 			if (!page)
468 				continue;
469 			rxb->page = page;
470 
471 			/* Get physical address of the RB */
472 			rxb->page_dma = dma_map_page(trans->dev, page, 0,
473 					PAGE_SIZE << trans_pcie->rx_page_order,
474 					DMA_FROM_DEVICE);
475 			if (dma_mapping_error(trans->dev, rxb->page_dma)) {
476 				rxb->page = NULL;
477 				__free_pages(page, trans_pcie->rx_page_order);
478 				continue;
479 			}
480 			/* dma address must be no more than 36 bits */
481 			BUG_ON(rxb->page_dma & ~DMA_BIT_MASK(36));
482 			/* and also 256 byte aligned! */
483 			BUG_ON(rxb->page_dma & DMA_BIT_MASK(8));
484 
485 			/* move the allocated entry to the out list */
486 			list_move(&rxb->list, &local_allocated);
487 			i++;
488 		}
489 
490 		pending--;
491 		if (!pending) {
492 			pending = atomic_xchg(&rba->req_pending, 0);
493 			IWL_DEBUG_RX(trans,
494 				     "Pending allocation requests = %d\n",
495 				     pending);
496 		}
497 
498 		spin_lock(&rba->lock);
499 		/* add the allocated rbds to the allocator allocated list */
500 		list_splice_tail(&local_allocated, &rba->rbd_allocated);
501 		/* get more empty RBDs for current pending requests */
502 		list_splice_tail_init(&rba->rbd_empty, &local_empty);
503 		spin_unlock(&rba->lock);
504 
505 		atomic_inc(&rba->req_ready);
506 	}
507 
508 	spin_lock(&rba->lock);
509 	/* return unused rbds to the allocator empty list */
510 	list_splice_tail(&local_empty, &rba->rbd_empty);
511 	spin_unlock(&rba->lock);
512 }
513 
514 /*
515  * iwl_pcie_rx_allocator_get - Returns the pre-allocated pages
516 .*
517 .* Called by queue when the queue posted allocation request and
518  * has freed 8 RBDs in order to restock itself.
519  */
520 static int iwl_pcie_rx_allocator_get(struct iwl_trans *trans,
521 				     struct iwl_rx_mem_buffer
522 				     *out[RX_CLAIM_REQ_ALLOC])
523 {
524 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
525 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
526 	int i;
527 
528 	/*
529 	 * atomic_dec_if_positive returns req_ready - 1 for any scenario.
530 	 * If req_ready is 0 atomic_dec_if_positive will return -1 and this
531 	 * function will return -ENOMEM, as there are no ready requests.
532 	 * atomic_dec_if_positive will perofrm the *actual* decrement only if
533 	 * req_ready > 0, i.e. - there are ready requests and the function
534 	 * hands one request to the caller.
535 	 */
536 	if (atomic_dec_if_positive(&rba->req_ready) < 0)
537 		return -ENOMEM;
538 
539 	spin_lock(&rba->lock);
540 	for (i = 0; i < RX_CLAIM_REQ_ALLOC; i++) {
541 		/* Get next free Rx buffer, remove it from free list */
542 		out[i] = list_first_entry(&rba->rbd_allocated,
543 			       struct iwl_rx_mem_buffer, list);
544 		list_del(&out[i]->list);
545 	}
546 	spin_unlock(&rba->lock);
547 
548 	return 0;
549 }
550 
551 static void iwl_pcie_rx_allocator_work(struct work_struct *data)
552 {
553 	struct iwl_rb_allocator *rba_p =
554 		container_of(data, struct iwl_rb_allocator, rx_alloc);
555 	struct iwl_trans_pcie *trans_pcie =
556 		container_of(rba_p, struct iwl_trans_pcie, rba);
557 
558 	iwl_pcie_rx_allocator(trans_pcie->trans);
559 }
560 
561 static int iwl_pcie_rx_alloc(struct iwl_trans *trans)
562 {
563 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
564 	struct iwl_rxq *rxq = &trans_pcie->rxq;
565 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
566 	struct device *dev = trans->dev;
567 
568 	memset(&trans_pcie->rxq, 0, sizeof(trans_pcie->rxq));
569 
570 	spin_lock_init(&rxq->lock);
571 	spin_lock_init(&rba->lock);
572 
573 	if (WARN_ON(rxq->bd || rxq->rb_stts))
574 		return -EINVAL;
575 
576 	/* Allocate the circular buffer of Read Buffer Descriptors (RBDs) */
577 	rxq->bd = dma_zalloc_coherent(dev, sizeof(__le32) * RX_QUEUE_SIZE,
578 				      &rxq->bd_dma, GFP_KERNEL);
579 	if (!rxq->bd)
580 		goto err_bd;
581 
582 	/*Allocate the driver's pointer to receive buffer status */
583 	rxq->rb_stts = dma_zalloc_coherent(dev, sizeof(*rxq->rb_stts),
584 					   &rxq->rb_stts_dma, GFP_KERNEL);
585 	if (!rxq->rb_stts)
586 		goto err_rb_stts;
587 
588 	return 0;
589 
590 err_rb_stts:
591 	dma_free_coherent(dev, sizeof(__le32) * RX_QUEUE_SIZE,
592 			  rxq->bd, rxq->bd_dma);
593 	rxq->bd_dma = 0;
594 	rxq->bd = NULL;
595 err_bd:
596 	return -ENOMEM;
597 }
598 
599 static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
600 {
601 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
602 	u32 rb_size;
603 	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
604 
605 	switch (trans_pcie->rx_buf_size) {
606 	case IWL_AMSDU_4K:
607 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
608 		break;
609 	case IWL_AMSDU_8K:
610 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
611 		break;
612 	case IWL_AMSDU_12K:
613 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_12K;
614 		break;
615 	default:
616 		WARN_ON(1);
617 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
618 	}
619 
620 	/* Stop Rx DMA */
621 	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
622 	/* reset and flush pointers */
623 	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
624 	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
625 	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
626 
627 	/* Reset driver's Rx queue write index */
628 	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
629 
630 	/* Tell device where to find RBD circular buffer in DRAM */
631 	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
632 			   (u32)(rxq->bd_dma >> 8));
633 
634 	/* Tell device where in DRAM to update its Rx status */
635 	iwl_write_direct32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
636 			   rxq->rb_stts_dma >> 4);
637 
638 	/* Enable Rx DMA
639 	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
640 	 *      the credit mechanism in 5000 HW RX FIFO
641 	 * Direct rx interrupts to hosts
642 	 * Rx buffer size 4 or 8k or 12k
643 	 * RB timeout 0x10
644 	 * 256 RBDs
645 	 */
646 	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
647 			   FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
648 			   FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
649 			   FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
650 			   rb_size|
651 			   (RX_RB_TIMEOUT << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS)|
652 			   (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));
653 
654 	/* Set interrupt coalescing timer to default (2048 usecs) */
655 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
656 
657 	/* W/A for interrupt coalescing bug in 7260 and 3160 */
658 	if (trans->cfg->host_interrupt_operation_mode)
659 		iwl_set_bit(trans, CSR_INT_COALESCING, IWL_HOST_INT_OPER_MODE);
660 }
661 
662 static void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq)
663 {
664 	int i;
665 
666 	lockdep_assert_held(&rxq->lock);
667 
668 	INIT_LIST_HEAD(&rxq->rx_free);
669 	INIT_LIST_HEAD(&rxq->rx_used);
670 	rxq->free_count = 0;
671 	rxq->used_count = 0;
672 
673 	for (i = 0; i < RX_QUEUE_SIZE; i++)
674 		list_add(&rxq->pool[i].list, &rxq->rx_used);
675 }
676 
677 static void iwl_pcie_rx_init_rba(struct iwl_rb_allocator *rba)
678 {
679 	int i;
680 
681 	lockdep_assert_held(&rba->lock);
682 
683 	INIT_LIST_HEAD(&rba->rbd_allocated);
684 	INIT_LIST_HEAD(&rba->rbd_empty);
685 
686 	for (i = 0; i < RX_POOL_SIZE; i++)
687 		list_add(&rba->pool[i].list, &rba->rbd_empty);
688 }
689 
690 static void iwl_pcie_rx_free_rba(struct iwl_trans *trans)
691 {
692 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
693 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
694 	int i;
695 
696 	lockdep_assert_held(&rba->lock);
697 
698 	for (i = 0; i < RX_POOL_SIZE; i++) {
699 		if (!rba->pool[i].page)
700 			continue;
701 		dma_unmap_page(trans->dev, rba->pool[i].page_dma,
702 			       PAGE_SIZE << trans_pcie->rx_page_order,
703 			       DMA_FROM_DEVICE);
704 		__free_pages(rba->pool[i].page, trans_pcie->rx_page_order);
705 		rba->pool[i].page = NULL;
706 	}
707 }
708 
709 int iwl_pcie_rx_init(struct iwl_trans *trans)
710 {
711 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
712 	struct iwl_rxq *rxq = &trans_pcie->rxq;
713 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
714 	int i, err;
715 
716 	if (!rxq->bd) {
717 		err = iwl_pcie_rx_alloc(trans);
718 		if (err)
719 			return err;
720 	}
721 	if (!rba->alloc_wq)
722 		rba->alloc_wq = alloc_workqueue("rb_allocator",
723 						WQ_HIGHPRI | WQ_UNBOUND, 1);
724 	INIT_WORK(&rba->rx_alloc, iwl_pcie_rx_allocator_work);
725 
726 	spin_lock(&rba->lock);
727 	atomic_set(&rba->req_pending, 0);
728 	atomic_set(&rba->req_ready, 0);
729 	/* free all first - we might be reconfigured for a different size */
730 	iwl_pcie_rx_free_rba(trans);
731 	iwl_pcie_rx_init_rba(rba);
732 	spin_unlock(&rba->lock);
733 
734 	spin_lock(&rxq->lock);
735 
736 	/* free all first - we might be reconfigured for a different size */
737 	iwl_pcie_rxq_free_rbs(trans);
738 	iwl_pcie_rx_init_rxb_lists(rxq);
739 
740 	for (i = 0; i < RX_QUEUE_SIZE; i++)
741 		rxq->queue[i] = NULL;
742 
743 	/* Set us so that we have processed and used all buffers, but have
744 	 * not restocked the Rx queue with fresh buffers */
745 	rxq->read = rxq->write = 0;
746 	rxq->write_actual = 0;
747 	memset(rxq->rb_stts, 0, sizeof(*rxq->rb_stts));
748 	spin_unlock(&rxq->lock);
749 
750 	iwl_pcie_rx_replenish(trans);
751 
752 	iwl_pcie_rx_hw_init(trans, rxq);
753 
754 	spin_lock(&rxq->lock);
755 	iwl_pcie_rxq_inc_wr_ptr(trans);
756 	spin_unlock(&rxq->lock);
757 
758 	return 0;
759 }
760 
761 void iwl_pcie_rx_free(struct iwl_trans *trans)
762 {
763 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
764 	struct iwl_rxq *rxq = &trans_pcie->rxq;
765 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
766 
767 	/*if rxq->bd is NULL, it means that nothing has been allocated,
768 	 * exit now */
769 	if (!rxq->bd) {
770 		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
771 		return;
772 	}
773 
774 	cancel_work_sync(&rba->rx_alloc);
775 	if (rba->alloc_wq) {
776 		destroy_workqueue(rba->alloc_wq);
777 		rba->alloc_wq = NULL;
778 	}
779 
780 	spin_lock(&rba->lock);
781 	iwl_pcie_rx_free_rba(trans);
782 	spin_unlock(&rba->lock);
783 
784 	spin_lock(&rxq->lock);
785 	iwl_pcie_rxq_free_rbs(trans);
786 	spin_unlock(&rxq->lock);
787 
788 	dma_free_coherent(trans->dev, sizeof(__le32) * RX_QUEUE_SIZE,
789 			  rxq->bd, rxq->bd_dma);
790 	rxq->bd_dma = 0;
791 	rxq->bd = NULL;
792 
793 	if (rxq->rb_stts)
794 		dma_free_coherent(trans->dev,
795 				  sizeof(struct iwl_rb_status),
796 				  rxq->rb_stts, rxq->rb_stts_dma);
797 	else
798 		IWL_DEBUG_INFO(trans, "Free rxq->rb_stts which is NULL\n");
799 	rxq->rb_stts_dma = 0;
800 	rxq->rb_stts = NULL;
801 }
802 
803 /*
804  * iwl_pcie_rx_reuse_rbd - Recycle used RBDs
805  *
806  * Called when a RBD can be reused. The RBD is transferred to the allocator.
807  * When there are 2 empty RBDs - a request for allocation is posted
808  */
809 static void iwl_pcie_rx_reuse_rbd(struct iwl_trans *trans,
810 				  struct iwl_rx_mem_buffer *rxb,
811 				  struct iwl_rxq *rxq, bool emergency)
812 {
813 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
814 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
815 
816 	/* Move the RBD to the used list, will be moved to allocator in batches
817 	 * before claiming or posting a request*/
818 	list_add_tail(&rxb->list, &rxq->rx_used);
819 
820 	if (unlikely(emergency))
821 		return;
822 
823 	/* Count the allocator owned RBDs */
824 	rxq->used_count++;
825 
826 	/* If we have RX_POST_REQ_ALLOC new released rx buffers -
827 	 * issue a request for allocator. Modulo RX_CLAIM_REQ_ALLOC is
828 	 * used for the case we failed to claim RX_CLAIM_REQ_ALLOC,
829 	 * after but we still need to post another request.
830 	 */
831 	if ((rxq->used_count % RX_CLAIM_REQ_ALLOC) == RX_POST_REQ_ALLOC) {
832 		/* Move the 2 RBDs to the allocator ownership.
833 		 Allocator has another 6 from pool for the request completion*/
834 		spin_lock(&rba->lock);
835 		list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
836 		spin_unlock(&rba->lock);
837 
838 		atomic_inc(&rba->req_pending);
839 		queue_work(rba->alloc_wq, &rba->rx_alloc);
840 	}
841 }
842 
843 static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
844 				struct iwl_rx_mem_buffer *rxb,
845 				bool emergency)
846 {
847 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
848 	struct iwl_rxq *rxq = &trans_pcie->rxq;
849 	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
850 	bool page_stolen = false;
851 	int max_len = PAGE_SIZE << trans_pcie->rx_page_order;
852 	u32 offset = 0;
853 
854 	if (WARN_ON(!rxb))
855 		return;
856 
857 	dma_unmap_page(trans->dev, rxb->page_dma, max_len, DMA_FROM_DEVICE);
858 
859 	while (offset + sizeof(u32) + sizeof(struct iwl_cmd_header) < max_len) {
860 		struct iwl_rx_packet *pkt;
861 		u16 sequence;
862 		bool reclaim;
863 		int index, cmd_index, len;
864 		struct iwl_rx_cmd_buffer rxcb = {
865 			._offset = offset,
866 			._rx_page_order = trans_pcie->rx_page_order,
867 			._page = rxb->page,
868 			._page_stolen = false,
869 			.truesize = max_len,
870 		};
871 
872 		pkt = rxb_addr(&rxcb);
873 
874 		if (pkt->len_n_flags == cpu_to_le32(FH_RSCSR_FRAME_INVALID))
875 			break;
876 
877 		IWL_DEBUG_RX(trans,
878 			     "cmd at offset %d: %s (0x%.2x, seq 0x%x)\n",
879 			     rxcb._offset,
880 			     iwl_get_cmd_string(trans,
881 						iwl_cmd_id(pkt->hdr.cmd,
882 							   pkt->hdr.group_id,
883 							   0)),
884 			     pkt->hdr.cmd, le16_to_cpu(pkt->hdr.sequence));
885 
886 		len = iwl_rx_packet_len(pkt);
887 		len += sizeof(u32); /* account for status word */
888 		trace_iwlwifi_dev_rx(trans->dev, trans, pkt, len);
889 		trace_iwlwifi_dev_rx_data(trans->dev, trans, pkt, len);
890 
891 		/* Reclaim a command buffer only if this packet is a response
892 		 *   to a (driver-originated) command.
893 		 * If the packet (e.g. Rx frame) originated from uCode,
894 		 *   there is no command buffer to reclaim.
895 		 * Ucode should set SEQ_RX_FRAME bit if ucode-originated,
896 		 *   but apparently a few don't get set; catch them here. */
897 		reclaim = !(pkt->hdr.sequence & SEQ_RX_FRAME);
898 		if (reclaim) {
899 			int i;
900 
901 			for (i = 0; i < trans_pcie->n_no_reclaim_cmds; i++) {
902 				if (trans_pcie->no_reclaim_cmds[i] ==
903 							pkt->hdr.cmd) {
904 					reclaim = false;
905 					break;
906 				}
907 			}
908 		}
909 
910 		sequence = le16_to_cpu(pkt->hdr.sequence);
911 		index = SEQ_TO_INDEX(sequence);
912 		cmd_index = get_cmd_index(&txq->q, index);
913 
914 		iwl_op_mode_rx(trans->op_mode, &trans_pcie->napi, &rxcb);
915 
916 		if (reclaim) {
917 			kzfree(txq->entries[cmd_index].free_buf);
918 			txq->entries[cmd_index].free_buf = NULL;
919 		}
920 
921 		/*
922 		 * After here, we should always check rxcb._page_stolen,
923 		 * if it is true then one of the handlers took the page.
924 		 */
925 
926 		if (reclaim) {
927 			/* Invoke any callbacks, transfer the buffer to caller,
928 			 * and fire off the (possibly) blocking
929 			 * iwl_trans_send_cmd()
930 			 * as we reclaim the driver command queue */
931 			if (!rxcb._page_stolen)
932 				iwl_pcie_hcmd_complete(trans, &rxcb);
933 			else
934 				IWL_WARN(trans, "Claim null rxb?\n");
935 		}
936 
937 		page_stolen |= rxcb._page_stolen;
938 		offset += ALIGN(len, FH_RSCSR_FRAME_ALIGN);
939 	}
940 
941 	/* page was stolen from us -- free our reference */
942 	if (page_stolen) {
943 		__free_pages(rxb->page, trans_pcie->rx_page_order);
944 		rxb->page = NULL;
945 	}
946 
947 	/* Reuse the page if possible. For notification packets and
948 	 * SKBs that fail to Rx correctly, add them back into the
949 	 * rx_free list for reuse later. */
950 	if (rxb->page != NULL) {
951 		rxb->page_dma =
952 			dma_map_page(trans->dev, rxb->page, 0,
953 				     PAGE_SIZE << trans_pcie->rx_page_order,
954 				     DMA_FROM_DEVICE);
955 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
956 			/*
957 			 * free the page(s) as well to not break
958 			 * the invariant that the items on the used
959 			 * list have no page(s)
960 			 */
961 			__free_pages(rxb->page, trans_pcie->rx_page_order);
962 			rxb->page = NULL;
963 			iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
964 		} else {
965 			list_add_tail(&rxb->list, &rxq->rx_free);
966 			rxq->free_count++;
967 		}
968 	} else
969 		iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
970 }
971 
972 /*
973  * iwl_pcie_rx_handle - Main entry function for receiving responses from fw
974  */
975 static void iwl_pcie_rx_handle(struct iwl_trans *trans)
976 {
977 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
978 	struct iwl_rxq *rxq = &trans_pcie->rxq;
979 	u32 r, i, j, count = 0;
980 	bool emergency = false;
981 
982 restart:
983 	spin_lock(&rxq->lock);
984 	/* uCode's read index (stored in shared DRAM) indicates the last Rx
985 	 * buffer that the driver may process (last buffer filled by ucode). */
986 	r = le16_to_cpu(ACCESS_ONCE(rxq->rb_stts->closed_rb_num)) & 0x0FFF;
987 	i = rxq->read;
988 
989 	/* Rx interrupt, but nothing sent from uCode */
990 	if (i == r)
991 		IWL_DEBUG_RX(trans, "HW = SW = %d\n", r);
992 
993 	while (i != r) {
994 		struct iwl_rx_mem_buffer *rxb;
995 
996 		if (unlikely(rxq->used_count == RX_QUEUE_SIZE / 2))
997 			emergency = true;
998 
999 		rxb = rxq->queue[i];
1000 		rxq->queue[i] = NULL;
1001 
1002 		IWL_DEBUG_RX(trans, "rxbuf: HW = %d, SW = %d\n", r, i);
1003 		iwl_pcie_rx_handle_rb(trans, rxb, emergency);
1004 
1005 		i = (i + 1) & RX_QUEUE_MASK;
1006 
1007 		/* If we have RX_CLAIM_REQ_ALLOC released rx buffers -
1008 		 * try to claim the pre-allocated buffers from the allocator */
1009 		if (rxq->used_count >= RX_CLAIM_REQ_ALLOC) {
1010 			struct iwl_rb_allocator *rba = &trans_pcie->rba;
1011 			struct iwl_rx_mem_buffer *out[RX_CLAIM_REQ_ALLOC];
1012 
1013 			if (rxq->used_count % RX_CLAIM_REQ_ALLOC == 0 &&
1014 			    !emergency) {
1015 				/* Add the remaining 6 empty RBDs
1016 				* for allocator use
1017 				 */
1018 				spin_lock(&rba->lock);
1019 				list_splice_tail_init(&rxq->rx_used,
1020 						      &rba->rbd_empty);
1021 				spin_unlock(&rba->lock);
1022 			}
1023 
1024 			/* If not ready - continue, will try to reclaim later.
1025 			* No need to reschedule work - allocator exits only on
1026 			* success */
1027 			if (!iwl_pcie_rx_allocator_get(trans, out)) {
1028 				/* If success - then RX_CLAIM_REQ_ALLOC
1029 				 * buffers were retrieved and should be added
1030 				 * to free list */
1031 				rxq->used_count -= RX_CLAIM_REQ_ALLOC;
1032 				for (j = 0; j < RX_CLAIM_REQ_ALLOC; j++) {
1033 					list_add_tail(&out[j]->list,
1034 						      &rxq->rx_free);
1035 					rxq->free_count++;
1036 				}
1037 			}
1038 		}
1039 		if (emergency) {
1040 			count++;
1041 			if (count == 8) {
1042 				count = 0;
1043 				if (rxq->used_count < RX_QUEUE_SIZE / 3)
1044 					emergency = false;
1045 				spin_unlock(&rxq->lock);
1046 				iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC);
1047 				spin_lock(&rxq->lock);
1048 			}
1049 		}
1050 		/* handle restock for three cases, can be all of them at once:
1051 		* - we just pulled buffers from the allocator
1052 		* - we have 8+ unstolen pages accumulated
1053 		* - we are in emergency and allocated buffers
1054 		 */
1055 		if (rxq->free_count >=  RX_CLAIM_REQ_ALLOC) {
1056 			rxq->read = i;
1057 			spin_unlock(&rxq->lock);
1058 			iwl_pcie_rxq_restock(trans);
1059 			goto restart;
1060 		}
1061 	}
1062 
1063 	/* Backtrack one entry */
1064 	rxq->read = i;
1065 	spin_unlock(&rxq->lock);
1066 
1067 	/*
1068 	 * handle a case where in emergency there are some unallocated RBDs.
1069 	 * those RBDs are in the used list, but are not tracked by the queue's
1070 	 * used_count which counts allocator owned RBDs.
1071 	 * unallocated emergency RBDs must be allocated on exit, otherwise
1072 	 * when called again the function may not be in emergency mode and
1073 	 * they will be handed to the allocator with no tracking in the RBD
1074 	 * allocator counters, which will lead to them never being claimed back
1075 	 * by the queue.
1076 	 * by allocating them here, they are now in the queue free list, and
1077 	 * will be restocked by the next call of iwl_pcie_rxq_restock.
1078 	 */
1079 	if (unlikely(emergency && count))
1080 		iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC);
1081 
1082 	if (trans_pcie->napi.poll)
1083 		napi_gro_flush(&trans_pcie->napi, false);
1084 }
1085 
1086 /*
1087  * iwl_pcie_irq_handle_error - called for HW or SW error interrupt from card
1088  */
1089 static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
1090 {
1091 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1092 	int i;
1093 
1094 	/* W/A for WiFi/WiMAX coex and WiMAX own the RF */
1095 	if (trans->cfg->internal_wimax_coex &&
1096 	    !trans->cfg->apmg_not_supported &&
1097 	    (!(iwl_read_prph(trans, APMG_CLK_CTRL_REG) &
1098 			     APMS_CLK_VAL_MRB_FUNC_MODE) ||
1099 	     (iwl_read_prph(trans, APMG_PS_CTRL_REG) &
1100 			    APMG_PS_CTRL_VAL_RESET_REQ))) {
1101 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1102 		iwl_op_mode_wimax_active(trans->op_mode);
1103 		wake_up(&trans_pcie->wait_command_queue);
1104 		return;
1105 	}
1106 
1107 	iwl_pcie_dump_csr(trans);
1108 	iwl_dump_fh(trans, NULL);
1109 
1110 	local_bh_disable();
1111 	/* The STATUS_FW_ERROR bit is set in this function. This must happen
1112 	 * before we wake up the command caller, to ensure a proper cleanup. */
1113 	iwl_trans_fw_error(trans);
1114 	local_bh_enable();
1115 
1116 	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++)
1117 		del_timer(&trans_pcie->txq[i].stuck_timer);
1118 
1119 	clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1120 	wake_up(&trans_pcie->wait_command_queue);
1121 }
1122 
1123 static u32 iwl_pcie_int_cause_non_ict(struct iwl_trans *trans)
1124 {
1125 	u32 inta;
1126 
1127 	lockdep_assert_held(&IWL_TRANS_GET_PCIE_TRANS(trans)->irq_lock);
1128 
1129 	trace_iwlwifi_dev_irq(trans->dev);
1130 
1131 	/* Discover which interrupts are active/pending */
1132 	inta = iwl_read32(trans, CSR_INT);
1133 
1134 	/* the thread will service interrupts and re-enable them */
1135 	return inta;
1136 }
1137 
1138 /* a device (PCI-E) page is 4096 bytes long */
1139 #define ICT_SHIFT	12
1140 #define ICT_SIZE	(1 << ICT_SHIFT)
1141 #define ICT_COUNT	(ICT_SIZE / sizeof(u32))
1142 
1143 /* interrupt handler using ict table, with this interrupt driver will
1144  * stop using INTA register to get device's interrupt, reading this register
1145  * is expensive, device will write interrupts in ICT dram table, increment
1146  * index then will fire interrupt to driver, driver will OR all ICT table
1147  * entries from current index up to table entry with 0 value. the result is
1148  * the interrupt we need to service, driver will set the entries back to 0 and
1149  * set index.
1150  */
1151 static u32 iwl_pcie_int_cause_ict(struct iwl_trans *trans)
1152 {
1153 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1154 	u32 inta;
1155 	u32 val = 0;
1156 	u32 read;
1157 
1158 	trace_iwlwifi_dev_irq(trans->dev);
1159 
1160 	/* Ignore interrupt if there's nothing in NIC to service.
1161 	 * This may be due to IRQ shared with another device,
1162 	 * or due to sporadic interrupts thrown from our NIC. */
1163 	read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1164 	trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index, read);
1165 	if (!read)
1166 		return 0;
1167 
1168 	/*
1169 	 * Collect all entries up to the first 0, starting from ict_index;
1170 	 * note we already read at ict_index.
1171 	 */
1172 	do {
1173 		val |= read;
1174 		IWL_DEBUG_ISR(trans, "ICT index %d value 0x%08X\n",
1175 				trans_pcie->ict_index, read);
1176 		trans_pcie->ict_tbl[trans_pcie->ict_index] = 0;
1177 		trans_pcie->ict_index =
1178 			((trans_pcie->ict_index + 1) & (ICT_COUNT - 1));
1179 
1180 		read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1181 		trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index,
1182 					   read);
1183 	} while (read);
1184 
1185 	/* We should not get this value, just ignore it. */
1186 	if (val == 0xffffffff)
1187 		val = 0;
1188 
1189 	/*
1190 	 * this is a w/a for a h/w bug. the h/w bug may cause the Rx bit
1191 	 * (bit 15 before shifting it to 31) to clear when using interrupt
1192 	 * coalescing. fortunately, bits 18 and 19 stay set when this happens
1193 	 * so we use them to decide on the real state of the Rx bit.
1194 	 * In order words, bit 15 is set if bit 18 or bit 19 are set.
1195 	 */
1196 	if (val & 0xC0000)
1197 		val |= 0x8000;
1198 
1199 	inta = (0xff & val) | ((0xff00 & val) << 16);
1200 	return inta;
1201 }
1202 
1203 irqreturn_t iwl_pcie_irq_handler(int irq, void *dev_id)
1204 {
1205 	struct iwl_trans *trans = dev_id;
1206 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1207 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1208 	u32 inta = 0;
1209 	u32 handled = 0;
1210 
1211 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1212 
1213 	spin_lock(&trans_pcie->irq_lock);
1214 
1215 	/* dram interrupt table not set yet,
1216 	 * use legacy interrupt.
1217 	 */
1218 	if (likely(trans_pcie->use_ict))
1219 		inta = iwl_pcie_int_cause_ict(trans);
1220 	else
1221 		inta = iwl_pcie_int_cause_non_ict(trans);
1222 
1223 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1224 		IWL_DEBUG_ISR(trans,
1225 			      "ISR inta 0x%08x, enabled 0x%08x(sw), enabled(hw) 0x%08x, fh 0x%08x\n",
1226 			      inta, trans_pcie->inta_mask,
1227 			      iwl_read32(trans, CSR_INT_MASK),
1228 			      iwl_read32(trans, CSR_FH_INT_STATUS));
1229 		if (inta & (~trans_pcie->inta_mask))
1230 			IWL_DEBUG_ISR(trans,
1231 				      "We got a masked interrupt (0x%08x)\n",
1232 				      inta & (~trans_pcie->inta_mask));
1233 	}
1234 
1235 	inta &= trans_pcie->inta_mask;
1236 
1237 	/*
1238 	 * Ignore interrupt if there's nothing in NIC to service.
1239 	 * This may be due to IRQ shared with another device,
1240 	 * or due to sporadic interrupts thrown from our NIC.
1241 	 */
1242 	if (unlikely(!inta)) {
1243 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
1244 		/*
1245 		 * Re-enable interrupts here since we don't
1246 		 * have anything to service
1247 		 */
1248 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1249 			iwl_enable_interrupts(trans);
1250 		spin_unlock(&trans_pcie->irq_lock);
1251 		lock_map_release(&trans->sync_cmd_lockdep_map);
1252 		return IRQ_NONE;
1253 	}
1254 
1255 	if (unlikely(inta == 0xFFFFFFFF || (inta & 0xFFFFFFF0) == 0xa5a5a5a0)) {
1256 		/*
1257 		 * Hardware disappeared. It might have
1258 		 * already raised an interrupt.
1259 		 */
1260 		IWL_WARN(trans, "HARDWARE GONE?? INTA == 0x%08x\n", inta);
1261 		spin_unlock(&trans_pcie->irq_lock);
1262 		goto out;
1263 	}
1264 
1265 	/* Ack/clear/reset pending uCode interrupts.
1266 	 * Note:  Some bits in CSR_INT are "OR" of bits in CSR_FH_INT_STATUS,
1267 	 */
1268 	/* There is a hardware bug in the interrupt mask function that some
1269 	 * interrupts (i.e. CSR_INT_BIT_SCD) can still be generated even if
1270 	 * they are disabled in the CSR_INT_MASK register. Furthermore the
1271 	 * ICT interrupt handling mechanism has another bug that might cause
1272 	 * these unmasked interrupts fail to be detected. We workaround the
1273 	 * hardware bugs here by ACKing all the possible interrupts so that
1274 	 * interrupt coalescing can still be achieved.
1275 	 */
1276 	iwl_write32(trans, CSR_INT, inta | ~trans_pcie->inta_mask);
1277 
1278 	if (iwl_have_debug_level(IWL_DL_ISR))
1279 		IWL_DEBUG_ISR(trans, "inta 0x%08x, enabled 0x%08x\n",
1280 			      inta, iwl_read32(trans, CSR_INT_MASK));
1281 
1282 	spin_unlock(&trans_pcie->irq_lock);
1283 
1284 	/* Now service all interrupt bits discovered above. */
1285 	if (inta & CSR_INT_BIT_HW_ERR) {
1286 		IWL_ERR(trans, "Hardware error detected.  Restarting.\n");
1287 
1288 		/* Tell the device to stop sending interrupts */
1289 		iwl_disable_interrupts(trans);
1290 
1291 		isr_stats->hw++;
1292 		iwl_pcie_irq_handle_error(trans);
1293 
1294 		handled |= CSR_INT_BIT_HW_ERR;
1295 
1296 		goto out;
1297 	}
1298 
1299 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1300 		/* NIC fires this, but we don't use it, redundant with WAKEUP */
1301 		if (inta & CSR_INT_BIT_SCD) {
1302 			IWL_DEBUG_ISR(trans,
1303 				      "Scheduler finished to transmit the frame/frames.\n");
1304 			isr_stats->sch++;
1305 		}
1306 
1307 		/* Alive notification via Rx interrupt will do the real work */
1308 		if (inta & CSR_INT_BIT_ALIVE) {
1309 			IWL_DEBUG_ISR(trans, "Alive interrupt\n");
1310 			isr_stats->alive++;
1311 		}
1312 	}
1313 
1314 	/* Safely ignore these bits for debug checks below */
1315 	inta &= ~(CSR_INT_BIT_SCD | CSR_INT_BIT_ALIVE);
1316 
1317 	/* HW RF KILL switch toggled */
1318 	if (inta & CSR_INT_BIT_RF_KILL) {
1319 		bool hw_rfkill;
1320 
1321 		hw_rfkill = iwl_is_rfkill_set(trans);
1322 		IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
1323 			 hw_rfkill ? "disable radio" : "enable radio");
1324 
1325 		isr_stats->rfkill++;
1326 
1327 		mutex_lock(&trans_pcie->mutex);
1328 		iwl_trans_pcie_rf_kill(trans, hw_rfkill);
1329 		mutex_unlock(&trans_pcie->mutex);
1330 		if (hw_rfkill) {
1331 			set_bit(STATUS_RFKILL, &trans->status);
1332 			if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
1333 					       &trans->status))
1334 				IWL_DEBUG_RF_KILL(trans,
1335 						  "Rfkill while SYNC HCMD in flight\n");
1336 			wake_up(&trans_pcie->wait_command_queue);
1337 		} else {
1338 			clear_bit(STATUS_RFKILL, &trans->status);
1339 		}
1340 
1341 		handled |= CSR_INT_BIT_RF_KILL;
1342 	}
1343 
1344 	/* Chip got too hot and stopped itself */
1345 	if (inta & CSR_INT_BIT_CT_KILL) {
1346 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
1347 		isr_stats->ctkill++;
1348 		handled |= CSR_INT_BIT_CT_KILL;
1349 	}
1350 
1351 	/* Error detected by uCode */
1352 	if (inta & CSR_INT_BIT_SW_ERR) {
1353 		IWL_ERR(trans, "Microcode SW error detected. "
1354 			" Restarting 0x%X.\n", inta);
1355 		isr_stats->sw++;
1356 		iwl_pcie_irq_handle_error(trans);
1357 		handled |= CSR_INT_BIT_SW_ERR;
1358 	}
1359 
1360 	/* uCode wakes up after power-down sleep */
1361 	if (inta & CSR_INT_BIT_WAKEUP) {
1362 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
1363 		iwl_pcie_rxq_check_wrptr(trans);
1364 		iwl_pcie_txq_check_wrptrs(trans);
1365 
1366 		isr_stats->wakeup++;
1367 
1368 		handled |= CSR_INT_BIT_WAKEUP;
1369 	}
1370 
1371 	/* All uCode command responses, including Tx command responses,
1372 	 * Rx "responses" (frame-received notification), and other
1373 	 * notifications from uCode come through here*/
1374 	if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX |
1375 		    CSR_INT_BIT_RX_PERIODIC)) {
1376 		IWL_DEBUG_ISR(trans, "Rx interrupt\n");
1377 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX)) {
1378 			handled |= (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX);
1379 			iwl_write32(trans, CSR_FH_INT_STATUS,
1380 					CSR_FH_INT_RX_MASK);
1381 		}
1382 		if (inta & CSR_INT_BIT_RX_PERIODIC) {
1383 			handled |= CSR_INT_BIT_RX_PERIODIC;
1384 			iwl_write32(trans,
1385 				CSR_INT, CSR_INT_BIT_RX_PERIODIC);
1386 		}
1387 		/* Sending RX interrupt require many steps to be done in the
1388 		 * the device:
1389 		 * 1- write interrupt to current index in ICT table.
1390 		 * 2- dma RX frame.
1391 		 * 3- update RX shared data to indicate last write index.
1392 		 * 4- send interrupt.
1393 		 * This could lead to RX race, driver could receive RX interrupt
1394 		 * but the shared data changes does not reflect this;
1395 		 * periodic interrupt will detect any dangling Rx activity.
1396 		 */
1397 
1398 		/* Disable periodic interrupt; we use it as just a one-shot. */
1399 		iwl_write8(trans, CSR_INT_PERIODIC_REG,
1400 			    CSR_INT_PERIODIC_DIS);
1401 
1402 		/*
1403 		 * Enable periodic interrupt in 8 msec only if we received
1404 		 * real RX interrupt (instead of just periodic int), to catch
1405 		 * any dangling Rx interrupt.  If it was just the periodic
1406 		 * interrupt, there was no dangling Rx activity, and no need
1407 		 * to extend the periodic interrupt; one-shot is enough.
1408 		 */
1409 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX))
1410 			iwl_write8(trans, CSR_INT_PERIODIC_REG,
1411 				   CSR_INT_PERIODIC_ENA);
1412 
1413 		isr_stats->rx++;
1414 
1415 		local_bh_disable();
1416 		iwl_pcie_rx_handle(trans);
1417 		local_bh_enable();
1418 	}
1419 
1420 	/* This "Tx" DMA channel is used only for loading uCode */
1421 	if (inta & CSR_INT_BIT_FH_TX) {
1422 		iwl_write32(trans, CSR_FH_INT_STATUS, CSR_FH_INT_TX_MASK);
1423 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
1424 		isr_stats->tx++;
1425 		handled |= CSR_INT_BIT_FH_TX;
1426 		/* Wake up uCode load routine, now that load is complete */
1427 		trans_pcie->ucode_write_complete = true;
1428 		wake_up(&trans_pcie->ucode_write_waitq);
1429 	}
1430 
1431 	if (inta & ~handled) {
1432 		IWL_ERR(trans, "Unhandled INTA bits 0x%08x\n", inta & ~handled);
1433 		isr_stats->unhandled++;
1434 	}
1435 
1436 	if (inta & ~(trans_pcie->inta_mask)) {
1437 		IWL_WARN(trans, "Disabled INTA bits 0x%08x were pending\n",
1438 			 inta & ~trans_pcie->inta_mask);
1439 	}
1440 
1441 	/* Re-enable all interrupts */
1442 	/* only Re-enable if disabled by irq */
1443 	if (test_bit(STATUS_INT_ENABLED, &trans->status))
1444 		iwl_enable_interrupts(trans);
1445 	/* Re-enable RF_KILL if it occurred */
1446 	else if (handled & CSR_INT_BIT_RF_KILL)
1447 		iwl_enable_rfkill_int(trans);
1448 
1449 out:
1450 	lock_map_release(&trans->sync_cmd_lockdep_map);
1451 	return IRQ_HANDLED;
1452 }
1453 
1454 /******************************************************************************
1455  *
1456  * ICT functions
1457  *
1458  ******************************************************************************/
1459 
1460 /* Free dram table */
1461 void iwl_pcie_free_ict(struct iwl_trans *trans)
1462 {
1463 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1464 
1465 	if (trans_pcie->ict_tbl) {
1466 		dma_free_coherent(trans->dev, ICT_SIZE,
1467 				  trans_pcie->ict_tbl,
1468 				  trans_pcie->ict_tbl_dma);
1469 		trans_pcie->ict_tbl = NULL;
1470 		trans_pcie->ict_tbl_dma = 0;
1471 	}
1472 }
1473 
1474 /*
1475  * allocate dram shared table, it is an aligned memory
1476  * block of ICT_SIZE.
1477  * also reset all data related to ICT table interrupt.
1478  */
1479 int iwl_pcie_alloc_ict(struct iwl_trans *trans)
1480 {
1481 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1482 
1483 	trans_pcie->ict_tbl =
1484 		dma_zalloc_coherent(trans->dev, ICT_SIZE,
1485 				   &trans_pcie->ict_tbl_dma,
1486 				   GFP_KERNEL);
1487 	if (!trans_pcie->ict_tbl)
1488 		return -ENOMEM;
1489 
1490 	/* just an API sanity check ... it is guaranteed to be aligned */
1491 	if (WARN_ON(trans_pcie->ict_tbl_dma & (ICT_SIZE - 1))) {
1492 		iwl_pcie_free_ict(trans);
1493 		return -EINVAL;
1494 	}
1495 
1496 	return 0;
1497 }
1498 
1499 /* Device is going up inform it about using ICT interrupt table,
1500  * also we need to tell the driver to start using ICT interrupt.
1501  */
1502 void iwl_pcie_reset_ict(struct iwl_trans *trans)
1503 {
1504 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1505 	u32 val;
1506 
1507 	if (!trans_pcie->ict_tbl)
1508 		return;
1509 
1510 	spin_lock(&trans_pcie->irq_lock);
1511 	iwl_disable_interrupts(trans);
1512 
1513 	memset(trans_pcie->ict_tbl, 0, ICT_SIZE);
1514 
1515 	val = trans_pcie->ict_tbl_dma >> ICT_SHIFT;
1516 
1517 	val |= CSR_DRAM_INT_TBL_ENABLE |
1518 	       CSR_DRAM_INIT_TBL_WRAP_CHECK |
1519 	       CSR_DRAM_INIT_TBL_WRITE_POINTER;
1520 
1521 	IWL_DEBUG_ISR(trans, "CSR_DRAM_INT_TBL_REG =0x%x\n", val);
1522 
1523 	iwl_write32(trans, CSR_DRAM_INT_TBL_REG, val);
1524 	trans_pcie->use_ict = true;
1525 	trans_pcie->ict_index = 0;
1526 	iwl_write32(trans, CSR_INT, trans_pcie->inta_mask);
1527 	iwl_enable_interrupts(trans);
1528 	spin_unlock(&trans_pcie->irq_lock);
1529 }
1530 
1531 /* Device is going down disable ict interrupt usage */
1532 void iwl_pcie_disable_ict(struct iwl_trans *trans)
1533 {
1534 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1535 
1536 	spin_lock(&trans_pcie->irq_lock);
1537 	trans_pcie->use_ict = false;
1538 	spin_unlock(&trans_pcie->irq_lock);
1539 }
1540 
1541 irqreturn_t iwl_pcie_isr(int irq, void *data)
1542 {
1543 	struct iwl_trans *trans = data;
1544 
1545 	if (!trans)
1546 		return IRQ_NONE;
1547 
1548 	/* Disable (but don't clear!) interrupts here to avoid
1549 	 * back-to-back ISRs and sporadic interrupts from our NIC.
1550 	 * If we have something to service, the tasklet will re-enable ints.
1551 	 * If we *don't* have something, we'll re-enable before leaving here.
1552 	 */
1553 	iwl_write32(trans, CSR_INT_MASK, 0x00000000);
1554 
1555 	return IRQ_WAKE_THREAD;
1556 }
1557