xref: /linux/drivers/net/wireless/intel/iwlwifi/pcie/gen1_2/rx.c (revision 8be4d31cb8aaeea27bde4b7ddb26e28a89062ebf)
1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /*
3  * Copyright (C) 2003-2014, 2018-2024 Intel Corporation
4  * Copyright (C) 2013-2015 Intel Mobile Communications GmbH
5  * Copyright (C) 2016-2017 Intel Deutschland GmbH
6  */
7 #include <linux/sched.h>
8 #include <linux/wait.h>
9 #include <linux/gfp.h>
10 
11 #include "iwl-prph.h"
12 #include "iwl-io.h"
13 #include "internal.h"
14 #include "iwl-op-mode.h"
15 #include "pcie/iwl-context-info-v2.h"
16 #include "fw/dbg.h"
17 
18 /******************************************************************************
19  *
20  * RX path functions
21  *
22  ******************************************************************************/
23 
24 /*
25  * Rx theory of operation
26  *
27  * Driver allocates a circular buffer of Receive Buffer Descriptors (RBDs),
28  * each of which point to Receive Buffers to be filled by the NIC.  These get
29  * used not only for Rx frames, but for any command response or notification
30  * from the NIC.  The driver and NIC manage the Rx buffers by means
31  * of indexes into the circular buffer.
32  *
33  * Rx Queue Indexes
34  * The host/firmware share two index registers for managing the Rx buffers.
35  *
36  * The READ index maps to the first position that the firmware may be writing
37  * to -- the driver can read up to (but not including) this position and get
38  * good data.
39  * The READ index is managed by the firmware once the card is enabled.
40  *
41  * The WRITE index maps to the last position the driver has read from -- the
42  * position preceding WRITE is the last slot the firmware can place a packet.
43  *
44  * The queue is empty (no good data) if WRITE = READ - 1, and is full if
45  * WRITE = READ.
46  *
47  * During initialization, the host sets up the READ queue position to the first
48  * INDEX position, and WRITE to the last (READ - 1 wrapped)
49  *
50  * When the firmware places a packet in a buffer, it will advance the READ index
51  * and fire the RX interrupt.  The driver can then query the READ index and
52  * process as many packets as possible, moving the WRITE index forward as it
53  * resets the Rx queue buffers with new memory.
54  *
55  * The management in the driver is as follows:
56  * + A list of pre-allocated RBDs is stored in iwl->rxq->rx_free.
57  *   When the interrupt handler is called, the request is processed.
58  *   The page is either stolen - transferred to the upper layer
59  *   or reused - added immediately to the iwl->rxq->rx_free list.
60  * + When the page is stolen - the driver updates the matching queue's used
61  *   count, detaches the RBD and transfers it to the queue used list.
62  *   When there are two used RBDs - they are transferred to the allocator empty
63  *   list. Work is then scheduled for the allocator to start allocating
64  *   eight buffers.
65  *   When there are another 6 used RBDs - they are transferred to the allocator
66  *   empty list and the driver tries to claim the pre-allocated buffers and
67  *   add them to iwl->rxq->rx_free. If it fails - it continues to claim them
68  *   until ready.
69  *   When there are 8+ buffers in the free list - either from allocation or from
70  *   8 reused unstolen pages - restock is called to update the FW and indexes.
71  * + In order to make sure the allocator always has RBDs to use for allocation
72  *   the allocator has initial pool in the size of num_queues*(8-2) - the
73  *   maximum missing RBDs per allocation request (request posted with 2
74  *    empty RBDs, there is no guarantee when the other 6 RBDs are supplied).
75  *   The queues supplies the recycle of the rest of the RBDs.
76  * + A received packet is processed and handed to the kernel network stack,
77  *   detached from the iwl->rxq.  The driver 'processed' index is updated.
78  * + If there are no allocated buffers in iwl->rxq->rx_free,
79  *   the READ INDEX is not incremented and iwl->status(RX_STALLED) is set.
80  *   If there were enough free buffers and RX_STALLED is set it is cleared.
81  *
82  *
83  * Driver sequence:
84  *
85  * iwl_rxq_alloc()            Allocates rx_free
86  * iwl_pcie_rx_replenish()    Replenishes rx_free list from rx_used, and calls
87  *                            iwl_pcie_rxq_restock.
88  *                            Used only during initialization.
89  * iwl_pcie_rxq_restock()     Moves available buffers from rx_free into Rx
90  *                            queue, updates firmware pointers, and updates
91  *                            the WRITE index.
92  * iwl_pcie_rx_allocator()     Background work for allocating pages.
93  *
94  * -- enable interrupts --
95  * ISR - iwl_rx()             Detach iwl_rx_mem_buffers from pool up to the
96  *                            READ INDEX, detaching the SKB from the pool.
97  *                            Moves the packet buffer from queue to rx_used.
98  *                            Posts and claims requests to the allocator.
99  *                            Calls iwl_pcie_rxq_restock to refill any empty
100  *                            slots.
101  *
102  * RBD life-cycle:
103  *
104  * Init:
105  * rxq.pool -> rxq.rx_used -> rxq.rx_free -> rxq.queue
106  *
107  * Regular Receive interrupt:
108  * Page Stolen:
109  * rxq.queue -> rxq.rx_used -> allocator.rbd_empty ->
110  * allocator.rbd_allocated -> rxq.rx_free -> rxq.queue
111  * Page not Stolen:
112  * rxq.queue -> rxq.rx_free -> rxq.queue
113  * ...
114  *
115  */
116 
117 /*
118  * iwl_rxq_space - Return number of free slots available in queue.
119  */
iwl_rxq_space(const struct iwl_rxq * rxq)120 static int iwl_rxq_space(const struct iwl_rxq *rxq)
121 {
122 	/* Make sure rx queue size is a power of 2 */
123 	WARN_ON(rxq->queue_size & (rxq->queue_size - 1));
124 
125 	/*
126 	 * There can be up to (RX_QUEUE_SIZE - 1) free slots, to avoid ambiguity
127 	 * between empty and completely full queues.
128 	 * The following is equivalent to modulo by RX_QUEUE_SIZE and is well
129 	 * defined for negative dividends.
130 	 */
131 	return (rxq->read - rxq->write - 1) & (rxq->queue_size - 1);
132 }
133 
134 /*
135  * iwl_dma_addr2rbd_ptr - convert a DMA address to a uCode read buffer ptr
136  */
iwl_pcie_dma_addr2rbd_ptr(dma_addr_t dma_addr)137 static inline __le32 iwl_pcie_dma_addr2rbd_ptr(dma_addr_t dma_addr)
138 {
139 	return cpu_to_le32((u32)(dma_addr >> 8));
140 }
141 
142 /*
143  * iwl_pcie_rx_stop - stops the Rx DMA
144  */
iwl_pcie_rx_stop(struct iwl_trans * trans)145 int iwl_pcie_rx_stop(struct iwl_trans *trans)
146 {
147 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
148 		/* TODO: remove this once fw does it */
149 		iwl_write_umac_prph(trans, RFH_RXF_DMA_CFG_AX210, 0);
150 		return iwl_poll_umac_prph_bit(trans, RFH_GEN_STATUS_AX210,
151 					      RXF_DMA_IDLE, RXF_DMA_IDLE, 1000);
152 	} else if (trans->mac_cfg->mq_rx_supported) {
153 		iwl_write_prph(trans, RFH_RXF_DMA_CFG, 0);
154 		return iwl_poll_prph_bit(trans, RFH_GEN_STATUS,
155 					   RXF_DMA_IDLE, RXF_DMA_IDLE, 1000);
156 	} else {
157 		iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
158 		return iwl_poll_direct_bit(trans, FH_MEM_RSSR_RX_STATUS_REG,
159 					   FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE,
160 					   1000);
161 	}
162 }
163 
164 /*
165  * iwl_pcie_rxq_inc_wr_ptr - Update the write pointer for the RX queue
166  */
iwl_pcie_rxq_inc_wr_ptr(struct iwl_trans * trans,struct iwl_rxq * rxq)167 static void iwl_pcie_rxq_inc_wr_ptr(struct iwl_trans *trans,
168 				    struct iwl_rxq *rxq)
169 {
170 	u32 reg;
171 
172 	lockdep_assert_held(&rxq->lock);
173 
174 	/*
175 	 * explicitly wake up the NIC if:
176 	 * 1. shadow registers aren't enabled
177 	 * 2. there is a chance that the NIC is asleep
178 	 */
179 	if (!trans->mac_cfg->base->shadow_reg_enable &&
180 	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
181 		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
182 
183 		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
184 			IWL_DEBUG_INFO(trans, "Rx queue requesting wakeup, GP1 = 0x%x\n",
185 				       reg);
186 			iwl_set_bit(trans, CSR_GP_CNTRL,
187 				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
188 			rxq->need_update = true;
189 			return;
190 		}
191 	}
192 
193 	rxq->write_actual = round_down(rxq->write, 8);
194 	if (!trans->mac_cfg->mq_rx_supported)
195 		iwl_write32(trans, FH_RSCSR_CHNL0_WPTR, rxq->write_actual);
196 	else if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_BZ)
197 		iwl_write32(trans, HBUS_TARG_WRPTR, rxq->write_actual |
198 			    HBUS_TARG_WRPTR_RX_Q(rxq->id));
199 	else
200 		iwl_write32(trans, RFH_Q_FRBDCB_WIDX_TRG(rxq->id),
201 			    rxq->write_actual);
202 }
203 
iwl_pcie_rxq_check_wrptr(struct iwl_trans * trans)204 static void iwl_pcie_rxq_check_wrptr(struct iwl_trans *trans)
205 {
206 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
207 	int i;
208 
209 	for (i = 0; i < trans->info.num_rxqs; i++) {
210 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
211 
212 		if (!rxq->need_update)
213 			continue;
214 		spin_lock_bh(&rxq->lock);
215 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
216 		rxq->need_update = false;
217 		spin_unlock_bh(&rxq->lock);
218 	}
219 }
220 
iwl_pcie_restock_bd(struct iwl_trans * trans,struct iwl_rxq * rxq,struct iwl_rx_mem_buffer * rxb)221 static void iwl_pcie_restock_bd(struct iwl_trans *trans,
222 				struct iwl_rxq *rxq,
223 				struct iwl_rx_mem_buffer *rxb)
224 {
225 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
226 		struct iwl_rx_transfer_desc *bd = rxq->bd;
227 
228 		BUILD_BUG_ON(sizeof(*bd) != 2 * sizeof(u64));
229 
230 		bd[rxq->write].addr = cpu_to_le64(rxb->page_dma);
231 		bd[rxq->write].rbid = cpu_to_le16(rxb->vid);
232 	} else {
233 		__le64 *bd = rxq->bd;
234 
235 		bd[rxq->write] = cpu_to_le64(rxb->page_dma | rxb->vid);
236 	}
237 
238 	IWL_DEBUG_RX(trans, "Assigned virtual RB ID %u to queue %d index %d\n",
239 		     (u32)rxb->vid, rxq->id, rxq->write);
240 }
241 
242 /*
243  * iwl_pcie_rxmq_restock - restock implementation for multi-queue rx
244  */
iwl_pcie_rxmq_restock(struct iwl_trans * trans,struct iwl_rxq * rxq)245 static void iwl_pcie_rxmq_restock(struct iwl_trans *trans,
246 				  struct iwl_rxq *rxq)
247 {
248 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
249 	struct iwl_rx_mem_buffer *rxb;
250 
251 	/*
252 	 * If the device isn't enabled - no need to try to add buffers...
253 	 * This can happen when we stop the device and still have an interrupt
254 	 * pending. We stop the APM before we sync the interrupts because we
255 	 * have to (see comment there). On the other hand, since the APM is
256 	 * stopped, we cannot access the HW (in particular not prph).
257 	 * So don't try to restock if the APM has been already stopped.
258 	 */
259 	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
260 		return;
261 
262 	spin_lock_bh(&rxq->lock);
263 	while (rxq->free_count) {
264 		/* Get next free Rx buffer, remove from free list */
265 		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
266 				       list);
267 		list_del(&rxb->list);
268 		rxb->invalid = false;
269 		/* some low bits are expected to be unset (depending on hw) */
270 		WARN_ON(rxb->page_dma & trans_pcie->supported_dma_mask);
271 		/* Point to Rx buffer via next RBD in circular buffer */
272 		iwl_pcie_restock_bd(trans, rxq, rxb);
273 		rxq->write = (rxq->write + 1) & (rxq->queue_size - 1);
274 		rxq->free_count--;
275 	}
276 	spin_unlock_bh(&rxq->lock);
277 
278 	/*
279 	 * If we've added more space for the firmware to place data, tell it.
280 	 * Increment device's write pointer in multiples of 8.
281 	 */
282 	if (rxq->write_actual != (rxq->write & ~0x7)) {
283 		spin_lock_bh(&rxq->lock);
284 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
285 		spin_unlock_bh(&rxq->lock);
286 	}
287 }
288 
289 /*
290  * iwl_pcie_rxsq_restock - restock implementation for single queue rx
291  */
iwl_pcie_rxsq_restock(struct iwl_trans * trans,struct iwl_rxq * rxq)292 static void iwl_pcie_rxsq_restock(struct iwl_trans *trans,
293 				  struct iwl_rxq *rxq)
294 {
295 	struct iwl_rx_mem_buffer *rxb;
296 
297 	/*
298 	 * If the device isn't enabled - not need to try to add buffers...
299 	 * This can happen when we stop the device and still have an interrupt
300 	 * pending. We stop the APM before we sync the interrupts because we
301 	 * have to (see comment there). On the other hand, since the APM is
302 	 * stopped, we cannot access the HW (in particular not prph).
303 	 * So don't try to restock if the APM has been already stopped.
304 	 */
305 	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
306 		return;
307 
308 	spin_lock_bh(&rxq->lock);
309 	while ((iwl_rxq_space(rxq) > 0) && (rxq->free_count)) {
310 		__le32 *bd = (__le32 *)rxq->bd;
311 		/* The overwritten rxb must be a used one */
312 		rxb = rxq->queue[rxq->write];
313 		BUG_ON(rxb && rxb->page);
314 
315 		/* Get next free Rx buffer, remove from free list */
316 		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
317 				       list);
318 		list_del(&rxb->list);
319 		rxb->invalid = false;
320 
321 		/* Point to Rx buffer via next RBD in circular buffer */
322 		bd[rxq->write] = iwl_pcie_dma_addr2rbd_ptr(rxb->page_dma);
323 		rxq->queue[rxq->write] = rxb;
324 		rxq->write = (rxq->write + 1) & RX_QUEUE_MASK;
325 		rxq->free_count--;
326 	}
327 	spin_unlock_bh(&rxq->lock);
328 
329 	/* If we've added more space for the firmware to place data, tell it.
330 	 * Increment device's write pointer in multiples of 8. */
331 	if (rxq->write_actual != (rxq->write & ~0x7)) {
332 		spin_lock_bh(&rxq->lock);
333 		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
334 		spin_unlock_bh(&rxq->lock);
335 	}
336 }
337 
338 /*
339  * iwl_pcie_rxq_restock - refill RX queue from pre-allocated pool
340  *
341  * If there are slots in the RX queue that need to be restocked,
342  * and we have free pre-allocated buffers, fill the ranks as much
343  * as we can, pulling from rx_free.
344  *
345  * This moves the 'write' index forward to catch up with 'processed', and
346  * also updates the memory address in the firmware to reference the new
347  * target buffer.
348  */
349 static
iwl_pcie_rxq_restock(struct iwl_trans * trans,struct iwl_rxq * rxq)350 void iwl_pcie_rxq_restock(struct iwl_trans *trans, struct iwl_rxq *rxq)
351 {
352 	if (trans->mac_cfg->mq_rx_supported)
353 		iwl_pcie_rxmq_restock(trans, rxq);
354 	else
355 		iwl_pcie_rxsq_restock(trans, rxq);
356 }
357 
358 /*
359  * iwl_pcie_rx_alloc_page - allocates and returns a page.
360  *
361  */
iwl_pcie_rx_alloc_page(struct iwl_trans * trans,u32 * offset,gfp_t priority)362 static struct page *iwl_pcie_rx_alloc_page(struct iwl_trans *trans,
363 					   u32 *offset, gfp_t priority)
364 {
365 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
366 	unsigned int allocsize = PAGE_SIZE << trans_pcie->rx_page_order;
367 	unsigned int rbsize = trans_pcie->rx_buf_bytes;
368 	struct page *page;
369 	gfp_t gfp_mask = priority;
370 
371 	if (trans_pcie->rx_page_order > 0)
372 		gfp_mask |= __GFP_COMP;
373 
374 	if (trans_pcie->alloc_page) {
375 		spin_lock_bh(&trans_pcie->alloc_page_lock);
376 		/* recheck */
377 		if (trans_pcie->alloc_page) {
378 			*offset = trans_pcie->alloc_page_used;
379 			page = trans_pcie->alloc_page;
380 			trans_pcie->alloc_page_used += rbsize;
381 			if (trans_pcie->alloc_page_used >= allocsize)
382 				trans_pcie->alloc_page = NULL;
383 			else
384 				get_page(page);
385 			spin_unlock_bh(&trans_pcie->alloc_page_lock);
386 			return page;
387 		}
388 		spin_unlock_bh(&trans_pcie->alloc_page_lock);
389 	}
390 
391 	/* Alloc a new receive buffer */
392 	page = alloc_pages(gfp_mask, trans_pcie->rx_page_order);
393 	if (!page) {
394 		if (net_ratelimit())
395 			IWL_DEBUG_INFO(trans, "alloc_pages failed, order: %d\n",
396 				       trans_pcie->rx_page_order);
397 		/*
398 		 * Issue an error if we don't have enough pre-allocated
399 		  * buffers.
400 		 */
401 		if (!(gfp_mask & __GFP_NOWARN) && net_ratelimit())
402 			IWL_CRIT(trans,
403 				 "Failed to alloc_pages\n");
404 		return NULL;
405 	}
406 
407 	if (2 * rbsize <= allocsize) {
408 		spin_lock_bh(&trans_pcie->alloc_page_lock);
409 		if (!trans_pcie->alloc_page) {
410 			get_page(page);
411 			trans_pcie->alloc_page = page;
412 			trans_pcie->alloc_page_used = rbsize;
413 		}
414 		spin_unlock_bh(&trans_pcie->alloc_page_lock);
415 	}
416 
417 	*offset = 0;
418 	return page;
419 }
420 
421 /*
422  * iwl_pcie_rxq_alloc_rbs - allocate a page for each used RBD
423  *
424  * A used RBD is an Rx buffer that has been given to the stack. To use it again
425  * a page must be allocated and the RBD must point to the page. This function
426  * doesn't change the HW pointer but handles the list of pages that is used by
427  * iwl_pcie_rxq_restock. The latter function will update the HW to use the newly
428  * allocated buffers.
429  */
iwl_pcie_rxq_alloc_rbs(struct iwl_trans * trans,gfp_t priority,struct iwl_rxq * rxq)430 void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans, gfp_t priority,
431 			    struct iwl_rxq *rxq)
432 {
433 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
434 	struct iwl_rx_mem_buffer *rxb;
435 	struct page *page;
436 
437 	while (1) {
438 		unsigned int offset;
439 
440 		spin_lock_bh(&rxq->lock);
441 		if (list_empty(&rxq->rx_used)) {
442 			spin_unlock_bh(&rxq->lock);
443 			return;
444 		}
445 		spin_unlock_bh(&rxq->lock);
446 
447 		page = iwl_pcie_rx_alloc_page(trans, &offset, priority);
448 		if (!page)
449 			return;
450 
451 		spin_lock_bh(&rxq->lock);
452 
453 		if (list_empty(&rxq->rx_used)) {
454 			spin_unlock_bh(&rxq->lock);
455 			__free_pages(page, trans_pcie->rx_page_order);
456 			return;
457 		}
458 		rxb = list_first_entry(&rxq->rx_used, struct iwl_rx_mem_buffer,
459 				       list);
460 		list_del(&rxb->list);
461 		spin_unlock_bh(&rxq->lock);
462 
463 		BUG_ON(rxb->page);
464 		rxb->page = page;
465 		rxb->offset = offset;
466 		/* Get physical address of the RB */
467 		rxb->page_dma =
468 			dma_map_page(trans->dev, page, rxb->offset,
469 				     trans_pcie->rx_buf_bytes,
470 				     DMA_FROM_DEVICE);
471 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
472 			rxb->page = NULL;
473 			spin_lock_bh(&rxq->lock);
474 			list_add(&rxb->list, &rxq->rx_used);
475 			spin_unlock_bh(&rxq->lock);
476 			__free_pages(page, trans_pcie->rx_page_order);
477 			return;
478 		}
479 
480 		spin_lock_bh(&rxq->lock);
481 
482 		list_add_tail(&rxb->list, &rxq->rx_free);
483 		rxq->free_count++;
484 
485 		spin_unlock_bh(&rxq->lock);
486 	}
487 }
488 
iwl_pcie_free_rbs_pool(struct iwl_trans * trans)489 void iwl_pcie_free_rbs_pool(struct iwl_trans *trans)
490 {
491 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
492 	int i;
493 
494 	if (!trans_pcie->rx_pool)
495 		return;
496 
497 	for (i = 0; i < RX_POOL_SIZE(trans_pcie->num_rx_bufs); i++) {
498 		if (!trans_pcie->rx_pool[i].page)
499 			continue;
500 		dma_unmap_page(trans->dev, trans_pcie->rx_pool[i].page_dma,
501 			       trans_pcie->rx_buf_bytes, DMA_FROM_DEVICE);
502 		__free_pages(trans_pcie->rx_pool[i].page,
503 			     trans_pcie->rx_page_order);
504 		trans_pcie->rx_pool[i].page = NULL;
505 	}
506 }
507 
508 /*
509  * iwl_pcie_rx_allocator - Allocates pages in the background for RX queues
510  *
511  * Allocates for each received request 8 pages
512  * Called as a scheduled work item.
513  */
iwl_pcie_rx_allocator(struct iwl_trans * trans)514 static void iwl_pcie_rx_allocator(struct iwl_trans *trans)
515 {
516 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
517 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
518 	struct list_head local_empty;
519 	int pending = atomic_read(&rba->req_pending);
520 
521 	IWL_DEBUG_TPT(trans, "Pending allocation requests = %d\n", pending);
522 
523 	/* If we were scheduled - there is at least one request */
524 	spin_lock_bh(&rba->lock);
525 	/* swap out the rba->rbd_empty to a local list */
526 	list_replace_init(&rba->rbd_empty, &local_empty);
527 	spin_unlock_bh(&rba->lock);
528 
529 	while (pending) {
530 		int i;
531 		LIST_HEAD(local_allocated);
532 		gfp_t gfp_mask = GFP_KERNEL;
533 
534 		/* Do not post a warning if there are only a few requests */
535 		if (pending < RX_PENDING_WATERMARK)
536 			gfp_mask |= __GFP_NOWARN;
537 
538 		for (i = 0; i < RX_CLAIM_REQ_ALLOC;) {
539 			struct iwl_rx_mem_buffer *rxb;
540 			struct page *page;
541 
542 			/* List should never be empty - each reused RBD is
543 			 * returned to the list, and initial pool covers any
544 			 * possible gap between the time the page is allocated
545 			 * to the time the RBD is added.
546 			 */
547 			BUG_ON(list_empty(&local_empty));
548 			/* Get the first rxb from the rbd list */
549 			rxb = list_first_entry(&local_empty,
550 					       struct iwl_rx_mem_buffer, list);
551 			BUG_ON(rxb->page);
552 
553 			/* Alloc a new receive buffer */
554 			page = iwl_pcie_rx_alloc_page(trans, &rxb->offset,
555 						      gfp_mask);
556 			if (!page)
557 				continue;
558 			rxb->page = page;
559 
560 			/* Get physical address of the RB */
561 			rxb->page_dma = dma_map_page(trans->dev, page,
562 						     rxb->offset,
563 						     trans_pcie->rx_buf_bytes,
564 						     DMA_FROM_DEVICE);
565 			if (dma_mapping_error(trans->dev, rxb->page_dma)) {
566 				rxb->page = NULL;
567 				__free_pages(page, trans_pcie->rx_page_order);
568 				continue;
569 			}
570 
571 			/* move the allocated entry to the out list */
572 			list_move(&rxb->list, &local_allocated);
573 			i++;
574 		}
575 
576 		atomic_dec(&rba->req_pending);
577 		pending--;
578 
579 		if (!pending) {
580 			pending = atomic_read(&rba->req_pending);
581 			if (pending)
582 				IWL_DEBUG_TPT(trans,
583 					      "Got more pending allocation requests = %d\n",
584 					      pending);
585 		}
586 
587 		spin_lock_bh(&rba->lock);
588 		/* add the allocated rbds to the allocator allocated list */
589 		list_splice_tail(&local_allocated, &rba->rbd_allocated);
590 		/* get more empty RBDs for current pending requests */
591 		list_splice_tail_init(&rba->rbd_empty, &local_empty);
592 		spin_unlock_bh(&rba->lock);
593 
594 		atomic_inc(&rba->req_ready);
595 
596 	}
597 
598 	spin_lock_bh(&rba->lock);
599 	/* return unused rbds to the allocator empty list */
600 	list_splice_tail(&local_empty, &rba->rbd_empty);
601 	spin_unlock_bh(&rba->lock);
602 
603 	IWL_DEBUG_TPT(trans, "%s, exit.\n", __func__);
604 }
605 
606 /*
607  * iwl_pcie_rx_allocator_get - returns the pre-allocated pages
608 .*
609 .* Called by queue when the queue posted allocation request and
610  * has freed 8 RBDs in order to restock itself.
611  * This function directly moves the allocated RBs to the queue's ownership
612  * and updates the relevant counters.
613  */
iwl_pcie_rx_allocator_get(struct iwl_trans * trans,struct iwl_rxq * rxq)614 static void iwl_pcie_rx_allocator_get(struct iwl_trans *trans,
615 				      struct iwl_rxq *rxq)
616 {
617 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
618 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
619 	int i;
620 
621 	lockdep_assert_held(&rxq->lock);
622 
623 	/*
624 	 * atomic_dec_if_positive returns req_ready - 1 for any scenario.
625 	 * If req_ready is 0 atomic_dec_if_positive will return -1 and this
626 	 * function will return early, as there are no ready requests.
627 	 * atomic_dec_if_positive will perofrm the *actual* decrement only if
628 	 * req_ready > 0, i.e. - there are ready requests and the function
629 	 * hands one request to the caller.
630 	 */
631 	if (atomic_dec_if_positive(&rba->req_ready) < 0)
632 		return;
633 
634 	spin_lock(&rba->lock);
635 	for (i = 0; i < RX_CLAIM_REQ_ALLOC; i++) {
636 		/* Get next free Rx buffer, remove it from free list */
637 		struct iwl_rx_mem_buffer *rxb =
638 			list_first_entry(&rba->rbd_allocated,
639 					 struct iwl_rx_mem_buffer, list);
640 
641 		list_move(&rxb->list, &rxq->rx_free);
642 	}
643 	spin_unlock(&rba->lock);
644 
645 	rxq->used_count -= RX_CLAIM_REQ_ALLOC;
646 	rxq->free_count += RX_CLAIM_REQ_ALLOC;
647 }
648 
iwl_pcie_rx_allocator_work(struct work_struct * data)649 void iwl_pcie_rx_allocator_work(struct work_struct *data)
650 {
651 	struct iwl_rb_allocator *rba_p =
652 		container_of(data, struct iwl_rb_allocator, rx_alloc);
653 	struct iwl_trans_pcie *trans_pcie =
654 		container_of(rba_p, struct iwl_trans_pcie, rba);
655 
656 	iwl_pcie_rx_allocator(trans_pcie->trans);
657 }
658 
iwl_pcie_free_bd_size(struct iwl_trans * trans)659 static int iwl_pcie_free_bd_size(struct iwl_trans *trans)
660 {
661 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_AX210)
662 		return sizeof(struct iwl_rx_transfer_desc);
663 
664 	return trans->mac_cfg->mq_rx_supported ?
665 			sizeof(__le64) : sizeof(__le32);
666 }
667 
iwl_pcie_used_bd_size(struct iwl_trans * trans)668 static int iwl_pcie_used_bd_size(struct iwl_trans *trans)
669 {
670 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_BZ)
671 		return sizeof(struct iwl_rx_completion_desc_bz);
672 
673 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_AX210)
674 		return sizeof(struct iwl_rx_completion_desc);
675 
676 	return sizeof(__le32);
677 }
678 
iwl_pcie_free_rxq_dma(struct iwl_trans * trans,struct iwl_rxq * rxq)679 static void iwl_pcie_free_rxq_dma(struct iwl_trans *trans,
680 				  struct iwl_rxq *rxq)
681 {
682 	int free_size = iwl_pcie_free_bd_size(trans);
683 
684 	if (rxq->bd)
685 		dma_free_coherent(trans->dev,
686 				  free_size * rxq->queue_size,
687 				  rxq->bd, rxq->bd_dma);
688 	rxq->bd_dma = 0;
689 	rxq->bd = NULL;
690 
691 	rxq->rb_stts_dma = 0;
692 	rxq->rb_stts = NULL;
693 
694 	if (rxq->used_bd)
695 		dma_free_coherent(trans->dev,
696 				  iwl_pcie_used_bd_size(trans) *
697 					rxq->queue_size,
698 				  rxq->used_bd, rxq->used_bd_dma);
699 	rxq->used_bd_dma = 0;
700 	rxq->used_bd = NULL;
701 }
702 
iwl_pcie_rb_stts_size(struct iwl_trans * trans)703 static size_t iwl_pcie_rb_stts_size(struct iwl_trans *trans)
704 {
705 	bool use_rx_td = (trans->mac_cfg->device_family >=
706 			  IWL_DEVICE_FAMILY_AX210);
707 
708 	if (use_rx_td)
709 		return sizeof(__le16);
710 
711 	return sizeof(struct iwl_rb_status);
712 }
713 
iwl_pcie_alloc_rxq_dma(struct iwl_trans * trans,struct iwl_rxq * rxq)714 static int iwl_pcie_alloc_rxq_dma(struct iwl_trans *trans,
715 				  struct iwl_rxq *rxq)
716 {
717 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
718 	size_t rb_stts_size = iwl_pcie_rb_stts_size(trans);
719 	struct device *dev = trans->dev;
720 	int i;
721 	int free_size;
722 
723 	spin_lock_init(&rxq->lock);
724 	if (trans->mac_cfg->mq_rx_supported)
725 		rxq->queue_size = iwl_trans_get_num_rbds(trans);
726 	else
727 		rxq->queue_size = RX_QUEUE_SIZE;
728 
729 	free_size = iwl_pcie_free_bd_size(trans);
730 
731 	/*
732 	 * Allocate the circular buffer of Read Buffer Descriptors
733 	 * (RBDs)
734 	 */
735 	rxq->bd = dma_alloc_coherent(dev, free_size * rxq->queue_size,
736 				     &rxq->bd_dma, GFP_KERNEL);
737 	if (!rxq->bd)
738 		goto err;
739 
740 	if (trans->mac_cfg->mq_rx_supported) {
741 		rxq->used_bd = dma_alloc_coherent(dev,
742 						  iwl_pcie_used_bd_size(trans) *
743 							rxq->queue_size,
744 						  &rxq->used_bd_dma,
745 						  GFP_KERNEL);
746 		if (!rxq->used_bd)
747 			goto err;
748 	}
749 
750 	rxq->rb_stts = (u8 *)trans_pcie->base_rb_stts + rxq->id * rb_stts_size;
751 	rxq->rb_stts_dma =
752 		trans_pcie->base_rb_stts_dma + rxq->id * rb_stts_size;
753 
754 	return 0;
755 
756 err:
757 	for (i = 0; i < trans->info.num_rxqs; i++) {
758 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
759 
760 		iwl_pcie_free_rxq_dma(trans, rxq);
761 	}
762 
763 	return -ENOMEM;
764 }
765 
iwl_pcie_rx_alloc(struct iwl_trans * trans)766 static int iwl_pcie_rx_alloc(struct iwl_trans *trans)
767 {
768 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
769 	size_t rb_stts_size = iwl_pcie_rb_stts_size(trans);
770 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
771 	int i, ret;
772 
773 	if (WARN_ON(trans_pcie->rxq))
774 		return -EINVAL;
775 
776 	trans_pcie->rxq = kcalloc(trans->info.num_rxqs, sizeof(struct iwl_rxq),
777 				  GFP_KERNEL);
778 	trans_pcie->rx_pool = kcalloc(RX_POOL_SIZE(trans_pcie->num_rx_bufs),
779 				      sizeof(trans_pcie->rx_pool[0]),
780 				      GFP_KERNEL);
781 	trans_pcie->global_table =
782 		kcalloc(RX_POOL_SIZE(trans_pcie->num_rx_bufs),
783 			sizeof(trans_pcie->global_table[0]),
784 			GFP_KERNEL);
785 	if (!trans_pcie->rxq || !trans_pcie->rx_pool ||
786 	    !trans_pcie->global_table) {
787 		ret = -ENOMEM;
788 		goto err;
789 	}
790 
791 	spin_lock_init(&rba->lock);
792 
793 	/*
794 	 * Allocate the driver's pointer to receive buffer status.
795 	 * Allocate for all queues continuously (HW requirement).
796 	 */
797 	trans_pcie->base_rb_stts =
798 			dma_alloc_coherent(trans->dev,
799 					   rb_stts_size * trans->info.num_rxqs,
800 					   &trans_pcie->base_rb_stts_dma,
801 					   GFP_KERNEL);
802 	if (!trans_pcie->base_rb_stts) {
803 		ret = -ENOMEM;
804 		goto err;
805 	}
806 
807 	for (i = 0; i < trans->info.num_rxqs; i++) {
808 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
809 
810 		rxq->id = i;
811 		ret = iwl_pcie_alloc_rxq_dma(trans, rxq);
812 		if (ret)
813 			goto err;
814 	}
815 	return 0;
816 
817 err:
818 	if (trans_pcie->base_rb_stts) {
819 		dma_free_coherent(trans->dev,
820 				  rb_stts_size * trans->info.num_rxqs,
821 				  trans_pcie->base_rb_stts,
822 				  trans_pcie->base_rb_stts_dma);
823 		trans_pcie->base_rb_stts = NULL;
824 		trans_pcie->base_rb_stts_dma = 0;
825 	}
826 	kfree(trans_pcie->rx_pool);
827 	trans_pcie->rx_pool = NULL;
828 	kfree(trans_pcie->global_table);
829 	trans_pcie->global_table = NULL;
830 	kfree(trans_pcie->rxq);
831 	trans_pcie->rxq = NULL;
832 
833 	return ret;
834 }
835 
iwl_pcie_rx_hw_init(struct iwl_trans * trans,struct iwl_rxq * rxq)836 static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
837 {
838 	u32 rb_size;
839 	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
840 
841 	switch (trans->conf.rx_buf_size) {
842 	case IWL_AMSDU_4K:
843 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
844 		break;
845 	case IWL_AMSDU_8K:
846 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
847 		break;
848 	case IWL_AMSDU_12K:
849 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_12K;
850 		break;
851 	default:
852 		WARN_ON(1);
853 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
854 	}
855 
856 	if (!iwl_trans_grab_nic_access(trans))
857 		return;
858 
859 	/* Stop Rx DMA */
860 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
861 	/* reset and flush pointers */
862 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
863 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
864 	iwl_write32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
865 
866 	/* Reset driver's Rx queue write index */
867 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
868 
869 	/* Tell device where to find RBD circular buffer in DRAM */
870 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
871 		    (u32)(rxq->bd_dma >> 8));
872 
873 	/* Tell device where in DRAM to update its Rx status */
874 	iwl_write32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
875 		    rxq->rb_stts_dma >> 4);
876 
877 	/* Enable Rx DMA
878 	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
879 	 *      the credit mechanism in 5000 HW RX FIFO
880 	 * Direct rx interrupts to hosts
881 	 * Rx buffer size 4 or 8k or 12k
882 	 * RB timeout 0x10
883 	 * 256 RBDs
884 	 */
885 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
886 		    FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
887 		    FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
888 		    FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
889 		    rb_size |
890 		    (RX_RB_TIMEOUT << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS) |
891 		    (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));
892 
893 	iwl_trans_release_nic_access(trans);
894 
895 	/* Set interrupt coalescing timer to default (2048 usecs) */
896 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
897 
898 	/* W/A for interrupt coalescing bug in 7260 and 3160 */
899 	if (trans->cfg->host_interrupt_operation_mode)
900 		iwl_set_bit(trans, CSR_INT_COALESCING, IWL_HOST_INT_OPER_MODE);
901 }
902 
iwl_pcie_rx_mq_hw_init(struct iwl_trans * trans)903 static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
904 {
905 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
906 	u32 rb_size, enabled = 0;
907 	int i;
908 
909 	switch (trans->conf.rx_buf_size) {
910 	case IWL_AMSDU_2K:
911 		rb_size = RFH_RXF_DMA_RB_SIZE_2K;
912 		break;
913 	case IWL_AMSDU_4K:
914 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
915 		break;
916 	case IWL_AMSDU_8K:
917 		rb_size = RFH_RXF_DMA_RB_SIZE_8K;
918 		break;
919 	case IWL_AMSDU_12K:
920 		rb_size = RFH_RXF_DMA_RB_SIZE_12K;
921 		break;
922 	default:
923 		WARN_ON(1);
924 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
925 	}
926 
927 	if (!iwl_trans_grab_nic_access(trans))
928 		return;
929 
930 	/* Stop Rx DMA */
931 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG, 0);
932 	/* disable free amd used rx queue operation */
933 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, 0);
934 
935 	for (i = 0; i < trans->info.num_rxqs; i++) {
936 		/* Tell device where to find RBD free table in DRAM */
937 		iwl_write_prph64_no_grab(trans,
938 					 RFH_Q_FRBDCB_BA_LSB(i),
939 					 trans_pcie->rxq[i].bd_dma);
940 		/* Tell device where to find RBD used table in DRAM */
941 		iwl_write_prph64_no_grab(trans,
942 					 RFH_Q_URBDCB_BA_LSB(i),
943 					 trans_pcie->rxq[i].used_bd_dma);
944 		/* Tell device where in DRAM to update its Rx status */
945 		iwl_write_prph64_no_grab(trans,
946 					 RFH_Q_URBD_STTS_WPTR_LSB(i),
947 					 trans_pcie->rxq[i].rb_stts_dma);
948 		/* Reset device indice tables */
949 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_WIDX(i), 0);
950 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_RIDX(i), 0);
951 		iwl_write_prph_no_grab(trans, RFH_Q_URBDCB_WIDX(i), 0);
952 
953 		enabled |= BIT(i) | BIT(i + 16);
954 	}
955 
956 	/*
957 	 * Enable Rx DMA
958 	 * Rx buffer size 4 or 8k or 12k
959 	 * Min RB size 4 or 8
960 	 * Drop frames that exceed RB size
961 	 * 512 RBDs
962 	 */
963 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG,
964 			       RFH_DMA_EN_ENABLE_VAL | rb_size |
965 			       RFH_RXF_DMA_MIN_RB_4_8 |
966 			       RFH_RXF_DMA_DROP_TOO_LARGE_MASK |
967 			       RFH_RXF_DMA_RBDCB_SIZE_512);
968 
969 	/*
970 	 * Activate DMA snooping.
971 	 * Set RX DMA chunk size to 64B for IOSF and 128B for PCIe
972 	 * Default queue is 0
973 	 */
974 	iwl_write_prph_no_grab(trans, RFH_GEN_CFG,
975 			       RFH_GEN_CFG_RFH_DMA_SNOOP |
976 			       RFH_GEN_CFG_VAL(DEFAULT_RXQ_NUM, 0) |
977 			       RFH_GEN_CFG_SERVICE_DMA_SNOOP |
978 			       RFH_GEN_CFG_VAL(RB_CHUNK_SIZE,
979 					       trans->mac_cfg->integrated ?
980 					       RFH_GEN_CFG_RB_CHUNK_SIZE_64 :
981 					       RFH_GEN_CFG_RB_CHUNK_SIZE_128));
982 	/* Enable the relevant rx queues */
983 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, enabled);
984 
985 	iwl_trans_release_nic_access(trans);
986 
987 	/* Set interrupt coalescing timer to default (2048 usecs) */
988 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
989 }
990 
iwl_pcie_rx_init_rxb_lists(struct iwl_rxq * rxq)991 void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq)
992 {
993 	lockdep_assert_held(&rxq->lock);
994 
995 	INIT_LIST_HEAD(&rxq->rx_free);
996 	INIT_LIST_HEAD(&rxq->rx_used);
997 	rxq->free_count = 0;
998 	rxq->used_count = 0;
999 }
1000 
1001 static int iwl_pcie_rx_handle(struct iwl_trans *trans, int queue, int budget);
1002 
iwl_netdev_to_trans_pcie(struct net_device * dev)1003 static inline struct iwl_trans_pcie *iwl_netdev_to_trans_pcie(struct net_device *dev)
1004 {
1005 	return *(struct iwl_trans_pcie **)netdev_priv(dev);
1006 }
1007 
iwl_pcie_napi_poll(struct napi_struct * napi,int budget)1008 static int iwl_pcie_napi_poll(struct napi_struct *napi, int budget)
1009 {
1010 	struct iwl_rxq *rxq = container_of(napi, struct iwl_rxq, napi);
1011 	struct iwl_trans_pcie *trans_pcie;
1012 	struct iwl_trans *trans;
1013 	int ret;
1014 
1015 	trans_pcie = iwl_netdev_to_trans_pcie(napi->dev);
1016 	trans = trans_pcie->trans;
1017 
1018 	ret = iwl_pcie_rx_handle(trans, rxq->id, budget);
1019 
1020 	IWL_DEBUG_ISR(trans, "[%d] handled %d, budget %d\n",
1021 		      rxq->id, ret, budget);
1022 
1023 	if (ret < budget) {
1024 		spin_lock(&trans_pcie->irq_lock);
1025 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1026 			_iwl_enable_interrupts(trans);
1027 		spin_unlock(&trans_pcie->irq_lock);
1028 
1029 		napi_complete_done(&rxq->napi, ret);
1030 	}
1031 
1032 	return ret;
1033 }
1034 
iwl_pcie_napi_poll_msix(struct napi_struct * napi,int budget)1035 static int iwl_pcie_napi_poll_msix(struct napi_struct *napi, int budget)
1036 {
1037 	struct iwl_rxq *rxq = container_of(napi, struct iwl_rxq, napi);
1038 	struct iwl_trans_pcie *trans_pcie;
1039 	struct iwl_trans *trans;
1040 	int ret;
1041 
1042 	trans_pcie = iwl_netdev_to_trans_pcie(napi->dev);
1043 	trans = trans_pcie->trans;
1044 
1045 	ret = iwl_pcie_rx_handle(trans, rxq->id, budget);
1046 	IWL_DEBUG_ISR(trans, "[%d] handled %d, budget %d\n", rxq->id, ret,
1047 		      budget);
1048 
1049 	if (ret < budget) {
1050 		int irq_line = rxq->id;
1051 
1052 		/* FIRST_RSS is shared with line 0 */
1053 		if (trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS &&
1054 		    rxq->id == 1)
1055 			irq_line = 0;
1056 
1057 		spin_lock(&trans_pcie->irq_lock);
1058 		iwl_pcie_clear_irq(trans, irq_line);
1059 		spin_unlock(&trans_pcie->irq_lock);
1060 
1061 		napi_complete_done(&rxq->napi, ret);
1062 	}
1063 
1064 	return ret;
1065 }
1066 
iwl_pcie_rx_napi_sync(struct iwl_trans * trans)1067 void iwl_pcie_rx_napi_sync(struct iwl_trans *trans)
1068 {
1069 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1070 	int i;
1071 
1072 	if (unlikely(!trans_pcie->rxq))
1073 		return;
1074 
1075 	for (i = 0; i < trans->info.num_rxqs; i++) {
1076 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1077 
1078 		if (rxq && rxq->napi.poll)
1079 			napi_synchronize(&rxq->napi);
1080 	}
1081 }
1082 
_iwl_pcie_rx_init(struct iwl_trans * trans)1083 static int _iwl_pcie_rx_init(struct iwl_trans *trans)
1084 {
1085 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1086 	struct iwl_rxq *def_rxq;
1087 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1088 	int i, err, queue_size, allocator_pool_size, num_alloc;
1089 
1090 	if (!trans_pcie->rxq) {
1091 		err = iwl_pcie_rx_alloc(trans);
1092 		if (err)
1093 			return err;
1094 	}
1095 	def_rxq = trans_pcie->rxq;
1096 
1097 	cancel_work_sync(&rba->rx_alloc);
1098 
1099 	spin_lock_bh(&rba->lock);
1100 	atomic_set(&rba->req_pending, 0);
1101 	atomic_set(&rba->req_ready, 0);
1102 	INIT_LIST_HEAD(&rba->rbd_allocated);
1103 	INIT_LIST_HEAD(&rba->rbd_empty);
1104 	spin_unlock_bh(&rba->lock);
1105 
1106 	/* free all first - we overwrite everything here */
1107 	iwl_pcie_free_rbs_pool(trans);
1108 
1109 	for (i = 0; i < RX_QUEUE_SIZE; i++)
1110 		def_rxq->queue[i] = NULL;
1111 
1112 	for (i = 0; i < trans->info.num_rxqs; i++) {
1113 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1114 
1115 		spin_lock_bh(&rxq->lock);
1116 		/*
1117 		 * Set read write pointer to reflect that we have processed
1118 		 * and used all buffers, but have not restocked the Rx queue
1119 		 * with fresh buffers
1120 		 */
1121 		rxq->read = 0;
1122 		rxq->write = 0;
1123 		rxq->write_actual = 0;
1124 		memset(rxq->rb_stts, 0,
1125 		       (trans->mac_cfg->device_family >=
1126 			IWL_DEVICE_FAMILY_AX210) ?
1127 		       sizeof(__le16) : sizeof(struct iwl_rb_status));
1128 
1129 		iwl_pcie_rx_init_rxb_lists(rxq);
1130 
1131 		spin_unlock_bh(&rxq->lock);
1132 
1133 		if (!rxq->napi.poll) {
1134 			int (*poll)(struct napi_struct *, int) = iwl_pcie_napi_poll;
1135 
1136 			if (trans_pcie->msix_enabled)
1137 				poll = iwl_pcie_napi_poll_msix;
1138 
1139 			netif_napi_add(trans_pcie->napi_dev, &rxq->napi,
1140 				       poll);
1141 			napi_enable(&rxq->napi);
1142 		}
1143 
1144 	}
1145 
1146 	/* move the pool to the default queue and allocator ownerships */
1147 	queue_size = trans->mac_cfg->mq_rx_supported ?
1148 			trans_pcie->num_rx_bufs - 1 : RX_QUEUE_SIZE;
1149 	allocator_pool_size = trans->info.num_rxqs *
1150 		(RX_CLAIM_REQ_ALLOC - RX_POST_REQ_ALLOC);
1151 	num_alloc = queue_size + allocator_pool_size;
1152 
1153 	for (i = 0; i < num_alloc; i++) {
1154 		struct iwl_rx_mem_buffer *rxb = &trans_pcie->rx_pool[i];
1155 
1156 		if (i < allocator_pool_size)
1157 			list_add(&rxb->list, &rba->rbd_empty);
1158 		else
1159 			list_add(&rxb->list, &def_rxq->rx_used);
1160 		trans_pcie->global_table[i] = rxb;
1161 		rxb->vid = (u16)(i + 1);
1162 		rxb->invalid = true;
1163 	}
1164 
1165 	iwl_pcie_rxq_alloc_rbs(trans, GFP_KERNEL, def_rxq);
1166 
1167 	return 0;
1168 }
1169 
iwl_pcie_rx_init(struct iwl_trans * trans)1170 int iwl_pcie_rx_init(struct iwl_trans *trans)
1171 {
1172 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1173 	int ret = _iwl_pcie_rx_init(trans);
1174 
1175 	if (ret)
1176 		return ret;
1177 
1178 	if (trans->mac_cfg->mq_rx_supported)
1179 		iwl_pcie_rx_mq_hw_init(trans);
1180 	else
1181 		iwl_pcie_rx_hw_init(trans, trans_pcie->rxq);
1182 
1183 	iwl_pcie_rxq_restock(trans, trans_pcie->rxq);
1184 
1185 	spin_lock_bh(&trans_pcie->rxq->lock);
1186 	iwl_pcie_rxq_inc_wr_ptr(trans, trans_pcie->rxq);
1187 	spin_unlock_bh(&trans_pcie->rxq->lock);
1188 
1189 	return 0;
1190 }
1191 
iwl_pcie_gen2_rx_init(struct iwl_trans * trans)1192 int iwl_pcie_gen2_rx_init(struct iwl_trans *trans)
1193 {
1194 	/* Set interrupt coalescing timer to default (2048 usecs) */
1195 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
1196 
1197 	/*
1198 	 * We don't configure the RFH.
1199 	 * Restock will be done at alive, after firmware configured the RFH.
1200 	 */
1201 	return _iwl_pcie_rx_init(trans);
1202 }
1203 
iwl_pcie_rx_free(struct iwl_trans * trans)1204 void iwl_pcie_rx_free(struct iwl_trans *trans)
1205 {
1206 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1207 	size_t rb_stts_size = iwl_pcie_rb_stts_size(trans);
1208 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1209 	int i;
1210 
1211 	/*
1212 	 * if rxq is NULL, it means that nothing has been allocated,
1213 	 * exit now
1214 	 */
1215 	if (!trans_pcie->rxq) {
1216 		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
1217 		return;
1218 	}
1219 
1220 	cancel_work_sync(&rba->rx_alloc);
1221 
1222 	iwl_pcie_free_rbs_pool(trans);
1223 
1224 	if (trans_pcie->base_rb_stts) {
1225 		dma_free_coherent(trans->dev,
1226 				  rb_stts_size * trans->info.num_rxqs,
1227 				  trans_pcie->base_rb_stts,
1228 				  trans_pcie->base_rb_stts_dma);
1229 		trans_pcie->base_rb_stts = NULL;
1230 		trans_pcie->base_rb_stts_dma = 0;
1231 	}
1232 
1233 	for (i = 0; i < trans->info.num_rxqs; i++) {
1234 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1235 
1236 		iwl_pcie_free_rxq_dma(trans, rxq);
1237 
1238 		if (rxq->napi.poll) {
1239 			napi_disable(&rxq->napi);
1240 			netif_napi_del(&rxq->napi);
1241 		}
1242 	}
1243 	kfree(trans_pcie->rx_pool);
1244 	kfree(trans_pcie->global_table);
1245 	kfree(trans_pcie->rxq);
1246 
1247 	if (trans_pcie->alloc_page)
1248 		__free_pages(trans_pcie->alloc_page, trans_pcie->rx_page_order);
1249 }
1250 
iwl_pcie_rx_move_to_allocator(struct iwl_rxq * rxq,struct iwl_rb_allocator * rba)1251 static void iwl_pcie_rx_move_to_allocator(struct iwl_rxq *rxq,
1252 					  struct iwl_rb_allocator *rba)
1253 {
1254 	spin_lock(&rba->lock);
1255 	list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
1256 	spin_unlock(&rba->lock);
1257 }
1258 
1259 /*
1260  * iwl_pcie_rx_reuse_rbd - Recycle used RBDs
1261  *
1262  * Called when a RBD can be reused. The RBD is transferred to the allocator.
1263  * When there are 2 empty RBDs - a request for allocation is posted
1264  */
iwl_pcie_rx_reuse_rbd(struct iwl_trans * trans,struct iwl_rx_mem_buffer * rxb,struct iwl_rxq * rxq,bool emergency)1265 static void iwl_pcie_rx_reuse_rbd(struct iwl_trans *trans,
1266 				  struct iwl_rx_mem_buffer *rxb,
1267 				  struct iwl_rxq *rxq, bool emergency)
1268 {
1269 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1270 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1271 
1272 	/* Move the RBD to the used list, will be moved to allocator in batches
1273 	 * before claiming or posting a request*/
1274 	list_add_tail(&rxb->list, &rxq->rx_used);
1275 
1276 	if (unlikely(emergency))
1277 		return;
1278 
1279 	/* Count the allocator owned RBDs */
1280 	rxq->used_count++;
1281 
1282 	/* If we have RX_POST_REQ_ALLOC new released rx buffers -
1283 	 * issue a request for allocator. Modulo RX_CLAIM_REQ_ALLOC is
1284 	 * used for the case we failed to claim RX_CLAIM_REQ_ALLOC,
1285 	 * after but we still need to post another request.
1286 	 */
1287 	if ((rxq->used_count % RX_CLAIM_REQ_ALLOC) == RX_POST_REQ_ALLOC) {
1288 		/* Move the 2 RBDs to the allocator ownership.
1289 		 Allocator has another 6 from pool for the request completion*/
1290 		iwl_pcie_rx_move_to_allocator(rxq, rba);
1291 
1292 		atomic_inc(&rba->req_pending);
1293 		queue_work(rba->alloc_wq, &rba->rx_alloc);
1294 	}
1295 }
1296 
iwl_pcie_rx_handle_rb(struct iwl_trans * trans,struct iwl_rxq * rxq,struct iwl_rx_mem_buffer * rxb,bool emergency,int i)1297 static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
1298 				struct iwl_rxq *rxq,
1299 				struct iwl_rx_mem_buffer *rxb,
1300 				bool emergency,
1301 				int i)
1302 {
1303 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1304 	struct iwl_txq *txq = trans_pcie->txqs.txq[trans->conf.cmd_queue];
1305 	bool page_stolen = false;
1306 	int max_len = trans_pcie->rx_buf_bytes;
1307 	u32 offset = 0;
1308 
1309 	if (WARN_ON(!rxb))
1310 		return;
1311 
1312 	dma_unmap_page(trans->dev, rxb->page_dma, max_len, DMA_FROM_DEVICE);
1313 
1314 	while (offset + sizeof(u32) + sizeof(struct iwl_cmd_header) < max_len) {
1315 		struct iwl_rx_packet *pkt;
1316 		bool reclaim;
1317 		int len;
1318 		struct iwl_rx_cmd_buffer rxcb = {
1319 			._offset = rxb->offset + offset,
1320 			._rx_page_order = trans_pcie->rx_page_order,
1321 			._page = rxb->page,
1322 			._page_stolen = false,
1323 			.truesize = max_len,
1324 		};
1325 
1326 		pkt = rxb_addr(&rxcb);
1327 
1328 		if (pkt->len_n_flags == cpu_to_le32(FH_RSCSR_FRAME_INVALID)) {
1329 			IWL_DEBUG_RX(trans,
1330 				     "Q %d: RB end marker at offset %d\n",
1331 				     rxq->id, offset);
1332 			break;
1333 		}
1334 
1335 		WARN((le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1336 			FH_RSCSR_RXQ_POS != rxq->id,
1337 		     "frame on invalid queue - is on %d and indicates %d\n",
1338 		     rxq->id,
1339 		     (le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1340 			FH_RSCSR_RXQ_POS);
1341 
1342 		IWL_DEBUG_RX(trans,
1343 			     "Q %d: cmd at offset %d: %s (%.2x.%2x, seq 0x%x)\n",
1344 			     rxq->id, offset,
1345 			     iwl_get_cmd_string(trans,
1346 						WIDE_ID(pkt->hdr.group_id, pkt->hdr.cmd)),
1347 			     pkt->hdr.group_id, pkt->hdr.cmd,
1348 			     le16_to_cpu(pkt->hdr.sequence));
1349 
1350 		len = iwl_rx_packet_len(pkt);
1351 		len += sizeof(u32); /* account for status word */
1352 
1353 		offset += ALIGN(len, FH_RSCSR_FRAME_ALIGN);
1354 
1355 		/* check that what the device tells us made sense */
1356 		if (len < sizeof(*pkt) || offset > max_len)
1357 			break;
1358 
1359 		maybe_trace_iwlwifi_dev_rx(trans, pkt, len);
1360 
1361 		/* Reclaim a command buffer only if this packet is a response
1362 		 *   to a (driver-originated) command.
1363 		 * If the packet (e.g. Rx frame) originated from uCode,
1364 		 *   there is no command buffer to reclaim.
1365 		 * Ucode should set SEQ_RX_FRAME bit if ucode-originated,
1366 		 *   but apparently a few don't get set; catch them here. */
1367 		reclaim = !(pkt->hdr.sequence & SEQ_RX_FRAME);
1368 		if (reclaim && !pkt->hdr.group_id) {
1369 			int i;
1370 
1371 			for (i = 0; i < trans->conf.n_no_reclaim_cmds; i++) {
1372 				if (trans->conf.no_reclaim_cmds[i] ==
1373 							pkt->hdr.cmd) {
1374 					reclaim = false;
1375 					break;
1376 				}
1377 			}
1378 		}
1379 
1380 		if (rxq->id == IWL_DEFAULT_RX_QUEUE)
1381 			iwl_op_mode_rx(trans->op_mode, &rxq->napi,
1382 				       &rxcb);
1383 		else
1384 			iwl_op_mode_rx_rss(trans->op_mode, &rxq->napi,
1385 					   &rxcb, rxq->id);
1386 
1387 		/*
1388 		 * After here, we should always check rxcb._page_stolen,
1389 		 * if it is true then one of the handlers took the page.
1390 		 */
1391 
1392 		if (reclaim && txq) {
1393 			u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1394 			int index = SEQ_TO_INDEX(sequence);
1395 			int cmd_index = iwl_txq_get_cmd_index(txq, index);
1396 
1397 			kfree_sensitive(txq->entries[cmd_index].free_buf);
1398 			txq->entries[cmd_index].free_buf = NULL;
1399 
1400 			/* Invoke any callbacks, transfer the buffer to caller,
1401 			 * and fire off the (possibly) blocking
1402 			 * iwl_trans_send_cmd()
1403 			 * as we reclaim the driver command queue */
1404 			if (!rxcb._page_stolen)
1405 				iwl_pcie_hcmd_complete(trans, &rxcb);
1406 			else
1407 				IWL_WARN(trans, "Claim null rxb?\n");
1408 		}
1409 
1410 		page_stolen |= rxcb._page_stolen;
1411 		if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_AX210)
1412 			break;
1413 	}
1414 
1415 	/* page was stolen from us -- free our reference */
1416 	if (page_stolen) {
1417 		__free_pages(rxb->page, trans_pcie->rx_page_order);
1418 		rxb->page = NULL;
1419 	}
1420 
1421 	/* Reuse the page if possible. For notification packets and
1422 	 * SKBs that fail to Rx correctly, add them back into the
1423 	 * rx_free list for reuse later. */
1424 	if (rxb->page != NULL) {
1425 		rxb->page_dma =
1426 			dma_map_page(trans->dev, rxb->page, rxb->offset,
1427 				     trans_pcie->rx_buf_bytes,
1428 				     DMA_FROM_DEVICE);
1429 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
1430 			/*
1431 			 * free the page(s) as well to not break
1432 			 * the invariant that the items on the used
1433 			 * list have no page(s)
1434 			 */
1435 			__free_pages(rxb->page, trans_pcie->rx_page_order);
1436 			rxb->page = NULL;
1437 			iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1438 		} else {
1439 			list_add_tail(&rxb->list, &rxq->rx_free);
1440 			rxq->free_count++;
1441 		}
1442 	} else
1443 		iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1444 }
1445 
iwl_pcie_get_rxb(struct iwl_trans * trans,struct iwl_rxq * rxq,int i,bool * join)1446 static struct iwl_rx_mem_buffer *iwl_pcie_get_rxb(struct iwl_trans *trans,
1447 						  struct iwl_rxq *rxq, int i,
1448 						  bool *join)
1449 {
1450 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1451 	struct iwl_rx_mem_buffer *rxb;
1452 	u16 vid;
1453 
1454 	BUILD_BUG_ON(sizeof(struct iwl_rx_completion_desc) != 32);
1455 	BUILD_BUG_ON(sizeof(struct iwl_rx_completion_desc_bz) != 4);
1456 
1457 	if (!trans->mac_cfg->mq_rx_supported) {
1458 		rxb = rxq->queue[i];
1459 		rxq->queue[i] = NULL;
1460 		return rxb;
1461 	}
1462 
1463 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) {
1464 		struct iwl_rx_completion_desc_bz *cd = rxq->used_bd;
1465 
1466 		vid = le16_to_cpu(cd[i].rbid);
1467 		*join = cd[i].flags & IWL_RX_CD_FLAGS_FRAGMENTED;
1468 	} else if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
1469 		struct iwl_rx_completion_desc *cd = rxq->used_bd;
1470 
1471 		vid = le16_to_cpu(cd[i].rbid);
1472 		*join = cd[i].flags & IWL_RX_CD_FLAGS_FRAGMENTED;
1473 	} else {
1474 		__le32 *cd = rxq->used_bd;
1475 
1476 		vid = le32_to_cpu(cd[i]) & 0x0FFF; /* 12-bit VID */
1477 	}
1478 
1479 	if (!vid || vid > RX_POOL_SIZE(trans_pcie->num_rx_bufs))
1480 		goto out_err;
1481 
1482 	rxb = trans_pcie->global_table[vid - 1];
1483 	if (rxb->invalid)
1484 		goto out_err;
1485 
1486 	IWL_DEBUG_RX(trans, "Got virtual RB ID %u\n", (u32)rxb->vid);
1487 
1488 	rxb->invalid = true;
1489 
1490 	return rxb;
1491 
1492 out_err:
1493 	WARN(1, "Invalid rxb from HW %u\n", (u32)vid);
1494 	iwl_force_nmi(trans);
1495 	return NULL;
1496 }
1497 
1498 /*
1499  * iwl_pcie_rx_handle - Main entry function for receiving responses from fw
1500  */
iwl_pcie_rx_handle(struct iwl_trans * trans,int queue,int budget)1501 static int iwl_pcie_rx_handle(struct iwl_trans *trans, int queue, int budget)
1502 {
1503 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1504 	struct iwl_rxq *rxq;
1505 	u32 r, i, count = 0, handled = 0;
1506 	bool emergency = false;
1507 
1508 	if (WARN_ON_ONCE(!trans_pcie->rxq || !trans_pcie->rxq[queue].bd))
1509 		return budget;
1510 
1511 	rxq = &trans_pcie->rxq[queue];
1512 
1513 restart:
1514 	spin_lock(&rxq->lock);
1515 	/* uCode's read index (stored in shared DRAM) indicates the last Rx
1516 	 * buffer that the driver may process (last buffer filled by ucode). */
1517 	r = iwl_get_closed_rb_stts(trans, rxq);
1518 	i = rxq->read;
1519 
1520 	/* W/A 9000 device step A0 wrap-around bug */
1521 	r &= (rxq->queue_size - 1);
1522 
1523 	/* Rx interrupt, but nothing sent from uCode */
1524 	if (i == r)
1525 		IWL_DEBUG_RX(trans, "Q %d: HW = SW = %d\n", rxq->id, r);
1526 
1527 	while (i != r && ++handled < budget) {
1528 		struct iwl_rb_allocator *rba = &trans_pcie->rba;
1529 		struct iwl_rx_mem_buffer *rxb;
1530 		/* number of RBDs still waiting for page allocation */
1531 		u32 rb_pending_alloc =
1532 			atomic_read(&trans_pcie->rba.req_pending) *
1533 			RX_CLAIM_REQ_ALLOC;
1534 		bool join = false;
1535 
1536 		if (unlikely(rb_pending_alloc >= rxq->queue_size / 2 &&
1537 			     !emergency)) {
1538 			iwl_pcie_rx_move_to_allocator(rxq, rba);
1539 			emergency = true;
1540 			IWL_DEBUG_TPT(trans,
1541 				      "RX path is in emergency. Pending allocations %d\n",
1542 				      rb_pending_alloc);
1543 		}
1544 
1545 		IWL_DEBUG_RX(trans, "Q %d: HW = %d, SW = %d\n", rxq->id, r, i);
1546 
1547 		rxb = iwl_pcie_get_rxb(trans, rxq, i, &join);
1548 		if (!rxb)
1549 			goto out;
1550 
1551 		if (unlikely(join || rxq->next_rb_is_fragment)) {
1552 			rxq->next_rb_is_fragment = join;
1553 			/*
1554 			 * We can only get a multi-RB in the following cases:
1555 			 *  - firmware issue, sending a too big notification
1556 			 *  - sniffer mode with a large A-MSDU
1557 			 *  - large MTU frames (>2k)
1558 			 * since the multi-RB functionality is limited to newer
1559 			 * hardware that cannot put multiple entries into a
1560 			 * single RB.
1561 			 *
1562 			 * Right now, the higher layers aren't set up to deal
1563 			 * with that, so discard all of these.
1564 			 */
1565 			list_add_tail(&rxb->list, &rxq->rx_free);
1566 			rxq->free_count++;
1567 		} else {
1568 			iwl_pcie_rx_handle_rb(trans, rxq, rxb, emergency, i);
1569 		}
1570 
1571 		i = (i + 1) & (rxq->queue_size - 1);
1572 
1573 		/*
1574 		 * If we have RX_CLAIM_REQ_ALLOC released rx buffers -
1575 		 * try to claim the pre-allocated buffers from the allocator.
1576 		 * If not ready - will try to reclaim next time.
1577 		 * There is no need to reschedule work - allocator exits only
1578 		 * on success
1579 		 */
1580 		if (rxq->used_count >= RX_CLAIM_REQ_ALLOC)
1581 			iwl_pcie_rx_allocator_get(trans, rxq);
1582 
1583 		if (rxq->used_count % RX_CLAIM_REQ_ALLOC == 0 && !emergency) {
1584 			/* Add the remaining empty RBDs for allocator use */
1585 			iwl_pcie_rx_move_to_allocator(rxq, rba);
1586 		} else if (emergency) {
1587 			count++;
1588 			if (count == 8) {
1589 				count = 0;
1590 				if (rb_pending_alloc < rxq->queue_size / 3) {
1591 					IWL_DEBUG_TPT(trans,
1592 						      "RX path exited emergency. Pending allocations %d\n",
1593 						      rb_pending_alloc);
1594 					emergency = false;
1595 				}
1596 
1597 				rxq->read = i;
1598 				spin_unlock(&rxq->lock);
1599 				iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1600 				iwl_pcie_rxq_restock(trans, rxq);
1601 				goto restart;
1602 			}
1603 		}
1604 	}
1605 out:
1606 	/* Backtrack one entry */
1607 	rxq->read = i;
1608 	spin_unlock(&rxq->lock);
1609 
1610 	/*
1611 	 * handle a case where in emergency there are some unallocated RBDs.
1612 	 * those RBDs are in the used list, but are not tracked by the queue's
1613 	 * used_count which counts allocator owned RBDs.
1614 	 * unallocated emergency RBDs must be allocated on exit, otherwise
1615 	 * when called again the function may not be in emergency mode and
1616 	 * they will be handed to the allocator with no tracking in the RBD
1617 	 * allocator counters, which will lead to them never being claimed back
1618 	 * by the queue.
1619 	 * by allocating them here, they are now in the queue free list, and
1620 	 * will be restocked by the next call of iwl_pcie_rxq_restock.
1621 	 */
1622 	if (unlikely(emergency && count))
1623 		iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1624 
1625 	iwl_pcie_rxq_restock(trans, rxq);
1626 
1627 	return handled;
1628 }
1629 
iwl_pcie_get_trans_pcie(struct msix_entry * entry)1630 static struct iwl_trans_pcie *iwl_pcie_get_trans_pcie(struct msix_entry *entry)
1631 {
1632 	u8 queue = entry->entry;
1633 	struct msix_entry *entries = entry - queue;
1634 
1635 	return container_of(entries, struct iwl_trans_pcie, msix_entries[0]);
1636 }
1637 
1638 /*
1639  * iwl_pcie_rx_msix_handle - Main entry function for receiving responses from fw
1640  * This interrupt handler should be used with RSS queue only.
1641  */
iwl_pcie_irq_rx_msix_handler(int irq,void * dev_id)1642 irqreturn_t iwl_pcie_irq_rx_msix_handler(int irq, void *dev_id)
1643 {
1644 	struct msix_entry *entry = dev_id;
1645 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
1646 	struct iwl_trans *trans = trans_pcie->trans;
1647 	struct iwl_rxq *rxq;
1648 
1649 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, false, 0, 0);
1650 
1651 	if (WARN_ON(entry->entry >= trans->info.num_rxqs))
1652 		return IRQ_NONE;
1653 
1654 	if (!trans_pcie->rxq) {
1655 		if (net_ratelimit())
1656 			IWL_ERR(trans,
1657 				"[%d] Got MSI-X interrupt before we have Rx queues\n",
1658 				entry->entry);
1659 		return IRQ_NONE;
1660 	}
1661 
1662 	rxq = &trans_pcie->rxq[entry->entry];
1663 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1664 	IWL_DEBUG_ISR(trans, "[%d] Got interrupt\n", entry->entry);
1665 
1666 	local_bh_disable();
1667 	if (!napi_schedule(&rxq->napi))
1668 		iwl_pcie_clear_irq(trans, entry->entry);
1669 	local_bh_enable();
1670 
1671 	lock_map_release(&trans->sync_cmd_lockdep_map);
1672 
1673 	return IRQ_HANDLED;
1674 }
1675 
1676 /*
1677  * iwl_pcie_irq_handle_error - called for HW or SW error interrupt from card
1678  */
iwl_pcie_irq_handle_error(struct iwl_trans * trans)1679 static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
1680 {
1681 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1682 	int i;
1683 
1684 	/* W/A for WiFi/WiMAX coex and WiMAX own the RF */
1685 	if (trans->cfg->internal_wimax_coex &&
1686 	    !trans->mac_cfg->base->apmg_not_supported &&
1687 	    (!(iwl_read_prph(trans, APMG_CLK_CTRL_REG) &
1688 			     APMS_CLK_VAL_MRB_FUNC_MODE) ||
1689 	     (iwl_read_prph(trans, APMG_PS_CTRL_REG) &
1690 			    APMG_PS_CTRL_VAL_RESET_REQ))) {
1691 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1692 		iwl_op_mode_wimax_active(trans->op_mode);
1693 		wake_up(&trans_pcie->wait_command_queue);
1694 		return;
1695 	}
1696 
1697 	for (i = 0; i < trans->mac_cfg->base->num_of_queues; i++) {
1698 		if (!trans_pcie->txqs.txq[i])
1699 			continue;
1700 		timer_delete(&trans_pcie->txqs.txq[i]->stuck_timer);
1701 	}
1702 
1703 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_SC) {
1704 		u32 val = iwl_read32(trans, CSR_IPC_STATE);
1705 
1706 		if (val & CSR_IPC_STATE_TOP_RESET_REQ) {
1707 			IWL_ERR(trans, "FW requested TOP reset for FSEQ\n");
1708 			trans->do_top_reset = 1;
1709 		}
1710 	}
1711 
1712 	/* The STATUS_FW_ERROR bit is set in this function. This must happen
1713 	 * before we wake up the command caller, to ensure a proper cleanup. */
1714 	iwl_trans_fw_error(trans, IWL_ERR_TYPE_IRQ);
1715 
1716 	clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1717 	wake_up(&trans_pcie->wait_command_queue);
1718 }
1719 
iwl_pcie_int_cause_non_ict(struct iwl_trans * trans)1720 static u32 iwl_pcie_int_cause_non_ict(struct iwl_trans *trans)
1721 {
1722 	u32 inta;
1723 
1724 	lockdep_assert_held(&IWL_TRANS_GET_PCIE_TRANS(trans)->irq_lock);
1725 
1726 	trace_iwlwifi_dev_irq(trans->dev);
1727 
1728 	/* Discover which interrupts are active/pending */
1729 	inta = iwl_read32(trans, CSR_INT);
1730 
1731 	/* the thread will service interrupts and re-enable them */
1732 	return inta;
1733 }
1734 
1735 /* a device (PCI-E) page is 4096 bytes long */
1736 #define ICT_SHIFT	12
1737 #define ICT_SIZE	(1 << ICT_SHIFT)
1738 #define ICT_COUNT	(ICT_SIZE / sizeof(u32))
1739 
1740 /* interrupt handler using ict table, with this interrupt driver will
1741  * stop using INTA register to get device's interrupt, reading this register
1742  * is expensive, device will write interrupts in ICT dram table, increment
1743  * index then will fire interrupt to driver, driver will OR all ICT table
1744  * entries from current index up to table entry with 0 value. the result is
1745  * the interrupt we need to service, driver will set the entries back to 0 and
1746  * set index.
1747  */
iwl_pcie_int_cause_ict(struct iwl_trans * trans)1748 static u32 iwl_pcie_int_cause_ict(struct iwl_trans *trans)
1749 {
1750 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1751 	u32 inta;
1752 	u32 val = 0;
1753 	u32 read;
1754 
1755 	trace_iwlwifi_dev_irq(trans->dev);
1756 
1757 	/* Ignore interrupt if there's nothing in NIC to service.
1758 	 * This may be due to IRQ shared with another device,
1759 	 * or due to sporadic interrupts thrown from our NIC. */
1760 	read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1761 	trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index, read);
1762 	if (!read)
1763 		return 0;
1764 
1765 	/*
1766 	 * Collect all entries up to the first 0, starting from ict_index;
1767 	 * note we already read at ict_index.
1768 	 */
1769 	do {
1770 		val |= read;
1771 		IWL_DEBUG_ISR(trans, "ICT index %d value 0x%08X\n",
1772 				trans_pcie->ict_index, read);
1773 		trans_pcie->ict_tbl[trans_pcie->ict_index] = 0;
1774 		trans_pcie->ict_index =
1775 			((trans_pcie->ict_index + 1) & (ICT_COUNT - 1));
1776 
1777 		read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1778 		trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index,
1779 					   read);
1780 	} while (read);
1781 
1782 	/* We should not get this value, just ignore it. */
1783 	if (val == 0xffffffff)
1784 		val = 0;
1785 
1786 	/*
1787 	 * this is a w/a for a h/w bug. the h/w bug may cause the Rx bit
1788 	 * (bit 15 before shifting it to 31) to clear when using interrupt
1789 	 * coalescing. fortunately, bits 18 and 19 stay set when this happens
1790 	 * so we use them to decide on the real state of the Rx bit.
1791 	 * In order words, bit 15 is set if bit 18 or bit 19 are set.
1792 	 */
1793 	if (val & 0xC0000)
1794 		val |= 0x8000;
1795 
1796 	inta = (0xff & val) | ((0xff00 & val) << 16);
1797 	return inta;
1798 }
1799 
iwl_pcie_handle_rfkill_irq(struct iwl_trans * trans,bool from_irq)1800 void iwl_pcie_handle_rfkill_irq(struct iwl_trans *trans, bool from_irq)
1801 {
1802 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1803 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1804 	bool hw_rfkill, prev, report;
1805 
1806 	mutex_lock(&trans_pcie->mutex);
1807 	prev = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1808 	hw_rfkill = iwl_is_rfkill_set(trans);
1809 	if (hw_rfkill) {
1810 		set_bit(STATUS_RFKILL_OPMODE, &trans->status);
1811 		set_bit(STATUS_RFKILL_HW, &trans->status);
1812 	}
1813 	if (trans_pcie->opmode_down)
1814 		report = hw_rfkill;
1815 	else
1816 		report = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1817 
1818 	IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
1819 		 hw_rfkill ? "disable radio" : "enable radio");
1820 
1821 	isr_stats->rfkill++;
1822 
1823 	if (prev != report)
1824 		iwl_trans_pcie_rf_kill(trans, report, from_irq);
1825 	mutex_unlock(&trans_pcie->mutex);
1826 
1827 	if (hw_rfkill) {
1828 		if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
1829 				       &trans->status))
1830 			IWL_DEBUG_RF_KILL(trans,
1831 					  "Rfkill while SYNC HCMD in flight\n");
1832 		wake_up(&trans_pcie->wait_command_queue);
1833 	} else {
1834 		clear_bit(STATUS_RFKILL_HW, &trans->status);
1835 		if (trans_pcie->opmode_down)
1836 			clear_bit(STATUS_RFKILL_OPMODE, &trans->status);
1837 	}
1838 }
1839 
iwl_trans_pcie_handle_reset_interrupt(struct iwl_trans * trans)1840 static void iwl_trans_pcie_handle_reset_interrupt(struct iwl_trans *trans)
1841 {
1842 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1843 	u32 state;
1844 
1845 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_SC) {
1846 		u32 val = iwl_read32(trans, CSR_IPC_STATE);
1847 
1848 		state = u32_get_bits(val, CSR_IPC_STATE_RESET);
1849 		IWL_DEBUG_ISR(trans, "IPC state = 0x%x/%d\n", val, state);
1850 	} else {
1851 		state = CSR_IPC_STATE_RESET_SW_READY;
1852 	}
1853 
1854 	switch (state) {
1855 	case CSR_IPC_STATE_RESET_SW_READY:
1856 		if (trans_pcie->fw_reset_state == FW_RESET_REQUESTED) {
1857 			IWL_DEBUG_ISR(trans, "Reset flow completed\n");
1858 			trans_pcie->fw_reset_state = FW_RESET_OK;
1859 			wake_up(&trans_pcie->fw_reset_waitq);
1860 			break;
1861 		}
1862 		fallthrough;
1863 	case CSR_IPC_STATE_RESET_TOP_READY:
1864 		if (trans_pcie->fw_reset_state == FW_RESET_TOP_REQUESTED) {
1865 			IWL_DEBUG_ISR(trans, "TOP Reset continues\n");
1866 			trans_pcie->fw_reset_state = FW_RESET_OK;
1867 			wake_up(&trans_pcie->fw_reset_waitq);
1868 			break;
1869 		}
1870 		fallthrough;
1871 	case CSR_IPC_STATE_RESET_NONE:
1872 		IWL_FW_CHECK_FAILED(trans,
1873 				    "Invalid reset interrupt (state=%d)!\n",
1874 				    state);
1875 		break;
1876 	case CSR_IPC_STATE_RESET_TOP_FOLLOWER:
1877 		if (trans_pcie->fw_reset_state == FW_RESET_REQUESTED) {
1878 			/* if we were in reset, wake that up */
1879 			IWL_INFO(trans,
1880 				 "TOP reset from BT while doing reset\n");
1881 			trans_pcie->fw_reset_state = FW_RESET_OK;
1882 			wake_up(&trans_pcie->fw_reset_waitq);
1883 		} else {
1884 			IWL_INFO(trans, "TOP reset from BT\n");
1885 			trans->state = IWL_TRANS_NO_FW;
1886 			iwl_trans_schedule_reset(trans,
1887 						 IWL_ERR_TYPE_TOP_RESET_BY_BT);
1888 		}
1889 		break;
1890 	}
1891 }
1892 
iwl_pcie_irq_handler(int irq,void * dev_id)1893 irqreturn_t iwl_pcie_irq_handler(int irq, void *dev_id)
1894 {
1895 	struct iwl_trans *trans = dev_id;
1896 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1897 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1898 	u32 inta = 0;
1899 	u32 handled = 0;
1900 	bool polling = false;
1901 
1902 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1903 
1904 	spin_lock_bh(&trans_pcie->irq_lock);
1905 
1906 	/* dram interrupt table not set yet,
1907 	 * use legacy interrupt.
1908 	 */
1909 	if (likely(trans_pcie->use_ict))
1910 		inta = iwl_pcie_int_cause_ict(trans);
1911 	else
1912 		inta = iwl_pcie_int_cause_non_ict(trans);
1913 
1914 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1915 		IWL_DEBUG_ISR(trans,
1916 			      "ISR inta 0x%08x, enabled 0x%08x(sw), enabled(hw) 0x%08x, fh 0x%08x\n",
1917 			      inta, trans_pcie->inta_mask,
1918 			      iwl_read32(trans, CSR_INT_MASK),
1919 			      iwl_read32(trans, CSR_FH_INT_STATUS));
1920 		if (inta & (~trans_pcie->inta_mask))
1921 			IWL_DEBUG_ISR(trans,
1922 				      "We got a masked interrupt (0x%08x)\n",
1923 				      inta & (~trans_pcie->inta_mask));
1924 	}
1925 
1926 	inta &= trans_pcie->inta_mask;
1927 
1928 	/*
1929 	 * Ignore interrupt if there's nothing in NIC to service.
1930 	 * This may be due to IRQ shared with another device,
1931 	 * or due to sporadic interrupts thrown from our NIC.
1932 	 */
1933 	if (unlikely(!inta)) {
1934 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
1935 		/*
1936 		 * Re-enable interrupts here since we don't
1937 		 * have anything to service
1938 		 */
1939 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1940 			_iwl_enable_interrupts(trans);
1941 		spin_unlock_bh(&trans_pcie->irq_lock);
1942 		lock_map_release(&trans->sync_cmd_lockdep_map);
1943 		return IRQ_NONE;
1944 	}
1945 
1946 	if (unlikely(inta == 0xFFFFFFFF || iwl_trans_is_hw_error_value(inta))) {
1947 		/*
1948 		 * Hardware disappeared. It might have
1949 		 * already raised an interrupt.
1950 		 */
1951 		IWL_WARN(trans, "HARDWARE GONE?? INTA == 0x%08x\n", inta);
1952 		spin_unlock_bh(&trans_pcie->irq_lock);
1953 		goto out;
1954 	}
1955 
1956 	/* Ack/clear/reset pending uCode interrupts.
1957 	 * Note:  Some bits in CSR_INT are "OR" of bits in CSR_FH_INT_STATUS,
1958 	 */
1959 	/* There is a hardware bug in the interrupt mask function that some
1960 	 * interrupts (i.e. CSR_INT_BIT_SCD) can still be generated even if
1961 	 * they are disabled in the CSR_INT_MASK register. Furthermore the
1962 	 * ICT interrupt handling mechanism has another bug that might cause
1963 	 * these unmasked interrupts fail to be detected. We workaround the
1964 	 * hardware bugs here by ACKing all the possible interrupts so that
1965 	 * interrupt coalescing can still be achieved.
1966 	 */
1967 	iwl_write32(trans, CSR_INT, inta | ~trans_pcie->inta_mask);
1968 
1969 	if (iwl_have_debug_level(IWL_DL_ISR))
1970 		IWL_DEBUG_ISR(trans, "inta 0x%08x, enabled 0x%08x\n",
1971 			      inta, iwl_read32(trans, CSR_INT_MASK));
1972 
1973 	spin_unlock_bh(&trans_pcie->irq_lock);
1974 
1975 	/* Now service all interrupt bits discovered above. */
1976 	if (inta & CSR_INT_BIT_HW_ERR) {
1977 		IWL_ERR(trans, "Hardware error detected.  Restarting.\n");
1978 
1979 		/* Tell the device to stop sending interrupts */
1980 		iwl_disable_interrupts(trans);
1981 
1982 		isr_stats->hw++;
1983 		iwl_pcie_irq_handle_error(trans);
1984 
1985 		handled |= CSR_INT_BIT_HW_ERR;
1986 
1987 		goto out;
1988 	}
1989 
1990 	/* NIC fires this, but we don't use it, redundant with WAKEUP */
1991 	if (inta & CSR_INT_BIT_SCD) {
1992 		IWL_DEBUG_ISR(trans,
1993 			      "Scheduler finished to transmit the frame/frames.\n");
1994 		isr_stats->sch++;
1995 	}
1996 
1997 	/* Alive notification via Rx interrupt will do the real work */
1998 	if (inta & CSR_INT_BIT_ALIVE) {
1999 		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
2000 		isr_stats->alive++;
2001 		if (trans->mac_cfg->gen2) {
2002 			/*
2003 			 * We can restock, since firmware configured
2004 			 * the RFH
2005 			 */
2006 			iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
2007 		}
2008 
2009 		handled |= CSR_INT_BIT_ALIVE;
2010 	}
2011 
2012 	if (inta & CSR_INT_BIT_RESET_DONE) {
2013 		iwl_trans_pcie_handle_reset_interrupt(trans);
2014 		handled |= CSR_INT_BIT_RESET_DONE;
2015 	}
2016 
2017 	/* Safely ignore these bits for debug checks below */
2018 	inta &= ~(CSR_INT_BIT_SCD | CSR_INT_BIT_ALIVE);
2019 
2020 	/* HW RF KILL switch toggled */
2021 	if (inta & CSR_INT_BIT_RF_KILL) {
2022 		iwl_pcie_handle_rfkill_irq(trans, true);
2023 		handled |= CSR_INT_BIT_RF_KILL;
2024 	}
2025 
2026 	/* Chip got too hot and stopped itself */
2027 	if (inta & CSR_INT_BIT_CT_KILL) {
2028 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
2029 		isr_stats->ctkill++;
2030 		handled |= CSR_INT_BIT_CT_KILL;
2031 	}
2032 
2033 	/* Error detected by uCode */
2034 	if (inta & CSR_INT_BIT_SW_ERR) {
2035 		IWL_ERR(trans, "Microcode SW error detected. "
2036 			" Restarting 0x%X.\n", inta);
2037 		isr_stats->sw++;
2038 		if (trans_pcie->fw_reset_state == FW_RESET_REQUESTED) {
2039 			trans_pcie->fw_reset_state = FW_RESET_ERROR;
2040 			wake_up(&trans_pcie->fw_reset_waitq);
2041 		} else {
2042 			iwl_pcie_irq_handle_error(trans);
2043 		}
2044 		handled |= CSR_INT_BIT_SW_ERR;
2045 	}
2046 
2047 	/* uCode wakes up after power-down sleep */
2048 	if (inta & CSR_INT_BIT_WAKEUP) {
2049 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
2050 		iwl_pcie_rxq_check_wrptr(trans);
2051 		iwl_pcie_txq_check_wrptrs(trans);
2052 
2053 		isr_stats->wakeup++;
2054 
2055 		handled |= CSR_INT_BIT_WAKEUP;
2056 	}
2057 
2058 	/* All uCode command responses, including Tx command responses,
2059 	 * Rx "responses" (frame-received notification), and other
2060 	 * notifications from uCode come through here*/
2061 	if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX |
2062 		    CSR_INT_BIT_RX_PERIODIC)) {
2063 		IWL_DEBUG_ISR(trans, "Rx interrupt\n");
2064 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX)) {
2065 			handled |= (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX);
2066 			iwl_write32(trans, CSR_FH_INT_STATUS,
2067 					CSR_FH_INT_RX_MASK);
2068 		}
2069 		if (inta & CSR_INT_BIT_RX_PERIODIC) {
2070 			handled |= CSR_INT_BIT_RX_PERIODIC;
2071 			iwl_write32(trans,
2072 				CSR_INT, CSR_INT_BIT_RX_PERIODIC);
2073 		}
2074 		/* Sending RX interrupt require many steps to be done in the
2075 		 * device:
2076 		 * 1- write interrupt to current index in ICT table.
2077 		 * 2- dma RX frame.
2078 		 * 3- update RX shared data to indicate last write index.
2079 		 * 4- send interrupt.
2080 		 * This could lead to RX race, driver could receive RX interrupt
2081 		 * but the shared data changes does not reflect this;
2082 		 * periodic interrupt will detect any dangling Rx activity.
2083 		 */
2084 
2085 		/* Disable periodic interrupt; we use it as just a one-shot. */
2086 		iwl_write8(trans, CSR_INT_PERIODIC_REG,
2087 			    CSR_INT_PERIODIC_DIS);
2088 
2089 		/*
2090 		 * Enable periodic interrupt in 8 msec only if we received
2091 		 * real RX interrupt (instead of just periodic int), to catch
2092 		 * any dangling Rx interrupt.  If it was just the periodic
2093 		 * interrupt, there was no dangling Rx activity, and no need
2094 		 * to extend the periodic interrupt; one-shot is enough.
2095 		 */
2096 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX))
2097 			iwl_write8(trans, CSR_INT_PERIODIC_REG,
2098 				   CSR_INT_PERIODIC_ENA);
2099 
2100 		isr_stats->rx++;
2101 
2102 		local_bh_disable();
2103 		if (napi_schedule_prep(&trans_pcie->rxq[0].napi)) {
2104 			polling = true;
2105 			__napi_schedule(&trans_pcie->rxq[0].napi);
2106 		}
2107 		local_bh_enable();
2108 	}
2109 
2110 	/* This "Tx" DMA channel is used only for loading uCode */
2111 	if (inta & CSR_INT_BIT_FH_TX) {
2112 		iwl_write32(trans, CSR_FH_INT_STATUS, CSR_FH_INT_TX_MASK);
2113 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
2114 		isr_stats->tx++;
2115 		handled |= CSR_INT_BIT_FH_TX;
2116 		/* Wake up uCode load routine, now that load is complete */
2117 		trans_pcie->ucode_write_complete = true;
2118 		wake_up(&trans_pcie->ucode_write_waitq);
2119 		/* Wake up IMR write routine, now that write to SRAM is complete */
2120 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2121 			trans_pcie->imr_status = IMR_D2S_COMPLETED;
2122 			wake_up(&trans_pcie->ucode_write_waitq);
2123 		}
2124 	}
2125 
2126 	if (inta & ~handled) {
2127 		IWL_ERR(trans, "Unhandled INTA bits 0x%08x\n", inta & ~handled);
2128 		isr_stats->unhandled++;
2129 	}
2130 
2131 	if (inta & ~(trans_pcie->inta_mask)) {
2132 		IWL_WARN(trans, "Disabled INTA bits 0x%08x were pending\n",
2133 			 inta & ~trans_pcie->inta_mask);
2134 	}
2135 
2136 	if (!polling) {
2137 		spin_lock_bh(&trans_pcie->irq_lock);
2138 		/* only Re-enable all interrupt if disabled by irq */
2139 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
2140 			_iwl_enable_interrupts(trans);
2141 		/* we are loading the firmware, enable FH_TX interrupt only */
2142 		else if (handled & CSR_INT_BIT_FH_TX)
2143 			iwl_enable_fw_load_int(trans);
2144 		/* Re-enable RF_KILL if it occurred */
2145 		else if (handled & CSR_INT_BIT_RF_KILL)
2146 			iwl_enable_rfkill_int(trans);
2147 		/* Re-enable the ALIVE / Rx interrupt if it occurred */
2148 		else if (handled & (CSR_INT_BIT_ALIVE | CSR_INT_BIT_FH_RX))
2149 			iwl_enable_fw_load_int_ctx_info(trans, false);
2150 		spin_unlock_bh(&trans_pcie->irq_lock);
2151 	}
2152 
2153 out:
2154 	lock_map_release(&trans->sync_cmd_lockdep_map);
2155 	return IRQ_HANDLED;
2156 }
2157 
2158 /******************************************************************************
2159  *
2160  * ICT functions
2161  *
2162  ******************************************************************************/
2163 
2164 /* Free dram table */
iwl_pcie_free_ict(struct iwl_trans * trans)2165 void iwl_pcie_free_ict(struct iwl_trans *trans)
2166 {
2167 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2168 
2169 	if (trans_pcie->ict_tbl) {
2170 		dma_free_coherent(trans->dev, ICT_SIZE,
2171 				  trans_pcie->ict_tbl,
2172 				  trans_pcie->ict_tbl_dma);
2173 		trans_pcie->ict_tbl = NULL;
2174 		trans_pcie->ict_tbl_dma = 0;
2175 	}
2176 }
2177 
2178 /*
2179  * allocate dram shared table, it is an aligned memory
2180  * block of ICT_SIZE.
2181  * also reset all data related to ICT table interrupt.
2182  */
iwl_pcie_alloc_ict(struct iwl_trans * trans)2183 int iwl_pcie_alloc_ict(struct iwl_trans *trans)
2184 {
2185 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2186 
2187 	trans_pcie->ict_tbl =
2188 		dma_alloc_coherent(trans->dev, ICT_SIZE,
2189 				   &trans_pcie->ict_tbl_dma, GFP_KERNEL);
2190 	if (!trans_pcie->ict_tbl)
2191 		return -ENOMEM;
2192 
2193 	/* just an API sanity check ... it is guaranteed to be aligned */
2194 	if (WARN_ON(trans_pcie->ict_tbl_dma & (ICT_SIZE - 1))) {
2195 		iwl_pcie_free_ict(trans);
2196 		return -EINVAL;
2197 	}
2198 
2199 	return 0;
2200 }
2201 
2202 /* Device is going up inform it about using ICT interrupt table,
2203  * also we need to tell the driver to start using ICT interrupt.
2204  */
iwl_pcie_reset_ict(struct iwl_trans * trans)2205 void iwl_pcie_reset_ict(struct iwl_trans *trans)
2206 {
2207 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2208 	u32 val;
2209 
2210 	if (!trans_pcie->ict_tbl)
2211 		return;
2212 
2213 	spin_lock_bh(&trans_pcie->irq_lock);
2214 	_iwl_disable_interrupts(trans);
2215 
2216 	memset(trans_pcie->ict_tbl, 0, ICT_SIZE);
2217 
2218 	val = trans_pcie->ict_tbl_dma >> ICT_SHIFT;
2219 
2220 	val |= CSR_DRAM_INT_TBL_ENABLE |
2221 	       CSR_DRAM_INIT_TBL_WRAP_CHECK |
2222 	       CSR_DRAM_INIT_TBL_WRITE_POINTER;
2223 
2224 	IWL_DEBUG_ISR(trans, "CSR_DRAM_INT_TBL_REG =0x%x\n", val);
2225 
2226 	iwl_write32(trans, CSR_DRAM_INT_TBL_REG, val);
2227 	trans_pcie->use_ict = true;
2228 	trans_pcie->ict_index = 0;
2229 	iwl_write32(trans, CSR_INT, trans_pcie->inta_mask);
2230 	_iwl_enable_interrupts(trans);
2231 	spin_unlock_bh(&trans_pcie->irq_lock);
2232 }
2233 
2234 /* Device is going down disable ict interrupt usage */
iwl_pcie_disable_ict(struct iwl_trans * trans)2235 void iwl_pcie_disable_ict(struct iwl_trans *trans)
2236 {
2237 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2238 
2239 	spin_lock_bh(&trans_pcie->irq_lock);
2240 	trans_pcie->use_ict = false;
2241 	spin_unlock_bh(&trans_pcie->irq_lock);
2242 }
2243 
iwl_pcie_isr(int irq,void * data)2244 irqreturn_t iwl_pcie_isr(int irq, void *data)
2245 {
2246 	struct iwl_trans *trans = data;
2247 
2248 	if (!trans)
2249 		return IRQ_NONE;
2250 
2251 	/* Disable (but don't clear!) interrupts here to avoid
2252 	 * back-to-back ISRs and sporadic interrupts from our NIC.
2253 	 * If we have something to service, the tasklet will re-enable ints.
2254 	 * If we *don't* have something, we'll re-enable before leaving here.
2255 	 */
2256 	iwl_write32(trans, CSR_INT_MASK, 0x00000000);
2257 
2258 	return IRQ_WAKE_THREAD;
2259 }
2260 
iwl_pcie_msix_isr(int irq,void * data)2261 irqreturn_t iwl_pcie_msix_isr(int irq, void *data)
2262 {
2263 	return IRQ_WAKE_THREAD;
2264 }
2265 
iwl_pcie_irq_msix_handler(int irq,void * dev_id)2266 irqreturn_t iwl_pcie_irq_msix_handler(int irq, void *dev_id)
2267 {
2268 	struct msix_entry *entry = dev_id;
2269 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
2270 	struct iwl_trans *trans = trans_pcie->trans;
2271 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
2272 	u32 inta_fh_msk = ~MSIX_FH_INT_CAUSES_DATA_QUEUE;
2273 	u32 inta_fh, inta_hw;
2274 	bool polling = false;
2275 	bool sw_err;
2276 
2277 	if (trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_NON_RX)
2278 		inta_fh_msk |= MSIX_FH_INT_CAUSES_Q0;
2279 
2280 	if (trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS)
2281 		inta_fh_msk |= MSIX_FH_INT_CAUSES_Q1;
2282 
2283 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
2284 
2285 	spin_lock_bh(&trans_pcie->irq_lock);
2286 	inta_fh = iwl_read32(trans, CSR_MSIX_FH_INT_CAUSES_AD);
2287 	inta_hw = iwl_read32(trans, CSR_MSIX_HW_INT_CAUSES_AD);
2288 	/*
2289 	 * Clear causes registers to avoid being handling the same cause.
2290 	 */
2291 	iwl_write32(trans, CSR_MSIX_FH_INT_CAUSES_AD, inta_fh & inta_fh_msk);
2292 	iwl_write32(trans, CSR_MSIX_HW_INT_CAUSES_AD, inta_hw);
2293 	spin_unlock_bh(&trans_pcie->irq_lock);
2294 
2295 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, true, inta_fh, inta_hw);
2296 
2297 	if (unlikely(!(inta_fh | inta_hw))) {
2298 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
2299 		lock_map_release(&trans->sync_cmd_lockdep_map);
2300 		return IRQ_NONE;
2301 	}
2302 
2303 	if (iwl_have_debug_level(IWL_DL_ISR)) {
2304 		IWL_DEBUG_ISR(trans,
2305 			      "ISR[%d] inta_fh 0x%08x, enabled (sw) 0x%08x (hw) 0x%08x\n",
2306 			      entry->entry, inta_fh, trans_pcie->fh_mask,
2307 			      iwl_read32(trans, CSR_MSIX_FH_INT_MASK_AD));
2308 		if (inta_fh & ~trans_pcie->fh_mask)
2309 			IWL_DEBUG_ISR(trans,
2310 				      "We got a masked interrupt (0x%08x)\n",
2311 				      inta_fh & ~trans_pcie->fh_mask);
2312 	}
2313 
2314 	inta_fh &= trans_pcie->fh_mask;
2315 
2316 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_NON_RX) &&
2317 	    inta_fh & MSIX_FH_INT_CAUSES_Q0) {
2318 		local_bh_disable();
2319 		if (napi_schedule_prep(&trans_pcie->rxq[0].napi)) {
2320 			polling = true;
2321 			__napi_schedule(&trans_pcie->rxq[0].napi);
2322 		}
2323 		local_bh_enable();
2324 	}
2325 
2326 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS) &&
2327 	    inta_fh & MSIX_FH_INT_CAUSES_Q1) {
2328 		local_bh_disable();
2329 		if (napi_schedule_prep(&trans_pcie->rxq[1].napi)) {
2330 			polling = true;
2331 			__napi_schedule(&trans_pcie->rxq[1].napi);
2332 		}
2333 		local_bh_enable();
2334 	}
2335 
2336 	/* This "Tx" DMA channel is used only for loading uCode */
2337 	if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM &&
2338 	    trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2339 		IWL_DEBUG_ISR(trans, "IMR Complete interrupt\n");
2340 		isr_stats->tx++;
2341 
2342 		/* Wake up IMR routine once write to SRAM is complete */
2343 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2344 			trans_pcie->imr_status = IMR_D2S_COMPLETED;
2345 			wake_up(&trans_pcie->ucode_write_waitq);
2346 		}
2347 	} else if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM) {
2348 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
2349 		isr_stats->tx++;
2350 		/*
2351 		 * Wake up uCode load routine,
2352 		 * now that load is complete
2353 		 */
2354 		trans_pcie->ucode_write_complete = true;
2355 		wake_up(&trans_pcie->ucode_write_waitq);
2356 
2357 		/* Wake up IMR routine once write to SRAM is complete */
2358 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2359 			trans_pcie->imr_status = IMR_D2S_COMPLETED;
2360 			wake_up(&trans_pcie->ucode_write_waitq);
2361 		}
2362 	}
2363 
2364 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_BZ)
2365 		sw_err = inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR_BZ;
2366 	else
2367 		sw_err = inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR;
2368 
2369 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_TOP_FATAL_ERR) {
2370 		IWL_ERR(trans, "TOP Fatal error detected, inta_hw=0x%x.\n",
2371 			inta_hw);
2372 		if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) {
2373 			trans->request_top_reset = 1;
2374 			iwl_op_mode_nic_error(trans->op_mode,
2375 					      IWL_ERR_TYPE_TOP_FATAL_ERROR);
2376 			iwl_trans_schedule_reset(trans,
2377 						 IWL_ERR_TYPE_TOP_FATAL_ERROR);
2378 		}
2379 	}
2380 
2381 	/* Error detected by uCode */
2382 	if ((inta_fh & MSIX_FH_INT_CAUSES_FH_ERR) || sw_err) {
2383 		IWL_ERR(trans,
2384 			"Microcode SW error detected. Restarting 0x%X.\n",
2385 			inta_fh);
2386 		isr_stats->sw++;
2387 		/* during FW reset flow report errors from there */
2388 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2389 			trans_pcie->imr_status = IMR_D2S_ERROR;
2390 			wake_up(&trans_pcie->imr_waitq);
2391 		} else if (trans_pcie->fw_reset_state == FW_RESET_REQUESTED) {
2392 			trans_pcie->fw_reset_state = FW_RESET_ERROR;
2393 			wake_up(&trans_pcie->fw_reset_waitq);
2394 		} else {
2395 			iwl_pcie_irq_handle_error(trans);
2396 		}
2397 
2398 		if (trans_pcie->sx_state == IWL_SX_WAITING) {
2399 			trans_pcie->sx_state = IWL_SX_ERROR;
2400 			wake_up(&trans_pcie->sx_waitq);
2401 		}
2402 	}
2403 
2404 	/* After checking FH register check HW register */
2405 	if (iwl_have_debug_level(IWL_DL_ISR)) {
2406 		IWL_DEBUG_ISR(trans,
2407 			      "ISR[%d] inta_hw 0x%08x, enabled (sw) 0x%08x (hw) 0x%08x\n",
2408 			      entry->entry, inta_hw, trans_pcie->hw_mask,
2409 			      iwl_read32(trans, CSR_MSIX_HW_INT_MASK_AD));
2410 		if (inta_hw & ~trans_pcie->hw_mask)
2411 			IWL_DEBUG_ISR(trans,
2412 				      "We got a masked interrupt 0x%08x\n",
2413 				      inta_hw & ~trans_pcie->hw_mask);
2414 	}
2415 
2416 	inta_hw &= trans_pcie->hw_mask;
2417 
2418 	/* Alive notification via Rx interrupt will do the real work */
2419 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_ALIVE) {
2420 		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
2421 		isr_stats->alive++;
2422 		if (trans->mac_cfg->gen2) {
2423 			/* We can restock, since firmware configured the RFH */
2424 			iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
2425 		}
2426 	}
2427 
2428 	/*
2429 	 * In some rare cases when the HW is in a bad state, we may
2430 	 * get this interrupt too early, when prph_info is still NULL.
2431 	 * So make sure that it's not NULL to prevent crashing.
2432 	 */
2433 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_WAKEUP && trans_pcie->prph_info) {
2434 		u32 sleep_notif =
2435 			le32_to_cpu(trans_pcie->prph_info->sleep_notif);
2436 
2437 		if (sleep_notif == IWL_D3_SLEEP_STATUS_SUSPEND ||
2438 		    sleep_notif == IWL_D3_SLEEP_STATUS_RESUME) {
2439 			IWL_DEBUG_ISR(trans,
2440 				      "Sx interrupt: sleep notification = 0x%x\n",
2441 				      sleep_notif);
2442 			if (trans_pcie->sx_state == IWL_SX_WAITING) {
2443 				trans_pcie->sx_state = IWL_SX_COMPLETE;
2444 				wake_up(&trans_pcie->sx_waitq);
2445 			} else {
2446 				IWL_ERR(trans,
2447 					"unexpected Sx interrupt (0x%x)\n",
2448 					sleep_notif);
2449 			}
2450 		} else {
2451 			/* uCode wakes up after power-down sleep */
2452 			IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
2453 			iwl_pcie_rxq_check_wrptr(trans);
2454 			iwl_pcie_txq_check_wrptrs(trans);
2455 
2456 			isr_stats->wakeup++;
2457 		}
2458 	}
2459 
2460 	/* Chip got too hot and stopped itself */
2461 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_CT_KILL) {
2462 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
2463 		isr_stats->ctkill++;
2464 	}
2465 
2466 	/* HW RF KILL switch toggled */
2467 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RF_KILL)
2468 		iwl_pcie_handle_rfkill_irq(trans, true);
2469 
2470 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_HW_ERR) {
2471 		IWL_ERR(trans,
2472 			"Hardware error detected. Restarting.\n");
2473 
2474 		isr_stats->hw++;
2475 		trans->dbg.hw_error = true;
2476 		iwl_pcie_irq_handle_error(trans);
2477 	}
2478 
2479 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RESET_DONE)
2480 		iwl_trans_pcie_handle_reset_interrupt(trans);
2481 
2482 	if (!polling)
2483 		iwl_pcie_clear_irq(trans, entry->entry);
2484 
2485 	lock_map_release(&trans->sync_cmd_lockdep_map);
2486 
2487 	return IRQ_HANDLED;
2488 }
2489