xref: /linux/drivers/net/wireless/intel/iwlwifi/pcie/gen1_2/rx.c (revision 189f164e573e18d9f8876dbd3ad8fcbe11f93037)
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 = kzalloc_objs(struct iwl_rxq, trans->info.num_rxqs);
777 	trans_pcie->rx_pool = kzalloc_objs(trans_pcie->rx_pool[0],
778 					   RX_POOL_SIZE(trans_pcie->num_rx_bufs));
779 	trans_pcie->global_table =
780 		kzalloc_objs(trans_pcie->global_table[0],
781 			     RX_POOL_SIZE(trans_pcie->num_rx_bufs));
782 	if (!trans_pcie->rxq || !trans_pcie->rx_pool ||
783 	    !trans_pcie->global_table) {
784 		ret = -ENOMEM;
785 		goto err;
786 	}
787 
788 	spin_lock_init(&rba->lock);
789 
790 	/*
791 	 * Allocate the driver's pointer to receive buffer status.
792 	 * Allocate for all queues continuously (HW requirement).
793 	 */
794 	trans_pcie->base_rb_stts =
795 			dma_alloc_coherent(trans->dev,
796 					   rb_stts_size * trans->info.num_rxqs,
797 					   &trans_pcie->base_rb_stts_dma,
798 					   GFP_KERNEL);
799 	if (!trans_pcie->base_rb_stts) {
800 		ret = -ENOMEM;
801 		goto err;
802 	}
803 
804 	for (i = 0; i < trans->info.num_rxqs; i++) {
805 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
806 
807 		rxq->id = i;
808 		ret = iwl_pcie_alloc_rxq_dma(trans, rxq);
809 		if (ret)
810 			goto err;
811 	}
812 	return 0;
813 
814 err:
815 	if (trans_pcie->base_rb_stts) {
816 		dma_free_coherent(trans->dev,
817 				  rb_stts_size * trans->info.num_rxqs,
818 				  trans_pcie->base_rb_stts,
819 				  trans_pcie->base_rb_stts_dma);
820 		trans_pcie->base_rb_stts = NULL;
821 		trans_pcie->base_rb_stts_dma = 0;
822 	}
823 	kfree(trans_pcie->rx_pool);
824 	trans_pcie->rx_pool = NULL;
825 	kfree(trans_pcie->global_table);
826 	trans_pcie->global_table = NULL;
827 	kfree(trans_pcie->rxq);
828 	trans_pcie->rxq = NULL;
829 
830 	return ret;
831 }
832 
iwl_pcie_rx_hw_init(struct iwl_trans * trans,struct iwl_rxq * rxq)833 static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
834 {
835 	u32 rb_size;
836 	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */
837 
838 	switch (trans->conf.rx_buf_size) {
839 	case IWL_AMSDU_4K:
840 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
841 		break;
842 	case IWL_AMSDU_8K:
843 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
844 		break;
845 	case IWL_AMSDU_12K:
846 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_12K;
847 		break;
848 	default:
849 		WARN_ON(1);
850 		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
851 	}
852 
853 	if (!iwl_trans_grab_nic_access(trans))
854 		return;
855 
856 	/* Stop Rx DMA */
857 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
858 	/* reset and flush pointers */
859 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
860 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
861 	iwl_write32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
862 
863 	/* Reset driver's Rx queue write index */
864 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);
865 
866 	/* Tell device where to find RBD circular buffer in DRAM */
867 	iwl_write32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
868 		    (u32)(rxq->bd_dma >> 8));
869 
870 	/* Tell device where in DRAM to update its Rx status */
871 	iwl_write32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
872 		    rxq->rb_stts_dma >> 4);
873 
874 	/* Enable Rx DMA
875 	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
876 	 *      the credit mechanism in 5000 HW RX FIFO
877 	 * Direct rx interrupts to hosts
878 	 * Rx buffer size 4 or 8k or 12k
879 	 * RB timeout 0x10
880 	 * 256 RBDs
881 	 */
882 	iwl_write32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
883 		    FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
884 		    FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
885 		    FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
886 		    rb_size |
887 		    (RX_RB_TIMEOUT << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS) |
888 		    (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));
889 
890 	iwl_trans_release_nic_access(trans);
891 
892 	/* Set interrupt coalescing timer to default (2048 usecs) */
893 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
894 
895 	/* W/A for interrupt coalescing bug in 7260 and 3160 */
896 	if (trans->cfg->host_interrupt_operation_mode)
897 		iwl_set_bit(trans, CSR_INT_COALESCING, IWL_HOST_INT_OPER_MODE);
898 }
899 
iwl_pcie_rx_mq_hw_init(struct iwl_trans * trans)900 static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
901 {
902 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
903 	u32 rb_size, enabled = 0;
904 	int i;
905 
906 	switch (trans->conf.rx_buf_size) {
907 	case IWL_AMSDU_2K:
908 		rb_size = RFH_RXF_DMA_RB_SIZE_2K;
909 		break;
910 	case IWL_AMSDU_4K:
911 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
912 		break;
913 	case IWL_AMSDU_8K:
914 		rb_size = RFH_RXF_DMA_RB_SIZE_8K;
915 		break;
916 	case IWL_AMSDU_12K:
917 		rb_size = RFH_RXF_DMA_RB_SIZE_12K;
918 		break;
919 	default:
920 		WARN_ON(1);
921 		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
922 	}
923 
924 	if (!iwl_trans_grab_nic_access(trans))
925 		return;
926 
927 	/* Stop Rx DMA */
928 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG, 0);
929 	/* disable free amd used rx queue operation */
930 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, 0);
931 
932 	for (i = 0; i < trans->info.num_rxqs; i++) {
933 		/* Tell device where to find RBD free table in DRAM */
934 		iwl_write_prph64_no_grab(trans,
935 					 RFH_Q_FRBDCB_BA_LSB(i),
936 					 trans_pcie->rxq[i].bd_dma);
937 		/* Tell device where to find RBD used table in DRAM */
938 		iwl_write_prph64_no_grab(trans,
939 					 RFH_Q_URBDCB_BA_LSB(i),
940 					 trans_pcie->rxq[i].used_bd_dma);
941 		/* Tell device where in DRAM to update its Rx status */
942 		iwl_write_prph64_no_grab(trans,
943 					 RFH_Q_URBD_STTS_WPTR_LSB(i),
944 					 trans_pcie->rxq[i].rb_stts_dma);
945 		/* Reset device indice tables */
946 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_WIDX(i), 0);
947 		iwl_write_prph_no_grab(trans, RFH_Q_FRBDCB_RIDX(i), 0);
948 		iwl_write_prph_no_grab(trans, RFH_Q_URBDCB_WIDX(i), 0);
949 
950 		enabled |= BIT(i) | BIT(i + 16);
951 	}
952 
953 	/*
954 	 * Enable Rx DMA
955 	 * Rx buffer size 4 or 8k or 12k
956 	 * Min RB size 4 or 8
957 	 * Drop frames that exceed RB size
958 	 * 512 RBDs
959 	 */
960 	iwl_write_prph_no_grab(trans, RFH_RXF_DMA_CFG,
961 			       RFH_DMA_EN_ENABLE_VAL | rb_size |
962 			       RFH_RXF_DMA_MIN_RB_4_8 |
963 			       RFH_RXF_DMA_DROP_TOO_LARGE_MASK |
964 			       RFH_RXF_DMA_RBDCB_SIZE_512);
965 
966 	/*
967 	 * Activate DMA snooping.
968 	 * Set RX DMA chunk size to 64B for IOSF and 128B for PCIe
969 	 * Default queue is 0
970 	 */
971 	iwl_write_prph_no_grab(trans, RFH_GEN_CFG,
972 			       RFH_GEN_CFG_RFH_DMA_SNOOP |
973 			       RFH_GEN_CFG_VAL(DEFAULT_RXQ_NUM, 0) |
974 			       RFH_GEN_CFG_SERVICE_DMA_SNOOP |
975 			       RFH_GEN_CFG_VAL(RB_CHUNK_SIZE,
976 					       trans->mac_cfg->integrated ?
977 					       RFH_GEN_CFG_RB_CHUNK_SIZE_64 :
978 					       RFH_GEN_CFG_RB_CHUNK_SIZE_128));
979 	/* Enable the relevant rx queues */
980 	iwl_write_prph_no_grab(trans, RFH_RXF_RXQ_ACTIVE, enabled);
981 
982 	iwl_trans_release_nic_access(trans);
983 
984 	/* Set interrupt coalescing timer to default (2048 usecs) */
985 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
986 }
987 
iwl_pcie_rx_init_rxb_lists(struct iwl_rxq * rxq)988 void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq)
989 {
990 	lockdep_assert_held(&rxq->lock);
991 
992 	INIT_LIST_HEAD(&rxq->rx_free);
993 	INIT_LIST_HEAD(&rxq->rx_used);
994 	rxq->free_count = 0;
995 	rxq->used_count = 0;
996 }
997 
998 static int iwl_pcie_rx_handle(struct iwl_trans *trans, int queue, int budget);
999 
iwl_netdev_to_trans_pcie(struct net_device * dev)1000 static inline struct iwl_trans_pcie *iwl_netdev_to_trans_pcie(struct net_device *dev)
1001 {
1002 	return *(struct iwl_trans_pcie **)netdev_priv(dev);
1003 }
1004 
iwl_pcie_napi_poll(struct napi_struct * napi,int budget)1005 static int iwl_pcie_napi_poll(struct napi_struct *napi, int budget)
1006 {
1007 	struct iwl_rxq *rxq = container_of(napi, struct iwl_rxq, napi);
1008 	struct iwl_trans_pcie *trans_pcie;
1009 	struct iwl_trans *trans;
1010 	int ret;
1011 
1012 	trans_pcie = iwl_netdev_to_trans_pcie(napi->dev);
1013 	trans = trans_pcie->trans;
1014 
1015 	ret = iwl_pcie_rx_handle(trans, rxq->id, budget);
1016 
1017 	IWL_DEBUG_ISR(trans, "[%d] handled %d, budget %d\n",
1018 		      rxq->id, ret, budget);
1019 
1020 	if (ret < budget) {
1021 		spin_lock(&trans_pcie->irq_lock);
1022 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1023 			_iwl_enable_interrupts(trans);
1024 		spin_unlock(&trans_pcie->irq_lock);
1025 
1026 		napi_complete_done(&rxq->napi, ret);
1027 	}
1028 
1029 	return ret;
1030 }
1031 
iwl_pcie_napi_poll_msix(struct napi_struct * napi,int budget)1032 static int iwl_pcie_napi_poll_msix(struct napi_struct *napi, int budget)
1033 {
1034 	struct iwl_rxq *rxq = container_of(napi, struct iwl_rxq, napi);
1035 	struct iwl_trans_pcie *trans_pcie;
1036 	struct iwl_trans *trans;
1037 	int ret;
1038 
1039 	trans_pcie = iwl_netdev_to_trans_pcie(napi->dev);
1040 	trans = trans_pcie->trans;
1041 
1042 	ret = iwl_pcie_rx_handle(trans, rxq->id, budget);
1043 	IWL_DEBUG_ISR(trans, "[%d] handled %d, budget %d\n", rxq->id, ret,
1044 		      budget);
1045 
1046 	if (ret < budget) {
1047 		int irq_line = rxq->id;
1048 
1049 		/* FIRST_RSS is shared with line 0 */
1050 		if (trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS &&
1051 		    rxq->id == 1)
1052 			irq_line = 0;
1053 
1054 		spin_lock(&trans_pcie->irq_lock);
1055 		iwl_pcie_clear_irq(trans, irq_line);
1056 		spin_unlock(&trans_pcie->irq_lock);
1057 
1058 		napi_complete_done(&rxq->napi, ret);
1059 	}
1060 
1061 	return ret;
1062 }
1063 
iwl_pcie_rx_napi_sync(struct iwl_trans * trans)1064 void iwl_pcie_rx_napi_sync(struct iwl_trans *trans)
1065 {
1066 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1067 	int i;
1068 
1069 	if (unlikely(!trans_pcie->rxq))
1070 		return;
1071 
1072 	for (i = 0; i < trans->info.num_rxqs; i++) {
1073 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1074 
1075 		if (rxq && rxq->napi.poll)
1076 			napi_synchronize(&rxq->napi);
1077 	}
1078 }
1079 
_iwl_pcie_rx_init(struct iwl_trans * trans)1080 static int _iwl_pcie_rx_init(struct iwl_trans *trans)
1081 {
1082 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1083 	struct iwl_rxq *def_rxq;
1084 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1085 	int i, err, queue_size, allocator_pool_size, num_alloc;
1086 
1087 	if (!trans_pcie->rxq) {
1088 		err = iwl_pcie_rx_alloc(trans);
1089 		if (err)
1090 			return err;
1091 	}
1092 	def_rxq = trans_pcie->rxq;
1093 
1094 	cancel_work_sync(&rba->rx_alloc);
1095 
1096 	spin_lock_bh(&rba->lock);
1097 	atomic_set(&rba->req_pending, 0);
1098 	atomic_set(&rba->req_ready, 0);
1099 	INIT_LIST_HEAD(&rba->rbd_allocated);
1100 	INIT_LIST_HEAD(&rba->rbd_empty);
1101 	spin_unlock_bh(&rba->lock);
1102 
1103 	/* free all first - we overwrite everything here */
1104 	iwl_pcie_free_rbs_pool(trans);
1105 
1106 	for (i = 0; i < RX_QUEUE_SIZE; i++)
1107 		def_rxq->queue[i] = NULL;
1108 
1109 	for (i = 0; i < trans->info.num_rxqs; i++) {
1110 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1111 
1112 		spin_lock_bh(&rxq->lock);
1113 		/*
1114 		 * Set read write pointer to reflect that we have processed
1115 		 * and used all buffers, but have not restocked the Rx queue
1116 		 * with fresh buffers
1117 		 */
1118 		rxq->read = 0;
1119 		rxq->write = 0;
1120 		rxq->write_actual = 0;
1121 		memset(rxq->rb_stts, 0,
1122 		       (trans->mac_cfg->device_family >=
1123 			IWL_DEVICE_FAMILY_AX210) ?
1124 		       sizeof(__le16) : sizeof(struct iwl_rb_status));
1125 
1126 		iwl_pcie_rx_init_rxb_lists(rxq);
1127 
1128 		spin_unlock_bh(&rxq->lock);
1129 
1130 		if (!rxq->napi.poll) {
1131 			int (*poll)(struct napi_struct *, int) = iwl_pcie_napi_poll;
1132 
1133 			if (trans_pcie->msix_enabled)
1134 				poll = iwl_pcie_napi_poll_msix;
1135 
1136 			netif_napi_add(trans_pcie->napi_dev, &rxq->napi,
1137 				       poll);
1138 			napi_enable(&rxq->napi);
1139 		}
1140 
1141 	}
1142 
1143 	/* move the pool to the default queue and allocator ownerships */
1144 	queue_size = trans->mac_cfg->mq_rx_supported ?
1145 			trans_pcie->num_rx_bufs - 1 : RX_QUEUE_SIZE;
1146 	allocator_pool_size = trans->info.num_rxqs *
1147 		(RX_CLAIM_REQ_ALLOC - RX_POST_REQ_ALLOC);
1148 	num_alloc = queue_size + allocator_pool_size;
1149 
1150 	for (i = 0; i < num_alloc; i++) {
1151 		struct iwl_rx_mem_buffer *rxb = &trans_pcie->rx_pool[i];
1152 
1153 		if (i < allocator_pool_size)
1154 			list_add(&rxb->list, &rba->rbd_empty);
1155 		else
1156 			list_add(&rxb->list, &def_rxq->rx_used);
1157 		trans_pcie->global_table[i] = rxb;
1158 		rxb->vid = (u16)(i + 1);
1159 		rxb->invalid = true;
1160 	}
1161 
1162 	iwl_pcie_rxq_alloc_rbs(trans, GFP_KERNEL, def_rxq);
1163 
1164 	return 0;
1165 }
1166 
iwl_pcie_rx_init(struct iwl_trans * trans)1167 int iwl_pcie_rx_init(struct iwl_trans *trans)
1168 {
1169 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1170 	int ret = _iwl_pcie_rx_init(trans);
1171 
1172 	if (ret)
1173 		return ret;
1174 
1175 	if (trans->mac_cfg->mq_rx_supported)
1176 		iwl_pcie_rx_mq_hw_init(trans);
1177 	else
1178 		iwl_pcie_rx_hw_init(trans, trans_pcie->rxq);
1179 
1180 	iwl_pcie_rxq_restock(trans, trans_pcie->rxq);
1181 
1182 	spin_lock_bh(&trans_pcie->rxq->lock);
1183 	iwl_pcie_rxq_inc_wr_ptr(trans, trans_pcie->rxq);
1184 	spin_unlock_bh(&trans_pcie->rxq->lock);
1185 
1186 	return 0;
1187 }
1188 
iwl_pcie_gen2_rx_init(struct iwl_trans * trans)1189 int iwl_pcie_gen2_rx_init(struct iwl_trans *trans)
1190 {
1191 	/* Set interrupt coalescing timer to default (2048 usecs) */
1192 	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
1193 
1194 	/*
1195 	 * We don't configure the RFH.
1196 	 * Restock will be done at alive, after firmware configured the RFH.
1197 	 */
1198 	return _iwl_pcie_rx_init(trans);
1199 }
1200 
iwl_pcie_rx_free(struct iwl_trans * trans)1201 void iwl_pcie_rx_free(struct iwl_trans *trans)
1202 {
1203 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1204 	size_t rb_stts_size = iwl_pcie_rb_stts_size(trans);
1205 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1206 	int i;
1207 
1208 	/*
1209 	 * if rxq is NULL, it means that nothing has been allocated,
1210 	 * exit now
1211 	 */
1212 	if (!trans_pcie->rxq) {
1213 		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
1214 		return;
1215 	}
1216 
1217 	cancel_work_sync(&rba->rx_alloc);
1218 
1219 	iwl_pcie_free_rbs_pool(trans);
1220 
1221 	if (trans_pcie->base_rb_stts) {
1222 		dma_free_coherent(trans->dev,
1223 				  rb_stts_size * trans->info.num_rxqs,
1224 				  trans_pcie->base_rb_stts,
1225 				  trans_pcie->base_rb_stts_dma);
1226 		trans_pcie->base_rb_stts = NULL;
1227 		trans_pcie->base_rb_stts_dma = 0;
1228 	}
1229 
1230 	for (i = 0; i < trans->info.num_rxqs; i++) {
1231 		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
1232 
1233 		iwl_pcie_free_rxq_dma(trans, rxq);
1234 
1235 		if (rxq->napi.poll) {
1236 			napi_disable(&rxq->napi);
1237 			netif_napi_del(&rxq->napi);
1238 		}
1239 	}
1240 	kfree(trans_pcie->rx_pool);
1241 	kfree(trans_pcie->global_table);
1242 	kfree(trans_pcie->rxq);
1243 
1244 	if (trans_pcie->alloc_page)
1245 		__free_pages(trans_pcie->alloc_page, trans_pcie->rx_page_order);
1246 }
1247 
iwl_pcie_rx_move_to_allocator(struct iwl_rxq * rxq,struct iwl_rb_allocator * rba)1248 static void iwl_pcie_rx_move_to_allocator(struct iwl_rxq *rxq,
1249 					  struct iwl_rb_allocator *rba)
1250 {
1251 	spin_lock(&rba->lock);
1252 	list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
1253 	spin_unlock(&rba->lock);
1254 }
1255 
1256 /*
1257  * iwl_pcie_rx_reuse_rbd - Recycle used RBDs
1258  *
1259  * Called when a RBD can be reused. The RBD is transferred to the allocator.
1260  * When there are 2 empty RBDs - a request for allocation is posted
1261  */
iwl_pcie_rx_reuse_rbd(struct iwl_trans * trans,struct iwl_rx_mem_buffer * rxb,struct iwl_rxq * rxq,bool emergency)1262 static void iwl_pcie_rx_reuse_rbd(struct iwl_trans *trans,
1263 				  struct iwl_rx_mem_buffer *rxb,
1264 				  struct iwl_rxq *rxq, bool emergency)
1265 {
1266 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1267 	struct iwl_rb_allocator *rba = &trans_pcie->rba;
1268 
1269 	/* Move the RBD to the used list, will be moved to allocator in batches
1270 	 * before claiming or posting a request*/
1271 	list_add_tail(&rxb->list, &rxq->rx_used);
1272 
1273 	if (unlikely(emergency))
1274 		return;
1275 
1276 	/* Count the allocator owned RBDs */
1277 	rxq->used_count++;
1278 
1279 	/* If we have RX_POST_REQ_ALLOC new released rx buffers -
1280 	 * issue a request for allocator. Modulo RX_CLAIM_REQ_ALLOC is
1281 	 * used for the case we failed to claim RX_CLAIM_REQ_ALLOC,
1282 	 * after but we still need to post another request.
1283 	 */
1284 	if ((rxq->used_count % RX_CLAIM_REQ_ALLOC) == RX_POST_REQ_ALLOC) {
1285 		/* Move the 2 RBDs to the allocator ownership.
1286 		 Allocator has another 6 from pool for the request completion*/
1287 		iwl_pcie_rx_move_to_allocator(rxq, rba);
1288 
1289 		atomic_inc(&rba->req_pending);
1290 		queue_work(rba->alloc_wq, &rba->rx_alloc);
1291 	}
1292 }
1293 
iwl_pcie_rx_handle_rb(struct iwl_trans * trans,struct iwl_rxq * rxq,struct iwl_rx_mem_buffer * rxb,bool emergency,int i)1294 static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
1295 				struct iwl_rxq *rxq,
1296 				struct iwl_rx_mem_buffer *rxb,
1297 				bool emergency,
1298 				int i)
1299 {
1300 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1301 	struct iwl_txq *txq = trans_pcie->txqs.txq[trans->conf.cmd_queue];
1302 	bool page_stolen = false;
1303 	int max_len = trans_pcie->rx_buf_bytes;
1304 	u32 offset = 0;
1305 
1306 	if (WARN_ON(!rxb))
1307 		return;
1308 
1309 	dma_unmap_page(trans->dev, rxb->page_dma, max_len, DMA_FROM_DEVICE);
1310 
1311 	while (offset + sizeof(u32) + sizeof(struct iwl_cmd_header) < max_len) {
1312 		struct iwl_rx_packet *pkt;
1313 		bool reclaim;
1314 		int len;
1315 		struct iwl_rx_cmd_buffer rxcb = {
1316 			._offset = rxb->offset + offset,
1317 			._rx_page_order = trans_pcie->rx_page_order,
1318 			._page = rxb->page,
1319 			._page_stolen = false,
1320 			.truesize = max_len,
1321 		};
1322 
1323 		pkt = rxb_addr(&rxcb);
1324 
1325 		if (pkt->len_n_flags == cpu_to_le32(FH_RSCSR_FRAME_INVALID)) {
1326 			IWL_DEBUG_RX(trans,
1327 				     "Q %d: RB end marker at offset %d\n",
1328 				     rxq->id, offset);
1329 			break;
1330 		}
1331 
1332 		WARN((le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1333 			FH_RSCSR_RXQ_POS != rxq->id,
1334 		     "frame on invalid queue - is on %d and indicates %d\n",
1335 		     rxq->id,
1336 		     (le32_to_cpu(pkt->len_n_flags) & FH_RSCSR_RXQ_MASK) >>
1337 			FH_RSCSR_RXQ_POS);
1338 
1339 		IWL_DEBUG_RX(trans,
1340 			     "Q %d: cmd at offset %d: %s (%.2x.%2x, seq 0x%x)\n",
1341 			     rxq->id, offset,
1342 			     iwl_get_cmd_string(trans,
1343 						WIDE_ID(pkt->hdr.group_id, pkt->hdr.cmd)),
1344 			     pkt->hdr.group_id, pkt->hdr.cmd,
1345 			     le16_to_cpu(pkt->hdr.sequence));
1346 
1347 		len = iwl_rx_packet_len(pkt);
1348 		len += sizeof(u32); /* account for status word */
1349 
1350 		offset += ALIGN(len, FH_RSCSR_FRAME_ALIGN);
1351 
1352 		/* check that what the device tells us made sense */
1353 		if (len < sizeof(*pkt) || offset > max_len)
1354 			break;
1355 
1356 		maybe_trace_iwlwifi_dev_rx(trans, pkt, len);
1357 
1358 		/* Reclaim a command buffer only if this packet is a response
1359 		 *   to a (driver-originated) command.
1360 		 * If the packet (e.g. Rx frame) originated from uCode,
1361 		 *   there is no command buffer to reclaim.
1362 		 * Ucode should set SEQ_RX_FRAME bit if ucode-originated,
1363 		 *   but apparently a few don't get set; catch them here. */
1364 		reclaim = !(pkt->hdr.sequence & SEQ_RX_FRAME);
1365 		if (reclaim && !pkt->hdr.group_id) {
1366 			int i;
1367 
1368 			for (i = 0; i < trans->conf.n_no_reclaim_cmds; i++) {
1369 				if (trans->conf.no_reclaim_cmds[i] ==
1370 							pkt->hdr.cmd) {
1371 					reclaim = false;
1372 					break;
1373 				}
1374 			}
1375 		}
1376 
1377 		if (rxq->id == IWL_DEFAULT_RX_QUEUE)
1378 			iwl_op_mode_rx(trans->op_mode, &rxq->napi,
1379 				       &rxcb);
1380 		else
1381 			iwl_op_mode_rx_rss(trans->op_mode, &rxq->napi,
1382 					   &rxcb, rxq->id);
1383 
1384 		/*
1385 		 * After here, we should always check rxcb._page_stolen,
1386 		 * if it is true then one of the handlers took the page.
1387 		 */
1388 
1389 		if (reclaim && txq) {
1390 			u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1391 			int index = SEQ_TO_INDEX(sequence);
1392 			int cmd_index = iwl_txq_get_cmd_index(txq, index);
1393 
1394 			kfree_sensitive(txq->entries[cmd_index].free_buf);
1395 			txq->entries[cmd_index].free_buf = NULL;
1396 
1397 			/* Invoke any callbacks, transfer the buffer to caller,
1398 			 * and fire off the (possibly) blocking
1399 			 * iwl_trans_send_cmd()
1400 			 * as we reclaim the driver command queue */
1401 			if (!rxcb._page_stolen)
1402 				iwl_pcie_hcmd_complete(trans, &rxcb);
1403 			else
1404 				IWL_WARN(trans, "Claim null rxb?\n");
1405 		}
1406 
1407 		page_stolen |= rxcb._page_stolen;
1408 		if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_AX210)
1409 			break;
1410 	}
1411 
1412 	/* page was stolen from us -- free our reference */
1413 	if (page_stolen) {
1414 		__free_pages(rxb->page, trans_pcie->rx_page_order);
1415 		rxb->page = NULL;
1416 	}
1417 
1418 	/* Reuse the page if possible. For notification packets and
1419 	 * SKBs that fail to Rx correctly, add them back into the
1420 	 * rx_free list for reuse later. */
1421 	if (rxb->page != NULL) {
1422 		rxb->page_dma =
1423 			dma_map_page(trans->dev, rxb->page, rxb->offset,
1424 				     trans_pcie->rx_buf_bytes,
1425 				     DMA_FROM_DEVICE);
1426 		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
1427 			/*
1428 			 * free the page(s) as well to not break
1429 			 * the invariant that the items on the used
1430 			 * list have no page(s)
1431 			 */
1432 			__free_pages(rxb->page, trans_pcie->rx_page_order);
1433 			rxb->page = NULL;
1434 			iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1435 		} else {
1436 			list_add_tail(&rxb->list, &rxq->rx_free);
1437 			rxq->free_count++;
1438 		}
1439 	} else
1440 		iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1441 }
1442 
iwl_pcie_get_rxb(struct iwl_trans * trans,struct iwl_rxq * rxq,int i,bool * join)1443 static struct iwl_rx_mem_buffer *iwl_pcie_get_rxb(struct iwl_trans *trans,
1444 						  struct iwl_rxq *rxq, int i,
1445 						  bool *join)
1446 {
1447 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1448 	struct iwl_rx_mem_buffer *rxb;
1449 	u16 vid;
1450 
1451 	BUILD_BUG_ON(sizeof(struct iwl_rx_completion_desc) != 32);
1452 	BUILD_BUG_ON(sizeof(struct iwl_rx_completion_desc_bz) != 4);
1453 
1454 	if (!trans->mac_cfg->mq_rx_supported) {
1455 		rxb = rxq->queue[i];
1456 		rxq->queue[i] = NULL;
1457 		return rxb;
1458 	}
1459 
1460 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) {
1461 		struct iwl_rx_completion_desc_bz *cd = rxq->used_bd;
1462 
1463 		vid = le16_to_cpu(cd[i].rbid);
1464 		*join = cd[i].flags & IWL_RX_CD_FLAGS_FRAGMENTED;
1465 	} else if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
1466 		struct iwl_rx_completion_desc *cd = rxq->used_bd;
1467 
1468 		vid = le16_to_cpu(cd[i].rbid);
1469 		*join = cd[i].flags & IWL_RX_CD_FLAGS_FRAGMENTED;
1470 	} else {
1471 		__le32 *cd = rxq->used_bd;
1472 
1473 		vid = le32_to_cpu(cd[i]) & 0x0FFF; /* 12-bit VID */
1474 	}
1475 
1476 	if (!vid || vid > RX_POOL_SIZE(trans_pcie->num_rx_bufs))
1477 		goto out_err;
1478 
1479 	rxb = trans_pcie->global_table[vid - 1];
1480 	if (rxb->invalid)
1481 		goto out_err;
1482 
1483 	IWL_DEBUG_RX(trans, "Got virtual RB ID %u\n", (u32)rxb->vid);
1484 
1485 	rxb->invalid = true;
1486 
1487 	return rxb;
1488 
1489 out_err:
1490 	WARN(1, "Invalid rxb from HW %u\n", (u32)vid);
1491 	iwl_force_nmi(trans);
1492 	return NULL;
1493 }
1494 
1495 /*
1496  * iwl_pcie_rx_handle - Main entry function for receiving responses from fw
1497  */
iwl_pcie_rx_handle(struct iwl_trans * trans,int queue,int budget)1498 static int iwl_pcie_rx_handle(struct iwl_trans *trans, int queue, int budget)
1499 {
1500 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1501 	struct iwl_rxq *rxq;
1502 	u32 r, i, count = 0, handled = 0;
1503 	bool emergency = false;
1504 
1505 	if (WARN_ON_ONCE(!trans_pcie->rxq || !trans_pcie->rxq[queue].bd))
1506 		return budget;
1507 
1508 	rxq = &trans_pcie->rxq[queue];
1509 
1510 restart:
1511 	spin_lock(&rxq->lock);
1512 	/* uCode's read index (stored in shared DRAM) indicates the last Rx
1513 	 * buffer that the driver may process (last buffer filled by ucode). */
1514 	r = iwl_get_closed_rb_stts(trans, rxq);
1515 	i = rxq->read;
1516 
1517 	/* W/A 9000 device step A0 wrap-around bug */
1518 	r &= (rxq->queue_size - 1);
1519 
1520 	/* Rx interrupt, but nothing sent from uCode */
1521 	if (i == r)
1522 		IWL_DEBUG_RX(trans, "Q %d: HW = SW = %d\n", rxq->id, r);
1523 
1524 	while (i != r && ++handled < budget) {
1525 		struct iwl_rb_allocator *rba = &trans_pcie->rba;
1526 		struct iwl_rx_mem_buffer *rxb;
1527 		/* number of RBDs still waiting for page allocation */
1528 		u32 rb_pending_alloc =
1529 			atomic_read(&trans_pcie->rba.req_pending) *
1530 			RX_CLAIM_REQ_ALLOC;
1531 		bool join = false;
1532 
1533 		if (unlikely(rb_pending_alloc >= rxq->queue_size / 2 &&
1534 			     !emergency)) {
1535 			iwl_pcie_rx_move_to_allocator(rxq, rba);
1536 			emergency = true;
1537 			IWL_DEBUG_TPT(trans,
1538 				      "RX path is in emergency. Pending allocations %d\n",
1539 				      rb_pending_alloc);
1540 		}
1541 
1542 		IWL_DEBUG_RX(trans, "Q %d: HW = %d, SW = %d\n", rxq->id, r, i);
1543 
1544 		rxb = iwl_pcie_get_rxb(trans, rxq, i, &join);
1545 		if (!rxb)
1546 			goto out;
1547 
1548 		if (unlikely(join || rxq->next_rb_is_fragment)) {
1549 			rxq->next_rb_is_fragment = join;
1550 			/*
1551 			 * We can only get a multi-RB in the following cases:
1552 			 *  - firmware issue, sending a too big notification
1553 			 *  - sniffer mode with a large A-MSDU
1554 			 *  - large MTU frames (>2k)
1555 			 * since the multi-RB functionality is limited to newer
1556 			 * hardware that cannot put multiple entries into a
1557 			 * single RB.
1558 			 *
1559 			 * Right now, the higher layers aren't set up to deal
1560 			 * with that, so discard all of these.
1561 			 */
1562 			list_add_tail(&rxb->list, &rxq->rx_free);
1563 			rxq->free_count++;
1564 		} else {
1565 			iwl_pcie_rx_handle_rb(trans, rxq, rxb, emergency, i);
1566 		}
1567 
1568 		i = (i + 1) & (rxq->queue_size - 1);
1569 
1570 		/*
1571 		 * If we have RX_CLAIM_REQ_ALLOC released rx buffers -
1572 		 * try to claim the pre-allocated buffers from the allocator.
1573 		 * If not ready - will try to reclaim next time.
1574 		 * There is no need to reschedule work - allocator exits only
1575 		 * on success
1576 		 */
1577 		if (rxq->used_count >= RX_CLAIM_REQ_ALLOC)
1578 			iwl_pcie_rx_allocator_get(trans, rxq);
1579 
1580 		if (rxq->used_count % RX_CLAIM_REQ_ALLOC == 0 && !emergency) {
1581 			/* Add the remaining empty RBDs for allocator use */
1582 			iwl_pcie_rx_move_to_allocator(rxq, rba);
1583 		} else if (emergency) {
1584 			count++;
1585 			if (count == 8) {
1586 				count = 0;
1587 				if (rb_pending_alloc < rxq->queue_size / 3) {
1588 					IWL_DEBUG_TPT(trans,
1589 						      "RX path exited emergency. Pending allocations %d\n",
1590 						      rb_pending_alloc);
1591 					emergency = false;
1592 				}
1593 
1594 				rxq->read = i;
1595 				spin_unlock(&rxq->lock);
1596 				iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1597 				iwl_pcie_rxq_restock(trans, rxq);
1598 				goto restart;
1599 			}
1600 		}
1601 	}
1602 out:
1603 	/* Backtrack one entry */
1604 	rxq->read = i;
1605 	spin_unlock(&rxq->lock);
1606 
1607 	/*
1608 	 * handle a case where in emergency there are some unallocated RBDs.
1609 	 * those RBDs are in the used list, but are not tracked by the queue's
1610 	 * used_count which counts allocator owned RBDs.
1611 	 * unallocated emergency RBDs must be allocated on exit, otherwise
1612 	 * when called again the function may not be in emergency mode and
1613 	 * they will be handed to the allocator with no tracking in the RBD
1614 	 * allocator counters, which will lead to them never being claimed back
1615 	 * by the queue.
1616 	 * by allocating them here, they are now in the queue free list, and
1617 	 * will be restocked by the next call of iwl_pcie_rxq_restock.
1618 	 */
1619 	if (unlikely(emergency && count))
1620 		iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1621 
1622 	iwl_pcie_rxq_restock(trans, rxq);
1623 
1624 	return handled;
1625 }
1626 
iwl_pcie_get_trans_pcie(struct msix_entry * entry)1627 static struct iwl_trans_pcie *iwl_pcie_get_trans_pcie(struct msix_entry *entry)
1628 {
1629 	u8 queue = entry->entry;
1630 	struct msix_entry *entries = entry - queue;
1631 
1632 	return container_of(entries, struct iwl_trans_pcie, msix_entries[0]);
1633 }
1634 
1635 /*
1636  * iwl_pcie_rx_msix_handle - Main entry function for receiving responses from fw
1637  * This interrupt handler should be used with RSS queue only.
1638  */
iwl_pcie_irq_rx_msix_handler(int irq,void * dev_id)1639 irqreturn_t iwl_pcie_irq_rx_msix_handler(int irq, void *dev_id)
1640 {
1641 	struct msix_entry *entry = dev_id;
1642 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
1643 	struct iwl_trans *trans = trans_pcie->trans;
1644 	struct iwl_rxq *rxq;
1645 
1646 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, false, 0, 0);
1647 
1648 	if (WARN_ON(entry->entry >= trans->info.num_rxqs))
1649 		return IRQ_NONE;
1650 
1651 	if (!trans_pcie->rxq) {
1652 		if (net_ratelimit())
1653 			IWL_ERR(trans,
1654 				"[%d] Got MSI-X interrupt before we have Rx queues\n",
1655 				entry->entry);
1656 		return IRQ_NONE;
1657 	}
1658 
1659 	rxq = &trans_pcie->rxq[entry->entry];
1660 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1661 	IWL_DEBUG_ISR(trans, "[%d] Got interrupt\n", entry->entry);
1662 
1663 	local_bh_disable();
1664 	if (!napi_schedule(&rxq->napi))
1665 		iwl_pcie_clear_irq(trans, entry->entry);
1666 	local_bh_enable();
1667 
1668 	lock_map_release(&trans->sync_cmd_lockdep_map);
1669 
1670 	return IRQ_HANDLED;
1671 }
1672 
1673 /*
1674  * iwl_pcie_irq_handle_error - called for HW or SW error interrupt from card
1675  */
iwl_pcie_irq_handle_error(struct iwl_trans * trans)1676 static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
1677 {
1678 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1679 	int i;
1680 
1681 	/* W/A for WiFi/WiMAX coex and WiMAX own the RF */
1682 	if (trans->cfg->internal_wimax_coex &&
1683 	    !trans->mac_cfg->base->apmg_not_supported &&
1684 	    (!(iwl_read_prph(trans, APMG_CLK_CTRL_REG) &
1685 			     APMS_CLK_VAL_MRB_FUNC_MODE) ||
1686 	     (iwl_read_prph(trans, APMG_PS_CTRL_REG) &
1687 			    APMG_PS_CTRL_VAL_RESET_REQ))) {
1688 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1689 		iwl_op_mode_wimax_active(trans->op_mode);
1690 		wake_up(&trans_pcie->wait_command_queue);
1691 		return;
1692 	}
1693 
1694 	for (i = 0; i < trans->mac_cfg->base->num_of_queues; i++) {
1695 		if (!trans_pcie->txqs.txq[i])
1696 			continue;
1697 		timer_delete(&trans_pcie->txqs.txq[i]->stuck_timer);
1698 	}
1699 
1700 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_SC) {
1701 		u32 val = iwl_read32(trans, CSR_IPC_STATE);
1702 
1703 		if (val & CSR_IPC_STATE_TOP_RESET_REQ) {
1704 			IWL_ERR(trans, "FW requested TOP reset for FSEQ\n");
1705 			trans->do_top_reset = 1;
1706 		}
1707 	}
1708 
1709 	/* The STATUS_FW_ERROR bit is set in this function. This must happen
1710 	 * before we wake up the command caller, to ensure a proper cleanup. */
1711 	iwl_trans_fw_error(trans, IWL_ERR_TYPE_IRQ);
1712 
1713 	clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1714 	wake_up(&trans_pcie->wait_command_queue);
1715 }
1716 
iwl_pcie_int_cause_non_ict(struct iwl_trans * trans)1717 static u32 iwl_pcie_int_cause_non_ict(struct iwl_trans *trans)
1718 {
1719 	u32 inta;
1720 
1721 	lockdep_assert_held(&IWL_TRANS_GET_PCIE_TRANS(trans)->irq_lock);
1722 
1723 	trace_iwlwifi_dev_irq(trans->dev);
1724 
1725 	/* Discover which interrupts are active/pending */
1726 	inta = iwl_read32(trans, CSR_INT);
1727 
1728 	/* the thread will service interrupts and re-enable them */
1729 	return inta;
1730 }
1731 
1732 /* a device (PCI-E) page is 4096 bytes long */
1733 #define ICT_SHIFT	12
1734 #define ICT_SIZE	(1 << ICT_SHIFT)
1735 #define ICT_COUNT	(ICT_SIZE / sizeof(u32))
1736 
1737 /* interrupt handler using ict table, with this interrupt driver will
1738  * stop using INTA register to get device's interrupt, reading this register
1739  * is expensive, device will write interrupts in ICT dram table, increment
1740  * index then will fire interrupt to driver, driver will OR all ICT table
1741  * entries from current index up to table entry with 0 value. the result is
1742  * the interrupt we need to service, driver will set the entries back to 0 and
1743  * set index.
1744  */
iwl_pcie_int_cause_ict(struct iwl_trans * trans)1745 static u32 iwl_pcie_int_cause_ict(struct iwl_trans *trans)
1746 {
1747 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1748 	u32 inta;
1749 	u32 val = 0;
1750 	u32 read;
1751 
1752 	trace_iwlwifi_dev_irq(trans->dev);
1753 
1754 	/* Ignore interrupt if there's nothing in NIC to service.
1755 	 * This may be due to IRQ shared with another device,
1756 	 * or due to sporadic interrupts thrown from our NIC. */
1757 	read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1758 	trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index, read);
1759 	if (!read)
1760 		return 0;
1761 
1762 	/*
1763 	 * Collect all entries up to the first 0, starting from ict_index;
1764 	 * note we already read at ict_index.
1765 	 */
1766 	do {
1767 		val |= read;
1768 		IWL_DEBUG_ISR(trans, "ICT index %d value 0x%08X\n",
1769 				trans_pcie->ict_index, read);
1770 		trans_pcie->ict_tbl[trans_pcie->ict_index] = 0;
1771 		trans_pcie->ict_index =
1772 			((trans_pcie->ict_index + 1) & (ICT_COUNT - 1));
1773 
1774 		read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
1775 		trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index,
1776 					   read);
1777 	} while (read);
1778 
1779 	/* We should not get this value, just ignore it. */
1780 	if (val == 0xffffffff)
1781 		val = 0;
1782 
1783 	/*
1784 	 * this is a w/a for a h/w bug. the h/w bug may cause the Rx bit
1785 	 * (bit 15 before shifting it to 31) to clear when using interrupt
1786 	 * coalescing. fortunately, bits 18 and 19 stay set when this happens
1787 	 * so we use them to decide on the real state of the Rx bit.
1788 	 * In order words, bit 15 is set if bit 18 or bit 19 are set.
1789 	 */
1790 	if (val & 0xC0000)
1791 		val |= 0x8000;
1792 
1793 	inta = (0xff & val) | ((0xff00 & val) << 16);
1794 	return inta;
1795 }
1796 
iwl_pcie_handle_rfkill_irq(struct iwl_trans * trans,bool from_irq)1797 void iwl_pcie_handle_rfkill_irq(struct iwl_trans *trans, bool from_irq)
1798 {
1799 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1800 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1801 	bool hw_rfkill, prev, report;
1802 
1803 	mutex_lock(&trans_pcie->mutex);
1804 	prev = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1805 	hw_rfkill = iwl_is_rfkill_set(trans);
1806 	if (hw_rfkill) {
1807 		set_bit(STATUS_RFKILL_OPMODE, &trans->status);
1808 		set_bit(STATUS_RFKILL_HW, &trans->status);
1809 	}
1810 	if (trans_pcie->opmode_down)
1811 		report = hw_rfkill;
1812 	else
1813 		report = test_bit(STATUS_RFKILL_OPMODE, &trans->status);
1814 
1815 	IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
1816 		 hw_rfkill ? "disable radio" : "enable radio");
1817 
1818 	isr_stats->rfkill++;
1819 
1820 	if (prev != report)
1821 		iwl_trans_pcie_rf_kill(trans, report, from_irq);
1822 	mutex_unlock(&trans_pcie->mutex);
1823 
1824 	if (hw_rfkill) {
1825 		if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
1826 				       &trans->status))
1827 			IWL_DEBUG_RF_KILL(trans,
1828 					  "Rfkill while SYNC HCMD in flight\n");
1829 		wake_up(&trans_pcie->wait_command_queue);
1830 	} else {
1831 		clear_bit(STATUS_RFKILL_HW, &trans->status);
1832 		if (trans_pcie->opmode_down)
1833 			clear_bit(STATUS_RFKILL_OPMODE, &trans->status);
1834 	}
1835 }
1836 
iwl_trans_pcie_handle_reset_interrupt(struct iwl_trans * trans)1837 static void iwl_trans_pcie_handle_reset_interrupt(struct iwl_trans *trans)
1838 {
1839 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1840 	u32 state;
1841 
1842 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_SC) {
1843 		u32 val = iwl_read32(trans, CSR_IPC_STATE);
1844 
1845 		state = u32_get_bits(val, CSR_IPC_STATE_RESET);
1846 		IWL_DEBUG_ISR(trans, "IPC state = 0x%x/%d\n", val, state);
1847 	} else {
1848 		state = CSR_IPC_STATE_RESET_SW_READY;
1849 	}
1850 
1851 	switch (state) {
1852 	case CSR_IPC_STATE_RESET_SW_READY:
1853 		if (trans_pcie->fw_reset_state == FW_RESET_REQUESTED) {
1854 			IWL_DEBUG_ISR(trans, "Reset flow completed\n");
1855 			trans_pcie->fw_reset_state = FW_RESET_OK;
1856 			wake_up(&trans_pcie->fw_reset_waitq);
1857 			break;
1858 		}
1859 		fallthrough;
1860 	case CSR_IPC_STATE_RESET_TOP_READY:
1861 		if (trans_pcie->fw_reset_state == FW_RESET_TOP_REQUESTED) {
1862 			IWL_DEBUG_ISR(trans, "TOP Reset continues\n");
1863 			trans_pcie->fw_reset_state = FW_RESET_OK;
1864 			wake_up(&trans_pcie->fw_reset_waitq);
1865 			break;
1866 		}
1867 		fallthrough;
1868 	case CSR_IPC_STATE_RESET_NONE:
1869 		IWL_FW_CHECK_FAILED(trans,
1870 				    "Invalid reset interrupt (state=%d)!\n",
1871 				    state);
1872 		break;
1873 	case CSR_IPC_STATE_RESET_TOP_FOLLOWER:
1874 		if (trans_pcie->fw_reset_state == FW_RESET_REQUESTED) {
1875 			/* if we were in reset, wake that up */
1876 			IWL_INFO(trans,
1877 				 "TOP reset from BT while doing reset\n");
1878 			trans_pcie->fw_reset_state = FW_RESET_OK;
1879 			wake_up(&trans_pcie->fw_reset_waitq);
1880 		} else {
1881 			IWL_INFO(trans, "TOP reset from BT\n");
1882 			trans->state = IWL_TRANS_NO_FW;
1883 			iwl_trans_schedule_reset(trans,
1884 						 IWL_ERR_TYPE_TOP_RESET_BY_BT);
1885 		}
1886 		break;
1887 	}
1888 }
1889 
iwl_pcie_irq_handler(int irq,void * dev_id)1890 irqreturn_t iwl_pcie_irq_handler(int irq, void *dev_id)
1891 {
1892 	struct iwl_trans *trans = dev_id;
1893 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1894 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1895 	u32 inta = 0;
1896 	u32 handled = 0;
1897 	bool polling = false;
1898 
1899 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
1900 
1901 	spin_lock_bh(&trans_pcie->irq_lock);
1902 
1903 	/* dram interrupt table not set yet,
1904 	 * use legacy interrupt.
1905 	 */
1906 	if (likely(trans_pcie->use_ict))
1907 		inta = iwl_pcie_int_cause_ict(trans);
1908 	else
1909 		inta = iwl_pcie_int_cause_non_ict(trans);
1910 
1911 	if (iwl_have_debug_level(IWL_DL_ISR)) {
1912 		IWL_DEBUG_ISR(trans,
1913 			      "ISR inta 0x%08x, enabled 0x%08x(sw), enabled(hw) 0x%08x, fh 0x%08x\n",
1914 			      inta, trans_pcie->inta_mask,
1915 			      iwl_read32(trans, CSR_INT_MASK),
1916 			      iwl_read32(trans, CSR_FH_INT_STATUS));
1917 		if (inta & (~trans_pcie->inta_mask))
1918 			IWL_DEBUG_ISR(trans,
1919 				      "We got a masked interrupt (0x%08x)\n",
1920 				      inta & (~trans_pcie->inta_mask));
1921 	}
1922 
1923 	inta &= trans_pcie->inta_mask;
1924 
1925 	/*
1926 	 * Ignore interrupt if there's nothing in NIC to service.
1927 	 * This may be due to IRQ shared with another device,
1928 	 * or due to sporadic interrupts thrown from our NIC.
1929 	 */
1930 	if (unlikely(!inta)) {
1931 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
1932 		/*
1933 		 * Re-enable interrupts here since we don't
1934 		 * have anything to service
1935 		 */
1936 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
1937 			_iwl_enable_interrupts(trans);
1938 		spin_unlock_bh(&trans_pcie->irq_lock);
1939 		lock_map_release(&trans->sync_cmd_lockdep_map);
1940 		return IRQ_NONE;
1941 	}
1942 
1943 	if (unlikely(inta == 0xFFFFFFFF || iwl_trans_is_hw_error_value(inta))) {
1944 		/*
1945 		 * Hardware disappeared. It might have
1946 		 * already raised an interrupt.
1947 		 */
1948 		IWL_WARN(trans, "HARDWARE GONE?? INTA == 0x%08x\n", inta);
1949 		spin_unlock_bh(&trans_pcie->irq_lock);
1950 		goto out;
1951 	}
1952 
1953 	/* Ack/clear/reset pending uCode interrupts.
1954 	 * Note:  Some bits in CSR_INT are "OR" of bits in CSR_FH_INT_STATUS,
1955 	 */
1956 	/* There is a hardware bug in the interrupt mask function that some
1957 	 * interrupts (i.e. CSR_INT_BIT_SCD) can still be generated even if
1958 	 * they are disabled in the CSR_INT_MASK register. Furthermore the
1959 	 * ICT interrupt handling mechanism has another bug that might cause
1960 	 * these unmasked interrupts fail to be detected. We workaround the
1961 	 * hardware bugs here by ACKing all the possible interrupts so that
1962 	 * interrupt coalescing can still be achieved.
1963 	 */
1964 	iwl_write32(trans, CSR_INT, inta | ~trans_pcie->inta_mask);
1965 
1966 	if (iwl_have_debug_level(IWL_DL_ISR))
1967 		IWL_DEBUG_ISR(trans, "inta 0x%08x, enabled 0x%08x\n",
1968 			      inta, iwl_read32(trans, CSR_INT_MASK));
1969 
1970 	spin_unlock_bh(&trans_pcie->irq_lock);
1971 
1972 	/* Now service all interrupt bits discovered above. */
1973 	if (inta & CSR_INT_BIT_HW_ERR) {
1974 		IWL_ERR(trans, "Hardware error detected.  Restarting.\n");
1975 
1976 		/* Tell the device to stop sending interrupts */
1977 		iwl_disable_interrupts(trans);
1978 
1979 		isr_stats->hw++;
1980 		iwl_pcie_irq_handle_error(trans);
1981 
1982 		handled |= CSR_INT_BIT_HW_ERR;
1983 
1984 		goto out;
1985 	}
1986 
1987 	/* NIC fires this, but we don't use it, redundant with WAKEUP */
1988 	if (inta & CSR_INT_BIT_SCD) {
1989 		IWL_DEBUG_ISR(trans,
1990 			      "Scheduler finished to transmit the frame/frames.\n");
1991 		isr_stats->sch++;
1992 	}
1993 
1994 	/* Alive notification via Rx interrupt will do the real work */
1995 	if (inta & CSR_INT_BIT_ALIVE) {
1996 		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
1997 		isr_stats->alive++;
1998 		if (trans->mac_cfg->gen2) {
1999 			/*
2000 			 * We can restock, since firmware configured
2001 			 * the RFH
2002 			 */
2003 			iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
2004 		}
2005 
2006 		handled |= CSR_INT_BIT_ALIVE;
2007 	}
2008 
2009 	if (inta & CSR_INT_BIT_RESET_DONE) {
2010 		iwl_trans_pcie_handle_reset_interrupt(trans);
2011 		handled |= CSR_INT_BIT_RESET_DONE;
2012 	}
2013 
2014 	/* Safely ignore these bits for debug checks below */
2015 	inta &= ~(CSR_INT_BIT_SCD | CSR_INT_BIT_ALIVE);
2016 
2017 	/* HW RF KILL switch toggled */
2018 	if (inta & CSR_INT_BIT_RF_KILL) {
2019 		iwl_pcie_handle_rfkill_irq(trans, true);
2020 		handled |= CSR_INT_BIT_RF_KILL;
2021 	}
2022 
2023 	/* Chip got too hot and stopped itself */
2024 	if (inta & CSR_INT_BIT_CT_KILL) {
2025 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
2026 		isr_stats->ctkill++;
2027 		handled |= CSR_INT_BIT_CT_KILL;
2028 	}
2029 
2030 	/* Error detected by uCode */
2031 	if (inta & CSR_INT_BIT_SW_ERR) {
2032 		IWL_ERR(trans, "Microcode SW error detected. "
2033 			" Restarting 0x%X.\n", inta);
2034 		isr_stats->sw++;
2035 		if (trans_pcie->fw_reset_state == FW_RESET_REQUESTED) {
2036 			trans_pcie->fw_reset_state = FW_RESET_ERROR;
2037 			wake_up(&trans_pcie->fw_reset_waitq);
2038 		} else {
2039 			iwl_pcie_irq_handle_error(trans);
2040 		}
2041 		handled |= CSR_INT_BIT_SW_ERR;
2042 	}
2043 
2044 	/* uCode wakes up after power-down sleep */
2045 	if (inta & CSR_INT_BIT_WAKEUP) {
2046 		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
2047 		iwl_pcie_rxq_check_wrptr(trans);
2048 		iwl_pcie_txq_check_wrptrs(trans);
2049 
2050 		isr_stats->wakeup++;
2051 
2052 		handled |= CSR_INT_BIT_WAKEUP;
2053 	}
2054 
2055 	/* All uCode command responses, including Tx command responses,
2056 	 * Rx "responses" (frame-received notification), and other
2057 	 * notifications from uCode come through here*/
2058 	if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX |
2059 		    CSR_INT_BIT_RX_PERIODIC)) {
2060 		IWL_DEBUG_ISR(trans, "Rx interrupt\n");
2061 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX)) {
2062 			handled |= (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX);
2063 			iwl_write32(trans, CSR_FH_INT_STATUS,
2064 					CSR_FH_INT_RX_MASK);
2065 		}
2066 		if (inta & CSR_INT_BIT_RX_PERIODIC) {
2067 			handled |= CSR_INT_BIT_RX_PERIODIC;
2068 			iwl_write32(trans,
2069 				CSR_INT, CSR_INT_BIT_RX_PERIODIC);
2070 		}
2071 		/* Sending RX interrupt require many steps to be done in the
2072 		 * device:
2073 		 * 1- write interrupt to current index in ICT table.
2074 		 * 2- dma RX frame.
2075 		 * 3- update RX shared data to indicate last write index.
2076 		 * 4- send interrupt.
2077 		 * This could lead to RX race, driver could receive RX interrupt
2078 		 * but the shared data changes does not reflect this;
2079 		 * periodic interrupt will detect any dangling Rx activity.
2080 		 */
2081 
2082 		/* Disable periodic interrupt; we use it as just a one-shot. */
2083 		iwl_write8(trans, CSR_INT_PERIODIC_REG,
2084 			    CSR_INT_PERIODIC_DIS);
2085 
2086 		/*
2087 		 * Enable periodic interrupt in 8 msec only if we received
2088 		 * real RX interrupt (instead of just periodic int), to catch
2089 		 * any dangling Rx interrupt.  If it was just the periodic
2090 		 * interrupt, there was no dangling Rx activity, and no need
2091 		 * to extend the periodic interrupt; one-shot is enough.
2092 		 */
2093 		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX))
2094 			iwl_write8(trans, CSR_INT_PERIODIC_REG,
2095 				   CSR_INT_PERIODIC_ENA);
2096 
2097 		isr_stats->rx++;
2098 
2099 		local_bh_disable();
2100 		if (napi_schedule_prep(&trans_pcie->rxq[0].napi)) {
2101 			polling = true;
2102 			__napi_schedule(&trans_pcie->rxq[0].napi);
2103 		}
2104 		local_bh_enable();
2105 	}
2106 
2107 	/* This "Tx" DMA channel is used only for loading uCode */
2108 	if (inta & CSR_INT_BIT_FH_TX) {
2109 		iwl_write32(trans, CSR_FH_INT_STATUS, CSR_FH_INT_TX_MASK);
2110 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
2111 		isr_stats->tx++;
2112 		handled |= CSR_INT_BIT_FH_TX;
2113 		/* Wake up uCode load routine, now that load is complete */
2114 		trans_pcie->ucode_write_complete = true;
2115 		wake_up(&trans_pcie->ucode_write_waitq);
2116 		/* Wake up IMR write routine, now that write to SRAM is complete */
2117 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2118 			trans_pcie->imr_status = IMR_D2S_COMPLETED;
2119 			wake_up(&trans_pcie->ucode_write_waitq);
2120 		}
2121 	}
2122 
2123 	if (inta & ~handled) {
2124 		IWL_ERR(trans, "Unhandled INTA bits 0x%08x\n", inta & ~handled);
2125 		isr_stats->unhandled++;
2126 	}
2127 
2128 	if (inta & ~(trans_pcie->inta_mask)) {
2129 		IWL_WARN(trans, "Disabled INTA bits 0x%08x were pending\n",
2130 			 inta & ~trans_pcie->inta_mask);
2131 	}
2132 
2133 	if (!polling) {
2134 		spin_lock_bh(&trans_pcie->irq_lock);
2135 		/* only Re-enable all interrupt if disabled by irq */
2136 		if (test_bit(STATUS_INT_ENABLED, &trans->status))
2137 			_iwl_enable_interrupts(trans);
2138 		/* we are loading the firmware, enable FH_TX interrupt only */
2139 		else if (handled & CSR_INT_BIT_FH_TX)
2140 			iwl_enable_fw_load_int(trans);
2141 		/* Re-enable RF_KILL if it occurred */
2142 		else if (handled & CSR_INT_BIT_RF_KILL)
2143 			iwl_enable_rfkill_int(trans);
2144 		/* Re-enable the ALIVE / Rx interrupt if it occurred */
2145 		else if (handled & (CSR_INT_BIT_ALIVE | CSR_INT_BIT_FH_RX))
2146 			iwl_enable_fw_load_int_ctx_info(trans, false);
2147 		spin_unlock_bh(&trans_pcie->irq_lock);
2148 	}
2149 
2150 out:
2151 	lock_map_release(&trans->sync_cmd_lockdep_map);
2152 	return IRQ_HANDLED;
2153 }
2154 
2155 /******************************************************************************
2156  *
2157  * ICT functions
2158  *
2159  ******************************************************************************/
2160 
2161 /* Free dram table */
iwl_pcie_free_ict(struct iwl_trans * trans)2162 void iwl_pcie_free_ict(struct iwl_trans *trans)
2163 {
2164 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2165 
2166 	if (trans_pcie->ict_tbl) {
2167 		dma_free_coherent(trans->dev, ICT_SIZE,
2168 				  trans_pcie->ict_tbl,
2169 				  trans_pcie->ict_tbl_dma);
2170 		trans_pcie->ict_tbl = NULL;
2171 		trans_pcie->ict_tbl_dma = 0;
2172 	}
2173 }
2174 
2175 /*
2176  * allocate dram shared table, it is an aligned memory
2177  * block of ICT_SIZE.
2178  * also reset all data related to ICT table interrupt.
2179  */
iwl_pcie_alloc_ict(struct iwl_trans * trans)2180 int iwl_pcie_alloc_ict(struct iwl_trans *trans)
2181 {
2182 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2183 
2184 	trans_pcie->ict_tbl =
2185 		dma_alloc_coherent(trans->dev, ICT_SIZE,
2186 				   &trans_pcie->ict_tbl_dma, GFP_KERNEL);
2187 	if (!trans_pcie->ict_tbl)
2188 		return -ENOMEM;
2189 
2190 	/* just an API sanity check ... it is guaranteed to be aligned */
2191 	if (WARN_ON(trans_pcie->ict_tbl_dma & (ICT_SIZE - 1))) {
2192 		iwl_pcie_free_ict(trans);
2193 		return -EINVAL;
2194 	}
2195 
2196 	return 0;
2197 }
2198 
2199 /* Device is going up inform it about using ICT interrupt table,
2200  * also we need to tell the driver to start using ICT interrupt.
2201  */
iwl_pcie_reset_ict(struct iwl_trans * trans)2202 void iwl_pcie_reset_ict(struct iwl_trans *trans)
2203 {
2204 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2205 	u32 val;
2206 
2207 	if (!trans_pcie->ict_tbl)
2208 		return;
2209 
2210 	spin_lock_bh(&trans_pcie->irq_lock);
2211 	_iwl_disable_interrupts(trans);
2212 
2213 	memset(trans_pcie->ict_tbl, 0, ICT_SIZE);
2214 
2215 	val = trans_pcie->ict_tbl_dma >> ICT_SHIFT;
2216 
2217 	val |= CSR_DRAM_INT_TBL_ENABLE |
2218 	       CSR_DRAM_INIT_TBL_WRAP_CHECK |
2219 	       CSR_DRAM_INIT_TBL_WRITE_POINTER;
2220 
2221 	IWL_DEBUG_ISR(trans, "CSR_DRAM_INT_TBL_REG =0x%x\n", val);
2222 
2223 	iwl_write32(trans, CSR_DRAM_INT_TBL_REG, val);
2224 	trans_pcie->use_ict = true;
2225 	trans_pcie->ict_index = 0;
2226 	iwl_write32(trans, CSR_INT, trans_pcie->inta_mask);
2227 	_iwl_enable_interrupts(trans);
2228 	spin_unlock_bh(&trans_pcie->irq_lock);
2229 }
2230 
2231 /* Device is going down disable ict interrupt usage */
iwl_pcie_disable_ict(struct iwl_trans * trans)2232 void iwl_pcie_disable_ict(struct iwl_trans *trans)
2233 {
2234 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2235 
2236 	spin_lock_bh(&trans_pcie->irq_lock);
2237 	trans_pcie->use_ict = false;
2238 	spin_unlock_bh(&trans_pcie->irq_lock);
2239 }
2240 
iwl_pcie_isr(int irq,void * data)2241 irqreturn_t iwl_pcie_isr(int irq, void *data)
2242 {
2243 	struct iwl_trans *trans = data;
2244 
2245 	if (!trans)
2246 		return IRQ_NONE;
2247 
2248 	/* Disable (but don't clear!) interrupts here to avoid
2249 	 * back-to-back ISRs and sporadic interrupts from our NIC.
2250 	 * If we have something to service, the tasklet will re-enable ints.
2251 	 * If we *don't* have something, we'll re-enable before leaving here.
2252 	 */
2253 	iwl_write32(trans, CSR_INT_MASK, 0x00000000);
2254 
2255 	return IRQ_WAKE_THREAD;
2256 }
2257 
iwl_pcie_msix_isr(int irq,void * data)2258 irqreturn_t iwl_pcie_msix_isr(int irq, void *data)
2259 {
2260 	return IRQ_WAKE_THREAD;
2261 }
2262 
iwl_pcie_irq_msix_handler(int irq,void * dev_id)2263 irqreturn_t iwl_pcie_irq_msix_handler(int irq, void *dev_id)
2264 {
2265 	struct msix_entry *entry = dev_id;
2266 	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
2267 	struct iwl_trans *trans = trans_pcie->trans;
2268 	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
2269 	u32 inta_fh_msk = ~MSIX_FH_INT_CAUSES_DATA_QUEUE;
2270 	u32 inta_fh, inta_hw;
2271 	bool polling = false;
2272 	bool sw_err;
2273 
2274 	if (trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_NON_RX)
2275 		inta_fh_msk |= MSIX_FH_INT_CAUSES_Q0;
2276 
2277 	if (trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS)
2278 		inta_fh_msk |= MSIX_FH_INT_CAUSES_Q1;
2279 
2280 	lock_map_acquire(&trans->sync_cmd_lockdep_map);
2281 
2282 	spin_lock_bh(&trans_pcie->irq_lock);
2283 	inta_fh = iwl_read32(trans, CSR_MSIX_FH_INT_CAUSES_AD);
2284 	inta_hw = iwl_read32(trans, CSR_MSIX_HW_INT_CAUSES_AD);
2285 	/*
2286 	 * Clear causes registers to avoid being handling the same cause.
2287 	 */
2288 	iwl_write32(trans, CSR_MSIX_FH_INT_CAUSES_AD, inta_fh & inta_fh_msk);
2289 	iwl_write32(trans, CSR_MSIX_HW_INT_CAUSES_AD, inta_hw);
2290 	spin_unlock_bh(&trans_pcie->irq_lock);
2291 
2292 	trace_iwlwifi_dev_irq_msix(trans->dev, entry, true, inta_fh, inta_hw);
2293 
2294 	if (unlikely(!(inta_fh | inta_hw))) {
2295 		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
2296 		lock_map_release(&trans->sync_cmd_lockdep_map);
2297 		return IRQ_NONE;
2298 	}
2299 
2300 	if (iwl_have_debug_level(IWL_DL_ISR)) {
2301 		IWL_DEBUG_ISR(trans,
2302 			      "ISR[%d] inta_fh 0x%08x, enabled (sw) 0x%08x (hw) 0x%08x\n",
2303 			      entry->entry, inta_fh, trans_pcie->fh_mask,
2304 			      iwl_read32(trans, CSR_MSIX_FH_INT_MASK_AD));
2305 		if (inta_fh & ~trans_pcie->fh_mask)
2306 			IWL_DEBUG_ISR(trans,
2307 				      "We got a masked interrupt (0x%08x)\n",
2308 				      inta_fh & ~trans_pcie->fh_mask);
2309 	}
2310 
2311 	inta_fh &= trans_pcie->fh_mask;
2312 
2313 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_NON_RX) &&
2314 	    inta_fh & MSIX_FH_INT_CAUSES_Q0) {
2315 		local_bh_disable();
2316 		if (napi_schedule_prep(&trans_pcie->rxq[0].napi)) {
2317 			polling = true;
2318 			__napi_schedule(&trans_pcie->rxq[0].napi);
2319 		}
2320 		local_bh_enable();
2321 	}
2322 
2323 	if ((trans_pcie->shared_vec_mask & IWL_SHARED_IRQ_FIRST_RSS) &&
2324 	    inta_fh & MSIX_FH_INT_CAUSES_Q1) {
2325 		local_bh_disable();
2326 		if (napi_schedule_prep(&trans_pcie->rxq[1].napi)) {
2327 			polling = true;
2328 			__napi_schedule(&trans_pcie->rxq[1].napi);
2329 		}
2330 		local_bh_enable();
2331 	}
2332 
2333 	/* This "Tx" DMA channel is used only for loading uCode */
2334 	if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM &&
2335 	    trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2336 		IWL_DEBUG_ISR(trans, "IMR Complete interrupt\n");
2337 		isr_stats->tx++;
2338 
2339 		/* Wake up IMR routine once write to SRAM is complete */
2340 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2341 			trans_pcie->imr_status = IMR_D2S_COMPLETED;
2342 			wake_up(&trans_pcie->ucode_write_waitq);
2343 		}
2344 	} else if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM) {
2345 		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
2346 		isr_stats->tx++;
2347 		/*
2348 		 * Wake up uCode load routine,
2349 		 * now that load is complete
2350 		 */
2351 		trans_pcie->ucode_write_complete = true;
2352 		wake_up(&trans_pcie->ucode_write_waitq);
2353 
2354 		/* Wake up IMR routine once write to SRAM is complete */
2355 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2356 			trans_pcie->imr_status = IMR_D2S_COMPLETED;
2357 			wake_up(&trans_pcie->ucode_write_waitq);
2358 		}
2359 	}
2360 
2361 	if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_BZ)
2362 		sw_err = inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR_BZ;
2363 	else
2364 		sw_err = inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR;
2365 
2366 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_TOP_FATAL_ERR) {
2367 		IWL_ERR(trans, "TOP Fatal error detected, inta_hw=0x%x.\n",
2368 			inta_hw);
2369 		if (trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) {
2370 			trans->request_top_reset = 1;
2371 			iwl_op_mode_nic_error(trans->op_mode,
2372 					      IWL_ERR_TYPE_TOP_FATAL_ERROR);
2373 			iwl_trans_schedule_reset(trans,
2374 						 IWL_ERR_TYPE_TOP_FATAL_ERROR);
2375 		}
2376 	}
2377 
2378 	/* Error detected by uCode */
2379 	if ((inta_fh & MSIX_FH_INT_CAUSES_FH_ERR) || sw_err) {
2380 		IWL_ERR(trans,
2381 			"Microcode SW error detected. Restarting 0x%X.\n",
2382 			inta_fh);
2383 		isr_stats->sw++;
2384 		/* during FW reset flow report errors from there */
2385 		if (trans_pcie->imr_status == IMR_D2S_REQUESTED) {
2386 			trans_pcie->imr_status = IMR_D2S_ERROR;
2387 			wake_up(&trans_pcie->imr_waitq);
2388 		} else if (trans_pcie->fw_reset_state == FW_RESET_REQUESTED) {
2389 			trans_pcie->fw_reset_state = FW_RESET_ERROR;
2390 			wake_up(&trans_pcie->fw_reset_waitq);
2391 		} else {
2392 			iwl_pcie_irq_handle_error(trans);
2393 		}
2394 
2395 		if (trans_pcie->sx_state == IWL_SX_WAITING) {
2396 			trans_pcie->sx_state = IWL_SX_ERROR;
2397 			wake_up(&trans_pcie->sx_waitq);
2398 		}
2399 	}
2400 
2401 	/* After checking FH register check HW register */
2402 	if (iwl_have_debug_level(IWL_DL_ISR)) {
2403 		IWL_DEBUG_ISR(trans,
2404 			      "ISR[%d] inta_hw 0x%08x, enabled (sw) 0x%08x (hw) 0x%08x\n",
2405 			      entry->entry, inta_hw, trans_pcie->hw_mask,
2406 			      iwl_read32(trans, CSR_MSIX_HW_INT_MASK_AD));
2407 		if (inta_hw & ~trans_pcie->hw_mask)
2408 			IWL_DEBUG_ISR(trans,
2409 				      "We got a masked interrupt 0x%08x\n",
2410 				      inta_hw & ~trans_pcie->hw_mask);
2411 	}
2412 
2413 	inta_hw &= trans_pcie->hw_mask;
2414 
2415 	/* Alive notification via Rx interrupt will do the real work */
2416 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_ALIVE) {
2417 		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
2418 		isr_stats->alive++;
2419 		if (trans->mac_cfg->gen2) {
2420 			/* We can restock, since firmware configured the RFH */
2421 			iwl_pcie_rxmq_restock(trans, trans_pcie->rxq);
2422 		}
2423 	}
2424 
2425 	/*
2426 	 * In some rare cases when the HW is in a bad state, we may
2427 	 * get this interrupt too early, when prph_info is still NULL.
2428 	 * So make sure that it's not NULL to prevent crashing.
2429 	 */
2430 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_WAKEUP && trans_pcie->prph_info) {
2431 		u32 sleep_notif =
2432 			le32_to_cpu(trans_pcie->prph_info->sleep_notif);
2433 
2434 		if (sleep_notif == IWL_D3_SLEEP_STATUS_SUSPEND ||
2435 		    sleep_notif == IWL_D3_SLEEP_STATUS_RESUME) {
2436 			IWL_DEBUG_ISR(trans,
2437 				      "Sx interrupt: sleep notification = 0x%x\n",
2438 				      sleep_notif);
2439 			if (trans_pcie->sx_state == IWL_SX_WAITING) {
2440 				trans_pcie->sx_state = IWL_SX_COMPLETE;
2441 				wake_up(&trans_pcie->sx_waitq);
2442 			} else {
2443 				IWL_ERR(trans,
2444 					"unexpected Sx interrupt (0x%x)\n",
2445 					sleep_notif);
2446 			}
2447 		} else {
2448 			/* uCode wakes up after power-down sleep */
2449 			IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
2450 			iwl_pcie_rxq_check_wrptr(trans);
2451 			iwl_pcie_txq_check_wrptrs(trans);
2452 
2453 			isr_stats->wakeup++;
2454 		}
2455 	}
2456 
2457 	/* Chip got too hot and stopped itself */
2458 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_CT_KILL) {
2459 		IWL_ERR(trans, "Microcode CT kill error detected.\n");
2460 		isr_stats->ctkill++;
2461 	}
2462 
2463 	/* HW RF KILL switch toggled */
2464 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RF_KILL)
2465 		iwl_pcie_handle_rfkill_irq(trans, true);
2466 
2467 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_HW_ERR) {
2468 		IWL_ERR(trans,
2469 			"Hardware error detected. Restarting.\n");
2470 
2471 		isr_stats->hw++;
2472 		trans->dbg.hw_error = true;
2473 		iwl_pcie_irq_handle_error(trans);
2474 	}
2475 
2476 	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RESET_DONE)
2477 		iwl_trans_pcie_handle_reset_interrupt(trans);
2478 
2479 	if (!polling)
2480 		iwl_pcie_clear_irq(trans, entry->entry);
2481 
2482 	lock_map_release(&trans->sync_cmd_lockdep_map);
2483 
2484 	return IRQ_HANDLED;
2485 }
2486