rx.c 57.1 KB
Newer Older
1 2
/******************************************************************************
 *
3
 * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
4
 * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
5
 * Copyright(c) 2016 Intel Deutschland GmbH
6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
 *
 * Portions of this file are derived from the ipw3945 project, as well
 * as portions of the ieee80211 subsystem header files.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of version 2 of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 *
 * You should have received a copy of the GNU General Public License along with
 * this program; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA
 *
 * The full GNU General Public License is included in this distribution in the
 * file called LICENSE.
 *
 * Contact Information:
27
 *  Intel Linux Wireless <linuxwifi@intel.com>
28 29 30 31 32
 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
 *
 *****************************************************************************/
#include <linux/sched.h>
#include <linux/wait.h>
33
#include <linux/gfp.h>
34

35
#include "iwl-prph.h"
36
#include "iwl-io.h"
37
#include "internal.h"
38
#include "iwl-op-mode.h"
39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77

/******************************************************************************
 *
 * RX path functions
 *
 ******************************************************************************/

/*
 * Rx theory of operation
 *
 * Driver allocates a circular buffer of Receive Buffer Descriptors (RBDs),
 * each of which point to Receive Buffers to be filled by the NIC.  These get
 * used not only for Rx frames, but for any command response or notification
 * from the NIC.  The driver and NIC manage the Rx buffers by means
 * of indexes into the circular buffer.
 *
 * Rx Queue Indexes
 * The host/firmware share two index registers for managing the Rx buffers.
 *
 * The READ index maps to the first position that the firmware may be writing
 * to -- the driver can read up to (but not including) this position and get
 * good data.
 * The READ index is managed by the firmware once the card is enabled.
 *
 * The WRITE index maps to the last position the driver has read from -- the
 * position preceding WRITE is the last slot the firmware can place a packet.
 *
 * The queue is empty (no good data) if WRITE = READ - 1, and is full if
 * WRITE = READ.
 *
 * During initialization, the host sets up the READ queue position to the first
 * INDEX position, and WRITE to the last (READ - 1 wrapped)
 *
 * When the firmware places a packet in a buffer, it will advance the READ index
 * and fire the RX interrupt.  The driver can then query the READ index and
 * process as many packets as possible, moving the WRITE index forward as it
 * resets the Rx queue buffers with new memory.
 *
 * The management in the driver is as follows:
78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97
 * + A list of pre-allocated RBDs is stored in iwl->rxq->rx_free.
 *   When the interrupt handler is called, the request is processed.
 *   The page is either stolen - transferred to the upper layer
 *   or reused - added immediately to the iwl->rxq->rx_free list.
 * + When the page is stolen - the driver updates the matching queue's used
 *   count, detaches the RBD and transfers it to the queue used list.
 *   When there are two used RBDs - they are transferred to the allocator empty
 *   list. Work is then scheduled for the allocator to start allocating
 *   eight buffers.
 *   When there are another 6 used RBDs - they are transferred to the allocator
 *   empty list and the driver tries to claim the pre-allocated buffers and
 *   add them to iwl->rxq->rx_free. If it fails - it continues to claim them
 *   until ready.
 *   When there are 8+ buffers in the free list - either from allocation or from
 *   8 reused unstolen pages - restock is called to update the FW and indexes.
 * + In order to make sure the allocator always has RBDs to use for allocation
 *   the allocator has initial pool in the size of num_queues*(8-2) - the
 *   maximum missing RBDs per allocation request (request posted with 2
 *    empty RBDs, there is no guarantee when the other 6 RBDs are supplied).
 *   The queues supplies the recycle of the rest of the RBDs.
98 99
 * + A received packet is processed and handed to the kernel network stack,
 *   detached from the iwl->rxq.  The driver 'processed' index is updated.
100
 * + If there are no allocated buffers in iwl->rxq->rx_free,
101 102
 *   the READ INDEX is not incremented and iwl->status(RX_STALLED) is set.
 *   If there were enough free buffers and RX_STALLED is set it is cleared.
103 104 105 106
 *
 *
 * Driver sequence:
 *
107 108
 * iwl_rxq_alloc()            Allocates rx_free
 * iwl_pcie_rx_replenish()    Replenishes rx_free list from rx_used, and calls
109 110
 *                            iwl_pcie_rxq_restock.
 *                            Used only during initialization.
111
 * iwl_pcie_rxq_restock()     Moves available buffers from rx_free into Rx
112
 *                            queue, updates firmware pointers, and updates
113 114
 *                            the WRITE index.
 * iwl_pcie_rx_allocator()     Background work for allocating pages.
115 116
 *
 * -- enable interrupts --
117
 * ISR - iwl_rx()             Detach iwl_rx_mem_buffers from pool up to the
118 119
 *                            READ INDEX, detaching the SKB from the pool.
 *                            Moves the packet buffer from queue to rx_used.
120
 *                            Posts and claims requests to the allocator.
121
 *                            Calls iwl_pcie_rxq_restock to refill any empty
122
 *                            slots.
123 124 125 126 127 128 129 130 131 132 133 134
 *
 * RBD life-cycle:
 *
 * Init:
 * rxq.pool -> rxq.rx_used -> rxq.rx_free -> rxq.queue
 *
 * Regular Receive interrupt:
 * Page Stolen:
 * rxq.queue -> rxq.rx_used -> allocator.rbd_empty ->
 * allocator.rbd_allocated -> rxq.rx_free -> rxq.queue
 * Page not Stolen:
 * rxq.queue -> rxq.rx_free -> rxq.queue
135 136 137 138
 * ...
 *
 */

139 140
/*
 * iwl_rxq_space - Return number of free slots available in queue.
141
 */
142
static int iwl_rxq_space(const struct iwl_rxq *rxq)
143
{
144 145
	/* Make sure rx queue size is a power of 2 */
	WARN_ON(rxq->queue_size & (rxq->queue_size - 1));
146

147 148 149 150 151 152
	/*
	 * There can be up to (RX_QUEUE_SIZE - 1) free slots, to avoid ambiguity
	 * between empty and completely full queues.
	 * The following is equivalent to modulo by RX_QUEUE_SIZE and is well
	 * defined for negative dividends.
	 */
153
	return (rxq->read - rxq->write - 1) & (rxq->queue_size - 1);
154 155
}

156 157 158 159 160 161 162 163
/*
 * iwl_dma_addr2rbd_ptr - convert a DMA address to a uCode read buffer ptr
 */
static inline __le32 iwl_pcie_dma_addr2rbd_ptr(dma_addr_t dma_addr)
{
	return cpu_to_le32((u32)(dma_addr >> 8));
}

164 165 166 167 168 169
static void iwl_pcie_write_prph_64(struct iwl_trans *trans, u64 ofs, u64 val)
{
	iwl_write_prph(trans, ofs, val & 0xffffffff);
	iwl_write_prph(trans, ofs + 4, val >> 32);
}

170 171 172
/*
 * iwl_pcie_rx_stop - stops the Rx DMA
 */
173 174 175 176 177 178 179
int iwl_pcie_rx_stop(struct iwl_trans *trans)
{
	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
	return iwl_poll_direct_bit(trans, FH_MEM_RSSR_RX_STATUS_REG,
				   FH_RSSR_CHNL0_RX_STATUS_CHNL_IDLE, 1000);
}

180 181
/*
 * iwl_pcie_rxq_inc_wr_ptr - Update the write pointer for the RX queue
182
 */
183 184
static void iwl_pcie_rxq_inc_wr_ptr(struct iwl_trans *trans,
				    struct iwl_rxq *rxq)
185 186 187
{
	u32 reg;

188
	lockdep_assert_held(&rxq->lock);
189

190 191 192 193 194 195 196 197 198 199 200 201 202 203
	/*
	 * explicitly wake up the NIC if:
	 * 1. shadow registers aren't enabled
	 * 2. there is a chance that the NIC is asleep
	 */
	if (!trans->cfg->base_params->shadow_reg_enable &&
	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);

		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
			IWL_DEBUG_INFO(trans, "Rx queue requesting wakeup, GP1 = 0x%x\n",
				       reg);
			iwl_set_bit(trans, CSR_GP_CNTRL,
				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
204 205
			rxq->need_update = true;
			return;
206 207
		}
	}
208 209

	rxq->write_actual = round_down(rxq->write, 8);
210 211 212
	if (trans->cfg->mq_rx_supported)
		iwl_write_prph(trans, RFH_Q_FRBDCB_WIDX(rxq->id),
			       rxq->write_actual);
213 214 215 216 217 218
	/*
	 * write to FH_RSCSR_CHNL0_WPTR register even in MQ as a W/A to
	 * hardware shadow registers bug - writing to RFH_Q_FRBDCB_WIDX will
	 * not wake the NIC.
	 */
	iwl_write32(trans, FH_RSCSR_CHNL0_WPTR, rxq->write_actual);
219 220 221 222 223
}

static void iwl_pcie_rxq_check_wrptr(struct iwl_trans *trans)
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
224
	int i;
225

226 227
	for (i = 0; i < trans->num_rx_queues; i++) {
		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
228

229 230 231 232 233 234 235
		if (!rxq->need_update)
			continue;
		spin_lock(&rxq->lock);
		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
		rxq->need_update = false;
		spin_unlock(&rxq->lock);
	}
236 237
}

238 239 240
/*
 * iwl_pcie_rxq_mq_restock - restock implementation for multi-queue rx
 */
241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285
static void iwl_pcie_rxq_mq_restock(struct iwl_trans *trans,
				    struct iwl_rxq *rxq)
{
	struct iwl_rx_mem_buffer *rxb;

	/*
	 * If the device isn't enabled - no need to try to add buffers...
	 * This can happen when we stop the device and still have an interrupt
	 * pending. We stop the APM before we sync the interrupts because we
	 * have to (see comment there). On the other hand, since the APM is
	 * stopped, we cannot access the HW (in particular not prph).
	 * So don't try to restock if the APM has been already stopped.
	 */
	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
		return;

	spin_lock(&rxq->lock);
	while (rxq->free_count) {
		__le64 *bd = (__le64 *)rxq->bd;

		/* Get next free Rx buffer, remove from free list */
		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
				       list);
		list_del(&rxb->list);

		/* 12 first bits are expected to be empty */
		WARN_ON(rxb->page_dma & DMA_BIT_MASK(12));
		/* Point to Rx buffer via next RBD in circular buffer */
		bd[rxq->write] = cpu_to_le64(rxb->page_dma | rxb->vid);
		rxq->write = (rxq->write + 1) & MQ_RX_TABLE_MASK;
		rxq->free_count--;
	}
	spin_unlock(&rxq->lock);

	/*
	 * If we've added more space for the firmware to place data, tell it.
	 * Increment device's write pointer in multiples of 8.
	 */
	if (rxq->write_actual != (rxq->write & ~0x7)) {
		spin_lock(&rxq->lock);
		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
		spin_unlock(&rxq->lock);
	}
}

286
/*
287
 * iwl_pcie_rxq_sq_restock - restock implementation for single queue rx
288
 */
289 290
static void iwl_pcie_rxq_sq_restock(struct iwl_trans *trans,
				    struct iwl_rxq *rxq)
291 292 293
{
	struct iwl_rx_mem_buffer *rxb;

294 295 296
	/*
	 * If the device isn't enabled - not need to try to add buffers...
	 * This can happen when we stop the device and still have an interrupt
297 298 299
	 * pending. We stop the APM before we sync the interrupts because we
	 * have to (see comment there). On the other hand, since the APM is
	 * stopped, we cannot access the HW (in particular not prph).
300 301
	 * So don't try to restock if the APM has been already stopped.
	 */
302
	if (!test_bit(STATUS_DEVICE_ENABLED, &trans->status))
303 304
		return;

305
	spin_lock(&rxq->lock);
306
	while ((iwl_rxq_space(rxq) > 0) && (rxq->free_count)) {
307
		__le32 *bd = (__le32 *)rxq->bd;
308 309 310 311 312
		/* The overwritten rxb must be a used one */
		rxb = rxq->queue[rxq->write];
		BUG_ON(rxb && rxb->page);

		/* Get next free Rx buffer, remove from free list */
J
Johannes Berg 已提交
313 314 315
		rxb = list_first_entry(&rxq->rx_free, struct iwl_rx_mem_buffer,
				       list);
		list_del(&rxb->list);
316 317

		/* Point to Rx buffer via next RBD in circular buffer */
318
		bd[rxq->write] = iwl_pcie_dma_addr2rbd_ptr(rxb->page_dma);
319 320 321 322
		rxq->queue[rxq->write] = rxb;
		rxq->write = (rxq->write + 1) & RX_QUEUE_MASK;
		rxq->free_count--;
	}
323
	spin_unlock(&rxq->lock);
324 325 326 327

	/* If we've added more space for the firmware to place data, tell it.
	 * Increment device's write pointer in multiples of 8. */
	if (rxq->write_actual != (rxq->write & ~0x7)) {
328
		spin_lock(&rxq->lock);
329
		iwl_pcie_rxq_inc_wr_ptr(trans, rxq);
330
		spin_unlock(&rxq->lock);
331 332 333
	}
}

334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
/*
 * iwl_pcie_rxq_restock - refill RX queue from pre-allocated pool
 *
 * If there are slots in the RX queue that need to be restocked,
 * and we have free pre-allocated buffers, fill the ranks as much
 * as we can, pulling from rx_free.
 *
 * This moves the 'write' index forward to catch up with 'processed', and
 * also updates the memory address in the firmware to reference the new
 * target buffer.
 */
static
void iwl_pcie_rxq_restock(struct iwl_trans *trans, struct iwl_rxq *rxq)
{
	if (trans->cfg->mq_rx_supported)
		iwl_pcie_rxq_mq_restock(trans, rxq);
	else
		iwl_pcie_rxq_sq_restock(trans, rxq);
}

354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373
/*
 * iwl_pcie_rx_alloc_page - allocates and returns a page.
 *
 */
static struct page *iwl_pcie_rx_alloc_page(struct iwl_trans *trans,
					   gfp_t priority)
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
	struct page *page;
	gfp_t gfp_mask = priority;

	if (trans_pcie->rx_page_order > 0)
		gfp_mask |= __GFP_COMP;

	/* Alloc a new receive buffer */
	page = alloc_pages(gfp_mask, trans_pcie->rx_page_order);
	if (!page) {
		if (net_ratelimit())
			IWL_DEBUG_INFO(trans, "alloc_pages failed, order: %d\n",
				       trans_pcie->rx_page_order);
374 375 376
		/*
		 * Issue an error if we don't have enough pre-allocated
		  * buffers.
377
`		 */
378
		if (!(gfp_mask & __GFP_NOWARN) && net_ratelimit())
379
			IWL_CRIT(trans,
380
				 "Failed to alloc_pages\n");
381 382 383 384 385
		return NULL;
	}
	return page;
}

386
/*
387
 * iwl_pcie_rxq_alloc_rbs - allocate a page for each used RBD
388
 *
389 390 391
 * A used RBD is an Rx buffer that has been given to the stack. To use it again
 * a page must be allocated and the RBD must point to the page. This function
 * doesn't change the HW pointer but handles the list of pages that is used by
392
 * iwl_pcie_rxq_restock. The latter function will update the HW to use the newly
393
 * allocated buffers.
394
 */
395 396
static void iwl_pcie_rxq_alloc_rbs(struct iwl_trans *trans, gfp_t priority,
				   struct iwl_rxq *rxq)
397
{
398
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
399 400 401 402
	struct iwl_rx_mem_buffer *rxb;
	struct page *page;

	while (1) {
403
		spin_lock(&rxq->lock);
404
		if (list_empty(&rxq->rx_used)) {
405
			spin_unlock(&rxq->lock);
406 407
			return;
		}
408
		spin_unlock(&rxq->lock);
409 410

		/* Alloc a new receive buffer */
411 412
		page = iwl_pcie_rx_alloc_page(trans, priority);
		if (!page)
413 414
			return;

415
		spin_lock(&rxq->lock);
416 417

		if (list_empty(&rxq->rx_used)) {
418
			spin_unlock(&rxq->lock);
419
			__free_pages(page, trans_pcie->rx_page_order);
420 421
			return;
		}
J
Johannes Berg 已提交
422 423 424
		rxb = list_first_entry(&rxq->rx_used, struct iwl_rx_mem_buffer,
				       list);
		list_del(&rxb->list);
425
		spin_unlock(&rxq->lock);
426 427 428 429

		BUG_ON(rxb->page);
		rxb->page = page;
		/* Get physical address of the RB */
430 431 432 433
		rxb->page_dma =
			dma_map_page(trans->dev, page, 0,
				     PAGE_SIZE << trans_pcie->rx_page_order,
				     DMA_FROM_DEVICE);
434 435
		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
			rxb->page = NULL;
436
			spin_lock(&rxq->lock);
437
			list_add(&rxb->list, &rxq->rx_used);
438
			spin_unlock(&rxq->lock);
439 440 441
			__free_pages(page, trans_pcie->rx_page_order);
			return;
		}
442

443
		spin_lock(&rxq->lock);
444 445 446 447

		list_add_tail(&rxb->list, &rxq->rx_free);
		rxq->free_count++;

448
		spin_unlock(&rxq->lock);
449 450 451
	}
}

452
static void iwl_pcie_free_rbs_pool(struct iwl_trans *trans)
453 454 455 456
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
	int i;

457
	for (i = 0; i < RX_POOL_SIZE; i++) {
458
		if (!trans_pcie->rx_pool[i].page)
459
			continue;
460
		dma_unmap_page(trans->dev, trans_pcie->rx_pool[i].page_dma,
461 462
			       PAGE_SIZE << trans_pcie->rx_page_order,
			       DMA_FROM_DEVICE);
463 464 465
		__free_pages(trans_pcie->rx_pool[i].page,
			     trans_pcie->rx_page_order);
		trans_pcie->rx_pool[i].page = NULL;
466 467 468
	}
}

469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492
/*
 * iwl_pcie_rx_allocator - Allocates pages in the background for RX queues
 *
 * Allocates for each received request 8 pages
 * Called as a scheduled work item.
 */
static void iwl_pcie_rx_allocator(struct iwl_trans *trans)
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
	struct iwl_rb_allocator *rba = &trans_pcie->rba;
	struct list_head local_empty;
	int pending = atomic_xchg(&rba->req_pending, 0);

	IWL_DEBUG_RX(trans, "Pending allocation requests = %d\n", pending);

	/* If we were scheduled - there is at least one request */
	spin_lock(&rba->lock);
	/* swap out the rba->rbd_empty to a local list */
	list_replace_init(&rba->rbd_empty, &local_empty);
	spin_unlock(&rba->lock);

	while (pending) {
		int i;
		struct list_head local_allocated;
493 494 495 496 497
		gfp_t gfp_mask = GFP_KERNEL;

		/* Do not post a warning if there are only a few requests */
		if (pending < RX_PENDING_WATERMARK)
			gfp_mask |= __GFP_NOWARN;
498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516

		INIT_LIST_HEAD(&local_allocated);

		for (i = 0; i < RX_CLAIM_REQ_ALLOC;) {
			struct iwl_rx_mem_buffer *rxb;
			struct page *page;

			/* List should never be empty - each reused RBD is
			 * returned to the list, and initial pool covers any
			 * possible gap between the time the page is allocated
			 * to the time the RBD is added.
			 */
			BUG_ON(list_empty(&local_empty));
			/* Get the first rxb from the rbd list */
			rxb = list_first_entry(&local_empty,
					       struct iwl_rx_mem_buffer, list);
			BUG_ON(rxb->page);

			/* Alloc a new receive buffer */
517
			page = iwl_pcie_rx_alloc_page(trans, gfp_mask);
518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561
			if (!page)
				continue;
			rxb->page = page;

			/* Get physical address of the RB */
			rxb->page_dma = dma_map_page(trans->dev, page, 0,
					PAGE_SIZE << trans_pcie->rx_page_order,
					DMA_FROM_DEVICE);
			if (dma_mapping_error(trans->dev, rxb->page_dma)) {
				rxb->page = NULL;
				__free_pages(page, trans_pcie->rx_page_order);
				continue;
			}

			/* move the allocated entry to the out list */
			list_move(&rxb->list, &local_allocated);
			i++;
		}

		pending--;
		if (!pending) {
			pending = atomic_xchg(&rba->req_pending, 0);
			IWL_DEBUG_RX(trans,
				     "Pending allocation requests = %d\n",
				     pending);
		}

		spin_lock(&rba->lock);
		/* add the allocated rbds to the allocator allocated list */
		list_splice_tail(&local_allocated, &rba->rbd_allocated);
		/* get more empty RBDs for current pending requests */
		list_splice_tail_init(&rba->rbd_empty, &local_empty);
		spin_unlock(&rba->lock);

		atomic_inc(&rba->req_ready);
	}

	spin_lock(&rba->lock);
	/* return unused rbds to the allocator empty list */
	list_splice_tail(&local_empty, &rba->rbd_empty);
	spin_unlock(&rba->lock);
}

/*
562
 * iwl_pcie_rx_allocator_get - returns the pre-allocated pages
563 564 565
.*
.* Called by queue when the queue posted allocation request and
 * has freed 8 RBDs in order to restock itself.
566 567
 * This function directly moves the allocated RBs to the queue's ownership
 * and updates the relevant counters.
568
 */
569 570
static void iwl_pcie_rx_allocator_get(struct iwl_trans *trans,
				      struct iwl_rxq *rxq)
571 572 573 574 575
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
	struct iwl_rb_allocator *rba = &trans_pcie->rba;
	int i;

576 577
	lockdep_assert_held(&rxq->lock);

578 579 580
	/*
	 * atomic_dec_if_positive returns req_ready - 1 for any scenario.
	 * If req_ready is 0 atomic_dec_if_positive will return -1 and this
581
	 * function will return early, as there are no ready requests.
582 583 584 585 586
	 * atomic_dec_if_positive will perofrm the *actual* decrement only if
	 * req_ready > 0, i.e. - there are ready requests and the function
	 * hands one request to the caller.
	 */
	if (atomic_dec_if_positive(&rba->req_ready) < 0)
587
		return;
588 589 590 591

	spin_lock(&rba->lock);
	for (i = 0; i < RX_CLAIM_REQ_ALLOC; i++) {
		/* Get next free Rx buffer, remove it from free list */
592 593 594 595 596
		struct iwl_rx_mem_buffer *rxb =
			list_first_entry(&rba->rbd_allocated,
					 struct iwl_rx_mem_buffer, list);

		list_move(&rxb->list, &rxq->rx_free);
597 598 599
	}
	spin_unlock(&rba->lock);

600 601
	rxq->used_count -= RX_CLAIM_REQ_ALLOC;
	rxq->free_count += RX_CLAIM_REQ_ALLOC;
602 603 604
}

static void iwl_pcie_rx_allocator_work(struct work_struct *data)
605
{
606 607
	struct iwl_rb_allocator *rba_p =
		container_of(data, struct iwl_rb_allocator, rx_alloc);
608
	struct iwl_trans_pcie *trans_pcie =
609
		container_of(rba_p, struct iwl_trans_pcie, rba);
610

611
	iwl_pcie_rx_allocator(trans_pcie->trans);
612 613
}

614 615 616
static int iwl_pcie_rx_alloc(struct iwl_trans *trans)
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
617
	struct iwl_rb_allocator *rba = &trans_pcie->rba;
618
	struct device *dev = trans->dev;
619
	int i;
620 621
	int free_size = trans->cfg->mq_rx_supported ? sizeof(__le64) :
						      sizeof(__le32);
622

623 624 625 626 627 628 629
	if (WARN_ON(trans_pcie->rxq))
		return -EINVAL;

	trans_pcie->rxq = kcalloc(trans->num_rx_queues, sizeof(struct iwl_rxq),
				  GFP_KERNEL);
	if (!trans_pcie->rxq)
		return -EINVAL;
630

631
	spin_lock_init(&rba->lock);
632

633 634
	for (i = 0; i < trans->num_rx_queues; i++) {
		struct iwl_rxq *rxq = &trans_pcie->rxq[i];
635

636
		spin_lock_init(&rxq->lock);
637 638 639 640 641
		if (trans->cfg->mq_rx_supported)
			rxq->queue_size = MQ_RX_TABLE_SIZE;
		else
			rxq->queue_size = RX_QUEUE_SIZE;

642 643 644 645 646
		/*
		 * Allocate the circular buffer of Read Buffer Descriptors
		 * (RBDs)
		 */
		rxq->bd = dma_zalloc_coherent(dev,
647 648
					     free_size * rxq->queue_size,
					     &rxq->bd_dma, GFP_KERNEL);
649 650
		if (!rxq->bd)
			goto err;
651

652 653 654 655 656 657 658 659 660
		if (trans->cfg->mq_rx_supported) {
			rxq->used_bd = dma_zalloc_coherent(dev,
							   sizeof(__le32) *
							   rxq->queue_size,
							   &rxq->used_bd_dma,
							   GFP_KERNEL);
			if (!rxq->used_bd)
				goto err;
		}
661

662 663 664 665 666 667 668
		/*Allocate the driver's pointer to receive buffer status */
		rxq->rb_stts = dma_zalloc_coherent(dev, sizeof(*rxq->rb_stts),
						   &rxq->rb_stts_dma,
						   GFP_KERNEL);
		if (!rxq->rb_stts)
			goto err;
	}
669 670
	return 0;

671 672 673 674 675
err:
	for (i = 0; i < trans->num_rx_queues; i++) {
		struct iwl_rxq *rxq = &trans_pcie->rxq[i];

		if (rxq->bd)
676
			dma_free_coherent(dev, free_size * rxq->queue_size,
677 678 679 680 681 682 683 684
					  rxq->bd, rxq->bd_dma);
		rxq->bd_dma = 0;
		rxq->bd = NULL;

		if (rxq->rb_stts)
			dma_free_coherent(trans->dev,
					  sizeof(struct iwl_rb_status),
					  rxq->rb_stts, rxq->rb_stts_dma);
685 686 687 688 689 690

		if (rxq->used_bd)
			dma_free_coherent(dev, sizeof(__le32) * rxq->queue_size,
					  rxq->used_bd, rxq->used_bd_dma);
		rxq->used_bd_dma = 0;
		rxq->used_bd = NULL;
691 692
	}
	kfree(trans_pcie->rxq);
693

694
	return -ENOMEM;
695 696
}

697 698 699 700 701 702
static void iwl_pcie_rx_hw_init(struct iwl_trans *trans, struct iwl_rxq *rxq)
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
	u32 rb_size;
	const u32 rfdnlog = RX_QUEUE_SIZE_LOG; /* 256 RBDs */

703 704 705 706 707
	switch (trans_pcie->rx_buf_size) {
	case IWL_AMSDU_4K:
		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
		break;
	case IWL_AMSDU_8K:
708
		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_8K;
709 710 711 712 713 714
		break;
	case IWL_AMSDU_12K:
		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_12K;
		break;
	default:
		WARN_ON(1);
715
		rb_size = FH_RCSR_RX_CONFIG_REG_VAL_RB_SIZE_4K;
716
	}
717 718 719

	/* Stop Rx DMA */
	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG, 0);
720 721 722 723
	/* reset and flush pointers */
	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_RBDCB_WPTR, 0);
	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_FLUSH_RB_REQ, 0);
	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RDPTR, 0);
724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739

	/* Reset driver's Rx queue write index */
	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RBDCB_WPTR_REG, 0);

	/* Tell device where to find RBD circular buffer in DRAM */
	iwl_write_direct32(trans, FH_RSCSR_CHNL0_RBDCB_BASE_REG,
			   (u32)(rxq->bd_dma >> 8));

	/* Tell device where in DRAM to update its Rx status */
	iwl_write_direct32(trans, FH_RSCSR_CHNL0_STTS_WPTR_REG,
			   rxq->rb_stts_dma >> 4);

	/* Enable Rx DMA
	 * FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY is set because of HW bug in
	 *      the credit mechanism in 5000 HW RX FIFO
	 * Direct rx interrupts to hosts
740
	 * Rx buffer size 4 or 8k or 12k
741 742 743 744 745 746 747 748
	 * RB timeout 0x10
	 * 256 RBDs
	 */
	iwl_write_direct32(trans, FH_MEM_RCSR_CHNL0_CONFIG_REG,
			   FH_RCSR_RX_CONFIG_CHNL_EN_ENABLE_VAL |
			   FH_RCSR_CHNL0_RX_IGNORE_RXF_EMPTY |
			   FH_RCSR_CHNL0_RX_CONFIG_IRQ_DEST_INT_HOST_VAL |
			   rb_size|
749
			   (RX_RB_TIMEOUT << FH_RCSR_RX_CONFIG_REG_IRQ_RBTH_POS)|
750 751 752 753
			   (rfdnlog << FH_RCSR_RX_CONFIG_RBDCB_SIZE_POS));

	/* Set interrupt coalescing timer to default (2048 usecs) */
	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
754 755 756 757

	/* W/A for interrupt coalescing bug in 7260 and 3160 */
	if (trans->cfg->host_interrupt_operation_mode)
		iwl_set_bit(trans, CSR_INT_COALESCING, IWL_HOST_INT_OPER_MODE);
758 759
}

760
static void iwl_pcie_rx_mq_hw_init(struct iwl_trans *trans)
761
{
762 763 764
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
	u32 rb_size, enabled = 0;
	int i;
765

766 767 768 769 770 771 772 773 774 775 776 777 778 779
	switch (trans_pcie->rx_buf_size) {
	case IWL_AMSDU_4K:
		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
		break;
	case IWL_AMSDU_8K:
		rb_size = RFH_RXF_DMA_RB_SIZE_8K;
		break;
	case IWL_AMSDU_12K:
		rb_size = RFH_RXF_DMA_RB_SIZE_12K;
		break;
	default:
		WARN_ON(1);
		rb_size = RFH_RXF_DMA_RB_SIZE_4K;
	}
780

781 782 783 784
	/* Stop Rx DMA */
	iwl_write_prph(trans, RFH_RXF_DMA_CFG, 0);
	/* disable free amd used rx queue operation */
	iwl_write_prph(trans, RFH_RXF_RXQ_ACTIVE, 0);
785

786 787 788
	for (i = 0; i < trans->num_rx_queues; i++) {
		/* Tell device where to find RBD free table in DRAM */
		iwl_pcie_write_prph_64(trans, RFH_Q_FRBDCB_BA_LSB(i),
789
				       (u64)(trans_pcie->rxq[i].bd_dma));
790 791
		/* Tell device where to find RBD used table in DRAM */
		iwl_pcie_write_prph_64(trans, RFH_Q_URBDCB_BA_LSB(i),
792
				       (u64)(trans_pcie->rxq[i].used_bd_dma));
793 794
		/* Tell device where in DRAM to update its Rx status */
		iwl_pcie_write_prph_64(trans, RFH_Q_URBD_STTS_WPTR_LSB(i),
795
				       trans_pcie->rxq[i].rb_stts_dma);
796 797 798 799 800 801 802
		/* Reset device indice tables */
		iwl_write_prph(trans, RFH_Q_FRBDCB_WIDX(i), 0);
		iwl_write_prph(trans, RFH_Q_FRBDCB_RIDX(i), 0);
		iwl_write_prph(trans, RFH_Q_URBDCB_WIDX(i), 0);

		enabled |= BIT(i) | BIT(i + 16);
	}
803

804 805 806 807 808 809 810 811
	/* restock default queue */
	iwl_pcie_rxq_mq_restock(trans, &trans_pcie->rxq[0]);

	/*
	 * Enable Rx DMA
	 * Single frame mode
	 * Rx buffer size 4 or 8k or 12k
	 * Min RB size 4 or 8
812
	 * Drop frames that exceed RB size
813 814 815 816 817 818
	 * 512 RBDs
	 */
	iwl_write_prph(trans, RFH_RXF_DMA_CFG,
		       RFH_DMA_EN_ENABLE_VAL |
		       rb_size | RFH_RXF_DMA_SINGLE_FRAME_MASK |
		       RFH_RXF_DMA_MIN_RB_4_8 |
819
		       RFH_RXF_DMA_DROP_TOO_LARGE_MASK |
820 821
		       RFH_RXF_DMA_RBDCB_SIZE_512);

822 823
	/*
	 * Activate DMA snooping.
824
	 * Set RX DMA chunk size to 64B
825 826
	 * Default queue is 0
	 */
827
	iwl_write_prph(trans, RFH_GEN_CFG, RFH_GEN_CFG_RFH_DMA_SNOOP |
828 829 830
		       (DEFAULT_RXQ_NUM << RFH_GEN_CFG_DEFAULT_RXQ_NUM_POS) |
		       RFH_GEN_CFG_SERVICE_DMA_SNOOP);
	/* Enable the relevant rx queues */
831
	iwl_write_prph(trans, RFH_RXF_RXQ_ACTIVE, enabled);
832

833 834
	/* Set interrupt coalescing timer to default (2048 usecs) */
	iwl_write8(trans, CSR_INT_COALESCING, IWL_HOST_INT_TIMEOUT_DEF);
835 836
}

837
static void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq)
838
{
839
	lockdep_assert_held(&rxq->lock);
840

841 842 843 844
	INIT_LIST_HEAD(&rxq->rx_free);
	INIT_LIST_HEAD(&rxq->rx_used);
	rxq->free_count = 0;
	rxq->used_count = 0;
845 846
}

847 848 849 850 851 852
static int iwl_pcie_dummy_napi_poll(struct napi_struct *napi, int budget)
{
	WARN_ON(1);
	return 0;
}

853 854 855
int iwl_pcie_rx_init(struct iwl_trans *trans)
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
856
	struct iwl_rxq *def_rxq;
857
	struct iwl_rb_allocator *rba = &trans_pcie->rba;
858
	int i, err, queue_size, allocator_pool_size, num_alloc;
859

860
	if (!trans_pcie->rxq) {
861 862 863 864
		err = iwl_pcie_rx_alloc(trans);
		if (err)
			return err;
	}
865
	def_rxq = trans_pcie->rxq;
866 867 868 869 870 871 872 873
	if (!rba->alloc_wq)
		rba->alloc_wq = alloc_workqueue("rb_allocator",
						WQ_HIGHPRI | WQ_UNBOUND, 1);
	INIT_WORK(&rba->rx_alloc, iwl_pcie_rx_allocator_work);

	spin_lock(&rba->lock);
	atomic_set(&rba->req_pending, 0);
	atomic_set(&rba->req_ready, 0);
874 875
	INIT_LIST_HEAD(&rba->rbd_allocated);
	INIT_LIST_HEAD(&rba->rbd_empty);
876
	spin_unlock(&rba->lock);
877

878
	/* free all first - we might be reconfigured for a different size */
879
	iwl_pcie_free_rbs_pool(trans);
880 881

	for (i = 0; i < RX_QUEUE_SIZE; i++)
882
		def_rxq->queue[i] = NULL;
883

884 885 886
	for (i = 0; i < trans->num_rx_queues; i++) {
		struct iwl_rxq *rxq = &trans_pcie->rxq[i];

887 888
		rxq->id = i;

889 890 891 892 893 894 895 896 897 898
		spin_lock(&rxq->lock);
		/*
		 * Set read write pointer to reflect that we have processed
		 * and used all buffers, but have not restocked the Rx queue
		 * with fresh buffers
		 */
		rxq->read = 0;
		rxq->write = 0;
		rxq->write_actual = 0;
		memset(rxq->rb_stts, 0, sizeof(*rxq->rb_stts));
899

900 901
		iwl_pcie_rx_init_rxb_lists(rxq);

902 903 904 905
		if (!rxq->napi.poll)
			netif_napi_add(&trans_pcie->napi_dev, &rxq->napi,
				       iwl_pcie_dummy_napi_poll, 64);

906 907
		spin_unlock(&rxq->lock);
	}
908

909
	/* move the pool to the default queue and allocator ownerships */
910 911
	queue_size = trans->cfg->mq_rx_supported ?
		     MQ_RX_NUM_RBDS : RX_QUEUE_SIZE;
912 913
	allocator_pool_size = trans->num_rx_queues *
		(RX_CLAIM_REQ_ALLOC - RX_POST_REQ_ALLOC);
914
	num_alloc = queue_size + allocator_pool_size;
915 916
	BUILD_BUG_ON(ARRAY_SIZE(trans_pcie->global_table) !=
		     ARRAY_SIZE(trans_pcie->rx_pool));
917
	for (i = 0; i < num_alloc; i++) {
918 919 920 921 922 923 924 925 926
		struct iwl_rx_mem_buffer *rxb = &trans_pcie->rx_pool[i];

		if (i < allocator_pool_size)
			list_add(&rxb->list, &rba->rbd_empty);
		else
			list_add(&rxb->list, &def_rxq->rx_used);
		trans_pcie->global_table[i] = rxb;
		rxb->vid = (u16)i;
	}
927

928
	iwl_pcie_rxq_alloc_rbs(trans, GFP_KERNEL, def_rxq);
929
	if (trans->cfg->mq_rx_supported) {
930
		iwl_pcie_rx_mq_hw_init(trans);
931
	} else {
932
		iwl_pcie_rxq_sq_restock(trans, def_rxq);
933 934
		iwl_pcie_rx_hw_init(trans, def_rxq);
	}
935 936 937 938

	spin_lock(&def_rxq->lock);
	iwl_pcie_rxq_inc_wr_ptr(trans, def_rxq);
	spin_unlock(&def_rxq->lock);
939 940 941 942 943 944 945

	return 0;
}

void iwl_pcie_rx_free(struct iwl_trans *trans)
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
946
	struct iwl_rb_allocator *rba = &trans_pcie->rba;
947 948
	int free_size = trans->cfg->mq_rx_supported ? sizeof(__le64) :
					      sizeof(__le32);
949
	int i;
950

951 952 953 954 955
	/*
	 * if rxq is NULL, it means that nothing has been allocated,
	 * exit now
	 */
	if (!trans_pcie->rxq) {
956 957 958 959
		IWL_DEBUG_INFO(trans, "Free NULL rx context\n");
		return;
	}

960 961 962 963 964 965
	cancel_work_sync(&rba->rx_alloc);
	if (rba->alloc_wq) {
		destroy_workqueue(rba->alloc_wq);
		rba->alloc_wq = NULL;
	}

966 967 968 969 970 971 972
	iwl_pcie_free_rbs_pool(trans);

	for (i = 0; i < trans->num_rx_queues; i++) {
		struct iwl_rxq *rxq = &trans_pcie->rxq[i];

		if (rxq->bd)
			dma_free_coherent(trans->dev,
973
					  free_size * rxq->queue_size,
974 975 976 977 978 979 980 981 982 983 984
					  rxq->bd, rxq->bd_dma);
		rxq->bd_dma = 0;
		rxq->bd = NULL;

		if (rxq->rb_stts)
			dma_free_coherent(trans->dev,
					  sizeof(struct iwl_rb_status),
					  rxq->rb_stts, rxq->rb_stts_dma);
		else
			IWL_DEBUG_INFO(trans,
				       "Free rxq->rb_stts which is NULL\n");
985

986 987 988 989 990 991
		if (rxq->used_bd)
			dma_free_coherent(trans->dev,
					  sizeof(__le32) * rxq->queue_size,
					  rxq->used_bd, rxq->used_bd_dma);
		rxq->used_bd_dma = 0;
		rxq->used_bd = NULL;
992 993 994

		if (rxq->napi.poll)
			netif_napi_del(&rxq->napi);
995
	}
996
	kfree(trans_pcie->rxq);
997 998
}

999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038
/*
 * iwl_pcie_rx_reuse_rbd - Recycle used RBDs
 *
 * Called when a RBD can be reused. The RBD is transferred to the allocator.
 * When there are 2 empty RBDs - a request for allocation is posted
 */
static void iwl_pcie_rx_reuse_rbd(struct iwl_trans *trans,
				  struct iwl_rx_mem_buffer *rxb,
				  struct iwl_rxq *rxq, bool emergency)
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
	struct iwl_rb_allocator *rba = &trans_pcie->rba;

	/* Move the RBD to the used list, will be moved to allocator in batches
	 * before claiming or posting a request*/
	list_add_tail(&rxb->list, &rxq->rx_used);

	if (unlikely(emergency))
		return;

	/* Count the allocator owned RBDs */
	rxq->used_count++;

	/* If we have RX_POST_REQ_ALLOC new released rx buffers -
	 * issue a request for allocator. Modulo RX_CLAIM_REQ_ALLOC is
	 * used for the case we failed to claim RX_CLAIM_REQ_ALLOC,
	 * after but we still need to post another request.
	 */
	if ((rxq->used_count % RX_CLAIM_REQ_ALLOC) == RX_POST_REQ_ALLOC) {
		/* Move the 2 RBDs to the allocator ownership.
		 Allocator has another 6 from pool for the request completion*/
		spin_lock(&rba->lock);
		list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
		spin_unlock(&rba->lock);

		atomic_inc(&rba->req_pending);
		queue_work(rba->alloc_wq, &rba->rx_alloc);
	}
}

1039
static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans,
1040
				struct iwl_rxq *rxq,
1041 1042
				struct iwl_rx_mem_buffer *rxb,
				bool emergency)
J
Johannes Berg 已提交
1043 1044
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1045
	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1046
	bool page_stolen = false;
1047
	int max_len = PAGE_SIZE << trans_pcie->rx_page_order;
1048
	u32 offset = 0;
J
Johannes Berg 已提交
1049 1050 1051 1052

	if (WARN_ON(!rxb))
		return;

1053 1054 1055 1056 1057 1058
	dma_unmap_page(trans->dev, rxb->page_dma, max_len, DMA_FROM_DEVICE);

	while (offset + sizeof(u32) + sizeof(struct iwl_cmd_header) < max_len) {
		struct iwl_rx_packet *pkt;
		u16 sequence;
		bool reclaim;
1059
		int index, cmd_index, len;
1060 1061
		struct iwl_rx_cmd_buffer rxcb = {
			._offset = offset,
1062
			._rx_page_order = trans_pcie->rx_page_order,
1063 1064
			._page = rxb->page,
			._page_stolen = false,
1065
			.truesize = max_len,
1066 1067 1068 1069 1070 1071 1072
		};

		pkt = rxb_addr(&rxcb);

		if (pkt->len_n_flags == cpu_to_le32(FH_RSCSR_FRAME_INVALID))
			break;

1073 1074 1075
		IWL_DEBUG_RX(trans,
			     "cmd at offset %d: %s (0x%.2x, seq 0x%x)\n",
			     rxcb._offset,
1076 1077 1078 1079
			     iwl_get_cmd_string(trans,
						iwl_cmd_id(pkt->hdr.cmd,
							   pkt->hdr.group_id,
							   0)),
1080
			     pkt->hdr.cmd, le16_to_cpu(pkt->hdr.sequence));
1081

1082
		len = iwl_rx_packet_len(pkt);
1083
		len += sizeof(u32); /* account for status word */
1084 1085
		trace_iwlwifi_dev_rx(trans->dev, trans, pkt, len);
		trace_iwlwifi_dev_rx_data(trans->dev, trans, pkt, len);
1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102

		/* Reclaim a command buffer only if this packet is a response
		 *   to a (driver-originated) command.
		 * If the packet (e.g. Rx frame) originated from uCode,
		 *   there is no command buffer to reclaim.
		 * Ucode should set SEQ_RX_FRAME bit if ucode-originated,
		 *   but apparently a few don't get set; catch them here. */
		reclaim = !(pkt->hdr.sequence & SEQ_RX_FRAME);
		if (reclaim) {
			int i;

			for (i = 0; i < trans_pcie->n_no_reclaim_cmds; i++) {
				if (trans_pcie->no_reclaim_cmds[i] ==
							pkt->hdr.cmd) {
					reclaim = false;
					break;
				}
1103 1104
			}
		}
J
Johannes Berg 已提交
1105

1106 1107 1108 1109
		sequence = le16_to_cpu(pkt->hdr.sequence);
		index = SEQ_TO_INDEX(sequence);
		cmd_index = get_cmd_index(&txq->q, index);

1110 1111 1112 1113 1114 1115
		if (rxq->id == 0)
			iwl_op_mode_rx(trans->op_mode, &rxq->napi,
				       &rxcb);
		else
			iwl_op_mode_rx_rss(trans->op_mode, &rxq->napi,
					   &rxcb, rxq->id);
1116

1117
		if (reclaim) {
1118
			kzfree(txq->entries[cmd_index].free_buf);
1119
			txq->entries[cmd_index].free_buf = NULL;
1120 1121
		}

1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132
		/*
		 * After here, we should always check rxcb._page_stolen,
		 * if it is true then one of the handlers took the page.
		 */

		if (reclaim) {
			/* Invoke any callbacks, transfer the buffer to caller,
			 * and fire off the (possibly) blocking
			 * iwl_trans_send_cmd()
			 * as we reclaim the driver command queue */
			if (!rxcb._page_stolen)
1133
				iwl_pcie_hcmd_complete(trans, &rxcb);
1134 1135 1136 1137 1138 1139
			else
				IWL_WARN(trans, "Claim null rxb?\n");
		}

		page_stolen |= rxcb._page_stolen;
		offset += ALIGN(len, FH_RSCSR_FRAME_ALIGN);
J
Johannes Berg 已提交
1140 1141
	}

1142 1143
	/* page was stolen from us -- free our reference */
	if (page_stolen) {
1144
		__free_pages(rxb->page, trans_pcie->rx_page_order);
J
Johannes Berg 已提交
1145
		rxb->page = NULL;
1146
	}
J
Johannes Berg 已提交
1147 1148 1149 1150 1151 1152 1153

	/* Reuse the page if possible. For notification packets and
	 * SKBs that fail to Rx correctly, add them back into the
	 * rx_free list for reuse later. */
	if (rxb->page != NULL) {
		rxb->page_dma =
			dma_map_page(trans->dev, rxb->page, 0,
1154 1155
				     PAGE_SIZE << trans_pcie->rx_page_order,
				     DMA_FROM_DEVICE);
1156 1157 1158 1159 1160 1161 1162 1163
		if (dma_mapping_error(trans->dev, rxb->page_dma)) {
			/*
			 * free the page(s) as well to not break
			 * the invariant that the items on the used
			 * list have no page(s)
			 */
			__free_pages(rxb->page, trans_pcie->rx_page_order);
			rxb->page = NULL;
1164
			iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
1165 1166 1167 1168
		} else {
			list_add_tail(&rxb->list, &rxq->rx_free);
			rxq->free_count++;
		}
J
Johannes Berg 已提交
1169
	} else
1170
		iwl_pcie_rx_reuse_rbd(trans, rxb, rxq, emergency);
J
Johannes Berg 已提交
1171 1172
}

1173 1174
/*
 * iwl_pcie_rx_handle - Main entry function for receiving responses from fw
1175
 */
1176
static void iwl_pcie_rx_handle(struct iwl_trans *trans, int queue)
1177
{
J
Johannes Berg 已提交
1178
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1179
	struct iwl_rxq *rxq = &trans_pcie->rxq[queue];
1180
	u32 r, i, count = 0;
1181
	bool emergency = false;
1182

1183 1184
restart:
	spin_lock(&rxq->lock);
1185 1186
	/* uCode's read index (stored in shared DRAM) indicates the last Rx
	 * buffer that the driver may process (last buffer filled by ucode). */
1187
	r = le16_to_cpu(ACCESS_ONCE(rxq->rb_stts->closed_rb_num)) & 0x0FFF;
1188 1189
	i = rxq->read;

1190 1191 1192
	/* W/A 9000 device step A0 wrap-around bug */
	r &= (rxq->queue_size - 1);

1193 1194
	/* Rx interrupt, but nothing sent from uCode */
	if (i == r)
1195
		IWL_DEBUG_RX(trans, "Q %d: HW = SW = %d\n", rxq->id, r);
1196 1197

	while (i != r) {
1198
		struct iwl_rx_mem_buffer *rxb;
1199

1200
		if (unlikely(rxq->used_count == rxq->queue_size / 2))
1201 1202
			emergency = true;

1203 1204 1205 1206 1207
		if (trans->cfg->mq_rx_supported) {
			/*
			 * used_bd is a 32 bit but only 12 are used to retrieve
			 * the vid
			 */
1208
			u16 vid = le32_to_cpu(rxq->used_bd[i]) & 0x0FFF;
1209

1210 1211 1212
			if (WARN(vid >= ARRAY_SIZE(trans_pcie->global_table),
				 "Invalid rxb index from HW %u\n", (u32)vid))
				goto out;
1213 1214 1215 1216 1217
			rxb = trans_pcie->global_table[vid];
		} else {
			rxb = rxq->queue[i];
			rxq->queue[i] = NULL;
		}
1218

1219
		IWL_DEBUG_RX(trans, "Q %d: HW = %d, SW = %d\n", rxq->id, r, i);
1220
		iwl_pcie_rx_handle_rb(trans, rxq, rxb, emergency);
1221

1222
		i = (i + 1) & (rxq->queue_size - 1);
1223

1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234
		/*
		 * If we have RX_CLAIM_REQ_ALLOC released rx buffers -
		 * try to claim the pre-allocated buffers from the allocator.
		 * If not ready - will try to reclaim next time.
		 * There is no need to reschedule work - allocator exits only
		 * on success
		 */
		if (rxq->used_count >= RX_CLAIM_REQ_ALLOC)
			iwl_pcie_rx_allocator_get(trans, rxq);

		if (rxq->used_count % RX_CLAIM_REQ_ALLOC == 0 && !emergency) {
1235 1236
			struct iwl_rb_allocator *rba = &trans_pcie->rba;

1237 1238 1239 1240 1241
			/* Add the remaining empty RBDs for allocator use */
			spin_lock(&rba->lock);
			list_splice_tail_init(&rxq->rx_used, &rba->rbd_empty);
			spin_unlock(&rba->lock);
		} else if (emergency) {
1242
			count++;
1243
			if (count == 8) {
1244
				count = 0;
1245
				if (rxq->used_count < rxq->queue_size / 3)
1246
					emergency = false;
1247 1248

				rxq->read = i;
1249
				spin_unlock(&rxq->lock);
1250
				iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1251
				iwl_pcie_rxq_restock(trans, rxq);
1252 1253
				goto restart;
			}
1254
		}
1255
	}
1256
out:
1257 1258
	/* Backtrack one entry */
	rxq->read = i;
1259 1260
	spin_unlock(&rxq->lock);

1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273
	/*
	 * handle a case where in emergency there are some unallocated RBDs.
	 * those RBDs are in the used list, but are not tracked by the queue's
	 * used_count which counts allocator owned RBDs.
	 * unallocated emergency RBDs must be allocated on exit, otherwise
	 * when called again the function may not be in emergency mode and
	 * they will be handed to the allocator with no tracking in the RBD
	 * allocator counters, which will lead to them never being claimed back
	 * by the queue.
	 * by allocating them here, they are now in the queue free list, and
	 * will be restocked by the next call of iwl_pcie_rxq_restock.
	 */
	if (unlikely(emergency && count))
1274
		iwl_pcie_rxq_alloc_rbs(trans, GFP_ATOMIC, rxq);
1275

1276 1277
	if (rxq->napi.poll)
		napi_gro_flush(&rxq->napi, false);
1278 1279

	iwl_pcie_rxq_restock(trans, rxq);
1280 1281
}

1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300
static struct iwl_trans_pcie *iwl_pcie_get_trans_pcie(struct msix_entry *entry)
{
	u8 queue = entry->entry;
	struct msix_entry *entries = entry - queue;

	return container_of(entries, struct iwl_trans_pcie, msix_entries[0]);
}

static inline void iwl_pcie_clear_irq(struct iwl_trans *trans,
				      struct msix_entry *entry)
{
	/*
	 * Before sending the interrupt the HW disables it to prevent
	 * a nested interrupt. This is done by writing 1 to the corresponding
	 * bit in the mask register. After handling the interrupt, it should be
	 * re-enabled by clearing this bit. This register is defined as
	 * write 1 clear (W1C) register, meaning that it's being clear
	 * by writing 1 to the bit.
	 */
1301
	iwl_write32(trans, CSR_MSIX_AUTOMASK_ST_AD, BIT(entry->entry));
1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313
}

/*
 * iwl_pcie_rx_msix_handle - Main entry function for receiving responses from fw
 * This interrupt handler should be used with RSS queue only.
 */
irqreturn_t iwl_pcie_irq_rx_msix_handler(int irq, void *dev_id)
{
	struct msix_entry *entry = dev_id;
	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
	struct iwl_trans *trans = trans_pcie->trans;

1314 1315 1316
	if (WARN_ON(entry->entry >= trans->num_rx_queues))
		return IRQ_NONE;

1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329
	lock_map_acquire(&trans->sync_cmd_lockdep_map);

	local_bh_disable();
	iwl_pcie_rx_handle(trans, entry->entry);
	local_bh_enable();

	iwl_pcie_clear_irq(trans, entry);

	lock_map_release(&trans->sync_cmd_lockdep_map);

	return IRQ_HANDLED;
}

1330 1331
/*
 * iwl_pcie_irq_handle_error - called for HW or SW error interrupt from card
1332
 */
1333
static void iwl_pcie_irq_handle_error(struct iwl_trans *trans)
1334
{
1335
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1336
	int i;
1337

1338
	/* W/A for WiFi/WiMAX coex and WiMAX own the RF */
1339
	if (trans->cfg->internal_wimax_coex &&
1340
	    !trans->cfg->apmg_not_supported &&
1341
	    (!(iwl_read_prph(trans, APMG_CLK_CTRL_REG) &
1342
			     APMS_CLK_VAL_MRB_FUNC_MODE) ||
1343
	     (iwl_read_prph(trans, APMG_PS_CTRL_REG) &
1344
			    APMG_PS_CTRL_VAL_RESET_REQ))) {
1345
		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1346
		iwl_op_mode_wimax_active(trans->op_mode);
1347
		wake_up(&trans_pcie->wait_command_queue);
1348 1349 1350
		return;
	}

1351
	iwl_pcie_dump_csr(trans);
1352
	iwl_dump_fh(trans, NULL);
1353

1354
	local_bh_disable();
1355 1356 1357
	/* The STATUS_FW_ERROR bit is set in this function. This must happen
	 * before we wake up the command caller, to ensure a proper cleanup. */
	iwl_trans_fw_error(trans);
1358
	local_bh_enable();
1359

1360 1361 1362
	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++)
		del_timer(&trans_pcie->txq[i].stuck_timer);

1363 1364
	clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
	wake_up(&trans_pcie->wait_command_queue);
1365 1366
}

1367
static u32 iwl_pcie_int_cause_non_ict(struct iwl_trans *trans)
1368 1369 1370
{
	u32 inta;

1371
	lockdep_assert_held(&IWL_TRANS_GET_PCIE_TRANS(trans)->irq_lock);
1372 1373 1374 1375 1376 1377 1378

	trace_iwlwifi_dev_irq(trans->dev);

	/* Discover which interrupts are active/pending */
	inta = iwl_read32(trans, CSR_INT);

	/* the thread will service interrupts and re-enable them */
1379
	return inta;
1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394
}

/* a device (PCI-E) page is 4096 bytes long */
#define ICT_SHIFT	12
#define ICT_SIZE	(1 << ICT_SHIFT)
#define ICT_COUNT	(ICT_SIZE / sizeof(u32))

/* interrupt handler using ict table, with this interrupt driver will
 * stop using INTA register to get device's interrupt, reading this register
 * is expensive, device will write interrupts in ICT dram table, increment
 * index then will fire interrupt to driver, driver will OR all ICT table
 * entries from current index up to table entry with 0 value. the result is
 * the interrupt we need to service, driver will set the entries back to 0 and
 * set index.
 */
1395
static u32 iwl_pcie_int_cause_ict(struct iwl_trans *trans)
1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408
{
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
	u32 inta;
	u32 val = 0;
	u32 read;

	trace_iwlwifi_dev_irq(trans->dev);

	/* Ignore interrupt if there's nothing in NIC to service.
	 * This may be due to IRQ shared with another device,
	 * or due to sporadic interrupts thrown from our NIC. */
	read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
	trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index, read);
1409 1410
	if (!read)
		return 0;
1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421

	/*
	 * Collect all entries up to the first 0, starting from ict_index;
	 * note we already read at ict_index.
	 */
	do {
		val |= read;
		IWL_DEBUG_ISR(trans, "ICT index %d value 0x%08X\n",
				trans_pcie->ict_index, read);
		trans_pcie->ict_tbl[trans_pcie->ict_index] = 0;
		trans_pcie->ict_index =
1422
			((trans_pcie->ict_index + 1) & (ICT_COUNT - 1));
1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443

		read = le32_to_cpu(trans_pcie->ict_tbl[trans_pcie->ict_index]);
		trace_iwlwifi_dev_ict_read(trans->dev, trans_pcie->ict_index,
					   read);
	} while (read);

	/* We should not get this value, just ignore it. */
	if (val == 0xffffffff)
		val = 0;

	/*
	 * this is a w/a for a h/w bug. the h/w bug may cause the Rx bit
	 * (bit 15 before shifting it to 31) to clear when using interrupt
	 * coalescing. fortunately, bits 18 and 19 stay set when this happens
	 * so we use them to decide on the real state of the Rx bit.
	 * In order words, bit 15 is set if bit 18 or bit 19 are set.
	 */
	if (val & 0xC0000)
		val |= 0x8000;

	inta = (0xff & val) | ((0xff00 & val) << 16);
1444
	return inta;
1445 1446
}

1447
irqreturn_t iwl_pcie_irq_handler(int irq, void *dev_id)
1448
{
1449
	struct iwl_trans *trans = dev_id;
1450 1451
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1452 1453 1454
	u32 inta = 0;
	u32 handled = 0;

1455 1456
	lock_map_acquire(&trans->sync_cmd_lockdep_map);

1457
	spin_lock(&trans_pcie->irq_lock);
1458

1459 1460 1461 1462
	/* dram interrupt table not set yet,
	 * use legacy interrupt.
	 */
	if (likely(trans_pcie->use_ict))
1463
		inta = iwl_pcie_int_cause_ict(trans);
1464
	else
1465
		inta = iwl_pcie_int_cause_non_ict(trans);
1466

1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485
	if (iwl_have_debug_level(IWL_DL_ISR)) {
		IWL_DEBUG_ISR(trans,
			      "ISR inta 0x%08x, enabled 0x%08x(sw), enabled(hw) 0x%08x, fh 0x%08x\n",
			      inta, trans_pcie->inta_mask,
			      iwl_read32(trans, CSR_INT_MASK),
			      iwl_read32(trans, CSR_FH_INT_STATUS));
		if (inta & (~trans_pcie->inta_mask))
			IWL_DEBUG_ISR(trans,
				      "We got a masked interrupt (0x%08x)\n",
				      inta & (~trans_pcie->inta_mask));
	}

	inta &= trans_pcie->inta_mask;

	/*
	 * Ignore interrupt if there's nothing in NIC to service.
	 * This may be due to IRQ shared with another device,
	 * or due to sporadic interrupts thrown from our NIC.
	 */
1486
	if (unlikely(!inta)) {
1487 1488 1489 1490 1491 1492 1493
		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
		/*
		 * Re-enable interrupts here since we don't
		 * have anything to service
		 */
		if (test_bit(STATUS_INT_ENABLED, &trans->status))
			iwl_enable_interrupts(trans);
1494
		spin_unlock(&trans_pcie->irq_lock);
1495 1496 1497 1498
		lock_map_release(&trans->sync_cmd_lockdep_map);
		return IRQ_NONE;
	}

1499 1500 1501 1502 1503 1504
	if (unlikely(inta == 0xFFFFFFFF || (inta & 0xFFFFFFF0) == 0xa5a5a5a0)) {
		/*
		 * Hardware disappeared. It might have
		 * already raised an interrupt.
		 */
		IWL_WARN(trans, "HARDWARE GONE?? INTA == 0x%08x\n", inta);
1505
		spin_unlock(&trans_pcie->irq_lock);
1506
		goto out;
1507 1508
	}

1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519
	/* Ack/clear/reset pending uCode interrupts.
	 * Note:  Some bits in CSR_INT are "OR" of bits in CSR_FH_INT_STATUS,
	 */
	/* There is a hardware bug in the interrupt mask function that some
	 * interrupts (i.e. CSR_INT_BIT_SCD) can still be generated even if
	 * they are disabled in the CSR_INT_MASK register. Furthermore the
	 * ICT interrupt handling mechanism has another bug that might cause
	 * these unmasked interrupts fail to be detected. We workaround the
	 * hardware bugs here by ACKing all the possible interrupts so that
	 * interrupt coalescing can still be achieved.
	 */
1520
	iwl_write32(trans, CSR_INT, inta | ~trans_pcie->inta_mask);
1521

1522
	if (iwl_have_debug_level(IWL_DL_ISR))
1523
		IWL_DEBUG_ISR(trans, "inta 0x%08x, enabled 0x%08x\n",
1524
			      inta, iwl_read32(trans, CSR_INT_MASK));
1525

1526
	spin_unlock(&trans_pcie->irq_lock);
1527

1528 1529
	/* Now service all interrupt bits discovered above. */
	if (inta & CSR_INT_BIT_HW_ERR) {
1530
		IWL_ERR(trans, "Hardware error detected.  Restarting.\n");
1531 1532

		/* Tell the device to stop sending interrupts */
1533
		iwl_disable_interrupts(trans);
1534

1535
		isr_stats->hw++;
1536
		iwl_pcie_irq_handle_error(trans);
1537 1538 1539

		handled |= CSR_INT_BIT_HW_ERR;

1540
		goto out;
1541 1542
	}

1543
	if (iwl_have_debug_level(IWL_DL_ISR)) {
1544 1545
		/* NIC fires this, but we don't use it, redundant with WAKEUP */
		if (inta & CSR_INT_BIT_SCD) {
1546 1547
			IWL_DEBUG_ISR(trans,
				      "Scheduler finished to transmit the frame/frames.\n");
1548
			isr_stats->sch++;
1549 1550 1551 1552
		}

		/* Alive notification via Rx interrupt will do the real work */
		if (inta & CSR_INT_BIT_ALIVE) {
1553
			IWL_DEBUG_ISR(trans, "Alive interrupt\n");
1554
			isr_stats->alive++;
1555 1556
		}
	}
1557

1558 1559 1560 1561 1562
	/* Safely ignore these bits for debug checks below */
	inta &= ~(CSR_INT_BIT_SCD | CSR_INT_BIT_ALIVE);

	/* HW RF KILL switch toggled */
	if (inta & CSR_INT_BIT_RF_KILL) {
1563
		bool hw_rfkill;
1564

1565
		hw_rfkill = iwl_is_rfkill_set(trans);
1566
		IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
1567
			 hw_rfkill ? "disable radio" : "enable radio");
1568

1569
		isr_stats->rfkill++;
1570

1571
		mutex_lock(&trans_pcie->mutex);
1572
		iwl_trans_pcie_rf_kill(trans, hw_rfkill);
1573
		mutex_unlock(&trans_pcie->mutex);
1574
		if (hw_rfkill) {
1575 1576 1577
			set_bit(STATUS_RFKILL, &trans->status);
			if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
					       &trans->status))
1578 1579 1580 1581
				IWL_DEBUG_RF_KILL(trans,
						  "Rfkill while SYNC HCMD in flight\n");
			wake_up(&trans_pcie->wait_command_queue);
		} else {
1582
			clear_bit(STATUS_RFKILL, &trans->status);
1583
		}
1584 1585 1586 1587 1588 1589

		handled |= CSR_INT_BIT_RF_KILL;
	}

	/* Chip got too hot and stopped itself */
	if (inta & CSR_INT_BIT_CT_KILL) {
1590
		IWL_ERR(trans, "Microcode CT kill error detected.\n");
1591
		isr_stats->ctkill++;
1592 1593 1594 1595 1596
		handled |= CSR_INT_BIT_CT_KILL;
	}

	/* Error detected by uCode */
	if (inta & CSR_INT_BIT_SW_ERR) {
1597
		IWL_ERR(trans, "Microcode SW error detected. "
1598
			" Restarting 0x%X.\n", inta);
1599
		isr_stats->sw++;
1600
		iwl_pcie_irq_handle_error(trans);
1601 1602 1603 1604 1605
		handled |= CSR_INT_BIT_SW_ERR;
	}

	/* uCode wakes up after power-down sleep */
	if (inta & CSR_INT_BIT_WAKEUP) {
1606
		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
1607
		iwl_pcie_rxq_check_wrptr(trans);
1608
		iwl_pcie_txq_check_wrptrs(trans);
1609

1610
		isr_stats->wakeup++;
1611 1612 1613 1614 1615 1616 1617 1618

		handled |= CSR_INT_BIT_WAKEUP;
	}

	/* All uCode command responses, including Tx command responses,
	 * Rx "responses" (frame-received notification), and other
	 * notifications from uCode come through here*/
	if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX |
1619
		    CSR_INT_BIT_RX_PERIODIC)) {
1620
		IWL_DEBUG_ISR(trans, "Rx interrupt\n");
1621 1622
		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX)) {
			handled |= (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX);
1623
			iwl_write32(trans, CSR_FH_INT_STATUS,
1624 1625 1626 1627
					CSR_FH_INT_RX_MASK);
		}
		if (inta & CSR_INT_BIT_RX_PERIODIC) {
			handled |= CSR_INT_BIT_RX_PERIODIC;
1628
			iwl_write32(trans,
1629
				CSR_INT, CSR_INT_BIT_RX_PERIODIC);
1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642
		}
		/* Sending RX interrupt require many steps to be done in the
		 * the device:
		 * 1- write interrupt to current index in ICT table.
		 * 2- dma RX frame.
		 * 3- update RX shared data to indicate last write index.
		 * 4- send interrupt.
		 * This could lead to RX race, driver could receive RX interrupt
		 * but the shared data changes does not reflect this;
		 * periodic interrupt will detect any dangling Rx activity.
		 */

		/* Disable periodic interrupt; we use it as just a one-shot. */
1643
		iwl_write8(trans, CSR_INT_PERIODIC_REG,
1644
			    CSR_INT_PERIODIC_DIS);
1645

1646 1647 1648 1649 1650 1651 1652 1653
		/*
		 * Enable periodic interrupt in 8 msec only if we received
		 * real RX interrupt (instead of just periodic int), to catch
		 * any dangling Rx interrupt.  If it was just the periodic
		 * interrupt, there was no dangling Rx activity, and no need
		 * to extend the periodic interrupt; one-shot is enough.
		 */
		if (inta & (CSR_INT_BIT_FH_RX | CSR_INT_BIT_SW_RX))
1654
			iwl_write8(trans, CSR_INT_PERIODIC_REG,
1655
				   CSR_INT_PERIODIC_ENA);
1656

1657
		isr_stats->rx++;
1658 1659

		local_bh_disable();
1660
		iwl_pcie_rx_handle(trans, 0);
1661
		local_bh_enable();
1662 1663 1664 1665
	}

	/* This "Tx" DMA channel is used only for loading uCode */
	if (inta & CSR_INT_BIT_FH_TX) {
1666
		iwl_write32(trans, CSR_FH_INT_STATUS, CSR_FH_INT_TX_MASK);
1667
		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
1668
		isr_stats->tx++;
1669 1670
		handled |= CSR_INT_BIT_FH_TX;
		/* Wake up uCode load routine, now that load is complete */
1671 1672
		trans_pcie->ucode_write_complete = true;
		wake_up(&trans_pcie->ucode_write_waitq);
1673 1674 1675
	}

	if (inta & ~handled) {
1676
		IWL_ERR(trans, "Unhandled INTA bits 0x%08x\n", inta & ~handled);
1677
		isr_stats->unhandled++;
1678 1679
	}

1680 1681 1682
	if (inta & ~(trans_pcie->inta_mask)) {
		IWL_WARN(trans, "Disabled INTA bits 0x%08x were pending\n",
			 inta & ~trans_pcie->inta_mask);
1683 1684
	}

1685 1686 1687 1688 1689
	/* we are loading the firmware, enable FH_TX interrupt only */
	if (handled & CSR_INT_BIT_FH_TX)
		iwl_enable_fw_load_int(trans);
	/* only Re-enable all interrupt if disabled by irq */
	else if (test_bit(STATUS_INT_ENABLED, &trans->status))
1690
		iwl_enable_interrupts(trans);
1691
	/* Re-enable RF_KILL if it occurred */
1692 1693
	else if (handled & CSR_INT_BIT_RF_KILL)
		iwl_enable_rfkill_int(trans);
1694 1695 1696 1697

out:
	lock_map_release(&trans->sync_cmd_lockdep_map);
	return IRQ_HANDLED;
1698 1699
}

1700 1701 1702 1703 1704
/******************************************************************************
 *
 * ICT functions
 *
 ******************************************************************************/
1705

1706
/* Free dram table */
1707
void iwl_pcie_free_ict(struct iwl_trans *trans)
1708
{
1709
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1710

1711
	if (trans_pcie->ict_tbl) {
1712
		dma_free_coherent(trans->dev, ICT_SIZE,
1713
				  trans_pcie->ict_tbl,
1714
				  trans_pcie->ict_tbl_dma);
1715 1716
		trans_pcie->ict_tbl = NULL;
		trans_pcie->ict_tbl_dma = 0;
1717 1718 1719
	}
}

1720 1721 1722
/*
 * allocate dram shared table, it is an aligned memory
 * block of ICT_SIZE.
1723 1724
 * also reset all data related to ICT table interrupt.
 */
1725
int iwl_pcie_alloc_ict(struct iwl_trans *trans)
1726
{
1727
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1728

1729
	trans_pcie->ict_tbl =
1730
		dma_zalloc_coherent(trans->dev, ICT_SIZE,
1731 1732 1733
				   &trans_pcie->ict_tbl_dma,
				   GFP_KERNEL);
	if (!trans_pcie->ict_tbl)
1734 1735
		return -ENOMEM;

1736 1737
	/* just an API sanity check ... it is guaranteed to be aligned */
	if (WARN_ON(trans_pcie->ict_tbl_dma & (ICT_SIZE - 1))) {
1738
		iwl_pcie_free_ict(trans);
1739 1740
		return -EINVAL;
	}
1741 1742 1743 1744 1745 1746 1747

	return 0;
}

/* Device is going up inform it about using ICT interrupt table,
 * also we need to tell the driver to start using ICT interrupt.
 */
1748
void iwl_pcie_reset_ict(struct iwl_trans *trans)
1749
{
1750
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1751 1752
	u32 val;

1753
	if (!trans_pcie->ict_tbl)
1754
		return;
1755

1756
	spin_lock(&trans_pcie->irq_lock);
1757
	iwl_disable_interrupts(trans);
1758

1759
	memset(trans_pcie->ict_tbl, 0, ICT_SIZE);
1760

1761
	val = trans_pcie->ict_tbl_dma >> ICT_SHIFT;
1762

1763 1764 1765
	val |= CSR_DRAM_INT_TBL_ENABLE |
	       CSR_DRAM_INIT_TBL_WRAP_CHECK |
	       CSR_DRAM_INIT_TBL_WRITE_POINTER;
1766

1767
	IWL_DEBUG_ISR(trans, "CSR_DRAM_INT_TBL_REG =0x%x\n", val);
1768

1769
	iwl_write32(trans, CSR_DRAM_INT_TBL_REG, val);
1770 1771
	trans_pcie->use_ict = true;
	trans_pcie->ict_index = 0;
1772
	iwl_write32(trans, CSR_INT, trans_pcie->inta_mask);
1773
	iwl_enable_interrupts(trans);
1774
	spin_unlock(&trans_pcie->irq_lock);
1775 1776 1777
}

/* Device is going down disable ict interrupt usage */
1778
void iwl_pcie_disable_ict(struct iwl_trans *trans)
1779
{
1780
	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1781

1782
	spin_lock(&trans_pcie->irq_lock);
1783
	trans_pcie->use_ict = false;
1784
	spin_unlock(&trans_pcie->irq_lock);
1785 1786
}

1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800
irqreturn_t iwl_pcie_isr(int irq, void *data)
{
	struct iwl_trans *trans = data;

	if (!trans)
		return IRQ_NONE;

	/* Disable (but don't clear!) interrupts here to avoid
	 * back-to-back ISRs and sporadic interrupts from our NIC.
	 * If we have something to service, the tasklet will re-enable ints.
	 * If we *don't* have something, we'll re-enable before leaving here.
	 */
	iwl_write32(trans, CSR_INT_MASK, 0x00000000);

1801
	return IRQ_WAKE_THREAD;
1802
}
1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813

irqreturn_t iwl_pcie_msix_isr(int irq, void *data)
{
	return IRQ_WAKE_THREAD;
}

irqreturn_t iwl_pcie_irq_msix_handler(int irq, void *dev_id)
{
	struct msix_entry *entry = dev_id;
	struct iwl_trans_pcie *trans_pcie = iwl_pcie_get_trans_pcie(entry);
	struct iwl_trans *trans = trans_pcie->trans;
1814
	struct isr_statistics *isr_stats = &trans_pcie->isr_stats;
1815 1816 1817 1818 1819
	u32 inta_fh, inta_hw;

	lock_map_acquire(&trans->sync_cmd_lockdep_map);

	spin_lock(&trans_pcie->irq_lock);
1820 1821
	inta_fh = iwl_read32(trans, CSR_MSIX_FH_INT_CAUSES_AD);
	inta_hw = iwl_read32(trans, CSR_MSIX_HW_INT_CAUSES_AD);
1822 1823 1824
	/*
	 * Clear causes registers to avoid being handling the same cause.
	 */
1825 1826
	iwl_write32(trans, CSR_MSIX_FH_INT_CAUSES_AD, inta_fh);
	iwl_write32(trans, CSR_MSIX_HW_INT_CAUSES_AD, inta_hw);
1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928
	spin_unlock(&trans_pcie->irq_lock);

	if (unlikely(!(inta_fh | inta_hw))) {
		IWL_DEBUG_ISR(trans, "Ignore interrupt, inta == 0\n");
		lock_map_release(&trans->sync_cmd_lockdep_map);
		return IRQ_NONE;
	}

	if (iwl_have_debug_level(IWL_DL_ISR))
		IWL_DEBUG_ISR(trans, "ISR inta_fh 0x%08x, enabled 0x%08x\n",
			      inta_fh,
			      iwl_read32(trans, CSR_MSIX_FH_INT_MASK_AD));

	/* This "Tx" DMA channel is used only for loading uCode */
	if (inta_fh & MSIX_FH_INT_CAUSES_D2S_CH0_NUM) {
		IWL_DEBUG_ISR(trans, "uCode load interrupt\n");
		isr_stats->tx++;
		/*
		 * Wake up uCode load routine,
		 * now that load is complete
		 */
		trans_pcie->ucode_write_complete = true;
		wake_up(&trans_pcie->ucode_write_waitq);
	}

	/* Error detected by uCode */
	if ((inta_fh & MSIX_FH_INT_CAUSES_FH_ERR) ||
	    (inta_hw & MSIX_HW_INT_CAUSES_REG_SW_ERR)) {
		IWL_ERR(trans,
			"Microcode SW error detected. Restarting 0x%X.\n",
			inta_fh);
		isr_stats->sw++;
		iwl_pcie_irq_handle_error(trans);
	}

	/* After checking FH register check HW register */
	if (iwl_have_debug_level(IWL_DL_ISR))
		IWL_DEBUG_ISR(trans,
			      "ISR inta_hw 0x%08x, enabled 0x%08x\n",
			      inta_hw,
			      iwl_read32(trans, CSR_MSIX_HW_INT_MASK_AD));

	/* Alive notification via Rx interrupt will do the real work */
	if (inta_hw & MSIX_HW_INT_CAUSES_REG_ALIVE) {
		IWL_DEBUG_ISR(trans, "Alive interrupt\n");
		isr_stats->alive++;
	}

	/* uCode wakes up after power-down sleep */
	if (inta_hw & MSIX_HW_INT_CAUSES_REG_WAKEUP) {
		IWL_DEBUG_ISR(trans, "Wakeup interrupt\n");
		iwl_pcie_rxq_check_wrptr(trans);
		iwl_pcie_txq_check_wrptrs(trans);

		isr_stats->wakeup++;
	}

	/* Chip got too hot and stopped itself */
	if (inta_hw & MSIX_HW_INT_CAUSES_REG_CT_KILL) {
		IWL_ERR(trans, "Microcode CT kill error detected.\n");
		isr_stats->ctkill++;
	}

	/* HW RF KILL switch toggled */
	if (inta_hw & MSIX_HW_INT_CAUSES_REG_RF_KILL) {
		bool hw_rfkill;

		hw_rfkill = iwl_is_rfkill_set(trans);
		IWL_WARN(trans, "RF_KILL bit toggled to %s.\n",
			 hw_rfkill ? "disable radio" : "enable radio");

		isr_stats->rfkill++;

		mutex_lock(&trans_pcie->mutex);
		iwl_trans_pcie_rf_kill(trans, hw_rfkill);
		mutex_unlock(&trans_pcie->mutex);
		if (hw_rfkill) {
			set_bit(STATUS_RFKILL, &trans->status);
			if (test_and_clear_bit(STATUS_SYNC_HCMD_ACTIVE,
					       &trans->status))
				IWL_DEBUG_RF_KILL(trans,
						  "Rfkill while SYNC HCMD in flight\n");
			wake_up(&trans_pcie->wait_command_queue);
		} else {
			clear_bit(STATUS_RFKILL, &trans->status);
		}
	}

	if (inta_hw & MSIX_HW_INT_CAUSES_REG_HW_ERR) {
		IWL_ERR(trans,
			"Hardware error detected. Restarting.\n");

		isr_stats->hw++;
		iwl_pcie_irq_handle_error(trans);
	}

	iwl_pcie_clear_irq(trans, entry);

	lock_map_release(&trans->sync_cmd_lockdep_map);

	return IRQ_HANDLED;
}