1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Tehuti Networks(R) Network Driver
4  * ethtool interface implementation
5  * Copyright (C) 2007 Tehuti Networks Ltd. All rights reserved
6  */
7 
8 /*
9  * RX HW/SW interaction overview
10  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
11  * There are 2 types of RX communication channels between driver and NIC.
12  * 1) RX Free Fifo - RXF - holds descriptors of empty buffers to accept incoming
13  * traffic. This Fifo is filled by SW and is readen by HW. Each descriptor holds
14  * info about buffer's location, size and ID. An ID field is used to identify a
15  * buffer when it's returned with data via RXD Fifo (see below)
16  * 2) RX Data Fifo - RXD - holds descriptors of full buffers. This Fifo is
17  * filled by HW and is readen by SW. Each descriptor holds status and ID.
18  * HW pops descriptor from RXF Fifo, stores ID, fills buffer with incoming data,
19  * via dma moves it into host memory, builds new RXD descriptor with same ID,
20  * pushes it into RXD Fifo and raises interrupt to indicate new RX data.
21  *
22  * Current NIC configuration (registers + firmware) makes NIC use 2 RXF Fifos.
23  * One holds 1.5K packets and another - 26K packets. Depending on incoming
24  * packet size, HW desides on a RXF Fifo to pop buffer from. When packet is
25  * filled with data, HW builds new RXD descriptor for it and push it into single
26  * RXD Fifo.
27  *
28  * RX SW Data Structures
29  * ~~~~~~~~~~~~~~~~~~~~~
30  * skb db - used to keep track of all skbs owned by SW and their dma addresses.
31  * For RX case, ownership lasts from allocating new empty skb for RXF until
32  * accepting full skb from RXD and passing it to OS. Each RXF Fifo has its own
33  * skb db. Implemented as array with bitmask.
34  * fifo - keeps info about fifo's size and location, relevant HW registers,
35  * usage and skb db. Each RXD and RXF Fifo has its own fifo structure.
36  * Implemented as simple struct.
37  *
38  * RX SW Execution Flow
39  * ~~~~~~~~~~~~~~~~~~~~
40  * Upon initialization (ifconfig up) driver creates RX fifos and initializes
41  * relevant registers. At the end of init phase, driver enables interrupts.
42  * NIC sees that there is no RXF buffers and raises
43  * RD_INTR interrupt, isr fills skbs and Rx begins.
44  * Driver has two receive operation modes:
45  *    NAPI - interrupt-driven mixed with polling
46  *    interrupt-driven only
47  *
48  * Interrupt-driven only flow is following. When buffer is ready, HW raises
49  * interrupt and isr is called. isr collects all available packets
50  * (bdx_rx_receive), refills skbs (bdx_rx_alloc_skbs) and exit.
51 
52  * Rx buffer allocation note
53  * ~~~~~~~~~~~~~~~~~~~~~~~~~
54  * Driver cares to feed such amount of RxF descriptors that respective amount of
55  * RxD descriptors can not fill entire RxD fifo. The main reason is lack of
56  * overflow check in Bordeaux for RxD fifo free/used size.
57  * FIXME: this is NOT fully implemented, more work should be done
58  *
59  */
60 
61 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
62 
63 #include "tehuti.h"
64 
65 static const struct pci_device_id bdx_pci_tbl[] = {
66 	{ PCI_VDEVICE(TEHUTI, 0x3009), },
67 	{ PCI_VDEVICE(TEHUTI, 0x3010), },
68 	{ PCI_VDEVICE(TEHUTI, 0x3014), },
69 	{ 0 }
70 };
71 
72 MODULE_DEVICE_TABLE(pci, bdx_pci_tbl);
73 
74 /* Definitions needed by ISR or NAPI functions */
75 static void bdx_rx_alloc_skbs(struct bdx_priv *priv, struct rxf_fifo *f);
76 static void bdx_tx_cleanup(struct bdx_priv *priv);
77 static int bdx_rx_receive(struct bdx_priv *priv, struct rxd_fifo *f, int budget);
78 
79 /* Definitions needed by FW loading */
80 static void bdx_tx_push_desc_safe(struct bdx_priv *priv, void *data, int size);
81 
82 /* Definitions needed by hw_start */
83 static int bdx_tx_init(struct bdx_priv *priv);
84 static int bdx_rx_init(struct bdx_priv *priv);
85 
86 /* Definitions needed by bdx_close */
87 static void bdx_rx_free(struct bdx_priv *priv);
88 static void bdx_tx_free(struct bdx_priv *priv);
89 
90 /* Definitions needed by bdx_probe */
91 static void bdx_set_ethtool_ops(struct net_device *netdev);
92 
93 /*************************************************************************
94  *    Print Info                                                         *
95  *************************************************************************/
96 
97 static void print_hw_id(struct pci_dev *pdev)
98 {
99 	struct pci_nic *nic = pci_get_drvdata(pdev);
100 	u16 pci_link_status = 0;
101 	u16 pci_ctrl = 0;
102 
103 	pci_read_config_word(pdev, PCI_LINK_STATUS_REG, &pci_link_status);
104 	pci_read_config_word(pdev, PCI_DEV_CTRL_REG, &pci_ctrl);
105 
106 	pr_info("%s%s\n", BDX_NIC_NAME,
107 		nic->port_num == 1 ? "" : ", 2-Port");
108 	pr_info("srom 0x%x fpga %d build %u lane# %d max_pl 0x%x mrrs 0x%x\n",
109 		readl(nic->regs + SROM_VER), readl(nic->regs + FPGA_VER) & 0xFFF,
110 		readl(nic->regs + FPGA_SEED),
111 		GET_LINK_STATUS_LANES(pci_link_status),
112 		GET_DEV_CTRL_MAXPL(pci_ctrl), GET_DEV_CTRL_MRRS(pci_ctrl));
113 }
114 
115 static void print_fw_id(struct pci_nic *nic)
116 {
117 	pr_info("fw 0x%x\n", readl(nic->regs + FW_VER));
118 }
119 
120 static void print_eth_id(struct net_device *ndev)
121 {
122 	netdev_info(ndev, "%s, Port %c\n",
123 		    BDX_NIC_NAME, (ndev->if_port == 0) ? 'A' : 'B');
124 
125 }
126 
127 /*************************************************************************
128  *    Code                                                               *
129  *************************************************************************/
130 
131 #define bdx_enable_interrupts(priv)	\
132 	do { WRITE_REG(priv, regIMR, IR_RUN); } while (0)
133 #define bdx_disable_interrupts(priv)	\
134 	do { WRITE_REG(priv, regIMR, 0); } while (0)
135 
136 /**
137  * bdx_fifo_init - create TX/RX descriptor fifo for host-NIC communication.
138  * @priv: NIC private structure
139  * @f: fifo to initialize
140  * @fsz_type: fifo size type: 0-4KB, 1-8KB, 2-16KB, 3-32KB
141  * @reg_CFG0: offsets of registers relative to base address
142  * @reg_CFG1: offsets of registers relative to base address
143  * @reg_RPTR: offsets of registers relative to base address
144  * @reg_WPTR: offsets of registers relative to base address
145  *
146  * 1K extra space is allocated at the end of the fifo to simplify
147  * processing of descriptors that wraps around fifo's end
148  *
149  * Returns 0 on success, negative value on failure
150  *
151  */
152 static int
153 bdx_fifo_init(struct bdx_priv *priv, struct fifo *f, int fsz_type,
154 	      u16 reg_CFG0, u16 reg_CFG1, u16 reg_RPTR, u16 reg_WPTR)
155 {
156 	u16 memsz = FIFO_SIZE * (1 << fsz_type);
157 
158 	memset(f, 0, sizeof(struct fifo));
159 	/* dma_alloc_coherent gives us 4k-aligned memory */
160 	f->va = dma_alloc_coherent(&priv->pdev->dev, memsz + FIFO_EXTRA_SPACE,
161 				   &f->da, GFP_ATOMIC);
162 	if (!f->va) {
163 		pr_err("dma_alloc_coherent failed\n");
164 		RET(-ENOMEM);
165 	}
166 	f->reg_CFG0 = reg_CFG0;
167 	f->reg_CFG1 = reg_CFG1;
168 	f->reg_RPTR = reg_RPTR;
169 	f->reg_WPTR = reg_WPTR;
170 	f->rptr = 0;
171 	f->wptr = 0;
172 	f->memsz = memsz;
173 	f->size_mask = memsz - 1;
174 	WRITE_REG(priv, reg_CFG0, (u32) ((f->da & TX_RX_CFG0_BASE) | fsz_type));
175 	WRITE_REG(priv, reg_CFG1, H32_64(f->da));
176 
177 	RET(0);
178 }
179 
180 /**
181  * bdx_fifo_free - free all resources used by fifo
182  * @priv: NIC private structure
183  * @f: fifo to release
184  */
185 static void bdx_fifo_free(struct bdx_priv *priv, struct fifo *f)
186 {
187 	ENTER;
188 	if (f->va) {
189 		dma_free_coherent(&priv->pdev->dev,
190 				  f->memsz + FIFO_EXTRA_SPACE, f->va, f->da);
191 		f->va = NULL;
192 	}
193 	RET();
194 }
195 
196 /**
197  * bdx_link_changed - notifies OS about hw link state.
198  * @priv: hw adapter structure
199  */
200 static void bdx_link_changed(struct bdx_priv *priv)
201 {
202 	u32 link = READ_REG(priv, regMAC_LNK_STAT) & MAC_LINK_STAT;
203 
204 	if (!link) {
205 		if (netif_carrier_ok(priv->ndev)) {
206 			netif_stop_queue(priv->ndev);
207 			netif_carrier_off(priv->ndev);
208 			netdev_err(priv->ndev, "Link Down\n");
209 		}
210 	} else {
211 		if (!netif_carrier_ok(priv->ndev)) {
212 			netif_wake_queue(priv->ndev);
213 			netif_carrier_on(priv->ndev);
214 			netdev_err(priv->ndev, "Link Up\n");
215 		}
216 	}
217 }
218 
219 static void bdx_isr_extra(struct bdx_priv *priv, u32 isr)
220 {
221 	if (isr & IR_RX_FREE_0) {
222 		bdx_rx_alloc_skbs(priv, &priv->rxf_fifo0);
223 		DBG("RX_FREE_0\n");
224 	}
225 
226 	if (isr & IR_LNKCHG0)
227 		bdx_link_changed(priv);
228 
229 	if (isr & IR_PCIE_LINK)
230 		netdev_err(priv->ndev, "PCI-E Link Fault\n");
231 
232 	if (isr & IR_PCIE_TOUT)
233 		netdev_err(priv->ndev, "PCI-E Time Out\n");
234 
235 }
236 
237 /**
238  * bdx_isr_napi - Interrupt Service Routine for Bordeaux NIC
239  * @irq: interrupt number
240  * @dev: network device
241  *
242  * Return IRQ_NONE if it was not our interrupt, IRQ_HANDLED - otherwise
243  *
244  * It reads ISR register to know interrupt reasons, and proceed them one by one.
245  * Reasons of interest are:
246  *    RX_DESC - new packet has arrived and RXD fifo holds its descriptor
247  *    RX_FREE - number of free Rx buffers in RXF fifo gets low
248  *    TX_FREE - packet was transmited and RXF fifo holds its descriptor
249  */
250 
251 static irqreturn_t bdx_isr_napi(int irq, void *dev)
252 {
253 	struct net_device *ndev = dev;
254 	struct bdx_priv *priv = netdev_priv(ndev);
255 	u32 isr;
256 
257 	ENTER;
258 	isr = (READ_REG(priv, regISR) & IR_RUN);
259 	if (unlikely(!isr)) {
260 		bdx_enable_interrupts(priv);
261 		return IRQ_NONE;	/* Not our interrupt */
262 	}
263 
264 	if (isr & IR_EXTRA)
265 		bdx_isr_extra(priv, isr);
266 
267 	if (isr & (IR_RX_DESC_0 | IR_TX_FREE_0)) {
268 		if (likely(napi_schedule_prep(&priv->napi))) {
269 			__napi_schedule(&priv->napi);
270 			RET(IRQ_HANDLED);
271 		} else {
272 			/* NOTE: we get here if intr has slipped into window
273 			 * between these lines in bdx_poll:
274 			 *    bdx_enable_interrupts(priv);
275 			 *    return 0;
276 			 * currently intrs are disabled (since we read ISR),
277 			 * and we have failed to register next poll.
278 			 * so we read the regs to trigger chip
279 			 * and allow further interupts. */
280 			READ_REG(priv, regTXF_WPTR_0);
281 			READ_REG(priv, regRXD_WPTR_0);
282 		}
283 	}
284 
285 	bdx_enable_interrupts(priv);
286 	RET(IRQ_HANDLED);
287 }
288 
289 static int bdx_poll(struct napi_struct *napi, int budget)
290 {
291 	struct bdx_priv *priv = container_of(napi, struct bdx_priv, napi);
292 	int work_done;
293 
294 	ENTER;
295 	bdx_tx_cleanup(priv);
296 	work_done = bdx_rx_receive(priv, &priv->rxd_fifo0, budget);
297 	if ((work_done < budget) ||
298 	    (priv->napi_stop++ >= 30)) {
299 		DBG("rx poll is done. backing to isr-driven\n");
300 
301 		/* from time to time we exit to let NAPI layer release
302 		 * device lock and allow waiting tasks (eg rmmod) to advance) */
303 		priv->napi_stop = 0;
304 
305 		napi_complete_done(napi, work_done);
306 		bdx_enable_interrupts(priv);
307 	}
308 	return work_done;
309 }
310 
311 /**
312  * bdx_fw_load - loads firmware to NIC
313  * @priv: NIC private structure
314  *
315  * Firmware is loaded via TXD fifo, so it must be initialized first.
316  * Firware must be loaded once per NIC not per PCI device provided by NIC (NIC
317  * can have few of them). So all drivers use semaphore register to choose one
318  * that will actually load FW to NIC.
319  */
320 
321 static int bdx_fw_load(struct bdx_priv *priv)
322 {
323 	const struct firmware *fw = NULL;
324 	int master, i;
325 	int rc;
326 
327 	ENTER;
328 	master = READ_REG(priv, regINIT_SEMAPHORE);
329 	if (!READ_REG(priv, regINIT_STATUS) && master) {
330 		rc = request_firmware(&fw, "tehuti/bdx.bin", &priv->pdev->dev);
331 		if (rc)
332 			goto out;
333 		bdx_tx_push_desc_safe(priv, (char *)fw->data, fw->size);
334 		mdelay(100);
335 	}
336 	for (i = 0; i < 200; i++) {
337 		if (READ_REG(priv, regINIT_STATUS)) {
338 			rc = 0;
339 			goto out;
340 		}
341 		mdelay(2);
342 	}
343 	rc = -EIO;
344 out:
345 	if (master)
346 		WRITE_REG(priv, regINIT_SEMAPHORE, 1);
347 
348 	release_firmware(fw);
349 
350 	if (rc) {
351 		netdev_err(priv->ndev, "firmware loading failed\n");
352 		if (rc == -EIO)
353 			DBG("VPC = 0x%x VIC = 0x%x INIT_STATUS = 0x%x i=%d\n",
354 			    READ_REG(priv, regVPC),
355 			    READ_REG(priv, regVIC),
356 			    READ_REG(priv, regINIT_STATUS), i);
357 		RET(rc);
358 	} else {
359 		DBG("%s: firmware loading success\n", priv->ndev->name);
360 		RET(0);
361 	}
362 }
363 
364 static void bdx_restore_mac(struct net_device *ndev, struct bdx_priv *priv)
365 {
366 	u32 val;
367 
368 	ENTER;
369 	DBG("mac0=%x mac1=%x mac2=%x\n",
370 	    READ_REG(priv, regUNC_MAC0_A),
371 	    READ_REG(priv, regUNC_MAC1_A), READ_REG(priv, regUNC_MAC2_A));
372 
373 	val = (ndev->dev_addr[0] << 8) | (ndev->dev_addr[1]);
374 	WRITE_REG(priv, regUNC_MAC2_A, val);
375 	val = (ndev->dev_addr[2] << 8) | (ndev->dev_addr[3]);
376 	WRITE_REG(priv, regUNC_MAC1_A, val);
377 	val = (ndev->dev_addr[4] << 8) | (ndev->dev_addr[5]);
378 	WRITE_REG(priv, regUNC_MAC0_A, val);
379 
380 	DBG("mac0=%x mac1=%x mac2=%x\n",
381 	    READ_REG(priv, regUNC_MAC0_A),
382 	    READ_REG(priv, regUNC_MAC1_A), READ_REG(priv, regUNC_MAC2_A));
383 	RET();
384 }
385 
386 /**
387  * bdx_hw_start - inits registers and starts HW's Rx and Tx engines
388  * @priv: NIC private structure
389  */
390 static int bdx_hw_start(struct bdx_priv *priv)
391 {
392 	int rc = -EIO;
393 	struct net_device *ndev = priv->ndev;
394 
395 	ENTER;
396 	bdx_link_changed(priv);
397 
398 	/* 10G overall max length (vlan, eth&ip header, ip payload, crc) */
399 	WRITE_REG(priv, regFRM_LENGTH, 0X3FE0);
400 	WRITE_REG(priv, regPAUSE_QUANT, 0x96);
401 	WRITE_REG(priv, regRX_FIFO_SECTION, 0x800010);
402 	WRITE_REG(priv, regTX_FIFO_SECTION, 0xE00010);
403 	WRITE_REG(priv, regRX_FULLNESS, 0);
404 	WRITE_REG(priv, regTX_FULLNESS, 0);
405 	WRITE_REG(priv, regCTRLST,
406 		  regCTRLST_BASE | regCTRLST_RX_ENA | regCTRLST_TX_ENA);
407 
408 	WRITE_REG(priv, regVGLB, 0);
409 	WRITE_REG(priv, regMAX_FRAME_A,
410 		  priv->rxf_fifo0.m.pktsz & MAX_FRAME_AB_VAL);
411 
412 	DBG("RDINTCM=%08x\n", priv->rdintcm);	/*NOTE: test script uses this */
413 	WRITE_REG(priv, regRDINTCM0, priv->rdintcm);
414 	WRITE_REG(priv, regRDINTCM2, 0);	/*cpu_to_le32(rcm.val)); */
415 
416 	DBG("TDINTCM=%08x\n", priv->tdintcm);	/*NOTE: test script uses this */
417 	WRITE_REG(priv, regTDINTCM0, priv->tdintcm);	/* old val = 0x300064 */
418 
419 	/* Enable timer interrupt once in 2 secs. */
420 	/*WRITE_REG(priv, regGTMR0, ((GTMR_SEC * 2) & GTMR_DATA)); */
421 	bdx_restore_mac(priv->ndev, priv);
422 
423 	WRITE_REG(priv, regGMAC_RXF_A, GMAC_RX_FILTER_OSEN |
424 		  GMAC_RX_FILTER_AM | GMAC_RX_FILTER_AB);
425 
426 #define BDX_IRQ_TYPE	((priv->nic->irq_type == IRQ_MSI) ? 0 : IRQF_SHARED)
427 
428 	rc = request_irq(priv->pdev->irq, bdx_isr_napi, BDX_IRQ_TYPE,
429 			 ndev->name, ndev);
430 	if (rc)
431 		goto err_irq;
432 	bdx_enable_interrupts(priv);
433 
434 	RET(0);
435 
436 err_irq:
437 	RET(rc);
438 }
439 
440 static void bdx_hw_stop(struct bdx_priv *priv)
441 {
442 	ENTER;
443 	bdx_disable_interrupts(priv);
444 	free_irq(priv->pdev->irq, priv->ndev);
445 
446 	netif_carrier_off(priv->ndev);
447 	netif_stop_queue(priv->ndev);
448 
449 	RET();
450 }
451 
452 static int bdx_hw_reset_direct(void __iomem *regs)
453 {
454 	u32 val, i;
455 	ENTER;
456 
457 	/* reset sequences: read, write 1, read, write 0 */
458 	val = readl(regs + regCLKPLL);
459 	writel((val | CLKPLL_SFTRST) + 0x8, regs + regCLKPLL);
460 	udelay(50);
461 	val = readl(regs + regCLKPLL);
462 	writel(val & ~CLKPLL_SFTRST, regs + regCLKPLL);
463 
464 	/* check that the PLLs are locked and reset ended */
465 	for (i = 0; i < 70; i++, mdelay(10))
466 		if ((readl(regs + regCLKPLL) & CLKPLL_LKD) == CLKPLL_LKD) {
467 			/* do any PCI-E read transaction */
468 			readl(regs + regRXD_CFG0_0);
469 			return 0;
470 		}
471 	pr_err("HW reset failed\n");
472 	return 1;		/* failure */
473 }
474 
475 static int bdx_hw_reset(struct bdx_priv *priv)
476 {
477 	u32 val, i;
478 	ENTER;
479 
480 	if (priv->port == 0) {
481 		/* reset sequences: read, write 1, read, write 0 */
482 		val = READ_REG(priv, regCLKPLL);
483 		WRITE_REG(priv, regCLKPLL, (val | CLKPLL_SFTRST) + 0x8);
484 		udelay(50);
485 		val = READ_REG(priv, regCLKPLL);
486 		WRITE_REG(priv, regCLKPLL, val & ~CLKPLL_SFTRST);
487 	}
488 	/* check that the PLLs are locked and reset ended */
489 	for (i = 0; i < 70; i++, mdelay(10))
490 		if ((READ_REG(priv, regCLKPLL) & CLKPLL_LKD) == CLKPLL_LKD) {
491 			/* do any PCI-E read transaction */
492 			READ_REG(priv, regRXD_CFG0_0);
493 			return 0;
494 		}
495 	pr_err("HW reset failed\n");
496 	return 1;		/* failure */
497 }
498 
499 static int bdx_sw_reset(struct bdx_priv *priv)
500 {
501 	int i;
502 
503 	ENTER;
504 	/* 1. load MAC (obsolete) */
505 	/* 2. disable Rx (and Tx) */
506 	WRITE_REG(priv, regGMAC_RXF_A, 0);
507 	mdelay(100);
508 	/* 3. disable port */
509 	WRITE_REG(priv, regDIS_PORT, 1);
510 	/* 4. disable queue */
511 	WRITE_REG(priv, regDIS_QU, 1);
512 	/* 5. wait until hw is disabled */
513 	for (i = 0; i < 50; i++) {
514 		if (READ_REG(priv, regRST_PORT) & 1)
515 			break;
516 		mdelay(10);
517 	}
518 	if (i == 50)
519 		netdev_err(priv->ndev, "SW reset timeout. continuing anyway\n");
520 
521 	/* 6. disable intrs */
522 	WRITE_REG(priv, regRDINTCM0, 0);
523 	WRITE_REG(priv, regTDINTCM0, 0);
524 	WRITE_REG(priv, regIMR, 0);
525 	READ_REG(priv, regISR);
526 
527 	/* 7. reset queue */
528 	WRITE_REG(priv, regRST_QU, 1);
529 	/* 8. reset port */
530 	WRITE_REG(priv, regRST_PORT, 1);
531 	/* 9. zero all read and write pointers */
532 	for (i = regTXD_WPTR_0; i <= regTXF_RPTR_3; i += 0x10)
533 		DBG("%x = %x\n", i, READ_REG(priv, i) & TXF_WPTR_WR_PTR);
534 	for (i = regTXD_WPTR_0; i <= regTXF_RPTR_3; i += 0x10)
535 		WRITE_REG(priv, i, 0);
536 	/* 10. unseet port disable */
537 	WRITE_REG(priv, regDIS_PORT, 0);
538 	/* 11. unset queue disable */
539 	WRITE_REG(priv, regDIS_QU, 0);
540 	/* 12. unset queue reset */
541 	WRITE_REG(priv, regRST_QU, 0);
542 	/* 13. unset port reset */
543 	WRITE_REG(priv, regRST_PORT, 0);
544 	/* 14. enable Rx */
545 	/* skiped. will be done later */
546 	/* 15. save MAC (obsolete) */
547 	for (i = regTXD_WPTR_0; i <= regTXF_RPTR_3; i += 0x10)
548 		DBG("%x = %x\n", i, READ_REG(priv, i) & TXF_WPTR_WR_PTR);
549 
550 	RET(0);
551 }
552 
553 /* bdx_reset - performs right type of reset depending on hw type */
554 static int bdx_reset(struct bdx_priv *priv)
555 {
556 	ENTER;
557 	RET((priv->pdev->device == 0x3009)
558 	    ? bdx_hw_reset(priv)
559 	    : bdx_sw_reset(priv));
560 }
561 
562 /**
563  * bdx_close - Disables a network interface
564  * @ndev: network interface device structure
565  *
566  * Returns 0, this is not allowed to fail
567  *
568  * The close entry point is called when an interface is de-activated
569  * by the OS.  The hardware is still under the drivers control, but
570  * needs to be disabled.  A global MAC reset is issued to stop the
571  * hardware, and all transmit and receive resources are freed.
572  **/
573 static int bdx_close(struct net_device *ndev)
574 {
575 	struct bdx_priv *priv = NULL;
576 
577 	ENTER;
578 	priv = netdev_priv(ndev);
579 
580 	napi_disable(&priv->napi);
581 
582 	bdx_reset(priv);
583 	bdx_hw_stop(priv);
584 	bdx_rx_free(priv);
585 	bdx_tx_free(priv);
586 	RET(0);
587 }
588 
589 /**
590  * bdx_open - Called when a network interface is made active
591  * @ndev: network interface device structure
592  *
593  * Returns 0 on success, negative value on failure
594  *
595  * The open entry point is called when a network interface is made
596  * active by the system (IFF_UP).  At this point all resources needed
597  * for transmit and receive operations are allocated, the interrupt
598  * handler is registered with the OS, the watchdog timer is started,
599  * and the stack is notified that the interface is ready.
600  **/
601 static int bdx_open(struct net_device *ndev)
602 {
603 	struct bdx_priv *priv;
604 	int rc;
605 
606 	ENTER;
607 	priv = netdev_priv(ndev);
608 	bdx_reset(priv);
609 	if (netif_running(ndev))
610 		netif_stop_queue(priv->ndev);
611 
612 	if ((rc = bdx_tx_init(priv)) ||
613 	    (rc = bdx_rx_init(priv)) ||
614 	    (rc = bdx_fw_load(priv)))
615 		goto err;
616 
617 	bdx_rx_alloc_skbs(priv, &priv->rxf_fifo0);
618 
619 	rc = bdx_hw_start(priv);
620 	if (rc)
621 		goto err;
622 
623 	napi_enable(&priv->napi);
624 
625 	print_fw_id(priv->nic);
626 
627 	RET(0);
628 
629 err:
630 	bdx_close(ndev);
631 	RET(rc);
632 }
633 
634 static int bdx_range_check(struct bdx_priv *priv, u32 offset)
635 {
636 	return (offset > (u32) (BDX_REGS_SIZE / priv->nic->port_num)) ?
637 		-EINVAL : 0;
638 }
639 
640 static int bdx_siocdevprivate(struct net_device *ndev, struct ifreq *ifr,
641 			      void __user *udata, int cmd)
642 {
643 	struct bdx_priv *priv = netdev_priv(ndev);
644 	u32 data[3];
645 	int error;
646 
647 	ENTER;
648 
649 	DBG("jiffies=%ld cmd=%d\n", jiffies, cmd);
650 	if (cmd != SIOCDEVPRIVATE) {
651 		error = copy_from_user(data, udata, sizeof(data));
652 		if (error) {
653 			pr_err("can't copy from user\n");
654 			RET(-EFAULT);
655 		}
656 		DBG("%d 0x%x 0x%x\n", data[0], data[1], data[2]);
657 	} else {
658 		return -EOPNOTSUPP;
659 	}
660 
661 	if (!capable(CAP_SYS_RAWIO))
662 		return -EPERM;
663 
664 	switch (data[0]) {
665 
666 	case BDX_OP_READ:
667 		error = bdx_range_check(priv, data[1]);
668 		if (error < 0)
669 			return error;
670 		data[2] = READ_REG(priv, data[1]);
671 		DBG("read_reg(0x%x)=0x%x (dec %d)\n", data[1], data[2],
672 		    data[2]);
673 		error = copy_to_user(udata, data, sizeof(data));
674 		if (error)
675 			RET(-EFAULT);
676 		break;
677 
678 	case BDX_OP_WRITE:
679 		error = bdx_range_check(priv, data[1]);
680 		if (error < 0)
681 			return error;
682 		WRITE_REG(priv, data[1], data[2]);
683 		DBG("write_reg(0x%x, 0x%x)\n", data[1], data[2]);
684 		break;
685 
686 	default:
687 		RET(-EOPNOTSUPP);
688 	}
689 	return 0;
690 }
691 
692 /**
693  * __bdx_vlan_rx_vid - private helper for adding/killing VLAN vid
694  * @ndev: network device
695  * @vid:  VLAN vid
696  * @enable: enable or disable vlan
697  *
698  * Passes VLAN filter table to hardware
699  */
700 static void __bdx_vlan_rx_vid(struct net_device *ndev, uint16_t vid, int enable)
701 {
702 	struct bdx_priv *priv = netdev_priv(ndev);
703 	u32 reg, bit, val;
704 
705 	ENTER;
706 	DBG2("vid=%d value=%d\n", (int)vid, enable);
707 	if (unlikely(vid >= 4096)) {
708 		pr_err("invalid VID: %u (> 4096)\n", vid);
709 		RET();
710 	}
711 	reg = regVLAN_0 + (vid / 32) * 4;
712 	bit = 1 << vid % 32;
713 	val = READ_REG(priv, reg);
714 	DBG2("reg=%x, val=%x, bit=%d\n", reg, val, bit);
715 	if (enable)
716 		val |= bit;
717 	else
718 		val &= ~bit;
719 	DBG2("new val %x\n", val);
720 	WRITE_REG(priv, reg, val);
721 	RET();
722 }
723 
724 /**
725  * bdx_vlan_rx_add_vid - kernel hook for adding VLAN vid to hw filtering table
726  * @ndev: network device
727  * @proto: unused
728  * @vid:  VLAN vid to add
729  */
730 static int bdx_vlan_rx_add_vid(struct net_device *ndev, __be16 proto, u16 vid)
731 {
732 	__bdx_vlan_rx_vid(ndev, vid, 1);
733 	return 0;
734 }
735 
736 /**
737  * bdx_vlan_rx_kill_vid - kernel hook for killing VLAN vid in hw filtering table
738  * @ndev: network device
739  * @proto: unused
740  * @vid:  VLAN vid to kill
741  */
742 static int bdx_vlan_rx_kill_vid(struct net_device *ndev, __be16 proto, u16 vid)
743 {
744 	__bdx_vlan_rx_vid(ndev, vid, 0);
745 	return 0;
746 }
747 
748 /**
749  * bdx_change_mtu - Change the Maximum Transfer Unit
750  * @ndev: network interface device structure
751  * @new_mtu: new value for maximum frame size
752  *
753  * Returns 0 on success, negative on failure
754  */
755 static int bdx_change_mtu(struct net_device *ndev, int new_mtu)
756 {
757 	ENTER;
758 
759 	ndev->mtu = new_mtu;
760 	if (netif_running(ndev)) {
761 		bdx_close(ndev);
762 		bdx_open(ndev);
763 	}
764 	RET(0);
765 }
766 
767 static void bdx_setmulti(struct net_device *ndev)
768 {
769 	struct bdx_priv *priv = netdev_priv(ndev);
770 
771 	u32 rxf_val =
772 	    GMAC_RX_FILTER_AM | GMAC_RX_FILTER_AB | GMAC_RX_FILTER_OSEN;
773 	int i;
774 
775 	ENTER;
776 	/* IMF - imperfect (hash) rx multicat filter */
777 	/* PMF - perfect rx multicat filter */
778 
779 	/* FIXME: RXE(OFF) */
780 	if (ndev->flags & IFF_PROMISC) {
781 		rxf_val |= GMAC_RX_FILTER_PRM;
782 	} else if (ndev->flags & IFF_ALLMULTI) {
783 		/* set IMF to accept all multicast frmaes */
784 		for (i = 0; i < MAC_MCST_HASH_NUM; i++)
785 			WRITE_REG(priv, regRX_MCST_HASH0 + i * 4, ~0);
786 	} else if (!netdev_mc_empty(ndev)) {
787 		u8 hash;
788 		struct netdev_hw_addr *ha;
789 		u32 reg, val;
790 
791 		/* set IMF to deny all multicast frames */
792 		for (i = 0; i < MAC_MCST_HASH_NUM; i++)
793 			WRITE_REG(priv, regRX_MCST_HASH0 + i * 4, 0);
794 		/* set PMF to deny all multicast frames */
795 		for (i = 0; i < MAC_MCST_NUM; i++) {
796 			WRITE_REG(priv, regRX_MAC_MCST0 + i * 8, 0);
797 			WRITE_REG(priv, regRX_MAC_MCST1 + i * 8, 0);
798 		}
799 
800 		/* use PMF to accept first MAC_MCST_NUM (15) addresses */
801 		/* TBD: sort addresses and write them in ascending order
802 		 * into RX_MAC_MCST regs. we skip this phase now and accept ALL
803 		 * multicast frames throu IMF */
804 		/* accept the rest of addresses throu IMF */
805 		netdev_for_each_mc_addr(ha, ndev) {
806 			hash = 0;
807 			for (i = 0; i < ETH_ALEN; i++)
808 				hash ^= ha->addr[i];
809 			reg = regRX_MCST_HASH0 + ((hash >> 5) << 2);
810 			val = READ_REG(priv, reg);
811 			val |= (1 << (hash % 32));
812 			WRITE_REG(priv, reg, val);
813 		}
814 
815 	} else {
816 		DBG("only own mac %d\n", netdev_mc_count(ndev));
817 		rxf_val |= GMAC_RX_FILTER_AB;
818 	}
819 	WRITE_REG(priv, regGMAC_RXF_A, rxf_val);
820 	/* enable RX */
821 	/* FIXME: RXE(ON) */
822 	RET();
823 }
824 
825 static int bdx_set_mac(struct net_device *ndev, void *p)
826 {
827 	struct bdx_priv *priv = netdev_priv(ndev);
828 	struct sockaddr *addr = p;
829 
830 	ENTER;
831 	/*
832 	   if (netif_running(dev))
833 	   return -EBUSY
834 	 */
835 	eth_hw_addr_set(ndev, addr->sa_data);
836 	bdx_restore_mac(ndev, priv);
837 	RET(0);
838 }
839 
840 static int bdx_read_mac(struct bdx_priv *priv)
841 {
842 	u16 macAddress[3], i;
843 	u8 addr[ETH_ALEN];
844 	ENTER;
845 
846 	macAddress[2] = READ_REG(priv, regUNC_MAC0_A);
847 	macAddress[2] = READ_REG(priv, regUNC_MAC0_A);
848 	macAddress[1] = READ_REG(priv, regUNC_MAC1_A);
849 	macAddress[1] = READ_REG(priv, regUNC_MAC1_A);
850 	macAddress[0] = READ_REG(priv, regUNC_MAC2_A);
851 	macAddress[0] = READ_REG(priv, regUNC_MAC2_A);
852 	for (i = 0; i < 3; i++) {
853 		addr[i * 2 + 1] = macAddress[i];
854 		addr[i * 2] = macAddress[i] >> 8;
855 	}
856 	eth_hw_addr_set(priv->ndev, addr);
857 	RET(0);
858 }
859 
860 static u64 bdx_read_l2stat(struct bdx_priv *priv, int reg)
861 {
862 	u64 val;
863 
864 	val = READ_REG(priv, reg);
865 	val |= ((u64) READ_REG(priv, reg + 8)) << 32;
866 	return val;
867 }
868 
869 /*Do the statistics-update work*/
870 static void bdx_update_stats(struct bdx_priv *priv)
871 {
872 	struct bdx_stats *stats = &priv->hw_stats;
873 	u64 *stats_vector = (u64 *) stats;
874 	int i;
875 	int addr;
876 
877 	/*Fill HW structure */
878 	addr = 0x7200;
879 	/*First 12 statistics - 0x7200 - 0x72B0 */
880 	for (i = 0; i < 12; i++) {
881 		stats_vector[i] = bdx_read_l2stat(priv, addr);
882 		addr += 0x10;
883 	}
884 	BDX_ASSERT(addr != 0x72C0);
885 	/* 0x72C0-0x72E0 RSRV */
886 	addr = 0x72F0;
887 	for (; i < 16; i++) {
888 		stats_vector[i] = bdx_read_l2stat(priv, addr);
889 		addr += 0x10;
890 	}
891 	BDX_ASSERT(addr != 0x7330);
892 	/* 0x7330-0x7360 RSRV */
893 	addr = 0x7370;
894 	for (; i < 19; i++) {
895 		stats_vector[i] = bdx_read_l2stat(priv, addr);
896 		addr += 0x10;
897 	}
898 	BDX_ASSERT(addr != 0x73A0);
899 	/* 0x73A0-0x73B0 RSRV */
900 	addr = 0x73C0;
901 	for (; i < 23; i++) {
902 		stats_vector[i] = bdx_read_l2stat(priv, addr);
903 		addr += 0x10;
904 	}
905 	BDX_ASSERT(addr != 0x7400);
906 	BDX_ASSERT((sizeof(struct bdx_stats) / sizeof(u64)) != i);
907 }
908 
909 static void print_rxdd(struct rxd_desc *rxdd, u32 rxd_val1, u16 len,
910 		       u16 rxd_vlan);
911 static void print_rxfd(struct rxf_desc *rxfd);
912 
913 /*************************************************************************
914  *     Rx DB                                                             *
915  *************************************************************************/
916 
917 static void bdx_rxdb_destroy(struct rxdb *db)
918 {
919 	vfree(db);
920 }
921 
922 static struct rxdb *bdx_rxdb_create(int nelem)
923 {
924 	struct rxdb *db;
925 	int i;
926 
927 	db = vmalloc(sizeof(struct rxdb)
928 		     + (nelem * sizeof(int))
929 		     + (nelem * sizeof(struct rx_map)));
930 	if (likely(db != NULL)) {
931 		db->stack = (int *)(db + 1);
932 		db->elems = (void *)(db->stack + nelem);
933 		db->nelem = nelem;
934 		db->top = nelem;
935 		for (i = 0; i < nelem; i++)
936 			db->stack[i] = nelem - i - 1;	/* to make first allocs
937 							   close to db struct*/
938 	}
939 
940 	return db;
941 }
942 
943 static inline int bdx_rxdb_alloc_elem(struct rxdb *db)
944 {
945 	BDX_ASSERT(db->top <= 0);
946 	return db->stack[--(db->top)];
947 }
948 
949 static inline void *bdx_rxdb_addr_elem(struct rxdb *db, int n)
950 {
951 	BDX_ASSERT((n < 0) || (n >= db->nelem));
952 	return db->elems + n;
953 }
954 
955 static inline int bdx_rxdb_available(struct rxdb *db)
956 {
957 	return db->top;
958 }
959 
960 static inline void bdx_rxdb_free_elem(struct rxdb *db, int n)
961 {
962 	BDX_ASSERT((n >= db->nelem) || (n < 0));
963 	db->stack[(db->top)++] = n;
964 }
965 
966 /*************************************************************************
967  *     Rx Init                                                           *
968  *************************************************************************/
969 
970 /**
971  * bdx_rx_init - initialize RX all related HW and SW resources
972  * @priv: NIC private structure
973  *
974  * Returns 0 on success, negative value on failure
975  *
976  * It creates rxf and rxd fifos, update relevant HW registers, preallocate
977  * skb for rx. It assumes that Rx is desabled in HW
978  * funcs are grouped for better cache usage
979  *
980  * RxD fifo is smaller than RxF fifo by design. Upon high load, RxD will be
981  * filled and packets will be dropped by nic without getting into host or
982  * cousing interrupt. Anyway, in that condition, host has no chance to process
983  * all packets, but dropping in nic is cheaper, since it takes 0 cpu cycles
984  */
985 
986 /* TBD: ensure proper packet size */
987 
988 static int bdx_rx_init(struct bdx_priv *priv)
989 {
990 	ENTER;
991 
992 	if (bdx_fifo_init(priv, &priv->rxd_fifo0.m, priv->rxd_size,
993 			  regRXD_CFG0_0, regRXD_CFG1_0,
994 			  regRXD_RPTR_0, regRXD_WPTR_0))
995 		goto err_mem;
996 	if (bdx_fifo_init(priv, &priv->rxf_fifo0.m, priv->rxf_size,
997 			  regRXF_CFG0_0, regRXF_CFG1_0,
998 			  regRXF_RPTR_0, regRXF_WPTR_0))
999 		goto err_mem;
1000 	priv->rxdb = bdx_rxdb_create(priv->rxf_fifo0.m.memsz /
1001 				     sizeof(struct rxf_desc));
1002 	if (!priv->rxdb)
1003 		goto err_mem;
1004 
1005 	priv->rxf_fifo0.m.pktsz = priv->ndev->mtu + VLAN_ETH_HLEN;
1006 	return 0;
1007 
1008 err_mem:
1009 	netdev_err(priv->ndev, "Rx init failed\n");
1010 	return -ENOMEM;
1011 }
1012 
1013 /**
1014  * bdx_rx_free_skbs - frees and unmaps all skbs allocated for the fifo
1015  * @priv: NIC private structure
1016  * @f: RXF fifo
1017  */
1018 static void bdx_rx_free_skbs(struct bdx_priv *priv, struct rxf_fifo *f)
1019 {
1020 	struct rx_map *dm;
1021 	struct rxdb *db = priv->rxdb;
1022 	u16 i;
1023 
1024 	ENTER;
1025 	DBG("total=%d free=%d busy=%d\n", db->nelem, bdx_rxdb_available(db),
1026 	    db->nelem - bdx_rxdb_available(db));
1027 	while (bdx_rxdb_available(db) > 0) {
1028 		i = bdx_rxdb_alloc_elem(db);
1029 		dm = bdx_rxdb_addr_elem(db, i);
1030 		dm->dma = 0;
1031 	}
1032 	for (i = 0; i < db->nelem; i++) {
1033 		dm = bdx_rxdb_addr_elem(db, i);
1034 		if (dm->dma) {
1035 			dma_unmap_single(&priv->pdev->dev, dm->dma,
1036 					 f->m.pktsz, DMA_FROM_DEVICE);
1037 			dev_kfree_skb(dm->skb);
1038 		}
1039 	}
1040 }
1041 
1042 /**
1043  * bdx_rx_free - release all Rx resources
1044  * @priv: NIC private structure
1045  *
1046  * It assumes that Rx is desabled in HW
1047  */
1048 static void bdx_rx_free(struct bdx_priv *priv)
1049 {
1050 	ENTER;
1051 	if (priv->rxdb) {
1052 		bdx_rx_free_skbs(priv, &priv->rxf_fifo0);
1053 		bdx_rxdb_destroy(priv->rxdb);
1054 		priv->rxdb = NULL;
1055 	}
1056 	bdx_fifo_free(priv, &priv->rxf_fifo0.m);
1057 	bdx_fifo_free(priv, &priv->rxd_fifo0.m);
1058 
1059 	RET();
1060 }
1061 
1062 /*************************************************************************
1063  *     Rx Engine                                                         *
1064  *************************************************************************/
1065 
1066 /**
1067  * bdx_rx_alloc_skbs - fill rxf fifo with new skbs
1068  * @priv: nic's private structure
1069  * @f: RXF fifo that needs skbs
1070  *
1071  * It allocates skbs, build rxf descs and push it (rxf descr) into rxf fifo.
1072  * skb's virtual and physical addresses are stored in skb db.
1073  * To calculate free space, func uses cached values of RPTR and WPTR
1074  * When needed, it also updates RPTR and WPTR.
1075  */
1076 
1077 /* TBD: do not update WPTR if no desc were written */
1078 
1079 static void bdx_rx_alloc_skbs(struct bdx_priv *priv, struct rxf_fifo *f)
1080 {
1081 	struct sk_buff *skb;
1082 	struct rxf_desc *rxfd;
1083 	struct rx_map *dm;
1084 	int dno, delta, idx;
1085 	struct rxdb *db = priv->rxdb;
1086 
1087 	ENTER;
1088 	dno = bdx_rxdb_available(db) - 1;
1089 	while (dno > 0) {
1090 		skb = netdev_alloc_skb(priv->ndev, f->m.pktsz + NET_IP_ALIGN);
1091 		if (!skb)
1092 			break;
1093 
1094 		skb_reserve(skb, NET_IP_ALIGN);
1095 
1096 		idx = bdx_rxdb_alloc_elem(db);
1097 		dm = bdx_rxdb_addr_elem(db, idx);
1098 		dm->dma = dma_map_single(&priv->pdev->dev, skb->data,
1099 					 f->m.pktsz, DMA_FROM_DEVICE);
1100 		dm->skb = skb;
1101 		rxfd = (struct rxf_desc *)(f->m.va + f->m.wptr);
1102 		rxfd->info = CPU_CHIP_SWAP32(0x10003);	/* INFO=1 BC=3 */
1103 		rxfd->va_lo = idx;
1104 		rxfd->pa_lo = CPU_CHIP_SWAP32(L32_64(dm->dma));
1105 		rxfd->pa_hi = CPU_CHIP_SWAP32(H32_64(dm->dma));
1106 		rxfd->len = CPU_CHIP_SWAP32(f->m.pktsz);
1107 		print_rxfd(rxfd);
1108 
1109 		f->m.wptr += sizeof(struct rxf_desc);
1110 		delta = f->m.wptr - f->m.memsz;
1111 		if (unlikely(delta >= 0)) {
1112 			f->m.wptr = delta;
1113 			if (delta > 0) {
1114 				memcpy(f->m.va, f->m.va + f->m.memsz, delta);
1115 				DBG("wrapped descriptor\n");
1116 			}
1117 		}
1118 		dno--;
1119 	}
1120 	/*TBD: to do - delayed rxf wptr like in txd */
1121 	WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1122 	RET();
1123 }
1124 
1125 static inline void
1126 NETIF_RX_MUX(struct bdx_priv *priv, u32 rxd_val1, u16 rxd_vlan,
1127 	     struct sk_buff *skb)
1128 {
1129 	ENTER;
1130 	DBG("rxdd->flags.bits.vtag=%d\n", GET_RXD_VTAG(rxd_val1));
1131 	if (GET_RXD_VTAG(rxd_val1)) {
1132 		DBG("%s: vlan rcv vlan '%x' vtag '%x'\n",
1133 		    priv->ndev->name,
1134 		    GET_RXD_VLAN_ID(rxd_vlan),
1135 		    GET_RXD_VTAG(rxd_val1));
1136 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), GET_RXD_VLAN_TCI(rxd_vlan));
1137 	}
1138 	netif_receive_skb(skb);
1139 }
1140 
1141 static void bdx_recycle_skb(struct bdx_priv *priv, struct rxd_desc *rxdd)
1142 {
1143 	struct rxf_desc *rxfd;
1144 	struct rx_map *dm;
1145 	struct rxf_fifo *f;
1146 	struct rxdb *db;
1147 	int delta;
1148 
1149 	ENTER;
1150 	DBG("priv=%p rxdd=%p\n", priv, rxdd);
1151 	f = &priv->rxf_fifo0;
1152 	db = priv->rxdb;
1153 	DBG("db=%p f=%p\n", db, f);
1154 	dm = bdx_rxdb_addr_elem(db, rxdd->va_lo);
1155 	DBG("dm=%p\n", dm);
1156 	rxfd = (struct rxf_desc *)(f->m.va + f->m.wptr);
1157 	rxfd->info = CPU_CHIP_SWAP32(0x10003);	/* INFO=1 BC=3 */
1158 	rxfd->va_lo = rxdd->va_lo;
1159 	rxfd->pa_lo = CPU_CHIP_SWAP32(L32_64(dm->dma));
1160 	rxfd->pa_hi = CPU_CHIP_SWAP32(H32_64(dm->dma));
1161 	rxfd->len = CPU_CHIP_SWAP32(f->m.pktsz);
1162 	print_rxfd(rxfd);
1163 
1164 	f->m.wptr += sizeof(struct rxf_desc);
1165 	delta = f->m.wptr - f->m.memsz;
1166 	if (unlikely(delta >= 0)) {
1167 		f->m.wptr = delta;
1168 		if (delta > 0) {
1169 			memcpy(f->m.va, f->m.va + f->m.memsz, delta);
1170 			DBG("wrapped descriptor\n");
1171 		}
1172 	}
1173 	RET();
1174 }
1175 
1176 /**
1177  * bdx_rx_receive - receives full packets from RXD fifo and pass them to OS
1178  * NOTE: a special treatment is given to non-continuous descriptors
1179  * that start near the end, wraps around and continue at the beginning. a second
1180  * part is copied right after the first, and then descriptor is interpreted as
1181  * normal. fifo has an extra space to allow such operations
1182  * @priv: nic's private structure
1183  * @f: RXF fifo that needs skbs
1184  * @budget: maximum number of packets to receive
1185  */
1186 
1187 /* TBD: replace memcpy func call by explicite inline asm */
1188 
1189 static int bdx_rx_receive(struct bdx_priv *priv, struct rxd_fifo *f, int budget)
1190 {
1191 	struct net_device *ndev = priv->ndev;
1192 	struct sk_buff *skb, *skb2;
1193 	struct rxd_desc *rxdd;
1194 	struct rx_map *dm;
1195 	struct rxf_fifo *rxf_fifo;
1196 	int tmp_len, size;
1197 	int done = 0;
1198 	int max_done = BDX_MAX_RX_DONE;
1199 	struct rxdb *db = NULL;
1200 	/* Unmarshalled descriptor - copy of descriptor in host order */
1201 	u32 rxd_val1;
1202 	u16 len;
1203 	u16 rxd_vlan;
1204 
1205 	ENTER;
1206 	max_done = budget;
1207 
1208 	f->m.wptr = READ_REG(priv, f->m.reg_WPTR) & TXF_WPTR_WR_PTR;
1209 
1210 	size = f->m.wptr - f->m.rptr;
1211 	if (size < 0)
1212 		size = f->m.memsz + size;	/* size is negative :-) */
1213 
1214 	while (size > 0) {
1215 
1216 		rxdd = (struct rxd_desc *)(f->m.va + f->m.rptr);
1217 		rxd_val1 = CPU_CHIP_SWAP32(rxdd->rxd_val1);
1218 
1219 		len = CPU_CHIP_SWAP16(rxdd->len);
1220 
1221 		rxd_vlan = CPU_CHIP_SWAP16(rxdd->rxd_vlan);
1222 
1223 		print_rxdd(rxdd, rxd_val1, len, rxd_vlan);
1224 
1225 		tmp_len = GET_RXD_BC(rxd_val1) << 3;
1226 		BDX_ASSERT(tmp_len <= 0);
1227 		size -= tmp_len;
1228 		if (size < 0)	/* test for partially arrived descriptor */
1229 			break;
1230 
1231 		f->m.rptr += tmp_len;
1232 
1233 		tmp_len = f->m.rptr - f->m.memsz;
1234 		if (unlikely(tmp_len >= 0)) {
1235 			f->m.rptr = tmp_len;
1236 			if (tmp_len > 0) {
1237 				DBG("wrapped desc rptr=%d tmp_len=%d\n",
1238 				    f->m.rptr, tmp_len);
1239 				memcpy(f->m.va + f->m.memsz, f->m.va, tmp_len);
1240 			}
1241 		}
1242 
1243 		if (unlikely(GET_RXD_ERR(rxd_val1))) {
1244 			DBG("rxd_err = 0x%x\n", GET_RXD_ERR(rxd_val1));
1245 			ndev->stats.rx_errors++;
1246 			bdx_recycle_skb(priv, rxdd);
1247 			continue;
1248 		}
1249 
1250 		rxf_fifo = &priv->rxf_fifo0;
1251 		db = priv->rxdb;
1252 		dm = bdx_rxdb_addr_elem(db, rxdd->va_lo);
1253 		skb = dm->skb;
1254 
1255 		if (len < BDX_COPYBREAK &&
1256 		    (skb2 = netdev_alloc_skb(priv->ndev, len + NET_IP_ALIGN))) {
1257 			skb_reserve(skb2, NET_IP_ALIGN);
1258 			/*skb_put(skb2, len); */
1259 			dma_sync_single_for_cpu(&priv->pdev->dev, dm->dma,
1260 						rxf_fifo->m.pktsz,
1261 						DMA_FROM_DEVICE);
1262 			memcpy(skb2->data, skb->data, len);
1263 			bdx_recycle_skb(priv, rxdd);
1264 			skb = skb2;
1265 		} else {
1266 			dma_unmap_single(&priv->pdev->dev, dm->dma,
1267 					 rxf_fifo->m.pktsz, DMA_FROM_DEVICE);
1268 			bdx_rxdb_free_elem(db, rxdd->va_lo);
1269 		}
1270 
1271 		ndev->stats.rx_bytes += len;
1272 
1273 		skb_put(skb, len);
1274 		skb->protocol = eth_type_trans(skb, ndev);
1275 
1276 		/* Non-IP packets aren't checksum-offloaded */
1277 		if (GET_RXD_PKT_ID(rxd_val1) == 0)
1278 			skb_checksum_none_assert(skb);
1279 		else
1280 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1281 
1282 		NETIF_RX_MUX(priv, rxd_val1, rxd_vlan, skb);
1283 
1284 		if (++done >= max_done)
1285 			break;
1286 	}
1287 
1288 	ndev->stats.rx_packets += done;
1289 
1290 	/* FIXME: do smth to minimize pci accesses    */
1291 	WRITE_REG(priv, f->m.reg_RPTR, f->m.rptr & TXF_WPTR_WR_PTR);
1292 
1293 	bdx_rx_alloc_skbs(priv, &priv->rxf_fifo0);
1294 
1295 	RET(done);
1296 }
1297 
1298 /*************************************************************************
1299  * Debug / Temprorary Code                                               *
1300  *************************************************************************/
1301 static void print_rxdd(struct rxd_desc *rxdd, u32 rxd_val1, u16 len,
1302 		       u16 rxd_vlan)
1303 {
1304 	DBG("ERROR: rxdd bc %d rxfq %d to %d type %d err %d rxp %d pkt_id %d vtag %d len %d vlan_id %d cfi %d prio %d va_lo %d va_hi %d\n",
1305 	    GET_RXD_BC(rxd_val1), GET_RXD_RXFQ(rxd_val1), GET_RXD_TO(rxd_val1),
1306 	    GET_RXD_TYPE(rxd_val1), GET_RXD_ERR(rxd_val1),
1307 	    GET_RXD_RXP(rxd_val1), GET_RXD_PKT_ID(rxd_val1),
1308 	    GET_RXD_VTAG(rxd_val1), len, GET_RXD_VLAN_ID(rxd_vlan),
1309 	    GET_RXD_CFI(rxd_vlan), GET_RXD_PRIO(rxd_vlan), rxdd->va_lo,
1310 	    rxdd->va_hi);
1311 }
1312 
1313 static void print_rxfd(struct rxf_desc *rxfd)
1314 {
1315 	DBG("=== RxF desc CHIP ORDER/ENDIANNESS =============\n"
1316 	    "info 0x%x va_lo %u pa_lo 0x%x pa_hi 0x%x len 0x%x\n",
1317 	    rxfd->info, rxfd->va_lo, rxfd->pa_lo, rxfd->pa_hi, rxfd->len);
1318 }
1319 
1320 /*
1321  * TX HW/SW interaction overview
1322  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1323  * There are 2 types of TX communication channels between driver and NIC.
1324  * 1) TX Free Fifo - TXF - holds ack descriptors for sent packets
1325  * 2) TX Data Fifo - TXD - holds descriptors of full buffers.
1326  *
1327  * Currently NIC supports TSO, checksuming and gather DMA
1328  * UFO and IP fragmentation is on the way
1329  *
1330  * RX SW Data Structures
1331  * ~~~~~~~~~~~~~~~~~~~~~
1332  * txdb - used to keep track of all skbs owned by SW and their dma addresses.
1333  * For TX case, ownership lasts from geting packet via hard_xmit and until HW
1334  * acknowledges sent by TXF descriptors.
1335  * Implemented as cyclic buffer.
1336  * fifo - keeps info about fifo's size and location, relevant HW registers,
1337  * usage and skb db. Each RXD and RXF Fifo has its own fifo structure.
1338  * Implemented as simple struct.
1339  *
1340  * TX SW Execution Flow
1341  * ~~~~~~~~~~~~~~~~~~~~
1342  * OS calls driver's hard_xmit method with packet to sent.
1343  * Driver creates DMA mappings, builds TXD descriptors and kicks HW
1344  * by updating TXD WPTR.
1345  * When packet is sent, HW write us TXF descriptor and SW frees original skb.
1346  * To prevent TXD fifo overflow without reading HW registers every time,
1347  * SW deploys "tx level" technique.
1348  * Upon strart up, tx level is initialized to TXD fifo length.
1349  * For every sent packet, SW gets its TXD descriptor sizei
1350  * (from precalculated array) and substructs it from tx level.
1351  * The size is also stored in txdb. When TXF ack arrives, SW fetch size of
1352  * original TXD descriptor from txdb and adds it to tx level.
1353  * When Tx level drops under some predefined treshhold, the driver
1354  * stops the TX queue. When TX level rises above that level,
1355  * the tx queue is enabled again.
1356  *
1357  * This technique avoids eccessive reading of RPTR and WPTR registers.
1358  * As our benchmarks shows, it adds 1.5 Gbit/sec to NIS's throuput.
1359  */
1360 
1361 /**
1362  * __bdx_tx_db_ptr_next - helper function, increment read/write pointer + wrap
1363  * @db: tx data base
1364  * @pptr: read or write pointer
1365  */
1366 static inline void __bdx_tx_db_ptr_next(struct txdb *db, struct tx_map **pptr)
1367 {
1368 	BDX_ASSERT(db == NULL || pptr == NULL);	/* sanity */
1369 
1370 	BDX_ASSERT(*pptr != db->rptr &&	/* expect either read */
1371 		   *pptr != db->wptr);	/* or write pointer */
1372 
1373 	BDX_ASSERT(*pptr < db->start ||	/* pointer has to be */
1374 		   *pptr >= db->end);	/* in range */
1375 
1376 	++*pptr;
1377 	if (unlikely(*pptr == db->end))
1378 		*pptr = db->start;
1379 }
1380 
1381 /**
1382  * bdx_tx_db_inc_rptr - increment read pointer
1383  * @db: tx data base
1384  */
1385 static inline void bdx_tx_db_inc_rptr(struct txdb *db)
1386 {
1387 	BDX_ASSERT(db->rptr == db->wptr);	/* can't read from empty db */
1388 	__bdx_tx_db_ptr_next(db, &db->rptr);
1389 }
1390 
1391 /**
1392  * bdx_tx_db_inc_wptr - increment write pointer
1393  * @db: tx data base
1394  */
1395 static inline void bdx_tx_db_inc_wptr(struct txdb *db)
1396 {
1397 	__bdx_tx_db_ptr_next(db, &db->wptr);
1398 	BDX_ASSERT(db->rptr == db->wptr);	/* we can not get empty db as
1399 						   a result of write */
1400 }
1401 
1402 /**
1403  * bdx_tx_db_init - creates and initializes tx db
1404  * @d: tx data base
1405  * @sz_type: size of tx fifo
1406  *
1407  * Returns 0 on success, error code otherwise
1408  */
1409 static int bdx_tx_db_init(struct txdb *d, int sz_type)
1410 {
1411 	int memsz = FIFO_SIZE * (1 << (sz_type + 1));
1412 
1413 	d->start = vmalloc(memsz);
1414 	if (!d->start)
1415 		return -ENOMEM;
1416 
1417 	/*
1418 	 * In order to differentiate between db is empty and db is full
1419 	 * states at least one element should always be empty in order to
1420 	 * avoid rptr == wptr which means db is empty
1421 	 */
1422 	d->size = memsz / sizeof(struct tx_map) - 1;
1423 	d->end = d->start + d->size + 1;	/* just after last element */
1424 
1425 	/* all dbs are created equally empty */
1426 	d->rptr = d->start;
1427 	d->wptr = d->start;
1428 
1429 	return 0;
1430 }
1431 
1432 /**
1433  * bdx_tx_db_close - closes tx db and frees all memory
1434  * @d: tx data base
1435  */
1436 static void bdx_tx_db_close(struct txdb *d)
1437 {
1438 	BDX_ASSERT(d == NULL);
1439 
1440 	vfree(d->start);
1441 	d->start = NULL;
1442 }
1443 
1444 /*************************************************************************
1445  *     Tx Engine                                                         *
1446  *************************************************************************/
1447 
1448 /* sizes of tx desc (including padding if needed) as function
1449  * of skb's frag number */
1450 static struct {
1451 	u16 bytes;
1452 	u16 qwords;		/* qword = 64 bit */
1453 } txd_sizes[MAX_SKB_FRAGS + 1];
1454 
1455 /**
1456  * bdx_tx_map_skb - creates and stores dma mappings for skb's data blocks
1457  * @priv: NIC private structure
1458  * @skb: socket buffer to map
1459  * @txdd: TX descriptor to use
1460  *
1461  * It makes dma mappings for skb's data blocks and writes them to PBL of
1462  * new tx descriptor. It also stores them in the tx db, so they could be
1463  * unmaped after data was sent. It is reponsibility of a caller to make
1464  * sure that there is enough space in the tx db. Last element holds pointer
1465  * to skb itself and marked with zero length
1466  */
1467 static inline void
1468 bdx_tx_map_skb(struct bdx_priv *priv, struct sk_buff *skb,
1469 	       struct txd_desc *txdd)
1470 {
1471 	struct txdb *db = &priv->txdb;
1472 	struct pbl *pbl = &txdd->pbl[0];
1473 	int nr_frags = skb_shinfo(skb)->nr_frags;
1474 	int i;
1475 
1476 	db->wptr->len = skb_headlen(skb);
1477 	db->wptr->addr.dma = dma_map_single(&priv->pdev->dev, skb->data,
1478 					    db->wptr->len, DMA_TO_DEVICE);
1479 	pbl->len = CPU_CHIP_SWAP32(db->wptr->len);
1480 	pbl->pa_lo = CPU_CHIP_SWAP32(L32_64(db->wptr->addr.dma));
1481 	pbl->pa_hi = CPU_CHIP_SWAP32(H32_64(db->wptr->addr.dma));
1482 	DBG("=== pbl   len: 0x%x ================\n", pbl->len);
1483 	DBG("=== pbl pa_lo: 0x%x ================\n", pbl->pa_lo);
1484 	DBG("=== pbl pa_hi: 0x%x ================\n", pbl->pa_hi);
1485 	bdx_tx_db_inc_wptr(db);
1486 
1487 	for (i = 0; i < nr_frags; i++) {
1488 		const skb_frag_t *frag;
1489 
1490 		frag = &skb_shinfo(skb)->frags[i];
1491 		db->wptr->len = skb_frag_size(frag);
1492 		db->wptr->addr.dma = skb_frag_dma_map(&priv->pdev->dev, frag,
1493 						      0, skb_frag_size(frag),
1494 						      DMA_TO_DEVICE);
1495 
1496 		pbl++;
1497 		pbl->len = CPU_CHIP_SWAP32(db->wptr->len);
1498 		pbl->pa_lo = CPU_CHIP_SWAP32(L32_64(db->wptr->addr.dma));
1499 		pbl->pa_hi = CPU_CHIP_SWAP32(H32_64(db->wptr->addr.dma));
1500 		bdx_tx_db_inc_wptr(db);
1501 	}
1502 
1503 	/* add skb clean up info. */
1504 	db->wptr->len = -txd_sizes[nr_frags].bytes;
1505 	db->wptr->addr.skb = skb;
1506 	bdx_tx_db_inc_wptr(db);
1507 }
1508 
1509 /* init_txd_sizes - precalculate sizes of descriptors for skbs up to 16 frags
1510  * number of frags is used as index to fetch correct descriptors size,
1511  * instead of calculating it each time */
1512 static void __init init_txd_sizes(void)
1513 {
1514 	int i, lwords;
1515 
1516 	/* 7 - is number of lwords in txd with one phys buffer
1517 	 * 3 - is number of lwords used for every additional phys buffer */
1518 	for (i = 0; i < MAX_SKB_FRAGS + 1; i++) {
1519 		lwords = 7 + (i * 3);
1520 		if (lwords & 1)
1521 			lwords++;	/* pad it with 1 lword */
1522 		txd_sizes[i].qwords = lwords >> 1;
1523 		txd_sizes[i].bytes = lwords << 2;
1524 	}
1525 }
1526 
1527 /* bdx_tx_init - initialize all Tx related stuff.
1528  * Namely, TXD and TXF fifos, database etc */
1529 static int bdx_tx_init(struct bdx_priv *priv)
1530 {
1531 	if (bdx_fifo_init(priv, &priv->txd_fifo0.m, priv->txd_size,
1532 			  regTXD_CFG0_0,
1533 			  regTXD_CFG1_0, regTXD_RPTR_0, regTXD_WPTR_0))
1534 		goto err_mem;
1535 	if (bdx_fifo_init(priv, &priv->txf_fifo0.m, priv->txf_size,
1536 			  regTXF_CFG0_0,
1537 			  regTXF_CFG1_0, regTXF_RPTR_0, regTXF_WPTR_0))
1538 		goto err_mem;
1539 
1540 	/* The TX db has to keep mappings for all packets sent (on TxD)
1541 	 * and not yet reclaimed (on TxF) */
1542 	if (bdx_tx_db_init(&priv->txdb, max(priv->txd_size, priv->txf_size)))
1543 		goto err_mem;
1544 
1545 	priv->tx_level = BDX_MAX_TX_LEVEL;
1546 #ifdef BDX_DELAY_WPTR
1547 	priv->tx_update_mark = priv->tx_level - 1024;
1548 #endif
1549 	return 0;
1550 
1551 err_mem:
1552 	netdev_err(priv->ndev, "Tx init failed\n");
1553 	return -ENOMEM;
1554 }
1555 
1556 /**
1557  * bdx_tx_space - calculates available space in TX fifo
1558  * @priv: NIC private structure
1559  *
1560  * Returns available space in TX fifo in bytes
1561  */
1562 static inline int bdx_tx_space(struct bdx_priv *priv)
1563 {
1564 	struct txd_fifo *f = &priv->txd_fifo0;
1565 	int fsize;
1566 
1567 	f->m.rptr = READ_REG(priv, f->m.reg_RPTR) & TXF_WPTR_WR_PTR;
1568 	fsize = f->m.rptr - f->m.wptr;
1569 	if (fsize <= 0)
1570 		fsize = f->m.memsz + fsize;
1571 	return fsize;
1572 }
1573 
1574 /**
1575  * bdx_tx_transmit - send packet to NIC
1576  * @skb: packet to send
1577  * @ndev: network device assigned to NIC
1578  * Return codes:
1579  * o NETDEV_TX_OK everything ok.
1580  * o NETDEV_TX_BUSY Cannot transmit packet, try later
1581  *   Usually a bug, means queue start/stop flow control is broken in
1582  *   the driver. Note: the driver must NOT put the skb in its DMA ring.
1583  */
1584 static netdev_tx_t bdx_tx_transmit(struct sk_buff *skb,
1585 				   struct net_device *ndev)
1586 {
1587 	struct bdx_priv *priv = netdev_priv(ndev);
1588 	struct txd_fifo *f = &priv->txd_fifo0;
1589 	int txd_checksum = 7;	/* full checksum */
1590 	int txd_lgsnd = 0;
1591 	int txd_vlan_id = 0;
1592 	int txd_vtag = 0;
1593 	int txd_mss = 0;
1594 
1595 	int nr_frags = skb_shinfo(skb)->nr_frags;
1596 	struct txd_desc *txdd;
1597 	int len;
1598 	unsigned long flags;
1599 
1600 	ENTER;
1601 	local_irq_save(flags);
1602 	spin_lock(&priv->tx_lock);
1603 
1604 	/* build tx descriptor */
1605 	BDX_ASSERT(f->m.wptr >= f->m.memsz);	/* started with valid wptr */
1606 	txdd = (struct txd_desc *)(f->m.va + f->m.wptr);
1607 	if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL))
1608 		txd_checksum = 0;
1609 
1610 	if (skb_shinfo(skb)->gso_size) {
1611 		txd_mss = skb_shinfo(skb)->gso_size;
1612 		txd_lgsnd = 1;
1613 		DBG("skb %p skb len %d gso size = %d\n", skb, skb->len,
1614 		    txd_mss);
1615 	}
1616 
1617 	if (skb_vlan_tag_present(skb)) {
1618 		/*Cut VLAN ID to 12 bits */
1619 		txd_vlan_id = skb_vlan_tag_get(skb) & BITS_MASK(12);
1620 		txd_vtag = 1;
1621 	}
1622 
1623 	txdd->length = CPU_CHIP_SWAP16(skb->len);
1624 	txdd->mss = CPU_CHIP_SWAP16(txd_mss);
1625 	txdd->txd_val1 =
1626 	    CPU_CHIP_SWAP32(TXD_W1_VAL
1627 			    (txd_sizes[nr_frags].qwords, txd_checksum, txd_vtag,
1628 			     txd_lgsnd, txd_vlan_id));
1629 	DBG("=== TxD desc =====================\n");
1630 	DBG("=== w1: 0x%x ================\n", txdd->txd_val1);
1631 	DBG("=== w2: mss 0x%x len 0x%x\n", txdd->mss, txdd->length);
1632 
1633 	bdx_tx_map_skb(priv, skb, txdd);
1634 
1635 	/* increment TXD write pointer. In case of
1636 	   fifo wrapping copy reminder of the descriptor
1637 	   to the beginning */
1638 	f->m.wptr += txd_sizes[nr_frags].bytes;
1639 	len = f->m.wptr - f->m.memsz;
1640 	if (unlikely(len >= 0)) {
1641 		f->m.wptr = len;
1642 		if (len > 0) {
1643 			BDX_ASSERT(len > f->m.memsz);
1644 			memcpy(f->m.va, f->m.va + f->m.memsz, len);
1645 		}
1646 	}
1647 	BDX_ASSERT(f->m.wptr >= f->m.memsz);	/* finished with valid wptr */
1648 
1649 	priv->tx_level -= txd_sizes[nr_frags].bytes;
1650 	BDX_ASSERT(priv->tx_level <= 0 || priv->tx_level > BDX_MAX_TX_LEVEL);
1651 #ifdef BDX_DELAY_WPTR
1652 	if (priv->tx_level > priv->tx_update_mark) {
1653 		/* Force memory writes to complete before letting h/w
1654 		   know there are new descriptors to fetch.
1655 		   (might be needed on platforms like IA64)
1656 		   wmb(); */
1657 		WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1658 	} else {
1659 		if (priv->tx_noupd++ > BDX_NO_UPD_PACKETS) {
1660 			priv->tx_noupd = 0;
1661 			WRITE_REG(priv, f->m.reg_WPTR,
1662 				  f->m.wptr & TXF_WPTR_WR_PTR);
1663 		}
1664 	}
1665 #else
1666 	/* Force memory writes to complete before letting h/w
1667 	   know there are new descriptors to fetch.
1668 	   (might be needed on platforms like IA64)
1669 	   wmb(); */
1670 	WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1671 
1672 #endif
1673 #ifdef BDX_LLTX
1674 	netif_trans_update(ndev); /* NETIF_F_LLTX driver :( */
1675 #endif
1676 	ndev->stats.tx_packets++;
1677 	ndev->stats.tx_bytes += skb->len;
1678 
1679 	if (priv->tx_level < BDX_MIN_TX_LEVEL) {
1680 		DBG("%s: %s: TX Q STOP level %d\n",
1681 		    BDX_DRV_NAME, ndev->name, priv->tx_level);
1682 		netif_stop_queue(ndev);
1683 	}
1684 
1685 	spin_unlock_irqrestore(&priv->tx_lock, flags);
1686 	return NETDEV_TX_OK;
1687 }
1688 
1689 /**
1690  * bdx_tx_cleanup - clean TXF fifo, run in the context of IRQ.
1691  * @priv: bdx adapter
1692  *
1693  * It scans TXF fifo for descriptors, frees DMA mappings and reports to OS
1694  * that those packets were sent
1695  */
1696 static void bdx_tx_cleanup(struct bdx_priv *priv)
1697 {
1698 	struct txf_fifo *f = &priv->txf_fifo0;
1699 	struct txdb *db = &priv->txdb;
1700 	int tx_level = 0;
1701 
1702 	ENTER;
1703 	f->m.wptr = READ_REG(priv, f->m.reg_WPTR) & TXF_WPTR_MASK;
1704 	BDX_ASSERT(f->m.rptr >= f->m.memsz);	/* started with valid rptr */
1705 
1706 	while (f->m.wptr != f->m.rptr) {
1707 		f->m.rptr += BDX_TXF_DESC_SZ;
1708 		f->m.rptr &= f->m.size_mask;
1709 
1710 		/* unmap all the fragments */
1711 		/* first has to come tx_maps containing dma */
1712 		BDX_ASSERT(db->rptr->len == 0);
1713 		do {
1714 			BDX_ASSERT(db->rptr->addr.dma == 0);
1715 			dma_unmap_page(&priv->pdev->dev, db->rptr->addr.dma,
1716 				       db->rptr->len, DMA_TO_DEVICE);
1717 			bdx_tx_db_inc_rptr(db);
1718 		} while (db->rptr->len > 0);
1719 		tx_level -= db->rptr->len;	/* '-' koz len is negative */
1720 
1721 		/* now should come skb pointer - free it */
1722 		dev_consume_skb_irq(db->rptr->addr.skb);
1723 		bdx_tx_db_inc_rptr(db);
1724 	}
1725 
1726 	/* let h/w know which TXF descriptors were cleaned */
1727 	BDX_ASSERT((f->m.wptr & TXF_WPTR_WR_PTR) >= f->m.memsz);
1728 	WRITE_REG(priv, f->m.reg_RPTR, f->m.rptr & TXF_WPTR_WR_PTR);
1729 
1730 	/* We reclaimed resources, so in case the Q is stopped by xmit callback,
1731 	 * we resume the transmission and use tx_lock to synchronize with xmit.*/
1732 	spin_lock(&priv->tx_lock);
1733 	priv->tx_level += tx_level;
1734 	BDX_ASSERT(priv->tx_level <= 0 || priv->tx_level > BDX_MAX_TX_LEVEL);
1735 #ifdef BDX_DELAY_WPTR
1736 	if (priv->tx_noupd) {
1737 		priv->tx_noupd = 0;
1738 		WRITE_REG(priv, priv->txd_fifo0.m.reg_WPTR,
1739 			  priv->txd_fifo0.m.wptr & TXF_WPTR_WR_PTR);
1740 	}
1741 #endif
1742 
1743 	if (unlikely(netif_queue_stopped(priv->ndev) &&
1744 		     netif_carrier_ok(priv->ndev) &&
1745 		     (priv->tx_level >= BDX_MIN_TX_LEVEL))) {
1746 		DBG("%s: %s: TX Q WAKE level %d\n",
1747 		    BDX_DRV_NAME, priv->ndev->name, priv->tx_level);
1748 		netif_wake_queue(priv->ndev);
1749 	}
1750 	spin_unlock(&priv->tx_lock);
1751 }
1752 
1753 /**
1754  * bdx_tx_free_skbs - frees all skbs from TXD fifo.
1755  * @priv: NIC private structure
1756  *
1757  * It gets called when OS stops this dev, eg upon "ifconfig down" or rmmod
1758  */
1759 static void bdx_tx_free_skbs(struct bdx_priv *priv)
1760 {
1761 	struct txdb *db = &priv->txdb;
1762 
1763 	ENTER;
1764 	while (db->rptr != db->wptr) {
1765 		if (likely(db->rptr->len))
1766 			dma_unmap_page(&priv->pdev->dev, db->rptr->addr.dma,
1767 				       db->rptr->len, DMA_TO_DEVICE);
1768 		else
1769 			dev_kfree_skb(db->rptr->addr.skb);
1770 		bdx_tx_db_inc_rptr(db);
1771 	}
1772 	RET();
1773 }
1774 
1775 /* bdx_tx_free - frees all Tx resources */
1776 static void bdx_tx_free(struct bdx_priv *priv)
1777 {
1778 	ENTER;
1779 	bdx_tx_free_skbs(priv);
1780 	bdx_fifo_free(priv, &priv->txd_fifo0.m);
1781 	bdx_fifo_free(priv, &priv->txf_fifo0.m);
1782 	bdx_tx_db_close(&priv->txdb);
1783 }
1784 
1785 /**
1786  * bdx_tx_push_desc - push descriptor to TxD fifo
1787  * @priv: NIC private structure
1788  * @data: desc's data
1789  * @size: desc's size
1790  *
1791  * Pushes desc to TxD fifo and overlaps it if needed.
1792  * NOTE: this func does not check for available space. this is responsibility
1793  *    of the caller. Neither does it check that data size is smaller than
1794  *    fifo size.
1795  */
1796 static void bdx_tx_push_desc(struct bdx_priv *priv, void *data, int size)
1797 {
1798 	struct txd_fifo *f = &priv->txd_fifo0;
1799 	int i = f->m.memsz - f->m.wptr;
1800 
1801 	if (size == 0)
1802 		return;
1803 
1804 	if (i > size) {
1805 		memcpy(f->m.va + f->m.wptr, data, size);
1806 		f->m.wptr += size;
1807 	} else {
1808 		memcpy(f->m.va + f->m.wptr, data, i);
1809 		f->m.wptr = size - i;
1810 		memcpy(f->m.va, data + i, f->m.wptr);
1811 	}
1812 	WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1813 }
1814 
1815 /**
1816  * bdx_tx_push_desc_safe - push descriptor to TxD fifo in a safe way
1817  * @priv: NIC private structure
1818  * @data: desc's data
1819  * @size: desc's size
1820  *
1821  * NOTE: this func does check for available space and, if necessary, waits for
1822  *   NIC to read existing data before writing new one.
1823  */
1824 static void bdx_tx_push_desc_safe(struct bdx_priv *priv, void *data, int size)
1825 {
1826 	int timer = 0;
1827 	ENTER;
1828 
1829 	while (size > 0) {
1830 		/* we substruct 8 because when fifo is full rptr == wptr
1831 		   which also means that fifo is empty, we can understand
1832 		   the difference, but could hw do the same ??? :) */
1833 		int avail = bdx_tx_space(priv) - 8;
1834 		if (avail <= 0) {
1835 			if (timer++ > 300) {	/* prevent endless loop */
1836 				DBG("timeout while writing desc to TxD fifo\n");
1837 				break;
1838 			}
1839 			udelay(50);	/* give hw a chance to clean fifo */
1840 			continue;
1841 		}
1842 		avail = min(avail, size);
1843 		DBG("about to push  %d bytes starting %p size %d\n", avail,
1844 		    data, size);
1845 		bdx_tx_push_desc(priv, data, avail);
1846 		size -= avail;
1847 		data += avail;
1848 	}
1849 	RET();
1850 }
1851 
1852 static const struct net_device_ops bdx_netdev_ops = {
1853 	.ndo_open		= bdx_open,
1854 	.ndo_stop		= bdx_close,
1855 	.ndo_start_xmit		= bdx_tx_transmit,
1856 	.ndo_validate_addr	= eth_validate_addr,
1857 	.ndo_siocdevprivate	= bdx_siocdevprivate,
1858 	.ndo_set_rx_mode	= bdx_setmulti,
1859 	.ndo_change_mtu		= bdx_change_mtu,
1860 	.ndo_set_mac_address	= bdx_set_mac,
1861 	.ndo_vlan_rx_add_vid	= bdx_vlan_rx_add_vid,
1862 	.ndo_vlan_rx_kill_vid	= bdx_vlan_rx_kill_vid,
1863 };
1864 
1865 /**
1866  * bdx_probe - Device Initialization Routine
1867  * @pdev: PCI device information struct
1868  * @ent: entry in bdx_pci_tbl
1869  *
1870  * Returns 0 on success, negative on failure
1871  *
1872  * bdx_probe initializes an adapter identified by a pci_dev structure.
1873  * The OS initialization, configuring of the adapter private structure,
1874  * and a hardware reset occur.
1875  *
1876  * functions and their order used as explained in
1877  * /usr/src/linux/Documentation/DMA-{API,mapping}.txt
1878  *
1879  */
1880 
1881 /* TBD: netif_msg should be checked and implemented. I disable it for now */
1882 static int
1883 bdx_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1884 {
1885 	struct net_device *ndev;
1886 	struct bdx_priv *priv;
1887 	unsigned long pciaddr;
1888 	u32 regionSize;
1889 	struct pci_nic *nic;
1890 	int err, port;
1891 
1892 	ENTER;
1893 
1894 	nic = vmalloc(sizeof(*nic));
1895 	if (!nic)
1896 		RET(-ENOMEM);
1897 
1898     /************** pci *****************/
1899 	err = pci_enable_device(pdev);
1900 	if (err)			/* it triggers interrupt, dunno why. */
1901 		goto err_pci;		/* it's not a problem though */
1902 
1903 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
1904 	if (err) {
1905 		pr_err("No usable DMA configuration, aborting\n");
1906 		goto err_dma;
1907 	}
1908 
1909 	err = pci_request_regions(pdev, BDX_DRV_NAME);
1910 	if (err)
1911 		goto err_dma;
1912 
1913 	pci_set_master(pdev);
1914 
1915 	pciaddr = pci_resource_start(pdev, 0);
1916 	if (!pciaddr) {
1917 		err = -EIO;
1918 		pr_err("no MMIO resource\n");
1919 		goto err_out_res;
1920 	}
1921 	regionSize = pci_resource_len(pdev, 0);
1922 	if (regionSize < BDX_REGS_SIZE) {
1923 		err = -EIO;
1924 		pr_err("MMIO resource (%x) too small\n", regionSize);
1925 		goto err_out_res;
1926 	}
1927 
1928 	nic->regs = ioremap(pciaddr, regionSize);
1929 	if (!nic->regs) {
1930 		err = -EIO;
1931 		pr_err("ioremap failed\n");
1932 		goto err_out_res;
1933 	}
1934 
1935 	if (pdev->irq < 2) {
1936 		err = -EIO;
1937 		pr_err("invalid irq (%d)\n", pdev->irq);
1938 		goto err_out_iomap;
1939 	}
1940 	pci_set_drvdata(pdev, nic);
1941 
1942 	if (pdev->device == 0x3014)
1943 		nic->port_num = 2;
1944 	else
1945 		nic->port_num = 1;
1946 
1947 	print_hw_id(pdev);
1948 
1949 	bdx_hw_reset_direct(nic->regs);
1950 
1951 	nic->irq_type = IRQ_INTX;
1952 #ifdef BDX_MSI
1953 	if ((readl(nic->regs + FPGA_VER) & 0xFFF) >= 378) {
1954 		err = pci_enable_msi(pdev);
1955 		if (err)
1956 			pr_err("Can't enable msi. error is %d\n", err);
1957 		else
1958 			nic->irq_type = IRQ_MSI;
1959 	} else
1960 		DBG("HW does not support MSI\n");
1961 #endif
1962 
1963     /************** netdev **************/
1964 	for (port = 0; port < nic->port_num; port++) {
1965 		ndev = alloc_etherdev(sizeof(struct bdx_priv));
1966 		if (!ndev) {
1967 			err = -ENOMEM;
1968 			goto err_out_iomap;
1969 		}
1970 
1971 		ndev->netdev_ops = &bdx_netdev_ops;
1972 		ndev->tx_queue_len = BDX_NDEV_TXQ_LEN;
1973 
1974 		bdx_set_ethtool_ops(ndev);	/* ethtool interface */
1975 
1976 		/* these fields are used for info purposes only
1977 		 * so we can have them same for all ports of the board */
1978 		ndev->if_port = port;
1979 		ndev->features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO |
1980 		    NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
1981 		    NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_RXCSUM |
1982 		    NETIF_F_HIGHDMA;
1983 
1984 		ndev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG |
1985 			NETIF_F_TSO | NETIF_F_HW_VLAN_CTAG_TX;
1986 
1987 	/************** priv ****************/
1988 		priv = nic->priv[port] = netdev_priv(ndev);
1989 
1990 		priv->pBdxRegs = nic->regs + port * 0x8000;
1991 		priv->port = port;
1992 		priv->pdev = pdev;
1993 		priv->ndev = ndev;
1994 		priv->nic = nic;
1995 		priv->msg_enable = BDX_DEF_MSG_ENABLE;
1996 
1997 		netif_napi_add(ndev, &priv->napi, bdx_poll, 64);
1998 
1999 		if ((readl(nic->regs + FPGA_VER) & 0xFFF) == 308) {
2000 			DBG("HW statistics not supported\n");
2001 			priv->stats_flag = 0;
2002 		} else {
2003 			priv->stats_flag = 1;
2004 		}
2005 
2006 		/* Initialize fifo sizes. */
2007 		priv->txd_size = 2;
2008 		priv->txf_size = 2;
2009 		priv->rxd_size = 2;
2010 		priv->rxf_size = 3;
2011 
2012 		/* Initialize the initial coalescing registers. */
2013 		priv->rdintcm = INT_REG_VAL(0x20, 1, 4, 12);
2014 		priv->tdintcm = INT_REG_VAL(0x20, 1, 0, 12);
2015 
2016 		/* ndev->xmit_lock spinlock is not used.
2017 		 * Private priv->tx_lock is used for synchronization
2018 		 * between transmit and TX irq cleanup.  In addition
2019 		 * set multicast list callback has to use priv->tx_lock.
2020 		 */
2021 #ifdef BDX_LLTX
2022 		ndev->features |= NETIF_F_LLTX;
2023 #endif
2024 		/* MTU range: 60 - 16384 */
2025 		ndev->min_mtu = ETH_ZLEN;
2026 		ndev->max_mtu = BDX_MAX_MTU;
2027 
2028 		spin_lock_init(&priv->tx_lock);
2029 
2030 		/*bdx_hw_reset(priv); */
2031 		if (bdx_read_mac(priv)) {
2032 			pr_err("load MAC address failed\n");
2033 			err = -EFAULT;
2034 			goto err_out_iomap;
2035 		}
2036 		SET_NETDEV_DEV(ndev, &pdev->dev);
2037 		err = register_netdev(ndev);
2038 		if (err) {
2039 			pr_err("register_netdev failed\n");
2040 			goto err_out_free;
2041 		}
2042 		netif_carrier_off(ndev);
2043 		netif_stop_queue(ndev);
2044 
2045 		print_eth_id(ndev);
2046 	}
2047 	RET(0);
2048 
2049 err_out_free:
2050 	free_netdev(ndev);
2051 err_out_iomap:
2052 	iounmap(nic->regs);
2053 err_out_res:
2054 	pci_release_regions(pdev);
2055 err_dma:
2056 	pci_disable_device(pdev);
2057 err_pci:
2058 	vfree(nic);
2059 
2060 	RET(err);
2061 }
2062 
2063 /****************** Ethtool interface *********************/
2064 /* get strings for statistics counters */
2065 static const char
2066  bdx_stat_names[][ETH_GSTRING_LEN] = {
2067 	"InUCast",		/* 0x7200 */
2068 	"InMCast",		/* 0x7210 */
2069 	"InBCast",		/* 0x7220 */
2070 	"InPkts",		/* 0x7230 */
2071 	"InErrors",		/* 0x7240 */
2072 	"InDropped",		/* 0x7250 */
2073 	"FrameTooLong",		/* 0x7260 */
2074 	"FrameSequenceErrors",	/* 0x7270 */
2075 	"InVLAN",		/* 0x7280 */
2076 	"InDroppedDFE",		/* 0x7290 */
2077 	"InDroppedIntFull",	/* 0x72A0 */
2078 	"InFrameAlignErrors",	/* 0x72B0 */
2079 
2080 	/* 0x72C0-0x72E0 RSRV */
2081 
2082 	"OutUCast",		/* 0x72F0 */
2083 	"OutMCast",		/* 0x7300 */
2084 	"OutBCast",		/* 0x7310 */
2085 	"OutPkts",		/* 0x7320 */
2086 
2087 	/* 0x7330-0x7360 RSRV */
2088 
2089 	"OutVLAN",		/* 0x7370 */
2090 	"InUCastOctects",	/* 0x7380 */
2091 	"OutUCastOctects",	/* 0x7390 */
2092 
2093 	/* 0x73A0-0x73B0 RSRV */
2094 
2095 	"InBCastOctects",	/* 0x73C0 */
2096 	"OutBCastOctects",	/* 0x73D0 */
2097 	"InOctects",		/* 0x73E0 */
2098 	"OutOctects",		/* 0x73F0 */
2099 };
2100 
2101 /*
2102  * bdx_get_link_ksettings - get device-specific settings
2103  * @netdev
2104  * @ecmd
2105  */
2106 static int bdx_get_link_ksettings(struct net_device *netdev,
2107 				  struct ethtool_link_ksettings *ecmd)
2108 {
2109 	ethtool_link_ksettings_zero_link_mode(ecmd, supported);
2110 	ethtool_link_ksettings_add_link_mode(ecmd, supported,
2111 					     10000baseT_Full);
2112 	ethtool_link_ksettings_add_link_mode(ecmd, supported, FIBRE);
2113 	ethtool_link_ksettings_zero_link_mode(ecmd, advertising);
2114 	ethtool_link_ksettings_add_link_mode(ecmd, advertising,
2115 					     10000baseT_Full);
2116 	ethtool_link_ksettings_add_link_mode(ecmd, advertising, FIBRE);
2117 
2118 	ecmd->base.speed = SPEED_10000;
2119 	ecmd->base.duplex = DUPLEX_FULL;
2120 	ecmd->base.port = PORT_FIBRE;
2121 	ecmd->base.autoneg = AUTONEG_DISABLE;
2122 
2123 	return 0;
2124 }
2125 
2126 /*
2127  * bdx_get_drvinfo - report driver information
2128  * @netdev
2129  * @drvinfo
2130  */
2131 static void
2132 bdx_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *drvinfo)
2133 {
2134 	struct bdx_priv *priv = netdev_priv(netdev);
2135 
2136 	strlcpy(drvinfo->driver, BDX_DRV_NAME, sizeof(drvinfo->driver));
2137 	strlcpy(drvinfo->version, BDX_DRV_VERSION, sizeof(drvinfo->version));
2138 	strlcpy(drvinfo->fw_version, "N/A", sizeof(drvinfo->fw_version));
2139 	strlcpy(drvinfo->bus_info, pci_name(priv->pdev),
2140 		sizeof(drvinfo->bus_info));
2141 }
2142 
2143 /*
2144  * bdx_get_coalesce - get interrupt coalescing parameters
2145  * @netdev
2146  * @ecoal
2147  */
2148 static int bdx_get_coalesce(struct net_device *netdev,
2149 			    struct ethtool_coalesce *ecoal,
2150 			    struct kernel_ethtool_coalesce *kernel_coal,
2151 			    struct netlink_ext_ack *extack)
2152 {
2153 	u32 rdintcm;
2154 	u32 tdintcm;
2155 	struct bdx_priv *priv = netdev_priv(netdev);
2156 
2157 	rdintcm = priv->rdintcm;
2158 	tdintcm = priv->tdintcm;
2159 
2160 	/* PCK_TH measures in multiples of FIFO bytes
2161 	   We translate to packets */
2162 	ecoal->rx_coalesce_usecs = GET_INT_COAL(rdintcm) * INT_COAL_MULT;
2163 	ecoal->rx_max_coalesced_frames =
2164 	    ((GET_PCK_TH(rdintcm) * PCK_TH_MULT) / sizeof(struct rxf_desc));
2165 
2166 	ecoal->tx_coalesce_usecs = GET_INT_COAL(tdintcm) * INT_COAL_MULT;
2167 	ecoal->tx_max_coalesced_frames =
2168 	    ((GET_PCK_TH(tdintcm) * PCK_TH_MULT) / BDX_TXF_DESC_SZ);
2169 
2170 	/* adaptive parameters ignored */
2171 	return 0;
2172 }
2173 
2174 /*
2175  * bdx_set_coalesce - set interrupt coalescing parameters
2176  * @netdev
2177  * @ecoal
2178  */
2179 static int bdx_set_coalesce(struct net_device *netdev,
2180 			    struct ethtool_coalesce *ecoal,
2181 			    struct kernel_ethtool_coalesce *kernel_coal,
2182 			    struct netlink_ext_ack *extack)
2183 {
2184 	u32 rdintcm;
2185 	u32 tdintcm;
2186 	struct bdx_priv *priv = netdev_priv(netdev);
2187 	int rx_coal;
2188 	int tx_coal;
2189 	int rx_max_coal;
2190 	int tx_max_coal;
2191 
2192 	/* Check for valid input */
2193 	rx_coal = ecoal->rx_coalesce_usecs / INT_COAL_MULT;
2194 	tx_coal = ecoal->tx_coalesce_usecs / INT_COAL_MULT;
2195 	rx_max_coal = ecoal->rx_max_coalesced_frames;
2196 	tx_max_coal = ecoal->tx_max_coalesced_frames;
2197 
2198 	/* Translate from packets to multiples of FIFO bytes */
2199 	rx_max_coal =
2200 	    (((rx_max_coal * sizeof(struct rxf_desc)) + PCK_TH_MULT - 1)
2201 	     / PCK_TH_MULT);
2202 	tx_max_coal =
2203 	    (((tx_max_coal * BDX_TXF_DESC_SZ) + PCK_TH_MULT - 1)
2204 	     / PCK_TH_MULT);
2205 
2206 	if ((rx_coal > 0x7FFF) || (tx_coal > 0x7FFF) ||
2207 	    (rx_max_coal > 0xF) || (tx_max_coal > 0xF))
2208 		return -EINVAL;
2209 
2210 	rdintcm = INT_REG_VAL(rx_coal, GET_INT_COAL_RC(priv->rdintcm),
2211 			      GET_RXF_TH(priv->rdintcm), rx_max_coal);
2212 	tdintcm = INT_REG_VAL(tx_coal, GET_INT_COAL_RC(priv->tdintcm), 0,
2213 			      tx_max_coal);
2214 
2215 	priv->rdintcm = rdintcm;
2216 	priv->tdintcm = tdintcm;
2217 
2218 	WRITE_REG(priv, regRDINTCM0, rdintcm);
2219 	WRITE_REG(priv, regTDINTCM0, tdintcm);
2220 
2221 	return 0;
2222 }
2223 
2224 /* Convert RX fifo size to number of pending packets */
2225 static inline int bdx_rx_fifo_size_to_packets(int rx_size)
2226 {
2227 	return (FIFO_SIZE * (1 << rx_size)) / sizeof(struct rxf_desc);
2228 }
2229 
2230 /* Convert TX fifo size to number of pending packets */
2231 static inline int bdx_tx_fifo_size_to_packets(int tx_size)
2232 {
2233 	return (FIFO_SIZE * (1 << tx_size)) / BDX_TXF_DESC_SZ;
2234 }
2235 
2236 /*
2237  * bdx_get_ringparam - report ring sizes
2238  * @netdev
2239  * @ring
2240  * @kernel_ring
2241  * @extack
2242  */
2243 static void
2244 bdx_get_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring,
2245 		  struct kernel_ethtool_ringparam *kernel_ring,
2246 		  struct netlink_ext_ack *extack)
2247 {
2248 	struct bdx_priv *priv = netdev_priv(netdev);
2249 
2250 	/*max_pending - the maximum-sized FIFO we allow */
2251 	ring->rx_max_pending = bdx_rx_fifo_size_to_packets(3);
2252 	ring->tx_max_pending = bdx_tx_fifo_size_to_packets(3);
2253 	ring->rx_pending = bdx_rx_fifo_size_to_packets(priv->rxf_size);
2254 	ring->tx_pending = bdx_tx_fifo_size_to_packets(priv->txd_size);
2255 }
2256 
2257 /*
2258  * bdx_set_ringparam - set ring sizes
2259  * @netdev
2260  * @ring
2261  * @kernel_ring
2262  * @extack
2263  */
2264 static int
2265 bdx_set_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring,
2266 		  struct kernel_ethtool_ringparam *kernel_ring,
2267 		  struct netlink_ext_ack *extack)
2268 {
2269 	struct bdx_priv *priv = netdev_priv(netdev);
2270 	int rx_size = 0;
2271 	int tx_size = 0;
2272 
2273 	for (; rx_size < 4; rx_size++) {
2274 		if (bdx_rx_fifo_size_to_packets(rx_size) >= ring->rx_pending)
2275 			break;
2276 	}
2277 	if (rx_size == 4)
2278 		rx_size = 3;
2279 
2280 	for (; tx_size < 4; tx_size++) {
2281 		if (bdx_tx_fifo_size_to_packets(tx_size) >= ring->tx_pending)
2282 			break;
2283 	}
2284 	if (tx_size == 4)
2285 		tx_size = 3;
2286 
2287 	/*Is there anything to do? */
2288 	if ((rx_size == priv->rxf_size) &&
2289 	    (tx_size == priv->txd_size))
2290 		return 0;
2291 
2292 	priv->rxf_size = rx_size;
2293 	if (rx_size > 1)
2294 		priv->rxd_size = rx_size - 1;
2295 	else
2296 		priv->rxd_size = rx_size;
2297 
2298 	priv->txf_size = priv->txd_size = tx_size;
2299 
2300 	if (netif_running(netdev)) {
2301 		bdx_close(netdev);
2302 		bdx_open(netdev);
2303 	}
2304 	return 0;
2305 }
2306 
2307 /*
2308  * bdx_get_strings - return a set of strings that describe the requested objects
2309  * @netdev
2310  * @data
2311  */
2312 static void bdx_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
2313 {
2314 	switch (stringset) {
2315 	case ETH_SS_STATS:
2316 		memcpy(data, *bdx_stat_names, sizeof(bdx_stat_names));
2317 		break;
2318 	}
2319 }
2320 
2321 /*
2322  * bdx_get_sset_count - return number of statistics or tests
2323  * @netdev
2324  */
2325 static int bdx_get_sset_count(struct net_device *netdev, int stringset)
2326 {
2327 	struct bdx_priv *priv = netdev_priv(netdev);
2328 
2329 	switch (stringset) {
2330 	case ETH_SS_STATS:
2331 		BDX_ASSERT(ARRAY_SIZE(bdx_stat_names)
2332 			   != sizeof(struct bdx_stats) / sizeof(u64));
2333 		return (priv->stats_flag) ? ARRAY_SIZE(bdx_stat_names)	: 0;
2334 	}
2335 
2336 	return -EINVAL;
2337 }
2338 
2339 /*
2340  * bdx_get_ethtool_stats - return device's hardware L2 statistics
2341  * @netdev
2342  * @stats
2343  * @data
2344  */
2345 static void bdx_get_ethtool_stats(struct net_device *netdev,
2346 				  struct ethtool_stats *stats, u64 *data)
2347 {
2348 	struct bdx_priv *priv = netdev_priv(netdev);
2349 
2350 	if (priv->stats_flag) {
2351 
2352 		/* Update stats from HW */
2353 		bdx_update_stats(priv);
2354 
2355 		/* Copy data to user buffer */
2356 		memcpy(data, &priv->hw_stats, sizeof(priv->hw_stats));
2357 	}
2358 }
2359 
2360 /*
2361  * bdx_set_ethtool_ops - ethtool interface implementation
2362  * @netdev
2363  */
2364 static void bdx_set_ethtool_ops(struct net_device *netdev)
2365 {
2366 	static const struct ethtool_ops bdx_ethtool_ops = {
2367 		.supported_coalesce_params = ETHTOOL_COALESCE_USECS |
2368 					     ETHTOOL_COALESCE_MAX_FRAMES,
2369 		.get_drvinfo = bdx_get_drvinfo,
2370 		.get_link = ethtool_op_get_link,
2371 		.get_coalesce = bdx_get_coalesce,
2372 		.set_coalesce = bdx_set_coalesce,
2373 		.get_ringparam = bdx_get_ringparam,
2374 		.set_ringparam = bdx_set_ringparam,
2375 		.get_strings = bdx_get_strings,
2376 		.get_sset_count = bdx_get_sset_count,
2377 		.get_ethtool_stats = bdx_get_ethtool_stats,
2378 		.get_link_ksettings = bdx_get_link_ksettings,
2379 	};
2380 
2381 	netdev->ethtool_ops = &bdx_ethtool_ops;
2382 }
2383 
2384 /**
2385  * bdx_remove - Device Removal Routine
2386  * @pdev: PCI device information struct
2387  *
2388  * bdx_remove is called by the PCI subsystem to alert the driver
2389  * that it should release a PCI device.  The could be caused by a
2390  * Hot-Plug event, or because the driver is going to be removed from
2391  * memory.
2392  **/
2393 static void bdx_remove(struct pci_dev *pdev)
2394 {
2395 	struct pci_nic *nic = pci_get_drvdata(pdev);
2396 	struct net_device *ndev;
2397 	int port;
2398 
2399 	for (port = 0; port < nic->port_num; port++) {
2400 		ndev = nic->priv[port]->ndev;
2401 		unregister_netdev(ndev);
2402 		free_netdev(ndev);
2403 	}
2404 
2405 	/*bdx_hw_reset_direct(nic->regs); */
2406 #ifdef BDX_MSI
2407 	if (nic->irq_type == IRQ_MSI)
2408 		pci_disable_msi(pdev);
2409 #endif
2410 
2411 	iounmap(nic->regs);
2412 	pci_release_regions(pdev);
2413 	pci_disable_device(pdev);
2414 	vfree(nic);
2415 
2416 	RET();
2417 }
2418 
2419 static struct pci_driver bdx_pci_driver = {
2420 	.name = BDX_DRV_NAME,
2421 	.id_table = bdx_pci_tbl,
2422 	.probe = bdx_probe,
2423 	.remove = bdx_remove,
2424 };
2425 
2426 /*
2427  * print_driver_id - print parameters of the driver build
2428  */
2429 static void __init print_driver_id(void)
2430 {
2431 	pr_info("%s, %s\n", BDX_DRV_DESC, BDX_DRV_VERSION);
2432 	pr_info("Options: hw_csum %s\n", BDX_MSI_STRING);
2433 }
2434 
2435 static int __init bdx_module_init(void)
2436 {
2437 	ENTER;
2438 	init_txd_sizes();
2439 	print_driver_id();
2440 	RET(pci_register_driver(&bdx_pci_driver));
2441 }
2442 
2443 module_init(bdx_module_init);
2444 
2445 static void __exit bdx_module_exit(void)
2446 {
2447 	ENTER;
2448 	pci_unregister_driver(&bdx_pci_driver);
2449 	RET();
2450 }
2451 
2452 module_exit(bdx_module_exit);
2453 
2454 MODULE_LICENSE("GPL");
2455 MODULE_AUTHOR(DRIVER_AUTHOR);
2456 MODULE_DESCRIPTION(BDX_DRV_DESC);
2457 MODULE_FIRMWARE("tehuti/bdx.bin");
2458