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